blob: b86b96fe1dc33926eb242a2a2e21061fe46145a5 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/pagemap.h>
Robert Love0eeca282005-07-12 17:06:03 -040023#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/personality.h>
25#include <linux/security.h>
Mimi Zohar6146f0d2009-02-04 09:06:57 -050026#include <linux/ima.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/syscalls.h>
28#include <linux/mount.h>
29#include <linux/audit.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080030#include <linux/capability.h>
Trond Myklebust834f2a42005-10-18 14:20:16 -070031#include <linux/file.h>
Ulrich Drepper5590ff02006-01-18 17:43:53 -080032#include <linux/fcntl.h>
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -070033#include <linux/device_cgroup.h>
Al Viro5ad4e532009-03-29 19:50:06 -040034#include <linux/fs_struct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <asm/uaccess.h>
36
Eric Parise81e3f42009-12-04 15:47:36 -050037#include "internal.h"
38
Linus Torvalds1da177e2005-04-16 15:20:36 -070039/* [Feb-1997 T. Schoebel-Theuer]
40 * Fundamental changes in the pathname lookup mechanisms (namei)
41 * were necessary because of omirr. The reason is that omirr needs
42 * to know the _real_ pathname, not the user-supplied one, in case
43 * of symlinks (and also when transname replacements occur).
44 *
45 * The new code replaces the old recursive symlink resolution with
46 * an iterative one (in case of non-nested symlink chains). It does
47 * this with calls to <fs>_follow_link().
48 * As a side effect, dir_namei(), _namei() and follow_link() are now
49 * replaced with a single function lookup_dentry() that can handle all
50 * the special cases of the former code.
51 *
52 * With the new dcache, the pathname is stored at each inode, at least as
53 * long as the refcount of the inode is positive. As a side effect, the
54 * size of the dcache depends on the inode cache and thus is dynamic.
55 *
56 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57 * resolution to correspond with current state of the code.
58 *
59 * Note that the symlink resolution is not *completely* iterative.
60 * There is still a significant amount of tail- and mid- recursion in
61 * the algorithm. Also, note that <fs>_readlink() is not used in
62 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63 * may return different results than <fs>_follow_link(). Many virtual
64 * filesystems (including /proc) exhibit this behavior.
65 */
66
67/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69 * and the name already exists in form of a symlink, try to create the new
70 * name indicated by the symlink. The old code always complained that the
71 * name already exists, due to not following the symlink even if its target
72 * is nonexistent. The new semantics affects also mknod() and link() when
73 * the name is a symlink pointing to a non-existant name.
74 *
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
82 */
83
84/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
86 *
87 * [10-Sep-98 Alan Modra] Another symlink change.
88 */
89
90/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
97 *
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
103 */
104/*
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800106 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 * any extra contention...
108 */
109
110/* In order to reduce some races, while at the same time doing additional
111 * checking and hopefully speeding things up, we copy filenames to the
112 * kernel data space before using them..
113 *
114 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
115 * PATH_MAX includes the nul terminator --RR.
116 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800117static int do_getname(const char __user *filename, char *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118{
119 int retval;
120 unsigned long len = PATH_MAX;
121
122 if (!segment_eq(get_fs(), KERNEL_DS)) {
123 if ((unsigned long) filename >= TASK_SIZE)
124 return -EFAULT;
125 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
126 len = TASK_SIZE - (unsigned long) filename;
127 }
128
129 retval = strncpy_from_user(page, filename, len);
130 if (retval > 0) {
131 if (retval < len)
132 return 0;
133 return -ENAMETOOLONG;
134 } else if (!retval)
135 retval = -ENOENT;
136 return retval;
137}
138
139char * getname(const char __user * filename)
140{
141 char *tmp, *result;
142
143 result = ERR_PTR(-ENOMEM);
144 tmp = __getname();
145 if (tmp) {
146 int retval = do_getname(filename, tmp);
147
148 result = tmp;
149 if (retval < 0) {
150 __putname(tmp);
151 result = ERR_PTR(retval);
152 }
153 }
154 audit_getname(result);
155 return result;
156}
157
158#ifdef CONFIG_AUDITSYSCALL
159void putname(const char *name)
160{
Al Viro5ac3a9c2006-07-16 06:38:45 -0400161 if (unlikely(!audit_dummy_context()))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 audit_putname(name);
163 else
164 __putname(name);
165}
166EXPORT_SYMBOL(putname);
167#endif
168
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700169/*
170 * This does basic POSIX ACL permission checking
171 */
172static int acl_permission_check(struct inode *inode, int mask,
173 int (*check_acl)(struct inode *inode, int mask))
174{
175 umode_t mode = inode->i_mode;
176
177 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
178
179 if (current_fsuid() == inode->i_uid)
180 mode >>= 6;
181 else {
182 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
183 int error = check_acl(inode, mask);
184 if (error != -EAGAIN)
185 return error;
186 }
187
188 if (in_group_p(inode->i_gid))
189 mode >>= 3;
190 }
191
192 /*
193 * If the DACs are ok we don't need any capability check.
194 */
195 if ((mask & ~mode) == 0)
196 return 0;
197 return -EACCES;
198}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199
200/**
201 * generic_permission - check for access rights on a Posix-like filesystem
202 * @inode: inode to check access rights for
203 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
204 * @check_acl: optional callback to check for Posix ACLs
205 *
206 * Used to check for read/write/execute permissions on a file.
207 * We use "fsuid" for this, letting us set arbitrary permissions
208 * for filesystem access without changing the "normal" uids which
209 * are used for other things..
210 */
211int generic_permission(struct inode *inode, int mask,
212 int (*check_acl)(struct inode *inode, int mask))
213{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700214 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215
216 /*
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700217 * Do the basic POSIX ACL permission checks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 */
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700219 ret = acl_permission_check(inode, mask, check_acl);
220 if (ret != -EACCES)
221 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 /*
224 * Read/write DACs are always overridable.
225 * Executable DACs are overridable if at least one exec bit is set.
226 */
Miklos Szeredif696a362008-07-31 13:41:58 +0200227 if (!(mask & MAY_EXEC) || execute_ok(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 if (capable(CAP_DAC_OVERRIDE))
229 return 0;
230
231 /*
232 * Searching includes executable on directories, else just read.
233 */
Serge E. Hallyn7ea66002009-12-29 14:50:19 -0600234 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
236 if (capable(CAP_DAC_READ_SEARCH))
237 return 0;
238
239 return -EACCES;
240}
241
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200242/**
243 * inode_permission - check for access rights to a given inode
244 * @inode: inode to check permission on
245 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
246 *
247 * Used to check for read/write/execute permissions on an inode.
248 * We use "fsuid" for this, letting us set arbitrary permissions
249 * for filesystem access without changing the "normal" uids which
250 * are used for other things.
251 */
Al Virof419a2e2008-07-22 00:07:17 -0400252int inode_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253{
Al Viroe6305c42008-07-15 21:03:57 -0400254 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255
256 if (mask & MAY_WRITE) {
Miklos Szeredi22590e42007-10-16 23:27:08 -0700257 umode_t mode = inode->i_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258
259 /*
260 * Nobody gets write access to a read-only fs.
261 */
262 if (IS_RDONLY(inode) &&
263 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
264 return -EROFS;
265
266 /*
267 * Nobody gets write access to an immutable file.
268 */
269 if (IS_IMMUTABLE(inode))
270 return -EACCES;
271 }
272
Al Viroacfa4382008-12-04 10:06:33 -0500273 if (inode->i_op->permission)
Al Virob77b0642008-07-17 09:37:02 -0400274 retval = inode->i_op->permission(inode, mask);
Miklos Szeredif696a362008-07-31 13:41:58 +0200275 else
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700276 retval = generic_permission(inode, mask, inode->i_op->check_acl);
Miklos Szeredif696a362008-07-31 13:41:58 +0200277
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 if (retval)
279 return retval;
280
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -0700281 retval = devcgroup_inode_permission(inode, mask);
282 if (retval)
283 return retval;
284
Al Viroe6305c42008-07-15 21:03:57 -0400285 return security_inode_permission(inode,
Stephen Smalleyf418b002008-07-28 13:32:38 -0400286 mask & (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287}
288
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800289/**
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800290 * file_permission - check for additional access rights to a given file
291 * @file: file to check access rights for
292 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
293 *
294 * Used to check for read/write/execute permissions on an already opened
295 * file.
296 *
297 * Note:
298 * Do not use this function in new code. All access checks should
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200299 * be done using inode_permission().
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800300 */
301int file_permission(struct file *file, int mask)
302{
Al Virof419a2e2008-07-22 00:07:17 -0400303 return inode_permission(file->f_path.dentry->d_inode, mask);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800304}
305
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306/*
307 * get_write_access() gets write permission for a file.
308 * put_write_access() releases this write permission.
309 * This is used for regular files.
310 * We cannot support write (and maybe mmap read-write shared) accesses and
311 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
312 * can have the following values:
313 * 0: no writers, no VM_DENYWRITE mappings
314 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
315 * > 0: (i_writecount) users are writing to the file.
316 *
317 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
318 * except for the cases where we don't hold i_writecount yet. Then we need to
319 * use {get,deny}_write_access() - these functions check the sign and refuse
320 * to do the change if sign is wrong. Exclusion between them is provided by
321 * the inode->i_lock spinlock.
322 */
323
324int get_write_access(struct inode * inode)
325{
326 spin_lock(&inode->i_lock);
327 if (atomic_read(&inode->i_writecount) < 0) {
328 spin_unlock(&inode->i_lock);
329 return -ETXTBSY;
330 }
331 atomic_inc(&inode->i_writecount);
332 spin_unlock(&inode->i_lock);
333
334 return 0;
335}
336
337int deny_write_access(struct file * file)
338{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800339 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340
341 spin_lock(&inode->i_lock);
342 if (atomic_read(&inode->i_writecount) > 0) {
343 spin_unlock(&inode->i_lock);
344 return -ETXTBSY;
345 }
346 atomic_dec(&inode->i_writecount);
347 spin_unlock(&inode->i_lock);
348
349 return 0;
350}
351
Jan Blunck1d957f92008-02-14 19:34:35 -0800352/**
Jan Blunck5dd784d02008-02-14 19:34:38 -0800353 * path_get - get a reference to a path
354 * @path: path to get the reference to
355 *
356 * Given a path increment the reference count to the dentry and the vfsmount.
357 */
358void path_get(struct path *path)
359{
360 mntget(path->mnt);
361 dget(path->dentry);
362}
363EXPORT_SYMBOL(path_get);
364
365/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800366 * path_put - put a reference to a path
367 * @path: path to put the reference to
368 *
369 * Given a path decrement the reference count to the dentry and the vfsmount.
370 */
371void path_put(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372{
Jan Blunck1d957f92008-02-14 19:34:35 -0800373 dput(path->dentry);
374 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375}
Jan Blunck1d957f92008-02-14 19:34:35 -0800376EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377
Trond Myklebust834f2a42005-10-18 14:20:16 -0700378/**
379 * release_open_intent - free up open intent resources
380 * @nd: pointer to nameidata
381 */
382void release_open_intent(struct nameidata *nd)
383{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800384 if (nd->intent.open.file->f_path.dentry == NULL)
Trond Myklebust834f2a42005-10-18 14:20:16 -0700385 put_filp(nd->intent.open.file);
386 else
387 fput(nd->intent.open.file);
388}
389
Ian Kentbcdc5e02006-09-27 01:50:44 -0700390static inline struct dentry *
391do_revalidate(struct dentry *dentry, struct nameidata *nd)
392{
393 int status = dentry->d_op->d_revalidate(dentry, nd);
394 if (unlikely(status <= 0)) {
395 /*
396 * The dentry failed validation.
397 * If d_revalidate returned 0 attempt to invalidate
398 * the dentry otherwise d_revalidate is asking us
399 * to return a fail status.
400 */
401 if (!status) {
402 if (!d_invalidate(dentry)) {
403 dput(dentry);
404 dentry = NULL;
405 }
406 } else {
407 dput(dentry);
408 dentry = ERR_PTR(status);
409 }
410 }
411 return dentry;
412}
413
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414/*
Jeff Layton39159de2009-12-07 12:01:50 -0500415 * force_reval_path - force revalidation of a dentry
416 *
417 * In some situations the path walking code will trust dentries without
418 * revalidating them. This causes problems for filesystems that depend on
419 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
420 * (which indicates that it's possible for the dentry to go stale), force
421 * a d_revalidate call before proceeding.
422 *
423 * Returns 0 if the revalidation was successful. If the revalidation fails,
424 * either return the error returned by d_revalidate or -ESTALE if the
425 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
426 * invalidate the dentry. It's up to the caller to handle putting references
427 * to the path if necessary.
428 */
429static int
430force_reval_path(struct path *path, struct nameidata *nd)
431{
432 int status;
433 struct dentry *dentry = path->dentry;
434
435 /*
436 * only check on filesystems where it's possible for the dentry to
437 * become stale. It's assumed that if this flag is set then the
438 * d_revalidate op will also be defined.
439 */
440 if (!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT))
441 return 0;
442
443 status = dentry->d_op->d_revalidate(dentry, nd);
444 if (status > 0)
445 return 0;
446
447 if (!status) {
448 d_invalidate(dentry);
449 status = -ESTALE;
450 }
451 return status;
452}
453
454/*
Al Virob75b5082009-12-16 01:01:38 -0500455 * Short-cut version of permission(), for calling on directories
456 * during pathname resolution. Combines parts of permission()
457 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 *
459 * If appropriate, check DAC only. If not appropriate, or
Al Virob75b5082009-12-16 01:01:38 -0500460 * short-cut DAC fails, then call ->permission() to do more
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 * complete permission check.
462 */
Al Virob75b5082009-12-16 01:01:38 -0500463static int exec_permission(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700465 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466
Linus Torvaldscb9179e2009-08-28 11:08:31 -0700467 if (inode->i_op->permission) {
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700468 ret = inode->i_op->permission(inode, MAY_EXEC);
Linus Torvaldscb9179e2009-08-28 11:08:31 -0700469 if (!ret)
470 goto ok;
471 return ret;
472 }
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700473 ret = acl_permission_check(inode, MAY_EXEC, inode->i_op->check_acl);
474 if (!ret)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 goto ok;
476
Linus Torvaldsf1ac9f62009-08-28 10:53:56 -0700477 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 goto ok;
479
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700480 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481ok:
Al Virob77b0642008-07-17 09:37:02 -0400482 return security_inode_permission(inode, MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483}
484
Al Viro2a737872009-04-07 11:49:53 -0400485static __always_inline void set_root(struct nameidata *nd)
486{
487 if (!nd->root.mnt) {
488 struct fs_struct *fs = current->fs;
489 read_lock(&fs->lock);
490 nd->root = fs->root;
491 path_get(&nd->root);
492 read_unlock(&fs->lock);
493 }
494}
495
Al Viro6de88d72009-08-09 01:41:57 +0400496static int link_path_walk(const char *, struct nameidata *);
497
Arjan van de Venf1662352006-01-14 13:21:31 -0800498static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 if (IS_ERR(link))
501 goto fail;
502
503 if (*link == '/') {
Al Viro2a737872009-04-07 11:49:53 -0400504 set_root(nd);
Jan Blunck1d957f92008-02-14 19:34:35 -0800505 path_put(&nd->path);
Al Viro2a737872009-04-07 11:49:53 -0400506 nd->path = nd->root;
507 path_get(&nd->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 }
Christoph Hellwigb4091d52008-11-05 15:07:21 +0100509
Al Virodef4af32009-12-26 08:37:05 -0500510 return link_path_walk(link, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511fail:
Jan Blunck1d957f92008-02-14 19:34:35 -0800512 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 return PTR_ERR(link);
514}
515
Jan Blunck1d957f92008-02-14 19:34:35 -0800516static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700517{
518 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800519 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700520 mntput(path->mnt);
521}
522
523static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
524{
Jan Blunck4ac91372008-02-14 19:34:32 -0800525 dput(nd->path.dentry);
526 if (nd->path.mnt != path->mnt)
527 mntput(nd->path.mnt);
528 nd->path.mnt = path->mnt;
529 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700530}
531
Al Virodef4af32009-12-26 08:37:05 -0500532static __always_inline int
533__do_follow_link(struct path *path, struct nameidata *nd, void **p)
Ian Kent051d3812006-03-27 01:14:53 -0800534{
535 int error;
Ian Kent051d3812006-03-27 01:14:53 -0800536 struct dentry *dentry = path->dentry;
537
538 touch_atime(path->mnt, dentry);
539 nd_set_link(nd, NULL);
540
Jan Blunck4ac91372008-02-14 19:34:32 -0800541 if (path->mnt != nd->path.mnt) {
Ian Kent051d3812006-03-27 01:14:53 -0800542 path_to_nameidata(path, nd);
543 dget(dentry);
544 }
545 mntget(path->mnt);
Al Viro86acdca12009-12-22 23:45:11 -0500546 nd->last_type = LAST_BIND;
Al Virodef4af32009-12-26 08:37:05 -0500547 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
548 error = PTR_ERR(*p);
549 if (!IS_ERR(*p)) {
Ian Kent051d3812006-03-27 01:14:53 -0800550 char *s = nd_get_link(nd);
551 error = 0;
552 if (s)
553 error = __vfs_follow_link(nd, s);
Jeff Layton39159de2009-12-07 12:01:50 -0500554 else if (nd->last_type == LAST_BIND) {
555 error = force_reval_path(&nd->path, nd);
556 if (error)
557 path_put(&nd->path);
558 }
Ian Kent051d3812006-03-27 01:14:53 -0800559 }
Ian Kent051d3812006-03-27 01:14:53 -0800560 return error;
561}
562
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563/*
564 * This limits recursive symlink follows to 8, while
565 * limiting consecutive symlinks to 40.
566 *
567 * Without that kind of total limit, nasty chains of consecutive
568 * symlinks can cause almost arbitrarily long lookups.
569 */
Al Viro90ebe562005-06-06 13:35:58 -0700570static inline int do_follow_link(struct path *path, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571{
Al Virodef4af32009-12-26 08:37:05 -0500572 void *cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 int err = -ELOOP;
574 if (current->link_count >= MAX_NESTED_LINKS)
575 goto loop;
576 if (current->total_link_count >= 40)
577 goto loop;
578 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
579 cond_resched();
Al Viro90ebe562005-06-06 13:35:58 -0700580 err = security_inode_follow_link(path->dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 if (err)
582 goto loop;
583 current->link_count++;
584 current->total_link_count++;
585 nd->depth++;
Al Virodef4af32009-12-26 08:37:05 -0500586 err = __do_follow_link(path, nd, &cookie);
587 if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link)
588 path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie);
Al Viro258fa992009-08-09 01:32:02 +0400589 path_put(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 current->link_count--;
591 nd->depth--;
592 return err;
593loop:
Jan Blunck1d957f92008-02-14 19:34:35 -0800594 path_put_conditional(path, nd);
595 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 return err;
597}
598
Al Virobab77eb2009-04-18 03:26:48 -0400599int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600{
601 struct vfsmount *parent;
602 struct dentry *mountpoint;
603 spin_lock(&vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400604 parent = path->mnt->mnt_parent;
605 if (parent == path->mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 spin_unlock(&vfsmount_lock);
607 return 0;
608 }
609 mntget(parent);
Al Virobab77eb2009-04-18 03:26:48 -0400610 mountpoint = dget(path->mnt->mnt_mountpoint);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 spin_unlock(&vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400612 dput(path->dentry);
613 path->dentry = mountpoint;
614 mntput(path->mnt);
615 path->mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 return 1;
617}
618
619/* no need for dcache_lock, as serialization is taken care in
620 * namespace.c
621 */
Al Viro463ffb22005-06-06 13:36:05 -0700622static int __follow_mount(struct path *path)
623{
624 int res = 0;
625 while (d_mountpoint(path->dentry)) {
Al Viro1c755af2009-04-18 14:06:57 -0400626 struct vfsmount *mounted = lookup_mnt(path);
Al Viro463ffb22005-06-06 13:36:05 -0700627 if (!mounted)
628 break;
629 dput(path->dentry);
630 if (res)
631 mntput(path->mnt);
632 path->mnt = mounted;
633 path->dentry = dget(mounted->mnt_root);
634 res = 1;
635 }
636 return res;
637}
638
Al Viro79ed0222009-04-18 13:59:41 -0400639static void follow_mount(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640{
Al Viro79ed0222009-04-18 13:59:41 -0400641 while (d_mountpoint(path->dentry)) {
Al Viro1c755af2009-04-18 14:06:57 -0400642 struct vfsmount *mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 if (!mounted)
644 break;
Al Viro79ed0222009-04-18 13:59:41 -0400645 dput(path->dentry);
646 mntput(path->mnt);
647 path->mnt = mounted;
648 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650}
651
652/* no need for dcache_lock, as serialization is taken care in
653 * namespace.c
654 */
Al Viro9393bd02009-04-18 13:58:15 -0400655int follow_down(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656{
657 struct vfsmount *mounted;
658
Al Viro1c755af2009-04-18 14:06:57 -0400659 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -0400661 dput(path->dentry);
662 mntput(path->mnt);
663 path->mnt = mounted;
664 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 return 1;
666 }
667 return 0;
668}
669
Arjan van de Venf1662352006-01-14 13:21:31 -0800670static __always_inline void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671{
Al Viro2a737872009-04-07 11:49:53 -0400672 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -0700673
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 while(1) {
Jan Blunck4ac91372008-02-14 19:34:32 -0800675 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676
Al Viro2a737872009-04-07 11:49:53 -0400677 if (nd->path.dentry == nd->root.dentry &&
678 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 break;
680 }
Jan Blunck4ac91372008-02-14 19:34:32 -0800681 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Al Viro3088dd72010-01-30 15:47:29 -0500682 /* rare case of legitimate dget_parent()... */
683 nd->path.dentry = dget_parent(nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 dput(old);
685 break;
686 }
Al Viro3088dd72010-01-30 15:47:29 -0500687 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 }
Al Viro79ed0222009-04-18 13:59:41 -0400690 follow_mount(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691}
692
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693/*
694 * It's more convoluted than I'd like it to be, but... it's still fairly
695 * small and for now I'd prefer to have fast path as straight as possible.
696 * It _is_ time-critical.
697 */
698static int do_lookup(struct nameidata *nd, struct qstr *name,
699 struct path *path)
700{
Jan Blunck4ac91372008-02-14 19:34:32 -0800701 struct vfsmount *mnt = nd->path.mnt;
Al Viro6e6b1bd2009-08-13 23:38:37 +0400702 struct dentry *dentry, *parent;
703 struct inode *dir;
Al Viro3cac2602009-08-13 18:27:43 +0400704 /*
705 * See if the low-level filesystem might want
706 * to use its own hash..
707 */
708 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
709 int err = nd->path.dentry->d_op->d_hash(nd->path.dentry, name);
710 if (err < 0)
711 return err;
712 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713
Al Viro3cac2602009-08-13 18:27:43 +0400714 dentry = __d_lookup(nd->path.dentry, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 if (!dentry)
716 goto need_lookup;
717 if (dentry->d_op && dentry->d_op->d_revalidate)
718 goto need_revalidate;
719done:
720 path->mnt = mnt;
721 path->dentry = dentry;
Al Viro634ee702005-06-06 13:36:13 -0700722 __follow_mount(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 return 0;
724
725need_lookup:
Al Viro6e6b1bd2009-08-13 23:38:37 +0400726 parent = nd->path.dentry;
727 dir = parent->d_inode;
728
729 mutex_lock(&dir->i_mutex);
730 /*
731 * First re-do the cached lookup just in case it was created
732 * while we waited for the directory semaphore..
733 *
734 * FIXME! This could use version numbering or similar to
735 * avoid unnecessary cache lookups.
736 *
737 * The "dcache_lock" is purely to protect the RCU list walker
738 * from concurrent renames at this point (we mustn't get false
739 * negatives from the RCU list walk here, unlike the optimistic
740 * fast walk).
741 *
742 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
743 */
744 dentry = d_lookup(parent, name);
745 if (!dentry) {
746 struct dentry *new;
747
748 /* Don't create child dentry for a dead directory. */
749 dentry = ERR_PTR(-ENOENT);
750 if (IS_DEADDIR(dir))
751 goto out_unlock;
752
753 new = d_alloc(parent, name);
754 dentry = ERR_PTR(-ENOMEM);
755 if (new) {
756 dentry = dir->i_op->lookup(dir, new, nd);
757 if (dentry)
758 dput(new);
759 else
760 dentry = new;
761 }
762out_unlock:
763 mutex_unlock(&dir->i_mutex);
764 if (IS_ERR(dentry))
765 goto fail;
766 goto done;
767 }
768
769 /*
770 * Uhhuh! Nasty case: the cache was re-populated while
771 * we waited on the semaphore. Need to revalidate.
772 */
773 mutex_unlock(&dir->i_mutex);
774 if (dentry->d_op && dentry->d_op->d_revalidate) {
775 dentry = do_revalidate(dentry, nd);
776 if (!dentry)
777 dentry = ERR_PTR(-ENOENT);
778 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 if (IS_ERR(dentry))
780 goto fail;
781 goto done;
782
783need_revalidate:
Ian Kentbcdc5e02006-09-27 01:50:44 -0700784 dentry = do_revalidate(dentry, nd);
785 if (!dentry)
786 goto need_lookup;
787 if (IS_ERR(dentry))
788 goto fail;
789 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790
791fail:
792 return PTR_ERR(dentry);
793}
794
795/*
Al Viroac278a92010-02-16 18:09:36 +0000796 * This is a temporary kludge to deal with "automount" symlinks; proper
797 * solution is to trigger them on follow_mount(), so that do_lookup()
798 * would DTRT. To be killed before 2.6.34-final.
799 */
800static inline int follow_on_final(struct inode *inode, unsigned lookup_flags)
801{
802 return inode && unlikely(inode->i_op->follow_link) &&
803 ((lookup_flags & LOOKUP_FOLLOW) || S_ISDIR(inode->i_mode));
804}
805
806/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +0100808 * This is the basic name resolution function, turning a pathname into
809 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 *
Prasanna Medaea3834d2005-04-29 16:00:17 +0100811 * Returns 0 and nd will have valid dentry and mnt on success.
812 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 */
Al Viro6de88d72009-08-09 01:41:57 +0400814static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815{
816 struct path next;
817 struct inode *inode;
818 int err;
819 unsigned int lookup_flags = nd->flags;
820
821 while (*name=='/')
822 name++;
823 if (!*name)
824 goto return_reval;
825
Jan Blunck4ac91372008-02-14 19:34:32 -0800826 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 if (nd->depth)
Trond Myklebustf55eab82006-02-04 23:28:01 -0800828 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829
830 /* At this point we know we have a real path component. */
831 for(;;) {
832 unsigned long hash;
833 struct qstr this;
834 unsigned int c;
835
Trond Myklebustcdce5d62005-10-18 14:20:18 -0700836 nd->flags |= LOOKUP_CONTINUE;
Al Virob75b5082009-12-16 01:01:38 -0500837 err = exec_permission(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 if (err)
839 break;
840
841 this.name = name;
842 c = *(const unsigned char *)name;
843
844 hash = init_name_hash();
845 do {
846 name++;
847 hash = partial_name_hash(c, hash);
848 c = *(const unsigned char *)name;
849 } while (c && (c != '/'));
850 this.len = name - (const char *) this.name;
851 this.hash = end_name_hash(hash);
852
853 /* remove trailing slashes? */
854 if (!c)
855 goto last_component;
856 while (*++name == '/');
857 if (!*name)
858 goto last_with_slashes;
859
860 /*
861 * "." and ".." are special - ".." especially so because it has
862 * to be able to know about the current root directory and
863 * parent relationships.
864 */
865 if (this.name[0] == '.') switch (this.len) {
866 default:
867 break;
868 case 2:
869 if (this.name[1] != '.')
870 break;
Al Viro58c465e2005-06-06 13:36:13 -0700871 follow_dotdot(nd);
Jan Blunck4ac91372008-02-14 19:34:32 -0800872 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 /* fallthrough */
874 case 1:
875 continue;
876 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 /* This does the actual lookups.. */
878 err = do_lookup(nd, &this, &next);
879 if (err)
880 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881
882 err = -ENOENT;
883 inode = next.dentry->d_inode;
884 if (!inode)
885 goto out_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886
887 if (inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700888 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 if (err)
890 goto return_err;
891 err = -ENOENT;
Jan Blunck4ac91372008-02-14 19:34:32 -0800892 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 if (!inode)
894 break;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700895 } else
896 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 err = -ENOTDIR;
898 if (!inode->i_op->lookup)
899 break;
900 continue;
901 /* here ends the main loop */
902
903last_with_slashes:
904 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
905last_component:
Trond Myklebustf55eab82006-02-04 23:28:01 -0800906 /* Clear LOOKUP_CONTINUE iff it was previously unset */
907 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 if (lookup_flags & LOOKUP_PARENT)
909 goto lookup_parent;
910 if (this.name[0] == '.') switch (this.len) {
911 default:
912 break;
913 case 2:
914 if (this.name[1] != '.')
915 break;
Al Viro58c465e2005-06-06 13:36:13 -0700916 follow_dotdot(nd);
Jan Blunck4ac91372008-02-14 19:34:32 -0800917 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 /* fallthrough */
919 case 1:
920 goto return_reval;
921 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 err = do_lookup(nd, &this, &next);
923 if (err)
924 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 inode = next.dentry->d_inode;
Al Viroac278a92010-02-16 18:09:36 +0000926 if (follow_on_final(inode, lookup_flags)) {
Al Viro90ebe562005-06-06 13:35:58 -0700927 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 if (err)
929 goto return_err;
Jan Blunck4ac91372008-02-14 19:34:32 -0800930 inode = nd->path.dentry->d_inode;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700931 } else
932 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 err = -ENOENT;
934 if (!inode)
935 break;
936 if (lookup_flags & LOOKUP_DIRECTORY) {
937 err = -ENOTDIR;
Al Viroacfa4382008-12-04 10:06:33 -0500938 if (!inode->i_op->lookup)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 break;
940 }
941 goto return_base;
942lookup_parent:
943 nd->last = this;
944 nd->last_type = LAST_NORM;
945 if (this.name[0] != '.')
946 goto return_base;
947 if (this.len == 1)
948 nd->last_type = LAST_DOT;
949 else if (this.len == 2 && this.name[1] == '.')
950 nd->last_type = LAST_DOTDOT;
951 else
952 goto return_base;
953return_reval:
954 /*
955 * We bypassed the ordinary revalidation routines.
956 * We may need to check the cached dentry for staleness.
957 */
Jan Blunck4ac91372008-02-14 19:34:32 -0800958 if (nd->path.dentry && nd->path.dentry->d_sb &&
959 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 err = -ESTALE;
961 /* Note: we do not d_invalidate() */
Jan Blunck4ac91372008-02-14 19:34:32 -0800962 if (!nd->path.dentry->d_op->d_revalidate(
963 nd->path.dentry, nd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 break;
965 }
966return_base:
967 return 0;
968out_dput:
Jan Blunck1d957f92008-02-14 19:34:35 -0800969 path_put_conditional(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 break;
971 }
Jan Blunck1d957f92008-02-14 19:34:35 -0800972 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973return_err:
974 return err;
975}
976
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800977static int path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978{
Al Viro6de88d72009-08-09 01:41:57 +0400979 struct path save = nd->path;
980 int result;
981
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 current->total_link_count = 0;
Al Viro6de88d72009-08-09 01:41:57 +0400983
984 /* make sure the stuff we saved doesn't go away */
985 path_get(&save);
986
987 result = link_path_walk(name, nd);
988 if (result == -ESTALE) {
989 /* nd->path had been dropped */
990 current->total_link_count = 0;
991 nd->path = save;
992 path_get(&nd->path);
993 nd->flags |= LOOKUP_REVAL;
994 result = link_path_walk(name, nd);
995 }
996
997 path_put(&save);
998
999 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000}
1001
Al Viro9b4a9b12009-04-07 11:44:16 -04001002static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001004 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001005 int fput_needed;
1006 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007
1008 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1009 nd->flags = flags;
1010 nd->depth = 0;
Al Viro2a737872009-04-07 11:49:53 -04001011 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 if (*name=='/') {
Al Viro2a737872009-04-07 11:49:53 -04001014 set_root(nd);
1015 nd->path = nd->root;
1016 path_get(&nd->root);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001017 } else if (dfd == AT_FDCWD) {
Al Viro2a737872009-04-07 11:49:53 -04001018 struct fs_struct *fs = current->fs;
Andreas Mohre518ddb2006-09-29 02:01:22 -07001019 read_lock(&fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08001020 nd->path = fs->pwd;
1021 path_get(&fs->pwd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001022 read_unlock(&fs->lock);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001023 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001024 struct dentry *dentry;
1025
1026 file = fget_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001027 retval = -EBADF;
1028 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001029 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001030
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001031 dentry = file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001032
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001033 retval = -ENOTDIR;
1034 if (!S_ISDIR(dentry->d_inode->i_mode))
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001035 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001036
1037 retval = file_permission(file, MAY_EXEC);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001038 if (retval)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001039 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001040
Jan Blunck5dd784d02008-02-14 19:34:38 -08001041 nd->path = file->f_path;
1042 path_get(&file->f_path);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001043
1044 fput_light(file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001046 return 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001047
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001048fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001049 fput_light(file, fput_needed);
Al Viro9b4a9b12009-04-07 11:44:16 -04001050out_fail:
1051 return retval;
1052}
1053
1054/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1055static int do_path_lookup(int dfd, const char *name,
1056 unsigned int flags, struct nameidata *nd)
1057{
1058 int retval = path_init(dfd, name, flags, nd);
1059 if (!retval)
1060 retval = path_walk(name, nd);
1061 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1062 nd->path.dentry->d_inode))
1063 audit_inode(name, nd->path.dentry);
Al Viro2a737872009-04-07 11:49:53 -04001064 if (nd->root.mnt) {
1065 path_put(&nd->root);
1066 nd->root.mnt = NULL;
1067 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001068 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069}
1070
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -08001071int path_lookup(const char *name, unsigned int flags,
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001072 struct nameidata *nd)
1073{
1074 return do_path_lookup(AT_FDCWD, name, flags, nd);
1075}
1076
Al Virod1811462008-08-02 00:49:18 -04001077int kern_path(const char *name, unsigned int flags, struct path *path)
1078{
1079 struct nameidata nd;
1080 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1081 if (!res)
1082 *path = nd.path;
1083 return res;
1084}
1085
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001086/**
1087 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1088 * @dentry: pointer to dentry of the base directory
1089 * @mnt: pointer to vfs mount of the base directory
1090 * @name: pointer to file name
1091 * @flags: lookup flags
1092 * @nd: pointer to nameidata
1093 */
1094int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1095 const char *name, unsigned int flags,
1096 struct nameidata *nd)
1097{
1098 int retval;
1099
1100 /* same as do_path_lookup */
1101 nd->last_type = LAST_ROOT;
1102 nd->flags = flags;
1103 nd->depth = 0;
1104
Jan Blunckc8e7f442008-06-09 16:40:35 -07001105 nd->path.dentry = dentry;
1106 nd->path.mnt = mnt;
1107 path_get(&nd->path);
Al Viro5b857112009-04-07 11:53:49 -04001108 nd->root = nd->path;
1109 path_get(&nd->root);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001110
1111 retval = path_walk(name, nd);
Jan Blunck4ac91372008-02-14 19:34:32 -08001112 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1113 nd->path.dentry->d_inode))
1114 audit_inode(name, nd->path.dentry);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001115
Al Viro5b857112009-04-07 11:53:49 -04001116 path_put(&nd->root);
1117 nd->root.mnt = NULL;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001118
Al Viro2a737872009-04-07 11:49:53 -04001119 return retval;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001120}
1121
Christoph Hellwigeead1912007-10-16 23:25:38 -07001122static struct dentry *__lookup_hash(struct qstr *name,
1123 struct dentry *base, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124{
James Morris057f6c02007-04-26 00:12:05 -07001125 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 struct inode *inode;
1127 int err;
1128
1129 inode = base->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130
1131 /*
1132 * See if the low-level filesystem might want
1133 * to use its own hash..
1134 */
1135 if (base->d_op && base->d_op->d_hash) {
1136 err = base->d_op->d_hash(base, name);
1137 dentry = ERR_PTR(err);
1138 if (err < 0)
1139 goto out;
1140 }
1141
Al Viro6e6b1bd2009-08-13 23:38:37 +04001142 dentry = __d_lookup(base, name);
1143
1144 /* lockess __d_lookup may fail due to concurrent d_move()
1145 * in some unrelated directory, so try with d_lookup
1146 */
1147 if (!dentry)
1148 dentry = d_lookup(base, name);
1149
1150 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
1151 dentry = do_revalidate(dentry, nd);
1152
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 if (!dentry) {
Miklos Szeredid70b67c2008-07-02 21:30:15 +02001154 struct dentry *new;
1155
1156 /* Don't create child dentry for a dead directory. */
1157 dentry = ERR_PTR(-ENOENT);
1158 if (IS_DEADDIR(inode))
1159 goto out;
1160
1161 new = d_alloc(base, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 dentry = ERR_PTR(-ENOMEM);
1163 if (!new)
1164 goto out;
1165 dentry = inode->i_op->lookup(inode, new, nd);
1166 if (!dentry)
1167 dentry = new;
1168 else
1169 dput(new);
1170 }
1171out:
1172 return dentry;
1173}
1174
James Morris057f6c02007-04-26 00:12:05 -07001175/*
1176 * Restricted form of lookup. Doesn't follow links, single-component only,
1177 * needs parent already locked. Doesn't follow mounts.
1178 * SMP-safe.
1179 */
Adrian Bunka244e162006-03-31 02:32:11 -08001180static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181{
Christoph Hellwigeead1912007-10-16 23:25:38 -07001182 int err;
1183
Al Virob75b5082009-12-16 01:01:38 -05001184 err = exec_permission(nd->path.dentry->d_inode);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001185 if (err)
1186 return ERR_PTR(err);
Jan Blunck4ac91372008-02-14 19:34:32 -08001187 return __lookup_hash(&nd->last, nd->path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188}
1189
Christoph Hellwigeead1912007-10-16 23:25:38 -07001190static int __lookup_one_len(const char *name, struct qstr *this,
1191 struct dentry *base, int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192{
1193 unsigned long hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 unsigned int c;
1195
James Morris057f6c02007-04-26 00:12:05 -07001196 this->name = name;
1197 this->len = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 if (!len)
James Morris057f6c02007-04-26 00:12:05 -07001199 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200
1201 hash = init_name_hash();
1202 while (len--) {
1203 c = *(const unsigned char *)name++;
1204 if (c == '/' || c == '\0')
James Morris057f6c02007-04-26 00:12:05 -07001205 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 hash = partial_name_hash(c, hash);
1207 }
James Morris057f6c02007-04-26 00:12:05 -07001208 this->hash = end_name_hash(hash);
1209 return 0;
1210}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211
Christoph Hellwigeead1912007-10-16 23:25:38 -07001212/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001213 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07001214 * @name: pathname component to lookup
1215 * @base: base directory to lookup from
1216 * @len: maximum length @len should be interpreted to
1217 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07001218 * Note that this routine is purely a helper for filesystem usage and should
1219 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07001220 * nameidata argument is passed to the filesystem methods and a filesystem
1221 * using this helper needs to be prepared for that.
1222 */
James Morris057f6c02007-04-26 00:12:05 -07001223struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1224{
1225 int err;
1226 struct qstr this;
1227
David Woodhouse2f9092e2009-04-20 23:18:37 +01001228 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1229
James Morris057f6c02007-04-26 00:12:05 -07001230 err = __lookup_one_len(name, &this, base, len);
1231 if (err)
1232 return ERR_PTR(err);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001233
Al Virob75b5082009-12-16 01:01:38 -05001234 err = exec_permission(base->d_inode);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001235 if (err)
1236 return ERR_PTR(err);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001237 return __lookup_hash(&this, base, NULL);
James Morris057f6c02007-04-26 00:12:05 -07001238}
1239
Al Viro2d8f3032008-07-22 09:59:21 -04001240int user_path_at(int dfd, const char __user *name, unsigned flags,
1241 struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242{
Al Viro2d8f3032008-07-22 09:59:21 -04001243 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 char *tmp = getname(name);
1245 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04001247
1248 BUG_ON(flags & LOOKUP_PARENT);
1249
1250 err = do_path_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04001252 if (!err)
1253 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 }
1255 return err;
1256}
1257
Al Viro2ad94ae2008-07-21 09:32:51 -04001258static int user_path_parent(int dfd, const char __user *path,
1259 struct nameidata *nd, char **name)
1260{
1261 char *s = getname(path);
1262 int error;
1263
1264 if (IS_ERR(s))
1265 return PTR_ERR(s);
1266
1267 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1268 if (error)
1269 putname(s);
1270 else
1271 *name = s;
1272
1273 return error;
1274}
1275
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276/*
1277 * It's inline, so penalty for filesystems that don't use sticky bit is
1278 * minimal.
1279 */
1280static inline int check_sticky(struct inode *dir, struct inode *inode)
1281{
David Howellsda9592e2008-11-14 10:39:05 +11001282 uid_t fsuid = current_fsuid();
1283
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 if (!(dir->i_mode & S_ISVTX))
1285 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001286 if (inode->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001288 if (dir->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 return 0;
1290 return !capable(CAP_FOWNER);
1291}
1292
1293/*
1294 * Check whether we can remove a link victim from directory dir, check
1295 * whether the type of victim is right.
1296 * 1. We can't do it if dir is read-only (done in permission())
1297 * 2. We should have write and exec permissions on dir
1298 * 3. We can't remove anything from append-only dir
1299 * 4. We can't do anything with immutable dir (done in permission())
1300 * 5. If the sticky bit on dir is set we should either
1301 * a. be owner of dir, or
1302 * b. be owner of victim, or
1303 * c. have CAP_FOWNER capability
1304 * 6. If the victim is append-only or immutable we can't do antyhing with
1305 * links pointing to it.
1306 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1307 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1308 * 9. We can't remove a root or mountpoint.
1309 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1310 * nfs_async_unlink().
1311 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001312static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313{
1314 int error;
1315
1316 if (!victim->d_inode)
1317 return -ENOENT;
1318
1319 BUG_ON(victim->d_parent->d_inode != dir);
Al Virocccc6bb2009-12-25 05:07:33 -05001320 audit_inode_child(victim, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321
Al Virof419a2e2008-07-22 00:07:17 -04001322 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 if (error)
1324 return error;
1325 if (IS_APPEND(dir))
1326 return -EPERM;
1327 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
Hugh Dickinsf9454542008-11-19 15:36:38 -08001328 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 return -EPERM;
1330 if (isdir) {
1331 if (!S_ISDIR(victim->d_inode->i_mode))
1332 return -ENOTDIR;
1333 if (IS_ROOT(victim))
1334 return -EBUSY;
1335 } else if (S_ISDIR(victim->d_inode->i_mode))
1336 return -EISDIR;
1337 if (IS_DEADDIR(dir))
1338 return -ENOENT;
1339 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1340 return -EBUSY;
1341 return 0;
1342}
1343
1344/* Check whether we can create an object with dentry child in directory
1345 * dir.
1346 * 1. We can't do it if child already exists (open has special treatment for
1347 * this case, but since we are inlined it's OK)
1348 * 2. We can't do it if dir is read-only (done in permission())
1349 * 3. We should have write and exec permissions on dir
1350 * 4. We can't do it if dir is immutable (done in permission())
1351 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02001352static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353{
1354 if (child->d_inode)
1355 return -EEXIST;
1356 if (IS_DEADDIR(dir))
1357 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04001358 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359}
1360
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361/*
1362 * p1 and p2 should be directories on the same fs.
1363 */
1364struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1365{
1366 struct dentry *p;
1367
1368 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001369 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 return NULL;
1371 }
1372
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001373 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001375 p = d_ancestor(p2, p1);
1376 if (p) {
1377 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1378 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1379 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 }
1381
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001382 p = d_ancestor(p1, p2);
1383 if (p) {
1384 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1385 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1386 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 }
1388
Ingo Molnarf2eace22006-07-03 00:25:05 -07001389 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1390 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 return NULL;
1392}
1393
1394void unlock_rename(struct dentry *p1, struct dentry *p2)
1395{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001396 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001398 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001399 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 }
1401}
1402
1403int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1404 struct nameidata *nd)
1405{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001406 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407
1408 if (error)
1409 return error;
1410
Al Viroacfa4382008-12-04 10:06:33 -05001411 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 return -EACCES; /* shouldn't it be ENOSYS? */
1413 mode &= S_IALLUGO;
1414 mode |= S_IFREG;
1415 error = security_inode_create(dir, dentry, mode);
1416 if (error)
1417 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001419 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001420 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 return error;
1422}
1423
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001424int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001426 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 struct inode *inode = dentry->d_inode;
1428 int error;
1429
1430 if (!inode)
1431 return -ENOENT;
1432
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001433 switch (inode->i_mode & S_IFMT) {
1434 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001436 case S_IFDIR:
1437 if (acc_mode & MAY_WRITE)
1438 return -EISDIR;
1439 break;
1440 case S_IFBLK:
1441 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001442 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001444 /*FALLTHRU*/
1445 case S_IFIFO:
1446 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001448 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001449 }
Dave Hansenb41572e2007-10-16 23:31:14 -07001450
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001451 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07001452 if (error)
1453 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05001454
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 /*
1456 * An append-only file must be opened in append mode for writing.
1457 */
1458 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05001459 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05001460 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05001462 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 }
1464
1465 /* O_NOATIME can only be set by the owner or superuser */
Al Viro7715b522009-12-16 03:54:00 -05001466 if (flag & O_NOATIME && !is_owner_or_cap(inode))
1467 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468
1469 /*
1470 * Ensure there are no outstanding leases on the file.
1471 */
Al Virob65a9cf2009-12-16 06:27:40 -05001472 return break_lease(inode, flag);
Al Viro7715b522009-12-16 03:54:00 -05001473}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474
Al Viro7715b522009-12-16 03:54:00 -05001475static int handle_truncate(struct path *path)
1476{
1477 struct inode *inode = path->dentry->d_inode;
1478 int error = get_write_access(inode);
1479 if (error)
1480 return error;
1481 /*
1482 * Refuse to truncate files with mandatory locks held on them.
1483 */
1484 error = locks_verify_locked(inode);
1485 if (!error)
1486 error = security_path_truncate(path, 0,
1487 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN);
1488 if (!error) {
1489 error = do_truncate(path->dentry, 0,
1490 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1491 NULL);
1492 }
1493 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04001494 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495}
1496
Dave Hansend57999e2008-02-15 14:37:27 -08001497/*
1498 * Be careful about ever adding any more callers of this
1499 * function. Its flags must be in the namei format, not
1500 * what get passed to sys_open().
1501 */
1502static int __open_namei_create(struct nameidata *nd, struct path *path,
Al Viro8737c932009-12-24 06:47:55 -05001503 int open_flag, int mode)
Dave Hansenaab520e2006-09-30 23:29:02 -07001504{
1505 int error;
Jan Blunck4ac91372008-02-14 19:34:32 -08001506 struct dentry *dir = nd->path.dentry;
Dave Hansenaab520e2006-09-30 23:29:02 -07001507
1508 if (!IS_POSIXACL(dir->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04001509 mode &= ~current_umask();
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001510 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1511 if (error)
1512 goto out_unlock;
Dave Hansenaab520e2006-09-30 23:29:02 -07001513 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001514out_unlock:
Dave Hansenaab520e2006-09-30 23:29:02 -07001515 mutex_unlock(&dir->d_inode->i_mutex);
Jan Blunck4ac91372008-02-14 19:34:32 -08001516 dput(nd->path.dentry);
1517 nd->path.dentry = path->dentry;
Dave Hansenaab520e2006-09-30 23:29:02 -07001518 if (error)
1519 return error;
1520 /* Don't check for write permission, don't truncate */
Al Viro8737c932009-12-24 06:47:55 -05001521 return may_open(&nd->path, 0, open_flag & ~O_TRUNC);
Dave Hansenaab520e2006-09-30 23:29:02 -07001522}
1523
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524/*
Dave Hansend57999e2008-02-15 14:37:27 -08001525 * Note that while the flag value (low two bits) for sys_open means:
1526 * 00 - read-only
1527 * 01 - write-only
1528 * 10 - read-write
1529 * 11 - special
1530 * it is changed into
1531 * 00 - no permissions needed
1532 * 01 - read-permission
1533 * 10 - write-permission
1534 * 11 - read-write
1535 * for the internal routines (ie open_namei()/follow_link() etc)
1536 * This is more logical, and also allows the 00 "no perm needed"
1537 * to be used for symlinks (where the permissions are checked
1538 * later).
1539 *
1540*/
1541static inline int open_to_namei_flags(int flag)
1542{
1543 if ((flag+1) & O_ACCMODE)
1544 flag++;
1545 return flag;
1546}
1547
Al Viro7715b522009-12-16 03:54:00 -05001548static int open_will_truncate(int flag, struct inode *inode)
Dave Hansen4a3fd212008-02-15 14:37:48 -08001549{
1550 /*
1551 * We'll never write to the fs underlying
1552 * a device file.
1553 */
1554 if (special_file(inode->i_mode))
1555 return 0;
1556 return (flag & O_TRUNC);
1557}
1558
Al Viro648fa862009-12-24 01:26:48 -05001559static struct file *finish_open(struct nameidata *nd,
Al Viro9a661792009-12-24 06:49:47 -05001560 int open_flag, int acc_mode)
Al Viro648fa862009-12-24 01:26:48 -05001561{
1562 struct file *filp;
1563 int will_truncate;
1564 int error;
1565
Al Viro9a661792009-12-24 06:49:47 -05001566 will_truncate = open_will_truncate(open_flag, nd->path.dentry->d_inode);
Al Viro648fa862009-12-24 01:26:48 -05001567 if (will_truncate) {
1568 error = mnt_want_write(nd->path.mnt);
1569 if (error)
1570 goto exit;
1571 }
1572 error = may_open(&nd->path, acc_mode, open_flag);
1573 if (error) {
1574 if (will_truncate)
1575 mnt_drop_write(nd->path.mnt);
1576 goto exit;
1577 }
1578 filp = nameidata_to_filp(nd);
1579 if (!IS_ERR(filp)) {
1580 error = ima_file_check(filp, acc_mode);
1581 if (error) {
1582 fput(filp);
1583 filp = ERR_PTR(error);
1584 }
1585 }
1586 if (!IS_ERR(filp)) {
Al Viro648fa862009-12-24 01:26:48 -05001587 if (will_truncate) {
1588 error = handle_truncate(&nd->path);
1589 if (error) {
1590 fput(filp);
1591 filp = ERR_PTR(error);
1592 }
1593 }
1594 }
1595 /*
1596 * It is now safe to drop the mnt write
1597 * because the filp has had a write taken
1598 * on its behalf.
1599 */
1600 if (will_truncate)
1601 mnt_drop_write(nd->path.mnt);
1602 return filp;
1603
1604exit:
1605 if (!IS_ERR(nd->intent.open.file))
1606 release_open_intent(nd);
1607 path_put(&nd->path);
1608 return ERR_PTR(error);
1609}
1610
Al Virofb1cc552009-12-24 01:58:28 -05001611static struct file *do_last(struct nameidata *nd, struct path *path,
Al Viro5b369df2009-12-24 06:51:13 -05001612 int open_flag, int acc_mode,
Al Viro3e297b62010-03-26 12:40:13 -04001613 int mode, const char *pathname)
Al Virofb1cc552009-12-24 01:58:28 -05001614{
Al Viroa1e28032009-12-24 02:12:06 -05001615 struct dentry *dir = nd->path.dentry;
Al Virofb1cc552009-12-24 01:58:28 -05001616 struct file *filp;
Al Viro1f36f772009-12-26 10:56:19 -05001617 int error = -EISDIR;
Al Virofb1cc552009-12-24 01:58:28 -05001618
Al Viro1f36f772009-12-26 10:56:19 -05001619 switch (nd->last_type) {
1620 case LAST_DOTDOT:
1621 follow_dotdot(nd);
1622 dir = nd->path.dentry;
1623 if (nd->path.mnt->mnt_sb->s_type->fs_flags & FS_REVAL_DOT) {
1624 if (!dir->d_op->d_revalidate(dir, nd)) {
1625 error = -ESTALE;
1626 goto exit;
1627 }
1628 }
1629 /* fallthrough */
1630 case LAST_DOT:
1631 case LAST_ROOT:
1632 if (open_flag & O_CREAT)
1633 goto exit;
1634 /* fallthrough */
1635 case LAST_BIND:
1636 audit_inode(pathname, dir);
Al Viro67ee3ad2009-12-26 07:01:01 -05001637 goto ok;
Al Viro1f36f772009-12-26 10:56:19 -05001638 }
Al Viro67ee3ad2009-12-26 07:01:01 -05001639
Al Viro1f36f772009-12-26 10:56:19 -05001640 /* trailing slashes? */
1641 if (nd->last.name[nd->last.len]) {
1642 if (open_flag & O_CREAT)
1643 goto exit;
Jan Kara002baee2010-05-13 12:52:57 +02001644 nd->flags |= LOOKUP_DIRECTORY | LOOKUP_FOLLOW;
Al Viro1f36f772009-12-26 10:56:19 -05001645 }
Al Viroa2c36b42009-12-24 03:39:50 -05001646
Al Viro1f36f772009-12-26 10:56:19 -05001647 /* just plain open? */
1648 if (!(open_flag & O_CREAT)) {
1649 error = do_lookup(nd, &nd->last, path);
1650 if (error)
1651 goto exit;
1652 error = -ENOENT;
1653 if (!path->dentry->d_inode)
1654 goto exit_dput;
1655 if (path->dentry->d_inode->i_op->follow_link)
1656 return NULL;
1657 error = -ENOTDIR;
Al Viro3e297b62010-03-26 12:40:13 -04001658 if (nd->flags & LOOKUP_DIRECTORY) {
1659 if (!path->dentry->d_inode->i_op->lookup)
1660 goto exit_dput;
1661 }
Al Viro1f36f772009-12-26 10:56:19 -05001662 path_to_nameidata(path, nd);
1663 audit_inode(pathname, nd->path.dentry);
1664 goto ok;
1665 }
1666
1667 /* OK, it's O_CREAT */
Al Viroa1e28032009-12-24 02:12:06 -05001668 mutex_lock(&dir->d_inode->i_mutex);
1669
1670 path->dentry = lookup_hash(nd);
1671 path->mnt = nd->path.mnt;
1672
Al Virofb1cc552009-12-24 01:58:28 -05001673 error = PTR_ERR(path->dentry);
1674 if (IS_ERR(path->dentry)) {
1675 mutex_unlock(&dir->d_inode->i_mutex);
1676 goto exit;
1677 }
1678
1679 if (IS_ERR(nd->intent.open.file)) {
1680 error = PTR_ERR(nd->intent.open.file);
1681 goto exit_mutex_unlock;
1682 }
1683
1684 /* Negative dentry, just create the file */
1685 if (!path->dentry->d_inode) {
1686 /*
1687 * This write is needed to ensure that a
1688 * ro->rw transition does not occur between
1689 * the time when the file is created and when
1690 * a permanent write count is taken through
1691 * the 'struct file' in nameidata_to_filp().
1692 */
1693 error = mnt_want_write(nd->path.mnt);
1694 if (error)
1695 goto exit_mutex_unlock;
1696 error = __open_namei_create(nd, path, open_flag, mode);
1697 if (error) {
1698 mnt_drop_write(nd->path.mnt);
1699 goto exit;
1700 }
1701 filp = nameidata_to_filp(nd);
1702 mnt_drop_write(nd->path.mnt);
Al Virofb1cc552009-12-24 01:58:28 -05001703 if (!IS_ERR(filp)) {
1704 error = ima_file_check(filp, acc_mode);
1705 if (error) {
1706 fput(filp);
1707 filp = ERR_PTR(error);
1708 }
1709 }
1710 return filp;
1711 }
1712
1713 /*
1714 * It already exists.
1715 */
1716 mutex_unlock(&dir->d_inode->i_mutex);
1717 audit_inode(pathname, path->dentry);
1718
1719 error = -EEXIST;
Al Viro5b369df2009-12-24 06:51:13 -05001720 if (open_flag & O_EXCL)
Al Virofb1cc552009-12-24 01:58:28 -05001721 goto exit_dput;
1722
1723 if (__follow_mount(path)) {
1724 error = -ELOOP;
Al Viro5b369df2009-12-24 06:51:13 -05001725 if (open_flag & O_NOFOLLOW)
Al Virofb1cc552009-12-24 01:58:28 -05001726 goto exit_dput;
1727 }
1728
1729 error = -ENOENT;
1730 if (!path->dentry->d_inode)
1731 goto exit_dput;
Al Viro9e67f362009-12-26 07:04:50 -05001732
1733 if (path->dentry->d_inode->i_op->follow_link)
Al Virofb1cc552009-12-24 01:58:28 -05001734 return NULL;
Al Virofb1cc552009-12-24 01:58:28 -05001735
1736 path_to_nameidata(path, nd);
1737 error = -EISDIR;
1738 if (S_ISDIR(path->dentry->d_inode->i_mode))
1739 goto exit;
Al Viro67ee3ad2009-12-26 07:01:01 -05001740ok:
Al Viro9a661792009-12-24 06:49:47 -05001741 filp = finish_open(nd, open_flag, acc_mode);
Al Virofb1cc552009-12-24 01:58:28 -05001742 return filp;
1743
1744exit_mutex_unlock:
1745 mutex_unlock(&dir->d_inode->i_mutex);
1746exit_dput:
1747 path_put_conditional(path, nd);
1748exit:
1749 if (!IS_ERR(nd->intent.open.file))
1750 release_open_intent(nd);
Al Virofb1cc552009-12-24 01:58:28 -05001751 path_put(&nd->path);
1752 return ERR_PTR(error);
1753}
1754
Dave Hansend57999e2008-02-15 14:37:27 -08001755/*
Dave Hansen4a3fd212008-02-15 14:37:48 -08001756 * Note that the low bits of the passed in "open_flag"
1757 * are not the same as in the local variable "flag". See
1758 * open_to_namei_flags() for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001760struct file *do_filp_open(int dfd, const char *pathname,
Al Viro6e8341a2009-04-06 11:16:22 -04001761 int open_flag, int mode, int acc_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762{
Dave Hansen4a3fd212008-02-15 14:37:48 -08001763 struct file *filp;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001764 struct nameidata nd;
Al Viro6e8341a2009-04-06 11:16:22 -04001765 int error;
Al Viro9850c052010-01-13 15:01:15 -05001766 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767 int count = 0;
Dave Hansend57999e2008-02-15 14:37:27 -08001768 int flag = open_to_namei_flags(open_flag);
Al Viro9850c052010-01-13 15:01:15 -05001769 int force_reval = 0;
Al Viro1f36f772009-12-26 10:56:19 -05001770
1771 if (!(open_flag & O_CREAT))
1772 mode = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773
Christoph Hellwig6b2f3d12009-10-27 11:05:28 +01001774 /*
1775 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1776 * check for O_DSYNC if the need any syncing at all we enforce it's
1777 * always set instead of having to deal with possibly weird behaviour
1778 * for malicious applications setting only __O_SYNC.
1779 */
1780 if (open_flag & __O_SYNC)
1781 open_flag |= O_DSYNC;
1782
Al Viro6e8341a2009-04-06 11:16:22 -04001783 if (!acc_mode)
Al Viro6d125522009-12-24 06:58:56 -05001784 acc_mode = MAY_OPEN | ACC_MODE(open_flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785
Trond Myklebust834f2a42005-10-18 14:20:16 -07001786 /* O_TRUNC implies we need access checks for write permissions */
Al Viro4296e2c2009-12-24 07:15:41 -05001787 if (open_flag & O_TRUNC)
Trond Myklebust834f2a42005-10-18 14:20:16 -07001788 acc_mode |= MAY_WRITE;
1789
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 /* Allow the LSM permission hook to distinguish append
1791 access from general write access. */
Al Viro4296e2c2009-12-24 07:15:41 -05001792 if (open_flag & O_APPEND)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 acc_mode |= MAY_APPEND;
1794
Al Viro1f36f772009-12-26 10:56:19 -05001795 /* find the parent */
Al Viro9850c052010-01-13 15:01:15 -05001796reval:
Al Viro9b4a9b12009-04-07 11:44:16 -04001797 error = path_init(dfd, pathname, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 if (error)
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001799 return ERR_PTR(error);
Al Viro9850c052010-01-13 15:01:15 -05001800 if (force_reval)
1801 nd.flags |= LOOKUP_REVAL;
Al Viro38662482009-12-26 07:21:48 -05001802
1803 current->total_link_count = 0;
1804 error = link_path_walk(pathname, &nd);
J. R. Okajima654f5622009-06-18 23:30:15 +09001805 if (error) {
Al Viro10fa8e62009-12-26 07:09:49 -05001806 filp = ERR_PTR(error);
1807 goto out;
J. R. Okajima654f5622009-06-18 23:30:15 +09001808 }
Al Viro1f36f772009-12-26 10:56:19 -05001809 if (unlikely(!audit_dummy_context()) && (open_flag & O_CREAT))
Al Viro9b4a9b12009-04-07 11:44:16 -04001810 audit_inode(pathname, nd.path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811
1812 /*
Al Viroa2c36b42009-12-24 03:39:50 -05001813 * We have the parent and last component.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815
Al Viro8737f3a2008-08-02 22:36:57 -04001816 error = -ENFILE;
1817 filp = get_empty_filp();
1818 if (filp == NULL)
1819 goto exit_parent;
1820 nd.intent.open.file = filp;
Al Viro482928d2009-12-19 10:10:39 -05001821 filp->f_flags = open_flag;
Al Viro8737f3a2008-08-02 22:36:57 -04001822 nd.intent.open.flags = flag;
1823 nd.intent.open.create_mode = mode;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001824 nd.flags &= ~LOOKUP_PARENT;
Al Viro1f36f772009-12-26 10:56:19 -05001825 nd.flags |= LOOKUP_OPEN;
1826 if (open_flag & O_CREAT) {
1827 nd.flags |= LOOKUP_CREATE;
1828 if (open_flag & O_EXCL)
1829 nd.flags |= LOOKUP_EXCL;
1830 }
Al Viro3e297b62010-03-26 12:40:13 -04001831 if (open_flag & O_DIRECTORY)
1832 nd.flags |= LOOKUP_DIRECTORY;
Jan Kara002baee2010-05-13 12:52:57 +02001833 if (!(open_flag & O_NOFOLLOW))
1834 nd.flags |= LOOKUP_FOLLOW;
Al Viro3e297b62010-03-26 12:40:13 -04001835 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
Al Viro806b6812009-12-26 07:16:40 -05001836 while (unlikely(!filp)) { /* trailing symlink */
Al Virodef4af32009-12-26 08:37:05 -05001837 struct path holder;
Al Viro1f36f772009-12-26 10:56:19 -05001838 struct inode *inode = path.dentry->d_inode;
Al Virodef4af32009-12-26 08:37:05 -05001839 void *cookie;
Al Viro806b6812009-12-26 07:16:40 -05001840 error = -ELOOP;
Al Viro1f36f772009-12-26 10:56:19 -05001841 /* S_ISDIR part is a temporary automount kludge */
Jan Kara002baee2010-05-13 12:52:57 +02001842 if (!(nd.flags & LOOKUP_FOLLOW) && !S_ISDIR(inode->i_mode))
Al Viro1f36f772009-12-26 10:56:19 -05001843 goto exit_dput;
1844 if (count++ == 32)
Al Viro806b6812009-12-26 07:16:40 -05001845 goto exit_dput;
1846 /*
1847 * This is subtle. Instead of calling do_follow_link() we do
1848 * the thing by hands. The reason is that this way we have zero
1849 * link_count and path_walk() (called from ->follow_link)
1850 * honoring LOOKUP_PARENT. After that we have the parent and
1851 * last component, i.e. we are in the same situation as after
1852 * the first path_walk(). Well, almost - if the last component
1853 * is normal we get its copy stored in nd->last.name and we will
1854 * have to putname() it when we are done. Procfs-like symlinks
1855 * just set LAST_BIND.
1856 */
1857 nd.flags |= LOOKUP_PARENT;
1858 error = security_inode_follow_link(path.dentry, &nd);
1859 if (error)
1860 goto exit_dput;
Al Virodef4af32009-12-26 08:37:05 -05001861 error = __do_follow_link(&path, &nd, &cookie);
1862 if (unlikely(error)) {
Al Viro806b6812009-12-26 07:16:40 -05001863 /* nd.path had been dropped */
Al Virodef4af32009-12-26 08:37:05 -05001864 if (!IS_ERR(cookie) && inode->i_op->put_link)
1865 inode->i_op->put_link(path.dentry, &nd, cookie);
1866 path_put(&path);
Al Viro806b6812009-12-26 07:16:40 -05001867 release_open_intent(&nd);
1868 filp = ERR_PTR(error);
1869 goto out;
1870 }
Al Virodef4af32009-12-26 08:37:05 -05001871 holder = path;
Al Viro806b6812009-12-26 07:16:40 -05001872 nd.flags &= ~LOOKUP_PARENT;
Al Viro3e297b62010-03-26 12:40:13 -04001873 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
Al Virodef4af32009-12-26 08:37:05 -05001874 if (inode->i_op->put_link)
1875 inode->i_op->put_link(holder.dentry, &nd, cookie);
1876 path_put(&holder);
Al Viro806b6812009-12-26 07:16:40 -05001877 }
Al Viro10fa8e62009-12-26 07:09:49 -05001878out:
Al Viro2a737872009-04-07 11:49:53 -04001879 if (nd.root.mnt)
1880 path_put(&nd.root);
Al Viro10fa8e62009-12-26 07:09:49 -05001881 if (filp == ERR_PTR(-ESTALE) && !force_reval) {
1882 force_reval = 1;
1883 goto reval;
1884 }
1885 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886
Al Viro806b6812009-12-26 07:16:40 -05001887exit_dput:
1888 path_put_conditional(&path, &nd);
1889 if (!IS_ERR(nd.intent.open.file))
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001890 release_open_intent(&nd);
Al Viro806b6812009-12-26 07:16:40 -05001891exit_parent:
1892 path_put(&nd.path);
1893 filp = ERR_PTR(error);
Al Viro10fa8e62009-12-26 07:09:49 -05001894 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895}
1896
1897/**
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001898 * filp_open - open file and return file pointer
1899 *
1900 * @filename: path to open
1901 * @flags: open flags as per the open(2) second argument
1902 * @mode: mode for the new file if O_CREAT is set, else ignored
1903 *
1904 * This is the helper to open a file from kernelspace if you really
1905 * have to. But in generally you should not do this, so please move
1906 * along, nothing to see here..
1907 */
1908struct file *filp_open(const char *filename, int flags, int mode)
1909{
Al Viro6e8341a2009-04-06 11:16:22 -04001910 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001911}
1912EXPORT_SYMBOL(filp_open);
1913
1914/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915 * lookup_create - lookup a dentry, creating it if it doesn't exist
1916 * @nd: nameidata info
1917 * @is_dir: directory flag
1918 *
1919 * Simple function to lookup and return a dentry and create it
1920 * if it doesn't exist. Is SMP-safe.
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001921 *
Jan Blunck4ac91372008-02-14 19:34:32 -08001922 * Returns with nd->path.dentry->d_inode->i_mutex locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923 */
1924struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1925{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001926 struct dentry *dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927
Jan Blunck4ac91372008-02-14 19:34:32 -08001928 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001929 /*
1930 * Yucky last component or no last component at all?
1931 * (foo/., foo/.., /////)
1932 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933 if (nd->last_type != LAST_NORM)
1934 goto fail;
1935 nd->flags &= ~LOOKUP_PARENT;
Al Viro35165862008-08-05 03:00:49 -04001936 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
ASANO Masahiroa6349042006-08-22 20:06:02 -04001937 nd->intent.open.flags = O_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001938
1939 /*
1940 * Do the final lookup.
1941 */
Christoph Hellwig49705b72005-11-08 21:35:06 -08001942 dentry = lookup_hash(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943 if (IS_ERR(dentry))
1944 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001945
Al Viroe9baf6e2008-05-15 04:49:12 -04001946 if (dentry->d_inode)
1947 goto eexist;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001948 /*
1949 * Special case - lookup gave negative, but... we had foo/bar/
1950 * From the vfs_mknod() POV we just have a negative dentry -
1951 * all is fine. Let's be bastards - you had / on the end, you've
1952 * been asking for (non-existent) directory. -ENOENT for you.
1953 */
Al Viroe9baf6e2008-05-15 04:49:12 -04001954 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1955 dput(dentry);
1956 dentry = ERR_PTR(-ENOENT);
1957 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 return dentry;
Al Viroe9baf6e2008-05-15 04:49:12 -04001959eexist:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 dput(dentry);
Al Viroe9baf6e2008-05-15 04:49:12 -04001961 dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962fail:
1963 return dentry;
1964}
Christoph Hellwigf81a0bf2005-05-19 12:26:43 -07001965EXPORT_SYMBOL_GPL(lookup_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966
1967int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1968{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001969 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970
1971 if (error)
1972 return error;
1973
1974 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1975 return -EPERM;
1976
Al Viroacfa4382008-12-04 10:06:33 -05001977 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978 return -EPERM;
1979
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07001980 error = devcgroup_inode_mknod(mode, dev);
1981 if (error)
1982 return error;
1983
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 error = security_inode_mknod(dir, dentry, mode, dev);
1985 if (error)
1986 return error;
1987
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001989 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001990 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991 return error;
1992}
1993
Dave Hansen463c3192008-02-15 14:37:57 -08001994static int may_mknod(mode_t mode)
1995{
1996 switch (mode & S_IFMT) {
1997 case S_IFREG:
1998 case S_IFCHR:
1999 case S_IFBLK:
2000 case S_IFIFO:
2001 case S_IFSOCK:
2002 case 0: /* zero mode translates to S_IFREG */
2003 return 0;
2004 case S_IFDIR:
2005 return -EPERM;
2006 default:
2007 return -EINVAL;
2008 }
2009}
2010
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002011SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2012 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013{
Al Viro2ad94ae2008-07-21 09:32:51 -04002014 int error;
2015 char *tmp;
2016 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 struct nameidata nd;
2018
2019 if (S_ISDIR(mode))
2020 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021
Al Viro2ad94ae2008-07-21 09:32:51 -04002022 error = user_path_parent(dfd, filename, &nd, &tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002024 return error;
2025
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026 dentry = lookup_create(&nd, 0);
Dave Hansen463c3192008-02-15 14:37:57 -08002027 if (IS_ERR(dentry)) {
2028 error = PTR_ERR(dentry);
2029 goto out_unlock;
2030 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002031 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002032 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002033 error = may_mknod(mode);
2034 if (error)
2035 goto out_dput;
2036 error = mnt_want_write(nd.path.mnt);
2037 if (error)
2038 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002039 error = security_path_mknod(&nd.path, dentry, mode, dev);
2040 if (error)
2041 goto out_drop_write;
Dave Hansen463c3192008-02-15 14:37:57 -08002042 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 case 0: case S_IFREG:
Jan Blunck4ac91372008-02-14 19:34:32 -08002044 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 break;
2046 case S_IFCHR: case S_IFBLK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002047 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 new_decode_dev(dev));
2049 break;
2050 case S_IFIFO: case S_IFSOCK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002051 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002054out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002055 mnt_drop_write(nd.path.mnt);
2056out_dput:
2057 dput(dentry);
2058out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002059 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002060 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 putname(tmp);
2062
2063 return error;
2064}
2065
Heiko Carstens3480b252009-01-14 14:14:16 +01002066SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002067{
2068 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2069}
2070
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2072{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002073 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074
2075 if (error)
2076 return error;
2077
Al Viroacfa4382008-12-04 10:06:33 -05002078 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079 return -EPERM;
2080
2081 mode &= (S_IRWXUGO|S_ISVTX);
2082 error = security_inode_mkdir(dir, dentry, mode);
2083 if (error)
2084 return error;
2085
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002087 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002088 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089 return error;
2090}
2091
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002092SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093{
2094 int error = 0;
2095 char * tmp;
Dave Hansen6902d922006-09-30 23:29:01 -07002096 struct dentry *dentry;
2097 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098
Al Viro2ad94ae2008-07-21 09:32:51 -04002099 error = user_path_parent(dfd, pathname, &nd, &tmp);
2100 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002101 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102
Dave Hansen6902d922006-09-30 23:29:01 -07002103 dentry = lookup_create(&nd, 1);
2104 error = PTR_ERR(dentry);
2105 if (IS_ERR(dentry))
2106 goto out_unlock;
2107
Jan Blunck4ac91372008-02-14 19:34:32 -08002108 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002109 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002110 error = mnt_want_write(nd.path.mnt);
2111 if (error)
2112 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002113 error = security_path_mkdir(&nd.path, dentry, mode);
2114 if (error)
2115 goto out_drop_write;
Jan Blunck4ac91372008-02-14 19:34:32 -08002116 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002117out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002118 mnt_drop_write(nd.path.mnt);
2119out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002120 dput(dentry);
2121out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002122 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002123 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002124 putname(tmp);
2125out_err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 return error;
2127}
2128
Heiko Carstens3cdad422009-01-14 14:14:22 +01002129SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002130{
2131 return sys_mkdirat(AT_FDCWD, pathname, mode);
2132}
2133
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134/*
2135 * We try to drop the dentry early: we should have
2136 * a usage count of 2 if we're the only user of this
2137 * dentry, and if that is true (possibly after pruning
2138 * the dcache), then we drop the dentry now.
2139 *
2140 * A low-level filesystem can, if it choses, legally
2141 * do a
2142 *
2143 * if (!d_unhashed(dentry))
2144 * return -EBUSY;
2145 *
2146 * if it cannot handle the case of removing a directory
2147 * that is still in use by something else..
2148 */
2149void dentry_unhash(struct dentry *dentry)
2150{
2151 dget(dentry);
Vasily Averindc168422006-12-06 20:37:07 -08002152 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153 spin_lock(&dcache_lock);
2154 spin_lock(&dentry->d_lock);
2155 if (atomic_read(&dentry->d_count) == 2)
2156 __d_drop(dentry);
2157 spin_unlock(&dentry->d_lock);
2158 spin_unlock(&dcache_lock);
2159}
2160
2161int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2162{
2163 int error = may_delete(dir, dentry, 1);
2164
2165 if (error)
2166 return error;
2167
Al Viroacfa4382008-12-04 10:06:33 -05002168 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169 return -EPERM;
2170
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002171 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172 dentry_unhash(dentry);
2173 if (d_mountpoint(dentry))
2174 error = -EBUSY;
2175 else {
2176 error = security_inode_rmdir(dir, dentry);
2177 if (!error) {
2178 error = dir->i_op->rmdir(dir, dentry);
Al Virod83c49f2010-04-30 17:17:09 -04002179 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 dentry->d_inode->i_flags |= S_DEAD;
Al Virod83c49f2010-04-30 17:17:09 -04002181 dont_mount(dentry);
2182 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183 }
2184 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002185 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187 d_delete(dentry);
2188 }
2189 dput(dentry);
2190
2191 return error;
2192}
2193
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002194static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195{
2196 int error = 0;
2197 char * name;
2198 struct dentry *dentry;
2199 struct nameidata nd;
2200
Al Viro2ad94ae2008-07-21 09:32:51 -04002201 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002203 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204
2205 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002206 case LAST_DOTDOT:
2207 error = -ENOTEMPTY;
2208 goto exit1;
2209 case LAST_DOT:
2210 error = -EINVAL;
2211 goto exit1;
2212 case LAST_ROOT:
2213 error = -EBUSY;
2214 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002216
2217 nd.flags &= ~LOOKUP_PARENT;
2218
Jan Blunck4ac91372008-02-14 19:34:32 -08002219 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002220 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002222 if (IS_ERR(dentry))
2223 goto exit2;
Dave Hansen06227532008-02-15 14:37:34 -08002224 error = mnt_want_write(nd.path.mnt);
2225 if (error)
2226 goto exit3;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002227 error = security_path_rmdir(&nd.path, dentry);
2228 if (error)
2229 goto exit4;
Jan Blunck4ac91372008-02-14 19:34:32 -08002230 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002231exit4:
Dave Hansen06227532008-02-15 14:37:34 -08002232 mnt_drop_write(nd.path.mnt);
2233exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07002234 dput(dentry);
2235exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08002236 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002238 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239 putname(name);
2240 return error;
2241}
2242
Heiko Carstens3cdad422009-01-14 14:14:22 +01002243SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002244{
2245 return do_rmdir(AT_FDCWD, pathname);
2246}
2247
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248int vfs_unlink(struct inode *dir, struct dentry *dentry)
2249{
2250 int error = may_delete(dir, dentry, 0);
2251
2252 if (error)
2253 return error;
2254
Al Viroacfa4382008-12-04 10:06:33 -05002255 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256 return -EPERM;
2257
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002258 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259 if (d_mountpoint(dentry))
2260 error = -EBUSY;
2261 else {
2262 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002263 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264 error = dir->i_op->unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002265 if (!error)
Al Virod83c49f2010-04-30 17:17:09 -04002266 dont_mount(dentry);
Al Virobec10522010-03-03 14:12:08 -05002267 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002269 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270
2271 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2272 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08002273 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275 }
Robert Love0eeca282005-07-12 17:06:03 -04002276
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277 return error;
2278}
2279
2280/*
2281 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002282 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283 * writeout happening, and we don't want to prevent access to the directory
2284 * while waiting on the I/O.
2285 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002286static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287{
Al Viro2ad94ae2008-07-21 09:32:51 -04002288 int error;
2289 char *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290 struct dentry *dentry;
2291 struct nameidata nd;
2292 struct inode *inode = NULL;
2293
Al Viro2ad94ae2008-07-21 09:32:51 -04002294 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002296 return error;
2297
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298 error = -EISDIR;
2299 if (nd.last_type != LAST_NORM)
2300 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002301
2302 nd.flags &= ~LOOKUP_PARENT;
2303
Jan Blunck4ac91372008-02-14 19:34:32 -08002304 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002305 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306 error = PTR_ERR(dentry);
2307 if (!IS_ERR(dentry)) {
2308 /* Why not before? Because we want correct error value */
2309 if (nd.last.name[nd.last.len])
2310 goto slashes;
2311 inode = dentry->d_inode;
2312 if (inode)
2313 atomic_inc(&inode->i_count);
Dave Hansen06227532008-02-15 14:37:34 -08002314 error = mnt_want_write(nd.path.mnt);
2315 if (error)
2316 goto exit2;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002317 error = security_path_unlink(&nd.path, dentry);
2318 if (error)
2319 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08002320 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002321exit3:
Dave Hansen06227532008-02-15 14:37:34 -08002322 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323 exit2:
2324 dput(dentry);
2325 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002326 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327 if (inode)
2328 iput(inode); /* truncate the inode here */
2329exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002330 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331 putname(name);
2332 return error;
2333
2334slashes:
2335 error = !dentry->d_inode ? -ENOENT :
2336 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2337 goto exit2;
2338}
2339
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002340SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002341{
2342 if ((flag & ~AT_REMOVEDIR) != 0)
2343 return -EINVAL;
2344
2345 if (flag & AT_REMOVEDIR)
2346 return do_rmdir(dfd, pathname);
2347
2348 return do_unlinkat(dfd, pathname);
2349}
2350
Heiko Carstens3480b252009-01-14 14:14:16 +01002351SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002352{
2353 return do_unlinkat(AT_FDCWD, pathname);
2354}
2355
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002356int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002358 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359
2360 if (error)
2361 return error;
2362
Al Viroacfa4382008-12-04 10:06:33 -05002363 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364 return -EPERM;
2365
2366 error = security_inode_symlink(dir, dentry, oldname);
2367 if (error)
2368 return error;
2369
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002371 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002372 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373 return error;
2374}
2375
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002376SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2377 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378{
Al Viro2ad94ae2008-07-21 09:32:51 -04002379 int error;
2380 char *from;
2381 char *to;
Dave Hansen6902d922006-09-30 23:29:01 -07002382 struct dentry *dentry;
2383 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384
2385 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04002386 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387 return PTR_ERR(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002388
2389 error = user_path_parent(newdfd, newname, &nd, &to);
2390 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002391 goto out_putname;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392
Dave Hansen6902d922006-09-30 23:29:01 -07002393 dentry = lookup_create(&nd, 0);
2394 error = PTR_ERR(dentry);
2395 if (IS_ERR(dentry))
2396 goto out_unlock;
2397
Dave Hansen75c3f292008-02-15 14:37:45 -08002398 error = mnt_want_write(nd.path.mnt);
2399 if (error)
2400 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002401 error = security_path_symlink(&nd.path, dentry, from);
2402 if (error)
2403 goto out_drop_write;
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002404 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002405out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002406 mnt_drop_write(nd.path.mnt);
2407out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002408 dput(dentry);
2409out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002410 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002411 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002412 putname(to);
2413out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414 putname(from);
2415 return error;
2416}
2417
Heiko Carstens3480b252009-01-14 14:14:16 +01002418SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002419{
2420 return sys_symlinkat(oldname, AT_FDCWD, newname);
2421}
2422
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2424{
2425 struct inode *inode = old_dentry->d_inode;
2426 int error;
2427
2428 if (!inode)
2429 return -ENOENT;
2430
Miklos Szeredia95164d2008-07-30 15:08:48 +02002431 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002432 if (error)
2433 return error;
2434
2435 if (dir->i_sb != inode->i_sb)
2436 return -EXDEV;
2437
2438 /*
2439 * A link to an append-only or immutable file cannot be created.
2440 */
2441 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2442 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05002443 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002445 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446 return -EPERM;
2447
2448 error = security_inode_link(old_dentry, dir, new_dentry);
2449 if (error)
2450 return error;
2451
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002452 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453 error = dir->i_op->link(old_dentry, dir, new_dentry);
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002454 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002455 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002456 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457 return error;
2458}
2459
2460/*
2461 * Hardlinks are often used in delicate situations. We avoid
2462 * security-related surprises by not following symlinks on the
2463 * newname. --KAB
2464 *
2465 * We don't follow them on the oldname either to be compatible
2466 * with linux 2.0, and to avoid hard-linking to directories
2467 * and other special files. --ADM
2468 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002469SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2470 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471{
2472 struct dentry *new_dentry;
Al Viro2d8f3032008-07-22 09:59:21 -04002473 struct nameidata nd;
2474 struct path old_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475 int error;
Al Viro2ad94ae2008-07-21 09:32:51 -04002476 char *to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477
Ulrich Drepper45c9b112006-06-25 05:49:11 -07002478 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002479 return -EINVAL;
2480
Al Viro2d8f3032008-07-22 09:59:21 -04002481 error = user_path_at(olddfd, oldname,
2482 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2483 &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002485 return error;
2486
2487 error = user_path_parent(newdfd, newname, &nd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488 if (error)
2489 goto out;
2490 error = -EXDEV;
Al Viro2d8f3032008-07-22 09:59:21 -04002491 if (old_path.mnt != nd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492 goto out_release;
2493 new_dentry = lookup_create(&nd, 0);
2494 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002495 if (IS_ERR(new_dentry))
2496 goto out_unlock;
Dave Hansen75c3f292008-02-15 14:37:45 -08002497 error = mnt_want_write(nd.path.mnt);
2498 if (error)
2499 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002500 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2501 if (error)
2502 goto out_drop_write;
Al Viro2d8f3032008-07-22 09:59:21 -04002503 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002504out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002505 mnt_drop_write(nd.path.mnt);
2506out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002507 dput(new_dentry);
2508out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002509 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510out_release:
Jan Blunck1d957f92008-02-14 19:34:35 -08002511 path_put(&nd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002512 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513out:
Al Viro2d8f3032008-07-22 09:59:21 -04002514 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515
2516 return error;
2517}
2518
Heiko Carstens3480b252009-01-14 14:14:16 +01002519SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002520{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002521 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002522}
2523
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524/*
2525 * The worst of all namespace operations - renaming directory. "Perverted"
2526 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2527 * Problems:
2528 * a) we can get into loop creation. Check is done in is_subdir().
2529 * b) race potential - two innocent renames can create a loop together.
2530 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002531 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532 * story.
2533 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002534 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535 * whether the target exists). Solution: try to be smart with locking
2536 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002537 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 * move will be locked. Thus we can rank directories by the tree
2539 * (ancestors first) and rank all non-directories after them.
2540 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002541 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542 * HOWEVER, it relies on the assumption that any object with ->lookup()
2543 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2544 * we'd better make sure that there's no link(2) for them.
2545 * d) some filesystems don't support opened-but-unlinked directories,
2546 * either because of layout or because they are not ready to deal with
2547 * all cases correctly. The latter will be fixed (taking this sort of
2548 * stuff into VFS), but the former is not going away. Solution: the same
2549 * trick as in rmdir().
2550 * e) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002551 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05002553 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554 * locking].
2555 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07002556static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2557 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558{
2559 int error = 0;
2560 struct inode *target;
2561
2562 /*
2563 * If we are going to change the parent - check write permissions,
2564 * we'll need to flip '..'.
2565 */
2566 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04002567 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568 if (error)
2569 return error;
2570 }
2571
2572 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2573 if (error)
2574 return error;
2575
2576 target = new_dentry->d_inode;
Al Virod83c49f2010-04-30 17:17:09 -04002577 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002578 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2580 error = -EBUSY;
Al Virod83c49f2010-04-30 17:17:09 -04002581 else {
2582 if (target)
2583 dentry_unhash(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
Al Virod83c49f2010-04-30 17:17:09 -04002585 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586 if (target) {
Al Virod83c49f2010-04-30 17:17:09 -04002587 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588 target->i_flags |= S_DEAD;
Al Virod83c49f2010-04-30 17:17:09 -04002589 dont_mount(new_dentry);
2590 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002591 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592 if (d_unhashed(new_dentry))
2593 d_rehash(new_dentry);
2594 dput(new_dentry);
2595 }
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002596 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07002597 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2598 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599 return error;
2600}
2601
Adrian Bunk75c96f82005-05-05 16:16:09 -07002602static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2603 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002604{
2605 struct inode *target;
2606 int error;
2607
2608 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2609 if (error)
2610 return error;
2611
2612 dget(new_dentry);
2613 target = new_dentry->d_inode;
2614 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002615 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2617 error = -EBUSY;
2618 else
2619 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2620 if (!error) {
Al Virobec10522010-03-03 14:12:08 -05002621 if (target)
Al Virod83c49f2010-04-30 17:17:09 -04002622 dont_mount(new_dentry);
Mark Fasheh349457c2006-09-08 14:22:21 -07002623 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002624 d_move(old_dentry, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625 }
2626 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002627 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628 dput(new_dentry);
2629 return error;
2630}
2631
2632int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2633 struct inode *new_dir, struct dentry *new_dentry)
2634{
2635 int error;
2636 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Robert Love0eeca282005-07-12 17:06:03 -04002637 const char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638
2639 if (old_dentry->d_inode == new_dentry->d_inode)
2640 return 0;
2641
2642 error = may_delete(old_dir, old_dentry, is_dir);
2643 if (error)
2644 return error;
2645
2646 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02002647 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648 else
2649 error = may_delete(new_dir, new_dentry, is_dir);
2650 if (error)
2651 return error;
2652
Al Viroacfa4382008-12-04 10:06:33 -05002653 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654 return -EPERM;
2655
Robert Love0eeca282005-07-12 17:06:03 -04002656 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2657
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658 if (is_dir)
2659 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2660 else
2661 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
Al Viro123df292009-12-25 04:57:57 -05002662 if (!error)
2663 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04002664 new_dentry->d_inode, old_dentry);
Robert Love0eeca282005-07-12 17:06:03 -04002665 fsnotify_oldname_free(old_name);
2666
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667 return error;
2668}
2669
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002670SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2671 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672{
Al Viro2ad94ae2008-07-21 09:32:51 -04002673 struct dentry *old_dir, *new_dir;
2674 struct dentry *old_dentry, *new_dentry;
2675 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676 struct nameidata oldnd, newnd;
Al Viro2ad94ae2008-07-21 09:32:51 -04002677 char *from;
2678 char *to;
2679 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680
Al Viro2ad94ae2008-07-21 09:32:51 -04002681 error = user_path_parent(olddfd, oldname, &oldnd, &from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682 if (error)
2683 goto exit;
2684
Al Viro2ad94ae2008-07-21 09:32:51 -04002685 error = user_path_parent(newdfd, newname, &newnd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686 if (error)
2687 goto exit1;
2688
2689 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08002690 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691 goto exit2;
2692
Jan Blunck4ac91372008-02-14 19:34:32 -08002693 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002694 error = -EBUSY;
2695 if (oldnd.last_type != LAST_NORM)
2696 goto exit2;
2697
Jan Blunck4ac91372008-02-14 19:34:32 -08002698 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699 if (newnd.last_type != LAST_NORM)
2700 goto exit2;
2701
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002702 oldnd.flags &= ~LOOKUP_PARENT;
2703 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09002704 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002705
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706 trap = lock_rename(new_dir, old_dir);
2707
Christoph Hellwig49705b72005-11-08 21:35:06 -08002708 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709 error = PTR_ERR(old_dentry);
2710 if (IS_ERR(old_dentry))
2711 goto exit3;
2712 /* source must exist */
2713 error = -ENOENT;
2714 if (!old_dentry->d_inode)
2715 goto exit4;
2716 /* unless the source is a directory trailing slashes give -ENOTDIR */
2717 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2718 error = -ENOTDIR;
2719 if (oldnd.last.name[oldnd.last.len])
2720 goto exit4;
2721 if (newnd.last.name[newnd.last.len])
2722 goto exit4;
2723 }
2724 /* source should not be ancestor of target */
2725 error = -EINVAL;
2726 if (old_dentry == trap)
2727 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08002728 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729 error = PTR_ERR(new_dentry);
2730 if (IS_ERR(new_dentry))
2731 goto exit4;
2732 /* target should not be an ancestor of source */
2733 error = -ENOTEMPTY;
2734 if (new_dentry == trap)
2735 goto exit5;
2736
Dave Hansen9079b1e2008-02-15 14:37:49 -08002737 error = mnt_want_write(oldnd.path.mnt);
2738 if (error)
2739 goto exit5;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002740 error = security_path_rename(&oldnd.path, old_dentry,
2741 &newnd.path, new_dentry);
2742 if (error)
2743 goto exit6;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744 error = vfs_rename(old_dir->d_inode, old_dentry,
2745 new_dir->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002746exit6:
Dave Hansen9079b1e2008-02-15 14:37:49 -08002747 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748exit5:
2749 dput(new_dentry);
2750exit4:
2751 dput(old_dentry);
2752exit3:
2753 unlock_rename(new_dir, old_dir);
2754exit2:
Jan Blunck1d957f92008-02-14 19:34:35 -08002755 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002756 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002757exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002758 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759 putname(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002760exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761 return error;
2762}
2763
Heiko Carstensa26eab22009-01-14 14:14:17 +01002764SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002765{
2766 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2767}
2768
Linus Torvalds1da177e2005-04-16 15:20:36 -07002769int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2770{
2771 int len;
2772
2773 len = PTR_ERR(link);
2774 if (IS_ERR(link))
2775 goto out;
2776
2777 len = strlen(link);
2778 if (len > (unsigned) buflen)
2779 len = buflen;
2780 if (copy_to_user(buffer, link, len))
2781 len = -EFAULT;
2782out:
2783 return len;
2784}
2785
2786/*
2787 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2788 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2789 * using) it for any given inode is up to filesystem.
2790 */
2791int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2792{
2793 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002794 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07002795 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002796
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002798 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07002799 if (IS_ERR(cookie))
2800 return PTR_ERR(cookie);
2801
2802 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2803 if (dentry->d_inode->i_op->put_link)
2804 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2805 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002806}
2807
2808int vfs_follow_link(struct nameidata *nd, const char *link)
2809{
2810 return __vfs_follow_link(nd, link);
2811}
2812
2813/* get the link contents into pagecache */
2814static char *page_getlink(struct dentry * dentry, struct page **ppage)
2815{
Duane Griffinebd09ab2008-12-19 20:47:12 +00002816 char *kaddr;
2817 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002818 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07002819 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07002821 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00002823 kaddr = kmap(page);
2824 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2825 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826}
2827
2828int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2829{
2830 struct page *page = NULL;
2831 char *s = page_getlink(dentry, &page);
2832 int res = vfs_readlink(dentry,buffer,buflen,s);
2833 if (page) {
2834 kunmap(page);
2835 page_cache_release(page);
2836 }
2837 return res;
2838}
2839
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002840void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002842 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002844 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845}
2846
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002847void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002848{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002849 struct page *page = cookie;
2850
2851 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852 kunmap(page);
2853 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002854 }
2855}
2856
Nick Piggin54566b22009-01-04 12:00:53 -08002857/*
2858 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2859 */
2860int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002861{
2862 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002863 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002864 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08002865 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002866 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08002867 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2868 if (nofs)
2869 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002870
NeilBrown7e53cac2006-03-25 03:07:57 -08002871retry:
Nick Pigginafddba42007-10-16 01:25:01 -07002872 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08002873 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002874 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07002875 goto fail;
2876
Linus Torvalds1da177e2005-04-16 15:20:36 -07002877 kaddr = kmap_atomic(page, KM_USER0);
2878 memcpy(kaddr, symname, len-1);
2879 kunmap_atomic(kaddr, KM_USER0);
Nick Pigginafddba42007-10-16 01:25:01 -07002880
2881 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2882 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883 if (err < 0)
2884 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07002885 if (err < len-1)
2886 goto retry;
2887
Linus Torvalds1da177e2005-04-16 15:20:36 -07002888 mark_inode_dirty(inode);
2889 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002890fail:
2891 return err;
2892}
2893
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002894int page_symlink(struct inode *inode, const char *symname, int len)
2895{
2896 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08002897 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002898}
2899
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002900const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002901 .readlink = generic_readlink,
2902 .follow_link = page_follow_link_light,
2903 .put_link = page_put_link,
2904};
2905
Al Viro2d8f3032008-07-22 09:59:21 -04002906EXPORT_SYMBOL(user_path_at);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907EXPORT_SYMBOL(follow_down);
2908EXPORT_SYMBOL(follow_up);
2909EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2910EXPORT_SYMBOL(getname);
2911EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002912EXPORT_SYMBOL(lookup_one_len);
2913EXPORT_SYMBOL(page_follow_link_light);
2914EXPORT_SYMBOL(page_put_link);
2915EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002916EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002917EXPORT_SYMBOL(page_symlink);
2918EXPORT_SYMBOL(page_symlink_inode_operations);
2919EXPORT_SYMBOL(path_lookup);
Al Virod1811462008-08-02 00:49:18 -04002920EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002921EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04002922EXPORT_SYMBOL(inode_permission);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -08002923EXPORT_SYMBOL(file_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002924EXPORT_SYMBOL(unlock_rename);
2925EXPORT_SYMBOL(vfs_create);
2926EXPORT_SYMBOL(vfs_follow_link);
2927EXPORT_SYMBOL(vfs_link);
2928EXPORT_SYMBOL(vfs_mkdir);
2929EXPORT_SYMBOL(vfs_mknod);
2930EXPORT_SYMBOL(generic_permission);
2931EXPORT_SYMBOL(vfs_readlink);
2932EXPORT_SYMBOL(vfs_rename);
2933EXPORT_SYMBOL(vfs_rmdir);
2934EXPORT_SYMBOL(vfs_symlink);
2935EXPORT_SYMBOL(vfs_unlink);
2936EXPORT_SYMBOL(dentry_unhash);
2937EXPORT_SYMBOL(generic_readlink);