blob: b55440baf7ab832e9d05ae8fdddc2337da57b5ed [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);
564 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
565 error = PTR_ERR(cookie);
566 if (!IS_ERR(cookie)) {
567 char *s = nd_get_link(nd);
568 error = 0;
569 if (s)
570 error = __vfs_follow_link(nd, s);
Jeff Layton39159de2009-12-07 12:01:50 -0500571 else if (nd->last_type == LAST_BIND) {
572 error = force_reval_path(&nd->path, nd);
573 if (error)
574 path_put(&nd->path);
575 }
Ian Kent051d3812006-03-27 01:14:53 -0800576 if (dentry->d_inode->i_op->put_link)
577 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
578 }
Ian Kent051d3812006-03-27 01:14:53 -0800579 return error;
580}
581
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582/*
583 * This limits recursive symlink follows to 8, while
584 * limiting consecutive symlinks to 40.
585 *
586 * Without that kind of total limit, nasty chains of consecutive
587 * symlinks can cause almost arbitrarily long lookups.
588 */
Al Viro90ebe562005-06-06 13:35:58 -0700589static inline int do_follow_link(struct path *path, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590{
591 int err = -ELOOP;
592 if (current->link_count >= MAX_NESTED_LINKS)
593 goto loop;
594 if (current->total_link_count >= 40)
595 goto loop;
596 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
597 cond_resched();
Al Viro90ebe562005-06-06 13:35:58 -0700598 err = security_inode_follow_link(path->dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 if (err)
600 goto loop;
601 current->link_count++;
602 current->total_link_count++;
603 nd->depth++;
Al Virocd4e91d2005-06-06 13:36:03 -0700604 err = __do_follow_link(path, nd);
Al Viro258fa992009-08-09 01:32:02 +0400605 path_put(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 current->link_count--;
607 nd->depth--;
608 return err;
609loop:
Jan Blunck1d957f92008-02-14 19:34:35 -0800610 path_put_conditional(path, nd);
611 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 return err;
613}
614
Al Virobab77eb2009-04-18 03:26:48 -0400615int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616{
617 struct vfsmount *parent;
618 struct dentry *mountpoint;
619 spin_lock(&vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400620 parent = path->mnt->mnt_parent;
621 if (parent == path->mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 spin_unlock(&vfsmount_lock);
623 return 0;
624 }
625 mntget(parent);
Al Virobab77eb2009-04-18 03:26:48 -0400626 mountpoint = dget(path->mnt->mnt_mountpoint);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 spin_unlock(&vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400628 dput(path->dentry);
629 path->dentry = mountpoint;
630 mntput(path->mnt);
631 path->mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 return 1;
633}
634
635/* no need for dcache_lock, as serialization is taken care in
636 * namespace.c
637 */
Al Viro463ffb22005-06-06 13:36:05 -0700638static int __follow_mount(struct path *path)
639{
640 int res = 0;
641 while (d_mountpoint(path->dentry)) {
Al Viro1c755af2009-04-18 14:06:57 -0400642 struct vfsmount *mounted = lookup_mnt(path);
Al Viro463ffb22005-06-06 13:36:05 -0700643 if (!mounted)
644 break;
645 dput(path->dentry);
646 if (res)
647 mntput(path->mnt);
648 path->mnt = mounted;
649 path->dentry = dget(mounted->mnt_root);
650 res = 1;
651 }
652 return res;
653}
654
Al Viro79ed0222009-04-18 13:59:41 -0400655static void follow_mount(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656{
Al Viro79ed0222009-04-18 13:59:41 -0400657 while (d_mountpoint(path->dentry)) {
Al Viro1c755af2009-04-18 14:06:57 -0400658 struct vfsmount *mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 if (!mounted)
660 break;
Al Viro79ed0222009-04-18 13:59:41 -0400661 dput(path->dentry);
662 mntput(path->mnt);
663 path->mnt = mounted;
664 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666}
667
668/* no need for dcache_lock, as serialization is taken care in
669 * namespace.c
670 */
Al Viro9393bd02009-04-18 13:58:15 -0400671int follow_down(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672{
673 struct vfsmount *mounted;
674
Al Viro1c755af2009-04-18 14:06:57 -0400675 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -0400677 dput(path->dentry);
678 mntput(path->mnt);
679 path->mnt = mounted;
680 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 return 1;
682 }
683 return 0;
684}
685
Arjan van de Venf1662352006-01-14 13:21:31 -0800686static __always_inline void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687{
Al Viro2a737872009-04-07 11:49:53 -0400688 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -0700689
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 while(1) {
691 struct vfsmount *parent;
Jan Blunck4ac91372008-02-14 19:34:32 -0800692 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693
Al Viro2a737872009-04-07 11:49:53 -0400694 if (nd->path.dentry == nd->root.dentry &&
695 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 break;
697 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 spin_lock(&dcache_lock);
Jan Blunck4ac91372008-02-14 19:34:32 -0800699 if (nd->path.dentry != nd->path.mnt->mnt_root) {
700 nd->path.dentry = dget(nd->path.dentry->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 spin_unlock(&dcache_lock);
702 dput(old);
703 break;
704 }
705 spin_unlock(&dcache_lock);
706 spin_lock(&vfsmount_lock);
Jan Blunck4ac91372008-02-14 19:34:32 -0800707 parent = nd->path.mnt->mnt_parent;
708 if (parent == nd->path.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 spin_unlock(&vfsmount_lock);
710 break;
711 }
712 mntget(parent);
Jan Blunck4ac91372008-02-14 19:34:32 -0800713 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 spin_unlock(&vfsmount_lock);
715 dput(old);
Jan Blunck4ac91372008-02-14 19:34:32 -0800716 mntput(nd->path.mnt);
717 nd->path.mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 }
Al Viro79ed0222009-04-18 13:59:41 -0400719 follow_mount(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720}
721
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722/*
723 * It's more convoluted than I'd like it to be, but... it's still fairly
724 * small and for now I'd prefer to have fast path as straight as possible.
725 * It _is_ time-critical.
726 */
727static int do_lookup(struct nameidata *nd, struct qstr *name,
728 struct path *path)
729{
Jan Blunck4ac91372008-02-14 19:34:32 -0800730 struct vfsmount *mnt = nd->path.mnt;
Al Viro6e6b1bd2009-08-13 23:38:37 +0400731 struct dentry *dentry, *parent;
732 struct inode *dir;
Al Viro3cac2602009-08-13 18:27:43 +0400733 /*
734 * See if the low-level filesystem might want
735 * to use its own hash..
736 */
737 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
738 int err = nd->path.dentry->d_op->d_hash(nd->path.dentry, name);
739 if (err < 0)
740 return err;
741 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742
Al Viro3cac2602009-08-13 18:27:43 +0400743 dentry = __d_lookup(nd->path.dentry, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 if (!dentry)
745 goto need_lookup;
746 if (dentry->d_op && dentry->d_op->d_revalidate)
747 goto need_revalidate;
748done:
749 path->mnt = mnt;
750 path->dentry = dentry;
Al Viro634ee702005-06-06 13:36:13 -0700751 __follow_mount(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 return 0;
753
754need_lookup:
Al Viro6e6b1bd2009-08-13 23:38:37 +0400755 parent = nd->path.dentry;
756 dir = parent->d_inode;
757
758 mutex_lock(&dir->i_mutex);
759 /*
760 * First re-do the cached lookup just in case it was created
761 * while we waited for the directory semaphore..
762 *
763 * FIXME! This could use version numbering or similar to
764 * avoid unnecessary cache lookups.
765 *
766 * The "dcache_lock" is purely to protect the RCU list walker
767 * from concurrent renames at this point (we mustn't get false
768 * negatives from the RCU list walk here, unlike the optimistic
769 * fast walk).
770 *
771 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
772 */
773 dentry = d_lookup(parent, name);
774 if (!dentry) {
775 struct dentry *new;
776
777 /* Don't create child dentry for a dead directory. */
778 dentry = ERR_PTR(-ENOENT);
779 if (IS_DEADDIR(dir))
780 goto out_unlock;
781
782 new = d_alloc(parent, name);
783 dentry = ERR_PTR(-ENOMEM);
784 if (new) {
785 dentry = dir->i_op->lookup(dir, new, nd);
786 if (dentry)
787 dput(new);
788 else
789 dentry = new;
790 }
791out_unlock:
792 mutex_unlock(&dir->i_mutex);
793 if (IS_ERR(dentry))
794 goto fail;
795 goto done;
796 }
797
798 /*
799 * Uhhuh! Nasty case: the cache was re-populated while
800 * we waited on the semaphore. Need to revalidate.
801 */
802 mutex_unlock(&dir->i_mutex);
803 if (dentry->d_op && dentry->d_op->d_revalidate) {
804 dentry = do_revalidate(dentry, nd);
805 if (!dentry)
806 dentry = ERR_PTR(-ENOENT);
807 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 if (IS_ERR(dentry))
809 goto fail;
810 goto done;
811
812need_revalidate:
Ian Kentbcdc5e02006-09-27 01:50:44 -0700813 dentry = do_revalidate(dentry, nd);
814 if (!dentry)
815 goto need_lookup;
816 if (IS_ERR(dentry))
817 goto fail;
818 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819
820fail:
821 return PTR_ERR(dentry);
822}
823
824/*
825 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +0100826 * This is the basic name resolution function, turning a pathname into
827 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 *
Prasanna Medaea3834d2005-04-29 16:00:17 +0100829 * Returns 0 and nd will have valid dentry and mnt on success.
830 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 */
Al Viro6de88d72009-08-09 01:41:57 +0400832static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833{
834 struct path next;
835 struct inode *inode;
836 int err;
837 unsigned int lookup_flags = nd->flags;
838
839 while (*name=='/')
840 name++;
841 if (!*name)
842 goto return_reval;
843
Jan Blunck4ac91372008-02-14 19:34:32 -0800844 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 if (nd->depth)
Trond Myklebustf55eab82006-02-04 23:28:01 -0800846 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847
848 /* At this point we know we have a real path component. */
849 for(;;) {
850 unsigned long hash;
851 struct qstr this;
852 unsigned int c;
853
Trond Myklebustcdce5d62005-10-18 14:20:18 -0700854 nd->flags |= LOOKUP_CONTINUE;
Al Virob75b5082009-12-16 01:01:38 -0500855 err = exec_permission(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 if (err)
857 break;
858
859 this.name = name;
860 c = *(const unsigned char *)name;
861
862 hash = init_name_hash();
863 do {
864 name++;
865 hash = partial_name_hash(c, hash);
866 c = *(const unsigned char *)name;
867 } while (c && (c != '/'));
868 this.len = name - (const char *) this.name;
869 this.hash = end_name_hash(hash);
870
871 /* remove trailing slashes? */
872 if (!c)
873 goto last_component;
874 while (*++name == '/');
875 if (!*name)
876 goto last_with_slashes;
877
878 /*
879 * "." and ".." are special - ".." especially so because it has
880 * to be able to know about the current root directory and
881 * parent relationships.
882 */
883 if (this.name[0] == '.') switch (this.len) {
884 default:
885 break;
886 case 2:
887 if (this.name[1] != '.')
888 break;
Al Viro58c465e2005-06-06 13:36:13 -0700889 follow_dotdot(nd);
Jan Blunck4ac91372008-02-14 19:34:32 -0800890 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 /* fallthrough */
892 case 1:
893 continue;
894 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 /* This does the actual lookups.. */
896 err = do_lookup(nd, &this, &next);
897 if (err)
898 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899
900 err = -ENOENT;
901 inode = next.dentry->d_inode;
902 if (!inode)
903 goto out_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904
905 if (inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700906 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 if (err)
908 goto return_err;
909 err = -ENOENT;
Jan Blunck4ac91372008-02-14 19:34:32 -0800910 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 if (!inode)
912 break;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700913 } else
914 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 err = -ENOTDIR;
916 if (!inode->i_op->lookup)
917 break;
918 continue;
919 /* here ends the main loop */
920
921last_with_slashes:
922 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
923last_component:
Trond Myklebustf55eab82006-02-04 23:28:01 -0800924 /* Clear LOOKUP_CONTINUE iff it was previously unset */
925 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 if (lookup_flags & LOOKUP_PARENT)
927 goto lookup_parent;
928 if (this.name[0] == '.') switch (this.len) {
929 default:
930 break;
931 case 2:
932 if (this.name[1] != '.')
933 break;
Al Viro58c465e2005-06-06 13:36:13 -0700934 follow_dotdot(nd);
Jan Blunck4ac91372008-02-14 19:34:32 -0800935 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 /* fallthrough */
937 case 1:
938 goto return_reval;
939 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 err = do_lookup(nd, &this, &next);
941 if (err)
942 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 inode = next.dentry->d_inode;
944 if ((lookup_flags & LOOKUP_FOLLOW)
Al Viroacfa4382008-12-04 10:06:33 -0500945 && inode && inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700946 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 if (err)
948 goto return_err;
Jan Blunck4ac91372008-02-14 19:34:32 -0800949 inode = nd->path.dentry->d_inode;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700950 } else
951 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 err = -ENOENT;
953 if (!inode)
954 break;
955 if (lookup_flags & LOOKUP_DIRECTORY) {
956 err = -ENOTDIR;
Al Viroacfa4382008-12-04 10:06:33 -0500957 if (!inode->i_op->lookup)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 break;
959 }
960 goto return_base;
961lookup_parent:
962 nd->last = this;
963 nd->last_type = LAST_NORM;
964 if (this.name[0] != '.')
965 goto return_base;
966 if (this.len == 1)
967 nd->last_type = LAST_DOT;
968 else if (this.len == 2 && this.name[1] == '.')
969 nd->last_type = LAST_DOTDOT;
970 else
971 goto return_base;
972return_reval:
973 /*
974 * We bypassed the ordinary revalidation routines.
975 * We may need to check the cached dentry for staleness.
976 */
Jan Blunck4ac91372008-02-14 19:34:32 -0800977 if (nd->path.dentry && nd->path.dentry->d_sb &&
978 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 err = -ESTALE;
980 /* Note: we do not d_invalidate() */
Jan Blunck4ac91372008-02-14 19:34:32 -0800981 if (!nd->path.dentry->d_op->d_revalidate(
982 nd->path.dentry, nd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 break;
984 }
985return_base:
986 return 0;
987out_dput:
Jan Blunck1d957f92008-02-14 19:34:35 -0800988 path_put_conditional(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 break;
990 }
Jan Blunck1d957f92008-02-14 19:34:35 -0800991 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992return_err:
993 return err;
994}
995
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800996static int path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997{
Al Viro6de88d72009-08-09 01:41:57 +0400998 struct path save = nd->path;
999 int result;
1000
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 current->total_link_count = 0;
Al Viro6de88d72009-08-09 01:41:57 +04001002
1003 /* make sure the stuff we saved doesn't go away */
1004 path_get(&save);
1005
1006 result = link_path_walk(name, nd);
1007 if (result == -ESTALE) {
1008 /* nd->path had been dropped */
1009 current->total_link_count = 0;
1010 nd->path = save;
1011 path_get(&nd->path);
1012 nd->flags |= LOOKUP_REVAL;
1013 result = link_path_walk(name, nd);
1014 }
1015
1016 path_put(&save);
1017
1018 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019}
1020
Al Viro9b4a9b12009-04-07 11:44:16 -04001021static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001023 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001024 int fput_needed;
1025 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026
1027 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1028 nd->flags = flags;
1029 nd->depth = 0;
Al Viro2a737872009-04-07 11:49:53 -04001030 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 if (*name=='/') {
Al Viro2a737872009-04-07 11:49:53 -04001033 set_root(nd);
1034 nd->path = nd->root;
1035 path_get(&nd->root);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001036 } else if (dfd == AT_FDCWD) {
Al Viro2a737872009-04-07 11:49:53 -04001037 struct fs_struct *fs = current->fs;
Andreas Mohre518ddb2006-09-29 02:01:22 -07001038 read_lock(&fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08001039 nd->path = fs->pwd;
1040 path_get(&fs->pwd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001041 read_unlock(&fs->lock);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001042 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001043 struct dentry *dentry;
1044
1045 file = fget_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001046 retval = -EBADF;
1047 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001048 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001049
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001050 dentry = file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001051
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001052 retval = -ENOTDIR;
1053 if (!S_ISDIR(dentry->d_inode->i_mode))
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001054 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001055
1056 retval = file_permission(file, MAY_EXEC);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001057 if (retval)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001058 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001059
Jan Blunck5dd784d2008-02-14 19:34:38 -08001060 nd->path = file->f_path;
1061 path_get(&file->f_path);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001062
1063 fput_light(file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001065 return 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001066
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001067fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001068 fput_light(file, fput_needed);
Al Viro9b4a9b12009-04-07 11:44:16 -04001069out_fail:
1070 return retval;
1071}
1072
1073/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1074static int do_path_lookup(int dfd, const char *name,
1075 unsigned int flags, struct nameidata *nd)
1076{
1077 int retval = path_init(dfd, name, flags, nd);
1078 if (!retval)
1079 retval = path_walk(name, nd);
1080 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1081 nd->path.dentry->d_inode))
1082 audit_inode(name, nd->path.dentry);
Al Viro2a737872009-04-07 11:49:53 -04001083 if (nd->root.mnt) {
1084 path_put(&nd->root);
1085 nd->root.mnt = NULL;
1086 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001087 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088}
1089
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -08001090int path_lookup(const char *name, unsigned int flags,
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001091 struct nameidata *nd)
1092{
1093 return do_path_lookup(AT_FDCWD, name, flags, nd);
1094}
1095
Al Virod1811462008-08-02 00:49:18 -04001096int kern_path(const char *name, unsigned int flags, struct path *path)
1097{
1098 struct nameidata nd;
1099 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1100 if (!res)
1101 *path = nd.path;
1102 return res;
1103}
1104
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001105/**
1106 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1107 * @dentry: pointer to dentry of the base directory
1108 * @mnt: pointer to vfs mount of the base directory
1109 * @name: pointer to file name
1110 * @flags: lookup flags
1111 * @nd: pointer to nameidata
1112 */
1113int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1114 const char *name, unsigned int flags,
1115 struct nameidata *nd)
1116{
1117 int retval;
1118
1119 /* same as do_path_lookup */
1120 nd->last_type = LAST_ROOT;
1121 nd->flags = flags;
1122 nd->depth = 0;
1123
Jan Blunckc8e7f442008-06-09 16:40:35 -07001124 nd->path.dentry = dentry;
1125 nd->path.mnt = mnt;
1126 path_get(&nd->path);
Al Viro5b857112009-04-07 11:53:49 -04001127 nd->root = nd->path;
1128 path_get(&nd->root);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001129
1130 retval = path_walk(name, nd);
Jan Blunck4ac91372008-02-14 19:34:32 -08001131 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1132 nd->path.dentry->d_inode))
1133 audit_inode(name, nd->path.dentry);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001134
Al Viro5b857112009-04-07 11:53:49 -04001135 path_put(&nd->root);
1136 nd->root.mnt = NULL;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001137
Al Viro2a737872009-04-07 11:49:53 -04001138 return retval;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001139}
1140
Christoph Hellwigeead1912007-10-16 23:25:38 -07001141static struct dentry *__lookup_hash(struct qstr *name,
1142 struct dentry *base, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143{
James Morris057f6c02007-04-26 00:12:05 -07001144 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 struct inode *inode;
1146 int err;
1147
1148 inode = base->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149
1150 /*
1151 * See if the low-level filesystem might want
1152 * to use its own hash..
1153 */
1154 if (base->d_op && base->d_op->d_hash) {
1155 err = base->d_op->d_hash(base, name);
1156 dentry = ERR_PTR(err);
1157 if (err < 0)
1158 goto out;
1159 }
1160
Al Viro6e6b1bd2009-08-13 23:38:37 +04001161 dentry = __d_lookup(base, name);
1162
1163 /* lockess __d_lookup may fail due to concurrent d_move()
1164 * in some unrelated directory, so try with d_lookup
1165 */
1166 if (!dentry)
1167 dentry = d_lookup(base, name);
1168
1169 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
1170 dentry = do_revalidate(dentry, nd);
1171
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 if (!dentry) {
Miklos Szeredid70b67c2008-07-02 21:30:15 +02001173 struct dentry *new;
1174
1175 /* Don't create child dentry for a dead directory. */
1176 dentry = ERR_PTR(-ENOENT);
1177 if (IS_DEADDIR(inode))
1178 goto out;
1179
1180 new = d_alloc(base, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 dentry = ERR_PTR(-ENOMEM);
1182 if (!new)
1183 goto out;
1184 dentry = inode->i_op->lookup(inode, new, nd);
1185 if (!dentry)
1186 dentry = new;
1187 else
1188 dput(new);
1189 }
1190out:
1191 return dentry;
1192}
1193
James Morris057f6c02007-04-26 00:12:05 -07001194/*
1195 * Restricted form of lookup. Doesn't follow links, single-component only,
1196 * needs parent already locked. Doesn't follow mounts.
1197 * SMP-safe.
1198 */
Adrian Bunka244e162006-03-31 02:32:11 -08001199static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200{
Christoph Hellwigeead1912007-10-16 23:25:38 -07001201 int err;
1202
Al Virob75b5082009-12-16 01:01:38 -05001203 err = exec_permission(nd->path.dentry->d_inode);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001204 if (err)
1205 return ERR_PTR(err);
Jan Blunck4ac91372008-02-14 19:34:32 -08001206 return __lookup_hash(&nd->last, nd->path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207}
1208
Christoph Hellwigeead1912007-10-16 23:25:38 -07001209static int __lookup_one_len(const char *name, struct qstr *this,
1210 struct dentry *base, int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211{
1212 unsigned long hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 unsigned int c;
1214
James Morris057f6c02007-04-26 00:12:05 -07001215 this->name = name;
1216 this->len = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 if (!len)
James Morris057f6c02007-04-26 00:12:05 -07001218 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219
1220 hash = init_name_hash();
1221 while (len--) {
1222 c = *(const unsigned char *)name++;
1223 if (c == '/' || c == '\0')
James Morris057f6c02007-04-26 00:12:05 -07001224 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 hash = partial_name_hash(c, hash);
1226 }
James Morris057f6c02007-04-26 00:12:05 -07001227 this->hash = end_name_hash(hash);
1228 return 0;
1229}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230
Christoph Hellwigeead1912007-10-16 23:25:38 -07001231/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001232 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07001233 * @name: pathname component to lookup
1234 * @base: base directory to lookup from
1235 * @len: maximum length @len should be interpreted to
1236 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07001237 * Note that this routine is purely a helper for filesystem usage and should
1238 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07001239 * nameidata argument is passed to the filesystem methods and a filesystem
1240 * using this helper needs to be prepared for that.
1241 */
James Morris057f6c02007-04-26 00:12:05 -07001242struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1243{
1244 int err;
1245 struct qstr this;
1246
David Woodhouse2f9092e2009-04-20 23:18:37 +01001247 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1248
James Morris057f6c02007-04-26 00:12:05 -07001249 err = __lookup_one_len(name, &this, base, len);
1250 if (err)
1251 return ERR_PTR(err);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001252
Al Virob75b5082009-12-16 01:01:38 -05001253 err = exec_permission(base->d_inode);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001254 if (err)
1255 return ERR_PTR(err);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001256 return __lookup_hash(&this, base, NULL);
James Morris057f6c02007-04-26 00:12:05 -07001257}
1258
Al Viro2d8f3032008-07-22 09:59:21 -04001259int user_path_at(int dfd, const char __user *name, unsigned flags,
1260 struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261{
Al Viro2d8f3032008-07-22 09:59:21 -04001262 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 char *tmp = getname(name);
1264 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04001266
1267 BUG_ON(flags & LOOKUP_PARENT);
1268
1269 err = do_path_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04001271 if (!err)
1272 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 }
1274 return err;
1275}
1276
Al Viro2ad94ae2008-07-21 09:32:51 -04001277static int user_path_parent(int dfd, const char __user *path,
1278 struct nameidata *nd, char **name)
1279{
1280 char *s = getname(path);
1281 int error;
1282
1283 if (IS_ERR(s))
1284 return PTR_ERR(s);
1285
1286 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1287 if (error)
1288 putname(s);
1289 else
1290 *name = s;
1291
1292 return error;
1293}
1294
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295/*
1296 * It's inline, so penalty for filesystems that don't use sticky bit is
1297 * minimal.
1298 */
1299static inline int check_sticky(struct inode *dir, struct inode *inode)
1300{
David Howellsda9592e2008-11-14 10:39:05 +11001301 uid_t fsuid = current_fsuid();
1302
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 if (!(dir->i_mode & S_ISVTX))
1304 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001305 if (inode->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001307 if (dir->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 return 0;
1309 return !capable(CAP_FOWNER);
1310}
1311
1312/*
1313 * Check whether we can remove a link victim from directory dir, check
1314 * whether the type of victim is right.
1315 * 1. We can't do it if dir is read-only (done in permission())
1316 * 2. We should have write and exec permissions on dir
1317 * 3. We can't remove anything from append-only dir
1318 * 4. We can't do anything with immutable dir (done in permission())
1319 * 5. If the sticky bit on dir is set we should either
1320 * a. be owner of dir, or
1321 * b. be owner of victim, or
1322 * c. have CAP_FOWNER capability
1323 * 6. If the victim is append-only or immutable we can't do antyhing with
1324 * links pointing to it.
1325 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1326 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1327 * 9. We can't remove a root or mountpoint.
1328 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1329 * nfs_async_unlink().
1330 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001331static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332{
1333 int error;
1334
1335 if (!victim->d_inode)
1336 return -ENOENT;
1337
1338 BUG_ON(victim->d_parent->d_inode != dir);
Al Viro5a190ae2007-06-07 12:19:32 -04001339 audit_inode_child(victim->d_name.name, victim, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340
Al Virof419a2e2008-07-22 00:07:17 -04001341 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 if (error)
1343 return error;
1344 if (IS_APPEND(dir))
1345 return -EPERM;
1346 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
Hugh Dickinsf9454542008-11-19 15:36:38 -08001347 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 return -EPERM;
1349 if (isdir) {
1350 if (!S_ISDIR(victim->d_inode->i_mode))
1351 return -ENOTDIR;
1352 if (IS_ROOT(victim))
1353 return -EBUSY;
1354 } else if (S_ISDIR(victim->d_inode->i_mode))
1355 return -EISDIR;
1356 if (IS_DEADDIR(dir))
1357 return -ENOENT;
1358 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1359 return -EBUSY;
1360 return 0;
1361}
1362
1363/* Check whether we can create an object with dentry child in directory
1364 * dir.
1365 * 1. We can't do it if child already exists (open has special treatment for
1366 * this case, but since we are inlined it's OK)
1367 * 2. We can't do it if dir is read-only (done in permission())
1368 * 3. We should have write and exec permissions on dir
1369 * 4. We can't do it if dir is immutable (done in permission())
1370 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02001371static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372{
1373 if (child->d_inode)
1374 return -EEXIST;
1375 if (IS_DEADDIR(dir))
1376 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04001377 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378}
1379
1380/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 * O_DIRECTORY translates into forcing a directory lookup.
1382 */
1383static inline int lookup_flags(unsigned int f)
1384{
1385 unsigned long retval = LOOKUP_FOLLOW;
1386
1387 if (f & O_NOFOLLOW)
1388 retval &= ~LOOKUP_FOLLOW;
1389
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 if (f & O_DIRECTORY)
1391 retval |= LOOKUP_DIRECTORY;
1392
1393 return retval;
1394}
1395
1396/*
1397 * p1 and p2 should be directories on the same fs.
1398 */
1399struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1400{
1401 struct dentry *p;
1402
1403 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001404 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 return NULL;
1406 }
1407
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001408 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001410 p = d_ancestor(p2, p1);
1411 if (p) {
1412 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1413 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1414 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 }
1416
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001417 p = d_ancestor(p1, p2);
1418 if (p) {
1419 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1420 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1421 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 }
1423
Ingo Molnarf2eace22006-07-03 00:25:05 -07001424 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1425 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 return NULL;
1427}
1428
1429void unlock_rename(struct dentry *p1, struct dentry *p2)
1430{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001431 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001433 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001434 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 }
1436}
1437
1438int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1439 struct nameidata *nd)
1440{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001441 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442
1443 if (error)
1444 return error;
1445
Al Viroacfa4382008-12-04 10:06:33 -05001446 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 return -EACCES; /* shouldn't it be ENOSYS? */
1448 mode &= S_IALLUGO;
1449 mode |= S_IFREG;
1450 error = security_inode_create(dir, dentry, mode);
1451 if (error)
1452 return error;
Jan Kara9e3509e2009-01-26 16:45:12 +01001453 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001455 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001456 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 return error;
1458}
1459
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001460int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001462 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 struct inode *inode = dentry->d_inode;
1464 int error;
1465
1466 if (!inode)
1467 return -ENOENT;
1468
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001469 switch (inode->i_mode & S_IFMT) {
1470 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001472 case S_IFDIR:
1473 if (acc_mode & MAY_WRITE)
1474 return -EISDIR;
1475 break;
1476 case S_IFBLK:
1477 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001478 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001480 /*FALLTHRU*/
1481 case S_IFIFO:
1482 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001484 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001485 }
Dave Hansenb41572e2007-10-16 23:31:14 -07001486
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001487 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07001488 if (error)
1489 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05001490
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 /*
1492 * An append-only file must be opened in append mode for writing.
1493 */
1494 if (IS_APPEND(inode)) {
1495 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05001496 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05001498 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 }
1500
1501 /* O_NOATIME can only be set by the owner or superuser */
Al Viro7715b522009-12-16 03:54:00 -05001502 if (flag & O_NOATIME && !is_owner_or_cap(inode))
1503 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504
1505 /*
1506 * Ensure there are no outstanding leases on the file.
1507 */
Al Virob65a9cf2009-12-16 06:27:40 -05001508 return break_lease(inode, flag);
Al Viro7715b522009-12-16 03:54:00 -05001509}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510
Al Viro7715b522009-12-16 03:54:00 -05001511static int handle_truncate(struct path *path)
1512{
1513 struct inode *inode = path->dentry->d_inode;
1514 int error = get_write_access(inode);
1515 if (error)
1516 return error;
1517 /*
1518 * Refuse to truncate files with mandatory locks held on them.
1519 */
1520 error = locks_verify_locked(inode);
1521 if (!error)
1522 error = security_path_truncate(path, 0,
1523 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN);
1524 if (!error) {
1525 error = do_truncate(path->dentry, 0,
1526 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1527 NULL);
1528 }
1529 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04001530 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531}
1532
Dave Hansend57999e2008-02-15 14:37:27 -08001533/*
1534 * Be careful about ever adding any more callers of this
1535 * function. Its flags must be in the namei format, not
1536 * what get passed to sys_open().
1537 */
1538static int __open_namei_create(struct nameidata *nd, struct path *path,
Dave Hansenaab520e2006-09-30 23:29:02 -07001539 int flag, int mode)
1540{
1541 int error;
Jan Blunck4ac91372008-02-14 19:34:32 -08001542 struct dentry *dir = nd->path.dentry;
Dave Hansenaab520e2006-09-30 23:29:02 -07001543
1544 if (!IS_POSIXACL(dir->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04001545 mode &= ~current_umask();
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001546 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1547 if (error)
1548 goto out_unlock;
Dave Hansenaab520e2006-09-30 23:29:02 -07001549 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001550out_unlock:
Dave Hansenaab520e2006-09-30 23:29:02 -07001551 mutex_unlock(&dir->d_inode->i_mutex);
Jan Blunck4ac91372008-02-14 19:34:32 -08001552 dput(nd->path.dentry);
1553 nd->path.dentry = path->dentry;
Dave Hansenaab520e2006-09-30 23:29:02 -07001554 if (error)
1555 return error;
1556 /* Don't check for write permission, don't truncate */
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001557 return may_open(&nd->path, 0, flag & ~O_TRUNC);
Dave Hansenaab520e2006-09-30 23:29:02 -07001558}
1559
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560/*
Dave Hansend57999e2008-02-15 14:37:27 -08001561 * Note that while the flag value (low two bits) for sys_open means:
1562 * 00 - read-only
1563 * 01 - write-only
1564 * 10 - read-write
1565 * 11 - special
1566 * it is changed into
1567 * 00 - no permissions needed
1568 * 01 - read-permission
1569 * 10 - write-permission
1570 * 11 - read-write
1571 * for the internal routines (ie open_namei()/follow_link() etc)
1572 * This is more logical, and also allows the 00 "no perm needed"
1573 * to be used for symlinks (where the permissions are checked
1574 * later).
1575 *
1576*/
1577static inline int open_to_namei_flags(int flag)
1578{
1579 if ((flag+1) & O_ACCMODE)
1580 flag++;
1581 return flag;
1582}
1583
Al Viro7715b522009-12-16 03:54:00 -05001584static int open_will_truncate(int flag, struct inode *inode)
Dave Hansen4a3fd212008-02-15 14:37:48 -08001585{
1586 /*
1587 * We'll never write to the fs underlying
1588 * a device file.
1589 */
1590 if (special_file(inode->i_mode))
1591 return 0;
1592 return (flag & O_TRUNC);
1593}
1594
Dave Hansend57999e2008-02-15 14:37:27 -08001595/*
Dave Hansen4a3fd212008-02-15 14:37:48 -08001596 * Note that the low bits of the passed in "open_flag"
1597 * are not the same as in the local variable "flag". See
1598 * open_to_namei_flags() for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001600struct file *do_filp_open(int dfd, const char *pathname,
Al Viro6e8341a2009-04-06 11:16:22 -04001601 int open_flag, int mode, int acc_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602{
Dave Hansen4a3fd212008-02-15 14:37:48 -08001603 struct file *filp;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001604 struct nameidata nd;
Al Viro6e8341a2009-04-06 11:16:22 -04001605 int error;
Al Viro6de88d72009-08-09 01:41:57 +04001606 struct path path, save;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 struct dentry *dir;
1608 int count = 0;
Al Viro7715b522009-12-16 03:54:00 -05001609 int will_truncate;
Dave Hansend57999e2008-02-15 14:37:27 -08001610 int flag = open_to_namei_flags(open_flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611
Christoph Hellwig6b2f3d12009-10-27 11:05:28 +01001612 /*
1613 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1614 * check for O_DSYNC if the need any syncing at all we enforce it's
1615 * always set instead of having to deal with possibly weird behaviour
1616 * for malicious applications setting only __O_SYNC.
1617 */
1618 if (open_flag & __O_SYNC)
1619 open_flag |= O_DSYNC;
1620
Al Viro6e8341a2009-04-06 11:16:22 -04001621 if (!acc_mode)
1622 acc_mode = MAY_OPEN | ACC_MODE(flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623
Trond Myklebust834f2a42005-10-18 14:20:16 -07001624 /* O_TRUNC implies we need access checks for write permissions */
1625 if (flag & O_TRUNC)
1626 acc_mode |= MAY_WRITE;
1627
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 /* Allow the LSM permission hook to distinguish append
1629 access from general write access. */
1630 if (flag & O_APPEND)
1631 acc_mode |= MAY_APPEND;
1632
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 /*
1634 * The simplest case - just a plain lookup.
1635 */
1636 if (!(flag & O_CREAT)) {
Al Viro2dd6d1f2009-08-13 20:40:45 +04001637 filp = get_empty_filp();
1638
1639 if (filp == NULL)
1640 return ERR_PTR(-ENFILE);
1641 nd.intent.open.file = filp;
Al Viro482928d2009-12-19 10:10:39 -05001642 filp->f_flags = open_flag;
Al Viro2dd6d1f2009-08-13 20:40:45 +04001643 nd.intent.open.flags = flag;
1644 nd.intent.open.create_mode = 0;
1645 error = do_path_lookup(dfd, pathname,
1646 lookup_flags(flag)|LOOKUP_OPEN, &nd);
1647 if (IS_ERR(nd.intent.open.file)) {
1648 if (error == 0) {
1649 error = PTR_ERR(nd.intent.open.file);
1650 path_put(&nd.path);
1651 }
1652 } else if (error)
1653 release_open_intent(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 if (error)
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001655 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 goto ok;
1657 }
1658
1659 /*
1660 * Create - we need to know the parent.
1661 */
Al Viro9b4a9b12009-04-07 11:44:16 -04001662 error = path_init(dfd, pathname, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 if (error)
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001664 return ERR_PTR(error);
Al Viro9b4a9b12009-04-07 11:44:16 -04001665 error = path_walk(pathname, &nd);
J. R. Okajima654f5622009-06-18 23:30:15 +09001666 if (error) {
1667 if (nd.root.mnt)
1668 path_put(&nd.root);
Al Viro9b4a9b12009-04-07 11:44:16 -04001669 return ERR_PTR(error);
J. R. Okajima654f5622009-06-18 23:30:15 +09001670 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001671 if (unlikely(!audit_dummy_context()))
1672 audit_inode(pathname, nd.path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673
1674 /*
1675 * We have the parent and last component. First of all, check
1676 * that we are not asked to creat(2) an obvious directory - that
1677 * will not do.
1678 */
1679 error = -EISDIR;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001680 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
Al Viro8737f3a2008-08-02 22:36:57 -04001681 goto exit_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682
Al Viro8737f3a2008-08-02 22:36:57 -04001683 error = -ENFILE;
1684 filp = get_empty_filp();
1685 if (filp == NULL)
1686 goto exit_parent;
1687 nd.intent.open.file = filp;
Al Viro482928d2009-12-19 10:10:39 -05001688 filp->f_flags = open_flag;
Al Viro8737f3a2008-08-02 22:36:57 -04001689 nd.intent.open.flags = flag;
1690 nd.intent.open.create_mode = mode;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001691 dir = nd.path.dentry;
1692 nd.flags &= ~LOOKUP_PARENT;
Al Viro8737f3a2008-08-02 22:36:57 -04001693 nd.flags |= LOOKUP_CREATE | LOOKUP_OPEN;
Al Viro35165862008-08-05 03:00:49 -04001694 if (flag & O_EXCL)
1695 nd.flags |= LOOKUP_EXCL;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001696 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001697 path.dentry = lookup_hash(&nd);
1698 path.mnt = nd.path.mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699
1700do_last:
Al Viro4e7506e2005-06-06 13:36:00 -07001701 error = PTR_ERR(path.dentry);
1702 if (IS_ERR(path.dentry)) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001703 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 goto exit;
1705 }
1706
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001707 if (IS_ERR(nd.intent.open.file)) {
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001708 error = PTR_ERR(nd.intent.open.file);
Dave Hansen4a3fd212008-02-15 14:37:48 -08001709 goto exit_mutex_unlock;
Oleg Drokin4af4c522006-03-25 03:06:54 -08001710 }
1711
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 /* Negative dentry, just create the file */
Al Viro4e7506e2005-06-06 13:36:00 -07001713 if (!path.dentry->d_inode) {
Dave Hansen4a3fd212008-02-15 14:37:48 -08001714 /*
1715 * This write is needed to ensure that a
1716 * ro->rw transition does not occur between
1717 * the time when the file is created and when
1718 * a permanent write count is taken through
1719 * the 'struct file' in nameidata_to_filp().
1720 */
1721 error = mnt_want_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 if (error)
Dave Hansen4a3fd212008-02-15 14:37:48 -08001723 goto exit_mutex_unlock;
1724 error = __open_namei_create(&nd, &path, flag, mode);
1725 if (error) {
1726 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 goto exit;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001728 }
Al Viro482928d2009-12-19 10:10:39 -05001729 filp = nameidata_to_filp(&nd);
Dave Hansen4a3fd212008-02-15 14:37:48 -08001730 mnt_drop_write(nd.path.mnt);
J. R. Okajima654f5622009-06-18 23:30:15 +09001731 if (nd.root.mnt)
1732 path_put(&nd.root);
Al Virob65a9cf2009-12-16 06:27:40 -05001733 if (!IS_ERR(filp)) {
1734 error = ima_path_check(&filp->f_path, filp->f_mode &
Al Viro1429b3e2009-12-16 06:38:01 -05001735 (MAY_READ | MAY_WRITE | MAY_EXEC));
Al Virob65a9cf2009-12-16 06:27:40 -05001736 if (error) {
1737 fput(filp);
1738 filp = ERR_PTR(error);
1739 }
1740 }
Dave Hansen4a3fd212008-02-15 14:37:48 -08001741 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 }
1743
1744 /*
1745 * It already exists.
1746 */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001747 mutex_unlock(&dir->d_inode->i_mutex);
Al Viro5a190ae2007-06-07 12:19:32 -04001748 audit_inode(pathname, path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749
1750 error = -EEXIST;
1751 if (flag & O_EXCL)
1752 goto exit_dput;
1753
Al Viroe13b2102005-06-06 13:36:06 -07001754 if (__follow_mount(&path)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 error = -ELOOP;
Al Viroba7a4c12005-06-06 13:36:08 -07001756 if (flag & O_NOFOLLOW)
1757 goto exit_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 }
Amy Griffis3e2efce2006-07-13 13:16:02 -04001759
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 error = -ENOENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001761 if (!path.dentry->d_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 goto exit_dput;
Al Viroacfa4382008-12-04 10:06:33 -05001763 if (path.dentry->d_inode->i_op->follow_link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 goto do_link;
1765
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001766 path_to_nameidata(&path, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767 error = -EISDIR;
Jeff Laytoncb598612009-11-16 12:05:20 -08001768 if (S_ISDIR(path.dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769 goto exit;
1770ok:
Dave Hansen4a3fd212008-02-15 14:37:48 -08001771 /*
1772 * Consider:
1773 * 1. may_open() truncates a file
1774 * 2. a rw->ro mount transition occurs
1775 * 3. nameidata_to_filp() fails due to
1776 * the ro mount.
1777 * That would be inconsistent, and should
1778 * be avoided. Taking this mnt write here
1779 * ensures that (2) can not occur.
1780 */
Al Viro7715b522009-12-16 03:54:00 -05001781 will_truncate = open_will_truncate(flag, nd.path.dentry->d_inode);
1782 if (will_truncate) {
Dave Hansen4a3fd212008-02-15 14:37:48 -08001783 error = mnt_want_write(nd.path.mnt);
1784 if (error)
1785 goto exit;
1786 }
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001787 error = may_open(&nd.path, acc_mode, flag);
Dave Hansen4a3fd212008-02-15 14:37:48 -08001788 if (error) {
Al Viro7715b522009-12-16 03:54:00 -05001789 if (will_truncate)
Dave Hansen4a3fd212008-02-15 14:37:48 -08001790 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 goto exit;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001792 }
Al Viro482928d2009-12-19 10:10:39 -05001793 filp = nameidata_to_filp(&nd);
Al Virob65a9cf2009-12-16 06:27:40 -05001794 if (!IS_ERR(filp)) {
1795 error = ima_path_check(&filp->f_path, filp->f_mode &
Al Viro1429b3e2009-12-16 06:38:01 -05001796 (MAY_READ | MAY_WRITE | MAY_EXEC));
Al Viro7715b522009-12-16 03:54:00 -05001797 if (error) {
Al Viro7715b522009-12-16 03:54:00 -05001798 fput(filp);
Al Virob65a9cf2009-12-16 06:27:40 -05001799 filp = ERR_PTR(error);
1800 }
1801 }
1802 if (!IS_ERR(filp)) {
1803 if (acc_mode & MAY_WRITE)
1804 vfs_dq_init(nd.path.dentry->d_inode);
1805
1806 if (will_truncate) {
1807 error = handle_truncate(&nd.path);
1808 if (error) {
1809 fput(filp);
1810 filp = ERR_PTR(error);
1811 }
Al Viro7715b522009-12-16 03:54:00 -05001812 }
1813 }
Dave Hansen4a3fd212008-02-15 14:37:48 -08001814 /*
1815 * It is now safe to drop the mnt write
1816 * because the filp has had a write taken
1817 * on its behalf.
1818 */
Al Viro7715b522009-12-16 03:54:00 -05001819 if (will_truncate)
Dave Hansen4a3fd212008-02-15 14:37:48 -08001820 mnt_drop_write(nd.path.mnt);
J. R. Okajima654f5622009-06-18 23:30:15 +09001821 if (nd.root.mnt)
1822 path_put(&nd.root);
Dave Hansen4a3fd212008-02-15 14:37:48 -08001823 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824
Dave Hansen4a3fd212008-02-15 14:37:48 -08001825exit_mutex_unlock:
1826 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827exit_dput:
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001828 path_put_conditional(&path, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829exit:
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001830 if (!IS_ERR(nd.intent.open.file))
1831 release_open_intent(&nd);
Al Viro8737f3a2008-08-02 22:36:57 -04001832exit_parent:
Al Viro2a737872009-04-07 11:49:53 -04001833 if (nd.root.mnt)
1834 path_put(&nd.root);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001835 path_put(&nd.path);
1836 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837
1838do_link:
1839 error = -ELOOP;
1840 if (flag & O_NOFOLLOW)
1841 goto exit_dput;
1842 /*
1843 * This is subtle. Instead of calling do_follow_link() we do the
1844 * thing by hands. The reason is that this way we have zero link_count
1845 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1846 * After that we have the parent and last component, i.e.
1847 * we are in the same situation as after the first path_walk().
1848 * Well, almost - if the last component is normal we get its copy
1849 * stored in nd->last.name and we will have to putname() it when we
1850 * are done. Procfs-like symlinks just set LAST_BIND.
1851 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001852 nd.flags |= LOOKUP_PARENT;
1853 error = security_inode_follow_link(path.dentry, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 if (error)
1855 goto exit_dput;
Al Viro6de88d72009-08-09 01:41:57 +04001856 save = nd.path;
1857 path_get(&save);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001858 error = __do_follow_link(&path, &nd);
Al Viro6de88d72009-08-09 01:41:57 +04001859 if (error == -ESTALE) {
1860 /* nd.path had been dropped */
1861 nd.path = save;
1862 path_get(&nd.path);
1863 nd.flags |= LOOKUP_REVAL;
1864 error = __do_follow_link(&path, &nd);
1865 }
1866 path_put(&save);
Al Viro258fa992009-08-09 01:32:02 +04001867 path_put(&path);
Kirill Korotaevde459212006-07-14 00:23:49 -07001868 if (error) {
1869 /* Does someone understand code flow here? Or it is only
1870 * me so stupid? Anathema to whoever designed this non-sense
1871 * with "intent.open".
1872 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001873 release_open_intent(&nd);
J. R. Okajima654f5622009-06-18 23:30:15 +09001874 if (nd.root.mnt)
1875 path_put(&nd.root);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001876 return ERR_PTR(error);
Kirill Korotaevde459212006-07-14 00:23:49 -07001877 }
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001878 nd.flags &= ~LOOKUP_PARENT;
1879 if (nd.last_type == LAST_BIND)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880 goto ok;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 error = -EISDIR;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001882 if (nd.last_type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883 goto exit;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001884 if (nd.last.name[nd.last.len]) {
1885 __putname(nd.last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 goto exit;
1887 }
1888 error = -ELOOP;
1889 if (count++==32) {
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001890 __putname(nd.last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891 goto exit;
1892 }
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001893 dir = nd.path.dentry;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001894 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001895 path.dentry = lookup_hash(&nd);
1896 path.mnt = nd.path.mnt;
1897 __putname(nd.last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898 goto do_last;
1899}
1900
1901/**
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001902 * filp_open - open file and return file pointer
1903 *
1904 * @filename: path to open
1905 * @flags: open flags as per the open(2) second argument
1906 * @mode: mode for the new file if O_CREAT is set, else ignored
1907 *
1908 * This is the helper to open a file from kernelspace if you really
1909 * have to. But in generally you should not do this, so please move
1910 * along, nothing to see here..
1911 */
1912struct file *filp_open(const char *filename, int flags, int mode)
1913{
Al Viro6e8341a2009-04-06 11:16:22 -04001914 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001915}
1916EXPORT_SYMBOL(filp_open);
1917
1918/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 * lookup_create - lookup a dentry, creating it if it doesn't exist
1920 * @nd: nameidata info
1921 * @is_dir: directory flag
1922 *
1923 * Simple function to lookup and return a dentry and create it
1924 * if it doesn't exist. Is SMP-safe.
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001925 *
Jan Blunck4ac91372008-02-14 19:34:32 -08001926 * Returns with nd->path.dentry->d_inode->i_mutex locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927 */
1928struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1929{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001930 struct dentry *dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931
Jan Blunck4ac91372008-02-14 19:34:32 -08001932 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001933 /*
1934 * Yucky last component or no last component at all?
1935 * (foo/., foo/.., /////)
1936 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 if (nd->last_type != LAST_NORM)
1938 goto fail;
1939 nd->flags &= ~LOOKUP_PARENT;
Al Viro35165862008-08-05 03:00:49 -04001940 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
ASANO Masahiroa6349042006-08-22 20:06:02 -04001941 nd->intent.open.flags = O_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001942
1943 /*
1944 * Do the final lookup.
1945 */
Christoph Hellwig49705b72005-11-08 21:35:06 -08001946 dentry = lookup_hash(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 if (IS_ERR(dentry))
1948 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001949
Al Viroe9baf6e2008-05-15 04:49:12 -04001950 if (dentry->d_inode)
1951 goto eexist;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001952 /*
1953 * Special case - lookup gave negative, but... we had foo/bar/
1954 * From the vfs_mknod() POV we just have a negative dentry -
1955 * all is fine. Let's be bastards - you had / on the end, you've
1956 * been asking for (non-existent) directory. -ENOENT for you.
1957 */
Al Viroe9baf6e2008-05-15 04:49:12 -04001958 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1959 dput(dentry);
1960 dentry = ERR_PTR(-ENOENT);
1961 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 return dentry;
Al Viroe9baf6e2008-05-15 04:49:12 -04001963eexist:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964 dput(dentry);
Al Viroe9baf6e2008-05-15 04:49:12 -04001965 dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966fail:
1967 return dentry;
1968}
Christoph Hellwigf81a0bf2005-05-19 12:26:43 -07001969EXPORT_SYMBOL_GPL(lookup_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970
1971int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1972{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001973 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974
1975 if (error)
1976 return error;
1977
1978 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1979 return -EPERM;
1980
Al Viroacfa4382008-12-04 10:06:33 -05001981 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 return -EPERM;
1983
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07001984 error = devcgroup_inode_mknod(mode, dev);
1985 if (error)
1986 return error;
1987
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 error = security_inode_mknod(dir, dentry, mode, dev);
1989 if (error)
1990 return error;
1991
Jan Kara9e3509e2009-01-26 16:45:12 +01001992 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001994 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001995 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 return error;
1997}
1998
Dave Hansen463c3192008-02-15 14:37:57 -08001999static int may_mknod(mode_t mode)
2000{
2001 switch (mode & S_IFMT) {
2002 case S_IFREG:
2003 case S_IFCHR:
2004 case S_IFBLK:
2005 case S_IFIFO:
2006 case S_IFSOCK:
2007 case 0: /* zero mode translates to S_IFREG */
2008 return 0;
2009 case S_IFDIR:
2010 return -EPERM;
2011 default:
2012 return -EINVAL;
2013 }
2014}
2015
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002016SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2017 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018{
Al Viro2ad94ae2008-07-21 09:32:51 -04002019 int error;
2020 char *tmp;
2021 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 struct nameidata nd;
2023
2024 if (S_ISDIR(mode))
2025 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026
Al Viro2ad94ae2008-07-21 09:32:51 -04002027 error = user_path_parent(dfd, filename, &nd, &tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002029 return error;
2030
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 dentry = lookup_create(&nd, 0);
Dave Hansen463c3192008-02-15 14:37:57 -08002032 if (IS_ERR(dentry)) {
2033 error = PTR_ERR(dentry);
2034 goto out_unlock;
2035 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002036 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002037 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002038 error = may_mknod(mode);
2039 if (error)
2040 goto out_dput;
2041 error = mnt_want_write(nd.path.mnt);
2042 if (error)
2043 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002044 error = security_path_mknod(&nd.path, dentry, mode, dev);
2045 if (error)
2046 goto out_drop_write;
Dave Hansen463c3192008-02-15 14:37:57 -08002047 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 case 0: case S_IFREG:
Jan Blunck4ac91372008-02-14 19:34:32 -08002049 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 break;
2051 case S_IFCHR: case S_IFBLK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002052 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 new_decode_dev(dev));
2054 break;
2055 case S_IFIFO: case S_IFSOCK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002056 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002059out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002060 mnt_drop_write(nd.path.mnt);
2061out_dput:
2062 dput(dentry);
2063out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002064 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002065 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066 putname(tmp);
2067
2068 return error;
2069}
2070
Heiko Carstens3480b252009-01-14 14:14:16 +01002071SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002072{
2073 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2074}
2075
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2077{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002078 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079
2080 if (error)
2081 return error;
2082
Al Viroacfa4382008-12-04 10:06:33 -05002083 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084 return -EPERM;
2085
2086 mode &= (S_IRWXUGO|S_ISVTX);
2087 error = security_inode_mkdir(dir, dentry, mode);
2088 if (error)
2089 return error;
2090
Jan Kara9e3509e2009-01-26 16:45:12 +01002091 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002093 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002094 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 return error;
2096}
2097
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002098SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099{
2100 int error = 0;
2101 char * tmp;
Dave Hansen6902d922006-09-30 23:29:01 -07002102 struct dentry *dentry;
2103 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104
Al Viro2ad94ae2008-07-21 09:32:51 -04002105 error = user_path_parent(dfd, pathname, &nd, &tmp);
2106 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002107 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108
Dave Hansen6902d922006-09-30 23:29:01 -07002109 dentry = lookup_create(&nd, 1);
2110 error = PTR_ERR(dentry);
2111 if (IS_ERR(dentry))
2112 goto out_unlock;
2113
Jan Blunck4ac91372008-02-14 19:34:32 -08002114 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002115 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002116 error = mnt_want_write(nd.path.mnt);
2117 if (error)
2118 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002119 error = security_path_mkdir(&nd.path, dentry, mode);
2120 if (error)
2121 goto out_drop_write;
Jan Blunck4ac91372008-02-14 19:34:32 -08002122 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002123out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002124 mnt_drop_write(nd.path.mnt);
2125out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002126 dput(dentry);
2127out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002128 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002129 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002130 putname(tmp);
2131out_err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132 return error;
2133}
2134
Heiko Carstens3cdad422009-01-14 14:14:22 +01002135SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002136{
2137 return sys_mkdirat(AT_FDCWD, pathname, mode);
2138}
2139
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140/*
2141 * We try to drop the dentry early: we should have
2142 * a usage count of 2 if we're the only user of this
2143 * dentry, and if that is true (possibly after pruning
2144 * the dcache), then we drop the dentry now.
2145 *
2146 * A low-level filesystem can, if it choses, legally
2147 * do a
2148 *
2149 * if (!d_unhashed(dentry))
2150 * return -EBUSY;
2151 *
2152 * if it cannot handle the case of removing a directory
2153 * that is still in use by something else..
2154 */
2155void dentry_unhash(struct dentry *dentry)
2156{
2157 dget(dentry);
Vasily Averindc168422006-12-06 20:37:07 -08002158 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159 spin_lock(&dcache_lock);
2160 spin_lock(&dentry->d_lock);
2161 if (atomic_read(&dentry->d_count) == 2)
2162 __d_drop(dentry);
2163 spin_unlock(&dentry->d_lock);
2164 spin_unlock(&dcache_lock);
2165}
2166
2167int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2168{
2169 int error = may_delete(dir, dentry, 1);
2170
2171 if (error)
2172 return error;
2173
Al Viroacfa4382008-12-04 10:06:33 -05002174 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 return -EPERM;
2176
Jan Kara9e3509e2009-01-26 16:45:12 +01002177 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002179 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 dentry_unhash(dentry);
2181 if (d_mountpoint(dentry))
2182 error = -EBUSY;
2183 else {
2184 error = security_inode_rmdir(dir, dentry);
2185 if (!error) {
2186 error = dir->i_op->rmdir(dir, dentry);
2187 if (!error)
2188 dentry->d_inode->i_flags |= S_DEAD;
2189 }
2190 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002191 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193 d_delete(dentry);
2194 }
2195 dput(dentry);
2196
2197 return error;
2198}
2199
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002200static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201{
2202 int error = 0;
2203 char * name;
2204 struct dentry *dentry;
2205 struct nameidata nd;
2206
Al Viro2ad94ae2008-07-21 09:32:51 -04002207 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002209 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210
2211 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002212 case LAST_DOTDOT:
2213 error = -ENOTEMPTY;
2214 goto exit1;
2215 case LAST_DOT:
2216 error = -EINVAL;
2217 goto exit1;
2218 case LAST_ROOT:
2219 error = -EBUSY;
2220 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002222
2223 nd.flags &= ~LOOKUP_PARENT;
2224
Jan Blunck4ac91372008-02-14 19:34:32 -08002225 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002226 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002228 if (IS_ERR(dentry))
2229 goto exit2;
Dave Hansen06227532008-02-15 14:37:34 -08002230 error = mnt_want_write(nd.path.mnt);
2231 if (error)
2232 goto exit3;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002233 error = security_path_rmdir(&nd.path, dentry);
2234 if (error)
2235 goto exit4;
Jan Blunck4ac91372008-02-14 19:34:32 -08002236 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002237exit4:
Dave Hansen06227532008-02-15 14:37:34 -08002238 mnt_drop_write(nd.path.mnt);
2239exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07002240 dput(dentry);
2241exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08002242 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002244 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245 putname(name);
2246 return error;
2247}
2248
Heiko Carstens3cdad422009-01-14 14:14:22 +01002249SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002250{
2251 return do_rmdir(AT_FDCWD, pathname);
2252}
2253
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254int vfs_unlink(struct inode *dir, struct dentry *dentry)
2255{
2256 int error = may_delete(dir, dentry, 0);
2257
2258 if (error)
2259 return error;
2260
Al Viroacfa4382008-12-04 10:06:33 -05002261 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262 return -EPERM;
2263
Jan Kara9e3509e2009-01-26 16:45:12 +01002264 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002266 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267 if (d_mountpoint(dentry))
2268 error = -EBUSY;
2269 else {
2270 error = security_inode_unlink(dir, dentry);
2271 if (!error)
2272 error = dir->i_op->unlink(dir, dentry);
2273 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002274 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275
2276 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2277 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08002278 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280 }
Robert Love0eeca282005-07-12 17:06:03 -04002281
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282 return error;
2283}
2284
2285/*
2286 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002287 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288 * writeout happening, and we don't want to prevent access to the directory
2289 * while waiting on the I/O.
2290 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002291static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292{
Al Viro2ad94ae2008-07-21 09:32:51 -04002293 int error;
2294 char *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 struct dentry *dentry;
2296 struct nameidata nd;
2297 struct inode *inode = NULL;
2298
Al Viro2ad94ae2008-07-21 09:32:51 -04002299 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002301 return error;
2302
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303 error = -EISDIR;
2304 if (nd.last_type != LAST_NORM)
2305 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002306
2307 nd.flags &= ~LOOKUP_PARENT;
2308
Jan Blunck4ac91372008-02-14 19:34:32 -08002309 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002310 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 error = PTR_ERR(dentry);
2312 if (!IS_ERR(dentry)) {
2313 /* Why not before? Because we want correct error value */
2314 if (nd.last.name[nd.last.len])
2315 goto slashes;
2316 inode = dentry->d_inode;
2317 if (inode)
2318 atomic_inc(&inode->i_count);
Dave Hansen06227532008-02-15 14:37:34 -08002319 error = mnt_want_write(nd.path.mnt);
2320 if (error)
2321 goto exit2;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002322 error = security_path_unlink(&nd.path, dentry);
2323 if (error)
2324 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08002325 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002326exit3:
Dave Hansen06227532008-02-15 14:37:34 -08002327 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328 exit2:
2329 dput(dentry);
2330 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002331 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332 if (inode)
2333 iput(inode); /* truncate the inode here */
2334exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002335 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 putname(name);
2337 return error;
2338
2339slashes:
2340 error = !dentry->d_inode ? -ENOENT :
2341 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2342 goto exit2;
2343}
2344
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002345SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002346{
2347 if ((flag & ~AT_REMOVEDIR) != 0)
2348 return -EINVAL;
2349
2350 if (flag & AT_REMOVEDIR)
2351 return do_rmdir(dfd, pathname);
2352
2353 return do_unlinkat(dfd, pathname);
2354}
2355
Heiko Carstens3480b252009-01-14 14:14:16 +01002356SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002357{
2358 return do_unlinkat(AT_FDCWD, pathname);
2359}
2360
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002361int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002363 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364
2365 if (error)
2366 return error;
2367
Al Viroacfa4382008-12-04 10:06:33 -05002368 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369 return -EPERM;
2370
2371 error = security_inode_symlink(dir, dentry, oldname);
2372 if (error)
2373 return error;
2374
Jan Kara9e3509e2009-01-26 16:45:12 +01002375 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002377 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002378 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379 return error;
2380}
2381
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002382SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2383 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384{
Al Viro2ad94ae2008-07-21 09:32:51 -04002385 int error;
2386 char *from;
2387 char *to;
Dave Hansen6902d922006-09-30 23:29:01 -07002388 struct dentry *dentry;
2389 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390
2391 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04002392 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393 return PTR_ERR(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002394
2395 error = user_path_parent(newdfd, newname, &nd, &to);
2396 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002397 goto out_putname;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398
Dave Hansen6902d922006-09-30 23:29:01 -07002399 dentry = lookup_create(&nd, 0);
2400 error = PTR_ERR(dentry);
2401 if (IS_ERR(dentry))
2402 goto out_unlock;
2403
Dave Hansen75c3f292008-02-15 14:37:45 -08002404 error = mnt_want_write(nd.path.mnt);
2405 if (error)
2406 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002407 error = security_path_symlink(&nd.path, dentry, from);
2408 if (error)
2409 goto out_drop_write;
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002410 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002411out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002412 mnt_drop_write(nd.path.mnt);
2413out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002414 dput(dentry);
2415out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002416 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002417 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002418 putname(to);
2419out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420 putname(from);
2421 return error;
2422}
2423
Heiko Carstens3480b252009-01-14 14:14:16 +01002424SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002425{
2426 return sys_symlinkat(oldname, AT_FDCWD, newname);
2427}
2428
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2430{
2431 struct inode *inode = old_dentry->d_inode;
2432 int error;
2433
2434 if (!inode)
2435 return -ENOENT;
2436
Miklos Szeredia95164d2008-07-30 15:08:48 +02002437 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438 if (error)
2439 return error;
2440
2441 if (dir->i_sb != inode->i_sb)
2442 return -EXDEV;
2443
2444 /*
2445 * A link to an append-only or immutable file cannot be created.
2446 */
2447 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2448 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05002449 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002451 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452 return -EPERM;
2453
2454 error = security_inode_link(old_dentry, dir, new_dentry);
2455 if (error)
2456 return error;
2457
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002458 mutex_lock(&inode->i_mutex);
Jan Kara9e3509e2009-01-26 16:45:12 +01002459 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460 error = dir->i_op->link(old_dentry, dir, new_dentry);
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002461 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002462 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002463 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464 return error;
2465}
2466
2467/*
2468 * Hardlinks are often used in delicate situations. We avoid
2469 * security-related surprises by not following symlinks on the
2470 * newname. --KAB
2471 *
2472 * We don't follow them on the oldname either to be compatible
2473 * with linux 2.0, and to avoid hard-linking to directories
2474 * and other special files. --ADM
2475 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002476SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2477 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478{
2479 struct dentry *new_dentry;
Al Viro2d8f3032008-07-22 09:59:21 -04002480 struct nameidata nd;
2481 struct path old_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482 int error;
Al Viro2ad94ae2008-07-21 09:32:51 -04002483 char *to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484
Ulrich Drepper45c9b112006-06-25 05:49:11 -07002485 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002486 return -EINVAL;
2487
Al Viro2d8f3032008-07-22 09:59:21 -04002488 error = user_path_at(olddfd, oldname,
2489 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2490 &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002492 return error;
2493
2494 error = user_path_parent(newdfd, newname, &nd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495 if (error)
2496 goto out;
2497 error = -EXDEV;
Al Viro2d8f3032008-07-22 09:59:21 -04002498 if (old_path.mnt != nd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002499 goto out_release;
2500 new_dentry = lookup_create(&nd, 0);
2501 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002502 if (IS_ERR(new_dentry))
2503 goto out_unlock;
Dave Hansen75c3f292008-02-15 14:37:45 -08002504 error = mnt_want_write(nd.path.mnt);
2505 if (error)
2506 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002507 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2508 if (error)
2509 goto out_drop_write;
Al Viro2d8f3032008-07-22 09:59:21 -04002510 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002511out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002512 mnt_drop_write(nd.path.mnt);
2513out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002514 dput(new_dentry);
2515out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002516 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517out_release:
Jan Blunck1d957f92008-02-14 19:34:35 -08002518 path_put(&nd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002519 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520out:
Al Viro2d8f3032008-07-22 09:59:21 -04002521 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522
2523 return error;
2524}
2525
Heiko Carstens3480b252009-01-14 14:14:16 +01002526SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002527{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002528 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002529}
2530
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531/*
2532 * The worst of all namespace operations - renaming directory. "Perverted"
2533 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2534 * Problems:
2535 * a) we can get into loop creation. Check is done in is_subdir().
2536 * b) race potential - two innocent renames can create a loop together.
2537 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002538 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539 * story.
2540 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002541 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542 * whether the target exists). Solution: try to be smart with locking
2543 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002544 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545 * move will be locked. Thus we can rank directories by the tree
2546 * (ancestors first) and rank all non-directories after them.
2547 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002548 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549 * HOWEVER, it relies on the assumption that any object with ->lookup()
2550 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2551 * we'd better make sure that there's no link(2) for them.
2552 * d) some filesystems don't support opened-but-unlinked directories,
2553 * either because of layout or because they are not ready to deal with
2554 * all cases correctly. The latter will be fixed (taking this sort of
2555 * stuff into VFS), but the former is not going away. Solution: the same
2556 * trick as in rmdir().
2557 * e) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002558 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002560 * ->i_mutex on parents, which works but leads to some truely excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561 * locking].
2562 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07002563static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2564 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565{
2566 int error = 0;
2567 struct inode *target;
2568
2569 /*
2570 * If we are going to change the parent - check write permissions,
2571 * we'll need to flip '..'.
2572 */
2573 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04002574 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575 if (error)
2576 return error;
2577 }
2578
2579 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2580 if (error)
2581 return error;
2582
2583 target = new_dentry->d_inode;
2584 if (target) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002585 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586 dentry_unhash(new_dentry);
2587 }
2588 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2589 error = -EBUSY;
2590 else
2591 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2592 if (target) {
2593 if (!error)
2594 target->i_flags |= S_DEAD;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002595 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596 if (d_unhashed(new_dentry))
2597 d_rehash(new_dentry);
2598 dput(new_dentry);
2599 }
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002600 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07002601 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2602 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603 return error;
2604}
2605
Adrian Bunk75c96f82005-05-05 16:16:09 -07002606static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2607 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608{
2609 struct inode *target;
2610 int error;
2611
2612 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2613 if (error)
2614 return error;
2615
2616 dget(new_dentry);
2617 target = new_dentry->d_inode;
2618 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002619 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2621 error = -EBUSY;
2622 else
2623 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2624 if (!error) {
Mark Fasheh349457c2006-09-08 14:22:21 -07002625 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626 d_move(old_dentry, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 }
2628 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002629 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630 dput(new_dentry);
2631 return error;
2632}
2633
2634int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2635 struct inode *new_dir, struct dentry *new_dentry)
2636{
2637 int error;
2638 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Robert Love0eeca282005-07-12 17:06:03 -04002639 const char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002640
2641 if (old_dentry->d_inode == new_dentry->d_inode)
2642 return 0;
2643
2644 error = may_delete(old_dir, old_dentry, is_dir);
2645 if (error)
2646 return error;
2647
2648 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02002649 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650 else
2651 error = may_delete(new_dir, new_dentry, is_dir);
2652 if (error)
2653 return error;
2654
Al Viroacfa4382008-12-04 10:06:33 -05002655 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002656 return -EPERM;
2657
Jan Kara9e3509e2009-01-26 16:45:12 +01002658 vfs_dq_init(old_dir);
2659 vfs_dq_init(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660
Robert Love0eeca282005-07-12 17:06:03 -04002661 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2662
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663 if (is_dir)
2664 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2665 else
2666 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2667 if (!error) {
Robert Love0eeca282005-07-12 17:06:03 -04002668 const char *new_name = old_dentry->d_name.name;
John McCutchan89204c42005-08-15 12:13:28 -04002669 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04002670 new_dentry->d_inode, old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671 }
Robert Love0eeca282005-07-12 17:06:03 -04002672 fsnotify_oldname_free(old_name);
2673
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674 return error;
2675}
2676
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002677SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2678 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679{
Al Viro2ad94ae2008-07-21 09:32:51 -04002680 struct dentry *old_dir, *new_dir;
2681 struct dentry *old_dentry, *new_dentry;
2682 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002683 struct nameidata oldnd, newnd;
Al Viro2ad94ae2008-07-21 09:32:51 -04002684 char *from;
2685 char *to;
2686 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687
Al Viro2ad94ae2008-07-21 09:32:51 -04002688 error = user_path_parent(olddfd, oldname, &oldnd, &from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002689 if (error)
2690 goto exit;
2691
Al Viro2ad94ae2008-07-21 09:32:51 -04002692 error = user_path_parent(newdfd, newname, &newnd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693 if (error)
2694 goto exit1;
2695
2696 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08002697 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002698 goto exit2;
2699
Jan Blunck4ac91372008-02-14 19:34:32 -08002700 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002701 error = -EBUSY;
2702 if (oldnd.last_type != LAST_NORM)
2703 goto exit2;
2704
Jan Blunck4ac91372008-02-14 19:34:32 -08002705 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706 if (newnd.last_type != LAST_NORM)
2707 goto exit2;
2708
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002709 oldnd.flags &= ~LOOKUP_PARENT;
2710 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09002711 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002712
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713 trap = lock_rename(new_dir, old_dir);
2714
Christoph Hellwig49705b72005-11-08 21:35:06 -08002715 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716 error = PTR_ERR(old_dentry);
2717 if (IS_ERR(old_dentry))
2718 goto exit3;
2719 /* source must exist */
2720 error = -ENOENT;
2721 if (!old_dentry->d_inode)
2722 goto exit4;
2723 /* unless the source is a directory trailing slashes give -ENOTDIR */
2724 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2725 error = -ENOTDIR;
2726 if (oldnd.last.name[oldnd.last.len])
2727 goto exit4;
2728 if (newnd.last.name[newnd.last.len])
2729 goto exit4;
2730 }
2731 /* source should not be ancestor of target */
2732 error = -EINVAL;
2733 if (old_dentry == trap)
2734 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08002735 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002736 error = PTR_ERR(new_dentry);
2737 if (IS_ERR(new_dentry))
2738 goto exit4;
2739 /* target should not be an ancestor of source */
2740 error = -ENOTEMPTY;
2741 if (new_dentry == trap)
2742 goto exit5;
2743
Dave Hansen9079b1e2008-02-15 14:37:49 -08002744 error = mnt_want_write(oldnd.path.mnt);
2745 if (error)
2746 goto exit5;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002747 error = security_path_rename(&oldnd.path, old_dentry,
2748 &newnd.path, new_dentry);
2749 if (error)
2750 goto exit6;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002751 error = vfs_rename(old_dir->d_inode, old_dentry,
2752 new_dir->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002753exit6:
Dave Hansen9079b1e2008-02-15 14:37:49 -08002754 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002755exit5:
2756 dput(new_dentry);
2757exit4:
2758 dput(old_dentry);
2759exit3:
2760 unlock_rename(new_dir, old_dir);
2761exit2:
Jan Blunck1d957f92008-02-14 19:34:35 -08002762 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002763 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002764exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002765 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002766 putname(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002767exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002768 return error;
2769}
2770
Heiko Carstensa26eab22009-01-14 14:14:17 +01002771SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002772{
2773 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2774}
2775
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2777{
2778 int len;
2779
2780 len = PTR_ERR(link);
2781 if (IS_ERR(link))
2782 goto out;
2783
2784 len = strlen(link);
2785 if (len > (unsigned) buflen)
2786 len = buflen;
2787 if (copy_to_user(buffer, link, len))
2788 len = -EFAULT;
2789out:
2790 return len;
2791}
2792
2793/*
2794 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2795 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2796 * using) it for any given inode is up to filesystem.
2797 */
2798int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2799{
2800 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002801 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07002802 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002803
Linus Torvalds1da177e2005-04-16 15:20:36 -07002804 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002805 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07002806 if (IS_ERR(cookie))
2807 return PTR_ERR(cookie);
2808
2809 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2810 if (dentry->d_inode->i_op->put_link)
2811 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2812 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002813}
2814
2815int vfs_follow_link(struct nameidata *nd, const char *link)
2816{
2817 return __vfs_follow_link(nd, link);
2818}
2819
2820/* get the link contents into pagecache */
2821static char *page_getlink(struct dentry * dentry, struct page **ppage)
2822{
Duane Griffinebd09ab2008-12-19 20:47:12 +00002823 char *kaddr;
2824 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002825 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07002826 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002827 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07002828 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002829 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00002830 kaddr = kmap(page);
2831 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2832 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002833}
2834
2835int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2836{
2837 struct page *page = NULL;
2838 char *s = page_getlink(dentry, &page);
2839 int res = vfs_readlink(dentry,buffer,buflen,s);
2840 if (page) {
2841 kunmap(page);
2842 page_cache_release(page);
2843 }
2844 return res;
2845}
2846
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002847void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002848{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002849 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002850 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002851 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852}
2853
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002854void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002856 struct page *page = cookie;
2857
2858 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002859 kunmap(page);
2860 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002861 }
2862}
2863
Nick Piggin54566b22009-01-04 12:00:53 -08002864/*
2865 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2866 */
2867int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868{
2869 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002870 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002871 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08002872 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002873 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08002874 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2875 if (nofs)
2876 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002877
NeilBrown7e53cac2006-03-25 03:07:57 -08002878retry:
Nick Pigginafddba42007-10-16 01:25:01 -07002879 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08002880 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07002882 goto fail;
2883
Linus Torvalds1da177e2005-04-16 15:20:36 -07002884 kaddr = kmap_atomic(page, KM_USER0);
2885 memcpy(kaddr, symname, len-1);
2886 kunmap_atomic(kaddr, KM_USER0);
Nick Pigginafddba42007-10-16 01:25:01 -07002887
2888 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2889 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002890 if (err < 0)
2891 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07002892 if (err < len-1)
2893 goto retry;
2894
Linus Torvalds1da177e2005-04-16 15:20:36 -07002895 mark_inode_dirty(inode);
2896 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002897fail:
2898 return err;
2899}
2900
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002901int page_symlink(struct inode *inode, const char *symname, int len)
2902{
2903 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08002904 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002905}
2906
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002907const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002908 .readlink = generic_readlink,
2909 .follow_link = page_follow_link_light,
2910 .put_link = page_put_link,
2911};
2912
Al Viro2d8f3032008-07-22 09:59:21 -04002913EXPORT_SYMBOL(user_path_at);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002914EXPORT_SYMBOL(follow_down);
2915EXPORT_SYMBOL(follow_up);
2916EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2917EXPORT_SYMBOL(getname);
2918EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002919EXPORT_SYMBOL(lookup_one_len);
2920EXPORT_SYMBOL(page_follow_link_light);
2921EXPORT_SYMBOL(page_put_link);
2922EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002923EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002924EXPORT_SYMBOL(page_symlink);
2925EXPORT_SYMBOL(page_symlink_inode_operations);
2926EXPORT_SYMBOL(path_lookup);
Al Virod1811462008-08-02 00:49:18 -04002927EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002928EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04002929EXPORT_SYMBOL(inode_permission);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -08002930EXPORT_SYMBOL(file_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002931EXPORT_SYMBOL(unlock_rename);
2932EXPORT_SYMBOL(vfs_create);
2933EXPORT_SYMBOL(vfs_follow_link);
2934EXPORT_SYMBOL(vfs_link);
2935EXPORT_SYMBOL(vfs_mkdir);
2936EXPORT_SYMBOL(vfs_mknod);
2937EXPORT_SYMBOL(generic_permission);
2938EXPORT_SYMBOL(vfs_readlink);
2939EXPORT_SYMBOL(vfs_rename);
2940EXPORT_SYMBOL(vfs_rmdir);
2941EXPORT_SYMBOL(vfs_symlink);
2942EXPORT_SYMBOL(vfs_unlink);
2943EXPORT_SYMBOL(dentry_unhash);
2944EXPORT_SYMBOL(generic_readlink);