blob: 5b961eb71cbf9781988b50559ff10842f49e483a [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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>
22#include <linux/quotaops.h>
23#include <linux/pagemap.h>
Robert Love0eeca282005-07-12 17:06:03 -040024#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/personality.h>
26#include <linux/security.h>
Mimi Zohar6146f0d2009-02-04 09:06:57 -050027#include <linux/ima.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/syscalls.h>
29#include <linux/mount.h>
30#include <linux/audit.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080031#include <linux/capability.h>
Trond Myklebust834f2a42005-10-18 14:20:16 -070032#include <linux/file.h>
Ulrich Drepper5590ff02006-01-18 17:43:53 -080033#include <linux/fcntl.h>
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -070034#include <linux/device_cgroup.h>
Al Viro5ad4e532009-03-29 19:50:06 -040035#include <linux/fs_struct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <asm/uaccess.h>
37
38#define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
39
40/* [Feb-1997 T. Schoebel-Theuer]
41 * Fundamental changes in the pathname lookup mechanisms (namei)
42 * were necessary because of omirr. The reason is that omirr needs
43 * to know the _real_ pathname, not the user-supplied one, in case
44 * of symlinks (and also when transname replacements occur).
45 *
46 * The new code replaces the old recursive symlink resolution with
47 * an iterative one (in case of non-nested symlink chains). It does
48 * this with calls to <fs>_follow_link().
49 * As a side effect, dir_namei(), _namei() and follow_link() are now
50 * replaced with a single function lookup_dentry() that can handle all
51 * the special cases of the former code.
52 *
53 * With the new dcache, the pathname is stored at each inode, at least as
54 * long as the refcount of the inode is positive. As a side effect, the
55 * size of the dcache depends on the inode cache and thus is dynamic.
56 *
57 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
58 * resolution to correspond with current state of the code.
59 *
60 * Note that the symlink resolution is not *completely* iterative.
61 * There is still a significant amount of tail- and mid- recursion in
62 * the algorithm. Also, note that <fs>_readlink() is not used in
63 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
64 * may return different results than <fs>_follow_link(). Many virtual
65 * filesystems (including /proc) exhibit this behavior.
66 */
67
68/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
69 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
70 * and the name already exists in form of a symlink, try to create the new
71 * name indicated by the symlink. The old code always complained that the
72 * name already exists, due to not following the symlink even if its target
73 * is nonexistent. The new semantics affects also mknod() and link() when
74 * the name is a symlink pointing to a non-existant name.
75 *
76 * I don't know which semantics is the right one, since I have no access
77 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
78 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
79 * "old" one. Personally, I think the new semantics is much more logical.
80 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
81 * file does succeed in both HP-UX and SunOs, but not in Solaris
82 * and in the old Linux semantics.
83 */
84
85/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
86 * semantics. See the comments in "open_namei" and "do_link" below.
87 *
88 * [10-Sep-98 Alan Modra] Another symlink change.
89 */
90
91/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
92 * inside the path - always follow.
93 * in the last component in creation/removal/renaming - never follow.
94 * if LOOKUP_FOLLOW passed - follow.
95 * if the pathname has trailing slashes - follow.
96 * otherwise - don't follow.
97 * (applied in that order).
98 *
99 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
100 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
101 * During the 2.4 we need to fix the userland stuff depending on it -
102 * hopefully we will be able to get rid of that wart in 2.5. So far only
103 * XEmacs seems to be relying on it...
104 */
105/*
106 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800107 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 * any extra contention...
109 */
110
Al Viroa02f76c2008-02-23 15:14:28 +0000111static int __link_path_walk(const char *name, struct nameidata *nd);
Josef 'Jeff' Sipekc4a78082007-07-19 01:48:22 -0700112
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113/* In order to reduce some races, while at the same time doing additional
114 * checking and hopefully speeding things up, we copy filenames to the
115 * kernel data space before using them..
116 *
117 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
118 * PATH_MAX includes the nul terminator --RR.
119 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800120static int do_getname(const char __user *filename, char *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121{
122 int retval;
123 unsigned long len = PATH_MAX;
124
125 if (!segment_eq(get_fs(), KERNEL_DS)) {
126 if ((unsigned long) filename >= TASK_SIZE)
127 return -EFAULT;
128 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
129 len = TASK_SIZE - (unsigned long) filename;
130 }
131
132 retval = strncpy_from_user(page, filename, len);
133 if (retval > 0) {
134 if (retval < len)
135 return 0;
136 return -ENAMETOOLONG;
137 } else if (!retval)
138 retval = -ENOENT;
139 return retval;
140}
141
142char * getname(const char __user * filename)
143{
144 char *tmp, *result;
145
146 result = ERR_PTR(-ENOMEM);
147 tmp = __getname();
148 if (tmp) {
149 int retval = do_getname(filename, tmp);
150
151 result = tmp;
152 if (retval < 0) {
153 __putname(tmp);
154 result = ERR_PTR(retval);
155 }
156 }
157 audit_getname(result);
158 return result;
159}
160
161#ifdef CONFIG_AUDITSYSCALL
162void putname(const char *name)
163{
Al Viro5ac3a9c2006-07-16 06:38:45 -0400164 if (unlikely(!audit_dummy_context()))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 audit_putname(name);
166 else
167 __putname(name);
168}
169EXPORT_SYMBOL(putname);
170#endif
171
172
173/**
174 * generic_permission - check for access rights on a Posix-like filesystem
175 * @inode: inode to check access rights for
176 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
177 * @check_acl: optional callback to check for Posix ACLs
178 *
179 * Used to check for read/write/execute permissions on a file.
180 * We use "fsuid" for this, letting us set arbitrary permissions
181 * for filesystem access without changing the "normal" uids which
182 * are used for other things..
183 */
184int generic_permission(struct inode *inode, int mask,
185 int (*check_acl)(struct inode *inode, int mask))
186{
187 umode_t mode = inode->i_mode;
188
Al Viroe6305c42008-07-15 21:03:57 -0400189 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
190
David Howellsda9592e2008-11-14 10:39:05 +1100191 if (current_fsuid() == inode->i_uid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 mode >>= 6;
193 else {
194 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
195 int error = check_acl(inode, mask);
196 if (error == -EACCES)
197 goto check_capabilities;
198 else if (error != -EAGAIN)
199 return error;
200 }
201
202 if (in_group_p(inode->i_gid))
203 mode >>= 3;
204 }
205
206 /*
207 * If the DACs are ok we don't need any capability check.
208 */
Al Viroe6305c42008-07-15 21:03:57 -0400209 if ((mask & ~mode) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 return 0;
211
212 check_capabilities:
213 /*
214 * Read/write DACs are always overridable.
215 * Executable DACs are overridable if at least one exec bit is set.
216 */
Miklos Szeredif696a362008-07-31 13:41:58 +0200217 if (!(mask & MAY_EXEC) || execute_ok(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 if (capable(CAP_DAC_OVERRIDE))
219 return 0;
220
221 /*
222 * Searching includes executable on directories, else just read.
223 */
224 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
225 if (capable(CAP_DAC_READ_SEARCH))
226 return 0;
227
228 return -EACCES;
229}
230
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200231/**
232 * inode_permission - check for access rights to a given inode
233 * @inode: inode to check permission on
234 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
235 *
236 * Used to check for read/write/execute permissions on an inode.
237 * We use "fsuid" for this, letting us set arbitrary permissions
238 * for filesystem access without changing the "normal" uids which
239 * are used for other things.
240 */
Al Virof419a2e2008-07-22 00:07:17 -0400241int inode_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242{
Al Viroe6305c42008-07-15 21:03:57 -0400243 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244
245 if (mask & MAY_WRITE) {
Miklos Szeredi22590e42007-10-16 23:27:08 -0700246 umode_t mode = inode->i_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247
248 /*
249 * Nobody gets write access to a read-only fs.
250 */
251 if (IS_RDONLY(inode) &&
252 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
253 return -EROFS;
254
255 /*
256 * Nobody gets write access to an immutable file.
257 */
258 if (IS_IMMUTABLE(inode))
259 return -EACCES;
260 }
261
Al Viroacfa4382008-12-04 10:06:33 -0500262 if (inode->i_op->permission)
Al Virob77b0642008-07-17 09:37:02 -0400263 retval = inode->i_op->permission(inode, mask);
Miklos Szeredif696a362008-07-31 13:41:58 +0200264 else
Al Viroe6305c42008-07-15 21:03:57 -0400265 retval = generic_permission(inode, mask, NULL);
Miklos Szeredif696a362008-07-31 13:41:58 +0200266
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 if (retval)
268 return retval;
269
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -0700270 retval = devcgroup_inode_permission(inode, mask);
271 if (retval)
272 return retval;
273
Al Viroe6305c42008-07-15 21:03:57 -0400274 return security_inode_permission(inode,
Stephen Smalleyf418b002008-07-28 13:32:38 -0400275 mask & (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276}
277
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800278/**
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800279 * file_permission - check for additional access rights to a given file
280 * @file: file to check access rights for
281 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
282 *
283 * Used to check for read/write/execute permissions on an already opened
284 * file.
285 *
286 * Note:
287 * Do not use this function in new code. All access checks should
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200288 * be done using inode_permission().
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800289 */
290int file_permission(struct file *file, int mask)
291{
Al Virof419a2e2008-07-22 00:07:17 -0400292 return inode_permission(file->f_path.dentry->d_inode, mask);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800293}
294
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295/*
296 * get_write_access() gets write permission for a file.
297 * put_write_access() releases this write permission.
298 * This is used for regular files.
299 * We cannot support write (and maybe mmap read-write shared) accesses and
300 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
301 * can have the following values:
302 * 0: no writers, no VM_DENYWRITE mappings
303 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
304 * > 0: (i_writecount) users are writing to the file.
305 *
306 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
307 * except for the cases where we don't hold i_writecount yet. Then we need to
308 * use {get,deny}_write_access() - these functions check the sign and refuse
309 * to do the change if sign is wrong. Exclusion between them is provided by
310 * the inode->i_lock spinlock.
311 */
312
313int get_write_access(struct inode * inode)
314{
315 spin_lock(&inode->i_lock);
316 if (atomic_read(&inode->i_writecount) < 0) {
317 spin_unlock(&inode->i_lock);
318 return -ETXTBSY;
319 }
320 atomic_inc(&inode->i_writecount);
321 spin_unlock(&inode->i_lock);
322
323 return 0;
324}
325
326int deny_write_access(struct file * file)
327{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800328 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329
330 spin_lock(&inode->i_lock);
331 if (atomic_read(&inode->i_writecount) > 0) {
332 spin_unlock(&inode->i_lock);
333 return -ETXTBSY;
334 }
335 atomic_dec(&inode->i_writecount);
336 spin_unlock(&inode->i_lock);
337
338 return 0;
339}
340
Jan Blunck1d957f92008-02-14 19:34:35 -0800341/**
Jan Blunck5dd784d2008-02-14 19:34:38 -0800342 * path_get - get a reference to a path
343 * @path: path to get the reference to
344 *
345 * Given a path increment the reference count to the dentry and the vfsmount.
346 */
347void path_get(struct path *path)
348{
349 mntget(path->mnt);
350 dget(path->dentry);
351}
352EXPORT_SYMBOL(path_get);
353
354/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800355 * path_put - put a reference to a path
356 * @path: path to put the reference to
357 *
358 * Given a path decrement the reference count to the dentry and the vfsmount.
359 */
360void path_put(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361{
Jan Blunck1d957f92008-02-14 19:34:35 -0800362 dput(path->dentry);
363 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364}
Jan Blunck1d957f92008-02-14 19:34:35 -0800365EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366
Trond Myklebust834f2a42005-10-18 14:20:16 -0700367/**
368 * release_open_intent - free up open intent resources
369 * @nd: pointer to nameidata
370 */
371void release_open_intent(struct nameidata *nd)
372{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800373 if (nd->intent.open.file->f_path.dentry == NULL)
Trond Myklebust834f2a42005-10-18 14:20:16 -0700374 put_filp(nd->intent.open.file);
375 else
376 fput(nd->intent.open.file);
377}
378
Ian Kentbcdc5e02006-09-27 01:50:44 -0700379static inline struct dentry *
380do_revalidate(struct dentry *dentry, struct nameidata *nd)
381{
382 int status = dentry->d_op->d_revalidate(dentry, nd);
383 if (unlikely(status <= 0)) {
384 /*
385 * The dentry failed validation.
386 * If d_revalidate returned 0 attempt to invalidate
387 * the dentry otherwise d_revalidate is asking us
388 * to return a fail status.
389 */
390 if (!status) {
391 if (!d_invalidate(dentry)) {
392 dput(dentry);
393 dentry = NULL;
394 }
395 } else {
396 dput(dentry);
397 dentry = ERR_PTR(status);
398 }
399 }
400 return dentry;
401}
402
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403/*
404 * Internal lookup() using the new generic dcache.
405 * SMP-safe
406 */
407static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
408{
409 struct dentry * dentry = __d_lookup(parent, name);
410
411 /* lockess __d_lookup may fail due to concurrent d_move()
412 * in some unrelated directory, so try with d_lookup
413 */
414 if (!dentry)
415 dentry = d_lookup(parent, name);
416
Ian Kentbcdc5e02006-09-27 01:50:44 -0700417 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
418 dentry = do_revalidate(dentry, nd);
419
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 return dentry;
421}
422
423/*
424 * Short-cut version of permission(), for calling by
425 * path_walk(), when dcache lock is held. Combines parts
426 * of permission() and generic_permission(), and tests ONLY for
427 * MAY_EXEC permission.
428 *
429 * If appropriate, check DAC only. If not appropriate, or
430 * short-cut DAC fails, then call permission() to do more
431 * complete permission check.
432 */
Al Viro672b16b2008-07-17 09:45:01 -0400433static int exec_permission_lite(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434{
435 umode_t mode = inode->i_mode;
436
Al Viroacfa4382008-12-04 10:06:33 -0500437 if (inode->i_op->permission)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 return -EAGAIN;
439
David Howellsda9592e2008-11-14 10:39:05 +1100440 if (current_fsuid() == inode->i_uid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 mode >>= 6;
442 else if (in_group_p(inode->i_gid))
443 mode >>= 3;
444
445 if (mode & MAY_EXEC)
446 goto ok;
447
448 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
449 goto ok;
450
451 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
452 goto ok;
453
454 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
455 goto ok;
456
457 return -EACCES;
458ok:
Al Virob77b0642008-07-17 09:37:02 -0400459 return security_inode_permission(inode, MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460}
461
462/*
463 * This is called when everything else fails, and we actually have
464 * to go to the low-level filesystem to find out what we should do..
465 *
466 * We get the directory semaphore, and after getting that we also
467 * make sure that nobody added the entry to the dcache in the meantime..
468 * SMP-safe
469 */
470static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
471{
472 struct dentry * result;
473 struct inode *dir = parent->d_inode;
474
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800475 mutex_lock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 /*
477 * First re-do the cached lookup just in case it was created
478 * while we waited for the directory semaphore..
479 *
480 * FIXME! This could use version numbering or similar to
481 * avoid unnecessary cache lookups.
482 *
483 * The "dcache_lock" is purely to protect the RCU list walker
484 * from concurrent renames at this point (we mustn't get false
485 * negatives from the RCU list walk here, unlike the optimistic
486 * fast walk).
487 *
488 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
489 */
490 result = d_lookup(parent, name);
491 if (!result) {
Miklos Szeredid70b67c2008-07-02 21:30:15 +0200492 struct dentry *dentry;
493
494 /* Don't create child dentry for a dead directory. */
495 result = ERR_PTR(-ENOENT);
496 if (IS_DEADDIR(dir))
497 goto out_unlock;
498
499 dentry = d_alloc(parent, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 result = ERR_PTR(-ENOMEM);
501 if (dentry) {
502 result = dir->i_op->lookup(dir, dentry, nd);
503 if (result)
504 dput(dentry);
505 else
506 result = dentry;
507 }
Miklos Szeredid70b67c2008-07-02 21:30:15 +0200508out_unlock:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800509 mutex_unlock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 return result;
511 }
512
513 /*
514 * Uhhuh! Nasty case: the cache was re-populated while
515 * we waited on the semaphore. Need to revalidate.
516 */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800517 mutex_unlock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 if (result->d_op && result->d_op->d_revalidate) {
Ian Kentbcdc5e02006-09-27 01:50:44 -0700519 result = do_revalidate(result, nd);
520 if (!result)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 result = ERR_PTR(-ENOENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 }
523 return result;
524}
525
Al Viroa02f76c2008-02-23 15:14:28 +0000526/*
527 * Wrapper to retry pathname resolution whenever the underlying
528 * file system returns an ESTALE.
529 *
530 * Retry the whole path once, forcing real lookup requests
531 * instead of relying on the dcache.
532 */
533static __always_inline int link_path_walk(const char *name, struct nameidata *nd)
534{
535 struct path save = nd->path;
536 int result;
537
538 /* make sure the stuff we saved doesn't go away */
Jan Blunckc8e7f442008-06-09 16:40:35 -0700539 path_get(&save);
Al Viroa02f76c2008-02-23 15:14:28 +0000540
541 result = __link_path_walk(name, nd);
542 if (result == -ESTALE) {
543 /* nd->path had been dropped */
544 nd->path = save;
Jan Blunckc8e7f442008-06-09 16:40:35 -0700545 path_get(&nd->path);
Al Viroa02f76c2008-02-23 15:14:28 +0000546 nd->flags |= LOOKUP_REVAL;
547 result = __link_path_walk(name, nd);
548 }
549
550 path_put(&save);
551
552 return result;
553}
554
Al Viro2a737872009-04-07 11:49:53 -0400555static __always_inline void set_root(struct nameidata *nd)
556{
557 if (!nd->root.mnt) {
558 struct fs_struct *fs = current->fs;
559 read_lock(&fs->lock);
560 nd->root = fs->root;
561 path_get(&nd->root);
562 read_unlock(&fs->lock);
563 }
564}
565
Arjan van de Venf1662352006-01-14 13:21:31 -0800566static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567{
568 int res = 0;
569 char *name;
570 if (IS_ERR(link))
571 goto fail;
572
573 if (*link == '/') {
Al Viro2a737872009-04-07 11:49:53 -0400574 set_root(nd);
Jan Blunck1d957f92008-02-14 19:34:35 -0800575 path_put(&nd->path);
Al Viro2a737872009-04-07 11:49:53 -0400576 nd->path = nd->root;
577 path_get(&nd->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 }
Christoph Hellwigb4091d52008-11-05 15:07:21 +0100579
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 res = link_path_walk(link, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 if (nd->depth || res || nd->last_type!=LAST_NORM)
582 return res;
583 /*
584 * If it is an iterative symlinks resolution in open_namei() we
585 * have to copy the last component. And all that crap because of
586 * bloody create() on broken symlinks. Furrfu...
587 */
588 name = __getname();
589 if (unlikely(!name)) {
Jan Blunck1d957f92008-02-14 19:34:35 -0800590 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 return -ENOMEM;
592 }
593 strcpy(name, nd->last.name);
594 nd->last.name = name;
595 return 0;
596fail:
Jan Blunck1d957f92008-02-14 19:34:35 -0800597 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 return PTR_ERR(link);
599}
600
Jan Blunck1d957f92008-02-14 19:34:35 -0800601static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700602{
603 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800604 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700605 mntput(path->mnt);
606}
607
608static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
609{
Jan Blunck4ac91372008-02-14 19:34:32 -0800610 dput(nd->path.dentry);
611 if (nd->path.mnt != path->mnt)
612 mntput(nd->path.mnt);
613 nd->path.mnt = path->mnt;
614 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700615}
616
Ian Kent051d3812006-03-27 01:14:53 -0800617static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
618{
619 int error;
620 void *cookie;
621 struct dentry *dentry = path->dentry;
622
623 touch_atime(path->mnt, dentry);
624 nd_set_link(nd, NULL);
625
Jan Blunck4ac91372008-02-14 19:34:32 -0800626 if (path->mnt != nd->path.mnt) {
Ian Kent051d3812006-03-27 01:14:53 -0800627 path_to_nameidata(path, nd);
628 dget(dentry);
629 }
630 mntget(path->mnt);
631 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
632 error = PTR_ERR(cookie);
633 if (!IS_ERR(cookie)) {
634 char *s = nd_get_link(nd);
635 error = 0;
636 if (s)
637 error = __vfs_follow_link(nd, s);
638 if (dentry->d_inode->i_op->put_link)
639 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
640 }
Jan Blunck09da59162008-02-14 19:34:37 -0800641 path_put(path);
Ian Kent051d3812006-03-27 01:14:53 -0800642
643 return error;
644}
645
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646/*
647 * This limits recursive symlink follows to 8, while
648 * limiting consecutive symlinks to 40.
649 *
650 * Without that kind of total limit, nasty chains of consecutive
651 * symlinks can cause almost arbitrarily long lookups.
652 */
Al Viro90ebe562005-06-06 13:35:58 -0700653static inline int do_follow_link(struct path *path, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654{
655 int err = -ELOOP;
656 if (current->link_count >= MAX_NESTED_LINKS)
657 goto loop;
658 if (current->total_link_count >= 40)
659 goto loop;
660 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
661 cond_resched();
Al Viro90ebe562005-06-06 13:35:58 -0700662 err = security_inode_follow_link(path->dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 if (err)
664 goto loop;
665 current->link_count++;
666 current->total_link_count++;
667 nd->depth++;
Al Virocd4e91d2005-06-06 13:36:03 -0700668 err = __do_follow_link(path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 current->link_count--;
670 nd->depth--;
671 return err;
672loop:
Jan Blunck1d957f92008-02-14 19:34:35 -0800673 path_put_conditional(path, nd);
674 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 return err;
676}
677
Al Virobab77eb2009-04-18 03:26:48 -0400678int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679{
680 struct vfsmount *parent;
681 struct dentry *mountpoint;
682 spin_lock(&vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400683 parent = path->mnt->mnt_parent;
684 if (parent == path->mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 spin_unlock(&vfsmount_lock);
686 return 0;
687 }
688 mntget(parent);
Al Virobab77eb2009-04-18 03:26:48 -0400689 mountpoint = dget(path->mnt->mnt_mountpoint);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 spin_unlock(&vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400691 dput(path->dentry);
692 path->dentry = mountpoint;
693 mntput(path->mnt);
694 path->mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 return 1;
696}
697
698/* no need for dcache_lock, as serialization is taken care in
699 * namespace.c
700 */
Al Viro463ffb22005-06-06 13:36:05 -0700701static int __follow_mount(struct path *path)
702{
703 int res = 0;
704 while (d_mountpoint(path->dentry)) {
Al Viro1c755af2009-04-18 14:06:57 -0400705 struct vfsmount *mounted = lookup_mnt(path);
Al Viro463ffb22005-06-06 13:36:05 -0700706 if (!mounted)
707 break;
708 dput(path->dentry);
709 if (res)
710 mntput(path->mnt);
711 path->mnt = mounted;
712 path->dentry = dget(mounted->mnt_root);
713 res = 1;
714 }
715 return res;
716}
717
Al Viro79ed0222009-04-18 13:59:41 -0400718static void follow_mount(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719{
Al Viro79ed0222009-04-18 13:59:41 -0400720 while (d_mountpoint(path->dentry)) {
Al Viro1c755af2009-04-18 14:06:57 -0400721 struct vfsmount *mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 if (!mounted)
723 break;
Al Viro79ed0222009-04-18 13:59:41 -0400724 dput(path->dentry);
725 mntput(path->mnt);
726 path->mnt = mounted;
727 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729}
730
731/* no need for dcache_lock, as serialization is taken care in
732 * namespace.c
733 */
Al Viro9393bd02009-04-18 13:58:15 -0400734int follow_down(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735{
736 struct vfsmount *mounted;
737
Al Viro1c755af2009-04-18 14:06:57 -0400738 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -0400740 dput(path->dentry);
741 mntput(path->mnt);
742 path->mnt = mounted;
743 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 return 1;
745 }
746 return 0;
747}
748
Arjan van de Venf1662352006-01-14 13:21:31 -0800749static __always_inline void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750{
Al Viro2a737872009-04-07 11:49:53 -0400751 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -0700752
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 while(1) {
754 struct vfsmount *parent;
Jan Blunck4ac91372008-02-14 19:34:32 -0800755 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756
Al Viro2a737872009-04-07 11:49:53 -0400757 if (nd->path.dentry == nd->root.dentry &&
758 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 break;
760 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 spin_lock(&dcache_lock);
Jan Blunck4ac91372008-02-14 19:34:32 -0800762 if (nd->path.dentry != nd->path.mnt->mnt_root) {
763 nd->path.dentry = dget(nd->path.dentry->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 spin_unlock(&dcache_lock);
765 dput(old);
766 break;
767 }
768 spin_unlock(&dcache_lock);
769 spin_lock(&vfsmount_lock);
Jan Blunck4ac91372008-02-14 19:34:32 -0800770 parent = nd->path.mnt->mnt_parent;
771 if (parent == nd->path.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 spin_unlock(&vfsmount_lock);
773 break;
774 }
775 mntget(parent);
Jan Blunck4ac91372008-02-14 19:34:32 -0800776 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 spin_unlock(&vfsmount_lock);
778 dput(old);
Jan Blunck4ac91372008-02-14 19:34:32 -0800779 mntput(nd->path.mnt);
780 nd->path.mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 }
Al Viro79ed0222009-04-18 13:59:41 -0400782 follow_mount(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783}
784
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785/*
786 * It's more convoluted than I'd like it to be, but... it's still fairly
787 * small and for now I'd prefer to have fast path as straight as possible.
788 * It _is_ time-critical.
789 */
790static int do_lookup(struct nameidata *nd, struct qstr *name,
791 struct path *path)
792{
Jan Blunck4ac91372008-02-14 19:34:32 -0800793 struct vfsmount *mnt = nd->path.mnt;
794 struct dentry *dentry = __d_lookup(nd->path.dentry, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795
796 if (!dentry)
797 goto need_lookup;
798 if (dentry->d_op && dentry->d_op->d_revalidate)
799 goto need_revalidate;
800done:
801 path->mnt = mnt;
802 path->dentry = dentry;
Al Viro634ee702005-06-06 13:36:13 -0700803 __follow_mount(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 return 0;
805
806need_lookup:
Jan Blunck4ac91372008-02-14 19:34:32 -0800807 dentry = real_lookup(nd->path.dentry, name, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 if (IS_ERR(dentry))
809 goto fail;
810 goto done;
811
812need_revalidate:
Ian Kentbcdc5e02006-09-27 01:50:44 -0700813 dentry = do_revalidate(dentry, nd);
814 if (!dentry)
815 goto need_lookup;
816 if (IS_ERR(dentry))
817 goto fail;
818 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819
820fail:
821 return PTR_ERR(dentry);
822}
823
824/*
825 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +0100826 * This is the basic name resolution function, turning a pathname into
827 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 *
Prasanna Medaea3834d2005-04-29 16:00:17 +0100829 * Returns 0 and nd will have valid dentry and mnt on success.
830 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800832static int __link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833{
834 struct path next;
835 struct inode *inode;
836 int err;
837 unsigned int lookup_flags = nd->flags;
838
839 while (*name=='/')
840 name++;
841 if (!*name)
842 goto return_reval;
843
Jan Blunck4ac91372008-02-14 19:34:32 -0800844 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 if (nd->depth)
Trond Myklebustf55eab82006-02-04 23:28:01 -0800846 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847
848 /* At this point we know we have a real path component. */
849 for(;;) {
850 unsigned long hash;
851 struct qstr this;
852 unsigned int c;
853
Trond Myklebustcdce5d62005-10-18 14:20:18 -0700854 nd->flags |= LOOKUP_CONTINUE;
Al Viro672b16b2008-07-17 09:45:01 -0400855 err = exec_permission_lite(inode);
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800856 if (err == -EAGAIN)
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200857 err = inode_permission(nd->path.dentry->d_inode,
858 MAY_EXEC);
Mimi Zohar6146f0d2009-02-04 09:06:57 -0500859 if (!err)
Mimi Zoharb9fc7452009-05-19 13:25:57 -0400860 err = ima_path_check(&nd->path, MAY_EXEC,
861 IMA_COUNT_UPDATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 if (err)
863 break;
864
865 this.name = name;
866 c = *(const unsigned char *)name;
867
868 hash = init_name_hash();
869 do {
870 name++;
871 hash = partial_name_hash(c, hash);
872 c = *(const unsigned char *)name;
873 } while (c && (c != '/'));
874 this.len = name - (const char *) this.name;
875 this.hash = end_name_hash(hash);
876
877 /* remove trailing slashes? */
878 if (!c)
879 goto last_component;
880 while (*++name == '/');
881 if (!*name)
882 goto last_with_slashes;
883
884 /*
885 * "." and ".." are special - ".." especially so because it has
886 * to be able to know about the current root directory and
887 * parent relationships.
888 */
889 if (this.name[0] == '.') switch (this.len) {
890 default:
891 break;
892 case 2:
893 if (this.name[1] != '.')
894 break;
Al Viro58c465e2005-06-06 13:36:13 -0700895 follow_dotdot(nd);
Jan Blunck4ac91372008-02-14 19:34:32 -0800896 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 /* fallthrough */
898 case 1:
899 continue;
900 }
901 /*
902 * See if the low-level filesystem might want
903 * to use its own hash..
904 */
Jan Blunck4ac91372008-02-14 19:34:32 -0800905 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
906 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
907 &this);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 if (err < 0)
909 break;
910 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 /* This does the actual lookups.. */
912 err = do_lookup(nd, &this, &next);
913 if (err)
914 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915
916 err = -ENOENT;
917 inode = next.dentry->d_inode;
918 if (!inode)
919 goto out_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920
921 if (inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700922 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 if (err)
924 goto return_err;
925 err = -ENOENT;
Jan Blunck4ac91372008-02-14 19:34:32 -0800926 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 if (!inode)
928 break;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700929 } else
930 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 err = -ENOTDIR;
932 if (!inode->i_op->lookup)
933 break;
934 continue;
935 /* here ends the main loop */
936
937last_with_slashes:
938 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
939last_component:
Trond Myklebustf55eab82006-02-04 23:28:01 -0800940 /* Clear LOOKUP_CONTINUE iff it was previously unset */
941 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 if (lookup_flags & LOOKUP_PARENT)
943 goto lookup_parent;
944 if (this.name[0] == '.') switch (this.len) {
945 default:
946 break;
947 case 2:
948 if (this.name[1] != '.')
949 break;
Al Viro58c465e2005-06-06 13:36:13 -0700950 follow_dotdot(nd);
Jan Blunck4ac91372008-02-14 19:34:32 -0800951 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 /* fallthrough */
953 case 1:
954 goto return_reval;
955 }
Jan Blunck4ac91372008-02-14 19:34:32 -0800956 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
957 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
958 &this);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 if (err < 0)
960 break;
961 }
962 err = do_lookup(nd, &this, &next);
963 if (err)
964 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 inode = next.dentry->d_inode;
966 if ((lookup_flags & LOOKUP_FOLLOW)
Al Viroacfa4382008-12-04 10:06:33 -0500967 && inode && inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700968 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 if (err)
970 goto return_err;
Jan Blunck4ac91372008-02-14 19:34:32 -0800971 inode = nd->path.dentry->d_inode;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700972 } else
973 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 err = -ENOENT;
975 if (!inode)
976 break;
977 if (lookup_flags & LOOKUP_DIRECTORY) {
978 err = -ENOTDIR;
Al Viroacfa4382008-12-04 10:06:33 -0500979 if (!inode->i_op->lookup)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 break;
981 }
982 goto return_base;
983lookup_parent:
984 nd->last = this;
985 nd->last_type = LAST_NORM;
986 if (this.name[0] != '.')
987 goto return_base;
988 if (this.len == 1)
989 nd->last_type = LAST_DOT;
990 else if (this.len == 2 && this.name[1] == '.')
991 nd->last_type = LAST_DOTDOT;
992 else
993 goto return_base;
994return_reval:
995 /*
996 * We bypassed the ordinary revalidation routines.
997 * We may need to check the cached dentry for staleness.
998 */
Jan Blunck4ac91372008-02-14 19:34:32 -0800999 if (nd->path.dentry && nd->path.dentry->d_sb &&
1000 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 err = -ESTALE;
1002 /* Note: we do not d_invalidate() */
Jan Blunck4ac91372008-02-14 19:34:32 -08001003 if (!nd->path.dentry->d_op->d_revalidate(
1004 nd->path.dentry, nd))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 break;
1006 }
1007return_base:
1008 return 0;
1009out_dput:
Jan Blunck1d957f92008-02-14 19:34:35 -08001010 path_put_conditional(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 break;
1012 }
Jan Blunck1d957f92008-02-14 19:34:35 -08001013 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014return_err:
1015 return err;
1016}
1017
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -08001018static int path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019{
1020 current->total_link_count = 0;
1021 return link_path_walk(name, nd);
1022}
1023
Al Viro9b4a9b12009-04-07 11:44:16 -04001024static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001026 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001027 int fput_needed;
1028 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029
1030 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1031 nd->flags = flags;
1032 nd->depth = 0;
Al Viro2a737872009-04-07 11:49:53 -04001033 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 if (*name=='/') {
Al Viro2a737872009-04-07 11:49:53 -04001036 set_root(nd);
1037 nd->path = nd->root;
1038 path_get(&nd->root);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001039 } else if (dfd == AT_FDCWD) {
Al Viro2a737872009-04-07 11:49:53 -04001040 struct fs_struct *fs = current->fs;
Andreas Mohre518ddb2006-09-29 02:01:22 -07001041 read_lock(&fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08001042 nd->path = fs->pwd;
1043 path_get(&fs->pwd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001044 read_unlock(&fs->lock);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001045 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001046 struct dentry *dentry;
1047
1048 file = fget_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001049 retval = -EBADF;
1050 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001051 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001052
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001053 dentry = file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001054
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001055 retval = -ENOTDIR;
1056 if (!S_ISDIR(dentry->d_inode->i_mode))
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001057 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001058
1059 retval = file_permission(file, MAY_EXEC);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001060 if (retval)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001061 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001062
Jan Blunck5dd784d2008-02-14 19:34:38 -08001063 nd->path = file->f_path;
1064 path_get(&file->f_path);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001065
1066 fput_light(file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001068 return 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001069
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001070fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001071 fput_light(file, fput_needed);
Al Viro9b4a9b12009-04-07 11:44:16 -04001072out_fail:
1073 return retval;
1074}
1075
1076/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1077static int do_path_lookup(int dfd, const char *name,
1078 unsigned int flags, struct nameidata *nd)
1079{
1080 int retval = path_init(dfd, name, flags, nd);
1081 if (!retval)
1082 retval = path_walk(name, nd);
1083 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1084 nd->path.dentry->d_inode))
1085 audit_inode(name, nd->path.dentry);
Al Viro2a737872009-04-07 11:49:53 -04001086 if (nd->root.mnt) {
1087 path_put(&nd->root);
1088 nd->root.mnt = NULL;
1089 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001090 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091}
1092
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -08001093int path_lookup(const char *name, unsigned int flags,
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001094 struct nameidata *nd)
1095{
1096 return do_path_lookup(AT_FDCWD, name, flags, nd);
1097}
1098
Al Virod1811462008-08-02 00:49:18 -04001099int kern_path(const char *name, unsigned int flags, struct path *path)
1100{
1101 struct nameidata nd;
1102 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1103 if (!res)
1104 *path = nd.path;
1105 return res;
1106}
1107
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001108/**
1109 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1110 * @dentry: pointer to dentry of the base directory
1111 * @mnt: pointer to vfs mount of the base directory
1112 * @name: pointer to file name
1113 * @flags: lookup flags
1114 * @nd: pointer to nameidata
1115 */
1116int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1117 const char *name, unsigned int flags,
1118 struct nameidata *nd)
1119{
1120 int retval;
1121
1122 /* same as do_path_lookup */
1123 nd->last_type = LAST_ROOT;
1124 nd->flags = flags;
1125 nd->depth = 0;
1126
Jan Blunckc8e7f442008-06-09 16:40:35 -07001127 nd->path.dentry = dentry;
1128 nd->path.mnt = mnt;
1129 path_get(&nd->path);
Al Viro5b857112009-04-07 11:53:49 -04001130 nd->root = nd->path;
1131 path_get(&nd->root);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001132
1133 retval = path_walk(name, nd);
Jan Blunck4ac91372008-02-14 19:34:32 -08001134 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1135 nd->path.dentry->d_inode))
1136 audit_inode(name, nd->path.dentry);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001137
Al Viro5b857112009-04-07 11:53:49 -04001138 path_put(&nd->root);
1139 nd->root.mnt = NULL;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001140
Al Viro2a737872009-04-07 11:49:53 -04001141 return retval;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001142}
1143
Trond Myklebust834f2a42005-10-18 14:20:16 -07001144/**
1145 * path_lookup_open - lookup a file path with open intent
Martin Waitz7045f372006-02-01 03:06:57 -08001146 * @dfd: the directory to use as base, or AT_FDCWD
Trond Myklebust834f2a42005-10-18 14:20:16 -07001147 * @name: pointer to file name
1148 * @lookup_flags: lookup intent flags
1149 * @nd: pointer to nameidata
1150 * @open_flags: open intent flags
1151 */
Al Viro6e8341a2009-04-06 11:16:22 -04001152static int path_lookup_open(int dfd, const char *name,
1153 unsigned int lookup_flags, struct nameidata *nd, int open_flags)
Trond Myklebust834f2a42005-10-18 14:20:16 -07001154{
Al Viro8737f3a2008-08-02 22:36:57 -04001155 struct file *filp = get_empty_filp();
1156 int err;
Trond Myklebust834f2a42005-10-18 14:20:16 -07001157
Al Viro8737f3a2008-08-02 22:36:57 -04001158 if (filp == NULL)
1159 return -ENFILE;
1160 nd->intent.open.file = filp;
1161 nd->intent.open.flags = open_flags;
1162 nd->intent.open.create_mode = 0;
1163 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1164 if (IS_ERR(nd->intent.open.file)) {
1165 if (err == 0) {
1166 err = PTR_ERR(nd->intent.open.file);
1167 path_put(&nd->path);
1168 }
1169 } else if (err != 0)
1170 release_open_intent(nd);
1171 return err;
Trond Myklebust834f2a42005-10-18 14:20:16 -07001172}
1173
Christoph Hellwigeead1912007-10-16 23:25:38 -07001174static struct dentry *__lookup_hash(struct qstr *name,
1175 struct dentry *base, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176{
James Morris057f6c02007-04-26 00:12:05 -07001177 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 struct inode *inode;
1179 int err;
1180
1181 inode = base->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182
1183 /*
1184 * See if the low-level filesystem might want
1185 * to use its own hash..
1186 */
1187 if (base->d_op && base->d_op->d_hash) {
1188 err = base->d_op->d_hash(base, name);
1189 dentry = ERR_PTR(err);
1190 if (err < 0)
1191 goto out;
1192 }
1193
1194 dentry = cached_lookup(base, name, nd);
1195 if (!dentry) {
Miklos Szeredid70b67c2008-07-02 21:30:15 +02001196 struct dentry *new;
1197
1198 /* Don't create child dentry for a dead directory. */
1199 dentry = ERR_PTR(-ENOENT);
1200 if (IS_DEADDIR(inode))
1201 goto out;
1202
1203 new = d_alloc(base, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 dentry = ERR_PTR(-ENOMEM);
1205 if (!new)
1206 goto out;
1207 dentry = inode->i_op->lookup(inode, new, nd);
1208 if (!dentry)
1209 dentry = new;
1210 else
1211 dput(new);
1212 }
1213out:
1214 return dentry;
1215}
1216
James Morris057f6c02007-04-26 00:12:05 -07001217/*
1218 * Restricted form of lookup. Doesn't follow links, single-component only,
1219 * needs parent already locked. Doesn't follow mounts.
1220 * SMP-safe.
1221 */
Adrian Bunka244e162006-03-31 02:32:11 -08001222static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223{
Christoph Hellwigeead1912007-10-16 23:25:38 -07001224 int err;
1225
Al Virof419a2e2008-07-22 00:07:17 -04001226 err = inode_permission(nd->path.dentry->d_inode, MAY_EXEC);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001227 if (err)
1228 return ERR_PTR(err);
Jan Blunck4ac91372008-02-14 19:34:32 -08001229 return __lookup_hash(&nd->last, nd->path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230}
1231
Christoph Hellwigeead1912007-10-16 23:25:38 -07001232static int __lookup_one_len(const char *name, struct qstr *this,
1233 struct dentry *base, int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234{
1235 unsigned long hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 unsigned int c;
1237
James Morris057f6c02007-04-26 00:12:05 -07001238 this->name = name;
1239 this->len = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 if (!len)
James Morris057f6c02007-04-26 00:12:05 -07001241 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242
1243 hash = init_name_hash();
1244 while (len--) {
1245 c = *(const unsigned char *)name++;
1246 if (c == '/' || c == '\0')
James Morris057f6c02007-04-26 00:12:05 -07001247 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 hash = partial_name_hash(c, hash);
1249 }
James Morris057f6c02007-04-26 00:12:05 -07001250 this->hash = end_name_hash(hash);
1251 return 0;
1252}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253
Christoph Hellwigeead1912007-10-16 23:25:38 -07001254/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001255 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07001256 * @name: pathname component to lookup
1257 * @base: base directory to lookup from
1258 * @len: maximum length @len should be interpreted to
1259 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07001260 * Note that this routine is purely a helper for filesystem usage and should
1261 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07001262 * nameidata argument is passed to the filesystem methods and a filesystem
1263 * using this helper needs to be prepared for that.
1264 */
James Morris057f6c02007-04-26 00:12:05 -07001265struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1266{
1267 int err;
1268 struct qstr this;
1269
David Woodhouse2f9092e2009-04-20 23:18:37 +01001270 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1271
James Morris057f6c02007-04-26 00:12:05 -07001272 err = __lookup_one_len(name, &this, base, len);
1273 if (err)
1274 return ERR_PTR(err);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001275
Al Virof419a2e2008-07-22 00:07:17 -04001276 err = inode_permission(base->d_inode, MAY_EXEC);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001277 if (err)
1278 return ERR_PTR(err);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001279 return __lookup_hash(&this, base, NULL);
James Morris057f6c02007-04-26 00:12:05 -07001280}
1281
Christoph Hellwigeead1912007-10-16 23:25:38 -07001282/**
1283 * lookup_one_noperm - bad hack for sysfs
1284 * @name: pathname component to lookup
1285 * @base: base directory to lookup from
1286 *
1287 * This is a variant of lookup_one_len that doesn't perform any permission
1288 * checks. It's a horrible hack to work around the braindead sysfs
1289 * architecture and should not be used anywhere else.
1290 *
1291 * DON'T USE THIS FUNCTION EVER, thanks.
1292 */
1293struct dentry *lookup_one_noperm(const char *name, struct dentry *base)
James Morris057f6c02007-04-26 00:12:05 -07001294{
1295 int err;
1296 struct qstr this;
1297
Christoph Hellwigeead1912007-10-16 23:25:38 -07001298 err = __lookup_one_len(name, &this, base, strlen(name));
James Morris057f6c02007-04-26 00:12:05 -07001299 if (err)
1300 return ERR_PTR(err);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001301 return __lookup_hash(&this, base, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302}
1303
Al Viro2d8f3032008-07-22 09:59:21 -04001304int user_path_at(int dfd, const char __user *name, unsigned flags,
1305 struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306{
Al Viro2d8f3032008-07-22 09:59:21 -04001307 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 char *tmp = getname(name);
1309 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04001311
1312 BUG_ON(flags & LOOKUP_PARENT);
1313
1314 err = do_path_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04001316 if (!err)
1317 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 }
1319 return err;
1320}
1321
Al Viro2ad94ae2008-07-21 09:32:51 -04001322static int user_path_parent(int dfd, const char __user *path,
1323 struct nameidata *nd, char **name)
1324{
1325 char *s = getname(path);
1326 int error;
1327
1328 if (IS_ERR(s))
1329 return PTR_ERR(s);
1330
1331 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1332 if (error)
1333 putname(s);
1334 else
1335 *name = s;
1336
1337 return error;
1338}
1339
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340/*
1341 * It's inline, so penalty for filesystems that don't use sticky bit is
1342 * minimal.
1343 */
1344static inline int check_sticky(struct inode *dir, struct inode *inode)
1345{
David Howellsda9592e2008-11-14 10:39:05 +11001346 uid_t fsuid = current_fsuid();
1347
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 if (!(dir->i_mode & S_ISVTX))
1349 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001350 if (inode->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001352 if (dir->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 return 0;
1354 return !capable(CAP_FOWNER);
1355}
1356
1357/*
1358 * Check whether we can remove a link victim from directory dir, check
1359 * whether the type of victim is right.
1360 * 1. We can't do it if dir is read-only (done in permission())
1361 * 2. We should have write and exec permissions on dir
1362 * 3. We can't remove anything from append-only dir
1363 * 4. We can't do anything with immutable dir (done in permission())
1364 * 5. If the sticky bit on dir is set we should either
1365 * a. be owner of dir, or
1366 * b. be owner of victim, or
1367 * c. have CAP_FOWNER capability
1368 * 6. If the victim is append-only or immutable we can't do antyhing with
1369 * links pointing to it.
1370 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1371 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1372 * 9. We can't remove a root or mountpoint.
1373 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1374 * nfs_async_unlink().
1375 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001376static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377{
1378 int error;
1379
1380 if (!victim->d_inode)
1381 return -ENOENT;
1382
1383 BUG_ON(victim->d_parent->d_inode != dir);
Al Viro5a190ae2007-06-07 12:19:32 -04001384 audit_inode_child(victim->d_name.name, victim, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385
Al Virof419a2e2008-07-22 00:07:17 -04001386 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 if (error)
1388 return error;
1389 if (IS_APPEND(dir))
1390 return -EPERM;
1391 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
Hugh Dickinsf9454542008-11-19 15:36:38 -08001392 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 return -EPERM;
1394 if (isdir) {
1395 if (!S_ISDIR(victim->d_inode->i_mode))
1396 return -ENOTDIR;
1397 if (IS_ROOT(victim))
1398 return -EBUSY;
1399 } else if (S_ISDIR(victim->d_inode->i_mode))
1400 return -EISDIR;
1401 if (IS_DEADDIR(dir))
1402 return -ENOENT;
1403 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1404 return -EBUSY;
1405 return 0;
1406}
1407
1408/* Check whether we can create an object with dentry child in directory
1409 * dir.
1410 * 1. We can't do it if child already exists (open has special treatment for
1411 * this case, but since we are inlined it's OK)
1412 * 2. We can't do it if dir is read-only (done in permission())
1413 * 3. We should have write and exec permissions on dir
1414 * 4. We can't do it if dir is immutable (done in permission())
1415 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02001416static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417{
1418 if (child->d_inode)
1419 return -EEXIST;
1420 if (IS_DEADDIR(dir))
1421 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04001422 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423}
1424
1425/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 * O_DIRECTORY translates into forcing a directory lookup.
1427 */
1428static inline int lookup_flags(unsigned int f)
1429{
1430 unsigned long retval = LOOKUP_FOLLOW;
1431
1432 if (f & O_NOFOLLOW)
1433 retval &= ~LOOKUP_FOLLOW;
1434
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 if (f & O_DIRECTORY)
1436 retval |= LOOKUP_DIRECTORY;
1437
1438 return retval;
1439}
1440
1441/*
1442 * p1 and p2 should be directories on the same fs.
1443 */
1444struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1445{
1446 struct dentry *p;
1447
1448 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001449 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 return NULL;
1451 }
1452
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001453 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001455 p = d_ancestor(p2, p1);
1456 if (p) {
1457 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1458 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1459 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 }
1461
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001462 p = d_ancestor(p1, p2);
1463 if (p) {
1464 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1465 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1466 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 }
1468
Ingo Molnarf2eace22006-07-03 00:25:05 -07001469 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1470 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 return NULL;
1472}
1473
1474void unlock_rename(struct dentry *p1, struct dentry *p2)
1475{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001476 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001478 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001479 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 }
1481}
1482
1483int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1484 struct nameidata *nd)
1485{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001486 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487
1488 if (error)
1489 return error;
1490
Al Viroacfa4382008-12-04 10:06:33 -05001491 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 return -EACCES; /* shouldn't it be ENOSYS? */
1493 mode &= S_IALLUGO;
1494 mode |= S_IFREG;
1495 error = security_inode_create(dir, dentry, mode);
1496 if (error)
1497 return error;
Jan Kara9e3509e2009-01-26 16:45:12 +01001498 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001500 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001501 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 return error;
1503}
1504
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001505int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001507 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 struct inode *inode = dentry->d_inode;
1509 int error;
1510
1511 if (!inode)
1512 return -ENOENT;
1513
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001514 switch (inode->i_mode & S_IFMT) {
1515 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001517 case S_IFDIR:
1518 if (acc_mode & MAY_WRITE)
1519 return -EISDIR;
1520 break;
1521 case S_IFBLK:
1522 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001523 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001525 /*FALLTHRU*/
1526 case S_IFIFO:
1527 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001529 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001530 }
Dave Hansenb41572e2007-10-16 23:31:14 -07001531
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001532 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07001533 if (error)
1534 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05001535
James Morriscb5629b2009-02-06 11:01:45 +11001536 error = ima_path_check(path,
Mimi Zoharb9fc7452009-05-19 13:25:57 -04001537 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC),
1538 IMA_COUNT_UPDATE);
Mimi Zohar6146f0d2009-02-04 09:06:57 -05001539 if (error)
1540 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 /*
1542 * An append-only file must be opened in append mode for writing.
1543 */
1544 if (IS_APPEND(inode)) {
1545 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1546 return -EPERM;
1547 if (flag & O_TRUNC)
1548 return -EPERM;
1549 }
1550
1551 /* O_NOATIME can only be set by the owner or superuser */
1552 if (flag & O_NOATIME)
Satyam Sharma3bd858a2007-07-17 15:00:08 +05301553 if (!is_owner_or_cap(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 return -EPERM;
1555
1556 /*
1557 * Ensure there are no outstanding leases on the file.
1558 */
1559 error = break_lease(inode, flag);
1560 if (error)
1561 return error;
1562
1563 if (flag & O_TRUNC) {
1564 error = get_write_access(inode);
1565 if (error)
1566 return error;
1567
1568 /*
1569 * Refuse to truncate files with mandatory locks held on them.
1570 */
1571 error = locks_verify_locked(inode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001572 if (!error)
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001573 error = security_path_truncate(path, 0,
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001574 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 if (!error) {
Jan Kara9e3509e2009-01-26 16:45:12 +01001576 vfs_dq_init(inode);
Miklos Szeredid139d7f2007-10-18 03:07:00 -07001577
1578 error = do_truncate(dentry, 0,
1579 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1580 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 }
1582 put_write_access(inode);
1583 if (error)
1584 return error;
1585 } else
1586 if (flag & FMODE_WRITE)
Jan Kara9e3509e2009-01-26 16:45:12 +01001587 vfs_dq_init(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588
1589 return 0;
1590}
1591
Dave Hansend57999e2008-02-15 14:37:27 -08001592/*
1593 * Be careful about ever adding any more callers of this
1594 * function. Its flags must be in the namei format, not
1595 * what get passed to sys_open().
1596 */
1597static int __open_namei_create(struct nameidata *nd, struct path *path,
Dave Hansenaab520e2006-09-30 23:29:02 -07001598 int flag, int mode)
1599{
1600 int error;
Jan Blunck4ac91372008-02-14 19:34:32 -08001601 struct dentry *dir = nd->path.dentry;
Dave Hansenaab520e2006-09-30 23:29:02 -07001602
1603 if (!IS_POSIXACL(dir->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04001604 mode &= ~current_umask();
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001605 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1606 if (error)
1607 goto out_unlock;
Dave Hansenaab520e2006-09-30 23:29:02 -07001608 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001609out_unlock:
Dave Hansenaab520e2006-09-30 23:29:02 -07001610 mutex_unlock(&dir->d_inode->i_mutex);
Jan Blunck4ac91372008-02-14 19:34:32 -08001611 dput(nd->path.dentry);
1612 nd->path.dentry = path->dentry;
Dave Hansenaab520e2006-09-30 23:29:02 -07001613 if (error)
1614 return error;
1615 /* Don't check for write permission, don't truncate */
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001616 return may_open(&nd->path, 0, flag & ~O_TRUNC);
Dave Hansenaab520e2006-09-30 23:29:02 -07001617}
1618
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619/*
Dave Hansend57999e2008-02-15 14:37:27 -08001620 * Note that while the flag value (low two bits) for sys_open means:
1621 * 00 - read-only
1622 * 01 - write-only
1623 * 10 - read-write
1624 * 11 - special
1625 * it is changed into
1626 * 00 - no permissions needed
1627 * 01 - read-permission
1628 * 10 - write-permission
1629 * 11 - read-write
1630 * for the internal routines (ie open_namei()/follow_link() etc)
1631 * This is more logical, and also allows the 00 "no perm needed"
1632 * to be used for symlinks (where the permissions are checked
1633 * later).
1634 *
1635*/
1636static inline int open_to_namei_flags(int flag)
1637{
1638 if ((flag+1) & O_ACCMODE)
1639 flag++;
1640 return flag;
1641}
1642
Dave Hansen4a3fd212008-02-15 14:37:48 -08001643static int open_will_write_to_fs(int flag, struct inode *inode)
1644{
1645 /*
1646 * We'll never write to the fs underlying
1647 * a device file.
1648 */
1649 if (special_file(inode->i_mode))
1650 return 0;
1651 return (flag & O_TRUNC);
1652}
1653
Dave Hansend57999e2008-02-15 14:37:27 -08001654/*
Dave Hansen4a3fd212008-02-15 14:37:48 -08001655 * Note that the low bits of the passed in "open_flag"
1656 * are not the same as in the local variable "flag". See
1657 * open_to_namei_flags() for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001659struct file *do_filp_open(int dfd, const char *pathname,
Al Viro6e8341a2009-04-06 11:16:22 -04001660 int open_flag, int mode, int acc_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661{
Dave Hansen4a3fd212008-02-15 14:37:48 -08001662 struct file *filp;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001663 struct nameidata nd;
Al Viro6e8341a2009-04-06 11:16:22 -04001664 int error;
Al Viro4e7506e2005-06-06 13:36:00 -07001665 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 struct dentry *dir;
1667 int count = 0;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001668 int will_write;
Dave Hansend57999e2008-02-15 14:37:27 -08001669 int flag = open_to_namei_flags(open_flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670
Al Viro6e8341a2009-04-06 11:16:22 -04001671 if (!acc_mode)
1672 acc_mode = MAY_OPEN | ACC_MODE(flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673
Trond Myklebust834f2a42005-10-18 14:20:16 -07001674 /* O_TRUNC implies we need access checks for write permissions */
1675 if (flag & O_TRUNC)
1676 acc_mode |= MAY_WRITE;
1677
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 /* Allow the LSM permission hook to distinguish append
1679 access from general write access. */
1680 if (flag & O_APPEND)
1681 acc_mode |= MAY_APPEND;
1682
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 /*
1684 * The simplest case - just a plain lookup.
1685 */
1686 if (!(flag & O_CREAT)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001687 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001688 &nd, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 if (error)
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001690 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691 goto ok;
1692 }
1693
1694 /*
1695 * Create - we need to know the parent.
1696 */
Al Viro9b4a9b12009-04-07 11:44:16 -04001697 error = path_init(dfd, pathname, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698 if (error)
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001699 return ERR_PTR(error);
Al Viro9b4a9b12009-04-07 11:44:16 -04001700 error = path_walk(pathname, &nd);
J. R. Okajima654f5622009-06-18 23:30:15 +09001701 if (error) {
1702 if (nd.root.mnt)
1703 path_put(&nd.root);
Al Viro9b4a9b12009-04-07 11:44:16 -04001704 return ERR_PTR(error);
J. R. Okajima654f5622009-06-18 23:30:15 +09001705 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001706 if (unlikely(!audit_dummy_context()))
1707 audit_inode(pathname, nd.path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708
1709 /*
1710 * We have the parent and last component. First of all, check
1711 * that we are not asked to creat(2) an obvious directory - that
1712 * will not do.
1713 */
1714 error = -EISDIR;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001715 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
Al Viro8737f3a2008-08-02 22:36:57 -04001716 goto exit_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717
Al Viro8737f3a2008-08-02 22:36:57 -04001718 error = -ENFILE;
1719 filp = get_empty_filp();
1720 if (filp == NULL)
1721 goto exit_parent;
1722 nd.intent.open.file = filp;
1723 nd.intent.open.flags = flag;
1724 nd.intent.open.create_mode = mode;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001725 dir = nd.path.dentry;
1726 nd.flags &= ~LOOKUP_PARENT;
Al Viro8737f3a2008-08-02 22:36:57 -04001727 nd.flags |= LOOKUP_CREATE | LOOKUP_OPEN;
Al Viro35165862008-08-05 03:00:49 -04001728 if (flag & O_EXCL)
1729 nd.flags |= LOOKUP_EXCL;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001730 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001731 path.dentry = lookup_hash(&nd);
1732 path.mnt = nd.path.mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733
1734do_last:
Al Viro4e7506e2005-06-06 13:36:00 -07001735 error = PTR_ERR(path.dentry);
1736 if (IS_ERR(path.dentry)) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001737 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738 goto exit;
1739 }
1740
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001741 if (IS_ERR(nd.intent.open.file)) {
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001742 error = PTR_ERR(nd.intent.open.file);
Dave Hansen4a3fd212008-02-15 14:37:48 -08001743 goto exit_mutex_unlock;
Oleg Drokin4af4c522006-03-25 03:06:54 -08001744 }
1745
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 /* Negative dentry, just create the file */
Al Viro4e7506e2005-06-06 13:36:00 -07001747 if (!path.dentry->d_inode) {
Dave Hansen4a3fd212008-02-15 14:37:48 -08001748 /*
1749 * This write is needed to ensure that a
1750 * ro->rw transition does not occur between
1751 * the time when the file is created and when
1752 * a permanent write count is taken through
1753 * the 'struct file' in nameidata_to_filp().
1754 */
1755 error = mnt_want_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 if (error)
Dave Hansen4a3fd212008-02-15 14:37:48 -08001757 goto exit_mutex_unlock;
1758 error = __open_namei_create(&nd, &path, flag, mode);
1759 if (error) {
1760 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 goto exit;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001762 }
1763 filp = nameidata_to_filp(&nd, open_flag);
1764 mnt_drop_write(nd.path.mnt);
J. R. Okajima654f5622009-06-18 23:30:15 +09001765 if (nd.root.mnt)
1766 path_put(&nd.root);
Dave Hansen4a3fd212008-02-15 14:37:48 -08001767 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 }
1769
1770 /*
1771 * It already exists.
1772 */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001773 mutex_unlock(&dir->d_inode->i_mutex);
Al Viro5a190ae2007-06-07 12:19:32 -04001774 audit_inode(pathname, path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775
1776 error = -EEXIST;
1777 if (flag & O_EXCL)
1778 goto exit_dput;
1779
Al Viroe13b2102005-06-06 13:36:06 -07001780 if (__follow_mount(&path)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781 error = -ELOOP;
Al Viroba7a4c12005-06-06 13:36:08 -07001782 if (flag & O_NOFOLLOW)
1783 goto exit_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 }
Amy Griffis3e2efce2006-07-13 13:16:02 -04001785
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 error = -ENOENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001787 if (!path.dentry->d_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 goto exit_dput;
Al Viroacfa4382008-12-04 10:06:33 -05001789 if (path.dentry->d_inode->i_op->follow_link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 goto do_link;
1791
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001792 path_to_nameidata(&path, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 error = -EISDIR;
Al Viro4e7506e2005-06-06 13:36:00 -07001794 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 goto exit;
1796ok:
Dave Hansen4a3fd212008-02-15 14:37:48 -08001797 /*
1798 * Consider:
1799 * 1. may_open() truncates a file
1800 * 2. a rw->ro mount transition occurs
1801 * 3. nameidata_to_filp() fails due to
1802 * the ro mount.
1803 * That would be inconsistent, and should
1804 * be avoided. Taking this mnt write here
1805 * ensures that (2) can not occur.
1806 */
1807 will_write = open_will_write_to_fs(flag, nd.path.dentry->d_inode);
1808 if (will_write) {
1809 error = mnt_want_write(nd.path.mnt);
1810 if (error)
1811 goto exit;
1812 }
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001813 error = may_open(&nd.path, acc_mode, flag);
Dave Hansen4a3fd212008-02-15 14:37:48 -08001814 if (error) {
1815 if (will_write)
1816 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 goto exit;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001818 }
1819 filp = nameidata_to_filp(&nd, open_flag);
1820 /*
1821 * It is now safe to drop the mnt write
1822 * because the filp has had a write taken
1823 * on its behalf.
1824 */
1825 if (will_write)
1826 mnt_drop_write(nd.path.mnt);
J. R. Okajima654f5622009-06-18 23:30:15 +09001827 if (nd.root.mnt)
1828 path_put(&nd.root);
Dave Hansen4a3fd212008-02-15 14:37:48 -08001829 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830
Dave Hansen4a3fd212008-02-15 14:37:48 -08001831exit_mutex_unlock:
1832 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833exit_dput:
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001834 path_put_conditional(&path, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835exit:
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001836 if (!IS_ERR(nd.intent.open.file))
1837 release_open_intent(&nd);
Al Viro8737f3a2008-08-02 22:36:57 -04001838exit_parent:
Al Viro2a737872009-04-07 11:49:53 -04001839 if (nd.root.mnt)
1840 path_put(&nd.root);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001841 path_put(&nd.path);
1842 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843
1844do_link:
1845 error = -ELOOP;
1846 if (flag & O_NOFOLLOW)
1847 goto exit_dput;
1848 /*
1849 * This is subtle. Instead of calling do_follow_link() we do the
1850 * thing by hands. The reason is that this way we have zero link_count
1851 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1852 * After that we have the parent and last component, i.e.
1853 * we are in the same situation as after the first path_walk().
1854 * Well, almost - if the last component is normal we get its copy
1855 * stored in nd->last.name and we will have to putname() it when we
1856 * are done. Procfs-like symlinks just set LAST_BIND.
1857 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001858 nd.flags |= LOOKUP_PARENT;
1859 error = security_inode_follow_link(path.dentry, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 if (error)
1861 goto exit_dput;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001862 error = __do_follow_link(&path, &nd);
Kirill Korotaevde459212006-07-14 00:23:49 -07001863 if (error) {
1864 /* Does someone understand code flow here? Or it is only
1865 * me so stupid? Anathema to whoever designed this non-sense
1866 * with "intent.open".
1867 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001868 release_open_intent(&nd);
J. R. Okajima654f5622009-06-18 23:30:15 +09001869 if (nd.root.mnt)
1870 path_put(&nd.root);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001871 return ERR_PTR(error);
Kirill Korotaevde459212006-07-14 00:23:49 -07001872 }
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001873 nd.flags &= ~LOOKUP_PARENT;
1874 if (nd.last_type == LAST_BIND)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875 goto ok;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876 error = -EISDIR;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001877 if (nd.last_type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 goto exit;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001879 if (nd.last.name[nd.last.len]) {
1880 __putname(nd.last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 goto exit;
1882 }
1883 error = -ELOOP;
1884 if (count++==32) {
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001885 __putname(nd.last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 goto exit;
1887 }
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001888 dir = nd.path.dentry;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001889 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001890 path.dentry = lookup_hash(&nd);
1891 path.mnt = nd.path.mnt;
1892 __putname(nd.last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893 goto do_last;
1894}
1895
1896/**
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001897 * filp_open - open file and return file pointer
1898 *
1899 * @filename: path to open
1900 * @flags: open flags as per the open(2) second argument
1901 * @mode: mode for the new file if O_CREAT is set, else ignored
1902 *
1903 * This is the helper to open a file from kernelspace if you really
1904 * have to. But in generally you should not do this, so please move
1905 * along, nothing to see here..
1906 */
1907struct file *filp_open(const char *filename, int flags, int mode)
1908{
Al Viro6e8341a2009-04-06 11:16:22 -04001909 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001910}
1911EXPORT_SYMBOL(filp_open);
1912
1913/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 * lookup_create - lookup a dentry, creating it if it doesn't exist
1915 * @nd: nameidata info
1916 * @is_dir: directory flag
1917 *
1918 * Simple function to lookup and return a dentry and create it
1919 * if it doesn't exist. Is SMP-safe.
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001920 *
Jan Blunck4ac91372008-02-14 19:34:32 -08001921 * Returns with nd->path.dentry->d_inode->i_mutex locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 */
1923struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1924{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001925 struct dentry *dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926
Jan Blunck4ac91372008-02-14 19:34:32 -08001927 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001928 /*
1929 * Yucky last component or no last component at all?
1930 * (foo/., foo/.., /////)
1931 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 if (nd->last_type != LAST_NORM)
1933 goto fail;
1934 nd->flags &= ~LOOKUP_PARENT;
Al Viro35165862008-08-05 03:00:49 -04001935 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
ASANO Masahiroa6349042006-08-22 20:06:02 -04001936 nd->intent.open.flags = O_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001937
1938 /*
1939 * Do the final lookup.
1940 */
Christoph Hellwig49705b72005-11-08 21:35:06 -08001941 dentry = lookup_hash(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942 if (IS_ERR(dentry))
1943 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001944
Al Viroe9baf6e2008-05-15 04:49:12 -04001945 if (dentry->d_inode)
1946 goto eexist;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001947 /*
1948 * Special case - lookup gave negative, but... we had foo/bar/
1949 * From the vfs_mknod() POV we just have a negative dentry -
1950 * all is fine. Let's be bastards - you had / on the end, you've
1951 * been asking for (non-existent) directory. -ENOENT for you.
1952 */
Al Viroe9baf6e2008-05-15 04:49:12 -04001953 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1954 dput(dentry);
1955 dentry = ERR_PTR(-ENOENT);
1956 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957 return dentry;
Al Viroe9baf6e2008-05-15 04:49:12 -04001958eexist:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959 dput(dentry);
Al Viroe9baf6e2008-05-15 04:49:12 -04001960 dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961fail:
1962 return dentry;
1963}
Christoph Hellwigf81a0bf2005-05-19 12:26:43 -07001964EXPORT_SYMBOL_GPL(lookup_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965
1966int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1967{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001968 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969
1970 if (error)
1971 return error;
1972
1973 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1974 return -EPERM;
1975
Al Viroacfa4382008-12-04 10:06:33 -05001976 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 return -EPERM;
1978
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07001979 error = devcgroup_inode_mknod(mode, dev);
1980 if (error)
1981 return error;
1982
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 error = security_inode_mknod(dir, dentry, mode, dev);
1984 if (error)
1985 return error;
1986
Jan Kara9e3509e2009-01-26 16:45:12 +01001987 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001989 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001990 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991 return error;
1992}
1993
Dave Hansen463c3192008-02-15 14:37:57 -08001994static int may_mknod(mode_t mode)
1995{
1996 switch (mode & S_IFMT) {
1997 case S_IFREG:
1998 case S_IFCHR:
1999 case S_IFBLK:
2000 case S_IFIFO:
2001 case S_IFSOCK:
2002 case 0: /* zero mode translates to S_IFREG */
2003 return 0;
2004 case S_IFDIR:
2005 return -EPERM;
2006 default:
2007 return -EINVAL;
2008 }
2009}
2010
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002011SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2012 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013{
Al Viro2ad94ae2008-07-21 09:32:51 -04002014 int error;
2015 char *tmp;
2016 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 struct nameidata nd;
2018
2019 if (S_ISDIR(mode))
2020 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021
Al Viro2ad94ae2008-07-21 09:32:51 -04002022 error = user_path_parent(dfd, filename, &nd, &tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002024 return error;
2025
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026 dentry = lookup_create(&nd, 0);
Dave Hansen463c3192008-02-15 14:37:57 -08002027 if (IS_ERR(dentry)) {
2028 error = PTR_ERR(dentry);
2029 goto out_unlock;
2030 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002031 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002032 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002033 error = may_mknod(mode);
2034 if (error)
2035 goto out_dput;
2036 error = mnt_want_write(nd.path.mnt);
2037 if (error)
2038 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002039 error = security_path_mknod(&nd.path, dentry, mode, dev);
2040 if (error)
2041 goto out_drop_write;
Dave Hansen463c3192008-02-15 14:37:57 -08002042 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 case 0: case S_IFREG:
Jan Blunck4ac91372008-02-14 19:34:32 -08002044 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 break;
2046 case S_IFCHR: case S_IFBLK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002047 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 new_decode_dev(dev));
2049 break;
2050 case S_IFIFO: case S_IFSOCK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002051 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002054out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002055 mnt_drop_write(nd.path.mnt);
2056out_dput:
2057 dput(dentry);
2058out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002059 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002060 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 putname(tmp);
2062
2063 return error;
2064}
2065
Heiko Carstens3480b252009-01-14 14:14:16 +01002066SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002067{
2068 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2069}
2070
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2072{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002073 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074
2075 if (error)
2076 return error;
2077
Al Viroacfa4382008-12-04 10:06:33 -05002078 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079 return -EPERM;
2080
2081 mode &= (S_IRWXUGO|S_ISVTX);
2082 error = security_inode_mkdir(dir, dentry, mode);
2083 if (error)
2084 return error;
2085
Jan Kara9e3509e2009-01-26 16:45:12 +01002086 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002088 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002089 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 return error;
2091}
2092
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002093SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094{
2095 int error = 0;
2096 char * tmp;
Dave Hansen6902d922006-09-30 23:29:01 -07002097 struct dentry *dentry;
2098 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099
Al Viro2ad94ae2008-07-21 09:32:51 -04002100 error = user_path_parent(dfd, pathname, &nd, &tmp);
2101 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002102 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103
Dave Hansen6902d922006-09-30 23:29:01 -07002104 dentry = lookup_create(&nd, 1);
2105 error = PTR_ERR(dentry);
2106 if (IS_ERR(dentry))
2107 goto out_unlock;
2108
Jan Blunck4ac91372008-02-14 19:34:32 -08002109 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002110 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002111 error = mnt_want_write(nd.path.mnt);
2112 if (error)
2113 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002114 error = security_path_mkdir(&nd.path, dentry, mode);
2115 if (error)
2116 goto out_drop_write;
Jan Blunck4ac91372008-02-14 19:34:32 -08002117 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002118out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002119 mnt_drop_write(nd.path.mnt);
2120out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002121 dput(dentry);
2122out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002123 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002124 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002125 putname(tmp);
2126out_err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127 return error;
2128}
2129
Heiko Carstens3cdad422009-01-14 14:14:22 +01002130SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002131{
2132 return sys_mkdirat(AT_FDCWD, pathname, mode);
2133}
2134
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135/*
2136 * We try to drop the dentry early: we should have
2137 * a usage count of 2 if we're the only user of this
2138 * dentry, and if that is true (possibly after pruning
2139 * the dcache), then we drop the dentry now.
2140 *
2141 * A low-level filesystem can, if it choses, legally
2142 * do a
2143 *
2144 * if (!d_unhashed(dentry))
2145 * return -EBUSY;
2146 *
2147 * if it cannot handle the case of removing a directory
2148 * that is still in use by something else..
2149 */
2150void dentry_unhash(struct dentry *dentry)
2151{
2152 dget(dentry);
Vasily Averindc168422006-12-06 20:37:07 -08002153 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154 spin_lock(&dcache_lock);
2155 spin_lock(&dentry->d_lock);
2156 if (atomic_read(&dentry->d_count) == 2)
2157 __d_drop(dentry);
2158 spin_unlock(&dentry->d_lock);
2159 spin_unlock(&dcache_lock);
2160}
2161
2162int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2163{
2164 int error = may_delete(dir, dentry, 1);
2165
2166 if (error)
2167 return error;
2168
Al Viroacfa4382008-12-04 10:06:33 -05002169 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 return -EPERM;
2171
Jan Kara9e3509e2009-01-26 16:45:12 +01002172 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002174 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 dentry_unhash(dentry);
2176 if (d_mountpoint(dentry))
2177 error = -EBUSY;
2178 else {
2179 error = security_inode_rmdir(dir, dentry);
2180 if (!error) {
2181 error = dir->i_op->rmdir(dir, dentry);
2182 if (!error)
2183 dentry->d_inode->i_flags |= S_DEAD;
2184 }
2185 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002186 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 d_delete(dentry);
2189 }
2190 dput(dentry);
2191
2192 return error;
2193}
2194
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002195static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196{
2197 int error = 0;
2198 char * name;
2199 struct dentry *dentry;
2200 struct nameidata nd;
2201
Al Viro2ad94ae2008-07-21 09:32:51 -04002202 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002204 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205
2206 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002207 case LAST_DOTDOT:
2208 error = -ENOTEMPTY;
2209 goto exit1;
2210 case LAST_DOT:
2211 error = -EINVAL;
2212 goto exit1;
2213 case LAST_ROOT:
2214 error = -EBUSY;
2215 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002217
2218 nd.flags &= ~LOOKUP_PARENT;
2219
Jan Blunck4ac91372008-02-14 19:34:32 -08002220 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002221 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002223 if (IS_ERR(dentry))
2224 goto exit2;
Dave Hansen06227532008-02-15 14:37:34 -08002225 error = mnt_want_write(nd.path.mnt);
2226 if (error)
2227 goto exit3;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002228 error = security_path_rmdir(&nd.path, dentry);
2229 if (error)
2230 goto exit4;
Jan Blunck4ac91372008-02-14 19:34:32 -08002231 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002232exit4:
Dave Hansen06227532008-02-15 14:37:34 -08002233 mnt_drop_write(nd.path.mnt);
2234exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07002235 dput(dentry);
2236exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08002237 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002239 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 putname(name);
2241 return error;
2242}
2243
Heiko Carstens3cdad422009-01-14 14:14:22 +01002244SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002245{
2246 return do_rmdir(AT_FDCWD, pathname);
2247}
2248
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249int vfs_unlink(struct inode *dir, struct dentry *dentry)
2250{
2251 int error = may_delete(dir, dentry, 0);
2252
2253 if (error)
2254 return error;
2255
Al Viroacfa4382008-12-04 10:06:33 -05002256 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257 return -EPERM;
2258
Jan Kara9e3509e2009-01-26 16:45:12 +01002259 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002261 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262 if (d_mountpoint(dentry))
2263 error = -EBUSY;
2264 else {
2265 error = security_inode_unlink(dir, dentry);
2266 if (!error)
2267 error = dir->i_op->unlink(dir, dentry);
2268 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002269 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270
2271 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2272 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08002273 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275 }
Robert Love0eeca282005-07-12 17:06:03 -04002276
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277 return error;
2278}
2279
2280/*
2281 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002282 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283 * writeout happening, and we don't want to prevent access to the directory
2284 * while waiting on the I/O.
2285 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002286static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287{
Al Viro2ad94ae2008-07-21 09:32:51 -04002288 int error;
2289 char *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290 struct dentry *dentry;
2291 struct nameidata nd;
2292 struct inode *inode = NULL;
2293
Al Viro2ad94ae2008-07-21 09:32:51 -04002294 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002296 return error;
2297
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298 error = -EISDIR;
2299 if (nd.last_type != LAST_NORM)
2300 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002301
2302 nd.flags &= ~LOOKUP_PARENT;
2303
Jan Blunck4ac91372008-02-14 19:34:32 -08002304 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002305 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306 error = PTR_ERR(dentry);
2307 if (!IS_ERR(dentry)) {
2308 /* Why not before? Because we want correct error value */
2309 if (nd.last.name[nd.last.len])
2310 goto slashes;
2311 inode = dentry->d_inode;
2312 if (inode)
2313 atomic_inc(&inode->i_count);
Dave Hansen06227532008-02-15 14:37:34 -08002314 error = mnt_want_write(nd.path.mnt);
2315 if (error)
2316 goto exit2;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002317 error = security_path_unlink(&nd.path, dentry);
2318 if (error)
2319 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08002320 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002321exit3:
Dave Hansen06227532008-02-15 14:37:34 -08002322 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323 exit2:
2324 dput(dentry);
2325 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002326 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327 if (inode)
2328 iput(inode); /* truncate the inode here */
2329exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002330 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331 putname(name);
2332 return error;
2333
2334slashes:
2335 error = !dentry->d_inode ? -ENOENT :
2336 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2337 goto exit2;
2338}
2339
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002340SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002341{
2342 if ((flag & ~AT_REMOVEDIR) != 0)
2343 return -EINVAL;
2344
2345 if (flag & AT_REMOVEDIR)
2346 return do_rmdir(dfd, pathname);
2347
2348 return do_unlinkat(dfd, pathname);
2349}
2350
Heiko Carstens3480b252009-01-14 14:14:16 +01002351SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002352{
2353 return do_unlinkat(AT_FDCWD, pathname);
2354}
2355
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002356int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002358 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359
2360 if (error)
2361 return error;
2362
Al Viroacfa4382008-12-04 10:06:33 -05002363 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364 return -EPERM;
2365
2366 error = security_inode_symlink(dir, dentry, oldname);
2367 if (error)
2368 return error;
2369
Jan Kara9e3509e2009-01-26 16:45:12 +01002370 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002372 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002373 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374 return error;
2375}
2376
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002377SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2378 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379{
Al Viro2ad94ae2008-07-21 09:32:51 -04002380 int error;
2381 char *from;
2382 char *to;
Dave Hansen6902d922006-09-30 23:29:01 -07002383 struct dentry *dentry;
2384 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385
2386 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04002387 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388 return PTR_ERR(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002389
2390 error = user_path_parent(newdfd, newname, &nd, &to);
2391 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002392 goto out_putname;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393
Dave Hansen6902d922006-09-30 23:29:01 -07002394 dentry = lookup_create(&nd, 0);
2395 error = PTR_ERR(dentry);
2396 if (IS_ERR(dentry))
2397 goto out_unlock;
2398
Dave Hansen75c3f292008-02-15 14:37:45 -08002399 error = mnt_want_write(nd.path.mnt);
2400 if (error)
2401 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002402 error = security_path_symlink(&nd.path, dentry, from);
2403 if (error)
2404 goto out_drop_write;
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002405 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002406out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002407 mnt_drop_write(nd.path.mnt);
2408out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002409 dput(dentry);
2410out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002411 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002412 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002413 putname(to);
2414out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415 putname(from);
2416 return error;
2417}
2418
Heiko Carstens3480b252009-01-14 14:14:16 +01002419SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002420{
2421 return sys_symlinkat(oldname, AT_FDCWD, newname);
2422}
2423
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2425{
2426 struct inode *inode = old_dentry->d_inode;
2427 int error;
2428
2429 if (!inode)
2430 return -ENOENT;
2431
Miklos Szeredia95164d2008-07-30 15:08:48 +02002432 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433 if (error)
2434 return error;
2435
2436 if (dir->i_sb != inode->i_sb)
2437 return -EXDEV;
2438
2439 /*
2440 * A link to an append-only or immutable file cannot be created.
2441 */
2442 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2443 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05002444 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002446 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447 return -EPERM;
2448
2449 error = security_inode_link(old_dentry, dir, new_dentry);
2450 if (error)
2451 return error;
2452
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002453 mutex_lock(&inode->i_mutex);
Jan Kara9e3509e2009-01-26 16:45:12 +01002454 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455 error = dir->i_op->link(old_dentry, dir, new_dentry);
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002456 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002457 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002458 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459 return error;
2460}
2461
2462/*
2463 * Hardlinks are often used in delicate situations. We avoid
2464 * security-related surprises by not following symlinks on the
2465 * newname. --KAB
2466 *
2467 * We don't follow them on the oldname either to be compatible
2468 * with linux 2.0, and to avoid hard-linking to directories
2469 * and other special files. --ADM
2470 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002471SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2472 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002473{
2474 struct dentry *new_dentry;
Al Viro2d8f3032008-07-22 09:59:21 -04002475 struct nameidata nd;
2476 struct path old_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477 int error;
Al Viro2ad94ae2008-07-21 09:32:51 -04002478 char *to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479
Ulrich Drepper45c9b112006-06-25 05:49:11 -07002480 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002481 return -EINVAL;
2482
Al Viro2d8f3032008-07-22 09:59:21 -04002483 error = user_path_at(olddfd, oldname,
2484 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2485 &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002486 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002487 return error;
2488
2489 error = user_path_parent(newdfd, newname, &nd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490 if (error)
2491 goto out;
2492 error = -EXDEV;
Al Viro2d8f3032008-07-22 09:59:21 -04002493 if (old_path.mnt != nd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494 goto out_release;
2495 new_dentry = lookup_create(&nd, 0);
2496 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002497 if (IS_ERR(new_dentry))
2498 goto out_unlock;
Dave Hansen75c3f292008-02-15 14:37:45 -08002499 error = mnt_want_write(nd.path.mnt);
2500 if (error)
2501 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002502 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2503 if (error)
2504 goto out_drop_write;
Al Viro2d8f3032008-07-22 09:59:21 -04002505 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002506out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002507 mnt_drop_write(nd.path.mnt);
2508out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002509 dput(new_dentry);
2510out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002511 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002512out_release:
Jan Blunck1d957f92008-02-14 19:34:35 -08002513 path_put(&nd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002514 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515out:
Al Viro2d8f3032008-07-22 09:59:21 -04002516 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517
2518 return error;
2519}
2520
Heiko Carstens3480b252009-01-14 14:14:16 +01002521SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002522{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002523 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002524}
2525
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526/*
2527 * The worst of all namespace operations - renaming directory. "Perverted"
2528 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2529 * Problems:
2530 * a) we can get into loop creation. Check is done in is_subdir().
2531 * b) race potential - two innocent renames can create a loop together.
2532 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002533 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534 * story.
2535 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002536 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537 * whether the target exists). Solution: try to be smart with locking
2538 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002539 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540 * move will be locked. Thus we can rank directories by the tree
2541 * (ancestors first) and rank all non-directories after them.
2542 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002543 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544 * HOWEVER, it relies on the assumption that any object with ->lookup()
2545 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2546 * we'd better make sure that there's no link(2) for them.
2547 * d) some filesystems don't support opened-but-unlinked directories,
2548 * either because of layout or because they are not ready to deal with
2549 * all cases correctly. The latter will be fixed (taking this sort of
2550 * stuff into VFS), but the former is not going away. Solution: the same
2551 * trick as in rmdir().
2552 * e) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002553 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002555 * ->i_mutex on parents, which works but leads to some truely excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556 * locking].
2557 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07002558static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2559 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560{
2561 int error = 0;
2562 struct inode *target;
2563
2564 /*
2565 * If we are going to change the parent - check write permissions,
2566 * we'll need to flip '..'.
2567 */
2568 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04002569 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570 if (error)
2571 return error;
2572 }
2573
2574 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2575 if (error)
2576 return error;
2577
2578 target = new_dentry->d_inode;
2579 if (target) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002580 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581 dentry_unhash(new_dentry);
2582 }
2583 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2584 error = -EBUSY;
2585 else
2586 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2587 if (target) {
2588 if (!error)
2589 target->i_flags |= S_DEAD;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002590 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591 if (d_unhashed(new_dentry))
2592 d_rehash(new_dentry);
2593 dput(new_dentry);
2594 }
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002595 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07002596 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2597 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598 return error;
2599}
2600
Adrian Bunk75c96f82005-05-05 16:16:09 -07002601static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2602 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603{
2604 struct inode *target;
2605 int error;
2606
2607 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2608 if (error)
2609 return error;
2610
2611 dget(new_dentry);
2612 target = new_dentry->d_inode;
2613 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002614 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2616 error = -EBUSY;
2617 else
2618 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2619 if (!error) {
Mark Fasheh349457c2006-09-08 14:22:21 -07002620 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621 d_move(old_dentry, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622 }
2623 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002624 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625 dput(new_dentry);
2626 return error;
2627}
2628
2629int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2630 struct inode *new_dir, struct dentry *new_dentry)
2631{
2632 int error;
2633 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Robert Love0eeca282005-07-12 17:06:03 -04002634 const char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635
2636 if (old_dentry->d_inode == new_dentry->d_inode)
2637 return 0;
2638
2639 error = may_delete(old_dir, old_dentry, is_dir);
2640 if (error)
2641 return error;
2642
2643 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02002644 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645 else
2646 error = may_delete(new_dir, new_dentry, is_dir);
2647 if (error)
2648 return error;
2649
Al Viroacfa4382008-12-04 10:06:33 -05002650 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651 return -EPERM;
2652
Jan Kara9e3509e2009-01-26 16:45:12 +01002653 vfs_dq_init(old_dir);
2654 vfs_dq_init(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655
Robert Love0eeca282005-07-12 17:06:03 -04002656 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2657
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658 if (is_dir)
2659 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2660 else
2661 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2662 if (!error) {
Robert Love0eeca282005-07-12 17:06:03 -04002663 const char *new_name = old_dentry->d_name.name;
John McCutchan89204c42005-08-15 12:13:28 -04002664 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04002665 new_dentry->d_inode, old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666 }
Robert Love0eeca282005-07-12 17:06:03 -04002667 fsnotify_oldname_free(old_name);
2668
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669 return error;
2670}
2671
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002672SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2673 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674{
Al Viro2ad94ae2008-07-21 09:32:51 -04002675 struct dentry *old_dir, *new_dir;
2676 struct dentry *old_dentry, *new_dentry;
2677 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678 struct nameidata oldnd, newnd;
Al Viro2ad94ae2008-07-21 09:32:51 -04002679 char *from;
2680 char *to;
2681 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682
Al Viro2ad94ae2008-07-21 09:32:51 -04002683 error = user_path_parent(olddfd, oldname, &oldnd, &from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002684 if (error)
2685 goto exit;
2686
Al Viro2ad94ae2008-07-21 09:32:51 -04002687 error = user_path_parent(newdfd, newname, &newnd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688 if (error)
2689 goto exit1;
2690
2691 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08002692 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693 goto exit2;
2694
Jan Blunck4ac91372008-02-14 19:34:32 -08002695 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002696 error = -EBUSY;
2697 if (oldnd.last_type != LAST_NORM)
2698 goto exit2;
2699
Jan Blunck4ac91372008-02-14 19:34:32 -08002700 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002701 if (newnd.last_type != LAST_NORM)
2702 goto exit2;
2703
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002704 oldnd.flags &= ~LOOKUP_PARENT;
2705 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09002706 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002707
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708 trap = lock_rename(new_dir, old_dir);
2709
Christoph Hellwig49705b72005-11-08 21:35:06 -08002710 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711 error = PTR_ERR(old_dentry);
2712 if (IS_ERR(old_dentry))
2713 goto exit3;
2714 /* source must exist */
2715 error = -ENOENT;
2716 if (!old_dentry->d_inode)
2717 goto exit4;
2718 /* unless the source is a directory trailing slashes give -ENOTDIR */
2719 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2720 error = -ENOTDIR;
2721 if (oldnd.last.name[oldnd.last.len])
2722 goto exit4;
2723 if (newnd.last.name[newnd.last.len])
2724 goto exit4;
2725 }
2726 /* source should not be ancestor of target */
2727 error = -EINVAL;
2728 if (old_dentry == trap)
2729 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08002730 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002731 error = PTR_ERR(new_dentry);
2732 if (IS_ERR(new_dentry))
2733 goto exit4;
2734 /* target should not be an ancestor of source */
2735 error = -ENOTEMPTY;
2736 if (new_dentry == trap)
2737 goto exit5;
2738
Dave Hansen9079b1e2008-02-15 14:37:49 -08002739 error = mnt_want_write(oldnd.path.mnt);
2740 if (error)
2741 goto exit5;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002742 error = security_path_rename(&oldnd.path, old_dentry,
2743 &newnd.path, new_dentry);
2744 if (error)
2745 goto exit6;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746 error = vfs_rename(old_dir->d_inode, old_dentry,
2747 new_dir->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002748exit6:
Dave Hansen9079b1e2008-02-15 14:37:49 -08002749 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750exit5:
2751 dput(new_dentry);
2752exit4:
2753 dput(old_dentry);
2754exit3:
2755 unlock_rename(new_dir, old_dir);
2756exit2:
Jan Blunck1d957f92008-02-14 19:34:35 -08002757 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002758 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002760 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761 putname(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002762exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002763 return error;
2764}
2765
Heiko Carstensa26eab22009-01-14 14:14:17 +01002766SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002767{
2768 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2769}
2770
Linus Torvalds1da177e2005-04-16 15:20:36 -07002771int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2772{
2773 int len;
2774
2775 len = PTR_ERR(link);
2776 if (IS_ERR(link))
2777 goto out;
2778
2779 len = strlen(link);
2780 if (len > (unsigned) buflen)
2781 len = buflen;
2782 if (copy_to_user(buffer, link, len))
2783 len = -EFAULT;
2784out:
2785 return len;
2786}
2787
2788/*
2789 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2790 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2791 * using) it for any given inode is up to filesystem.
2792 */
2793int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2794{
2795 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002796 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07002797 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002798
Linus Torvalds1da177e2005-04-16 15:20:36 -07002799 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002800 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07002801 if (IS_ERR(cookie))
2802 return PTR_ERR(cookie);
2803
2804 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2805 if (dentry->d_inode->i_op->put_link)
2806 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2807 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002808}
2809
2810int vfs_follow_link(struct nameidata *nd, const char *link)
2811{
2812 return __vfs_follow_link(nd, link);
2813}
2814
2815/* get the link contents into pagecache */
2816static char *page_getlink(struct dentry * dentry, struct page **ppage)
2817{
Duane Griffinebd09ab2008-12-19 20:47:12 +00002818 char *kaddr;
2819 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07002821 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07002823 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002824 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00002825 kaddr = kmap(page);
2826 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2827 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002828}
2829
2830int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2831{
2832 struct page *page = NULL;
2833 char *s = page_getlink(dentry, &page);
2834 int res = vfs_readlink(dentry,buffer,buflen,s);
2835 if (page) {
2836 kunmap(page);
2837 page_cache_release(page);
2838 }
2839 return res;
2840}
2841
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002842void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002844 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002846 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002847}
2848
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002849void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002850{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002851 struct page *page = cookie;
2852
2853 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002854 kunmap(page);
2855 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002856 }
2857}
2858
Nick Piggin54566b22009-01-04 12:00:53 -08002859/*
2860 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2861 */
2862int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863{
2864 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002865 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002866 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08002867 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08002869 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2870 if (nofs)
2871 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872
NeilBrown7e53cac2006-03-25 03:07:57 -08002873retry:
Nick Pigginafddba42007-10-16 01:25:01 -07002874 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08002875 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002876 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07002877 goto fail;
2878
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879 kaddr = kmap_atomic(page, KM_USER0);
2880 memcpy(kaddr, symname, len-1);
2881 kunmap_atomic(kaddr, KM_USER0);
Nick Pigginafddba42007-10-16 01:25:01 -07002882
2883 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2884 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002885 if (err < 0)
2886 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07002887 if (err < len-1)
2888 goto retry;
2889
Linus Torvalds1da177e2005-04-16 15:20:36 -07002890 mark_inode_dirty(inode);
2891 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002892fail:
2893 return err;
2894}
2895
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002896int page_symlink(struct inode *inode, const char *symname, int len)
2897{
2898 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08002899 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002900}
2901
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002902const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002903 .readlink = generic_readlink,
2904 .follow_link = page_follow_link_light,
2905 .put_link = page_put_link,
2906};
2907
Al Viro2d8f3032008-07-22 09:59:21 -04002908EXPORT_SYMBOL(user_path_at);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909EXPORT_SYMBOL(follow_down);
2910EXPORT_SYMBOL(follow_up);
2911EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2912EXPORT_SYMBOL(getname);
2913EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002914EXPORT_SYMBOL(lookup_one_len);
2915EXPORT_SYMBOL(page_follow_link_light);
2916EXPORT_SYMBOL(page_put_link);
2917EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002918EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002919EXPORT_SYMBOL(page_symlink);
2920EXPORT_SYMBOL(page_symlink_inode_operations);
2921EXPORT_SYMBOL(path_lookup);
Al Virod1811462008-08-02 00:49:18 -04002922EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002923EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04002924EXPORT_SYMBOL(inode_permission);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -08002925EXPORT_SYMBOL(file_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002926EXPORT_SYMBOL(unlock_rename);
2927EXPORT_SYMBOL(vfs_create);
2928EXPORT_SYMBOL(vfs_follow_link);
2929EXPORT_SYMBOL(vfs_link);
2930EXPORT_SYMBOL(vfs_mkdir);
2931EXPORT_SYMBOL(vfs_mknod);
2932EXPORT_SYMBOL(generic_permission);
2933EXPORT_SYMBOL(vfs_readlink);
2934EXPORT_SYMBOL(vfs_rename);
2935EXPORT_SYMBOL(vfs_rmdir);
2936EXPORT_SYMBOL(vfs_symlink);
2937EXPORT_SYMBOL(vfs_unlink);
2938EXPORT_SYMBOL(dentry_unhash);
2939EXPORT_SYMBOL(generic_readlink);