blob: 05480a64d7b7acec9446db567b9c3560ff0f5762 [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>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050018#include <linux/export.h>
David S. Miller44696902012-05-23 20:12:50 -070019#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/slab.h>
21#include <linux/fs.h>
22#include <linux/namei.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#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 Torvaldse77819e2011-07-22 19:30:19 -070036#include <linux/posix_acl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include <asm/uaccess.h>
38
Eric Parise81e3f42009-12-04 15:47:36 -050039#include "internal.h"
Al Viroc7105362011-11-24 18:22:03 -050040#include "mount.h"
Eric Parise81e3f42009-12-04 15:47:36 -050041
Linus Torvalds1da177e2005-04-16 15:20:36 -070042/* [Feb-1997 T. Schoebel-Theuer]
43 * Fundamental changes in the pathname lookup mechanisms (namei)
44 * were necessary because of omirr. The reason is that omirr needs
45 * to know the _real_ pathname, not the user-supplied one, in case
46 * of symlinks (and also when transname replacements occur).
47 *
48 * The new code replaces the old recursive symlink resolution with
49 * an iterative one (in case of non-nested symlink chains). It does
50 * this with calls to <fs>_follow_link().
51 * As a side effect, dir_namei(), _namei() and follow_link() are now
52 * replaced with a single function lookup_dentry() that can handle all
53 * the special cases of the former code.
54 *
55 * With the new dcache, the pathname is stored at each inode, at least as
56 * long as the refcount of the inode is positive. As a side effect, the
57 * size of the dcache depends on the inode cache and thus is dynamic.
58 *
59 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
60 * resolution to correspond with current state of the code.
61 *
62 * Note that the symlink resolution is not *completely* iterative.
63 * There is still a significant amount of tail- and mid- recursion in
64 * the algorithm. Also, note that <fs>_readlink() is not used in
65 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
66 * may return different results than <fs>_follow_link(). Many virtual
67 * filesystems (including /proc) exhibit this behavior.
68 */
69
70/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
71 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
72 * and the name already exists in form of a symlink, try to create the new
73 * name indicated by the symlink. The old code always complained that the
74 * name already exists, due to not following the symlink even if its target
75 * is nonexistent. The new semantics affects also mknod() and link() when
Lucas De Marchi25985ed2011-03-30 22:57:33 -030076 * the name is a symlink pointing to a non-existent name.
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 *
78 * I don't know which semantics is the right one, since I have no access
79 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
80 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
81 * "old" one. Personally, I think the new semantics is much more logical.
82 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
83 * file does succeed in both HP-UX and SunOs, but not in Solaris
84 * and in the old Linux semantics.
85 */
86
87/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
88 * semantics. See the comments in "open_namei" and "do_link" below.
89 *
90 * [10-Sep-98 Alan Modra] Another symlink change.
91 */
92
93/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
94 * inside the path - always follow.
95 * in the last component in creation/removal/renaming - never follow.
96 * if LOOKUP_FOLLOW passed - follow.
97 * if the pathname has trailing slashes - follow.
98 * otherwise - don't follow.
99 * (applied in that order).
100 *
101 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
102 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
103 * During the 2.4 we need to fix the userland stuff depending on it -
104 * hopefully we will be able to get rid of that wart in 2.5. So far only
105 * XEmacs seems to be relying on it...
106 */
107/*
108 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800109 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 * any extra contention...
111 */
112
113/* In order to reduce some races, while at the same time doing additional
114 * checking and hopefully speeding things up, we copy filenames to the
115 * kernel data space before using them..
116 *
117 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
118 * PATH_MAX includes the nul terminator --RR.
119 */
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +0100120static char *getname_flags(const char __user *filename, int flags, int *empty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121{
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700122 char *result = __getname(), *err;
123 int len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700125 if (unlikely(!result))
Eric Paris4043cde2012-01-03 14:23:08 -0500126 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700128 len = strncpy_from_user(result, filename, PATH_MAX);
129 err = ERR_PTR(len);
130 if (unlikely(len < 0))
131 goto error;
132
133 /* The empty path is special. */
134 if (unlikely(!len)) {
135 if (empty)
Eric Paris4043cde2012-01-03 14:23:08 -0500136 *empty = 1;
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700137 err = ERR_PTR(-ENOENT);
138 if (!(flags & LOOKUP_EMPTY))
139 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 }
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700141
142 err = ERR_PTR(-ENAMETOOLONG);
143 if (likely(len < PATH_MAX)) {
144 audit_getname(result);
145 return result;
146 }
147
148error:
149 __putname(result);
150 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151}
152
Al Virof52e0c12011-03-14 18:56:51 -0400153char *getname(const char __user * filename)
154{
Linus Torvaldsf7493e52012-03-22 16:10:40 -0700155 return getname_flags(filename, 0, NULL);
Al Virof52e0c12011-03-14 18:56:51 -0400156}
157
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158#ifdef CONFIG_AUDITSYSCALL
159void putname(const char *name)
160{
Al Viro5ac3a9c2006-07-16 06:38:45 -0400161 if (unlikely(!audit_dummy_context()))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 audit_putname(name);
163 else
164 __putname(name);
165}
166EXPORT_SYMBOL(putname);
167#endif
168
Linus Torvaldse77819e2011-07-22 19:30:19 -0700169static int check_acl(struct inode *inode, int mask)
170{
Linus Torvalds84635d62011-07-25 22:47:03 -0700171#ifdef CONFIG_FS_POSIX_ACL
Linus Torvaldse77819e2011-07-22 19:30:19 -0700172 struct posix_acl *acl;
173
Linus Torvaldse77819e2011-07-22 19:30:19 -0700174 if (mask & MAY_NOT_BLOCK) {
Al Viro35678662011-08-02 21:32:13 -0400175 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
176 if (!acl)
Linus Torvaldse77819e2011-07-22 19:30:19 -0700177 return -EAGAIN;
Al Viro35678662011-08-02 21:32:13 -0400178 /* no ->get_acl() calls in RCU mode... */
179 if (acl == ACL_NOT_CACHED)
180 return -ECHILD;
Ari Savolainen206b1d02011-08-06 19:43:07 +0300181 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
Linus Torvaldse77819e2011-07-22 19:30:19 -0700182 }
183
184 acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
185
186 /*
Christoph Hellwig4e34e712011-07-23 17:37:31 +0200187 * A filesystem can force a ACL callback by just never filling the
188 * ACL cache. But normally you'd fill the cache either at inode
189 * instantiation time, or on the first ->get_acl call.
Linus Torvaldse77819e2011-07-22 19:30:19 -0700190 *
Christoph Hellwig4e34e712011-07-23 17:37:31 +0200191 * If the filesystem doesn't have a get_acl() function at all, we'll
192 * just create the negative cache entry.
Linus Torvaldse77819e2011-07-22 19:30:19 -0700193 */
194 if (acl == ACL_NOT_CACHED) {
Christoph Hellwig4e34e712011-07-23 17:37:31 +0200195 if (inode->i_op->get_acl) {
196 acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS);
197 if (IS_ERR(acl))
198 return PTR_ERR(acl);
199 } else {
200 set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
201 return -EAGAIN;
202 }
Linus Torvaldse77819e2011-07-22 19:30:19 -0700203 }
204
205 if (acl) {
206 int error = posix_acl_permission(inode, acl, mask);
207 posix_acl_release(acl);
208 return error;
209 }
Linus Torvalds84635d62011-07-25 22:47:03 -0700210#endif
Linus Torvaldse77819e2011-07-22 19:30:19 -0700211
212 return -EAGAIN;
213}
214
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700215/*
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530216 * This does the basic permission checking
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700217 */
Al Viro7e401452011-06-20 19:12:17 -0400218static int acl_permission_check(struct inode *inode, int mask)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700219{
Linus Torvalds26cf46b2011-05-13 11:51:01 -0700220 unsigned int mode = inode->i_mode;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700221
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -0800222 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700223 mode >>= 6;
224 else {
Linus Torvaldse77819e2011-07-22 19:30:19 -0700225 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
Al Viro7e401452011-06-20 19:12:17 -0400226 int error = check_acl(inode, mask);
Nick Pigginb74c79e2011-01-07 17:49:58 +1100227 if (error != -EAGAIN)
228 return error;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700229 }
230
231 if (in_group_p(inode->i_gid))
232 mode >>= 3;
233 }
234
235 /*
236 * If the DACs are ok we don't need any capability check.
237 */
Al Viro9c2c7032011-06-20 19:06:22 -0400238 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700239 return 0;
240 return -EACCES;
241}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242
243/**
Nick Pigginb74c79e2011-01-07 17:49:58 +1100244 * generic_permission - check for access rights on a Posix-like filesystem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 * @inode: inode to check access rights for
Andreas Gruenbacher8fd90c82011-10-23 23:13:30 +0530246 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 *
248 * Used to check for read/write/execute permissions on a file.
249 * We use "fsuid" for this, letting us set arbitrary permissions
250 * for filesystem access without changing the "normal" uids which
Nick Pigginb74c79e2011-01-07 17:49:58 +1100251 * are used for other things.
252 *
253 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
254 * request cannot be satisfied (eg. requires blocking or too much complexity).
255 * It would then be called again in ref-walk mode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 */
Al Viro2830ba72011-06-20 19:16:29 -0400257int generic_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700259 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260
261 /*
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530262 * Do the basic permission checks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 */
Al Viro7e401452011-06-20 19:12:17 -0400264 ret = acl_permission_check(inode, mask);
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700265 if (ret != -EACCES)
266 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267
Al Virod594e7e2011-06-20 19:55:42 -0400268 if (S_ISDIR(inode->i_mode)) {
269 /* DACs are overridable for directories */
Eric W. Biederman1a48e2a2011-11-14 16:24:06 -0800270 if (inode_capable(inode, CAP_DAC_OVERRIDE))
Al Virod594e7e2011-06-20 19:55:42 -0400271 return 0;
272 if (!(mask & MAY_WRITE))
Eric W. Biederman1a48e2a2011-11-14 16:24:06 -0800273 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
Al Virod594e7e2011-06-20 19:55:42 -0400274 return 0;
275 return -EACCES;
276 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 /*
278 * Read/write DACs are always overridable.
Al Virod594e7e2011-06-20 19:55:42 -0400279 * Executable DACs are overridable when there is
280 * at least one exec bit set.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 */
Al Virod594e7e2011-06-20 19:55:42 -0400282 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
Eric W. Biederman1a48e2a2011-11-14 16:24:06 -0800283 if (inode_capable(inode, CAP_DAC_OVERRIDE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 return 0;
285
286 /*
287 * Searching includes executable on directories, else just read.
288 */
Serge E. Hallyn7ea66002009-12-29 14:50:19 -0600289 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
Al Virod594e7e2011-06-20 19:55:42 -0400290 if (mask == MAY_READ)
Eric W. Biederman1a48e2a2011-11-14 16:24:06 -0800291 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 return 0;
293
294 return -EACCES;
295}
296
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700297/*
298 * We _really_ want to just do "generic_permission()" without
299 * even looking at the inode->i_op values. So we keep a cache
300 * flag in inode->i_opflags, that says "this has not special
301 * permission function, use the fast case".
302 */
303static inline int do_inode_permission(struct inode *inode, int mask)
304{
305 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
306 if (likely(inode->i_op->permission))
307 return inode->i_op->permission(inode, mask);
308
309 /* This gets set once for the inode lifetime */
310 spin_lock(&inode->i_lock);
311 inode->i_opflags |= IOP_FASTPERM;
312 spin_unlock(&inode->i_lock);
313 }
314 return generic_permission(inode, mask);
315}
316
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200317/**
David Howells0bdaea92012-06-25 12:55:46 +0100318 * __inode_permission - Check for access rights to a given inode
319 * @inode: Inode to check permission on
320 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200321 *
David Howells0bdaea92012-06-25 12:55:46 +0100322 * Check for read/write/execute permissions on an inode.
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530323 *
324 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
David Howells0bdaea92012-06-25 12:55:46 +0100325 *
326 * This does not check for a read-only file system. You probably want
327 * inode_permission().
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200328 */
David Howells0bdaea92012-06-25 12:55:46 +0100329int __inode_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330{
Al Viroe6305c42008-07-15 21:03:57 -0400331 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700333 if (unlikely(mask & MAY_WRITE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 /*
335 * Nobody gets write access to an immutable file.
336 */
337 if (IS_IMMUTABLE(inode))
338 return -EACCES;
339 }
340
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700341 retval = do_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 if (retval)
343 return retval;
344
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -0700345 retval = devcgroup_inode_permission(inode, mask);
346 if (retval)
347 return retval;
348
Eric Parisd09ca732010-07-23 11:43:57 -0400349 return security_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350}
351
Al Virof4d6ff82011-06-19 13:14:21 -0400352/**
David Howells0bdaea92012-06-25 12:55:46 +0100353 * sb_permission - Check superblock-level permissions
354 * @sb: Superblock of inode to check permission on
355 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
356 *
357 * Separate out file-system wide checks from inode-specific permission checks.
358 */
359static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
360{
361 if (unlikely(mask & MAY_WRITE)) {
362 umode_t mode = inode->i_mode;
363
364 /* Nobody gets write access to a read-only fs. */
365 if ((sb->s_flags & MS_RDONLY) &&
366 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
367 return -EROFS;
368 }
369 return 0;
370}
371
372/**
373 * inode_permission - Check for access rights to a given inode
374 * @inode: Inode to check permission on
375 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
376 *
377 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
378 * this, letting us set arbitrary permissions for filesystem access without
379 * changing the "normal" UIDs which are used for other things.
380 *
381 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
382 */
383int inode_permission(struct inode *inode, int mask)
384{
385 int retval;
386
387 retval = sb_permission(inode->i_sb, inode, mask);
388 if (retval)
389 return retval;
390 return __inode_permission(inode, mask);
391}
392
393/**
Jan Blunck5dd784d02008-02-14 19:34:38 -0800394 * path_get - get a reference to a path
395 * @path: path to get the reference to
396 *
397 * Given a path increment the reference count to the dentry and the vfsmount.
398 */
399void path_get(struct path *path)
400{
401 mntget(path->mnt);
402 dget(path->dentry);
403}
404EXPORT_SYMBOL(path_get);
405
406/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800407 * path_put - put a reference to a path
408 * @path: path to put the reference to
409 *
410 * Given a path decrement the reference count to the dentry and the vfsmount.
411 */
412void path_put(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413{
Jan Blunck1d957f92008-02-14 19:34:35 -0800414 dput(path->dentry);
415 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416}
Jan Blunck1d957f92008-02-14 19:34:35 -0800417EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418
Al Viro19660af2011-03-25 10:32:48 -0400419/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100420 * Path walking has 2 modes, rcu-walk and ref-walk (see
Al Viro19660af2011-03-25 10:32:48 -0400421 * Documentation/filesystems/path-lookup.txt). In situations when we can't
422 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
423 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
424 * mode. Refcounts are grabbed at the last known good point before rcu-walk
425 * got stuck, so ref-walk may continue from there. If this is not successful
426 * (eg. a seqcount has changed), then failure is returned and it's up to caller
427 * to restart the path walk from the beginning in ref-walk mode.
Nick Piggin31e6b012011-01-07 17:49:52 +1100428 */
Nick Piggin31e6b012011-01-07 17:49:52 +1100429
Al Viro32a79912012-07-18 20:43:19 +0400430static inline void lock_rcu_walk(void)
431{
432 br_read_lock(&vfsmount_lock);
433 rcu_read_lock();
434}
435
436static inline void unlock_rcu_walk(void)
437{
438 rcu_read_unlock();
439 br_read_unlock(&vfsmount_lock);
440}
441
Nick Piggin31e6b012011-01-07 17:49:52 +1100442/**
Al Viro19660af2011-03-25 10:32:48 -0400443 * unlazy_walk - try to switch to ref-walk mode.
444 * @nd: nameidata pathwalk data
445 * @dentry: child of nd->path.dentry or NULL
Randy Dunlap39191622011-01-08 19:36:21 -0800446 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100447 *
Al Viro19660af2011-03-25 10:32:48 -0400448 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
449 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
450 * @nd or NULL. Must be called from rcu-walk context.
Nick Piggin31e6b012011-01-07 17:49:52 +1100451 */
Al Viro19660af2011-03-25 10:32:48 -0400452static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
Nick Piggin31e6b012011-01-07 17:49:52 +1100453{
454 struct fs_struct *fs = current->fs;
455 struct dentry *parent = nd->path.dentry;
Al Viro5b6ca022011-03-09 23:04:47 -0500456 int want_root = 0;
Nick Piggin31e6b012011-01-07 17:49:52 +1100457
458 BUG_ON(!(nd->flags & LOOKUP_RCU));
Al Viro5b6ca022011-03-09 23:04:47 -0500459 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
460 want_root = 1;
Nick Piggin31e6b012011-01-07 17:49:52 +1100461 spin_lock(&fs->lock);
462 if (nd->root.mnt != fs->root.mnt ||
463 nd->root.dentry != fs->root.dentry)
464 goto err_root;
465 }
466 spin_lock(&parent->d_lock);
Al Viro19660af2011-03-25 10:32:48 -0400467 if (!dentry) {
468 if (!__d_rcu_to_refcount(parent, nd->seq))
469 goto err_parent;
470 BUG_ON(nd->inode != parent->d_inode);
471 } else {
Al Viro94c0d4e2011-07-12 21:40:23 -0400472 if (dentry->d_parent != parent)
473 goto err_parent;
Al Viro19660af2011-03-25 10:32:48 -0400474 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
475 if (!__d_rcu_to_refcount(dentry, nd->seq))
476 goto err_child;
477 /*
478 * If the sequence check on the child dentry passed, then
479 * the child has not been removed from its parent. This
480 * means the parent dentry must be valid and able to take
481 * a reference at this point.
482 */
483 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
484 BUG_ON(!parent->d_count);
485 parent->d_count++;
486 spin_unlock(&dentry->d_lock);
487 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100488 spin_unlock(&parent->d_lock);
Al Viro5b6ca022011-03-09 23:04:47 -0500489 if (want_root) {
Nick Piggin31e6b012011-01-07 17:49:52 +1100490 path_get(&nd->root);
491 spin_unlock(&fs->lock);
492 }
493 mntget(nd->path.mnt);
494
Al Viro32a79912012-07-18 20:43:19 +0400495 unlock_rcu_walk();
Nick Piggin31e6b012011-01-07 17:49:52 +1100496 nd->flags &= ~LOOKUP_RCU;
497 return 0;
Al Viro19660af2011-03-25 10:32:48 -0400498
499err_child:
Nick Piggin31e6b012011-01-07 17:49:52 +1100500 spin_unlock(&dentry->d_lock);
Al Viro19660af2011-03-25 10:32:48 -0400501err_parent:
Nick Piggin31e6b012011-01-07 17:49:52 +1100502 spin_unlock(&parent->d_lock);
503err_root:
Al Viro5b6ca022011-03-09 23:04:47 -0500504 if (want_root)
Nick Piggin31e6b012011-01-07 17:49:52 +1100505 spin_unlock(&fs->lock);
506 return -ECHILD;
507}
508
Al Viro4ce16ef32012-06-10 16:10:59 -0400509static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
Nick Piggin34286d62011-01-07 17:49:57 +1100510{
Al Viro4ce16ef32012-06-10 16:10:59 -0400511 return dentry->d_op->d_revalidate(dentry, flags);
Nick Piggin34286d62011-01-07 17:49:57 +1100512}
513
Al Viro9f1fafe2011-03-25 11:00:12 -0400514/**
515 * complete_walk - successful completion of path walk
516 * @nd: pointer nameidata
Jeff Layton39159de2009-12-07 12:01:50 -0500517 *
Al Viro9f1fafe2011-03-25 11:00:12 -0400518 * If we had been in RCU mode, drop out of it and legitimize nd->path.
519 * Revalidate the final result, unless we'd already done that during
520 * the path walk or the filesystem doesn't ask for it. Return 0 on
521 * success, -error on failure. In case of failure caller does not
522 * need to drop nd->path.
Jeff Layton39159de2009-12-07 12:01:50 -0500523 */
Al Viro9f1fafe2011-03-25 11:00:12 -0400524static int complete_walk(struct nameidata *nd)
Jeff Layton39159de2009-12-07 12:01:50 -0500525{
Al Viro16c2cd72011-02-22 15:50:10 -0500526 struct dentry *dentry = nd->path.dentry;
Jeff Layton39159de2009-12-07 12:01:50 -0500527 int status;
Jeff Layton39159de2009-12-07 12:01:50 -0500528
Al Viro9f1fafe2011-03-25 11:00:12 -0400529 if (nd->flags & LOOKUP_RCU) {
530 nd->flags &= ~LOOKUP_RCU;
531 if (!(nd->flags & LOOKUP_ROOT))
532 nd->root.mnt = NULL;
533 spin_lock(&dentry->d_lock);
534 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
535 spin_unlock(&dentry->d_lock);
Al Viro32a79912012-07-18 20:43:19 +0400536 unlock_rcu_walk();
Al Viro9f1fafe2011-03-25 11:00:12 -0400537 return -ECHILD;
538 }
539 BUG_ON(nd->inode != dentry->d_inode);
540 spin_unlock(&dentry->d_lock);
541 mntget(nd->path.mnt);
Al Viro32a79912012-07-18 20:43:19 +0400542 unlock_rcu_walk();
Al Viro9f1fafe2011-03-25 11:00:12 -0400543 }
544
Al Viro16c2cd72011-02-22 15:50:10 -0500545 if (likely(!(nd->flags & LOOKUP_JUMPED)))
Jeff Layton39159de2009-12-07 12:01:50 -0500546 return 0;
547
Al Viro16c2cd72011-02-22 15:50:10 -0500548 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
549 return 0;
550
551 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
552 return 0;
553
554 /* Note: we do not d_invalidate() */
Al Viro4ce16ef32012-06-10 16:10:59 -0400555 status = d_revalidate(dentry, nd->flags);
Jeff Layton39159de2009-12-07 12:01:50 -0500556 if (status > 0)
557 return 0;
558
Al Viro16c2cd72011-02-22 15:50:10 -0500559 if (!status)
Jeff Layton39159de2009-12-07 12:01:50 -0500560 status = -ESTALE;
Al Viro16c2cd72011-02-22 15:50:10 -0500561
Al Viro9f1fafe2011-03-25 11:00:12 -0400562 path_put(&nd->path);
Jeff Layton39159de2009-12-07 12:01:50 -0500563 return status;
564}
565
Al Viro2a737872009-04-07 11:49:53 -0400566static __always_inline void set_root(struct nameidata *nd)
567{
Miklos Szeredif7ad3c62010-08-10 11:41:36 +0200568 if (!nd->root.mnt)
569 get_fs_root(current->fs, &nd->root);
Al Viro2a737872009-04-07 11:49:53 -0400570}
571
Al Viro6de88d72009-08-09 01:41:57 +0400572static int link_path_walk(const char *, struct nameidata *);
573
Nick Piggin31e6b012011-01-07 17:49:52 +1100574static __always_inline void set_root_rcu(struct nameidata *nd)
575{
576 if (!nd->root.mnt) {
577 struct fs_struct *fs = current->fs;
Nick Pigginc28cc362011-01-07 17:49:53 +1100578 unsigned seq;
579
580 do {
581 seq = read_seqcount_begin(&fs->seq);
582 nd->root = fs->root;
Tim Chenc1530012011-04-15 11:39:29 -0700583 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
Nick Pigginc28cc362011-01-07 17:49:53 +1100584 } while (read_seqcount_retry(&fs->seq, seq));
Nick Piggin31e6b012011-01-07 17:49:52 +1100585 }
586}
587
Arjan van de Venf1662352006-01-14 13:21:31 -0800588static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589{
Nick Piggin31e6b012011-01-07 17:49:52 +1100590 int ret;
591
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 if (IS_ERR(link))
593 goto fail;
594
595 if (*link == '/') {
Al Viro2a737872009-04-07 11:49:53 -0400596 set_root(nd);
Jan Blunck1d957f92008-02-14 19:34:35 -0800597 path_put(&nd->path);
Al Viro2a737872009-04-07 11:49:53 -0400598 nd->path = nd->root;
599 path_get(&nd->root);
Al Viro16c2cd72011-02-22 15:50:10 -0500600 nd->flags |= LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100602 nd->inode = nd->path.dentry->d_inode;
Christoph Hellwigb4091d52008-11-05 15:07:21 +0100603
Nick Piggin31e6b012011-01-07 17:49:52 +1100604 ret = link_path_walk(link, nd);
605 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606fail:
Jan Blunck1d957f92008-02-14 19:34:35 -0800607 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 return PTR_ERR(link);
609}
610
Jan Blunck1d957f92008-02-14 19:34:35 -0800611static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700612{
613 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800614 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700615 mntput(path->mnt);
616}
617
Nick Piggin7b9337a2011-01-14 08:42:43 +0000618static inline void path_to_nameidata(const struct path *path,
619 struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700620{
Nick Piggin31e6b012011-01-07 17:49:52 +1100621 if (!(nd->flags & LOOKUP_RCU)) {
622 dput(nd->path.dentry);
623 if (nd->path.mnt != path->mnt)
624 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800625 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100626 nd->path.mnt = path->mnt;
Jan Blunck4ac91372008-02-14 19:34:32 -0800627 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700628}
629
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400630/*
631 * Helper to directly jump to a known parsed path from ->follow_link,
632 * caller must have taken a reference to path beforehand.
633 */
634void nd_jump_link(struct nameidata *nd, struct path *path)
635{
636 path_put(&nd->path);
637
638 nd->path = *path;
639 nd->inode = nd->path.dentry->d_inode;
640 nd->flags |= LOOKUP_JUMPED;
641
642 BUG_ON(nd->inode->i_op->follow_link);
643}
644
Al Viro574197e2011-03-14 22:20:34 -0400645static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
646{
647 struct inode *inode = link->dentry->d_inode;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400648 if (inode->i_op->put_link)
Al Viro574197e2011-03-14 22:20:34 -0400649 inode->i_op->put_link(link->dentry, nd, cookie);
650 path_put(link);
651}
652
Kees Cook800179c2012-07-25 17:29:07 -0700653int sysctl_protected_symlinks __read_mostly = 1;
654int sysctl_protected_hardlinks __read_mostly = 1;
655
656/**
657 * may_follow_link - Check symlink following for unsafe situations
658 * @link: The path of the symlink
659 *
660 * In the case of the sysctl_protected_symlinks sysctl being enabled,
661 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
662 * in a sticky world-writable directory. This is to protect privileged
663 * processes from failing races against path names that may change out
664 * from under them by way of other users creating malicious symlinks.
665 * It will permit symlinks to be followed only when outside a sticky
666 * world-writable directory, or when the uid of the symlink and follower
667 * match, or when the directory owner matches the symlink's owner.
668 *
669 * Returns 0 if following the symlink is allowed, -ve on error.
670 */
671static inline int may_follow_link(struct path *link, struct nameidata *nd)
672{
673 const struct inode *inode;
674 const struct inode *parent;
675
676 if (!sysctl_protected_symlinks)
677 return 0;
678
679 /* Allowed if owner and follower match. */
680 inode = link->dentry->d_inode;
Eric W. Biederman81abe272012-08-03 09:38:08 -0700681 if (uid_eq(current_cred()->fsuid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700682 return 0;
683
684 /* Allowed if parent directory not sticky and world-writable. */
685 parent = nd->path.dentry->d_inode;
686 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
687 return 0;
688
689 /* Allowed if parent directory and link owner match. */
Eric W. Biederman81abe272012-08-03 09:38:08 -0700690 if (uid_eq(parent->i_uid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700691 return 0;
692
693 path_put_conditional(link, nd);
694 path_put(&nd->path);
Kees Cooka51d9ea2012-07-25 17:29:08 -0700695 audit_log_link_denied("follow_link", link);
Kees Cook800179c2012-07-25 17:29:07 -0700696 return -EACCES;
697}
698
699/**
700 * safe_hardlink_source - Check for safe hardlink conditions
701 * @inode: the source inode to hardlink from
702 *
703 * Return false if at least one of the following conditions:
704 * - inode is not a regular file
705 * - inode is setuid
706 * - inode is setgid and group-exec
707 * - access failure for read and write
708 *
709 * Otherwise returns true.
710 */
711static bool safe_hardlink_source(struct inode *inode)
712{
713 umode_t mode = inode->i_mode;
714
715 /* Special files should not get pinned to the filesystem. */
716 if (!S_ISREG(mode))
717 return false;
718
719 /* Setuid files should not get pinned to the filesystem. */
720 if (mode & S_ISUID)
721 return false;
722
723 /* Executable setgid files should not get pinned to the filesystem. */
724 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
725 return false;
726
727 /* Hardlinking to unreadable or unwritable sources is dangerous. */
728 if (inode_permission(inode, MAY_READ | MAY_WRITE))
729 return false;
730
731 return true;
732}
733
734/**
735 * may_linkat - Check permissions for creating a hardlink
736 * @link: the source to hardlink from
737 *
738 * Block hardlink when all of:
739 * - sysctl_protected_hardlinks enabled
740 * - fsuid does not match inode
741 * - hardlink source is unsafe (see safe_hardlink_source() above)
742 * - not CAP_FOWNER
743 *
744 * Returns 0 if successful, -ve on error.
745 */
746static int may_linkat(struct path *link)
747{
748 const struct cred *cred;
749 struct inode *inode;
750
751 if (!sysctl_protected_hardlinks)
752 return 0;
753
754 cred = current_cred();
755 inode = link->dentry->d_inode;
756
757 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
758 * otherwise, it must be a safe source.
759 */
Eric W. Biederman81abe272012-08-03 09:38:08 -0700760 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
Kees Cook800179c2012-07-25 17:29:07 -0700761 capable(CAP_FOWNER))
762 return 0;
763
Kees Cooka51d9ea2012-07-25 17:29:08 -0700764 audit_log_link_denied("linkat", link);
Kees Cook800179c2012-07-25 17:29:07 -0700765 return -EPERM;
766}
767
Al Virodef4af32009-12-26 08:37:05 -0500768static __always_inline int
Al Viro574197e2011-03-14 22:20:34 -0400769follow_link(struct path *link, struct nameidata *nd, void **p)
Ian Kent051d3812006-03-27 01:14:53 -0800770{
Nick Piggin7b9337a2011-01-14 08:42:43 +0000771 struct dentry *dentry = link->dentry;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400772 int error;
773 char *s;
Ian Kent051d3812006-03-27 01:14:53 -0800774
Al Viro844a3912011-02-15 00:38:26 -0500775 BUG_ON(nd->flags & LOOKUP_RCU);
776
Al Viro0e794582011-03-16 02:45:02 -0400777 if (link->mnt == nd->path.mnt)
778 mntget(link->mnt);
779
Al Viro6d7b5aa2012-06-10 04:15:17 -0400780 error = -ELOOP;
781 if (unlikely(current->total_link_count >= 40))
782 goto out_put_nd_path;
783
Al Viro574197e2011-03-14 22:20:34 -0400784 cond_resched();
785 current->total_link_count++;
786
Al Viro68ac1232012-03-15 08:21:57 -0400787 touch_atime(link);
Ian Kent051d3812006-03-27 01:14:53 -0800788 nd_set_link(nd, NULL);
789
Al Viro36f3b4f2011-02-22 21:24:38 -0500790 error = security_inode_follow_link(link->dentry, nd);
Al Viro6d7b5aa2012-06-10 04:15:17 -0400791 if (error)
792 goto out_put_nd_path;
Al Viro36f3b4f2011-02-22 21:24:38 -0500793
Al Viro86acdca12009-12-22 23:45:11 -0500794 nd->last_type = LAST_BIND;
Al Virodef4af32009-12-26 08:37:05 -0500795 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
796 error = PTR_ERR(*p);
Al Viro6d7b5aa2012-06-10 04:15:17 -0400797 if (IS_ERR(*p))
Christoph Hellwig408ef012012-06-18 10:47:03 -0400798 goto out_put_nd_path;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400799
800 error = 0;
801 s = nd_get_link(nd);
802 if (s) {
803 error = __vfs_follow_link(nd, s);
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400804 if (unlikely(error))
805 put_link(nd, link, *p);
Ian Kent051d3812006-03-27 01:14:53 -0800806 }
Al Viro6d7b5aa2012-06-10 04:15:17 -0400807
808 return error;
809
810out_put_nd_path:
811 path_put(&nd->path);
Al Viro6d7b5aa2012-06-10 04:15:17 -0400812 path_put(link);
Ian Kent051d3812006-03-27 01:14:53 -0800813 return error;
814}
815
Nick Piggin31e6b012011-01-07 17:49:52 +1100816static int follow_up_rcu(struct path *path)
817{
Al Viro0714a532011-11-24 22:19:58 -0500818 struct mount *mnt = real_mount(path->mnt);
819 struct mount *parent;
Nick Piggin31e6b012011-01-07 17:49:52 +1100820 struct dentry *mountpoint;
821
Al Viro0714a532011-11-24 22:19:58 -0500822 parent = mnt->mnt_parent;
823 if (&parent->mnt == path->mnt)
Nick Piggin31e6b012011-01-07 17:49:52 +1100824 return 0;
Al Viroa73324d2011-11-24 22:25:07 -0500825 mountpoint = mnt->mnt_mountpoint;
Nick Piggin31e6b012011-01-07 17:49:52 +1100826 path->dentry = mountpoint;
Al Viro0714a532011-11-24 22:19:58 -0500827 path->mnt = &parent->mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +1100828 return 1;
829}
830
David Howellsf015f1262012-06-25 12:55:28 +0100831/*
832 * follow_up - Find the mountpoint of path's vfsmount
833 *
834 * Given a path, find the mountpoint of its source file system.
835 * Replace @path with the path of the mountpoint in the parent mount.
836 * Up is towards /.
837 *
838 * Return 1 if we went up a level and 0 if we were already at the
839 * root.
840 */
Al Virobab77eb2009-04-18 03:26:48 -0400841int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842{
Al Viro0714a532011-11-24 22:19:58 -0500843 struct mount *mnt = real_mount(path->mnt);
844 struct mount *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +1000846
Andi Kleen962830d2012-05-08 13:32:02 +0930847 br_read_lock(&vfsmount_lock);
Al Viro0714a532011-11-24 22:19:58 -0500848 parent = mnt->mnt_parent;
Al Viro3c0a6162012-07-18 17:32:50 +0400849 if (parent == mnt) {
Andi Kleen962830d2012-05-08 13:32:02 +0930850 br_read_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 return 0;
852 }
Al Viro0714a532011-11-24 22:19:58 -0500853 mntget(&parent->mnt);
Al Viroa73324d2011-11-24 22:25:07 -0500854 mountpoint = dget(mnt->mnt_mountpoint);
Andi Kleen962830d2012-05-08 13:32:02 +0930855 br_read_unlock(&vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400856 dput(path->dentry);
857 path->dentry = mountpoint;
858 mntput(path->mnt);
Al Viro0714a532011-11-24 22:19:58 -0500859 path->mnt = &parent->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 return 1;
861}
862
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100863/*
David Howells9875cf82011-01-14 18:45:21 +0000864 * Perform an automount
865 * - return -EISDIR to tell follow_managed() to stop and return the path we
866 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 */
David Howells9875cf82011-01-14 18:45:21 +0000868static int follow_automount(struct path *path, unsigned flags,
869 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +1100870{
David Howells9875cf82011-01-14 18:45:21 +0000871 struct vfsmount *mnt;
David Howellsea5b7782011-01-14 19:10:03 +0000872 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +1100873
David Howells9875cf82011-01-14 18:45:21 +0000874 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
875 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -0700876
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200877 /* We don't want to mount if someone's just doing a stat -
878 * unless they're stat'ing a directory and appended a '/' to
879 * the name.
880 *
881 * We do, however, want to mount if someone wants to open or
882 * create a file of any type under the mountpoint, wants to
883 * traverse through the mountpoint or wants to open the
884 * mounted directory. Also, autofs may mark negative dentries
885 * as being automount points. These will need the attentions
886 * of the daemon to instantiate them before they can be used.
David Howells9875cf82011-01-14 18:45:21 +0000887 */
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200888 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
Linus Torvaldsd94c1772011-09-26 17:44:55 -0700889 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200890 path->dentry->d_inode)
891 return -EISDIR;
892
David Howells9875cf82011-01-14 18:45:21 +0000893 current->total_link_count++;
894 if (current->total_link_count >= 40)
895 return -ELOOP;
896
897 mnt = path->dentry->d_op->d_automount(path);
898 if (IS_ERR(mnt)) {
899 /*
900 * The filesystem is allowed to return -EISDIR here to indicate
901 * it doesn't want to automount. For instance, autofs would do
902 * this so that its userspace daemon can mount on this dentry.
903 *
904 * However, we can only permit this if it's a terminal point in
905 * the path being looked up; if it wasn't then the remainder of
906 * the path is inaccessible and we should say so.
907 */
Al Viro49084c32011-06-25 21:59:52 -0400908 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
David Howells9875cf82011-01-14 18:45:21 +0000909 return -EREMOTE;
910 return PTR_ERR(mnt);
911 }
David Howellsea5b7782011-01-14 19:10:03 +0000912
David Howells9875cf82011-01-14 18:45:21 +0000913 if (!mnt) /* mount collision */
914 return 0;
915
Al Viro8aef1882011-06-16 15:10:06 +0100916 if (!*need_mntput) {
917 /* lock_mount() may release path->mnt on error */
918 mntget(path->mnt);
919 *need_mntput = true;
920 }
Al Viro19a167a2011-01-17 01:35:23 -0500921 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +0000922
David Howellsea5b7782011-01-14 19:10:03 +0000923 switch (err) {
924 case -EBUSY:
925 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -0500926 return 0;
David Howellsea5b7782011-01-14 19:10:03 +0000927 case 0:
Al Viro8aef1882011-06-16 15:10:06 +0100928 path_put(path);
David Howellsea5b7782011-01-14 19:10:03 +0000929 path->mnt = mnt;
930 path->dentry = dget(mnt->mnt_root);
David Howellsea5b7782011-01-14 19:10:03 +0000931 return 0;
Al Viro19a167a2011-01-17 01:35:23 -0500932 default:
933 return err;
David Howellsea5b7782011-01-14 19:10:03 +0000934 }
Al Viro19a167a2011-01-17 01:35:23 -0500935
David Howells9875cf82011-01-14 18:45:21 +0000936}
937
938/*
939 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +0000940 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +0000941 * - Flagged as mountpoint
942 * - Flagged as automount point
943 *
944 * This may only be called in refwalk mode.
945 *
946 * Serialization is taken care of in namespace.c
947 */
948static int follow_managed(struct path *path, unsigned flags)
949{
Al Viro8aef1882011-06-16 15:10:06 +0100950 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
David Howells9875cf82011-01-14 18:45:21 +0000951 unsigned managed;
952 bool need_mntput = false;
Al Viro8aef1882011-06-16 15:10:06 +0100953 int ret = 0;
David Howells9875cf82011-01-14 18:45:21 +0000954
955 /* Given that we're not holding a lock here, we retain the value in a
956 * local variable for each dentry as we look at it so that we don't see
957 * the components of that value change under us */
958 while (managed = ACCESS_ONCE(path->dentry->d_flags),
959 managed &= DCACHE_MANAGED_DENTRY,
960 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +0000961 /* Allow the filesystem to manage the transit without i_mutex
962 * being held. */
963 if (managed & DCACHE_MANAGE_TRANSIT) {
964 BUG_ON(!path->dentry->d_op);
965 BUG_ON(!path->dentry->d_op->d_manage);
Al Viro1aed3e42011-03-18 09:09:02 -0400966 ret = path->dentry->d_op->d_manage(path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +0000967 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +0100968 break;
David Howellscc53ce52011-01-14 18:45:26 +0000969 }
970
David Howells9875cf82011-01-14 18:45:21 +0000971 /* Transit to a mounted filesystem. */
972 if (managed & DCACHE_MOUNTED) {
973 struct vfsmount *mounted = lookup_mnt(path);
974 if (mounted) {
975 dput(path->dentry);
976 if (need_mntput)
977 mntput(path->mnt);
978 path->mnt = mounted;
979 path->dentry = dget(mounted->mnt_root);
980 need_mntput = true;
981 continue;
982 }
983
984 /* Something is mounted on this dentry in another
985 * namespace and/or whatever was mounted there in this
986 * namespace got unmounted before we managed to get the
987 * vfsmount_lock */
988 }
989
990 /* Handle an automount point */
991 if (managed & DCACHE_NEED_AUTOMOUNT) {
992 ret = follow_automount(path, flags, &need_mntput);
993 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +0100994 break;
David Howells9875cf82011-01-14 18:45:21 +0000995 continue;
996 }
997
998 /* We didn't change the current path point */
999 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 }
Al Viro8aef1882011-06-16 15:10:06 +01001001
1002 if (need_mntput && path->mnt == mnt)
1003 mntput(path->mnt);
1004 if (ret == -EISDIR)
1005 ret = 0;
Al Viroa3fbbde2011-11-07 21:21:26 +00001006 return ret < 0 ? ret : need_mntput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007}
1008
David Howellscc53ce52011-01-14 18:45:26 +00001009int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010{
1011 struct vfsmount *mounted;
1012
Al Viro1c755af2009-04-18 14:06:57 -04001013 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -04001015 dput(path->dentry);
1016 mntput(path->mnt);
1017 path->mnt = mounted;
1018 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 return 1;
1020 }
1021 return 0;
1022}
1023
Ian Kent62a73752011-03-25 01:51:02 +08001024static inline bool managed_dentry_might_block(struct dentry *dentry)
1025{
1026 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
1027 dentry->d_op->d_manage(dentry, true) < 0);
1028}
1029
David Howells9875cf82011-01-14 18:45:21 +00001030/*
Al Viro287548e2011-05-27 06:50:06 -04001031 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1032 * we meet a managed dentry that would need blocking.
David Howells9875cf82011-01-14 18:45:21 +00001033 */
1034static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
Al Viro287548e2011-05-27 06:50:06 -04001035 struct inode **inode)
David Howells9875cf82011-01-14 18:45:21 +00001036{
Ian Kent62a73752011-03-25 01:51:02 +08001037 for (;;) {
Al Viroc7105362011-11-24 18:22:03 -05001038 struct mount *mounted;
Ian Kent62a73752011-03-25 01:51:02 +08001039 /*
1040 * Don't forget we might have a non-mountpoint managed dentry
1041 * that wants to block transit.
1042 */
Al Viro287548e2011-05-27 06:50:06 -04001043 if (unlikely(managed_dentry_might_block(path->dentry)))
David Howellsab909112011-01-14 18:46:51 +00001044 return false;
Ian Kent62a73752011-03-25 01:51:02 +08001045
1046 if (!d_mountpoint(path->dentry))
1047 break;
1048
David Howells9875cf82011-01-14 18:45:21 +00001049 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
1050 if (!mounted)
1051 break;
Al Viroc7105362011-11-24 18:22:03 -05001052 path->mnt = &mounted->mnt;
1053 path->dentry = mounted->mnt.mnt_root;
Al Viroa3fbbde2011-11-07 21:21:26 +00001054 nd->flags |= LOOKUP_JUMPED;
David Howells9875cf82011-01-14 18:45:21 +00001055 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
Linus Torvalds59430262011-07-18 15:43:29 -07001056 /*
1057 * Update the inode too. We don't need to re-check the
1058 * dentry sequence number here after this d_inode read,
1059 * because a mount-point is always pinned.
1060 */
1061 *inode = path->dentry->d_inode;
David Howells9875cf82011-01-14 18:45:21 +00001062 }
David Howells9875cf82011-01-14 18:45:21 +00001063 return true;
1064}
1065
Al Virodea39372011-05-27 06:53:39 -04001066static void follow_mount_rcu(struct nameidata *nd)
Al Viro287548e2011-05-27 06:50:06 -04001067{
Al Virodea39372011-05-27 06:53:39 -04001068 while (d_mountpoint(nd->path.dentry)) {
Al Viroc7105362011-11-24 18:22:03 -05001069 struct mount *mounted;
Al Virodea39372011-05-27 06:53:39 -04001070 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
Al Viro287548e2011-05-27 06:50:06 -04001071 if (!mounted)
1072 break;
Al Viroc7105362011-11-24 18:22:03 -05001073 nd->path.mnt = &mounted->mnt;
1074 nd->path.dentry = mounted->mnt.mnt_root;
Al Virodea39372011-05-27 06:53:39 -04001075 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro287548e2011-05-27 06:50:06 -04001076 }
1077}
1078
Nick Piggin31e6b012011-01-07 17:49:52 +11001079static int follow_dotdot_rcu(struct nameidata *nd)
1080{
Nick Piggin31e6b012011-01-07 17:49:52 +11001081 set_root_rcu(nd);
1082
David Howells9875cf82011-01-14 18:45:21 +00001083 while (1) {
Nick Piggin31e6b012011-01-07 17:49:52 +11001084 if (nd->path.dentry == nd->root.dentry &&
1085 nd->path.mnt == nd->root.mnt) {
1086 break;
1087 }
1088 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1089 struct dentry *old = nd->path.dentry;
1090 struct dentry *parent = old->d_parent;
1091 unsigned seq;
1092
1093 seq = read_seqcount_begin(&parent->d_seq);
1094 if (read_seqcount_retry(&old->d_seq, nd->seq))
Al Viroef7562d2011-03-04 14:35:59 -05001095 goto failed;
Nick Piggin31e6b012011-01-07 17:49:52 +11001096 nd->path.dentry = parent;
1097 nd->seq = seq;
1098 break;
1099 }
1100 if (!follow_up_rcu(&nd->path))
1101 break;
1102 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11001103 }
Al Virodea39372011-05-27 06:53:39 -04001104 follow_mount_rcu(nd);
1105 nd->inode = nd->path.dentry->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001106 return 0;
Al Viroef7562d2011-03-04 14:35:59 -05001107
1108failed:
1109 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001110 if (!(nd->flags & LOOKUP_ROOT))
1111 nd->root.mnt = NULL;
Al Viro32a79912012-07-18 20:43:19 +04001112 unlock_rcu_walk();
Al Viroef7562d2011-03-04 14:35:59 -05001113 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001114}
1115
David Howells9875cf82011-01-14 18:45:21 +00001116/*
David Howellscc53ce52011-01-14 18:45:26 +00001117 * Follow down to the covering mount currently visible to userspace. At each
1118 * point, the filesystem owning that dentry may be queried as to whether the
1119 * caller is permitted to proceed or not.
David Howellscc53ce52011-01-14 18:45:26 +00001120 */
Al Viro7cc90cc2011-03-18 09:04:20 -04001121int follow_down(struct path *path)
David Howellscc53ce52011-01-14 18:45:26 +00001122{
1123 unsigned managed;
1124 int ret;
1125
1126 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1127 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1128 /* Allow the filesystem to manage the transit without i_mutex
1129 * being held.
1130 *
1131 * We indicate to the filesystem if someone is trying to mount
1132 * something here. This gives autofs the chance to deny anyone
1133 * other than its daemon the right to mount on its
1134 * superstructure.
1135 *
1136 * The filesystem may sleep at this point.
1137 */
1138 if (managed & DCACHE_MANAGE_TRANSIT) {
1139 BUG_ON(!path->dentry->d_op);
1140 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +00001141 ret = path->dentry->d_op->d_manage(
Al Viro1aed3e42011-03-18 09:09:02 -04001142 path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001143 if (ret < 0)
1144 return ret == -EISDIR ? 0 : ret;
1145 }
1146
1147 /* Transit to a mounted filesystem. */
1148 if (managed & DCACHE_MOUNTED) {
1149 struct vfsmount *mounted = lookup_mnt(path);
1150 if (!mounted)
1151 break;
1152 dput(path->dentry);
1153 mntput(path->mnt);
1154 path->mnt = mounted;
1155 path->dentry = dget(mounted->mnt_root);
1156 continue;
1157 }
1158
1159 /* Don't handle automount points here */
1160 break;
1161 }
1162 return 0;
1163}
1164
1165/*
David Howells9875cf82011-01-14 18:45:21 +00001166 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1167 */
1168static void follow_mount(struct path *path)
1169{
1170 while (d_mountpoint(path->dentry)) {
1171 struct vfsmount *mounted = lookup_mnt(path);
1172 if (!mounted)
1173 break;
1174 dput(path->dentry);
1175 mntput(path->mnt);
1176 path->mnt = mounted;
1177 path->dentry = dget(mounted->mnt_root);
1178 }
1179}
1180
Nick Piggin31e6b012011-01-07 17:49:52 +11001181static void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182{
Al Viro2a737872009-04-07 11:49:53 -04001183 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001184
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 while(1) {
Jan Blunck4ac91372008-02-14 19:34:32 -08001186 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187
Al Viro2a737872009-04-07 11:49:53 -04001188 if (nd->path.dentry == nd->root.dentry &&
1189 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 break;
1191 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001192 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Al Viro3088dd72010-01-30 15:47:29 -05001193 /* rare case of legitimate dget_parent()... */
1194 nd->path.dentry = dget_parent(nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 dput(old);
1196 break;
1197 }
Al Viro3088dd72010-01-30 15:47:29 -05001198 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 }
Al Viro79ed0222009-04-18 13:59:41 -04001201 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001202 nd->inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203}
1204
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205/*
Miklos Szeredibad61182012-03-26 12:54:24 +02001206 * This looks up the name in dcache, possibly revalidates the old dentry and
1207 * allocates a new one if not found or not valid. In the need_lookup argument
1208 * returns whether i_op->lookup is necessary.
1209 *
1210 * dir->d_inode->i_mutex must be held
Nick Pigginbaa03892010-08-18 04:37:31 +10001211 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001212static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
Al Viro201f9562012-06-22 12:42:10 +04001213 unsigned int flags, bool *need_lookup)
Nick Pigginbaa03892010-08-18 04:37:31 +10001214{
Nick Pigginbaa03892010-08-18 04:37:31 +10001215 struct dentry *dentry;
Miklos Szeredibad61182012-03-26 12:54:24 +02001216 int error;
Nick Pigginbaa03892010-08-18 04:37:31 +10001217
Miklos Szeredibad61182012-03-26 12:54:24 +02001218 *need_lookup = false;
1219 dentry = d_lookup(dir, name);
1220 if (dentry) {
1221 if (d_need_lookup(dentry)) {
1222 *need_lookup = true;
1223 } else if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
Al Viro201f9562012-06-22 12:42:10 +04001224 error = d_revalidate(dentry, flags);
Miklos Szeredibad61182012-03-26 12:54:24 +02001225 if (unlikely(error <= 0)) {
1226 if (error < 0) {
1227 dput(dentry);
1228 return ERR_PTR(error);
1229 } else if (!d_invalidate(dentry)) {
1230 dput(dentry);
1231 dentry = NULL;
1232 }
1233 }
1234 }
1235 }
Nick Pigginbaa03892010-08-18 04:37:31 +10001236
Miklos Szeredibad61182012-03-26 12:54:24 +02001237 if (!dentry) {
1238 dentry = d_alloc(dir, name);
1239 if (unlikely(!dentry))
1240 return ERR_PTR(-ENOMEM);
Nick Pigginbaa03892010-08-18 04:37:31 +10001241
Miklos Szeredibad61182012-03-26 12:54:24 +02001242 *need_lookup = true;
Nick Pigginbaa03892010-08-18 04:37:31 +10001243 }
1244 return dentry;
1245}
1246
1247/*
Miklos Szeredibad61182012-03-26 12:54:24 +02001248 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1249 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1250 *
1251 * dir->d_inode->i_mutex must be held
Josef Bacik44396f42011-05-31 11:58:49 -04001252 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001253static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
Al Viro72bd8662012-06-10 17:17:17 -04001254 unsigned int flags)
Josef Bacik44396f42011-05-31 11:58:49 -04001255{
Josef Bacik44396f42011-05-31 11:58:49 -04001256 struct dentry *old;
1257
1258 /* Don't create child dentry for a dead directory. */
Miklos Szeredibad61182012-03-26 12:54:24 +02001259 if (unlikely(IS_DEADDIR(dir))) {
Miklos Szeredie188dc02012-02-03 14:25:18 +01001260 dput(dentry);
Josef Bacik44396f42011-05-31 11:58:49 -04001261 return ERR_PTR(-ENOENT);
Miklos Szeredie188dc02012-02-03 14:25:18 +01001262 }
Josef Bacik44396f42011-05-31 11:58:49 -04001263
Al Viro72bd8662012-06-10 17:17:17 -04001264 old = dir->i_op->lookup(dir, dentry, flags);
Josef Bacik44396f42011-05-31 11:58:49 -04001265 if (unlikely(old)) {
1266 dput(dentry);
1267 dentry = old;
1268 }
1269 return dentry;
1270}
1271
Al Viroa3255542012-03-30 14:41:51 -04001272static struct dentry *__lookup_hash(struct qstr *name,
Al Viro72bd8662012-06-10 17:17:17 -04001273 struct dentry *base, unsigned int flags)
Al Viroa3255542012-03-30 14:41:51 -04001274{
Miklos Szeredibad61182012-03-26 12:54:24 +02001275 bool need_lookup;
Al Viroa3255542012-03-30 14:41:51 -04001276 struct dentry *dentry;
1277
Al Viro72bd8662012-06-10 17:17:17 -04001278 dentry = lookup_dcache(name, base, flags, &need_lookup);
Miklos Szeredibad61182012-03-26 12:54:24 +02001279 if (!need_lookup)
1280 return dentry;
Al Viroa3255542012-03-30 14:41:51 -04001281
Al Viro72bd8662012-06-10 17:17:17 -04001282 return lookup_real(base->d_inode, dentry, flags);
Al Viroa3255542012-03-30 14:41:51 -04001283}
1284
Josef Bacik44396f42011-05-31 11:58:49 -04001285/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 * It's more convoluted than I'd like it to be, but... it's still fairly
1287 * small and for now I'd prefer to have fast path as straight as possible.
1288 * It _is_ time-critical.
1289 */
Miklos Szeredi697f5142012-05-21 17:30:05 +02001290static int lookup_fast(struct nameidata *nd, struct qstr *name,
1291 struct path *path, struct inode **inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292{
Jan Blunck4ac91372008-02-14 19:34:32 -08001293 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001294 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001295 int need_reval = 1;
1296 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001297 int err;
1298
Al Viro3cac2602009-08-13 18:27:43 +04001299 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001300 * Rename seqlock is not required here because in the off chance
1301 * of a false negative due to a concurrent rename, we're going to
1302 * do the non-racy lookup, below.
1303 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001304 if (nd->flags & LOOKUP_RCU) {
1305 unsigned seq;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001306 dentry = __d_lookup_rcu(parent, name, &seq, nd->inode);
Al Viro5a18fff2011-03-11 04:44:53 -05001307 if (!dentry)
1308 goto unlazy;
1309
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001310 /*
1311 * This sequence count validates that the inode matches
1312 * the dentry name information from lookup.
1313 */
1314 *inode = dentry->d_inode;
1315 if (read_seqcount_retry(&dentry->d_seq, seq))
1316 return -ECHILD;
1317
1318 /*
1319 * This sequence count validates that the parent had no
1320 * changes while we did the lookup of the dentry above.
1321 *
1322 * The memory barrier in read_seqcount_begin of child is
1323 * enough, we can use __read_seqcount_retry here.
1324 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001325 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1326 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001327 nd->seq = seq;
Al Viro5a18fff2011-03-11 04:44:53 -05001328
Miklos Szeredifa4ee152012-03-26 12:54:19 +02001329 if (unlikely(d_need_lookup(dentry)))
1330 goto unlazy;
Al Viro24643082011-02-15 01:26:22 -05001331 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
Al Viro4ce16ef32012-06-10 16:10:59 -04001332 status = d_revalidate(dentry, nd->flags);
Al Viro5a18fff2011-03-11 04:44:53 -05001333 if (unlikely(status <= 0)) {
1334 if (status != -ECHILD)
1335 need_reval = 0;
1336 goto unlazy;
1337 }
Al Viro24643082011-02-15 01:26:22 -05001338 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001339 path->mnt = mnt;
1340 path->dentry = dentry;
Al Virod6e9bd22011-05-27 07:03:15 -04001341 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1342 goto unlazy;
1343 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1344 goto unlazy;
1345 return 0;
Al Viro5a18fff2011-03-11 04:44:53 -05001346unlazy:
Al Viro19660af2011-03-25 10:32:48 -04001347 if (unlazy_walk(nd, dentry))
1348 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001349 } else {
1350 dentry = __d_lookup(parent, name);
Nick Piggin31e6b012011-01-07 17:49:52 +11001351 }
Al Viro5a18fff2011-03-11 04:44:53 -05001352
Al Viro81e6f522012-03-30 14:48:04 -04001353 if (unlikely(!dentry))
1354 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001355
Al Viro81e6f522012-03-30 14:48:04 -04001356 if (unlikely(d_need_lookup(dentry))) {
1357 dput(dentry);
1358 goto need_lookup;
Al Viro24643082011-02-15 01:26:22 -05001359 }
Al Viro81e6f522012-03-30 14:48:04 -04001360
Al Viro5a18fff2011-03-11 04:44:53 -05001361 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
Al Viro4ce16ef32012-06-10 16:10:59 -04001362 status = d_revalidate(dentry, nd->flags);
Al Viro5a18fff2011-03-11 04:44:53 -05001363 if (unlikely(status <= 0)) {
1364 if (status < 0) {
1365 dput(dentry);
1366 return status;
1367 }
1368 if (!d_invalidate(dentry)) {
1369 dput(dentry);
Al Viro81e6f522012-03-30 14:48:04 -04001370 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001371 }
1372 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001373
David Howells9875cf82011-01-14 18:45:21 +00001374 path->mnt = mnt;
1375 path->dentry = dentry;
1376 err = follow_managed(path, nd->flags);
Ian Kent89312212011-01-18 12:06:10 +08001377 if (unlikely(err < 0)) {
1378 path_put_conditional(path, nd);
David Howells9875cf82011-01-14 18:45:21 +00001379 return err;
Ian Kent89312212011-01-18 12:06:10 +08001380 }
Al Viroa3fbbde2011-11-07 21:21:26 +00001381 if (err)
1382 nd->flags |= LOOKUP_JUMPED;
David Howells9875cf82011-01-14 18:45:21 +00001383 *inode = path->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 return 0;
Al Viro81e6f522012-03-30 14:48:04 -04001385
1386need_lookup:
Miklos Szeredi697f5142012-05-21 17:30:05 +02001387 return 1;
1388}
1389
1390/* Fast lookup failed, do it the slow way */
1391static int lookup_slow(struct nameidata *nd, struct qstr *name,
1392 struct path *path)
1393{
1394 struct dentry *dentry, *parent;
1395 int err;
1396
1397 parent = nd->path.dentry;
Al Viro81e6f522012-03-30 14:48:04 -04001398 BUG_ON(nd->inode != parent->d_inode);
1399
1400 mutex_lock(&parent->d_inode->i_mutex);
Al Viro72bd8662012-06-10 17:17:17 -04001401 dentry = __lookup_hash(name, parent, nd->flags);
Al Viro81e6f522012-03-30 14:48:04 -04001402 mutex_unlock(&parent->d_inode->i_mutex);
1403 if (IS_ERR(dentry))
1404 return PTR_ERR(dentry);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001405 path->mnt = nd->path.mnt;
1406 path->dentry = dentry;
1407 err = follow_managed(path, nd->flags);
1408 if (unlikely(err < 0)) {
1409 path_put_conditional(path, nd);
1410 return err;
1411 }
1412 if (err)
1413 nd->flags |= LOOKUP_JUMPED;
1414 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415}
1416
Al Viro52094c82011-02-21 21:34:47 -05001417static inline int may_lookup(struct nameidata *nd)
1418{
1419 if (nd->flags & LOOKUP_RCU) {
Al Viro4ad5abb2011-06-20 19:57:03 -04001420 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
Al Viro52094c82011-02-21 21:34:47 -05001421 if (err != -ECHILD)
1422 return err;
Al Viro19660af2011-03-25 10:32:48 -04001423 if (unlazy_walk(nd, NULL))
Al Viro52094c82011-02-21 21:34:47 -05001424 return -ECHILD;
1425 }
Al Viro4ad5abb2011-06-20 19:57:03 -04001426 return inode_permission(nd->inode, MAY_EXEC);
Al Viro52094c82011-02-21 21:34:47 -05001427}
1428
Al Viro9856fa12011-03-04 14:22:06 -05001429static inline int handle_dots(struct nameidata *nd, int type)
1430{
1431 if (type == LAST_DOTDOT) {
1432 if (nd->flags & LOOKUP_RCU) {
1433 if (follow_dotdot_rcu(nd))
1434 return -ECHILD;
1435 } else
1436 follow_dotdot(nd);
1437 }
1438 return 0;
1439}
1440
Al Viro951361f2011-03-04 14:44:37 -05001441static void terminate_walk(struct nameidata *nd)
1442{
1443 if (!(nd->flags & LOOKUP_RCU)) {
1444 path_put(&nd->path);
1445 } else {
1446 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001447 if (!(nd->flags & LOOKUP_ROOT))
1448 nd->root.mnt = NULL;
Al Viro32a79912012-07-18 20:43:19 +04001449 unlock_rcu_walk();
Al Viro951361f2011-03-04 14:44:37 -05001450 }
1451}
1452
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001453/*
1454 * Do we need to follow links? We _really_ want to be able
1455 * to do this check without having to look at inode->i_op,
1456 * so we keep a cache of "no, this doesn't need follow_link"
1457 * for the common case.
1458 */
Linus Torvalds7813b942011-08-07 09:53:20 -07001459static inline int should_follow_link(struct inode *inode, int follow)
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001460{
1461 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1462 if (likely(inode->i_op->follow_link))
1463 return follow;
1464
1465 /* This gets set once for the inode lifetime */
1466 spin_lock(&inode->i_lock);
1467 inode->i_opflags |= IOP_NOFOLLOW;
1468 spin_unlock(&inode->i_lock);
1469 }
1470 return 0;
1471}
1472
Al Viroce57dfc2011-03-13 19:58:58 -04001473static inline int walk_component(struct nameidata *nd, struct path *path,
1474 struct qstr *name, int type, int follow)
1475{
1476 struct inode *inode;
1477 int err;
1478 /*
1479 * "." and ".." are special - ".." especially so because it has
1480 * to be able to know about the current root directory and
1481 * parent relationships.
1482 */
1483 if (unlikely(type != LAST_NORM))
1484 return handle_dots(nd, type);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001485 err = lookup_fast(nd, name, path, &inode);
Al Viroce57dfc2011-03-13 19:58:58 -04001486 if (unlikely(err)) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001487 if (err < 0)
1488 goto out_err;
1489
1490 err = lookup_slow(nd, name, path);
1491 if (err < 0)
1492 goto out_err;
1493
1494 inode = path->dentry->d_inode;
Al Viroce57dfc2011-03-13 19:58:58 -04001495 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001496 err = -ENOENT;
1497 if (!inode)
1498 goto out_path_put;
1499
Linus Torvalds7813b942011-08-07 09:53:20 -07001500 if (should_follow_link(inode, follow)) {
Al Viro19660af2011-03-25 10:32:48 -04001501 if (nd->flags & LOOKUP_RCU) {
1502 if (unlikely(unlazy_walk(nd, path->dentry))) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001503 err = -ECHILD;
1504 goto out_err;
Al Viro19660af2011-03-25 10:32:48 -04001505 }
1506 }
Al Viroce57dfc2011-03-13 19:58:58 -04001507 BUG_ON(inode != path->dentry->d_inode);
1508 return 1;
1509 }
1510 path_to_nameidata(path, nd);
1511 nd->inode = inode;
1512 return 0;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001513
1514out_path_put:
1515 path_to_nameidata(path, nd);
1516out_err:
1517 terminate_walk(nd);
1518 return err;
Al Viroce57dfc2011-03-13 19:58:58 -04001519}
1520
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521/*
Al Virob3563792011-03-14 21:54:55 -04001522 * This limits recursive symlink follows to 8, while
1523 * limiting consecutive symlinks to 40.
1524 *
1525 * Without that kind of total limit, nasty chains of consecutive
1526 * symlinks can cause almost arbitrarily long lookups.
1527 */
1528static inline int nested_symlink(struct path *path, struct nameidata *nd)
1529{
1530 int res;
1531
Al Virob3563792011-03-14 21:54:55 -04001532 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1533 path_put_conditional(path, nd);
1534 path_put(&nd->path);
1535 return -ELOOP;
1536 }
Erez Zadok1a4022f2011-05-21 01:19:59 -04001537 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
Al Virob3563792011-03-14 21:54:55 -04001538
1539 nd->depth++;
1540 current->link_count++;
1541
1542 do {
1543 struct path link = *path;
1544 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04001545
1546 res = follow_link(&link, nd, &cookie);
Al Viro6d7b5aa2012-06-10 04:15:17 -04001547 if (res)
1548 break;
1549 res = walk_component(nd, path, &nd->last,
1550 nd->last_type, LOOKUP_FOLLOW);
Al Viro574197e2011-03-14 22:20:34 -04001551 put_link(nd, &link, cookie);
Al Virob3563792011-03-14 21:54:55 -04001552 } while (res > 0);
1553
1554 current->link_count--;
1555 nd->depth--;
1556 return res;
1557}
1558
1559/*
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001560 * We really don't want to look at inode->i_op->lookup
1561 * when we don't have to. So we keep a cache bit in
1562 * the inode ->i_opflags field that says "yes, we can
1563 * do lookup on this inode".
1564 */
1565static inline int can_lookup(struct inode *inode)
1566{
1567 if (likely(inode->i_opflags & IOP_LOOKUP))
1568 return 1;
1569 if (likely(!inode->i_op->lookup))
1570 return 0;
1571
1572 /* We do this once for the lifetime of the inode */
1573 spin_lock(&inode->i_lock);
1574 inode->i_opflags |= IOP_LOOKUP;
1575 spin_unlock(&inode->i_lock);
1576 return 1;
1577}
1578
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001579/*
1580 * We can do the critical dentry name comparison and hashing
1581 * operations one word at a time, but we are limited to:
1582 *
1583 * - Architectures with fast unaligned word accesses. We could
1584 * do a "get_unaligned()" if this helps and is sufficiently
1585 * fast.
1586 *
1587 * - Little-endian machines (so that we can generate the mask
1588 * of low bytes efficiently). Again, we *could* do a byte
1589 * swapping load on big-endian architectures if that is not
1590 * expensive enough to make the optimization worthless.
1591 *
1592 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1593 * do not trap on the (extremely unlikely) case of a page
1594 * crossing operation.
1595 *
1596 * - Furthermore, we need an efficient 64-bit compile for the
1597 * 64-bit case in order to generate the "number of bytes in
1598 * the final mask". Again, that could be replaced with a
1599 * efficient population count instruction or similar.
1600 */
1601#ifdef CONFIG_DCACHE_WORD_ACCESS
1602
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001603#include <asm/word-at-a-time.h>
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001604
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001605#ifdef CONFIG_64BIT
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001606
1607static inline unsigned int fold_hash(unsigned long hash)
1608{
1609 hash += hash >> (8*sizeof(int));
1610 return hash;
1611}
1612
1613#else /* 32-bit case */
1614
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001615#define fold_hash(x) (x)
1616
1617#endif
1618
1619unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1620{
1621 unsigned long a, mask;
1622 unsigned long hash = 0;
1623
1624 for (;;) {
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001625 a = load_unaligned_zeropad(name);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001626 if (len < sizeof(unsigned long))
1627 break;
1628 hash += a;
Al Virof132c5b2012-03-22 21:59:52 +00001629 hash *= 9;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001630 name += sizeof(unsigned long);
1631 len -= sizeof(unsigned long);
1632 if (!len)
1633 goto done;
1634 }
1635 mask = ~(~0ul << len*8);
1636 hash += mask & a;
1637done:
1638 return fold_hash(hash);
1639}
1640EXPORT_SYMBOL(full_name_hash);
1641
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001642/*
1643 * Calculate the length and hash of the path component, and
1644 * return the length of the component;
1645 */
1646static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1647{
Linus Torvalds36126f82012-05-26 10:43:17 -07001648 unsigned long a, b, adata, bdata, mask, hash, len;
1649 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001650
1651 hash = a = 0;
1652 len = -sizeof(unsigned long);
1653 do {
1654 hash = (hash + a) * 9;
1655 len += sizeof(unsigned long);
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001656 a = load_unaligned_zeropad(name+len);
Linus Torvalds36126f82012-05-26 10:43:17 -07001657 b = a ^ REPEAT_BYTE('/');
1658 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001659
Linus Torvalds36126f82012-05-26 10:43:17 -07001660 adata = prep_zero_mask(a, adata, &constants);
1661 bdata = prep_zero_mask(b, bdata, &constants);
1662
1663 mask = create_zero_mask(adata | bdata);
1664
1665 hash += a & zero_bytemask(mask);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001666 *hashp = fold_hash(hash);
1667
Linus Torvalds36126f82012-05-26 10:43:17 -07001668 return len + find_zero(mask);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001669}
1670
1671#else
1672
Linus Torvalds0145acc2012-03-02 14:32:59 -08001673unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1674{
1675 unsigned long hash = init_name_hash();
1676 while (len--)
1677 hash = partial_name_hash(*name++, hash);
1678 return end_name_hash(hash);
1679}
Linus Torvaldsae942ae2012-03-02 19:40:57 -08001680EXPORT_SYMBOL(full_name_hash);
Linus Torvalds0145acc2012-03-02 14:32:59 -08001681
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001682/*
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001683 * We know there's a real path component here of at least
1684 * one character.
1685 */
1686static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1687{
1688 unsigned long hash = init_name_hash();
1689 unsigned long len = 0, c;
1690
1691 c = (unsigned char)*name;
1692 do {
1693 len++;
1694 hash = partial_name_hash(c, hash);
1695 c = (unsigned char)name[len];
1696 } while (c && c != '/');
1697 *hashp = end_name_hash(hash);
1698 return len;
1699}
1700
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001701#endif
1702
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001703/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01001705 * This is the basic name resolution function, turning a pathname into
1706 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01001708 * Returns 0 and nd will have valid dentry and mnt on success.
1709 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 */
Al Viro6de88d72009-08-09 01:41:57 +04001711static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712{
1713 struct path next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715
1716 while (*name=='/')
1717 name++;
1718 if (!*name)
Al Viro086e1832011-02-22 20:56:27 -05001719 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 /* At this point we know we have a real path component. */
1722 for(;;) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 struct qstr this;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001724 long len;
Al Virofe479a52011-02-22 15:10:03 -05001725 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726
Al Viro52094c82011-02-21 21:34:47 -05001727 err = may_lookup(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 if (err)
1729 break;
1730
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001731 len = hash_name(name, &this.hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732 this.name = name;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001733 this.len = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734
Al Virofe479a52011-02-22 15:10:03 -05001735 type = LAST_NORM;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001736 if (name[0] == '.') switch (len) {
Al Virofe479a52011-02-22 15:10:03 -05001737 case 2:
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001738 if (name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05001739 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05001740 nd->flags |= LOOKUP_JUMPED;
1741 }
Al Virofe479a52011-02-22 15:10:03 -05001742 break;
1743 case 1:
1744 type = LAST_DOT;
1745 }
Al Viro5a202bc2011-03-08 14:17:44 -05001746 if (likely(type == LAST_NORM)) {
1747 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05001748 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05001749 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1750 err = parent->d_op->d_hash(parent, nd->inode,
1751 &this);
1752 if (err < 0)
1753 break;
1754 }
1755 }
Al Virofe479a52011-02-22 15:10:03 -05001756
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001757 if (!name[len])
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 goto last_component;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001759 /*
1760 * If it wasn't NUL, we know it was '/'. Skip that
1761 * slash, and continue until no more slashes.
1762 */
1763 do {
1764 len++;
1765 } while (unlikely(name[len] == '/'));
1766 if (!name[len])
Al Virob3563792011-03-14 21:54:55 -04001767 goto last_component;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001768 name += len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769
Al Viroce57dfc2011-03-13 19:58:58 -04001770 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1771 if (err < 0)
1772 return err;
Al Virofe479a52011-02-22 15:10:03 -05001773
Al Viroce57dfc2011-03-13 19:58:58 -04001774 if (err) {
Al Virob3563792011-03-14 21:54:55 -04001775 err = nested_symlink(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 if (err)
Al Viroa7472ba2011-03-04 14:39:30 -05001777 return err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001778 }
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001779 if (can_lookup(nd->inode))
1780 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781 err = -ENOTDIR;
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001782 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 /* here ends the main loop */
1784
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785last_component:
Al Virob3563792011-03-14 21:54:55 -04001786 nd->last = this;
1787 nd->last_type = type;
Al Viro086e1832011-02-22 20:56:27 -05001788 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 }
Al Viro951361f2011-03-04 14:44:37 -05001790 terminate_walk(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 return err;
1792}
1793
Al Viro70e9b352011-03-05 21:12:22 -05001794static int path_init(int dfd, const char *name, unsigned int flags,
1795 struct nameidata *nd, struct file **fp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001797 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001798 int fput_needed;
1799 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800
1801 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro16c2cd72011-02-22 15:50:10 -05001802 nd->flags = flags | LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 nd->depth = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05001804 if (flags & LOOKUP_ROOT) {
1805 struct inode *inode = nd->root.dentry->d_inode;
Al Viro73d049a2011-03-11 12:08:24 -05001806 if (*name) {
1807 if (!inode->i_op->lookup)
1808 return -ENOTDIR;
1809 retval = inode_permission(inode, MAY_EXEC);
1810 if (retval)
1811 return retval;
1812 }
Al Viro5b6ca022011-03-09 23:04:47 -05001813 nd->path = nd->root;
1814 nd->inode = inode;
1815 if (flags & LOOKUP_RCU) {
Al Viro32a79912012-07-18 20:43:19 +04001816 lock_rcu_walk();
Al Viro5b6ca022011-03-09 23:04:47 -05001817 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1818 } else {
1819 path_get(&nd->path);
1820 }
1821 return 0;
1822 }
1823
Al Viro2a737872009-04-07 11:49:53 -04001824 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 if (*name=='/') {
Al Viroe41f7d42011-02-22 14:02:58 -05001827 if (flags & LOOKUP_RCU) {
Al Viro32a79912012-07-18 20:43:19 +04001828 lock_rcu_walk();
Al Viroe41f7d42011-02-22 14:02:58 -05001829 set_root_rcu(nd);
1830 } else {
1831 set_root(nd);
1832 path_get(&nd->root);
1833 }
Al Viro2a737872009-04-07 11:49:53 -04001834 nd->path = nd->root;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001835 } else if (dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05001836 if (flags & LOOKUP_RCU) {
1837 struct fs_struct *fs = current->fs;
1838 unsigned seq;
1839
Al Viro32a79912012-07-18 20:43:19 +04001840 lock_rcu_walk();
Al Viroe41f7d42011-02-22 14:02:58 -05001841
1842 do {
1843 seq = read_seqcount_begin(&fs->seq);
1844 nd->path = fs->pwd;
1845 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1846 } while (read_seqcount_retry(&fs->seq, seq));
1847 } else {
1848 get_fs_pwd(current->fs, &nd->path);
1849 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001850 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001851 struct dentry *dentry;
1852
Al Viro1abf0c72011-03-13 03:51:11 -04001853 file = fget_raw_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001854 retval = -EBADF;
1855 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001856 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001857
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001858 dentry = file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001859
Al Virof52e0c12011-03-14 18:56:51 -04001860 if (*name) {
1861 retval = -ENOTDIR;
1862 if (!S_ISDIR(dentry->d_inode->i_mode))
1863 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001864
Al Viro4ad5abb2011-06-20 19:57:03 -04001865 retval = inode_permission(dentry->d_inode, MAY_EXEC);
Al Virof52e0c12011-03-14 18:56:51 -04001866 if (retval)
1867 goto fput_fail;
1868 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001869
Jan Blunck5dd784d02008-02-14 19:34:38 -08001870 nd->path = file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05001871 if (flags & LOOKUP_RCU) {
1872 if (fput_needed)
Al Viro70e9b352011-03-05 21:12:22 -05001873 *fp = file;
Al Viroe41f7d42011-02-22 14:02:58 -05001874 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro32a79912012-07-18 20:43:19 +04001875 lock_rcu_walk();
Al Viroe41f7d42011-02-22 14:02:58 -05001876 } else {
1877 path_get(&file->f_path);
1878 fput_light(file, fput_needed);
1879 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880 }
Al Viroe41f7d42011-02-22 14:02:58 -05001881
Nick Piggin31e6b012011-01-07 17:49:52 +11001882 nd->inode = nd->path.dentry->d_inode;
Al Viro9b4a9b12009-04-07 11:44:16 -04001883 return 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001884
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001885fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001886 fput_light(file, fput_needed);
Al Viro9b4a9b12009-04-07 11:44:16 -04001887out_fail:
1888 return retval;
1889}
1890
Al Virobd92d7f2011-03-14 19:54:59 -04001891static inline int lookup_last(struct nameidata *nd, struct path *path)
1892{
1893 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1894 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1895
1896 nd->flags &= ~LOOKUP_PARENT;
1897 return walk_component(nd, path, &nd->last, nd->last_type,
1898 nd->flags & LOOKUP_FOLLOW);
1899}
1900
Al Viro9b4a9b12009-04-07 11:44:16 -04001901/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroee0827c2011-02-21 23:38:09 -05001902static int path_lookupat(int dfd, const char *name,
Al Viro9b4a9b12009-04-07 11:44:16 -04001903 unsigned int flags, struct nameidata *nd)
1904{
Al Viro70e9b352011-03-05 21:12:22 -05001905 struct file *base = NULL;
Al Virobd92d7f2011-03-14 19:54:59 -04001906 struct path path;
1907 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001908
1909 /*
1910 * Path walking is largely split up into 2 different synchronisation
1911 * schemes, rcu-walk and ref-walk (explained in
1912 * Documentation/filesystems/path-lookup.txt). These share much of the
1913 * path walk code, but some things particularly setup, cleanup, and
1914 * following mounts are sufficiently divergent that functions are
1915 * duplicated. Typically there is a function foo(), and its RCU
1916 * analogue, foo_rcu().
1917 *
1918 * -ECHILD is the error number of choice (just to avoid clashes) that
1919 * is returned if some aspect of an rcu-walk fails. Such an error must
1920 * be handled by restarting a traditional ref-walk (which will always
1921 * be able to complete).
1922 */
Al Virobd92d7f2011-03-14 19:54:59 -04001923 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
Al Viroee0827c2011-02-21 23:38:09 -05001924
Al Virobd92d7f2011-03-14 19:54:59 -04001925 if (unlikely(err))
1926 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001927
1928 current->total_link_count = 0;
Al Virobd92d7f2011-03-14 19:54:59 -04001929 err = link_path_walk(name, nd);
1930
1931 if (!err && !(flags & LOOKUP_PARENT)) {
Al Virobd92d7f2011-03-14 19:54:59 -04001932 err = lookup_last(nd, &path);
1933 while (err > 0) {
1934 void *cookie;
1935 struct path link = path;
Kees Cook800179c2012-07-25 17:29:07 -07001936 err = may_follow_link(&link, nd);
1937 if (unlikely(err))
1938 break;
Al Virobd92d7f2011-03-14 19:54:59 -04001939 nd->flags |= LOOKUP_PARENT;
Al Viro574197e2011-03-14 22:20:34 -04001940 err = follow_link(&link, nd, &cookie);
Al Viro6d7b5aa2012-06-10 04:15:17 -04001941 if (err)
1942 break;
1943 err = lookup_last(nd, &path);
Al Viro574197e2011-03-14 22:20:34 -04001944 put_link(nd, &link, cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04001945 }
1946 }
Al Viroee0827c2011-02-21 23:38:09 -05001947
Al Viro9f1fafe2011-03-25 11:00:12 -04001948 if (!err)
1949 err = complete_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04001950
1951 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1952 if (!nd->inode->i_op->lookup) {
1953 path_put(&nd->path);
Al Virobd23a532011-03-23 09:56:30 -04001954 err = -ENOTDIR;
Al Virobd92d7f2011-03-14 19:54:59 -04001955 }
1956 }
Al Viro16c2cd72011-02-22 15:50:10 -05001957
Al Viro70e9b352011-03-05 21:12:22 -05001958 if (base)
1959 fput(base);
Al Viroee0827c2011-02-21 23:38:09 -05001960
Al Viro5b6ca022011-03-09 23:04:47 -05001961 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
Al Viro2a737872009-04-07 11:49:53 -04001962 path_put(&nd->root);
1963 nd->root.mnt = NULL;
1964 }
Al Virobd92d7f2011-03-14 19:54:59 -04001965 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001966}
Nick Piggin31e6b012011-01-07 17:49:52 +11001967
Al Viroee0827c2011-02-21 23:38:09 -05001968static int do_path_lookup(int dfd, const char *name,
1969 unsigned int flags, struct nameidata *nd)
1970{
1971 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1972 if (unlikely(retval == -ECHILD))
1973 retval = path_lookupat(dfd, name, flags, nd);
1974 if (unlikely(retval == -ESTALE))
1975 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11001976
1977 if (likely(!retval)) {
1978 if (unlikely(!audit_dummy_context())) {
1979 if (nd->path.dentry && nd->inode)
1980 audit_inode(name, nd->path.dentry);
1981 }
1982 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001983 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984}
1985
Al Viro79714f72012-06-15 03:01:42 +04001986/* does lookup, returns the object with parent locked */
1987struct dentry *kern_path_locked(const char *name, struct path *path)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001988{
Al Viro79714f72012-06-15 03:01:42 +04001989 struct nameidata nd;
1990 struct dentry *d;
1991 int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd);
1992 if (err)
1993 return ERR_PTR(err);
1994 if (nd.last_type != LAST_NORM) {
1995 path_put(&nd.path);
1996 return ERR_PTR(-EINVAL);
1997 }
1998 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Al Viro1e0ea002012-07-22 23:46:21 +04001999 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
Al Viro79714f72012-06-15 03:01:42 +04002000 if (IS_ERR(d)) {
2001 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2002 path_put(&nd.path);
2003 return d;
2004 }
2005 *path = nd.path;
2006 return d;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002007}
2008
Al Virod1811462008-08-02 00:49:18 -04002009int kern_path(const char *name, unsigned int flags, struct path *path)
2010{
2011 struct nameidata nd;
2012 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
2013 if (!res)
2014 *path = nd.path;
2015 return res;
2016}
2017
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002018/**
2019 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2020 * @dentry: pointer to dentry of the base directory
2021 * @mnt: pointer to vfs mount of the base directory
2022 * @name: pointer to file name
2023 * @flags: lookup flags
Al Viroe0a01242011-06-27 17:00:37 -04002024 * @path: pointer to struct path to fill
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002025 */
2026int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2027 const char *name, unsigned int flags,
Al Viroe0a01242011-06-27 17:00:37 -04002028 struct path *path)
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002029{
Al Viroe0a01242011-06-27 17:00:37 -04002030 struct nameidata nd;
2031 int err;
2032 nd.root.dentry = dentry;
2033 nd.root.mnt = mnt;
2034 BUG_ON(flags & LOOKUP_PARENT);
Al Viro5b6ca022011-03-09 23:04:47 -05002035 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
Al Viroe0a01242011-06-27 17:00:37 -04002036 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
2037 if (!err)
2038 *path = nd.path;
2039 return err;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002040}
2041
James Morris057f6c02007-04-26 00:12:05 -07002042/*
2043 * Restricted form of lookup. Doesn't follow links, single-component only,
2044 * needs parent already locked. Doesn't follow mounts.
2045 * SMP-safe.
2046 */
Adrian Bunka244e162006-03-31 02:32:11 -08002047static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048{
Al Viro72bd8662012-06-10 17:17:17 -04002049 return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050}
2051
Christoph Hellwigeead1912007-10-16 23:25:38 -07002052/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07002053 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07002054 * @name: pathname component to lookup
2055 * @base: base directory to lookup from
2056 * @len: maximum length @len should be interpreted to
2057 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07002058 * Note that this routine is purely a helper for filesystem usage and should
2059 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07002060 * nameidata argument is passed to the filesystem methods and a filesystem
2061 * using this helper needs to be prepared for that.
2062 */
James Morris057f6c02007-04-26 00:12:05 -07002063struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2064{
James Morris057f6c02007-04-26 00:12:05 -07002065 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05002066 unsigned int c;
Miklos Szeredicda309d2012-03-26 12:54:21 +02002067 int err;
James Morris057f6c02007-04-26 00:12:05 -07002068
David Woodhouse2f9092e2009-04-20 23:18:37 +01002069 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2070
Al Viro6a96ba52011-03-07 23:49:20 -05002071 this.name = name;
2072 this.len = len;
Linus Torvalds0145acc2012-03-02 14:32:59 -08002073 this.hash = full_name_hash(name, len);
Al Viro6a96ba52011-03-07 23:49:20 -05002074 if (!len)
2075 return ERR_PTR(-EACCES);
2076
Al Viro6a96ba52011-03-07 23:49:20 -05002077 while (len--) {
2078 c = *(const unsigned char *)name++;
2079 if (c == '/' || c == '\0')
2080 return ERR_PTR(-EACCES);
Al Viro6a96ba52011-03-07 23:49:20 -05002081 }
Al Viro5a202bc2011-03-08 14:17:44 -05002082 /*
2083 * See if the low-level filesystem might want
2084 * to use its own hash..
2085 */
2086 if (base->d_flags & DCACHE_OP_HASH) {
2087 int err = base->d_op->d_hash(base, base->d_inode, &this);
2088 if (err < 0)
2089 return ERR_PTR(err);
2090 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07002091
Miklos Szeredicda309d2012-03-26 12:54:21 +02002092 err = inode_permission(base->d_inode, MAY_EXEC);
2093 if (err)
2094 return ERR_PTR(err);
2095
Al Viro72bd8662012-06-10 17:17:17 -04002096 return __lookup_hash(&this, base, 0);
James Morris057f6c02007-04-26 00:12:05 -07002097}
2098
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002099int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2100 struct path *path, int *empty)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101{
Al Viro2d8f3032008-07-22 09:59:21 -04002102 struct nameidata nd;
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002103 char *tmp = getname_flags(name, flags, empty);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04002106
2107 BUG_ON(flags & LOOKUP_PARENT);
2108
2109 err = do_path_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04002111 if (!err)
2112 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113 }
2114 return err;
2115}
2116
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002117int user_path_at(int dfd, const char __user *name, unsigned flags,
2118 struct path *path)
2119{
Linus Torvaldsf7493e52012-03-22 16:10:40 -07002120 return user_path_at_empty(dfd, name, flags, path, NULL);
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002121}
2122
Al Viro2ad94ae2008-07-21 09:32:51 -04002123static int user_path_parent(int dfd, const char __user *path,
2124 struct nameidata *nd, char **name)
2125{
2126 char *s = getname(path);
2127 int error;
2128
2129 if (IS_ERR(s))
2130 return PTR_ERR(s);
2131
2132 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
2133 if (error)
2134 putname(s);
2135 else
2136 *name = s;
2137
2138 return error;
2139}
2140
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141/*
2142 * It's inline, so penalty for filesystems that don't use sticky bit is
2143 * minimal.
2144 */
2145static inline int check_sticky(struct inode *dir, struct inode *inode)
2146{
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002147 kuid_t fsuid = current_fsuid();
David Howellsda9592e2008-11-14 10:39:05 +11002148
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 if (!(dir->i_mode & S_ISVTX))
2150 return 0;
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002151 if (uid_eq(inode->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 return 0;
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002153 if (uid_eq(dir->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154 return 0;
Eric W. Biederman1a48e2a2011-11-14 16:24:06 -08002155 return !inode_capable(inode, CAP_FOWNER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156}
2157
2158/*
2159 * Check whether we can remove a link victim from directory dir, check
2160 * whether the type of victim is right.
2161 * 1. We can't do it if dir is read-only (done in permission())
2162 * 2. We should have write and exec permissions on dir
2163 * 3. We can't remove anything from append-only dir
2164 * 4. We can't do anything with immutable dir (done in permission())
2165 * 5. If the sticky bit on dir is set we should either
2166 * a. be owner of dir, or
2167 * b. be owner of victim, or
2168 * c. have CAP_FOWNER capability
2169 * 6. If the victim is append-only or immutable we can't do antyhing with
2170 * links pointing to it.
2171 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2172 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2173 * 9. We can't remove a root or mountpoint.
2174 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2175 * nfs_async_unlink().
2176 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08002177static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178{
2179 int error;
2180
2181 if (!victim->d_inode)
2182 return -ENOENT;
2183
2184 BUG_ON(victim->d_parent->d_inode != dir);
Al Virocccc6bb2009-12-25 05:07:33 -05002185 audit_inode_child(victim, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186
Al Virof419a2e2008-07-22 00:07:17 -04002187 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 if (error)
2189 return error;
2190 if (IS_APPEND(dir))
2191 return -EPERM;
2192 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
Hugh Dickinsf9454542008-11-19 15:36:38 -08002193 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 return -EPERM;
2195 if (isdir) {
2196 if (!S_ISDIR(victim->d_inode->i_mode))
2197 return -ENOTDIR;
2198 if (IS_ROOT(victim))
2199 return -EBUSY;
2200 } else if (S_ISDIR(victim->d_inode->i_mode))
2201 return -EISDIR;
2202 if (IS_DEADDIR(dir))
2203 return -ENOENT;
2204 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2205 return -EBUSY;
2206 return 0;
2207}
2208
2209/* Check whether we can create an object with dentry child in directory
2210 * dir.
2211 * 1. We can't do it if child already exists (open has special treatment for
2212 * this case, but since we are inlined it's OK)
2213 * 2. We can't do it if dir is read-only (done in permission())
2214 * 3. We should have write and exec permissions on dir
2215 * 4. We can't do it if dir is immutable (done in permission())
2216 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02002217static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218{
2219 if (child->d_inode)
2220 return -EEXIST;
2221 if (IS_DEADDIR(dir))
2222 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04002223 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224}
2225
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226/*
2227 * p1 and p2 should be directories on the same fs.
2228 */
2229struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2230{
2231 struct dentry *p;
2232
2233 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07002234 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 return NULL;
2236 }
2237
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002238 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002240 p = d_ancestor(p2, p1);
2241 if (p) {
2242 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2243 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2244 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245 }
2246
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002247 p = d_ancestor(p1, p2);
2248 if (p) {
2249 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2250 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2251 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252 }
2253
Ingo Molnarf2eace22006-07-03 00:25:05 -07002254 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2255 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256 return NULL;
2257}
2258
2259void unlock_rename(struct dentry *p1, struct dentry *p2)
2260{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002261 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002263 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002264 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265 }
2266}
2267
Al Viro4acdaf22011-07-26 01:42:34 -04002268int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viro312b63f2012-06-10 18:09:36 -04002269 bool want_excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002271 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 if (error)
2273 return error;
2274
Al Viroacfa4382008-12-04 10:06:33 -05002275 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276 return -EACCES; /* shouldn't it be ENOSYS? */
2277 mode &= S_IALLUGO;
2278 mode |= S_IFREG;
2279 error = security_inode_create(dir, dentry, mode);
2280 if (error)
2281 return error;
Al Viro312b63f2012-06-10 18:09:36 -04002282 error = dir->i_op->create(dir, dentry, mode, want_excl);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002283 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002284 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285 return error;
2286}
2287
Al Viro73d049a2011-03-11 12:08:24 -05002288static int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002290 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291 struct inode *inode = dentry->d_inode;
2292 int error;
2293
Al Virobcda7652011-03-13 16:42:14 -04002294 /* O_PATH? */
2295 if (!acc_mode)
2296 return 0;
2297
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298 if (!inode)
2299 return -ENOENT;
2300
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002301 switch (inode->i_mode & S_IFMT) {
2302 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002304 case S_IFDIR:
2305 if (acc_mode & MAY_WRITE)
2306 return -EISDIR;
2307 break;
2308 case S_IFBLK:
2309 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002310 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002312 /*FALLTHRU*/
2313 case S_IFIFO:
2314 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002316 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002317 }
Dave Hansenb41572e2007-10-16 23:31:14 -07002318
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002319 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07002320 if (error)
2321 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05002322
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323 /*
2324 * An append-only file must be opened in append mode for writing.
2325 */
2326 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05002327 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05002328 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05002330 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331 }
2332
2333 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002334 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05002335 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336
J. Bruce Fieldsf3c7691e2011-09-21 10:58:13 -04002337 return 0;
Al Viro7715b522009-12-16 03:54:00 -05002338}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339
Jeff Laytone1181ee2010-12-07 16:19:50 -05002340static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002341{
Jeff Laytone1181ee2010-12-07 16:19:50 -05002342 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002343 struct inode *inode = path->dentry->d_inode;
2344 int error = get_write_access(inode);
2345 if (error)
2346 return error;
2347 /*
2348 * Refuse to truncate files with mandatory locks held on them.
2349 */
2350 error = locks_verify_locked(inode);
2351 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002352 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002353 if (!error) {
2354 error = do_truncate(path->dentry, 0,
2355 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002356 filp);
Al Viro7715b522009-12-16 03:54:00 -05002357 }
2358 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002359 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360}
2361
Dave Hansend57999e2008-02-15 14:37:27 -08002362static inline int open_to_namei_flags(int flag)
2363{
Al Viro8a5e9292011-06-25 19:15:54 -04002364 if ((flag & O_ACCMODE) == 3)
2365 flag--;
Dave Hansend57999e2008-02-15 14:37:27 -08002366 return flag;
2367}
2368
Miklos Szeredid18e9002012-06-05 15:10:17 +02002369static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2370{
2371 int error = security_path_mknod(dir, dentry, mode, 0);
2372 if (error)
2373 return error;
2374
2375 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2376 if (error)
2377 return error;
2378
2379 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2380}
2381
David Howells1acf0af2012-06-14 16:13:46 +01002382/*
2383 * Attempt to atomically look up, create and open a file from a negative
2384 * dentry.
2385 *
2386 * Returns 0 if successful. The file will have been created and attached to
2387 * @file by the filesystem calling finish_open().
2388 *
2389 * Returns 1 if the file was looked up only or didn't need creating. The
2390 * caller will need to perform the open themselves. @path will have been
2391 * updated to point to the new dentry. This may be negative.
2392 *
2393 * Returns an error code otherwise.
2394 */
Al Viro2675a4e2012-06-22 12:41:10 +04002395static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2396 struct path *path, struct file *file,
2397 const struct open_flags *op,
Al Viro64894cf2012-07-31 00:53:35 +04002398 bool got_write, bool need_lookup,
Al Viro2675a4e2012-06-22 12:41:10 +04002399 int *opened)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002400{
2401 struct inode *dir = nd->path.dentry->d_inode;
2402 unsigned open_flag = open_to_namei_flags(op->open_flag);
2403 umode_t mode;
2404 int error;
2405 int acc_mode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002406 int create_error = 0;
2407 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2408
2409 BUG_ON(dentry->d_inode);
2410
2411 /* Don't create child dentry for a dead directory. */
2412 if (unlikely(IS_DEADDIR(dir))) {
Al Viro2675a4e2012-06-22 12:41:10 +04002413 error = -ENOENT;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002414 goto out;
2415 }
2416
2417 mode = op->mode & S_IALLUGO;
2418 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2419 mode &= ~current_umask();
2420
Al Virof8310c52012-07-30 11:50:30 +04002421 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT)) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002422 open_flag &= ~O_TRUNC;
Al Viro47237682012-06-10 05:01:45 -04002423 *opened |= FILE_CREATED;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002424 }
2425
2426 /*
2427 * Checking write permission is tricky, bacuse we don't know if we are
2428 * going to actually need it: O_CREAT opens should work as long as the
2429 * file exists. But checking existence breaks atomicity. The trick is
2430 * to check access and if not granted clear O_CREAT from the flags.
2431 *
2432 * Another problem is returing the "right" error value (e.g. for an
2433 * O_EXCL open we want to return EEXIST not EROFS).
2434 */
Al Viro64894cf2012-07-31 00:53:35 +04002435 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2436 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2437 if (!(open_flag & O_CREAT)) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002438 /*
2439 * No O_CREATE -> atomicity not a requirement -> fall
2440 * back to lookup + open
2441 */
2442 goto no_open;
2443 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2444 /* Fall back and fail with the right error */
Al Viro64894cf2012-07-31 00:53:35 +04002445 create_error = -EROFS;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002446 goto no_open;
2447 } else {
2448 /* No side effects, safe to clear O_CREAT */
Al Viro64894cf2012-07-31 00:53:35 +04002449 create_error = -EROFS;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002450 open_flag &= ~O_CREAT;
2451 }
2452 }
2453
2454 if (open_flag & O_CREAT) {
2455 error = may_o_create(&nd->path, dentry, op->mode);
2456 if (error) {
2457 create_error = error;
2458 if (open_flag & O_EXCL)
2459 goto no_open;
2460 open_flag &= ~O_CREAT;
2461 }
2462 }
2463
2464 if (nd->flags & LOOKUP_DIRECTORY)
2465 open_flag |= O_DIRECTORY;
2466
Al Viro30d90492012-06-22 12:40:19 +04002467 file->f_path.dentry = DENTRY_NOT_SET;
2468 file->f_path.mnt = nd->path.mnt;
2469 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
Al Viro47237682012-06-10 05:01:45 -04002470 opened);
Al Virod9585272012-06-22 12:39:14 +04002471 if (error < 0) {
Al Virod9585272012-06-22 12:39:14 +04002472 if (create_error && error == -ENOENT)
2473 error = create_error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002474 goto out;
2475 }
2476
2477 acc_mode = op->acc_mode;
Al Viro47237682012-06-10 05:01:45 -04002478 if (*opened & FILE_CREATED) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002479 fsnotify_create(dir, dentry);
2480 acc_mode = MAY_OPEN;
2481 }
2482
Al Virod9585272012-06-22 12:39:14 +04002483 if (error) { /* returned 1, that is */
Al Viro30d90492012-06-22 12:40:19 +04002484 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
Al Viro2675a4e2012-06-22 12:41:10 +04002485 error = -EIO;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002486 goto out;
2487 }
Al Viro30d90492012-06-22 12:40:19 +04002488 if (file->f_path.dentry) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002489 dput(dentry);
Al Viro30d90492012-06-22 12:40:19 +04002490 dentry = file->f_path.dentry;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002491 }
2492 goto looked_up;
2493 }
2494
2495 /*
2496 * We didn't have the inode before the open, so check open permission
2497 * here.
2498 */
Al Viro2675a4e2012-06-22 12:41:10 +04002499 error = may_open(&file->f_path, acc_mode, open_flag);
2500 if (error)
2501 fput(file);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002502
2503out:
2504 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04002505 return error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002506
Miklos Szeredid18e9002012-06-05 15:10:17 +02002507no_open:
2508 if (need_lookup) {
Al Viro72bd8662012-06-10 17:17:17 -04002509 dentry = lookup_real(dir, dentry, nd->flags);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002510 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002511 return PTR_ERR(dentry);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002512
2513 if (create_error) {
2514 int open_flag = op->open_flag;
2515
Al Viro2675a4e2012-06-22 12:41:10 +04002516 error = create_error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002517 if ((open_flag & O_EXCL)) {
2518 if (!dentry->d_inode)
2519 goto out;
2520 } else if (!dentry->d_inode) {
2521 goto out;
2522 } else if ((open_flag & O_TRUNC) &&
2523 S_ISREG(dentry->d_inode->i_mode)) {
2524 goto out;
2525 }
2526 /* will fail later, go on to get the right error */
2527 }
2528 }
2529looked_up:
2530 path->dentry = dentry;
2531 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04002532 return 1;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002533}
2534
Nick Piggin31e6b012011-01-07 17:49:52 +11002535/*
David Howells1acf0af2012-06-14 16:13:46 +01002536 * Look up and maybe create and open the last component.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002537 *
2538 * Must be called with i_mutex held on parent.
2539 *
David Howells1acf0af2012-06-14 16:13:46 +01002540 * Returns 0 if the file was successfully atomically created (if necessary) and
2541 * opened. In this case the file will be returned attached to @file.
2542 *
2543 * Returns 1 if the file was not completely opened at this time, though lookups
2544 * and creations will have been performed and the dentry returned in @path will
2545 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2546 * specified then a negative dentry may be returned.
2547 *
2548 * An error code is returned otherwise.
2549 *
2550 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2551 * cleared otherwise prior to returning.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002552 */
Al Viro2675a4e2012-06-22 12:41:10 +04002553static int lookup_open(struct nameidata *nd, struct path *path,
2554 struct file *file,
2555 const struct open_flags *op,
Al Viro64894cf2012-07-31 00:53:35 +04002556 bool got_write, int *opened)
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002557{
2558 struct dentry *dir = nd->path.dentry;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002559 struct inode *dir_inode = dir->d_inode;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002560 struct dentry *dentry;
2561 int error;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002562 bool need_lookup;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002563
Al Viro47237682012-06-10 05:01:45 -04002564 *opened &= ~FILE_CREATED;
Al Viro201f9562012-06-22 12:42:10 +04002565 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002566 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002567 return PTR_ERR(dentry);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002568
Miklos Szeredid18e9002012-06-05 15:10:17 +02002569 /* Cached positive dentry: will open in f_op->open */
2570 if (!need_lookup && dentry->d_inode)
2571 goto out_no_open;
2572
2573 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
Al Viro64894cf2012-07-31 00:53:35 +04002574 return atomic_open(nd, dentry, path, file, op, got_write,
Al Viro47237682012-06-10 05:01:45 -04002575 need_lookup, opened);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002576 }
2577
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002578 if (need_lookup) {
2579 BUG_ON(dentry->d_inode);
2580
Al Viro72bd8662012-06-10 17:17:17 -04002581 dentry = lookup_real(dir_inode, dentry, nd->flags);
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002582 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002583 return PTR_ERR(dentry);
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002584 }
2585
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002586 /* Negative dentry, just create the file */
2587 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2588 umode_t mode = op->mode;
2589 if (!IS_POSIXACL(dir->d_inode))
2590 mode &= ~current_umask();
2591 /*
2592 * This write is needed to ensure that a
2593 * rw->ro transition does not occur between
2594 * the time when the file is created and when
2595 * a permanent write count is taken through
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002596 * the 'struct file' in finish_open().
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002597 */
Al Viro64894cf2012-07-31 00:53:35 +04002598 if (!got_write) {
2599 error = -EROFS;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002600 goto out_dput;
Al Viro64894cf2012-07-31 00:53:35 +04002601 }
Al Viro47237682012-06-10 05:01:45 -04002602 *opened |= FILE_CREATED;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002603 error = security_path_mknod(&nd->path, dentry, mode, 0);
2604 if (error)
2605 goto out_dput;
Al Viro312b63f2012-06-10 18:09:36 -04002606 error = vfs_create(dir->d_inode, dentry, mode,
2607 nd->flags & LOOKUP_EXCL);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002608 if (error)
2609 goto out_dput;
2610 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02002611out_no_open:
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002612 path->dentry = dentry;
2613 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04002614 return 1;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002615
2616out_dput:
2617 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04002618 return error;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002619}
2620
2621/*
Al Virofe2d35f2011-03-05 22:58:25 -05002622 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11002623 */
Al Viro2675a4e2012-06-22 12:41:10 +04002624static int do_last(struct nameidata *nd, struct path *path,
2625 struct file *file, const struct open_flags *op,
2626 int *opened, const char *pathname)
Al Virofb1cc552009-12-24 01:58:28 -05002627{
Al Viroa1e28032009-12-24 02:12:06 -05002628 struct dentry *dir = nd->path.dentry;
Al Viroca344a892011-03-09 00:36:45 -05002629 int open_flag = op->open_flag;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002630 bool will_truncate = (open_flag & O_TRUNC) != 0;
Al Viro64894cf2012-07-31 00:53:35 +04002631 bool got_write = false;
Al Virobcda7652011-03-13 16:42:14 -04002632 int acc_mode = op->acc_mode;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02002633 struct inode *inode;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002634 bool symlink_ok = false;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002635 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2636 bool retried = false;
Al Viro16c2cd72011-02-22 15:50:10 -05002637 int error;
Al Virofb1cc552009-12-24 01:58:28 -05002638
Al Viroc3e380b2011-02-23 13:39:45 -05002639 nd->flags &= ~LOOKUP_PARENT;
2640 nd->flags |= op->intent;
2641
Al Viro1f36f772009-12-26 10:56:19 -05002642 switch (nd->last_type) {
2643 case LAST_DOTDOT:
Neil Brown176306f2010-05-24 16:57:56 +10002644 case LAST_DOT:
Al Virofe2d35f2011-03-05 22:58:25 -05002645 error = handle_dots(nd, nd->last_type);
2646 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002647 return error;
Al Viro1f36f772009-12-26 10:56:19 -05002648 /* fallthrough */
Al Viro1f36f772009-12-26 10:56:19 -05002649 case LAST_ROOT:
Al Viro9f1fafe2011-03-25 11:00:12 -04002650 error = complete_walk(nd);
Al Viro16c2cd72011-02-22 15:50:10 -05002651 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002652 return error;
Al Virofe2d35f2011-03-05 22:58:25 -05002653 audit_inode(pathname, nd->path.dentry);
Al Viroca344a892011-03-09 00:36:45 -05002654 if (open_flag & O_CREAT) {
Al Virofe2d35f2011-03-05 22:58:25 -05002655 error = -EISDIR;
Al Viro2675a4e2012-06-22 12:41:10 +04002656 goto out;
Al Virofe2d35f2011-03-05 22:58:25 -05002657 }
Miklos Szeredie83db162012-06-05 15:10:29 +02002658 goto finish_open;
Al Viro1f36f772009-12-26 10:56:19 -05002659 case LAST_BIND:
Al Viro9f1fafe2011-03-25 11:00:12 -04002660 error = complete_walk(nd);
Al Viro16c2cd72011-02-22 15:50:10 -05002661 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002662 return error;
Al Viro1f36f772009-12-26 10:56:19 -05002663 audit_inode(pathname, dir);
Miklos Szeredie83db162012-06-05 15:10:29 +02002664 goto finish_open;
Al Viro1f36f772009-12-26 10:56:19 -05002665 }
Al Viro67ee3ad2009-12-26 07:01:01 -05002666
Al Viroca344a892011-03-09 00:36:45 -05002667 if (!(open_flag & O_CREAT)) {
Al Virofe2d35f2011-03-05 22:58:25 -05002668 if (nd->last.name[nd->last.len])
2669 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
Al Virobcda7652011-03-13 16:42:14 -04002670 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002671 symlink_ok = true;
Al Virofe2d35f2011-03-05 22:58:25 -05002672 /* we _can_ be in RCU mode here */
Miklos Szeredia1eb3312012-05-21 17:30:07 +02002673 error = lookup_fast(nd, &nd->last, path, &inode);
Miklos Szeredi71574862012-06-05 15:10:14 +02002674 if (likely(!error))
2675 goto finish_lookup;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02002676
Miklos Szeredi71574862012-06-05 15:10:14 +02002677 if (error < 0)
Al Viro2675a4e2012-06-22 12:41:10 +04002678 goto out;
Miklos Szeredi37d7fff2012-06-05 15:10:12 +02002679
Miklos Szeredi71574862012-06-05 15:10:14 +02002680 BUG_ON(nd->inode != dir->d_inode);
Miklos Szeredib6183df2012-06-05 15:10:13 +02002681 } else {
2682 /* create side of things */
2683 /*
2684 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2685 * has been cleared when we got to the last component we are
2686 * about to look up
2687 */
2688 error = complete_walk(nd);
2689 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002690 return error;
Miklos Szeredib6183df2012-06-05 15:10:13 +02002691
2692 audit_inode(pathname, dir);
2693 error = -EISDIR;
2694 /* trailing slashes? */
2695 if (nd->last.name[nd->last.len])
Al Viro2675a4e2012-06-22 12:41:10 +04002696 goto out;
Al Virofe2d35f2011-03-05 22:58:25 -05002697 }
2698
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002699retry_lookup:
Al Viro64894cf2012-07-31 00:53:35 +04002700 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2701 error = mnt_want_write(nd->path.mnt);
2702 if (!error)
2703 got_write = true;
2704 /*
2705 * do _not_ fail yet - we might not need that or fail with
2706 * a different error; let lookup_open() decide; we'll be
2707 * dropping this one anyway.
2708 */
2709 }
Al Viroa1e28032009-12-24 02:12:06 -05002710 mutex_lock(&dir->d_inode->i_mutex);
Al Viro64894cf2012-07-31 00:53:35 +04002711 error = lookup_open(nd, path, file, op, got_write, opened);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002712 mutex_unlock(&dir->d_inode->i_mutex);
Al Viroa1e28032009-12-24 02:12:06 -05002713
Al Viro2675a4e2012-06-22 12:41:10 +04002714 if (error <= 0) {
2715 if (error)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002716 goto out;
2717
Al Viro47237682012-06-10 05:01:45 -04002718 if ((*opened & FILE_CREATED) ||
Al Viro2675a4e2012-06-22 12:41:10 +04002719 !S_ISREG(file->f_path.dentry->d_inode->i_mode))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002720 will_truncate = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002721
Al Viro2675a4e2012-06-22 12:41:10 +04002722 audit_inode(pathname, file->f_path.dentry);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002723 goto opened;
2724 }
Al Virofb1cc552009-12-24 01:58:28 -05002725
Al Viro47237682012-06-10 05:01:45 -04002726 if (*opened & FILE_CREATED) {
Al Viro9b44f1b2011-03-09 00:17:27 -05002727 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05002728 open_flag &= ~O_TRUNC;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002729 will_truncate = false;
Al Virobcda7652011-03-13 16:42:14 -04002730 acc_mode = MAY_OPEN;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002731 path_to_nameidata(path, nd);
Miklos Szeredie83db162012-06-05 15:10:29 +02002732 goto finish_open_created;
Al Virofb1cc552009-12-24 01:58:28 -05002733 }
2734
2735 /*
Jeff Layton3134f372012-07-25 10:19:47 -04002736 * create/update audit record if it already exists.
Al Virofb1cc552009-12-24 01:58:28 -05002737 */
Jeff Layton3134f372012-07-25 10:19:47 -04002738 if (path->dentry->d_inode)
2739 audit_inode(pathname, path->dentry);
Al Virofb1cc552009-12-24 01:58:28 -05002740
Miklos Szeredid18e9002012-06-05 15:10:17 +02002741 /*
2742 * If atomic_open() acquired write access it is dropped now due to
2743 * possible mount and symlink following (this might be optimized away if
2744 * necessary...)
2745 */
Al Viro64894cf2012-07-31 00:53:35 +04002746 if (got_write) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002747 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04002748 got_write = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002749 }
2750
Al Virofb1cc552009-12-24 01:58:28 -05002751 error = -EEXIST;
Al Virof8310c52012-07-30 11:50:30 +04002752 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
Al Virofb1cc552009-12-24 01:58:28 -05002753 goto exit_dput;
2754
David Howells9875cf82011-01-14 18:45:21 +00002755 error = follow_managed(path, nd->flags);
2756 if (error < 0)
2757 goto exit_dput;
Al Virofb1cc552009-12-24 01:58:28 -05002758
Al Viroa3fbbde2011-11-07 21:21:26 +00002759 if (error)
2760 nd->flags |= LOOKUP_JUMPED;
2761
Miklos Szeredidecf3402012-05-21 17:30:08 +02002762 BUG_ON(nd->flags & LOOKUP_RCU);
2763 inode = path->dentry->d_inode;
Miklos Szeredi5f5daac2012-05-21 17:30:14 +02002764finish_lookup:
2765 /* we _can_ be in RCU mode here */
Al Virofb1cc552009-12-24 01:58:28 -05002766 error = -ENOENT;
Miklos Szeredi54c33e72012-05-21 17:30:10 +02002767 if (!inode) {
2768 path_to_nameidata(path, nd);
Al Viro2675a4e2012-06-22 12:41:10 +04002769 goto out;
Miklos Szeredi54c33e72012-05-21 17:30:10 +02002770 }
Al Viro9e67f362009-12-26 07:04:50 -05002771
Miklos Szeredid45ea862012-05-21 17:30:09 +02002772 if (should_follow_link(inode, !symlink_ok)) {
2773 if (nd->flags & LOOKUP_RCU) {
2774 if (unlikely(unlazy_walk(nd, path->dentry))) {
2775 error = -ECHILD;
Al Viro2675a4e2012-06-22 12:41:10 +04002776 goto out;
Miklos Szeredid45ea862012-05-21 17:30:09 +02002777 }
2778 }
2779 BUG_ON(inode != path->dentry->d_inode);
Al Viro2675a4e2012-06-22 12:41:10 +04002780 return 1;
Miklos Szeredid45ea862012-05-21 17:30:09 +02002781 }
Al Virofb1cc552009-12-24 01:58:28 -05002782
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002783 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
2784 path_to_nameidata(path, nd);
2785 } else {
2786 save_parent.dentry = nd->path.dentry;
2787 save_parent.mnt = mntget(path->mnt);
2788 nd->path.dentry = path->dentry;
2789
2790 }
Miklos Szeredidecf3402012-05-21 17:30:08 +02002791 nd->inode = inode;
Al Viroa3fbbde2011-11-07 21:21:26 +00002792 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2793 error = complete_walk(nd);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002794 if (error) {
2795 path_put(&save_parent);
Al Viro2675a4e2012-06-22 12:41:10 +04002796 return error;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002797 }
Al Virofb1cc552009-12-24 01:58:28 -05002798 error = -EISDIR;
Miklos Szeredi050ac842012-05-21 17:30:12 +02002799 if ((open_flag & O_CREAT) && S_ISDIR(nd->inode->i_mode))
Al Viro2675a4e2012-06-22 12:41:10 +04002800 goto out;
Miklos Szerediaf2f5542012-05-21 17:30:11 +02002801 error = -ENOTDIR;
2802 if ((nd->flags & LOOKUP_DIRECTORY) && !nd->inode->i_op->lookup)
Al Viro2675a4e2012-06-22 12:41:10 +04002803 goto out;
Miklos Szeredid7fdd7f2012-05-21 17:30:13 +02002804 audit_inode(pathname, nd->path.dentry);
Miklos Szeredie83db162012-06-05 15:10:29 +02002805finish_open:
Al Viro6c0d46c2011-03-09 00:59:59 -05002806 if (!S_ISREG(nd->inode->i_mode))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002807 will_truncate = false;
Al Viro6c0d46c2011-03-09 00:59:59 -05002808
Al Viro0f9d1a12011-03-09 00:13:14 -05002809 if (will_truncate) {
2810 error = mnt_want_write(nd->path.mnt);
2811 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002812 goto out;
Al Viro64894cf2012-07-31 00:53:35 +04002813 got_write = true;
Al Viro0f9d1a12011-03-09 00:13:14 -05002814 }
Miklos Szeredie83db162012-06-05 15:10:29 +02002815finish_open_created:
Al Virobcda7652011-03-13 16:42:14 -04002816 error = may_open(&nd->path, acc_mode, open_flag);
Al Viroca344a892011-03-09 00:36:45 -05002817 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002818 goto out;
Al Viro30d90492012-06-22 12:40:19 +04002819 file->f_path.mnt = nd->path.mnt;
2820 error = finish_open(file, nd->path.dentry, NULL, opened);
2821 if (error) {
Al Viro30d90492012-06-22 12:40:19 +04002822 if (error == -EOPENSTALE)
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02002823 goto stale_open;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002824 goto out;
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02002825 }
Miklos Szeredia8277b92012-06-05 15:10:32 +02002826opened:
Al Viro2675a4e2012-06-22 12:41:10 +04002827 error = open_check_o_direct(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002828 if (error)
2829 goto exit_fput;
Al Viro2675a4e2012-06-22 12:41:10 +04002830 error = ima_file_check(file, op->acc_mode);
Miklos Szerediaa4caad2012-06-05 15:10:28 +02002831 if (error)
2832 goto exit_fput;
2833
2834 if (will_truncate) {
Al Viro2675a4e2012-06-22 12:41:10 +04002835 error = handle_truncate(file);
Miklos Szerediaa4caad2012-06-05 15:10:28 +02002836 if (error)
2837 goto exit_fput;
Al Viro0f9d1a12011-03-09 00:13:14 -05002838 }
Al Viroca344a892011-03-09 00:36:45 -05002839out:
Al Viro64894cf2012-07-31 00:53:35 +04002840 if (got_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05002841 mnt_drop_write(nd->path.mnt);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002842 path_put(&save_parent);
Miklos Szeredie276ae62012-05-21 17:30:06 +02002843 terminate_walk(nd);
Al Viro2675a4e2012-06-22 12:41:10 +04002844 return error;
Al Virofb1cc552009-12-24 01:58:28 -05002845
Al Virofb1cc552009-12-24 01:58:28 -05002846exit_dput:
2847 path_put_conditional(path, nd);
Al Viroca344a892011-03-09 00:36:45 -05002848 goto out;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002849exit_fput:
Al Viro2675a4e2012-06-22 12:41:10 +04002850 fput(file);
2851 goto out;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002852
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02002853stale_open:
2854 /* If no saved parent or already retried then can't retry */
2855 if (!save_parent.dentry || retried)
2856 goto out;
2857
2858 BUG_ON(save_parent.dentry != dir);
2859 path_put(&nd->path);
2860 nd->path = save_parent;
2861 nd->inode = dir->d_inode;
2862 save_parent.mnt = NULL;
2863 save_parent.dentry = NULL;
Al Viro64894cf2012-07-31 00:53:35 +04002864 if (got_write) {
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02002865 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04002866 got_write = false;
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02002867 }
2868 retried = true;
2869 goto retry_lookup;
Al Virofb1cc552009-12-24 01:58:28 -05002870}
2871
Al Viro13aab422011-02-23 17:54:08 -05002872static struct file *path_openat(int dfd, const char *pathname,
Al Viro73d049a2011-03-11 12:08:24 -05002873 struct nameidata *nd, const struct open_flags *op, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002874{
Al Virofe2d35f2011-03-05 22:58:25 -05002875 struct file *base = NULL;
Al Viro30d90492012-06-22 12:40:19 +04002876 struct file *file;
Al Viro9850c052010-01-13 15:01:15 -05002877 struct path path;
Al Viro47237682012-06-10 05:01:45 -04002878 int opened = 0;
Al Viro13aab422011-02-23 17:54:08 -05002879 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11002880
Al Viro30d90492012-06-22 12:40:19 +04002881 file = get_empty_filp();
2882 if (!file)
Nick Piggin31e6b012011-01-07 17:49:52 +11002883 return ERR_PTR(-ENFILE);
2884
Al Viro30d90492012-06-22 12:40:19 +04002885 file->f_flags = op->open_flag;
Nick Piggin31e6b012011-01-07 17:49:52 +11002886
Al Viro73d049a2011-03-11 12:08:24 -05002887 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
Nick Piggin31e6b012011-01-07 17:49:52 +11002888 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04002889 goto out;
Nick Piggin31e6b012011-01-07 17:49:52 +11002890
Al Virofe2d35f2011-03-05 22:58:25 -05002891 current->total_link_count = 0;
Al Viro73d049a2011-03-11 12:08:24 -05002892 error = link_path_walk(pathname, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11002893 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04002894 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002895
Al Viro2675a4e2012-06-22 12:41:10 +04002896 error = do_last(nd, &path, file, op, &opened, pathname);
2897 while (unlikely(error > 0)) { /* trailing symlink */
Nick Piggin7b9337a2011-01-14 08:42:43 +00002898 struct path link = path;
Al Virodef4af32009-12-26 08:37:05 -05002899 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04002900 if (!(nd->flags & LOOKUP_FOLLOW)) {
Al Viro73d049a2011-03-11 12:08:24 -05002901 path_put_conditional(&path, nd);
2902 path_put(&nd->path);
Al Viro2675a4e2012-06-22 12:41:10 +04002903 error = -ELOOP;
Al Viro40b39132011-03-09 16:22:18 -05002904 break;
2905 }
Kees Cook800179c2012-07-25 17:29:07 -07002906 error = may_follow_link(&link, nd);
2907 if (unlikely(error))
2908 break;
Al Viro73d049a2011-03-11 12:08:24 -05002909 nd->flags |= LOOKUP_PARENT;
2910 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
Al Viro574197e2011-03-14 22:20:34 -04002911 error = follow_link(&link, nd, &cookie);
Al Viroc3e380b2011-02-23 13:39:45 -05002912 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04002913 break;
2914 error = do_last(nd, &path, file, op, &opened, pathname);
Al Viro574197e2011-03-14 22:20:34 -04002915 put_link(nd, &link, cookie);
Al Viro806b6812009-12-26 07:16:40 -05002916 }
Al Viro10fa8e62009-12-26 07:09:49 -05002917out:
Al Viro73d049a2011-03-11 12:08:24 -05002918 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2919 path_put(&nd->root);
Al Virofe2d35f2011-03-05 22:58:25 -05002920 if (base)
2921 fput(base);
Al Viro2675a4e2012-06-22 12:41:10 +04002922 if (!(opened & FILE_OPENED)) {
2923 BUG_ON(!error);
Al Viro30d90492012-06-22 12:40:19 +04002924 put_filp(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002925 }
Al Viro2675a4e2012-06-22 12:41:10 +04002926 if (unlikely(error)) {
2927 if (error == -EOPENSTALE) {
2928 if (flags & LOOKUP_RCU)
2929 error = -ECHILD;
2930 else
2931 error = -ESTALE;
2932 }
2933 file = ERR_PTR(error);
2934 }
2935 return file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002936}
2937
Al Viro13aab422011-02-23 17:54:08 -05002938struct file *do_filp_open(int dfd, const char *pathname,
2939 const struct open_flags *op, int flags)
2940{
Al Viro73d049a2011-03-11 12:08:24 -05002941 struct nameidata nd;
Al Viro13aab422011-02-23 17:54:08 -05002942 struct file *filp;
2943
Al Viro73d049a2011-03-11 12:08:24 -05002944 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
Al Viro13aab422011-02-23 17:54:08 -05002945 if (unlikely(filp == ERR_PTR(-ECHILD)))
Al Viro73d049a2011-03-11 12:08:24 -05002946 filp = path_openat(dfd, pathname, &nd, op, flags);
Al Viro13aab422011-02-23 17:54:08 -05002947 if (unlikely(filp == ERR_PTR(-ESTALE)))
Al Viro73d049a2011-03-11 12:08:24 -05002948 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
Al Viro13aab422011-02-23 17:54:08 -05002949 return filp;
2950}
2951
Al Viro73d049a2011-03-11 12:08:24 -05002952struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2953 const char *name, const struct open_flags *op, int flags)
2954{
2955 struct nameidata nd;
2956 struct file *file;
2957
2958 nd.root.mnt = mnt;
2959 nd.root.dentry = dentry;
2960
2961 flags |= LOOKUP_ROOT;
2962
Al Virobcda7652011-03-13 16:42:14 -04002963 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
Al Viro73d049a2011-03-11 12:08:24 -05002964 return ERR_PTR(-ELOOP);
2965
2966 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2967 if (unlikely(file == ERR_PTR(-ECHILD)))
2968 file = path_openat(-1, name, &nd, op, flags);
2969 if (unlikely(file == ERR_PTR(-ESTALE)))
2970 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2971 return file;
2972}
2973
Al Viroed75e952011-06-27 16:53:43 -04002974struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002975{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002976 struct dentry *dentry = ERR_PTR(-EEXIST);
Al Viroed75e952011-06-27 16:53:43 -04002977 struct nameidata nd;
Jan Karac30dabf2012-06-12 16:20:30 +02002978 int err2;
Al Viroed75e952011-06-27 16:53:43 -04002979 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2980 if (error)
2981 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002982
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002983 /*
2984 * Yucky last component or no last component at all?
2985 * (foo/., foo/.., /////)
2986 */
Al Viroed75e952011-06-27 16:53:43 -04002987 if (nd.last_type != LAST_NORM)
2988 goto out;
2989 nd.flags &= ~LOOKUP_PARENT;
2990 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002991
Jan Karac30dabf2012-06-12 16:20:30 +02002992 /* don't fail immediately if it's r/o, at least try to report other errors */
2993 err2 = mnt_want_write(nd.path.mnt);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002994 /*
2995 * Do the final lookup.
2996 */
Al Viroed75e952011-06-27 16:53:43 -04002997 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2998 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002999 if (IS_ERR(dentry))
Al Viroa8104a92012-07-20 02:25:00 +04003000 goto unlock;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003001
Al Viroa8104a92012-07-20 02:25:00 +04003002 error = -EEXIST;
Al Viroe9baf6e2008-05-15 04:49:12 -04003003 if (dentry->d_inode)
Al Viroa8104a92012-07-20 02:25:00 +04003004 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003005 /*
3006 * Special case - lookup gave negative, but... we had foo/bar/
3007 * From the vfs_mknod() POV we just have a negative dentry -
3008 * all is fine. Let's be bastards - you had / on the end, you've
3009 * been asking for (non-existent) directory. -ENOENT for you.
3010 */
Al Viroed75e952011-06-27 16:53:43 -04003011 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
Al Viroa8104a92012-07-20 02:25:00 +04003012 error = -ENOENT;
Al Viroed75e952011-06-27 16:53:43 -04003013 goto fail;
Al Viroe9baf6e2008-05-15 04:49:12 -04003014 }
Jan Karac30dabf2012-06-12 16:20:30 +02003015 if (unlikely(err2)) {
3016 error = err2;
Al Viroa8104a92012-07-20 02:25:00 +04003017 goto fail;
Jan Karac30dabf2012-06-12 16:20:30 +02003018 }
Al Viroed75e952011-06-27 16:53:43 -04003019 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003020 return dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021fail:
Al Viroa8104a92012-07-20 02:25:00 +04003022 dput(dentry);
3023 dentry = ERR_PTR(error);
3024unlock:
Al Viroed75e952011-06-27 16:53:43 -04003025 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Karac30dabf2012-06-12 16:20:30 +02003026 if (!err2)
3027 mnt_drop_write(nd.path.mnt);
Al Viroed75e952011-06-27 16:53:43 -04003028out:
3029 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003030 return dentry;
3031}
Al Virodae6ad82011-06-26 11:50:15 -04003032EXPORT_SYMBOL(kern_path_create);
3033
Al Viro921a1652012-07-20 01:15:31 +04003034void done_path_create(struct path *path, struct dentry *dentry)
3035{
3036 dput(dentry);
3037 mutex_unlock(&path->dentry->d_inode->i_mutex);
Al Viroa8104a92012-07-20 02:25:00 +04003038 mnt_drop_write(path->mnt);
Al Viro921a1652012-07-20 01:15:31 +04003039 path_put(path);
3040}
3041EXPORT_SYMBOL(done_path_create);
3042
Al Virodae6ad82011-06-26 11:50:15 -04003043struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
3044{
3045 char *tmp = getname(pathname);
3046 struct dentry *res;
3047 if (IS_ERR(tmp))
3048 return ERR_CAST(tmp);
3049 res = kern_path_create(dfd, tmp, path, is_dir);
3050 putname(tmp);
3051 return res;
3052}
3053EXPORT_SYMBOL(user_path_create);
3054
Al Viro1a67aaf2011-07-26 01:52:52 -04003055int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003056{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003057 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003058
3059 if (error)
3060 return error;
3061
Eric W. Biederman975d6b32011-11-13 12:16:43 -08003062 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003063 return -EPERM;
3064
Al Viroacfa4382008-12-04 10:06:33 -05003065 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003066 return -EPERM;
3067
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07003068 error = devcgroup_inode_mknod(mode, dev);
3069 if (error)
3070 return error;
3071
Linus Torvalds1da177e2005-04-16 15:20:36 -07003072 error = security_inode_mknod(dir, dentry, mode, dev);
3073 if (error)
3074 return error;
3075
Linus Torvalds1da177e2005-04-16 15:20:36 -07003076 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003077 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003078 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003079 return error;
3080}
3081
Al Virof69aac02011-07-26 04:31:40 -04003082static int may_mknod(umode_t mode)
Dave Hansen463c3192008-02-15 14:37:57 -08003083{
3084 switch (mode & S_IFMT) {
3085 case S_IFREG:
3086 case S_IFCHR:
3087 case S_IFBLK:
3088 case S_IFIFO:
3089 case S_IFSOCK:
3090 case 0: /* zero mode translates to S_IFREG */
3091 return 0;
3092 case S_IFDIR:
3093 return -EPERM;
3094 default:
3095 return -EINVAL;
3096 }
3097}
3098
Al Viro8208a222011-07-25 17:32:17 -04003099SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003100 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003101{
Al Viro2ad94ae2008-07-21 09:32:51 -04003102 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003103 struct path path;
3104 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003105
Al Viro8e4bfca2012-07-20 01:17:26 +04003106 error = may_mknod(mode);
3107 if (error)
3108 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003109
Al Virodae6ad82011-06-26 11:50:15 -04003110 dentry = user_path_create(dfd, filename, &path, 0);
3111 if (IS_ERR(dentry))
3112 return PTR_ERR(dentry);
Al Viro2ad94ae2008-07-21 09:32:51 -04003113
Al Virodae6ad82011-06-26 11:50:15 -04003114 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003115 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003116 error = security_path_mknod(&path, dentry, mode, dev);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003117 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04003118 goto out;
Dave Hansen463c3192008-02-15 14:37:57 -08003119 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003120 case 0: case S_IFREG:
Al Viro312b63f2012-06-10 18:09:36 -04003121 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003122 break;
3123 case S_IFCHR: case S_IFBLK:
Al Virodae6ad82011-06-26 11:50:15 -04003124 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003125 new_decode_dev(dev));
3126 break;
3127 case S_IFIFO: case S_IFSOCK:
Al Virodae6ad82011-06-26 11:50:15 -04003128 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003129 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003130 }
Al Viroa8104a92012-07-20 02:25:00 +04003131out:
Al Viro921a1652012-07-20 01:15:31 +04003132 done_path_create(&path, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003133 return error;
3134}
3135
Al Viro8208a222011-07-25 17:32:17 -04003136SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003137{
3138 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3139}
3140
Al Viro18bb1db2011-07-26 01:41:39 -04003141int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003142{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003143 int error = may_create(dir, dentry);
Al Viro8de52772012-02-06 12:45:27 -05003144 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003145
3146 if (error)
3147 return error;
3148
Al Viroacfa4382008-12-04 10:06:33 -05003149 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003150 return -EPERM;
3151
3152 mode &= (S_IRWXUGO|S_ISVTX);
3153 error = security_inode_mkdir(dir, dentry, mode);
3154 if (error)
3155 return error;
3156
Al Viro8de52772012-02-06 12:45:27 -05003157 if (max_links && dir->i_nlink >= max_links)
3158 return -EMLINK;
3159
Linus Torvalds1da177e2005-04-16 15:20:36 -07003160 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003161 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003162 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003163 return error;
3164}
3165
Al Viroa218d0f2011-11-21 14:59:34 -05003166SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003167{
Dave Hansen6902d922006-09-30 23:29:01 -07003168 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003169 struct path path;
3170 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003171
Al Virodae6ad82011-06-26 11:50:15 -04003172 dentry = user_path_create(dfd, pathname, &path, 1);
Dave Hansen6902d922006-09-30 23:29:01 -07003173 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003174 return PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003175
Al Virodae6ad82011-06-26 11:50:15 -04003176 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003177 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003178 error = security_path_mkdir(&path, dentry, mode);
Al Viroa8104a92012-07-20 02:25:00 +04003179 if (!error)
3180 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
Al Viro921a1652012-07-20 01:15:31 +04003181 done_path_create(&path, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003182 return error;
3183}
3184
Al Viroa218d0f2011-11-21 14:59:34 -05003185SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003186{
3187 return sys_mkdirat(AT_FDCWD, pathname, mode);
3188}
3189
Linus Torvalds1da177e2005-04-16 15:20:36 -07003190/*
Sage Weila71905f2011-05-24 13:06:08 -07003191 * The dentry_unhash() helper will try to drop the dentry early: we
J. Bruce Fieldsc0d02592012-02-15 11:48:40 -05003192 * should have a usage count of 1 if we're the only user of this
Sage Weila71905f2011-05-24 13:06:08 -07003193 * dentry, and if that is true (possibly after pruning the dcache),
3194 * then we drop the dentry now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003195 *
3196 * A low-level filesystem can, if it choses, legally
3197 * do a
3198 *
3199 * if (!d_unhashed(dentry))
3200 * return -EBUSY;
3201 *
3202 * if it cannot handle the case of removing a directory
3203 * that is still in use by something else..
3204 */
3205void dentry_unhash(struct dentry *dentry)
3206{
Vasily Averindc168422006-12-06 20:37:07 -08003207 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003208 spin_lock(&dentry->d_lock);
Sage Weil64252c72011-05-24 13:06:05 -07003209 if (dentry->d_count == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003210 __d_drop(dentry);
3211 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003212}
3213
3214int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3215{
3216 int error = may_delete(dir, dentry, 1);
3217
3218 if (error)
3219 return error;
3220
Al Viroacfa4382008-12-04 10:06:33 -05003221 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003222 return -EPERM;
3223
Al Viro1d2ef592011-09-14 18:55:41 +01003224 dget(dentry);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003225 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003226
Sage Weil912dbc12011-05-24 13:06:11 -07003227 error = -EBUSY;
3228 if (d_mountpoint(dentry))
3229 goto out;
3230
3231 error = security_inode_rmdir(dir, dentry);
3232 if (error)
3233 goto out;
3234
Sage Weil3cebde22011-05-29 21:20:59 -07003235 shrink_dcache_parent(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003236 error = dir->i_op->rmdir(dir, dentry);
3237 if (error)
3238 goto out;
3239
3240 dentry->d_inode->i_flags |= S_DEAD;
3241 dont_mount(dentry);
3242
3243out:
3244 mutex_unlock(&dentry->d_inode->i_mutex);
Al Viro1d2ef592011-09-14 18:55:41 +01003245 dput(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003246 if (!error)
3247 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003248 return error;
3249}
3250
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003251static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003252{
3253 int error = 0;
3254 char * name;
3255 struct dentry *dentry;
3256 struct nameidata nd;
3257
Al Viro2ad94ae2008-07-21 09:32:51 -04003258 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003259 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04003260 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003261
3262 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003263 case LAST_DOTDOT:
3264 error = -ENOTEMPTY;
3265 goto exit1;
3266 case LAST_DOT:
3267 error = -EINVAL;
3268 goto exit1;
3269 case LAST_ROOT:
3270 error = -EBUSY;
3271 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003272 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003273
3274 nd.flags &= ~LOOKUP_PARENT;
Jan Karac30dabf2012-06-12 16:20:30 +02003275 error = mnt_want_write(nd.path.mnt);
3276 if (error)
3277 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003278
Jan Blunck4ac91372008-02-14 19:34:32 -08003279 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08003280 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003281 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003282 if (IS_ERR(dentry))
3283 goto exit2;
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003284 if (!dentry->d_inode) {
3285 error = -ENOENT;
3286 goto exit3;
3287 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003288 error = security_path_rmdir(&nd.path, dentry);
3289 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003290 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08003291 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Dave Hansen06227532008-02-15 14:37:34 -08003292exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07003293 dput(dentry);
3294exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08003295 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Karac30dabf2012-06-12 16:20:30 +02003296 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003297exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003298 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003299 putname(name);
3300 return error;
3301}
3302
Heiko Carstens3cdad422009-01-14 14:14:22 +01003303SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003304{
3305 return do_rmdir(AT_FDCWD, pathname);
3306}
3307
Linus Torvalds1da177e2005-04-16 15:20:36 -07003308int vfs_unlink(struct inode *dir, struct dentry *dentry)
3309{
3310 int error = may_delete(dir, dentry, 0);
3311
3312 if (error)
3313 return error;
3314
Al Viroacfa4382008-12-04 10:06:33 -05003315 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003316 return -EPERM;
3317
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003318 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003319 if (d_mountpoint(dentry))
3320 error = -EBUSY;
3321 else {
3322 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05003323 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003324 error = dir->i_op->unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05003325 if (!error)
Al Virod83c49f2010-04-30 17:17:09 -04003326 dont_mount(dentry);
Al Virobec10522010-03-03 14:12:08 -05003327 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003328 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003329 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003330
3331 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3332 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08003333 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003334 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003335 }
Robert Love0eeca282005-07-12 17:06:03 -04003336
Linus Torvalds1da177e2005-04-16 15:20:36 -07003337 return error;
3338}
3339
3340/*
3341 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003342 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07003343 * writeout happening, and we don't want to prevent access to the directory
3344 * while waiting on the I/O.
3345 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003346static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003347{
Al Viro2ad94ae2008-07-21 09:32:51 -04003348 int error;
3349 char *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003350 struct dentry *dentry;
3351 struct nameidata nd;
3352 struct inode *inode = NULL;
3353
Al Viro2ad94ae2008-07-21 09:32:51 -04003354 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003355 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04003356 return error;
3357
Linus Torvalds1da177e2005-04-16 15:20:36 -07003358 error = -EISDIR;
3359 if (nd.last_type != LAST_NORM)
3360 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003361
3362 nd.flags &= ~LOOKUP_PARENT;
Jan Karac30dabf2012-06-12 16:20:30 +02003363 error = mnt_want_write(nd.path.mnt);
3364 if (error)
3365 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003366
Jan Blunck4ac91372008-02-14 19:34:32 -08003367 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08003368 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003369 error = PTR_ERR(dentry);
3370 if (!IS_ERR(dentry)) {
3371 /* Why not before? Because we want correct error value */
Török Edwin50338b82011-06-16 00:06:14 +03003372 if (nd.last.name[nd.last.len])
3373 goto slashes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003374 inode = dentry->d_inode;
Török Edwin50338b82011-06-16 00:06:14 +03003375 if (!inode)
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003376 goto slashes;
3377 ihold(inode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003378 error = security_path_unlink(&nd.path, dentry);
3379 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003380 goto exit2;
Jan Blunck4ac91372008-02-14 19:34:32 -08003381 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Jan Karac30dabf2012-06-12 16:20:30 +02003382exit2:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003383 dput(dentry);
3384 }
Jan Blunck4ac91372008-02-14 19:34:32 -08003385 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003386 if (inode)
3387 iput(inode); /* truncate the inode here */
Jan Karac30dabf2012-06-12 16:20:30 +02003388 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003389exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003390 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003391 putname(name);
3392 return error;
3393
3394slashes:
3395 error = !dentry->d_inode ? -ENOENT :
3396 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
3397 goto exit2;
3398}
3399
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003400SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003401{
3402 if ((flag & ~AT_REMOVEDIR) != 0)
3403 return -EINVAL;
3404
3405 if (flag & AT_REMOVEDIR)
3406 return do_rmdir(dfd, pathname);
3407
3408 return do_unlinkat(dfd, pathname);
3409}
3410
Heiko Carstens3480b252009-01-14 14:14:16 +01003411SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003412{
3413 return do_unlinkat(AT_FDCWD, pathname);
3414}
3415
Miklos Szeredidb2e7472008-06-24 16:50:16 +02003416int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003417{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003418 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003419
3420 if (error)
3421 return error;
3422
Al Viroacfa4382008-12-04 10:06:33 -05003423 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003424 return -EPERM;
3425
3426 error = security_inode_symlink(dir, dentry, oldname);
3427 if (error)
3428 return error;
3429
Linus Torvalds1da177e2005-04-16 15:20:36 -07003430 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003431 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003432 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003433 return error;
3434}
3435
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003436SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3437 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003438{
Al Viro2ad94ae2008-07-21 09:32:51 -04003439 int error;
3440 char *from;
Dave Hansen6902d922006-09-30 23:29:01 -07003441 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003442 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003443
3444 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04003445 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003446 return PTR_ERR(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04003447
Al Virodae6ad82011-06-26 11:50:15 -04003448 dentry = user_path_create(newdfd, newname, &path, 0);
Dave Hansen6902d922006-09-30 23:29:01 -07003449 error = PTR_ERR(dentry);
3450 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003451 goto out_putname;
Dave Hansen6902d922006-09-30 23:29:01 -07003452
Al Virodae6ad82011-06-26 11:50:15 -04003453 error = security_path_symlink(&path, dentry, from);
Al Viroa8104a92012-07-20 02:25:00 +04003454 if (!error)
3455 error = vfs_symlink(path.dentry->d_inode, dentry, from);
Al Viro921a1652012-07-20 01:15:31 +04003456 done_path_create(&path, dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003457out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003458 putname(from);
3459 return error;
3460}
3461
Heiko Carstens3480b252009-01-14 14:14:16 +01003462SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003463{
3464 return sys_symlinkat(oldname, AT_FDCWD, newname);
3465}
3466
Linus Torvalds1da177e2005-04-16 15:20:36 -07003467int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
3468{
3469 struct inode *inode = old_dentry->d_inode;
Al Viro8de52772012-02-06 12:45:27 -05003470 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003471 int error;
3472
3473 if (!inode)
3474 return -ENOENT;
3475
Miklos Szeredia95164d2008-07-30 15:08:48 +02003476 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003477 if (error)
3478 return error;
3479
3480 if (dir->i_sb != inode->i_sb)
3481 return -EXDEV;
3482
3483 /*
3484 * A link to an append-only or immutable file cannot be created.
3485 */
3486 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3487 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05003488 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003489 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003490 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003491 return -EPERM;
3492
3493 error = security_inode_link(old_dentry, dir, new_dentry);
3494 if (error)
3495 return error;
3496
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003497 mutex_lock(&inode->i_mutex);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05303498 /* Make sure we don't allow creating hardlink to an unlinked file */
3499 if (inode->i_nlink == 0)
3500 error = -ENOENT;
Al Viro8de52772012-02-06 12:45:27 -05003501 else if (max_links && inode->i_nlink >= max_links)
3502 error = -EMLINK;
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05303503 else
3504 error = dir->i_op->link(old_dentry, dir, new_dentry);
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003505 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07003506 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003507 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003508 return error;
3509}
3510
3511/*
3512 * Hardlinks are often used in delicate situations. We avoid
3513 * security-related surprises by not following symlinks on the
3514 * newname. --KAB
3515 *
3516 * We don't follow them on the oldname either to be compatible
3517 * with linux 2.0, and to avoid hard-linking to directories
3518 * and other special files. --ADM
3519 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003520SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3521 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003522{
3523 struct dentry *new_dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003524 struct path old_path, new_path;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303525 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003526 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003527
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303528 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003529 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303530 /*
3531 * To use null names we require CAP_DAC_READ_SEARCH
3532 * This ensures that not everyone will be able to create
3533 * handlink using the passed filedescriptor.
3534 */
3535 if (flags & AT_EMPTY_PATH) {
3536 if (!capable(CAP_DAC_READ_SEARCH))
3537 return -ENOENT;
3538 how = LOOKUP_EMPTY;
3539 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003540
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303541 if (flags & AT_SYMLINK_FOLLOW)
3542 how |= LOOKUP_FOLLOW;
3543
3544 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003545 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04003546 return error;
3547
Al Virodae6ad82011-06-26 11:50:15 -04003548 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003549 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003550 if (IS_ERR(new_dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003551 goto out;
3552
3553 error = -EXDEV;
3554 if (old_path.mnt != new_path.mnt)
3555 goto out_dput;
Kees Cook800179c2012-07-25 17:29:07 -07003556 error = may_linkat(&old_path);
3557 if (unlikely(error))
3558 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04003559 error = security_path_link(old_path.dentry, &new_path, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003560 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04003561 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04003562 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
Dave Hansen75c3f292008-02-15 14:37:45 -08003563out_dput:
Al Viro921a1652012-07-20 01:15:31 +04003564 done_path_create(&new_path, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003565out:
Al Viro2d8f3032008-07-22 09:59:21 -04003566 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003567
3568 return error;
3569}
3570
Heiko Carstens3480b252009-01-14 14:14:16 +01003571SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003572{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003573 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003574}
3575
Linus Torvalds1da177e2005-04-16 15:20:36 -07003576/*
3577 * The worst of all namespace operations - renaming directory. "Perverted"
3578 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3579 * Problems:
3580 * a) we can get into loop creation. Check is done in is_subdir().
3581 * b) race potential - two innocent renames can create a loop together.
3582 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003583 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07003584 * story.
3585 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003586 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07003587 * whether the target exists). Solution: try to be smart with locking
3588 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003589 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07003590 * move will be locked. Thus we can rank directories by the tree
3591 * (ancestors first) and rank all non-directories after them.
3592 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003593 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003594 * HOWEVER, it relies on the assumption that any object with ->lookup()
3595 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3596 * we'd better make sure that there's no link(2) for them.
Sage Weile4eaac02011-05-24 13:06:07 -07003597 * d) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003598 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07003599 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05003600 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07003601 * locking].
3602 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07003603static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3604 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003605{
3606 int error = 0;
Sage Weil9055cba2011-05-24 13:06:12 -07003607 struct inode *target = new_dentry->d_inode;
Al Viro8de52772012-02-06 12:45:27 -05003608 unsigned max_links = new_dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003609
3610 /*
3611 * If we are going to change the parent - check write permissions,
3612 * we'll need to flip '..'.
3613 */
3614 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04003615 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003616 if (error)
3617 return error;
3618 }
3619
3620 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3621 if (error)
3622 return error;
3623
Al Viro1d2ef592011-09-14 18:55:41 +01003624 dget(new_dentry);
Al Virod83c49f2010-04-30 17:17:09 -04003625 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003626 mutex_lock(&target->i_mutex);
Sage Weil9055cba2011-05-24 13:06:12 -07003627
3628 error = -EBUSY;
3629 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3630 goto out;
3631
Al Viro8de52772012-02-06 12:45:27 -05003632 error = -EMLINK;
3633 if (max_links && !target && new_dir != old_dir &&
3634 new_dir->i_nlink >= max_links)
3635 goto out;
3636
Sage Weil3cebde22011-05-29 21:20:59 -07003637 if (target)
3638 shrink_dcache_parent(new_dentry);
Sage Weil9055cba2011-05-24 13:06:12 -07003639 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3640 if (error)
3641 goto out;
3642
Linus Torvalds1da177e2005-04-16 15:20:36 -07003643 if (target) {
Sage Weil9055cba2011-05-24 13:06:12 -07003644 target->i_flags |= S_DEAD;
3645 dont_mount(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003646 }
Sage Weil9055cba2011-05-24 13:06:12 -07003647out:
3648 if (target)
3649 mutex_unlock(&target->i_mutex);
Al Viro1d2ef592011-09-14 18:55:41 +01003650 dput(new_dentry);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07003651 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07003652 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3653 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003654 return error;
3655}
3656
Adrian Bunk75c96f82005-05-05 16:16:09 -07003657static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3658 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003659{
Sage Weil51892bb2011-05-24 13:06:13 -07003660 struct inode *target = new_dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003661 int error;
3662
3663 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3664 if (error)
3665 return error;
3666
3667 dget(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003668 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003669 mutex_lock(&target->i_mutex);
Sage Weil51892bb2011-05-24 13:06:13 -07003670
3671 error = -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003672 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
Sage Weil51892bb2011-05-24 13:06:13 -07003673 goto out;
3674
3675 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3676 if (error)
3677 goto out;
3678
3679 if (target)
3680 dont_mount(new_dentry);
3681 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3682 d_move(old_dentry, new_dentry);
3683out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003684 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003685 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003686 dput(new_dentry);
3687 return error;
3688}
3689
3690int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3691 struct inode *new_dir, struct dentry *new_dentry)
3692{
3693 int error;
3694 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Eric Paris59b0df22010-02-08 12:53:52 -05003695 const unsigned char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003696
3697 if (old_dentry->d_inode == new_dentry->d_inode)
3698 return 0;
3699
3700 error = may_delete(old_dir, old_dentry, is_dir);
3701 if (error)
3702 return error;
3703
3704 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02003705 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003706 else
3707 error = may_delete(new_dir, new_dentry, is_dir);
3708 if (error)
3709 return error;
3710
Al Viroacfa4382008-12-04 10:06:33 -05003711 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003712 return -EPERM;
3713
Robert Love0eeca282005-07-12 17:06:03 -04003714 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3715
Linus Torvalds1da177e2005-04-16 15:20:36 -07003716 if (is_dir)
3717 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3718 else
3719 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
Al Viro123df292009-12-25 04:57:57 -05003720 if (!error)
3721 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04003722 new_dentry->d_inode, old_dentry);
Robert Love0eeca282005-07-12 17:06:03 -04003723 fsnotify_oldname_free(old_name);
3724
Linus Torvalds1da177e2005-04-16 15:20:36 -07003725 return error;
3726}
3727
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003728SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3729 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003730{
Al Viro2ad94ae2008-07-21 09:32:51 -04003731 struct dentry *old_dir, *new_dir;
3732 struct dentry *old_dentry, *new_dentry;
3733 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003734 struct nameidata oldnd, newnd;
Al Viro2ad94ae2008-07-21 09:32:51 -04003735 char *from;
3736 char *to;
3737 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003738
Al Viro2ad94ae2008-07-21 09:32:51 -04003739 error = user_path_parent(olddfd, oldname, &oldnd, &from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003740 if (error)
3741 goto exit;
3742
Al Viro2ad94ae2008-07-21 09:32:51 -04003743 error = user_path_parent(newdfd, newname, &newnd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003744 if (error)
3745 goto exit1;
3746
3747 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08003748 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003749 goto exit2;
3750
Jan Blunck4ac91372008-02-14 19:34:32 -08003751 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003752 error = -EBUSY;
3753 if (oldnd.last_type != LAST_NORM)
3754 goto exit2;
3755
Jan Blunck4ac91372008-02-14 19:34:32 -08003756 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003757 if (newnd.last_type != LAST_NORM)
3758 goto exit2;
3759
Jan Karac30dabf2012-06-12 16:20:30 +02003760 error = mnt_want_write(oldnd.path.mnt);
3761 if (error)
3762 goto exit2;
3763
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003764 oldnd.flags &= ~LOOKUP_PARENT;
3765 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09003766 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003767
Linus Torvalds1da177e2005-04-16 15:20:36 -07003768 trap = lock_rename(new_dir, old_dir);
3769
Christoph Hellwig49705b72005-11-08 21:35:06 -08003770 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003771 error = PTR_ERR(old_dentry);
3772 if (IS_ERR(old_dentry))
3773 goto exit3;
3774 /* source must exist */
3775 error = -ENOENT;
3776 if (!old_dentry->d_inode)
3777 goto exit4;
3778 /* unless the source is a directory trailing slashes give -ENOTDIR */
3779 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3780 error = -ENOTDIR;
3781 if (oldnd.last.name[oldnd.last.len])
3782 goto exit4;
3783 if (newnd.last.name[newnd.last.len])
3784 goto exit4;
3785 }
3786 /* source should not be ancestor of target */
3787 error = -EINVAL;
3788 if (old_dentry == trap)
3789 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08003790 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003791 error = PTR_ERR(new_dentry);
3792 if (IS_ERR(new_dentry))
3793 goto exit4;
3794 /* target should not be an ancestor of source */
3795 error = -ENOTEMPTY;
3796 if (new_dentry == trap)
3797 goto exit5;
3798
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003799 error = security_path_rename(&oldnd.path, old_dentry,
3800 &newnd.path, new_dentry);
3801 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003802 goto exit5;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003803 error = vfs_rename(old_dir->d_inode, old_dentry,
3804 new_dir->d_inode, new_dentry);
3805exit5:
3806 dput(new_dentry);
3807exit4:
3808 dput(old_dentry);
3809exit3:
3810 unlock_rename(new_dir, old_dir);
Jan Karac30dabf2012-06-12 16:20:30 +02003811 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003812exit2:
Jan Blunck1d957f92008-02-14 19:34:35 -08003813 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04003814 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003815exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003816 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003817 putname(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04003818exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003819 return error;
3820}
3821
Heiko Carstensa26eab22009-01-14 14:14:17 +01003822SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003823{
3824 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3825}
3826
Linus Torvalds1da177e2005-04-16 15:20:36 -07003827int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3828{
3829 int len;
3830
3831 len = PTR_ERR(link);
3832 if (IS_ERR(link))
3833 goto out;
3834
3835 len = strlen(link);
3836 if (len > (unsigned) buflen)
3837 len = buflen;
3838 if (copy_to_user(buffer, link, len))
3839 len = -EFAULT;
3840out:
3841 return len;
3842}
3843
3844/*
3845 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3846 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3847 * using) it for any given inode is up to filesystem.
3848 */
3849int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3850{
3851 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003852 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07003853 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003854
Linus Torvalds1da177e2005-04-16 15:20:36 -07003855 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003856 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07003857 if (IS_ERR(cookie))
3858 return PTR_ERR(cookie);
3859
3860 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3861 if (dentry->d_inode->i_op->put_link)
3862 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3863 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003864}
3865
3866int vfs_follow_link(struct nameidata *nd, const char *link)
3867{
3868 return __vfs_follow_link(nd, link);
3869}
3870
3871/* get the link contents into pagecache */
3872static char *page_getlink(struct dentry * dentry, struct page **ppage)
3873{
Duane Griffinebd09ab2008-12-19 20:47:12 +00003874 char *kaddr;
3875 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003876 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07003877 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003878 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07003879 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003880 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00003881 kaddr = kmap(page);
3882 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3883 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003884}
3885
3886int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3887{
3888 struct page *page = NULL;
3889 char *s = page_getlink(dentry, &page);
3890 int res = vfs_readlink(dentry,buffer,buflen,s);
3891 if (page) {
3892 kunmap(page);
3893 page_cache_release(page);
3894 }
3895 return res;
3896}
3897
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003898void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003899{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003900 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003901 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003902 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003903}
3904
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003905void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003906{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003907 struct page *page = cookie;
3908
3909 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003910 kunmap(page);
3911 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003912 }
3913}
3914
Nick Piggin54566b22009-01-04 12:00:53 -08003915/*
3916 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3917 */
3918int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003919{
3920 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003921 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07003922 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08003923 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003924 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08003925 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3926 if (nofs)
3927 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003928
NeilBrown7e53cac2006-03-25 03:07:57 -08003929retry:
Nick Pigginafddba42007-10-16 01:25:01 -07003930 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08003931 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003932 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07003933 goto fail;
3934
Cong Wange8e3c3d2011-11-25 23:14:27 +08003935 kaddr = kmap_atomic(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003936 memcpy(kaddr, symname, len-1);
Cong Wange8e3c3d2011-11-25 23:14:27 +08003937 kunmap_atomic(kaddr);
Nick Pigginafddba42007-10-16 01:25:01 -07003938
3939 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3940 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003941 if (err < 0)
3942 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07003943 if (err < len-1)
3944 goto retry;
3945
Linus Torvalds1da177e2005-04-16 15:20:36 -07003946 mark_inode_dirty(inode);
3947 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003948fail:
3949 return err;
3950}
3951
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003952int page_symlink(struct inode *inode, const char *symname, int len)
3953{
3954 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08003955 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003956}
3957
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003958const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003959 .readlink = generic_readlink,
3960 .follow_link = page_follow_link_light,
3961 .put_link = page_put_link,
3962};
3963
Al Viro2d8f3032008-07-22 09:59:21 -04003964EXPORT_SYMBOL(user_path_at);
David Howellscc53ce52011-01-14 18:45:26 +00003965EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003966EXPORT_SYMBOL(follow_down);
3967EXPORT_SYMBOL(follow_up);
3968EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3969EXPORT_SYMBOL(getname);
3970EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003971EXPORT_SYMBOL(lookup_one_len);
3972EXPORT_SYMBOL(page_follow_link_light);
3973EXPORT_SYMBOL(page_put_link);
3974EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003975EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003976EXPORT_SYMBOL(page_symlink);
3977EXPORT_SYMBOL(page_symlink_inode_operations);
Al Virod1811462008-08-02 00:49:18 -04003978EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07003979EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04003980EXPORT_SYMBOL(inode_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003981EXPORT_SYMBOL(unlock_rename);
3982EXPORT_SYMBOL(vfs_create);
3983EXPORT_SYMBOL(vfs_follow_link);
3984EXPORT_SYMBOL(vfs_link);
3985EXPORT_SYMBOL(vfs_mkdir);
3986EXPORT_SYMBOL(vfs_mknod);
3987EXPORT_SYMBOL(generic_permission);
3988EXPORT_SYMBOL(vfs_readlink);
3989EXPORT_SYMBOL(vfs_rename);
3990EXPORT_SYMBOL(vfs_rmdir);
3991EXPORT_SYMBOL(vfs_symlink);
3992EXPORT_SYMBOL(vfs_unlink);
3993EXPORT_SYMBOL(dentry_unhash);
3994EXPORT_SYMBOL(generic_readlink);