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