blob: 8b61d103a8a7a15b0c5e8eab070e92f940a64b88 [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 */
Al Virodcf787f2013-03-01 23:51:07 -0500454void path_get(const struct path *path)
Jan Blunck5dd784d02008-02-14 19:34:38 -0800455{
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 */
Al Virodcf787f2013-03-01 23:51:07 -0500467void path_put(const 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
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500603 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
Al Viro16c2cd72011-02-22 15:50:10 -0500604 return 0;
605
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500606 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
Jeff Layton39159de2009-12-07 12:01:50 -0500607 if (status > 0)
608 return 0;
609
Al Viro16c2cd72011-02-22 15:50:10 -0500610 if (!status)
Jeff Layton39159de2009-12-07 12:01:50 -0500611 status = -ESTALE;
Al Viro16c2cd72011-02-22 15:50:10 -0500612
Al Viro9f1fafe2011-03-25 11:00:12 -0400613 path_put(&nd->path);
Jeff Layton39159de2009-12-07 12:01:50 -0500614 return status;
615}
616
Al Viro2a737872009-04-07 11:49:53 -0400617static __always_inline void set_root(struct nameidata *nd)
618{
Miklos Szeredif7ad3c62010-08-10 11:41:36 +0200619 if (!nd->root.mnt)
620 get_fs_root(current->fs, &nd->root);
Al Viro2a737872009-04-07 11:49:53 -0400621}
622
Al Viro6de88d72009-08-09 01:41:57 +0400623static int link_path_walk(const char *, struct nameidata *);
624
Nick Piggin31e6b012011-01-07 17:49:52 +1100625static __always_inline void set_root_rcu(struct nameidata *nd)
626{
627 if (!nd->root.mnt) {
628 struct fs_struct *fs = current->fs;
Nick Pigginc28cc362011-01-07 17:49:53 +1100629 unsigned seq;
630
631 do {
632 seq = read_seqcount_begin(&fs->seq);
633 nd->root = fs->root;
Tim Chenc1530012011-04-15 11:39:29 -0700634 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
Nick Pigginc28cc362011-01-07 17:49:53 +1100635 } while (read_seqcount_retry(&fs->seq, seq));
Nick Piggin31e6b012011-01-07 17:49:52 +1100636 }
637}
638
Arjan van de Venf1662352006-01-14 13:21:31 -0800639static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640{
Nick Piggin31e6b012011-01-07 17:49:52 +1100641 int ret;
642
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 if (IS_ERR(link))
644 goto fail;
645
646 if (*link == '/') {
Al Viro2a737872009-04-07 11:49:53 -0400647 set_root(nd);
Jan Blunck1d957f92008-02-14 19:34:35 -0800648 path_put(&nd->path);
Al Viro2a737872009-04-07 11:49:53 -0400649 nd->path = nd->root;
650 path_get(&nd->root);
Al Viro16c2cd72011-02-22 15:50:10 -0500651 nd->flags |= LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100653 nd->inode = nd->path.dentry->d_inode;
Christoph Hellwigb4091d52008-11-05 15:07:21 +0100654
Nick Piggin31e6b012011-01-07 17:49:52 +1100655 ret = link_path_walk(link, nd);
656 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657fail:
Jan Blunck1d957f92008-02-14 19:34:35 -0800658 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 return PTR_ERR(link);
660}
661
Jan Blunck1d957f92008-02-14 19:34:35 -0800662static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700663{
664 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800665 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700666 mntput(path->mnt);
667}
668
Nick Piggin7b9337a2011-01-14 08:42:43 +0000669static inline void path_to_nameidata(const struct path *path,
670 struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700671{
Nick Piggin31e6b012011-01-07 17:49:52 +1100672 if (!(nd->flags & LOOKUP_RCU)) {
673 dput(nd->path.dentry);
674 if (nd->path.mnt != path->mnt)
675 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800676 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100677 nd->path.mnt = path->mnt;
Jan Blunck4ac91372008-02-14 19:34:32 -0800678 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700679}
680
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400681/*
682 * Helper to directly jump to a known parsed path from ->follow_link,
683 * caller must have taken a reference to path beforehand.
684 */
685void nd_jump_link(struct nameidata *nd, struct path *path)
686{
687 path_put(&nd->path);
688
689 nd->path = *path;
690 nd->inode = nd->path.dentry->d_inode;
691 nd->flags |= LOOKUP_JUMPED;
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400692}
693
Al Viro574197e2011-03-14 22:20:34 -0400694static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
695{
696 struct inode *inode = link->dentry->d_inode;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400697 if (inode->i_op->put_link)
Al Viro574197e2011-03-14 22:20:34 -0400698 inode->i_op->put_link(link->dentry, nd, cookie);
699 path_put(link);
700}
701
Linus Torvalds561ec642012-10-26 10:05:07 -0700702int sysctl_protected_symlinks __read_mostly = 0;
703int sysctl_protected_hardlinks __read_mostly = 0;
Kees Cook800179c2012-07-25 17:29:07 -0700704
705/**
706 * may_follow_link - Check symlink following for unsafe situations
707 * @link: The path of the symlink
Randy Dunlap55852632012-08-18 17:39:25 -0700708 * @nd: nameidata pathwalk data
Kees Cook800179c2012-07-25 17:29:07 -0700709 *
710 * In the case of the sysctl_protected_symlinks sysctl being enabled,
711 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
712 * in a sticky world-writable directory. This is to protect privileged
713 * processes from failing races against path names that may change out
714 * from under them by way of other users creating malicious symlinks.
715 * It will permit symlinks to be followed only when outside a sticky
716 * world-writable directory, or when the uid of the symlink and follower
717 * match, or when the directory owner matches the symlink's owner.
718 *
719 * Returns 0 if following the symlink is allowed, -ve on error.
720 */
721static inline int may_follow_link(struct path *link, struct nameidata *nd)
722{
723 const struct inode *inode;
724 const struct inode *parent;
725
726 if (!sysctl_protected_symlinks)
727 return 0;
728
729 /* Allowed if owner and follower match. */
730 inode = link->dentry->d_inode;
Eric W. Biederman81abe272012-08-03 09:38:08 -0700731 if (uid_eq(current_cred()->fsuid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700732 return 0;
733
734 /* Allowed if parent directory not sticky and world-writable. */
735 parent = nd->path.dentry->d_inode;
736 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
737 return 0;
738
739 /* Allowed if parent directory and link owner match. */
Eric W. Biederman81abe272012-08-03 09:38:08 -0700740 if (uid_eq(parent->i_uid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700741 return 0;
742
Sasha Levinffd8d102012-10-04 19:56:40 -0400743 audit_log_link_denied("follow_link", link);
Kees Cook800179c2012-07-25 17:29:07 -0700744 path_put_conditional(link, nd);
745 path_put(&nd->path);
746 return -EACCES;
747}
748
749/**
750 * safe_hardlink_source - Check for safe hardlink conditions
751 * @inode: the source inode to hardlink from
752 *
753 * Return false if at least one of the following conditions:
754 * - inode is not a regular file
755 * - inode is setuid
756 * - inode is setgid and group-exec
757 * - access failure for read and write
758 *
759 * Otherwise returns true.
760 */
761static bool safe_hardlink_source(struct inode *inode)
762{
763 umode_t mode = inode->i_mode;
764
765 /* Special files should not get pinned to the filesystem. */
766 if (!S_ISREG(mode))
767 return false;
768
769 /* Setuid files should not get pinned to the filesystem. */
770 if (mode & S_ISUID)
771 return false;
772
773 /* Executable setgid files should not get pinned to the filesystem. */
774 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
775 return false;
776
777 /* Hardlinking to unreadable or unwritable sources is dangerous. */
778 if (inode_permission(inode, MAY_READ | MAY_WRITE))
779 return false;
780
781 return true;
782}
783
784/**
785 * may_linkat - Check permissions for creating a hardlink
786 * @link: the source to hardlink from
787 *
788 * Block hardlink when all of:
789 * - sysctl_protected_hardlinks enabled
790 * - fsuid does not match inode
791 * - hardlink source is unsafe (see safe_hardlink_source() above)
792 * - not CAP_FOWNER
793 *
794 * Returns 0 if successful, -ve on error.
795 */
796static int may_linkat(struct path *link)
797{
798 const struct cred *cred;
799 struct inode *inode;
800
801 if (!sysctl_protected_hardlinks)
802 return 0;
803
804 cred = current_cred();
805 inode = link->dentry->d_inode;
806
807 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
808 * otherwise, it must be a safe source.
809 */
Eric W. Biederman81abe272012-08-03 09:38:08 -0700810 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
Kees Cook800179c2012-07-25 17:29:07 -0700811 capable(CAP_FOWNER))
812 return 0;
813
Kees Cooka51d9ea2012-07-25 17:29:08 -0700814 audit_log_link_denied("linkat", link);
Kees Cook800179c2012-07-25 17:29:07 -0700815 return -EPERM;
816}
817
Al Virodef4af32009-12-26 08:37:05 -0500818static __always_inline int
Al Viro574197e2011-03-14 22:20:34 -0400819follow_link(struct path *link, struct nameidata *nd, void **p)
Ian Kent051d3812006-03-27 01:14:53 -0800820{
Nick Piggin7b9337a2011-01-14 08:42:43 +0000821 struct dentry *dentry = link->dentry;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400822 int error;
823 char *s;
Ian Kent051d3812006-03-27 01:14:53 -0800824
Al Viro844a3912011-02-15 00:38:26 -0500825 BUG_ON(nd->flags & LOOKUP_RCU);
826
Al Viro0e794582011-03-16 02:45:02 -0400827 if (link->mnt == nd->path.mnt)
828 mntget(link->mnt);
829
Al Viro6d7b5aa2012-06-10 04:15:17 -0400830 error = -ELOOP;
831 if (unlikely(current->total_link_count >= 40))
832 goto out_put_nd_path;
833
Al Viro574197e2011-03-14 22:20:34 -0400834 cond_resched();
835 current->total_link_count++;
836
Al Viro68ac1232012-03-15 08:21:57 -0400837 touch_atime(link);
Ian Kent051d3812006-03-27 01:14:53 -0800838 nd_set_link(nd, NULL);
839
Al Viro36f3b4f2011-02-22 21:24:38 -0500840 error = security_inode_follow_link(link->dentry, nd);
Al Viro6d7b5aa2012-06-10 04:15:17 -0400841 if (error)
842 goto out_put_nd_path;
Al Viro36f3b4f2011-02-22 21:24:38 -0500843
Al Viro86acdca12009-12-22 23:45:11 -0500844 nd->last_type = LAST_BIND;
Al Virodef4af32009-12-26 08:37:05 -0500845 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
846 error = PTR_ERR(*p);
Al Viro6d7b5aa2012-06-10 04:15:17 -0400847 if (IS_ERR(*p))
Christoph Hellwig408ef012012-06-18 10:47:03 -0400848 goto out_put_nd_path;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400849
850 error = 0;
851 s = nd_get_link(nd);
852 if (s) {
853 error = __vfs_follow_link(nd, s);
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400854 if (unlikely(error))
855 put_link(nd, link, *p);
Ian Kent051d3812006-03-27 01:14:53 -0800856 }
Al Viro6d7b5aa2012-06-10 04:15:17 -0400857
858 return error;
859
860out_put_nd_path:
Arnd Bergmann98f6ef62012-10-11 13:20:00 +0000861 *p = NULL;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400862 path_put(&nd->path);
Al Viro6d7b5aa2012-06-10 04:15:17 -0400863 path_put(link);
Ian Kent051d3812006-03-27 01:14:53 -0800864 return error;
865}
866
Nick Piggin31e6b012011-01-07 17:49:52 +1100867static int follow_up_rcu(struct path *path)
868{
Al Viro0714a532011-11-24 22:19:58 -0500869 struct mount *mnt = real_mount(path->mnt);
870 struct mount *parent;
Nick Piggin31e6b012011-01-07 17:49:52 +1100871 struct dentry *mountpoint;
872
Al Viro0714a532011-11-24 22:19:58 -0500873 parent = mnt->mnt_parent;
874 if (&parent->mnt == path->mnt)
Nick Piggin31e6b012011-01-07 17:49:52 +1100875 return 0;
Al Viroa73324d2011-11-24 22:25:07 -0500876 mountpoint = mnt->mnt_mountpoint;
Nick Piggin31e6b012011-01-07 17:49:52 +1100877 path->dentry = mountpoint;
Al Viro0714a532011-11-24 22:19:58 -0500878 path->mnt = &parent->mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +1100879 return 1;
880}
881
David Howellsf015f1262012-06-25 12:55:28 +0100882/*
883 * follow_up - Find the mountpoint of path's vfsmount
884 *
885 * Given a path, find the mountpoint of its source file system.
886 * Replace @path with the path of the mountpoint in the parent mount.
887 * Up is towards /.
888 *
889 * Return 1 if we went up a level and 0 if we were already at the
890 * root.
891 */
Al Virobab77eb2009-04-18 03:26:48 -0400892int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893{
Al Viro0714a532011-11-24 22:19:58 -0500894 struct mount *mnt = real_mount(path->mnt);
895 struct mount *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +1000897
Andi Kleen962830d2012-05-08 13:32:02 +0930898 br_read_lock(&vfsmount_lock);
Al Viro0714a532011-11-24 22:19:58 -0500899 parent = mnt->mnt_parent;
Al Viro3c0a6162012-07-18 17:32:50 +0400900 if (parent == mnt) {
Andi Kleen962830d2012-05-08 13:32:02 +0930901 br_read_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 return 0;
903 }
Al Viro0714a532011-11-24 22:19:58 -0500904 mntget(&parent->mnt);
Al Viroa73324d2011-11-24 22:25:07 -0500905 mountpoint = dget(mnt->mnt_mountpoint);
Andi Kleen962830d2012-05-08 13:32:02 +0930906 br_read_unlock(&vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400907 dput(path->dentry);
908 path->dentry = mountpoint;
909 mntput(path->mnt);
Al Viro0714a532011-11-24 22:19:58 -0500910 path->mnt = &parent->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 return 1;
912}
913
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100914/*
David Howells9875cf82011-01-14 18:45:21 +0000915 * Perform an automount
916 * - return -EISDIR to tell follow_managed() to stop and return the path we
917 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 */
David Howells9875cf82011-01-14 18:45:21 +0000919static int follow_automount(struct path *path, unsigned flags,
920 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +1100921{
David Howells9875cf82011-01-14 18:45:21 +0000922 struct vfsmount *mnt;
David Howellsea5b7782011-01-14 19:10:03 +0000923 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +1100924
David Howells9875cf82011-01-14 18:45:21 +0000925 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
926 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -0700927
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200928 /* We don't want to mount if someone's just doing a stat -
929 * unless they're stat'ing a directory and appended a '/' to
930 * the name.
931 *
932 * We do, however, want to mount if someone wants to open or
933 * create a file of any type under the mountpoint, wants to
934 * traverse through the mountpoint or wants to open the
935 * mounted directory. Also, autofs may mark negative dentries
936 * as being automount points. These will need the attentions
937 * of the daemon to instantiate them before they can be used.
David Howells9875cf82011-01-14 18:45:21 +0000938 */
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200939 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
Linus Torvaldsd94c1772011-09-26 17:44:55 -0700940 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200941 path->dentry->d_inode)
942 return -EISDIR;
943
David Howells9875cf82011-01-14 18:45:21 +0000944 current->total_link_count++;
945 if (current->total_link_count >= 40)
946 return -ELOOP;
947
948 mnt = path->dentry->d_op->d_automount(path);
949 if (IS_ERR(mnt)) {
950 /*
951 * The filesystem is allowed to return -EISDIR here to indicate
952 * it doesn't want to automount. For instance, autofs would do
953 * this so that its userspace daemon can mount on this dentry.
954 *
955 * However, we can only permit this if it's a terminal point in
956 * the path being looked up; if it wasn't then the remainder of
957 * the path is inaccessible and we should say so.
958 */
Al Viro49084c32011-06-25 21:59:52 -0400959 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
David Howells9875cf82011-01-14 18:45:21 +0000960 return -EREMOTE;
961 return PTR_ERR(mnt);
962 }
David Howellsea5b7782011-01-14 19:10:03 +0000963
David Howells9875cf82011-01-14 18:45:21 +0000964 if (!mnt) /* mount collision */
965 return 0;
966
Al Viro8aef1882011-06-16 15:10:06 +0100967 if (!*need_mntput) {
968 /* lock_mount() may release path->mnt on error */
969 mntget(path->mnt);
970 *need_mntput = true;
971 }
Al Viro19a167a2011-01-17 01:35:23 -0500972 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +0000973
David Howellsea5b7782011-01-14 19:10:03 +0000974 switch (err) {
975 case -EBUSY:
976 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -0500977 return 0;
David Howellsea5b7782011-01-14 19:10:03 +0000978 case 0:
Al Viro8aef1882011-06-16 15:10:06 +0100979 path_put(path);
David Howellsea5b7782011-01-14 19:10:03 +0000980 path->mnt = mnt;
981 path->dentry = dget(mnt->mnt_root);
David Howellsea5b7782011-01-14 19:10:03 +0000982 return 0;
Al Viro19a167a2011-01-17 01:35:23 -0500983 default:
984 return err;
David Howellsea5b7782011-01-14 19:10:03 +0000985 }
Al Viro19a167a2011-01-17 01:35:23 -0500986
David Howells9875cf82011-01-14 18:45:21 +0000987}
988
989/*
990 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +0000991 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +0000992 * - Flagged as mountpoint
993 * - Flagged as automount point
994 *
995 * This may only be called in refwalk mode.
996 *
997 * Serialization is taken care of in namespace.c
998 */
999static int follow_managed(struct path *path, unsigned flags)
1000{
Al Viro8aef1882011-06-16 15:10:06 +01001001 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
David Howells9875cf82011-01-14 18:45:21 +00001002 unsigned managed;
1003 bool need_mntput = false;
Al Viro8aef1882011-06-16 15:10:06 +01001004 int ret = 0;
David Howells9875cf82011-01-14 18:45:21 +00001005
1006 /* Given that we're not holding a lock here, we retain the value in a
1007 * local variable for each dentry as we look at it so that we don't see
1008 * the components of that value change under us */
1009 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1010 managed &= DCACHE_MANAGED_DENTRY,
1011 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +00001012 /* Allow the filesystem to manage the transit without i_mutex
1013 * being held. */
1014 if (managed & DCACHE_MANAGE_TRANSIT) {
1015 BUG_ON(!path->dentry->d_op);
1016 BUG_ON(!path->dentry->d_op->d_manage);
Al Viro1aed3e42011-03-18 09:09:02 -04001017 ret = path->dentry->d_op->d_manage(path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001018 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001019 break;
David Howellscc53ce52011-01-14 18:45:26 +00001020 }
1021
David Howells9875cf82011-01-14 18:45:21 +00001022 /* Transit to a mounted filesystem. */
1023 if (managed & DCACHE_MOUNTED) {
1024 struct vfsmount *mounted = lookup_mnt(path);
1025 if (mounted) {
1026 dput(path->dentry);
1027 if (need_mntput)
1028 mntput(path->mnt);
1029 path->mnt = mounted;
1030 path->dentry = dget(mounted->mnt_root);
1031 need_mntput = true;
1032 continue;
1033 }
1034
1035 /* Something is mounted on this dentry in another
1036 * namespace and/or whatever was mounted there in this
1037 * namespace got unmounted before we managed to get the
1038 * vfsmount_lock */
1039 }
1040
1041 /* Handle an automount point */
1042 if (managed & DCACHE_NEED_AUTOMOUNT) {
1043 ret = follow_automount(path, flags, &need_mntput);
1044 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001045 break;
David Howells9875cf82011-01-14 18:45:21 +00001046 continue;
1047 }
1048
1049 /* We didn't change the current path point */
1050 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 }
Al Viro8aef1882011-06-16 15:10:06 +01001052
1053 if (need_mntput && path->mnt == mnt)
1054 mntput(path->mnt);
1055 if (ret == -EISDIR)
1056 ret = 0;
Al Viroa3fbbde2011-11-07 21:21:26 +00001057 return ret < 0 ? ret : need_mntput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058}
1059
David Howellscc53ce52011-01-14 18:45:26 +00001060int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061{
1062 struct vfsmount *mounted;
1063
Al Viro1c755af2009-04-18 14:06:57 -04001064 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -04001066 dput(path->dentry);
1067 mntput(path->mnt);
1068 path->mnt = mounted;
1069 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 return 1;
1071 }
1072 return 0;
1073}
1074
Ian Kent62a73752011-03-25 01:51:02 +08001075static inline bool managed_dentry_might_block(struct dentry *dentry)
1076{
1077 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
1078 dentry->d_op->d_manage(dentry, true) < 0);
1079}
1080
David Howells9875cf82011-01-14 18:45:21 +00001081/*
Al Viro287548e2011-05-27 06:50:06 -04001082 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1083 * we meet a managed dentry that would need blocking.
David Howells9875cf82011-01-14 18:45:21 +00001084 */
1085static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
Al Viro287548e2011-05-27 06:50:06 -04001086 struct inode **inode)
David Howells9875cf82011-01-14 18:45:21 +00001087{
Ian Kent62a73752011-03-25 01:51:02 +08001088 for (;;) {
Al Viroc7105362011-11-24 18:22:03 -05001089 struct mount *mounted;
Ian Kent62a73752011-03-25 01:51:02 +08001090 /*
1091 * Don't forget we might have a non-mountpoint managed dentry
1092 * that wants to block transit.
1093 */
Al Viro287548e2011-05-27 06:50:06 -04001094 if (unlikely(managed_dentry_might_block(path->dentry)))
David Howellsab909112011-01-14 18:46:51 +00001095 return false;
Ian Kent62a73752011-03-25 01:51:02 +08001096
1097 if (!d_mountpoint(path->dentry))
1098 break;
1099
David Howells9875cf82011-01-14 18:45:21 +00001100 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
1101 if (!mounted)
1102 break;
Al Viroc7105362011-11-24 18:22:03 -05001103 path->mnt = &mounted->mnt;
1104 path->dentry = mounted->mnt.mnt_root;
Al Viroa3fbbde2011-11-07 21:21:26 +00001105 nd->flags |= LOOKUP_JUMPED;
David Howells9875cf82011-01-14 18:45:21 +00001106 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
Linus Torvalds59430262011-07-18 15:43:29 -07001107 /*
1108 * Update the inode too. We don't need to re-check the
1109 * dentry sequence number here after this d_inode read,
1110 * because a mount-point is always pinned.
1111 */
1112 *inode = path->dentry->d_inode;
David Howells9875cf82011-01-14 18:45:21 +00001113 }
David Howells9875cf82011-01-14 18:45:21 +00001114 return true;
1115}
1116
Al Virodea39372011-05-27 06:53:39 -04001117static void follow_mount_rcu(struct nameidata *nd)
Al Viro287548e2011-05-27 06:50:06 -04001118{
Al Virodea39372011-05-27 06:53:39 -04001119 while (d_mountpoint(nd->path.dentry)) {
Al Viroc7105362011-11-24 18:22:03 -05001120 struct mount *mounted;
Al Virodea39372011-05-27 06:53:39 -04001121 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
Al Viro287548e2011-05-27 06:50:06 -04001122 if (!mounted)
1123 break;
Al Viroc7105362011-11-24 18:22:03 -05001124 nd->path.mnt = &mounted->mnt;
1125 nd->path.dentry = mounted->mnt.mnt_root;
Al Virodea39372011-05-27 06:53:39 -04001126 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro287548e2011-05-27 06:50:06 -04001127 }
1128}
1129
Nick Piggin31e6b012011-01-07 17:49:52 +11001130static int follow_dotdot_rcu(struct nameidata *nd)
1131{
Nick Piggin31e6b012011-01-07 17:49:52 +11001132 set_root_rcu(nd);
1133
David Howells9875cf82011-01-14 18:45:21 +00001134 while (1) {
Nick Piggin31e6b012011-01-07 17:49:52 +11001135 if (nd->path.dentry == nd->root.dentry &&
1136 nd->path.mnt == nd->root.mnt) {
1137 break;
1138 }
1139 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1140 struct dentry *old = nd->path.dentry;
1141 struct dentry *parent = old->d_parent;
1142 unsigned seq;
1143
1144 seq = read_seqcount_begin(&parent->d_seq);
1145 if (read_seqcount_retry(&old->d_seq, nd->seq))
Al Viroef7562d2011-03-04 14:35:59 -05001146 goto failed;
Nick Piggin31e6b012011-01-07 17:49:52 +11001147 nd->path.dentry = parent;
1148 nd->seq = seq;
1149 break;
1150 }
1151 if (!follow_up_rcu(&nd->path))
1152 break;
1153 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11001154 }
Al Virodea39372011-05-27 06:53:39 -04001155 follow_mount_rcu(nd);
1156 nd->inode = nd->path.dentry->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001157 return 0;
Al Viroef7562d2011-03-04 14:35:59 -05001158
1159failed:
1160 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001161 if (!(nd->flags & LOOKUP_ROOT))
1162 nd->root.mnt = NULL;
Al Viro32a79912012-07-18 20:43:19 +04001163 unlock_rcu_walk();
Al Viroef7562d2011-03-04 14:35:59 -05001164 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001165}
1166
David Howells9875cf82011-01-14 18:45:21 +00001167/*
David Howellscc53ce52011-01-14 18:45:26 +00001168 * Follow down to the covering mount currently visible to userspace. At each
1169 * point, the filesystem owning that dentry may be queried as to whether the
1170 * caller is permitted to proceed or not.
David Howellscc53ce52011-01-14 18:45:26 +00001171 */
Al Viro7cc90cc2011-03-18 09:04:20 -04001172int follow_down(struct path *path)
David Howellscc53ce52011-01-14 18:45:26 +00001173{
1174 unsigned managed;
1175 int ret;
1176
1177 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1178 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1179 /* Allow the filesystem to manage the transit without i_mutex
1180 * being held.
1181 *
1182 * We indicate to the filesystem if someone is trying to mount
1183 * something here. This gives autofs the chance to deny anyone
1184 * other than its daemon the right to mount on its
1185 * superstructure.
1186 *
1187 * The filesystem may sleep at this point.
1188 */
1189 if (managed & DCACHE_MANAGE_TRANSIT) {
1190 BUG_ON(!path->dentry->d_op);
1191 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +00001192 ret = path->dentry->d_op->d_manage(
Al Viro1aed3e42011-03-18 09:09:02 -04001193 path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001194 if (ret < 0)
1195 return ret == -EISDIR ? 0 : ret;
1196 }
1197
1198 /* Transit to a mounted filesystem. */
1199 if (managed & DCACHE_MOUNTED) {
1200 struct vfsmount *mounted = lookup_mnt(path);
1201 if (!mounted)
1202 break;
1203 dput(path->dentry);
1204 mntput(path->mnt);
1205 path->mnt = mounted;
1206 path->dentry = dget(mounted->mnt_root);
1207 continue;
1208 }
1209
1210 /* Don't handle automount points here */
1211 break;
1212 }
1213 return 0;
1214}
1215
1216/*
David Howells9875cf82011-01-14 18:45:21 +00001217 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1218 */
1219static void follow_mount(struct path *path)
1220{
1221 while (d_mountpoint(path->dentry)) {
1222 struct vfsmount *mounted = lookup_mnt(path);
1223 if (!mounted)
1224 break;
1225 dput(path->dentry);
1226 mntput(path->mnt);
1227 path->mnt = mounted;
1228 path->dentry = dget(mounted->mnt_root);
1229 }
1230}
1231
Nick Piggin31e6b012011-01-07 17:49:52 +11001232static void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233{
Al Viro2a737872009-04-07 11:49:53 -04001234 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001235
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 while(1) {
Jan Blunck4ac91372008-02-14 19:34:32 -08001237 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238
Al Viro2a737872009-04-07 11:49:53 -04001239 if (nd->path.dentry == nd->root.dentry &&
1240 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 break;
1242 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001243 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Al Viro3088dd72010-01-30 15:47:29 -05001244 /* rare case of legitimate dget_parent()... */
1245 nd->path.dentry = dget_parent(nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 dput(old);
1247 break;
1248 }
Al Viro3088dd72010-01-30 15:47:29 -05001249 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 }
Al Viro79ed0222009-04-18 13:59:41 -04001252 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001253 nd->inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254}
1255
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256/*
Miklos Szeredibad61182012-03-26 12:54:24 +02001257 * This looks up the name in dcache, possibly revalidates the old dentry and
1258 * allocates a new one if not found or not valid. In the need_lookup argument
1259 * returns whether i_op->lookup is necessary.
1260 *
1261 * dir->d_inode->i_mutex must be held
Nick Pigginbaa03892010-08-18 04:37:31 +10001262 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001263static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
Al Viro201f9562012-06-22 12:42:10 +04001264 unsigned int flags, bool *need_lookup)
Nick Pigginbaa03892010-08-18 04:37:31 +10001265{
Nick Pigginbaa03892010-08-18 04:37:31 +10001266 struct dentry *dentry;
Miklos Szeredibad61182012-03-26 12:54:24 +02001267 int error;
Nick Pigginbaa03892010-08-18 04:37:31 +10001268
Miklos Szeredibad61182012-03-26 12:54:24 +02001269 *need_lookup = false;
1270 dentry = d_lookup(dir, name);
1271 if (dentry) {
Jeff Layton39e3c952012-11-28 11:30:53 -05001272 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
Al Viro201f9562012-06-22 12:42:10 +04001273 error = d_revalidate(dentry, flags);
Miklos Szeredibad61182012-03-26 12:54:24 +02001274 if (unlikely(error <= 0)) {
1275 if (error < 0) {
1276 dput(dentry);
1277 return ERR_PTR(error);
1278 } else if (!d_invalidate(dentry)) {
1279 dput(dentry);
1280 dentry = NULL;
1281 }
1282 }
1283 }
1284 }
Nick Pigginbaa03892010-08-18 04:37:31 +10001285
Miklos Szeredibad61182012-03-26 12:54:24 +02001286 if (!dentry) {
1287 dentry = d_alloc(dir, name);
1288 if (unlikely(!dentry))
1289 return ERR_PTR(-ENOMEM);
Nick Pigginbaa03892010-08-18 04:37:31 +10001290
Miklos Szeredibad61182012-03-26 12:54:24 +02001291 *need_lookup = true;
Nick Pigginbaa03892010-08-18 04:37:31 +10001292 }
1293 return dentry;
1294}
1295
1296/*
Miklos Szeredibad61182012-03-26 12:54:24 +02001297 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1298 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1299 *
1300 * dir->d_inode->i_mutex must be held
Josef Bacik44396f42011-05-31 11:58:49 -04001301 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001302static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
Al Viro72bd8662012-06-10 17:17:17 -04001303 unsigned int flags)
Josef Bacik44396f42011-05-31 11:58:49 -04001304{
Josef Bacik44396f42011-05-31 11:58:49 -04001305 struct dentry *old;
1306
1307 /* Don't create child dentry for a dead directory. */
Miklos Szeredibad61182012-03-26 12:54:24 +02001308 if (unlikely(IS_DEADDIR(dir))) {
Miklos Szeredie188dc02012-02-03 14:25:18 +01001309 dput(dentry);
Josef Bacik44396f42011-05-31 11:58:49 -04001310 return ERR_PTR(-ENOENT);
Miklos Szeredie188dc02012-02-03 14:25:18 +01001311 }
Josef Bacik44396f42011-05-31 11:58:49 -04001312
Al Viro72bd8662012-06-10 17:17:17 -04001313 old = dir->i_op->lookup(dir, dentry, flags);
Josef Bacik44396f42011-05-31 11:58:49 -04001314 if (unlikely(old)) {
1315 dput(dentry);
1316 dentry = old;
1317 }
1318 return dentry;
1319}
1320
Al Viroa3255542012-03-30 14:41:51 -04001321static struct dentry *__lookup_hash(struct qstr *name,
Al Viro72bd8662012-06-10 17:17:17 -04001322 struct dentry *base, unsigned int flags)
Al Viroa3255542012-03-30 14:41:51 -04001323{
Miklos Szeredibad61182012-03-26 12:54:24 +02001324 bool need_lookup;
Al Viroa3255542012-03-30 14:41:51 -04001325 struct dentry *dentry;
1326
Al Viro72bd8662012-06-10 17:17:17 -04001327 dentry = lookup_dcache(name, base, flags, &need_lookup);
Miklos Szeredibad61182012-03-26 12:54:24 +02001328 if (!need_lookup)
1329 return dentry;
Al Viroa3255542012-03-30 14:41:51 -04001330
Al Viro72bd8662012-06-10 17:17:17 -04001331 return lookup_real(base->d_inode, dentry, flags);
Al Viroa3255542012-03-30 14:41:51 -04001332}
1333
Josef Bacik44396f42011-05-31 11:58:49 -04001334/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 * It's more convoluted than I'd like it to be, but... it's still fairly
1336 * small and for now I'd prefer to have fast path as straight as possible.
1337 * It _is_ time-critical.
1338 */
Al Viroe97cdc82013-01-24 18:16:00 -05001339static int lookup_fast(struct nameidata *nd,
Miklos Szeredi697f5142012-05-21 17:30:05 +02001340 struct path *path, struct inode **inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341{
Jan Blunck4ac91372008-02-14 19:34:32 -08001342 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001343 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001344 int need_reval = 1;
1345 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001346 int err;
1347
Al Viro3cac2602009-08-13 18:27:43 +04001348 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001349 * Rename seqlock is not required here because in the off chance
1350 * of a false negative due to a concurrent rename, we're going to
1351 * do the non-racy lookup, below.
1352 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001353 if (nd->flags & LOOKUP_RCU) {
1354 unsigned seq;
Linus Torvaldsda53be12013-05-21 15:22:44 -07001355 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
Al Viro5a18fff2011-03-11 04:44:53 -05001356 if (!dentry)
1357 goto unlazy;
1358
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001359 /*
1360 * This sequence count validates that the inode matches
1361 * the dentry name information from lookup.
1362 */
1363 *inode = dentry->d_inode;
1364 if (read_seqcount_retry(&dentry->d_seq, seq))
1365 return -ECHILD;
1366
1367 /*
1368 * This sequence count validates that the parent had no
1369 * changes while we did the lookup of the dentry above.
1370 *
1371 * The memory barrier in read_seqcount_begin of child is
1372 * enough, we can use __read_seqcount_retry here.
1373 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001374 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1375 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001376 nd->seq = seq;
Al Viro5a18fff2011-03-11 04:44:53 -05001377
Al Viro24643082011-02-15 01:26:22 -05001378 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
Al Viro4ce16ef32012-06-10 16:10:59 -04001379 status = d_revalidate(dentry, nd->flags);
Al Viro5a18fff2011-03-11 04:44:53 -05001380 if (unlikely(status <= 0)) {
1381 if (status != -ECHILD)
1382 need_reval = 0;
1383 goto unlazy;
1384 }
Al Viro24643082011-02-15 01:26:22 -05001385 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001386 path->mnt = mnt;
1387 path->dentry = dentry;
Al Virod6e9bd22011-05-27 07:03:15 -04001388 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1389 goto unlazy;
1390 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1391 goto unlazy;
1392 return 0;
Al Viro5a18fff2011-03-11 04:44:53 -05001393unlazy:
Al Viro19660af2011-03-25 10:32:48 -04001394 if (unlazy_walk(nd, dentry))
1395 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001396 } else {
Al Viroe97cdc82013-01-24 18:16:00 -05001397 dentry = __d_lookup(parent, &nd->last);
Nick Piggin31e6b012011-01-07 17:49:52 +11001398 }
Al Viro5a18fff2011-03-11 04:44:53 -05001399
Al Viro81e6f522012-03-30 14:48:04 -04001400 if (unlikely(!dentry))
1401 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001402
Al Viro5a18fff2011-03-11 04:44:53 -05001403 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
Al Viro4ce16ef32012-06-10 16:10:59 -04001404 status = d_revalidate(dentry, nd->flags);
Al Viro5a18fff2011-03-11 04:44:53 -05001405 if (unlikely(status <= 0)) {
1406 if (status < 0) {
1407 dput(dentry);
1408 return status;
1409 }
1410 if (!d_invalidate(dentry)) {
1411 dput(dentry);
Al Viro81e6f522012-03-30 14:48:04 -04001412 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001413 }
1414 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001415
David Howells9875cf82011-01-14 18:45:21 +00001416 path->mnt = mnt;
1417 path->dentry = dentry;
1418 err = follow_managed(path, nd->flags);
Ian Kent89312212011-01-18 12:06:10 +08001419 if (unlikely(err < 0)) {
1420 path_put_conditional(path, nd);
David Howells9875cf82011-01-14 18:45:21 +00001421 return err;
Ian Kent89312212011-01-18 12:06:10 +08001422 }
Al Viroa3fbbde2011-11-07 21:21:26 +00001423 if (err)
1424 nd->flags |= LOOKUP_JUMPED;
David Howells9875cf82011-01-14 18:45:21 +00001425 *inode = path->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 return 0;
Al Viro81e6f522012-03-30 14:48:04 -04001427
1428need_lookup:
Miklos Szeredi697f5142012-05-21 17:30:05 +02001429 return 1;
1430}
1431
1432/* Fast lookup failed, do it the slow way */
Al Virocc2a5272013-01-24 18:19:49 -05001433static int lookup_slow(struct nameidata *nd, struct path *path)
Miklos Szeredi697f5142012-05-21 17:30:05 +02001434{
1435 struct dentry *dentry, *parent;
1436 int err;
1437
1438 parent = nd->path.dentry;
Al Viro81e6f522012-03-30 14:48:04 -04001439 BUG_ON(nd->inode != parent->d_inode);
1440
1441 mutex_lock(&parent->d_inode->i_mutex);
Al Virocc2a5272013-01-24 18:19:49 -05001442 dentry = __lookup_hash(&nd->last, parent, nd->flags);
Al Viro81e6f522012-03-30 14:48:04 -04001443 mutex_unlock(&parent->d_inode->i_mutex);
1444 if (IS_ERR(dentry))
1445 return PTR_ERR(dentry);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001446 path->mnt = nd->path.mnt;
1447 path->dentry = dentry;
1448 err = follow_managed(path, nd->flags);
1449 if (unlikely(err < 0)) {
1450 path_put_conditional(path, nd);
1451 return err;
1452 }
1453 if (err)
1454 nd->flags |= LOOKUP_JUMPED;
1455 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456}
1457
Al Viro52094c82011-02-21 21:34:47 -05001458static inline int may_lookup(struct nameidata *nd)
1459{
1460 if (nd->flags & LOOKUP_RCU) {
Al Viro4ad5abb2011-06-20 19:57:03 -04001461 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
Al Viro52094c82011-02-21 21:34:47 -05001462 if (err != -ECHILD)
1463 return err;
Al Viro19660af2011-03-25 10:32:48 -04001464 if (unlazy_walk(nd, NULL))
Al Viro52094c82011-02-21 21:34:47 -05001465 return -ECHILD;
1466 }
Al Viro4ad5abb2011-06-20 19:57:03 -04001467 return inode_permission(nd->inode, MAY_EXEC);
Al Viro52094c82011-02-21 21:34:47 -05001468}
1469
Al Viro9856fa12011-03-04 14:22:06 -05001470static inline int handle_dots(struct nameidata *nd, int type)
1471{
1472 if (type == LAST_DOTDOT) {
1473 if (nd->flags & LOOKUP_RCU) {
1474 if (follow_dotdot_rcu(nd))
1475 return -ECHILD;
1476 } else
1477 follow_dotdot(nd);
1478 }
1479 return 0;
1480}
1481
Al Viro951361f2011-03-04 14:44:37 -05001482static void terminate_walk(struct nameidata *nd)
1483{
1484 if (!(nd->flags & LOOKUP_RCU)) {
1485 path_put(&nd->path);
1486 } else {
1487 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001488 if (!(nd->flags & LOOKUP_ROOT))
1489 nd->root.mnt = NULL;
Al Viro32a79912012-07-18 20:43:19 +04001490 unlock_rcu_walk();
Al Viro951361f2011-03-04 14:44:37 -05001491 }
1492}
1493
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001494/*
1495 * Do we need to follow links? We _really_ want to be able
1496 * to do this check without having to look at inode->i_op,
1497 * so we keep a cache of "no, this doesn't need follow_link"
1498 * for the common case.
1499 */
Linus Torvalds7813b942011-08-07 09:53:20 -07001500static inline int should_follow_link(struct inode *inode, int follow)
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001501{
1502 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1503 if (likely(inode->i_op->follow_link))
1504 return follow;
1505
1506 /* This gets set once for the inode lifetime */
1507 spin_lock(&inode->i_lock);
1508 inode->i_opflags |= IOP_NOFOLLOW;
1509 spin_unlock(&inode->i_lock);
1510 }
1511 return 0;
1512}
1513
Al Viroce57dfc2011-03-13 19:58:58 -04001514static inline int walk_component(struct nameidata *nd, struct path *path,
Al Viro21b9b072013-01-24 18:10:25 -05001515 int follow)
Al Viroce57dfc2011-03-13 19:58:58 -04001516{
1517 struct inode *inode;
1518 int err;
1519 /*
1520 * "." and ".." are special - ".." especially so because it has
1521 * to be able to know about the current root directory and
1522 * parent relationships.
1523 */
Al Viro21b9b072013-01-24 18:10:25 -05001524 if (unlikely(nd->last_type != LAST_NORM))
1525 return handle_dots(nd, nd->last_type);
Al Viroe97cdc82013-01-24 18:16:00 -05001526 err = lookup_fast(nd, path, &inode);
Al Viroce57dfc2011-03-13 19:58:58 -04001527 if (unlikely(err)) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001528 if (err < 0)
1529 goto out_err;
1530
Al Virocc2a5272013-01-24 18:19:49 -05001531 err = lookup_slow(nd, path);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001532 if (err < 0)
1533 goto out_err;
1534
1535 inode = path->dentry->d_inode;
Al Viroce57dfc2011-03-13 19:58:58 -04001536 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001537 err = -ENOENT;
1538 if (!inode)
1539 goto out_path_put;
1540
Linus Torvalds7813b942011-08-07 09:53:20 -07001541 if (should_follow_link(inode, follow)) {
Al Viro19660af2011-03-25 10:32:48 -04001542 if (nd->flags & LOOKUP_RCU) {
1543 if (unlikely(unlazy_walk(nd, path->dentry))) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001544 err = -ECHILD;
1545 goto out_err;
Al Viro19660af2011-03-25 10:32:48 -04001546 }
1547 }
Al Viroce57dfc2011-03-13 19:58:58 -04001548 BUG_ON(inode != path->dentry->d_inode);
1549 return 1;
1550 }
1551 path_to_nameidata(path, nd);
1552 nd->inode = inode;
1553 return 0;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001554
1555out_path_put:
1556 path_to_nameidata(path, nd);
1557out_err:
1558 terminate_walk(nd);
1559 return err;
Al Viroce57dfc2011-03-13 19:58:58 -04001560}
1561
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562/*
Al Virob3563792011-03-14 21:54:55 -04001563 * This limits recursive symlink follows to 8, while
1564 * limiting consecutive symlinks to 40.
1565 *
1566 * Without that kind of total limit, nasty chains of consecutive
1567 * symlinks can cause almost arbitrarily long lookups.
1568 */
1569static inline int nested_symlink(struct path *path, struct nameidata *nd)
1570{
1571 int res;
1572
Al Virob3563792011-03-14 21:54:55 -04001573 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1574 path_put_conditional(path, nd);
1575 path_put(&nd->path);
1576 return -ELOOP;
1577 }
Erez Zadok1a4022f2011-05-21 01:19:59 -04001578 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
Al Virob3563792011-03-14 21:54:55 -04001579
1580 nd->depth++;
1581 current->link_count++;
1582
1583 do {
1584 struct path link = *path;
1585 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04001586
1587 res = follow_link(&link, nd, &cookie);
Al Viro6d7b5aa2012-06-10 04:15:17 -04001588 if (res)
1589 break;
Al Viro21b9b072013-01-24 18:10:25 -05001590 res = walk_component(nd, path, LOOKUP_FOLLOW);
Al Viro574197e2011-03-14 22:20:34 -04001591 put_link(nd, &link, cookie);
Al Virob3563792011-03-14 21:54:55 -04001592 } while (res > 0);
1593
1594 current->link_count--;
1595 nd->depth--;
1596 return res;
1597}
1598
1599/*
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001600 * We really don't want to look at inode->i_op->lookup
1601 * when we don't have to. So we keep a cache bit in
1602 * the inode ->i_opflags field that says "yes, we can
1603 * do lookup on this inode".
1604 */
1605static inline int can_lookup(struct inode *inode)
1606{
1607 if (likely(inode->i_opflags & IOP_LOOKUP))
1608 return 1;
1609 if (likely(!inode->i_op->lookup))
1610 return 0;
1611
1612 /* We do this once for the lifetime of the inode */
1613 spin_lock(&inode->i_lock);
1614 inode->i_opflags |= IOP_LOOKUP;
1615 spin_unlock(&inode->i_lock);
1616 return 1;
1617}
1618
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001619/*
1620 * We can do the critical dentry name comparison and hashing
1621 * operations one word at a time, but we are limited to:
1622 *
1623 * - Architectures with fast unaligned word accesses. We could
1624 * do a "get_unaligned()" if this helps and is sufficiently
1625 * fast.
1626 *
1627 * - Little-endian machines (so that we can generate the mask
1628 * of low bytes efficiently). Again, we *could* do a byte
1629 * swapping load on big-endian architectures if that is not
1630 * expensive enough to make the optimization worthless.
1631 *
1632 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1633 * do not trap on the (extremely unlikely) case of a page
1634 * crossing operation.
1635 *
1636 * - Furthermore, we need an efficient 64-bit compile for the
1637 * 64-bit case in order to generate the "number of bytes in
1638 * the final mask". Again, that could be replaced with a
1639 * efficient population count instruction or similar.
1640 */
1641#ifdef CONFIG_DCACHE_WORD_ACCESS
1642
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001643#include <asm/word-at-a-time.h>
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001644
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001645#ifdef CONFIG_64BIT
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001646
1647static inline unsigned int fold_hash(unsigned long hash)
1648{
1649 hash += hash >> (8*sizeof(int));
1650 return hash;
1651}
1652
1653#else /* 32-bit case */
1654
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001655#define fold_hash(x) (x)
1656
1657#endif
1658
1659unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1660{
1661 unsigned long a, mask;
1662 unsigned long hash = 0;
1663
1664 for (;;) {
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001665 a = load_unaligned_zeropad(name);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001666 if (len < sizeof(unsigned long))
1667 break;
1668 hash += a;
Al Virof132c5b2012-03-22 21:59:52 +00001669 hash *= 9;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001670 name += sizeof(unsigned long);
1671 len -= sizeof(unsigned long);
1672 if (!len)
1673 goto done;
1674 }
1675 mask = ~(~0ul << len*8);
1676 hash += mask & a;
1677done:
1678 return fold_hash(hash);
1679}
1680EXPORT_SYMBOL(full_name_hash);
1681
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001682/*
1683 * Calculate the length and hash of the path component, and
1684 * return the length of the component;
1685 */
1686static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1687{
Linus Torvalds36126f82012-05-26 10:43:17 -07001688 unsigned long a, b, adata, bdata, mask, hash, len;
1689 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001690
1691 hash = a = 0;
1692 len = -sizeof(unsigned long);
1693 do {
1694 hash = (hash + a) * 9;
1695 len += sizeof(unsigned long);
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001696 a = load_unaligned_zeropad(name+len);
Linus Torvalds36126f82012-05-26 10:43:17 -07001697 b = a ^ REPEAT_BYTE('/');
1698 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001699
Linus Torvalds36126f82012-05-26 10:43:17 -07001700 adata = prep_zero_mask(a, adata, &constants);
1701 bdata = prep_zero_mask(b, bdata, &constants);
1702
1703 mask = create_zero_mask(adata | bdata);
1704
1705 hash += a & zero_bytemask(mask);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001706 *hashp = fold_hash(hash);
1707
Linus Torvalds36126f82012-05-26 10:43:17 -07001708 return len + find_zero(mask);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001709}
1710
1711#else
1712
Linus Torvalds0145acc2012-03-02 14:32:59 -08001713unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1714{
1715 unsigned long hash = init_name_hash();
1716 while (len--)
1717 hash = partial_name_hash(*name++, hash);
1718 return end_name_hash(hash);
1719}
Linus Torvaldsae942ae2012-03-02 19:40:57 -08001720EXPORT_SYMBOL(full_name_hash);
Linus Torvalds0145acc2012-03-02 14:32:59 -08001721
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001722/*
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001723 * We know there's a real path component here of at least
1724 * one character.
1725 */
1726static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1727{
1728 unsigned long hash = init_name_hash();
1729 unsigned long len = 0, c;
1730
1731 c = (unsigned char)*name;
1732 do {
1733 len++;
1734 hash = partial_name_hash(c, hash);
1735 c = (unsigned char)name[len];
1736 } while (c && c != '/');
1737 *hashp = end_name_hash(hash);
1738 return len;
1739}
1740
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001741#endif
1742
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001743/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01001745 * This is the basic name resolution function, turning a pathname into
1746 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01001748 * Returns 0 and nd will have valid dentry and mnt on success.
1749 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 */
Al Viro6de88d72009-08-09 01:41:57 +04001751static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752{
1753 struct path next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755
1756 while (*name=='/')
1757 name++;
1758 if (!*name)
Al Viro086e1832011-02-22 20:56:27 -05001759 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 /* At this point we know we have a real path component. */
1762 for(;;) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 struct qstr this;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001764 long len;
Al Virofe479a52011-02-22 15:10:03 -05001765 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766
Al Viro52094c82011-02-21 21:34:47 -05001767 err = may_lookup(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 if (err)
1769 break;
1770
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001771 len = hash_name(name, &this.hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 this.name = name;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001773 this.len = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774
Al Virofe479a52011-02-22 15:10:03 -05001775 type = LAST_NORM;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001776 if (name[0] == '.') switch (len) {
Al Virofe479a52011-02-22 15:10:03 -05001777 case 2:
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001778 if (name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05001779 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05001780 nd->flags |= LOOKUP_JUMPED;
1781 }
Al Virofe479a52011-02-22 15:10:03 -05001782 break;
1783 case 1:
1784 type = LAST_DOT;
1785 }
Al Viro5a202bc2011-03-08 14:17:44 -05001786 if (likely(type == LAST_NORM)) {
1787 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05001788 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05001789 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07001790 err = parent->d_op->d_hash(parent, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05001791 if (err < 0)
1792 break;
1793 }
1794 }
Al Virofe479a52011-02-22 15:10:03 -05001795
Al Viro5f4a6a62013-01-24 18:04:22 -05001796 nd->last = this;
1797 nd->last_type = type;
1798
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001799 if (!name[len])
Al Viro5f4a6a62013-01-24 18:04:22 -05001800 return 0;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001801 /*
1802 * If it wasn't NUL, we know it was '/'. Skip that
1803 * slash, and continue until no more slashes.
1804 */
1805 do {
1806 len++;
1807 } while (unlikely(name[len] == '/'));
1808 if (!name[len])
Al Viro5f4a6a62013-01-24 18:04:22 -05001809 return 0;
1810
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001811 name += len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812
Al Viro21b9b072013-01-24 18:10:25 -05001813 err = walk_component(nd, &next, LOOKUP_FOLLOW);
Al Viroce57dfc2011-03-13 19:58:58 -04001814 if (err < 0)
1815 return err;
Al Virofe479a52011-02-22 15:10:03 -05001816
Al Viroce57dfc2011-03-13 19:58:58 -04001817 if (err) {
Al Virob3563792011-03-14 21:54:55 -04001818 err = nested_symlink(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 if (err)
Al Viroa7472ba2011-03-04 14:39:30 -05001820 return err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001821 }
Al Viro5f4a6a62013-01-24 18:04:22 -05001822 if (!can_lookup(nd->inode)) {
1823 err = -ENOTDIR;
1824 break;
1825 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 }
Al Viro951361f2011-03-04 14:44:37 -05001827 terminate_walk(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 return err;
1829}
1830
Al Viro70e9b352011-03-05 21:12:22 -05001831static int path_init(int dfd, const char *name, unsigned int flags,
1832 struct nameidata *nd, struct file **fp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001834 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835
1836 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro16c2cd72011-02-22 15:50:10 -05001837 nd->flags = flags | LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 nd->depth = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05001839 if (flags & LOOKUP_ROOT) {
1840 struct inode *inode = nd->root.dentry->d_inode;
Al Viro73d049a2011-03-11 12:08:24 -05001841 if (*name) {
Al Viro741b7c32012-12-20 13:41:28 -05001842 if (!can_lookup(inode))
Al Viro73d049a2011-03-11 12:08:24 -05001843 return -ENOTDIR;
1844 retval = inode_permission(inode, MAY_EXEC);
1845 if (retval)
1846 return retval;
1847 }
Al Viro5b6ca022011-03-09 23:04:47 -05001848 nd->path = nd->root;
1849 nd->inode = inode;
1850 if (flags & LOOKUP_RCU) {
Al Viro32a79912012-07-18 20:43:19 +04001851 lock_rcu_walk();
Al Viro5b6ca022011-03-09 23:04:47 -05001852 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1853 } else {
1854 path_get(&nd->path);
1855 }
1856 return 0;
1857 }
1858
Al Viro2a737872009-04-07 11:49:53 -04001859 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 if (*name=='/') {
Al Viroe41f7d42011-02-22 14:02:58 -05001862 if (flags & LOOKUP_RCU) {
Al Viro32a79912012-07-18 20:43:19 +04001863 lock_rcu_walk();
Al Viroe41f7d42011-02-22 14:02:58 -05001864 set_root_rcu(nd);
1865 } else {
1866 set_root(nd);
1867 path_get(&nd->root);
1868 }
Al Viro2a737872009-04-07 11:49:53 -04001869 nd->path = nd->root;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001870 } else if (dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05001871 if (flags & LOOKUP_RCU) {
1872 struct fs_struct *fs = current->fs;
1873 unsigned seq;
1874
Al Viro32a79912012-07-18 20:43:19 +04001875 lock_rcu_walk();
Al Viroe41f7d42011-02-22 14:02:58 -05001876
1877 do {
1878 seq = read_seqcount_begin(&fs->seq);
1879 nd->path = fs->pwd;
1880 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1881 } while (read_seqcount_retry(&fs->seq, seq));
1882 } else {
1883 get_fs_pwd(current->fs, &nd->path);
1884 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001885 } else {
Jeff Layton582aa642012-12-11 08:56:16 -05001886 /* Caller must check execute permissions on the starting path component */
Al Viro2903ff02012-08-28 12:52:22 -04001887 struct fd f = fdget_raw(dfd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001888 struct dentry *dentry;
1889
Al Viro2903ff02012-08-28 12:52:22 -04001890 if (!f.file)
1891 return -EBADF;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001892
Al Viro2903ff02012-08-28 12:52:22 -04001893 dentry = f.file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001894
Al Virof52e0c12011-03-14 18:56:51 -04001895 if (*name) {
Al Viro741b7c32012-12-20 13:41:28 -05001896 if (!can_lookup(dentry->d_inode)) {
Al Viro2903ff02012-08-28 12:52:22 -04001897 fdput(f);
1898 return -ENOTDIR;
1899 }
Al Virof52e0c12011-03-14 18:56:51 -04001900 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001901
Al Viro2903ff02012-08-28 12:52:22 -04001902 nd->path = f.file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05001903 if (flags & LOOKUP_RCU) {
Al Viro2903ff02012-08-28 12:52:22 -04001904 if (f.need_put)
1905 *fp = f.file;
Al Viroe41f7d42011-02-22 14:02:58 -05001906 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro32a79912012-07-18 20:43:19 +04001907 lock_rcu_walk();
Al Viroe41f7d42011-02-22 14:02:58 -05001908 } else {
Al Viro2903ff02012-08-28 12:52:22 -04001909 path_get(&nd->path);
1910 fdput(f);
Al Viroe41f7d42011-02-22 14:02:58 -05001911 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 }
Al Viroe41f7d42011-02-22 14:02:58 -05001913
Nick Piggin31e6b012011-01-07 17:49:52 +11001914 nd->inode = nd->path.dentry->d_inode;
Al Viro9b4a9b12009-04-07 11:44:16 -04001915 return 0;
Al Viro9b4a9b12009-04-07 11:44:16 -04001916}
1917
Al Virobd92d7f2011-03-14 19:54:59 -04001918static inline int lookup_last(struct nameidata *nd, struct path *path)
1919{
1920 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1921 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1922
1923 nd->flags &= ~LOOKUP_PARENT;
Al Viro21b9b072013-01-24 18:10:25 -05001924 return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW);
Al Virobd92d7f2011-03-14 19:54:59 -04001925}
1926
Al Viro9b4a9b12009-04-07 11:44:16 -04001927/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroee0827c2011-02-21 23:38:09 -05001928static int path_lookupat(int dfd, const char *name,
Al Viro9b4a9b12009-04-07 11:44:16 -04001929 unsigned int flags, struct nameidata *nd)
1930{
Al Viro70e9b352011-03-05 21:12:22 -05001931 struct file *base = NULL;
Al Virobd92d7f2011-03-14 19:54:59 -04001932 struct path path;
1933 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001934
1935 /*
1936 * Path walking is largely split up into 2 different synchronisation
1937 * schemes, rcu-walk and ref-walk (explained in
1938 * Documentation/filesystems/path-lookup.txt). These share much of the
1939 * path walk code, but some things particularly setup, cleanup, and
1940 * following mounts are sufficiently divergent that functions are
1941 * duplicated. Typically there is a function foo(), and its RCU
1942 * analogue, foo_rcu().
1943 *
1944 * -ECHILD is the error number of choice (just to avoid clashes) that
1945 * is returned if some aspect of an rcu-walk fails. Such an error must
1946 * be handled by restarting a traditional ref-walk (which will always
1947 * be able to complete).
1948 */
Al Virobd92d7f2011-03-14 19:54:59 -04001949 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
Al Viroee0827c2011-02-21 23:38:09 -05001950
Al Virobd92d7f2011-03-14 19:54:59 -04001951 if (unlikely(err))
1952 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001953
1954 current->total_link_count = 0;
Al Virobd92d7f2011-03-14 19:54:59 -04001955 err = link_path_walk(name, nd);
1956
1957 if (!err && !(flags & LOOKUP_PARENT)) {
Al Virobd92d7f2011-03-14 19:54:59 -04001958 err = lookup_last(nd, &path);
1959 while (err > 0) {
1960 void *cookie;
1961 struct path link = path;
Kees Cook800179c2012-07-25 17:29:07 -07001962 err = may_follow_link(&link, nd);
1963 if (unlikely(err))
1964 break;
Al Virobd92d7f2011-03-14 19:54:59 -04001965 nd->flags |= LOOKUP_PARENT;
Al Viro574197e2011-03-14 22:20:34 -04001966 err = follow_link(&link, nd, &cookie);
Al Viro6d7b5aa2012-06-10 04:15:17 -04001967 if (err)
1968 break;
1969 err = lookup_last(nd, &path);
Al Viro574197e2011-03-14 22:20:34 -04001970 put_link(nd, &link, cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04001971 }
1972 }
Al Viroee0827c2011-02-21 23:38:09 -05001973
Al Viro9f1fafe2011-03-25 11:00:12 -04001974 if (!err)
1975 err = complete_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04001976
1977 if (!err && nd->flags & LOOKUP_DIRECTORY) {
Al Viro05252902013-06-06 19:33:47 -04001978 if (!can_lookup(nd->inode)) {
Al Virobd92d7f2011-03-14 19:54:59 -04001979 path_put(&nd->path);
Al Virobd23a532011-03-23 09:56:30 -04001980 err = -ENOTDIR;
Al Virobd92d7f2011-03-14 19:54:59 -04001981 }
1982 }
Al Viro16c2cd72011-02-22 15:50:10 -05001983
Al Viro70e9b352011-03-05 21:12:22 -05001984 if (base)
1985 fput(base);
Al Viroee0827c2011-02-21 23:38:09 -05001986
Al Viro5b6ca022011-03-09 23:04:47 -05001987 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
Al Viro2a737872009-04-07 11:49:53 -04001988 path_put(&nd->root);
1989 nd->root.mnt = NULL;
1990 }
Al Virobd92d7f2011-03-14 19:54:59 -04001991 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001992}
Nick Piggin31e6b012011-01-07 17:49:52 +11001993
Jeff Layton873f1ee2012-10-10 15:25:29 -04001994static int filename_lookup(int dfd, struct filename *name,
1995 unsigned int flags, struct nameidata *nd)
1996{
1997 int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd);
1998 if (unlikely(retval == -ECHILD))
1999 retval = path_lookupat(dfd, name->name, flags, nd);
2000 if (unlikely(retval == -ESTALE))
2001 retval = path_lookupat(dfd, name->name,
2002 flags | LOOKUP_REVAL, nd);
2003
2004 if (likely(!retval))
Jeff Laytonadb5c242012-10-10 16:43:13 -04002005 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002006 return retval;
2007}
2008
Al Viroee0827c2011-02-21 23:38:09 -05002009static int do_path_lookup(int dfd, const char *name,
2010 unsigned int flags, struct nameidata *nd)
2011{
Jeff Layton873f1ee2012-10-10 15:25:29 -04002012 struct filename filename = { .name = name };
Nick Piggin31e6b012011-01-07 17:49:52 +11002013
Jeff Layton873f1ee2012-10-10 15:25:29 -04002014 return filename_lookup(dfd, &filename, flags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015}
2016
Al Viro79714f72012-06-15 03:01:42 +04002017/* does lookup, returns the object with parent locked */
2018struct dentry *kern_path_locked(const char *name, struct path *path)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002019{
Al Viro79714f72012-06-15 03:01:42 +04002020 struct nameidata nd;
2021 struct dentry *d;
2022 int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd);
2023 if (err)
2024 return ERR_PTR(err);
2025 if (nd.last_type != LAST_NORM) {
2026 path_put(&nd.path);
2027 return ERR_PTR(-EINVAL);
2028 }
2029 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Al Viro1e0ea002012-07-22 23:46:21 +04002030 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
Al Viro79714f72012-06-15 03:01:42 +04002031 if (IS_ERR(d)) {
2032 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2033 path_put(&nd.path);
2034 return d;
2035 }
2036 *path = nd.path;
2037 return d;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002038}
2039
Al Virod1811462008-08-02 00:49:18 -04002040int kern_path(const char *name, unsigned int flags, struct path *path)
2041{
2042 struct nameidata nd;
2043 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
2044 if (!res)
2045 *path = nd.path;
2046 return res;
2047}
2048
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002049/**
2050 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2051 * @dentry: pointer to dentry of the base directory
2052 * @mnt: pointer to vfs mount of the base directory
2053 * @name: pointer to file name
2054 * @flags: lookup flags
Al Viroe0a01242011-06-27 17:00:37 -04002055 * @path: pointer to struct path to fill
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002056 */
2057int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2058 const char *name, unsigned int flags,
Al Viroe0a01242011-06-27 17:00:37 -04002059 struct path *path)
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002060{
Al Viroe0a01242011-06-27 17:00:37 -04002061 struct nameidata nd;
2062 int err;
2063 nd.root.dentry = dentry;
2064 nd.root.mnt = mnt;
2065 BUG_ON(flags & LOOKUP_PARENT);
Al Viro5b6ca022011-03-09 23:04:47 -05002066 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
Al Viroe0a01242011-06-27 17:00:37 -04002067 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
2068 if (!err)
2069 *path = nd.path;
2070 return err;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002071}
2072
James Morris057f6c02007-04-26 00:12:05 -07002073/*
2074 * Restricted form of lookup. Doesn't follow links, single-component only,
2075 * needs parent already locked. Doesn't follow mounts.
2076 * SMP-safe.
2077 */
Adrian Bunka244e162006-03-31 02:32:11 -08002078static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079{
Al Viro72bd8662012-06-10 17:17:17 -04002080 return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081}
2082
Christoph Hellwigeead1912007-10-16 23:25:38 -07002083/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07002084 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07002085 * @name: pathname component to lookup
2086 * @base: base directory to lookup from
2087 * @len: maximum length @len should be interpreted to
2088 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07002089 * Note that this routine is purely a helper for filesystem usage and should
2090 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07002091 * nameidata argument is passed to the filesystem methods and a filesystem
2092 * using this helper needs to be prepared for that.
2093 */
James Morris057f6c02007-04-26 00:12:05 -07002094struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2095{
James Morris057f6c02007-04-26 00:12:05 -07002096 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05002097 unsigned int c;
Miklos Szeredicda309d2012-03-26 12:54:21 +02002098 int err;
James Morris057f6c02007-04-26 00:12:05 -07002099
David Woodhouse2f9092e2009-04-20 23:18:37 +01002100 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2101
Al Viro6a96ba52011-03-07 23:49:20 -05002102 this.name = name;
2103 this.len = len;
Linus Torvalds0145acc2012-03-02 14:32:59 -08002104 this.hash = full_name_hash(name, len);
Al Viro6a96ba52011-03-07 23:49:20 -05002105 if (!len)
2106 return ERR_PTR(-EACCES);
2107
Al Viro21d8a152012-11-29 22:17:21 -05002108 if (unlikely(name[0] == '.')) {
2109 if (len < 2 || (len == 2 && name[1] == '.'))
2110 return ERR_PTR(-EACCES);
2111 }
2112
Al Viro6a96ba52011-03-07 23:49:20 -05002113 while (len--) {
2114 c = *(const unsigned char *)name++;
2115 if (c == '/' || c == '\0')
2116 return ERR_PTR(-EACCES);
Al Viro6a96ba52011-03-07 23:49:20 -05002117 }
Al Viro5a202bc2011-03-08 14:17:44 -05002118 /*
2119 * See if the low-level filesystem might want
2120 * to use its own hash..
2121 */
2122 if (base->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002123 int err = base->d_op->d_hash(base, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05002124 if (err < 0)
2125 return ERR_PTR(err);
2126 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07002127
Miklos Szeredicda309d2012-03-26 12:54:21 +02002128 err = inode_permission(base->d_inode, MAY_EXEC);
2129 if (err)
2130 return ERR_PTR(err);
2131
Al Viro72bd8662012-06-10 17:17:17 -04002132 return __lookup_hash(&this, base, 0);
James Morris057f6c02007-04-26 00:12:05 -07002133}
2134
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002135int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2136 struct path *path, int *empty)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137{
Al Viro2d8f3032008-07-22 09:59:21 -04002138 struct nameidata nd;
Jeff Layton91a27b22012-10-10 15:25:28 -04002139 struct filename *tmp = getname_flags(name, flags, empty);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04002142
2143 BUG_ON(flags & LOOKUP_PARENT);
2144
Jeff Layton873f1ee2012-10-10 15:25:29 -04002145 err = filename_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04002147 if (!err)
2148 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 }
2150 return err;
2151}
2152
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002153int user_path_at(int dfd, const char __user *name, unsigned flags,
2154 struct path *path)
2155{
Linus Torvaldsf7493e52012-03-22 16:10:40 -07002156 return user_path_at_empty(dfd, name, flags, path, NULL);
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002157}
2158
Jeff Layton873f1ee2012-10-10 15:25:29 -04002159/*
2160 * NB: most callers don't do anything directly with the reference to the
2161 * to struct filename, but the nd->last pointer points into the name string
2162 * allocated by getname. So we must hold the reference to it until all
2163 * path-walking is complete.
2164 */
Jeff Layton91a27b22012-10-10 15:25:28 -04002165static struct filename *
Jeff Layton9e790bd2012-12-11 12:10:09 -05002166user_path_parent(int dfd, const char __user *path, struct nameidata *nd,
2167 unsigned int flags)
Al Viro2ad94ae2008-07-21 09:32:51 -04002168{
Jeff Layton91a27b22012-10-10 15:25:28 -04002169 struct filename *s = getname(path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002170 int error;
2171
Jeff Layton9e790bd2012-12-11 12:10:09 -05002172 /* only LOOKUP_REVAL is allowed in extra flags */
2173 flags &= LOOKUP_REVAL;
2174
Al Viro2ad94ae2008-07-21 09:32:51 -04002175 if (IS_ERR(s))
Jeff Layton91a27b22012-10-10 15:25:28 -04002176 return s;
Al Viro2ad94ae2008-07-21 09:32:51 -04002177
Jeff Layton9e790bd2012-12-11 12:10:09 -05002178 error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd);
Jeff Layton91a27b22012-10-10 15:25:28 -04002179 if (error) {
Al Viro2ad94ae2008-07-21 09:32:51 -04002180 putname(s);
Jeff Layton91a27b22012-10-10 15:25:28 -04002181 return ERR_PTR(error);
2182 }
Al Viro2ad94ae2008-07-21 09:32:51 -04002183
Jeff Layton91a27b22012-10-10 15:25:28 -04002184 return s;
Al Viro2ad94ae2008-07-21 09:32:51 -04002185}
2186
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187/*
2188 * It's inline, so penalty for filesystems that don't use sticky bit is
2189 * minimal.
2190 */
2191static inline int check_sticky(struct inode *dir, struct inode *inode)
2192{
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002193 kuid_t fsuid = current_fsuid();
David Howellsda9592e2008-11-14 10:39:05 +11002194
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195 if (!(dir->i_mode & S_ISVTX))
2196 return 0;
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002197 if (uid_eq(inode->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198 return 0;
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002199 if (uid_eq(dir->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 return 0;
Eric W. Biederman1a48e2a2011-11-14 16:24:06 -08002201 return !inode_capable(inode, CAP_FOWNER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202}
2203
2204/*
2205 * Check whether we can remove a link victim from directory dir, check
2206 * whether the type of victim is right.
2207 * 1. We can't do it if dir is read-only (done in permission())
2208 * 2. We should have write and exec permissions on dir
2209 * 3. We can't remove anything from append-only dir
2210 * 4. We can't do anything with immutable dir (done in permission())
2211 * 5. If the sticky bit on dir is set we should either
2212 * a. be owner of dir, or
2213 * b. be owner of victim, or
2214 * c. have CAP_FOWNER capability
2215 * 6. If the victim is append-only or immutable we can't do antyhing with
2216 * links pointing to it.
2217 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2218 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2219 * 9. We can't remove a root or mountpoint.
2220 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2221 * nfs_async_unlink().
2222 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08002223static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224{
2225 int error;
2226
2227 if (!victim->d_inode)
2228 return -ENOENT;
2229
2230 BUG_ON(victim->d_parent->d_inode != dir);
Jeff Layton4fa6b5e2012-10-10 15:25:25 -04002231 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232
Al Virof419a2e2008-07-22 00:07:17 -04002233 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234 if (error)
2235 return error;
2236 if (IS_APPEND(dir))
2237 return -EPERM;
2238 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
Hugh Dickinsf9454542008-11-19 15:36:38 -08002239 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 return -EPERM;
2241 if (isdir) {
2242 if (!S_ISDIR(victim->d_inode->i_mode))
2243 return -ENOTDIR;
2244 if (IS_ROOT(victim))
2245 return -EBUSY;
2246 } else if (S_ISDIR(victim->d_inode->i_mode))
2247 return -EISDIR;
2248 if (IS_DEADDIR(dir))
2249 return -ENOENT;
2250 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2251 return -EBUSY;
2252 return 0;
2253}
2254
2255/* Check whether we can create an object with dentry child in directory
2256 * dir.
2257 * 1. We can't do it if child already exists (open has special treatment for
2258 * this case, but since we are inlined it's OK)
2259 * 2. We can't do it if dir is read-only (done in permission())
2260 * 3. We should have write and exec permissions on dir
2261 * 4. We can't do it if dir is immutable (done in permission())
2262 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02002263static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264{
2265 if (child->d_inode)
2266 return -EEXIST;
2267 if (IS_DEADDIR(dir))
2268 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04002269 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270}
2271
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272/*
2273 * p1 and p2 should be directories on the same fs.
2274 */
2275struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2276{
2277 struct dentry *p;
2278
2279 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07002280 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 return NULL;
2282 }
2283
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002284 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002286 p = d_ancestor(p2, p1);
2287 if (p) {
2288 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2289 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2290 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291 }
2292
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002293 p = d_ancestor(p1, p2);
2294 if (p) {
2295 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2296 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2297 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298 }
2299
Ingo Molnarf2eace22006-07-03 00:25:05 -07002300 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2301 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302 return NULL;
2303}
2304
2305void unlock_rename(struct dentry *p1, struct dentry *p2)
2306{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002307 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002309 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002310 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 }
2312}
2313
Al Viro4acdaf22011-07-26 01:42:34 -04002314int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viro312b63f2012-06-10 18:09:36 -04002315 bool want_excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002317 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318 if (error)
2319 return error;
2320
Al Viroacfa4382008-12-04 10:06:33 -05002321 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322 return -EACCES; /* shouldn't it be ENOSYS? */
2323 mode &= S_IALLUGO;
2324 mode |= S_IFREG;
2325 error = security_inode_create(dir, dentry, mode);
2326 if (error)
2327 return error;
Al Viro312b63f2012-06-10 18:09:36 -04002328 error = dir->i_op->create(dir, dentry, mode, want_excl);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002329 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002330 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331 return error;
2332}
2333
Al Viro73d049a2011-03-11 12:08:24 -05002334static int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002336 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337 struct inode *inode = dentry->d_inode;
2338 int error;
2339
Al Virobcda7652011-03-13 16:42:14 -04002340 /* O_PATH? */
2341 if (!acc_mode)
2342 return 0;
2343
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344 if (!inode)
2345 return -ENOENT;
2346
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002347 switch (inode->i_mode & S_IFMT) {
2348 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002350 case S_IFDIR:
2351 if (acc_mode & MAY_WRITE)
2352 return -EISDIR;
2353 break;
2354 case S_IFBLK:
2355 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002356 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002358 /*FALLTHRU*/
2359 case S_IFIFO:
2360 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002362 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002363 }
Dave Hansenb41572e2007-10-16 23:31:14 -07002364
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002365 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07002366 if (error)
2367 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05002368
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369 /*
2370 * An append-only file must be opened in append mode for writing.
2371 */
2372 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05002373 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05002374 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05002376 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377 }
2378
2379 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002380 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05002381 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382
J. Bruce Fieldsf3c7691e2011-09-21 10:58:13 -04002383 return 0;
Al Viro7715b522009-12-16 03:54:00 -05002384}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385
Jeff Laytone1181ee2010-12-07 16:19:50 -05002386static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002387{
Jeff Laytone1181ee2010-12-07 16:19:50 -05002388 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002389 struct inode *inode = path->dentry->d_inode;
2390 int error = get_write_access(inode);
2391 if (error)
2392 return error;
2393 /*
2394 * Refuse to truncate files with mandatory locks held on them.
2395 */
2396 error = locks_verify_locked(inode);
2397 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002398 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002399 if (!error) {
2400 error = do_truncate(path->dentry, 0,
2401 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002402 filp);
Al Viro7715b522009-12-16 03:54:00 -05002403 }
2404 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002405 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406}
2407
Dave Hansend57999e2008-02-15 14:37:27 -08002408static inline int open_to_namei_flags(int flag)
2409{
Al Viro8a5e9292011-06-25 19:15:54 -04002410 if ((flag & O_ACCMODE) == 3)
2411 flag--;
Dave Hansend57999e2008-02-15 14:37:27 -08002412 return flag;
2413}
2414
Miklos Szeredid18e9002012-06-05 15:10:17 +02002415static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2416{
2417 int error = security_path_mknod(dir, dentry, mode, 0);
2418 if (error)
2419 return error;
2420
2421 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2422 if (error)
2423 return error;
2424
2425 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2426}
2427
David Howells1acf0af2012-06-14 16:13:46 +01002428/*
2429 * Attempt to atomically look up, create and open a file from a negative
2430 * dentry.
2431 *
2432 * Returns 0 if successful. The file will have been created and attached to
2433 * @file by the filesystem calling finish_open().
2434 *
2435 * Returns 1 if the file was looked up only or didn't need creating. The
2436 * caller will need to perform the open themselves. @path will have been
2437 * updated to point to the new dentry. This may be negative.
2438 *
2439 * Returns an error code otherwise.
2440 */
Al Viro2675a4e2012-06-22 12:41:10 +04002441static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2442 struct path *path, struct file *file,
2443 const struct open_flags *op,
Al Viro64894cf2012-07-31 00:53:35 +04002444 bool got_write, bool need_lookup,
Al Viro2675a4e2012-06-22 12:41:10 +04002445 int *opened)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002446{
2447 struct inode *dir = nd->path.dentry->d_inode;
2448 unsigned open_flag = open_to_namei_flags(op->open_flag);
2449 umode_t mode;
2450 int error;
2451 int acc_mode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002452 int create_error = 0;
2453 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2454
2455 BUG_ON(dentry->d_inode);
2456
2457 /* Don't create child dentry for a dead directory. */
2458 if (unlikely(IS_DEADDIR(dir))) {
Al Viro2675a4e2012-06-22 12:41:10 +04002459 error = -ENOENT;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002460 goto out;
2461 }
2462
Miklos Szeredi62b259d2012-08-15 13:01:24 +02002463 mode = op->mode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002464 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2465 mode &= ~current_umask();
2466
Al Virof8310c52012-07-30 11:50:30 +04002467 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT)) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002468 open_flag &= ~O_TRUNC;
Al Viro47237682012-06-10 05:01:45 -04002469 *opened |= FILE_CREATED;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002470 }
2471
2472 /*
2473 * Checking write permission is tricky, bacuse we don't know if we are
2474 * going to actually need it: O_CREAT opens should work as long as the
2475 * file exists. But checking existence breaks atomicity. The trick is
2476 * to check access and if not granted clear O_CREAT from the flags.
2477 *
2478 * Another problem is returing the "right" error value (e.g. for an
2479 * O_EXCL open we want to return EEXIST not EROFS).
2480 */
Al Viro64894cf2012-07-31 00:53:35 +04002481 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2482 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2483 if (!(open_flag & O_CREAT)) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002484 /*
2485 * No O_CREATE -> atomicity not a requirement -> fall
2486 * back to lookup + open
2487 */
2488 goto no_open;
2489 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2490 /* Fall back and fail with the right error */
Al Viro64894cf2012-07-31 00:53:35 +04002491 create_error = -EROFS;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002492 goto no_open;
2493 } else {
2494 /* No side effects, safe to clear O_CREAT */
Al Viro64894cf2012-07-31 00:53:35 +04002495 create_error = -EROFS;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002496 open_flag &= ~O_CREAT;
2497 }
2498 }
2499
2500 if (open_flag & O_CREAT) {
Miklos Szeredi38227f72012-08-15 13:01:24 +02002501 error = may_o_create(&nd->path, dentry, mode);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002502 if (error) {
2503 create_error = error;
2504 if (open_flag & O_EXCL)
2505 goto no_open;
2506 open_flag &= ~O_CREAT;
2507 }
2508 }
2509
2510 if (nd->flags & LOOKUP_DIRECTORY)
2511 open_flag |= O_DIRECTORY;
2512
Al Viro30d90492012-06-22 12:40:19 +04002513 file->f_path.dentry = DENTRY_NOT_SET;
2514 file->f_path.mnt = nd->path.mnt;
2515 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
Al Viro47237682012-06-10 05:01:45 -04002516 opened);
Al Virod9585272012-06-22 12:39:14 +04002517 if (error < 0) {
Al Virod9585272012-06-22 12:39:14 +04002518 if (create_error && error == -ENOENT)
2519 error = create_error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002520 goto out;
2521 }
2522
2523 acc_mode = op->acc_mode;
Al Viro47237682012-06-10 05:01:45 -04002524 if (*opened & FILE_CREATED) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002525 fsnotify_create(dir, dentry);
2526 acc_mode = MAY_OPEN;
2527 }
2528
Al Virod9585272012-06-22 12:39:14 +04002529 if (error) { /* returned 1, that is */
Al Viro30d90492012-06-22 12:40:19 +04002530 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
Al Viro2675a4e2012-06-22 12:41:10 +04002531 error = -EIO;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002532 goto out;
2533 }
Al Viro30d90492012-06-22 12:40:19 +04002534 if (file->f_path.dentry) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002535 dput(dentry);
Al Viro30d90492012-06-22 12:40:19 +04002536 dentry = file->f_path.dentry;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002537 }
Sage Weil62b2ce92012-08-15 13:30:12 -07002538 if (create_error && dentry->d_inode == NULL) {
2539 error = create_error;
2540 goto out;
2541 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02002542 goto looked_up;
2543 }
2544
2545 /*
2546 * We didn't have the inode before the open, so check open permission
2547 * here.
2548 */
Al Viro2675a4e2012-06-22 12:41:10 +04002549 error = may_open(&file->f_path, acc_mode, open_flag);
2550 if (error)
2551 fput(file);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002552
2553out:
2554 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04002555 return error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002556
Miklos Szeredid18e9002012-06-05 15:10:17 +02002557no_open:
2558 if (need_lookup) {
Al Viro72bd8662012-06-10 17:17:17 -04002559 dentry = lookup_real(dir, dentry, nd->flags);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002560 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002561 return PTR_ERR(dentry);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002562
2563 if (create_error) {
2564 int open_flag = op->open_flag;
2565
Al Viro2675a4e2012-06-22 12:41:10 +04002566 error = create_error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002567 if ((open_flag & O_EXCL)) {
2568 if (!dentry->d_inode)
2569 goto out;
2570 } else if (!dentry->d_inode) {
2571 goto out;
2572 } else if ((open_flag & O_TRUNC) &&
2573 S_ISREG(dentry->d_inode->i_mode)) {
2574 goto out;
2575 }
2576 /* will fail later, go on to get the right error */
2577 }
2578 }
2579looked_up:
2580 path->dentry = dentry;
2581 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04002582 return 1;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002583}
2584
Nick Piggin31e6b012011-01-07 17:49:52 +11002585/*
David Howells1acf0af2012-06-14 16:13:46 +01002586 * Look up and maybe create and open the last component.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002587 *
2588 * Must be called with i_mutex held on parent.
2589 *
David Howells1acf0af2012-06-14 16:13:46 +01002590 * Returns 0 if the file was successfully atomically created (if necessary) and
2591 * opened. In this case the file will be returned attached to @file.
2592 *
2593 * Returns 1 if the file was not completely opened at this time, though lookups
2594 * and creations will have been performed and the dentry returned in @path will
2595 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2596 * specified then a negative dentry may be returned.
2597 *
2598 * An error code is returned otherwise.
2599 *
2600 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2601 * cleared otherwise prior to returning.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002602 */
Al Viro2675a4e2012-06-22 12:41:10 +04002603static int lookup_open(struct nameidata *nd, struct path *path,
2604 struct file *file,
2605 const struct open_flags *op,
Al Viro64894cf2012-07-31 00:53:35 +04002606 bool got_write, int *opened)
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002607{
2608 struct dentry *dir = nd->path.dentry;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002609 struct inode *dir_inode = dir->d_inode;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002610 struct dentry *dentry;
2611 int error;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002612 bool need_lookup;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002613
Al Viro47237682012-06-10 05:01:45 -04002614 *opened &= ~FILE_CREATED;
Al Viro201f9562012-06-22 12:42:10 +04002615 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002616 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002617 return PTR_ERR(dentry);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002618
Miklos Szeredid18e9002012-06-05 15:10:17 +02002619 /* Cached positive dentry: will open in f_op->open */
2620 if (!need_lookup && dentry->d_inode)
2621 goto out_no_open;
2622
2623 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
Al Viro64894cf2012-07-31 00:53:35 +04002624 return atomic_open(nd, dentry, path, file, op, got_write,
Al Viro47237682012-06-10 05:01:45 -04002625 need_lookup, opened);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002626 }
2627
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002628 if (need_lookup) {
2629 BUG_ON(dentry->d_inode);
2630
Al Viro72bd8662012-06-10 17:17:17 -04002631 dentry = lookup_real(dir_inode, dentry, nd->flags);
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002632 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002633 return PTR_ERR(dentry);
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002634 }
2635
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002636 /* Negative dentry, just create the file */
2637 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2638 umode_t mode = op->mode;
2639 if (!IS_POSIXACL(dir->d_inode))
2640 mode &= ~current_umask();
2641 /*
2642 * This write is needed to ensure that a
2643 * rw->ro transition does not occur between
2644 * the time when the file is created and when
2645 * a permanent write count is taken through
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002646 * the 'struct file' in finish_open().
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002647 */
Al Viro64894cf2012-07-31 00:53:35 +04002648 if (!got_write) {
2649 error = -EROFS;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002650 goto out_dput;
Al Viro64894cf2012-07-31 00:53:35 +04002651 }
Al Viro47237682012-06-10 05:01:45 -04002652 *opened |= FILE_CREATED;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002653 error = security_path_mknod(&nd->path, dentry, mode, 0);
2654 if (error)
2655 goto out_dput;
Al Viro312b63f2012-06-10 18:09:36 -04002656 error = vfs_create(dir->d_inode, dentry, mode,
2657 nd->flags & LOOKUP_EXCL);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002658 if (error)
2659 goto out_dput;
2660 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02002661out_no_open:
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002662 path->dentry = dentry;
2663 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04002664 return 1;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002665
2666out_dput:
2667 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04002668 return error;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002669}
2670
2671/*
Al Virofe2d35f2011-03-05 22:58:25 -05002672 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11002673 */
Al Viro2675a4e2012-06-22 12:41:10 +04002674static int do_last(struct nameidata *nd, struct path *path,
2675 struct file *file, const struct open_flags *op,
Jeff Layton669abf42012-10-10 16:43:10 -04002676 int *opened, struct filename *name)
Al Virofb1cc552009-12-24 01:58:28 -05002677{
Al Viroa1e28032009-12-24 02:12:06 -05002678 struct dentry *dir = nd->path.dentry;
Al Viroca344a892011-03-09 00:36:45 -05002679 int open_flag = op->open_flag;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002680 bool will_truncate = (open_flag & O_TRUNC) != 0;
Al Viro64894cf2012-07-31 00:53:35 +04002681 bool got_write = false;
Al Virobcda7652011-03-13 16:42:14 -04002682 int acc_mode = op->acc_mode;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02002683 struct inode *inode;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002684 bool symlink_ok = false;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002685 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2686 bool retried = false;
Al Viro16c2cd72011-02-22 15:50:10 -05002687 int error;
Al Virofb1cc552009-12-24 01:58:28 -05002688
Al Viroc3e380b2011-02-23 13:39:45 -05002689 nd->flags &= ~LOOKUP_PARENT;
2690 nd->flags |= op->intent;
2691
Al Virobc77daa2013-06-06 09:12:33 -04002692 if (nd->last_type != LAST_NORM) {
Al Virofe2d35f2011-03-05 22:58:25 -05002693 error = handle_dots(nd, nd->last_type);
2694 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002695 return error;
Miklos Szeredie83db162012-06-05 15:10:29 +02002696 goto finish_open;
Al Viro1f36f772009-12-26 10:56:19 -05002697 }
Al Viro67ee3ad2009-12-26 07:01:01 -05002698
Al Viroca344a892011-03-09 00:36:45 -05002699 if (!(open_flag & O_CREAT)) {
Al Virofe2d35f2011-03-05 22:58:25 -05002700 if (nd->last.name[nd->last.len])
2701 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
Al Virobcda7652011-03-13 16:42:14 -04002702 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002703 symlink_ok = true;
Al Virofe2d35f2011-03-05 22:58:25 -05002704 /* we _can_ be in RCU mode here */
Al Viroe97cdc82013-01-24 18:16:00 -05002705 error = lookup_fast(nd, path, &inode);
Miklos Szeredi71574862012-06-05 15:10:14 +02002706 if (likely(!error))
2707 goto finish_lookup;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02002708
Miklos Szeredi71574862012-06-05 15:10:14 +02002709 if (error < 0)
Al Viro2675a4e2012-06-22 12:41:10 +04002710 goto out;
Miklos Szeredi37d7fff2012-06-05 15:10:12 +02002711
Miklos Szeredi71574862012-06-05 15:10:14 +02002712 BUG_ON(nd->inode != dir->d_inode);
Miklos Szeredib6183df2012-06-05 15:10:13 +02002713 } else {
2714 /* create side of things */
2715 /*
2716 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2717 * has been cleared when we got to the last component we are
2718 * about to look up
2719 */
2720 error = complete_walk(nd);
2721 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002722 return error;
Miklos Szeredib6183df2012-06-05 15:10:13 +02002723
Jeff Layton33e22082013-04-12 15:16:32 -04002724 audit_inode(name, dir, LOOKUP_PARENT);
Miklos Szeredib6183df2012-06-05 15:10:13 +02002725 error = -EISDIR;
2726 /* trailing slashes? */
2727 if (nd->last.name[nd->last.len])
Al Viro2675a4e2012-06-22 12:41:10 +04002728 goto out;
Al Virofe2d35f2011-03-05 22:58:25 -05002729 }
2730
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002731retry_lookup:
Al Viro64894cf2012-07-31 00:53:35 +04002732 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2733 error = mnt_want_write(nd->path.mnt);
2734 if (!error)
2735 got_write = true;
2736 /*
2737 * do _not_ fail yet - we might not need that or fail with
2738 * a different error; let lookup_open() decide; we'll be
2739 * dropping this one anyway.
2740 */
2741 }
Al Viroa1e28032009-12-24 02:12:06 -05002742 mutex_lock(&dir->d_inode->i_mutex);
Al Viro64894cf2012-07-31 00:53:35 +04002743 error = lookup_open(nd, path, file, op, got_write, opened);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002744 mutex_unlock(&dir->d_inode->i_mutex);
Al Viroa1e28032009-12-24 02:12:06 -05002745
Al Viro2675a4e2012-06-22 12:41:10 +04002746 if (error <= 0) {
2747 if (error)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002748 goto out;
2749
Al Viro47237682012-06-10 05:01:45 -04002750 if ((*opened & FILE_CREATED) ||
Al Viro496ad9a2013-01-23 17:07:38 -05002751 !S_ISREG(file_inode(file)->i_mode))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002752 will_truncate = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002753
Jeff Laytonadb5c242012-10-10 16:43:13 -04002754 audit_inode(name, file->f_path.dentry, 0);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002755 goto opened;
2756 }
Al Virofb1cc552009-12-24 01:58:28 -05002757
Al Viro47237682012-06-10 05:01:45 -04002758 if (*opened & FILE_CREATED) {
Al Viro9b44f1b2011-03-09 00:17:27 -05002759 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05002760 open_flag &= ~O_TRUNC;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002761 will_truncate = false;
Al Virobcda7652011-03-13 16:42:14 -04002762 acc_mode = MAY_OPEN;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002763 path_to_nameidata(path, nd);
Miklos Szeredie83db162012-06-05 15:10:29 +02002764 goto finish_open_created;
Al Virofb1cc552009-12-24 01:58:28 -05002765 }
2766
2767 /*
Jeff Layton3134f372012-07-25 10:19:47 -04002768 * create/update audit record if it already exists.
Al Virofb1cc552009-12-24 01:58:28 -05002769 */
Jeff Layton3134f372012-07-25 10:19:47 -04002770 if (path->dentry->d_inode)
Jeff Laytonadb5c242012-10-10 16:43:13 -04002771 audit_inode(name, path->dentry, 0);
Al Virofb1cc552009-12-24 01:58:28 -05002772
Miklos Szeredid18e9002012-06-05 15:10:17 +02002773 /*
2774 * If atomic_open() acquired write access it is dropped now due to
2775 * possible mount and symlink following (this might be optimized away if
2776 * necessary...)
2777 */
Al Viro64894cf2012-07-31 00:53:35 +04002778 if (got_write) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002779 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04002780 got_write = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002781 }
2782
Al Virofb1cc552009-12-24 01:58:28 -05002783 error = -EEXIST;
Al Virof8310c52012-07-30 11:50:30 +04002784 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
Al Virofb1cc552009-12-24 01:58:28 -05002785 goto exit_dput;
2786
David Howells9875cf82011-01-14 18:45:21 +00002787 error = follow_managed(path, nd->flags);
2788 if (error < 0)
2789 goto exit_dput;
Al Virofb1cc552009-12-24 01:58:28 -05002790
Al Viroa3fbbde2011-11-07 21:21:26 +00002791 if (error)
2792 nd->flags |= LOOKUP_JUMPED;
2793
Miklos Szeredidecf3402012-05-21 17:30:08 +02002794 BUG_ON(nd->flags & LOOKUP_RCU);
2795 inode = path->dentry->d_inode;
Miklos Szeredi5f5daac2012-05-21 17:30:14 +02002796finish_lookup:
2797 /* we _can_ be in RCU mode here */
Al Virofb1cc552009-12-24 01:58:28 -05002798 error = -ENOENT;
Miklos Szeredi54c33e72012-05-21 17:30:10 +02002799 if (!inode) {
2800 path_to_nameidata(path, nd);
Al Viro2675a4e2012-06-22 12:41:10 +04002801 goto out;
Miklos Szeredi54c33e72012-05-21 17:30:10 +02002802 }
Al Viro9e67f362009-12-26 07:04:50 -05002803
Miklos Szeredid45ea862012-05-21 17:30:09 +02002804 if (should_follow_link(inode, !symlink_ok)) {
2805 if (nd->flags & LOOKUP_RCU) {
2806 if (unlikely(unlazy_walk(nd, path->dentry))) {
2807 error = -ECHILD;
Al Viro2675a4e2012-06-22 12:41:10 +04002808 goto out;
Miklos Szeredid45ea862012-05-21 17:30:09 +02002809 }
2810 }
2811 BUG_ON(inode != path->dentry->d_inode);
Al Viro2675a4e2012-06-22 12:41:10 +04002812 return 1;
Miklos Szeredid45ea862012-05-21 17:30:09 +02002813 }
Al Virofb1cc552009-12-24 01:58:28 -05002814
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002815 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
2816 path_to_nameidata(path, nd);
2817 } else {
2818 save_parent.dentry = nd->path.dentry;
2819 save_parent.mnt = mntget(path->mnt);
2820 nd->path.dentry = path->dentry;
2821
2822 }
Miklos Szeredidecf3402012-05-21 17:30:08 +02002823 nd->inode = inode;
Al Viroa3fbbde2011-11-07 21:21:26 +00002824 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
Al Virobc77daa2013-06-06 09:12:33 -04002825finish_open:
Al Viroa3fbbde2011-11-07 21:21:26 +00002826 error = complete_walk(nd);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002827 if (error) {
2828 path_put(&save_parent);
Al Viro2675a4e2012-06-22 12:41:10 +04002829 return error;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002830 }
Al Virobc77daa2013-06-06 09:12:33 -04002831 audit_inode(name, nd->path.dentry, 0);
Al Virofb1cc552009-12-24 01:58:28 -05002832 error = -EISDIR;
Miklos Szeredi050ac842012-05-21 17:30:12 +02002833 if ((open_flag & O_CREAT) && S_ISDIR(nd->inode->i_mode))
Al Viro2675a4e2012-06-22 12:41:10 +04002834 goto out;
Miklos Szerediaf2f5542012-05-21 17:30:11 +02002835 error = -ENOTDIR;
Al Viro05252902013-06-06 19:33:47 -04002836 if ((nd->flags & LOOKUP_DIRECTORY) && !can_lookup(nd->inode))
Al Viro2675a4e2012-06-22 12:41:10 +04002837 goto out;
Al Viro6c0d46c2011-03-09 00:59:59 -05002838 if (!S_ISREG(nd->inode->i_mode))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002839 will_truncate = false;
Al Viro6c0d46c2011-03-09 00:59:59 -05002840
Al Viro0f9d1a12011-03-09 00:13:14 -05002841 if (will_truncate) {
2842 error = mnt_want_write(nd->path.mnt);
2843 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002844 goto out;
Al Viro64894cf2012-07-31 00:53:35 +04002845 got_write = true;
Al Viro0f9d1a12011-03-09 00:13:14 -05002846 }
Miklos Szeredie83db162012-06-05 15:10:29 +02002847finish_open_created:
Al Virobcda7652011-03-13 16:42:14 -04002848 error = may_open(&nd->path, acc_mode, open_flag);
Al Viroca344a892011-03-09 00:36:45 -05002849 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002850 goto out;
Al Viro30d90492012-06-22 12:40:19 +04002851 file->f_path.mnt = nd->path.mnt;
2852 error = finish_open(file, nd->path.dentry, NULL, opened);
2853 if (error) {
Al Viro30d90492012-06-22 12:40:19 +04002854 if (error == -EOPENSTALE)
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02002855 goto stale_open;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002856 goto out;
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02002857 }
Miklos Szeredia8277b92012-06-05 15:10:32 +02002858opened:
Al Viro2675a4e2012-06-22 12:41:10 +04002859 error = open_check_o_direct(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002860 if (error)
2861 goto exit_fput;
Al Viro2675a4e2012-06-22 12:41:10 +04002862 error = ima_file_check(file, op->acc_mode);
Miklos Szerediaa4caad2012-06-05 15:10:28 +02002863 if (error)
2864 goto exit_fput;
2865
2866 if (will_truncate) {
Al Viro2675a4e2012-06-22 12:41:10 +04002867 error = handle_truncate(file);
Miklos Szerediaa4caad2012-06-05 15:10:28 +02002868 if (error)
2869 goto exit_fput;
Al Viro0f9d1a12011-03-09 00:13:14 -05002870 }
Al Viroca344a892011-03-09 00:36:45 -05002871out:
Al Viro64894cf2012-07-31 00:53:35 +04002872 if (got_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05002873 mnt_drop_write(nd->path.mnt);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002874 path_put(&save_parent);
Miklos Szeredie276ae62012-05-21 17:30:06 +02002875 terminate_walk(nd);
Al Viro2675a4e2012-06-22 12:41:10 +04002876 return error;
Al Virofb1cc552009-12-24 01:58:28 -05002877
Al Virofb1cc552009-12-24 01:58:28 -05002878exit_dput:
2879 path_put_conditional(path, nd);
Al Viroca344a892011-03-09 00:36:45 -05002880 goto out;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002881exit_fput:
Al Viro2675a4e2012-06-22 12:41:10 +04002882 fput(file);
2883 goto out;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002884
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02002885stale_open:
2886 /* If no saved parent or already retried then can't retry */
2887 if (!save_parent.dentry || retried)
2888 goto out;
2889
2890 BUG_ON(save_parent.dentry != dir);
2891 path_put(&nd->path);
2892 nd->path = save_parent;
2893 nd->inode = dir->d_inode;
2894 save_parent.mnt = NULL;
2895 save_parent.dentry = NULL;
Al Viro64894cf2012-07-31 00:53:35 +04002896 if (got_write) {
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02002897 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04002898 got_write = false;
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02002899 }
2900 retried = true;
2901 goto retry_lookup;
Al Virofb1cc552009-12-24 01:58:28 -05002902}
2903
Al Viro60545d02013-06-07 01:20:27 -04002904static int do_tmpfile(int dfd, struct filename *pathname,
2905 struct nameidata *nd, int flags,
2906 const struct open_flags *op,
2907 struct file *file, int *opened)
2908{
2909 static const struct qstr name = QSTR_INIT("/", 1);
2910 struct dentry *dentry, *child;
2911 struct inode *dir;
2912 int error = path_lookupat(dfd, pathname->name,
2913 flags | LOOKUP_DIRECTORY, nd);
2914 if (unlikely(error))
2915 return error;
2916 error = mnt_want_write(nd->path.mnt);
2917 if (unlikely(error))
2918 goto out;
2919 /* we want directory to be writable */
2920 error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
2921 if (error)
2922 goto out2;
2923 dentry = nd->path.dentry;
2924 dir = dentry->d_inode;
2925 if (!dir->i_op->tmpfile) {
2926 error = -EOPNOTSUPP;
2927 goto out2;
2928 }
2929 child = d_alloc(dentry, &name);
2930 if (unlikely(!child)) {
2931 error = -ENOMEM;
2932 goto out2;
2933 }
2934 nd->flags &= ~LOOKUP_DIRECTORY;
2935 nd->flags |= op->intent;
2936 dput(nd->path.dentry);
2937 nd->path.dentry = child;
2938 error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
2939 if (error)
2940 goto out2;
2941 audit_inode(pathname, nd->path.dentry, 0);
2942 error = may_open(&nd->path, op->acc_mode, op->open_flag);
2943 if (error)
2944 goto out2;
2945 file->f_path.mnt = nd->path.mnt;
2946 error = finish_open(file, nd->path.dentry, NULL, opened);
2947 if (error)
2948 goto out2;
2949 error = open_check_o_direct(file);
Al Virof4e0c302013-06-11 08:34:36 +04002950 if (error) {
Al Viro60545d02013-06-07 01:20:27 -04002951 fput(file);
Al Virof4e0c302013-06-11 08:34:36 +04002952 } else if (!(op->open_flag & O_EXCL)) {
2953 struct inode *inode = file_inode(file);
2954 spin_lock(&inode->i_lock);
2955 inode->i_state |= I_LINKABLE;
2956 spin_unlock(&inode->i_lock);
2957 }
Al Viro60545d02013-06-07 01:20:27 -04002958out2:
2959 mnt_drop_write(nd->path.mnt);
2960out:
2961 path_put(&nd->path);
2962 return error;
2963}
2964
Jeff Layton669abf42012-10-10 16:43:10 -04002965static struct file *path_openat(int dfd, struct filename *pathname,
Al Viro73d049a2011-03-11 12:08:24 -05002966 struct nameidata *nd, const struct open_flags *op, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002967{
Al Virofe2d35f2011-03-05 22:58:25 -05002968 struct file *base = NULL;
Al Viro30d90492012-06-22 12:40:19 +04002969 struct file *file;
Al Viro9850c052010-01-13 15:01:15 -05002970 struct path path;
Al Viro47237682012-06-10 05:01:45 -04002971 int opened = 0;
Al Viro13aab422011-02-23 17:54:08 -05002972 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11002973
Al Viro30d90492012-06-22 12:40:19 +04002974 file = get_empty_filp();
Al Viro1afc99b2013-02-14 20:41:04 -05002975 if (IS_ERR(file))
2976 return file;
Nick Piggin31e6b012011-01-07 17:49:52 +11002977
Al Viro30d90492012-06-22 12:40:19 +04002978 file->f_flags = op->open_flag;
Nick Piggin31e6b012011-01-07 17:49:52 +11002979
Al Virobb458c62013-07-13 13:26:37 +04002980 if (unlikely(file->f_flags & __O_TMPFILE)) {
Al Viro60545d02013-06-07 01:20:27 -04002981 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
2982 goto out;
2983 }
2984
Jeff Layton669abf42012-10-10 16:43:10 -04002985 error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base);
Nick Piggin31e6b012011-01-07 17:49:52 +11002986 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04002987 goto out;
Nick Piggin31e6b012011-01-07 17:49:52 +11002988
Al Virofe2d35f2011-03-05 22:58:25 -05002989 current->total_link_count = 0;
Jeff Layton669abf42012-10-10 16:43:10 -04002990 error = link_path_walk(pathname->name, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11002991 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04002992 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002993
Al Viro2675a4e2012-06-22 12:41:10 +04002994 error = do_last(nd, &path, file, op, &opened, pathname);
2995 while (unlikely(error > 0)) { /* trailing symlink */
Nick Piggin7b9337a2011-01-14 08:42:43 +00002996 struct path link = path;
Al Virodef4af32009-12-26 08:37:05 -05002997 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04002998 if (!(nd->flags & LOOKUP_FOLLOW)) {
Al Viro73d049a2011-03-11 12:08:24 -05002999 path_put_conditional(&path, nd);
3000 path_put(&nd->path);
Al Viro2675a4e2012-06-22 12:41:10 +04003001 error = -ELOOP;
Al Viro40b39132011-03-09 16:22:18 -05003002 break;
3003 }
Kees Cook800179c2012-07-25 17:29:07 -07003004 error = may_follow_link(&link, nd);
3005 if (unlikely(error))
3006 break;
Al Viro73d049a2011-03-11 12:08:24 -05003007 nd->flags |= LOOKUP_PARENT;
3008 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
Al Viro574197e2011-03-14 22:20:34 -04003009 error = follow_link(&link, nd, &cookie);
Al Viroc3e380b2011-02-23 13:39:45 -05003010 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04003011 break;
3012 error = do_last(nd, &path, file, op, &opened, pathname);
Al Viro574197e2011-03-14 22:20:34 -04003013 put_link(nd, &link, cookie);
Al Viro806b6812009-12-26 07:16:40 -05003014 }
Al Viro10fa8e62009-12-26 07:09:49 -05003015out:
Al Viro73d049a2011-03-11 12:08:24 -05003016 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
3017 path_put(&nd->root);
Al Virofe2d35f2011-03-05 22:58:25 -05003018 if (base)
3019 fput(base);
Al Viro2675a4e2012-06-22 12:41:10 +04003020 if (!(opened & FILE_OPENED)) {
3021 BUG_ON(!error);
Al Viro30d90492012-06-22 12:40:19 +04003022 put_filp(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003023 }
Al Viro2675a4e2012-06-22 12:41:10 +04003024 if (unlikely(error)) {
3025 if (error == -EOPENSTALE) {
3026 if (flags & LOOKUP_RCU)
3027 error = -ECHILD;
3028 else
3029 error = -ESTALE;
3030 }
3031 file = ERR_PTR(error);
3032 }
3033 return file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003034}
3035
Jeff Layton669abf42012-10-10 16:43:10 -04003036struct file *do_filp_open(int dfd, struct filename *pathname,
Al Virof9652e12013-06-11 08:23:01 +04003037 const struct open_flags *op)
Al Viro13aab422011-02-23 17:54:08 -05003038{
Al Viro73d049a2011-03-11 12:08:24 -05003039 struct nameidata nd;
Al Virof9652e12013-06-11 08:23:01 +04003040 int flags = op->lookup_flags;
Al Viro13aab422011-02-23 17:54:08 -05003041 struct file *filp;
3042
Al Viro73d049a2011-03-11 12:08:24 -05003043 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
Al Viro13aab422011-02-23 17:54:08 -05003044 if (unlikely(filp == ERR_PTR(-ECHILD)))
Al Viro73d049a2011-03-11 12:08:24 -05003045 filp = path_openat(dfd, pathname, &nd, op, flags);
Al Viro13aab422011-02-23 17:54:08 -05003046 if (unlikely(filp == ERR_PTR(-ESTALE)))
Al Viro73d049a2011-03-11 12:08:24 -05003047 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
Al Viro13aab422011-02-23 17:54:08 -05003048 return filp;
3049}
3050
Al Viro73d049a2011-03-11 12:08:24 -05003051struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
Al Virof9652e12013-06-11 08:23:01 +04003052 const char *name, const struct open_flags *op)
Al Viro73d049a2011-03-11 12:08:24 -05003053{
3054 struct nameidata nd;
3055 struct file *file;
Jeff Layton669abf42012-10-10 16:43:10 -04003056 struct filename filename = { .name = name };
Al Virof9652e12013-06-11 08:23:01 +04003057 int flags = op->lookup_flags | LOOKUP_ROOT;
Al Viro73d049a2011-03-11 12:08:24 -05003058
3059 nd.root.mnt = mnt;
3060 nd.root.dentry = dentry;
3061
Al Virobcda7652011-03-13 16:42:14 -04003062 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
Al Viro73d049a2011-03-11 12:08:24 -05003063 return ERR_PTR(-ELOOP);
3064
Jeff Layton669abf42012-10-10 16:43:10 -04003065 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU);
Al Viro73d049a2011-03-11 12:08:24 -05003066 if (unlikely(file == ERR_PTR(-ECHILD)))
Jeff Layton669abf42012-10-10 16:43:10 -04003067 file = path_openat(-1, &filename, &nd, op, flags);
Al Viro73d049a2011-03-11 12:08:24 -05003068 if (unlikely(file == ERR_PTR(-ESTALE)))
Jeff Layton669abf42012-10-10 16:43:10 -04003069 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL);
Al Viro73d049a2011-03-11 12:08:24 -05003070 return file;
3071}
3072
Jeff Layton1ac12b42012-12-11 12:10:06 -05003073struct dentry *kern_path_create(int dfd, const char *pathname,
3074 struct path *path, unsigned int lookup_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003075{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003076 struct dentry *dentry = ERR_PTR(-EEXIST);
Al Viroed75e952011-06-27 16:53:43 -04003077 struct nameidata nd;
Jan Karac30dabf2012-06-12 16:20:30 +02003078 int err2;
Jeff Layton1ac12b42012-12-11 12:10:06 -05003079 int error;
3080 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3081
3082 /*
3083 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3084 * other flags passed in are ignored!
3085 */
3086 lookup_flags &= LOOKUP_REVAL;
3087
3088 error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd);
Al Viroed75e952011-06-27 16:53:43 -04003089 if (error)
3090 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003091
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003092 /*
3093 * Yucky last component or no last component at all?
3094 * (foo/., foo/.., /////)
3095 */
Al Viroed75e952011-06-27 16:53:43 -04003096 if (nd.last_type != LAST_NORM)
3097 goto out;
3098 nd.flags &= ~LOOKUP_PARENT;
3099 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003100
Jan Karac30dabf2012-06-12 16:20:30 +02003101 /* don't fail immediately if it's r/o, at least try to report other errors */
3102 err2 = mnt_want_write(nd.path.mnt);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003103 /*
3104 * Do the final lookup.
3105 */
Al Viroed75e952011-06-27 16:53:43 -04003106 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3107 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003108 if (IS_ERR(dentry))
Al Viroa8104a92012-07-20 02:25:00 +04003109 goto unlock;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003110
Al Viroa8104a92012-07-20 02:25:00 +04003111 error = -EEXIST;
Al Viroe9baf6e2008-05-15 04:49:12 -04003112 if (dentry->d_inode)
Al Viroa8104a92012-07-20 02:25:00 +04003113 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003114 /*
3115 * Special case - lookup gave negative, but... we had foo/bar/
3116 * From the vfs_mknod() POV we just have a negative dentry -
3117 * all is fine. Let's be bastards - you had / on the end, you've
3118 * been asking for (non-existent) directory. -ENOENT for you.
3119 */
Al Viroed75e952011-06-27 16:53:43 -04003120 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
Al Viroa8104a92012-07-20 02:25:00 +04003121 error = -ENOENT;
Al Viroed75e952011-06-27 16:53:43 -04003122 goto fail;
Al Viroe9baf6e2008-05-15 04:49:12 -04003123 }
Jan Karac30dabf2012-06-12 16:20:30 +02003124 if (unlikely(err2)) {
3125 error = err2;
Al Viroa8104a92012-07-20 02:25:00 +04003126 goto fail;
Jan Karac30dabf2012-06-12 16:20:30 +02003127 }
Al Viroed75e952011-06-27 16:53:43 -04003128 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003129 return dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003130fail:
Al Viroa8104a92012-07-20 02:25:00 +04003131 dput(dentry);
3132 dentry = ERR_PTR(error);
3133unlock:
Al Viroed75e952011-06-27 16:53:43 -04003134 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Karac30dabf2012-06-12 16:20:30 +02003135 if (!err2)
3136 mnt_drop_write(nd.path.mnt);
Al Viroed75e952011-06-27 16:53:43 -04003137out:
3138 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003139 return dentry;
3140}
Al Virodae6ad82011-06-26 11:50:15 -04003141EXPORT_SYMBOL(kern_path_create);
3142
Al Viro921a1652012-07-20 01:15:31 +04003143void done_path_create(struct path *path, struct dentry *dentry)
3144{
3145 dput(dentry);
3146 mutex_unlock(&path->dentry->d_inode->i_mutex);
Al Viroa8104a92012-07-20 02:25:00 +04003147 mnt_drop_write(path->mnt);
Al Viro921a1652012-07-20 01:15:31 +04003148 path_put(path);
3149}
3150EXPORT_SYMBOL(done_path_create);
3151
Jeff Layton1ac12b42012-12-11 12:10:06 -05003152struct dentry *user_path_create(int dfd, const char __user *pathname,
3153 struct path *path, unsigned int lookup_flags)
Al Virodae6ad82011-06-26 11:50:15 -04003154{
Jeff Layton91a27b22012-10-10 15:25:28 -04003155 struct filename *tmp = getname(pathname);
Al Virodae6ad82011-06-26 11:50:15 -04003156 struct dentry *res;
3157 if (IS_ERR(tmp))
3158 return ERR_CAST(tmp);
Jeff Layton1ac12b42012-12-11 12:10:06 -05003159 res = kern_path_create(dfd, tmp->name, path, lookup_flags);
Al Virodae6ad82011-06-26 11:50:15 -04003160 putname(tmp);
3161 return res;
3162}
3163EXPORT_SYMBOL(user_path_create);
3164
Al Viro1a67aaf2011-07-26 01:52:52 -04003165int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003166{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003167 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003168
3169 if (error)
3170 return error;
3171
Eric W. Biederman975d6b32011-11-13 12:16:43 -08003172 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003173 return -EPERM;
3174
Al Viroacfa4382008-12-04 10:06:33 -05003175 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003176 return -EPERM;
3177
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07003178 error = devcgroup_inode_mknod(mode, dev);
3179 if (error)
3180 return error;
3181
Linus Torvalds1da177e2005-04-16 15:20:36 -07003182 error = security_inode_mknod(dir, dentry, mode, dev);
3183 if (error)
3184 return error;
3185
Linus Torvalds1da177e2005-04-16 15:20:36 -07003186 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003187 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003188 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003189 return error;
3190}
3191
Al Virof69aac02011-07-26 04:31:40 -04003192static int may_mknod(umode_t mode)
Dave Hansen463c3192008-02-15 14:37:57 -08003193{
3194 switch (mode & S_IFMT) {
3195 case S_IFREG:
3196 case S_IFCHR:
3197 case S_IFBLK:
3198 case S_IFIFO:
3199 case S_IFSOCK:
3200 case 0: /* zero mode translates to S_IFREG */
3201 return 0;
3202 case S_IFDIR:
3203 return -EPERM;
3204 default:
3205 return -EINVAL;
3206 }
3207}
3208
Al Viro8208a222011-07-25 17:32:17 -04003209SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003210 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003211{
Al Viro2ad94ae2008-07-21 09:32:51 -04003212 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003213 struct path path;
3214 int error;
Jeff Layton972567f2012-12-20 16:00:10 -05003215 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003216
Al Viro8e4bfca2012-07-20 01:17:26 +04003217 error = may_mknod(mode);
3218 if (error)
3219 return error;
Jeff Layton972567f2012-12-20 16:00:10 -05003220retry:
3221 dentry = user_path_create(dfd, filename, &path, lookup_flags);
Al Virodae6ad82011-06-26 11:50:15 -04003222 if (IS_ERR(dentry))
3223 return PTR_ERR(dentry);
Al Viro2ad94ae2008-07-21 09:32:51 -04003224
Al Virodae6ad82011-06-26 11:50:15 -04003225 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003226 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003227 error = security_path_mknod(&path, dentry, mode, dev);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003228 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04003229 goto out;
Dave Hansen463c3192008-02-15 14:37:57 -08003230 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003231 case 0: case S_IFREG:
Al Viro312b63f2012-06-10 18:09:36 -04003232 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003233 break;
3234 case S_IFCHR: case S_IFBLK:
Al Virodae6ad82011-06-26 11:50:15 -04003235 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003236 new_decode_dev(dev));
3237 break;
3238 case S_IFIFO: case S_IFSOCK:
Al Virodae6ad82011-06-26 11:50:15 -04003239 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003240 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003241 }
Al Viroa8104a92012-07-20 02:25:00 +04003242out:
Al Viro921a1652012-07-20 01:15:31 +04003243 done_path_create(&path, dentry);
Jeff Layton972567f2012-12-20 16:00:10 -05003244 if (retry_estale(error, lookup_flags)) {
3245 lookup_flags |= LOOKUP_REVAL;
3246 goto retry;
3247 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003248 return error;
3249}
3250
Al Viro8208a222011-07-25 17:32:17 -04003251SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003252{
3253 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3254}
3255
Al Viro18bb1db2011-07-26 01:41:39 -04003256int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003257{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003258 int error = may_create(dir, dentry);
Al Viro8de52772012-02-06 12:45:27 -05003259 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003260
3261 if (error)
3262 return error;
3263
Al Viroacfa4382008-12-04 10:06:33 -05003264 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003265 return -EPERM;
3266
3267 mode &= (S_IRWXUGO|S_ISVTX);
3268 error = security_inode_mkdir(dir, dentry, mode);
3269 if (error)
3270 return error;
3271
Al Viro8de52772012-02-06 12:45:27 -05003272 if (max_links && dir->i_nlink >= max_links)
3273 return -EMLINK;
3274
Linus Torvalds1da177e2005-04-16 15:20:36 -07003275 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003276 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003277 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003278 return error;
3279}
3280
Al Viroa218d0f2011-11-21 14:59:34 -05003281SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003282{
Dave Hansen6902d922006-09-30 23:29:01 -07003283 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003284 struct path path;
3285 int error;
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003286 unsigned int lookup_flags = LOOKUP_DIRECTORY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003287
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003288retry:
3289 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07003290 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003291 return PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003292
Al Virodae6ad82011-06-26 11:50:15 -04003293 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003294 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003295 error = security_path_mkdir(&path, dentry, mode);
Al Viroa8104a92012-07-20 02:25:00 +04003296 if (!error)
3297 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
Al Viro921a1652012-07-20 01:15:31 +04003298 done_path_create(&path, dentry);
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003299 if (retry_estale(error, lookup_flags)) {
3300 lookup_flags |= LOOKUP_REVAL;
3301 goto retry;
3302 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003303 return error;
3304}
3305
Al Viroa218d0f2011-11-21 14:59:34 -05003306SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003307{
3308 return sys_mkdirat(AT_FDCWD, pathname, mode);
3309}
3310
Linus Torvalds1da177e2005-04-16 15:20:36 -07003311/*
Sage Weila71905f2011-05-24 13:06:08 -07003312 * The dentry_unhash() helper will try to drop the dentry early: we
J. Bruce Fieldsc0d02592012-02-15 11:48:40 -05003313 * should have a usage count of 1 if we're the only user of this
Sage Weila71905f2011-05-24 13:06:08 -07003314 * dentry, and if that is true (possibly after pruning the dcache),
3315 * then we drop the dentry now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003316 *
3317 * A low-level filesystem can, if it choses, legally
3318 * do a
3319 *
3320 * if (!d_unhashed(dentry))
3321 * return -EBUSY;
3322 *
3323 * if it cannot handle the case of removing a directory
3324 * that is still in use by something else..
3325 */
3326void dentry_unhash(struct dentry *dentry)
3327{
Vasily Averindc168422006-12-06 20:37:07 -08003328 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003329 spin_lock(&dentry->d_lock);
Sage Weil64252c72011-05-24 13:06:05 -07003330 if (dentry->d_count == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003331 __d_drop(dentry);
3332 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003333}
3334
3335int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3336{
3337 int error = may_delete(dir, dentry, 1);
3338
3339 if (error)
3340 return error;
3341
Al Viroacfa4382008-12-04 10:06:33 -05003342 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003343 return -EPERM;
3344
Al Viro1d2ef592011-09-14 18:55:41 +01003345 dget(dentry);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003346 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003347
Sage Weil912dbc12011-05-24 13:06:11 -07003348 error = -EBUSY;
3349 if (d_mountpoint(dentry))
3350 goto out;
3351
3352 error = security_inode_rmdir(dir, dentry);
3353 if (error)
3354 goto out;
3355
Sage Weil3cebde22011-05-29 21:20:59 -07003356 shrink_dcache_parent(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003357 error = dir->i_op->rmdir(dir, dentry);
3358 if (error)
3359 goto out;
3360
3361 dentry->d_inode->i_flags |= S_DEAD;
3362 dont_mount(dentry);
3363
3364out:
3365 mutex_unlock(&dentry->d_inode->i_mutex);
Al Viro1d2ef592011-09-14 18:55:41 +01003366 dput(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003367 if (!error)
3368 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003369 return error;
3370}
3371
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003372static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003373{
3374 int error = 0;
Jeff Layton91a27b22012-10-10 15:25:28 -04003375 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003376 struct dentry *dentry;
3377 struct nameidata nd;
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003378 unsigned int lookup_flags = 0;
3379retry:
3380 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04003381 if (IS_ERR(name))
3382 return PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003383
3384 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003385 case LAST_DOTDOT:
3386 error = -ENOTEMPTY;
3387 goto exit1;
3388 case LAST_DOT:
3389 error = -EINVAL;
3390 goto exit1;
3391 case LAST_ROOT:
3392 error = -EBUSY;
3393 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003394 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003395
3396 nd.flags &= ~LOOKUP_PARENT;
Jan Karac30dabf2012-06-12 16:20:30 +02003397 error = mnt_want_write(nd.path.mnt);
3398 if (error)
3399 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003400
Jan Blunck4ac91372008-02-14 19:34:32 -08003401 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08003402 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003403 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003404 if (IS_ERR(dentry))
3405 goto exit2;
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003406 if (!dentry->d_inode) {
3407 error = -ENOENT;
3408 goto exit3;
3409 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003410 error = security_path_rmdir(&nd.path, dentry);
3411 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003412 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08003413 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Dave Hansen06227532008-02-15 14:37:34 -08003414exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07003415 dput(dentry);
3416exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08003417 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Karac30dabf2012-06-12 16:20:30 +02003418 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003419exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003420 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003421 putname(name);
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003422 if (retry_estale(error, lookup_flags)) {
3423 lookup_flags |= LOOKUP_REVAL;
3424 goto retry;
3425 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003426 return error;
3427}
3428
Heiko Carstens3cdad422009-01-14 14:14:22 +01003429SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003430{
3431 return do_rmdir(AT_FDCWD, pathname);
3432}
3433
Linus Torvalds1da177e2005-04-16 15:20:36 -07003434int vfs_unlink(struct inode *dir, struct dentry *dentry)
3435{
3436 int error = may_delete(dir, dentry, 0);
3437
3438 if (error)
3439 return error;
3440
Al Viroacfa4382008-12-04 10:06:33 -05003441 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003442 return -EPERM;
3443
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003444 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003445 if (d_mountpoint(dentry))
3446 error = -EBUSY;
3447 else {
3448 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05003449 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003450 error = dir->i_op->unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05003451 if (!error)
Al Virod83c49f2010-04-30 17:17:09 -04003452 dont_mount(dentry);
Al Virobec10522010-03-03 14:12:08 -05003453 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003454 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003455 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003456
3457 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3458 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08003459 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003460 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003461 }
Robert Love0eeca282005-07-12 17:06:03 -04003462
Linus Torvalds1da177e2005-04-16 15:20:36 -07003463 return error;
3464}
3465
3466/*
3467 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003468 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07003469 * writeout happening, and we don't want to prevent access to the directory
3470 * while waiting on the I/O.
3471 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003472static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003473{
Al Viro2ad94ae2008-07-21 09:32:51 -04003474 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04003475 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003476 struct dentry *dentry;
3477 struct nameidata nd;
3478 struct inode *inode = NULL;
Jeff Layton5d18f812012-12-20 16:38:04 -05003479 unsigned int lookup_flags = 0;
3480retry:
3481 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04003482 if (IS_ERR(name))
3483 return PTR_ERR(name);
Al Viro2ad94ae2008-07-21 09:32:51 -04003484
Linus Torvalds1da177e2005-04-16 15:20:36 -07003485 error = -EISDIR;
3486 if (nd.last_type != LAST_NORM)
3487 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003488
3489 nd.flags &= ~LOOKUP_PARENT;
Jan Karac30dabf2012-06-12 16:20:30 +02003490 error = mnt_want_write(nd.path.mnt);
3491 if (error)
3492 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003493
Jan Blunck4ac91372008-02-14 19:34:32 -08003494 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08003495 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003496 error = PTR_ERR(dentry);
3497 if (!IS_ERR(dentry)) {
3498 /* Why not before? Because we want correct error value */
Török Edwin50338b82011-06-16 00:06:14 +03003499 if (nd.last.name[nd.last.len])
3500 goto slashes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003501 inode = dentry->d_inode;
Török Edwin50338b82011-06-16 00:06:14 +03003502 if (!inode)
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003503 goto slashes;
3504 ihold(inode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003505 error = security_path_unlink(&nd.path, dentry);
3506 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003507 goto exit2;
Jan Blunck4ac91372008-02-14 19:34:32 -08003508 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Jan Karac30dabf2012-06-12 16:20:30 +02003509exit2:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003510 dput(dentry);
3511 }
Jan Blunck4ac91372008-02-14 19:34:32 -08003512 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003513 if (inode)
3514 iput(inode); /* truncate the inode here */
Jan Karac30dabf2012-06-12 16:20:30 +02003515 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003516exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003517 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003518 putname(name);
Jeff Layton5d18f812012-12-20 16:38:04 -05003519 if (retry_estale(error, lookup_flags)) {
3520 lookup_flags |= LOOKUP_REVAL;
3521 inode = NULL;
3522 goto retry;
3523 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003524 return error;
3525
3526slashes:
3527 error = !dentry->d_inode ? -ENOENT :
3528 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
3529 goto exit2;
3530}
3531
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003532SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003533{
3534 if ((flag & ~AT_REMOVEDIR) != 0)
3535 return -EINVAL;
3536
3537 if (flag & AT_REMOVEDIR)
3538 return do_rmdir(dfd, pathname);
3539
3540 return do_unlinkat(dfd, pathname);
3541}
3542
Heiko Carstens3480b252009-01-14 14:14:16 +01003543SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003544{
3545 return do_unlinkat(AT_FDCWD, pathname);
3546}
3547
Miklos Szeredidb2e7472008-06-24 16:50:16 +02003548int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003549{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003550 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003551
3552 if (error)
3553 return error;
3554
Al Viroacfa4382008-12-04 10:06:33 -05003555 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003556 return -EPERM;
3557
3558 error = security_inode_symlink(dir, dentry, oldname);
3559 if (error)
3560 return error;
3561
Linus Torvalds1da177e2005-04-16 15:20:36 -07003562 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003563 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003564 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003565 return error;
3566}
3567
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003568SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3569 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003570{
Al Viro2ad94ae2008-07-21 09:32:51 -04003571 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04003572 struct filename *from;
Dave Hansen6902d922006-09-30 23:29:01 -07003573 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003574 struct path path;
Jeff Laytonf46d3562012-12-11 12:10:08 -05003575 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003576
3577 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04003578 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003579 return PTR_ERR(from);
Jeff Laytonf46d3562012-12-11 12:10:08 -05003580retry:
3581 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07003582 error = PTR_ERR(dentry);
3583 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003584 goto out_putname;
Dave Hansen6902d922006-09-30 23:29:01 -07003585
Jeff Layton91a27b22012-10-10 15:25:28 -04003586 error = security_path_symlink(&path, dentry, from->name);
Al Viroa8104a92012-07-20 02:25:00 +04003587 if (!error)
Jeff Layton91a27b22012-10-10 15:25:28 -04003588 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
Al Viro921a1652012-07-20 01:15:31 +04003589 done_path_create(&path, dentry);
Jeff Laytonf46d3562012-12-11 12:10:08 -05003590 if (retry_estale(error, lookup_flags)) {
3591 lookup_flags |= LOOKUP_REVAL;
3592 goto retry;
3593 }
Dave Hansen6902d922006-09-30 23:29:01 -07003594out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003595 putname(from);
3596 return error;
3597}
3598
Heiko Carstens3480b252009-01-14 14:14:16 +01003599SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003600{
3601 return sys_symlinkat(oldname, AT_FDCWD, newname);
3602}
3603
Linus Torvalds1da177e2005-04-16 15:20:36 -07003604int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
3605{
3606 struct inode *inode = old_dentry->d_inode;
Al Viro8de52772012-02-06 12:45:27 -05003607 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003608 int error;
3609
3610 if (!inode)
3611 return -ENOENT;
3612
Miklos Szeredia95164d2008-07-30 15:08:48 +02003613 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003614 if (error)
3615 return error;
3616
3617 if (dir->i_sb != inode->i_sb)
3618 return -EXDEV;
3619
3620 /*
3621 * A link to an append-only or immutable file cannot be created.
3622 */
3623 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3624 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05003625 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003626 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003627 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003628 return -EPERM;
3629
3630 error = security_inode_link(old_dentry, dir, new_dentry);
3631 if (error)
3632 return error;
3633
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003634 mutex_lock(&inode->i_mutex);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05303635 /* Make sure we don't allow creating hardlink to an unlinked file */
Al Virof4e0c302013-06-11 08:34:36 +04003636 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05303637 error = -ENOENT;
Al Viro8de52772012-02-06 12:45:27 -05003638 else if (max_links && inode->i_nlink >= max_links)
3639 error = -EMLINK;
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05303640 else
3641 error = dir->i_op->link(old_dentry, dir, new_dentry);
Al Virof4e0c302013-06-11 08:34:36 +04003642
3643 if (!error && (inode->i_state & I_LINKABLE)) {
3644 spin_lock(&inode->i_lock);
3645 inode->i_state &= ~I_LINKABLE;
3646 spin_unlock(&inode->i_lock);
3647 }
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003648 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07003649 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003650 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003651 return error;
3652}
3653
3654/*
3655 * Hardlinks are often used in delicate situations. We avoid
3656 * security-related surprises by not following symlinks on the
3657 * newname. --KAB
3658 *
3659 * We don't follow them on the oldname either to be compatible
3660 * with linux 2.0, and to avoid hard-linking to directories
3661 * and other special files. --ADM
3662 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003663SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3664 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003665{
3666 struct dentry *new_dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003667 struct path old_path, new_path;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303668 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003669 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003670
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303671 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003672 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303673 /*
3674 * To use null names we require CAP_DAC_READ_SEARCH
3675 * This ensures that not everyone will be able to create
3676 * handlink using the passed filedescriptor.
3677 */
3678 if (flags & AT_EMPTY_PATH) {
3679 if (!capable(CAP_DAC_READ_SEARCH))
3680 return -ENOENT;
3681 how = LOOKUP_EMPTY;
3682 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003683
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303684 if (flags & AT_SYMLINK_FOLLOW)
3685 how |= LOOKUP_FOLLOW;
Jeff Layton442e31c2012-12-20 16:15:38 -05003686retry:
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303687 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003688 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04003689 return error;
3690
Jeff Layton442e31c2012-12-20 16:15:38 -05003691 new_dentry = user_path_create(newdfd, newname, &new_path,
3692 (how & LOOKUP_REVAL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003693 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003694 if (IS_ERR(new_dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003695 goto out;
3696
3697 error = -EXDEV;
3698 if (old_path.mnt != new_path.mnt)
3699 goto out_dput;
Kees Cook800179c2012-07-25 17:29:07 -07003700 error = may_linkat(&old_path);
3701 if (unlikely(error))
3702 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04003703 error = security_path_link(old_path.dentry, &new_path, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003704 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04003705 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04003706 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
Dave Hansen75c3f292008-02-15 14:37:45 -08003707out_dput:
Al Viro921a1652012-07-20 01:15:31 +04003708 done_path_create(&new_path, new_dentry);
Jeff Layton442e31c2012-12-20 16:15:38 -05003709 if (retry_estale(error, how)) {
3710 how |= LOOKUP_REVAL;
3711 goto retry;
3712 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003713out:
Al Viro2d8f3032008-07-22 09:59:21 -04003714 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003715
3716 return error;
3717}
3718
Heiko Carstens3480b252009-01-14 14:14:16 +01003719SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003720{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003721 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003722}
3723
Linus Torvalds1da177e2005-04-16 15:20:36 -07003724/*
3725 * The worst of all namespace operations - renaming directory. "Perverted"
3726 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3727 * Problems:
3728 * a) we can get into loop creation. Check is done in is_subdir().
3729 * b) race potential - two innocent renames can create a loop together.
3730 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003731 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07003732 * story.
3733 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003734 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07003735 * whether the target exists). Solution: try to be smart with locking
3736 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003737 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07003738 * move will be locked. Thus we can rank directories by the tree
3739 * (ancestors first) and rank all non-directories after them.
3740 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003741 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003742 * HOWEVER, it relies on the assumption that any object with ->lookup()
3743 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3744 * we'd better make sure that there's no link(2) for them.
Sage Weile4eaac02011-05-24 13:06:07 -07003745 * d) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003746 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07003747 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05003748 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07003749 * locking].
3750 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07003751static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3752 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003753{
3754 int error = 0;
Sage Weil9055cba2011-05-24 13:06:12 -07003755 struct inode *target = new_dentry->d_inode;
Al Viro8de52772012-02-06 12:45:27 -05003756 unsigned max_links = new_dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003757
3758 /*
3759 * If we are going to change the parent - check write permissions,
3760 * we'll need to flip '..'.
3761 */
3762 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04003763 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003764 if (error)
3765 return error;
3766 }
3767
3768 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3769 if (error)
3770 return error;
3771
Al Viro1d2ef592011-09-14 18:55:41 +01003772 dget(new_dentry);
Al Virod83c49f2010-04-30 17:17:09 -04003773 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003774 mutex_lock(&target->i_mutex);
Sage Weil9055cba2011-05-24 13:06:12 -07003775
3776 error = -EBUSY;
3777 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3778 goto out;
3779
Al Viro8de52772012-02-06 12:45:27 -05003780 error = -EMLINK;
3781 if (max_links && !target && new_dir != old_dir &&
3782 new_dir->i_nlink >= max_links)
3783 goto out;
3784
Sage Weil3cebde22011-05-29 21:20:59 -07003785 if (target)
3786 shrink_dcache_parent(new_dentry);
Sage Weil9055cba2011-05-24 13:06:12 -07003787 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3788 if (error)
3789 goto out;
3790
Linus Torvalds1da177e2005-04-16 15:20:36 -07003791 if (target) {
Sage Weil9055cba2011-05-24 13:06:12 -07003792 target->i_flags |= S_DEAD;
3793 dont_mount(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003794 }
Sage Weil9055cba2011-05-24 13:06:12 -07003795out:
3796 if (target)
3797 mutex_unlock(&target->i_mutex);
Al Viro1d2ef592011-09-14 18:55:41 +01003798 dput(new_dentry);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07003799 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07003800 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3801 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003802 return error;
3803}
3804
Adrian Bunk75c96f82005-05-05 16:16:09 -07003805static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3806 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003807{
Sage Weil51892bb2011-05-24 13:06:13 -07003808 struct inode *target = new_dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003809 int error;
3810
3811 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3812 if (error)
3813 return error;
3814
3815 dget(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003816 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003817 mutex_lock(&target->i_mutex);
Sage Weil51892bb2011-05-24 13:06:13 -07003818
3819 error = -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003820 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
Sage Weil51892bb2011-05-24 13:06:13 -07003821 goto out;
3822
3823 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3824 if (error)
3825 goto out;
3826
3827 if (target)
3828 dont_mount(new_dentry);
3829 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3830 d_move(old_dentry, new_dentry);
3831out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003832 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003833 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003834 dput(new_dentry);
3835 return error;
3836}
3837
3838int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3839 struct inode *new_dir, struct dentry *new_dentry)
3840{
3841 int error;
3842 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Eric Paris59b0df22010-02-08 12:53:52 -05003843 const unsigned char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003844
3845 if (old_dentry->d_inode == new_dentry->d_inode)
3846 return 0;
3847
3848 error = may_delete(old_dir, old_dentry, is_dir);
3849 if (error)
3850 return error;
3851
3852 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02003853 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003854 else
3855 error = may_delete(new_dir, new_dentry, is_dir);
3856 if (error)
3857 return error;
3858
Al Viroacfa4382008-12-04 10:06:33 -05003859 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003860 return -EPERM;
3861
Robert Love0eeca282005-07-12 17:06:03 -04003862 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3863
Linus Torvalds1da177e2005-04-16 15:20:36 -07003864 if (is_dir)
3865 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3866 else
3867 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
Al Viro123df292009-12-25 04:57:57 -05003868 if (!error)
3869 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04003870 new_dentry->d_inode, old_dentry);
Robert Love0eeca282005-07-12 17:06:03 -04003871 fsnotify_oldname_free(old_name);
3872
Linus Torvalds1da177e2005-04-16 15:20:36 -07003873 return error;
3874}
3875
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003876SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3877 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003878{
Al Viro2ad94ae2008-07-21 09:32:51 -04003879 struct dentry *old_dir, *new_dir;
3880 struct dentry *old_dentry, *new_dentry;
3881 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003882 struct nameidata oldnd, newnd;
Jeff Layton91a27b22012-10-10 15:25:28 -04003883 struct filename *from;
3884 struct filename *to;
Jeff Laytonc6a94282012-12-11 12:10:10 -05003885 unsigned int lookup_flags = 0;
3886 bool should_retry = false;
Al Viro2ad94ae2008-07-21 09:32:51 -04003887 int error;
Jeff Laytonc6a94282012-12-11 12:10:10 -05003888retry:
3889 from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04003890 if (IS_ERR(from)) {
3891 error = PTR_ERR(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003892 goto exit;
Jeff Layton91a27b22012-10-10 15:25:28 -04003893 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003894
Jeff Laytonc6a94282012-12-11 12:10:10 -05003895 to = user_path_parent(newdfd, newname, &newnd, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04003896 if (IS_ERR(to)) {
3897 error = PTR_ERR(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003898 goto exit1;
Jeff Layton91a27b22012-10-10 15:25:28 -04003899 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003900
3901 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08003902 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003903 goto exit2;
3904
Jan Blunck4ac91372008-02-14 19:34:32 -08003905 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003906 error = -EBUSY;
3907 if (oldnd.last_type != LAST_NORM)
3908 goto exit2;
3909
Jan Blunck4ac91372008-02-14 19:34:32 -08003910 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003911 if (newnd.last_type != LAST_NORM)
3912 goto exit2;
3913
Jan Karac30dabf2012-06-12 16:20:30 +02003914 error = mnt_want_write(oldnd.path.mnt);
3915 if (error)
3916 goto exit2;
3917
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003918 oldnd.flags &= ~LOOKUP_PARENT;
3919 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09003920 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003921
Linus Torvalds1da177e2005-04-16 15:20:36 -07003922 trap = lock_rename(new_dir, old_dir);
3923
Christoph Hellwig49705b72005-11-08 21:35:06 -08003924 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003925 error = PTR_ERR(old_dentry);
3926 if (IS_ERR(old_dentry))
3927 goto exit3;
3928 /* source must exist */
3929 error = -ENOENT;
3930 if (!old_dentry->d_inode)
3931 goto exit4;
3932 /* unless the source is a directory trailing slashes give -ENOTDIR */
3933 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3934 error = -ENOTDIR;
3935 if (oldnd.last.name[oldnd.last.len])
3936 goto exit4;
3937 if (newnd.last.name[newnd.last.len])
3938 goto exit4;
3939 }
3940 /* source should not be ancestor of target */
3941 error = -EINVAL;
3942 if (old_dentry == trap)
3943 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08003944 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003945 error = PTR_ERR(new_dentry);
3946 if (IS_ERR(new_dentry))
3947 goto exit4;
3948 /* target should not be an ancestor of source */
3949 error = -ENOTEMPTY;
3950 if (new_dentry == trap)
3951 goto exit5;
3952
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003953 error = security_path_rename(&oldnd.path, old_dentry,
3954 &newnd.path, new_dentry);
3955 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003956 goto exit5;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003957 error = vfs_rename(old_dir->d_inode, old_dentry,
3958 new_dir->d_inode, new_dentry);
3959exit5:
3960 dput(new_dentry);
3961exit4:
3962 dput(old_dentry);
3963exit3:
3964 unlock_rename(new_dir, old_dir);
Jan Karac30dabf2012-06-12 16:20:30 +02003965 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003966exit2:
Jeff Laytonc6a94282012-12-11 12:10:10 -05003967 if (retry_estale(error, lookup_flags))
3968 should_retry = true;
Jan Blunck1d957f92008-02-14 19:34:35 -08003969 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04003970 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003971exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003972 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003973 putname(from);
Jeff Laytonc6a94282012-12-11 12:10:10 -05003974 if (should_retry) {
3975 should_retry = false;
3976 lookup_flags |= LOOKUP_REVAL;
3977 goto retry;
3978 }
Al Viro2ad94ae2008-07-21 09:32:51 -04003979exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003980 return error;
3981}
3982
Heiko Carstensa26eab22009-01-14 14:14:17 +01003983SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003984{
3985 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3986}
3987
Linus Torvalds1da177e2005-04-16 15:20:36 -07003988int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3989{
3990 int len;
3991
3992 len = PTR_ERR(link);
3993 if (IS_ERR(link))
3994 goto out;
3995
3996 len = strlen(link);
3997 if (len > (unsigned) buflen)
3998 len = buflen;
3999 if (copy_to_user(buffer, link, len))
4000 len = -EFAULT;
4001out:
4002 return len;
4003}
4004
4005/*
4006 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4007 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4008 * using) it for any given inode is up to filesystem.
4009 */
4010int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4011{
4012 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004013 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07004014 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004015
Linus Torvalds1da177e2005-04-16 15:20:36 -07004016 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004017 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07004018 if (IS_ERR(cookie))
4019 return PTR_ERR(cookie);
4020
4021 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
4022 if (dentry->d_inode->i_op->put_link)
4023 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
4024 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004025}
4026
4027int vfs_follow_link(struct nameidata *nd, const char *link)
4028{
4029 return __vfs_follow_link(nd, link);
4030}
4031
4032/* get the link contents into pagecache */
4033static char *page_getlink(struct dentry * dentry, struct page **ppage)
4034{
Duane Griffinebd09ab2008-12-19 20:47:12 +00004035 char *kaddr;
4036 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004037 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07004038 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004039 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07004040 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004041 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00004042 kaddr = kmap(page);
4043 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4044 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004045}
4046
4047int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4048{
4049 struct page *page = NULL;
4050 char *s = page_getlink(dentry, &page);
4051 int res = vfs_readlink(dentry,buffer,buflen,s);
4052 if (page) {
4053 kunmap(page);
4054 page_cache_release(page);
4055 }
4056 return res;
4057}
4058
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004059void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004060{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004061 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004062 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004063 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004064}
4065
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004066void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004067{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004068 struct page *page = cookie;
4069
4070 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004071 kunmap(page);
4072 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004073 }
4074}
4075
Nick Piggin54566b22009-01-04 12:00:53 -08004076/*
4077 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4078 */
4079int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004080{
4081 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004082 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07004083 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08004084 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004085 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08004086 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4087 if (nofs)
4088 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004089
NeilBrown7e53cac2006-03-25 03:07:57 -08004090retry:
Nick Pigginafddba42007-10-16 01:25:01 -07004091 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08004092 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004093 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07004094 goto fail;
4095
Cong Wange8e3c3d2011-11-25 23:14:27 +08004096 kaddr = kmap_atomic(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004097 memcpy(kaddr, symname, len-1);
Cong Wange8e3c3d2011-11-25 23:14:27 +08004098 kunmap_atomic(kaddr);
Nick Pigginafddba42007-10-16 01:25:01 -07004099
4100 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4101 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004102 if (err < 0)
4103 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07004104 if (err < len-1)
4105 goto retry;
4106
Linus Torvalds1da177e2005-04-16 15:20:36 -07004107 mark_inode_dirty(inode);
4108 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004109fail:
4110 return err;
4111}
4112
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004113int page_symlink(struct inode *inode, const char *symname, int len)
4114{
4115 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08004116 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004117}
4118
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08004119const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004120 .readlink = generic_readlink,
4121 .follow_link = page_follow_link_light,
4122 .put_link = page_put_link,
4123};
4124
Al Viro2d8f3032008-07-22 09:59:21 -04004125EXPORT_SYMBOL(user_path_at);
David Howellscc53ce52011-01-14 18:45:26 +00004126EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004127EXPORT_SYMBOL(follow_down);
4128EXPORT_SYMBOL(follow_up);
Al Virof6d2ac52012-08-26 12:55:54 -04004129EXPORT_SYMBOL(get_write_access); /* nfsd */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004130EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004131EXPORT_SYMBOL(lookup_one_len);
4132EXPORT_SYMBOL(page_follow_link_light);
4133EXPORT_SYMBOL(page_put_link);
4134EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004135EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004136EXPORT_SYMBOL(page_symlink);
4137EXPORT_SYMBOL(page_symlink_inode_operations);
Al Virod1811462008-08-02 00:49:18 -04004138EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07004139EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04004140EXPORT_SYMBOL(inode_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004141EXPORT_SYMBOL(unlock_rename);
4142EXPORT_SYMBOL(vfs_create);
4143EXPORT_SYMBOL(vfs_follow_link);
4144EXPORT_SYMBOL(vfs_link);
4145EXPORT_SYMBOL(vfs_mkdir);
4146EXPORT_SYMBOL(vfs_mknod);
4147EXPORT_SYMBOL(generic_permission);
4148EXPORT_SYMBOL(vfs_readlink);
4149EXPORT_SYMBOL(vfs_rename);
4150EXPORT_SYMBOL(vfs_rmdir);
4151EXPORT_SYMBOL(vfs_symlink);
4152EXPORT_SYMBOL(vfs_unlink);
4153EXPORT_SYMBOL(dentry_unhash);
4154EXPORT_SYMBOL(generic_readlink);