projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
New kind of open files - "location only".
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / fs / namei.c
CommitLineData
1da177e4
LT		 1/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
18#include <linux/module.h>
19#include <linux/slab.h>
20#include <linux/fs.h>
21#include <linux/namei.h>
1da177e4 22#include <linux/pagemap.h>
0eeca283 23#include <linux/fsnotify.h>
1da177e4
LT		 24#include <linux/personality.h>
25#include <linux/security.h>
6146f0d5 26#include <linux/ima.h>
1da177e4
LT		 27#include <linux/syscalls.h>
28#include <linux/mount.h>
29#include <linux/audit.h>
16f7e0fe 30#include <linux/capability.h>
834f2a4a 31#include <linux/file.h>
5590ff0d 32#include <linux/fcntl.h>
08ce5f16 33#include <linux/device_cgroup.h>
5ad4e53b 34#include <linux/fs_struct.h>
1da177e4
LT		 35#include <asm/uaccess.h>
36
e81e3f4d
EP		 37#include "internal.h"
38
1da177e4
LT		 39/* [Feb-1997 T. Schoebel-Theuer]
40 * Fundamental changes in the pathname lookup mechanisms (namei)
41 * were necessary because of omirr. The reason is that omirr needs
42 * to know the _real_ pathname, not the user-supplied one, in case
43 * of symlinks (and also when transname replacements occur).
44 *
45 * The new code replaces the old recursive symlink resolution with
46 * an iterative one (in case of non-nested symlink chains). It does
47 * this with calls to <fs>_follow_link().
48 * As a side effect, dir_namei(), _namei() and follow_link() are now
49 * replaced with a single function lookup_dentry() that can handle all
50 * the special cases of the former code.
51 *
52 * With the new dcache, the pathname is stored at each inode, at least as
53 * long as the refcount of the inode is positive. As a side effect, the
54 * size of the dcache depends on the inode cache and thus is dynamic.
55 *
56 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57 * resolution to correspond with current state of the code.
58 *
59 * Note that the symlink resolution is not *completely* iterative.
60 * There is still a significant amount of tail- and mid- recursion in
61 * the algorithm. Also, note that <fs>_readlink() is not used in
62 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63 * may return different results than <fs>_follow_link(). Many virtual
64 * filesystems (including /proc) exhibit this behavior.
65 */
66
67/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69 * and the name already exists in form of a symlink, try to create the new
70 * name indicated by the symlink. The old code always complained that the
71 * name already exists, due to not following the symlink even if its target
72 * is nonexistent. The new semantics affects also mknod() and link() when
73 * the name is a symlink pointing to a non-existant name.
74 *
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
82 */
83
84/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
86 *
87 * [10-Sep-98 Alan Modra] Another symlink change.
88 */
89
90/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
97 *
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
103 */
104/*
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
a11f3a05 106 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
1da177e4
LT		 107 * any extra contention...
108 */
109
110/* In order to reduce some races, while at the same time doing additional
111 * checking and hopefully speeding things up, we copy filenames to the
112 * kernel data space before using them..
113 *
114 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
115 * PATH_MAX includes the nul terminator --RR.
116 */
858119e1 117static int do_getname(const char __user *filename, char *page)
1da177e4
LT		 118{
119 int retval;
120 unsigned long len = PATH_MAX;
121
122 if (!segment_eq(get_fs(), KERNEL_DS)) {
123 if ((unsigned long) filename >= TASK_SIZE)
124 return -EFAULT;
125 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
126 len = TASK_SIZE - (unsigned long) filename;
127 }
128
129 retval = strncpy_from_user(page, filename, len);
130 if (retval > 0) {
131 if (retval < len)
132 return 0;
133 return -ENAMETOOLONG;
134 } else if (!retval)
135 retval = -ENOENT;
136 return retval;
137}
138
f52e0c11 139static char *getname_flags(const char __user * filename, int flags)
1da177e4
LT		 140{
141 char *tmp, *result;
142
143 result = ERR_PTR(-ENOMEM);
144 tmp = __getname();
145 if (tmp) {
146 int retval = do_getname(filename, tmp);
147
148 result = tmp;
149 if (retval < 0) {
f52e0c11
AV		 150 if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) {
151 __putname(tmp);
152 result = ERR_PTR(retval);
153 }
1da177e4
LT		 154 }
155 }
156 audit_getname(result);
157 return result;
158}
159
f52e0c11
AV		 160char *getname(const char __user * filename)
161{
162 return getname_flags(filename, 0);
163}
164
1da177e4
LT		 165#ifdef CONFIG_AUDITSYSCALL
166void putname(const char *name)
167{
5ac3a9c2 168 if (unlikely(!audit_dummy_context()))
1da177e4
LT		 169 audit_putname(name);
170 else
171 __putname(name);
172}
173EXPORT_SYMBOL(putname);
174#endif
175
5909ccaa
LT		 176/*
177 * This does basic POSIX ACL permission checking
1da177e4 178 */
b74c79e9
NP		 179static int acl_permission_check(struct inode *inode, int mask, unsigned int flags,
180 int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
1da177e4
LT		 181{
182 umode_t mode = inode->i_mode;
183
e6305c43
AV		 184 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
185
da9592ed 186 if (current_fsuid() == inode->i_uid)
1da177e4
LT		 187 mode >>= 6;
188 else {
189 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
b74c79e9
NP		 190 int error = check_acl(inode, mask, flags);
191 if (error != -EAGAIN)
192 return error;
1da177e4
LT		 193 }
194
195 if (in_group_p(inode->i_gid))
196 mode >>= 3;
197 }
198
199 /*
200 * If the DACs are ok we don't need any capability check.
201 */
e6305c43 202 if ((mask & ~mode) == 0)
1da177e4 203 return 0;
5909ccaa
LT		 204 return -EACCES;
205}
206
207/**
b74c79e9 208 * generic_permission - check for access rights on a Posix-like filesystem
5909ccaa
LT		 209 * @inode: inode to check access rights for
210 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
211 * @check_acl: optional callback to check for Posix ACLs
39191628 212 * @flags: IPERM_FLAG_ flags.
5909ccaa
LT		 213 *
214 * Used to check for read/write/execute permissions on a file.
215 * We use "fsuid" for this, letting us set arbitrary permissions
216 * for filesystem access without changing the "normal" uids which
b74c79e9
NP		 217 * are used for other things.
218 *
219 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
220 * request cannot be satisfied (eg. requires blocking or too much complexity).
221 * It would then be called again in ref-walk mode.
5909ccaa 222 */
b74c79e9
NP		 223int generic_permission(struct inode *inode, int mask, unsigned int flags,
224 int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
5909ccaa
LT		 225{
226 int ret;
227
228 /*
229 * Do the basic POSIX ACL permission checks.
230 */
b74c79e9 231 ret = acl_permission_check(inode, mask, flags, check_acl);
5909ccaa
LT		 232 if (ret != -EACCES)
233 return ret;
1da177e4 234
1da177e4
LT		 235 /*
236 * Read/write DACs are always overridable.
237 * Executable DACs are overridable if at least one exec bit is set.
238 */
f696a365 239 if (!(mask & MAY_EXEC) || execute_ok(inode))
1da177e4
LT		 240 if (capable(CAP_DAC_OVERRIDE))
241 return 0;
242
243 /*
244 * Searching includes executable on directories, else just read.
245 */
7ea66001 246 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
1da177e4
LT		 247 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
248 if (capable(CAP_DAC_READ_SEARCH))
249 return 0;
250
251 return -EACCES;
252}
253
cb23beb5
CH		 254/**
255 * inode_permission - check for access rights to a given inode
256 * @inode: inode to check permission on
257 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
258 *
259 * Used to check for read/write/execute permissions on an inode.
260 * We use "fsuid" for this, letting us set arbitrary permissions
261 * for filesystem access without changing the "normal" uids which
262 * are used for other things.
263 */
f419a2e3 264int inode_permission(struct inode *inode, int mask)
1da177e4 265{
e6305c43 266 int retval;
1da177e4
LT		 267
268 if (mask & MAY_WRITE) {
22590e41 269 umode_t mode = inode->i_mode;
1da177e4
LT		 270
271 /*
272 * Nobody gets write access to a read-only fs.
273 */
274 if (IS_RDONLY(inode) &&
275 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
276 return -EROFS;
277
278 /*
279 * Nobody gets write access to an immutable file.
280 */
281 if (IS_IMMUTABLE(inode))
282 return -EACCES;
283 }
284
acfa4380 285 if (inode->i_op->permission)
b74c79e9 286 retval = inode->i_op->permission(inode, mask, 0);
f696a365 287 else
b74c79e9
NP		 288 retval = generic_permission(inode, mask, 0,
289 inode->i_op->check_acl);
f696a365 290
1da177e4
LT		 291 if (retval)
292 return retval;
293
08ce5f16
SH		 294 retval = devcgroup_inode_permission(inode, mask);
295 if (retval)
296 return retval;
297
d09ca739 298 return security_inode_permission(inode, mask);
1da177e4
LT		 299}
300
8c744fb8
CH		 301/**
302 * file_permission - check for additional access rights to a given file
303 * @file: file to check access rights for
304 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
305 *
306 * Used to check for read/write/execute permissions on an already opened
307 * file.
308 *
309 * Note:
310 * Do not use this function in new code. All access checks should
cb23beb5 311 * be done using inode_permission().
8c744fb8
CH		 312 */
313int file_permission(struct file *file, int mask)
314{
f419a2e3 315 return inode_permission(file->f_path.dentry->d_inode, mask);
8c744fb8
CH		 316}
317
1da177e4
LT		 318/*
319 * get_write_access() gets write permission for a file.
320 * put_write_access() releases this write permission.
321 * This is used for regular files.
322 * We cannot support write (and maybe mmap read-write shared) accesses and
323 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
324 * can have the following values:
325 * 0: no writers, no VM_DENYWRITE mappings
326 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
327 * > 0: (i_writecount) users are writing to the file.
328 *
329 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
330 * except for the cases where we don't hold i_writecount yet. Then we need to
331 * use {get,deny}_write_access() - these functions check the sign and refuse
332 * to do the change if sign is wrong. Exclusion between them is provided by
333 * the inode->i_lock spinlock.
334 */
335
336int get_write_access(struct inode * inode)
337{
338 spin_lock(&inode->i_lock);
339 if (atomic_read(&inode->i_writecount) < 0) {
340 spin_unlock(&inode->i_lock);
341 return -ETXTBSY;
342 }
343 atomic_inc(&inode->i_writecount);
344 spin_unlock(&inode->i_lock);
345
346 return 0;
347}
348
349int deny_write_access(struct file * file)
350{
0f7fc9e4 351 struct inode *inode = file->f_path.dentry->d_inode;
1da177e4
LT		 352
353 spin_lock(&inode->i_lock);
354 if (atomic_read(&inode->i_writecount) > 0) {
355 spin_unlock(&inode->i_lock);
356 return -ETXTBSY;
357 }
358 atomic_dec(&inode->i_writecount);
359 spin_unlock(&inode->i_lock);
360
361 return 0;
362}
363
5dd784d0
JB		 364/**
365 * path_get - get a reference to a path
366 * @path: path to get the reference to
367 *
368 * Given a path increment the reference count to the dentry and the vfsmount.
369 */
370void path_get(struct path *path)
371{
372 mntget(path->mnt);
373 dget(path->dentry);
374}
375EXPORT_SYMBOL(path_get);
376
1d957f9b
JB		 377/**
378 * path_put - put a reference to a path
379 * @path: path to put the reference to
380 *
381 * Given a path decrement the reference count to the dentry and the vfsmount.
382 */
383void path_put(struct path *path)
1da177e4 384{
1d957f9b
JB		 385 dput(path->dentry);
386 mntput(path->mnt);
1da177e4 387}
1d957f9b 388EXPORT_SYMBOL(path_put);
1da177e4 389
31e6b01f
NP		 390/**
391 * nameidata_drop_rcu - drop this nameidata out of rcu-walk
392 * @nd: nameidata pathwalk data to drop
39191628 393 * Returns: 0 on success, -ECHILD on failure
31e6b01f
NP		 394 *
395 * Path walking has 2 modes, rcu-walk and ref-walk (see
396 * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt
397 * to drop out of rcu-walk mode and take normal reference counts on dentries
398 * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take
399 * refcounts at the last known good point before rcu-walk got stuck, so
400 * ref-walk may continue from there. If this is not successful (eg. a seqcount
401 * has changed), then failure is returned and path walk restarts from the
402 * beginning in ref-walk mode.
403 *
404 * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into
405 * ref-walk. Must be called from rcu-walk context.
406 */
407static int nameidata_drop_rcu(struct nameidata *nd)
408{
409 struct fs_struct *fs = current->fs;
410 struct dentry *dentry = nd->path.dentry;
5b6ca027 411 int want_root = 0;
31e6b01f
NP		 412
413 BUG_ON(!(nd->flags & LOOKUP_RCU));
5b6ca027
AV		 414 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
415 want_root = 1;
31e6b01f
NP		 416 spin_lock(&fs->lock);
417 if (nd->root.mnt != fs->root.mnt ||
418 nd->root.dentry != fs->root.dentry)
419 goto err_root;
420 }
421 spin_lock(&dentry->d_lock);
422 if (!__d_rcu_to_refcount(dentry, nd->seq))
423 goto err;
424 BUG_ON(nd->inode != dentry->d_inode);
425 spin_unlock(&dentry->d_lock);
5b6ca027 426 if (want_root) {
31e6b01f
NP		 427 path_get(&nd->root);
428 spin_unlock(&fs->lock);
429 }
430 mntget(nd->path.mnt);
431
432 rcu_read_unlock();
433 br_read_unlock(vfsmount_lock);
434 nd->flags &= ~LOOKUP_RCU;
435 return 0;
436err:
437 spin_unlock(&dentry->d_lock);
438err_root:
5b6ca027 439 if (want_root)
31e6b01f
NP		 440 spin_unlock(&fs->lock);
441 return -ECHILD;
442}
443
444/* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
445static inline int nameidata_drop_rcu_maybe(struct nameidata *nd)
446{
447 if (nd->flags & LOOKUP_RCU)
448 return nameidata_drop_rcu(nd);
449 return 0;
450}
451
452/**
453 * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk
454 * @nd: nameidata pathwalk data to drop
455 * @dentry: dentry to drop
39191628 456 * Returns: 0 on success, -ECHILD on failure
31e6b01f
NP		 457 *
458 * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root,
459 * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on
460 * @nd. Must be called from rcu-walk context.
461 */
462static int nameidata_dentry_drop_rcu(struct nameidata *nd, struct dentry *dentry)
463{
464 struct fs_struct *fs = current->fs;
465 struct dentry *parent = nd->path.dentry;
5b6ca027 466 int want_root = 0;
31e6b01f
NP		 467
468 BUG_ON(!(nd->flags & LOOKUP_RCU));
5b6ca027
AV		 469 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
470 want_root = 1;
31e6b01f
NP		 471 spin_lock(&fs->lock);
472 if (nd->root.mnt != fs->root.mnt ||
473 nd->root.dentry != fs->root.dentry)
474 goto err_root;
475 }
476 spin_lock(&parent->d_lock);
477 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
478 if (!__d_rcu_to_refcount(dentry, nd->seq))
479 goto err;
480 /*
481 * If the sequence check on the child dentry passed, then the child has
482 * not been removed from its parent. This means the parent dentry must
483 * be valid and able to take a reference at this point.
484 */
485 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
486 BUG_ON(!parent->d_count);
487 parent->d_count++;
488 spin_unlock(&dentry->d_lock);
489 spin_unlock(&parent->d_lock);
5b6ca027 490 if (want_root) {
31e6b01f
NP		 491 path_get(&nd->root);
492 spin_unlock(&fs->lock);
493 }
494 mntget(nd->path.mnt);
495
496 rcu_read_unlock();
497 br_read_unlock(vfsmount_lock);
498 nd->flags &= ~LOOKUP_RCU;
499 return 0;
500err:
501 spin_unlock(&dentry->d_lock);
502 spin_unlock(&parent->d_lock);
503err_root:
5b6ca027 504 if (want_root)
31e6b01f
NP		 505 spin_unlock(&fs->lock);
506 return -ECHILD;
507}
508
509/* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
510static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata *nd, struct dentry *dentry)
511{
a7472bab
AV		 512 if (nd->flags & LOOKUP_RCU) {
513 if (unlikely(nameidata_dentry_drop_rcu(nd, dentry))) {
514 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 515 if (!(nd->flags & LOOKUP_ROOT))
516 nd->root.mnt = NULL;
a7472bab
AV		 517 rcu_read_unlock();
518 br_read_unlock(vfsmount_lock);
519 return -ECHILD;
520 }
521 }
31e6b01f
NP		 522 return 0;
523}
524
525/**
526 * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk
527 * @nd: nameidata pathwalk data to drop
39191628 528 * Returns: 0 on success, -ECHILD on failure
31e6b01f
NP		 529 *
530 * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk.
531 * nd->path should be the final element of the lookup, so nd->root is discarded.
532 * Must be called from rcu-walk context.
533 */
534static int nameidata_drop_rcu_last(struct nameidata *nd)
535{
536 struct dentry *dentry = nd->path.dentry;
537
538 BUG_ON(!(nd->flags & LOOKUP_RCU));
539 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 540 if (!(nd->flags & LOOKUP_ROOT))
541 nd->root.mnt = NULL;
31e6b01f
NP		 542 spin_lock(&dentry->d_lock);
543 if (!__d_rcu_to_refcount(dentry, nd->seq))
544 goto err_unlock;
545 BUG_ON(nd->inode != dentry->d_inode);
546 spin_unlock(&dentry->d_lock);
547
548 mntget(nd->path.mnt);
549
550 rcu_read_unlock();
551 br_read_unlock(vfsmount_lock);
552
553 return 0;
554
555err_unlock:
556 spin_unlock(&dentry->d_lock);
557 rcu_read_unlock();
558 br_read_unlock(vfsmount_lock);
559 return -ECHILD;
560}
561
834f2a4a
TM		 562/**
563 * release_open_intent - free up open intent resources
564 * @nd: pointer to nameidata
565 */
566void release_open_intent(struct nameidata *nd)
567{
2dab5974
LT		 568 struct file *file = nd->intent.open.file;
569
570 if (file && !IS_ERR(file)) {
571 if (file->f_path.dentry == NULL)
572 put_filp(file);
573 else
574 fput(file);
575 }
834f2a4a
TM		 576}
577
f60aef7e 578static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
34286d66 579{
f60aef7e 580 return dentry->d_op->d_revalidate(dentry, nd);
34286d66
NP		 581}
582
f5e1c1c1 583static struct dentry *
bcdc5e01
IK		 584do_revalidate(struct dentry *dentry, struct nameidata *nd)
585{
f5e1c1c1 586 int status = d_revalidate(dentry, nd);
bcdc5e01
IK		 587 if (unlikely(status <= 0)) {
588 /*
589 * The dentry failed validation.
590 * If d_revalidate returned 0 attempt to invalidate
591 * the dentry otherwise d_revalidate is asking us
592 * to return a fail status.
593 */
34286d66 594 if (status < 0) {
f5e1c1c1 595 dput(dentry);
34286d66 596 dentry = ERR_PTR(status);
f5e1c1c1
AV		 597 } else if (!d_invalidate(dentry)) {
598 dput(dentry);
599 dentry = NULL;
bcdc5e01
IK		 600 }
601 }
602 return dentry;
603}
604
39159de2 605/*
16c2cd71 606 * handle_reval_path - force revalidation of a dentry
39159de2
JL		 607 *
608 * In some situations the path walking code will trust dentries without
609 * revalidating them. This causes problems for filesystems that depend on
610 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
611 * (which indicates that it's possible for the dentry to go stale), force
612 * a d_revalidate call before proceeding.
613 *
614 * Returns 0 if the revalidation was successful. If the revalidation fails,
615 * either return the error returned by d_revalidate or -ESTALE if the
616 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
617 * invalidate the dentry. It's up to the caller to handle putting references
618 * to the path if necessary.
619 */
16c2cd71 620static inline int handle_reval_path(struct nameidata *nd)
39159de2 621{
16c2cd71 622 struct dentry *dentry = nd->path.dentry;
39159de2 623 int status;
39159de2 624
16c2cd71
AV		 625 if (likely(!(nd->flags & LOOKUP_JUMPED)))
626 return 0;
627
628 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
39159de2
JL		 629 return 0;
630
16c2cd71
AV		 631 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
632 return 0;
633
634 /* Note: we do not d_invalidate() */
34286d66 635 status = d_revalidate(dentry, nd);
39159de2
JL		 636 if (status > 0)
637 return 0;
638
16c2cd71 639 if (!status)
39159de2 640 status = -ESTALE;
16c2cd71 641
39159de2
JL		 642 return status;
643}
644
1da177e4 645/*
b75b5086
AV		 646 * Short-cut version of permission(), for calling on directories
647 * during pathname resolution. Combines parts of permission()
648 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
1da177e4
LT		 649 *
650 * If appropriate, check DAC only. If not appropriate, or
b75b5086 651 * short-cut DAC fails, then call ->permission() to do more
1da177e4
LT		 652 * complete permission check.
653 */
b74c79e9 654static inline int exec_permission(struct inode *inode, unsigned int flags)
1da177e4 655{
5909ccaa 656 int ret;
1da177e4 657
cb9179ea 658 if (inode->i_op->permission) {
b74c79e9
NP		 659 ret = inode->i_op->permission(inode, MAY_EXEC, flags);
660 } else {
661 ret = acl_permission_check(inode, MAY_EXEC, flags,
662 inode->i_op->check_acl);
cb9179ea 663 }
b74c79e9 664 if (likely(!ret))
1da177e4 665 goto ok;
b74c79e9 666 if (ret == -ECHILD)
31e6b01f 667 return ret;
1da177e4 668
f1ac9f6b 669 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
1da177e4
LT		 670 goto ok;
671
5909ccaa 672 return ret;
1da177e4 673ok:
b74c79e9 674 return security_inode_exec_permission(inode, flags);
1da177e4
LT		 675}
676
2a737871
AV		 677static __always_inline void set_root(struct nameidata *nd)
678{
f7ad3c6b
MS		 679 if (!nd->root.mnt)
680 get_fs_root(current->fs, &nd->root);
2a737871
AV		 681}
682
6de88d72
AV		 683static int link_path_walk(const char *, struct nameidata *);
684
31e6b01f
NP		 685static __always_inline void set_root_rcu(struct nameidata *nd)
686{
687 if (!nd->root.mnt) {
688 struct fs_struct *fs = current->fs;
c28cc364
NP		 689 unsigned seq;
690
691 do {
692 seq = read_seqcount_begin(&fs->seq);
693 nd->root = fs->root;
694 } while (read_seqcount_retry(&fs->seq, seq));
31e6b01f
NP		 695 }
696}
697
f1662356 698static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
1da177e4 699{
31e6b01f
NP		 700 int ret;
701
1da177e4
LT		 702 if (IS_ERR(link))
703 goto fail;
704
705 if (*link == '/') {
2a737871 706 set_root(nd);
1d957f9b 707 path_put(&nd->path);
2a737871
AV		 708 nd->path = nd->root;
709 path_get(&nd->root);
16c2cd71 710 nd->flags |= LOOKUP_JUMPED;
1da177e4 711 }
31e6b01f 712 nd->inode = nd->path.dentry->d_inode;
b4091d5f 713
31e6b01f
NP		 714 ret = link_path_walk(link, nd);
715 return ret;
1da177e4 716fail:
1d957f9b 717 path_put(&nd->path);
1da177e4
LT		 718 return PTR_ERR(link);
719}
720
1d957f9b 721static void path_put_conditional(struct path *path, struct nameidata *nd)
051d3812
IK		 722{
723 dput(path->dentry);
4ac91378 724 if (path->mnt != nd->path.mnt)
051d3812
IK		 725 mntput(path->mnt);
726}
727
7b9337aa
NP		 728static inline void path_to_nameidata(const struct path *path,
729 struct nameidata *nd)
051d3812 730{
31e6b01f
NP		 731 if (!(nd->flags & LOOKUP_RCU)) {
732 dput(nd->path.dentry);
733 if (nd->path.mnt != path->mnt)
734 mntput(nd->path.mnt);
9a229683 735 }
31e6b01f 736 nd->path.mnt = path->mnt;
4ac91378 737 nd->path.dentry = path->dentry;
051d3812
IK		 738}
739
def4af30 740static __always_inline int
7b9337aa 741__do_follow_link(const struct path *link, struct nameidata *nd, void **p)
1da177e4
LT		 742{
743 int error;
7b9337aa 744 struct dentry *dentry = link->dentry;
1da177e4 745
844a3917
AV		 746 BUG_ON(nd->flags & LOOKUP_RCU);
747
7b9337aa 748 touch_atime(link->mnt, dentry);
1da177e4 749 nd_set_link(nd, NULL);
cd4e91d3 750
87556ef1
DH		 751 if (link->mnt == nd->path.mnt)
752 mntget(link->mnt);
31e6b01f 753
36f3b4f6
AV		 754 error = security_inode_follow_link(link->dentry, nd);
755 if (error) {
756 *p = ERR_PTR(error); /* no ->put_link(), please */
757 path_put(&nd->path);
758 return error;
759 }
760
86acdca1 761 nd->last_type = LAST_BIND;
def4af30
AV		 762 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
763 error = PTR_ERR(*p);
764 if (!IS_ERR(*p)) {
1da177e4 765 char *s = nd_get_link(nd);
cc314eef 766 error = 0;
1da177e4
LT		 767 if (s)
768 error = __vfs_follow_link(nd, s);
16c2cd71
AV		 769 else if (nd->last_type == LAST_BIND)
770 nd->flags |= LOOKUP_JUMPED;
1da177e4 771 }
1da177e4
LT		 772 return error;
773}
774
775/*
776 * This limits recursive symlink follows to 8, while
777 * limiting consecutive symlinks to 40.
778 *
779 * Without that kind of total limit, nasty chains of consecutive
780 * symlinks can cause almost arbitrarily long lookups.
781 */
3abb17e8 782static inline int do_follow_link(struct inode *inode, struct path *path, struct nameidata *nd)
1da177e4 783{
def4af30 784 void *cookie;
1da177e4 785 int err = -ELOOP;
844a3917
AV		 786
787 /* We drop rcu-walk here */
788 if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry))
789 return -ECHILD;
3abb17e8 790 BUG_ON(inode != path->dentry->d_inode);
844a3917 791
1da177e4
LT		 792 if (current->link_count >= MAX_NESTED_LINKS)
793 goto loop;
794 if (current->total_link_count >= 40)
795 goto loop;
796 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
797 cond_resched();
1da177e4
LT		 798 current->link_count++;
799 current->total_link_count++;
800 nd->depth++;
def4af30
AV		 801 err = __do_follow_link(path, nd, &cookie);
802 if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link)
803 path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie);
258fa999 804 path_put(path);
839d9f93
AV		 805 current->link_count--;
806 nd->depth--;
1da177e4
LT		 807 return err;
808loop:
1d957f9b
JB		 809 path_put_conditional(path, nd);
810 path_put(&nd->path);
1da177e4
LT		 811 return err;
812}
813
31e6b01f
NP		 814static int follow_up_rcu(struct path *path)
815{
816 struct vfsmount *parent;
817 struct dentry *mountpoint;
818
819 parent = path->mnt->mnt_parent;
820 if (parent == path->mnt)
821 return 0;
822 mountpoint = path->mnt->mnt_mountpoint;
823 path->dentry = mountpoint;
824 path->mnt = parent;
825 return 1;
826}
827
bab77ebf 828int follow_up(struct path *path)
1da177e4
LT		 829{
830 struct vfsmount *parent;
831 struct dentry *mountpoint;
99b7db7b
NP		 832
833 br_read_lock(vfsmount_lock);
bab77ebf
AV		 834 parent = path->mnt->mnt_parent;
835 if (parent == path->mnt) {
99b7db7b 836 br_read_unlock(vfsmount_lock);
1da177e4
LT		 837 return 0;
838 }
839 mntget(parent);
bab77ebf 840 mountpoint = dget(path->mnt->mnt_mountpoint);
99b7db7b 841 br_read_unlock(vfsmount_lock);
bab77ebf
AV		 842 dput(path->dentry);
843 path->dentry = mountpoint;
844 mntput(path->mnt);
845 path->mnt = parent;
1da177e4
LT		 846 return 1;
847}
848
b5c84bf6 849/*
9875cf80
DH		 850 * Perform an automount
851 * - return -EISDIR to tell follow_managed() to stop and return the path we
852 * were called with.
1da177e4 853 */
9875cf80
DH		 854static int follow_automount(struct path *path, unsigned flags,
855 bool *need_mntput)
31e6b01f 856{
9875cf80 857 struct vfsmount *mnt;
ea5b778a 858 int err;
9875cf80
DH		 859
860 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
861 return -EREMOTE;
862
6f45b656
DH		 863 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
864 * and this is the terminal part of the path.
865 */
866 if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE))
867 return -EISDIR; /* we actually want to stop here */
868
9875cf80
DH		 869 /* We want to mount if someone is trying to open/create a file of any
870 * type under the mountpoint, wants to traverse through the mountpoint
871 * or wants to open the mounted directory.
872 *
873 * We don't want to mount if someone's just doing a stat and they've
874 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
875 * appended a '/' to the name.
876 */
877 if (!(flags & LOOKUP_FOLLOW) &&
878 !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY |
879 LOOKUP_OPEN | LOOKUP_CREATE)))
880 return -EISDIR;
881
882 current->total_link_count++;
883 if (current->total_link_count >= 40)
884 return -ELOOP;
885
886 mnt = path->dentry->d_op->d_automount(path);
887 if (IS_ERR(mnt)) {
888 /*
889 * The filesystem is allowed to return -EISDIR here to indicate
890 * it doesn't want to automount. For instance, autofs would do
891 * this so that its userspace daemon can mount on this dentry.
892 *
893 * However, we can only permit this if it's a terminal point in
894 * the path being looked up; if it wasn't then the remainder of
895 * the path is inaccessible and we should say so.
896 */
897 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE))
898 return -EREMOTE;
899 return PTR_ERR(mnt);
31e6b01f 900 }
ea5b778a 901
9875cf80
DH		 902 if (!mnt) /* mount collision */
903 return 0;
31e6b01f 904
19a167af 905 err = finish_automount(mnt, path);
9875cf80 906
ea5b778a
DH		 907 switch (err) {
908 case -EBUSY:
909 /* Someone else made a mount here whilst we were busy */
19a167af 910 return 0;
ea5b778a 911 case 0:
ea5b778a
DH		 912 dput(path->dentry);
913 if (*need_mntput)
914 mntput(path->mnt);
915 path->mnt = mnt;
916 path->dentry = dget(mnt->mnt_root);
917 *need_mntput = true;
918 return 0;
19a167af
AV		 919 default:
920 return err;
ea5b778a 921 }
19a167af 922
463ffb2e
AV		 923}
924
9875cf80
DH		 925/*
926 * Handle a dentry that is managed in some way.
cc53ce53 927 * - Flagged for transit management (autofs)
9875cf80
DH		 928 * - Flagged as mountpoint
929 * - Flagged as automount point
930 *
931 * This may only be called in refwalk mode.
932 *
933 * Serialization is taken care of in namespace.c
934 */
935static int follow_managed(struct path *path, unsigned flags)
1da177e4 936{
9875cf80
DH		 937 unsigned managed;
938 bool need_mntput = false;
939 int ret;
940
941 /* Given that we're not holding a lock here, we retain the value in a
942 * local variable for each dentry as we look at it so that we don't see
943 * the components of that value change under us */
944 while (managed = ACCESS_ONCE(path->dentry->d_flags),
945 managed &= DCACHE_MANAGED_DENTRY,
946 unlikely(managed != 0)) {
cc53ce53
DH		 947 /* Allow the filesystem to manage the transit without i_mutex
948 * being held. */
949 if (managed & DCACHE_MANAGE_TRANSIT) {
950 BUG_ON(!path->dentry->d_op);
951 BUG_ON(!path->dentry->d_op->d_manage);
ab90911f
DH		 952 ret = path->dentry->d_op->d_manage(path->dentry,
953 false, false);
cc53ce53
DH		 954 if (ret < 0)
955 return ret == -EISDIR ? 0 : ret;
956 }
957
9875cf80
DH		 958 /* Transit to a mounted filesystem. */
959 if (managed & DCACHE_MOUNTED) {
960 struct vfsmount *mounted = lookup_mnt(path);
961 if (mounted) {
962 dput(path->dentry);
963 if (need_mntput)
964 mntput(path->mnt);
965 path->mnt = mounted;
966 path->dentry = dget(mounted->mnt_root);
967 need_mntput = true;
968 continue;
969 }
970
971 /* Something is mounted on this dentry in another
972 * namespace and/or whatever was mounted there in this
973 * namespace got unmounted before we managed to get the
974 * vfsmount_lock */
975 }
976
977 /* Handle an automount point */
978 if (managed & DCACHE_NEED_AUTOMOUNT) {
979 ret = follow_automount(path, flags, &need_mntput);
980 if (ret < 0)
981 return ret == -EISDIR ? 0 : ret;
982 continue;
983 }
984
985 /* We didn't change the current path point */
986 break;
1da177e4 987 }
9875cf80 988 return 0;
1da177e4
LT		 989}
990
cc53ce53 991int follow_down_one(struct path *path)
1da177e4
LT		 992{
993 struct vfsmount *mounted;
994
1c755af4 995 mounted = lookup_mnt(path);
1da177e4 996 if (mounted) {
9393bd07
AV		 997 dput(path->dentry);
998 mntput(path->mnt);
999 path->mnt = mounted;
1000 path->dentry = dget(mounted->mnt_root);
1da177e4
LT		 1001 return 1;
1002 }
1003 return 0;
1004}
1005
9875cf80
DH		 1006/*
1007 * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we
cc53ce53 1008 * meet a managed dentry and we're not walking to "..". True is returned to
9875cf80
DH		 1009 * continue, false to abort.
1010 */
1011static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1012 struct inode **inode, bool reverse_transit)
1013{
1014 while (d_mountpoint(path->dentry)) {
1015 struct vfsmount *mounted;
ab90911f
DH		 1016 if (unlikely(path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) &&
1017 !reverse_transit &&
1018 path->dentry->d_op->d_manage(path->dentry, false, true) < 0)
1019 return false;
9875cf80
DH		 1020 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
1021 if (!mounted)
1022 break;
1023 path->mnt = mounted;
1024 path->dentry = mounted->mnt_root;
1025 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1026 *inode = path->dentry->d_inode;
1027 }
1028
1029 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1030 return reverse_transit;
1031 return true;
1032}
1033
31e6b01f
NP		 1034static int follow_dotdot_rcu(struct nameidata *nd)
1035{
1036 struct inode *inode = nd->inode;
1037
1038 set_root_rcu(nd);
1039
9875cf80 1040 while (1) {
31e6b01f
NP		 1041 if (nd->path.dentry == nd->root.dentry &&
1042 nd->path.mnt == nd->root.mnt) {
1043 break;
1044 }
1045 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1046 struct dentry *old = nd->path.dentry;
1047 struct dentry *parent = old->d_parent;
1048 unsigned seq;
1049
1050 seq = read_seqcount_begin(&parent->d_seq);
1051 if (read_seqcount_retry(&old->d_seq, nd->seq))
ef7562d5 1052 goto failed;
31e6b01f
NP		 1053 inode = parent->d_inode;
1054 nd->path.dentry = parent;
1055 nd->seq = seq;
1056 break;
1057 }
1058 if (!follow_up_rcu(&nd->path))
1059 break;
1060 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1061 inode = nd->path.dentry->d_inode;
1062 }
9875cf80 1063 __follow_mount_rcu(nd, &nd->path, &inode, true);
31e6b01f 1064 nd->inode = inode;
31e6b01f 1065 return 0;
ef7562d5
AV		 1066
1067failed:
1068 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 1069 if (!(nd->flags & LOOKUP_ROOT))
1070 nd->root.mnt = NULL;
ef7562d5
AV		 1071 rcu_read_unlock();
1072 br_read_unlock(vfsmount_lock);
1073 return -ECHILD;
31e6b01f
NP		 1074}
1075
cc53ce53
DH		 1076/*
1077 * Follow down to the covering mount currently visible to userspace. At each
1078 * point, the filesystem owning that dentry may be queried as to whether the
1079 * caller is permitted to proceed or not.
1080 *
1081 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1082 * being true).
1083 */
1084int follow_down(struct path *path, bool mounting_here)
1085{
1086 unsigned managed;
1087 int ret;
1088
1089 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1090 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1091 /* Allow the filesystem to manage the transit without i_mutex
1092 * being held.
1093 *
1094 * We indicate to the filesystem if someone is trying to mount
1095 * something here. This gives autofs the chance to deny anyone
1096 * other than its daemon the right to mount on its
1097 * superstructure.
1098 *
1099 * The filesystem may sleep at this point.
1100 */
1101 if (managed & DCACHE_MANAGE_TRANSIT) {
1102 BUG_ON(!path->dentry->d_op);
1103 BUG_ON(!path->dentry->d_op->d_manage);
ab90911f
DH		 1104 ret = path->dentry->d_op->d_manage(
1105 path->dentry, mounting_here, false);
cc53ce53
DH		 1106 if (ret < 0)
1107 return ret == -EISDIR ? 0 : ret;
1108 }
1109
1110 /* Transit to a mounted filesystem. */
1111 if (managed & DCACHE_MOUNTED) {
1112 struct vfsmount *mounted = lookup_mnt(path);
1113 if (!mounted)
1114 break;
1115 dput(path->dentry);
1116 mntput(path->mnt);
1117 path->mnt = mounted;
1118 path->dentry = dget(mounted->mnt_root);
1119 continue;
1120 }
1121
1122 /* Don't handle automount points here */
1123 break;
1124 }
1125 return 0;
1126}
1127
9875cf80
DH		 1128/*
1129 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1130 */
1131static void follow_mount(struct path *path)
1132{
1133 while (d_mountpoint(path->dentry)) {
1134 struct vfsmount *mounted = lookup_mnt(path);
1135 if (!mounted)
1136 break;
1137 dput(path->dentry);
1138 mntput(path->mnt);
1139 path->mnt = mounted;
1140 path->dentry = dget(mounted->mnt_root);
1141 }
1142}
1143
31e6b01f 1144static void follow_dotdot(struct nameidata *nd)
1da177e4 1145{
2a737871 1146 set_root(nd);
e518ddb7 1147
1da177e4 1148 while(1) {
4ac91378 1149 struct dentry *old = nd->path.dentry;
1da177e4 1150
2a737871
AV		 1151 if (nd->path.dentry == nd->root.dentry &&
1152 nd->path.mnt == nd->root.mnt) {
1da177e4
LT		 1153 break;
1154 }
4ac91378 1155 if (nd->path.dentry != nd->path.mnt->mnt_root) {
3088dd70
AV		 1156 /* rare case of legitimate dget_parent()... */
1157 nd->path.dentry = dget_parent(nd->path.dentry);
1da177e4
LT		 1158 dput(old);
1159 break;
1160 }
3088dd70 1161 if (!follow_up(&nd->path))
1da177e4 1162 break;
1da177e4 1163 }
79ed0226 1164 follow_mount(&nd->path);
31e6b01f 1165 nd->inode = nd->path.dentry->d_inode;
1da177e4
LT		 1166}
1167
baa03890
NP		 1168/*
1169 * Allocate a dentry with name and parent, and perform a parent
1170 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1171 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1172 * have verified that no child exists while under i_mutex.
1173 */
1174static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1175 struct qstr *name, struct nameidata *nd)
1176{
1177 struct inode *inode = parent->d_inode;
1178 struct dentry *dentry;
1179 struct dentry *old;
1180
1181 /* Don't create child dentry for a dead directory. */
1182 if (unlikely(IS_DEADDIR(inode)))
1183 return ERR_PTR(-ENOENT);
1184
1185 dentry = d_alloc(parent, name);
1186 if (unlikely(!dentry))
1187 return ERR_PTR(-ENOMEM);
1188
1189 old = inode->i_op->lookup(inode, dentry, nd);
1190 if (unlikely(old)) {
1191 dput(dentry);
1192 dentry = old;
1193 }
1194 return dentry;
1195}
1196
1da177e4
LT		 1197/*
1198 * It's more convoluted than I'd like it to be, but... it's still fairly
1199 * small and for now I'd prefer to have fast path as straight as possible.
1200 * It _is_ time-critical.
1201 */
1202static int do_lookup(struct nameidata *nd, struct qstr *name,
31e6b01f 1203 struct path *path, struct inode **inode)
1da177e4 1204{
4ac91378 1205 struct vfsmount *mnt = nd->path.mnt;
31e6b01f 1206 struct dentry *dentry, *parent = nd->path.dentry;
5a18fff2
AV		 1207 int need_reval = 1;
1208 int status = 1;
9875cf80
DH		 1209 int err;
1210
b04f784e
NP		 1211 /*
1212 * Rename seqlock is not required here because in the off chance
1213 * of a false negative due to a concurrent rename, we're going to
1214 * do the non-racy lookup, below.
1215 */
31e6b01f
NP		 1216 if (nd->flags & LOOKUP_RCU) {
1217 unsigned seq;
31e6b01f
NP		 1218 *inode = nd->inode;
1219 dentry = __d_lookup_rcu(parent, name, &seq, inode);
5a18fff2
AV		 1220 if (!dentry)
1221 goto unlazy;
1222
31e6b01f
NP		 1223 /* Memory barrier in read_seqcount_begin of child is enough */
1224 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1225 return -ECHILD;
31e6b01f 1226 nd->seq = seq;
5a18fff2 1227
24643087 1228 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
5a18fff2
AV		 1229 status = d_revalidate(dentry, nd);
1230 if (unlikely(status <= 0)) {
1231 if (status != -ECHILD)
1232 need_reval = 0;
1233 goto unlazy;
1234 }
24643087 1235 }
31e6b01f
NP		 1236 path->mnt = mnt;
1237 path->dentry = dentry;
9875cf80
DH		 1238 if (likely(__follow_mount_rcu(nd, path, inode, false)))
1239 return 0;
5a18fff2
AV		 1240unlazy:
1241 if (dentry) {
1242 if (nameidata_dentry_drop_rcu(nd, dentry))
1243 return -ECHILD;
1244 } else {
1245 if (nameidata_drop_rcu(nd))
1246 return -ECHILD;
1247 }
1248 } else {
1249 dentry = __d_lookup(parent, name);
9875cf80 1250 }
5a18fff2
AV		 1251
1252retry:
1253 if (unlikely(!dentry)) {
1254 struct inode *dir = parent->d_inode;
1255 BUG_ON(nd->inode != dir);
1256
1257 mutex_lock(&dir->i_mutex);
1258 dentry = d_lookup(parent, name);
1259 if (likely(!dentry)) {
1260 dentry = d_alloc_and_lookup(parent, name, nd);
1261 if (IS_ERR(dentry)) {
1262 mutex_unlock(&dir->i_mutex);
1263 return PTR_ERR(dentry);
1264 }
1265 /* known good */
1266 need_reval = 0;
1267 status = 1;
1268 }
1269 mutex_unlock(&dir->i_mutex);
1270 }
1271 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1272 status = d_revalidate(dentry, nd);
1273 if (unlikely(status <= 0)) {
1274 if (status < 0) {
1275 dput(dentry);
1276 return status;
1277 }
1278 if (!d_invalidate(dentry)) {
1279 dput(dentry);
1280 dentry = NULL;
1281 need_reval = 1;
1282 goto retry;
1283 }
24643087 1284 }
5a18fff2 1285
9875cf80
DH		 1286 path->mnt = mnt;
1287 path->dentry = dentry;
1288 err = follow_managed(path, nd->flags);
89312214
IK		 1289 if (unlikely(err < 0)) {
1290 path_put_conditional(path, nd);
9875cf80 1291 return err;
89312214 1292 }
9875cf80 1293 *inode = path->dentry->d_inode;
1da177e4 1294 return 0;
1da177e4
LT		 1295}
1296
52094c8a
AV		 1297static inline int may_lookup(struct nameidata *nd)
1298{
1299 if (nd->flags & LOOKUP_RCU) {
1300 int err = exec_permission(nd->inode, IPERM_FLAG_RCU);
1301 if (err != -ECHILD)
1302 return err;
1303 if (nameidata_drop_rcu(nd))
1304 return -ECHILD;
1305 }
1306 return exec_permission(nd->inode, 0);
1307}
1308
9856fa1b
AV		 1309static inline int handle_dots(struct nameidata *nd, int type)
1310{
1311 if (type == LAST_DOTDOT) {
1312 if (nd->flags & LOOKUP_RCU) {
1313 if (follow_dotdot_rcu(nd))
1314 return -ECHILD;
1315 } else
1316 follow_dotdot(nd);
1317 }
1318 return 0;
1319}
1320
951361f9
AV		 1321static void terminate_walk(struct nameidata *nd)
1322{
1323 if (!(nd->flags & LOOKUP_RCU)) {
1324 path_put(&nd->path);
1325 } else {
1326 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 1327 if (!(nd->flags & LOOKUP_ROOT))
1328 nd->root.mnt = NULL;
951361f9
AV		 1329 rcu_read_unlock();
1330 br_read_unlock(vfsmount_lock);
1331 }
1332}
1333
1da177e4
LT		 1334/*
1335 * Name resolution.
ea3834d9
PM		 1336 * This is the basic name resolution function, turning a pathname into
1337 * the final dentry. We expect 'base' to be positive and a directory.
1da177e4 1338 *
ea3834d9
PM		 1339 * Returns 0 and nd will have valid dentry and mnt on success.
1340 * Returns error and drops reference to input namei data on failure.
1da177e4 1341 */
6de88d72 1342static int link_path_walk(const char *name, struct nameidata *nd)
1da177e4
LT		 1343{
1344 struct path next;
1da177e4
LT		 1345 int err;
1346 unsigned int lookup_flags = nd->flags;
1347
1348 while (*name=='/')
1349 name++;
1350 if (!*name)
086e183a 1351 return 0;
1da177e4 1352
1da177e4 1353 if (nd->depth)
f55eab82 1354 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
1da177e4
LT		 1355
1356 /* At this point we know we have a real path component. */
1357 for(;;) {
31e6b01f 1358 struct inode *inode;
1da177e4
LT		 1359 unsigned long hash;
1360 struct qstr this;
1361 unsigned int c;
fe479a58 1362 int type;
1da177e4 1363
cdce5d6b 1364 nd->flags |= LOOKUP_CONTINUE;
52094c8a
AV		 1365
1366 err = may_lookup(nd);
1da177e4
LT		 1367 if (err)
1368 break;
1369
1370 this.name = name;
1371 c = *(const unsigned char *)name;
1372
1373 hash = init_name_hash();
1374 do {
1375 name++;
1376 hash = partial_name_hash(c, hash);
1377 c = *(const unsigned char *)name;
1378 } while (c && (c != '/'));
1379 this.len = name - (const char *) this.name;
1380 this.hash = end_name_hash(hash);
1381
fe479a58
AV		 1382 type = LAST_NORM;
1383 if (this.name[0] == '.') switch (this.len) {
1384 case 2:
16c2cd71 1385 if (this.name[1] == '.') {
fe479a58 1386 type = LAST_DOTDOT;
16c2cd71
AV		 1387 nd->flags |= LOOKUP_JUMPED;
1388 }
fe479a58
AV		 1389 break;
1390 case 1:
1391 type = LAST_DOT;
1392 }
5a202bcd
AV		 1393 if (likely(type == LAST_NORM)) {
1394 struct dentry *parent = nd->path.dentry;
16c2cd71 1395 nd->flags &= ~LOOKUP_JUMPED;
5a202bcd
AV		 1396 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1397 err = parent->d_op->d_hash(parent, nd->inode,
1398 &this);
1399 if (err < 0)
1400 break;
1401 }
1402 }
fe479a58 1403
1da177e4
LT		 1404 /* remove trailing slashes? */
1405 if (!c)
1406 goto last_component;
1407 while (*++name == '/');
1408 if (!*name)
1409 goto last_with_slashes;
1410
1411 /*
1412 * "." and ".." are special - ".." especially so because it has
1413 * to be able to know about the current root directory and
1414 * parent relationships.
1415 */
fe479a58 1416 if (unlikely(type != LAST_NORM)) {
ef7562d5
AV		 1417 if (handle_dots(nd, type))
1418 return -ECHILD;
fe479a58 1419 continue;
1da177e4 1420 }
fe479a58 1421
1da177e4 1422 /* This does the actual lookups.. */
31e6b01f 1423 err = do_lookup(nd, &this, &next, &inode);
1da177e4
LT		 1424 if (err)
1425 break;
1da177e4 1426
7bc055d1 1427 if (inode && inode->i_op->follow_link) {
3abb17e8 1428 err = do_follow_link(inode, &next, nd);
1da177e4 1429 if (err)
a7472bab 1430 return err;
31e6b01f 1431 nd->inode = nd->path.dentry->d_inode;
31e6b01f 1432 } else {
09dd17d3 1433 path_to_nameidata(&next, nd);
31e6b01f
NP		 1434 nd->inode = inode;
1435 }
7bc055d1
AV		 1436 err = -ENOENT;
1437 if (!nd->inode)
1438 break;
1da177e4 1439 err = -ENOTDIR;
31e6b01f 1440 if (!nd->inode->i_op->lookup)
1da177e4
LT		 1441 break;
1442 continue;
1443 /* here ends the main loop */
1444
1445last_with_slashes:
1446 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1447last_component:
f55eab82
TM		 1448 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1449 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
1da177e4
LT		 1450 if (lookup_flags & LOOKUP_PARENT)
1451 goto lookup_parent;
ef7562d5
AV		 1452 if (unlikely(type != LAST_NORM))
1453 return handle_dots(nd, type);
31e6b01f 1454 err = do_lookup(nd, &this, &next, &inode);
1da177e4
LT		 1455 if (err)
1456 break;
db372915
DH		 1457 if (inode && unlikely(inode->i_op->follow_link) &&
1458 (lookup_flags & LOOKUP_FOLLOW)) {
3abb17e8 1459 err = do_follow_link(inode, &next, nd);
1da177e4 1460 if (err)
a7472bab 1461 return err;
31e6b01f
NP		 1462 nd->inode = nd->path.dentry->d_inode;
1463 } else {
09dd17d3 1464 path_to_nameidata(&next, nd);
31e6b01f
NP		 1465 nd->inode = inode;
1466 }
1da177e4 1467 err = -ENOENT;
31e6b01f 1468 if (!nd->inode)
1da177e4
LT		 1469 break;
1470 if (lookup_flags & LOOKUP_DIRECTORY) {
1471 err = -ENOTDIR;
31e6b01f 1472 if (!nd->inode->i_op->lookup)
1da177e4
LT		 1473 break;
1474 }
086e183a 1475 return 0;
1da177e4
LT		 1476lookup_parent:
1477 nd->last = this;
fe479a58 1478 nd->last_type = type;
1da177e4 1479 return 0;
1da177e4 1480 }
951361f9 1481 terminate_walk(nd);
1da177e4
LT		 1482 return err;
1483}
1484
70e9b357
AV		 1485static int path_init(int dfd, const char *name, unsigned int flags,
1486 struct nameidata *nd, struct file **fp)
31e6b01f
NP		 1487{
1488 int retval = 0;
1489 int fput_needed;
1490 struct file *file;
1491
1492 nd->last_type = LAST_ROOT; /* if there are only slashes... */
16c2cd71 1493 nd->flags = flags | LOOKUP_JUMPED;
31e6b01f 1494 nd->depth = 0;
5b6ca027
AV		 1495 if (flags & LOOKUP_ROOT) {
1496 struct inode *inode = nd->root.dentry->d_inode;
73d049a4
AV		 1497 if (*name) {
1498 if (!inode->i_op->lookup)
1499 return -ENOTDIR;
1500 retval = inode_permission(inode, MAY_EXEC);
1501 if (retval)
1502 return retval;
1503 }
5b6ca027
AV		 1504 nd->path = nd->root;
1505 nd->inode = inode;
1506 if (flags & LOOKUP_RCU) {
1507 br_read_lock(vfsmount_lock);
1508 rcu_read_lock();
1509 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1510 } else {
1511 path_get(&nd->path);
1512 }
1513 return 0;
1514 }
1515
31e6b01f 1516 nd->root.mnt = NULL;
31e6b01f
NP		 1517
1518 if (*name=='/') {
e41f7d4e
AV		 1519 if (flags & LOOKUP_RCU) {
1520 br_read_lock(vfsmount_lock);
1521 rcu_read_lock();
1522 set_root_rcu(nd);
1523 } else {
1524 set_root(nd);
1525 path_get(&nd->root);
1526 }
1527 nd->path = nd->root;
31e6b01f 1528 } else if (dfd == AT_FDCWD) {
e41f7d4e
AV		 1529 if (flags & LOOKUP_RCU) {
1530 struct fs_struct *fs = current->fs;
1531 unsigned seq;
31e6b01f 1532
e41f7d4e
AV		 1533 br_read_lock(vfsmount_lock);
1534 rcu_read_lock();
c28cc364 1535
e41f7d4e
AV		 1536 do {
1537 seq = read_seqcount_begin(&fs->seq);
1538 nd->path = fs->pwd;
1539 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1540 } while (read_seqcount_retry(&fs->seq, seq));
1541 } else {
1542 get_fs_pwd(current->fs, &nd->path);
1543 }
31e6b01f
NP		 1544 } else {
1545 struct dentry *dentry;
1546
1abf0c71 1547 file = fget_raw_light(dfd, &fput_needed);
31e6b01f
NP		 1548 retval = -EBADF;
1549 if (!file)
1550 goto out_fail;
1551
1552 dentry = file->f_path.dentry;
1553
f52e0c11
AV		 1554 if (*name) {
1555 retval = -ENOTDIR;
1556 if (!S_ISDIR(dentry->d_inode->i_mode))
1557 goto fput_fail;
31e6b01f 1558
f52e0c11
AV		 1559 retval = file_permission(file, MAY_EXEC);
1560 if (retval)
1561 goto fput_fail;
1562 }
31e6b01f
NP		 1563
1564 nd->path = file->f_path;
e41f7d4e
AV		 1565 if (flags & LOOKUP_RCU) {
1566 if (fput_needed)
70e9b357 1567 *fp = file;
e41f7d4e
AV		 1568 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1569 br_read_lock(vfsmount_lock);
1570 rcu_read_lock();
1571 } else {
1572 path_get(&file->f_path);
1573 fput_light(file, fput_needed);
1574 }
31e6b01f 1575 }
31e6b01f 1576
31e6b01f 1577 nd->inode = nd->path.dentry->d_inode;
9b4a9b14 1578 return 0;
2dfdd266 1579
9b4a9b14
AV		 1580fput_fail:
1581 fput_light(file, fput_needed);
1582out_fail:
1583 return retval;
1584}
1585
1586/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
ee0827cd 1587static int path_lookupat(int dfd, const char *name,
9b4a9b14
AV		 1588 unsigned int flags, struct nameidata *nd)
1589{
70e9b357 1590 struct file *base = NULL;
31e6b01f
NP		 1591 int retval;
1592
1593 /*
1594 * Path walking is largely split up into 2 different synchronisation
1595 * schemes, rcu-walk and ref-walk (explained in
1596 * Documentation/filesystems/path-lookup.txt). These share much of the
1597 * path walk code, but some things particularly setup, cleanup, and
1598 * following mounts are sufficiently divergent that functions are
1599 * duplicated. Typically there is a function foo(), and its RCU
1600 * analogue, foo_rcu().
1601 *
1602 * -ECHILD is the error number of choice (just to avoid clashes) that
1603 * is returned if some aspect of an rcu-walk fails. Such an error must
1604 * be handled by restarting a traditional ref-walk (which will always
1605 * be able to complete).
1606 */
70e9b357 1607 retval = path_init(dfd, name, flags, nd, &base);
ee0827cd 1608
31e6b01f
NP		 1609 if (unlikely(retval))
1610 return retval;
ee0827cd
AV		 1611
1612 current->total_link_count = 0;
1613 retval = link_path_walk(name, nd);
1614
1615 if (nd->flags & LOOKUP_RCU) {
4455ca62
AV		 1616 /* went all way through without dropping RCU */
1617 BUG_ON(retval);
1618 if (nameidata_drop_rcu_last(nd))
1619 retval = -ECHILD;
ee0827cd
AV		 1620 }
1621
16c2cd71
AV		 1622 if (!retval)
1623 retval = handle_reval_path(nd);
1624
70e9b357
AV		 1625 if (base)
1626 fput(base);
ee0827cd 1627
5b6ca027 1628 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
2a737871
AV		 1629 path_put(&nd->root);
1630 nd->root.mnt = NULL;
1631 }
ee0827cd
AV		 1632 return retval;
1633}
31e6b01f 1634
ee0827cd
AV		 1635static int do_path_lookup(int dfd, const char *name,
1636 unsigned int flags, struct nameidata *nd)
1637{
1638 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1639 if (unlikely(retval == -ECHILD))
1640 retval = path_lookupat(dfd, name, flags, nd);
1641 if (unlikely(retval == -ESTALE))
1642 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
31e6b01f
NP		 1643
1644 if (likely(!retval)) {
1645 if (unlikely(!audit_dummy_context())) {
1646 if (nd->path.dentry && nd->inode)
1647 audit_inode(name, nd->path.dentry);
1648 }
1649 }
170aa3d0 1650 return retval;
1da177e4
LT		 1651}
1652
c9c6cac0 1653int kern_path_parent(const char *name, struct nameidata *nd)
5590ff0d 1654{
c9c6cac0 1655 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
5590ff0d
UD		 1656}
1657
d1811465
AV		 1658int kern_path(const char *name, unsigned int flags, struct path *path)
1659{
1660 struct nameidata nd;
1661 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1662 if (!res)
1663 *path = nd.path;
1664 return res;
1665}
1666
16f18200
JJS		 1667/**
1668 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1669 * @dentry: pointer to dentry of the base directory
1670 * @mnt: pointer to vfs mount of the base directory
1671 * @name: pointer to file name
1672 * @flags: lookup flags
1673 * @nd: pointer to nameidata
1674 */
1675int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1676 const char *name, unsigned int flags,
1677 struct nameidata *nd)
1678{
5b6ca027
AV		 1679 nd->root.dentry = dentry;
1680 nd->root.mnt = mnt;
1681 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1682 return do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, nd);
16f18200
JJS		 1683}
1684
eead1911
CH		 1685static struct dentry *__lookup_hash(struct qstr *name,
1686 struct dentry *base, struct nameidata *nd)
1da177e4 1687{
81fca444 1688 struct inode *inode = base->d_inode;
057f6c01 1689 struct dentry *dentry;
1da177e4
LT		 1690 int err;
1691
b74c79e9 1692 err = exec_permission(inode, 0);
81fca444
CH		 1693 if (err)
1694 return ERR_PTR(err);
1da177e4 1695
b04f784e
NP		 1696 /*
1697 * Don't bother with __d_lookup: callers are for creat as
1698 * well as unlink, so a lot of the time it would cost
1699 * a double lookup.
6e6b1bd1 1700 */
b04f784e 1701 dentry = d_lookup(base, name);
6e6b1bd1 1702
fb045adb 1703 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE))
6e6b1bd1
AV		 1704 dentry = do_revalidate(dentry, nd);
1705
baa03890
NP		 1706 if (!dentry)
1707 dentry = d_alloc_and_lookup(base, name, nd);
5a202bcd 1708
1da177e4
LT		 1709 return dentry;
1710}
1711
057f6c01
JM		 1712/*
1713 * Restricted form of lookup. Doesn't follow links, single-component only,
1714 * needs parent already locked. Doesn't follow mounts.
1715 * SMP-safe.
1716 */
eead1911 1717static struct dentry *lookup_hash(struct nameidata *nd)
057f6c01 1718{
4ac91378 1719 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1da177e4
LT		 1720}
1721
eead1911 1722/**
a6b91919 1723 * lookup_one_len - filesystem helper to lookup single pathname component
eead1911
CH		 1724 * @name: pathname component to lookup
1725 * @base: base directory to lookup from
1726 * @len: maximum length @len should be interpreted to
1727 *
a6b91919
RD		 1728 * Note that this routine is purely a helper for filesystem usage and should
1729 * not be called by generic code. Also note that by using this function the
eead1911
CH		 1730 * nameidata argument is passed to the filesystem methods and a filesystem
1731 * using this helper needs to be prepared for that.
1732 */
057f6c01
JM		 1733struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1734{
057f6c01 1735 struct qstr this;
6a96ba54
AV		 1736 unsigned long hash;
1737 unsigned int c;
057f6c01 1738
2f9092e1
DW		 1739 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1740
6a96ba54
AV		 1741 this.name = name;
1742 this.len = len;
1743 if (!len)
1744 return ERR_PTR(-EACCES);
1745
1746 hash = init_name_hash();
1747 while (len--) {
1748 c = *(const unsigned char *)name++;
1749 if (c == '/' || c == '\0')
1750 return ERR_PTR(-EACCES);
1751 hash = partial_name_hash(c, hash);
1752 }
1753 this.hash = end_name_hash(hash);
5a202bcd
AV		 1754 /*
1755 * See if the low-level filesystem might want
1756 * to use its own hash..
1757 */
1758 if (base->d_flags & DCACHE_OP_HASH) {
1759 int err = base->d_op->d_hash(base, base->d_inode, &this);
1760 if (err < 0)
1761 return ERR_PTR(err);
1762 }
eead1911 1763
49705b77 1764 return __lookup_hash(&this, base, NULL);
057f6c01
JM		 1765}
1766
2d8f3038
AV		 1767int user_path_at(int dfd, const char __user *name, unsigned flags,
1768 struct path *path)
1da177e4 1769{
2d8f3038 1770 struct nameidata nd;
f52e0c11 1771 char *tmp = getname_flags(name, flags);
1da177e4 1772 int err = PTR_ERR(tmp);
1da177e4 1773 if (!IS_ERR(tmp)) {
2d8f3038
AV		 1774
1775 BUG_ON(flags & LOOKUP_PARENT);
1776
1777 err = do_path_lookup(dfd, tmp, flags, &nd);
1da177e4 1778 putname(tmp);
2d8f3038
AV		 1779 if (!err)
1780 *path = nd.path;
1da177e4
LT		 1781 }
1782 return err;
1783}
1784
2ad94ae6
AV		 1785static int user_path_parent(int dfd, const char __user *path,
1786 struct nameidata *nd, char **name)
1787{
1788 char *s = getname(path);
1789 int error;
1790
1791 if (IS_ERR(s))
1792 return PTR_ERR(s);
1793
1794 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1795 if (error)
1796 putname(s);
1797 else
1798 *name = s;
1799
1800 return error;
1801}
1802
1da177e4
LT		 1803/*
1804 * It's inline, so penalty for filesystems that don't use sticky bit is
1805 * minimal.
1806 */
1807static inline int check_sticky(struct inode *dir, struct inode *inode)
1808{
da9592ed
DH		 1809 uid_t fsuid = current_fsuid();
1810
1da177e4
LT		 1811 if (!(dir->i_mode & S_ISVTX))
1812 return 0;
da9592ed 1813 if (inode->i_uid == fsuid)
1da177e4 1814 return 0;
da9592ed 1815 if (dir->i_uid == fsuid)
1da177e4
LT		 1816 return 0;
1817 return !capable(CAP_FOWNER);
1818}
1819
1820/*
1821 * Check whether we can remove a link victim from directory dir, check
1822 * whether the type of victim is right.
1823 * 1. We can't do it if dir is read-only (done in permission())
1824 * 2. We should have write and exec permissions on dir
1825 * 3. We can't remove anything from append-only dir
1826 * 4. We can't do anything with immutable dir (done in permission())
1827 * 5. If the sticky bit on dir is set we should either
1828 * a. be owner of dir, or
1829 * b. be owner of victim, or
1830 * c. have CAP_FOWNER capability
1831 * 6. If the victim is append-only or immutable we can't do antyhing with
1832 * links pointing to it.
1833 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1834 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1835 * 9. We can't remove a root or mountpoint.
1836 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1837 * nfs_async_unlink().
1838 */
858119e1 1839static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1da177e4
LT		 1840{
1841 int error;
1842
1843 if (!victim->d_inode)
1844 return -ENOENT;
1845
1846 BUG_ON(victim->d_parent->d_inode != dir);
cccc6bba 1847 audit_inode_child(victim, dir);
1da177e4 1848
f419a2e3 1849 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1850 if (error)
1851 return error;
1852 if (IS_APPEND(dir))
1853 return -EPERM;
1854 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
f9454548 1855 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1da177e4
LT		 1856 return -EPERM;
1857 if (isdir) {
1858 if (!S_ISDIR(victim->d_inode->i_mode))
1859 return -ENOTDIR;
1860 if (IS_ROOT(victim))
1861 return -EBUSY;
1862 } else if (S_ISDIR(victim->d_inode->i_mode))
1863 return -EISDIR;
1864 if (IS_DEADDIR(dir))
1865 return -ENOENT;
1866 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1867 return -EBUSY;
1868 return 0;
1869}
1870
1871/* Check whether we can create an object with dentry child in directory
1872 * dir.
1873 * 1. We can't do it if child already exists (open has special treatment for
1874 * this case, but since we are inlined it's OK)
1875 * 2. We can't do it if dir is read-only (done in permission())
1876 * 3. We should have write and exec permissions on dir
1877 * 4. We can't do it if dir is immutable (done in permission())
1878 */
a95164d9 1879static inline int may_create(struct inode *dir, struct dentry *child)
1da177e4
LT		 1880{
1881 if (child->d_inode)
1882 return -EEXIST;
1883 if (IS_DEADDIR(dir))
1884 return -ENOENT;
f419a2e3 1885 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1886}
1887
1da177e4
LT		 1888/*
1889 * p1 and p2 should be directories on the same fs.
1890 */
1891struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1892{
1893 struct dentry *p;
1894
1895 if (p1 == p2) {
f2eace23 1896 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1da177e4
LT		 1897 return NULL;
1898 }
1899
a11f3a05 1900 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4 1901
e2761a11
OH		 1902 p = d_ancestor(p2, p1);
1903 if (p) {
1904 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1905 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1906 return p;
1da177e4
LT		 1907 }
1908
e2761a11
OH		 1909 p = d_ancestor(p1, p2);
1910 if (p) {
1911 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1912 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1913 return p;
1da177e4
LT		 1914 }
1915
f2eace23
IM		 1916 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1917 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1da177e4
LT		 1918 return NULL;
1919}
1920
1921void unlock_rename(struct dentry *p1, struct dentry *p2)
1922{
1b1dcc1b 1923 mutex_unlock(&p1->d_inode->i_mutex);
1da177e4 1924 if (p1 != p2) {
1b1dcc1b 1925 mutex_unlock(&p2->d_inode->i_mutex);
a11f3a05 1926 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4
LT		 1927 }
1928}
1929
1930int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1931 struct nameidata *nd)
1932{
a95164d9 1933 int error = may_create(dir, dentry);
1da177e4
LT		 1934
1935 if (error)
1936 return error;
1937
acfa4380 1938 if (!dir->i_op->create)
1da177e4
LT		 1939 return -EACCES; /* shouldn't it be ENOSYS? */
1940 mode &= S_IALLUGO;
1941 mode |= S_IFREG;
1942 error = security_inode_create(dir, dentry, mode);
1943 if (error)
1944 return error;
1da177e4 1945 error = dir->i_op->create(dir, dentry, mode, nd);
a74574aa 1946 if (!error)
f38aa942 1947 fsnotify_create(dir, dentry);
1da177e4
LT		 1948 return error;
1949}
1950
73d049a4 1951static int may_open(struct path *path, int acc_mode, int flag)
1da177e4 1952{
3fb64190 1953 struct dentry *dentry = path->dentry;
1da177e4
LT		 1954 struct inode *inode = dentry->d_inode;
1955 int error;
1956
1957 if (!inode)
1958 return -ENOENT;
1959
c8fe8f30
CH		 1960 switch (inode->i_mode & S_IFMT) {
1961 case S_IFLNK:
1da177e4 1962 return -ELOOP;
c8fe8f30
CH		 1963 case S_IFDIR:
1964 if (acc_mode & MAY_WRITE)
1965 return -EISDIR;
1966 break;
1967 case S_IFBLK:
1968 case S_IFCHR:
3fb64190 1969 if (path->mnt->mnt_flags & MNT_NODEV)
1da177e4 1970 return -EACCES;
c8fe8f30
CH		 1971 /*FALLTHRU*/
1972 case S_IFIFO:
1973 case S_IFSOCK:
1da177e4 1974 flag &= ~O_TRUNC;
c8fe8f30 1975 break;
4a3fd211 1976 }
b41572e9 1977
3fb64190 1978 error = inode_permission(inode, acc_mode);
b41572e9
DH		 1979 if (error)
1980 return error;
6146f0d5 1981
1da177e4
LT		 1982 /*
1983 * An append-only file must be opened in append mode for writing.
1984 */
1985 if (IS_APPEND(inode)) {
8737c930 1986 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
7715b521 1987 return -EPERM;
1da177e4 1988 if (flag & O_TRUNC)
7715b521 1989 return -EPERM;
1da177e4
LT		 1990 }
1991
1992 /* O_NOATIME can only be set by the owner or superuser */
7715b521
AV		 1993 if (flag & O_NOATIME && !is_owner_or_cap(inode))
1994 return -EPERM;
1da177e4
LT		 1995
1996 /*
1997 * Ensure there are no outstanding leases on the file.
1998 */
b65a9cfc 1999 return break_lease(inode, flag);
7715b521 2000}
1da177e4 2001
e1181ee6 2002static int handle_truncate(struct file *filp)
7715b521 2003{
e1181ee6 2004 struct path *path = &filp->f_path;
7715b521
AV		 2005 struct inode *inode = path->dentry->d_inode;
2006 int error = get_write_access(inode);
2007 if (error)
2008 return error;
2009 /*
2010 * Refuse to truncate files with mandatory locks held on them.
2011 */
2012 error = locks_verify_locked(inode);
2013 if (!error)
ea0d3ab2 2014 error = security_path_truncate(path);
7715b521
AV		 2015 if (!error) {
2016 error = do_truncate(path->dentry, 0,
2017 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
e1181ee6 2018 filp);
7715b521
AV		 2019 }
2020 put_write_access(inode);
acd0c935 2021 return error;
1da177e4
LT		 2022}
2023
d57999e1
DH		 2024/*
2025 * Note that while the flag value (low two bits) for sys_open means:
2026 * 00 - read-only
2027 * 01 - write-only
2028 * 10 - read-write
2029 * 11 - special
2030 * it is changed into
2031 * 00 - no permissions needed
2032 * 01 - read-permission
2033 * 10 - write-permission
2034 * 11 - read-write
2035 * for the internal routines (ie open_namei()/follow_link() etc)
2036 * This is more logical, and also allows the 00 "no perm needed"
2037 * to be used for symlinks (where the permissions are checked
2038 * later).
2039 *
2040*/
2041static inline int open_to_namei_flags(int flag)
2042{
2043 if ((flag+1) & O_ACCMODE)
2044 flag++;
2045 return flag;
2046}
2047
31e6b01f 2048/*
fe2d35ff 2049 * Handle the last step of open()
31e6b01f 2050 */
fb1cc555 2051static struct file *do_last(struct nameidata *nd, struct path *path,
c3e380b0 2052 const struct open_flags *op, const char *pathname)
fb1cc555 2053{
a1e28038 2054 struct dentry *dir = nd->path.dentry;
6c0d46c4 2055 struct dentry *dentry;
ca344a89 2056 int open_flag = op->open_flag;
6c0d46c4 2057 int will_truncate = open_flag & O_TRUNC;
ca344a89
AV		 2058 int want_write = 0;
2059 int skip_perm = 0;
fb1cc555 2060 struct file *filp;
fe2d35ff 2061 struct inode *inode;
16c2cd71 2062 int error;
1f36f774 2063
c3e380b0
AV		 2064 nd->flags &= ~LOOKUP_PARENT;
2065 nd->flags |= op->intent;
2066
1f36f774
AV		 2067 switch (nd->last_type) {
2068 case LAST_DOTDOT:
176306f5 2069 case LAST_DOT:
fe2d35ff
AV		 2070 error = handle_dots(nd, nd->last_type);
2071 if (error)
2072 return ERR_PTR(error);
1f36f774 2073 /* fallthrough */
1f36f774 2074 case LAST_ROOT:
fe2d35ff
AV		 2075 if (nd->flags & LOOKUP_RCU) {
2076 if (nameidata_drop_rcu_last(nd))
2077 return ERR_PTR(-ECHILD);
2078 }
16c2cd71
AV		 2079 error = handle_reval_path(nd);
2080 if (error)
2081 goto exit;
fe2d35ff 2082 audit_inode(pathname, nd->path.dentry);
ca344a89 2083 if (open_flag & O_CREAT) {
fe2d35ff
AV		 2084 error = -EISDIR;
2085 goto exit;
2086 }
2087 goto ok;
1f36f774 2088 case LAST_BIND:
fe2d35ff 2089 /* can't be RCU mode here */
16c2cd71
AV		 2090 error = handle_reval_path(nd);
2091 if (error)
2092 goto exit;
1f36f774 2093 audit_inode(pathname, dir);
67ee3ad2 2094 goto ok;
1f36f774 2095 }
67ee3ad2 2096
ca344a89 2097 if (!(open_flag & O_CREAT)) {
fe2d35ff
AV		 2098 if (nd->last.name[nd->last.len])
2099 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2100 /* we _can_ be in RCU mode here */
2101 error = do_lookup(nd, &nd->last, path, &inode);
2102 if (error) {
2103 terminate_walk(nd);
2104 return ERR_PTR(error);
2105 }
2106 if (!inode) {
2107 path_to_nameidata(path, nd);
2108 terminate_walk(nd);
2109 return ERR_PTR(-ENOENT);
2110 }
2111 if (unlikely(inode->i_op->follow_link)) {
2112 /* We drop rcu-walk here */
2113 if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry))
2114 return ERR_PTR(-ECHILD);
2115 return NULL;
2116 }
2117 path_to_nameidata(path, nd);
2118 nd->inode = inode;
2119 /* sayonara */
2120 if (nd->flags & LOOKUP_RCU) {
2121 if (nameidata_drop_rcu_last(nd))
2122 return ERR_PTR(-ECHILD);
2123 }
2124
2125 error = -ENOTDIR;
2126 if (nd->flags & LOOKUP_DIRECTORY) {
2127 if (!inode->i_op->lookup)
2128 goto exit;
2129 }
2130 audit_inode(pathname, nd->path.dentry);
2131 goto ok;
2132 }
2133
2134 /* create side of things */
2135
2136 if (nd->flags & LOOKUP_RCU) {
2137 if (nameidata_drop_rcu_last(nd))
2138 return ERR_PTR(-ECHILD);
2139 }
2140
2141 audit_inode(pathname, dir);
16c2cd71 2142 error = -EISDIR;
1f36f774 2143 /* trailing slashes? */
31e6b01f
NP		 2144 if (nd->last.name[nd->last.len])
2145 goto exit;
a2c36b45 2146
a1e28038
AV		 2147 mutex_lock(&dir->d_inode->i_mutex);
2148
6c0d46c4
AV		 2149 dentry = lookup_hash(nd);
2150 error = PTR_ERR(dentry);
2151 if (IS_ERR(dentry)) {
fb1cc555
AV		 2152 mutex_unlock(&dir->d_inode->i_mutex);
2153 goto exit;
2154 }
2155
6c0d46c4
AV		 2156 path->dentry = dentry;
2157 path->mnt = nd->path.mnt;
2158
fb1cc555 2159 /* Negative dentry, just create the file */
6c0d46c4
AV		 2160 if (!dentry->d_inode) {
2161 int mode = op->mode;
2162 if (!IS_POSIXACL(dir->d_inode))
2163 mode &= ~current_umask();
fb1cc555
AV		 2164 /*
2165 * This write is needed to ensure that a
6c0d46c4 2166 * rw->ro transition does not occur between
fb1cc555
AV		 2167 * the time when the file is created and when
2168 * a permanent write count is taken through
2169 * the 'struct file' in nameidata_to_filp().
2170 */
2171 error = mnt_want_write(nd->path.mnt);
2172 if (error)
2173 goto exit_mutex_unlock;
ca344a89 2174 want_write = 1;
9b44f1b3 2175 /* Don't check for write permission, don't truncate */
ca344a89 2176 open_flag &= ~O_TRUNC;
6c0d46c4 2177 will_truncate = 0;
ca344a89 2178 skip_perm = 1;
6c0d46c4
AV		 2179 error = security_path_mknod(&nd->path, dentry, mode, 0);
2180 if (error)
2181 goto exit_mutex_unlock;
2182 error = vfs_create(dir->d_inode, dentry, mode, nd);
2183 if (error)
2184 goto exit_mutex_unlock;
2185 mutex_unlock(&dir->d_inode->i_mutex);
2186 dput(nd->path.dentry);
2187 nd->path.dentry = dentry;
ca344a89 2188 goto common;
fb1cc555
AV		 2189 }
2190
2191 /*
2192 * It already exists.
2193 */
2194 mutex_unlock(&dir->d_inode->i_mutex);
2195 audit_inode(pathname, path->dentry);
2196
2197 error = -EEXIST;
ca344a89 2198 if (open_flag & O_EXCL)
fb1cc555
AV		 2199 goto exit_dput;
2200
9875cf80
DH		 2201 error = follow_managed(path, nd->flags);
2202 if (error < 0)
2203 goto exit_dput;
fb1cc555
AV		 2204
2205 error = -ENOENT;
2206 if (!path->dentry->d_inode)
2207 goto exit_dput;
9e67f361
AV		 2208
2209 if (path->dentry->d_inode->i_op->follow_link)
fb1cc555 2210 return NULL;
fb1cc555
AV		 2211
2212 path_to_nameidata(path, nd);
31e6b01f 2213 nd->inode = path->dentry->d_inode;
fb1cc555 2214 error = -EISDIR;
31e6b01f 2215 if (S_ISDIR(nd->inode->i_mode))
fb1cc555 2216 goto exit;
67ee3ad2 2217ok:
6c0d46c4
AV		 2218 if (!S_ISREG(nd->inode->i_mode))
2219 will_truncate = 0;
2220
0f9d1a10
AV		 2221 if (will_truncate) {
2222 error = mnt_want_write(nd->path.mnt);
2223 if (error)
2224 goto exit;
ca344a89 2225 want_write = 1;
0f9d1a10 2226 }
ca344a89
AV		 2227common:
2228 error = may_open(&nd->path, skip_perm ? 0 : op->acc_mode, open_flag);
2229 if (error)
0f9d1a10 2230 goto exit;
0f9d1a10
AV		 2231 filp = nameidata_to_filp(nd);
2232 if (!IS_ERR(filp)) {
2233 error = ima_file_check(filp, op->acc_mode);
2234 if (error) {
2235 fput(filp);
2236 filp = ERR_PTR(error);
2237 }
2238 }
2239 if (!IS_ERR(filp)) {
2240 if (will_truncate) {
2241 error = handle_truncate(filp);
2242 if (error) {
2243 fput(filp);
2244 filp = ERR_PTR(error);
2245 }
2246 }
2247 }
ca344a89
AV		 2248out:
2249 if (want_write)
0f9d1a10
AV		 2250 mnt_drop_write(nd->path.mnt);
2251 path_put(&nd->path);
fb1cc555
AV		 2252 return filp;
2253
2254exit_mutex_unlock:
2255 mutex_unlock(&dir->d_inode->i_mutex);
2256exit_dput:
2257 path_put_conditional(path, nd);
2258exit:
ca344a89
AV		 2259 filp = ERR_PTR(error);
2260 goto out;
fb1cc555
AV		 2261}
2262
13aab428 2263static struct file *path_openat(int dfd, const char *pathname,
73d049a4 2264 struct nameidata *nd, const struct open_flags *op, int flags)
1da177e4 2265{
fe2d35ff 2266 struct file *base = NULL;
4a3fd211 2267 struct file *filp;
9850c056 2268 struct path path;
1da177e4 2269 int count = 0;
13aab428 2270 int error;
31e6b01f
NP		 2271
2272 filp = get_empty_filp();
2273 if (!filp)
2274 return ERR_PTR(-ENFILE);
2275
47c805dc 2276 filp->f_flags = op->open_flag;
73d049a4
AV		 2277 nd->intent.open.file = filp;
2278 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2279 nd->intent.open.create_mode = op->mode;
31e6b01f 2280
73d049a4 2281 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
31e6b01f 2282 if (unlikely(error))
13aab428 2283 goto out_filp;
31e6b01f 2284
fe2d35ff 2285 current->total_link_count = 0;
73d049a4 2286 error = link_path_walk(pathname, nd);
31e6b01f
NP		 2287 if (unlikely(error))
2288 goto out_filp;
1da177e4 2289
73d049a4 2290 filp = do_last(nd, &path, op, pathname);
806b681c 2291 while (unlikely(!filp)) { /* trailing symlink */
7b9337aa
NP		 2292 struct path link = path;
2293 struct inode *linki = link.dentry->d_inode;
def4af30 2294 void *cookie;
73d049a4
AV		 2295 if (!(nd->flags & LOOKUP_FOLLOW) || count++ == 32) {
2296 path_put_conditional(&path, nd);
2297 path_put(&nd->path);
40b39136
AV		 2298 filp = ERR_PTR(-ELOOP);
2299 break;
2300 }
806b681c
AV		 2301 /*
2302 * This is subtle. Instead of calling do_follow_link() we do
2303 * the thing by hands. The reason is that this way we have zero
2304 * link_count and path_walk() (called from ->follow_link)
2305 * honoring LOOKUP_PARENT. After that we have the parent and
2306 * last component, i.e. we are in the same situation as after
2307 * the first path_walk(). Well, almost - if the last component
2308 * is normal we get its copy stored in nd->last.name and we will
2309 * have to putname() it when we are done. Procfs-like symlinks
2310 * just set LAST_BIND.
2311 */
73d049a4
AV		 2312 nd->flags |= LOOKUP_PARENT;
2313 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
2314 error = __do_follow_link(&link, nd, &cookie);
c3e380b0 2315 if (unlikely(error))
f1afe9ef 2316 filp = ERR_PTR(error);
c3e380b0 2317 else
73d049a4 2318 filp = do_last(nd, &path, op, pathname);
f1afe9ef 2319 if (!IS_ERR(cookie) && linki->i_op->put_link)
73d049a4 2320 linki->i_op->put_link(link.dentry, nd, cookie);
7b9337aa 2321 path_put(&link);
806b681c 2322 }
10fa8e62 2323out:
73d049a4
AV		 2324 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2325 path_put(&nd->root);
fe2d35ff
AV		 2326 if (base)
2327 fput(base);
73d049a4 2328 release_open_intent(nd);
10fa8e62 2329 return filp;
1da177e4 2330
31e6b01f 2331out_filp:
806b681c 2332 filp = ERR_PTR(error);
10fa8e62 2333 goto out;
1da177e4
LT		 2334}
2335
13aab428
AV		 2336struct file *do_filp_open(int dfd, const char *pathname,
2337 const struct open_flags *op, int flags)
2338{
73d049a4 2339 struct nameidata nd;
13aab428
AV		 2340 struct file *filp;
2341
73d049a4 2342 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
13aab428 2343 if (unlikely(filp == ERR_PTR(-ECHILD)))
73d049a4 2344 filp = path_openat(dfd, pathname, &nd, op, flags);
13aab428 2345 if (unlikely(filp == ERR_PTR(-ESTALE)))
73d049a4 2346 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
13aab428
AV		 2347 return filp;
2348}
2349
73d049a4
AV		 2350struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2351 const char *name, const struct open_flags *op, int flags)
2352{
2353 struct nameidata nd;
2354 struct file *file;
2355
2356 nd.root.mnt = mnt;
2357 nd.root.dentry = dentry;
2358
2359 flags |= LOOKUP_ROOT;
2360
2361 if (dentry->d_inode->i_op->follow_link)
2362 return ERR_PTR(-ELOOP);
2363
2364 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2365 if (unlikely(file == ERR_PTR(-ECHILD)))
2366 file = path_openat(-1, name, &nd, op, flags);
2367 if (unlikely(file == ERR_PTR(-ESTALE)))
2368 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2369 return file;
2370}
2371
1da177e4
LT		 2372/**
2373 * lookup_create - lookup a dentry, creating it if it doesn't exist
2374 * @nd: nameidata info
2375 * @is_dir: directory flag
2376 *
2377 * Simple function to lookup and return a dentry and create it
2378 * if it doesn't exist. Is SMP-safe.
c663e5d8 2379 *
4ac91378 2380 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1da177e4
LT		 2381 */
2382struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2383{
c663e5d8 2384 struct dentry *dentry = ERR_PTR(-EEXIST);
1da177e4 2385
4ac91378 2386 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
c663e5d8
CH		 2387 /*
2388 * Yucky last component or no last component at all?
2389 * (foo/., foo/.., /////)
2390 */
1da177e4
LT		 2391 if (nd->last_type != LAST_NORM)
2392 goto fail;
2393 nd->flags &= ~LOOKUP_PARENT;
3516586a 2394 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
a634904a 2395 nd->intent.open.flags = O_EXCL;
c663e5d8
CH		 2396
2397 /*
2398 * Do the final lookup.
2399 */
49705b77 2400 dentry = lookup_hash(nd);
1da177e4
LT		 2401 if (IS_ERR(dentry))
2402 goto fail;
c663e5d8 2403
e9baf6e5
AV		 2404 if (dentry->d_inode)
2405 goto eexist;
c663e5d8
CH		 2406 /*
2407 * Special case - lookup gave negative, but... we had foo/bar/
2408 * From the vfs_mknod() POV we just have a negative dentry -
2409 * all is fine. Let's be bastards - you had / on the end, you've
2410 * been asking for (non-existent) directory. -ENOENT for you.
2411 */
e9baf6e5
AV		 2412 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2413 dput(dentry);
2414 dentry = ERR_PTR(-ENOENT);
2415 }
1da177e4 2416 return dentry;
e9baf6e5 2417eexist:
1da177e4 2418 dput(dentry);
e9baf6e5 2419 dentry = ERR_PTR(-EEXIST);
1da177e4
LT		 2420fail:
2421 return dentry;
2422}
f81a0bff 2423EXPORT_SYMBOL_GPL(lookup_create);
1da177e4
LT		 2424
2425int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2426{
a95164d9 2427 int error = may_create(dir, dentry);
1da177e4
LT		 2428
2429 if (error)
2430 return error;
2431
2432 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2433 return -EPERM;
2434
acfa4380 2435 if (!dir->i_op->mknod)
1da177e4
LT		 2436 return -EPERM;
2437
08ce5f16
SH		 2438 error = devcgroup_inode_mknod(mode, dev);
2439 if (error)
2440 return error;
2441
1da177e4
LT		 2442 error = security_inode_mknod(dir, dentry, mode, dev);
2443 if (error)
2444 return error;
2445
1da177e4 2446 error = dir->i_op->mknod(dir, dentry, mode, dev);
a74574aa 2447 if (!error)
f38aa942 2448 fsnotify_create(dir, dentry);
1da177e4
LT		 2449 return error;
2450}
2451
463c3197
DH		 2452static int may_mknod(mode_t mode)
2453{
2454 switch (mode & S_IFMT) {
2455 case S_IFREG:
2456 case S_IFCHR:
2457 case S_IFBLK:
2458 case S_IFIFO:
2459 case S_IFSOCK:
2460 case 0: /* zero mode translates to S_IFREG */
2461 return 0;
2462 case S_IFDIR:
2463 return -EPERM;
2464 default:
2465 return -EINVAL;
2466 }
2467}
2468
2e4d0924
HC		 2469SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2470 unsigned, dev)
1da177e4 2471{
2ad94ae6
AV		 2472 int error;
2473 char *tmp;
2474 struct dentry *dentry;
1da177e4
LT		 2475 struct nameidata nd;
2476
2477 if (S_ISDIR(mode))
2478 return -EPERM;
1da177e4 2479
2ad94ae6 2480 error = user_path_parent(dfd, filename, &nd, &tmp);
1da177e4 2481 if (error)
2ad94ae6
AV		 2482 return error;
2483
1da177e4 2484 dentry = lookup_create(&nd, 0);
463c3197
DH		 2485 if (IS_ERR(dentry)) {
2486 error = PTR_ERR(dentry);
2487 goto out_unlock;
2488 }
4ac91378 2489 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2490 mode &= ~current_umask();
463c3197
DH		 2491 error = may_mknod(mode);
2492 if (error)
2493 goto out_dput;
2494 error = mnt_want_write(nd.path.mnt);
2495 if (error)
2496 goto out_dput;
be6d3e56
KT		 2497 error = security_path_mknod(&nd.path, dentry, mode, dev);
2498 if (error)
2499 goto out_drop_write;
463c3197 2500 switch (mode & S_IFMT) {
1da177e4 2501 case 0: case S_IFREG:
4ac91378 2502 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
1da177e4
LT		 2503 break;
2504 case S_IFCHR: case S_IFBLK:
4ac91378 2505 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
1da177e4
LT		 2506 new_decode_dev(dev));
2507 break;
2508 case S_IFIFO: case S_IFSOCK:
4ac91378 2509 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
1da177e4 2510 break;
1da177e4 2511 }
be6d3e56 2512out_drop_write:
463c3197
DH		 2513 mnt_drop_write(nd.path.mnt);
2514out_dput:
2515 dput(dentry);
2516out_unlock:
4ac91378 2517 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2518 path_put(&nd.path);
1da177e4
LT		 2519 putname(tmp);
2520
2521 return error;
2522}
2523
3480b257 2524SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
5590ff0d
UD		 2525{
2526 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2527}
2528
1da177e4
LT		 2529int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2530{
a95164d9 2531 int error = may_create(dir, dentry);
1da177e4
LT		 2532
2533 if (error)
2534 return error;
2535
acfa4380 2536 if (!dir->i_op->mkdir)
1da177e4
LT		 2537 return -EPERM;
2538
2539 mode &= (S_IRWXUGO|S_ISVTX);
2540 error = security_inode_mkdir(dir, dentry, mode);
2541 if (error)
2542 return error;
2543
1da177e4 2544 error = dir->i_op->mkdir(dir, dentry, mode);
a74574aa 2545 if (!error)
f38aa942 2546 fsnotify_mkdir(dir, dentry);
1da177e4
LT		 2547 return error;
2548}
2549
2e4d0924 2550SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
1da177e4
LT		 2551{
2552 int error = 0;
2553 char * tmp;
6902d925
DH		 2554 struct dentry *dentry;
2555 struct nameidata nd;
1da177e4 2556
2ad94ae6
AV		 2557 error = user_path_parent(dfd, pathname, &nd, &tmp);
2558 if (error)
6902d925 2559 goto out_err;
1da177e4 2560
6902d925
DH		 2561 dentry = lookup_create(&nd, 1);
2562 error = PTR_ERR(dentry);
2563 if (IS_ERR(dentry))
2564 goto out_unlock;
1da177e4 2565
4ac91378 2566 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2567 mode &= ~current_umask();
463c3197
DH		 2568 error = mnt_want_write(nd.path.mnt);
2569 if (error)
2570 goto out_dput;
be6d3e56
KT		 2571 error = security_path_mkdir(&nd.path, dentry, mode);
2572 if (error)
2573 goto out_drop_write;
4ac91378 2574 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
be6d3e56 2575out_drop_write:
463c3197
DH		 2576 mnt_drop_write(nd.path.mnt);
2577out_dput:
6902d925
DH		 2578 dput(dentry);
2579out_unlock:
4ac91378 2580 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2581 path_put(&nd.path);
6902d925
DH		 2582 putname(tmp);
2583out_err:
1da177e4
LT		 2584 return error;
2585}
2586
3cdad428 2587SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
5590ff0d
UD		 2588{
2589 return sys_mkdirat(AT_FDCWD, pathname, mode);
2590}
2591
1da177e4
LT		 2592/*
2593 * We try to drop the dentry early: we should have
2594 * a usage count of 2 if we're the only user of this
2595 * dentry, and if that is true (possibly after pruning
2596 * the dcache), then we drop the dentry now.
2597 *
2598 * A low-level filesystem can, if it choses, legally
2599 * do a
2600 *
2601 * if (!d_unhashed(dentry))
2602 * return -EBUSY;
2603 *
2604 * if it cannot handle the case of removing a directory
2605 * that is still in use by something else..
2606 */
2607void dentry_unhash(struct dentry *dentry)
2608{
2609 dget(dentry);
dc168427 2610 shrink_dcache_parent(dentry);
1da177e4 2611 spin_lock(&dentry->d_lock);
b7ab39f6 2612 if (dentry->d_count == 2)
1da177e4
LT		 2613 __d_drop(dentry);
2614 spin_unlock(&dentry->d_lock);
1da177e4
LT		 2615}
2616
2617int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2618{
2619 int error = may_delete(dir, dentry, 1);
2620
2621 if (error)
2622 return error;
2623
acfa4380 2624 if (!dir->i_op->rmdir)
1da177e4
LT		 2625 return -EPERM;
2626
1b1dcc1b 2627 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2628 dentry_unhash(dentry);
2629 if (d_mountpoint(dentry))
2630 error = -EBUSY;
2631 else {
2632 error = security_inode_rmdir(dir, dentry);
2633 if (!error) {
2634 error = dir->i_op->rmdir(dir, dentry);
d83c49f3 2635 if (!error) {
1da177e4 2636 dentry->d_inode->i_flags |= S_DEAD;
d83c49f3
AV		 2637 dont_mount(dentry);
2638 }
1da177e4
LT		 2639 }
2640 }
1b1dcc1b 2641 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4 2642 if (!error) {
1da177e4
LT		 2643 d_delete(dentry);
2644 }
2645 dput(dentry);
2646
2647 return error;
2648}
2649
5590ff0d 2650static long do_rmdir(int dfd, const char __user *pathname)
1da177e4
LT		 2651{
2652 int error = 0;
2653 char * name;
2654 struct dentry *dentry;
2655 struct nameidata nd;
2656
2ad94ae6 2657 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2658 if (error)
2ad94ae6 2659 return error;
1da177e4
LT		 2660
2661 switch(nd.last_type) {
0612d9fb
OH		 2662 case LAST_DOTDOT:
2663 error = -ENOTEMPTY;
2664 goto exit1;
2665 case LAST_DOT:
2666 error = -EINVAL;
2667 goto exit1;
2668 case LAST_ROOT:
2669 error = -EBUSY;
2670 goto exit1;
1da177e4 2671 }
0612d9fb
OH		 2672
2673 nd.flags &= ~LOOKUP_PARENT;
2674
4ac91378 2675 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2676 dentry = lookup_hash(&nd);
1da177e4 2677 error = PTR_ERR(dentry);
6902d925
DH		 2678 if (IS_ERR(dentry))
2679 goto exit2;
0622753b
DH		 2680 error = mnt_want_write(nd.path.mnt);
2681 if (error)
2682 goto exit3;
be6d3e56
KT		 2683 error = security_path_rmdir(&nd.path, dentry);
2684 if (error)
2685 goto exit4;
4ac91378 2686 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
be6d3e56 2687exit4:
0622753b
DH		 2688 mnt_drop_write(nd.path.mnt);
2689exit3:
6902d925
DH		 2690 dput(dentry);
2691exit2:
4ac91378 2692 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2693exit1:
1d957f9b 2694 path_put(&nd.path);
1da177e4
LT		 2695 putname(name);
2696 return error;
2697}
2698
3cdad428 2699SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
5590ff0d
UD		 2700{
2701 return do_rmdir(AT_FDCWD, pathname);
2702}
2703
1da177e4
LT		 2704int vfs_unlink(struct inode *dir, struct dentry *dentry)
2705{
2706 int error = may_delete(dir, dentry, 0);
2707
2708 if (error)
2709 return error;
2710
acfa4380 2711 if (!dir->i_op->unlink)
1da177e4
LT		 2712 return -EPERM;
2713
1b1dcc1b 2714 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2715 if (d_mountpoint(dentry))
2716 error = -EBUSY;
2717 else {
2718 error = security_inode_unlink(dir, dentry);
bec1052e 2719 if (!error) {
1da177e4 2720 error = dir->i_op->unlink(dir, dentry);
bec1052e 2721 if (!error)
d83c49f3 2722 dont_mount(dentry);
bec1052e 2723 }
1da177e4 2724 }
1b1dcc1b 2725 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2726
2727 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2728 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
ece95912 2729 fsnotify_link_count(dentry->d_inode);
e234f35c 2730 d_delete(dentry);
1da177e4 2731 }
0eeca283 2732
1da177e4
LT		 2733 return error;
2734}
2735
2736/*
2737 * Make sure that the actual truncation of the file will occur outside its
1b1dcc1b 2738 * directory's i_mutex. Truncate can take a long time if there is a lot of
1da177e4
LT		 2739 * writeout happening, and we don't want to prevent access to the directory
2740 * while waiting on the I/O.
2741 */
5590ff0d 2742static long do_unlinkat(int dfd, const char __user *pathname)
1da177e4 2743{
2ad94ae6
AV		 2744 int error;
2745 char *name;
1da177e4
LT		 2746 struct dentry *dentry;
2747 struct nameidata nd;
2748 struct inode *inode = NULL;
2749
2ad94ae6 2750 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2751 if (error)
2ad94ae6
AV		 2752 return error;
2753
1da177e4
LT		 2754 error = -EISDIR;
2755 if (nd.last_type != LAST_NORM)
2756 goto exit1;
0612d9fb
OH		 2757
2758 nd.flags &= ~LOOKUP_PARENT;
2759
4ac91378 2760 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2761 dentry = lookup_hash(&nd);
1da177e4
LT		 2762 error = PTR_ERR(dentry);
2763 if (!IS_ERR(dentry)) {
2764 /* Why not before? Because we want correct error value */
2765 if (nd.last.name[nd.last.len])
2766 goto slashes;
2767 inode = dentry->d_inode;
2768 if (inode)
7de9c6ee 2769 ihold(inode);
0622753b
DH		 2770 error = mnt_want_write(nd.path.mnt);
2771 if (error)
2772 goto exit2;
be6d3e56
KT		 2773 error = security_path_unlink(&nd.path, dentry);
2774 if (error)
2775 goto exit3;
4ac91378 2776 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
be6d3e56 2777exit3:
0622753b 2778 mnt_drop_write(nd.path.mnt);
1da177e4
LT		 2779 exit2:
2780 dput(dentry);
2781 }
4ac91378 2782 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4
LT		 2783 if (inode)
2784 iput(inode); /* truncate the inode here */
2785exit1:
1d957f9b 2786 path_put(&nd.path);
1da177e4
LT		 2787 putname(name);
2788 return error;
2789
2790slashes:
2791 error = !dentry->d_inode ? -ENOENT :
2792 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2793 goto exit2;
2794}
2795
2e4d0924 2796SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
5590ff0d
UD		 2797{
2798 if ((flag & ~AT_REMOVEDIR) != 0)
2799 return -EINVAL;
2800
2801 if (flag & AT_REMOVEDIR)
2802 return do_rmdir(dfd, pathname);
2803
2804 return do_unlinkat(dfd, pathname);
2805}
2806
3480b257 2807SYSCALL_DEFINE1(unlink, const char __user *, pathname)
5590ff0d
UD		 2808{
2809 return do_unlinkat(AT_FDCWD, pathname);
2810}
2811
db2e747b 2812int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
1da177e4 2813{
a95164d9 2814 int error = may_create(dir, dentry);
1da177e4
LT		 2815
2816 if (error)
2817 return error;
2818
acfa4380 2819 if (!dir->i_op->symlink)
1da177e4
LT		 2820 return -EPERM;
2821
2822 error = security_inode_symlink(dir, dentry, oldname);
2823 if (error)
2824 return error;
2825
1da177e4 2826 error = dir->i_op->symlink(dir, dentry, oldname);
a74574aa 2827 if (!error)
f38aa942 2828 fsnotify_create(dir, dentry);
1da177e4
LT		 2829 return error;
2830}
2831
2e4d0924
HC		 2832SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2833 int, newdfd, const char __user *, newname)
1da177e4 2834{
2ad94ae6
AV		 2835 int error;
2836 char *from;
2837 char *to;
6902d925
DH		 2838 struct dentry *dentry;
2839 struct nameidata nd;
1da177e4
LT		 2840
2841 from = getname(oldname);
2ad94ae6 2842 if (IS_ERR(from))
1da177e4 2843 return PTR_ERR(from);
1da177e4 2844
2ad94ae6 2845 error = user_path_parent(newdfd, newname, &nd, &to);
6902d925 2846 if (error)
2ad94ae6
AV		 2847 goto out_putname;
2848
6902d925
DH		 2849 dentry = lookup_create(&nd, 0);
2850 error = PTR_ERR(dentry);
2851 if (IS_ERR(dentry))
2852 goto out_unlock;
2853
75c3f29d
DH		 2854 error = mnt_want_write(nd.path.mnt);
2855 if (error)
2856 goto out_dput;
be6d3e56
KT		 2857 error = security_path_symlink(&nd.path, dentry, from);
2858 if (error)
2859 goto out_drop_write;
db2e747b 2860 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
be6d3e56 2861out_drop_write:
75c3f29d
DH		 2862 mnt_drop_write(nd.path.mnt);
2863out_dput:
6902d925
DH		 2864 dput(dentry);
2865out_unlock:
4ac91378 2866 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2867 path_put(&nd.path);
6902d925
DH		 2868 putname(to);
2869out_putname:
1da177e4
LT		 2870 putname(from);
2871 return error;
2872}
2873
3480b257 2874SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 2875{
2876 return sys_symlinkat(oldname, AT_FDCWD, newname);
2877}
2878
1da177e4
LT		 2879int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2880{
2881 struct inode *inode = old_dentry->d_inode;
2882 int error;
2883
2884 if (!inode)
2885 return -ENOENT;
2886
a95164d9 2887 error = may_create(dir, new_dentry);
1da177e4
LT		 2888 if (error)
2889 return error;
2890
2891 if (dir->i_sb != inode->i_sb)
2892 return -EXDEV;
2893
2894 /*
2895 * A link to an append-only or immutable file cannot be created.
2896 */
2897 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2898 return -EPERM;
acfa4380 2899 if (!dir->i_op->link)
1da177e4 2900 return -EPERM;
7e79eedb 2901 if (S_ISDIR(inode->i_mode))
1da177e4
LT		 2902 return -EPERM;
2903
2904 error = security_inode_link(old_dentry, dir, new_dentry);
2905 if (error)
2906 return error;
2907
7e79eedb 2908 mutex_lock(&inode->i_mutex);
aae8a97d
AK		 2909 /* Make sure we don't allow creating hardlink to an unlinked file */
2910 if (inode->i_nlink == 0)
2911 error = -ENOENT;
2912 else
2913 error = dir->i_op->link(old_dentry, dir, new_dentry);
7e79eedb 2914 mutex_unlock(&inode->i_mutex);
e31e14ec 2915 if (!error)
7e79eedb 2916 fsnotify_link(dir, inode, new_dentry);
1da177e4
LT		 2917 return error;
2918}
2919
2920/*
2921 * Hardlinks are often used in delicate situations. We avoid
2922 * security-related surprises by not following symlinks on the
2923 * newname. --KAB
2924 *
2925 * We don't follow them on the oldname either to be compatible
2926 * with linux 2.0, and to avoid hard-linking to directories
2927 * and other special files. --ADM
2928 */
2e4d0924
HC		 2929SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2930 int, newdfd, const char __user *, newname, int, flags)
1da177e4
LT		 2931{
2932 struct dentry *new_dentry;
2d8f3038
AV		 2933 struct nameidata nd;
2934 struct path old_path;
1da177e4 2935 int error;
2ad94ae6 2936 char *to;
1da177e4 2937
45c9b11a 2938 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
c04030e1
UD		 2939 return -EINVAL;
2940
2d8f3038
AV		 2941 error = user_path_at(olddfd, oldname,
2942 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2943 &old_path);
1da177e4 2944 if (error)
2ad94ae6
AV		 2945 return error;
2946
2947 error = user_path_parent(newdfd, newname, &nd, &to);
1da177e4
LT		 2948 if (error)
2949 goto out;
2950 error = -EXDEV;
2d8f3038 2951 if (old_path.mnt != nd.path.mnt)
1da177e4
LT		 2952 goto out_release;
2953 new_dentry = lookup_create(&nd, 0);
2954 error = PTR_ERR(new_dentry);
6902d925
DH		 2955 if (IS_ERR(new_dentry))
2956 goto out_unlock;
75c3f29d
DH		 2957 error = mnt_want_write(nd.path.mnt);
2958 if (error)
2959 goto out_dput;
be6d3e56
KT		 2960 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2961 if (error)
2962 goto out_drop_write;
2d8f3038 2963 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
be6d3e56 2964out_drop_write:
75c3f29d
DH		 2965 mnt_drop_write(nd.path.mnt);
2966out_dput:
6902d925
DH		 2967 dput(new_dentry);
2968out_unlock:
4ac91378 2969 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2970out_release:
1d957f9b 2971 path_put(&nd.path);
2ad94ae6 2972 putname(to);
1da177e4 2973out:
2d8f3038 2974 path_put(&old_path);
1da177e4
LT		 2975
2976 return error;
2977}
2978
3480b257 2979SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
5590ff0d 2980{
c04030e1 2981 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
5590ff0d
UD		 2982}
2983
1da177e4
LT		 2984/*
2985 * The worst of all namespace operations - renaming directory. "Perverted"
2986 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2987 * Problems:
2988 * a) we can get into loop creation. Check is done in is_subdir().
2989 * b) race potential - two innocent renames can create a loop together.
2990 * That's where 4.4 screws up. Current fix: serialization on
a11f3a05 2991 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
1da177e4
LT		 2992 * story.
2993 * c) we have to lock _three_ objects - parents and victim (if it exists).
1b1dcc1b 2994 * And that - after we got ->i_mutex on parents (until then we don't know
1da177e4
LT		 2995 * whether the target exists). Solution: try to be smart with locking
2996 * order for inodes. We rely on the fact that tree topology may change
a11f3a05 2997 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
1da177e4
LT		 2998 * move will be locked. Thus we can rank directories by the tree
2999 * (ancestors first) and rank all non-directories after them.
3000 * That works since everybody except rename does "lock parent, lookup,
a11f3a05 3001 * lock child" and rename is under ->s_vfs_rename_mutex.
1da177e4
LT		 3002 * HOWEVER, it relies on the assumption that any object with ->lookup()
3003 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3004 * we'd better make sure that there's no link(2) for them.
3005 * d) some filesystems don't support opened-but-unlinked directories,
3006 * either because of layout or because they are not ready to deal with
3007 * all cases correctly. The latter will be fixed (taking this sort of
3008 * stuff into VFS), but the former is not going away. Solution: the same
3009 * trick as in rmdir().
3010 * e) conversion from fhandle to dentry may come in the wrong moment - when
1b1dcc1b 3011 * we are removing the target. Solution: we will have to grab ->i_mutex
1da177e4 3012 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
c41b20e7 3013 * ->i_mutex on parents, which works but leads to some truly excessive
1da177e4
LT		 3014 * locking].
3015 */
75c96f85
AB		 3016static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3017 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3018{
3019 int error = 0;
3020 struct inode *target;
3021
3022 /*
3023 * If we are going to change the parent - check write permissions,
3024 * we'll need to flip '..'.
3025 */
3026 if (new_dir != old_dir) {
f419a2e3 3027 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
1da177e4
LT		 3028 if (error)
3029 return error;
3030 }
3031
3032 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3033 if (error)
3034 return error;
3035
3036 target = new_dentry->d_inode;
d83c49f3 3037 if (target)
1b1dcc1b 3038 mutex_lock(&target->i_mutex);
1da177e4
LT		 3039 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3040 error = -EBUSY;
d83c49f3
AV		 3041 else {
3042 if (target)
3043 dentry_unhash(new_dentry);
1da177e4 3044 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
d83c49f3 3045 }
1da177e4 3046 if (target) {
d83c49f3 3047 if (!error) {
1da177e4 3048 target->i_flags |= S_DEAD;
d83c49f3
AV		 3049 dont_mount(new_dentry);
3050 }
1b1dcc1b 3051 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3052 if (d_unhashed(new_dentry))
3053 d_rehash(new_dentry);
3054 dput(new_dentry);
3055 }
e31e14ec 3056 if (!error)
349457cc
MF		 3057 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3058 d_move(old_dentry,new_dentry);
1da177e4
LT		 3059 return error;
3060}
3061
75c96f85
AB		 3062static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3063 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3064{
3065 struct inode *target;
3066 int error;
3067
3068 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3069 if (error)
3070 return error;
3071
3072 dget(new_dentry);
3073 target = new_dentry->d_inode;
3074 if (target)
1b1dcc1b 3075 mutex_lock(&target->i_mutex);
1da177e4
LT		 3076 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3077 error = -EBUSY;
3078 else
3079 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3080 if (!error) {
bec1052e 3081 if (target)
d83c49f3 3082 dont_mount(new_dentry);
349457cc 3083 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
1da177e4 3084 d_move(old_dentry, new_dentry);
1da177e4
LT		 3085 }
3086 if (target)
1b1dcc1b 3087 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3088 dput(new_dentry);
3089 return error;
3090}
3091
3092int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3093 struct inode *new_dir, struct dentry *new_dentry)
3094{
3095 int error;
3096 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
59b0df21 3097 const unsigned char *old_name;
1da177e4
LT		 3098
3099 if (old_dentry->d_inode == new_dentry->d_inode)
3100 return 0;
3101
3102 error = may_delete(old_dir, old_dentry, is_dir);
3103 if (error)
3104 return error;
3105
3106 if (!new_dentry->d_inode)
a95164d9 3107 error = may_create(new_dir, new_dentry);
1da177e4
LT		 3108 else
3109 error = may_delete(new_dir, new_dentry, is_dir);
3110 if (error)
3111 return error;
3112
acfa4380 3113 if (!old_dir->i_op->rename)
1da177e4
LT		 3114 return -EPERM;
3115
0eeca283
RL		 3116 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3117
1da177e4
LT		 3118 if (is_dir)
3119 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3120 else
3121 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
123df294
AV		 3122 if (!error)
3123 fsnotify_move(old_dir, new_dir, old_name, is_dir,
5a190ae6 3124 new_dentry->d_inode, old_dentry);
0eeca283
RL		 3125 fsnotify_oldname_free(old_name);
3126
1da177e4
LT		 3127 return error;
3128}
3129
2e4d0924
HC		 3130SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3131 int, newdfd, const char __user *, newname)
1da177e4 3132{
2ad94ae6
AV		 3133 struct dentry *old_dir, *new_dir;
3134 struct dentry *old_dentry, *new_dentry;
3135 struct dentry *trap;
1da177e4 3136 struct nameidata oldnd, newnd;
2ad94ae6
AV		 3137 char *from;
3138 char *to;
3139 int error;
1da177e4 3140
2ad94ae6 3141 error = user_path_parent(olddfd, oldname, &oldnd, &from);
1da177e4
LT		 3142 if (error)
3143 goto exit;
3144
2ad94ae6 3145 error = user_path_parent(newdfd, newname, &newnd, &to);
1da177e4
LT		 3146 if (error)
3147 goto exit1;
3148
3149 error = -EXDEV;
4ac91378 3150 if (oldnd.path.mnt != newnd.path.mnt)
1da177e4
LT		 3151 goto exit2;
3152
4ac91378 3153 old_dir = oldnd.path.dentry;
1da177e4
LT		 3154 error = -EBUSY;
3155 if (oldnd.last_type != LAST_NORM)
3156 goto exit2;
3157
4ac91378 3158 new_dir = newnd.path.dentry;
1da177e4
LT		 3159 if (newnd.last_type != LAST_NORM)
3160 goto exit2;
3161
0612d9fb
OH		 3162 oldnd.flags &= ~LOOKUP_PARENT;
3163 newnd.flags &= ~LOOKUP_PARENT;
4e9ed2f8 3164 newnd.flags |= LOOKUP_RENAME_TARGET;
0612d9fb 3165
1da177e4
LT		 3166 trap = lock_rename(new_dir, old_dir);
3167
49705b77 3168 old_dentry = lookup_hash(&oldnd);
1da177e4
LT		 3169 error = PTR_ERR(old_dentry);
3170 if (IS_ERR(old_dentry))
3171 goto exit3;
3172 /* source must exist */
3173 error = -ENOENT;
3174 if (!old_dentry->d_inode)
3175 goto exit4;
3176 /* unless the source is a directory trailing slashes give -ENOTDIR */
3177 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3178 error = -ENOTDIR;
3179 if (oldnd.last.name[oldnd.last.len])
3180 goto exit4;
3181 if (newnd.last.name[newnd.last.len])
3182 goto exit4;
3183 }
3184 /* source should not be ancestor of target */
3185 error = -EINVAL;
3186 if (old_dentry == trap)
3187 goto exit4;
49705b77 3188 new_dentry = lookup_hash(&newnd);
1da177e4
LT		 3189 error = PTR_ERR(new_dentry);
3190 if (IS_ERR(new_dentry))
3191 goto exit4;
3192 /* target should not be an ancestor of source */
3193 error = -ENOTEMPTY;
3194 if (new_dentry == trap)
3195 goto exit5;
3196
9079b1eb
DH		 3197 error = mnt_want_write(oldnd.path.mnt);
3198 if (error)
3199 goto exit5;
be6d3e56
KT		 3200 error = security_path_rename(&oldnd.path, old_dentry,
3201 &newnd.path, new_dentry);
3202 if (error)
3203 goto exit6;
1da177e4
LT		 3204 error = vfs_rename(old_dir->d_inode, old_dentry,
3205 new_dir->d_inode, new_dentry);
be6d3e56 3206exit6:
9079b1eb 3207 mnt_drop_write(oldnd.path.mnt);
1da177e4
LT		 3208exit5:
3209 dput(new_dentry);
3210exit4:
3211 dput(old_dentry);
3212exit3:
3213 unlock_rename(new_dir, old_dir);
3214exit2:
1d957f9b 3215 path_put(&newnd.path);
2ad94ae6 3216 putname(to);
1da177e4 3217exit1:
1d957f9b 3218 path_put(&oldnd.path);
1da177e4 3219 putname(from);
2ad94ae6 3220exit:
1da177e4
LT		 3221 return error;
3222}
3223
a26eab24 3224SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 3225{
3226 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3227}
3228
1da177e4
LT		 3229int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3230{
3231 int len;
3232
3233 len = PTR_ERR(link);
3234 if (IS_ERR(link))
3235 goto out;
3236
3237 len = strlen(link);
3238 if (len > (unsigned) buflen)
3239 len = buflen;
3240 if (copy_to_user(buffer, link, len))
3241 len = -EFAULT;
3242out:
3243 return len;
3244}
3245
3246/*
3247 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3248 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3249 * using) it for any given inode is up to filesystem.
3250 */
3251int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3252{
3253 struct nameidata nd;
cc314eef 3254 void *cookie;
694a1764 3255 int res;
cc314eef 3256
1da177e4 3257 nd.depth = 0;
cc314eef 3258 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
694a1764
MS		 3259 if (IS_ERR(cookie))
3260 return PTR_ERR(cookie);
3261
3262 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3263 if (dentry->d_inode->i_op->put_link)
3264 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3265 return res;
1da177e4
LT		 3266}
3267
3268int vfs_follow_link(struct nameidata *nd, const char *link)
3269{
3270 return __vfs_follow_link(nd, link);
3271}
3272
3273/* get the link contents into pagecache */
3274static char *page_getlink(struct dentry * dentry, struct page **ppage)
3275{
ebd09abb
DG		 3276 char *kaddr;
3277 struct page *page;
1da177e4 3278 struct address_space *mapping = dentry->d_inode->i_mapping;
090d2b18 3279 page = read_mapping_page(mapping, 0, NULL);
1da177e4 3280 if (IS_ERR(page))
6fe6900e 3281 return (char*)page;
1da177e4 3282 *ppage = page;
ebd09abb
DG		 3283 kaddr = kmap(page);
3284 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3285 return kaddr;
1da177e4
LT		 3286}
3287
3288int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3289{
3290 struct page *page = NULL;
3291 char *s = page_getlink(dentry, &page);
3292 int res = vfs_readlink(dentry,buffer,buflen,s);
3293 if (page) {
3294 kunmap(page);
3295 page_cache_release(page);
3296 }
3297 return res;
3298}
3299
cc314eef 3300void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
1da177e4 3301{
cc314eef 3302 struct page *page = NULL;
1da177e4 3303 nd_set_link(nd, page_getlink(dentry, &page));
cc314eef 3304 return page;
1da177e4
LT		 3305}
3306
cc314eef 3307void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1da177e4 3308{
cc314eef
LT		 3309 struct page *page = cookie;
3310
3311 if (page) {
1da177e4
LT		 3312 kunmap(page);
3313 page_cache_release(page);
1da177e4
LT		 3314 }
3315}
3316
54566b2c
NP		 3317/*
3318 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3319 */
3320int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
1da177e4
LT		 3321{
3322 struct address_space *mapping = inode->i_mapping;
0adb25d2 3323 struct page *page;
afddba49 3324 void *fsdata;
beb497ab 3325 int err;
1da177e4 3326 char *kaddr;
54566b2c
NP		 3327 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3328 if (nofs)
3329 flags |= AOP_FLAG_NOFS;
1da177e4 3330
7e53cac4 3331retry:
afddba49 3332 err = pagecache_write_begin(NULL, mapping, 0, len-1,
54566b2c 3333 flags, &page, &fsdata);
1da177e4 3334 if (err)
afddba49
NP		 3335 goto fail;
3336
1da177e4
LT		 3337 kaddr = kmap_atomic(page, KM_USER0);
3338 memcpy(kaddr, symname, len-1);
3339 kunmap_atomic(kaddr, KM_USER0);
afddba49
NP		 3340
3341 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3342 page, fsdata);
1da177e4
LT		 3343 if (err < 0)
3344 goto fail;
afddba49
NP		 3345 if (err < len-1)
3346 goto retry;
3347
1da177e4
LT		 3348 mark_inode_dirty(inode);
3349 return 0;
1da177e4
LT		 3350fail:
3351 return err;
3352}
3353
0adb25d2
KK		 3354int page_symlink(struct inode *inode, const char *symname, int len)
3355{
3356 return __page_symlink(inode, symname, len,
54566b2c 3357 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
0adb25d2
KK		 3358}
3359
92e1d5be 3360const struct inode_operations page_symlink_inode_operations = {
1da177e4
LT		 3361 .readlink = generic_readlink,
3362 .follow_link = page_follow_link_light,
3363 .put_link = page_put_link,
3364};
3365
2d8f3038 3366EXPORT_SYMBOL(user_path_at);
cc53ce53 3367EXPORT_SYMBOL(follow_down_one);
1da177e4
LT		 3368EXPORT_SYMBOL(follow_down);
3369EXPORT_SYMBOL(follow_up);
3370EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3371EXPORT_SYMBOL(getname);
3372EXPORT_SYMBOL(lock_rename);
1da177e4
LT		 3373EXPORT_SYMBOL(lookup_one_len);
3374EXPORT_SYMBOL(page_follow_link_light);
3375EXPORT_SYMBOL(page_put_link);
3376EXPORT_SYMBOL(page_readlink);
0adb25d2 3377EXPORT_SYMBOL(__page_symlink);
1da177e4
LT		 3378EXPORT_SYMBOL(page_symlink);
3379EXPORT_SYMBOL(page_symlink_inode_operations);
c9c6cac0 3380EXPORT_SYMBOL(kern_path_parent);
d1811465 3381EXPORT_SYMBOL(kern_path);
16f18200 3382EXPORT_SYMBOL(vfs_path_lookup);
f419a2e3 3383EXPORT_SYMBOL(inode_permission);
8c744fb8 3384EXPORT_SYMBOL(file_permission);
1da177e4
LT		 3385EXPORT_SYMBOL(unlock_rename);
3386EXPORT_SYMBOL(vfs_create);
3387EXPORT_SYMBOL(vfs_follow_link);
3388EXPORT_SYMBOL(vfs_link);
3389EXPORT_SYMBOL(vfs_mkdir);
3390EXPORT_SYMBOL(vfs_mknod);
3391EXPORT_SYMBOL(generic_permission);
3392EXPORT_SYMBOL(vfs_readlink);
3393EXPORT_SYMBOL(vfs_rename);
3394EXPORT_SYMBOL(vfs_rmdir);
3395EXPORT_SYMBOL(vfs_symlink);
3396EXPORT_SYMBOL(vfs_unlink);
3397EXPORT_SYMBOL(dentry_unhash);
3398EXPORT_SYMBOL(generic_readlink);
kernel source for telekom puls (alcatel/mediatek)