blob: 96ca11dea4a20c56b89ca126274b68089235346f [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 Torvalds99d263d2014-09-13 11:30:10 -070037#include <linux/hash.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include <asm/uaccess.h>
39
Eric Parise81e3f42009-12-04 15:47:36 -050040#include "internal.h"
Al Viroc7105362011-11-24 18:22:03 -050041#include "mount.h"
Eric Parise81e3f42009-12-04 15:47:36 -050042
Linus Torvalds1da177e2005-04-16 15:20:36 -070043/* [Feb-1997 T. Schoebel-Theuer]
44 * Fundamental changes in the pathname lookup mechanisms (namei)
45 * were necessary because of omirr. The reason is that omirr needs
46 * to know the _real_ pathname, not the user-supplied one, in case
47 * of symlinks (and also when transname replacements occur).
48 *
49 * The new code replaces the old recursive symlink resolution with
50 * an iterative one (in case of non-nested symlink chains). It does
51 * this with calls to <fs>_follow_link().
52 * As a side effect, dir_namei(), _namei() and follow_link() are now
53 * replaced with a single function lookup_dentry() that can handle all
54 * the special cases of the former code.
55 *
56 * With the new dcache, the pathname is stored at each inode, at least as
57 * long as the refcount of the inode is positive. As a side effect, the
58 * size of the dcache depends on the inode cache and thus is dynamic.
59 *
60 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
61 * resolution to correspond with current state of the code.
62 *
63 * Note that the symlink resolution is not *completely* iterative.
64 * There is still a significant amount of tail- and mid- recursion in
65 * the algorithm. Also, note that <fs>_readlink() is not used in
66 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
67 * may return different results than <fs>_follow_link(). Many virtual
68 * filesystems (including /proc) exhibit this behavior.
69 */
70
71/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
72 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
73 * and the name already exists in form of a symlink, try to create the new
74 * name indicated by the symlink. The old code always complained that the
75 * name already exists, due to not following the symlink even if its target
76 * is nonexistent. The new semantics affects also mknod() and link() when
Lucas De Marchi25985ed2011-03-30 22:57:33 -030077 * the name is a symlink pointing to a non-existent name.
Linus Torvalds1da177e2005-04-16 15:20:36 -070078 *
79 * I don't know which semantics is the right one, since I have no access
80 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
81 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
82 * "old" one. Personally, I think the new semantics is much more logical.
83 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
84 * file does succeed in both HP-UX and SunOs, but not in Solaris
85 * and in the old Linux semantics.
86 */
87
88/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
89 * semantics. See the comments in "open_namei" and "do_link" below.
90 *
91 * [10-Sep-98 Alan Modra] Another symlink change.
92 */
93
94/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
95 * inside the path - always follow.
96 * in the last component in creation/removal/renaming - never follow.
97 * if LOOKUP_FOLLOW passed - follow.
98 * if the pathname has trailing slashes - follow.
99 * otherwise - don't follow.
100 * (applied in that order).
101 *
102 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
103 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
104 * During the 2.4 we need to fix the userland stuff depending on it -
105 * hopefully we will be able to get rid of that wart in 2.5. So far only
106 * XEmacs seems to be relying on it...
107 */
108/*
109 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800110 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 * any extra contention...
112 */
113
114/* In order to reduce some races, while at the same time doing additional
115 * checking and hopefully speeding things up, we copy filenames to the
116 * kernel data space before using them..
117 *
118 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
119 * PATH_MAX includes the nul terminator --RR.
120 */
Jeff Layton91a27b22012-10-10 15:25:28 -0400121
Jeff Layton7950e382012-10-10 16:43:13 -0400122#define EMBEDDED_NAME_MAX (PATH_MAX - sizeof(struct filename))
123
David Drysdale51f39a12014-12-12 16:57:29 -0800124struct filename *
Jeff Layton91a27b22012-10-10 15:25:28 -0400125getname_flags(const char __user *filename, int flags, int *empty)
126{
127 struct filename *result, *err;
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700128 int len;
Jeff Layton7950e382012-10-10 16:43:13 -0400129 long max;
130 char *kname;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131
Jeff Layton7ac86262012-10-10 15:25:28 -0400132 result = audit_reusename(filename);
133 if (result)
134 return result;
135
Jeff Layton7950e382012-10-10 16:43:13 -0400136 result = __getname();
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700137 if (unlikely(!result))
Eric Paris4043cde2012-01-03 14:23:08 -0500138 return ERR_PTR(-ENOMEM);
Paul Moore55422d02015-01-22 00:00:23 -0500139 result->refcnt = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140
Jeff Layton7950e382012-10-10 16:43:13 -0400141 /*
142 * First, try to embed the struct filename inside the names_cache
143 * allocation
144 */
145 kname = (char *)result + sizeof(*result);
Jeff Layton91a27b22012-10-10 15:25:28 -0400146 result->name = kname;
Jeff Layton7950e382012-10-10 16:43:13 -0400147 result->separate = false;
148 max = EMBEDDED_NAME_MAX;
149
150recopy:
151 len = strncpy_from_user(kname, filename, max);
Jeff Layton91a27b22012-10-10 15:25:28 -0400152 if (unlikely(len < 0)) {
153 err = ERR_PTR(len);
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700154 goto error;
Jeff Layton91a27b22012-10-10 15:25:28 -0400155 }
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700156
Jeff Layton7950e382012-10-10 16:43:13 -0400157 /*
158 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
159 * separate struct filename so we can dedicate the entire
160 * names_cache allocation for the pathname, and re-do the copy from
161 * userland.
162 */
163 if (len == EMBEDDED_NAME_MAX && max == EMBEDDED_NAME_MAX) {
164 kname = (char *)result;
165
166 result = kzalloc(sizeof(*result), GFP_KERNEL);
167 if (!result) {
168 err = ERR_PTR(-ENOMEM);
169 result = (struct filename *)kname;
170 goto error;
171 }
172 result->name = kname;
173 result->separate = true;
Paul Moore55422d02015-01-22 00:00:23 -0500174 result->refcnt = 1;
Jeff Layton7950e382012-10-10 16:43:13 -0400175 max = PATH_MAX;
176 goto recopy;
177 }
178
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700179 /* The empty path is special. */
180 if (unlikely(!len)) {
181 if (empty)
Eric Paris4043cde2012-01-03 14:23:08 -0500182 *empty = 1;
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700183 err = ERR_PTR(-ENOENT);
184 if (!(flags & LOOKUP_EMPTY))
185 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 }
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700187
188 err = ERR_PTR(-ENAMETOOLONG);
Jeff Layton7950e382012-10-10 16:43:13 -0400189 if (unlikely(len >= PATH_MAX))
190 goto error;
191
192 result->uptr = filename;
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800193 result->aname = NULL;
Jeff Layton7950e382012-10-10 16:43:13 -0400194 audit_getname(result);
195 return result;
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700196
197error:
Paul Moore55422d02015-01-22 00:00:23 -0500198 putname(result);
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700199 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200}
201
Jeff Layton91a27b22012-10-10 15:25:28 -0400202struct filename *
203getname(const char __user * filename)
Al Virof52e0c12011-03-14 18:56:51 -0400204{
Linus Torvaldsf7493e52012-03-22 16:10:40 -0700205 return getname_flags(filename, 0, NULL);
Al Virof52e0c12011-03-14 18:56:51 -0400206}
207
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800208struct filename *
209getname_kernel(const char * filename)
210{
211 struct filename *result;
Paul Moore08518542015-01-21 23:59:56 -0500212 int len = strlen(filename) + 1;
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800213
214 result = __getname();
215 if (unlikely(!result))
216 return ERR_PTR(-ENOMEM);
217
Paul Moore08518542015-01-21 23:59:56 -0500218 if (len <= EMBEDDED_NAME_MAX) {
219 result->name = (char *)(result) + sizeof(*result);
220 result->separate = false;
221 } else if (len <= PATH_MAX) {
222 struct filename *tmp;
223
224 tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
225 if (unlikely(!tmp)) {
226 __putname(result);
227 return ERR_PTR(-ENOMEM);
228 }
229 tmp->name = (char *)result;
230 tmp->separate = true;
231 result = tmp;
232 } else {
233 __putname(result);
234 return ERR_PTR(-ENAMETOOLONG);
235 }
236 memcpy((char *)result->name, filename, len);
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800237 result->uptr = NULL;
238 result->aname = NULL;
Paul Moore55422d02015-01-22 00:00:23 -0500239 result->refcnt = 1;
Paul Moorefd3522f2015-01-22 00:00:10 -0500240 audit_getname(result);
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800241
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800242 return result;
243}
244
Jeff Layton91a27b22012-10-10 15:25:28 -0400245void putname(struct filename *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246{
Paul Moore55422d02015-01-22 00:00:23 -0500247 BUG_ON(name->refcnt <= 0);
248
249 if (--name->refcnt > 0)
250 return;
251
252 if (name->separate) {
253 __putname(name->name);
254 kfree(name);
255 } else
256 __putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258
Linus Torvaldse77819e2011-07-22 19:30:19 -0700259static int check_acl(struct inode *inode, int mask)
260{
Linus Torvalds84635d62011-07-25 22:47:03 -0700261#ifdef CONFIG_FS_POSIX_ACL
Linus Torvaldse77819e2011-07-22 19:30:19 -0700262 struct posix_acl *acl;
263
Linus Torvaldse77819e2011-07-22 19:30:19 -0700264 if (mask & MAY_NOT_BLOCK) {
Al Viro35678662011-08-02 21:32:13 -0400265 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
266 if (!acl)
Linus Torvaldse77819e2011-07-22 19:30:19 -0700267 return -EAGAIN;
Al Viro35678662011-08-02 21:32:13 -0400268 /* no ->get_acl() calls in RCU mode... */
269 if (acl == ACL_NOT_CACHED)
270 return -ECHILD;
Ari Savolainen206b1d02011-08-06 19:43:07 +0300271 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
Linus Torvaldse77819e2011-07-22 19:30:19 -0700272 }
273
Christoph Hellwig2982baa2013-12-20 05:16:38 -0800274 acl = get_acl(inode, ACL_TYPE_ACCESS);
275 if (IS_ERR(acl))
276 return PTR_ERR(acl);
Linus Torvaldse77819e2011-07-22 19:30:19 -0700277 if (acl) {
278 int error = posix_acl_permission(inode, acl, mask);
279 posix_acl_release(acl);
280 return error;
281 }
Linus Torvalds84635d62011-07-25 22:47:03 -0700282#endif
Linus Torvaldse77819e2011-07-22 19:30:19 -0700283
284 return -EAGAIN;
285}
286
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700287/*
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530288 * This does the basic permission checking
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700289 */
Al Viro7e401452011-06-20 19:12:17 -0400290static int acl_permission_check(struct inode *inode, int mask)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700291{
Linus Torvalds26cf46b2011-05-13 11:51:01 -0700292 unsigned int mode = inode->i_mode;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700293
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -0800294 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700295 mode >>= 6;
296 else {
Linus Torvaldse77819e2011-07-22 19:30:19 -0700297 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
Al Viro7e401452011-06-20 19:12:17 -0400298 int error = check_acl(inode, mask);
Nick Pigginb74c79e2011-01-07 17:49:58 +1100299 if (error != -EAGAIN)
300 return error;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700301 }
302
303 if (in_group_p(inode->i_gid))
304 mode >>= 3;
305 }
306
307 /*
308 * If the DACs are ok we don't need any capability check.
309 */
Al Viro9c2c7032011-06-20 19:06:22 -0400310 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700311 return 0;
312 return -EACCES;
313}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314
315/**
Nick Pigginb74c79e2011-01-07 17:49:58 +1100316 * generic_permission - check for access rights on a Posix-like filesystem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 * @inode: inode to check access rights for
Andreas Gruenbacher8fd90c82011-10-23 23:13:30 +0530318 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 *
320 * Used to check for read/write/execute permissions on a file.
321 * We use "fsuid" for this, letting us set arbitrary permissions
322 * for filesystem access without changing the "normal" uids which
Nick Pigginb74c79e2011-01-07 17:49:58 +1100323 * are used for other things.
324 *
325 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
326 * request cannot be satisfied (eg. requires blocking or too much complexity).
327 * It would then be called again in ref-walk mode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 */
Al Viro2830ba72011-06-20 19:16:29 -0400329int generic_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700331 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332
333 /*
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530334 * Do the basic permission checks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 */
Al Viro7e401452011-06-20 19:12:17 -0400336 ret = acl_permission_check(inode, mask);
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700337 if (ret != -EACCES)
338 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339
Al Virod594e7e2011-06-20 19:55:42 -0400340 if (S_ISDIR(inode->i_mode)) {
341 /* DACs are overridable for directories */
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700342 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
Al Virod594e7e2011-06-20 19:55:42 -0400343 return 0;
344 if (!(mask & MAY_WRITE))
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700345 if (capable_wrt_inode_uidgid(inode,
346 CAP_DAC_READ_SEARCH))
Al Virod594e7e2011-06-20 19:55:42 -0400347 return 0;
348 return -EACCES;
349 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 /*
351 * Read/write DACs are always overridable.
Al Virod594e7e2011-06-20 19:55:42 -0400352 * Executable DACs are overridable when there is
353 * at least one exec bit set.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 */
Al Virod594e7e2011-06-20 19:55:42 -0400355 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700356 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 return 0;
358
359 /*
360 * Searching includes executable on directories, else just read.
361 */
Serge E. Hallyn7ea66002009-12-29 14:50:19 -0600362 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
Al Virod594e7e2011-06-20 19:55:42 -0400363 if (mask == MAY_READ)
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700364 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 return 0;
366
367 return -EACCES;
368}
Al Viro4d359502014-03-14 12:20:17 -0400369EXPORT_SYMBOL(generic_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700371/*
372 * We _really_ want to just do "generic_permission()" without
373 * even looking at the inode->i_op values. So we keep a cache
374 * flag in inode->i_opflags, that says "this has not special
375 * permission function, use the fast case".
376 */
377static inline int do_inode_permission(struct inode *inode, int mask)
378{
379 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
380 if (likely(inode->i_op->permission))
381 return inode->i_op->permission(inode, mask);
382
383 /* This gets set once for the inode lifetime */
384 spin_lock(&inode->i_lock);
385 inode->i_opflags |= IOP_FASTPERM;
386 spin_unlock(&inode->i_lock);
387 }
388 return generic_permission(inode, mask);
389}
390
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200391/**
David Howells0bdaea92012-06-25 12:55:46 +0100392 * __inode_permission - Check for access rights to a given inode
393 * @inode: Inode to check permission on
394 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200395 *
David Howells0bdaea92012-06-25 12:55:46 +0100396 * Check for read/write/execute permissions on an inode.
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530397 *
398 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
David Howells0bdaea92012-06-25 12:55:46 +0100399 *
400 * This does not check for a read-only file system. You probably want
401 * inode_permission().
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200402 */
David Howells0bdaea92012-06-25 12:55:46 +0100403int __inode_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404{
Al Viroe6305c42008-07-15 21:03:57 -0400405 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700407 if (unlikely(mask & MAY_WRITE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 /*
409 * Nobody gets write access to an immutable file.
410 */
411 if (IS_IMMUTABLE(inode))
412 return -EACCES;
413 }
414
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700415 retval = do_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 if (retval)
417 return retval;
418
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -0700419 retval = devcgroup_inode_permission(inode, mask);
420 if (retval)
421 return retval;
422
Eric Parisd09ca732010-07-23 11:43:57 -0400423 return security_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424}
Miklos Szeredibd5d0852014-10-24 00:14:35 +0200425EXPORT_SYMBOL(__inode_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426
Al Virof4d6ff82011-06-19 13:14:21 -0400427/**
David Howells0bdaea92012-06-25 12:55:46 +0100428 * sb_permission - Check superblock-level permissions
429 * @sb: Superblock of inode to check permission on
Randy Dunlap55852632012-08-18 17:39:25 -0700430 * @inode: Inode to check permission on
David Howells0bdaea92012-06-25 12:55:46 +0100431 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
432 *
433 * Separate out file-system wide checks from inode-specific permission checks.
434 */
435static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
436{
437 if (unlikely(mask & MAY_WRITE)) {
438 umode_t mode = inode->i_mode;
439
440 /* Nobody gets write access to a read-only fs. */
441 if ((sb->s_flags & MS_RDONLY) &&
442 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
443 return -EROFS;
444 }
445 return 0;
446}
447
448/**
449 * inode_permission - Check for access rights to a given inode
450 * @inode: Inode to check permission on
451 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
452 *
453 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
454 * this, letting us set arbitrary permissions for filesystem access without
455 * changing the "normal" UIDs which are used for other things.
456 *
457 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
458 */
459int inode_permission(struct inode *inode, int mask)
460{
461 int retval;
462
463 retval = sb_permission(inode->i_sb, inode, mask);
464 if (retval)
465 return retval;
466 return __inode_permission(inode, mask);
467}
Al Viro4d359502014-03-14 12:20:17 -0400468EXPORT_SYMBOL(inode_permission);
David Howells0bdaea92012-06-25 12:55:46 +0100469
470/**
Jan Blunck5dd784d02008-02-14 19:34:38 -0800471 * path_get - get a reference to a path
472 * @path: path to get the reference to
473 *
474 * Given a path increment the reference count to the dentry and the vfsmount.
475 */
Al Virodcf787f2013-03-01 23:51:07 -0500476void path_get(const struct path *path)
Jan Blunck5dd784d02008-02-14 19:34:38 -0800477{
478 mntget(path->mnt);
479 dget(path->dentry);
480}
481EXPORT_SYMBOL(path_get);
482
483/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800484 * path_put - put a reference to a path
485 * @path: path to put the reference to
486 *
487 * Given a path decrement the reference count to the dentry and the vfsmount.
488 */
Al Virodcf787f2013-03-01 23:51:07 -0500489void path_put(const struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490{
Jan Blunck1d957f92008-02-14 19:34:35 -0800491 dput(path->dentry);
492 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493}
Jan Blunck1d957f92008-02-14 19:34:35 -0800494EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495
Al Viro1f55a6e2014-11-01 19:30:41 -0400496struct nameidata {
497 struct path path;
498 struct qstr last;
499 struct path root;
500 struct inode *inode; /* path.dentry.d_inode */
501 unsigned int flags;
502 unsigned seq, m_seq;
503 int last_type;
504 unsigned depth;
Al Viro5e530842014-11-20 14:14:42 -0500505 struct file *base;
Al Viro1f55a6e2014-11-01 19:30:41 -0400506 char *saved_names[MAX_NESTED_LINKS + 1];
507};
508
Al Viro19660af2011-03-25 10:32:48 -0400509/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100510 * Path walking has 2 modes, rcu-walk and ref-walk (see
Al Viro19660af2011-03-25 10:32:48 -0400511 * Documentation/filesystems/path-lookup.txt). In situations when we can't
512 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
513 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
514 * mode. Refcounts are grabbed at the last known good point before rcu-walk
515 * got stuck, so ref-walk may continue from there. If this is not successful
516 * (eg. a seqcount has changed), then failure is returned and it's up to caller
517 * to restart the path walk from the beginning in ref-walk mode.
Nick Piggin31e6b012011-01-07 17:49:52 +1100518 */
Nick Piggin31e6b012011-01-07 17:49:52 +1100519
520/**
Al Viro19660af2011-03-25 10:32:48 -0400521 * unlazy_walk - try to switch to ref-walk mode.
522 * @nd: nameidata pathwalk data
523 * @dentry: child of nd->path.dentry or NULL
Randy Dunlap39191622011-01-08 19:36:21 -0800524 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100525 *
Al Viro19660af2011-03-25 10:32:48 -0400526 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
527 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
528 * @nd or NULL. Must be called from rcu-walk context.
Nick Piggin31e6b012011-01-07 17:49:52 +1100529 */
Al Viro19660af2011-03-25 10:32:48 -0400530static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
Nick Piggin31e6b012011-01-07 17:49:52 +1100531{
532 struct fs_struct *fs = current->fs;
533 struct dentry *parent = nd->path.dentry;
534
535 BUG_ON(!(nd->flags & LOOKUP_RCU));
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700536
537 /*
Al Viro48a066e2013-09-29 22:06:07 -0400538 * After legitimizing the bastards, terminate_walk()
539 * will do the right thing for non-RCU mode, and all our
540 * subsequent exit cases should rcu_read_unlock()
541 * before returning. Do vfsmount first; if dentry
542 * can't be legitimized, just set nd->path.dentry to NULL
543 * and rely on dput(NULL) being a no-op.
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700544 */
Al Viro48a066e2013-09-29 22:06:07 -0400545 if (!legitimize_mnt(nd->path.mnt, nd->m_seq))
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700546 return -ECHILD;
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700547 nd->flags &= ~LOOKUP_RCU;
Linus Torvalds15570082013-09-02 11:38:06 -0700548
Al Viro48a066e2013-09-29 22:06:07 -0400549 if (!lockref_get_not_dead(&parent->d_lockref)) {
550 nd->path.dentry = NULL;
Al Virod870b4a2013-11-29 01:48:32 -0500551 goto out;
Al Viro48a066e2013-09-29 22:06:07 -0400552 }
553
Linus Torvalds15570082013-09-02 11:38:06 -0700554 /*
555 * For a negative lookup, the lookup sequence point is the parents
556 * sequence point, and it only needs to revalidate the parent dentry.
557 *
558 * For a positive lookup, we need to move both the parent and the
559 * dentry from the RCU domain to be properly refcounted. And the
560 * sequence number in the dentry validates *both* dentry counters,
561 * since we checked the sequence number of the parent after we got
562 * the child sequence number. So we know the parent must still
563 * be valid if the child sequence number is still valid.
564 */
Al Viro19660af2011-03-25 10:32:48 -0400565 if (!dentry) {
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700566 if (read_seqcount_retry(&parent->d_seq, nd->seq))
567 goto out;
Al Viro19660af2011-03-25 10:32:48 -0400568 BUG_ON(nd->inode != parent->d_inode);
569 } else {
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700570 if (!lockref_get_not_dead(&dentry->d_lockref))
571 goto out;
572 if (read_seqcount_retry(&dentry->d_seq, nd->seq))
573 goto drop_dentry;
Al Viro19660af2011-03-25 10:32:48 -0400574 }
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700575
576 /*
577 * Sequence counts matched. Now make sure that the root is
578 * still valid and get it if required.
579 */
580 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
581 spin_lock(&fs->lock);
582 if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry)
583 goto unlock_and_drop_dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +1100584 path_get(&nd->root);
585 spin_unlock(&fs->lock);
586 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100587
Al Viro8b61e742013-11-08 12:45:01 -0500588 rcu_read_unlock();
Nick Piggin31e6b012011-01-07 17:49:52 +1100589 return 0;
Al Viro19660af2011-03-25 10:32:48 -0400590
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700591unlock_and_drop_dentry:
592 spin_unlock(&fs->lock);
593drop_dentry:
Al Viro8b61e742013-11-08 12:45:01 -0500594 rcu_read_unlock();
Linus Torvalds15570082013-09-02 11:38:06 -0700595 dput(dentry);
Linus Torvaldsd0d27272013-09-10 12:17:49 -0700596 goto drop_root_mnt;
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700597out:
Al Viro8b61e742013-11-08 12:45:01 -0500598 rcu_read_unlock();
Linus Torvaldsd0d27272013-09-10 12:17:49 -0700599drop_root_mnt:
600 if (!(nd->flags & LOOKUP_ROOT))
601 nd->root.mnt = NULL;
Nick Piggin31e6b012011-01-07 17:49:52 +1100602 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
Al Viro48a066e2013-09-29 22:06:07 -0400630 if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) {
Al Viro8b61e742013-11-08 12:45:01 -0500631 rcu_read_unlock();
Al Viro48a066e2013-09-29 22:06:07 -0400632 return -ECHILD;
633 }
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700634 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) {
Al Viro8b61e742013-11-08 12:45:01 -0500635 rcu_read_unlock();
Al Viro48a066e2013-09-29 22:06:07 -0400636 mntput(nd->path.mnt);
Al Viro9f1fafe2011-03-25 11:00:12 -0400637 return -ECHILD;
638 }
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700639 if (read_seqcount_retry(&dentry->d_seq, nd->seq)) {
Al Viro8b61e742013-11-08 12:45:01 -0500640 rcu_read_unlock();
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700641 dput(dentry);
Al Viro48a066e2013-09-29 22:06:07 -0400642 mntput(nd->path.mnt);
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700643 return -ECHILD;
644 }
Al Viro8b61e742013-11-08 12:45:01 -0500645 rcu_read_unlock();
Al Viro9f1fafe2011-03-25 11:00:12 -0400646 }
647
Al Viro16c2cd72011-02-22 15:50:10 -0500648 if (likely(!(nd->flags & LOOKUP_JUMPED)))
Jeff Layton39159de2009-12-07 12:01:50 -0500649 return 0;
650
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500651 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
Al Viro16c2cd72011-02-22 15:50:10 -0500652 return 0;
653
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500654 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
Jeff Layton39159de2009-12-07 12:01:50 -0500655 if (status > 0)
656 return 0;
657
Al Viro16c2cd72011-02-22 15:50:10 -0500658 if (!status)
Jeff Layton39159de2009-12-07 12:01:50 -0500659 status = -ESTALE;
Al Viro16c2cd72011-02-22 15:50:10 -0500660
Al Viro9f1fafe2011-03-25 11:00:12 -0400661 path_put(&nd->path);
Jeff Layton39159de2009-12-07 12:01:50 -0500662 return status;
663}
664
Al Viro2a737872009-04-07 11:49:53 -0400665static __always_inline void set_root(struct nameidata *nd)
666{
Al Viro7bd88372014-09-13 21:55:46 -0400667 get_fs_root(current->fs, &nd->root);
Al Viro2a737872009-04-07 11:49:53 -0400668}
669
Al Viro6de88d72009-08-09 01:41:57 +0400670static int link_path_walk(const char *, struct nameidata *);
671
Al Viro7bd88372014-09-13 21:55:46 -0400672static __always_inline unsigned set_root_rcu(struct nameidata *nd)
Nick Piggin31e6b012011-01-07 17:49:52 +1100673{
Al Viro7bd88372014-09-13 21:55:46 -0400674 struct fs_struct *fs = current->fs;
675 unsigned seq, res;
Nick Pigginc28cc362011-01-07 17:49:53 +1100676
Al Viro7bd88372014-09-13 21:55:46 -0400677 do {
678 seq = read_seqcount_begin(&fs->seq);
679 nd->root = fs->root;
680 res = __read_seqcount_begin(&nd->root.dentry->d_seq);
681 } while (read_seqcount_retry(&fs->seq, seq));
682 return res;
Nick Piggin31e6b012011-01-07 17:49:52 +1100683}
684
Jan Blunck1d957f92008-02-14 19:34:35 -0800685static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700686{
687 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800688 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700689 mntput(path->mnt);
690}
691
Nick Piggin7b9337a2011-01-14 08:42:43 +0000692static inline void path_to_nameidata(const struct path *path,
693 struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700694{
Nick Piggin31e6b012011-01-07 17:49:52 +1100695 if (!(nd->flags & LOOKUP_RCU)) {
696 dput(nd->path.dentry);
697 if (nd->path.mnt != path->mnt)
698 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800699 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100700 nd->path.mnt = path->mnt;
Jan Blunck4ac91372008-02-14 19:34:32 -0800701 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700702}
703
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400704/*
705 * Helper to directly jump to a known parsed path from ->follow_link,
706 * caller must have taken a reference to path beforehand.
707 */
708void nd_jump_link(struct nameidata *nd, struct path *path)
709{
710 path_put(&nd->path);
711
712 nd->path = *path;
713 nd->inode = nd->path.dentry->d_inode;
714 nd->flags |= LOOKUP_JUMPED;
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400715}
716
Al Viro1f55a6e2014-11-01 19:30:41 -0400717void nd_set_link(struct nameidata *nd, char *path)
718{
719 nd->saved_names[nd->depth] = path;
720}
721EXPORT_SYMBOL(nd_set_link);
722
723char *nd_get_link(struct nameidata *nd)
724{
725 return nd->saved_names[nd->depth];
726}
727EXPORT_SYMBOL(nd_get_link);
728
Al Viro574197e2011-03-14 22:20:34 -0400729static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
730{
731 struct inode *inode = link->dentry->d_inode;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400732 if (inode->i_op->put_link)
Al Viro574197e2011-03-14 22:20:34 -0400733 inode->i_op->put_link(link->dentry, nd, cookie);
734 path_put(link);
735}
736
Linus Torvalds561ec642012-10-26 10:05:07 -0700737int sysctl_protected_symlinks __read_mostly = 0;
738int sysctl_protected_hardlinks __read_mostly = 0;
Kees Cook800179c2012-07-25 17:29:07 -0700739
740/**
741 * may_follow_link - Check symlink following for unsafe situations
742 * @link: The path of the symlink
Randy Dunlap55852632012-08-18 17:39:25 -0700743 * @nd: nameidata pathwalk data
Kees Cook800179c2012-07-25 17:29:07 -0700744 *
745 * In the case of the sysctl_protected_symlinks sysctl being enabled,
746 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
747 * in a sticky world-writable directory. This is to protect privileged
748 * processes from failing races against path names that may change out
749 * from under them by way of other users creating malicious symlinks.
750 * It will permit symlinks to be followed only when outside a sticky
751 * world-writable directory, or when the uid of the symlink and follower
752 * match, or when the directory owner matches the symlink's owner.
753 *
754 * Returns 0 if following the symlink is allowed, -ve on error.
755 */
756static inline int may_follow_link(struct path *link, struct nameidata *nd)
757{
758 const struct inode *inode;
759 const struct inode *parent;
760
761 if (!sysctl_protected_symlinks)
762 return 0;
763
764 /* Allowed if owner and follower match. */
765 inode = link->dentry->d_inode;
Eric W. Biederman81abe272012-08-03 09:38:08 -0700766 if (uid_eq(current_cred()->fsuid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700767 return 0;
768
769 /* Allowed if parent directory not sticky and world-writable. */
770 parent = nd->path.dentry->d_inode;
771 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
772 return 0;
773
774 /* Allowed if parent directory and link owner match. */
Eric W. Biederman81abe272012-08-03 09:38:08 -0700775 if (uid_eq(parent->i_uid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700776 return 0;
777
Sasha Levinffd8d102012-10-04 19:56:40 -0400778 audit_log_link_denied("follow_link", link);
Kees Cook800179c2012-07-25 17:29:07 -0700779 path_put_conditional(link, nd);
780 path_put(&nd->path);
781 return -EACCES;
782}
783
784/**
785 * safe_hardlink_source - Check for safe hardlink conditions
786 * @inode: the source inode to hardlink from
787 *
788 * Return false if at least one of the following conditions:
789 * - inode is not a regular file
790 * - inode is setuid
791 * - inode is setgid and group-exec
792 * - access failure for read and write
793 *
794 * Otherwise returns true.
795 */
796static bool safe_hardlink_source(struct inode *inode)
797{
798 umode_t mode = inode->i_mode;
799
800 /* Special files should not get pinned to the filesystem. */
801 if (!S_ISREG(mode))
802 return false;
803
804 /* Setuid files should not get pinned to the filesystem. */
805 if (mode & S_ISUID)
806 return false;
807
808 /* Executable setgid files should not get pinned to the filesystem. */
809 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
810 return false;
811
812 /* Hardlinking to unreadable or unwritable sources is dangerous. */
813 if (inode_permission(inode, MAY_READ | MAY_WRITE))
814 return false;
815
816 return true;
817}
818
819/**
820 * may_linkat - Check permissions for creating a hardlink
821 * @link: the source to hardlink from
822 *
823 * Block hardlink when all of:
824 * - sysctl_protected_hardlinks enabled
825 * - fsuid does not match inode
826 * - hardlink source is unsafe (see safe_hardlink_source() above)
827 * - not CAP_FOWNER
828 *
829 * Returns 0 if successful, -ve on error.
830 */
831static int may_linkat(struct path *link)
832{
833 const struct cred *cred;
834 struct inode *inode;
835
836 if (!sysctl_protected_hardlinks)
837 return 0;
838
839 cred = current_cred();
840 inode = link->dentry->d_inode;
841
842 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
843 * otherwise, it must be a safe source.
844 */
Eric W. Biederman81abe272012-08-03 09:38:08 -0700845 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
Kees Cook800179c2012-07-25 17:29:07 -0700846 capable(CAP_FOWNER))
847 return 0;
848
Kees Cooka51d9ea2012-07-25 17:29:08 -0700849 audit_log_link_denied("linkat", link);
Kees Cook800179c2012-07-25 17:29:07 -0700850 return -EPERM;
851}
852
Al Virodef4af32009-12-26 08:37:05 -0500853static __always_inline int
Al Viro574197e2011-03-14 22:20:34 -0400854follow_link(struct path *link, struct nameidata *nd, void **p)
Ian Kent051d3812006-03-27 01:14:53 -0800855{
Nick Piggin7b9337a2011-01-14 08:42:43 +0000856 struct dentry *dentry = link->dentry;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400857 int error;
858 char *s;
Ian Kent051d3812006-03-27 01:14:53 -0800859
Al Viro844a3912011-02-15 00:38:26 -0500860 BUG_ON(nd->flags & LOOKUP_RCU);
861
Al Viro0e794582011-03-16 02:45:02 -0400862 if (link->mnt == nd->path.mnt)
863 mntget(link->mnt);
864
Al Viro6d7b5aa2012-06-10 04:15:17 -0400865 error = -ELOOP;
866 if (unlikely(current->total_link_count >= 40))
867 goto out_put_nd_path;
868
Al Viro574197e2011-03-14 22:20:34 -0400869 cond_resched();
870 current->total_link_count++;
871
Al Viro68ac1232012-03-15 08:21:57 -0400872 touch_atime(link);
Ian Kent051d3812006-03-27 01:14:53 -0800873 nd_set_link(nd, NULL);
874
Al Viro36f3b4f2011-02-22 21:24:38 -0500875 error = security_inode_follow_link(link->dentry, nd);
Al Viro6d7b5aa2012-06-10 04:15:17 -0400876 if (error)
877 goto out_put_nd_path;
Al Viro36f3b4f2011-02-22 21:24:38 -0500878
Al Viro86acdca12009-12-22 23:45:11 -0500879 nd->last_type = LAST_BIND;
Al Virodef4af32009-12-26 08:37:05 -0500880 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
881 error = PTR_ERR(*p);
Al Viro6d7b5aa2012-06-10 04:15:17 -0400882 if (IS_ERR(*p))
Christoph Hellwig408ef012012-06-18 10:47:03 -0400883 goto out_put_nd_path;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400884
885 error = 0;
886 s = nd_get_link(nd);
887 if (s) {
Al Viro443ed252013-09-10 12:00:43 -0400888 if (unlikely(IS_ERR(s))) {
889 path_put(&nd->path);
890 put_link(nd, link, *p);
891 return PTR_ERR(s);
892 }
893 if (*s == '/') {
Al Viro7bd88372014-09-13 21:55:46 -0400894 if (!nd->root.mnt)
895 set_root(nd);
Al Viro443ed252013-09-10 12:00:43 -0400896 path_put(&nd->path);
897 nd->path = nd->root;
898 path_get(&nd->root);
899 nd->flags |= LOOKUP_JUMPED;
900 }
901 nd->inode = nd->path.dentry->d_inode;
902 error = link_path_walk(s, nd);
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400903 if (unlikely(error))
904 put_link(nd, link, *p);
Ian Kent051d3812006-03-27 01:14:53 -0800905 }
Al Viro6d7b5aa2012-06-10 04:15:17 -0400906
907 return error;
908
909out_put_nd_path:
Arnd Bergmann98f6ef62012-10-11 13:20:00 +0000910 *p = NULL;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400911 path_put(&nd->path);
Al Viro6d7b5aa2012-06-10 04:15:17 -0400912 path_put(link);
Ian Kent051d3812006-03-27 01:14:53 -0800913 return error;
914}
915
Nick Piggin31e6b012011-01-07 17:49:52 +1100916static int follow_up_rcu(struct path *path)
917{
Al Viro0714a532011-11-24 22:19:58 -0500918 struct mount *mnt = real_mount(path->mnt);
919 struct mount *parent;
Nick Piggin31e6b012011-01-07 17:49:52 +1100920 struct dentry *mountpoint;
921
Al Viro0714a532011-11-24 22:19:58 -0500922 parent = mnt->mnt_parent;
923 if (&parent->mnt == path->mnt)
Nick Piggin31e6b012011-01-07 17:49:52 +1100924 return 0;
Al Viroa73324d2011-11-24 22:25:07 -0500925 mountpoint = mnt->mnt_mountpoint;
Nick Piggin31e6b012011-01-07 17:49:52 +1100926 path->dentry = mountpoint;
Al Viro0714a532011-11-24 22:19:58 -0500927 path->mnt = &parent->mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +1100928 return 1;
929}
930
David Howellsf015f1262012-06-25 12:55:28 +0100931/*
932 * follow_up - Find the mountpoint of path's vfsmount
933 *
934 * Given a path, find the mountpoint of its source file system.
935 * Replace @path with the path of the mountpoint in the parent mount.
936 * Up is towards /.
937 *
938 * Return 1 if we went up a level and 0 if we were already at the
939 * root.
940 */
Al Virobab77eb2009-04-18 03:26:48 -0400941int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942{
Al Viro0714a532011-11-24 22:19:58 -0500943 struct mount *mnt = real_mount(path->mnt);
944 struct mount *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +1000946
Al Viro48a066e2013-09-29 22:06:07 -0400947 read_seqlock_excl(&mount_lock);
Al Viro0714a532011-11-24 22:19:58 -0500948 parent = mnt->mnt_parent;
Al Viro3c0a6162012-07-18 17:32:50 +0400949 if (parent == mnt) {
Al Viro48a066e2013-09-29 22:06:07 -0400950 read_sequnlock_excl(&mount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 return 0;
952 }
Al Viro0714a532011-11-24 22:19:58 -0500953 mntget(&parent->mnt);
Al Viroa73324d2011-11-24 22:25:07 -0500954 mountpoint = dget(mnt->mnt_mountpoint);
Al Viro48a066e2013-09-29 22:06:07 -0400955 read_sequnlock_excl(&mount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400956 dput(path->dentry);
957 path->dentry = mountpoint;
958 mntput(path->mnt);
Al Viro0714a532011-11-24 22:19:58 -0500959 path->mnt = &parent->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 return 1;
961}
Al Viro4d359502014-03-14 12:20:17 -0400962EXPORT_SYMBOL(follow_up);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100964/*
David Howells9875cf82011-01-14 18:45:21 +0000965 * Perform an automount
966 * - return -EISDIR to tell follow_managed() to stop and return the path we
967 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 */
David Howells9875cf82011-01-14 18:45:21 +0000969static int follow_automount(struct path *path, unsigned flags,
970 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +1100971{
David Howells9875cf82011-01-14 18:45:21 +0000972 struct vfsmount *mnt;
David Howellsea5b7782011-01-14 19:10:03 +0000973 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +1100974
David Howells9875cf82011-01-14 18:45:21 +0000975 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
976 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -0700977
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200978 /* We don't want to mount if someone's just doing a stat -
979 * unless they're stat'ing a directory and appended a '/' to
980 * the name.
981 *
982 * We do, however, want to mount if someone wants to open or
983 * create a file of any type under the mountpoint, wants to
984 * traverse through the mountpoint or wants to open the
985 * mounted directory. Also, autofs may mark negative dentries
986 * as being automount points. These will need the attentions
987 * of the daemon to instantiate them before they can be used.
David Howells9875cf82011-01-14 18:45:21 +0000988 */
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200989 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
Linus Torvaldsd94c1772011-09-26 17:44:55 -0700990 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200991 path->dentry->d_inode)
992 return -EISDIR;
993
David Howells9875cf82011-01-14 18:45:21 +0000994 current->total_link_count++;
995 if (current->total_link_count >= 40)
996 return -ELOOP;
997
998 mnt = path->dentry->d_op->d_automount(path);
999 if (IS_ERR(mnt)) {
1000 /*
1001 * The filesystem is allowed to return -EISDIR here to indicate
1002 * it doesn't want to automount. For instance, autofs would do
1003 * this so that its userspace daemon can mount on this dentry.
1004 *
1005 * However, we can only permit this if it's a terminal point in
1006 * the path being looked up; if it wasn't then the remainder of
1007 * the path is inaccessible and we should say so.
1008 */
Al Viro49084c32011-06-25 21:59:52 -04001009 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
David Howells9875cf82011-01-14 18:45:21 +00001010 return -EREMOTE;
1011 return PTR_ERR(mnt);
1012 }
David Howellsea5b7782011-01-14 19:10:03 +00001013
David Howells9875cf82011-01-14 18:45:21 +00001014 if (!mnt) /* mount collision */
1015 return 0;
1016
Al Viro8aef1882011-06-16 15:10:06 +01001017 if (!*need_mntput) {
1018 /* lock_mount() may release path->mnt on error */
1019 mntget(path->mnt);
1020 *need_mntput = true;
1021 }
Al Viro19a167a2011-01-17 01:35:23 -05001022 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +00001023
David Howellsea5b7782011-01-14 19:10:03 +00001024 switch (err) {
1025 case -EBUSY:
1026 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -05001027 return 0;
David Howellsea5b7782011-01-14 19:10:03 +00001028 case 0:
Al Viro8aef1882011-06-16 15:10:06 +01001029 path_put(path);
David Howellsea5b7782011-01-14 19:10:03 +00001030 path->mnt = mnt;
1031 path->dentry = dget(mnt->mnt_root);
David Howellsea5b7782011-01-14 19:10:03 +00001032 return 0;
Al Viro19a167a2011-01-17 01:35:23 -05001033 default:
1034 return err;
David Howellsea5b7782011-01-14 19:10:03 +00001035 }
Al Viro19a167a2011-01-17 01:35:23 -05001036
David Howells9875cf82011-01-14 18:45:21 +00001037}
1038
1039/*
1040 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +00001041 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +00001042 * - Flagged as mountpoint
1043 * - Flagged as automount point
1044 *
1045 * This may only be called in refwalk mode.
1046 *
1047 * Serialization is taken care of in namespace.c
1048 */
1049static int follow_managed(struct path *path, unsigned flags)
1050{
Al Viro8aef1882011-06-16 15:10:06 +01001051 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
David Howells9875cf82011-01-14 18:45:21 +00001052 unsigned managed;
1053 bool need_mntput = false;
Al Viro8aef1882011-06-16 15:10:06 +01001054 int ret = 0;
David Howells9875cf82011-01-14 18:45:21 +00001055
1056 /* Given that we're not holding a lock here, we retain the value in a
1057 * local variable for each dentry as we look at it so that we don't see
1058 * the components of that value change under us */
1059 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1060 managed &= DCACHE_MANAGED_DENTRY,
1061 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +00001062 /* Allow the filesystem to manage the transit without i_mutex
1063 * being held. */
1064 if (managed & DCACHE_MANAGE_TRANSIT) {
1065 BUG_ON(!path->dentry->d_op);
1066 BUG_ON(!path->dentry->d_op->d_manage);
Al Viro1aed3e42011-03-18 09:09:02 -04001067 ret = path->dentry->d_op->d_manage(path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001068 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001069 break;
David Howellscc53ce52011-01-14 18:45:26 +00001070 }
1071
David Howells9875cf82011-01-14 18:45:21 +00001072 /* Transit to a mounted filesystem. */
1073 if (managed & DCACHE_MOUNTED) {
1074 struct vfsmount *mounted = lookup_mnt(path);
1075 if (mounted) {
1076 dput(path->dentry);
1077 if (need_mntput)
1078 mntput(path->mnt);
1079 path->mnt = mounted;
1080 path->dentry = dget(mounted->mnt_root);
1081 need_mntput = true;
1082 continue;
1083 }
1084
1085 /* Something is mounted on this dentry in another
1086 * namespace and/or whatever was mounted there in this
Al Viro48a066e2013-09-29 22:06:07 -04001087 * namespace got unmounted before lookup_mnt() could
1088 * get it */
David Howells9875cf82011-01-14 18:45:21 +00001089 }
1090
1091 /* Handle an automount point */
1092 if (managed & DCACHE_NEED_AUTOMOUNT) {
1093 ret = follow_automount(path, flags, &need_mntput);
1094 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001095 break;
David Howells9875cf82011-01-14 18:45:21 +00001096 continue;
1097 }
1098
1099 /* We didn't change the current path point */
1100 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 }
Al Viro8aef1882011-06-16 15:10:06 +01001102
1103 if (need_mntput && path->mnt == mnt)
1104 mntput(path->mnt);
1105 if (ret == -EISDIR)
1106 ret = 0;
Al Viroa3fbbde2011-11-07 21:21:26 +00001107 return ret < 0 ? ret : need_mntput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108}
1109
David Howellscc53ce52011-01-14 18:45:26 +00001110int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111{
1112 struct vfsmount *mounted;
1113
Al Viro1c755af2009-04-18 14:06:57 -04001114 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -04001116 dput(path->dentry);
1117 mntput(path->mnt);
1118 path->mnt = mounted;
1119 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 return 1;
1121 }
1122 return 0;
1123}
Al Viro4d359502014-03-14 12:20:17 -04001124EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125
NeilBrownb8faf032014-08-04 17:06:29 +10001126static inline int managed_dentry_rcu(struct dentry *dentry)
Ian Kent62a73752011-03-25 01:51:02 +08001127{
NeilBrownb8faf032014-08-04 17:06:29 +10001128 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1129 dentry->d_op->d_manage(dentry, true) : 0;
Ian Kent62a73752011-03-25 01:51:02 +08001130}
1131
David Howells9875cf82011-01-14 18:45:21 +00001132/*
Al Viro287548e2011-05-27 06:50:06 -04001133 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1134 * we meet a managed dentry that would need blocking.
David Howells9875cf82011-01-14 18:45:21 +00001135 */
1136static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
Al Viro287548e2011-05-27 06:50:06 -04001137 struct inode **inode)
David Howells9875cf82011-01-14 18:45:21 +00001138{
Ian Kent62a73752011-03-25 01:51:02 +08001139 for (;;) {
Al Viroc7105362011-11-24 18:22:03 -05001140 struct mount *mounted;
Ian Kent62a73752011-03-25 01:51:02 +08001141 /*
1142 * Don't forget we might have a non-mountpoint managed dentry
1143 * that wants to block transit.
1144 */
NeilBrownb8faf032014-08-04 17:06:29 +10001145 switch (managed_dentry_rcu(path->dentry)) {
1146 case -ECHILD:
1147 default:
David Howellsab909112011-01-14 18:46:51 +00001148 return false;
NeilBrownb8faf032014-08-04 17:06:29 +10001149 case -EISDIR:
1150 return true;
1151 case 0:
1152 break;
1153 }
Ian Kent62a73752011-03-25 01:51:02 +08001154
1155 if (!d_mountpoint(path->dentry))
NeilBrownb8faf032014-08-04 17:06:29 +10001156 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
Ian Kent62a73752011-03-25 01:51:02 +08001157
Al Viro474279d2013-10-01 16:11:26 -04001158 mounted = __lookup_mnt(path->mnt, path->dentry);
David Howells9875cf82011-01-14 18:45:21 +00001159 if (!mounted)
1160 break;
Al Viroc7105362011-11-24 18:22:03 -05001161 path->mnt = &mounted->mnt;
1162 path->dentry = mounted->mnt.mnt_root;
Al Viroa3fbbde2011-11-07 21:21:26 +00001163 nd->flags |= LOOKUP_JUMPED;
David Howells9875cf82011-01-14 18:45:21 +00001164 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
Linus Torvalds59430262011-07-18 15:43:29 -07001165 /*
1166 * Update the inode too. We don't need to re-check the
1167 * dentry sequence number here after this d_inode read,
1168 * because a mount-point is always pinned.
1169 */
1170 *inode = path->dentry->d_inode;
David Howells9875cf82011-01-14 18:45:21 +00001171 }
Al Virof5be3e29122014-09-13 21:50:45 -04001172 return !read_seqretry(&mount_lock, nd->m_seq) &&
NeilBrownb8faf032014-08-04 17:06:29 +10001173 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
Al Viro287548e2011-05-27 06:50:06 -04001174}
1175
Nick Piggin31e6b012011-01-07 17:49:52 +11001176static int follow_dotdot_rcu(struct nameidata *nd)
1177{
Al Viro4023bfc2014-09-13 21:59:43 -04001178 struct inode *inode = nd->inode;
Al Viro7bd88372014-09-13 21:55:46 -04001179 if (!nd->root.mnt)
1180 set_root_rcu(nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11001181
David Howells9875cf82011-01-14 18:45:21 +00001182 while (1) {
Nick Piggin31e6b012011-01-07 17:49:52 +11001183 if (nd->path.dentry == nd->root.dentry &&
1184 nd->path.mnt == nd->root.mnt) {
1185 break;
1186 }
1187 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1188 struct dentry *old = nd->path.dentry;
1189 struct dentry *parent = old->d_parent;
1190 unsigned seq;
1191
Al Viro4023bfc2014-09-13 21:59:43 -04001192 inode = parent->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001193 seq = read_seqcount_begin(&parent->d_seq);
1194 if (read_seqcount_retry(&old->d_seq, nd->seq))
Al Viroef7562d2011-03-04 14:35:59 -05001195 goto failed;
Nick Piggin31e6b012011-01-07 17:49:52 +11001196 nd->path.dentry = parent;
1197 nd->seq = seq;
1198 break;
1199 }
1200 if (!follow_up_rcu(&nd->path))
1201 break;
Al Viro4023bfc2014-09-13 21:59:43 -04001202 inode = nd->path.dentry->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001203 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11001204 }
Al Virob37199e2014-03-20 15:18:22 -04001205 while (d_mountpoint(nd->path.dentry)) {
1206 struct mount *mounted;
1207 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1208 if (!mounted)
1209 break;
1210 nd->path.mnt = &mounted->mnt;
1211 nd->path.dentry = mounted->mnt.mnt_root;
Al Viro4023bfc2014-09-13 21:59:43 -04001212 inode = nd->path.dentry->d_inode;
Al Virob37199e2014-03-20 15:18:22 -04001213 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Virof5be3e29122014-09-13 21:50:45 -04001214 if (read_seqretry(&mount_lock, nd->m_seq))
Al Virob37199e2014-03-20 15:18:22 -04001215 goto failed;
1216 }
Al Viro4023bfc2014-09-13 21:59:43 -04001217 nd->inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001218 return 0;
Al Viroef7562d2011-03-04 14:35:59 -05001219
1220failed:
1221 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001222 if (!(nd->flags & LOOKUP_ROOT))
1223 nd->root.mnt = NULL;
Al Viro8b61e742013-11-08 12:45:01 -05001224 rcu_read_unlock();
Al Viroef7562d2011-03-04 14:35:59 -05001225 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001226}
1227
David Howells9875cf82011-01-14 18:45:21 +00001228/*
David Howellscc53ce52011-01-14 18:45:26 +00001229 * Follow down to the covering mount currently visible to userspace. At each
1230 * point, the filesystem owning that dentry may be queried as to whether the
1231 * caller is permitted to proceed or not.
David Howellscc53ce52011-01-14 18:45:26 +00001232 */
Al Viro7cc90cc2011-03-18 09:04:20 -04001233int follow_down(struct path *path)
David Howellscc53ce52011-01-14 18:45:26 +00001234{
1235 unsigned managed;
1236 int ret;
1237
1238 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1239 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1240 /* Allow the filesystem to manage the transit without i_mutex
1241 * being held.
1242 *
1243 * We indicate to the filesystem if someone is trying to mount
1244 * something here. This gives autofs the chance to deny anyone
1245 * other than its daemon the right to mount on its
1246 * superstructure.
1247 *
1248 * The filesystem may sleep at this point.
1249 */
1250 if (managed & DCACHE_MANAGE_TRANSIT) {
1251 BUG_ON(!path->dentry->d_op);
1252 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +00001253 ret = path->dentry->d_op->d_manage(
Al Viro1aed3e42011-03-18 09:09:02 -04001254 path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001255 if (ret < 0)
1256 return ret == -EISDIR ? 0 : ret;
1257 }
1258
1259 /* Transit to a mounted filesystem. */
1260 if (managed & DCACHE_MOUNTED) {
1261 struct vfsmount *mounted = lookup_mnt(path);
1262 if (!mounted)
1263 break;
1264 dput(path->dentry);
1265 mntput(path->mnt);
1266 path->mnt = mounted;
1267 path->dentry = dget(mounted->mnt_root);
1268 continue;
1269 }
1270
1271 /* Don't handle automount points here */
1272 break;
1273 }
1274 return 0;
1275}
Al Viro4d359502014-03-14 12:20:17 -04001276EXPORT_SYMBOL(follow_down);
David Howellscc53ce52011-01-14 18:45:26 +00001277
1278/*
David Howells9875cf82011-01-14 18:45:21 +00001279 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1280 */
1281static void follow_mount(struct path *path)
1282{
1283 while (d_mountpoint(path->dentry)) {
1284 struct vfsmount *mounted = lookup_mnt(path);
1285 if (!mounted)
1286 break;
1287 dput(path->dentry);
1288 mntput(path->mnt);
1289 path->mnt = mounted;
1290 path->dentry = dget(mounted->mnt_root);
1291 }
1292}
1293
Nick Piggin31e6b012011-01-07 17:49:52 +11001294static void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295{
Al Viro7bd88372014-09-13 21:55:46 -04001296 if (!nd->root.mnt)
1297 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001298
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 while(1) {
Jan Blunck4ac91372008-02-14 19:34:32 -08001300 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301
Al Viro2a737872009-04-07 11:49:53 -04001302 if (nd->path.dentry == nd->root.dentry &&
1303 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 break;
1305 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001306 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Al Viro3088dd72010-01-30 15:47:29 -05001307 /* rare case of legitimate dget_parent()... */
1308 nd->path.dentry = dget_parent(nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 dput(old);
1310 break;
1311 }
Al Viro3088dd72010-01-30 15:47:29 -05001312 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 }
Al Viro79ed0222009-04-18 13:59:41 -04001315 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001316 nd->inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317}
1318
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319/*
Miklos Szeredibad61182012-03-26 12:54:24 +02001320 * This looks up the name in dcache, possibly revalidates the old dentry and
1321 * allocates a new one if not found or not valid. In the need_lookup argument
1322 * returns whether i_op->lookup is necessary.
1323 *
1324 * dir->d_inode->i_mutex must be held
Nick Pigginbaa03892010-08-18 04:37:31 +10001325 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001326static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
Al Viro201f9562012-06-22 12:42:10 +04001327 unsigned int flags, bool *need_lookup)
Nick Pigginbaa03892010-08-18 04:37:31 +10001328{
Nick Pigginbaa03892010-08-18 04:37:31 +10001329 struct dentry *dentry;
Miklos Szeredibad61182012-03-26 12:54:24 +02001330 int error;
Nick Pigginbaa03892010-08-18 04:37:31 +10001331
Miklos Szeredibad61182012-03-26 12:54:24 +02001332 *need_lookup = false;
1333 dentry = d_lookup(dir, name);
1334 if (dentry) {
Jeff Layton39e3c952012-11-28 11:30:53 -05001335 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
Al Viro201f9562012-06-22 12:42:10 +04001336 error = d_revalidate(dentry, flags);
Miklos Szeredibad61182012-03-26 12:54:24 +02001337 if (unlikely(error <= 0)) {
1338 if (error < 0) {
1339 dput(dentry);
1340 return ERR_PTR(error);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001341 } else {
1342 d_invalidate(dentry);
Miklos Szeredibad61182012-03-26 12:54:24 +02001343 dput(dentry);
1344 dentry = NULL;
1345 }
1346 }
1347 }
1348 }
Nick Pigginbaa03892010-08-18 04:37:31 +10001349
Miklos Szeredibad61182012-03-26 12:54:24 +02001350 if (!dentry) {
1351 dentry = d_alloc(dir, name);
1352 if (unlikely(!dentry))
1353 return ERR_PTR(-ENOMEM);
Nick Pigginbaa03892010-08-18 04:37:31 +10001354
Miklos Szeredibad61182012-03-26 12:54:24 +02001355 *need_lookup = true;
Nick Pigginbaa03892010-08-18 04:37:31 +10001356 }
1357 return dentry;
1358}
1359
1360/*
J. Bruce Fields13a2c3b2013-10-23 16:09:16 -04001361 * Call i_op->lookup on the dentry. The dentry must be negative and
1362 * unhashed.
Miklos Szeredibad61182012-03-26 12:54:24 +02001363 *
1364 * dir->d_inode->i_mutex must be held
Josef Bacik44396f42011-05-31 11:58:49 -04001365 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001366static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
Al Viro72bd8662012-06-10 17:17:17 -04001367 unsigned int flags)
Josef Bacik44396f42011-05-31 11:58:49 -04001368{
Josef Bacik44396f42011-05-31 11:58:49 -04001369 struct dentry *old;
1370
1371 /* Don't create child dentry for a dead directory. */
Miklos Szeredibad61182012-03-26 12:54:24 +02001372 if (unlikely(IS_DEADDIR(dir))) {
Miklos Szeredie188dc02012-02-03 14:25:18 +01001373 dput(dentry);
Josef Bacik44396f42011-05-31 11:58:49 -04001374 return ERR_PTR(-ENOENT);
Miklos Szeredie188dc02012-02-03 14:25:18 +01001375 }
Josef Bacik44396f42011-05-31 11:58:49 -04001376
Al Viro72bd8662012-06-10 17:17:17 -04001377 old = dir->i_op->lookup(dir, dentry, flags);
Josef Bacik44396f42011-05-31 11:58:49 -04001378 if (unlikely(old)) {
1379 dput(dentry);
1380 dentry = old;
1381 }
1382 return dentry;
1383}
1384
Al Viroa3255542012-03-30 14:41:51 -04001385static struct dentry *__lookup_hash(struct qstr *name,
Al Viro72bd8662012-06-10 17:17:17 -04001386 struct dentry *base, unsigned int flags)
Al Viroa3255542012-03-30 14:41:51 -04001387{
Miklos Szeredibad61182012-03-26 12:54:24 +02001388 bool need_lookup;
Al Viroa3255542012-03-30 14:41:51 -04001389 struct dentry *dentry;
1390
Al Viro72bd8662012-06-10 17:17:17 -04001391 dentry = lookup_dcache(name, base, flags, &need_lookup);
Miklos Szeredibad61182012-03-26 12:54:24 +02001392 if (!need_lookup)
1393 return dentry;
Al Viroa3255542012-03-30 14:41:51 -04001394
Al Viro72bd8662012-06-10 17:17:17 -04001395 return lookup_real(base->d_inode, dentry, flags);
Al Viroa3255542012-03-30 14:41:51 -04001396}
1397
Josef Bacik44396f42011-05-31 11:58:49 -04001398/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 * It's more convoluted than I'd like it to be, but... it's still fairly
1400 * small and for now I'd prefer to have fast path as straight as possible.
1401 * It _is_ time-critical.
1402 */
Al Viroe97cdc82013-01-24 18:16:00 -05001403static int lookup_fast(struct nameidata *nd,
Miklos Szeredi697f5142012-05-21 17:30:05 +02001404 struct path *path, struct inode **inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405{
Jan Blunck4ac91372008-02-14 19:34:32 -08001406 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001407 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001408 int need_reval = 1;
1409 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001410 int err;
1411
Al Viro3cac2602009-08-13 18:27:43 +04001412 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001413 * Rename seqlock is not required here because in the off chance
1414 * of a false negative due to a concurrent rename, we're going to
1415 * do the non-racy lookup, below.
1416 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001417 if (nd->flags & LOOKUP_RCU) {
1418 unsigned seq;
Linus Torvaldsda53be12013-05-21 15:22:44 -07001419 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
Al Viro5a18fff2011-03-11 04:44:53 -05001420 if (!dentry)
1421 goto unlazy;
1422
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001423 /*
1424 * This sequence count validates that the inode matches
1425 * the dentry name information from lookup.
1426 */
1427 *inode = dentry->d_inode;
1428 if (read_seqcount_retry(&dentry->d_seq, seq))
1429 return -ECHILD;
1430
1431 /*
1432 * This sequence count validates that the parent had no
1433 * changes while we did the lookup of the dentry above.
1434 *
1435 * The memory barrier in read_seqcount_begin of child is
1436 * enough, we can use __read_seqcount_retry here.
1437 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001438 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1439 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001440 nd->seq = seq;
Al Viro5a18fff2011-03-11 04:44:53 -05001441
Al Viro24643082011-02-15 01:26:22 -05001442 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
Al Viro4ce16ef32012-06-10 16:10:59 -04001443 status = d_revalidate(dentry, nd->flags);
Al Viro5a18fff2011-03-11 04:44:53 -05001444 if (unlikely(status <= 0)) {
1445 if (status != -ECHILD)
1446 need_reval = 0;
1447 goto unlazy;
1448 }
Al Viro24643082011-02-15 01:26:22 -05001449 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001450 path->mnt = mnt;
1451 path->dentry = dentry;
NeilBrownb8faf032014-08-04 17:06:29 +10001452 if (likely(__follow_mount_rcu(nd, path, inode)))
1453 return 0;
Al Viro5a18fff2011-03-11 04:44:53 -05001454unlazy:
Al Viro19660af2011-03-25 10:32:48 -04001455 if (unlazy_walk(nd, dentry))
1456 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001457 } else {
Al Viroe97cdc82013-01-24 18:16:00 -05001458 dentry = __d_lookup(parent, &nd->last);
Nick Piggin31e6b012011-01-07 17:49:52 +11001459 }
Al Viro5a18fff2011-03-11 04:44:53 -05001460
Al Viro81e6f522012-03-30 14:48:04 -04001461 if (unlikely(!dentry))
1462 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001463
Al Viro5a18fff2011-03-11 04:44:53 -05001464 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
Al Viro4ce16ef32012-06-10 16:10:59 -04001465 status = d_revalidate(dentry, nd->flags);
Al Viro5a18fff2011-03-11 04:44:53 -05001466 if (unlikely(status <= 0)) {
1467 if (status < 0) {
1468 dput(dentry);
1469 return status;
1470 }
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001471 d_invalidate(dentry);
1472 dput(dentry);
1473 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001474 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001475
David Howells9875cf82011-01-14 18:45:21 +00001476 path->mnt = mnt;
1477 path->dentry = dentry;
1478 err = follow_managed(path, nd->flags);
Ian Kent89312212011-01-18 12:06:10 +08001479 if (unlikely(err < 0)) {
1480 path_put_conditional(path, nd);
David Howells9875cf82011-01-14 18:45:21 +00001481 return err;
Ian Kent89312212011-01-18 12:06:10 +08001482 }
Al Viroa3fbbde2011-11-07 21:21:26 +00001483 if (err)
1484 nd->flags |= LOOKUP_JUMPED;
David Howells9875cf82011-01-14 18:45:21 +00001485 *inode = path->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 return 0;
Al Viro81e6f522012-03-30 14:48:04 -04001487
1488need_lookup:
Miklos Szeredi697f5142012-05-21 17:30:05 +02001489 return 1;
1490}
1491
1492/* Fast lookup failed, do it the slow way */
Al Virocc2a5272013-01-24 18:19:49 -05001493static int lookup_slow(struct nameidata *nd, struct path *path)
Miklos Szeredi697f5142012-05-21 17:30:05 +02001494{
1495 struct dentry *dentry, *parent;
1496 int err;
1497
1498 parent = nd->path.dentry;
Al Viro81e6f522012-03-30 14:48:04 -04001499 BUG_ON(nd->inode != parent->d_inode);
1500
1501 mutex_lock(&parent->d_inode->i_mutex);
Al Virocc2a5272013-01-24 18:19:49 -05001502 dentry = __lookup_hash(&nd->last, parent, nd->flags);
Al Viro81e6f522012-03-30 14:48:04 -04001503 mutex_unlock(&parent->d_inode->i_mutex);
1504 if (IS_ERR(dentry))
1505 return PTR_ERR(dentry);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001506 path->mnt = nd->path.mnt;
1507 path->dentry = dentry;
1508 err = follow_managed(path, nd->flags);
1509 if (unlikely(err < 0)) {
1510 path_put_conditional(path, nd);
1511 return err;
1512 }
1513 if (err)
1514 nd->flags |= LOOKUP_JUMPED;
1515 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516}
1517
Al Viro52094c82011-02-21 21:34:47 -05001518static inline int may_lookup(struct nameidata *nd)
1519{
1520 if (nd->flags & LOOKUP_RCU) {
Al Viro4ad5abb2011-06-20 19:57:03 -04001521 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
Al Viro52094c82011-02-21 21:34:47 -05001522 if (err != -ECHILD)
1523 return err;
Al Viro19660af2011-03-25 10:32:48 -04001524 if (unlazy_walk(nd, NULL))
Al Viro52094c82011-02-21 21:34:47 -05001525 return -ECHILD;
1526 }
Al Viro4ad5abb2011-06-20 19:57:03 -04001527 return inode_permission(nd->inode, MAY_EXEC);
Al Viro52094c82011-02-21 21:34:47 -05001528}
1529
Al Viro9856fa12011-03-04 14:22:06 -05001530static inline int handle_dots(struct nameidata *nd, int type)
1531{
1532 if (type == LAST_DOTDOT) {
1533 if (nd->flags & LOOKUP_RCU) {
1534 if (follow_dotdot_rcu(nd))
1535 return -ECHILD;
1536 } else
1537 follow_dotdot(nd);
1538 }
1539 return 0;
1540}
1541
Al Viro951361f2011-03-04 14:44:37 -05001542static void terminate_walk(struct nameidata *nd)
1543{
1544 if (!(nd->flags & LOOKUP_RCU)) {
1545 path_put(&nd->path);
1546 } else {
1547 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001548 if (!(nd->flags & LOOKUP_ROOT))
1549 nd->root.mnt = NULL;
Al Viro8b61e742013-11-08 12:45:01 -05001550 rcu_read_unlock();
Al Viro951361f2011-03-04 14:44:37 -05001551 }
1552}
1553
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001554/*
1555 * Do we need to follow links? We _really_ want to be able
1556 * to do this check without having to look at inode->i_op,
1557 * so we keep a cache of "no, this doesn't need follow_link"
1558 * for the common case.
1559 */
David Howellsb18825a2013-09-12 19:22:53 +01001560static inline int should_follow_link(struct dentry *dentry, int follow)
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001561{
David Howellsb18825a2013-09-12 19:22:53 +01001562 return unlikely(d_is_symlink(dentry)) ? follow : 0;
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001563}
1564
Al Viroce57dfc2011-03-13 19:58:58 -04001565static inline int walk_component(struct nameidata *nd, struct path *path,
Al Viro21b9b072013-01-24 18:10:25 -05001566 int follow)
Al Viroce57dfc2011-03-13 19:58:58 -04001567{
1568 struct inode *inode;
1569 int err;
1570 /*
1571 * "." and ".." are special - ".." especially so because it has
1572 * to be able to know about the current root directory and
1573 * parent relationships.
1574 */
Al Viro21b9b072013-01-24 18:10:25 -05001575 if (unlikely(nd->last_type != LAST_NORM))
1576 return handle_dots(nd, nd->last_type);
Al Viroe97cdc82013-01-24 18:16:00 -05001577 err = lookup_fast(nd, path, &inode);
Al Viroce57dfc2011-03-13 19:58:58 -04001578 if (unlikely(err)) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001579 if (err < 0)
1580 goto out_err;
1581
Al Virocc2a5272013-01-24 18:19:49 -05001582 err = lookup_slow(nd, path);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001583 if (err < 0)
1584 goto out_err;
1585
1586 inode = path->dentry->d_inode;
Al Viroce57dfc2011-03-13 19:58:58 -04001587 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001588 err = -ENOENT;
Al Viro22213312014-04-19 12:30:58 -04001589 if (!inode || d_is_negative(path->dentry))
Miklos Szeredi697f5142012-05-21 17:30:05 +02001590 goto out_path_put;
1591
David Howellsb18825a2013-09-12 19:22:53 +01001592 if (should_follow_link(path->dentry, follow)) {
Al Viro19660af2011-03-25 10:32:48 -04001593 if (nd->flags & LOOKUP_RCU) {
1594 if (unlikely(unlazy_walk(nd, path->dentry))) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001595 err = -ECHILD;
1596 goto out_err;
Al Viro19660af2011-03-25 10:32:48 -04001597 }
1598 }
Al Viroce57dfc2011-03-13 19:58:58 -04001599 BUG_ON(inode != path->dentry->d_inode);
1600 return 1;
1601 }
1602 path_to_nameidata(path, nd);
1603 nd->inode = inode;
1604 return 0;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001605
1606out_path_put:
1607 path_to_nameidata(path, nd);
1608out_err:
1609 terminate_walk(nd);
1610 return err;
Al Viroce57dfc2011-03-13 19:58:58 -04001611}
1612
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613/*
Al Virob3563792011-03-14 21:54:55 -04001614 * This limits recursive symlink follows to 8, while
1615 * limiting consecutive symlinks to 40.
1616 *
1617 * Without that kind of total limit, nasty chains of consecutive
1618 * symlinks can cause almost arbitrarily long lookups.
1619 */
1620static inline int nested_symlink(struct path *path, struct nameidata *nd)
1621{
1622 int res;
1623
Al Virob3563792011-03-14 21:54:55 -04001624 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1625 path_put_conditional(path, nd);
1626 path_put(&nd->path);
1627 return -ELOOP;
1628 }
Erez Zadok1a4022f2011-05-21 01:19:59 -04001629 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
Al Virob3563792011-03-14 21:54:55 -04001630
1631 nd->depth++;
1632 current->link_count++;
1633
1634 do {
1635 struct path link = *path;
1636 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04001637
1638 res = follow_link(&link, nd, &cookie);
Al Viro6d7b5aa2012-06-10 04:15:17 -04001639 if (res)
1640 break;
Al Viro21b9b072013-01-24 18:10:25 -05001641 res = walk_component(nd, path, LOOKUP_FOLLOW);
Al Viro574197e2011-03-14 22:20:34 -04001642 put_link(nd, &link, cookie);
Al Virob3563792011-03-14 21:54:55 -04001643 } while (res > 0);
1644
1645 current->link_count--;
1646 nd->depth--;
1647 return res;
1648}
1649
1650/*
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001651 * We can do the critical dentry name comparison and hashing
1652 * operations one word at a time, but we are limited to:
1653 *
1654 * - Architectures with fast unaligned word accesses. We could
1655 * do a "get_unaligned()" if this helps and is sufficiently
1656 * fast.
1657 *
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001658 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1659 * do not trap on the (extremely unlikely) case of a page
1660 * crossing operation.
1661 *
1662 * - Furthermore, we need an efficient 64-bit compile for the
1663 * 64-bit case in order to generate the "number of bytes in
1664 * the final mask". Again, that could be replaced with a
1665 * efficient population count instruction or similar.
1666 */
1667#ifdef CONFIG_DCACHE_WORD_ACCESS
1668
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001669#include <asm/word-at-a-time.h>
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001670
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001671#ifdef CONFIG_64BIT
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001672
1673static inline unsigned int fold_hash(unsigned long hash)
1674{
Linus Torvalds99d263d2014-09-13 11:30:10 -07001675 return hash_64(hash, 32);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001676}
1677
1678#else /* 32-bit case */
1679
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001680#define fold_hash(x) (x)
1681
1682#endif
1683
1684unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1685{
1686 unsigned long a, mask;
1687 unsigned long hash = 0;
1688
1689 for (;;) {
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001690 a = load_unaligned_zeropad(name);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001691 if (len < sizeof(unsigned long))
1692 break;
1693 hash += a;
Al Virof132c5b2012-03-22 21:59:52 +00001694 hash *= 9;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001695 name += sizeof(unsigned long);
1696 len -= sizeof(unsigned long);
1697 if (!len)
1698 goto done;
1699 }
Will Deacona5c21dc2013-12-12 17:40:21 +00001700 mask = bytemask_from_count(len);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001701 hash += mask & a;
1702done:
1703 return fold_hash(hash);
1704}
1705EXPORT_SYMBOL(full_name_hash);
1706
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001707/*
1708 * Calculate the length and hash of the path component, and
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001709 * return the "hash_len" as the result.
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001710 */
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001711static inline u64 hash_name(const char *name)
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001712{
Linus Torvalds36126f82012-05-26 10:43:17 -07001713 unsigned long a, b, adata, bdata, mask, hash, len;
1714 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001715
1716 hash = a = 0;
1717 len = -sizeof(unsigned long);
1718 do {
1719 hash = (hash + a) * 9;
1720 len += sizeof(unsigned long);
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001721 a = load_unaligned_zeropad(name+len);
Linus Torvalds36126f82012-05-26 10:43:17 -07001722 b = a ^ REPEAT_BYTE('/');
1723 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001724
Linus Torvalds36126f82012-05-26 10:43:17 -07001725 adata = prep_zero_mask(a, adata, &constants);
1726 bdata = prep_zero_mask(b, bdata, &constants);
1727
1728 mask = create_zero_mask(adata | bdata);
1729
1730 hash += a & zero_bytemask(mask);
Linus Torvalds9226b5b2014-09-14 17:28:32 -07001731 len += find_zero(mask);
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001732 return hashlen_create(fold_hash(hash), len);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001733}
1734
1735#else
1736
Linus Torvalds0145acc2012-03-02 14:32:59 -08001737unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1738{
1739 unsigned long hash = init_name_hash();
1740 while (len--)
1741 hash = partial_name_hash(*name++, hash);
1742 return end_name_hash(hash);
1743}
Linus Torvaldsae942ae2012-03-02 19:40:57 -08001744EXPORT_SYMBOL(full_name_hash);
Linus Torvalds0145acc2012-03-02 14:32:59 -08001745
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001746/*
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001747 * We know there's a real path component here of at least
1748 * one character.
1749 */
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001750static inline u64 hash_name(const char *name)
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001751{
1752 unsigned long hash = init_name_hash();
1753 unsigned long len = 0, c;
1754
1755 c = (unsigned char)*name;
1756 do {
1757 len++;
1758 hash = partial_name_hash(c, hash);
1759 c = (unsigned char)name[len];
1760 } while (c && c != '/');
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001761 return hashlen_create(end_name_hash(hash), len);
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001762}
1763
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001764#endif
1765
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001766/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01001768 * This is the basic name resolution function, turning a pathname into
1769 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01001771 * Returns 0 and nd will have valid dentry and mnt on success.
1772 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 */
Al Viro6de88d72009-08-09 01:41:57 +04001774static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775{
1776 struct path next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778
1779 while (*name=='/')
1780 name++;
1781 if (!*name)
Al Viro086e1832011-02-22 20:56:27 -05001782 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 /* At this point we know we have a real path component. */
1785 for(;;) {
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001786 u64 hash_len;
Al Virofe479a52011-02-22 15:10:03 -05001787 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788
Al Viro52094c82011-02-21 21:34:47 -05001789 err = may_lookup(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 if (err)
1791 break;
1792
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001793 hash_len = hash_name(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794
Al Virofe479a52011-02-22 15:10:03 -05001795 type = LAST_NORM;
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001796 if (name[0] == '.') switch (hashlen_len(hash_len)) {
Al Virofe479a52011-02-22 15:10:03 -05001797 case 2:
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001798 if (name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05001799 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05001800 nd->flags |= LOOKUP_JUMPED;
1801 }
Al Virofe479a52011-02-22 15:10:03 -05001802 break;
1803 case 1:
1804 type = LAST_DOT;
1805 }
Al Viro5a202bc2011-03-08 14:17:44 -05001806 if (likely(type == LAST_NORM)) {
1807 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05001808 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05001809 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
James Hogana060dc52014-09-16 13:07:35 +01001810 struct qstr this = { { .hash_len = hash_len }, .name = name };
Linus Torvaldsda53be12013-05-21 15:22:44 -07001811 err = parent->d_op->d_hash(parent, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05001812 if (err < 0)
1813 break;
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001814 hash_len = this.hash_len;
1815 name = this.name;
Al Viro5a202bc2011-03-08 14:17:44 -05001816 }
1817 }
Al Virofe479a52011-02-22 15:10:03 -05001818
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001819 nd->last.hash_len = hash_len;
1820 nd->last.name = name;
Al Viro5f4a6a62013-01-24 18:04:22 -05001821 nd->last_type = type;
1822
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001823 name += hashlen_len(hash_len);
1824 if (!*name)
Al Viro5f4a6a62013-01-24 18:04:22 -05001825 return 0;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001826 /*
1827 * If it wasn't NUL, we know it was '/'. Skip that
1828 * slash, and continue until no more slashes.
1829 */
1830 do {
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001831 name++;
1832 } while (unlikely(*name == '/'));
1833 if (!*name)
Al Viro5f4a6a62013-01-24 18:04:22 -05001834 return 0;
1835
Al Viro21b9b072013-01-24 18:10:25 -05001836 err = walk_component(nd, &next, LOOKUP_FOLLOW);
Al Viroce57dfc2011-03-13 19:58:58 -04001837 if (err < 0)
1838 return err;
Al Virofe479a52011-02-22 15:10:03 -05001839
Al Viroce57dfc2011-03-13 19:58:58 -04001840 if (err) {
Al Virob3563792011-03-14 21:54:55 -04001841 err = nested_symlink(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842 if (err)
Al Viroa7472ba2011-03-04 14:39:30 -05001843 return err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001844 }
Miklos Szeredi44b1d532014-04-01 17:08:41 +02001845 if (!d_can_lookup(nd->path.dentry)) {
Al Viro5f4a6a62013-01-24 18:04:22 -05001846 err = -ENOTDIR;
1847 break;
1848 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 }
Al Viro951361f2011-03-04 14:44:37 -05001850 terminate_walk(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851 return err;
1852}
1853
Al Viro70e9b352011-03-05 21:12:22 -05001854static int path_init(int dfd, const char *name, unsigned int flags,
Al Viro5e530842014-11-20 14:14:42 -05001855 struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001857 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858
1859 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro980f3ea2014-11-20 14:20:24 -05001860 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 nd->depth = 0;
Al Viro5e530842014-11-20 14:14:42 -05001862 nd->base = NULL;
Al Viro5b6ca022011-03-09 23:04:47 -05001863 if (flags & LOOKUP_ROOT) {
David Howellsb18825a2013-09-12 19:22:53 +01001864 struct dentry *root = nd->root.dentry;
1865 struct inode *inode = root->d_inode;
Al Viro73d049a2011-03-11 12:08:24 -05001866 if (*name) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02001867 if (!d_can_lookup(root))
Al Viro73d049a2011-03-11 12:08:24 -05001868 return -ENOTDIR;
1869 retval = inode_permission(inode, MAY_EXEC);
1870 if (retval)
1871 return retval;
1872 }
Al Viro5b6ca022011-03-09 23:04:47 -05001873 nd->path = nd->root;
1874 nd->inode = inode;
1875 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05001876 rcu_read_lock();
Al Viro5b6ca022011-03-09 23:04:47 -05001877 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro48a066e2013-09-29 22:06:07 -04001878 nd->m_seq = read_seqbegin(&mount_lock);
Al Viro5b6ca022011-03-09 23:04:47 -05001879 } else {
1880 path_get(&nd->path);
1881 }
Al Virod4658872014-11-20 14:23:33 -05001882 goto done;
Al Viro5b6ca022011-03-09 23:04:47 -05001883 }
1884
Al Viro2a737872009-04-07 11:49:53 -04001885 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886
Al Viro48a066e2013-09-29 22:06:07 -04001887 nd->m_seq = read_seqbegin(&mount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888 if (*name=='/') {
Al Viroe41f7d42011-02-22 14:02:58 -05001889 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05001890 rcu_read_lock();
Al Viro7bd88372014-09-13 21:55:46 -04001891 nd->seq = set_root_rcu(nd);
Al Viroe41f7d42011-02-22 14:02:58 -05001892 } else {
1893 set_root(nd);
1894 path_get(&nd->root);
1895 }
Al Viro2a737872009-04-07 11:49:53 -04001896 nd->path = nd->root;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001897 } else if (dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05001898 if (flags & LOOKUP_RCU) {
1899 struct fs_struct *fs = current->fs;
1900 unsigned seq;
1901
Al Viro8b61e742013-11-08 12:45:01 -05001902 rcu_read_lock();
Al Viroe41f7d42011-02-22 14:02:58 -05001903
1904 do {
1905 seq = read_seqcount_begin(&fs->seq);
1906 nd->path = fs->pwd;
1907 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1908 } while (read_seqcount_retry(&fs->seq, seq));
1909 } else {
1910 get_fs_pwd(current->fs, &nd->path);
1911 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001912 } else {
Jeff Layton582aa642012-12-11 08:56:16 -05001913 /* Caller must check execute permissions on the starting path component */
Al Viro2903ff02012-08-28 12:52:22 -04001914 struct fd f = fdget_raw(dfd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001915 struct dentry *dentry;
1916
Al Viro2903ff02012-08-28 12:52:22 -04001917 if (!f.file)
1918 return -EBADF;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001919
Al Viro2903ff02012-08-28 12:52:22 -04001920 dentry = f.file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001921
Al Virof52e0c12011-03-14 18:56:51 -04001922 if (*name) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02001923 if (!d_can_lookup(dentry)) {
Al Viro2903ff02012-08-28 12:52:22 -04001924 fdput(f);
1925 return -ENOTDIR;
1926 }
Al Virof52e0c12011-03-14 18:56:51 -04001927 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001928
Al Viro2903ff02012-08-28 12:52:22 -04001929 nd->path = f.file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05001930 if (flags & LOOKUP_RCU) {
Linus Torvalds9c225f22014-03-03 09:36:58 -08001931 if (f.flags & FDPUT_FPUT)
Al Viro5e530842014-11-20 14:14:42 -05001932 nd->base = f.file;
Al Viroe41f7d42011-02-22 14:02:58 -05001933 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro8b61e742013-11-08 12:45:01 -05001934 rcu_read_lock();
Al Viroe41f7d42011-02-22 14:02:58 -05001935 } else {
Al Viro2903ff02012-08-28 12:52:22 -04001936 path_get(&nd->path);
1937 fdput(f);
Al Viroe41f7d42011-02-22 14:02:58 -05001938 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 }
Al Viroe41f7d42011-02-22 14:02:58 -05001940
Nick Piggin31e6b012011-01-07 17:49:52 +11001941 nd->inode = nd->path.dentry->d_inode;
Al Viro4023bfc2014-09-13 21:59:43 -04001942 if (!(flags & LOOKUP_RCU))
Al Virod4658872014-11-20 14:23:33 -05001943 goto done;
Al Viro4023bfc2014-09-13 21:59:43 -04001944 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
Al Virod4658872014-11-20 14:23:33 -05001945 goto done;
Al Viro4023bfc2014-09-13 21:59:43 -04001946 if (!(nd->flags & LOOKUP_ROOT))
1947 nd->root.mnt = NULL;
1948 rcu_read_unlock();
1949 return -ECHILD;
Al Virod4658872014-11-20 14:23:33 -05001950done:
1951 current->total_link_count = 0;
1952 return link_path_walk(name, nd);
Al Viro9b4a9b12009-04-07 11:44:16 -04001953}
1954
Al Viro893b7772014-11-20 14:18:09 -05001955static void path_cleanup(struct nameidata *nd)
1956{
1957 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1958 path_put(&nd->root);
1959 nd->root.mnt = NULL;
1960 }
1961 if (unlikely(nd->base))
1962 fput(nd->base);
Al Viro9b4a9b12009-04-07 11:44:16 -04001963}
1964
Al Virobd92d7f2011-03-14 19:54:59 -04001965static inline int lookup_last(struct nameidata *nd, struct path *path)
1966{
1967 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1968 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1969
1970 nd->flags &= ~LOOKUP_PARENT;
Al Viro21b9b072013-01-24 18:10:25 -05001971 return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW);
Al Virobd92d7f2011-03-14 19:54:59 -04001972}
1973
Al Viro9b4a9b12009-04-07 11:44:16 -04001974/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroee0827c2011-02-21 23:38:09 -05001975static int path_lookupat(int dfd, const char *name,
Al Viro9b4a9b12009-04-07 11:44:16 -04001976 unsigned int flags, struct nameidata *nd)
1977{
Al Virobd92d7f2011-03-14 19:54:59 -04001978 struct path path;
1979 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001980
1981 /*
1982 * Path walking is largely split up into 2 different synchronisation
1983 * schemes, rcu-walk and ref-walk (explained in
1984 * Documentation/filesystems/path-lookup.txt). These share much of the
1985 * path walk code, but some things particularly setup, cleanup, and
1986 * following mounts are sufficiently divergent that functions are
1987 * duplicated. Typically there is a function foo(), and its RCU
1988 * analogue, foo_rcu().
1989 *
1990 * -ECHILD is the error number of choice (just to avoid clashes) that
1991 * is returned if some aspect of an rcu-walk fails. Such an error must
1992 * be handled by restarting a traditional ref-walk (which will always
1993 * be able to complete).
1994 */
Al Viro980f3ea2014-11-20 14:20:24 -05001995 err = path_init(dfd, name, flags, nd);
Al Virobd92d7f2011-03-14 19:54:59 -04001996 if (!err && !(flags & LOOKUP_PARENT)) {
Al Virobd92d7f2011-03-14 19:54:59 -04001997 err = lookup_last(nd, &path);
1998 while (err > 0) {
1999 void *cookie;
2000 struct path link = path;
Kees Cook800179c2012-07-25 17:29:07 -07002001 err = may_follow_link(&link, nd);
2002 if (unlikely(err))
2003 break;
Al Virobd92d7f2011-03-14 19:54:59 -04002004 nd->flags |= LOOKUP_PARENT;
Al Viro574197e2011-03-14 22:20:34 -04002005 err = follow_link(&link, nd, &cookie);
Al Viro6d7b5aa2012-06-10 04:15:17 -04002006 if (err)
2007 break;
2008 err = lookup_last(nd, &path);
Al Viro574197e2011-03-14 22:20:34 -04002009 put_link(nd, &link, cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04002010 }
2011 }
Al Viroee0827c2011-02-21 23:38:09 -05002012
Al Viro9f1fafe2011-03-25 11:00:12 -04002013 if (!err)
2014 err = complete_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04002015
2016 if (!err && nd->flags & LOOKUP_DIRECTORY) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002017 if (!d_can_lookup(nd->path.dentry)) {
Al Virobd92d7f2011-03-14 19:54:59 -04002018 path_put(&nd->path);
Al Virobd23a532011-03-23 09:56:30 -04002019 err = -ENOTDIR;
Al Virobd92d7f2011-03-14 19:54:59 -04002020 }
2021 }
Al Viro16c2cd72011-02-22 15:50:10 -05002022
Al Viro893b7772014-11-20 14:18:09 -05002023 path_cleanup(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04002024 return err;
Al Viroee0827c2011-02-21 23:38:09 -05002025}
Nick Piggin31e6b012011-01-07 17:49:52 +11002026
Jeff Layton873f1ee2012-10-10 15:25:29 -04002027static int filename_lookup(int dfd, struct filename *name,
2028 unsigned int flags, struct nameidata *nd)
2029{
2030 int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd);
2031 if (unlikely(retval == -ECHILD))
2032 retval = path_lookupat(dfd, name->name, flags, nd);
2033 if (unlikely(retval == -ESTALE))
2034 retval = path_lookupat(dfd, name->name,
2035 flags | LOOKUP_REVAL, nd);
2036
2037 if (likely(!retval))
Jeff Laytonadb5c242012-10-10 16:43:13 -04002038 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002039 return retval;
2040}
2041
Al Viroee0827c2011-02-21 23:38:09 -05002042static int do_path_lookup(int dfd, const char *name,
2043 unsigned int flags, struct nameidata *nd)
2044{
Paul Moore51689102015-01-22 00:00:03 -05002045 struct filename *filename = getname_kernel(name);
2046 int retval = PTR_ERR(filename);
Nick Piggin31e6b012011-01-07 17:49:52 +11002047
Paul Moore51689102015-01-22 00:00:03 -05002048 if (!IS_ERR(filename)) {
2049 retval = filename_lookup(dfd, filename, flags, nd);
2050 putname(filename);
2051 }
2052 return retval;
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{
Paul Moore51689102015-01-22 00:00:03 -05002058 struct filename *filename = getname_kernel(name);
Al Viro79714f72012-06-15 03:01:42 +04002059 struct nameidata nd;
2060 struct dentry *d;
Paul Moore51689102015-01-22 00:00:03 -05002061 int err;
2062
2063 if (IS_ERR(filename))
2064 return ERR_CAST(filename);
2065
2066 err = filename_lookup(AT_FDCWD, filename, LOOKUP_PARENT, &nd);
2067 if (err) {
2068 d = ERR_PTR(err);
2069 goto out;
2070 }
Al Viro79714f72012-06-15 03:01:42 +04002071 if (nd.last_type != LAST_NORM) {
2072 path_put(&nd.path);
Paul Moore51689102015-01-22 00:00:03 -05002073 d = ERR_PTR(-EINVAL);
2074 goto out;
Al Viro79714f72012-06-15 03:01:42 +04002075 }
2076 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Al Viro1e0ea002012-07-22 23:46:21 +04002077 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
Al Viro79714f72012-06-15 03:01:42 +04002078 if (IS_ERR(d)) {
2079 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2080 path_put(&nd.path);
Paul Moore51689102015-01-22 00:00:03 -05002081 goto out;
Al Viro79714f72012-06-15 03:01:42 +04002082 }
2083 *path = nd.path;
Paul Moore51689102015-01-22 00:00:03 -05002084out:
2085 putname(filename);
Al Viro79714f72012-06-15 03:01:42 +04002086 return d;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002087}
2088
Al Virod1811462008-08-02 00:49:18 -04002089int kern_path(const char *name, unsigned int flags, struct path *path)
2090{
2091 struct nameidata nd;
2092 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
2093 if (!res)
2094 *path = nd.path;
2095 return res;
2096}
Al Viro4d359502014-03-14 12:20:17 -04002097EXPORT_SYMBOL(kern_path);
Al Virod1811462008-08-02 00:49:18 -04002098
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002099/**
2100 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2101 * @dentry: pointer to dentry of the base directory
2102 * @mnt: pointer to vfs mount of the base directory
2103 * @name: pointer to file name
2104 * @flags: lookup flags
Al Viroe0a01242011-06-27 17:00:37 -04002105 * @path: pointer to struct path to fill
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002106 */
2107int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2108 const char *name, unsigned int flags,
Al Viroe0a01242011-06-27 17:00:37 -04002109 struct path *path)
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002110{
Al Viroe0a01242011-06-27 17:00:37 -04002111 struct nameidata nd;
2112 int err;
2113 nd.root.dentry = dentry;
2114 nd.root.mnt = mnt;
2115 BUG_ON(flags & LOOKUP_PARENT);
Al Viro5b6ca022011-03-09 23:04:47 -05002116 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
Al Viroe0a01242011-06-27 17:00:37 -04002117 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
2118 if (!err)
2119 *path = nd.path;
2120 return err;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002121}
Al Viro4d359502014-03-14 12:20:17 -04002122EXPORT_SYMBOL(vfs_path_lookup);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002123
James Morris057f6c02007-04-26 00:12:05 -07002124/*
2125 * Restricted form of lookup. Doesn't follow links, single-component only,
2126 * needs parent already locked. Doesn't follow mounts.
2127 * SMP-safe.
2128 */
Adrian Bunka244e162006-03-31 02:32:11 -08002129static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130{
Al Viro72bd8662012-06-10 17:17:17 -04002131 return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132}
2133
Christoph Hellwigeead1912007-10-16 23:25:38 -07002134/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07002135 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07002136 * @name: pathname component to lookup
2137 * @base: base directory to lookup from
2138 * @len: maximum length @len should be interpreted to
2139 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07002140 * Note that this routine is purely a helper for filesystem usage and should
2141 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07002142 * nameidata argument is passed to the filesystem methods and a filesystem
2143 * using this helper needs to be prepared for that.
2144 */
James Morris057f6c02007-04-26 00:12:05 -07002145struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2146{
James Morris057f6c02007-04-26 00:12:05 -07002147 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05002148 unsigned int c;
Miklos Szeredicda309d2012-03-26 12:54:21 +02002149 int err;
James Morris057f6c02007-04-26 00:12:05 -07002150
David Woodhouse2f9092e2009-04-20 23:18:37 +01002151 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2152
Al Viro6a96ba52011-03-07 23:49:20 -05002153 this.name = name;
2154 this.len = len;
Linus Torvalds0145acc2012-03-02 14:32:59 -08002155 this.hash = full_name_hash(name, len);
Al Viro6a96ba52011-03-07 23:49:20 -05002156 if (!len)
2157 return ERR_PTR(-EACCES);
2158
Al Viro21d8a152012-11-29 22:17:21 -05002159 if (unlikely(name[0] == '.')) {
2160 if (len < 2 || (len == 2 && name[1] == '.'))
2161 return ERR_PTR(-EACCES);
2162 }
2163
Al Viro6a96ba52011-03-07 23:49:20 -05002164 while (len--) {
2165 c = *(const unsigned char *)name++;
2166 if (c == '/' || c == '\0')
2167 return ERR_PTR(-EACCES);
Al Viro6a96ba52011-03-07 23:49:20 -05002168 }
Al Viro5a202bc2011-03-08 14:17:44 -05002169 /*
2170 * See if the low-level filesystem might want
2171 * to use its own hash..
2172 */
2173 if (base->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002174 int err = base->d_op->d_hash(base, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05002175 if (err < 0)
2176 return ERR_PTR(err);
2177 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07002178
Miklos Szeredicda309d2012-03-26 12:54:21 +02002179 err = inode_permission(base->d_inode, MAY_EXEC);
2180 if (err)
2181 return ERR_PTR(err);
2182
Al Viro72bd8662012-06-10 17:17:17 -04002183 return __lookup_hash(&this, base, 0);
James Morris057f6c02007-04-26 00:12:05 -07002184}
Al Viro4d359502014-03-14 12:20:17 -04002185EXPORT_SYMBOL(lookup_one_len);
James Morris057f6c02007-04-26 00:12:05 -07002186
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002187int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2188 struct path *path, int *empty)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189{
Al Viro2d8f3032008-07-22 09:59:21 -04002190 struct nameidata nd;
Jeff Layton91a27b22012-10-10 15:25:28 -04002191 struct filename *tmp = getname_flags(name, flags, empty);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04002194
2195 BUG_ON(flags & LOOKUP_PARENT);
2196
Jeff Layton873f1ee2012-10-10 15:25:29 -04002197 err = filename_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04002199 if (!err)
2200 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201 }
2202 return err;
2203}
2204
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002205int user_path_at(int dfd, const char __user *name, unsigned flags,
2206 struct path *path)
2207{
Linus Torvaldsf7493e52012-03-22 16:10:40 -07002208 return user_path_at_empty(dfd, name, flags, path, NULL);
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002209}
Al Viro4d359502014-03-14 12:20:17 -04002210EXPORT_SYMBOL(user_path_at);
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002211
Jeff Layton873f1ee2012-10-10 15:25:29 -04002212/*
2213 * NB: most callers don't do anything directly with the reference to the
2214 * to struct filename, but the nd->last pointer points into the name string
2215 * allocated by getname. So we must hold the reference to it until all
2216 * path-walking is complete.
2217 */
Jeff Layton91a27b22012-10-10 15:25:28 -04002218static struct filename *
Jeff Layton9e790bd2012-12-11 12:10:09 -05002219user_path_parent(int dfd, const char __user *path, struct nameidata *nd,
2220 unsigned int flags)
Al Viro2ad94ae2008-07-21 09:32:51 -04002221{
Jeff Layton91a27b22012-10-10 15:25:28 -04002222 struct filename *s = getname(path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002223 int error;
2224
Jeff Layton9e790bd2012-12-11 12:10:09 -05002225 /* only LOOKUP_REVAL is allowed in extra flags */
2226 flags &= LOOKUP_REVAL;
2227
Al Viro2ad94ae2008-07-21 09:32:51 -04002228 if (IS_ERR(s))
Jeff Layton91a27b22012-10-10 15:25:28 -04002229 return s;
Al Viro2ad94ae2008-07-21 09:32:51 -04002230
Jeff Layton9e790bd2012-12-11 12:10:09 -05002231 error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd);
Jeff Layton91a27b22012-10-10 15:25:28 -04002232 if (error) {
Al Viro2ad94ae2008-07-21 09:32:51 -04002233 putname(s);
Jeff Layton91a27b22012-10-10 15:25:28 -04002234 return ERR_PTR(error);
2235 }
Al Viro2ad94ae2008-07-21 09:32:51 -04002236
Jeff Layton91a27b22012-10-10 15:25:28 -04002237 return s;
Al Viro2ad94ae2008-07-21 09:32:51 -04002238}
2239
Jeff Layton80334262013-07-26 06:23:25 -04002240/**
Al Viro197df042013-09-08 14:03:27 -04002241 * mountpoint_last - look up last component for umount
Jeff Layton80334262013-07-26 06:23:25 -04002242 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2243 * @path: pointer to container for result
2244 *
2245 * This is a special lookup_last function just for umount. In this case, we
2246 * need to resolve the path without doing any revalidation.
2247 *
2248 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2249 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2250 * in almost all cases, this lookup will be served out of the dcache. The only
2251 * cases where it won't are if nd->last refers to a symlink or the path is
2252 * bogus and it doesn't exist.
2253 *
2254 * Returns:
2255 * -error: if there was an error during lookup. This includes -ENOENT if the
2256 * lookup found a negative dentry. The nd->path reference will also be
2257 * put in this case.
2258 *
2259 * 0: if we successfully resolved nd->path and found it to not to be a
2260 * symlink that needs to be followed. "path" will also be populated.
2261 * The nd->path reference will also be put.
2262 *
2263 * 1: if we successfully resolved nd->last and found it to be a symlink
2264 * that needs to be followed. "path" will be populated with the path
2265 * to the link, and nd->path will *not* be put.
2266 */
2267static int
Al Viro197df042013-09-08 14:03:27 -04002268mountpoint_last(struct nameidata *nd, struct path *path)
Jeff Layton80334262013-07-26 06:23:25 -04002269{
2270 int error = 0;
2271 struct dentry *dentry;
2272 struct dentry *dir = nd->path.dentry;
2273
Al Viro35759522013-09-08 13:41:33 -04002274 /* If we're in rcuwalk, drop out of it to handle last component */
2275 if (nd->flags & LOOKUP_RCU) {
2276 if (unlazy_walk(nd, NULL)) {
2277 error = -ECHILD;
2278 goto out;
2279 }
Jeff Layton80334262013-07-26 06:23:25 -04002280 }
2281
2282 nd->flags &= ~LOOKUP_PARENT;
2283
2284 if (unlikely(nd->last_type != LAST_NORM)) {
2285 error = handle_dots(nd, nd->last_type);
Al Viro35759522013-09-08 13:41:33 -04002286 if (error)
2287 goto out;
2288 dentry = dget(nd->path.dentry);
2289 goto done;
Jeff Layton80334262013-07-26 06:23:25 -04002290 }
2291
2292 mutex_lock(&dir->d_inode->i_mutex);
2293 dentry = d_lookup(dir, &nd->last);
2294 if (!dentry) {
2295 /*
2296 * No cached dentry. Mounted dentries are pinned in the cache,
2297 * so that means that this dentry is probably a symlink or the
2298 * path doesn't actually point to a mounted dentry.
2299 */
2300 dentry = d_alloc(dir, &nd->last);
2301 if (!dentry) {
2302 error = -ENOMEM;
Dave Jonesbcceeeb2013-09-10 17:04:25 -04002303 mutex_unlock(&dir->d_inode->i_mutex);
Al Viro35759522013-09-08 13:41:33 -04002304 goto out;
Jeff Layton80334262013-07-26 06:23:25 -04002305 }
Al Viro35759522013-09-08 13:41:33 -04002306 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2307 error = PTR_ERR(dentry);
Dave Jonesbcceeeb2013-09-10 17:04:25 -04002308 if (IS_ERR(dentry)) {
2309 mutex_unlock(&dir->d_inode->i_mutex);
Al Viro35759522013-09-08 13:41:33 -04002310 goto out;
Dave Jonesbcceeeb2013-09-10 17:04:25 -04002311 }
Jeff Layton80334262013-07-26 06:23:25 -04002312 }
2313 mutex_unlock(&dir->d_inode->i_mutex);
2314
Al Viro35759522013-09-08 13:41:33 -04002315done:
Al Viro22213312014-04-19 12:30:58 -04002316 if (!dentry->d_inode || d_is_negative(dentry)) {
Al Viro35759522013-09-08 13:41:33 -04002317 error = -ENOENT;
2318 dput(dentry);
2319 goto out;
Jeff Layton80334262013-07-26 06:23:25 -04002320 }
Al Viro35759522013-09-08 13:41:33 -04002321 path->dentry = dentry;
Vasily Averin295dc392014-07-21 12:30:23 +04002322 path->mnt = nd->path.mnt;
David Howellsb18825a2013-09-12 19:22:53 +01002323 if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW))
Al Viro35759522013-09-08 13:41:33 -04002324 return 1;
Vasily Averin295dc392014-07-21 12:30:23 +04002325 mntget(path->mnt);
Al Viro35759522013-09-08 13:41:33 -04002326 follow_mount(path);
2327 error = 0;
2328out:
Jeff Layton80334262013-07-26 06:23:25 -04002329 terminate_walk(nd);
2330 return error;
2331}
2332
2333/**
Al Viro197df042013-09-08 14:03:27 -04002334 * path_mountpoint - look up a path to be umounted
Jeff Layton80334262013-07-26 06:23:25 -04002335 * @dfd: directory file descriptor to start walk from
2336 * @name: full pathname to walk
Randy Dunlap606d6fe2013-10-19 14:56:55 -07002337 * @path: pointer to container for result
Jeff Layton80334262013-07-26 06:23:25 -04002338 * @flags: lookup flags
Jeff Layton80334262013-07-26 06:23:25 -04002339 *
2340 * Look up the given name, but don't attempt to revalidate the last component.
Randy Dunlap606d6fe2013-10-19 14:56:55 -07002341 * Returns 0 and "path" will be valid on success; Returns error otherwise.
Jeff Layton80334262013-07-26 06:23:25 -04002342 */
2343static int
Al Viro197df042013-09-08 14:03:27 -04002344path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags)
Jeff Layton80334262013-07-26 06:23:25 -04002345{
Jeff Layton80334262013-07-26 06:23:25 -04002346 struct nameidata nd;
2347 int err;
2348
Al Viro980f3ea2014-11-20 14:20:24 -05002349 err = path_init(dfd, name, flags, &nd);
Jeff Layton80334262013-07-26 06:23:25 -04002350 if (unlikely(err))
Al Viro115cbfd2014-10-11 23:05:52 -04002351 goto out;
Jeff Layton80334262013-07-26 06:23:25 -04002352
Al Viro197df042013-09-08 14:03:27 -04002353 err = mountpoint_last(&nd, path);
Jeff Layton80334262013-07-26 06:23:25 -04002354 while (err > 0) {
2355 void *cookie;
2356 struct path link = *path;
2357 err = may_follow_link(&link, &nd);
2358 if (unlikely(err))
2359 break;
2360 nd.flags |= LOOKUP_PARENT;
2361 err = follow_link(&link, &nd, &cookie);
2362 if (err)
2363 break;
Al Viro197df042013-09-08 14:03:27 -04002364 err = mountpoint_last(&nd, path);
Jeff Layton80334262013-07-26 06:23:25 -04002365 put_link(&nd, &link, cookie);
2366 }
2367out:
Al Viro893b7772014-11-20 14:18:09 -05002368 path_cleanup(&nd);
Jeff Layton80334262013-07-26 06:23:25 -04002369 return err;
2370}
2371
Al Viro2d864652013-09-08 20:18:44 -04002372static int
2373filename_mountpoint(int dfd, struct filename *s, struct path *path,
2374 unsigned int flags)
2375{
Al Virocbaab2d2015-01-22 02:49:00 -05002376 int error;
2377 if (IS_ERR(s))
2378 return PTR_ERR(s);
2379 error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU);
Al Viro2d864652013-09-08 20:18:44 -04002380 if (unlikely(error == -ECHILD))
2381 error = path_mountpoint(dfd, s->name, path, flags);
2382 if (unlikely(error == -ESTALE))
2383 error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL);
2384 if (likely(!error))
2385 audit_inode(s, path->dentry, 0);
Al Virocbaab2d2015-01-22 02:49:00 -05002386 putname(s);
Al Viro2d864652013-09-08 20:18:44 -04002387 return error;
2388}
2389
Jeff Layton80334262013-07-26 06:23:25 -04002390/**
Al Viro197df042013-09-08 14:03:27 -04002391 * user_path_mountpoint_at - lookup a path from userland in order to umount it
Jeff Layton80334262013-07-26 06:23:25 -04002392 * @dfd: directory file descriptor
2393 * @name: pathname from userland
2394 * @flags: lookup flags
2395 * @path: pointer to container to hold result
2396 *
2397 * A umount is a special case for path walking. We're not actually interested
2398 * in the inode in this situation, and ESTALE errors can be a problem. We
2399 * simply want track down the dentry and vfsmount attached at the mountpoint
2400 * and avoid revalidating the last component.
2401 *
2402 * Returns 0 and populates "path" on success.
2403 */
2404int
Al Viro197df042013-09-08 14:03:27 -04002405user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
Jeff Layton80334262013-07-26 06:23:25 -04002406 struct path *path)
2407{
Al Virocbaab2d2015-01-22 02:49:00 -05002408 return filename_mountpoint(dfd, getname(name), path, flags);
Jeff Layton80334262013-07-26 06:23:25 -04002409}
2410
Al Viro2d864652013-09-08 20:18:44 -04002411int
2412kern_path_mountpoint(int dfd, const char *name, struct path *path,
2413 unsigned int flags)
2414{
Al Virocbaab2d2015-01-22 02:49:00 -05002415 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
Al Viro2d864652013-09-08 20:18:44 -04002416}
2417EXPORT_SYMBOL(kern_path_mountpoint);
2418
Miklos Szeredicbdf35b2014-10-24 00:14:36 +02002419int __check_sticky(struct inode *dir, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420{
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002421 kuid_t fsuid = current_fsuid();
David Howellsda9592e2008-11-14 10:39:05 +11002422
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002423 if (uid_eq(inode->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 return 0;
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002425 if (uid_eq(dir->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426 return 0;
Andy Lutomirski23adbe12014-06-10 12:45:42 -07002427 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428}
Miklos Szeredicbdf35b2014-10-24 00:14:36 +02002429EXPORT_SYMBOL(__check_sticky);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430
2431/*
2432 * Check whether we can remove a link victim from directory dir, check
2433 * whether the type of victim is right.
2434 * 1. We can't do it if dir is read-only (done in permission())
2435 * 2. We should have write and exec permissions on dir
2436 * 3. We can't remove anything from append-only dir
2437 * 4. We can't do anything with immutable dir (done in permission())
2438 * 5. If the sticky bit on dir is set we should either
2439 * a. be owner of dir, or
2440 * b. be owner of victim, or
2441 * c. have CAP_FOWNER capability
2442 * 6. If the victim is append-only or immutable we can't do antyhing with
2443 * links pointing to it.
2444 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2445 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2446 * 9. We can't remove a root or mountpoint.
2447 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2448 * nfs_async_unlink().
2449 */
David Howellsb18825a2013-09-12 19:22:53 +01002450static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451{
David Howellsb18825a2013-09-12 19:22:53 +01002452 struct inode *inode = victim->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453 int error;
2454
David Howellsb18825a2013-09-12 19:22:53 +01002455 if (d_is_negative(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456 return -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01002457 BUG_ON(!inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458
2459 BUG_ON(victim->d_parent->d_inode != dir);
Jeff Layton4fa6b5e2012-10-10 15:25:25 -04002460 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461
Al Virof419a2e2008-07-22 00:07:17 -04002462 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463 if (error)
2464 return error;
2465 if (IS_APPEND(dir))
2466 return -EPERM;
David Howellsb18825a2013-09-12 19:22:53 +01002467
2468 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2469 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470 return -EPERM;
2471 if (isdir) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002472 if (!d_is_dir(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002473 return -ENOTDIR;
2474 if (IS_ROOT(victim))
2475 return -EBUSY;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002476 } else if (d_is_dir(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477 return -EISDIR;
2478 if (IS_DEADDIR(dir))
2479 return -ENOENT;
2480 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2481 return -EBUSY;
2482 return 0;
2483}
2484
2485/* Check whether we can create an object with dentry child in directory
2486 * dir.
2487 * 1. We can't do it if child already exists (open has special treatment for
2488 * this case, but since we are inlined it's OK)
2489 * 2. We can't do it if dir is read-only (done in permission())
2490 * 3. We should have write and exec permissions on dir
2491 * 4. We can't do it if dir is immutable (done in permission())
2492 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02002493static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494{
Jeff Layton14e972b2013-05-08 10:25:58 -04002495 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496 if (child->d_inode)
2497 return -EEXIST;
2498 if (IS_DEADDIR(dir))
2499 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04002500 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002501}
2502
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503/*
2504 * p1 and p2 should be directories on the same fs.
2505 */
2506struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2507{
2508 struct dentry *p;
2509
2510 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07002511 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002512 return NULL;
2513 }
2514
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002515 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002517 p = d_ancestor(p2, p1);
2518 if (p) {
2519 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2520 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2521 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522 }
2523
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002524 p = d_ancestor(p1, p2);
2525 if (p) {
2526 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2527 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2528 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529 }
2530
Ingo Molnarf2eace22006-07-03 00:25:05 -07002531 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Miklos Szeredid1b72cc2014-10-27 15:42:01 +01002532 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533 return NULL;
2534}
Al Viro4d359502014-03-14 12:20:17 -04002535EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536
2537void unlock_rename(struct dentry *p1, struct dentry *p2)
2538{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002539 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002541 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002542 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543 }
2544}
Al Viro4d359502014-03-14 12:20:17 -04002545EXPORT_SYMBOL(unlock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546
Al Viro4acdaf22011-07-26 01:42:34 -04002547int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viro312b63f2012-06-10 18:09:36 -04002548 bool want_excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002550 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 if (error)
2552 return error;
2553
Al Viroacfa4382008-12-04 10:06:33 -05002554 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 return -EACCES; /* shouldn't it be ENOSYS? */
2556 mode &= S_IALLUGO;
2557 mode |= S_IFREG;
2558 error = security_inode_create(dir, dentry, mode);
2559 if (error)
2560 return error;
Al Viro312b63f2012-06-10 18:09:36 -04002561 error = dir->i_op->create(dir, dentry, mode, want_excl);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002562 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002563 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564 return error;
2565}
Al Viro4d359502014-03-14 12:20:17 -04002566EXPORT_SYMBOL(vfs_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567
Al Viro73d049a2011-03-11 12:08:24 -05002568static int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002570 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571 struct inode *inode = dentry->d_inode;
2572 int error;
2573
Al Virobcda7652011-03-13 16:42:14 -04002574 /* O_PATH? */
2575 if (!acc_mode)
2576 return 0;
2577
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578 if (!inode)
2579 return -ENOENT;
2580
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002581 switch (inode->i_mode & S_IFMT) {
2582 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002584 case S_IFDIR:
2585 if (acc_mode & MAY_WRITE)
2586 return -EISDIR;
2587 break;
2588 case S_IFBLK:
2589 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002590 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002592 /*FALLTHRU*/
2593 case S_IFIFO:
2594 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002596 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002597 }
Dave Hansenb41572e2007-10-16 23:31:14 -07002598
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002599 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07002600 if (error)
2601 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05002602
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603 /*
2604 * An append-only file must be opened in append mode for writing.
2605 */
2606 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05002607 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05002608 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05002610 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611 }
2612
2613 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002614 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05002615 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616
J. Bruce Fieldsf3c7691e2011-09-21 10:58:13 -04002617 return 0;
Al Viro7715b522009-12-16 03:54:00 -05002618}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619
Jeff Laytone1181ee2010-12-07 16:19:50 -05002620static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002621{
Jeff Laytone1181ee2010-12-07 16:19:50 -05002622 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002623 struct inode *inode = path->dentry->d_inode;
2624 int error = get_write_access(inode);
2625 if (error)
2626 return error;
2627 /*
2628 * Refuse to truncate files with mandatory locks held on them.
2629 */
Jeff Laytond7a06982014-03-10 09:54:15 -04002630 error = locks_verify_locked(filp);
Al Viro7715b522009-12-16 03:54:00 -05002631 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002632 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002633 if (!error) {
2634 error = do_truncate(path->dentry, 0,
2635 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002636 filp);
Al Viro7715b522009-12-16 03:54:00 -05002637 }
2638 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002639 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002640}
2641
Dave Hansend57999e2008-02-15 14:37:27 -08002642static inline int open_to_namei_flags(int flag)
2643{
Al Viro8a5e9292011-06-25 19:15:54 -04002644 if ((flag & O_ACCMODE) == 3)
2645 flag--;
Dave Hansend57999e2008-02-15 14:37:27 -08002646 return flag;
2647}
2648
Miklos Szeredid18e9002012-06-05 15:10:17 +02002649static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2650{
2651 int error = security_path_mknod(dir, dentry, mode, 0);
2652 if (error)
2653 return error;
2654
2655 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2656 if (error)
2657 return error;
2658
2659 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2660}
2661
David Howells1acf0af2012-06-14 16:13:46 +01002662/*
2663 * Attempt to atomically look up, create and open a file from a negative
2664 * dentry.
2665 *
2666 * Returns 0 if successful. The file will have been created and attached to
2667 * @file by the filesystem calling finish_open().
2668 *
2669 * Returns 1 if the file was looked up only or didn't need creating. The
2670 * caller will need to perform the open themselves. @path will have been
2671 * updated to point to the new dentry. This may be negative.
2672 *
2673 * Returns an error code otherwise.
2674 */
Al Viro2675a4e2012-06-22 12:41:10 +04002675static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2676 struct path *path, struct file *file,
2677 const struct open_flags *op,
Al Viro64894cf2012-07-31 00:53:35 +04002678 bool got_write, bool need_lookup,
Al Viro2675a4e2012-06-22 12:41:10 +04002679 int *opened)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002680{
2681 struct inode *dir = nd->path.dentry->d_inode;
2682 unsigned open_flag = open_to_namei_flags(op->open_flag);
2683 umode_t mode;
2684 int error;
2685 int acc_mode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002686 int create_error = 0;
2687 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
Miklos Szeredi116cc022013-09-16 14:52:05 +02002688 bool excl;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002689
2690 BUG_ON(dentry->d_inode);
2691
2692 /* Don't create child dentry for a dead directory. */
2693 if (unlikely(IS_DEADDIR(dir))) {
Al Viro2675a4e2012-06-22 12:41:10 +04002694 error = -ENOENT;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002695 goto out;
2696 }
2697
Miklos Szeredi62b259d2012-08-15 13:01:24 +02002698 mode = op->mode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002699 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2700 mode &= ~current_umask();
2701
Miklos Szeredi116cc022013-09-16 14:52:05 +02002702 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2703 if (excl)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002704 open_flag &= ~O_TRUNC;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002705
2706 /*
2707 * Checking write permission is tricky, bacuse we don't know if we are
2708 * going to actually need it: O_CREAT opens should work as long as the
2709 * file exists. But checking existence breaks atomicity. The trick is
2710 * to check access and if not granted clear O_CREAT from the flags.
2711 *
2712 * Another problem is returing the "right" error value (e.g. for an
2713 * O_EXCL open we want to return EEXIST not EROFS).
2714 */
Al Viro64894cf2012-07-31 00:53:35 +04002715 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2716 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2717 if (!(open_flag & O_CREAT)) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002718 /*
2719 * No O_CREATE -> atomicity not a requirement -> fall
2720 * back to lookup + open
2721 */
2722 goto no_open;
2723 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2724 /* Fall back and fail with the right error */
Al Viro64894cf2012-07-31 00:53:35 +04002725 create_error = -EROFS;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002726 goto no_open;
2727 } else {
2728 /* No side effects, safe to clear O_CREAT */
Al Viro64894cf2012-07-31 00:53:35 +04002729 create_error = -EROFS;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002730 open_flag &= ~O_CREAT;
2731 }
2732 }
2733
2734 if (open_flag & O_CREAT) {
Miklos Szeredi38227f72012-08-15 13:01:24 +02002735 error = may_o_create(&nd->path, dentry, mode);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002736 if (error) {
2737 create_error = error;
2738 if (open_flag & O_EXCL)
2739 goto no_open;
2740 open_flag &= ~O_CREAT;
2741 }
2742 }
2743
2744 if (nd->flags & LOOKUP_DIRECTORY)
2745 open_flag |= O_DIRECTORY;
2746
Al Viro30d90492012-06-22 12:40:19 +04002747 file->f_path.dentry = DENTRY_NOT_SET;
2748 file->f_path.mnt = nd->path.mnt;
2749 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
Al Viro47237682012-06-10 05:01:45 -04002750 opened);
Al Virod9585272012-06-22 12:39:14 +04002751 if (error < 0) {
Al Virod9585272012-06-22 12:39:14 +04002752 if (create_error && error == -ENOENT)
2753 error = create_error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002754 goto out;
2755 }
2756
Al Virod9585272012-06-22 12:39:14 +04002757 if (error) { /* returned 1, that is */
Al Viro30d90492012-06-22 12:40:19 +04002758 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
Al Viro2675a4e2012-06-22 12:41:10 +04002759 error = -EIO;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002760 goto out;
2761 }
Al Viro30d90492012-06-22 12:40:19 +04002762 if (file->f_path.dentry) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002763 dput(dentry);
Al Viro30d90492012-06-22 12:40:19 +04002764 dentry = file->f_path.dentry;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002765 }
Al Viro03da6332013-09-16 19:22:33 -04002766 if (*opened & FILE_CREATED)
2767 fsnotify_create(dir, dentry);
2768 if (!dentry->d_inode) {
2769 WARN_ON(*opened & FILE_CREATED);
2770 if (create_error) {
2771 error = create_error;
2772 goto out;
2773 }
2774 } else {
2775 if (excl && !(*opened & FILE_CREATED)) {
2776 error = -EEXIST;
2777 goto out;
2778 }
Sage Weil62b2ce92012-08-15 13:30:12 -07002779 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02002780 goto looked_up;
2781 }
2782
2783 /*
2784 * We didn't have the inode before the open, so check open permission
2785 * here.
2786 */
Al Viro03da6332013-09-16 19:22:33 -04002787 acc_mode = op->acc_mode;
2788 if (*opened & FILE_CREATED) {
2789 WARN_ON(!(open_flag & O_CREAT));
2790 fsnotify_create(dir, dentry);
2791 acc_mode = MAY_OPEN;
2792 }
Al Viro2675a4e2012-06-22 12:41:10 +04002793 error = may_open(&file->f_path, acc_mode, open_flag);
2794 if (error)
2795 fput(file);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002796
2797out:
2798 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04002799 return error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002800
Miklos Szeredid18e9002012-06-05 15:10:17 +02002801no_open:
2802 if (need_lookup) {
Al Viro72bd8662012-06-10 17:17:17 -04002803 dentry = lookup_real(dir, dentry, nd->flags);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002804 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002805 return PTR_ERR(dentry);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002806
2807 if (create_error) {
2808 int open_flag = op->open_flag;
2809
Al Viro2675a4e2012-06-22 12:41:10 +04002810 error = create_error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002811 if ((open_flag & O_EXCL)) {
2812 if (!dentry->d_inode)
2813 goto out;
2814 } else if (!dentry->d_inode) {
2815 goto out;
2816 } else if ((open_flag & O_TRUNC) &&
2817 S_ISREG(dentry->d_inode->i_mode)) {
2818 goto out;
2819 }
2820 /* will fail later, go on to get the right error */
2821 }
2822 }
2823looked_up:
2824 path->dentry = dentry;
2825 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04002826 return 1;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002827}
2828
Nick Piggin31e6b012011-01-07 17:49:52 +11002829/*
David Howells1acf0af2012-06-14 16:13:46 +01002830 * Look up and maybe create and open the last component.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002831 *
2832 * Must be called with i_mutex held on parent.
2833 *
David Howells1acf0af2012-06-14 16:13:46 +01002834 * Returns 0 if the file was successfully atomically created (if necessary) and
2835 * opened. In this case the file will be returned attached to @file.
2836 *
2837 * Returns 1 if the file was not completely opened at this time, though lookups
2838 * and creations will have been performed and the dentry returned in @path will
2839 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2840 * specified then a negative dentry may be returned.
2841 *
2842 * An error code is returned otherwise.
2843 *
2844 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2845 * cleared otherwise prior to returning.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002846 */
Al Viro2675a4e2012-06-22 12:41:10 +04002847static int lookup_open(struct nameidata *nd, struct path *path,
2848 struct file *file,
2849 const struct open_flags *op,
Al Viro64894cf2012-07-31 00:53:35 +04002850 bool got_write, int *opened)
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002851{
2852 struct dentry *dir = nd->path.dentry;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002853 struct inode *dir_inode = dir->d_inode;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002854 struct dentry *dentry;
2855 int error;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002856 bool need_lookup;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002857
Al Viro47237682012-06-10 05:01:45 -04002858 *opened &= ~FILE_CREATED;
Al Viro201f9562012-06-22 12:42:10 +04002859 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002860 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002861 return PTR_ERR(dentry);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002862
Miklos Szeredid18e9002012-06-05 15:10:17 +02002863 /* Cached positive dentry: will open in f_op->open */
2864 if (!need_lookup && dentry->d_inode)
2865 goto out_no_open;
2866
2867 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
Al Viro64894cf2012-07-31 00:53:35 +04002868 return atomic_open(nd, dentry, path, file, op, got_write,
Al Viro47237682012-06-10 05:01:45 -04002869 need_lookup, opened);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002870 }
2871
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002872 if (need_lookup) {
2873 BUG_ON(dentry->d_inode);
2874
Al Viro72bd8662012-06-10 17:17:17 -04002875 dentry = lookup_real(dir_inode, dentry, nd->flags);
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002876 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002877 return PTR_ERR(dentry);
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002878 }
2879
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002880 /* Negative dentry, just create the file */
2881 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2882 umode_t mode = op->mode;
2883 if (!IS_POSIXACL(dir->d_inode))
2884 mode &= ~current_umask();
2885 /*
2886 * This write is needed to ensure that a
2887 * rw->ro transition does not occur between
2888 * the time when the file is created and when
2889 * a permanent write count is taken through
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002890 * the 'struct file' in finish_open().
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002891 */
Al Viro64894cf2012-07-31 00:53:35 +04002892 if (!got_write) {
2893 error = -EROFS;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002894 goto out_dput;
Al Viro64894cf2012-07-31 00:53:35 +04002895 }
Al Viro47237682012-06-10 05:01:45 -04002896 *opened |= FILE_CREATED;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002897 error = security_path_mknod(&nd->path, dentry, mode, 0);
2898 if (error)
2899 goto out_dput;
Al Viro312b63f2012-06-10 18:09:36 -04002900 error = vfs_create(dir->d_inode, dentry, mode,
2901 nd->flags & LOOKUP_EXCL);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002902 if (error)
2903 goto out_dput;
2904 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02002905out_no_open:
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002906 path->dentry = dentry;
2907 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04002908 return 1;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002909
2910out_dput:
2911 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04002912 return error;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002913}
2914
2915/*
Al Virofe2d35f2011-03-05 22:58:25 -05002916 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11002917 */
Al Viro2675a4e2012-06-22 12:41:10 +04002918static int do_last(struct nameidata *nd, struct path *path,
2919 struct file *file, const struct open_flags *op,
Jeff Layton669abf42012-10-10 16:43:10 -04002920 int *opened, struct filename *name)
Al Virofb1cc552009-12-24 01:58:28 -05002921{
Al Viroa1e28032009-12-24 02:12:06 -05002922 struct dentry *dir = nd->path.dentry;
Al Viroca344a892011-03-09 00:36:45 -05002923 int open_flag = op->open_flag;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002924 bool will_truncate = (open_flag & O_TRUNC) != 0;
Al Viro64894cf2012-07-31 00:53:35 +04002925 bool got_write = false;
Al Virobcda7652011-03-13 16:42:14 -04002926 int acc_mode = op->acc_mode;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02002927 struct inode *inode;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002928 bool symlink_ok = false;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002929 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2930 bool retried = false;
Al Viro16c2cd72011-02-22 15:50:10 -05002931 int error;
Al Virofb1cc552009-12-24 01:58:28 -05002932
Al Viroc3e380b2011-02-23 13:39:45 -05002933 nd->flags &= ~LOOKUP_PARENT;
2934 nd->flags |= op->intent;
2935
Al Virobc77daa2013-06-06 09:12:33 -04002936 if (nd->last_type != LAST_NORM) {
Al Virofe2d35f2011-03-05 22:58:25 -05002937 error = handle_dots(nd, nd->last_type);
2938 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002939 return error;
Miklos Szeredie83db162012-06-05 15:10:29 +02002940 goto finish_open;
Al Viro1f36f772009-12-26 10:56:19 -05002941 }
Al Viro67ee3ad2009-12-26 07:01:01 -05002942
Al Viroca344a892011-03-09 00:36:45 -05002943 if (!(open_flag & O_CREAT)) {
Al Virofe2d35f2011-03-05 22:58:25 -05002944 if (nd->last.name[nd->last.len])
2945 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
Al Virobcda7652011-03-13 16:42:14 -04002946 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002947 symlink_ok = true;
Al Virofe2d35f2011-03-05 22:58:25 -05002948 /* we _can_ be in RCU mode here */
Al Viroe97cdc82013-01-24 18:16:00 -05002949 error = lookup_fast(nd, path, &inode);
Miklos Szeredi71574862012-06-05 15:10:14 +02002950 if (likely(!error))
2951 goto finish_lookup;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02002952
Miklos Szeredi71574862012-06-05 15:10:14 +02002953 if (error < 0)
Al Viro2675a4e2012-06-22 12:41:10 +04002954 goto out;
Miklos Szeredi37d7fff2012-06-05 15:10:12 +02002955
Miklos Szeredi71574862012-06-05 15:10:14 +02002956 BUG_ON(nd->inode != dir->d_inode);
Miklos Szeredib6183df2012-06-05 15:10:13 +02002957 } else {
2958 /* create side of things */
2959 /*
2960 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2961 * has been cleared when we got to the last component we are
2962 * about to look up
2963 */
2964 error = complete_walk(nd);
2965 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002966 return error;
Miklos Szeredib6183df2012-06-05 15:10:13 +02002967
Jeff Layton33e22082013-04-12 15:16:32 -04002968 audit_inode(name, dir, LOOKUP_PARENT);
Miklos Szeredib6183df2012-06-05 15:10:13 +02002969 error = -EISDIR;
2970 /* trailing slashes? */
2971 if (nd->last.name[nd->last.len])
Al Viro2675a4e2012-06-22 12:41:10 +04002972 goto out;
Al Virofe2d35f2011-03-05 22:58:25 -05002973 }
2974
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002975retry_lookup:
Al Viro64894cf2012-07-31 00:53:35 +04002976 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2977 error = mnt_want_write(nd->path.mnt);
2978 if (!error)
2979 got_write = true;
2980 /*
2981 * do _not_ fail yet - we might not need that or fail with
2982 * a different error; let lookup_open() decide; we'll be
2983 * dropping this one anyway.
2984 */
2985 }
Al Viroa1e28032009-12-24 02:12:06 -05002986 mutex_lock(&dir->d_inode->i_mutex);
Al Viro64894cf2012-07-31 00:53:35 +04002987 error = lookup_open(nd, path, file, op, got_write, opened);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002988 mutex_unlock(&dir->d_inode->i_mutex);
Al Viroa1e28032009-12-24 02:12:06 -05002989
Al Viro2675a4e2012-06-22 12:41:10 +04002990 if (error <= 0) {
2991 if (error)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002992 goto out;
2993
Al Viro47237682012-06-10 05:01:45 -04002994 if ((*opened & FILE_CREATED) ||
Al Viro496ad9a2013-01-23 17:07:38 -05002995 !S_ISREG(file_inode(file)->i_mode))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002996 will_truncate = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002997
Jeff Laytonadb5c242012-10-10 16:43:13 -04002998 audit_inode(name, file->f_path.dentry, 0);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002999 goto opened;
3000 }
Al Virofb1cc552009-12-24 01:58:28 -05003001
Al Viro47237682012-06-10 05:01:45 -04003002 if (*opened & FILE_CREATED) {
Al Viro9b44f1b2011-03-09 00:17:27 -05003003 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05003004 open_flag &= ~O_TRUNC;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003005 will_truncate = false;
Al Virobcda7652011-03-13 16:42:14 -04003006 acc_mode = MAY_OPEN;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003007 path_to_nameidata(path, nd);
Miklos Szeredie83db162012-06-05 15:10:29 +02003008 goto finish_open_created;
Al Virofb1cc552009-12-24 01:58:28 -05003009 }
3010
3011 /*
Jeff Layton3134f372012-07-25 10:19:47 -04003012 * create/update audit record if it already exists.
Al Virofb1cc552009-12-24 01:58:28 -05003013 */
David Howellsb18825a2013-09-12 19:22:53 +01003014 if (d_is_positive(path->dentry))
Jeff Laytonadb5c242012-10-10 16:43:13 -04003015 audit_inode(name, path->dentry, 0);
Al Virofb1cc552009-12-24 01:58:28 -05003016
Miklos Szeredid18e9002012-06-05 15:10:17 +02003017 /*
3018 * If atomic_open() acquired write access it is dropped now due to
3019 * possible mount and symlink following (this might be optimized away if
3020 * necessary...)
3021 */
Al Viro64894cf2012-07-31 00:53:35 +04003022 if (got_write) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02003023 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04003024 got_write = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003025 }
3026
Al Virofb1cc552009-12-24 01:58:28 -05003027 error = -EEXIST;
Al Virof8310c52012-07-30 11:50:30 +04003028 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
Al Virofb1cc552009-12-24 01:58:28 -05003029 goto exit_dput;
3030
David Howells9875cf82011-01-14 18:45:21 +00003031 error = follow_managed(path, nd->flags);
3032 if (error < 0)
3033 goto exit_dput;
Al Virofb1cc552009-12-24 01:58:28 -05003034
Al Viroa3fbbde2011-11-07 21:21:26 +00003035 if (error)
3036 nd->flags |= LOOKUP_JUMPED;
3037
Miklos Szeredidecf3402012-05-21 17:30:08 +02003038 BUG_ON(nd->flags & LOOKUP_RCU);
3039 inode = path->dentry->d_inode;
Miklos Szeredi5f5daac2012-05-21 17:30:14 +02003040finish_lookup:
3041 /* we _can_ be in RCU mode here */
Al Virofb1cc552009-12-24 01:58:28 -05003042 error = -ENOENT;
Al Viro22213312014-04-19 12:30:58 -04003043 if (!inode || d_is_negative(path->dentry)) {
Miklos Szeredi54c33e72012-05-21 17:30:10 +02003044 path_to_nameidata(path, nd);
Al Viro2675a4e2012-06-22 12:41:10 +04003045 goto out;
Miklos Szeredi54c33e72012-05-21 17:30:10 +02003046 }
Al Viro9e67f362009-12-26 07:04:50 -05003047
David Howellsb18825a2013-09-12 19:22:53 +01003048 if (should_follow_link(path->dentry, !symlink_ok)) {
Miklos Szeredid45ea862012-05-21 17:30:09 +02003049 if (nd->flags & LOOKUP_RCU) {
3050 if (unlikely(unlazy_walk(nd, path->dentry))) {
3051 error = -ECHILD;
Al Viro2675a4e2012-06-22 12:41:10 +04003052 goto out;
Miklos Szeredid45ea862012-05-21 17:30:09 +02003053 }
3054 }
3055 BUG_ON(inode != path->dentry->d_inode);
Al Viro2675a4e2012-06-22 12:41:10 +04003056 return 1;
Miklos Szeredid45ea862012-05-21 17:30:09 +02003057 }
Al Virofb1cc552009-12-24 01:58:28 -05003058
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003059 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
3060 path_to_nameidata(path, nd);
3061 } else {
3062 save_parent.dentry = nd->path.dentry;
3063 save_parent.mnt = mntget(path->mnt);
3064 nd->path.dentry = path->dentry;
3065
3066 }
Miklos Szeredidecf3402012-05-21 17:30:08 +02003067 nd->inode = inode;
Al Viroa3fbbde2011-11-07 21:21:26 +00003068 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
Al Virobc77daa2013-06-06 09:12:33 -04003069finish_open:
Al Viroa3fbbde2011-11-07 21:21:26 +00003070 error = complete_walk(nd);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003071 if (error) {
3072 path_put(&save_parent);
Al Viro2675a4e2012-06-22 12:41:10 +04003073 return error;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003074 }
Al Virobc77daa2013-06-06 09:12:33 -04003075 audit_inode(name, nd->path.dentry, 0);
Al Virofb1cc552009-12-24 01:58:28 -05003076 error = -EISDIR;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003077 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04003078 goto out;
Miklos Szerediaf2f5542012-05-21 17:30:11 +02003079 error = -ENOTDIR;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003080 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04003081 goto out;
Al Viro6c0d46c2011-03-09 00:59:59 -05003082 if (!S_ISREG(nd->inode->i_mode))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003083 will_truncate = false;
Al Viro6c0d46c2011-03-09 00:59:59 -05003084
Al Viro0f9d1a12011-03-09 00:13:14 -05003085 if (will_truncate) {
3086 error = mnt_want_write(nd->path.mnt);
3087 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003088 goto out;
Al Viro64894cf2012-07-31 00:53:35 +04003089 got_write = true;
Al Viro0f9d1a12011-03-09 00:13:14 -05003090 }
Miklos Szeredie83db162012-06-05 15:10:29 +02003091finish_open_created:
Al Virobcda7652011-03-13 16:42:14 -04003092 error = may_open(&nd->path, acc_mode, open_flag);
Al Viroca344a892011-03-09 00:36:45 -05003093 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003094 goto out;
Miklos Szeredi4aa7c632014-10-24 00:14:35 +02003095
3096 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3097 error = vfs_open(&nd->path, file, current_cred());
3098 if (!error) {
3099 *opened |= FILE_OPENED;
3100 } else {
Al Viro30d90492012-06-22 12:40:19 +04003101 if (error == -EOPENSTALE)
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003102 goto stale_open;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003103 goto out;
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003104 }
Miklos Szeredia8277b92012-06-05 15:10:32 +02003105opened:
Al Viro2675a4e2012-06-22 12:41:10 +04003106 error = open_check_o_direct(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003107 if (error)
3108 goto exit_fput;
Dmitry Kasatkin3034a142014-06-27 18:15:44 +03003109 error = ima_file_check(file, op->acc_mode, *opened);
Miklos Szerediaa4caad2012-06-05 15:10:28 +02003110 if (error)
3111 goto exit_fput;
3112
3113 if (will_truncate) {
Al Viro2675a4e2012-06-22 12:41:10 +04003114 error = handle_truncate(file);
Miklos Szerediaa4caad2012-06-05 15:10:28 +02003115 if (error)
3116 goto exit_fput;
Al Viro0f9d1a12011-03-09 00:13:14 -05003117 }
Al Viroca344a892011-03-09 00:36:45 -05003118out:
Al Viro64894cf2012-07-31 00:53:35 +04003119 if (got_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05003120 mnt_drop_write(nd->path.mnt);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003121 path_put(&save_parent);
Miklos Szeredie276ae62012-05-21 17:30:06 +02003122 terminate_walk(nd);
Al Viro2675a4e2012-06-22 12:41:10 +04003123 return error;
Al Virofb1cc552009-12-24 01:58:28 -05003124
Al Virofb1cc552009-12-24 01:58:28 -05003125exit_dput:
3126 path_put_conditional(path, nd);
Al Viroca344a892011-03-09 00:36:45 -05003127 goto out;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003128exit_fput:
Al Viro2675a4e2012-06-22 12:41:10 +04003129 fput(file);
3130 goto out;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003131
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003132stale_open:
3133 /* If no saved parent or already retried then can't retry */
3134 if (!save_parent.dentry || retried)
3135 goto out;
3136
3137 BUG_ON(save_parent.dentry != dir);
3138 path_put(&nd->path);
3139 nd->path = save_parent;
3140 nd->inode = dir->d_inode;
3141 save_parent.mnt = NULL;
3142 save_parent.dentry = NULL;
Al Viro64894cf2012-07-31 00:53:35 +04003143 if (got_write) {
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003144 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04003145 got_write = false;
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003146 }
3147 retried = true;
3148 goto retry_lookup;
Al Virofb1cc552009-12-24 01:58:28 -05003149}
3150
Al Viro60545d02013-06-07 01:20:27 -04003151static int do_tmpfile(int dfd, struct filename *pathname,
3152 struct nameidata *nd, int flags,
3153 const struct open_flags *op,
3154 struct file *file, int *opened)
3155{
3156 static const struct qstr name = QSTR_INIT("/", 1);
3157 struct dentry *dentry, *child;
3158 struct inode *dir;
3159 int error = path_lookupat(dfd, pathname->name,
3160 flags | LOOKUP_DIRECTORY, nd);
3161 if (unlikely(error))
3162 return error;
3163 error = mnt_want_write(nd->path.mnt);
3164 if (unlikely(error))
3165 goto out;
3166 /* we want directory to be writable */
3167 error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3168 if (error)
3169 goto out2;
3170 dentry = nd->path.dentry;
3171 dir = dentry->d_inode;
3172 if (!dir->i_op->tmpfile) {
3173 error = -EOPNOTSUPP;
3174 goto out2;
3175 }
3176 child = d_alloc(dentry, &name);
3177 if (unlikely(!child)) {
3178 error = -ENOMEM;
3179 goto out2;
3180 }
3181 nd->flags &= ~LOOKUP_DIRECTORY;
3182 nd->flags |= op->intent;
3183 dput(nd->path.dentry);
3184 nd->path.dentry = child;
3185 error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3186 if (error)
3187 goto out2;
3188 audit_inode(pathname, nd->path.dentry, 0);
Eric Rannaud69a91c22014-10-30 01:51:01 -07003189 /* Don't check for other permissions, the inode was just created */
3190 error = may_open(&nd->path, MAY_OPEN, op->open_flag);
Al Viro60545d02013-06-07 01:20:27 -04003191 if (error)
3192 goto out2;
3193 file->f_path.mnt = nd->path.mnt;
3194 error = finish_open(file, nd->path.dentry, NULL, opened);
3195 if (error)
3196 goto out2;
3197 error = open_check_o_direct(file);
Al Virof4e0c302013-06-11 08:34:36 +04003198 if (error) {
Al Viro60545d02013-06-07 01:20:27 -04003199 fput(file);
Al Virof4e0c302013-06-11 08:34:36 +04003200 } else if (!(op->open_flag & O_EXCL)) {
3201 struct inode *inode = file_inode(file);
3202 spin_lock(&inode->i_lock);
3203 inode->i_state |= I_LINKABLE;
3204 spin_unlock(&inode->i_lock);
3205 }
Al Viro60545d02013-06-07 01:20:27 -04003206out2:
3207 mnt_drop_write(nd->path.mnt);
3208out:
3209 path_put(&nd->path);
3210 return error;
3211}
3212
Jeff Layton669abf42012-10-10 16:43:10 -04003213static struct file *path_openat(int dfd, struct filename *pathname,
Al Viro73d049a2011-03-11 12:08:24 -05003214 struct nameidata *nd, const struct open_flags *op, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003215{
Al Viro30d90492012-06-22 12:40:19 +04003216 struct file *file;
Al Viro9850c052010-01-13 15:01:15 -05003217 struct path path;
Al Viro47237682012-06-10 05:01:45 -04003218 int opened = 0;
Al Viro13aab422011-02-23 17:54:08 -05003219 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11003220
Al Viro30d90492012-06-22 12:40:19 +04003221 file = get_empty_filp();
Al Viro1afc99b2013-02-14 20:41:04 -05003222 if (IS_ERR(file))
3223 return file;
Nick Piggin31e6b012011-01-07 17:49:52 +11003224
Al Viro30d90492012-06-22 12:40:19 +04003225 file->f_flags = op->open_flag;
Nick Piggin31e6b012011-01-07 17:49:52 +11003226
Al Virobb458c62013-07-13 13:26:37 +04003227 if (unlikely(file->f_flags & __O_TMPFILE)) {
Al Viro60545d02013-06-07 01:20:27 -04003228 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3229 goto out;
3230 }
3231
Al Viro980f3ea2014-11-20 14:20:24 -05003232 error = path_init(dfd, pathname->name, flags, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11003233 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04003234 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003235
Al Viro2675a4e2012-06-22 12:41:10 +04003236 error = do_last(nd, &path, file, op, &opened, pathname);
3237 while (unlikely(error > 0)) { /* trailing symlink */
Nick Piggin7b9337a2011-01-14 08:42:43 +00003238 struct path link = path;
Al Virodef4af32009-12-26 08:37:05 -05003239 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04003240 if (!(nd->flags & LOOKUP_FOLLOW)) {
Al Viro73d049a2011-03-11 12:08:24 -05003241 path_put_conditional(&path, nd);
3242 path_put(&nd->path);
Al Viro2675a4e2012-06-22 12:41:10 +04003243 error = -ELOOP;
Al Viro40b39132011-03-09 16:22:18 -05003244 break;
3245 }
Kees Cook800179c2012-07-25 17:29:07 -07003246 error = may_follow_link(&link, nd);
3247 if (unlikely(error))
3248 break;
Al Viro73d049a2011-03-11 12:08:24 -05003249 nd->flags |= LOOKUP_PARENT;
3250 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
Al Viro574197e2011-03-14 22:20:34 -04003251 error = follow_link(&link, nd, &cookie);
Al Viroc3e380b2011-02-23 13:39:45 -05003252 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04003253 break;
3254 error = do_last(nd, &path, file, op, &opened, pathname);
Al Viro574197e2011-03-14 22:20:34 -04003255 put_link(nd, &link, cookie);
Al Viro806b6812009-12-26 07:16:40 -05003256 }
Al Viro10fa8e62009-12-26 07:09:49 -05003257out:
Al Viro893b7772014-11-20 14:18:09 -05003258 path_cleanup(nd);
Al Viro2675a4e2012-06-22 12:41:10 +04003259 if (!(opened & FILE_OPENED)) {
3260 BUG_ON(!error);
Al Viro30d90492012-06-22 12:40:19 +04003261 put_filp(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003262 }
Al Viro2675a4e2012-06-22 12:41:10 +04003263 if (unlikely(error)) {
3264 if (error == -EOPENSTALE) {
3265 if (flags & LOOKUP_RCU)
3266 error = -ECHILD;
3267 else
3268 error = -ESTALE;
3269 }
3270 file = ERR_PTR(error);
3271 }
3272 return file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003273}
3274
Jeff Layton669abf42012-10-10 16:43:10 -04003275struct file *do_filp_open(int dfd, struct filename *pathname,
Al Virof9652e12013-06-11 08:23:01 +04003276 const struct open_flags *op)
Al Viro13aab422011-02-23 17:54:08 -05003277{
Al Viro73d049a2011-03-11 12:08:24 -05003278 struct nameidata nd;
Al Virof9652e12013-06-11 08:23:01 +04003279 int flags = op->lookup_flags;
Al Viro13aab422011-02-23 17:54:08 -05003280 struct file *filp;
3281
Al Viro73d049a2011-03-11 12:08:24 -05003282 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
Al Viro13aab422011-02-23 17:54:08 -05003283 if (unlikely(filp == ERR_PTR(-ECHILD)))
Al Viro73d049a2011-03-11 12:08:24 -05003284 filp = path_openat(dfd, pathname, &nd, op, flags);
Al Viro13aab422011-02-23 17:54:08 -05003285 if (unlikely(filp == ERR_PTR(-ESTALE)))
Al Viro73d049a2011-03-11 12:08:24 -05003286 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
Al Viro13aab422011-02-23 17:54:08 -05003287 return filp;
3288}
3289
Al Viro73d049a2011-03-11 12:08:24 -05003290struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
Al Virof9652e12013-06-11 08:23:01 +04003291 const char *name, const struct open_flags *op)
Al Viro73d049a2011-03-11 12:08:24 -05003292{
3293 struct nameidata nd;
3294 struct file *file;
Paul Moore51689102015-01-22 00:00:03 -05003295 struct filename *filename;
Al Virof9652e12013-06-11 08:23:01 +04003296 int flags = op->lookup_flags | LOOKUP_ROOT;
Al Viro73d049a2011-03-11 12:08:24 -05003297
3298 nd.root.mnt = mnt;
3299 nd.root.dentry = dentry;
3300
David Howellsb18825a2013-09-12 19:22:53 +01003301 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
Al Viro73d049a2011-03-11 12:08:24 -05003302 return ERR_PTR(-ELOOP);
3303
Paul Moore51689102015-01-22 00:00:03 -05003304 filename = getname_kernel(name);
3305 if (unlikely(IS_ERR(filename)))
3306 return ERR_CAST(filename);
3307
3308 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_RCU);
Al Viro73d049a2011-03-11 12:08:24 -05003309 if (unlikely(file == ERR_PTR(-ECHILD)))
Paul Moore51689102015-01-22 00:00:03 -05003310 file = path_openat(-1, filename, &nd, op, flags);
Al Viro73d049a2011-03-11 12:08:24 -05003311 if (unlikely(file == ERR_PTR(-ESTALE)))
Paul Moore51689102015-01-22 00:00:03 -05003312 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_REVAL);
3313 putname(filename);
Al Viro73d049a2011-03-11 12:08:24 -05003314 return file;
3315}
3316
Al Virofa14a0b2015-01-22 02:16:49 -05003317static struct dentry *filename_create(int dfd, struct filename *name,
Jeff Layton1ac12b42012-12-11 12:10:06 -05003318 struct path *path, unsigned int lookup_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003319{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003320 struct dentry *dentry = ERR_PTR(-EEXIST);
Al Viroed75e952011-06-27 16:53:43 -04003321 struct nameidata nd;
Jan Karac30dabf2012-06-12 16:20:30 +02003322 int err2;
Jeff Layton1ac12b42012-12-11 12:10:06 -05003323 int error;
3324 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3325
3326 /*
3327 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3328 * other flags passed in are ignored!
3329 */
3330 lookup_flags &= LOOKUP_REVAL;
3331
Al Virofa14a0b2015-01-22 02:16:49 -05003332 error = filename_lookup(dfd, name, LOOKUP_PARENT|lookup_flags, &nd);
Al Viroed75e952011-06-27 16:53:43 -04003333 if (error)
3334 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003335
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003336 /*
3337 * Yucky last component or no last component at all?
3338 * (foo/., foo/.., /////)
3339 */
Al Viroed75e952011-06-27 16:53:43 -04003340 if (nd.last_type != LAST_NORM)
3341 goto out;
3342 nd.flags &= ~LOOKUP_PARENT;
3343 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003344
Jan Karac30dabf2012-06-12 16:20:30 +02003345 /* don't fail immediately if it's r/o, at least try to report other errors */
3346 err2 = mnt_want_write(nd.path.mnt);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003347 /*
3348 * Do the final lookup.
3349 */
Al Viroed75e952011-06-27 16:53:43 -04003350 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3351 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003352 if (IS_ERR(dentry))
Al Viroa8104a92012-07-20 02:25:00 +04003353 goto unlock;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003354
Al Viroa8104a92012-07-20 02:25:00 +04003355 error = -EEXIST;
David Howellsb18825a2013-09-12 19:22:53 +01003356 if (d_is_positive(dentry))
Al Viroa8104a92012-07-20 02:25:00 +04003357 goto fail;
David Howellsb18825a2013-09-12 19:22:53 +01003358
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003359 /*
3360 * Special case - lookup gave negative, but... we had foo/bar/
3361 * From the vfs_mknod() POV we just have a negative dentry -
3362 * all is fine. Let's be bastards - you had / on the end, you've
3363 * been asking for (non-existent) directory. -ENOENT for you.
3364 */
Al Viroed75e952011-06-27 16:53:43 -04003365 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
Al Viroa8104a92012-07-20 02:25:00 +04003366 error = -ENOENT;
Al Viroed75e952011-06-27 16:53:43 -04003367 goto fail;
Al Viroe9baf6e2008-05-15 04:49:12 -04003368 }
Jan Karac30dabf2012-06-12 16:20:30 +02003369 if (unlikely(err2)) {
3370 error = err2;
Al Viroa8104a92012-07-20 02:25:00 +04003371 goto fail;
Jan Karac30dabf2012-06-12 16:20:30 +02003372 }
Al Viroed75e952011-06-27 16:53:43 -04003373 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003374 return dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003375fail:
Al Viroa8104a92012-07-20 02:25:00 +04003376 dput(dentry);
3377 dentry = ERR_PTR(error);
3378unlock:
Al Viroed75e952011-06-27 16:53:43 -04003379 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Karac30dabf2012-06-12 16:20:30 +02003380 if (!err2)
3381 mnt_drop_write(nd.path.mnt);
Al Viroed75e952011-06-27 16:53:43 -04003382out:
3383 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003384 return dentry;
3385}
Al Virofa14a0b2015-01-22 02:16:49 -05003386
3387struct dentry *kern_path_create(int dfd, const char *pathname,
3388 struct path *path, unsigned int lookup_flags)
3389{
Paul Moore51689102015-01-22 00:00:03 -05003390 struct filename *filename = getname_kernel(pathname);
3391 struct dentry *res;
3392
3393 if (IS_ERR(filename))
3394 return ERR_CAST(filename);
3395 res = filename_create(dfd, filename, path, lookup_flags);
3396 putname(filename);
3397 return res;
Al Virofa14a0b2015-01-22 02:16:49 -05003398}
Al Virodae6ad82011-06-26 11:50:15 -04003399EXPORT_SYMBOL(kern_path_create);
3400
Al Viro921a1652012-07-20 01:15:31 +04003401void done_path_create(struct path *path, struct dentry *dentry)
3402{
3403 dput(dentry);
3404 mutex_unlock(&path->dentry->d_inode->i_mutex);
Al Viroa8104a92012-07-20 02:25:00 +04003405 mnt_drop_write(path->mnt);
Al Viro921a1652012-07-20 01:15:31 +04003406 path_put(path);
3407}
3408EXPORT_SYMBOL(done_path_create);
3409
Jeff Layton1ac12b42012-12-11 12:10:06 -05003410struct dentry *user_path_create(int dfd, const char __user *pathname,
3411 struct path *path, unsigned int lookup_flags)
Al Virodae6ad82011-06-26 11:50:15 -04003412{
Jeff Layton91a27b22012-10-10 15:25:28 -04003413 struct filename *tmp = getname(pathname);
Al Virodae6ad82011-06-26 11:50:15 -04003414 struct dentry *res;
3415 if (IS_ERR(tmp))
3416 return ERR_CAST(tmp);
Al Virofa14a0b2015-01-22 02:16:49 -05003417 res = filename_create(dfd, tmp, path, lookup_flags);
Al Virodae6ad82011-06-26 11:50:15 -04003418 putname(tmp);
3419 return res;
3420}
3421EXPORT_SYMBOL(user_path_create);
3422
Al Viro1a67aaf2011-07-26 01:52:52 -04003423int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003424{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003425 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003426
3427 if (error)
3428 return error;
3429
Eric W. Biederman975d6b32011-11-13 12:16:43 -08003430 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003431 return -EPERM;
3432
Al Viroacfa4382008-12-04 10:06:33 -05003433 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003434 return -EPERM;
3435
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07003436 error = devcgroup_inode_mknod(mode, dev);
3437 if (error)
3438 return error;
3439
Linus Torvalds1da177e2005-04-16 15:20:36 -07003440 error = security_inode_mknod(dir, dentry, mode, dev);
3441 if (error)
3442 return error;
3443
Linus Torvalds1da177e2005-04-16 15:20:36 -07003444 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003445 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003446 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003447 return error;
3448}
Al Viro4d359502014-03-14 12:20:17 -04003449EXPORT_SYMBOL(vfs_mknod);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003450
Al Virof69aac02011-07-26 04:31:40 -04003451static int may_mknod(umode_t mode)
Dave Hansen463c3192008-02-15 14:37:57 -08003452{
3453 switch (mode & S_IFMT) {
3454 case S_IFREG:
3455 case S_IFCHR:
3456 case S_IFBLK:
3457 case S_IFIFO:
3458 case S_IFSOCK:
3459 case 0: /* zero mode translates to S_IFREG */
3460 return 0;
3461 case S_IFDIR:
3462 return -EPERM;
3463 default:
3464 return -EINVAL;
3465 }
3466}
3467
Al Viro8208a222011-07-25 17:32:17 -04003468SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003469 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003470{
Al Viro2ad94ae2008-07-21 09:32:51 -04003471 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003472 struct path path;
3473 int error;
Jeff Layton972567f2012-12-20 16:00:10 -05003474 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003475
Al Viro8e4bfca2012-07-20 01:17:26 +04003476 error = may_mknod(mode);
3477 if (error)
3478 return error;
Jeff Layton972567f2012-12-20 16:00:10 -05003479retry:
3480 dentry = user_path_create(dfd, filename, &path, lookup_flags);
Al Virodae6ad82011-06-26 11:50:15 -04003481 if (IS_ERR(dentry))
3482 return PTR_ERR(dentry);
Al Viro2ad94ae2008-07-21 09:32:51 -04003483
Al Virodae6ad82011-06-26 11:50:15 -04003484 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003485 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003486 error = security_path_mknod(&path, dentry, mode, dev);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003487 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04003488 goto out;
Dave Hansen463c3192008-02-15 14:37:57 -08003489 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003490 case 0: case S_IFREG:
Al Viro312b63f2012-06-10 18:09:36 -04003491 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003492 break;
3493 case S_IFCHR: case S_IFBLK:
Al Virodae6ad82011-06-26 11:50:15 -04003494 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003495 new_decode_dev(dev));
3496 break;
3497 case S_IFIFO: case S_IFSOCK:
Al Virodae6ad82011-06-26 11:50:15 -04003498 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003499 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003500 }
Al Viroa8104a92012-07-20 02:25:00 +04003501out:
Al Viro921a1652012-07-20 01:15:31 +04003502 done_path_create(&path, dentry);
Jeff Layton972567f2012-12-20 16:00:10 -05003503 if (retry_estale(error, lookup_flags)) {
3504 lookup_flags |= LOOKUP_REVAL;
3505 goto retry;
3506 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003507 return error;
3508}
3509
Al Viro8208a222011-07-25 17:32:17 -04003510SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003511{
3512 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3513}
3514
Al Viro18bb1db2011-07-26 01:41:39 -04003515int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003516{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003517 int error = may_create(dir, dentry);
Al Viro8de52772012-02-06 12:45:27 -05003518 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003519
3520 if (error)
3521 return error;
3522
Al Viroacfa4382008-12-04 10:06:33 -05003523 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003524 return -EPERM;
3525
3526 mode &= (S_IRWXUGO|S_ISVTX);
3527 error = security_inode_mkdir(dir, dentry, mode);
3528 if (error)
3529 return error;
3530
Al Viro8de52772012-02-06 12:45:27 -05003531 if (max_links && dir->i_nlink >= max_links)
3532 return -EMLINK;
3533
Linus Torvalds1da177e2005-04-16 15:20:36 -07003534 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003535 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003536 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003537 return error;
3538}
Al Viro4d359502014-03-14 12:20:17 -04003539EXPORT_SYMBOL(vfs_mkdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003540
Al Viroa218d0f2011-11-21 14:59:34 -05003541SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003542{
Dave Hansen6902d922006-09-30 23:29:01 -07003543 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003544 struct path path;
3545 int error;
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003546 unsigned int lookup_flags = LOOKUP_DIRECTORY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003547
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003548retry:
3549 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07003550 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003551 return PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003552
Al Virodae6ad82011-06-26 11:50:15 -04003553 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003554 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003555 error = security_path_mkdir(&path, dentry, mode);
Al Viroa8104a92012-07-20 02:25:00 +04003556 if (!error)
3557 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
Al Viro921a1652012-07-20 01:15:31 +04003558 done_path_create(&path, dentry);
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003559 if (retry_estale(error, lookup_flags)) {
3560 lookup_flags |= LOOKUP_REVAL;
3561 goto retry;
3562 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003563 return error;
3564}
3565
Al Viroa218d0f2011-11-21 14:59:34 -05003566SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003567{
3568 return sys_mkdirat(AT_FDCWD, pathname, mode);
3569}
3570
Linus Torvalds1da177e2005-04-16 15:20:36 -07003571/*
Sage Weila71905f2011-05-24 13:06:08 -07003572 * The dentry_unhash() helper will try to drop the dentry early: we
J. Bruce Fieldsc0d02592012-02-15 11:48:40 -05003573 * should have a usage count of 1 if we're the only user of this
Sage Weila71905f2011-05-24 13:06:08 -07003574 * dentry, and if that is true (possibly after pruning the dcache),
3575 * then we drop the dentry now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003576 *
3577 * A low-level filesystem can, if it choses, legally
3578 * do a
3579 *
3580 * if (!d_unhashed(dentry))
3581 * return -EBUSY;
3582 *
3583 * if it cannot handle the case of removing a directory
3584 * that is still in use by something else..
3585 */
3586void dentry_unhash(struct dentry *dentry)
3587{
Vasily Averindc168422006-12-06 20:37:07 -08003588 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003589 spin_lock(&dentry->d_lock);
Waiman Long98474232013-08-28 18:24:59 -07003590 if (dentry->d_lockref.count == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003591 __d_drop(dentry);
3592 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003593}
Al Viro4d359502014-03-14 12:20:17 -04003594EXPORT_SYMBOL(dentry_unhash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003595
3596int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3597{
3598 int error = may_delete(dir, dentry, 1);
3599
3600 if (error)
3601 return error;
3602
Al Viroacfa4382008-12-04 10:06:33 -05003603 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003604 return -EPERM;
3605
Al Viro1d2ef592011-09-14 18:55:41 +01003606 dget(dentry);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003607 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003608
Sage Weil912dbc12011-05-24 13:06:11 -07003609 error = -EBUSY;
Eric W. Biederman7af13642013-10-04 19:15:13 -07003610 if (is_local_mountpoint(dentry))
Sage Weil912dbc12011-05-24 13:06:11 -07003611 goto out;
3612
3613 error = security_inode_rmdir(dir, dentry);
3614 if (error)
3615 goto out;
3616
Sage Weil3cebde22011-05-29 21:20:59 -07003617 shrink_dcache_parent(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003618 error = dir->i_op->rmdir(dir, dentry);
3619 if (error)
3620 goto out;
3621
3622 dentry->d_inode->i_flags |= S_DEAD;
3623 dont_mount(dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003624 detach_mounts(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003625
3626out:
3627 mutex_unlock(&dentry->d_inode->i_mutex);
Al Viro1d2ef592011-09-14 18:55:41 +01003628 dput(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003629 if (!error)
3630 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003631 return error;
3632}
Al Viro4d359502014-03-14 12:20:17 -04003633EXPORT_SYMBOL(vfs_rmdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003634
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003635static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003636{
3637 int error = 0;
Jeff Layton91a27b22012-10-10 15:25:28 -04003638 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003639 struct dentry *dentry;
3640 struct nameidata nd;
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003641 unsigned int lookup_flags = 0;
3642retry:
3643 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04003644 if (IS_ERR(name))
3645 return PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003646
3647 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003648 case LAST_DOTDOT:
3649 error = -ENOTEMPTY;
3650 goto exit1;
3651 case LAST_DOT:
3652 error = -EINVAL;
3653 goto exit1;
3654 case LAST_ROOT:
3655 error = -EBUSY;
3656 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003657 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003658
3659 nd.flags &= ~LOOKUP_PARENT;
Jan Karac30dabf2012-06-12 16:20:30 +02003660 error = mnt_want_write(nd.path.mnt);
3661 if (error)
3662 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003663
Jan Blunck4ac91372008-02-14 19:34:32 -08003664 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08003665 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003666 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003667 if (IS_ERR(dentry))
3668 goto exit2;
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003669 if (!dentry->d_inode) {
3670 error = -ENOENT;
3671 goto exit3;
3672 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003673 error = security_path_rmdir(&nd.path, dentry);
3674 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003675 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08003676 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Dave Hansen06227532008-02-15 14:37:34 -08003677exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07003678 dput(dentry);
3679exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08003680 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Karac30dabf2012-06-12 16:20:30 +02003681 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003682exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003683 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003684 putname(name);
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003685 if (retry_estale(error, lookup_flags)) {
3686 lookup_flags |= LOOKUP_REVAL;
3687 goto retry;
3688 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003689 return error;
3690}
3691
Heiko Carstens3cdad422009-01-14 14:14:22 +01003692SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003693{
3694 return do_rmdir(AT_FDCWD, pathname);
3695}
3696
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003697/**
3698 * vfs_unlink - unlink a filesystem object
3699 * @dir: parent directory
3700 * @dentry: victim
3701 * @delegated_inode: returns victim inode, if the inode is delegated.
3702 *
3703 * The caller must hold dir->i_mutex.
3704 *
3705 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3706 * return a reference to the inode in delegated_inode. The caller
3707 * should then break the delegation on that inode and retry. Because
3708 * breaking a delegation may take a long time, the caller should drop
3709 * dir->i_mutex before doing so.
3710 *
3711 * Alternatively, a caller may pass NULL for delegated_inode. This may
3712 * be appropriate for callers that expect the underlying filesystem not
3713 * to be NFS exported.
3714 */
3715int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003716{
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003717 struct inode *target = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003718 int error = may_delete(dir, dentry, 0);
3719
3720 if (error)
3721 return error;
3722
Al Viroacfa4382008-12-04 10:06:33 -05003723 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003724 return -EPERM;
3725
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003726 mutex_lock(&target->i_mutex);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003727 if (is_local_mountpoint(dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003728 error = -EBUSY;
3729 else {
3730 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05003731 if (!error) {
J. Bruce Fields5a146962012-08-28 07:50:40 -07003732 error = try_break_deleg(target, delegated_inode);
3733 if (error)
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003734 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003735 error = dir->i_op->unlink(dir, dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003736 if (!error) {
Al Virod83c49f2010-04-30 17:17:09 -04003737 dont_mount(dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003738 detach_mounts(dentry);
3739 }
Al Virobec10522010-03-03 14:12:08 -05003740 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003741 }
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003742out:
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003743 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003744
3745 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3746 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003747 fsnotify_link_count(target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003748 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003749 }
Robert Love0eeca282005-07-12 17:06:03 -04003750
Linus Torvalds1da177e2005-04-16 15:20:36 -07003751 return error;
3752}
Al Viro4d359502014-03-14 12:20:17 -04003753EXPORT_SYMBOL(vfs_unlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003754
3755/*
3756 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003757 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07003758 * writeout happening, and we don't want to prevent access to the directory
3759 * while waiting on the I/O.
3760 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003761static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003762{
Al Viro2ad94ae2008-07-21 09:32:51 -04003763 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04003764 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003765 struct dentry *dentry;
3766 struct nameidata nd;
3767 struct inode *inode = NULL;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003768 struct inode *delegated_inode = NULL;
Jeff Layton5d18f812012-12-20 16:38:04 -05003769 unsigned int lookup_flags = 0;
3770retry:
3771 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04003772 if (IS_ERR(name))
3773 return PTR_ERR(name);
Al Viro2ad94ae2008-07-21 09:32:51 -04003774
Linus Torvalds1da177e2005-04-16 15:20:36 -07003775 error = -EISDIR;
3776 if (nd.last_type != LAST_NORM)
3777 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003778
3779 nd.flags &= ~LOOKUP_PARENT;
Jan Karac30dabf2012-06-12 16:20:30 +02003780 error = mnt_want_write(nd.path.mnt);
3781 if (error)
3782 goto exit1;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003783retry_deleg:
Jan Blunck4ac91372008-02-14 19:34:32 -08003784 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08003785 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003786 error = PTR_ERR(dentry);
3787 if (!IS_ERR(dentry)) {
3788 /* Why not before? Because we want correct error value */
Török Edwin50338b82011-06-16 00:06:14 +03003789 if (nd.last.name[nd.last.len])
3790 goto slashes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003791 inode = dentry->d_inode;
David Howellsb18825a2013-09-12 19:22:53 +01003792 if (d_is_negative(dentry))
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003793 goto slashes;
3794 ihold(inode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003795 error = security_path_unlink(&nd.path, dentry);
3796 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003797 goto exit2;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003798 error = vfs_unlink(nd.path.dentry->d_inode, dentry, &delegated_inode);
Jan Karac30dabf2012-06-12 16:20:30 +02003799exit2:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003800 dput(dentry);
3801 }
Jan Blunck4ac91372008-02-14 19:34:32 -08003802 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003803 if (inode)
3804 iput(inode); /* truncate the inode here */
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003805 inode = NULL;
3806 if (delegated_inode) {
J. Bruce Fields5a146962012-08-28 07:50:40 -07003807 error = break_deleg_wait(&delegated_inode);
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003808 if (!error)
3809 goto retry_deleg;
3810 }
Jan Karac30dabf2012-06-12 16:20:30 +02003811 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003812exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003813 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003814 putname(name);
Jeff Layton5d18f812012-12-20 16:38:04 -05003815 if (retry_estale(error, lookup_flags)) {
3816 lookup_flags |= LOOKUP_REVAL;
3817 inode = NULL;
3818 goto retry;
3819 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003820 return error;
3821
3822slashes:
David Howellsb18825a2013-09-12 19:22:53 +01003823 if (d_is_negative(dentry))
3824 error = -ENOENT;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003825 else if (d_is_dir(dentry))
David Howellsb18825a2013-09-12 19:22:53 +01003826 error = -EISDIR;
3827 else
3828 error = -ENOTDIR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003829 goto exit2;
3830}
3831
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003832SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003833{
3834 if ((flag & ~AT_REMOVEDIR) != 0)
3835 return -EINVAL;
3836
3837 if (flag & AT_REMOVEDIR)
3838 return do_rmdir(dfd, pathname);
3839
3840 return do_unlinkat(dfd, pathname);
3841}
3842
Heiko Carstens3480b252009-01-14 14:14:16 +01003843SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003844{
3845 return do_unlinkat(AT_FDCWD, pathname);
3846}
3847
Miklos Szeredidb2e7472008-06-24 16:50:16 +02003848int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003849{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003850 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003851
3852 if (error)
3853 return error;
3854
Al Viroacfa4382008-12-04 10:06:33 -05003855 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003856 return -EPERM;
3857
3858 error = security_inode_symlink(dir, dentry, oldname);
3859 if (error)
3860 return error;
3861
Linus Torvalds1da177e2005-04-16 15:20:36 -07003862 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003863 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003864 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003865 return error;
3866}
Al Viro4d359502014-03-14 12:20:17 -04003867EXPORT_SYMBOL(vfs_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003868
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003869SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3870 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003871{
Al Viro2ad94ae2008-07-21 09:32:51 -04003872 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04003873 struct filename *from;
Dave Hansen6902d922006-09-30 23:29:01 -07003874 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003875 struct path path;
Jeff Laytonf46d3562012-12-11 12:10:08 -05003876 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003877
3878 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04003879 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003880 return PTR_ERR(from);
Jeff Laytonf46d3562012-12-11 12:10:08 -05003881retry:
3882 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07003883 error = PTR_ERR(dentry);
3884 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003885 goto out_putname;
Dave Hansen6902d922006-09-30 23:29:01 -07003886
Jeff Layton91a27b22012-10-10 15:25:28 -04003887 error = security_path_symlink(&path, dentry, from->name);
Al Viroa8104a92012-07-20 02:25:00 +04003888 if (!error)
Jeff Layton91a27b22012-10-10 15:25:28 -04003889 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
Al Viro921a1652012-07-20 01:15:31 +04003890 done_path_create(&path, dentry);
Jeff Laytonf46d3562012-12-11 12:10:08 -05003891 if (retry_estale(error, lookup_flags)) {
3892 lookup_flags |= LOOKUP_REVAL;
3893 goto retry;
3894 }
Dave Hansen6902d922006-09-30 23:29:01 -07003895out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003896 putname(from);
3897 return error;
3898}
3899
Heiko Carstens3480b252009-01-14 14:14:16 +01003900SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003901{
3902 return sys_symlinkat(oldname, AT_FDCWD, newname);
3903}
3904
J. Bruce Fields146a8592011-09-20 17:14:31 -04003905/**
3906 * vfs_link - create a new link
3907 * @old_dentry: object to be linked
3908 * @dir: new parent
3909 * @new_dentry: where to create the new link
3910 * @delegated_inode: returns inode needing a delegation break
3911 *
3912 * The caller must hold dir->i_mutex
3913 *
3914 * If vfs_link discovers a delegation on the to-be-linked file in need
3915 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3916 * inode in delegated_inode. The caller should then break the delegation
3917 * and retry. Because breaking a delegation may take a long time, the
3918 * caller should drop the i_mutex before doing so.
3919 *
3920 * Alternatively, a caller may pass NULL for delegated_inode. This may
3921 * be appropriate for callers that expect the underlying filesystem not
3922 * to be NFS exported.
3923 */
3924int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003925{
3926 struct inode *inode = old_dentry->d_inode;
Al Viro8de52772012-02-06 12:45:27 -05003927 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003928 int error;
3929
3930 if (!inode)
3931 return -ENOENT;
3932
Miklos Szeredia95164d2008-07-30 15:08:48 +02003933 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003934 if (error)
3935 return error;
3936
3937 if (dir->i_sb != inode->i_sb)
3938 return -EXDEV;
3939
3940 /*
3941 * A link to an append-only or immutable file cannot be created.
3942 */
3943 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3944 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05003945 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003946 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003947 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003948 return -EPERM;
3949
3950 error = security_inode_link(old_dentry, dir, new_dentry);
3951 if (error)
3952 return error;
3953
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003954 mutex_lock(&inode->i_mutex);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05303955 /* Make sure we don't allow creating hardlink to an unlinked file */
Al Virof4e0c302013-06-11 08:34:36 +04003956 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05303957 error = -ENOENT;
Al Viro8de52772012-02-06 12:45:27 -05003958 else if (max_links && inode->i_nlink >= max_links)
3959 error = -EMLINK;
J. Bruce Fields146a8592011-09-20 17:14:31 -04003960 else {
3961 error = try_break_deleg(inode, delegated_inode);
3962 if (!error)
3963 error = dir->i_op->link(old_dentry, dir, new_dentry);
3964 }
Al Virof4e0c302013-06-11 08:34:36 +04003965
3966 if (!error && (inode->i_state & I_LINKABLE)) {
3967 spin_lock(&inode->i_lock);
3968 inode->i_state &= ~I_LINKABLE;
3969 spin_unlock(&inode->i_lock);
3970 }
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003971 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07003972 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003973 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003974 return error;
3975}
Al Viro4d359502014-03-14 12:20:17 -04003976EXPORT_SYMBOL(vfs_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003977
3978/*
3979 * Hardlinks are often used in delicate situations. We avoid
3980 * security-related surprises by not following symlinks on the
3981 * newname. --KAB
3982 *
3983 * We don't follow them on the oldname either to be compatible
3984 * with linux 2.0, and to avoid hard-linking to directories
3985 * and other special files. --ADM
3986 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003987SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3988 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003989{
3990 struct dentry *new_dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003991 struct path old_path, new_path;
J. Bruce Fields146a8592011-09-20 17:14:31 -04003992 struct inode *delegated_inode = NULL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303993 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003994 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003995
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303996 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003997 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303998 /*
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07003999 * To use null names we require CAP_DAC_READ_SEARCH
4000 * This ensures that not everyone will be able to create
4001 * handlink using the passed filedescriptor.
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304002 */
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004003 if (flags & AT_EMPTY_PATH) {
4004 if (!capable(CAP_DAC_READ_SEARCH))
4005 return -ENOENT;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304006 how = LOOKUP_EMPTY;
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004007 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004008
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304009 if (flags & AT_SYMLINK_FOLLOW)
4010 how |= LOOKUP_FOLLOW;
Jeff Layton442e31c2012-12-20 16:15:38 -05004011retry:
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304012 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004013 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04004014 return error;
4015
Jeff Layton442e31c2012-12-20 16:15:38 -05004016 new_dentry = user_path_create(newdfd, newname, &new_path,
4017 (how & LOOKUP_REVAL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004018 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07004019 if (IS_ERR(new_dentry))
Al Virodae6ad82011-06-26 11:50:15 -04004020 goto out;
4021
4022 error = -EXDEV;
4023 if (old_path.mnt != new_path.mnt)
4024 goto out_dput;
Kees Cook800179c2012-07-25 17:29:07 -07004025 error = may_linkat(&old_path);
4026 if (unlikely(error))
4027 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04004028 error = security_path_link(old_path.dentry, &new_path, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004029 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04004030 goto out_dput;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004031 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
Dave Hansen75c3f292008-02-15 14:37:45 -08004032out_dput:
Al Viro921a1652012-07-20 01:15:31 +04004033 done_path_create(&new_path, new_dentry);
J. Bruce Fields146a8592011-09-20 17:14:31 -04004034 if (delegated_inode) {
4035 error = break_deleg_wait(&delegated_inode);
Oleg Drokind22e6332014-01-31 15:41:58 -05004036 if (!error) {
4037 path_put(&old_path);
J. Bruce Fields146a8592011-09-20 17:14:31 -04004038 goto retry;
Oleg Drokind22e6332014-01-31 15:41:58 -05004039 }
J. Bruce Fields146a8592011-09-20 17:14:31 -04004040 }
Jeff Layton442e31c2012-12-20 16:15:38 -05004041 if (retry_estale(error, how)) {
Oleg Drokind22e6332014-01-31 15:41:58 -05004042 path_put(&old_path);
Jeff Layton442e31c2012-12-20 16:15:38 -05004043 how |= LOOKUP_REVAL;
4044 goto retry;
4045 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004046out:
Al Viro2d8f3032008-07-22 09:59:21 -04004047 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004048
4049 return error;
4050}
4051
Heiko Carstens3480b252009-01-14 14:14:16 +01004052SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004053{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004054 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004055}
4056
Miklos Szeredibc270272014-04-01 17:08:42 +02004057/**
4058 * vfs_rename - rename a filesystem object
4059 * @old_dir: parent of source
4060 * @old_dentry: source
4061 * @new_dir: parent of destination
4062 * @new_dentry: destination
4063 * @delegated_inode: returns an inode needing a delegation break
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004064 * @flags: rename flags
Miklos Szeredibc270272014-04-01 17:08:42 +02004065 *
4066 * The caller must hold multiple mutexes--see lock_rename()).
4067 *
4068 * If vfs_rename discovers a delegation in need of breaking at either
4069 * the source or destination, it will return -EWOULDBLOCK and return a
4070 * reference to the inode in delegated_inode. The caller should then
4071 * break the delegation and retry. Because breaking a delegation may
4072 * take a long time, the caller should drop all locks before doing
4073 * so.
4074 *
4075 * Alternatively, a caller may pass NULL for delegated_inode. This may
4076 * be appropriate for callers that expect the underlying filesystem not
4077 * to be NFS exported.
4078 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07004079 * The worst of all namespace operations - renaming directory. "Perverted"
4080 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4081 * Problems:
J. Bruce Fieldsd03b29a2014-02-17 16:52:33 -05004082 * a) we can get into loop creation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004083 * b) race potential - two innocent renames can create a loop together.
4084 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004085 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07004086 * story.
J. Bruce Fields6cedba82012-03-05 11:40:41 -05004087 * c) we have to lock _four_ objects - parents and victim (if it exists),
4088 * and source (if it is not a directory).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004089 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07004090 * whether the target exists). Solution: try to be smart with locking
4091 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004092 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07004093 * move will be locked. Thus we can rank directories by the tree
4094 * (ancestors first) and rank all non-directories after them.
4095 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004096 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004097 * HOWEVER, it relies on the assumption that any object with ->lookup()
4098 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4099 * we'd better make sure that there's no link(2) for them.
Sage Weile4eaac02011-05-24 13:06:07 -07004100 * d) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004101 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07004102 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05004103 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07004104 * locking].
4105 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004106int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004107 struct inode *new_dir, struct dentry *new_dentry,
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004108 struct inode **delegated_inode, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004109{
4110 int error;
Miklos Szeredibc270272014-04-01 17:08:42 +02004111 bool is_dir = d_is_dir(old_dentry);
Eric Paris59b0df22010-02-08 12:53:52 -05004112 const unsigned char *old_name;
Miklos Szeredibc270272014-04-01 17:08:42 +02004113 struct inode *source = old_dentry->d_inode;
4114 struct inode *target = new_dentry->d_inode;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004115 bool new_is_dir = false;
4116 unsigned max_links = new_dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004117
Miklos Szeredibc270272014-04-01 17:08:42 +02004118 if (source == target)
4119 return 0;
4120
Linus Torvalds1da177e2005-04-16 15:20:36 -07004121 error = may_delete(old_dir, old_dentry, is_dir);
4122 if (error)
4123 return error;
4124
Miklos Szeredida1ce062014-04-01 17:08:43 +02004125 if (!target) {
Miklos Szeredia95164d2008-07-30 15:08:48 +02004126 error = may_create(new_dir, new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004127 } else {
4128 new_is_dir = d_is_dir(new_dentry);
4129
4130 if (!(flags & RENAME_EXCHANGE))
4131 error = may_delete(new_dir, new_dentry, is_dir);
4132 else
4133 error = may_delete(new_dir, new_dentry, new_is_dir);
4134 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004135 if (error)
4136 return error;
4137
Miklos Szeredi7177a9c2014-07-23 15:15:30 +02004138 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004139 return -EPERM;
4140
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004141 if (flags && !old_dir->i_op->rename2)
4142 return -EINVAL;
4143
Miklos Szeredibc270272014-04-01 17:08:42 +02004144 /*
4145 * If we are going to change the parent - check write permissions,
4146 * we'll need to flip '..'.
4147 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02004148 if (new_dir != old_dir) {
4149 if (is_dir) {
4150 error = inode_permission(source, MAY_WRITE);
4151 if (error)
4152 return error;
4153 }
4154 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4155 error = inode_permission(target, MAY_WRITE);
4156 if (error)
4157 return error;
4158 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004159 }
Robert Love0eeca282005-07-12 17:06:03 -04004160
Miklos Szeredi0b3974e2014-04-01 17:08:43 +02004161 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4162 flags);
Miklos Szeredibc270272014-04-01 17:08:42 +02004163 if (error)
4164 return error;
4165
4166 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4167 dget(new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004168 if (!is_dir || (flags & RENAME_EXCHANGE))
Miklos Szeredibc270272014-04-01 17:08:42 +02004169 lock_two_nondirectories(source, target);
4170 else if (target)
4171 mutex_lock(&target->i_mutex);
4172
4173 error = -EBUSY;
Eric W. Biederman7af13642013-10-04 19:15:13 -07004174 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
Miklos Szeredibc270272014-04-01 17:08:42 +02004175 goto out;
4176
Miklos Szeredida1ce062014-04-01 17:08:43 +02004177 if (max_links && new_dir != old_dir) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004178 error = -EMLINK;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004179 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
Miklos Szeredibc270272014-04-01 17:08:42 +02004180 goto out;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004181 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4182 old_dir->i_nlink >= max_links)
4183 goto out;
4184 }
4185 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4186 shrink_dcache_parent(new_dentry);
4187 if (!is_dir) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004188 error = try_break_deleg(source, delegated_inode);
4189 if (error)
4190 goto out;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004191 }
4192 if (target && !new_is_dir) {
4193 error = try_break_deleg(target, delegated_inode);
4194 if (error)
4195 goto out;
Miklos Szeredibc270272014-04-01 17:08:42 +02004196 }
Miklos Szeredi7177a9c2014-07-23 15:15:30 +02004197 if (!old_dir->i_op->rename2) {
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004198 error = old_dir->i_op->rename(old_dir, old_dentry,
4199 new_dir, new_dentry);
4200 } else {
Miklos Szeredi7177a9c2014-07-23 15:15:30 +02004201 WARN_ON(old_dir->i_op->rename != NULL);
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004202 error = old_dir->i_op->rename2(old_dir, old_dentry,
4203 new_dir, new_dentry, flags);
4204 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004205 if (error)
4206 goto out;
4207
Miklos Szeredida1ce062014-04-01 17:08:43 +02004208 if (!(flags & RENAME_EXCHANGE) && target) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004209 if (is_dir)
4210 target->i_flags |= S_DEAD;
4211 dont_mount(new_dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07004212 detach_mounts(new_dentry);
Miklos Szeredibc270272014-04-01 17:08:42 +02004213 }
Miklos Szeredida1ce062014-04-01 17:08:43 +02004214 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4215 if (!(flags & RENAME_EXCHANGE))
4216 d_move(old_dentry, new_dentry);
4217 else
4218 d_exchange(old_dentry, new_dentry);
4219 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004220out:
Miklos Szeredida1ce062014-04-01 17:08:43 +02004221 if (!is_dir || (flags & RENAME_EXCHANGE))
Miklos Szeredibc270272014-04-01 17:08:42 +02004222 unlock_two_nondirectories(source, target);
4223 else if (target)
4224 mutex_unlock(&target->i_mutex);
4225 dput(new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004226 if (!error) {
Al Viro123df292009-12-25 04:57:57 -05004227 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Miklos Szeredida1ce062014-04-01 17:08:43 +02004228 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4229 if (flags & RENAME_EXCHANGE) {
4230 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4231 new_is_dir, NULL, new_dentry);
4232 }
4233 }
Robert Love0eeca282005-07-12 17:06:03 -04004234 fsnotify_oldname_free(old_name);
4235
Linus Torvalds1da177e2005-04-16 15:20:36 -07004236 return error;
4237}
Al Viro4d359502014-03-14 12:20:17 -04004238EXPORT_SYMBOL(vfs_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004239
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004240SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4241 int, newdfd, const char __user *, newname, unsigned int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004242{
Al Viro2ad94ae2008-07-21 09:32:51 -04004243 struct dentry *old_dir, *new_dir;
4244 struct dentry *old_dentry, *new_dentry;
4245 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004246 struct nameidata oldnd, newnd;
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004247 struct inode *delegated_inode = NULL;
Jeff Layton91a27b22012-10-10 15:25:28 -04004248 struct filename *from;
4249 struct filename *to;
Jeff Laytonc6a94282012-12-11 12:10:10 -05004250 unsigned int lookup_flags = 0;
4251 bool should_retry = false;
Al Viro2ad94ae2008-07-21 09:32:51 -04004252 int error;
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004253
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004254 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredida1ce062014-04-01 17:08:43 +02004255 return -EINVAL;
4256
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004257 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4258 (flags & RENAME_EXCHANGE))
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004259 return -EINVAL;
4260
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004261 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4262 return -EPERM;
4263
Jeff Laytonc6a94282012-12-11 12:10:10 -05004264retry:
4265 from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04004266 if (IS_ERR(from)) {
4267 error = PTR_ERR(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004268 goto exit;
Jeff Layton91a27b22012-10-10 15:25:28 -04004269 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004270
Jeff Laytonc6a94282012-12-11 12:10:10 -05004271 to = user_path_parent(newdfd, newname, &newnd, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04004272 if (IS_ERR(to)) {
4273 error = PTR_ERR(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004274 goto exit1;
Jeff Layton91a27b22012-10-10 15:25:28 -04004275 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004276
4277 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08004278 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004279 goto exit2;
4280
Jan Blunck4ac91372008-02-14 19:34:32 -08004281 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004282 error = -EBUSY;
4283 if (oldnd.last_type != LAST_NORM)
4284 goto exit2;
4285
Jan Blunck4ac91372008-02-14 19:34:32 -08004286 new_dir = newnd.path.dentry;
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004287 if (flags & RENAME_NOREPLACE)
4288 error = -EEXIST;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004289 if (newnd.last_type != LAST_NORM)
4290 goto exit2;
4291
Jan Karac30dabf2012-06-12 16:20:30 +02004292 error = mnt_want_write(oldnd.path.mnt);
4293 if (error)
4294 goto exit2;
4295
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09004296 oldnd.flags &= ~LOOKUP_PARENT;
4297 newnd.flags &= ~LOOKUP_PARENT;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004298 if (!(flags & RENAME_EXCHANGE))
4299 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09004300
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004301retry_deleg:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004302 trap = lock_rename(new_dir, old_dir);
4303
Christoph Hellwig49705b72005-11-08 21:35:06 -08004304 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004305 error = PTR_ERR(old_dentry);
4306 if (IS_ERR(old_dentry))
4307 goto exit3;
4308 /* source must exist */
4309 error = -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01004310 if (d_is_negative(old_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004311 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08004312 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004313 error = PTR_ERR(new_dentry);
4314 if (IS_ERR(new_dentry))
4315 goto exit4;
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004316 error = -EEXIST;
4317 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4318 goto exit5;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004319 if (flags & RENAME_EXCHANGE) {
4320 error = -ENOENT;
4321 if (d_is_negative(new_dentry))
4322 goto exit5;
4323
4324 if (!d_is_dir(new_dentry)) {
4325 error = -ENOTDIR;
4326 if (newnd.last.name[newnd.last.len])
4327 goto exit5;
4328 }
4329 }
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004330 /* unless the source is a directory trailing slashes give -ENOTDIR */
4331 if (!d_is_dir(old_dentry)) {
4332 error = -ENOTDIR;
4333 if (oldnd.last.name[oldnd.last.len])
4334 goto exit5;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004335 if (!(flags & RENAME_EXCHANGE) && newnd.last.name[newnd.last.len])
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004336 goto exit5;
4337 }
4338 /* source should not be ancestor of target */
4339 error = -EINVAL;
4340 if (old_dentry == trap)
4341 goto exit5;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004342 /* target should not be an ancestor of source */
Miklos Szeredida1ce062014-04-01 17:08:43 +02004343 if (!(flags & RENAME_EXCHANGE))
4344 error = -ENOTEMPTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004345 if (new_dentry == trap)
4346 goto exit5;
4347
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004348 error = security_path_rename(&oldnd.path, old_dentry,
Miklos Szeredi0b3974e2014-04-01 17:08:43 +02004349 &newnd.path, new_dentry, flags);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004350 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02004351 goto exit5;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004352 error = vfs_rename(old_dir->d_inode, old_dentry,
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004353 new_dir->d_inode, new_dentry,
4354 &delegated_inode, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004355exit5:
4356 dput(new_dentry);
4357exit4:
4358 dput(old_dentry);
4359exit3:
4360 unlock_rename(new_dir, old_dir);
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004361 if (delegated_inode) {
4362 error = break_deleg_wait(&delegated_inode);
4363 if (!error)
4364 goto retry_deleg;
4365 }
Jan Karac30dabf2012-06-12 16:20:30 +02004366 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004367exit2:
Jeff Laytonc6a94282012-12-11 12:10:10 -05004368 if (retry_estale(error, lookup_flags))
4369 should_retry = true;
Jan Blunck1d957f92008-02-14 19:34:35 -08004370 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04004371 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004372exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08004373 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004374 putname(from);
Jeff Laytonc6a94282012-12-11 12:10:10 -05004375 if (should_retry) {
4376 should_retry = false;
4377 lookup_flags |= LOOKUP_REVAL;
4378 goto retry;
4379 }
Al Viro2ad94ae2008-07-21 09:32:51 -04004380exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004381 return error;
4382}
4383
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004384SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4385 int, newdfd, const char __user *, newname)
4386{
4387 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4388}
4389
Heiko Carstensa26eab22009-01-14 14:14:17 +01004390SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004391{
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004392 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004393}
4394
Miklos Szeredi787fb6b2014-10-24 00:14:36 +02004395int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4396{
4397 int error = may_create(dir, dentry);
4398 if (error)
4399 return error;
4400
4401 if (!dir->i_op->mknod)
4402 return -EPERM;
4403
4404 return dir->i_op->mknod(dir, dentry,
4405 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4406}
4407EXPORT_SYMBOL(vfs_whiteout);
4408
Al Viro5d826c82014-03-14 13:42:45 -04004409int readlink_copy(char __user *buffer, int buflen, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004410{
Al Viro5d826c82014-03-14 13:42:45 -04004411 int len = PTR_ERR(link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004412 if (IS_ERR(link))
4413 goto out;
4414
4415 len = strlen(link);
4416 if (len > (unsigned) buflen)
4417 len = buflen;
4418 if (copy_to_user(buffer, link, len))
4419 len = -EFAULT;
4420out:
4421 return len;
4422}
Al Viro5d826c82014-03-14 13:42:45 -04004423EXPORT_SYMBOL(readlink_copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004424
4425/*
4426 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4427 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4428 * using) it for any given inode is up to filesystem.
4429 */
4430int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4431{
4432 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004433 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07004434 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004435
Linus Torvalds1da177e2005-04-16 15:20:36 -07004436 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004437 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07004438 if (IS_ERR(cookie))
4439 return PTR_ERR(cookie);
4440
Al Viro5d826c82014-03-14 13:42:45 -04004441 res = readlink_copy(buffer, buflen, nd_get_link(&nd));
Marcin Slusarz694a1762008-06-09 16:40:37 -07004442 if (dentry->d_inode->i_op->put_link)
4443 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
4444 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004445}
Al Viro4d359502014-03-14 12:20:17 -04004446EXPORT_SYMBOL(generic_readlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004447
Linus Torvalds1da177e2005-04-16 15:20:36 -07004448/* get the link contents into pagecache */
4449static char *page_getlink(struct dentry * dentry, struct page **ppage)
4450{
Duane Griffinebd09ab2008-12-19 20:47:12 +00004451 char *kaddr;
4452 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004453 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07004454 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004455 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07004456 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004457 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00004458 kaddr = kmap(page);
4459 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4460 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004461}
4462
4463int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4464{
4465 struct page *page = NULL;
Al Viro5d826c82014-03-14 13:42:45 -04004466 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004467 if (page) {
4468 kunmap(page);
4469 page_cache_release(page);
4470 }
4471 return res;
4472}
Al Viro4d359502014-03-14 12:20:17 -04004473EXPORT_SYMBOL(page_readlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004474
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004475void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004476{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004477 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004478 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004479 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004480}
Al Viro4d359502014-03-14 12:20:17 -04004481EXPORT_SYMBOL(page_follow_link_light);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004482
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004483void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004484{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004485 struct page *page = cookie;
4486
4487 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004488 kunmap(page);
4489 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004490 }
4491}
Al Viro4d359502014-03-14 12:20:17 -04004492EXPORT_SYMBOL(page_put_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004493
Nick Piggin54566b22009-01-04 12:00:53 -08004494/*
4495 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4496 */
4497int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004498{
4499 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004500 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07004501 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08004502 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004503 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08004504 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4505 if (nofs)
4506 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004507
NeilBrown7e53cac2006-03-25 03:07:57 -08004508retry:
Nick Pigginafddba42007-10-16 01:25:01 -07004509 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08004510 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004511 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07004512 goto fail;
4513
Cong Wange8e3c3d2011-11-25 23:14:27 +08004514 kaddr = kmap_atomic(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004515 memcpy(kaddr, symname, len-1);
Cong Wange8e3c3d2011-11-25 23:14:27 +08004516 kunmap_atomic(kaddr);
Nick Pigginafddba42007-10-16 01:25:01 -07004517
4518 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4519 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004520 if (err < 0)
4521 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07004522 if (err < len-1)
4523 goto retry;
4524
Linus Torvalds1da177e2005-04-16 15:20:36 -07004525 mark_inode_dirty(inode);
4526 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004527fail:
4528 return err;
4529}
Al Viro4d359502014-03-14 12:20:17 -04004530EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004531
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004532int page_symlink(struct inode *inode, const char *symname, int len)
4533{
4534 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08004535 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004536}
Al Viro4d359502014-03-14 12:20:17 -04004537EXPORT_SYMBOL(page_symlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004538
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08004539const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004540 .readlink = generic_readlink,
4541 .follow_link = page_follow_link_light,
4542 .put_link = page_put_link,
4543};
Linus Torvalds1da177e2005-04-16 15:20:36 -07004544EXPORT_SYMBOL(page_symlink_inode_operations);