blob: 94a5e60779f917eb9336dc9d6e4bc8bde09482e4 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
18#include <linux/module.h>
19#include <linux/slab.h>
20#include <linux/fs.h>
21#include <linux/namei.h>
22#include <linux/quotaops.h>
23#include <linux/pagemap.h>
Robert Love0eeca282005-07-12 17:06:03 -040024#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/personality.h>
26#include <linux/security.h>
Mimi Zohar6146f0d2009-02-04 09:06:57 -050027#include <linux/ima.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/syscalls.h>
29#include <linux/mount.h>
30#include <linux/audit.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080031#include <linux/capability.h>
Trond Myklebust834f2a42005-10-18 14:20:16 -070032#include <linux/file.h>
Ulrich Drepper5590ff02006-01-18 17:43:53 -080033#include <linux/fcntl.h>
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -070034#include <linux/device_cgroup.h>
Al Viro5ad4e532009-03-29 19:50:06 -040035#include <linux/fs_struct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <asm/uaccess.h>
37
Eric Parise81e3f42009-12-04 15:47:36 -050038#include "internal.h"
39
Linus Torvalds1da177e2005-04-16 15:20:36 -070040/* [Feb-1997 T. Schoebel-Theuer]
41 * Fundamental changes in the pathname lookup mechanisms (namei)
42 * were necessary because of omirr. The reason is that omirr needs
43 * to know the _real_ pathname, not the user-supplied one, in case
44 * of symlinks (and also when transname replacements occur).
45 *
46 * The new code replaces the old recursive symlink resolution with
47 * an iterative one (in case of non-nested symlink chains). It does
48 * this with calls to <fs>_follow_link().
49 * As a side effect, dir_namei(), _namei() and follow_link() are now
50 * replaced with a single function lookup_dentry() that can handle all
51 * the special cases of the former code.
52 *
53 * With the new dcache, the pathname is stored at each inode, at least as
54 * long as the refcount of the inode is positive. As a side effect, the
55 * size of the dcache depends on the inode cache and thus is dynamic.
56 *
57 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
58 * resolution to correspond with current state of the code.
59 *
60 * Note that the symlink resolution is not *completely* iterative.
61 * There is still a significant amount of tail- and mid- recursion in
62 * the algorithm. Also, note that <fs>_readlink() is not used in
63 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
64 * may return different results than <fs>_follow_link(). Many virtual
65 * filesystems (including /proc) exhibit this behavior.
66 */
67
68/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
69 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
70 * and the name already exists in form of a symlink, try to create the new
71 * name indicated by the symlink. The old code always complained that the
72 * name already exists, due to not following the symlink even if its target
73 * is nonexistent. The new semantics affects also mknod() and link() when
74 * the name is a symlink pointing to a non-existant name.
75 *
76 * I don't know which semantics is the right one, since I have no access
77 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
78 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
79 * "old" one. Personally, I think the new semantics is much more logical.
80 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
81 * file does succeed in both HP-UX and SunOs, but not in Solaris
82 * and in the old Linux semantics.
83 */
84
85/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
86 * semantics. See the comments in "open_namei" and "do_link" below.
87 *
88 * [10-Sep-98 Alan Modra] Another symlink change.
89 */
90
91/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
92 * inside the path - always follow.
93 * in the last component in creation/removal/renaming - never follow.
94 * if LOOKUP_FOLLOW passed - follow.
95 * if the pathname has trailing slashes - follow.
96 * otherwise - don't follow.
97 * (applied in that order).
98 *
99 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
100 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
101 * During the 2.4 we need to fix the userland stuff depending on it -
102 * hopefully we will be able to get rid of that wart in 2.5. So far only
103 * XEmacs seems to be relying on it...
104 */
105/*
106 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800107 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 * any extra contention...
109 */
110
111/* In order to reduce some races, while at the same time doing additional
112 * checking and hopefully speeding things up, we copy filenames to the
113 * kernel data space before using them..
114 *
115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
116 * PATH_MAX includes the nul terminator --RR.
117 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800118static int do_getname(const char __user *filename, char *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119{
120 int retval;
121 unsigned long len = PATH_MAX;
122
123 if (!segment_eq(get_fs(), KERNEL_DS)) {
124 if ((unsigned long) filename >= TASK_SIZE)
125 return -EFAULT;
126 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
127 len = TASK_SIZE - (unsigned long) filename;
128 }
129
130 retval = strncpy_from_user(page, filename, len);
131 if (retval > 0) {
132 if (retval < len)
133 return 0;
134 return -ENAMETOOLONG;
135 } else if (!retval)
136 retval = -ENOENT;
137 return retval;
138}
139
140char * getname(const char __user * filename)
141{
142 char *tmp, *result;
143
144 result = ERR_PTR(-ENOMEM);
145 tmp = __getname();
146 if (tmp) {
147 int retval = do_getname(filename, tmp);
148
149 result = tmp;
150 if (retval < 0) {
151 __putname(tmp);
152 result = ERR_PTR(retval);
153 }
154 }
155 audit_getname(result);
156 return result;
157}
158
159#ifdef CONFIG_AUDITSYSCALL
160void putname(const char *name)
161{
Al Viro5ac3a9c2006-07-16 06:38:45 -0400162 if (unlikely(!audit_dummy_context()))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 audit_putname(name);
164 else
165 __putname(name);
166}
167EXPORT_SYMBOL(putname);
168#endif
169
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700170/*
171 * This does basic POSIX ACL permission checking
172 */
173static int acl_permission_check(struct inode *inode, int mask,
174 int (*check_acl)(struct inode *inode, int mask))
175{
176 umode_t mode = inode->i_mode;
177
178 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
179
180 if (current_fsuid() == inode->i_uid)
181 mode >>= 6;
182 else {
183 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
184 int error = check_acl(inode, mask);
185 if (error != -EAGAIN)
186 return error;
187 }
188
189 if (in_group_p(inode->i_gid))
190 mode >>= 3;
191 }
192
193 /*
194 * If the DACs are ok we don't need any capability check.
195 */
196 if ((mask & ~mode) == 0)
197 return 0;
198 return -EACCES;
199}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200
201/**
202 * generic_permission - check for access rights on a Posix-like filesystem
203 * @inode: inode to check access rights for
204 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
205 * @check_acl: optional callback to check for Posix ACLs
206 *
207 * Used to check for read/write/execute permissions on a file.
208 * We use "fsuid" for this, letting us set arbitrary permissions
209 * for filesystem access without changing the "normal" uids which
210 * are used for other things..
211 */
212int generic_permission(struct inode *inode, int mask,
213 int (*check_acl)(struct inode *inode, int mask))
214{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700215 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216
217 /*
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700218 * Do the basic POSIX ACL permission checks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 */
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700220 ret = acl_permission_check(inode, mask, check_acl);
221 if (ret != -EACCES)
222 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 /*
225 * Read/write DACs are always overridable.
226 * Executable DACs are overridable if at least one exec bit is set.
227 */
Miklos Szeredif696a362008-07-31 13:41:58 +0200228 if (!(mask & MAY_EXEC) || execute_ok(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 if (capable(CAP_DAC_OVERRIDE))
230 return 0;
231
232 /*
233 * Searching includes executable on directories, else just read.
234 */
Serge E. Hallyn7ea66002009-12-29 14:50:19 -0600235 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
237 if (capable(CAP_DAC_READ_SEARCH))
238 return 0;
239
240 return -EACCES;
241}
242
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200243/**
244 * inode_permission - check for access rights to a given inode
245 * @inode: inode to check permission on
246 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
247 *
248 * Used to check for read/write/execute permissions on an inode.
249 * We use "fsuid" for this, letting us set arbitrary permissions
250 * for filesystem access without changing the "normal" uids which
251 * are used for other things.
252 */
Al Virof419a2e2008-07-22 00:07:17 -0400253int inode_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254{
Al Viroe6305c42008-07-15 21:03:57 -0400255 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256
257 if (mask & MAY_WRITE) {
Miklos Szeredi22590e42007-10-16 23:27:08 -0700258 umode_t mode = inode->i_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259
260 /*
261 * Nobody gets write access to a read-only fs.
262 */
263 if (IS_RDONLY(inode) &&
264 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
265 return -EROFS;
266
267 /*
268 * Nobody gets write access to an immutable file.
269 */
270 if (IS_IMMUTABLE(inode))
271 return -EACCES;
272 }
273
Al Viroacfa4382008-12-04 10:06:33 -0500274 if (inode->i_op->permission)
Al Virob77b0642008-07-17 09:37:02 -0400275 retval = inode->i_op->permission(inode, mask);
Miklos Szeredif696a362008-07-31 13:41:58 +0200276 else
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700277 retval = generic_permission(inode, mask, inode->i_op->check_acl);
Miklos Szeredif696a362008-07-31 13:41:58 +0200278
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 if (retval)
280 return retval;
281
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -0700282 retval = devcgroup_inode_permission(inode, mask);
283 if (retval)
284 return retval;
285
Al Viroe6305c42008-07-15 21:03:57 -0400286 return security_inode_permission(inode,
Stephen Smalleyf418b002008-07-28 13:32:38 -0400287 mask & (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288}
289
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800290/**
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800291 * file_permission - check for additional access rights to a given file
292 * @file: file to check access rights for
293 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
294 *
295 * Used to check for read/write/execute permissions on an already opened
296 * file.
297 *
298 * Note:
299 * Do not use this function in new code. All access checks should
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200300 * be done using inode_permission().
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800301 */
302int file_permission(struct file *file, int mask)
303{
Al Virof419a2e2008-07-22 00:07:17 -0400304 return inode_permission(file->f_path.dentry->d_inode, mask);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800305}
306
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307/*
308 * get_write_access() gets write permission for a file.
309 * put_write_access() releases this write permission.
310 * This is used for regular files.
311 * We cannot support write (and maybe mmap read-write shared) accesses and
312 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
313 * can have the following values:
314 * 0: no writers, no VM_DENYWRITE mappings
315 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
316 * > 0: (i_writecount) users are writing to the file.
317 *
318 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
319 * except for the cases where we don't hold i_writecount yet. Then we need to
320 * use {get,deny}_write_access() - these functions check the sign and refuse
321 * to do the change if sign is wrong. Exclusion between them is provided by
322 * the inode->i_lock spinlock.
323 */
324
325int get_write_access(struct inode * inode)
326{
327 spin_lock(&inode->i_lock);
328 if (atomic_read(&inode->i_writecount) < 0) {
329 spin_unlock(&inode->i_lock);
330 return -ETXTBSY;
331 }
332 atomic_inc(&inode->i_writecount);
333 spin_unlock(&inode->i_lock);
334
335 return 0;
336}
337
338int deny_write_access(struct file * file)
339{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800340 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341
342 spin_lock(&inode->i_lock);
343 if (atomic_read(&inode->i_writecount) > 0) {
344 spin_unlock(&inode->i_lock);
345 return -ETXTBSY;
346 }
347 atomic_dec(&inode->i_writecount);
348 spin_unlock(&inode->i_lock);
349
350 return 0;
351}
352
Jan Blunck1d957f92008-02-14 19:34:35 -0800353/**
Jan Blunck5dd784d02008-02-14 19:34:38 -0800354 * path_get - get a reference to a path
355 * @path: path to get the reference to
356 *
357 * Given a path increment the reference count to the dentry and the vfsmount.
358 */
359void path_get(struct path *path)
360{
361 mntget(path->mnt);
362 dget(path->dentry);
363}
364EXPORT_SYMBOL(path_get);
365
366/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800367 * path_put - put a reference to a path
368 * @path: path to put the reference to
369 *
370 * Given a path decrement the reference count to the dentry and the vfsmount.
371 */
372void path_put(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373{
Jan Blunck1d957f92008-02-14 19:34:35 -0800374 dput(path->dentry);
375 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376}
Jan Blunck1d957f92008-02-14 19:34:35 -0800377EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378
Trond Myklebust834f2a42005-10-18 14:20:16 -0700379/**
380 * release_open_intent - free up open intent resources
381 * @nd: pointer to nameidata
382 */
383void release_open_intent(struct nameidata *nd)
384{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800385 if (nd->intent.open.file->f_path.dentry == NULL)
Trond Myklebust834f2a42005-10-18 14:20:16 -0700386 put_filp(nd->intent.open.file);
387 else
388 fput(nd->intent.open.file);
389}
390
Ian Kentbcdc5e02006-09-27 01:50:44 -0700391static inline struct dentry *
392do_revalidate(struct dentry *dentry, struct nameidata *nd)
393{
394 int status = dentry->d_op->d_revalidate(dentry, nd);
395 if (unlikely(status <= 0)) {
396 /*
397 * The dentry failed validation.
398 * If d_revalidate returned 0 attempt to invalidate
399 * the dentry otherwise d_revalidate is asking us
400 * to return a fail status.
401 */
402 if (!status) {
403 if (!d_invalidate(dentry)) {
404 dput(dentry);
405 dentry = NULL;
406 }
407 } else {
408 dput(dentry);
409 dentry = ERR_PTR(status);
410 }
411 }
412 return dentry;
413}
414
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415/*
Jeff Layton39159de2009-12-07 12:01:50 -0500416 * force_reval_path - force revalidation of a dentry
417 *
418 * In some situations the path walking code will trust dentries without
419 * revalidating them. This causes problems for filesystems that depend on
420 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
421 * (which indicates that it's possible for the dentry to go stale), force
422 * a d_revalidate call before proceeding.
423 *
424 * Returns 0 if the revalidation was successful. If the revalidation fails,
425 * either return the error returned by d_revalidate or -ESTALE if the
426 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
427 * invalidate the dentry. It's up to the caller to handle putting references
428 * to the path if necessary.
429 */
430static int
431force_reval_path(struct path *path, struct nameidata *nd)
432{
433 int status;
434 struct dentry *dentry = path->dentry;
435
436 /*
437 * only check on filesystems where it's possible for the dentry to
438 * become stale. It's assumed that if this flag is set then the
439 * d_revalidate op will also be defined.
440 */
441 if (!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT))
442 return 0;
443
444 status = dentry->d_op->d_revalidate(dentry, nd);
445 if (status > 0)
446 return 0;
447
448 if (!status) {
449 d_invalidate(dentry);
450 status = -ESTALE;
451 }
452 return status;
453}
454
455/*
Al Virob75b5082009-12-16 01:01:38 -0500456 * Short-cut version of permission(), for calling on directories
457 * during pathname resolution. Combines parts of permission()
458 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 *
460 * If appropriate, check DAC only. If not appropriate, or
Al Virob75b5082009-12-16 01:01:38 -0500461 * short-cut DAC fails, then call ->permission() to do more
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 * complete permission check.
463 */
Al Virob75b5082009-12-16 01:01:38 -0500464static int exec_permission(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700466 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467
Linus Torvaldscb9179e2009-08-28 11:08:31 -0700468 if (inode->i_op->permission) {
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700469 ret = inode->i_op->permission(inode, MAY_EXEC);
Linus Torvaldscb9179e2009-08-28 11:08:31 -0700470 if (!ret)
471 goto ok;
472 return ret;
473 }
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700474 ret = acl_permission_check(inode, MAY_EXEC, inode->i_op->check_acl);
475 if (!ret)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 goto ok;
477
Linus Torvaldsf1ac9f62009-08-28 10:53:56 -0700478 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 goto ok;
480
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700481 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482ok:
Al Virob77b0642008-07-17 09:37:02 -0400483 return security_inode_permission(inode, MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484}
485
Al Viro2a737872009-04-07 11:49:53 -0400486static __always_inline void set_root(struct nameidata *nd)
487{
488 if (!nd->root.mnt) {
489 struct fs_struct *fs = current->fs;
490 read_lock(&fs->lock);
491 nd->root = fs->root;
492 path_get(&nd->root);
493 read_unlock(&fs->lock);
494 }
495}
496
Al Viro6de88d72009-08-09 01:41:57 +0400497static int link_path_walk(const char *, struct nameidata *);
498
Arjan van de Venf1662352006-01-14 13:21:31 -0800499static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500{
501 int res = 0;
502 char *name;
503 if (IS_ERR(link))
504 goto fail;
505
506 if (*link == '/') {
Al Viro2a737872009-04-07 11:49:53 -0400507 set_root(nd);
Jan Blunck1d957f92008-02-14 19:34:35 -0800508 path_put(&nd->path);
Al Viro2a737872009-04-07 11:49:53 -0400509 nd->path = nd->root;
510 path_get(&nd->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 }
Christoph Hellwigb4091d52008-11-05 15:07:21 +0100512
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 res = link_path_walk(link, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 if (nd->depth || res || nd->last_type!=LAST_NORM)
515 return res;
516 /*
517 * If it is an iterative symlinks resolution in open_namei() we
518 * have to copy the last component. And all that crap because of
519 * bloody create() on broken symlinks. Furrfu...
520 */
521 name = __getname();
522 if (unlikely(!name)) {
Jan Blunck1d957f92008-02-14 19:34:35 -0800523 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 return -ENOMEM;
525 }
526 strcpy(name, nd->last.name);
527 nd->last.name = name;
528 return 0;
529fail:
Jan Blunck1d957f92008-02-14 19:34:35 -0800530 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 return PTR_ERR(link);
532}
533
Jan Blunck1d957f92008-02-14 19:34:35 -0800534static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700535{
536 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800537 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700538 mntput(path->mnt);
539}
540
541static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
542{
Jan Blunck4ac91372008-02-14 19:34:32 -0800543 dput(nd->path.dentry);
544 if (nd->path.mnt != path->mnt)
545 mntput(nd->path.mnt);
546 nd->path.mnt = path->mnt;
547 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700548}
549
Ian Kent051d3812006-03-27 01:14:53 -0800550static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
551{
552 int error;
553 void *cookie;
554 struct dentry *dentry = path->dentry;
555
556 touch_atime(path->mnt, dentry);
557 nd_set_link(nd, NULL);
558
Jan Blunck4ac91372008-02-14 19:34:32 -0800559 if (path->mnt != nd->path.mnt) {
Ian Kent051d3812006-03-27 01:14:53 -0800560 path_to_nameidata(path, nd);
561 dget(dentry);
562 }
563 mntget(path->mnt);
Al Viro86acdca12009-12-22 23:45:11 -0500564 nd->last_type = LAST_BIND;
Ian Kent051d3812006-03-27 01:14:53 -0800565 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
566 error = PTR_ERR(cookie);
567 if (!IS_ERR(cookie)) {
568 char *s = nd_get_link(nd);
569 error = 0;
570 if (s)
571 error = __vfs_follow_link(nd, s);
Jeff Layton39159de2009-12-07 12:01:50 -0500572 else if (nd->last_type == LAST_BIND) {
573 error = force_reval_path(&nd->path, nd);
574 if (error)
575 path_put(&nd->path);
576 }
Ian Kent051d3812006-03-27 01:14:53 -0800577 if (dentry->d_inode->i_op->put_link)
578 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
579 }
Ian Kent051d3812006-03-27 01:14:53 -0800580 return error;
581}
582
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583/*
584 * This limits recursive symlink follows to 8, while
585 * limiting consecutive symlinks to 40.
586 *
587 * Without that kind of total limit, nasty chains of consecutive
588 * symlinks can cause almost arbitrarily long lookups.
589 */
Al Viro90ebe562005-06-06 13:35:58 -0700590static inline int do_follow_link(struct path *path, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591{
592 int err = -ELOOP;
593 if (current->link_count >= MAX_NESTED_LINKS)
594 goto loop;
595 if (current->total_link_count >= 40)
596 goto loop;
597 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
598 cond_resched();
Al Viro90ebe562005-06-06 13:35:58 -0700599 err = security_inode_follow_link(path->dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 if (err)
601 goto loop;
602 current->link_count++;
603 current->total_link_count++;
604 nd->depth++;
Al Virocd4e91d2005-06-06 13:36:03 -0700605 err = __do_follow_link(path, nd);
Al Viro258fa992009-08-09 01:32:02 +0400606 path_put(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 current->link_count--;
608 nd->depth--;
609 return err;
610loop:
Jan Blunck1d957f92008-02-14 19:34:35 -0800611 path_put_conditional(path, nd);
612 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 return err;
614}
615
Al Virobab77eb2009-04-18 03:26:48 -0400616int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617{
618 struct vfsmount *parent;
619 struct dentry *mountpoint;
620 spin_lock(&vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400621 parent = path->mnt->mnt_parent;
622 if (parent == path->mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 spin_unlock(&vfsmount_lock);
624 return 0;
625 }
626 mntget(parent);
Al Virobab77eb2009-04-18 03:26:48 -0400627 mountpoint = dget(path->mnt->mnt_mountpoint);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 spin_unlock(&vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400629 dput(path->dentry);
630 path->dentry = mountpoint;
631 mntput(path->mnt);
632 path->mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 return 1;
634}
635
636/* no need for dcache_lock, as serialization is taken care in
637 * namespace.c
638 */
Al Viro463ffb22005-06-06 13:36:05 -0700639static int __follow_mount(struct path *path)
640{
641 int res = 0;
642 while (d_mountpoint(path->dentry)) {
Al Viro1c755af2009-04-18 14:06:57 -0400643 struct vfsmount *mounted = lookup_mnt(path);
Al Viro463ffb22005-06-06 13:36:05 -0700644 if (!mounted)
645 break;
646 dput(path->dentry);
647 if (res)
648 mntput(path->mnt);
649 path->mnt = mounted;
650 path->dentry = dget(mounted->mnt_root);
651 res = 1;
652 }
653 return res;
654}
655
Al Viro79ed0222009-04-18 13:59:41 -0400656static void follow_mount(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657{
Al Viro79ed0222009-04-18 13:59:41 -0400658 while (d_mountpoint(path->dentry)) {
Al Viro1c755af2009-04-18 14:06:57 -0400659 struct vfsmount *mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 if (!mounted)
661 break;
Al Viro79ed0222009-04-18 13:59:41 -0400662 dput(path->dentry);
663 mntput(path->mnt);
664 path->mnt = mounted;
665 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667}
668
669/* no need for dcache_lock, as serialization is taken care in
670 * namespace.c
671 */
Al Viro9393bd02009-04-18 13:58:15 -0400672int follow_down(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673{
674 struct vfsmount *mounted;
675
Al Viro1c755af2009-04-18 14:06:57 -0400676 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -0400678 dput(path->dentry);
679 mntput(path->mnt);
680 path->mnt = mounted;
681 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 return 1;
683 }
684 return 0;
685}
686
Arjan van de Venf1662352006-01-14 13:21:31 -0800687static __always_inline void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688{
Al Viro2a737872009-04-07 11:49:53 -0400689 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -0700690
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 while(1) {
692 struct vfsmount *parent;
Jan Blunck4ac91372008-02-14 19:34:32 -0800693 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694
Al Viro2a737872009-04-07 11:49:53 -0400695 if (nd->path.dentry == nd->root.dentry &&
696 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 break;
698 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 spin_lock(&dcache_lock);
Jan Blunck4ac91372008-02-14 19:34:32 -0800700 if (nd->path.dentry != nd->path.mnt->mnt_root) {
701 nd->path.dentry = dget(nd->path.dentry->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 spin_unlock(&dcache_lock);
703 dput(old);
704 break;
705 }
706 spin_unlock(&dcache_lock);
707 spin_lock(&vfsmount_lock);
Jan Blunck4ac91372008-02-14 19:34:32 -0800708 parent = nd->path.mnt->mnt_parent;
709 if (parent == nd->path.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 spin_unlock(&vfsmount_lock);
711 break;
712 }
713 mntget(parent);
Jan Blunck4ac91372008-02-14 19:34:32 -0800714 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 spin_unlock(&vfsmount_lock);
716 dput(old);
Jan Blunck4ac91372008-02-14 19:34:32 -0800717 mntput(nd->path.mnt);
718 nd->path.mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 }
Al Viro79ed0222009-04-18 13:59:41 -0400720 follow_mount(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721}
722
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723/*
724 * It's more convoluted than I'd like it to be, but... it's still fairly
725 * small and for now I'd prefer to have fast path as straight as possible.
726 * It _is_ time-critical.
727 */
728static int do_lookup(struct nameidata *nd, struct qstr *name,
729 struct path *path)
730{
Jan Blunck4ac91372008-02-14 19:34:32 -0800731 struct vfsmount *mnt = nd->path.mnt;
Al Viro6e6b1bd2009-08-13 23:38:37 +0400732 struct dentry *dentry, *parent;
733 struct inode *dir;
Al Viro3cac2602009-08-13 18:27:43 +0400734 /*
735 * See if the low-level filesystem might want
736 * to use its own hash..
737 */
738 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
739 int err = nd->path.dentry->d_op->d_hash(nd->path.dentry, name);
740 if (err < 0)
741 return err;
742 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743
Al Viro3cac2602009-08-13 18:27:43 +0400744 dentry = __d_lookup(nd->path.dentry, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 if (!dentry)
746 goto need_lookup;
747 if (dentry->d_op && dentry->d_op->d_revalidate)
748 goto need_revalidate;
749done:
750 path->mnt = mnt;
751 path->dentry = dentry;
Al Viro634ee702005-06-06 13:36:13 -0700752 __follow_mount(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 return 0;
754
755need_lookup:
Al Viro6e6b1bd2009-08-13 23:38:37 +0400756 parent = nd->path.dentry;
757 dir = parent->d_inode;
758
759 mutex_lock(&dir->i_mutex);
760 /*
761 * First re-do the cached lookup just in case it was created
762 * while we waited for the directory semaphore..
763 *
764 * FIXME! This could use version numbering or similar to
765 * avoid unnecessary cache lookups.
766 *
767 * The "dcache_lock" is purely to protect the RCU list walker
768 * from concurrent renames at this point (we mustn't get false
769 * negatives from the RCU list walk here, unlike the optimistic
770 * fast walk).
771 *
772 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
773 */
774 dentry = d_lookup(parent, name);
775 if (!dentry) {
776 struct dentry *new;
777
778 /* Don't create child dentry for a dead directory. */
779 dentry = ERR_PTR(-ENOENT);
780 if (IS_DEADDIR(dir))
781 goto out_unlock;
782
783 new = d_alloc(parent, name);
784 dentry = ERR_PTR(-ENOMEM);
785 if (new) {
786 dentry = dir->i_op->lookup(dir, new, nd);
787 if (dentry)
788 dput(new);
789 else
790 dentry = new;
791 }
792out_unlock:
793 mutex_unlock(&dir->i_mutex);
794 if (IS_ERR(dentry))
795 goto fail;
796 goto done;
797 }
798
799 /*
800 * Uhhuh! Nasty case: the cache was re-populated while
801 * we waited on the semaphore. Need to revalidate.
802 */
803 mutex_unlock(&dir->i_mutex);
804 if (dentry->d_op && dentry->d_op->d_revalidate) {
805 dentry = do_revalidate(dentry, nd);
806 if (!dentry)
807 dentry = ERR_PTR(-ENOENT);
808 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 if (IS_ERR(dentry))
810 goto fail;
811 goto done;
812
813need_revalidate:
Ian Kentbcdc5e02006-09-27 01:50:44 -0700814 dentry = do_revalidate(dentry, nd);
815 if (!dentry)
816 goto need_lookup;
817 if (IS_ERR(dentry))
818 goto fail;
819 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820
821fail:
822 return PTR_ERR(dentry);
823}
824
825/*
826 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +0100827 * This is the basic name resolution function, turning a pathname into
828 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 *
Prasanna Medaea3834d2005-04-29 16:00:17 +0100830 * Returns 0 and nd will have valid dentry and mnt on success.
831 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 */
Al Viro6de88d72009-08-09 01:41:57 +0400833static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834{
835 struct path next;
836 struct inode *inode;
837 int err;
838 unsigned int lookup_flags = nd->flags;
839
840 while (*name=='/')
841 name++;
842 if (!*name)
843 goto return_reval;
844
Jan Blunck4ac91372008-02-14 19:34:32 -0800845 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 if (nd->depth)
Trond Myklebustf55eab82006-02-04 23:28:01 -0800847 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848
849 /* At this point we know we have a real path component. */
850 for(;;) {
851 unsigned long hash;
852 struct qstr this;
853 unsigned int c;
854
Trond Myklebustcdce5d62005-10-18 14:20:18 -0700855 nd->flags |= LOOKUP_CONTINUE;
Al Virob75b5082009-12-16 01:01:38 -0500856 err = exec_permission(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 if (err)
858 break;
859
860 this.name = name;
861 c = *(const unsigned char *)name;
862
863 hash = init_name_hash();
864 do {
865 name++;
866 hash = partial_name_hash(c, hash);
867 c = *(const unsigned char *)name;
868 } while (c && (c != '/'));
869 this.len = name - (const char *) this.name;
870 this.hash = end_name_hash(hash);
871
872 /* remove trailing slashes? */
873 if (!c)
874 goto last_component;
875 while (*++name == '/');
876 if (!*name)
877 goto last_with_slashes;
878
879 /*
880 * "." and ".." are special - ".." especially so because it has
881 * to be able to know about the current root directory and
882 * parent relationships.
883 */
884 if (this.name[0] == '.') switch (this.len) {
885 default:
886 break;
887 case 2:
888 if (this.name[1] != '.')
889 break;
Al Viro58c465e2005-06-06 13:36:13 -0700890 follow_dotdot(nd);
Jan Blunck4ac91372008-02-14 19:34:32 -0800891 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 /* fallthrough */
893 case 1:
894 continue;
895 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 /* This does the actual lookups.. */
897 err = do_lookup(nd, &this, &next);
898 if (err)
899 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900
901 err = -ENOENT;
902 inode = next.dentry->d_inode;
903 if (!inode)
904 goto out_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905
906 if (inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700907 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 if (err)
909 goto return_err;
910 err = -ENOENT;
Jan Blunck4ac91372008-02-14 19:34:32 -0800911 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 if (!inode)
913 break;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700914 } else
915 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 err = -ENOTDIR;
917 if (!inode->i_op->lookup)
918 break;
919 continue;
920 /* here ends the main loop */
921
922last_with_slashes:
923 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
924last_component:
Trond Myklebustf55eab82006-02-04 23:28:01 -0800925 /* Clear LOOKUP_CONTINUE iff it was previously unset */
926 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 if (lookup_flags & LOOKUP_PARENT)
928 goto lookup_parent;
929 if (this.name[0] == '.') switch (this.len) {
930 default:
931 break;
932 case 2:
933 if (this.name[1] != '.')
934 break;
Al Viro58c465e2005-06-06 13:36:13 -0700935 follow_dotdot(nd);
Jan Blunck4ac91372008-02-14 19:34:32 -0800936 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 /* fallthrough */
938 case 1:
939 goto return_reval;
940 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 err = do_lookup(nd, &this, &next);
942 if (err)
943 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 inode = next.dentry->d_inode;
945 if ((lookup_flags & LOOKUP_FOLLOW)
Al Viroacfa4382008-12-04 10:06:33 -0500946 && inode && inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700947 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 if (err)
949 goto return_err;
Jan Blunck4ac91372008-02-14 19:34:32 -0800950 inode = nd->path.dentry->d_inode;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700951 } else
952 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 err = -ENOENT;
954 if (!inode)
955 break;
956 if (lookup_flags & LOOKUP_DIRECTORY) {
957 err = -ENOTDIR;
Al Viroacfa4382008-12-04 10:06:33 -0500958 if (!inode->i_op->lookup)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 break;
960 }
961 goto return_base;
962lookup_parent:
963 nd->last = this;
964 nd->last_type = LAST_NORM;
965 if (this.name[0] != '.')
966 goto return_base;
967 if (this.len == 1)
968 nd->last_type = LAST_DOT;
969 else if (this.len == 2 && this.name[1] == '.')
970 nd->last_type = LAST_DOTDOT;
971 else
972 goto return_base;
973return_reval:
974 /*
975 * We bypassed the ordinary revalidation routines.
976 * We may need to check the cached dentry for staleness.
977 */
Jan Blunck4ac91372008-02-14 19:34:32 -0800978 if (nd->path.dentry && nd->path.dentry->d_sb &&
979 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 err = -ESTALE;
981 /* Note: we do not d_invalidate() */
Jan Blunck4ac91372008-02-14 19:34:32 -0800982 if (!nd->path.dentry->d_op->d_revalidate(
983 nd->path.dentry, nd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 break;
985 }
986return_base:
987 return 0;
988out_dput:
Jan Blunck1d957f92008-02-14 19:34:35 -0800989 path_put_conditional(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 break;
991 }
Jan Blunck1d957f92008-02-14 19:34:35 -0800992 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993return_err:
994 return err;
995}
996
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800997static int path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998{
Al Viro6de88d72009-08-09 01:41:57 +0400999 struct path save = nd->path;
1000 int result;
1001
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 current->total_link_count = 0;
Al Viro6de88d72009-08-09 01:41:57 +04001003
1004 /* make sure the stuff we saved doesn't go away */
1005 path_get(&save);
1006
1007 result = link_path_walk(name, nd);
1008 if (result == -ESTALE) {
1009 /* nd->path had been dropped */
1010 current->total_link_count = 0;
1011 nd->path = save;
1012 path_get(&nd->path);
1013 nd->flags |= LOOKUP_REVAL;
1014 result = link_path_walk(name, nd);
1015 }
1016
1017 path_put(&save);
1018
1019 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020}
1021
Al Viro9b4a9b12009-04-07 11:44:16 -04001022static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001024 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001025 int fput_needed;
1026 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027
1028 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1029 nd->flags = flags;
1030 nd->depth = 0;
Al Viro2a737872009-04-07 11:49:53 -04001031 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 if (*name=='/') {
Al Viro2a737872009-04-07 11:49:53 -04001034 set_root(nd);
1035 nd->path = nd->root;
1036 path_get(&nd->root);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001037 } else if (dfd == AT_FDCWD) {
Al Viro2a737872009-04-07 11:49:53 -04001038 struct fs_struct *fs = current->fs;
Andreas Mohre518ddb2006-09-29 02:01:22 -07001039 read_lock(&fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08001040 nd->path = fs->pwd;
1041 path_get(&fs->pwd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001042 read_unlock(&fs->lock);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001043 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001044 struct dentry *dentry;
1045
1046 file = fget_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001047 retval = -EBADF;
1048 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001049 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001050
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001051 dentry = file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001052
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001053 retval = -ENOTDIR;
1054 if (!S_ISDIR(dentry->d_inode->i_mode))
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001055 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001056
1057 retval = file_permission(file, MAY_EXEC);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001058 if (retval)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001059 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001060
Jan Blunck5dd784d02008-02-14 19:34:38 -08001061 nd->path = file->f_path;
1062 path_get(&file->f_path);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001063
1064 fput_light(file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001066 return 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001067
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001068fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001069 fput_light(file, fput_needed);
Al Viro9b4a9b12009-04-07 11:44:16 -04001070out_fail:
1071 return retval;
1072}
1073
1074/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1075static int do_path_lookup(int dfd, const char *name,
1076 unsigned int flags, struct nameidata *nd)
1077{
1078 int retval = path_init(dfd, name, flags, nd);
1079 if (!retval)
1080 retval = path_walk(name, nd);
1081 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1082 nd->path.dentry->d_inode))
1083 audit_inode(name, nd->path.dentry);
Al Viro2a737872009-04-07 11:49:53 -04001084 if (nd->root.mnt) {
1085 path_put(&nd->root);
1086 nd->root.mnt = NULL;
1087 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001088 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089}
1090
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -08001091int path_lookup(const char *name, unsigned int flags,
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001092 struct nameidata *nd)
1093{
1094 return do_path_lookup(AT_FDCWD, name, flags, nd);
1095}
1096
Al Virod1811462008-08-02 00:49:18 -04001097int kern_path(const char *name, unsigned int flags, struct path *path)
1098{
1099 struct nameidata nd;
1100 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1101 if (!res)
1102 *path = nd.path;
1103 return res;
1104}
1105
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001106/**
1107 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1108 * @dentry: pointer to dentry of the base directory
1109 * @mnt: pointer to vfs mount of the base directory
1110 * @name: pointer to file name
1111 * @flags: lookup flags
1112 * @nd: pointer to nameidata
1113 */
1114int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1115 const char *name, unsigned int flags,
1116 struct nameidata *nd)
1117{
1118 int retval;
1119
1120 /* same as do_path_lookup */
1121 nd->last_type = LAST_ROOT;
1122 nd->flags = flags;
1123 nd->depth = 0;
1124
Jan Blunckc8e7f442008-06-09 16:40:35 -07001125 nd->path.dentry = dentry;
1126 nd->path.mnt = mnt;
1127 path_get(&nd->path);
Al Viro5b857112009-04-07 11:53:49 -04001128 nd->root = nd->path;
1129 path_get(&nd->root);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001130
1131 retval = path_walk(name, nd);
Jan Blunck4ac91372008-02-14 19:34:32 -08001132 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1133 nd->path.dentry->d_inode))
1134 audit_inode(name, nd->path.dentry);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001135
Al Viro5b857112009-04-07 11:53:49 -04001136 path_put(&nd->root);
1137 nd->root.mnt = NULL;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001138
Al Viro2a737872009-04-07 11:49:53 -04001139 return retval;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001140}
1141
Christoph Hellwigeead1912007-10-16 23:25:38 -07001142static struct dentry *__lookup_hash(struct qstr *name,
1143 struct dentry *base, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144{
James Morris057f6c02007-04-26 00:12:05 -07001145 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 struct inode *inode;
1147 int err;
1148
1149 inode = base->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150
1151 /*
1152 * See if the low-level filesystem might want
1153 * to use its own hash..
1154 */
1155 if (base->d_op && base->d_op->d_hash) {
1156 err = base->d_op->d_hash(base, name);
1157 dentry = ERR_PTR(err);
1158 if (err < 0)
1159 goto out;
1160 }
1161
Al Viro6e6b1bd2009-08-13 23:38:37 +04001162 dentry = __d_lookup(base, name);
1163
1164 /* lockess __d_lookup may fail due to concurrent d_move()
1165 * in some unrelated directory, so try with d_lookup
1166 */
1167 if (!dentry)
1168 dentry = d_lookup(base, name);
1169
1170 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
1171 dentry = do_revalidate(dentry, nd);
1172
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 if (!dentry) {
Miklos Szeredid70b67c2008-07-02 21:30:15 +02001174 struct dentry *new;
1175
1176 /* Don't create child dentry for a dead directory. */
1177 dentry = ERR_PTR(-ENOENT);
1178 if (IS_DEADDIR(inode))
1179 goto out;
1180
1181 new = d_alloc(base, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 dentry = ERR_PTR(-ENOMEM);
1183 if (!new)
1184 goto out;
1185 dentry = inode->i_op->lookup(inode, new, nd);
1186 if (!dentry)
1187 dentry = new;
1188 else
1189 dput(new);
1190 }
1191out:
1192 return dentry;
1193}
1194
James Morris057f6c02007-04-26 00:12:05 -07001195/*
1196 * Restricted form of lookup. Doesn't follow links, single-component only,
1197 * needs parent already locked. Doesn't follow mounts.
1198 * SMP-safe.
1199 */
Adrian Bunka244e162006-03-31 02:32:11 -08001200static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201{
Christoph Hellwigeead1912007-10-16 23:25:38 -07001202 int err;
1203
Al Virob75b5082009-12-16 01:01:38 -05001204 err = exec_permission(nd->path.dentry->d_inode);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001205 if (err)
1206 return ERR_PTR(err);
Jan Blunck4ac91372008-02-14 19:34:32 -08001207 return __lookup_hash(&nd->last, nd->path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208}
1209
Christoph Hellwigeead1912007-10-16 23:25:38 -07001210static int __lookup_one_len(const char *name, struct qstr *this,
1211 struct dentry *base, int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212{
1213 unsigned long hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 unsigned int c;
1215
James Morris057f6c02007-04-26 00:12:05 -07001216 this->name = name;
1217 this->len = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 if (!len)
James Morris057f6c02007-04-26 00:12:05 -07001219 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220
1221 hash = init_name_hash();
1222 while (len--) {
1223 c = *(const unsigned char *)name++;
1224 if (c == '/' || c == '\0')
James Morris057f6c02007-04-26 00:12:05 -07001225 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 hash = partial_name_hash(c, hash);
1227 }
James Morris057f6c02007-04-26 00:12:05 -07001228 this->hash = end_name_hash(hash);
1229 return 0;
1230}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231
Christoph Hellwigeead1912007-10-16 23:25:38 -07001232/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001233 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07001234 * @name: pathname component to lookup
1235 * @base: base directory to lookup from
1236 * @len: maximum length @len should be interpreted to
1237 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07001238 * Note that this routine is purely a helper for filesystem usage and should
1239 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07001240 * nameidata argument is passed to the filesystem methods and a filesystem
1241 * using this helper needs to be prepared for that.
1242 */
James Morris057f6c02007-04-26 00:12:05 -07001243struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1244{
1245 int err;
1246 struct qstr this;
1247
David Woodhouse2f9092e2009-04-20 23:18:37 +01001248 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1249
James Morris057f6c02007-04-26 00:12:05 -07001250 err = __lookup_one_len(name, &this, base, len);
1251 if (err)
1252 return ERR_PTR(err);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001253
Al Virob75b5082009-12-16 01:01:38 -05001254 err = exec_permission(base->d_inode);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001255 if (err)
1256 return ERR_PTR(err);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001257 return __lookup_hash(&this, base, NULL);
James Morris057f6c02007-04-26 00:12:05 -07001258}
1259
Al Viro2d8f3032008-07-22 09:59:21 -04001260int user_path_at(int dfd, const char __user *name, unsigned flags,
1261 struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262{
Al Viro2d8f3032008-07-22 09:59:21 -04001263 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 char *tmp = getname(name);
1265 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04001267
1268 BUG_ON(flags & LOOKUP_PARENT);
1269
1270 err = do_path_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04001272 if (!err)
1273 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 }
1275 return err;
1276}
1277
Al Viro2ad94ae2008-07-21 09:32:51 -04001278static int user_path_parent(int dfd, const char __user *path,
1279 struct nameidata *nd, char **name)
1280{
1281 char *s = getname(path);
1282 int error;
1283
1284 if (IS_ERR(s))
1285 return PTR_ERR(s);
1286
1287 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1288 if (error)
1289 putname(s);
1290 else
1291 *name = s;
1292
1293 return error;
1294}
1295
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296/*
1297 * It's inline, so penalty for filesystems that don't use sticky bit is
1298 * minimal.
1299 */
1300static inline int check_sticky(struct inode *dir, struct inode *inode)
1301{
David Howellsda9592e2008-11-14 10:39:05 +11001302 uid_t fsuid = current_fsuid();
1303
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 if (!(dir->i_mode & S_ISVTX))
1305 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001306 if (inode->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001308 if (dir->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 return 0;
1310 return !capable(CAP_FOWNER);
1311}
1312
1313/*
1314 * Check whether we can remove a link victim from directory dir, check
1315 * whether the type of victim is right.
1316 * 1. We can't do it if dir is read-only (done in permission())
1317 * 2. We should have write and exec permissions on dir
1318 * 3. We can't remove anything from append-only dir
1319 * 4. We can't do anything with immutable dir (done in permission())
1320 * 5. If the sticky bit on dir is set we should either
1321 * a. be owner of dir, or
1322 * b. be owner of victim, or
1323 * c. have CAP_FOWNER capability
1324 * 6. If the victim is append-only or immutable we can't do antyhing with
1325 * links pointing to it.
1326 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1327 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1328 * 9. We can't remove a root or mountpoint.
1329 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1330 * nfs_async_unlink().
1331 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001332static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333{
1334 int error;
1335
1336 if (!victim->d_inode)
1337 return -ENOENT;
1338
1339 BUG_ON(victim->d_parent->d_inode != dir);
Al Viro5a190ae2007-06-07 12:19:32 -04001340 audit_inode_child(victim->d_name.name, victim, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341
Al Virof419a2e2008-07-22 00:07:17 -04001342 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 if (error)
1344 return error;
1345 if (IS_APPEND(dir))
1346 return -EPERM;
1347 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
Hugh Dickinsf9454542008-11-19 15:36:38 -08001348 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 return -EPERM;
1350 if (isdir) {
1351 if (!S_ISDIR(victim->d_inode->i_mode))
1352 return -ENOTDIR;
1353 if (IS_ROOT(victim))
1354 return -EBUSY;
1355 } else if (S_ISDIR(victim->d_inode->i_mode))
1356 return -EISDIR;
1357 if (IS_DEADDIR(dir))
1358 return -ENOENT;
1359 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1360 return -EBUSY;
1361 return 0;
1362}
1363
1364/* Check whether we can create an object with dentry child in directory
1365 * dir.
1366 * 1. We can't do it if child already exists (open has special treatment for
1367 * this case, but since we are inlined it's OK)
1368 * 2. We can't do it if dir is read-only (done in permission())
1369 * 3. We should have write and exec permissions on dir
1370 * 4. We can't do it if dir is immutable (done in permission())
1371 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02001372static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373{
1374 if (child->d_inode)
1375 return -EEXIST;
1376 if (IS_DEADDIR(dir))
1377 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04001378 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379}
1380
1381/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 * O_DIRECTORY translates into forcing a directory lookup.
1383 */
1384static inline int lookup_flags(unsigned int f)
1385{
1386 unsigned long retval = LOOKUP_FOLLOW;
1387
1388 if (f & O_NOFOLLOW)
1389 retval &= ~LOOKUP_FOLLOW;
1390
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 if (f & O_DIRECTORY)
1392 retval |= LOOKUP_DIRECTORY;
1393
1394 return retval;
1395}
1396
1397/*
1398 * p1 and p2 should be directories on the same fs.
1399 */
1400struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1401{
1402 struct dentry *p;
1403
1404 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001405 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 return NULL;
1407 }
1408
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001409 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001411 p = d_ancestor(p2, p1);
1412 if (p) {
1413 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1414 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1415 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 }
1417
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001418 p = d_ancestor(p1, p2);
1419 if (p) {
1420 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1421 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1422 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 }
1424
Ingo Molnarf2eace22006-07-03 00:25:05 -07001425 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1426 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 return NULL;
1428}
1429
1430void unlock_rename(struct dentry *p1, struct dentry *p2)
1431{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001432 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001434 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001435 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 }
1437}
1438
1439int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1440 struct nameidata *nd)
1441{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001442 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443
1444 if (error)
1445 return error;
1446
Al Viroacfa4382008-12-04 10:06:33 -05001447 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 return -EACCES; /* shouldn't it be ENOSYS? */
1449 mode &= S_IALLUGO;
1450 mode |= S_IFREG;
1451 error = security_inode_create(dir, dentry, mode);
1452 if (error)
1453 return error;
Jan Kara9e3509e2009-01-26 16:45:12 +01001454 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001456 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001457 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 return error;
1459}
1460
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001461int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001463 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 struct inode *inode = dentry->d_inode;
1465 int error;
1466
1467 if (!inode)
1468 return -ENOENT;
1469
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001470 switch (inode->i_mode & S_IFMT) {
1471 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001473 case S_IFDIR:
1474 if (acc_mode & MAY_WRITE)
1475 return -EISDIR;
1476 break;
1477 case S_IFBLK:
1478 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001479 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001481 /*FALLTHRU*/
1482 case S_IFIFO:
1483 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001485 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001486 }
Dave Hansenb41572e2007-10-16 23:31:14 -07001487
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001488 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07001489 if (error)
1490 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05001491
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 /*
1493 * An append-only file must be opened in append mode for writing.
1494 */
1495 if (IS_APPEND(inode)) {
1496 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05001497 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05001499 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 }
1501
1502 /* O_NOATIME can only be set by the owner or superuser */
Al Viro7715b522009-12-16 03:54:00 -05001503 if (flag & O_NOATIME && !is_owner_or_cap(inode))
1504 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505
1506 /*
1507 * Ensure there are no outstanding leases on the file.
1508 */
Al Virob65a9cf2009-12-16 06:27:40 -05001509 return break_lease(inode, flag);
Al Viro7715b522009-12-16 03:54:00 -05001510}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511
Al Viro7715b522009-12-16 03:54:00 -05001512static int handle_truncate(struct path *path)
1513{
1514 struct inode *inode = path->dentry->d_inode;
1515 int error = get_write_access(inode);
1516 if (error)
1517 return error;
1518 /*
1519 * Refuse to truncate files with mandatory locks held on them.
1520 */
1521 error = locks_verify_locked(inode);
1522 if (!error)
1523 error = security_path_truncate(path, 0,
1524 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN);
1525 if (!error) {
1526 error = do_truncate(path->dentry, 0,
1527 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1528 NULL);
1529 }
1530 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04001531 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532}
1533
Dave Hansend57999e2008-02-15 14:37:27 -08001534/*
1535 * Be careful about ever adding any more callers of this
1536 * function. Its flags must be in the namei format, not
1537 * what get passed to sys_open().
1538 */
1539static int __open_namei_create(struct nameidata *nd, struct path *path,
Dave Hansenaab520e2006-09-30 23:29:02 -07001540 int flag, int mode)
1541{
1542 int error;
Jan Blunck4ac91372008-02-14 19:34:32 -08001543 struct dentry *dir = nd->path.dentry;
Dave Hansenaab520e2006-09-30 23:29:02 -07001544
1545 if (!IS_POSIXACL(dir->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04001546 mode &= ~current_umask();
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001547 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1548 if (error)
1549 goto out_unlock;
Dave Hansenaab520e2006-09-30 23:29:02 -07001550 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001551out_unlock:
Dave Hansenaab520e2006-09-30 23:29:02 -07001552 mutex_unlock(&dir->d_inode->i_mutex);
Jan Blunck4ac91372008-02-14 19:34:32 -08001553 dput(nd->path.dentry);
1554 nd->path.dentry = path->dentry;
Dave Hansenaab520e2006-09-30 23:29:02 -07001555 if (error)
1556 return error;
1557 /* Don't check for write permission, don't truncate */
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001558 return may_open(&nd->path, 0, flag & ~O_TRUNC);
Dave Hansenaab520e2006-09-30 23:29:02 -07001559}
1560
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561/*
Dave Hansend57999e2008-02-15 14:37:27 -08001562 * Note that while the flag value (low two bits) for sys_open means:
1563 * 00 - read-only
1564 * 01 - write-only
1565 * 10 - read-write
1566 * 11 - special
1567 * it is changed into
1568 * 00 - no permissions needed
1569 * 01 - read-permission
1570 * 10 - write-permission
1571 * 11 - read-write
1572 * for the internal routines (ie open_namei()/follow_link() etc)
1573 * This is more logical, and also allows the 00 "no perm needed"
1574 * to be used for symlinks (where the permissions are checked
1575 * later).
1576 *
1577*/
1578static inline int open_to_namei_flags(int flag)
1579{
1580 if ((flag+1) & O_ACCMODE)
1581 flag++;
1582 return flag;
1583}
1584
Al Viro7715b522009-12-16 03:54:00 -05001585static int open_will_truncate(int flag, struct inode *inode)
Dave Hansen4a3fd212008-02-15 14:37:48 -08001586{
1587 /*
1588 * We'll never write to the fs underlying
1589 * a device file.
1590 */
1591 if (special_file(inode->i_mode))
1592 return 0;
1593 return (flag & O_TRUNC);
1594}
1595
Dave Hansend57999e2008-02-15 14:37:27 -08001596/*
Dave Hansen4a3fd212008-02-15 14:37:48 -08001597 * Note that the low bits of the passed in "open_flag"
1598 * are not the same as in the local variable "flag". See
1599 * open_to_namei_flags() for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001601struct file *do_filp_open(int dfd, const char *pathname,
Al Viro6e8341a2009-04-06 11:16:22 -04001602 int open_flag, int mode, int acc_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603{
Dave Hansen4a3fd212008-02-15 14:37:48 -08001604 struct file *filp;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001605 struct nameidata nd;
Al Viro6e8341a2009-04-06 11:16:22 -04001606 int error;
Al Viro9850c052010-01-13 15:01:15 -05001607 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 struct dentry *dir;
1609 int count = 0;
Al Viro7715b522009-12-16 03:54:00 -05001610 int will_truncate;
Dave Hansend57999e2008-02-15 14:37:27 -08001611 int flag = open_to_namei_flags(open_flag);
Al Viro9850c052010-01-13 15:01:15 -05001612 int force_reval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613
Christoph Hellwig6b2f3d12009-10-27 11:05:28 +01001614 /*
1615 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1616 * check for O_DSYNC if the need any syncing at all we enforce it's
1617 * always set instead of having to deal with possibly weird behaviour
1618 * for malicious applications setting only __O_SYNC.
1619 */
1620 if (open_flag & __O_SYNC)
1621 open_flag |= O_DSYNC;
1622
Al Viro6e8341a2009-04-06 11:16:22 -04001623 if (!acc_mode)
Al Viro6d125522009-12-24 06:58:56 -05001624 acc_mode = MAY_OPEN | ACC_MODE(open_flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625
Trond Myklebust834f2a42005-10-18 14:20:16 -07001626 /* O_TRUNC implies we need access checks for write permissions */
1627 if (flag & O_TRUNC)
1628 acc_mode |= MAY_WRITE;
1629
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 /* Allow the LSM permission hook to distinguish append
1631 access from general write access. */
1632 if (flag & O_APPEND)
1633 acc_mode |= MAY_APPEND;
1634
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 /*
1636 * The simplest case - just a plain lookup.
1637 */
1638 if (!(flag & O_CREAT)) {
Al Viro2dd6d1f2009-08-13 20:40:45 +04001639 filp = get_empty_filp();
1640
1641 if (filp == NULL)
1642 return ERR_PTR(-ENFILE);
1643 nd.intent.open.file = filp;
Al Viro482928d2009-12-19 10:10:39 -05001644 filp->f_flags = open_flag;
Al Viro2dd6d1f2009-08-13 20:40:45 +04001645 nd.intent.open.flags = flag;
1646 nd.intent.open.create_mode = 0;
1647 error = do_path_lookup(dfd, pathname,
1648 lookup_flags(flag)|LOOKUP_OPEN, &nd);
1649 if (IS_ERR(nd.intent.open.file)) {
1650 if (error == 0) {
1651 error = PTR_ERR(nd.intent.open.file);
1652 path_put(&nd.path);
1653 }
1654 } else if (error)
1655 release_open_intent(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 if (error)
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001657 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 goto ok;
1659 }
1660
1661 /*
1662 * Create - we need to know the parent.
1663 */
Al Viro9850c052010-01-13 15:01:15 -05001664reval:
Al Viro9b4a9b12009-04-07 11:44:16 -04001665 error = path_init(dfd, pathname, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 if (error)
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001667 return ERR_PTR(error);
Al Viro9850c052010-01-13 15:01:15 -05001668 if (force_reval)
1669 nd.flags |= LOOKUP_REVAL;
Al Viro9b4a9b12009-04-07 11:44:16 -04001670 error = path_walk(pathname, &nd);
J. R. Okajima654f5622009-06-18 23:30:15 +09001671 if (error) {
1672 if (nd.root.mnt)
1673 path_put(&nd.root);
Al Viro9b4a9b12009-04-07 11:44:16 -04001674 return ERR_PTR(error);
J. R. Okajima654f5622009-06-18 23:30:15 +09001675 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001676 if (unlikely(!audit_dummy_context()))
1677 audit_inode(pathname, nd.path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678
1679 /*
1680 * We have the parent and last component. First of all, check
1681 * that we are not asked to creat(2) an obvious directory - that
1682 * will not do.
1683 */
1684 error = -EISDIR;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001685 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
Al Viro8737f3a2008-08-02 22:36:57 -04001686 goto exit_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687
Al Viro8737f3a2008-08-02 22:36:57 -04001688 error = -ENFILE;
1689 filp = get_empty_filp();
1690 if (filp == NULL)
1691 goto exit_parent;
1692 nd.intent.open.file = filp;
Al Viro482928d2009-12-19 10:10:39 -05001693 filp->f_flags = open_flag;
Al Viro8737f3a2008-08-02 22:36:57 -04001694 nd.intent.open.flags = flag;
1695 nd.intent.open.create_mode = mode;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001696 dir = nd.path.dentry;
1697 nd.flags &= ~LOOKUP_PARENT;
Al Viro8737f3a2008-08-02 22:36:57 -04001698 nd.flags |= LOOKUP_CREATE | LOOKUP_OPEN;
Al Viro35165862008-08-05 03:00:49 -04001699 if (flag & O_EXCL)
1700 nd.flags |= LOOKUP_EXCL;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001701 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001702 path.dentry = lookup_hash(&nd);
1703 path.mnt = nd.path.mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704
1705do_last:
Al Viro4e7506e2005-06-06 13:36:00 -07001706 error = PTR_ERR(path.dentry);
1707 if (IS_ERR(path.dentry)) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001708 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 goto exit;
1710 }
1711
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001712 if (IS_ERR(nd.intent.open.file)) {
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001713 error = PTR_ERR(nd.intent.open.file);
Dave Hansen4a3fd212008-02-15 14:37:48 -08001714 goto exit_mutex_unlock;
Oleg Drokin4af4c522006-03-25 03:06:54 -08001715 }
1716
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 /* Negative dentry, just create the file */
Al Viro4e7506e2005-06-06 13:36:00 -07001718 if (!path.dentry->d_inode) {
Dave Hansen4a3fd212008-02-15 14:37:48 -08001719 /*
1720 * This write is needed to ensure that a
1721 * ro->rw transition does not occur between
1722 * the time when the file is created and when
1723 * a permanent write count is taken through
1724 * the 'struct file' in nameidata_to_filp().
1725 */
1726 error = mnt_want_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 if (error)
Dave Hansen4a3fd212008-02-15 14:37:48 -08001728 goto exit_mutex_unlock;
1729 error = __open_namei_create(&nd, &path, flag, mode);
1730 if (error) {
1731 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732 goto exit;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001733 }
Al Viro482928d2009-12-19 10:10:39 -05001734 filp = nameidata_to_filp(&nd);
Dave Hansen4a3fd212008-02-15 14:37:48 -08001735 mnt_drop_write(nd.path.mnt);
J. R. Okajima654f5622009-06-18 23:30:15 +09001736 if (nd.root.mnt)
1737 path_put(&nd.root);
Al Virob65a9cf2009-12-16 06:27:40 -05001738 if (!IS_ERR(filp)) {
1739 error = ima_path_check(&filp->f_path, filp->f_mode &
Al Viro1429b3e2009-12-16 06:38:01 -05001740 (MAY_READ | MAY_WRITE | MAY_EXEC));
Al Virob65a9cf2009-12-16 06:27:40 -05001741 if (error) {
1742 fput(filp);
1743 filp = ERR_PTR(error);
1744 }
1745 }
Dave Hansen4a3fd212008-02-15 14:37:48 -08001746 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 }
1748
1749 /*
1750 * It already exists.
1751 */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001752 mutex_unlock(&dir->d_inode->i_mutex);
Al Viro5a190ae2007-06-07 12:19:32 -04001753 audit_inode(pathname, path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754
1755 error = -EEXIST;
1756 if (flag & O_EXCL)
1757 goto exit_dput;
1758
Al Viroe13b2102005-06-06 13:36:06 -07001759 if (__follow_mount(&path)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 error = -ELOOP;
Al Viroba7a4c12005-06-06 13:36:08 -07001761 if (flag & O_NOFOLLOW)
1762 goto exit_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 }
Amy Griffis3e2efce2006-07-13 13:16:02 -04001764
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 error = -ENOENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001766 if (!path.dentry->d_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767 goto exit_dput;
Al Viroacfa4382008-12-04 10:06:33 -05001768 if (path.dentry->d_inode->i_op->follow_link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769 goto do_link;
1770
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001771 path_to_nameidata(&path, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 error = -EISDIR;
Jeff Laytoncb598612009-11-16 12:05:20 -08001773 if (S_ISDIR(path.dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 goto exit;
1775ok:
Dave Hansen4a3fd212008-02-15 14:37:48 -08001776 /*
1777 * Consider:
1778 * 1. may_open() truncates a file
1779 * 2. a rw->ro mount transition occurs
1780 * 3. nameidata_to_filp() fails due to
1781 * the ro mount.
1782 * That would be inconsistent, and should
1783 * be avoided. Taking this mnt write here
1784 * ensures that (2) can not occur.
1785 */
Al Viro7715b522009-12-16 03:54:00 -05001786 will_truncate = open_will_truncate(flag, nd.path.dentry->d_inode);
1787 if (will_truncate) {
Dave Hansen4a3fd212008-02-15 14:37:48 -08001788 error = mnt_want_write(nd.path.mnt);
1789 if (error)
1790 goto exit;
1791 }
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001792 error = may_open(&nd.path, acc_mode, flag);
Dave Hansen4a3fd212008-02-15 14:37:48 -08001793 if (error) {
Al Viro7715b522009-12-16 03:54:00 -05001794 if (will_truncate)
Dave Hansen4a3fd212008-02-15 14:37:48 -08001795 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796 goto exit;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001797 }
Al Viro482928d2009-12-19 10:10:39 -05001798 filp = nameidata_to_filp(&nd);
Al Virob65a9cf2009-12-16 06:27:40 -05001799 if (!IS_ERR(filp)) {
1800 error = ima_path_check(&filp->f_path, filp->f_mode &
Al Viro1429b3e2009-12-16 06:38:01 -05001801 (MAY_READ | MAY_WRITE | MAY_EXEC));
Al Viro7715b522009-12-16 03:54:00 -05001802 if (error) {
Al Viro7715b522009-12-16 03:54:00 -05001803 fput(filp);
Al Virob65a9cf2009-12-16 06:27:40 -05001804 filp = ERR_PTR(error);
1805 }
1806 }
1807 if (!IS_ERR(filp)) {
1808 if (acc_mode & MAY_WRITE)
1809 vfs_dq_init(nd.path.dentry->d_inode);
1810
1811 if (will_truncate) {
1812 error = handle_truncate(&nd.path);
1813 if (error) {
1814 fput(filp);
1815 filp = ERR_PTR(error);
1816 }
Al Viro7715b522009-12-16 03:54:00 -05001817 }
1818 }
Dave Hansen4a3fd212008-02-15 14:37:48 -08001819 /*
1820 * It is now safe to drop the mnt write
1821 * because the filp has had a write taken
1822 * on its behalf.
1823 */
Al Viro7715b522009-12-16 03:54:00 -05001824 if (will_truncate)
Dave Hansen4a3fd212008-02-15 14:37:48 -08001825 mnt_drop_write(nd.path.mnt);
J. R. Okajima654f5622009-06-18 23:30:15 +09001826 if (nd.root.mnt)
1827 path_put(&nd.root);
Dave Hansen4a3fd212008-02-15 14:37:48 -08001828 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829
Dave Hansen4a3fd212008-02-15 14:37:48 -08001830exit_mutex_unlock:
1831 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832exit_dput:
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001833 path_put_conditional(&path, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834exit:
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001835 if (!IS_ERR(nd.intent.open.file))
1836 release_open_intent(&nd);
Al Viro8737f3a2008-08-02 22:36:57 -04001837exit_parent:
Al Viro2a737872009-04-07 11:49:53 -04001838 if (nd.root.mnt)
1839 path_put(&nd.root);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001840 path_put(&nd.path);
1841 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842
1843do_link:
1844 error = -ELOOP;
1845 if (flag & O_NOFOLLOW)
1846 goto exit_dput;
1847 /*
1848 * This is subtle. Instead of calling do_follow_link() we do the
1849 * thing by hands. The reason is that this way we have zero link_count
1850 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1851 * After that we have the parent and last component, i.e.
1852 * we are in the same situation as after the first path_walk().
1853 * Well, almost - if the last component is normal we get its copy
1854 * stored in nd->last.name and we will have to putname() it when we
1855 * are done. Procfs-like symlinks just set LAST_BIND.
1856 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001857 nd.flags |= LOOKUP_PARENT;
1858 error = security_inode_follow_link(path.dentry, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859 if (error)
1860 goto exit_dput;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001861 error = __do_follow_link(&path, &nd);
Al Viro258fa992009-08-09 01:32:02 +04001862 path_put(&path);
Kirill Korotaevde459212006-07-14 00:23:49 -07001863 if (error) {
1864 /* Does someone understand code flow here? Or it is only
1865 * me so stupid? Anathema to whoever designed this non-sense
1866 * with "intent.open".
1867 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001868 release_open_intent(&nd);
J. R. Okajima654f5622009-06-18 23:30:15 +09001869 if (nd.root.mnt)
1870 path_put(&nd.root);
Al Viro9850c052010-01-13 15:01:15 -05001871 if (error == -ESTALE && !force_reval) {
1872 force_reval = 1;
1873 goto reval;
1874 }
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001875 return ERR_PTR(error);
Kirill Korotaevde459212006-07-14 00:23:49 -07001876 }
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001877 nd.flags &= ~LOOKUP_PARENT;
1878 if (nd.last_type == LAST_BIND)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 goto ok;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880 error = -EISDIR;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001881 if (nd.last_type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 goto exit;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001883 if (nd.last.name[nd.last.len]) {
1884 __putname(nd.last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885 goto exit;
1886 }
1887 error = -ELOOP;
1888 if (count++==32) {
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001889 __putname(nd.last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 goto exit;
1891 }
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001892 dir = nd.path.dentry;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001893 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001894 path.dentry = lookup_hash(&nd);
1895 path.mnt = nd.path.mnt;
1896 __putname(nd.last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897 goto do_last;
1898}
1899
1900/**
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001901 * filp_open - open file and return file pointer
1902 *
1903 * @filename: path to open
1904 * @flags: open flags as per the open(2) second argument
1905 * @mode: mode for the new file if O_CREAT is set, else ignored
1906 *
1907 * This is the helper to open a file from kernelspace if you really
1908 * have to. But in generally you should not do this, so please move
1909 * along, nothing to see here..
1910 */
1911struct file *filp_open(const char *filename, int flags, int mode)
1912{
Al Viro6e8341a2009-04-06 11:16:22 -04001913 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001914}
1915EXPORT_SYMBOL(filp_open);
1916
1917/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918 * lookup_create - lookup a dentry, creating it if it doesn't exist
1919 * @nd: nameidata info
1920 * @is_dir: directory flag
1921 *
1922 * Simple function to lookup and return a dentry and create it
1923 * if it doesn't exist. Is SMP-safe.
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001924 *
Jan Blunck4ac91372008-02-14 19:34:32 -08001925 * Returns with nd->path.dentry->d_inode->i_mutex locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 */
1927struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1928{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001929 struct dentry *dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930
Jan Blunck4ac91372008-02-14 19:34:32 -08001931 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001932 /*
1933 * Yucky last component or no last component at all?
1934 * (foo/., foo/.., /////)
1935 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936 if (nd->last_type != LAST_NORM)
1937 goto fail;
1938 nd->flags &= ~LOOKUP_PARENT;
Al Viro35165862008-08-05 03:00:49 -04001939 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
ASANO Masahiroa6349042006-08-22 20:06:02 -04001940 nd->intent.open.flags = O_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001941
1942 /*
1943 * Do the final lookup.
1944 */
Christoph Hellwig49705b72005-11-08 21:35:06 -08001945 dentry = lookup_hash(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 if (IS_ERR(dentry))
1947 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001948
Al Viroe9baf6e2008-05-15 04:49:12 -04001949 if (dentry->d_inode)
1950 goto eexist;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001951 /*
1952 * Special case - lookup gave negative, but... we had foo/bar/
1953 * From the vfs_mknod() POV we just have a negative dentry -
1954 * all is fine. Let's be bastards - you had / on the end, you've
1955 * been asking for (non-existent) directory. -ENOENT for you.
1956 */
Al Viroe9baf6e2008-05-15 04:49:12 -04001957 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1958 dput(dentry);
1959 dentry = ERR_PTR(-ENOENT);
1960 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 return dentry;
Al Viroe9baf6e2008-05-15 04:49:12 -04001962eexist:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963 dput(dentry);
Al Viroe9baf6e2008-05-15 04:49:12 -04001964 dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965fail:
1966 return dentry;
1967}
Christoph Hellwigf81a0bf2005-05-19 12:26:43 -07001968EXPORT_SYMBOL_GPL(lookup_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969
1970int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1971{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001972 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973
1974 if (error)
1975 return error;
1976
1977 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1978 return -EPERM;
1979
Al Viroacfa4382008-12-04 10:06:33 -05001980 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 return -EPERM;
1982
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07001983 error = devcgroup_inode_mknod(mode, dev);
1984 if (error)
1985 return error;
1986
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 error = security_inode_mknod(dir, dentry, mode, dev);
1988 if (error)
1989 return error;
1990
Jan Kara9e3509e2009-01-26 16:45:12 +01001991 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001993 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001994 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 return error;
1996}
1997
Dave Hansen463c3192008-02-15 14:37:57 -08001998static int may_mknod(mode_t mode)
1999{
2000 switch (mode & S_IFMT) {
2001 case S_IFREG:
2002 case S_IFCHR:
2003 case S_IFBLK:
2004 case S_IFIFO:
2005 case S_IFSOCK:
2006 case 0: /* zero mode translates to S_IFREG */
2007 return 0;
2008 case S_IFDIR:
2009 return -EPERM;
2010 default:
2011 return -EINVAL;
2012 }
2013}
2014
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002015SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2016 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017{
Al Viro2ad94ae2008-07-21 09:32:51 -04002018 int error;
2019 char *tmp;
2020 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 struct nameidata nd;
2022
2023 if (S_ISDIR(mode))
2024 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025
Al Viro2ad94ae2008-07-21 09:32:51 -04002026 error = user_path_parent(dfd, filename, &nd, &tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002028 return error;
2029
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030 dentry = lookup_create(&nd, 0);
Dave Hansen463c3192008-02-15 14:37:57 -08002031 if (IS_ERR(dentry)) {
2032 error = PTR_ERR(dentry);
2033 goto out_unlock;
2034 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002035 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002036 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002037 error = may_mknod(mode);
2038 if (error)
2039 goto out_dput;
2040 error = mnt_want_write(nd.path.mnt);
2041 if (error)
2042 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002043 error = security_path_mknod(&nd.path, dentry, mode, dev);
2044 if (error)
2045 goto out_drop_write;
Dave Hansen463c3192008-02-15 14:37:57 -08002046 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 case 0: case S_IFREG:
Jan Blunck4ac91372008-02-14 19:34:32 -08002048 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 break;
2050 case S_IFCHR: case S_IFBLK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002051 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 new_decode_dev(dev));
2053 break;
2054 case S_IFIFO: case S_IFSOCK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002055 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002058out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002059 mnt_drop_write(nd.path.mnt);
2060out_dput:
2061 dput(dentry);
2062out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002063 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002064 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 putname(tmp);
2066
2067 return error;
2068}
2069
Heiko Carstens3480b252009-01-14 14:14:16 +01002070SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002071{
2072 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2073}
2074
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2076{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002077 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078
2079 if (error)
2080 return error;
2081
Al Viroacfa4382008-12-04 10:06:33 -05002082 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083 return -EPERM;
2084
2085 mode &= (S_IRWXUGO|S_ISVTX);
2086 error = security_inode_mkdir(dir, dentry, mode);
2087 if (error)
2088 return error;
2089
Jan Kara9e3509e2009-01-26 16:45:12 +01002090 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002092 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002093 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094 return error;
2095}
2096
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002097SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098{
2099 int error = 0;
2100 char * tmp;
Dave Hansen6902d922006-09-30 23:29:01 -07002101 struct dentry *dentry;
2102 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103
Al Viro2ad94ae2008-07-21 09:32:51 -04002104 error = user_path_parent(dfd, pathname, &nd, &tmp);
2105 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002106 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107
Dave Hansen6902d922006-09-30 23:29:01 -07002108 dentry = lookup_create(&nd, 1);
2109 error = PTR_ERR(dentry);
2110 if (IS_ERR(dentry))
2111 goto out_unlock;
2112
Jan Blunck4ac91372008-02-14 19:34:32 -08002113 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002114 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002115 error = mnt_want_write(nd.path.mnt);
2116 if (error)
2117 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002118 error = security_path_mkdir(&nd.path, dentry, mode);
2119 if (error)
2120 goto out_drop_write;
Jan Blunck4ac91372008-02-14 19:34:32 -08002121 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002122out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002123 mnt_drop_write(nd.path.mnt);
2124out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002125 dput(dentry);
2126out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002127 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002128 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002129 putname(tmp);
2130out_err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131 return error;
2132}
2133
Heiko Carstens3cdad422009-01-14 14:14:22 +01002134SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002135{
2136 return sys_mkdirat(AT_FDCWD, pathname, mode);
2137}
2138
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139/*
2140 * We try to drop the dentry early: we should have
2141 * a usage count of 2 if we're the only user of this
2142 * dentry, and if that is true (possibly after pruning
2143 * the dcache), then we drop the dentry now.
2144 *
2145 * A low-level filesystem can, if it choses, legally
2146 * do a
2147 *
2148 * if (!d_unhashed(dentry))
2149 * return -EBUSY;
2150 *
2151 * if it cannot handle the case of removing a directory
2152 * that is still in use by something else..
2153 */
2154void dentry_unhash(struct dentry *dentry)
2155{
2156 dget(dentry);
Vasily Averindc168422006-12-06 20:37:07 -08002157 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158 spin_lock(&dcache_lock);
2159 spin_lock(&dentry->d_lock);
2160 if (atomic_read(&dentry->d_count) == 2)
2161 __d_drop(dentry);
2162 spin_unlock(&dentry->d_lock);
2163 spin_unlock(&dcache_lock);
2164}
2165
2166int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2167{
2168 int error = may_delete(dir, dentry, 1);
2169
2170 if (error)
2171 return error;
2172
Al Viroacfa4382008-12-04 10:06:33 -05002173 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 return -EPERM;
2175
Jan Kara9e3509e2009-01-26 16:45:12 +01002176 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002178 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179 dentry_unhash(dentry);
2180 if (d_mountpoint(dentry))
2181 error = -EBUSY;
2182 else {
2183 error = security_inode_rmdir(dir, dentry);
2184 if (!error) {
2185 error = dir->i_op->rmdir(dir, dentry);
2186 if (!error)
2187 dentry->d_inode->i_flags |= S_DEAD;
2188 }
2189 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002190 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 d_delete(dentry);
2193 }
2194 dput(dentry);
2195
2196 return error;
2197}
2198
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002199static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200{
2201 int error = 0;
2202 char * name;
2203 struct dentry *dentry;
2204 struct nameidata nd;
2205
Al Viro2ad94ae2008-07-21 09:32:51 -04002206 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002208 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209
2210 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002211 case LAST_DOTDOT:
2212 error = -ENOTEMPTY;
2213 goto exit1;
2214 case LAST_DOT:
2215 error = -EINVAL;
2216 goto exit1;
2217 case LAST_ROOT:
2218 error = -EBUSY;
2219 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002221
2222 nd.flags &= ~LOOKUP_PARENT;
2223
Jan Blunck4ac91372008-02-14 19:34:32 -08002224 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002225 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002227 if (IS_ERR(dentry))
2228 goto exit2;
Dave Hansen06227532008-02-15 14:37:34 -08002229 error = mnt_want_write(nd.path.mnt);
2230 if (error)
2231 goto exit3;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002232 error = security_path_rmdir(&nd.path, dentry);
2233 if (error)
2234 goto exit4;
Jan Blunck4ac91372008-02-14 19:34:32 -08002235 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002236exit4:
Dave Hansen06227532008-02-15 14:37:34 -08002237 mnt_drop_write(nd.path.mnt);
2238exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07002239 dput(dentry);
2240exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08002241 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002243 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244 putname(name);
2245 return error;
2246}
2247
Heiko Carstens3cdad422009-01-14 14:14:22 +01002248SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002249{
2250 return do_rmdir(AT_FDCWD, pathname);
2251}
2252
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253int vfs_unlink(struct inode *dir, struct dentry *dentry)
2254{
2255 int error = may_delete(dir, dentry, 0);
2256
2257 if (error)
2258 return error;
2259
Al Viroacfa4382008-12-04 10:06:33 -05002260 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261 return -EPERM;
2262
Jan Kara9e3509e2009-01-26 16:45:12 +01002263 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002265 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266 if (d_mountpoint(dentry))
2267 error = -EBUSY;
2268 else {
2269 error = security_inode_unlink(dir, dentry);
2270 if (!error)
2271 error = dir->i_op->unlink(dir, dentry);
2272 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002273 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274
2275 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2276 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08002277 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279 }
Robert Love0eeca282005-07-12 17:06:03 -04002280
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 return error;
2282}
2283
2284/*
2285 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002286 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 * writeout happening, and we don't want to prevent access to the directory
2288 * while waiting on the I/O.
2289 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002290static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291{
Al Viro2ad94ae2008-07-21 09:32:51 -04002292 int error;
2293 char *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 struct dentry *dentry;
2295 struct nameidata nd;
2296 struct inode *inode = NULL;
2297
Al Viro2ad94ae2008-07-21 09:32:51 -04002298 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002300 return error;
2301
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302 error = -EISDIR;
2303 if (nd.last_type != LAST_NORM)
2304 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002305
2306 nd.flags &= ~LOOKUP_PARENT;
2307
Jan Blunck4ac91372008-02-14 19:34:32 -08002308 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002309 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310 error = PTR_ERR(dentry);
2311 if (!IS_ERR(dentry)) {
2312 /* Why not before? Because we want correct error value */
2313 if (nd.last.name[nd.last.len])
2314 goto slashes;
2315 inode = dentry->d_inode;
2316 if (inode)
2317 atomic_inc(&inode->i_count);
Dave Hansen06227532008-02-15 14:37:34 -08002318 error = mnt_want_write(nd.path.mnt);
2319 if (error)
2320 goto exit2;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002321 error = security_path_unlink(&nd.path, dentry);
2322 if (error)
2323 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08002324 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002325exit3:
Dave Hansen06227532008-02-15 14:37:34 -08002326 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327 exit2:
2328 dput(dentry);
2329 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002330 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331 if (inode)
2332 iput(inode); /* truncate the inode here */
2333exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002334 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335 putname(name);
2336 return error;
2337
2338slashes:
2339 error = !dentry->d_inode ? -ENOENT :
2340 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2341 goto exit2;
2342}
2343
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002344SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002345{
2346 if ((flag & ~AT_REMOVEDIR) != 0)
2347 return -EINVAL;
2348
2349 if (flag & AT_REMOVEDIR)
2350 return do_rmdir(dfd, pathname);
2351
2352 return do_unlinkat(dfd, pathname);
2353}
2354
Heiko Carstens3480b252009-01-14 14:14:16 +01002355SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002356{
2357 return do_unlinkat(AT_FDCWD, pathname);
2358}
2359
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002360int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002362 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363
2364 if (error)
2365 return error;
2366
Al Viroacfa4382008-12-04 10:06:33 -05002367 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368 return -EPERM;
2369
2370 error = security_inode_symlink(dir, dentry, oldname);
2371 if (error)
2372 return error;
2373
Jan Kara9e3509e2009-01-26 16:45:12 +01002374 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002376 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002377 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378 return error;
2379}
2380
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002381SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2382 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383{
Al Viro2ad94ae2008-07-21 09:32:51 -04002384 int error;
2385 char *from;
2386 char *to;
Dave Hansen6902d922006-09-30 23:29:01 -07002387 struct dentry *dentry;
2388 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389
2390 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04002391 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392 return PTR_ERR(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002393
2394 error = user_path_parent(newdfd, newname, &nd, &to);
2395 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002396 goto out_putname;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002397
Dave Hansen6902d922006-09-30 23:29:01 -07002398 dentry = lookup_create(&nd, 0);
2399 error = PTR_ERR(dentry);
2400 if (IS_ERR(dentry))
2401 goto out_unlock;
2402
Dave Hansen75c3f292008-02-15 14:37:45 -08002403 error = mnt_want_write(nd.path.mnt);
2404 if (error)
2405 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002406 error = security_path_symlink(&nd.path, dentry, from);
2407 if (error)
2408 goto out_drop_write;
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002409 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002410out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002411 mnt_drop_write(nd.path.mnt);
2412out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002413 dput(dentry);
2414out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002415 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002416 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002417 putname(to);
2418out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419 putname(from);
2420 return error;
2421}
2422
Heiko Carstens3480b252009-01-14 14:14:16 +01002423SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002424{
2425 return sys_symlinkat(oldname, AT_FDCWD, newname);
2426}
2427
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2429{
2430 struct inode *inode = old_dentry->d_inode;
2431 int error;
2432
2433 if (!inode)
2434 return -ENOENT;
2435
Miklos Szeredia95164d2008-07-30 15:08:48 +02002436 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437 if (error)
2438 return error;
2439
2440 if (dir->i_sb != inode->i_sb)
2441 return -EXDEV;
2442
2443 /*
2444 * A link to an append-only or immutable file cannot be created.
2445 */
2446 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2447 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05002448 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002450 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451 return -EPERM;
2452
2453 error = security_inode_link(old_dentry, dir, new_dentry);
2454 if (error)
2455 return error;
2456
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002457 mutex_lock(&inode->i_mutex);
Jan Kara9e3509e2009-01-26 16:45:12 +01002458 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459 error = dir->i_op->link(old_dentry, dir, new_dentry);
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002460 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002461 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002462 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463 return error;
2464}
2465
2466/*
2467 * Hardlinks are often used in delicate situations. We avoid
2468 * security-related surprises by not following symlinks on the
2469 * newname. --KAB
2470 *
2471 * We don't follow them on the oldname either to be compatible
2472 * with linux 2.0, and to avoid hard-linking to directories
2473 * and other special files. --ADM
2474 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002475SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2476 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477{
2478 struct dentry *new_dentry;
Al Viro2d8f3032008-07-22 09:59:21 -04002479 struct nameidata nd;
2480 struct path old_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481 int error;
Al Viro2ad94ae2008-07-21 09:32:51 -04002482 char *to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483
Ulrich Drepper45c9b112006-06-25 05:49:11 -07002484 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002485 return -EINVAL;
2486
Al Viro2d8f3032008-07-22 09:59:21 -04002487 error = user_path_at(olddfd, oldname,
2488 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2489 &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002491 return error;
2492
2493 error = user_path_parent(newdfd, newname, &nd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494 if (error)
2495 goto out;
2496 error = -EXDEV;
Al Viro2d8f3032008-07-22 09:59:21 -04002497 if (old_path.mnt != nd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498 goto out_release;
2499 new_dentry = lookup_create(&nd, 0);
2500 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002501 if (IS_ERR(new_dentry))
2502 goto out_unlock;
Dave Hansen75c3f292008-02-15 14:37:45 -08002503 error = mnt_want_write(nd.path.mnt);
2504 if (error)
2505 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002506 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2507 if (error)
2508 goto out_drop_write;
Al Viro2d8f3032008-07-22 09:59:21 -04002509 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002510out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002511 mnt_drop_write(nd.path.mnt);
2512out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002513 dput(new_dentry);
2514out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002515 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516out_release:
Jan Blunck1d957f92008-02-14 19:34:35 -08002517 path_put(&nd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002518 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519out:
Al Viro2d8f3032008-07-22 09:59:21 -04002520 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521
2522 return error;
2523}
2524
Heiko Carstens3480b252009-01-14 14:14:16 +01002525SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002526{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002527 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002528}
2529
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530/*
2531 * The worst of all namespace operations - renaming directory. "Perverted"
2532 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2533 * Problems:
2534 * a) we can get into loop creation. Check is done in is_subdir().
2535 * b) race potential - two innocent renames can create a loop together.
2536 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002537 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 * story.
2539 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002540 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07002541 * whether the target exists). Solution: try to be smart with locking
2542 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002543 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544 * move will be locked. Thus we can rank directories by the tree
2545 * (ancestors first) and rank all non-directories after them.
2546 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002547 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548 * HOWEVER, it relies on the assumption that any object with ->lookup()
2549 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2550 * we'd better make sure that there's no link(2) for them.
2551 * d) some filesystems don't support opened-but-unlinked directories,
2552 * either because of layout or because they are not ready to deal with
2553 * all cases correctly. The latter will be fixed (taking this sort of
2554 * stuff into VFS), but the former is not going away. Solution: the same
2555 * trick as in rmdir().
2556 * e) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002557 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002559 * ->i_mutex on parents, which works but leads to some truely excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560 * locking].
2561 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07002562static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2563 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564{
2565 int error = 0;
2566 struct inode *target;
2567
2568 /*
2569 * If we are going to change the parent - check write permissions,
2570 * we'll need to flip '..'.
2571 */
2572 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04002573 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574 if (error)
2575 return error;
2576 }
2577
2578 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2579 if (error)
2580 return error;
2581
2582 target = new_dentry->d_inode;
2583 if (target) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002584 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585 dentry_unhash(new_dentry);
2586 }
2587 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2588 error = -EBUSY;
2589 else
2590 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2591 if (target) {
2592 if (!error)
2593 target->i_flags |= S_DEAD;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002594 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595 if (d_unhashed(new_dentry))
2596 d_rehash(new_dentry);
2597 dput(new_dentry);
2598 }
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002599 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07002600 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2601 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602 return error;
2603}
2604
Adrian Bunk75c96f82005-05-05 16:16:09 -07002605static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2606 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607{
2608 struct inode *target;
2609 int error;
2610
2611 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2612 if (error)
2613 return error;
2614
2615 dget(new_dentry);
2616 target = new_dentry->d_inode;
2617 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002618 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2620 error = -EBUSY;
2621 else
2622 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2623 if (!error) {
Mark Fasheh349457c2006-09-08 14:22:21 -07002624 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625 d_move(old_dentry, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626 }
2627 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002628 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629 dput(new_dentry);
2630 return error;
2631}
2632
2633int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2634 struct inode *new_dir, struct dentry *new_dentry)
2635{
2636 int error;
2637 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Robert Love0eeca282005-07-12 17:06:03 -04002638 const char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639
2640 if (old_dentry->d_inode == new_dentry->d_inode)
2641 return 0;
2642
2643 error = may_delete(old_dir, old_dentry, is_dir);
2644 if (error)
2645 return error;
2646
2647 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02002648 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649 else
2650 error = may_delete(new_dir, new_dentry, is_dir);
2651 if (error)
2652 return error;
2653
Al Viroacfa4382008-12-04 10:06:33 -05002654 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655 return -EPERM;
2656
Jan Kara9e3509e2009-01-26 16:45:12 +01002657 vfs_dq_init(old_dir);
2658 vfs_dq_init(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659
Robert Love0eeca282005-07-12 17:06:03 -04002660 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2661
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662 if (is_dir)
2663 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2664 else
2665 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2666 if (!error) {
Robert Love0eeca282005-07-12 17:06:03 -04002667 const char *new_name = old_dentry->d_name.name;
John McCutchan89204c42005-08-15 12:13:28 -04002668 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04002669 new_dentry->d_inode, old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670 }
Robert Love0eeca282005-07-12 17:06:03 -04002671 fsnotify_oldname_free(old_name);
2672
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673 return error;
2674}
2675
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002676SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2677 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678{
Al Viro2ad94ae2008-07-21 09:32:51 -04002679 struct dentry *old_dir, *new_dir;
2680 struct dentry *old_dentry, *new_dentry;
2681 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682 struct nameidata oldnd, newnd;
Al Viro2ad94ae2008-07-21 09:32:51 -04002683 char *from;
2684 char *to;
2685 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686
Al Viro2ad94ae2008-07-21 09:32:51 -04002687 error = user_path_parent(olddfd, oldname, &oldnd, &from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688 if (error)
2689 goto exit;
2690
Al Viro2ad94ae2008-07-21 09:32:51 -04002691 error = user_path_parent(newdfd, newname, &newnd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692 if (error)
2693 goto exit1;
2694
2695 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08002696 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697 goto exit2;
2698
Jan Blunck4ac91372008-02-14 19:34:32 -08002699 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700 error = -EBUSY;
2701 if (oldnd.last_type != LAST_NORM)
2702 goto exit2;
2703
Jan Blunck4ac91372008-02-14 19:34:32 -08002704 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705 if (newnd.last_type != LAST_NORM)
2706 goto exit2;
2707
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002708 oldnd.flags &= ~LOOKUP_PARENT;
2709 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09002710 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002711
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712 trap = lock_rename(new_dir, old_dir);
2713
Christoph Hellwig49705b72005-11-08 21:35:06 -08002714 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715 error = PTR_ERR(old_dentry);
2716 if (IS_ERR(old_dentry))
2717 goto exit3;
2718 /* source must exist */
2719 error = -ENOENT;
2720 if (!old_dentry->d_inode)
2721 goto exit4;
2722 /* unless the source is a directory trailing slashes give -ENOTDIR */
2723 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2724 error = -ENOTDIR;
2725 if (oldnd.last.name[oldnd.last.len])
2726 goto exit4;
2727 if (newnd.last.name[newnd.last.len])
2728 goto exit4;
2729 }
2730 /* source should not be ancestor of target */
2731 error = -EINVAL;
2732 if (old_dentry == trap)
2733 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08002734 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735 error = PTR_ERR(new_dentry);
2736 if (IS_ERR(new_dentry))
2737 goto exit4;
2738 /* target should not be an ancestor of source */
2739 error = -ENOTEMPTY;
2740 if (new_dentry == trap)
2741 goto exit5;
2742
Dave Hansen9079b1e2008-02-15 14:37:49 -08002743 error = mnt_want_write(oldnd.path.mnt);
2744 if (error)
2745 goto exit5;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002746 error = security_path_rename(&oldnd.path, old_dentry,
2747 &newnd.path, new_dentry);
2748 if (error)
2749 goto exit6;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750 error = vfs_rename(old_dir->d_inode, old_dentry,
2751 new_dir->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002752exit6:
Dave Hansen9079b1e2008-02-15 14:37:49 -08002753 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002754exit5:
2755 dput(new_dentry);
2756exit4:
2757 dput(old_dentry);
2758exit3:
2759 unlock_rename(new_dir, old_dir);
2760exit2:
Jan Blunck1d957f92008-02-14 19:34:35 -08002761 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002762 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002763exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002764 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002765 putname(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002766exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002767 return error;
2768}
2769
Heiko Carstensa26eab22009-01-14 14:14:17 +01002770SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002771{
2772 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2773}
2774
Linus Torvalds1da177e2005-04-16 15:20:36 -07002775int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2776{
2777 int len;
2778
2779 len = PTR_ERR(link);
2780 if (IS_ERR(link))
2781 goto out;
2782
2783 len = strlen(link);
2784 if (len > (unsigned) buflen)
2785 len = buflen;
2786 if (copy_to_user(buffer, link, len))
2787 len = -EFAULT;
2788out:
2789 return len;
2790}
2791
2792/*
2793 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2794 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2795 * using) it for any given inode is up to filesystem.
2796 */
2797int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2798{
2799 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002800 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07002801 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002802
Linus Torvalds1da177e2005-04-16 15:20:36 -07002803 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002804 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07002805 if (IS_ERR(cookie))
2806 return PTR_ERR(cookie);
2807
2808 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2809 if (dentry->d_inode->i_op->put_link)
2810 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2811 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812}
2813
2814int vfs_follow_link(struct nameidata *nd, const char *link)
2815{
2816 return __vfs_follow_link(nd, link);
2817}
2818
2819/* get the link contents into pagecache */
2820static char *page_getlink(struct dentry * dentry, struct page **ppage)
2821{
Duane Griffinebd09ab2008-12-19 20:47:12 +00002822 char *kaddr;
2823 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002824 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07002825 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07002827 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002828 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00002829 kaddr = kmap(page);
2830 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2831 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002832}
2833
2834int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2835{
2836 struct page *page = NULL;
2837 char *s = page_getlink(dentry, &page);
2838 int res = vfs_readlink(dentry,buffer,buflen,s);
2839 if (page) {
2840 kunmap(page);
2841 page_cache_release(page);
2842 }
2843 return res;
2844}
2845
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002846void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002847{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002848 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002849 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002850 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851}
2852
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002853void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002854{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002855 struct page *page = cookie;
2856
2857 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002858 kunmap(page);
2859 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002860 }
2861}
2862
Nick Piggin54566b22009-01-04 12:00:53 -08002863/*
2864 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2865 */
2866int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867{
2868 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002869 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002870 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08002871 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08002873 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2874 if (nofs)
2875 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002876
NeilBrown7e53cac2006-03-25 03:07:57 -08002877retry:
Nick Pigginafddba42007-10-16 01:25:01 -07002878 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08002879 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002880 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07002881 goto fail;
2882
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883 kaddr = kmap_atomic(page, KM_USER0);
2884 memcpy(kaddr, symname, len-1);
2885 kunmap_atomic(kaddr, KM_USER0);
Nick Pigginafddba42007-10-16 01:25:01 -07002886
2887 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2888 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002889 if (err < 0)
2890 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07002891 if (err < len-1)
2892 goto retry;
2893
Linus Torvalds1da177e2005-04-16 15:20:36 -07002894 mark_inode_dirty(inode);
2895 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002896fail:
2897 return err;
2898}
2899
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002900int page_symlink(struct inode *inode, const char *symname, int len)
2901{
2902 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08002903 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002904}
2905
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002906const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907 .readlink = generic_readlink,
2908 .follow_link = page_follow_link_light,
2909 .put_link = page_put_link,
2910};
2911
Al Viro2d8f3032008-07-22 09:59:21 -04002912EXPORT_SYMBOL(user_path_at);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002913EXPORT_SYMBOL(follow_down);
2914EXPORT_SYMBOL(follow_up);
2915EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2916EXPORT_SYMBOL(getname);
2917EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002918EXPORT_SYMBOL(lookup_one_len);
2919EXPORT_SYMBOL(page_follow_link_light);
2920EXPORT_SYMBOL(page_put_link);
2921EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002922EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002923EXPORT_SYMBOL(page_symlink);
2924EXPORT_SYMBOL(page_symlink_inode_operations);
2925EXPORT_SYMBOL(path_lookup);
Al Virod1811462008-08-02 00:49:18 -04002926EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002927EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04002928EXPORT_SYMBOL(inode_permission);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -08002929EXPORT_SYMBOL(file_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002930EXPORT_SYMBOL(unlock_rename);
2931EXPORT_SYMBOL(vfs_create);
2932EXPORT_SYMBOL(vfs_follow_link);
2933EXPORT_SYMBOL(vfs_link);
2934EXPORT_SYMBOL(vfs_mkdir);
2935EXPORT_SYMBOL(vfs_mknod);
2936EXPORT_SYMBOL(generic_permission);
2937EXPORT_SYMBOL(vfs_readlink);
2938EXPORT_SYMBOL(vfs_rename);
2939EXPORT_SYMBOL(vfs_rmdir);
2940EXPORT_SYMBOL(vfs_symlink);
2941EXPORT_SYMBOL(vfs_unlink);
2942EXPORT_SYMBOL(dentry_unhash);
2943EXPORT_SYMBOL(generic_readlink);