blob: 2c30c84d4ea1da5e17381caeea6de5ae27e47a0e [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
Linus Torvalds15570082013-09-02 11:38:06 -0700497/*
498 * When we move over from the RCU domain to properly refcounted
499 * long-lived dentries, we need to check the sequence numbers
500 * we got before lookup very carefully.
501 *
502 * We cannot blindly increment a dentry refcount - even if it
503 * is not locked - if it is zero, because it may have gone
504 * through the final d_kill() logic already.
505 *
506 * So for a zero refcount, we need to get the spinlock (which is
507 * safe even for a dead dentry because the de-allocation is
508 * RCU-delayed), and check the sequence count under the lock.
509 *
510 * Once we have checked the sequence count, we know it is live,
511 * and since we hold the spinlock it cannot die from under us.
512 *
513 * In contrast, if the reference count wasn't zero, we can just
514 * increment the lockref without having to take the spinlock.
515 * Even if the sequence number ends up being stale, we haven't
516 * gone through the final dput() and killed the dentry yet.
517 */
518static inline int d_rcu_to_refcount(struct dentry *dentry, seqcount_t *validate, unsigned seq)
519{
520 int gotref;
521
522 gotref = lockref_get_or_lock(&dentry->d_lockref);
523
524 /* Does the sequence number still match? */
525 if (read_seqcount_retry(validate, seq)) {
526 if (gotref)
527 dput(dentry);
528 else
529 spin_unlock(&dentry->d_lock);
530 return -ECHILD;
531 }
532
533 /* Get the ref now, if we couldn't get it originally */
534 if (!gotref) {
535 dentry->d_lockref.count++;
536 spin_unlock(&dentry->d_lock);
537 }
538 return 0;
539}
540
Nick Piggin31e6b012011-01-07 17:49:52 +1100541/**
Al Viro19660af2011-03-25 10:32:48 -0400542 * unlazy_walk - try to switch to ref-walk mode.
543 * @nd: nameidata pathwalk data
544 * @dentry: child of nd->path.dentry or NULL
Randy Dunlap39191622011-01-08 19:36:21 -0800545 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100546 *
Al Viro19660af2011-03-25 10:32:48 -0400547 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
548 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
549 * @nd or NULL. Must be called from rcu-walk context.
Nick Piggin31e6b012011-01-07 17:49:52 +1100550 */
Al Viro19660af2011-03-25 10:32:48 -0400551static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
Nick Piggin31e6b012011-01-07 17:49:52 +1100552{
553 struct fs_struct *fs = current->fs;
554 struct dentry *parent = nd->path.dentry;
Al Viro5b6ca022011-03-09 23:04:47 -0500555 int want_root = 0;
Nick Piggin31e6b012011-01-07 17:49:52 +1100556
557 BUG_ON(!(nd->flags & LOOKUP_RCU));
Al Viro5b6ca022011-03-09 23:04:47 -0500558 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
559 want_root = 1;
Nick Piggin31e6b012011-01-07 17:49:52 +1100560 spin_lock(&fs->lock);
561 if (nd->root.mnt != fs->root.mnt ||
562 nd->root.dentry != fs->root.dentry)
563 goto err_root;
564 }
Linus Torvalds15570082013-09-02 11:38:06 -0700565
566 /*
567 * For a negative lookup, the lookup sequence point is the parents
568 * sequence point, and it only needs to revalidate the parent dentry.
569 *
570 * For a positive lookup, we need to move both the parent and the
571 * dentry from the RCU domain to be properly refcounted. And the
572 * sequence number in the dentry validates *both* dentry counters,
573 * since we checked the sequence number of the parent after we got
574 * the child sequence number. So we know the parent must still
575 * be valid if the child sequence number is still valid.
576 */
Al Viro19660af2011-03-25 10:32:48 -0400577 if (!dentry) {
Linus Torvalds15570082013-09-02 11:38:06 -0700578 if (d_rcu_to_refcount(parent, &parent->d_seq, nd->seq) < 0)
579 goto err_root;
Al Viro19660af2011-03-25 10:32:48 -0400580 BUG_ON(nd->inode != parent->d_inode);
581 } else {
Linus Torvalds15570082013-09-02 11:38:06 -0700582 if (d_rcu_to_refcount(dentry, &dentry->d_seq, nd->seq) < 0)
583 goto err_root;
584 if (d_rcu_to_refcount(parent, &dentry->d_seq, nd->seq) < 0)
Al Viro94c0d4e2011-07-12 21:40:23 -0400585 goto err_parent;
Al Viro19660af2011-03-25 10:32:48 -0400586 }
Al Viro5b6ca022011-03-09 23:04:47 -0500587 if (want_root) {
Nick Piggin31e6b012011-01-07 17:49:52 +1100588 path_get(&nd->root);
589 spin_unlock(&fs->lock);
590 }
591 mntget(nd->path.mnt);
592
Al Viro32a79912012-07-18 20:43:19 +0400593 unlock_rcu_walk();
Nick Piggin31e6b012011-01-07 17:49:52 +1100594 nd->flags &= ~LOOKUP_RCU;
595 return 0;
Al Viro19660af2011-03-25 10:32:48 -0400596
Al Viro19660af2011-03-25 10:32:48 -0400597err_parent:
Linus Torvalds15570082013-09-02 11:38:06 -0700598 dput(dentry);
Nick Piggin31e6b012011-01-07 17:49:52 +1100599err_root:
Al Viro5b6ca022011-03-09 23:04:47 -0500600 if (want_root)
Nick Piggin31e6b012011-01-07 17:49:52 +1100601 spin_unlock(&fs->lock);
602 return -ECHILD;
603}
604
Al Viro4ce16ef32012-06-10 16:10:59 -0400605static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
Nick Piggin34286d62011-01-07 17:49:57 +1100606{
Al Viro4ce16ef32012-06-10 16:10:59 -0400607 return dentry->d_op->d_revalidate(dentry, flags);
Nick Piggin34286d62011-01-07 17:49:57 +1100608}
609
Al Viro9f1fafe2011-03-25 11:00:12 -0400610/**
611 * complete_walk - successful completion of path walk
612 * @nd: pointer nameidata
Jeff Layton39159de2009-12-07 12:01:50 -0500613 *
Al Viro9f1fafe2011-03-25 11:00:12 -0400614 * If we had been in RCU mode, drop out of it and legitimize nd->path.
615 * Revalidate the final result, unless we'd already done that during
616 * the path walk or the filesystem doesn't ask for it. Return 0 on
617 * success, -error on failure. In case of failure caller does not
618 * need to drop nd->path.
Jeff Layton39159de2009-12-07 12:01:50 -0500619 */
Al Viro9f1fafe2011-03-25 11:00:12 -0400620static int complete_walk(struct nameidata *nd)
Jeff Layton39159de2009-12-07 12:01:50 -0500621{
Al Viro16c2cd72011-02-22 15:50:10 -0500622 struct dentry *dentry = nd->path.dentry;
Jeff Layton39159de2009-12-07 12:01:50 -0500623 int status;
Jeff Layton39159de2009-12-07 12:01:50 -0500624
Al Viro9f1fafe2011-03-25 11:00:12 -0400625 if (nd->flags & LOOKUP_RCU) {
626 nd->flags &= ~LOOKUP_RCU;
627 if (!(nd->flags & LOOKUP_ROOT))
628 nd->root.mnt = NULL;
Linus Torvalds15570082013-09-02 11:38:06 -0700629
630 if (d_rcu_to_refcount(dentry, &dentry->d_seq, nd->seq) < 0) {
Al Viro32a79912012-07-18 20:43:19 +0400631 unlock_rcu_walk();
Al Viro9f1fafe2011-03-25 11:00:12 -0400632 return -ECHILD;
633 }
Al Viro9f1fafe2011-03-25 11:00:12 -0400634 mntget(nd->path.mnt);
Al Viro32a79912012-07-18 20:43:19 +0400635 unlock_rcu_walk();
Al Viro9f1fafe2011-03-25 11:00:12 -0400636 }
637
Al Viro16c2cd72011-02-22 15:50:10 -0500638 if (likely(!(nd->flags & LOOKUP_JUMPED)))
Jeff Layton39159de2009-12-07 12:01:50 -0500639 return 0;
640
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500641 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
Al Viro16c2cd72011-02-22 15:50:10 -0500642 return 0;
643
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500644 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
Jeff Layton39159de2009-12-07 12:01:50 -0500645 if (status > 0)
646 return 0;
647
Al Viro16c2cd72011-02-22 15:50:10 -0500648 if (!status)
Jeff Layton39159de2009-12-07 12:01:50 -0500649 status = -ESTALE;
Al Viro16c2cd72011-02-22 15:50:10 -0500650
Al Viro9f1fafe2011-03-25 11:00:12 -0400651 path_put(&nd->path);
Jeff Layton39159de2009-12-07 12:01:50 -0500652 return status;
653}
654
Al Viro2a737872009-04-07 11:49:53 -0400655static __always_inline void set_root(struct nameidata *nd)
656{
Miklos Szeredif7ad3c62010-08-10 11:41:36 +0200657 if (!nd->root.mnt)
658 get_fs_root(current->fs, &nd->root);
Al Viro2a737872009-04-07 11:49:53 -0400659}
660
Al Viro6de88d72009-08-09 01:41:57 +0400661static int link_path_walk(const char *, struct nameidata *);
662
Nick Piggin31e6b012011-01-07 17:49:52 +1100663static __always_inline void set_root_rcu(struct nameidata *nd)
664{
665 if (!nd->root.mnt) {
666 struct fs_struct *fs = current->fs;
Nick Pigginc28cc362011-01-07 17:49:53 +1100667 unsigned seq;
668
669 do {
670 seq = read_seqcount_begin(&fs->seq);
671 nd->root = fs->root;
Tim Chenc1530012011-04-15 11:39:29 -0700672 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
Nick Pigginc28cc362011-01-07 17:49:53 +1100673 } while (read_seqcount_retry(&fs->seq, seq));
Nick Piggin31e6b012011-01-07 17:49:52 +1100674 }
675}
676
Arjan van de Venf1662352006-01-14 13:21:31 -0800677static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678{
Nick Piggin31e6b012011-01-07 17:49:52 +1100679 int ret;
680
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 if (IS_ERR(link))
682 goto fail;
683
684 if (*link == '/') {
Al Viro2a737872009-04-07 11:49:53 -0400685 set_root(nd);
Jan Blunck1d957f92008-02-14 19:34:35 -0800686 path_put(&nd->path);
Al Viro2a737872009-04-07 11:49:53 -0400687 nd->path = nd->root;
688 path_get(&nd->root);
Al Viro16c2cd72011-02-22 15:50:10 -0500689 nd->flags |= LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100691 nd->inode = nd->path.dentry->d_inode;
Christoph Hellwigb4091d52008-11-05 15:07:21 +0100692
Nick Piggin31e6b012011-01-07 17:49:52 +1100693 ret = link_path_walk(link, nd);
694 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695fail:
Jan Blunck1d957f92008-02-14 19:34:35 -0800696 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 return PTR_ERR(link);
698}
699
Jan Blunck1d957f92008-02-14 19:34:35 -0800700static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700701{
702 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800703 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700704 mntput(path->mnt);
705}
706
Nick Piggin7b9337a2011-01-14 08:42:43 +0000707static inline void path_to_nameidata(const struct path *path,
708 struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700709{
Nick Piggin31e6b012011-01-07 17:49:52 +1100710 if (!(nd->flags & LOOKUP_RCU)) {
711 dput(nd->path.dentry);
712 if (nd->path.mnt != path->mnt)
713 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800714 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100715 nd->path.mnt = path->mnt;
Jan Blunck4ac91372008-02-14 19:34:32 -0800716 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700717}
718
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400719/*
720 * Helper to directly jump to a known parsed path from ->follow_link,
721 * caller must have taken a reference to path beforehand.
722 */
723void nd_jump_link(struct nameidata *nd, struct path *path)
724{
725 path_put(&nd->path);
726
727 nd->path = *path;
728 nd->inode = nd->path.dentry->d_inode;
729 nd->flags |= LOOKUP_JUMPED;
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400730}
731
Al Viro574197e2011-03-14 22:20:34 -0400732static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
733{
734 struct inode *inode = link->dentry->d_inode;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400735 if (inode->i_op->put_link)
Al Viro574197e2011-03-14 22:20:34 -0400736 inode->i_op->put_link(link->dentry, nd, cookie);
737 path_put(link);
738}
739
Linus Torvalds561ec642012-10-26 10:05:07 -0700740int sysctl_protected_symlinks __read_mostly = 0;
741int sysctl_protected_hardlinks __read_mostly = 0;
Kees Cook800179c2012-07-25 17:29:07 -0700742
743/**
744 * may_follow_link - Check symlink following for unsafe situations
745 * @link: The path of the symlink
Randy Dunlap55852632012-08-18 17:39:25 -0700746 * @nd: nameidata pathwalk data
Kees Cook800179c2012-07-25 17:29:07 -0700747 *
748 * In the case of the sysctl_protected_symlinks sysctl being enabled,
749 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
750 * in a sticky world-writable directory. This is to protect privileged
751 * processes from failing races against path names that may change out
752 * from under them by way of other users creating malicious symlinks.
753 * It will permit symlinks to be followed only when outside a sticky
754 * world-writable directory, or when the uid of the symlink and follower
755 * match, or when the directory owner matches the symlink's owner.
756 *
757 * Returns 0 if following the symlink is allowed, -ve on error.
758 */
759static inline int may_follow_link(struct path *link, struct nameidata *nd)
760{
761 const struct inode *inode;
762 const struct inode *parent;
763
764 if (!sysctl_protected_symlinks)
765 return 0;
766
767 /* Allowed if owner and follower match. */
768 inode = link->dentry->d_inode;
Eric W. Biederman81abe272012-08-03 09:38:08 -0700769 if (uid_eq(current_cred()->fsuid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700770 return 0;
771
772 /* Allowed if parent directory not sticky and world-writable. */
773 parent = nd->path.dentry->d_inode;
774 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
775 return 0;
776
777 /* Allowed if parent directory and link owner match. */
Eric W. Biederman81abe272012-08-03 09:38:08 -0700778 if (uid_eq(parent->i_uid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700779 return 0;
780
Sasha Levinffd8d102012-10-04 19:56:40 -0400781 audit_log_link_denied("follow_link", link);
Kees Cook800179c2012-07-25 17:29:07 -0700782 path_put_conditional(link, nd);
783 path_put(&nd->path);
784 return -EACCES;
785}
786
787/**
788 * safe_hardlink_source - Check for safe hardlink conditions
789 * @inode: the source inode to hardlink from
790 *
791 * Return false if at least one of the following conditions:
792 * - inode is not a regular file
793 * - inode is setuid
794 * - inode is setgid and group-exec
795 * - access failure for read and write
796 *
797 * Otherwise returns true.
798 */
799static bool safe_hardlink_source(struct inode *inode)
800{
801 umode_t mode = inode->i_mode;
802
803 /* Special files should not get pinned to the filesystem. */
804 if (!S_ISREG(mode))
805 return false;
806
807 /* Setuid files should not get pinned to the filesystem. */
808 if (mode & S_ISUID)
809 return false;
810
811 /* Executable setgid files should not get pinned to the filesystem. */
812 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
813 return false;
814
815 /* Hardlinking to unreadable or unwritable sources is dangerous. */
816 if (inode_permission(inode, MAY_READ | MAY_WRITE))
817 return false;
818
819 return true;
820}
821
822/**
823 * may_linkat - Check permissions for creating a hardlink
824 * @link: the source to hardlink from
825 *
826 * Block hardlink when all of:
827 * - sysctl_protected_hardlinks enabled
828 * - fsuid does not match inode
829 * - hardlink source is unsafe (see safe_hardlink_source() above)
830 * - not CAP_FOWNER
831 *
832 * Returns 0 if successful, -ve on error.
833 */
834static int may_linkat(struct path *link)
835{
836 const struct cred *cred;
837 struct inode *inode;
838
839 if (!sysctl_protected_hardlinks)
840 return 0;
841
842 cred = current_cred();
843 inode = link->dentry->d_inode;
844
845 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
846 * otherwise, it must be a safe source.
847 */
Eric W. Biederman81abe272012-08-03 09:38:08 -0700848 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
Kees Cook800179c2012-07-25 17:29:07 -0700849 capable(CAP_FOWNER))
850 return 0;
851
Kees Cooka51d9ea2012-07-25 17:29:08 -0700852 audit_log_link_denied("linkat", link);
Kees Cook800179c2012-07-25 17:29:07 -0700853 return -EPERM;
854}
855
Al Virodef4af32009-12-26 08:37:05 -0500856static __always_inline int
Al Viro574197e2011-03-14 22:20:34 -0400857follow_link(struct path *link, struct nameidata *nd, void **p)
Ian Kent051d3812006-03-27 01:14:53 -0800858{
Nick Piggin7b9337a2011-01-14 08:42:43 +0000859 struct dentry *dentry = link->dentry;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400860 int error;
861 char *s;
Ian Kent051d3812006-03-27 01:14:53 -0800862
Al Viro844a3912011-02-15 00:38:26 -0500863 BUG_ON(nd->flags & LOOKUP_RCU);
864
Al Viro0e794582011-03-16 02:45:02 -0400865 if (link->mnt == nd->path.mnt)
866 mntget(link->mnt);
867
Al Viro6d7b5aa2012-06-10 04:15:17 -0400868 error = -ELOOP;
869 if (unlikely(current->total_link_count >= 40))
870 goto out_put_nd_path;
871
Al Viro574197e2011-03-14 22:20:34 -0400872 cond_resched();
873 current->total_link_count++;
874
Al Viro68ac1232012-03-15 08:21:57 -0400875 touch_atime(link);
Ian Kent051d3812006-03-27 01:14:53 -0800876 nd_set_link(nd, NULL);
877
Al Viro36f3b4f2011-02-22 21:24:38 -0500878 error = security_inode_follow_link(link->dentry, nd);
Al Viro6d7b5aa2012-06-10 04:15:17 -0400879 if (error)
880 goto out_put_nd_path;
Al Viro36f3b4f2011-02-22 21:24:38 -0500881
Al Viro86acdca12009-12-22 23:45:11 -0500882 nd->last_type = LAST_BIND;
Al Virodef4af32009-12-26 08:37:05 -0500883 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
884 error = PTR_ERR(*p);
Al Viro6d7b5aa2012-06-10 04:15:17 -0400885 if (IS_ERR(*p))
Christoph Hellwig408ef012012-06-18 10:47:03 -0400886 goto out_put_nd_path;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400887
888 error = 0;
889 s = nd_get_link(nd);
890 if (s) {
891 error = __vfs_follow_link(nd, s);
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400892 if (unlikely(error))
893 put_link(nd, link, *p);
Ian Kent051d3812006-03-27 01:14:53 -0800894 }
Al Viro6d7b5aa2012-06-10 04:15:17 -0400895
896 return error;
897
898out_put_nd_path:
Arnd Bergmann98f6ef62012-10-11 13:20:00 +0000899 *p = NULL;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400900 path_put(&nd->path);
Al Viro6d7b5aa2012-06-10 04:15:17 -0400901 path_put(link);
Ian Kent051d3812006-03-27 01:14:53 -0800902 return error;
903}
904
Nick Piggin31e6b012011-01-07 17:49:52 +1100905static int follow_up_rcu(struct path *path)
906{
Al Viro0714a532011-11-24 22:19:58 -0500907 struct mount *mnt = real_mount(path->mnt);
908 struct mount *parent;
Nick Piggin31e6b012011-01-07 17:49:52 +1100909 struct dentry *mountpoint;
910
Al Viro0714a532011-11-24 22:19:58 -0500911 parent = mnt->mnt_parent;
912 if (&parent->mnt == path->mnt)
Nick Piggin31e6b012011-01-07 17:49:52 +1100913 return 0;
Al Viroa73324d2011-11-24 22:25:07 -0500914 mountpoint = mnt->mnt_mountpoint;
Nick Piggin31e6b012011-01-07 17:49:52 +1100915 path->dentry = mountpoint;
Al Viro0714a532011-11-24 22:19:58 -0500916 path->mnt = &parent->mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +1100917 return 1;
918}
919
David Howellsf015f1262012-06-25 12:55:28 +0100920/*
921 * follow_up - Find the mountpoint of path's vfsmount
922 *
923 * Given a path, find the mountpoint of its source file system.
924 * Replace @path with the path of the mountpoint in the parent mount.
925 * Up is towards /.
926 *
927 * Return 1 if we went up a level and 0 if we were already at the
928 * root.
929 */
Al Virobab77eb2009-04-18 03:26:48 -0400930int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931{
Al Viro0714a532011-11-24 22:19:58 -0500932 struct mount *mnt = real_mount(path->mnt);
933 struct mount *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +1000935
Andi Kleen962830d2012-05-08 13:32:02 +0930936 br_read_lock(&vfsmount_lock);
Al Viro0714a532011-11-24 22:19:58 -0500937 parent = mnt->mnt_parent;
Al Viro3c0a6162012-07-18 17:32:50 +0400938 if (parent == mnt) {
Andi Kleen962830d2012-05-08 13:32:02 +0930939 br_read_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 return 0;
941 }
Al Viro0714a532011-11-24 22:19:58 -0500942 mntget(&parent->mnt);
Al Viroa73324d2011-11-24 22:25:07 -0500943 mountpoint = dget(mnt->mnt_mountpoint);
Andi Kleen962830d2012-05-08 13:32:02 +0930944 br_read_unlock(&vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400945 dput(path->dentry);
946 path->dentry = mountpoint;
947 mntput(path->mnt);
Al Viro0714a532011-11-24 22:19:58 -0500948 path->mnt = &parent->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 return 1;
950}
951
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100952/*
David Howells9875cf82011-01-14 18:45:21 +0000953 * Perform an automount
954 * - return -EISDIR to tell follow_managed() to stop and return the path we
955 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 */
David Howells9875cf82011-01-14 18:45:21 +0000957static int follow_automount(struct path *path, unsigned flags,
958 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +1100959{
David Howells9875cf82011-01-14 18:45:21 +0000960 struct vfsmount *mnt;
David Howellsea5b7782011-01-14 19:10:03 +0000961 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +1100962
David Howells9875cf82011-01-14 18:45:21 +0000963 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
964 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -0700965
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200966 /* We don't want to mount if someone's just doing a stat -
967 * unless they're stat'ing a directory and appended a '/' to
968 * the name.
969 *
970 * We do, however, want to mount if someone wants to open or
971 * create a file of any type under the mountpoint, wants to
972 * traverse through the mountpoint or wants to open the
973 * mounted directory. Also, autofs may mark negative dentries
974 * as being automount points. These will need the attentions
975 * of the daemon to instantiate them before they can be used.
David Howells9875cf82011-01-14 18:45:21 +0000976 */
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200977 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
Linus Torvaldsd94c1772011-09-26 17:44:55 -0700978 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200979 path->dentry->d_inode)
980 return -EISDIR;
981
David Howells9875cf82011-01-14 18:45:21 +0000982 current->total_link_count++;
983 if (current->total_link_count >= 40)
984 return -ELOOP;
985
986 mnt = path->dentry->d_op->d_automount(path);
987 if (IS_ERR(mnt)) {
988 /*
989 * The filesystem is allowed to return -EISDIR here to indicate
990 * it doesn't want to automount. For instance, autofs would do
991 * this so that its userspace daemon can mount on this dentry.
992 *
993 * However, we can only permit this if it's a terminal point in
994 * the path being looked up; if it wasn't then the remainder of
995 * the path is inaccessible and we should say so.
996 */
Al Viro49084c32011-06-25 21:59:52 -0400997 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
David Howells9875cf82011-01-14 18:45:21 +0000998 return -EREMOTE;
999 return PTR_ERR(mnt);
1000 }
David Howellsea5b7782011-01-14 19:10:03 +00001001
David Howells9875cf82011-01-14 18:45:21 +00001002 if (!mnt) /* mount collision */
1003 return 0;
1004
Al Viro8aef1882011-06-16 15:10:06 +01001005 if (!*need_mntput) {
1006 /* lock_mount() may release path->mnt on error */
1007 mntget(path->mnt);
1008 *need_mntput = true;
1009 }
Al Viro19a167a2011-01-17 01:35:23 -05001010 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +00001011
David Howellsea5b7782011-01-14 19:10:03 +00001012 switch (err) {
1013 case -EBUSY:
1014 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -05001015 return 0;
David Howellsea5b7782011-01-14 19:10:03 +00001016 case 0:
Al Viro8aef1882011-06-16 15:10:06 +01001017 path_put(path);
David Howellsea5b7782011-01-14 19:10:03 +00001018 path->mnt = mnt;
1019 path->dentry = dget(mnt->mnt_root);
David Howellsea5b7782011-01-14 19:10:03 +00001020 return 0;
Al Viro19a167a2011-01-17 01:35:23 -05001021 default:
1022 return err;
David Howellsea5b7782011-01-14 19:10:03 +00001023 }
Al Viro19a167a2011-01-17 01:35:23 -05001024
David Howells9875cf82011-01-14 18:45:21 +00001025}
1026
1027/*
1028 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +00001029 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +00001030 * - Flagged as mountpoint
1031 * - Flagged as automount point
1032 *
1033 * This may only be called in refwalk mode.
1034 *
1035 * Serialization is taken care of in namespace.c
1036 */
1037static int follow_managed(struct path *path, unsigned flags)
1038{
Al Viro8aef1882011-06-16 15:10:06 +01001039 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
David Howells9875cf82011-01-14 18:45:21 +00001040 unsigned managed;
1041 bool need_mntput = false;
Al Viro8aef1882011-06-16 15:10:06 +01001042 int ret = 0;
David Howells9875cf82011-01-14 18:45:21 +00001043
1044 /* Given that we're not holding a lock here, we retain the value in a
1045 * local variable for each dentry as we look at it so that we don't see
1046 * the components of that value change under us */
1047 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1048 managed &= DCACHE_MANAGED_DENTRY,
1049 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +00001050 /* Allow the filesystem to manage the transit without i_mutex
1051 * being held. */
1052 if (managed & DCACHE_MANAGE_TRANSIT) {
1053 BUG_ON(!path->dentry->d_op);
1054 BUG_ON(!path->dentry->d_op->d_manage);
Al Viro1aed3e42011-03-18 09:09:02 -04001055 ret = path->dentry->d_op->d_manage(path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001056 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001057 break;
David Howellscc53ce52011-01-14 18:45:26 +00001058 }
1059
David Howells9875cf82011-01-14 18:45:21 +00001060 /* Transit to a mounted filesystem. */
1061 if (managed & DCACHE_MOUNTED) {
1062 struct vfsmount *mounted = lookup_mnt(path);
1063 if (mounted) {
1064 dput(path->dentry);
1065 if (need_mntput)
1066 mntput(path->mnt);
1067 path->mnt = mounted;
1068 path->dentry = dget(mounted->mnt_root);
1069 need_mntput = true;
1070 continue;
1071 }
1072
1073 /* Something is mounted on this dentry in another
1074 * namespace and/or whatever was mounted there in this
1075 * namespace got unmounted before we managed to get the
1076 * vfsmount_lock */
1077 }
1078
1079 /* Handle an automount point */
1080 if (managed & DCACHE_NEED_AUTOMOUNT) {
1081 ret = follow_automount(path, flags, &need_mntput);
1082 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001083 break;
David Howells9875cf82011-01-14 18:45:21 +00001084 continue;
1085 }
1086
1087 /* We didn't change the current path point */
1088 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 }
Al Viro8aef1882011-06-16 15:10:06 +01001090
1091 if (need_mntput && path->mnt == mnt)
1092 mntput(path->mnt);
1093 if (ret == -EISDIR)
1094 ret = 0;
Al Viroa3fbbde2011-11-07 21:21:26 +00001095 return ret < 0 ? ret : need_mntput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096}
1097
David Howellscc53ce52011-01-14 18:45:26 +00001098int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099{
1100 struct vfsmount *mounted;
1101
Al Viro1c755af2009-04-18 14:06:57 -04001102 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -04001104 dput(path->dentry);
1105 mntput(path->mnt);
1106 path->mnt = mounted;
1107 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 return 1;
1109 }
1110 return 0;
1111}
1112
Ian Kent62a73752011-03-25 01:51:02 +08001113static inline bool managed_dentry_might_block(struct dentry *dentry)
1114{
1115 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
1116 dentry->d_op->d_manage(dentry, true) < 0);
1117}
1118
David Howells9875cf82011-01-14 18:45:21 +00001119/*
Al Viro287548e2011-05-27 06:50:06 -04001120 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1121 * we meet a managed dentry that would need blocking.
David Howells9875cf82011-01-14 18:45:21 +00001122 */
1123static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
Al Viro287548e2011-05-27 06:50:06 -04001124 struct inode **inode)
David Howells9875cf82011-01-14 18:45:21 +00001125{
Ian Kent62a73752011-03-25 01:51:02 +08001126 for (;;) {
Al Viroc7105362011-11-24 18:22:03 -05001127 struct mount *mounted;
Ian Kent62a73752011-03-25 01:51:02 +08001128 /*
1129 * Don't forget we might have a non-mountpoint managed dentry
1130 * that wants to block transit.
1131 */
Al Viro287548e2011-05-27 06:50:06 -04001132 if (unlikely(managed_dentry_might_block(path->dentry)))
David Howellsab909112011-01-14 18:46:51 +00001133 return false;
Ian Kent62a73752011-03-25 01:51:02 +08001134
1135 if (!d_mountpoint(path->dentry))
1136 break;
1137
David Howells9875cf82011-01-14 18:45:21 +00001138 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
1139 if (!mounted)
1140 break;
Al Viroc7105362011-11-24 18:22:03 -05001141 path->mnt = &mounted->mnt;
1142 path->dentry = mounted->mnt.mnt_root;
Al Viroa3fbbde2011-11-07 21:21:26 +00001143 nd->flags |= LOOKUP_JUMPED;
David Howells9875cf82011-01-14 18:45:21 +00001144 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
Linus Torvalds59430262011-07-18 15:43:29 -07001145 /*
1146 * Update the inode too. We don't need to re-check the
1147 * dentry sequence number here after this d_inode read,
1148 * because a mount-point is always pinned.
1149 */
1150 *inode = path->dentry->d_inode;
David Howells9875cf82011-01-14 18:45:21 +00001151 }
David Howells9875cf82011-01-14 18:45:21 +00001152 return true;
1153}
1154
Al Virodea39372011-05-27 06:53:39 -04001155static void follow_mount_rcu(struct nameidata *nd)
Al Viro287548e2011-05-27 06:50:06 -04001156{
Al Virodea39372011-05-27 06:53:39 -04001157 while (d_mountpoint(nd->path.dentry)) {
Al Viroc7105362011-11-24 18:22:03 -05001158 struct mount *mounted;
Al Virodea39372011-05-27 06:53:39 -04001159 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
Al Viro287548e2011-05-27 06:50:06 -04001160 if (!mounted)
1161 break;
Al Viroc7105362011-11-24 18:22:03 -05001162 nd->path.mnt = &mounted->mnt;
1163 nd->path.dentry = mounted->mnt.mnt_root;
Al Virodea39372011-05-27 06:53:39 -04001164 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro287548e2011-05-27 06:50:06 -04001165 }
1166}
1167
Nick Piggin31e6b012011-01-07 17:49:52 +11001168static int follow_dotdot_rcu(struct nameidata *nd)
1169{
Nick Piggin31e6b012011-01-07 17:49:52 +11001170 set_root_rcu(nd);
1171
David Howells9875cf82011-01-14 18:45:21 +00001172 while (1) {
Nick Piggin31e6b012011-01-07 17:49:52 +11001173 if (nd->path.dentry == nd->root.dentry &&
1174 nd->path.mnt == nd->root.mnt) {
1175 break;
1176 }
1177 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1178 struct dentry *old = nd->path.dentry;
1179 struct dentry *parent = old->d_parent;
1180 unsigned seq;
1181
1182 seq = read_seqcount_begin(&parent->d_seq);
1183 if (read_seqcount_retry(&old->d_seq, nd->seq))
Al Viroef7562d2011-03-04 14:35:59 -05001184 goto failed;
Nick Piggin31e6b012011-01-07 17:49:52 +11001185 nd->path.dentry = parent;
1186 nd->seq = seq;
1187 break;
1188 }
1189 if (!follow_up_rcu(&nd->path))
1190 break;
1191 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11001192 }
Al Virodea39372011-05-27 06:53:39 -04001193 follow_mount_rcu(nd);
1194 nd->inode = nd->path.dentry->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001195 return 0;
Al Viroef7562d2011-03-04 14:35:59 -05001196
1197failed:
1198 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001199 if (!(nd->flags & LOOKUP_ROOT))
1200 nd->root.mnt = NULL;
Al Viro32a79912012-07-18 20:43:19 +04001201 unlock_rcu_walk();
Al Viroef7562d2011-03-04 14:35:59 -05001202 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001203}
1204
David Howells9875cf82011-01-14 18:45:21 +00001205/*
David Howellscc53ce52011-01-14 18:45:26 +00001206 * Follow down to the covering mount currently visible to userspace. At each
1207 * point, the filesystem owning that dentry may be queried as to whether the
1208 * caller is permitted to proceed or not.
David Howellscc53ce52011-01-14 18:45:26 +00001209 */
Al Viro7cc90cc2011-03-18 09:04:20 -04001210int follow_down(struct path *path)
David Howellscc53ce52011-01-14 18:45:26 +00001211{
1212 unsigned managed;
1213 int ret;
1214
1215 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1216 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1217 /* Allow the filesystem to manage the transit without i_mutex
1218 * being held.
1219 *
1220 * We indicate to the filesystem if someone is trying to mount
1221 * something here. This gives autofs the chance to deny anyone
1222 * other than its daemon the right to mount on its
1223 * superstructure.
1224 *
1225 * The filesystem may sleep at this point.
1226 */
1227 if (managed & DCACHE_MANAGE_TRANSIT) {
1228 BUG_ON(!path->dentry->d_op);
1229 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +00001230 ret = path->dentry->d_op->d_manage(
Al Viro1aed3e42011-03-18 09:09:02 -04001231 path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001232 if (ret < 0)
1233 return ret == -EISDIR ? 0 : ret;
1234 }
1235
1236 /* Transit to a mounted filesystem. */
1237 if (managed & DCACHE_MOUNTED) {
1238 struct vfsmount *mounted = lookup_mnt(path);
1239 if (!mounted)
1240 break;
1241 dput(path->dentry);
1242 mntput(path->mnt);
1243 path->mnt = mounted;
1244 path->dentry = dget(mounted->mnt_root);
1245 continue;
1246 }
1247
1248 /* Don't handle automount points here */
1249 break;
1250 }
1251 return 0;
1252}
1253
1254/*
David Howells9875cf82011-01-14 18:45:21 +00001255 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1256 */
1257static void follow_mount(struct path *path)
1258{
1259 while (d_mountpoint(path->dentry)) {
1260 struct vfsmount *mounted = lookup_mnt(path);
1261 if (!mounted)
1262 break;
1263 dput(path->dentry);
1264 mntput(path->mnt);
1265 path->mnt = mounted;
1266 path->dentry = dget(mounted->mnt_root);
1267 }
1268}
1269
Nick Piggin31e6b012011-01-07 17:49:52 +11001270static void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271{
Al Viro2a737872009-04-07 11:49:53 -04001272 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001273
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 while(1) {
Jan Blunck4ac91372008-02-14 19:34:32 -08001275 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276
Al Viro2a737872009-04-07 11:49:53 -04001277 if (nd->path.dentry == nd->root.dentry &&
1278 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 break;
1280 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001281 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Al Viro3088dd72010-01-30 15:47:29 -05001282 /* rare case of legitimate dget_parent()... */
1283 nd->path.dentry = dget_parent(nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 dput(old);
1285 break;
1286 }
Al Viro3088dd72010-01-30 15:47:29 -05001287 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 }
Al Viro79ed0222009-04-18 13:59:41 -04001290 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001291 nd->inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292}
1293
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294/*
Miklos Szeredibad61182012-03-26 12:54:24 +02001295 * This looks up the name in dcache, possibly revalidates the old dentry and
1296 * allocates a new one if not found or not valid. In the need_lookup argument
1297 * returns whether i_op->lookup is necessary.
1298 *
1299 * dir->d_inode->i_mutex must be held
Nick Pigginbaa03892010-08-18 04:37:31 +10001300 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001301static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
Al Viro201f9562012-06-22 12:42:10 +04001302 unsigned int flags, bool *need_lookup)
Nick Pigginbaa03892010-08-18 04:37:31 +10001303{
Nick Pigginbaa03892010-08-18 04:37:31 +10001304 struct dentry *dentry;
Miklos Szeredibad61182012-03-26 12:54:24 +02001305 int error;
Nick Pigginbaa03892010-08-18 04:37:31 +10001306
Miklos Szeredibad61182012-03-26 12:54:24 +02001307 *need_lookup = false;
1308 dentry = d_lookup(dir, name);
1309 if (dentry) {
Jeff Layton39e3c952012-11-28 11:30:53 -05001310 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
Al Viro201f9562012-06-22 12:42:10 +04001311 error = d_revalidate(dentry, flags);
Miklos Szeredibad61182012-03-26 12:54:24 +02001312 if (unlikely(error <= 0)) {
1313 if (error < 0) {
1314 dput(dentry);
1315 return ERR_PTR(error);
1316 } else if (!d_invalidate(dentry)) {
1317 dput(dentry);
1318 dentry = NULL;
1319 }
1320 }
1321 }
1322 }
Nick Pigginbaa03892010-08-18 04:37:31 +10001323
Miklos Szeredibad61182012-03-26 12:54:24 +02001324 if (!dentry) {
1325 dentry = d_alloc(dir, name);
1326 if (unlikely(!dentry))
1327 return ERR_PTR(-ENOMEM);
Nick Pigginbaa03892010-08-18 04:37:31 +10001328
Miklos Szeredibad61182012-03-26 12:54:24 +02001329 *need_lookup = true;
Nick Pigginbaa03892010-08-18 04:37:31 +10001330 }
1331 return dentry;
1332}
1333
1334/*
Miklos Szeredibad61182012-03-26 12:54:24 +02001335 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1336 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1337 *
1338 * dir->d_inode->i_mutex must be held
Josef Bacik44396f42011-05-31 11:58:49 -04001339 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001340static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
Al Viro72bd8662012-06-10 17:17:17 -04001341 unsigned int flags)
Josef Bacik44396f42011-05-31 11:58:49 -04001342{
Josef Bacik44396f42011-05-31 11:58:49 -04001343 struct dentry *old;
1344
1345 /* Don't create child dentry for a dead directory. */
Miklos Szeredibad61182012-03-26 12:54:24 +02001346 if (unlikely(IS_DEADDIR(dir))) {
Miklos Szeredie188dc02012-02-03 14:25:18 +01001347 dput(dentry);
Josef Bacik44396f42011-05-31 11:58:49 -04001348 return ERR_PTR(-ENOENT);
Miklos Szeredie188dc02012-02-03 14:25:18 +01001349 }
Josef Bacik44396f42011-05-31 11:58:49 -04001350
Al Viro72bd8662012-06-10 17:17:17 -04001351 old = dir->i_op->lookup(dir, dentry, flags);
Josef Bacik44396f42011-05-31 11:58:49 -04001352 if (unlikely(old)) {
1353 dput(dentry);
1354 dentry = old;
1355 }
1356 return dentry;
1357}
1358
Al Viroa3255542012-03-30 14:41:51 -04001359static struct dentry *__lookup_hash(struct qstr *name,
Al Viro72bd8662012-06-10 17:17:17 -04001360 struct dentry *base, unsigned int flags)
Al Viroa3255542012-03-30 14:41:51 -04001361{
Miklos Szeredibad61182012-03-26 12:54:24 +02001362 bool need_lookup;
Al Viroa3255542012-03-30 14:41:51 -04001363 struct dentry *dentry;
1364
Al Viro72bd8662012-06-10 17:17:17 -04001365 dentry = lookup_dcache(name, base, flags, &need_lookup);
Miklos Szeredibad61182012-03-26 12:54:24 +02001366 if (!need_lookup)
1367 return dentry;
Al Viroa3255542012-03-30 14:41:51 -04001368
Al Viro72bd8662012-06-10 17:17:17 -04001369 return lookup_real(base->d_inode, dentry, flags);
Al Viroa3255542012-03-30 14:41:51 -04001370}
1371
Josef Bacik44396f42011-05-31 11:58:49 -04001372/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 * It's more convoluted than I'd like it to be, but... it's still fairly
1374 * small and for now I'd prefer to have fast path as straight as possible.
1375 * It _is_ time-critical.
1376 */
Al Viroe97cdc82013-01-24 18:16:00 -05001377static int lookup_fast(struct nameidata *nd,
Miklos Szeredi697f5142012-05-21 17:30:05 +02001378 struct path *path, struct inode **inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379{
Jan Blunck4ac91372008-02-14 19:34:32 -08001380 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001381 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001382 int need_reval = 1;
1383 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001384 int err;
1385
Al Viro3cac2602009-08-13 18:27:43 +04001386 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001387 * Rename seqlock is not required here because in the off chance
1388 * of a false negative due to a concurrent rename, we're going to
1389 * do the non-racy lookup, below.
1390 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001391 if (nd->flags & LOOKUP_RCU) {
1392 unsigned seq;
Linus Torvaldsda53be12013-05-21 15:22:44 -07001393 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
Al Viro5a18fff2011-03-11 04:44:53 -05001394 if (!dentry)
1395 goto unlazy;
1396
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001397 /*
1398 * This sequence count validates that the inode matches
1399 * the dentry name information from lookup.
1400 */
1401 *inode = dentry->d_inode;
1402 if (read_seqcount_retry(&dentry->d_seq, seq))
1403 return -ECHILD;
1404
1405 /*
1406 * This sequence count validates that the parent had no
1407 * changes while we did the lookup of the dentry above.
1408 *
1409 * The memory barrier in read_seqcount_begin of child is
1410 * enough, we can use __read_seqcount_retry here.
1411 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001412 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1413 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001414 nd->seq = seq;
Al Viro5a18fff2011-03-11 04:44:53 -05001415
Al Viro24643082011-02-15 01:26:22 -05001416 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
Al Viro4ce16ef32012-06-10 16:10:59 -04001417 status = d_revalidate(dentry, nd->flags);
Al Viro5a18fff2011-03-11 04:44:53 -05001418 if (unlikely(status <= 0)) {
1419 if (status != -ECHILD)
1420 need_reval = 0;
1421 goto unlazy;
1422 }
Al Viro24643082011-02-15 01:26:22 -05001423 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001424 path->mnt = mnt;
1425 path->dentry = dentry;
Al Virod6e9bd22011-05-27 07:03:15 -04001426 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1427 goto unlazy;
1428 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1429 goto unlazy;
1430 return 0;
Al Viro5a18fff2011-03-11 04:44:53 -05001431unlazy:
Al Viro19660af2011-03-25 10:32:48 -04001432 if (unlazy_walk(nd, dentry))
1433 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001434 } else {
Al Viroe97cdc82013-01-24 18:16:00 -05001435 dentry = __d_lookup(parent, &nd->last);
Nick Piggin31e6b012011-01-07 17:49:52 +11001436 }
Al Viro5a18fff2011-03-11 04:44:53 -05001437
Al Viro81e6f522012-03-30 14:48:04 -04001438 if (unlikely(!dentry))
1439 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001440
Al Viro5a18fff2011-03-11 04:44:53 -05001441 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
Al Viro4ce16ef32012-06-10 16:10:59 -04001442 status = d_revalidate(dentry, nd->flags);
Al Viro5a18fff2011-03-11 04:44:53 -05001443 if (unlikely(status <= 0)) {
1444 if (status < 0) {
1445 dput(dentry);
1446 return status;
1447 }
1448 if (!d_invalidate(dentry)) {
1449 dput(dentry);
Al Viro81e6f522012-03-30 14:48:04 -04001450 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001451 }
1452 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001453
David Howells9875cf82011-01-14 18:45:21 +00001454 path->mnt = mnt;
1455 path->dentry = dentry;
1456 err = follow_managed(path, nd->flags);
Ian Kent89312212011-01-18 12:06:10 +08001457 if (unlikely(err < 0)) {
1458 path_put_conditional(path, nd);
David Howells9875cf82011-01-14 18:45:21 +00001459 return err;
Ian Kent89312212011-01-18 12:06:10 +08001460 }
Al Viroa3fbbde2011-11-07 21:21:26 +00001461 if (err)
1462 nd->flags |= LOOKUP_JUMPED;
David Howells9875cf82011-01-14 18:45:21 +00001463 *inode = path->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 return 0;
Al Viro81e6f522012-03-30 14:48:04 -04001465
1466need_lookup:
Miklos Szeredi697f5142012-05-21 17:30:05 +02001467 return 1;
1468}
1469
1470/* Fast lookup failed, do it the slow way */
Al Virocc2a5272013-01-24 18:19:49 -05001471static int lookup_slow(struct nameidata *nd, struct path *path)
Miklos Szeredi697f5142012-05-21 17:30:05 +02001472{
1473 struct dentry *dentry, *parent;
1474 int err;
1475
1476 parent = nd->path.dentry;
Al Viro81e6f522012-03-30 14:48:04 -04001477 BUG_ON(nd->inode != parent->d_inode);
1478
1479 mutex_lock(&parent->d_inode->i_mutex);
Al Virocc2a5272013-01-24 18:19:49 -05001480 dentry = __lookup_hash(&nd->last, parent, nd->flags);
Al Viro81e6f522012-03-30 14:48:04 -04001481 mutex_unlock(&parent->d_inode->i_mutex);
1482 if (IS_ERR(dentry))
1483 return PTR_ERR(dentry);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001484 path->mnt = nd->path.mnt;
1485 path->dentry = dentry;
1486 err = follow_managed(path, nd->flags);
1487 if (unlikely(err < 0)) {
1488 path_put_conditional(path, nd);
1489 return err;
1490 }
1491 if (err)
1492 nd->flags |= LOOKUP_JUMPED;
1493 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494}
1495
Al Viro52094c82011-02-21 21:34:47 -05001496static inline int may_lookup(struct nameidata *nd)
1497{
1498 if (nd->flags & LOOKUP_RCU) {
Al Viro4ad5abb2011-06-20 19:57:03 -04001499 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
Al Viro52094c82011-02-21 21:34:47 -05001500 if (err != -ECHILD)
1501 return err;
Al Viro19660af2011-03-25 10:32:48 -04001502 if (unlazy_walk(nd, NULL))
Al Viro52094c82011-02-21 21:34:47 -05001503 return -ECHILD;
1504 }
Al Viro4ad5abb2011-06-20 19:57:03 -04001505 return inode_permission(nd->inode, MAY_EXEC);
Al Viro52094c82011-02-21 21:34:47 -05001506}
1507
Al Viro9856fa12011-03-04 14:22:06 -05001508static inline int handle_dots(struct nameidata *nd, int type)
1509{
1510 if (type == LAST_DOTDOT) {
1511 if (nd->flags & LOOKUP_RCU) {
1512 if (follow_dotdot_rcu(nd))
1513 return -ECHILD;
1514 } else
1515 follow_dotdot(nd);
1516 }
1517 return 0;
1518}
1519
Al Viro951361f2011-03-04 14:44:37 -05001520static void terminate_walk(struct nameidata *nd)
1521{
1522 if (!(nd->flags & LOOKUP_RCU)) {
1523 path_put(&nd->path);
1524 } else {
1525 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001526 if (!(nd->flags & LOOKUP_ROOT))
1527 nd->root.mnt = NULL;
Al Viro32a79912012-07-18 20:43:19 +04001528 unlock_rcu_walk();
Al Viro951361f2011-03-04 14:44:37 -05001529 }
1530}
1531
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001532/*
1533 * Do we need to follow links? We _really_ want to be able
1534 * to do this check without having to look at inode->i_op,
1535 * so we keep a cache of "no, this doesn't need follow_link"
1536 * for the common case.
1537 */
Linus Torvalds7813b942011-08-07 09:53:20 -07001538static inline int should_follow_link(struct inode *inode, int follow)
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001539{
1540 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1541 if (likely(inode->i_op->follow_link))
1542 return follow;
1543
1544 /* This gets set once for the inode lifetime */
1545 spin_lock(&inode->i_lock);
1546 inode->i_opflags |= IOP_NOFOLLOW;
1547 spin_unlock(&inode->i_lock);
1548 }
1549 return 0;
1550}
1551
Al Viroce57dfc2011-03-13 19:58:58 -04001552static inline int walk_component(struct nameidata *nd, struct path *path,
Al Viro21b9b072013-01-24 18:10:25 -05001553 int follow)
Al Viroce57dfc2011-03-13 19:58:58 -04001554{
1555 struct inode *inode;
1556 int err;
1557 /*
1558 * "." and ".." are special - ".." especially so because it has
1559 * to be able to know about the current root directory and
1560 * parent relationships.
1561 */
Al Viro21b9b072013-01-24 18:10:25 -05001562 if (unlikely(nd->last_type != LAST_NORM))
1563 return handle_dots(nd, nd->last_type);
Al Viroe97cdc82013-01-24 18:16:00 -05001564 err = lookup_fast(nd, path, &inode);
Al Viroce57dfc2011-03-13 19:58:58 -04001565 if (unlikely(err)) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001566 if (err < 0)
1567 goto out_err;
1568
Al Virocc2a5272013-01-24 18:19:49 -05001569 err = lookup_slow(nd, path);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001570 if (err < 0)
1571 goto out_err;
1572
1573 inode = path->dentry->d_inode;
Al Viroce57dfc2011-03-13 19:58:58 -04001574 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001575 err = -ENOENT;
1576 if (!inode)
1577 goto out_path_put;
1578
Linus Torvalds7813b942011-08-07 09:53:20 -07001579 if (should_follow_link(inode, follow)) {
Al Viro19660af2011-03-25 10:32:48 -04001580 if (nd->flags & LOOKUP_RCU) {
1581 if (unlikely(unlazy_walk(nd, path->dentry))) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001582 err = -ECHILD;
1583 goto out_err;
Al Viro19660af2011-03-25 10:32:48 -04001584 }
1585 }
Al Viroce57dfc2011-03-13 19:58:58 -04001586 BUG_ON(inode != path->dentry->d_inode);
1587 return 1;
1588 }
1589 path_to_nameidata(path, nd);
1590 nd->inode = inode;
1591 return 0;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001592
1593out_path_put:
1594 path_to_nameidata(path, nd);
1595out_err:
1596 terminate_walk(nd);
1597 return err;
Al Viroce57dfc2011-03-13 19:58:58 -04001598}
1599
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600/*
Al Virob3563792011-03-14 21:54:55 -04001601 * This limits recursive symlink follows to 8, while
1602 * limiting consecutive symlinks to 40.
1603 *
1604 * Without that kind of total limit, nasty chains of consecutive
1605 * symlinks can cause almost arbitrarily long lookups.
1606 */
1607static inline int nested_symlink(struct path *path, struct nameidata *nd)
1608{
1609 int res;
1610
Al Virob3563792011-03-14 21:54:55 -04001611 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1612 path_put_conditional(path, nd);
1613 path_put(&nd->path);
1614 return -ELOOP;
1615 }
Erez Zadok1a4022f2011-05-21 01:19:59 -04001616 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
Al Virob3563792011-03-14 21:54:55 -04001617
1618 nd->depth++;
1619 current->link_count++;
1620
1621 do {
1622 struct path link = *path;
1623 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04001624
1625 res = follow_link(&link, nd, &cookie);
Al Viro6d7b5aa2012-06-10 04:15:17 -04001626 if (res)
1627 break;
Al Viro21b9b072013-01-24 18:10:25 -05001628 res = walk_component(nd, path, LOOKUP_FOLLOW);
Al Viro574197e2011-03-14 22:20:34 -04001629 put_link(nd, &link, cookie);
Al Virob3563792011-03-14 21:54:55 -04001630 } while (res > 0);
1631
1632 current->link_count--;
1633 nd->depth--;
1634 return res;
1635}
1636
1637/*
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001638 * We really don't want to look at inode->i_op->lookup
1639 * when we don't have to. So we keep a cache bit in
1640 * the inode ->i_opflags field that says "yes, we can
1641 * do lookup on this inode".
1642 */
1643static inline int can_lookup(struct inode *inode)
1644{
1645 if (likely(inode->i_opflags & IOP_LOOKUP))
1646 return 1;
1647 if (likely(!inode->i_op->lookup))
1648 return 0;
1649
1650 /* We do this once for the lifetime of the inode */
1651 spin_lock(&inode->i_lock);
1652 inode->i_opflags |= IOP_LOOKUP;
1653 spin_unlock(&inode->i_lock);
1654 return 1;
1655}
1656
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001657/*
1658 * We can do the critical dentry name comparison and hashing
1659 * operations one word at a time, but we are limited to:
1660 *
1661 * - Architectures with fast unaligned word accesses. We could
1662 * do a "get_unaligned()" if this helps and is sufficiently
1663 * fast.
1664 *
1665 * - Little-endian machines (so that we can generate the mask
1666 * of low bytes efficiently). Again, we *could* do a byte
1667 * swapping load on big-endian architectures if that is not
1668 * expensive enough to make the optimization worthless.
1669 *
1670 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1671 * do not trap on the (extremely unlikely) case of a page
1672 * crossing operation.
1673 *
1674 * - Furthermore, we need an efficient 64-bit compile for the
1675 * 64-bit case in order to generate the "number of bytes in
1676 * the final mask". Again, that could be replaced with a
1677 * efficient population count instruction or similar.
1678 */
1679#ifdef CONFIG_DCACHE_WORD_ACCESS
1680
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001681#include <asm/word-at-a-time.h>
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001682
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001683#ifdef CONFIG_64BIT
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001684
1685static inline unsigned int fold_hash(unsigned long hash)
1686{
1687 hash += hash >> (8*sizeof(int));
1688 return hash;
1689}
1690
1691#else /* 32-bit case */
1692
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001693#define fold_hash(x) (x)
1694
1695#endif
1696
1697unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1698{
1699 unsigned long a, mask;
1700 unsigned long hash = 0;
1701
1702 for (;;) {
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001703 a = load_unaligned_zeropad(name);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001704 if (len < sizeof(unsigned long))
1705 break;
1706 hash += a;
Al Virof132c5b2012-03-22 21:59:52 +00001707 hash *= 9;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001708 name += sizeof(unsigned long);
1709 len -= sizeof(unsigned long);
1710 if (!len)
1711 goto done;
1712 }
1713 mask = ~(~0ul << len*8);
1714 hash += mask & a;
1715done:
1716 return fold_hash(hash);
1717}
1718EXPORT_SYMBOL(full_name_hash);
1719
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001720/*
1721 * Calculate the length and hash of the path component, and
1722 * return the length of the component;
1723 */
1724static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1725{
Linus Torvalds36126f82012-05-26 10:43:17 -07001726 unsigned long a, b, adata, bdata, mask, hash, len;
1727 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001728
1729 hash = a = 0;
1730 len = -sizeof(unsigned long);
1731 do {
1732 hash = (hash + a) * 9;
1733 len += sizeof(unsigned long);
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001734 a = load_unaligned_zeropad(name+len);
Linus Torvalds36126f82012-05-26 10:43:17 -07001735 b = a ^ REPEAT_BYTE('/');
1736 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001737
Linus Torvalds36126f82012-05-26 10:43:17 -07001738 adata = prep_zero_mask(a, adata, &constants);
1739 bdata = prep_zero_mask(b, bdata, &constants);
1740
1741 mask = create_zero_mask(adata | bdata);
1742
1743 hash += a & zero_bytemask(mask);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001744 *hashp = fold_hash(hash);
1745
Linus Torvalds36126f82012-05-26 10:43:17 -07001746 return len + find_zero(mask);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001747}
1748
1749#else
1750
Linus Torvalds0145acc2012-03-02 14:32:59 -08001751unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1752{
1753 unsigned long hash = init_name_hash();
1754 while (len--)
1755 hash = partial_name_hash(*name++, hash);
1756 return end_name_hash(hash);
1757}
Linus Torvaldsae942ae2012-03-02 19:40:57 -08001758EXPORT_SYMBOL(full_name_hash);
Linus Torvalds0145acc2012-03-02 14:32:59 -08001759
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001760/*
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001761 * We know there's a real path component here of at least
1762 * one character.
1763 */
1764static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1765{
1766 unsigned long hash = init_name_hash();
1767 unsigned long len = 0, c;
1768
1769 c = (unsigned char)*name;
1770 do {
1771 len++;
1772 hash = partial_name_hash(c, hash);
1773 c = (unsigned char)name[len];
1774 } while (c && c != '/');
1775 *hashp = end_name_hash(hash);
1776 return len;
1777}
1778
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001779#endif
1780
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001781/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01001783 * This is the basic name resolution function, turning a pathname into
1784 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01001786 * Returns 0 and nd will have valid dentry and mnt on success.
1787 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 */
Al Viro6de88d72009-08-09 01:41:57 +04001789static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790{
1791 struct path next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793
1794 while (*name=='/')
1795 name++;
1796 if (!*name)
Al Viro086e1832011-02-22 20:56:27 -05001797 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 /* At this point we know we have a real path component. */
1800 for(;;) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 struct qstr this;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001802 long len;
Al Virofe479a52011-02-22 15:10:03 -05001803 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804
Al Viro52094c82011-02-21 21:34:47 -05001805 err = may_lookup(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806 if (err)
1807 break;
1808
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001809 len = hash_name(name, &this.hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 this.name = name;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001811 this.len = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812
Al Virofe479a52011-02-22 15:10:03 -05001813 type = LAST_NORM;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001814 if (name[0] == '.') switch (len) {
Al Virofe479a52011-02-22 15:10:03 -05001815 case 2:
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001816 if (name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05001817 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05001818 nd->flags |= LOOKUP_JUMPED;
1819 }
Al Virofe479a52011-02-22 15:10:03 -05001820 break;
1821 case 1:
1822 type = LAST_DOT;
1823 }
Al Viro5a202bc2011-03-08 14:17:44 -05001824 if (likely(type == LAST_NORM)) {
1825 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05001826 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05001827 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07001828 err = parent->d_op->d_hash(parent, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05001829 if (err < 0)
1830 break;
1831 }
1832 }
Al Virofe479a52011-02-22 15:10:03 -05001833
Al Viro5f4a6a62013-01-24 18:04:22 -05001834 nd->last = this;
1835 nd->last_type = type;
1836
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001837 if (!name[len])
Al Viro5f4a6a62013-01-24 18:04:22 -05001838 return 0;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001839 /*
1840 * If it wasn't NUL, we know it was '/'. Skip that
1841 * slash, and continue until no more slashes.
1842 */
1843 do {
1844 len++;
1845 } while (unlikely(name[len] == '/'));
1846 if (!name[len])
Al Viro5f4a6a62013-01-24 18:04:22 -05001847 return 0;
1848
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001849 name += len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850
Al Viro21b9b072013-01-24 18:10:25 -05001851 err = walk_component(nd, &next, LOOKUP_FOLLOW);
Al Viroce57dfc2011-03-13 19:58:58 -04001852 if (err < 0)
1853 return err;
Al Virofe479a52011-02-22 15:10:03 -05001854
Al Viroce57dfc2011-03-13 19:58:58 -04001855 if (err) {
Al Virob3563792011-03-14 21:54:55 -04001856 err = nested_symlink(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857 if (err)
Al Viroa7472ba2011-03-04 14:39:30 -05001858 return err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001859 }
Al Viro5f4a6a62013-01-24 18:04:22 -05001860 if (!can_lookup(nd->inode)) {
1861 err = -ENOTDIR;
1862 break;
1863 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864 }
Al Viro951361f2011-03-04 14:44:37 -05001865 terminate_walk(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 return err;
1867}
1868
Al Viro70e9b352011-03-05 21:12:22 -05001869static int path_init(int dfd, const char *name, unsigned int flags,
1870 struct nameidata *nd, struct file **fp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001872 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873
1874 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro16c2cd72011-02-22 15:50:10 -05001875 nd->flags = flags | LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876 nd->depth = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05001877 if (flags & LOOKUP_ROOT) {
1878 struct inode *inode = nd->root.dentry->d_inode;
Al Viro73d049a2011-03-11 12:08:24 -05001879 if (*name) {
Al Viro741b7c32012-12-20 13:41:28 -05001880 if (!can_lookup(inode))
Al Viro73d049a2011-03-11 12:08:24 -05001881 return -ENOTDIR;
1882 retval = inode_permission(inode, MAY_EXEC);
1883 if (retval)
1884 return retval;
1885 }
Al Viro5b6ca022011-03-09 23:04:47 -05001886 nd->path = nd->root;
1887 nd->inode = inode;
1888 if (flags & LOOKUP_RCU) {
Al Viro32a79912012-07-18 20:43:19 +04001889 lock_rcu_walk();
Al Viro5b6ca022011-03-09 23:04:47 -05001890 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1891 } else {
1892 path_get(&nd->path);
1893 }
1894 return 0;
1895 }
1896
Al Viro2a737872009-04-07 11:49:53 -04001897 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 if (*name=='/') {
Al Viroe41f7d42011-02-22 14:02:58 -05001900 if (flags & LOOKUP_RCU) {
Al Viro32a79912012-07-18 20:43:19 +04001901 lock_rcu_walk();
Al Viroe41f7d42011-02-22 14:02:58 -05001902 set_root_rcu(nd);
1903 } else {
1904 set_root(nd);
1905 path_get(&nd->root);
1906 }
Al Viro2a737872009-04-07 11:49:53 -04001907 nd->path = nd->root;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001908 } else if (dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05001909 if (flags & LOOKUP_RCU) {
1910 struct fs_struct *fs = current->fs;
1911 unsigned seq;
1912
Al Viro32a79912012-07-18 20:43:19 +04001913 lock_rcu_walk();
Al Viroe41f7d42011-02-22 14:02:58 -05001914
1915 do {
1916 seq = read_seqcount_begin(&fs->seq);
1917 nd->path = fs->pwd;
1918 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1919 } while (read_seqcount_retry(&fs->seq, seq));
1920 } else {
1921 get_fs_pwd(current->fs, &nd->path);
1922 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001923 } else {
Jeff Layton582aa642012-12-11 08:56:16 -05001924 /* Caller must check execute permissions on the starting path component */
Al Viro2903ff02012-08-28 12:52:22 -04001925 struct fd f = fdget_raw(dfd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001926 struct dentry *dentry;
1927
Al Viro2903ff02012-08-28 12:52:22 -04001928 if (!f.file)
1929 return -EBADF;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001930
Al Viro2903ff02012-08-28 12:52:22 -04001931 dentry = f.file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001932
Al Virof52e0c12011-03-14 18:56:51 -04001933 if (*name) {
Al Viro741b7c32012-12-20 13:41:28 -05001934 if (!can_lookup(dentry->d_inode)) {
Al Viro2903ff02012-08-28 12:52:22 -04001935 fdput(f);
1936 return -ENOTDIR;
1937 }
Al Virof52e0c12011-03-14 18:56:51 -04001938 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001939
Al Viro2903ff02012-08-28 12:52:22 -04001940 nd->path = f.file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05001941 if (flags & LOOKUP_RCU) {
Al Viro2903ff02012-08-28 12:52:22 -04001942 if (f.need_put)
1943 *fp = f.file;
Al Viroe41f7d42011-02-22 14:02:58 -05001944 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro32a79912012-07-18 20:43:19 +04001945 lock_rcu_walk();
Al Viroe41f7d42011-02-22 14:02:58 -05001946 } else {
Al Viro2903ff02012-08-28 12:52:22 -04001947 path_get(&nd->path);
1948 fdput(f);
Al Viroe41f7d42011-02-22 14:02:58 -05001949 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 }
Al Viroe41f7d42011-02-22 14:02:58 -05001951
Nick Piggin31e6b012011-01-07 17:49:52 +11001952 nd->inode = nd->path.dentry->d_inode;
Al Viro9b4a9b12009-04-07 11:44:16 -04001953 return 0;
Al Viro9b4a9b12009-04-07 11:44:16 -04001954}
1955
Al Virobd92d7f2011-03-14 19:54:59 -04001956static inline int lookup_last(struct nameidata *nd, struct path *path)
1957{
1958 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1959 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1960
1961 nd->flags &= ~LOOKUP_PARENT;
Al Viro21b9b072013-01-24 18:10:25 -05001962 return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW);
Al Virobd92d7f2011-03-14 19:54:59 -04001963}
1964
Al Viro9b4a9b12009-04-07 11:44:16 -04001965/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroee0827c2011-02-21 23:38:09 -05001966static int path_lookupat(int dfd, const char *name,
Al Viro9b4a9b12009-04-07 11:44:16 -04001967 unsigned int flags, struct nameidata *nd)
1968{
Al Viro70e9b352011-03-05 21:12:22 -05001969 struct file *base = NULL;
Al Virobd92d7f2011-03-14 19:54:59 -04001970 struct path path;
1971 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001972
1973 /*
1974 * Path walking is largely split up into 2 different synchronisation
1975 * schemes, rcu-walk and ref-walk (explained in
1976 * Documentation/filesystems/path-lookup.txt). These share much of the
1977 * path walk code, but some things particularly setup, cleanup, and
1978 * following mounts are sufficiently divergent that functions are
1979 * duplicated. Typically there is a function foo(), and its RCU
1980 * analogue, foo_rcu().
1981 *
1982 * -ECHILD is the error number of choice (just to avoid clashes) that
1983 * is returned if some aspect of an rcu-walk fails. Such an error must
1984 * be handled by restarting a traditional ref-walk (which will always
1985 * be able to complete).
1986 */
Al Virobd92d7f2011-03-14 19:54:59 -04001987 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
Al Viroee0827c2011-02-21 23:38:09 -05001988
Al Virobd92d7f2011-03-14 19:54:59 -04001989 if (unlikely(err))
1990 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001991
1992 current->total_link_count = 0;
Al Virobd92d7f2011-03-14 19:54:59 -04001993 err = link_path_walk(name, nd);
1994
1995 if (!err && !(flags & LOOKUP_PARENT)) {
Al Virobd92d7f2011-03-14 19:54:59 -04001996 err = lookup_last(nd, &path);
1997 while (err > 0) {
1998 void *cookie;
1999 struct path link = path;
Kees Cook800179c2012-07-25 17:29:07 -07002000 err = may_follow_link(&link, nd);
2001 if (unlikely(err))
2002 break;
Al Virobd92d7f2011-03-14 19:54:59 -04002003 nd->flags |= LOOKUP_PARENT;
Al Viro574197e2011-03-14 22:20:34 -04002004 err = follow_link(&link, nd, &cookie);
Al Viro6d7b5aa2012-06-10 04:15:17 -04002005 if (err)
2006 break;
2007 err = lookup_last(nd, &path);
Al Viro574197e2011-03-14 22:20:34 -04002008 put_link(nd, &link, cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04002009 }
2010 }
Al Viroee0827c2011-02-21 23:38:09 -05002011
Al Viro9f1fafe2011-03-25 11:00:12 -04002012 if (!err)
2013 err = complete_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04002014
2015 if (!err && nd->flags & LOOKUP_DIRECTORY) {
Al Viro05252902013-06-06 19:33:47 -04002016 if (!can_lookup(nd->inode)) {
Al Virobd92d7f2011-03-14 19:54:59 -04002017 path_put(&nd->path);
Al Virobd23a532011-03-23 09:56:30 -04002018 err = -ENOTDIR;
Al Virobd92d7f2011-03-14 19:54:59 -04002019 }
2020 }
Al Viro16c2cd72011-02-22 15:50:10 -05002021
Al Viro70e9b352011-03-05 21:12:22 -05002022 if (base)
2023 fput(base);
Al Viroee0827c2011-02-21 23:38:09 -05002024
Al Viro5b6ca022011-03-09 23:04:47 -05002025 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
Al Viro2a737872009-04-07 11:49:53 -04002026 path_put(&nd->root);
2027 nd->root.mnt = NULL;
2028 }
Al Virobd92d7f2011-03-14 19:54:59 -04002029 return err;
Al Viroee0827c2011-02-21 23:38:09 -05002030}
Nick Piggin31e6b012011-01-07 17:49:52 +11002031
Jeff Layton873f1ee2012-10-10 15:25:29 -04002032static int filename_lookup(int dfd, struct filename *name,
2033 unsigned int flags, struct nameidata *nd)
2034{
2035 int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd);
2036 if (unlikely(retval == -ECHILD))
2037 retval = path_lookupat(dfd, name->name, flags, nd);
2038 if (unlikely(retval == -ESTALE))
2039 retval = path_lookupat(dfd, name->name,
2040 flags | LOOKUP_REVAL, nd);
2041
2042 if (likely(!retval))
Jeff Laytonadb5c242012-10-10 16:43:13 -04002043 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002044 return retval;
2045}
2046
Al Viroee0827c2011-02-21 23:38:09 -05002047static int do_path_lookup(int dfd, const char *name,
2048 unsigned int flags, struct nameidata *nd)
2049{
Jeff Layton873f1ee2012-10-10 15:25:29 -04002050 struct filename filename = { .name = name };
Nick Piggin31e6b012011-01-07 17:49:52 +11002051
Jeff Layton873f1ee2012-10-10 15:25:29 -04002052 return filename_lookup(dfd, &filename, flags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053}
2054
Al Viro79714f72012-06-15 03:01:42 +04002055/* does lookup, returns the object with parent locked */
2056struct dentry *kern_path_locked(const char *name, struct path *path)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002057{
Al Viro79714f72012-06-15 03:01:42 +04002058 struct nameidata nd;
2059 struct dentry *d;
2060 int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd);
2061 if (err)
2062 return ERR_PTR(err);
2063 if (nd.last_type != LAST_NORM) {
2064 path_put(&nd.path);
2065 return ERR_PTR(-EINVAL);
2066 }
2067 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Al Viro1e0ea002012-07-22 23:46:21 +04002068 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
Al Viro79714f72012-06-15 03:01:42 +04002069 if (IS_ERR(d)) {
2070 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2071 path_put(&nd.path);
2072 return d;
2073 }
2074 *path = nd.path;
2075 return d;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002076}
2077
Al Virod1811462008-08-02 00:49:18 -04002078int kern_path(const char *name, unsigned int flags, struct path *path)
2079{
2080 struct nameidata nd;
2081 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
2082 if (!res)
2083 *path = nd.path;
2084 return res;
2085}
2086
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002087/**
2088 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2089 * @dentry: pointer to dentry of the base directory
2090 * @mnt: pointer to vfs mount of the base directory
2091 * @name: pointer to file name
2092 * @flags: lookup flags
Al Viroe0a01242011-06-27 17:00:37 -04002093 * @path: pointer to struct path to fill
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002094 */
2095int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2096 const char *name, unsigned int flags,
Al Viroe0a01242011-06-27 17:00:37 -04002097 struct path *path)
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002098{
Al Viroe0a01242011-06-27 17:00:37 -04002099 struct nameidata nd;
2100 int err;
2101 nd.root.dentry = dentry;
2102 nd.root.mnt = mnt;
2103 BUG_ON(flags & LOOKUP_PARENT);
Al Viro5b6ca022011-03-09 23:04:47 -05002104 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
Al Viroe0a01242011-06-27 17:00:37 -04002105 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
2106 if (!err)
2107 *path = nd.path;
2108 return err;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002109}
2110
James Morris057f6c02007-04-26 00:12:05 -07002111/*
2112 * Restricted form of lookup. Doesn't follow links, single-component only,
2113 * needs parent already locked. Doesn't follow mounts.
2114 * SMP-safe.
2115 */
Adrian Bunka244e162006-03-31 02:32:11 -08002116static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117{
Al Viro72bd8662012-06-10 17:17:17 -04002118 return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119}
2120
Christoph Hellwigeead1912007-10-16 23:25:38 -07002121/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07002122 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07002123 * @name: pathname component to lookup
2124 * @base: base directory to lookup from
2125 * @len: maximum length @len should be interpreted to
2126 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07002127 * Note that this routine is purely a helper for filesystem usage and should
2128 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07002129 * nameidata argument is passed to the filesystem methods and a filesystem
2130 * using this helper needs to be prepared for that.
2131 */
James Morris057f6c02007-04-26 00:12:05 -07002132struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2133{
James Morris057f6c02007-04-26 00:12:05 -07002134 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05002135 unsigned int c;
Miklos Szeredicda309d2012-03-26 12:54:21 +02002136 int err;
James Morris057f6c02007-04-26 00:12:05 -07002137
David Woodhouse2f9092e2009-04-20 23:18:37 +01002138 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2139
Al Viro6a96ba52011-03-07 23:49:20 -05002140 this.name = name;
2141 this.len = len;
Linus Torvalds0145acc2012-03-02 14:32:59 -08002142 this.hash = full_name_hash(name, len);
Al Viro6a96ba52011-03-07 23:49:20 -05002143 if (!len)
2144 return ERR_PTR(-EACCES);
2145
Al Viro21d8a152012-11-29 22:17:21 -05002146 if (unlikely(name[0] == '.')) {
2147 if (len < 2 || (len == 2 && name[1] == '.'))
2148 return ERR_PTR(-EACCES);
2149 }
2150
Al Viro6a96ba52011-03-07 23:49:20 -05002151 while (len--) {
2152 c = *(const unsigned char *)name++;
2153 if (c == '/' || c == '\0')
2154 return ERR_PTR(-EACCES);
Al Viro6a96ba52011-03-07 23:49:20 -05002155 }
Al Viro5a202bc2011-03-08 14:17:44 -05002156 /*
2157 * See if the low-level filesystem might want
2158 * to use its own hash..
2159 */
2160 if (base->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002161 int err = base->d_op->d_hash(base, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05002162 if (err < 0)
2163 return ERR_PTR(err);
2164 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07002165
Miklos Szeredicda309d2012-03-26 12:54:21 +02002166 err = inode_permission(base->d_inode, MAY_EXEC);
2167 if (err)
2168 return ERR_PTR(err);
2169
Al Viro72bd8662012-06-10 17:17:17 -04002170 return __lookup_hash(&this, base, 0);
James Morris057f6c02007-04-26 00:12:05 -07002171}
2172
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002173int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2174 struct path *path, int *empty)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175{
Al Viro2d8f3032008-07-22 09:59:21 -04002176 struct nameidata nd;
Jeff Layton91a27b22012-10-10 15:25:28 -04002177 struct filename *tmp = getname_flags(name, flags, empty);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04002180
2181 BUG_ON(flags & LOOKUP_PARENT);
2182
Jeff Layton873f1ee2012-10-10 15:25:29 -04002183 err = filename_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04002185 if (!err)
2186 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187 }
2188 return err;
2189}
2190
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002191int user_path_at(int dfd, const char __user *name, unsigned flags,
2192 struct path *path)
2193{
Linus Torvaldsf7493e52012-03-22 16:10:40 -07002194 return user_path_at_empty(dfd, name, flags, path, NULL);
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002195}
2196
Jeff Layton873f1ee2012-10-10 15:25:29 -04002197/*
2198 * NB: most callers don't do anything directly with the reference to the
2199 * to struct filename, but the nd->last pointer points into the name string
2200 * allocated by getname. So we must hold the reference to it until all
2201 * path-walking is complete.
2202 */
Jeff Layton91a27b22012-10-10 15:25:28 -04002203static struct filename *
Jeff Layton9e790bd2012-12-11 12:10:09 -05002204user_path_parent(int dfd, const char __user *path, struct nameidata *nd,
2205 unsigned int flags)
Al Viro2ad94ae2008-07-21 09:32:51 -04002206{
Jeff Layton91a27b22012-10-10 15:25:28 -04002207 struct filename *s = getname(path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002208 int error;
2209
Jeff Layton9e790bd2012-12-11 12:10:09 -05002210 /* only LOOKUP_REVAL is allowed in extra flags */
2211 flags &= LOOKUP_REVAL;
2212
Al Viro2ad94ae2008-07-21 09:32:51 -04002213 if (IS_ERR(s))
Jeff Layton91a27b22012-10-10 15:25:28 -04002214 return s;
Al Viro2ad94ae2008-07-21 09:32:51 -04002215
Jeff Layton9e790bd2012-12-11 12:10:09 -05002216 error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd);
Jeff Layton91a27b22012-10-10 15:25:28 -04002217 if (error) {
Al Viro2ad94ae2008-07-21 09:32:51 -04002218 putname(s);
Jeff Layton91a27b22012-10-10 15:25:28 -04002219 return ERR_PTR(error);
2220 }
Al Viro2ad94ae2008-07-21 09:32:51 -04002221
Jeff Layton91a27b22012-10-10 15:25:28 -04002222 return s;
Al Viro2ad94ae2008-07-21 09:32:51 -04002223}
2224
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225/*
2226 * It's inline, so penalty for filesystems that don't use sticky bit is
2227 * minimal.
2228 */
2229static inline int check_sticky(struct inode *dir, struct inode *inode)
2230{
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002231 kuid_t fsuid = current_fsuid();
David Howellsda9592e2008-11-14 10:39:05 +11002232
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233 if (!(dir->i_mode & S_ISVTX))
2234 return 0;
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002235 if (uid_eq(inode->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236 return 0;
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002237 if (uid_eq(dir->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238 return 0;
Eric W. Biederman1a48e2a2011-11-14 16:24:06 -08002239 return !inode_capable(inode, CAP_FOWNER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240}
2241
2242/*
2243 * Check whether we can remove a link victim from directory dir, check
2244 * whether the type of victim is right.
2245 * 1. We can't do it if dir is read-only (done in permission())
2246 * 2. We should have write and exec permissions on dir
2247 * 3. We can't remove anything from append-only dir
2248 * 4. We can't do anything with immutable dir (done in permission())
2249 * 5. If the sticky bit on dir is set we should either
2250 * a. be owner of dir, or
2251 * b. be owner of victim, or
2252 * c. have CAP_FOWNER capability
2253 * 6. If the victim is append-only or immutable we can't do antyhing with
2254 * links pointing to it.
2255 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2256 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2257 * 9. We can't remove a root or mountpoint.
2258 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2259 * nfs_async_unlink().
2260 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08002261static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262{
2263 int error;
2264
2265 if (!victim->d_inode)
2266 return -ENOENT;
2267
2268 BUG_ON(victim->d_parent->d_inode != dir);
Jeff Layton4fa6b5e2012-10-10 15:25:25 -04002269 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270
Al Virof419a2e2008-07-22 00:07:17 -04002271 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 if (error)
2273 return error;
2274 if (IS_APPEND(dir))
2275 return -EPERM;
2276 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
Hugh Dickinsf9454542008-11-19 15:36:38 -08002277 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278 return -EPERM;
2279 if (isdir) {
2280 if (!S_ISDIR(victim->d_inode->i_mode))
2281 return -ENOTDIR;
2282 if (IS_ROOT(victim))
2283 return -EBUSY;
2284 } else if (S_ISDIR(victim->d_inode->i_mode))
2285 return -EISDIR;
2286 if (IS_DEADDIR(dir))
2287 return -ENOENT;
2288 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2289 return -EBUSY;
2290 return 0;
2291}
2292
2293/* Check whether we can create an object with dentry child in directory
2294 * dir.
2295 * 1. We can't do it if child already exists (open has special treatment for
2296 * this case, but since we are inlined it's OK)
2297 * 2. We can't do it if dir is read-only (done in permission())
2298 * 3. We should have write and exec permissions on dir
2299 * 4. We can't do it if dir is immutable (done in permission())
2300 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02002301static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302{
2303 if (child->d_inode)
2304 return -EEXIST;
2305 if (IS_DEADDIR(dir))
2306 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04002307 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308}
2309
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310/*
2311 * p1 and p2 should be directories on the same fs.
2312 */
2313struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2314{
2315 struct dentry *p;
2316
2317 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07002318 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319 return NULL;
2320 }
2321
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002322 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002324 p = d_ancestor(p2, p1);
2325 if (p) {
2326 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2327 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2328 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329 }
2330
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002331 p = d_ancestor(p1, p2);
2332 if (p) {
2333 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2334 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2335 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 }
2337
Ingo Molnarf2eace22006-07-03 00:25:05 -07002338 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2339 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340 return NULL;
2341}
2342
2343void unlock_rename(struct dentry *p1, struct dentry *p2)
2344{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002345 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002347 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002348 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 }
2350}
2351
Al Viro4acdaf22011-07-26 01:42:34 -04002352int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viro312b63f2012-06-10 18:09:36 -04002353 bool want_excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002355 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356 if (error)
2357 return error;
2358
Al Viroacfa4382008-12-04 10:06:33 -05002359 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360 return -EACCES; /* shouldn't it be ENOSYS? */
2361 mode &= S_IALLUGO;
2362 mode |= S_IFREG;
2363 error = security_inode_create(dir, dentry, mode);
2364 if (error)
2365 return error;
Al Viro312b63f2012-06-10 18:09:36 -04002366 error = dir->i_op->create(dir, dentry, mode, want_excl);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002367 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002368 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369 return error;
2370}
2371
Al Viro73d049a2011-03-11 12:08:24 -05002372static int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002374 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375 struct inode *inode = dentry->d_inode;
2376 int error;
2377
Al Virobcda7652011-03-13 16:42:14 -04002378 /* O_PATH? */
2379 if (!acc_mode)
2380 return 0;
2381
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 if (!inode)
2383 return -ENOENT;
2384
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002385 switch (inode->i_mode & S_IFMT) {
2386 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002388 case S_IFDIR:
2389 if (acc_mode & MAY_WRITE)
2390 return -EISDIR;
2391 break;
2392 case S_IFBLK:
2393 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002394 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002396 /*FALLTHRU*/
2397 case S_IFIFO:
2398 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002400 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002401 }
Dave Hansenb41572e2007-10-16 23:31:14 -07002402
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002403 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07002404 if (error)
2405 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05002406
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407 /*
2408 * An append-only file must be opened in append mode for writing.
2409 */
2410 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05002411 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05002412 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05002414 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415 }
2416
2417 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002418 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05002419 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420
J. Bruce Fieldsf3c7691e2011-09-21 10:58:13 -04002421 return 0;
Al Viro7715b522009-12-16 03:54:00 -05002422}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423
Jeff Laytone1181ee2010-12-07 16:19:50 -05002424static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002425{
Jeff Laytone1181ee2010-12-07 16:19:50 -05002426 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002427 struct inode *inode = path->dentry->d_inode;
2428 int error = get_write_access(inode);
2429 if (error)
2430 return error;
2431 /*
2432 * Refuse to truncate files with mandatory locks held on them.
2433 */
2434 error = locks_verify_locked(inode);
2435 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002436 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002437 if (!error) {
2438 error = do_truncate(path->dentry, 0,
2439 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002440 filp);
Al Viro7715b522009-12-16 03:54:00 -05002441 }
2442 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002443 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444}
2445
Dave Hansend57999e2008-02-15 14:37:27 -08002446static inline int open_to_namei_flags(int flag)
2447{
Al Viro8a5e9292011-06-25 19:15:54 -04002448 if ((flag & O_ACCMODE) == 3)
2449 flag--;
Dave Hansend57999e2008-02-15 14:37:27 -08002450 return flag;
2451}
2452
Miklos Szeredid18e9002012-06-05 15:10:17 +02002453static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2454{
2455 int error = security_path_mknod(dir, dentry, mode, 0);
2456 if (error)
2457 return error;
2458
2459 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2460 if (error)
2461 return error;
2462
2463 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2464}
2465
David Howells1acf0af2012-06-14 16:13:46 +01002466/*
2467 * Attempt to atomically look up, create and open a file from a negative
2468 * dentry.
2469 *
2470 * Returns 0 if successful. The file will have been created and attached to
2471 * @file by the filesystem calling finish_open().
2472 *
2473 * Returns 1 if the file was looked up only or didn't need creating. The
2474 * caller will need to perform the open themselves. @path will have been
2475 * updated to point to the new dentry. This may be negative.
2476 *
2477 * Returns an error code otherwise.
2478 */
Al Viro2675a4e2012-06-22 12:41:10 +04002479static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2480 struct path *path, struct file *file,
2481 const struct open_flags *op,
Al Viro64894cf2012-07-31 00:53:35 +04002482 bool got_write, bool need_lookup,
Al Viro2675a4e2012-06-22 12:41:10 +04002483 int *opened)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002484{
2485 struct inode *dir = nd->path.dentry->d_inode;
2486 unsigned open_flag = open_to_namei_flags(op->open_flag);
2487 umode_t mode;
2488 int error;
2489 int acc_mode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002490 int create_error = 0;
2491 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2492
2493 BUG_ON(dentry->d_inode);
2494
2495 /* Don't create child dentry for a dead directory. */
2496 if (unlikely(IS_DEADDIR(dir))) {
Al Viro2675a4e2012-06-22 12:41:10 +04002497 error = -ENOENT;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002498 goto out;
2499 }
2500
Miklos Szeredi62b259d2012-08-15 13:01:24 +02002501 mode = op->mode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002502 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2503 mode &= ~current_umask();
2504
Al Virof8310c52012-07-30 11:50:30 +04002505 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT)) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002506 open_flag &= ~O_TRUNC;
Al Viro47237682012-06-10 05:01:45 -04002507 *opened |= FILE_CREATED;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002508 }
2509
2510 /*
2511 * Checking write permission is tricky, bacuse we don't know if we are
2512 * going to actually need it: O_CREAT opens should work as long as the
2513 * file exists. But checking existence breaks atomicity. The trick is
2514 * to check access and if not granted clear O_CREAT from the flags.
2515 *
2516 * Another problem is returing the "right" error value (e.g. for an
2517 * O_EXCL open we want to return EEXIST not EROFS).
2518 */
Al Viro64894cf2012-07-31 00:53:35 +04002519 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2520 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2521 if (!(open_flag & O_CREAT)) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002522 /*
2523 * No O_CREATE -> atomicity not a requirement -> fall
2524 * back to lookup + open
2525 */
2526 goto no_open;
2527 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2528 /* Fall back and fail with the right error */
Al Viro64894cf2012-07-31 00:53:35 +04002529 create_error = -EROFS;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002530 goto no_open;
2531 } else {
2532 /* No side effects, safe to clear O_CREAT */
Al Viro64894cf2012-07-31 00:53:35 +04002533 create_error = -EROFS;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002534 open_flag &= ~O_CREAT;
2535 }
2536 }
2537
2538 if (open_flag & O_CREAT) {
Miklos Szeredi38227f72012-08-15 13:01:24 +02002539 error = may_o_create(&nd->path, dentry, mode);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002540 if (error) {
2541 create_error = error;
2542 if (open_flag & O_EXCL)
2543 goto no_open;
2544 open_flag &= ~O_CREAT;
2545 }
2546 }
2547
2548 if (nd->flags & LOOKUP_DIRECTORY)
2549 open_flag |= O_DIRECTORY;
2550
Al Viro30d90492012-06-22 12:40:19 +04002551 file->f_path.dentry = DENTRY_NOT_SET;
2552 file->f_path.mnt = nd->path.mnt;
2553 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
Al Viro47237682012-06-10 05:01:45 -04002554 opened);
Al Virod9585272012-06-22 12:39:14 +04002555 if (error < 0) {
Al Virod9585272012-06-22 12:39:14 +04002556 if (create_error && error == -ENOENT)
2557 error = create_error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002558 goto out;
2559 }
2560
2561 acc_mode = op->acc_mode;
Al Viro47237682012-06-10 05:01:45 -04002562 if (*opened & FILE_CREATED) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002563 fsnotify_create(dir, dentry);
2564 acc_mode = MAY_OPEN;
2565 }
2566
Al Virod9585272012-06-22 12:39:14 +04002567 if (error) { /* returned 1, that is */
Al Viro30d90492012-06-22 12:40:19 +04002568 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
Al Viro2675a4e2012-06-22 12:41:10 +04002569 error = -EIO;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002570 goto out;
2571 }
Al Viro30d90492012-06-22 12:40:19 +04002572 if (file->f_path.dentry) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002573 dput(dentry);
Al Viro30d90492012-06-22 12:40:19 +04002574 dentry = file->f_path.dentry;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002575 }
Sage Weil62b2ce92012-08-15 13:30:12 -07002576 if (create_error && dentry->d_inode == NULL) {
2577 error = create_error;
2578 goto out;
2579 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02002580 goto looked_up;
2581 }
2582
2583 /*
2584 * We didn't have the inode before the open, so check open permission
2585 * here.
2586 */
Al Viro2675a4e2012-06-22 12:41:10 +04002587 error = may_open(&file->f_path, acc_mode, open_flag);
2588 if (error)
2589 fput(file);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002590
2591out:
2592 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04002593 return error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002594
Miklos Szeredid18e9002012-06-05 15:10:17 +02002595no_open:
2596 if (need_lookup) {
Al Viro72bd8662012-06-10 17:17:17 -04002597 dentry = lookup_real(dir, dentry, nd->flags);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002598 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002599 return PTR_ERR(dentry);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002600
2601 if (create_error) {
2602 int open_flag = op->open_flag;
2603
Al Viro2675a4e2012-06-22 12:41:10 +04002604 error = create_error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002605 if ((open_flag & O_EXCL)) {
2606 if (!dentry->d_inode)
2607 goto out;
2608 } else if (!dentry->d_inode) {
2609 goto out;
2610 } else if ((open_flag & O_TRUNC) &&
2611 S_ISREG(dentry->d_inode->i_mode)) {
2612 goto out;
2613 }
2614 /* will fail later, go on to get the right error */
2615 }
2616 }
2617looked_up:
2618 path->dentry = dentry;
2619 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04002620 return 1;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002621}
2622
Nick Piggin31e6b012011-01-07 17:49:52 +11002623/*
David Howells1acf0af2012-06-14 16:13:46 +01002624 * Look up and maybe create and open the last component.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002625 *
2626 * Must be called with i_mutex held on parent.
2627 *
David Howells1acf0af2012-06-14 16:13:46 +01002628 * Returns 0 if the file was successfully atomically created (if necessary) and
2629 * opened. In this case the file will be returned attached to @file.
2630 *
2631 * Returns 1 if the file was not completely opened at this time, though lookups
2632 * and creations will have been performed and the dentry returned in @path will
2633 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2634 * specified then a negative dentry may be returned.
2635 *
2636 * An error code is returned otherwise.
2637 *
2638 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2639 * cleared otherwise prior to returning.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002640 */
Al Viro2675a4e2012-06-22 12:41:10 +04002641static int lookup_open(struct nameidata *nd, struct path *path,
2642 struct file *file,
2643 const struct open_flags *op,
Al Viro64894cf2012-07-31 00:53:35 +04002644 bool got_write, int *opened)
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002645{
2646 struct dentry *dir = nd->path.dentry;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002647 struct inode *dir_inode = dir->d_inode;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002648 struct dentry *dentry;
2649 int error;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002650 bool need_lookup;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002651
Al Viro47237682012-06-10 05:01:45 -04002652 *opened &= ~FILE_CREATED;
Al Viro201f9562012-06-22 12:42:10 +04002653 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002654 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002655 return PTR_ERR(dentry);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002656
Miklos Szeredid18e9002012-06-05 15:10:17 +02002657 /* Cached positive dentry: will open in f_op->open */
2658 if (!need_lookup && dentry->d_inode)
2659 goto out_no_open;
2660
2661 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
Al Viro64894cf2012-07-31 00:53:35 +04002662 return atomic_open(nd, dentry, path, file, op, got_write,
Al Viro47237682012-06-10 05:01:45 -04002663 need_lookup, opened);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002664 }
2665
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002666 if (need_lookup) {
2667 BUG_ON(dentry->d_inode);
2668
Al Viro72bd8662012-06-10 17:17:17 -04002669 dentry = lookup_real(dir_inode, dentry, nd->flags);
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002670 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002671 return PTR_ERR(dentry);
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002672 }
2673
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002674 /* Negative dentry, just create the file */
2675 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2676 umode_t mode = op->mode;
2677 if (!IS_POSIXACL(dir->d_inode))
2678 mode &= ~current_umask();
2679 /*
2680 * This write is needed to ensure that a
2681 * rw->ro transition does not occur between
2682 * the time when the file is created and when
2683 * a permanent write count is taken through
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002684 * the 'struct file' in finish_open().
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002685 */
Al Viro64894cf2012-07-31 00:53:35 +04002686 if (!got_write) {
2687 error = -EROFS;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002688 goto out_dput;
Al Viro64894cf2012-07-31 00:53:35 +04002689 }
Al Viro47237682012-06-10 05:01:45 -04002690 *opened |= FILE_CREATED;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002691 error = security_path_mknod(&nd->path, dentry, mode, 0);
2692 if (error)
2693 goto out_dput;
Al Viro312b63f2012-06-10 18:09:36 -04002694 error = vfs_create(dir->d_inode, dentry, mode,
2695 nd->flags & LOOKUP_EXCL);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002696 if (error)
2697 goto out_dput;
2698 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02002699out_no_open:
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002700 path->dentry = dentry;
2701 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04002702 return 1;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002703
2704out_dput:
2705 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04002706 return error;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002707}
2708
2709/*
Al Virofe2d35f2011-03-05 22:58:25 -05002710 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11002711 */
Al Viro2675a4e2012-06-22 12:41:10 +04002712static int do_last(struct nameidata *nd, struct path *path,
2713 struct file *file, const struct open_flags *op,
Jeff Layton669abf42012-10-10 16:43:10 -04002714 int *opened, struct filename *name)
Al Virofb1cc552009-12-24 01:58:28 -05002715{
Al Viroa1e28032009-12-24 02:12:06 -05002716 struct dentry *dir = nd->path.dentry;
Al Viroca344a892011-03-09 00:36:45 -05002717 int open_flag = op->open_flag;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002718 bool will_truncate = (open_flag & O_TRUNC) != 0;
Al Viro64894cf2012-07-31 00:53:35 +04002719 bool got_write = false;
Al Virobcda7652011-03-13 16:42:14 -04002720 int acc_mode = op->acc_mode;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02002721 struct inode *inode;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002722 bool symlink_ok = false;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002723 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2724 bool retried = false;
Al Viro16c2cd72011-02-22 15:50:10 -05002725 int error;
Al Virofb1cc552009-12-24 01:58:28 -05002726
Al Viroc3e380b2011-02-23 13:39:45 -05002727 nd->flags &= ~LOOKUP_PARENT;
2728 nd->flags |= op->intent;
2729
Al Virobc77daa2013-06-06 09:12:33 -04002730 if (nd->last_type != LAST_NORM) {
Al Virofe2d35f2011-03-05 22:58:25 -05002731 error = handle_dots(nd, nd->last_type);
2732 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002733 return error;
Miklos Szeredie83db162012-06-05 15:10:29 +02002734 goto finish_open;
Al Viro1f36f772009-12-26 10:56:19 -05002735 }
Al Viro67ee3ad2009-12-26 07:01:01 -05002736
Al Viroca344a892011-03-09 00:36:45 -05002737 if (!(open_flag & O_CREAT)) {
Al Virofe2d35f2011-03-05 22:58:25 -05002738 if (nd->last.name[nd->last.len])
2739 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
Al Virobcda7652011-03-13 16:42:14 -04002740 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002741 symlink_ok = true;
Al Virofe2d35f2011-03-05 22:58:25 -05002742 /* we _can_ be in RCU mode here */
Al Viroe97cdc82013-01-24 18:16:00 -05002743 error = lookup_fast(nd, path, &inode);
Miklos Szeredi71574862012-06-05 15:10:14 +02002744 if (likely(!error))
2745 goto finish_lookup;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02002746
Miklos Szeredi71574862012-06-05 15:10:14 +02002747 if (error < 0)
Al Viro2675a4e2012-06-22 12:41:10 +04002748 goto out;
Miklos Szeredi37d7fff2012-06-05 15:10:12 +02002749
Miklos Szeredi71574862012-06-05 15:10:14 +02002750 BUG_ON(nd->inode != dir->d_inode);
Miklos Szeredib6183df2012-06-05 15:10:13 +02002751 } else {
2752 /* create side of things */
2753 /*
2754 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2755 * has been cleared when we got to the last component we are
2756 * about to look up
2757 */
2758 error = complete_walk(nd);
2759 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002760 return error;
Miklos Szeredib6183df2012-06-05 15:10:13 +02002761
Jeff Layton33e22082013-04-12 15:16:32 -04002762 audit_inode(name, dir, LOOKUP_PARENT);
Miklos Szeredib6183df2012-06-05 15:10:13 +02002763 error = -EISDIR;
2764 /* trailing slashes? */
2765 if (nd->last.name[nd->last.len])
Al Viro2675a4e2012-06-22 12:41:10 +04002766 goto out;
Al Virofe2d35f2011-03-05 22:58:25 -05002767 }
2768
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002769retry_lookup:
Al Viro64894cf2012-07-31 00:53:35 +04002770 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2771 error = mnt_want_write(nd->path.mnt);
2772 if (!error)
2773 got_write = true;
2774 /*
2775 * do _not_ fail yet - we might not need that or fail with
2776 * a different error; let lookup_open() decide; we'll be
2777 * dropping this one anyway.
2778 */
2779 }
Al Viroa1e28032009-12-24 02:12:06 -05002780 mutex_lock(&dir->d_inode->i_mutex);
Al Viro64894cf2012-07-31 00:53:35 +04002781 error = lookup_open(nd, path, file, op, got_write, opened);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002782 mutex_unlock(&dir->d_inode->i_mutex);
Al Viroa1e28032009-12-24 02:12:06 -05002783
Al Viro2675a4e2012-06-22 12:41:10 +04002784 if (error <= 0) {
2785 if (error)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002786 goto out;
2787
Al Viro47237682012-06-10 05:01:45 -04002788 if ((*opened & FILE_CREATED) ||
Al Viro496ad9a2013-01-23 17:07:38 -05002789 !S_ISREG(file_inode(file)->i_mode))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002790 will_truncate = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002791
Jeff Laytonadb5c242012-10-10 16:43:13 -04002792 audit_inode(name, file->f_path.dentry, 0);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002793 goto opened;
2794 }
Al Virofb1cc552009-12-24 01:58:28 -05002795
Al Viro47237682012-06-10 05:01:45 -04002796 if (*opened & FILE_CREATED) {
Al Viro9b44f1b2011-03-09 00:17:27 -05002797 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05002798 open_flag &= ~O_TRUNC;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002799 will_truncate = false;
Al Virobcda7652011-03-13 16:42:14 -04002800 acc_mode = MAY_OPEN;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002801 path_to_nameidata(path, nd);
Miklos Szeredie83db162012-06-05 15:10:29 +02002802 goto finish_open_created;
Al Virofb1cc552009-12-24 01:58:28 -05002803 }
2804
2805 /*
Jeff Layton3134f372012-07-25 10:19:47 -04002806 * create/update audit record if it already exists.
Al Virofb1cc552009-12-24 01:58:28 -05002807 */
Jeff Layton3134f372012-07-25 10:19:47 -04002808 if (path->dentry->d_inode)
Jeff Laytonadb5c242012-10-10 16:43:13 -04002809 audit_inode(name, path->dentry, 0);
Al Virofb1cc552009-12-24 01:58:28 -05002810
Miklos Szeredid18e9002012-06-05 15:10:17 +02002811 /*
2812 * If atomic_open() acquired write access it is dropped now due to
2813 * possible mount and symlink following (this might be optimized away if
2814 * necessary...)
2815 */
Al Viro64894cf2012-07-31 00:53:35 +04002816 if (got_write) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002817 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04002818 got_write = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002819 }
2820
Al Virofb1cc552009-12-24 01:58:28 -05002821 error = -EEXIST;
Al Virof8310c52012-07-30 11:50:30 +04002822 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
Al Virofb1cc552009-12-24 01:58:28 -05002823 goto exit_dput;
2824
David Howells9875cf82011-01-14 18:45:21 +00002825 error = follow_managed(path, nd->flags);
2826 if (error < 0)
2827 goto exit_dput;
Al Virofb1cc552009-12-24 01:58:28 -05002828
Al Viroa3fbbde2011-11-07 21:21:26 +00002829 if (error)
2830 nd->flags |= LOOKUP_JUMPED;
2831
Miklos Szeredidecf3402012-05-21 17:30:08 +02002832 BUG_ON(nd->flags & LOOKUP_RCU);
2833 inode = path->dentry->d_inode;
Miklos Szeredi5f5daac2012-05-21 17:30:14 +02002834finish_lookup:
2835 /* we _can_ be in RCU mode here */
Al Virofb1cc552009-12-24 01:58:28 -05002836 error = -ENOENT;
Miklos Szeredi54c33e72012-05-21 17:30:10 +02002837 if (!inode) {
2838 path_to_nameidata(path, nd);
Al Viro2675a4e2012-06-22 12:41:10 +04002839 goto out;
Miklos Szeredi54c33e72012-05-21 17:30:10 +02002840 }
Al Viro9e67f362009-12-26 07:04:50 -05002841
Miklos Szeredid45ea862012-05-21 17:30:09 +02002842 if (should_follow_link(inode, !symlink_ok)) {
2843 if (nd->flags & LOOKUP_RCU) {
2844 if (unlikely(unlazy_walk(nd, path->dentry))) {
2845 error = -ECHILD;
Al Viro2675a4e2012-06-22 12:41:10 +04002846 goto out;
Miklos Szeredid45ea862012-05-21 17:30:09 +02002847 }
2848 }
2849 BUG_ON(inode != path->dentry->d_inode);
Al Viro2675a4e2012-06-22 12:41:10 +04002850 return 1;
Miklos Szeredid45ea862012-05-21 17:30:09 +02002851 }
Al Virofb1cc552009-12-24 01:58:28 -05002852
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002853 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
2854 path_to_nameidata(path, nd);
2855 } else {
2856 save_parent.dentry = nd->path.dentry;
2857 save_parent.mnt = mntget(path->mnt);
2858 nd->path.dentry = path->dentry;
2859
2860 }
Miklos Szeredidecf3402012-05-21 17:30:08 +02002861 nd->inode = inode;
Al Viroa3fbbde2011-11-07 21:21:26 +00002862 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
Al Virobc77daa2013-06-06 09:12:33 -04002863finish_open:
Al Viroa3fbbde2011-11-07 21:21:26 +00002864 error = complete_walk(nd);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002865 if (error) {
2866 path_put(&save_parent);
Al Viro2675a4e2012-06-22 12:41:10 +04002867 return error;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002868 }
Al Virobc77daa2013-06-06 09:12:33 -04002869 audit_inode(name, nd->path.dentry, 0);
Al Virofb1cc552009-12-24 01:58:28 -05002870 error = -EISDIR;
Miklos Szeredi050ac842012-05-21 17:30:12 +02002871 if ((open_flag & O_CREAT) && S_ISDIR(nd->inode->i_mode))
Al Viro2675a4e2012-06-22 12:41:10 +04002872 goto out;
Miklos Szerediaf2f5542012-05-21 17:30:11 +02002873 error = -ENOTDIR;
Al Viro05252902013-06-06 19:33:47 -04002874 if ((nd->flags & LOOKUP_DIRECTORY) && !can_lookup(nd->inode))
Al Viro2675a4e2012-06-22 12:41:10 +04002875 goto out;
Al Viro6c0d46c2011-03-09 00:59:59 -05002876 if (!S_ISREG(nd->inode->i_mode))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002877 will_truncate = false;
Al Viro6c0d46c2011-03-09 00:59:59 -05002878
Al Viro0f9d1a12011-03-09 00:13:14 -05002879 if (will_truncate) {
2880 error = mnt_want_write(nd->path.mnt);
2881 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002882 goto out;
Al Viro64894cf2012-07-31 00:53:35 +04002883 got_write = true;
Al Viro0f9d1a12011-03-09 00:13:14 -05002884 }
Miklos Szeredie83db162012-06-05 15:10:29 +02002885finish_open_created:
Al Virobcda7652011-03-13 16:42:14 -04002886 error = may_open(&nd->path, acc_mode, open_flag);
Al Viroca344a892011-03-09 00:36:45 -05002887 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002888 goto out;
Al Viro30d90492012-06-22 12:40:19 +04002889 file->f_path.mnt = nd->path.mnt;
2890 error = finish_open(file, nd->path.dentry, NULL, opened);
2891 if (error) {
Al Viro30d90492012-06-22 12:40:19 +04002892 if (error == -EOPENSTALE)
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02002893 goto stale_open;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002894 goto out;
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02002895 }
Miklos Szeredia8277b92012-06-05 15:10:32 +02002896opened:
Al Viro2675a4e2012-06-22 12:41:10 +04002897 error = open_check_o_direct(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002898 if (error)
2899 goto exit_fput;
Al Viro2675a4e2012-06-22 12:41:10 +04002900 error = ima_file_check(file, op->acc_mode);
Miklos Szerediaa4caad2012-06-05 15:10:28 +02002901 if (error)
2902 goto exit_fput;
2903
2904 if (will_truncate) {
Al Viro2675a4e2012-06-22 12:41:10 +04002905 error = handle_truncate(file);
Miklos Szerediaa4caad2012-06-05 15:10:28 +02002906 if (error)
2907 goto exit_fput;
Al Viro0f9d1a12011-03-09 00:13:14 -05002908 }
Al Viroca344a892011-03-09 00:36:45 -05002909out:
Al Viro64894cf2012-07-31 00:53:35 +04002910 if (got_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05002911 mnt_drop_write(nd->path.mnt);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002912 path_put(&save_parent);
Miklos Szeredie276ae62012-05-21 17:30:06 +02002913 terminate_walk(nd);
Al Viro2675a4e2012-06-22 12:41:10 +04002914 return error;
Al Virofb1cc552009-12-24 01:58:28 -05002915
Al Virofb1cc552009-12-24 01:58:28 -05002916exit_dput:
2917 path_put_conditional(path, nd);
Al Viroca344a892011-03-09 00:36:45 -05002918 goto out;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002919exit_fput:
Al Viro2675a4e2012-06-22 12:41:10 +04002920 fput(file);
2921 goto out;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002922
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02002923stale_open:
2924 /* If no saved parent or already retried then can't retry */
2925 if (!save_parent.dentry || retried)
2926 goto out;
2927
2928 BUG_ON(save_parent.dentry != dir);
2929 path_put(&nd->path);
2930 nd->path = save_parent;
2931 nd->inode = dir->d_inode;
2932 save_parent.mnt = NULL;
2933 save_parent.dentry = NULL;
Al Viro64894cf2012-07-31 00:53:35 +04002934 if (got_write) {
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02002935 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04002936 got_write = false;
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02002937 }
2938 retried = true;
2939 goto retry_lookup;
Al Virofb1cc552009-12-24 01:58:28 -05002940}
2941
Al Viro60545d02013-06-07 01:20:27 -04002942static int do_tmpfile(int dfd, struct filename *pathname,
2943 struct nameidata *nd, int flags,
2944 const struct open_flags *op,
2945 struct file *file, int *opened)
2946{
2947 static const struct qstr name = QSTR_INIT("/", 1);
2948 struct dentry *dentry, *child;
2949 struct inode *dir;
2950 int error = path_lookupat(dfd, pathname->name,
2951 flags | LOOKUP_DIRECTORY, nd);
2952 if (unlikely(error))
2953 return error;
2954 error = mnt_want_write(nd->path.mnt);
2955 if (unlikely(error))
2956 goto out;
2957 /* we want directory to be writable */
2958 error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
2959 if (error)
2960 goto out2;
2961 dentry = nd->path.dentry;
2962 dir = dentry->d_inode;
2963 if (!dir->i_op->tmpfile) {
2964 error = -EOPNOTSUPP;
2965 goto out2;
2966 }
2967 child = d_alloc(dentry, &name);
2968 if (unlikely(!child)) {
2969 error = -ENOMEM;
2970 goto out2;
2971 }
2972 nd->flags &= ~LOOKUP_DIRECTORY;
2973 nd->flags |= op->intent;
2974 dput(nd->path.dentry);
2975 nd->path.dentry = child;
2976 error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
2977 if (error)
2978 goto out2;
2979 audit_inode(pathname, nd->path.dentry, 0);
2980 error = may_open(&nd->path, op->acc_mode, op->open_flag);
2981 if (error)
2982 goto out2;
2983 file->f_path.mnt = nd->path.mnt;
2984 error = finish_open(file, nd->path.dentry, NULL, opened);
2985 if (error)
2986 goto out2;
2987 error = open_check_o_direct(file);
Al Virof4e0c302013-06-11 08:34:36 +04002988 if (error) {
Al Viro60545d02013-06-07 01:20:27 -04002989 fput(file);
Al Virof4e0c302013-06-11 08:34:36 +04002990 } else if (!(op->open_flag & O_EXCL)) {
2991 struct inode *inode = file_inode(file);
2992 spin_lock(&inode->i_lock);
2993 inode->i_state |= I_LINKABLE;
2994 spin_unlock(&inode->i_lock);
2995 }
Al Viro60545d02013-06-07 01:20:27 -04002996out2:
2997 mnt_drop_write(nd->path.mnt);
2998out:
2999 path_put(&nd->path);
3000 return error;
3001}
3002
Jeff Layton669abf42012-10-10 16:43:10 -04003003static struct file *path_openat(int dfd, struct filename *pathname,
Al Viro73d049a2011-03-11 12:08:24 -05003004 struct nameidata *nd, const struct open_flags *op, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005{
Al Virofe2d35f2011-03-05 22:58:25 -05003006 struct file *base = NULL;
Al Viro30d90492012-06-22 12:40:19 +04003007 struct file *file;
Al Viro9850c052010-01-13 15:01:15 -05003008 struct path path;
Al Viro47237682012-06-10 05:01:45 -04003009 int opened = 0;
Al Viro13aab422011-02-23 17:54:08 -05003010 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11003011
Al Viro30d90492012-06-22 12:40:19 +04003012 file = get_empty_filp();
Al Viro1afc99b2013-02-14 20:41:04 -05003013 if (IS_ERR(file))
3014 return file;
Nick Piggin31e6b012011-01-07 17:49:52 +11003015
Al Viro30d90492012-06-22 12:40:19 +04003016 file->f_flags = op->open_flag;
Nick Piggin31e6b012011-01-07 17:49:52 +11003017
Al Virobb458c62013-07-13 13:26:37 +04003018 if (unlikely(file->f_flags & __O_TMPFILE)) {
Al Viro60545d02013-06-07 01:20:27 -04003019 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3020 goto out;
3021 }
3022
Jeff Layton669abf42012-10-10 16:43:10 -04003023 error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base);
Nick Piggin31e6b012011-01-07 17:49:52 +11003024 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04003025 goto out;
Nick Piggin31e6b012011-01-07 17:49:52 +11003026
Al Virofe2d35f2011-03-05 22:58:25 -05003027 current->total_link_count = 0;
Jeff Layton669abf42012-10-10 16:43:10 -04003028 error = link_path_walk(pathname->name, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11003029 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04003030 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003031
Al Viro2675a4e2012-06-22 12:41:10 +04003032 error = do_last(nd, &path, file, op, &opened, pathname);
3033 while (unlikely(error > 0)) { /* trailing symlink */
Nick Piggin7b9337a2011-01-14 08:42:43 +00003034 struct path link = path;
Al Virodef4af32009-12-26 08:37:05 -05003035 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04003036 if (!(nd->flags & LOOKUP_FOLLOW)) {
Al Viro73d049a2011-03-11 12:08:24 -05003037 path_put_conditional(&path, nd);
3038 path_put(&nd->path);
Al Viro2675a4e2012-06-22 12:41:10 +04003039 error = -ELOOP;
Al Viro40b39132011-03-09 16:22:18 -05003040 break;
3041 }
Kees Cook800179c2012-07-25 17:29:07 -07003042 error = may_follow_link(&link, nd);
3043 if (unlikely(error))
3044 break;
Al Viro73d049a2011-03-11 12:08:24 -05003045 nd->flags |= LOOKUP_PARENT;
3046 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
Al Viro574197e2011-03-14 22:20:34 -04003047 error = follow_link(&link, nd, &cookie);
Al Viroc3e380b2011-02-23 13:39:45 -05003048 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04003049 break;
3050 error = do_last(nd, &path, file, op, &opened, pathname);
Al Viro574197e2011-03-14 22:20:34 -04003051 put_link(nd, &link, cookie);
Al Viro806b6812009-12-26 07:16:40 -05003052 }
Al Viro10fa8e62009-12-26 07:09:49 -05003053out:
Al Viro73d049a2011-03-11 12:08:24 -05003054 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
3055 path_put(&nd->root);
Al Virofe2d35f2011-03-05 22:58:25 -05003056 if (base)
3057 fput(base);
Al Viro2675a4e2012-06-22 12:41:10 +04003058 if (!(opened & FILE_OPENED)) {
3059 BUG_ON(!error);
Al Viro30d90492012-06-22 12:40:19 +04003060 put_filp(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003061 }
Al Viro2675a4e2012-06-22 12:41:10 +04003062 if (unlikely(error)) {
3063 if (error == -EOPENSTALE) {
3064 if (flags & LOOKUP_RCU)
3065 error = -ECHILD;
3066 else
3067 error = -ESTALE;
3068 }
3069 file = ERR_PTR(error);
3070 }
3071 return file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003072}
3073
Jeff Layton669abf42012-10-10 16:43:10 -04003074struct file *do_filp_open(int dfd, struct filename *pathname,
Al Virof9652e12013-06-11 08:23:01 +04003075 const struct open_flags *op)
Al Viro13aab422011-02-23 17:54:08 -05003076{
Al Viro73d049a2011-03-11 12:08:24 -05003077 struct nameidata nd;
Al Virof9652e12013-06-11 08:23:01 +04003078 int flags = op->lookup_flags;
Al Viro13aab422011-02-23 17:54:08 -05003079 struct file *filp;
3080
Al Viro73d049a2011-03-11 12:08:24 -05003081 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
Al Viro13aab422011-02-23 17:54:08 -05003082 if (unlikely(filp == ERR_PTR(-ECHILD)))
Al Viro73d049a2011-03-11 12:08:24 -05003083 filp = path_openat(dfd, pathname, &nd, op, flags);
Al Viro13aab422011-02-23 17:54:08 -05003084 if (unlikely(filp == ERR_PTR(-ESTALE)))
Al Viro73d049a2011-03-11 12:08:24 -05003085 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
Al Viro13aab422011-02-23 17:54:08 -05003086 return filp;
3087}
3088
Al Viro73d049a2011-03-11 12:08:24 -05003089struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
Al Virof9652e12013-06-11 08:23:01 +04003090 const char *name, const struct open_flags *op)
Al Viro73d049a2011-03-11 12:08:24 -05003091{
3092 struct nameidata nd;
3093 struct file *file;
Jeff Layton669abf42012-10-10 16:43:10 -04003094 struct filename filename = { .name = name };
Al Virof9652e12013-06-11 08:23:01 +04003095 int flags = op->lookup_flags | LOOKUP_ROOT;
Al Viro73d049a2011-03-11 12:08:24 -05003096
3097 nd.root.mnt = mnt;
3098 nd.root.dentry = dentry;
3099
Al Virobcda7652011-03-13 16:42:14 -04003100 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
Al Viro73d049a2011-03-11 12:08:24 -05003101 return ERR_PTR(-ELOOP);
3102
Jeff Layton669abf42012-10-10 16:43:10 -04003103 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU);
Al Viro73d049a2011-03-11 12:08:24 -05003104 if (unlikely(file == ERR_PTR(-ECHILD)))
Jeff Layton669abf42012-10-10 16:43:10 -04003105 file = path_openat(-1, &filename, &nd, op, flags);
Al Viro73d049a2011-03-11 12:08:24 -05003106 if (unlikely(file == ERR_PTR(-ESTALE)))
Jeff Layton669abf42012-10-10 16:43:10 -04003107 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL);
Al Viro73d049a2011-03-11 12:08:24 -05003108 return file;
3109}
3110
Jeff Layton1ac12b42012-12-11 12:10:06 -05003111struct dentry *kern_path_create(int dfd, const char *pathname,
3112 struct path *path, unsigned int lookup_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003113{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003114 struct dentry *dentry = ERR_PTR(-EEXIST);
Al Viroed75e952011-06-27 16:53:43 -04003115 struct nameidata nd;
Jan Karac30dabf2012-06-12 16:20:30 +02003116 int err2;
Jeff Layton1ac12b42012-12-11 12:10:06 -05003117 int error;
3118 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3119
3120 /*
3121 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3122 * other flags passed in are ignored!
3123 */
3124 lookup_flags &= LOOKUP_REVAL;
3125
3126 error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd);
Al Viroed75e952011-06-27 16:53:43 -04003127 if (error)
3128 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003129
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003130 /*
3131 * Yucky last component or no last component at all?
3132 * (foo/., foo/.., /////)
3133 */
Al Viroed75e952011-06-27 16:53:43 -04003134 if (nd.last_type != LAST_NORM)
3135 goto out;
3136 nd.flags &= ~LOOKUP_PARENT;
3137 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003138
Jan Karac30dabf2012-06-12 16:20:30 +02003139 /* don't fail immediately if it's r/o, at least try to report other errors */
3140 err2 = mnt_want_write(nd.path.mnt);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003141 /*
3142 * Do the final lookup.
3143 */
Al Viroed75e952011-06-27 16:53:43 -04003144 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3145 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003146 if (IS_ERR(dentry))
Al Viroa8104a92012-07-20 02:25:00 +04003147 goto unlock;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003148
Al Viroa8104a92012-07-20 02:25:00 +04003149 error = -EEXIST;
Al Viroe9baf6e2008-05-15 04:49:12 -04003150 if (dentry->d_inode)
Al Viroa8104a92012-07-20 02:25:00 +04003151 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003152 /*
3153 * Special case - lookup gave negative, but... we had foo/bar/
3154 * From the vfs_mknod() POV we just have a negative dentry -
3155 * all is fine. Let's be bastards - you had / on the end, you've
3156 * been asking for (non-existent) directory. -ENOENT for you.
3157 */
Al Viroed75e952011-06-27 16:53:43 -04003158 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
Al Viroa8104a92012-07-20 02:25:00 +04003159 error = -ENOENT;
Al Viroed75e952011-06-27 16:53:43 -04003160 goto fail;
Al Viroe9baf6e2008-05-15 04:49:12 -04003161 }
Jan Karac30dabf2012-06-12 16:20:30 +02003162 if (unlikely(err2)) {
3163 error = err2;
Al Viroa8104a92012-07-20 02:25:00 +04003164 goto fail;
Jan Karac30dabf2012-06-12 16:20:30 +02003165 }
Al Viroed75e952011-06-27 16:53:43 -04003166 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003167 return dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003168fail:
Al Viroa8104a92012-07-20 02:25:00 +04003169 dput(dentry);
3170 dentry = ERR_PTR(error);
3171unlock:
Al Viroed75e952011-06-27 16:53:43 -04003172 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Karac30dabf2012-06-12 16:20:30 +02003173 if (!err2)
3174 mnt_drop_write(nd.path.mnt);
Al Viroed75e952011-06-27 16:53:43 -04003175out:
3176 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003177 return dentry;
3178}
Al Virodae6ad82011-06-26 11:50:15 -04003179EXPORT_SYMBOL(kern_path_create);
3180
Al Viro921a1652012-07-20 01:15:31 +04003181void done_path_create(struct path *path, struct dentry *dentry)
3182{
3183 dput(dentry);
3184 mutex_unlock(&path->dentry->d_inode->i_mutex);
Al Viroa8104a92012-07-20 02:25:00 +04003185 mnt_drop_write(path->mnt);
Al Viro921a1652012-07-20 01:15:31 +04003186 path_put(path);
3187}
3188EXPORT_SYMBOL(done_path_create);
3189
Jeff Layton1ac12b42012-12-11 12:10:06 -05003190struct dentry *user_path_create(int dfd, const char __user *pathname,
3191 struct path *path, unsigned int lookup_flags)
Al Virodae6ad82011-06-26 11:50:15 -04003192{
Jeff Layton91a27b22012-10-10 15:25:28 -04003193 struct filename *tmp = getname(pathname);
Al Virodae6ad82011-06-26 11:50:15 -04003194 struct dentry *res;
3195 if (IS_ERR(tmp))
3196 return ERR_CAST(tmp);
Jeff Layton1ac12b42012-12-11 12:10:06 -05003197 res = kern_path_create(dfd, tmp->name, path, lookup_flags);
Al Virodae6ad82011-06-26 11:50:15 -04003198 putname(tmp);
3199 return res;
3200}
3201EXPORT_SYMBOL(user_path_create);
3202
Al Viro1a67aaf2011-07-26 01:52:52 -04003203int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003204{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003205 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003206
3207 if (error)
3208 return error;
3209
Eric W. Biederman975d6b32011-11-13 12:16:43 -08003210 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003211 return -EPERM;
3212
Al Viroacfa4382008-12-04 10:06:33 -05003213 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003214 return -EPERM;
3215
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07003216 error = devcgroup_inode_mknod(mode, dev);
3217 if (error)
3218 return error;
3219
Linus Torvalds1da177e2005-04-16 15:20:36 -07003220 error = security_inode_mknod(dir, dentry, mode, dev);
3221 if (error)
3222 return error;
3223
Linus Torvalds1da177e2005-04-16 15:20:36 -07003224 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003225 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003226 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003227 return error;
3228}
3229
Al Virof69aac02011-07-26 04:31:40 -04003230static int may_mknod(umode_t mode)
Dave Hansen463c3192008-02-15 14:37:57 -08003231{
3232 switch (mode & S_IFMT) {
3233 case S_IFREG:
3234 case S_IFCHR:
3235 case S_IFBLK:
3236 case S_IFIFO:
3237 case S_IFSOCK:
3238 case 0: /* zero mode translates to S_IFREG */
3239 return 0;
3240 case S_IFDIR:
3241 return -EPERM;
3242 default:
3243 return -EINVAL;
3244 }
3245}
3246
Al Viro8208a222011-07-25 17:32:17 -04003247SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003248 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003249{
Al Viro2ad94ae2008-07-21 09:32:51 -04003250 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003251 struct path path;
3252 int error;
Jeff Layton972567f2012-12-20 16:00:10 -05003253 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003254
Al Viro8e4bfca2012-07-20 01:17:26 +04003255 error = may_mknod(mode);
3256 if (error)
3257 return error;
Jeff Layton972567f2012-12-20 16:00:10 -05003258retry:
3259 dentry = user_path_create(dfd, filename, &path, lookup_flags);
Al Virodae6ad82011-06-26 11:50:15 -04003260 if (IS_ERR(dentry))
3261 return PTR_ERR(dentry);
Al Viro2ad94ae2008-07-21 09:32:51 -04003262
Al Virodae6ad82011-06-26 11:50:15 -04003263 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003264 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003265 error = security_path_mknod(&path, dentry, mode, dev);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003266 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04003267 goto out;
Dave Hansen463c3192008-02-15 14:37:57 -08003268 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003269 case 0: case S_IFREG:
Al Viro312b63f2012-06-10 18:09:36 -04003270 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003271 break;
3272 case S_IFCHR: case S_IFBLK:
Al Virodae6ad82011-06-26 11:50:15 -04003273 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003274 new_decode_dev(dev));
3275 break;
3276 case S_IFIFO: case S_IFSOCK:
Al Virodae6ad82011-06-26 11:50:15 -04003277 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003278 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003279 }
Al Viroa8104a92012-07-20 02:25:00 +04003280out:
Al Viro921a1652012-07-20 01:15:31 +04003281 done_path_create(&path, dentry);
Jeff Layton972567f2012-12-20 16:00:10 -05003282 if (retry_estale(error, lookup_flags)) {
3283 lookup_flags |= LOOKUP_REVAL;
3284 goto retry;
3285 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003286 return error;
3287}
3288
Al Viro8208a222011-07-25 17:32:17 -04003289SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003290{
3291 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3292}
3293
Al Viro18bb1db2011-07-26 01:41:39 -04003294int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003295{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003296 int error = may_create(dir, dentry);
Al Viro8de52772012-02-06 12:45:27 -05003297 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003298
3299 if (error)
3300 return error;
3301
Al Viroacfa4382008-12-04 10:06:33 -05003302 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003303 return -EPERM;
3304
3305 mode &= (S_IRWXUGO|S_ISVTX);
3306 error = security_inode_mkdir(dir, dentry, mode);
3307 if (error)
3308 return error;
3309
Al Viro8de52772012-02-06 12:45:27 -05003310 if (max_links && dir->i_nlink >= max_links)
3311 return -EMLINK;
3312
Linus Torvalds1da177e2005-04-16 15:20:36 -07003313 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003314 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003315 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003316 return error;
3317}
3318
Al Viroa218d0f2011-11-21 14:59:34 -05003319SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003320{
Dave Hansen6902d922006-09-30 23:29:01 -07003321 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003322 struct path path;
3323 int error;
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003324 unsigned int lookup_flags = LOOKUP_DIRECTORY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003325
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003326retry:
3327 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07003328 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003329 return PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003330
Al Virodae6ad82011-06-26 11:50:15 -04003331 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003332 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003333 error = security_path_mkdir(&path, dentry, mode);
Al Viroa8104a92012-07-20 02:25:00 +04003334 if (!error)
3335 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
Al Viro921a1652012-07-20 01:15:31 +04003336 done_path_create(&path, dentry);
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003337 if (retry_estale(error, lookup_flags)) {
3338 lookup_flags |= LOOKUP_REVAL;
3339 goto retry;
3340 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003341 return error;
3342}
3343
Al Viroa218d0f2011-11-21 14:59:34 -05003344SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003345{
3346 return sys_mkdirat(AT_FDCWD, pathname, mode);
3347}
3348
Linus Torvalds1da177e2005-04-16 15:20:36 -07003349/*
Sage Weila71905f2011-05-24 13:06:08 -07003350 * The dentry_unhash() helper will try to drop the dentry early: we
J. Bruce Fieldsc0d02592012-02-15 11:48:40 -05003351 * should have a usage count of 1 if we're the only user of this
Sage Weila71905f2011-05-24 13:06:08 -07003352 * dentry, and if that is true (possibly after pruning the dcache),
3353 * then we drop the dentry now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003354 *
3355 * A low-level filesystem can, if it choses, legally
3356 * do a
3357 *
3358 * if (!d_unhashed(dentry))
3359 * return -EBUSY;
3360 *
3361 * if it cannot handle the case of removing a directory
3362 * that is still in use by something else..
3363 */
3364void dentry_unhash(struct dentry *dentry)
3365{
Vasily Averindc168422006-12-06 20:37:07 -08003366 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003367 spin_lock(&dentry->d_lock);
Waiman Long98474232013-08-28 18:24:59 -07003368 if (dentry->d_lockref.count == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003369 __d_drop(dentry);
3370 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003371}
3372
3373int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3374{
3375 int error = may_delete(dir, dentry, 1);
3376
3377 if (error)
3378 return error;
3379
Al Viroacfa4382008-12-04 10:06:33 -05003380 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003381 return -EPERM;
3382
Al Viro1d2ef592011-09-14 18:55:41 +01003383 dget(dentry);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003384 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003385
Sage Weil912dbc12011-05-24 13:06:11 -07003386 error = -EBUSY;
3387 if (d_mountpoint(dentry))
3388 goto out;
3389
3390 error = security_inode_rmdir(dir, dentry);
3391 if (error)
3392 goto out;
3393
Sage Weil3cebde22011-05-29 21:20:59 -07003394 shrink_dcache_parent(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003395 error = dir->i_op->rmdir(dir, dentry);
3396 if (error)
3397 goto out;
3398
3399 dentry->d_inode->i_flags |= S_DEAD;
3400 dont_mount(dentry);
3401
3402out:
3403 mutex_unlock(&dentry->d_inode->i_mutex);
Al Viro1d2ef592011-09-14 18:55:41 +01003404 dput(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003405 if (!error)
3406 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003407 return error;
3408}
3409
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003410static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003411{
3412 int error = 0;
Jeff Layton91a27b22012-10-10 15:25:28 -04003413 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003414 struct dentry *dentry;
3415 struct nameidata nd;
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003416 unsigned int lookup_flags = 0;
3417retry:
3418 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04003419 if (IS_ERR(name))
3420 return PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003421
3422 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003423 case LAST_DOTDOT:
3424 error = -ENOTEMPTY;
3425 goto exit1;
3426 case LAST_DOT:
3427 error = -EINVAL;
3428 goto exit1;
3429 case LAST_ROOT:
3430 error = -EBUSY;
3431 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003432 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003433
3434 nd.flags &= ~LOOKUP_PARENT;
Jan Karac30dabf2012-06-12 16:20:30 +02003435 error = mnt_want_write(nd.path.mnt);
3436 if (error)
3437 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003438
Jan Blunck4ac91372008-02-14 19:34:32 -08003439 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08003440 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003441 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003442 if (IS_ERR(dentry))
3443 goto exit2;
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003444 if (!dentry->d_inode) {
3445 error = -ENOENT;
3446 goto exit3;
3447 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003448 error = security_path_rmdir(&nd.path, dentry);
3449 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003450 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08003451 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Dave Hansen06227532008-02-15 14:37:34 -08003452exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07003453 dput(dentry);
3454exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08003455 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Karac30dabf2012-06-12 16:20:30 +02003456 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003457exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003458 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003459 putname(name);
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003460 if (retry_estale(error, lookup_flags)) {
3461 lookup_flags |= LOOKUP_REVAL;
3462 goto retry;
3463 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003464 return error;
3465}
3466
Heiko Carstens3cdad422009-01-14 14:14:22 +01003467SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003468{
3469 return do_rmdir(AT_FDCWD, pathname);
3470}
3471
Linus Torvalds1da177e2005-04-16 15:20:36 -07003472int vfs_unlink(struct inode *dir, struct dentry *dentry)
3473{
3474 int error = may_delete(dir, dentry, 0);
3475
3476 if (error)
3477 return error;
3478
Al Viroacfa4382008-12-04 10:06:33 -05003479 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003480 return -EPERM;
3481
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003482 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003483 if (d_mountpoint(dentry))
3484 error = -EBUSY;
3485 else {
3486 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05003487 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003488 error = dir->i_op->unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05003489 if (!error)
Al Virod83c49f2010-04-30 17:17:09 -04003490 dont_mount(dentry);
Al Virobec10522010-03-03 14:12:08 -05003491 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003492 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003493 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003494
3495 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3496 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08003497 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003498 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003499 }
Robert Love0eeca282005-07-12 17:06:03 -04003500
Linus Torvalds1da177e2005-04-16 15:20:36 -07003501 return error;
3502}
3503
3504/*
3505 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003506 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07003507 * writeout happening, and we don't want to prevent access to the directory
3508 * while waiting on the I/O.
3509 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003510static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003511{
Al Viro2ad94ae2008-07-21 09:32:51 -04003512 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04003513 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003514 struct dentry *dentry;
3515 struct nameidata nd;
3516 struct inode *inode = NULL;
Jeff Layton5d18f812012-12-20 16:38:04 -05003517 unsigned int lookup_flags = 0;
3518retry:
3519 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04003520 if (IS_ERR(name))
3521 return PTR_ERR(name);
Al Viro2ad94ae2008-07-21 09:32:51 -04003522
Linus Torvalds1da177e2005-04-16 15:20:36 -07003523 error = -EISDIR;
3524 if (nd.last_type != LAST_NORM)
3525 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003526
3527 nd.flags &= ~LOOKUP_PARENT;
Jan Karac30dabf2012-06-12 16:20:30 +02003528 error = mnt_want_write(nd.path.mnt);
3529 if (error)
3530 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003531
Jan Blunck4ac91372008-02-14 19:34:32 -08003532 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08003533 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003534 error = PTR_ERR(dentry);
3535 if (!IS_ERR(dentry)) {
3536 /* Why not before? Because we want correct error value */
Török Edwin50338b82011-06-16 00:06:14 +03003537 if (nd.last.name[nd.last.len])
3538 goto slashes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003539 inode = dentry->d_inode;
Török Edwin50338b82011-06-16 00:06:14 +03003540 if (!inode)
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003541 goto slashes;
3542 ihold(inode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003543 error = security_path_unlink(&nd.path, dentry);
3544 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003545 goto exit2;
Jan Blunck4ac91372008-02-14 19:34:32 -08003546 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Jan Karac30dabf2012-06-12 16:20:30 +02003547exit2:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003548 dput(dentry);
3549 }
Jan Blunck4ac91372008-02-14 19:34:32 -08003550 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003551 if (inode)
3552 iput(inode); /* truncate the inode here */
Jan Karac30dabf2012-06-12 16:20:30 +02003553 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003554exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003555 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003556 putname(name);
Jeff Layton5d18f812012-12-20 16:38:04 -05003557 if (retry_estale(error, lookup_flags)) {
3558 lookup_flags |= LOOKUP_REVAL;
3559 inode = NULL;
3560 goto retry;
3561 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003562 return error;
3563
3564slashes:
3565 error = !dentry->d_inode ? -ENOENT :
3566 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
3567 goto exit2;
3568}
3569
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003570SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003571{
3572 if ((flag & ~AT_REMOVEDIR) != 0)
3573 return -EINVAL;
3574
3575 if (flag & AT_REMOVEDIR)
3576 return do_rmdir(dfd, pathname);
3577
3578 return do_unlinkat(dfd, pathname);
3579}
3580
Heiko Carstens3480b252009-01-14 14:14:16 +01003581SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003582{
3583 return do_unlinkat(AT_FDCWD, pathname);
3584}
3585
Miklos Szeredidb2e7472008-06-24 16:50:16 +02003586int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003587{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003588 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003589
3590 if (error)
3591 return error;
3592
Al Viroacfa4382008-12-04 10:06:33 -05003593 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003594 return -EPERM;
3595
3596 error = security_inode_symlink(dir, dentry, oldname);
3597 if (error)
3598 return error;
3599
Linus Torvalds1da177e2005-04-16 15:20:36 -07003600 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003601 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003602 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003603 return error;
3604}
3605
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003606SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3607 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003608{
Al Viro2ad94ae2008-07-21 09:32:51 -04003609 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04003610 struct filename *from;
Dave Hansen6902d922006-09-30 23:29:01 -07003611 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003612 struct path path;
Jeff Laytonf46d3562012-12-11 12:10:08 -05003613 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003614
3615 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04003616 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003617 return PTR_ERR(from);
Jeff Laytonf46d3562012-12-11 12:10:08 -05003618retry:
3619 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07003620 error = PTR_ERR(dentry);
3621 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003622 goto out_putname;
Dave Hansen6902d922006-09-30 23:29:01 -07003623
Jeff Layton91a27b22012-10-10 15:25:28 -04003624 error = security_path_symlink(&path, dentry, from->name);
Al Viroa8104a92012-07-20 02:25:00 +04003625 if (!error)
Jeff Layton91a27b22012-10-10 15:25:28 -04003626 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
Al Viro921a1652012-07-20 01:15:31 +04003627 done_path_create(&path, dentry);
Jeff Laytonf46d3562012-12-11 12:10:08 -05003628 if (retry_estale(error, lookup_flags)) {
3629 lookup_flags |= LOOKUP_REVAL;
3630 goto retry;
3631 }
Dave Hansen6902d922006-09-30 23:29:01 -07003632out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003633 putname(from);
3634 return error;
3635}
3636
Heiko Carstens3480b252009-01-14 14:14:16 +01003637SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003638{
3639 return sys_symlinkat(oldname, AT_FDCWD, newname);
3640}
3641
Linus Torvalds1da177e2005-04-16 15:20:36 -07003642int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
3643{
3644 struct inode *inode = old_dentry->d_inode;
Al Viro8de52772012-02-06 12:45:27 -05003645 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003646 int error;
3647
3648 if (!inode)
3649 return -ENOENT;
3650
Miklos Szeredia95164d2008-07-30 15:08:48 +02003651 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003652 if (error)
3653 return error;
3654
3655 if (dir->i_sb != inode->i_sb)
3656 return -EXDEV;
3657
3658 /*
3659 * A link to an append-only or immutable file cannot be created.
3660 */
3661 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3662 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05003663 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003664 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003665 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003666 return -EPERM;
3667
3668 error = security_inode_link(old_dentry, dir, new_dentry);
3669 if (error)
3670 return error;
3671
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003672 mutex_lock(&inode->i_mutex);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05303673 /* Make sure we don't allow creating hardlink to an unlinked file */
Al Virof4e0c302013-06-11 08:34:36 +04003674 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05303675 error = -ENOENT;
Al Viro8de52772012-02-06 12:45:27 -05003676 else if (max_links && inode->i_nlink >= max_links)
3677 error = -EMLINK;
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05303678 else
3679 error = dir->i_op->link(old_dentry, dir, new_dentry);
Al Virof4e0c302013-06-11 08:34:36 +04003680
3681 if (!error && (inode->i_state & I_LINKABLE)) {
3682 spin_lock(&inode->i_lock);
3683 inode->i_state &= ~I_LINKABLE;
3684 spin_unlock(&inode->i_lock);
3685 }
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003686 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07003687 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003688 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003689 return error;
3690}
3691
3692/*
3693 * Hardlinks are often used in delicate situations. We avoid
3694 * security-related surprises by not following symlinks on the
3695 * newname. --KAB
3696 *
3697 * We don't follow them on the oldname either to be compatible
3698 * with linux 2.0, and to avoid hard-linking to directories
3699 * and other special files. --ADM
3700 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003701SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3702 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003703{
3704 struct dentry *new_dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003705 struct path old_path, new_path;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303706 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003707 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003708
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303709 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003710 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303711 /*
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07003712 * To use null names we require CAP_DAC_READ_SEARCH
3713 * This ensures that not everyone will be able to create
3714 * handlink using the passed filedescriptor.
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303715 */
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07003716 if (flags & AT_EMPTY_PATH) {
3717 if (!capable(CAP_DAC_READ_SEARCH))
3718 return -ENOENT;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303719 how = LOOKUP_EMPTY;
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07003720 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003721
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303722 if (flags & AT_SYMLINK_FOLLOW)
3723 how |= LOOKUP_FOLLOW;
Jeff Layton442e31c2012-12-20 16:15:38 -05003724retry:
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303725 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003726 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04003727 return error;
3728
Jeff Layton442e31c2012-12-20 16:15:38 -05003729 new_dentry = user_path_create(newdfd, newname, &new_path,
3730 (how & LOOKUP_REVAL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003731 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003732 if (IS_ERR(new_dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003733 goto out;
3734
3735 error = -EXDEV;
3736 if (old_path.mnt != new_path.mnt)
3737 goto out_dput;
Kees Cook800179c2012-07-25 17:29:07 -07003738 error = may_linkat(&old_path);
3739 if (unlikely(error))
3740 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04003741 error = security_path_link(old_path.dentry, &new_path, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003742 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04003743 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04003744 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
Dave Hansen75c3f292008-02-15 14:37:45 -08003745out_dput:
Al Viro921a1652012-07-20 01:15:31 +04003746 done_path_create(&new_path, new_dentry);
Jeff Layton442e31c2012-12-20 16:15:38 -05003747 if (retry_estale(error, how)) {
3748 how |= LOOKUP_REVAL;
3749 goto retry;
3750 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003751out:
Al Viro2d8f3032008-07-22 09:59:21 -04003752 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003753
3754 return error;
3755}
3756
Heiko Carstens3480b252009-01-14 14:14:16 +01003757SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003758{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003759 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003760}
3761
Linus Torvalds1da177e2005-04-16 15:20:36 -07003762/*
3763 * The worst of all namespace operations - renaming directory. "Perverted"
3764 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3765 * Problems:
3766 * a) we can get into loop creation. Check is done in is_subdir().
3767 * b) race potential - two innocent renames can create a loop together.
3768 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003769 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07003770 * story.
3771 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003772 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07003773 * whether the target exists). Solution: try to be smart with locking
3774 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003775 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07003776 * move will be locked. Thus we can rank directories by the tree
3777 * (ancestors first) and rank all non-directories after them.
3778 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003779 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003780 * HOWEVER, it relies on the assumption that any object with ->lookup()
3781 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3782 * we'd better make sure that there's no link(2) for them.
Sage Weile4eaac02011-05-24 13:06:07 -07003783 * d) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003784 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07003785 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05003786 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07003787 * locking].
3788 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07003789static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3790 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003791{
3792 int error = 0;
Sage Weil9055cba2011-05-24 13:06:12 -07003793 struct inode *target = new_dentry->d_inode;
Al Viro8de52772012-02-06 12:45:27 -05003794 unsigned max_links = new_dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003795
3796 /*
3797 * If we are going to change the parent - check write permissions,
3798 * we'll need to flip '..'.
3799 */
3800 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04003801 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003802 if (error)
3803 return error;
3804 }
3805
3806 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3807 if (error)
3808 return error;
3809
Al Viro1d2ef592011-09-14 18:55:41 +01003810 dget(new_dentry);
Al Virod83c49f2010-04-30 17:17:09 -04003811 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003812 mutex_lock(&target->i_mutex);
Sage Weil9055cba2011-05-24 13:06:12 -07003813
3814 error = -EBUSY;
3815 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3816 goto out;
3817
Al Viro8de52772012-02-06 12:45:27 -05003818 error = -EMLINK;
3819 if (max_links && !target && new_dir != old_dir &&
3820 new_dir->i_nlink >= max_links)
3821 goto out;
3822
Sage Weil3cebde22011-05-29 21:20:59 -07003823 if (target)
3824 shrink_dcache_parent(new_dentry);
Sage Weil9055cba2011-05-24 13:06:12 -07003825 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3826 if (error)
3827 goto out;
3828
Linus Torvalds1da177e2005-04-16 15:20:36 -07003829 if (target) {
Sage Weil9055cba2011-05-24 13:06:12 -07003830 target->i_flags |= S_DEAD;
3831 dont_mount(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003832 }
Sage Weil9055cba2011-05-24 13:06:12 -07003833out:
3834 if (target)
3835 mutex_unlock(&target->i_mutex);
Al Viro1d2ef592011-09-14 18:55:41 +01003836 dput(new_dentry);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07003837 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07003838 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3839 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003840 return error;
3841}
3842
Adrian Bunk75c96f82005-05-05 16:16:09 -07003843static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3844 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003845{
Sage Weil51892bb2011-05-24 13:06:13 -07003846 struct inode *target = new_dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003847 int error;
3848
3849 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3850 if (error)
3851 return error;
3852
3853 dget(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003854 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003855 mutex_lock(&target->i_mutex);
Sage Weil51892bb2011-05-24 13:06:13 -07003856
3857 error = -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003858 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
Sage Weil51892bb2011-05-24 13:06:13 -07003859 goto out;
3860
3861 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3862 if (error)
3863 goto out;
3864
3865 if (target)
3866 dont_mount(new_dentry);
3867 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3868 d_move(old_dentry, new_dentry);
3869out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003870 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003871 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003872 dput(new_dentry);
3873 return error;
3874}
3875
3876int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3877 struct inode *new_dir, struct dentry *new_dentry)
3878{
3879 int error;
3880 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Eric Paris59b0df22010-02-08 12:53:52 -05003881 const unsigned char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003882
3883 if (old_dentry->d_inode == new_dentry->d_inode)
3884 return 0;
3885
3886 error = may_delete(old_dir, old_dentry, is_dir);
3887 if (error)
3888 return error;
3889
3890 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02003891 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003892 else
3893 error = may_delete(new_dir, new_dentry, is_dir);
3894 if (error)
3895 return error;
3896
Al Viroacfa4382008-12-04 10:06:33 -05003897 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003898 return -EPERM;
3899
Robert Love0eeca282005-07-12 17:06:03 -04003900 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3901
Linus Torvalds1da177e2005-04-16 15:20:36 -07003902 if (is_dir)
3903 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3904 else
3905 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
Al Viro123df292009-12-25 04:57:57 -05003906 if (!error)
3907 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04003908 new_dentry->d_inode, old_dentry);
Robert Love0eeca282005-07-12 17:06:03 -04003909 fsnotify_oldname_free(old_name);
3910
Linus Torvalds1da177e2005-04-16 15:20:36 -07003911 return error;
3912}
3913
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003914SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3915 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003916{
Al Viro2ad94ae2008-07-21 09:32:51 -04003917 struct dentry *old_dir, *new_dir;
3918 struct dentry *old_dentry, *new_dentry;
3919 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003920 struct nameidata oldnd, newnd;
Jeff Layton91a27b22012-10-10 15:25:28 -04003921 struct filename *from;
3922 struct filename *to;
Jeff Laytonc6a94282012-12-11 12:10:10 -05003923 unsigned int lookup_flags = 0;
3924 bool should_retry = false;
Al Viro2ad94ae2008-07-21 09:32:51 -04003925 int error;
Jeff Laytonc6a94282012-12-11 12:10:10 -05003926retry:
3927 from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04003928 if (IS_ERR(from)) {
3929 error = PTR_ERR(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003930 goto exit;
Jeff Layton91a27b22012-10-10 15:25:28 -04003931 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003932
Jeff Laytonc6a94282012-12-11 12:10:10 -05003933 to = user_path_parent(newdfd, newname, &newnd, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04003934 if (IS_ERR(to)) {
3935 error = PTR_ERR(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003936 goto exit1;
Jeff Layton91a27b22012-10-10 15:25:28 -04003937 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003938
3939 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08003940 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003941 goto exit2;
3942
Jan Blunck4ac91372008-02-14 19:34:32 -08003943 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003944 error = -EBUSY;
3945 if (oldnd.last_type != LAST_NORM)
3946 goto exit2;
3947
Jan Blunck4ac91372008-02-14 19:34:32 -08003948 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003949 if (newnd.last_type != LAST_NORM)
3950 goto exit2;
3951
Jan Karac30dabf2012-06-12 16:20:30 +02003952 error = mnt_want_write(oldnd.path.mnt);
3953 if (error)
3954 goto exit2;
3955
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003956 oldnd.flags &= ~LOOKUP_PARENT;
3957 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09003958 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003959
Linus Torvalds1da177e2005-04-16 15:20:36 -07003960 trap = lock_rename(new_dir, old_dir);
3961
Christoph Hellwig49705b72005-11-08 21:35:06 -08003962 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003963 error = PTR_ERR(old_dentry);
3964 if (IS_ERR(old_dentry))
3965 goto exit3;
3966 /* source must exist */
3967 error = -ENOENT;
3968 if (!old_dentry->d_inode)
3969 goto exit4;
3970 /* unless the source is a directory trailing slashes give -ENOTDIR */
3971 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3972 error = -ENOTDIR;
3973 if (oldnd.last.name[oldnd.last.len])
3974 goto exit4;
3975 if (newnd.last.name[newnd.last.len])
3976 goto exit4;
3977 }
3978 /* source should not be ancestor of target */
3979 error = -EINVAL;
3980 if (old_dentry == trap)
3981 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08003982 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003983 error = PTR_ERR(new_dentry);
3984 if (IS_ERR(new_dentry))
3985 goto exit4;
3986 /* target should not be an ancestor of source */
3987 error = -ENOTEMPTY;
3988 if (new_dentry == trap)
3989 goto exit5;
3990
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003991 error = security_path_rename(&oldnd.path, old_dentry,
3992 &newnd.path, new_dentry);
3993 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003994 goto exit5;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003995 error = vfs_rename(old_dir->d_inode, old_dentry,
3996 new_dir->d_inode, new_dentry);
3997exit5:
3998 dput(new_dentry);
3999exit4:
4000 dput(old_dentry);
4001exit3:
4002 unlock_rename(new_dir, old_dir);
Jan Karac30dabf2012-06-12 16:20:30 +02004003 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004004exit2:
Jeff Laytonc6a94282012-12-11 12:10:10 -05004005 if (retry_estale(error, lookup_flags))
4006 should_retry = true;
Jan Blunck1d957f92008-02-14 19:34:35 -08004007 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04004008 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004009exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08004010 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004011 putname(from);
Jeff Laytonc6a94282012-12-11 12:10:10 -05004012 if (should_retry) {
4013 should_retry = false;
4014 lookup_flags |= LOOKUP_REVAL;
4015 goto retry;
4016 }
Al Viro2ad94ae2008-07-21 09:32:51 -04004017exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004018 return error;
4019}
4020
Heiko Carstensa26eab22009-01-14 14:14:17 +01004021SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004022{
4023 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
4024}
4025
Linus Torvalds1da177e2005-04-16 15:20:36 -07004026int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
4027{
4028 int len;
4029
4030 len = PTR_ERR(link);
4031 if (IS_ERR(link))
4032 goto out;
4033
4034 len = strlen(link);
4035 if (len > (unsigned) buflen)
4036 len = buflen;
4037 if (copy_to_user(buffer, link, len))
4038 len = -EFAULT;
4039out:
4040 return len;
4041}
4042
4043/*
4044 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4045 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4046 * using) it for any given inode is up to filesystem.
4047 */
4048int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4049{
4050 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004051 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07004052 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004053
Linus Torvalds1da177e2005-04-16 15:20:36 -07004054 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004055 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07004056 if (IS_ERR(cookie))
4057 return PTR_ERR(cookie);
4058
4059 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
4060 if (dentry->d_inode->i_op->put_link)
4061 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
4062 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004063}
4064
4065int vfs_follow_link(struct nameidata *nd, const char *link)
4066{
4067 return __vfs_follow_link(nd, link);
4068}
4069
4070/* get the link contents into pagecache */
4071static char *page_getlink(struct dentry * dentry, struct page **ppage)
4072{
Duane Griffinebd09ab2008-12-19 20:47:12 +00004073 char *kaddr;
4074 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004075 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07004076 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004077 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07004078 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004079 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00004080 kaddr = kmap(page);
4081 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4082 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004083}
4084
4085int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4086{
4087 struct page *page = NULL;
4088 char *s = page_getlink(dentry, &page);
4089 int res = vfs_readlink(dentry,buffer,buflen,s);
4090 if (page) {
4091 kunmap(page);
4092 page_cache_release(page);
4093 }
4094 return res;
4095}
4096
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004097void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004098{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004099 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004100 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004101 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004102}
4103
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004104void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004105{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004106 struct page *page = cookie;
4107
4108 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004109 kunmap(page);
4110 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004111 }
4112}
4113
Nick Piggin54566b22009-01-04 12:00:53 -08004114/*
4115 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4116 */
4117int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004118{
4119 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004120 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07004121 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08004122 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004123 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08004124 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4125 if (nofs)
4126 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004127
NeilBrown7e53cac2006-03-25 03:07:57 -08004128retry:
Nick Pigginafddba42007-10-16 01:25:01 -07004129 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08004130 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004131 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07004132 goto fail;
4133
Cong Wange8e3c3d2011-11-25 23:14:27 +08004134 kaddr = kmap_atomic(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004135 memcpy(kaddr, symname, len-1);
Cong Wange8e3c3d2011-11-25 23:14:27 +08004136 kunmap_atomic(kaddr);
Nick Pigginafddba42007-10-16 01:25:01 -07004137
4138 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4139 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004140 if (err < 0)
4141 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07004142 if (err < len-1)
4143 goto retry;
4144
Linus Torvalds1da177e2005-04-16 15:20:36 -07004145 mark_inode_dirty(inode);
4146 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004147fail:
4148 return err;
4149}
4150
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004151int page_symlink(struct inode *inode, const char *symname, int len)
4152{
4153 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08004154 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004155}
4156
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08004157const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004158 .readlink = generic_readlink,
4159 .follow_link = page_follow_link_light,
4160 .put_link = page_put_link,
4161};
4162
Al Viro2d8f3032008-07-22 09:59:21 -04004163EXPORT_SYMBOL(user_path_at);
David Howellscc53ce52011-01-14 18:45:26 +00004164EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004165EXPORT_SYMBOL(follow_down);
4166EXPORT_SYMBOL(follow_up);
Al Virof6d2ac52012-08-26 12:55:54 -04004167EXPORT_SYMBOL(get_write_access); /* nfsd */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004168EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004169EXPORT_SYMBOL(lookup_one_len);
4170EXPORT_SYMBOL(page_follow_link_light);
4171EXPORT_SYMBOL(page_put_link);
4172EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004173EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004174EXPORT_SYMBOL(page_symlink);
4175EXPORT_SYMBOL(page_symlink_inode_operations);
Al Virod1811462008-08-02 00:49:18 -04004176EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07004177EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04004178EXPORT_SYMBOL(inode_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004179EXPORT_SYMBOL(unlock_rename);
4180EXPORT_SYMBOL(vfs_create);
4181EXPORT_SYMBOL(vfs_follow_link);
4182EXPORT_SYMBOL(vfs_link);
4183EXPORT_SYMBOL(vfs_mkdir);
4184EXPORT_SYMBOL(vfs_mknod);
4185EXPORT_SYMBOL(generic_permission);
4186EXPORT_SYMBOL(vfs_readlink);
4187EXPORT_SYMBOL(vfs_rename);
4188EXPORT_SYMBOL(vfs_rmdir);
4189EXPORT_SYMBOL(vfs_symlink);
4190EXPORT_SYMBOL(vfs_unlink);
4191EXPORT_SYMBOL(dentry_unhash);
4192EXPORT_SYMBOL(generic_readlink);