blob: b8433ebfae055424c5d953ff3ff9c96bc7c7a628 [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 Blunck5dd784d02008-02-14 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
Arjan van de Venf1662352006-01-14 13:21:31 -0800555static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556{
557 int res = 0;
558 char *name;
559 if (IS_ERR(link))
560 goto fail;
561
562 if (*link == '/') {
Christoph Hellwigb4091d52008-11-05 15:07:21 +0100563 struct fs_struct *fs = current->fs;
564
Jan Blunck1d957f92008-02-14 19:34:35 -0800565 path_put(&nd->path);
Christoph Hellwigb4091d52008-11-05 15:07:21 +0100566
567 read_lock(&fs->lock);
568 nd->path = fs->root;
569 path_get(&fs->root);
570 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 }
Christoph Hellwigb4091d52008-11-05 15:07:21 +0100572
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 res = link_path_walk(link, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 if (nd->depth || res || nd->last_type!=LAST_NORM)
575 return res;
576 /*
577 * If it is an iterative symlinks resolution in open_namei() we
578 * have to copy the last component. And all that crap because of
579 * bloody create() on broken symlinks. Furrfu...
580 */
581 name = __getname();
582 if (unlikely(!name)) {
Jan Blunck1d957f92008-02-14 19:34:35 -0800583 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 return -ENOMEM;
585 }
586 strcpy(name, nd->last.name);
587 nd->last.name = name;
588 return 0;
589fail:
Jan Blunck1d957f92008-02-14 19:34:35 -0800590 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 return PTR_ERR(link);
592}
593
Jan Blunck1d957f92008-02-14 19:34:35 -0800594static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700595{
596 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800597 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700598 mntput(path->mnt);
599}
600
601static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
602{
Jan Blunck4ac91372008-02-14 19:34:32 -0800603 dput(nd->path.dentry);
604 if (nd->path.mnt != path->mnt)
605 mntput(nd->path.mnt);
606 nd->path.mnt = path->mnt;
607 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700608}
609
Ian Kent051d3812006-03-27 01:14:53 -0800610static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
611{
612 int error;
613 void *cookie;
614 struct dentry *dentry = path->dentry;
615
616 touch_atime(path->mnt, dentry);
617 nd_set_link(nd, NULL);
618
Jan Blunck4ac91372008-02-14 19:34:32 -0800619 if (path->mnt != nd->path.mnt) {
Ian Kent051d3812006-03-27 01:14:53 -0800620 path_to_nameidata(path, nd);
621 dget(dentry);
622 }
623 mntget(path->mnt);
624 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
625 error = PTR_ERR(cookie);
626 if (!IS_ERR(cookie)) {
627 char *s = nd_get_link(nd);
628 error = 0;
629 if (s)
630 error = __vfs_follow_link(nd, s);
631 if (dentry->d_inode->i_op->put_link)
632 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
633 }
Jan Blunck09da59162008-02-14 19:34:37 -0800634 path_put(path);
Ian Kent051d3812006-03-27 01:14:53 -0800635
636 return error;
637}
638
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639/*
640 * This limits recursive symlink follows to 8, while
641 * limiting consecutive symlinks to 40.
642 *
643 * Without that kind of total limit, nasty chains of consecutive
644 * symlinks can cause almost arbitrarily long lookups.
645 */
Al Viro90ebe562005-06-06 13:35:58 -0700646static inline int do_follow_link(struct path *path, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647{
648 int err = -ELOOP;
649 if (current->link_count >= MAX_NESTED_LINKS)
650 goto loop;
651 if (current->total_link_count >= 40)
652 goto loop;
653 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
654 cond_resched();
Al Viro90ebe562005-06-06 13:35:58 -0700655 err = security_inode_follow_link(path->dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 if (err)
657 goto loop;
658 current->link_count++;
659 current->total_link_count++;
660 nd->depth++;
Al Virocd4e91d2005-06-06 13:36:03 -0700661 err = __do_follow_link(path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 current->link_count--;
663 nd->depth--;
664 return err;
665loop:
Jan Blunck1d957f92008-02-14 19:34:35 -0800666 path_put_conditional(path, nd);
667 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 return err;
669}
670
671int follow_up(struct vfsmount **mnt, struct dentry **dentry)
672{
673 struct vfsmount *parent;
674 struct dentry *mountpoint;
675 spin_lock(&vfsmount_lock);
676 parent=(*mnt)->mnt_parent;
677 if (parent == *mnt) {
678 spin_unlock(&vfsmount_lock);
679 return 0;
680 }
681 mntget(parent);
682 mountpoint=dget((*mnt)->mnt_mountpoint);
683 spin_unlock(&vfsmount_lock);
684 dput(*dentry);
685 *dentry = mountpoint;
686 mntput(*mnt);
687 *mnt = parent;
688 return 1;
689}
690
691/* no need for dcache_lock, as serialization is taken care in
692 * namespace.c
693 */
Al Viro463ffb22005-06-06 13:36:05 -0700694static int __follow_mount(struct path *path)
695{
696 int res = 0;
697 while (d_mountpoint(path->dentry)) {
698 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
699 if (!mounted)
700 break;
701 dput(path->dentry);
702 if (res)
703 mntput(path->mnt);
704 path->mnt = mounted;
705 path->dentry = dget(mounted->mnt_root);
706 res = 1;
707 }
708 return res;
709}
710
Al Viro58c465e2005-06-06 13:36:13 -0700711static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 while (d_mountpoint(*dentry)) {
714 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
715 if (!mounted)
716 break;
Al Viro58c465e2005-06-06 13:36:13 -0700717 dput(*dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 mntput(*mnt);
719 *mnt = mounted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 *dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722}
723
724/* no need for dcache_lock, as serialization is taken care in
725 * namespace.c
726 */
Al Viroe13b2102005-06-06 13:36:06 -0700727int follow_down(struct vfsmount **mnt, struct dentry **dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728{
729 struct vfsmount *mounted;
730
731 mounted = lookup_mnt(*mnt, *dentry);
732 if (mounted) {
Al Viroe13b2102005-06-06 13:36:06 -0700733 dput(*dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 mntput(*mnt);
735 *mnt = mounted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 *dentry = dget(mounted->mnt_root);
737 return 1;
738 }
739 return 0;
740}
741
Arjan van de Venf1662352006-01-14 13:21:31 -0800742static __always_inline void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743{
Andreas Mohre518ddb2006-09-29 02:01:22 -0700744 struct fs_struct *fs = current->fs;
745
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 while(1) {
747 struct vfsmount *parent;
Jan Blunck4ac91372008-02-14 19:34:32 -0800748 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749
Andreas Mohre518ddb2006-09-29 02:01:22 -0700750 read_lock(&fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -0800751 if (nd->path.dentry == fs->root.dentry &&
752 nd->path.mnt == fs->root.mnt) {
Andreas Mohre518ddb2006-09-29 02:01:22 -0700753 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 break;
755 }
Andreas Mohre518ddb2006-09-29 02:01:22 -0700756 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 spin_lock(&dcache_lock);
Jan Blunck4ac91372008-02-14 19:34:32 -0800758 if (nd->path.dentry != nd->path.mnt->mnt_root) {
759 nd->path.dentry = dget(nd->path.dentry->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 spin_unlock(&dcache_lock);
761 dput(old);
762 break;
763 }
764 spin_unlock(&dcache_lock);
765 spin_lock(&vfsmount_lock);
Jan Blunck4ac91372008-02-14 19:34:32 -0800766 parent = nd->path.mnt->mnt_parent;
767 if (parent == nd->path.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 spin_unlock(&vfsmount_lock);
769 break;
770 }
771 mntget(parent);
Jan Blunck4ac91372008-02-14 19:34:32 -0800772 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 spin_unlock(&vfsmount_lock);
774 dput(old);
Jan Blunck4ac91372008-02-14 19:34:32 -0800775 mntput(nd->path.mnt);
776 nd->path.mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 }
Jan Blunck4ac91372008-02-14 19:34:32 -0800778 follow_mount(&nd->path.mnt, &nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779}
780
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781/*
782 * It's more convoluted than I'd like it to be, but... it's still fairly
783 * small and for now I'd prefer to have fast path as straight as possible.
784 * It _is_ time-critical.
785 */
786static int do_lookup(struct nameidata *nd, struct qstr *name,
787 struct path *path)
788{
Jan Blunck4ac91372008-02-14 19:34:32 -0800789 struct vfsmount *mnt = nd->path.mnt;
790 struct dentry *dentry = __d_lookup(nd->path.dentry, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791
792 if (!dentry)
793 goto need_lookup;
794 if (dentry->d_op && dentry->d_op->d_revalidate)
795 goto need_revalidate;
796done:
797 path->mnt = mnt;
798 path->dentry = dentry;
Al Viro634ee702005-06-06 13:36:13 -0700799 __follow_mount(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 return 0;
801
802need_lookup:
Jan Blunck4ac91372008-02-14 19:34:32 -0800803 dentry = real_lookup(nd->path.dentry, name, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 if (IS_ERR(dentry))
805 goto fail;
806 goto done;
807
808need_revalidate:
Ian Kentbcdc5e02006-09-27 01:50:44 -0700809 dentry = do_revalidate(dentry, nd);
810 if (!dentry)
811 goto need_lookup;
812 if (IS_ERR(dentry))
813 goto fail;
814 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815
816fail:
817 return PTR_ERR(dentry);
818}
819
820/*
821 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +0100822 * This is the basic name resolution function, turning a pathname into
823 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 *
Prasanna Medaea3834d2005-04-29 16:00:17 +0100825 * Returns 0 and nd will have valid dentry and mnt on success.
826 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800828static int __link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829{
830 struct path next;
831 struct inode *inode;
832 int err;
833 unsigned int lookup_flags = nd->flags;
834
835 while (*name=='/')
836 name++;
837 if (!*name)
838 goto return_reval;
839
Jan Blunck4ac91372008-02-14 19:34:32 -0800840 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 if (nd->depth)
Trond Myklebustf55eab82006-02-04 23:28:01 -0800842 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843
844 /* At this point we know we have a real path component. */
845 for(;;) {
846 unsigned long hash;
847 struct qstr this;
848 unsigned int c;
849
Trond Myklebustcdce5d62005-10-18 14:20:18 -0700850 nd->flags |= LOOKUP_CONTINUE;
Al Viro672b16b2008-07-17 09:45:01 -0400851 err = exec_permission_lite(inode);
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800852 if (err == -EAGAIN)
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200853 err = inode_permission(nd->path.dentry->d_inode,
854 MAY_EXEC);
Mimi Zohar6146f0d2009-02-04 09:06:57 -0500855 if (!err)
856 err = ima_path_check(&nd->path, MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 if (err)
858 break;
859
860 this.name = name;
861 c = *(const unsigned char *)name;
862
863 hash = init_name_hash();
864 do {
865 name++;
866 hash = partial_name_hash(c, hash);
867 c = *(const unsigned char *)name;
868 } while (c && (c != '/'));
869 this.len = name - (const char *) this.name;
870 this.hash = end_name_hash(hash);
871
872 /* remove trailing slashes? */
873 if (!c)
874 goto last_component;
875 while (*++name == '/');
876 if (!*name)
877 goto last_with_slashes;
878
879 /*
880 * "." and ".." are special - ".." especially so because it has
881 * to be able to know about the current root directory and
882 * parent relationships.
883 */
884 if (this.name[0] == '.') switch (this.len) {
885 default:
886 break;
887 case 2:
888 if (this.name[1] != '.')
889 break;
Al Viro58c465e2005-06-06 13:36:13 -0700890 follow_dotdot(nd);
Jan Blunck4ac91372008-02-14 19:34:32 -0800891 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 /* fallthrough */
893 case 1:
894 continue;
895 }
896 /*
897 * See if the low-level filesystem might want
898 * to use its own hash..
899 */
Jan Blunck4ac91372008-02-14 19:34:32 -0800900 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
901 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
902 &this);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 if (err < 0)
904 break;
905 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 /* This does the actual lookups.. */
907 err = do_lookup(nd, &this, &next);
908 if (err)
909 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910
911 err = -ENOENT;
912 inode = next.dentry->d_inode;
913 if (!inode)
914 goto out_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915
916 if (inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700917 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 if (err)
919 goto return_err;
920 err = -ENOENT;
Jan Blunck4ac91372008-02-14 19:34:32 -0800921 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 if (!inode)
923 break;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700924 } else
925 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 err = -ENOTDIR;
927 if (!inode->i_op->lookup)
928 break;
929 continue;
930 /* here ends the main loop */
931
932last_with_slashes:
933 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
934last_component:
Trond Myklebustf55eab82006-02-04 23:28:01 -0800935 /* Clear LOOKUP_CONTINUE iff it was previously unset */
936 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 if (lookup_flags & LOOKUP_PARENT)
938 goto lookup_parent;
939 if (this.name[0] == '.') switch (this.len) {
940 default:
941 break;
942 case 2:
943 if (this.name[1] != '.')
944 break;
Al Viro58c465e2005-06-06 13:36:13 -0700945 follow_dotdot(nd);
Jan Blunck4ac91372008-02-14 19:34:32 -0800946 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 /* fallthrough */
948 case 1:
949 goto return_reval;
950 }
Jan Blunck4ac91372008-02-14 19:34:32 -0800951 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
952 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
953 &this);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 if (err < 0)
955 break;
956 }
957 err = do_lookup(nd, &this, &next);
958 if (err)
959 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 inode = next.dentry->d_inode;
961 if ((lookup_flags & LOOKUP_FOLLOW)
Al Viroacfa4382008-12-04 10:06:33 -0500962 && inode && inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700963 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 if (err)
965 goto return_err;
Jan Blunck4ac91372008-02-14 19:34:32 -0800966 inode = nd->path.dentry->d_inode;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700967 } else
968 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 err = -ENOENT;
970 if (!inode)
971 break;
972 if (lookup_flags & LOOKUP_DIRECTORY) {
973 err = -ENOTDIR;
Al Viroacfa4382008-12-04 10:06:33 -0500974 if (!inode->i_op->lookup)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 break;
976 }
977 goto return_base;
978lookup_parent:
979 nd->last = this;
980 nd->last_type = LAST_NORM;
981 if (this.name[0] != '.')
982 goto return_base;
983 if (this.len == 1)
984 nd->last_type = LAST_DOT;
985 else if (this.len == 2 && this.name[1] == '.')
986 nd->last_type = LAST_DOTDOT;
987 else
988 goto return_base;
989return_reval:
990 /*
991 * We bypassed the ordinary revalidation routines.
992 * We may need to check the cached dentry for staleness.
993 */
Jan Blunck4ac91372008-02-14 19:34:32 -0800994 if (nd->path.dentry && nd->path.dentry->d_sb &&
995 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 err = -ESTALE;
997 /* Note: we do not d_invalidate() */
Jan Blunck4ac91372008-02-14 19:34:32 -0800998 if (!nd->path.dentry->d_op->d_revalidate(
999 nd->path.dentry, nd))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 break;
1001 }
1002return_base:
1003 return 0;
1004out_dput:
Jan Blunck1d957f92008-02-14 19:34:35 -08001005 path_put_conditional(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 break;
1007 }
Jan Blunck1d957f92008-02-14 19:34:35 -08001008 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009return_err:
1010 return err;
1011}
1012
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -08001013static int path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014{
1015 current->total_link_count = 0;
1016 return link_path_walk(name, nd);
1017}
1018
Prasanna Medaea3834d2005-04-29 16:00:17 +01001019/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -08001020static int do_path_lookup(int dfd, const char *name,
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001021 unsigned int flags, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001023 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001024 int fput_needed;
1025 struct file *file;
Andreas Mohre518ddb2006-09-29 02:01:22 -07001026 struct fs_struct *fs = current->fs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027
1028 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1029 nd->flags = flags;
1030 nd->depth = 0;
1031
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 if (*name=='/') {
Andreas Mohre518ddb2006-09-29 02:01:22 -07001033 read_lock(&fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08001034 nd->path = fs->root;
1035 path_get(&fs->root);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001036 read_unlock(&fs->lock);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001037 } else if (dfd == AT_FDCWD) {
Andreas Mohre518ddb2006-09-29 02:01:22 -07001038 read_lock(&fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08001039 nd->path = fs->pwd;
1040 path_get(&fs->pwd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001041 read_unlock(&fs->lock);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001042 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001043 struct dentry *dentry;
1044
1045 file = fget_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001046 retval = -EBADF;
1047 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001048 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001049
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001050 dentry = file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001051
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001052 retval = -ENOTDIR;
1053 if (!S_ISDIR(dentry->d_inode->i_mode))
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001054 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001055
1056 retval = file_permission(file, MAY_EXEC);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001057 if (retval)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001058 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001059
Jan Blunck5dd784d02008-02-14 19:34:38 -08001060 nd->path = file->f_path;
1061 path_get(&file->f_path);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001062
1063 fput_light(file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 }
Josef 'Jeff' Sipek2dfdd262007-05-09 02:33:41 -07001065
1066 retval = path_walk(name, nd);
Jan Blunck4ac91372008-02-14 19:34:32 -08001067 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1068 nd->path.dentry->d_inode))
1069 audit_inode(name, nd->path.dentry);
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001070out_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001071 return retval;
1072
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001073fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001074 fput_light(file, fput_needed);
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001075 goto out_fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076}
1077
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -08001078int path_lookup(const char *name, unsigned int flags,
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001079 struct nameidata *nd)
1080{
1081 return do_path_lookup(AT_FDCWD, name, flags, nd);
1082}
1083
Al Virod1811462008-08-02 00:49:18 -04001084int kern_path(const char *name, unsigned int flags, struct path *path)
1085{
1086 struct nameidata nd;
1087 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1088 if (!res)
1089 *path = nd.path;
1090 return res;
1091}
1092
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001093/**
1094 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1095 * @dentry: pointer to dentry of the base directory
1096 * @mnt: pointer to vfs mount of the base directory
1097 * @name: pointer to file name
1098 * @flags: lookup flags
1099 * @nd: pointer to nameidata
1100 */
1101int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1102 const char *name, unsigned int flags,
1103 struct nameidata *nd)
1104{
1105 int retval;
1106
1107 /* same as do_path_lookup */
1108 nd->last_type = LAST_ROOT;
1109 nd->flags = flags;
1110 nd->depth = 0;
1111
Jan Blunckc8e7f442008-06-09 16:40:35 -07001112 nd->path.dentry = dentry;
1113 nd->path.mnt = mnt;
1114 path_get(&nd->path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001115
1116 retval = path_walk(name, nd);
Jan Blunck4ac91372008-02-14 19:34:32 -08001117 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1118 nd->path.dentry->d_inode))
1119 audit_inode(name, nd->path.dentry);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001120
1121 return retval;
1122
1123}
1124
Trond Myklebust834f2a42005-10-18 14:20:16 -07001125/**
1126 * path_lookup_open - lookup a file path with open intent
Martin Waitz7045f372006-02-01 03:06:57 -08001127 * @dfd: the directory to use as base, or AT_FDCWD
Trond Myklebust834f2a42005-10-18 14:20:16 -07001128 * @name: pointer to file name
1129 * @lookup_flags: lookup intent flags
1130 * @nd: pointer to nameidata
1131 * @open_flags: open intent flags
1132 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001133int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
Trond Myklebust834f2a42005-10-18 14:20:16 -07001134 struct nameidata *nd, int open_flags)
1135{
Al Viro8737f3a2008-08-02 22:36:57 -04001136 struct file *filp = get_empty_filp();
1137 int err;
Trond Myklebust834f2a42005-10-18 14:20:16 -07001138
Al Viro8737f3a2008-08-02 22:36:57 -04001139 if (filp == NULL)
1140 return -ENFILE;
1141 nd->intent.open.file = filp;
1142 nd->intent.open.flags = open_flags;
1143 nd->intent.open.create_mode = 0;
1144 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1145 if (IS_ERR(nd->intent.open.file)) {
1146 if (err == 0) {
1147 err = PTR_ERR(nd->intent.open.file);
1148 path_put(&nd->path);
1149 }
1150 } else if (err != 0)
1151 release_open_intent(nd);
1152 return err;
Trond Myklebust834f2a42005-10-18 14:20:16 -07001153}
1154
Christoph Hellwigeead1912007-10-16 23:25:38 -07001155static struct dentry *__lookup_hash(struct qstr *name,
1156 struct dentry *base, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157{
James Morris057f6c02007-04-26 00:12:05 -07001158 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 struct inode *inode;
1160 int err;
1161
1162 inode = base->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163
1164 /*
1165 * See if the low-level filesystem might want
1166 * to use its own hash..
1167 */
1168 if (base->d_op && base->d_op->d_hash) {
1169 err = base->d_op->d_hash(base, name);
1170 dentry = ERR_PTR(err);
1171 if (err < 0)
1172 goto out;
1173 }
1174
1175 dentry = cached_lookup(base, name, nd);
1176 if (!dentry) {
Miklos Szeredid70b67c2008-07-02 21:30:15 +02001177 struct dentry *new;
1178
1179 /* Don't create child dentry for a dead directory. */
1180 dentry = ERR_PTR(-ENOENT);
1181 if (IS_DEADDIR(inode))
1182 goto out;
1183
1184 new = d_alloc(base, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 dentry = ERR_PTR(-ENOMEM);
1186 if (!new)
1187 goto out;
1188 dentry = inode->i_op->lookup(inode, new, nd);
1189 if (!dentry)
1190 dentry = new;
1191 else
1192 dput(new);
1193 }
1194out:
1195 return dentry;
1196}
1197
James Morris057f6c02007-04-26 00:12:05 -07001198/*
1199 * Restricted form of lookup. Doesn't follow links, single-component only,
1200 * needs parent already locked. Doesn't follow mounts.
1201 * SMP-safe.
1202 */
Adrian Bunka244e162006-03-31 02:32:11 -08001203static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204{
Christoph Hellwigeead1912007-10-16 23:25:38 -07001205 int err;
1206
Al Virof419a2e2008-07-22 00:07:17 -04001207 err = inode_permission(nd->path.dentry->d_inode, MAY_EXEC);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001208 if (err)
1209 return ERR_PTR(err);
Jan Blunck4ac91372008-02-14 19:34:32 -08001210 return __lookup_hash(&nd->last, nd->path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211}
1212
Christoph Hellwigeead1912007-10-16 23:25:38 -07001213static int __lookup_one_len(const char *name, struct qstr *this,
1214 struct dentry *base, int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215{
1216 unsigned long hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 unsigned int c;
1218
James Morris057f6c02007-04-26 00:12:05 -07001219 this->name = name;
1220 this->len = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 if (!len)
James Morris057f6c02007-04-26 00:12:05 -07001222 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223
1224 hash = init_name_hash();
1225 while (len--) {
1226 c = *(const unsigned char *)name++;
1227 if (c == '/' || c == '\0')
James Morris057f6c02007-04-26 00:12:05 -07001228 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 hash = partial_name_hash(c, hash);
1230 }
James Morris057f6c02007-04-26 00:12:05 -07001231 this->hash = end_name_hash(hash);
1232 return 0;
1233}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234
Christoph Hellwigeead1912007-10-16 23:25:38 -07001235/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001236 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07001237 * @name: pathname component to lookup
1238 * @base: base directory to lookup from
1239 * @len: maximum length @len should be interpreted to
1240 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07001241 * Note that this routine is purely a helper for filesystem usage and should
1242 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07001243 * nameidata argument is passed to the filesystem methods and a filesystem
1244 * using this helper needs to be prepared for that.
1245 */
James Morris057f6c02007-04-26 00:12:05 -07001246struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1247{
1248 int err;
1249 struct qstr this;
1250
1251 err = __lookup_one_len(name, &this, base, len);
1252 if (err)
1253 return ERR_PTR(err);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001254
Al Virof419a2e2008-07-22 00:07:17 -04001255 err = inode_permission(base->d_inode, MAY_EXEC);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001256 if (err)
1257 return ERR_PTR(err);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001258 return __lookup_hash(&this, base, NULL);
James Morris057f6c02007-04-26 00:12:05 -07001259}
1260
Christoph Hellwigeead1912007-10-16 23:25:38 -07001261/**
1262 * lookup_one_noperm - bad hack for sysfs
1263 * @name: pathname component to lookup
1264 * @base: base directory to lookup from
1265 *
1266 * This is a variant of lookup_one_len that doesn't perform any permission
1267 * checks. It's a horrible hack to work around the braindead sysfs
1268 * architecture and should not be used anywhere else.
1269 *
1270 * DON'T USE THIS FUNCTION EVER, thanks.
1271 */
1272struct dentry *lookup_one_noperm(const char *name, struct dentry *base)
James Morris057f6c02007-04-26 00:12:05 -07001273{
1274 int err;
1275 struct qstr this;
1276
Christoph Hellwigeead1912007-10-16 23:25:38 -07001277 err = __lookup_one_len(name, &this, base, strlen(name));
James Morris057f6c02007-04-26 00:12:05 -07001278 if (err)
1279 return ERR_PTR(err);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001280 return __lookup_hash(&this, base, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281}
1282
Al Viro2d8f3032008-07-22 09:59:21 -04001283int user_path_at(int dfd, const char __user *name, unsigned flags,
1284 struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285{
Al Viro2d8f3032008-07-22 09:59:21 -04001286 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 char *tmp = getname(name);
1288 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04001290
1291 BUG_ON(flags & LOOKUP_PARENT);
1292
1293 err = do_path_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04001295 if (!err)
1296 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 }
1298 return err;
1299}
1300
Al Viro2ad94ae2008-07-21 09:32:51 -04001301static int user_path_parent(int dfd, const char __user *path,
1302 struct nameidata *nd, char **name)
1303{
1304 char *s = getname(path);
1305 int error;
1306
1307 if (IS_ERR(s))
1308 return PTR_ERR(s);
1309
1310 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1311 if (error)
1312 putname(s);
1313 else
1314 *name = s;
1315
1316 return error;
1317}
1318
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319/*
1320 * It's inline, so penalty for filesystems that don't use sticky bit is
1321 * minimal.
1322 */
1323static inline int check_sticky(struct inode *dir, struct inode *inode)
1324{
David Howellsda9592e2008-11-14 10:39:05 +11001325 uid_t fsuid = current_fsuid();
1326
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 if (!(dir->i_mode & S_ISVTX))
1328 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001329 if (inode->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001331 if (dir->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 return 0;
1333 return !capable(CAP_FOWNER);
1334}
1335
1336/*
1337 * Check whether we can remove a link victim from directory dir, check
1338 * whether the type of victim is right.
1339 * 1. We can't do it if dir is read-only (done in permission())
1340 * 2. We should have write and exec permissions on dir
1341 * 3. We can't remove anything from append-only dir
1342 * 4. We can't do anything with immutable dir (done in permission())
1343 * 5. If the sticky bit on dir is set we should either
1344 * a. be owner of dir, or
1345 * b. be owner of victim, or
1346 * c. have CAP_FOWNER capability
1347 * 6. If the victim is append-only or immutable we can't do antyhing with
1348 * links pointing to it.
1349 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1350 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1351 * 9. We can't remove a root or mountpoint.
1352 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1353 * nfs_async_unlink().
1354 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001355static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356{
1357 int error;
1358
1359 if (!victim->d_inode)
1360 return -ENOENT;
1361
1362 BUG_ON(victim->d_parent->d_inode != dir);
Al Viro5a190ae2007-06-07 12:19:32 -04001363 audit_inode_child(victim->d_name.name, victim, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364
Al Virof419a2e2008-07-22 00:07:17 -04001365 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 if (error)
1367 return error;
1368 if (IS_APPEND(dir))
1369 return -EPERM;
1370 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
Hugh Dickinsf9454542008-11-19 15:36:38 -08001371 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 return -EPERM;
1373 if (isdir) {
1374 if (!S_ISDIR(victim->d_inode->i_mode))
1375 return -ENOTDIR;
1376 if (IS_ROOT(victim))
1377 return -EBUSY;
1378 } else if (S_ISDIR(victim->d_inode->i_mode))
1379 return -EISDIR;
1380 if (IS_DEADDIR(dir))
1381 return -ENOENT;
1382 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1383 return -EBUSY;
1384 return 0;
1385}
1386
1387/* Check whether we can create an object with dentry child in directory
1388 * dir.
1389 * 1. We can't do it if child already exists (open has special treatment for
1390 * this case, but since we are inlined it's OK)
1391 * 2. We can't do it if dir is read-only (done in permission())
1392 * 3. We should have write and exec permissions on dir
1393 * 4. We can't do it if dir is immutable (done in permission())
1394 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02001395static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396{
1397 if (child->d_inode)
1398 return -EEXIST;
1399 if (IS_DEADDIR(dir))
1400 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04001401 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402}
1403
1404/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 * O_DIRECTORY translates into forcing a directory lookup.
1406 */
1407static inline int lookup_flags(unsigned int f)
1408{
1409 unsigned long retval = LOOKUP_FOLLOW;
1410
1411 if (f & O_NOFOLLOW)
1412 retval &= ~LOOKUP_FOLLOW;
1413
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 if (f & O_DIRECTORY)
1415 retval |= LOOKUP_DIRECTORY;
1416
1417 return retval;
1418}
1419
1420/*
1421 * p1 and p2 should be directories on the same fs.
1422 */
1423struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1424{
1425 struct dentry *p;
1426
1427 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001428 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 return NULL;
1430 }
1431
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001432 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001434 p = d_ancestor(p2, p1);
1435 if (p) {
1436 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1437 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1438 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 }
1440
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001441 p = d_ancestor(p1, p2);
1442 if (p) {
1443 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1444 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1445 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 }
1447
Ingo Molnarf2eace22006-07-03 00:25:05 -07001448 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1449 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 return NULL;
1451}
1452
1453void unlock_rename(struct dentry *p1, struct dentry *p2)
1454{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001455 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001457 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001458 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 }
1460}
1461
1462int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1463 struct nameidata *nd)
1464{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001465 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466
1467 if (error)
1468 return error;
1469
Al Viroacfa4382008-12-04 10:06:33 -05001470 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 return -EACCES; /* shouldn't it be ENOSYS? */
1472 mode &= S_IALLUGO;
1473 mode |= S_IFREG;
1474 error = security_inode_create(dir, dentry, mode);
1475 if (error)
1476 return error;
Jan Kara9e3509e2009-01-26 16:45:12 +01001477 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001479 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001480 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 return error;
1482}
1483
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001484int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001486 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 struct inode *inode = dentry->d_inode;
1488 int error;
1489
1490 if (!inode)
1491 return -ENOENT;
1492
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001493 switch (inode->i_mode & S_IFMT) {
1494 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001496 case S_IFDIR:
1497 if (acc_mode & MAY_WRITE)
1498 return -EISDIR;
1499 break;
1500 case S_IFBLK:
1501 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001502 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001504 /*FALLTHRU*/
1505 case S_IFIFO:
1506 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001508 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001509 }
Dave Hansenb41572e2007-10-16 23:31:14 -07001510
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001511 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07001512 if (error)
1513 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05001514
James Morriscb5629b2009-02-06 11:01:45 +11001515 error = ima_path_check(path,
Mimi Zohar6146f0d2009-02-04 09:06:57 -05001516 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC));
1517 if (error)
1518 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 /*
1520 * An append-only file must be opened in append mode for writing.
1521 */
1522 if (IS_APPEND(inode)) {
1523 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1524 return -EPERM;
1525 if (flag & O_TRUNC)
1526 return -EPERM;
1527 }
1528
1529 /* O_NOATIME can only be set by the owner or superuser */
1530 if (flag & O_NOATIME)
Satyam Sharma3bd858a2007-07-17 15:00:08 +05301531 if (!is_owner_or_cap(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 return -EPERM;
1533
1534 /*
1535 * Ensure there are no outstanding leases on the file.
1536 */
1537 error = break_lease(inode, flag);
1538 if (error)
1539 return error;
1540
1541 if (flag & O_TRUNC) {
1542 error = get_write_access(inode);
1543 if (error)
1544 return error;
1545
1546 /*
1547 * Refuse to truncate files with mandatory locks held on them.
1548 */
1549 error = locks_verify_locked(inode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001550 if (!error)
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001551 error = security_path_truncate(path, 0,
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001552 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 if (!error) {
Jan Kara9e3509e2009-01-26 16:45:12 +01001554 vfs_dq_init(inode);
Miklos Szeredid139d7f2007-10-18 03:07:00 -07001555
1556 error = do_truncate(dentry, 0,
1557 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1558 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 }
1560 put_write_access(inode);
1561 if (error)
1562 return error;
1563 } else
1564 if (flag & FMODE_WRITE)
Jan Kara9e3509e2009-01-26 16:45:12 +01001565 vfs_dq_init(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566
1567 return 0;
1568}
1569
Dave Hansend57999e2008-02-15 14:37:27 -08001570/*
1571 * Be careful about ever adding any more callers of this
1572 * function. Its flags must be in the namei format, not
1573 * what get passed to sys_open().
1574 */
1575static int __open_namei_create(struct nameidata *nd, struct path *path,
Dave Hansenaab520e2006-09-30 23:29:02 -07001576 int flag, int mode)
1577{
1578 int error;
Jan Blunck4ac91372008-02-14 19:34:32 -08001579 struct dentry *dir = nd->path.dentry;
Dave Hansenaab520e2006-09-30 23:29:02 -07001580
1581 if (!IS_POSIXACL(dir->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04001582 mode &= ~current_umask();
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001583 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1584 if (error)
1585 goto out_unlock;
Dave Hansenaab520e2006-09-30 23:29:02 -07001586 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001587out_unlock:
Dave Hansenaab520e2006-09-30 23:29:02 -07001588 mutex_unlock(&dir->d_inode->i_mutex);
Jan Blunck4ac91372008-02-14 19:34:32 -08001589 dput(nd->path.dentry);
1590 nd->path.dentry = path->dentry;
Dave Hansenaab520e2006-09-30 23:29:02 -07001591 if (error)
1592 return error;
1593 /* Don't check for write permission, don't truncate */
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001594 return may_open(&nd->path, 0, flag & ~O_TRUNC);
Dave Hansenaab520e2006-09-30 23:29:02 -07001595}
1596
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597/*
Dave Hansend57999e2008-02-15 14:37:27 -08001598 * Note that while the flag value (low two bits) for sys_open means:
1599 * 00 - read-only
1600 * 01 - write-only
1601 * 10 - read-write
1602 * 11 - special
1603 * it is changed into
1604 * 00 - no permissions needed
1605 * 01 - read-permission
1606 * 10 - write-permission
1607 * 11 - read-write
1608 * for the internal routines (ie open_namei()/follow_link() etc)
1609 * This is more logical, and also allows the 00 "no perm needed"
1610 * to be used for symlinks (where the permissions are checked
1611 * later).
1612 *
1613*/
1614static inline int open_to_namei_flags(int flag)
1615{
1616 if ((flag+1) & O_ACCMODE)
1617 flag++;
1618 return flag;
1619}
1620
Dave Hansen4a3fd212008-02-15 14:37:48 -08001621static int open_will_write_to_fs(int flag, struct inode *inode)
1622{
1623 /*
1624 * We'll never write to the fs underlying
1625 * a device file.
1626 */
1627 if (special_file(inode->i_mode))
1628 return 0;
1629 return (flag & O_TRUNC);
1630}
1631
Dave Hansend57999e2008-02-15 14:37:27 -08001632/*
Dave Hansen4a3fd212008-02-15 14:37:48 -08001633 * Note that the low bits of the passed in "open_flag"
1634 * are not the same as in the local variable "flag". See
1635 * open_to_namei_flags() for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001637struct file *do_filp_open(int dfd, const char *pathname,
1638 int open_flag, int mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639{
Dave Hansen4a3fd212008-02-15 14:37:48 -08001640 struct file *filp;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001641 struct nameidata nd;
Trond Myklebust834f2a42005-10-18 14:20:16 -07001642 int acc_mode, error;
Al Viro4e7506e2005-06-06 13:36:00 -07001643 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 struct dentry *dir;
1645 int count = 0;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001646 int will_write;
Dave Hansend57999e2008-02-15 14:37:27 -08001647 int flag = open_to_namei_flags(open_flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648
Al Virob77b0642008-07-17 09:37:02 -04001649 acc_mode = MAY_OPEN | ACC_MODE(flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650
Trond Myklebust834f2a42005-10-18 14:20:16 -07001651 /* O_TRUNC implies we need access checks for write permissions */
1652 if (flag & O_TRUNC)
1653 acc_mode |= MAY_WRITE;
1654
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 /* Allow the LSM permission hook to distinguish append
1656 access from general write access. */
1657 if (flag & O_APPEND)
1658 acc_mode |= MAY_APPEND;
1659
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 /*
1661 * The simplest case - just a plain lookup.
1662 */
1663 if (!(flag & O_CREAT)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001664 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001665 &nd, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 if (error)
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001667 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 goto ok;
1669 }
1670
1671 /*
1672 * Create - we need to know the parent.
1673 */
Al Viro8737f3a2008-08-02 22:36:57 -04001674 error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 if (error)
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001676 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677
1678 /*
1679 * We have the parent and last component. First of all, check
1680 * that we are not asked to creat(2) an obvious directory - that
1681 * will not do.
1682 */
1683 error = -EISDIR;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001684 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
Al Viro8737f3a2008-08-02 22:36:57 -04001685 goto exit_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686
Al Viro8737f3a2008-08-02 22:36:57 -04001687 error = -ENFILE;
1688 filp = get_empty_filp();
1689 if (filp == NULL)
1690 goto exit_parent;
1691 nd.intent.open.file = filp;
1692 nd.intent.open.flags = flag;
1693 nd.intent.open.create_mode = mode;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001694 dir = nd.path.dentry;
1695 nd.flags &= ~LOOKUP_PARENT;
Al Viro8737f3a2008-08-02 22:36:57 -04001696 nd.flags |= LOOKUP_CREATE | LOOKUP_OPEN;
Al Viro35165862008-08-05 03:00:49 -04001697 if (flag & O_EXCL)
1698 nd.flags |= LOOKUP_EXCL;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001699 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001700 path.dentry = lookup_hash(&nd);
1701 path.mnt = nd.path.mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702
1703do_last:
Al Viro4e7506e2005-06-06 13:36:00 -07001704 error = PTR_ERR(path.dentry);
1705 if (IS_ERR(path.dentry)) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001706 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 goto exit;
1708 }
1709
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001710 if (IS_ERR(nd.intent.open.file)) {
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001711 error = PTR_ERR(nd.intent.open.file);
Dave Hansen4a3fd212008-02-15 14:37:48 -08001712 goto exit_mutex_unlock;
Oleg Drokin4af4c522006-03-25 03:06:54 -08001713 }
1714
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 /* Negative dentry, just create the file */
Al Viro4e7506e2005-06-06 13:36:00 -07001716 if (!path.dentry->d_inode) {
Dave Hansen4a3fd212008-02-15 14:37:48 -08001717 /*
1718 * This write is needed to ensure that a
1719 * ro->rw transition does not occur between
1720 * the time when the file is created and when
1721 * a permanent write count is taken through
1722 * the 'struct file' in nameidata_to_filp().
1723 */
1724 error = mnt_want_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 if (error)
Dave Hansen4a3fd212008-02-15 14:37:48 -08001726 goto exit_mutex_unlock;
1727 error = __open_namei_create(&nd, &path, flag, mode);
1728 if (error) {
1729 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 goto exit;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001731 }
1732 filp = nameidata_to_filp(&nd, open_flag);
1733 mnt_drop_write(nd.path.mnt);
1734 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 }
1736
1737 /*
1738 * It already exists.
1739 */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001740 mutex_unlock(&dir->d_inode->i_mutex);
Al Viro5a190ae2007-06-07 12:19:32 -04001741 audit_inode(pathname, path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742
1743 error = -EEXIST;
1744 if (flag & O_EXCL)
1745 goto exit_dput;
1746
Al Viroe13b2102005-06-06 13:36:06 -07001747 if (__follow_mount(&path)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 error = -ELOOP;
Al Viroba7a4c12005-06-06 13:36:08 -07001749 if (flag & O_NOFOLLOW)
1750 goto exit_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 }
Amy Griffis3e2efce2006-07-13 13:16:02 -04001752
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 error = -ENOENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001754 if (!path.dentry->d_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 goto exit_dput;
Al Viroacfa4382008-12-04 10:06:33 -05001756 if (path.dentry->d_inode->i_op->follow_link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 goto do_link;
1758
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001759 path_to_nameidata(&path, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 error = -EISDIR;
Al Viro4e7506e2005-06-06 13:36:00 -07001761 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 goto exit;
1763ok:
Dave Hansen4a3fd212008-02-15 14:37:48 -08001764 /*
1765 * Consider:
1766 * 1. may_open() truncates a file
1767 * 2. a rw->ro mount transition occurs
1768 * 3. nameidata_to_filp() fails due to
1769 * the ro mount.
1770 * That would be inconsistent, and should
1771 * be avoided. Taking this mnt write here
1772 * ensures that (2) can not occur.
1773 */
1774 will_write = open_will_write_to_fs(flag, nd.path.dentry->d_inode);
1775 if (will_write) {
1776 error = mnt_want_write(nd.path.mnt);
1777 if (error)
1778 goto exit;
1779 }
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001780 error = may_open(&nd.path, acc_mode, flag);
Dave Hansen4a3fd212008-02-15 14:37:48 -08001781 if (error) {
1782 if (will_write)
1783 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 goto exit;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001785 }
1786 filp = nameidata_to_filp(&nd, open_flag);
1787 /*
1788 * It is now safe to drop the mnt write
1789 * because the filp has had a write taken
1790 * on its behalf.
1791 */
1792 if (will_write)
1793 mnt_drop_write(nd.path.mnt);
1794 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795
Dave Hansen4a3fd212008-02-15 14:37:48 -08001796exit_mutex_unlock:
1797 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798exit_dput:
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001799 path_put_conditional(&path, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800exit:
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001801 if (!IS_ERR(nd.intent.open.file))
1802 release_open_intent(&nd);
Al Viro8737f3a2008-08-02 22:36:57 -04001803exit_parent:
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001804 path_put(&nd.path);
1805 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806
1807do_link:
1808 error = -ELOOP;
1809 if (flag & O_NOFOLLOW)
1810 goto exit_dput;
1811 /*
1812 * This is subtle. Instead of calling do_follow_link() we do the
1813 * thing by hands. The reason is that this way we have zero link_count
1814 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1815 * After that we have the parent and last component, i.e.
1816 * we are in the same situation as after the first path_walk().
1817 * Well, almost - if the last component is normal we get its copy
1818 * stored in nd->last.name and we will have to putname() it when we
1819 * are done. Procfs-like symlinks just set LAST_BIND.
1820 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001821 nd.flags |= LOOKUP_PARENT;
1822 error = security_inode_follow_link(path.dentry, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823 if (error)
1824 goto exit_dput;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001825 error = __do_follow_link(&path, &nd);
Kirill Korotaevde459212006-07-14 00:23:49 -07001826 if (error) {
1827 /* Does someone understand code flow here? Or it is only
1828 * me so stupid? Anathema to whoever designed this non-sense
1829 * with "intent.open".
1830 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001831 release_open_intent(&nd);
1832 return ERR_PTR(error);
Kirill Korotaevde459212006-07-14 00:23:49 -07001833 }
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001834 nd.flags &= ~LOOKUP_PARENT;
1835 if (nd.last_type == LAST_BIND)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 goto ok;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 error = -EISDIR;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001838 if (nd.last_type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 goto exit;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001840 if (nd.last.name[nd.last.len]) {
1841 __putname(nd.last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842 goto exit;
1843 }
1844 error = -ELOOP;
1845 if (count++==32) {
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001846 __putname(nd.last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 goto exit;
1848 }
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001849 dir = nd.path.dentry;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001850 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001851 path.dentry = lookup_hash(&nd);
1852 path.mnt = nd.path.mnt;
1853 __putname(nd.last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 goto do_last;
1855}
1856
1857/**
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001858 * filp_open - open file and return file pointer
1859 *
1860 * @filename: path to open
1861 * @flags: open flags as per the open(2) second argument
1862 * @mode: mode for the new file if O_CREAT is set, else ignored
1863 *
1864 * This is the helper to open a file from kernelspace if you really
1865 * have to. But in generally you should not do this, so please move
1866 * along, nothing to see here..
1867 */
1868struct file *filp_open(const char *filename, int flags, int mode)
1869{
1870 return do_filp_open(AT_FDCWD, filename, flags, mode);
1871}
1872EXPORT_SYMBOL(filp_open);
1873
1874/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875 * lookup_create - lookup a dentry, creating it if it doesn't exist
1876 * @nd: nameidata info
1877 * @is_dir: directory flag
1878 *
1879 * Simple function to lookup and return a dentry and create it
1880 * if it doesn't exist. Is SMP-safe.
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001881 *
Jan Blunck4ac91372008-02-14 19:34:32 -08001882 * Returns with nd->path.dentry->d_inode->i_mutex locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883 */
1884struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1885{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001886 struct dentry *dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887
Jan Blunck4ac91372008-02-14 19:34:32 -08001888 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001889 /*
1890 * Yucky last component or no last component at all?
1891 * (foo/., foo/.., /////)
1892 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893 if (nd->last_type != LAST_NORM)
1894 goto fail;
1895 nd->flags &= ~LOOKUP_PARENT;
Al Viro35165862008-08-05 03:00:49 -04001896 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
ASANO Masahiroa6349042006-08-22 20:06:02 -04001897 nd->intent.open.flags = O_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001898
1899 /*
1900 * Do the final lookup.
1901 */
Christoph Hellwig49705b72005-11-08 21:35:06 -08001902 dentry = lookup_hash(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 if (IS_ERR(dentry))
1904 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001905
Al Viroe9baf6e2008-05-15 04:49:12 -04001906 if (dentry->d_inode)
1907 goto eexist;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001908 /*
1909 * Special case - lookup gave negative, but... we had foo/bar/
1910 * From the vfs_mknod() POV we just have a negative dentry -
1911 * all is fine. Let's be bastards - you had / on the end, you've
1912 * been asking for (non-existent) directory. -ENOENT for you.
1913 */
Al Viroe9baf6e2008-05-15 04:49:12 -04001914 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1915 dput(dentry);
1916 dentry = ERR_PTR(-ENOENT);
1917 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918 return dentry;
Al Viroe9baf6e2008-05-15 04:49:12 -04001919eexist:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920 dput(dentry);
Al Viroe9baf6e2008-05-15 04:49:12 -04001921 dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922fail:
1923 return dentry;
1924}
Christoph Hellwigf81a0bf2005-05-19 12:26:43 -07001925EXPORT_SYMBOL_GPL(lookup_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926
1927int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1928{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001929 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930
1931 if (error)
1932 return error;
1933
1934 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1935 return -EPERM;
1936
Al Viroacfa4382008-12-04 10:06:33 -05001937 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 return -EPERM;
1939
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07001940 error = devcgroup_inode_mknod(mode, dev);
1941 if (error)
1942 return error;
1943
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 error = security_inode_mknod(dir, dentry, mode, dev);
1945 if (error)
1946 return error;
1947
Jan Kara9e3509e2009-01-26 16:45:12 +01001948 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001950 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001951 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 return error;
1953}
1954
Dave Hansen463c3192008-02-15 14:37:57 -08001955static int may_mknod(mode_t mode)
1956{
1957 switch (mode & S_IFMT) {
1958 case S_IFREG:
1959 case S_IFCHR:
1960 case S_IFBLK:
1961 case S_IFIFO:
1962 case S_IFSOCK:
1963 case 0: /* zero mode translates to S_IFREG */
1964 return 0;
1965 case S_IFDIR:
1966 return -EPERM;
1967 default:
1968 return -EINVAL;
1969 }
1970}
1971
Heiko Carstens2e4d0922009-01-14 14:14:31 +01001972SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
1973 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974{
Al Viro2ad94ae2008-07-21 09:32:51 -04001975 int error;
1976 char *tmp;
1977 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978 struct nameidata nd;
1979
1980 if (S_ISDIR(mode))
1981 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982
Al Viro2ad94ae2008-07-21 09:32:51 -04001983 error = user_path_parent(dfd, filename, &nd, &tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04001985 return error;
1986
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 dentry = lookup_create(&nd, 0);
Dave Hansen463c3192008-02-15 14:37:57 -08001988 if (IS_ERR(dentry)) {
1989 error = PTR_ERR(dentry);
1990 goto out_unlock;
1991 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001992 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04001993 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08001994 error = may_mknod(mode);
1995 if (error)
1996 goto out_dput;
1997 error = mnt_want_write(nd.path.mnt);
1998 if (error)
1999 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002000 error = security_path_mknod(&nd.path, dentry, mode, dev);
2001 if (error)
2002 goto out_drop_write;
Dave Hansen463c3192008-02-15 14:37:57 -08002003 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 case 0: case S_IFREG:
Jan Blunck4ac91372008-02-14 19:34:32 -08002005 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006 break;
2007 case S_IFCHR: case S_IFBLK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002008 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 new_decode_dev(dev));
2010 break;
2011 case S_IFIFO: case S_IFSOCK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002012 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002015out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002016 mnt_drop_write(nd.path.mnt);
2017out_dput:
2018 dput(dentry);
2019out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002020 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002021 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 putname(tmp);
2023
2024 return error;
2025}
2026
Heiko Carstens3480b252009-01-14 14:14:16 +01002027SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002028{
2029 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2030}
2031
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2033{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002034 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035
2036 if (error)
2037 return error;
2038
Al Viroacfa4382008-12-04 10:06:33 -05002039 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 return -EPERM;
2041
2042 mode &= (S_IRWXUGO|S_ISVTX);
2043 error = security_inode_mkdir(dir, dentry, mode);
2044 if (error)
2045 return error;
2046
Jan Kara9e3509e2009-01-26 16:45:12 +01002047 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002049 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002050 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 return error;
2052}
2053
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002054SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055{
2056 int error = 0;
2057 char * tmp;
Dave Hansen6902d922006-09-30 23:29:01 -07002058 struct dentry *dentry;
2059 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060
Al Viro2ad94ae2008-07-21 09:32:51 -04002061 error = user_path_parent(dfd, pathname, &nd, &tmp);
2062 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002063 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064
Dave Hansen6902d922006-09-30 23:29:01 -07002065 dentry = lookup_create(&nd, 1);
2066 error = PTR_ERR(dentry);
2067 if (IS_ERR(dentry))
2068 goto out_unlock;
2069
Jan Blunck4ac91372008-02-14 19:34:32 -08002070 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002071 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002072 error = mnt_want_write(nd.path.mnt);
2073 if (error)
2074 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002075 error = security_path_mkdir(&nd.path, dentry, mode);
2076 if (error)
2077 goto out_drop_write;
Jan Blunck4ac91372008-02-14 19:34:32 -08002078 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002079out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002080 mnt_drop_write(nd.path.mnt);
2081out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002082 dput(dentry);
2083out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002084 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002085 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002086 putname(tmp);
2087out_err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088 return error;
2089}
2090
Heiko Carstens3cdad422009-01-14 14:14:22 +01002091SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002092{
2093 return sys_mkdirat(AT_FDCWD, pathname, mode);
2094}
2095
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096/*
2097 * We try to drop the dentry early: we should have
2098 * a usage count of 2 if we're the only user of this
2099 * dentry, and if that is true (possibly after pruning
2100 * the dcache), then we drop the dentry now.
2101 *
2102 * A low-level filesystem can, if it choses, legally
2103 * do a
2104 *
2105 * if (!d_unhashed(dentry))
2106 * return -EBUSY;
2107 *
2108 * if it cannot handle the case of removing a directory
2109 * that is still in use by something else..
2110 */
2111void dentry_unhash(struct dentry *dentry)
2112{
2113 dget(dentry);
Vasily Averindc168422006-12-06 20:37:07 -08002114 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 spin_lock(&dcache_lock);
2116 spin_lock(&dentry->d_lock);
2117 if (atomic_read(&dentry->d_count) == 2)
2118 __d_drop(dentry);
2119 spin_unlock(&dentry->d_lock);
2120 spin_unlock(&dcache_lock);
2121}
2122
2123int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2124{
2125 int error = may_delete(dir, dentry, 1);
2126
2127 if (error)
2128 return error;
2129
Al Viroacfa4382008-12-04 10:06:33 -05002130 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131 return -EPERM;
2132
Jan Kara9e3509e2009-01-26 16:45:12 +01002133 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002135 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 dentry_unhash(dentry);
2137 if (d_mountpoint(dentry))
2138 error = -EBUSY;
2139 else {
2140 error = security_inode_rmdir(dir, dentry);
2141 if (!error) {
2142 error = dir->i_op->rmdir(dir, dentry);
2143 if (!error)
2144 dentry->d_inode->i_flags |= S_DEAD;
2145 }
2146 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002147 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 d_delete(dentry);
2150 }
2151 dput(dentry);
2152
2153 return error;
2154}
2155
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002156static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157{
2158 int error = 0;
2159 char * name;
2160 struct dentry *dentry;
2161 struct nameidata nd;
2162
Al Viro2ad94ae2008-07-21 09:32:51 -04002163 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002165 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166
2167 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002168 case LAST_DOTDOT:
2169 error = -ENOTEMPTY;
2170 goto exit1;
2171 case LAST_DOT:
2172 error = -EINVAL;
2173 goto exit1;
2174 case LAST_ROOT:
2175 error = -EBUSY;
2176 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002178
2179 nd.flags &= ~LOOKUP_PARENT;
2180
Jan Blunck4ac91372008-02-14 19:34:32 -08002181 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002182 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002184 if (IS_ERR(dentry))
2185 goto exit2;
Dave Hansen06227532008-02-15 14:37:34 -08002186 error = mnt_want_write(nd.path.mnt);
2187 if (error)
2188 goto exit3;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002189 error = security_path_rmdir(&nd.path, dentry);
2190 if (error)
2191 goto exit4;
Jan Blunck4ac91372008-02-14 19:34:32 -08002192 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002193exit4:
Dave Hansen06227532008-02-15 14:37:34 -08002194 mnt_drop_write(nd.path.mnt);
2195exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07002196 dput(dentry);
2197exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08002198 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002200 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201 putname(name);
2202 return error;
2203}
2204
Heiko Carstens3cdad422009-01-14 14:14:22 +01002205SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002206{
2207 return do_rmdir(AT_FDCWD, pathname);
2208}
2209
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210int vfs_unlink(struct inode *dir, struct dentry *dentry)
2211{
2212 int error = may_delete(dir, dentry, 0);
2213
2214 if (error)
2215 return error;
2216
Al Viroacfa4382008-12-04 10:06:33 -05002217 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 return -EPERM;
2219
Jan Kara9e3509e2009-01-26 16:45:12 +01002220 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002222 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223 if (d_mountpoint(dentry))
2224 error = -EBUSY;
2225 else {
2226 error = security_inode_unlink(dir, dentry);
2227 if (!error)
2228 error = dir->i_op->unlink(dir, dentry);
2229 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002230 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231
2232 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2233 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08002234 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236 }
Robert Love0eeca282005-07-12 17:06:03 -04002237
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238 return error;
2239}
2240
2241/*
2242 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002243 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244 * writeout happening, and we don't want to prevent access to the directory
2245 * while waiting on the I/O.
2246 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002247static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248{
Al Viro2ad94ae2008-07-21 09:32:51 -04002249 int error;
2250 char *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251 struct dentry *dentry;
2252 struct nameidata nd;
2253 struct inode *inode = NULL;
2254
Al Viro2ad94ae2008-07-21 09:32:51 -04002255 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002257 return error;
2258
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259 error = -EISDIR;
2260 if (nd.last_type != LAST_NORM)
2261 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002262
2263 nd.flags &= ~LOOKUP_PARENT;
2264
Jan Blunck4ac91372008-02-14 19:34:32 -08002265 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002266 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267 error = PTR_ERR(dentry);
2268 if (!IS_ERR(dentry)) {
2269 /* Why not before? Because we want correct error value */
2270 if (nd.last.name[nd.last.len])
2271 goto slashes;
2272 inode = dentry->d_inode;
2273 if (inode)
2274 atomic_inc(&inode->i_count);
Dave Hansen06227532008-02-15 14:37:34 -08002275 error = mnt_want_write(nd.path.mnt);
2276 if (error)
2277 goto exit2;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002278 error = security_path_unlink(&nd.path, dentry);
2279 if (error)
2280 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08002281 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002282exit3:
Dave Hansen06227532008-02-15 14:37:34 -08002283 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284 exit2:
2285 dput(dentry);
2286 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002287 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288 if (inode)
2289 iput(inode); /* truncate the inode here */
2290exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002291 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 putname(name);
2293 return error;
2294
2295slashes:
2296 error = !dentry->d_inode ? -ENOENT :
2297 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2298 goto exit2;
2299}
2300
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002301SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002302{
2303 if ((flag & ~AT_REMOVEDIR) != 0)
2304 return -EINVAL;
2305
2306 if (flag & AT_REMOVEDIR)
2307 return do_rmdir(dfd, pathname);
2308
2309 return do_unlinkat(dfd, pathname);
2310}
2311
Heiko Carstens3480b252009-01-14 14:14:16 +01002312SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002313{
2314 return do_unlinkat(AT_FDCWD, pathname);
2315}
2316
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002317int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002319 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320
2321 if (error)
2322 return error;
2323
Al Viroacfa4382008-12-04 10:06:33 -05002324 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325 return -EPERM;
2326
2327 error = security_inode_symlink(dir, dentry, oldname);
2328 if (error)
2329 return error;
2330
Jan Kara9e3509e2009-01-26 16:45:12 +01002331 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002333 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002334 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335 return error;
2336}
2337
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002338SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2339 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340{
Al Viro2ad94ae2008-07-21 09:32:51 -04002341 int error;
2342 char *from;
2343 char *to;
Dave Hansen6902d922006-09-30 23:29:01 -07002344 struct dentry *dentry;
2345 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346
2347 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04002348 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 return PTR_ERR(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002350
2351 error = user_path_parent(newdfd, newname, &nd, &to);
2352 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002353 goto out_putname;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354
Dave Hansen6902d922006-09-30 23:29:01 -07002355 dentry = lookup_create(&nd, 0);
2356 error = PTR_ERR(dentry);
2357 if (IS_ERR(dentry))
2358 goto out_unlock;
2359
Dave Hansen75c3f292008-02-15 14:37:45 -08002360 error = mnt_want_write(nd.path.mnt);
2361 if (error)
2362 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002363 error = security_path_symlink(&nd.path, dentry, from);
2364 if (error)
2365 goto out_drop_write;
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002366 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002367out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002368 mnt_drop_write(nd.path.mnt);
2369out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002370 dput(dentry);
2371out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002372 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002373 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002374 putname(to);
2375out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376 putname(from);
2377 return error;
2378}
2379
Heiko Carstens3480b252009-01-14 14:14:16 +01002380SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002381{
2382 return sys_symlinkat(oldname, AT_FDCWD, newname);
2383}
2384
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2386{
2387 struct inode *inode = old_dentry->d_inode;
2388 int error;
2389
2390 if (!inode)
2391 return -ENOENT;
2392
Miklos Szeredia95164d2008-07-30 15:08:48 +02002393 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394 if (error)
2395 return error;
2396
2397 if (dir->i_sb != inode->i_sb)
2398 return -EXDEV;
2399
2400 /*
2401 * A link to an append-only or immutable file cannot be created.
2402 */
2403 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2404 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05002405 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002407 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408 return -EPERM;
2409
2410 error = security_inode_link(old_dentry, dir, new_dentry);
2411 if (error)
2412 return error;
2413
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002414 mutex_lock(&inode->i_mutex);
Jan Kara9e3509e2009-01-26 16:45:12 +01002415 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416 error = dir->i_op->link(old_dentry, dir, new_dentry);
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002417 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002418 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002419 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420 return error;
2421}
2422
2423/*
2424 * Hardlinks are often used in delicate situations. We avoid
2425 * security-related surprises by not following symlinks on the
2426 * newname. --KAB
2427 *
2428 * We don't follow them on the oldname either to be compatible
2429 * with linux 2.0, and to avoid hard-linking to directories
2430 * and other special files. --ADM
2431 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002432SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2433 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434{
2435 struct dentry *new_dentry;
Al Viro2d8f3032008-07-22 09:59:21 -04002436 struct nameidata nd;
2437 struct path old_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438 int error;
Al Viro2ad94ae2008-07-21 09:32:51 -04002439 char *to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440
Ulrich Drepper45c9b112006-06-25 05:49:11 -07002441 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002442 return -EINVAL;
2443
Al Viro2d8f3032008-07-22 09:59:21 -04002444 error = user_path_at(olddfd, oldname,
2445 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2446 &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002448 return error;
2449
2450 error = user_path_parent(newdfd, newname, &nd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451 if (error)
2452 goto out;
2453 error = -EXDEV;
Al Viro2d8f3032008-07-22 09:59:21 -04002454 if (old_path.mnt != nd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455 goto out_release;
2456 new_dentry = lookup_create(&nd, 0);
2457 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002458 if (IS_ERR(new_dentry))
2459 goto out_unlock;
Dave Hansen75c3f292008-02-15 14:37:45 -08002460 error = mnt_want_write(nd.path.mnt);
2461 if (error)
2462 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002463 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2464 if (error)
2465 goto out_drop_write;
Al Viro2d8f3032008-07-22 09:59:21 -04002466 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002467out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002468 mnt_drop_write(nd.path.mnt);
2469out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002470 dput(new_dentry);
2471out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002472 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002473out_release:
Jan Blunck1d957f92008-02-14 19:34:35 -08002474 path_put(&nd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002475 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476out:
Al Viro2d8f3032008-07-22 09:59:21 -04002477 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478
2479 return error;
2480}
2481
Heiko Carstens3480b252009-01-14 14:14:16 +01002482SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002483{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002484 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002485}
2486
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487/*
2488 * The worst of all namespace operations - renaming directory. "Perverted"
2489 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2490 * Problems:
2491 * a) we can get into loop creation. Check is done in is_subdir().
2492 * b) race potential - two innocent renames can create a loop together.
2493 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002494 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495 * story.
2496 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002497 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498 * whether the target exists). Solution: try to be smart with locking
2499 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002500 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07002501 * move will be locked. Thus we can rank directories by the tree
2502 * (ancestors first) and rank all non-directories after them.
2503 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002504 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505 * HOWEVER, it relies on the assumption that any object with ->lookup()
2506 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2507 * we'd better make sure that there's no link(2) for them.
2508 * d) some filesystems don't support opened-but-unlinked directories,
2509 * either because of layout or because they are not ready to deal with
2510 * all cases correctly. The latter will be fixed (taking this sort of
2511 * stuff into VFS), but the former is not going away. Solution: the same
2512 * trick as in rmdir().
2513 * e) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002514 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002516 * ->i_mutex on parents, which works but leads to some truely excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517 * locking].
2518 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07002519static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2520 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521{
2522 int error = 0;
2523 struct inode *target;
2524
2525 /*
2526 * If we are going to change the parent - check write permissions,
2527 * we'll need to flip '..'.
2528 */
2529 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04002530 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531 if (error)
2532 return error;
2533 }
2534
2535 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2536 if (error)
2537 return error;
2538
2539 target = new_dentry->d_inode;
2540 if (target) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002541 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542 dentry_unhash(new_dentry);
2543 }
2544 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2545 error = -EBUSY;
2546 else
2547 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2548 if (target) {
2549 if (!error)
2550 target->i_flags |= S_DEAD;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002551 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552 if (d_unhashed(new_dentry))
2553 d_rehash(new_dentry);
2554 dput(new_dentry);
2555 }
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002556 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07002557 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2558 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 return error;
2560}
2561
Adrian Bunk75c96f82005-05-05 16:16:09 -07002562static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2563 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564{
2565 struct inode *target;
2566 int error;
2567
2568 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2569 if (error)
2570 return error;
2571
2572 dget(new_dentry);
2573 target = new_dentry->d_inode;
2574 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002575 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002576 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2577 error = -EBUSY;
2578 else
2579 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2580 if (!error) {
Mark Fasheh349457c2006-09-08 14:22:21 -07002581 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582 d_move(old_dentry, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583 }
2584 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002585 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586 dput(new_dentry);
2587 return error;
2588}
2589
2590int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2591 struct inode *new_dir, struct dentry *new_dentry)
2592{
2593 int error;
2594 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Robert Love0eeca282005-07-12 17:06:03 -04002595 const char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596
2597 if (old_dentry->d_inode == new_dentry->d_inode)
2598 return 0;
2599
2600 error = may_delete(old_dir, old_dentry, is_dir);
2601 if (error)
2602 return error;
2603
2604 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02002605 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002606 else
2607 error = may_delete(new_dir, new_dentry, is_dir);
2608 if (error)
2609 return error;
2610
Al Viroacfa4382008-12-04 10:06:33 -05002611 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612 return -EPERM;
2613
Jan Kara9e3509e2009-01-26 16:45:12 +01002614 vfs_dq_init(old_dir);
2615 vfs_dq_init(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616
Robert Love0eeca282005-07-12 17:06:03 -04002617 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2618
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619 if (is_dir)
2620 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2621 else
2622 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2623 if (!error) {
Robert Love0eeca282005-07-12 17:06:03 -04002624 const char *new_name = old_dentry->d_name.name;
John McCutchan89204c42005-08-15 12:13:28 -04002625 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04002626 new_dentry->d_inode, old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 }
Robert Love0eeca282005-07-12 17:06:03 -04002628 fsnotify_oldname_free(old_name);
2629
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630 return error;
2631}
2632
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002633SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2634 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635{
Al Viro2ad94ae2008-07-21 09:32:51 -04002636 struct dentry *old_dir, *new_dir;
2637 struct dentry *old_dentry, *new_dentry;
2638 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639 struct nameidata oldnd, newnd;
Al Viro2ad94ae2008-07-21 09:32:51 -04002640 char *from;
2641 char *to;
2642 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002643
Al Viro2ad94ae2008-07-21 09:32:51 -04002644 error = user_path_parent(olddfd, oldname, &oldnd, &from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645 if (error)
2646 goto exit;
2647
Al Viro2ad94ae2008-07-21 09:32:51 -04002648 error = user_path_parent(newdfd, newname, &newnd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649 if (error)
2650 goto exit1;
2651
2652 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08002653 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654 goto exit2;
2655
Jan Blunck4ac91372008-02-14 19:34:32 -08002656 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657 error = -EBUSY;
2658 if (oldnd.last_type != LAST_NORM)
2659 goto exit2;
2660
Jan Blunck4ac91372008-02-14 19:34:32 -08002661 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662 if (newnd.last_type != LAST_NORM)
2663 goto exit2;
2664
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002665 oldnd.flags &= ~LOOKUP_PARENT;
2666 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09002667 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002668
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669 trap = lock_rename(new_dir, old_dir);
2670
Christoph Hellwig49705b72005-11-08 21:35:06 -08002671 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672 error = PTR_ERR(old_dentry);
2673 if (IS_ERR(old_dentry))
2674 goto exit3;
2675 /* source must exist */
2676 error = -ENOENT;
2677 if (!old_dentry->d_inode)
2678 goto exit4;
2679 /* unless the source is a directory trailing slashes give -ENOTDIR */
2680 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2681 error = -ENOTDIR;
2682 if (oldnd.last.name[oldnd.last.len])
2683 goto exit4;
2684 if (newnd.last.name[newnd.last.len])
2685 goto exit4;
2686 }
2687 /* source should not be ancestor of target */
2688 error = -EINVAL;
2689 if (old_dentry == trap)
2690 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08002691 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692 error = PTR_ERR(new_dentry);
2693 if (IS_ERR(new_dentry))
2694 goto exit4;
2695 /* target should not be an ancestor of source */
2696 error = -ENOTEMPTY;
2697 if (new_dentry == trap)
2698 goto exit5;
2699
Dave Hansen9079b1e2008-02-15 14:37:49 -08002700 error = mnt_want_write(oldnd.path.mnt);
2701 if (error)
2702 goto exit5;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002703 error = security_path_rename(&oldnd.path, old_dentry,
2704 &newnd.path, new_dentry);
2705 if (error)
2706 goto exit6;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707 error = vfs_rename(old_dir->d_inode, old_dentry,
2708 new_dir->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002709exit6:
Dave Hansen9079b1e2008-02-15 14:37:49 -08002710 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711exit5:
2712 dput(new_dentry);
2713exit4:
2714 dput(old_dentry);
2715exit3:
2716 unlock_rename(new_dir, old_dir);
2717exit2:
Jan Blunck1d957f92008-02-14 19:34:35 -08002718 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002719 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002721 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722 putname(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002723exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002724 return error;
2725}
2726
Heiko Carstensa26eab22009-01-14 14:14:17 +01002727SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002728{
2729 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2730}
2731
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2733{
2734 int len;
2735
2736 len = PTR_ERR(link);
2737 if (IS_ERR(link))
2738 goto out;
2739
2740 len = strlen(link);
2741 if (len > (unsigned) buflen)
2742 len = buflen;
2743 if (copy_to_user(buffer, link, len))
2744 len = -EFAULT;
2745out:
2746 return len;
2747}
2748
2749/*
2750 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2751 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2752 * using) it for any given inode is up to filesystem.
2753 */
2754int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2755{
2756 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002757 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07002758 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002759
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002761 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07002762 if (IS_ERR(cookie))
2763 return PTR_ERR(cookie);
2764
2765 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2766 if (dentry->d_inode->i_op->put_link)
2767 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2768 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002769}
2770
2771int vfs_follow_link(struct nameidata *nd, const char *link)
2772{
2773 return __vfs_follow_link(nd, link);
2774}
2775
2776/* get the link contents into pagecache */
2777static char *page_getlink(struct dentry * dentry, struct page **ppage)
2778{
Duane Griffinebd09ab2008-12-19 20:47:12 +00002779 char *kaddr;
2780 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002781 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07002782 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002783 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07002784 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00002786 kaddr = kmap(page);
2787 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2788 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002789}
2790
2791int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2792{
2793 struct page *page = NULL;
2794 char *s = page_getlink(dentry, &page);
2795 int res = vfs_readlink(dentry,buffer,buflen,s);
2796 if (page) {
2797 kunmap(page);
2798 page_cache_release(page);
2799 }
2800 return res;
2801}
2802
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002803void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002804{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002805 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002806 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002807 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002808}
2809
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002810void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002812 struct page *page = cookie;
2813
2814 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815 kunmap(page);
2816 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817 }
2818}
2819
Nick Piggin54566b22009-01-04 12:00:53 -08002820/*
2821 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2822 */
2823int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002824{
2825 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002826 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002827 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08002828 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002829 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08002830 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2831 if (nofs)
2832 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002833
NeilBrown7e53cac2006-03-25 03:07:57 -08002834retry:
Nick Pigginafddba42007-10-16 01:25:01 -07002835 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08002836 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07002838 goto fail;
2839
Linus Torvalds1da177e2005-04-16 15:20:36 -07002840 kaddr = kmap_atomic(page, KM_USER0);
2841 memcpy(kaddr, symname, len-1);
2842 kunmap_atomic(kaddr, KM_USER0);
Nick Pigginafddba42007-10-16 01:25:01 -07002843
2844 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2845 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002846 if (err < 0)
2847 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07002848 if (err < len-1)
2849 goto retry;
2850
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851 mark_inode_dirty(inode);
2852 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002853fail:
2854 return err;
2855}
2856
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002857int page_symlink(struct inode *inode, const char *symname, int len)
2858{
2859 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08002860 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002861}
2862
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002863const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864 .readlink = generic_readlink,
2865 .follow_link = page_follow_link_light,
2866 .put_link = page_put_link,
2867};
2868
Al Viro2d8f3032008-07-22 09:59:21 -04002869EXPORT_SYMBOL(user_path_at);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002870EXPORT_SYMBOL(follow_down);
2871EXPORT_SYMBOL(follow_up);
2872EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2873EXPORT_SYMBOL(getname);
2874EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002875EXPORT_SYMBOL(lookup_one_len);
2876EXPORT_SYMBOL(page_follow_link_light);
2877EXPORT_SYMBOL(page_put_link);
2878EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002879EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002880EXPORT_SYMBOL(page_symlink);
2881EXPORT_SYMBOL(page_symlink_inode_operations);
2882EXPORT_SYMBOL(path_lookup);
Al Virod1811462008-08-02 00:49:18 -04002883EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002884EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04002885EXPORT_SYMBOL(inode_permission);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -08002886EXPORT_SYMBOL(file_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002887EXPORT_SYMBOL(unlock_rename);
2888EXPORT_SYMBOL(vfs_create);
2889EXPORT_SYMBOL(vfs_follow_link);
2890EXPORT_SYMBOL(vfs_link);
2891EXPORT_SYMBOL(vfs_mkdir);
2892EXPORT_SYMBOL(vfs_mknod);
2893EXPORT_SYMBOL(generic_permission);
2894EXPORT_SYMBOL(vfs_readlink);
2895EXPORT_SYMBOL(vfs_rename);
2896EXPORT_SYMBOL(vfs_rmdir);
2897EXPORT_SYMBOL(vfs_symlink);
2898EXPORT_SYMBOL(vfs_unlink);
2899EXPORT_SYMBOL(dentry_unhash);
2900EXPORT_SYMBOL(generic_readlink);