projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
fs: rcu-walk aware d_revalidate method
[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 */
31e6b01f
NP		 172static inline int __acl_permission_check(struct inode *inode, int mask,
173 int (*check_acl)(struct inode *inode, int mask), int rcu)
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) {
31e6b01f
NP		 183 if (rcu) {
184 return -ECHILD;
185 } else {
186 int error = check_acl(inode, mask);
187 if (error != -EAGAIN)
188 return error;
189 }
1da177e4
LT		 190 }
191
192 if (in_group_p(inode->i_gid))
193 mode >>= 3;
194 }
195
196 /*
197 * If the DACs are ok we don't need any capability check.
198 */
e6305c43 199 if ((mask & ~mode) == 0)
1da177e4 200 return 0;
5909ccaa
LT		 201 return -EACCES;
202}
203
31e6b01f
NP		 204static inline int acl_permission_check(struct inode *inode, int mask,
205 int (*check_acl)(struct inode *inode, int mask))
206{
207 return __acl_permission_check(inode, mask, check_acl, 0);
208}
209
5909ccaa
LT		 210/**
211 * generic_permission - check for access rights on a Posix-like filesystem
212 * @inode: inode to check access rights for
213 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
214 * @check_acl: optional callback to check for Posix ACLs
215 *
216 * Used to check for read/write/execute permissions on a file.
217 * We use "fsuid" for this, letting us set arbitrary permissions
218 * for filesystem access without changing the "normal" uids which
219 * are used for other things..
220 */
221int generic_permission(struct inode *inode, int mask,
222 int (*check_acl)(struct inode *inode, int mask))
223{
224 int ret;
225
226 /*
227 * Do the basic POSIX ACL permission checks.
228 */
229 ret = acl_permission_check(inode, mask, check_acl);
230 if (ret != -EACCES)
231 return ret;
1da177e4 232
1da177e4
LT		 233 /*
234 * Read/write DACs are always overridable.
235 * Executable DACs are overridable if at least one exec bit is set.
236 */
f696a365 237 if (!(mask & MAY_EXEC) || execute_ok(inode))
1da177e4
LT		 238 if (capable(CAP_DAC_OVERRIDE))
239 return 0;
240
241 /*
242 * Searching includes executable on directories, else just read.
243 */
7ea66001 244 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
1da177e4
LT		 245 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
246 if (capable(CAP_DAC_READ_SEARCH))
247 return 0;
248
249 return -EACCES;
250}
251
cb23beb5
CH		 252/**
253 * inode_permission - check for access rights to a given inode
254 * @inode: inode to check permission on
255 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
256 *
257 * Used to check for read/write/execute permissions on an inode.
258 * We use "fsuid" for this, letting us set arbitrary permissions
259 * for filesystem access without changing the "normal" uids which
260 * are used for other things.
261 */
f419a2e3 262int inode_permission(struct inode *inode, int mask)
1da177e4 263{
e6305c43 264 int retval;
1da177e4
LT		 265
266 if (mask & MAY_WRITE) {
22590e41 267 umode_t mode = inode->i_mode;
1da177e4
LT		 268
269 /*
270 * Nobody gets write access to a read-only fs.
271 */
272 if (IS_RDONLY(inode) &&
273 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
274 return -EROFS;
275
276 /*
277 * Nobody gets write access to an immutable file.
278 */
279 if (IS_IMMUTABLE(inode))
280 return -EACCES;
281 }
282
acfa4380 283 if (inode->i_op->permission)
b77b0646 284 retval = inode->i_op->permission(inode, mask);
f696a365 285 else
5909ccaa 286 retval = generic_permission(inode, mask, inode->i_op->check_acl);
f696a365 287
1da177e4
LT		 288 if (retval)
289 return retval;
290
08ce5f16
SH		 291 retval = devcgroup_inode_permission(inode, mask);
292 if (retval)
293 return retval;
294
d09ca739 295 return security_inode_permission(inode, mask);
1da177e4
LT		 296}
297
8c744fb8
CH		 298/**
299 * file_permission - check for additional access rights to a given file
300 * @file: file to check access rights for
301 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
302 *
303 * Used to check for read/write/execute permissions on an already opened
304 * file.
305 *
306 * Note:
307 * Do not use this function in new code. All access checks should
cb23beb5 308 * be done using inode_permission().
8c744fb8
CH		 309 */
310int file_permission(struct file *file, int mask)
311{
f419a2e3 312 return inode_permission(file->f_path.dentry->d_inode, mask);
8c744fb8
CH		 313}
314
1da177e4
LT		 315/*
316 * get_write_access() gets write permission for a file.
317 * put_write_access() releases this write permission.
318 * This is used for regular files.
319 * We cannot support write (and maybe mmap read-write shared) accesses and
320 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
321 * can have the following values:
322 * 0: no writers, no VM_DENYWRITE mappings
323 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
324 * > 0: (i_writecount) users are writing to the file.
325 *
326 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
327 * except for the cases where we don't hold i_writecount yet. Then we need to
328 * use {get,deny}_write_access() - these functions check the sign and refuse
329 * to do the change if sign is wrong. Exclusion between them is provided by
330 * the inode->i_lock spinlock.
331 */
332
333int get_write_access(struct inode * inode)
334{
335 spin_lock(&inode->i_lock);
336 if (atomic_read(&inode->i_writecount) < 0) {
337 spin_unlock(&inode->i_lock);
338 return -ETXTBSY;
339 }
340 atomic_inc(&inode->i_writecount);
341 spin_unlock(&inode->i_lock);
342
343 return 0;
344}
345
346int deny_write_access(struct file * file)
347{
0f7fc9e4 348 struct inode *inode = file->f_path.dentry->d_inode;
1da177e4
LT		 349
350 spin_lock(&inode->i_lock);
351 if (atomic_read(&inode->i_writecount) > 0) {
352 spin_unlock(&inode->i_lock);
353 return -ETXTBSY;
354 }
355 atomic_dec(&inode->i_writecount);
356 spin_unlock(&inode->i_lock);
357
358 return 0;
359}
360
5dd784d0
JB		 361/**
362 * path_get - get a reference to a path
363 * @path: path to get the reference to
364 *
365 * Given a path increment the reference count to the dentry and the vfsmount.
366 */
367void path_get(struct path *path)
368{
369 mntget(path->mnt);
370 dget(path->dentry);
371}
372EXPORT_SYMBOL(path_get);
373
1d957f9b
JB		 374/**
375 * path_put - put a reference to a path
376 * @path: path to put the reference to
377 *
378 * Given a path decrement the reference count to the dentry and the vfsmount.
379 */
380void path_put(struct path *path)
1da177e4 381{
1d957f9b
JB		 382 dput(path->dentry);
383 mntput(path->mnt);
1da177e4 384}
1d957f9b 385EXPORT_SYMBOL(path_put);
1da177e4 386
31e6b01f
NP		 387/**
388 * nameidata_drop_rcu - drop this nameidata out of rcu-walk
389 * @nd: nameidata pathwalk data to drop
390 * @Returns: 0 on success, -ECHLID on failure
391 *
392 * Path walking has 2 modes, rcu-walk and ref-walk (see
393 * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt
394 * to drop out of rcu-walk mode and take normal reference counts on dentries
395 * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take
396 * refcounts at the last known good point before rcu-walk got stuck, so
397 * ref-walk may continue from there. If this is not successful (eg. a seqcount
398 * has changed), then failure is returned and path walk restarts from the
399 * beginning in ref-walk mode.
400 *
401 * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into
402 * ref-walk. Must be called from rcu-walk context.
403 */
404static int nameidata_drop_rcu(struct nameidata *nd)
405{
406 struct fs_struct *fs = current->fs;
407 struct dentry *dentry = nd->path.dentry;
408
409 BUG_ON(!(nd->flags & LOOKUP_RCU));
410 if (nd->root.mnt) {
411 spin_lock(&fs->lock);
412 if (nd->root.mnt != fs->root.mnt ||
413 nd->root.dentry != fs->root.dentry)
414 goto err_root;
415 }
416 spin_lock(&dentry->d_lock);
417 if (!__d_rcu_to_refcount(dentry, nd->seq))
418 goto err;
419 BUG_ON(nd->inode != dentry->d_inode);
420 spin_unlock(&dentry->d_lock);
421 if (nd->root.mnt) {
422 path_get(&nd->root);
423 spin_unlock(&fs->lock);
424 }
425 mntget(nd->path.mnt);
426
427 rcu_read_unlock();
428 br_read_unlock(vfsmount_lock);
429 nd->flags &= ~LOOKUP_RCU;
430 return 0;
431err:
432 spin_unlock(&dentry->d_lock);
433err_root:
434 if (nd->root.mnt)
435 spin_unlock(&fs->lock);
436 return -ECHILD;
437}
438
439/* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
440static inline int nameidata_drop_rcu_maybe(struct nameidata *nd)
441{
442 if (nd->flags & LOOKUP_RCU)
443 return nameidata_drop_rcu(nd);
444 return 0;
445}
446
447/**
448 * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk
449 * @nd: nameidata pathwalk data to drop
450 * @dentry: dentry to drop
451 * @Returns: 0 on success, -ECHLID on failure
452 *
453 * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root,
454 * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on
455 * @nd. Must be called from rcu-walk context.
456 */
457static int nameidata_dentry_drop_rcu(struct nameidata *nd, struct dentry *dentry)
458{
459 struct fs_struct *fs = current->fs;
460 struct dentry *parent = nd->path.dentry;
461
462 BUG_ON(!(nd->flags & LOOKUP_RCU));
463 if (nd->root.mnt) {
464 spin_lock(&fs->lock);
465 if (nd->root.mnt != fs->root.mnt ||
466 nd->root.dentry != fs->root.dentry)
467 goto err_root;
468 }
469 spin_lock(&parent->d_lock);
470 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
471 if (!__d_rcu_to_refcount(dentry, nd->seq))
472 goto err;
473 /*
474 * If the sequence check on the child dentry passed, then the child has
475 * not been removed from its parent. This means the parent dentry must
476 * be valid and able to take a reference at this point.
477 */
478 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
479 BUG_ON(!parent->d_count);
480 parent->d_count++;
481 spin_unlock(&dentry->d_lock);
482 spin_unlock(&parent->d_lock);
483 if (nd->root.mnt) {
484 path_get(&nd->root);
485 spin_unlock(&fs->lock);
486 }
487 mntget(nd->path.mnt);
488
489 rcu_read_unlock();
490 br_read_unlock(vfsmount_lock);
491 nd->flags &= ~LOOKUP_RCU;
492 return 0;
493err:
494 spin_unlock(&dentry->d_lock);
495 spin_unlock(&parent->d_lock);
496err_root:
497 if (nd->root.mnt)
498 spin_unlock(&fs->lock);
499 return -ECHILD;
500}
501
502/* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
503static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata *nd, struct dentry *dentry)
504{
505 if (nd->flags & LOOKUP_RCU)
506 return nameidata_dentry_drop_rcu(nd, dentry);
507 return 0;
508}
509
510/**
511 * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk
512 * @nd: nameidata pathwalk data to drop
513 * @Returns: 0 on success, -ECHLID on failure
514 *
515 * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk.
516 * nd->path should be the final element of the lookup, so nd->root is discarded.
517 * Must be called from rcu-walk context.
518 */
519static int nameidata_drop_rcu_last(struct nameidata *nd)
520{
521 struct dentry *dentry = nd->path.dentry;
522
523 BUG_ON(!(nd->flags & LOOKUP_RCU));
524 nd->flags &= ~LOOKUP_RCU;
525 nd->root.mnt = NULL;
526 spin_lock(&dentry->d_lock);
527 if (!__d_rcu_to_refcount(dentry, nd->seq))
528 goto err_unlock;
529 BUG_ON(nd->inode != dentry->d_inode);
530 spin_unlock(&dentry->d_lock);
531
532 mntget(nd->path.mnt);
533
534 rcu_read_unlock();
535 br_read_unlock(vfsmount_lock);
536
537 return 0;
538
539err_unlock:
540 spin_unlock(&dentry->d_lock);
541 rcu_read_unlock();
542 br_read_unlock(vfsmount_lock);
543 return -ECHILD;
544}
545
546/* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
547static inline int nameidata_drop_rcu_last_maybe(struct nameidata *nd)
548{
549 if (likely(nd->flags & LOOKUP_RCU))
550 return nameidata_drop_rcu_last(nd);
551 return 0;
552}
553
834f2a4a
TM		 554/**
555 * release_open_intent - free up open intent resources
556 * @nd: pointer to nameidata
557 */
558void release_open_intent(struct nameidata *nd)
559{
0f7fc9e4 560 if (nd->intent.open.file->f_path.dentry == NULL)
834f2a4a
TM		 561 put_filp(nd->intent.open.file);
562 else
563 fput(nd->intent.open.file);
564}
565
34286d66
NP		 566static int d_revalidate(struct dentry *dentry, struct nameidata *nd)
567{
568 int status;
569
570 status = dentry->d_op->d_revalidate(dentry, nd);
571 if (status == -ECHILD) {
572 if (nameidata_dentry_drop_rcu(nd, dentry))
573 return status;
574 status = dentry->d_op->d_revalidate(dentry, nd);
575 }
576
577 return status;
578}
579
bcdc5e01
IK		 580static inline struct dentry *
581do_revalidate(struct dentry *dentry, struct nameidata *nd)
582{
34286d66
NP		 583 int status;
584
585 status = d_revalidate(dentry, nd);
bcdc5e01
IK		 586 if (unlikely(status <= 0)) {
587 /*
588 * The dentry failed validation.
589 * If d_revalidate returned 0 attempt to invalidate
590 * the dentry otherwise d_revalidate is asking us
591 * to return a fail status.
592 */
34286d66
NP		 593 if (status < 0) {
594 /* If we're in rcu-walk, we don't have a ref */
595 if (!(nd->flags & LOOKUP_RCU))
596 dput(dentry);
597 dentry = ERR_PTR(status);
598
599 } else {
600 /* Don't d_invalidate in rcu-walk mode */
601 if (nameidata_dentry_drop_rcu_maybe(nd, dentry))
602 return ERR_PTR(-ECHILD);
bcdc5e01
IK		 603 if (!d_invalidate(dentry)) {
604 dput(dentry);
605 dentry = NULL;
606 }
bcdc5e01
IK		 607 }
608 }
609 return dentry;
610}
611
fb045adb
NP		 612static inline int need_reval_dot(struct dentry *dentry)
613{
614 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
615 return 0;
616
617 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
618 return 0;
619
620 return 1;
621}
622
39159de2
JL		 623/*
624 * force_reval_path - force revalidation of a dentry
625 *
626 * In some situations the path walking code will trust dentries without
627 * revalidating them. This causes problems for filesystems that depend on
628 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
629 * (which indicates that it's possible for the dentry to go stale), force
630 * a d_revalidate call before proceeding.
631 *
632 * Returns 0 if the revalidation was successful. If the revalidation fails,
633 * either return the error returned by d_revalidate or -ESTALE if the
634 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
635 * invalidate the dentry. It's up to the caller to handle putting references
636 * to the path if necessary.
637 */
638static int
639force_reval_path(struct path *path, struct nameidata *nd)
640{
641 int status;
642 struct dentry *dentry = path->dentry;
643
644 /*
645 * only check on filesystems where it's possible for the dentry to
fb045adb 646 * become stale.
39159de2 647 */
fb045adb 648 if (!need_reval_dot(dentry))
39159de2
JL		 649 return 0;
650
34286d66 651 status = d_revalidate(dentry, nd);
39159de2
JL		 652 if (status > 0)
653 return 0;
654
655 if (!status) {
656 d_invalidate(dentry);
657 status = -ESTALE;
658 }
659 return status;
660}
661
1da177e4 662/*
b75b5086
AV		 663 * Short-cut version of permission(), for calling on directories
664 * during pathname resolution. Combines parts of permission()
665 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
1da177e4
LT		 666 *
667 * If appropriate, check DAC only. If not appropriate, or
b75b5086 668 * short-cut DAC fails, then call ->permission() to do more
1da177e4
LT		 669 * complete permission check.
670 */
31e6b01f 671static inline int __exec_permission(struct inode *inode, int rcu)
1da177e4 672{
5909ccaa 673 int ret;
1da177e4 674
cb9179ea 675 if (inode->i_op->permission) {
31e6b01f
NP		 676 if (rcu)
677 return -ECHILD;
5909ccaa 678 ret = inode->i_op->permission(inode, MAY_EXEC);
cb9179ea
LT		 679 if (!ret)
680 goto ok;
681 return ret;
682 }
31e6b01f 683 ret = __acl_permission_check(inode, MAY_EXEC, inode->i_op->check_acl, rcu);
5909ccaa 684 if (!ret)
1da177e4 685 goto ok;
31e6b01f
NP		 686 if (rcu && ret == -ECHILD)
687 return ret;
1da177e4 688
f1ac9f6b 689 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
1da177e4
LT		 690 goto ok;
691
5909ccaa 692 return ret;
1da177e4 693ok:
31e6b01f
NP		 694 return security_inode_exec_permission(inode, rcu);
695}
696
697static int exec_permission(struct inode *inode)
698{
699 return __exec_permission(inode, 0);
700}
701
702static int exec_permission_rcu(struct inode *inode)
703{
704 return __exec_permission(inode, 1);
1da177e4
LT		 705}
706
2a737871
AV		 707static __always_inline void set_root(struct nameidata *nd)
708{
f7ad3c6b
MS		 709 if (!nd->root.mnt)
710 get_fs_root(current->fs, &nd->root);
2a737871
AV		 711}
712
6de88d72
AV		 713static int link_path_walk(const char *, struct nameidata *);
714
31e6b01f
NP		 715static __always_inline void set_root_rcu(struct nameidata *nd)
716{
717 if (!nd->root.mnt) {
718 struct fs_struct *fs = current->fs;
c28cc364
NP		 719 unsigned seq;
720
721 do {
722 seq = read_seqcount_begin(&fs->seq);
723 nd->root = fs->root;
724 } while (read_seqcount_retry(&fs->seq, seq));
31e6b01f
NP		 725 }
726}
727
f1662356 728static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
1da177e4 729{
31e6b01f
NP		 730 int ret;
731
1da177e4
LT		 732 if (IS_ERR(link))
733 goto fail;
734
735 if (*link == '/') {
2a737871 736 set_root(nd);
1d957f9b 737 path_put(&nd->path);
2a737871
AV		 738 nd->path = nd->root;
739 path_get(&nd->root);
1da177e4 740 }
31e6b01f 741 nd->inode = nd->path.dentry->d_inode;
b4091d5f 742
31e6b01f
NP		 743 ret = link_path_walk(link, nd);
744 return ret;
1da177e4 745fail:
1d957f9b 746 path_put(&nd->path);
1da177e4
LT		 747 return PTR_ERR(link);
748}
749
1d957f9b 750static void path_put_conditional(struct path *path, struct nameidata *nd)
051d3812
IK		 751{
752 dput(path->dentry);
4ac91378 753 if (path->mnt != nd->path.mnt)
051d3812
IK		 754 mntput(path->mnt);
755}
756
757static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
758{
31e6b01f
NP		 759 if (!(nd->flags & LOOKUP_RCU)) {
760 dput(nd->path.dentry);
761 if (nd->path.mnt != path->mnt)
762 mntput(nd->path.mnt);
9a229683 763 }
31e6b01f 764 nd->path.mnt = path->mnt;
4ac91378 765 nd->path.dentry = path->dentry;
051d3812
IK		 766}
767
def4af30
AV		 768static __always_inline int
769__do_follow_link(struct path *path, struct nameidata *nd, void **p)
1da177e4
LT		 770{
771 int error;
cd4e91d3 772 struct dentry *dentry = path->dentry;
1da177e4 773
d671a1cb 774 touch_atime(path->mnt, dentry);
1da177e4 775 nd_set_link(nd, NULL);
cd4e91d3 776
4ac91378 777 if (path->mnt != nd->path.mnt) {
051d3812 778 path_to_nameidata(path, nd);
31e6b01f 779 nd->inode = nd->path.dentry->d_inode;
051d3812
IK		 780 dget(dentry);
781 }
782 mntget(path->mnt);
31e6b01f 783
86acdca1 784 nd->last_type = LAST_BIND;
def4af30
AV		 785 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
786 error = PTR_ERR(*p);
787 if (!IS_ERR(*p)) {
1da177e4 788 char *s = nd_get_link(nd);
cc314eef 789 error = 0;
1da177e4
LT		 790 if (s)
791 error = __vfs_follow_link(nd, s);
39159de2
JL		 792 else if (nd->last_type == LAST_BIND) {
793 error = force_reval_path(&nd->path, nd);
794 if (error)
795 path_put(&nd->path);
796 }
1da177e4 797 }
1da177e4
LT		 798 return error;
799}
800
801/*
802 * This limits recursive symlink follows to 8, while
803 * limiting consecutive symlinks to 40.
804 *
805 * Without that kind of total limit, nasty chains of consecutive
806 * symlinks can cause almost arbitrarily long lookups.
807 */
90ebe565 808static inline int do_follow_link(struct path *path, struct nameidata *nd)
1da177e4 809{
def4af30 810 void *cookie;
1da177e4
LT		 811 int err = -ELOOP;
812 if (current->link_count >= MAX_NESTED_LINKS)
813 goto loop;
814 if (current->total_link_count >= 40)
815 goto loop;
816 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
817 cond_resched();
90ebe565 818 err = security_inode_follow_link(path->dentry, nd);
1da177e4
LT		 819 if (err)
820 goto loop;
821 current->link_count++;
822 current->total_link_count++;
823 nd->depth++;
def4af30
AV		 824 err = __do_follow_link(path, nd, &cookie);
825 if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link)
826 path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie);
258fa999 827 path_put(path);
839d9f93
AV		 828 current->link_count--;
829 nd->depth--;
1da177e4
LT		 830 return err;
831loop:
1d957f9b
JB		 832 path_put_conditional(path, nd);
833 path_put(&nd->path);
1da177e4
LT		 834 return err;
835}
836
31e6b01f
NP		 837static int follow_up_rcu(struct path *path)
838{
839 struct vfsmount *parent;
840 struct dentry *mountpoint;
841
842 parent = path->mnt->mnt_parent;
843 if (parent == path->mnt)
844 return 0;
845 mountpoint = path->mnt->mnt_mountpoint;
846 path->dentry = mountpoint;
847 path->mnt = parent;
848 return 1;
849}
850
bab77ebf 851int follow_up(struct path *path)
1da177e4
LT		 852{
853 struct vfsmount *parent;
854 struct dentry *mountpoint;
99b7db7b
NP		 855
856 br_read_lock(vfsmount_lock);
bab77ebf
AV		 857 parent = path->mnt->mnt_parent;
858 if (parent == path->mnt) {
99b7db7b 859 br_read_unlock(vfsmount_lock);
1da177e4
LT		 860 return 0;
861 }
862 mntget(parent);
bab77ebf 863 mountpoint = dget(path->mnt->mnt_mountpoint);
99b7db7b 864 br_read_unlock(vfsmount_lock);
bab77ebf
AV		 865 dput(path->dentry);
866 path->dentry = mountpoint;
867 mntput(path->mnt);
868 path->mnt = parent;
1da177e4
LT		 869 return 1;
870}
871
b5c84bf6
NP		 872/*
873 * serialization is taken care of in namespace.c
1da177e4 874 */
31e6b01f
NP		 875static void __follow_mount_rcu(struct nameidata *nd, struct path *path,
876 struct inode **inode)
877{
878 while (d_mountpoint(path->dentry)) {
879 struct vfsmount *mounted;
880 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
881 if (!mounted)
882 return;
883 path->mnt = mounted;
884 path->dentry = mounted->mnt_root;
885 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
886 *inode = path->dentry->d_inode;
887 }
888}
889
463ffb2e
AV		 890static int __follow_mount(struct path *path)
891{
892 int res = 0;
893 while (d_mountpoint(path->dentry)) {
1c755af4 894 struct vfsmount *mounted = lookup_mnt(path);
463ffb2e
AV		 895 if (!mounted)
896 break;
897 dput(path->dentry);
898 if (res)
899 mntput(path->mnt);
900 path->mnt = mounted;
901 path->dentry = dget(mounted->mnt_root);
902 res = 1;
903 }
904 return res;
905}
906
79ed0226 907static void follow_mount(struct path *path)
1da177e4 908{
79ed0226 909 while (d_mountpoint(path->dentry)) {
1c755af4 910 struct vfsmount *mounted = lookup_mnt(path);
1da177e4
LT		 911 if (!mounted)
912 break;
79ed0226
AV		 913 dput(path->dentry);
914 mntput(path->mnt);
915 path->mnt = mounted;
916 path->dentry = dget(mounted->mnt_root);
1da177e4 917 }
1da177e4
LT		 918}
919
9393bd07 920int follow_down(struct path *path)
1da177e4
LT		 921{
922 struct vfsmount *mounted;
923
1c755af4 924 mounted = lookup_mnt(path);
1da177e4 925 if (mounted) {
9393bd07
AV		 926 dput(path->dentry);
927 mntput(path->mnt);
928 path->mnt = mounted;
929 path->dentry = dget(mounted->mnt_root);
1da177e4
LT		 930 return 1;
931 }
932 return 0;
933}
934
31e6b01f
NP		 935static int follow_dotdot_rcu(struct nameidata *nd)
936{
937 struct inode *inode = nd->inode;
938
939 set_root_rcu(nd);
940
941 while(1) {
942 if (nd->path.dentry == nd->root.dentry &&
943 nd->path.mnt == nd->root.mnt) {
944 break;
945 }
946 if (nd->path.dentry != nd->path.mnt->mnt_root) {
947 struct dentry *old = nd->path.dentry;
948 struct dentry *parent = old->d_parent;
949 unsigned seq;
950
951 seq = read_seqcount_begin(&parent->d_seq);
952 if (read_seqcount_retry(&old->d_seq, nd->seq))
953 return -ECHILD;
954 inode = parent->d_inode;
955 nd->path.dentry = parent;
956 nd->seq = seq;
957 break;
958 }
959 if (!follow_up_rcu(&nd->path))
960 break;
961 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
962 inode = nd->path.dentry->d_inode;
963 }
964 __follow_mount_rcu(nd, &nd->path, &inode);
965 nd->inode = inode;
966
967 return 0;
968}
969
970static void follow_dotdot(struct nameidata *nd)
1da177e4 971{
2a737871 972 set_root(nd);
e518ddb7 973
1da177e4 974 while(1) {
4ac91378 975 struct dentry *old = nd->path.dentry;
1da177e4 976
2a737871
AV		 977 if (nd->path.dentry == nd->root.dentry &&
978 nd->path.mnt == nd->root.mnt) {
1da177e4
LT		 979 break;
980 }
4ac91378 981 if (nd->path.dentry != nd->path.mnt->mnt_root) {
3088dd70
AV		 982 /* rare case of legitimate dget_parent()... */
983 nd->path.dentry = dget_parent(nd->path.dentry);
1da177e4
LT		 984 dput(old);
985 break;
986 }
3088dd70 987 if (!follow_up(&nd->path))
1da177e4 988 break;
1da177e4 989 }
79ed0226 990 follow_mount(&nd->path);
31e6b01f 991 nd->inode = nd->path.dentry->d_inode;
1da177e4
LT		 992}
993
baa03890
NP		 994/*
995 * Allocate a dentry with name and parent, and perform a parent
996 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
997 * on error. parent->d_inode->i_mutex must be held. d_lookup must
998 * have verified that no child exists while under i_mutex.
999 */
1000static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1001 struct qstr *name, struct nameidata *nd)
1002{
1003 struct inode *inode = parent->d_inode;
1004 struct dentry *dentry;
1005 struct dentry *old;
1006
1007 /* Don't create child dentry for a dead directory. */
1008 if (unlikely(IS_DEADDIR(inode)))
1009 return ERR_PTR(-ENOENT);
1010
1011 dentry = d_alloc(parent, name);
1012 if (unlikely(!dentry))
1013 return ERR_PTR(-ENOMEM);
1014
1015 old = inode->i_op->lookup(inode, dentry, nd);
1016 if (unlikely(old)) {
1017 dput(dentry);
1018 dentry = old;
1019 }
1020 return dentry;
1021}
1022
1da177e4
LT		 1023/*
1024 * It's more convoluted than I'd like it to be, but... it's still fairly
1025 * small and for now I'd prefer to have fast path as straight as possible.
1026 * It _is_ time-critical.
1027 */
1028static int do_lookup(struct nameidata *nd, struct qstr *name,
31e6b01f 1029 struct path *path, struct inode **inode)
1da177e4 1030{
4ac91378 1031 struct vfsmount *mnt = nd->path.mnt;
31e6b01f 1032 struct dentry *dentry, *parent = nd->path.dentry;
6e6b1bd1 1033 struct inode *dir;
3cac260a
AV		 1034 /*
1035 * See if the low-level filesystem might want
1036 * to use its own hash..
1037 */
fb045adb 1038 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
31e6b01f 1039 int err = parent->d_op->d_hash(parent, nd->inode, name);
3cac260a
AV		 1040 if (err < 0)
1041 return err;
1042 }
1da177e4 1043
b04f784e
NP		 1044 /*
1045 * Rename seqlock is not required here because in the off chance
1046 * of a false negative due to a concurrent rename, we're going to
1047 * do the non-racy lookup, below.
1048 */
31e6b01f
NP		 1049 if (nd->flags & LOOKUP_RCU) {
1050 unsigned seq;
1051
1052 *inode = nd->inode;
1053 dentry = __d_lookup_rcu(parent, name, &seq, inode);
1054 if (!dentry) {
1055 if (nameidata_drop_rcu(nd))
1056 return -ECHILD;
1057 goto need_lookup;
1058 }
1059 /* Memory barrier in read_seqcount_begin of child is enough */
1060 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1061 return -ECHILD;
1062
1063 nd->seq = seq;
34286d66 1064 if (dentry->d_flags & DCACHE_OP_REVALIDATE)
31e6b01f 1065 goto need_revalidate;
31e6b01f
NP		 1066 path->mnt = mnt;
1067 path->dentry = dentry;
1068 __follow_mount_rcu(nd, path, inode);
1069 } else {
1070 dentry = __d_lookup(parent, name);
1071 if (!dentry)
1072 goto need_lookup;
2e2e88ea 1073found:
fb045adb 1074 if (dentry->d_flags & DCACHE_OP_REVALIDATE)
31e6b01f 1075 goto need_revalidate;
1da177e4 1076done:
31e6b01f
NP		 1077 path->mnt = mnt;
1078 path->dentry = dentry;
1079 __follow_mount(path);
1080 *inode = path->dentry->d_inode;
1081 }
1da177e4
LT		 1082 return 0;
1083
1084need_lookup:
6e6b1bd1 1085 dir = parent->d_inode;
31e6b01f 1086 BUG_ON(nd->inode != dir);
6e6b1bd1
AV		 1087
1088 mutex_lock(&dir->i_mutex);
1089 /*
1090 * First re-do the cached lookup just in case it was created
b04f784e
NP		 1091 * while we waited for the directory semaphore, or the first
1092 * lookup failed due to an unrelated rename.
6e6b1bd1 1093 *
b04f784e
NP		 1094 * This could use version numbering or similar to avoid unnecessary
1095 * cache lookups, but then we'd have to do the first lookup in the
1096 * non-racy way. However in the common case here, everything should
1097 * be hot in cache, so would it be a big win?
6e6b1bd1
AV		 1098 */
1099 dentry = d_lookup(parent, name);
baa03890
NP		 1100 if (likely(!dentry)) {
1101 dentry = d_alloc_and_lookup(parent, name, nd);
6e6b1bd1
AV		 1102 mutex_unlock(&dir->i_mutex);
1103 if (IS_ERR(dentry))
1104 goto fail;
1105 goto done;
1106 }
6e6b1bd1
AV		 1107 /*
1108 * Uhhuh! Nasty case: the cache was re-populated while
1109 * we waited on the semaphore. Need to revalidate.
1110 */
1111 mutex_unlock(&dir->i_mutex);
2e2e88ea 1112 goto found;
1da177e4
LT		 1113
1114need_revalidate:
bcdc5e01
IK		 1115 dentry = do_revalidate(dentry, nd);
1116 if (!dentry)
1117 goto need_lookup;
1118 if (IS_ERR(dentry))
1119 goto fail;
1120 goto done;
1da177e4
LT		 1121
1122fail:
1123 return PTR_ERR(dentry);
1124}
1125
ac278a9c
AV		 1126/*
1127 * This is a temporary kludge to deal with "automount" symlinks; proper
1128 * solution is to trigger them on follow_mount(), so that do_lookup()
1129 * would DTRT. To be killed before 2.6.34-final.
1130 */
1131static inline int follow_on_final(struct inode *inode, unsigned lookup_flags)
1132{
1133 return inode && unlikely(inode->i_op->follow_link) &&
1134 ((lookup_flags & LOOKUP_FOLLOW) || S_ISDIR(inode->i_mode));
1135}
1136
1da177e4
LT		 1137/*
1138 * Name resolution.
ea3834d9
PM		 1139 * This is the basic name resolution function, turning a pathname into
1140 * the final dentry. We expect 'base' to be positive and a directory.
1da177e4 1141 *
ea3834d9
PM		 1142 * Returns 0 and nd will have valid dentry and mnt on success.
1143 * Returns error and drops reference to input namei data on failure.
1da177e4 1144 */
6de88d72 1145static int link_path_walk(const char *name, struct nameidata *nd)
1da177e4
LT		 1146{
1147 struct path next;
1da177e4
LT		 1148 int err;
1149 unsigned int lookup_flags = nd->flags;
1150
1151 while (*name=='/')
1152 name++;
1153 if (!*name)
1154 goto return_reval;
1155
1da177e4 1156 if (nd->depth)
f55eab82 1157 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
1da177e4
LT		 1158
1159 /* At this point we know we have a real path component. */
1160 for(;;) {
31e6b01f 1161 struct inode *inode;
1da177e4
LT		 1162 unsigned long hash;
1163 struct qstr this;
1164 unsigned int c;
1165
cdce5d6b 1166 nd->flags |= LOOKUP_CONTINUE;
31e6b01f
NP		 1167 if (nd->flags & LOOKUP_RCU) {
1168 err = exec_permission_rcu(nd->inode);
1169 if (err == -ECHILD) {
1170 if (nameidata_drop_rcu(nd))
1171 return -ECHILD;
1172 goto exec_again;
1173 }
1174 } else {
1175exec_again:
1176 err = exec_permission(nd->inode);
1177 }
1da177e4
LT		 1178 if (err)
1179 break;
1180
1181 this.name = name;
1182 c = *(const unsigned char *)name;
1183
1184 hash = init_name_hash();
1185 do {
1186 name++;
1187 hash = partial_name_hash(c, hash);
1188 c = *(const unsigned char *)name;
1189 } while (c && (c != '/'));
1190 this.len = name - (const char *) this.name;
1191 this.hash = end_name_hash(hash);
1192
1193 /* remove trailing slashes? */
1194 if (!c)
1195 goto last_component;
1196 while (*++name == '/');
1197 if (!*name)
1198 goto last_with_slashes;
1199
1200 /*
1201 * "." and ".." are special - ".." especially so because it has
1202 * to be able to know about the current root directory and
1203 * parent relationships.
1204 */
1205 if (this.name[0] == '.') switch (this.len) {
1206 default:
1207 break;
31e6b01f 1208 case 2:
1da177e4
LT		 1209 if (this.name[1] != '.')
1210 break;
31e6b01f
NP		 1211 if (nd->flags & LOOKUP_RCU) {
1212 if (follow_dotdot_rcu(nd))
1213 return -ECHILD;
1214 } else
1215 follow_dotdot(nd);
1da177e4
LT		 1216 /* fallthrough */
1217 case 1:
1218 continue;
1219 }
1da177e4 1220 /* This does the actual lookups.. */
31e6b01f 1221 err = do_lookup(nd, &this, &next, &inode);
1da177e4
LT		 1222 if (err)
1223 break;
1da177e4 1224 err = -ENOENT;
1da177e4
LT		 1225 if (!inode)
1226 goto out_dput;
1da177e4
LT		 1227
1228 if (inode->i_op->follow_link) {
31e6b01f
NP		 1229 /* We commonly drop rcu-walk here */
1230 if (nameidata_dentry_drop_rcu_maybe(nd, next.dentry))
1231 return -ECHILD;
1232 BUG_ON(inode != next.dentry->d_inode);
90ebe565 1233 err = do_follow_link(&next, nd);
1da177e4
LT		 1234 if (err)
1235 goto return_err;
31e6b01f 1236 nd->inode = nd->path.dentry->d_inode;
1da177e4 1237 err = -ENOENT;
31e6b01f 1238 if (!nd->inode)
1da177e4 1239 break;
31e6b01f 1240 } else {
09dd17d3 1241 path_to_nameidata(&next, nd);
31e6b01f
NP		 1242 nd->inode = inode;
1243 }
1da177e4 1244 err = -ENOTDIR;
31e6b01f 1245 if (!nd->inode->i_op->lookup)
1da177e4
LT		 1246 break;
1247 continue;
1248 /* here ends the main loop */
1249
1250last_with_slashes:
1251 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1252last_component:
f55eab82
TM		 1253 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1254 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
1da177e4
LT		 1255 if (lookup_flags & LOOKUP_PARENT)
1256 goto lookup_parent;
1257 if (this.name[0] == '.') switch (this.len) {
1258 default:
1259 break;
31e6b01f 1260 case 2:
1da177e4
LT		 1261 if (this.name[1] != '.')
1262 break;
31e6b01f
NP		 1263 if (nd->flags & LOOKUP_RCU) {
1264 if (follow_dotdot_rcu(nd))
1265 return -ECHILD;
1266 } else
1267 follow_dotdot(nd);
1da177e4
LT		 1268 /* fallthrough */
1269 case 1:
1270 goto return_reval;
1271 }
31e6b01f 1272 err = do_lookup(nd, &this, &next, &inode);
1da177e4
LT		 1273 if (err)
1274 break;
ac278a9c 1275 if (follow_on_final(inode, lookup_flags)) {
31e6b01f
NP		 1276 if (nameidata_dentry_drop_rcu_maybe(nd, next.dentry))
1277 return -ECHILD;
1278 BUG_ON(inode != next.dentry->d_inode);
90ebe565 1279 err = do_follow_link(&next, nd);
1da177e4
LT		 1280 if (err)
1281 goto return_err;
31e6b01f
NP		 1282 nd->inode = nd->path.dentry->d_inode;
1283 } else {
09dd17d3 1284 path_to_nameidata(&next, nd);
31e6b01f
NP		 1285 nd->inode = inode;
1286 }
1da177e4 1287 err = -ENOENT;
31e6b01f 1288 if (!nd->inode)
1da177e4
LT		 1289 break;
1290 if (lookup_flags & LOOKUP_DIRECTORY) {
1291 err = -ENOTDIR;
31e6b01f 1292 if (!nd->inode->i_op->lookup)
1da177e4
LT		 1293 break;
1294 }
1295 goto return_base;
1296lookup_parent:
1297 nd->last = this;
1298 nd->last_type = LAST_NORM;
1299 if (this.name[0] != '.')
1300 goto return_base;
1301 if (this.len == 1)
1302 nd->last_type = LAST_DOT;
1303 else if (this.len == 2 && this.name[1] == '.')
1304 nd->last_type = LAST_DOTDOT;
1305 else
1306 goto return_base;
1307return_reval:
1308 /*
1309 * We bypassed the ordinary revalidation routines.
1310 * We may need to check the cached dentry for staleness.
1311 */
fb045adb 1312 if (need_reval_dot(nd->path.dentry)) {
1da177e4 1313 /* Note: we do not d_invalidate() */
34286d66
NP		 1314 err = d_revalidate(nd->path.dentry, nd);
1315 if (!err)
1316 err = -ESTALE;
1317 if (err < 0)
1da177e4
LT		 1318 break;
1319 }
1320return_base:
31e6b01f
NP		 1321 if (nameidata_drop_rcu_last_maybe(nd))
1322 return -ECHILD;
1da177e4
LT		 1323 return 0;
1324out_dput:
31e6b01f
NP		 1325 if (!(nd->flags & LOOKUP_RCU))
1326 path_put_conditional(&next, nd);
1da177e4
LT		 1327 break;
1328 }
31e6b01f
NP		 1329 if (!(nd->flags & LOOKUP_RCU))
1330 path_put(&nd->path);
1da177e4
LT		 1331return_err:
1332 return err;
1333}
1334
31e6b01f
NP		 1335static inline int path_walk_rcu(const char *name, struct nameidata *nd)
1336{
1337 current->total_link_count = 0;
1338
1339 return link_path_walk(name, nd);
1340}
1341
1342static inline int path_walk_simple(const char *name, struct nameidata *nd)
1343{
1344 current->total_link_count = 0;
1345
1346 return link_path_walk(name, nd);
1347}
1348
fc9b52cd 1349static int path_walk(const char *name, struct nameidata *nd)
1da177e4 1350{
6de88d72
AV		 1351 struct path save = nd->path;
1352 int result;
1353
1da177e4 1354 current->total_link_count = 0;
6de88d72
AV		 1355
1356 /* make sure the stuff we saved doesn't go away */
1357 path_get(&save);
1358
1359 result = link_path_walk(name, nd);
1360 if (result == -ESTALE) {
1361 /* nd->path had been dropped */
1362 current->total_link_count = 0;
1363 nd->path = save;
1364 path_get(&nd->path);
1365 nd->flags |= LOOKUP_REVAL;
1366 result = link_path_walk(name, nd);
1367 }
1368
1369 path_put(&save);
1370
1371 return result;
1da177e4
LT		 1372}
1373
31e6b01f
NP		 1374static void path_finish_rcu(struct nameidata *nd)
1375{
1376 if (nd->flags & LOOKUP_RCU) {
1377 /* RCU dangling. Cancel it. */
1378 nd->flags &= ~LOOKUP_RCU;
1379 nd->root.mnt = NULL;
1380 rcu_read_unlock();
1381 br_read_unlock(vfsmount_lock);
1382 }
1383 if (nd->file)
1384 fput(nd->file);
1385}
1386
1387static int path_init_rcu(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1388{
1389 int retval = 0;
1390 int fput_needed;
1391 struct file *file;
1392
1393 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1394 nd->flags = flags | LOOKUP_RCU;
1395 nd->depth = 0;
1396 nd->root.mnt = NULL;
1397 nd->file = NULL;
1398
1399 if (*name=='/') {
1400 struct fs_struct *fs = current->fs;
c28cc364 1401 unsigned seq;
31e6b01f
NP		 1402
1403 br_read_lock(vfsmount_lock);
1404 rcu_read_lock();
1405
c28cc364
NP		 1406 do {
1407 seq = read_seqcount_begin(&fs->seq);
1408 nd->root = fs->root;
1409 nd->path = nd->root;
1410 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1411 } while (read_seqcount_retry(&fs->seq, seq));
31e6b01f
NP		 1412
1413 } else if (dfd == AT_FDCWD) {
1414 struct fs_struct *fs = current->fs;
c28cc364 1415 unsigned seq;
31e6b01f
NP		 1416
1417 br_read_lock(vfsmount_lock);
1418 rcu_read_lock();
1419
c28cc364
NP		 1420 do {
1421 seq = read_seqcount_begin(&fs->seq);
1422 nd->path = fs->pwd;
1423 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1424 } while (read_seqcount_retry(&fs->seq, seq));
1425
31e6b01f
NP		 1426 } else {
1427 struct dentry *dentry;
1428
1429 file = fget_light(dfd, &fput_needed);
1430 retval = -EBADF;
1431 if (!file)
1432 goto out_fail;
1433
1434 dentry = file->f_path.dentry;
1435
1436 retval = -ENOTDIR;
1437 if (!S_ISDIR(dentry->d_inode->i_mode))
1438 goto fput_fail;
1439
1440 retval = file_permission(file, MAY_EXEC);
1441 if (retval)
1442 goto fput_fail;
1443
1444 nd->path = file->f_path;
1445 if (fput_needed)
1446 nd->file = file;
1447
c28cc364 1448 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
31e6b01f
NP		 1449 br_read_lock(vfsmount_lock);
1450 rcu_read_lock();
1451 }
1452 nd->inode = nd->path.dentry->d_inode;
1453 return 0;
1454
1455fput_fail:
1456 fput_light(file, fput_needed);
1457out_fail:
1458 return retval;
1459}
1460
9b4a9b14 1461static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1da177e4 1462{
ea3834d9 1463 int retval = 0;
170aa3d0
UD		 1464 int fput_needed;
1465 struct file *file;
1da177e4
LT		 1466
1467 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1468 nd->flags = flags;
1469 nd->depth = 0;
2a737871 1470 nd->root.mnt = NULL;
1da177e4 1471
1da177e4 1472 if (*name=='/') {
2a737871
AV		 1473 set_root(nd);
1474 nd->path = nd->root;
1475 path_get(&nd->root);
5590ff0d 1476 } else if (dfd == AT_FDCWD) {
f7ad3c6b 1477 get_fs_pwd(current->fs, &nd->path);
5590ff0d 1478 } else {
5590ff0d
UD		 1479 struct dentry *dentry;
1480
1481 file = fget_light(dfd, &fput_needed);
170aa3d0
UD		 1482 retval = -EBADF;
1483 if (!file)
6d09bb62 1484 goto out_fail;
5590ff0d 1485
0f7fc9e4 1486 dentry = file->f_path.dentry;
5590ff0d 1487
170aa3d0
UD		 1488 retval = -ENOTDIR;
1489 if (!S_ISDIR(dentry->d_inode->i_mode))
6d09bb62 1490 goto fput_fail;
5590ff0d
UD		 1491
1492 retval = file_permission(file, MAY_EXEC);
170aa3d0 1493 if (retval)
6d09bb62 1494 goto fput_fail;
5590ff0d 1495
5dd784d0
JB		 1496 nd->path = file->f_path;
1497 path_get(&file->f_path);
5590ff0d
UD		 1498
1499 fput_light(file, fput_needed);
1da177e4 1500 }
31e6b01f 1501 nd->inode = nd->path.dentry->d_inode;
9b4a9b14 1502 return 0;
2dfdd266 1503
9b4a9b14
AV		 1504fput_fail:
1505 fput_light(file, fput_needed);
1506out_fail:
1507 return retval;
1508}
1509
1510/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1511static int do_path_lookup(int dfd, const char *name,
1512 unsigned int flags, struct nameidata *nd)
1513{
31e6b01f
NP		 1514 int retval;
1515
1516 /*
1517 * Path walking is largely split up into 2 different synchronisation
1518 * schemes, rcu-walk and ref-walk (explained in
1519 * Documentation/filesystems/path-lookup.txt). These share much of the
1520 * path walk code, but some things particularly setup, cleanup, and
1521 * following mounts are sufficiently divergent that functions are
1522 * duplicated. Typically there is a function foo(), and its RCU
1523 * analogue, foo_rcu().
1524 *
1525 * -ECHILD is the error number of choice (just to avoid clashes) that
1526 * is returned if some aspect of an rcu-walk fails. Such an error must
1527 * be handled by restarting a traditional ref-walk (which will always
1528 * be able to complete).
1529 */
1530 retval = path_init_rcu(dfd, name, flags, nd);
1531 if (unlikely(retval))
1532 return retval;
1533 retval = path_walk_rcu(name, nd);
1534 path_finish_rcu(nd);
2a737871
AV		 1535 if (nd->root.mnt) {
1536 path_put(&nd->root);
1537 nd->root.mnt = NULL;
1538 }
31e6b01f
NP		 1539
1540 if (unlikely(retval == -ECHILD || retval == -ESTALE)) {
1541 /* slower, locked walk */
1542 if (retval == -ESTALE)
1543 flags |= LOOKUP_REVAL;
1544 retval = path_init(dfd, name, flags, nd);
1545 if (unlikely(retval))
1546 return retval;
1547 retval = path_walk(name, nd);
1548 if (nd->root.mnt) {
1549 path_put(&nd->root);
1550 nd->root.mnt = NULL;
1551 }
1552 }
1553
1554 if (likely(!retval)) {
1555 if (unlikely(!audit_dummy_context())) {
1556 if (nd->path.dentry && nd->inode)
1557 audit_inode(name, nd->path.dentry);
1558 }
1559 }
1560
170aa3d0 1561 return retval;
1da177e4
LT		 1562}
1563
fc9b52cd 1564int path_lookup(const char *name, unsigned int flags,
5590ff0d
UD		 1565 struct nameidata *nd)
1566{
1567 return do_path_lookup(AT_FDCWD, name, flags, nd);
1568}
1569
d1811465
AV		 1570int kern_path(const char *name, unsigned int flags, struct path *path)
1571{
1572 struct nameidata nd;
1573 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1574 if (!res)
1575 *path = nd.path;
1576 return res;
1577}
1578
16f18200
JJS		 1579/**
1580 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1581 * @dentry: pointer to dentry of the base directory
1582 * @mnt: pointer to vfs mount of the base directory
1583 * @name: pointer to file name
1584 * @flags: lookup flags
1585 * @nd: pointer to nameidata
1586 */
1587int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1588 const char *name, unsigned int flags,
1589 struct nameidata *nd)
1590{
1591 int retval;
1592
1593 /* same as do_path_lookup */
1594 nd->last_type = LAST_ROOT;
1595 nd->flags = flags;
1596 nd->depth = 0;
1597
c8e7f449
JB		 1598 nd->path.dentry = dentry;
1599 nd->path.mnt = mnt;
1600 path_get(&nd->path);
5b857119
AV		 1601 nd->root = nd->path;
1602 path_get(&nd->root);
31e6b01f 1603 nd->inode = nd->path.dentry->d_inode;
16f18200
JJS		 1604
1605 retval = path_walk(name, nd);
4ac91378 1606 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
31e6b01f 1607 nd->inode))
4ac91378 1608 audit_inode(name, nd->path.dentry);
16f18200 1609
5b857119
AV		 1610 path_put(&nd->root);
1611 nd->root.mnt = NULL;
16f18200 1612
2a737871 1613 return retval;
16f18200
JJS		 1614}
1615
eead1911
CH		 1616static struct dentry *__lookup_hash(struct qstr *name,
1617 struct dentry *base, struct nameidata *nd)
1da177e4 1618{
81fca444 1619 struct inode *inode = base->d_inode;
057f6c01 1620 struct dentry *dentry;
1da177e4
LT		 1621 int err;
1622
81fca444
CH		 1623 err = exec_permission(inode);
1624 if (err)
1625 return ERR_PTR(err);
1da177e4
LT		 1626
1627 /*
1628 * See if the low-level filesystem might want
1629 * to use its own hash..
1630 */
fb045adb 1631 if (base->d_flags & DCACHE_OP_HASH) {
b1e6a015 1632 err = base->d_op->d_hash(base, inode, name);
1da177e4
LT		 1633 dentry = ERR_PTR(err);
1634 if (err < 0)
1635 goto out;
1636 }
1637
b04f784e
NP		 1638 /*
1639 * Don't bother with __d_lookup: callers are for creat as
1640 * well as unlink, so a lot of the time it would cost
1641 * a double lookup.
6e6b1bd1 1642 */
b04f784e 1643 dentry = d_lookup(base, name);
6e6b1bd1 1644
fb045adb 1645 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE))
6e6b1bd1
AV		 1646 dentry = do_revalidate(dentry, nd);
1647
baa03890
NP		 1648 if (!dentry)
1649 dentry = d_alloc_and_lookup(base, name, nd);
1da177e4
LT		 1650out:
1651 return dentry;
1652}
1653
057f6c01
JM		 1654/*
1655 * Restricted form of lookup. Doesn't follow links, single-component only,
1656 * needs parent already locked. Doesn't follow mounts.
1657 * SMP-safe.
1658 */
eead1911 1659static struct dentry *lookup_hash(struct nameidata *nd)
057f6c01 1660{
4ac91378 1661 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1da177e4
LT		 1662}
1663
eead1911
CH		 1664static int __lookup_one_len(const char *name, struct qstr *this,
1665 struct dentry *base, int len)
1da177e4
LT		 1666{
1667 unsigned long hash;
1da177e4
LT		 1668 unsigned int c;
1669
057f6c01
JM		 1670 this->name = name;
1671 this->len = len;
1da177e4 1672 if (!len)
057f6c01 1673 return -EACCES;
1da177e4
LT		 1674
1675 hash = init_name_hash();
1676 while (len--) {
1677 c = *(const unsigned char *)name++;
1678 if (c == '/' || c == '\0')
057f6c01 1679 return -EACCES;
1da177e4
LT		 1680 hash = partial_name_hash(c, hash);
1681 }
057f6c01
JM		 1682 this->hash = end_name_hash(hash);
1683 return 0;
1684}
1da177e4 1685
eead1911 1686/**
a6b91919 1687 * lookup_one_len - filesystem helper to lookup single pathname component
eead1911
CH		 1688 * @name: pathname component to lookup
1689 * @base: base directory to lookup from
1690 * @len: maximum length @len should be interpreted to
1691 *
a6b91919
RD		 1692 * Note that this routine is purely a helper for filesystem usage and should
1693 * not be called by generic code. Also note that by using this function the
eead1911
CH		 1694 * nameidata argument is passed to the filesystem methods and a filesystem
1695 * using this helper needs to be prepared for that.
1696 */
057f6c01
JM		 1697struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1698{
1699 int err;
1700 struct qstr this;
1701
2f9092e1
DW		 1702 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1703
057f6c01 1704 err = __lookup_one_len(name, &this, base, len);
eead1911
CH		 1705 if (err)
1706 return ERR_PTR(err);
1707
49705b77 1708 return __lookup_hash(&this, base, NULL);
057f6c01
JM		 1709}
1710
2d8f3038
AV		 1711int user_path_at(int dfd, const char __user *name, unsigned flags,
1712 struct path *path)
1da177e4 1713{
2d8f3038 1714 struct nameidata nd;
1da177e4
LT		 1715 char *tmp = getname(name);
1716 int err = PTR_ERR(tmp);
1da177e4 1717 if (!IS_ERR(tmp)) {
2d8f3038
AV		 1718
1719 BUG_ON(flags & LOOKUP_PARENT);
1720
1721 err = do_path_lookup(dfd, tmp, flags, &nd);
1da177e4 1722 putname(tmp);
2d8f3038
AV		 1723 if (!err)
1724 *path = nd.path;
1da177e4
LT		 1725 }
1726 return err;
1727}
1728
2ad94ae6
AV		 1729static int user_path_parent(int dfd, const char __user *path,
1730 struct nameidata *nd, char **name)
1731{
1732 char *s = getname(path);
1733 int error;
1734
1735 if (IS_ERR(s))
1736 return PTR_ERR(s);
1737
1738 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1739 if (error)
1740 putname(s);
1741 else
1742 *name = s;
1743
1744 return error;
1745}
1746
1da177e4
LT		 1747/*
1748 * It's inline, so penalty for filesystems that don't use sticky bit is
1749 * minimal.
1750 */
1751static inline int check_sticky(struct inode *dir, struct inode *inode)
1752{
da9592ed
DH		 1753 uid_t fsuid = current_fsuid();
1754
1da177e4
LT		 1755 if (!(dir->i_mode & S_ISVTX))
1756 return 0;
da9592ed 1757 if (inode->i_uid == fsuid)
1da177e4 1758 return 0;
da9592ed 1759 if (dir->i_uid == fsuid)
1da177e4
LT		 1760 return 0;
1761 return !capable(CAP_FOWNER);
1762}
1763
1764/*
1765 * Check whether we can remove a link victim from directory dir, check
1766 * whether the type of victim is right.
1767 * 1. We can't do it if dir is read-only (done in permission())
1768 * 2. We should have write and exec permissions on dir
1769 * 3. We can't remove anything from append-only dir
1770 * 4. We can't do anything with immutable dir (done in permission())
1771 * 5. If the sticky bit on dir is set we should either
1772 * a. be owner of dir, or
1773 * b. be owner of victim, or
1774 * c. have CAP_FOWNER capability
1775 * 6. If the victim is append-only or immutable we can't do antyhing with
1776 * links pointing to it.
1777 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1778 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1779 * 9. We can't remove a root or mountpoint.
1780 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1781 * nfs_async_unlink().
1782 */
858119e1 1783static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1da177e4
LT		 1784{
1785 int error;
1786
1787 if (!victim->d_inode)
1788 return -ENOENT;
1789
1790 BUG_ON(victim->d_parent->d_inode != dir);
cccc6bba 1791 audit_inode_child(victim, dir);
1da177e4 1792
f419a2e3 1793 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1794 if (error)
1795 return error;
1796 if (IS_APPEND(dir))
1797 return -EPERM;
1798 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
f9454548 1799 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1da177e4
LT		 1800 return -EPERM;
1801 if (isdir) {
1802 if (!S_ISDIR(victim->d_inode->i_mode))
1803 return -ENOTDIR;
1804 if (IS_ROOT(victim))
1805 return -EBUSY;
1806 } else if (S_ISDIR(victim->d_inode->i_mode))
1807 return -EISDIR;
1808 if (IS_DEADDIR(dir))
1809 return -ENOENT;
1810 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1811 return -EBUSY;
1812 return 0;
1813}
1814
1815/* Check whether we can create an object with dentry child in directory
1816 * dir.
1817 * 1. We can't do it if child already exists (open has special treatment for
1818 * this case, but since we are inlined it's OK)
1819 * 2. We can't do it if dir is read-only (done in permission())
1820 * 3. We should have write and exec permissions on dir
1821 * 4. We can't do it if dir is immutable (done in permission())
1822 */
a95164d9 1823static inline int may_create(struct inode *dir, struct dentry *child)
1da177e4
LT		 1824{
1825 if (child->d_inode)
1826 return -EEXIST;
1827 if (IS_DEADDIR(dir))
1828 return -ENOENT;
f419a2e3 1829 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1830}
1831
1da177e4
LT		 1832/*
1833 * p1 and p2 should be directories on the same fs.
1834 */
1835struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1836{
1837 struct dentry *p;
1838
1839 if (p1 == p2) {
f2eace23 1840 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1da177e4
LT		 1841 return NULL;
1842 }
1843
a11f3a05 1844 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4 1845
e2761a11
OH		 1846 p = d_ancestor(p2, p1);
1847 if (p) {
1848 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1849 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1850 return p;
1da177e4
LT		 1851 }
1852
e2761a11
OH		 1853 p = d_ancestor(p1, p2);
1854 if (p) {
1855 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1856 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1857 return p;
1da177e4
LT		 1858 }
1859
f2eace23
IM		 1860 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1861 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1da177e4
LT		 1862 return NULL;
1863}
1864
1865void unlock_rename(struct dentry *p1, struct dentry *p2)
1866{
1b1dcc1b 1867 mutex_unlock(&p1->d_inode->i_mutex);
1da177e4 1868 if (p1 != p2) {
1b1dcc1b 1869 mutex_unlock(&p2->d_inode->i_mutex);
a11f3a05 1870 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4
LT		 1871 }
1872}
1873
1874int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1875 struct nameidata *nd)
1876{
a95164d9 1877 int error = may_create(dir, dentry);
1da177e4
LT		 1878
1879 if (error)
1880 return error;
1881
acfa4380 1882 if (!dir->i_op->create)
1da177e4
LT		 1883 return -EACCES; /* shouldn't it be ENOSYS? */
1884 mode &= S_IALLUGO;
1885 mode |= S_IFREG;
1886 error = security_inode_create(dir, dentry, mode);
1887 if (error)
1888 return error;
1da177e4 1889 error = dir->i_op->create(dir, dentry, mode, nd);
a74574aa 1890 if (!error)
f38aa942 1891 fsnotify_create(dir, dentry);
1da177e4
LT		 1892 return error;
1893}
1894
3fb64190 1895int may_open(struct path *path, int acc_mode, int flag)
1da177e4 1896{
3fb64190 1897 struct dentry *dentry = path->dentry;
1da177e4
LT		 1898 struct inode *inode = dentry->d_inode;
1899 int error;
1900
1901 if (!inode)
1902 return -ENOENT;
1903
c8fe8f30
CH		 1904 switch (inode->i_mode & S_IFMT) {
1905 case S_IFLNK:
1da177e4 1906 return -ELOOP;
c8fe8f30
CH		 1907 case S_IFDIR:
1908 if (acc_mode & MAY_WRITE)
1909 return -EISDIR;
1910 break;
1911 case S_IFBLK:
1912 case S_IFCHR:
3fb64190 1913 if (path->mnt->mnt_flags & MNT_NODEV)
1da177e4 1914 return -EACCES;
c8fe8f30
CH		 1915 /*FALLTHRU*/
1916 case S_IFIFO:
1917 case S_IFSOCK:
1da177e4 1918 flag &= ~O_TRUNC;
c8fe8f30 1919 break;
4a3fd211 1920 }
b41572e9 1921
3fb64190 1922 error = inode_permission(inode, acc_mode);
b41572e9
DH		 1923 if (error)
1924 return error;
6146f0d5 1925
1da177e4
LT		 1926 /*
1927 * An append-only file must be opened in append mode for writing.
1928 */
1929 if (IS_APPEND(inode)) {
8737c930 1930 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
7715b521 1931 return -EPERM;
1da177e4 1932 if (flag & O_TRUNC)
7715b521 1933 return -EPERM;
1da177e4
LT		 1934 }
1935
1936 /* O_NOATIME can only be set by the owner or superuser */
7715b521
AV		 1937 if (flag & O_NOATIME && !is_owner_or_cap(inode))
1938 return -EPERM;
1da177e4
LT		 1939
1940 /*
1941 * Ensure there are no outstanding leases on the file.
1942 */
b65a9cfc 1943 return break_lease(inode, flag);
7715b521 1944}
1da177e4 1945
7715b521
AV		 1946static int handle_truncate(struct path *path)
1947{
1948 struct inode *inode = path->dentry->d_inode;
1949 int error = get_write_access(inode);
1950 if (error)
1951 return error;
1952 /*
1953 * Refuse to truncate files with mandatory locks held on them.
1954 */
1955 error = locks_verify_locked(inode);
1956 if (!error)
ea0d3ab2 1957 error = security_path_truncate(path);
7715b521
AV		 1958 if (!error) {
1959 error = do_truncate(path->dentry, 0,
1960 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1961 NULL);
1962 }
1963 put_write_access(inode);
acd0c935 1964 return error;
1da177e4
LT		 1965}
1966
d57999e1
DH		 1967/*
1968 * Be careful about ever adding any more callers of this
1969 * function. Its flags must be in the namei format, not
1970 * what get passed to sys_open().
1971 */
1972static int __open_namei_create(struct nameidata *nd, struct path *path,
8737c930 1973 int open_flag, int mode)
aab520e2
DH		 1974{
1975 int error;
4ac91378 1976 struct dentry *dir = nd->path.dentry;
aab520e2
DH		 1977
1978 if (!IS_POSIXACL(dir->d_inode))
ce3b0f8d 1979 mode &= ~current_umask();
be6d3e56
KT		 1980 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1981 if (error)
1982 goto out_unlock;
aab520e2 1983 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
be6d3e56 1984out_unlock:
aab520e2 1985 mutex_unlock(&dir->d_inode->i_mutex);
4ac91378
JB		 1986 dput(nd->path.dentry);
1987 nd->path.dentry = path->dentry;
31e6b01f 1988
aab520e2
DH		 1989 if (error)
1990 return error;
1991 /* Don't check for write permission, don't truncate */
8737c930 1992 return may_open(&nd->path, 0, open_flag & ~O_TRUNC);
aab520e2
DH		 1993}
1994
d57999e1
DH		 1995/*
1996 * Note that while the flag value (low two bits) for sys_open means:
1997 * 00 - read-only
1998 * 01 - write-only
1999 * 10 - read-write
2000 * 11 - special
2001 * it is changed into
2002 * 00 - no permissions needed
2003 * 01 - read-permission
2004 * 10 - write-permission
2005 * 11 - read-write
2006 * for the internal routines (ie open_namei()/follow_link() etc)
2007 * This is more logical, and also allows the 00 "no perm needed"
2008 * to be used for symlinks (where the permissions are checked
2009 * later).
2010 *
2011*/
2012static inline int open_to_namei_flags(int flag)
2013{
2014 if ((flag+1) & O_ACCMODE)
2015 flag++;
2016 return flag;
2017}
2018
7715b521 2019static int open_will_truncate(int flag, struct inode *inode)
4a3fd211
DH		 2020{
2021 /*
2022 * We'll never write to the fs underlying
2023 * a device file.
2024 */
2025 if (special_file(inode->i_mode))
2026 return 0;
2027 return (flag & O_TRUNC);
2028}
2029
648fa861 2030static struct file *finish_open(struct nameidata *nd,
9a66179e 2031 int open_flag, int acc_mode)
648fa861
AV		 2032{
2033 struct file *filp;
2034 int will_truncate;
2035 int error;
2036
9a66179e 2037 will_truncate = open_will_truncate(open_flag, nd->path.dentry->d_inode);
648fa861
AV		 2038 if (will_truncate) {
2039 error = mnt_want_write(nd->path.mnt);
2040 if (error)
2041 goto exit;
2042 }
2043 error = may_open(&nd->path, acc_mode, open_flag);
2044 if (error) {
2045 if (will_truncate)
2046 mnt_drop_write(nd->path.mnt);
2047 goto exit;
2048 }
2049 filp = nameidata_to_filp(nd);
2050 if (!IS_ERR(filp)) {
2051 error = ima_file_check(filp, acc_mode);
2052 if (error) {
2053 fput(filp);
2054 filp = ERR_PTR(error);
2055 }
2056 }
2057 if (!IS_ERR(filp)) {
648fa861
AV		 2058 if (will_truncate) {
2059 error = handle_truncate(&nd->path);
2060 if (error) {
2061 fput(filp);
2062 filp = ERR_PTR(error);
2063 }
2064 }
2065 }
2066 /*
2067 * It is now safe to drop the mnt write
2068 * because the filp has had a write taken
2069 * on its behalf.
2070 */
2071 if (will_truncate)
2072 mnt_drop_write(nd->path.mnt);
d893f1bc 2073 path_put(&nd->path);
648fa861
AV		 2074 return filp;
2075
2076exit:
2077 if (!IS_ERR(nd->intent.open.file))
2078 release_open_intent(nd);
2079 path_put(&nd->path);
2080 return ERR_PTR(error);
2081}
2082
31e6b01f
NP		 2083/*
2084 * Handle O_CREAT case for do_filp_open
2085 */
fb1cc555 2086static struct file *do_last(struct nameidata *nd, struct path *path,
5b369df8 2087 int open_flag, int acc_mode,
3e297b61 2088 int mode, const char *pathname)
fb1cc555 2089{
a1e28038 2090 struct dentry *dir = nd->path.dentry;
fb1cc555 2091 struct file *filp;
1f36f774
AV		 2092 int error = -EISDIR;
2093
2094 switch (nd->last_type) {
2095 case LAST_DOTDOT:
2096 follow_dotdot(nd);
2097 dir = nd->path.dentry;
176306f5 2098 case LAST_DOT:
fb045adb 2099 if (need_reval_dot(dir)) {
34286d66
NP		 2100 error = d_revalidate(nd->path.dentry, nd);
2101 if (!error)
1f36f774 2102 error = -ESTALE;
34286d66 2103 if (error < 0)
1f36f774 2104 goto exit;
1f36f774
AV		 2105 }
2106 /* fallthrough */
1f36f774 2107 case LAST_ROOT:
31e6b01f 2108 goto exit;
1f36f774
AV		 2109 case LAST_BIND:
2110 audit_inode(pathname, dir);
67ee3ad2 2111 goto ok;
1f36f774 2112 }
67ee3ad2 2113
1f36f774 2114 /* trailing slashes? */
31e6b01f
NP		 2115 if (nd->last.name[nd->last.len])
2116 goto exit;
a2c36b45 2117
a1e28038
AV		 2118 mutex_lock(&dir->d_inode->i_mutex);
2119
2120 path->dentry = lookup_hash(nd);
2121 path->mnt = nd->path.mnt;
2122
fb1cc555
AV		 2123 error = PTR_ERR(path->dentry);
2124 if (IS_ERR(path->dentry)) {
2125 mutex_unlock(&dir->d_inode->i_mutex);
2126 goto exit;
2127 }
2128
2129 if (IS_ERR(nd->intent.open.file)) {
2130 error = PTR_ERR(nd->intent.open.file);
2131 goto exit_mutex_unlock;
2132 }
2133
2134 /* Negative dentry, just create the file */
2135 if (!path->dentry->d_inode) {
2136 /*
2137 * This write is needed to ensure that a
2138 * ro->rw transition does not occur between
2139 * the time when the file is created and when
2140 * a permanent write count is taken through
2141 * the 'struct file' in nameidata_to_filp().
2142 */
2143 error = mnt_want_write(nd->path.mnt);
2144 if (error)
2145 goto exit_mutex_unlock;
2146 error = __open_namei_create(nd, path, open_flag, mode);
2147 if (error) {
2148 mnt_drop_write(nd->path.mnt);
2149 goto exit;
2150 }
2151 filp = nameidata_to_filp(nd);
2152 mnt_drop_write(nd->path.mnt);
d893f1bc 2153 path_put(&nd->path);
fb1cc555
AV		 2154 if (!IS_ERR(filp)) {
2155 error = ima_file_check(filp, acc_mode);
2156 if (error) {
2157 fput(filp);
2158 filp = ERR_PTR(error);
2159 }
2160 }
2161 return filp;
2162 }
2163
2164 /*
2165 * It already exists.
2166 */
2167 mutex_unlock(&dir->d_inode->i_mutex);
2168 audit_inode(pathname, path->dentry);
2169
2170 error = -EEXIST;
5b369df8 2171 if (open_flag & O_EXCL)
fb1cc555
AV		 2172 goto exit_dput;
2173
2174 if (__follow_mount(path)) {
2175 error = -ELOOP;
5b369df8 2176 if (open_flag & O_NOFOLLOW)
fb1cc555
AV		 2177 goto exit_dput;
2178 }
2179
2180 error = -ENOENT;
2181 if (!path->dentry->d_inode)
2182 goto exit_dput;
9e67f361
AV		 2183
2184 if (path->dentry->d_inode->i_op->follow_link)
fb1cc555 2185 return NULL;
fb1cc555
AV		 2186
2187 path_to_nameidata(path, nd);
31e6b01f 2188 nd->inode = path->dentry->d_inode;
fb1cc555 2189 error = -EISDIR;
31e6b01f 2190 if (S_ISDIR(nd->inode->i_mode))
fb1cc555 2191 goto exit;
67ee3ad2 2192ok:
9a66179e 2193 filp = finish_open(nd, open_flag, acc_mode);
fb1cc555
AV		 2194 return filp;
2195
2196exit_mutex_unlock:
2197 mutex_unlock(&dir->d_inode->i_mutex);
2198exit_dput:
2199 path_put_conditional(path, nd);
2200exit:
2201 if (!IS_ERR(nd->intent.open.file))
2202 release_open_intent(nd);
fb1cc555
AV		 2203 path_put(&nd->path);
2204 return ERR_PTR(error);
2205}
2206
1da177e4 2207/*
4a3fd211
DH		 2208 * Note that the low bits of the passed in "open_flag"
2209 * are not the same as in the local variable "flag". See
2210 * open_to_namei_flags() for more details.
1da177e4 2211 */
a70e65df 2212struct file *do_filp_open(int dfd, const char *pathname,
6e8341a1 2213 int open_flag, int mode, int acc_mode)
1da177e4 2214{
4a3fd211 2215 struct file *filp;
a70e65df 2216 struct nameidata nd;
6e8341a1 2217 int error;
9850c056 2218 struct path path;
1da177e4 2219 int count = 0;
d57999e1 2220 int flag = open_to_namei_flags(open_flag);
31e6b01f 2221 int flags;
1f36f774
AV		 2222
2223 if (!(open_flag & O_CREAT))
2224 mode = 0;
1da177e4 2225
b1085ba8
LS		 2226 /* Must never be set by userspace */
2227 open_flag &= ~FMODE_NONOTIFY;
2228
6b2f3d1f
CH		 2229 /*
2230 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
2231 * check for O_DSYNC if the need any syncing at all we enforce it's
2232 * always set instead of having to deal with possibly weird behaviour
2233 * for malicious applications setting only __O_SYNC.
2234 */
2235 if (open_flag & __O_SYNC)
2236 open_flag |= O_DSYNC;
2237
6e8341a1 2238 if (!acc_mode)
6d125529 2239 acc_mode = MAY_OPEN | ACC_MODE(open_flag);
1da177e4 2240
834f2a4a 2241 /* O_TRUNC implies we need access checks for write permissions */
4296e2cb 2242 if (open_flag & O_TRUNC)
834f2a4a
TM		 2243 acc_mode |= MAY_WRITE;
2244
1da177e4
LT		 2245 /* Allow the LSM permission hook to distinguish append
2246 access from general write access. */
4296e2cb 2247 if (open_flag & O_APPEND)
1da177e4
LT		 2248 acc_mode |= MAY_APPEND;
2249
31e6b01f
NP		 2250 flags = LOOKUP_OPEN;
2251 if (open_flag & O_CREAT) {
2252 flags |= LOOKUP_CREATE;
2253 if (open_flag & O_EXCL)
2254 flags |= LOOKUP_EXCL;
2255 }
2256 if (open_flag & O_DIRECTORY)
2257 flags |= LOOKUP_DIRECTORY;
2258 if (!(open_flag & O_NOFOLLOW))
2259 flags |= LOOKUP_FOLLOW;
2260
2261 filp = get_empty_filp();
2262 if (!filp)
2263 return ERR_PTR(-ENFILE);
2264
2265 filp->f_flags = open_flag;
2266 nd.intent.open.file = filp;
2267 nd.intent.open.flags = flag;
2268 nd.intent.open.create_mode = mode;
2269
2270 if (open_flag & O_CREAT)
2271 goto creat;
2272
2273 /* !O_CREAT, simple open */
2274 error = do_path_lookup(dfd, pathname, flags, &nd);
2275 if (unlikely(error))
2276 goto out_filp;
2277 error = -ELOOP;
2278 if (!(nd.flags & LOOKUP_FOLLOW)) {
2279 if (nd.inode->i_op->follow_link)
2280 goto out_path;
2281 }
2282 error = -ENOTDIR;
2283 if (nd.flags & LOOKUP_DIRECTORY) {
2284 if (!nd.inode->i_op->lookup)
2285 goto out_path;
2286 }
2287 audit_inode(pathname, nd.path.dentry);
2288 filp = finish_open(&nd, open_flag, acc_mode);
2289 return filp;
2290
2291creat:
2292 /* OK, have to create the file. Find the parent. */
2293 error = path_init_rcu(dfd, pathname,
2294 LOOKUP_PARENT | (flags & LOOKUP_REVAL), &nd);
1da177e4 2295 if (error)
31e6b01f
NP		 2296 goto out_filp;
2297 error = path_walk_rcu(pathname, &nd);
2298 path_finish_rcu(&nd);
2299 if (unlikely(error == -ECHILD || error == -ESTALE)) {
2300 /* slower, locked walk */
2301 if (error == -ESTALE) {
2302reval:
2303 flags |= LOOKUP_REVAL;
2304 }
2305 error = path_init(dfd, pathname,
2306 LOOKUP_PARENT | (flags & LOOKUP_REVAL), &nd);
2307 if (error)
2308 goto out_filp;
3866248e 2309
31e6b01f 2310 error = path_walk_simple(pathname, &nd);
654f562c 2311 }
31e6b01f
NP		 2312 if (unlikely(error))
2313 goto out_filp;
2314 if (unlikely(!audit_dummy_context()))
9b4a9b14 2315 audit_inode(pathname, nd.path.dentry);
1da177e4
LT		 2316
2317 /*
a2c36b45 2318 * We have the parent and last component.
1da177e4 2319 */
31e6b01f 2320 nd.flags = flags;
3e297b61 2321 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
806b681c 2322 while (unlikely(!filp)) { /* trailing symlink */
def4af30 2323 struct path holder;
def4af30 2324 void *cookie;
806b681c 2325 error = -ELOOP;
1f36f774 2326 /* S_ISDIR part is a temporary automount kludge */
31e6b01f 2327 if (!(nd.flags & LOOKUP_FOLLOW) && !S_ISDIR(nd.inode->i_mode))
1f36f774
AV		 2328 goto exit_dput;
2329 if (count++ == 32)
806b681c
AV		 2330 goto exit_dput;
2331 /*
2332 * This is subtle. Instead of calling do_follow_link() we do
2333 * the thing by hands. The reason is that this way we have zero
2334 * link_count and path_walk() (called from ->follow_link)
2335 * honoring LOOKUP_PARENT. After that we have the parent and
2336 * last component, i.e. we are in the same situation as after
2337 * the first path_walk(). Well, almost - if the last component
2338 * is normal we get its copy stored in nd->last.name and we will
2339 * have to putname() it when we are done. Procfs-like symlinks
2340 * just set LAST_BIND.
2341 */
2342 nd.flags |= LOOKUP_PARENT;
2343 error = security_inode_follow_link(path.dentry, &nd);
2344 if (error)
2345 goto exit_dput;
def4af30
AV		 2346 error = __do_follow_link(&path, &nd, &cookie);
2347 if (unlikely(error)) {
31e6b01f
NP		 2348 if (!IS_ERR(cookie) && nd.inode->i_op->put_link)
2349 nd.inode->i_op->put_link(path.dentry, &nd, cookie);
806b681c 2350 /* nd.path had been dropped */
31e6b01f
NP		 2351 nd.path = path;
2352 goto out_path;
806b681c 2353 }
def4af30 2354 holder = path;
806b681c 2355 nd.flags &= ~LOOKUP_PARENT;
3e297b61 2356 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
31e6b01f
NP		 2357 if (nd.inode->i_op->put_link)
2358 nd.inode->i_op->put_link(holder.dentry, &nd, cookie);
def4af30 2359 path_put(&holder);
806b681c 2360 }
10fa8e62 2361out:
2a737871
AV		 2362 if (nd.root.mnt)
2363 path_put(&nd.root);
31e6b01f 2364 if (filp == ERR_PTR(-ESTALE) && !(flags & LOOKUP_REVAL))
10fa8e62 2365 goto reval;
10fa8e62 2366 return filp;
1da177e4 2367
806b681c
AV		 2368exit_dput:
2369 path_put_conditional(&path, &nd);
31e6b01f
NP		 2370out_path:
2371 path_put(&nd.path);
2372out_filp:
806b681c 2373 if (!IS_ERR(nd.intent.open.file))
a70e65df 2374 release_open_intent(&nd);
806b681c 2375 filp = ERR_PTR(error);
10fa8e62 2376 goto out;
1da177e4
LT		 2377}
2378
a70e65df
CH		 2379/**
2380 * filp_open - open file and return file pointer
2381 *
2382 * @filename: path to open
2383 * @flags: open flags as per the open(2) second argument
2384 * @mode: mode for the new file if O_CREAT is set, else ignored
2385 *
2386 * This is the helper to open a file from kernelspace if you really
2387 * have to. But in generally you should not do this, so please move
2388 * along, nothing to see here..
2389 */
2390struct file *filp_open(const char *filename, int flags, int mode)
2391{
6e8341a1 2392 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
a70e65df
CH		 2393}
2394EXPORT_SYMBOL(filp_open);
2395
1da177e4
LT		 2396/**
2397 * lookup_create - lookup a dentry, creating it if it doesn't exist
2398 * @nd: nameidata info
2399 * @is_dir: directory flag
2400 *
2401 * Simple function to lookup and return a dentry and create it
2402 * if it doesn't exist. Is SMP-safe.
c663e5d8 2403 *
4ac91378 2404 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1da177e4
LT		 2405 */
2406struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2407{
c663e5d8 2408 struct dentry *dentry = ERR_PTR(-EEXIST);
1da177e4 2409
4ac91378 2410 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
c663e5d8
CH		 2411 /*
2412 * Yucky last component or no last component at all?
2413 * (foo/., foo/.., /////)
2414 */
1da177e4
LT		 2415 if (nd->last_type != LAST_NORM)
2416 goto fail;
2417 nd->flags &= ~LOOKUP_PARENT;
3516586a 2418 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
a634904a 2419 nd->intent.open.flags = O_EXCL;
c663e5d8
CH		 2420
2421 /*
2422 * Do the final lookup.
2423 */
49705b77 2424 dentry = lookup_hash(nd);
1da177e4
LT		 2425 if (IS_ERR(dentry))
2426 goto fail;
c663e5d8 2427
e9baf6e5
AV		 2428 if (dentry->d_inode)
2429 goto eexist;
c663e5d8
CH		 2430 /*
2431 * Special case - lookup gave negative, but... we had foo/bar/
2432 * From the vfs_mknod() POV we just have a negative dentry -
2433 * all is fine. Let's be bastards - you had / on the end, you've
2434 * been asking for (non-existent) directory. -ENOENT for you.
2435 */
e9baf6e5
AV		 2436 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2437 dput(dentry);
2438 dentry = ERR_PTR(-ENOENT);
2439 }
1da177e4 2440 return dentry;
e9baf6e5 2441eexist:
1da177e4 2442 dput(dentry);
e9baf6e5 2443 dentry = ERR_PTR(-EEXIST);
1da177e4
LT		 2444fail:
2445 return dentry;
2446}
f81a0bff 2447EXPORT_SYMBOL_GPL(lookup_create);
1da177e4
LT		 2448
2449int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2450{
a95164d9 2451 int error = may_create(dir, dentry);
1da177e4
LT		 2452
2453 if (error)
2454 return error;
2455
2456 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2457 return -EPERM;
2458
acfa4380 2459 if (!dir->i_op->mknod)
1da177e4
LT		 2460 return -EPERM;
2461
08ce5f16
SH		 2462 error = devcgroup_inode_mknod(mode, dev);
2463 if (error)
2464 return error;
2465
1da177e4
LT		 2466 error = security_inode_mknod(dir, dentry, mode, dev);
2467 if (error)
2468 return error;
2469
1da177e4 2470 error = dir->i_op->mknod(dir, dentry, mode, dev);
a74574aa 2471 if (!error)
f38aa942 2472 fsnotify_create(dir, dentry);
1da177e4
LT		 2473 return error;
2474}
2475
463c3197
DH		 2476static int may_mknod(mode_t mode)
2477{
2478 switch (mode & S_IFMT) {
2479 case S_IFREG:
2480 case S_IFCHR:
2481 case S_IFBLK:
2482 case S_IFIFO:
2483 case S_IFSOCK:
2484 case 0: /* zero mode translates to S_IFREG */
2485 return 0;
2486 case S_IFDIR:
2487 return -EPERM;
2488 default:
2489 return -EINVAL;
2490 }
2491}
2492
2e4d0924
HC		 2493SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2494 unsigned, dev)
1da177e4 2495{
2ad94ae6
AV		 2496 int error;
2497 char *tmp;
2498 struct dentry *dentry;
1da177e4
LT		 2499 struct nameidata nd;
2500
2501 if (S_ISDIR(mode))
2502 return -EPERM;
1da177e4 2503
2ad94ae6 2504 error = user_path_parent(dfd, filename, &nd, &tmp);
1da177e4 2505 if (error)
2ad94ae6
AV		 2506 return error;
2507
1da177e4 2508 dentry = lookup_create(&nd, 0);
463c3197
DH		 2509 if (IS_ERR(dentry)) {
2510 error = PTR_ERR(dentry);
2511 goto out_unlock;
2512 }
4ac91378 2513 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2514 mode &= ~current_umask();
463c3197
DH		 2515 error = may_mknod(mode);
2516 if (error)
2517 goto out_dput;
2518 error = mnt_want_write(nd.path.mnt);
2519 if (error)
2520 goto out_dput;
be6d3e56
KT		 2521 error = security_path_mknod(&nd.path, dentry, mode, dev);
2522 if (error)
2523 goto out_drop_write;
463c3197 2524 switch (mode & S_IFMT) {
1da177e4 2525 case 0: case S_IFREG:
4ac91378 2526 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
1da177e4
LT		 2527 break;
2528 case S_IFCHR: case S_IFBLK:
4ac91378 2529 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
1da177e4
LT		 2530 new_decode_dev(dev));
2531 break;
2532 case S_IFIFO: case S_IFSOCK:
4ac91378 2533 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
1da177e4 2534 break;
1da177e4 2535 }
be6d3e56 2536out_drop_write:
463c3197
DH		 2537 mnt_drop_write(nd.path.mnt);
2538out_dput:
2539 dput(dentry);
2540out_unlock:
4ac91378 2541 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2542 path_put(&nd.path);
1da177e4
LT		 2543 putname(tmp);
2544
2545 return error;
2546}
2547
3480b257 2548SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
5590ff0d
UD		 2549{
2550 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2551}
2552
1da177e4
LT		 2553int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2554{
a95164d9 2555 int error = may_create(dir, dentry);
1da177e4
LT		 2556
2557 if (error)
2558 return error;
2559
acfa4380 2560 if (!dir->i_op->mkdir)
1da177e4
LT		 2561 return -EPERM;
2562
2563 mode &= (S_IRWXUGO|S_ISVTX);
2564 error = security_inode_mkdir(dir, dentry, mode);
2565 if (error)
2566 return error;
2567
1da177e4 2568 error = dir->i_op->mkdir(dir, dentry, mode);
a74574aa 2569 if (!error)
f38aa942 2570 fsnotify_mkdir(dir, dentry);
1da177e4
LT		 2571 return error;
2572}
2573
2e4d0924 2574SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
1da177e4
LT		 2575{
2576 int error = 0;
2577 char * tmp;
6902d925
DH		 2578 struct dentry *dentry;
2579 struct nameidata nd;
1da177e4 2580
2ad94ae6
AV		 2581 error = user_path_parent(dfd, pathname, &nd, &tmp);
2582 if (error)
6902d925 2583 goto out_err;
1da177e4 2584
6902d925
DH		 2585 dentry = lookup_create(&nd, 1);
2586 error = PTR_ERR(dentry);
2587 if (IS_ERR(dentry))
2588 goto out_unlock;
1da177e4 2589
4ac91378 2590 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2591 mode &= ~current_umask();
463c3197
DH		 2592 error = mnt_want_write(nd.path.mnt);
2593 if (error)
2594 goto out_dput;
be6d3e56
KT		 2595 error = security_path_mkdir(&nd.path, dentry, mode);
2596 if (error)
2597 goto out_drop_write;
4ac91378 2598 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
be6d3e56 2599out_drop_write:
463c3197
DH		 2600 mnt_drop_write(nd.path.mnt);
2601out_dput:
6902d925
DH		 2602 dput(dentry);
2603out_unlock:
4ac91378 2604 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2605 path_put(&nd.path);
6902d925
DH		 2606 putname(tmp);
2607out_err:
1da177e4
LT		 2608 return error;
2609}
2610
3cdad428 2611SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
5590ff0d
UD		 2612{
2613 return sys_mkdirat(AT_FDCWD, pathname, mode);
2614}
2615
1da177e4
LT		 2616/*
2617 * We try to drop the dentry early: we should have
2618 * a usage count of 2 if we're the only user of this
2619 * dentry, and if that is true (possibly after pruning
2620 * the dcache), then we drop the dentry now.
2621 *
2622 * A low-level filesystem can, if it choses, legally
2623 * do a
2624 *
2625 * if (!d_unhashed(dentry))
2626 * return -EBUSY;
2627 *
2628 * if it cannot handle the case of removing a directory
2629 * that is still in use by something else..
2630 */
2631void dentry_unhash(struct dentry *dentry)
2632{
2633 dget(dentry);
dc168427 2634 shrink_dcache_parent(dentry);
1da177e4 2635 spin_lock(&dentry->d_lock);
b7ab39f6 2636 if (dentry->d_count == 2)
1da177e4
LT		 2637 __d_drop(dentry);
2638 spin_unlock(&dentry->d_lock);
1da177e4
LT		 2639}
2640
2641int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2642{
2643 int error = may_delete(dir, dentry, 1);
2644
2645 if (error)
2646 return error;
2647
acfa4380 2648 if (!dir->i_op->rmdir)
1da177e4
LT		 2649 return -EPERM;
2650
1b1dcc1b 2651 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2652 dentry_unhash(dentry);
2653 if (d_mountpoint(dentry))
2654 error = -EBUSY;
2655 else {
2656 error = security_inode_rmdir(dir, dentry);
2657 if (!error) {
2658 error = dir->i_op->rmdir(dir, dentry);
d83c49f3 2659 if (!error) {
1da177e4 2660 dentry->d_inode->i_flags |= S_DEAD;
d83c49f3
AV		 2661 dont_mount(dentry);
2662 }
1da177e4
LT		 2663 }
2664 }
1b1dcc1b 2665 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4 2666 if (!error) {
1da177e4
LT		 2667 d_delete(dentry);
2668 }
2669 dput(dentry);
2670
2671 return error;
2672}
2673
5590ff0d 2674static long do_rmdir(int dfd, const char __user *pathname)
1da177e4
LT		 2675{
2676 int error = 0;
2677 char * name;
2678 struct dentry *dentry;
2679 struct nameidata nd;
2680
2ad94ae6 2681 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2682 if (error)
2ad94ae6 2683 return error;
1da177e4
LT		 2684
2685 switch(nd.last_type) {
0612d9fb
OH		 2686 case LAST_DOTDOT:
2687 error = -ENOTEMPTY;
2688 goto exit1;
2689 case LAST_DOT:
2690 error = -EINVAL;
2691 goto exit1;
2692 case LAST_ROOT:
2693 error = -EBUSY;
2694 goto exit1;
1da177e4 2695 }
0612d9fb
OH		 2696
2697 nd.flags &= ~LOOKUP_PARENT;
2698
4ac91378 2699 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2700 dentry = lookup_hash(&nd);
1da177e4 2701 error = PTR_ERR(dentry);
6902d925
DH		 2702 if (IS_ERR(dentry))
2703 goto exit2;
0622753b
DH		 2704 error = mnt_want_write(nd.path.mnt);
2705 if (error)
2706 goto exit3;
be6d3e56
KT		 2707 error = security_path_rmdir(&nd.path, dentry);
2708 if (error)
2709 goto exit4;
4ac91378 2710 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
be6d3e56 2711exit4:
0622753b
DH		 2712 mnt_drop_write(nd.path.mnt);
2713exit3:
6902d925
DH		 2714 dput(dentry);
2715exit2:
4ac91378 2716 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2717exit1:
1d957f9b 2718 path_put(&nd.path);
1da177e4
LT		 2719 putname(name);
2720 return error;
2721}
2722
3cdad428 2723SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
5590ff0d
UD		 2724{
2725 return do_rmdir(AT_FDCWD, pathname);
2726}
2727
1da177e4
LT		 2728int vfs_unlink(struct inode *dir, struct dentry *dentry)
2729{
2730 int error = may_delete(dir, dentry, 0);
2731
2732 if (error)
2733 return error;
2734
acfa4380 2735 if (!dir->i_op->unlink)
1da177e4
LT		 2736 return -EPERM;
2737
1b1dcc1b 2738 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2739 if (d_mountpoint(dentry))
2740 error = -EBUSY;
2741 else {
2742 error = security_inode_unlink(dir, dentry);
bec1052e 2743 if (!error) {
1da177e4 2744 error = dir->i_op->unlink(dir, dentry);
bec1052e 2745 if (!error)
d83c49f3 2746 dont_mount(dentry);
bec1052e 2747 }
1da177e4 2748 }
1b1dcc1b 2749 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2750
2751 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2752 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
ece95912 2753 fsnotify_link_count(dentry->d_inode);
e234f35c 2754 d_delete(dentry);
1da177e4 2755 }
0eeca283 2756
1da177e4
LT		 2757 return error;
2758}
2759
2760/*
2761 * Make sure that the actual truncation of the file will occur outside its
1b1dcc1b 2762 * directory's i_mutex. Truncate can take a long time if there is a lot of
1da177e4
LT		 2763 * writeout happening, and we don't want to prevent access to the directory
2764 * while waiting on the I/O.
2765 */
5590ff0d 2766static long do_unlinkat(int dfd, const char __user *pathname)
1da177e4 2767{
2ad94ae6
AV		 2768 int error;
2769 char *name;
1da177e4
LT		 2770 struct dentry *dentry;
2771 struct nameidata nd;
2772 struct inode *inode = NULL;
2773
2ad94ae6 2774 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2775 if (error)
2ad94ae6
AV		 2776 return error;
2777
1da177e4
LT		 2778 error = -EISDIR;
2779 if (nd.last_type != LAST_NORM)
2780 goto exit1;
0612d9fb
OH		 2781
2782 nd.flags &= ~LOOKUP_PARENT;
2783
4ac91378 2784 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2785 dentry = lookup_hash(&nd);
1da177e4
LT		 2786 error = PTR_ERR(dentry);
2787 if (!IS_ERR(dentry)) {
2788 /* Why not before? Because we want correct error value */
2789 if (nd.last.name[nd.last.len])
2790 goto slashes;
2791 inode = dentry->d_inode;
2792 if (inode)
7de9c6ee 2793 ihold(inode);
0622753b
DH		 2794 error = mnt_want_write(nd.path.mnt);
2795 if (error)
2796 goto exit2;
be6d3e56
KT		 2797 error = security_path_unlink(&nd.path, dentry);
2798 if (error)
2799 goto exit3;
4ac91378 2800 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
be6d3e56 2801exit3:
0622753b 2802 mnt_drop_write(nd.path.mnt);
1da177e4
LT		 2803 exit2:
2804 dput(dentry);
2805 }
4ac91378 2806 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4
LT		 2807 if (inode)
2808 iput(inode); /* truncate the inode here */
2809exit1:
1d957f9b 2810 path_put(&nd.path);
1da177e4
LT		 2811 putname(name);
2812 return error;
2813
2814slashes:
2815 error = !dentry->d_inode ? -ENOENT :
2816 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2817 goto exit2;
2818}
2819
2e4d0924 2820SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
5590ff0d
UD		 2821{
2822 if ((flag & ~AT_REMOVEDIR) != 0)
2823 return -EINVAL;
2824
2825 if (flag & AT_REMOVEDIR)
2826 return do_rmdir(dfd, pathname);
2827
2828 return do_unlinkat(dfd, pathname);
2829}
2830
3480b257 2831SYSCALL_DEFINE1(unlink, const char __user *, pathname)
5590ff0d
UD		 2832{
2833 return do_unlinkat(AT_FDCWD, pathname);
2834}
2835
db2e747b 2836int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
1da177e4 2837{
a95164d9 2838 int error = may_create(dir, dentry);
1da177e4
LT		 2839
2840 if (error)
2841 return error;
2842
acfa4380 2843 if (!dir->i_op->symlink)
1da177e4
LT		 2844 return -EPERM;
2845
2846 error = security_inode_symlink(dir, dentry, oldname);
2847 if (error)
2848 return error;
2849
1da177e4 2850 error = dir->i_op->symlink(dir, dentry, oldname);
a74574aa 2851 if (!error)
f38aa942 2852 fsnotify_create(dir, dentry);
1da177e4
LT		 2853 return error;
2854}
2855
2e4d0924
HC		 2856SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2857 int, newdfd, const char __user *, newname)
1da177e4 2858{
2ad94ae6
AV		 2859 int error;
2860 char *from;
2861 char *to;
6902d925
DH		 2862 struct dentry *dentry;
2863 struct nameidata nd;
1da177e4
LT		 2864
2865 from = getname(oldname);
2ad94ae6 2866 if (IS_ERR(from))
1da177e4 2867 return PTR_ERR(from);
1da177e4 2868
2ad94ae6 2869 error = user_path_parent(newdfd, newname, &nd, &to);
6902d925 2870 if (error)
2ad94ae6
AV		 2871 goto out_putname;
2872
6902d925
DH		 2873 dentry = lookup_create(&nd, 0);
2874 error = PTR_ERR(dentry);
2875 if (IS_ERR(dentry))
2876 goto out_unlock;
2877
75c3f29d
DH		 2878 error = mnt_want_write(nd.path.mnt);
2879 if (error)
2880 goto out_dput;
be6d3e56
KT		 2881 error = security_path_symlink(&nd.path, dentry, from);
2882 if (error)
2883 goto out_drop_write;
db2e747b 2884 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
be6d3e56 2885out_drop_write:
75c3f29d
DH		 2886 mnt_drop_write(nd.path.mnt);
2887out_dput:
6902d925
DH		 2888 dput(dentry);
2889out_unlock:
4ac91378 2890 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2891 path_put(&nd.path);
6902d925
DH		 2892 putname(to);
2893out_putname:
1da177e4
LT		 2894 putname(from);
2895 return error;
2896}
2897
3480b257 2898SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 2899{
2900 return sys_symlinkat(oldname, AT_FDCWD, newname);
2901}
2902
1da177e4
LT		 2903int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2904{
2905 struct inode *inode = old_dentry->d_inode;
2906 int error;
2907
2908 if (!inode)
2909 return -ENOENT;
2910
a95164d9 2911 error = may_create(dir, new_dentry);
1da177e4
LT		 2912 if (error)
2913 return error;
2914
2915 if (dir->i_sb != inode->i_sb)
2916 return -EXDEV;
2917
2918 /*
2919 * A link to an append-only or immutable file cannot be created.
2920 */
2921 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2922 return -EPERM;
acfa4380 2923 if (!dir->i_op->link)
1da177e4 2924 return -EPERM;
7e79eedb 2925 if (S_ISDIR(inode->i_mode))
1da177e4
LT		 2926 return -EPERM;
2927
2928 error = security_inode_link(old_dentry, dir, new_dentry);
2929 if (error)
2930 return error;
2931
7e79eedb 2932 mutex_lock(&inode->i_mutex);
1da177e4 2933 error = dir->i_op->link(old_dentry, dir, new_dentry);
7e79eedb 2934 mutex_unlock(&inode->i_mutex);
e31e14ec 2935 if (!error)
7e79eedb 2936 fsnotify_link(dir, inode, new_dentry);
1da177e4
LT		 2937 return error;
2938}
2939
2940/*
2941 * Hardlinks are often used in delicate situations. We avoid
2942 * security-related surprises by not following symlinks on the
2943 * newname. --KAB
2944 *
2945 * We don't follow them on the oldname either to be compatible
2946 * with linux 2.0, and to avoid hard-linking to directories
2947 * and other special files. --ADM
2948 */
2e4d0924
HC		 2949SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2950 int, newdfd, const char __user *, newname, int, flags)
1da177e4
LT		 2951{
2952 struct dentry *new_dentry;
2d8f3038
AV		 2953 struct nameidata nd;
2954 struct path old_path;
1da177e4 2955 int error;
2ad94ae6 2956 char *to;
1da177e4 2957
45c9b11a 2958 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
c04030e1
UD		 2959 return -EINVAL;
2960
2d8f3038
AV		 2961 error = user_path_at(olddfd, oldname,
2962 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2963 &old_path);
1da177e4 2964 if (error)
2ad94ae6
AV		 2965 return error;
2966
2967 error = user_path_parent(newdfd, newname, &nd, &to);
1da177e4
LT		 2968 if (error)
2969 goto out;
2970 error = -EXDEV;
2d8f3038 2971 if (old_path.mnt != nd.path.mnt)
1da177e4
LT		 2972 goto out_release;
2973 new_dentry = lookup_create(&nd, 0);
2974 error = PTR_ERR(new_dentry);
6902d925
DH		 2975 if (IS_ERR(new_dentry))
2976 goto out_unlock;
75c3f29d
DH		 2977 error = mnt_want_write(nd.path.mnt);
2978 if (error)
2979 goto out_dput;
be6d3e56
KT		 2980 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2981 if (error)
2982 goto out_drop_write;
2d8f3038 2983 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
be6d3e56 2984out_drop_write:
75c3f29d
DH		 2985 mnt_drop_write(nd.path.mnt);
2986out_dput:
6902d925
DH		 2987 dput(new_dentry);
2988out_unlock:
4ac91378 2989 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2990out_release:
1d957f9b 2991 path_put(&nd.path);
2ad94ae6 2992 putname(to);
1da177e4 2993out:
2d8f3038 2994 path_put(&old_path);
1da177e4
LT		 2995
2996 return error;
2997}
2998
3480b257 2999SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
5590ff0d 3000{
c04030e1 3001 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
5590ff0d
UD		 3002}
3003
1da177e4
LT		 3004/*
3005 * The worst of all namespace operations - renaming directory. "Perverted"
3006 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3007 * Problems:
3008 * a) we can get into loop creation. Check is done in is_subdir().
3009 * b) race potential - two innocent renames can create a loop together.
3010 * That's where 4.4 screws up. Current fix: serialization on
a11f3a05 3011 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
1da177e4
LT		 3012 * story.
3013 * c) we have to lock _three_ objects - parents and victim (if it exists).
1b1dcc1b 3014 * And that - after we got ->i_mutex on parents (until then we don't know
1da177e4
LT		 3015 * whether the target exists). Solution: try to be smart with locking
3016 * order for inodes. We rely on the fact that tree topology may change
a11f3a05 3017 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
1da177e4
LT		 3018 * move will be locked. Thus we can rank directories by the tree
3019 * (ancestors first) and rank all non-directories after them.
3020 * That works since everybody except rename does "lock parent, lookup,
a11f3a05 3021 * lock child" and rename is under ->s_vfs_rename_mutex.
1da177e4
LT		 3022 * HOWEVER, it relies on the assumption that any object with ->lookup()
3023 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3024 * we'd better make sure that there's no link(2) for them.
3025 * d) some filesystems don't support opened-but-unlinked directories,
3026 * either because of layout or because they are not ready to deal with
3027 * all cases correctly. The latter will be fixed (taking this sort of
3028 * stuff into VFS), but the former is not going away. Solution: the same
3029 * trick as in rmdir().
3030 * e) conversion from fhandle to dentry may come in the wrong moment - when
1b1dcc1b 3031 * we are removing the target. Solution: we will have to grab ->i_mutex
1da177e4 3032 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
c41b20e7 3033 * ->i_mutex on parents, which works but leads to some truly excessive
1da177e4
LT		 3034 * locking].
3035 */
75c96f85
AB		 3036static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3037 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3038{
3039 int error = 0;
3040 struct inode *target;
3041
3042 /*
3043 * If we are going to change the parent - check write permissions,
3044 * we'll need to flip '..'.
3045 */
3046 if (new_dir != old_dir) {
f419a2e3 3047 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
1da177e4
LT		 3048 if (error)
3049 return error;
3050 }
3051
3052 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3053 if (error)
3054 return error;
3055
3056 target = new_dentry->d_inode;
d83c49f3 3057 if (target)
1b1dcc1b 3058 mutex_lock(&target->i_mutex);
1da177e4
LT		 3059 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3060 error = -EBUSY;
d83c49f3
AV		 3061 else {
3062 if (target)
3063 dentry_unhash(new_dentry);
1da177e4 3064 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
d83c49f3 3065 }
1da177e4 3066 if (target) {
d83c49f3 3067 if (!error) {
1da177e4 3068 target->i_flags |= S_DEAD;
d83c49f3
AV		 3069 dont_mount(new_dentry);
3070 }
1b1dcc1b 3071 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3072 if (d_unhashed(new_dentry))
3073 d_rehash(new_dentry);
3074 dput(new_dentry);
3075 }
e31e14ec 3076 if (!error)
349457cc
MF		 3077 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3078 d_move(old_dentry,new_dentry);
1da177e4
LT		 3079 return error;
3080}
3081
75c96f85
AB		 3082static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3083 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3084{
3085 struct inode *target;
3086 int error;
3087
3088 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3089 if (error)
3090 return error;
3091
3092 dget(new_dentry);
3093 target = new_dentry->d_inode;
3094 if (target)
1b1dcc1b 3095 mutex_lock(&target->i_mutex);
1da177e4
LT		 3096 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3097 error = -EBUSY;
3098 else
3099 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3100 if (!error) {
bec1052e 3101 if (target)
d83c49f3 3102 dont_mount(new_dentry);
349457cc 3103 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
1da177e4 3104 d_move(old_dentry, new_dentry);
1da177e4
LT		 3105 }
3106 if (target)
1b1dcc1b 3107 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3108 dput(new_dentry);
3109 return error;
3110}
3111
3112int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3113 struct inode *new_dir, struct dentry *new_dentry)
3114{
3115 int error;
3116 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
59b0df21 3117 const unsigned char *old_name;
1da177e4
LT		 3118
3119 if (old_dentry->d_inode == new_dentry->d_inode)
3120 return 0;
3121
3122 error = may_delete(old_dir, old_dentry, is_dir);
3123 if (error)
3124 return error;
3125
3126 if (!new_dentry->d_inode)
a95164d9 3127 error = may_create(new_dir, new_dentry);
1da177e4
LT		 3128 else
3129 error = may_delete(new_dir, new_dentry, is_dir);
3130 if (error)
3131 return error;
3132
acfa4380 3133 if (!old_dir->i_op->rename)
1da177e4
LT		 3134 return -EPERM;
3135
0eeca283
RL		 3136 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3137
1da177e4
LT		 3138 if (is_dir)
3139 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3140 else
3141 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
123df294
AV		 3142 if (!error)
3143 fsnotify_move(old_dir, new_dir, old_name, is_dir,
5a190ae6 3144 new_dentry->d_inode, old_dentry);
0eeca283
RL		 3145 fsnotify_oldname_free(old_name);
3146
1da177e4
LT		 3147 return error;
3148}
3149
2e4d0924
HC		 3150SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3151 int, newdfd, const char __user *, newname)
1da177e4 3152{
2ad94ae6
AV		 3153 struct dentry *old_dir, *new_dir;
3154 struct dentry *old_dentry, *new_dentry;
3155 struct dentry *trap;
1da177e4 3156 struct nameidata oldnd, newnd;
2ad94ae6
AV		 3157 char *from;
3158 char *to;
3159 int error;
1da177e4 3160
2ad94ae6 3161 error = user_path_parent(olddfd, oldname, &oldnd, &from);
1da177e4
LT		 3162 if (error)
3163 goto exit;
3164
2ad94ae6 3165 error = user_path_parent(newdfd, newname, &newnd, &to);
1da177e4
LT		 3166 if (error)
3167 goto exit1;
3168
3169 error = -EXDEV;
4ac91378 3170 if (oldnd.path.mnt != newnd.path.mnt)
1da177e4
LT		 3171 goto exit2;
3172
4ac91378 3173 old_dir = oldnd.path.dentry;
1da177e4
LT		 3174 error = -EBUSY;
3175 if (oldnd.last_type != LAST_NORM)
3176 goto exit2;
3177
4ac91378 3178 new_dir = newnd.path.dentry;
1da177e4
LT		 3179 if (newnd.last_type != LAST_NORM)
3180 goto exit2;
3181
0612d9fb
OH		 3182 oldnd.flags &= ~LOOKUP_PARENT;
3183 newnd.flags &= ~LOOKUP_PARENT;
4e9ed2f8 3184 newnd.flags |= LOOKUP_RENAME_TARGET;
0612d9fb 3185
1da177e4
LT		 3186 trap = lock_rename(new_dir, old_dir);
3187
49705b77 3188 old_dentry = lookup_hash(&oldnd);
1da177e4
LT		 3189 error = PTR_ERR(old_dentry);
3190 if (IS_ERR(old_dentry))
3191 goto exit3;
3192 /* source must exist */
3193 error = -ENOENT;
3194 if (!old_dentry->d_inode)
3195 goto exit4;
3196 /* unless the source is a directory trailing slashes give -ENOTDIR */
3197 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3198 error = -ENOTDIR;
3199 if (oldnd.last.name[oldnd.last.len])
3200 goto exit4;
3201 if (newnd.last.name[newnd.last.len])
3202 goto exit4;
3203 }
3204 /* source should not be ancestor of target */
3205 error = -EINVAL;
3206 if (old_dentry == trap)
3207 goto exit4;
49705b77 3208 new_dentry = lookup_hash(&newnd);
1da177e4
LT		 3209 error = PTR_ERR(new_dentry);
3210 if (IS_ERR(new_dentry))
3211 goto exit4;
3212 /* target should not be an ancestor of source */
3213 error = -ENOTEMPTY;
3214 if (new_dentry == trap)
3215 goto exit5;
3216
9079b1eb
DH		 3217 error = mnt_want_write(oldnd.path.mnt);
3218 if (error)
3219 goto exit5;
be6d3e56
KT		 3220 error = security_path_rename(&oldnd.path, old_dentry,
3221 &newnd.path, new_dentry);
3222 if (error)
3223 goto exit6;
1da177e4
LT		 3224 error = vfs_rename(old_dir->d_inode, old_dentry,
3225 new_dir->d_inode, new_dentry);
be6d3e56 3226exit6:
9079b1eb 3227 mnt_drop_write(oldnd.path.mnt);
1da177e4
LT		 3228exit5:
3229 dput(new_dentry);
3230exit4:
3231 dput(old_dentry);
3232exit3:
3233 unlock_rename(new_dir, old_dir);
3234exit2:
1d957f9b 3235 path_put(&newnd.path);
2ad94ae6 3236 putname(to);
1da177e4 3237exit1:
1d957f9b 3238 path_put(&oldnd.path);
1da177e4 3239 putname(from);
2ad94ae6 3240exit:
1da177e4
LT		 3241 return error;
3242}
3243
a26eab24 3244SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 3245{
3246 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3247}
3248
1da177e4
LT		 3249int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3250{
3251 int len;
3252
3253 len = PTR_ERR(link);
3254 if (IS_ERR(link))
3255 goto out;
3256
3257 len = strlen(link);
3258 if (len > (unsigned) buflen)
3259 len = buflen;
3260 if (copy_to_user(buffer, link, len))
3261 len = -EFAULT;
3262out:
3263 return len;
3264}
3265
3266/*
3267 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3268 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3269 * using) it for any given inode is up to filesystem.
3270 */
3271int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3272{
3273 struct nameidata nd;
cc314eef 3274 void *cookie;
694a1764 3275 int res;
cc314eef 3276
1da177e4 3277 nd.depth = 0;
cc314eef 3278 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
694a1764
MS		 3279 if (IS_ERR(cookie))
3280 return PTR_ERR(cookie);
3281
3282 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3283 if (dentry->d_inode->i_op->put_link)
3284 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3285 return res;
1da177e4
LT		 3286}
3287
3288int vfs_follow_link(struct nameidata *nd, const char *link)
3289{
3290 return __vfs_follow_link(nd, link);
3291}
3292
3293/* get the link contents into pagecache */
3294static char *page_getlink(struct dentry * dentry, struct page **ppage)
3295{
ebd09abb
DG		 3296 char *kaddr;
3297 struct page *page;
1da177e4 3298 struct address_space *mapping = dentry->d_inode->i_mapping;
090d2b18 3299 page = read_mapping_page(mapping, 0, NULL);
1da177e4 3300 if (IS_ERR(page))
6fe6900e 3301 return (char*)page;
1da177e4 3302 *ppage = page;
ebd09abb
DG		 3303 kaddr = kmap(page);
3304 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3305 return kaddr;
1da177e4
LT		 3306}
3307
3308int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3309{
3310 struct page *page = NULL;
3311 char *s = page_getlink(dentry, &page);
3312 int res = vfs_readlink(dentry,buffer,buflen,s);
3313 if (page) {
3314 kunmap(page);
3315 page_cache_release(page);
3316 }
3317 return res;
3318}
3319
cc314eef 3320void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
1da177e4 3321{
cc314eef 3322 struct page *page = NULL;
1da177e4 3323 nd_set_link(nd, page_getlink(dentry, &page));
cc314eef 3324 return page;
1da177e4
LT		 3325}
3326
cc314eef 3327void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1da177e4 3328{
cc314eef
LT		 3329 struct page *page = cookie;
3330
3331 if (page) {
1da177e4
LT		 3332 kunmap(page);
3333 page_cache_release(page);
1da177e4
LT		 3334 }
3335}
3336
54566b2c
NP		 3337/*
3338 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3339 */
3340int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
1da177e4
LT		 3341{
3342 struct address_space *mapping = inode->i_mapping;
0adb25d2 3343 struct page *page;
afddba49 3344 void *fsdata;
beb497ab 3345 int err;
1da177e4 3346 char *kaddr;
54566b2c
NP		 3347 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3348 if (nofs)
3349 flags |= AOP_FLAG_NOFS;
1da177e4 3350
7e53cac4 3351retry:
afddba49 3352 err = pagecache_write_begin(NULL, mapping, 0, len-1,
54566b2c 3353 flags, &page, &fsdata);
1da177e4 3354 if (err)
afddba49
NP		 3355 goto fail;
3356
1da177e4
LT		 3357 kaddr = kmap_atomic(page, KM_USER0);
3358 memcpy(kaddr, symname, len-1);
3359 kunmap_atomic(kaddr, KM_USER0);
afddba49
NP		 3360
3361 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3362 page, fsdata);
1da177e4
LT		 3363 if (err < 0)
3364 goto fail;
afddba49
NP		 3365 if (err < len-1)
3366 goto retry;
3367
1da177e4
LT		 3368 mark_inode_dirty(inode);
3369 return 0;
1da177e4
LT		 3370fail:
3371 return err;
3372}
3373
0adb25d2
KK		 3374int page_symlink(struct inode *inode, const char *symname, int len)
3375{
3376 return __page_symlink(inode, symname, len,
54566b2c 3377 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
0adb25d2
KK		 3378}
3379
92e1d5be 3380const struct inode_operations page_symlink_inode_operations = {
1da177e4
LT		 3381 .readlink = generic_readlink,
3382 .follow_link = page_follow_link_light,
3383 .put_link = page_put_link,
3384};
3385
2d8f3038 3386EXPORT_SYMBOL(user_path_at);
1da177e4
LT		 3387EXPORT_SYMBOL(follow_down);
3388EXPORT_SYMBOL(follow_up);
3389EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3390EXPORT_SYMBOL(getname);
3391EXPORT_SYMBOL(lock_rename);
1da177e4
LT		 3392EXPORT_SYMBOL(lookup_one_len);
3393EXPORT_SYMBOL(page_follow_link_light);
3394EXPORT_SYMBOL(page_put_link);
3395EXPORT_SYMBOL(page_readlink);
0adb25d2 3396EXPORT_SYMBOL(__page_symlink);
1da177e4
LT		 3397EXPORT_SYMBOL(page_symlink);
3398EXPORT_SYMBOL(page_symlink_inode_operations);
3399EXPORT_SYMBOL(path_lookup);
d1811465 3400EXPORT_SYMBOL(kern_path);
16f18200 3401EXPORT_SYMBOL(vfs_path_lookup);
f419a2e3 3402EXPORT_SYMBOL(inode_permission);
8c744fb8 3403EXPORT_SYMBOL(file_permission);
1da177e4
LT		 3404EXPORT_SYMBOL(unlock_rename);
3405EXPORT_SYMBOL(vfs_create);
3406EXPORT_SYMBOL(vfs_follow_link);
3407EXPORT_SYMBOL(vfs_link);
3408EXPORT_SYMBOL(vfs_mkdir);
3409EXPORT_SYMBOL(vfs_mknod);
3410EXPORT_SYMBOL(generic_permission);
3411EXPORT_SYMBOL(vfs_readlink);
3412EXPORT_SYMBOL(vfs_rename);
3413EXPORT_SYMBOL(vfs_rmdir);
3414EXPORT_SYMBOL(vfs_symlink);
3415EXPORT_SYMBOL(vfs_unlink);
3416EXPORT_SYMBOL(dentry_unhash);
3417EXPORT_SYMBOL(generic_readlink);
kernel source for telekom puls (alcatel/mediatek)