blob: 70580ab1445c89f8379756fc16e33223aca4ab0d [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>
George Spelvin2a18da7a2016-05-23 07:43:58 -040038#include <linux/bitops.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include <asm/uaccess.h>
40
Eric Parise81e3f42009-12-04 15:47:36 -050041#include "internal.h"
Al Viroc7105362011-11-24 18:22:03 -050042#include "mount.h"
Eric Parise81e3f42009-12-04 15:47:36 -050043
Linus Torvalds1da177e2005-04-16 15:20:36 -070044/* [Feb-1997 T. Schoebel-Theuer]
45 * Fundamental changes in the pathname lookup mechanisms (namei)
46 * were necessary because of omirr. The reason is that omirr needs
47 * to know the _real_ pathname, not the user-supplied one, in case
48 * of symlinks (and also when transname replacements occur).
49 *
50 * The new code replaces the old recursive symlink resolution with
51 * an iterative one (in case of non-nested symlink chains). It does
52 * this with calls to <fs>_follow_link().
53 * As a side effect, dir_namei(), _namei() and follow_link() are now
54 * replaced with a single function lookup_dentry() that can handle all
55 * the special cases of the former code.
56 *
57 * With the new dcache, the pathname is stored at each inode, at least as
58 * long as the refcount of the inode is positive. As a side effect, the
59 * size of the dcache depends on the inode cache and thus is dynamic.
60 *
61 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
62 * resolution to correspond with current state of the code.
63 *
64 * Note that the symlink resolution is not *completely* iterative.
65 * There is still a significant amount of tail- and mid- recursion in
66 * the algorithm. Also, note that <fs>_readlink() is not used in
67 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
68 * may return different results than <fs>_follow_link(). Many virtual
69 * filesystems (including /proc) exhibit this behavior.
70 */
71
72/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
73 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
74 * and the name already exists in form of a symlink, try to create the new
75 * name indicated by the symlink. The old code always complained that the
76 * name already exists, due to not following the symlink even if its target
77 * is nonexistent. The new semantics affects also mknod() and link() when
Lucas De Marchi25985ed2011-03-30 22:57:33 -030078 * the name is a symlink pointing to a non-existent name.
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 *
80 * I don't know which semantics is the right one, since I have no access
81 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
82 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
83 * "old" one. Personally, I think the new semantics is much more logical.
84 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
85 * file does succeed in both HP-UX and SunOs, but not in Solaris
86 * and in the old Linux semantics.
87 */
88
89/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
90 * semantics. See the comments in "open_namei" and "do_link" below.
91 *
92 * [10-Sep-98 Alan Modra] Another symlink change.
93 */
94
95/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
96 * inside the path - always follow.
97 * in the last component in creation/removal/renaming - never follow.
98 * if LOOKUP_FOLLOW passed - follow.
99 * if the pathname has trailing slashes - follow.
100 * otherwise - don't follow.
101 * (applied in that order).
102 *
103 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
104 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
105 * During the 2.4 we need to fix the userland stuff depending on it -
106 * hopefully we will be able to get rid of that wart in 2.5. So far only
107 * XEmacs seems to be relying on it...
108 */
109/*
110 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800111 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 * any extra contention...
113 */
114
115/* In order to reduce some races, while at the same time doing additional
116 * checking and hopefully speeding things up, we copy filenames to the
117 * kernel data space before using them..
118 *
119 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
120 * PATH_MAX includes the nul terminator --RR.
121 */
Jeff Layton91a27b22012-10-10 15:25:28 -0400122
Al Virofd2f7cb2015-02-22 20:07:13 -0500123#define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
Jeff Layton7950e382012-10-10 16:43:13 -0400124
David Drysdale51f39a12014-12-12 16:57:29 -0800125struct filename *
Jeff Layton91a27b22012-10-10 15:25:28 -0400126getname_flags(const char __user *filename, int flags, int *empty)
127{
Al Viro94b5d262015-02-22 19:38:03 -0500128 struct filename *result;
Jeff Layton7950e382012-10-10 16:43:13 -0400129 char *kname;
Al Viro94b5d262015-02-22 19:38:03 -0500130 int len;
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);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139
Jeff Layton7950e382012-10-10 16:43:13 -0400140 /*
141 * First, try to embed the struct filename inside the names_cache
142 * allocation
143 */
Al Virofd2f7cb2015-02-22 20:07:13 -0500144 kname = (char *)result->iname;
Jeff Layton91a27b22012-10-10 15:25:28 -0400145 result->name = kname;
Jeff Layton7950e382012-10-10 16:43:13 -0400146
Al Viro94b5d262015-02-22 19:38:03 -0500147 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
Jeff Layton91a27b22012-10-10 15:25:28 -0400148 if (unlikely(len < 0)) {
Al Viro94b5d262015-02-22 19:38:03 -0500149 __putname(result);
150 return ERR_PTR(len);
Jeff Layton91a27b22012-10-10 15:25:28 -0400151 }
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700152
Jeff Layton7950e382012-10-10 16:43:13 -0400153 /*
154 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
155 * separate struct filename so we can dedicate the entire
156 * names_cache allocation for the pathname, and re-do the copy from
157 * userland.
158 */
Al Viro94b5d262015-02-22 19:38:03 -0500159 if (unlikely(len == EMBEDDED_NAME_MAX)) {
Al Virofd2f7cb2015-02-22 20:07:13 -0500160 const size_t size = offsetof(struct filename, iname[1]);
Jeff Layton7950e382012-10-10 16:43:13 -0400161 kname = (char *)result;
162
Al Virofd2f7cb2015-02-22 20:07:13 -0500163 /*
164 * size is chosen that way we to guarantee that
165 * result->iname[0] is within the same object and that
166 * kname can't be equal to result->iname, no matter what.
167 */
168 result = kzalloc(size, GFP_KERNEL);
Al Viro94b5d262015-02-22 19:38:03 -0500169 if (unlikely(!result)) {
170 __putname(kname);
171 return ERR_PTR(-ENOMEM);
Jeff Layton7950e382012-10-10 16:43:13 -0400172 }
173 result->name = kname;
Al Viro94b5d262015-02-22 19:38:03 -0500174 len = strncpy_from_user(kname, filename, PATH_MAX);
175 if (unlikely(len < 0)) {
176 __putname(kname);
177 kfree(result);
178 return ERR_PTR(len);
179 }
180 if (unlikely(len == PATH_MAX)) {
181 __putname(kname);
182 kfree(result);
183 return ERR_PTR(-ENAMETOOLONG);
184 }
Jeff Layton7950e382012-10-10 16:43:13 -0400185 }
186
Al Viro94b5d262015-02-22 19:38:03 -0500187 result->refcnt = 1;
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700188 /* The empty path is special. */
189 if (unlikely(!len)) {
190 if (empty)
Eric Paris4043cde2012-01-03 14:23:08 -0500191 *empty = 1;
Al Viro94b5d262015-02-22 19:38:03 -0500192 if (!(flags & LOOKUP_EMPTY)) {
193 putname(result);
194 return ERR_PTR(-ENOENT);
195 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 }
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700197
Jeff Layton7950e382012-10-10 16:43:13 -0400198 result->uptr = filename;
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800199 result->aname = NULL;
Jeff Layton7950e382012-10-10 16:43:13 -0400200 audit_getname(result);
201 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202}
203
Jeff Layton91a27b22012-10-10 15:25:28 -0400204struct filename *
205getname(const char __user * filename)
Al Virof52e0c12011-03-14 18:56:51 -0400206{
Linus Torvaldsf7493e52012-03-22 16:10:40 -0700207 return getname_flags(filename, 0, NULL);
Al Virof52e0c12011-03-14 18:56:51 -0400208}
209
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800210struct filename *
211getname_kernel(const char * filename)
212{
213 struct filename *result;
Paul Moore08518542015-01-21 23:59:56 -0500214 int len = strlen(filename) + 1;
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800215
216 result = __getname();
217 if (unlikely(!result))
218 return ERR_PTR(-ENOMEM);
219
Paul Moore08518542015-01-21 23:59:56 -0500220 if (len <= EMBEDDED_NAME_MAX) {
Al Virofd2f7cb2015-02-22 20:07:13 -0500221 result->name = (char *)result->iname;
Paul Moore08518542015-01-21 23:59:56 -0500222 } else if (len <= PATH_MAX) {
223 struct filename *tmp;
224
225 tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
226 if (unlikely(!tmp)) {
227 __putname(result);
228 return ERR_PTR(-ENOMEM);
229 }
230 tmp->name = (char *)result;
Paul Moore08518542015-01-21 23:59:56 -0500231 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
Al Virofd2f7cb2015-02-22 20:07:13 -0500252 if (name->name != name->iname) {
Paul Moore55422d02015-01-22 00:00:23 -0500253 __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... */
Andreas Gruenbacherb8a7a3a2016-03-24 14:38:37 +0100269 if (is_uncached_acl(acl))
Al Viro35678662011-08-02 21:32:13 -0400270 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 Viro894bc8c2015-05-02 07:16:16 -0400496#define EMBEDDED_LEVELS 2
Al Viro1f55a6e2014-11-01 19:30:41 -0400497struct nameidata {
498 struct path path;
Al Viro1cf26652015-05-06 16:01:56 -0400499 struct qstr last;
Al Viro1f55a6e2014-11-01 19:30:41 -0400500 struct path root;
501 struct inode *inode; /* path.dentry.d_inode */
502 unsigned int flags;
Al Viro9883d182015-05-13 07:28:08 -0400503 unsigned seq, m_seq;
Al Viro1f55a6e2014-11-01 19:30:41 -0400504 int last_type;
505 unsigned depth;
NeilBrown756daf22015-03-23 13:37:38 +1100506 int total_link_count;
Al Viro697fc6c2015-05-02 19:38:35 -0400507 struct saved {
508 struct path link;
Al Virofceef392015-12-29 15:58:39 -0500509 struct delayed_call done;
Al Viro697fc6c2015-05-02 19:38:35 -0400510 const char *name;
Al Viro0450b2d2015-05-08 13:23:53 -0400511 unsigned seq;
Al Viro894bc8c2015-05-02 07:16:16 -0400512 } *stack, internal[EMBEDDED_LEVELS];
Al Viro9883d182015-05-13 07:28:08 -0400513 struct filename *name;
514 struct nameidata *saved;
Al Virofceef392015-12-29 15:58:39 -0500515 struct inode *link_inode;
Al Viro9883d182015-05-13 07:28:08 -0400516 unsigned root_seq;
517 int dfd;
Al Viro1f55a6e2014-11-01 19:30:41 -0400518};
519
Al Viro9883d182015-05-13 07:28:08 -0400520static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
Al Viro894bc8c2015-05-02 07:16:16 -0400521{
NeilBrown756daf22015-03-23 13:37:38 +1100522 struct nameidata *old = current->nameidata;
523 p->stack = p->internal;
Al Viroc8a53ee2015-05-12 18:43:07 -0400524 p->dfd = dfd;
525 p->name = name;
NeilBrown756daf22015-03-23 13:37:38 +1100526 p->total_link_count = old ? old->total_link_count : 0;
Al Viro9883d182015-05-13 07:28:08 -0400527 p->saved = old;
NeilBrown756daf22015-03-23 13:37:38 +1100528 current->nameidata = p;
Al Viro894bc8c2015-05-02 07:16:16 -0400529}
530
Al Viro9883d182015-05-13 07:28:08 -0400531static void restore_nameidata(void)
Al Viro894bc8c2015-05-02 07:16:16 -0400532{
Al Viro9883d182015-05-13 07:28:08 -0400533 struct nameidata *now = current->nameidata, *old = now->saved;
NeilBrown756daf22015-03-23 13:37:38 +1100534
535 current->nameidata = old;
536 if (old)
537 old->total_link_count = now->total_link_count;
Al Viroe1a63bb2015-12-05 21:06:33 -0500538 if (now->stack != now->internal)
NeilBrown756daf22015-03-23 13:37:38 +1100539 kfree(now->stack);
Al Viro894bc8c2015-05-02 07:16:16 -0400540}
541
542static int __nd_alloc_stack(struct nameidata *nd)
543{
Al Virobc40aee2015-05-09 13:04:24 -0400544 struct saved *p;
545
546 if (nd->flags & LOOKUP_RCU) {
547 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
548 GFP_ATOMIC);
549 if (unlikely(!p))
550 return -ECHILD;
551 } else {
552 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
Al Viro894bc8c2015-05-02 07:16:16 -0400553 GFP_KERNEL);
Al Virobc40aee2015-05-09 13:04:24 -0400554 if (unlikely(!p))
555 return -ENOMEM;
556 }
Al Viro894bc8c2015-05-02 07:16:16 -0400557 memcpy(p, nd->internal, sizeof(nd->internal));
558 nd->stack = p;
559 return 0;
560}
561
Eric W. Biederman397d4252015-08-15 20:27:13 -0500562/**
563 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
564 * @path: nameidate to verify
565 *
566 * Rename can sometimes move a file or directory outside of a bind
567 * mount, path_connected allows those cases to be detected.
568 */
569static bool path_connected(const struct path *path)
570{
571 struct vfsmount *mnt = path->mnt;
572
573 /* Only bind mounts can have disconnected paths */
574 if (mnt->mnt_root == mnt->mnt_sb->s_root)
575 return true;
576
577 return is_subdir(path->dentry, mnt->mnt_root);
578}
579
Al Viro894bc8c2015-05-02 07:16:16 -0400580static inline int nd_alloc_stack(struct nameidata *nd)
581{
Al Viroda4e0be2015-05-03 20:52:15 -0400582 if (likely(nd->depth != EMBEDDED_LEVELS))
Al Viro894bc8c2015-05-02 07:16:16 -0400583 return 0;
584 if (likely(nd->stack != nd->internal))
585 return 0;
586 return __nd_alloc_stack(nd);
587}
588
Al Viro79733872015-05-09 12:55:43 -0400589static void drop_links(struct nameidata *nd)
590{
591 int i = nd->depth;
592 while (i--) {
593 struct saved *last = nd->stack + i;
Al Virofceef392015-12-29 15:58:39 -0500594 do_delayed_call(&last->done);
595 clear_delayed_call(&last->done);
Al Viro79733872015-05-09 12:55:43 -0400596 }
597}
598
599static void terminate_walk(struct nameidata *nd)
600{
601 drop_links(nd);
602 if (!(nd->flags & LOOKUP_RCU)) {
603 int i;
604 path_put(&nd->path);
605 for (i = 0; i < nd->depth; i++)
606 path_put(&nd->stack[i].link);
Al Viro102b8af2015-05-12 17:35:52 -0400607 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
608 path_put(&nd->root);
609 nd->root.mnt = NULL;
610 }
Al Viro79733872015-05-09 12:55:43 -0400611 } else {
612 nd->flags &= ~LOOKUP_RCU;
613 if (!(nd->flags & LOOKUP_ROOT))
614 nd->root.mnt = NULL;
615 rcu_read_unlock();
616 }
617 nd->depth = 0;
618}
619
620/* path_put is needed afterwards regardless of success or failure */
621static bool legitimize_path(struct nameidata *nd,
622 struct path *path, unsigned seq)
623{
624 int res = __legitimize_mnt(path->mnt, nd->m_seq);
625 if (unlikely(res)) {
626 if (res > 0)
627 path->mnt = NULL;
628 path->dentry = NULL;
629 return false;
630 }
631 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
632 path->dentry = NULL;
633 return false;
634 }
635 return !read_seqcount_retry(&path->dentry->d_seq, seq);
636}
637
638static bool legitimize_links(struct nameidata *nd)
639{
640 int i;
641 for (i = 0; i < nd->depth; i++) {
642 struct saved *last = nd->stack + i;
643 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
644 drop_links(nd);
645 nd->depth = i + 1;
646 return false;
647 }
648 }
649 return true;
650}
651
Al Viro19660af2011-03-25 10:32:48 -0400652/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100653 * Path walking has 2 modes, rcu-walk and ref-walk (see
Al Viro19660af2011-03-25 10:32:48 -0400654 * Documentation/filesystems/path-lookup.txt). In situations when we can't
655 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
Mike Marshall57e37152015-11-30 11:11:59 -0500656 * normal reference counts on dentries and vfsmounts to transition to ref-walk
Al Viro19660af2011-03-25 10:32:48 -0400657 * mode. Refcounts are grabbed at the last known good point before rcu-walk
658 * got stuck, so ref-walk may continue from there. If this is not successful
659 * (eg. a seqcount has changed), then failure is returned and it's up to caller
660 * to restart the path walk from the beginning in ref-walk mode.
Nick Piggin31e6b012011-01-07 17:49:52 +1100661 */
Nick Piggin31e6b012011-01-07 17:49:52 +1100662
663/**
Al Viro19660af2011-03-25 10:32:48 -0400664 * unlazy_walk - try to switch to ref-walk mode.
665 * @nd: nameidata pathwalk data
666 * @dentry: child of nd->path.dentry or NULL
Al Viro6e9918b2015-05-05 09:26:05 -0400667 * @seq: seq number to check dentry against
Randy Dunlap39191622011-01-08 19:36:21 -0800668 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100669 *
Al Viro19660af2011-03-25 10:32:48 -0400670 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
671 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
672 * @nd or NULL. Must be called from rcu-walk context.
Al Viro79733872015-05-09 12:55:43 -0400673 * Nothing should touch nameidata between unlazy_walk() failure and
674 * terminate_walk().
Nick Piggin31e6b012011-01-07 17:49:52 +1100675 */
Al Viro6e9918b2015-05-05 09:26:05 -0400676static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq)
Nick Piggin31e6b012011-01-07 17:49:52 +1100677{
Nick Piggin31e6b012011-01-07 17:49:52 +1100678 struct dentry *parent = nd->path.dentry;
679
680 BUG_ON(!(nd->flags & LOOKUP_RCU));
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700681
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700682 nd->flags &= ~LOOKUP_RCU;
Al Viro79733872015-05-09 12:55:43 -0400683 if (unlikely(!legitimize_links(nd)))
684 goto out2;
685 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
686 goto out2;
687 if (unlikely(!lockref_get_not_dead(&parent->d_lockref)))
688 goto out1;
Al Viro48a066e2013-09-29 22:06:07 -0400689
Linus Torvalds15570082013-09-02 11:38:06 -0700690 /*
691 * For a negative lookup, the lookup sequence point is the parents
692 * sequence point, and it only needs to revalidate the parent dentry.
693 *
694 * For a positive lookup, we need to move both the parent and the
695 * dentry from the RCU domain to be properly refcounted. And the
696 * sequence number in the dentry validates *both* dentry counters,
697 * since we checked the sequence number of the parent after we got
698 * the child sequence number. So we know the parent must still
699 * be valid if the child sequence number is still valid.
700 */
Al Viro19660af2011-03-25 10:32:48 -0400701 if (!dentry) {
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700702 if (read_seqcount_retry(&parent->d_seq, nd->seq))
703 goto out;
Al Viro19660af2011-03-25 10:32:48 -0400704 BUG_ON(nd->inode != parent->d_inode);
705 } else {
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700706 if (!lockref_get_not_dead(&dentry->d_lockref))
707 goto out;
Al Viro6e9918b2015-05-05 09:26:05 -0400708 if (read_seqcount_retry(&dentry->d_seq, seq))
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700709 goto drop_dentry;
Al Viro19660af2011-03-25 10:32:48 -0400710 }
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700711
712 /*
713 * Sequence counts matched. Now make sure that the root is
714 * still valid and get it if required.
715 */
716 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
Al Viro5a8d87e2015-05-12 00:10:18 -0400717 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
718 rcu_read_unlock();
719 dput(dentry);
720 return -ECHILD;
Al Viro79733872015-05-09 12:55:43 -0400721 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100722 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100723
Al Viro8b61e742013-11-08 12:45:01 -0500724 rcu_read_unlock();
Nick Piggin31e6b012011-01-07 17:49:52 +1100725 return 0;
Al Viro19660af2011-03-25 10:32:48 -0400726
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700727drop_dentry:
Al Viro8b61e742013-11-08 12:45:01 -0500728 rcu_read_unlock();
Linus Torvalds15570082013-09-02 11:38:06 -0700729 dput(dentry);
Linus Torvaldsd0d27272013-09-10 12:17:49 -0700730 goto drop_root_mnt;
Al Viro79733872015-05-09 12:55:43 -0400731out2:
732 nd->path.mnt = NULL;
733out1:
734 nd->path.dentry = NULL;
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700735out:
Al Viro8b61e742013-11-08 12:45:01 -0500736 rcu_read_unlock();
Linus Torvaldsd0d27272013-09-10 12:17:49 -0700737drop_root_mnt:
738 if (!(nd->flags & LOOKUP_ROOT))
739 nd->root.mnt = NULL;
Nick Piggin31e6b012011-01-07 17:49:52 +1100740 return -ECHILD;
741}
742
Al Viro79733872015-05-09 12:55:43 -0400743static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq)
744{
745 if (unlikely(!legitimize_path(nd, link, seq))) {
746 drop_links(nd);
747 nd->depth = 0;
748 nd->flags &= ~LOOKUP_RCU;
749 nd->path.mnt = NULL;
750 nd->path.dentry = NULL;
751 if (!(nd->flags & LOOKUP_ROOT))
752 nd->root.mnt = NULL;
753 rcu_read_unlock();
754 } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) {
755 return 0;
756 }
757 path_put(link);
758 return -ECHILD;
759}
760
Al Viro4ce16ef32012-06-10 16:10:59 -0400761static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
Nick Piggin34286d62011-01-07 17:49:57 +1100762{
Al Viro4ce16ef32012-06-10 16:10:59 -0400763 return dentry->d_op->d_revalidate(dentry, flags);
Nick Piggin34286d62011-01-07 17:49:57 +1100764}
765
Al Viro9f1fafe2011-03-25 11:00:12 -0400766/**
767 * complete_walk - successful completion of path walk
768 * @nd: pointer nameidata
Jeff Layton39159de2009-12-07 12:01:50 -0500769 *
Al Viro9f1fafe2011-03-25 11:00:12 -0400770 * If we had been in RCU mode, drop out of it and legitimize nd->path.
771 * Revalidate the final result, unless we'd already done that during
772 * the path walk or the filesystem doesn't ask for it. Return 0 on
773 * success, -error on failure. In case of failure caller does not
774 * need to drop nd->path.
Jeff Layton39159de2009-12-07 12:01:50 -0500775 */
Al Viro9f1fafe2011-03-25 11:00:12 -0400776static int complete_walk(struct nameidata *nd)
Jeff Layton39159de2009-12-07 12:01:50 -0500777{
Al Viro16c2cd72011-02-22 15:50:10 -0500778 struct dentry *dentry = nd->path.dentry;
Jeff Layton39159de2009-12-07 12:01:50 -0500779 int status;
Jeff Layton39159de2009-12-07 12:01:50 -0500780
Al Viro9f1fafe2011-03-25 11:00:12 -0400781 if (nd->flags & LOOKUP_RCU) {
Al Viro9f1fafe2011-03-25 11:00:12 -0400782 if (!(nd->flags & LOOKUP_ROOT))
783 nd->root.mnt = NULL;
Al Viro6e9918b2015-05-05 09:26:05 -0400784 if (unlikely(unlazy_walk(nd, NULL, 0)))
Al Viro48a066e2013-09-29 22:06:07 -0400785 return -ECHILD;
Al Viro9f1fafe2011-03-25 11:00:12 -0400786 }
787
Al Viro16c2cd72011-02-22 15:50:10 -0500788 if (likely(!(nd->flags & LOOKUP_JUMPED)))
Jeff Layton39159de2009-12-07 12:01:50 -0500789 return 0;
790
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500791 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
Al Viro16c2cd72011-02-22 15:50:10 -0500792 return 0;
793
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500794 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
Jeff Layton39159de2009-12-07 12:01:50 -0500795 if (status > 0)
796 return 0;
797
Al Viro16c2cd72011-02-22 15:50:10 -0500798 if (!status)
Jeff Layton39159de2009-12-07 12:01:50 -0500799 status = -ESTALE;
Al Viro16c2cd72011-02-22 15:50:10 -0500800
Jeff Layton39159de2009-12-07 12:01:50 -0500801 return status;
802}
803
Al Viro18d8c862015-05-12 16:32:34 -0400804static void set_root(struct nameidata *nd)
Al Viro2a737872009-04-07 11:49:53 -0400805{
Al Viro7bd88372014-09-13 21:55:46 -0400806 struct fs_struct *fs = current->fs;
Nick Pigginc28cc362011-01-07 17:49:53 +1100807
Al Viro9e6697e2015-12-05 20:07:21 -0500808 if (nd->flags & LOOKUP_RCU) {
809 unsigned seq;
810
811 do {
812 seq = read_seqcount_begin(&fs->seq);
813 nd->root = fs->root;
814 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
815 } while (read_seqcount_retry(&fs->seq, seq));
816 } else {
817 get_fs_root(fs, &nd->root);
818 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100819}
820
Jan Blunck1d957f92008-02-14 19:34:35 -0800821static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700822{
823 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800824 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700825 mntput(path->mnt);
826}
827
Nick Piggin7b9337a2011-01-14 08:42:43 +0000828static inline void path_to_nameidata(const struct path *path,
829 struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700830{
Nick Piggin31e6b012011-01-07 17:49:52 +1100831 if (!(nd->flags & LOOKUP_RCU)) {
832 dput(nd->path.dentry);
833 if (nd->path.mnt != path->mnt)
834 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800835 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100836 nd->path.mnt = path->mnt;
Jan Blunck4ac91372008-02-14 19:34:32 -0800837 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700838}
839
Al Viro248fb5b2015-12-05 20:51:58 -0500840static int nd_jump_root(struct nameidata *nd)
841{
842 if (nd->flags & LOOKUP_RCU) {
843 struct dentry *d;
844 nd->path = nd->root;
845 d = nd->path.dentry;
846 nd->inode = d->d_inode;
847 nd->seq = nd->root_seq;
848 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
849 return -ECHILD;
850 } else {
851 path_put(&nd->path);
852 nd->path = nd->root;
853 path_get(&nd->path);
854 nd->inode = nd->path.dentry->d_inode;
855 }
856 nd->flags |= LOOKUP_JUMPED;
857 return 0;
858}
859
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400860/*
Al Viro6b255392015-11-17 10:20:54 -0500861 * Helper to directly jump to a known parsed path from ->get_link,
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400862 * caller must have taken a reference to path beforehand.
863 */
Al Viro6e771372015-05-02 13:37:52 -0400864void nd_jump_link(struct path *path)
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400865{
Al Viro6e771372015-05-02 13:37:52 -0400866 struct nameidata *nd = current->nameidata;
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400867 path_put(&nd->path);
868
869 nd->path = *path;
870 nd->inode = nd->path.dentry->d_inode;
871 nd->flags |= LOOKUP_JUMPED;
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400872}
873
Al Virob9ff4422015-05-02 20:19:23 -0400874static inline void put_link(struct nameidata *nd)
Al Viro574197e2011-03-14 22:20:34 -0400875{
Al Viro21c30032015-05-03 21:06:24 -0400876 struct saved *last = nd->stack + --nd->depth;
Al Virofceef392015-12-29 15:58:39 -0500877 do_delayed_call(&last->done);
Al Viro6548fae2015-05-07 20:32:22 -0400878 if (!(nd->flags & LOOKUP_RCU))
879 path_put(&last->link);
Al Viro574197e2011-03-14 22:20:34 -0400880}
881
Linus Torvalds561ec642012-10-26 10:05:07 -0700882int sysctl_protected_symlinks __read_mostly = 0;
883int sysctl_protected_hardlinks __read_mostly = 0;
Kees Cook800179c2012-07-25 17:29:07 -0700884
885/**
886 * may_follow_link - Check symlink following for unsafe situations
Randy Dunlap55852632012-08-18 17:39:25 -0700887 * @nd: nameidata pathwalk data
Kees Cook800179c2012-07-25 17:29:07 -0700888 *
889 * In the case of the sysctl_protected_symlinks sysctl being enabled,
890 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
891 * in a sticky world-writable directory. This is to protect privileged
892 * processes from failing races against path names that may change out
893 * from under them by way of other users creating malicious symlinks.
894 * It will permit symlinks to be followed only when outside a sticky
895 * world-writable directory, or when the uid of the symlink and follower
896 * match, or when the directory owner matches the symlink's owner.
897 *
898 * Returns 0 if following the symlink is allowed, -ve on error.
899 */
Al Virofec2fa22015-05-06 15:58:18 -0400900static inline int may_follow_link(struct nameidata *nd)
Kees Cook800179c2012-07-25 17:29:07 -0700901{
902 const struct inode *inode;
903 const struct inode *parent;
904
905 if (!sysctl_protected_symlinks)
906 return 0;
907
908 /* Allowed if owner and follower match. */
Al Virofceef392015-12-29 15:58:39 -0500909 inode = nd->link_inode;
Eric W. Biederman81abe272012-08-03 09:38:08 -0700910 if (uid_eq(current_cred()->fsuid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700911 return 0;
912
913 /* Allowed if parent directory not sticky and world-writable. */
Al Viroaa65fa32015-08-04 23:23:50 -0400914 parent = nd->inode;
Kees Cook800179c2012-07-25 17:29:07 -0700915 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
916 return 0;
917
918 /* Allowed if parent directory and link owner match. */
Eric W. Biederman81abe272012-08-03 09:38:08 -0700919 if (uid_eq(parent->i_uid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700920 return 0;
921
Al Viro319565022015-05-07 20:37:40 -0400922 if (nd->flags & LOOKUP_RCU)
923 return -ECHILD;
924
Al Viro1cf26652015-05-06 16:01:56 -0400925 audit_log_link_denied("follow_link", &nd->stack[0].link);
Kees Cook800179c2012-07-25 17:29:07 -0700926 return -EACCES;
927}
928
929/**
930 * safe_hardlink_source - Check for safe hardlink conditions
931 * @inode: the source inode to hardlink from
932 *
933 * Return false if at least one of the following conditions:
934 * - inode is not a regular file
935 * - inode is setuid
936 * - inode is setgid and group-exec
937 * - access failure for read and write
938 *
939 * Otherwise returns true.
940 */
941static bool safe_hardlink_source(struct inode *inode)
942{
943 umode_t mode = inode->i_mode;
944
945 /* Special files should not get pinned to the filesystem. */
946 if (!S_ISREG(mode))
947 return false;
948
949 /* Setuid files should not get pinned to the filesystem. */
950 if (mode & S_ISUID)
951 return false;
952
953 /* Executable setgid files should not get pinned to the filesystem. */
954 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
955 return false;
956
957 /* Hardlinking to unreadable or unwritable sources is dangerous. */
958 if (inode_permission(inode, MAY_READ | MAY_WRITE))
959 return false;
960
961 return true;
962}
963
964/**
965 * may_linkat - Check permissions for creating a hardlink
966 * @link: the source to hardlink from
967 *
968 * Block hardlink when all of:
969 * - sysctl_protected_hardlinks enabled
970 * - fsuid does not match inode
971 * - hardlink source is unsafe (see safe_hardlink_source() above)
Dirk Steinmetzf2ca3792015-10-20 16:09:19 +0200972 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
Kees Cook800179c2012-07-25 17:29:07 -0700973 *
974 * Returns 0 if successful, -ve on error.
975 */
976static int may_linkat(struct path *link)
977{
Kees Cook800179c2012-07-25 17:29:07 -0700978 struct inode *inode;
979
980 if (!sysctl_protected_hardlinks)
981 return 0;
982
Kees Cook800179c2012-07-25 17:29:07 -0700983 inode = link->dentry->d_inode;
984
985 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
986 * otherwise, it must be a safe source.
987 */
Dirk Steinmetzf2ca3792015-10-20 16:09:19 +0200988 if (inode_owner_or_capable(inode) || safe_hardlink_source(inode))
Kees Cook800179c2012-07-25 17:29:07 -0700989 return 0;
990
Kees Cooka51d9ea2012-07-25 17:29:08 -0700991 audit_log_link_denied("linkat", link);
Kees Cook800179c2012-07-25 17:29:07 -0700992 return -EPERM;
993}
994
Al Viro3b2e7f72015-04-19 00:53:50 -0400995static __always_inline
996const char *get_link(struct nameidata *nd)
Ian Kent051d3812006-03-27 01:14:53 -0800997{
Al Viroab104922015-05-10 11:50:01 -0400998 struct saved *last = nd->stack + nd->depth - 1;
Al Viro1cf26652015-05-06 16:01:56 -0400999 struct dentry *dentry = last->link.dentry;
Al Virofceef392015-12-29 15:58:39 -05001000 struct inode *inode = nd->link_inode;
Al Viro6d7b5aa2012-06-10 04:15:17 -04001001 int error;
Al Viro0a959df2015-04-18 18:23:41 -04001002 const char *res;
Ian Kent051d3812006-03-27 01:14:53 -08001003
NeilBrown8fa9dd22015-03-23 13:37:40 +11001004 if (!(nd->flags & LOOKUP_RCU)) {
1005 touch_atime(&last->link);
1006 cond_resched();
1007 } else if (atime_needs_update(&last->link, inode)) {
1008 if (unlikely(unlazy_walk(nd, NULL, 0)))
1009 return ERR_PTR(-ECHILD);
1010 touch_atime(&last->link);
1011 }
1012
NeilBrownbda0be72015-03-23 13:37:39 +11001013 error = security_inode_follow_link(dentry, inode,
1014 nd->flags & LOOKUP_RCU);
1015 if (unlikely(error))
Al Viro6920a442015-05-10 10:43:46 -04001016 return ERR_PTR(error);
Al Viro36f3b4f2011-02-22 21:24:38 -05001017
Al Viro86acdca12009-12-22 23:45:11 -05001018 nd->last_type = LAST_BIND;
Al Virod4dee482015-04-30 20:08:02 -04001019 res = inode->i_link;
1020 if (!res) {
Al Virofceef392015-12-29 15:58:39 -05001021 const char * (*get)(struct dentry *, struct inode *,
1022 struct delayed_call *);
1023 get = inode->i_op->get_link;
Al Viro8c1b4562015-05-09 18:15:21 -04001024 if (nd->flags & LOOKUP_RCU) {
Al Virofceef392015-12-29 15:58:39 -05001025 res = get(NULL, inode, &last->done);
Al Viro6b255392015-11-17 10:20:54 -05001026 if (res == ERR_PTR(-ECHILD)) {
1027 if (unlikely(unlazy_walk(nd, NULL, 0)))
1028 return ERR_PTR(-ECHILD);
Al Virofceef392015-12-29 15:58:39 -05001029 res = get(dentry, inode, &last->done);
Al Viro6b255392015-11-17 10:20:54 -05001030 }
1031 } else {
Al Virofceef392015-12-29 15:58:39 -05001032 res = get(dentry, inode, &last->done);
Al Viro8c1b4562015-05-09 18:15:21 -04001033 }
Al Virofceef392015-12-29 15:58:39 -05001034 if (IS_ERR_OR_NULL(res))
Al Virofab51e82015-05-10 11:01:00 -04001035 return res;
Ian Kent051d3812006-03-27 01:14:53 -08001036 }
Al Virofab51e82015-05-10 11:01:00 -04001037 if (*res == '/') {
Al Viro9e6697e2015-12-05 20:07:21 -05001038 if (!nd->root.mnt)
1039 set_root(nd);
Al Viro248fb5b2015-12-05 20:51:58 -05001040 if (unlikely(nd_jump_root(nd)))
1041 return ERR_PTR(-ECHILD);
Al Virofab51e82015-05-10 11:01:00 -04001042 while (unlikely(*++res == '/'))
1043 ;
1044 }
1045 if (!*res)
1046 res = NULL;
Al Viro0a959df2015-04-18 18:23:41 -04001047 return res;
1048}
Al Viro6d7b5aa2012-06-10 04:15:17 -04001049
David Howellsf015f1262012-06-25 12:55:28 +01001050/*
1051 * follow_up - Find the mountpoint of path's vfsmount
1052 *
1053 * Given a path, find the mountpoint of its source file system.
1054 * Replace @path with the path of the mountpoint in the parent mount.
1055 * Up is towards /.
1056 *
1057 * Return 1 if we went up a level and 0 if we were already at the
1058 * root.
1059 */
Al Virobab77eb2009-04-18 03:26:48 -04001060int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061{
Al Viro0714a532011-11-24 22:19:58 -05001062 struct mount *mnt = real_mount(path->mnt);
1063 struct mount *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +10001065
Al Viro48a066e2013-09-29 22:06:07 -04001066 read_seqlock_excl(&mount_lock);
Al Viro0714a532011-11-24 22:19:58 -05001067 parent = mnt->mnt_parent;
Al Viro3c0a6162012-07-18 17:32:50 +04001068 if (parent == mnt) {
Al Viro48a066e2013-09-29 22:06:07 -04001069 read_sequnlock_excl(&mount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 return 0;
1071 }
Al Viro0714a532011-11-24 22:19:58 -05001072 mntget(&parent->mnt);
Al Viroa73324d2011-11-24 22:25:07 -05001073 mountpoint = dget(mnt->mnt_mountpoint);
Al Viro48a066e2013-09-29 22:06:07 -04001074 read_sequnlock_excl(&mount_lock);
Al Virobab77eb2009-04-18 03:26:48 -04001075 dput(path->dentry);
1076 path->dentry = mountpoint;
1077 mntput(path->mnt);
Al Viro0714a532011-11-24 22:19:58 -05001078 path->mnt = &parent->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 return 1;
1080}
Al Viro4d359502014-03-14 12:20:17 -04001081EXPORT_SYMBOL(follow_up);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001083/*
David Howells9875cf82011-01-14 18:45:21 +00001084 * Perform an automount
1085 * - return -EISDIR to tell follow_managed() to stop and return the path we
1086 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 */
NeilBrown756daf22015-03-23 13:37:38 +11001088static int follow_automount(struct path *path, struct nameidata *nd,
David Howells9875cf82011-01-14 18:45:21 +00001089 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +11001090{
David Howells9875cf82011-01-14 18:45:21 +00001091 struct vfsmount *mnt;
David Howellsea5b7782011-01-14 19:10:03 +00001092 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001093
David Howells9875cf82011-01-14 18:45:21 +00001094 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1095 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -07001096
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +02001097 /* We don't want to mount if someone's just doing a stat -
1098 * unless they're stat'ing a directory and appended a '/' to
1099 * the name.
1100 *
1101 * We do, however, want to mount if someone wants to open or
1102 * create a file of any type under the mountpoint, wants to
1103 * traverse through the mountpoint or wants to open the
1104 * mounted directory. Also, autofs may mark negative dentries
1105 * as being automount points. These will need the attentions
1106 * of the daemon to instantiate them before they can be used.
David Howells9875cf82011-01-14 18:45:21 +00001107 */
NeilBrown756daf22015-03-23 13:37:38 +11001108 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1109 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +02001110 path->dentry->d_inode)
1111 return -EISDIR;
1112
NeilBrown756daf22015-03-23 13:37:38 +11001113 nd->total_link_count++;
1114 if (nd->total_link_count >= 40)
David Howells9875cf82011-01-14 18:45:21 +00001115 return -ELOOP;
1116
1117 mnt = path->dentry->d_op->d_automount(path);
1118 if (IS_ERR(mnt)) {
1119 /*
1120 * The filesystem is allowed to return -EISDIR here to indicate
1121 * it doesn't want to automount. For instance, autofs would do
1122 * this so that its userspace daemon can mount on this dentry.
1123 *
1124 * However, we can only permit this if it's a terminal point in
1125 * the path being looked up; if it wasn't then the remainder of
1126 * the path is inaccessible and we should say so.
1127 */
NeilBrown756daf22015-03-23 13:37:38 +11001128 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
David Howells9875cf82011-01-14 18:45:21 +00001129 return -EREMOTE;
1130 return PTR_ERR(mnt);
1131 }
David Howellsea5b7782011-01-14 19:10:03 +00001132
David Howells9875cf82011-01-14 18:45:21 +00001133 if (!mnt) /* mount collision */
1134 return 0;
1135
Al Viro8aef1882011-06-16 15:10:06 +01001136 if (!*need_mntput) {
1137 /* lock_mount() may release path->mnt on error */
1138 mntget(path->mnt);
1139 *need_mntput = true;
1140 }
Al Viro19a167a2011-01-17 01:35:23 -05001141 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +00001142
David Howellsea5b7782011-01-14 19:10:03 +00001143 switch (err) {
1144 case -EBUSY:
1145 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -05001146 return 0;
David Howellsea5b7782011-01-14 19:10:03 +00001147 case 0:
Al Viro8aef1882011-06-16 15:10:06 +01001148 path_put(path);
David Howellsea5b7782011-01-14 19:10:03 +00001149 path->mnt = mnt;
1150 path->dentry = dget(mnt->mnt_root);
David Howellsea5b7782011-01-14 19:10:03 +00001151 return 0;
Al Viro19a167a2011-01-17 01:35:23 -05001152 default:
1153 return err;
David Howellsea5b7782011-01-14 19:10:03 +00001154 }
Al Viro19a167a2011-01-17 01:35:23 -05001155
David Howells9875cf82011-01-14 18:45:21 +00001156}
1157
1158/*
1159 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +00001160 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +00001161 * - Flagged as mountpoint
1162 * - Flagged as automount point
1163 *
1164 * This may only be called in refwalk mode.
1165 *
1166 * Serialization is taken care of in namespace.c
1167 */
NeilBrown756daf22015-03-23 13:37:38 +11001168static int follow_managed(struct path *path, struct nameidata *nd)
David Howells9875cf82011-01-14 18:45:21 +00001169{
Al Viro8aef1882011-06-16 15:10:06 +01001170 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
David Howells9875cf82011-01-14 18:45:21 +00001171 unsigned managed;
1172 bool need_mntput = false;
Al Viro8aef1882011-06-16 15:10:06 +01001173 int ret = 0;
David Howells9875cf82011-01-14 18:45:21 +00001174
1175 /* Given that we're not holding a lock here, we retain the value in a
1176 * local variable for each dentry as we look at it so that we don't see
1177 * the components of that value change under us */
1178 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1179 managed &= DCACHE_MANAGED_DENTRY,
1180 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +00001181 /* Allow the filesystem to manage the transit without i_mutex
1182 * being held. */
1183 if (managed & DCACHE_MANAGE_TRANSIT) {
1184 BUG_ON(!path->dentry->d_op);
1185 BUG_ON(!path->dentry->d_op->d_manage);
Al Viro1aed3e42011-03-18 09:09:02 -04001186 ret = path->dentry->d_op->d_manage(path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001187 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001188 break;
David Howellscc53ce52011-01-14 18:45:26 +00001189 }
1190
David Howells9875cf82011-01-14 18:45:21 +00001191 /* Transit to a mounted filesystem. */
1192 if (managed & DCACHE_MOUNTED) {
1193 struct vfsmount *mounted = lookup_mnt(path);
1194 if (mounted) {
1195 dput(path->dentry);
1196 if (need_mntput)
1197 mntput(path->mnt);
1198 path->mnt = mounted;
1199 path->dentry = dget(mounted->mnt_root);
1200 need_mntput = true;
1201 continue;
1202 }
1203
1204 /* Something is mounted on this dentry in another
1205 * namespace and/or whatever was mounted there in this
Al Viro48a066e2013-09-29 22:06:07 -04001206 * namespace got unmounted before lookup_mnt() could
1207 * get it */
David Howells9875cf82011-01-14 18:45:21 +00001208 }
1209
1210 /* Handle an automount point */
1211 if (managed & DCACHE_NEED_AUTOMOUNT) {
NeilBrown756daf22015-03-23 13:37:38 +11001212 ret = follow_automount(path, nd, &need_mntput);
David Howells9875cf82011-01-14 18:45:21 +00001213 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001214 break;
David Howells9875cf82011-01-14 18:45:21 +00001215 continue;
1216 }
1217
1218 /* We didn't change the current path point */
1219 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 }
Al Viro8aef1882011-06-16 15:10:06 +01001221
1222 if (need_mntput && path->mnt == mnt)
1223 mntput(path->mnt);
Al Viroe9742b52016-03-05 22:04:59 -05001224 if (ret == -EISDIR || !ret)
1225 ret = 1;
Al Viro84027522015-04-22 10:30:08 -04001226 if (need_mntput)
1227 nd->flags |= LOOKUP_JUMPED;
1228 if (unlikely(ret < 0))
1229 path_put_conditional(path, nd);
1230 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231}
1232
David Howellscc53ce52011-01-14 18:45:26 +00001233int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234{
1235 struct vfsmount *mounted;
1236
Al Viro1c755af2009-04-18 14:06:57 -04001237 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -04001239 dput(path->dentry);
1240 mntput(path->mnt);
1241 path->mnt = mounted;
1242 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 return 1;
1244 }
1245 return 0;
1246}
Al Viro4d359502014-03-14 12:20:17 -04001247EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248
NeilBrownb8faf032014-08-04 17:06:29 +10001249static inline int managed_dentry_rcu(struct dentry *dentry)
Ian Kent62a73752011-03-25 01:51:02 +08001250{
NeilBrownb8faf032014-08-04 17:06:29 +10001251 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1252 dentry->d_op->d_manage(dentry, true) : 0;
Ian Kent62a73752011-03-25 01:51:02 +08001253}
1254
David Howells9875cf82011-01-14 18:45:21 +00001255/*
Al Viro287548e2011-05-27 06:50:06 -04001256 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1257 * we meet a managed dentry that would need blocking.
David Howells9875cf82011-01-14 18:45:21 +00001258 */
1259static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
Al Viro254cf582015-05-05 09:40:46 -04001260 struct inode **inode, unsigned *seqp)
David Howells9875cf82011-01-14 18:45:21 +00001261{
Ian Kent62a73752011-03-25 01:51:02 +08001262 for (;;) {
Al Viroc7105362011-11-24 18:22:03 -05001263 struct mount *mounted;
Ian Kent62a73752011-03-25 01:51:02 +08001264 /*
1265 * Don't forget we might have a non-mountpoint managed dentry
1266 * that wants to block transit.
1267 */
NeilBrownb8faf032014-08-04 17:06:29 +10001268 switch (managed_dentry_rcu(path->dentry)) {
1269 case -ECHILD:
1270 default:
David Howellsab909112011-01-14 18:46:51 +00001271 return false;
NeilBrownb8faf032014-08-04 17:06:29 +10001272 case -EISDIR:
1273 return true;
1274 case 0:
1275 break;
1276 }
Ian Kent62a73752011-03-25 01:51:02 +08001277
1278 if (!d_mountpoint(path->dentry))
NeilBrownb8faf032014-08-04 17:06:29 +10001279 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
Ian Kent62a73752011-03-25 01:51:02 +08001280
Al Viro474279d2013-10-01 16:11:26 -04001281 mounted = __lookup_mnt(path->mnt, path->dentry);
David Howells9875cf82011-01-14 18:45:21 +00001282 if (!mounted)
1283 break;
Al Viroc7105362011-11-24 18:22:03 -05001284 path->mnt = &mounted->mnt;
1285 path->dentry = mounted->mnt.mnt_root;
Al Viroa3fbbde2011-11-07 21:21:26 +00001286 nd->flags |= LOOKUP_JUMPED;
Al Viro254cf582015-05-05 09:40:46 -04001287 *seqp = read_seqcount_begin(&path->dentry->d_seq);
Linus Torvalds59430262011-07-18 15:43:29 -07001288 /*
1289 * Update the inode too. We don't need to re-check the
1290 * dentry sequence number here after this d_inode read,
1291 * because a mount-point is always pinned.
1292 */
1293 *inode = path->dentry->d_inode;
David Howells9875cf82011-01-14 18:45:21 +00001294 }
Al Virof5be3e29122014-09-13 21:50:45 -04001295 return !read_seqretry(&mount_lock, nd->m_seq) &&
NeilBrownb8faf032014-08-04 17:06:29 +10001296 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
Al Viro287548e2011-05-27 06:50:06 -04001297}
1298
Nick Piggin31e6b012011-01-07 17:49:52 +11001299static int follow_dotdot_rcu(struct nameidata *nd)
1300{
Al Viro4023bfc2014-09-13 21:59:43 -04001301 struct inode *inode = nd->inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001302
David Howells9875cf82011-01-14 18:45:21 +00001303 while (1) {
Al Viroaed434a2015-05-12 12:22:47 -04001304 if (path_equal(&nd->path, &nd->root))
Nick Piggin31e6b012011-01-07 17:49:52 +11001305 break;
Nick Piggin31e6b012011-01-07 17:49:52 +11001306 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1307 struct dentry *old = nd->path.dentry;
1308 struct dentry *parent = old->d_parent;
1309 unsigned seq;
1310
Al Viro4023bfc2014-09-13 21:59:43 -04001311 inode = parent->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001312 seq = read_seqcount_begin(&parent->d_seq);
Al Viroaed434a2015-05-12 12:22:47 -04001313 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1314 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001315 nd->path.dentry = parent;
1316 nd->seq = seq;
Eric W. Biederman397d4252015-08-15 20:27:13 -05001317 if (unlikely(!path_connected(&nd->path)))
1318 return -ENOENT;
Nick Piggin31e6b012011-01-07 17:49:52 +11001319 break;
Al Viroaed434a2015-05-12 12:22:47 -04001320 } else {
1321 struct mount *mnt = real_mount(nd->path.mnt);
1322 struct mount *mparent = mnt->mnt_parent;
1323 struct dentry *mountpoint = mnt->mnt_mountpoint;
1324 struct inode *inode2 = mountpoint->d_inode;
1325 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1326 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1327 return -ECHILD;
1328 if (&mparent->mnt == nd->path.mnt)
1329 break;
1330 /* we know that mountpoint was pinned */
1331 nd->path.dentry = mountpoint;
1332 nd->path.mnt = &mparent->mnt;
1333 inode = inode2;
1334 nd->seq = seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11001335 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001336 }
Al Viroaed434a2015-05-12 12:22:47 -04001337 while (unlikely(d_mountpoint(nd->path.dentry))) {
Al Virob37199e2014-03-20 15:18:22 -04001338 struct mount *mounted;
1339 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
Al Viroaed434a2015-05-12 12:22:47 -04001340 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1341 return -ECHILD;
Al Virob37199e2014-03-20 15:18:22 -04001342 if (!mounted)
1343 break;
1344 nd->path.mnt = &mounted->mnt;
1345 nd->path.dentry = mounted->mnt.mnt_root;
Al Viro4023bfc2014-09-13 21:59:43 -04001346 inode = nd->path.dentry->d_inode;
Al Virob37199e2014-03-20 15:18:22 -04001347 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Virob37199e2014-03-20 15:18:22 -04001348 }
Al Viro4023bfc2014-09-13 21:59:43 -04001349 nd->inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001350 return 0;
1351}
1352
David Howells9875cf82011-01-14 18:45:21 +00001353/*
David Howellscc53ce52011-01-14 18:45:26 +00001354 * Follow down to the covering mount currently visible to userspace. At each
1355 * point, the filesystem owning that dentry may be queried as to whether the
1356 * caller is permitted to proceed or not.
David Howellscc53ce52011-01-14 18:45:26 +00001357 */
Al Viro7cc90cc2011-03-18 09:04:20 -04001358int follow_down(struct path *path)
David Howellscc53ce52011-01-14 18:45:26 +00001359{
1360 unsigned managed;
1361 int ret;
1362
1363 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1364 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1365 /* Allow the filesystem to manage the transit without i_mutex
1366 * being held.
1367 *
1368 * We indicate to the filesystem if someone is trying to mount
1369 * something here. This gives autofs the chance to deny anyone
1370 * other than its daemon the right to mount on its
1371 * superstructure.
1372 *
1373 * The filesystem may sleep at this point.
1374 */
1375 if (managed & DCACHE_MANAGE_TRANSIT) {
1376 BUG_ON(!path->dentry->d_op);
1377 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +00001378 ret = path->dentry->d_op->d_manage(
Al Viro1aed3e42011-03-18 09:09:02 -04001379 path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001380 if (ret < 0)
1381 return ret == -EISDIR ? 0 : ret;
1382 }
1383
1384 /* Transit to a mounted filesystem. */
1385 if (managed & DCACHE_MOUNTED) {
1386 struct vfsmount *mounted = lookup_mnt(path);
1387 if (!mounted)
1388 break;
1389 dput(path->dentry);
1390 mntput(path->mnt);
1391 path->mnt = mounted;
1392 path->dentry = dget(mounted->mnt_root);
1393 continue;
1394 }
1395
1396 /* Don't handle automount points here */
1397 break;
1398 }
1399 return 0;
1400}
Al Viro4d359502014-03-14 12:20:17 -04001401EXPORT_SYMBOL(follow_down);
David Howellscc53ce52011-01-14 18:45:26 +00001402
1403/*
David Howells9875cf82011-01-14 18:45:21 +00001404 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1405 */
1406static void follow_mount(struct path *path)
1407{
1408 while (d_mountpoint(path->dentry)) {
1409 struct vfsmount *mounted = lookup_mnt(path);
1410 if (!mounted)
1411 break;
1412 dput(path->dentry);
1413 mntput(path->mnt);
1414 path->mnt = mounted;
1415 path->dentry = dget(mounted->mnt_root);
1416 }
1417}
1418
Eric W. Biedermaneedf2652016-06-02 10:29:47 -05001419static int path_parent_directory(struct path *path)
1420{
1421 struct dentry *old = path->dentry;
1422 /* rare case of legitimate dget_parent()... */
1423 path->dentry = dget_parent(path->dentry);
1424 dput(old);
1425 if (unlikely(!path_connected(path)))
1426 return -ENOENT;
1427 return 0;
1428}
1429
Eric W. Biederman397d4252015-08-15 20:27:13 -05001430static int follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431{
1432 while(1) {
Al Viro2a737872009-04-07 11:49:53 -04001433 if (nd->path.dentry == nd->root.dentry &&
1434 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 break;
1436 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001437 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Eric W. Biedermaneedf2652016-06-02 10:29:47 -05001438 int ret = path_parent_directory(&nd->path);
1439 if (ret)
1440 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 break;
1442 }
Al Viro3088dd72010-01-30 15:47:29 -05001443 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 }
Al Viro79ed0222009-04-18 13:59:41 -04001446 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001447 nd->inode = nd->path.dentry->d_inode;
Eric W. Biederman397d4252015-08-15 20:27:13 -05001448 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449}
1450
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451/*
Miklos Szeredibad61182012-03-26 12:54:24 +02001452 * This looks up the name in dcache, possibly revalidates the old dentry and
1453 * allocates a new one if not found or not valid. In the need_lookup argument
1454 * returns whether i_op->lookup is necessary.
Nick Pigginbaa03892010-08-18 04:37:31 +10001455 */
Al Viroe3c13922016-03-06 14:03:27 -05001456static struct dentry *lookup_dcache(const struct qstr *name,
1457 struct dentry *dir,
Al Viro6c51e512016-03-05 20:09:32 -05001458 unsigned int flags)
Nick Pigginbaa03892010-08-18 04:37:31 +10001459{
Nick Pigginbaa03892010-08-18 04:37:31 +10001460 struct dentry *dentry;
Miklos Szeredibad61182012-03-26 12:54:24 +02001461 int error;
Nick Pigginbaa03892010-08-18 04:37:31 +10001462
Miklos Szeredibad61182012-03-26 12:54:24 +02001463 dentry = d_lookup(dir, name);
1464 if (dentry) {
Jeff Layton39e3c952012-11-28 11:30:53 -05001465 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
Al Viro201f9562012-06-22 12:42:10 +04001466 error = d_revalidate(dentry, flags);
Miklos Szeredibad61182012-03-26 12:54:24 +02001467 if (unlikely(error <= 0)) {
Al Viro74ff0ff2016-03-05 22:37:46 -05001468 if (!error)
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001469 d_invalidate(dentry);
Al Viro74ff0ff2016-03-05 22:37:46 -05001470 dput(dentry);
1471 return ERR_PTR(error);
Miklos Szeredibad61182012-03-26 12:54:24 +02001472 }
1473 }
1474 }
Nick Pigginbaa03892010-08-18 04:37:31 +10001475 return dentry;
1476}
1477
1478/*
J. Bruce Fields13a2c3b2013-10-23 16:09:16 -04001479 * Call i_op->lookup on the dentry. The dentry must be negative and
1480 * unhashed.
Miklos Szeredibad61182012-03-26 12:54:24 +02001481 *
1482 * dir->d_inode->i_mutex must be held
Josef Bacik44396f42011-05-31 11:58:49 -04001483 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001484static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
Al Viro72bd8662012-06-10 17:17:17 -04001485 unsigned int flags)
Josef Bacik44396f42011-05-31 11:58:49 -04001486{
Josef Bacik44396f42011-05-31 11:58:49 -04001487 struct dentry *old;
1488
1489 /* Don't create child dentry for a dead directory. */
Miklos Szeredibad61182012-03-26 12:54:24 +02001490 if (unlikely(IS_DEADDIR(dir))) {
Miklos Szeredie188dc02012-02-03 14:25:18 +01001491 dput(dentry);
Josef Bacik44396f42011-05-31 11:58:49 -04001492 return ERR_PTR(-ENOENT);
Miklos Szeredie188dc02012-02-03 14:25:18 +01001493 }
Josef Bacik44396f42011-05-31 11:58:49 -04001494
Al Viro72bd8662012-06-10 17:17:17 -04001495 old = dir->i_op->lookup(dir, dentry, flags);
Josef Bacik44396f42011-05-31 11:58:49 -04001496 if (unlikely(old)) {
1497 dput(dentry);
1498 dentry = old;
1499 }
1500 return dentry;
1501}
1502
Al Viroe3c13922016-03-06 14:03:27 -05001503static struct dentry *__lookup_hash(const struct qstr *name,
Al Viro72bd8662012-06-10 17:17:17 -04001504 struct dentry *base, unsigned int flags)
Al Viroa3255542012-03-30 14:41:51 -04001505{
Al Viro6c51e512016-03-05 20:09:32 -05001506 struct dentry *dentry = lookup_dcache(name, base, flags);
Al Viroa3255542012-03-30 14:41:51 -04001507
Al Viro6c51e512016-03-05 20:09:32 -05001508 if (dentry)
Miklos Szeredibad61182012-03-26 12:54:24 +02001509 return dentry;
Al Viroa3255542012-03-30 14:41:51 -04001510
Al Viro6c51e512016-03-05 20:09:32 -05001511 dentry = d_alloc(base, name);
1512 if (unlikely(!dentry))
1513 return ERR_PTR(-ENOMEM);
1514
Al Viro72bd8662012-06-10 17:17:17 -04001515 return lookup_real(base->d_inode, dentry, flags);
Al Viroa3255542012-03-30 14:41:51 -04001516}
1517
Al Viroe97cdc82013-01-24 18:16:00 -05001518static int lookup_fast(struct nameidata *nd,
Al Viro254cf582015-05-05 09:40:46 -04001519 struct path *path, struct inode **inode,
1520 unsigned *seqp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521{
Jan Blunck4ac91372008-02-14 19:34:32 -08001522 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001523 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001524 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001525 int err;
1526
Al Viro3cac2602009-08-13 18:27:43 +04001527 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001528 * Rename seqlock is not required here because in the off chance
Al Viro5d0f49c2016-03-05 21:32:53 -05001529 * of a false negative due to a concurrent rename, the caller is
1530 * going to fall back to non-racy lookup.
Nick Pigginb04f7842010-08-18 04:37:34 +10001531 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001532 if (nd->flags & LOOKUP_RCU) {
1533 unsigned seq;
Al Viro766c4cb2015-05-07 19:24:57 -04001534 bool negative;
Linus Torvaldsda53be12013-05-21 15:22:44 -07001535 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
Al Viro5d0f49c2016-03-05 21:32:53 -05001536 if (unlikely(!dentry)) {
1537 if (unlazy_walk(nd, NULL, 0))
1538 return -ECHILD;
Al Viroe9742b52016-03-05 22:04:59 -05001539 return 0;
Al Viro5d0f49c2016-03-05 21:32:53 -05001540 }
Al Viro5a18fff2011-03-11 04:44:53 -05001541
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001542 /*
1543 * This sequence count validates that the inode matches
1544 * the dentry name information from lookup.
1545 */
David Howells63afdfc2015-05-06 15:59:00 +01001546 *inode = d_backing_inode(dentry);
Al Viro766c4cb2015-05-07 19:24:57 -04001547 negative = d_is_negative(dentry);
Al Viro5d0f49c2016-03-05 21:32:53 -05001548 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001549 return -ECHILD;
1550
1551 /*
1552 * This sequence count validates that the parent had no
1553 * changes while we did the lookup of the dentry above.
1554 *
1555 * The memory barrier in read_seqcount_begin of child is
1556 * enough, we can use __read_seqcount_retry here.
1557 */
Al Viro5d0f49c2016-03-05 21:32:53 -05001558 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
Nick Piggin31e6b012011-01-07 17:49:52 +11001559 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001560
Al Viro254cf582015-05-05 09:40:46 -04001561 *seqp = seq;
Al Viro5d0f49c2016-03-05 21:32:53 -05001562 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
Al Viro4ce16ef32012-06-10 16:10:59 -04001563 status = d_revalidate(dentry, nd->flags);
Al Viro5d0f49c2016-03-05 21:32:53 -05001564 if (unlikely(status <= 0)) {
1565 if (unlazy_walk(nd, dentry, seq))
1566 return -ECHILD;
1567 if (status == -ECHILD)
1568 status = d_revalidate(dentry, nd->flags);
1569 } else {
1570 /*
1571 * Note: do negative dentry check after revalidation in
1572 * case that drops it.
1573 */
1574 if (unlikely(negative))
1575 return -ENOENT;
1576 path->mnt = mnt;
1577 path->dentry = dentry;
1578 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
Al Viroe9742b52016-03-05 22:04:59 -05001579 return 1;
Al Viro5d0f49c2016-03-05 21:32:53 -05001580 if (unlazy_walk(nd, dentry, seq))
1581 return -ECHILD;
Al Viro24643082011-02-15 01:26:22 -05001582 }
Al Viro5a18fff2011-03-11 04:44:53 -05001583 } else {
Al Viroe97cdc82013-01-24 18:16:00 -05001584 dentry = __d_lookup(parent, &nd->last);
Al Viro5d0f49c2016-03-05 21:32:53 -05001585 if (unlikely(!dentry))
Al Viroe9742b52016-03-05 22:04:59 -05001586 return 0;
Al Viro5d0f49c2016-03-05 21:32:53 -05001587 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
1588 status = d_revalidate(dentry, nd->flags);
Nick Piggin31e6b012011-01-07 17:49:52 +11001589 }
Al Viro5a18fff2011-03-11 04:44:53 -05001590 if (unlikely(status <= 0)) {
Al Viroe9742b52016-03-05 22:04:59 -05001591 if (!status)
Al Viro5d0f49c2016-03-05 21:32:53 -05001592 d_invalidate(dentry);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001593 dput(dentry);
Al Viro5d0f49c2016-03-05 21:32:53 -05001594 return status;
Al Viro5a18fff2011-03-11 04:44:53 -05001595 }
Al Viro766c4cb2015-05-07 19:24:57 -04001596 if (unlikely(d_is_negative(dentry))) {
1597 dput(dentry);
1598 return -ENOENT;
1599 }
Al Viro5d0f49c2016-03-05 21:32:53 -05001600
David Howells9875cf82011-01-14 18:45:21 +00001601 path->mnt = mnt;
1602 path->dentry = dentry;
NeilBrown756daf22015-03-23 13:37:38 +11001603 err = follow_managed(path, nd);
Al Viroe9742b52016-03-05 22:04:59 -05001604 if (likely(err > 0))
David Howells63afdfc2015-05-06 15:59:00 +01001605 *inode = d_backing_inode(path->dentry);
Al Viro84027522015-04-22 10:30:08 -04001606 return err;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001607}
1608
1609/* Fast lookup failed, do it the slow way */
Al Viroe3c13922016-03-06 14:03:27 -05001610static struct dentry *lookup_slow(const struct qstr *name,
1611 struct dentry *dir,
1612 unsigned int flags)
Miklos Szeredi697f5142012-05-21 17:30:05 +02001613{
Al Viro94bdd652016-04-15 02:42:04 -04001614 struct dentry *dentry = ERR_PTR(-ENOENT), *old;
Al Viro19363862016-04-14 19:33:34 -04001615 struct inode *inode = dir->d_inode;
Al Virod9171b92016-04-15 03:33:13 -04001616 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
Al Viro19363862016-04-14 19:33:34 -04001617
Al Viro9902af72016-04-15 15:08:36 -04001618 inode_lock_shared(inode);
Al Viro19363862016-04-14 19:33:34 -04001619 /* Don't go there if it's already dead */
Al Viro94bdd652016-04-15 02:42:04 -04001620 if (unlikely(IS_DEADDIR(inode)))
1621 goto out;
1622again:
Al Virod9171b92016-04-15 03:33:13 -04001623 dentry = d_alloc_parallel(dir, name, &wq);
Al Viro94bdd652016-04-15 02:42:04 -04001624 if (IS_ERR(dentry))
1625 goto out;
1626 if (unlikely(!d_in_lookup(dentry))) {
Al Viro949a8522016-03-06 14:20:52 -05001627 if ((dentry->d_flags & DCACHE_OP_REVALIDATE) &&
1628 !(flags & LOOKUP_NO_REVAL)) {
1629 int error = d_revalidate(dentry, flags);
1630 if (unlikely(error <= 0)) {
Al Viro94bdd652016-04-15 02:42:04 -04001631 if (!error) {
Al Viro949a8522016-03-06 14:20:52 -05001632 d_invalidate(dentry);
Al Viro94bdd652016-04-15 02:42:04 -04001633 dput(dentry);
1634 goto again;
1635 }
Al Viro949a8522016-03-06 14:20:52 -05001636 dput(dentry);
1637 dentry = ERR_PTR(error);
1638 }
1639 }
Al Viro94bdd652016-04-15 02:42:04 -04001640 } else {
1641 old = inode->i_op->lookup(inode, dentry, flags);
1642 d_lookup_done(dentry);
1643 if (unlikely(old)) {
1644 dput(dentry);
1645 dentry = old;
Al Viro949a8522016-03-06 14:20:52 -05001646 }
1647 }
Al Viro94bdd652016-04-15 02:42:04 -04001648out:
Al Viro9902af72016-04-15 15:08:36 -04001649 inode_unlock_shared(inode);
Al Viroe3c13922016-03-06 14:03:27 -05001650 return dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651}
1652
Al Viro52094c82011-02-21 21:34:47 -05001653static inline int may_lookup(struct nameidata *nd)
1654{
1655 if (nd->flags & LOOKUP_RCU) {
Al Viro4ad5abb2011-06-20 19:57:03 -04001656 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
Al Viro52094c82011-02-21 21:34:47 -05001657 if (err != -ECHILD)
1658 return err;
Al Viro6e9918b2015-05-05 09:26:05 -04001659 if (unlazy_walk(nd, NULL, 0))
Al Viro52094c82011-02-21 21:34:47 -05001660 return -ECHILD;
1661 }
Al Viro4ad5abb2011-06-20 19:57:03 -04001662 return inode_permission(nd->inode, MAY_EXEC);
Al Viro52094c82011-02-21 21:34:47 -05001663}
1664
Al Viro9856fa12011-03-04 14:22:06 -05001665static inline int handle_dots(struct nameidata *nd, int type)
1666{
1667 if (type == LAST_DOTDOT) {
Al Viro9e6697e2015-12-05 20:07:21 -05001668 if (!nd->root.mnt)
1669 set_root(nd);
Al Viro9856fa12011-03-04 14:22:06 -05001670 if (nd->flags & LOOKUP_RCU) {
Al Viro70291ae2015-05-04 07:53:00 -04001671 return follow_dotdot_rcu(nd);
Al Viro9856fa12011-03-04 14:22:06 -05001672 } else
Eric W. Biederman397d4252015-08-15 20:27:13 -05001673 return follow_dotdot(nd);
Al Viro9856fa12011-03-04 14:22:06 -05001674 }
1675 return 0;
1676}
1677
Al Viro181548c2015-05-07 19:54:34 -04001678static int pick_link(struct nameidata *nd, struct path *link,
1679 struct inode *inode, unsigned seq)
Al Virod63ff282015-05-04 18:13:23 -04001680{
Al Viro626de992015-05-04 18:26:59 -04001681 int error;
Al Viro1cf26652015-05-06 16:01:56 -04001682 struct saved *last;
NeilBrown756daf22015-03-23 13:37:38 +11001683 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
Al Viro626de992015-05-04 18:26:59 -04001684 path_to_nameidata(link, nd);
1685 return -ELOOP;
1686 }
Al Virobc40aee2015-05-09 13:04:24 -04001687 if (!(nd->flags & LOOKUP_RCU)) {
Al Viro79733872015-05-09 12:55:43 -04001688 if (link->mnt == nd->path.mnt)
1689 mntget(link->mnt);
Al Virod63ff282015-05-04 18:13:23 -04001690 }
Al Viro626de992015-05-04 18:26:59 -04001691 error = nd_alloc_stack(nd);
1692 if (unlikely(error)) {
Al Virobc40aee2015-05-09 13:04:24 -04001693 if (error == -ECHILD) {
1694 if (unlikely(unlazy_link(nd, link, seq)))
1695 return -ECHILD;
1696 error = nd_alloc_stack(nd);
1697 }
1698 if (error) {
1699 path_put(link);
1700 return error;
1701 }
Al Viro626de992015-05-04 18:26:59 -04001702 }
1703
Al Viroab104922015-05-10 11:50:01 -04001704 last = nd->stack + nd->depth++;
Al Viro1cf26652015-05-06 16:01:56 -04001705 last->link = *link;
Al Virofceef392015-12-29 15:58:39 -05001706 clear_delayed_call(&last->done);
1707 nd->link_inode = inode;
Al Viro0450b2d2015-05-08 13:23:53 -04001708 last->seq = seq;
Al Virod63ff282015-05-04 18:13:23 -04001709 return 1;
1710}
1711
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001712/*
1713 * Do we need to follow links? We _really_ want to be able
1714 * to do this check without having to look at inode->i_op,
1715 * so we keep a cache of "no, this doesn't need follow_link"
1716 * for the common case.
1717 */
Al Viro254cf582015-05-05 09:40:46 -04001718static inline int should_follow_link(struct nameidata *nd, struct path *link,
Al Viro181548c2015-05-07 19:54:34 -04001719 int follow,
1720 struct inode *inode, unsigned seq)
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001721{
Al Virod63ff282015-05-04 18:13:23 -04001722 if (likely(!d_is_symlink(link->dentry)))
1723 return 0;
1724 if (!follow)
1725 return 0;
Al Viroa7f77542016-02-27 19:31:01 -05001726 /* make sure that d_is_symlink above matches inode */
1727 if (nd->flags & LOOKUP_RCU) {
1728 if (read_seqcount_retry(&link->dentry->d_seq, seq))
1729 return -ECHILD;
1730 }
Al Viro181548c2015-05-07 19:54:34 -04001731 return pick_link(nd, link, inode, seq);
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001732}
1733
Al Viro4693a542015-05-04 17:47:11 -04001734enum {WALK_GET = 1, WALK_PUT = 2};
1735
1736static int walk_component(struct nameidata *nd, int flags)
Al Viroce57dfc2011-03-13 19:58:58 -04001737{
Al Virocaa85632015-04-22 17:52:47 -04001738 struct path path;
Al Viroce57dfc2011-03-13 19:58:58 -04001739 struct inode *inode;
Al Viro254cf582015-05-05 09:40:46 -04001740 unsigned seq;
Al Viroce57dfc2011-03-13 19:58:58 -04001741 int err;
1742 /*
1743 * "." and ".." are special - ".." especially so because it has
1744 * to be able to know about the current root directory and
1745 * parent relationships.
1746 */
Al Viro4693a542015-05-04 17:47:11 -04001747 if (unlikely(nd->last_type != LAST_NORM)) {
1748 err = handle_dots(nd, nd->last_type);
1749 if (flags & WALK_PUT)
1750 put_link(nd);
1751 return err;
1752 }
Al Viro254cf582015-05-05 09:40:46 -04001753 err = lookup_fast(nd, &path, &inode, &seq);
Al Viroe9742b52016-03-05 22:04:59 -05001754 if (unlikely(err <= 0)) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001755 if (err < 0)
Al Virof0a9ba72015-05-04 07:59:30 -04001756 return err;
Al Viroe3c13922016-03-06 14:03:27 -05001757 path.dentry = lookup_slow(&nd->last, nd->path.dentry,
1758 nd->flags);
1759 if (IS_ERR(path.dentry))
1760 return PTR_ERR(path.dentry);
Al Viro7500c382016-03-31 00:23:05 -04001761
Al Viroe3c13922016-03-06 14:03:27 -05001762 path.mnt = nd->path.mnt;
1763 err = follow_managed(&path, nd);
1764 if (unlikely(err < 0))
Al Virof0a9ba72015-05-04 07:59:30 -04001765 return err;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001766
Al Viro7500c382016-03-31 00:23:05 -04001767 if (unlikely(d_is_negative(path.dentry))) {
1768 path_to_nameidata(&path, nd);
1769 return -ENOENT;
1770 }
1771
Al Viro254cf582015-05-05 09:40:46 -04001772 seq = 0; /* we are already out of RCU mode */
Al Virod4565642016-02-27 19:23:16 -05001773 inode = d_backing_inode(path.dentry);
Al Viroce57dfc2011-03-13 19:58:58 -04001774 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001775
Al Viro4693a542015-05-04 17:47:11 -04001776 if (flags & WALK_PUT)
1777 put_link(nd);
Al Viro181548c2015-05-07 19:54:34 -04001778 err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq);
Al Virod63ff282015-05-04 18:13:23 -04001779 if (unlikely(err))
1780 return err;
Al Virocaa85632015-04-22 17:52:47 -04001781 path_to_nameidata(&path, nd);
Al Viroce57dfc2011-03-13 19:58:58 -04001782 nd->inode = inode;
Al Viro254cf582015-05-05 09:40:46 -04001783 nd->seq = seq;
Al Viroce57dfc2011-03-13 19:58:58 -04001784 return 0;
1785}
1786
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787/*
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001788 * We can do the critical dentry name comparison and hashing
1789 * operations one word at a time, but we are limited to:
1790 *
1791 * - Architectures with fast unaligned word accesses. We could
1792 * do a "get_unaligned()" if this helps and is sufficiently
1793 * fast.
1794 *
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001795 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1796 * do not trap on the (extremely unlikely) case of a page
1797 * crossing operation.
1798 *
1799 * - Furthermore, we need an efficient 64-bit compile for the
1800 * 64-bit case in order to generate the "number of bytes in
1801 * the final mask". Again, that could be replaced with a
1802 * efficient population count instruction or similar.
1803 */
1804#ifdef CONFIG_DCACHE_WORD_ACCESS
1805
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001806#include <asm/word-at-a-time.h>
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001807
George Spelvin468a9422016-05-26 22:11:51 -04001808#ifdef HASH_MIX
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001809
George Spelvin468a9422016-05-26 22:11:51 -04001810/* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1811
1812#elif defined(CONFIG_64BIT)
George Spelvin2a18da7a2016-05-23 07:43:58 -04001813/*
1814 * Register pressure in the mixing function is an issue, particularly
1815 * on 32-bit x86, but almost any function requires one state value and
1816 * one temporary. Instead, use a function designed for two state values
1817 * and no temporaries.
1818 *
1819 * This function cannot create a collision in only two iterations, so
1820 * we have two iterations to achieve avalanche. In those two iterations,
1821 * we have six layers of mixing, which is enough to spread one bit's
1822 * influence out to 2^6 = 64 state bits.
1823 *
1824 * Rotate constants are scored by considering either 64 one-bit input
1825 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1826 * probability of that delta causing a change to each of the 128 output
1827 * bits, using a sample of random initial states.
1828 *
1829 * The Shannon entropy of the computed probabilities is then summed
1830 * to produce a score. Ideally, any input change has a 50% chance of
1831 * toggling any given output bit.
1832 *
1833 * Mixing scores (in bits) for (12,45):
1834 * Input delta: 1-bit 2-bit
1835 * 1 round: 713.3 42542.6
1836 * 2 rounds: 2753.7 140389.8
1837 * 3 rounds: 5954.1 233458.2
1838 * 4 rounds: 7862.6 256672.2
1839 * Perfect: 8192 258048
1840 * (64*128) (64*63/2 * 128)
1841 */
1842#define HASH_MIX(x, y, a) \
1843 ( x ^= (a), \
1844 y ^= x, x = rol64(x,12),\
1845 x += y, y = rol64(y,45),\
1846 y *= 9 )
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001847
George Spelvin0fed3ac2016-05-02 06:31:01 -04001848/*
George Spelvin2a18da7a2016-05-23 07:43:58 -04001849 * Fold two longs into one 32-bit hash value. This must be fast, but
1850 * latency isn't quite as critical, as there is a fair bit of additional
1851 * work done before the hash value is used.
George Spelvin0fed3ac2016-05-02 06:31:01 -04001852 */
George Spelvin2a18da7a2016-05-23 07:43:58 -04001853static inline unsigned int fold_hash(unsigned long x, unsigned long y)
George Spelvin0fed3ac2016-05-02 06:31:01 -04001854{
George Spelvin2a18da7a2016-05-23 07:43:58 -04001855 y ^= x * GOLDEN_RATIO_64;
1856 y *= GOLDEN_RATIO_64;
1857 return y >> 32;
George Spelvin0fed3ac2016-05-02 06:31:01 -04001858}
1859
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001860#else /* 32-bit case */
1861
George Spelvin2a18da7a2016-05-23 07:43:58 -04001862/*
1863 * Mixing scores (in bits) for (7,20):
1864 * Input delta: 1-bit 2-bit
1865 * 1 round: 330.3 9201.6
1866 * 2 rounds: 1246.4 25475.4
1867 * 3 rounds: 1907.1 31295.1
1868 * 4 rounds: 2042.3 31718.6
1869 * Perfect: 2048 31744
1870 * (32*64) (32*31/2 * 64)
1871 */
1872#define HASH_MIX(x, y, a) \
1873 ( x ^= (a), \
1874 y ^= x, x = rol32(x, 7),\
1875 x += y, y = rol32(y,20),\
1876 y *= 9 )
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001877
George Spelvin2a18da7a2016-05-23 07:43:58 -04001878static inline unsigned int fold_hash(unsigned long x, unsigned long y)
George Spelvin0fed3ac2016-05-02 06:31:01 -04001879{
George Spelvin2a18da7a2016-05-23 07:43:58 -04001880 /* Use arch-optimized multiply if one exists */
1881 return __hash_32(y ^ __hash_32(x));
George Spelvin0fed3ac2016-05-02 06:31:01 -04001882}
1883
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001884#endif
1885
George Spelvin2a18da7a2016-05-23 07:43:58 -04001886/*
1887 * Return the hash of a string of known length. This is carfully
1888 * designed to match hash_name(), which is the more critical function.
1889 * In particular, we must end by hashing a final word containing 0..7
1890 * payload bytes, to match the way that hash_name() iterates until it
1891 * finds the delimiter after the name.
1892 */
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001893unsigned int full_name_hash(const char *name, unsigned int len)
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001894{
George Spelvin2a18da7a2016-05-23 07:43:58 -04001895 unsigned long a, x = 0, y = 0;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001896
1897 for (;;) {
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001898 if (!len)
1899 goto done;
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001900 a = load_unaligned_zeropad(name);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001901 if (len < sizeof(unsigned long))
1902 break;
George Spelvin2a18da7a2016-05-23 07:43:58 -04001903 HASH_MIX(x, y, a);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001904 name += sizeof(unsigned long);
1905 len -= sizeof(unsigned long);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001906 }
George Spelvin2a18da7a2016-05-23 07:43:58 -04001907 x ^= a & bytemask_from_count(len);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001908done:
George Spelvin2a18da7a2016-05-23 07:43:58 -04001909 return fold_hash(x, y);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001910}
1911EXPORT_SYMBOL(full_name_hash);
1912
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001913/* Return the "hash_len" (hash and length) of a null-terminated string */
1914u64 hashlen_string(const char *name)
1915{
George Spelvin2a18da7a2016-05-23 07:43:58 -04001916 unsigned long a = 0, x = 0, y = 0, adata, mask, len;
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001917 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1918
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001919 len = -sizeof(unsigned long);
1920 do {
George Spelvin2a18da7a2016-05-23 07:43:58 -04001921 HASH_MIX(x, y, a);
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001922 len += sizeof(unsigned long);
1923 a = load_unaligned_zeropad(name+len);
1924 } while (!has_zero(a, &adata, &constants));
1925
1926 adata = prep_zero_mask(a, adata, &constants);
1927 mask = create_zero_mask(adata);
George Spelvin2a18da7a2016-05-23 07:43:58 -04001928 x ^= a & zero_bytemask(mask);
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001929
George Spelvin2a18da7a2016-05-23 07:43:58 -04001930 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001931}
1932EXPORT_SYMBOL(hashlen_string);
1933
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001934/*
1935 * Calculate the length and hash of the path component, and
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001936 * return the "hash_len" as the result.
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001937 */
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001938static inline u64 hash_name(const char *name)
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001939{
George Spelvin2a18da7a2016-05-23 07:43:58 -04001940 unsigned long a = 0, b, x = 0, y = 0, adata, bdata, mask, len;
Linus Torvalds36126f82012-05-26 10:43:17 -07001941 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001942
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001943 len = -sizeof(unsigned long);
1944 do {
George Spelvin2a18da7a2016-05-23 07:43:58 -04001945 HASH_MIX(x, y, a);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001946 len += sizeof(unsigned long);
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001947 a = load_unaligned_zeropad(name+len);
Linus Torvalds36126f82012-05-26 10:43:17 -07001948 b = a ^ REPEAT_BYTE('/');
1949 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001950
Linus Torvalds36126f82012-05-26 10:43:17 -07001951 adata = prep_zero_mask(a, adata, &constants);
1952 bdata = prep_zero_mask(b, bdata, &constants);
Linus Torvalds36126f82012-05-26 10:43:17 -07001953 mask = create_zero_mask(adata | bdata);
George Spelvin2a18da7a2016-05-23 07:43:58 -04001954 x ^= a & zero_bytemask(mask);
Linus Torvalds36126f82012-05-26 10:43:17 -07001955
George Spelvin2a18da7a2016-05-23 07:43:58 -04001956 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001957}
1958
George Spelvin2a18da7a2016-05-23 07:43:58 -04001959#else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001960
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001961/* Return the hash of a string of known length */
1962unsigned int full_name_hash(const char *name, unsigned int len)
Linus Torvalds0145acc2012-03-02 14:32:59 -08001963{
1964 unsigned long hash = init_name_hash();
1965 while (len--)
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001966 hash = partial_name_hash((unsigned char)*name++, hash);
Linus Torvalds0145acc2012-03-02 14:32:59 -08001967 return end_name_hash(hash);
1968}
Linus Torvaldsae942ae2012-03-02 19:40:57 -08001969EXPORT_SYMBOL(full_name_hash);
Linus Torvalds0145acc2012-03-02 14:32:59 -08001970
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001971/* Return the "hash_len" (hash and length) of a null-terminated string */
George Spelvinf2a031b2016-05-29 01:26:41 -04001972u64 hashlen_string(const char *name)
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001973{
1974 unsigned long hash = init_name_hash();
1975 unsigned long len = 0, c;
1976
1977 c = (unsigned char)*name;
George Spelvine0ab7af2016-05-29 08:05:56 -04001978 while (c) {
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001979 len++;
1980 hash = partial_name_hash(c, hash);
1981 c = (unsigned char)name[len];
George Spelvine0ab7af2016-05-29 08:05:56 -04001982 }
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001983 return hashlen_create(end_name_hash(hash), len);
1984}
George Spelvinf2a031b2016-05-29 01:26:41 -04001985EXPORT_SYMBOL(hashlen_string);
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001986
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001987/*
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001988 * We know there's a real path component here of at least
1989 * one character.
1990 */
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001991static inline u64 hash_name(const char *name)
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001992{
1993 unsigned long hash = init_name_hash();
1994 unsigned long len = 0, c;
1995
1996 c = (unsigned char)*name;
1997 do {
1998 len++;
1999 hash = partial_name_hash(c, hash);
2000 c = (unsigned char)name[len];
2001 } while (c && c != '/');
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002002 return hashlen_create(end_name_hash(hash), len);
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002003}
2004
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08002005#endif
2006
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002007/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01002009 * This is the basic name resolution function, turning a pathname into
2010 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01002012 * Returns 0 and nd will have valid dentry and mnt on success.
2013 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014 */
Al Viro6de88d72009-08-09 01:41:57 +04002015static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 int err;
Al Viro32cd7462015-04-18 20:30:49 -04002018
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 while (*name=='/')
2020 name++;
2021 if (!*name)
Al Viro9e18f102015-04-18 20:44:34 -04002022 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024 /* At this point we know we have a real path component. */
2025 for(;;) {
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002026 u64 hash_len;
Al Virofe479a52011-02-22 15:10:03 -05002027 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028
Al Viro52094c82011-02-21 21:34:47 -05002029 err = may_lookup(nd);
George Spelvin2a18da7a2016-05-23 07:43:58 -04002030 if (err)
Al Viro3595e232015-05-09 16:54:45 -04002031 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002033 hash_len = hash_name(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034
Al Virofe479a52011-02-22 15:10:03 -05002035 type = LAST_NORM;
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002036 if (name[0] == '.') switch (hashlen_len(hash_len)) {
Al Virofe479a52011-02-22 15:10:03 -05002037 case 2:
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002038 if (name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05002039 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05002040 nd->flags |= LOOKUP_JUMPED;
2041 }
Al Virofe479a52011-02-22 15:10:03 -05002042 break;
2043 case 1:
2044 type = LAST_DOT;
2045 }
Al Viro5a202bc2011-03-08 14:17:44 -05002046 if (likely(type == LAST_NORM)) {
2047 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05002048 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05002049 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
James Hogana060dc52014-09-16 13:07:35 +01002050 struct qstr this = { { .hash_len = hash_len }, .name = name };
Linus Torvaldsda53be12013-05-21 15:22:44 -07002051 err = parent->d_op->d_hash(parent, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05002052 if (err < 0)
Al Viro3595e232015-05-09 16:54:45 -04002053 return err;
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002054 hash_len = this.hash_len;
2055 name = this.name;
Al Viro5a202bc2011-03-08 14:17:44 -05002056 }
2057 }
Al Virofe479a52011-02-22 15:10:03 -05002058
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002059 nd->last.hash_len = hash_len;
2060 nd->last.name = name;
Al Viro5f4a6a62013-01-24 18:04:22 -05002061 nd->last_type = type;
2062
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002063 name += hashlen_len(hash_len);
2064 if (!*name)
Al Virobdf6cbf2015-04-18 20:21:40 -04002065 goto OK;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002066 /*
2067 * If it wasn't NUL, we know it was '/'. Skip that
2068 * slash, and continue until no more slashes.
2069 */
2070 do {
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002071 name++;
2072 } while (unlikely(*name == '/'));
Al Viro8620c232015-05-04 08:58:35 -04002073 if (unlikely(!*name)) {
2074OK:
Al Viro368ee9b2015-05-08 17:19:59 -04002075 /* pathname body, done */
Al Viro8620c232015-05-04 08:58:35 -04002076 if (!nd->depth)
2077 return 0;
2078 name = nd->stack[nd->depth - 1].name;
Al Viro368ee9b2015-05-08 17:19:59 -04002079 /* trailing symlink, done */
Al Viro8620c232015-05-04 08:58:35 -04002080 if (!name)
2081 return 0;
2082 /* last component of nested symlink */
Al Viro4693a542015-05-04 17:47:11 -04002083 err = walk_component(nd, WALK_GET | WALK_PUT);
Al Viro8620c232015-05-04 08:58:35 -04002084 } else {
Al Viro4693a542015-05-04 17:47:11 -04002085 err = walk_component(nd, WALK_GET);
Al Viro8620c232015-05-04 08:58:35 -04002086 }
Al Viroce57dfc2011-03-13 19:58:58 -04002087 if (err < 0)
Al Viro3595e232015-05-09 16:54:45 -04002088 return err;
Al Virofe479a52011-02-22 15:10:03 -05002089
Al Viroce57dfc2011-03-13 19:58:58 -04002090 if (err) {
Al Viro626de992015-05-04 18:26:59 -04002091 const char *s = get_link(nd);
Al Viro5a460272015-04-17 23:44:45 -04002092
Viresh Kumara1c83682015-08-12 15:59:44 +05302093 if (IS_ERR(s))
Al Viro3595e232015-05-09 16:54:45 -04002094 return PTR_ERR(s);
Al Virod40bcc02015-04-18 20:03:03 -04002095 err = 0;
2096 if (unlikely(!s)) {
2097 /* jumped */
Al Virob9ff4422015-05-02 20:19:23 -04002098 put_link(nd);
Al Virod40bcc02015-04-18 20:03:03 -04002099 } else {
Al Virofab51e82015-05-10 11:01:00 -04002100 nd->stack[nd->depth - 1].name = name;
2101 name = s;
2102 continue;
Al Virod40bcc02015-04-18 20:03:03 -04002103 }
Nick Piggin31e6b012011-01-07 17:49:52 +11002104 }
Al Viro97242f92015-08-01 19:59:28 -04002105 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2106 if (nd->flags & LOOKUP_RCU) {
2107 if (unlazy_walk(nd, NULL, 0))
2108 return -ECHILD;
2109 }
Al Viro3595e232015-05-09 16:54:45 -04002110 return -ENOTDIR;
Al Viro97242f92015-08-01 19:59:28 -04002111 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113}
2114
Al Viroc8a53ee2015-05-12 18:43:07 -04002115static const char *path_init(struct nameidata *nd, unsigned flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116{
Prasanna Medaea3834d2005-04-29 16:00:17 +01002117 int retval = 0;
Al Viroc8a53ee2015-05-12 18:43:07 -04002118 const char *s = nd->name->name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119
2120 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro980f3ea2014-11-20 14:20:24 -05002121 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122 nd->depth = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05002123 if (flags & LOOKUP_ROOT) {
David Howellsb18825a2013-09-12 19:22:53 +01002124 struct dentry *root = nd->root.dentry;
2125 struct inode *inode = root->d_inode;
Al Virofd2f7cb2015-02-22 20:07:13 -05002126 if (*s) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002127 if (!d_can_lookup(root))
Al Viro368ee9b2015-05-08 17:19:59 -04002128 return ERR_PTR(-ENOTDIR);
Al Viro73d049a2011-03-11 12:08:24 -05002129 retval = inode_permission(inode, MAY_EXEC);
2130 if (retval)
Al Viro368ee9b2015-05-08 17:19:59 -04002131 return ERR_PTR(retval);
Al Viro73d049a2011-03-11 12:08:24 -05002132 }
Al Viro5b6ca022011-03-09 23:04:47 -05002133 nd->path = nd->root;
2134 nd->inode = inode;
2135 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05002136 rcu_read_lock();
Al Viro5b6ca022011-03-09 23:04:47 -05002137 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro8f47a0162015-05-09 19:02:01 -04002138 nd->root_seq = nd->seq;
Al Viro48a066e2013-09-29 22:06:07 -04002139 nd->m_seq = read_seqbegin(&mount_lock);
Al Viro5b6ca022011-03-09 23:04:47 -05002140 } else {
2141 path_get(&nd->path);
2142 }
Al Viro368ee9b2015-05-08 17:19:59 -04002143 return s;
Al Viro5b6ca022011-03-09 23:04:47 -05002144 }
2145
Al Viro2a737872009-04-07 11:49:53 -04002146 nd->root.mnt = NULL;
Al Viro248fb5b2015-12-05 20:51:58 -05002147 nd->path.mnt = NULL;
2148 nd->path.dentry = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149
Al Viro48a066e2013-09-29 22:06:07 -04002150 nd->m_seq = read_seqbegin(&mount_lock);
Al Virofd2f7cb2015-02-22 20:07:13 -05002151 if (*s == '/') {
Al Viro9e6697e2015-12-05 20:07:21 -05002152 if (flags & LOOKUP_RCU)
Al Viro8b61e742013-11-08 12:45:01 -05002153 rcu_read_lock();
Al Viro9e6697e2015-12-05 20:07:21 -05002154 set_root(nd);
Al Viro248fb5b2015-12-05 20:51:58 -05002155 if (likely(!nd_jump_root(nd)))
Al Viroef55d912015-12-05 20:25:06 -05002156 return s;
Al Viro248fb5b2015-12-05 20:51:58 -05002157 nd->root.mnt = NULL;
Al Viroef55d912015-12-05 20:25:06 -05002158 rcu_read_unlock();
2159 return ERR_PTR(-ECHILD);
Al Viroc8a53ee2015-05-12 18:43:07 -04002160 } else if (nd->dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05002161 if (flags & LOOKUP_RCU) {
2162 struct fs_struct *fs = current->fs;
2163 unsigned seq;
2164
Al Viro8b61e742013-11-08 12:45:01 -05002165 rcu_read_lock();
Al Viroe41f7d42011-02-22 14:02:58 -05002166
2167 do {
2168 seq = read_seqcount_begin(&fs->seq);
2169 nd->path = fs->pwd;
Al Viroef55d912015-12-05 20:25:06 -05002170 nd->inode = nd->path.dentry->d_inode;
Al Viroe41f7d42011-02-22 14:02:58 -05002171 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2172 } while (read_seqcount_retry(&fs->seq, seq));
2173 } else {
2174 get_fs_pwd(current->fs, &nd->path);
Al Viroef55d912015-12-05 20:25:06 -05002175 nd->inode = nd->path.dentry->d_inode;
Al Viroe41f7d42011-02-22 14:02:58 -05002176 }
Al Viroef55d912015-12-05 20:25:06 -05002177 return s;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002178 } else {
Jeff Layton582aa642012-12-11 08:56:16 -05002179 /* Caller must check execute permissions on the starting path component */
Al Viroc8a53ee2015-05-12 18:43:07 -04002180 struct fd f = fdget_raw(nd->dfd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002181 struct dentry *dentry;
2182
Al Viro2903ff02012-08-28 12:52:22 -04002183 if (!f.file)
Al Viro368ee9b2015-05-08 17:19:59 -04002184 return ERR_PTR(-EBADF);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002185
Al Viro2903ff02012-08-28 12:52:22 -04002186 dentry = f.file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002187
Al Virofd2f7cb2015-02-22 20:07:13 -05002188 if (*s) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002189 if (!d_can_lookup(dentry)) {
Al Viro2903ff02012-08-28 12:52:22 -04002190 fdput(f);
Al Viro368ee9b2015-05-08 17:19:59 -04002191 return ERR_PTR(-ENOTDIR);
Al Viro2903ff02012-08-28 12:52:22 -04002192 }
Al Virof52e0c12011-03-14 18:56:51 -04002193 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002194
Al Viro2903ff02012-08-28 12:52:22 -04002195 nd->path = f.file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05002196 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05002197 rcu_read_lock();
Al Viro34a26b92015-05-11 08:05:05 -04002198 nd->inode = nd->path.dentry->d_inode;
2199 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viroe41f7d42011-02-22 14:02:58 -05002200 } else {
Al Viro2903ff02012-08-28 12:52:22 -04002201 path_get(&nd->path);
Al Viro34a26b92015-05-11 08:05:05 -04002202 nd->inode = nd->path.dentry->d_inode;
Al Viroe41f7d42011-02-22 14:02:58 -05002203 }
Al Viro34a26b92015-05-11 08:05:05 -04002204 fdput(f);
Al Viro368ee9b2015-05-08 17:19:59 -04002205 return s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 }
Al Viro9b4a9b12009-04-07 11:44:16 -04002207}
2208
Al Viro3bdba282015-05-08 17:37:07 -04002209static const char *trailing_symlink(struct nameidata *nd)
Al Viro95fa25d2015-04-22 13:46:57 -04002210{
2211 const char *s;
Al Virofec2fa22015-05-06 15:58:18 -04002212 int error = may_follow_link(nd);
Al Virodeb106c2015-05-08 18:05:21 -04002213 if (unlikely(error))
Al Viro3bdba282015-05-08 17:37:07 -04002214 return ERR_PTR(error);
Al Viro95fa25d2015-04-22 13:46:57 -04002215 nd->flags |= LOOKUP_PARENT;
Al Virofab51e82015-05-10 11:01:00 -04002216 nd->stack[0].name = NULL;
Al Viro3b2e7f72015-04-19 00:53:50 -04002217 s = get_link(nd);
Al Virodeb106c2015-05-08 18:05:21 -04002218 return s ? s : "";
Al Viro95fa25d2015-04-22 13:46:57 -04002219}
2220
Al Virocaa85632015-04-22 17:52:47 -04002221static inline int lookup_last(struct nameidata *nd)
Al Virobd92d7f2011-03-14 19:54:59 -04002222{
2223 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2224 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2225
2226 nd->flags &= ~LOOKUP_PARENT;
Al Virodeb106c2015-05-08 18:05:21 -04002227 return walk_component(nd,
Al Viro4693a542015-05-04 17:47:11 -04002228 nd->flags & LOOKUP_FOLLOW
2229 ? nd->depth
2230 ? WALK_PUT | WALK_GET
2231 : WALK_GET
2232 : 0);
Al Virobd92d7f2011-03-14 19:54:59 -04002233}
2234
Al Viro9b4a9b12009-04-07 11:44:16 -04002235/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroc8a53ee2015-05-12 18:43:07 -04002236static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
Al Viro9b4a9b12009-04-07 11:44:16 -04002237{
Al Viroc8a53ee2015-05-12 18:43:07 -04002238 const char *s = path_init(nd, flags);
Al Virobd92d7f2011-03-14 19:54:59 -04002239 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11002240
Al Viro368ee9b2015-05-08 17:19:59 -04002241 if (IS_ERR(s))
2242 return PTR_ERR(s);
Al Viro3bdba282015-05-08 17:37:07 -04002243 while (!(err = link_path_walk(s, nd))
2244 && ((err = lookup_last(nd)) > 0)) {
2245 s = trailing_symlink(nd);
2246 if (IS_ERR(s)) {
2247 err = PTR_ERR(s);
2248 break;
Al Virobd92d7f2011-03-14 19:54:59 -04002249 }
2250 }
Al Viro9f1fafe2011-03-25 11:00:12 -04002251 if (!err)
2252 err = complete_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04002253
Al Virodeb106c2015-05-08 18:05:21 -04002254 if (!err && nd->flags & LOOKUP_DIRECTORY)
2255 if (!d_can_lookup(nd->path.dentry))
Al Virobd23a532011-03-23 09:56:30 -04002256 err = -ENOTDIR;
Al Viro625b6d12015-05-12 16:36:12 -04002257 if (!err) {
2258 *path = nd->path;
2259 nd->path.mnt = NULL;
2260 nd->path.dentry = NULL;
2261 }
2262 terminate_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04002263 return err;
Al Viroee0827c2011-02-21 23:38:09 -05002264}
Nick Piggin31e6b012011-01-07 17:49:52 +11002265
Al Viro625b6d12015-05-12 16:36:12 -04002266static int filename_lookup(int dfd, struct filename *name, unsigned flags,
Al Viro9ad1aaa2015-05-12 16:44:39 -04002267 struct path *path, struct path *root)
Jeff Layton873f1ee2012-10-10 15:25:29 -04002268{
Al Viro894bc8c2015-05-02 07:16:16 -04002269 int retval;
Al Viro9883d182015-05-13 07:28:08 -04002270 struct nameidata nd;
Al Viroabc9f5b2015-05-12 16:53:42 -04002271 if (IS_ERR(name))
2272 return PTR_ERR(name);
Al Viro9ad1aaa2015-05-12 16:44:39 -04002273 if (unlikely(root)) {
2274 nd.root = *root;
2275 flags |= LOOKUP_ROOT;
2276 }
Al Viro9883d182015-05-13 07:28:08 -04002277 set_nameidata(&nd, dfd, name);
Al Viroc8a53ee2015-05-12 18:43:07 -04002278 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002279 if (unlikely(retval == -ECHILD))
Al Viroc8a53ee2015-05-12 18:43:07 -04002280 retval = path_lookupat(&nd, flags, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002281 if (unlikely(retval == -ESTALE))
Al Viroc8a53ee2015-05-12 18:43:07 -04002282 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002283
2284 if (likely(!retval))
Al Viro625b6d12015-05-12 16:36:12 -04002285 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
Al Viro9883d182015-05-13 07:28:08 -04002286 restore_nameidata();
Al Viroe4bd1c12015-05-12 16:40:39 -04002287 putname(name);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002288 return retval;
2289}
2290
Al Viro8bcb77f2015-05-08 16:59:20 -04002291/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroc8a53ee2015-05-12 18:43:07 -04002292static int path_parentat(struct nameidata *nd, unsigned flags,
Al Viro391172c2015-05-09 11:19:16 -04002293 struct path *parent)
Al Viro8bcb77f2015-05-08 16:59:20 -04002294{
Al Viroc8a53ee2015-05-12 18:43:07 -04002295 const char *s = path_init(nd, flags);
Al Viro368ee9b2015-05-08 17:19:59 -04002296 int err;
2297 if (IS_ERR(s))
2298 return PTR_ERR(s);
2299 err = link_path_walk(s, nd);
Al Viro8bcb77f2015-05-08 16:59:20 -04002300 if (!err)
2301 err = complete_walk(nd);
Al Viro391172c2015-05-09 11:19:16 -04002302 if (!err) {
2303 *parent = nd->path;
2304 nd->path.mnt = NULL;
2305 nd->path.dentry = NULL;
2306 }
2307 terminate_walk(nd);
Al Viro8bcb77f2015-05-08 16:59:20 -04002308 return err;
2309}
2310
Al Viro5c31b6c2015-05-12 17:32:54 -04002311static struct filename *filename_parentat(int dfd, struct filename *name,
Al Viro391172c2015-05-09 11:19:16 -04002312 unsigned int flags, struct path *parent,
2313 struct qstr *last, int *type)
Al Viro8bcb77f2015-05-08 16:59:20 -04002314{
2315 int retval;
Al Viro9883d182015-05-13 07:28:08 -04002316 struct nameidata nd;
Al Viro8bcb77f2015-05-08 16:59:20 -04002317
Al Viro5c31b6c2015-05-12 17:32:54 -04002318 if (IS_ERR(name))
2319 return name;
Al Viro9883d182015-05-13 07:28:08 -04002320 set_nameidata(&nd, dfd, name);
Al Viroc8a53ee2015-05-12 18:43:07 -04002321 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
Al Viro8bcb77f2015-05-08 16:59:20 -04002322 if (unlikely(retval == -ECHILD))
Al Viroc8a53ee2015-05-12 18:43:07 -04002323 retval = path_parentat(&nd, flags, parent);
Al Viro8bcb77f2015-05-08 16:59:20 -04002324 if (unlikely(retval == -ESTALE))
Al Viroc8a53ee2015-05-12 18:43:07 -04002325 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
Al Viro391172c2015-05-09 11:19:16 -04002326 if (likely(!retval)) {
2327 *last = nd.last;
2328 *type = nd.last_type;
2329 audit_inode(name, parent->dentry, LOOKUP_PARENT);
Al Viro5c31b6c2015-05-12 17:32:54 -04002330 } else {
2331 putname(name);
2332 name = ERR_PTR(retval);
Al Viro391172c2015-05-09 11:19:16 -04002333 }
Al Viro9883d182015-05-13 07:28:08 -04002334 restore_nameidata();
Al Viro5c31b6c2015-05-12 17:32:54 -04002335 return name;
Al Viro8bcb77f2015-05-08 16:59:20 -04002336}
2337
Al Viro79714f72012-06-15 03:01:42 +04002338/* does lookup, returns the object with parent locked */
2339struct dentry *kern_path_locked(const char *name, struct path *path)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002340{
Al Viro5c31b6c2015-05-12 17:32:54 -04002341 struct filename *filename;
2342 struct dentry *d;
Al Viro391172c2015-05-09 11:19:16 -04002343 struct qstr last;
2344 int type;
Paul Moore51689102015-01-22 00:00:03 -05002345
Al Viro5c31b6c2015-05-12 17:32:54 -04002346 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2347 &last, &type);
Paul Moore51689102015-01-22 00:00:03 -05002348 if (IS_ERR(filename))
2349 return ERR_CAST(filename);
Al Viro5c31b6c2015-05-12 17:32:54 -04002350 if (unlikely(type != LAST_NORM)) {
Al Viro391172c2015-05-09 11:19:16 -04002351 path_put(path);
Al Viro5c31b6c2015-05-12 17:32:54 -04002352 putname(filename);
2353 return ERR_PTR(-EINVAL);
Al Viro79714f72012-06-15 03:01:42 +04002354 }
Al Viro59551022016-01-22 15:40:57 -05002355 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
Al Viro391172c2015-05-09 11:19:16 -04002356 d = __lookup_hash(&last, path->dentry, 0);
Al Viro79714f72012-06-15 03:01:42 +04002357 if (IS_ERR(d)) {
Al Viro59551022016-01-22 15:40:57 -05002358 inode_unlock(path->dentry->d_inode);
Al Viro391172c2015-05-09 11:19:16 -04002359 path_put(path);
Al Viro79714f72012-06-15 03:01:42 +04002360 }
Paul Moore51689102015-01-22 00:00:03 -05002361 putname(filename);
Al Viro79714f72012-06-15 03:01:42 +04002362 return d;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002363}
2364
Al Virod1811462008-08-02 00:49:18 -04002365int kern_path(const char *name, unsigned int flags, struct path *path)
2366{
Al Viroabc9f5b2015-05-12 16:53:42 -04002367 return filename_lookup(AT_FDCWD, getname_kernel(name),
2368 flags, path, NULL);
Al Virod1811462008-08-02 00:49:18 -04002369}
Al Viro4d359502014-03-14 12:20:17 -04002370EXPORT_SYMBOL(kern_path);
Al Virod1811462008-08-02 00:49:18 -04002371
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002372/**
2373 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2374 * @dentry: pointer to dentry of the base directory
2375 * @mnt: pointer to vfs mount of the base directory
2376 * @name: pointer to file name
2377 * @flags: lookup flags
Al Viroe0a01242011-06-27 17:00:37 -04002378 * @path: pointer to struct path to fill
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002379 */
2380int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2381 const char *name, unsigned int flags,
Al Viroe0a01242011-06-27 17:00:37 -04002382 struct path *path)
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002383{
Al Viro9ad1aaa2015-05-12 16:44:39 -04002384 struct path root = {.mnt = mnt, .dentry = dentry};
Al Viro9ad1aaa2015-05-12 16:44:39 -04002385 /* the first argument of filename_lookup() is ignored with root */
Al Viroabc9f5b2015-05-12 16:53:42 -04002386 return filename_lookup(AT_FDCWD, getname_kernel(name),
2387 flags , path, &root);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002388}
Al Viro4d359502014-03-14 12:20:17 -04002389EXPORT_SYMBOL(vfs_path_lookup);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002390
Christoph Hellwigeead1912007-10-16 23:25:38 -07002391/**
Miklos Szeredi3c9fe8c2016-05-11 01:16:37 +02002392 * lookup_hash - lookup single pathname component on already hashed name
2393 * @name: name and hash to lookup
2394 * @base: base directory to lookup from
2395 *
2396 * The name must have been verified and hashed (see lookup_one_len()). Using
2397 * this after just full_name_hash() is unsafe.
2398 *
2399 * This function also doesn't check for search permission on base directory.
2400 *
2401 * Use lookup_one_len_unlocked() instead, unless you really know what you are
2402 * doing.
2403 *
2404 * Do not hold i_mutex; this helper takes i_mutex if necessary.
2405 */
2406struct dentry *lookup_hash(const struct qstr *name, struct dentry *base)
2407{
2408 struct dentry *ret;
2409
2410 ret = lookup_dcache(name, base, 0);
2411 if (!ret)
2412 ret = lookup_slow(name, base, 0);
2413
2414 return ret;
2415}
2416EXPORT_SYMBOL(lookup_hash);
2417
2418/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07002419 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07002420 * @name: pathname component to lookup
2421 * @base: base directory to lookup from
2422 * @len: maximum length @len should be interpreted to
2423 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07002424 * Note that this routine is purely a helper for filesystem usage and should
Al Viro9e7543e2015-02-23 02:49:48 -05002425 * not be called by generic code.
NeilBrownbbddca82016-01-07 16:08:20 -05002426 *
2427 * The caller must hold base->i_mutex.
Christoph Hellwigeead1912007-10-16 23:25:38 -07002428 */
James Morris057f6c02007-04-26 00:12:05 -07002429struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2430{
James Morris057f6c02007-04-26 00:12:05 -07002431 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05002432 unsigned int c;
Miklos Szeredicda309d2012-03-26 12:54:21 +02002433 int err;
James Morris057f6c02007-04-26 00:12:05 -07002434
Al Viro59551022016-01-22 15:40:57 -05002435 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
David Woodhouse2f9092e2009-04-20 23:18:37 +01002436
Al Viro6a96ba52011-03-07 23:49:20 -05002437 this.name = name;
2438 this.len = len;
Linus Torvalds0145acc2012-03-02 14:32:59 -08002439 this.hash = full_name_hash(name, len);
Al Viro6a96ba52011-03-07 23:49:20 -05002440 if (!len)
2441 return ERR_PTR(-EACCES);
2442
Al Viro21d8a152012-11-29 22:17:21 -05002443 if (unlikely(name[0] == '.')) {
2444 if (len < 2 || (len == 2 && name[1] == '.'))
2445 return ERR_PTR(-EACCES);
2446 }
2447
Al Viro6a96ba52011-03-07 23:49:20 -05002448 while (len--) {
2449 c = *(const unsigned char *)name++;
2450 if (c == '/' || c == '\0')
2451 return ERR_PTR(-EACCES);
Al Viro6a96ba52011-03-07 23:49:20 -05002452 }
Al Viro5a202bc2011-03-08 14:17:44 -05002453 /*
2454 * See if the low-level filesystem might want
2455 * to use its own hash..
2456 */
2457 if (base->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002458 int err = base->d_op->d_hash(base, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05002459 if (err < 0)
2460 return ERR_PTR(err);
2461 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07002462
Miklos Szeredicda309d2012-03-26 12:54:21 +02002463 err = inode_permission(base->d_inode, MAY_EXEC);
2464 if (err)
2465 return ERR_PTR(err);
2466
Al Viro72bd8662012-06-10 17:17:17 -04002467 return __lookup_hash(&this, base, 0);
James Morris057f6c02007-04-26 00:12:05 -07002468}
Al Viro4d359502014-03-14 12:20:17 -04002469EXPORT_SYMBOL(lookup_one_len);
James Morris057f6c02007-04-26 00:12:05 -07002470
NeilBrownbbddca82016-01-07 16:08:20 -05002471/**
2472 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2473 * @name: pathname component to lookup
2474 * @base: base directory to lookup from
2475 * @len: maximum length @len should be interpreted to
2476 *
2477 * Note that this routine is purely a helper for filesystem usage and should
2478 * not be called by generic code.
2479 *
2480 * Unlike lookup_one_len, it should be called without the parent
2481 * i_mutex held, and will take the i_mutex itself if necessary.
2482 */
2483struct dentry *lookup_one_len_unlocked(const char *name,
2484 struct dentry *base, int len)
2485{
2486 struct qstr this;
2487 unsigned int c;
2488 int err;
NeilBrownbbddca82016-01-07 16:08:20 -05002489
2490 this.name = name;
2491 this.len = len;
2492 this.hash = full_name_hash(name, len);
2493 if (!len)
2494 return ERR_PTR(-EACCES);
2495
2496 if (unlikely(name[0] == '.')) {
2497 if (len < 2 || (len == 2 && name[1] == '.'))
2498 return ERR_PTR(-EACCES);
2499 }
2500
2501 while (len--) {
2502 c = *(const unsigned char *)name++;
2503 if (c == '/' || c == '\0')
2504 return ERR_PTR(-EACCES);
2505 }
2506 /*
2507 * See if the low-level filesystem might want
2508 * to use its own hash..
2509 */
2510 if (base->d_flags & DCACHE_OP_HASH) {
2511 int err = base->d_op->d_hash(base, &this);
2512 if (err < 0)
2513 return ERR_PTR(err);
2514 }
2515
2516 err = inode_permission(base->d_inode, MAY_EXEC);
2517 if (err)
2518 return ERR_PTR(err);
2519
Miklos Szeredi3c9fe8c2016-05-11 01:16:37 +02002520 return lookup_hash(&this, base);
NeilBrownbbddca82016-01-07 16:08:20 -05002521}
2522EXPORT_SYMBOL(lookup_one_len_unlocked);
2523
Eric W. Biedermaneedf2652016-06-02 10:29:47 -05002524#ifdef CONFIG_UNIX98_PTYS
2525int path_pts(struct path *path)
2526{
2527 /* Find something mounted on "pts" in the same directory as
2528 * the input path.
2529 */
2530 struct dentry *child, *parent;
2531 struct qstr this;
2532 int ret;
2533
2534 ret = path_parent_directory(path);
2535 if (ret)
2536 return ret;
2537
2538 parent = path->dentry;
2539 this.name = "pts";
2540 this.len = 3;
2541 child = d_hash_and_lookup(parent, &this);
2542 if (!child)
2543 return -ENOENT;
2544
2545 path->dentry = child;
2546 dput(parent);
2547 follow_mount(path);
2548 return 0;
2549}
2550#endif
2551
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002552int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2553 struct path *path, int *empty)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554{
Al Viroabc9f5b2015-05-12 16:53:42 -04002555 return filename_lookup(dfd, getname_flags(name, flags, empty),
2556 flags, path, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557}
Al Virob853a162015-05-13 09:12:02 -04002558EXPORT_SYMBOL(user_path_at_empty);
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002559
Jeff Layton873f1ee2012-10-10 15:25:29 -04002560/*
2561 * NB: most callers don't do anything directly with the reference to the
2562 * to struct filename, but the nd->last pointer points into the name string
2563 * allocated by getname. So we must hold the reference to it until all
2564 * path-walking is complete.
2565 */
Al Viroa2ec4a22015-05-13 06:57:49 -04002566static inline struct filename *
Al Virof5beed72015-04-30 16:09:11 -04002567user_path_parent(int dfd, const char __user *path,
2568 struct path *parent,
2569 struct qstr *last,
2570 int *type,
Jeff Layton9e790bd2012-12-11 12:10:09 -05002571 unsigned int flags)
Al Viro2ad94ae2008-07-21 09:32:51 -04002572{
Jeff Layton9e790bd2012-12-11 12:10:09 -05002573 /* only LOOKUP_REVAL is allowed in extra flags */
Al Viro5c31b6c2015-05-12 17:32:54 -04002574 return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL,
2575 parent, last, type);
Al Viro2ad94ae2008-07-21 09:32:51 -04002576}
2577
Jeff Layton80334262013-07-26 06:23:25 -04002578/**
Al Viro197df042013-09-08 14:03:27 -04002579 * mountpoint_last - look up last component for umount
Jeff Layton80334262013-07-26 06:23:25 -04002580 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2581 * @path: pointer to container for result
2582 *
2583 * This is a special lookup_last function just for umount. In this case, we
2584 * need to resolve the path without doing any revalidation.
2585 *
2586 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2587 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2588 * in almost all cases, this lookup will be served out of the dcache. The only
2589 * cases where it won't are if nd->last refers to a symlink or the path is
2590 * bogus and it doesn't exist.
2591 *
2592 * Returns:
2593 * -error: if there was an error during lookup. This includes -ENOENT if the
2594 * lookup found a negative dentry. The nd->path reference will also be
2595 * put in this case.
2596 *
2597 * 0: if we successfully resolved nd->path and found it to not to be a
2598 * symlink that needs to be followed. "path" will also be populated.
2599 * The nd->path reference will also be put.
2600 *
2601 * 1: if we successfully resolved nd->last and found it to be a symlink
2602 * that needs to be followed. "path" will be populated with the path
2603 * to the link, and nd->path will *not* be put.
2604 */
2605static int
Al Viro197df042013-09-08 14:03:27 -04002606mountpoint_last(struct nameidata *nd, struct path *path)
Jeff Layton80334262013-07-26 06:23:25 -04002607{
2608 int error = 0;
2609 struct dentry *dentry;
2610 struct dentry *dir = nd->path.dentry;
2611
Al Viro35759522013-09-08 13:41:33 -04002612 /* If we're in rcuwalk, drop out of it to handle last component */
2613 if (nd->flags & LOOKUP_RCU) {
Al Viro6e9918b2015-05-05 09:26:05 -04002614 if (unlazy_walk(nd, NULL, 0))
Al Virodeb106c2015-05-08 18:05:21 -04002615 return -ECHILD;
Jeff Layton80334262013-07-26 06:23:25 -04002616 }
2617
2618 nd->flags &= ~LOOKUP_PARENT;
2619
2620 if (unlikely(nd->last_type != LAST_NORM)) {
2621 error = handle_dots(nd, nd->last_type);
Al Viro35759522013-09-08 13:41:33 -04002622 if (error)
Al Virodeb106c2015-05-08 18:05:21 -04002623 return error;
Al Viro35759522013-09-08 13:41:33 -04002624 dentry = dget(nd->path.dentry);
Al Viro949a8522016-03-06 14:20:52 -05002625 } else {
2626 dentry = d_lookup(dir, &nd->last);
Jeff Layton80334262013-07-26 06:23:25 -04002627 if (!dentry) {
Al Viro949a8522016-03-06 14:20:52 -05002628 /*
2629 * No cached dentry. Mounted dentries are pinned in the
2630 * cache, so that means that this dentry is probably
2631 * a symlink or the path doesn't actually point
2632 * to a mounted dentry.
2633 */
2634 dentry = lookup_slow(&nd->last, dir,
2635 nd->flags | LOOKUP_NO_REVAL);
2636 if (IS_ERR(dentry))
2637 return PTR_ERR(dentry);
Dave Jonesbcceeeb2013-09-10 17:04:25 -04002638 }
Jeff Layton80334262013-07-26 06:23:25 -04002639 }
David Howells698934d2015-03-17 17:33:52 +00002640 if (d_is_negative(dentry)) {
Al Viro35759522013-09-08 13:41:33 -04002641 dput(dentry);
Al Virodeb106c2015-05-08 18:05:21 -04002642 return -ENOENT;
Jeff Layton80334262013-07-26 06:23:25 -04002643 }
Al Viro191d7f72015-05-04 08:26:45 -04002644 if (nd->depth)
2645 put_link(nd);
Al Viro35759522013-09-08 13:41:33 -04002646 path->dentry = dentry;
Vasily Averin295dc392014-07-21 12:30:23 +04002647 path->mnt = nd->path.mnt;
Al Viro181548c2015-05-07 19:54:34 -04002648 error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW,
2649 d_backing_inode(dentry), 0);
Al Virodeb106c2015-05-08 18:05:21 -04002650 if (unlikely(error))
Al Virod63ff282015-05-04 18:13:23 -04002651 return error;
Vasily Averin295dc392014-07-21 12:30:23 +04002652 mntget(path->mnt);
Al Viro35759522013-09-08 13:41:33 -04002653 follow_mount(path);
Al Virodeb106c2015-05-08 18:05:21 -04002654 return 0;
Jeff Layton80334262013-07-26 06:23:25 -04002655}
2656
2657/**
Al Viro197df042013-09-08 14:03:27 -04002658 * path_mountpoint - look up a path to be umounted
Masanari Iida2a78b852015-09-09 15:39:23 -07002659 * @nd: lookup context
Jeff Layton80334262013-07-26 06:23:25 -04002660 * @flags: lookup flags
Al Viroc8a53ee2015-05-12 18:43:07 -04002661 * @path: pointer to container for result
Jeff Layton80334262013-07-26 06:23:25 -04002662 *
2663 * Look up the given name, but don't attempt to revalidate the last component.
Randy Dunlap606d6fe2013-10-19 14:56:55 -07002664 * Returns 0 and "path" will be valid on success; Returns error otherwise.
Jeff Layton80334262013-07-26 06:23:25 -04002665 */
2666static int
Al Viroc8a53ee2015-05-12 18:43:07 -04002667path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
Jeff Layton80334262013-07-26 06:23:25 -04002668{
Al Viroc8a53ee2015-05-12 18:43:07 -04002669 const char *s = path_init(nd, flags);
Al Viro368ee9b2015-05-08 17:19:59 -04002670 int err;
2671 if (IS_ERR(s))
2672 return PTR_ERR(s);
Al Viro3bdba282015-05-08 17:37:07 -04002673 while (!(err = link_path_walk(s, nd)) &&
2674 (err = mountpoint_last(nd, path)) > 0) {
2675 s = trailing_symlink(nd);
2676 if (IS_ERR(s)) {
2677 err = PTR_ERR(s);
Jeff Layton80334262013-07-26 06:23:25 -04002678 break;
Al Viro3bdba282015-05-08 17:37:07 -04002679 }
Jeff Layton80334262013-07-26 06:23:25 -04002680 }
Al Virodeb106c2015-05-08 18:05:21 -04002681 terminate_walk(nd);
Jeff Layton80334262013-07-26 06:23:25 -04002682 return err;
2683}
2684
Al Viro2d864652013-09-08 20:18:44 -04002685static int
Al Viro668696d2015-02-22 19:44:00 -05002686filename_mountpoint(int dfd, struct filename *name, struct path *path,
Al Viro2d864652013-09-08 20:18:44 -04002687 unsigned int flags)
2688{
Al Viro9883d182015-05-13 07:28:08 -04002689 struct nameidata nd;
Al Virocbaab2d2015-01-22 02:49:00 -05002690 int error;
Al Viro668696d2015-02-22 19:44:00 -05002691 if (IS_ERR(name))
2692 return PTR_ERR(name);
Al Viro9883d182015-05-13 07:28:08 -04002693 set_nameidata(&nd, dfd, name);
Al Viroc8a53ee2015-05-12 18:43:07 -04002694 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
Al Viro2d864652013-09-08 20:18:44 -04002695 if (unlikely(error == -ECHILD))
Al Viroc8a53ee2015-05-12 18:43:07 -04002696 error = path_mountpoint(&nd, flags, path);
Al Viro2d864652013-09-08 20:18:44 -04002697 if (unlikely(error == -ESTALE))
Al Viroc8a53ee2015-05-12 18:43:07 -04002698 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
Al Viro2d864652013-09-08 20:18:44 -04002699 if (likely(!error))
Al Viro668696d2015-02-22 19:44:00 -05002700 audit_inode(name, path->dentry, 0);
Al Viro9883d182015-05-13 07:28:08 -04002701 restore_nameidata();
Al Viro668696d2015-02-22 19:44:00 -05002702 putname(name);
Al Viro2d864652013-09-08 20:18:44 -04002703 return error;
2704}
2705
Jeff Layton80334262013-07-26 06:23:25 -04002706/**
Al Viro197df042013-09-08 14:03:27 -04002707 * user_path_mountpoint_at - lookup a path from userland in order to umount it
Jeff Layton80334262013-07-26 06:23:25 -04002708 * @dfd: directory file descriptor
2709 * @name: pathname from userland
2710 * @flags: lookup flags
2711 * @path: pointer to container to hold result
2712 *
2713 * A umount is a special case for path walking. We're not actually interested
2714 * in the inode in this situation, and ESTALE errors can be a problem. We
2715 * simply want track down the dentry and vfsmount attached at the mountpoint
2716 * and avoid revalidating the last component.
2717 *
2718 * Returns 0 and populates "path" on success.
2719 */
2720int
Al Viro197df042013-09-08 14:03:27 -04002721user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
Jeff Layton80334262013-07-26 06:23:25 -04002722 struct path *path)
2723{
Al Virocbaab2d2015-01-22 02:49:00 -05002724 return filename_mountpoint(dfd, getname(name), path, flags);
Jeff Layton80334262013-07-26 06:23:25 -04002725}
2726
Al Viro2d864652013-09-08 20:18:44 -04002727int
2728kern_path_mountpoint(int dfd, const char *name, struct path *path,
2729 unsigned int flags)
2730{
Al Virocbaab2d2015-01-22 02:49:00 -05002731 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
Al Viro2d864652013-09-08 20:18:44 -04002732}
2733EXPORT_SYMBOL(kern_path_mountpoint);
2734
Miklos Szeredicbdf35b2014-10-24 00:14:36 +02002735int __check_sticky(struct inode *dir, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002736{
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002737 kuid_t fsuid = current_fsuid();
David Howellsda9592e2008-11-14 10:39:05 +11002738
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002739 if (uid_eq(inode->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002740 return 0;
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002741 if (uid_eq(dir->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742 return 0;
Andy Lutomirski23adbe12014-06-10 12:45:42 -07002743 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744}
Miklos Szeredicbdf35b2014-10-24 00:14:36 +02002745EXPORT_SYMBOL(__check_sticky);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746
2747/*
2748 * Check whether we can remove a link victim from directory dir, check
2749 * whether the type of victim is right.
2750 * 1. We can't do it if dir is read-only (done in permission())
2751 * 2. We should have write and exec permissions on dir
2752 * 3. We can't remove anything from append-only dir
2753 * 4. We can't do anything with immutable dir (done in permission())
2754 * 5. If the sticky bit on dir is set we should either
2755 * a. be owner of dir, or
2756 * b. be owner of victim, or
2757 * c. have CAP_FOWNER capability
2758 * 6. If the victim is append-only or immutable we can't do antyhing with
2759 * links pointing to it.
2760 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2761 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2762 * 9. We can't remove a root or mountpoint.
2763 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2764 * nfs_async_unlink().
2765 */
David Howellsb18825a2013-09-12 19:22:53 +01002766static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002767{
David Howells63afdfc2015-05-06 15:59:00 +01002768 struct inode *inode = d_backing_inode(victim);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002769 int error;
2770
David Howellsb18825a2013-09-12 19:22:53 +01002771 if (d_is_negative(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002772 return -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01002773 BUG_ON(!inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774
2775 BUG_ON(victim->d_parent->d_inode != dir);
Jeff Layton4fa6b5e2012-10-10 15:25:25 -04002776 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777
Al Virof419a2e2008-07-22 00:07:17 -04002778 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779 if (error)
2780 return error;
2781 if (IS_APPEND(dir))
2782 return -EPERM;
David Howellsb18825a2013-09-12 19:22:53 +01002783
2784 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2785 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786 return -EPERM;
2787 if (isdir) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002788 if (!d_is_dir(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002789 return -ENOTDIR;
2790 if (IS_ROOT(victim))
2791 return -EBUSY;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002792 } else if (d_is_dir(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002793 return -EISDIR;
2794 if (IS_DEADDIR(dir))
2795 return -ENOENT;
2796 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2797 return -EBUSY;
2798 return 0;
2799}
2800
2801/* Check whether we can create an object with dentry child in directory
2802 * dir.
2803 * 1. We can't do it if child already exists (open has special treatment for
2804 * this case, but since we are inlined it's OK)
2805 * 2. We can't do it if dir is read-only (done in permission())
2806 * 3. We should have write and exec permissions on dir
2807 * 4. We can't do it if dir is immutable (done in permission())
2808 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02002809static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002810{
Jeff Layton14e972b2013-05-08 10:25:58 -04002811 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812 if (child->d_inode)
2813 return -EEXIST;
2814 if (IS_DEADDIR(dir))
2815 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04002816 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817}
2818
Linus Torvalds1da177e2005-04-16 15:20:36 -07002819/*
2820 * p1 and p2 should be directories on the same fs.
2821 */
2822struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2823{
2824 struct dentry *p;
2825
2826 if (p1 == p2) {
Al Viro59551022016-01-22 15:40:57 -05002827 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002828 return NULL;
2829 }
2830
Al Virofc640052016-04-10 01:33:30 -04002831 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002832
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002833 p = d_ancestor(p2, p1);
2834 if (p) {
Al Viro59551022016-01-22 15:40:57 -05002835 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2836 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002837 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838 }
2839
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002840 p = d_ancestor(p1, p2);
2841 if (p) {
Al Viro59551022016-01-22 15:40:57 -05002842 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2843 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002844 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845 }
2846
Al Viro59551022016-01-22 15:40:57 -05002847 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2848 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002849 return NULL;
2850}
Al Viro4d359502014-03-14 12:20:17 -04002851EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852
2853void unlock_rename(struct dentry *p1, struct dentry *p2)
2854{
Al Viro59551022016-01-22 15:40:57 -05002855 inode_unlock(p1->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002856 if (p1 != p2) {
Al Viro59551022016-01-22 15:40:57 -05002857 inode_unlock(p2->d_inode);
Al Virofc640052016-04-10 01:33:30 -04002858 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002859 }
2860}
Al Viro4d359502014-03-14 12:20:17 -04002861EXPORT_SYMBOL(unlock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862
Al Viro4acdaf22011-07-26 01:42:34 -04002863int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viro312b63f2012-06-10 18:09:36 -04002864 bool want_excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002865{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002866 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867 if (error)
2868 return error;
2869
Al Viroacfa4382008-12-04 10:06:33 -05002870 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002871 return -EACCES; /* shouldn't it be ENOSYS? */
2872 mode &= S_IALLUGO;
2873 mode |= S_IFREG;
2874 error = security_inode_create(dir, dentry, mode);
2875 if (error)
2876 return error;
Al Viro312b63f2012-06-10 18:09:36 -04002877 error = dir->i_op->create(dir, dentry, mode, want_excl);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002878 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002879 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002880 return error;
2881}
Al Viro4d359502014-03-14 12:20:17 -04002882EXPORT_SYMBOL(vfs_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883
Al Viro73d049a2011-03-11 12:08:24 -05002884static int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002885{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002886 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002887 struct inode *inode = dentry->d_inode;
2888 int error;
2889
2890 if (!inode)
2891 return -ENOENT;
2892
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002893 switch (inode->i_mode & S_IFMT) {
2894 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002895 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002896 case S_IFDIR:
2897 if (acc_mode & MAY_WRITE)
2898 return -EISDIR;
2899 break;
2900 case S_IFBLK:
2901 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002902 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002903 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002904 /*FALLTHRU*/
2905 case S_IFIFO:
2906 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002908 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002909 }
Dave Hansenb41572e2007-10-16 23:31:14 -07002910
Al Viro62fb4a12015-12-26 22:33:24 -05002911 error = inode_permission(inode, MAY_OPEN | acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07002912 if (error)
2913 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05002914
Linus Torvalds1da177e2005-04-16 15:20:36 -07002915 /*
2916 * An append-only file must be opened in append mode for writing.
2917 */
2918 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05002919 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05002920 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002921 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05002922 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002923 }
2924
2925 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002926 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05002927 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002928
J. Bruce Fieldsf3c7691e2011-09-21 10:58:13 -04002929 return 0;
Al Viro7715b522009-12-16 03:54:00 -05002930}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002931
Jeff Laytone1181ee2010-12-07 16:19:50 -05002932static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002933{
Jeff Laytone1181ee2010-12-07 16:19:50 -05002934 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002935 struct inode *inode = path->dentry->d_inode;
2936 int error = get_write_access(inode);
2937 if (error)
2938 return error;
2939 /*
2940 * Refuse to truncate files with mandatory locks held on them.
2941 */
Jeff Laytond7a06982014-03-10 09:54:15 -04002942 error = locks_verify_locked(filp);
Al Viro7715b522009-12-16 03:54:00 -05002943 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002944 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002945 if (!error) {
2946 error = do_truncate(path->dentry, 0,
2947 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002948 filp);
Al Viro7715b522009-12-16 03:54:00 -05002949 }
2950 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002951 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002952}
2953
Dave Hansend57999e2008-02-15 14:37:27 -08002954static inline int open_to_namei_flags(int flag)
2955{
Al Viro8a5e9292011-06-25 19:15:54 -04002956 if ((flag & O_ACCMODE) == 3)
2957 flag--;
Dave Hansend57999e2008-02-15 14:37:27 -08002958 return flag;
2959}
2960
Al Virod3607752016-03-25 15:21:09 -04002961static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002962{
2963 int error = security_path_mknod(dir, dentry, mode, 0);
2964 if (error)
2965 return error;
2966
2967 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2968 if (error)
2969 return error;
2970
2971 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2972}
2973
David Howells1acf0af2012-06-14 16:13:46 +01002974/*
2975 * Attempt to atomically look up, create and open a file from a negative
2976 * dentry.
2977 *
2978 * Returns 0 if successful. The file will have been created and attached to
2979 * @file by the filesystem calling finish_open().
2980 *
2981 * Returns 1 if the file was looked up only or didn't need creating. The
2982 * caller will need to perform the open themselves. @path will have been
2983 * updated to point to the new dentry. This may be negative.
2984 *
2985 * Returns an error code otherwise.
2986 */
Al Viro2675a4e2012-06-22 12:41:10 +04002987static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2988 struct path *path, struct file *file,
2989 const struct open_flags *op,
Al Viro1643b432016-04-27 19:14:10 -04002990 int open_flag, umode_t mode,
Al Viro2675a4e2012-06-22 12:41:10 +04002991 int *opened)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002992{
Miklos Szeredid18e9002012-06-05 15:10:17 +02002993 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002994 struct inode *dir = nd->path.dentry->d_inode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002995 int error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002996
Al Viro384f26e2016-04-28 02:03:55 -04002997 if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */
Miklos Szeredid18e9002012-06-05 15:10:17 +02002998 open_flag &= ~O_TRUNC;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002999
Miklos Szeredid18e9002012-06-05 15:10:17 +02003000 if (nd->flags & LOOKUP_DIRECTORY)
3001 open_flag |= O_DIRECTORY;
3002
Al Viro30d90492012-06-22 12:40:19 +04003003 file->f_path.dentry = DENTRY_NOT_SET;
3004 file->f_path.mnt = nd->path.mnt;
Al Viro0fb1ea02016-04-27 14:13:10 -04003005 error = dir->i_op->atomic_open(dir, dentry, file,
3006 open_to_namei_flags(open_flag),
3007 mode, opened);
Al Viro6fbd0712016-04-28 11:50:59 -04003008 d_lookup_done(dentry);
Al Viro384f26e2016-04-28 02:03:55 -04003009 if (!error) {
3010 /*
3011 * We didn't have the inode before the open, so check open
3012 * permission here.
3013 */
3014 int acc_mode = op->acc_mode;
3015 if (*opened & FILE_CREATED) {
3016 WARN_ON(!(open_flag & O_CREAT));
3017 fsnotify_create(dir, dentry);
3018 acc_mode = 0;
3019 }
3020 error = may_open(&file->f_path, acc_mode, open_flag);
3021 if (WARN_ON(error > 0))
3022 error = -EINVAL;
3023 } else if (error > 0) {
Al Viro30d90492012-06-22 12:40:19 +04003024 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
Al Viro2675a4e2012-06-22 12:41:10 +04003025 error = -EIO;
Al Viro03da6332013-09-16 19:22:33 -04003026 } else {
Al Viro384f26e2016-04-28 02:03:55 -04003027 if (file->f_path.dentry) {
3028 dput(dentry);
3029 dentry = file->f_path.dentry;
Al Viro03da6332013-09-16 19:22:33 -04003030 }
Al Viro384f26e2016-04-28 02:03:55 -04003031 if (*opened & FILE_CREATED)
3032 fsnotify_create(dir, dentry);
Al Viroa01e7182016-06-07 21:53:51 -04003033 if (unlikely(d_is_negative(dentry))) {
3034 error = -ENOENT;
3035 } else {
3036 path->dentry = dentry;
3037 path->mnt = nd->path.mnt;
3038 return 1;
3039 }
Sage Weil62b2ce92012-08-15 13:30:12 -07003040 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02003041 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02003042 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04003043 return error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003044}
3045
Nick Piggin31e6b012011-01-07 17:49:52 +11003046/*
David Howells1acf0af2012-06-14 16:13:46 +01003047 * Look up and maybe create and open the last component.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003048 *
3049 * Must be called with i_mutex held on parent.
3050 *
David Howells1acf0af2012-06-14 16:13:46 +01003051 * Returns 0 if the file was successfully atomically created (if necessary) and
3052 * opened. In this case the file will be returned attached to @file.
3053 *
3054 * Returns 1 if the file was not completely opened at this time, though lookups
3055 * and creations will have been performed and the dentry returned in @path will
3056 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
3057 * specified then a negative dentry may be returned.
3058 *
3059 * An error code is returned otherwise.
3060 *
3061 * FILE_CREATE will be set in @*opened if the dentry was created and will be
3062 * cleared otherwise prior to returning.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003063 */
Al Viro2675a4e2012-06-22 12:41:10 +04003064static int lookup_open(struct nameidata *nd, struct path *path,
3065 struct file *file,
3066 const struct open_flags *op,
Al Viro64894cf2012-07-31 00:53:35 +04003067 bool got_write, int *opened)
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003068{
3069 struct dentry *dir = nd->path.dentry;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02003070 struct inode *dir_inode = dir->d_inode;
Al Viro1643b432016-04-27 19:14:10 -04003071 int open_flag = op->open_flag;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003072 struct dentry *dentry;
Al Viro1643b432016-04-27 19:14:10 -04003073 int error, create_error = 0;
Al Viro1643b432016-04-27 19:14:10 -04003074 umode_t mode = op->mode;
Al Viro6fbd0712016-04-28 11:50:59 -04003075 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003076
Al Viroce8644f2016-04-26 14:17:56 -04003077 if (unlikely(IS_DEADDIR(dir_inode)))
3078 return -ENOENT;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003079
Al Viro47237682012-06-10 05:01:45 -04003080 *opened &= ~FILE_CREATED;
Al Viro6fbd0712016-04-28 11:50:59 -04003081 dentry = d_lookup(dir, &nd->last);
3082 for (;;) {
3083 if (!dentry) {
3084 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3085 if (IS_ERR(dentry))
3086 return PTR_ERR(dentry);
3087 }
3088 if (d_in_lookup(dentry))
3089 break;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003090
Al Viro6fbd0712016-04-28 11:50:59 -04003091 if (!(dentry->d_flags & DCACHE_OP_REVALIDATE))
3092 break;
3093
3094 error = d_revalidate(dentry, nd->flags);
3095 if (likely(error > 0))
3096 break;
3097 if (error)
3098 goto out_dput;
3099 d_invalidate(dentry);
3100 dput(dentry);
3101 dentry = NULL;
3102 }
3103 if (dentry->d_inode) {
Al Viro6c51e512016-03-05 20:09:32 -05003104 /* Cached positive dentry: will open in f_op->open */
Miklos Szeredid18e9002012-06-05 15:10:17 +02003105 goto out_no_open;
Al Viro6c51e512016-03-05 20:09:32 -05003106 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02003107
Al Viro1643b432016-04-27 19:14:10 -04003108 /*
3109 * Checking write permission is tricky, bacuse we don't know if we are
3110 * going to actually need it: O_CREAT opens should work as long as the
3111 * file exists. But checking existence breaks atomicity. The trick is
3112 * to check access and if not granted clear O_CREAT from the flags.
3113 *
3114 * Another problem is returing the "right" error value (e.g. for an
3115 * O_EXCL open we want to return EEXIST not EROFS).
3116 */
3117 if (open_flag & O_CREAT) {
3118 if (!IS_POSIXACL(dir->d_inode))
3119 mode &= ~current_umask();
3120 if (unlikely(!got_write)) {
3121 create_error = -EROFS;
3122 open_flag &= ~O_CREAT;
3123 if (open_flag & (O_EXCL | O_TRUNC))
3124 goto no_open;
3125 /* No side effects, safe to clear O_CREAT */
3126 } else {
3127 create_error = may_o_create(&nd->path, dentry, mode);
3128 if (create_error) {
3129 open_flag &= ~O_CREAT;
3130 if (open_flag & O_EXCL)
3131 goto no_open;
3132 }
3133 }
3134 } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3135 unlikely(!got_write)) {
3136 /*
3137 * No O_CREATE -> atomicity not a requirement -> fall
3138 * back to lookup + open
3139 */
3140 goto no_open;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003141 }
3142
Al Viro1643b432016-04-27 19:14:10 -04003143 if (dir_inode->i_op->atomic_open) {
3144 error = atomic_open(nd, dentry, path, file, op, open_flag,
3145 mode, opened);
3146 if (unlikely(error == -ENOENT) && create_error)
3147 error = create_error;
3148 return error;
3149 }
Miklos Szeredi54ef4872012-06-05 15:10:16 +02003150
Al Viro1643b432016-04-27 19:14:10 -04003151no_open:
Al Viro6fbd0712016-04-28 11:50:59 -04003152 if (d_in_lookup(dentry)) {
Al Viro12fa5e22016-04-28 11:19:43 -04003153 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3154 nd->flags);
Al Viro6fbd0712016-04-28 11:50:59 -04003155 d_lookup_done(dentry);
Al Viro12fa5e22016-04-28 11:19:43 -04003156 if (unlikely(res)) {
3157 if (IS_ERR(res)) {
3158 error = PTR_ERR(res);
3159 goto out_dput;
3160 }
3161 dput(dentry);
3162 dentry = res;
3163 }
Miklos Szeredi54ef4872012-06-05 15:10:16 +02003164 }
3165
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003166 /* Negative dentry, just create the file */
Al Viro1643b432016-04-27 19:14:10 -04003167 if (!dentry->d_inode && (open_flag & O_CREAT)) {
Al Viro47237682012-06-10 05:01:45 -04003168 *opened |= FILE_CREATED;
Al Viroce8644f2016-04-26 14:17:56 -04003169 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
Al Viroce8644f2016-04-26 14:17:56 -04003170 if (!dir_inode->i_op->create) {
3171 error = -EACCES;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003172 goto out_dput;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003173 }
Al Viroce8644f2016-04-26 14:17:56 -04003174 error = dir_inode->i_op->create(dir_inode, dentry, mode,
Al Viro1643b432016-04-27 19:14:10 -04003175 open_flag & O_EXCL);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003176 if (error)
3177 goto out_dput;
Al Viroce8644f2016-04-26 14:17:56 -04003178 fsnotify_create(dir_inode, dentry);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003179 }
Al Viro1643b432016-04-27 19:14:10 -04003180 if (unlikely(create_error) && !dentry->d_inode) {
3181 error = create_error;
3182 goto out_dput;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003183 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02003184out_no_open:
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003185 path->dentry = dentry;
3186 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04003187 return 1;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003188
3189out_dput:
3190 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04003191 return error;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003192}
3193
3194/*
Al Virofe2d35f2011-03-05 22:58:25 -05003195 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11003196 */
Al Viro896475d2015-04-22 18:02:17 -04003197static int do_last(struct nameidata *nd,
Al Viro2675a4e2012-06-22 12:41:10 +04003198 struct file *file, const struct open_flags *op,
Al Viro76ae2a52015-05-12 18:44:32 -04003199 int *opened)
Al Virofb1cc552009-12-24 01:58:28 -05003200{
Al Viroa1e28032009-12-24 02:12:06 -05003201 struct dentry *dir = nd->path.dentry;
Al Viroca344a892011-03-09 00:36:45 -05003202 int open_flag = op->open_flag;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003203 bool will_truncate = (open_flag & O_TRUNC) != 0;
Al Viro64894cf2012-07-31 00:53:35 +04003204 bool got_write = false;
Al Virobcda7652011-03-13 16:42:14 -04003205 int acc_mode = op->acc_mode;
Al Viro254cf582015-05-05 09:40:46 -04003206 unsigned seq;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02003207 struct inode *inode;
Al Viro896475d2015-04-22 18:02:17 -04003208 struct path path;
Al Viro16c2cd72011-02-22 15:50:10 -05003209 int error;
Al Virofb1cc552009-12-24 01:58:28 -05003210
Al Viroc3e380b2011-02-23 13:39:45 -05003211 nd->flags &= ~LOOKUP_PARENT;
3212 nd->flags |= op->intent;
3213
Al Virobc77daa2013-06-06 09:12:33 -04003214 if (nd->last_type != LAST_NORM) {
Al Virofe2d35f2011-03-05 22:58:25 -05003215 error = handle_dots(nd, nd->last_type);
Al Virodeb106c2015-05-08 18:05:21 -04003216 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04003217 return error;
Miklos Szeredie83db162012-06-05 15:10:29 +02003218 goto finish_open;
Al Viro1f36f772009-12-26 10:56:19 -05003219 }
Al Viro67ee3ad2009-12-26 07:01:01 -05003220
Al Viroca344a892011-03-09 00:36:45 -05003221 if (!(open_flag & O_CREAT)) {
Al Virofe2d35f2011-03-05 22:58:25 -05003222 if (nd->last.name[nd->last.len])
3223 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3224 /* we _can_ be in RCU mode here */
Al Viro254cf582015-05-05 09:40:46 -04003225 error = lookup_fast(nd, &path, &inode, &seq);
Al Viroe9742b52016-03-05 22:04:59 -05003226 if (likely(error > 0))
Miklos Szeredi71574862012-06-05 15:10:14 +02003227 goto finish_lookup;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02003228
Miklos Szeredi71574862012-06-05 15:10:14 +02003229 if (error < 0)
Al Virodeb106c2015-05-08 18:05:21 -04003230 return error;
Miklos Szeredi37d7fff2012-06-05 15:10:12 +02003231
Miklos Szeredi71574862012-06-05 15:10:14 +02003232 BUG_ON(nd->inode != dir->d_inode);
Al Viro6583fe22016-03-05 18:14:03 -05003233 BUG_ON(nd->flags & LOOKUP_RCU);
Miklos Szeredib6183df2012-06-05 15:10:13 +02003234 } else {
3235 /* create side of things */
3236 /*
3237 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3238 * has been cleared when we got to the last component we are
3239 * about to look up
3240 */
3241 error = complete_walk(nd);
Al Viroe8bb73d2015-05-08 16:28:42 -04003242 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003243 return error;
Miklos Szeredib6183df2012-06-05 15:10:13 +02003244
Al Viro76ae2a52015-05-12 18:44:32 -04003245 audit_inode(nd->name, dir, LOOKUP_PARENT);
Miklos Szeredib6183df2012-06-05 15:10:13 +02003246 /* trailing slashes? */
Al Virodeb106c2015-05-08 18:05:21 -04003247 if (unlikely(nd->last.name[nd->last.len]))
3248 return -EISDIR;
Al Virofe2d35f2011-03-05 22:58:25 -05003249 }
3250
Al Viro9cf843e2016-04-28 19:35:16 -04003251 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
Al Viro64894cf2012-07-31 00:53:35 +04003252 error = mnt_want_write(nd->path.mnt);
3253 if (!error)
3254 got_write = true;
3255 /*
3256 * do _not_ fail yet - we might not need that or fail with
3257 * a different error; let lookup_open() decide; we'll be
3258 * dropping this one anyway.
3259 */
3260 }
Al Viro9cf843e2016-04-28 19:35:16 -04003261 if (open_flag & O_CREAT)
3262 inode_lock(dir->d_inode);
3263 else
3264 inode_lock_shared(dir->d_inode);
Al Viro896475d2015-04-22 18:02:17 -04003265 error = lookup_open(nd, &path, file, op, got_write, opened);
Al Viro9cf843e2016-04-28 19:35:16 -04003266 if (open_flag & O_CREAT)
3267 inode_unlock(dir->d_inode);
3268 else
3269 inode_unlock_shared(dir->d_inode);
Al Viroa1e28032009-12-24 02:12:06 -05003270
Al Viro2675a4e2012-06-22 12:41:10 +04003271 if (error <= 0) {
3272 if (error)
Miklos Szeredid18e9002012-06-05 15:10:17 +02003273 goto out;
3274
Al Viro47237682012-06-10 05:01:45 -04003275 if ((*opened & FILE_CREATED) ||
Al Viro496ad9a2013-01-23 17:07:38 -05003276 !S_ISREG(file_inode(file)->i_mode))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003277 will_truncate = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003278
Al Viro76ae2a52015-05-12 18:44:32 -04003279 audit_inode(nd->name, file->f_path.dentry, 0);
Miklos Szeredid18e9002012-06-05 15:10:17 +02003280 goto opened;
3281 }
Al Virofb1cc552009-12-24 01:58:28 -05003282
Al Viro47237682012-06-10 05:01:45 -04003283 if (*opened & FILE_CREATED) {
Al Viro9b44f1b2011-03-09 00:17:27 -05003284 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05003285 open_flag &= ~O_TRUNC;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003286 will_truncate = false;
Al Viro62fb4a12015-12-26 22:33:24 -05003287 acc_mode = 0;
Al Viro896475d2015-04-22 18:02:17 -04003288 path_to_nameidata(&path, nd);
Miklos Szeredie83db162012-06-05 15:10:29 +02003289 goto finish_open_created;
Al Virofb1cc552009-12-24 01:58:28 -05003290 }
3291
3292 /*
Miklos Szeredid18e9002012-06-05 15:10:17 +02003293 * If atomic_open() acquired write access it is dropped now due to
3294 * possible mount and symlink following (this might be optimized away if
3295 * necessary...)
3296 */
Al Viro64894cf2012-07-31 00:53:35 +04003297 if (got_write) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02003298 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04003299 got_write = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003300 }
3301
Al Viroe6ec03a2016-06-05 00:23:09 -04003302 error = follow_managed(&path, nd);
3303 if (unlikely(error < 0))
3304 return error;
3305
Al Viro6583fe22016-03-05 18:14:03 -05003306 if (unlikely(d_is_negative(path.dentry))) {
3307 path_to_nameidata(&path, nd);
3308 return -ENOENT;
3309 }
3310
3311 /*
3312 * create/update audit record if it already exists.
3313 */
3314 audit_inode(nd->name, path.dentry, 0);
3315
Al Virodeb106c2015-05-08 18:05:21 -04003316 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3317 path_to_nameidata(&path, nd);
3318 return -EEXIST;
3319 }
Al Virofb1cc552009-12-24 01:58:28 -05003320
Al Viro254cf582015-05-05 09:40:46 -04003321 seq = 0; /* out of RCU mode, so the value doesn't matter */
Al Virod4565642016-02-27 19:23:16 -05003322 inode = d_backing_inode(path.dentry);
Al Viro766c4cb2015-05-07 19:24:57 -04003323finish_lookup:
Al Virod63ff282015-05-04 18:13:23 -04003324 if (nd->depth)
3325 put_link(nd);
Al Viro181548c2015-05-07 19:54:34 -04003326 error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW,
3327 inode, seq);
Al Virodeb106c2015-05-08 18:05:21 -04003328 if (unlikely(error))
Al Virod63ff282015-05-04 18:13:23 -04003329 return error;
Al Virofb1cc552009-12-24 01:58:28 -05003330
Al Virofac7d192016-06-04 11:41:49 -04003331 path_to_nameidata(&path, nd);
Miklos Szeredidecf3402012-05-21 17:30:08 +02003332 nd->inode = inode;
Al Viro254cf582015-05-05 09:40:46 -04003333 nd->seq = seq;
Al Viroa3fbbde2011-11-07 21:21:26 +00003334 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
Al Virobc77daa2013-06-06 09:12:33 -04003335finish_open:
Al Viroa3fbbde2011-11-07 21:21:26 +00003336 error = complete_walk(nd);
Al Virofac7d192016-06-04 11:41:49 -04003337 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003338 return error;
Al Viro76ae2a52015-05-12 18:44:32 -04003339 audit_inode(nd->name, nd->path.dentry, 0);
Al Virofb1cc552009-12-24 01:58:28 -05003340 error = -EISDIR;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003341 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04003342 goto out;
Miklos Szerediaf2f5542012-05-21 17:30:11 +02003343 error = -ENOTDIR;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003344 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04003345 goto out;
David Howells4bbcbd3b2015-03-17 22:16:40 +00003346 if (!d_is_reg(nd->path.dentry))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003347 will_truncate = false;
Al Viro6c0d46c2011-03-09 00:59:59 -05003348
Al Viro0f9d1a12011-03-09 00:13:14 -05003349 if (will_truncate) {
3350 error = mnt_want_write(nd->path.mnt);
3351 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003352 goto out;
Al Viro64894cf2012-07-31 00:53:35 +04003353 got_write = true;
Al Viro0f9d1a12011-03-09 00:13:14 -05003354 }
Miklos Szeredie83db162012-06-05 15:10:29 +02003355finish_open_created:
Al Viro6ac08702016-04-26 00:02:50 -04003356 error = may_open(&nd->path, acc_mode, open_flag);
3357 if (error)
3358 goto out;
Miklos Szeredi4aa7c632014-10-24 00:14:35 +02003359 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3360 error = vfs_open(&nd->path, file, current_cred());
Al Virofac7d192016-06-04 11:41:49 -04003361 if (error)
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003362 goto out;
Al Virofac7d192016-06-04 11:41:49 -04003363 *opened |= FILE_OPENED;
Miklos Szeredia8277b92012-06-05 15:10:32 +02003364opened:
Al Viro2675a4e2012-06-22 12:41:10 +04003365 error = open_check_o_direct(file);
Al Virofe9ec822016-04-27 03:14:20 -04003366 if (!error)
3367 error = ima_file_check(file, op->acc_mode, *opened);
3368 if (!error && will_truncate)
Al Viro2675a4e2012-06-22 12:41:10 +04003369 error = handle_truncate(file);
Al Viroca344a892011-03-09 00:36:45 -05003370out:
Al Virofe9ec822016-04-27 03:14:20 -04003371 if (unlikely(error) && (*opened & FILE_OPENED))
3372 fput(file);
Al Viroc80567c2016-02-27 19:17:33 -05003373 if (unlikely(error > 0)) {
3374 WARN_ON(1);
3375 error = -EINVAL;
3376 }
Al Viro64894cf2012-07-31 00:53:35 +04003377 if (got_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05003378 mnt_drop_write(nd->path.mnt);
Al Viro2675a4e2012-06-22 12:41:10 +04003379 return error;
Al Virofb1cc552009-12-24 01:58:28 -05003380}
3381
Al Viroc8a53ee2015-05-12 18:43:07 -04003382static int do_tmpfile(struct nameidata *nd, unsigned flags,
Al Viro60545d02013-06-07 01:20:27 -04003383 const struct open_flags *op,
3384 struct file *file, int *opened)
3385{
3386 static const struct qstr name = QSTR_INIT("/", 1);
Al Viro625b6d12015-05-12 16:36:12 -04003387 struct dentry *child;
Al Viro60545d02013-06-07 01:20:27 -04003388 struct inode *dir;
Al Viro625b6d12015-05-12 16:36:12 -04003389 struct path path;
Al Viroc8a53ee2015-05-12 18:43:07 -04003390 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
Al Viro60545d02013-06-07 01:20:27 -04003391 if (unlikely(error))
3392 return error;
Al Viro625b6d12015-05-12 16:36:12 -04003393 error = mnt_want_write(path.mnt);
Al Viro60545d02013-06-07 01:20:27 -04003394 if (unlikely(error))
3395 goto out;
Al Viro625b6d12015-05-12 16:36:12 -04003396 dir = path.dentry->d_inode;
Al Viro60545d02013-06-07 01:20:27 -04003397 /* we want directory to be writable */
Al Viro625b6d12015-05-12 16:36:12 -04003398 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Al Viro60545d02013-06-07 01:20:27 -04003399 if (error)
3400 goto out2;
Al Viro60545d02013-06-07 01:20:27 -04003401 if (!dir->i_op->tmpfile) {
3402 error = -EOPNOTSUPP;
3403 goto out2;
3404 }
Al Viro625b6d12015-05-12 16:36:12 -04003405 child = d_alloc(path.dentry, &name);
Al Viro60545d02013-06-07 01:20:27 -04003406 if (unlikely(!child)) {
3407 error = -ENOMEM;
3408 goto out2;
3409 }
Al Viro625b6d12015-05-12 16:36:12 -04003410 dput(path.dentry);
3411 path.dentry = child;
3412 error = dir->i_op->tmpfile(dir, child, op->mode);
Al Viro60545d02013-06-07 01:20:27 -04003413 if (error)
3414 goto out2;
Al Viroc8a53ee2015-05-12 18:43:07 -04003415 audit_inode(nd->name, child, 0);
Eric Rannaud69a91c22014-10-30 01:51:01 -07003416 /* Don't check for other permissions, the inode was just created */
Al Viro62fb4a12015-12-26 22:33:24 -05003417 error = may_open(&path, 0, op->open_flag);
Al Viro60545d02013-06-07 01:20:27 -04003418 if (error)
3419 goto out2;
Al Viro625b6d12015-05-12 16:36:12 -04003420 file->f_path.mnt = path.mnt;
3421 error = finish_open(file, child, NULL, opened);
Al Viro60545d02013-06-07 01:20:27 -04003422 if (error)
3423 goto out2;
3424 error = open_check_o_direct(file);
Al Virof4e0c302013-06-11 08:34:36 +04003425 if (error) {
Al Viro60545d02013-06-07 01:20:27 -04003426 fput(file);
Al Virof4e0c302013-06-11 08:34:36 +04003427 } else if (!(op->open_flag & O_EXCL)) {
3428 struct inode *inode = file_inode(file);
3429 spin_lock(&inode->i_lock);
3430 inode->i_state |= I_LINKABLE;
3431 spin_unlock(&inode->i_lock);
3432 }
Al Viro60545d02013-06-07 01:20:27 -04003433out2:
Al Viro625b6d12015-05-12 16:36:12 -04003434 mnt_drop_write(path.mnt);
Al Viro60545d02013-06-07 01:20:27 -04003435out:
Al Viro625b6d12015-05-12 16:36:12 -04003436 path_put(&path);
Al Viro60545d02013-06-07 01:20:27 -04003437 return error;
3438}
3439
Al Viro6ac08702016-04-26 00:02:50 -04003440static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3441{
3442 struct path path;
3443 int error = path_lookupat(nd, flags, &path);
3444 if (!error) {
3445 audit_inode(nd->name, path.dentry, 0);
3446 error = vfs_open(&path, file, current_cred());
3447 path_put(&path);
3448 }
3449 return error;
3450}
3451
Al Viroc8a53ee2015-05-12 18:43:07 -04003452static struct file *path_openat(struct nameidata *nd,
3453 const struct open_flags *op, unsigned flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003454{
Al Viro368ee9b2015-05-08 17:19:59 -04003455 const char *s;
Al Viro30d90492012-06-22 12:40:19 +04003456 struct file *file;
Al Viro47237682012-06-10 05:01:45 -04003457 int opened = 0;
Al Viro13aab422011-02-23 17:54:08 -05003458 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11003459
Al Viro30d90492012-06-22 12:40:19 +04003460 file = get_empty_filp();
Al Viro1afc99b2013-02-14 20:41:04 -05003461 if (IS_ERR(file))
3462 return file;
Nick Piggin31e6b012011-01-07 17:49:52 +11003463
Al Viro30d90492012-06-22 12:40:19 +04003464 file->f_flags = op->open_flag;
Nick Piggin31e6b012011-01-07 17:49:52 +11003465
Al Virobb458c62013-07-13 13:26:37 +04003466 if (unlikely(file->f_flags & __O_TMPFILE)) {
Al Viroc8a53ee2015-05-12 18:43:07 -04003467 error = do_tmpfile(nd, flags, op, file, &opened);
Al Virof15133d2015-05-08 22:53:15 -04003468 goto out2;
Al Viro60545d02013-06-07 01:20:27 -04003469 }
3470
Al Viro6ac08702016-04-26 00:02:50 -04003471 if (unlikely(file->f_flags & O_PATH)) {
3472 error = do_o_path(nd, flags, file);
3473 if (!error)
3474 opened |= FILE_OPENED;
3475 goto out2;
3476 }
3477
Al Viroc8a53ee2015-05-12 18:43:07 -04003478 s = path_init(nd, flags);
Al Viro368ee9b2015-05-08 17:19:59 -04003479 if (IS_ERR(s)) {
3480 put_filp(file);
3481 return ERR_CAST(s);
3482 }
Al Viro3bdba282015-05-08 17:37:07 -04003483 while (!(error = link_path_walk(s, nd)) &&
Al Viro76ae2a52015-05-12 18:44:32 -04003484 (error = do_last(nd, file, op, &opened)) > 0) {
Al Viro73d049a2011-03-11 12:08:24 -05003485 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
Al Viro3bdba282015-05-08 17:37:07 -04003486 s = trailing_symlink(nd);
3487 if (IS_ERR(s)) {
3488 error = PTR_ERR(s);
Al Viro2675a4e2012-06-22 12:41:10 +04003489 break;
Al Viro3bdba282015-05-08 17:37:07 -04003490 }
Al Viro806b6812009-12-26 07:16:40 -05003491 }
Al Virodeb106c2015-05-08 18:05:21 -04003492 terminate_walk(nd);
Al Virof15133d2015-05-08 22:53:15 -04003493out2:
Al Viro2675a4e2012-06-22 12:41:10 +04003494 if (!(opened & FILE_OPENED)) {
3495 BUG_ON(!error);
Al Viro30d90492012-06-22 12:40:19 +04003496 put_filp(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003497 }
Al Viro2675a4e2012-06-22 12:41:10 +04003498 if (unlikely(error)) {
3499 if (error == -EOPENSTALE) {
3500 if (flags & LOOKUP_RCU)
3501 error = -ECHILD;
3502 else
3503 error = -ESTALE;
3504 }
3505 file = ERR_PTR(error);
3506 }
3507 return file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003508}
3509
Jeff Layton669abf42012-10-10 16:43:10 -04003510struct file *do_filp_open(int dfd, struct filename *pathname,
Al Virof9652e12013-06-11 08:23:01 +04003511 const struct open_flags *op)
Al Viro13aab422011-02-23 17:54:08 -05003512{
Al Viro9883d182015-05-13 07:28:08 -04003513 struct nameidata nd;
Al Virof9652e12013-06-11 08:23:01 +04003514 int flags = op->lookup_flags;
Al Viro13aab422011-02-23 17:54:08 -05003515 struct file *filp;
3516
Al Viro9883d182015-05-13 07:28:08 -04003517 set_nameidata(&nd, dfd, pathname);
Al Viroc8a53ee2015-05-12 18:43:07 -04003518 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
Al Viro13aab422011-02-23 17:54:08 -05003519 if (unlikely(filp == ERR_PTR(-ECHILD)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003520 filp = path_openat(&nd, op, flags);
Al Viro13aab422011-02-23 17:54:08 -05003521 if (unlikely(filp == ERR_PTR(-ESTALE)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003522 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
Al Viro9883d182015-05-13 07:28:08 -04003523 restore_nameidata();
Al Viro13aab422011-02-23 17:54:08 -05003524 return filp;
3525}
3526
Al Viro73d049a2011-03-11 12:08:24 -05003527struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
Al Virof9652e12013-06-11 08:23:01 +04003528 const char *name, const struct open_flags *op)
Al Viro73d049a2011-03-11 12:08:24 -05003529{
Al Viro9883d182015-05-13 07:28:08 -04003530 struct nameidata nd;
Al Viro73d049a2011-03-11 12:08:24 -05003531 struct file *file;
Paul Moore51689102015-01-22 00:00:03 -05003532 struct filename *filename;
Al Virof9652e12013-06-11 08:23:01 +04003533 int flags = op->lookup_flags | LOOKUP_ROOT;
Al Viro73d049a2011-03-11 12:08:24 -05003534
3535 nd.root.mnt = mnt;
3536 nd.root.dentry = dentry;
3537
David Howellsb18825a2013-09-12 19:22:53 +01003538 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
Al Viro73d049a2011-03-11 12:08:24 -05003539 return ERR_PTR(-ELOOP);
3540
Paul Moore51689102015-01-22 00:00:03 -05003541 filename = getname_kernel(name);
Viresh Kumara1c83682015-08-12 15:59:44 +05303542 if (IS_ERR(filename))
Paul Moore51689102015-01-22 00:00:03 -05003543 return ERR_CAST(filename);
3544
Al Viro9883d182015-05-13 07:28:08 -04003545 set_nameidata(&nd, -1, filename);
Al Viroc8a53ee2015-05-12 18:43:07 -04003546 file = path_openat(&nd, op, flags | LOOKUP_RCU);
Al Viro73d049a2011-03-11 12:08:24 -05003547 if (unlikely(file == ERR_PTR(-ECHILD)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003548 file = path_openat(&nd, op, flags);
Al Viro73d049a2011-03-11 12:08:24 -05003549 if (unlikely(file == ERR_PTR(-ESTALE)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003550 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
Al Viro9883d182015-05-13 07:28:08 -04003551 restore_nameidata();
Paul Moore51689102015-01-22 00:00:03 -05003552 putname(filename);
Al Viro73d049a2011-03-11 12:08:24 -05003553 return file;
3554}
3555
Al Virofa14a0b2015-01-22 02:16:49 -05003556static struct dentry *filename_create(int dfd, struct filename *name,
Jeff Layton1ac12b42012-12-11 12:10:06 -05003557 struct path *path, unsigned int lookup_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003558{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003559 struct dentry *dentry = ERR_PTR(-EEXIST);
Al Viro391172c2015-05-09 11:19:16 -04003560 struct qstr last;
3561 int type;
Jan Karac30dabf2012-06-12 16:20:30 +02003562 int err2;
Jeff Layton1ac12b42012-12-11 12:10:06 -05003563 int error;
3564 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3565
3566 /*
3567 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3568 * other flags passed in are ignored!
3569 */
3570 lookup_flags &= LOOKUP_REVAL;
3571
Al Viro5c31b6c2015-05-12 17:32:54 -04003572 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3573 if (IS_ERR(name))
3574 return ERR_CAST(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003575
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003576 /*
3577 * Yucky last component or no last component at all?
3578 * (foo/., foo/.., /////)
3579 */
Al Viro5c31b6c2015-05-12 17:32:54 -04003580 if (unlikely(type != LAST_NORM))
Al Viroed75e952011-06-27 16:53:43 -04003581 goto out;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003582
Jan Karac30dabf2012-06-12 16:20:30 +02003583 /* don't fail immediately if it's r/o, at least try to report other errors */
Al Viro391172c2015-05-09 11:19:16 -04003584 err2 = mnt_want_write(path->mnt);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003585 /*
3586 * Do the final lookup.
3587 */
Al Viro391172c2015-05-09 11:19:16 -04003588 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
Al Viro59551022016-01-22 15:40:57 -05003589 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
Al Viro391172c2015-05-09 11:19:16 -04003590 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003591 if (IS_ERR(dentry))
Al Viroa8104a92012-07-20 02:25:00 +04003592 goto unlock;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003593
Al Viroa8104a92012-07-20 02:25:00 +04003594 error = -EEXIST;
David Howellsb18825a2013-09-12 19:22:53 +01003595 if (d_is_positive(dentry))
Al Viroa8104a92012-07-20 02:25:00 +04003596 goto fail;
David Howellsb18825a2013-09-12 19:22:53 +01003597
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003598 /*
3599 * Special case - lookup gave negative, but... we had foo/bar/
3600 * From the vfs_mknod() POV we just have a negative dentry -
3601 * all is fine. Let's be bastards - you had / on the end, you've
3602 * been asking for (non-existent) directory. -ENOENT for you.
3603 */
Al Viro391172c2015-05-09 11:19:16 -04003604 if (unlikely(!is_dir && last.name[last.len])) {
Al Viroa8104a92012-07-20 02:25:00 +04003605 error = -ENOENT;
Al Viroed75e952011-06-27 16:53:43 -04003606 goto fail;
Al Viroe9baf6e2008-05-15 04:49:12 -04003607 }
Jan Karac30dabf2012-06-12 16:20:30 +02003608 if (unlikely(err2)) {
3609 error = err2;
Al Viroa8104a92012-07-20 02:25:00 +04003610 goto fail;
Jan Karac30dabf2012-06-12 16:20:30 +02003611 }
Al Viro181c37b2015-05-12 17:21:25 -04003612 putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003613 return dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003614fail:
Al Viroa8104a92012-07-20 02:25:00 +04003615 dput(dentry);
3616 dentry = ERR_PTR(error);
3617unlock:
Al Viro59551022016-01-22 15:40:57 -05003618 inode_unlock(path->dentry->d_inode);
Jan Karac30dabf2012-06-12 16:20:30 +02003619 if (!err2)
Al Viro391172c2015-05-09 11:19:16 -04003620 mnt_drop_write(path->mnt);
Al Viroed75e952011-06-27 16:53:43 -04003621out:
Al Viro391172c2015-05-09 11:19:16 -04003622 path_put(path);
Al Viro181c37b2015-05-12 17:21:25 -04003623 putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003624 return dentry;
3625}
Al Virofa14a0b2015-01-22 02:16:49 -05003626
3627struct dentry *kern_path_create(int dfd, const char *pathname,
3628 struct path *path, unsigned int lookup_flags)
3629{
Al Viro181c37b2015-05-12 17:21:25 -04003630 return filename_create(dfd, getname_kernel(pathname),
3631 path, lookup_flags);
Al Virofa14a0b2015-01-22 02:16:49 -05003632}
Al Virodae6ad82011-06-26 11:50:15 -04003633EXPORT_SYMBOL(kern_path_create);
3634
Al Viro921a1652012-07-20 01:15:31 +04003635void done_path_create(struct path *path, struct dentry *dentry)
3636{
3637 dput(dentry);
Al Viro59551022016-01-22 15:40:57 -05003638 inode_unlock(path->dentry->d_inode);
Al Viroa8104a92012-07-20 02:25:00 +04003639 mnt_drop_write(path->mnt);
Al Viro921a1652012-07-20 01:15:31 +04003640 path_put(path);
3641}
3642EXPORT_SYMBOL(done_path_create);
3643
Al Viro520ae682015-05-13 07:00:28 -04003644inline struct dentry *user_path_create(int dfd, const char __user *pathname,
Jeff Layton1ac12b42012-12-11 12:10:06 -05003645 struct path *path, unsigned int lookup_flags)
Al Virodae6ad82011-06-26 11:50:15 -04003646{
Al Viro181c37b2015-05-12 17:21:25 -04003647 return filename_create(dfd, getname(pathname), path, lookup_flags);
Al Virodae6ad82011-06-26 11:50:15 -04003648}
3649EXPORT_SYMBOL(user_path_create);
3650
Al Viro1a67aaf2011-07-26 01:52:52 -04003651int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003652{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003653 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003654
3655 if (error)
3656 return error;
3657
Eric W. Biederman975d6b32011-11-13 12:16:43 -08003658 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003659 return -EPERM;
3660
Al Viroacfa4382008-12-04 10:06:33 -05003661 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003662 return -EPERM;
3663
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07003664 error = devcgroup_inode_mknod(mode, dev);
3665 if (error)
3666 return error;
3667
Linus Torvalds1da177e2005-04-16 15:20:36 -07003668 error = security_inode_mknod(dir, dentry, mode, dev);
3669 if (error)
3670 return error;
3671
Linus Torvalds1da177e2005-04-16 15:20:36 -07003672 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003673 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003674 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003675 return error;
3676}
Al Viro4d359502014-03-14 12:20:17 -04003677EXPORT_SYMBOL(vfs_mknod);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003678
Al Virof69aac02011-07-26 04:31:40 -04003679static int may_mknod(umode_t mode)
Dave Hansen463c3192008-02-15 14:37:57 -08003680{
3681 switch (mode & S_IFMT) {
3682 case S_IFREG:
3683 case S_IFCHR:
3684 case S_IFBLK:
3685 case S_IFIFO:
3686 case S_IFSOCK:
3687 case 0: /* zero mode translates to S_IFREG */
3688 return 0;
3689 case S_IFDIR:
3690 return -EPERM;
3691 default:
3692 return -EINVAL;
3693 }
3694}
3695
Al Viro8208a222011-07-25 17:32:17 -04003696SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003697 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003698{
Al Viro2ad94ae2008-07-21 09:32:51 -04003699 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003700 struct path path;
3701 int error;
Jeff Layton972567f2012-12-20 16:00:10 -05003702 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003703
Al Viro8e4bfca2012-07-20 01:17:26 +04003704 error = may_mknod(mode);
3705 if (error)
3706 return error;
Jeff Layton972567f2012-12-20 16:00:10 -05003707retry:
3708 dentry = user_path_create(dfd, filename, &path, lookup_flags);
Al Virodae6ad82011-06-26 11:50:15 -04003709 if (IS_ERR(dentry))
3710 return PTR_ERR(dentry);
Al Viro2ad94ae2008-07-21 09:32:51 -04003711
Al Virodae6ad82011-06-26 11:50:15 -04003712 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003713 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003714 error = security_path_mknod(&path, dentry, mode, dev);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003715 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04003716 goto out;
Dave Hansen463c3192008-02-15 14:37:57 -08003717 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003718 case 0: case S_IFREG:
Al Viro312b63f2012-06-10 18:09:36 -04003719 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
Mimi Zohar05d1a712016-02-29 19:52:05 -05003720 if (!error)
3721 ima_post_path_mknod(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003722 break;
3723 case S_IFCHR: case S_IFBLK:
Al Virodae6ad82011-06-26 11:50:15 -04003724 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003725 new_decode_dev(dev));
3726 break;
3727 case S_IFIFO: case S_IFSOCK:
Al Virodae6ad82011-06-26 11:50:15 -04003728 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003729 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003730 }
Al Viroa8104a92012-07-20 02:25:00 +04003731out:
Al Viro921a1652012-07-20 01:15:31 +04003732 done_path_create(&path, dentry);
Jeff Layton972567f2012-12-20 16:00:10 -05003733 if (retry_estale(error, lookup_flags)) {
3734 lookup_flags |= LOOKUP_REVAL;
3735 goto retry;
3736 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003737 return error;
3738}
3739
Al Viro8208a222011-07-25 17:32:17 -04003740SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003741{
3742 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3743}
3744
Al Viro18bb1db2011-07-26 01:41:39 -04003745int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003746{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003747 int error = may_create(dir, dentry);
Al Viro8de52772012-02-06 12:45:27 -05003748 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003749
3750 if (error)
3751 return error;
3752
Al Viroacfa4382008-12-04 10:06:33 -05003753 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003754 return -EPERM;
3755
3756 mode &= (S_IRWXUGO|S_ISVTX);
3757 error = security_inode_mkdir(dir, dentry, mode);
3758 if (error)
3759 return error;
3760
Al Viro8de52772012-02-06 12:45:27 -05003761 if (max_links && dir->i_nlink >= max_links)
3762 return -EMLINK;
3763
Linus Torvalds1da177e2005-04-16 15:20:36 -07003764 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003765 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003766 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003767 return error;
3768}
Al Viro4d359502014-03-14 12:20:17 -04003769EXPORT_SYMBOL(vfs_mkdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003770
Al Viroa218d0f2011-11-21 14:59:34 -05003771SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003772{
Dave Hansen6902d922006-09-30 23:29:01 -07003773 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003774 struct path path;
3775 int error;
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003776 unsigned int lookup_flags = LOOKUP_DIRECTORY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003777
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003778retry:
3779 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07003780 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003781 return PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003782
Al Virodae6ad82011-06-26 11:50:15 -04003783 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003784 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003785 error = security_path_mkdir(&path, dentry, mode);
Al Viroa8104a92012-07-20 02:25:00 +04003786 if (!error)
3787 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
Al Viro921a1652012-07-20 01:15:31 +04003788 done_path_create(&path, dentry);
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003789 if (retry_estale(error, lookup_flags)) {
3790 lookup_flags |= LOOKUP_REVAL;
3791 goto retry;
3792 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003793 return error;
3794}
3795
Al Viroa218d0f2011-11-21 14:59:34 -05003796SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003797{
3798 return sys_mkdirat(AT_FDCWD, pathname, mode);
3799}
3800
Linus Torvalds1da177e2005-04-16 15:20:36 -07003801int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3802{
3803 int error = may_delete(dir, dentry, 1);
3804
3805 if (error)
3806 return error;
3807
Al Viroacfa4382008-12-04 10:06:33 -05003808 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003809 return -EPERM;
3810
Al Viro1d2ef592011-09-14 18:55:41 +01003811 dget(dentry);
Al Viro59551022016-01-22 15:40:57 -05003812 inode_lock(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003813
Sage Weil912dbc12011-05-24 13:06:11 -07003814 error = -EBUSY;
Eric W. Biederman7af13642013-10-04 19:15:13 -07003815 if (is_local_mountpoint(dentry))
Sage Weil912dbc12011-05-24 13:06:11 -07003816 goto out;
3817
3818 error = security_inode_rmdir(dir, dentry);
3819 if (error)
3820 goto out;
3821
Sage Weil3cebde22011-05-29 21:20:59 -07003822 shrink_dcache_parent(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003823 error = dir->i_op->rmdir(dir, dentry);
3824 if (error)
3825 goto out;
3826
3827 dentry->d_inode->i_flags |= S_DEAD;
3828 dont_mount(dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003829 detach_mounts(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003830
3831out:
Al Viro59551022016-01-22 15:40:57 -05003832 inode_unlock(dentry->d_inode);
Al Viro1d2ef592011-09-14 18:55:41 +01003833 dput(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003834 if (!error)
3835 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003836 return error;
3837}
Al Viro4d359502014-03-14 12:20:17 -04003838EXPORT_SYMBOL(vfs_rmdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003839
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003840static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003841{
3842 int error = 0;
Jeff Layton91a27b22012-10-10 15:25:28 -04003843 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003844 struct dentry *dentry;
Al Virof5beed72015-04-30 16:09:11 -04003845 struct path path;
3846 struct qstr last;
3847 int type;
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003848 unsigned int lookup_flags = 0;
3849retry:
Al Virof5beed72015-04-30 16:09:11 -04003850 name = user_path_parent(dfd, pathname,
3851 &path, &last, &type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04003852 if (IS_ERR(name))
3853 return PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003854
Al Virof5beed72015-04-30 16:09:11 -04003855 switch (type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003856 case LAST_DOTDOT:
3857 error = -ENOTEMPTY;
3858 goto exit1;
3859 case LAST_DOT:
3860 error = -EINVAL;
3861 goto exit1;
3862 case LAST_ROOT:
3863 error = -EBUSY;
3864 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003865 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003866
Al Virof5beed72015-04-30 16:09:11 -04003867 error = mnt_want_write(path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02003868 if (error)
3869 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003870
Al Viro59551022016-01-22 15:40:57 -05003871 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
Al Virof5beed72015-04-30 16:09:11 -04003872 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003873 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003874 if (IS_ERR(dentry))
3875 goto exit2;
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003876 if (!dentry->d_inode) {
3877 error = -ENOENT;
3878 goto exit3;
3879 }
Al Virof5beed72015-04-30 16:09:11 -04003880 error = security_path_rmdir(&path, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003881 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003882 goto exit3;
Al Virof5beed72015-04-30 16:09:11 -04003883 error = vfs_rmdir(path.dentry->d_inode, dentry);
Dave Hansen06227532008-02-15 14:37:34 -08003884exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07003885 dput(dentry);
3886exit2:
Al Viro59551022016-01-22 15:40:57 -05003887 inode_unlock(path.dentry->d_inode);
Al Virof5beed72015-04-30 16:09:11 -04003888 mnt_drop_write(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003889exit1:
Al Virof5beed72015-04-30 16:09:11 -04003890 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003891 putname(name);
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003892 if (retry_estale(error, lookup_flags)) {
3893 lookup_flags |= LOOKUP_REVAL;
3894 goto retry;
3895 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003896 return error;
3897}
3898
Heiko Carstens3cdad422009-01-14 14:14:22 +01003899SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003900{
3901 return do_rmdir(AT_FDCWD, pathname);
3902}
3903
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003904/**
3905 * vfs_unlink - unlink a filesystem object
3906 * @dir: parent directory
3907 * @dentry: victim
3908 * @delegated_inode: returns victim inode, if the inode is delegated.
3909 *
3910 * The caller must hold dir->i_mutex.
3911 *
3912 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3913 * return a reference to the inode in delegated_inode. The caller
3914 * should then break the delegation on that inode and retry. Because
3915 * breaking a delegation may take a long time, the caller should drop
3916 * dir->i_mutex before doing so.
3917 *
3918 * Alternatively, a caller may pass NULL for delegated_inode. This may
3919 * be appropriate for callers that expect the underlying filesystem not
3920 * to be NFS exported.
3921 */
3922int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003923{
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003924 struct inode *target = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003925 int error = may_delete(dir, dentry, 0);
3926
3927 if (error)
3928 return error;
3929
Al Viroacfa4382008-12-04 10:06:33 -05003930 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003931 return -EPERM;
3932
Al Viro59551022016-01-22 15:40:57 -05003933 inode_lock(target);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003934 if (is_local_mountpoint(dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003935 error = -EBUSY;
3936 else {
3937 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05003938 if (!error) {
J. Bruce Fields5a146962012-08-28 07:50:40 -07003939 error = try_break_deleg(target, delegated_inode);
3940 if (error)
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003941 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003942 error = dir->i_op->unlink(dir, dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003943 if (!error) {
Al Virod83c49f2010-04-30 17:17:09 -04003944 dont_mount(dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003945 detach_mounts(dentry);
3946 }
Al Virobec10522010-03-03 14:12:08 -05003947 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003948 }
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003949out:
Al Viro59551022016-01-22 15:40:57 -05003950 inode_unlock(target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003951
3952 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3953 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003954 fsnotify_link_count(target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003955 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003956 }
Robert Love0eeca282005-07-12 17:06:03 -04003957
Linus Torvalds1da177e2005-04-16 15:20:36 -07003958 return error;
3959}
Al Viro4d359502014-03-14 12:20:17 -04003960EXPORT_SYMBOL(vfs_unlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003961
3962/*
3963 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003964 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07003965 * writeout happening, and we don't want to prevent access to the directory
3966 * while waiting on the I/O.
3967 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003968static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003969{
Al Viro2ad94ae2008-07-21 09:32:51 -04003970 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04003971 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003972 struct dentry *dentry;
Al Virof5beed72015-04-30 16:09:11 -04003973 struct path path;
3974 struct qstr last;
3975 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003976 struct inode *inode = NULL;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003977 struct inode *delegated_inode = NULL;
Jeff Layton5d18f812012-12-20 16:38:04 -05003978 unsigned int lookup_flags = 0;
3979retry:
Al Virof5beed72015-04-30 16:09:11 -04003980 name = user_path_parent(dfd, pathname,
3981 &path, &last, &type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04003982 if (IS_ERR(name))
3983 return PTR_ERR(name);
Al Viro2ad94ae2008-07-21 09:32:51 -04003984
Linus Torvalds1da177e2005-04-16 15:20:36 -07003985 error = -EISDIR;
Al Virof5beed72015-04-30 16:09:11 -04003986 if (type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003987 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003988
Al Virof5beed72015-04-30 16:09:11 -04003989 error = mnt_want_write(path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02003990 if (error)
3991 goto exit1;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003992retry_deleg:
Al Viro59551022016-01-22 15:40:57 -05003993 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
Al Virof5beed72015-04-30 16:09:11 -04003994 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003995 error = PTR_ERR(dentry);
3996 if (!IS_ERR(dentry)) {
3997 /* Why not before? Because we want correct error value */
Al Virof5beed72015-04-30 16:09:11 -04003998 if (last.name[last.len])
Török Edwin50338b82011-06-16 00:06:14 +03003999 goto slashes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004000 inode = dentry->d_inode;
David Howellsb18825a2013-09-12 19:22:53 +01004001 if (d_is_negative(dentry))
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04004002 goto slashes;
4003 ihold(inode);
Al Virof5beed72015-04-30 16:09:11 -04004004 error = security_path_unlink(&path, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004005 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02004006 goto exit2;
Al Virof5beed72015-04-30 16:09:11 -04004007 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
Jan Karac30dabf2012-06-12 16:20:30 +02004008exit2:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004009 dput(dentry);
4010 }
Al Viro59551022016-01-22 15:40:57 -05004011 inode_unlock(path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004012 if (inode)
4013 iput(inode); /* truncate the inode here */
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04004014 inode = NULL;
4015 if (delegated_inode) {
J. Bruce Fields5a146962012-08-28 07:50:40 -07004016 error = break_deleg_wait(&delegated_inode);
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04004017 if (!error)
4018 goto retry_deleg;
4019 }
Al Virof5beed72015-04-30 16:09:11 -04004020 mnt_drop_write(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004021exit1:
Al Virof5beed72015-04-30 16:09:11 -04004022 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004023 putname(name);
Jeff Layton5d18f812012-12-20 16:38:04 -05004024 if (retry_estale(error, lookup_flags)) {
4025 lookup_flags |= LOOKUP_REVAL;
4026 inode = NULL;
4027 goto retry;
4028 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004029 return error;
4030
4031slashes:
David Howellsb18825a2013-09-12 19:22:53 +01004032 if (d_is_negative(dentry))
4033 error = -ENOENT;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02004034 else if (d_is_dir(dentry))
David Howellsb18825a2013-09-12 19:22:53 +01004035 error = -EISDIR;
4036 else
4037 error = -ENOTDIR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004038 goto exit2;
4039}
4040
Heiko Carstens2e4d0922009-01-14 14:14:31 +01004041SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004042{
4043 if ((flag & ~AT_REMOVEDIR) != 0)
4044 return -EINVAL;
4045
4046 if (flag & AT_REMOVEDIR)
4047 return do_rmdir(dfd, pathname);
4048
4049 return do_unlinkat(dfd, pathname);
4050}
4051
Heiko Carstens3480b252009-01-14 14:14:16 +01004052SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004053{
4054 return do_unlinkat(AT_FDCWD, pathname);
4055}
4056
Miklos Szeredidb2e7472008-06-24 16:50:16 +02004057int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004058{
Miklos Szeredia95164d2008-07-30 15:08:48 +02004059 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004060
4061 if (error)
4062 return error;
4063
Al Viroacfa4382008-12-04 10:06:33 -05004064 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004065 return -EPERM;
4066
4067 error = security_inode_symlink(dir, dentry, oldname);
4068 if (error)
4069 return error;
4070
Linus Torvalds1da177e2005-04-16 15:20:36 -07004071 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07004072 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00004073 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004074 return error;
4075}
Al Viro4d359502014-03-14 12:20:17 -04004076EXPORT_SYMBOL(vfs_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004077
Heiko Carstens2e4d0922009-01-14 14:14:31 +01004078SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4079 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004080{
Al Viro2ad94ae2008-07-21 09:32:51 -04004081 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04004082 struct filename *from;
Dave Hansen6902d922006-09-30 23:29:01 -07004083 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04004084 struct path path;
Jeff Laytonf46d3562012-12-11 12:10:08 -05004085 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004086
4087 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04004088 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004089 return PTR_ERR(from);
Jeff Laytonf46d3562012-12-11 12:10:08 -05004090retry:
4091 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07004092 error = PTR_ERR(dentry);
4093 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04004094 goto out_putname;
Dave Hansen6902d922006-09-30 23:29:01 -07004095
Jeff Layton91a27b22012-10-10 15:25:28 -04004096 error = security_path_symlink(&path, dentry, from->name);
Al Viroa8104a92012-07-20 02:25:00 +04004097 if (!error)
Jeff Layton91a27b22012-10-10 15:25:28 -04004098 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
Al Viro921a1652012-07-20 01:15:31 +04004099 done_path_create(&path, dentry);
Jeff Laytonf46d3562012-12-11 12:10:08 -05004100 if (retry_estale(error, lookup_flags)) {
4101 lookup_flags |= LOOKUP_REVAL;
4102 goto retry;
4103 }
Dave Hansen6902d922006-09-30 23:29:01 -07004104out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004105 putname(from);
4106 return error;
4107}
4108
Heiko Carstens3480b252009-01-14 14:14:16 +01004109SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004110{
4111 return sys_symlinkat(oldname, AT_FDCWD, newname);
4112}
4113
J. Bruce Fields146a8592011-09-20 17:14:31 -04004114/**
4115 * vfs_link - create a new link
4116 * @old_dentry: object to be linked
4117 * @dir: new parent
4118 * @new_dentry: where to create the new link
4119 * @delegated_inode: returns inode needing a delegation break
4120 *
4121 * The caller must hold dir->i_mutex
4122 *
4123 * If vfs_link discovers a delegation on the to-be-linked file in need
4124 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4125 * inode in delegated_inode. The caller should then break the delegation
4126 * and retry. Because breaking a delegation may take a long time, the
4127 * caller should drop the i_mutex before doing so.
4128 *
4129 * Alternatively, a caller may pass NULL for delegated_inode. This may
4130 * be appropriate for callers that expect the underlying filesystem not
4131 * to be NFS exported.
4132 */
4133int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004134{
4135 struct inode *inode = old_dentry->d_inode;
Al Viro8de52772012-02-06 12:45:27 -05004136 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004137 int error;
4138
4139 if (!inode)
4140 return -ENOENT;
4141
Miklos Szeredia95164d2008-07-30 15:08:48 +02004142 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004143 if (error)
4144 return error;
4145
4146 if (dir->i_sb != inode->i_sb)
4147 return -EXDEV;
4148
4149 /*
4150 * A link to an append-only or immutable file cannot be created.
4151 */
4152 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4153 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05004154 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004155 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02004156 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004157 return -EPERM;
4158
4159 error = security_inode_link(old_dentry, dir, new_dentry);
4160 if (error)
4161 return error;
4162
Al Viro59551022016-01-22 15:40:57 -05004163 inode_lock(inode);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05304164 /* Make sure we don't allow creating hardlink to an unlinked file */
Al Virof4e0c302013-06-11 08:34:36 +04004165 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05304166 error = -ENOENT;
Al Viro8de52772012-02-06 12:45:27 -05004167 else if (max_links && inode->i_nlink >= max_links)
4168 error = -EMLINK;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004169 else {
4170 error = try_break_deleg(inode, delegated_inode);
4171 if (!error)
4172 error = dir->i_op->link(old_dentry, dir, new_dentry);
4173 }
Al Virof4e0c302013-06-11 08:34:36 +04004174
4175 if (!error && (inode->i_state & I_LINKABLE)) {
4176 spin_lock(&inode->i_lock);
4177 inode->i_state &= ~I_LINKABLE;
4178 spin_unlock(&inode->i_lock);
4179 }
Al Viro59551022016-01-22 15:40:57 -05004180 inode_unlock(inode);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07004181 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02004182 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004183 return error;
4184}
Al Viro4d359502014-03-14 12:20:17 -04004185EXPORT_SYMBOL(vfs_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004186
4187/*
4188 * Hardlinks are often used in delicate situations. We avoid
4189 * security-related surprises by not following symlinks on the
4190 * newname. --KAB
4191 *
4192 * We don't follow them on the oldname either to be compatible
4193 * with linux 2.0, and to avoid hard-linking to directories
4194 * and other special files. --ADM
4195 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01004196SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4197 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004198{
4199 struct dentry *new_dentry;
Al Virodae6ad82011-06-26 11:50:15 -04004200 struct path old_path, new_path;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004201 struct inode *delegated_inode = NULL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304202 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004203 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004204
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304205 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004206 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304207 /*
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004208 * To use null names we require CAP_DAC_READ_SEARCH
4209 * This ensures that not everyone will be able to create
4210 * handlink using the passed filedescriptor.
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304211 */
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004212 if (flags & AT_EMPTY_PATH) {
4213 if (!capable(CAP_DAC_READ_SEARCH))
4214 return -ENOENT;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304215 how = LOOKUP_EMPTY;
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004216 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004217
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304218 if (flags & AT_SYMLINK_FOLLOW)
4219 how |= LOOKUP_FOLLOW;
Jeff Layton442e31c2012-12-20 16:15:38 -05004220retry:
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304221 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004222 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04004223 return error;
4224
Jeff Layton442e31c2012-12-20 16:15:38 -05004225 new_dentry = user_path_create(newdfd, newname, &new_path,
4226 (how & LOOKUP_REVAL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004227 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07004228 if (IS_ERR(new_dentry))
Al Virodae6ad82011-06-26 11:50:15 -04004229 goto out;
4230
4231 error = -EXDEV;
4232 if (old_path.mnt != new_path.mnt)
4233 goto out_dput;
Kees Cook800179c2012-07-25 17:29:07 -07004234 error = may_linkat(&old_path);
4235 if (unlikely(error))
4236 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04004237 error = security_path_link(old_path.dentry, &new_path, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004238 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04004239 goto out_dput;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004240 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
Dave Hansen75c3f292008-02-15 14:37:45 -08004241out_dput:
Al Viro921a1652012-07-20 01:15:31 +04004242 done_path_create(&new_path, new_dentry);
J. Bruce Fields146a8592011-09-20 17:14:31 -04004243 if (delegated_inode) {
4244 error = break_deleg_wait(&delegated_inode);
Oleg Drokind22e6332014-01-31 15:41:58 -05004245 if (!error) {
4246 path_put(&old_path);
J. Bruce Fields146a8592011-09-20 17:14:31 -04004247 goto retry;
Oleg Drokind22e6332014-01-31 15:41:58 -05004248 }
J. Bruce Fields146a8592011-09-20 17:14:31 -04004249 }
Jeff Layton442e31c2012-12-20 16:15:38 -05004250 if (retry_estale(error, how)) {
Oleg Drokind22e6332014-01-31 15:41:58 -05004251 path_put(&old_path);
Jeff Layton442e31c2012-12-20 16:15:38 -05004252 how |= LOOKUP_REVAL;
4253 goto retry;
4254 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004255out:
Al Viro2d8f3032008-07-22 09:59:21 -04004256 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004257
4258 return error;
4259}
4260
Heiko Carstens3480b252009-01-14 14:14:16 +01004261SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004262{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004263 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004264}
4265
Miklos Szeredibc270272014-04-01 17:08:42 +02004266/**
4267 * vfs_rename - rename a filesystem object
4268 * @old_dir: parent of source
4269 * @old_dentry: source
4270 * @new_dir: parent of destination
4271 * @new_dentry: destination
4272 * @delegated_inode: returns an inode needing a delegation break
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004273 * @flags: rename flags
Miklos Szeredibc270272014-04-01 17:08:42 +02004274 *
4275 * The caller must hold multiple mutexes--see lock_rename()).
4276 *
4277 * If vfs_rename discovers a delegation in need of breaking at either
4278 * the source or destination, it will return -EWOULDBLOCK and return a
4279 * reference to the inode in delegated_inode. The caller should then
4280 * break the delegation and retry. Because breaking a delegation may
4281 * take a long time, the caller should drop all locks before doing
4282 * so.
4283 *
4284 * Alternatively, a caller may pass NULL for delegated_inode. This may
4285 * be appropriate for callers that expect the underlying filesystem not
4286 * to be NFS exported.
4287 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07004288 * The worst of all namespace operations - renaming directory. "Perverted"
4289 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4290 * Problems:
J. Bruce Fieldsd03b29a2014-02-17 16:52:33 -05004291 * a) we can get into loop creation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004292 * b) race potential - two innocent renames can create a loop together.
4293 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004294 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07004295 * story.
J. Bruce Fields6cedba82012-03-05 11:40:41 -05004296 * c) we have to lock _four_ objects - parents and victim (if it exists),
4297 * and source (if it is not a directory).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004298 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07004299 * whether the target exists). Solution: try to be smart with locking
4300 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004301 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07004302 * move will be locked. Thus we can rank directories by the tree
4303 * (ancestors first) and rank all non-directories after them.
4304 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004305 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004306 * HOWEVER, it relies on the assumption that any object with ->lookup()
4307 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4308 * we'd better make sure that there's no link(2) for them.
Sage Weile4eaac02011-05-24 13:06:07 -07004309 * d) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004310 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07004311 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05004312 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07004313 * locking].
4314 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004315int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004316 struct inode *new_dir, struct dentry *new_dentry,
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004317 struct inode **delegated_inode, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004318{
4319 int error;
Miklos Szeredibc270272014-04-01 17:08:42 +02004320 bool is_dir = d_is_dir(old_dentry);
Eric Paris59b0df22010-02-08 12:53:52 -05004321 const unsigned char *old_name;
Miklos Szeredibc270272014-04-01 17:08:42 +02004322 struct inode *source = old_dentry->d_inode;
4323 struct inode *target = new_dentry->d_inode;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004324 bool new_is_dir = false;
4325 unsigned max_links = new_dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004326
Miklos Szeredi9409e222016-05-11 01:16:37 +02004327 /*
4328 * Check source == target.
4329 * On overlayfs need to look at underlying inodes.
4330 */
4331 if (vfs_select_inode(old_dentry, 0) == vfs_select_inode(new_dentry, 0))
Miklos Szeredibc270272014-04-01 17:08:42 +02004332 return 0;
4333
Linus Torvalds1da177e2005-04-16 15:20:36 -07004334 error = may_delete(old_dir, old_dentry, is_dir);
4335 if (error)
4336 return error;
4337
Miklos Szeredida1ce062014-04-01 17:08:43 +02004338 if (!target) {
Miklos Szeredia95164d2008-07-30 15:08:48 +02004339 error = may_create(new_dir, new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004340 } else {
4341 new_is_dir = d_is_dir(new_dentry);
4342
4343 if (!(flags & RENAME_EXCHANGE))
4344 error = may_delete(new_dir, new_dentry, is_dir);
4345 else
4346 error = may_delete(new_dir, new_dentry, new_is_dir);
4347 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004348 if (error)
4349 return error;
4350
Miklos Szeredi7177a9c2014-07-23 15:15:30 +02004351 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004352 return -EPERM;
4353
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004354 if (flags && !old_dir->i_op->rename2)
4355 return -EINVAL;
4356
Miklos Szeredibc270272014-04-01 17:08:42 +02004357 /*
4358 * If we are going to change the parent - check write permissions,
4359 * we'll need to flip '..'.
4360 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02004361 if (new_dir != old_dir) {
4362 if (is_dir) {
4363 error = inode_permission(source, MAY_WRITE);
4364 if (error)
4365 return error;
4366 }
4367 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4368 error = inode_permission(target, MAY_WRITE);
4369 if (error)
4370 return error;
4371 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004372 }
Robert Love0eeca282005-07-12 17:06:03 -04004373
Miklos Szeredi0b3974e2014-04-01 17:08:43 +02004374 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4375 flags);
Miklos Szeredibc270272014-04-01 17:08:42 +02004376 if (error)
4377 return error;
4378
4379 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4380 dget(new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004381 if (!is_dir || (flags & RENAME_EXCHANGE))
Miklos Szeredibc270272014-04-01 17:08:42 +02004382 lock_two_nondirectories(source, target);
4383 else if (target)
Al Viro59551022016-01-22 15:40:57 -05004384 inode_lock(target);
Miklos Szeredibc270272014-04-01 17:08:42 +02004385
4386 error = -EBUSY;
Eric W. Biederman7af13642013-10-04 19:15:13 -07004387 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
Miklos Szeredibc270272014-04-01 17:08:42 +02004388 goto out;
4389
Miklos Szeredida1ce062014-04-01 17:08:43 +02004390 if (max_links && new_dir != old_dir) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004391 error = -EMLINK;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004392 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
Miklos Szeredibc270272014-04-01 17:08:42 +02004393 goto out;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004394 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4395 old_dir->i_nlink >= max_links)
4396 goto out;
4397 }
4398 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4399 shrink_dcache_parent(new_dentry);
4400 if (!is_dir) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004401 error = try_break_deleg(source, delegated_inode);
4402 if (error)
4403 goto out;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004404 }
4405 if (target && !new_is_dir) {
4406 error = try_break_deleg(target, delegated_inode);
4407 if (error)
4408 goto out;
Miklos Szeredibc270272014-04-01 17:08:42 +02004409 }
Miklos Szeredi7177a9c2014-07-23 15:15:30 +02004410 if (!old_dir->i_op->rename2) {
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004411 error = old_dir->i_op->rename(old_dir, old_dentry,
4412 new_dir, new_dentry);
4413 } else {
Miklos Szeredi7177a9c2014-07-23 15:15:30 +02004414 WARN_ON(old_dir->i_op->rename != NULL);
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004415 error = old_dir->i_op->rename2(old_dir, old_dentry,
4416 new_dir, new_dentry, flags);
4417 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004418 if (error)
4419 goto out;
4420
Miklos Szeredida1ce062014-04-01 17:08:43 +02004421 if (!(flags & RENAME_EXCHANGE) && target) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004422 if (is_dir)
4423 target->i_flags |= S_DEAD;
4424 dont_mount(new_dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07004425 detach_mounts(new_dentry);
Miklos Szeredibc270272014-04-01 17:08:42 +02004426 }
Miklos Szeredida1ce062014-04-01 17:08:43 +02004427 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4428 if (!(flags & RENAME_EXCHANGE))
4429 d_move(old_dentry, new_dentry);
4430 else
4431 d_exchange(old_dentry, new_dentry);
4432 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004433out:
Miklos Szeredida1ce062014-04-01 17:08:43 +02004434 if (!is_dir || (flags & RENAME_EXCHANGE))
Miklos Szeredibc270272014-04-01 17:08:42 +02004435 unlock_two_nondirectories(source, target);
4436 else if (target)
Al Viro59551022016-01-22 15:40:57 -05004437 inode_unlock(target);
Miklos Szeredibc270272014-04-01 17:08:42 +02004438 dput(new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004439 if (!error) {
Al Viro123df292009-12-25 04:57:57 -05004440 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Miklos Szeredida1ce062014-04-01 17:08:43 +02004441 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4442 if (flags & RENAME_EXCHANGE) {
4443 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4444 new_is_dir, NULL, new_dentry);
4445 }
4446 }
Robert Love0eeca282005-07-12 17:06:03 -04004447 fsnotify_oldname_free(old_name);
4448
Linus Torvalds1da177e2005-04-16 15:20:36 -07004449 return error;
4450}
Al Viro4d359502014-03-14 12:20:17 -04004451EXPORT_SYMBOL(vfs_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004452
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004453SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4454 int, newdfd, const char __user *, newname, unsigned int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004455{
Al Viro2ad94ae2008-07-21 09:32:51 -04004456 struct dentry *old_dentry, *new_dentry;
4457 struct dentry *trap;
Al Virof5beed72015-04-30 16:09:11 -04004458 struct path old_path, new_path;
4459 struct qstr old_last, new_last;
4460 int old_type, new_type;
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004461 struct inode *delegated_inode = NULL;
Jeff Layton91a27b22012-10-10 15:25:28 -04004462 struct filename *from;
4463 struct filename *to;
Al Virof5beed72015-04-30 16:09:11 -04004464 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
Jeff Laytonc6a94282012-12-11 12:10:10 -05004465 bool should_retry = false;
Al Viro2ad94ae2008-07-21 09:32:51 -04004466 int error;
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004467
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004468 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredida1ce062014-04-01 17:08:43 +02004469 return -EINVAL;
4470
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004471 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4472 (flags & RENAME_EXCHANGE))
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004473 return -EINVAL;
4474
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004475 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4476 return -EPERM;
4477
Al Virof5beed72015-04-30 16:09:11 -04004478 if (flags & RENAME_EXCHANGE)
4479 target_flags = 0;
4480
Jeff Laytonc6a94282012-12-11 12:10:10 -05004481retry:
Al Virof5beed72015-04-30 16:09:11 -04004482 from = user_path_parent(olddfd, oldname,
4483 &old_path, &old_last, &old_type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04004484 if (IS_ERR(from)) {
4485 error = PTR_ERR(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004486 goto exit;
Jeff Layton91a27b22012-10-10 15:25:28 -04004487 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004488
Al Virof5beed72015-04-30 16:09:11 -04004489 to = user_path_parent(newdfd, newname,
4490 &new_path, &new_last, &new_type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04004491 if (IS_ERR(to)) {
4492 error = PTR_ERR(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004493 goto exit1;
Jeff Layton91a27b22012-10-10 15:25:28 -04004494 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004495
4496 error = -EXDEV;
Al Virof5beed72015-04-30 16:09:11 -04004497 if (old_path.mnt != new_path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004498 goto exit2;
4499
Linus Torvalds1da177e2005-04-16 15:20:36 -07004500 error = -EBUSY;
Al Virof5beed72015-04-30 16:09:11 -04004501 if (old_type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004502 goto exit2;
4503
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004504 if (flags & RENAME_NOREPLACE)
4505 error = -EEXIST;
Al Virof5beed72015-04-30 16:09:11 -04004506 if (new_type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004507 goto exit2;
4508
Al Virof5beed72015-04-30 16:09:11 -04004509 error = mnt_want_write(old_path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02004510 if (error)
4511 goto exit2;
4512
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004513retry_deleg:
Al Virof5beed72015-04-30 16:09:11 -04004514 trap = lock_rename(new_path.dentry, old_path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004515
Al Virof5beed72015-04-30 16:09:11 -04004516 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004517 error = PTR_ERR(old_dentry);
4518 if (IS_ERR(old_dentry))
4519 goto exit3;
4520 /* source must exist */
4521 error = -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01004522 if (d_is_negative(old_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004523 goto exit4;
Al Virof5beed72015-04-30 16:09:11 -04004524 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004525 error = PTR_ERR(new_dentry);
4526 if (IS_ERR(new_dentry))
4527 goto exit4;
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004528 error = -EEXIST;
4529 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4530 goto exit5;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004531 if (flags & RENAME_EXCHANGE) {
4532 error = -ENOENT;
4533 if (d_is_negative(new_dentry))
4534 goto exit5;
4535
4536 if (!d_is_dir(new_dentry)) {
4537 error = -ENOTDIR;
Al Virof5beed72015-04-30 16:09:11 -04004538 if (new_last.name[new_last.len])
Miklos Szeredida1ce062014-04-01 17:08:43 +02004539 goto exit5;
4540 }
4541 }
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004542 /* unless the source is a directory trailing slashes give -ENOTDIR */
4543 if (!d_is_dir(old_dentry)) {
4544 error = -ENOTDIR;
Al Virof5beed72015-04-30 16:09:11 -04004545 if (old_last.name[old_last.len])
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004546 goto exit5;
Al Virof5beed72015-04-30 16:09:11 -04004547 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004548 goto exit5;
4549 }
4550 /* source should not be ancestor of target */
4551 error = -EINVAL;
4552 if (old_dentry == trap)
4553 goto exit5;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004554 /* target should not be an ancestor of source */
Miklos Szeredida1ce062014-04-01 17:08:43 +02004555 if (!(flags & RENAME_EXCHANGE))
4556 error = -ENOTEMPTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004557 if (new_dentry == trap)
4558 goto exit5;
4559
Al Virof5beed72015-04-30 16:09:11 -04004560 error = security_path_rename(&old_path, old_dentry,
4561 &new_path, new_dentry, flags);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004562 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02004563 goto exit5;
Al Virof5beed72015-04-30 16:09:11 -04004564 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4565 new_path.dentry->d_inode, new_dentry,
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004566 &delegated_inode, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004567exit5:
4568 dput(new_dentry);
4569exit4:
4570 dput(old_dentry);
4571exit3:
Al Virof5beed72015-04-30 16:09:11 -04004572 unlock_rename(new_path.dentry, old_path.dentry);
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004573 if (delegated_inode) {
4574 error = break_deleg_wait(&delegated_inode);
4575 if (!error)
4576 goto retry_deleg;
4577 }
Al Virof5beed72015-04-30 16:09:11 -04004578 mnt_drop_write(old_path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004579exit2:
Jeff Laytonc6a94282012-12-11 12:10:10 -05004580 if (retry_estale(error, lookup_flags))
4581 should_retry = true;
Al Virof5beed72015-04-30 16:09:11 -04004582 path_put(&new_path);
Al Viro2ad94ae2008-07-21 09:32:51 -04004583 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004584exit1:
Al Virof5beed72015-04-30 16:09:11 -04004585 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004586 putname(from);
Jeff Laytonc6a94282012-12-11 12:10:10 -05004587 if (should_retry) {
4588 should_retry = false;
4589 lookup_flags |= LOOKUP_REVAL;
4590 goto retry;
4591 }
Al Viro2ad94ae2008-07-21 09:32:51 -04004592exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004593 return error;
4594}
4595
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004596SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4597 int, newdfd, const char __user *, newname)
4598{
4599 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4600}
4601
Heiko Carstensa26eab22009-01-14 14:14:17 +01004602SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004603{
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004604 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004605}
4606
Miklos Szeredi787fb6b2014-10-24 00:14:36 +02004607int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4608{
4609 int error = may_create(dir, dentry);
4610 if (error)
4611 return error;
4612
4613 if (!dir->i_op->mknod)
4614 return -EPERM;
4615
4616 return dir->i_op->mknod(dir, dentry,
4617 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4618}
4619EXPORT_SYMBOL(vfs_whiteout);
4620
Al Viro5d826c82014-03-14 13:42:45 -04004621int readlink_copy(char __user *buffer, int buflen, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004622{
Al Viro5d826c82014-03-14 13:42:45 -04004623 int len = PTR_ERR(link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004624 if (IS_ERR(link))
4625 goto out;
4626
4627 len = strlen(link);
4628 if (len > (unsigned) buflen)
4629 len = buflen;
4630 if (copy_to_user(buffer, link, len))
4631 len = -EFAULT;
4632out:
4633 return len;
4634}
4635
4636/*
4637 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
Al Viro6b255392015-11-17 10:20:54 -05004638 * have ->get_link() not calling nd_jump_link(). Using (or not using) it
4639 * for any given inode is up to filesystem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004640 */
4641int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4642{
Al Virofceef392015-12-29 15:58:39 -05004643 DEFINE_DELAYED_CALL(done);
Al Viro5f2c4172015-05-07 11:14:26 -04004644 struct inode *inode = d_inode(dentry);
4645 const char *link = inode->i_link;
Marcin Slusarz694a1762008-06-09 16:40:37 -07004646 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004647
Al Virod4dee482015-04-30 20:08:02 -04004648 if (!link) {
Al Virofceef392015-12-29 15:58:39 -05004649 link = inode->i_op->get_link(dentry, inode, &done);
Al Virod4dee482015-04-30 20:08:02 -04004650 if (IS_ERR(link))
4651 return PTR_ERR(link);
4652 }
Al Viro680baac2015-05-02 13:32:22 -04004653 res = readlink_copy(buffer, buflen, link);
Al Virofceef392015-12-29 15:58:39 -05004654 do_delayed_call(&done);
Marcin Slusarz694a1762008-06-09 16:40:37 -07004655 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004656}
Al Viro4d359502014-03-14 12:20:17 -04004657EXPORT_SYMBOL(generic_readlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004658
Linus Torvalds1da177e2005-04-16 15:20:36 -07004659/* get the link contents into pagecache */
Al Viro6b255392015-11-17 10:20:54 -05004660const char *page_get_link(struct dentry *dentry, struct inode *inode,
Al Virofceef392015-12-29 15:58:39 -05004661 struct delayed_call *callback)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004662{
Duane Griffinebd09ab2008-12-19 20:47:12 +00004663 char *kaddr;
4664 struct page *page;
Al Viro6b255392015-11-17 10:20:54 -05004665 struct address_space *mapping = inode->i_mapping;
4666
Al Virod3883d42015-11-17 10:41:04 -05004667 if (!dentry) {
4668 page = find_get_page(mapping, 0);
4669 if (!page)
4670 return ERR_PTR(-ECHILD);
4671 if (!PageUptodate(page)) {
4672 put_page(page);
4673 return ERR_PTR(-ECHILD);
4674 }
4675 } else {
4676 page = read_mapping_page(mapping, 0, NULL);
4677 if (IS_ERR(page))
4678 return (char*)page;
4679 }
Al Virofceef392015-12-29 15:58:39 -05004680 set_delayed_call(callback, page_put_link, page);
Al Viro21fc61c2015-11-17 01:07:57 -05004681 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4682 kaddr = page_address(page);
Al Viro6b255392015-11-17 10:20:54 -05004683 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
Duane Griffinebd09ab2008-12-19 20:47:12 +00004684 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004685}
4686
Al Viro6b255392015-11-17 10:20:54 -05004687EXPORT_SYMBOL(page_get_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004688
Al Virofceef392015-12-29 15:58:39 -05004689void page_put_link(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004690{
Al Virofceef392015-12-29 15:58:39 -05004691 put_page(arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004692}
Al Viro4d359502014-03-14 12:20:17 -04004693EXPORT_SYMBOL(page_put_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004694
Al Viroaa80dea2015-11-16 18:26:34 -05004695int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4696{
Al Virofceef392015-12-29 15:58:39 -05004697 DEFINE_DELAYED_CALL(done);
Al Viro6b255392015-11-17 10:20:54 -05004698 int res = readlink_copy(buffer, buflen,
4699 page_get_link(dentry, d_inode(dentry),
Al Virofceef392015-12-29 15:58:39 -05004700 &done));
4701 do_delayed_call(&done);
Al Viroaa80dea2015-11-16 18:26:34 -05004702 return res;
4703}
4704EXPORT_SYMBOL(page_readlink);
4705
Nick Piggin54566b22009-01-04 12:00:53 -08004706/*
4707 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4708 */
4709int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004710{
4711 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004712 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07004713 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08004714 int err;
Nick Piggin54566b22009-01-04 12:00:53 -08004715 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4716 if (nofs)
4717 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004718
NeilBrown7e53cac2006-03-25 03:07:57 -08004719retry:
Nick Pigginafddba42007-10-16 01:25:01 -07004720 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08004721 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004722 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07004723 goto fail;
4724
Al Viro21fc61c2015-11-17 01:07:57 -05004725 memcpy(page_address(page), symname, len-1);
Nick Pigginafddba42007-10-16 01:25:01 -07004726
4727 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4728 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004729 if (err < 0)
4730 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07004731 if (err < len-1)
4732 goto retry;
4733
Linus Torvalds1da177e2005-04-16 15:20:36 -07004734 mark_inode_dirty(inode);
4735 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004736fail:
4737 return err;
4738}
Al Viro4d359502014-03-14 12:20:17 -04004739EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004740
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004741int page_symlink(struct inode *inode, const char *symname, int len)
4742{
4743 return __page_symlink(inode, symname, len,
Michal Hockoc62d2552015-11-06 16:28:49 -08004744 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004745}
Al Viro4d359502014-03-14 12:20:17 -04004746EXPORT_SYMBOL(page_symlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004747
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08004748const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004749 .readlink = generic_readlink,
Al Viro6b255392015-11-17 10:20:54 -05004750 .get_link = page_get_link,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004751};
Linus Torvalds1da177e2005-04-16 15:20:36 -07004752EXPORT_SYMBOL(page_symlink_inode_operations);