blob: 5f4cdf3ad913fec928835d87fbe499efdf7d67e5 [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 */
Jeff Layton91a27b22012-10-10 15:25:28 -0400120void final_putname(struct filename *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121{
Jeff Layton7950e382012-10-10 16:43:13 -0400122 if (name->separate) {
123 __putname(name->name);
124 kfree(name);
125 } else {
126 __putname(name);
127 }
Jeff Layton91a27b22012-10-10 15:25:28 -0400128}
129
Jeff Layton7950e382012-10-10 16:43:13 -0400130#define EMBEDDED_NAME_MAX (PATH_MAX - sizeof(struct filename))
131
Jeff Layton91a27b22012-10-10 15:25:28 -0400132static struct filename *
133getname_flags(const char __user *filename, int flags, int *empty)
134{
135 struct filename *result, *err;
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700136 int len;
Jeff Layton7950e382012-10-10 16:43:13 -0400137 long max;
138 char *kname;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139
Jeff Layton7ac86262012-10-10 15:25:28 -0400140 result = audit_reusename(filename);
141 if (result)
142 return result;
143
Jeff Layton7950e382012-10-10 16:43:13 -0400144 result = __getname();
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700145 if (unlikely(!result))
Eric Paris4043cde2012-01-03 14:23:08 -0500146 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147
Jeff Layton7950e382012-10-10 16:43:13 -0400148 /*
149 * First, try to embed the struct filename inside the names_cache
150 * allocation
151 */
152 kname = (char *)result + sizeof(*result);
Jeff Layton91a27b22012-10-10 15:25:28 -0400153 result->name = kname;
Jeff Layton7950e382012-10-10 16:43:13 -0400154 result->separate = false;
155 max = EMBEDDED_NAME_MAX;
156
157recopy:
158 len = strncpy_from_user(kname, filename, max);
Jeff Layton91a27b22012-10-10 15:25:28 -0400159 if (unlikely(len < 0)) {
160 err = ERR_PTR(len);
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700161 goto error;
Jeff Layton91a27b22012-10-10 15:25:28 -0400162 }
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700163
Jeff Layton7950e382012-10-10 16:43:13 -0400164 /*
165 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
166 * separate struct filename so we can dedicate the entire
167 * names_cache allocation for the pathname, and re-do the copy from
168 * userland.
169 */
170 if (len == EMBEDDED_NAME_MAX && max == EMBEDDED_NAME_MAX) {
171 kname = (char *)result;
172
173 result = kzalloc(sizeof(*result), GFP_KERNEL);
174 if (!result) {
175 err = ERR_PTR(-ENOMEM);
176 result = (struct filename *)kname;
177 goto error;
178 }
179 result->name = kname;
180 result->separate = true;
181 max = PATH_MAX;
182 goto recopy;
183 }
184
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700185 /* The empty path is special. */
186 if (unlikely(!len)) {
187 if (empty)
Eric Paris4043cde2012-01-03 14:23:08 -0500188 *empty = 1;
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700189 err = ERR_PTR(-ENOENT);
190 if (!(flags & LOOKUP_EMPTY))
191 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 }
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700193
194 err = ERR_PTR(-ENAMETOOLONG);
Jeff Layton7950e382012-10-10 16:43:13 -0400195 if (unlikely(len >= PATH_MAX))
196 goto error;
197
198 result->uptr = filename;
199 audit_getname(result);
200 return result;
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700201
202error:
Jeff Layton7950e382012-10-10 16:43:13 -0400203 final_putname(result);
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700204 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205}
206
Jeff Layton91a27b22012-10-10 15:25:28 -0400207struct filename *
208getname(const char __user * filename)
Al Virof52e0c12011-03-14 18:56:51 -0400209{
Linus Torvaldsf7493e52012-03-22 16:10:40 -0700210 return getname_flags(filename, 0, NULL);
Al Virof52e0c12011-03-14 18:56:51 -0400211}
Jeff Layton91a27b22012-10-10 15:25:28 -0400212EXPORT_SYMBOL(getname);
Al Virof52e0c12011-03-14 18:56:51 -0400213
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214#ifdef CONFIG_AUDITSYSCALL
Jeff Layton91a27b22012-10-10 15:25:28 -0400215void putname(struct filename *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216{
Al Viro5ac3a9c2006-07-16 06:38:45 -0400217 if (unlikely(!audit_dummy_context()))
Jeff Layton91a27b22012-10-10 15:25:28 -0400218 return audit_putname(name);
219 final_putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221#endif
222
Linus Torvaldse77819e2011-07-22 19:30:19 -0700223static int check_acl(struct inode *inode, int mask)
224{
Linus Torvalds84635d62011-07-25 22:47:03 -0700225#ifdef CONFIG_FS_POSIX_ACL
Linus Torvaldse77819e2011-07-22 19:30:19 -0700226 struct posix_acl *acl;
227
Linus Torvaldse77819e2011-07-22 19:30:19 -0700228 if (mask & MAY_NOT_BLOCK) {
Al Viro35678662011-08-02 21:32:13 -0400229 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
230 if (!acl)
Linus Torvaldse77819e2011-07-22 19:30:19 -0700231 return -EAGAIN;
Al Viro35678662011-08-02 21:32:13 -0400232 /* no ->get_acl() calls in RCU mode... */
233 if (acl == ACL_NOT_CACHED)
234 return -ECHILD;
Ari Savolainen206b1d02011-08-06 19:43:07 +0300235 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
Linus Torvaldse77819e2011-07-22 19:30:19 -0700236 }
237
238 acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
239
240 /*
Christoph Hellwig4e34e712011-07-23 17:37:31 +0200241 * A filesystem can force a ACL callback by just never filling the
242 * ACL cache. But normally you'd fill the cache either at inode
243 * instantiation time, or on the first ->get_acl call.
Linus Torvaldse77819e2011-07-22 19:30:19 -0700244 *
Christoph Hellwig4e34e712011-07-23 17:37:31 +0200245 * If the filesystem doesn't have a get_acl() function at all, we'll
246 * just create the negative cache entry.
Linus Torvaldse77819e2011-07-22 19:30:19 -0700247 */
248 if (acl == ACL_NOT_CACHED) {
Christoph Hellwig4e34e712011-07-23 17:37:31 +0200249 if (inode->i_op->get_acl) {
250 acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS);
251 if (IS_ERR(acl))
252 return PTR_ERR(acl);
253 } else {
254 set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
255 return -EAGAIN;
256 }
Linus Torvaldse77819e2011-07-22 19:30:19 -0700257 }
258
259 if (acl) {
260 int error = posix_acl_permission(inode, acl, mask);
261 posix_acl_release(acl);
262 return error;
263 }
Linus Torvalds84635d62011-07-25 22:47:03 -0700264#endif
Linus Torvaldse77819e2011-07-22 19:30:19 -0700265
266 return -EAGAIN;
267}
268
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700269/*
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530270 * This does the basic permission checking
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700271 */
Al Viro7e401452011-06-20 19:12:17 -0400272static int acl_permission_check(struct inode *inode, int mask)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700273{
Linus Torvalds26cf46b2011-05-13 11:51:01 -0700274 unsigned int mode = inode->i_mode;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700275
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -0800276 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700277 mode >>= 6;
278 else {
Linus Torvaldse77819e2011-07-22 19:30:19 -0700279 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
Al Viro7e401452011-06-20 19:12:17 -0400280 int error = check_acl(inode, mask);
Nick Pigginb74c79e2011-01-07 17:49:58 +1100281 if (error != -EAGAIN)
282 return error;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700283 }
284
285 if (in_group_p(inode->i_gid))
286 mode >>= 3;
287 }
288
289 /*
290 * If the DACs are ok we don't need any capability check.
291 */
Al Viro9c2c7032011-06-20 19:06:22 -0400292 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700293 return 0;
294 return -EACCES;
295}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296
297/**
Nick Pigginb74c79e2011-01-07 17:49:58 +1100298 * generic_permission - check for access rights on a Posix-like filesystem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 * @inode: inode to check access rights for
Andreas Gruenbacher8fd90c82011-10-23 23:13:30 +0530300 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 *
302 * Used to check for read/write/execute permissions on a file.
303 * We use "fsuid" for this, letting us set arbitrary permissions
304 * for filesystem access without changing the "normal" uids which
Nick Pigginb74c79e2011-01-07 17:49:58 +1100305 * are used for other things.
306 *
307 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
308 * request cannot be satisfied (eg. requires blocking or too much complexity).
309 * It would then be called again in ref-walk mode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 */
Al Viro2830ba72011-06-20 19:16:29 -0400311int generic_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700313 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314
315 /*
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530316 * Do the basic permission checks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 */
Al Viro7e401452011-06-20 19:12:17 -0400318 ret = acl_permission_check(inode, mask);
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700319 if (ret != -EACCES)
320 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321
Al Virod594e7e2011-06-20 19:55:42 -0400322 if (S_ISDIR(inode->i_mode)) {
323 /* DACs are overridable for directories */
Eric W. Biederman1a48e2a2011-11-14 16:24:06 -0800324 if (inode_capable(inode, CAP_DAC_OVERRIDE))
Al Virod594e7e2011-06-20 19:55:42 -0400325 return 0;
326 if (!(mask & MAY_WRITE))
Eric W. Biederman1a48e2a2011-11-14 16:24:06 -0800327 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
Al Virod594e7e2011-06-20 19:55:42 -0400328 return 0;
329 return -EACCES;
330 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 /*
332 * Read/write DACs are always overridable.
Al Virod594e7e2011-06-20 19:55:42 -0400333 * Executable DACs are overridable when there is
334 * at least one exec bit set.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 */
Al Virod594e7e2011-06-20 19:55:42 -0400336 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
Eric W. Biederman1a48e2a2011-11-14 16:24:06 -0800337 if (inode_capable(inode, CAP_DAC_OVERRIDE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 return 0;
339
340 /*
341 * Searching includes executable on directories, else just read.
342 */
Serge E. Hallyn7ea66002009-12-29 14:50:19 -0600343 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
Al Virod594e7e2011-06-20 19:55:42 -0400344 if (mask == MAY_READ)
Eric W. Biederman1a48e2a2011-11-14 16:24:06 -0800345 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 return 0;
347
348 return -EACCES;
349}
350
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700351/*
352 * We _really_ want to just do "generic_permission()" without
353 * even looking at the inode->i_op values. So we keep a cache
354 * flag in inode->i_opflags, that says "this has not special
355 * permission function, use the fast case".
356 */
357static inline int do_inode_permission(struct inode *inode, int mask)
358{
359 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
360 if (likely(inode->i_op->permission))
361 return inode->i_op->permission(inode, mask);
362
363 /* This gets set once for the inode lifetime */
364 spin_lock(&inode->i_lock);
365 inode->i_opflags |= IOP_FASTPERM;
366 spin_unlock(&inode->i_lock);
367 }
368 return generic_permission(inode, mask);
369}
370
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200371/**
David Howells0bdaea92012-06-25 12:55:46 +0100372 * __inode_permission - Check for access rights to a given inode
373 * @inode: Inode to check permission on
374 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200375 *
David Howells0bdaea92012-06-25 12:55:46 +0100376 * Check for read/write/execute permissions on an inode.
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530377 *
378 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
David Howells0bdaea92012-06-25 12:55:46 +0100379 *
380 * This does not check for a read-only file system. You probably want
381 * inode_permission().
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200382 */
David Howells0bdaea92012-06-25 12:55:46 +0100383int __inode_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384{
Al Viroe6305c42008-07-15 21:03:57 -0400385 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700387 if (unlikely(mask & MAY_WRITE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 /*
389 * Nobody gets write access to an immutable file.
390 */
391 if (IS_IMMUTABLE(inode))
392 return -EACCES;
393 }
394
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700395 retval = do_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 if (retval)
397 return retval;
398
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -0700399 retval = devcgroup_inode_permission(inode, mask);
400 if (retval)
401 return retval;
402
Eric Parisd09ca732010-07-23 11:43:57 -0400403 return security_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404}
405
Al Virof4d6ff82011-06-19 13:14:21 -0400406/**
David Howells0bdaea92012-06-25 12:55:46 +0100407 * sb_permission - Check superblock-level permissions
408 * @sb: Superblock of inode to check permission on
Randy Dunlap55852632012-08-18 17:39:25 -0700409 * @inode: Inode to check permission on
David Howells0bdaea92012-06-25 12:55:46 +0100410 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
411 *
412 * Separate out file-system wide checks from inode-specific permission checks.
413 */
414static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
415{
416 if (unlikely(mask & MAY_WRITE)) {
417 umode_t mode = inode->i_mode;
418
419 /* Nobody gets write access to a read-only fs. */
420 if ((sb->s_flags & MS_RDONLY) &&
421 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
422 return -EROFS;
423 }
424 return 0;
425}
426
427/**
428 * inode_permission - Check for access rights to a given inode
429 * @inode: Inode to check permission on
430 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
431 *
432 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
433 * this, letting us set arbitrary permissions for filesystem access without
434 * changing the "normal" UIDs which are used for other things.
435 *
436 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
437 */
438int inode_permission(struct inode *inode, int mask)
439{
440 int retval;
441
442 retval = sb_permission(inode->i_sb, inode, mask);
443 if (retval)
444 return retval;
445 return __inode_permission(inode, mask);
446}
447
448/**
Jan Blunck5dd784d02008-02-14 19:34:38 -0800449 * path_get - get a reference to a path
450 * @path: path to get the reference to
451 *
452 * Given a path increment the reference count to the dentry and the vfsmount.
453 */
454void path_get(struct path *path)
455{
456 mntget(path->mnt);
457 dget(path->dentry);
458}
459EXPORT_SYMBOL(path_get);
460
461/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800462 * path_put - put a reference to a path
463 * @path: path to put the reference to
464 *
465 * Given a path decrement the reference count to the dentry and the vfsmount.
466 */
467void path_put(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468{
Jan Blunck1d957f92008-02-14 19:34:35 -0800469 dput(path->dentry);
470 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471}
Jan Blunck1d957f92008-02-14 19:34:35 -0800472EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473
Al Viro19660af2011-03-25 10:32:48 -0400474/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100475 * Path walking has 2 modes, rcu-walk and ref-walk (see
Al Viro19660af2011-03-25 10:32:48 -0400476 * Documentation/filesystems/path-lookup.txt). In situations when we can't
477 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
478 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
479 * mode. Refcounts are grabbed at the last known good point before rcu-walk
480 * got stuck, so ref-walk may continue from there. If this is not successful
481 * (eg. a seqcount has changed), then failure is returned and it's up to caller
482 * to restart the path walk from the beginning in ref-walk mode.
Nick Piggin31e6b012011-01-07 17:49:52 +1100483 */
Nick Piggin31e6b012011-01-07 17:49:52 +1100484
Al Viro32a79912012-07-18 20:43:19 +0400485static inline void lock_rcu_walk(void)
486{
487 br_read_lock(&vfsmount_lock);
488 rcu_read_lock();
489}
490
491static inline void unlock_rcu_walk(void)
492{
493 rcu_read_unlock();
494 br_read_unlock(&vfsmount_lock);
495}
496
Nick Piggin31e6b012011-01-07 17:49:52 +1100497/**
Al Viro19660af2011-03-25 10:32:48 -0400498 * unlazy_walk - try to switch to ref-walk mode.
499 * @nd: nameidata pathwalk data
500 * @dentry: child of nd->path.dentry or NULL
Randy Dunlap39191622011-01-08 19:36:21 -0800501 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100502 *
Al Viro19660af2011-03-25 10:32:48 -0400503 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
504 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
505 * @nd or NULL. Must be called from rcu-walk context.
Nick Piggin31e6b012011-01-07 17:49:52 +1100506 */
Al Viro19660af2011-03-25 10:32:48 -0400507static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
Nick Piggin31e6b012011-01-07 17:49:52 +1100508{
509 struct fs_struct *fs = current->fs;
510 struct dentry *parent = nd->path.dentry;
Al Viro5b6ca022011-03-09 23:04:47 -0500511 int want_root = 0;
Nick Piggin31e6b012011-01-07 17:49:52 +1100512
513 BUG_ON(!(nd->flags & LOOKUP_RCU));
Al Viro5b6ca022011-03-09 23:04:47 -0500514 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
515 want_root = 1;
Nick Piggin31e6b012011-01-07 17:49:52 +1100516 spin_lock(&fs->lock);
517 if (nd->root.mnt != fs->root.mnt ||
518 nd->root.dentry != fs->root.dentry)
519 goto err_root;
520 }
521 spin_lock(&parent->d_lock);
Al Viro19660af2011-03-25 10:32:48 -0400522 if (!dentry) {
523 if (!__d_rcu_to_refcount(parent, nd->seq))
524 goto err_parent;
525 BUG_ON(nd->inode != parent->d_inode);
526 } else {
Al Viro94c0d4e2011-07-12 21:40:23 -0400527 if (dentry->d_parent != parent)
528 goto err_parent;
Al Viro19660af2011-03-25 10:32:48 -0400529 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
530 if (!__d_rcu_to_refcount(dentry, nd->seq))
531 goto err_child;
532 /*
533 * If the sequence check on the child dentry passed, then
534 * the child has not been removed from its parent. This
535 * means the parent dentry must be valid and able to take
536 * a reference at this point.
537 */
538 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
539 BUG_ON(!parent->d_count);
540 parent->d_count++;
541 spin_unlock(&dentry->d_lock);
542 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100543 spin_unlock(&parent->d_lock);
Al Viro5b6ca022011-03-09 23:04:47 -0500544 if (want_root) {
Nick Piggin31e6b012011-01-07 17:49:52 +1100545 path_get(&nd->root);
546 spin_unlock(&fs->lock);
547 }
548 mntget(nd->path.mnt);
549
Al Viro32a79912012-07-18 20:43:19 +0400550 unlock_rcu_walk();
Nick Piggin31e6b012011-01-07 17:49:52 +1100551 nd->flags &= ~LOOKUP_RCU;
552 return 0;
Al Viro19660af2011-03-25 10:32:48 -0400553
554err_child:
Nick Piggin31e6b012011-01-07 17:49:52 +1100555 spin_unlock(&dentry->d_lock);
Al Viro19660af2011-03-25 10:32:48 -0400556err_parent:
Nick Piggin31e6b012011-01-07 17:49:52 +1100557 spin_unlock(&parent->d_lock);
558err_root:
Al Viro5b6ca022011-03-09 23:04:47 -0500559 if (want_root)
Nick Piggin31e6b012011-01-07 17:49:52 +1100560 spin_unlock(&fs->lock);
561 return -ECHILD;
562}
563
Al Viro4ce16ef32012-06-10 16:10:59 -0400564static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
Nick Piggin34286d62011-01-07 17:49:57 +1100565{
Al Viro4ce16ef32012-06-10 16:10:59 -0400566 return dentry->d_op->d_revalidate(dentry, flags);
Nick Piggin34286d62011-01-07 17:49:57 +1100567}
568
Al Viro9f1fafe2011-03-25 11:00:12 -0400569/**
570 * complete_walk - successful completion of path walk
571 * @nd: pointer nameidata
Jeff Layton39159de2009-12-07 12:01:50 -0500572 *
Al Viro9f1fafe2011-03-25 11:00:12 -0400573 * If we had been in RCU mode, drop out of it and legitimize nd->path.
574 * Revalidate the final result, unless we'd already done that during
575 * the path walk or the filesystem doesn't ask for it. Return 0 on
576 * success, -error on failure. In case of failure caller does not
577 * need to drop nd->path.
Jeff Layton39159de2009-12-07 12:01:50 -0500578 */
Al Viro9f1fafe2011-03-25 11:00:12 -0400579static int complete_walk(struct nameidata *nd)
Jeff Layton39159de2009-12-07 12:01:50 -0500580{
Al Viro16c2cd72011-02-22 15:50:10 -0500581 struct dentry *dentry = nd->path.dentry;
Jeff Layton39159de2009-12-07 12:01:50 -0500582 int status;
Jeff Layton39159de2009-12-07 12:01:50 -0500583
Al Viro9f1fafe2011-03-25 11:00:12 -0400584 if (nd->flags & LOOKUP_RCU) {
585 nd->flags &= ~LOOKUP_RCU;
586 if (!(nd->flags & LOOKUP_ROOT))
587 nd->root.mnt = NULL;
588 spin_lock(&dentry->d_lock);
589 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
590 spin_unlock(&dentry->d_lock);
Al Viro32a79912012-07-18 20:43:19 +0400591 unlock_rcu_walk();
Al Viro9f1fafe2011-03-25 11:00:12 -0400592 return -ECHILD;
593 }
594 BUG_ON(nd->inode != dentry->d_inode);
595 spin_unlock(&dentry->d_lock);
596 mntget(nd->path.mnt);
Al Viro32a79912012-07-18 20:43:19 +0400597 unlock_rcu_walk();
Al Viro9f1fafe2011-03-25 11:00:12 -0400598 }
599
Al Viro16c2cd72011-02-22 15:50:10 -0500600 if (likely(!(nd->flags & LOOKUP_JUMPED)))
Jeff Layton39159de2009-12-07 12:01:50 -0500601 return 0;
602
Al Viro16c2cd72011-02-22 15:50:10 -0500603 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
604 return 0;
605
606 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
607 return 0;
608
609 /* Note: we do not d_invalidate() */
Al Viro4ce16ef32012-06-10 16:10:59 -0400610 status = d_revalidate(dentry, nd->flags);
Jeff Layton39159de2009-12-07 12:01:50 -0500611 if (status > 0)
612 return 0;
613
Al Viro16c2cd72011-02-22 15:50:10 -0500614 if (!status)
Jeff Layton39159de2009-12-07 12:01:50 -0500615 status = -ESTALE;
Al Viro16c2cd72011-02-22 15:50:10 -0500616
Al Viro9f1fafe2011-03-25 11:00:12 -0400617 path_put(&nd->path);
Jeff Layton39159de2009-12-07 12:01:50 -0500618 return status;
619}
620
Al Viro2a737872009-04-07 11:49:53 -0400621static __always_inline void set_root(struct nameidata *nd)
622{
Miklos Szeredif7ad3c62010-08-10 11:41:36 +0200623 if (!nd->root.mnt)
624 get_fs_root(current->fs, &nd->root);
Al Viro2a737872009-04-07 11:49:53 -0400625}
626
Al Viro6de88d72009-08-09 01:41:57 +0400627static int link_path_walk(const char *, struct nameidata *);
628
Nick Piggin31e6b012011-01-07 17:49:52 +1100629static __always_inline void set_root_rcu(struct nameidata *nd)
630{
631 if (!nd->root.mnt) {
632 struct fs_struct *fs = current->fs;
Nick Pigginc28cc362011-01-07 17:49:53 +1100633 unsigned seq;
634
635 do {
636 seq = read_seqcount_begin(&fs->seq);
637 nd->root = fs->root;
Tim Chenc1530012011-04-15 11:39:29 -0700638 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
Nick Pigginc28cc362011-01-07 17:49:53 +1100639 } while (read_seqcount_retry(&fs->seq, seq));
Nick Piggin31e6b012011-01-07 17:49:52 +1100640 }
641}
642
Arjan van de Venf1662352006-01-14 13:21:31 -0800643static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644{
Nick Piggin31e6b012011-01-07 17:49:52 +1100645 int ret;
646
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 if (IS_ERR(link))
648 goto fail;
649
650 if (*link == '/') {
Al Viro2a737872009-04-07 11:49:53 -0400651 set_root(nd);
Jan Blunck1d957f92008-02-14 19:34:35 -0800652 path_put(&nd->path);
Al Viro2a737872009-04-07 11:49:53 -0400653 nd->path = nd->root;
654 path_get(&nd->root);
Al Viro16c2cd72011-02-22 15:50:10 -0500655 nd->flags |= LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100657 nd->inode = nd->path.dentry->d_inode;
Christoph Hellwigb4091d52008-11-05 15:07:21 +0100658
Nick Piggin31e6b012011-01-07 17:49:52 +1100659 ret = link_path_walk(link, nd);
660 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661fail:
Jan Blunck1d957f92008-02-14 19:34:35 -0800662 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 return PTR_ERR(link);
664}
665
Jan Blunck1d957f92008-02-14 19:34:35 -0800666static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700667{
668 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800669 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700670 mntput(path->mnt);
671}
672
Nick Piggin7b9337a2011-01-14 08:42:43 +0000673static inline void path_to_nameidata(const struct path *path,
674 struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700675{
Nick Piggin31e6b012011-01-07 17:49:52 +1100676 if (!(nd->flags & LOOKUP_RCU)) {
677 dput(nd->path.dentry);
678 if (nd->path.mnt != path->mnt)
679 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800680 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100681 nd->path.mnt = path->mnt;
Jan Blunck4ac91372008-02-14 19:34:32 -0800682 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700683}
684
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400685/*
686 * Helper to directly jump to a known parsed path from ->follow_link,
687 * caller must have taken a reference to path beforehand.
688 */
689void nd_jump_link(struct nameidata *nd, struct path *path)
690{
691 path_put(&nd->path);
692
693 nd->path = *path;
694 nd->inode = nd->path.dentry->d_inode;
695 nd->flags |= LOOKUP_JUMPED;
696
697 BUG_ON(nd->inode->i_op->follow_link);
698}
699
Al Viro574197e2011-03-14 22:20:34 -0400700static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
701{
702 struct inode *inode = link->dentry->d_inode;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400703 if (inode->i_op->put_link)
Al Viro574197e2011-03-14 22:20:34 -0400704 inode->i_op->put_link(link->dentry, nd, cookie);
705 path_put(link);
706}
707
Linus Torvalds561ec642012-10-26 10:05:07 -0700708int sysctl_protected_symlinks __read_mostly = 0;
709int sysctl_protected_hardlinks __read_mostly = 0;
Kees Cook800179c2012-07-25 17:29:07 -0700710
711/**
712 * may_follow_link - Check symlink following for unsafe situations
713 * @link: The path of the symlink
Randy Dunlap55852632012-08-18 17:39:25 -0700714 * @nd: nameidata pathwalk data
Kees Cook800179c2012-07-25 17:29:07 -0700715 *
716 * In the case of the sysctl_protected_symlinks sysctl being enabled,
717 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
718 * in a sticky world-writable directory. This is to protect privileged
719 * processes from failing races against path names that may change out
720 * from under them by way of other users creating malicious symlinks.
721 * It will permit symlinks to be followed only when outside a sticky
722 * world-writable directory, or when the uid of the symlink and follower
723 * match, or when the directory owner matches the symlink's owner.
724 *
725 * Returns 0 if following the symlink is allowed, -ve on error.
726 */
727static inline int may_follow_link(struct path *link, struct nameidata *nd)
728{
729 const struct inode *inode;
730 const struct inode *parent;
731
732 if (!sysctl_protected_symlinks)
733 return 0;
734
735 /* Allowed if owner and follower match. */
736 inode = link->dentry->d_inode;
Eric W. Biederman81abe272012-08-03 09:38:08 -0700737 if (uid_eq(current_cred()->fsuid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700738 return 0;
739
740 /* Allowed if parent directory not sticky and world-writable. */
741 parent = nd->path.dentry->d_inode;
742 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
743 return 0;
744
745 /* Allowed if parent directory and link owner match. */
Eric W. Biederman81abe272012-08-03 09:38:08 -0700746 if (uid_eq(parent->i_uid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700747 return 0;
748
Sasha Levinffd8d102012-10-04 19:56:40 -0400749 audit_log_link_denied("follow_link", link);
Kees Cook800179c2012-07-25 17:29:07 -0700750 path_put_conditional(link, nd);
751 path_put(&nd->path);
752 return -EACCES;
753}
754
755/**
756 * safe_hardlink_source - Check for safe hardlink conditions
757 * @inode: the source inode to hardlink from
758 *
759 * Return false if at least one of the following conditions:
760 * - inode is not a regular file
761 * - inode is setuid
762 * - inode is setgid and group-exec
763 * - access failure for read and write
764 *
765 * Otherwise returns true.
766 */
767static bool safe_hardlink_source(struct inode *inode)
768{
769 umode_t mode = inode->i_mode;
770
771 /* Special files should not get pinned to the filesystem. */
772 if (!S_ISREG(mode))
773 return false;
774
775 /* Setuid files should not get pinned to the filesystem. */
776 if (mode & S_ISUID)
777 return false;
778
779 /* Executable setgid files should not get pinned to the filesystem. */
780 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
781 return false;
782
783 /* Hardlinking to unreadable or unwritable sources is dangerous. */
784 if (inode_permission(inode, MAY_READ | MAY_WRITE))
785 return false;
786
787 return true;
788}
789
790/**
791 * may_linkat - Check permissions for creating a hardlink
792 * @link: the source to hardlink from
793 *
794 * Block hardlink when all of:
795 * - sysctl_protected_hardlinks enabled
796 * - fsuid does not match inode
797 * - hardlink source is unsafe (see safe_hardlink_source() above)
798 * - not CAP_FOWNER
799 *
800 * Returns 0 if successful, -ve on error.
801 */
802static int may_linkat(struct path *link)
803{
804 const struct cred *cred;
805 struct inode *inode;
806
807 if (!sysctl_protected_hardlinks)
808 return 0;
809
810 cred = current_cred();
811 inode = link->dentry->d_inode;
812
813 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
814 * otherwise, it must be a safe source.
815 */
Eric W. Biederman81abe272012-08-03 09:38:08 -0700816 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
Kees Cook800179c2012-07-25 17:29:07 -0700817 capable(CAP_FOWNER))
818 return 0;
819
Kees Cooka51d9ea2012-07-25 17:29:08 -0700820 audit_log_link_denied("linkat", link);
Kees Cook800179c2012-07-25 17:29:07 -0700821 return -EPERM;
822}
823
Al Virodef4af32009-12-26 08:37:05 -0500824static __always_inline int
Al Viro574197e2011-03-14 22:20:34 -0400825follow_link(struct path *link, struct nameidata *nd, void **p)
Ian Kent051d3812006-03-27 01:14:53 -0800826{
Nick Piggin7b9337a2011-01-14 08:42:43 +0000827 struct dentry *dentry = link->dentry;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400828 int error;
829 char *s;
Ian Kent051d3812006-03-27 01:14:53 -0800830
Al Viro844a3912011-02-15 00:38:26 -0500831 BUG_ON(nd->flags & LOOKUP_RCU);
832
Al Viro0e794582011-03-16 02:45:02 -0400833 if (link->mnt == nd->path.mnt)
834 mntget(link->mnt);
835
Al Viro6d7b5aa2012-06-10 04:15:17 -0400836 error = -ELOOP;
837 if (unlikely(current->total_link_count >= 40))
838 goto out_put_nd_path;
839
Al Viro574197e2011-03-14 22:20:34 -0400840 cond_resched();
841 current->total_link_count++;
842
Al Viro68ac1232012-03-15 08:21:57 -0400843 touch_atime(link);
Ian Kent051d3812006-03-27 01:14:53 -0800844 nd_set_link(nd, NULL);
845
Al Viro36f3b4f2011-02-22 21:24:38 -0500846 error = security_inode_follow_link(link->dentry, nd);
Al Viro6d7b5aa2012-06-10 04:15:17 -0400847 if (error)
848 goto out_put_nd_path;
Al Viro36f3b4f2011-02-22 21:24:38 -0500849
Al Viro86acdca12009-12-22 23:45:11 -0500850 nd->last_type = LAST_BIND;
Al Virodef4af32009-12-26 08:37:05 -0500851 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
852 error = PTR_ERR(*p);
Al Viro6d7b5aa2012-06-10 04:15:17 -0400853 if (IS_ERR(*p))
Christoph Hellwig408ef012012-06-18 10:47:03 -0400854 goto out_put_nd_path;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400855
856 error = 0;
857 s = nd_get_link(nd);
858 if (s) {
859 error = __vfs_follow_link(nd, s);
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400860 if (unlikely(error))
861 put_link(nd, link, *p);
Ian Kent051d3812006-03-27 01:14:53 -0800862 }
Al Viro6d7b5aa2012-06-10 04:15:17 -0400863
864 return error;
865
866out_put_nd_path:
Arnd Bergmann98f6ef62012-10-11 13:20:00 +0000867 *p = NULL;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400868 path_put(&nd->path);
Al Viro6d7b5aa2012-06-10 04:15:17 -0400869 path_put(link);
Ian Kent051d3812006-03-27 01:14:53 -0800870 return error;
871}
872
Nick Piggin31e6b012011-01-07 17:49:52 +1100873static int follow_up_rcu(struct path *path)
874{
Al Viro0714a532011-11-24 22:19:58 -0500875 struct mount *mnt = real_mount(path->mnt);
876 struct mount *parent;
Nick Piggin31e6b012011-01-07 17:49:52 +1100877 struct dentry *mountpoint;
878
Al Viro0714a532011-11-24 22:19:58 -0500879 parent = mnt->mnt_parent;
880 if (&parent->mnt == path->mnt)
Nick Piggin31e6b012011-01-07 17:49:52 +1100881 return 0;
Al Viroa73324d2011-11-24 22:25:07 -0500882 mountpoint = mnt->mnt_mountpoint;
Nick Piggin31e6b012011-01-07 17:49:52 +1100883 path->dentry = mountpoint;
Al Viro0714a532011-11-24 22:19:58 -0500884 path->mnt = &parent->mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +1100885 return 1;
886}
887
David Howellsf015f1262012-06-25 12:55:28 +0100888/*
889 * follow_up - Find the mountpoint of path's vfsmount
890 *
891 * Given a path, find the mountpoint of its source file system.
892 * Replace @path with the path of the mountpoint in the parent mount.
893 * Up is towards /.
894 *
895 * Return 1 if we went up a level and 0 if we were already at the
896 * root.
897 */
Al Virobab77eb2009-04-18 03:26:48 -0400898int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899{
Al Viro0714a532011-11-24 22:19:58 -0500900 struct mount *mnt = real_mount(path->mnt);
901 struct mount *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +1000903
Andi Kleen962830d2012-05-08 13:32:02 +0930904 br_read_lock(&vfsmount_lock);
Al Viro0714a532011-11-24 22:19:58 -0500905 parent = mnt->mnt_parent;
Al Viro3c0a6162012-07-18 17:32:50 +0400906 if (parent == mnt) {
Andi Kleen962830d2012-05-08 13:32:02 +0930907 br_read_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 return 0;
909 }
Al Viro0714a532011-11-24 22:19:58 -0500910 mntget(&parent->mnt);
Al Viroa73324d2011-11-24 22:25:07 -0500911 mountpoint = dget(mnt->mnt_mountpoint);
Andi Kleen962830d2012-05-08 13:32:02 +0930912 br_read_unlock(&vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400913 dput(path->dentry);
914 path->dentry = mountpoint;
915 mntput(path->mnt);
Al Viro0714a532011-11-24 22:19:58 -0500916 path->mnt = &parent->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 return 1;
918}
919
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100920/*
David Howells9875cf82011-01-14 18:45:21 +0000921 * Perform an automount
922 * - return -EISDIR to tell follow_managed() to stop and return the path we
923 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 */
David Howells9875cf82011-01-14 18:45:21 +0000925static int follow_automount(struct path *path, unsigned flags,
926 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +1100927{
David Howells9875cf82011-01-14 18:45:21 +0000928 struct vfsmount *mnt;
David Howellsea5b7782011-01-14 19:10:03 +0000929 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +1100930
David Howells9875cf82011-01-14 18:45:21 +0000931 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
932 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -0700933
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200934 /* We don't want to mount if someone's just doing a stat -
935 * unless they're stat'ing a directory and appended a '/' to
936 * the name.
937 *
938 * We do, however, want to mount if someone wants to open or
939 * create a file of any type under the mountpoint, wants to
940 * traverse through the mountpoint or wants to open the
941 * mounted directory. Also, autofs may mark negative dentries
942 * as being automount points. These will need the attentions
943 * of the daemon to instantiate them before they can be used.
David Howells9875cf82011-01-14 18:45:21 +0000944 */
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200945 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
Linus Torvaldsd94c1772011-09-26 17:44:55 -0700946 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200947 path->dentry->d_inode)
948 return -EISDIR;
949
David Howells9875cf82011-01-14 18:45:21 +0000950 current->total_link_count++;
951 if (current->total_link_count >= 40)
952 return -ELOOP;
953
954 mnt = path->dentry->d_op->d_automount(path);
955 if (IS_ERR(mnt)) {
956 /*
957 * The filesystem is allowed to return -EISDIR here to indicate
958 * it doesn't want to automount. For instance, autofs would do
959 * this so that its userspace daemon can mount on this dentry.
960 *
961 * However, we can only permit this if it's a terminal point in
962 * the path being looked up; if it wasn't then the remainder of
963 * the path is inaccessible and we should say so.
964 */
Al Viro49084c32011-06-25 21:59:52 -0400965 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
David Howells9875cf82011-01-14 18:45:21 +0000966 return -EREMOTE;
967 return PTR_ERR(mnt);
968 }
David Howellsea5b7782011-01-14 19:10:03 +0000969
David Howells9875cf82011-01-14 18:45:21 +0000970 if (!mnt) /* mount collision */
971 return 0;
972
Al Viro8aef1882011-06-16 15:10:06 +0100973 if (!*need_mntput) {
974 /* lock_mount() may release path->mnt on error */
975 mntget(path->mnt);
976 *need_mntput = true;
977 }
Al Viro19a167a2011-01-17 01:35:23 -0500978 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +0000979
David Howellsea5b7782011-01-14 19:10:03 +0000980 switch (err) {
981 case -EBUSY:
982 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -0500983 return 0;
David Howellsea5b7782011-01-14 19:10:03 +0000984 case 0:
Al Viro8aef1882011-06-16 15:10:06 +0100985 path_put(path);
David Howellsea5b7782011-01-14 19:10:03 +0000986 path->mnt = mnt;
987 path->dentry = dget(mnt->mnt_root);
David Howellsea5b7782011-01-14 19:10:03 +0000988 return 0;
Al Viro19a167a2011-01-17 01:35:23 -0500989 default:
990 return err;
David Howellsea5b7782011-01-14 19:10:03 +0000991 }
Al Viro19a167a2011-01-17 01:35:23 -0500992
David Howells9875cf82011-01-14 18:45:21 +0000993}
994
995/*
996 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +0000997 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +0000998 * - Flagged as mountpoint
999 * - Flagged as automount point
1000 *
1001 * This may only be called in refwalk mode.
1002 *
1003 * Serialization is taken care of in namespace.c
1004 */
1005static int follow_managed(struct path *path, unsigned flags)
1006{
Al Viro8aef1882011-06-16 15:10:06 +01001007 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
David Howells9875cf82011-01-14 18:45:21 +00001008 unsigned managed;
1009 bool need_mntput = false;
Al Viro8aef1882011-06-16 15:10:06 +01001010 int ret = 0;
David Howells9875cf82011-01-14 18:45:21 +00001011
1012 /* Given that we're not holding a lock here, we retain the value in a
1013 * local variable for each dentry as we look at it so that we don't see
1014 * the components of that value change under us */
1015 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1016 managed &= DCACHE_MANAGED_DENTRY,
1017 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +00001018 /* Allow the filesystem to manage the transit without i_mutex
1019 * being held. */
1020 if (managed & DCACHE_MANAGE_TRANSIT) {
1021 BUG_ON(!path->dentry->d_op);
1022 BUG_ON(!path->dentry->d_op->d_manage);
Al Viro1aed3e42011-03-18 09:09:02 -04001023 ret = path->dentry->d_op->d_manage(path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001024 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001025 break;
David Howellscc53ce52011-01-14 18:45:26 +00001026 }
1027
David Howells9875cf82011-01-14 18:45:21 +00001028 /* Transit to a mounted filesystem. */
1029 if (managed & DCACHE_MOUNTED) {
1030 struct vfsmount *mounted = lookup_mnt(path);
1031 if (mounted) {
1032 dput(path->dentry);
1033 if (need_mntput)
1034 mntput(path->mnt);
1035 path->mnt = mounted;
1036 path->dentry = dget(mounted->mnt_root);
1037 need_mntput = true;
1038 continue;
1039 }
1040
1041 /* Something is mounted on this dentry in another
1042 * namespace and/or whatever was mounted there in this
1043 * namespace got unmounted before we managed to get the
1044 * vfsmount_lock */
1045 }
1046
1047 /* Handle an automount point */
1048 if (managed & DCACHE_NEED_AUTOMOUNT) {
1049 ret = follow_automount(path, flags, &need_mntput);
1050 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001051 break;
David Howells9875cf82011-01-14 18:45:21 +00001052 continue;
1053 }
1054
1055 /* We didn't change the current path point */
1056 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 }
Al Viro8aef1882011-06-16 15:10:06 +01001058
1059 if (need_mntput && path->mnt == mnt)
1060 mntput(path->mnt);
1061 if (ret == -EISDIR)
1062 ret = 0;
Al Viroa3fbbde2011-11-07 21:21:26 +00001063 return ret < 0 ? ret : need_mntput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064}
1065
David Howellscc53ce52011-01-14 18:45:26 +00001066int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067{
1068 struct vfsmount *mounted;
1069
Al Viro1c755af2009-04-18 14:06:57 -04001070 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -04001072 dput(path->dentry);
1073 mntput(path->mnt);
1074 path->mnt = mounted;
1075 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 return 1;
1077 }
1078 return 0;
1079}
1080
Ian Kent62a73752011-03-25 01:51:02 +08001081static inline bool managed_dentry_might_block(struct dentry *dentry)
1082{
1083 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
1084 dentry->d_op->d_manage(dentry, true) < 0);
1085}
1086
David Howells9875cf82011-01-14 18:45:21 +00001087/*
Al Viro287548e2011-05-27 06:50:06 -04001088 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1089 * we meet a managed dentry that would need blocking.
David Howells9875cf82011-01-14 18:45:21 +00001090 */
1091static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
Al Viro287548e2011-05-27 06:50:06 -04001092 struct inode **inode)
David Howells9875cf82011-01-14 18:45:21 +00001093{
Ian Kent62a73752011-03-25 01:51:02 +08001094 for (;;) {
Al Viroc7105362011-11-24 18:22:03 -05001095 struct mount *mounted;
Ian Kent62a73752011-03-25 01:51:02 +08001096 /*
1097 * Don't forget we might have a non-mountpoint managed dentry
1098 * that wants to block transit.
1099 */
Al Viro287548e2011-05-27 06:50:06 -04001100 if (unlikely(managed_dentry_might_block(path->dentry)))
David Howellsab909112011-01-14 18:46:51 +00001101 return false;
Ian Kent62a73752011-03-25 01:51:02 +08001102
1103 if (!d_mountpoint(path->dentry))
1104 break;
1105
David Howells9875cf82011-01-14 18:45:21 +00001106 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
1107 if (!mounted)
1108 break;
Al Viroc7105362011-11-24 18:22:03 -05001109 path->mnt = &mounted->mnt;
1110 path->dentry = mounted->mnt.mnt_root;
Al Viroa3fbbde2011-11-07 21:21:26 +00001111 nd->flags |= LOOKUP_JUMPED;
David Howells9875cf82011-01-14 18:45:21 +00001112 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
Linus Torvalds59430262011-07-18 15:43:29 -07001113 /*
1114 * Update the inode too. We don't need to re-check the
1115 * dentry sequence number here after this d_inode read,
1116 * because a mount-point is always pinned.
1117 */
1118 *inode = path->dentry->d_inode;
David Howells9875cf82011-01-14 18:45:21 +00001119 }
David Howells9875cf82011-01-14 18:45:21 +00001120 return true;
1121}
1122
Al Virodea39372011-05-27 06:53:39 -04001123static void follow_mount_rcu(struct nameidata *nd)
Al Viro287548e2011-05-27 06:50:06 -04001124{
Al Virodea39372011-05-27 06:53:39 -04001125 while (d_mountpoint(nd->path.dentry)) {
Al Viroc7105362011-11-24 18:22:03 -05001126 struct mount *mounted;
Al Virodea39372011-05-27 06:53:39 -04001127 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
Al Viro287548e2011-05-27 06:50:06 -04001128 if (!mounted)
1129 break;
Al Viroc7105362011-11-24 18:22:03 -05001130 nd->path.mnt = &mounted->mnt;
1131 nd->path.dentry = mounted->mnt.mnt_root;
Al Virodea39372011-05-27 06:53:39 -04001132 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro287548e2011-05-27 06:50:06 -04001133 }
1134}
1135
Nick Piggin31e6b012011-01-07 17:49:52 +11001136static int follow_dotdot_rcu(struct nameidata *nd)
1137{
Nick Piggin31e6b012011-01-07 17:49:52 +11001138 set_root_rcu(nd);
1139
David Howells9875cf82011-01-14 18:45:21 +00001140 while (1) {
Nick Piggin31e6b012011-01-07 17:49:52 +11001141 if (nd->path.dentry == nd->root.dentry &&
1142 nd->path.mnt == nd->root.mnt) {
1143 break;
1144 }
1145 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1146 struct dentry *old = nd->path.dentry;
1147 struct dentry *parent = old->d_parent;
1148 unsigned seq;
1149
1150 seq = read_seqcount_begin(&parent->d_seq);
1151 if (read_seqcount_retry(&old->d_seq, nd->seq))
Al Viroef7562d2011-03-04 14:35:59 -05001152 goto failed;
Nick Piggin31e6b012011-01-07 17:49:52 +11001153 nd->path.dentry = parent;
1154 nd->seq = seq;
1155 break;
1156 }
1157 if (!follow_up_rcu(&nd->path))
1158 break;
1159 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11001160 }
Al Virodea39372011-05-27 06:53:39 -04001161 follow_mount_rcu(nd);
1162 nd->inode = nd->path.dentry->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001163 return 0;
Al Viroef7562d2011-03-04 14:35:59 -05001164
1165failed:
1166 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001167 if (!(nd->flags & LOOKUP_ROOT))
1168 nd->root.mnt = NULL;
Al Viro32a79912012-07-18 20:43:19 +04001169 unlock_rcu_walk();
Al Viroef7562d2011-03-04 14:35:59 -05001170 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001171}
1172
David Howells9875cf82011-01-14 18:45:21 +00001173/*
David Howellscc53ce52011-01-14 18:45:26 +00001174 * Follow down to the covering mount currently visible to userspace. At each
1175 * point, the filesystem owning that dentry may be queried as to whether the
1176 * caller is permitted to proceed or not.
David Howellscc53ce52011-01-14 18:45:26 +00001177 */
Al Viro7cc90cc2011-03-18 09:04:20 -04001178int follow_down(struct path *path)
David Howellscc53ce52011-01-14 18:45:26 +00001179{
1180 unsigned managed;
1181 int ret;
1182
1183 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1184 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1185 /* Allow the filesystem to manage the transit without i_mutex
1186 * being held.
1187 *
1188 * We indicate to the filesystem if someone is trying to mount
1189 * something here. This gives autofs the chance to deny anyone
1190 * other than its daemon the right to mount on its
1191 * superstructure.
1192 *
1193 * The filesystem may sleep at this point.
1194 */
1195 if (managed & DCACHE_MANAGE_TRANSIT) {
1196 BUG_ON(!path->dentry->d_op);
1197 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +00001198 ret = path->dentry->d_op->d_manage(
Al Viro1aed3e42011-03-18 09:09:02 -04001199 path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001200 if (ret < 0)
1201 return ret == -EISDIR ? 0 : ret;
1202 }
1203
1204 /* Transit to a mounted filesystem. */
1205 if (managed & DCACHE_MOUNTED) {
1206 struct vfsmount *mounted = lookup_mnt(path);
1207 if (!mounted)
1208 break;
1209 dput(path->dentry);
1210 mntput(path->mnt);
1211 path->mnt = mounted;
1212 path->dentry = dget(mounted->mnt_root);
1213 continue;
1214 }
1215
1216 /* Don't handle automount points here */
1217 break;
1218 }
1219 return 0;
1220}
1221
1222/*
David Howells9875cf82011-01-14 18:45:21 +00001223 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1224 */
1225static void follow_mount(struct path *path)
1226{
1227 while (d_mountpoint(path->dentry)) {
1228 struct vfsmount *mounted = lookup_mnt(path);
1229 if (!mounted)
1230 break;
1231 dput(path->dentry);
1232 mntput(path->mnt);
1233 path->mnt = mounted;
1234 path->dentry = dget(mounted->mnt_root);
1235 }
1236}
1237
Nick Piggin31e6b012011-01-07 17:49:52 +11001238static void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239{
Al Viro2a737872009-04-07 11:49:53 -04001240 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001241
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 while(1) {
Jan Blunck4ac91372008-02-14 19:34:32 -08001243 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244
Al Viro2a737872009-04-07 11:49:53 -04001245 if (nd->path.dentry == nd->root.dentry &&
1246 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 break;
1248 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001249 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Al Viro3088dd72010-01-30 15:47:29 -05001250 /* rare case of legitimate dget_parent()... */
1251 nd->path.dentry = dget_parent(nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 dput(old);
1253 break;
1254 }
Al Viro3088dd72010-01-30 15:47:29 -05001255 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 }
Al Viro79ed0222009-04-18 13:59:41 -04001258 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001259 nd->inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260}
1261
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262/*
Miklos Szeredibad61182012-03-26 12:54:24 +02001263 * This looks up the name in dcache, possibly revalidates the old dentry and
1264 * allocates a new one if not found or not valid. In the need_lookup argument
1265 * returns whether i_op->lookup is necessary.
1266 *
1267 * dir->d_inode->i_mutex must be held
Nick Pigginbaa03892010-08-18 04:37:31 +10001268 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001269static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
Al Viro201f9562012-06-22 12:42:10 +04001270 unsigned int flags, bool *need_lookup)
Nick Pigginbaa03892010-08-18 04:37:31 +10001271{
Nick Pigginbaa03892010-08-18 04:37:31 +10001272 struct dentry *dentry;
Miklos Szeredibad61182012-03-26 12:54:24 +02001273 int error;
Nick Pigginbaa03892010-08-18 04:37:31 +10001274
Miklos Szeredibad61182012-03-26 12:54:24 +02001275 *need_lookup = false;
1276 dentry = d_lookup(dir, name);
1277 if (dentry) {
1278 if (d_need_lookup(dentry)) {
1279 *need_lookup = true;
1280 } else if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
Al Viro201f9562012-06-22 12:42:10 +04001281 error = d_revalidate(dentry, flags);
Miklos Szeredibad61182012-03-26 12:54:24 +02001282 if (unlikely(error <= 0)) {
1283 if (error < 0) {
1284 dput(dentry);
1285 return ERR_PTR(error);
1286 } else if (!d_invalidate(dentry)) {
1287 dput(dentry);
1288 dentry = NULL;
1289 }
1290 }
1291 }
1292 }
Nick Pigginbaa03892010-08-18 04:37:31 +10001293
Miklos Szeredibad61182012-03-26 12:54:24 +02001294 if (!dentry) {
1295 dentry = d_alloc(dir, name);
1296 if (unlikely(!dentry))
1297 return ERR_PTR(-ENOMEM);
Nick Pigginbaa03892010-08-18 04:37:31 +10001298
Miklos Szeredibad61182012-03-26 12:54:24 +02001299 *need_lookup = true;
Nick Pigginbaa03892010-08-18 04:37:31 +10001300 }
1301 return dentry;
1302}
1303
1304/*
Miklos Szeredibad61182012-03-26 12:54:24 +02001305 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1306 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1307 *
1308 * dir->d_inode->i_mutex must be held
Josef Bacik44396f42011-05-31 11:58:49 -04001309 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001310static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
Al Viro72bd8662012-06-10 17:17:17 -04001311 unsigned int flags)
Josef Bacik44396f42011-05-31 11:58:49 -04001312{
Josef Bacik44396f42011-05-31 11:58:49 -04001313 struct dentry *old;
1314
1315 /* Don't create child dentry for a dead directory. */
Miklos Szeredibad61182012-03-26 12:54:24 +02001316 if (unlikely(IS_DEADDIR(dir))) {
Miklos Szeredie188dc02012-02-03 14:25:18 +01001317 dput(dentry);
Josef Bacik44396f42011-05-31 11:58:49 -04001318 return ERR_PTR(-ENOENT);
Miklos Szeredie188dc02012-02-03 14:25:18 +01001319 }
Josef Bacik44396f42011-05-31 11:58:49 -04001320
Al Viro72bd8662012-06-10 17:17:17 -04001321 old = dir->i_op->lookup(dir, dentry, flags);
Josef Bacik44396f42011-05-31 11:58:49 -04001322 if (unlikely(old)) {
1323 dput(dentry);
1324 dentry = old;
1325 }
1326 return dentry;
1327}
1328
Al Viroa3255542012-03-30 14:41:51 -04001329static struct dentry *__lookup_hash(struct qstr *name,
Al Viro72bd8662012-06-10 17:17:17 -04001330 struct dentry *base, unsigned int flags)
Al Viroa3255542012-03-30 14:41:51 -04001331{
Miklos Szeredibad61182012-03-26 12:54:24 +02001332 bool need_lookup;
Al Viroa3255542012-03-30 14:41:51 -04001333 struct dentry *dentry;
1334
Al Viro72bd8662012-06-10 17:17:17 -04001335 dentry = lookup_dcache(name, base, flags, &need_lookup);
Miklos Szeredibad61182012-03-26 12:54:24 +02001336 if (!need_lookup)
1337 return dentry;
Al Viroa3255542012-03-30 14:41:51 -04001338
Al Viro72bd8662012-06-10 17:17:17 -04001339 return lookup_real(base->d_inode, dentry, flags);
Al Viroa3255542012-03-30 14:41:51 -04001340}
1341
Josef Bacik44396f42011-05-31 11:58:49 -04001342/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 * It's more convoluted than I'd like it to be, but... it's still fairly
1344 * small and for now I'd prefer to have fast path as straight as possible.
1345 * It _is_ time-critical.
1346 */
Miklos Szeredi697f5142012-05-21 17:30:05 +02001347static int lookup_fast(struct nameidata *nd, struct qstr *name,
1348 struct path *path, struct inode **inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349{
Jan Blunck4ac91372008-02-14 19:34:32 -08001350 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001351 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001352 int need_reval = 1;
1353 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001354 int err;
1355
Al Viro3cac2602009-08-13 18:27:43 +04001356 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001357 * Rename seqlock is not required here because in the off chance
1358 * of a false negative due to a concurrent rename, we're going to
1359 * do the non-racy lookup, below.
1360 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001361 if (nd->flags & LOOKUP_RCU) {
1362 unsigned seq;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001363 dentry = __d_lookup_rcu(parent, name, &seq, nd->inode);
Al Viro5a18fff2011-03-11 04:44:53 -05001364 if (!dentry)
1365 goto unlazy;
1366
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001367 /*
1368 * This sequence count validates that the inode matches
1369 * the dentry name information from lookup.
1370 */
1371 *inode = dentry->d_inode;
1372 if (read_seqcount_retry(&dentry->d_seq, seq))
1373 return -ECHILD;
1374
1375 /*
1376 * This sequence count validates that the parent had no
1377 * changes while we did the lookup of the dentry above.
1378 *
1379 * The memory barrier in read_seqcount_begin of child is
1380 * enough, we can use __read_seqcount_retry here.
1381 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001382 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1383 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001384 nd->seq = seq;
Al Viro5a18fff2011-03-11 04:44:53 -05001385
Miklos Szeredifa4ee152012-03-26 12:54:19 +02001386 if (unlikely(d_need_lookup(dentry)))
1387 goto unlazy;
Al Viro24643082011-02-15 01:26:22 -05001388 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
Al Viro4ce16ef32012-06-10 16:10:59 -04001389 status = d_revalidate(dentry, nd->flags);
Al Viro5a18fff2011-03-11 04:44:53 -05001390 if (unlikely(status <= 0)) {
1391 if (status != -ECHILD)
1392 need_reval = 0;
1393 goto unlazy;
1394 }
Al Viro24643082011-02-15 01:26:22 -05001395 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001396 path->mnt = mnt;
1397 path->dentry = dentry;
Al Virod6e9bd22011-05-27 07:03:15 -04001398 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1399 goto unlazy;
1400 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1401 goto unlazy;
1402 return 0;
Al Viro5a18fff2011-03-11 04:44:53 -05001403unlazy:
Al Viro19660af2011-03-25 10:32:48 -04001404 if (unlazy_walk(nd, dentry))
1405 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001406 } else {
1407 dentry = __d_lookup(parent, name);
Nick Piggin31e6b012011-01-07 17:49:52 +11001408 }
Al Viro5a18fff2011-03-11 04:44:53 -05001409
Al Viro81e6f522012-03-30 14:48:04 -04001410 if (unlikely(!dentry))
1411 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001412
Al Viro81e6f522012-03-30 14:48:04 -04001413 if (unlikely(d_need_lookup(dentry))) {
1414 dput(dentry);
1415 goto need_lookup;
Al Viro24643082011-02-15 01:26:22 -05001416 }
Al Viro81e6f522012-03-30 14:48:04 -04001417
Al Viro5a18fff2011-03-11 04:44:53 -05001418 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
Al Viro4ce16ef32012-06-10 16:10:59 -04001419 status = d_revalidate(dentry, nd->flags);
Al Viro5a18fff2011-03-11 04:44:53 -05001420 if (unlikely(status <= 0)) {
1421 if (status < 0) {
1422 dput(dentry);
1423 return status;
1424 }
1425 if (!d_invalidate(dentry)) {
1426 dput(dentry);
Al Viro81e6f522012-03-30 14:48:04 -04001427 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001428 }
1429 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001430
David Howells9875cf82011-01-14 18:45:21 +00001431 path->mnt = mnt;
1432 path->dentry = dentry;
1433 err = follow_managed(path, nd->flags);
Ian Kent89312212011-01-18 12:06:10 +08001434 if (unlikely(err < 0)) {
1435 path_put_conditional(path, nd);
David Howells9875cf82011-01-14 18:45:21 +00001436 return err;
Ian Kent89312212011-01-18 12:06:10 +08001437 }
Al Viroa3fbbde2011-11-07 21:21:26 +00001438 if (err)
1439 nd->flags |= LOOKUP_JUMPED;
David Howells9875cf82011-01-14 18:45:21 +00001440 *inode = path->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 return 0;
Al Viro81e6f522012-03-30 14:48:04 -04001442
1443need_lookup:
Miklos Szeredi697f5142012-05-21 17:30:05 +02001444 return 1;
1445}
1446
1447/* Fast lookup failed, do it the slow way */
1448static int lookup_slow(struct nameidata *nd, struct qstr *name,
1449 struct path *path)
1450{
1451 struct dentry *dentry, *parent;
1452 int err;
1453
1454 parent = nd->path.dentry;
Al Viro81e6f522012-03-30 14:48:04 -04001455 BUG_ON(nd->inode != parent->d_inode);
1456
1457 mutex_lock(&parent->d_inode->i_mutex);
Al Viro72bd8662012-06-10 17:17:17 -04001458 dentry = __lookup_hash(name, parent, nd->flags);
Al Viro81e6f522012-03-30 14:48:04 -04001459 mutex_unlock(&parent->d_inode->i_mutex);
1460 if (IS_ERR(dentry))
1461 return PTR_ERR(dentry);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001462 path->mnt = nd->path.mnt;
1463 path->dentry = dentry;
1464 err = follow_managed(path, nd->flags);
1465 if (unlikely(err < 0)) {
1466 path_put_conditional(path, nd);
1467 return err;
1468 }
1469 if (err)
1470 nd->flags |= LOOKUP_JUMPED;
1471 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472}
1473
Al Viro52094c82011-02-21 21:34:47 -05001474static inline int may_lookup(struct nameidata *nd)
1475{
1476 if (nd->flags & LOOKUP_RCU) {
Al Viro4ad5abb2011-06-20 19:57:03 -04001477 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
Al Viro52094c82011-02-21 21:34:47 -05001478 if (err != -ECHILD)
1479 return err;
Al Viro19660af2011-03-25 10:32:48 -04001480 if (unlazy_walk(nd, NULL))
Al Viro52094c82011-02-21 21:34:47 -05001481 return -ECHILD;
1482 }
Al Viro4ad5abb2011-06-20 19:57:03 -04001483 return inode_permission(nd->inode, MAY_EXEC);
Al Viro52094c82011-02-21 21:34:47 -05001484}
1485
Al Viro9856fa12011-03-04 14:22:06 -05001486static inline int handle_dots(struct nameidata *nd, int type)
1487{
1488 if (type == LAST_DOTDOT) {
1489 if (nd->flags & LOOKUP_RCU) {
1490 if (follow_dotdot_rcu(nd))
1491 return -ECHILD;
1492 } else
1493 follow_dotdot(nd);
1494 }
1495 return 0;
1496}
1497
Al Viro951361f2011-03-04 14:44:37 -05001498static void terminate_walk(struct nameidata *nd)
1499{
1500 if (!(nd->flags & LOOKUP_RCU)) {
1501 path_put(&nd->path);
1502 } else {
1503 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001504 if (!(nd->flags & LOOKUP_ROOT))
1505 nd->root.mnt = NULL;
Al Viro32a79912012-07-18 20:43:19 +04001506 unlock_rcu_walk();
Al Viro951361f2011-03-04 14:44:37 -05001507 }
1508}
1509
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001510/*
1511 * Do we need to follow links? We _really_ want to be able
1512 * to do this check without having to look at inode->i_op,
1513 * so we keep a cache of "no, this doesn't need follow_link"
1514 * for the common case.
1515 */
Linus Torvalds7813b942011-08-07 09:53:20 -07001516static inline int should_follow_link(struct inode *inode, int follow)
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001517{
1518 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1519 if (likely(inode->i_op->follow_link))
1520 return follow;
1521
1522 /* This gets set once for the inode lifetime */
1523 spin_lock(&inode->i_lock);
1524 inode->i_opflags |= IOP_NOFOLLOW;
1525 spin_unlock(&inode->i_lock);
1526 }
1527 return 0;
1528}
1529
Al Viroce57dfc2011-03-13 19:58:58 -04001530static inline int walk_component(struct nameidata *nd, struct path *path,
1531 struct qstr *name, int type, int follow)
1532{
1533 struct inode *inode;
1534 int err;
1535 /*
1536 * "." and ".." are special - ".." especially so because it has
1537 * to be able to know about the current root directory and
1538 * parent relationships.
1539 */
1540 if (unlikely(type != LAST_NORM))
1541 return handle_dots(nd, type);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001542 err = lookup_fast(nd, name, path, &inode);
Al Viroce57dfc2011-03-13 19:58:58 -04001543 if (unlikely(err)) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001544 if (err < 0)
1545 goto out_err;
1546
1547 err = lookup_slow(nd, name, path);
1548 if (err < 0)
1549 goto out_err;
1550
1551 inode = path->dentry->d_inode;
Al Viroce57dfc2011-03-13 19:58:58 -04001552 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001553 err = -ENOENT;
1554 if (!inode)
1555 goto out_path_put;
1556
Linus Torvalds7813b942011-08-07 09:53:20 -07001557 if (should_follow_link(inode, follow)) {
Al Viro19660af2011-03-25 10:32:48 -04001558 if (nd->flags & LOOKUP_RCU) {
1559 if (unlikely(unlazy_walk(nd, path->dentry))) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001560 err = -ECHILD;
1561 goto out_err;
Al Viro19660af2011-03-25 10:32:48 -04001562 }
1563 }
Al Viroce57dfc2011-03-13 19:58:58 -04001564 BUG_ON(inode != path->dentry->d_inode);
1565 return 1;
1566 }
1567 path_to_nameidata(path, nd);
1568 nd->inode = inode;
1569 return 0;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001570
1571out_path_put:
1572 path_to_nameidata(path, nd);
1573out_err:
1574 terminate_walk(nd);
1575 return err;
Al Viroce57dfc2011-03-13 19:58:58 -04001576}
1577
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578/*
Al Virob3563792011-03-14 21:54:55 -04001579 * This limits recursive symlink follows to 8, while
1580 * limiting consecutive symlinks to 40.
1581 *
1582 * Without that kind of total limit, nasty chains of consecutive
1583 * symlinks can cause almost arbitrarily long lookups.
1584 */
1585static inline int nested_symlink(struct path *path, struct nameidata *nd)
1586{
1587 int res;
1588
Al Virob3563792011-03-14 21:54:55 -04001589 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1590 path_put_conditional(path, nd);
1591 path_put(&nd->path);
1592 return -ELOOP;
1593 }
Erez Zadok1a4022f2011-05-21 01:19:59 -04001594 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
Al Virob3563792011-03-14 21:54:55 -04001595
1596 nd->depth++;
1597 current->link_count++;
1598
1599 do {
1600 struct path link = *path;
1601 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04001602
1603 res = follow_link(&link, nd, &cookie);
Al Viro6d7b5aa2012-06-10 04:15:17 -04001604 if (res)
1605 break;
1606 res = walk_component(nd, path, &nd->last,
1607 nd->last_type, LOOKUP_FOLLOW);
Al Viro574197e2011-03-14 22:20:34 -04001608 put_link(nd, &link, cookie);
Al Virob3563792011-03-14 21:54:55 -04001609 } while (res > 0);
1610
1611 current->link_count--;
1612 nd->depth--;
1613 return res;
1614}
1615
1616/*
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001617 * We really don't want to look at inode->i_op->lookup
1618 * when we don't have to. So we keep a cache bit in
1619 * the inode ->i_opflags field that says "yes, we can
1620 * do lookup on this inode".
1621 */
1622static inline int can_lookup(struct inode *inode)
1623{
1624 if (likely(inode->i_opflags & IOP_LOOKUP))
1625 return 1;
1626 if (likely(!inode->i_op->lookup))
1627 return 0;
1628
1629 /* We do this once for the lifetime of the inode */
1630 spin_lock(&inode->i_lock);
1631 inode->i_opflags |= IOP_LOOKUP;
1632 spin_unlock(&inode->i_lock);
1633 return 1;
1634}
1635
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001636/*
1637 * We can do the critical dentry name comparison and hashing
1638 * operations one word at a time, but we are limited to:
1639 *
1640 * - Architectures with fast unaligned word accesses. We could
1641 * do a "get_unaligned()" if this helps and is sufficiently
1642 * fast.
1643 *
1644 * - Little-endian machines (so that we can generate the mask
1645 * of low bytes efficiently). Again, we *could* do a byte
1646 * swapping load on big-endian architectures if that is not
1647 * expensive enough to make the optimization worthless.
1648 *
1649 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1650 * do not trap on the (extremely unlikely) case of a page
1651 * crossing operation.
1652 *
1653 * - Furthermore, we need an efficient 64-bit compile for the
1654 * 64-bit case in order to generate the "number of bytes in
1655 * the final mask". Again, that could be replaced with a
1656 * efficient population count instruction or similar.
1657 */
1658#ifdef CONFIG_DCACHE_WORD_ACCESS
1659
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001660#include <asm/word-at-a-time.h>
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001661
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001662#ifdef CONFIG_64BIT
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001663
1664static inline unsigned int fold_hash(unsigned long hash)
1665{
1666 hash += hash >> (8*sizeof(int));
1667 return hash;
1668}
1669
1670#else /* 32-bit case */
1671
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001672#define fold_hash(x) (x)
1673
1674#endif
1675
1676unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1677{
1678 unsigned long a, mask;
1679 unsigned long hash = 0;
1680
1681 for (;;) {
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001682 a = load_unaligned_zeropad(name);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001683 if (len < sizeof(unsigned long))
1684 break;
1685 hash += a;
Al Virof132c5b2012-03-22 21:59:52 +00001686 hash *= 9;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001687 name += sizeof(unsigned long);
1688 len -= sizeof(unsigned long);
1689 if (!len)
1690 goto done;
1691 }
1692 mask = ~(~0ul << len*8);
1693 hash += mask & a;
1694done:
1695 return fold_hash(hash);
1696}
1697EXPORT_SYMBOL(full_name_hash);
1698
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001699/*
1700 * Calculate the length and hash of the path component, and
1701 * return the length of the component;
1702 */
1703static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1704{
Linus Torvalds36126f82012-05-26 10:43:17 -07001705 unsigned long a, b, adata, bdata, mask, hash, len;
1706 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001707
1708 hash = a = 0;
1709 len = -sizeof(unsigned long);
1710 do {
1711 hash = (hash + a) * 9;
1712 len += sizeof(unsigned long);
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001713 a = load_unaligned_zeropad(name+len);
Linus Torvalds36126f82012-05-26 10:43:17 -07001714 b = a ^ REPEAT_BYTE('/');
1715 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001716
Linus Torvalds36126f82012-05-26 10:43:17 -07001717 adata = prep_zero_mask(a, adata, &constants);
1718 bdata = prep_zero_mask(b, bdata, &constants);
1719
1720 mask = create_zero_mask(adata | bdata);
1721
1722 hash += a & zero_bytemask(mask);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001723 *hashp = fold_hash(hash);
1724
Linus Torvalds36126f82012-05-26 10:43:17 -07001725 return len + find_zero(mask);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001726}
1727
1728#else
1729
Linus Torvalds0145acc2012-03-02 14:32:59 -08001730unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1731{
1732 unsigned long hash = init_name_hash();
1733 while (len--)
1734 hash = partial_name_hash(*name++, hash);
1735 return end_name_hash(hash);
1736}
Linus Torvaldsae942ae2012-03-02 19:40:57 -08001737EXPORT_SYMBOL(full_name_hash);
Linus Torvalds0145acc2012-03-02 14:32:59 -08001738
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001739/*
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001740 * We know there's a real path component here of at least
1741 * one character.
1742 */
1743static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1744{
1745 unsigned long hash = init_name_hash();
1746 unsigned long len = 0, c;
1747
1748 c = (unsigned char)*name;
1749 do {
1750 len++;
1751 hash = partial_name_hash(c, hash);
1752 c = (unsigned char)name[len];
1753 } while (c && c != '/');
1754 *hashp = end_name_hash(hash);
1755 return len;
1756}
1757
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001758#endif
1759
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001760/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01001762 * This is the basic name resolution function, turning a pathname into
1763 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01001765 * Returns 0 and nd will have valid dentry and mnt on success.
1766 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767 */
Al Viro6de88d72009-08-09 01:41:57 +04001768static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769{
1770 struct path next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772
1773 while (*name=='/')
1774 name++;
1775 if (!*name)
Al Viro086e1832011-02-22 20:56:27 -05001776 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 /* At this point we know we have a real path component. */
1779 for(;;) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 struct qstr this;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001781 long len;
Al Virofe479a52011-02-22 15:10:03 -05001782 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783
Al Viro52094c82011-02-21 21:34:47 -05001784 err = may_lookup(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 if (err)
1786 break;
1787
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001788 len = hash_name(name, &this.hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 this.name = name;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001790 this.len = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791
Al Virofe479a52011-02-22 15:10:03 -05001792 type = LAST_NORM;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001793 if (name[0] == '.') switch (len) {
Al Virofe479a52011-02-22 15:10:03 -05001794 case 2:
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001795 if (name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05001796 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05001797 nd->flags |= LOOKUP_JUMPED;
1798 }
Al Virofe479a52011-02-22 15:10:03 -05001799 break;
1800 case 1:
1801 type = LAST_DOT;
1802 }
Al Viro5a202bc2011-03-08 14:17:44 -05001803 if (likely(type == LAST_NORM)) {
1804 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05001805 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05001806 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1807 err = parent->d_op->d_hash(parent, nd->inode,
1808 &this);
1809 if (err < 0)
1810 break;
1811 }
1812 }
Al Virofe479a52011-02-22 15:10:03 -05001813
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001814 if (!name[len])
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 goto last_component;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001816 /*
1817 * If it wasn't NUL, we know it was '/'. Skip that
1818 * slash, and continue until no more slashes.
1819 */
1820 do {
1821 len++;
1822 } while (unlikely(name[len] == '/'));
1823 if (!name[len])
Al Virob3563792011-03-14 21:54:55 -04001824 goto last_component;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001825 name += len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826
Al Viroce57dfc2011-03-13 19:58:58 -04001827 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1828 if (err < 0)
1829 return err;
Al Virofe479a52011-02-22 15:10:03 -05001830
Al Viroce57dfc2011-03-13 19:58:58 -04001831 if (err) {
Al Virob3563792011-03-14 21:54:55 -04001832 err = nested_symlink(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833 if (err)
Al Viroa7472ba2011-03-04 14:39:30 -05001834 return err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001835 }
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001836 if (can_lookup(nd->inode))
1837 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 err = -ENOTDIR;
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001839 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840 /* here ends the main loop */
1841
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842last_component:
Al Virob3563792011-03-14 21:54:55 -04001843 nd->last = this;
1844 nd->last_type = type;
Al Viro086e1832011-02-22 20:56:27 -05001845 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846 }
Al Viro951361f2011-03-04 14:44:37 -05001847 terminate_walk(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848 return err;
1849}
1850
Al Viro70e9b352011-03-05 21:12:22 -05001851static int path_init(int dfd, const char *name, unsigned int flags,
1852 struct nameidata *nd, struct file **fp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001854 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855
1856 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro16c2cd72011-02-22 15:50:10 -05001857 nd->flags = flags | LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 nd->depth = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05001859 if (flags & LOOKUP_ROOT) {
1860 struct inode *inode = nd->root.dentry->d_inode;
Al Viro73d049a2011-03-11 12:08:24 -05001861 if (*name) {
1862 if (!inode->i_op->lookup)
1863 return -ENOTDIR;
1864 retval = inode_permission(inode, MAY_EXEC);
1865 if (retval)
1866 return retval;
1867 }
Al Viro5b6ca022011-03-09 23:04:47 -05001868 nd->path = nd->root;
1869 nd->inode = inode;
1870 if (flags & LOOKUP_RCU) {
Al Viro32a79912012-07-18 20:43:19 +04001871 lock_rcu_walk();
Al Viro5b6ca022011-03-09 23:04:47 -05001872 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1873 } else {
1874 path_get(&nd->path);
1875 }
1876 return 0;
1877 }
1878
Al Viro2a737872009-04-07 11:49:53 -04001879 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 if (*name=='/') {
Al Viroe41f7d42011-02-22 14:02:58 -05001882 if (flags & LOOKUP_RCU) {
Al Viro32a79912012-07-18 20:43:19 +04001883 lock_rcu_walk();
Al Viroe41f7d42011-02-22 14:02:58 -05001884 set_root_rcu(nd);
1885 } else {
1886 set_root(nd);
1887 path_get(&nd->root);
1888 }
Al Viro2a737872009-04-07 11:49:53 -04001889 nd->path = nd->root;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001890 } else if (dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05001891 if (flags & LOOKUP_RCU) {
1892 struct fs_struct *fs = current->fs;
1893 unsigned seq;
1894
Al Viro32a79912012-07-18 20:43:19 +04001895 lock_rcu_walk();
Al Viroe41f7d42011-02-22 14:02:58 -05001896
1897 do {
1898 seq = read_seqcount_begin(&fs->seq);
1899 nd->path = fs->pwd;
1900 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1901 } while (read_seqcount_retry(&fs->seq, seq));
1902 } else {
1903 get_fs_pwd(current->fs, &nd->path);
1904 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001905 } else {
Al Viro2903ff02012-08-28 12:52:22 -04001906 struct fd f = fdget_raw(dfd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001907 struct dentry *dentry;
1908
Al Viro2903ff02012-08-28 12:52:22 -04001909 if (!f.file)
1910 return -EBADF;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001911
Al Viro2903ff02012-08-28 12:52:22 -04001912 dentry = f.file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001913
Al Virof52e0c12011-03-14 18:56:51 -04001914 if (*name) {
Al Viro2903ff02012-08-28 12:52:22 -04001915 if (!S_ISDIR(dentry->d_inode->i_mode)) {
1916 fdput(f);
1917 return -ENOTDIR;
1918 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001919
Al Viro4ad5abb2011-06-20 19:57:03 -04001920 retval = inode_permission(dentry->d_inode, MAY_EXEC);
Al Viro2903ff02012-08-28 12:52:22 -04001921 if (retval) {
1922 fdput(f);
1923 return retval;
1924 }
Al Virof52e0c12011-03-14 18:56:51 -04001925 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001926
Al Viro2903ff02012-08-28 12:52:22 -04001927 nd->path = f.file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05001928 if (flags & LOOKUP_RCU) {
Al Viro2903ff02012-08-28 12:52:22 -04001929 if (f.need_put)
1930 *fp = f.file;
Al Viroe41f7d42011-02-22 14:02:58 -05001931 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro32a79912012-07-18 20:43:19 +04001932 lock_rcu_walk();
Al Viroe41f7d42011-02-22 14:02:58 -05001933 } else {
Al Viro2903ff02012-08-28 12:52:22 -04001934 path_get(&nd->path);
1935 fdput(f);
Al Viroe41f7d42011-02-22 14:02:58 -05001936 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 }
Al Viroe41f7d42011-02-22 14:02:58 -05001938
Nick Piggin31e6b012011-01-07 17:49:52 +11001939 nd->inode = nd->path.dentry->d_inode;
Al Viro9b4a9b12009-04-07 11:44:16 -04001940 return 0;
Al Viro9b4a9b12009-04-07 11:44:16 -04001941}
1942
Al Virobd92d7f2011-03-14 19:54:59 -04001943static inline int lookup_last(struct nameidata *nd, struct path *path)
1944{
1945 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1946 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1947
1948 nd->flags &= ~LOOKUP_PARENT;
1949 return walk_component(nd, path, &nd->last, nd->last_type,
1950 nd->flags & LOOKUP_FOLLOW);
1951}
1952
Al Viro9b4a9b12009-04-07 11:44:16 -04001953/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroee0827c2011-02-21 23:38:09 -05001954static int path_lookupat(int dfd, const char *name,
Al Viro9b4a9b12009-04-07 11:44:16 -04001955 unsigned int flags, struct nameidata *nd)
1956{
Al Viro70e9b352011-03-05 21:12:22 -05001957 struct file *base = NULL;
Al Virobd92d7f2011-03-14 19:54:59 -04001958 struct path path;
1959 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001960
1961 /*
1962 * Path walking is largely split up into 2 different synchronisation
1963 * schemes, rcu-walk and ref-walk (explained in
1964 * Documentation/filesystems/path-lookup.txt). These share much of the
1965 * path walk code, but some things particularly setup, cleanup, and
1966 * following mounts are sufficiently divergent that functions are
1967 * duplicated. Typically there is a function foo(), and its RCU
1968 * analogue, foo_rcu().
1969 *
1970 * -ECHILD is the error number of choice (just to avoid clashes) that
1971 * is returned if some aspect of an rcu-walk fails. Such an error must
1972 * be handled by restarting a traditional ref-walk (which will always
1973 * be able to complete).
1974 */
Al Virobd92d7f2011-03-14 19:54:59 -04001975 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
Al Viroee0827c2011-02-21 23:38:09 -05001976
Al Virobd92d7f2011-03-14 19:54:59 -04001977 if (unlikely(err))
1978 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001979
1980 current->total_link_count = 0;
Al Virobd92d7f2011-03-14 19:54:59 -04001981 err = link_path_walk(name, nd);
1982
1983 if (!err && !(flags & LOOKUP_PARENT)) {
Al Virobd92d7f2011-03-14 19:54:59 -04001984 err = lookup_last(nd, &path);
1985 while (err > 0) {
1986 void *cookie;
1987 struct path link = path;
Kees Cook800179c2012-07-25 17:29:07 -07001988 err = may_follow_link(&link, nd);
1989 if (unlikely(err))
1990 break;
Al Virobd92d7f2011-03-14 19:54:59 -04001991 nd->flags |= LOOKUP_PARENT;
Al Viro574197e2011-03-14 22:20:34 -04001992 err = follow_link(&link, nd, &cookie);
Al Viro6d7b5aa2012-06-10 04:15:17 -04001993 if (err)
1994 break;
1995 err = lookup_last(nd, &path);
Al Viro574197e2011-03-14 22:20:34 -04001996 put_link(nd, &link, cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04001997 }
1998 }
Al Viroee0827c2011-02-21 23:38:09 -05001999
Al Viro9f1fafe2011-03-25 11:00:12 -04002000 if (!err)
2001 err = complete_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04002002
2003 if (!err && nd->flags & LOOKUP_DIRECTORY) {
2004 if (!nd->inode->i_op->lookup) {
2005 path_put(&nd->path);
Al Virobd23a532011-03-23 09:56:30 -04002006 err = -ENOTDIR;
Al Virobd92d7f2011-03-14 19:54:59 -04002007 }
2008 }
Al Viro16c2cd72011-02-22 15:50:10 -05002009
Al Viro70e9b352011-03-05 21:12:22 -05002010 if (base)
2011 fput(base);
Al Viroee0827c2011-02-21 23:38:09 -05002012
Al Viro5b6ca022011-03-09 23:04:47 -05002013 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
Al Viro2a737872009-04-07 11:49:53 -04002014 path_put(&nd->root);
2015 nd->root.mnt = NULL;
2016 }
Al Virobd92d7f2011-03-14 19:54:59 -04002017 return err;
Al Viroee0827c2011-02-21 23:38:09 -05002018}
Nick Piggin31e6b012011-01-07 17:49:52 +11002019
Jeff Layton873f1ee2012-10-10 15:25:29 -04002020static int filename_lookup(int dfd, struct filename *name,
2021 unsigned int flags, struct nameidata *nd)
2022{
2023 int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd);
2024 if (unlikely(retval == -ECHILD))
2025 retval = path_lookupat(dfd, name->name, flags, nd);
2026 if (unlikely(retval == -ESTALE))
2027 retval = path_lookupat(dfd, name->name,
2028 flags | LOOKUP_REVAL, nd);
2029
2030 if (likely(!retval))
Jeff Laytonadb5c242012-10-10 16:43:13 -04002031 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002032 return retval;
2033}
2034
Al Viroee0827c2011-02-21 23:38:09 -05002035static int do_path_lookup(int dfd, const char *name,
2036 unsigned int flags, struct nameidata *nd)
2037{
Jeff Layton873f1ee2012-10-10 15:25:29 -04002038 struct filename filename = { .name = name };
Nick Piggin31e6b012011-01-07 17:49:52 +11002039
Jeff Layton873f1ee2012-10-10 15:25:29 -04002040 return filename_lookup(dfd, &filename, flags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041}
2042
Al Viro79714f72012-06-15 03:01:42 +04002043/* does lookup, returns the object with parent locked */
2044struct dentry *kern_path_locked(const char *name, struct path *path)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002045{
Al Viro79714f72012-06-15 03:01:42 +04002046 struct nameidata nd;
2047 struct dentry *d;
2048 int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd);
2049 if (err)
2050 return ERR_PTR(err);
2051 if (nd.last_type != LAST_NORM) {
2052 path_put(&nd.path);
2053 return ERR_PTR(-EINVAL);
2054 }
2055 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Al Viro1e0ea002012-07-22 23:46:21 +04002056 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
Al Viro79714f72012-06-15 03:01:42 +04002057 if (IS_ERR(d)) {
2058 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2059 path_put(&nd.path);
2060 return d;
2061 }
2062 *path = nd.path;
2063 return d;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002064}
2065
Al Virod1811462008-08-02 00:49:18 -04002066int kern_path(const char *name, unsigned int flags, struct path *path)
2067{
2068 struct nameidata nd;
2069 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
2070 if (!res)
2071 *path = nd.path;
2072 return res;
2073}
2074
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002075/**
2076 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2077 * @dentry: pointer to dentry of the base directory
2078 * @mnt: pointer to vfs mount of the base directory
2079 * @name: pointer to file name
2080 * @flags: lookup flags
Al Viroe0a01242011-06-27 17:00:37 -04002081 * @path: pointer to struct path to fill
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002082 */
2083int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2084 const char *name, unsigned int flags,
Al Viroe0a01242011-06-27 17:00:37 -04002085 struct path *path)
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002086{
Al Viroe0a01242011-06-27 17:00:37 -04002087 struct nameidata nd;
2088 int err;
2089 nd.root.dentry = dentry;
2090 nd.root.mnt = mnt;
2091 BUG_ON(flags & LOOKUP_PARENT);
Al Viro5b6ca022011-03-09 23:04:47 -05002092 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
Al Viroe0a01242011-06-27 17:00:37 -04002093 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
2094 if (!err)
2095 *path = nd.path;
2096 return err;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002097}
2098
James Morris057f6c02007-04-26 00:12:05 -07002099/*
2100 * Restricted form of lookup. Doesn't follow links, single-component only,
2101 * needs parent already locked. Doesn't follow mounts.
2102 * SMP-safe.
2103 */
Adrian Bunka244e162006-03-31 02:32:11 -08002104static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105{
Al Viro72bd8662012-06-10 17:17:17 -04002106 return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107}
2108
Christoph Hellwigeead1912007-10-16 23:25:38 -07002109/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07002110 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07002111 * @name: pathname component to lookup
2112 * @base: base directory to lookup from
2113 * @len: maximum length @len should be interpreted to
2114 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07002115 * Note that this routine is purely a helper for filesystem usage and should
2116 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07002117 * nameidata argument is passed to the filesystem methods and a filesystem
2118 * using this helper needs to be prepared for that.
2119 */
James Morris057f6c02007-04-26 00:12:05 -07002120struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2121{
James Morris057f6c02007-04-26 00:12:05 -07002122 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05002123 unsigned int c;
Miklos Szeredicda309d2012-03-26 12:54:21 +02002124 int err;
James Morris057f6c02007-04-26 00:12:05 -07002125
David Woodhouse2f9092e2009-04-20 23:18:37 +01002126 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2127
Al Viro6a96ba52011-03-07 23:49:20 -05002128 this.name = name;
2129 this.len = len;
Linus Torvalds0145acc2012-03-02 14:32:59 -08002130 this.hash = full_name_hash(name, len);
Al Viro6a96ba52011-03-07 23:49:20 -05002131 if (!len)
2132 return ERR_PTR(-EACCES);
2133
Al Viro21d8a152012-11-29 22:17:21 -05002134 if (unlikely(name[0] == '.')) {
2135 if (len < 2 || (len == 2 && name[1] == '.'))
2136 return ERR_PTR(-EACCES);
2137 }
2138
Al Viro6a96ba52011-03-07 23:49:20 -05002139 while (len--) {
2140 c = *(const unsigned char *)name++;
2141 if (c == '/' || c == '\0')
2142 return ERR_PTR(-EACCES);
Al Viro6a96ba52011-03-07 23:49:20 -05002143 }
Al Viro5a202bc2011-03-08 14:17:44 -05002144 /*
2145 * See if the low-level filesystem might want
2146 * to use its own hash..
2147 */
2148 if (base->d_flags & DCACHE_OP_HASH) {
2149 int err = base->d_op->d_hash(base, base->d_inode, &this);
2150 if (err < 0)
2151 return ERR_PTR(err);
2152 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07002153
Miklos Szeredicda309d2012-03-26 12:54:21 +02002154 err = inode_permission(base->d_inode, MAY_EXEC);
2155 if (err)
2156 return ERR_PTR(err);
2157
Al Viro72bd8662012-06-10 17:17:17 -04002158 return __lookup_hash(&this, base, 0);
James Morris057f6c02007-04-26 00:12:05 -07002159}
2160
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002161int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2162 struct path *path, int *empty)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163{
Al Viro2d8f3032008-07-22 09:59:21 -04002164 struct nameidata nd;
Jeff Layton91a27b22012-10-10 15:25:28 -04002165 struct filename *tmp = getname_flags(name, flags, empty);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04002168
2169 BUG_ON(flags & LOOKUP_PARENT);
2170
Jeff Layton873f1ee2012-10-10 15:25:29 -04002171 err = filename_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04002173 if (!err)
2174 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 }
2176 return err;
2177}
2178
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002179int user_path_at(int dfd, const char __user *name, unsigned flags,
2180 struct path *path)
2181{
Linus Torvaldsf7493e52012-03-22 16:10:40 -07002182 return user_path_at_empty(dfd, name, flags, path, NULL);
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002183}
2184
Jeff Layton873f1ee2012-10-10 15:25:29 -04002185/*
2186 * NB: most callers don't do anything directly with the reference to the
2187 * to struct filename, but the nd->last pointer points into the name string
2188 * allocated by getname. So we must hold the reference to it until all
2189 * path-walking is complete.
2190 */
Jeff Layton91a27b22012-10-10 15:25:28 -04002191static struct filename *
2192user_path_parent(int dfd, const char __user *path, struct nameidata *nd)
Al Viro2ad94ae2008-07-21 09:32:51 -04002193{
Jeff Layton91a27b22012-10-10 15:25:28 -04002194 struct filename *s = getname(path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002195 int error;
2196
2197 if (IS_ERR(s))
Jeff Layton91a27b22012-10-10 15:25:28 -04002198 return s;
Al Viro2ad94ae2008-07-21 09:32:51 -04002199
Jeff Layton873f1ee2012-10-10 15:25:29 -04002200 error = filename_lookup(dfd, s, LOOKUP_PARENT, nd);
Jeff Layton91a27b22012-10-10 15:25:28 -04002201 if (error) {
Al Viro2ad94ae2008-07-21 09:32:51 -04002202 putname(s);
Jeff Layton91a27b22012-10-10 15:25:28 -04002203 return ERR_PTR(error);
2204 }
Al Viro2ad94ae2008-07-21 09:32:51 -04002205
Jeff Layton91a27b22012-10-10 15:25:28 -04002206 return s;
Al Viro2ad94ae2008-07-21 09:32:51 -04002207}
2208
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209/*
2210 * It's inline, so penalty for filesystems that don't use sticky bit is
2211 * minimal.
2212 */
2213static inline int check_sticky(struct inode *dir, struct inode *inode)
2214{
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002215 kuid_t fsuid = current_fsuid();
David Howellsda9592e2008-11-14 10:39:05 +11002216
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217 if (!(dir->i_mode & S_ISVTX))
2218 return 0;
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002219 if (uid_eq(inode->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220 return 0;
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002221 if (uid_eq(dir->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 return 0;
Eric W. Biederman1a48e2a2011-11-14 16:24:06 -08002223 return !inode_capable(inode, CAP_FOWNER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224}
2225
2226/*
2227 * Check whether we can remove a link victim from directory dir, check
2228 * whether the type of victim is right.
2229 * 1. We can't do it if dir is read-only (done in permission())
2230 * 2. We should have write and exec permissions on dir
2231 * 3. We can't remove anything from append-only dir
2232 * 4. We can't do anything with immutable dir (done in permission())
2233 * 5. If the sticky bit on dir is set we should either
2234 * a. be owner of dir, or
2235 * b. be owner of victim, or
2236 * c. have CAP_FOWNER capability
2237 * 6. If the victim is append-only or immutable we can't do antyhing with
2238 * links pointing to it.
2239 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2240 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2241 * 9. We can't remove a root or mountpoint.
2242 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2243 * nfs_async_unlink().
2244 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08002245static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246{
2247 int error;
2248
2249 if (!victim->d_inode)
2250 return -ENOENT;
2251
2252 BUG_ON(victim->d_parent->d_inode != dir);
Jeff Layton4fa6b5e2012-10-10 15:25:25 -04002253 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254
Al Virof419a2e2008-07-22 00:07:17 -04002255 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256 if (error)
2257 return error;
2258 if (IS_APPEND(dir))
2259 return -EPERM;
2260 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
Hugh Dickinsf9454542008-11-19 15:36:38 -08002261 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262 return -EPERM;
2263 if (isdir) {
2264 if (!S_ISDIR(victim->d_inode->i_mode))
2265 return -ENOTDIR;
2266 if (IS_ROOT(victim))
2267 return -EBUSY;
2268 } else if (S_ISDIR(victim->d_inode->i_mode))
2269 return -EISDIR;
2270 if (IS_DEADDIR(dir))
2271 return -ENOENT;
2272 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2273 return -EBUSY;
2274 return 0;
2275}
2276
2277/* Check whether we can create an object with dentry child in directory
2278 * dir.
2279 * 1. We can't do it if child already exists (open has special treatment for
2280 * this case, but since we are inlined it's OK)
2281 * 2. We can't do it if dir is read-only (done in permission())
2282 * 3. We should have write and exec permissions on dir
2283 * 4. We can't do it if dir is immutable (done in permission())
2284 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02002285static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286{
2287 if (child->d_inode)
2288 return -EEXIST;
2289 if (IS_DEADDIR(dir))
2290 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04002291 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292}
2293
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294/*
2295 * p1 and p2 should be directories on the same fs.
2296 */
2297struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2298{
2299 struct dentry *p;
2300
2301 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07002302 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303 return NULL;
2304 }
2305
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002306 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002308 p = d_ancestor(p2, p1);
2309 if (p) {
2310 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2311 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2312 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313 }
2314
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002315 p = d_ancestor(p1, p2);
2316 if (p) {
2317 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2318 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2319 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 }
2321
Ingo Molnarf2eace22006-07-03 00:25:05 -07002322 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2323 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324 return NULL;
2325}
2326
2327void unlock_rename(struct dentry *p1, struct dentry *p2)
2328{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002329 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002331 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002332 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 }
2334}
2335
Al Viro4acdaf22011-07-26 01:42:34 -04002336int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viro312b63f2012-06-10 18:09:36 -04002337 bool want_excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002339 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340 if (error)
2341 return error;
2342
Al Viroacfa4382008-12-04 10:06:33 -05002343 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344 return -EACCES; /* shouldn't it be ENOSYS? */
2345 mode &= S_IALLUGO;
2346 mode |= S_IFREG;
2347 error = security_inode_create(dir, dentry, mode);
2348 if (error)
2349 return error;
Al Viro312b63f2012-06-10 18:09:36 -04002350 error = dir->i_op->create(dir, dentry, mode, want_excl);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002351 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002352 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353 return error;
2354}
2355
Al Viro73d049a2011-03-11 12:08:24 -05002356static int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002358 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359 struct inode *inode = dentry->d_inode;
2360 int error;
2361
Al Virobcda7652011-03-13 16:42:14 -04002362 /* O_PATH? */
2363 if (!acc_mode)
2364 return 0;
2365
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366 if (!inode)
2367 return -ENOENT;
2368
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002369 switch (inode->i_mode & S_IFMT) {
2370 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002372 case S_IFDIR:
2373 if (acc_mode & MAY_WRITE)
2374 return -EISDIR;
2375 break;
2376 case S_IFBLK:
2377 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002378 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002380 /*FALLTHRU*/
2381 case S_IFIFO:
2382 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002384 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002385 }
Dave Hansenb41572e2007-10-16 23:31:14 -07002386
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002387 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07002388 if (error)
2389 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05002390
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391 /*
2392 * An append-only file must be opened in append mode for writing.
2393 */
2394 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05002395 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05002396 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002397 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05002398 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399 }
2400
2401 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002402 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05002403 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404
J. Bruce Fieldsf3c7691e2011-09-21 10:58:13 -04002405 return 0;
Al Viro7715b522009-12-16 03:54:00 -05002406}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407
Jeff Laytone1181ee2010-12-07 16:19:50 -05002408static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002409{
Jeff Laytone1181ee2010-12-07 16:19:50 -05002410 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002411 struct inode *inode = path->dentry->d_inode;
2412 int error = get_write_access(inode);
2413 if (error)
2414 return error;
2415 /*
2416 * Refuse to truncate files with mandatory locks held on them.
2417 */
2418 error = locks_verify_locked(inode);
2419 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002420 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002421 if (!error) {
2422 error = do_truncate(path->dentry, 0,
2423 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002424 filp);
Al Viro7715b522009-12-16 03:54:00 -05002425 }
2426 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002427 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428}
2429
Dave Hansend57999e2008-02-15 14:37:27 -08002430static inline int open_to_namei_flags(int flag)
2431{
Al Viro8a5e9292011-06-25 19:15:54 -04002432 if ((flag & O_ACCMODE) == 3)
2433 flag--;
Dave Hansend57999e2008-02-15 14:37:27 -08002434 return flag;
2435}
2436
Miklos Szeredid18e9002012-06-05 15:10:17 +02002437static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2438{
2439 int error = security_path_mknod(dir, dentry, mode, 0);
2440 if (error)
2441 return error;
2442
2443 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2444 if (error)
2445 return error;
2446
2447 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2448}
2449
David Howells1acf0af2012-06-14 16:13:46 +01002450/*
2451 * Attempt to atomically look up, create and open a file from a negative
2452 * dentry.
2453 *
2454 * Returns 0 if successful. The file will have been created and attached to
2455 * @file by the filesystem calling finish_open().
2456 *
2457 * Returns 1 if the file was looked up only or didn't need creating. The
2458 * caller will need to perform the open themselves. @path will have been
2459 * updated to point to the new dentry. This may be negative.
2460 *
2461 * Returns an error code otherwise.
2462 */
Al Viro2675a4e2012-06-22 12:41:10 +04002463static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2464 struct path *path, struct file *file,
2465 const struct open_flags *op,
Al Viro64894cf2012-07-31 00:53:35 +04002466 bool got_write, bool need_lookup,
Al Viro2675a4e2012-06-22 12:41:10 +04002467 int *opened)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002468{
2469 struct inode *dir = nd->path.dentry->d_inode;
2470 unsigned open_flag = open_to_namei_flags(op->open_flag);
2471 umode_t mode;
2472 int error;
2473 int acc_mode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002474 int create_error = 0;
2475 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2476
2477 BUG_ON(dentry->d_inode);
2478
2479 /* Don't create child dentry for a dead directory. */
2480 if (unlikely(IS_DEADDIR(dir))) {
Al Viro2675a4e2012-06-22 12:41:10 +04002481 error = -ENOENT;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002482 goto out;
2483 }
2484
Miklos Szeredi62b259d2012-08-15 13:01:24 +02002485 mode = op->mode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002486 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2487 mode &= ~current_umask();
2488
Al Virof8310c52012-07-30 11:50:30 +04002489 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT)) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002490 open_flag &= ~O_TRUNC;
Al Viro47237682012-06-10 05:01:45 -04002491 *opened |= FILE_CREATED;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002492 }
2493
2494 /*
2495 * Checking write permission is tricky, bacuse we don't know if we are
2496 * going to actually need it: O_CREAT opens should work as long as the
2497 * file exists. But checking existence breaks atomicity. The trick is
2498 * to check access and if not granted clear O_CREAT from the flags.
2499 *
2500 * Another problem is returing the "right" error value (e.g. for an
2501 * O_EXCL open we want to return EEXIST not EROFS).
2502 */
Al Viro64894cf2012-07-31 00:53:35 +04002503 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2504 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2505 if (!(open_flag & O_CREAT)) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002506 /*
2507 * No O_CREATE -> atomicity not a requirement -> fall
2508 * back to lookup + open
2509 */
2510 goto no_open;
2511 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2512 /* Fall back and fail with the right error */
Al Viro64894cf2012-07-31 00:53:35 +04002513 create_error = -EROFS;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002514 goto no_open;
2515 } else {
2516 /* No side effects, safe to clear O_CREAT */
Al Viro64894cf2012-07-31 00:53:35 +04002517 create_error = -EROFS;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002518 open_flag &= ~O_CREAT;
2519 }
2520 }
2521
2522 if (open_flag & O_CREAT) {
Miklos Szeredi38227f72012-08-15 13:01:24 +02002523 error = may_o_create(&nd->path, dentry, mode);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002524 if (error) {
2525 create_error = error;
2526 if (open_flag & O_EXCL)
2527 goto no_open;
2528 open_flag &= ~O_CREAT;
2529 }
2530 }
2531
2532 if (nd->flags & LOOKUP_DIRECTORY)
2533 open_flag |= O_DIRECTORY;
2534
Al Viro30d90492012-06-22 12:40:19 +04002535 file->f_path.dentry = DENTRY_NOT_SET;
2536 file->f_path.mnt = nd->path.mnt;
2537 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
Al Viro47237682012-06-10 05:01:45 -04002538 opened);
Al Virod9585272012-06-22 12:39:14 +04002539 if (error < 0) {
Al Virod9585272012-06-22 12:39:14 +04002540 if (create_error && error == -ENOENT)
2541 error = create_error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002542 goto out;
2543 }
2544
2545 acc_mode = op->acc_mode;
Al Viro47237682012-06-10 05:01:45 -04002546 if (*opened & FILE_CREATED) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002547 fsnotify_create(dir, dentry);
2548 acc_mode = MAY_OPEN;
2549 }
2550
Al Virod9585272012-06-22 12:39:14 +04002551 if (error) { /* returned 1, that is */
Al Viro30d90492012-06-22 12:40:19 +04002552 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
Al Viro2675a4e2012-06-22 12:41:10 +04002553 error = -EIO;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002554 goto out;
2555 }
Al Viro30d90492012-06-22 12:40:19 +04002556 if (file->f_path.dentry) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002557 dput(dentry);
Al Viro30d90492012-06-22 12:40:19 +04002558 dentry = file->f_path.dentry;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002559 }
Sage Weil62b2ce92012-08-15 13:30:12 -07002560 if (create_error && dentry->d_inode == NULL) {
2561 error = create_error;
2562 goto out;
2563 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02002564 goto looked_up;
2565 }
2566
2567 /*
2568 * We didn't have the inode before the open, so check open permission
2569 * here.
2570 */
Al Viro2675a4e2012-06-22 12:41:10 +04002571 error = may_open(&file->f_path, acc_mode, open_flag);
2572 if (error)
2573 fput(file);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002574
2575out:
2576 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04002577 return error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002578
Miklos Szeredid18e9002012-06-05 15:10:17 +02002579no_open:
2580 if (need_lookup) {
Al Viro72bd8662012-06-10 17:17:17 -04002581 dentry = lookup_real(dir, dentry, nd->flags);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002582 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002583 return PTR_ERR(dentry);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002584
2585 if (create_error) {
2586 int open_flag = op->open_flag;
2587
Al Viro2675a4e2012-06-22 12:41:10 +04002588 error = create_error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002589 if ((open_flag & O_EXCL)) {
2590 if (!dentry->d_inode)
2591 goto out;
2592 } else if (!dentry->d_inode) {
2593 goto out;
2594 } else if ((open_flag & O_TRUNC) &&
2595 S_ISREG(dentry->d_inode->i_mode)) {
2596 goto out;
2597 }
2598 /* will fail later, go on to get the right error */
2599 }
2600 }
2601looked_up:
2602 path->dentry = dentry;
2603 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04002604 return 1;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002605}
2606
Nick Piggin31e6b012011-01-07 17:49:52 +11002607/*
David Howells1acf0af2012-06-14 16:13:46 +01002608 * Look up and maybe create and open the last component.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002609 *
2610 * Must be called with i_mutex held on parent.
2611 *
David Howells1acf0af2012-06-14 16:13:46 +01002612 * Returns 0 if the file was successfully atomically created (if necessary) and
2613 * opened. In this case the file will be returned attached to @file.
2614 *
2615 * Returns 1 if the file was not completely opened at this time, though lookups
2616 * and creations will have been performed and the dentry returned in @path will
2617 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2618 * specified then a negative dentry may be returned.
2619 *
2620 * An error code is returned otherwise.
2621 *
2622 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2623 * cleared otherwise prior to returning.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002624 */
Al Viro2675a4e2012-06-22 12:41:10 +04002625static int lookup_open(struct nameidata *nd, struct path *path,
2626 struct file *file,
2627 const struct open_flags *op,
Al Viro64894cf2012-07-31 00:53:35 +04002628 bool got_write, int *opened)
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002629{
2630 struct dentry *dir = nd->path.dentry;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002631 struct inode *dir_inode = dir->d_inode;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002632 struct dentry *dentry;
2633 int error;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002634 bool need_lookup;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002635
Al Viro47237682012-06-10 05:01:45 -04002636 *opened &= ~FILE_CREATED;
Al Viro201f9562012-06-22 12:42:10 +04002637 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002638 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002639 return PTR_ERR(dentry);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002640
Miklos Szeredid18e9002012-06-05 15:10:17 +02002641 /* Cached positive dentry: will open in f_op->open */
2642 if (!need_lookup && dentry->d_inode)
2643 goto out_no_open;
2644
2645 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
Al Viro64894cf2012-07-31 00:53:35 +04002646 return atomic_open(nd, dentry, path, file, op, got_write,
Al Viro47237682012-06-10 05:01:45 -04002647 need_lookup, opened);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002648 }
2649
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002650 if (need_lookup) {
2651 BUG_ON(dentry->d_inode);
2652
Al Viro72bd8662012-06-10 17:17:17 -04002653 dentry = lookup_real(dir_inode, dentry, nd->flags);
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002654 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002655 return PTR_ERR(dentry);
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002656 }
2657
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002658 /* Negative dentry, just create the file */
2659 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2660 umode_t mode = op->mode;
2661 if (!IS_POSIXACL(dir->d_inode))
2662 mode &= ~current_umask();
2663 /*
2664 * This write is needed to ensure that a
2665 * rw->ro transition does not occur between
2666 * the time when the file is created and when
2667 * a permanent write count is taken through
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002668 * the 'struct file' in finish_open().
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002669 */
Al Viro64894cf2012-07-31 00:53:35 +04002670 if (!got_write) {
2671 error = -EROFS;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002672 goto out_dput;
Al Viro64894cf2012-07-31 00:53:35 +04002673 }
Al Viro47237682012-06-10 05:01:45 -04002674 *opened |= FILE_CREATED;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002675 error = security_path_mknod(&nd->path, dentry, mode, 0);
2676 if (error)
2677 goto out_dput;
Al Viro312b63f2012-06-10 18:09:36 -04002678 error = vfs_create(dir->d_inode, dentry, mode,
2679 nd->flags & LOOKUP_EXCL);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002680 if (error)
2681 goto out_dput;
2682 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02002683out_no_open:
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002684 path->dentry = dentry;
2685 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04002686 return 1;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002687
2688out_dput:
2689 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04002690 return error;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002691}
2692
2693/*
Al Virofe2d35f2011-03-05 22:58:25 -05002694 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11002695 */
Al Viro2675a4e2012-06-22 12:41:10 +04002696static int do_last(struct nameidata *nd, struct path *path,
2697 struct file *file, const struct open_flags *op,
Jeff Layton669abf42012-10-10 16:43:10 -04002698 int *opened, struct filename *name)
Al Virofb1cc552009-12-24 01:58:28 -05002699{
Al Viroa1e28032009-12-24 02:12:06 -05002700 struct dentry *dir = nd->path.dentry;
Al Viroca344a892011-03-09 00:36:45 -05002701 int open_flag = op->open_flag;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002702 bool will_truncate = (open_flag & O_TRUNC) != 0;
Al Viro64894cf2012-07-31 00:53:35 +04002703 bool got_write = false;
Al Virobcda7652011-03-13 16:42:14 -04002704 int acc_mode = op->acc_mode;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02002705 struct inode *inode;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002706 bool symlink_ok = false;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002707 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2708 bool retried = false;
Al Viro16c2cd72011-02-22 15:50:10 -05002709 int error;
Al Virofb1cc552009-12-24 01:58:28 -05002710
Al Viroc3e380b2011-02-23 13:39:45 -05002711 nd->flags &= ~LOOKUP_PARENT;
2712 nd->flags |= op->intent;
2713
Al Viro1f36f772009-12-26 10:56:19 -05002714 switch (nd->last_type) {
2715 case LAST_DOTDOT:
Neil Brown176306f2010-05-24 16:57:56 +10002716 case LAST_DOT:
Al Virofe2d35f2011-03-05 22:58:25 -05002717 error = handle_dots(nd, nd->last_type);
2718 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002719 return error;
Al Viro1f36f772009-12-26 10:56:19 -05002720 /* fallthrough */
Al Viro1f36f772009-12-26 10:56:19 -05002721 case LAST_ROOT:
Al Viro9f1fafe2011-03-25 11:00:12 -04002722 error = complete_walk(nd);
Al Viro16c2cd72011-02-22 15:50:10 -05002723 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002724 return error;
Jeff Laytonadb5c242012-10-10 16:43:13 -04002725 audit_inode(name, nd->path.dentry, 0);
Al Viroca344a892011-03-09 00:36:45 -05002726 if (open_flag & O_CREAT) {
Al Virofe2d35f2011-03-05 22:58:25 -05002727 error = -EISDIR;
Al Viro2675a4e2012-06-22 12:41:10 +04002728 goto out;
Al Virofe2d35f2011-03-05 22:58:25 -05002729 }
Miklos Szeredie83db162012-06-05 15:10:29 +02002730 goto finish_open;
Al Viro1f36f772009-12-26 10:56:19 -05002731 case LAST_BIND:
Al Viro9f1fafe2011-03-25 11:00:12 -04002732 error = complete_walk(nd);
Al Viro16c2cd72011-02-22 15:50:10 -05002733 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002734 return error;
Jeff Laytonadb5c242012-10-10 16:43:13 -04002735 audit_inode(name, dir, 0);
Miklos Szeredie83db162012-06-05 15:10:29 +02002736 goto finish_open;
Al Viro1f36f772009-12-26 10:56:19 -05002737 }
Al Viro67ee3ad2009-12-26 07:01:01 -05002738
Al Viroca344a892011-03-09 00:36:45 -05002739 if (!(open_flag & O_CREAT)) {
Al Virofe2d35f2011-03-05 22:58:25 -05002740 if (nd->last.name[nd->last.len])
2741 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
Al Virobcda7652011-03-13 16:42:14 -04002742 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002743 symlink_ok = true;
Al Virofe2d35f2011-03-05 22:58:25 -05002744 /* we _can_ be in RCU mode here */
Miklos Szeredia1eb3312012-05-21 17:30:07 +02002745 error = lookup_fast(nd, &nd->last, path, &inode);
Miklos Szeredi71574862012-06-05 15:10:14 +02002746 if (likely(!error))
2747 goto finish_lookup;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02002748
Miklos Szeredi71574862012-06-05 15:10:14 +02002749 if (error < 0)
Al Viro2675a4e2012-06-22 12:41:10 +04002750 goto out;
Miklos Szeredi37d7fff2012-06-05 15:10:12 +02002751
Miklos Szeredi71574862012-06-05 15:10:14 +02002752 BUG_ON(nd->inode != dir->d_inode);
Miklos Szeredib6183df2012-06-05 15:10:13 +02002753 } else {
2754 /* create side of things */
2755 /*
2756 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2757 * has been cleared when we got to the last component we are
2758 * about to look up
2759 */
2760 error = complete_walk(nd);
2761 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002762 return error;
Miklos Szeredib6183df2012-06-05 15:10:13 +02002763
Jeff Laytonadb5c242012-10-10 16:43:13 -04002764 audit_inode(name, dir, 0);
Miklos Szeredib6183df2012-06-05 15:10:13 +02002765 error = -EISDIR;
2766 /* trailing slashes? */
2767 if (nd->last.name[nd->last.len])
Al Viro2675a4e2012-06-22 12:41:10 +04002768 goto out;
Al Virofe2d35f2011-03-05 22:58:25 -05002769 }
2770
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002771retry_lookup:
Al Viro64894cf2012-07-31 00:53:35 +04002772 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2773 error = mnt_want_write(nd->path.mnt);
2774 if (!error)
2775 got_write = true;
2776 /*
2777 * do _not_ fail yet - we might not need that or fail with
2778 * a different error; let lookup_open() decide; we'll be
2779 * dropping this one anyway.
2780 */
2781 }
Al Viroa1e28032009-12-24 02:12:06 -05002782 mutex_lock(&dir->d_inode->i_mutex);
Al Viro64894cf2012-07-31 00:53:35 +04002783 error = lookup_open(nd, path, file, op, got_write, opened);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002784 mutex_unlock(&dir->d_inode->i_mutex);
Al Viroa1e28032009-12-24 02:12:06 -05002785
Al Viro2675a4e2012-06-22 12:41:10 +04002786 if (error <= 0) {
2787 if (error)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002788 goto out;
2789
Al Viro47237682012-06-10 05:01:45 -04002790 if ((*opened & FILE_CREATED) ||
Al Viro2675a4e2012-06-22 12:41:10 +04002791 !S_ISREG(file->f_path.dentry->d_inode->i_mode))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002792 will_truncate = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002793
Jeff Laytonadb5c242012-10-10 16:43:13 -04002794 audit_inode(name, file->f_path.dentry, 0);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002795 goto opened;
2796 }
Al Virofb1cc552009-12-24 01:58:28 -05002797
Al Viro47237682012-06-10 05:01:45 -04002798 if (*opened & FILE_CREATED) {
Al Viro9b44f1b2011-03-09 00:17:27 -05002799 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05002800 open_flag &= ~O_TRUNC;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002801 will_truncate = false;
Al Virobcda7652011-03-13 16:42:14 -04002802 acc_mode = MAY_OPEN;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002803 path_to_nameidata(path, nd);
Miklos Szeredie83db162012-06-05 15:10:29 +02002804 goto finish_open_created;
Al Virofb1cc552009-12-24 01:58:28 -05002805 }
2806
2807 /*
Jeff Layton3134f372012-07-25 10:19:47 -04002808 * create/update audit record if it already exists.
Al Virofb1cc552009-12-24 01:58:28 -05002809 */
Jeff Layton3134f372012-07-25 10:19:47 -04002810 if (path->dentry->d_inode)
Jeff Laytonadb5c242012-10-10 16:43:13 -04002811 audit_inode(name, path->dentry, 0);
Al Virofb1cc552009-12-24 01:58:28 -05002812
Miklos Szeredid18e9002012-06-05 15:10:17 +02002813 /*
2814 * If atomic_open() acquired write access it is dropped now due to
2815 * possible mount and symlink following (this might be optimized away if
2816 * necessary...)
2817 */
Al Viro64894cf2012-07-31 00:53:35 +04002818 if (got_write) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002819 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04002820 got_write = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002821 }
2822
Al Virofb1cc552009-12-24 01:58:28 -05002823 error = -EEXIST;
Al Virof8310c52012-07-30 11:50:30 +04002824 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
Al Virofb1cc552009-12-24 01:58:28 -05002825 goto exit_dput;
2826
David Howells9875cf82011-01-14 18:45:21 +00002827 error = follow_managed(path, nd->flags);
2828 if (error < 0)
2829 goto exit_dput;
Al Virofb1cc552009-12-24 01:58:28 -05002830
Al Viroa3fbbde2011-11-07 21:21:26 +00002831 if (error)
2832 nd->flags |= LOOKUP_JUMPED;
2833
Miklos Szeredidecf3402012-05-21 17:30:08 +02002834 BUG_ON(nd->flags & LOOKUP_RCU);
2835 inode = path->dentry->d_inode;
Miklos Szeredi5f5daac2012-05-21 17:30:14 +02002836finish_lookup:
2837 /* we _can_ be in RCU mode here */
Al Virofb1cc552009-12-24 01:58:28 -05002838 error = -ENOENT;
Miklos Szeredi54c33e72012-05-21 17:30:10 +02002839 if (!inode) {
2840 path_to_nameidata(path, nd);
Al Viro2675a4e2012-06-22 12:41:10 +04002841 goto out;
Miklos Szeredi54c33e72012-05-21 17:30:10 +02002842 }
Al Viro9e67f362009-12-26 07:04:50 -05002843
Miklos Szeredid45ea862012-05-21 17:30:09 +02002844 if (should_follow_link(inode, !symlink_ok)) {
2845 if (nd->flags & LOOKUP_RCU) {
2846 if (unlikely(unlazy_walk(nd, path->dentry))) {
2847 error = -ECHILD;
Al Viro2675a4e2012-06-22 12:41:10 +04002848 goto out;
Miklos Szeredid45ea862012-05-21 17:30:09 +02002849 }
2850 }
2851 BUG_ON(inode != path->dentry->d_inode);
Al Viro2675a4e2012-06-22 12:41:10 +04002852 return 1;
Miklos Szeredid45ea862012-05-21 17:30:09 +02002853 }
Al Virofb1cc552009-12-24 01:58:28 -05002854
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002855 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
2856 path_to_nameidata(path, nd);
2857 } else {
2858 save_parent.dentry = nd->path.dentry;
2859 save_parent.mnt = mntget(path->mnt);
2860 nd->path.dentry = path->dentry;
2861
2862 }
Miklos Szeredidecf3402012-05-21 17:30:08 +02002863 nd->inode = inode;
Al Viroa3fbbde2011-11-07 21:21:26 +00002864 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2865 error = complete_walk(nd);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002866 if (error) {
2867 path_put(&save_parent);
Al Viro2675a4e2012-06-22 12:41:10 +04002868 return error;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002869 }
Al Virofb1cc552009-12-24 01:58:28 -05002870 error = -EISDIR;
Miklos Szeredi050ac842012-05-21 17:30:12 +02002871 if ((open_flag & O_CREAT) && S_ISDIR(nd->inode->i_mode))
Al Viro2675a4e2012-06-22 12:41:10 +04002872 goto out;
Miklos Szerediaf2f5542012-05-21 17:30:11 +02002873 error = -ENOTDIR;
2874 if ((nd->flags & LOOKUP_DIRECTORY) && !nd->inode->i_op->lookup)
Al Viro2675a4e2012-06-22 12:41:10 +04002875 goto out;
Jeff Laytonadb5c242012-10-10 16:43:13 -04002876 audit_inode(name, nd->path.dentry, 0);
Miklos Szeredie83db162012-06-05 15:10:29 +02002877finish_open:
Al Viro6c0d46c2011-03-09 00:59:59 -05002878 if (!S_ISREG(nd->inode->i_mode))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002879 will_truncate = false;
Al Viro6c0d46c2011-03-09 00:59:59 -05002880
Al Viro0f9d1a12011-03-09 00:13:14 -05002881 if (will_truncate) {
2882 error = mnt_want_write(nd->path.mnt);
2883 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002884 goto out;
Al Viro64894cf2012-07-31 00:53:35 +04002885 got_write = true;
Al Viro0f9d1a12011-03-09 00:13:14 -05002886 }
Miklos Szeredie83db162012-06-05 15:10:29 +02002887finish_open_created:
Al Virobcda7652011-03-13 16:42:14 -04002888 error = may_open(&nd->path, acc_mode, open_flag);
Al Viroca344a892011-03-09 00:36:45 -05002889 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002890 goto out;
Al Viro30d90492012-06-22 12:40:19 +04002891 file->f_path.mnt = nd->path.mnt;
2892 error = finish_open(file, nd->path.dentry, NULL, opened);
2893 if (error) {
Al Viro30d90492012-06-22 12:40:19 +04002894 if (error == -EOPENSTALE)
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02002895 goto stale_open;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002896 goto out;
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02002897 }
Miklos Szeredia8277b92012-06-05 15:10:32 +02002898opened:
Al Viro2675a4e2012-06-22 12:41:10 +04002899 error = open_check_o_direct(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002900 if (error)
2901 goto exit_fput;
Al Viro2675a4e2012-06-22 12:41:10 +04002902 error = ima_file_check(file, op->acc_mode);
Miklos Szerediaa4caad2012-06-05 15:10:28 +02002903 if (error)
2904 goto exit_fput;
2905
2906 if (will_truncate) {
Al Viro2675a4e2012-06-22 12:41:10 +04002907 error = handle_truncate(file);
Miklos Szerediaa4caad2012-06-05 15:10:28 +02002908 if (error)
2909 goto exit_fput;
Al Viro0f9d1a12011-03-09 00:13:14 -05002910 }
Al Viroca344a892011-03-09 00:36:45 -05002911out:
Al Viro64894cf2012-07-31 00:53:35 +04002912 if (got_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05002913 mnt_drop_write(nd->path.mnt);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002914 path_put(&save_parent);
Miklos Szeredie276ae62012-05-21 17:30:06 +02002915 terminate_walk(nd);
Al Viro2675a4e2012-06-22 12:41:10 +04002916 return error;
Al Virofb1cc552009-12-24 01:58:28 -05002917
Al Virofb1cc552009-12-24 01:58:28 -05002918exit_dput:
2919 path_put_conditional(path, nd);
Al Viroca344a892011-03-09 00:36:45 -05002920 goto out;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002921exit_fput:
Al Viro2675a4e2012-06-22 12:41:10 +04002922 fput(file);
2923 goto out;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002924
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02002925stale_open:
2926 /* If no saved parent or already retried then can't retry */
2927 if (!save_parent.dentry || retried)
2928 goto out;
2929
2930 BUG_ON(save_parent.dentry != dir);
2931 path_put(&nd->path);
2932 nd->path = save_parent;
2933 nd->inode = dir->d_inode;
2934 save_parent.mnt = NULL;
2935 save_parent.dentry = NULL;
Al Viro64894cf2012-07-31 00:53:35 +04002936 if (got_write) {
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02002937 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04002938 got_write = false;
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02002939 }
2940 retried = true;
2941 goto retry_lookup;
Al Virofb1cc552009-12-24 01:58:28 -05002942}
2943
Jeff Layton669abf42012-10-10 16:43:10 -04002944static struct file *path_openat(int dfd, struct filename *pathname,
Al Viro73d049a2011-03-11 12:08:24 -05002945 struct nameidata *nd, const struct open_flags *op, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002946{
Al Virofe2d35f2011-03-05 22:58:25 -05002947 struct file *base = NULL;
Al Viro30d90492012-06-22 12:40:19 +04002948 struct file *file;
Al Viro9850c052010-01-13 15:01:15 -05002949 struct path path;
Al Viro47237682012-06-10 05:01:45 -04002950 int opened = 0;
Al Viro13aab422011-02-23 17:54:08 -05002951 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11002952
Al Viro30d90492012-06-22 12:40:19 +04002953 file = get_empty_filp();
2954 if (!file)
Nick Piggin31e6b012011-01-07 17:49:52 +11002955 return ERR_PTR(-ENFILE);
2956
Al Viro30d90492012-06-22 12:40:19 +04002957 file->f_flags = op->open_flag;
Nick Piggin31e6b012011-01-07 17:49:52 +11002958
Jeff Layton669abf42012-10-10 16:43:10 -04002959 error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base);
Nick Piggin31e6b012011-01-07 17:49:52 +11002960 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04002961 goto out;
Nick Piggin31e6b012011-01-07 17:49:52 +11002962
Al Virofe2d35f2011-03-05 22:58:25 -05002963 current->total_link_count = 0;
Jeff Layton669abf42012-10-10 16:43:10 -04002964 error = link_path_walk(pathname->name, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11002965 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04002966 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002967
Al Viro2675a4e2012-06-22 12:41:10 +04002968 error = do_last(nd, &path, file, op, &opened, pathname);
2969 while (unlikely(error > 0)) { /* trailing symlink */
Nick Piggin7b9337a2011-01-14 08:42:43 +00002970 struct path link = path;
Al Virodef4af32009-12-26 08:37:05 -05002971 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04002972 if (!(nd->flags & LOOKUP_FOLLOW)) {
Al Viro73d049a2011-03-11 12:08:24 -05002973 path_put_conditional(&path, nd);
2974 path_put(&nd->path);
Al Viro2675a4e2012-06-22 12:41:10 +04002975 error = -ELOOP;
Al Viro40b39132011-03-09 16:22:18 -05002976 break;
2977 }
Kees Cook800179c2012-07-25 17:29:07 -07002978 error = may_follow_link(&link, nd);
2979 if (unlikely(error))
2980 break;
Al Viro73d049a2011-03-11 12:08:24 -05002981 nd->flags |= LOOKUP_PARENT;
2982 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
Al Viro574197e2011-03-14 22:20:34 -04002983 error = follow_link(&link, nd, &cookie);
Al Viroc3e380b2011-02-23 13:39:45 -05002984 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04002985 break;
2986 error = do_last(nd, &path, file, op, &opened, pathname);
Al Viro574197e2011-03-14 22:20:34 -04002987 put_link(nd, &link, cookie);
Al Viro806b6812009-12-26 07:16:40 -05002988 }
Al Viro10fa8e62009-12-26 07:09:49 -05002989out:
Al Viro73d049a2011-03-11 12:08:24 -05002990 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2991 path_put(&nd->root);
Al Virofe2d35f2011-03-05 22:58:25 -05002992 if (base)
2993 fput(base);
Al Viro2675a4e2012-06-22 12:41:10 +04002994 if (!(opened & FILE_OPENED)) {
2995 BUG_ON(!error);
Al Viro30d90492012-06-22 12:40:19 +04002996 put_filp(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002997 }
Al Viro2675a4e2012-06-22 12:41:10 +04002998 if (unlikely(error)) {
2999 if (error == -EOPENSTALE) {
3000 if (flags & LOOKUP_RCU)
3001 error = -ECHILD;
3002 else
3003 error = -ESTALE;
3004 }
3005 file = ERR_PTR(error);
3006 }
3007 return file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003008}
3009
Jeff Layton669abf42012-10-10 16:43:10 -04003010struct file *do_filp_open(int dfd, struct filename *pathname,
Al Viro13aab422011-02-23 17:54:08 -05003011 const struct open_flags *op, int flags)
3012{
Al Viro73d049a2011-03-11 12:08:24 -05003013 struct nameidata nd;
Al Viro13aab422011-02-23 17:54:08 -05003014 struct file *filp;
3015
Al Viro73d049a2011-03-11 12:08:24 -05003016 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
Al Viro13aab422011-02-23 17:54:08 -05003017 if (unlikely(filp == ERR_PTR(-ECHILD)))
Al Viro73d049a2011-03-11 12:08:24 -05003018 filp = path_openat(dfd, pathname, &nd, op, flags);
Al Viro13aab422011-02-23 17:54:08 -05003019 if (unlikely(filp == ERR_PTR(-ESTALE)))
Al Viro73d049a2011-03-11 12:08:24 -05003020 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
Al Viro13aab422011-02-23 17:54:08 -05003021 return filp;
3022}
3023
Al Viro73d049a2011-03-11 12:08:24 -05003024struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3025 const char *name, const struct open_flags *op, int flags)
3026{
3027 struct nameidata nd;
3028 struct file *file;
Jeff Layton669abf42012-10-10 16:43:10 -04003029 struct filename filename = { .name = name };
Al Viro73d049a2011-03-11 12:08:24 -05003030
3031 nd.root.mnt = mnt;
3032 nd.root.dentry = dentry;
3033
3034 flags |= LOOKUP_ROOT;
3035
Al Virobcda7652011-03-13 16:42:14 -04003036 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
Al Viro73d049a2011-03-11 12:08:24 -05003037 return ERR_PTR(-ELOOP);
3038
Jeff Layton669abf42012-10-10 16:43:10 -04003039 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU);
Al Viro73d049a2011-03-11 12:08:24 -05003040 if (unlikely(file == ERR_PTR(-ECHILD)))
Jeff Layton669abf42012-10-10 16:43:10 -04003041 file = path_openat(-1, &filename, &nd, op, flags);
Al Viro73d049a2011-03-11 12:08:24 -05003042 if (unlikely(file == ERR_PTR(-ESTALE)))
Jeff Layton669abf42012-10-10 16:43:10 -04003043 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL);
Al Viro73d049a2011-03-11 12:08:24 -05003044 return file;
3045}
3046
Al Viroed75e952011-06-27 16:53:43 -04003047struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003048{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003049 struct dentry *dentry = ERR_PTR(-EEXIST);
Al Viroed75e952011-06-27 16:53:43 -04003050 struct nameidata nd;
Jan Karac30dabf2012-06-12 16:20:30 +02003051 int err2;
Al Viroed75e952011-06-27 16:53:43 -04003052 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
3053 if (error)
3054 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003055
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003056 /*
3057 * Yucky last component or no last component at all?
3058 * (foo/., foo/.., /////)
3059 */
Al Viroed75e952011-06-27 16:53:43 -04003060 if (nd.last_type != LAST_NORM)
3061 goto out;
3062 nd.flags &= ~LOOKUP_PARENT;
3063 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003064
Jan Karac30dabf2012-06-12 16:20:30 +02003065 /* don't fail immediately if it's r/o, at least try to report other errors */
3066 err2 = mnt_want_write(nd.path.mnt);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003067 /*
3068 * Do the final lookup.
3069 */
Al Viroed75e952011-06-27 16:53:43 -04003070 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3071 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003072 if (IS_ERR(dentry))
Al Viroa8104a92012-07-20 02:25:00 +04003073 goto unlock;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003074
Al Viroa8104a92012-07-20 02:25:00 +04003075 error = -EEXIST;
Al Viroe9baf6e2008-05-15 04:49:12 -04003076 if (dentry->d_inode)
Al Viroa8104a92012-07-20 02:25:00 +04003077 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003078 /*
3079 * Special case - lookup gave negative, but... we had foo/bar/
3080 * From the vfs_mknod() POV we just have a negative dentry -
3081 * all is fine. Let's be bastards - you had / on the end, you've
3082 * been asking for (non-existent) directory. -ENOENT for you.
3083 */
Al Viroed75e952011-06-27 16:53:43 -04003084 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
Al Viroa8104a92012-07-20 02:25:00 +04003085 error = -ENOENT;
Al Viroed75e952011-06-27 16:53:43 -04003086 goto fail;
Al Viroe9baf6e2008-05-15 04:49:12 -04003087 }
Jan Karac30dabf2012-06-12 16:20:30 +02003088 if (unlikely(err2)) {
3089 error = err2;
Al Viroa8104a92012-07-20 02:25:00 +04003090 goto fail;
Jan Karac30dabf2012-06-12 16:20:30 +02003091 }
Al Viroed75e952011-06-27 16:53:43 -04003092 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003093 return dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003094fail:
Al Viroa8104a92012-07-20 02:25:00 +04003095 dput(dentry);
3096 dentry = ERR_PTR(error);
3097unlock:
Al Viroed75e952011-06-27 16:53:43 -04003098 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Karac30dabf2012-06-12 16:20:30 +02003099 if (!err2)
3100 mnt_drop_write(nd.path.mnt);
Al Viroed75e952011-06-27 16:53:43 -04003101out:
3102 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003103 return dentry;
3104}
Al Virodae6ad82011-06-26 11:50:15 -04003105EXPORT_SYMBOL(kern_path_create);
3106
Al Viro921a1652012-07-20 01:15:31 +04003107void done_path_create(struct path *path, struct dentry *dentry)
3108{
3109 dput(dentry);
3110 mutex_unlock(&path->dentry->d_inode->i_mutex);
Al Viroa8104a92012-07-20 02:25:00 +04003111 mnt_drop_write(path->mnt);
Al Viro921a1652012-07-20 01:15:31 +04003112 path_put(path);
3113}
3114EXPORT_SYMBOL(done_path_create);
3115
Al Virodae6ad82011-06-26 11:50:15 -04003116struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
3117{
Jeff Layton91a27b22012-10-10 15:25:28 -04003118 struct filename *tmp = getname(pathname);
Al Virodae6ad82011-06-26 11:50:15 -04003119 struct dentry *res;
3120 if (IS_ERR(tmp))
3121 return ERR_CAST(tmp);
Jeff Layton91a27b22012-10-10 15:25:28 -04003122 res = kern_path_create(dfd, tmp->name, path, is_dir);
Al Virodae6ad82011-06-26 11:50:15 -04003123 putname(tmp);
3124 return res;
3125}
3126EXPORT_SYMBOL(user_path_create);
3127
Al Viro1a67aaf2011-07-26 01:52:52 -04003128int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003129{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003130 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003131
3132 if (error)
3133 return error;
3134
Eric W. Biederman975d6b32011-11-13 12:16:43 -08003135 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003136 return -EPERM;
3137
Al Viroacfa4382008-12-04 10:06:33 -05003138 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003139 return -EPERM;
3140
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07003141 error = devcgroup_inode_mknod(mode, dev);
3142 if (error)
3143 return error;
3144
Linus Torvalds1da177e2005-04-16 15:20:36 -07003145 error = security_inode_mknod(dir, dentry, mode, dev);
3146 if (error)
3147 return error;
3148
Linus Torvalds1da177e2005-04-16 15:20:36 -07003149 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003150 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003151 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003152 return error;
3153}
3154
Al Virof69aac02011-07-26 04:31:40 -04003155static int may_mknod(umode_t mode)
Dave Hansen463c3192008-02-15 14:37:57 -08003156{
3157 switch (mode & S_IFMT) {
3158 case S_IFREG:
3159 case S_IFCHR:
3160 case S_IFBLK:
3161 case S_IFIFO:
3162 case S_IFSOCK:
3163 case 0: /* zero mode translates to S_IFREG */
3164 return 0;
3165 case S_IFDIR:
3166 return -EPERM;
3167 default:
3168 return -EINVAL;
3169 }
3170}
3171
Al Viro8208a222011-07-25 17:32:17 -04003172SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003173 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003174{
Al Viro2ad94ae2008-07-21 09:32:51 -04003175 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003176 struct path path;
3177 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003178
Al Viro8e4bfca2012-07-20 01:17:26 +04003179 error = may_mknod(mode);
3180 if (error)
3181 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003182
Al Virodae6ad82011-06-26 11:50:15 -04003183 dentry = user_path_create(dfd, filename, &path, 0);
3184 if (IS_ERR(dentry))
3185 return PTR_ERR(dentry);
Al Viro2ad94ae2008-07-21 09:32:51 -04003186
Al Virodae6ad82011-06-26 11:50:15 -04003187 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003188 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003189 error = security_path_mknod(&path, dentry, mode, dev);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003190 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04003191 goto out;
Dave Hansen463c3192008-02-15 14:37:57 -08003192 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003193 case 0: case S_IFREG:
Al Viro312b63f2012-06-10 18:09:36 -04003194 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003195 break;
3196 case S_IFCHR: case S_IFBLK:
Al Virodae6ad82011-06-26 11:50:15 -04003197 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003198 new_decode_dev(dev));
3199 break;
3200 case S_IFIFO: case S_IFSOCK:
Al Virodae6ad82011-06-26 11:50:15 -04003201 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003202 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003203 }
Al Viroa8104a92012-07-20 02:25:00 +04003204out:
Al Viro921a1652012-07-20 01:15:31 +04003205 done_path_create(&path, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003206 return error;
3207}
3208
Al Viro8208a222011-07-25 17:32:17 -04003209SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003210{
3211 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3212}
3213
Al Viro18bb1db2011-07-26 01:41:39 -04003214int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003215{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003216 int error = may_create(dir, dentry);
Al Viro8de52772012-02-06 12:45:27 -05003217 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003218
3219 if (error)
3220 return error;
3221
Al Viroacfa4382008-12-04 10:06:33 -05003222 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003223 return -EPERM;
3224
3225 mode &= (S_IRWXUGO|S_ISVTX);
3226 error = security_inode_mkdir(dir, dentry, mode);
3227 if (error)
3228 return error;
3229
Al Viro8de52772012-02-06 12:45:27 -05003230 if (max_links && dir->i_nlink >= max_links)
3231 return -EMLINK;
3232
Linus Torvalds1da177e2005-04-16 15:20:36 -07003233 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003234 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003235 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003236 return error;
3237}
3238
Al Viroa218d0f2011-11-21 14:59:34 -05003239SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003240{
Dave Hansen6902d922006-09-30 23:29:01 -07003241 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003242 struct path path;
3243 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003244
Al Virodae6ad82011-06-26 11:50:15 -04003245 dentry = user_path_create(dfd, pathname, &path, 1);
Dave Hansen6902d922006-09-30 23:29:01 -07003246 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003247 return PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003248
Al Virodae6ad82011-06-26 11:50:15 -04003249 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003250 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003251 error = security_path_mkdir(&path, dentry, mode);
Al Viroa8104a92012-07-20 02:25:00 +04003252 if (!error)
3253 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
Al Viro921a1652012-07-20 01:15:31 +04003254 done_path_create(&path, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003255 return error;
3256}
3257
Al Viroa218d0f2011-11-21 14:59:34 -05003258SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003259{
3260 return sys_mkdirat(AT_FDCWD, pathname, mode);
3261}
3262
Linus Torvalds1da177e2005-04-16 15:20:36 -07003263/*
Sage Weila71905f2011-05-24 13:06:08 -07003264 * The dentry_unhash() helper will try to drop the dentry early: we
J. Bruce Fieldsc0d02592012-02-15 11:48:40 -05003265 * should have a usage count of 1 if we're the only user of this
Sage Weila71905f2011-05-24 13:06:08 -07003266 * dentry, and if that is true (possibly after pruning the dcache),
3267 * then we drop the dentry now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003268 *
3269 * A low-level filesystem can, if it choses, legally
3270 * do a
3271 *
3272 * if (!d_unhashed(dentry))
3273 * return -EBUSY;
3274 *
3275 * if it cannot handle the case of removing a directory
3276 * that is still in use by something else..
3277 */
3278void dentry_unhash(struct dentry *dentry)
3279{
Vasily Averindc168422006-12-06 20:37:07 -08003280 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003281 spin_lock(&dentry->d_lock);
Sage Weil64252c72011-05-24 13:06:05 -07003282 if (dentry->d_count == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003283 __d_drop(dentry);
3284 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003285}
3286
3287int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3288{
3289 int error = may_delete(dir, dentry, 1);
3290
3291 if (error)
3292 return error;
3293
Al Viroacfa4382008-12-04 10:06:33 -05003294 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003295 return -EPERM;
3296
Al Viro1d2ef592011-09-14 18:55:41 +01003297 dget(dentry);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003298 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003299
Sage Weil912dbc12011-05-24 13:06:11 -07003300 error = -EBUSY;
3301 if (d_mountpoint(dentry))
3302 goto out;
3303
3304 error = security_inode_rmdir(dir, dentry);
3305 if (error)
3306 goto out;
3307
Sage Weil3cebde22011-05-29 21:20:59 -07003308 shrink_dcache_parent(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003309 error = dir->i_op->rmdir(dir, dentry);
3310 if (error)
3311 goto out;
3312
3313 dentry->d_inode->i_flags |= S_DEAD;
3314 dont_mount(dentry);
3315
3316out:
3317 mutex_unlock(&dentry->d_inode->i_mutex);
Al Viro1d2ef592011-09-14 18:55:41 +01003318 dput(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003319 if (!error)
3320 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003321 return error;
3322}
3323
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003324static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003325{
3326 int error = 0;
Jeff Layton91a27b22012-10-10 15:25:28 -04003327 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003328 struct dentry *dentry;
3329 struct nameidata nd;
3330
Jeff Layton91a27b22012-10-10 15:25:28 -04003331 name = user_path_parent(dfd, pathname, &nd);
3332 if (IS_ERR(name))
3333 return PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003334
3335 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003336 case LAST_DOTDOT:
3337 error = -ENOTEMPTY;
3338 goto exit1;
3339 case LAST_DOT:
3340 error = -EINVAL;
3341 goto exit1;
3342 case LAST_ROOT:
3343 error = -EBUSY;
3344 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003345 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003346
3347 nd.flags &= ~LOOKUP_PARENT;
Jan Karac30dabf2012-06-12 16:20:30 +02003348 error = mnt_want_write(nd.path.mnt);
3349 if (error)
3350 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003351
Jan Blunck4ac91372008-02-14 19:34:32 -08003352 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08003353 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003354 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003355 if (IS_ERR(dentry))
3356 goto exit2;
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003357 if (!dentry->d_inode) {
3358 error = -ENOENT;
3359 goto exit3;
3360 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003361 error = security_path_rmdir(&nd.path, dentry);
3362 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003363 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08003364 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Dave Hansen06227532008-02-15 14:37:34 -08003365exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07003366 dput(dentry);
3367exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08003368 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Karac30dabf2012-06-12 16:20:30 +02003369 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003370exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003371 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003372 putname(name);
3373 return error;
3374}
3375
Heiko Carstens3cdad422009-01-14 14:14:22 +01003376SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003377{
3378 return do_rmdir(AT_FDCWD, pathname);
3379}
3380
Linus Torvalds1da177e2005-04-16 15:20:36 -07003381int vfs_unlink(struct inode *dir, struct dentry *dentry)
3382{
3383 int error = may_delete(dir, dentry, 0);
3384
3385 if (error)
3386 return error;
3387
Al Viroacfa4382008-12-04 10:06:33 -05003388 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003389 return -EPERM;
3390
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003391 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003392 if (d_mountpoint(dentry))
3393 error = -EBUSY;
3394 else {
3395 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05003396 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003397 error = dir->i_op->unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05003398 if (!error)
Al Virod83c49f2010-04-30 17:17:09 -04003399 dont_mount(dentry);
Al Virobec10522010-03-03 14:12:08 -05003400 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003401 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003402 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003403
3404 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3405 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08003406 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003407 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003408 }
Robert Love0eeca282005-07-12 17:06:03 -04003409
Linus Torvalds1da177e2005-04-16 15:20:36 -07003410 return error;
3411}
3412
3413/*
3414 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003415 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07003416 * writeout happening, and we don't want to prevent access to the directory
3417 * while waiting on the I/O.
3418 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003419static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003420{
Al Viro2ad94ae2008-07-21 09:32:51 -04003421 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04003422 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003423 struct dentry *dentry;
3424 struct nameidata nd;
3425 struct inode *inode = NULL;
3426
Jeff Layton91a27b22012-10-10 15:25:28 -04003427 name = user_path_parent(dfd, pathname, &nd);
3428 if (IS_ERR(name))
3429 return PTR_ERR(name);
Al Viro2ad94ae2008-07-21 09:32:51 -04003430
Linus Torvalds1da177e2005-04-16 15:20:36 -07003431 error = -EISDIR;
3432 if (nd.last_type != LAST_NORM)
3433 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003434
3435 nd.flags &= ~LOOKUP_PARENT;
Jan Karac30dabf2012-06-12 16:20:30 +02003436 error = mnt_want_write(nd.path.mnt);
3437 if (error)
3438 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003439
Jan Blunck4ac91372008-02-14 19:34:32 -08003440 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08003441 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003442 error = PTR_ERR(dentry);
3443 if (!IS_ERR(dentry)) {
3444 /* Why not before? Because we want correct error value */
Török Edwin50338b82011-06-16 00:06:14 +03003445 if (nd.last.name[nd.last.len])
3446 goto slashes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003447 inode = dentry->d_inode;
Török Edwin50338b82011-06-16 00:06:14 +03003448 if (!inode)
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003449 goto slashes;
3450 ihold(inode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003451 error = security_path_unlink(&nd.path, dentry);
3452 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003453 goto exit2;
Jan Blunck4ac91372008-02-14 19:34:32 -08003454 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Jan Karac30dabf2012-06-12 16:20:30 +02003455exit2:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003456 dput(dentry);
3457 }
Jan Blunck4ac91372008-02-14 19:34:32 -08003458 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003459 if (inode)
3460 iput(inode); /* truncate the inode here */
Jan Karac30dabf2012-06-12 16:20:30 +02003461 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003462exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003463 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003464 putname(name);
3465 return error;
3466
3467slashes:
3468 error = !dentry->d_inode ? -ENOENT :
3469 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
3470 goto exit2;
3471}
3472
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003473SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003474{
3475 if ((flag & ~AT_REMOVEDIR) != 0)
3476 return -EINVAL;
3477
3478 if (flag & AT_REMOVEDIR)
3479 return do_rmdir(dfd, pathname);
3480
3481 return do_unlinkat(dfd, pathname);
3482}
3483
Heiko Carstens3480b252009-01-14 14:14:16 +01003484SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003485{
3486 return do_unlinkat(AT_FDCWD, pathname);
3487}
3488
Miklos Szeredidb2e7472008-06-24 16:50:16 +02003489int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003490{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003491 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003492
3493 if (error)
3494 return error;
3495
Al Viroacfa4382008-12-04 10:06:33 -05003496 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003497 return -EPERM;
3498
3499 error = security_inode_symlink(dir, dentry, oldname);
3500 if (error)
3501 return error;
3502
Linus Torvalds1da177e2005-04-16 15:20:36 -07003503 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003504 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003505 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003506 return error;
3507}
3508
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003509SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3510 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003511{
Al Viro2ad94ae2008-07-21 09:32:51 -04003512 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04003513 struct filename *from;
Dave Hansen6902d922006-09-30 23:29:01 -07003514 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003515 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003516
3517 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04003518 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003519 return PTR_ERR(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04003520
Al Virodae6ad82011-06-26 11:50:15 -04003521 dentry = user_path_create(newdfd, newname, &path, 0);
Dave Hansen6902d922006-09-30 23:29:01 -07003522 error = PTR_ERR(dentry);
3523 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003524 goto out_putname;
Dave Hansen6902d922006-09-30 23:29:01 -07003525
Jeff Layton91a27b22012-10-10 15:25:28 -04003526 error = security_path_symlink(&path, dentry, from->name);
Al Viroa8104a92012-07-20 02:25:00 +04003527 if (!error)
Jeff Layton91a27b22012-10-10 15:25:28 -04003528 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
Al Viro921a1652012-07-20 01:15:31 +04003529 done_path_create(&path, dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003530out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003531 putname(from);
3532 return error;
3533}
3534
Heiko Carstens3480b252009-01-14 14:14:16 +01003535SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003536{
3537 return sys_symlinkat(oldname, AT_FDCWD, newname);
3538}
3539
Linus Torvalds1da177e2005-04-16 15:20:36 -07003540int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
3541{
3542 struct inode *inode = old_dentry->d_inode;
Al Viro8de52772012-02-06 12:45:27 -05003543 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003544 int error;
3545
3546 if (!inode)
3547 return -ENOENT;
3548
Miklos Szeredia95164d2008-07-30 15:08:48 +02003549 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003550 if (error)
3551 return error;
3552
3553 if (dir->i_sb != inode->i_sb)
3554 return -EXDEV;
3555
3556 /*
3557 * A link to an append-only or immutable file cannot be created.
3558 */
3559 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3560 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05003561 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003562 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003563 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003564 return -EPERM;
3565
3566 error = security_inode_link(old_dentry, dir, new_dentry);
3567 if (error)
3568 return error;
3569
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003570 mutex_lock(&inode->i_mutex);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05303571 /* Make sure we don't allow creating hardlink to an unlinked file */
3572 if (inode->i_nlink == 0)
3573 error = -ENOENT;
Al Viro8de52772012-02-06 12:45:27 -05003574 else if (max_links && inode->i_nlink >= max_links)
3575 error = -EMLINK;
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05303576 else
3577 error = dir->i_op->link(old_dentry, dir, new_dentry);
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003578 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07003579 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003580 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003581 return error;
3582}
3583
3584/*
3585 * Hardlinks are often used in delicate situations. We avoid
3586 * security-related surprises by not following symlinks on the
3587 * newname. --KAB
3588 *
3589 * We don't follow them on the oldname either to be compatible
3590 * with linux 2.0, and to avoid hard-linking to directories
3591 * and other special files. --ADM
3592 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003593SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3594 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003595{
3596 struct dentry *new_dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003597 struct path old_path, new_path;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303598 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003599 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003600
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303601 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003602 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303603 /*
3604 * To use null names we require CAP_DAC_READ_SEARCH
3605 * This ensures that not everyone will be able to create
3606 * handlink using the passed filedescriptor.
3607 */
3608 if (flags & AT_EMPTY_PATH) {
3609 if (!capable(CAP_DAC_READ_SEARCH))
3610 return -ENOENT;
3611 how = LOOKUP_EMPTY;
3612 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003613
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303614 if (flags & AT_SYMLINK_FOLLOW)
3615 how |= LOOKUP_FOLLOW;
3616
3617 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003618 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04003619 return error;
3620
Al Virodae6ad82011-06-26 11:50:15 -04003621 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003622 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003623 if (IS_ERR(new_dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003624 goto out;
3625
3626 error = -EXDEV;
3627 if (old_path.mnt != new_path.mnt)
3628 goto out_dput;
Kees Cook800179c2012-07-25 17:29:07 -07003629 error = may_linkat(&old_path);
3630 if (unlikely(error))
3631 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04003632 error = security_path_link(old_path.dentry, &new_path, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003633 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04003634 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04003635 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
Dave Hansen75c3f292008-02-15 14:37:45 -08003636out_dput:
Al Viro921a1652012-07-20 01:15:31 +04003637 done_path_create(&new_path, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003638out:
Al Viro2d8f3032008-07-22 09:59:21 -04003639 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003640
3641 return error;
3642}
3643
Heiko Carstens3480b252009-01-14 14:14:16 +01003644SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003645{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003646 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003647}
3648
Linus Torvalds1da177e2005-04-16 15:20:36 -07003649/*
3650 * The worst of all namespace operations - renaming directory. "Perverted"
3651 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3652 * Problems:
3653 * a) we can get into loop creation. Check is done in is_subdir().
3654 * b) race potential - two innocent renames can create a loop together.
3655 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003656 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07003657 * story.
3658 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003659 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07003660 * whether the target exists). Solution: try to be smart with locking
3661 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003662 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07003663 * move will be locked. Thus we can rank directories by the tree
3664 * (ancestors first) and rank all non-directories after them.
3665 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003666 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003667 * HOWEVER, it relies on the assumption that any object with ->lookup()
3668 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3669 * we'd better make sure that there's no link(2) for them.
Sage Weile4eaac02011-05-24 13:06:07 -07003670 * d) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003671 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07003672 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05003673 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07003674 * locking].
3675 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07003676static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3677 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003678{
3679 int error = 0;
Sage Weil9055cba2011-05-24 13:06:12 -07003680 struct inode *target = new_dentry->d_inode;
Al Viro8de52772012-02-06 12:45:27 -05003681 unsigned max_links = new_dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003682
3683 /*
3684 * If we are going to change the parent - check write permissions,
3685 * we'll need to flip '..'.
3686 */
3687 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04003688 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003689 if (error)
3690 return error;
3691 }
3692
3693 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3694 if (error)
3695 return error;
3696
Al Viro1d2ef592011-09-14 18:55:41 +01003697 dget(new_dentry);
Al Virod83c49f2010-04-30 17:17:09 -04003698 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003699 mutex_lock(&target->i_mutex);
Sage Weil9055cba2011-05-24 13:06:12 -07003700
3701 error = -EBUSY;
3702 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3703 goto out;
3704
Al Viro8de52772012-02-06 12:45:27 -05003705 error = -EMLINK;
3706 if (max_links && !target && new_dir != old_dir &&
3707 new_dir->i_nlink >= max_links)
3708 goto out;
3709
Sage Weil3cebde22011-05-29 21:20:59 -07003710 if (target)
3711 shrink_dcache_parent(new_dentry);
Sage Weil9055cba2011-05-24 13:06:12 -07003712 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3713 if (error)
3714 goto out;
3715
Linus Torvalds1da177e2005-04-16 15:20:36 -07003716 if (target) {
Sage Weil9055cba2011-05-24 13:06:12 -07003717 target->i_flags |= S_DEAD;
3718 dont_mount(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003719 }
Sage Weil9055cba2011-05-24 13:06:12 -07003720out:
3721 if (target)
3722 mutex_unlock(&target->i_mutex);
Al Viro1d2ef592011-09-14 18:55:41 +01003723 dput(new_dentry);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07003724 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07003725 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3726 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003727 return error;
3728}
3729
Adrian Bunk75c96f82005-05-05 16:16:09 -07003730static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3731 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003732{
Sage Weil51892bb2011-05-24 13:06:13 -07003733 struct inode *target = new_dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003734 int error;
3735
3736 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3737 if (error)
3738 return error;
3739
3740 dget(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003741 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003742 mutex_lock(&target->i_mutex);
Sage Weil51892bb2011-05-24 13:06:13 -07003743
3744 error = -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003745 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
Sage Weil51892bb2011-05-24 13:06:13 -07003746 goto out;
3747
3748 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3749 if (error)
3750 goto out;
3751
3752 if (target)
3753 dont_mount(new_dentry);
3754 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3755 d_move(old_dentry, new_dentry);
3756out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003757 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003758 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003759 dput(new_dentry);
3760 return error;
3761}
3762
3763int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3764 struct inode *new_dir, struct dentry *new_dentry)
3765{
3766 int error;
3767 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Eric Paris59b0df22010-02-08 12:53:52 -05003768 const unsigned char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003769
3770 if (old_dentry->d_inode == new_dentry->d_inode)
3771 return 0;
3772
3773 error = may_delete(old_dir, old_dentry, is_dir);
3774 if (error)
3775 return error;
3776
3777 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02003778 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003779 else
3780 error = may_delete(new_dir, new_dentry, is_dir);
3781 if (error)
3782 return error;
3783
Al Viroacfa4382008-12-04 10:06:33 -05003784 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003785 return -EPERM;
3786
Robert Love0eeca282005-07-12 17:06:03 -04003787 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3788
Linus Torvalds1da177e2005-04-16 15:20:36 -07003789 if (is_dir)
3790 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3791 else
3792 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
Al Viro123df292009-12-25 04:57:57 -05003793 if (!error)
3794 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04003795 new_dentry->d_inode, old_dentry);
Robert Love0eeca282005-07-12 17:06:03 -04003796 fsnotify_oldname_free(old_name);
3797
Linus Torvalds1da177e2005-04-16 15:20:36 -07003798 return error;
3799}
3800
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003801SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3802 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003803{
Al Viro2ad94ae2008-07-21 09:32:51 -04003804 struct dentry *old_dir, *new_dir;
3805 struct dentry *old_dentry, *new_dentry;
3806 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003807 struct nameidata oldnd, newnd;
Jeff Layton91a27b22012-10-10 15:25:28 -04003808 struct filename *from;
3809 struct filename *to;
Al Viro2ad94ae2008-07-21 09:32:51 -04003810 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003811
Jeff Layton91a27b22012-10-10 15:25:28 -04003812 from = user_path_parent(olddfd, oldname, &oldnd);
3813 if (IS_ERR(from)) {
3814 error = PTR_ERR(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003815 goto exit;
Jeff Layton91a27b22012-10-10 15:25:28 -04003816 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003817
Jeff Layton91a27b22012-10-10 15:25:28 -04003818 to = user_path_parent(newdfd, newname, &newnd);
3819 if (IS_ERR(to)) {
3820 error = PTR_ERR(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003821 goto exit1;
Jeff Layton91a27b22012-10-10 15:25:28 -04003822 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003823
3824 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08003825 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003826 goto exit2;
3827
Jan Blunck4ac91372008-02-14 19:34:32 -08003828 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003829 error = -EBUSY;
3830 if (oldnd.last_type != LAST_NORM)
3831 goto exit2;
3832
Jan Blunck4ac91372008-02-14 19:34:32 -08003833 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003834 if (newnd.last_type != LAST_NORM)
3835 goto exit2;
3836
Jan Karac30dabf2012-06-12 16:20:30 +02003837 error = mnt_want_write(oldnd.path.mnt);
3838 if (error)
3839 goto exit2;
3840
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003841 oldnd.flags &= ~LOOKUP_PARENT;
3842 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09003843 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003844
Linus Torvalds1da177e2005-04-16 15:20:36 -07003845 trap = lock_rename(new_dir, old_dir);
3846
Christoph Hellwig49705b72005-11-08 21:35:06 -08003847 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003848 error = PTR_ERR(old_dentry);
3849 if (IS_ERR(old_dentry))
3850 goto exit3;
3851 /* source must exist */
3852 error = -ENOENT;
3853 if (!old_dentry->d_inode)
3854 goto exit4;
3855 /* unless the source is a directory trailing slashes give -ENOTDIR */
3856 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3857 error = -ENOTDIR;
3858 if (oldnd.last.name[oldnd.last.len])
3859 goto exit4;
3860 if (newnd.last.name[newnd.last.len])
3861 goto exit4;
3862 }
3863 /* source should not be ancestor of target */
3864 error = -EINVAL;
3865 if (old_dentry == trap)
3866 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08003867 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003868 error = PTR_ERR(new_dentry);
3869 if (IS_ERR(new_dentry))
3870 goto exit4;
3871 /* target should not be an ancestor of source */
3872 error = -ENOTEMPTY;
3873 if (new_dentry == trap)
3874 goto exit5;
3875
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003876 error = security_path_rename(&oldnd.path, old_dentry,
3877 &newnd.path, new_dentry);
3878 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003879 goto exit5;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003880 error = vfs_rename(old_dir->d_inode, old_dentry,
3881 new_dir->d_inode, new_dentry);
3882exit5:
3883 dput(new_dentry);
3884exit4:
3885 dput(old_dentry);
3886exit3:
3887 unlock_rename(new_dir, old_dir);
Jan Karac30dabf2012-06-12 16:20:30 +02003888 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003889exit2:
Jan Blunck1d957f92008-02-14 19:34:35 -08003890 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04003891 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003892exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003893 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003894 putname(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04003895exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003896 return error;
3897}
3898
Heiko Carstensa26eab22009-01-14 14:14:17 +01003899SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003900{
3901 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3902}
3903
Linus Torvalds1da177e2005-04-16 15:20:36 -07003904int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3905{
3906 int len;
3907
3908 len = PTR_ERR(link);
3909 if (IS_ERR(link))
3910 goto out;
3911
3912 len = strlen(link);
3913 if (len > (unsigned) buflen)
3914 len = buflen;
3915 if (copy_to_user(buffer, link, len))
3916 len = -EFAULT;
3917out:
3918 return len;
3919}
3920
3921/*
3922 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3923 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3924 * using) it for any given inode is up to filesystem.
3925 */
3926int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3927{
3928 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003929 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07003930 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003931
Linus Torvalds1da177e2005-04-16 15:20:36 -07003932 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003933 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07003934 if (IS_ERR(cookie))
3935 return PTR_ERR(cookie);
3936
3937 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3938 if (dentry->d_inode->i_op->put_link)
3939 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3940 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003941}
3942
3943int vfs_follow_link(struct nameidata *nd, const char *link)
3944{
3945 return __vfs_follow_link(nd, link);
3946}
3947
3948/* get the link contents into pagecache */
3949static char *page_getlink(struct dentry * dentry, struct page **ppage)
3950{
Duane Griffinebd09ab2008-12-19 20:47:12 +00003951 char *kaddr;
3952 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003953 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07003954 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003955 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07003956 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003957 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00003958 kaddr = kmap(page);
3959 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3960 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003961}
3962
3963int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3964{
3965 struct page *page = NULL;
3966 char *s = page_getlink(dentry, &page);
3967 int res = vfs_readlink(dentry,buffer,buflen,s);
3968 if (page) {
3969 kunmap(page);
3970 page_cache_release(page);
3971 }
3972 return res;
3973}
3974
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003975void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003976{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003977 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003978 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003979 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003980}
3981
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003982void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003983{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003984 struct page *page = cookie;
3985
3986 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003987 kunmap(page);
3988 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003989 }
3990}
3991
Nick Piggin54566b22009-01-04 12:00:53 -08003992/*
3993 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3994 */
3995int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003996{
3997 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003998 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07003999 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08004000 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004001 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08004002 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4003 if (nofs)
4004 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004005
NeilBrown7e53cac2006-03-25 03:07:57 -08004006retry:
Nick Pigginafddba42007-10-16 01:25:01 -07004007 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08004008 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004009 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07004010 goto fail;
4011
Cong Wange8e3c3d2011-11-25 23:14:27 +08004012 kaddr = kmap_atomic(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004013 memcpy(kaddr, symname, len-1);
Cong Wange8e3c3d2011-11-25 23:14:27 +08004014 kunmap_atomic(kaddr);
Nick Pigginafddba42007-10-16 01:25:01 -07004015
4016 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4017 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004018 if (err < 0)
4019 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07004020 if (err < len-1)
4021 goto retry;
4022
Linus Torvalds1da177e2005-04-16 15:20:36 -07004023 mark_inode_dirty(inode);
4024 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004025fail:
4026 return err;
4027}
4028
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004029int page_symlink(struct inode *inode, const char *symname, int len)
4030{
4031 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08004032 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004033}
4034
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08004035const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004036 .readlink = generic_readlink,
4037 .follow_link = page_follow_link_light,
4038 .put_link = page_put_link,
4039};
4040
Al Viro2d8f3032008-07-22 09:59:21 -04004041EXPORT_SYMBOL(user_path_at);
David Howellscc53ce52011-01-14 18:45:26 +00004042EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004043EXPORT_SYMBOL(follow_down);
4044EXPORT_SYMBOL(follow_up);
Al Virof6d2ac52012-08-26 12:55:54 -04004045EXPORT_SYMBOL(get_write_access); /* nfsd */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004046EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004047EXPORT_SYMBOL(lookup_one_len);
4048EXPORT_SYMBOL(page_follow_link_light);
4049EXPORT_SYMBOL(page_put_link);
4050EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004051EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004052EXPORT_SYMBOL(page_symlink);
4053EXPORT_SYMBOL(page_symlink_inode_operations);
Al Virod1811462008-08-02 00:49:18 -04004054EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07004055EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04004056EXPORT_SYMBOL(inode_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004057EXPORT_SYMBOL(unlock_rename);
4058EXPORT_SYMBOL(vfs_create);
4059EXPORT_SYMBOL(vfs_follow_link);
4060EXPORT_SYMBOL(vfs_link);
4061EXPORT_SYMBOL(vfs_mkdir);
4062EXPORT_SYMBOL(vfs_mknod);
4063EXPORT_SYMBOL(generic_permission);
4064EXPORT_SYMBOL(vfs_readlink);
4065EXPORT_SYMBOL(vfs_rename);
4066EXPORT_SYMBOL(vfs_rmdir);
4067EXPORT_SYMBOL(vfs_symlink);
4068EXPORT_SYMBOL(vfs_unlink);
4069EXPORT_SYMBOL(dentry_unhash);
4070EXPORT_SYMBOL(generic_readlink);