blob: 5b9016006913bb388eba701dd204d72dce2df124 [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 Blunck5dd784d2008-02-14 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) {
Jan Blunck4ac91372008-02-14 19:34:32 -0800692 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693
Al Viro2a737872009-04-07 11:49:53 -0400694 if (nd->path.dentry == nd->root.dentry &&
695 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 break;
697 }
Jan Blunck4ac91372008-02-14 19:34:32 -0800698 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Al Viro3088dd72010-01-30 15:47:29 -0500699 /* rare case of legitimate dget_parent()... */
700 nd->path.dentry = dget_parent(nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 dput(old);
702 break;
703 }
Al Viro3088dd72010-01-30 15:47:29 -0500704 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 }
Al Viro79ed0222009-04-18 13:59:41 -0400707 follow_mount(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708}
709
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710/*
711 * It's more convoluted than I'd like it to be, but... it's still fairly
712 * small and for now I'd prefer to have fast path as straight as possible.
713 * It _is_ time-critical.
714 */
715static int do_lookup(struct nameidata *nd, struct qstr *name,
716 struct path *path)
717{
Jan Blunck4ac91372008-02-14 19:34:32 -0800718 struct vfsmount *mnt = nd->path.mnt;
Al Viro6e6b1bd2009-08-13 23:38:37 +0400719 struct dentry *dentry, *parent;
720 struct inode *dir;
Al Viro3cac2602009-08-13 18:27:43 +0400721 /*
722 * See if the low-level filesystem might want
723 * to use its own hash..
724 */
725 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
726 int err = nd->path.dentry->d_op->d_hash(nd->path.dentry, name);
727 if (err < 0)
728 return err;
729 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730
Al Viro3cac2602009-08-13 18:27:43 +0400731 dentry = __d_lookup(nd->path.dentry, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 if (!dentry)
733 goto need_lookup;
734 if (dentry->d_op && dentry->d_op->d_revalidate)
735 goto need_revalidate;
736done:
737 path->mnt = mnt;
738 path->dentry = dentry;
Al Viro634ee702005-06-06 13:36:13 -0700739 __follow_mount(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 return 0;
741
742need_lookup:
Al Viro6e6b1bd2009-08-13 23:38:37 +0400743 parent = nd->path.dentry;
744 dir = parent->d_inode;
745
746 mutex_lock(&dir->i_mutex);
747 /*
748 * First re-do the cached lookup just in case it was created
749 * while we waited for the directory semaphore..
750 *
751 * FIXME! This could use version numbering or similar to
752 * avoid unnecessary cache lookups.
753 *
754 * The "dcache_lock" is purely to protect the RCU list walker
755 * from concurrent renames at this point (we mustn't get false
756 * negatives from the RCU list walk here, unlike the optimistic
757 * fast walk).
758 *
759 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
760 */
761 dentry = d_lookup(parent, name);
762 if (!dentry) {
763 struct dentry *new;
764
765 /* Don't create child dentry for a dead directory. */
766 dentry = ERR_PTR(-ENOENT);
767 if (IS_DEADDIR(dir))
768 goto out_unlock;
769
770 new = d_alloc(parent, name);
771 dentry = ERR_PTR(-ENOMEM);
772 if (new) {
773 dentry = dir->i_op->lookup(dir, new, nd);
774 if (dentry)
775 dput(new);
776 else
777 dentry = new;
778 }
779out_unlock:
780 mutex_unlock(&dir->i_mutex);
781 if (IS_ERR(dentry))
782 goto fail;
783 goto done;
784 }
785
786 /*
787 * Uhhuh! Nasty case: the cache was re-populated while
788 * we waited on the semaphore. Need to revalidate.
789 */
790 mutex_unlock(&dir->i_mutex);
791 if (dentry->d_op && dentry->d_op->d_revalidate) {
792 dentry = do_revalidate(dentry, nd);
793 if (!dentry)
794 dentry = ERR_PTR(-ENOENT);
795 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 if (IS_ERR(dentry))
797 goto fail;
798 goto done;
799
800need_revalidate:
Ian Kentbcdc5e02006-09-27 01:50:44 -0700801 dentry = do_revalidate(dentry, nd);
802 if (!dentry)
803 goto need_lookup;
804 if (IS_ERR(dentry))
805 goto fail;
806 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807
808fail:
809 return PTR_ERR(dentry);
810}
811
812/*
Al Viroac278a92010-02-16 18:09:36 +0000813 * This is a temporary kludge to deal with "automount" symlinks; proper
814 * solution is to trigger them on follow_mount(), so that do_lookup()
815 * would DTRT. To be killed before 2.6.34-final.
816 */
817static inline int follow_on_final(struct inode *inode, unsigned lookup_flags)
818{
819 return inode && unlikely(inode->i_op->follow_link) &&
820 ((lookup_flags & LOOKUP_FOLLOW) || S_ISDIR(inode->i_mode));
821}
822
823/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +0100825 * This is the basic name resolution function, turning a pathname into
826 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 *
Prasanna Medaea3834d2005-04-29 16:00:17 +0100828 * Returns 0 and nd will have valid dentry and mnt on success.
829 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 */
Al Viro6de88d72009-08-09 01:41:57 +0400831static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832{
833 struct path next;
834 struct inode *inode;
835 int err;
836 unsigned int lookup_flags = nd->flags;
837
838 while (*name=='/')
839 name++;
840 if (!*name)
841 goto return_reval;
842
Jan Blunck4ac91372008-02-14 19:34:32 -0800843 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 if (nd->depth)
Trond Myklebustf55eab82006-02-04 23:28:01 -0800845 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846
847 /* At this point we know we have a real path component. */
848 for(;;) {
849 unsigned long hash;
850 struct qstr this;
851 unsigned int c;
852
Trond Myklebustcdce5d62005-10-18 14:20:18 -0700853 nd->flags |= LOOKUP_CONTINUE;
Al Virob75b5082009-12-16 01:01:38 -0500854 err = exec_permission(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 if (err)
856 break;
857
858 this.name = name;
859 c = *(const unsigned char *)name;
860
861 hash = init_name_hash();
862 do {
863 name++;
864 hash = partial_name_hash(c, hash);
865 c = *(const unsigned char *)name;
866 } while (c && (c != '/'));
867 this.len = name - (const char *) this.name;
868 this.hash = end_name_hash(hash);
869
870 /* remove trailing slashes? */
871 if (!c)
872 goto last_component;
873 while (*++name == '/');
874 if (!*name)
875 goto last_with_slashes;
876
877 /*
878 * "." and ".." are special - ".." especially so because it has
879 * to be able to know about the current root directory and
880 * parent relationships.
881 */
882 if (this.name[0] == '.') switch (this.len) {
883 default:
884 break;
885 case 2:
886 if (this.name[1] != '.')
887 break;
Al Viro58c465e2005-06-06 13:36:13 -0700888 follow_dotdot(nd);
Jan Blunck4ac91372008-02-14 19:34:32 -0800889 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 /* fallthrough */
891 case 1:
892 continue;
893 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 /* This does the actual lookups.. */
895 err = do_lookup(nd, &this, &next);
896 if (err)
897 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898
899 err = -ENOENT;
900 inode = next.dentry->d_inode;
901 if (!inode)
902 goto out_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903
904 if (inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700905 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 if (err)
907 goto return_err;
908 err = -ENOENT;
Jan Blunck4ac91372008-02-14 19:34:32 -0800909 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 if (!inode)
911 break;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700912 } else
913 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 err = -ENOTDIR;
915 if (!inode->i_op->lookup)
916 break;
917 continue;
918 /* here ends the main loop */
919
920last_with_slashes:
921 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
922last_component:
Trond Myklebustf55eab82006-02-04 23:28:01 -0800923 /* Clear LOOKUP_CONTINUE iff it was previously unset */
924 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 if (lookup_flags & LOOKUP_PARENT)
926 goto lookup_parent;
927 if (this.name[0] == '.') switch (this.len) {
928 default:
929 break;
930 case 2:
931 if (this.name[1] != '.')
932 break;
Al Viro58c465e2005-06-06 13:36:13 -0700933 follow_dotdot(nd);
Jan Blunck4ac91372008-02-14 19:34:32 -0800934 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 /* fallthrough */
936 case 1:
937 goto return_reval;
938 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 err = do_lookup(nd, &this, &next);
940 if (err)
941 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 inode = next.dentry->d_inode;
Al Viroac278a92010-02-16 18:09:36 +0000943 if (follow_on_final(inode, lookup_flags)) {
Al Viro90ebe562005-06-06 13:35:58 -0700944 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 if (err)
946 goto return_err;
Jan Blunck4ac91372008-02-14 19:34:32 -0800947 inode = nd->path.dentry->d_inode;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700948 } else
949 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 err = -ENOENT;
951 if (!inode)
952 break;
953 if (lookup_flags & LOOKUP_DIRECTORY) {
954 err = -ENOTDIR;
Al Viroacfa4382008-12-04 10:06:33 -0500955 if (!inode->i_op->lookup)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 break;
957 }
958 goto return_base;
959lookup_parent:
960 nd->last = this;
961 nd->last_type = LAST_NORM;
962 if (this.name[0] != '.')
963 goto return_base;
964 if (this.len == 1)
965 nd->last_type = LAST_DOT;
966 else if (this.len == 2 && this.name[1] == '.')
967 nd->last_type = LAST_DOTDOT;
968 else
969 goto return_base;
970return_reval:
971 /*
972 * We bypassed the ordinary revalidation routines.
973 * We may need to check the cached dentry for staleness.
974 */
Jan Blunck4ac91372008-02-14 19:34:32 -0800975 if (nd->path.dentry && nd->path.dentry->d_sb &&
976 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 err = -ESTALE;
978 /* Note: we do not d_invalidate() */
Jan Blunck4ac91372008-02-14 19:34:32 -0800979 if (!nd->path.dentry->d_op->d_revalidate(
980 nd->path.dentry, nd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 break;
982 }
983return_base:
984 return 0;
985out_dput:
Jan Blunck1d957f92008-02-14 19:34:35 -0800986 path_put_conditional(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 break;
988 }
Jan Blunck1d957f92008-02-14 19:34:35 -0800989 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990return_err:
991 return err;
992}
993
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800994static int path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995{
Al Viro6de88d72009-08-09 01:41:57 +0400996 struct path save = nd->path;
997 int result;
998
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 current->total_link_count = 0;
Al Viro6de88d72009-08-09 01:41:57 +04001000
1001 /* make sure the stuff we saved doesn't go away */
1002 path_get(&save);
1003
1004 result = link_path_walk(name, nd);
1005 if (result == -ESTALE) {
1006 /* nd->path had been dropped */
1007 current->total_link_count = 0;
1008 nd->path = save;
1009 path_get(&nd->path);
1010 nd->flags |= LOOKUP_REVAL;
1011 result = link_path_walk(name, nd);
1012 }
1013
1014 path_put(&save);
1015
1016 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017}
1018
Al Viro9b4a9b12009-04-07 11:44:16 -04001019static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001021 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001022 int fput_needed;
1023 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024
1025 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1026 nd->flags = flags;
1027 nd->depth = 0;
Al Viro2a737872009-04-07 11:49:53 -04001028 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 if (*name=='/') {
Al Viro2a737872009-04-07 11:49:53 -04001031 set_root(nd);
1032 nd->path = nd->root;
1033 path_get(&nd->root);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001034 } else if (dfd == AT_FDCWD) {
Al Viro2a737872009-04-07 11:49:53 -04001035 struct fs_struct *fs = current->fs;
Andreas Mohre518ddb2006-09-29 02:01:22 -07001036 read_lock(&fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08001037 nd->path = fs->pwd;
1038 path_get(&fs->pwd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001039 read_unlock(&fs->lock);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001040 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001041 struct dentry *dentry;
1042
1043 file = fget_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001044 retval = -EBADF;
1045 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001046 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001047
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001048 dentry = file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001049
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001050 retval = -ENOTDIR;
1051 if (!S_ISDIR(dentry->d_inode->i_mode))
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001052 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001053
1054 retval = file_permission(file, MAY_EXEC);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001055 if (retval)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001056 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001057
Jan Blunck5dd784d2008-02-14 19:34:38 -08001058 nd->path = file->f_path;
1059 path_get(&file->f_path);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001060
1061 fput_light(file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001063 return 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001064
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001065fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001066 fput_light(file, fput_needed);
Al Viro9b4a9b12009-04-07 11:44:16 -04001067out_fail:
1068 return retval;
1069}
1070
1071/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1072static int do_path_lookup(int dfd, const char *name,
1073 unsigned int flags, struct nameidata *nd)
1074{
1075 int retval = path_init(dfd, name, flags, nd);
1076 if (!retval)
1077 retval = path_walk(name, nd);
1078 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1079 nd->path.dentry->d_inode))
1080 audit_inode(name, nd->path.dentry);
Al Viro2a737872009-04-07 11:49:53 -04001081 if (nd->root.mnt) {
1082 path_put(&nd->root);
1083 nd->root.mnt = NULL;
1084 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001085 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086}
1087
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -08001088int path_lookup(const char *name, unsigned int flags,
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001089 struct nameidata *nd)
1090{
1091 return do_path_lookup(AT_FDCWD, name, flags, nd);
1092}
1093
Al Virod1811462008-08-02 00:49:18 -04001094int kern_path(const char *name, unsigned int flags, struct path *path)
1095{
1096 struct nameidata nd;
1097 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1098 if (!res)
1099 *path = nd.path;
1100 return res;
1101}
1102
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001103/**
1104 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1105 * @dentry: pointer to dentry of the base directory
1106 * @mnt: pointer to vfs mount of the base directory
1107 * @name: pointer to file name
1108 * @flags: lookup flags
1109 * @nd: pointer to nameidata
1110 */
1111int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1112 const char *name, unsigned int flags,
1113 struct nameidata *nd)
1114{
1115 int retval;
1116
1117 /* same as do_path_lookup */
1118 nd->last_type = LAST_ROOT;
1119 nd->flags = flags;
1120 nd->depth = 0;
1121
Jan Blunckc8e7f442008-06-09 16:40:35 -07001122 nd->path.dentry = dentry;
1123 nd->path.mnt = mnt;
1124 path_get(&nd->path);
Al Viro5b857112009-04-07 11:53:49 -04001125 nd->root = nd->path;
1126 path_get(&nd->root);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001127
1128 retval = path_walk(name, nd);
Jan Blunck4ac91372008-02-14 19:34:32 -08001129 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1130 nd->path.dentry->d_inode))
1131 audit_inode(name, nd->path.dentry);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001132
Al Viro5b857112009-04-07 11:53:49 -04001133 path_put(&nd->root);
1134 nd->root.mnt = NULL;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001135
Al Viro2a737872009-04-07 11:49:53 -04001136 return retval;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001137}
1138
Christoph Hellwigeead1912007-10-16 23:25:38 -07001139static struct dentry *__lookup_hash(struct qstr *name,
1140 struct dentry *base, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141{
James Morris057f6c02007-04-26 00:12:05 -07001142 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 struct inode *inode;
1144 int err;
1145
1146 inode = base->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147
1148 /*
1149 * See if the low-level filesystem might want
1150 * to use its own hash..
1151 */
1152 if (base->d_op && base->d_op->d_hash) {
1153 err = base->d_op->d_hash(base, name);
1154 dentry = ERR_PTR(err);
1155 if (err < 0)
1156 goto out;
1157 }
1158
Al Viro6e6b1bd2009-08-13 23:38:37 +04001159 dentry = __d_lookup(base, name);
1160
1161 /* lockess __d_lookup may fail due to concurrent d_move()
1162 * in some unrelated directory, so try with d_lookup
1163 */
1164 if (!dentry)
1165 dentry = d_lookup(base, name);
1166
1167 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
1168 dentry = do_revalidate(dentry, nd);
1169
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 if (!dentry) {
Miklos Szeredid70b67c2008-07-02 21:30:15 +02001171 struct dentry *new;
1172
1173 /* Don't create child dentry for a dead directory. */
1174 dentry = ERR_PTR(-ENOENT);
1175 if (IS_DEADDIR(inode))
1176 goto out;
1177
1178 new = d_alloc(base, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 dentry = ERR_PTR(-ENOMEM);
1180 if (!new)
1181 goto out;
1182 dentry = inode->i_op->lookup(inode, new, nd);
1183 if (!dentry)
1184 dentry = new;
1185 else
1186 dput(new);
1187 }
1188out:
1189 return dentry;
1190}
1191
James Morris057f6c02007-04-26 00:12:05 -07001192/*
1193 * Restricted form of lookup. Doesn't follow links, single-component only,
1194 * needs parent already locked. Doesn't follow mounts.
1195 * SMP-safe.
1196 */
Adrian Bunka244e162006-03-31 02:32:11 -08001197static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198{
Christoph Hellwigeead1912007-10-16 23:25:38 -07001199 int err;
1200
Al Virob75b5082009-12-16 01:01:38 -05001201 err = exec_permission(nd->path.dentry->d_inode);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001202 if (err)
1203 return ERR_PTR(err);
Jan Blunck4ac91372008-02-14 19:34:32 -08001204 return __lookup_hash(&nd->last, nd->path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205}
1206
Christoph Hellwigeead1912007-10-16 23:25:38 -07001207static int __lookup_one_len(const char *name, struct qstr *this,
1208 struct dentry *base, int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209{
1210 unsigned long hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 unsigned int c;
1212
James Morris057f6c02007-04-26 00:12:05 -07001213 this->name = name;
1214 this->len = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 if (!len)
James Morris057f6c02007-04-26 00:12:05 -07001216 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217
1218 hash = init_name_hash();
1219 while (len--) {
1220 c = *(const unsigned char *)name++;
1221 if (c == '/' || c == '\0')
James Morris057f6c02007-04-26 00:12:05 -07001222 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 hash = partial_name_hash(c, hash);
1224 }
James Morris057f6c02007-04-26 00:12:05 -07001225 this->hash = end_name_hash(hash);
1226 return 0;
1227}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228
Christoph Hellwigeead1912007-10-16 23:25:38 -07001229/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001230 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07001231 * @name: pathname component to lookup
1232 * @base: base directory to lookup from
1233 * @len: maximum length @len should be interpreted to
1234 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07001235 * Note that this routine is purely a helper for filesystem usage and should
1236 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07001237 * nameidata argument is passed to the filesystem methods and a filesystem
1238 * using this helper needs to be prepared for that.
1239 */
James Morris057f6c02007-04-26 00:12:05 -07001240struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1241{
1242 int err;
1243 struct qstr this;
1244
David Woodhouse2f9092e2009-04-20 23:18:37 +01001245 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1246
James Morris057f6c02007-04-26 00:12:05 -07001247 err = __lookup_one_len(name, &this, base, len);
1248 if (err)
1249 return ERR_PTR(err);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001250
Al Virob75b5082009-12-16 01:01:38 -05001251 err = exec_permission(base->d_inode);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001252 if (err)
1253 return ERR_PTR(err);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001254 return __lookup_hash(&this, base, NULL);
James Morris057f6c02007-04-26 00:12:05 -07001255}
1256
Al Viro2d8f3032008-07-22 09:59:21 -04001257int user_path_at(int dfd, const char __user *name, unsigned flags,
1258 struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259{
Al Viro2d8f3032008-07-22 09:59:21 -04001260 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 char *tmp = getname(name);
1262 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04001264
1265 BUG_ON(flags & LOOKUP_PARENT);
1266
1267 err = do_path_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04001269 if (!err)
1270 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 }
1272 return err;
1273}
1274
Al Viro2ad94ae2008-07-21 09:32:51 -04001275static int user_path_parent(int dfd, const char __user *path,
1276 struct nameidata *nd, char **name)
1277{
1278 char *s = getname(path);
1279 int error;
1280
1281 if (IS_ERR(s))
1282 return PTR_ERR(s);
1283
1284 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1285 if (error)
1286 putname(s);
1287 else
1288 *name = s;
1289
1290 return error;
1291}
1292
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293/*
1294 * It's inline, so penalty for filesystems that don't use sticky bit is
1295 * minimal.
1296 */
1297static inline int check_sticky(struct inode *dir, struct inode *inode)
1298{
David Howellsda9592e2008-11-14 10:39:05 +11001299 uid_t fsuid = current_fsuid();
1300
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 if (!(dir->i_mode & S_ISVTX))
1302 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001303 if (inode->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001305 if (dir->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 return 0;
1307 return !capable(CAP_FOWNER);
1308}
1309
1310/*
1311 * Check whether we can remove a link victim from directory dir, check
1312 * whether the type of victim is right.
1313 * 1. We can't do it if dir is read-only (done in permission())
1314 * 2. We should have write and exec permissions on dir
1315 * 3. We can't remove anything from append-only dir
1316 * 4. We can't do anything with immutable dir (done in permission())
1317 * 5. If the sticky bit on dir is set we should either
1318 * a. be owner of dir, or
1319 * b. be owner of victim, or
1320 * c. have CAP_FOWNER capability
1321 * 6. If the victim is append-only or immutable we can't do antyhing with
1322 * links pointing to it.
1323 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1324 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1325 * 9. We can't remove a root or mountpoint.
1326 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1327 * nfs_async_unlink().
1328 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001329static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330{
1331 int error;
1332
1333 if (!victim->d_inode)
1334 return -ENOENT;
1335
1336 BUG_ON(victim->d_parent->d_inode != dir);
Al Virocccc6bb2009-12-25 05:07:33 -05001337 audit_inode_child(victim, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338
Al Virof419a2e2008-07-22 00:07:17 -04001339 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 if (error)
1341 return error;
1342 if (IS_APPEND(dir))
1343 return -EPERM;
1344 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
Hugh Dickinsf9454542008-11-19 15:36:38 -08001345 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 return -EPERM;
1347 if (isdir) {
1348 if (!S_ISDIR(victim->d_inode->i_mode))
1349 return -ENOTDIR;
1350 if (IS_ROOT(victim))
1351 return -EBUSY;
1352 } else if (S_ISDIR(victim->d_inode->i_mode))
1353 return -EISDIR;
1354 if (IS_DEADDIR(dir))
1355 return -ENOENT;
1356 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1357 return -EBUSY;
1358 return 0;
1359}
1360
1361/* Check whether we can create an object with dentry child in directory
1362 * dir.
1363 * 1. We can't do it if child already exists (open has special treatment for
1364 * this case, but since we are inlined it's OK)
1365 * 2. We can't do it if dir is read-only (done in permission())
1366 * 3. We should have write and exec permissions on dir
1367 * 4. We can't do it if dir is immutable (done in permission())
1368 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02001369static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370{
1371 if (child->d_inode)
1372 return -EEXIST;
1373 if (IS_DEADDIR(dir))
1374 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04001375 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376}
1377
1378/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 * O_DIRECTORY translates into forcing a directory lookup.
1380 */
1381static inline int lookup_flags(unsigned int f)
1382{
1383 unsigned long retval = LOOKUP_FOLLOW;
1384
1385 if (f & O_NOFOLLOW)
1386 retval &= ~LOOKUP_FOLLOW;
1387
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 if (f & O_DIRECTORY)
1389 retval |= LOOKUP_DIRECTORY;
1390
1391 return retval;
1392}
1393
1394/*
1395 * p1 and p2 should be directories on the same fs.
1396 */
1397struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1398{
1399 struct dentry *p;
1400
1401 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001402 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 return NULL;
1404 }
1405
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001406 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001408 p = d_ancestor(p2, p1);
1409 if (p) {
1410 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1411 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1412 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 }
1414
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001415 p = d_ancestor(p1, p2);
1416 if (p) {
1417 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1418 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1419 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 }
1421
Ingo Molnarf2eace22006-07-03 00:25:05 -07001422 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1423 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 return NULL;
1425}
1426
1427void unlock_rename(struct dentry *p1, struct dentry *p2)
1428{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001429 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001431 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001432 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 }
1434}
1435
1436int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1437 struct nameidata *nd)
1438{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001439 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440
1441 if (error)
1442 return error;
1443
Al Viroacfa4382008-12-04 10:06:33 -05001444 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 return -EACCES; /* shouldn't it be ENOSYS? */
1446 mode &= S_IALLUGO;
1447 mode |= S_IFREG;
1448 error = security_inode_create(dir, dentry, mode);
1449 if (error)
1450 return error;
Jan Kara9e3509e2009-01-26 16:45:12 +01001451 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001453 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001454 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 return error;
1456}
1457
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001458int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001460 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 struct inode *inode = dentry->d_inode;
1462 int error;
1463
1464 if (!inode)
1465 return -ENOENT;
1466
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001467 switch (inode->i_mode & S_IFMT) {
1468 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001470 case S_IFDIR:
1471 if (acc_mode & MAY_WRITE)
1472 return -EISDIR;
1473 break;
1474 case S_IFBLK:
1475 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001476 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001478 /*FALLTHRU*/
1479 case S_IFIFO:
1480 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001482 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001483 }
Dave Hansenb41572e2007-10-16 23:31:14 -07001484
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001485 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07001486 if (error)
1487 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05001488
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 /*
1490 * An append-only file must be opened in append mode for writing.
1491 */
1492 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05001493 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05001494 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05001496 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 }
1498
1499 /* O_NOATIME can only be set by the owner or superuser */
Al Viro7715b522009-12-16 03:54:00 -05001500 if (flag & O_NOATIME && !is_owner_or_cap(inode))
1501 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502
1503 /*
1504 * Ensure there are no outstanding leases on the file.
1505 */
Al Virob65a9cf2009-12-16 06:27:40 -05001506 return break_lease(inode, flag);
Al Viro7715b522009-12-16 03:54:00 -05001507}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508
Al Viro7715b522009-12-16 03:54:00 -05001509static int handle_truncate(struct path *path)
1510{
1511 struct inode *inode = path->dentry->d_inode;
1512 int error = get_write_access(inode);
1513 if (error)
1514 return error;
1515 /*
1516 * Refuse to truncate files with mandatory locks held on them.
1517 */
1518 error = locks_verify_locked(inode);
1519 if (!error)
1520 error = security_path_truncate(path, 0,
1521 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN);
1522 if (!error) {
1523 error = do_truncate(path->dentry, 0,
1524 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1525 NULL);
1526 }
1527 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04001528 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529}
1530
Dave Hansend57999e2008-02-15 14:37:27 -08001531/*
1532 * Be careful about ever adding any more callers of this
1533 * function. Its flags must be in the namei format, not
1534 * what get passed to sys_open().
1535 */
1536static int __open_namei_create(struct nameidata *nd, struct path *path,
Al Viro8737c932009-12-24 06:47:55 -05001537 int open_flag, int mode)
Dave Hansenaab520e2006-09-30 23:29:02 -07001538{
1539 int error;
Jan Blunck4ac91372008-02-14 19:34:32 -08001540 struct dentry *dir = nd->path.dentry;
Dave Hansenaab520e2006-09-30 23:29:02 -07001541
1542 if (!IS_POSIXACL(dir->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04001543 mode &= ~current_umask();
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001544 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1545 if (error)
1546 goto out_unlock;
Dave Hansenaab520e2006-09-30 23:29:02 -07001547 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001548out_unlock:
Dave Hansenaab520e2006-09-30 23:29:02 -07001549 mutex_unlock(&dir->d_inode->i_mutex);
Jan Blunck4ac91372008-02-14 19:34:32 -08001550 dput(nd->path.dentry);
1551 nd->path.dentry = path->dentry;
Dave Hansenaab520e2006-09-30 23:29:02 -07001552 if (error)
1553 return error;
1554 /* Don't check for write permission, don't truncate */
Al Viro8737c932009-12-24 06:47:55 -05001555 return may_open(&nd->path, 0, open_flag & ~O_TRUNC);
Dave Hansenaab520e2006-09-30 23:29:02 -07001556}
1557
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558/*
Dave Hansend57999e2008-02-15 14:37:27 -08001559 * Note that while the flag value (low two bits) for sys_open means:
1560 * 00 - read-only
1561 * 01 - write-only
1562 * 10 - read-write
1563 * 11 - special
1564 * it is changed into
1565 * 00 - no permissions needed
1566 * 01 - read-permission
1567 * 10 - write-permission
1568 * 11 - read-write
1569 * for the internal routines (ie open_namei()/follow_link() etc)
1570 * This is more logical, and also allows the 00 "no perm needed"
1571 * to be used for symlinks (where the permissions are checked
1572 * later).
1573 *
1574*/
1575static inline int open_to_namei_flags(int flag)
1576{
1577 if ((flag+1) & O_ACCMODE)
1578 flag++;
1579 return flag;
1580}
1581
Al Viro7715b522009-12-16 03:54:00 -05001582static int open_will_truncate(int flag, struct inode *inode)
Dave Hansen4a3fd212008-02-15 14:37:48 -08001583{
1584 /*
1585 * We'll never write to the fs underlying
1586 * a device file.
1587 */
1588 if (special_file(inode->i_mode))
1589 return 0;
1590 return (flag & O_TRUNC);
1591}
1592
Al Viro648fa862009-12-24 01:26:48 -05001593static struct file *finish_open(struct nameidata *nd,
Al Viro9a661792009-12-24 06:49:47 -05001594 int open_flag, int acc_mode)
Al Viro648fa862009-12-24 01:26:48 -05001595{
1596 struct file *filp;
1597 int will_truncate;
1598 int error;
1599
Al Viro9a661792009-12-24 06:49:47 -05001600 will_truncate = open_will_truncate(open_flag, nd->path.dentry->d_inode);
Al Viro648fa862009-12-24 01:26:48 -05001601 if (will_truncate) {
1602 error = mnt_want_write(nd->path.mnt);
1603 if (error)
1604 goto exit;
1605 }
1606 error = may_open(&nd->path, acc_mode, open_flag);
1607 if (error) {
1608 if (will_truncate)
1609 mnt_drop_write(nd->path.mnt);
1610 goto exit;
1611 }
1612 filp = nameidata_to_filp(nd);
1613 if (!IS_ERR(filp)) {
1614 error = ima_file_check(filp, acc_mode);
1615 if (error) {
1616 fput(filp);
1617 filp = ERR_PTR(error);
1618 }
1619 }
1620 if (!IS_ERR(filp)) {
1621 if (acc_mode & MAY_WRITE)
1622 vfs_dq_init(nd->path.dentry->d_inode);
1623
1624 if (will_truncate) {
1625 error = handle_truncate(&nd->path);
1626 if (error) {
1627 fput(filp);
1628 filp = ERR_PTR(error);
1629 }
1630 }
1631 }
1632 /*
1633 * It is now safe to drop the mnt write
1634 * because the filp has had a write taken
1635 * on its behalf.
1636 */
1637 if (will_truncate)
1638 mnt_drop_write(nd->path.mnt);
1639 return filp;
1640
1641exit:
1642 if (!IS_ERR(nd->intent.open.file))
1643 release_open_intent(nd);
1644 path_put(&nd->path);
1645 return ERR_PTR(error);
1646}
1647
Al Virofb1cc552009-12-24 01:58:28 -05001648static struct file *do_last(struct nameidata *nd, struct path *path,
1649 int open_flag, int flag, int acc_mode,
1650 int mode, const char *pathname,
Al Viroa1e28032009-12-24 02:12:06 -05001651 int *is_link)
Al Virofb1cc552009-12-24 01:58:28 -05001652{
Al Viroa1e28032009-12-24 02:12:06 -05001653 struct dentry *dir = nd->path.dentry;
Al Virofb1cc552009-12-24 01:58:28 -05001654 struct file *filp;
1655 int error;
1656
1657 *is_link = 0;
1658
Al Viroa2c36b42009-12-24 03:39:50 -05001659 error = -EISDIR;
1660 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1661 goto exit;
1662
Al Viroa1e28032009-12-24 02:12:06 -05001663 mutex_lock(&dir->d_inode->i_mutex);
1664
1665 path->dentry = lookup_hash(nd);
1666 path->mnt = nd->path.mnt;
1667
Al Virofb1cc552009-12-24 01:58:28 -05001668 error = PTR_ERR(path->dentry);
1669 if (IS_ERR(path->dentry)) {
1670 mutex_unlock(&dir->d_inode->i_mutex);
1671 goto exit;
1672 }
1673
1674 if (IS_ERR(nd->intent.open.file)) {
1675 error = PTR_ERR(nd->intent.open.file);
1676 goto exit_mutex_unlock;
1677 }
1678
1679 /* Negative dentry, just create the file */
1680 if (!path->dentry->d_inode) {
1681 /*
1682 * This write is needed to ensure that a
1683 * ro->rw transition does not occur between
1684 * the time when the file is created and when
1685 * a permanent write count is taken through
1686 * the 'struct file' in nameidata_to_filp().
1687 */
1688 error = mnt_want_write(nd->path.mnt);
1689 if (error)
1690 goto exit_mutex_unlock;
1691 error = __open_namei_create(nd, path, open_flag, mode);
1692 if (error) {
1693 mnt_drop_write(nd->path.mnt);
1694 goto exit;
1695 }
1696 filp = nameidata_to_filp(nd);
1697 mnt_drop_write(nd->path.mnt);
Al Virofb1cc552009-12-24 01:58:28 -05001698 if (!IS_ERR(filp)) {
1699 error = ima_file_check(filp, acc_mode);
1700 if (error) {
1701 fput(filp);
1702 filp = ERR_PTR(error);
1703 }
1704 }
1705 return filp;
1706 }
1707
1708 /*
1709 * It already exists.
1710 */
1711 mutex_unlock(&dir->d_inode->i_mutex);
1712 audit_inode(pathname, path->dentry);
1713
1714 error = -EEXIST;
1715 if (flag & O_EXCL)
1716 goto exit_dput;
1717
1718 if (__follow_mount(path)) {
1719 error = -ELOOP;
1720 if (flag & O_NOFOLLOW)
1721 goto exit_dput;
1722 }
1723
1724 error = -ENOENT;
1725 if (!path->dentry->d_inode)
1726 goto exit_dput;
1727 if (path->dentry->d_inode->i_op->follow_link) {
1728 *is_link = 1;
1729 return NULL;
1730 }
1731
1732 path_to_nameidata(path, nd);
1733 error = -EISDIR;
1734 if (S_ISDIR(path->dentry->d_inode->i_mode))
1735 goto exit;
Al Viro9a661792009-12-24 06:49:47 -05001736 filp = finish_open(nd, open_flag, acc_mode);
Al Virofb1cc552009-12-24 01:58:28 -05001737 return filp;
1738
1739exit_mutex_unlock:
1740 mutex_unlock(&dir->d_inode->i_mutex);
1741exit_dput:
1742 path_put_conditional(path, nd);
1743exit:
1744 if (!IS_ERR(nd->intent.open.file))
1745 release_open_intent(nd);
Al Virofb1cc552009-12-24 01:58:28 -05001746 path_put(&nd->path);
1747 return ERR_PTR(error);
1748}
1749
Dave Hansend57999e2008-02-15 14:37:27 -08001750/*
Dave Hansen4a3fd212008-02-15 14:37:48 -08001751 * Note that the low bits of the passed in "open_flag"
1752 * are not the same as in the local variable "flag". See
1753 * open_to_namei_flags() for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001755struct file *do_filp_open(int dfd, const char *pathname,
Al Viro6e8341a2009-04-06 11:16:22 -04001756 int open_flag, int mode, int acc_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757{
Dave Hansen4a3fd212008-02-15 14:37:48 -08001758 struct file *filp;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001759 struct nameidata nd;
Al Viro6e8341a2009-04-06 11:16:22 -04001760 int error;
Al Viro9850c052010-01-13 15:01:15 -05001761 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 int count = 0;
Dave Hansend57999e2008-02-15 14:37:27 -08001763 int flag = open_to_namei_flags(open_flag);
Al Viro9850c052010-01-13 15:01:15 -05001764 int force_reval = 0;
Al Virofb1cc552009-12-24 01:58:28 -05001765 int is_link;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766
Christoph Hellwig6b2f3d12009-10-27 11:05:28 +01001767 /*
1768 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1769 * check for O_DSYNC if the need any syncing at all we enforce it's
1770 * always set instead of having to deal with possibly weird behaviour
1771 * for malicious applications setting only __O_SYNC.
1772 */
1773 if (open_flag & __O_SYNC)
1774 open_flag |= O_DSYNC;
1775
Al Viro6e8341a2009-04-06 11:16:22 -04001776 if (!acc_mode)
Al Viro6d125522009-12-24 06:58:56 -05001777 acc_mode = MAY_OPEN | ACC_MODE(open_flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778
Trond Myklebust834f2a42005-10-18 14:20:16 -07001779 /* O_TRUNC implies we need access checks for write permissions */
1780 if (flag & O_TRUNC)
1781 acc_mode |= MAY_WRITE;
1782
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 /* Allow the LSM permission hook to distinguish append
1784 access from general write access. */
1785 if (flag & O_APPEND)
1786 acc_mode |= MAY_APPEND;
1787
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 /*
1789 * The simplest case - just a plain lookup.
1790 */
1791 if (!(flag & O_CREAT)) {
Al Viro2dd6d1f2009-08-13 20:40:45 +04001792 filp = get_empty_filp();
1793
1794 if (filp == NULL)
1795 return ERR_PTR(-ENFILE);
1796 nd.intent.open.file = filp;
Al Viro482928d2009-12-19 10:10:39 -05001797 filp->f_flags = open_flag;
Al Viro2dd6d1f2009-08-13 20:40:45 +04001798 nd.intent.open.flags = flag;
1799 nd.intent.open.create_mode = 0;
1800 error = do_path_lookup(dfd, pathname,
1801 lookup_flags(flag)|LOOKUP_OPEN, &nd);
1802 if (IS_ERR(nd.intent.open.file)) {
1803 if (error == 0) {
1804 error = PTR_ERR(nd.intent.open.file);
1805 path_put(&nd.path);
1806 }
1807 } else if (error)
1808 release_open_intent(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 if (error)
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001810 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 goto ok;
1812 }
1813
1814 /*
1815 * Create - we need to know the parent.
1816 */
Al Viro9850c052010-01-13 15:01:15 -05001817reval:
Al Viro9b4a9b12009-04-07 11:44:16 -04001818 error = path_init(dfd, pathname, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 if (error)
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001820 return ERR_PTR(error);
Al Viro9850c052010-01-13 15:01:15 -05001821 if (force_reval)
1822 nd.flags |= LOOKUP_REVAL;
Al Viro9b4a9b12009-04-07 11:44:16 -04001823 error = path_walk(pathname, &nd);
J. R. Okajima654f5622009-06-18 23:30:15 +09001824 if (error) {
1825 if (nd.root.mnt)
1826 path_put(&nd.root);
Al Viro9b4a9b12009-04-07 11:44:16 -04001827 return ERR_PTR(error);
J. R. Okajima654f5622009-06-18 23:30:15 +09001828 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001829 if (unlikely(!audit_dummy_context()))
1830 audit_inode(pathname, nd.path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831
1832 /*
Al Viroa2c36b42009-12-24 03:39:50 -05001833 * We have the parent and last component.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835
Al Viro8737f3a2008-08-02 22:36:57 -04001836 error = -ENFILE;
1837 filp = get_empty_filp();
1838 if (filp == NULL)
1839 goto exit_parent;
1840 nd.intent.open.file = filp;
Al Viro482928d2009-12-19 10:10:39 -05001841 filp->f_flags = open_flag;
Al Viro8737f3a2008-08-02 22:36:57 -04001842 nd.intent.open.flags = flag;
1843 nd.intent.open.create_mode = mode;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001844 nd.flags &= ~LOOKUP_PARENT;
Al Viro8737f3a2008-08-02 22:36:57 -04001845 nd.flags |= LOOKUP_CREATE | LOOKUP_OPEN;
Al Viro35165862008-08-05 03:00:49 -04001846 if (flag & O_EXCL)
1847 nd.flags |= LOOKUP_EXCL;
Al Virofb1cc552009-12-24 01:58:28 -05001848 filp = do_last(&nd, &path, open_flag, flag, acc_mode, mode,
Al Viroa1e28032009-12-24 02:12:06 -05001849 pathname, &is_link);
Al Virofb1cc552009-12-24 01:58:28 -05001850 if (is_link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851 goto do_link;
Al Viro3343eb82009-12-24 02:02:38 -05001852 if (nd.root.mnt)
1853 path_put(&nd.root);
Al Virofb1cc552009-12-24 01:58:28 -05001854 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856ok:
Al Viro9a661792009-12-24 06:49:47 -05001857 filp = finish_open(&nd, open_flag, acc_mode);
J. R. Okajima654f5622009-06-18 23:30:15 +09001858 if (nd.root.mnt)
1859 path_put(&nd.root);
Dave Hansen4a3fd212008-02-15 14:37:48 -08001860 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861
1862exit_dput:
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001863 path_put_conditional(&path, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864exit:
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001865 if (!IS_ERR(nd.intent.open.file))
1866 release_open_intent(&nd);
Al Viro8737f3a2008-08-02 22:36:57 -04001867exit_parent:
Al Viro2a737872009-04-07 11:49:53 -04001868 if (nd.root.mnt)
1869 path_put(&nd.root);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001870 path_put(&nd.path);
1871 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872
1873do_link:
1874 error = -ELOOP;
Al Viroc99658f2009-12-24 02:27:30 -05001875 if ((flag & O_NOFOLLOW) || count++ == 32)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876 goto exit_dput;
1877 /*
1878 * This is subtle. Instead of calling do_follow_link() we do the
1879 * thing by hands. The reason is that this way we have zero link_count
1880 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1881 * After that we have the parent and last component, i.e.
1882 * we are in the same situation as after the first path_walk().
1883 * Well, almost - if the last component is normal we get its copy
1884 * stored in nd->last.name and we will have to putname() it when we
1885 * are done. Procfs-like symlinks just set LAST_BIND.
1886 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001887 nd.flags |= LOOKUP_PARENT;
1888 error = security_inode_follow_link(path.dentry, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 if (error)
1890 goto exit_dput;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001891 error = __do_follow_link(&path, &nd);
Al Viro258fa992009-08-09 01:32:02 +04001892 path_put(&path);
Kirill Korotaevde459212006-07-14 00:23:49 -07001893 if (error) {
1894 /* Does someone understand code flow here? Or it is only
1895 * me so stupid? Anathema to whoever designed this non-sense
1896 * with "intent.open".
1897 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001898 release_open_intent(&nd);
J. R. Okajima654f5622009-06-18 23:30:15 +09001899 if (nd.root.mnt)
1900 path_put(&nd.root);
Al Viro9850c052010-01-13 15:01:15 -05001901 if (error == -ESTALE && !force_reval) {
1902 force_reval = 1;
1903 goto reval;
1904 }
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001905 return ERR_PTR(error);
Kirill Korotaevde459212006-07-14 00:23:49 -07001906 }
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001907 nd.flags &= ~LOOKUP_PARENT;
1908 if (nd.last_type == LAST_BIND)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 goto ok;
Al Viro27bff342009-12-24 02:05:43 -05001910 filp = do_last(&nd, &path, open_flag, flag, acc_mode, mode,
Al Viroa1e28032009-12-24 02:12:06 -05001911 pathname, &is_link);
Al Viroa2c36b42009-12-24 03:39:50 -05001912 if (nd.last_type == LAST_NORM)
1913 __putname(nd.last.name);
Al Viro27bff342009-12-24 02:05:43 -05001914 if (is_link)
1915 goto do_link;
1916 if (nd.root.mnt)
1917 path_put(&nd.root);
1918 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919}
1920
1921/**
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001922 * filp_open - open file and return file pointer
1923 *
1924 * @filename: path to open
1925 * @flags: open flags as per the open(2) second argument
1926 * @mode: mode for the new file if O_CREAT is set, else ignored
1927 *
1928 * This is the helper to open a file from kernelspace if you really
1929 * have to. But in generally you should not do this, so please move
1930 * along, nothing to see here..
1931 */
1932struct file *filp_open(const char *filename, int flags, int mode)
1933{
Al Viro6e8341a2009-04-06 11:16:22 -04001934 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001935}
1936EXPORT_SYMBOL(filp_open);
1937
1938/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 * lookup_create - lookup a dentry, creating it if it doesn't exist
1940 * @nd: nameidata info
1941 * @is_dir: directory flag
1942 *
1943 * Simple function to lookup and return a dentry and create it
1944 * if it doesn't exist. Is SMP-safe.
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001945 *
Jan Blunck4ac91372008-02-14 19:34:32 -08001946 * Returns with nd->path.dentry->d_inode->i_mutex locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 */
1948struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1949{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001950 struct dentry *dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951
Jan Blunck4ac91372008-02-14 19:34:32 -08001952 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001953 /*
1954 * Yucky last component or no last component at all?
1955 * (foo/., foo/.., /////)
1956 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957 if (nd->last_type != LAST_NORM)
1958 goto fail;
1959 nd->flags &= ~LOOKUP_PARENT;
Al Viro35165862008-08-05 03:00:49 -04001960 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
ASANO Masahiroa6349042006-08-22 20:06:02 -04001961 nd->intent.open.flags = O_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001962
1963 /*
1964 * Do the final lookup.
1965 */
Christoph Hellwig49705b72005-11-08 21:35:06 -08001966 dentry = lookup_hash(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 if (IS_ERR(dentry))
1968 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001969
Al Viroe9baf6e2008-05-15 04:49:12 -04001970 if (dentry->d_inode)
1971 goto eexist;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001972 /*
1973 * Special case - lookup gave negative, but... we had foo/bar/
1974 * From the vfs_mknod() POV we just have a negative dentry -
1975 * all is fine. Let's be bastards - you had / on the end, you've
1976 * been asking for (non-existent) directory. -ENOENT for you.
1977 */
Al Viroe9baf6e2008-05-15 04:49:12 -04001978 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1979 dput(dentry);
1980 dentry = ERR_PTR(-ENOENT);
1981 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 return dentry;
Al Viroe9baf6e2008-05-15 04:49:12 -04001983eexist:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 dput(dentry);
Al Viroe9baf6e2008-05-15 04:49:12 -04001985 dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986fail:
1987 return dentry;
1988}
Christoph Hellwigf81a0bf2005-05-19 12:26:43 -07001989EXPORT_SYMBOL_GPL(lookup_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990
1991int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1992{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001993 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994
1995 if (error)
1996 return error;
1997
1998 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1999 return -EPERM;
2000
Al Viroacfa4382008-12-04 10:06:33 -05002001 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 return -EPERM;
2003
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07002004 error = devcgroup_inode_mknod(mode, dev);
2005 if (error)
2006 return error;
2007
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008 error = security_inode_mknod(dir, dentry, mode, dev);
2009 if (error)
2010 return error;
2011
Jan Kara9e3509e2009-01-26 16:45:12 +01002012 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002014 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002015 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 return error;
2017}
2018
Dave Hansen463c3192008-02-15 14:37:57 -08002019static int may_mknod(mode_t mode)
2020{
2021 switch (mode & S_IFMT) {
2022 case S_IFREG:
2023 case S_IFCHR:
2024 case S_IFBLK:
2025 case S_IFIFO:
2026 case S_IFSOCK:
2027 case 0: /* zero mode translates to S_IFREG */
2028 return 0;
2029 case S_IFDIR:
2030 return -EPERM;
2031 default:
2032 return -EINVAL;
2033 }
2034}
2035
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002036SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2037 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038{
Al Viro2ad94ae2008-07-21 09:32:51 -04002039 int error;
2040 char *tmp;
2041 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 struct nameidata nd;
2043
2044 if (S_ISDIR(mode))
2045 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046
Al Viro2ad94ae2008-07-21 09:32:51 -04002047 error = user_path_parent(dfd, filename, &nd, &tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002049 return error;
2050
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 dentry = lookup_create(&nd, 0);
Dave Hansen463c3192008-02-15 14:37:57 -08002052 if (IS_ERR(dentry)) {
2053 error = PTR_ERR(dentry);
2054 goto out_unlock;
2055 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002056 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002057 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002058 error = may_mknod(mode);
2059 if (error)
2060 goto out_dput;
2061 error = mnt_want_write(nd.path.mnt);
2062 if (error)
2063 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002064 error = security_path_mknod(&nd.path, dentry, mode, dev);
2065 if (error)
2066 goto out_drop_write;
Dave Hansen463c3192008-02-15 14:37:57 -08002067 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068 case 0: case S_IFREG:
Jan Blunck4ac91372008-02-14 19:34:32 -08002069 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 break;
2071 case S_IFCHR: case S_IFBLK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002072 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073 new_decode_dev(dev));
2074 break;
2075 case S_IFIFO: case S_IFSOCK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002076 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002079out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002080 mnt_drop_write(nd.path.mnt);
2081out_dput:
2082 dput(dentry);
2083out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002084 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002085 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086 putname(tmp);
2087
2088 return error;
2089}
2090
Heiko Carstens3480b252009-01-14 14:14:16 +01002091SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002092{
2093 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2094}
2095
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2097{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002098 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099
2100 if (error)
2101 return error;
2102
Al Viroacfa4382008-12-04 10:06:33 -05002103 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104 return -EPERM;
2105
2106 mode &= (S_IRWXUGO|S_ISVTX);
2107 error = security_inode_mkdir(dir, dentry, mode);
2108 if (error)
2109 return error;
2110
Jan Kara9e3509e2009-01-26 16:45:12 +01002111 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002113 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002114 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 return error;
2116}
2117
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002118SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119{
2120 int error = 0;
2121 char * tmp;
Dave Hansen6902d922006-09-30 23:29:01 -07002122 struct dentry *dentry;
2123 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124
Al Viro2ad94ae2008-07-21 09:32:51 -04002125 error = user_path_parent(dfd, pathname, &nd, &tmp);
2126 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002127 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128
Dave Hansen6902d922006-09-30 23:29:01 -07002129 dentry = lookup_create(&nd, 1);
2130 error = PTR_ERR(dentry);
2131 if (IS_ERR(dentry))
2132 goto out_unlock;
2133
Jan Blunck4ac91372008-02-14 19:34:32 -08002134 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002135 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002136 error = mnt_want_write(nd.path.mnt);
2137 if (error)
2138 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002139 error = security_path_mkdir(&nd.path, dentry, mode);
2140 if (error)
2141 goto out_drop_write;
Jan Blunck4ac91372008-02-14 19:34:32 -08002142 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002143out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002144 mnt_drop_write(nd.path.mnt);
2145out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002146 dput(dentry);
2147out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002148 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002149 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002150 putname(tmp);
2151out_err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 return error;
2153}
2154
Heiko Carstens3cdad422009-01-14 14:14:22 +01002155SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002156{
2157 return sys_mkdirat(AT_FDCWD, pathname, mode);
2158}
2159
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160/*
2161 * We try to drop the dentry early: we should have
2162 * a usage count of 2 if we're the only user of this
2163 * dentry, and if that is true (possibly after pruning
2164 * the dcache), then we drop the dentry now.
2165 *
2166 * A low-level filesystem can, if it choses, legally
2167 * do a
2168 *
2169 * if (!d_unhashed(dentry))
2170 * return -EBUSY;
2171 *
2172 * if it cannot handle the case of removing a directory
2173 * that is still in use by something else..
2174 */
2175void dentry_unhash(struct dentry *dentry)
2176{
2177 dget(dentry);
Vasily Averindc168422006-12-06 20:37:07 -08002178 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179 spin_lock(&dcache_lock);
2180 spin_lock(&dentry->d_lock);
2181 if (atomic_read(&dentry->d_count) == 2)
2182 __d_drop(dentry);
2183 spin_unlock(&dentry->d_lock);
2184 spin_unlock(&dcache_lock);
2185}
2186
2187int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2188{
2189 int error = may_delete(dir, dentry, 1);
2190
2191 if (error)
2192 return error;
2193
Al Viroacfa4382008-12-04 10:06:33 -05002194 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195 return -EPERM;
2196
Jan Kara9e3509e2009-01-26 16:45:12 +01002197 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002199 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 dentry_unhash(dentry);
2201 if (d_mountpoint(dentry))
2202 error = -EBUSY;
2203 else {
2204 error = security_inode_rmdir(dir, dentry);
2205 if (!error) {
2206 error = dir->i_op->rmdir(dir, dentry);
2207 if (!error)
2208 dentry->d_inode->i_flags |= S_DEAD;
2209 }
2210 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002211 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213 d_delete(dentry);
2214 }
2215 dput(dentry);
2216
2217 return error;
2218}
2219
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002220static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221{
2222 int error = 0;
2223 char * name;
2224 struct dentry *dentry;
2225 struct nameidata nd;
2226
Al Viro2ad94ae2008-07-21 09:32:51 -04002227 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002229 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230
2231 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002232 case LAST_DOTDOT:
2233 error = -ENOTEMPTY;
2234 goto exit1;
2235 case LAST_DOT:
2236 error = -EINVAL;
2237 goto exit1;
2238 case LAST_ROOT:
2239 error = -EBUSY;
2240 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002242
2243 nd.flags &= ~LOOKUP_PARENT;
2244
Jan Blunck4ac91372008-02-14 19:34:32 -08002245 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002246 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002248 if (IS_ERR(dentry))
2249 goto exit2;
Dave Hansen06227532008-02-15 14:37:34 -08002250 error = mnt_want_write(nd.path.mnt);
2251 if (error)
2252 goto exit3;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002253 error = security_path_rmdir(&nd.path, dentry);
2254 if (error)
2255 goto exit4;
Jan Blunck4ac91372008-02-14 19:34:32 -08002256 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002257exit4:
Dave Hansen06227532008-02-15 14:37:34 -08002258 mnt_drop_write(nd.path.mnt);
2259exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07002260 dput(dentry);
2261exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08002262 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002264 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265 putname(name);
2266 return error;
2267}
2268
Heiko Carstens3cdad422009-01-14 14:14:22 +01002269SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002270{
2271 return do_rmdir(AT_FDCWD, pathname);
2272}
2273
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274int vfs_unlink(struct inode *dir, struct dentry *dentry)
2275{
2276 int error = may_delete(dir, dentry, 0);
2277
2278 if (error)
2279 return error;
2280
Al Viroacfa4382008-12-04 10:06:33 -05002281 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282 return -EPERM;
2283
Jan Kara9e3509e2009-01-26 16:45:12 +01002284 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002286 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 if (d_mountpoint(dentry))
2288 error = -EBUSY;
2289 else {
2290 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002291 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 error = dir->i_op->unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002293 if (!error)
2294 dentry->d_inode->i_flags |= S_DEAD;
2295 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002297 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298
2299 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2300 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08002301 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303 }
Robert Love0eeca282005-07-12 17:06:03 -04002304
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305 return error;
2306}
2307
2308/*
2309 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002310 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 * writeout happening, and we don't want to prevent access to the directory
2312 * while waiting on the I/O.
2313 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002314static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315{
Al Viro2ad94ae2008-07-21 09:32:51 -04002316 int error;
2317 char *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318 struct dentry *dentry;
2319 struct nameidata nd;
2320 struct inode *inode = NULL;
2321
Al Viro2ad94ae2008-07-21 09:32:51 -04002322 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002324 return error;
2325
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326 error = -EISDIR;
2327 if (nd.last_type != LAST_NORM)
2328 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002329
2330 nd.flags &= ~LOOKUP_PARENT;
2331
Jan Blunck4ac91372008-02-14 19:34:32 -08002332 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002333 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334 error = PTR_ERR(dentry);
2335 if (!IS_ERR(dentry)) {
2336 /* Why not before? Because we want correct error value */
2337 if (nd.last.name[nd.last.len])
2338 goto slashes;
2339 inode = dentry->d_inode;
2340 if (inode)
2341 atomic_inc(&inode->i_count);
Dave Hansen06227532008-02-15 14:37:34 -08002342 error = mnt_want_write(nd.path.mnt);
2343 if (error)
2344 goto exit2;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002345 error = security_path_unlink(&nd.path, dentry);
2346 if (error)
2347 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08002348 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002349exit3:
Dave Hansen06227532008-02-15 14:37:34 -08002350 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002351 exit2:
2352 dput(dentry);
2353 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002354 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355 if (inode)
2356 iput(inode); /* truncate the inode here */
2357exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002358 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359 putname(name);
2360 return error;
2361
2362slashes:
2363 error = !dentry->d_inode ? -ENOENT :
2364 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2365 goto exit2;
2366}
2367
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002368SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002369{
2370 if ((flag & ~AT_REMOVEDIR) != 0)
2371 return -EINVAL;
2372
2373 if (flag & AT_REMOVEDIR)
2374 return do_rmdir(dfd, pathname);
2375
2376 return do_unlinkat(dfd, pathname);
2377}
2378
Heiko Carstens3480b252009-01-14 14:14:16 +01002379SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002380{
2381 return do_unlinkat(AT_FDCWD, pathname);
2382}
2383
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002384int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002386 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387
2388 if (error)
2389 return error;
2390
Al Viroacfa4382008-12-04 10:06:33 -05002391 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392 return -EPERM;
2393
2394 error = security_inode_symlink(dir, dentry, oldname);
2395 if (error)
2396 return error;
2397
Jan Kara9e3509e2009-01-26 16:45:12 +01002398 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002400 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002401 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402 return error;
2403}
2404
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002405SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2406 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407{
Al Viro2ad94ae2008-07-21 09:32:51 -04002408 int error;
2409 char *from;
2410 char *to;
Dave Hansen6902d922006-09-30 23:29:01 -07002411 struct dentry *dentry;
2412 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413
2414 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04002415 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416 return PTR_ERR(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002417
2418 error = user_path_parent(newdfd, newname, &nd, &to);
2419 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002420 goto out_putname;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421
Dave Hansen6902d922006-09-30 23:29:01 -07002422 dentry = lookup_create(&nd, 0);
2423 error = PTR_ERR(dentry);
2424 if (IS_ERR(dentry))
2425 goto out_unlock;
2426
Dave Hansen75c3f292008-02-15 14:37:45 -08002427 error = mnt_want_write(nd.path.mnt);
2428 if (error)
2429 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002430 error = security_path_symlink(&nd.path, dentry, from);
2431 if (error)
2432 goto out_drop_write;
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002433 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002434out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002435 mnt_drop_write(nd.path.mnt);
2436out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002437 dput(dentry);
2438out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002439 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002440 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002441 putname(to);
2442out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443 putname(from);
2444 return error;
2445}
2446
Heiko Carstens3480b252009-01-14 14:14:16 +01002447SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002448{
2449 return sys_symlinkat(oldname, AT_FDCWD, newname);
2450}
2451
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2453{
2454 struct inode *inode = old_dentry->d_inode;
2455 int error;
2456
2457 if (!inode)
2458 return -ENOENT;
2459
Miklos Szeredia95164d2008-07-30 15:08:48 +02002460 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461 if (error)
2462 return error;
2463
2464 if (dir->i_sb != inode->i_sb)
2465 return -EXDEV;
2466
2467 /*
2468 * A link to an append-only or immutable file cannot be created.
2469 */
2470 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2471 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05002472 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002473 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002474 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475 return -EPERM;
2476
2477 error = security_inode_link(old_dentry, dir, new_dentry);
2478 if (error)
2479 return error;
2480
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002481 mutex_lock(&inode->i_mutex);
Jan Kara9e3509e2009-01-26 16:45:12 +01002482 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483 error = dir->i_op->link(old_dentry, dir, new_dentry);
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002484 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002485 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002486 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487 return error;
2488}
2489
2490/*
2491 * Hardlinks are often used in delicate situations. We avoid
2492 * security-related surprises by not following symlinks on the
2493 * newname. --KAB
2494 *
2495 * We don't follow them on the oldname either to be compatible
2496 * with linux 2.0, and to avoid hard-linking to directories
2497 * and other special files. --ADM
2498 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002499SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2500 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002501{
2502 struct dentry *new_dentry;
Al Viro2d8f3032008-07-22 09:59:21 -04002503 struct nameidata nd;
2504 struct path old_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505 int error;
Al Viro2ad94ae2008-07-21 09:32:51 -04002506 char *to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507
Ulrich Drepper45c9b112006-06-25 05:49:11 -07002508 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002509 return -EINVAL;
2510
Al Viro2d8f3032008-07-22 09:59:21 -04002511 error = user_path_at(olddfd, oldname,
2512 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2513 &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002515 return error;
2516
2517 error = user_path_parent(newdfd, newname, &nd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518 if (error)
2519 goto out;
2520 error = -EXDEV;
Al Viro2d8f3032008-07-22 09:59:21 -04002521 if (old_path.mnt != nd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522 goto out_release;
2523 new_dentry = lookup_create(&nd, 0);
2524 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002525 if (IS_ERR(new_dentry))
2526 goto out_unlock;
Dave Hansen75c3f292008-02-15 14:37:45 -08002527 error = mnt_want_write(nd.path.mnt);
2528 if (error)
2529 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002530 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2531 if (error)
2532 goto out_drop_write;
Al Viro2d8f3032008-07-22 09:59:21 -04002533 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002534out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002535 mnt_drop_write(nd.path.mnt);
2536out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002537 dput(new_dentry);
2538out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002539 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540out_release:
Jan Blunck1d957f92008-02-14 19:34:35 -08002541 path_put(&nd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002542 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543out:
Al Viro2d8f3032008-07-22 09:59:21 -04002544 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545
2546 return error;
2547}
2548
Heiko Carstens3480b252009-01-14 14:14:16 +01002549SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002550{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002551 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002552}
2553
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554/*
2555 * The worst of all namespace operations - renaming directory. "Perverted"
2556 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2557 * Problems:
2558 * a) we can get into loop creation. Check is done in is_subdir().
2559 * b) race potential - two innocent renames can create a loop together.
2560 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002561 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562 * story.
2563 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002564 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565 * whether the target exists). Solution: try to be smart with locking
2566 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002567 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568 * move will be locked. Thus we can rank directories by the tree
2569 * (ancestors first) and rank all non-directories after them.
2570 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002571 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572 * HOWEVER, it relies on the assumption that any object with ->lookup()
2573 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2574 * we'd better make sure that there's no link(2) for them.
2575 * d) some filesystems don't support opened-but-unlinked directories,
2576 * either because of layout or because they are not ready to deal with
2577 * all cases correctly. The latter will be fixed (taking this sort of
2578 * stuff into VFS), but the former is not going away. Solution: the same
2579 * trick as in rmdir().
2580 * e) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002581 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002583 * ->i_mutex on parents, which works but leads to some truely excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 * locking].
2585 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07002586static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2587 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588{
2589 int error = 0;
2590 struct inode *target;
2591
2592 /*
2593 * If we are going to change the parent - check write permissions,
2594 * we'll need to flip '..'.
2595 */
2596 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04002597 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598 if (error)
2599 return error;
2600 }
2601
2602 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2603 if (error)
2604 return error;
2605
2606 target = new_dentry->d_inode;
2607 if (target) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002608 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609 dentry_unhash(new_dentry);
2610 }
2611 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2612 error = -EBUSY;
2613 else
2614 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2615 if (target) {
2616 if (!error)
2617 target->i_flags |= S_DEAD;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002618 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619 if (d_unhashed(new_dentry))
2620 d_rehash(new_dentry);
2621 dput(new_dentry);
2622 }
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002623 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07002624 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2625 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626 return error;
2627}
2628
Adrian Bunk75c96f82005-05-05 16:16:09 -07002629static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2630 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631{
2632 struct inode *target;
2633 int error;
2634
2635 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2636 if (error)
2637 return error;
2638
2639 dget(new_dentry);
2640 target = new_dentry->d_inode;
2641 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002642 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002643 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2644 error = -EBUSY;
2645 else
2646 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2647 if (!error) {
Al Virobec10522010-03-03 14:12:08 -05002648 if (target)
2649 target->i_flags |= S_DEAD;
Mark Fasheh349457c2006-09-08 14:22:21 -07002650 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651 d_move(old_dentry, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652 }
2653 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002654 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655 dput(new_dentry);
2656 return error;
2657}
2658
2659int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2660 struct inode *new_dir, struct dentry *new_dentry)
2661{
2662 int error;
2663 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Robert Love0eeca282005-07-12 17:06:03 -04002664 const char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665
2666 if (old_dentry->d_inode == new_dentry->d_inode)
2667 return 0;
2668
2669 error = may_delete(old_dir, old_dentry, is_dir);
2670 if (error)
2671 return error;
2672
2673 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02002674 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675 else
2676 error = may_delete(new_dir, new_dentry, is_dir);
2677 if (error)
2678 return error;
2679
Al Viroacfa4382008-12-04 10:06:33 -05002680 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681 return -EPERM;
2682
Jan Kara9e3509e2009-01-26 16:45:12 +01002683 vfs_dq_init(old_dir);
2684 vfs_dq_init(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002685
Robert Love0eeca282005-07-12 17:06:03 -04002686 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2687
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688 if (is_dir)
2689 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2690 else
2691 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
Al Viro123df292009-12-25 04:57:57 -05002692 if (!error)
2693 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04002694 new_dentry->d_inode, old_dentry);
Robert Love0eeca282005-07-12 17:06:03 -04002695 fsnotify_oldname_free(old_name);
2696
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697 return error;
2698}
2699
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002700SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2701 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702{
Al Viro2ad94ae2008-07-21 09:32:51 -04002703 struct dentry *old_dir, *new_dir;
2704 struct dentry *old_dentry, *new_dentry;
2705 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706 struct nameidata oldnd, newnd;
Al Viro2ad94ae2008-07-21 09:32:51 -04002707 char *from;
2708 char *to;
2709 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710
Al Viro2ad94ae2008-07-21 09:32:51 -04002711 error = user_path_parent(olddfd, oldname, &oldnd, &from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712 if (error)
2713 goto exit;
2714
Al Viro2ad94ae2008-07-21 09:32:51 -04002715 error = user_path_parent(newdfd, newname, &newnd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716 if (error)
2717 goto exit1;
2718
2719 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08002720 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002721 goto exit2;
2722
Jan Blunck4ac91372008-02-14 19:34:32 -08002723 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002724 error = -EBUSY;
2725 if (oldnd.last_type != LAST_NORM)
2726 goto exit2;
2727
Jan Blunck4ac91372008-02-14 19:34:32 -08002728 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729 if (newnd.last_type != LAST_NORM)
2730 goto exit2;
2731
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002732 oldnd.flags &= ~LOOKUP_PARENT;
2733 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09002734 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002735
Linus Torvalds1da177e2005-04-16 15:20:36 -07002736 trap = lock_rename(new_dir, old_dir);
2737
Christoph Hellwig49705b72005-11-08 21:35:06 -08002738 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002739 error = PTR_ERR(old_dentry);
2740 if (IS_ERR(old_dentry))
2741 goto exit3;
2742 /* source must exist */
2743 error = -ENOENT;
2744 if (!old_dentry->d_inode)
2745 goto exit4;
2746 /* unless the source is a directory trailing slashes give -ENOTDIR */
2747 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2748 error = -ENOTDIR;
2749 if (oldnd.last.name[oldnd.last.len])
2750 goto exit4;
2751 if (newnd.last.name[newnd.last.len])
2752 goto exit4;
2753 }
2754 /* source should not be ancestor of target */
2755 error = -EINVAL;
2756 if (old_dentry == trap)
2757 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08002758 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759 error = PTR_ERR(new_dentry);
2760 if (IS_ERR(new_dentry))
2761 goto exit4;
2762 /* target should not be an ancestor of source */
2763 error = -ENOTEMPTY;
2764 if (new_dentry == trap)
2765 goto exit5;
2766
Dave Hansen9079b1e2008-02-15 14:37:49 -08002767 error = mnt_want_write(oldnd.path.mnt);
2768 if (error)
2769 goto exit5;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002770 error = security_path_rename(&oldnd.path, old_dentry,
2771 &newnd.path, new_dentry);
2772 if (error)
2773 goto exit6;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774 error = vfs_rename(old_dir->d_inode, old_dentry,
2775 new_dir->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002776exit6:
Dave Hansen9079b1e2008-02-15 14:37:49 -08002777 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002778exit5:
2779 dput(new_dentry);
2780exit4:
2781 dput(old_dentry);
2782exit3:
2783 unlock_rename(new_dir, old_dir);
2784exit2:
Jan Blunck1d957f92008-02-14 19:34:35 -08002785 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002786 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002787exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002788 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002789 putname(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002790exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791 return error;
2792}
2793
Heiko Carstensa26eab22009-01-14 14:14:17 +01002794SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002795{
2796 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2797}
2798
Linus Torvalds1da177e2005-04-16 15:20:36 -07002799int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2800{
2801 int len;
2802
2803 len = PTR_ERR(link);
2804 if (IS_ERR(link))
2805 goto out;
2806
2807 len = strlen(link);
2808 if (len > (unsigned) buflen)
2809 len = buflen;
2810 if (copy_to_user(buffer, link, len))
2811 len = -EFAULT;
2812out:
2813 return len;
2814}
2815
2816/*
2817 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2818 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2819 * using) it for any given inode is up to filesystem.
2820 */
2821int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2822{
2823 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002824 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07002825 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002826
Linus Torvalds1da177e2005-04-16 15:20:36 -07002827 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002828 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07002829 if (IS_ERR(cookie))
2830 return PTR_ERR(cookie);
2831
2832 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2833 if (dentry->d_inode->i_op->put_link)
2834 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2835 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836}
2837
2838int vfs_follow_link(struct nameidata *nd, const char *link)
2839{
2840 return __vfs_follow_link(nd, link);
2841}
2842
2843/* get the link contents into pagecache */
2844static char *page_getlink(struct dentry * dentry, struct page **ppage)
2845{
Duane Griffinebd09ab2008-12-19 20:47:12 +00002846 char *kaddr;
2847 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002848 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07002849 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002850 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07002851 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00002853 kaddr = kmap(page);
2854 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2855 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002856}
2857
2858int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2859{
2860 struct page *page = NULL;
2861 char *s = page_getlink(dentry, &page);
2862 int res = vfs_readlink(dentry,buffer,buflen,s);
2863 if (page) {
2864 kunmap(page);
2865 page_cache_release(page);
2866 }
2867 return res;
2868}
2869
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002870void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002871{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002872 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002873 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002874 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002875}
2876
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002877void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002879 struct page *page = cookie;
2880
2881 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002882 kunmap(page);
2883 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002884 }
2885}
2886
Nick Piggin54566b22009-01-04 12:00:53 -08002887/*
2888 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2889 */
2890int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891{
2892 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002893 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002894 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08002895 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002896 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08002897 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2898 if (nofs)
2899 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002900
NeilBrown7e53cac2006-03-25 03:07:57 -08002901retry:
Nick Pigginafddba42007-10-16 01:25:01 -07002902 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08002903 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002904 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07002905 goto fail;
2906
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907 kaddr = kmap_atomic(page, KM_USER0);
2908 memcpy(kaddr, symname, len-1);
2909 kunmap_atomic(kaddr, KM_USER0);
Nick Pigginafddba42007-10-16 01:25:01 -07002910
2911 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2912 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002913 if (err < 0)
2914 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07002915 if (err < len-1)
2916 goto retry;
2917
Linus Torvalds1da177e2005-04-16 15:20:36 -07002918 mark_inode_dirty(inode);
2919 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002920fail:
2921 return err;
2922}
2923
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002924int page_symlink(struct inode *inode, const char *symname, int len)
2925{
2926 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08002927 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002928}
2929
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002930const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002931 .readlink = generic_readlink,
2932 .follow_link = page_follow_link_light,
2933 .put_link = page_put_link,
2934};
2935
Al Viro2d8f3032008-07-22 09:59:21 -04002936EXPORT_SYMBOL(user_path_at);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002937EXPORT_SYMBOL(follow_down);
2938EXPORT_SYMBOL(follow_up);
2939EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2940EXPORT_SYMBOL(getname);
2941EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002942EXPORT_SYMBOL(lookup_one_len);
2943EXPORT_SYMBOL(page_follow_link_light);
2944EXPORT_SYMBOL(page_put_link);
2945EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002946EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002947EXPORT_SYMBOL(page_symlink);
2948EXPORT_SYMBOL(page_symlink_inode_operations);
2949EXPORT_SYMBOL(path_lookup);
Al Virod1811462008-08-02 00:49:18 -04002950EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002951EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04002952EXPORT_SYMBOL(inode_permission);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -08002953EXPORT_SYMBOL(file_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002954EXPORT_SYMBOL(unlock_rename);
2955EXPORT_SYMBOL(vfs_create);
2956EXPORT_SYMBOL(vfs_follow_link);
2957EXPORT_SYMBOL(vfs_link);
2958EXPORT_SYMBOL(vfs_mkdir);
2959EXPORT_SYMBOL(vfs_mknod);
2960EXPORT_SYMBOL(generic_permission);
2961EXPORT_SYMBOL(vfs_readlink);
2962EXPORT_SYMBOL(vfs_rename);
2963EXPORT_SYMBOL(vfs_rmdir);
2964EXPORT_SYMBOL(vfs_symlink);
2965EXPORT_SYMBOL(vfs_unlink);
2966EXPORT_SYMBOL(dentry_unhash);
2967EXPORT_SYMBOL(generic_readlink);