blob: a4855af776a8b6670b753aa3af71203519de94a7 [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) {
692 struct vfsmount *parent;
Jan Blunck4ac91372008-02-14 19:34:32 -0800693 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694
Al Viro2a737872009-04-07 11:49:53 -0400695 if (nd->path.dentry == nd->root.dentry &&
696 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 break;
698 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 spin_lock(&dcache_lock);
Jan Blunck4ac91372008-02-14 19:34:32 -0800700 if (nd->path.dentry != nd->path.mnt->mnt_root) {
701 nd->path.dentry = dget(nd->path.dentry->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 spin_unlock(&dcache_lock);
703 dput(old);
704 break;
705 }
706 spin_unlock(&dcache_lock);
707 spin_lock(&vfsmount_lock);
Jan Blunck4ac91372008-02-14 19:34:32 -0800708 parent = nd->path.mnt->mnt_parent;
709 if (parent == nd->path.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 spin_unlock(&vfsmount_lock);
711 break;
712 }
713 mntget(parent);
Jan Blunck4ac91372008-02-14 19:34:32 -0800714 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 spin_unlock(&vfsmount_lock);
716 dput(old);
Jan Blunck4ac91372008-02-14 19:34:32 -0800717 mntput(nd->path.mnt);
718 nd->path.mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 }
Al Viro79ed0222009-04-18 13:59:41 -0400720 follow_mount(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721}
722
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723/*
724 * It's more convoluted than I'd like it to be, but... it's still fairly
725 * small and for now I'd prefer to have fast path as straight as possible.
726 * It _is_ time-critical.
727 */
728static int do_lookup(struct nameidata *nd, struct qstr *name,
729 struct path *path)
730{
Jan Blunck4ac91372008-02-14 19:34:32 -0800731 struct vfsmount *mnt = nd->path.mnt;
Al Viro6e6b1bd2009-08-13 23:38:37 +0400732 struct dentry *dentry, *parent;
733 struct inode *dir;
Al Viro3cac2602009-08-13 18:27:43 +0400734 /*
735 * See if the low-level filesystem might want
736 * to use its own hash..
737 */
738 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
739 int err = nd->path.dentry->d_op->d_hash(nd->path.dentry, name);
740 if (err < 0)
741 return err;
742 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743
Al Viro3cac2602009-08-13 18:27:43 +0400744 dentry = __d_lookup(nd->path.dentry, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 if (!dentry)
746 goto need_lookup;
747 if (dentry->d_op && dentry->d_op->d_revalidate)
748 goto need_revalidate;
749done:
750 path->mnt = mnt;
751 path->dentry = dentry;
Al Viro634ee702005-06-06 13:36:13 -0700752 __follow_mount(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 return 0;
754
755need_lookup:
Al Viro6e6b1bd2009-08-13 23:38:37 +0400756 parent = nd->path.dentry;
757 dir = parent->d_inode;
758
759 mutex_lock(&dir->i_mutex);
760 /*
761 * First re-do the cached lookup just in case it was created
762 * while we waited for the directory semaphore..
763 *
764 * FIXME! This could use version numbering or similar to
765 * avoid unnecessary cache lookups.
766 *
767 * The "dcache_lock" is purely to protect the RCU list walker
768 * from concurrent renames at this point (we mustn't get false
769 * negatives from the RCU list walk here, unlike the optimistic
770 * fast walk).
771 *
772 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
773 */
774 dentry = d_lookup(parent, name);
775 if (!dentry) {
776 struct dentry *new;
777
778 /* Don't create child dentry for a dead directory. */
779 dentry = ERR_PTR(-ENOENT);
780 if (IS_DEADDIR(dir))
781 goto out_unlock;
782
783 new = d_alloc(parent, name);
784 dentry = ERR_PTR(-ENOMEM);
785 if (new) {
786 dentry = dir->i_op->lookup(dir, new, nd);
787 if (dentry)
788 dput(new);
789 else
790 dentry = new;
791 }
792out_unlock:
793 mutex_unlock(&dir->i_mutex);
794 if (IS_ERR(dentry))
795 goto fail;
796 goto done;
797 }
798
799 /*
800 * Uhhuh! Nasty case: the cache was re-populated while
801 * we waited on the semaphore. Need to revalidate.
802 */
803 mutex_unlock(&dir->i_mutex);
804 if (dentry->d_op && dentry->d_op->d_revalidate) {
805 dentry = do_revalidate(dentry, nd);
806 if (!dentry)
807 dentry = ERR_PTR(-ENOENT);
808 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 if (IS_ERR(dentry))
810 goto fail;
811 goto done;
812
813need_revalidate:
Ian Kentbcdc5e02006-09-27 01:50:44 -0700814 dentry = do_revalidate(dentry, nd);
815 if (!dentry)
816 goto need_lookup;
817 if (IS_ERR(dentry))
818 goto fail;
819 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820
821fail:
822 return PTR_ERR(dentry);
823}
824
825/*
Al Viroac278a92010-02-16 18:09:36 +0000826 * This is a temporary kludge to deal with "automount" symlinks; proper
827 * solution is to trigger them on follow_mount(), so that do_lookup()
828 * would DTRT. To be killed before 2.6.34-final.
829 */
830static inline int follow_on_final(struct inode *inode, unsigned lookup_flags)
831{
832 return inode && unlikely(inode->i_op->follow_link) &&
833 ((lookup_flags & LOOKUP_FOLLOW) || S_ISDIR(inode->i_mode));
834}
835
836/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +0100838 * This is the basic name resolution function, turning a pathname into
839 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 *
Prasanna Medaea3834d2005-04-29 16:00:17 +0100841 * Returns 0 and nd will have valid dentry and mnt on success.
842 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 */
Al Viro6de88d72009-08-09 01:41:57 +0400844static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845{
846 struct path next;
847 struct inode *inode;
848 int err;
849 unsigned int lookup_flags = nd->flags;
850
851 while (*name=='/')
852 name++;
853 if (!*name)
854 goto return_reval;
855
Jan Blunck4ac91372008-02-14 19:34:32 -0800856 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 if (nd->depth)
Trond Myklebustf55eab82006-02-04 23:28:01 -0800858 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859
860 /* At this point we know we have a real path component. */
861 for(;;) {
862 unsigned long hash;
863 struct qstr this;
864 unsigned int c;
865
Trond Myklebustcdce5d62005-10-18 14:20:18 -0700866 nd->flags |= LOOKUP_CONTINUE;
Al Virob75b5082009-12-16 01:01:38 -0500867 err = exec_permission(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 if (err)
869 break;
870
871 this.name = name;
872 c = *(const unsigned char *)name;
873
874 hash = init_name_hash();
875 do {
876 name++;
877 hash = partial_name_hash(c, hash);
878 c = *(const unsigned char *)name;
879 } while (c && (c != '/'));
880 this.len = name - (const char *) this.name;
881 this.hash = end_name_hash(hash);
882
883 /* remove trailing slashes? */
884 if (!c)
885 goto last_component;
886 while (*++name == '/');
887 if (!*name)
888 goto last_with_slashes;
889
890 /*
891 * "." and ".." are special - ".." especially so because it has
892 * to be able to know about the current root directory and
893 * parent relationships.
894 */
895 if (this.name[0] == '.') switch (this.len) {
896 default:
897 break;
898 case 2:
899 if (this.name[1] != '.')
900 break;
Al Viro58c465e2005-06-06 13:36:13 -0700901 follow_dotdot(nd);
Jan Blunck4ac91372008-02-14 19:34:32 -0800902 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 /* fallthrough */
904 case 1:
905 continue;
906 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 /* This does the actual lookups.. */
908 err = do_lookup(nd, &this, &next);
909 if (err)
910 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911
912 err = -ENOENT;
913 inode = next.dentry->d_inode;
914 if (!inode)
915 goto out_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916
917 if (inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700918 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 if (err)
920 goto return_err;
921 err = -ENOENT;
Jan Blunck4ac91372008-02-14 19:34:32 -0800922 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 if (!inode)
924 break;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700925 } else
926 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 err = -ENOTDIR;
928 if (!inode->i_op->lookup)
929 break;
930 continue;
931 /* here ends the main loop */
932
933last_with_slashes:
934 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
935last_component:
Trond Myklebustf55eab82006-02-04 23:28:01 -0800936 /* Clear LOOKUP_CONTINUE iff it was previously unset */
937 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 if (lookup_flags & LOOKUP_PARENT)
939 goto lookup_parent;
940 if (this.name[0] == '.') switch (this.len) {
941 default:
942 break;
943 case 2:
944 if (this.name[1] != '.')
945 break;
Al Viro58c465e2005-06-06 13:36:13 -0700946 follow_dotdot(nd);
Jan Blunck4ac91372008-02-14 19:34:32 -0800947 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 /* fallthrough */
949 case 1:
950 goto return_reval;
951 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 err = do_lookup(nd, &this, &next);
953 if (err)
954 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 inode = next.dentry->d_inode;
Al Viroac278a92010-02-16 18:09:36 +0000956 if (follow_on_final(inode, lookup_flags)) {
Al Viro90ebe562005-06-06 13:35:58 -0700957 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 if (err)
959 goto return_err;
Jan Blunck4ac91372008-02-14 19:34:32 -0800960 inode = nd->path.dentry->d_inode;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700961 } else
962 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 err = -ENOENT;
964 if (!inode)
965 break;
966 if (lookup_flags & LOOKUP_DIRECTORY) {
967 err = -ENOTDIR;
Al Viroacfa4382008-12-04 10:06:33 -0500968 if (!inode->i_op->lookup)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 break;
970 }
971 goto return_base;
972lookup_parent:
973 nd->last = this;
974 nd->last_type = LAST_NORM;
975 if (this.name[0] != '.')
976 goto return_base;
977 if (this.len == 1)
978 nd->last_type = LAST_DOT;
979 else if (this.len == 2 && this.name[1] == '.')
980 nd->last_type = LAST_DOTDOT;
981 else
982 goto return_base;
983return_reval:
984 /*
985 * We bypassed the ordinary revalidation routines.
986 * We may need to check the cached dentry for staleness.
987 */
Jan Blunck4ac91372008-02-14 19:34:32 -0800988 if (nd->path.dentry && nd->path.dentry->d_sb &&
989 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 err = -ESTALE;
991 /* Note: we do not d_invalidate() */
Jan Blunck4ac91372008-02-14 19:34:32 -0800992 if (!nd->path.dentry->d_op->d_revalidate(
993 nd->path.dentry, nd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 break;
995 }
996return_base:
997 return 0;
998out_dput:
Jan Blunck1d957f92008-02-14 19:34:35 -0800999 path_put_conditional(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 break;
1001 }
Jan Blunck1d957f92008-02-14 19:34:35 -08001002 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003return_err:
1004 return err;
1005}
1006
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -08001007static int path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008{
Al Viro6de88d72009-08-09 01:41:57 +04001009 struct path save = nd->path;
1010 int result;
1011
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 current->total_link_count = 0;
Al Viro6de88d72009-08-09 01:41:57 +04001013
1014 /* make sure the stuff we saved doesn't go away */
1015 path_get(&save);
1016
1017 result = link_path_walk(name, nd);
1018 if (result == -ESTALE) {
1019 /* nd->path had been dropped */
1020 current->total_link_count = 0;
1021 nd->path = save;
1022 path_get(&nd->path);
1023 nd->flags |= LOOKUP_REVAL;
1024 result = link_path_walk(name, nd);
1025 }
1026
1027 path_put(&save);
1028
1029 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030}
1031
Al Viro9b4a9b12009-04-07 11:44:16 -04001032static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001034 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001035 int fput_needed;
1036 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037
1038 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1039 nd->flags = flags;
1040 nd->depth = 0;
Al Viro2a737872009-04-07 11:49:53 -04001041 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 if (*name=='/') {
Al Viro2a737872009-04-07 11:49:53 -04001044 set_root(nd);
1045 nd->path = nd->root;
1046 path_get(&nd->root);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001047 } else if (dfd == AT_FDCWD) {
Al Viro2a737872009-04-07 11:49:53 -04001048 struct fs_struct *fs = current->fs;
Andreas Mohre518ddb2006-09-29 02:01:22 -07001049 read_lock(&fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08001050 nd->path = fs->pwd;
1051 path_get(&fs->pwd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001052 read_unlock(&fs->lock);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001053 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001054 struct dentry *dentry;
1055
1056 file = fget_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001057 retval = -EBADF;
1058 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001059 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001060
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001061 dentry = file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001062
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001063 retval = -ENOTDIR;
1064 if (!S_ISDIR(dentry->d_inode->i_mode))
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001065 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001066
1067 retval = file_permission(file, MAY_EXEC);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001068 if (retval)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001069 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001070
Jan Blunck5dd784d2008-02-14 19:34:38 -08001071 nd->path = file->f_path;
1072 path_get(&file->f_path);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001073
1074 fput_light(file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001076 return 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001077
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001078fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001079 fput_light(file, fput_needed);
Al Viro9b4a9b12009-04-07 11:44:16 -04001080out_fail:
1081 return retval;
1082}
1083
1084/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1085static int do_path_lookup(int dfd, const char *name,
1086 unsigned int flags, struct nameidata *nd)
1087{
1088 int retval = path_init(dfd, name, flags, nd);
1089 if (!retval)
1090 retval = path_walk(name, nd);
1091 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1092 nd->path.dentry->d_inode))
1093 audit_inode(name, nd->path.dentry);
Al Viro2a737872009-04-07 11:49:53 -04001094 if (nd->root.mnt) {
1095 path_put(&nd->root);
1096 nd->root.mnt = NULL;
1097 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001098 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099}
1100
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -08001101int path_lookup(const char *name, unsigned int flags,
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001102 struct nameidata *nd)
1103{
1104 return do_path_lookup(AT_FDCWD, name, flags, nd);
1105}
1106
Al Virod1811462008-08-02 00:49:18 -04001107int kern_path(const char *name, unsigned int flags, struct path *path)
1108{
1109 struct nameidata nd;
1110 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1111 if (!res)
1112 *path = nd.path;
1113 return res;
1114}
1115
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001116/**
1117 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1118 * @dentry: pointer to dentry of the base directory
1119 * @mnt: pointer to vfs mount of the base directory
1120 * @name: pointer to file name
1121 * @flags: lookup flags
1122 * @nd: pointer to nameidata
1123 */
1124int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1125 const char *name, unsigned int flags,
1126 struct nameidata *nd)
1127{
1128 int retval;
1129
1130 /* same as do_path_lookup */
1131 nd->last_type = LAST_ROOT;
1132 nd->flags = flags;
1133 nd->depth = 0;
1134
Jan Blunckc8e7f442008-06-09 16:40:35 -07001135 nd->path.dentry = dentry;
1136 nd->path.mnt = mnt;
1137 path_get(&nd->path);
Al Viro5b857112009-04-07 11:53:49 -04001138 nd->root = nd->path;
1139 path_get(&nd->root);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001140
1141 retval = path_walk(name, nd);
Jan Blunck4ac91372008-02-14 19:34:32 -08001142 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1143 nd->path.dentry->d_inode))
1144 audit_inode(name, nd->path.dentry);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001145
Al Viro5b857112009-04-07 11:53:49 -04001146 path_put(&nd->root);
1147 nd->root.mnt = NULL;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001148
Al Viro2a737872009-04-07 11:49:53 -04001149 return retval;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001150}
1151
Christoph Hellwigeead1912007-10-16 23:25:38 -07001152static struct dentry *__lookup_hash(struct qstr *name,
1153 struct dentry *base, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154{
James Morris057f6c02007-04-26 00:12:05 -07001155 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 struct inode *inode;
1157 int err;
1158
1159 inode = base->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160
1161 /*
1162 * See if the low-level filesystem might want
1163 * to use its own hash..
1164 */
1165 if (base->d_op && base->d_op->d_hash) {
1166 err = base->d_op->d_hash(base, name);
1167 dentry = ERR_PTR(err);
1168 if (err < 0)
1169 goto out;
1170 }
1171
Al Viro6e6b1bd2009-08-13 23:38:37 +04001172 dentry = __d_lookup(base, name);
1173
1174 /* lockess __d_lookup may fail due to concurrent d_move()
1175 * in some unrelated directory, so try with d_lookup
1176 */
1177 if (!dentry)
1178 dentry = d_lookup(base, name);
1179
1180 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
1181 dentry = do_revalidate(dentry, nd);
1182
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 if (!dentry) {
Miklos Szeredid70b67c2008-07-02 21:30:15 +02001184 struct dentry *new;
1185
1186 /* Don't create child dentry for a dead directory. */
1187 dentry = ERR_PTR(-ENOENT);
1188 if (IS_DEADDIR(inode))
1189 goto out;
1190
1191 new = d_alloc(base, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 dentry = ERR_PTR(-ENOMEM);
1193 if (!new)
1194 goto out;
1195 dentry = inode->i_op->lookup(inode, new, nd);
1196 if (!dentry)
1197 dentry = new;
1198 else
1199 dput(new);
1200 }
1201out:
1202 return dentry;
1203}
1204
James Morris057f6c02007-04-26 00:12:05 -07001205/*
1206 * Restricted form of lookup. Doesn't follow links, single-component only,
1207 * needs parent already locked. Doesn't follow mounts.
1208 * SMP-safe.
1209 */
Adrian Bunka244e162006-03-31 02:32:11 -08001210static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211{
Christoph Hellwigeead1912007-10-16 23:25:38 -07001212 int err;
1213
Al Virob75b5082009-12-16 01:01:38 -05001214 err = exec_permission(nd->path.dentry->d_inode);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001215 if (err)
1216 return ERR_PTR(err);
Jan Blunck4ac91372008-02-14 19:34:32 -08001217 return __lookup_hash(&nd->last, nd->path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218}
1219
Christoph Hellwigeead1912007-10-16 23:25:38 -07001220static int __lookup_one_len(const char *name, struct qstr *this,
1221 struct dentry *base, int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222{
1223 unsigned long hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 unsigned int c;
1225
James Morris057f6c02007-04-26 00:12:05 -07001226 this->name = name;
1227 this->len = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 if (!len)
James Morris057f6c02007-04-26 00:12:05 -07001229 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230
1231 hash = init_name_hash();
1232 while (len--) {
1233 c = *(const unsigned char *)name++;
1234 if (c == '/' || c == '\0')
James Morris057f6c02007-04-26 00:12:05 -07001235 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 hash = partial_name_hash(c, hash);
1237 }
James Morris057f6c02007-04-26 00:12:05 -07001238 this->hash = end_name_hash(hash);
1239 return 0;
1240}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241
Christoph Hellwigeead1912007-10-16 23:25:38 -07001242/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001243 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07001244 * @name: pathname component to lookup
1245 * @base: base directory to lookup from
1246 * @len: maximum length @len should be interpreted to
1247 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07001248 * Note that this routine is purely a helper for filesystem usage and should
1249 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07001250 * nameidata argument is passed to the filesystem methods and a filesystem
1251 * using this helper needs to be prepared for that.
1252 */
James Morris057f6c02007-04-26 00:12:05 -07001253struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1254{
1255 int err;
1256 struct qstr this;
1257
David Woodhouse2f9092e2009-04-20 23:18:37 +01001258 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1259
James Morris057f6c02007-04-26 00:12:05 -07001260 err = __lookup_one_len(name, &this, base, len);
1261 if (err)
1262 return ERR_PTR(err);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001263
Al Virob75b5082009-12-16 01:01:38 -05001264 err = exec_permission(base->d_inode);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001265 if (err)
1266 return ERR_PTR(err);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001267 return __lookup_hash(&this, base, NULL);
James Morris057f6c02007-04-26 00:12:05 -07001268}
1269
Al Viro2d8f3032008-07-22 09:59:21 -04001270int user_path_at(int dfd, const char __user *name, unsigned flags,
1271 struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272{
Al Viro2d8f3032008-07-22 09:59:21 -04001273 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 char *tmp = getname(name);
1275 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04001277
1278 BUG_ON(flags & LOOKUP_PARENT);
1279
1280 err = do_path_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04001282 if (!err)
1283 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 }
1285 return err;
1286}
1287
Al Viro2ad94ae2008-07-21 09:32:51 -04001288static int user_path_parent(int dfd, const char __user *path,
1289 struct nameidata *nd, char **name)
1290{
1291 char *s = getname(path);
1292 int error;
1293
1294 if (IS_ERR(s))
1295 return PTR_ERR(s);
1296
1297 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1298 if (error)
1299 putname(s);
1300 else
1301 *name = s;
1302
1303 return error;
1304}
1305
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306/*
1307 * It's inline, so penalty for filesystems that don't use sticky bit is
1308 * minimal.
1309 */
1310static inline int check_sticky(struct inode *dir, struct inode *inode)
1311{
David Howellsda9592e2008-11-14 10:39:05 +11001312 uid_t fsuid = current_fsuid();
1313
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 if (!(dir->i_mode & S_ISVTX))
1315 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001316 if (inode->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001318 if (dir->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 return 0;
1320 return !capable(CAP_FOWNER);
1321}
1322
1323/*
1324 * Check whether we can remove a link victim from directory dir, check
1325 * whether the type of victim is right.
1326 * 1. We can't do it if dir is read-only (done in permission())
1327 * 2. We should have write and exec permissions on dir
1328 * 3. We can't remove anything from append-only dir
1329 * 4. We can't do anything with immutable dir (done in permission())
1330 * 5. If the sticky bit on dir is set we should either
1331 * a. be owner of dir, or
1332 * b. be owner of victim, or
1333 * c. have CAP_FOWNER capability
1334 * 6. If the victim is append-only or immutable we can't do antyhing with
1335 * links pointing to it.
1336 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1337 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1338 * 9. We can't remove a root or mountpoint.
1339 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1340 * nfs_async_unlink().
1341 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001342static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343{
1344 int error;
1345
1346 if (!victim->d_inode)
1347 return -ENOENT;
1348
1349 BUG_ON(victim->d_parent->d_inode != dir);
Al Viro5a190ae2007-06-07 12:19:32 -04001350 audit_inode_child(victim->d_name.name, victim, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351
Al Virof419a2e2008-07-22 00:07:17 -04001352 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 if (error)
1354 return error;
1355 if (IS_APPEND(dir))
1356 return -EPERM;
1357 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
Hugh Dickinsf9454542008-11-19 15:36:38 -08001358 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 return -EPERM;
1360 if (isdir) {
1361 if (!S_ISDIR(victim->d_inode->i_mode))
1362 return -ENOTDIR;
1363 if (IS_ROOT(victim))
1364 return -EBUSY;
1365 } else if (S_ISDIR(victim->d_inode->i_mode))
1366 return -EISDIR;
1367 if (IS_DEADDIR(dir))
1368 return -ENOENT;
1369 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1370 return -EBUSY;
1371 return 0;
1372}
1373
1374/* Check whether we can create an object with dentry child in directory
1375 * dir.
1376 * 1. We can't do it if child already exists (open has special treatment for
1377 * this case, but since we are inlined it's OK)
1378 * 2. We can't do it if dir is read-only (done in permission())
1379 * 3. We should have write and exec permissions on dir
1380 * 4. We can't do it if dir is immutable (done in permission())
1381 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02001382static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383{
1384 if (child->d_inode)
1385 return -EEXIST;
1386 if (IS_DEADDIR(dir))
1387 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04001388 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389}
1390
1391/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 * O_DIRECTORY translates into forcing a directory lookup.
1393 */
1394static inline int lookup_flags(unsigned int f)
1395{
1396 unsigned long retval = LOOKUP_FOLLOW;
1397
1398 if (f & O_NOFOLLOW)
1399 retval &= ~LOOKUP_FOLLOW;
1400
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 if (f & O_DIRECTORY)
1402 retval |= LOOKUP_DIRECTORY;
1403
1404 return retval;
1405}
1406
1407/*
1408 * p1 and p2 should be directories on the same fs.
1409 */
1410struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1411{
1412 struct dentry *p;
1413
1414 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001415 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 return NULL;
1417 }
1418
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001419 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001421 p = d_ancestor(p2, p1);
1422 if (p) {
1423 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1424 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1425 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 }
1427
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001428 p = d_ancestor(p1, p2);
1429 if (p) {
1430 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1431 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1432 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 }
1434
Ingo Molnarf2eace22006-07-03 00:25:05 -07001435 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1436 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 return NULL;
1438}
1439
1440void unlock_rename(struct dentry *p1, struct dentry *p2)
1441{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001442 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001444 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001445 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 }
1447}
1448
1449int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1450 struct nameidata *nd)
1451{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001452 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453
1454 if (error)
1455 return error;
1456
Al Viroacfa4382008-12-04 10:06:33 -05001457 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 return -EACCES; /* shouldn't it be ENOSYS? */
1459 mode &= S_IALLUGO;
1460 mode |= S_IFREG;
1461 error = security_inode_create(dir, dentry, mode);
1462 if (error)
1463 return error;
Jan Kara9e3509e2009-01-26 16:45:12 +01001464 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001466 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001467 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 return error;
1469}
1470
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001471int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001473 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 struct inode *inode = dentry->d_inode;
1475 int error;
1476
1477 if (!inode)
1478 return -ENOENT;
1479
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001480 switch (inode->i_mode & S_IFMT) {
1481 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001483 case S_IFDIR:
1484 if (acc_mode & MAY_WRITE)
1485 return -EISDIR;
1486 break;
1487 case S_IFBLK:
1488 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001489 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001491 /*FALLTHRU*/
1492 case S_IFIFO:
1493 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001495 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001496 }
Dave Hansenb41572e2007-10-16 23:31:14 -07001497
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001498 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07001499 if (error)
1500 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05001501
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 /*
1503 * An append-only file must be opened in append mode for writing.
1504 */
1505 if (IS_APPEND(inode)) {
1506 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05001507 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05001509 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 }
1511
1512 /* O_NOATIME can only be set by the owner or superuser */
Al Viro7715b522009-12-16 03:54:00 -05001513 if (flag & O_NOATIME && !is_owner_or_cap(inode))
1514 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515
1516 /*
1517 * Ensure there are no outstanding leases on the file.
1518 */
Al Virob65a9cf2009-12-16 06:27:40 -05001519 return break_lease(inode, flag);
Al Viro7715b522009-12-16 03:54:00 -05001520}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521
Al Viro7715b522009-12-16 03:54:00 -05001522static int handle_truncate(struct path *path)
1523{
1524 struct inode *inode = path->dentry->d_inode;
1525 int error = get_write_access(inode);
1526 if (error)
1527 return error;
1528 /*
1529 * Refuse to truncate files with mandatory locks held on them.
1530 */
1531 error = locks_verify_locked(inode);
1532 if (!error)
1533 error = security_path_truncate(path, 0,
1534 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN);
1535 if (!error) {
1536 error = do_truncate(path->dentry, 0,
1537 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1538 NULL);
1539 }
1540 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04001541 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542}
1543
Dave Hansend57999e2008-02-15 14:37:27 -08001544/*
1545 * Be careful about ever adding any more callers of this
1546 * function. Its flags must be in the namei format, not
1547 * what get passed to sys_open().
1548 */
1549static int __open_namei_create(struct nameidata *nd, struct path *path,
Dave Hansenaab520e2006-09-30 23:29:02 -07001550 int flag, int mode)
1551{
1552 int error;
Jan Blunck4ac91372008-02-14 19:34:32 -08001553 struct dentry *dir = nd->path.dentry;
Dave Hansenaab520e2006-09-30 23:29:02 -07001554
1555 if (!IS_POSIXACL(dir->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04001556 mode &= ~current_umask();
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001557 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1558 if (error)
1559 goto out_unlock;
Dave Hansenaab520e2006-09-30 23:29:02 -07001560 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001561out_unlock:
Dave Hansenaab520e2006-09-30 23:29:02 -07001562 mutex_unlock(&dir->d_inode->i_mutex);
Jan Blunck4ac91372008-02-14 19:34:32 -08001563 dput(nd->path.dentry);
1564 nd->path.dentry = path->dentry;
Dave Hansenaab520e2006-09-30 23:29:02 -07001565 if (error)
1566 return error;
1567 /* Don't check for write permission, don't truncate */
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001568 return may_open(&nd->path, 0, flag & ~O_TRUNC);
Dave Hansenaab520e2006-09-30 23:29:02 -07001569}
1570
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571/*
Dave Hansend57999e2008-02-15 14:37:27 -08001572 * Note that while the flag value (low two bits) for sys_open means:
1573 * 00 - read-only
1574 * 01 - write-only
1575 * 10 - read-write
1576 * 11 - special
1577 * it is changed into
1578 * 00 - no permissions needed
1579 * 01 - read-permission
1580 * 10 - write-permission
1581 * 11 - read-write
1582 * for the internal routines (ie open_namei()/follow_link() etc)
1583 * This is more logical, and also allows the 00 "no perm needed"
1584 * to be used for symlinks (where the permissions are checked
1585 * later).
1586 *
1587*/
1588static inline int open_to_namei_flags(int flag)
1589{
1590 if ((flag+1) & O_ACCMODE)
1591 flag++;
1592 return flag;
1593}
1594
Al Viro7715b522009-12-16 03:54:00 -05001595static int open_will_truncate(int flag, struct inode *inode)
Dave Hansen4a3fd212008-02-15 14:37:48 -08001596{
1597 /*
1598 * We'll never write to the fs underlying
1599 * a device file.
1600 */
1601 if (special_file(inode->i_mode))
1602 return 0;
1603 return (flag & O_TRUNC);
1604}
1605
Dave Hansend57999e2008-02-15 14:37:27 -08001606/*
Dave Hansen4a3fd212008-02-15 14:37:48 -08001607 * Note that the low bits of the passed in "open_flag"
1608 * are not the same as in the local variable "flag". See
1609 * open_to_namei_flags() for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001611struct file *do_filp_open(int dfd, const char *pathname,
Al Viro6e8341a2009-04-06 11:16:22 -04001612 int open_flag, int mode, int acc_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613{
Dave Hansen4a3fd212008-02-15 14:37:48 -08001614 struct file *filp;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001615 struct nameidata nd;
Al Viro6e8341a2009-04-06 11:16:22 -04001616 int error;
Al Viro9850c052010-01-13 15:01:15 -05001617 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 struct dentry *dir;
1619 int count = 0;
Al Viro7715b522009-12-16 03:54:00 -05001620 int will_truncate;
Dave Hansend57999e2008-02-15 14:37:27 -08001621 int flag = open_to_namei_flags(open_flag);
Al Viro9850c052010-01-13 15:01:15 -05001622 int force_reval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623
Christoph Hellwig6b2f3d12009-10-27 11:05:28 +01001624 /*
1625 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1626 * check for O_DSYNC if the need any syncing at all we enforce it's
1627 * always set instead of having to deal with possibly weird behaviour
1628 * for malicious applications setting only __O_SYNC.
1629 */
1630 if (open_flag & __O_SYNC)
1631 open_flag |= O_DSYNC;
1632
Al Viro6e8341a2009-04-06 11:16:22 -04001633 if (!acc_mode)
Al Viro6d125522009-12-24 06:58:56 -05001634 acc_mode = MAY_OPEN | ACC_MODE(open_flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635
Trond Myklebust834f2a42005-10-18 14:20:16 -07001636 /* O_TRUNC implies we need access checks for write permissions */
1637 if (flag & O_TRUNC)
1638 acc_mode |= MAY_WRITE;
1639
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 /* Allow the LSM permission hook to distinguish append
1641 access from general write access. */
1642 if (flag & O_APPEND)
1643 acc_mode |= MAY_APPEND;
1644
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 /*
1646 * The simplest case - just a plain lookup.
1647 */
1648 if (!(flag & O_CREAT)) {
Al Viro2dd6d1f2009-08-13 20:40:45 +04001649 filp = get_empty_filp();
1650
1651 if (filp == NULL)
1652 return ERR_PTR(-ENFILE);
1653 nd.intent.open.file = filp;
Al Viro482928d2009-12-19 10:10:39 -05001654 filp->f_flags = open_flag;
Al Viro2dd6d1f2009-08-13 20:40:45 +04001655 nd.intent.open.flags = flag;
1656 nd.intent.open.create_mode = 0;
1657 error = do_path_lookup(dfd, pathname,
1658 lookup_flags(flag)|LOOKUP_OPEN, &nd);
1659 if (IS_ERR(nd.intent.open.file)) {
1660 if (error == 0) {
1661 error = PTR_ERR(nd.intent.open.file);
1662 path_put(&nd.path);
1663 }
1664 } else if (error)
1665 release_open_intent(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 if (error)
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001667 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 goto ok;
1669 }
1670
1671 /*
1672 * Create - we need to know the parent.
1673 */
Al Viro9850c052010-01-13 15:01:15 -05001674reval:
Al Viro9b4a9b12009-04-07 11:44:16 -04001675 error = path_init(dfd, pathname, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 if (error)
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001677 return ERR_PTR(error);
Al Viro9850c052010-01-13 15:01:15 -05001678 if (force_reval)
1679 nd.flags |= LOOKUP_REVAL;
Al Viro9b4a9b12009-04-07 11:44:16 -04001680 error = path_walk(pathname, &nd);
J. R. Okajima654f5622009-06-18 23:30:15 +09001681 if (error) {
1682 if (nd.root.mnt)
1683 path_put(&nd.root);
Al Viro9b4a9b12009-04-07 11:44:16 -04001684 return ERR_PTR(error);
J. R. Okajima654f5622009-06-18 23:30:15 +09001685 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001686 if (unlikely(!audit_dummy_context()))
1687 audit_inode(pathname, nd.path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688
1689 /*
1690 * We have the parent and last component. First of all, check
1691 * that we are not asked to creat(2) an obvious directory - that
1692 * will not do.
1693 */
1694 error = -EISDIR;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001695 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
Al Viro8737f3a2008-08-02 22:36:57 -04001696 goto exit_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697
Al Viro8737f3a2008-08-02 22:36:57 -04001698 error = -ENFILE;
1699 filp = get_empty_filp();
1700 if (filp == NULL)
1701 goto exit_parent;
1702 nd.intent.open.file = filp;
Al Viro482928d2009-12-19 10:10:39 -05001703 filp->f_flags = open_flag;
Al Viro8737f3a2008-08-02 22:36:57 -04001704 nd.intent.open.flags = flag;
1705 nd.intent.open.create_mode = mode;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001706 dir = nd.path.dentry;
1707 nd.flags &= ~LOOKUP_PARENT;
Al Viro8737f3a2008-08-02 22:36:57 -04001708 nd.flags |= LOOKUP_CREATE | LOOKUP_OPEN;
Al Viro35165862008-08-05 03:00:49 -04001709 if (flag & O_EXCL)
1710 nd.flags |= LOOKUP_EXCL;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001711 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001712 path.dentry = lookup_hash(&nd);
1713 path.mnt = nd.path.mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714
1715do_last:
Al Viro4e7506e2005-06-06 13:36:00 -07001716 error = PTR_ERR(path.dentry);
1717 if (IS_ERR(path.dentry)) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001718 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 goto exit;
1720 }
1721
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001722 if (IS_ERR(nd.intent.open.file)) {
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001723 error = PTR_ERR(nd.intent.open.file);
Dave Hansen4a3fd212008-02-15 14:37:48 -08001724 goto exit_mutex_unlock;
Oleg Drokin4af4c522006-03-25 03:06:54 -08001725 }
1726
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 /* Negative dentry, just create the file */
Al Viro4e7506e2005-06-06 13:36:00 -07001728 if (!path.dentry->d_inode) {
Dave Hansen4a3fd212008-02-15 14:37:48 -08001729 /*
1730 * This write is needed to ensure that a
1731 * ro->rw transition does not occur between
1732 * the time when the file is created and when
1733 * a permanent write count is taken through
1734 * the 'struct file' in nameidata_to_filp().
1735 */
1736 error = mnt_want_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 if (error)
Dave Hansen4a3fd212008-02-15 14:37:48 -08001738 goto exit_mutex_unlock;
1739 error = __open_namei_create(&nd, &path, flag, mode);
1740 if (error) {
1741 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 goto exit;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001743 }
Al Viro482928d2009-12-19 10:10:39 -05001744 filp = nameidata_to_filp(&nd);
Dave Hansen4a3fd212008-02-15 14:37:48 -08001745 mnt_drop_write(nd.path.mnt);
J. R. Okajima654f5622009-06-18 23:30:15 +09001746 if (nd.root.mnt)
1747 path_put(&nd.root);
Al Virob65a9cf2009-12-16 06:27:40 -05001748 if (!IS_ERR(filp)) {
Mimi Zohar9bbb6ca2010-01-26 17:02:40 -05001749 error = ima_file_check(filp, acc_mode);
Al Virob65a9cf2009-12-16 06:27:40 -05001750 if (error) {
1751 fput(filp);
1752 filp = ERR_PTR(error);
1753 }
1754 }
Dave Hansen4a3fd212008-02-15 14:37:48 -08001755 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 }
1757
1758 /*
1759 * It already exists.
1760 */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001761 mutex_unlock(&dir->d_inode->i_mutex);
Al Viro5a190ae2007-06-07 12:19:32 -04001762 audit_inode(pathname, path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763
1764 error = -EEXIST;
1765 if (flag & O_EXCL)
1766 goto exit_dput;
1767
Al Viroe13b2102005-06-06 13:36:06 -07001768 if (__follow_mount(&path)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769 error = -ELOOP;
Al Viroba7a4c12005-06-06 13:36:08 -07001770 if (flag & O_NOFOLLOW)
1771 goto exit_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 }
Amy Griffis3e2efce2006-07-13 13:16:02 -04001773
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 error = -ENOENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001775 if (!path.dentry->d_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 goto exit_dput;
Al Viroacfa4382008-12-04 10:06:33 -05001777 if (path.dentry->d_inode->i_op->follow_link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 goto do_link;
1779
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001780 path_to_nameidata(&path, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781 error = -EISDIR;
Jeff Laytoncb598612009-11-16 12:05:20 -08001782 if (S_ISDIR(path.dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 goto exit;
1784ok:
Dave Hansen4a3fd212008-02-15 14:37:48 -08001785 /*
1786 * Consider:
1787 * 1. may_open() truncates a file
1788 * 2. a rw->ro mount transition occurs
1789 * 3. nameidata_to_filp() fails due to
1790 * the ro mount.
1791 * That would be inconsistent, and should
1792 * be avoided. Taking this mnt write here
1793 * ensures that (2) can not occur.
1794 */
Al Viro7715b522009-12-16 03:54:00 -05001795 will_truncate = open_will_truncate(flag, nd.path.dentry->d_inode);
1796 if (will_truncate) {
Dave Hansen4a3fd212008-02-15 14:37:48 -08001797 error = mnt_want_write(nd.path.mnt);
1798 if (error)
1799 goto exit;
1800 }
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001801 error = may_open(&nd.path, acc_mode, flag);
Dave Hansen4a3fd212008-02-15 14:37:48 -08001802 if (error) {
Al Viro7715b522009-12-16 03:54:00 -05001803 if (will_truncate)
Dave Hansen4a3fd212008-02-15 14:37:48 -08001804 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 goto exit;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001806 }
Al Viro482928d2009-12-19 10:10:39 -05001807 filp = nameidata_to_filp(&nd);
Al Virob65a9cf2009-12-16 06:27:40 -05001808 if (!IS_ERR(filp)) {
Mimi Zohar9bbb6ca2010-01-26 17:02:40 -05001809 error = ima_file_check(filp, acc_mode);
Al Viro7715b522009-12-16 03:54:00 -05001810 if (error) {
Al Viro7715b522009-12-16 03:54:00 -05001811 fput(filp);
Al Virob65a9cf2009-12-16 06:27:40 -05001812 filp = ERR_PTR(error);
1813 }
1814 }
1815 if (!IS_ERR(filp)) {
1816 if (acc_mode & MAY_WRITE)
1817 vfs_dq_init(nd.path.dentry->d_inode);
1818
1819 if (will_truncate) {
1820 error = handle_truncate(&nd.path);
1821 if (error) {
1822 fput(filp);
1823 filp = ERR_PTR(error);
1824 }
Al Viro7715b522009-12-16 03:54:00 -05001825 }
1826 }
Dave Hansen4a3fd212008-02-15 14:37:48 -08001827 /*
1828 * It is now safe to drop the mnt write
1829 * because the filp has had a write taken
1830 * on its behalf.
1831 */
Al Viro7715b522009-12-16 03:54:00 -05001832 if (will_truncate)
Dave Hansen4a3fd212008-02-15 14:37:48 -08001833 mnt_drop_write(nd.path.mnt);
J. R. Okajima654f5622009-06-18 23:30:15 +09001834 if (nd.root.mnt)
1835 path_put(&nd.root);
Dave Hansen4a3fd212008-02-15 14:37:48 -08001836 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837
Dave Hansen4a3fd212008-02-15 14:37:48 -08001838exit_mutex_unlock:
1839 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840exit_dput:
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001841 path_put_conditional(&path, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842exit:
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001843 if (!IS_ERR(nd.intent.open.file))
1844 release_open_intent(&nd);
Al Viro8737f3a2008-08-02 22:36:57 -04001845exit_parent:
Al Viro2a737872009-04-07 11:49:53 -04001846 if (nd.root.mnt)
1847 path_put(&nd.root);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001848 path_put(&nd.path);
1849 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850
1851do_link:
1852 error = -ELOOP;
1853 if (flag & O_NOFOLLOW)
1854 goto exit_dput;
1855 /*
1856 * This is subtle. Instead of calling do_follow_link() we do the
1857 * thing by hands. The reason is that this way we have zero link_count
1858 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1859 * After that we have the parent and last component, i.e.
1860 * we are in the same situation as after the first path_walk().
1861 * Well, almost - if the last component is normal we get its copy
1862 * stored in nd->last.name and we will have to putname() it when we
1863 * are done. Procfs-like symlinks just set LAST_BIND.
1864 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001865 nd.flags |= LOOKUP_PARENT;
1866 error = security_inode_follow_link(path.dentry, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867 if (error)
1868 goto exit_dput;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001869 error = __do_follow_link(&path, &nd);
Al Viro258fa992009-08-09 01:32:02 +04001870 path_put(&path);
Kirill Korotaevde459212006-07-14 00:23:49 -07001871 if (error) {
1872 /* Does someone understand code flow here? Or it is only
1873 * me so stupid? Anathema to whoever designed this non-sense
1874 * with "intent.open".
1875 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001876 release_open_intent(&nd);
J. R. Okajima654f5622009-06-18 23:30:15 +09001877 if (nd.root.mnt)
1878 path_put(&nd.root);
Al Viro9850c052010-01-13 15:01:15 -05001879 if (error == -ESTALE && !force_reval) {
1880 force_reval = 1;
1881 goto reval;
1882 }
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001883 return ERR_PTR(error);
Kirill Korotaevde459212006-07-14 00:23:49 -07001884 }
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001885 nd.flags &= ~LOOKUP_PARENT;
1886 if (nd.last_type == LAST_BIND)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 goto ok;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888 error = -EISDIR;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001889 if (nd.last_type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 goto exit;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001891 if (nd.last.name[nd.last.len]) {
1892 __putname(nd.last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893 goto exit;
1894 }
1895 error = -ELOOP;
1896 if (count++==32) {
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001897 __putname(nd.last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898 goto exit;
1899 }
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001900 dir = nd.path.dentry;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001901 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001902 path.dentry = lookup_hash(&nd);
1903 path.mnt = nd.path.mnt;
1904 __putname(nd.last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905 goto do_last;
1906}
1907
1908/**
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001909 * filp_open - open file and return file pointer
1910 *
1911 * @filename: path to open
1912 * @flags: open flags as per the open(2) second argument
1913 * @mode: mode for the new file if O_CREAT is set, else ignored
1914 *
1915 * This is the helper to open a file from kernelspace if you really
1916 * have to. But in generally you should not do this, so please move
1917 * along, nothing to see here..
1918 */
1919struct file *filp_open(const char *filename, int flags, int mode)
1920{
Al Viro6e8341a2009-04-06 11:16:22 -04001921 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001922}
1923EXPORT_SYMBOL(filp_open);
1924
1925/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 * lookup_create - lookup a dentry, creating it if it doesn't exist
1927 * @nd: nameidata info
1928 * @is_dir: directory flag
1929 *
1930 * Simple function to lookup and return a dentry and create it
1931 * if it doesn't exist. Is SMP-safe.
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001932 *
Jan Blunck4ac91372008-02-14 19:34:32 -08001933 * Returns with nd->path.dentry->d_inode->i_mutex locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 */
1935struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1936{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001937 struct dentry *dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938
Jan Blunck4ac91372008-02-14 19:34:32 -08001939 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001940 /*
1941 * Yucky last component or no last component at all?
1942 * (foo/., foo/.., /////)
1943 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 if (nd->last_type != LAST_NORM)
1945 goto fail;
1946 nd->flags &= ~LOOKUP_PARENT;
Al Viro35165862008-08-05 03:00:49 -04001947 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
ASANO Masahiroa6349042006-08-22 20:06:02 -04001948 nd->intent.open.flags = O_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001949
1950 /*
1951 * Do the final lookup.
1952 */
Christoph Hellwig49705b72005-11-08 21:35:06 -08001953 dentry = lookup_hash(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 if (IS_ERR(dentry))
1955 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001956
Al Viroe9baf6e2008-05-15 04:49:12 -04001957 if (dentry->d_inode)
1958 goto eexist;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001959 /*
1960 * Special case - lookup gave negative, but... we had foo/bar/
1961 * From the vfs_mknod() POV we just have a negative dentry -
1962 * all is fine. Let's be bastards - you had / on the end, you've
1963 * been asking for (non-existent) directory. -ENOENT for you.
1964 */
Al Viroe9baf6e2008-05-15 04:49:12 -04001965 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1966 dput(dentry);
1967 dentry = ERR_PTR(-ENOENT);
1968 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969 return dentry;
Al Viroe9baf6e2008-05-15 04:49:12 -04001970eexist:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 dput(dentry);
Al Viroe9baf6e2008-05-15 04:49:12 -04001972 dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973fail:
1974 return dentry;
1975}
Christoph Hellwigf81a0bf2005-05-19 12:26:43 -07001976EXPORT_SYMBOL_GPL(lookup_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977
1978int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1979{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001980 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981
1982 if (error)
1983 return error;
1984
1985 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1986 return -EPERM;
1987
Al Viroacfa4382008-12-04 10:06:33 -05001988 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989 return -EPERM;
1990
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07001991 error = devcgroup_inode_mknod(mode, dev);
1992 if (error)
1993 return error;
1994
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 error = security_inode_mknod(dir, dentry, mode, dev);
1996 if (error)
1997 return error;
1998
Jan Kara9e3509e2009-01-26 16:45:12 +01001999 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002001 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002002 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003 return error;
2004}
2005
Dave Hansen463c3192008-02-15 14:37:57 -08002006static int may_mknod(mode_t mode)
2007{
2008 switch (mode & S_IFMT) {
2009 case S_IFREG:
2010 case S_IFCHR:
2011 case S_IFBLK:
2012 case S_IFIFO:
2013 case S_IFSOCK:
2014 case 0: /* zero mode translates to S_IFREG */
2015 return 0;
2016 case S_IFDIR:
2017 return -EPERM;
2018 default:
2019 return -EINVAL;
2020 }
2021}
2022
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002023SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2024 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025{
Al Viro2ad94ae2008-07-21 09:32:51 -04002026 int error;
2027 char *tmp;
2028 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029 struct nameidata nd;
2030
2031 if (S_ISDIR(mode))
2032 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033
Al Viro2ad94ae2008-07-21 09:32:51 -04002034 error = user_path_parent(dfd, filename, &nd, &tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002036 return error;
2037
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 dentry = lookup_create(&nd, 0);
Dave Hansen463c3192008-02-15 14:37:57 -08002039 if (IS_ERR(dentry)) {
2040 error = PTR_ERR(dentry);
2041 goto out_unlock;
2042 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002043 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002044 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002045 error = may_mknod(mode);
2046 if (error)
2047 goto out_dput;
2048 error = mnt_want_write(nd.path.mnt);
2049 if (error)
2050 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002051 error = security_path_mknod(&nd.path, dentry, mode, dev);
2052 if (error)
2053 goto out_drop_write;
Dave Hansen463c3192008-02-15 14:37:57 -08002054 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 case 0: case S_IFREG:
Jan Blunck4ac91372008-02-14 19:34:32 -08002056 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057 break;
2058 case S_IFCHR: case S_IFBLK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002059 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060 new_decode_dev(dev));
2061 break;
2062 case S_IFIFO: case S_IFSOCK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002063 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002066out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002067 mnt_drop_write(nd.path.mnt);
2068out_dput:
2069 dput(dentry);
2070out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002071 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002072 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073 putname(tmp);
2074
2075 return error;
2076}
2077
Heiko Carstens3480b252009-01-14 14:14:16 +01002078SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002079{
2080 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2081}
2082
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2084{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002085 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086
2087 if (error)
2088 return error;
2089
Al Viroacfa4382008-12-04 10:06:33 -05002090 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091 return -EPERM;
2092
2093 mode &= (S_IRWXUGO|S_ISVTX);
2094 error = security_inode_mkdir(dir, dentry, mode);
2095 if (error)
2096 return error;
2097
Jan Kara9e3509e2009-01-26 16:45:12 +01002098 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002100 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002101 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102 return error;
2103}
2104
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002105SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106{
2107 int error = 0;
2108 char * tmp;
Dave Hansen6902d922006-09-30 23:29:01 -07002109 struct dentry *dentry;
2110 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111
Al Viro2ad94ae2008-07-21 09:32:51 -04002112 error = user_path_parent(dfd, pathname, &nd, &tmp);
2113 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002114 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115
Dave Hansen6902d922006-09-30 23:29:01 -07002116 dentry = lookup_create(&nd, 1);
2117 error = PTR_ERR(dentry);
2118 if (IS_ERR(dentry))
2119 goto out_unlock;
2120
Jan Blunck4ac91372008-02-14 19:34:32 -08002121 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002122 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002123 error = mnt_want_write(nd.path.mnt);
2124 if (error)
2125 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002126 error = security_path_mkdir(&nd.path, dentry, mode);
2127 if (error)
2128 goto out_drop_write;
Jan Blunck4ac91372008-02-14 19:34:32 -08002129 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002130out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002131 mnt_drop_write(nd.path.mnt);
2132out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002133 dput(dentry);
2134out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002135 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002136 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002137 putname(tmp);
2138out_err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139 return error;
2140}
2141
Heiko Carstens3cdad422009-01-14 14:14:22 +01002142SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002143{
2144 return sys_mkdirat(AT_FDCWD, pathname, mode);
2145}
2146
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147/*
2148 * We try to drop the dentry early: we should have
2149 * a usage count of 2 if we're the only user of this
2150 * dentry, and if that is true (possibly after pruning
2151 * the dcache), then we drop the dentry now.
2152 *
2153 * A low-level filesystem can, if it choses, legally
2154 * do a
2155 *
2156 * if (!d_unhashed(dentry))
2157 * return -EBUSY;
2158 *
2159 * if it cannot handle the case of removing a directory
2160 * that is still in use by something else..
2161 */
2162void dentry_unhash(struct dentry *dentry)
2163{
2164 dget(dentry);
Vasily Averindc168422006-12-06 20:37:07 -08002165 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166 spin_lock(&dcache_lock);
2167 spin_lock(&dentry->d_lock);
2168 if (atomic_read(&dentry->d_count) == 2)
2169 __d_drop(dentry);
2170 spin_unlock(&dentry->d_lock);
2171 spin_unlock(&dcache_lock);
2172}
2173
2174int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2175{
2176 int error = may_delete(dir, dentry, 1);
2177
2178 if (error)
2179 return error;
2180
Al Viroacfa4382008-12-04 10:06:33 -05002181 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 return -EPERM;
2183
Jan Kara9e3509e2009-01-26 16:45:12 +01002184 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002186 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187 dentry_unhash(dentry);
2188 if (d_mountpoint(dentry))
2189 error = -EBUSY;
2190 else {
2191 error = security_inode_rmdir(dir, dentry);
2192 if (!error) {
2193 error = dir->i_op->rmdir(dir, dentry);
2194 if (!error)
2195 dentry->d_inode->i_flags |= S_DEAD;
2196 }
2197 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002198 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 d_delete(dentry);
2201 }
2202 dput(dentry);
2203
2204 return error;
2205}
2206
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002207static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208{
2209 int error = 0;
2210 char * name;
2211 struct dentry *dentry;
2212 struct nameidata nd;
2213
Al Viro2ad94ae2008-07-21 09:32:51 -04002214 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002216 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217
2218 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002219 case LAST_DOTDOT:
2220 error = -ENOTEMPTY;
2221 goto exit1;
2222 case LAST_DOT:
2223 error = -EINVAL;
2224 goto exit1;
2225 case LAST_ROOT:
2226 error = -EBUSY;
2227 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002229
2230 nd.flags &= ~LOOKUP_PARENT;
2231
Jan Blunck4ac91372008-02-14 19:34:32 -08002232 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002233 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002235 if (IS_ERR(dentry))
2236 goto exit2;
Dave Hansen06227532008-02-15 14:37:34 -08002237 error = mnt_want_write(nd.path.mnt);
2238 if (error)
2239 goto exit3;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002240 error = security_path_rmdir(&nd.path, dentry);
2241 if (error)
2242 goto exit4;
Jan Blunck4ac91372008-02-14 19:34:32 -08002243 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002244exit4:
Dave Hansen06227532008-02-15 14:37:34 -08002245 mnt_drop_write(nd.path.mnt);
2246exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07002247 dput(dentry);
2248exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08002249 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002251 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252 putname(name);
2253 return error;
2254}
2255
Heiko Carstens3cdad422009-01-14 14:14:22 +01002256SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002257{
2258 return do_rmdir(AT_FDCWD, pathname);
2259}
2260
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261int vfs_unlink(struct inode *dir, struct dentry *dentry)
2262{
2263 int error = may_delete(dir, dentry, 0);
2264
2265 if (error)
2266 return error;
2267
Al Viroacfa4382008-12-04 10:06:33 -05002268 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269 return -EPERM;
2270
Jan Kara9e3509e2009-01-26 16:45:12 +01002271 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002273 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274 if (d_mountpoint(dentry))
2275 error = -EBUSY;
2276 else {
2277 error = security_inode_unlink(dir, dentry);
2278 if (!error)
2279 error = dir->i_op->unlink(dir, dentry);
2280 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002281 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282
2283 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2284 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08002285 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 }
Robert Love0eeca282005-07-12 17:06:03 -04002288
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289 return error;
2290}
2291
2292/*
2293 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002294 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 * writeout happening, and we don't want to prevent access to the directory
2296 * while waiting on the I/O.
2297 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002298static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299{
Al Viro2ad94ae2008-07-21 09:32:51 -04002300 int error;
2301 char *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302 struct dentry *dentry;
2303 struct nameidata nd;
2304 struct inode *inode = NULL;
2305
Al Viro2ad94ae2008-07-21 09:32:51 -04002306 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002308 return error;
2309
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310 error = -EISDIR;
2311 if (nd.last_type != LAST_NORM)
2312 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002313
2314 nd.flags &= ~LOOKUP_PARENT;
2315
Jan Blunck4ac91372008-02-14 19:34:32 -08002316 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002317 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318 error = PTR_ERR(dentry);
2319 if (!IS_ERR(dentry)) {
2320 /* Why not before? Because we want correct error value */
2321 if (nd.last.name[nd.last.len])
2322 goto slashes;
2323 inode = dentry->d_inode;
2324 if (inode)
2325 atomic_inc(&inode->i_count);
Dave Hansen06227532008-02-15 14:37:34 -08002326 error = mnt_want_write(nd.path.mnt);
2327 if (error)
2328 goto exit2;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002329 error = security_path_unlink(&nd.path, dentry);
2330 if (error)
2331 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08002332 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002333exit3:
Dave Hansen06227532008-02-15 14:37:34 -08002334 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335 exit2:
2336 dput(dentry);
2337 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002338 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339 if (inode)
2340 iput(inode); /* truncate the inode here */
2341exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002342 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 putname(name);
2344 return error;
2345
2346slashes:
2347 error = !dentry->d_inode ? -ENOENT :
2348 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2349 goto exit2;
2350}
2351
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002352SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002353{
2354 if ((flag & ~AT_REMOVEDIR) != 0)
2355 return -EINVAL;
2356
2357 if (flag & AT_REMOVEDIR)
2358 return do_rmdir(dfd, pathname);
2359
2360 return do_unlinkat(dfd, pathname);
2361}
2362
Heiko Carstens3480b252009-01-14 14:14:16 +01002363SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002364{
2365 return do_unlinkat(AT_FDCWD, pathname);
2366}
2367
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002368int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002370 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371
2372 if (error)
2373 return error;
2374
Al Viroacfa4382008-12-04 10:06:33 -05002375 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376 return -EPERM;
2377
2378 error = security_inode_symlink(dir, dentry, oldname);
2379 if (error)
2380 return error;
2381
Jan Kara9e3509e2009-01-26 16:45:12 +01002382 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002384 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002385 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386 return error;
2387}
2388
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002389SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2390 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391{
Al Viro2ad94ae2008-07-21 09:32:51 -04002392 int error;
2393 char *from;
2394 char *to;
Dave Hansen6902d922006-09-30 23:29:01 -07002395 struct dentry *dentry;
2396 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002397
2398 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04002399 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400 return PTR_ERR(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002401
2402 error = user_path_parent(newdfd, newname, &nd, &to);
2403 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002404 goto out_putname;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405
Dave Hansen6902d922006-09-30 23:29:01 -07002406 dentry = lookup_create(&nd, 0);
2407 error = PTR_ERR(dentry);
2408 if (IS_ERR(dentry))
2409 goto out_unlock;
2410
Dave Hansen75c3f292008-02-15 14:37:45 -08002411 error = mnt_want_write(nd.path.mnt);
2412 if (error)
2413 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002414 error = security_path_symlink(&nd.path, dentry, from);
2415 if (error)
2416 goto out_drop_write;
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002417 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002418out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002419 mnt_drop_write(nd.path.mnt);
2420out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002421 dput(dentry);
2422out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002423 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002424 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002425 putname(to);
2426out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 putname(from);
2428 return error;
2429}
2430
Heiko Carstens3480b252009-01-14 14:14:16 +01002431SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002432{
2433 return sys_symlinkat(oldname, AT_FDCWD, newname);
2434}
2435
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2437{
2438 struct inode *inode = old_dentry->d_inode;
2439 int error;
2440
2441 if (!inode)
2442 return -ENOENT;
2443
Miklos Szeredia95164d2008-07-30 15:08:48 +02002444 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445 if (error)
2446 return error;
2447
2448 if (dir->i_sb != inode->i_sb)
2449 return -EXDEV;
2450
2451 /*
2452 * A link to an append-only or immutable file cannot be created.
2453 */
2454 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2455 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05002456 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002458 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459 return -EPERM;
2460
2461 error = security_inode_link(old_dentry, dir, new_dentry);
2462 if (error)
2463 return error;
2464
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002465 mutex_lock(&inode->i_mutex);
Jan Kara9e3509e2009-01-26 16:45:12 +01002466 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467 error = dir->i_op->link(old_dentry, dir, new_dentry);
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002468 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002469 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002470 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471 return error;
2472}
2473
2474/*
2475 * Hardlinks are often used in delicate situations. We avoid
2476 * security-related surprises by not following symlinks on the
2477 * newname. --KAB
2478 *
2479 * We don't follow them on the oldname either to be compatible
2480 * with linux 2.0, and to avoid hard-linking to directories
2481 * and other special files. --ADM
2482 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002483SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2484 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485{
2486 struct dentry *new_dentry;
Al Viro2d8f3032008-07-22 09:59:21 -04002487 struct nameidata nd;
2488 struct path old_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489 int error;
Al Viro2ad94ae2008-07-21 09:32:51 -04002490 char *to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491
Ulrich Drepper45c9b112006-06-25 05:49:11 -07002492 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002493 return -EINVAL;
2494
Al Viro2d8f3032008-07-22 09:59:21 -04002495 error = user_path_at(olddfd, oldname,
2496 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2497 &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002499 return error;
2500
2501 error = user_path_parent(newdfd, newname, &nd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502 if (error)
2503 goto out;
2504 error = -EXDEV;
Al Viro2d8f3032008-07-22 09:59:21 -04002505 if (old_path.mnt != nd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506 goto out_release;
2507 new_dentry = lookup_create(&nd, 0);
2508 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002509 if (IS_ERR(new_dentry))
2510 goto out_unlock;
Dave Hansen75c3f292008-02-15 14:37:45 -08002511 error = mnt_want_write(nd.path.mnt);
2512 if (error)
2513 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002514 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2515 if (error)
2516 goto out_drop_write;
Al Viro2d8f3032008-07-22 09:59:21 -04002517 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002518out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002519 mnt_drop_write(nd.path.mnt);
2520out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002521 dput(new_dentry);
2522out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002523 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524out_release:
Jan Blunck1d957f92008-02-14 19:34:35 -08002525 path_put(&nd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002526 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527out:
Al Viro2d8f3032008-07-22 09:59:21 -04002528 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529
2530 return error;
2531}
2532
Heiko Carstens3480b252009-01-14 14:14:16 +01002533SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002534{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002535 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002536}
2537
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538/*
2539 * The worst of all namespace operations - renaming directory. "Perverted"
2540 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2541 * Problems:
2542 * a) we can get into loop creation. Check is done in is_subdir().
2543 * b) race potential - two innocent renames can create a loop together.
2544 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002545 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546 * story.
2547 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002548 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549 * whether the target exists). Solution: try to be smart with locking
2550 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002551 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552 * move will be locked. Thus we can rank directories by the tree
2553 * (ancestors first) and rank all non-directories after them.
2554 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002555 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556 * HOWEVER, it relies on the assumption that any object with ->lookup()
2557 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2558 * we'd better make sure that there's no link(2) for them.
2559 * d) some filesystems don't support opened-but-unlinked directories,
2560 * either because of layout or because they are not ready to deal with
2561 * all cases correctly. The latter will be fixed (taking this sort of
2562 * stuff into VFS), but the former is not going away. Solution: the same
2563 * trick as in rmdir().
2564 * e) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002565 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002567 * ->i_mutex on parents, which works but leads to some truely excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568 * locking].
2569 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07002570static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2571 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572{
2573 int error = 0;
2574 struct inode *target;
2575
2576 /*
2577 * If we are going to change the parent - check write permissions,
2578 * we'll need to flip '..'.
2579 */
2580 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04002581 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582 if (error)
2583 return error;
2584 }
2585
2586 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2587 if (error)
2588 return error;
2589
2590 target = new_dentry->d_inode;
2591 if (target) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002592 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593 dentry_unhash(new_dentry);
2594 }
2595 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2596 error = -EBUSY;
2597 else
2598 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2599 if (target) {
2600 if (!error)
2601 target->i_flags |= S_DEAD;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002602 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603 if (d_unhashed(new_dentry))
2604 d_rehash(new_dentry);
2605 dput(new_dentry);
2606 }
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002607 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07002608 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2609 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610 return error;
2611}
2612
Adrian Bunk75c96f82005-05-05 16:16:09 -07002613static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2614 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615{
2616 struct inode *target;
2617 int error;
2618
2619 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2620 if (error)
2621 return error;
2622
2623 dget(new_dentry);
2624 target = new_dentry->d_inode;
2625 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002626 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2628 error = -EBUSY;
2629 else
2630 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2631 if (!error) {
Mark Fasheh349457c2006-09-08 14:22:21 -07002632 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002633 d_move(old_dentry, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634 }
2635 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002636 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002637 dput(new_dentry);
2638 return error;
2639}
2640
2641int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2642 struct inode *new_dir, struct dentry *new_dentry)
2643{
2644 int error;
2645 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Robert Love0eeca282005-07-12 17:06:03 -04002646 const char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647
2648 if (old_dentry->d_inode == new_dentry->d_inode)
2649 return 0;
2650
2651 error = may_delete(old_dir, old_dentry, is_dir);
2652 if (error)
2653 return error;
2654
2655 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02002656 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657 else
2658 error = may_delete(new_dir, new_dentry, is_dir);
2659 if (error)
2660 return error;
2661
Al Viroacfa4382008-12-04 10:06:33 -05002662 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663 return -EPERM;
2664
Jan Kara9e3509e2009-01-26 16:45:12 +01002665 vfs_dq_init(old_dir);
2666 vfs_dq_init(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667
Robert Love0eeca282005-07-12 17:06:03 -04002668 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2669
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670 if (is_dir)
2671 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2672 else
2673 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2674 if (!error) {
Robert Love0eeca282005-07-12 17:06:03 -04002675 const char *new_name = old_dentry->d_name.name;
John McCutchan89204c42005-08-15 12:13:28 -04002676 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04002677 new_dentry->d_inode, old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678 }
Robert Love0eeca282005-07-12 17:06:03 -04002679 fsnotify_oldname_free(old_name);
2680
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681 return error;
2682}
2683
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002684SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2685 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686{
Al Viro2ad94ae2008-07-21 09:32:51 -04002687 struct dentry *old_dir, *new_dir;
2688 struct dentry *old_dentry, *new_dentry;
2689 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690 struct nameidata oldnd, newnd;
Al Viro2ad94ae2008-07-21 09:32:51 -04002691 char *from;
2692 char *to;
2693 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002694
Al Viro2ad94ae2008-07-21 09:32:51 -04002695 error = user_path_parent(olddfd, oldname, &oldnd, &from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002696 if (error)
2697 goto exit;
2698
Al Viro2ad94ae2008-07-21 09:32:51 -04002699 error = user_path_parent(newdfd, newname, &newnd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700 if (error)
2701 goto exit1;
2702
2703 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08002704 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705 goto exit2;
2706
Jan Blunck4ac91372008-02-14 19:34:32 -08002707 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708 error = -EBUSY;
2709 if (oldnd.last_type != LAST_NORM)
2710 goto exit2;
2711
Jan Blunck4ac91372008-02-14 19:34:32 -08002712 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713 if (newnd.last_type != LAST_NORM)
2714 goto exit2;
2715
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002716 oldnd.flags &= ~LOOKUP_PARENT;
2717 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09002718 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002719
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720 trap = lock_rename(new_dir, old_dir);
2721
Christoph Hellwig49705b72005-11-08 21:35:06 -08002722 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002723 error = PTR_ERR(old_dentry);
2724 if (IS_ERR(old_dentry))
2725 goto exit3;
2726 /* source must exist */
2727 error = -ENOENT;
2728 if (!old_dentry->d_inode)
2729 goto exit4;
2730 /* unless the source is a directory trailing slashes give -ENOTDIR */
2731 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2732 error = -ENOTDIR;
2733 if (oldnd.last.name[oldnd.last.len])
2734 goto exit4;
2735 if (newnd.last.name[newnd.last.len])
2736 goto exit4;
2737 }
2738 /* source should not be ancestor of target */
2739 error = -EINVAL;
2740 if (old_dentry == trap)
2741 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08002742 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002743 error = PTR_ERR(new_dentry);
2744 if (IS_ERR(new_dentry))
2745 goto exit4;
2746 /* target should not be an ancestor of source */
2747 error = -ENOTEMPTY;
2748 if (new_dentry == trap)
2749 goto exit5;
2750
Dave Hansen9079b1e2008-02-15 14:37:49 -08002751 error = mnt_want_write(oldnd.path.mnt);
2752 if (error)
2753 goto exit5;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002754 error = security_path_rename(&oldnd.path, old_dentry,
2755 &newnd.path, new_dentry);
2756 if (error)
2757 goto exit6;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002758 error = vfs_rename(old_dir->d_inode, old_dentry,
2759 new_dir->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002760exit6:
Dave Hansen9079b1e2008-02-15 14:37:49 -08002761 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002762exit5:
2763 dput(new_dentry);
2764exit4:
2765 dput(old_dentry);
2766exit3:
2767 unlock_rename(new_dir, old_dir);
2768exit2:
Jan Blunck1d957f92008-02-14 19:34:35 -08002769 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002770 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002771exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002772 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773 putname(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002774exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002775 return error;
2776}
2777
Heiko Carstensa26eab22009-01-14 14:14:17 +01002778SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002779{
2780 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2781}
2782
Linus Torvalds1da177e2005-04-16 15:20:36 -07002783int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2784{
2785 int len;
2786
2787 len = PTR_ERR(link);
2788 if (IS_ERR(link))
2789 goto out;
2790
2791 len = strlen(link);
2792 if (len > (unsigned) buflen)
2793 len = buflen;
2794 if (copy_to_user(buffer, link, len))
2795 len = -EFAULT;
2796out:
2797 return len;
2798}
2799
2800/*
2801 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2802 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2803 * using) it for any given inode is up to filesystem.
2804 */
2805int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2806{
2807 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002808 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07002809 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002810
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002812 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07002813 if (IS_ERR(cookie))
2814 return PTR_ERR(cookie);
2815
2816 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2817 if (dentry->d_inode->i_op->put_link)
2818 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2819 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820}
2821
2822int vfs_follow_link(struct nameidata *nd, const char *link)
2823{
2824 return __vfs_follow_link(nd, link);
2825}
2826
2827/* get the link contents into pagecache */
2828static char *page_getlink(struct dentry * dentry, struct page **ppage)
2829{
Duane Griffinebd09ab2008-12-19 20:47:12 +00002830 char *kaddr;
2831 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002832 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07002833 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002834 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07002835 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00002837 kaddr = kmap(page);
2838 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2839 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002840}
2841
2842int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2843{
2844 struct page *page = NULL;
2845 char *s = page_getlink(dentry, &page);
2846 int res = vfs_readlink(dentry,buffer,buflen,s);
2847 if (page) {
2848 kunmap(page);
2849 page_cache_release(page);
2850 }
2851 return res;
2852}
2853
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002854void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002856 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002858 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002859}
2860
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002861void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002863 struct page *page = cookie;
2864
2865 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002866 kunmap(page);
2867 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868 }
2869}
2870
Nick Piggin54566b22009-01-04 12:00:53 -08002871/*
2872 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2873 */
2874int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002875{
2876 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002877 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002878 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08002879 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002880 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08002881 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2882 if (nofs)
2883 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002884
NeilBrown7e53cac2006-03-25 03:07:57 -08002885retry:
Nick Pigginafddba42007-10-16 01:25:01 -07002886 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08002887 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002888 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07002889 goto fail;
2890
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891 kaddr = kmap_atomic(page, KM_USER0);
2892 memcpy(kaddr, symname, len-1);
2893 kunmap_atomic(kaddr, KM_USER0);
Nick Pigginafddba42007-10-16 01:25:01 -07002894
2895 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2896 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002897 if (err < 0)
2898 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07002899 if (err < len-1)
2900 goto retry;
2901
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902 mark_inode_dirty(inode);
2903 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002904fail:
2905 return err;
2906}
2907
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002908int page_symlink(struct inode *inode, const char *symname, int len)
2909{
2910 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08002911 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002912}
2913
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002914const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002915 .readlink = generic_readlink,
2916 .follow_link = page_follow_link_light,
2917 .put_link = page_put_link,
2918};
2919
Al Viro2d8f3032008-07-22 09:59:21 -04002920EXPORT_SYMBOL(user_path_at);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002921EXPORT_SYMBOL(follow_down);
2922EXPORT_SYMBOL(follow_up);
2923EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2924EXPORT_SYMBOL(getname);
2925EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002926EXPORT_SYMBOL(lookup_one_len);
2927EXPORT_SYMBOL(page_follow_link_light);
2928EXPORT_SYMBOL(page_put_link);
2929EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002930EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002931EXPORT_SYMBOL(page_symlink);
2932EXPORT_SYMBOL(page_symlink_inode_operations);
2933EXPORT_SYMBOL(path_lookup);
Al Virod1811462008-08-02 00:49:18 -04002934EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002935EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04002936EXPORT_SYMBOL(inode_permission);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -08002937EXPORT_SYMBOL(file_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002938EXPORT_SYMBOL(unlock_rename);
2939EXPORT_SYMBOL(vfs_create);
2940EXPORT_SYMBOL(vfs_follow_link);
2941EXPORT_SYMBOL(vfs_link);
2942EXPORT_SYMBOL(vfs_mkdir);
2943EXPORT_SYMBOL(vfs_mknod);
2944EXPORT_SYMBOL(generic_permission);
2945EXPORT_SYMBOL(vfs_readlink);
2946EXPORT_SYMBOL(vfs_rename);
2947EXPORT_SYMBOL(vfs_rmdir);
2948EXPORT_SYMBOL(vfs_symlink);
2949EXPORT_SYMBOL(vfs_unlink);
2950EXPORT_SYMBOL(dentry_unhash);
2951EXPORT_SYMBOL(generic_readlink);