projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
switch non-create side of open() to use of do_last()
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / fs / namei.c
CommitLineData
1da177e4
LT		 1/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
18#include <linux/module.h>
19#include <linux/slab.h>
20#include <linux/fs.h>
21#include <linux/namei.h>
1da177e4 22#include <linux/pagemap.h>
0eeca283 23#include <linux/fsnotify.h>
1da177e4
LT		 24#include <linux/personality.h>
25#include <linux/security.h>
6146f0d5 26#include <linux/ima.h>
1da177e4
LT		 27#include <linux/syscalls.h>
28#include <linux/mount.h>
29#include <linux/audit.h>
16f7e0fe 30#include <linux/capability.h>
834f2a4a 31#include <linux/file.h>
5590ff0d 32#include <linux/fcntl.h>
08ce5f16 33#include <linux/device_cgroup.h>
5ad4e53b 34#include <linux/fs_struct.h>
1da177e4
LT		 35#include <asm/uaccess.h>
36
e81e3f4d
EP		 37#include "internal.h"
38
1da177e4
LT		 39/* [Feb-1997 T. Schoebel-Theuer]
40 * Fundamental changes in the pathname lookup mechanisms (namei)
41 * were necessary because of omirr. The reason is that omirr needs
42 * to know the _real_ pathname, not the user-supplied one, in case
43 * of symlinks (and also when transname replacements occur).
44 *
45 * The new code replaces the old recursive symlink resolution with
46 * an iterative one (in case of non-nested symlink chains). It does
47 * this with calls to <fs>_follow_link().
48 * As a side effect, dir_namei(), _namei() and follow_link() are now
49 * replaced with a single function lookup_dentry() that can handle all
50 * the special cases of the former code.
51 *
52 * With the new dcache, the pathname is stored at each inode, at least as
53 * long as the refcount of the inode is positive. As a side effect, the
54 * size of the dcache depends on the inode cache and thus is dynamic.
55 *
56 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57 * resolution to correspond with current state of the code.
58 *
59 * Note that the symlink resolution is not *completely* iterative.
60 * There is still a significant amount of tail- and mid- recursion in
61 * the algorithm. Also, note that <fs>_readlink() is not used in
62 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63 * may return different results than <fs>_follow_link(). Many virtual
64 * filesystems (including /proc) exhibit this behavior.
65 */
66
67/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69 * and the name already exists in form of a symlink, try to create the new
70 * name indicated by the symlink. The old code always complained that the
71 * name already exists, due to not following the symlink even if its target
72 * is nonexistent. The new semantics affects also mknod() and link() when
73 * the name is a symlink pointing to a non-existant name.
74 *
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
82 */
83
84/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
86 *
87 * [10-Sep-98 Alan Modra] Another symlink change.
88 */
89
90/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
97 *
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
103 */
104/*
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
a11f3a05 106 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
1da177e4
LT		 107 * any extra contention...
108 */
109
110/* In order to reduce some races, while at the same time doing additional
111 * checking and hopefully speeding things up, we copy filenames to the
112 * kernel data space before using them..
113 *
114 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
115 * PATH_MAX includes the nul terminator --RR.
116 */
858119e1 117static int do_getname(const char __user *filename, char *page)
1da177e4
LT		 118{
119 int retval;
120 unsigned long len = PATH_MAX;
121
122 if (!segment_eq(get_fs(), KERNEL_DS)) {
123 if ((unsigned long) filename >= TASK_SIZE)
124 return -EFAULT;
125 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
126 len = TASK_SIZE - (unsigned long) filename;
127 }
128
129 retval = strncpy_from_user(page, filename, len);
130 if (retval > 0) {
131 if (retval < len)
132 return 0;
133 return -ENAMETOOLONG;
134 } else if (!retval)
135 retval = -ENOENT;
136 return retval;
137}
138
139char * getname(const char __user * filename)
140{
141 char *tmp, *result;
142
143 result = ERR_PTR(-ENOMEM);
144 tmp = __getname();
145 if (tmp) {
146 int retval = do_getname(filename, tmp);
147
148 result = tmp;
149 if (retval < 0) {
150 __putname(tmp);
151 result = ERR_PTR(retval);
152 }
153 }
154 audit_getname(result);
155 return result;
156}
157
158#ifdef CONFIG_AUDITSYSCALL
159void putname(const char *name)
160{
5ac3a9c2 161 if (unlikely(!audit_dummy_context()))
1da177e4
LT		 162 audit_putname(name);
163 else
164 __putname(name);
165}
166EXPORT_SYMBOL(putname);
167#endif
168
5909ccaa
LT		 169/*
170 * This does basic POSIX ACL permission checking
1da177e4 171 */
b74c79e9
NP		 172static int acl_permission_check(struct inode *inode, int mask, unsigned int flags,
173 int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
1da177e4
LT		 174{
175 umode_t mode = inode->i_mode;
176
e6305c43
AV		 177 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
178
da9592ed 179 if (current_fsuid() == inode->i_uid)
1da177e4
LT		 180 mode >>= 6;
181 else {
182 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
b74c79e9
NP		 183 int error = check_acl(inode, mask, flags);
184 if (error != -EAGAIN)
185 return error;
1da177e4
LT		 186 }
187
188 if (in_group_p(inode->i_gid))
189 mode >>= 3;
190 }
191
192 /*
193 * If the DACs are ok we don't need any capability check.
194 */
e6305c43 195 if ((mask & ~mode) == 0)
1da177e4 196 return 0;
5909ccaa
LT		 197 return -EACCES;
198}
199
200/**
b74c79e9 201 * generic_permission - check for access rights on a Posix-like filesystem
5909ccaa
LT		 202 * @inode: inode to check access rights for
203 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
204 * @check_acl: optional callback to check for Posix ACLs
39191628 205 * @flags: IPERM_FLAG_ flags.
5909ccaa
LT		 206 *
207 * Used to check for read/write/execute permissions on a file.
208 * We use "fsuid" for this, letting us set arbitrary permissions
209 * for filesystem access without changing the "normal" uids which
b74c79e9
NP		 210 * are used for other things.
211 *
212 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
213 * request cannot be satisfied (eg. requires blocking or too much complexity).
214 * It would then be called again in ref-walk mode.
5909ccaa 215 */
b74c79e9
NP		 216int generic_permission(struct inode *inode, int mask, unsigned int flags,
217 int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
5909ccaa
LT		 218{
219 int ret;
220
221 /*
222 * Do the basic POSIX ACL permission checks.
223 */
b74c79e9 224 ret = acl_permission_check(inode, mask, flags, check_acl);
5909ccaa
LT		 225 if (ret != -EACCES)
226 return ret;
1da177e4 227
1da177e4
LT		 228 /*
229 * Read/write DACs are always overridable.
230 * Executable DACs are overridable if at least one exec bit is set.
231 */
f696a365 232 if (!(mask & MAY_EXEC) || execute_ok(inode))
1da177e4
LT		 233 if (capable(CAP_DAC_OVERRIDE))
234 return 0;
235
236 /*
237 * Searching includes executable on directories, else just read.
238 */
7ea66001 239 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
1da177e4
LT		 240 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
241 if (capable(CAP_DAC_READ_SEARCH))
242 return 0;
243
244 return -EACCES;
245}
246
cb23beb5
CH		 247/**
248 * inode_permission - check for access rights to a given inode
249 * @inode: inode to check permission on
250 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
251 *
252 * Used to check for read/write/execute permissions on an inode.
253 * We use "fsuid" for this, letting us set arbitrary permissions
254 * for filesystem access without changing the "normal" uids which
255 * are used for other things.
256 */
f419a2e3 257int inode_permission(struct inode *inode, int mask)
1da177e4 258{
e6305c43 259 int retval;
1da177e4
LT		 260
261 if (mask & MAY_WRITE) {
22590e41 262 umode_t mode = inode->i_mode;
1da177e4
LT		 263
264 /*
265 * Nobody gets write access to a read-only fs.
266 */
267 if (IS_RDONLY(inode) &&
268 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
269 return -EROFS;
270
271 /*
272 * Nobody gets write access to an immutable file.
273 */
274 if (IS_IMMUTABLE(inode))
275 return -EACCES;
276 }
277
acfa4380 278 if (inode->i_op->permission)
b74c79e9 279 retval = inode->i_op->permission(inode, mask, 0);
f696a365 280 else
b74c79e9
NP		 281 retval = generic_permission(inode, mask, 0,
282 inode->i_op->check_acl);
f696a365 283
1da177e4
LT		 284 if (retval)
285 return retval;
286
08ce5f16
SH		 287 retval = devcgroup_inode_permission(inode, mask);
288 if (retval)
289 return retval;
290
d09ca739 291 return security_inode_permission(inode, mask);
1da177e4
LT		 292}
293
8c744fb8
CH		 294/**
295 * file_permission - check for additional access rights to a given file
296 * @file: file to check access rights for
297 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
298 *
299 * Used to check for read/write/execute permissions on an already opened
300 * file.
301 *
302 * Note:
303 * Do not use this function in new code. All access checks should
cb23beb5 304 * be done using inode_permission().
8c744fb8
CH		 305 */
306int file_permission(struct file *file, int mask)
307{
f419a2e3 308 return inode_permission(file->f_path.dentry->d_inode, mask);
8c744fb8
CH		 309}
310
1da177e4
LT		 311/*
312 * get_write_access() gets write permission for a file.
313 * put_write_access() releases this write permission.
314 * This is used for regular files.
315 * We cannot support write (and maybe mmap read-write shared) accesses and
316 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
317 * can have the following values:
318 * 0: no writers, no VM_DENYWRITE mappings
319 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
320 * > 0: (i_writecount) users are writing to the file.
321 *
322 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
323 * except for the cases where we don't hold i_writecount yet. Then we need to
324 * use {get,deny}_write_access() - these functions check the sign and refuse
325 * to do the change if sign is wrong. Exclusion between them is provided by
326 * the inode->i_lock spinlock.
327 */
328
329int get_write_access(struct inode * inode)
330{
331 spin_lock(&inode->i_lock);
332 if (atomic_read(&inode->i_writecount) < 0) {
333 spin_unlock(&inode->i_lock);
334 return -ETXTBSY;
335 }
336 atomic_inc(&inode->i_writecount);
337 spin_unlock(&inode->i_lock);
338
339 return 0;
340}
341
342int deny_write_access(struct file * file)
343{
0f7fc9e4 344 struct inode *inode = file->f_path.dentry->d_inode;
1da177e4
LT		 345
346 spin_lock(&inode->i_lock);
347 if (atomic_read(&inode->i_writecount) > 0) {
348 spin_unlock(&inode->i_lock);
349 return -ETXTBSY;
350 }
351 atomic_dec(&inode->i_writecount);
352 spin_unlock(&inode->i_lock);
353
354 return 0;
355}
356
5dd784d0
JB		 357/**
358 * path_get - get a reference to a path
359 * @path: path to get the reference to
360 *
361 * Given a path increment the reference count to the dentry and the vfsmount.
362 */
363void path_get(struct path *path)
364{
365 mntget(path->mnt);
366 dget(path->dentry);
367}
368EXPORT_SYMBOL(path_get);
369
1d957f9b
JB		 370/**
371 * path_put - put a reference to a path
372 * @path: path to put the reference to
373 *
374 * Given a path decrement the reference count to the dentry and the vfsmount.
375 */
376void path_put(struct path *path)
1da177e4 377{
1d957f9b
JB		 378 dput(path->dentry);
379 mntput(path->mnt);
1da177e4 380}
1d957f9b 381EXPORT_SYMBOL(path_put);
1da177e4 382
31e6b01f
NP		 383/**
384 * nameidata_drop_rcu - drop this nameidata out of rcu-walk
385 * @nd: nameidata pathwalk data to drop
39191628 386 * Returns: 0 on success, -ECHILD on failure
31e6b01f
NP		 387 *
388 * Path walking has 2 modes, rcu-walk and ref-walk (see
389 * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt
390 * to drop out of rcu-walk mode and take normal reference counts on dentries
391 * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take
392 * refcounts at the last known good point before rcu-walk got stuck, so
393 * ref-walk may continue from there. If this is not successful (eg. a seqcount
394 * has changed), then failure is returned and path walk restarts from the
395 * beginning in ref-walk mode.
396 *
397 * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into
398 * ref-walk. Must be called from rcu-walk context.
399 */
400static int nameidata_drop_rcu(struct nameidata *nd)
401{
402 struct fs_struct *fs = current->fs;
403 struct dentry *dentry = nd->path.dentry;
404
405 BUG_ON(!(nd->flags & LOOKUP_RCU));
406 if (nd->root.mnt) {
407 spin_lock(&fs->lock);
408 if (nd->root.mnt != fs->root.mnt ||
409 nd->root.dentry != fs->root.dentry)
410 goto err_root;
411 }
412 spin_lock(&dentry->d_lock);
413 if (!__d_rcu_to_refcount(dentry, nd->seq))
414 goto err;
415 BUG_ON(nd->inode != dentry->d_inode);
416 spin_unlock(&dentry->d_lock);
417 if (nd->root.mnt) {
418 path_get(&nd->root);
419 spin_unlock(&fs->lock);
420 }
421 mntget(nd->path.mnt);
422
423 rcu_read_unlock();
424 br_read_unlock(vfsmount_lock);
425 nd->flags &= ~LOOKUP_RCU;
426 return 0;
427err:
428 spin_unlock(&dentry->d_lock);
429err_root:
430 if (nd->root.mnt)
431 spin_unlock(&fs->lock);
432 return -ECHILD;
433}
434
435/* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
436static inline int nameidata_drop_rcu_maybe(struct nameidata *nd)
437{
438 if (nd->flags & LOOKUP_RCU)
439 return nameidata_drop_rcu(nd);
440 return 0;
441}
442
443/**
444 * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk
445 * @nd: nameidata pathwalk data to drop
446 * @dentry: dentry to drop
39191628 447 * Returns: 0 on success, -ECHILD on failure
31e6b01f
NP		 448 *
449 * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root,
450 * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on
451 * @nd. Must be called from rcu-walk context.
452 */
453static int nameidata_dentry_drop_rcu(struct nameidata *nd, struct dentry *dentry)
454{
455 struct fs_struct *fs = current->fs;
456 struct dentry *parent = nd->path.dentry;
457
458 BUG_ON(!(nd->flags & LOOKUP_RCU));
459 if (nd->root.mnt) {
460 spin_lock(&fs->lock);
461 if (nd->root.mnt != fs->root.mnt ||
462 nd->root.dentry != fs->root.dentry)
463 goto err_root;
464 }
465 spin_lock(&parent->d_lock);
466 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
467 if (!__d_rcu_to_refcount(dentry, nd->seq))
468 goto err;
469 /*
470 * If the sequence check on the child dentry passed, then the child has
471 * not been removed from its parent. This means the parent dentry must
472 * be valid and able to take a reference at this point.
473 */
474 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
475 BUG_ON(!parent->d_count);
476 parent->d_count++;
477 spin_unlock(&dentry->d_lock);
478 spin_unlock(&parent->d_lock);
479 if (nd->root.mnt) {
480 path_get(&nd->root);
481 spin_unlock(&fs->lock);
482 }
483 mntget(nd->path.mnt);
484
485 rcu_read_unlock();
486 br_read_unlock(vfsmount_lock);
487 nd->flags &= ~LOOKUP_RCU;
488 return 0;
489err:
490 spin_unlock(&dentry->d_lock);
491 spin_unlock(&parent->d_lock);
492err_root:
493 if (nd->root.mnt)
494 spin_unlock(&fs->lock);
495 return -ECHILD;
496}
497
498/* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
499static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata *nd, struct dentry *dentry)
500{
a7472bab
AV		 501 if (nd->flags & LOOKUP_RCU) {
502 if (unlikely(nameidata_dentry_drop_rcu(nd, dentry))) {
503 nd->flags &= ~LOOKUP_RCU;
504 nd->root.mnt = NULL;
505 rcu_read_unlock();
506 br_read_unlock(vfsmount_lock);
507 return -ECHILD;
508 }
509 }
31e6b01f
NP		 510 return 0;
511}
512
513/**
514 * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk
515 * @nd: nameidata pathwalk data to drop
39191628 516 * Returns: 0 on success, -ECHILD on failure
31e6b01f
NP		 517 *
518 * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk.
519 * nd->path should be the final element of the lookup, so nd->root is discarded.
520 * Must be called from rcu-walk context.
521 */
522static int nameidata_drop_rcu_last(struct nameidata *nd)
523{
524 struct dentry *dentry = nd->path.dentry;
525
526 BUG_ON(!(nd->flags & LOOKUP_RCU));
527 nd->flags &= ~LOOKUP_RCU;
528 nd->root.mnt = NULL;
529 spin_lock(&dentry->d_lock);
530 if (!__d_rcu_to_refcount(dentry, nd->seq))
531 goto err_unlock;
532 BUG_ON(nd->inode != dentry->d_inode);
533 spin_unlock(&dentry->d_lock);
534
535 mntget(nd->path.mnt);
536
537 rcu_read_unlock();
538 br_read_unlock(vfsmount_lock);
539
540 return 0;
541
542err_unlock:
543 spin_unlock(&dentry->d_lock);
544 rcu_read_unlock();
545 br_read_unlock(vfsmount_lock);
546 return -ECHILD;
547}
548
834f2a4a
TM		 549/**
550 * release_open_intent - free up open intent resources
551 * @nd: pointer to nameidata
552 */
553void release_open_intent(struct nameidata *nd)
554{
2dab5974
LT		 555 struct file *file = nd->intent.open.file;
556
557 if (file && !IS_ERR(file)) {
558 if (file->f_path.dentry == NULL)
559 put_filp(file);
560 else
561 fput(file);
562 }
834f2a4a
TM		 563}
564
f60aef7e 565static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
34286d66 566{
f60aef7e 567 return dentry->d_op->d_revalidate(dentry, nd);
34286d66
NP		 568}
569
f5e1c1c1 570static struct dentry *
bcdc5e01
IK		 571do_revalidate(struct dentry *dentry, struct nameidata *nd)
572{
f5e1c1c1 573 int status = d_revalidate(dentry, nd);
bcdc5e01
IK		 574 if (unlikely(status <= 0)) {
575 /*
576 * The dentry failed validation.
577 * If d_revalidate returned 0 attempt to invalidate
578 * the dentry otherwise d_revalidate is asking us
579 * to return a fail status.
580 */
34286d66 581 if (status < 0) {
f5e1c1c1 582 dput(dentry);
34286d66 583 dentry = ERR_PTR(status);
f5e1c1c1
AV		 584 } else if (!d_invalidate(dentry)) {
585 dput(dentry);
586 dentry = NULL;
bcdc5e01
IK		 587 }
588 }
589 return dentry;
590}
591
f5e1c1c1
AV		 592static inline struct dentry *
593do_revalidate_rcu(struct dentry *dentry, struct nameidata *nd)
594{
f60aef7e 595 int status = d_revalidate(dentry, nd);
f5e1c1c1
AV		 596 if (likely(status > 0))
597 return dentry;
598 if (status == -ECHILD) {
599 if (nameidata_dentry_drop_rcu(nd, dentry))
600 return ERR_PTR(-ECHILD);
601 return do_revalidate(dentry, nd);
602 }
603 if (status < 0)
604 return ERR_PTR(status);
605 /* Don't d_invalidate in rcu-walk mode */
606 if (nameidata_dentry_drop_rcu(nd, dentry))
607 return ERR_PTR(-ECHILD);
608 if (!d_invalidate(dentry)) {
609 dput(dentry);
610 dentry = NULL;
611 }
612 return dentry;
613}
614
39159de2 615/*
16c2cd71 616 * handle_reval_path - force revalidation of a dentry
39159de2
JL		 617 *
618 * In some situations the path walking code will trust dentries without
619 * revalidating them. This causes problems for filesystems that depend on
620 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
621 * (which indicates that it's possible for the dentry to go stale), force
622 * a d_revalidate call before proceeding.
623 *
624 * Returns 0 if the revalidation was successful. If the revalidation fails,
625 * either return the error returned by d_revalidate or -ESTALE if the
626 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
627 * invalidate the dentry. It's up to the caller to handle putting references
628 * to the path if necessary.
629 */
16c2cd71 630static inline int handle_reval_path(struct nameidata *nd)
39159de2 631{
16c2cd71 632 struct dentry *dentry = nd->path.dentry;
39159de2 633 int status;
39159de2 634
16c2cd71
AV		 635 if (likely(!(nd->flags & LOOKUP_JUMPED)))
636 return 0;
637
638 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
39159de2
JL		 639 return 0;
640
16c2cd71
AV		 641 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
642 return 0;
643
644 /* Note: we do not d_invalidate() */
34286d66 645 status = d_revalidate(dentry, nd);
39159de2
JL		 646 if (status > 0)
647 return 0;
648
16c2cd71 649 if (!status)
39159de2 650 status = -ESTALE;
16c2cd71 651
39159de2
JL		 652 return status;
653}
654
1da177e4 655/*
b75b5086
AV		 656 * Short-cut version of permission(), for calling on directories
657 * during pathname resolution. Combines parts of permission()
658 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
1da177e4
LT		 659 *
660 * If appropriate, check DAC only. If not appropriate, or
b75b5086 661 * short-cut DAC fails, then call ->permission() to do more
1da177e4
LT		 662 * complete permission check.
663 */
b74c79e9 664static inline int exec_permission(struct inode *inode, unsigned int flags)
1da177e4 665{
5909ccaa 666 int ret;
1da177e4 667
cb9179ea 668 if (inode->i_op->permission) {
b74c79e9
NP		 669 ret = inode->i_op->permission(inode, MAY_EXEC, flags);
670 } else {
671 ret = acl_permission_check(inode, MAY_EXEC, flags,
672 inode->i_op->check_acl);
cb9179ea 673 }
b74c79e9 674 if (likely(!ret))
1da177e4 675 goto ok;
b74c79e9 676 if (ret == -ECHILD)
31e6b01f 677 return ret;
1da177e4 678
f1ac9f6b 679 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
1da177e4
LT		 680 goto ok;
681
5909ccaa 682 return ret;
1da177e4 683ok:
b74c79e9 684 return security_inode_exec_permission(inode, flags);
1da177e4
LT		 685}
686
2a737871
AV		 687static __always_inline void set_root(struct nameidata *nd)
688{
f7ad3c6b
MS		 689 if (!nd->root.mnt)
690 get_fs_root(current->fs, &nd->root);
2a737871
AV		 691}
692
6de88d72
AV		 693static int link_path_walk(const char *, struct nameidata *);
694
31e6b01f
NP		 695static __always_inline void set_root_rcu(struct nameidata *nd)
696{
697 if (!nd->root.mnt) {
698 struct fs_struct *fs = current->fs;
c28cc364
NP		 699 unsigned seq;
700
701 do {
702 seq = read_seqcount_begin(&fs->seq);
703 nd->root = fs->root;
704 } while (read_seqcount_retry(&fs->seq, seq));
31e6b01f
NP		 705 }
706}
707
f1662356 708static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
1da177e4 709{
31e6b01f
NP		 710 int ret;
711
1da177e4
LT		 712 if (IS_ERR(link))
713 goto fail;
714
715 if (*link == '/') {
2a737871 716 set_root(nd);
1d957f9b 717 path_put(&nd->path);
2a737871
AV		 718 nd->path = nd->root;
719 path_get(&nd->root);
16c2cd71 720 nd->flags |= LOOKUP_JUMPED;
1da177e4 721 }
31e6b01f 722 nd->inode = nd->path.dentry->d_inode;
b4091d5f 723
31e6b01f
NP		 724 ret = link_path_walk(link, nd);
725 return ret;
1da177e4 726fail:
1d957f9b 727 path_put(&nd->path);
1da177e4
LT		 728 return PTR_ERR(link);
729}
730
1d957f9b 731static void path_put_conditional(struct path *path, struct nameidata *nd)
051d3812
IK		 732{
733 dput(path->dentry);
4ac91378 734 if (path->mnt != nd->path.mnt)
051d3812
IK		 735 mntput(path->mnt);
736}
737
7b9337aa
NP		 738static inline void path_to_nameidata(const struct path *path,
739 struct nameidata *nd)
051d3812 740{
31e6b01f
NP		 741 if (!(nd->flags & LOOKUP_RCU)) {
742 dput(nd->path.dentry);
743 if (nd->path.mnt != path->mnt)
744 mntput(nd->path.mnt);
9a229683 745 }
31e6b01f 746 nd->path.mnt = path->mnt;
4ac91378 747 nd->path.dentry = path->dentry;
051d3812
IK		 748}
749
def4af30 750static __always_inline int
7b9337aa 751__do_follow_link(const struct path *link, struct nameidata *nd, void **p)
1da177e4
LT		 752{
753 int error;
7b9337aa 754 struct dentry *dentry = link->dentry;
1da177e4 755
844a3917
AV		 756 BUG_ON(nd->flags & LOOKUP_RCU);
757
7b9337aa 758 touch_atime(link->mnt, dentry);
1da177e4 759 nd_set_link(nd, NULL);
cd4e91d3 760
87556ef1
DH		 761 if (link->mnt == nd->path.mnt)
762 mntget(link->mnt);
31e6b01f 763
36f3b4f6
AV		 764 error = security_inode_follow_link(link->dentry, nd);
765 if (error) {
766 *p = ERR_PTR(error); /* no ->put_link(), please */
767 path_put(&nd->path);
768 return error;
769 }
770
86acdca1 771 nd->last_type = LAST_BIND;
def4af30
AV		 772 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
773 error = PTR_ERR(*p);
774 if (!IS_ERR(*p)) {
1da177e4 775 char *s = nd_get_link(nd);
cc314eef 776 error = 0;
1da177e4
LT		 777 if (s)
778 error = __vfs_follow_link(nd, s);
16c2cd71
AV		 779 else if (nd->last_type == LAST_BIND)
780 nd->flags |= LOOKUP_JUMPED;
1da177e4 781 }
1da177e4
LT		 782 return error;
783}
784
785/*
786 * This limits recursive symlink follows to 8, while
787 * limiting consecutive symlinks to 40.
788 *
789 * Without that kind of total limit, nasty chains of consecutive
790 * symlinks can cause almost arbitrarily long lookups.
791 */
3abb17e8 792static inline int do_follow_link(struct inode *inode, struct path *path, struct nameidata *nd)
1da177e4 793{
def4af30 794 void *cookie;
1da177e4 795 int err = -ELOOP;
844a3917
AV		 796
797 /* We drop rcu-walk here */
798 if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry))
799 return -ECHILD;
3abb17e8 800 BUG_ON(inode != path->dentry->d_inode);
844a3917 801
1da177e4
LT		 802 if (current->link_count >= MAX_NESTED_LINKS)
803 goto loop;
804 if (current->total_link_count >= 40)
805 goto loop;
806 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
807 cond_resched();
1da177e4
LT		 808 current->link_count++;
809 current->total_link_count++;
810 nd->depth++;
def4af30
AV		 811 err = __do_follow_link(path, nd, &cookie);
812 if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link)
813 path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie);
258fa999 814 path_put(path);
839d9f93
AV		 815 current->link_count--;
816 nd->depth--;
1da177e4
LT		 817 return err;
818loop:
1d957f9b
JB		 819 path_put_conditional(path, nd);
820 path_put(&nd->path);
1da177e4
LT		 821 return err;
822}
823
31e6b01f
NP		 824static int follow_up_rcu(struct path *path)
825{
826 struct vfsmount *parent;
827 struct dentry *mountpoint;
828
829 parent = path->mnt->mnt_parent;
830 if (parent == path->mnt)
831 return 0;
832 mountpoint = path->mnt->mnt_mountpoint;
833 path->dentry = mountpoint;
834 path->mnt = parent;
835 return 1;
836}
837
bab77ebf 838int follow_up(struct path *path)
1da177e4
LT		 839{
840 struct vfsmount *parent;
841 struct dentry *mountpoint;
99b7db7b
NP		 842
843 br_read_lock(vfsmount_lock);
bab77ebf
AV		 844 parent = path->mnt->mnt_parent;
845 if (parent == path->mnt) {
99b7db7b 846 br_read_unlock(vfsmount_lock);
1da177e4
LT		 847 return 0;
848 }
849 mntget(parent);
bab77ebf 850 mountpoint = dget(path->mnt->mnt_mountpoint);
99b7db7b 851 br_read_unlock(vfsmount_lock);
bab77ebf
AV		 852 dput(path->dentry);
853 path->dentry = mountpoint;
854 mntput(path->mnt);
855 path->mnt = parent;
1da177e4
LT		 856 return 1;
857}
858
b5c84bf6 859/*
9875cf80
DH		 860 * Perform an automount
861 * - return -EISDIR to tell follow_managed() to stop and return the path we
862 * were called with.
1da177e4 863 */
9875cf80
DH		 864static int follow_automount(struct path *path, unsigned flags,
865 bool *need_mntput)
31e6b01f 866{
9875cf80 867 struct vfsmount *mnt;
ea5b778a 868 int err;
9875cf80
DH		 869
870 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
871 return -EREMOTE;
872
6f45b656
DH		 873 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
874 * and this is the terminal part of the path.
875 */
876 if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE))
877 return -EISDIR; /* we actually want to stop here */
878
9875cf80
DH		 879 /* We want to mount if someone is trying to open/create a file of any
880 * type under the mountpoint, wants to traverse through the mountpoint
881 * or wants to open the mounted directory.
882 *
883 * We don't want to mount if someone's just doing a stat and they've
884 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
885 * appended a '/' to the name.
886 */
887 if (!(flags & LOOKUP_FOLLOW) &&
888 !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY |
889 LOOKUP_OPEN | LOOKUP_CREATE)))
890 return -EISDIR;
891
892 current->total_link_count++;
893 if (current->total_link_count >= 40)
894 return -ELOOP;
895
896 mnt = path->dentry->d_op->d_automount(path);
897 if (IS_ERR(mnt)) {
898 /*
899 * The filesystem is allowed to return -EISDIR here to indicate
900 * it doesn't want to automount. For instance, autofs would do
901 * this so that its userspace daemon can mount on this dentry.
902 *
903 * However, we can only permit this if it's a terminal point in
904 * the path being looked up; if it wasn't then the remainder of
905 * the path is inaccessible and we should say so.
906 */
907 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE))
908 return -EREMOTE;
909 return PTR_ERR(mnt);
31e6b01f 910 }
ea5b778a 911
9875cf80
DH		 912 if (!mnt) /* mount collision */
913 return 0;
31e6b01f 914
19a167af 915 err = finish_automount(mnt, path);
9875cf80 916
ea5b778a
DH		 917 switch (err) {
918 case -EBUSY:
919 /* Someone else made a mount here whilst we were busy */
19a167af 920 return 0;
ea5b778a 921 case 0:
ea5b778a
DH		 922 dput(path->dentry);
923 if (*need_mntput)
924 mntput(path->mnt);
925 path->mnt = mnt;
926 path->dentry = dget(mnt->mnt_root);
927 *need_mntput = true;
928 return 0;
19a167af
AV		 929 default:
930 return err;
ea5b778a 931 }
19a167af 932
463ffb2e
AV		 933}
934
9875cf80
DH		 935/*
936 * Handle a dentry that is managed in some way.
cc53ce53 937 * - Flagged for transit management (autofs)
9875cf80
DH		 938 * - Flagged as mountpoint
939 * - Flagged as automount point
940 *
941 * This may only be called in refwalk mode.
942 *
943 * Serialization is taken care of in namespace.c
944 */
945static int follow_managed(struct path *path, unsigned flags)
1da177e4 946{
9875cf80
DH		 947 unsigned managed;
948 bool need_mntput = false;
949 int ret;
950
951 /* Given that we're not holding a lock here, we retain the value in a
952 * local variable for each dentry as we look at it so that we don't see
953 * the components of that value change under us */
954 while (managed = ACCESS_ONCE(path->dentry->d_flags),
955 managed &= DCACHE_MANAGED_DENTRY,
956 unlikely(managed != 0)) {
cc53ce53
DH		 957 /* Allow the filesystem to manage the transit without i_mutex
958 * being held. */
959 if (managed & DCACHE_MANAGE_TRANSIT) {
960 BUG_ON(!path->dentry->d_op);
961 BUG_ON(!path->dentry->d_op->d_manage);
ab90911f
DH		 962 ret = path->dentry->d_op->d_manage(path->dentry,
963 false, false);
cc53ce53
DH		 964 if (ret < 0)
965 return ret == -EISDIR ? 0 : ret;
966 }
967
9875cf80
DH		 968 /* Transit to a mounted filesystem. */
969 if (managed & DCACHE_MOUNTED) {
970 struct vfsmount *mounted = lookup_mnt(path);
971 if (mounted) {
972 dput(path->dentry);
973 if (need_mntput)
974 mntput(path->mnt);
975 path->mnt = mounted;
976 path->dentry = dget(mounted->mnt_root);
977 need_mntput = true;
978 continue;
979 }
980
981 /* Something is mounted on this dentry in another
982 * namespace and/or whatever was mounted there in this
983 * namespace got unmounted before we managed to get the
984 * vfsmount_lock */
985 }
986
987 /* Handle an automount point */
988 if (managed & DCACHE_NEED_AUTOMOUNT) {
989 ret = follow_automount(path, flags, &need_mntput);
990 if (ret < 0)
991 return ret == -EISDIR ? 0 : ret;
992 continue;
993 }
994
995 /* We didn't change the current path point */
996 break;
1da177e4 997 }
9875cf80 998 return 0;
1da177e4
LT		 999}
1000
cc53ce53 1001int follow_down_one(struct path *path)
1da177e4
LT		 1002{
1003 struct vfsmount *mounted;
1004
1c755af4 1005 mounted = lookup_mnt(path);
1da177e4 1006 if (mounted) {
9393bd07
AV		 1007 dput(path->dentry);
1008 mntput(path->mnt);
1009 path->mnt = mounted;
1010 path->dentry = dget(mounted->mnt_root);
1da177e4
LT		 1011 return 1;
1012 }
1013 return 0;
1014}
1015
9875cf80
DH		 1016/*
1017 * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we
cc53ce53 1018 * meet a managed dentry and we're not walking to "..". True is returned to
9875cf80
DH		 1019 * continue, false to abort.
1020 */
1021static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1022 struct inode **inode, bool reverse_transit)
1023{
1024 while (d_mountpoint(path->dentry)) {
1025 struct vfsmount *mounted;
ab90911f
DH		 1026 if (unlikely(path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) &&
1027 !reverse_transit &&
1028 path->dentry->d_op->d_manage(path->dentry, false, true) < 0)
1029 return false;
9875cf80
DH		 1030 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
1031 if (!mounted)
1032 break;
1033 path->mnt = mounted;
1034 path->dentry = mounted->mnt_root;
1035 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1036 *inode = path->dentry->d_inode;
1037 }
1038
1039 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1040 return reverse_transit;
1041 return true;
1042}
1043
31e6b01f
NP		 1044static int follow_dotdot_rcu(struct nameidata *nd)
1045{
1046 struct inode *inode = nd->inode;
1047
1048 set_root_rcu(nd);
1049
9875cf80 1050 while (1) {
31e6b01f
NP		 1051 if (nd->path.dentry == nd->root.dentry &&
1052 nd->path.mnt == nd->root.mnt) {
1053 break;
1054 }
1055 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1056 struct dentry *old = nd->path.dentry;
1057 struct dentry *parent = old->d_parent;
1058 unsigned seq;
1059
1060 seq = read_seqcount_begin(&parent->d_seq);
1061 if (read_seqcount_retry(&old->d_seq, nd->seq))
ef7562d5 1062 goto failed;
31e6b01f
NP		 1063 inode = parent->d_inode;
1064 nd->path.dentry = parent;
1065 nd->seq = seq;
1066 break;
1067 }
1068 if (!follow_up_rcu(&nd->path))
1069 break;
1070 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1071 inode = nd->path.dentry->d_inode;
1072 }
9875cf80 1073 __follow_mount_rcu(nd, &nd->path, &inode, true);
31e6b01f 1074 nd->inode = inode;
31e6b01f 1075 return 0;
ef7562d5
AV		 1076
1077failed:
1078 nd->flags &= ~LOOKUP_RCU;
1079 nd->root.mnt = NULL;
1080 rcu_read_unlock();
1081 br_read_unlock(vfsmount_lock);
1082 return -ECHILD;
31e6b01f
NP		 1083}
1084
cc53ce53
DH		 1085/*
1086 * Follow down to the covering mount currently visible to userspace. At each
1087 * point, the filesystem owning that dentry may be queried as to whether the
1088 * caller is permitted to proceed or not.
1089 *
1090 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1091 * being true).
1092 */
1093int follow_down(struct path *path, bool mounting_here)
1094{
1095 unsigned managed;
1096 int ret;
1097
1098 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1099 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1100 /* Allow the filesystem to manage the transit without i_mutex
1101 * being held.
1102 *
1103 * We indicate to the filesystem if someone is trying to mount
1104 * something here. This gives autofs the chance to deny anyone
1105 * other than its daemon the right to mount on its
1106 * superstructure.
1107 *
1108 * The filesystem may sleep at this point.
1109 */
1110 if (managed & DCACHE_MANAGE_TRANSIT) {
1111 BUG_ON(!path->dentry->d_op);
1112 BUG_ON(!path->dentry->d_op->d_manage);
ab90911f
DH		 1113 ret = path->dentry->d_op->d_manage(
1114 path->dentry, mounting_here, false);
cc53ce53
DH		 1115 if (ret < 0)
1116 return ret == -EISDIR ? 0 : ret;
1117 }
1118
1119 /* Transit to a mounted filesystem. */
1120 if (managed & DCACHE_MOUNTED) {
1121 struct vfsmount *mounted = lookup_mnt(path);
1122 if (!mounted)
1123 break;
1124 dput(path->dentry);
1125 mntput(path->mnt);
1126 path->mnt = mounted;
1127 path->dentry = dget(mounted->mnt_root);
1128 continue;
1129 }
1130
1131 /* Don't handle automount points here */
1132 break;
1133 }
1134 return 0;
1135}
1136
9875cf80
DH		 1137/*
1138 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1139 */
1140static void follow_mount(struct path *path)
1141{
1142 while (d_mountpoint(path->dentry)) {
1143 struct vfsmount *mounted = lookup_mnt(path);
1144 if (!mounted)
1145 break;
1146 dput(path->dentry);
1147 mntput(path->mnt);
1148 path->mnt = mounted;
1149 path->dentry = dget(mounted->mnt_root);
1150 }
1151}
1152
31e6b01f 1153static void follow_dotdot(struct nameidata *nd)
1da177e4 1154{
2a737871 1155 set_root(nd);
e518ddb7 1156
1da177e4 1157 while(1) {
4ac91378 1158 struct dentry *old = nd->path.dentry;
1da177e4 1159
2a737871
AV		 1160 if (nd->path.dentry == nd->root.dentry &&
1161 nd->path.mnt == nd->root.mnt) {
1da177e4
LT		 1162 break;
1163 }
4ac91378 1164 if (nd->path.dentry != nd->path.mnt->mnt_root) {
3088dd70
AV		 1165 /* rare case of legitimate dget_parent()... */
1166 nd->path.dentry = dget_parent(nd->path.dentry);
1da177e4
LT		 1167 dput(old);
1168 break;
1169 }
3088dd70 1170 if (!follow_up(&nd->path))
1da177e4 1171 break;
1da177e4 1172 }
79ed0226 1173 follow_mount(&nd->path);
31e6b01f 1174 nd->inode = nd->path.dentry->d_inode;
1da177e4
LT		 1175}
1176
baa03890
NP		 1177/*
1178 * Allocate a dentry with name and parent, and perform a parent
1179 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1180 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1181 * have verified that no child exists while under i_mutex.
1182 */
1183static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1184 struct qstr *name, struct nameidata *nd)
1185{
1186 struct inode *inode = parent->d_inode;
1187 struct dentry *dentry;
1188 struct dentry *old;
1189
1190 /* Don't create child dentry for a dead directory. */
1191 if (unlikely(IS_DEADDIR(inode)))
1192 return ERR_PTR(-ENOENT);
1193
1194 dentry = d_alloc(parent, name);
1195 if (unlikely(!dentry))
1196 return ERR_PTR(-ENOMEM);
1197
1198 old = inode->i_op->lookup(inode, dentry, nd);
1199 if (unlikely(old)) {
1200 dput(dentry);
1201 dentry = old;
1202 }
1203 return dentry;
1204}
1205
1da177e4
LT		 1206/*
1207 * It's more convoluted than I'd like it to be, but... it's still fairly
1208 * small and for now I'd prefer to have fast path as straight as possible.
1209 * It _is_ time-critical.
1210 */
1211static int do_lookup(struct nameidata *nd, struct qstr *name,
31e6b01f 1212 struct path *path, struct inode **inode)
1da177e4 1213{
4ac91378 1214 struct vfsmount *mnt = nd->path.mnt;
31e6b01f 1215 struct dentry *dentry, *parent = nd->path.dentry;
6e6b1bd1 1216 struct inode *dir;
9875cf80
DH		 1217 int err;
1218
3cac260a
AV		 1219 /*
1220 * See if the low-level filesystem might want
1221 * to use its own hash..
1222 */
fb045adb 1223 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
9875cf80 1224 err = parent->d_op->d_hash(parent, nd->inode, name);
3cac260a
AV		 1225 if (err < 0)
1226 return err;
1227 }
1da177e4 1228
b04f784e
NP		 1229 /*
1230 * Rename seqlock is not required here because in the off chance
1231 * of a false negative due to a concurrent rename, we're going to
1232 * do the non-racy lookup, below.
1233 */
31e6b01f
NP		 1234 if (nd->flags & LOOKUP_RCU) {
1235 unsigned seq;
1236
1237 *inode = nd->inode;
1238 dentry = __d_lookup_rcu(parent, name, &seq, inode);
1239 if (!dentry) {
1240 if (nameidata_drop_rcu(nd))
1241 return -ECHILD;
1242 goto need_lookup;
1243 }
1244 /* Memory barrier in read_seqcount_begin of child is enough */
1245 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1246 return -ECHILD;
1247
1248 nd->seq = seq;
24643087 1249 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
f5e1c1c1 1250 dentry = do_revalidate_rcu(dentry, nd);
24643087
AV		 1251 if (!dentry)
1252 goto need_lookup;
1253 if (IS_ERR(dentry))
1254 goto fail;
1255 if (!(nd->flags & LOOKUP_RCU))
1256 goto done;
1257 }
31e6b01f
NP		 1258 path->mnt = mnt;
1259 path->dentry = dentry;
9875cf80
DH		 1260 if (likely(__follow_mount_rcu(nd, path, inode, false)))
1261 return 0;
1262 if (nameidata_drop_rcu(nd))
1263 return -ECHILD;
1264 /* fallthru */
1265 }
1266 dentry = __d_lookup(parent, name);
1267 if (!dentry)
1268 goto need_lookup;
2e2e88ea 1269found:
24643087
AV		 1270 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1271 dentry = do_revalidate(dentry, nd);
1272 if (!dentry)
1273 goto need_lookup;
1274 if (IS_ERR(dentry))
1275 goto fail;
1276 }
1da177e4 1277done:
9875cf80
DH		 1278 path->mnt = mnt;
1279 path->dentry = dentry;
1280 err = follow_managed(path, nd->flags);
89312214
IK		 1281 if (unlikely(err < 0)) {
1282 path_put_conditional(path, nd);
9875cf80 1283 return err;
89312214 1284 }
9875cf80 1285 *inode = path->dentry->d_inode;
1da177e4
LT		 1286 return 0;
1287
1288need_lookup:
6e6b1bd1 1289 dir = parent->d_inode;
31e6b01f 1290 BUG_ON(nd->inode != dir);
6e6b1bd1
AV		 1291
1292 mutex_lock(&dir->i_mutex);
1293 /*
1294 * First re-do the cached lookup just in case it was created
b04f784e
NP		 1295 * while we waited for the directory semaphore, or the first
1296 * lookup failed due to an unrelated rename.
6e6b1bd1 1297 *
b04f784e
NP		 1298 * This could use version numbering or similar to avoid unnecessary
1299 * cache lookups, but then we'd have to do the first lookup in the
1300 * non-racy way. However in the common case here, everything should
1301 * be hot in cache, so would it be a big win?
6e6b1bd1
AV		 1302 */
1303 dentry = d_lookup(parent, name);
baa03890
NP		 1304 if (likely(!dentry)) {
1305 dentry = d_alloc_and_lookup(parent, name, nd);
6e6b1bd1
AV		 1306 mutex_unlock(&dir->i_mutex);
1307 if (IS_ERR(dentry))
1308 goto fail;
1309 goto done;
1310 }
6e6b1bd1
AV		 1311 /*
1312 * Uhhuh! Nasty case: the cache was re-populated while
1313 * we waited on the semaphore. Need to revalidate.
1314 */
1315 mutex_unlock(&dir->i_mutex);
2e2e88ea 1316 goto found;
1da177e4 1317
1da177e4
LT		 1318fail:
1319 return PTR_ERR(dentry);
1320}
1321
52094c8a
AV		 1322static inline int may_lookup(struct nameidata *nd)
1323{
1324 if (nd->flags & LOOKUP_RCU) {
1325 int err = exec_permission(nd->inode, IPERM_FLAG_RCU);
1326 if (err != -ECHILD)
1327 return err;
1328 if (nameidata_drop_rcu(nd))
1329 return -ECHILD;
1330 }
1331 return exec_permission(nd->inode, 0);
1332}
1333
9856fa1b
AV		 1334static inline int handle_dots(struct nameidata *nd, int type)
1335{
1336 if (type == LAST_DOTDOT) {
1337 if (nd->flags & LOOKUP_RCU) {
1338 if (follow_dotdot_rcu(nd))
1339 return -ECHILD;
1340 } else
1341 follow_dotdot(nd);
1342 }
1343 return 0;
1344}
1345
951361f9
AV		 1346static void terminate_walk(struct nameidata *nd)
1347{
1348 if (!(nd->flags & LOOKUP_RCU)) {
1349 path_put(&nd->path);
1350 } else {
1351 nd->flags &= ~LOOKUP_RCU;
1352 nd->root.mnt = NULL;
1353 rcu_read_unlock();
1354 br_read_unlock(vfsmount_lock);
1355 }
1356}
1357
1da177e4
LT		 1358/*
1359 * Name resolution.
ea3834d9
PM		 1360 * This is the basic name resolution function, turning a pathname into
1361 * the final dentry. We expect 'base' to be positive and a directory.
1da177e4 1362 *
ea3834d9
PM		 1363 * Returns 0 and nd will have valid dentry and mnt on success.
1364 * Returns error and drops reference to input namei data on failure.
1da177e4 1365 */
6de88d72 1366static int link_path_walk(const char *name, struct nameidata *nd)
1da177e4
LT		 1367{
1368 struct path next;
1da177e4
LT		 1369 int err;
1370 unsigned int lookup_flags = nd->flags;
1371
1372 while (*name=='/')
1373 name++;
1374 if (!*name)
086e183a 1375 return 0;
1da177e4 1376
1da177e4 1377 if (nd->depth)
f55eab82 1378 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
1da177e4
LT		 1379
1380 /* At this point we know we have a real path component. */
1381 for(;;) {
31e6b01f 1382 struct inode *inode;
1da177e4
LT		 1383 unsigned long hash;
1384 struct qstr this;
1385 unsigned int c;
fe479a58 1386 int type;
1da177e4 1387
cdce5d6b 1388 nd->flags |= LOOKUP_CONTINUE;
52094c8a
AV		 1389
1390 err = may_lookup(nd);
1da177e4
LT		 1391 if (err)
1392 break;
1393
1394 this.name = name;
1395 c = *(const unsigned char *)name;
1396
1397 hash = init_name_hash();
1398 do {
1399 name++;
1400 hash = partial_name_hash(c, hash);
1401 c = *(const unsigned char *)name;
1402 } while (c && (c != '/'));
1403 this.len = name - (const char *) this.name;
1404 this.hash = end_name_hash(hash);
1405
fe479a58
AV		 1406 type = LAST_NORM;
1407 if (this.name[0] == '.') switch (this.len) {
1408 case 2:
16c2cd71 1409 if (this.name[1] == '.') {
fe479a58 1410 type = LAST_DOTDOT;
16c2cd71
AV		 1411 nd->flags |= LOOKUP_JUMPED;
1412 }
fe479a58
AV		 1413 break;
1414 case 1:
1415 type = LAST_DOT;
1416 }
16c2cd71
AV		 1417 if (likely(type == LAST_NORM))
1418 nd->flags &= ~LOOKUP_JUMPED;
fe479a58 1419
1da177e4
LT		 1420 /* remove trailing slashes? */
1421 if (!c)
1422 goto last_component;
1423 while (*++name == '/');
1424 if (!*name)
1425 goto last_with_slashes;
1426
1427 /*
1428 * "." and ".." are special - ".." especially so because it has
1429 * to be able to know about the current root directory and
1430 * parent relationships.
1431 */
fe479a58 1432 if (unlikely(type != LAST_NORM)) {
ef7562d5
AV		 1433 if (handle_dots(nd, type))
1434 return -ECHILD;
fe479a58 1435 continue;
1da177e4 1436 }
fe479a58 1437
1da177e4 1438 /* This does the actual lookups.. */
31e6b01f 1439 err = do_lookup(nd, &this, &next, &inode);
1da177e4
LT		 1440 if (err)
1441 break;
1da177e4 1442
7bc055d1 1443 if (inode && inode->i_op->follow_link) {
3abb17e8 1444 err = do_follow_link(inode, &next, nd);
1da177e4 1445 if (err)
a7472bab 1446 return err;
31e6b01f 1447 nd->inode = nd->path.dentry->d_inode;
31e6b01f 1448 } else {
09dd17d3 1449 path_to_nameidata(&next, nd);
31e6b01f
NP		 1450 nd->inode = inode;
1451 }
7bc055d1
AV		 1452 err = -ENOENT;
1453 if (!nd->inode)
1454 break;
1da177e4 1455 err = -ENOTDIR;
31e6b01f 1456 if (!nd->inode->i_op->lookup)
1da177e4
LT		 1457 break;
1458 continue;
1459 /* here ends the main loop */
1460
1461last_with_slashes:
1462 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1463last_component:
f55eab82
TM		 1464 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1465 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
1da177e4
LT		 1466 if (lookup_flags & LOOKUP_PARENT)
1467 goto lookup_parent;
ef7562d5
AV		 1468 if (unlikely(type != LAST_NORM))
1469 return handle_dots(nd, type);
31e6b01f 1470 err = do_lookup(nd, &this, &next, &inode);
1da177e4
LT		 1471 if (err)
1472 break;
db372915
DH		 1473 if (inode && unlikely(inode->i_op->follow_link) &&
1474 (lookup_flags & LOOKUP_FOLLOW)) {
3abb17e8 1475 err = do_follow_link(inode, &next, nd);
1da177e4 1476 if (err)
a7472bab 1477 return err;
31e6b01f
NP		 1478 nd->inode = nd->path.dentry->d_inode;
1479 } else {
09dd17d3 1480 path_to_nameidata(&next, nd);
31e6b01f
NP		 1481 nd->inode = inode;
1482 }
1da177e4 1483 err = -ENOENT;
31e6b01f 1484 if (!nd->inode)
1da177e4
LT		 1485 break;
1486 if (lookup_flags & LOOKUP_DIRECTORY) {
1487 err = -ENOTDIR;
31e6b01f 1488 if (!nd->inode->i_op->lookup)
1da177e4
LT		 1489 break;
1490 }
086e183a 1491 return 0;
1da177e4
LT		 1492lookup_parent:
1493 nd->last = this;
fe479a58 1494 nd->last_type = type;
1da177e4 1495 return 0;
1da177e4 1496 }
951361f9 1497 terminate_walk(nd);
1da177e4
LT		 1498 return err;
1499}
1500
70e9b357
AV		 1501static int path_init(int dfd, const char *name, unsigned int flags,
1502 struct nameidata *nd, struct file **fp)
31e6b01f
NP		 1503{
1504 int retval = 0;
1505 int fput_needed;
1506 struct file *file;
1507
1508 nd->last_type = LAST_ROOT; /* if there are only slashes... */
16c2cd71 1509 nd->flags = flags | LOOKUP_JUMPED;
31e6b01f
NP		 1510 nd->depth = 0;
1511 nd->root.mnt = NULL;
31e6b01f
NP		 1512
1513 if (*name=='/') {
e41f7d4e
AV		 1514 if (flags & LOOKUP_RCU) {
1515 br_read_lock(vfsmount_lock);
1516 rcu_read_lock();
1517 set_root_rcu(nd);
1518 } else {
1519 set_root(nd);
1520 path_get(&nd->root);
1521 }
1522 nd->path = nd->root;
31e6b01f 1523 } else if (dfd == AT_FDCWD) {
e41f7d4e
AV		 1524 if (flags & LOOKUP_RCU) {
1525 struct fs_struct *fs = current->fs;
1526 unsigned seq;
31e6b01f 1527
e41f7d4e
AV		 1528 br_read_lock(vfsmount_lock);
1529 rcu_read_lock();
c28cc364 1530
e41f7d4e
AV		 1531 do {
1532 seq = read_seqcount_begin(&fs->seq);
1533 nd->path = fs->pwd;
1534 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1535 } while (read_seqcount_retry(&fs->seq, seq));
1536 } else {
1537 get_fs_pwd(current->fs, &nd->path);
1538 }
31e6b01f
NP		 1539 } else {
1540 struct dentry *dentry;
1541
1542 file = fget_light(dfd, &fput_needed);
1543 retval = -EBADF;
1544 if (!file)
1545 goto out_fail;
1546
1547 dentry = file->f_path.dentry;
1548
1549 retval = -ENOTDIR;
1550 if (!S_ISDIR(dentry->d_inode->i_mode))
1551 goto fput_fail;
1552
1553 retval = file_permission(file, MAY_EXEC);
1554 if (retval)
1555 goto fput_fail;
1556
1557 nd->path = file->f_path;
e41f7d4e
AV		 1558 if (flags & LOOKUP_RCU) {
1559 if (fput_needed)
70e9b357 1560 *fp = file;
e41f7d4e
AV		 1561 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1562 br_read_lock(vfsmount_lock);
1563 rcu_read_lock();
1564 } else {
1565 path_get(&file->f_path);
1566 fput_light(file, fput_needed);
1567 }
31e6b01f 1568 }
31e6b01f 1569
31e6b01f 1570 nd->inode = nd->path.dentry->d_inode;
9b4a9b14 1571 return 0;
2dfdd266 1572
9b4a9b14
AV		 1573fput_fail:
1574 fput_light(file, fput_needed);
1575out_fail:
1576 return retval;
1577}
1578
1579/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
ee0827cd 1580static int path_lookupat(int dfd, const char *name,
9b4a9b14
AV		 1581 unsigned int flags, struct nameidata *nd)
1582{
70e9b357 1583 struct file *base = NULL;
31e6b01f
NP		 1584 int retval;
1585
1586 /*
1587 * Path walking is largely split up into 2 different synchronisation
1588 * schemes, rcu-walk and ref-walk (explained in
1589 * Documentation/filesystems/path-lookup.txt). These share much of the
1590 * path walk code, but some things particularly setup, cleanup, and
1591 * following mounts are sufficiently divergent that functions are
1592 * duplicated. Typically there is a function foo(), and its RCU
1593 * analogue, foo_rcu().
1594 *
1595 * -ECHILD is the error number of choice (just to avoid clashes) that
1596 * is returned if some aspect of an rcu-walk fails. Such an error must
1597 * be handled by restarting a traditional ref-walk (which will always
1598 * be able to complete).
1599 */
70e9b357 1600 retval = path_init(dfd, name, flags, nd, &base);
ee0827cd 1601
31e6b01f
NP		 1602 if (unlikely(retval))
1603 return retval;
ee0827cd
AV		 1604
1605 current->total_link_count = 0;
1606 retval = link_path_walk(name, nd);
1607
1608 if (nd->flags & LOOKUP_RCU) {
4455ca62
AV		 1609 /* went all way through without dropping RCU */
1610 BUG_ON(retval);
1611 if (nameidata_drop_rcu_last(nd))
1612 retval = -ECHILD;
ee0827cd
AV		 1613 }
1614
16c2cd71
AV		 1615 if (!retval)
1616 retval = handle_reval_path(nd);
1617
70e9b357
AV		 1618 if (base)
1619 fput(base);
ee0827cd 1620
2a737871
AV		 1621 if (nd->root.mnt) {
1622 path_put(&nd->root);
1623 nd->root.mnt = NULL;
1624 }
ee0827cd
AV		 1625 return retval;
1626}
31e6b01f 1627
ee0827cd
AV		 1628static int do_path_lookup(int dfd, const char *name,
1629 unsigned int flags, struct nameidata *nd)
1630{
1631 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1632 if (unlikely(retval == -ECHILD))
1633 retval = path_lookupat(dfd, name, flags, nd);
1634 if (unlikely(retval == -ESTALE))
1635 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
31e6b01f
NP		 1636
1637 if (likely(!retval)) {
1638 if (unlikely(!audit_dummy_context())) {
1639 if (nd->path.dentry && nd->inode)
1640 audit_inode(name, nd->path.dentry);
1641 }
1642 }
170aa3d0 1643 return retval;
1da177e4
LT		 1644}
1645
c9c6cac0 1646int kern_path_parent(const char *name, struct nameidata *nd)
5590ff0d 1647{
c9c6cac0 1648 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
5590ff0d
UD		 1649}
1650
d1811465
AV		 1651int kern_path(const char *name, unsigned int flags, struct path *path)
1652{
1653 struct nameidata nd;
1654 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1655 if (!res)
1656 *path = nd.path;
1657 return res;
1658}
1659
16f18200
JJS		 1660/**
1661 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1662 * @dentry: pointer to dentry of the base directory
1663 * @mnt: pointer to vfs mount of the base directory
1664 * @name: pointer to file name
1665 * @flags: lookup flags
1666 * @nd: pointer to nameidata
1667 */
1668int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1669 const char *name, unsigned int flags,
1670 struct nameidata *nd)
1671{
ee0827cd 1672 int result;
16f18200
JJS		 1673
1674 /* same as do_path_lookup */
1675 nd->last_type = LAST_ROOT;
16c2cd71 1676 nd->flags = flags | LOOKUP_JUMPED;
16f18200
JJS		 1677 nd->depth = 0;
1678
c8e7f449
JB		 1679 nd->path.dentry = dentry;
1680 nd->path.mnt = mnt;
1681 path_get(&nd->path);
5b857119
AV		 1682 nd->root = nd->path;
1683 path_get(&nd->root);
31e6b01f 1684 nd->inode = nd->path.dentry->d_inode;
16f18200 1685
ee0827cd
AV		 1686 current->total_link_count = 0;
1687
1688 result = link_path_walk(name, nd);
16c2cd71
AV		 1689 if (!result)
1690 result = handle_reval_path(nd);
ee0827cd
AV		 1691 if (result == -ESTALE) {
1692 /* nd->path had been dropped */
1693 current->total_link_count = 0;
1694 nd->path.dentry = dentry;
1695 nd->path.mnt = mnt;
1696 nd->inode = dentry->d_inode;
1697 path_get(&nd->path);
16c2cd71
AV		 1698 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_REVAL;
1699
ee0827cd 1700 result = link_path_walk(name, nd);
16c2cd71
AV		 1701 if (!result)
1702 result = handle_reval_path(nd);
ee0827cd
AV		 1703 }
1704 if (unlikely(!result && !audit_dummy_context() && nd->path.dentry &&
31e6b01f 1705 nd->inode))
4ac91378 1706 audit_inode(name, nd->path.dentry);
16f18200 1707
5b857119
AV		 1708 path_put(&nd->root);
1709 nd->root.mnt = NULL;
16f18200 1710
ee0827cd 1711 return result;
16f18200
JJS		 1712}
1713
eead1911
CH		 1714static struct dentry *__lookup_hash(struct qstr *name,
1715 struct dentry *base, struct nameidata *nd)
1da177e4 1716{
81fca444 1717 struct inode *inode = base->d_inode;
057f6c01 1718 struct dentry *dentry;
1da177e4
LT		 1719 int err;
1720
b74c79e9 1721 err = exec_permission(inode, 0);
81fca444
CH		 1722 if (err)
1723 return ERR_PTR(err);
1da177e4
LT		 1724
1725 /*
1726 * See if the low-level filesystem might want
1727 * to use its own hash..
1728 */
fb045adb 1729 if (base->d_flags & DCACHE_OP_HASH) {
b1e6a015 1730 err = base->d_op->d_hash(base, inode, name);
1da177e4
LT		 1731 dentry = ERR_PTR(err);
1732 if (err < 0)
1733 goto out;
1734 }
1735
b04f784e
NP		 1736 /*
1737 * Don't bother with __d_lookup: callers are for creat as
1738 * well as unlink, so a lot of the time it would cost
1739 * a double lookup.
6e6b1bd1 1740 */
b04f784e 1741 dentry = d_lookup(base, name);
6e6b1bd1 1742
fb045adb 1743 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE))
6e6b1bd1
AV		 1744 dentry = do_revalidate(dentry, nd);
1745
baa03890
NP		 1746 if (!dentry)
1747 dentry = d_alloc_and_lookup(base, name, nd);
1da177e4
LT		 1748out:
1749 return dentry;
1750}
1751
057f6c01
JM		 1752/*
1753 * Restricted form of lookup. Doesn't follow links, single-component only,
1754 * needs parent already locked. Doesn't follow mounts.
1755 * SMP-safe.
1756 */
eead1911 1757static struct dentry *lookup_hash(struct nameidata *nd)
057f6c01 1758{
4ac91378 1759 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1da177e4
LT		 1760}
1761
eead1911
CH		 1762static int __lookup_one_len(const char *name, struct qstr *this,
1763 struct dentry *base, int len)
1da177e4
LT		 1764{
1765 unsigned long hash;
1da177e4
LT		 1766 unsigned int c;
1767
057f6c01
JM		 1768 this->name = name;
1769 this->len = len;
1da177e4 1770 if (!len)
057f6c01 1771 return -EACCES;
1da177e4
LT		 1772
1773 hash = init_name_hash();
1774 while (len--) {
1775 c = *(const unsigned char *)name++;
1776 if (c == '/' || c == '\0')
057f6c01 1777 return -EACCES;
1da177e4
LT		 1778 hash = partial_name_hash(c, hash);
1779 }
057f6c01
JM		 1780 this->hash = end_name_hash(hash);
1781 return 0;
1782}
1da177e4 1783
eead1911 1784/**
a6b91919 1785 * lookup_one_len - filesystem helper to lookup single pathname component
eead1911
CH		 1786 * @name: pathname component to lookup
1787 * @base: base directory to lookup from
1788 * @len: maximum length @len should be interpreted to
1789 *
a6b91919
RD		 1790 * Note that this routine is purely a helper for filesystem usage and should
1791 * not be called by generic code. Also note that by using this function the
eead1911
CH		 1792 * nameidata argument is passed to the filesystem methods and a filesystem
1793 * using this helper needs to be prepared for that.
1794 */
057f6c01
JM		 1795struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1796{
1797 int err;
1798 struct qstr this;
1799
2f9092e1
DW		 1800 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1801
057f6c01 1802 err = __lookup_one_len(name, &this, base, len);
eead1911
CH		 1803 if (err)
1804 return ERR_PTR(err);
1805
49705b77 1806 return __lookup_hash(&this, base, NULL);
057f6c01
JM		 1807}
1808
2d8f3038
AV		 1809int user_path_at(int dfd, const char __user *name, unsigned flags,
1810 struct path *path)
1da177e4 1811{
2d8f3038 1812 struct nameidata nd;
1da177e4
LT		 1813 char *tmp = getname(name);
1814 int err = PTR_ERR(tmp);
1da177e4 1815 if (!IS_ERR(tmp)) {
2d8f3038
AV		 1816
1817 BUG_ON(flags & LOOKUP_PARENT);
1818
1819 err = do_path_lookup(dfd, tmp, flags, &nd);
1da177e4 1820 putname(tmp);
2d8f3038
AV		 1821 if (!err)
1822 *path = nd.path;
1da177e4
LT		 1823 }
1824 return err;
1825}
1826
2ad94ae6
AV		 1827static int user_path_parent(int dfd, const char __user *path,
1828 struct nameidata *nd, char **name)
1829{
1830 char *s = getname(path);
1831 int error;
1832
1833 if (IS_ERR(s))
1834 return PTR_ERR(s);
1835
1836 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1837 if (error)
1838 putname(s);
1839 else
1840 *name = s;
1841
1842 return error;
1843}
1844
1da177e4
LT		 1845/*
1846 * It's inline, so penalty for filesystems that don't use sticky bit is
1847 * minimal.
1848 */
1849static inline int check_sticky(struct inode *dir, struct inode *inode)
1850{
da9592ed
DH		 1851 uid_t fsuid = current_fsuid();
1852
1da177e4
LT		 1853 if (!(dir->i_mode & S_ISVTX))
1854 return 0;
da9592ed 1855 if (inode->i_uid == fsuid)
1da177e4 1856 return 0;
da9592ed 1857 if (dir->i_uid == fsuid)
1da177e4
LT		 1858 return 0;
1859 return !capable(CAP_FOWNER);
1860}
1861
1862/*
1863 * Check whether we can remove a link victim from directory dir, check
1864 * whether the type of victim is right.
1865 * 1. We can't do it if dir is read-only (done in permission())
1866 * 2. We should have write and exec permissions on dir
1867 * 3. We can't remove anything from append-only dir
1868 * 4. We can't do anything with immutable dir (done in permission())
1869 * 5. If the sticky bit on dir is set we should either
1870 * a. be owner of dir, or
1871 * b. be owner of victim, or
1872 * c. have CAP_FOWNER capability
1873 * 6. If the victim is append-only or immutable we can't do antyhing with
1874 * links pointing to it.
1875 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1876 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1877 * 9. We can't remove a root or mountpoint.
1878 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1879 * nfs_async_unlink().
1880 */
858119e1 1881static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1da177e4
LT		 1882{
1883 int error;
1884
1885 if (!victim->d_inode)
1886 return -ENOENT;
1887
1888 BUG_ON(victim->d_parent->d_inode != dir);
cccc6bba 1889 audit_inode_child(victim, dir);
1da177e4 1890
f419a2e3 1891 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1892 if (error)
1893 return error;
1894 if (IS_APPEND(dir))
1895 return -EPERM;
1896 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
f9454548 1897 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1da177e4
LT		 1898 return -EPERM;
1899 if (isdir) {
1900 if (!S_ISDIR(victim->d_inode->i_mode))
1901 return -ENOTDIR;
1902 if (IS_ROOT(victim))
1903 return -EBUSY;
1904 } else if (S_ISDIR(victim->d_inode->i_mode))
1905 return -EISDIR;
1906 if (IS_DEADDIR(dir))
1907 return -ENOENT;
1908 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1909 return -EBUSY;
1910 return 0;
1911}
1912
1913/* Check whether we can create an object with dentry child in directory
1914 * dir.
1915 * 1. We can't do it if child already exists (open has special treatment for
1916 * this case, but since we are inlined it's OK)
1917 * 2. We can't do it if dir is read-only (done in permission())
1918 * 3. We should have write and exec permissions on dir
1919 * 4. We can't do it if dir is immutable (done in permission())
1920 */
a95164d9 1921static inline int may_create(struct inode *dir, struct dentry *child)
1da177e4
LT		 1922{
1923 if (child->d_inode)
1924 return -EEXIST;
1925 if (IS_DEADDIR(dir))
1926 return -ENOENT;
f419a2e3 1927 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1928}
1929
1da177e4
LT		 1930/*
1931 * p1 and p2 should be directories on the same fs.
1932 */
1933struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1934{
1935 struct dentry *p;
1936
1937 if (p1 == p2) {
f2eace23 1938 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1da177e4
LT		 1939 return NULL;
1940 }
1941
a11f3a05 1942 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4 1943
e2761a11
OH		 1944 p = d_ancestor(p2, p1);
1945 if (p) {
1946 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1947 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1948 return p;
1da177e4
LT		 1949 }
1950
e2761a11
OH		 1951 p = d_ancestor(p1, p2);
1952 if (p) {
1953 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1954 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1955 return p;
1da177e4
LT		 1956 }
1957
f2eace23
IM		 1958 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1959 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1da177e4
LT		 1960 return NULL;
1961}
1962
1963void unlock_rename(struct dentry *p1, struct dentry *p2)
1964{
1b1dcc1b 1965 mutex_unlock(&p1->d_inode->i_mutex);
1da177e4 1966 if (p1 != p2) {
1b1dcc1b 1967 mutex_unlock(&p2->d_inode->i_mutex);
a11f3a05 1968 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4
LT		 1969 }
1970}
1971
1972int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1973 struct nameidata *nd)
1974{
a95164d9 1975 int error = may_create(dir, dentry);
1da177e4
LT		 1976
1977 if (error)
1978 return error;
1979
acfa4380 1980 if (!dir->i_op->create)
1da177e4
LT		 1981 return -EACCES; /* shouldn't it be ENOSYS? */
1982 mode &= S_IALLUGO;
1983 mode |= S_IFREG;
1984 error = security_inode_create(dir, dentry, mode);
1985 if (error)
1986 return error;
1da177e4 1987 error = dir->i_op->create(dir, dentry, mode, nd);
a74574aa 1988 if (!error)
f38aa942 1989 fsnotify_create(dir, dentry);
1da177e4
LT		 1990 return error;
1991}
1992
3fb64190 1993int may_open(struct path *path, int acc_mode, int flag)
1da177e4 1994{
3fb64190 1995 struct dentry *dentry = path->dentry;
1da177e4
LT		 1996 struct inode *inode = dentry->d_inode;
1997 int error;
1998
1999 if (!inode)
2000 return -ENOENT;
2001
c8fe8f30
CH		 2002 switch (inode->i_mode & S_IFMT) {
2003 case S_IFLNK:
1da177e4 2004 return -ELOOP;
c8fe8f30
CH		 2005 case S_IFDIR:
2006 if (acc_mode & MAY_WRITE)
2007 return -EISDIR;
2008 break;
2009 case S_IFBLK:
2010 case S_IFCHR:
3fb64190 2011 if (path->mnt->mnt_flags & MNT_NODEV)
1da177e4 2012 return -EACCES;
c8fe8f30
CH		 2013 /*FALLTHRU*/
2014 case S_IFIFO:
2015 case S_IFSOCK:
1da177e4 2016 flag &= ~O_TRUNC;
c8fe8f30 2017 break;
4a3fd211 2018 }
b41572e9 2019
3fb64190 2020 error = inode_permission(inode, acc_mode);
b41572e9
DH		 2021 if (error)
2022 return error;
6146f0d5 2023
1da177e4
LT		 2024 /*
2025 * An append-only file must be opened in append mode for writing.
2026 */
2027 if (IS_APPEND(inode)) {
8737c930 2028 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
7715b521 2029 return -EPERM;
1da177e4 2030 if (flag & O_TRUNC)
7715b521 2031 return -EPERM;
1da177e4
LT		 2032 }
2033
2034 /* O_NOATIME can only be set by the owner or superuser */
7715b521
AV		 2035 if (flag & O_NOATIME && !is_owner_or_cap(inode))
2036 return -EPERM;
1da177e4
LT		 2037
2038 /*
2039 * Ensure there are no outstanding leases on the file.
2040 */
b65a9cfc 2041 return break_lease(inode, flag);
7715b521 2042}
1da177e4 2043
e1181ee6 2044static int handle_truncate(struct file *filp)
7715b521 2045{
e1181ee6 2046 struct path *path = &filp->f_path;
7715b521
AV		 2047 struct inode *inode = path->dentry->d_inode;
2048 int error = get_write_access(inode);
2049 if (error)
2050 return error;
2051 /*
2052 * Refuse to truncate files with mandatory locks held on them.
2053 */
2054 error = locks_verify_locked(inode);
2055 if (!error)
ea0d3ab2 2056 error = security_path_truncate(path);
7715b521
AV		 2057 if (!error) {
2058 error = do_truncate(path->dentry, 0,
2059 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
e1181ee6 2060 filp);
7715b521
AV		 2061 }
2062 put_write_access(inode);
acd0c935 2063 return error;
1da177e4
LT		 2064}
2065
d57999e1
DH		 2066/*
2067 * Be careful about ever adding any more callers of this
2068 * function. Its flags must be in the namei format, not
2069 * what get passed to sys_open().
2070 */
2071static int __open_namei_create(struct nameidata *nd, struct path *path,
8737c930 2072 int open_flag, int mode)
aab520e2
DH		 2073{
2074 int error;
4ac91378 2075 struct dentry *dir = nd->path.dentry;
aab520e2
DH		 2076
2077 if (!IS_POSIXACL(dir->d_inode))
ce3b0f8d 2078 mode &= ~current_umask();
be6d3e56
KT		 2079 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
2080 if (error)
2081 goto out_unlock;
aab520e2 2082 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
be6d3e56 2083out_unlock:
aab520e2 2084 mutex_unlock(&dir->d_inode->i_mutex);
4ac91378
JB		 2085 dput(nd->path.dentry);
2086 nd->path.dentry = path->dentry;
31e6b01f 2087
aab520e2
DH		 2088 if (error)
2089 return error;
2090 /* Don't check for write permission, don't truncate */
8737c930 2091 return may_open(&nd->path, 0, open_flag & ~O_TRUNC);
aab520e2
DH		 2092}
2093
d57999e1
DH		 2094/*
2095 * Note that while the flag value (low two bits) for sys_open means:
2096 * 00 - read-only
2097 * 01 - write-only
2098 * 10 - read-write
2099 * 11 - special
2100 * it is changed into
2101 * 00 - no permissions needed
2102 * 01 - read-permission
2103 * 10 - write-permission
2104 * 11 - read-write
2105 * for the internal routines (ie open_namei()/follow_link() etc)
2106 * This is more logical, and also allows the 00 "no perm needed"
2107 * to be used for symlinks (where the permissions are checked
2108 * later).
2109 *
2110*/
2111static inline int open_to_namei_flags(int flag)
2112{
2113 if ((flag+1) & O_ACCMODE)
2114 flag++;
2115 return flag;
2116}
2117
7715b521 2118static int open_will_truncate(int flag, struct inode *inode)
4a3fd211
DH		 2119{
2120 /*
2121 * We'll never write to the fs underlying
2122 * a device file.
2123 */
2124 if (special_file(inode->i_mode))
2125 return 0;
2126 return (flag & O_TRUNC);
2127}
2128
648fa861 2129static struct file *finish_open(struct nameidata *nd,
9a66179e 2130 int open_flag, int acc_mode)
648fa861
AV		 2131{
2132 struct file *filp;
2133 int will_truncate;
2134 int error;
2135
9a66179e 2136 will_truncate = open_will_truncate(open_flag, nd->path.dentry->d_inode);
648fa861
AV		 2137 if (will_truncate) {
2138 error = mnt_want_write(nd->path.mnt);
2139 if (error)
2140 goto exit;
2141 }
2142 error = may_open(&nd->path, acc_mode, open_flag);
2143 if (error) {
2144 if (will_truncate)
2145 mnt_drop_write(nd->path.mnt);
2146 goto exit;
2147 }
2148 filp = nameidata_to_filp(nd);
2149 if (!IS_ERR(filp)) {
2150 error = ima_file_check(filp, acc_mode);
2151 if (error) {
2152 fput(filp);
2153 filp = ERR_PTR(error);
2154 }
2155 }
2156 if (!IS_ERR(filp)) {
648fa861 2157 if (will_truncate) {
e1181ee6 2158 error = handle_truncate(filp);
648fa861
AV		 2159 if (error) {
2160 fput(filp);
2161 filp = ERR_PTR(error);
2162 }
2163 }
2164 }
2165 /*
2166 * It is now safe to drop the mnt write
2167 * because the filp has had a write taken
2168 * on its behalf.
2169 */
2170 if (will_truncate)
2171 mnt_drop_write(nd->path.mnt);
d893f1bc 2172 path_put(&nd->path);
648fa861
AV		 2173 return filp;
2174
2175exit:
648fa861
AV		 2176 path_put(&nd->path);
2177 return ERR_PTR(error);
2178}
2179
31e6b01f 2180/*
fe2d35ff 2181 * Handle the last step of open()
31e6b01f 2182 */
fb1cc555 2183static struct file *do_last(struct nameidata *nd, struct path *path,
c3e380b0 2184 const struct open_flags *op, const char *pathname)
fb1cc555 2185{
a1e28038 2186 struct dentry *dir = nd->path.dentry;
fb1cc555 2187 struct file *filp;
fe2d35ff 2188 struct inode *inode;
16c2cd71 2189 int error;
1f36f774 2190
c3e380b0
AV		 2191 nd->flags &= ~LOOKUP_PARENT;
2192 nd->flags |= op->intent;
2193
1f36f774
AV		 2194 switch (nd->last_type) {
2195 case LAST_DOTDOT:
176306f5 2196 case LAST_DOT:
fe2d35ff
AV		 2197 error = handle_dots(nd, nd->last_type);
2198 if (error)
2199 return ERR_PTR(error);
1f36f774 2200 /* fallthrough */
1f36f774 2201 case LAST_ROOT:
fe2d35ff
AV		 2202 if (nd->flags & LOOKUP_RCU) {
2203 if (nameidata_drop_rcu_last(nd))
2204 return ERR_PTR(-ECHILD);
2205 }
16c2cd71
AV		 2206 error = handle_reval_path(nd);
2207 if (error)
2208 goto exit;
fe2d35ff
AV		 2209 audit_inode(pathname, nd->path.dentry);
2210 if (op->open_flag & O_CREAT) {
2211 error = -EISDIR;
2212 goto exit;
2213 }
2214 goto ok;
1f36f774 2215 case LAST_BIND:
fe2d35ff 2216 /* can't be RCU mode here */
16c2cd71
AV		 2217 error = handle_reval_path(nd);
2218 if (error)
2219 goto exit;
1f36f774 2220 audit_inode(pathname, dir);
67ee3ad2 2221 goto ok;
1f36f774 2222 }
67ee3ad2 2223
fe2d35ff
AV		 2224 if (!(op->open_flag & O_CREAT)) {
2225 if (nd->last.name[nd->last.len])
2226 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2227 /* we _can_ be in RCU mode here */
2228 error = do_lookup(nd, &nd->last, path, &inode);
2229 if (error) {
2230 terminate_walk(nd);
2231 return ERR_PTR(error);
2232 }
2233 if (!inode) {
2234 path_to_nameidata(path, nd);
2235 terminate_walk(nd);
2236 return ERR_PTR(-ENOENT);
2237 }
2238 if (unlikely(inode->i_op->follow_link)) {
2239 /* We drop rcu-walk here */
2240 if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry))
2241 return ERR_PTR(-ECHILD);
2242 return NULL;
2243 }
2244 path_to_nameidata(path, nd);
2245 nd->inode = inode;
2246 /* sayonara */
2247 if (nd->flags & LOOKUP_RCU) {
2248 if (nameidata_drop_rcu_last(nd))
2249 return ERR_PTR(-ECHILD);
2250 }
2251
2252 error = -ENOTDIR;
2253 if (nd->flags & LOOKUP_DIRECTORY) {
2254 if (!inode->i_op->lookup)
2255 goto exit;
2256 }
2257 audit_inode(pathname, nd->path.dentry);
2258 goto ok;
2259 }
2260
2261 /* create side of things */
2262
2263 if (nd->flags & LOOKUP_RCU) {
2264 if (nameidata_drop_rcu_last(nd))
2265 return ERR_PTR(-ECHILD);
2266 }
2267
2268 audit_inode(pathname, dir);
16c2cd71 2269 error = -EISDIR;
1f36f774 2270 /* trailing slashes? */
31e6b01f
NP		 2271 if (nd->last.name[nd->last.len])
2272 goto exit;
a2c36b45 2273
a1e28038
AV		 2274 mutex_lock(&dir->d_inode->i_mutex);
2275
2276 path->dentry = lookup_hash(nd);
2277 path->mnt = nd->path.mnt;
2278
fb1cc555
AV		 2279 error = PTR_ERR(path->dentry);
2280 if (IS_ERR(path->dentry)) {
2281 mutex_unlock(&dir->d_inode->i_mutex);
2282 goto exit;
2283 }
2284
2285 if (IS_ERR(nd->intent.open.file)) {
2286 error = PTR_ERR(nd->intent.open.file);
2287 goto exit_mutex_unlock;
2288 }
2289
2290 /* Negative dentry, just create the file */
2291 if (!path->dentry->d_inode) {
2292 /*
2293 * This write is needed to ensure that a
2294 * ro->rw transition does not occur between
2295 * the time when the file is created and when
2296 * a permanent write count is taken through
2297 * the 'struct file' in nameidata_to_filp().
2298 */
2299 error = mnt_want_write(nd->path.mnt);
2300 if (error)
2301 goto exit_mutex_unlock;
c3e380b0 2302 error = __open_namei_create(nd, path, op->open_flag, op->mode);
fb1cc555
AV		 2303 if (error) {
2304 mnt_drop_write(nd->path.mnt);
2305 goto exit;
2306 }
2307 filp = nameidata_to_filp(nd);
2308 mnt_drop_write(nd->path.mnt);
d893f1bc 2309 path_put(&nd->path);
fb1cc555 2310 if (!IS_ERR(filp)) {
c3e380b0 2311 error = ima_file_check(filp, op->acc_mode);
fb1cc555
AV		 2312 if (error) {
2313 fput(filp);
2314 filp = ERR_PTR(error);
2315 }
2316 }
2317 return filp;
2318 }
2319
2320 /*
2321 * It already exists.
2322 */
2323 mutex_unlock(&dir->d_inode->i_mutex);
2324 audit_inode(pathname, path->dentry);
2325
2326 error = -EEXIST;
c3e380b0 2327 if (op->open_flag & O_EXCL)
fb1cc555
AV		 2328 goto exit_dput;
2329
9875cf80
DH		 2330 error = follow_managed(path, nd->flags);
2331 if (error < 0)
2332 goto exit_dput;
fb1cc555
AV		 2333
2334 error = -ENOENT;
2335 if (!path->dentry->d_inode)
2336 goto exit_dput;
9e67f361
AV		 2337
2338 if (path->dentry->d_inode->i_op->follow_link)
fb1cc555 2339 return NULL;
fb1cc555
AV		 2340
2341 path_to_nameidata(path, nd);
31e6b01f 2342 nd->inode = path->dentry->d_inode;
fb1cc555 2343 error = -EISDIR;
31e6b01f 2344 if (S_ISDIR(nd->inode->i_mode))
fb1cc555 2345 goto exit;
67ee3ad2 2346ok:
c3e380b0 2347 filp = finish_open(nd, op->open_flag, op->acc_mode);
fb1cc555
AV		 2348 return filp;
2349
2350exit_mutex_unlock:
2351 mutex_unlock(&dir->d_inode->i_mutex);
2352exit_dput:
2353 path_put_conditional(path, nd);
2354exit:
fb1cc555
AV		 2355 path_put(&nd->path);
2356 return ERR_PTR(error);
2357}
2358
13aab428 2359static struct file *path_openat(int dfd, const char *pathname,
47c805dc 2360 const struct open_flags *op, int flags)
1da177e4 2361{
fe2d35ff 2362 struct file *base = NULL;
4a3fd211 2363 struct file *filp;
a70e65df 2364 struct nameidata nd;
9850c056 2365 struct path path;
1da177e4 2366 int count = 0;
13aab428 2367 int error;
31e6b01f
NP		 2368
2369 filp = get_empty_filp();
2370 if (!filp)
2371 return ERR_PTR(-ENFILE);
2372
47c805dc 2373 filp->f_flags = op->open_flag;
31e6b01f 2374 nd.intent.open.file = filp;
47c805dc
AV		 2375 nd.intent.open.flags = open_to_namei_flags(op->open_flag);
2376 nd.intent.open.create_mode = op->mode;
31e6b01f 2377
fe2d35ff 2378 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, &nd, &base);
31e6b01f 2379 if (unlikely(error))
13aab428 2380 goto out_filp;
31e6b01f 2381
fe2d35ff
AV		 2382 current->total_link_count = 0;
2383 error = link_path_walk(pathname, &nd);
31e6b01f
NP		 2384 if (unlikely(error))
2385 goto out_filp;
1da177e4 2386
47c805dc 2387 filp = do_last(&nd, &path, op, pathname);
806b681c 2388 while (unlikely(!filp)) { /* trailing symlink */
7b9337aa
NP		 2389 struct path link = path;
2390 struct inode *linki = link.dentry->d_inode;
def4af30 2391 void *cookie;
806b681c 2392 error = -ELOOP;
db372915 2393 if (!(nd.flags & LOOKUP_FOLLOW))
1f36f774
AV		 2394 goto exit_dput;
2395 if (count++ == 32)
806b681c
AV		 2396 goto exit_dput;
2397 /*
2398 * This is subtle. Instead of calling do_follow_link() we do
2399 * the thing by hands. The reason is that this way we have zero
2400 * link_count and path_walk() (called from ->follow_link)
2401 * honoring LOOKUP_PARENT. After that we have the parent and
2402 * last component, i.e. we are in the same situation as after
2403 * the first path_walk(). Well, almost - if the last component
2404 * is normal we get its copy stored in nd->last.name and we will
2405 * have to putname() it when we are done. Procfs-like symlinks
2406 * just set LAST_BIND.
2407 */
2408 nd.flags |= LOOKUP_PARENT;
c3e380b0 2409 nd.flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
7b9337aa 2410 error = __do_follow_link(&link, &nd, &cookie);
c3e380b0 2411 if (unlikely(error))
f1afe9ef 2412 filp = ERR_PTR(error);
c3e380b0 2413 else
47c805dc 2414 filp = do_last(&nd, &path, op, pathname);
f1afe9ef 2415 if (!IS_ERR(cookie) && linki->i_op->put_link)
7b9337aa
NP		 2416 linki->i_op->put_link(link.dentry, &nd, cookie);
2417 path_put(&link);
806b681c 2418 }
10fa8e62 2419out:
2a737871
AV		 2420 if (nd.root.mnt)
2421 path_put(&nd.root);
fe2d35ff
AV		 2422 if (base)
2423 fput(base);
2dab5974 2424 release_open_intent(&nd);
10fa8e62 2425 return filp;
1da177e4 2426
806b681c
AV		 2427exit_dput:
2428 path_put_conditional(&path, &nd);
31e6b01f
NP		 2429 path_put(&nd.path);
2430out_filp:
806b681c 2431 filp = ERR_PTR(error);
10fa8e62 2432 goto out;
1da177e4
LT		 2433}
2434
13aab428
AV		 2435struct file *do_filp_open(int dfd, const char *pathname,
2436 const struct open_flags *op, int flags)
2437{
2438 struct file *filp;
2439
2440 filp = path_openat(dfd, pathname, op, flags | LOOKUP_RCU);
2441 if (unlikely(filp == ERR_PTR(-ECHILD)))
2442 filp = path_openat(dfd, pathname, op, flags);
2443 if (unlikely(filp == ERR_PTR(-ESTALE)))
2444 filp = path_openat(dfd, pathname, op, flags | LOOKUP_REVAL);
2445 return filp;
2446}
2447
1da177e4
LT		 2448/**
2449 * lookup_create - lookup a dentry, creating it if it doesn't exist
2450 * @nd: nameidata info
2451 * @is_dir: directory flag
2452 *
2453 * Simple function to lookup and return a dentry and create it
2454 * if it doesn't exist. Is SMP-safe.
c663e5d8 2455 *
4ac91378 2456 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1da177e4
LT		 2457 */
2458struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2459{
c663e5d8 2460 struct dentry *dentry = ERR_PTR(-EEXIST);
1da177e4 2461
4ac91378 2462 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
c663e5d8
CH		 2463 /*
2464 * Yucky last component or no last component at all?
2465 * (foo/., foo/.., /////)
2466 */
1da177e4
LT		 2467 if (nd->last_type != LAST_NORM)
2468 goto fail;
2469 nd->flags &= ~LOOKUP_PARENT;
3516586a 2470 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
a634904a 2471 nd->intent.open.flags = O_EXCL;
c663e5d8
CH		 2472
2473 /*
2474 * Do the final lookup.
2475 */
49705b77 2476 dentry = lookup_hash(nd);
1da177e4
LT		 2477 if (IS_ERR(dentry))
2478 goto fail;
c663e5d8 2479
e9baf6e5
AV		 2480 if (dentry->d_inode)
2481 goto eexist;
c663e5d8
CH		 2482 /*
2483 * Special case - lookup gave negative, but... we had foo/bar/
2484 * From the vfs_mknod() POV we just have a negative dentry -
2485 * all is fine. Let's be bastards - you had / on the end, you've
2486 * been asking for (non-existent) directory. -ENOENT for you.
2487 */
e9baf6e5
AV		 2488 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2489 dput(dentry);
2490 dentry = ERR_PTR(-ENOENT);
2491 }
1da177e4 2492 return dentry;
e9baf6e5 2493eexist:
1da177e4 2494 dput(dentry);
e9baf6e5 2495 dentry = ERR_PTR(-EEXIST);
1da177e4
LT		 2496fail:
2497 return dentry;
2498}
f81a0bff 2499EXPORT_SYMBOL_GPL(lookup_create);
1da177e4
LT		 2500
2501int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2502{
a95164d9 2503 int error = may_create(dir, dentry);
1da177e4
LT		 2504
2505 if (error)
2506 return error;
2507
2508 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2509 return -EPERM;
2510
acfa4380 2511 if (!dir->i_op->mknod)
1da177e4
LT		 2512 return -EPERM;
2513
08ce5f16
SH		 2514 error = devcgroup_inode_mknod(mode, dev);
2515 if (error)
2516 return error;
2517
1da177e4
LT		 2518 error = security_inode_mknod(dir, dentry, mode, dev);
2519 if (error)
2520 return error;
2521
1da177e4 2522 error = dir->i_op->mknod(dir, dentry, mode, dev);
a74574aa 2523 if (!error)
f38aa942 2524 fsnotify_create(dir, dentry);
1da177e4
LT		 2525 return error;
2526}
2527
463c3197
DH		 2528static int may_mknod(mode_t mode)
2529{
2530 switch (mode & S_IFMT) {
2531 case S_IFREG:
2532 case S_IFCHR:
2533 case S_IFBLK:
2534 case S_IFIFO:
2535 case S_IFSOCK:
2536 case 0: /* zero mode translates to S_IFREG */
2537 return 0;
2538 case S_IFDIR:
2539 return -EPERM;
2540 default:
2541 return -EINVAL;
2542 }
2543}
2544
2e4d0924
HC		 2545SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2546 unsigned, dev)
1da177e4 2547{
2ad94ae6
AV		 2548 int error;
2549 char *tmp;
2550 struct dentry *dentry;
1da177e4
LT		 2551 struct nameidata nd;
2552
2553 if (S_ISDIR(mode))
2554 return -EPERM;
1da177e4 2555
2ad94ae6 2556 error = user_path_parent(dfd, filename, &nd, &tmp);
1da177e4 2557 if (error)
2ad94ae6
AV		 2558 return error;
2559
1da177e4 2560 dentry = lookup_create(&nd, 0);
463c3197
DH		 2561 if (IS_ERR(dentry)) {
2562 error = PTR_ERR(dentry);
2563 goto out_unlock;
2564 }
4ac91378 2565 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2566 mode &= ~current_umask();
463c3197
DH		 2567 error = may_mknod(mode);
2568 if (error)
2569 goto out_dput;
2570 error = mnt_want_write(nd.path.mnt);
2571 if (error)
2572 goto out_dput;
be6d3e56
KT		 2573 error = security_path_mknod(&nd.path, dentry, mode, dev);
2574 if (error)
2575 goto out_drop_write;
463c3197 2576 switch (mode & S_IFMT) {
1da177e4 2577 case 0: case S_IFREG:
4ac91378 2578 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
1da177e4
LT		 2579 break;
2580 case S_IFCHR: case S_IFBLK:
4ac91378 2581 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
1da177e4
LT		 2582 new_decode_dev(dev));
2583 break;
2584 case S_IFIFO: case S_IFSOCK:
4ac91378 2585 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
1da177e4 2586 break;
1da177e4 2587 }
be6d3e56 2588out_drop_write:
463c3197
DH		 2589 mnt_drop_write(nd.path.mnt);
2590out_dput:
2591 dput(dentry);
2592out_unlock:
4ac91378 2593 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2594 path_put(&nd.path);
1da177e4
LT		 2595 putname(tmp);
2596
2597 return error;
2598}
2599
3480b257 2600SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
5590ff0d
UD		 2601{
2602 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2603}
2604
1da177e4
LT		 2605int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2606{
a95164d9 2607 int error = may_create(dir, dentry);
1da177e4
LT		 2608
2609 if (error)
2610 return error;
2611
acfa4380 2612 if (!dir->i_op->mkdir)
1da177e4
LT		 2613 return -EPERM;
2614
2615 mode &= (S_IRWXUGO|S_ISVTX);
2616 error = security_inode_mkdir(dir, dentry, mode);
2617 if (error)
2618 return error;
2619
1da177e4 2620 error = dir->i_op->mkdir(dir, dentry, mode);
a74574aa 2621 if (!error)
f38aa942 2622 fsnotify_mkdir(dir, dentry);
1da177e4
LT		 2623 return error;
2624}
2625
2e4d0924 2626SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
1da177e4
LT		 2627{
2628 int error = 0;
2629 char * tmp;
6902d925
DH		 2630 struct dentry *dentry;
2631 struct nameidata nd;
1da177e4 2632
2ad94ae6
AV		 2633 error = user_path_parent(dfd, pathname, &nd, &tmp);
2634 if (error)
6902d925 2635 goto out_err;
1da177e4 2636
6902d925
DH		 2637 dentry = lookup_create(&nd, 1);
2638 error = PTR_ERR(dentry);
2639 if (IS_ERR(dentry))
2640 goto out_unlock;
1da177e4 2641
4ac91378 2642 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2643 mode &= ~current_umask();
463c3197
DH		 2644 error = mnt_want_write(nd.path.mnt);
2645 if (error)
2646 goto out_dput;
be6d3e56
KT		 2647 error = security_path_mkdir(&nd.path, dentry, mode);
2648 if (error)
2649 goto out_drop_write;
4ac91378 2650 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
be6d3e56 2651out_drop_write:
463c3197
DH		 2652 mnt_drop_write(nd.path.mnt);
2653out_dput:
6902d925
DH		 2654 dput(dentry);
2655out_unlock:
4ac91378 2656 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2657 path_put(&nd.path);
6902d925
DH		 2658 putname(tmp);
2659out_err:
1da177e4
LT		 2660 return error;
2661}
2662
3cdad428 2663SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
5590ff0d
UD		 2664{
2665 return sys_mkdirat(AT_FDCWD, pathname, mode);
2666}
2667
1da177e4
LT		 2668/*
2669 * We try to drop the dentry early: we should have
2670 * a usage count of 2 if we're the only user of this
2671 * dentry, and if that is true (possibly after pruning
2672 * the dcache), then we drop the dentry now.
2673 *
2674 * A low-level filesystem can, if it choses, legally
2675 * do a
2676 *
2677 * if (!d_unhashed(dentry))
2678 * return -EBUSY;
2679 *
2680 * if it cannot handle the case of removing a directory
2681 * that is still in use by something else..
2682 */
2683void dentry_unhash(struct dentry *dentry)
2684{
2685 dget(dentry);
dc168427 2686 shrink_dcache_parent(dentry);
1da177e4 2687 spin_lock(&dentry->d_lock);
b7ab39f6 2688 if (dentry->d_count == 2)
1da177e4
LT		 2689 __d_drop(dentry);
2690 spin_unlock(&dentry->d_lock);
1da177e4
LT		 2691}
2692
2693int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2694{
2695 int error = may_delete(dir, dentry, 1);
2696
2697 if (error)
2698 return error;
2699
acfa4380 2700 if (!dir->i_op->rmdir)
1da177e4
LT		 2701 return -EPERM;
2702
1b1dcc1b 2703 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2704 dentry_unhash(dentry);
2705 if (d_mountpoint(dentry))
2706 error = -EBUSY;
2707 else {
2708 error = security_inode_rmdir(dir, dentry);
2709 if (!error) {
2710 error = dir->i_op->rmdir(dir, dentry);
d83c49f3 2711 if (!error) {
1da177e4 2712 dentry->d_inode->i_flags |= S_DEAD;
d83c49f3
AV		 2713 dont_mount(dentry);
2714 }
1da177e4
LT		 2715 }
2716 }
1b1dcc1b 2717 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4 2718 if (!error) {
1da177e4
LT		 2719 d_delete(dentry);
2720 }
2721 dput(dentry);
2722
2723 return error;
2724}
2725
5590ff0d 2726static long do_rmdir(int dfd, const char __user *pathname)
1da177e4
LT		 2727{
2728 int error = 0;
2729 char * name;
2730 struct dentry *dentry;
2731 struct nameidata nd;
2732
2ad94ae6 2733 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2734 if (error)
2ad94ae6 2735 return error;
1da177e4
LT		 2736
2737 switch(nd.last_type) {
0612d9fb
OH		 2738 case LAST_DOTDOT:
2739 error = -ENOTEMPTY;
2740 goto exit1;
2741 case LAST_DOT:
2742 error = -EINVAL;
2743 goto exit1;
2744 case LAST_ROOT:
2745 error = -EBUSY;
2746 goto exit1;
1da177e4 2747 }
0612d9fb
OH		 2748
2749 nd.flags &= ~LOOKUP_PARENT;
2750
4ac91378 2751 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2752 dentry = lookup_hash(&nd);
1da177e4 2753 error = PTR_ERR(dentry);
6902d925
DH		 2754 if (IS_ERR(dentry))
2755 goto exit2;
0622753b
DH		 2756 error = mnt_want_write(nd.path.mnt);
2757 if (error)
2758 goto exit3;
be6d3e56
KT		 2759 error = security_path_rmdir(&nd.path, dentry);
2760 if (error)
2761 goto exit4;
4ac91378 2762 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
be6d3e56 2763exit4:
0622753b
DH		 2764 mnt_drop_write(nd.path.mnt);
2765exit3:
6902d925
DH		 2766 dput(dentry);
2767exit2:
4ac91378 2768 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2769exit1:
1d957f9b 2770 path_put(&nd.path);
1da177e4
LT		 2771 putname(name);
2772 return error;
2773}
2774
3cdad428 2775SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
5590ff0d
UD		 2776{
2777 return do_rmdir(AT_FDCWD, pathname);
2778}
2779
1da177e4
LT		 2780int vfs_unlink(struct inode *dir, struct dentry *dentry)
2781{
2782 int error = may_delete(dir, dentry, 0);
2783
2784 if (error)
2785 return error;
2786
acfa4380 2787 if (!dir->i_op->unlink)
1da177e4
LT		 2788 return -EPERM;
2789
1b1dcc1b 2790 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2791 if (d_mountpoint(dentry))
2792 error = -EBUSY;
2793 else {
2794 error = security_inode_unlink(dir, dentry);
bec1052e 2795 if (!error) {
1da177e4 2796 error = dir->i_op->unlink(dir, dentry);
bec1052e 2797 if (!error)
d83c49f3 2798 dont_mount(dentry);
bec1052e 2799 }
1da177e4 2800 }
1b1dcc1b 2801 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2802
2803 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2804 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
ece95912 2805 fsnotify_link_count(dentry->d_inode);
e234f35c 2806 d_delete(dentry);
1da177e4 2807 }
0eeca283 2808
1da177e4
LT		 2809 return error;
2810}
2811
2812/*
2813 * Make sure that the actual truncation of the file will occur outside its
1b1dcc1b 2814 * directory's i_mutex. Truncate can take a long time if there is a lot of
1da177e4
LT		 2815 * writeout happening, and we don't want to prevent access to the directory
2816 * while waiting on the I/O.
2817 */
5590ff0d 2818static long do_unlinkat(int dfd, const char __user *pathname)
1da177e4 2819{
2ad94ae6
AV		 2820 int error;
2821 char *name;
1da177e4
LT		 2822 struct dentry *dentry;
2823 struct nameidata nd;
2824 struct inode *inode = NULL;
2825
2ad94ae6 2826 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2827 if (error)
2ad94ae6
AV		 2828 return error;
2829
1da177e4
LT		 2830 error = -EISDIR;
2831 if (nd.last_type != LAST_NORM)
2832 goto exit1;
0612d9fb
OH		 2833
2834 nd.flags &= ~LOOKUP_PARENT;
2835
4ac91378 2836 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2837 dentry = lookup_hash(&nd);
1da177e4
LT		 2838 error = PTR_ERR(dentry);
2839 if (!IS_ERR(dentry)) {
2840 /* Why not before? Because we want correct error value */
2841 if (nd.last.name[nd.last.len])
2842 goto slashes;
2843 inode = dentry->d_inode;
2844 if (inode)
7de9c6ee 2845 ihold(inode);
0622753b
DH		 2846 error = mnt_want_write(nd.path.mnt);
2847 if (error)
2848 goto exit2;
be6d3e56
KT		 2849 error = security_path_unlink(&nd.path, dentry);
2850 if (error)
2851 goto exit3;
4ac91378 2852 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
be6d3e56 2853exit3:
0622753b 2854 mnt_drop_write(nd.path.mnt);
1da177e4
LT		 2855 exit2:
2856 dput(dentry);
2857 }
4ac91378 2858 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4
LT		 2859 if (inode)
2860 iput(inode); /* truncate the inode here */
2861exit1:
1d957f9b 2862 path_put(&nd.path);
1da177e4
LT		 2863 putname(name);
2864 return error;
2865
2866slashes:
2867 error = !dentry->d_inode ? -ENOENT :
2868 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2869 goto exit2;
2870}
2871
2e4d0924 2872SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
5590ff0d
UD		 2873{
2874 if ((flag & ~AT_REMOVEDIR) != 0)
2875 return -EINVAL;
2876
2877 if (flag & AT_REMOVEDIR)
2878 return do_rmdir(dfd, pathname);
2879
2880 return do_unlinkat(dfd, pathname);
2881}
2882
3480b257 2883SYSCALL_DEFINE1(unlink, const char __user *, pathname)
5590ff0d
UD		 2884{
2885 return do_unlinkat(AT_FDCWD, pathname);
2886}
2887
db2e747b 2888int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
1da177e4 2889{
a95164d9 2890 int error = may_create(dir, dentry);
1da177e4
LT		 2891
2892 if (error)
2893 return error;
2894
acfa4380 2895 if (!dir->i_op->symlink)
1da177e4
LT		 2896 return -EPERM;
2897
2898 error = security_inode_symlink(dir, dentry, oldname);
2899 if (error)
2900 return error;
2901
1da177e4 2902 error = dir->i_op->symlink(dir, dentry, oldname);
a74574aa 2903 if (!error)
f38aa942 2904 fsnotify_create(dir, dentry);
1da177e4
LT		 2905 return error;
2906}
2907
2e4d0924
HC		 2908SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2909 int, newdfd, const char __user *, newname)
1da177e4 2910{
2ad94ae6
AV		 2911 int error;
2912 char *from;
2913 char *to;
6902d925
DH		 2914 struct dentry *dentry;
2915 struct nameidata nd;
1da177e4
LT		 2916
2917 from = getname(oldname);
2ad94ae6 2918 if (IS_ERR(from))
1da177e4 2919 return PTR_ERR(from);
1da177e4 2920
2ad94ae6 2921 error = user_path_parent(newdfd, newname, &nd, &to);
6902d925 2922 if (error)
2ad94ae6
AV		 2923 goto out_putname;
2924
6902d925
DH		 2925 dentry = lookup_create(&nd, 0);
2926 error = PTR_ERR(dentry);
2927 if (IS_ERR(dentry))
2928 goto out_unlock;
2929
75c3f29d
DH		 2930 error = mnt_want_write(nd.path.mnt);
2931 if (error)
2932 goto out_dput;
be6d3e56
KT		 2933 error = security_path_symlink(&nd.path, dentry, from);
2934 if (error)
2935 goto out_drop_write;
db2e747b 2936 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
be6d3e56 2937out_drop_write:
75c3f29d
DH		 2938 mnt_drop_write(nd.path.mnt);
2939out_dput:
6902d925
DH		 2940 dput(dentry);
2941out_unlock:
4ac91378 2942 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2943 path_put(&nd.path);
6902d925
DH		 2944 putname(to);
2945out_putname:
1da177e4
LT		 2946 putname(from);
2947 return error;
2948}
2949
3480b257 2950SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 2951{
2952 return sys_symlinkat(oldname, AT_FDCWD, newname);
2953}
2954
1da177e4
LT		 2955int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2956{
2957 struct inode *inode = old_dentry->d_inode;
2958 int error;
2959
2960 if (!inode)
2961 return -ENOENT;
2962
a95164d9 2963 error = may_create(dir, new_dentry);
1da177e4
LT		 2964 if (error)
2965 return error;
2966
2967 if (dir->i_sb != inode->i_sb)
2968 return -EXDEV;
2969
2970 /*
2971 * A link to an append-only or immutable file cannot be created.
2972 */
2973 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2974 return -EPERM;
acfa4380 2975 if (!dir->i_op->link)
1da177e4 2976 return -EPERM;
7e79eedb 2977 if (S_ISDIR(inode->i_mode))
1da177e4
LT		 2978 return -EPERM;
2979
2980 error = security_inode_link(old_dentry, dir, new_dentry);
2981 if (error)
2982 return error;
2983
7e79eedb 2984 mutex_lock(&inode->i_mutex);
1da177e4 2985 error = dir->i_op->link(old_dentry, dir, new_dentry);
7e79eedb 2986 mutex_unlock(&inode->i_mutex);
e31e14ec 2987 if (!error)
7e79eedb 2988 fsnotify_link(dir, inode, new_dentry);
1da177e4
LT		 2989 return error;
2990}
2991
2992/*
2993 * Hardlinks are often used in delicate situations. We avoid
2994 * security-related surprises by not following symlinks on the
2995 * newname. --KAB
2996 *
2997 * We don't follow them on the oldname either to be compatible
2998 * with linux 2.0, and to avoid hard-linking to directories
2999 * and other special files. --ADM
3000 */
2e4d0924
HC		 3001SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3002 int, newdfd, const char __user *, newname, int, flags)
1da177e4
LT		 3003{
3004 struct dentry *new_dentry;
2d8f3038
AV		 3005 struct nameidata nd;
3006 struct path old_path;
1da177e4 3007 int error;
2ad94ae6 3008 char *to;
1da177e4 3009
45c9b11a 3010 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
c04030e1
UD		 3011 return -EINVAL;
3012
2d8f3038
AV		 3013 error = user_path_at(olddfd, oldname,
3014 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
3015 &old_path);
1da177e4 3016 if (error)
2ad94ae6
AV		 3017 return error;
3018
3019 error = user_path_parent(newdfd, newname, &nd, &to);
1da177e4
LT		 3020 if (error)
3021 goto out;
3022 error = -EXDEV;
2d8f3038 3023 if (old_path.mnt != nd.path.mnt)
1da177e4
LT		 3024 goto out_release;
3025 new_dentry = lookup_create(&nd, 0);
3026 error = PTR_ERR(new_dentry);
6902d925
DH		 3027 if (IS_ERR(new_dentry))
3028 goto out_unlock;
75c3f29d
DH		 3029 error = mnt_want_write(nd.path.mnt);
3030 if (error)
3031 goto out_dput;
be6d3e56
KT		 3032 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
3033 if (error)
3034 goto out_drop_write;
2d8f3038 3035 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
be6d3e56 3036out_drop_write:
75c3f29d
DH		 3037 mnt_drop_write(nd.path.mnt);
3038out_dput:
6902d925
DH		 3039 dput(new_dentry);
3040out_unlock:
4ac91378 3041 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 3042out_release:
1d957f9b 3043 path_put(&nd.path);
2ad94ae6 3044 putname(to);
1da177e4 3045out:
2d8f3038 3046 path_put(&old_path);
1da177e4
LT		 3047
3048 return error;
3049}
3050
3480b257 3051SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
5590ff0d 3052{
c04030e1 3053 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
5590ff0d
UD		 3054}
3055
1da177e4
LT		 3056/*
3057 * The worst of all namespace operations - renaming directory. "Perverted"
3058 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3059 * Problems:
3060 * a) we can get into loop creation. Check is done in is_subdir().
3061 * b) race potential - two innocent renames can create a loop together.
3062 * That's where 4.4 screws up. Current fix: serialization on
a11f3a05 3063 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
1da177e4
LT		 3064 * story.
3065 * c) we have to lock _three_ objects - parents and victim (if it exists).
1b1dcc1b 3066 * And that - after we got ->i_mutex on parents (until then we don't know
1da177e4
LT		 3067 * whether the target exists). Solution: try to be smart with locking
3068 * order for inodes. We rely on the fact that tree topology may change
a11f3a05 3069 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
1da177e4
LT		 3070 * move will be locked. Thus we can rank directories by the tree
3071 * (ancestors first) and rank all non-directories after them.
3072 * That works since everybody except rename does "lock parent, lookup,
a11f3a05 3073 * lock child" and rename is under ->s_vfs_rename_mutex.
1da177e4
LT		 3074 * HOWEVER, it relies on the assumption that any object with ->lookup()
3075 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3076 * we'd better make sure that there's no link(2) for them.
3077 * d) some filesystems don't support opened-but-unlinked directories,
3078 * either because of layout or because they are not ready to deal with
3079 * all cases correctly. The latter will be fixed (taking this sort of
3080 * stuff into VFS), but the former is not going away. Solution: the same
3081 * trick as in rmdir().
3082 * e) conversion from fhandle to dentry may come in the wrong moment - when
1b1dcc1b 3083 * we are removing the target. Solution: we will have to grab ->i_mutex
1da177e4 3084 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
c41b20e7 3085 * ->i_mutex on parents, which works but leads to some truly excessive
1da177e4
LT		 3086 * locking].
3087 */
75c96f85
AB		 3088static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3089 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3090{
3091 int error = 0;
3092 struct inode *target;
3093
3094 /*
3095 * If we are going to change the parent - check write permissions,
3096 * we'll need to flip '..'.
3097 */
3098 if (new_dir != old_dir) {
f419a2e3 3099 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
1da177e4
LT		 3100 if (error)
3101 return error;
3102 }
3103
3104 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3105 if (error)
3106 return error;
3107
3108 target = new_dentry->d_inode;
d83c49f3 3109 if (target)
1b1dcc1b 3110 mutex_lock(&target->i_mutex);
1da177e4
LT		 3111 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3112 error = -EBUSY;
d83c49f3
AV		 3113 else {
3114 if (target)
3115 dentry_unhash(new_dentry);
1da177e4 3116 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
d83c49f3 3117 }
1da177e4 3118 if (target) {
d83c49f3 3119 if (!error) {
1da177e4 3120 target->i_flags |= S_DEAD;
d83c49f3
AV		 3121 dont_mount(new_dentry);
3122 }
1b1dcc1b 3123 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3124 if (d_unhashed(new_dentry))
3125 d_rehash(new_dentry);
3126 dput(new_dentry);
3127 }
e31e14ec 3128 if (!error)
349457cc
MF		 3129 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3130 d_move(old_dentry,new_dentry);
1da177e4
LT		 3131 return error;
3132}
3133
75c96f85
AB		 3134static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3135 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3136{
3137 struct inode *target;
3138 int error;
3139
3140 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3141 if (error)
3142 return error;
3143
3144 dget(new_dentry);
3145 target = new_dentry->d_inode;
3146 if (target)
1b1dcc1b 3147 mutex_lock(&target->i_mutex);
1da177e4
LT		 3148 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3149 error = -EBUSY;
3150 else
3151 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3152 if (!error) {
bec1052e 3153 if (target)
d83c49f3 3154 dont_mount(new_dentry);
349457cc 3155 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
1da177e4 3156 d_move(old_dentry, new_dentry);
1da177e4
LT		 3157 }
3158 if (target)
1b1dcc1b 3159 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3160 dput(new_dentry);
3161 return error;
3162}
3163
3164int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3165 struct inode *new_dir, struct dentry *new_dentry)
3166{
3167 int error;
3168 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
59b0df21 3169 const unsigned char *old_name;
1da177e4
LT		 3170
3171 if (old_dentry->d_inode == new_dentry->d_inode)
3172 return 0;
3173
3174 error = may_delete(old_dir, old_dentry, is_dir);
3175 if (error)
3176 return error;
3177
3178 if (!new_dentry->d_inode)
a95164d9 3179 error = may_create(new_dir, new_dentry);
1da177e4
LT		 3180 else
3181 error = may_delete(new_dir, new_dentry, is_dir);
3182 if (error)
3183 return error;
3184
acfa4380 3185 if (!old_dir->i_op->rename)
1da177e4
LT		 3186 return -EPERM;
3187
0eeca283
RL		 3188 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3189
1da177e4
LT		 3190 if (is_dir)
3191 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3192 else
3193 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
123df294
AV		 3194 if (!error)
3195 fsnotify_move(old_dir, new_dir, old_name, is_dir,
5a190ae6 3196 new_dentry->d_inode, old_dentry);
0eeca283
RL		 3197 fsnotify_oldname_free(old_name);
3198
1da177e4
LT		 3199 return error;
3200}
3201
2e4d0924
HC		 3202SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3203 int, newdfd, const char __user *, newname)
1da177e4 3204{
2ad94ae6
AV		 3205 struct dentry *old_dir, *new_dir;
3206 struct dentry *old_dentry, *new_dentry;
3207 struct dentry *trap;
1da177e4 3208 struct nameidata oldnd, newnd;
2ad94ae6
AV		 3209 char *from;
3210 char *to;
3211 int error;
1da177e4 3212
2ad94ae6 3213 error = user_path_parent(olddfd, oldname, &oldnd, &from);
1da177e4
LT		 3214 if (error)
3215 goto exit;
3216
2ad94ae6 3217 error = user_path_parent(newdfd, newname, &newnd, &to);
1da177e4
LT		 3218 if (error)
3219 goto exit1;
3220
3221 error = -EXDEV;
4ac91378 3222 if (oldnd.path.mnt != newnd.path.mnt)
1da177e4
LT		 3223 goto exit2;
3224
4ac91378 3225 old_dir = oldnd.path.dentry;
1da177e4
LT		 3226 error = -EBUSY;
3227 if (oldnd.last_type != LAST_NORM)
3228 goto exit2;
3229
4ac91378 3230 new_dir = newnd.path.dentry;
1da177e4
LT		 3231 if (newnd.last_type != LAST_NORM)
3232 goto exit2;
3233
0612d9fb
OH		 3234 oldnd.flags &= ~LOOKUP_PARENT;
3235 newnd.flags &= ~LOOKUP_PARENT;
4e9ed2f8 3236 newnd.flags |= LOOKUP_RENAME_TARGET;
0612d9fb 3237
1da177e4
LT		 3238 trap = lock_rename(new_dir, old_dir);
3239
49705b77 3240 old_dentry = lookup_hash(&oldnd);
1da177e4
LT		 3241 error = PTR_ERR(old_dentry);
3242 if (IS_ERR(old_dentry))
3243 goto exit3;
3244 /* source must exist */
3245 error = -ENOENT;
3246 if (!old_dentry->d_inode)
3247 goto exit4;
3248 /* unless the source is a directory trailing slashes give -ENOTDIR */
3249 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3250 error = -ENOTDIR;
3251 if (oldnd.last.name[oldnd.last.len])
3252 goto exit4;
3253 if (newnd.last.name[newnd.last.len])
3254 goto exit4;
3255 }
3256 /* source should not be ancestor of target */
3257 error = -EINVAL;
3258 if (old_dentry == trap)
3259 goto exit4;
49705b77 3260 new_dentry = lookup_hash(&newnd);
1da177e4
LT		 3261 error = PTR_ERR(new_dentry);
3262 if (IS_ERR(new_dentry))
3263 goto exit4;
3264 /* target should not be an ancestor of source */
3265 error = -ENOTEMPTY;
3266 if (new_dentry == trap)
3267 goto exit5;
3268
9079b1eb
DH		 3269 error = mnt_want_write(oldnd.path.mnt);
3270 if (error)
3271 goto exit5;
be6d3e56
KT		 3272 error = security_path_rename(&oldnd.path, old_dentry,
3273 &newnd.path, new_dentry);
3274 if (error)
3275 goto exit6;
1da177e4
LT		 3276 error = vfs_rename(old_dir->d_inode, old_dentry,
3277 new_dir->d_inode, new_dentry);
be6d3e56 3278exit6:
9079b1eb 3279 mnt_drop_write(oldnd.path.mnt);
1da177e4
LT		 3280exit5:
3281 dput(new_dentry);
3282exit4:
3283 dput(old_dentry);
3284exit3:
3285 unlock_rename(new_dir, old_dir);
3286exit2:
1d957f9b 3287 path_put(&newnd.path);
2ad94ae6 3288 putname(to);
1da177e4 3289exit1:
1d957f9b 3290 path_put(&oldnd.path);
1da177e4 3291 putname(from);
2ad94ae6 3292exit:
1da177e4
LT		 3293 return error;
3294}
3295
a26eab24 3296SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 3297{
3298 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3299}
3300
1da177e4
LT		 3301int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3302{
3303 int len;
3304
3305 len = PTR_ERR(link);
3306 if (IS_ERR(link))
3307 goto out;
3308
3309 len = strlen(link);
3310 if (len > (unsigned) buflen)
3311 len = buflen;
3312 if (copy_to_user(buffer, link, len))
3313 len = -EFAULT;
3314out:
3315 return len;
3316}
3317
3318/*
3319 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3320 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3321 * using) it for any given inode is up to filesystem.
3322 */
3323int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3324{
3325 struct nameidata nd;
cc314eef 3326 void *cookie;
694a1764 3327 int res;
cc314eef 3328
1da177e4 3329 nd.depth = 0;
cc314eef 3330 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
694a1764
MS		 3331 if (IS_ERR(cookie))
3332 return PTR_ERR(cookie);
3333
3334 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3335 if (dentry->d_inode->i_op->put_link)
3336 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3337 return res;
1da177e4
LT		 3338}
3339
3340int vfs_follow_link(struct nameidata *nd, const char *link)
3341{
3342 return __vfs_follow_link(nd, link);
3343}
3344
3345/* get the link contents into pagecache */
3346static char *page_getlink(struct dentry * dentry, struct page **ppage)
3347{
ebd09abb
DG		 3348 char *kaddr;
3349 struct page *page;
1da177e4 3350 struct address_space *mapping = dentry->d_inode->i_mapping;
090d2b18 3351 page = read_mapping_page(mapping, 0, NULL);
1da177e4 3352 if (IS_ERR(page))
6fe6900e 3353 return (char*)page;
1da177e4 3354 *ppage = page;
ebd09abb
DG		 3355 kaddr = kmap(page);
3356 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3357 return kaddr;
1da177e4
LT		 3358}
3359
3360int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3361{
3362 struct page *page = NULL;
3363 char *s = page_getlink(dentry, &page);
3364 int res = vfs_readlink(dentry,buffer,buflen,s);
3365 if (page) {
3366 kunmap(page);
3367 page_cache_release(page);
3368 }
3369 return res;
3370}
3371
cc314eef 3372void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
1da177e4 3373{
cc314eef 3374 struct page *page = NULL;
1da177e4 3375 nd_set_link(nd, page_getlink(dentry, &page));
cc314eef 3376 return page;
1da177e4
LT		 3377}
3378
cc314eef 3379void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1da177e4 3380{
cc314eef
LT		 3381 struct page *page = cookie;
3382
3383 if (page) {
1da177e4
LT		 3384 kunmap(page);
3385 page_cache_release(page);
1da177e4
LT		 3386 }
3387}
3388
54566b2c
NP		 3389/*
3390 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3391 */
3392int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
1da177e4
LT		 3393{
3394 struct address_space *mapping = inode->i_mapping;
0adb25d2 3395 struct page *page;
afddba49 3396 void *fsdata;
beb497ab 3397 int err;
1da177e4 3398 char *kaddr;
54566b2c
NP		 3399 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3400 if (nofs)
3401 flags |= AOP_FLAG_NOFS;
1da177e4 3402
7e53cac4 3403retry:
afddba49 3404 err = pagecache_write_begin(NULL, mapping, 0, len-1,
54566b2c 3405 flags, &page, &fsdata);
1da177e4 3406 if (err)
afddba49
NP		 3407 goto fail;
3408
1da177e4
LT		 3409 kaddr = kmap_atomic(page, KM_USER0);
3410 memcpy(kaddr, symname, len-1);
3411 kunmap_atomic(kaddr, KM_USER0);
afddba49
NP		 3412
3413 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3414 page, fsdata);
1da177e4
LT		 3415 if (err < 0)
3416 goto fail;
afddba49
NP		 3417 if (err < len-1)
3418 goto retry;
3419
1da177e4
LT		 3420 mark_inode_dirty(inode);
3421 return 0;
1da177e4
LT		 3422fail:
3423 return err;
3424}
3425
0adb25d2
KK		 3426int page_symlink(struct inode *inode, const char *symname, int len)
3427{
3428 return __page_symlink(inode, symname, len,
54566b2c 3429 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
0adb25d2
KK		 3430}
3431
92e1d5be 3432const struct inode_operations page_symlink_inode_operations = {
1da177e4
LT		 3433 .readlink = generic_readlink,
3434 .follow_link = page_follow_link_light,
3435 .put_link = page_put_link,
3436};
3437
2d8f3038 3438EXPORT_SYMBOL(user_path_at);
cc53ce53 3439EXPORT_SYMBOL(follow_down_one);
1da177e4
LT		 3440EXPORT_SYMBOL(follow_down);
3441EXPORT_SYMBOL(follow_up);
3442EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3443EXPORT_SYMBOL(getname);
3444EXPORT_SYMBOL(lock_rename);
1da177e4
LT		 3445EXPORT_SYMBOL(lookup_one_len);
3446EXPORT_SYMBOL(page_follow_link_light);
3447EXPORT_SYMBOL(page_put_link);
3448EXPORT_SYMBOL(page_readlink);
0adb25d2 3449EXPORT_SYMBOL(__page_symlink);
1da177e4
LT		 3450EXPORT_SYMBOL(page_symlink);
3451EXPORT_SYMBOL(page_symlink_inode_operations);
c9c6cac0 3452EXPORT_SYMBOL(kern_path_parent);
d1811465 3453EXPORT_SYMBOL(kern_path);
16f18200 3454EXPORT_SYMBOL(vfs_path_lookup);
f419a2e3 3455EXPORT_SYMBOL(inode_permission);
8c744fb8 3456EXPORT_SYMBOL(file_permission);
1da177e4
LT		 3457EXPORT_SYMBOL(unlock_rename);
3458EXPORT_SYMBOL(vfs_create);
3459EXPORT_SYMBOL(vfs_follow_link);
3460EXPORT_SYMBOL(vfs_link);
3461EXPORT_SYMBOL(vfs_mkdir);
3462EXPORT_SYMBOL(vfs_mknod);
3463EXPORT_SYMBOL(generic_permission);
3464EXPORT_SYMBOL(vfs_readlink);
3465EXPORT_SYMBOL(vfs_rename);
3466EXPORT_SYMBOL(vfs_rmdir);
3467EXPORT_SYMBOL(vfs_symlink);
3468EXPORT_SYMBOL(vfs_unlink);
3469EXPORT_SYMBOL(dentry_unhash);
3470EXPORT_SYMBOL(generic_readlink);
kernel source for telekom puls (alcatel/mediatek)