blob: 1c0fca6e899eef7aa768ed704e2216c3d024c7df [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 Viro1f36f772009-12-26 10:56:19 -05001613 int mode, const char *pathname,
1614 int *want_dir)
Al Virofb1cc552009-12-24 01:58:28 -05001615{
Al Viroa1e28032009-12-24 02:12:06 -05001616 struct dentry *dir = nd->path.dentry;
Al Virofb1cc552009-12-24 01:58:28 -05001617 struct file *filp;
Al Viro1f36f772009-12-26 10:56:19 -05001618 int error = -EISDIR;
Al Virofb1cc552009-12-24 01:58:28 -05001619
Al Viro1f36f772009-12-26 10:56:19 -05001620 switch (nd->last_type) {
1621 case LAST_DOTDOT:
1622 follow_dotdot(nd);
1623 dir = nd->path.dentry;
1624 if (nd->path.mnt->mnt_sb->s_type->fs_flags & FS_REVAL_DOT) {
1625 if (!dir->d_op->d_revalidate(dir, nd)) {
1626 error = -ESTALE;
1627 goto exit;
1628 }
1629 }
1630 /* fallthrough */
1631 case LAST_DOT:
1632 case LAST_ROOT:
1633 if (open_flag & O_CREAT)
1634 goto exit;
1635 /* fallthrough */
1636 case LAST_BIND:
1637 audit_inode(pathname, dir);
Al Viro67ee3ad2009-12-26 07:01:01 -05001638 goto ok;
Al Viro1f36f772009-12-26 10:56:19 -05001639 }
Al Viro67ee3ad2009-12-26 07:01:01 -05001640
Al Viro1f36f772009-12-26 10:56:19 -05001641 /* trailing slashes? */
1642 if (nd->last.name[nd->last.len]) {
1643 if (open_flag & O_CREAT)
1644 goto exit;
1645 *want_dir = 1;
1646 }
Al Viroa2c36b42009-12-24 03:39:50 -05001647
Al Viro1f36f772009-12-26 10:56:19 -05001648 /* just plain open? */
1649 if (!(open_flag & O_CREAT)) {
1650 error = do_lookup(nd, &nd->last, path);
1651 if (error)
1652 goto exit;
1653 error = -ENOENT;
1654 if (!path->dentry->d_inode)
1655 goto exit_dput;
1656 if (path->dentry->d_inode->i_op->follow_link)
1657 return NULL;
1658 error = -ENOTDIR;
Al Viro781b1672010-03-06 18:41:07 +00001659 if (*want_dir && !path->dentry->d_inode->i_op->lookup)
Al Viro1f36f772009-12-26 10:56:19 -05001660 goto exit_dput;
1661 path_to_nameidata(path, nd);
1662 audit_inode(pathname, nd->path.dentry);
1663 goto ok;
1664 }
1665
1666 /* OK, it's O_CREAT */
Al Viroa1e28032009-12-24 02:12:06 -05001667 mutex_lock(&dir->d_inode->i_mutex);
1668
1669 path->dentry = lookup_hash(nd);
1670 path->mnt = nd->path.mnt;
1671
Al Virofb1cc552009-12-24 01:58:28 -05001672 error = PTR_ERR(path->dentry);
1673 if (IS_ERR(path->dentry)) {
1674 mutex_unlock(&dir->d_inode->i_mutex);
1675 goto exit;
1676 }
1677
1678 if (IS_ERR(nd->intent.open.file)) {
1679 error = PTR_ERR(nd->intent.open.file);
1680 goto exit_mutex_unlock;
1681 }
1682
1683 /* Negative dentry, just create the file */
1684 if (!path->dentry->d_inode) {
1685 /*
1686 * This write is needed to ensure that a
1687 * ro->rw transition does not occur between
1688 * the time when the file is created and when
1689 * a permanent write count is taken through
1690 * the 'struct file' in nameidata_to_filp().
1691 */
1692 error = mnt_want_write(nd->path.mnt);
1693 if (error)
1694 goto exit_mutex_unlock;
1695 error = __open_namei_create(nd, path, open_flag, mode);
1696 if (error) {
1697 mnt_drop_write(nd->path.mnt);
1698 goto exit;
1699 }
1700 filp = nameidata_to_filp(nd);
1701 mnt_drop_write(nd->path.mnt);
Al Virofb1cc552009-12-24 01:58:28 -05001702 if (!IS_ERR(filp)) {
1703 error = ima_file_check(filp, acc_mode);
1704 if (error) {
1705 fput(filp);
1706 filp = ERR_PTR(error);
1707 }
1708 }
1709 return filp;
1710 }
1711
1712 /*
1713 * It already exists.
1714 */
1715 mutex_unlock(&dir->d_inode->i_mutex);
1716 audit_inode(pathname, path->dentry);
1717
1718 error = -EEXIST;
Al Viro5b369df2009-12-24 06:51:13 -05001719 if (open_flag & O_EXCL)
Al Virofb1cc552009-12-24 01:58:28 -05001720 goto exit_dput;
1721
1722 if (__follow_mount(path)) {
1723 error = -ELOOP;
Al Viro5b369df2009-12-24 06:51:13 -05001724 if (open_flag & O_NOFOLLOW)
Al Virofb1cc552009-12-24 01:58:28 -05001725 goto exit_dput;
1726 }
1727
1728 error = -ENOENT;
1729 if (!path->dentry->d_inode)
1730 goto exit_dput;
Al Viro9e67f362009-12-26 07:04:50 -05001731
1732 if (path->dentry->d_inode->i_op->follow_link)
Al Virofb1cc552009-12-24 01:58:28 -05001733 return NULL;
Al Virofb1cc552009-12-24 01:58:28 -05001734
1735 path_to_nameidata(path, nd);
1736 error = -EISDIR;
1737 if (S_ISDIR(path->dentry->d_inode->i_mode))
1738 goto exit;
Al Viro67ee3ad2009-12-26 07:01:01 -05001739ok:
Al Viro9a661792009-12-24 06:49:47 -05001740 filp = finish_open(nd, open_flag, acc_mode);
Al Virofb1cc552009-12-24 01:58:28 -05001741 return filp;
1742
1743exit_mutex_unlock:
1744 mutex_unlock(&dir->d_inode->i_mutex);
1745exit_dput:
1746 path_put_conditional(path, nd);
1747exit:
1748 if (!IS_ERR(nd->intent.open.file))
1749 release_open_intent(nd);
Al Virofb1cc552009-12-24 01:58:28 -05001750 path_put(&nd->path);
1751 return ERR_PTR(error);
1752}
1753
Dave Hansend57999e2008-02-15 14:37:27 -08001754/*
Dave Hansen4a3fd212008-02-15 14:37:48 -08001755 * Note that the low bits of the passed in "open_flag"
1756 * are not the same as in the local variable "flag". See
1757 * open_to_namei_flags() for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001759struct file *do_filp_open(int dfd, const char *pathname,
Al Viro6e8341a2009-04-06 11:16:22 -04001760 int open_flag, int mode, int acc_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761{
Dave Hansen4a3fd212008-02-15 14:37:48 -08001762 struct file *filp;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001763 struct nameidata nd;
Al Viro6e8341a2009-04-06 11:16:22 -04001764 int error;
Al Viro9850c052010-01-13 15:01:15 -05001765 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 int count = 0;
Dave Hansend57999e2008-02-15 14:37:27 -08001767 int flag = open_to_namei_flags(open_flag);
Al Viro9850c052010-01-13 15:01:15 -05001768 int force_reval = 0;
Al Viro1f36f772009-12-26 10:56:19 -05001769 int want_dir = open_flag & O_DIRECTORY;
1770
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 }
1831 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname, &want_dir);
Al Viro806b6812009-12-26 07:16:40 -05001832 while (unlikely(!filp)) { /* trailing symlink */
Al Virodef4af32009-12-26 08:37:05 -05001833 struct path holder;
Al Viro1f36f772009-12-26 10:56:19 -05001834 struct inode *inode = path.dentry->d_inode;
Al Virodef4af32009-12-26 08:37:05 -05001835 void *cookie;
Al Viro806b6812009-12-26 07:16:40 -05001836 error = -ELOOP;
Al Viro1f36f772009-12-26 10:56:19 -05001837 /* S_ISDIR part is a temporary automount kludge */
1838 if ((open_flag & O_NOFOLLOW) && !S_ISDIR(inode->i_mode))
1839 goto exit_dput;
1840 if (count++ == 32)
Al Viro806b6812009-12-26 07:16:40 -05001841 goto exit_dput;
1842 /*
1843 * This is subtle. Instead of calling do_follow_link() we do
1844 * the thing by hands. The reason is that this way we have zero
1845 * link_count and path_walk() (called from ->follow_link)
1846 * honoring LOOKUP_PARENT. After that we have the parent and
1847 * last component, i.e. we are in the same situation as after
1848 * the first path_walk(). Well, almost - if the last component
1849 * is normal we get its copy stored in nd->last.name and we will
1850 * have to putname() it when we are done. Procfs-like symlinks
1851 * just set LAST_BIND.
1852 */
1853 nd.flags |= LOOKUP_PARENT;
1854 error = security_inode_follow_link(path.dentry, &nd);
1855 if (error)
1856 goto exit_dput;
Al Virodef4af32009-12-26 08:37:05 -05001857 error = __do_follow_link(&path, &nd, &cookie);
1858 if (unlikely(error)) {
Al Viro806b6812009-12-26 07:16:40 -05001859 /* nd.path had been dropped */
Al Virodef4af32009-12-26 08:37:05 -05001860 if (!IS_ERR(cookie) && inode->i_op->put_link)
1861 inode->i_op->put_link(path.dentry, &nd, cookie);
1862 path_put(&path);
Al Viro806b6812009-12-26 07:16:40 -05001863 release_open_intent(&nd);
1864 filp = ERR_PTR(error);
1865 goto out;
1866 }
Al Virodef4af32009-12-26 08:37:05 -05001867 holder = path;
Al Viro806b6812009-12-26 07:16:40 -05001868 nd.flags &= ~LOOKUP_PARENT;
Al Viro1f36f772009-12-26 10:56:19 -05001869 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname, &want_dir);
Al Virodef4af32009-12-26 08:37:05 -05001870 if (inode->i_op->put_link)
1871 inode->i_op->put_link(holder.dentry, &nd, cookie);
1872 path_put(&holder);
Al Viro806b6812009-12-26 07:16:40 -05001873 }
Al Viro10fa8e62009-12-26 07:09:49 -05001874out:
Al Viro2a737872009-04-07 11:49:53 -04001875 if (nd.root.mnt)
1876 path_put(&nd.root);
Al Viro10fa8e62009-12-26 07:09:49 -05001877 if (filp == ERR_PTR(-ESTALE) && !force_reval) {
1878 force_reval = 1;
1879 goto reval;
1880 }
1881 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882
Al Viro806b6812009-12-26 07:16:40 -05001883exit_dput:
1884 path_put_conditional(&path, &nd);
1885 if (!IS_ERR(nd.intent.open.file))
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001886 release_open_intent(&nd);
Al Viro806b6812009-12-26 07:16:40 -05001887exit_parent:
1888 path_put(&nd.path);
1889 filp = ERR_PTR(error);
Al Viro10fa8e62009-12-26 07:09:49 -05001890 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891}
1892
1893/**
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001894 * filp_open - open file and return file pointer
1895 *
1896 * @filename: path to open
1897 * @flags: open flags as per the open(2) second argument
1898 * @mode: mode for the new file if O_CREAT is set, else ignored
1899 *
1900 * This is the helper to open a file from kernelspace if you really
1901 * have to. But in generally you should not do this, so please move
1902 * along, nothing to see here..
1903 */
1904struct file *filp_open(const char *filename, int flags, int mode)
1905{
Al Viro6e8341a2009-04-06 11:16:22 -04001906 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001907}
1908EXPORT_SYMBOL(filp_open);
1909
1910/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911 * lookup_create - lookup a dentry, creating it if it doesn't exist
1912 * @nd: nameidata info
1913 * @is_dir: directory flag
1914 *
1915 * Simple function to lookup and return a dentry and create it
1916 * if it doesn't exist. Is SMP-safe.
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001917 *
Jan Blunck4ac91372008-02-14 19:34:32 -08001918 * Returns with nd->path.dentry->d_inode->i_mutex locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 */
1920struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1921{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001922 struct dentry *dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923
Jan Blunck4ac91372008-02-14 19:34:32 -08001924 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001925 /*
1926 * Yucky last component or no last component at all?
1927 * (foo/., foo/.., /////)
1928 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929 if (nd->last_type != LAST_NORM)
1930 goto fail;
1931 nd->flags &= ~LOOKUP_PARENT;
Al Viro35165862008-08-05 03:00:49 -04001932 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
ASANO Masahiroa6349042006-08-22 20:06:02 -04001933 nd->intent.open.flags = O_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001934
1935 /*
1936 * Do the final lookup.
1937 */
Christoph Hellwig49705b72005-11-08 21:35:06 -08001938 dentry = lookup_hash(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 if (IS_ERR(dentry))
1940 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001941
Al Viroe9baf6e2008-05-15 04:49:12 -04001942 if (dentry->d_inode)
1943 goto eexist;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001944 /*
1945 * Special case - lookup gave negative, but... we had foo/bar/
1946 * From the vfs_mknod() POV we just have a negative dentry -
1947 * all is fine. Let's be bastards - you had / on the end, you've
1948 * been asking for (non-existent) directory. -ENOENT for you.
1949 */
Al Viroe9baf6e2008-05-15 04:49:12 -04001950 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1951 dput(dentry);
1952 dentry = ERR_PTR(-ENOENT);
1953 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 return dentry;
Al Viroe9baf6e2008-05-15 04:49:12 -04001955eexist:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 dput(dentry);
Al Viroe9baf6e2008-05-15 04:49:12 -04001957 dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958fail:
1959 return dentry;
1960}
Christoph Hellwigf81a0bf2005-05-19 12:26:43 -07001961EXPORT_SYMBOL_GPL(lookup_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962
1963int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1964{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001965 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966
1967 if (error)
1968 return error;
1969
1970 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1971 return -EPERM;
1972
Al Viroacfa4382008-12-04 10:06:33 -05001973 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 return -EPERM;
1975
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07001976 error = devcgroup_inode_mknod(mode, dev);
1977 if (error)
1978 return error;
1979
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 error = security_inode_mknod(dir, dentry, mode, dev);
1981 if (error)
1982 return error;
1983
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001985 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001986 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 return error;
1988}
1989
Dave Hansen463c3192008-02-15 14:37:57 -08001990static int may_mknod(mode_t mode)
1991{
1992 switch (mode & S_IFMT) {
1993 case S_IFREG:
1994 case S_IFCHR:
1995 case S_IFBLK:
1996 case S_IFIFO:
1997 case S_IFSOCK:
1998 case 0: /* zero mode translates to S_IFREG */
1999 return 0;
2000 case S_IFDIR:
2001 return -EPERM;
2002 default:
2003 return -EINVAL;
2004 }
2005}
2006
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002007SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2008 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009{
Al Viro2ad94ae2008-07-21 09:32:51 -04002010 int error;
2011 char *tmp;
2012 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 struct nameidata nd;
2014
2015 if (S_ISDIR(mode))
2016 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017
Al Viro2ad94ae2008-07-21 09:32:51 -04002018 error = user_path_parent(dfd, filename, &nd, &tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002020 return error;
2021
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 dentry = lookup_create(&nd, 0);
Dave Hansen463c3192008-02-15 14:37:57 -08002023 if (IS_ERR(dentry)) {
2024 error = PTR_ERR(dentry);
2025 goto out_unlock;
2026 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002027 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002028 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002029 error = may_mknod(mode);
2030 if (error)
2031 goto out_dput;
2032 error = mnt_want_write(nd.path.mnt);
2033 if (error)
2034 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002035 error = security_path_mknod(&nd.path, dentry, mode, dev);
2036 if (error)
2037 goto out_drop_write;
Dave Hansen463c3192008-02-15 14:37:57 -08002038 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039 case 0: case S_IFREG:
Jan Blunck4ac91372008-02-14 19:34:32 -08002040 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 break;
2042 case S_IFCHR: case S_IFBLK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002043 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 new_decode_dev(dev));
2045 break;
2046 case S_IFIFO: case S_IFSOCK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002047 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002050out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002051 mnt_drop_write(nd.path.mnt);
2052out_dput:
2053 dput(dentry);
2054out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002055 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002056 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057 putname(tmp);
2058
2059 return error;
2060}
2061
Heiko Carstens3480b252009-01-14 14:14:16 +01002062SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002063{
2064 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2065}
2066
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2068{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002069 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070
2071 if (error)
2072 return error;
2073
Al Viroacfa4382008-12-04 10:06:33 -05002074 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 return -EPERM;
2076
2077 mode &= (S_IRWXUGO|S_ISVTX);
2078 error = security_inode_mkdir(dir, dentry, mode);
2079 if (error)
2080 return error;
2081
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002083 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002084 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 return error;
2086}
2087
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002088SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089{
2090 int error = 0;
2091 char * tmp;
Dave Hansen6902d922006-09-30 23:29:01 -07002092 struct dentry *dentry;
2093 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094
Al Viro2ad94ae2008-07-21 09:32:51 -04002095 error = user_path_parent(dfd, pathname, &nd, &tmp);
2096 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002097 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098
Dave Hansen6902d922006-09-30 23:29:01 -07002099 dentry = lookup_create(&nd, 1);
2100 error = PTR_ERR(dentry);
2101 if (IS_ERR(dentry))
2102 goto out_unlock;
2103
Jan Blunck4ac91372008-02-14 19:34:32 -08002104 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002105 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002106 error = mnt_want_write(nd.path.mnt);
2107 if (error)
2108 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002109 error = security_path_mkdir(&nd.path, dentry, mode);
2110 if (error)
2111 goto out_drop_write;
Jan Blunck4ac91372008-02-14 19:34:32 -08002112 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002113out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002114 mnt_drop_write(nd.path.mnt);
2115out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002116 dput(dentry);
2117out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002118 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002119 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002120 putname(tmp);
2121out_err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122 return error;
2123}
2124
Heiko Carstens3cdad422009-01-14 14:14:22 +01002125SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002126{
2127 return sys_mkdirat(AT_FDCWD, pathname, mode);
2128}
2129
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130/*
2131 * We try to drop the dentry early: we should have
2132 * a usage count of 2 if we're the only user of this
2133 * dentry, and if that is true (possibly after pruning
2134 * the dcache), then we drop the dentry now.
2135 *
2136 * A low-level filesystem can, if it choses, legally
2137 * do a
2138 *
2139 * if (!d_unhashed(dentry))
2140 * return -EBUSY;
2141 *
2142 * if it cannot handle the case of removing a directory
2143 * that is still in use by something else..
2144 */
2145void dentry_unhash(struct dentry *dentry)
2146{
2147 dget(dentry);
Vasily Averindc168422006-12-06 20:37:07 -08002148 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 spin_lock(&dcache_lock);
2150 spin_lock(&dentry->d_lock);
2151 if (atomic_read(&dentry->d_count) == 2)
2152 __d_drop(dentry);
2153 spin_unlock(&dentry->d_lock);
2154 spin_unlock(&dcache_lock);
2155}
2156
2157int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2158{
2159 int error = may_delete(dir, dentry, 1);
2160
2161 if (error)
2162 return error;
2163
Al Viroacfa4382008-12-04 10:06:33 -05002164 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 return -EPERM;
2166
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002167 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 dentry_unhash(dentry);
2169 if (d_mountpoint(dentry))
2170 error = -EBUSY;
2171 else {
2172 error = security_inode_rmdir(dir, dentry);
2173 if (!error) {
2174 error = dir->i_op->rmdir(dir, dentry);
2175 if (!error)
2176 dentry->d_inode->i_flags |= S_DEAD;
2177 }
2178 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002179 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181 d_delete(dentry);
2182 }
2183 dput(dentry);
2184
2185 return error;
2186}
2187
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002188static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189{
2190 int error = 0;
2191 char * name;
2192 struct dentry *dentry;
2193 struct nameidata nd;
2194
Al Viro2ad94ae2008-07-21 09:32:51 -04002195 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002197 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198
2199 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002200 case LAST_DOTDOT:
2201 error = -ENOTEMPTY;
2202 goto exit1;
2203 case LAST_DOT:
2204 error = -EINVAL;
2205 goto exit1;
2206 case LAST_ROOT:
2207 error = -EBUSY;
2208 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002210
2211 nd.flags &= ~LOOKUP_PARENT;
2212
Jan Blunck4ac91372008-02-14 19:34:32 -08002213 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002214 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002216 if (IS_ERR(dentry))
2217 goto exit2;
Dave Hansen06227532008-02-15 14:37:34 -08002218 error = mnt_want_write(nd.path.mnt);
2219 if (error)
2220 goto exit3;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002221 error = security_path_rmdir(&nd.path, dentry);
2222 if (error)
2223 goto exit4;
Jan Blunck4ac91372008-02-14 19:34:32 -08002224 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002225exit4:
Dave Hansen06227532008-02-15 14:37:34 -08002226 mnt_drop_write(nd.path.mnt);
2227exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07002228 dput(dentry);
2229exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08002230 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002232 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233 putname(name);
2234 return error;
2235}
2236
Heiko Carstens3cdad422009-01-14 14:14:22 +01002237SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002238{
2239 return do_rmdir(AT_FDCWD, pathname);
2240}
2241
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242int vfs_unlink(struct inode *dir, struct dentry *dentry)
2243{
2244 int error = may_delete(dir, dentry, 0);
2245
2246 if (error)
2247 return error;
2248
Al Viroacfa4382008-12-04 10:06:33 -05002249 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 return -EPERM;
2251
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002252 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 if (d_mountpoint(dentry))
2254 error = -EBUSY;
2255 else {
2256 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002257 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 error = dir->i_op->unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002259 if (!error)
2260 dentry->d_inode->i_flags |= S_DEAD;
2261 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002263 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264
2265 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2266 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08002267 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269 }
Robert Love0eeca282005-07-12 17:06:03 -04002270
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271 return error;
2272}
2273
2274/*
2275 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002276 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277 * writeout happening, and we don't want to prevent access to the directory
2278 * while waiting on the I/O.
2279 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002280static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281{
Al Viro2ad94ae2008-07-21 09:32:51 -04002282 int error;
2283 char *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284 struct dentry *dentry;
2285 struct nameidata nd;
2286 struct inode *inode = NULL;
2287
Al Viro2ad94ae2008-07-21 09:32:51 -04002288 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002290 return error;
2291
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 error = -EISDIR;
2293 if (nd.last_type != LAST_NORM)
2294 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002295
2296 nd.flags &= ~LOOKUP_PARENT;
2297
Jan Blunck4ac91372008-02-14 19:34:32 -08002298 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002299 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 error = PTR_ERR(dentry);
2301 if (!IS_ERR(dentry)) {
2302 /* Why not before? Because we want correct error value */
2303 if (nd.last.name[nd.last.len])
2304 goto slashes;
2305 inode = dentry->d_inode;
2306 if (inode)
2307 atomic_inc(&inode->i_count);
Dave Hansen06227532008-02-15 14:37:34 -08002308 error = mnt_want_write(nd.path.mnt);
2309 if (error)
2310 goto exit2;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002311 error = security_path_unlink(&nd.path, dentry);
2312 if (error)
2313 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08002314 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002315exit3:
Dave Hansen06227532008-02-15 14:37:34 -08002316 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317 exit2:
2318 dput(dentry);
2319 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002320 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321 if (inode)
2322 iput(inode); /* truncate the inode here */
2323exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002324 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325 putname(name);
2326 return error;
2327
2328slashes:
2329 error = !dentry->d_inode ? -ENOENT :
2330 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2331 goto exit2;
2332}
2333
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002334SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002335{
2336 if ((flag & ~AT_REMOVEDIR) != 0)
2337 return -EINVAL;
2338
2339 if (flag & AT_REMOVEDIR)
2340 return do_rmdir(dfd, pathname);
2341
2342 return do_unlinkat(dfd, pathname);
2343}
2344
Heiko Carstens3480b252009-01-14 14:14:16 +01002345SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002346{
2347 return do_unlinkat(AT_FDCWD, pathname);
2348}
2349
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002350int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002351{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002352 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353
2354 if (error)
2355 return error;
2356
Al Viroacfa4382008-12-04 10:06:33 -05002357 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358 return -EPERM;
2359
2360 error = security_inode_symlink(dir, dentry, oldname);
2361 if (error)
2362 return error;
2363
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002365 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002366 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367 return error;
2368}
2369
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002370SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2371 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372{
Al Viro2ad94ae2008-07-21 09:32:51 -04002373 int error;
2374 char *from;
2375 char *to;
Dave Hansen6902d922006-09-30 23:29:01 -07002376 struct dentry *dentry;
2377 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378
2379 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04002380 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381 return PTR_ERR(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002382
2383 error = user_path_parent(newdfd, newname, &nd, &to);
2384 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002385 goto out_putname;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386
Dave Hansen6902d922006-09-30 23:29:01 -07002387 dentry = lookup_create(&nd, 0);
2388 error = PTR_ERR(dentry);
2389 if (IS_ERR(dentry))
2390 goto out_unlock;
2391
Dave Hansen75c3f292008-02-15 14:37:45 -08002392 error = mnt_want_write(nd.path.mnt);
2393 if (error)
2394 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002395 error = security_path_symlink(&nd.path, dentry, from);
2396 if (error)
2397 goto out_drop_write;
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002398 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002399out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002400 mnt_drop_write(nd.path.mnt);
2401out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002402 dput(dentry);
2403out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002404 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002405 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002406 putname(to);
2407out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408 putname(from);
2409 return error;
2410}
2411
Heiko Carstens3480b252009-01-14 14:14:16 +01002412SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002413{
2414 return sys_symlinkat(oldname, AT_FDCWD, newname);
2415}
2416
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2418{
2419 struct inode *inode = old_dentry->d_inode;
2420 int error;
2421
2422 if (!inode)
2423 return -ENOENT;
2424
Miklos Szeredia95164d2008-07-30 15:08:48 +02002425 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426 if (error)
2427 return error;
2428
2429 if (dir->i_sb != inode->i_sb)
2430 return -EXDEV;
2431
2432 /*
2433 * A link to an append-only or immutable file cannot be created.
2434 */
2435 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2436 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05002437 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002439 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440 return -EPERM;
2441
2442 error = security_inode_link(old_dentry, dir, new_dentry);
2443 if (error)
2444 return error;
2445
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002446 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447 error = dir->i_op->link(old_dentry, dir, new_dentry);
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002448 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002449 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002450 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451 return error;
2452}
2453
2454/*
2455 * Hardlinks are often used in delicate situations. We avoid
2456 * security-related surprises by not following symlinks on the
2457 * newname. --KAB
2458 *
2459 * We don't follow them on the oldname either to be compatible
2460 * with linux 2.0, and to avoid hard-linking to directories
2461 * and other special files. --ADM
2462 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002463SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2464 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465{
2466 struct dentry *new_dentry;
Al Viro2d8f3032008-07-22 09:59:21 -04002467 struct nameidata nd;
2468 struct path old_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469 int error;
Al Viro2ad94ae2008-07-21 09:32:51 -04002470 char *to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471
Ulrich Drepper45c9b112006-06-25 05:49:11 -07002472 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002473 return -EINVAL;
2474
Al Viro2d8f3032008-07-22 09:59:21 -04002475 error = user_path_at(olddfd, oldname,
2476 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2477 &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002479 return error;
2480
2481 error = user_path_parent(newdfd, newname, &nd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482 if (error)
2483 goto out;
2484 error = -EXDEV;
Al Viro2d8f3032008-07-22 09:59:21 -04002485 if (old_path.mnt != nd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002486 goto out_release;
2487 new_dentry = lookup_create(&nd, 0);
2488 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002489 if (IS_ERR(new_dentry))
2490 goto out_unlock;
Dave Hansen75c3f292008-02-15 14:37:45 -08002491 error = mnt_want_write(nd.path.mnt);
2492 if (error)
2493 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002494 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2495 if (error)
2496 goto out_drop_write;
Al Viro2d8f3032008-07-22 09:59:21 -04002497 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002498out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002499 mnt_drop_write(nd.path.mnt);
2500out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002501 dput(new_dentry);
2502out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002503 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504out_release:
Jan Blunck1d957f92008-02-14 19:34:35 -08002505 path_put(&nd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002506 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507out:
Al Viro2d8f3032008-07-22 09:59:21 -04002508 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509
2510 return error;
2511}
2512
Heiko Carstens3480b252009-01-14 14:14:16 +01002513SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002514{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002515 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002516}
2517
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518/*
2519 * The worst of all namespace operations - renaming directory. "Perverted"
2520 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2521 * Problems:
2522 * a) we can get into loop creation. Check is done in is_subdir().
2523 * b) race potential - two innocent renames can create a loop together.
2524 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002525 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526 * story.
2527 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002528 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529 * whether the target exists). Solution: try to be smart with locking
2530 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002531 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532 * move will be locked. Thus we can rank directories by the tree
2533 * (ancestors first) and rank all non-directories after them.
2534 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002535 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536 * HOWEVER, it relies on the assumption that any object with ->lookup()
2537 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2538 * we'd better make sure that there's no link(2) for them.
2539 * d) some filesystems don't support opened-but-unlinked directories,
2540 * either because of layout or because they are not ready to deal with
2541 * all cases correctly. The latter will be fixed (taking this sort of
2542 * stuff into VFS), but the former is not going away. Solution: the same
2543 * trick as in rmdir().
2544 * e) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002545 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05002547 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548 * locking].
2549 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07002550static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2551 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552{
2553 int error = 0;
2554 struct inode *target;
2555
2556 /*
2557 * If we are going to change the parent - check write permissions,
2558 * we'll need to flip '..'.
2559 */
2560 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04002561 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562 if (error)
2563 return error;
2564 }
2565
2566 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2567 if (error)
2568 return error;
2569
2570 target = new_dentry->d_inode;
2571 if (target) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002572 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573 dentry_unhash(new_dentry);
2574 }
2575 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2576 error = -EBUSY;
2577 else
2578 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2579 if (target) {
2580 if (!error)
2581 target->i_flags |= S_DEAD;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002582 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583 if (d_unhashed(new_dentry))
2584 d_rehash(new_dentry);
2585 dput(new_dentry);
2586 }
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002587 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07002588 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2589 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590 return error;
2591}
2592
Adrian Bunk75c96f82005-05-05 16:16:09 -07002593static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2594 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595{
2596 struct inode *target;
2597 int error;
2598
2599 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2600 if (error)
2601 return error;
2602
2603 dget(new_dentry);
2604 target = new_dentry->d_inode;
2605 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002606 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2608 error = -EBUSY;
2609 else
2610 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2611 if (!error) {
Al Virobec10522010-03-03 14:12:08 -05002612 if (target)
2613 target->i_flags |= S_DEAD;
Mark Fasheh349457c2006-09-08 14:22:21 -07002614 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615 d_move(old_dentry, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616 }
2617 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002618 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619 dput(new_dentry);
2620 return error;
2621}
2622
2623int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2624 struct inode *new_dir, struct dentry *new_dentry)
2625{
2626 int error;
2627 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Robert Love0eeca282005-07-12 17:06:03 -04002628 const char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629
2630 if (old_dentry->d_inode == new_dentry->d_inode)
2631 return 0;
2632
2633 error = may_delete(old_dir, old_dentry, is_dir);
2634 if (error)
2635 return error;
2636
2637 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02002638 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639 else
2640 error = may_delete(new_dir, new_dentry, is_dir);
2641 if (error)
2642 return error;
2643
Al Viroacfa4382008-12-04 10:06:33 -05002644 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645 return -EPERM;
2646
Robert Love0eeca282005-07-12 17:06:03 -04002647 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2648
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649 if (is_dir)
2650 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2651 else
2652 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
Al Viro123df292009-12-25 04:57:57 -05002653 if (!error)
2654 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04002655 new_dentry->d_inode, old_dentry);
Robert Love0eeca282005-07-12 17:06:03 -04002656 fsnotify_oldname_free(old_name);
2657
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658 return error;
2659}
2660
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002661SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2662 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663{
Al Viro2ad94ae2008-07-21 09:32:51 -04002664 struct dentry *old_dir, *new_dir;
2665 struct dentry *old_dentry, *new_dentry;
2666 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667 struct nameidata oldnd, newnd;
Al Viro2ad94ae2008-07-21 09:32:51 -04002668 char *from;
2669 char *to;
2670 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671
Al Viro2ad94ae2008-07-21 09:32:51 -04002672 error = user_path_parent(olddfd, oldname, &oldnd, &from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673 if (error)
2674 goto exit;
2675
Al Viro2ad94ae2008-07-21 09:32:51 -04002676 error = user_path_parent(newdfd, newname, &newnd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677 if (error)
2678 goto exit1;
2679
2680 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08002681 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682 goto exit2;
2683
Jan Blunck4ac91372008-02-14 19:34:32 -08002684 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002685 error = -EBUSY;
2686 if (oldnd.last_type != LAST_NORM)
2687 goto exit2;
2688
Jan Blunck4ac91372008-02-14 19:34:32 -08002689 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690 if (newnd.last_type != LAST_NORM)
2691 goto exit2;
2692
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002693 oldnd.flags &= ~LOOKUP_PARENT;
2694 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09002695 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002696
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697 trap = lock_rename(new_dir, old_dir);
2698
Christoph Hellwig49705b72005-11-08 21:35:06 -08002699 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700 error = PTR_ERR(old_dentry);
2701 if (IS_ERR(old_dentry))
2702 goto exit3;
2703 /* source must exist */
2704 error = -ENOENT;
2705 if (!old_dentry->d_inode)
2706 goto exit4;
2707 /* unless the source is a directory trailing slashes give -ENOTDIR */
2708 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2709 error = -ENOTDIR;
2710 if (oldnd.last.name[oldnd.last.len])
2711 goto exit4;
2712 if (newnd.last.name[newnd.last.len])
2713 goto exit4;
2714 }
2715 /* source should not be ancestor of target */
2716 error = -EINVAL;
2717 if (old_dentry == trap)
2718 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08002719 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720 error = PTR_ERR(new_dentry);
2721 if (IS_ERR(new_dentry))
2722 goto exit4;
2723 /* target should not be an ancestor of source */
2724 error = -ENOTEMPTY;
2725 if (new_dentry == trap)
2726 goto exit5;
2727
Dave Hansen9079b1e2008-02-15 14:37:49 -08002728 error = mnt_want_write(oldnd.path.mnt);
2729 if (error)
2730 goto exit5;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002731 error = security_path_rename(&oldnd.path, old_dentry,
2732 &newnd.path, new_dentry);
2733 if (error)
2734 goto exit6;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735 error = vfs_rename(old_dir->d_inode, old_dentry,
2736 new_dir->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002737exit6:
Dave Hansen9079b1e2008-02-15 14:37:49 -08002738 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002739exit5:
2740 dput(new_dentry);
2741exit4:
2742 dput(old_dentry);
2743exit3:
2744 unlock_rename(new_dir, old_dir);
2745exit2:
Jan Blunck1d957f92008-02-14 19:34:35 -08002746 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002747 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002749 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750 putname(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002751exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002752 return error;
2753}
2754
Heiko Carstensa26eab22009-01-14 14:14:17 +01002755SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002756{
2757 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2758}
2759
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2761{
2762 int len;
2763
2764 len = PTR_ERR(link);
2765 if (IS_ERR(link))
2766 goto out;
2767
2768 len = strlen(link);
2769 if (len > (unsigned) buflen)
2770 len = buflen;
2771 if (copy_to_user(buffer, link, len))
2772 len = -EFAULT;
2773out:
2774 return len;
2775}
2776
2777/*
2778 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2779 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2780 * using) it for any given inode is up to filesystem.
2781 */
2782int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2783{
2784 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002785 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07002786 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002787
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002789 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07002790 if (IS_ERR(cookie))
2791 return PTR_ERR(cookie);
2792
2793 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2794 if (dentry->d_inode->i_op->put_link)
2795 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2796 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797}
2798
2799int vfs_follow_link(struct nameidata *nd, const char *link)
2800{
2801 return __vfs_follow_link(nd, link);
2802}
2803
2804/* get the link contents into pagecache */
2805static char *page_getlink(struct dentry * dentry, struct page **ppage)
2806{
Duane Griffinebd09ab2008-12-19 20:47:12 +00002807 char *kaddr;
2808 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002809 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07002810 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07002812 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002813 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00002814 kaddr = kmap(page);
2815 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2816 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817}
2818
2819int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2820{
2821 struct page *page = NULL;
2822 char *s = page_getlink(dentry, &page);
2823 int res = vfs_readlink(dentry,buffer,buflen,s);
2824 if (page) {
2825 kunmap(page);
2826 page_cache_release(page);
2827 }
2828 return res;
2829}
2830
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002831void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002832{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002833 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002834 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002835 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836}
2837
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002838void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002840 struct page *page = cookie;
2841
2842 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843 kunmap(page);
2844 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845 }
2846}
2847
Nick Piggin54566b22009-01-04 12:00:53 -08002848/*
2849 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2850 */
2851int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852{
2853 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002854 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002855 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08002856 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08002858 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2859 if (nofs)
2860 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002861
NeilBrown7e53cac2006-03-25 03:07:57 -08002862retry:
Nick Pigginafddba42007-10-16 01:25:01 -07002863 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08002864 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002865 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07002866 goto fail;
2867
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868 kaddr = kmap_atomic(page, KM_USER0);
2869 memcpy(kaddr, symname, len-1);
2870 kunmap_atomic(kaddr, KM_USER0);
Nick Pigginafddba42007-10-16 01:25:01 -07002871
2872 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2873 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002874 if (err < 0)
2875 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07002876 if (err < len-1)
2877 goto retry;
2878
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879 mark_inode_dirty(inode);
2880 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881fail:
2882 return err;
2883}
2884
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002885int page_symlink(struct inode *inode, const char *symname, int len)
2886{
2887 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08002888 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002889}
2890
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002891const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002892 .readlink = generic_readlink,
2893 .follow_link = page_follow_link_light,
2894 .put_link = page_put_link,
2895};
2896
Al Viro2d8f3032008-07-22 09:59:21 -04002897EXPORT_SYMBOL(user_path_at);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002898EXPORT_SYMBOL(follow_down);
2899EXPORT_SYMBOL(follow_up);
2900EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2901EXPORT_SYMBOL(getname);
2902EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002903EXPORT_SYMBOL(lookup_one_len);
2904EXPORT_SYMBOL(page_follow_link_light);
2905EXPORT_SYMBOL(page_put_link);
2906EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002907EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002908EXPORT_SYMBOL(page_symlink);
2909EXPORT_SYMBOL(page_symlink_inode_operations);
2910EXPORT_SYMBOL(path_lookup);
Al Virod1811462008-08-02 00:49:18 -04002911EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002912EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04002913EXPORT_SYMBOL(inode_permission);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -08002914EXPORT_SYMBOL(file_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002915EXPORT_SYMBOL(unlock_rename);
2916EXPORT_SYMBOL(vfs_create);
2917EXPORT_SYMBOL(vfs_follow_link);
2918EXPORT_SYMBOL(vfs_link);
2919EXPORT_SYMBOL(vfs_mkdir);
2920EXPORT_SYMBOL(vfs_mknod);
2921EXPORT_SYMBOL(generic_permission);
2922EXPORT_SYMBOL(vfs_readlink);
2923EXPORT_SYMBOL(vfs_rename);
2924EXPORT_SYMBOL(vfs_rmdir);
2925EXPORT_SYMBOL(vfs_symlink);
2926EXPORT_SYMBOL(vfs_unlink);
2927EXPORT_SYMBOL(dentry_unhash);
2928EXPORT_SYMBOL(generic_readlink);