projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
xfs: provide simple rcu-walk ACL implementation
[deliverable/linux.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
b74c79e9 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
386 * @Returns: 0 on success, -ECHLID on failure
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
447 * @Returns: 0 on success, -ECHLID on failure
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
509 * @Returns: 0 on success, -ECHLID on failure
510 *
511 * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk.
512 * nd->path should be the final element of the lookup, so nd->root is discarded.
513 * Must be called from rcu-walk context.
514 */
515static int nameidata_drop_rcu_last(struct nameidata *nd)
516{
517 struct dentry *dentry = nd->path.dentry;
518
519 BUG_ON(!(nd->flags & LOOKUP_RCU));
520 nd->flags &= ~LOOKUP_RCU;
521 nd->root.mnt = NULL;
522 spin_lock(&dentry->d_lock);
523 if (!__d_rcu_to_refcount(dentry, nd->seq))
524 goto err_unlock;
525 BUG_ON(nd->inode != dentry->d_inode);
526 spin_unlock(&dentry->d_lock);
527
528 mntget(nd->path.mnt);
529
530 rcu_read_unlock();
531 br_read_unlock(vfsmount_lock);
532
533 return 0;
534
535err_unlock:
536 spin_unlock(&dentry->d_lock);
537 rcu_read_unlock();
538 br_read_unlock(vfsmount_lock);
539 return -ECHILD;
540}
541
542/* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
543static inline int nameidata_drop_rcu_last_maybe(struct nameidata *nd)
544{
545 if (likely(nd->flags & LOOKUP_RCU))
546 return nameidata_drop_rcu_last(nd);
547 return 0;
548}
549
834f2a4a
TM		 550/**
551 * release_open_intent - free up open intent resources
552 * @nd: pointer to nameidata
553 */
554void release_open_intent(struct nameidata *nd)
555{
0f7fc9e4 556 if (nd->intent.open.file->f_path.dentry == NULL)
834f2a4a
TM		 557 put_filp(nd->intent.open.file);
558 else
559 fput(nd->intent.open.file);
560}
561
34286d66
NP		 562static int d_revalidate(struct dentry *dentry, struct nameidata *nd)
563{
564 int status;
565
566 status = dentry->d_op->d_revalidate(dentry, nd);
567 if (status == -ECHILD) {
568 if (nameidata_dentry_drop_rcu(nd, dentry))
569 return status;
570 status = dentry->d_op->d_revalidate(dentry, nd);
571 }
572
573 return status;
574}
575
bcdc5e01
IK		 576static inline struct dentry *
577do_revalidate(struct dentry *dentry, struct nameidata *nd)
578{
34286d66
NP		 579 int status;
580
581 status = d_revalidate(dentry, nd);
bcdc5e01
IK		 582 if (unlikely(status <= 0)) {
583 /*
584 * The dentry failed validation.
585 * If d_revalidate returned 0 attempt to invalidate
586 * the dentry otherwise d_revalidate is asking us
587 * to return a fail status.
588 */
34286d66
NP		 589 if (status < 0) {
590 /* If we're in rcu-walk, we don't have a ref */
591 if (!(nd->flags & LOOKUP_RCU))
592 dput(dentry);
593 dentry = ERR_PTR(status);
594
595 } else {
596 /* Don't d_invalidate in rcu-walk mode */
597 if (nameidata_dentry_drop_rcu_maybe(nd, dentry))
598 return ERR_PTR(-ECHILD);
bcdc5e01
IK		 599 if (!d_invalidate(dentry)) {
600 dput(dentry);
601 dentry = NULL;
602 }
bcdc5e01
IK		 603 }
604 }
605 return dentry;
606}
607
fb045adb
NP		 608static inline int need_reval_dot(struct dentry *dentry)
609{
610 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
611 return 0;
612
613 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
614 return 0;
615
616 return 1;
617}
618
39159de2
JL		 619/*
620 * force_reval_path - force revalidation of a dentry
621 *
622 * In some situations the path walking code will trust dentries without
623 * revalidating them. This causes problems for filesystems that depend on
624 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
625 * (which indicates that it's possible for the dentry to go stale), force
626 * a d_revalidate call before proceeding.
627 *
628 * Returns 0 if the revalidation was successful. If the revalidation fails,
629 * either return the error returned by d_revalidate or -ESTALE if the
630 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
631 * invalidate the dentry. It's up to the caller to handle putting references
632 * to the path if necessary.
633 */
634static int
635force_reval_path(struct path *path, struct nameidata *nd)
636{
637 int status;
638 struct dentry *dentry = path->dentry;
639
640 /*
641 * only check on filesystems where it's possible for the dentry to
fb045adb 642 * become stale.
39159de2 643 */
fb045adb 644 if (!need_reval_dot(dentry))
39159de2
JL		 645 return 0;
646
34286d66 647 status = d_revalidate(dentry, nd);
39159de2
JL		 648 if (status > 0)
649 return 0;
650
651 if (!status) {
652 d_invalidate(dentry);
653 status = -ESTALE;
654 }
655 return status;
656}
657
1da177e4 658/*
b75b5086
AV		 659 * Short-cut version of permission(), for calling on directories
660 * during pathname resolution. Combines parts of permission()
661 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
1da177e4
LT		 662 *
663 * If appropriate, check DAC only. If not appropriate, or
b75b5086 664 * short-cut DAC fails, then call ->permission() to do more
1da177e4
LT		 665 * complete permission check.
666 */
b74c79e9 667static inline int exec_permission(struct inode *inode, unsigned int flags)
1da177e4 668{
5909ccaa 669 int ret;
1da177e4 670
cb9179ea 671 if (inode->i_op->permission) {
b74c79e9
NP		 672 ret = inode->i_op->permission(inode, MAY_EXEC, flags);
673 } else {
674 ret = acl_permission_check(inode, MAY_EXEC, flags,
675 inode->i_op->check_acl);
cb9179ea 676 }
b74c79e9 677 if (likely(!ret))
1da177e4 678 goto ok;
b74c79e9 679 if (ret == -ECHILD)
31e6b01f 680 return ret;
1da177e4 681
f1ac9f6b 682 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
1da177e4
LT		 683 goto ok;
684
5909ccaa 685 return ret;
1da177e4 686ok:
b74c79e9 687 return security_inode_exec_permission(inode, flags);
1da177e4
LT		 688}
689
2a737871
AV		 690static __always_inline void set_root(struct nameidata *nd)
691{
f7ad3c6b
MS		 692 if (!nd->root.mnt)
693 get_fs_root(current->fs, &nd->root);
2a737871
AV		 694}
695
6de88d72
AV		 696static int link_path_walk(const char *, struct nameidata *);
697
31e6b01f
NP		 698static __always_inline void set_root_rcu(struct nameidata *nd)
699{
700 if (!nd->root.mnt) {
701 struct fs_struct *fs = current->fs;
c28cc364
NP		 702 unsigned seq;
703
704 do {
705 seq = read_seqcount_begin(&fs->seq);
706 nd->root = fs->root;
707 } while (read_seqcount_retry(&fs->seq, seq));
31e6b01f
NP		 708 }
709}
710
f1662356 711static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
1da177e4 712{
31e6b01f
NP		 713 int ret;
714
1da177e4
LT		 715 if (IS_ERR(link))
716 goto fail;
717
718 if (*link == '/') {
2a737871 719 set_root(nd);
1d957f9b 720 path_put(&nd->path);
2a737871
AV		 721 nd->path = nd->root;
722 path_get(&nd->root);
1da177e4 723 }
31e6b01f 724 nd->inode = nd->path.dentry->d_inode;
b4091d5f 725
31e6b01f
NP		 726 ret = link_path_walk(link, nd);
727 return ret;
1da177e4 728fail:
1d957f9b 729 path_put(&nd->path);
1da177e4
LT		 730 return PTR_ERR(link);
731}
732
1d957f9b 733static void path_put_conditional(struct path *path, struct nameidata *nd)
051d3812
IK		 734{
735 dput(path->dentry);
4ac91378 736 if (path->mnt != nd->path.mnt)
051d3812
IK		 737 mntput(path->mnt);
738}
739
740static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
741{
31e6b01f
NP		 742 if (!(nd->flags & LOOKUP_RCU)) {
743 dput(nd->path.dentry);
744 if (nd->path.mnt != path->mnt)
745 mntput(nd->path.mnt);
9a229683 746 }
31e6b01f 747 nd->path.mnt = path->mnt;
4ac91378 748 nd->path.dentry = path->dentry;
051d3812
IK		 749}
750
def4af30
AV		 751static __always_inline int
752__do_follow_link(struct path *path, struct nameidata *nd, void **p)
1da177e4
LT		 753{
754 int error;
cd4e91d3 755 struct dentry *dentry = path->dentry;
1da177e4 756
d671a1cb 757 touch_atime(path->mnt, dentry);
1da177e4 758 nd_set_link(nd, NULL);
cd4e91d3 759
4ac91378 760 if (path->mnt != nd->path.mnt) {
051d3812 761 path_to_nameidata(path, nd);
31e6b01f 762 nd->inode = nd->path.dentry->d_inode;
051d3812
IK		 763 dget(dentry);
764 }
765 mntget(path->mnt);
31e6b01f 766
86acdca1 767 nd->last_type = LAST_BIND;
def4af30
AV		 768 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
769 error = PTR_ERR(*p);
770 if (!IS_ERR(*p)) {
1da177e4 771 char *s = nd_get_link(nd);
cc314eef 772 error = 0;
1da177e4
LT		 773 if (s)
774 error = __vfs_follow_link(nd, s);
39159de2
JL		 775 else if (nd->last_type == LAST_BIND) {
776 error = force_reval_path(&nd->path, nd);
777 if (error)
778 path_put(&nd->path);
779 }
1da177e4 780 }
1da177e4
LT		 781 return error;
782}
783
784/*
785 * This limits recursive symlink follows to 8, while
786 * limiting consecutive symlinks to 40.
787 *
788 * Without that kind of total limit, nasty chains of consecutive
789 * symlinks can cause almost arbitrarily long lookups.
790 */
90ebe565 791static inline int do_follow_link(struct path *path, struct nameidata *nd)
1da177e4 792{
def4af30 793 void *cookie;
1da177e4
LT		 794 int err = -ELOOP;
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();
90ebe565 801 err = security_inode_follow_link(path->dentry, nd);
1da177e4
LT		 802 if (err)
803 goto loop;
804 current->link_count++;
805 current->total_link_count++;
806 nd->depth++;
def4af30
AV		 807 err = __do_follow_link(path, nd, &cookie);
808 if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link)
809 path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie);
258fa999 810 path_put(path);
839d9f93
AV		 811 current->link_count--;
812 nd->depth--;
1da177e4
LT		 813 return err;
814loop:
1d957f9b
JB		 815 path_put_conditional(path, nd);
816 path_put(&nd->path);
1da177e4
LT		 817 return err;
818}
819
31e6b01f
NP		 820static int follow_up_rcu(struct path *path)
821{
822 struct vfsmount *parent;
823 struct dentry *mountpoint;
824
825 parent = path->mnt->mnt_parent;
826 if (parent == path->mnt)
827 return 0;
828 mountpoint = path->mnt->mnt_mountpoint;
829 path->dentry = mountpoint;
830 path->mnt = parent;
831 return 1;
832}
833
bab77ebf 834int follow_up(struct path *path)
1da177e4
LT		 835{
836 struct vfsmount *parent;
837 struct dentry *mountpoint;
99b7db7b
NP		 838
839 br_read_lock(vfsmount_lock);
bab77ebf
AV		 840 parent = path->mnt->mnt_parent;
841 if (parent == path->mnt) {
99b7db7b 842 br_read_unlock(vfsmount_lock);
1da177e4
LT		 843 return 0;
844 }
845 mntget(parent);
bab77ebf 846 mountpoint = dget(path->mnt->mnt_mountpoint);
99b7db7b 847 br_read_unlock(vfsmount_lock);
bab77ebf
AV		 848 dput(path->dentry);
849 path->dentry = mountpoint;
850 mntput(path->mnt);
851 path->mnt = parent;
1da177e4
LT		 852 return 1;
853}
854
b5c84bf6
NP		 855/*
856 * serialization is taken care of in namespace.c
1da177e4 857 */
31e6b01f
NP		 858static void __follow_mount_rcu(struct nameidata *nd, struct path *path,
859 struct inode **inode)
860{
861 while (d_mountpoint(path->dentry)) {
862 struct vfsmount *mounted;
863 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
864 if (!mounted)
865 return;
866 path->mnt = mounted;
867 path->dentry = mounted->mnt_root;
868 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
869 *inode = path->dentry->d_inode;
870 }
871}
872
463ffb2e
AV		 873static int __follow_mount(struct path *path)
874{
875 int res = 0;
876 while (d_mountpoint(path->dentry)) {
1c755af4 877 struct vfsmount *mounted = lookup_mnt(path);
463ffb2e
AV		 878 if (!mounted)
879 break;
880 dput(path->dentry);
881 if (res)
882 mntput(path->mnt);
883 path->mnt = mounted;
884 path->dentry = dget(mounted->mnt_root);
885 res = 1;
886 }
887 return res;
888}
889
79ed0226 890static void follow_mount(struct path *path)
1da177e4 891{
79ed0226 892 while (d_mountpoint(path->dentry)) {
1c755af4 893 struct vfsmount *mounted = lookup_mnt(path);
1da177e4
LT		 894 if (!mounted)
895 break;
79ed0226
AV		 896 dput(path->dentry);
897 mntput(path->mnt);
898 path->mnt = mounted;
899 path->dentry = dget(mounted->mnt_root);
1da177e4 900 }
1da177e4
LT		 901}
902
9393bd07 903int follow_down(struct path *path)
1da177e4
LT		 904{
905 struct vfsmount *mounted;
906
1c755af4 907 mounted = lookup_mnt(path);
1da177e4 908 if (mounted) {
9393bd07
AV		 909 dput(path->dentry);
910 mntput(path->mnt);
911 path->mnt = mounted;
912 path->dentry = dget(mounted->mnt_root);
1da177e4
LT		 913 return 1;
914 }
915 return 0;
916}
917
31e6b01f
NP		 918static int follow_dotdot_rcu(struct nameidata *nd)
919{
920 struct inode *inode = nd->inode;
921
922 set_root_rcu(nd);
923
924 while(1) {
925 if (nd->path.dentry == nd->root.dentry &&
926 nd->path.mnt == nd->root.mnt) {
927 break;
928 }
929 if (nd->path.dentry != nd->path.mnt->mnt_root) {
930 struct dentry *old = nd->path.dentry;
931 struct dentry *parent = old->d_parent;
932 unsigned seq;
933
934 seq = read_seqcount_begin(&parent->d_seq);
935 if (read_seqcount_retry(&old->d_seq, nd->seq))
936 return -ECHILD;
937 inode = parent->d_inode;
938 nd->path.dentry = parent;
939 nd->seq = seq;
940 break;
941 }
942 if (!follow_up_rcu(&nd->path))
943 break;
944 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
945 inode = nd->path.dentry->d_inode;
946 }
947 __follow_mount_rcu(nd, &nd->path, &inode);
948 nd->inode = inode;
949
950 return 0;
951}
952
953static void follow_dotdot(struct nameidata *nd)
1da177e4 954{
2a737871 955 set_root(nd);
e518ddb7 956
1da177e4 957 while(1) {
4ac91378 958 struct dentry *old = nd->path.dentry;
1da177e4 959
2a737871
AV		 960 if (nd->path.dentry == nd->root.dentry &&
961 nd->path.mnt == nd->root.mnt) {
1da177e4
LT		 962 break;
963 }
4ac91378 964 if (nd->path.dentry != nd->path.mnt->mnt_root) {
3088dd70
AV		 965 /* rare case of legitimate dget_parent()... */
966 nd->path.dentry = dget_parent(nd->path.dentry);
1da177e4
LT		 967 dput(old);
968 break;
969 }
3088dd70 970 if (!follow_up(&nd->path))
1da177e4 971 break;
1da177e4 972 }
79ed0226 973 follow_mount(&nd->path);
31e6b01f 974 nd->inode = nd->path.dentry->d_inode;
1da177e4
LT		 975}
976
baa03890
NP		 977/*
978 * Allocate a dentry with name and parent, and perform a parent
979 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
980 * on error. parent->d_inode->i_mutex must be held. d_lookup must
981 * have verified that no child exists while under i_mutex.
982 */
983static struct dentry *d_alloc_and_lookup(struct dentry *parent,
984 struct qstr *name, struct nameidata *nd)
985{
986 struct inode *inode = parent->d_inode;
987 struct dentry *dentry;
988 struct dentry *old;
989
990 /* Don't create child dentry for a dead directory. */
991 if (unlikely(IS_DEADDIR(inode)))
992 return ERR_PTR(-ENOENT);
993
994 dentry = d_alloc(parent, name);
995 if (unlikely(!dentry))
996 return ERR_PTR(-ENOMEM);
997
998 old = inode->i_op->lookup(inode, dentry, nd);
999 if (unlikely(old)) {
1000 dput(dentry);
1001 dentry = old;
1002 }
1003 return dentry;
1004}
1005
1da177e4
LT		 1006/*
1007 * It's more convoluted than I'd like it to be, but... it's still fairly
1008 * small and for now I'd prefer to have fast path as straight as possible.
1009 * It _is_ time-critical.
1010 */
1011static int do_lookup(struct nameidata *nd, struct qstr *name,
31e6b01f 1012 struct path *path, struct inode **inode)
1da177e4 1013{
4ac91378 1014 struct vfsmount *mnt = nd->path.mnt;
31e6b01f 1015 struct dentry *dentry, *parent = nd->path.dentry;
6e6b1bd1 1016 struct inode *dir;
3cac260a
AV		 1017 /*
1018 * See if the low-level filesystem might want
1019 * to use its own hash..
1020 */
fb045adb 1021 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
31e6b01f 1022 int err = parent->d_op->d_hash(parent, nd->inode, name);
3cac260a
AV		 1023 if (err < 0)
1024 return err;
1025 }
1da177e4 1026
b04f784e
NP		 1027 /*
1028 * Rename seqlock is not required here because in the off chance
1029 * of a false negative due to a concurrent rename, we're going to
1030 * do the non-racy lookup, below.
1031 */
31e6b01f
NP		 1032 if (nd->flags & LOOKUP_RCU) {
1033 unsigned seq;
1034
1035 *inode = nd->inode;
1036 dentry = __d_lookup_rcu(parent, name, &seq, inode);
1037 if (!dentry) {
1038 if (nameidata_drop_rcu(nd))
1039 return -ECHILD;
1040 goto need_lookup;
1041 }
1042 /* Memory barrier in read_seqcount_begin of child is enough */
1043 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1044 return -ECHILD;
1045
1046 nd->seq = seq;
34286d66 1047 if (dentry->d_flags & DCACHE_OP_REVALIDATE)
31e6b01f 1048 goto need_revalidate;
31e6b01f
NP		 1049 path->mnt = mnt;
1050 path->dentry = dentry;
1051 __follow_mount_rcu(nd, path, inode);
1052 } else {
1053 dentry = __d_lookup(parent, name);
1054 if (!dentry)
1055 goto need_lookup;
2e2e88ea 1056found:
fb045adb 1057 if (dentry->d_flags & DCACHE_OP_REVALIDATE)
31e6b01f 1058 goto need_revalidate;
1da177e4 1059done:
31e6b01f
NP		 1060 path->mnt = mnt;
1061 path->dentry = dentry;
1062 __follow_mount(path);
1063 *inode = path->dentry->d_inode;
1064 }
1da177e4
LT		 1065 return 0;
1066
1067need_lookup:
6e6b1bd1 1068 dir = parent->d_inode;
31e6b01f 1069 BUG_ON(nd->inode != dir);
6e6b1bd1
AV		 1070
1071 mutex_lock(&dir->i_mutex);
1072 /*
1073 * First re-do the cached lookup just in case it was created
b04f784e
NP		 1074 * while we waited for the directory semaphore, or the first
1075 * lookup failed due to an unrelated rename.
6e6b1bd1 1076 *
b04f784e
NP		 1077 * This could use version numbering or similar to avoid unnecessary
1078 * cache lookups, but then we'd have to do the first lookup in the
1079 * non-racy way. However in the common case here, everything should
1080 * be hot in cache, so would it be a big win?
6e6b1bd1
AV		 1081 */
1082 dentry = d_lookup(parent, name);
baa03890
NP		 1083 if (likely(!dentry)) {
1084 dentry = d_alloc_and_lookup(parent, name, nd);
6e6b1bd1
AV		 1085 mutex_unlock(&dir->i_mutex);
1086 if (IS_ERR(dentry))
1087 goto fail;
1088 goto done;
1089 }
6e6b1bd1
AV		 1090 /*
1091 * Uhhuh! Nasty case: the cache was re-populated while
1092 * we waited on the semaphore. Need to revalidate.
1093 */
1094 mutex_unlock(&dir->i_mutex);
2e2e88ea 1095 goto found;
1da177e4
LT		 1096
1097need_revalidate:
bcdc5e01
IK		 1098 dentry = do_revalidate(dentry, nd);
1099 if (!dentry)
1100 goto need_lookup;
1101 if (IS_ERR(dentry))
1102 goto fail;
1103 goto done;
1da177e4
LT		 1104
1105fail:
1106 return PTR_ERR(dentry);
1107}
1108
ac278a9c
AV		 1109/*
1110 * This is a temporary kludge to deal with "automount" symlinks; proper
1111 * solution is to trigger them on follow_mount(), so that do_lookup()
1112 * would DTRT. To be killed before 2.6.34-final.
1113 */
1114static inline int follow_on_final(struct inode *inode, unsigned lookup_flags)
1115{
1116 return inode && unlikely(inode->i_op->follow_link) &&
1117 ((lookup_flags & LOOKUP_FOLLOW) || S_ISDIR(inode->i_mode));
1118}
1119
1da177e4
LT		 1120/*
1121 * Name resolution.
ea3834d9
PM		 1122 * This is the basic name resolution function, turning a pathname into
1123 * the final dentry. We expect 'base' to be positive and a directory.
1da177e4 1124 *
ea3834d9
PM		 1125 * Returns 0 and nd will have valid dentry and mnt on success.
1126 * Returns error and drops reference to input namei data on failure.
1da177e4 1127 */
6de88d72 1128static int link_path_walk(const char *name, struct nameidata *nd)
1da177e4
LT		 1129{
1130 struct path next;
1da177e4
LT		 1131 int err;
1132 unsigned int lookup_flags = nd->flags;
1133
1134 while (*name=='/')
1135 name++;
1136 if (!*name)
1137 goto return_reval;
1138
1da177e4 1139 if (nd->depth)
f55eab82 1140 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
1da177e4
LT		 1141
1142 /* At this point we know we have a real path component. */
1143 for(;;) {
31e6b01f 1144 struct inode *inode;
1da177e4
LT		 1145 unsigned long hash;
1146 struct qstr this;
1147 unsigned int c;
1148
cdce5d6b 1149 nd->flags |= LOOKUP_CONTINUE;
31e6b01f 1150 if (nd->flags & LOOKUP_RCU) {
b74c79e9 1151 err = exec_permission(nd->inode, IPERM_FLAG_RCU);
31e6b01f
NP		 1152 if (err == -ECHILD) {
1153 if (nameidata_drop_rcu(nd))
1154 return -ECHILD;
1155 goto exec_again;
1156 }
1157 } else {
1158exec_again:
b74c79e9 1159 err = exec_permission(nd->inode, 0);
31e6b01f 1160 }
1da177e4
LT		 1161 if (err)
1162 break;
1163
1164 this.name = name;
1165 c = *(const unsigned char *)name;
1166
1167 hash = init_name_hash();
1168 do {
1169 name++;
1170 hash = partial_name_hash(c, hash);
1171 c = *(const unsigned char *)name;
1172 } while (c && (c != '/'));
1173 this.len = name - (const char *) this.name;
1174 this.hash = end_name_hash(hash);
1175
1176 /* remove trailing slashes? */
1177 if (!c)
1178 goto last_component;
1179 while (*++name == '/');
1180 if (!*name)
1181 goto last_with_slashes;
1182
1183 /*
1184 * "." and ".." are special - ".." especially so because it has
1185 * to be able to know about the current root directory and
1186 * parent relationships.
1187 */
1188 if (this.name[0] == '.') switch (this.len) {
1189 default:
1190 break;
31e6b01f 1191 case 2:
1da177e4
LT		 1192 if (this.name[1] != '.')
1193 break;
31e6b01f
NP		 1194 if (nd->flags & LOOKUP_RCU) {
1195 if (follow_dotdot_rcu(nd))
1196 return -ECHILD;
1197 } else
1198 follow_dotdot(nd);
1da177e4
LT		 1199 /* fallthrough */
1200 case 1:
1201 continue;
1202 }
1da177e4 1203 /* This does the actual lookups.. */
31e6b01f 1204 err = do_lookup(nd, &this, &next, &inode);
1da177e4
LT		 1205 if (err)
1206 break;
1da177e4 1207 err = -ENOENT;
1da177e4
LT		 1208 if (!inode)
1209 goto out_dput;
1da177e4
LT		 1210
1211 if (inode->i_op->follow_link) {
31e6b01f
NP		 1212 /* We commonly drop rcu-walk here */
1213 if (nameidata_dentry_drop_rcu_maybe(nd, next.dentry))
1214 return -ECHILD;
1215 BUG_ON(inode != next.dentry->d_inode);
90ebe565 1216 err = do_follow_link(&next, nd);
1da177e4
LT		 1217 if (err)
1218 goto return_err;
31e6b01f 1219 nd->inode = nd->path.dentry->d_inode;
1da177e4 1220 err = -ENOENT;
31e6b01f 1221 if (!nd->inode)
1da177e4 1222 break;
31e6b01f 1223 } else {
09dd17d3 1224 path_to_nameidata(&next, nd);
31e6b01f
NP		 1225 nd->inode = inode;
1226 }
1da177e4 1227 err = -ENOTDIR;
31e6b01f 1228 if (!nd->inode->i_op->lookup)
1da177e4
LT		 1229 break;
1230 continue;
1231 /* here ends the main loop */
1232
1233last_with_slashes:
1234 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1235last_component:
f55eab82
TM		 1236 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1237 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
1da177e4
LT		 1238 if (lookup_flags & LOOKUP_PARENT)
1239 goto lookup_parent;
1240 if (this.name[0] == '.') switch (this.len) {
1241 default:
1242 break;
31e6b01f 1243 case 2:
1da177e4
LT		 1244 if (this.name[1] != '.')
1245 break;
31e6b01f
NP		 1246 if (nd->flags & LOOKUP_RCU) {
1247 if (follow_dotdot_rcu(nd))
1248 return -ECHILD;
1249 } else
1250 follow_dotdot(nd);
1da177e4
LT		 1251 /* fallthrough */
1252 case 1:
1253 goto return_reval;
1254 }
31e6b01f 1255 err = do_lookup(nd, &this, &next, &inode);
1da177e4
LT		 1256 if (err)
1257 break;
ac278a9c 1258 if (follow_on_final(inode, lookup_flags)) {
31e6b01f
NP		 1259 if (nameidata_dentry_drop_rcu_maybe(nd, next.dentry))
1260 return -ECHILD;
1261 BUG_ON(inode != next.dentry->d_inode);
90ebe565 1262 err = do_follow_link(&next, nd);
1da177e4
LT		 1263 if (err)
1264 goto return_err;
31e6b01f
NP		 1265 nd->inode = nd->path.dentry->d_inode;
1266 } else {
09dd17d3 1267 path_to_nameidata(&next, nd);
31e6b01f
NP		 1268 nd->inode = inode;
1269 }
1da177e4 1270 err = -ENOENT;
31e6b01f 1271 if (!nd->inode)
1da177e4
LT		 1272 break;
1273 if (lookup_flags & LOOKUP_DIRECTORY) {
1274 err = -ENOTDIR;
31e6b01f 1275 if (!nd->inode->i_op->lookup)
1da177e4
LT		 1276 break;
1277 }
1278 goto return_base;
1279lookup_parent:
1280 nd->last = this;
1281 nd->last_type = LAST_NORM;
1282 if (this.name[0] != '.')
1283 goto return_base;
1284 if (this.len == 1)
1285 nd->last_type = LAST_DOT;
1286 else if (this.len == 2 && this.name[1] == '.')
1287 nd->last_type = LAST_DOTDOT;
1288 else
1289 goto return_base;
1290return_reval:
1291 /*
1292 * We bypassed the ordinary revalidation routines.
1293 * We may need to check the cached dentry for staleness.
1294 */
fb045adb 1295 if (need_reval_dot(nd->path.dentry)) {
1da177e4 1296 /* Note: we do not d_invalidate() */
34286d66
NP		 1297 err = d_revalidate(nd->path.dentry, nd);
1298 if (!err)
1299 err = -ESTALE;
1300 if (err < 0)
1da177e4
LT		 1301 break;
1302 }
1303return_base:
31e6b01f
NP		 1304 if (nameidata_drop_rcu_last_maybe(nd))
1305 return -ECHILD;
1da177e4
LT		 1306 return 0;
1307out_dput:
31e6b01f
NP		 1308 if (!(nd->flags & LOOKUP_RCU))
1309 path_put_conditional(&next, nd);
1da177e4
LT		 1310 break;
1311 }
31e6b01f
NP		 1312 if (!(nd->flags & LOOKUP_RCU))
1313 path_put(&nd->path);
1da177e4
LT		 1314return_err:
1315 return err;
1316}
1317
31e6b01f
NP		 1318static inline int path_walk_rcu(const char *name, struct nameidata *nd)
1319{
1320 current->total_link_count = 0;
1321
1322 return link_path_walk(name, nd);
1323}
1324
1325static inline int path_walk_simple(const char *name, struct nameidata *nd)
1326{
1327 current->total_link_count = 0;
1328
1329 return link_path_walk(name, nd);
1330}
1331
fc9b52cd 1332static int path_walk(const char *name, struct nameidata *nd)
1da177e4 1333{
6de88d72
AV		 1334 struct path save = nd->path;
1335 int result;
1336
1da177e4 1337 current->total_link_count = 0;
6de88d72
AV		 1338
1339 /* make sure the stuff we saved doesn't go away */
1340 path_get(&save);
1341
1342 result = link_path_walk(name, nd);
1343 if (result == -ESTALE) {
1344 /* nd->path had been dropped */
1345 current->total_link_count = 0;
1346 nd->path = save;
1347 path_get(&nd->path);
1348 nd->flags |= LOOKUP_REVAL;
1349 result = link_path_walk(name, nd);
1350 }
1351
1352 path_put(&save);
1353
1354 return result;
1da177e4
LT		 1355}
1356
31e6b01f
NP		 1357static void path_finish_rcu(struct nameidata *nd)
1358{
1359 if (nd->flags & LOOKUP_RCU) {
1360 /* RCU dangling. Cancel it. */
1361 nd->flags &= ~LOOKUP_RCU;
1362 nd->root.mnt = NULL;
1363 rcu_read_unlock();
1364 br_read_unlock(vfsmount_lock);
1365 }
1366 if (nd->file)
1367 fput(nd->file);
1368}
1369
1370static int path_init_rcu(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1371{
1372 int retval = 0;
1373 int fput_needed;
1374 struct file *file;
1375
1376 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1377 nd->flags = flags | LOOKUP_RCU;
1378 nd->depth = 0;
1379 nd->root.mnt = NULL;
1380 nd->file = NULL;
1381
1382 if (*name=='/') {
1383 struct fs_struct *fs = current->fs;
c28cc364 1384 unsigned seq;
31e6b01f
NP		 1385
1386 br_read_lock(vfsmount_lock);
1387 rcu_read_lock();
1388
c28cc364
NP		 1389 do {
1390 seq = read_seqcount_begin(&fs->seq);
1391 nd->root = fs->root;
1392 nd->path = nd->root;
1393 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1394 } while (read_seqcount_retry(&fs->seq, seq));
31e6b01f
NP		 1395
1396 } else if (dfd == AT_FDCWD) {
1397 struct fs_struct *fs = current->fs;
c28cc364 1398 unsigned seq;
31e6b01f
NP		 1399
1400 br_read_lock(vfsmount_lock);
1401 rcu_read_lock();
1402
c28cc364
NP		 1403 do {
1404 seq = read_seqcount_begin(&fs->seq);
1405 nd->path = fs->pwd;
1406 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1407 } while (read_seqcount_retry(&fs->seq, seq));
1408
31e6b01f
NP		 1409 } else {
1410 struct dentry *dentry;
1411
1412 file = fget_light(dfd, &fput_needed);
1413 retval = -EBADF;
1414 if (!file)
1415 goto out_fail;
1416
1417 dentry = file->f_path.dentry;
1418
1419 retval = -ENOTDIR;
1420 if (!S_ISDIR(dentry->d_inode->i_mode))
1421 goto fput_fail;
1422
1423 retval = file_permission(file, MAY_EXEC);
1424 if (retval)
1425 goto fput_fail;
1426
1427 nd->path = file->f_path;
1428 if (fput_needed)
1429 nd->file = file;
1430
c28cc364 1431 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
31e6b01f
NP		 1432 br_read_lock(vfsmount_lock);
1433 rcu_read_lock();
1434 }
1435 nd->inode = nd->path.dentry->d_inode;
1436 return 0;
1437
1438fput_fail:
1439 fput_light(file, fput_needed);
1440out_fail:
1441 return retval;
1442}
1443
9b4a9b14 1444static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1da177e4 1445{
ea3834d9 1446 int retval = 0;
170aa3d0
UD		 1447 int fput_needed;
1448 struct file *file;
1da177e4
LT		 1449
1450 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1451 nd->flags = flags;
1452 nd->depth = 0;
2a737871 1453 nd->root.mnt = NULL;
1da177e4 1454
1da177e4 1455 if (*name=='/') {
2a737871
AV		 1456 set_root(nd);
1457 nd->path = nd->root;
1458 path_get(&nd->root);
5590ff0d 1459 } else if (dfd == AT_FDCWD) {
f7ad3c6b 1460 get_fs_pwd(current->fs, &nd->path);
5590ff0d 1461 } else {
5590ff0d
UD		 1462 struct dentry *dentry;
1463
1464 file = fget_light(dfd, &fput_needed);
170aa3d0
UD		 1465 retval = -EBADF;
1466 if (!file)
6d09bb62 1467 goto out_fail;
5590ff0d 1468
0f7fc9e4 1469 dentry = file->f_path.dentry;
5590ff0d 1470
170aa3d0
UD		 1471 retval = -ENOTDIR;
1472 if (!S_ISDIR(dentry->d_inode->i_mode))
6d09bb62 1473 goto fput_fail;
5590ff0d
UD		 1474
1475 retval = file_permission(file, MAY_EXEC);
170aa3d0 1476 if (retval)
6d09bb62 1477 goto fput_fail;
5590ff0d 1478
5dd784d0
JB		 1479 nd->path = file->f_path;
1480 path_get(&file->f_path);
5590ff0d
UD		 1481
1482 fput_light(file, fput_needed);
1da177e4 1483 }
31e6b01f 1484 nd->inode = nd->path.dentry->d_inode;
9b4a9b14 1485 return 0;
2dfdd266 1486
9b4a9b14
AV		 1487fput_fail:
1488 fput_light(file, fput_needed);
1489out_fail:
1490 return retval;
1491}
1492
1493/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1494static int do_path_lookup(int dfd, const char *name,
1495 unsigned int flags, struct nameidata *nd)
1496{
31e6b01f
NP		 1497 int retval;
1498
1499 /*
1500 * Path walking is largely split up into 2 different synchronisation
1501 * schemes, rcu-walk and ref-walk (explained in
1502 * Documentation/filesystems/path-lookup.txt). These share much of the
1503 * path walk code, but some things particularly setup, cleanup, and
1504 * following mounts are sufficiently divergent that functions are
1505 * duplicated. Typically there is a function foo(), and its RCU
1506 * analogue, foo_rcu().
1507 *
1508 * -ECHILD is the error number of choice (just to avoid clashes) that
1509 * is returned if some aspect of an rcu-walk fails. Such an error must
1510 * be handled by restarting a traditional ref-walk (which will always
1511 * be able to complete).
1512 */
1513 retval = path_init_rcu(dfd, name, flags, nd);
1514 if (unlikely(retval))
1515 return retval;
1516 retval = path_walk_rcu(name, nd);
1517 path_finish_rcu(nd);
2a737871
AV		 1518 if (nd->root.mnt) {
1519 path_put(&nd->root);
1520 nd->root.mnt = NULL;
1521 }
31e6b01f
NP		 1522
1523 if (unlikely(retval == -ECHILD || retval == -ESTALE)) {
1524 /* slower, locked walk */
1525 if (retval == -ESTALE)
1526 flags |= LOOKUP_REVAL;
1527 retval = path_init(dfd, name, flags, nd);
1528 if (unlikely(retval))
1529 return retval;
1530 retval = path_walk(name, nd);
1531 if (nd->root.mnt) {
1532 path_put(&nd->root);
1533 nd->root.mnt = NULL;
1534 }
1535 }
1536
1537 if (likely(!retval)) {
1538 if (unlikely(!audit_dummy_context())) {
1539 if (nd->path.dentry && nd->inode)
1540 audit_inode(name, nd->path.dentry);
1541 }
1542 }
1543
170aa3d0 1544 return retval;
1da177e4
LT		 1545}
1546
fc9b52cd 1547int path_lookup(const char *name, unsigned int flags,
5590ff0d
UD		 1548 struct nameidata *nd)
1549{
1550 return do_path_lookup(AT_FDCWD, name, flags, nd);
1551}
1552
d1811465
AV		 1553int kern_path(const char *name, unsigned int flags, struct path *path)
1554{
1555 struct nameidata nd;
1556 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1557 if (!res)
1558 *path = nd.path;
1559 return res;
1560}
1561
16f18200
JJS		 1562/**
1563 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1564 * @dentry: pointer to dentry of the base directory
1565 * @mnt: pointer to vfs mount of the base directory
1566 * @name: pointer to file name
1567 * @flags: lookup flags
1568 * @nd: pointer to nameidata
1569 */
1570int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1571 const char *name, unsigned int flags,
1572 struct nameidata *nd)
1573{
1574 int retval;
1575
1576 /* same as do_path_lookup */
1577 nd->last_type = LAST_ROOT;
1578 nd->flags = flags;
1579 nd->depth = 0;
1580
c8e7f449
JB		 1581 nd->path.dentry = dentry;
1582 nd->path.mnt = mnt;
1583 path_get(&nd->path);
5b857119
AV		 1584 nd->root = nd->path;
1585 path_get(&nd->root);
31e6b01f 1586 nd->inode = nd->path.dentry->d_inode;
16f18200
JJS		 1587
1588 retval = path_walk(name, nd);
4ac91378 1589 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
31e6b01f 1590 nd->inode))
4ac91378 1591 audit_inode(name, nd->path.dentry);
16f18200 1592
5b857119
AV		 1593 path_put(&nd->root);
1594 nd->root.mnt = NULL;
16f18200 1595
2a737871 1596 return retval;
16f18200
JJS		 1597}
1598
eead1911
CH		 1599static struct dentry *__lookup_hash(struct qstr *name,
1600 struct dentry *base, struct nameidata *nd)
1da177e4 1601{
81fca444 1602 struct inode *inode = base->d_inode;
057f6c01 1603 struct dentry *dentry;
1da177e4
LT		 1604 int err;
1605
b74c79e9 1606 err = exec_permission(inode, 0);
81fca444
CH		 1607 if (err)
1608 return ERR_PTR(err);
1da177e4
LT		 1609
1610 /*
1611 * See if the low-level filesystem might want
1612 * to use its own hash..
1613 */
fb045adb 1614 if (base->d_flags & DCACHE_OP_HASH) {
b1e6a015 1615 err = base->d_op->d_hash(base, inode, name);
1da177e4
LT		 1616 dentry = ERR_PTR(err);
1617 if (err < 0)
1618 goto out;
1619 }
1620
b04f784e
NP		 1621 /*
1622 * Don't bother with __d_lookup: callers are for creat as
1623 * well as unlink, so a lot of the time it would cost
1624 * a double lookup.
6e6b1bd1 1625 */
b04f784e 1626 dentry = d_lookup(base, name);
6e6b1bd1 1627
fb045adb 1628 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE))
6e6b1bd1
AV		 1629 dentry = do_revalidate(dentry, nd);
1630
baa03890
NP		 1631 if (!dentry)
1632 dentry = d_alloc_and_lookup(base, name, nd);
1da177e4
LT		 1633out:
1634 return dentry;
1635}
1636
057f6c01
JM		 1637/*
1638 * Restricted form of lookup. Doesn't follow links, single-component only,
1639 * needs parent already locked. Doesn't follow mounts.
1640 * SMP-safe.
1641 */
eead1911 1642static struct dentry *lookup_hash(struct nameidata *nd)
057f6c01 1643{
4ac91378 1644 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1da177e4
LT		 1645}
1646
eead1911
CH		 1647static int __lookup_one_len(const char *name, struct qstr *this,
1648 struct dentry *base, int len)
1da177e4
LT		 1649{
1650 unsigned long hash;
1da177e4
LT		 1651 unsigned int c;
1652
057f6c01
JM		 1653 this->name = name;
1654 this->len = len;
1da177e4 1655 if (!len)
057f6c01 1656 return -EACCES;
1da177e4
LT		 1657
1658 hash = init_name_hash();
1659 while (len--) {
1660 c = *(const unsigned char *)name++;
1661 if (c == '/' || c == '\0')
057f6c01 1662 return -EACCES;
1da177e4
LT		 1663 hash = partial_name_hash(c, hash);
1664 }
057f6c01
JM		 1665 this->hash = end_name_hash(hash);
1666 return 0;
1667}
1da177e4 1668
eead1911 1669/**
a6b91919 1670 * lookup_one_len - filesystem helper to lookup single pathname component
eead1911
CH		 1671 * @name: pathname component to lookup
1672 * @base: base directory to lookup from
1673 * @len: maximum length @len should be interpreted to
1674 *
a6b91919
RD		 1675 * Note that this routine is purely a helper for filesystem usage and should
1676 * not be called by generic code. Also note that by using this function the
eead1911
CH		 1677 * nameidata argument is passed to the filesystem methods and a filesystem
1678 * using this helper needs to be prepared for that.
1679 */
057f6c01
JM		 1680struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1681{
1682 int err;
1683 struct qstr this;
1684
2f9092e1
DW		 1685 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1686
057f6c01 1687 err = __lookup_one_len(name, &this, base, len);
eead1911
CH		 1688 if (err)
1689 return ERR_PTR(err);
1690
49705b77 1691 return __lookup_hash(&this, base, NULL);
057f6c01
JM		 1692}
1693
2d8f3038
AV		 1694int user_path_at(int dfd, const char __user *name, unsigned flags,
1695 struct path *path)
1da177e4 1696{
2d8f3038 1697 struct nameidata nd;
1da177e4
LT		 1698 char *tmp = getname(name);
1699 int err = PTR_ERR(tmp);
1da177e4 1700 if (!IS_ERR(tmp)) {
2d8f3038
AV		 1701
1702 BUG_ON(flags & LOOKUP_PARENT);
1703
1704 err = do_path_lookup(dfd, tmp, flags, &nd);
1da177e4 1705 putname(tmp);
2d8f3038
AV		 1706 if (!err)
1707 *path = nd.path;
1da177e4
LT		 1708 }
1709 return err;
1710}
1711
2ad94ae6
AV		 1712static int user_path_parent(int dfd, const char __user *path,
1713 struct nameidata *nd, char **name)
1714{
1715 char *s = getname(path);
1716 int error;
1717
1718 if (IS_ERR(s))
1719 return PTR_ERR(s);
1720
1721 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1722 if (error)
1723 putname(s);
1724 else
1725 *name = s;
1726
1727 return error;
1728}
1729
1da177e4
LT		 1730/*
1731 * It's inline, so penalty for filesystems that don't use sticky bit is
1732 * minimal.
1733 */
1734static inline int check_sticky(struct inode *dir, struct inode *inode)
1735{
da9592ed
DH		 1736 uid_t fsuid = current_fsuid();
1737
1da177e4
LT		 1738 if (!(dir->i_mode & S_ISVTX))
1739 return 0;
da9592ed 1740 if (inode->i_uid == fsuid)
1da177e4 1741 return 0;
da9592ed 1742 if (dir->i_uid == fsuid)
1da177e4
LT		 1743 return 0;
1744 return !capable(CAP_FOWNER);
1745}
1746
1747/*
1748 * Check whether we can remove a link victim from directory dir, check
1749 * whether the type of victim is right.
1750 * 1. We can't do it if dir is read-only (done in permission())
1751 * 2. We should have write and exec permissions on dir
1752 * 3. We can't remove anything from append-only dir
1753 * 4. We can't do anything with immutable dir (done in permission())
1754 * 5. If the sticky bit on dir is set we should either
1755 * a. be owner of dir, or
1756 * b. be owner of victim, or
1757 * c. have CAP_FOWNER capability
1758 * 6. If the victim is append-only or immutable we can't do antyhing with
1759 * links pointing to it.
1760 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1761 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1762 * 9. We can't remove a root or mountpoint.
1763 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1764 * nfs_async_unlink().
1765 */
858119e1 1766static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1da177e4
LT		 1767{
1768 int error;
1769
1770 if (!victim->d_inode)
1771 return -ENOENT;
1772
1773 BUG_ON(victim->d_parent->d_inode != dir);
cccc6bba 1774 audit_inode_child(victim, dir);
1da177e4 1775
f419a2e3 1776 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1777 if (error)
1778 return error;
1779 if (IS_APPEND(dir))
1780 return -EPERM;
1781 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
f9454548 1782 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1da177e4
LT		 1783 return -EPERM;
1784 if (isdir) {
1785 if (!S_ISDIR(victim->d_inode->i_mode))
1786 return -ENOTDIR;
1787 if (IS_ROOT(victim))
1788 return -EBUSY;
1789 } else if (S_ISDIR(victim->d_inode->i_mode))
1790 return -EISDIR;
1791 if (IS_DEADDIR(dir))
1792 return -ENOENT;
1793 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1794 return -EBUSY;
1795 return 0;
1796}
1797
1798/* Check whether we can create an object with dentry child in directory
1799 * dir.
1800 * 1. We can't do it if child already exists (open has special treatment for
1801 * this case, but since we are inlined it's OK)
1802 * 2. We can't do it if dir is read-only (done in permission())
1803 * 3. We should have write and exec permissions on dir
1804 * 4. We can't do it if dir is immutable (done in permission())
1805 */
a95164d9 1806static inline int may_create(struct inode *dir, struct dentry *child)
1da177e4
LT		 1807{
1808 if (child->d_inode)
1809 return -EEXIST;
1810 if (IS_DEADDIR(dir))
1811 return -ENOENT;
f419a2e3 1812 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1813}
1814
1da177e4
LT		 1815/*
1816 * p1 and p2 should be directories on the same fs.
1817 */
1818struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1819{
1820 struct dentry *p;
1821
1822 if (p1 == p2) {
f2eace23 1823 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1da177e4
LT		 1824 return NULL;
1825 }
1826
a11f3a05 1827 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4 1828
e2761a11
OH		 1829 p = d_ancestor(p2, p1);
1830 if (p) {
1831 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1832 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1833 return p;
1da177e4
LT		 1834 }
1835
e2761a11
OH		 1836 p = d_ancestor(p1, p2);
1837 if (p) {
1838 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1839 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1840 return p;
1da177e4
LT		 1841 }
1842
f2eace23
IM		 1843 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1844 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1da177e4
LT		 1845 return NULL;
1846}
1847
1848void unlock_rename(struct dentry *p1, struct dentry *p2)
1849{
1b1dcc1b 1850 mutex_unlock(&p1->d_inode->i_mutex);
1da177e4 1851 if (p1 != p2) {
1b1dcc1b 1852 mutex_unlock(&p2->d_inode->i_mutex);
a11f3a05 1853 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4
LT		 1854 }
1855}
1856
1857int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1858 struct nameidata *nd)
1859{
a95164d9 1860 int error = may_create(dir, dentry);
1da177e4
LT		 1861
1862 if (error)
1863 return error;
1864
acfa4380 1865 if (!dir->i_op->create)
1da177e4
LT		 1866 return -EACCES; /* shouldn't it be ENOSYS? */
1867 mode &= S_IALLUGO;
1868 mode |= S_IFREG;
1869 error = security_inode_create(dir, dentry, mode);
1870 if (error)
1871 return error;
1da177e4 1872 error = dir->i_op->create(dir, dentry, mode, nd);
a74574aa 1873 if (!error)
f38aa942 1874 fsnotify_create(dir, dentry);
1da177e4
LT		 1875 return error;
1876}
1877
3fb64190 1878int may_open(struct path *path, int acc_mode, int flag)
1da177e4 1879{
3fb64190 1880 struct dentry *dentry = path->dentry;
1da177e4
LT		 1881 struct inode *inode = dentry->d_inode;
1882 int error;
1883
1884 if (!inode)
1885 return -ENOENT;
1886
c8fe8f30
CH		 1887 switch (inode->i_mode & S_IFMT) {
1888 case S_IFLNK:
1da177e4 1889 return -ELOOP;
c8fe8f30
CH		 1890 case S_IFDIR:
1891 if (acc_mode & MAY_WRITE)
1892 return -EISDIR;
1893 break;
1894 case S_IFBLK:
1895 case S_IFCHR:
3fb64190 1896 if (path->mnt->mnt_flags & MNT_NODEV)
1da177e4 1897 return -EACCES;
c8fe8f30
CH		 1898 /*FALLTHRU*/
1899 case S_IFIFO:
1900 case S_IFSOCK:
1da177e4 1901 flag &= ~O_TRUNC;
c8fe8f30 1902 break;
4a3fd211 1903 }
b41572e9 1904
3fb64190 1905 error = inode_permission(inode, acc_mode);
b41572e9
DH		 1906 if (error)
1907 return error;
6146f0d5 1908
1da177e4
LT		 1909 /*
1910 * An append-only file must be opened in append mode for writing.
1911 */
1912 if (IS_APPEND(inode)) {
8737c930 1913 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
7715b521 1914 return -EPERM;
1da177e4 1915 if (flag & O_TRUNC)
7715b521 1916 return -EPERM;
1da177e4
LT		 1917 }
1918
1919 /* O_NOATIME can only be set by the owner or superuser */
7715b521
AV		 1920 if (flag & O_NOATIME && !is_owner_or_cap(inode))
1921 return -EPERM;
1da177e4
LT		 1922
1923 /*
1924 * Ensure there are no outstanding leases on the file.
1925 */
b65a9cfc 1926 return break_lease(inode, flag);
7715b521 1927}
1da177e4 1928
7715b521
AV		 1929static int handle_truncate(struct path *path)
1930{
1931 struct inode *inode = path->dentry->d_inode;
1932 int error = get_write_access(inode);
1933 if (error)
1934 return error;
1935 /*
1936 * Refuse to truncate files with mandatory locks held on them.
1937 */
1938 error = locks_verify_locked(inode);
1939 if (!error)
ea0d3ab2 1940 error = security_path_truncate(path);
7715b521
AV		 1941 if (!error) {
1942 error = do_truncate(path->dentry, 0,
1943 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1944 NULL);
1945 }
1946 put_write_access(inode);
acd0c935 1947 return error;
1da177e4
LT		 1948}
1949
d57999e1
DH		 1950/*
1951 * Be careful about ever adding any more callers of this
1952 * function. Its flags must be in the namei format, not
1953 * what get passed to sys_open().
1954 */
1955static int __open_namei_create(struct nameidata *nd, struct path *path,
8737c930 1956 int open_flag, int mode)
aab520e2
DH		 1957{
1958 int error;
4ac91378 1959 struct dentry *dir = nd->path.dentry;
aab520e2
DH		 1960
1961 if (!IS_POSIXACL(dir->d_inode))
ce3b0f8d 1962 mode &= ~current_umask();
be6d3e56
KT		 1963 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1964 if (error)
1965 goto out_unlock;
aab520e2 1966 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
be6d3e56 1967out_unlock:
aab520e2 1968 mutex_unlock(&dir->d_inode->i_mutex);
4ac91378
JB		 1969 dput(nd->path.dentry);
1970 nd->path.dentry = path->dentry;
31e6b01f 1971
aab520e2
DH		 1972 if (error)
1973 return error;
1974 /* Don't check for write permission, don't truncate */
8737c930 1975 return may_open(&nd->path, 0, open_flag & ~O_TRUNC);
aab520e2
DH		 1976}
1977
d57999e1
DH		 1978/*
1979 * Note that while the flag value (low two bits) for sys_open means:
1980 * 00 - read-only
1981 * 01 - write-only
1982 * 10 - read-write
1983 * 11 - special
1984 * it is changed into
1985 * 00 - no permissions needed
1986 * 01 - read-permission
1987 * 10 - write-permission
1988 * 11 - read-write
1989 * for the internal routines (ie open_namei()/follow_link() etc)
1990 * This is more logical, and also allows the 00 "no perm needed"
1991 * to be used for symlinks (where the permissions are checked
1992 * later).
1993 *
1994*/
1995static inline int open_to_namei_flags(int flag)
1996{
1997 if ((flag+1) & O_ACCMODE)
1998 flag++;
1999 return flag;
2000}
2001
7715b521 2002static int open_will_truncate(int flag, struct inode *inode)
4a3fd211
DH		 2003{
2004 /*
2005 * We'll never write to the fs underlying
2006 * a device file.
2007 */
2008 if (special_file(inode->i_mode))
2009 return 0;
2010 return (flag & O_TRUNC);
2011}
2012
648fa861 2013static struct file *finish_open(struct nameidata *nd,
9a66179e 2014 int open_flag, int acc_mode)
648fa861
AV		 2015{
2016 struct file *filp;
2017 int will_truncate;
2018 int error;
2019
9a66179e 2020 will_truncate = open_will_truncate(open_flag, nd->path.dentry->d_inode);
648fa861
AV		 2021 if (will_truncate) {
2022 error = mnt_want_write(nd->path.mnt);
2023 if (error)
2024 goto exit;
2025 }
2026 error = may_open(&nd->path, acc_mode, open_flag);
2027 if (error) {
2028 if (will_truncate)
2029 mnt_drop_write(nd->path.mnt);
2030 goto exit;
2031 }
2032 filp = nameidata_to_filp(nd);
2033 if (!IS_ERR(filp)) {
2034 error = ima_file_check(filp, acc_mode);
2035 if (error) {
2036 fput(filp);
2037 filp = ERR_PTR(error);
2038 }
2039 }
2040 if (!IS_ERR(filp)) {
648fa861
AV		 2041 if (will_truncate) {
2042 error = handle_truncate(&nd->path);
2043 if (error) {
2044 fput(filp);
2045 filp = ERR_PTR(error);
2046 }
2047 }
2048 }
2049 /*
2050 * It is now safe to drop the mnt write
2051 * because the filp has had a write taken
2052 * on its behalf.
2053 */
2054 if (will_truncate)
2055 mnt_drop_write(nd->path.mnt);
d893f1bc 2056 path_put(&nd->path);
648fa861
AV		 2057 return filp;
2058
2059exit:
2060 if (!IS_ERR(nd->intent.open.file))
2061 release_open_intent(nd);
2062 path_put(&nd->path);
2063 return ERR_PTR(error);
2064}
2065
31e6b01f
NP		 2066/*
2067 * Handle O_CREAT case for do_filp_open
2068 */
fb1cc555 2069static struct file *do_last(struct nameidata *nd, struct path *path,
5b369df8 2070 int open_flag, int acc_mode,
3e297b61 2071 int mode, const char *pathname)
fb1cc555 2072{
a1e28038 2073 struct dentry *dir = nd->path.dentry;
fb1cc555 2074 struct file *filp;
1f36f774
AV		 2075 int error = -EISDIR;
2076
2077 switch (nd->last_type) {
2078 case LAST_DOTDOT:
2079 follow_dotdot(nd);
2080 dir = nd->path.dentry;
176306f5 2081 case LAST_DOT:
fb045adb 2082 if (need_reval_dot(dir)) {
34286d66
NP		 2083 error = d_revalidate(nd->path.dentry, nd);
2084 if (!error)
1f36f774 2085 error = -ESTALE;
34286d66 2086 if (error < 0)
1f36f774 2087 goto exit;
1f36f774
AV		 2088 }
2089 /* fallthrough */
1f36f774 2090 case LAST_ROOT:
31e6b01f 2091 goto exit;
1f36f774
AV		 2092 case LAST_BIND:
2093 audit_inode(pathname, dir);
67ee3ad2 2094 goto ok;
1f36f774 2095 }
67ee3ad2 2096
1f36f774 2097 /* trailing slashes? */
31e6b01f
NP		 2098 if (nd->last.name[nd->last.len])
2099 goto exit;
a2c36b45 2100
a1e28038
AV		 2101 mutex_lock(&dir->d_inode->i_mutex);
2102
2103 path->dentry = lookup_hash(nd);
2104 path->mnt = nd->path.mnt;
2105
fb1cc555
AV		 2106 error = PTR_ERR(path->dentry);
2107 if (IS_ERR(path->dentry)) {
2108 mutex_unlock(&dir->d_inode->i_mutex);
2109 goto exit;
2110 }
2111
2112 if (IS_ERR(nd->intent.open.file)) {
2113 error = PTR_ERR(nd->intent.open.file);
2114 goto exit_mutex_unlock;
2115 }
2116
2117 /* Negative dentry, just create the file */
2118 if (!path->dentry->d_inode) {
2119 /*
2120 * This write is needed to ensure that a
2121 * ro->rw transition does not occur between
2122 * the time when the file is created and when
2123 * a permanent write count is taken through
2124 * the 'struct file' in nameidata_to_filp().
2125 */
2126 error = mnt_want_write(nd->path.mnt);
2127 if (error)
2128 goto exit_mutex_unlock;
2129 error = __open_namei_create(nd, path, open_flag, mode);
2130 if (error) {
2131 mnt_drop_write(nd->path.mnt);
2132 goto exit;
2133 }
2134 filp = nameidata_to_filp(nd);
2135 mnt_drop_write(nd->path.mnt);
d893f1bc 2136 path_put(&nd->path);
fb1cc555
AV		 2137 if (!IS_ERR(filp)) {
2138 error = ima_file_check(filp, acc_mode);
2139 if (error) {
2140 fput(filp);
2141 filp = ERR_PTR(error);
2142 }
2143 }
2144 return filp;
2145 }
2146
2147 /*
2148 * It already exists.
2149 */
2150 mutex_unlock(&dir->d_inode->i_mutex);
2151 audit_inode(pathname, path->dentry);
2152
2153 error = -EEXIST;
5b369df8 2154 if (open_flag & O_EXCL)
fb1cc555
AV		 2155 goto exit_dput;
2156
2157 if (__follow_mount(path)) {
2158 error = -ELOOP;
5b369df8 2159 if (open_flag & O_NOFOLLOW)
fb1cc555
AV		 2160 goto exit_dput;
2161 }
2162
2163 error = -ENOENT;
2164 if (!path->dentry->d_inode)
2165 goto exit_dput;
9e67f361
AV		 2166
2167 if (path->dentry->d_inode->i_op->follow_link)
fb1cc555 2168 return NULL;
fb1cc555
AV		 2169
2170 path_to_nameidata(path, nd);
31e6b01f 2171 nd->inode = path->dentry->d_inode;
fb1cc555 2172 error = -EISDIR;
31e6b01f 2173 if (S_ISDIR(nd->inode->i_mode))
fb1cc555 2174 goto exit;
67ee3ad2 2175ok:
9a66179e 2176 filp = finish_open(nd, open_flag, acc_mode);
fb1cc555
AV		 2177 return filp;
2178
2179exit_mutex_unlock:
2180 mutex_unlock(&dir->d_inode->i_mutex);
2181exit_dput:
2182 path_put_conditional(path, nd);
2183exit:
2184 if (!IS_ERR(nd->intent.open.file))
2185 release_open_intent(nd);
fb1cc555
AV		 2186 path_put(&nd->path);
2187 return ERR_PTR(error);
2188}
2189
1da177e4 2190/*
4a3fd211
DH		 2191 * Note that the low bits of the passed in "open_flag"
2192 * are not the same as in the local variable "flag". See
2193 * open_to_namei_flags() for more details.
1da177e4 2194 */
a70e65df 2195struct file *do_filp_open(int dfd, const char *pathname,
6e8341a1 2196 int open_flag, int mode, int acc_mode)
1da177e4 2197{
4a3fd211 2198 struct file *filp;
a70e65df 2199 struct nameidata nd;
6e8341a1 2200 int error;
9850c056 2201 struct path path;
1da177e4 2202 int count = 0;
d57999e1 2203 int flag = open_to_namei_flags(open_flag);
31e6b01f 2204 int flags;
1f36f774
AV		 2205
2206 if (!(open_flag & O_CREAT))
2207 mode = 0;
1da177e4 2208
b1085ba8
LS		 2209 /* Must never be set by userspace */
2210 open_flag &= ~FMODE_NONOTIFY;
2211
6b2f3d1f
CH		 2212 /*
2213 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
2214 * check for O_DSYNC if the need any syncing at all we enforce it's
2215 * always set instead of having to deal with possibly weird behaviour
2216 * for malicious applications setting only __O_SYNC.
2217 */
2218 if (open_flag & __O_SYNC)
2219 open_flag |= O_DSYNC;
2220
6e8341a1 2221 if (!acc_mode)
6d125529 2222 acc_mode = MAY_OPEN | ACC_MODE(open_flag);
1da177e4 2223
834f2a4a 2224 /* O_TRUNC implies we need access checks for write permissions */
4296e2cb 2225 if (open_flag & O_TRUNC)
834f2a4a
TM		 2226 acc_mode |= MAY_WRITE;
2227
1da177e4
LT		 2228 /* Allow the LSM permission hook to distinguish append
2229 access from general write access. */
4296e2cb 2230 if (open_flag & O_APPEND)
1da177e4
LT		 2231 acc_mode |= MAY_APPEND;
2232
31e6b01f
NP		 2233 flags = LOOKUP_OPEN;
2234 if (open_flag & O_CREAT) {
2235 flags |= LOOKUP_CREATE;
2236 if (open_flag & O_EXCL)
2237 flags |= LOOKUP_EXCL;
2238 }
2239 if (open_flag & O_DIRECTORY)
2240 flags |= LOOKUP_DIRECTORY;
2241 if (!(open_flag & O_NOFOLLOW))
2242 flags |= LOOKUP_FOLLOW;
2243
2244 filp = get_empty_filp();
2245 if (!filp)
2246 return ERR_PTR(-ENFILE);
2247
2248 filp->f_flags = open_flag;
2249 nd.intent.open.file = filp;
2250 nd.intent.open.flags = flag;
2251 nd.intent.open.create_mode = mode;
2252
2253 if (open_flag & O_CREAT)
2254 goto creat;
2255
2256 /* !O_CREAT, simple open */
2257 error = do_path_lookup(dfd, pathname, flags, &nd);
2258 if (unlikely(error))
2259 goto out_filp;
2260 error = -ELOOP;
2261 if (!(nd.flags & LOOKUP_FOLLOW)) {
2262 if (nd.inode->i_op->follow_link)
2263 goto out_path;
2264 }
2265 error = -ENOTDIR;
2266 if (nd.flags & LOOKUP_DIRECTORY) {
2267 if (!nd.inode->i_op->lookup)
2268 goto out_path;
2269 }
2270 audit_inode(pathname, nd.path.dentry);
2271 filp = finish_open(&nd, open_flag, acc_mode);
2272 return filp;
2273
2274creat:
2275 /* OK, have to create the file. Find the parent. */
2276 error = path_init_rcu(dfd, pathname,
2277 LOOKUP_PARENT | (flags & LOOKUP_REVAL), &nd);
1da177e4 2278 if (error)
31e6b01f
NP		 2279 goto out_filp;
2280 error = path_walk_rcu(pathname, &nd);
2281 path_finish_rcu(&nd);
2282 if (unlikely(error == -ECHILD || error == -ESTALE)) {
2283 /* slower, locked walk */
2284 if (error == -ESTALE) {
2285reval:
2286 flags |= LOOKUP_REVAL;
2287 }
2288 error = path_init(dfd, pathname,
2289 LOOKUP_PARENT | (flags & LOOKUP_REVAL), &nd);
2290 if (error)
2291 goto out_filp;
3866248e 2292
31e6b01f 2293 error = path_walk_simple(pathname, &nd);
654f562c 2294 }
31e6b01f
NP		 2295 if (unlikely(error))
2296 goto out_filp;
2297 if (unlikely(!audit_dummy_context()))
9b4a9b14 2298 audit_inode(pathname, nd.path.dentry);
1da177e4
LT		 2299
2300 /*
a2c36b45 2301 * We have the parent and last component.
1da177e4 2302 */
31e6b01f 2303 nd.flags = flags;
3e297b61 2304 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
806b681c 2305 while (unlikely(!filp)) { /* trailing symlink */
def4af30 2306 struct path holder;
def4af30 2307 void *cookie;
806b681c 2308 error = -ELOOP;
1f36f774 2309 /* S_ISDIR part is a temporary automount kludge */
31e6b01f 2310 if (!(nd.flags & LOOKUP_FOLLOW) && !S_ISDIR(nd.inode->i_mode))
1f36f774
AV		 2311 goto exit_dput;
2312 if (count++ == 32)
806b681c
AV		 2313 goto exit_dput;
2314 /*
2315 * This is subtle. Instead of calling do_follow_link() we do
2316 * the thing by hands. The reason is that this way we have zero
2317 * link_count and path_walk() (called from ->follow_link)
2318 * honoring LOOKUP_PARENT. After that we have the parent and
2319 * last component, i.e. we are in the same situation as after
2320 * the first path_walk(). Well, almost - if the last component
2321 * is normal we get its copy stored in nd->last.name and we will
2322 * have to putname() it when we are done. Procfs-like symlinks
2323 * just set LAST_BIND.
2324 */
2325 nd.flags |= LOOKUP_PARENT;
2326 error = security_inode_follow_link(path.dentry, &nd);
2327 if (error)
2328 goto exit_dput;
def4af30
AV		 2329 error = __do_follow_link(&path, &nd, &cookie);
2330 if (unlikely(error)) {
31e6b01f
NP		 2331 if (!IS_ERR(cookie) && nd.inode->i_op->put_link)
2332 nd.inode->i_op->put_link(path.dentry, &nd, cookie);
806b681c 2333 /* nd.path had been dropped */
31e6b01f
NP		 2334 nd.path = path;
2335 goto out_path;
806b681c 2336 }
def4af30 2337 holder = path;
806b681c 2338 nd.flags &= ~LOOKUP_PARENT;
3e297b61 2339 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
31e6b01f
NP		 2340 if (nd.inode->i_op->put_link)
2341 nd.inode->i_op->put_link(holder.dentry, &nd, cookie);
def4af30 2342 path_put(&holder);
806b681c 2343 }
10fa8e62 2344out:
2a737871
AV		 2345 if (nd.root.mnt)
2346 path_put(&nd.root);
31e6b01f 2347 if (filp == ERR_PTR(-ESTALE) && !(flags & LOOKUP_REVAL))
10fa8e62 2348 goto reval;
10fa8e62 2349 return filp;
1da177e4 2350
806b681c
AV		 2351exit_dput:
2352 path_put_conditional(&path, &nd);
31e6b01f
NP		 2353out_path:
2354 path_put(&nd.path);
2355out_filp:
806b681c 2356 if (!IS_ERR(nd.intent.open.file))
a70e65df 2357 release_open_intent(&nd);
806b681c 2358 filp = ERR_PTR(error);
10fa8e62 2359 goto out;
1da177e4
LT		 2360}
2361
a70e65df
CH		 2362/**
2363 * filp_open - open file and return file pointer
2364 *
2365 * @filename: path to open
2366 * @flags: open flags as per the open(2) second argument
2367 * @mode: mode for the new file if O_CREAT is set, else ignored
2368 *
2369 * This is the helper to open a file from kernelspace if you really
2370 * have to. But in generally you should not do this, so please move
2371 * along, nothing to see here..
2372 */
2373struct file *filp_open(const char *filename, int flags, int mode)
2374{
6e8341a1 2375 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
a70e65df
CH		 2376}
2377EXPORT_SYMBOL(filp_open);
2378
1da177e4
LT		 2379/**
2380 * lookup_create - lookup a dentry, creating it if it doesn't exist
2381 * @nd: nameidata info
2382 * @is_dir: directory flag
2383 *
2384 * Simple function to lookup and return a dentry and create it
2385 * if it doesn't exist. Is SMP-safe.
c663e5d8 2386 *
4ac91378 2387 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1da177e4
LT		 2388 */
2389struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2390{
c663e5d8 2391 struct dentry *dentry = ERR_PTR(-EEXIST);
1da177e4 2392
4ac91378 2393 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
c663e5d8
CH		 2394 /*
2395 * Yucky last component or no last component at all?
2396 * (foo/., foo/.., /////)
2397 */
1da177e4
LT		 2398 if (nd->last_type != LAST_NORM)
2399 goto fail;
2400 nd->flags &= ~LOOKUP_PARENT;
3516586a 2401 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
a634904a 2402 nd->intent.open.flags = O_EXCL;
c663e5d8
CH		 2403
2404 /*
2405 * Do the final lookup.
2406 */
49705b77 2407 dentry = lookup_hash(nd);
1da177e4
LT		 2408 if (IS_ERR(dentry))
2409 goto fail;
c663e5d8 2410
e9baf6e5
AV		 2411 if (dentry->d_inode)
2412 goto eexist;
c663e5d8
CH		 2413 /*
2414 * Special case - lookup gave negative, but... we had foo/bar/
2415 * From the vfs_mknod() POV we just have a negative dentry -
2416 * all is fine. Let's be bastards - you had / on the end, you've
2417 * been asking for (non-existent) directory. -ENOENT for you.
2418 */
e9baf6e5
AV		 2419 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2420 dput(dentry);
2421 dentry = ERR_PTR(-ENOENT);
2422 }
1da177e4 2423 return dentry;
e9baf6e5 2424eexist:
1da177e4 2425 dput(dentry);
e9baf6e5 2426 dentry = ERR_PTR(-EEXIST);
1da177e4
LT		 2427fail:
2428 return dentry;
2429}
f81a0bff 2430EXPORT_SYMBOL_GPL(lookup_create);
1da177e4
LT		 2431
2432int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2433{
a95164d9 2434 int error = may_create(dir, dentry);
1da177e4
LT		 2435
2436 if (error)
2437 return error;
2438
2439 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2440 return -EPERM;
2441
acfa4380 2442 if (!dir->i_op->mknod)
1da177e4
LT		 2443 return -EPERM;
2444
08ce5f16
SH		 2445 error = devcgroup_inode_mknod(mode, dev);
2446 if (error)
2447 return error;
2448
1da177e4
LT		 2449 error = security_inode_mknod(dir, dentry, mode, dev);
2450 if (error)
2451 return error;
2452
1da177e4 2453 error = dir->i_op->mknod(dir, dentry, mode, dev);
a74574aa 2454 if (!error)
f38aa942 2455 fsnotify_create(dir, dentry);
1da177e4
LT		 2456 return error;
2457}
2458
463c3197
DH		 2459static int may_mknod(mode_t mode)
2460{
2461 switch (mode & S_IFMT) {
2462 case S_IFREG:
2463 case S_IFCHR:
2464 case S_IFBLK:
2465 case S_IFIFO:
2466 case S_IFSOCK:
2467 case 0: /* zero mode translates to S_IFREG */
2468 return 0;
2469 case S_IFDIR:
2470 return -EPERM;
2471 default:
2472 return -EINVAL;
2473 }
2474}
2475
2e4d0924
HC		 2476SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2477 unsigned, dev)
1da177e4 2478{
2ad94ae6
AV		 2479 int error;
2480 char *tmp;
2481 struct dentry *dentry;
1da177e4
LT		 2482 struct nameidata nd;
2483
2484 if (S_ISDIR(mode))
2485 return -EPERM;
1da177e4 2486
2ad94ae6 2487 error = user_path_parent(dfd, filename, &nd, &tmp);
1da177e4 2488 if (error)
2ad94ae6
AV		 2489 return error;
2490
1da177e4 2491 dentry = lookup_create(&nd, 0);
463c3197
DH		 2492 if (IS_ERR(dentry)) {
2493 error = PTR_ERR(dentry);
2494 goto out_unlock;
2495 }
4ac91378 2496 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2497 mode &= ~current_umask();
463c3197
DH		 2498 error = may_mknod(mode);
2499 if (error)
2500 goto out_dput;
2501 error = mnt_want_write(nd.path.mnt);
2502 if (error)
2503 goto out_dput;
be6d3e56
KT		 2504 error = security_path_mknod(&nd.path, dentry, mode, dev);
2505 if (error)
2506 goto out_drop_write;
463c3197 2507 switch (mode & S_IFMT) {
1da177e4 2508 case 0: case S_IFREG:
4ac91378 2509 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
1da177e4
LT		 2510 break;
2511 case S_IFCHR: case S_IFBLK:
4ac91378 2512 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
1da177e4
LT		 2513 new_decode_dev(dev));
2514 break;
2515 case S_IFIFO: case S_IFSOCK:
4ac91378 2516 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
1da177e4 2517 break;
1da177e4 2518 }
be6d3e56 2519out_drop_write:
463c3197
DH		 2520 mnt_drop_write(nd.path.mnt);
2521out_dput:
2522 dput(dentry);
2523out_unlock:
4ac91378 2524 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2525 path_put(&nd.path);
1da177e4
LT		 2526 putname(tmp);
2527
2528 return error;
2529}
2530
3480b257 2531SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
5590ff0d
UD		 2532{
2533 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2534}
2535
1da177e4
LT		 2536int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2537{
a95164d9 2538 int error = may_create(dir, dentry);
1da177e4
LT		 2539
2540 if (error)
2541 return error;
2542
acfa4380 2543 if (!dir->i_op->mkdir)
1da177e4
LT		 2544 return -EPERM;
2545
2546 mode &= (S_IRWXUGO|S_ISVTX);
2547 error = security_inode_mkdir(dir, dentry, mode);
2548 if (error)
2549 return error;
2550
1da177e4 2551 error = dir->i_op->mkdir(dir, dentry, mode);
a74574aa 2552 if (!error)
f38aa942 2553 fsnotify_mkdir(dir, dentry);
1da177e4
LT		 2554 return error;
2555}
2556
2e4d0924 2557SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
1da177e4
LT		 2558{
2559 int error = 0;
2560 char * tmp;
6902d925
DH		 2561 struct dentry *dentry;
2562 struct nameidata nd;
1da177e4 2563
2ad94ae6
AV		 2564 error = user_path_parent(dfd, pathname, &nd, &tmp);
2565 if (error)
6902d925 2566 goto out_err;
1da177e4 2567
6902d925
DH		 2568 dentry = lookup_create(&nd, 1);
2569 error = PTR_ERR(dentry);
2570 if (IS_ERR(dentry))
2571 goto out_unlock;
1da177e4 2572
4ac91378 2573 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2574 mode &= ~current_umask();
463c3197
DH		 2575 error = mnt_want_write(nd.path.mnt);
2576 if (error)
2577 goto out_dput;
be6d3e56
KT		 2578 error = security_path_mkdir(&nd.path, dentry, mode);
2579 if (error)
2580 goto out_drop_write;
4ac91378 2581 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
be6d3e56 2582out_drop_write:
463c3197
DH		 2583 mnt_drop_write(nd.path.mnt);
2584out_dput:
6902d925
DH		 2585 dput(dentry);
2586out_unlock:
4ac91378 2587 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2588 path_put(&nd.path);
6902d925
DH		 2589 putname(tmp);
2590out_err:
1da177e4
LT		 2591 return error;
2592}
2593
3cdad428 2594SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
5590ff0d
UD		 2595{
2596 return sys_mkdirat(AT_FDCWD, pathname, mode);
2597}
2598
1da177e4
LT		 2599/*
2600 * We try to drop the dentry early: we should have
2601 * a usage count of 2 if we're the only user of this
2602 * dentry, and if that is true (possibly after pruning
2603 * the dcache), then we drop the dentry now.
2604 *
2605 * A low-level filesystem can, if it choses, legally
2606 * do a
2607 *
2608 * if (!d_unhashed(dentry))
2609 * return -EBUSY;
2610 *
2611 * if it cannot handle the case of removing a directory
2612 * that is still in use by something else..
2613 */
2614void dentry_unhash(struct dentry *dentry)
2615{
2616 dget(dentry);
dc168427 2617 shrink_dcache_parent(dentry);
1da177e4 2618 spin_lock(&dentry->d_lock);
b7ab39f6 2619 if (dentry->d_count == 2)
1da177e4
LT		 2620 __d_drop(dentry);
2621 spin_unlock(&dentry->d_lock);
1da177e4
LT		 2622}
2623
2624int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2625{
2626 int error = may_delete(dir, dentry, 1);
2627
2628 if (error)
2629 return error;
2630
acfa4380 2631 if (!dir->i_op->rmdir)
1da177e4
LT		 2632 return -EPERM;
2633
1b1dcc1b 2634 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2635 dentry_unhash(dentry);
2636 if (d_mountpoint(dentry))
2637 error = -EBUSY;
2638 else {
2639 error = security_inode_rmdir(dir, dentry);
2640 if (!error) {
2641 error = dir->i_op->rmdir(dir, dentry);
d83c49f3 2642 if (!error) {
1da177e4 2643 dentry->d_inode->i_flags |= S_DEAD;
d83c49f3
AV		 2644 dont_mount(dentry);
2645 }
1da177e4
LT		 2646 }
2647 }
1b1dcc1b 2648 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4 2649 if (!error) {
1da177e4
LT		 2650 d_delete(dentry);
2651 }
2652 dput(dentry);
2653
2654 return error;
2655}
2656
5590ff0d 2657static long do_rmdir(int dfd, const char __user *pathname)
1da177e4
LT		 2658{
2659 int error = 0;
2660 char * name;
2661 struct dentry *dentry;
2662 struct nameidata nd;
2663
2ad94ae6 2664 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2665 if (error)
2ad94ae6 2666 return error;
1da177e4
LT		 2667
2668 switch(nd.last_type) {
0612d9fb
OH		 2669 case LAST_DOTDOT:
2670 error = -ENOTEMPTY;
2671 goto exit1;
2672 case LAST_DOT:
2673 error = -EINVAL;
2674 goto exit1;
2675 case LAST_ROOT:
2676 error = -EBUSY;
2677 goto exit1;
1da177e4 2678 }
0612d9fb
OH		 2679
2680 nd.flags &= ~LOOKUP_PARENT;
2681
4ac91378 2682 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2683 dentry = lookup_hash(&nd);
1da177e4 2684 error = PTR_ERR(dentry);
6902d925
DH		 2685 if (IS_ERR(dentry))
2686 goto exit2;
0622753b
DH		 2687 error = mnt_want_write(nd.path.mnt);
2688 if (error)
2689 goto exit3;
be6d3e56
KT		 2690 error = security_path_rmdir(&nd.path, dentry);
2691 if (error)
2692 goto exit4;
4ac91378 2693 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
be6d3e56 2694exit4:
0622753b
DH		 2695 mnt_drop_write(nd.path.mnt);
2696exit3:
6902d925
DH		 2697 dput(dentry);
2698exit2:
4ac91378 2699 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2700exit1:
1d957f9b 2701 path_put(&nd.path);
1da177e4
LT		 2702 putname(name);
2703 return error;
2704}
2705
3cdad428 2706SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
5590ff0d
UD		 2707{
2708 return do_rmdir(AT_FDCWD, pathname);
2709}
2710
1da177e4
LT		 2711int vfs_unlink(struct inode *dir, struct dentry *dentry)
2712{
2713 int error = may_delete(dir, dentry, 0);
2714
2715 if (error)
2716 return error;
2717
acfa4380 2718 if (!dir->i_op->unlink)
1da177e4
LT		 2719 return -EPERM;
2720
1b1dcc1b 2721 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2722 if (d_mountpoint(dentry))
2723 error = -EBUSY;
2724 else {
2725 error = security_inode_unlink(dir, dentry);
bec1052e 2726 if (!error) {
1da177e4 2727 error = dir->i_op->unlink(dir, dentry);
bec1052e 2728 if (!error)
d83c49f3 2729 dont_mount(dentry);
bec1052e 2730 }
1da177e4 2731 }
1b1dcc1b 2732 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2733
2734 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2735 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
ece95912 2736 fsnotify_link_count(dentry->d_inode);
e234f35c 2737 d_delete(dentry);
1da177e4 2738 }
0eeca283 2739
1da177e4
LT		 2740 return error;
2741}
2742
2743/*
2744 * Make sure that the actual truncation of the file will occur outside its
1b1dcc1b 2745 * directory's i_mutex. Truncate can take a long time if there is a lot of
1da177e4
LT		 2746 * writeout happening, and we don't want to prevent access to the directory
2747 * while waiting on the I/O.
2748 */
5590ff0d 2749static long do_unlinkat(int dfd, const char __user *pathname)
1da177e4 2750{
2ad94ae6
AV		 2751 int error;
2752 char *name;
1da177e4
LT		 2753 struct dentry *dentry;
2754 struct nameidata nd;
2755 struct inode *inode = NULL;
2756
2ad94ae6 2757 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2758 if (error)
2ad94ae6
AV		 2759 return error;
2760
1da177e4
LT		 2761 error = -EISDIR;
2762 if (nd.last_type != LAST_NORM)
2763 goto exit1;
0612d9fb
OH		 2764
2765 nd.flags &= ~LOOKUP_PARENT;
2766
4ac91378 2767 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2768 dentry = lookup_hash(&nd);
1da177e4
LT		 2769 error = PTR_ERR(dentry);
2770 if (!IS_ERR(dentry)) {
2771 /* Why not before? Because we want correct error value */
2772 if (nd.last.name[nd.last.len])
2773 goto slashes;
2774 inode = dentry->d_inode;
2775 if (inode)
7de9c6ee 2776 ihold(inode);
0622753b
DH		 2777 error = mnt_want_write(nd.path.mnt);
2778 if (error)
2779 goto exit2;
be6d3e56
KT		 2780 error = security_path_unlink(&nd.path, dentry);
2781 if (error)
2782 goto exit3;
4ac91378 2783 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
be6d3e56 2784exit3:
0622753b 2785 mnt_drop_write(nd.path.mnt);
1da177e4
LT		 2786 exit2:
2787 dput(dentry);
2788 }
4ac91378 2789 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4
LT		 2790 if (inode)
2791 iput(inode); /* truncate the inode here */
2792exit1:
1d957f9b 2793 path_put(&nd.path);
1da177e4
LT		 2794 putname(name);
2795 return error;
2796
2797slashes:
2798 error = !dentry->d_inode ? -ENOENT :
2799 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2800 goto exit2;
2801}
2802
2e4d0924 2803SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
5590ff0d
UD		 2804{
2805 if ((flag & ~AT_REMOVEDIR) != 0)
2806 return -EINVAL;
2807
2808 if (flag & AT_REMOVEDIR)
2809 return do_rmdir(dfd, pathname);
2810
2811 return do_unlinkat(dfd, pathname);
2812}
2813
3480b257 2814SYSCALL_DEFINE1(unlink, const char __user *, pathname)
5590ff0d
UD		 2815{
2816 return do_unlinkat(AT_FDCWD, pathname);
2817}
2818
db2e747b 2819int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
1da177e4 2820{
a95164d9 2821 int error = may_create(dir, dentry);
1da177e4
LT		 2822
2823 if (error)
2824 return error;
2825
acfa4380 2826 if (!dir->i_op->symlink)
1da177e4
LT		 2827 return -EPERM;
2828
2829 error = security_inode_symlink(dir, dentry, oldname);
2830 if (error)
2831 return error;
2832
1da177e4 2833 error = dir->i_op->symlink(dir, dentry, oldname);
a74574aa 2834 if (!error)
f38aa942 2835 fsnotify_create(dir, dentry);
1da177e4
LT		 2836 return error;
2837}
2838
2e4d0924
HC		 2839SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2840 int, newdfd, const char __user *, newname)
1da177e4 2841{
2ad94ae6
AV		 2842 int error;
2843 char *from;
2844 char *to;
6902d925
DH		 2845 struct dentry *dentry;
2846 struct nameidata nd;
1da177e4
LT		 2847
2848 from = getname(oldname);
2ad94ae6 2849 if (IS_ERR(from))
1da177e4 2850 return PTR_ERR(from);
1da177e4 2851
2ad94ae6 2852 error = user_path_parent(newdfd, newname, &nd, &to);
6902d925 2853 if (error)
2ad94ae6
AV		 2854 goto out_putname;
2855
6902d925
DH		 2856 dentry = lookup_create(&nd, 0);
2857 error = PTR_ERR(dentry);
2858 if (IS_ERR(dentry))
2859 goto out_unlock;
2860
75c3f29d
DH		 2861 error = mnt_want_write(nd.path.mnt);
2862 if (error)
2863 goto out_dput;
be6d3e56
KT		 2864 error = security_path_symlink(&nd.path, dentry, from);
2865 if (error)
2866 goto out_drop_write;
db2e747b 2867 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
be6d3e56 2868out_drop_write:
75c3f29d
DH		 2869 mnt_drop_write(nd.path.mnt);
2870out_dput:
6902d925
DH		 2871 dput(dentry);
2872out_unlock:
4ac91378 2873 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2874 path_put(&nd.path);
6902d925
DH		 2875 putname(to);
2876out_putname:
1da177e4
LT		 2877 putname(from);
2878 return error;
2879}
2880
3480b257 2881SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 2882{
2883 return sys_symlinkat(oldname, AT_FDCWD, newname);
2884}
2885
1da177e4
LT		 2886int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2887{
2888 struct inode *inode = old_dentry->d_inode;
2889 int error;
2890
2891 if (!inode)
2892 return -ENOENT;
2893
a95164d9 2894 error = may_create(dir, new_dentry);
1da177e4
LT		 2895 if (error)
2896 return error;
2897
2898 if (dir->i_sb != inode->i_sb)
2899 return -EXDEV;
2900
2901 /*
2902 * A link to an append-only or immutable file cannot be created.
2903 */
2904 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2905 return -EPERM;
acfa4380 2906 if (!dir->i_op->link)
1da177e4 2907 return -EPERM;
7e79eedb 2908 if (S_ISDIR(inode->i_mode))
1da177e4
LT		 2909 return -EPERM;
2910
2911 error = security_inode_link(old_dentry, dir, new_dentry);
2912 if (error)
2913 return error;
2914
7e79eedb 2915 mutex_lock(&inode->i_mutex);
1da177e4 2916 error = dir->i_op->link(old_dentry, dir, new_dentry);
7e79eedb 2917 mutex_unlock(&inode->i_mutex);
e31e14ec 2918 if (!error)
7e79eedb 2919 fsnotify_link(dir, inode, new_dentry);
1da177e4
LT		 2920 return error;
2921}
2922
2923/*
2924 * Hardlinks are often used in delicate situations. We avoid
2925 * security-related surprises by not following symlinks on the
2926 * newname. --KAB
2927 *
2928 * We don't follow them on the oldname either to be compatible
2929 * with linux 2.0, and to avoid hard-linking to directories
2930 * and other special files. --ADM
2931 */
2e4d0924
HC		 2932SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2933 int, newdfd, const char __user *, newname, int, flags)
1da177e4
LT		 2934{
2935 struct dentry *new_dentry;
2d8f3038
AV		 2936 struct nameidata nd;
2937 struct path old_path;
1da177e4 2938 int error;
2ad94ae6 2939 char *to;
1da177e4 2940
45c9b11a 2941 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
c04030e1
UD		 2942 return -EINVAL;
2943
2d8f3038
AV		 2944 error = user_path_at(olddfd, oldname,
2945 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2946 &old_path);
1da177e4 2947 if (error)
2ad94ae6
AV		 2948 return error;
2949
2950 error = user_path_parent(newdfd, newname, &nd, &to);
1da177e4
LT		 2951 if (error)
2952 goto out;
2953 error = -EXDEV;
2d8f3038 2954 if (old_path.mnt != nd.path.mnt)
1da177e4
LT		 2955 goto out_release;
2956 new_dentry = lookup_create(&nd, 0);
2957 error = PTR_ERR(new_dentry);
6902d925
DH		 2958 if (IS_ERR(new_dentry))
2959 goto out_unlock;
75c3f29d
DH		 2960 error = mnt_want_write(nd.path.mnt);
2961 if (error)
2962 goto out_dput;
be6d3e56
KT		 2963 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2964 if (error)
2965 goto out_drop_write;
2d8f3038 2966 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
be6d3e56 2967out_drop_write:
75c3f29d
DH		 2968 mnt_drop_write(nd.path.mnt);
2969out_dput:
6902d925
DH		 2970 dput(new_dentry);
2971out_unlock:
4ac91378 2972 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2973out_release:
1d957f9b 2974 path_put(&nd.path);
2ad94ae6 2975 putname(to);
1da177e4 2976out:
2d8f3038 2977 path_put(&old_path);
1da177e4
LT		 2978
2979 return error;
2980}
2981
3480b257 2982SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
5590ff0d 2983{
c04030e1 2984 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
5590ff0d
UD		 2985}
2986
1da177e4
LT		 2987/*
2988 * The worst of all namespace operations - renaming directory. "Perverted"
2989 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2990 * Problems:
2991 * a) we can get into loop creation. Check is done in is_subdir().
2992 * b) race potential - two innocent renames can create a loop together.
2993 * That's where 4.4 screws up. Current fix: serialization on
a11f3a05 2994 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
1da177e4
LT		 2995 * story.
2996 * c) we have to lock _three_ objects - parents and victim (if it exists).
1b1dcc1b 2997 * And that - after we got ->i_mutex on parents (until then we don't know
1da177e4
LT		 2998 * whether the target exists). Solution: try to be smart with locking
2999 * order for inodes. We rely on the fact that tree topology may change
a11f3a05 3000 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
1da177e4
LT		 3001 * move will be locked. Thus we can rank directories by the tree
3002 * (ancestors first) and rank all non-directories after them.
3003 * That works since everybody except rename does "lock parent, lookup,
a11f3a05 3004 * lock child" and rename is under ->s_vfs_rename_mutex.
1da177e4
LT		 3005 * HOWEVER, it relies on the assumption that any object with ->lookup()
3006 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3007 * we'd better make sure that there's no link(2) for them.
3008 * d) some filesystems don't support opened-but-unlinked directories,
3009 * either because of layout or because they are not ready to deal with
3010 * all cases correctly. The latter will be fixed (taking this sort of
3011 * stuff into VFS), but the former is not going away. Solution: the same
3012 * trick as in rmdir().
3013 * e) conversion from fhandle to dentry may come in the wrong moment - when
1b1dcc1b 3014 * we are removing the target. Solution: we will have to grab ->i_mutex
1da177e4 3015 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
c41b20e7 3016 * ->i_mutex on parents, which works but leads to some truly excessive
1da177e4
LT		 3017 * locking].
3018 */
75c96f85
AB		 3019static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3020 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3021{
3022 int error = 0;
3023 struct inode *target;
3024
3025 /*
3026 * If we are going to change the parent - check write permissions,
3027 * we'll need to flip '..'.
3028 */
3029 if (new_dir != old_dir) {
f419a2e3 3030 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
1da177e4
LT		 3031 if (error)
3032 return error;
3033 }
3034
3035 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3036 if (error)
3037 return error;
3038
3039 target = new_dentry->d_inode;
d83c49f3 3040 if (target)
1b1dcc1b 3041 mutex_lock(&target->i_mutex);
1da177e4
LT		 3042 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3043 error = -EBUSY;
d83c49f3
AV		 3044 else {
3045 if (target)
3046 dentry_unhash(new_dentry);
1da177e4 3047 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
d83c49f3 3048 }
1da177e4 3049 if (target) {
d83c49f3 3050 if (!error) {
1da177e4 3051 target->i_flags |= S_DEAD;
d83c49f3
AV		 3052 dont_mount(new_dentry);
3053 }
1b1dcc1b 3054 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3055 if (d_unhashed(new_dentry))
3056 d_rehash(new_dentry);
3057 dput(new_dentry);
3058 }
e31e14ec 3059 if (!error)
349457cc
MF		 3060 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3061 d_move(old_dentry,new_dentry);
1da177e4
LT		 3062 return error;
3063}
3064
75c96f85
AB		 3065static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3066 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3067{
3068 struct inode *target;
3069 int error;
3070
3071 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3072 if (error)
3073 return error;
3074
3075 dget(new_dentry);
3076 target = new_dentry->d_inode;
3077 if (target)
1b1dcc1b 3078 mutex_lock(&target->i_mutex);
1da177e4
LT		 3079 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3080 error = -EBUSY;
3081 else
3082 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3083 if (!error) {
bec1052e 3084 if (target)
d83c49f3 3085 dont_mount(new_dentry);
349457cc 3086 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
1da177e4 3087 d_move(old_dentry, new_dentry);
1da177e4
LT		 3088 }
3089 if (target)
1b1dcc1b 3090 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3091 dput(new_dentry);
3092 return error;
3093}
3094
3095int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3096 struct inode *new_dir, struct dentry *new_dentry)
3097{
3098 int error;
3099 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
59b0df21 3100 const unsigned char *old_name;
1da177e4
LT		 3101
3102 if (old_dentry->d_inode == new_dentry->d_inode)
3103 return 0;
3104
3105 error = may_delete(old_dir, old_dentry, is_dir);
3106 if (error)
3107 return error;
3108
3109 if (!new_dentry->d_inode)
a95164d9 3110 error = may_create(new_dir, new_dentry);
1da177e4
LT		 3111 else
3112 error = may_delete(new_dir, new_dentry, is_dir);
3113 if (error)
3114 return error;
3115
acfa4380 3116 if (!old_dir->i_op->rename)
1da177e4
LT		 3117 return -EPERM;
3118
0eeca283
RL		 3119 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3120
1da177e4
LT		 3121 if (is_dir)
3122 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3123 else
3124 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
123df294
AV		 3125 if (!error)
3126 fsnotify_move(old_dir, new_dir, old_name, is_dir,
5a190ae6 3127 new_dentry->d_inode, old_dentry);
0eeca283
RL		 3128 fsnotify_oldname_free(old_name);
3129
1da177e4
LT		 3130 return error;
3131}
3132
2e4d0924
HC		 3133SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3134 int, newdfd, const char __user *, newname)
1da177e4 3135{
2ad94ae6
AV		 3136 struct dentry *old_dir, *new_dir;
3137 struct dentry *old_dentry, *new_dentry;
3138 struct dentry *trap;
1da177e4 3139 struct nameidata oldnd, newnd;
2ad94ae6
AV		 3140 char *from;
3141 char *to;
3142 int error;
1da177e4 3143
2ad94ae6 3144 error = user_path_parent(olddfd, oldname, &oldnd, &from);
1da177e4
LT		 3145 if (error)
3146 goto exit;
3147
2ad94ae6 3148 error = user_path_parent(newdfd, newname, &newnd, &to);
1da177e4
LT		 3149 if (error)
3150 goto exit1;
3151
3152 error = -EXDEV;
4ac91378 3153 if (oldnd.path.mnt != newnd.path.mnt)
1da177e4
LT		 3154 goto exit2;
3155
4ac91378 3156 old_dir = oldnd.path.dentry;
1da177e4
LT		 3157 error = -EBUSY;
3158 if (oldnd.last_type != LAST_NORM)
3159 goto exit2;
3160
4ac91378 3161 new_dir = newnd.path.dentry;
1da177e4
LT		 3162 if (newnd.last_type != LAST_NORM)
3163 goto exit2;
3164
0612d9fb
OH		 3165 oldnd.flags &= ~LOOKUP_PARENT;
3166 newnd.flags &= ~LOOKUP_PARENT;
4e9ed2f8 3167 newnd.flags |= LOOKUP_RENAME_TARGET;
0612d9fb 3168
1da177e4
LT		 3169 trap = lock_rename(new_dir, old_dir);
3170
49705b77 3171 old_dentry = lookup_hash(&oldnd);
1da177e4
LT		 3172 error = PTR_ERR(old_dentry);
3173 if (IS_ERR(old_dentry))
3174 goto exit3;
3175 /* source must exist */
3176 error = -ENOENT;
3177 if (!old_dentry->d_inode)
3178 goto exit4;
3179 /* unless the source is a directory trailing slashes give -ENOTDIR */
3180 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3181 error = -ENOTDIR;
3182 if (oldnd.last.name[oldnd.last.len])
3183 goto exit4;
3184 if (newnd.last.name[newnd.last.len])
3185 goto exit4;
3186 }
3187 /* source should not be ancestor of target */
3188 error = -EINVAL;
3189 if (old_dentry == trap)
3190 goto exit4;
49705b77 3191 new_dentry = lookup_hash(&newnd);
1da177e4
LT		 3192 error = PTR_ERR(new_dentry);
3193 if (IS_ERR(new_dentry))
3194 goto exit4;
3195 /* target should not be an ancestor of source */
3196 error = -ENOTEMPTY;
3197 if (new_dentry == trap)
3198 goto exit5;
3199
9079b1eb
DH		 3200 error = mnt_want_write(oldnd.path.mnt);
3201 if (error)
3202 goto exit5;
be6d3e56
KT		 3203 error = security_path_rename(&oldnd.path, old_dentry,
3204 &newnd.path, new_dentry);
3205 if (error)
3206 goto exit6;
1da177e4
LT		 3207 error = vfs_rename(old_dir->d_inode, old_dentry,
3208 new_dir->d_inode, new_dentry);
be6d3e56 3209exit6:
9079b1eb 3210 mnt_drop_write(oldnd.path.mnt);
1da177e4
LT		 3211exit5:
3212 dput(new_dentry);
3213exit4:
3214 dput(old_dentry);
3215exit3:
3216 unlock_rename(new_dir, old_dir);
3217exit2:
1d957f9b 3218 path_put(&newnd.path);
2ad94ae6 3219 putname(to);
1da177e4 3220exit1:
1d957f9b 3221 path_put(&oldnd.path);
1da177e4 3222 putname(from);
2ad94ae6 3223exit:
1da177e4
LT		 3224 return error;
3225}
3226
a26eab24 3227SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 3228{
3229 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3230}
3231
1da177e4
LT		 3232int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3233{
3234 int len;
3235
3236 len = PTR_ERR(link);
3237 if (IS_ERR(link))
3238 goto out;
3239
3240 len = strlen(link);
3241 if (len > (unsigned) buflen)
3242 len = buflen;
3243 if (copy_to_user(buffer, link, len))
3244 len = -EFAULT;
3245out:
3246 return len;
3247}
3248
3249/*
3250 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3251 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3252 * using) it for any given inode is up to filesystem.
3253 */
3254int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3255{
3256 struct nameidata nd;
cc314eef 3257 void *cookie;
694a1764 3258 int res;
cc314eef 3259
1da177e4 3260 nd.depth = 0;
cc314eef 3261 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
694a1764
MS		 3262 if (IS_ERR(cookie))
3263 return PTR_ERR(cookie);
3264
3265 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3266 if (dentry->d_inode->i_op->put_link)
3267 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3268 return res;
1da177e4
LT		 3269}
3270
3271int vfs_follow_link(struct nameidata *nd, const char *link)
3272{
3273 return __vfs_follow_link(nd, link);
3274}
3275
3276/* get the link contents into pagecache */
3277static char *page_getlink(struct dentry * dentry, struct page **ppage)
3278{
ebd09abb
DG		 3279 char *kaddr;
3280 struct page *page;
1da177e4 3281 struct address_space *mapping = dentry->d_inode->i_mapping;
090d2b18 3282 page = read_mapping_page(mapping, 0, NULL);
1da177e4 3283 if (IS_ERR(page))
6fe6900e 3284 return (char*)page;
1da177e4 3285 *ppage = page;
ebd09abb
DG		 3286 kaddr = kmap(page);
3287 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3288 return kaddr;
1da177e4
LT		 3289}
3290
3291int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3292{
3293 struct page *page = NULL;
3294 char *s = page_getlink(dentry, &page);
3295 int res = vfs_readlink(dentry,buffer,buflen,s);
3296 if (page) {
3297 kunmap(page);
3298 page_cache_release(page);
3299 }
3300 return res;
3301}
3302
cc314eef 3303void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
1da177e4 3304{
cc314eef 3305 struct page *page = NULL;
1da177e4 3306 nd_set_link(nd, page_getlink(dentry, &page));
cc314eef 3307 return page;
1da177e4
LT		 3308}
3309
cc314eef 3310void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1da177e4 3311{
cc314eef
LT		 3312 struct page *page = cookie;
3313
3314 if (page) {
1da177e4
LT		 3315 kunmap(page);
3316 page_cache_release(page);
1da177e4
LT		 3317 }
3318}
3319
54566b2c
NP		 3320/*
3321 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3322 */
3323int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
1da177e4
LT		 3324{
3325 struct address_space *mapping = inode->i_mapping;
0adb25d2 3326 struct page *page;
afddba49 3327 void *fsdata;
beb497ab 3328 int err;
1da177e4 3329 char *kaddr;
54566b2c
NP		 3330 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3331 if (nofs)
3332 flags |= AOP_FLAG_NOFS;
1da177e4 3333
7e53cac4 3334retry:
afddba49 3335 err = pagecache_write_begin(NULL, mapping, 0, len-1,
54566b2c 3336 flags, &page, &fsdata);
1da177e4 3337 if (err)
afddba49
NP		 3338 goto fail;
3339
1da177e4
LT		 3340 kaddr = kmap_atomic(page, KM_USER0);
3341 memcpy(kaddr, symname, len-1);
3342 kunmap_atomic(kaddr, KM_USER0);
afddba49
NP		 3343
3344 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3345 page, fsdata);
1da177e4
LT		 3346 if (err < 0)
3347 goto fail;
afddba49
NP		 3348 if (err < len-1)
3349 goto retry;
3350
1da177e4
LT		 3351 mark_inode_dirty(inode);
3352 return 0;
1da177e4
LT		 3353fail:
3354 return err;
3355}
3356
0adb25d2
KK		 3357int page_symlink(struct inode *inode, const char *symname, int len)
3358{
3359 return __page_symlink(inode, symname, len,
54566b2c 3360 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
0adb25d2
KK		 3361}
3362
92e1d5be 3363const struct inode_operations page_symlink_inode_operations = {
1da177e4
LT		 3364 .readlink = generic_readlink,
3365 .follow_link = page_follow_link_light,
3366 .put_link = page_put_link,
3367};
3368
2d8f3038 3369EXPORT_SYMBOL(user_path_at);
1da177e4
LT		 3370EXPORT_SYMBOL(follow_down);
3371EXPORT_SYMBOL(follow_up);
3372EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3373EXPORT_SYMBOL(getname);
3374EXPORT_SYMBOL(lock_rename);
1da177e4
LT		 3375EXPORT_SYMBOL(lookup_one_len);
3376EXPORT_SYMBOL(page_follow_link_light);
3377EXPORT_SYMBOL(page_put_link);
3378EXPORT_SYMBOL(page_readlink);
0adb25d2 3379EXPORT_SYMBOL(__page_symlink);
1da177e4
LT		 3380EXPORT_SYMBOL(page_symlink);
3381EXPORT_SYMBOL(page_symlink_inode_operations);
3382EXPORT_SYMBOL(path_lookup);
d1811465 3383EXPORT_SYMBOL(kern_path);
16f18200 3384EXPORT_SYMBOL(vfs_path_lookup);
f419a2e3 3385EXPORT_SYMBOL(inode_permission);
8c744fb8 3386EXPORT_SYMBOL(file_permission);
1da177e4
LT		 3387EXPORT_SYMBOL(unlock_rename);
3388EXPORT_SYMBOL(vfs_create);
3389EXPORT_SYMBOL(vfs_follow_link);
3390EXPORT_SYMBOL(vfs_link);
3391EXPORT_SYMBOL(vfs_mkdir);
3392EXPORT_SYMBOL(vfs_mknod);
3393EXPORT_SYMBOL(generic_permission);
3394EXPORT_SYMBOL(vfs_readlink);
3395EXPORT_SYMBOL(vfs_rename);
3396EXPORT_SYMBOL(vfs_rmdir);
3397EXPORT_SYMBOL(vfs_symlink);
3398EXPORT_SYMBOL(vfs_unlink);
3399EXPORT_SYMBOL(dentry_unhash);
3400EXPORT_SYMBOL(generic_readlink);
Linux kernel - Deliverables
This page took 1.314985 seconds and 5 git commands to generate.