projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
take RCU-dependent stuff around exec_permission() into a new helper
[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
542/* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
543static inline int nameidata_drop_rcu_last_maybe(struct nameidata *nd)
544{
545 if (likely(nd->flags & LOOKUP_RCU))
546 return nameidata_drop_rcu_last(nd);
547 return 0;
548}
549
834f2a4a
TM		 550/**
551 * release_open_intent - free up open intent resources
552 * @nd: pointer to nameidata
553 */
554void release_open_intent(struct nameidata *nd)
555{
2dab5974
LT		 556 struct file *file = nd->intent.open.file;
557
558 if (file && !IS_ERR(file)) {
559 if (file->f_path.dentry == NULL)
560 put_filp(file);
561 else
562 fput(file);
563 }
834f2a4a
TM		 564}
565
f60aef7e 566static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
34286d66 567{
f60aef7e 568 return dentry->d_op->d_revalidate(dentry, nd);
34286d66
NP		 569}
570
f5e1c1c1 571static struct dentry *
bcdc5e01
IK		 572do_revalidate(struct dentry *dentry, struct nameidata *nd)
573{
f5e1c1c1 574 int status = d_revalidate(dentry, nd);
bcdc5e01
IK		 575 if (unlikely(status <= 0)) {
576 /*
577 * The dentry failed validation.
578 * If d_revalidate returned 0 attempt to invalidate
579 * the dentry otherwise d_revalidate is asking us
580 * to return a fail status.
581 */
34286d66 582 if (status < 0) {
f5e1c1c1 583 dput(dentry);
34286d66 584 dentry = ERR_PTR(status);
f5e1c1c1
AV		 585 } else if (!d_invalidate(dentry)) {
586 dput(dentry);
587 dentry = NULL;
bcdc5e01
IK		 588 }
589 }
590 return dentry;
591}
592
f5e1c1c1
AV		 593static inline struct dentry *
594do_revalidate_rcu(struct dentry *dentry, struct nameidata *nd)
595{
f60aef7e 596 int status = d_revalidate(dentry, nd);
f5e1c1c1
AV		 597 if (likely(status > 0))
598 return dentry;
599 if (status == -ECHILD) {
600 if (nameidata_dentry_drop_rcu(nd, dentry))
601 return ERR_PTR(-ECHILD);
602 return do_revalidate(dentry, nd);
603 }
604 if (status < 0)
605 return ERR_PTR(status);
606 /* Don't d_invalidate in rcu-walk mode */
607 if (nameidata_dentry_drop_rcu(nd, dentry))
608 return ERR_PTR(-ECHILD);
609 if (!d_invalidate(dentry)) {
610 dput(dentry);
611 dentry = NULL;
612 }
613 return dentry;
614}
615
fb045adb
NP		 616static inline int need_reval_dot(struct dentry *dentry)
617{
618 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
619 return 0;
620
621 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
622 return 0;
623
624 return 1;
625}
626
39159de2
JL		 627/*
628 * force_reval_path - force revalidation of a dentry
629 *
630 * In some situations the path walking code will trust dentries without
631 * revalidating them. This causes problems for filesystems that depend on
632 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
633 * (which indicates that it's possible for the dentry to go stale), force
634 * a d_revalidate call before proceeding.
635 *
636 * Returns 0 if the revalidation was successful. If the revalidation fails,
637 * either return the error returned by d_revalidate or -ESTALE if the
638 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
639 * invalidate the dentry. It's up to the caller to handle putting references
640 * to the path if necessary.
641 */
642static int
643force_reval_path(struct path *path, struct nameidata *nd)
644{
645 int status;
646 struct dentry *dentry = path->dentry;
647
648 /*
649 * only check on filesystems where it's possible for the dentry to
fb045adb 650 * become stale.
39159de2 651 */
fb045adb 652 if (!need_reval_dot(dentry))
39159de2
JL		 653 return 0;
654
34286d66 655 status = d_revalidate(dentry, nd);
39159de2
JL		 656 if (status > 0)
657 return 0;
658
659 if (!status) {
660 d_invalidate(dentry);
661 status = -ESTALE;
662 }
663 return status;
664}
665
1da177e4 666/*
b75b5086
AV		 667 * Short-cut version of permission(), for calling on directories
668 * during pathname resolution. Combines parts of permission()
669 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
1da177e4
LT		 670 *
671 * If appropriate, check DAC only. If not appropriate, or
b75b5086 672 * short-cut DAC fails, then call ->permission() to do more
1da177e4
LT		 673 * complete permission check.
674 */
b74c79e9 675static inline int exec_permission(struct inode *inode, unsigned int flags)
1da177e4 676{
5909ccaa 677 int ret;
1da177e4 678
cb9179ea 679 if (inode->i_op->permission) {
b74c79e9
NP		 680 ret = inode->i_op->permission(inode, MAY_EXEC, flags);
681 } else {
682 ret = acl_permission_check(inode, MAY_EXEC, flags,
683 inode->i_op->check_acl);
cb9179ea 684 }
b74c79e9 685 if (likely(!ret))
1da177e4 686 goto ok;
b74c79e9 687 if (ret == -ECHILD)
31e6b01f 688 return ret;
1da177e4 689
f1ac9f6b 690 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
1da177e4
LT		 691 goto ok;
692
5909ccaa 693 return ret;
1da177e4 694ok:
b74c79e9 695 return security_inode_exec_permission(inode, flags);
1da177e4
LT		 696}
697
2a737871
AV		 698static __always_inline void set_root(struct nameidata *nd)
699{
f7ad3c6b
MS		 700 if (!nd->root.mnt)
701 get_fs_root(current->fs, &nd->root);
2a737871
AV		 702}
703
6de88d72
AV		 704static int link_path_walk(const char *, struct nameidata *);
705
31e6b01f
NP		 706static __always_inline void set_root_rcu(struct nameidata *nd)
707{
708 if (!nd->root.mnt) {
709 struct fs_struct *fs = current->fs;
c28cc364
NP		 710 unsigned seq;
711
712 do {
713 seq = read_seqcount_begin(&fs->seq);
714 nd->root = fs->root;
715 } while (read_seqcount_retry(&fs->seq, seq));
31e6b01f
NP		 716 }
717}
718
f1662356 719static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
1da177e4 720{
31e6b01f
NP		 721 int ret;
722
1da177e4
LT		 723 if (IS_ERR(link))
724 goto fail;
725
726 if (*link == '/') {
2a737871 727 set_root(nd);
1d957f9b 728 path_put(&nd->path);
2a737871
AV		 729 nd->path = nd->root;
730 path_get(&nd->root);
1da177e4 731 }
31e6b01f 732 nd->inode = nd->path.dentry->d_inode;
b4091d5f 733
31e6b01f
NP		 734 ret = link_path_walk(link, nd);
735 return ret;
1da177e4 736fail:
1d957f9b 737 path_put(&nd->path);
1da177e4
LT		 738 return PTR_ERR(link);
739}
740
1d957f9b 741static void path_put_conditional(struct path *path, struct nameidata *nd)
051d3812
IK		 742{
743 dput(path->dentry);
4ac91378 744 if (path->mnt != nd->path.mnt)
051d3812
IK		 745 mntput(path->mnt);
746}
747
7b9337aa
NP		 748static inline void path_to_nameidata(const struct path *path,
749 struct nameidata *nd)
051d3812 750{
31e6b01f
NP		 751 if (!(nd->flags & LOOKUP_RCU)) {
752 dput(nd->path.dentry);
753 if (nd->path.mnt != path->mnt)
754 mntput(nd->path.mnt);
9a229683 755 }
31e6b01f 756 nd->path.mnt = path->mnt;
4ac91378 757 nd->path.dentry = path->dentry;
051d3812
IK		 758}
759
def4af30 760static __always_inline int
7b9337aa 761__do_follow_link(const struct path *link, struct nameidata *nd, void **p)
1da177e4
LT		 762{
763 int error;
7b9337aa 764 struct dentry *dentry = link->dentry;
1da177e4 765
844a3917
AV		 766 BUG_ON(nd->flags & LOOKUP_RCU);
767
7b9337aa 768 touch_atime(link->mnt, dentry);
1da177e4 769 nd_set_link(nd, NULL);
cd4e91d3 770
87556ef1
DH		 771 if (link->mnt == nd->path.mnt)
772 mntget(link->mnt);
31e6b01f 773
86acdca1 774 nd->last_type = LAST_BIND;
def4af30
AV		 775 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
776 error = PTR_ERR(*p);
777 if (!IS_ERR(*p)) {
1da177e4 778 char *s = nd_get_link(nd);
cc314eef 779 error = 0;
1da177e4
LT		 780 if (s)
781 error = __vfs_follow_link(nd, s);
39159de2
JL		 782 else if (nd->last_type == LAST_BIND) {
783 error = force_reval_path(&nd->path, nd);
784 if (error)
785 path_put(&nd->path);
786 }
1da177e4 787 }
1da177e4
LT		 788 return error;
789}
790
791/*
792 * This limits recursive symlink follows to 8, while
793 * limiting consecutive symlinks to 40.
794 *
795 * Without that kind of total limit, nasty chains of consecutive
796 * symlinks can cause almost arbitrarily long lookups.
797 */
3abb17e8 798static inline int do_follow_link(struct inode *inode, struct path *path, struct nameidata *nd)
1da177e4 799{
def4af30 800 void *cookie;
1da177e4 801 int err = -ELOOP;
844a3917
AV		 802
803 /* We drop rcu-walk here */
804 if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry))
805 return -ECHILD;
3abb17e8 806 BUG_ON(inode != path->dentry->d_inode);
844a3917 807
1da177e4
LT		 808 if (current->link_count >= MAX_NESTED_LINKS)
809 goto loop;
810 if (current->total_link_count >= 40)
811 goto loop;
812 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
813 cond_resched();
90ebe565 814 err = security_inode_follow_link(path->dentry, nd);
1da177e4
LT		 815 if (err)
816 goto loop;
817 current->link_count++;
818 current->total_link_count++;
819 nd->depth++;
def4af30
AV		 820 err = __do_follow_link(path, nd, &cookie);
821 if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link)
822 path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie);
258fa999 823 path_put(path);
839d9f93
AV		 824 current->link_count--;
825 nd->depth--;
1da177e4
LT		 826 return err;
827loop:
1d957f9b
JB		 828 path_put_conditional(path, nd);
829 path_put(&nd->path);
1da177e4
LT		 830 return err;
831}
832
31e6b01f
NP		 833static int follow_up_rcu(struct path *path)
834{
835 struct vfsmount *parent;
836 struct dentry *mountpoint;
837
838 parent = path->mnt->mnt_parent;
839 if (parent == path->mnt)
840 return 0;
841 mountpoint = path->mnt->mnt_mountpoint;
842 path->dentry = mountpoint;
843 path->mnt = parent;
844 return 1;
845}
846
bab77ebf 847int follow_up(struct path *path)
1da177e4
LT		 848{
849 struct vfsmount *parent;
850 struct dentry *mountpoint;
99b7db7b
NP		 851
852 br_read_lock(vfsmount_lock);
bab77ebf
AV		 853 parent = path->mnt->mnt_parent;
854 if (parent == path->mnt) {
99b7db7b 855 br_read_unlock(vfsmount_lock);
1da177e4
LT		 856 return 0;
857 }
858 mntget(parent);
bab77ebf 859 mountpoint = dget(path->mnt->mnt_mountpoint);
99b7db7b 860 br_read_unlock(vfsmount_lock);
bab77ebf
AV		 861 dput(path->dentry);
862 path->dentry = mountpoint;
863 mntput(path->mnt);
864 path->mnt = parent;
1da177e4
LT		 865 return 1;
866}
867
b5c84bf6 868/*
9875cf80
DH		 869 * Perform an automount
870 * - return -EISDIR to tell follow_managed() to stop and return the path we
871 * were called with.
1da177e4 872 */
9875cf80
DH		 873static int follow_automount(struct path *path, unsigned flags,
874 bool *need_mntput)
31e6b01f 875{
9875cf80 876 struct vfsmount *mnt;
ea5b778a 877 int err;
9875cf80
DH		 878
879 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
880 return -EREMOTE;
881
6f45b656
DH		 882 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
883 * and this is the terminal part of the path.
884 */
885 if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE))
886 return -EISDIR; /* we actually want to stop here */
887
9875cf80
DH		 888 /* We want to mount if someone is trying to open/create a file of any
889 * type under the mountpoint, wants to traverse through the mountpoint
890 * or wants to open the mounted directory.
891 *
892 * We don't want to mount if someone's just doing a stat and they've
893 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
894 * appended a '/' to the name.
895 */
896 if (!(flags & LOOKUP_FOLLOW) &&
897 !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY |
898 LOOKUP_OPEN | LOOKUP_CREATE)))
899 return -EISDIR;
900
901 current->total_link_count++;
902 if (current->total_link_count >= 40)
903 return -ELOOP;
904
905 mnt = path->dentry->d_op->d_automount(path);
906 if (IS_ERR(mnt)) {
907 /*
908 * The filesystem is allowed to return -EISDIR here to indicate
909 * it doesn't want to automount. For instance, autofs would do
910 * this so that its userspace daemon can mount on this dentry.
911 *
912 * However, we can only permit this if it's a terminal point in
913 * the path being looked up; if it wasn't then the remainder of
914 * the path is inaccessible and we should say so.
915 */
916 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE))
917 return -EREMOTE;
918 return PTR_ERR(mnt);
31e6b01f 919 }
ea5b778a 920
9875cf80
DH		 921 if (!mnt) /* mount collision */
922 return 0;
31e6b01f 923
19a167af 924 err = finish_automount(mnt, path);
9875cf80 925
ea5b778a
DH		 926 switch (err) {
927 case -EBUSY:
928 /* Someone else made a mount here whilst we were busy */
19a167af 929 return 0;
ea5b778a 930 case 0:
ea5b778a
DH		 931 dput(path->dentry);
932 if (*need_mntput)
933 mntput(path->mnt);
934 path->mnt = mnt;
935 path->dentry = dget(mnt->mnt_root);
936 *need_mntput = true;
937 return 0;
19a167af
AV		 938 default:
939 return err;
ea5b778a 940 }
19a167af 941
463ffb2e
AV		 942}
943
9875cf80
DH		 944/*
945 * Handle a dentry that is managed in some way.
cc53ce53 946 * - Flagged for transit management (autofs)
9875cf80
DH		 947 * - Flagged as mountpoint
948 * - Flagged as automount point
949 *
950 * This may only be called in refwalk mode.
951 *
952 * Serialization is taken care of in namespace.c
953 */
954static int follow_managed(struct path *path, unsigned flags)
1da177e4 955{
9875cf80
DH		 956 unsigned managed;
957 bool need_mntput = false;
958 int ret;
959
960 /* Given that we're not holding a lock here, we retain the value in a
961 * local variable for each dentry as we look at it so that we don't see
962 * the components of that value change under us */
963 while (managed = ACCESS_ONCE(path->dentry->d_flags),
964 managed &= DCACHE_MANAGED_DENTRY,
965 unlikely(managed != 0)) {
cc53ce53
DH		 966 /* Allow the filesystem to manage the transit without i_mutex
967 * being held. */
968 if (managed & DCACHE_MANAGE_TRANSIT) {
969 BUG_ON(!path->dentry->d_op);
970 BUG_ON(!path->dentry->d_op->d_manage);
ab90911f
DH		 971 ret = path->dentry->d_op->d_manage(path->dentry,
972 false, false);
cc53ce53
DH		 973 if (ret < 0)
974 return ret == -EISDIR ? 0 : ret;
975 }
976
9875cf80
DH		 977 /* Transit to a mounted filesystem. */
978 if (managed & DCACHE_MOUNTED) {
979 struct vfsmount *mounted = lookup_mnt(path);
980 if (mounted) {
981 dput(path->dentry);
982 if (need_mntput)
983 mntput(path->mnt);
984 path->mnt = mounted;
985 path->dentry = dget(mounted->mnt_root);
986 need_mntput = true;
987 continue;
988 }
989
990 /* Something is mounted on this dentry in another
991 * namespace and/or whatever was mounted there in this
992 * namespace got unmounted before we managed to get the
993 * vfsmount_lock */
994 }
995
996 /* Handle an automount point */
997 if (managed & DCACHE_NEED_AUTOMOUNT) {
998 ret = follow_automount(path, flags, &need_mntput);
999 if (ret < 0)
1000 return ret == -EISDIR ? 0 : ret;
1001 continue;
1002 }
1003
1004 /* We didn't change the current path point */
1005 break;
1da177e4 1006 }
9875cf80 1007 return 0;
1da177e4
LT		 1008}
1009
cc53ce53 1010int follow_down_one(struct path *path)
1da177e4
LT		 1011{
1012 struct vfsmount *mounted;
1013
1c755af4 1014 mounted = lookup_mnt(path);
1da177e4 1015 if (mounted) {
9393bd07
AV		 1016 dput(path->dentry);
1017 mntput(path->mnt);
1018 path->mnt = mounted;
1019 path->dentry = dget(mounted->mnt_root);
1da177e4
LT		 1020 return 1;
1021 }
1022 return 0;
1023}
1024
9875cf80
DH		 1025/*
1026 * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we
cc53ce53 1027 * meet a managed dentry and we're not walking to "..". True is returned to
9875cf80
DH		 1028 * continue, false to abort.
1029 */
1030static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1031 struct inode **inode, bool reverse_transit)
1032{
1033 while (d_mountpoint(path->dentry)) {
1034 struct vfsmount *mounted;
ab90911f
DH		 1035 if (unlikely(path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) &&
1036 !reverse_transit &&
1037 path->dentry->d_op->d_manage(path->dentry, false, true) < 0)
1038 return false;
9875cf80
DH		 1039 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
1040 if (!mounted)
1041 break;
1042 path->mnt = mounted;
1043 path->dentry = mounted->mnt_root;
1044 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1045 *inode = path->dentry->d_inode;
1046 }
1047
1048 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1049 return reverse_transit;
1050 return true;
1051}
1052
31e6b01f
NP		 1053static int follow_dotdot_rcu(struct nameidata *nd)
1054{
1055 struct inode *inode = nd->inode;
1056
1057 set_root_rcu(nd);
1058
9875cf80 1059 while (1) {
31e6b01f
NP		 1060 if (nd->path.dentry == nd->root.dentry &&
1061 nd->path.mnt == nd->root.mnt) {
1062 break;
1063 }
1064 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1065 struct dentry *old = nd->path.dentry;
1066 struct dentry *parent = old->d_parent;
1067 unsigned seq;
1068
1069 seq = read_seqcount_begin(&parent->d_seq);
1070 if (read_seqcount_retry(&old->d_seq, nd->seq))
1071 return -ECHILD;
1072 inode = parent->d_inode;
1073 nd->path.dentry = parent;
1074 nd->seq = seq;
1075 break;
1076 }
1077 if (!follow_up_rcu(&nd->path))
1078 break;
1079 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1080 inode = nd->path.dentry->d_inode;
1081 }
9875cf80 1082 __follow_mount_rcu(nd, &nd->path, &inode, true);
31e6b01f
NP		 1083 nd->inode = inode;
1084
1085 return 0;
1086}
1087
cc53ce53
DH		 1088/*
1089 * Follow down to the covering mount currently visible to userspace. At each
1090 * point, the filesystem owning that dentry may be queried as to whether the
1091 * caller is permitted to proceed or not.
1092 *
1093 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1094 * being true).
1095 */
1096int follow_down(struct path *path, bool mounting_here)
1097{
1098 unsigned managed;
1099 int ret;
1100
1101 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1102 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1103 /* Allow the filesystem to manage the transit without i_mutex
1104 * being held.
1105 *
1106 * We indicate to the filesystem if someone is trying to mount
1107 * something here. This gives autofs the chance to deny anyone
1108 * other than its daemon the right to mount on its
1109 * superstructure.
1110 *
1111 * The filesystem may sleep at this point.
1112 */
1113 if (managed & DCACHE_MANAGE_TRANSIT) {
1114 BUG_ON(!path->dentry->d_op);
1115 BUG_ON(!path->dentry->d_op->d_manage);
ab90911f
DH		 1116 ret = path->dentry->d_op->d_manage(
1117 path->dentry, mounting_here, false);
cc53ce53
DH		 1118 if (ret < 0)
1119 return ret == -EISDIR ? 0 : ret;
1120 }
1121
1122 /* Transit to a mounted filesystem. */
1123 if (managed & DCACHE_MOUNTED) {
1124 struct vfsmount *mounted = lookup_mnt(path);
1125 if (!mounted)
1126 break;
1127 dput(path->dentry);
1128 mntput(path->mnt);
1129 path->mnt = mounted;
1130 path->dentry = dget(mounted->mnt_root);
1131 continue;
1132 }
1133
1134 /* Don't handle automount points here */
1135 break;
1136 }
1137 return 0;
1138}
1139
9875cf80
DH		 1140/*
1141 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1142 */
1143static void follow_mount(struct path *path)
1144{
1145 while (d_mountpoint(path->dentry)) {
1146 struct vfsmount *mounted = lookup_mnt(path);
1147 if (!mounted)
1148 break;
1149 dput(path->dentry);
1150 mntput(path->mnt);
1151 path->mnt = mounted;
1152 path->dentry = dget(mounted->mnt_root);
1153 }
1154}
1155
31e6b01f 1156static void follow_dotdot(struct nameidata *nd)
1da177e4 1157{
2a737871 1158 set_root(nd);
e518ddb7 1159
1da177e4 1160 while(1) {
4ac91378 1161 struct dentry *old = nd->path.dentry;
1da177e4 1162
2a737871
AV		 1163 if (nd->path.dentry == nd->root.dentry &&
1164 nd->path.mnt == nd->root.mnt) {
1da177e4
LT		 1165 break;
1166 }
4ac91378 1167 if (nd->path.dentry != nd->path.mnt->mnt_root) {
3088dd70
AV		 1168 /* rare case of legitimate dget_parent()... */
1169 nd->path.dentry = dget_parent(nd->path.dentry);
1da177e4
LT		 1170 dput(old);
1171 break;
1172 }
3088dd70 1173 if (!follow_up(&nd->path))
1da177e4 1174 break;
1da177e4 1175 }
79ed0226 1176 follow_mount(&nd->path);
31e6b01f 1177 nd->inode = nd->path.dentry->d_inode;
1da177e4
LT		 1178}
1179
baa03890
NP		 1180/*
1181 * Allocate a dentry with name and parent, and perform a parent
1182 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1183 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1184 * have verified that no child exists while under i_mutex.
1185 */
1186static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1187 struct qstr *name, struct nameidata *nd)
1188{
1189 struct inode *inode = parent->d_inode;
1190 struct dentry *dentry;
1191 struct dentry *old;
1192
1193 /* Don't create child dentry for a dead directory. */
1194 if (unlikely(IS_DEADDIR(inode)))
1195 return ERR_PTR(-ENOENT);
1196
1197 dentry = d_alloc(parent, name);
1198 if (unlikely(!dentry))
1199 return ERR_PTR(-ENOMEM);
1200
1201 old = inode->i_op->lookup(inode, dentry, nd);
1202 if (unlikely(old)) {
1203 dput(dentry);
1204 dentry = old;
1205 }
1206 return dentry;
1207}
1208
1da177e4
LT		 1209/*
1210 * It's more convoluted than I'd like it to be, but... it's still fairly
1211 * small and for now I'd prefer to have fast path as straight as possible.
1212 * It _is_ time-critical.
1213 */
1214static int do_lookup(struct nameidata *nd, struct qstr *name,
31e6b01f 1215 struct path *path, struct inode **inode)
1da177e4 1216{
4ac91378 1217 struct vfsmount *mnt = nd->path.mnt;
31e6b01f 1218 struct dentry *dentry, *parent = nd->path.dentry;
6e6b1bd1 1219 struct inode *dir;
9875cf80
DH		 1220 int err;
1221
3cac260a
AV		 1222 /*
1223 * See if the low-level filesystem might want
1224 * to use its own hash..
1225 */
fb045adb 1226 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
9875cf80 1227 err = parent->d_op->d_hash(parent, nd->inode, name);
3cac260a
AV		 1228 if (err < 0)
1229 return err;
1230 }
1da177e4 1231
b04f784e
NP		 1232 /*
1233 * Rename seqlock is not required here because in the off chance
1234 * of a false negative due to a concurrent rename, we're going to
1235 * do the non-racy lookup, below.
1236 */
31e6b01f
NP		 1237 if (nd->flags & LOOKUP_RCU) {
1238 unsigned seq;
1239
1240 *inode = nd->inode;
1241 dentry = __d_lookup_rcu(parent, name, &seq, inode);
1242 if (!dentry) {
1243 if (nameidata_drop_rcu(nd))
1244 return -ECHILD;
1245 goto need_lookup;
1246 }
1247 /* Memory barrier in read_seqcount_begin of child is enough */
1248 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1249 return -ECHILD;
1250
1251 nd->seq = seq;
24643087 1252 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
f5e1c1c1 1253 dentry = do_revalidate_rcu(dentry, nd);
24643087
AV		 1254 if (!dentry)
1255 goto need_lookup;
1256 if (IS_ERR(dentry))
1257 goto fail;
1258 if (!(nd->flags & LOOKUP_RCU))
1259 goto done;
1260 }
31e6b01f
NP		 1261 path->mnt = mnt;
1262 path->dentry = dentry;
9875cf80
DH		 1263 if (likely(__follow_mount_rcu(nd, path, inode, false)))
1264 return 0;
1265 if (nameidata_drop_rcu(nd))
1266 return -ECHILD;
1267 /* fallthru */
1268 }
1269 dentry = __d_lookup(parent, name);
1270 if (!dentry)
1271 goto need_lookup;
2e2e88ea 1272found:
24643087
AV		 1273 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1274 dentry = do_revalidate(dentry, nd);
1275 if (!dentry)
1276 goto need_lookup;
1277 if (IS_ERR(dentry))
1278 goto fail;
1279 }
1da177e4 1280done:
9875cf80
DH		 1281 path->mnt = mnt;
1282 path->dentry = dentry;
1283 err = follow_managed(path, nd->flags);
89312214
IK		 1284 if (unlikely(err < 0)) {
1285 path_put_conditional(path, nd);
9875cf80 1286 return err;
89312214 1287 }
9875cf80 1288 *inode = path->dentry->d_inode;
1da177e4
LT		 1289 return 0;
1290
1291need_lookup:
6e6b1bd1 1292 dir = parent->d_inode;
31e6b01f 1293 BUG_ON(nd->inode != dir);
6e6b1bd1
AV		 1294
1295 mutex_lock(&dir->i_mutex);
1296 /*
1297 * First re-do the cached lookup just in case it was created
b04f784e
NP		 1298 * while we waited for the directory semaphore, or the first
1299 * lookup failed due to an unrelated rename.
6e6b1bd1 1300 *
b04f784e
NP		 1301 * This could use version numbering or similar to avoid unnecessary
1302 * cache lookups, but then we'd have to do the first lookup in the
1303 * non-racy way. However in the common case here, everything should
1304 * be hot in cache, so would it be a big win?
6e6b1bd1
AV		 1305 */
1306 dentry = d_lookup(parent, name);
baa03890
NP		 1307 if (likely(!dentry)) {
1308 dentry = d_alloc_and_lookup(parent, name, nd);
6e6b1bd1
AV		 1309 mutex_unlock(&dir->i_mutex);
1310 if (IS_ERR(dentry))
1311 goto fail;
1312 goto done;
1313 }
6e6b1bd1
AV		 1314 /*
1315 * Uhhuh! Nasty case: the cache was re-populated while
1316 * we waited on the semaphore. Need to revalidate.
1317 */
1318 mutex_unlock(&dir->i_mutex);
2e2e88ea 1319 goto found;
1da177e4 1320
1da177e4
LT		 1321fail:
1322 return PTR_ERR(dentry);
1323}
1324
52094c8a
AV		 1325static inline int may_lookup(struct nameidata *nd)
1326{
1327 if (nd->flags & LOOKUP_RCU) {
1328 int err = exec_permission(nd->inode, IPERM_FLAG_RCU);
1329 if (err != -ECHILD)
1330 return err;
1331 if (nameidata_drop_rcu(nd))
1332 return -ECHILD;
1333 }
1334 return exec_permission(nd->inode, 0);
1335}
1336
1da177e4
LT		 1337/*
1338 * Name resolution.
ea3834d9
PM		 1339 * This is the basic name resolution function, turning a pathname into
1340 * the final dentry. We expect 'base' to be positive and a directory.
1da177e4 1341 *
ea3834d9
PM		 1342 * Returns 0 and nd will have valid dentry and mnt on success.
1343 * Returns error and drops reference to input namei data on failure.
1da177e4 1344 */
6de88d72 1345static int link_path_walk(const char *name, struct nameidata *nd)
1da177e4
LT		 1346{
1347 struct path next;
1da177e4
LT		 1348 int err;
1349 unsigned int lookup_flags = nd->flags;
1350
1351 while (*name=='/')
1352 name++;
1353 if (!*name)
1354 goto return_reval;
1355
1da177e4 1356 if (nd->depth)
f55eab82 1357 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
1da177e4
LT		 1358
1359 /* At this point we know we have a real path component. */
1360 for(;;) {
31e6b01f 1361 struct inode *inode;
1da177e4
LT		 1362 unsigned long hash;
1363 struct qstr this;
1364 unsigned int c;
1365
cdce5d6b 1366 nd->flags |= LOOKUP_CONTINUE;
52094c8a
AV		 1367
1368 err = may_lookup(nd);
1da177e4
LT		 1369 if (err)
1370 break;
1371
1372 this.name = name;
1373 c = *(const unsigned char *)name;
1374
1375 hash = init_name_hash();
1376 do {
1377 name++;
1378 hash = partial_name_hash(c, hash);
1379 c = *(const unsigned char *)name;
1380 } while (c && (c != '/'));
1381 this.len = name - (const char *) this.name;
1382 this.hash = end_name_hash(hash);
1383
1384 /* remove trailing slashes? */
1385 if (!c)
1386 goto last_component;
1387 while (*++name == '/');
1388 if (!*name)
1389 goto last_with_slashes;
1390
1391 /*
1392 * "." and ".." are special - ".." especially so because it has
1393 * to be able to know about the current root directory and
1394 * parent relationships.
1395 */
1396 if (this.name[0] == '.') switch (this.len) {
1397 default:
1398 break;
31e6b01f 1399 case 2:
1da177e4
LT		 1400 if (this.name[1] != '.')
1401 break;
31e6b01f
NP		 1402 if (nd->flags & LOOKUP_RCU) {
1403 if (follow_dotdot_rcu(nd))
1404 return -ECHILD;
1405 } else
1406 follow_dotdot(nd);
1da177e4
LT		 1407 /* fallthrough */
1408 case 1:
1409 continue;
1410 }
1da177e4 1411 /* This does the actual lookups.. */
31e6b01f 1412 err = do_lookup(nd, &this, &next, &inode);
1da177e4
LT		 1413 if (err)
1414 break;
1da177e4 1415 err = -ENOENT;
1da177e4
LT		 1416 if (!inode)
1417 goto out_dput;
1da177e4
LT		 1418
1419 if (inode->i_op->follow_link) {
3abb17e8 1420 err = do_follow_link(inode, &next, nd);
1da177e4
LT		 1421 if (err)
1422 goto return_err;
31e6b01f 1423 nd->inode = nd->path.dentry->d_inode;
1da177e4 1424 err = -ENOENT;
31e6b01f 1425 if (!nd->inode)
1da177e4 1426 break;
31e6b01f 1427 } else {
09dd17d3 1428 path_to_nameidata(&next, nd);
31e6b01f
NP		 1429 nd->inode = inode;
1430 }
1da177e4 1431 err = -ENOTDIR;
31e6b01f 1432 if (!nd->inode->i_op->lookup)
1da177e4
LT		 1433 break;
1434 continue;
1435 /* here ends the main loop */
1436
1437last_with_slashes:
1438 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1439last_component:
f55eab82
TM		 1440 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1441 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
1da177e4
LT		 1442 if (lookup_flags & LOOKUP_PARENT)
1443 goto lookup_parent;
1444 if (this.name[0] == '.') switch (this.len) {
1445 default:
1446 break;
31e6b01f 1447 case 2:
1da177e4
LT		 1448 if (this.name[1] != '.')
1449 break;
31e6b01f
NP		 1450 if (nd->flags & LOOKUP_RCU) {
1451 if (follow_dotdot_rcu(nd))
1452 return -ECHILD;
1453 } else
1454 follow_dotdot(nd);
1da177e4
LT		 1455 /* fallthrough */
1456 case 1:
1457 goto return_reval;
1458 }
31e6b01f 1459 err = do_lookup(nd, &this, &next, &inode);
1da177e4
LT		 1460 if (err)
1461 break;
db372915
DH		 1462 if (inode && unlikely(inode->i_op->follow_link) &&
1463 (lookup_flags & LOOKUP_FOLLOW)) {
3abb17e8 1464 err = do_follow_link(inode, &next, nd);
1da177e4
LT		 1465 if (err)
1466 goto return_err;
31e6b01f
NP		 1467 nd->inode = nd->path.dentry->d_inode;
1468 } else {
09dd17d3 1469 path_to_nameidata(&next, nd);
31e6b01f
NP		 1470 nd->inode = inode;
1471 }
1da177e4 1472 err = -ENOENT;
31e6b01f 1473 if (!nd->inode)
1da177e4
LT		 1474 break;
1475 if (lookup_flags & LOOKUP_DIRECTORY) {
1476 err = -ENOTDIR;
31e6b01f 1477 if (!nd->inode->i_op->lookup)
1da177e4
LT		 1478 break;
1479 }
1480 goto return_base;
1481lookup_parent:
1482 nd->last = this;
1483 nd->last_type = LAST_NORM;
1484 if (this.name[0] != '.')
1485 goto return_base;
1486 if (this.len == 1)
1487 nd->last_type = LAST_DOT;
1488 else if (this.len == 2 && this.name[1] == '.')
1489 nd->last_type = LAST_DOTDOT;
1490 else
1491 goto return_base;
1492return_reval:
1493 /*
1494 * We bypassed the ordinary revalidation routines.
1495 * We may need to check the cached dentry for staleness.
1496 */
fb045adb 1497 if (need_reval_dot(nd->path.dentry)) {
f60aef7e
AV		 1498 if (nameidata_drop_rcu_last_maybe(nd))
1499 return -ECHILD;
1da177e4 1500 /* Note: we do not d_invalidate() */
34286d66
NP		 1501 err = d_revalidate(nd->path.dentry, nd);
1502 if (!err)
1503 err = -ESTALE;
1504 if (err < 0)
1da177e4 1505 break;
f60aef7e 1506 return 0;
1da177e4
LT		 1507 }
1508return_base:
31e6b01f
NP		 1509 if (nameidata_drop_rcu_last_maybe(nd))
1510 return -ECHILD;
1da177e4
LT		 1511 return 0;
1512out_dput:
31e6b01f
NP		 1513 if (!(nd->flags & LOOKUP_RCU))
1514 path_put_conditional(&next, nd);
1da177e4
LT		 1515 break;
1516 }
31e6b01f
NP		 1517 if (!(nd->flags & LOOKUP_RCU))
1518 path_put(&nd->path);
1da177e4
LT		 1519return_err:
1520 return err;
1521}
1522
31e6b01f
NP		 1523static inline int path_walk_rcu(const char *name, struct nameidata *nd)
1524{
1525 current->total_link_count = 0;
1526
1527 return link_path_walk(name, nd);
1528}
1529
1530static inline int path_walk_simple(const char *name, struct nameidata *nd)
1531{
1532 current->total_link_count = 0;
1533
1534 return link_path_walk(name, nd);
1535}
1536
fc9b52cd 1537static int path_walk(const char *name, struct nameidata *nd)
1da177e4 1538{
6de88d72
AV		 1539 struct path save = nd->path;
1540 int result;
1541
1da177e4 1542 current->total_link_count = 0;
6de88d72
AV		 1543
1544 /* make sure the stuff we saved doesn't go away */
1545 path_get(&save);
1546
1547 result = link_path_walk(name, nd);
1548 if (result == -ESTALE) {
1549 /* nd->path had been dropped */
1550 current->total_link_count = 0;
1551 nd->path = save;
b306419a 1552 nd->inode = save.dentry->d_inode;
6de88d72
AV		 1553 path_get(&nd->path);
1554 nd->flags |= LOOKUP_REVAL;
1555 result = link_path_walk(name, nd);
1556 }
1557
1558 path_put(&save);
1559
1560 return result;
1da177e4
LT		 1561}
1562
31e6b01f
NP		 1563static void path_finish_rcu(struct nameidata *nd)
1564{
1565 if (nd->flags & LOOKUP_RCU) {
1566 /* RCU dangling. Cancel it. */
1567 nd->flags &= ~LOOKUP_RCU;
1568 nd->root.mnt = NULL;
1569 rcu_read_unlock();
1570 br_read_unlock(vfsmount_lock);
1571 }
1572 if (nd->file)
1573 fput(nd->file);
1574}
1575
1576static int path_init_rcu(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1577{
1578 int retval = 0;
1579 int fput_needed;
1580 struct file *file;
1581
1582 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1583 nd->flags = flags | LOOKUP_RCU;
1584 nd->depth = 0;
1585 nd->root.mnt = NULL;
1586 nd->file = NULL;
1587
1588 if (*name=='/') {
1589 struct fs_struct *fs = current->fs;
c28cc364 1590 unsigned seq;
31e6b01f
NP		 1591
1592 br_read_lock(vfsmount_lock);
1593 rcu_read_lock();
1594
c28cc364
NP		 1595 do {
1596 seq = read_seqcount_begin(&fs->seq);
1597 nd->root = fs->root;
1598 nd->path = nd->root;
1599 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1600 } while (read_seqcount_retry(&fs->seq, seq));
31e6b01f
NP		 1601
1602 } else if (dfd == AT_FDCWD) {
1603 struct fs_struct *fs = current->fs;
c28cc364 1604 unsigned seq;
31e6b01f
NP		 1605
1606 br_read_lock(vfsmount_lock);
1607 rcu_read_lock();
1608
c28cc364
NP		 1609 do {
1610 seq = read_seqcount_begin(&fs->seq);
1611 nd->path = fs->pwd;
1612 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1613 } while (read_seqcount_retry(&fs->seq, seq));
1614
31e6b01f
NP		 1615 } else {
1616 struct dentry *dentry;
1617
1618 file = fget_light(dfd, &fput_needed);
1619 retval = -EBADF;
1620 if (!file)
1621 goto out_fail;
1622
1623 dentry = file->f_path.dentry;
1624
1625 retval = -ENOTDIR;
1626 if (!S_ISDIR(dentry->d_inode->i_mode))
1627 goto fput_fail;
1628
1629 retval = file_permission(file, MAY_EXEC);
1630 if (retval)
1631 goto fput_fail;
1632
1633 nd->path = file->f_path;
1634 if (fput_needed)
1635 nd->file = file;
1636
c28cc364 1637 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
31e6b01f
NP		 1638 br_read_lock(vfsmount_lock);
1639 rcu_read_lock();
1640 }
1641 nd->inode = nd->path.dentry->d_inode;
1642 return 0;
1643
1644fput_fail:
1645 fput_light(file, fput_needed);
1646out_fail:
1647 return retval;
1648}
1649
9b4a9b14 1650static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1da177e4 1651{
ea3834d9 1652 int retval = 0;
170aa3d0
UD		 1653 int fput_needed;
1654 struct file *file;
1da177e4
LT		 1655
1656 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1657 nd->flags = flags;
1658 nd->depth = 0;
2a737871 1659 nd->root.mnt = NULL;
1da177e4 1660
1da177e4 1661 if (*name=='/') {
2a737871
AV		 1662 set_root(nd);
1663 nd->path = nd->root;
1664 path_get(&nd->root);
5590ff0d 1665 } else if (dfd == AT_FDCWD) {
f7ad3c6b 1666 get_fs_pwd(current->fs, &nd->path);
5590ff0d 1667 } else {
5590ff0d
UD		 1668 struct dentry *dentry;
1669
1670 file = fget_light(dfd, &fput_needed);
170aa3d0
UD		 1671 retval = -EBADF;
1672 if (!file)
6d09bb62 1673 goto out_fail;
5590ff0d 1674
0f7fc9e4 1675 dentry = file->f_path.dentry;
5590ff0d 1676
170aa3d0
UD		 1677 retval = -ENOTDIR;
1678 if (!S_ISDIR(dentry->d_inode->i_mode))
6d09bb62 1679 goto fput_fail;
5590ff0d
UD		 1680
1681 retval = file_permission(file, MAY_EXEC);
170aa3d0 1682 if (retval)
6d09bb62 1683 goto fput_fail;
5590ff0d 1684
5dd784d0
JB		 1685 nd->path = file->f_path;
1686 path_get(&file->f_path);
5590ff0d
UD		 1687
1688 fput_light(file, fput_needed);
1da177e4 1689 }
31e6b01f 1690 nd->inode = nd->path.dentry->d_inode;
9b4a9b14 1691 return 0;
2dfdd266 1692
9b4a9b14
AV		 1693fput_fail:
1694 fput_light(file, fput_needed);
1695out_fail:
1696 return retval;
1697}
1698
1699/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1700static int do_path_lookup(int dfd, const char *name,
1701 unsigned int flags, struct nameidata *nd)
1702{
31e6b01f
NP		 1703 int retval;
1704
1705 /*
1706 * Path walking is largely split up into 2 different synchronisation
1707 * schemes, rcu-walk and ref-walk (explained in
1708 * Documentation/filesystems/path-lookup.txt). These share much of the
1709 * path walk code, but some things particularly setup, cleanup, and
1710 * following mounts are sufficiently divergent that functions are
1711 * duplicated. Typically there is a function foo(), and its RCU
1712 * analogue, foo_rcu().
1713 *
1714 * -ECHILD is the error number of choice (just to avoid clashes) that
1715 * is returned if some aspect of an rcu-walk fails. Such an error must
1716 * be handled by restarting a traditional ref-walk (which will always
1717 * be able to complete).
1718 */
1719 retval = path_init_rcu(dfd, name, flags, nd);
1720 if (unlikely(retval))
1721 return retval;
1722 retval = path_walk_rcu(name, nd);
1723 path_finish_rcu(nd);
2a737871
AV		 1724 if (nd->root.mnt) {
1725 path_put(&nd->root);
1726 nd->root.mnt = NULL;
1727 }
31e6b01f
NP		 1728
1729 if (unlikely(retval == -ECHILD || retval == -ESTALE)) {
1730 /* slower, locked walk */
1731 if (retval == -ESTALE)
1732 flags |= LOOKUP_REVAL;
1733 retval = path_init(dfd, name, flags, nd);
1734 if (unlikely(retval))
1735 return retval;
1736 retval = path_walk(name, nd);
1737 if (nd->root.mnt) {
1738 path_put(&nd->root);
1739 nd->root.mnt = NULL;
1740 }
1741 }
1742
1743 if (likely(!retval)) {
1744 if (unlikely(!audit_dummy_context())) {
1745 if (nd->path.dentry && nd->inode)
1746 audit_inode(name, nd->path.dentry);
1747 }
1748 }
1749
170aa3d0 1750 return retval;
1da177e4
LT		 1751}
1752
c9c6cac0 1753int kern_path_parent(const char *name, struct nameidata *nd)
5590ff0d 1754{
c9c6cac0 1755 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
5590ff0d
UD		 1756}
1757
d1811465
AV		 1758int kern_path(const char *name, unsigned int flags, struct path *path)
1759{
1760 struct nameidata nd;
1761 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1762 if (!res)
1763 *path = nd.path;
1764 return res;
1765}
1766
16f18200
JJS		 1767/**
1768 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1769 * @dentry: pointer to dentry of the base directory
1770 * @mnt: pointer to vfs mount of the base directory
1771 * @name: pointer to file name
1772 * @flags: lookup flags
1773 * @nd: pointer to nameidata
1774 */
1775int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1776 const char *name, unsigned int flags,
1777 struct nameidata *nd)
1778{
1779 int retval;
1780
1781 /* same as do_path_lookup */
1782 nd->last_type = LAST_ROOT;
1783 nd->flags = flags;
1784 nd->depth = 0;
1785
c8e7f449
JB		 1786 nd->path.dentry = dentry;
1787 nd->path.mnt = mnt;
1788 path_get(&nd->path);
5b857119
AV		 1789 nd->root = nd->path;
1790 path_get(&nd->root);
31e6b01f 1791 nd->inode = nd->path.dentry->d_inode;
16f18200
JJS		 1792
1793 retval = path_walk(name, nd);
4ac91378 1794 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
31e6b01f 1795 nd->inode))
4ac91378 1796 audit_inode(name, nd->path.dentry);
16f18200 1797
5b857119
AV		 1798 path_put(&nd->root);
1799 nd->root.mnt = NULL;
16f18200 1800
2a737871 1801 return retval;
16f18200
JJS		 1802}
1803
eead1911
CH		 1804static struct dentry *__lookup_hash(struct qstr *name,
1805 struct dentry *base, struct nameidata *nd)
1da177e4 1806{
81fca444 1807 struct inode *inode = base->d_inode;
057f6c01 1808 struct dentry *dentry;
1da177e4
LT		 1809 int err;
1810
b74c79e9 1811 err = exec_permission(inode, 0);
81fca444
CH		 1812 if (err)
1813 return ERR_PTR(err);
1da177e4
LT		 1814
1815 /*
1816 * See if the low-level filesystem might want
1817 * to use its own hash..
1818 */
fb045adb 1819 if (base->d_flags & DCACHE_OP_HASH) {
b1e6a015 1820 err = base->d_op->d_hash(base, inode, name);
1da177e4
LT		 1821 dentry = ERR_PTR(err);
1822 if (err < 0)
1823 goto out;
1824 }
1825
b04f784e
NP		 1826 /*
1827 * Don't bother with __d_lookup: callers are for creat as
1828 * well as unlink, so a lot of the time it would cost
1829 * a double lookup.
6e6b1bd1 1830 */
b04f784e 1831 dentry = d_lookup(base, name);
6e6b1bd1 1832
fb045adb 1833 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE))
6e6b1bd1
AV		 1834 dentry = do_revalidate(dentry, nd);
1835
baa03890
NP		 1836 if (!dentry)
1837 dentry = d_alloc_and_lookup(base, name, nd);
1da177e4
LT		 1838out:
1839 return dentry;
1840}
1841
057f6c01
JM		 1842/*
1843 * Restricted form of lookup. Doesn't follow links, single-component only,
1844 * needs parent already locked. Doesn't follow mounts.
1845 * SMP-safe.
1846 */
eead1911 1847static struct dentry *lookup_hash(struct nameidata *nd)
057f6c01 1848{
4ac91378 1849 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1da177e4
LT		 1850}
1851
eead1911
CH		 1852static int __lookup_one_len(const char *name, struct qstr *this,
1853 struct dentry *base, int len)
1da177e4
LT		 1854{
1855 unsigned long hash;
1da177e4
LT		 1856 unsigned int c;
1857
057f6c01
JM		 1858 this->name = name;
1859 this->len = len;
1da177e4 1860 if (!len)
057f6c01 1861 return -EACCES;
1da177e4
LT		 1862
1863 hash = init_name_hash();
1864 while (len--) {
1865 c = *(const unsigned char *)name++;
1866 if (c == '/' || c == '\0')
057f6c01 1867 return -EACCES;
1da177e4
LT		 1868 hash = partial_name_hash(c, hash);
1869 }
057f6c01
JM		 1870 this->hash = end_name_hash(hash);
1871 return 0;
1872}
1da177e4 1873
eead1911 1874/**
a6b91919 1875 * lookup_one_len - filesystem helper to lookup single pathname component
eead1911
CH		 1876 * @name: pathname component to lookup
1877 * @base: base directory to lookup from
1878 * @len: maximum length @len should be interpreted to
1879 *
a6b91919
RD		 1880 * Note that this routine is purely a helper for filesystem usage and should
1881 * not be called by generic code. Also note that by using this function the
eead1911
CH		 1882 * nameidata argument is passed to the filesystem methods and a filesystem
1883 * using this helper needs to be prepared for that.
1884 */
057f6c01
JM		 1885struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1886{
1887 int err;
1888 struct qstr this;
1889
2f9092e1
DW		 1890 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1891
057f6c01 1892 err = __lookup_one_len(name, &this, base, len);
eead1911
CH		 1893 if (err)
1894 return ERR_PTR(err);
1895
49705b77 1896 return __lookup_hash(&this, base, NULL);
057f6c01
JM		 1897}
1898
2d8f3038
AV		 1899int user_path_at(int dfd, const char __user *name, unsigned flags,
1900 struct path *path)
1da177e4 1901{
2d8f3038 1902 struct nameidata nd;
1da177e4
LT		 1903 char *tmp = getname(name);
1904 int err = PTR_ERR(tmp);
1da177e4 1905 if (!IS_ERR(tmp)) {
2d8f3038
AV		 1906
1907 BUG_ON(flags & LOOKUP_PARENT);
1908
1909 err = do_path_lookup(dfd, tmp, flags, &nd);
1da177e4 1910 putname(tmp);
2d8f3038
AV		 1911 if (!err)
1912 *path = nd.path;
1da177e4
LT		 1913 }
1914 return err;
1915}
1916
2ad94ae6
AV		 1917static int user_path_parent(int dfd, const char __user *path,
1918 struct nameidata *nd, char **name)
1919{
1920 char *s = getname(path);
1921 int error;
1922
1923 if (IS_ERR(s))
1924 return PTR_ERR(s);
1925
1926 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1927 if (error)
1928 putname(s);
1929 else
1930 *name = s;
1931
1932 return error;
1933}
1934
1da177e4
LT		 1935/*
1936 * It's inline, so penalty for filesystems that don't use sticky bit is
1937 * minimal.
1938 */
1939static inline int check_sticky(struct inode *dir, struct inode *inode)
1940{
da9592ed
DH		 1941 uid_t fsuid = current_fsuid();
1942
1da177e4
LT		 1943 if (!(dir->i_mode & S_ISVTX))
1944 return 0;
da9592ed 1945 if (inode->i_uid == fsuid)
1da177e4 1946 return 0;
da9592ed 1947 if (dir->i_uid == fsuid)
1da177e4
LT		 1948 return 0;
1949 return !capable(CAP_FOWNER);
1950}
1951
1952/*
1953 * Check whether we can remove a link victim from directory dir, check
1954 * whether the type of victim is right.
1955 * 1. We can't do it if dir is read-only (done in permission())
1956 * 2. We should have write and exec permissions on dir
1957 * 3. We can't remove anything from append-only dir
1958 * 4. We can't do anything with immutable dir (done in permission())
1959 * 5. If the sticky bit on dir is set we should either
1960 * a. be owner of dir, or
1961 * b. be owner of victim, or
1962 * c. have CAP_FOWNER capability
1963 * 6. If the victim is append-only or immutable we can't do antyhing with
1964 * links pointing to it.
1965 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1966 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1967 * 9. We can't remove a root or mountpoint.
1968 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1969 * nfs_async_unlink().
1970 */
858119e1 1971static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1da177e4
LT		 1972{
1973 int error;
1974
1975 if (!victim->d_inode)
1976 return -ENOENT;
1977
1978 BUG_ON(victim->d_parent->d_inode != dir);
cccc6bba 1979 audit_inode_child(victim, dir);
1da177e4 1980
f419a2e3 1981 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1982 if (error)
1983 return error;
1984 if (IS_APPEND(dir))
1985 return -EPERM;
1986 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
f9454548 1987 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1da177e4
LT		 1988 return -EPERM;
1989 if (isdir) {
1990 if (!S_ISDIR(victim->d_inode->i_mode))
1991 return -ENOTDIR;
1992 if (IS_ROOT(victim))
1993 return -EBUSY;
1994 } else if (S_ISDIR(victim->d_inode->i_mode))
1995 return -EISDIR;
1996 if (IS_DEADDIR(dir))
1997 return -ENOENT;
1998 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1999 return -EBUSY;
2000 return 0;
2001}
2002
2003/* Check whether we can create an object with dentry child in directory
2004 * dir.
2005 * 1. We can't do it if child already exists (open has special treatment for
2006 * this case, but since we are inlined it's OK)
2007 * 2. We can't do it if dir is read-only (done in permission())
2008 * 3. We should have write and exec permissions on dir
2009 * 4. We can't do it if dir is immutable (done in permission())
2010 */
a95164d9 2011static inline int may_create(struct inode *dir, struct dentry *child)
1da177e4
LT		 2012{
2013 if (child->d_inode)
2014 return -EEXIST;
2015 if (IS_DEADDIR(dir))
2016 return -ENOENT;
f419a2e3 2017 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 2018}
2019
1da177e4
LT		 2020/*
2021 * p1 and p2 should be directories on the same fs.
2022 */
2023struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2024{
2025 struct dentry *p;
2026
2027 if (p1 == p2) {
f2eace23 2028 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1da177e4
LT		 2029 return NULL;
2030 }
2031
a11f3a05 2032 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4 2033
e2761a11
OH		 2034 p = d_ancestor(p2, p1);
2035 if (p) {
2036 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2037 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2038 return p;
1da177e4
LT		 2039 }
2040
e2761a11
OH		 2041 p = d_ancestor(p1, p2);
2042 if (p) {
2043 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2044 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2045 return p;
1da177e4
LT		 2046 }
2047
f2eace23
IM		 2048 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2049 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1da177e4
LT		 2050 return NULL;
2051}
2052
2053void unlock_rename(struct dentry *p1, struct dentry *p2)
2054{
1b1dcc1b 2055 mutex_unlock(&p1->d_inode->i_mutex);
1da177e4 2056 if (p1 != p2) {
1b1dcc1b 2057 mutex_unlock(&p2->d_inode->i_mutex);
a11f3a05 2058 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4
LT		 2059 }
2060}
2061
2062int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
2063 struct nameidata *nd)
2064{
a95164d9 2065 int error = may_create(dir, dentry);
1da177e4
LT		 2066
2067 if (error)
2068 return error;
2069
acfa4380 2070 if (!dir->i_op->create)
1da177e4
LT		 2071 return -EACCES; /* shouldn't it be ENOSYS? */
2072 mode &= S_IALLUGO;
2073 mode |= S_IFREG;
2074 error = security_inode_create(dir, dentry, mode);
2075 if (error)
2076 return error;
1da177e4 2077 error = dir->i_op->create(dir, dentry, mode, nd);
a74574aa 2078 if (!error)
f38aa942 2079 fsnotify_create(dir, dentry);
1da177e4
LT		 2080 return error;
2081}
2082
3fb64190 2083int may_open(struct path *path, int acc_mode, int flag)
1da177e4 2084{
3fb64190 2085 struct dentry *dentry = path->dentry;
1da177e4
LT		 2086 struct inode *inode = dentry->d_inode;
2087 int error;
2088
2089 if (!inode)
2090 return -ENOENT;
2091
c8fe8f30
CH		 2092 switch (inode->i_mode & S_IFMT) {
2093 case S_IFLNK:
1da177e4 2094 return -ELOOP;
c8fe8f30
CH		 2095 case S_IFDIR:
2096 if (acc_mode & MAY_WRITE)
2097 return -EISDIR;
2098 break;
2099 case S_IFBLK:
2100 case S_IFCHR:
3fb64190 2101 if (path->mnt->mnt_flags & MNT_NODEV)
1da177e4 2102 return -EACCES;
c8fe8f30
CH		 2103 /*FALLTHRU*/
2104 case S_IFIFO:
2105 case S_IFSOCK:
1da177e4 2106 flag &= ~O_TRUNC;
c8fe8f30 2107 break;
4a3fd211 2108 }
b41572e9 2109
3fb64190 2110 error = inode_permission(inode, acc_mode);
b41572e9
DH		 2111 if (error)
2112 return error;
6146f0d5 2113
1da177e4
LT		 2114 /*
2115 * An append-only file must be opened in append mode for writing.
2116 */
2117 if (IS_APPEND(inode)) {
8737c930 2118 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
7715b521 2119 return -EPERM;
1da177e4 2120 if (flag & O_TRUNC)
7715b521 2121 return -EPERM;
1da177e4
LT		 2122 }
2123
2124 /* O_NOATIME can only be set by the owner or superuser */
7715b521
AV		 2125 if (flag & O_NOATIME && !is_owner_or_cap(inode))
2126 return -EPERM;
1da177e4
LT		 2127
2128 /*
2129 * Ensure there are no outstanding leases on the file.
2130 */
b65a9cfc 2131 return break_lease(inode, flag);
7715b521 2132}
1da177e4 2133
e1181ee6 2134static int handle_truncate(struct file *filp)
7715b521 2135{
e1181ee6 2136 struct path *path = &filp->f_path;
7715b521
AV		 2137 struct inode *inode = path->dentry->d_inode;
2138 int error = get_write_access(inode);
2139 if (error)
2140 return error;
2141 /*
2142 * Refuse to truncate files with mandatory locks held on them.
2143 */
2144 error = locks_verify_locked(inode);
2145 if (!error)
ea0d3ab2 2146 error = security_path_truncate(path);
7715b521
AV		 2147 if (!error) {
2148 error = do_truncate(path->dentry, 0,
2149 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
e1181ee6 2150 filp);
7715b521
AV		 2151 }
2152 put_write_access(inode);
acd0c935 2153 return error;
1da177e4
LT		 2154}
2155
d57999e1
DH		 2156/*
2157 * Be careful about ever adding any more callers of this
2158 * function. Its flags must be in the namei format, not
2159 * what get passed to sys_open().
2160 */
2161static int __open_namei_create(struct nameidata *nd, struct path *path,
8737c930 2162 int open_flag, int mode)
aab520e2
DH		 2163{
2164 int error;
4ac91378 2165 struct dentry *dir = nd->path.dentry;
aab520e2
DH		 2166
2167 if (!IS_POSIXACL(dir->d_inode))
ce3b0f8d 2168 mode &= ~current_umask();
be6d3e56
KT		 2169 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
2170 if (error)
2171 goto out_unlock;
aab520e2 2172 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
be6d3e56 2173out_unlock:
aab520e2 2174 mutex_unlock(&dir->d_inode->i_mutex);
4ac91378
JB		 2175 dput(nd->path.dentry);
2176 nd->path.dentry = path->dentry;
31e6b01f 2177
aab520e2
DH		 2178 if (error)
2179 return error;
2180 /* Don't check for write permission, don't truncate */
8737c930 2181 return may_open(&nd->path, 0, open_flag & ~O_TRUNC);
aab520e2
DH		 2182}
2183
d57999e1
DH		 2184/*
2185 * Note that while the flag value (low two bits) for sys_open means:
2186 * 00 - read-only
2187 * 01 - write-only
2188 * 10 - read-write
2189 * 11 - special
2190 * it is changed into
2191 * 00 - no permissions needed
2192 * 01 - read-permission
2193 * 10 - write-permission
2194 * 11 - read-write
2195 * for the internal routines (ie open_namei()/follow_link() etc)
2196 * This is more logical, and also allows the 00 "no perm needed"
2197 * to be used for symlinks (where the permissions are checked
2198 * later).
2199 *
2200*/
2201static inline int open_to_namei_flags(int flag)
2202{
2203 if ((flag+1) & O_ACCMODE)
2204 flag++;
2205 return flag;
2206}
2207
7715b521 2208static int open_will_truncate(int flag, struct inode *inode)
4a3fd211
DH		 2209{
2210 /*
2211 * We'll never write to the fs underlying
2212 * a device file.
2213 */
2214 if (special_file(inode->i_mode))
2215 return 0;
2216 return (flag & O_TRUNC);
2217}
2218
648fa861 2219static struct file *finish_open(struct nameidata *nd,
9a66179e 2220 int open_flag, int acc_mode)
648fa861
AV		 2221{
2222 struct file *filp;
2223 int will_truncate;
2224 int error;
2225
9a66179e 2226 will_truncate = open_will_truncate(open_flag, nd->path.dentry->d_inode);
648fa861
AV		 2227 if (will_truncate) {
2228 error = mnt_want_write(nd->path.mnt);
2229 if (error)
2230 goto exit;
2231 }
2232 error = may_open(&nd->path, acc_mode, open_flag);
2233 if (error) {
2234 if (will_truncate)
2235 mnt_drop_write(nd->path.mnt);
2236 goto exit;
2237 }
2238 filp = nameidata_to_filp(nd);
2239 if (!IS_ERR(filp)) {
2240 error = ima_file_check(filp, acc_mode);
2241 if (error) {
2242 fput(filp);
2243 filp = ERR_PTR(error);
2244 }
2245 }
2246 if (!IS_ERR(filp)) {
648fa861 2247 if (will_truncate) {
e1181ee6 2248 error = handle_truncate(filp);
648fa861
AV		 2249 if (error) {
2250 fput(filp);
2251 filp = ERR_PTR(error);
2252 }
2253 }
2254 }
2255 /*
2256 * It is now safe to drop the mnt write
2257 * because the filp has had a write taken
2258 * on its behalf.
2259 */
2260 if (will_truncate)
2261 mnt_drop_write(nd->path.mnt);
d893f1bc 2262 path_put(&nd->path);
648fa861
AV		 2263 return filp;
2264
2265exit:
648fa861
AV		 2266 path_put(&nd->path);
2267 return ERR_PTR(error);
2268}
2269
31e6b01f
NP		 2270/*
2271 * Handle O_CREAT case for do_filp_open
2272 */
fb1cc555 2273static struct file *do_last(struct nameidata *nd, struct path *path,
5b369df8 2274 int open_flag, int acc_mode,
3e297b61 2275 int mode, const char *pathname)
fb1cc555 2276{
a1e28038 2277 struct dentry *dir = nd->path.dentry;
fb1cc555 2278 struct file *filp;
1f36f774
AV		 2279 int error = -EISDIR;
2280
2281 switch (nd->last_type) {
2282 case LAST_DOTDOT:
2283 follow_dotdot(nd);
2284 dir = nd->path.dentry;
176306f5 2285 case LAST_DOT:
fb045adb 2286 if (need_reval_dot(dir)) {
f20877d9
O		 2287 int status = d_revalidate(nd->path.dentry, nd);
2288 if (!status)
2289 status = -ESTALE;
2290 if (status < 0) {
2291 error = status;
1f36f774 2292 goto exit;
f20877d9 2293 }
1f36f774
AV		 2294 }
2295 /* fallthrough */
1f36f774 2296 case LAST_ROOT:
31e6b01f 2297 goto exit;
1f36f774
AV		 2298 case LAST_BIND:
2299 audit_inode(pathname, dir);
67ee3ad2 2300 goto ok;
1f36f774 2301 }
67ee3ad2 2302
1f36f774 2303 /* trailing slashes? */
31e6b01f
NP		 2304 if (nd->last.name[nd->last.len])
2305 goto exit;
a2c36b45 2306
a1e28038
AV		 2307 mutex_lock(&dir->d_inode->i_mutex);
2308
2309 path->dentry = lookup_hash(nd);
2310 path->mnt = nd->path.mnt;
2311
fb1cc555
AV		 2312 error = PTR_ERR(path->dentry);
2313 if (IS_ERR(path->dentry)) {
2314 mutex_unlock(&dir->d_inode->i_mutex);
2315 goto exit;
2316 }
2317
2318 if (IS_ERR(nd->intent.open.file)) {
2319 error = PTR_ERR(nd->intent.open.file);
2320 goto exit_mutex_unlock;
2321 }
2322
2323 /* Negative dentry, just create the file */
2324 if (!path->dentry->d_inode) {
2325 /*
2326 * This write is needed to ensure that a
2327 * ro->rw transition does not occur between
2328 * the time when the file is created and when
2329 * a permanent write count is taken through
2330 * the 'struct file' in nameidata_to_filp().
2331 */
2332 error = mnt_want_write(nd->path.mnt);
2333 if (error)
2334 goto exit_mutex_unlock;
2335 error = __open_namei_create(nd, path, open_flag, mode);
2336 if (error) {
2337 mnt_drop_write(nd->path.mnt);
2338 goto exit;
2339 }
2340 filp = nameidata_to_filp(nd);
2341 mnt_drop_write(nd->path.mnt);
d893f1bc 2342 path_put(&nd->path);
fb1cc555
AV		 2343 if (!IS_ERR(filp)) {
2344 error = ima_file_check(filp, acc_mode);
2345 if (error) {
2346 fput(filp);
2347 filp = ERR_PTR(error);
2348 }
2349 }
2350 return filp;
2351 }
2352
2353 /*
2354 * It already exists.
2355 */
2356 mutex_unlock(&dir->d_inode->i_mutex);
2357 audit_inode(pathname, path->dentry);
2358
2359 error = -EEXIST;
5b369df8 2360 if (open_flag & O_EXCL)
fb1cc555
AV		 2361 goto exit_dput;
2362
9875cf80
DH		 2363 error = follow_managed(path, nd->flags);
2364 if (error < 0)
2365 goto exit_dput;
fb1cc555
AV		 2366
2367 error = -ENOENT;
2368 if (!path->dentry->d_inode)
2369 goto exit_dput;
9e67f361
AV		 2370
2371 if (path->dentry->d_inode->i_op->follow_link)
fb1cc555 2372 return NULL;
fb1cc555
AV		 2373
2374 path_to_nameidata(path, nd);
31e6b01f 2375 nd->inode = path->dentry->d_inode;
fb1cc555 2376 error = -EISDIR;
31e6b01f 2377 if (S_ISDIR(nd->inode->i_mode))
fb1cc555 2378 goto exit;
67ee3ad2 2379ok:
9a66179e 2380 filp = finish_open(nd, open_flag, acc_mode);
fb1cc555
AV		 2381 return filp;
2382
2383exit_mutex_unlock:
2384 mutex_unlock(&dir->d_inode->i_mutex);
2385exit_dput:
2386 path_put_conditional(path, nd);
2387exit:
fb1cc555
AV		 2388 path_put(&nd->path);
2389 return ERR_PTR(error);
2390}
2391
1da177e4 2392/*
4a3fd211
DH		 2393 * Note that the low bits of the passed in "open_flag"
2394 * are not the same as in the local variable "flag". See
2395 * open_to_namei_flags() for more details.
1da177e4 2396 */
a70e65df 2397struct file *do_filp_open(int dfd, const char *pathname,
6e8341a1 2398 int open_flag, int mode, int acc_mode)
1da177e4 2399{
4a3fd211 2400 struct file *filp;
a70e65df 2401 struct nameidata nd;
6e8341a1 2402 int error;
9850c056 2403 struct path path;
1da177e4 2404 int count = 0;
d57999e1 2405 int flag = open_to_namei_flags(open_flag);
31e6b01f 2406 int flags;
1f36f774
AV		 2407
2408 if (!(open_flag & O_CREAT))
2409 mode = 0;
1da177e4 2410
b1085ba8
LS		 2411 /* Must never be set by userspace */
2412 open_flag &= ~FMODE_NONOTIFY;
2413
6b2f3d1f
CH		 2414 /*
2415 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
2416 * check for O_DSYNC if the need any syncing at all we enforce it's
2417 * always set instead of having to deal with possibly weird behaviour
2418 * for malicious applications setting only __O_SYNC.
2419 */
2420 if (open_flag & __O_SYNC)
2421 open_flag |= O_DSYNC;
2422
6e8341a1 2423 if (!acc_mode)
6d125529 2424 acc_mode = MAY_OPEN | ACC_MODE(open_flag);
1da177e4 2425
834f2a4a 2426 /* O_TRUNC implies we need access checks for write permissions */
4296e2cb 2427 if (open_flag & O_TRUNC)
834f2a4a
TM		 2428 acc_mode |= MAY_WRITE;
2429
1da177e4
LT		 2430 /* Allow the LSM permission hook to distinguish append
2431 access from general write access. */
4296e2cb 2432 if (open_flag & O_APPEND)
1da177e4
LT		 2433 acc_mode |= MAY_APPEND;
2434
31e6b01f
NP		 2435 flags = LOOKUP_OPEN;
2436 if (open_flag & O_CREAT) {
2437 flags |= LOOKUP_CREATE;
2438 if (open_flag & O_EXCL)
2439 flags |= LOOKUP_EXCL;
2440 }
2441 if (open_flag & O_DIRECTORY)
2442 flags |= LOOKUP_DIRECTORY;
2443 if (!(open_flag & O_NOFOLLOW))
2444 flags |= LOOKUP_FOLLOW;
2445
2446 filp = get_empty_filp();
2447 if (!filp)
2448 return ERR_PTR(-ENFILE);
2449
2450 filp->f_flags = open_flag;
2451 nd.intent.open.file = filp;
2452 nd.intent.open.flags = flag;
2453 nd.intent.open.create_mode = mode;
2454
2455 if (open_flag & O_CREAT)
2456 goto creat;
2457
2458 /* !O_CREAT, simple open */
2459 error = do_path_lookup(dfd, pathname, flags, &nd);
2460 if (unlikely(error))
1858efd4 2461 goto out_filp2;
31e6b01f
NP		 2462 error = -ELOOP;
2463 if (!(nd.flags & LOOKUP_FOLLOW)) {
2464 if (nd.inode->i_op->follow_link)
1858efd4 2465 goto out_path2;
31e6b01f
NP		 2466 }
2467 error = -ENOTDIR;
2468 if (nd.flags & LOOKUP_DIRECTORY) {
2469 if (!nd.inode->i_op->lookup)
1858efd4 2470 goto out_path2;
31e6b01f
NP		 2471 }
2472 audit_inode(pathname, nd.path.dentry);
2473 filp = finish_open(&nd, open_flag, acc_mode);
1858efd4 2474out2:
2dab5974 2475 release_open_intent(&nd);
31e6b01f
NP		 2476 return filp;
2477
1858efd4
AV		 2478out_path2:
2479 path_put(&nd.path);
2480out_filp2:
2481 filp = ERR_PTR(error);
2482 goto out2;
2483
31e6b01f
NP		 2484creat:
2485 /* OK, have to create the file. Find the parent. */
2486 error = path_init_rcu(dfd, pathname,
2487 LOOKUP_PARENT | (flags & LOOKUP_REVAL), &nd);
1da177e4 2488 if (error)
31e6b01f
NP		 2489 goto out_filp;
2490 error = path_walk_rcu(pathname, &nd);
2491 path_finish_rcu(&nd);
2492 if (unlikely(error == -ECHILD || error == -ESTALE)) {
2493 /* slower, locked walk */
2494 if (error == -ESTALE) {
2495reval:
2496 flags |= LOOKUP_REVAL;
2497 }
2498 error = path_init(dfd, pathname,
2499 LOOKUP_PARENT | (flags & LOOKUP_REVAL), &nd);
2500 if (error)
2501 goto out_filp;
3866248e 2502
31e6b01f 2503 error = path_walk_simple(pathname, &nd);
654f562c 2504 }
31e6b01f
NP		 2505 if (unlikely(error))
2506 goto out_filp;
2507 if (unlikely(!audit_dummy_context()))
9b4a9b14 2508 audit_inode(pathname, nd.path.dentry);
1da177e4
LT		 2509
2510 /*
a2c36b45 2511 * We have the parent and last component.
1da177e4 2512 */
31e6b01f 2513 nd.flags = flags;
3e297b61 2514 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
806b681c 2515 while (unlikely(!filp)) { /* trailing symlink */
7b9337aa
NP		 2516 struct path link = path;
2517 struct inode *linki = link.dentry->d_inode;
def4af30 2518 void *cookie;
806b681c 2519 error = -ELOOP;
db372915 2520 if (!(nd.flags & LOOKUP_FOLLOW))
1f36f774
AV		 2521 goto exit_dput;
2522 if (count++ == 32)
806b681c
AV		 2523 goto exit_dput;
2524 /*
2525 * This is subtle. Instead of calling do_follow_link() we do
2526 * the thing by hands. The reason is that this way we have zero
2527 * link_count and path_walk() (called from ->follow_link)
2528 * honoring LOOKUP_PARENT. After that we have the parent and
2529 * last component, i.e. we are in the same situation as after
2530 * the first path_walk(). Well, almost - if the last component
2531 * is normal we get its copy stored in nd->last.name and we will
2532 * have to putname() it when we are done. Procfs-like symlinks
2533 * just set LAST_BIND.
2534 */
2535 nd.flags |= LOOKUP_PARENT;
7b9337aa 2536 error = security_inode_follow_link(link.dentry, &nd);
806b681c
AV		 2537 if (error)
2538 goto exit_dput;
7b9337aa 2539 error = __do_follow_link(&link, &nd, &cookie);
def4af30 2540 if (unlikely(error)) {
7b9337aa
NP		 2541 if (!IS_ERR(cookie) && linki->i_op->put_link)
2542 linki->i_op->put_link(link.dentry, &nd, cookie);
806b681c 2543 /* nd.path had been dropped */
7b9337aa 2544 nd.path = link;
31e6b01f 2545 goto out_path;
806b681c
AV		 2546 }
2547 nd.flags &= ~LOOKUP_PARENT;
3e297b61 2548 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
7b9337aa
NP		 2549 if (linki->i_op->put_link)
2550 linki->i_op->put_link(link.dentry, &nd, cookie);
2551 path_put(&link);
806b681c 2552 }
10fa8e62 2553out:
2a737871
AV		 2554 if (nd.root.mnt)
2555 path_put(&nd.root);
31e6b01f 2556 if (filp == ERR_PTR(-ESTALE) && !(flags & LOOKUP_REVAL))
10fa8e62 2557 goto reval;
2dab5974 2558 release_open_intent(&nd);
10fa8e62 2559 return filp;
1da177e4 2560
806b681c
AV		 2561exit_dput:
2562 path_put_conditional(&path, &nd);
31e6b01f
NP		 2563out_path:
2564 path_put(&nd.path);
2565out_filp:
806b681c 2566 filp = ERR_PTR(error);
10fa8e62 2567 goto out;
1da177e4
LT		 2568}
2569
a70e65df
CH		 2570/**
2571 * filp_open - open file and return file pointer
2572 *
2573 * @filename: path to open
2574 * @flags: open flags as per the open(2) second argument
2575 * @mode: mode for the new file if O_CREAT is set, else ignored
2576 *
2577 * This is the helper to open a file from kernelspace if you really
2578 * have to. But in generally you should not do this, so please move
2579 * along, nothing to see here..
2580 */
2581struct file *filp_open(const char *filename, int flags, int mode)
2582{
6e8341a1 2583 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
a70e65df
CH		 2584}
2585EXPORT_SYMBOL(filp_open);
2586
1da177e4
LT		 2587/**
2588 * lookup_create - lookup a dentry, creating it if it doesn't exist
2589 * @nd: nameidata info
2590 * @is_dir: directory flag
2591 *
2592 * Simple function to lookup and return a dentry and create it
2593 * if it doesn't exist. Is SMP-safe.
c663e5d8 2594 *
4ac91378 2595 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1da177e4
LT		 2596 */
2597struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2598{
c663e5d8 2599 struct dentry *dentry = ERR_PTR(-EEXIST);
1da177e4 2600
4ac91378 2601 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
c663e5d8
CH		 2602 /*
2603 * Yucky last component or no last component at all?
2604 * (foo/., foo/.., /////)
2605 */
1da177e4
LT		 2606 if (nd->last_type != LAST_NORM)
2607 goto fail;
2608 nd->flags &= ~LOOKUP_PARENT;
3516586a 2609 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
a634904a 2610 nd->intent.open.flags = O_EXCL;
c663e5d8
CH		 2611
2612 /*
2613 * Do the final lookup.
2614 */
49705b77 2615 dentry = lookup_hash(nd);
1da177e4
LT		 2616 if (IS_ERR(dentry))
2617 goto fail;
c663e5d8 2618
e9baf6e5
AV		 2619 if (dentry->d_inode)
2620 goto eexist;
c663e5d8
CH		 2621 /*
2622 * Special case - lookup gave negative, but... we had foo/bar/
2623 * From the vfs_mknod() POV we just have a negative dentry -
2624 * all is fine. Let's be bastards - you had / on the end, you've
2625 * been asking for (non-existent) directory. -ENOENT for you.
2626 */
e9baf6e5
AV		 2627 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2628 dput(dentry);
2629 dentry = ERR_PTR(-ENOENT);
2630 }
1da177e4 2631 return dentry;
e9baf6e5 2632eexist:
1da177e4 2633 dput(dentry);
e9baf6e5 2634 dentry = ERR_PTR(-EEXIST);
1da177e4
LT		 2635fail:
2636 return dentry;
2637}
f81a0bff 2638EXPORT_SYMBOL_GPL(lookup_create);
1da177e4
LT		 2639
2640int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2641{
a95164d9 2642 int error = may_create(dir, dentry);
1da177e4
LT		 2643
2644 if (error)
2645 return error;
2646
2647 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2648 return -EPERM;
2649
acfa4380 2650 if (!dir->i_op->mknod)
1da177e4
LT		 2651 return -EPERM;
2652
08ce5f16
SH		 2653 error = devcgroup_inode_mknod(mode, dev);
2654 if (error)
2655 return error;
2656
1da177e4
LT		 2657 error = security_inode_mknod(dir, dentry, mode, dev);
2658 if (error)
2659 return error;
2660
1da177e4 2661 error = dir->i_op->mknod(dir, dentry, mode, dev);
a74574aa 2662 if (!error)
f38aa942 2663 fsnotify_create(dir, dentry);
1da177e4
LT		 2664 return error;
2665}
2666
463c3197
DH		 2667static int may_mknod(mode_t mode)
2668{
2669 switch (mode & S_IFMT) {
2670 case S_IFREG:
2671 case S_IFCHR:
2672 case S_IFBLK:
2673 case S_IFIFO:
2674 case S_IFSOCK:
2675 case 0: /* zero mode translates to S_IFREG */
2676 return 0;
2677 case S_IFDIR:
2678 return -EPERM;
2679 default:
2680 return -EINVAL;
2681 }
2682}
2683
2e4d0924
HC		 2684SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2685 unsigned, dev)
1da177e4 2686{
2ad94ae6
AV		 2687 int error;
2688 char *tmp;
2689 struct dentry *dentry;
1da177e4
LT		 2690 struct nameidata nd;
2691
2692 if (S_ISDIR(mode))
2693 return -EPERM;
1da177e4 2694
2ad94ae6 2695 error = user_path_parent(dfd, filename, &nd, &tmp);
1da177e4 2696 if (error)
2ad94ae6
AV		 2697 return error;
2698
1da177e4 2699 dentry = lookup_create(&nd, 0);
463c3197
DH		 2700 if (IS_ERR(dentry)) {
2701 error = PTR_ERR(dentry);
2702 goto out_unlock;
2703 }
4ac91378 2704 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2705 mode &= ~current_umask();
463c3197
DH		 2706 error = may_mknod(mode);
2707 if (error)
2708 goto out_dput;
2709 error = mnt_want_write(nd.path.mnt);
2710 if (error)
2711 goto out_dput;
be6d3e56
KT		 2712 error = security_path_mknod(&nd.path, dentry, mode, dev);
2713 if (error)
2714 goto out_drop_write;
463c3197 2715 switch (mode & S_IFMT) {
1da177e4 2716 case 0: case S_IFREG:
4ac91378 2717 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
1da177e4
LT		 2718 break;
2719 case S_IFCHR: case S_IFBLK:
4ac91378 2720 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
1da177e4
LT		 2721 new_decode_dev(dev));
2722 break;
2723 case S_IFIFO: case S_IFSOCK:
4ac91378 2724 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
1da177e4 2725 break;
1da177e4 2726 }
be6d3e56 2727out_drop_write:
463c3197
DH		 2728 mnt_drop_write(nd.path.mnt);
2729out_dput:
2730 dput(dentry);
2731out_unlock:
4ac91378 2732 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2733 path_put(&nd.path);
1da177e4
LT		 2734 putname(tmp);
2735
2736 return error;
2737}
2738
3480b257 2739SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
5590ff0d
UD		 2740{
2741 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2742}
2743
1da177e4
LT		 2744int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2745{
a95164d9 2746 int error = may_create(dir, dentry);
1da177e4
LT		 2747
2748 if (error)
2749 return error;
2750
acfa4380 2751 if (!dir->i_op->mkdir)
1da177e4
LT		 2752 return -EPERM;
2753
2754 mode &= (S_IRWXUGO|S_ISVTX);
2755 error = security_inode_mkdir(dir, dentry, mode);
2756 if (error)
2757 return error;
2758
1da177e4 2759 error = dir->i_op->mkdir(dir, dentry, mode);
a74574aa 2760 if (!error)
f38aa942 2761 fsnotify_mkdir(dir, dentry);
1da177e4
LT		 2762 return error;
2763}
2764
2e4d0924 2765SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
1da177e4
LT		 2766{
2767 int error = 0;
2768 char * tmp;
6902d925
DH		 2769 struct dentry *dentry;
2770 struct nameidata nd;
1da177e4 2771
2ad94ae6
AV		 2772 error = user_path_parent(dfd, pathname, &nd, &tmp);
2773 if (error)
6902d925 2774 goto out_err;
1da177e4 2775
6902d925
DH		 2776 dentry = lookup_create(&nd, 1);
2777 error = PTR_ERR(dentry);
2778 if (IS_ERR(dentry))
2779 goto out_unlock;
1da177e4 2780
4ac91378 2781 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2782 mode &= ~current_umask();
463c3197
DH		 2783 error = mnt_want_write(nd.path.mnt);
2784 if (error)
2785 goto out_dput;
be6d3e56
KT		 2786 error = security_path_mkdir(&nd.path, dentry, mode);
2787 if (error)
2788 goto out_drop_write;
4ac91378 2789 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
be6d3e56 2790out_drop_write:
463c3197
DH		 2791 mnt_drop_write(nd.path.mnt);
2792out_dput:
6902d925
DH		 2793 dput(dentry);
2794out_unlock:
4ac91378 2795 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2796 path_put(&nd.path);
6902d925
DH		 2797 putname(tmp);
2798out_err:
1da177e4
LT		 2799 return error;
2800}
2801
3cdad428 2802SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
5590ff0d
UD		 2803{
2804 return sys_mkdirat(AT_FDCWD, pathname, mode);
2805}
2806
1da177e4
LT		 2807/*
2808 * We try to drop the dentry early: we should have
2809 * a usage count of 2 if we're the only user of this
2810 * dentry, and if that is true (possibly after pruning
2811 * the dcache), then we drop the dentry now.
2812 *
2813 * A low-level filesystem can, if it choses, legally
2814 * do a
2815 *
2816 * if (!d_unhashed(dentry))
2817 * return -EBUSY;
2818 *
2819 * if it cannot handle the case of removing a directory
2820 * that is still in use by something else..
2821 */
2822void dentry_unhash(struct dentry *dentry)
2823{
2824 dget(dentry);
dc168427 2825 shrink_dcache_parent(dentry);
1da177e4 2826 spin_lock(&dentry->d_lock);
b7ab39f6 2827 if (dentry->d_count == 2)
1da177e4
LT		 2828 __d_drop(dentry);
2829 spin_unlock(&dentry->d_lock);
1da177e4
LT		 2830}
2831
2832int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2833{
2834 int error = may_delete(dir, dentry, 1);
2835
2836 if (error)
2837 return error;
2838
acfa4380 2839 if (!dir->i_op->rmdir)
1da177e4
LT		 2840 return -EPERM;
2841
1b1dcc1b 2842 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2843 dentry_unhash(dentry);
2844 if (d_mountpoint(dentry))
2845 error = -EBUSY;
2846 else {
2847 error = security_inode_rmdir(dir, dentry);
2848 if (!error) {
2849 error = dir->i_op->rmdir(dir, dentry);
d83c49f3 2850 if (!error) {
1da177e4 2851 dentry->d_inode->i_flags |= S_DEAD;
d83c49f3
AV		 2852 dont_mount(dentry);
2853 }
1da177e4
LT		 2854 }
2855 }
1b1dcc1b 2856 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4 2857 if (!error) {
1da177e4
LT		 2858 d_delete(dentry);
2859 }
2860 dput(dentry);
2861
2862 return error;
2863}
2864
5590ff0d 2865static long do_rmdir(int dfd, const char __user *pathname)
1da177e4
LT		 2866{
2867 int error = 0;
2868 char * name;
2869 struct dentry *dentry;
2870 struct nameidata nd;
2871
2ad94ae6 2872 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2873 if (error)
2ad94ae6 2874 return error;
1da177e4
LT		 2875
2876 switch(nd.last_type) {
0612d9fb
OH		 2877 case LAST_DOTDOT:
2878 error = -ENOTEMPTY;
2879 goto exit1;
2880 case LAST_DOT:
2881 error = -EINVAL;
2882 goto exit1;
2883 case LAST_ROOT:
2884 error = -EBUSY;
2885 goto exit1;
1da177e4 2886 }
0612d9fb
OH		 2887
2888 nd.flags &= ~LOOKUP_PARENT;
2889
4ac91378 2890 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2891 dentry = lookup_hash(&nd);
1da177e4 2892 error = PTR_ERR(dentry);
6902d925
DH		 2893 if (IS_ERR(dentry))
2894 goto exit2;
0622753b
DH		 2895 error = mnt_want_write(nd.path.mnt);
2896 if (error)
2897 goto exit3;
be6d3e56
KT		 2898 error = security_path_rmdir(&nd.path, dentry);
2899 if (error)
2900 goto exit4;
4ac91378 2901 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
be6d3e56 2902exit4:
0622753b
DH		 2903 mnt_drop_write(nd.path.mnt);
2904exit3:
6902d925
DH		 2905 dput(dentry);
2906exit2:
4ac91378 2907 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2908exit1:
1d957f9b 2909 path_put(&nd.path);
1da177e4
LT		 2910 putname(name);
2911 return error;
2912}
2913
3cdad428 2914SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
5590ff0d
UD		 2915{
2916 return do_rmdir(AT_FDCWD, pathname);
2917}
2918
1da177e4
LT		 2919int vfs_unlink(struct inode *dir, struct dentry *dentry)
2920{
2921 int error = may_delete(dir, dentry, 0);
2922
2923 if (error)
2924 return error;
2925
acfa4380 2926 if (!dir->i_op->unlink)
1da177e4
LT		 2927 return -EPERM;
2928
1b1dcc1b 2929 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2930 if (d_mountpoint(dentry))
2931 error = -EBUSY;
2932 else {
2933 error = security_inode_unlink(dir, dentry);
bec1052e 2934 if (!error) {
1da177e4 2935 error = dir->i_op->unlink(dir, dentry);
bec1052e 2936 if (!error)
d83c49f3 2937 dont_mount(dentry);
bec1052e 2938 }
1da177e4 2939 }
1b1dcc1b 2940 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2941
2942 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2943 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
ece95912 2944 fsnotify_link_count(dentry->d_inode);
e234f35c 2945 d_delete(dentry);
1da177e4 2946 }
0eeca283 2947
1da177e4
LT		 2948 return error;
2949}
2950
2951/*
2952 * Make sure that the actual truncation of the file will occur outside its
1b1dcc1b 2953 * directory's i_mutex. Truncate can take a long time if there is a lot of
1da177e4
LT		 2954 * writeout happening, and we don't want to prevent access to the directory
2955 * while waiting on the I/O.
2956 */
5590ff0d 2957static long do_unlinkat(int dfd, const char __user *pathname)
1da177e4 2958{
2ad94ae6
AV		 2959 int error;
2960 char *name;
1da177e4
LT		 2961 struct dentry *dentry;
2962 struct nameidata nd;
2963 struct inode *inode = NULL;
2964
2ad94ae6 2965 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2966 if (error)
2ad94ae6
AV		 2967 return error;
2968
1da177e4
LT		 2969 error = -EISDIR;
2970 if (nd.last_type != LAST_NORM)
2971 goto exit1;
0612d9fb
OH		 2972
2973 nd.flags &= ~LOOKUP_PARENT;
2974
4ac91378 2975 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2976 dentry = lookup_hash(&nd);
1da177e4
LT		 2977 error = PTR_ERR(dentry);
2978 if (!IS_ERR(dentry)) {
2979 /* Why not before? Because we want correct error value */
2980 if (nd.last.name[nd.last.len])
2981 goto slashes;
2982 inode = dentry->d_inode;
2983 if (inode)
7de9c6ee 2984 ihold(inode);
0622753b
DH		 2985 error = mnt_want_write(nd.path.mnt);
2986 if (error)
2987 goto exit2;
be6d3e56
KT		 2988 error = security_path_unlink(&nd.path, dentry);
2989 if (error)
2990 goto exit3;
4ac91378 2991 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
be6d3e56 2992exit3:
0622753b 2993 mnt_drop_write(nd.path.mnt);
1da177e4
LT		 2994 exit2:
2995 dput(dentry);
2996 }
4ac91378 2997 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4
LT		 2998 if (inode)
2999 iput(inode); /* truncate the inode here */
3000exit1:
1d957f9b 3001 path_put(&nd.path);
1da177e4
LT		 3002 putname(name);
3003 return error;
3004
3005slashes:
3006 error = !dentry->d_inode ? -ENOENT :
3007 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
3008 goto exit2;
3009}
3010
2e4d0924 3011SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
5590ff0d
UD		 3012{
3013 if ((flag & ~AT_REMOVEDIR) != 0)
3014 return -EINVAL;
3015
3016 if (flag & AT_REMOVEDIR)
3017 return do_rmdir(dfd, pathname);
3018
3019 return do_unlinkat(dfd, pathname);
3020}
3021
3480b257 3022SYSCALL_DEFINE1(unlink, const char __user *, pathname)
5590ff0d
UD		 3023{
3024 return do_unlinkat(AT_FDCWD, pathname);
3025}
3026
db2e747b 3027int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
1da177e4 3028{
a95164d9 3029 int error = may_create(dir, dentry);
1da177e4
LT		 3030
3031 if (error)
3032 return error;
3033
acfa4380 3034 if (!dir->i_op->symlink)
1da177e4
LT		 3035 return -EPERM;
3036
3037 error = security_inode_symlink(dir, dentry, oldname);
3038 if (error)
3039 return error;
3040
1da177e4 3041 error = dir->i_op->symlink(dir, dentry, oldname);
a74574aa 3042 if (!error)
f38aa942 3043 fsnotify_create(dir, dentry);
1da177e4
LT		 3044 return error;
3045}
3046
2e4d0924
HC		 3047SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3048 int, newdfd, const char __user *, newname)
1da177e4 3049{
2ad94ae6
AV		 3050 int error;
3051 char *from;
3052 char *to;
6902d925
DH		 3053 struct dentry *dentry;
3054 struct nameidata nd;
1da177e4
LT		 3055
3056 from = getname(oldname);
2ad94ae6 3057 if (IS_ERR(from))
1da177e4 3058 return PTR_ERR(from);
1da177e4 3059
2ad94ae6 3060 error = user_path_parent(newdfd, newname, &nd, &to);
6902d925 3061 if (error)
2ad94ae6
AV		 3062 goto out_putname;
3063
6902d925
DH		 3064 dentry = lookup_create(&nd, 0);
3065 error = PTR_ERR(dentry);
3066 if (IS_ERR(dentry))
3067 goto out_unlock;
3068
75c3f29d
DH		 3069 error = mnt_want_write(nd.path.mnt);
3070 if (error)
3071 goto out_dput;
be6d3e56
KT		 3072 error = security_path_symlink(&nd.path, dentry, from);
3073 if (error)
3074 goto out_drop_write;
db2e747b 3075 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
be6d3e56 3076out_drop_write:
75c3f29d
DH		 3077 mnt_drop_write(nd.path.mnt);
3078out_dput:
6902d925
DH		 3079 dput(dentry);
3080out_unlock:
4ac91378 3081 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 3082 path_put(&nd.path);
6902d925
DH		 3083 putname(to);
3084out_putname:
1da177e4
LT		 3085 putname(from);
3086 return error;
3087}
3088
3480b257 3089SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 3090{
3091 return sys_symlinkat(oldname, AT_FDCWD, newname);
3092}
3093
1da177e4
LT		 3094int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
3095{
3096 struct inode *inode = old_dentry->d_inode;
3097 int error;
3098
3099 if (!inode)
3100 return -ENOENT;
3101
a95164d9 3102 error = may_create(dir, new_dentry);
1da177e4
LT		 3103 if (error)
3104 return error;
3105
3106 if (dir->i_sb != inode->i_sb)
3107 return -EXDEV;
3108
3109 /*
3110 * A link to an append-only or immutable file cannot be created.
3111 */
3112 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3113 return -EPERM;
acfa4380 3114 if (!dir->i_op->link)
1da177e4 3115 return -EPERM;
7e79eedb 3116 if (S_ISDIR(inode->i_mode))
1da177e4
LT		 3117 return -EPERM;
3118
3119 error = security_inode_link(old_dentry, dir, new_dentry);
3120 if (error)
3121 return error;
3122
7e79eedb 3123 mutex_lock(&inode->i_mutex);
1da177e4 3124 error = dir->i_op->link(old_dentry, dir, new_dentry);
7e79eedb 3125 mutex_unlock(&inode->i_mutex);
e31e14ec 3126 if (!error)
7e79eedb 3127 fsnotify_link(dir, inode, new_dentry);
1da177e4
LT		 3128 return error;
3129}
3130
3131/*
3132 * Hardlinks are often used in delicate situations. We avoid
3133 * security-related surprises by not following symlinks on the
3134 * newname. --KAB
3135 *
3136 * We don't follow them on the oldname either to be compatible
3137 * with linux 2.0, and to avoid hard-linking to directories
3138 * and other special files. --ADM
3139 */
2e4d0924
HC		 3140SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3141 int, newdfd, const char __user *, newname, int, flags)
1da177e4
LT		 3142{
3143 struct dentry *new_dentry;
2d8f3038
AV		 3144 struct nameidata nd;
3145 struct path old_path;
1da177e4 3146 int error;
2ad94ae6 3147 char *to;
1da177e4 3148
45c9b11a 3149 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
c04030e1
UD		 3150 return -EINVAL;
3151
2d8f3038
AV		 3152 error = user_path_at(olddfd, oldname,
3153 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
3154 &old_path);
1da177e4 3155 if (error)
2ad94ae6
AV		 3156 return error;
3157
3158 error = user_path_parent(newdfd, newname, &nd, &to);
1da177e4
LT		 3159 if (error)
3160 goto out;
3161 error = -EXDEV;
2d8f3038 3162 if (old_path.mnt != nd.path.mnt)
1da177e4
LT		 3163 goto out_release;
3164 new_dentry = lookup_create(&nd, 0);
3165 error = PTR_ERR(new_dentry);
6902d925
DH		 3166 if (IS_ERR(new_dentry))
3167 goto out_unlock;
75c3f29d
DH		 3168 error = mnt_want_write(nd.path.mnt);
3169 if (error)
3170 goto out_dput;
be6d3e56
KT		 3171 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
3172 if (error)
3173 goto out_drop_write;
2d8f3038 3174 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
be6d3e56 3175out_drop_write:
75c3f29d
DH		 3176 mnt_drop_write(nd.path.mnt);
3177out_dput:
6902d925
DH		 3178 dput(new_dentry);
3179out_unlock:
4ac91378 3180 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 3181out_release:
1d957f9b 3182 path_put(&nd.path);
2ad94ae6 3183 putname(to);
1da177e4 3184out:
2d8f3038 3185 path_put(&old_path);
1da177e4
LT		 3186
3187 return error;
3188}
3189
3480b257 3190SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
5590ff0d 3191{
c04030e1 3192 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
5590ff0d
UD		 3193}
3194
1da177e4
LT		 3195/*
3196 * The worst of all namespace operations - renaming directory. "Perverted"
3197 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3198 * Problems:
3199 * a) we can get into loop creation. Check is done in is_subdir().
3200 * b) race potential - two innocent renames can create a loop together.
3201 * That's where 4.4 screws up. Current fix: serialization on
a11f3a05 3202 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
1da177e4
LT		 3203 * story.
3204 * c) we have to lock _three_ objects - parents and victim (if it exists).
1b1dcc1b 3205 * And that - after we got ->i_mutex on parents (until then we don't know
1da177e4
LT		 3206 * whether the target exists). Solution: try to be smart with locking
3207 * order for inodes. We rely on the fact that tree topology may change
a11f3a05 3208 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
1da177e4
LT		 3209 * move will be locked. Thus we can rank directories by the tree
3210 * (ancestors first) and rank all non-directories after them.
3211 * That works since everybody except rename does "lock parent, lookup,
a11f3a05 3212 * lock child" and rename is under ->s_vfs_rename_mutex.
1da177e4
LT		 3213 * HOWEVER, it relies on the assumption that any object with ->lookup()
3214 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3215 * we'd better make sure that there's no link(2) for them.
3216 * d) some filesystems don't support opened-but-unlinked directories,
3217 * either because of layout or because they are not ready to deal with
3218 * all cases correctly. The latter will be fixed (taking this sort of
3219 * stuff into VFS), but the former is not going away. Solution: the same
3220 * trick as in rmdir().
3221 * e) conversion from fhandle to dentry may come in the wrong moment - when
1b1dcc1b 3222 * we are removing the target. Solution: we will have to grab ->i_mutex
1da177e4 3223 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
c41b20e7 3224 * ->i_mutex on parents, which works but leads to some truly excessive
1da177e4
LT		 3225 * locking].
3226 */
75c96f85
AB		 3227static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3228 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3229{
3230 int error = 0;
3231 struct inode *target;
3232
3233 /*
3234 * If we are going to change the parent - check write permissions,
3235 * we'll need to flip '..'.
3236 */
3237 if (new_dir != old_dir) {
f419a2e3 3238 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
1da177e4
LT		 3239 if (error)
3240 return error;
3241 }
3242
3243 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3244 if (error)
3245 return error;
3246
3247 target = new_dentry->d_inode;
d83c49f3 3248 if (target)
1b1dcc1b 3249 mutex_lock(&target->i_mutex);
1da177e4
LT		 3250 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3251 error = -EBUSY;
d83c49f3
AV		 3252 else {
3253 if (target)
3254 dentry_unhash(new_dentry);
1da177e4 3255 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
d83c49f3 3256 }
1da177e4 3257 if (target) {
d83c49f3 3258 if (!error) {
1da177e4 3259 target->i_flags |= S_DEAD;
d83c49f3
AV		 3260 dont_mount(new_dentry);
3261 }
1b1dcc1b 3262 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3263 if (d_unhashed(new_dentry))
3264 d_rehash(new_dentry);
3265 dput(new_dentry);
3266 }
e31e14ec 3267 if (!error)
349457cc
MF		 3268 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3269 d_move(old_dentry,new_dentry);
1da177e4
LT		 3270 return error;
3271}
3272
75c96f85
AB		 3273static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3274 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3275{
3276 struct inode *target;
3277 int error;
3278
3279 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3280 if (error)
3281 return error;
3282
3283 dget(new_dentry);
3284 target = new_dentry->d_inode;
3285 if (target)
1b1dcc1b 3286 mutex_lock(&target->i_mutex);
1da177e4
LT		 3287 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3288 error = -EBUSY;
3289 else
3290 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3291 if (!error) {
bec1052e 3292 if (target)
d83c49f3 3293 dont_mount(new_dentry);
349457cc 3294 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
1da177e4 3295 d_move(old_dentry, new_dentry);
1da177e4
LT		 3296 }
3297 if (target)
1b1dcc1b 3298 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3299 dput(new_dentry);
3300 return error;
3301}
3302
3303int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3304 struct inode *new_dir, struct dentry *new_dentry)
3305{
3306 int error;
3307 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
59b0df21 3308 const unsigned char *old_name;
1da177e4
LT		 3309
3310 if (old_dentry->d_inode == new_dentry->d_inode)
3311 return 0;
3312
3313 error = may_delete(old_dir, old_dentry, is_dir);
3314 if (error)
3315 return error;
3316
3317 if (!new_dentry->d_inode)
a95164d9 3318 error = may_create(new_dir, new_dentry);
1da177e4
LT		 3319 else
3320 error = may_delete(new_dir, new_dentry, is_dir);
3321 if (error)
3322 return error;
3323
acfa4380 3324 if (!old_dir->i_op->rename)
1da177e4
LT		 3325 return -EPERM;
3326
0eeca283
RL		 3327 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3328
1da177e4
LT		 3329 if (is_dir)
3330 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3331 else
3332 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
123df294
AV		 3333 if (!error)
3334 fsnotify_move(old_dir, new_dir, old_name, is_dir,
5a190ae6 3335 new_dentry->d_inode, old_dentry);
0eeca283
RL		 3336 fsnotify_oldname_free(old_name);
3337
1da177e4
LT		 3338 return error;
3339}
3340
2e4d0924
HC		 3341SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3342 int, newdfd, const char __user *, newname)
1da177e4 3343{
2ad94ae6
AV		 3344 struct dentry *old_dir, *new_dir;
3345 struct dentry *old_dentry, *new_dentry;
3346 struct dentry *trap;
1da177e4 3347 struct nameidata oldnd, newnd;
2ad94ae6
AV		 3348 char *from;
3349 char *to;
3350 int error;
1da177e4 3351
2ad94ae6 3352 error = user_path_parent(olddfd, oldname, &oldnd, &from);
1da177e4
LT		 3353 if (error)
3354 goto exit;
3355
2ad94ae6 3356 error = user_path_parent(newdfd, newname, &newnd, &to);
1da177e4
LT		 3357 if (error)
3358 goto exit1;
3359
3360 error = -EXDEV;
4ac91378 3361 if (oldnd.path.mnt != newnd.path.mnt)
1da177e4
LT		 3362 goto exit2;
3363
4ac91378 3364 old_dir = oldnd.path.dentry;
1da177e4
LT		 3365 error = -EBUSY;
3366 if (oldnd.last_type != LAST_NORM)
3367 goto exit2;
3368
4ac91378 3369 new_dir = newnd.path.dentry;
1da177e4
LT		 3370 if (newnd.last_type != LAST_NORM)
3371 goto exit2;
3372
0612d9fb
OH		 3373 oldnd.flags &= ~LOOKUP_PARENT;
3374 newnd.flags &= ~LOOKUP_PARENT;
4e9ed2f8 3375 newnd.flags |= LOOKUP_RENAME_TARGET;
0612d9fb 3376
1da177e4
LT		 3377 trap = lock_rename(new_dir, old_dir);
3378
49705b77 3379 old_dentry = lookup_hash(&oldnd);
1da177e4
LT		 3380 error = PTR_ERR(old_dentry);
3381 if (IS_ERR(old_dentry))
3382 goto exit3;
3383 /* source must exist */
3384 error = -ENOENT;
3385 if (!old_dentry->d_inode)
3386 goto exit4;
3387 /* unless the source is a directory trailing slashes give -ENOTDIR */
3388 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3389 error = -ENOTDIR;
3390 if (oldnd.last.name[oldnd.last.len])
3391 goto exit4;
3392 if (newnd.last.name[newnd.last.len])
3393 goto exit4;
3394 }
3395 /* source should not be ancestor of target */
3396 error = -EINVAL;
3397 if (old_dentry == trap)
3398 goto exit4;
49705b77 3399 new_dentry = lookup_hash(&newnd);
1da177e4
LT		 3400 error = PTR_ERR(new_dentry);
3401 if (IS_ERR(new_dentry))
3402 goto exit4;
3403 /* target should not be an ancestor of source */
3404 error = -ENOTEMPTY;
3405 if (new_dentry == trap)
3406 goto exit5;
3407
9079b1eb
DH		 3408 error = mnt_want_write(oldnd.path.mnt);
3409 if (error)
3410 goto exit5;
be6d3e56
KT		 3411 error = security_path_rename(&oldnd.path, old_dentry,
3412 &newnd.path, new_dentry);
3413 if (error)
3414 goto exit6;
1da177e4
LT		 3415 error = vfs_rename(old_dir->d_inode, old_dentry,
3416 new_dir->d_inode, new_dentry);
be6d3e56 3417exit6:
9079b1eb 3418 mnt_drop_write(oldnd.path.mnt);
1da177e4
LT		 3419exit5:
3420 dput(new_dentry);
3421exit4:
3422 dput(old_dentry);
3423exit3:
3424 unlock_rename(new_dir, old_dir);
3425exit2:
1d957f9b 3426 path_put(&newnd.path);
2ad94ae6 3427 putname(to);
1da177e4 3428exit1:
1d957f9b 3429 path_put(&oldnd.path);
1da177e4 3430 putname(from);
2ad94ae6 3431exit:
1da177e4
LT		 3432 return error;
3433}
3434
a26eab24 3435SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 3436{
3437 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3438}
3439
1da177e4
LT		 3440int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3441{
3442 int len;
3443
3444 len = PTR_ERR(link);
3445 if (IS_ERR(link))
3446 goto out;
3447
3448 len = strlen(link);
3449 if (len > (unsigned) buflen)
3450 len = buflen;
3451 if (copy_to_user(buffer, link, len))
3452 len = -EFAULT;
3453out:
3454 return len;
3455}
3456
3457/*
3458 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3459 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3460 * using) it for any given inode is up to filesystem.
3461 */
3462int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3463{
3464 struct nameidata nd;
cc314eef 3465 void *cookie;
694a1764 3466 int res;
cc314eef 3467
1da177e4 3468 nd.depth = 0;
cc314eef 3469 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
694a1764
MS		 3470 if (IS_ERR(cookie))
3471 return PTR_ERR(cookie);
3472
3473 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3474 if (dentry->d_inode->i_op->put_link)
3475 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3476 return res;
1da177e4
LT		 3477}
3478
3479int vfs_follow_link(struct nameidata *nd, const char *link)
3480{
3481 return __vfs_follow_link(nd, link);
3482}
3483
3484/* get the link contents into pagecache */
3485static char *page_getlink(struct dentry * dentry, struct page **ppage)
3486{
ebd09abb
DG		 3487 char *kaddr;
3488 struct page *page;
1da177e4 3489 struct address_space *mapping = dentry->d_inode->i_mapping;
090d2b18 3490 page = read_mapping_page(mapping, 0, NULL);
1da177e4 3491 if (IS_ERR(page))
6fe6900e 3492 return (char*)page;
1da177e4 3493 *ppage = page;
ebd09abb
DG		 3494 kaddr = kmap(page);
3495 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3496 return kaddr;
1da177e4
LT		 3497}
3498
3499int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3500{
3501 struct page *page = NULL;
3502 char *s = page_getlink(dentry, &page);
3503 int res = vfs_readlink(dentry,buffer,buflen,s);
3504 if (page) {
3505 kunmap(page);
3506 page_cache_release(page);
3507 }
3508 return res;
3509}
3510
cc314eef 3511void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
1da177e4 3512{
cc314eef 3513 struct page *page = NULL;
1da177e4 3514 nd_set_link(nd, page_getlink(dentry, &page));
cc314eef 3515 return page;
1da177e4
LT		 3516}
3517
cc314eef 3518void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1da177e4 3519{
cc314eef
LT		 3520 struct page *page = cookie;
3521
3522 if (page) {
1da177e4
LT		 3523 kunmap(page);
3524 page_cache_release(page);
1da177e4
LT		 3525 }
3526}
3527
54566b2c
NP		 3528/*
3529 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3530 */
3531int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
1da177e4
LT		 3532{
3533 struct address_space *mapping = inode->i_mapping;
0adb25d2 3534 struct page *page;
afddba49 3535 void *fsdata;
beb497ab 3536 int err;
1da177e4 3537 char *kaddr;
54566b2c
NP		 3538 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3539 if (nofs)
3540 flags |= AOP_FLAG_NOFS;
1da177e4 3541
7e53cac4 3542retry:
afddba49 3543 err = pagecache_write_begin(NULL, mapping, 0, len-1,
54566b2c 3544 flags, &page, &fsdata);
1da177e4 3545 if (err)
afddba49
NP		 3546 goto fail;
3547
1da177e4
LT		 3548 kaddr = kmap_atomic(page, KM_USER0);
3549 memcpy(kaddr, symname, len-1);
3550 kunmap_atomic(kaddr, KM_USER0);
afddba49
NP		 3551
3552 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3553 page, fsdata);
1da177e4
LT		 3554 if (err < 0)
3555 goto fail;
afddba49
NP		 3556 if (err < len-1)
3557 goto retry;
3558
1da177e4
LT		 3559 mark_inode_dirty(inode);
3560 return 0;
1da177e4
LT		 3561fail:
3562 return err;
3563}
3564
0adb25d2
KK		 3565int page_symlink(struct inode *inode, const char *symname, int len)
3566{
3567 return __page_symlink(inode, symname, len,
54566b2c 3568 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
0adb25d2
KK		 3569}
3570
92e1d5be 3571const struct inode_operations page_symlink_inode_operations = {
1da177e4
LT		 3572 .readlink = generic_readlink,
3573 .follow_link = page_follow_link_light,
3574 .put_link = page_put_link,
3575};
3576
2d8f3038 3577EXPORT_SYMBOL(user_path_at);
cc53ce53 3578EXPORT_SYMBOL(follow_down_one);
1da177e4
LT		 3579EXPORT_SYMBOL(follow_down);
3580EXPORT_SYMBOL(follow_up);
3581EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3582EXPORT_SYMBOL(getname);
3583EXPORT_SYMBOL(lock_rename);
1da177e4
LT		 3584EXPORT_SYMBOL(lookup_one_len);
3585EXPORT_SYMBOL(page_follow_link_light);
3586EXPORT_SYMBOL(page_put_link);
3587EXPORT_SYMBOL(page_readlink);
0adb25d2 3588EXPORT_SYMBOL(__page_symlink);
1da177e4
LT		 3589EXPORT_SYMBOL(page_symlink);
3590EXPORT_SYMBOL(page_symlink_inode_operations);
c9c6cac0 3591EXPORT_SYMBOL(kern_path_parent);
d1811465 3592EXPORT_SYMBOL(kern_path);
16f18200 3593EXPORT_SYMBOL(vfs_path_lookup);
f419a2e3 3594EXPORT_SYMBOL(inode_permission);
8c744fb8 3595EXPORT_SYMBOL(file_permission);
1da177e4
LT		 3596EXPORT_SYMBOL(unlock_rename);
3597EXPORT_SYMBOL(vfs_create);
3598EXPORT_SYMBOL(vfs_follow_link);
3599EXPORT_SYMBOL(vfs_link);
3600EXPORT_SYMBOL(vfs_mkdir);
3601EXPORT_SYMBOL(vfs_mknod);
3602EXPORT_SYMBOL(generic_permission);
3603EXPORT_SYMBOL(vfs_readlink);
3604EXPORT_SYMBOL(vfs_rename);
3605EXPORT_SYMBOL(vfs_rmdir);
3606EXPORT_SYMBOL(vfs_symlink);
3607EXPORT_SYMBOL(vfs_unlink);
3608EXPORT_SYMBOL(dentry_unhash);
3609EXPORT_SYMBOL(generic_readlink);
kernel source for telekom puls (alcatel/mediatek)