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