blob: d11f404667e962ababa87f3678a52e78c50794dd [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
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700172/*
173 * This does basic POSIX ACL permission checking
174 */
175static int acl_permission_check(struct inode *inode, int mask,
176 int (*check_acl)(struct inode *inode, int mask))
177{
178 umode_t mode = inode->i_mode;
179
180 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
181
182 if (current_fsuid() == inode->i_uid)
183 mode >>= 6;
184 else {
185 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
186 int error = check_acl(inode, mask);
187 if (error != -EAGAIN)
188 return error;
189 }
190
191 if (in_group_p(inode->i_gid))
192 mode >>= 3;
193 }
194
195 /*
196 * If the DACs are ok we don't need any capability check.
197 */
198 if ((mask & ~mode) == 0)
199 return 0;
200 return -EACCES;
201}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202
203/**
204 * generic_permission - check for access rights on a Posix-like filesystem
205 * @inode: inode to check access rights for
206 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
207 * @check_acl: optional callback to check for Posix ACLs
208 *
209 * Used to check for read/write/execute permissions on a file.
210 * We use "fsuid" for this, letting us set arbitrary permissions
211 * for filesystem access without changing the "normal" uids which
212 * are used for other things..
213 */
214int generic_permission(struct inode *inode, int mask,
215 int (*check_acl)(struct inode *inode, int mask))
216{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700217 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218
219 /*
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700220 * Do the basic POSIX ACL permission checks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 */
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700222 ret = acl_permission_check(inode, mask, check_acl);
223 if (ret != -EACCES)
224 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 /*
227 * Read/write DACs are always overridable.
228 * Executable DACs are overridable if at least one exec bit is set.
229 */
Miklos Szeredif696a362008-07-31 13:41:58 +0200230 if (!(mask & MAY_EXEC) || execute_ok(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 if (capable(CAP_DAC_OVERRIDE))
232 return 0;
233
234 /*
235 * Searching includes executable on directories, else just read.
236 */
237 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
238 if (capable(CAP_DAC_READ_SEARCH))
239 return 0;
240
241 return -EACCES;
242}
243
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200244/**
245 * inode_permission - check for access rights to a given inode
246 * @inode: inode to check permission on
247 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
248 *
249 * Used to check for read/write/execute permissions on an inode.
250 * We use "fsuid" for this, letting us set arbitrary permissions
251 * for filesystem access without changing the "normal" uids which
252 * are used for other things.
253 */
Al Virof419a2e2008-07-22 00:07:17 -0400254int inode_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255{
Al Viroe6305c42008-07-15 21:03:57 -0400256 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257
258 if (mask & MAY_WRITE) {
Miklos Szeredi22590e42007-10-16 23:27:08 -0700259 umode_t mode = inode->i_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260
261 /*
262 * Nobody gets write access to a read-only fs.
263 */
264 if (IS_RDONLY(inode) &&
265 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
266 return -EROFS;
267
268 /*
269 * Nobody gets write access to an immutable file.
270 */
271 if (IS_IMMUTABLE(inode))
272 return -EACCES;
273 }
274
Al Viroacfa4382008-12-04 10:06:33 -0500275 if (inode->i_op->permission)
Al Virob77b0642008-07-17 09:37:02 -0400276 retval = inode->i_op->permission(inode, mask);
Miklos Szeredif696a362008-07-31 13:41:58 +0200277 else
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700278 retval = generic_permission(inode, mask, inode->i_op->check_acl);
Miklos Szeredif696a362008-07-31 13:41:58 +0200279
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 if (retval)
281 return retval;
282
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -0700283 retval = devcgroup_inode_permission(inode, mask);
284 if (retval)
285 return retval;
286
Al Viroe6305c42008-07-15 21:03:57 -0400287 return security_inode_permission(inode,
Stephen Smalleyf418b002008-07-28 13:32:38 -0400288 mask & (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289}
290
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800291/**
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800292 * file_permission - check for additional access rights to a given file
293 * @file: file to check access rights for
294 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
295 *
296 * Used to check for read/write/execute permissions on an already opened
297 * file.
298 *
299 * Note:
300 * Do not use this function in new code. All access checks should
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200301 * be done using inode_permission().
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800302 */
303int file_permission(struct file *file, int mask)
304{
Al Virof419a2e2008-07-22 00:07:17 -0400305 return inode_permission(file->f_path.dentry->d_inode, mask);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800306}
307
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308/*
309 * get_write_access() gets write permission for a file.
310 * put_write_access() releases this write permission.
311 * This is used for regular files.
312 * We cannot support write (and maybe mmap read-write shared) accesses and
313 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
314 * can have the following values:
315 * 0: no writers, no VM_DENYWRITE mappings
316 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
317 * > 0: (i_writecount) users are writing to the file.
318 *
319 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
320 * except for the cases where we don't hold i_writecount yet. Then we need to
321 * use {get,deny}_write_access() - these functions check the sign and refuse
322 * to do the change if sign is wrong. Exclusion between them is provided by
323 * the inode->i_lock spinlock.
324 */
325
326int get_write_access(struct inode * inode)
327{
328 spin_lock(&inode->i_lock);
329 if (atomic_read(&inode->i_writecount) < 0) {
330 spin_unlock(&inode->i_lock);
331 return -ETXTBSY;
332 }
333 atomic_inc(&inode->i_writecount);
334 spin_unlock(&inode->i_lock);
335
336 return 0;
337}
338
339int deny_write_access(struct file * file)
340{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800341 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342
343 spin_lock(&inode->i_lock);
344 if (atomic_read(&inode->i_writecount) > 0) {
345 spin_unlock(&inode->i_lock);
346 return -ETXTBSY;
347 }
348 atomic_dec(&inode->i_writecount);
349 spin_unlock(&inode->i_lock);
350
351 return 0;
352}
353
Jan Blunck1d957f92008-02-14 19:34:35 -0800354/**
Jan Blunck5dd784d02008-02-14 19:34:38 -0800355 * path_get - get a reference to a path
356 * @path: path to get the reference to
357 *
358 * Given a path increment the reference count to the dentry and the vfsmount.
359 */
360void path_get(struct path *path)
361{
362 mntget(path->mnt);
363 dget(path->dentry);
364}
365EXPORT_SYMBOL(path_get);
366
367/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800368 * path_put - put a reference to a path
369 * @path: path to put the reference to
370 *
371 * Given a path decrement the reference count to the dentry and the vfsmount.
372 */
373void path_put(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374{
Jan Blunck1d957f92008-02-14 19:34:35 -0800375 dput(path->dentry);
376 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377}
Jan Blunck1d957f92008-02-14 19:34:35 -0800378EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379
Trond Myklebust834f2a42005-10-18 14:20:16 -0700380/**
381 * release_open_intent - free up open intent resources
382 * @nd: pointer to nameidata
383 */
384void release_open_intent(struct nameidata *nd)
385{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800386 if (nd->intent.open.file->f_path.dentry == NULL)
Trond Myklebust834f2a42005-10-18 14:20:16 -0700387 put_filp(nd->intent.open.file);
388 else
389 fput(nd->intent.open.file);
390}
391
Ian Kentbcdc5e02006-09-27 01:50:44 -0700392static inline struct dentry *
393do_revalidate(struct dentry *dentry, struct nameidata *nd)
394{
395 int status = dentry->d_op->d_revalidate(dentry, nd);
396 if (unlikely(status <= 0)) {
397 /*
398 * The dentry failed validation.
399 * If d_revalidate returned 0 attempt to invalidate
400 * the dentry otherwise d_revalidate is asking us
401 * to return a fail status.
402 */
403 if (!status) {
404 if (!d_invalidate(dentry)) {
405 dput(dentry);
406 dentry = NULL;
407 }
408 } else {
409 dput(dentry);
410 dentry = ERR_PTR(status);
411 }
412 }
413 return dentry;
414}
415
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416/*
417 * Internal lookup() using the new generic dcache.
418 * SMP-safe
419 */
420static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
421{
422 struct dentry * dentry = __d_lookup(parent, name);
423
424 /* lockess __d_lookup may fail due to concurrent d_move()
425 * in some unrelated directory, so try with d_lookup
426 */
427 if (!dentry)
428 dentry = d_lookup(parent, name);
429
Ian Kentbcdc5e02006-09-27 01:50:44 -0700430 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
431 dentry = do_revalidate(dentry, nd);
432
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 return dentry;
434}
435
436/*
437 * Short-cut version of permission(), for calling by
438 * path_walk(), when dcache lock is held. Combines parts
439 * of permission() and generic_permission(), and tests ONLY for
440 * MAY_EXEC permission.
441 *
442 * If appropriate, check DAC only. If not appropriate, or
443 * short-cut DAC fails, then call permission() to do more
444 * complete permission check.
445 */
Al Viro672b16b2008-07-17 09:45:01 -0400446static int exec_permission_lite(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700448 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449
Linus Torvaldscb9179e2009-08-28 11:08:31 -0700450 if (inode->i_op->permission) {
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700451 ret = inode->i_op->permission(inode, MAY_EXEC);
Linus Torvaldscb9179e2009-08-28 11:08:31 -0700452 if (!ret)
453 goto ok;
454 return ret;
455 }
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700456 ret = acl_permission_check(inode, MAY_EXEC, inode->i_op->check_acl);
457 if (!ret)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 goto ok;
459
Linus Torvaldsf1ac9f62009-08-28 10:53:56 -0700460 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 goto ok;
462
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700463 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464ok:
Al Virob77b0642008-07-17 09:37:02 -0400465 return security_inode_permission(inode, MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466}
467
468/*
469 * This is called when everything else fails, and we actually have
470 * to go to the low-level filesystem to find out what we should do..
471 *
472 * We get the directory semaphore, and after getting that we also
473 * make sure that nobody added the entry to the dcache in the meantime..
474 * SMP-safe
475 */
476static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
477{
478 struct dentry * result;
479 struct inode *dir = parent->d_inode;
480
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800481 mutex_lock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 /*
483 * First re-do the cached lookup just in case it was created
484 * while we waited for the directory semaphore..
485 *
486 * FIXME! This could use version numbering or similar to
487 * avoid unnecessary cache lookups.
488 *
489 * The "dcache_lock" is purely to protect the RCU list walker
490 * from concurrent renames at this point (we mustn't get false
491 * negatives from the RCU list walk here, unlike the optimistic
492 * fast walk).
493 *
494 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
495 */
496 result = d_lookup(parent, name);
497 if (!result) {
Miklos Szeredid70b67c2008-07-02 21:30:15 +0200498 struct dentry *dentry;
499
500 /* Don't create child dentry for a dead directory. */
501 result = ERR_PTR(-ENOENT);
502 if (IS_DEADDIR(dir))
503 goto out_unlock;
504
505 dentry = d_alloc(parent, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 result = ERR_PTR(-ENOMEM);
507 if (dentry) {
508 result = dir->i_op->lookup(dir, dentry, nd);
509 if (result)
510 dput(dentry);
511 else
512 result = dentry;
513 }
Miklos Szeredid70b67c2008-07-02 21:30:15 +0200514out_unlock:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800515 mutex_unlock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 return result;
517 }
518
519 /*
520 * Uhhuh! Nasty case: the cache was re-populated while
521 * we waited on the semaphore. Need to revalidate.
522 */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800523 mutex_unlock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 if (result->d_op && result->d_op->d_revalidate) {
Ian Kentbcdc5e02006-09-27 01:50:44 -0700525 result = do_revalidate(result, nd);
526 if (!result)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 result = ERR_PTR(-ENOENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 }
529 return result;
530}
531
Al Viroa02f76c2008-02-23 15:14:28 +0000532/*
533 * Wrapper to retry pathname resolution whenever the underlying
534 * file system returns an ESTALE.
535 *
536 * Retry the whole path once, forcing real lookup requests
537 * instead of relying on the dcache.
538 */
539static __always_inline int link_path_walk(const char *name, struct nameidata *nd)
540{
541 struct path save = nd->path;
542 int result;
543
544 /* make sure the stuff we saved doesn't go away */
Jan Blunckc8e7f442008-06-09 16:40:35 -0700545 path_get(&save);
Al Viroa02f76c2008-02-23 15:14:28 +0000546
547 result = __link_path_walk(name, nd);
548 if (result == -ESTALE) {
549 /* nd->path had been dropped */
550 nd->path = save;
Jan Blunckc8e7f442008-06-09 16:40:35 -0700551 path_get(&nd->path);
Al Viroa02f76c2008-02-23 15:14:28 +0000552 nd->flags |= LOOKUP_REVAL;
553 result = __link_path_walk(name, nd);
554 }
555
556 path_put(&save);
557
558 return result;
559}
560
Al Viro2a737872009-04-07 11:49:53 -0400561static __always_inline void set_root(struct nameidata *nd)
562{
563 if (!nd->root.mnt) {
564 struct fs_struct *fs = current->fs;
565 read_lock(&fs->lock);
566 nd->root = fs->root;
567 path_get(&nd->root);
568 read_unlock(&fs->lock);
569 }
570}
571
Arjan van de Venf1662352006-01-14 13:21:31 -0800572static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573{
574 int res = 0;
575 char *name;
576 if (IS_ERR(link))
577 goto fail;
578
579 if (*link == '/') {
Al Viro2a737872009-04-07 11:49:53 -0400580 set_root(nd);
Jan Blunck1d957f92008-02-14 19:34:35 -0800581 path_put(&nd->path);
Al Viro2a737872009-04-07 11:49:53 -0400582 nd->path = nd->root;
583 path_get(&nd->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 }
Christoph Hellwigb4091d52008-11-05 15:07:21 +0100585
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 res = link_path_walk(link, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 if (nd->depth || res || nd->last_type!=LAST_NORM)
588 return res;
589 /*
590 * If it is an iterative symlinks resolution in open_namei() we
591 * have to copy the last component. And all that crap because of
592 * bloody create() on broken symlinks. Furrfu...
593 */
594 name = __getname();
595 if (unlikely(!name)) {
Jan Blunck1d957f92008-02-14 19:34:35 -0800596 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 return -ENOMEM;
598 }
599 strcpy(name, nd->last.name);
600 nd->last.name = name;
601 return 0;
602fail:
Jan Blunck1d957f92008-02-14 19:34:35 -0800603 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 return PTR_ERR(link);
605}
606
Jan Blunck1d957f92008-02-14 19:34:35 -0800607static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700608{
609 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800610 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700611 mntput(path->mnt);
612}
613
614static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
615{
Jan Blunck4ac91372008-02-14 19:34:32 -0800616 dput(nd->path.dentry);
617 if (nd->path.mnt != path->mnt)
618 mntput(nd->path.mnt);
619 nd->path.mnt = path->mnt;
620 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700621}
622
Ian Kent051d3812006-03-27 01:14:53 -0800623static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
624{
625 int error;
626 void *cookie;
627 struct dentry *dentry = path->dentry;
628
629 touch_atime(path->mnt, dentry);
630 nd_set_link(nd, NULL);
631
Jan Blunck4ac91372008-02-14 19:34:32 -0800632 if (path->mnt != nd->path.mnt) {
Ian Kent051d3812006-03-27 01:14:53 -0800633 path_to_nameidata(path, nd);
634 dget(dentry);
635 }
636 mntget(path->mnt);
637 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
638 error = PTR_ERR(cookie);
639 if (!IS_ERR(cookie)) {
640 char *s = nd_get_link(nd);
641 error = 0;
642 if (s)
643 error = __vfs_follow_link(nd, s);
644 if (dentry->d_inode->i_op->put_link)
645 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
646 }
Jan Blunck09da59162008-02-14 19:34:37 -0800647 path_put(path);
Ian Kent051d3812006-03-27 01:14:53 -0800648
649 return error;
650}
651
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652/*
653 * This limits recursive symlink follows to 8, while
654 * limiting consecutive symlinks to 40.
655 *
656 * Without that kind of total limit, nasty chains of consecutive
657 * symlinks can cause almost arbitrarily long lookups.
658 */
Al Viro90ebe562005-06-06 13:35:58 -0700659static inline int do_follow_link(struct path *path, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660{
661 int err = -ELOOP;
662 if (current->link_count >= MAX_NESTED_LINKS)
663 goto loop;
664 if (current->total_link_count >= 40)
665 goto loop;
666 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
667 cond_resched();
Al Viro90ebe562005-06-06 13:35:58 -0700668 err = security_inode_follow_link(path->dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 if (err)
670 goto loop;
671 current->link_count++;
672 current->total_link_count++;
673 nd->depth++;
Al Virocd4e91d2005-06-06 13:36:03 -0700674 err = __do_follow_link(path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 current->link_count--;
676 nd->depth--;
677 return err;
678loop:
Jan Blunck1d957f92008-02-14 19:34:35 -0800679 path_put_conditional(path, nd);
680 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 return err;
682}
683
Al Virobab77eb2009-04-18 03:26:48 -0400684int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685{
686 struct vfsmount *parent;
687 struct dentry *mountpoint;
688 spin_lock(&vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400689 parent = path->mnt->mnt_parent;
690 if (parent == path->mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 spin_unlock(&vfsmount_lock);
692 return 0;
693 }
694 mntget(parent);
Al Virobab77eb2009-04-18 03:26:48 -0400695 mountpoint = dget(path->mnt->mnt_mountpoint);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 spin_unlock(&vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400697 dput(path->dentry);
698 path->dentry = mountpoint;
699 mntput(path->mnt);
700 path->mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 return 1;
702}
703
704/* no need for dcache_lock, as serialization is taken care in
705 * namespace.c
706 */
Al Viro463ffb22005-06-06 13:36:05 -0700707static int __follow_mount(struct path *path)
708{
709 int res = 0;
710 while (d_mountpoint(path->dentry)) {
Al Viro1c755af2009-04-18 14:06:57 -0400711 struct vfsmount *mounted = lookup_mnt(path);
Al Viro463ffb22005-06-06 13:36:05 -0700712 if (!mounted)
713 break;
714 dput(path->dentry);
715 if (res)
716 mntput(path->mnt);
717 path->mnt = mounted;
718 path->dentry = dget(mounted->mnt_root);
719 res = 1;
720 }
721 return res;
722}
723
Al Viro79ed0222009-04-18 13:59:41 -0400724static void follow_mount(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725{
Al Viro79ed0222009-04-18 13:59:41 -0400726 while (d_mountpoint(path->dentry)) {
Al Viro1c755af2009-04-18 14:06:57 -0400727 struct vfsmount *mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 if (!mounted)
729 break;
Al Viro79ed0222009-04-18 13:59:41 -0400730 dput(path->dentry);
731 mntput(path->mnt);
732 path->mnt = mounted;
733 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735}
736
737/* no need for dcache_lock, as serialization is taken care in
738 * namespace.c
739 */
Al Viro9393bd02009-04-18 13:58:15 -0400740int follow_down(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741{
742 struct vfsmount *mounted;
743
Al Viro1c755af2009-04-18 14:06:57 -0400744 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -0400746 dput(path->dentry);
747 mntput(path->mnt);
748 path->mnt = mounted;
749 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 return 1;
751 }
752 return 0;
753}
754
Arjan van de Venf1662352006-01-14 13:21:31 -0800755static __always_inline void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756{
Al Viro2a737872009-04-07 11:49:53 -0400757 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -0700758
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 while(1) {
760 struct vfsmount *parent;
Jan Blunck4ac91372008-02-14 19:34:32 -0800761 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762
Al Viro2a737872009-04-07 11:49:53 -0400763 if (nd->path.dentry == nd->root.dentry &&
764 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 break;
766 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 spin_lock(&dcache_lock);
Jan Blunck4ac91372008-02-14 19:34:32 -0800768 if (nd->path.dentry != nd->path.mnt->mnt_root) {
769 nd->path.dentry = dget(nd->path.dentry->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 spin_unlock(&dcache_lock);
771 dput(old);
772 break;
773 }
774 spin_unlock(&dcache_lock);
775 spin_lock(&vfsmount_lock);
Jan Blunck4ac91372008-02-14 19:34:32 -0800776 parent = nd->path.mnt->mnt_parent;
777 if (parent == nd->path.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 spin_unlock(&vfsmount_lock);
779 break;
780 }
781 mntget(parent);
Jan Blunck4ac91372008-02-14 19:34:32 -0800782 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 spin_unlock(&vfsmount_lock);
784 dput(old);
Jan Blunck4ac91372008-02-14 19:34:32 -0800785 mntput(nd->path.mnt);
786 nd->path.mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 }
Al Viro79ed0222009-04-18 13:59:41 -0400788 follow_mount(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789}
790
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791/*
792 * It's more convoluted than I'd like it to be, but... it's still fairly
793 * small and for now I'd prefer to have fast path as straight as possible.
794 * It _is_ time-critical.
795 */
796static int do_lookup(struct nameidata *nd, struct qstr *name,
797 struct path *path)
798{
Jan Blunck4ac91372008-02-14 19:34:32 -0800799 struct vfsmount *mnt = nd->path.mnt;
800 struct dentry *dentry = __d_lookup(nd->path.dentry, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801
802 if (!dentry)
803 goto need_lookup;
804 if (dentry->d_op && dentry->d_op->d_revalidate)
805 goto need_revalidate;
806done:
807 path->mnt = mnt;
808 path->dentry = dentry;
Al Viro634ee702005-06-06 13:36:13 -0700809 __follow_mount(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 return 0;
811
812need_lookup:
Jan Blunck4ac91372008-02-14 19:34:32 -0800813 dentry = real_lookup(nd->path.dentry, name, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 if (IS_ERR(dentry))
815 goto fail;
816 goto done;
817
818need_revalidate:
Ian Kentbcdc5e02006-09-27 01:50:44 -0700819 dentry = do_revalidate(dentry, nd);
820 if (!dentry)
821 goto need_lookup;
822 if (IS_ERR(dentry))
823 goto fail;
824 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825
826fail:
827 return PTR_ERR(dentry);
828}
829
830/*
831 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +0100832 * This is the basic name resolution function, turning a pathname into
833 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 *
Prasanna Medaea3834d2005-04-29 16:00:17 +0100835 * Returns 0 and nd will have valid dentry and mnt on success.
836 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800838static int __link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839{
840 struct path next;
841 struct inode *inode;
842 int err;
843 unsigned int lookup_flags = nd->flags;
844
845 while (*name=='/')
846 name++;
847 if (!*name)
848 goto return_reval;
849
Jan Blunck4ac91372008-02-14 19:34:32 -0800850 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 if (nd->depth)
Trond Myklebustf55eab82006-02-04 23:28:01 -0800852 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853
854 /* At this point we know we have a real path component. */
855 for(;;) {
856 unsigned long hash;
857 struct qstr this;
858 unsigned int c;
859
Trond Myklebustcdce5d62005-10-18 14:20:18 -0700860 nd->flags |= LOOKUP_CONTINUE;
Al Viro672b16b2008-07-17 09:45:01 -0400861 err = exec_permission_lite(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 if (err)
863 break;
864
865 this.name = name;
866 c = *(const unsigned char *)name;
867
868 hash = init_name_hash();
869 do {
870 name++;
871 hash = partial_name_hash(c, hash);
872 c = *(const unsigned char *)name;
873 } while (c && (c != '/'));
874 this.len = name - (const char *) this.name;
875 this.hash = end_name_hash(hash);
876
877 /* remove trailing slashes? */
878 if (!c)
879 goto last_component;
880 while (*++name == '/');
881 if (!*name)
882 goto last_with_slashes;
883
884 /*
885 * "." and ".." are special - ".." especially so because it has
886 * to be able to know about the current root directory and
887 * parent relationships.
888 */
889 if (this.name[0] == '.') switch (this.len) {
890 default:
891 break;
892 case 2:
893 if (this.name[1] != '.')
894 break;
Al Viro58c465e2005-06-06 13:36:13 -0700895 follow_dotdot(nd);
Jan Blunck4ac91372008-02-14 19:34:32 -0800896 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 /* fallthrough */
898 case 1:
899 continue;
900 }
901 /*
902 * See if the low-level filesystem might want
903 * to use its own hash..
904 */
Jan Blunck4ac91372008-02-14 19:34:32 -0800905 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
906 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
907 &this);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 if (err < 0)
909 break;
910 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 /* This does the actual lookups.. */
912 err = do_lookup(nd, &this, &next);
913 if (err)
914 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915
916 err = -ENOENT;
917 inode = next.dentry->d_inode;
918 if (!inode)
919 goto out_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920
921 if (inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700922 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 if (err)
924 goto return_err;
925 err = -ENOENT;
Jan Blunck4ac91372008-02-14 19:34:32 -0800926 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 if (!inode)
928 break;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700929 } else
930 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 err = -ENOTDIR;
932 if (!inode->i_op->lookup)
933 break;
934 continue;
935 /* here ends the main loop */
936
937last_with_slashes:
938 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
939last_component:
Trond Myklebustf55eab82006-02-04 23:28:01 -0800940 /* Clear LOOKUP_CONTINUE iff it was previously unset */
941 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 if (lookup_flags & LOOKUP_PARENT)
943 goto lookup_parent;
944 if (this.name[0] == '.') switch (this.len) {
945 default:
946 break;
947 case 2:
948 if (this.name[1] != '.')
949 break;
Al Viro58c465e2005-06-06 13:36:13 -0700950 follow_dotdot(nd);
Jan Blunck4ac91372008-02-14 19:34:32 -0800951 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 /* fallthrough */
953 case 1:
954 goto return_reval;
955 }
Jan Blunck4ac91372008-02-14 19:34:32 -0800956 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
957 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
958 &this);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 if (err < 0)
960 break;
961 }
962 err = do_lookup(nd, &this, &next);
963 if (err)
964 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 inode = next.dentry->d_inode;
966 if ((lookup_flags & LOOKUP_FOLLOW)
Al Viroacfa4382008-12-04 10:06:33 -0500967 && inode && inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700968 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 if (err)
970 goto return_err;
Jan Blunck4ac91372008-02-14 19:34:32 -0800971 inode = nd->path.dentry->d_inode;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700972 } else
973 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 err = -ENOENT;
975 if (!inode)
976 break;
977 if (lookup_flags & LOOKUP_DIRECTORY) {
978 err = -ENOTDIR;
Al Viroacfa4382008-12-04 10:06:33 -0500979 if (!inode->i_op->lookup)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 break;
981 }
982 goto return_base;
983lookup_parent:
984 nd->last = this;
985 nd->last_type = LAST_NORM;
986 if (this.name[0] != '.')
987 goto return_base;
988 if (this.len == 1)
989 nd->last_type = LAST_DOT;
990 else if (this.len == 2 && this.name[1] == '.')
991 nd->last_type = LAST_DOTDOT;
992 else
993 goto return_base;
994return_reval:
995 /*
996 * We bypassed the ordinary revalidation routines.
997 * We may need to check the cached dentry for staleness.
998 */
Jan Blunck4ac91372008-02-14 19:34:32 -0800999 if (nd->path.dentry && nd->path.dentry->d_sb &&
1000 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 err = -ESTALE;
1002 /* Note: we do not d_invalidate() */
Jan Blunck4ac91372008-02-14 19:34:32 -08001003 if (!nd->path.dentry->d_op->d_revalidate(
1004 nd->path.dentry, nd))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 break;
1006 }
1007return_base:
1008 return 0;
1009out_dput:
Jan Blunck1d957f92008-02-14 19:34:35 -08001010 path_put_conditional(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 break;
1012 }
Jan Blunck1d957f92008-02-14 19:34:35 -08001013 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014return_err:
1015 return err;
1016}
1017
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -08001018static int path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019{
1020 current->total_link_count = 0;
1021 return link_path_walk(name, nd);
1022}
1023
Al Viro9b4a9b12009-04-07 11:44:16 -04001024static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001026 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001027 int fput_needed;
1028 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029
1030 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1031 nd->flags = flags;
1032 nd->depth = 0;
Al Viro2a737872009-04-07 11:49:53 -04001033 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 if (*name=='/') {
Al Viro2a737872009-04-07 11:49:53 -04001036 set_root(nd);
1037 nd->path = nd->root;
1038 path_get(&nd->root);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001039 } else if (dfd == AT_FDCWD) {
Al Viro2a737872009-04-07 11:49:53 -04001040 struct fs_struct *fs = current->fs;
Andreas Mohre518ddb2006-09-29 02:01:22 -07001041 read_lock(&fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08001042 nd->path = fs->pwd;
1043 path_get(&fs->pwd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001044 read_unlock(&fs->lock);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001045 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001046 struct dentry *dentry;
1047
1048 file = fget_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001049 retval = -EBADF;
1050 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001051 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001052
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001053 dentry = file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001054
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001055 retval = -ENOTDIR;
1056 if (!S_ISDIR(dentry->d_inode->i_mode))
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001057 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001058
1059 retval = file_permission(file, MAY_EXEC);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001060 if (retval)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001061 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001062
Jan Blunck5dd784d02008-02-14 19:34:38 -08001063 nd->path = file->f_path;
1064 path_get(&file->f_path);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001065
1066 fput_light(file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001068 return 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001069
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001070fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001071 fput_light(file, fput_needed);
Al Viro9b4a9b12009-04-07 11:44:16 -04001072out_fail:
1073 return retval;
1074}
1075
1076/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1077static int do_path_lookup(int dfd, const char *name,
1078 unsigned int flags, struct nameidata *nd)
1079{
1080 int retval = path_init(dfd, name, flags, nd);
1081 if (!retval)
1082 retval = path_walk(name, nd);
1083 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1084 nd->path.dentry->d_inode))
1085 audit_inode(name, nd->path.dentry);
Al Viro2a737872009-04-07 11:49:53 -04001086 if (nd->root.mnt) {
1087 path_put(&nd->root);
1088 nd->root.mnt = NULL;
1089 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001090 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091}
1092
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -08001093int path_lookup(const char *name, unsigned int flags,
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001094 struct nameidata *nd)
1095{
1096 return do_path_lookup(AT_FDCWD, name, flags, nd);
1097}
1098
Al Virod1811462008-08-02 00:49:18 -04001099int kern_path(const char *name, unsigned int flags, struct path *path)
1100{
1101 struct nameidata nd;
1102 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1103 if (!res)
1104 *path = nd.path;
1105 return res;
1106}
1107
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001108/**
1109 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1110 * @dentry: pointer to dentry of the base directory
1111 * @mnt: pointer to vfs mount of the base directory
1112 * @name: pointer to file name
1113 * @flags: lookup flags
1114 * @nd: pointer to nameidata
1115 */
1116int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1117 const char *name, unsigned int flags,
1118 struct nameidata *nd)
1119{
1120 int retval;
1121
1122 /* same as do_path_lookup */
1123 nd->last_type = LAST_ROOT;
1124 nd->flags = flags;
1125 nd->depth = 0;
1126
Jan Blunckc8e7f442008-06-09 16:40:35 -07001127 nd->path.dentry = dentry;
1128 nd->path.mnt = mnt;
1129 path_get(&nd->path);
Al Viro5b857112009-04-07 11:53:49 -04001130 nd->root = nd->path;
1131 path_get(&nd->root);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001132
1133 retval = path_walk(name, nd);
Jan Blunck4ac91372008-02-14 19:34:32 -08001134 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1135 nd->path.dentry->d_inode))
1136 audit_inode(name, nd->path.dentry);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001137
Al Viro5b857112009-04-07 11:53:49 -04001138 path_put(&nd->root);
1139 nd->root.mnt = NULL;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001140
Al Viro2a737872009-04-07 11:49:53 -04001141 return retval;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001142}
1143
Trond Myklebust834f2a42005-10-18 14:20:16 -07001144/**
1145 * path_lookup_open - lookup a file path with open intent
Martin Waitz7045f372006-02-01 03:06:57 -08001146 * @dfd: the directory to use as base, or AT_FDCWD
Trond Myklebust834f2a42005-10-18 14:20:16 -07001147 * @name: pointer to file name
1148 * @lookup_flags: lookup intent flags
1149 * @nd: pointer to nameidata
1150 * @open_flags: open intent flags
1151 */
Al Viro6e8341a2009-04-06 11:16:22 -04001152static int path_lookup_open(int dfd, const char *name,
1153 unsigned int lookup_flags, struct nameidata *nd, int open_flags)
Trond Myklebust834f2a42005-10-18 14:20:16 -07001154{
Al Viro8737f3a2008-08-02 22:36:57 -04001155 struct file *filp = get_empty_filp();
1156 int err;
Trond Myklebust834f2a42005-10-18 14:20:16 -07001157
Al Viro8737f3a2008-08-02 22:36:57 -04001158 if (filp == NULL)
1159 return -ENFILE;
1160 nd->intent.open.file = filp;
1161 nd->intent.open.flags = open_flags;
1162 nd->intent.open.create_mode = 0;
1163 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1164 if (IS_ERR(nd->intent.open.file)) {
1165 if (err == 0) {
1166 err = PTR_ERR(nd->intent.open.file);
1167 path_put(&nd->path);
1168 }
1169 } else if (err != 0)
1170 release_open_intent(nd);
1171 return err;
Trond Myklebust834f2a42005-10-18 14:20:16 -07001172}
1173
Christoph Hellwigeead1912007-10-16 23:25:38 -07001174static struct dentry *__lookup_hash(struct qstr *name,
1175 struct dentry *base, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176{
James Morris057f6c02007-04-26 00:12:05 -07001177 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 struct inode *inode;
1179 int err;
1180
1181 inode = base->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182
1183 /*
1184 * See if the low-level filesystem might want
1185 * to use its own hash..
1186 */
1187 if (base->d_op && base->d_op->d_hash) {
1188 err = base->d_op->d_hash(base, name);
1189 dentry = ERR_PTR(err);
1190 if (err < 0)
1191 goto out;
1192 }
1193
1194 dentry = cached_lookup(base, name, nd);
1195 if (!dentry) {
Miklos Szeredid70b67c2008-07-02 21:30:15 +02001196 struct dentry *new;
1197
1198 /* Don't create child dentry for a dead directory. */
1199 dentry = ERR_PTR(-ENOENT);
1200 if (IS_DEADDIR(inode))
1201 goto out;
1202
1203 new = d_alloc(base, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 dentry = ERR_PTR(-ENOMEM);
1205 if (!new)
1206 goto out;
1207 dentry = inode->i_op->lookup(inode, new, nd);
1208 if (!dentry)
1209 dentry = new;
1210 else
1211 dput(new);
1212 }
1213out:
1214 return dentry;
1215}
1216
James Morris057f6c02007-04-26 00:12:05 -07001217/*
1218 * Restricted form of lookup. Doesn't follow links, single-component only,
1219 * needs parent already locked. Doesn't follow mounts.
1220 * SMP-safe.
1221 */
Adrian Bunka244e162006-03-31 02:32:11 -08001222static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223{
Christoph Hellwigeead1912007-10-16 23:25:38 -07001224 int err;
1225
Al Virof419a2e2008-07-22 00:07:17 -04001226 err = inode_permission(nd->path.dentry->d_inode, MAY_EXEC);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001227 if (err)
1228 return ERR_PTR(err);
Jan Blunck4ac91372008-02-14 19:34:32 -08001229 return __lookup_hash(&nd->last, nd->path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230}
1231
Christoph Hellwigeead1912007-10-16 23:25:38 -07001232static int __lookup_one_len(const char *name, struct qstr *this,
1233 struct dentry *base, int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234{
1235 unsigned long hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 unsigned int c;
1237
James Morris057f6c02007-04-26 00:12:05 -07001238 this->name = name;
1239 this->len = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 if (!len)
James Morris057f6c02007-04-26 00:12:05 -07001241 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242
1243 hash = init_name_hash();
1244 while (len--) {
1245 c = *(const unsigned char *)name++;
1246 if (c == '/' || c == '\0')
James Morris057f6c02007-04-26 00:12:05 -07001247 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 hash = partial_name_hash(c, hash);
1249 }
James Morris057f6c02007-04-26 00:12:05 -07001250 this->hash = end_name_hash(hash);
1251 return 0;
1252}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253
Christoph Hellwigeead1912007-10-16 23:25:38 -07001254/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001255 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07001256 * @name: pathname component to lookup
1257 * @base: base directory to lookup from
1258 * @len: maximum length @len should be interpreted to
1259 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07001260 * Note that this routine is purely a helper for filesystem usage and should
1261 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07001262 * nameidata argument is passed to the filesystem methods and a filesystem
1263 * using this helper needs to be prepared for that.
1264 */
James Morris057f6c02007-04-26 00:12:05 -07001265struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1266{
1267 int err;
1268 struct qstr this;
1269
David Woodhouse2f9092e2009-04-20 23:18:37 +01001270 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1271
James Morris057f6c02007-04-26 00:12:05 -07001272 err = __lookup_one_len(name, &this, base, len);
1273 if (err)
1274 return ERR_PTR(err);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001275
Al Virof419a2e2008-07-22 00:07:17 -04001276 err = inode_permission(base->d_inode, MAY_EXEC);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001277 if (err)
1278 return ERR_PTR(err);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001279 return __lookup_hash(&this, base, NULL);
James Morris057f6c02007-04-26 00:12:05 -07001280}
1281
Christoph Hellwigeead1912007-10-16 23:25:38 -07001282/**
1283 * lookup_one_noperm - bad hack for sysfs
1284 * @name: pathname component to lookup
1285 * @base: base directory to lookup from
1286 *
1287 * This is a variant of lookup_one_len that doesn't perform any permission
1288 * checks. It's a horrible hack to work around the braindead sysfs
1289 * architecture and should not be used anywhere else.
1290 *
1291 * DON'T USE THIS FUNCTION EVER, thanks.
1292 */
1293struct dentry *lookup_one_noperm(const char *name, struct dentry *base)
James Morris057f6c02007-04-26 00:12:05 -07001294{
1295 int err;
1296 struct qstr this;
1297
Christoph Hellwigeead1912007-10-16 23:25:38 -07001298 err = __lookup_one_len(name, &this, base, strlen(name));
James Morris057f6c02007-04-26 00:12:05 -07001299 if (err)
1300 return ERR_PTR(err);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001301 return __lookup_hash(&this, base, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302}
1303
Al Viro2d8f3032008-07-22 09:59:21 -04001304int user_path_at(int dfd, const char __user *name, unsigned flags,
1305 struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306{
Al Viro2d8f3032008-07-22 09:59:21 -04001307 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 char *tmp = getname(name);
1309 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04001311
1312 BUG_ON(flags & LOOKUP_PARENT);
1313
1314 err = do_path_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04001316 if (!err)
1317 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 }
1319 return err;
1320}
1321
Al Viro2ad94ae2008-07-21 09:32:51 -04001322static int user_path_parent(int dfd, const char __user *path,
1323 struct nameidata *nd, char **name)
1324{
1325 char *s = getname(path);
1326 int error;
1327
1328 if (IS_ERR(s))
1329 return PTR_ERR(s);
1330
1331 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1332 if (error)
1333 putname(s);
1334 else
1335 *name = s;
1336
1337 return error;
1338}
1339
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340/*
1341 * It's inline, so penalty for filesystems that don't use sticky bit is
1342 * minimal.
1343 */
1344static inline int check_sticky(struct inode *dir, struct inode *inode)
1345{
David Howellsda9592e2008-11-14 10:39:05 +11001346 uid_t fsuid = current_fsuid();
1347
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 if (!(dir->i_mode & S_ISVTX))
1349 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001350 if (inode->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001352 if (dir->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 return 0;
1354 return !capable(CAP_FOWNER);
1355}
1356
1357/*
1358 * Check whether we can remove a link victim from directory dir, check
1359 * whether the type of victim is right.
1360 * 1. We can't do it if dir is read-only (done in permission())
1361 * 2. We should have write and exec permissions on dir
1362 * 3. We can't remove anything from append-only dir
1363 * 4. We can't do anything with immutable dir (done in permission())
1364 * 5. If the sticky bit on dir is set we should either
1365 * a. be owner of dir, or
1366 * b. be owner of victim, or
1367 * c. have CAP_FOWNER capability
1368 * 6. If the victim is append-only or immutable we can't do antyhing with
1369 * links pointing to it.
1370 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1371 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1372 * 9. We can't remove a root or mountpoint.
1373 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1374 * nfs_async_unlink().
1375 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001376static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377{
1378 int error;
1379
1380 if (!victim->d_inode)
1381 return -ENOENT;
1382
1383 BUG_ON(victim->d_parent->d_inode != dir);
Al Viro5a190ae2007-06-07 12:19:32 -04001384 audit_inode_child(victim->d_name.name, victim, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385
Al Virof419a2e2008-07-22 00:07:17 -04001386 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 if (error)
1388 return error;
1389 if (IS_APPEND(dir))
1390 return -EPERM;
1391 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
Hugh Dickinsf9454542008-11-19 15:36:38 -08001392 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 return -EPERM;
1394 if (isdir) {
1395 if (!S_ISDIR(victim->d_inode->i_mode))
1396 return -ENOTDIR;
1397 if (IS_ROOT(victim))
1398 return -EBUSY;
1399 } else if (S_ISDIR(victim->d_inode->i_mode))
1400 return -EISDIR;
1401 if (IS_DEADDIR(dir))
1402 return -ENOENT;
1403 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1404 return -EBUSY;
1405 return 0;
1406}
1407
1408/* Check whether we can create an object with dentry child in directory
1409 * dir.
1410 * 1. We can't do it if child already exists (open has special treatment for
1411 * this case, but since we are inlined it's OK)
1412 * 2. We can't do it if dir is read-only (done in permission())
1413 * 3. We should have write and exec permissions on dir
1414 * 4. We can't do it if dir is immutable (done in permission())
1415 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02001416static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417{
1418 if (child->d_inode)
1419 return -EEXIST;
1420 if (IS_DEADDIR(dir))
1421 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04001422 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423}
1424
1425/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 * O_DIRECTORY translates into forcing a directory lookup.
1427 */
1428static inline int lookup_flags(unsigned int f)
1429{
1430 unsigned long retval = LOOKUP_FOLLOW;
1431
1432 if (f & O_NOFOLLOW)
1433 retval &= ~LOOKUP_FOLLOW;
1434
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 if (f & O_DIRECTORY)
1436 retval |= LOOKUP_DIRECTORY;
1437
1438 return retval;
1439}
1440
1441/*
1442 * p1 and p2 should be directories on the same fs.
1443 */
1444struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1445{
1446 struct dentry *p;
1447
1448 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001449 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 return NULL;
1451 }
1452
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001453 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001455 p = d_ancestor(p2, p1);
1456 if (p) {
1457 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1458 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1459 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 }
1461
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001462 p = d_ancestor(p1, p2);
1463 if (p) {
1464 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1465 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1466 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 }
1468
Ingo Molnarf2eace22006-07-03 00:25:05 -07001469 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1470 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 return NULL;
1472}
1473
1474void unlock_rename(struct dentry *p1, struct dentry *p2)
1475{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001476 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001478 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001479 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 }
1481}
1482
1483int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1484 struct nameidata *nd)
1485{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001486 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487
1488 if (error)
1489 return error;
1490
Al Viroacfa4382008-12-04 10:06:33 -05001491 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 return -EACCES; /* shouldn't it be ENOSYS? */
1493 mode &= S_IALLUGO;
1494 mode |= S_IFREG;
1495 error = security_inode_create(dir, dentry, mode);
1496 if (error)
1497 return error;
Jan Kara9e3509e2009-01-26 16:45:12 +01001498 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001500 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001501 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 return error;
1503}
1504
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001505int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001507 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 struct inode *inode = dentry->d_inode;
1509 int error;
1510
1511 if (!inode)
1512 return -ENOENT;
1513
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001514 switch (inode->i_mode & S_IFMT) {
1515 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001517 case S_IFDIR:
1518 if (acc_mode & MAY_WRITE)
1519 return -EISDIR;
1520 break;
1521 case S_IFBLK:
1522 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001523 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001525 /*FALLTHRU*/
1526 case S_IFIFO:
1527 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001529 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001530 }
Dave Hansenb41572e2007-10-16 23:31:14 -07001531
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001532 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07001533 if (error)
1534 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05001535
Mimi Zohar6c1488f2009-09-02 11:40:32 -04001536 error = ima_path_check(path, acc_mode ?
1537 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC) :
1538 ACC_MODE(flag) & (MAY_READ | MAY_WRITE),
Mimi Zoharb9fc7452009-05-19 13:25:57 -04001539 IMA_COUNT_UPDATE);
Mimi Zohar6c1488f2009-09-02 11:40:32 -04001540
Mimi Zohar6146f0d2009-02-04 09:06:57 -05001541 if (error)
1542 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 /*
1544 * An append-only file must be opened in append mode for writing.
1545 */
1546 if (IS_APPEND(inode)) {
Mimi Zoharacd0c932009-09-04 13:08:46 -04001547 error = -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
Mimi Zoharacd0c932009-09-04 13:08:46 -04001549 goto err_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 if (flag & O_TRUNC)
Mimi Zoharacd0c932009-09-04 13:08:46 -04001551 goto err_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 }
1553
1554 /* O_NOATIME can only be set by the owner or superuser */
1555 if (flag & O_NOATIME)
Mimi Zoharacd0c932009-09-04 13:08:46 -04001556 if (!is_owner_or_cap(inode)) {
1557 error = -EPERM;
1558 goto err_out;
1559 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560
1561 /*
1562 * Ensure there are no outstanding leases on the file.
1563 */
1564 error = break_lease(inode, flag);
1565 if (error)
Mimi Zoharacd0c932009-09-04 13:08:46 -04001566 goto err_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567
1568 if (flag & O_TRUNC) {
1569 error = get_write_access(inode);
1570 if (error)
Mimi Zoharacd0c932009-09-04 13:08:46 -04001571 goto err_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572
1573 /*
1574 * Refuse to truncate files with mandatory locks held on them.
1575 */
1576 error = locks_verify_locked(inode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001577 if (!error)
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001578 error = security_path_truncate(path, 0,
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001579 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 if (!error) {
Jan Kara9e3509e2009-01-26 16:45:12 +01001581 vfs_dq_init(inode);
Miklos Szeredid139d7f2007-10-18 03:07:00 -07001582
1583 error = do_truncate(dentry, 0,
1584 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1585 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 }
1587 put_write_access(inode);
1588 if (error)
Mimi Zoharacd0c932009-09-04 13:08:46 -04001589 goto err_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 } else
1591 if (flag & FMODE_WRITE)
Jan Kara9e3509e2009-01-26 16:45:12 +01001592 vfs_dq_init(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593
1594 return 0;
Mimi Zoharacd0c932009-09-04 13:08:46 -04001595err_out:
1596 ima_counts_put(path, acc_mode ?
1597 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC) :
1598 ACC_MODE(flag) & (MAY_READ | MAY_WRITE));
1599 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600}
1601
Dave Hansend57999e2008-02-15 14:37:27 -08001602/*
1603 * Be careful about ever adding any more callers of this
1604 * function. Its flags must be in the namei format, not
1605 * what get passed to sys_open().
1606 */
1607static int __open_namei_create(struct nameidata *nd, struct path *path,
Dave Hansenaab520e2006-09-30 23:29:02 -07001608 int flag, int mode)
1609{
1610 int error;
Jan Blunck4ac91372008-02-14 19:34:32 -08001611 struct dentry *dir = nd->path.dentry;
Dave Hansenaab520e2006-09-30 23:29:02 -07001612
1613 if (!IS_POSIXACL(dir->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04001614 mode &= ~current_umask();
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001615 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1616 if (error)
1617 goto out_unlock;
Dave Hansenaab520e2006-09-30 23:29:02 -07001618 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001619out_unlock:
Dave Hansenaab520e2006-09-30 23:29:02 -07001620 mutex_unlock(&dir->d_inode->i_mutex);
Jan Blunck4ac91372008-02-14 19:34:32 -08001621 dput(nd->path.dentry);
1622 nd->path.dentry = path->dentry;
Dave Hansenaab520e2006-09-30 23:29:02 -07001623 if (error)
1624 return error;
1625 /* Don't check for write permission, don't truncate */
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001626 return may_open(&nd->path, 0, flag & ~O_TRUNC);
Dave Hansenaab520e2006-09-30 23:29:02 -07001627}
1628
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629/*
Dave Hansend57999e2008-02-15 14:37:27 -08001630 * Note that while the flag value (low two bits) for sys_open means:
1631 * 00 - read-only
1632 * 01 - write-only
1633 * 10 - read-write
1634 * 11 - special
1635 * it is changed into
1636 * 00 - no permissions needed
1637 * 01 - read-permission
1638 * 10 - write-permission
1639 * 11 - read-write
1640 * for the internal routines (ie open_namei()/follow_link() etc)
1641 * This is more logical, and also allows the 00 "no perm needed"
1642 * to be used for symlinks (where the permissions are checked
1643 * later).
1644 *
1645*/
1646static inline int open_to_namei_flags(int flag)
1647{
1648 if ((flag+1) & O_ACCMODE)
1649 flag++;
1650 return flag;
1651}
1652
Dave Hansen4a3fd212008-02-15 14:37:48 -08001653static int open_will_write_to_fs(int flag, struct inode *inode)
1654{
1655 /*
1656 * We'll never write to the fs underlying
1657 * a device file.
1658 */
1659 if (special_file(inode->i_mode))
1660 return 0;
1661 return (flag & O_TRUNC);
1662}
1663
Dave Hansend57999e2008-02-15 14:37:27 -08001664/*
Dave Hansen4a3fd212008-02-15 14:37:48 -08001665 * Note that the low bits of the passed in "open_flag"
1666 * are not the same as in the local variable "flag". See
1667 * open_to_namei_flags() for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001669struct file *do_filp_open(int dfd, const char *pathname,
Al Viro6e8341a2009-04-06 11:16:22 -04001670 int open_flag, int mode, int acc_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671{
Dave Hansen4a3fd212008-02-15 14:37:48 -08001672 struct file *filp;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001673 struct nameidata nd;
Al Viro6e8341a2009-04-06 11:16:22 -04001674 int error;
Al Viro4e7506e2005-06-06 13:36:00 -07001675 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 struct dentry *dir;
1677 int count = 0;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001678 int will_write;
Dave Hansend57999e2008-02-15 14:37:27 -08001679 int flag = open_to_namei_flags(open_flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680
Al Viro6e8341a2009-04-06 11:16:22 -04001681 if (!acc_mode)
1682 acc_mode = MAY_OPEN | ACC_MODE(flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683
Trond Myklebust834f2a42005-10-18 14:20:16 -07001684 /* O_TRUNC implies we need access checks for write permissions */
1685 if (flag & O_TRUNC)
1686 acc_mode |= MAY_WRITE;
1687
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688 /* Allow the LSM permission hook to distinguish append
1689 access from general write access. */
1690 if (flag & O_APPEND)
1691 acc_mode |= MAY_APPEND;
1692
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 /*
1694 * The simplest case - just a plain lookup.
1695 */
1696 if (!(flag & O_CREAT)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001697 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001698 &nd, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 if (error)
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001700 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 goto ok;
1702 }
1703
1704 /*
1705 * Create - we need to know the parent.
1706 */
Al Viro9b4a9b12009-04-07 11:44:16 -04001707 error = path_init(dfd, pathname, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 if (error)
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001709 return ERR_PTR(error);
Al Viro9b4a9b12009-04-07 11:44:16 -04001710 error = path_walk(pathname, &nd);
J. R. Okajima654f5622009-06-18 23:30:15 +09001711 if (error) {
1712 if (nd.root.mnt)
1713 path_put(&nd.root);
Al Viro9b4a9b12009-04-07 11:44:16 -04001714 return ERR_PTR(error);
J. R. Okajima654f5622009-06-18 23:30:15 +09001715 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001716 if (unlikely(!audit_dummy_context()))
1717 audit_inode(pathname, nd.path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718
1719 /*
1720 * We have the parent and last component. First of all, check
1721 * that we are not asked to creat(2) an obvious directory - that
1722 * will not do.
1723 */
1724 error = -EISDIR;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001725 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
Al Viro8737f3a2008-08-02 22:36:57 -04001726 goto exit_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727
Al Viro8737f3a2008-08-02 22:36:57 -04001728 error = -ENFILE;
1729 filp = get_empty_filp();
1730 if (filp == NULL)
1731 goto exit_parent;
1732 nd.intent.open.file = filp;
1733 nd.intent.open.flags = flag;
1734 nd.intent.open.create_mode = mode;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001735 dir = nd.path.dentry;
1736 nd.flags &= ~LOOKUP_PARENT;
Al Viro8737f3a2008-08-02 22:36:57 -04001737 nd.flags |= LOOKUP_CREATE | LOOKUP_OPEN;
Al Viro35165862008-08-05 03:00:49 -04001738 if (flag & O_EXCL)
1739 nd.flags |= LOOKUP_EXCL;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001740 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001741 path.dentry = lookup_hash(&nd);
1742 path.mnt = nd.path.mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743
1744do_last:
Al Viro4e7506e2005-06-06 13:36:00 -07001745 error = PTR_ERR(path.dentry);
1746 if (IS_ERR(path.dentry)) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001747 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 goto exit;
1749 }
1750
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001751 if (IS_ERR(nd.intent.open.file)) {
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001752 error = PTR_ERR(nd.intent.open.file);
Dave Hansen4a3fd212008-02-15 14:37:48 -08001753 goto exit_mutex_unlock;
Oleg Drokin4af4c522006-03-25 03:06:54 -08001754 }
1755
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 /* Negative dentry, just create the file */
Al Viro4e7506e2005-06-06 13:36:00 -07001757 if (!path.dentry->d_inode) {
Dave Hansen4a3fd212008-02-15 14:37:48 -08001758 /*
1759 * This write is needed to ensure that a
1760 * ro->rw transition does not occur between
1761 * the time when the file is created and when
1762 * a permanent write count is taken through
1763 * the 'struct file' in nameidata_to_filp().
1764 */
1765 error = mnt_want_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 if (error)
Dave Hansen4a3fd212008-02-15 14:37:48 -08001767 goto exit_mutex_unlock;
1768 error = __open_namei_create(&nd, &path, flag, mode);
1769 if (error) {
1770 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771 goto exit;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001772 }
1773 filp = nameidata_to_filp(&nd, open_flag);
Mimi Zohar94e5d712009-06-26 14:05:27 -04001774 if (IS_ERR(filp))
1775 ima_counts_put(&nd.path,
1776 acc_mode & (MAY_READ | MAY_WRITE |
1777 MAY_EXEC));
Dave Hansen4a3fd212008-02-15 14:37:48 -08001778 mnt_drop_write(nd.path.mnt);
J. R. Okajima654f5622009-06-18 23:30:15 +09001779 if (nd.root.mnt)
1780 path_put(&nd.root);
Dave Hansen4a3fd212008-02-15 14:37:48 -08001781 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 }
1783
1784 /*
1785 * It already exists.
1786 */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001787 mutex_unlock(&dir->d_inode->i_mutex);
Al Viro5a190ae2007-06-07 12:19:32 -04001788 audit_inode(pathname, path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789
1790 error = -EEXIST;
1791 if (flag & O_EXCL)
1792 goto exit_dput;
1793
Al Viroe13b2102005-06-06 13:36:06 -07001794 if (__follow_mount(&path)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 error = -ELOOP;
Al Viroba7a4c12005-06-06 13:36:08 -07001796 if (flag & O_NOFOLLOW)
1797 goto exit_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 }
Amy Griffis3e2efce2006-07-13 13:16:02 -04001799
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800 error = -ENOENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001801 if (!path.dentry->d_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 goto exit_dput;
Al Viroacfa4382008-12-04 10:06:33 -05001803 if (path.dentry->d_inode->i_op->follow_link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 goto do_link;
1805
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001806 path_to_nameidata(&path, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807 error = -EISDIR;
Al Viro4e7506e2005-06-06 13:36:00 -07001808 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 goto exit;
1810ok:
Dave Hansen4a3fd212008-02-15 14:37:48 -08001811 /*
1812 * Consider:
1813 * 1. may_open() truncates a file
1814 * 2. a rw->ro mount transition occurs
1815 * 3. nameidata_to_filp() fails due to
1816 * the ro mount.
1817 * That would be inconsistent, and should
1818 * be avoided. Taking this mnt write here
1819 * ensures that (2) can not occur.
1820 */
1821 will_write = open_will_write_to_fs(flag, nd.path.dentry->d_inode);
1822 if (will_write) {
1823 error = mnt_want_write(nd.path.mnt);
1824 if (error)
1825 goto exit;
1826 }
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001827 error = may_open(&nd.path, acc_mode, flag);
Dave Hansen4a3fd212008-02-15 14:37:48 -08001828 if (error) {
1829 if (will_write)
1830 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 goto exit;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001832 }
1833 filp = nameidata_to_filp(&nd, open_flag);
Mimi Zohar94e5d712009-06-26 14:05:27 -04001834 if (IS_ERR(filp))
1835 ima_counts_put(&nd.path,
1836 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC));
Dave Hansen4a3fd212008-02-15 14:37:48 -08001837 /*
1838 * It is now safe to drop the mnt write
1839 * because the filp has had a write taken
1840 * on its behalf.
1841 */
1842 if (will_write)
1843 mnt_drop_write(nd.path.mnt);
J. R. Okajima654f5622009-06-18 23:30:15 +09001844 if (nd.root.mnt)
1845 path_put(&nd.root);
Dave Hansen4a3fd212008-02-15 14:37:48 -08001846 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847
Dave Hansen4a3fd212008-02-15 14:37:48 -08001848exit_mutex_unlock:
1849 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850exit_dput:
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001851 path_put_conditional(&path, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852exit:
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001853 if (!IS_ERR(nd.intent.open.file))
1854 release_open_intent(&nd);
Al Viro8737f3a2008-08-02 22:36:57 -04001855exit_parent:
Al Viro2a737872009-04-07 11:49:53 -04001856 if (nd.root.mnt)
1857 path_put(&nd.root);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001858 path_put(&nd.path);
1859 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860
1861do_link:
1862 error = -ELOOP;
1863 if (flag & O_NOFOLLOW)
1864 goto exit_dput;
1865 /*
1866 * This is subtle. Instead of calling do_follow_link() we do the
1867 * thing by hands. The reason is that this way we have zero link_count
1868 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1869 * After that we have the parent and last component, i.e.
1870 * we are in the same situation as after the first path_walk().
1871 * Well, almost - if the last component is normal we get its copy
1872 * stored in nd->last.name and we will have to putname() it when we
1873 * are done. Procfs-like symlinks just set LAST_BIND.
1874 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001875 nd.flags |= LOOKUP_PARENT;
1876 error = security_inode_follow_link(path.dentry, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 if (error)
1878 goto exit_dput;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001879 error = __do_follow_link(&path, &nd);
Kirill Korotaevde459212006-07-14 00:23:49 -07001880 if (error) {
1881 /* Does someone understand code flow here? Or it is only
1882 * me so stupid? Anathema to whoever designed this non-sense
1883 * with "intent.open".
1884 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001885 release_open_intent(&nd);
J. R. Okajima654f5622009-06-18 23:30:15 +09001886 if (nd.root.mnt)
1887 path_put(&nd.root);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001888 return ERR_PTR(error);
Kirill Korotaevde459212006-07-14 00:23:49 -07001889 }
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001890 nd.flags &= ~LOOKUP_PARENT;
1891 if (nd.last_type == LAST_BIND)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 goto ok;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893 error = -EISDIR;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001894 if (nd.last_type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895 goto exit;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001896 if (nd.last.name[nd.last.len]) {
1897 __putname(nd.last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898 goto exit;
1899 }
1900 error = -ELOOP;
1901 if (count++==32) {
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001902 __putname(nd.last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 goto exit;
1904 }
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001905 dir = nd.path.dentry;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001906 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001907 path.dentry = lookup_hash(&nd);
1908 path.mnt = nd.path.mnt;
1909 __putname(nd.last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910 goto do_last;
1911}
1912
1913/**
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001914 * filp_open - open file and return file pointer
1915 *
1916 * @filename: path to open
1917 * @flags: open flags as per the open(2) second argument
1918 * @mode: mode for the new file if O_CREAT is set, else ignored
1919 *
1920 * This is the helper to open a file from kernelspace if you really
1921 * have to. But in generally you should not do this, so please move
1922 * along, nothing to see here..
1923 */
1924struct file *filp_open(const char *filename, int flags, int mode)
1925{
Al Viro6e8341a2009-04-06 11:16:22 -04001926 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001927}
1928EXPORT_SYMBOL(filp_open);
1929
1930/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 * lookup_create - lookup a dentry, creating it if it doesn't exist
1932 * @nd: nameidata info
1933 * @is_dir: directory flag
1934 *
1935 * Simple function to lookup and return a dentry and create it
1936 * if it doesn't exist. Is SMP-safe.
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001937 *
Jan Blunck4ac91372008-02-14 19:34:32 -08001938 * Returns with nd->path.dentry->d_inode->i_mutex locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 */
1940struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1941{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001942 struct dentry *dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943
Jan Blunck4ac91372008-02-14 19:34:32 -08001944 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001945 /*
1946 * Yucky last component or no last component at all?
1947 * (foo/., foo/.., /////)
1948 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 if (nd->last_type != LAST_NORM)
1950 goto fail;
1951 nd->flags &= ~LOOKUP_PARENT;
Al Viro35165862008-08-05 03:00:49 -04001952 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
ASANO Masahiroa6349042006-08-22 20:06:02 -04001953 nd->intent.open.flags = O_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001954
1955 /*
1956 * Do the final lookup.
1957 */
Christoph Hellwig49705b72005-11-08 21:35:06 -08001958 dentry = lookup_hash(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959 if (IS_ERR(dentry))
1960 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001961
Al Viroe9baf6e2008-05-15 04:49:12 -04001962 if (dentry->d_inode)
1963 goto eexist;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001964 /*
1965 * Special case - lookup gave negative, but... we had foo/bar/
1966 * From the vfs_mknod() POV we just have a negative dentry -
1967 * all is fine. Let's be bastards - you had / on the end, you've
1968 * been asking for (non-existent) directory. -ENOENT for you.
1969 */
Al Viroe9baf6e2008-05-15 04:49:12 -04001970 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1971 dput(dentry);
1972 dentry = ERR_PTR(-ENOENT);
1973 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 return dentry;
Al Viroe9baf6e2008-05-15 04:49:12 -04001975eexist:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 dput(dentry);
Al Viroe9baf6e2008-05-15 04:49:12 -04001977 dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978fail:
1979 return dentry;
1980}
Christoph Hellwigf81a0bf2005-05-19 12:26:43 -07001981EXPORT_SYMBOL_GPL(lookup_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982
1983int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1984{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001985 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986
1987 if (error)
1988 return error;
1989
1990 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1991 return -EPERM;
1992
Al Viroacfa4382008-12-04 10:06:33 -05001993 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 return -EPERM;
1995
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07001996 error = devcgroup_inode_mknod(mode, dev);
1997 if (error)
1998 return error;
1999
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 error = security_inode_mknod(dir, dentry, mode, dev);
2001 if (error)
2002 return error;
2003
Jan Kara9e3509e2009-01-26 16:45:12 +01002004 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002006 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002007 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008 return error;
2009}
2010
Dave Hansen463c3192008-02-15 14:37:57 -08002011static int may_mknod(mode_t mode)
2012{
2013 switch (mode & S_IFMT) {
2014 case S_IFREG:
2015 case S_IFCHR:
2016 case S_IFBLK:
2017 case S_IFIFO:
2018 case S_IFSOCK:
2019 case 0: /* zero mode translates to S_IFREG */
2020 return 0;
2021 case S_IFDIR:
2022 return -EPERM;
2023 default:
2024 return -EINVAL;
2025 }
2026}
2027
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002028SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2029 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030{
Al Viro2ad94ae2008-07-21 09:32:51 -04002031 int error;
2032 char *tmp;
2033 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 struct nameidata nd;
2035
2036 if (S_ISDIR(mode))
2037 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038
Al Viro2ad94ae2008-07-21 09:32:51 -04002039 error = user_path_parent(dfd, filename, &nd, &tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002041 return error;
2042
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 dentry = lookup_create(&nd, 0);
Dave Hansen463c3192008-02-15 14:37:57 -08002044 if (IS_ERR(dentry)) {
2045 error = PTR_ERR(dentry);
2046 goto out_unlock;
2047 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002048 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002049 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002050 error = may_mknod(mode);
2051 if (error)
2052 goto out_dput;
2053 error = mnt_want_write(nd.path.mnt);
2054 if (error)
2055 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002056 error = security_path_mknod(&nd.path, dentry, mode, dev);
2057 if (error)
2058 goto out_drop_write;
Dave Hansen463c3192008-02-15 14:37:57 -08002059 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060 case 0: case S_IFREG:
Jan Blunck4ac91372008-02-14 19:34:32 -08002061 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 break;
2063 case S_IFCHR: case S_IFBLK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002064 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 new_decode_dev(dev));
2066 break;
2067 case S_IFIFO: case S_IFSOCK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002068 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002071out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002072 mnt_drop_write(nd.path.mnt);
2073out_dput:
2074 dput(dentry);
2075out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002076 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002077 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 putname(tmp);
2079
2080 return error;
2081}
2082
Heiko Carstens3480b252009-01-14 14:14:16 +01002083SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002084{
2085 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2086}
2087
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2089{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002090 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091
2092 if (error)
2093 return error;
2094
Al Viroacfa4382008-12-04 10:06:33 -05002095 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096 return -EPERM;
2097
2098 mode &= (S_IRWXUGO|S_ISVTX);
2099 error = security_inode_mkdir(dir, dentry, mode);
2100 if (error)
2101 return error;
2102
Jan Kara9e3509e2009-01-26 16:45:12 +01002103 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002105 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002106 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107 return error;
2108}
2109
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002110SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111{
2112 int error = 0;
2113 char * tmp;
Dave Hansen6902d922006-09-30 23:29:01 -07002114 struct dentry *dentry;
2115 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116
Al Viro2ad94ae2008-07-21 09:32:51 -04002117 error = user_path_parent(dfd, pathname, &nd, &tmp);
2118 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002119 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120
Dave Hansen6902d922006-09-30 23:29:01 -07002121 dentry = lookup_create(&nd, 1);
2122 error = PTR_ERR(dentry);
2123 if (IS_ERR(dentry))
2124 goto out_unlock;
2125
Jan Blunck4ac91372008-02-14 19:34:32 -08002126 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002127 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002128 error = mnt_want_write(nd.path.mnt);
2129 if (error)
2130 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002131 error = security_path_mkdir(&nd.path, dentry, mode);
2132 if (error)
2133 goto out_drop_write;
Jan Blunck4ac91372008-02-14 19:34:32 -08002134 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002135out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002136 mnt_drop_write(nd.path.mnt);
2137out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002138 dput(dentry);
2139out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002140 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002141 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002142 putname(tmp);
2143out_err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 return error;
2145}
2146
Heiko Carstens3cdad422009-01-14 14:14:22 +01002147SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002148{
2149 return sys_mkdirat(AT_FDCWD, pathname, mode);
2150}
2151
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152/*
2153 * We try to drop the dentry early: we should have
2154 * a usage count of 2 if we're the only user of this
2155 * dentry, and if that is true (possibly after pruning
2156 * the dcache), then we drop the dentry now.
2157 *
2158 * A low-level filesystem can, if it choses, legally
2159 * do a
2160 *
2161 * if (!d_unhashed(dentry))
2162 * return -EBUSY;
2163 *
2164 * if it cannot handle the case of removing a directory
2165 * that is still in use by something else..
2166 */
2167void dentry_unhash(struct dentry *dentry)
2168{
2169 dget(dentry);
Vasily Averindc168422006-12-06 20:37:07 -08002170 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 spin_lock(&dcache_lock);
2172 spin_lock(&dentry->d_lock);
2173 if (atomic_read(&dentry->d_count) == 2)
2174 __d_drop(dentry);
2175 spin_unlock(&dentry->d_lock);
2176 spin_unlock(&dcache_lock);
2177}
2178
2179int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2180{
2181 int error = may_delete(dir, dentry, 1);
2182
2183 if (error)
2184 return error;
2185
Al Viroacfa4382008-12-04 10:06:33 -05002186 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187 return -EPERM;
2188
Jan Kara9e3509e2009-01-26 16:45:12 +01002189 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002191 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 dentry_unhash(dentry);
2193 if (d_mountpoint(dentry))
2194 error = -EBUSY;
2195 else {
2196 error = security_inode_rmdir(dir, dentry);
2197 if (!error) {
2198 error = dir->i_op->rmdir(dir, dentry);
2199 if (!error)
2200 dentry->d_inode->i_flags |= S_DEAD;
2201 }
2202 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002203 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 d_delete(dentry);
2206 }
2207 dput(dentry);
2208
2209 return error;
2210}
2211
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002212static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213{
2214 int error = 0;
2215 char * name;
2216 struct dentry *dentry;
2217 struct nameidata nd;
2218
Al Viro2ad94ae2008-07-21 09:32:51 -04002219 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002221 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222
2223 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002224 case LAST_DOTDOT:
2225 error = -ENOTEMPTY;
2226 goto exit1;
2227 case LAST_DOT:
2228 error = -EINVAL;
2229 goto exit1;
2230 case LAST_ROOT:
2231 error = -EBUSY;
2232 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002234
2235 nd.flags &= ~LOOKUP_PARENT;
2236
Jan Blunck4ac91372008-02-14 19:34:32 -08002237 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002238 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002240 if (IS_ERR(dentry))
2241 goto exit2;
Dave Hansen06227532008-02-15 14:37:34 -08002242 error = mnt_want_write(nd.path.mnt);
2243 if (error)
2244 goto exit3;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002245 error = security_path_rmdir(&nd.path, dentry);
2246 if (error)
2247 goto exit4;
Jan Blunck4ac91372008-02-14 19:34:32 -08002248 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002249exit4:
Dave Hansen06227532008-02-15 14:37:34 -08002250 mnt_drop_write(nd.path.mnt);
2251exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07002252 dput(dentry);
2253exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08002254 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002256 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257 putname(name);
2258 return error;
2259}
2260
Heiko Carstens3cdad422009-01-14 14:14:22 +01002261SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002262{
2263 return do_rmdir(AT_FDCWD, pathname);
2264}
2265
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266int vfs_unlink(struct inode *dir, struct dentry *dentry)
2267{
2268 int error = may_delete(dir, dentry, 0);
2269
2270 if (error)
2271 return error;
2272
Al Viroacfa4382008-12-04 10:06:33 -05002273 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274 return -EPERM;
2275
Jan Kara9e3509e2009-01-26 16:45:12 +01002276 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002278 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279 if (d_mountpoint(dentry))
2280 error = -EBUSY;
2281 else {
2282 error = security_inode_unlink(dir, dentry);
2283 if (!error)
2284 error = dir->i_op->unlink(dir, dentry);
2285 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002286 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287
2288 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2289 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08002290 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 }
Robert Love0eeca282005-07-12 17:06:03 -04002293
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 return error;
2295}
2296
2297/*
2298 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002299 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 * writeout happening, and we don't want to prevent access to the directory
2301 * while waiting on the I/O.
2302 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002303static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304{
Al Viro2ad94ae2008-07-21 09:32:51 -04002305 int error;
2306 char *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307 struct dentry *dentry;
2308 struct nameidata nd;
2309 struct inode *inode = NULL;
2310
Al Viro2ad94ae2008-07-21 09:32:51 -04002311 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002313 return error;
2314
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 error = -EISDIR;
2316 if (nd.last_type != LAST_NORM)
2317 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002318
2319 nd.flags &= ~LOOKUP_PARENT;
2320
Jan Blunck4ac91372008-02-14 19:34:32 -08002321 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002322 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323 error = PTR_ERR(dentry);
2324 if (!IS_ERR(dentry)) {
2325 /* Why not before? Because we want correct error value */
2326 if (nd.last.name[nd.last.len])
2327 goto slashes;
2328 inode = dentry->d_inode;
2329 if (inode)
2330 atomic_inc(&inode->i_count);
Dave Hansen06227532008-02-15 14:37:34 -08002331 error = mnt_want_write(nd.path.mnt);
2332 if (error)
2333 goto exit2;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002334 error = security_path_unlink(&nd.path, dentry);
2335 if (error)
2336 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08002337 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002338exit3:
Dave Hansen06227532008-02-15 14:37:34 -08002339 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340 exit2:
2341 dput(dentry);
2342 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002343 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344 if (inode)
2345 iput(inode); /* truncate the inode here */
2346exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002347 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348 putname(name);
2349 return error;
2350
2351slashes:
2352 error = !dentry->d_inode ? -ENOENT :
2353 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2354 goto exit2;
2355}
2356
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002357SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002358{
2359 if ((flag & ~AT_REMOVEDIR) != 0)
2360 return -EINVAL;
2361
2362 if (flag & AT_REMOVEDIR)
2363 return do_rmdir(dfd, pathname);
2364
2365 return do_unlinkat(dfd, pathname);
2366}
2367
Heiko Carstens3480b252009-01-14 14:14:16 +01002368SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002369{
2370 return do_unlinkat(AT_FDCWD, pathname);
2371}
2372
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002373int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002375 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376
2377 if (error)
2378 return error;
2379
Al Viroacfa4382008-12-04 10:06:33 -05002380 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381 return -EPERM;
2382
2383 error = security_inode_symlink(dir, dentry, oldname);
2384 if (error)
2385 return error;
2386
Jan Kara9e3509e2009-01-26 16:45:12 +01002387 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002389 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002390 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391 return error;
2392}
2393
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002394SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2395 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396{
Al Viro2ad94ae2008-07-21 09:32:51 -04002397 int error;
2398 char *from;
2399 char *to;
Dave Hansen6902d922006-09-30 23:29:01 -07002400 struct dentry *dentry;
2401 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402
2403 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04002404 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405 return PTR_ERR(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002406
2407 error = user_path_parent(newdfd, newname, &nd, &to);
2408 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002409 goto out_putname;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410
Dave Hansen6902d922006-09-30 23:29:01 -07002411 dentry = lookup_create(&nd, 0);
2412 error = PTR_ERR(dentry);
2413 if (IS_ERR(dentry))
2414 goto out_unlock;
2415
Dave Hansen75c3f292008-02-15 14:37:45 -08002416 error = mnt_want_write(nd.path.mnt);
2417 if (error)
2418 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002419 error = security_path_symlink(&nd.path, dentry, from);
2420 if (error)
2421 goto out_drop_write;
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002422 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002423out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002424 mnt_drop_write(nd.path.mnt);
2425out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002426 dput(dentry);
2427out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002428 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002429 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002430 putname(to);
2431out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002432 putname(from);
2433 return error;
2434}
2435
Heiko Carstens3480b252009-01-14 14:14:16 +01002436SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002437{
2438 return sys_symlinkat(oldname, AT_FDCWD, newname);
2439}
2440
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2442{
2443 struct inode *inode = old_dentry->d_inode;
2444 int error;
2445
2446 if (!inode)
2447 return -ENOENT;
2448
Miklos Szeredia95164d2008-07-30 15:08:48 +02002449 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450 if (error)
2451 return error;
2452
2453 if (dir->i_sb != inode->i_sb)
2454 return -EXDEV;
2455
2456 /*
2457 * A link to an append-only or immutable file cannot be created.
2458 */
2459 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2460 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05002461 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002463 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464 return -EPERM;
2465
2466 error = security_inode_link(old_dentry, dir, new_dentry);
2467 if (error)
2468 return error;
2469
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002470 mutex_lock(&inode->i_mutex);
Jan Kara9e3509e2009-01-26 16:45:12 +01002471 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472 error = dir->i_op->link(old_dentry, dir, new_dentry);
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002473 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002474 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002475 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476 return error;
2477}
2478
2479/*
2480 * Hardlinks are often used in delicate situations. We avoid
2481 * security-related surprises by not following symlinks on the
2482 * newname. --KAB
2483 *
2484 * We don't follow them on the oldname either to be compatible
2485 * with linux 2.0, and to avoid hard-linking to directories
2486 * and other special files. --ADM
2487 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002488SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2489 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490{
2491 struct dentry *new_dentry;
Al Viro2d8f3032008-07-22 09:59:21 -04002492 struct nameidata nd;
2493 struct path old_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494 int error;
Al Viro2ad94ae2008-07-21 09:32:51 -04002495 char *to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496
Ulrich Drepper45c9b112006-06-25 05:49:11 -07002497 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002498 return -EINVAL;
2499
Al Viro2d8f3032008-07-22 09:59:21 -04002500 error = user_path_at(olddfd, oldname,
2501 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2502 &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002504 return error;
2505
2506 error = user_path_parent(newdfd, newname, &nd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507 if (error)
2508 goto out;
2509 error = -EXDEV;
Al Viro2d8f3032008-07-22 09:59:21 -04002510 if (old_path.mnt != nd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511 goto out_release;
2512 new_dentry = lookup_create(&nd, 0);
2513 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002514 if (IS_ERR(new_dentry))
2515 goto out_unlock;
Dave Hansen75c3f292008-02-15 14:37:45 -08002516 error = mnt_want_write(nd.path.mnt);
2517 if (error)
2518 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002519 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2520 if (error)
2521 goto out_drop_write;
Al Viro2d8f3032008-07-22 09:59:21 -04002522 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002523out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002524 mnt_drop_write(nd.path.mnt);
2525out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002526 dput(new_dentry);
2527out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002528 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529out_release:
Jan Blunck1d957f92008-02-14 19:34:35 -08002530 path_put(&nd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002531 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532out:
Al Viro2d8f3032008-07-22 09:59:21 -04002533 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534
2535 return error;
2536}
2537
Heiko Carstens3480b252009-01-14 14:14:16 +01002538SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002539{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002540 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002541}
2542
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543/*
2544 * The worst of all namespace operations - renaming directory. "Perverted"
2545 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2546 * Problems:
2547 * a) we can get into loop creation. Check is done in is_subdir().
2548 * b) race potential - two innocent renames can create a loop together.
2549 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002550 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 * story.
2552 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002553 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554 * whether the target exists). Solution: try to be smart with locking
2555 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002556 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557 * move will be locked. Thus we can rank directories by the tree
2558 * (ancestors first) and rank all non-directories after them.
2559 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002560 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561 * HOWEVER, it relies on the assumption that any object with ->lookup()
2562 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2563 * we'd better make sure that there's no link(2) for them.
2564 * d) some filesystems don't support opened-but-unlinked directories,
2565 * either because of layout or because they are not ready to deal with
2566 * all cases correctly. The latter will be fixed (taking this sort of
2567 * stuff into VFS), but the former is not going away. Solution: the same
2568 * trick as in rmdir().
2569 * e) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002570 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002572 * ->i_mutex on parents, which works but leads to some truely excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573 * locking].
2574 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07002575static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2576 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577{
2578 int error = 0;
2579 struct inode *target;
2580
2581 /*
2582 * If we are going to change the parent - check write permissions,
2583 * we'll need to flip '..'.
2584 */
2585 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04002586 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587 if (error)
2588 return error;
2589 }
2590
2591 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2592 if (error)
2593 return error;
2594
2595 target = new_dentry->d_inode;
2596 if (target) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002597 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598 dentry_unhash(new_dentry);
2599 }
2600 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2601 error = -EBUSY;
2602 else
2603 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2604 if (target) {
2605 if (!error)
2606 target->i_flags |= S_DEAD;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002607 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608 if (d_unhashed(new_dentry))
2609 d_rehash(new_dentry);
2610 dput(new_dentry);
2611 }
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002612 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07002613 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2614 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615 return error;
2616}
2617
Adrian Bunk75c96f82005-05-05 16:16:09 -07002618static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2619 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620{
2621 struct inode *target;
2622 int error;
2623
2624 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2625 if (error)
2626 return error;
2627
2628 dget(new_dentry);
2629 target = new_dentry->d_inode;
2630 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002631 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2633 error = -EBUSY;
2634 else
2635 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2636 if (!error) {
Mark Fasheh349457c2006-09-08 14:22:21 -07002637 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638 d_move(old_dentry, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639 }
2640 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002641 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642 dput(new_dentry);
2643 return error;
2644}
2645
2646int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2647 struct inode *new_dir, struct dentry *new_dentry)
2648{
2649 int error;
2650 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Robert Love0eeca282005-07-12 17:06:03 -04002651 const char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652
2653 if (old_dentry->d_inode == new_dentry->d_inode)
2654 return 0;
2655
2656 error = may_delete(old_dir, old_dentry, is_dir);
2657 if (error)
2658 return error;
2659
2660 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02002661 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662 else
2663 error = may_delete(new_dir, new_dentry, is_dir);
2664 if (error)
2665 return error;
2666
Al Viroacfa4382008-12-04 10:06:33 -05002667 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668 return -EPERM;
2669
Jan Kara9e3509e2009-01-26 16:45:12 +01002670 vfs_dq_init(old_dir);
2671 vfs_dq_init(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672
Robert Love0eeca282005-07-12 17:06:03 -04002673 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2674
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675 if (is_dir)
2676 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2677 else
2678 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2679 if (!error) {
Robert Love0eeca282005-07-12 17:06:03 -04002680 const char *new_name = old_dentry->d_name.name;
John McCutchan89204c42005-08-15 12:13:28 -04002681 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04002682 new_dentry->d_inode, old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002683 }
Robert Love0eeca282005-07-12 17:06:03 -04002684 fsnotify_oldname_free(old_name);
2685
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686 return error;
2687}
2688
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002689SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2690 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691{
Al Viro2ad94ae2008-07-21 09:32:51 -04002692 struct dentry *old_dir, *new_dir;
2693 struct dentry *old_dentry, *new_dentry;
2694 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695 struct nameidata oldnd, newnd;
Al Viro2ad94ae2008-07-21 09:32:51 -04002696 char *from;
2697 char *to;
2698 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699
Al Viro2ad94ae2008-07-21 09:32:51 -04002700 error = user_path_parent(olddfd, oldname, &oldnd, &from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002701 if (error)
2702 goto exit;
2703
Al Viro2ad94ae2008-07-21 09:32:51 -04002704 error = user_path_parent(newdfd, newname, &newnd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705 if (error)
2706 goto exit1;
2707
2708 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08002709 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710 goto exit2;
2711
Jan Blunck4ac91372008-02-14 19:34:32 -08002712 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713 error = -EBUSY;
2714 if (oldnd.last_type != LAST_NORM)
2715 goto exit2;
2716
Jan Blunck4ac91372008-02-14 19:34:32 -08002717 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718 if (newnd.last_type != LAST_NORM)
2719 goto exit2;
2720
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002721 oldnd.flags &= ~LOOKUP_PARENT;
2722 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09002723 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002724
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725 trap = lock_rename(new_dir, old_dir);
2726
Christoph Hellwig49705b72005-11-08 21:35:06 -08002727 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728 error = PTR_ERR(old_dentry);
2729 if (IS_ERR(old_dentry))
2730 goto exit3;
2731 /* source must exist */
2732 error = -ENOENT;
2733 if (!old_dentry->d_inode)
2734 goto exit4;
2735 /* unless the source is a directory trailing slashes give -ENOTDIR */
2736 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2737 error = -ENOTDIR;
2738 if (oldnd.last.name[oldnd.last.len])
2739 goto exit4;
2740 if (newnd.last.name[newnd.last.len])
2741 goto exit4;
2742 }
2743 /* source should not be ancestor of target */
2744 error = -EINVAL;
2745 if (old_dentry == trap)
2746 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08002747 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748 error = PTR_ERR(new_dentry);
2749 if (IS_ERR(new_dentry))
2750 goto exit4;
2751 /* target should not be an ancestor of source */
2752 error = -ENOTEMPTY;
2753 if (new_dentry == trap)
2754 goto exit5;
2755
Dave Hansen9079b1e2008-02-15 14:37:49 -08002756 error = mnt_want_write(oldnd.path.mnt);
2757 if (error)
2758 goto exit5;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002759 error = security_path_rename(&oldnd.path, old_dentry,
2760 &newnd.path, new_dentry);
2761 if (error)
2762 goto exit6;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002763 error = vfs_rename(old_dir->d_inode, old_dentry,
2764 new_dir->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002765exit6:
Dave Hansen9079b1e2008-02-15 14:37:49 -08002766 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002767exit5:
2768 dput(new_dentry);
2769exit4:
2770 dput(old_dentry);
2771exit3:
2772 unlock_rename(new_dir, old_dir);
2773exit2:
Jan Blunck1d957f92008-02-14 19:34:35 -08002774 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002775 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002777 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002778 putname(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002779exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780 return error;
2781}
2782
Heiko Carstensa26eab22009-01-14 14:14:17 +01002783SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002784{
2785 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2786}
2787
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2789{
2790 int len;
2791
2792 len = PTR_ERR(link);
2793 if (IS_ERR(link))
2794 goto out;
2795
2796 len = strlen(link);
2797 if (len > (unsigned) buflen)
2798 len = buflen;
2799 if (copy_to_user(buffer, link, len))
2800 len = -EFAULT;
2801out:
2802 return len;
2803}
2804
2805/*
2806 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2807 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2808 * using) it for any given inode is up to filesystem.
2809 */
2810int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2811{
2812 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002813 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07002814 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002815
Linus Torvalds1da177e2005-04-16 15:20:36 -07002816 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002817 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07002818 if (IS_ERR(cookie))
2819 return PTR_ERR(cookie);
2820
2821 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2822 if (dentry->d_inode->i_op->put_link)
2823 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2824 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002825}
2826
2827int vfs_follow_link(struct nameidata *nd, const char *link)
2828{
2829 return __vfs_follow_link(nd, link);
2830}
2831
2832/* get the link contents into pagecache */
2833static char *page_getlink(struct dentry * dentry, struct page **ppage)
2834{
Duane Griffinebd09ab2008-12-19 20:47:12 +00002835 char *kaddr;
2836 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07002838 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07002840 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00002842 kaddr = kmap(page);
2843 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2844 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845}
2846
2847int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2848{
2849 struct page *page = NULL;
2850 char *s = page_getlink(dentry, &page);
2851 int res = vfs_readlink(dentry,buffer,buflen,s);
2852 if (page) {
2853 kunmap(page);
2854 page_cache_release(page);
2855 }
2856 return res;
2857}
2858
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002859void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002860{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002861 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002863 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864}
2865
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002866void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002868 struct page *page = cookie;
2869
2870 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002871 kunmap(page);
2872 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002873 }
2874}
2875
Nick Piggin54566b22009-01-04 12:00:53 -08002876/*
2877 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2878 */
2879int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002880{
2881 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002882 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002883 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08002884 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002885 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08002886 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2887 if (nofs)
2888 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002889
NeilBrown7e53cac2006-03-25 03:07:57 -08002890retry:
Nick Pigginafddba42007-10-16 01:25:01 -07002891 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08002892 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002893 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07002894 goto fail;
2895
Linus Torvalds1da177e2005-04-16 15:20:36 -07002896 kaddr = kmap_atomic(page, KM_USER0);
2897 memcpy(kaddr, symname, len-1);
2898 kunmap_atomic(kaddr, KM_USER0);
Nick Pigginafddba42007-10-16 01:25:01 -07002899
2900 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2901 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902 if (err < 0)
2903 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07002904 if (err < len-1)
2905 goto retry;
2906
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907 mark_inode_dirty(inode);
2908 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909fail:
2910 return err;
2911}
2912
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002913int page_symlink(struct inode *inode, const char *symname, int len)
2914{
2915 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08002916 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002917}
2918
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002919const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002920 .readlink = generic_readlink,
2921 .follow_link = page_follow_link_light,
2922 .put_link = page_put_link,
2923};
2924
Al Viro2d8f3032008-07-22 09:59:21 -04002925EXPORT_SYMBOL(user_path_at);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002926EXPORT_SYMBOL(follow_down);
2927EXPORT_SYMBOL(follow_up);
2928EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2929EXPORT_SYMBOL(getname);
2930EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002931EXPORT_SYMBOL(lookup_one_len);
2932EXPORT_SYMBOL(page_follow_link_light);
2933EXPORT_SYMBOL(page_put_link);
2934EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002935EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002936EXPORT_SYMBOL(page_symlink);
2937EXPORT_SYMBOL(page_symlink_inode_operations);
2938EXPORT_SYMBOL(path_lookup);
Al Virod1811462008-08-02 00:49:18 -04002939EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002940EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04002941EXPORT_SYMBOL(inode_permission);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -08002942EXPORT_SYMBOL(file_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002943EXPORT_SYMBOL(unlock_rename);
2944EXPORT_SYMBOL(vfs_create);
2945EXPORT_SYMBOL(vfs_follow_link);
2946EXPORT_SYMBOL(vfs_link);
2947EXPORT_SYMBOL(vfs_mkdir);
2948EXPORT_SYMBOL(vfs_mknod);
2949EXPORT_SYMBOL(generic_permission);
2950EXPORT_SYMBOL(vfs_readlink);
2951EXPORT_SYMBOL(vfs_rename);
2952EXPORT_SYMBOL(vfs_rmdir);
2953EXPORT_SYMBOL(vfs_symlink);
2954EXPORT_SYMBOL(vfs_unlink);
2955EXPORT_SYMBOL(dentry_unhash);
2956EXPORT_SYMBOL(generic_readlink);