blob: 85ac38b99065c6c88537cdbccd49908c0e38430e [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>
Eric W. Biedermanaeaa4a72016-07-23 11:20:44 -050039#include <linux/init_task.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include <asm/uaccess.h>
41
Eric Parise81e3f42009-12-04 15:47:36 -050042#include "internal.h"
Al Viroc7105362011-11-24 18:22:03 -050043#include "mount.h"
Eric Parise81e3f42009-12-04 15:47:36 -050044
Linus Torvalds1da177e2005-04-16 15:20:36 -070045/* [Feb-1997 T. Schoebel-Theuer]
46 * Fundamental changes in the pathname lookup mechanisms (namei)
47 * were necessary because of omirr. The reason is that omirr needs
48 * to know the _real_ pathname, not the user-supplied one, in case
49 * of symlinks (and also when transname replacements occur).
50 *
51 * The new code replaces the old recursive symlink resolution with
52 * an iterative one (in case of non-nested symlink chains). It does
53 * this with calls to <fs>_follow_link().
54 * As a side effect, dir_namei(), _namei() and follow_link() are now
55 * replaced with a single function lookup_dentry() that can handle all
56 * the special cases of the former code.
57 *
58 * With the new dcache, the pathname is stored at each inode, at least as
59 * long as the refcount of the inode is positive. As a side effect, the
60 * size of the dcache depends on the inode cache and thus is dynamic.
61 *
62 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
63 * resolution to correspond with current state of the code.
64 *
65 * Note that the symlink resolution is not *completely* iterative.
66 * There is still a significant amount of tail- and mid- recursion in
67 * the algorithm. Also, note that <fs>_readlink() is not used in
68 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
69 * may return different results than <fs>_follow_link(). Many virtual
70 * filesystems (including /proc) exhibit this behavior.
71 */
72
73/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
74 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
75 * and the name already exists in form of a symlink, try to create the new
76 * name indicated by the symlink. The old code always complained that the
77 * name already exists, due to not following the symlink even if its target
78 * is nonexistent. The new semantics affects also mknod() and link() when
Lucas De Marchi25985ed2011-03-30 22:57:33 -030079 * the name is a symlink pointing to a non-existent name.
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 *
81 * I don't know which semantics is the right one, since I have no access
82 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
83 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
84 * "old" one. Personally, I think the new semantics is much more logical.
85 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
86 * file does succeed in both HP-UX and SunOs, but not in Solaris
87 * and in the old Linux semantics.
88 */
89
90/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
91 * semantics. See the comments in "open_namei" and "do_link" below.
92 *
93 * [10-Sep-98 Alan Modra] Another symlink change.
94 */
95
96/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
97 * inside the path - always follow.
98 * in the last component in creation/removal/renaming - never follow.
99 * if LOOKUP_FOLLOW passed - follow.
100 * if the pathname has trailing slashes - follow.
101 * otherwise - don't follow.
102 * (applied in that order).
103 *
104 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
105 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
106 * During the 2.4 we need to fix the userland stuff depending on it -
107 * hopefully we will be able to get rid of that wart in 2.5. So far only
108 * XEmacs seems to be relying on it...
109 */
110/*
111 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800112 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 * any extra contention...
114 */
115
116/* In order to reduce some races, while at the same time doing additional
117 * checking and hopefully speeding things up, we copy filenames to the
118 * kernel data space before using them..
119 *
120 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
121 * PATH_MAX includes the nul terminator --RR.
122 */
Jeff Layton91a27b22012-10-10 15:25:28 -0400123
Al Virofd2f7cb2015-02-22 20:07:13 -0500124#define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
Jeff Layton7950e382012-10-10 16:43:13 -0400125
David Drysdale51f39a12014-12-12 16:57:29 -0800126struct filename *
Jeff Layton91a27b22012-10-10 15:25:28 -0400127getname_flags(const char __user *filename, int flags, int *empty)
128{
Al Viro94b5d262015-02-22 19:38:03 -0500129 struct filename *result;
Jeff Layton7950e382012-10-10 16:43:13 -0400130 char *kname;
Al Viro94b5d262015-02-22 19:38:03 -0500131 int len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132
Jeff Layton7ac86262012-10-10 15:25:28 -0400133 result = audit_reusename(filename);
134 if (result)
135 return result;
136
Jeff Layton7950e382012-10-10 16:43:13 -0400137 result = __getname();
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700138 if (unlikely(!result))
Eric Paris4043cde2012-01-03 14:23:08 -0500139 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140
Jeff Layton7950e382012-10-10 16:43:13 -0400141 /*
142 * First, try to embed the struct filename inside the names_cache
143 * allocation
144 */
Al Virofd2f7cb2015-02-22 20:07:13 -0500145 kname = (char *)result->iname;
Jeff Layton91a27b22012-10-10 15:25:28 -0400146 result->name = kname;
Jeff Layton7950e382012-10-10 16:43:13 -0400147
Al Viro94b5d262015-02-22 19:38:03 -0500148 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
Jeff Layton91a27b22012-10-10 15:25:28 -0400149 if (unlikely(len < 0)) {
Al Viro94b5d262015-02-22 19:38:03 -0500150 __putname(result);
151 return ERR_PTR(len);
Jeff Layton91a27b22012-10-10 15:25:28 -0400152 }
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700153
Jeff Layton7950e382012-10-10 16:43:13 -0400154 /*
155 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
156 * separate struct filename so we can dedicate the entire
157 * names_cache allocation for the pathname, and re-do the copy from
158 * userland.
159 */
Al Viro94b5d262015-02-22 19:38:03 -0500160 if (unlikely(len == EMBEDDED_NAME_MAX)) {
Al Virofd2f7cb2015-02-22 20:07:13 -0500161 const size_t size = offsetof(struct filename, iname[1]);
Jeff Layton7950e382012-10-10 16:43:13 -0400162 kname = (char *)result;
163
Al Virofd2f7cb2015-02-22 20:07:13 -0500164 /*
165 * size is chosen that way we to guarantee that
166 * result->iname[0] is within the same object and that
167 * kname can't be equal to result->iname, no matter what.
168 */
169 result = kzalloc(size, GFP_KERNEL);
Al Viro94b5d262015-02-22 19:38:03 -0500170 if (unlikely(!result)) {
171 __putname(kname);
172 return ERR_PTR(-ENOMEM);
Jeff Layton7950e382012-10-10 16:43:13 -0400173 }
174 result->name = kname;
Al Viro94b5d262015-02-22 19:38:03 -0500175 len = strncpy_from_user(kname, filename, PATH_MAX);
176 if (unlikely(len < 0)) {
177 __putname(kname);
178 kfree(result);
179 return ERR_PTR(len);
180 }
181 if (unlikely(len == PATH_MAX)) {
182 __putname(kname);
183 kfree(result);
184 return ERR_PTR(-ENAMETOOLONG);
185 }
Jeff Layton7950e382012-10-10 16:43:13 -0400186 }
187
Al Viro94b5d262015-02-22 19:38:03 -0500188 result->refcnt = 1;
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700189 /* The empty path is special. */
190 if (unlikely(!len)) {
191 if (empty)
Eric Paris4043cde2012-01-03 14:23:08 -0500192 *empty = 1;
Al Viro94b5d262015-02-22 19:38:03 -0500193 if (!(flags & LOOKUP_EMPTY)) {
194 putname(result);
195 return ERR_PTR(-ENOENT);
196 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 }
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700198
Jeff Layton7950e382012-10-10 16:43:13 -0400199 result->uptr = filename;
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800200 result->aname = NULL;
Jeff Layton7950e382012-10-10 16:43:13 -0400201 audit_getname(result);
202 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203}
204
Jeff Layton91a27b22012-10-10 15:25:28 -0400205struct filename *
206getname(const char __user * filename)
Al Virof52e0c12011-03-14 18:56:51 -0400207{
Linus Torvaldsf7493e52012-03-22 16:10:40 -0700208 return getname_flags(filename, 0, NULL);
Al Virof52e0c12011-03-14 18:56:51 -0400209}
210
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800211struct filename *
212getname_kernel(const char * filename)
213{
214 struct filename *result;
Paul Moore08518542015-01-21 23:59:56 -0500215 int len = strlen(filename) + 1;
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800216
217 result = __getname();
218 if (unlikely(!result))
219 return ERR_PTR(-ENOMEM);
220
Paul Moore08518542015-01-21 23:59:56 -0500221 if (len <= EMBEDDED_NAME_MAX) {
Al Virofd2f7cb2015-02-22 20:07:13 -0500222 result->name = (char *)result->iname;
Paul Moore08518542015-01-21 23:59:56 -0500223 } else if (len <= PATH_MAX) {
Al Viro960534a2018-04-08 11:57:10 -0400224 const size_t size = offsetof(struct filename, iname[1]);
Paul Moore08518542015-01-21 23:59:56 -0500225 struct filename *tmp;
226
Al Viro960534a2018-04-08 11:57:10 -0400227 tmp = kmalloc(size, GFP_KERNEL);
Paul Moore08518542015-01-21 23:59:56 -0500228 if (unlikely(!tmp)) {
229 __putname(result);
230 return ERR_PTR(-ENOMEM);
231 }
232 tmp->name = (char *)result;
Paul Moore08518542015-01-21 23:59:56 -0500233 result = tmp;
234 } else {
235 __putname(result);
236 return ERR_PTR(-ENAMETOOLONG);
237 }
238 memcpy((char *)result->name, filename, len);
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800239 result->uptr = NULL;
240 result->aname = NULL;
Paul Moore55422d02015-01-22 00:00:23 -0500241 result->refcnt = 1;
Paul Moorefd3522f2015-01-22 00:00:10 -0500242 audit_getname(result);
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800243
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800244 return result;
245}
246
Jeff Layton91a27b22012-10-10 15:25:28 -0400247void putname(struct filename *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248{
Paul Moore55422d02015-01-22 00:00:23 -0500249 BUG_ON(name->refcnt <= 0);
250
251 if (--name->refcnt > 0)
252 return;
253
Al Virofd2f7cb2015-02-22 20:07:13 -0500254 if (name->name != name->iname) {
Paul Moore55422d02015-01-22 00:00:23 -0500255 __putname(name->name);
256 kfree(name);
257 } else
258 __putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260
Linus Torvaldse77819e2011-07-22 19:30:19 -0700261static int check_acl(struct inode *inode, int mask)
262{
Linus Torvalds84635d62011-07-25 22:47:03 -0700263#ifdef CONFIG_FS_POSIX_ACL
Linus Torvaldse77819e2011-07-22 19:30:19 -0700264 struct posix_acl *acl;
265
Linus Torvaldse77819e2011-07-22 19:30:19 -0700266 if (mask & MAY_NOT_BLOCK) {
Al Viro35678662011-08-02 21:32:13 -0400267 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
268 if (!acl)
Linus Torvaldse77819e2011-07-22 19:30:19 -0700269 return -EAGAIN;
Al Viro35678662011-08-02 21:32:13 -0400270 /* no ->get_acl() calls in RCU mode... */
Andreas Gruenbacherb8a7a3a2016-03-24 14:38:37 +0100271 if (is_uncached_acl(acl))
Al Viro35678662011-08-02 21:32:13 -0400272 return -ECHILD;
Ari Savolainen206b1d02011-08-06 19:43:07 +0300273 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
Linus Torvaldse77819e2011-07-22 19:30:19 -0700274 }
275
Christoph Hellwig2982baa2013-12-20 05:16:38 -0800276 acl = get_acl(inode, ACL_TYPE_ACCESS);
277 if (IS_ERR(acl))
278 return PTR_ERR(acl);
Linus Torvaldse77819e2011-07-22 19:30:19 -0700279 if (acl) {
280 int error = posix_acl_permission(inode, acl, mask);
281 posix_acl_release(acl);
282 return error;
283 }
Linus Torvalds84635d62011-07-25 22:47:03 -0700284#endif
Linus Torvaldse77819e2011-07-22 19:30:19 -0700285
286 return -EAGAIN;
287}
288
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700289/*
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530290 * This does the basic permission checking
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700291 */
Al Viro7e401452011-06-20 19:12:17 -0400292static int acl_permission_check(struct inode *inode, int mask)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700293{
Linus Torvalds26cf46b2011-05-13 11:51:01 -0700294 unsigned int mode = inode->i_mode;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700295
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -0800296 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700297 mode >>= 6;
298 else {
Linus Torvaldse77819e2011-07-22 19:30:19 -0700299 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
Al Viro7e401452011-06-20 19:12:17 -0400300 int error = check_acl(inode, mask);
Nick Pigginb74c79e2011-01-07 17:49:58 +1100301 if (error != -EAGAIN)
302 return error;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700303 }
304
305 if (in_group_p(inode->i_gid))
306 mode >>= 3;
307 }
308
309 /*
310 * If the DACs are ok we don't need any capability check.
311 */
Al Viro9c2c7032011-06-20 19:06:22 -0400312 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700313 return 0;
314 return -EACCES;
315}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316
317/**
Nick Pigginb74c79e2011-01-07 17:49:58 +1100318 * generic_permission - check for access rights on a Posix-like filesystem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 * @inode: inode to check access rights for
Andreas Gruenbacher8fd90c82011-10-23 23:13:30 +0530320 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 *
322 * Used to check for read/write/execute permissions on a file.
323 * We use "fsuid" for this, letting us set arbitrary permissions
324 * for filesystem access without changing the "normal" uids which
Nick Pigginb74c79e2011-01-07 17:49:58 +1100325 * are used for other things.
326 *
327 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
328 * request cannot be satisfied (eg. requires blocking or too much complexity).
329 * It would then be called again in ref-walk mode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 */
Al Viro2830ba72011-06-20 19:16:29 -0400331int generic_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700333 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334
335 /*
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530336 * Do the basic permission checks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 */
Al Viro7e401452011-06-20 19:12:17 -0400338 ret = acl_permission_check(inode, mask);
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700339 if (ret != -EACCES)
340 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341
Al Virod594e7e2011-06-20 19:55:42 -0400342 if (S_ISDIR(inode->i_mode)) {
343 /* DACs are overridable for directories */
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700344 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
Al Virod594e7e2011-06-20 19:55:42 -0400345 return 0;
346 if (!(mask & MAY_WRITE))
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700347 if (capable_wrt_inode_uidgid(inode,
348 CAP_DAC_READ_SEARCH))
Al Virod594e7e2011-06-20 19:55:42 -0400349 return 0;
350 return -EACCES;
351 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 /*
353 * Read/write DACs are always overridable.
Al Virod594e7e2011-06-20 19:55:42 -0400354 * Executable DACs are overridable when there is
355 * at least one exec bit set.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 */
Al Virod594e7e2011-06-20 19:55:42 -0400357 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700358 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 return 0;
360
361 /*
362 * Searching includes executable on directories, else just read.
363 */
Serge E. Hallyn7ea66002009-12-29 14:50:19 -0600364 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
Al Virod594e7e2011-06-20 19:55:42 -0400365 if (mask == MAY_READ)
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700366 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 return 0;
368
369 return -EACCES;
370}
Al Viro4d359502014-03-14 12:20:17 -0400371EXPORT_SYMBOL(generic_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700373/*
374 * We _really_ want to just do "generic_permission()" without
375 * even looking at the inode->i_op values. So we keep a cache
376 * flag in inode->i_opflags, that says "this has not special
377 * permission function, use the fast case".
378 */
379static inline int do_inode_permission(struct inode *inode, int mask)
380{
381 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
382 if (likely(inode->i_op->permission))
383 return inode->i_op->permission(inode, mask);
384
385 /* This gets set once for the inode lifetime */
386 spin_lock(&inode->i_lock);
387 inode->i_opflags |= IOP_FASTPERM;
388 spin_unlock(&inode->i_lock);
389 }
390 return generic_permission(inode, mask);
391}
392
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200393/**
David Howells0bdaea92012-06-25 12:55:46 +0100394 * __inode_permission - Check for access rights to a given inode
395 * @inode: Inode to check permission on
396 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200397 *
David Howells0bdaea92012-06-25 12:55:46 +0100398 * Check for read/write/execute permissions on an inode.
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530399 *
400 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
David Howells0bdaea92012-06-25 12:55:46 +0100401 *
402 * This does not check for a read-only file system. You probably want
403 * inode_permission().
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200404 */
David Howells0bdaea92012-06-25 12:55:46 +0100405int __inode_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406{
Al Viroe6305c42008-07-15 21:03:57 -0400407 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700409 if (unlikely(mask & MAY_WRITE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 /*
411 * Nobody gets write access to an immutable file.
412 */
413 if (IS_IMMUTABLE(inode))
Eryu Guan337684a2016-08-02 19:58:28 +0800414 return -EPERM;
Eric W. Biederman0bd23d092016-06-29 14:54:46 -0500415
416 /*
417 * Updating mtime will likely cause i_uid and i_gid to be
418 * written back improperly if their true value is unknown
419 * to the vfs.
420 */
421 if (HAS_UNMAPPED_ID(inode))
422 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 }
424
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700425 retval = do_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 if (retval)
427 return retval;
428
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -0700429 retval = devcgroup_inode_permission(inode, mask);
430 if (retval)
431 return retval;
432
Eric Parisd09ca732010-07-23 11:43:57 -0400433 return security_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434}
Miklos Szeredibd5d0852014-10-24 00:14:35 +0200435EXPORT_SYMBOL(__inode_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436
Al Virof4d6ff82011-06-19 13:14:21 -0400437/**
David Howells0bdaea92012-06-25 12:55:46 +0100438 * sb_permission - Check superblock-level permissions
439 * @sb: Superblock of inode to check permission on
Randy Dunlap55852632012-08-18 17:39:25 -0700440 * @inode: Inode to check permission on
David Howells0bdaea92012-06-25 12:55:46 +0100441 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
442 *
443 * Separate out file-system wide checks from inode-specific permission checks.
444 */
445static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
446{
447 if (unlikely(mask & MAY_WRITE)) {
448 umode_t mode = inode->i_mode;
449
450 /* Nobody gets write access to a read-only fs. */
451 if ((sb->s_flags & MS_RDONLY) &&
452 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
453 return -EROFS;
454 }
455 return 0;
456}
457
458/**
459 * inode_permission - Check for access rights to a given inode
460 * @inode: Inode to check permission on
461 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
462 *
463 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
464 * this, letting us set arbitrary permissions for filesystem access without
465 * changing the "normal" UIDs which are used for other things.
466 *
467 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
468 */
469int inode_permission(struct inode *inode, int mask)
470{
471 int retval;
472
473 retval = sb_permission(inode->i_sb, inode, mask);
474 if (retval)
475 return retval;
476 return __inode_permission(inode, mask);
477}
Al Viro4d359502014-03-14 12:20:17 -0400478EXPORT_SYMBOL(inode_permission);
David Howells0bdaea92012-06-25 12:55:46 +0100479
480/**
Jan Blunck5dd784d02008-02-14 19:34:38 -0800481 * path_get - get a reference to a path
482 * @path: path to get the reference to
483 *
484 * Given a path increment the reference count to the dentry and the vfsmount.
485 */
Al Virodcf787f2013-03-01 23:51:07 -0500486void path_get(const struct path *path)
Jan Blunck5dd784d02008-02-14 19:34:38 -0800487{
488 mntget(path->mnt);
489 dget(path->dentry);
490}
491EXPORT_SYMBOL(path_get);
492
493/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800494 * path_put - put a reference to a path
495 * @path: path to put the reference to
496 *
497 * Given a path decrement the reference count to the dentry and the vfsmount.
498 */
Al Virodcf787f2013-03-01 23:51:07 -0500499void path_put(const struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500{
Jan Blunck1d957f92008-02-14 19:34:35 -0800501 dput(path->dentry);
502 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503}
Jan Blunck1d957f92008-02-14 19:34:35 -0800504EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505
Al Viro894bc8c2015-05-02 07:16:16 -0400506#define EMBEDDED_LEVELS 2
Al Viro1f55a6e2014-11-01 19:30:41 -0400507struct nameidata {
508 struct path path;
Al Viro1cf26652015-05-06 16:01:56 -0400509 struct qstr last;
Al Viro1f55a6e2014-11-01 19:30:41 -0400510 struct path root;
511 struct inode *inode; /* path.dentry.d_inode */
512 unsigned int flags;
Al Viro9883d182015-05-13 07:28:08 -0400513 unsigned seq, m_seq;
Al Viro1f55a6e2014-11-01 19:30:41 -0400514 int last_type;
515 unsigned depth;
NeilBrown756daf22015-03-23 13:37:38 +1100516 int total_link_count;
Al Viro697fc6c2015-05-02 19:38:35 -0400517 struct saved {
518 struct path link;
Al Virofceef392015-12-29 15:58:39 -0500519 struct delayed_call done;
Al Viro697fc6c2015-05-02 19:38:35 -0400520 const char *name;
Al Viro0450b2d2015-05-08 13:23:53 -0400521 unsigned seq;
Al Viro894bc8c2015-05-02 07:16:16 -0400522 } *stack, internal[EMBEDDED_LEVELS];
Al Viro9883d182015-05-13 07:28:08 -0400523 struct filename *name;
524 struct nameidata *saved;
Al Virofceef392015-12-29 15:58:39 -0500525 struct inode *link_inode;
Al Viro9883d182015-05-13 07:28:08 -0400526 unsigned root_seq;
527 int dfd;
Al Viro1f55a6e2014-11-01 19:30:41 -0400528};
529
Al Viro9883d182015-05-13 07:28:08 -0400530static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
Al Viro894bc8c2015-05-02 07:16:16 -0400531{
NeilBrown756daf22015-03-23 13:37:38 +1100532 struct nameidata *old = current->nameidata;
533 p->stack = p->internal;
Al Viroc8a53ee2015-05-12 18:43:07 -0400534 p->dfd = dfd;
535 p->name = name;
NeilBrown756daf22015-03-23 13:37:38 +1100536 p->total_link_count = old ? old->total_link_count : 0;
Al Viro9883d182015-05-13 07:28:08 -0400537 p->saved = old;
NeilBrown756daf22015-03-23 13:37:38 +1100538 current->nameidata = p;
Al Viro894bc8c2015-05-02 07:16:16 -0400539}
540
Al Viro9883d182015-05-13 07:28:08 -0400541static void restore_nameidata(void)
Al Viro894bc8c2015-05-02 07:16:16 -0400542{
Al Viro9883d182015-05-13 07:28:08 -0400543 struct nameidata *now = current->nameidata, *old = now->saved;
NeilBrown756daf22015-03-23 13:37:38 +1100544
545 current->nameidata = old;
546 if (old)
547 old->total_link_count = now->total_link_count;
Al Viroe1a63bb2015-12-05 21:06:33 -0500548 if (now->stack != now->internal)
NeilBrown756daf22015-03-23 13:37:38 +1100549 kfree(now->stack);
Al Viro894bc8c2015-05-02 07:16:16 -0400550}
551
552static int __nd_alloc_stack(struct nameidata *nd)
553{
Al Virobc40aee2015-05-09 13:04:24 -0400554 struct saved *p;
555
556 if (nd->flags & LOOKUP_RCU) {
557 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
558 GFP_ATOMIC);
559 if (unlikely(!p))
560 return -ECHILD;
561 } else {
562 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
Al Viro894bc8c2015-05-02 07:16:16 -0400563 GFP_KERNEL);
Al Virobc40aee2015-05-09 13:04:24 -0400564 if (unlikely(!p))
565 return -ENOMEM;
566 }
Al Viro894bc8c2015-05-02 07:16:16 -0400567 memcpy(p, nd->internal, sizeof(nd->internal));
568 nd->stack = p;
569 return 0;
570}
571
Eric W. Biederman397d4252015-08-15 20:27:13 -0500572/**
573 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
574 * @path: nameidate to verify
575 *
576 * Rename can sometimes move a file or directory outside of a bind
577 * mount, path_connected allows those cases to be detected.
578 */
579static bool path_connected(const struct path *path)
580{
581 struct vfsmount *mnt = path->mnt;
Eric W. Biedermaneaa95922018-03-14 18:20:29 -0500582 struct super_block *sb = mnt->mnt_sb;
Eric W. Biederman397d4252015-08-15 20:27:13 -0500583
Eric W. Biedermaneaa95922018-03-14 18:20:29 -0500584 /* Bind mounts and multi-root filesystems can have disconnected paths */
585 if (!(sb->s_iflags & SB_I_MULTIROOT) && (mnt->mnt_root == sb->s_root))
Eric W. Biederman397d4252015-08-15 20:27:13 -0500586 return true;
587
588 return is_subdir(path->dentry, mnt->mnt_root);
589}
590
Al Viro894bc8c2015-05-02 07:16:16 -0400591static inline int nd_alloc_stack(struct nameidata *nd)
592{
Al Viroda4e0be2015-05-03 20:52:15 -0400593 if (likely(nd->depth != EMBEDDED_LEVELS))
Al Viro894bc8c2015-05-02 07:16:16 -0400594 return 0;
595 if (likely(nd->stack != nd->internal))
596 return 0;
597 return __nd_alloc_stack(nd);
598}
599
Al Viro79733872015-05-09 12:55:43 -0400600static void drop_links(struct nameidata *nd)
601{
602 int i = nd->depth;
603 while (i--) {
604 struct saved *last = nd->stack + i;
Al Virofceef392015-12-29 15:58:39 -0500605 do_delayed_call(&last->done);
606 clear_delayed_call(&last->done);
Al Viro79733872015-05-09 12:55:43 -0400607 }
608}
609
610static void terminate_walk(struct nameidata *nd)
611{
612 drop_links(nd);
613 if (!(nd->flags & LOOKUP_RCU)) {
614 int i;
615 path_put(&nd->path);
616 for (i = 0; i < nd->depth; i++)
617 path_put(&nd->stack[i].link);
Al Viro102b8af2015-05-12 17:35:52 -0400618 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
619 path_put(&nd->root);
620 nd->root.mnt = NULL;
621 }
Al Viro79733872015-05-09 12:55:43 -0400622 } else {
623 nd->flags &= ~LOOKUP_RCU;
624 if (!(nd->flags & LOOKUP_ROOT))
625 nd->root.mnt = NULL;
626 rcu_read_unlock();
627 }
628 nd->depth = 0;
629}
630
631/* path_put is needed afterwards regardless of success or failure */
632static bool legitimize_path(struct nameidata *nd,
633 struct path *path, unsigned seq)
634{
635 int res = __legitimize_mnt(path->mnt, nd->m_seq);
636 if (unlikely(res)) {
637 if (res > 0)
638 path->mnt = NULL;
639 path->dentry = NULL;
640 return false;
641 }
642 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
643 path->dentry = NULL;
644 return false;
645 }
646 return !read_seqcount_retry(&path->dentry->d_seq, seq);
647}
648
649static bool legitimize_links(struct nameidata *nd)
650{
651 int i;
652 for (i = 0; i < nd->depth; i++) {
653 struct saved *last = nd->stack + i;
654 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
655 drop_links(nd);
656 nd->depth = i + 1;
657 return false;
658 }
659 }
660 return true;
661}
662
Al Viro19660af2011-03-25 10:32:48 -0400663/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100664 * Path walking has 2 modes, rcu-walk and ref-walk (see
Al Viro19660af2011-03-25 10:32:48 -0400665 * Documentation/filesystems/path-lookup.txt). In situations when we can't
666 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
Mike Marshall57e37152015-11-30 11:11:59 -0500667 * normal reference counts on dentries and vfsmounts to transition to ref-walk
Al Viro19660af2011-03-25 10:32:48 -0400668 * mode. Refcounts are grabbed at the last known good point before rcu-walk
669 * got stuck, so ref-walk may continue from there. If this is not successful
670 * (eg. a seqcount has changed), then failure is returned and it's up to caller
671 * to restart the path walk from the beginning in ref-walk mode.
Nick Piggin31e6b012011-01-07 17:49:52 +1100672 */
Nick Piggin31e6b012011-01-07 17:49:52 +1100673
674/**
Al Viro19660af2011-03-25 10:32:48 -0400675 * unlazy_walk - try to switch to ref-walk mode.
676 * @nd: nameidata pathwalk data
677 * @dentry: child of nd->path.dentry or NULL
Al Viro6e9918b2015-05-05 09:26:05 -0400678 * @seq: seq number to check dentry against
Randy Dunlap39191622011-01-08 19:36:21 -0800679 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100680 *
Al Viro19660af2011-03-25 10:32:48 -0400681 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
682 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
683 * @nd or NULL. Must be called from rcu-walk context.
Al Viro79733872015-05-09 12:55:43 -0400684 * Nothing should touch nameidata between unlazy_walk() failure and
685 * terminate_walk().
Nick Piggin31e6b012011-01-07 17:49:52 +1100686 */
Al Viro6e9918b2015-05-05 09:26:05 -0400687static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq)
Nick Piggin31e6b012011-01-07 17:49:52 +1100688{
Nick Piggin31e6b012011-01-07 17:49:52 +1100689 struct dentry *parent = nd->path.dentry;
690
691 BUG_ON(!(nd->flags & LOOKUP_RCU));
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700692
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700693 nd->flags &= ~LOOKUP_RCU;
Al Viro79733872015-05-09 12:55:43 -0400694 if (unlikely(!legitimize_links(nd)))
695 goto out2;
696 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
697 goto out2;
698 if (unlikely(!lockref_get_not_dead(&parent->d_lockref)))
699 goto out1;
Al Viro48a066e2013-09-29 22:06:07 -0400700
Linus Torvalds15570082013-09-02 11:38:06 -0700701 /*
702 * For a negative lookup, the lookup sequence point is the parents
703 * sequence point, and it only needs to revalidate the parent dentry.
704 *
705 * For a positive lookup, we need to move both the parent and the
706 * dentry from the RCU domain to be properly refcounted. And the
707 * sequence number in the dentry validates *both* dentry counters,
708 * since we checked the sequence number of the parent after we got
709 * the child sequence number. So we know the parent must still
710 * be valid if the child sequence number is still valid.
711 */
Al Viro19660af2011-03-25 10:32:48 -0400712 if (!dentry) {
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700713 if (read_seqcount_retry(&parent->d_seq, nd->seq))
714 goto out;
Al Viro19660af2011-03-25 10:32:48 -0400715 BUG_ON(nd->inode != parent->d_inode);
716 } else {
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700717 if (!lockref_get_not_dead(&dentry->d_lockref))
718 goto out;
Al Viro6e9918b2015-05-05 09:26:05 -0400719 if (read_seqcount_retry(&dentry->d_seq, seq))
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700720 goto drop_dentry;
Al Viro19660af2011-03-25 10:32:48 -0400721 }
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700722
723 /*
724 * Sequence counts matched. Now make sure that the root is
725 * still valid and get it if required.
726 */
727 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
Al Viro5a8d87e2015-05-12 00:10:18 -0400728 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
729 rcu_read_unlock();
730 dput(dentry);
731 return -ECHILD;
Al Viro79733872015-05-09 12:55:43 -0400732 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100733 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100734
Al Viro8b61e742013-11-08 12:45:01 -0500735 rcu_read_unlock();
Nick Piggin31e6b012011-01-07 17:49:52 +1100736 return 0;
Al Viro19660af2011-03-25 10:32:48 -0400737
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700738drop_dentry:
Al Viro8b61e742013-11-08 12:45:01 -0500739 rcu_read_unlock();
Linus Torvalds15570082013-09-02 11:38:06 -0700740 dput(dentry);
Linus Torvaldsd0d27272013-09-10 12:17:49 -0700741 goto drop_root_mnt;
Al Viro79733872015-05-09 12:55:43 -0400742out2:
743 nd->path.mnt = NULL;
744out1:
745 nd->path.dentry = NULL;
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700746out:
Al Viro8b61e742013-11-08 12:45:01 -0500747 rcu_read_unlock();
Linus Torvaldsd0d27272013-09-10 12:17:49 -0700748drop_root_mnt:
749 if (!(nd->flags & LOOKUP_ROOT))
750 nd->root.mnt = NULL;
Nick Piggin31e6b012011-01-07 17:49:52 +1100751 return -ECHILD;
752}
753
Al Viro79733872015-05-09 12:55:43 -0400754static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq)
755{
756 if (unlikely(!legitimize_path(nd, link, seq))) {
757 drop_links(nd);
758 nd->depth = 0;
759 nd->flags &= ~LOOKUP_RCU;
760 nd->path.mnt = NULL;
761 nd->path.dentry = NULL;
762 if (!(nd->flags & LOOKUP_ROOT))
763 nd->root.mnt = NULL;
764 rcu_read_unlock();
765 } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) {
766 return 0;
767 }
768 path_put(link);
769 return -ECHILD;
770}
771
Al Viro4ce16ef32012-06-10 16:10:59 -0400772static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
Nick Piggin34286d62011-01-07 17:49:57 +1100773{
Al Viro4ce16ef32012-06-10 16:10:59 -0400774 return dentry->d_op->d_revalidate(dentry, flags);
Nick Piggin34286d62011-01-07 17:49:57 +1100775}
776
Al Viro9f1fafe2011-03-25 11:00:12 -0400777/**
778 * complete_walk - successful completion of path walk
779 * @nd: pointer nameidata
Jeff Layton39159de2009-12-07 12:01:50 -0500780 *
Al Viro9f1fafe2011-03-25 11:00:12 -0400781 * If we had been in RCU mode, drop out of it and legitimize nd->path.
782 * Revalidate the final result, unless we'd already done that during
783 * the path walk or the filesystem doesn't ask for it. Return 0 on
784 * success, -error on failure. In case of failure caller does not
785 * need to drop nd->path.
Jeff Layton39159de2009-12-07 12:01:50 -0500786 */
Al Viro9f1fafe2011-03-25 11:00:12 -0400787static int complete_walk(struct nameidata *nd)
Jeff Layton39159de2009-12-07 12:01:50 -0500788{
Al Viro16c2cd72011-02-22 15:50:10 -0500789 struct dentry *dentry = nd->path.dentry;
Jeff Layton39159de2009-12-07 12:01:50 -0500790 int status;
Jeff Layton39159de2009-12-07 12:01:50 -0500791
Al Viro9f1fafe2011-03-25 11:00:12 -0400792 if (nd->flags & LOOKUP_RCU) {
Al Viro9f1fafe2011-03-25 11:00:12 -0400793 if (!(nd->flags & LOOKUP_ROOT))
794 nd->root.mnt = NULL;
Al Viro6e9918b2015-05-05 09:26:05 -0400795 if (unlikely(unlazy_walk(nd, NULL, 0)))
Al Viro48a066e2013-09-29 22:06:07 -0400796 return -ECHILD;
Al Viro9f1fafe2011-03-25 11:00:12 -0400797 }
798
Al Viro16c2cd72011-02-22 15:50:10 -0500799 if (likely(!(nd->flags & LOOKUP_JUMPED)))
Jeff Layton39159de2009-12-07 12:01:50 -0500800 return 0;
801
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500802 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
Al Viro16c2cd72011-02-22 15:50:10 -0500803 return 0;
804
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500805 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
Jeff Layton39159de2009-12-07 12:01:50 -0500806 if (status > 0)
807 return 0;
808
Al Viro16c2cd72011-02-22 15:50:10 -0500809 if (!status)
Jeff Layton39159de2009-12-07 12:01:50 -0500810 status = -ESTALE;
Al Viro16c2cd72011-02-22 15:50:10 -0500811
Jeff Layton39159de2009-12-07 12:01:50 -0500812 return status;
813}
814
Al Viro18d8c862015-05-12 16:32:34 -0400815static void set_root(struct nameidata *nd)
Al Viro2a737872009-04-07 11:49:53 -0400816{
Al Viro7bd88372014-09-13 21:55:46 -0400817 struct fs_struct *fs = current->fs;
Nick Pigginc28cc362011-01-07 17:49:53 +1100818
Al Viro9e6697e2015-12-05 20:07:21 -0500819 if (nd->flags & LOOKUP_RCU) {
820 unsigned seq;
821
822 do {
823 seq = read_seqcount_begin(&fs->seq);
824 nd->root = fs->root;
825 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
826 } while (read_seqcount_retry(&fs->seq, seq));
827 } else {
828 get_fs_root(fs, &nd->root);
829 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100830}
831
Jan Blunck1d957f92008-02-14 19:34:35 -0800832static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700833{
834 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800835 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700836 mntput(path->mnt);
837}
838
Nick Piggin7b9337a2011-01-14 08:42:43 +0000839static inline void path_to_nameidata(const struct path *path,
840 struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700841{
Nick Piggin31e6b012011-01-07 17:49:52 +1100842 if (!(nd->flags & LOOKUP_RCU)) {
843 dput(nd->path.dentry);
844 if (nd->path.mnt != path->mnt)
845 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800846 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100847 nd->path.mnt = path->mnt;
Jan Blunck4ac91372008-02-14 19:34:32 -0800848 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700849}
850
Al Viro248fb5b2015-12-05 20:51:58 -0500851static int nd_jump_root(struct nameidata *nd)
852{
853 if (nd->flags & LOOKUP_RCU) {
854 struct dentry *d;
855 nd->path = nd->root;
856 d = nd->path.dentry;
857 nd->inode = d->d_inode;
858 nd->seq = nd->root_seq;
859 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
860 return -ECHILD;
861 } else {
862 path_put(&nd->path);
863 nd->path = nd->root;
864 path_get(&nd->path);
865 nd->inode = nd->path.dentry->d_inode;
866 }
867 nd->flags |= LOOKUP_JUMPED;
868 return 0;
869}
870
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400871/*
Al Viro6b255392015-11-17 10:20:54 -0500872 * Helper to directly jump to a known parsed path from ->get_link,
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400873 * caller must have taken a reference to path beforehand.
874 */
Al Viro6e771372015-05-02 13:37:52 -0400875void nd_jump_link(struct path *path)
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400876{
Al Viro6e771372015-05-02 13:37:52 -0400877 struct nameidata *nd = current->nameidata;
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400878 path_put(&nd->path);
879
880 nd->path = *path;
881 nd->inode = nd->path.dentry->d_inode;
882 nd->flags |= LOOKUP_JUMPED;
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400883}
884
Al Virob9ff4422015-05-02 20:19:23 -0400885static inline void put_link(struct nameidata *nd)
Al Viro574197e2011-03-14 22:20:34 -0400886{
Al Viro21c30032015-05-03 21:06:24 -0400887 struct saved *last = nd->stack + --nd->depth;
Al Virofceef392015-12-29 15:58:39 -0500888 do_delayed_call(&last->done);
Al Viro6548fae2015-05-07 20:32:22 -0400889 if (!(nd->flags & LOOKUP_RCU))
890 path_put(&last->link);
Al Viro574197e2011-03-14 22:20:34 -0400891}
892
Linus Torvalds561ec642012-10-26 10:05:07 -0700893int sysctl_protected_symlinks __read_mostly = 0;
894int sysctl_protected_hardlinks __read_mostly = 0;
Kees Cook800179c2012-07-25 17:29:07 -0700895
896/**
897 * may_follow_link - Check symlink following for unsafe situations
Randy Dunlap55852632012-08-18 17:39:25 -0700898 * @nd: nameidata pathwalk data
Kees Cook800179c2012-07-25 17:29:07 -0700899 *
900 * In the case of the sysctl_protected_symlinks sysctl being enabled,
901 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
902 * in a sticky world-writable directory. This is to protect privileged
903 * processes from failing races against path names that may change out
904 * from under them by way of other users creating malicious symlinks.
905 * It will permit symlinks to be followed only when outside a sticky
906 * world-writable directory, or when the uid of the symlink and follower
907 * match, or when the directory owner matches the symlink's owner.
908 *
909 * Returns 0 if following the symlink is allowed, -ve on error.
910 */
Al Virofec2fa22015-05-06 15:58:18 -0400911static inline int may_follow_link(struct nameidata *nd)
Kees Cook800179c2012-07-25 17:29:07 -0700912{
913 const struct inode *inode;
914 const struct inode *parent;
Seth Forshee2d7f9e22016-04-26 14:36:23 -0500915 kuid_t puid;
Kees Cook800179c2012-07-25 17:29:07 -0700916
917 if (!sysctl_protected_symlinks)
918 return 0;
919
920 /* Allowed if owner and follower match. */
Al Virofceef392015-12-29 15:58:39 -0500921 inode = nd->link_inode;
Eric W. Biederman81abe272012-08-03 09:38:08 -0700922 if (uid_eq(current_cred()->fsuid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700923 return 0;
924
925 /* Allowed if parent directory not sticky and world-writable. */
Al Viroaa65fa32015-08-04 23:23:50 -0400926 parent = nd->inode;
Kees Cook800179c2012-07-25 17:29:07 -0700927 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
928 return 0;
929
930 /* Allowed if parent directory and link owner match. */
Seth Forshee2d7f9e22016-04-26 14:36:23 -0500931 puid = parent->i_uid;
932 if (uid_valid(puid) && uid_eq(puid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700933 return 0;
934
Al Viro319565022015-05-07 20:37:40 -0400935 if (nd->flags & LOOKUP_RCU)
936 return -ECHILD;
937
Al Viro1cf26652015-05-06 16:01:56 -0400938 audit_log_link_denied("follow_link", &nd->stack[0].link);
Kees Cook800179c2012-07-25 17:29:07 -0700939 return -EACCES;
940}
941
942/**
943 * safe_hardlink_source - Check for safe hardlink conditions
944 * @inode: the source inode to hardlink from
945 *
946 * Return false if at least one of the following conditions:
947 * - inode is not a regular file
948 * - inode is setuid
949 * - inode is setgid and group-exec
950 * - access failure for read and write
951 *
952 * Otherwise returns true.
953 */
954static bool safe_hardlink_source(struct inode *inode)
955{
956 umode_t mode = inode->i_mode;
957
958 /* Special files should not get pinned to the filesystem. */
959 if (!S_ISREG(mode))
960 return false;
961
962 /* Setuid files should not get pinned to the filesystem. */
963 if (mode & S_ISUID)
964 return false;
965
966 /* Executable setgid files should not get pinned to the filesystem. */
967 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
968 return false;
969
970 /* Hardlinking to unreadable or unwritable sources is dangerous. */
971 if (inode_permission(inode, MAY_READ | MAY_WRITE))
972 return false;
973
974 return true;
975}
976
977/**
978 * may_linkat - Check permissions for creating a hardlink
979 * @link: the source to hardlink from
980 *
981 * Block hardlink when all of:
982 * - sysctl_protected_hardlinks enabled
983 * - fsuid does not match inode
984 * - hardlink source is unsafe (see safe_hardlink_source() above)
Dirk Steinmetzf2ca3792015-10-20 16:09:19 +0200985 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
Kees Cook800179c2012-07-25 17:29:07 -0700986 *
987 * Returns 0 if successful, -ve on error.
988 */
989static int may_linkat(struct path *link)
990{
Kees Cook800179c2012-07-25 17:29:07 -0700991 struct inode *inode;
992
993 if (!sysctl_protected_hardlinks)
994 return 0;
995
Kees Cook800179c2012-07-25 17:29:07 -0700996 inode = link->dentry->d_inode;
997
998 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
999 * otherwise, it must be a safe source.
1000 */
Dirk Steinmetzf2ca3792015-10-20 16:09:19 +02001001 if (inode_owner_or_capable(inode) || safe_hardlink_source(inode))
Kees Cook800179c2012-07-25 17:29:07 -07001002 return 0;
1003
Kees Cooka51d9ea2012-07-25 17:29:08 -07001004 audit_log_link_denied("linkat", link);
Kees Cook800179c2012-07-25 17:29:07 -07001005 return -EPERM;
1006}
1007
Al Viro3b2e7f72015-04-19 00:53:50 -04001008static __always_inline
1009const char *get_link(struct nameidata *nd)
Ian Kent051d3812006-03-27 01:14:53 -08001010{
Al Viroab104922015-05-10 11:50:01 -04001011 struct saved *last = nd->stack + nd->depth - 1;
Al Viro1cf26652015-05-06 16:01:56 -04001012 struct dentry *dentry = last->link.dentry;
Al Virofceef392015-12-29 15:58:39 -05001013 struct inode *inode = nd->link_inode;
Al Viro6d7b5aa2012-06-10 04:15:17 -04001014 int error;
Al Viro0a959df2015-04-18 18:23:41 -04001015 const char *res;
Ian Kent051d3812006-03-27 01:14:53 -08001016
NeilBrown8fa9dd22015-03-23 13:37:40 +11001017 if (!(nd->flags & LOOKUP_RCU)) {
1018 touch_atime(&last->link);
1019 cond_resched();
Miklos Szeredi598e3c82016-09-16 12:44:20 +02001020 } else if (atime_needs_update_rcu(&last->link, inode)) {
NeilBrown8fa9dd22015-03-23 13:37:40 +11001021 if (unlikely(unlazy_walk(nd, NULL, 0)))
1022 return ERR_PTR(-ECHILD);
1023 touch_atime(&last->link);
1024 }
1025
NeilBrownbda0be72015-03-23 13:37:39 +11001026 error = security_inode_follow_link(dentry, inode,
1027 nd->flags & LOOKUP_RCU);
1028 if (unlikely(error))
Al Viro6920a442015-05-10 10:43:46 -04001029 return ERR_PTR(error);
Al Viro36f3b4f2011-02-22 21:24:38 -05001030
Al Viro86acdca12009-12-22 23:45:11 -05001031 nd->last_type = LAST_BIND;
Al Virod4dee482015-04-30 20:08:02 -04001032 res = inode->i_link;
1033 if (!res) {
Al Virofceef392015-12-29 15:58:39 -05001034 const char * (*get)(struct dentry *, struct inode *,
1035 struct delayed_call *);
1036 get = inode->i_op->get_link;
Al Viro8c1b4562015-05-09 18:15:21 -04001037 if (nd->flags & LOOKUP_RCU) {
Al Virofceef392015-12-29 15:58:39 -05001038 res = get(NULL, inode, &last->done);
Al Viro6b255392015-11-17 10:20:54 -05001039 if (res == ERR_PTR(-ECHILD)) {
1040 if (unlikely(unlazy_walk(nd, NULL, 0)))
1041 return ERR_PTR(-ECHILD);
Al Virofceef392015-12-29 15:58:39 -05001042 res = get(dentry, inode, &last->done);
Al Viro6b255392015-11-17 10:20:54 -05001043 }
1044 } else {
Al Virofceef392015-12-29 15:58:39 -05001045 res = get(dentry, inode, &last->done);
Al Viro8c1b4562015-05-09 18:15:21 -04001046 }
Al Virofceef392015-12-29 15:58:39 -05001047 if (IS_ERR_OR_NULL(res))
Al Virofab51e82015-05-10 11:01:00 -04001048 return res;
Ian Kent051d3812006-03-27 01:14:53 -08001049 }
Al Virofab51e82015-05-10 11:01:00 -04001050 if (*res == '/') {
Al Viro9e6697e2015-12-05 20:07:21 -05001051 if (!nd->root.mnt)
1052 set_root(nd);
Al Viro248fb5b2015-12-05 20:51:58 -05001053 if (unlikely(nd_jump_root(nd)))
1054 return ERR_PTR(-ECHILD);
Al Virofab51e82015-05-10 11:01:00 -04001055 while (unlikely(*++res == '/'))
1056 ;
1057 }
1058 if (!*res)
1059 res = NULL;
Al Viro0a959df2015-04-18 18:23:41 -04001060 return res;
1061}
Al Viro6d7b5aa2012-06-10 04:15:17 -04001062
David Howellsf015f1262012-06-25 12:55:28 +01001063/*
1064 * follow_up - Find the mountpoint of path's vfsmount
1065 *
1066 * Given a path, find the mountpoint of its source file system.
1067 * Replace @path with the path of the mountpoint in the parent mount.
1068 * Up is towards /.
1069 *
1070 * Return 1 if we went up a level and 0 if we were already at the
1071 * root.
1072 */
Al Virobab77eb2009-04-18 03:26:48 -04001073int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074{
Al Viro0714a532011-11-24 22:19:58 -05001075 struct mount *mnt = real_mount(path->mnt);
1076 struct mount *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +10001078
Al Viro48a066e2013-09-29 22:06:07 -04001079 read_seqlock_excl(&mount_lock);
Al Viro0714a532011-11-24 22:19:58 -05001080 parent = mnt->mnt_parent;
Al Viro3c0a6162012-07-18 17:32:50 +04001081 if (parent == mnt) {
Al Viro48a066e2013-09-29 22:06:07 -04001082 read_sequnlock_excl(&mount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 return 0;
1084 }
Al Viro0714a532011-11-24 22:19:58 -05001085 mntget(&parent->mnt);
Al Viroa73324d2011-11-24 22:25:07 -05001086 mountpoint = dget(mnt->mnt_mountpoint);
Al Viro48a066e2013-09-29 22:06:07 -04001087 read_sequnlock_excl(&mount_lock);
Al Virobab77eb2009-04-18 03:26:48 -04001088 dput(path->dentry);
1089 path->dentry = mountpoint;
1090 mntput(path->mnt);
Al Viro0714a532011-11-24 22:19:58 -05001091 path->mnt = &parent->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 return 1;
1093}
Al Viro4d359502014-03-14 12:20:17 -04001094EXPORT_SYMBOL(follow_up);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001096/*
David Howells9875cf82011-01-14 18:45:21 +00001097 * Perform an automount
1098 * - return -EISDIR to tell follow_managed() to stop and return the path we
1099 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 */
NeilBrown756daf22015-03-23 13:37:38 +11001101static int follow_automount(struct path *path, struct nameidata *nd,
David Howells9875cf82011-01-14 18:45:21 +00001102 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +11001103{
David Howells9875cf82011-01-14 18:45:21 +00001104 struct vfsmount *mnt;
David Howellsea5b7782011-01-14 19:10:03 +00001105 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001106
David Howells9875cf82011-01-14 18:45:21 +00001107 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1108 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -07001109
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +02001110 /* We don't want to mount if someone's just doing a stat -
1111 * unless they're stat'ing a directory and appended a '/' to
1112 * the name.
1113 *
1114 * We do, however, want to mount if someone wants to open or
1115 * create a file of any type under the mountpoint, wants to
1116 * traverse through the mountpoint or wants to open the
1117 * mounted directory. Also, autofs may mark negative dentries
1118 * as being automount points. These will need the attentions
1119 * of the daemon to instantiate them before they can be used.
David Howells9875cf82011-01-14 18:45:21 +00001120 */
NeilBrown756daf22015-03-23 13:37:38 +11001121 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1122 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +02001123 path->dentry->d_inode)
1124 return -EISDIR;
1125
NeilBrown756daf22015-03-23 13:37:38 +11001126 nd->total_link_count++;
1127 if (nd->total_link_count >= 40)
David Howells9875cf82011-01-14 18:45:21 +00001128 return -ELOOP;
1129
1130 mnt = path->dentry->d_op->d_automount(path);
1131 if (IS_ERR(mnt)) {
1132 /*
1133 * The filesystem is allowed to return -EISDIR here to indicate
1134 * it doesn't want to automount. For instance, autofs would do
1135 * this so that its userspace daemon can mount on this dentry.
1136 *
1137 * However, we can only permit this if it's a terminal point in
1138 * the path being looked up; if it wasn't then the remainder of
1139 * the path is inaccessible and we should say so.
1140 */
NeilBrown756daf22015-03-23 13:37:38 +11001141 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
David Howells9875cf82011-01-14 18:45:21 +00001142 return -EREMOTE;
1143 return PTR_ERR(mnt);
1144 }
David Howellsea5b7782011-01-14 19:10:03 +00001145
David Howells9875cf82011-01-14 18:45:21 +00001146 if (!mnt) /* mount collision */
1147 return 0;
1148
Al Viro8aef1882011-06-16 15:10:06 +01001149 if (!*need_mntput) {
1150 /* lock_mount() may release path->mnt on error */
1151 mntget(path->mnt);
1152 *need_mntput = true;
1153 }
Al Viro19a167a2011-01-17 01:35:23 -05001154 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +00001155
David Howellsea5b7782011-01-14 19:10:03 +00001156 switch (err) {
1157 case -EBUSY:
1158 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -05001159 return 0;
David Howellsea5b7782011-01-14 19:10:03 +00001160 case 0:
Al Viro8aef1882011-06-16 15:10:06 +01001161 path_put(path);
David Howellsea5b7782011-01-14 19:10:03 +00001162 path->mnt = mnt;
1163 path->dentry = dget(mnt->mnt_root);
David Howellsea5b7782011-01-14 19:10:03 +00001164 return 0;
Al Viro19a167a2011-01-17 01:35:23 -05001165 default:
1166 return err;
David Howellsea5b7782011-01-14 19:10:03 +00001167 }
Al Viro19a167a2011-01-17 01:35:23 -05001168
David Howells9875cf82011-01-14 18:45:21 +00001169}
1170
1171/*
1172 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +00001173 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +00001174 * - Flagged as mountpoint
1175 * - Flagged as automount point
1176 *
1177 * This may only be called in refwalk mode.
1178 *
1179 * Serialization is taken care of in namespace.c
1180 */
NeilBrown756daf22015-03-23 13:37:38 +11001181static int follow_managed(struct path *path, struct nameidata *nd)
David Howells9875cf82011-01-14 18:45:21 +00001182{
Al Viro8aef1882011-06-16 15:10:06 +01001183 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
David Howells9875cf82011-01-14 18:45:21 +00001184 unsigned managed;
1185 bool need_mntput = false;
Al Viro8aef1882011-06-16 15:10:06 +01001186 int ret = 0;
David Howells9875cf82011-01-14 18:45:21 +00001187
1188 /* Given that we're not holding a lock here, we retain the value in a
1189 * local variable for each dentry as we look at it so that we don't see
1190 * the components of that value change under us */
1191 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1192 managed &= DCACHE_MANAGED_DENTRY,
1193 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +00001194 /* Allow the filesystem to manage the transit without i_mutex
1195 * being held. */
1196 if (managed & DCACHE_MANAGE_TRANSIT) {
1197 BUG_ON(!path->dentry->d_op);
1198 BUG_ON(!path->dentry->d_op->d_manage);
Al Viro1aed3e42011-03-18 09:09:02 -04001199 ret = path->dentry->d_op->d_manage(path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001200 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001201 break;
David Howellscc53ce52011-01-14 18:45:26 +00001202 }
1203
David Howells9875cf82011-01-14 18:45:21 +00001204 /* Transit to a mounted filesystem. */
1205 if (managed & DCACHE_MOUNTED) {
1206 struct vfsmount *mounted = lookup_mnt(path);
1207 if (mounted) {
1208 dput(path->dentry);
1209 if (need_mntput)
1210 mntput(path->mnt);
1211 path->mnt = mounted;
1212 path->dentry = dget(mounted->mnt_root);
1213 need_mntput = true;
1214 continue;
1215 }
1216
1217 /* Something is mounted on this dentry in another
1218 * namespace and/or whatever was mounted there in this
Al Viro48a066e2013-09-29 22:06:07 -04001219 * namespace got unmounted before lookup_mnt() could
1220 * get it */
David Howells9875cf82011-01-14 18:45:21 +00001221 }
1222
1223 /* Handle an automount point */
1224 if (managed & DCACHE_NEED_AUTOMOUNT) {
NeilBrown756daf22015-03-23 13:37:38 +11001225 ret = follow_automount(path, nd, &need_mntput);
David Howells9875cf82011-01-14 18:45:21 +00001226 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001227 break;
David Howells9875cf82011-01-14 18:45:21 +00001228 continue;
1229 }
1230
1231 /* We didn't change the current path point */
1232 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 }
Al Viro8aef1882011-06-16 15:10:06 +01001234
1235 if (need_mntput && path->mnt == mnt)
1236 mntput(path->mnt);
Al Viroe9742b52016-03-05 22:04:59 -05001237 if (ret == -EISDIR || !ret)
1238 ret = 1;
Al Viro84027522015-04-22 10:30:08 -04001239 if (need_mntput)
1240 nd->flags |= LOOKUP_JUMPED;
1241 if (unlikely(ret < 0))
1242 path_put_conditional(path, nd);
1243 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244}
1245
David Howellscc53ce52011-01-14 18:45:26 +00001246int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247{
1248 struct vfsmount *mounted;
1249
Al Viro1c755af2009-04-18 14:06:57 -04001250 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -04001252 dput(path->dentry);
1253 mntput(path->mnt);
1254 path->mnt = mounted;
1255 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 return 1;
1257 }
1258 return 0;
1259}
Al Viro4d359502014-03-14 12:20:17 -04001260EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261
NeilBrownb8faf032014-08-04 17:06:29 +10001262static inline int managed_dentry_rcu(struct dentry *dentry)
Ian Kent62a73752011-03-25 01:51:02 +08001263{
NeilBrownb8faf032014-08-04 17:06:29 +10001264 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1265 dentry->d_op->d_manage(dentry, true) : 0;
Ian Kent62a73752011-03-25 01:51:02 +08001266}
1267
David Howells9875cf82011-01-14 18:45:21 +00001268/*
Al Viro287548e2011-05-27 06:50:06 -04001269 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1270 * we meet a managed dentry that would need blocking.
David Howells9875cf82011-01-14 18:45:21 +00001271 */
1272static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
Al Viro254cf582015-05-05 09:40:46 -04001273 struct inode **inode, unsigned *seqp)
David Howells9875cf82011-01-14 18:45:21 +00001274{
Ian Kent62a73752011-03-25 01:51:02 +08001275 for (;;) {
Al Viroc7105362011-11-24 18:22:03 -05001276 struct mount *mounted;
Ian Kent62a73752011-03-25 01:51:02 +08001277 /*
1278 * Don't forget we might have a non-mountpoint managed dentry
1279 * that wants to block transit.
1280 */
NeilBrownb8faf032014-08-04 17:06:29 +10001281 switch (managed_dentry_rcu(path->dentry)) {
1282 case -ECHILD:
1283 default:
David Howellsab909112011-01-14 18:46:51 +00001284 return false;
NeilBrownb8faf032014-08-04 17:06:29 +10001285 case -EISDIR:
1286 return true;
1287 case 0:
1288 break;
1289 }
Ian Kent62a73752011-03-25 01:51:02 +08001290
1291 if (!d_mountpoint(path->dentry))
NeilBrownb8faf032014-08-04 17:06:29 +10001292 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
Ian Kent62a73752011-03-25 01:51:02 +08001293
Al Viro474279d2013-10-01 16:11:26 -04001294 mounted = __lookup_mnt(path->mnt, path->dentry);
David Howells9875cf82011-01-14 18:45:21 +00001295 if (!mounted)
1296 break;
Al Viroc7105362011-11-24 18:22:03 -05001297 path->mnt = &mounted->mnt;
1298 path->dentry = mounted->mnt.mnt_root;
Al Viroa3fbbde2011-11-07 21:21:26 +00001299 nd->flags |= LOOKUP_JUMPED;
Al Viro254cf582015-05-05 09:40:46 -04001300 *seqp = read_seqcount_begin(&path->dentry->d_seq);
Linus Torvalds59430262011-07-18 15:43:29 -07001301 /*
1302 * Update the inode too. We don't need to re-check the
1303 * dentry sequence number here after this d_inode read,
1304 * because a mount-point is always pinned.
1305 */
1306 *inode = path->dentry->d_inode;
David Howells9875cf82011-01-14 18:45:21 +00001307 }
Al Virof5be3e29122014-09-13 21:50:45 -04001308 return !read_seqretry(&mount_lock, nd->m_seq) &&
NeilBrownb8faf032014-08-04 17:06:29 +10001309 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
Al Viro287548e2011-05-27 06:50:06 -04001310}
1311
Nick Piggin31e6b012011-01-07 17:49:52 +11001312static int follow_dotdot_rcu(struct nameidata *nd)
1313{
Al Viro4023bfc2014-09-13 21:59:43 -04001314 struct inode *inode = nd->inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001315
David Howells9875cf82011-01-14 18:45:21 +00001316 while (1) {
Al Viroaed434a2015-05-12 12:22:47 -04001317 if (path_equal(&nd->path, &nd->root))
Nick Piggin31e6b012011-01-07 17:49:52 +11001318 break;
Nick Piggin31e6b012011-01-07 17:49:52 +11001319 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1320 struct dentry *old = nd->path.dentry;
1321 struct dentry *parent = old->d_parent;
1322 unsigned seq;
1323
Al Viro4023bfc2014-09-13 21:59:43 -04001324 inode = parent->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001325 seq = read_seqcount_begin(&parent->d_seq);
Al Viroaed434a2015-05-12 12:22:47 -04001326 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1327 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001328 nd->path.dentry = parent;
1329 nd->seq = seq;
Eric W. Biederman397d4252015-08-15 20:27:13 -05001330 if (unlikely(!path_connected(&nd->path)))
1331 return -ENOENT;
Nick Piggin31e6b012011-01-07 17:49:52 +11001332 break;
Al Viroaed434a2015-05-12 12:22:47 -04001333 } else {
1334 struct mount *mnt = real_mount(nd->path.mnt);
1335 struct mount *mparent = mnt->mnt_parent;
1336 struct dentry *mountpoint = mnt->mnt_mountpoint;
1337 struct inode *inode2 = mountpoint->d_inode;
1338 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1339 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1340 return -ECHILD;
1341 if (&mparent->mnt == nd->path.mnt)
1342 break;
1343 /* we know that mountpoint was pinned */
1344 nd->path.dentry = mountpoint;
1345 nd->path.mnt = &mparent->mnt;
1346 inode = inode2;
1347 nd->seq = seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11001348 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001349 }
Al Viroaed434a2015-05-12 12:22:47 -04001350 while (unlikely(d_mountpoint(nd->path.dentry))) {
Al Virob37199e2014-03-20 15:18:22 -04001351 struct mount *mounted;
1352 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
Al Viroaed434a2015-05-12 12:22:47 -04001353 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1354 return -ECHILD;
Al Virob37199e2014-03-20 15:18:22 -04001355 if (!mounted)
1356 break;
1357 nd->path.mnt = &mounted->mnt;
1358 nd->path.dentry = mounted->mnt.mnt_root;
Al Viro4023bfc2014-09-13 21:59:43 -04001359 inode = nd->path.dentry->d_inode;
Al Virob37199e2014-03-20 15:18:22 -04001360 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Virob37199e2014-03-20 15:18:22 -04001361 }
Al Viro4023bfc2014-09-13 21:59:43 -04001362 nd->inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001363 return 0;
1364}
1365
David Howells9875cf82011-01-14 18:45:21 +00001366/*
David Howellscc53ce52011-01-14 18:45:26 +00001367 * Follow down to the covering mount currently visible to userspace. At each
1368 * point, the filesystem owning that dentry may be queried as to whether the
1369 * caller is permitted to proceed or not.
David Howellscc53ce52011-01-14 18:45:26 +00001370 */
Al Viro7cc90cc2011-03-18 09:04:20 -04001371int follow_down(struct path *path)
David Howellscc53ce52011-01-14 18:45:26 +00001372{
1373 unsigned managed;
1374 int ret;
1375
1376 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1377 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1378 /* Allow the filesystem to manage the transit without i_mutex
1379 * being held.
1380 *
1381 * We indicate to the filesystem if someone is trying to mount
1382 * something here. This gives autofs the chance to deny anyone
1383 * other than its daemon the right to mount on its
1384 * superstructure.
1385 *
1386 * The filesystem may sleep at this point.
1387 */
1388 if (managed & DCACHE_MANAGE_TRANSIT) {
1389 BUG_ON(!path->dentry->d_op);
1390 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +00001391 ret = path->dentry->d_op->d_manage(
Al Viro1aed3e42011-03-18 09:09:02 -04001392 path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001393 if (ret < 0)
1394 return ret == -EISDIR ? 0 : ret;
1395 }
1396
1397 /* Transit to a mounted filesystem. */
1398 if (managed & DCACHE_MOUNTED) {
1399 struct vfsmount *mounted = lookup_mnt(path);
1400 if (!mounted)
1401 break;
1402 dput(path->dentry);
1403 mntput(path->mnt);
1404 path->mnt = mounted;
1405 path->dentry = dget(mounted->mnt_root);
1406 continue;
1407 }
1408
1409 /* Don't handle automount points here */
1410 break;
1411 }
1412 return 0;
1413}
Al Viro4d359502014-03-14 12:20:17 -04001414EXPORT_SYMBOL(follow_down);
David Howellscc53ce52011-01-14 18:45:26 +00001415
1416/*
David Howells9875cf82011-01-14 18:45:21 +00001417 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1418 */
1419static void follow_mount(struct path *path)
1420{
1421 while (d_mountpoint(path->dentry)) {
1422 struct vfsmount *mounted = lookup_mnt(path);
1423 if (!mounted)
1424 break;
1425 dput(path->dentry);
1426 mntput(path->mnt);
1427 path->mnt = mounted;
1428 path->dentry = dget(mounted->mnt_root);
1429 }
1430}
1431
Eric W. Biedermaneedf2652016-06-02 10:29:47 -05001432static int path_parent_directory(struct path *path)
1433{
1434 struct dentry *old = path->dentry;
1435 /* rare case of legitimate dget_parent()... */
1436 path->dentry = dget_parent(path->dentry);
1437 dput(old);
1438 if (unlikely(!path_connected(path)))
1439 return -ENOENT;
1440 return 0;
1441}
1442
Eric W. Biederman397d4252015-08-15 20:27:13 -05001443static int follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444{
1445 while(1) {
Al Viro2a737872009-04-07 11:49:53 -04001446 if (nd->path.dentry == nd->root.dentry &&
1447 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 break;
1449 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001450 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Eric W. Biedermaneedf2652016-06-02 10:29:47 -05001451 int ret = path_parent_directory(&nd->path);
1452 if (ret)
1453 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 break;
1455 }
Al Viro3088dd72010-01-30 15:47:29 -05001456 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 }
Al Viro79ed0222009-04-18 13:59:41 -04001459 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001460 nd->inode = nd->path.dentry->d_inode;
Eric W. Biederman397d4252015-08-15 20:27:13 -05001461 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462}
1463
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464/*
Oleg Drokinf4fdace2016-07-07 22:04:04 -04001465 * This looks up the name in dcache and possibly revalidates the found dentry.
1466 * NULL is returned if the dentry does not exist in the cache.
Nick Pigginbaa03892010-08-18 04:37:31 +10001467 */
Al Viroe3c13922016-03-06 14:03:27 -05001468static struct dentry *lookup_dcache(const struct qstr *name,
1469 struct dentry *dir,
Al Viro6c51e512016-03-05 20:09:32 -05001470 unsigned int flags)
Nick Pigginbaa03892010-08-18 04:37:31 +10001471{
Nick Pigginbaa03892010-08-18 04:37:31 +10001472 struct dentry *dentry;
Miklos Szeredibad61182012-03-26 12:54:24 +02001473 int error;
Nick Pigginbaa03892010-08-18 04:37:31 +10001474
Miklos Szeredibad61182012-03-26 12:54:24 +02001475 dentry = d_lookup(dir, name);
1476 if (dentry) {
Jeff Layton39e3c952012-11-28 11:30:53 -05001477 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
Al Viro201f9562012-06-22 12:42:10 +04001478 error = d_revalidate(dentry, flags);
Miklos Szeredibad61182012-03-26 12:54:24 +02001479 if (unlikely(error <= 0)) {
Al Viro74ff0ff2016-03-05 22:37:46 -05001480 if (!error)
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001481 d_invalidate(dentry);
Al Viro74ff0ff2016-03-05 22:37:46 -05001482 dput(dentry);
1483 return ERR_PTR(error);
Miklos Szeredibad61182012-03-26 12:54:24 +02001484 }
1485 }
1486 }
Nick Pigginbaa03892010-08-18 04:37:31 +10001487 return dentry;
1488}
1489
1490/*
J. Bruce Fields13a2c3b2013-10-23 16:09:16 -04001491 * Call i_op->lookup on the dentry. The dentry must be negative and
1492 * unhashed.
Miklos Szeredibad61182012-03-26 12:54:24 +02001493 *
1494 * dir->d_inode->i_mutex must be held
Josef Bacik44396f42011-05-31 11:58:49 -04001495 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001496static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
Al Viro72bd8662012-06-10 17:17:17 -04001497 unsigned int flags)
Josef Bacik44396f42011-05-31 11:58:49 -04001498{
Josef Bacik44396f42011-05-31 11:58:49 -04001499 struct dentry *old;
1500
1501 /* Don't create child dentry for a dead directory. */
Miklos Szeredibad61182012-03-26 12:54:24 +02001502 if (unlikely(IS_DEADDIR(dir))) {
Miklos Szeredie188dc02012-02-03 14:25:18 +01001503 dput(dentry);
Josef Bacik44396f42011-05-31 11:58:49 -04001504 return ERR_PTR(-ENOENT);
Miklos Szeredie188dc02012-02-03 14:25:18 +01001505 }
Josef Bacik44396f42011-05-31 11:58:49 -04001506
Al Viro72bd8662012-06-10 17:17:17 -04001507 old = dir->i_op->lookup(dir, dentry, flags);
Josef Bacik44396f42011-05-31 11:58:49 -04001508 if (unlikely(old)) {
1509 dput(dentry);
1510 dentry = old;
1511 }
1512 return dentry;
1513}
1514
Al Viroe3c13922016-03-06 14:03:27 -05001515static struct dentry *__lookup_hash(const struct qstr *name,
Al Viro72bd8662012-06-10 17:17:17 -04001516 struct dentry *base, unsigned int flags)
Al Viroa3255542012-03-30 14:41:51 -04001517{
Al Viro6c51e512016-03-05 20:09:32 -05001518 struct dentry *dentry = lookup_dcache(name, base, flags);
Al Viroa3255542012-03-30 14:41:51 -04001519
Al Viro6c51e512016-03-05 20:09:32 -05001520 if (dentry)
Miklos Szeredibad61182012-03-26 12:54:24 +02001521 return dentry;
Al Viroa3255542012-03-30 14:41:51 -04001522
Al Viro6c51e512016-03-05 20:09:32 -05001523 dentry = d_alloc(base, name);
1524 if (unlikely(!dentry))
1525 return ERR_PTR(-ENOMEM);
1526
Al Viro72bd8662012-06-10 17:17:17 -04001527 return lookup_real(base->d_inode, dentry, flags);
Al Viroa3255542012-03-30 14:41:51 -04001528}
1529
Al Viroe97cdc82013-01-24 18:16:00 -05001530static int lookup_fast(struct nameidata *nd,
Al Viro254cf582015-05-05 09:40:46 -04001531 struct path *path, struct inode **inode,
1532 unsigned *seqp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533{
Jan Blunck4ac91372008-02-14 19:34:32 -08001534 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001535 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001536 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001537 int err;
1538
Al Viro3cac2602009-08-13 18:27:43 +04001539 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001540 * Rename seqlock is not required here because in the off chance
Al Viro5d0f49c2016-03-05 21:32:53 -05001541 * of a false negative due to a concurrent rename, the caller is
1542 * going to fall back to non-racy lookup.
Nick Pigginb04f7842010-08-18 04:37:34 +10001543 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001544 if (nd->flags & LOOKUP_RCU) {
1545 unsigned seq;
Al Viro766c4cb2015-05-07 19:24:57 -04001546 bool negative;
Linus Torvaldsda53be12013-05-21 15:22:44 -07001547 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
Al Viro5d0f49c2016-03-05 21:32:53 -05001548 if (unlikely(!dentry)) {
1549 if (unlazy_walk(nd, NULL, 0))
1550 return -ECHILD;
Al Viroe9742b52016-03-05 22:04:59 -05001551 return 0;
Al Viro5d0f49c2016-03-05 21:32:53 -05001552 }
Al Viro5a18fff2011-03-11 04:44:53 -05001553
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001554 /*
1555 * This sequence count validates that the inode matches
1556 * the dentry name information from lookup.
1557 */
David Howells63afdfc2015-05-06 15:59:00 +01001558 *inode = d_backing_inode(dentry);
Al Viro766c4cb2015-05-07 19:24:57 -04001559 negative = d_is_negative(dentry);
Al Viro5d0f49c2016-03-05 21:32:53 -05001560 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001561 return -ECHILD;
1562
1563 /*
1564 * This sequence count validates that the parent had no
1565 * changes while we did the lookup of the dentry above.
1566 *
1567 * The memory barrier in read_seqcount_begin of child is
1568 * enough, we can use __read_seqcount_retry here.
1569 */
Al Viro5d0f49c2016-03-05 21:32:53 -05001570 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
Nick Piggin31e6b012011-01-07 17:49:52 +11001571 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001572
Al Viro254cf582015-05-05 09:40:46 -04001573 *seqp = seq;
Al Viro5d0f49c2016-03-05 21:32:53 -05001574 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
Al Viro4ce16ef32012-06-10 16:10:59 -04001575 status = d_revalidate(dentry, nd->flags);
Al Viro5d0f49c2016-03-05 21:32:53 -05001576 if (unlikely(status <= 0)) {
1577 if (unlazy_walk(nd, dentry, seq))
1578 return -ECHILD;
1579 if (status == -ECHILD)
1580 status = d_revalidate(dentry, nd->flags);
1581 } else {
1582 /*
1583 * Note: do negative dentry check after revalidation in
1584 * case that drops it.
1585 */
1586 if (unlikely(negative))
1587 return -ENOENT;
1588 path->mnt = mnt;
1589 path->dentry = dentry;
1590 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
Al Viroe9742b52016-03-05 22:04:59 -05001591 return 1;
Al Viro5d0f49c2016-03-05 21:32:53 -05001592 if (unlazy_walk(nd, dentry, seq))
1593 return -ECHILD;
Al Viro24643082011-02-15 01:26:22 -05001594 }
Al Viro5a18fff2011-03-11 04:44:53 -05001595 } else {
Al Viroe97cdc82013-01-24 18:16:00 -05001596 dentry = __d_lookup(parent, &nd->last);
Al Viro5d0f49c2016-03-05 21:32:53 -05001597 if (unlikely(!dentry))
Al Viroe9742b52016-03-05 22:04:59 -05001598 return 0;
Al Viro5d0f49c2016-03-05 21:32:53 -05001599 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
1600 status = d_revalidate(dentry, nd->flags);
Nick Piggin31e6b012011-01-07 17:49:52 +11001601 }
Al Viro5a18fff2011-03-11 04:44:53 -05001602 if (unlikely(status <= 0)) {
Al Viroe9742b52016-03-05 22:04:59 -05001603 if (!status)
Al Viro5d0f49c2016-03-05 21:32:53 -05001604 d_invalidate(dentry);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001605 dput(dentry);
Al Viro5d0f49c2016-03-05 21:32:53 -05001606 return status;
Al Viro5a18fff2011-03-11 04:44:53 -05001607 }
Al Viro766c4cb2015-05-07 19:24:57 -04001608 if (unlikely(d_is_negative(dentry))) {
1609 dput(dentry);
1610 return -ENOENT;
1611 }
Al Viro5d0f49c2016-03-05 21:32:53 -05001612
David Howells9875cf82011-01-14 18:45:21 +00001613 path->mnt = mnt;
1614 path->dentry = dentry;
NeilBrown756daf22015-03-23 13:37:38 +11001615 err = follow_managed(path, nd);
Al Viroe9742b52016-03-05 22:04:59 -05001616 if (likely(err > 0))
David Howells63afdfc2015-05-06 15:59:00 +01001617 *inode = d_backing_inode(path->dentry);
Al Viro84027522015-04-22 10:30:08 -04001618 return err;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001619}
1620
1621/* Fast lookup failed, do it the slow way */
Al Viroe3c13922016-03-06 14:03:27 -05001622static struct dentry *lookup_slow(const struct qstr *name,
1623 struct dentry *dir,
1624 unsigned int flags)
Miklos Szeredi697f5142012-05-21 17:30:05 +02001625{
Al Viro94bdd652016-04-15 02:42:04 -04001626 struct dentry *dentry = ERR_PTR(-ENOENT), *old;
Al Viro19363862016-04-14 19:33:34 -04001627 struct inode *inode = dir->d_inode;
Al Virod9171b92016-04-15 03:33:13 -04001628 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
Al Viro19363862016-04-14 19:33:34 -04001629
Al Viro9902af72016-04-15 15:08:36 -04001630 inode_lock_shared(inode);
Al Viro19363862016-04-14 19:33:34 -04001631 /* Don't go there if it's already dead */
Al Viro94bdd652016-04-15 02:42:04 -04001632 if (unlikely(IS_DEADDIR(inode)))
1633 goto out;
1634again:
Al Virod9171b92016-04-15 03:33:13 -04001635 dentry = d_alloc_parallel(dir, name, &wq);
Al Viro94bdd652016-04-15 02:42:04 -04001636 if (IS_ERR(dentry))
1637 goto out;
1638 if (unlikely(!d_in_lookup(dentry))) {
Al Viro949a8522016-03-06 14:20:52 -05001639 if ((dentry->d_flags & DCACHE_OP_REVALIDATE) &&
1640 !(flags & LOOKUP_NO_REVAL)) {
1641 int error = d_revalidate(dentry, flags);
1642 if (unlikely(error <= 0)) {
Al Viro94bdd652016-04-15 02:42:04 -04001643 if (!error) {
Al Viro949a8522016-03-06 14:20:52 -05001644 d_invalidate(dentry);
Al Viro94bdd652016-04-15 02:42:04 -04001645 dput(dentry);
1646 goto again;
1647 }
Al Viro949a8522016-03-06 14:20:52 -05001648 dput(dentry);
1649 dentry = ERR_PTR(error);
1650 }
1651 }
Al Viro94bdd652016-04-15 02:42:04 -04001652 } else {
1653 old = inode->i_op->lookup(inode, dentry, flags);
1654 d_lookup_done(dentry);
1655 if (unlikely(old)) {
1656 dput(dentry);
1657 dentry = old;
Al Viro949a8522016-03-06 14:20:52 -05001658 }
1659 }
Al Viro94bdd652016-04-15 02:42:04 -04001660out:
Al Viro9902af72016-04-15 15:08:36 -04001661 inode_unlock_shared(inode);
Al Viroe3c13922016-03-06 14:03:27 -05001662 return dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663}
1664
Al Viro52094c82011-02-21 21:34:47 -05001665static inline int may_lookup(struct nameidata *nd)
1666{
1667 if (nd->flags & LOOKUP_RCU) {
Al Viro4ad5abb2011-06-20 19:57:03 -04001668 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
Al Viro52094c82011-02-21 21:34:47 -05001669 if (err != -ECHILD)
1670 return err;
Al Viro6e9918b2015-05-05 09:26:05 -04001671 if (unlazy_walk(nd, NULL, 0))
Al Viro52094c82011-02-21 21:34:47 -05001672 return -ECHILD;
1673 }
Al Viro4ad5abb2011-06-20 19:57:03 -04001674 return inode_permission(nd->inode, MAY_EXEC);
Al Viro52094c82011-02-21 21:34:47 -05001675}
1676
Al Viro9856fa12011-03-04 14:22:06 -05001677static inline int handle_dots(struct nameidata *nd, int type)
1678{
1679 if (type == LAST_DOTDOT) {
Al Viro9e6697e2015-12-05 20:07:21 -05001680 if (!nd->root.mnt)
1681 set_root(nd);
Al Viro9856fa12011-03-04 14:22:06 -05001682 if (nd->flags & LOOKUP_RCU) {
Al Viro70291ae2015-05-04 07:53:00 -04001683 return follow_dotdot_rcu(nd);
Al Viro9856fa12011-03-04 14:22:06 -05001684 } else
Eric W. Biederman397d4252015-08-15 20:27:13 -05001685 return follow_dotdot(nd);
Al Viro9856fa12011-03-04 14:22:06 -05001686 }
1687 return 0;
1688}
1689
Al Viro181548c2015-05-07 19:54:34 -04001690static int pick_link(struct nameidata *nd, struct path *link,
1691 struct inode *inode, unsigned seq)
Al Virod63ff282015-05-04 18:13:23 -04001692{
Al Viro626de992015-05-04 18:26:59 -04001693 int error;
Al Viro1cf26652015-05-06 16:01:56 -04001694 struct saved *last;
NeilBrown756daf22015-03-23 13:37:38 +11001695 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
Al Viro626de992015-05-04 18:26:59 -04001696 path_to_nameidata(link, nd);
1697 return -ELOOP;
1698 }
Al Virobc40aee2015-05-09 13:04:24 -04001699 if (!(nd->flags & LOOKUP_RCU)) {
Al Viro79733872015-05-09 12:55:43 -04001700 if (link->mnt == nd->path.mnt)
1701 mntget(link->mnt);
Al Virod63ff282015-05-04 18:13:23 -04001702 }
Al Viro626de992015-05-04 18:26:59 -04001703 error = nd_alloc_stack(nd);
1704 if (unlikely(error)) {
Al Virobc40aee2015-05-09 13:04:24 -04001705 if (error == -ECHILD) {
1706 if (unlikely(unlazy_link(nd, link, seq)))
1707 return -ECHILD;
1708 error = nd_alloc_stack(nd);
1709 }
1710 if (error) {
1711 path_put(link);
1712 return error;
1713 }
Al Viro626de992015-05-04 18:26:59 -04001714 }
1715
Al Viroab104922015-05-10 11:50:01 -04001716 last = nd->stack + nd->depth++;
Al Viro1cf26652015-05-06 16:01:56 -04001717 last->link = *link;
Al Virofceef392015-12-29 15:58:39 -05001718 clear_delayed_call(&last->done);
1719 nd->link_inode = inode;
Al Viro0450b2d2015-05-08 13:23:53 -04001720 last->seq = seq;
Al Virod63ff282015-05-04 18:13:23 -04001721 return 1;
1722}
1723
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001724/*
1725 * Do we need to follow links? We _really_ want to be able
1726 * to do this check without having to look at inode->i_op,
1727 * so we keep a cache of "no, this doesn't need follow_link"
1728 * for the common case.
1729 */
Al Viro254cf582015-05-05 09:40:46 -04001730static inline int should_follow_link(struct nameidata *nd, struct path *link,
Al Viro181548c2015-05-07 19:54:34 -04001731 int follow,
1732 struct inode *inode, unsigned seq)
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001733{
Al Virod63ff282015-05-04 18:13:23 -04001734 if (likely(!d_is_symlink(link->dentry)))
1735 return 0;
1736 if (!follow)
1737 return 0;
Al Viroa7f77542016-02-27 19:31:01 -05001738 /* make sure that d_is_symlink above matches inode */
1739 if (nd->flags & LOOKUP_RCU) {
1740 if (read_seqcount_retry(&link->dentry->d_seq, seq))
1741 return -ECHILD;
1742 }
Al Viro181548c2015-05-07 19:54:34 -04001743 return pick_link(nd, link, inode, seq);
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001744}
1745
Al Viro4693a542015-05-04 17:47:11 -04001746enum {WALK_GET = 1, WALK_PUT = 2};
1747
1748static int walk_component(struct nameidata *nd, int flags)
Al Viroce57dfc2011-03-13 19:58:58 -04001749{
Al Virocaa85632015-04-22 17:52:47 -04001750 struct path path;
Al Viroce57dfc2011-03-13 19:58:58 -04001751 struct inode *inode;
Al Viro254cf582015-05-05 09:40:46 -04001752 unsigned seq;
Al Viroce57dfc2011-03-13 19:58:58 -04001753 int err;
1754 /*
1755 * "." and ".." are special - ".." especially so because it has
1756 * to be able to know about the current root directory and
1757 * parent relationships.
1758 */
Al Viro4693a542015-05-04 17:47:11 -04001759 if (unlikely(nd->last_type != LAST_NORM)) {
1760 err = handle_dots(nd, nd->last_type);
1761 if (flags & WALK_PUT)
1762 put_link(nd);
1763 return err;
1764 }
Al Viro254cf582015-05-05 09:40:46 -04001765 err = lookup_fast(nd, &path, &inode, &seq);
Al Viroe9742b52016-03-05 22:04:59 -05001766 if (unlikely(err <= 0)) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001767 if (err < 0)
Al Virof0a9ba72015-05-04 07:59:30 -04001768 return err;
Al Viroe3c13922016-03-06 14:03:27 -05001769 path.dentry = lookup_slow(&nd->last, nd->path.dentry,
1770 nd->flags);
1771 if (IS_ERR(path.dentry))
1772 return PTR_ERR(path.dentry);
Al Viro7500c382016-03-31 00:23:05 -04001773
Al Viroe3c13922016-03-06 14:03:27 -05001774 path.mnt = nd->path.mnt;
1775 err = follow_managed(&path, nd);
1776 if (unlikely(err < 0))
Al Virof0a9ba72015-05-04 07:59:30 -04001777 return err;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001778
Al Viro7500c382016-03-31 00:23:05 -04001779 if (unlikely(d_is_negative(path.dentry))) {
1780 path_to_nameidata(&path, nd);
1781 return -ENOENT;
1782 }
1783
Al Viro254cf582015-05-05 09:40:46 -04001784 seq = 0; /* we are already out of RCU mode */
Al Virod4565642016-02-27 19:23:16 -05001785 inode = d_backing_inode(path.dentry);
Al Viroce57dfc2011-03-13 19:58:58 -04001786 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001787
Al Viro4693a542015-05-04 17:47:11 -04001788 if (flags & WALK_PUT)
1789 put_link(nd);
Al Viro181548c2015-05-07 19:54:34 -04001790 err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq);
Al Virod63ff282015-05-04 18:13:23 -04001791 if (unlikely(err))
1792 return err;
Al Virocaa85632015-04-22 17:52:47 -04001793 path_to_nameidata(&path, nd);
Al Viroce57dfc2011-03-13 19:58:58 -04001794 nd->inode = inode;
Al Viro254cf582015-05-05 09:40:46 -04001795 nd->seq = seq;
Al Viroce57dfc2011-03-13 19:58:58 -04001796 return 0;
1797}
1798
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799/*
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001800 * We can do the critical dentry name comparison and hashing
1801 * operations one word at a time, but we are limited to:
1802 *
1803 * - Architectures with fast unaligned word accesses. We could
1804 * do a "get_unaligned()" if this helps and is sufficiently
1805 * fast.
1806 *
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001807 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1808 * do not trap on the (extremely unlikely) case of a page
1809 * crossing operation.
1810 *
1811 * - Furthermore, we need an efficient 64-bit compile for the
1812 * 64-bit case in order to generate the "number of bytes in
1813 * the final mask". Again, that could be replaced with a
1814 * efficient population count instruction or similar.
1815 */
1816#ifdef CONFIG_DCACHE_WORD_ACCESS
1817
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001818#include <asm/word-at-a-time.h>
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001819
George Spelvin468a9422016-05-26 22:11:51 -04001820#ifdef HASH_MIX
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001821
George Spelvin468a9422016-05-26 22:11:51 -04001822/* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1823
1824#elif defined(CONFIG_64BIT)
George Spelvin2a18da7a2016-05-23 07:43:58 -04001825/*
1826 * Register pressure in the mixing function is an issue, particularly
1827 * on 32-bit x86, but almost any function requires one state value and
1828 * one temporary. Instead, use a function designed for two state values
1829 * and no temporaries.
1830 *
1831 * This function cannot create a collision in only two iterations, so
1832 * we have two iterations to achieve avalanche. In those two iterations,
1833 * we have six layers of mixing, which is enough to spread one bit's
1834 * influence out to 2^6 = 64 state bits.
1835 *
1836 * Rotate constants are scored by considering either 64 one-bit input
1837 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1838 * probability of that delta causing a change to each of the 128 output
1839 * bits, using a sample of random initial states.
1840 *
1841 * The Shannon entropy of the computed probabilities is then summed
1842 * to produce a score. Ideally, any input change has a 50% chance of
1843 * toggling any given output bit.
1844 *
1845 * Mixing scores (in bits) for (12,45):
1846 * Input delta: 1-bit 2-bit
1847 * 1 round: 713.3 42542.6
1848 * 2 rounds: 2753.7 140389.8
1849 * 3 rounds: 5954.1 233458.2
1850 * 4 rounds: 7862.6 256672.2
1851 * Perfect: 8192 258048
1852 * (64*128) (64*63/2 * 128)
1853 */
1854#define HASH_MIX(x, y, a) \
1855 ( x ^= (a), \
1856 y ^= x, x = rol64(x,12),\
1857 x += y, y = rol64(y,45),\
1858 y *= 9 )
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001859
George Spelvin0fed3ac2016-05-02 06:31:01 -04001860/*
George Spelvin2a18da7a2016-05-23 07:43:58 -04001861 * Fold two longs into one 32-bit hash value. This must be fast, but
1862 * latency isn't quite as critical, as there is a fair bit of additional
1863 * work done before the hash value is used.
George Spelvin0fed3ac2016-05-02 06:31:01 -04001864 */
George Spelvin2a18da7a2016-05-23 07:43:58 -04001865static inline unsigned int fold_hash(unsigned long x, unsigned long y)
George Spelvin0fed3ac2016-05-02 06:31:01 -04001866{
George Spelvin2a18da7a2016-05-23 07:43:58 -04001867 y ^= x * GOLDEN_RATIO_64;
1868 y *= GOLDEN_RATIO_64;
1869 return y >> 32;
George Spelvin0fed3ac2016-05-02 06:31:01 -04001870}
1871
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001872#else /* 32-bit case */
1873
George Spelvin2a18da7a2016-05-23 07:43:58 -04001874/*
1875 * Mixing scores (in bits) for (7,20):
1876 * Input delta: 1-bit 2-bit
1877 * 1 round: 330.3 9201.6
1878 * 2 rounds: 1246.4 25475.4
1879 * 3 rounds: 1907.1 31295.1
1880 * 4 rounds: 2042.3 31718.6
1881 * Perfect: 2048 31744
1882 * (32*64) (32*31/2 * 64)
1883 */
1884#define HASH_MIX(x, y, a) \
1885 ( x ^= (a), \
1886 y ^= x, x = rol32(x, 7),\
1887 x += y, y = rol32(y,20),\
1888 y *= 9 )
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001889
George Spelvin2a18da7a2016-05-23 07:43:58 -04001890static inline unsigned int fold_hash(unsigned long x, unsigned long y)
George Spelvin0fed3ac2016-05-02 06:31:01 -04001891{
George Spelvin2a18da7a2016-05-23 07:43:58 -04001892 /* Use arch-optimized multiply if one exists */
1893 return __hash_32(y ^ __hash_32(x));
George Spelvin0fed3ac2016-05-02 06:31:01 -04001894}
1895
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001896#endif
1897
George Spelvin2a18da7a2016-05-23 07:43:58 -04001898/*
1899 * Return the hash of a string of known length. This is carfully
1900 * designed to match hash_name(), which is the more critical function.
1901 * In particular, we must end by hashing a final word containing 0..7
1902 * payload bytes, to match the way that hash_name() iterates until it
1903 * finds the delimiter after the name.
1904 */
Linus Torvalds8387ff22016-06-10 07:51:30 -07001905unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001906{
Linus Torvalds8387ff22016-06-10 07:51:30 -07001907 unsigned long a, x = 0, y = (unsigned long)salt;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001908
1909 for (;;) {
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001910 if (!len)
1911 goto done;
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001912 a = load_unaligned_zeropad(name);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001913 if (len < sizeof(unsigned long))
1914 break;
George Spelvin2a18da7a2016-05-23 07:43:58 -04001915 HASH_MIX(x, y, a);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001916 name += sizeof(unsigned long);
1917 len -= sizeof(unsigned long);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001918 }
George Spelvin2a18da7a2016-05-23 07:43:58 -04001919 x ^= a & bytemask_from_count(len);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001920done:
George Spelvin2a18da7a2016-05-23 07:43:58 -04001921 return fold_hash(x, y);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001922}
1923EXPORT_SYMBOL(full_name_hash);
1924
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001925/* Return the "hash_len" (hash and length) of a null-terminated string */
Linus Torvalds8387ff22016-06-10 07:51:30 -07001926u64 hashlen_string(const void *salt, const char *name)
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001927{
Linus Torvalds8387ff22016-06-10 07:51:30 -07001928 unsigned long a = 0, x = 0, y = (unsigned long)salt;
1929 unsigned long adata, mask, len;
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001930 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1931
Linus Torvalds8387ff22016-06-10 07:51:30 -07001932 len = 0;
1933 goto inside;
1934
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001935 do {
George Spelvin2a18da7a2016-05-23 07:43:58 -04001936 HASH_MIX(x, y, a);
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001937 len += sizeof(unsigned long);
Linus Torvalds8387ff22016-06-10 07:51:30 -07001938inside:
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001939 a = load_unaligned_zeropad(name+len);
1940 } while (!has_zero(a, &adata, &constants));
1941
1942 adata = prep_zero_mask(a, adata, &constants);
1943 mask = create_zero_mask(adata);
George Spelvin2a18da7a2016-05-23 07:43:58 -04001944 x ^= a & zero_bytemask(mask);
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001945
George Spelvin2a18da7a2016-05-23 07:43:58 -04001946 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001947}
1948EXPORT_SYMBOL(hashlen_string);
1949
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001950/*
1951 * Calculate the length and hash of the path component, and
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001952 * return the "hash_len" as the result.
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001953 */
Linus Torvalds8387ff22016-06-10 07:51:30 -07001954static inline u64 hash_name(const void *salt, const char *name)
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001955{
Linus Torvalds8387ff22016-06-10 07:51:30 -07001956 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
1957 unsigned long adata, bdata, mask, len;
Linus Torvalds36126f82012-05-26 10:43:17 -07001958 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001959
Linus Torvalds8387ff22016-06-10 07:51:30 -07001960 len = 0;
1961 goto inside;
1962
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001963 do {
George Spelvin2a18da7a2016-05-23 07:43:58 -04001964 HASH_MIX(x, y, a);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001965 len += sizeof(unsigned long);
Linus Torvalds8387ff22016-06-10 07:51:30 -07001966inside:
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001967 a = load_unaligned_zeropad(name+len);
Linus Torvalds36126f82012-05-26 10:43:17 -07001968 b = a ^ REPEAT_BYTE('/');
1969 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001970
Linus Torvalds36126f82012-05-26 10:43:17 -07001971 adata = prep_zero_mask(a, adata, &constants);
1972 bdata = prep_zero_mask(b, bdata, &constants);
Linus Torvalds36126f82012-05-26 10:43:17 -07001973 mask = create_zero_mask(adata | bdata);
George Spelvin2a18da7a2016-05-23 07:43:58 -04001974 x ^= a & zero_bytemask(mask);
Linus Torvalds36126f82012-05-26 10:43:17 -07001975
George Spelvin2a18da7a2016-05-23 07:43:58 -04001976 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001977}
1978
George Spelvin2a18da7a2016-05-23 07:43:58 -04001979#else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001980
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001981/* Return the hash of a string of known length */
Linus Torvalds8387ff22016-06-10 07:51:30 -07001982unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
Linus Torvalds0145acc2012-03-02 14:32:59 -08001983{
Linus Torvalds8387ff22016-06-10 07:51:30 -07001984 unsigned long hash = init_name_hash(salt);
Linus Torvalds0145acc2012-03-02 14:32:59 -08001985 while (len--)
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001986 hash = partial_name_hash((unsigned char)*name++, hash);
Linus Torvalds0145acc2012-03-02 14:32:59 -08001987 return end_name_hash(hash);
1988}
Linus Torvaldsae942ae2012-03-02 19:40:57 -08001989EXPORT_SYMBOL(full_name_hash);
Linus Torvalds0145acc2012-03-02 14:32:59 -08001990
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001991/* Return the "hash_len" (hash and length) of a null-terminated string */
Linus Torvalds8387ff22016-06-10 07:51:30 -07001992u64 hashlen_string(const void *salt, const char *name)
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001993{
Linus Torvalds8387ff22016-06-10 07:51:30 -07001994 unsigned long hash = init_name_hash(salt);
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001995 unsigned long len = 0, c;
1996
1997 c = (unsigned char)*name;
George Spelvine0ab7af2016-05-29 08:05:56 -04001998 while (c) {
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001999 len++;
2000 hash = partial_name_hash(c, hash);
2001 c = (unsigned char)name[len];
George Spelvine0ab7af2016-05-29 08:05:56 -04002002 }
George Spelvinfcfd2fb2016-05-20 08:41:37 -04002003 return hashlen_create(end_name_hash(hash), len);
2004}
George Spelvinf2a031b2016-05-29 01:26:41 -04002005EXPORT_SYMBOL(hashlen_string);
George Spelvinfcfd2fb2016-05-20 08:41:37 -04002006
Linus Torvalds3ddcd052011-08-06 22:45:50 -07002007/*
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002008 * We know there's a real path component here of at least
2009 * one character.
2010 */
Linus Torvalds8387ff22016-06-10 07:51:30 -07002011static inline u64 hash_name(const void *salt, const char *name)
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002012{
Linus Torvalds8387ff22016-06-10 07:51:30 -07002013 unsigned long hash = init_name_hash(salt);
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002014 unsigned long len = 0, c;
2015
2016 c = (unsigned char)*name;
2017 do {
2018 len++;
2019 hash = partial_name_hash(c, hash);
2020 c = (unsigned char)name[len];
2021 } while (c && c != '/');
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002022 return hashlen_create(end_name_hash(hash), len);
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002023}
2024
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08002025#endif
2026
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002027/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01002029 * This is the basic name resolution function, turning a pathname into
2030 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01002032 * Returns 0 and nd will have valid dentry and mnt on success.
2033 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 */
Al Viro6de88d72009-08-09 01:41:57 +04002035static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037 int err;
Al Viro32cd7462015-04-18 20:30:49 -04002038
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039 while (*name=='/')
2040 name++;
2041 if (!*name)
Al Viro9e18f102015-04-18 20:44:34 -04002042 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 /* At this point we know we have a real path component. */
2045 for(;;) {
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002046 u64 hash_len;
Al Virofe479a52011-02-22 15:10:03 -05002047 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048
Al Viro52094c82011-02-21 21:34:47 -05002049 err = may_lookup(nd);
George Spelvin2a18da7a2016-05-23 07:43:58 -04002050 if (err)
Al Viro3595e232015-05-09 16:54:45 -04002051 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052
Linus Torvalds8387ff22016-06-10 07:51:30 -07002053 hash_len = hash_name(nd->path.dentry, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054
Al Virofe479a52011-02-22 15:10:03 -05002055 type = LAST_NORM;
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002056 if (name[0] == '.') switch (hashlen_len(hash_len)) {
Al Virofe479a52011-02-22 15:10:03 -05002057 case 2:
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002058 if (name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05002059 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05002060 nd->flags |= LOOKUP_JUMPED;
2061 }
Al Virofe479a52011-02-22 15:10:03 -05002062 break;
2063 case 1:
2064 type = LAST_DOT;
2065 }
Al Viro5a202bc2011-03-08 14:17:44 -05002066 if (likely(type == LAST_NORM)) {
2067 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05002068 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05002069 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
James Hogana060dc52014-09-16 13:07:35 +01002070 struct qstr this = { { .hash_len = hash_len }, .name = name };
Linus Torvaldsda53be12013-05-21 15:22:44 -07002071 err = parent->d_op->d_hash(parent, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05002072 if (err < 0)
Al Viro3595e232015-05-09 16:54:45 -04002073 return err;
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002074 hash_len = this.hash_len;
2075 name = this.name;
Al Viro5a202bc2011-03-08 14:17:44 -05002076 }
2077 }
Al Virofe479a52011-02-22 15:10:03 -05002078
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002079 nd->last.hash_len = hash_len;
2080 nd->last.name = name;
Al Viro5f4a6a62013-01-24 18:04:22 -05002081 nd->last_type = type;
2082
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002083 name += hashlen_len(hash_len);
2084 if (!*name)
Al Virobdf6cbf2015-04-18 20:21:40 -04002085 goto OK;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002086 /*
2087 * If it wasn't NUL, we know it was '/'. Skip that
2088 * slash, and continue until no more slashes.
2089 */
2090 do {
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002091 name++;
2092 } while (unlikely(*name == '/'));
Al Viro8620c232015-05-04 08:58:35 -04002093 if (unlikely(!*name)) {
2094OK:
Al Viro368ee9b2015-05-08 17:19:59 -04002095 /* pathname body, done */
Al Viro8620c232015-05-04 08:58:35 -04002096 if (!nd->depth)
2097 return 0;
2098 name = nd->stack[nd->depth - 1].name;
Al Viro368ee9b2015-05-08 17:19:59 -04002099 /* trailing symlink, done */
Al Viro8620c232015-05-04 08:58:35 -04002100 if (!name)
2101 return 0;
2102 /* last component of nested symlink */
Al Viro4693a542015-05-04 17:47:11 -04002103 err = walk_component(nd, WALK_GET | WALK_PUT);
Al Viro8620c232015-05-04 08:58:35 -04002104 } else {
Al Viro4693a542015-05-04 17:47:11 -04002105 err = walk_component(nd, WALK_GET);
Al Viro8620c232015-05-04 08:58:35 -04002106 }
Al Viroce57dfc2011-03-13 19:58:58 -04002107 if (err < 0)
Al Viro3595e232015-05-09 16:54:45 -04002108 return err;
Al Virofe479a52011-02-22 15:10:03 -05002109
Al Viroce57dfc2011-03-13 19:58:58 -04002110 if (err) {
Al Viro626de992015-05-04 18:26:59 -04002111 const char *s = get_link(nd);
Al Viro5a460272015-04-17 23:44:45 -04002112
Viresh Kumara1c83682015-08-12 15:59:44 +05302113 if (IS_ERR(s))
Al Viro3595e232015-05-09 16:54:45 -04002114 return PTR_ERR(s);
Al Virod40bcc02015-04-18 20:03:03 -04002115 err = 0;
2116 if (unlikely(!s)) {
2117 /* jumped */
Al Virob9ff4422015-05-02 20:19:23 -04002118 put_link(nd);
Al Virod40bcc02015-04-18 20:03:03 -04002119 } else {
Al Virofab51e82015-05-10 11:01:00 -04002120 nd->stack[nd->depth - 1].name = name;
2121 name = s;
2122 continue;
Al Virod40bcc02015-04-18 20:03:03 -04002123 }
Nick Piggin31e6b012011-01-07 17:49:52 +11002124 }
Al Viro97242f92015-08-01 19:59:28 -04002125 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2126 if (nd->flags & LOOKUP_RCU) {
2127 if (unlazy_walk(nd, NULL, 0))
2128 return -ECHILD;
2129 }
Al Viro3595e232015-05-09 16:54:45 -04002130 return -ENOTDIR;
Al Viro97242f92015-08-01 19:59:28 -04002131 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133}
2134
Al Viroc8a53ee2015-05-12 18:43:07 -04002135static const char *path_init(struct nameidata *nd, unsigned flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136{
Prasanna Medaea3834d2005-04-29 16:00:17 +01002137 int retval = 0;
Al Viroc8a53ee2015-05-12 18:43:07 -04002138 const char *s = nd->name->name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139
Linus Torvalds012e79b2017-04-02 17:10:08 -07002140 if (!*s)
2141 flags &= ~LOOKUP_RCU;
2142
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro980f3ea2014-11-20 14:20:24 -05002144 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 nd->depth = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05002146 if (flags & LOOKUP_ROOT) {
David Howellsb18825a2013-09-12 19:22:53 +01002147 struct dentry *root = nd->root.dentry;
2148 struct inode *inode = root->d_inode;
Al Virofd2f7cb2015-02-22 20:07:13 -05002149 if (*s) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002150 if (!d_can_lookup(root))
Al Viro368ee9b2015-05-08 17:19:59 -04002151 return ERR_PTR(-ENOTDIR);
Al Viro73d049a2011-03-11 12:08:24 -05002152 retval = inode_permission(inode, MAY_EXEC);
2153 if (retval)
Al Viro368ee9b2015-05-08 17:19:59 -04002154 return ERR_PTR(retval);
Al Viro73d049a2011-03-11 12:08:24 -05002155 }
Al Viro5b6ca022011-03-09 23:04:47 -05002156 nd->path = nd->root;
2157 nd->inode = inode;
2158 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05002159 rcu_read_lock();
Al Viro5b6ca022011-03-09 23:04:47 -05002160 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro8f47a0162015-05-09 19:02:01 -04002161 nd->root_seq = nd->seq;
Al Viro48a066e2013-09-29 22:06:07 -04002162 nd->m_seq = read_seqbegin(&mount_lock);
Al Viro5b6ca022011-03-09 23:04:47 -05002163 } else {
2164 path_get(&nd->path);
2165 }
Al Viro368ee9b2015-05-08 17:19:59 -04002166 return s;
Al Viro5b6ca022011-03-09 23:04:47 -05002167 }
2168
Al Viro2a737872009-04-07 11:49:53 -04002169 nd->root.mnt = NULL;
Al Viro248fb5b2015-12-05 20:51:58 -05002170 nd->path.mnt = NULL;
2171 nd->path.dentry = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172
Al Viro48a066e2013-09-29 22:06:07 -04002173 nd->m_seq = read_seqbegin(&mount_lock);
Al Virofd2f7cb2015-02-22 20:07:13 -05002174 if (*s == '/') {
Al Viro9e6697e2015-12-05 20:07:21 -05002175 if (flags & LOOKUP_RCU)
Al Viro8b61e742013-11-08 12:45:01 -05002176 rcu_read_lock();
Al Viro9e6697e2015-12-05 20:07:21 -05002177 set_root(nd);
Al Viro248fb5b2015-12-05 20:51:58 -05002178 if (likely(!nd_jump_root(nd)))
Al Viroef55d912015-12-05 20:25:06 -05002179 return s;
Al Viro248fb5b2015-12-05 20:51:58 -05002180 nd->root.mnt = NULL;
Al Viroef55d912015-12-05 20:25:06 -05002181 rcu_read_unlock();
2182 return ERR_PTR(-ECHILD);
Al Viroc8a53ee2015-05-12 18:43:07 -04002183 } else if (nd->dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05002184 if (flags & LOOKUP_RCU) {
2185 struct fs_struct *fs = current->fs;
2186 unsigned seq;
2187
Al Viro8b61e742013-11-08 12:45:01 -05002188 rcu_read_lock();
Al Viroe41f7d42011-02-22 14:02:58 -05002189
2190 do {
2191 seq = read_seqcount_begin(&fs->seq);
2192 nd->path = fs->pwd;
Al Viroef55d912015-12-05 20:25:06 -05002193 nd->inode = nd->path.dentry->d_inode;
Al Viroe41f7d42011-02-22 14:02:58 -05002194 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2195 } while (read_seqcount_retry(&fs->seq, seq));
2196 } else {
2197 get_fs_pwd(current->fs, &nd->path);
Al Viroef55d912015-12-05 20:25:06 -05002198 nd->inode = nd->path.dentry->d_inode;
Al Viroe41f7d42011-02-22 14:02:58 -05002199 }
Al Viroef55d912015-12-05 20:25:06 -05002200 return s;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002201 } else {
Jeff Layton582aa642012-12-11 08:56:16 -05002202 /* Caller must check execute permissions on the starting path component */
Al Viroc8a53ee2015-05-12 18:43:07 -04002203 struct fd f = fdget_raw(nd->dfd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002204 struct dentry *dentry;
2205
Al Viro2903ff02012-08-28 12:52:22 -04002206 if (!f.file)
Al Viro368ee9b2015-05-08 17:19:59 -04002207 return ERR_PTR(-EBADF);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002208
Al Viro2903ff02012-08-28 12:52:22 -04002209 dentry = f.file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002210
Al Virofd2f7cb2015-02-22 20:07:13 -05002211 if (*s) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002212 if (!d_can_lookup(dentry)) {
Al Viro2903ff02012-08-28 12:52:22 -04002213 fdput(f);
Al Viro368ee9b2015-05-08 17:19:59 -04002214 return ERR_PTR(-ENOTDIR);
Al Viro2903ff02012-08-28 12:52:22 -04002215 }
Al Virof52e0c12011-03-14 18:56:51 -04002216 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002217
Al Viro2903ff02012-08-28 12:52:22 -04002218 nd->path = f.file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05002219 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05002220 rcu_read_lock();
Al Viro34a26b92015-05-11 08:05:05 -04002221 nd->inode = nd->path.dentry->d_inode;
2222 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viroe41f7d42011-02-22 14:02:58 -05002223 } else {
Al Viro2903ff02012-08-28 12:52:22 -04002224 path_get(&nd->path);
Al Viro34a26b92015-05-11 08:05:05 -04002225 nd->inode = nd->path.dentry->d_inode;
Al Viroe41f7d42011-02-22 14:02:58 -05002226 }
Al Viro34a26b92015-05-11 08:05:05 -04002227 fdput(f);
Al Viro368ee9b2015-05-08 17:19:59 -04002228 return s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229 }
Al Viro9b4a9b12009-04-07 11:44:16 -04002230}
2231
Al Viro3bdba282015-05-08 17:37:07 -04002232static const char *trailing_symlink(struct nameidata *nd)
Al Viro95fa25d2015-04-22 13:46:57 -04002233{
2234 const char *s;
Al Virofec2fa22015-05-06 15:58:18 -04002235 int error = may_follow_link(nd);
Al Virodeb106c2015-05-08 18:05:21 -04002236 if (unlikely(error))
Al Viro3bdba282015-05-08 17:37:07 -04002237 return ERR_PTR(error);
Al Viro95fa25d2015-04-22 13:46:57 -04002238 nd->flags |= LOOKUP_PARENT;
Al Virofab51e82015-05-10 11:01:00 -04002239 nd->stack[0].name = NULL;
Al Viro3b2e7f72015-04-19 00:53:50 -04002240 s = get_link(nd);
Al Virodeb106c2015-05-08 18:05:21 -04002241 return s ? s : "";
Al Viro95fa25d2015-04-22 13:46:57 -04002242}
2243
Al Virocaa85632015-04-22 17:52:47 -04002244static inline int lookup_last(struct nameidata *nd)
Al Virobd92d7f2011-03-14 19:54:59 -04002245{
2246 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2247 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2248
2249 nd->flags &= ~LOOKUP_PARENT;
Al Virodeb106c2015-05-08 18:05:21 -04002250 return walk_component(nd,
Al Viro4693a542015-05-04 17:47:11 -04002251 nd->flags & LOOKUP_FOLLOW
2252 ? nd->depth
2253 ? WALK_PUT | WALK_GET
2254 : WALK_GET
2255 : 0);
Al Virobd92d7f2011-03-14 19:54:59 -04002256}
2257
Al Viro9b4a9b12009-04-07 11:44:16 -04002258/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroc8a53ee2015-05-12 18:43:07 -04002259static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
Al Viro9b4a9b12009-04-07 11:44:16 -04002260{
Al Viroc8a53ee2015-05-12 18:43:07 -04002261 const char *s = path_init(nd, flags);
Al Virobd92d7f2011-03-14 19:54:59 -04002262 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11002263
Al Viro368ee9b2015-05-08 17:19:59 -04002264 if (IS_ERR(s))
2265 return PTR_ERR(s);
Al Viro3bdba282015-05-08 17:37:07 -04002266 while (!(err = link_path_walk(s, nd))
2267 && ((err = lookup_last(nd)) > 0)) {
2268 s = trailing_symlink(nd);
2269 if (IS_ERR(s)) {
2270 err = PTR_ERR(s);
2271 break;
Al Virobd92d7f2011-03-14 19:54:59 -04002272 }
2273 }
Al Viro9f1fafe2011-03-25 11:00:12 -04002274 if (!err)
2275 err = complete_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04002276
Al Virodeb106c2015-05-08 18:05:21 -04002277 if (!err && nd->flags & LOOKUP_DIRECTORY)
2278 if (!d_can_lookup(nd->path.dentry))
Al Virobd23a532011-03-23 09:56:30 -04002279 err = -ENOTDIR;
Al Viro625b6d12015-05-12 16:36:12 -04002280 if (!err) {
2281 *path = nd->path;
2282 nd->path.mnt = NULL;
2283 nd->path.dentry = NULL;
2284 }
2285 terminate_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04002286 return err;
Al Viroee0827c2011-02-21 23:38:09 -05002287}
Nick Piggin31e6b012011-01-07 17:49:52 +11002288
Al Viro625b6d12015-05-12 16:36:12 -04002289static int filename_lookup(int dfd, struct filename *name, unsigned flags,
Al Viro9ad1aaa2015-05-12 16:44:39 -04002290 struct path *path, struct path *root)
Jeff Layton873f1ee2012-10-10 15:25:29 -04002291{
Al Viro894bc8c2015-05-02 07:16:16 -04002292 int retval;
Al Viro9883d182015-05-13 07:28:08 -04002293 struct nameidata nd;
Al Viroabc9f5b2015-05-12 16:53:42 -04002294 if (IS_ERR(name))
2295 return PTR_ERR(name);
Al Viro9ad1aaa2015-05-12 16:44:39 -04002296 if (unlikely(root)) {
2297 nd.root = *root;
2298 flags |= LOOKUP_ROOT;
2299 }
Al Viro9883d182015-05-13 07:28:08 -04002300 set_nameidata(&nd, dfd, name);
Al Viroc8a53ee2015-05-12 18:43:07 -04002301 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002302 if (unlikely(retval == -ECHILD))
Al Viroc8a53ee2015-05-12 18:43:07 -04002303 retval = path_lookupat(&nd, flags, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002304 if (unlikely(retval == -ESTALE))
Al Viroc8a53ee2015-05-12 18:43:07 -04002305 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002306
2307 if (likely(!retval))
Al Viro625b6d12015-05-12 16:36:12 -04002308 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
Al Viro9883d182015-05-13 07:28:08 -04002309 restore_nameidata();
Al Viroe4bd1c12015-05-12 16:40:39 -04002310 putname(name);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002311 return retval;
2312}
2313
Al Viro8bcb77f2015-05-08 16:59:20 -04002314/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroc8a53ee2015-05-12 18:43:07 -04002315static int path_parentat(struct nameidata *nd, unsigned flags,
Al Viro391172c2015-05-09 11:19:16 -04002316 struct path *parent)
Al Viro8bcb77f2015-05-08 16:59:20 -04002317{
Al Viroc8a53ee2015-05-12 18:43:07 -04002318 const char *s = path_init(nd, flags);
Al Viro368ee9b2015-05-08 17:19:59 -04002319 int err;
2320 if (IS_ERR(s))
2321 return PTR_ERR(s);
2322 err = link_path_walk(s, nd);
Al Viro8bcb77f2015-05-08 16:59:20 -04002323 if (!err)
2324 err = complete_walk(nd);
Al Viro391172c2015-05-09 11:19:16 -04002325 if (!err) {
2326 *parent = nd->path;
2327 nd->path.mnt = NULL;
2328 nd->path.dentry = NULL;
2329 }
2330 terminate_walk(nd);
Al Viro8bcb77f2015-05-08 16:59:20 -04002331 return err;
2332}
2333
Al Viro5c31b6c2015-05-12 17:32:54 -04002334static struct filename *filename_parentat(int dfd, struct filename *name,
Al Viro391172c2015-05-09 11:19:16 -04002335 unsigned int flags, struct path *parent,
2336 struct qstr *last, int *type)
Al Viro8bcb77f2015-05-08 16:59:20 -04002337{
2338 int retval;
Al Viro9883d182015-05-13 07:28:08 -04002339 struct nameidata nd;
Al Viro8bcb77f2015-05-08 16:59:20 -04002340
Al Viro5c31b6c2015-05-12 17:32:54 -04002341 if (IS_ERR(name))
2342 return name;
Al Viro9883d182015-05-13 07:28:08 -04002343 set_nameidata(&nd, dfd, name);
Al Viroc8a53ee2015-05-12 18:43:07 -04002344 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
Al Viro8bcb77f2015-05-08 16:59:20 -04002345 if (unlikely(retval == -ECHILD))
Al Viroc8a53ee2015-05-12 18:43:07 -04002346 retval = path_parentat(&nd, flags, parent);
Al Viro8bcb77f2015-05-08 16:59:20 -04002347 if (unlikely(retval == -ESTALE))
Al Viroc8a53ee2015-05-12 18:43:07 -04002348 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
Al Viro391172c2015-05-09 11:19:16 -04002349 if (likely(!retval)) {
2350 *last = nd.last;
2351 *type = nd.last_type;
2352 audit_inode(name, parent->dentry, LOOKUP_PARENT);
Al Viro5c31b6c2015-05-12 17:32:54 -04002353 } else {
2354 putname(name);
2355 name = ERR_PTR(retval);
Al Viro391172c2015-05-09 11:19:16 -04002356 }
Al Viro9883d182015-05-13 07:28:08 -04002357 restore_nameidata();
Al Viro5c31b6c2015-05-12 17:32:54 -04002358 return name;
Al Viro8bcb77f2015-05-08 16:59:20 -04002359}
2360
Al Viro79714f72012-06-15 03:01:42 +04002361/* does lookup, returns the object with parent locked */
2362struct dentry *kern_path_locked(const char *name, struct path *path)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002363{
Al Viro5c31b6c2015-05-12 17:32:54 -04002364 struct filename *filename;
2365 struct dentry *d;
Al Viro391172c2015-05-09 11:19:16 -04002366 struct qstr last;
2367 int type;
Paul Moore51689102015-01-22 00:00:03 -05002368
Al Viro5c31b6c2015-05-12 17:32:54 -04002369 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2370 &last, &type);
Paul Moore51689102015-01-22 00:00:03 -05002371 if (IS_ERR(filename))
2372 return ERR_CAST(filename);
Al Viro5c31b6c2015-05-12 17:32:54 -04002373 if (unlikely(type != LAST_NORM)) {
Al Viro391172c2015-05-09 11:19:16 -04002374 path_put(path);
Al Viro5c31b6c2015-05-12 17:32:54 -04002375 putname(filename);
2376 return ERR_PTR(-EINVAL);
Al Viro79714f72012-06-15 03:01:42 +04002377 }
Al Viro59551022016-01-22 15:40:57 -05002378 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
Al Viro391172c2015-05-09 11:19:16 -04002379 d = __lookup_hash(&last, path->dentry, 0);
Al Viro79714f72012-06-15 03:01:42 +04002380 if (IS_ERR(d)) {
Al Viro59551022016-01-22 15:40:57 -05002381 inode_unlock(path->dentry->d_inode);
Al Viro391172c2015-05-09 11:19:16 -04002382 path_put(path);
Al Viro79714f72012-06-15 03:01:42 +04002383 }
Paul Moore51689102015-01-22 00:00:03 -05002384 putname(filename);
Al Viro79714f72012-06-15 03:01:42 +04002385 return d;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002386}
2387
Al Virod1811462008-08-02 00:49:18 -04002388int kern_path(const char *name, unsigned int flags, struct path *path)
2389{
Al Viroabc9f5b2015-05-12 16:53:42 -04002390 return filename_lookup(AT_FDCWD, getname_kernel(name),
2391 flags, path, NULL);
Al Virod1811462008-08-02 00:49:18 -04002392}
Al Viro4d359502014-03-14 12:20:17 -04002393EXPORT_SYMBOL(kern_path);
Al Virod1811462008-08-02 00:49:18 -04002394
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002395/**
2396 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2397 * @dentry: pointer to dentry of the base directory
2398 * @mnt: pointer to vfs mount of the base directory
2399 * @name: pointer to file name
2400 * @flags: lookup flags
Al Viroe0a01242011-06-27 17:00:37 -04002401 * @path: pointer to struct path to fill
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002402 */
2403int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2404 const char *name, unsigned int flags,
Al Viroe0a01242011-06-27 17:00:37 -04002405 struct path *path)
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002406{
Al Viro9ad1aaa2015-05-12 16:44:39 -04002407 struct path root = {.mnt = mnt, .dentry = dentry};
Al Viro9ad1aaa2015-05-12 16:44:39 -04002408 /* the first argument of filename_lookup() is ignored with root */
Al Viroabc9f5b2015-05-12 16:53:42 -04002409 return filename_lookup(AT_FDCWD, getname_kernel(name),
2410 flags , path, &root);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002411}
Al Viro4d359502014-03-14 12:20:17 -04002412EXPORT_SYMBOL(vfs_path_lookup);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002413
Christoph Hellwigeead1912007-10-16 23:25:38 -07002414/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07002415 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07002416 * @name: pathname component to lookup
2417 * @base: base directory to lookup from
2418 * @len: maximum length @len should be interpreted to
2419 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07002420 * Note that this routine is purely a helper for filesystem usage and should
Al Viro9e7543e2015-02-23 02:49:48 -05002421 * not be called by generic code.
NeilBrownbbddca82016-01-07 16:08:20 -05002422 *
2423 * The caller must hold base->i_mutex.
Christoph Hellwigeead1912007-10-16 23:25:38 -07002424 */
James Morris057f6c02007-04-26 00:12:05 -07002425struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2426{
James Morris057f6c02007-04-26 00:12:05 -07002427 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05002428 unsigned int c;
Miklos Szeredicda309d2012-03-26 12:54:21 +02002429 int err;
James Morris057f6c02007-04-26 00:12:05 -07002430
Al Viro59551022016-01-22 15:40:57 -05002431 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
David Woodhouse2f9092e2009-04-20 23:18:37 +01002432
Al Viro6a96ba52011-03-07 23:49:20 -05002433 this.name = name;
2434 this.len = len;
Linus Torvalds8387ff22016-06-10 07:51:30 -07002435 this.hash = full_name_hash(base, name, len);
Al Viro6a96ba52011-03-07 23:49:20 -05002436 if (!len)
2437 return ERR_PTR(-EACCES);
2438
Al Viro21d8a152012-11-29 22:17:21 -05002439 if (unlikely(name[0] == '.')) {
2440 if (len < 2 || (len == 2 && name[1] == '.'))
2441 return ERR_PTR(-EACCES);
2442 }
2443
Al Viro6a96ba52011-03-07 23:49:20 -05002444 while (len--) {
2445 c = *(const unsigned char *)name++;
2446 if (c == '/' || c == '\0')
2447 return ERR_PTR(-EACCES);
Al Viro6a96ba52011-03-07 23:49:20 -05002448 }
Al Viro5a202bc2011-03-08 14:17:44 -05002449 /*
2450 * See if the low-level filesystem might want
2451 * to use its own hash..
2452 */
2453 if (base->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002454 int err = base->d_op->d_hash(base, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05002455 if (err < 0)
2456 return ERR_PTR(err);
2457 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07002458
Miklos Szeredicda309d2012-03-26 12:54:21 +02002459 err = inode_permission(base->d_inode, MAY_EXEC);
2460 if (err)
2461 return ERR_PTR(err);
2462
Al Viro72bd8662012-06-10 17:17:17 -04002463 return __lookup_hash(&this, base, 0);
James Morris057f6c02007-04-26 00:12:05 -07002464}
Al Viro4d359502014-03-14 12:20:17 -04002465EXPORT_SYMBOL(lookup_one_len);
James Morris057f6c02007-04-26 00:12:05 -07002466
NeilBrownbbddca82016-01-07 16:08:20 -05002467/**
2468 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2469 * @name: pathname component to lookup
2470 * @base: base directory to lookup from
2471 * @len: maximum length @len should be interpreted to
2472 *
2473 * Note that this routine is purely a helper for filesystem usage and should
2474 * not be called by generic code.
2475 *
2476 * Unlike lookup_one_len, it should be called without the parent
2477 * i_mutex held, and will take the i_mutex itself if necessary.
2478 */
2479struct dentry *lookup_one_len_unlocked(const char *name,
2480 struct dentry *base, int len)
2481{
2482 struct qstr this;
2483 unsigned int c;
2484 int err;
Linus Torvalds20d00ee2016-07-29 12:17:52 -07002485 struct dentry *ret;
NeilBrownbbddca82016-01-07 16:08:20 -05002486
2487 this.name = name;
2488 this.len = len;
Linus Torvalds8387ff22016-06-10 07:51:30 -07002489 this.hash = full_name_hash(base, name, len);
NeilBrownbbddca82016-01-07 16:08:20 -05002490 if (!len)
2491 return ERR_PTR(-EACCES);
2492
2493 if (unlikely(name[0] == '.')) {
2494 if (len < 2 || (len == 2 && name[1] == '.'))
2495 return ERR_PTR(-EACCES);
2496 }
2497
2498 while (len--) {
2499 c = *(const unsigned char *)name++;
2500 if (c == '/' || c == '\0')
2501 return ERR_PTR(-EACCES);
2502 }
2503 /*
2504 * See if the low-level filesystem might want
2505 * to use its own hash..
2506 */
2507 if (base->d_flags & DCACHE_OP_HASH) {
2508 int err = base->d_op->d_hash(base, &this);
2509 if (err < 0)
2510 return ERR_PTR(err);
2511 }
2512
2513 err = inode_permission(base->d_inode, MAY_EXEC);
2514 if (err)
2515 return ERR_PTR(err);
2516
Linus Torvalds20d00ee2016-07-29 12:17:52 -07002517 ret = lookup_dcache(&this, base, 0);
2518 if (!ret)
2519 ret = lookup_slow(&this, base, 0);
2520 return ret;
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.
Eric W. Biederman0bd23d092016-06-29 14:54:46 -05002760 * 7. If the victim has an unknown uid or gid we can't change the inode.
2761 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2762 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2763 * 10. We can't remove a root or mountpoint.
2764 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
Linus Torvalds1da177e2005-04-16 15:20:36 -07002765 * nfs_async_unlink().
2766 */
David Howellsb18825a2013-09-12 19:22:53 +01002767static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002768{
David Howells63afdfc2015-05-06 15:59:00 +01002769 struct inode *inode = d_backing_inode(victim);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770 int error;
2771
David Howellsb18825a2013-09-12 19:22:53 +01002772 if (d_is_negative(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773 return -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01002774 BUG_ON(!inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002775
2776 BUG_ON(victim->d_parent->d_inode != dir);
Jeff Layton4fa6b5e2012-10-10 15:25:25 -04002777 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002778
Al Virof419a2e2008-07-22 00:07:17 -04002779 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780 if (error)
2781 return error;
2782 if (IS_APPEND(dir))
2783 return -EPERM;
David Howellsb18825a2013-09-12 19:22:53 +01002784
2785 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
Eric W. Biederman0bd23d092016-06-29 14:54:46 -05002786 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002787 return -EPERM;
2788 if (isdir) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002789 if (!d_is_dir(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002790 return -ENOTDIR;
2791 if (IS_ROOT(victim))
2792 return -EBUSY;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002793 } else if (d_is_dir(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794 return -EISDIR;
2795 if (IS_DEADDIR(dir))
2796 return -ENOENT;
2797 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2798 return -EBUSY;
2799 return 0;
2800}
2801
2802/* Check whether we can create an object with dentry child in directory
2803 * dir.
2804 * 1. We can't do it if child already exists (open has special treatment for
2805 * this case, but since we are inlined it's OK)
2806 * 2. We can't do it if dir is read-only (done in permission())
Eric W. Biederman036d5232016-07-01 12:52:06 -05002807 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2808 * 4. We should have write and exec permissions on dir
2809 * 5. We can't do it if dir is immutable (done in permission())
Linus Torvalds1da177e2005-04-16 15:20:36 -07002810 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02002811static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812{
Eric W. Biederman036d5232016-07-01 12:52:06 -05002813 struct user_namespace *s_user_ns;
Jeff Layton14e972b2013-05-08 10:25:58 -04002814 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815 if (child->d_inode)
2816 return -EEXIST;
2817 if (IS_DEADDIR(dir))
2818 return -ENOENT;
Eric W. Biederman036d5232016-07-01 12:52:06 -05002819 s_user_ns = dir->i_sb->s_user_ns;
2820 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2821 !kgid_has_mapping(s_user_ns, current_fsgid()))
2822 return -EOVERFLOW;
Al Virof419a2e2008-07-22 00:07:17 -04002823 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002824}
2825
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826/*
2827 * p1 and p2 should be directories on the same fs.
2828 */
2829struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2830{
2831 struct dentry *p;
2832
2833 if (p1 == p2) {
Al Viro59551022016-01-22 15:40:57 -05002834 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002835 return NULL;
2836 }
2837
Al Virofc640052016-04-10 01:33:30 -04002838 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002840 p = d_ancestor(p2, p1);
2841 if (p) {
Al Viro59551022016-01-22 15:40:57 -05002842 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2843 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002844 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845 }
2846
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002847 p = d_ancestor(p1, p2);
2848 if (p) {
Al Viro59551022016-01-22 15:40:57 -05002849 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2850 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002851 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852 }
2853
Al Viro59551022016-01-22 15:40:57 -05002854 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2855 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002856 return NULL;
2857}
Al Viro4d359502014-03-14 12:20:17 -04002858EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002859
2860void unlock_rename(struct dentry *p1, struct dentry *p2)
2861{
Al Viro59551022016-01-22 15:40:57 -05002862 inode_unlock(p1->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863 if (p1 != p2) {
Al Viro59551022016-01-22 15:40:57 -05002864 inode_unlock(p2->d_inode);
Al Virofc640052016-04-10 01:33:30 -04002865 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002866 }
2867}
Al Viro4d359502014-03-14 12:20:17 -04002868EXPORT_SYMBOL(unlock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002869
Al Viro4acdaf22011-07-26 01:42:34 -04002870int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viro312b63f2012-06-10 18:09:36 -04002871 bool want_excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002873 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002874 if (error)
2875 return error;
2876
Al Viroacfa4382008-12-04 10:06:33 -05002877 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878 return -EACCES; /* shouldn't it be ENOSYS? */
2879 mode &= S_IALLUGO;
2880 mode |= S_IFREG;
2881 error = security_inode_create(dir, dentry, mode);
2882 if (error)
2883 return error;
Al Viro312b63f2012-06-10 18:09:36 -04002884 error = dir->i_op->create(dir, dentry, mode, want_excl);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002885 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002886 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002887 return error;
2888}
Al Viro4d359502014-03-14 12:20:17 -04002889EXPORT_SYMBOL(vfs_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002890
Eric W. Biedermana2982cc2016-06-09 15:34:02 -05002891bool may_open_dev(const struct path *path)
2892{
2893 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2894 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2895}
2896
Al Viro73d049a2011-03-11 12:08:24 -05002897static int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002898{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002899 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002900 struct inode *inode = dentry->d_inode;
2901 int error;
2902
2903 if (!inode)
2904 return -ENOENT;
2905
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002906 switch (inode->i_mode & S_IFMT) {
2907 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002908 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002909 case S_IFDIR:
2910 if (acc_mode & MAY_WRITE)
2911 return -EISDIR;
2912 break;
2913 case S_IFBLK:
2914 case S_IFCHR:
Eric W. Biedermana2982cc2016-06-09 15:34:02 -05002915 if (!may_open_dev(path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002916 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002917 /*FALLTHRU*/
2918 case S_IFIFO:
2919 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002920 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002921 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002922 }
Dave Hansenb41572e2007-10-16 23:31:14 -07002923
Al Viro62fb4a12015-12-26 22:33:24 -05002924 error = inode_permission(inode, MAY_OPEN | acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07002925 if (error)
2926 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05002927
Linus Torvalds1da177e2005-04-16 15:20:36 -07002928 /*
2929 * An append-only file must be opened in append mode for writing.
2930 */
2931 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05002932 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05002933 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002934 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05002935 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002936 }
2937
2938 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002939 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05002940 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002941
J. Bruce Fieldsf3c7691e2011-09-21 10:58:13 -04002942 return 0;
Al Viro7715b522009-12-16 03:54:00 -05002943}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002944
Jeff Laytone1181ee2010-12-07 16:19:50 -05002945static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002946{
Jeff Laytone1181ee2010-12-07 16:19:50 -05002947 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002948 struct inode *inode = path->dentry->d_inode;
2949 int error = get_write_access(inode);
2950 if (error)
2951 return error;
2952 /*
2953 * Refuse to truncate files with mandatory locks held on them.
2954 */
Jeff Laytond7a06982014-03-10 09:54:15 -04002955 error = locks_verify_locked(filp);
Al Viro7715b522009-12-16 03:54:00 -05002956 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002957 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002958 if (!error) {
2959 error = do_truncate(path->dentry, 0,
2960 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002961 filp);
Al Viro7715b522009-12-16 03:54:00 -05002962 }
2963 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002964 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002965}
2966
Dave Hansend57999e2008-02-15 14:37:27 -08002967static inline int open_to_namei_flags(int flag)
2968{
Al Viro8a5e9292011-06-25 19:15:54 -04002969 if ((flag & O_ACCMODE) == 3)
2970 flag--;
Dave Hansend57999e2008-02-15 14:37:27 -08002971 return flag;
2972}
2973
Al Virod3607752016-03-25 15:21:09 -04002974static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002975{
Seth Forshee2e727b82017-10-07 22:36:55 +00002976 struct user_namespace *s_user_ns;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002977 int error = security_path_mknod(dir, dentry, mode, 0);
2978 if (error)
2979 return error;
2980
Seth Forshee2e727b82017-10-07 22:36:55 +00002981 s_user_ns = dir->dentry->d_sb->s_user_ns;
2982 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2983 !kgid_has_mapping(s_user_ns, current_fsgid()))
2984 return -EOVERFLOW;
2985
Miklos Szeredid18e9002012-06-05 15:10:17 +02002986 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2987 if (error)
2988 return error;
2989
2990 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2991}
2992
David Howells1acf0af2012-06-14 16:13:46 +01002993/*
2994 * Attempt to atomically look up, create and open a file from a negative
2995 * dentry.
2996 *
2997 * Returns 0 if successful. The file will have been created and attached to
2998 * @file by the filesystem calling finish_open().
2999 *
3000 * Returns 1 if the file was looked up only or didn't need creating. The
3001 * caller will need to perform the open themselves. @path will have been
3002 * updated to point to the new dentry. This may be negative.
3003 *
3004 * Returns an error code otherwise.
3005 */
Al Viro2675a4e2012-06-22 12:41:10 +04003006static int atomic_open(struct nameidata *nd, struct dentry *dentry,
3007 struct path *path, struct file *file,
3008 const struct open_flags *op,
Al Viro1643b432016-04-27 19:14:10 -04003009 int open_flag, umode_t mode,
Al Viro2675a4e2012-06-22 12:41:10 +04003010 int *opened)
Miklos Szeredid18e9002012-06-05 15:10:17 +02003011{
Miklos Szeredid18e9002012-06-05 15:10:17 +02003012 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003013 struct inode *dir = nd->path.dentry->d_inode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003014 int error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003015
Al Viro384f26e2016-04-28 02:03:55 -04003016 if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */
Miklos Szeredid18e9002012-06-05 15:10:17 +02003017 open_flag &= ~O_TRUNC;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003018
Miklos Szeredid18e9002012-06-05 15:10:17 +02003019 if (nd->flags & LOOKUP_DIRECTORY)
3020 open_flag |= O_DIRECTORY;
3021
Al Viro30d90492012-06-22 12:40:19 +04003022 file->f_path.dentry = DENTRY_NOT_SET;
3023 file->f_path.mnt = nd->path.mnt;
Al Viro0fb1ea02016-04-27 14:13:10 -04003024 error = dir->i_op->atomic_open(dir, dentry, file,
3025 open_to_namei_flags(open_flag),
3026 mode, opened);
Al Viro6fbd0712016-04-28 11:50:59 -04003027 d_lookup_done(dentry);
Al Viro384f26e2016-04-28 02:03:55 -04003028 if (!error) {
3029 /*
3030 * We didn't have the inode before the open, so check open
3031 * permission here.
3032 */
3033 int acc_mode = op->acc_mode;
3034 if (*opened & FILE_CREATED) {
3035 WARN_ON(!(open_flag & O_CREAT));
3036 fsnotify_create(dir, dentry);
3037 acc_mode = 0;
3038 }
3039 error = may_open(&file->f_path, acc_mode, open_flag);
3040 if (WARN_ON(error > 0))
3041 error = -EINVAL;
3042 } else if (error > 0) {
Al Viro30d90492012-06-22 12:40:19 +04003043 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
Al Viro2675a4e2012-06-22 12:41:10 +04003044 error = -EIO;
Al Viro03da6332013-09-16 19:22:33 -04003045 } else {
Al Viro384f26e2016-04-28 02:03:55 -04003046 if (file->f_path.dentry) {
3047 dput(dentry);
3048 dentry = file->f_path.dentry;
Al Viro03da6332013-09-16 19:22:33 -04003049 }
Al Viro384f26e2016-04-28 02:03:55 -04003050 if (*opened & FILE_CREATED)
3051 fsnotify_create(dir, dentry);
Al Viroa01e7182016-06-07 21:53:51 -04003052 if (unlikely(d_is_negative(dentry))) {
3053 error = -ENOENT;
3054 } else {
3055 path->dentry = dentry;
3056 path->mnt = nd->path.mnt;
3057 return 1;
3058 }
Sage Weil62b2ce92012-08-15 13:30:12 -07003059 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02003060 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02003061 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04003062 return error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003063}
3064
Nick Piggin31e6b012011-01-07 17:49:52 +11003065/*
David Howells1acf0af2012-06-14 16:13:46 +01003066 * Look up and maybe create and open the last component.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003067 *
3068 * Must be called with i_mutex held on parent.
3069 *
David Howells1acf0af2012-06-14 16:13:46 +01003070 * Returns 0 if the file was successfully atomically created (if necessary) and
3071 * opened. In this case the file will be returned attached to @file.
3072 *
3073 * Returns 1 if the file was not completely opened at this time, though lookups
3074 * and creations will have been performed and the dentry returned in @path will
3075 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
3076 * specified then a negative dentry may be returned.
3077 *
3078 * An error code is returned otherwise.
3079 *
3080 * FILE_CREATE will be set in @*opened if the dentry was created and will be
3081 * cleared otherwise prior to returning.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003082 */
Al Viro2675a4e2012-06-22 12:41:10 +04003083static int lookup_open(struct nameidata *nd, struct path *path,
3084 struct file *file,
3085 const struct open_flags *op,
Al Viro64894cf2012-07-31 00:53:35 +04003086 bool got_write, int *opened)
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003087{
3088 struct dentry *dir = nd->path.dentry;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02003089 struct inode *dir_inode = dir->d_inode;
Al Viro1643b432016-04-27 19:14:10 -04003090 int open_flag = op->open_flag;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003091 struct dentry *dentry;
Al Viro1643b432016-04-27 19:14:10 -04003092 int error, create_error = 0;
Al Viro1643b432016-04-27 19:14:10 -04003093 umode_t mode = op->mode;
Al Viro6fbd0712016-04-28 11:50:59 -04003094 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003095
Al Viroce8644f2016-04-26 14:17:56 -04003096 if (unlikely(IS_DEADDIR(dir_inode)))
3097 return -ENOENT;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003098
Al Viro47237682012-06-10 05:01:45 -04003099 *opened &= ~FILE_CREATED;
Al Viro6fbd0712016-04-28 11:50:59 -04003100 dentry = d_lookup(dir, &nd->last);
3101 for (;;) {
3102 if (!dentry) {
3103 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3104 if (IS_ERR(dentry))
3105 return PTR_ERR(dentry);
3106 }
3107 if (d_in_lookup(dentry))
3108 break;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003109
Al Viro6fbd0712016-04-28 11:50:59 -04003110 if (!(dentry->d_flags & DCACHE_OP_REVALIDATE))
3111 break;
3112
3113 error = d_revalidate(dentry, nd->flags);
3114 if (likely(error > 0))
3115 break;
3116 if (error)
3117 goto out_dput;
3118 d_invalidate(dentry);
3119 dput(dentry);
3120 dentry = NULL;
3121 }
3122 if (dentry->d_inode) {
Al Viro6c51e512016-03-05 20:09:32 -05003123 /* Cached positive dentry: will open in f_op->open */
Miklos Szeredid18e9002012-06-05 15:10:17 +02003124 goto out_no_open;
Al Viro6c51e512016-03-05 20:09:32 -05003125 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02003126
Al Viro1643b432016-04-27 19:14:10 -04003127 /*
3128 * Checking write permission is tricky, bacuse we don't know if we are
3129 * going to actually need it: O_CREAT opens should work as long as the
3130 * file exists. But checking existence breaks atomicity. The trick is
3131 * to check access and if not granted clear O_CREAT from the flags.
3132 *
3133 * Another problem is returing the "right" error value (e.g. for an
3134 * O_EXCL open we want to return EEXIST not EROFS).
3135 */
3136 if (open_flag & O_CREAT) {
3137 if (!IS_POSIXACL(dir->d_inode))
3138 mode &= ~current_umask();
3139 if (unlikely(!got_write)) {
3140 create_error = -EROFS;
3141 open_flag &= ~O_CREAT;
3142 if (open_flag & (O_EXCL | O_TRUNC))
3143 goto no_open;
3144 /* No side effects, safe to clear O_CREAT */
3145 } else {
3146 create_error = may_o_create(&nd->path, dentry, mode);
3147 if (create_error) {
3148 open_flag &= ~O_CREAT;
3149 if (open_flag & O_EXCL)
3150 goto no_open;
3151 }
3152 }
3153 } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3154 unlikely(!got_write)) {
3155 /*
3156 * No O_CREATE -> atomicity not a requirement -> fall
3157 * back to lookup + open
3158 */
3159 goto no_open;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003160 }
3161
Al Viro1643b432016-04-27 19:14:10 -04003162 if (dir_inode->i_op->atomic_open) {
3163 error = atomic_open(nd, dentry, path, file, op, open_flag,
3164 mode, opened);
3165 if (unlikely(error == -ENOENT) && create_error)
3166 error = create_error;
3167 return error;
3168 }
Miklos Szeredi54ef4872012-06-05 15:10:16 +02003169
Al Viro1643b432016-04-27 19:14:10 -04003170no_open:
Al Viro6fbd0712016-04-28 11:50:59 -04003171 if (d_in_lookup(dentry)) {
Al Viro12fa5e22016-04-28 11:19:43 -04003172 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3173 nd->flags);
Al Viro6fbd0712016-04-28 11:50:59 -04003174 d_lookup_done(dentry);
Al Viro12fa5e22016-04-28 11:19:43 -04003175 if (unlikely(res)) {
3176 if (IS_ERR(res)) {
3177 error = PTR_ERR(res);
3178 goto out_dput;
3179 }
3180 dput(dentry);
3181 dentry = res;
3182 }
Miklos Szeredi54ef4872012-06-05 15:10:16 +02003183 }
3184
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003185 /* Negative dentry, just create the file */
Al Viro1643b432016-04-27 19:14:10 -04003186 if (!dentry->d_inode && (open_flag & O_CREAT)) {
Al Viro47237682012-06-10 05:01:45 -04003187 *opened |= FILE_CREATED;
Al Viroce8644f2016-04-26 14:17:56 -04003188 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
Al Viroce8644f2016-04-26 14:17:56 -04003189 if (!dir_inode->i_op->create) {
3190 error = -EACCES;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003191 goto out_dput;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003192 }
Al Viroce8644f2016-04-26 14:17:56 -04003193 error = dir_inode->i_op->create(dir_inode, dentry, mode,
Al Viro1643b432016-04-27 19:14:10 -04003194 open_flag & O_EXCL);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003195 if (error)
3196 goto out_dput;
Al Viroce8644f2016-04-26 14:17:56 -04003197 fsnotify_create(dir_inode, dentry);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003198 }
Al Viro1643b432016-04-27 19:14:10 -04003199 if (unlikely(create_error) && !dentry->d_inode) {
3200 error = create_error;
3201 goto out_dput;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003202 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02003203out_no_open:
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003204 path->dentry = dentry;
3205 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04003206 return 1;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003207
3208out_dput:
3209 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04003210 return error;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003211}
3212
3213/*
Al Virofe2d35f2011-03-05 22:58:25 -05003214 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11003215 */
Al Viro896475d2015-04-22 18:02:17 -04003216static int do_last(struct nameidata *nd,
Al Viro2675a4e2012-06-22 12:41:10 +04003217 struct file *file, const struct open_flags *op,
Al Viro76ae2a52015-05-12 18:44:32 -04003218 int *opened)
Al Virofb1cc552009-12-24 01:58:28 -05003219{
Al Viroa1e28032009-12-24 02:12:06 -05003220 struct dentry *dir = nd->path.dentry;
Al Viroca344a892011-03-09 00:36:45 -05003221 int open_flag = op->open_flag;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003222 bool will_truncate = (open_flag & O_TRUNC) != 0;
Al Viro64894cf2012-07-31 00:53:35 +04003223 bool got_write = false;
Al Virobcda7652011-03-13 16:42:14 -04003224 int acc_mode = op->acc_mode;
Al Viro254cf582015-05-05 09:40:46 -04003225 unsigned seq;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02003226 struct inode *inode;
Al Viro896475d2015-04-22 18:02:17 -04003227 struct path path;
Al Viro16c2cd72011-02-22 15:50:10 -05003228 int error;
Al Virofb1cc552009-12-24 01:58:28 -05003229
Al Viroc3e380b2011-02-23 13:39:45 -05003230 nd->flags &= ~LOOKUP_PARENT;
3231 nd->flags |= op->intent;
3232
Al Virobc77daa2013-06-06 09:12:33 -04003233 if (nd->last_type != LAST_NORM) {
Al Virofe2d35f2011-03-05 22:58:25 -05003234 error = handle_dots(nd, nd->last_type);
Al Virodeb106c2015-05-08 18:05:21 -04003235 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04003236 return error;
Miklos Szeredie83db162012-06-05 15:10:29 +02003237 goto finish_open;
Al Viro1f36f772009-12-26 10:56:19 -05003238 }
Al Viro67ee3ad2009-12-26 07:01:01 -05003239
Al Viroca344a892011-03-09 00:36:45 -05003240 if (!(open_flag & O_CREAT)) {
Al Virofe2d35f2011-03-05 22:58:25 -05003241 if (nd->last.name[nd->last.len])
3242 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3243 /* we _can_ be in RCU mode here */
Al Viro254cf582015-05-05 09:40:46 -04003244 error = lookup_fast(nd, &path, &inode, &seq);
Al Viroe9742b52016-03-05 22:04:59 -05003245 if (likely(error > 0))
Miklos Szeredi71574862012-06-05 15:10:14 +02003246 goto finish_lookup;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02003247
Miklos Szeredi71574862012-06-05 15:10:14 +02003248 if (error < 0)
Al Virodeb106c2015-05-08 18:05:21 -04003249 return error;
Miklos Szeredi37d7fff2012-06-05 15:10:12 +02003250
Miklos Szeredi71574862012-06-05 15:10:14 +02003251 BUG_ON(nd->inode != dir->d_inode);
Al Viro6583fe22016-03-05 18:14:03 -05003252 BUG_ON(nd->flags & LOOKUP_RCU);
Miklos Szeredib6183df2012-06-05 15:10:13 +02003253 } else {
3254 /* create side of things */
3255 /*
3256 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3257 * has been cleared when we got to the last component we are
3258 * about to look up
3259 */
3260 error = complete_walk(nd);
Al Viroe8bb73d2015-05-08 16:28:42 -04003261 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003262 return error;
Miklos Szeredib6183df2012-06-05 15:10:13 +02003263
Al Viro76ae2a52015-05-12 18:44:32 -04003264 audit_inode(nd->name, dir, LOOKUP_PARENT);
Miklos Szeredib6183df2012-06-05 15:10:13 +02003265 /* trailing slashes? */
Al Virodeb106c2015-05-08 18:05:21 -04003266 if (unlikely(nd->last.name[nd->last.len]))
3267 return -EISDIR;
Al Virofe2d35f2011-03-05 22:58:25 -05003268 }
3269
Al Viro9cf843e2016-04-28 19:35:16 -04003270 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
Al Viro64894cf2012-07-31 00:53:35 +04003271 error = mnt_want_write(nd->path.mnt);
3272 if (!error)
3273 got_write = true;
3274 /*
3275 * do _not_ fail yet - we might not need that or fail with
3276 * a different error; let lookup_open() decide; we'll be
3277 * dropping this one anyway.
3278 */
3279 }
Al Viro9cf843e2016-04-28 19:35:16 -04003280 if (open_flag & O_CREAT)
3281 inode_lock(dir->d_inode);
3282 else
3283 inode_lock_shared(dir->d_inode);
Al Viro896475d2015-04-22 18:02:17 -04003284 error = lookup_open(nd, &path, file, op, got_write, opened);
Al Viro9cf843e2016-04-28 19:35:16 -04003285 if (open_flag & O_CREAT)
3286 inode_unlock(dir->d_inode);
3287 else
3288 inode_unlock_shared(dir->d_inode);
Al Viroa1e28032009-12-24 02:12:06 -05003289
Al Viro2675a4e2012-06-22 12:41:10 +04003290 if (error <= 0) {
3291 if (error)
Miklos Szeredid18e9002012-06-05 15:10:17 +02003292 goto out;
3293
Al Viro47237682012-06-10 05:01:45 -04003294 if ((*opened & FILE_CREATED) ||
Al Viro496ad9a2013-01-23 17:07:38 -05003295 !S_ISREG(file_inode(file)->i_mode))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003296 will_truncate = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003297
Al Viro76ae2a52015-05-12 18:44:32 -04003298 audit_inode(nd->name, file->f_path.dentry, 0);
Miklos Szeredid18e9002012-06-05 15:10:17 +02003299 goto opened;
3300 }
Al Virofb1cc552009-12-24 01:58:28 -05003301
Al Viro47237682012-06-10 05:01:45 -04003302 if (*opened & FILE_CREATED) {
Al Viro9b44f1b2011-03-09 00:17:27 -05003303 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05003304 open_flag &= ~O_TRUNC;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003305 will_truncate = false;
Al Viro62fb4a12015-12-26 22:33:24 -05003306 acc_mode = 0;
Al Viro896475d2015-04-22 18:02:17 -04003307 path_to_nameidata(&path, nd);
Miklos Szeredie83db162012-06-05 15:10:29 +02003308 goto finish_open_created;
Al Virofb1cc552009-12-24 01:58:28 -05003309 }
3310
3311 /*
Miklos Szeredid18e9002012-06-05 15:10:17 +02003312 * If atomic_open() acquired write access it is dropped now due to
3313 * possible mount and symlink following (this might be optimized away if
3314 * necessary...)
3315 */
Al Viro64894cf2012-07-31 00:53:35 +04003316 if (got_write) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02003317 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04003318 got_write = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003319 }
3320
Al Viroe6ec03a2016-06-05 00:23:09 -04003321 error = follow_managed(&path, nd);
3322 if (unlikely(error < 0))
3323 return error;
3324
Al Viro6583fe22016-03-05 18:14:03 -05003325 if (unlikely(d_is_negative(path.dentry))) {
3326 path_to_nameidata(&path, nd);
3327 return -ENOENT;
3328 }
3329
3330 /*
3331 * create/update audit record if it already exists.
3332 */
3333 audit_inode(nd->name, path.dentry, 0);
3334
Al Virodeb106c2015-05-08 18:05:21 -04003335 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3336 path_to_nameidata(&path, nd);
3337 return -EEXIST;
3338 }
Al Virofb1cc552009-12-24 01:58:28 -05003339
Al Viro254cf582015-05-05 09:40:46 -04003340 seq = 0; /* out of RCU mode, so the value doesn't matter */
Al Virod4565642016-02-27 19:23:16 -05003341 inode = d_backing_inode(path.dentry);
Al Viro766c4cb2015-05-07 19:24:57 -04003342finish_lookup:
Al Virod63ff282015-05-04 18:13:23 -04003343 if (nd->depth)
3344 put_link(nd);
Al Viro181548c2015-05-07 19:54:34 -04003345 error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW,
3346 inode, seq);
Al Virodeb106c2015-05-08 18:05:21 -04003347 if (unlikely(error))
Al Virod63ff282015-05-04 18:13:23 -04003348 return error;
Al Virofb1cc552009-12-24 01:58:28 -05003349
Al Virofac7d192016-06-04 11:41:49 -04003350 path_to_nameidata(&path, nd);
Miklos Szeredidecf3402012-05-21 17:30:08 +02003351 nd->inode = inode;
Al Viro254cf582015-05-05 09:40:46 -04003352 nd->seq = seq;
Al Viroa3fbbde2011-11-07 21:21:26 +00003353 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
Al Virobc77daa2013-06-06 09:12:33 -04003354finish_open:
Al Viroa3fbbde2011-11-07 21:21:26 +00003355 error = complete_walk(nd);
Al Virofac7d192016-06-04 11:41:49 -04003356 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003357 return error;
Al Viro76ae2a52015-05-12 18:44:32 -04003358 audit_inode(nd->name, nd->path.dentry, 0);
Al Virofb1cc552009-12-24 01:58:28 -05003359 error = -EISDIR;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003360 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04003361 goto out;
Miklos Szerediaf2f5542012-05-21 17:30:11 +02003362 error = -ENOTDIR;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003363 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04003364 goto out;
David Howells4bbcbd32015-03-17 22:16:40 +00003365 if (!d_is_reg(nd->path.dentry))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003366 will_truncate = false;
Al Viro6c0d46c2011-03-09 00:59:59 -05003367
Al Viro0f9d1a12011-03-09 00:13:14 -05003368 if (will_truncate) {
3369 error = mnt_want_write(nd->path.mnt);
3370 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003371 goto out;
Al Viro64894cf2012-07-31 00:53:35 +04003372 got_write = true;
Al Viro0f9d1a12011-03-09 00:13:14 -05003373 }
Miklos Szeredie83db162012-06-05 15:10:29 +02003374finish_open_created:
Al Viro6ac08702016-04-26 00:02:50 -04003375 error = may_open(&nd->path, acc_mode, open_flag);
3376 if (error)
3377 goto out;
Miklos Szeredi4aa7c632014-10-24 00:14:35 +02003378 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3379 error = vfs_open(&nd->path, file, current_cred());
Al Virofac7d192016-06-04 11:41:49 -04003380 if (error)
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003381 goto out;
Al Virofac7d192016-06-04 11:41:49 -04003382 *opened |= FILE_OPENED;
Miklos Szeredia8277b92012-06-05 15:10:32 +02003383opened:
Al Viro2675a4e2012-06-22 12:41:10 +04003384 error = open_check_o_direct(file);
Al Virofe9ec822016-04-27 03:14:20 -04003385 if (!error)
3386 error = ima_file_check(file, op->acc_mode, *opened);
3387 if (!error && will_truncate)
Al Viro2675a4e2012-06-22 12:41:10 +04003388 error = handle_truncate(file);
Al Viroca344a892011-03-09 00:36:45 -05003389out:
Al Virofe9ec822016-04-27 03:14:20 -04003390 if (unlikely(error) && (*opened & FILE_OPENED))
3391 fput(file);
Al Viroc80567c2016-02-27 19:17:33 -05003392 if (unlikely(error > 0)) {
3393 WARN_ON(1);
3394 error = -EINVAL;
3395 }
Al Viro64894cf2012-07-31 00:53:35 +04003396 if (got_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05003397 mnt_drop_write(nd->path.mnt);
Al Viro2675a4e2012-06-22 12:41:10 +04003398 return error;
Al Virofb1cc552009-12-24 01:58:28 -05003399}
3400
Al Viroc8a53ee2015-05-12 18:43:07 -04003401static int do_tmpfile(struct nameidata *nd, unsigned flags,
Al Viro60545d02013-06-07 01:20:27 -04003402 const struct open_flags *op,
3403 struct file *file, int *opened)
3404{
3405 static const struct qstr name = QSTR_INIT("/", 1);
Al Viro625b6d12015-05-12 16:36:12 -04003406 struct dentry *child;
Al Viro60545d02013-06-07 01:20:27 -04003407 struct inode *dir;
Al Viro625b6d12015-05-12 16:36:12 -04003408 struct path path;
Al Viroc8a53ee2015-05-12 18:43:07 -04003409 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
Al Viro60545d02013-06-07 01:20:27 -04003410 if (unlikely(error))
3411 return error;
Al Viro625b6d12015-05-12 16:36:12 -04003412 error = mnt_want_write(path.mnt);
Al Viro60545d02013-06-07 01:20:27 -04003413 if (unlikely(error))
3414 goto out;
Al Viro625b6d12015-05-12 16:36:12 -04003415 dir = path.dentry->d_inode;
Al Viro60545d02013-06-07 01:20:27 -04003416 /* we want directory to be writable */
Al Viro625b6d12015-05-12 16:36:12 -04003417 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Al Viro60545d02013-06-07 01:20:27 -04003418 if (error)
3419 goto out2;
Al Viro60545d02013-06-07 01:20:27 -04003420 if (!dir->i_op->tmpfile) {
3421 error = -EOPNOTSUPP;
3422 goto out2;
3423 }
Al Viro625b6d12015-05-12 16:36:12 -04003424 child = d_alloc(path.dentry, &name);
Al Viro60545d02013-06-07 01:20:27 -04003425 if (unlikely(!child)) {
3426 error = -ENOMEM;
3427 goto out2;
3428 }
Al Viro625b6d12015-05-12 16:36:12 -04003429 dput(path.dentry);
3430 path.dentry = child;
3431 error = dir->i_op->tmpfile(dir, child, op->mode);
Al Viro60545d02013-06-07 01:20:27 -04003432 if (error)
3433 goto out2;
Al Viroc8a53ee2015-05-12 18:43:07 -04003434 audit_inode(nd->name, child, 0);
Eric Rannaud69a91c22014-10-30 01:51:01 -07003435 /* Don't check for other permissions, the inode was just created */
Al Viro62fb4a12015-12-26 22:33:24 -05003436 error = may_open(&path, 0, op->open_flag);
Al Viro60545d02013-06-07 01:20:27 -04003437 if (error)
3438 goto out2;
Al Viro625b6d12015-05-12 16:36:12 -04003439 file->f_path.mnt = path.mnt;
3440 error = finish_open(file, child, NULL, opened);
Al Viro60545d02013-06-07 01:20:27 -04003441 if (error)
3442 goto out2;
3443 error = open_check_o_direct(file);
Al Virof4e0c302013-06-11 08:34:36 +04003444 if (error) {
Al Viro60545d02013-06-07 01:20:27 -04003445 fput(file);
Al Virof4e0c302013-06-11 08:34:36 +04003446 } else if (!(op->open_flag & O_EXCL)) {
3447 struct inode *inode = file_inode(file);
3448 spin_lock(&inode->i_lock);
3449 inode->i_state |= I_LINKABLE;
3450 spin_unlock(&inode->i_lock);
3451 }
Al Viro60545d02013-06-07 01:20:27 -04003452out2:
Al Viro625b6d12015-05-12 16:36:12 -04003453 mnt_drop_write(path.mnt);
Al Viro60545d02013-06-07 01:20:27 -04003454out:
Al Viro625b6d12015-05-12 16:36:12 -04003455 path_put(&path);
Al Viro60545d02013-06-07 01:20:27 -04003456 return error;
3457}
3458
Al Viro6ac08702016-04-26 00:02:50 -04003459static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3460{
3461 struct path path;
3462 int error = path_lookupat(nd, flags, &path);
3463 if (!error) {
3464 audit_inode(nd->name, path.dentry, 0);
3465 error = vfs_open(&path, file, current_cred());
3466 path_put(&path);
3467 }
3468 return error;
3469}
3470
Al Viroc8a53ee2015-05-12 18:43:07 -04003471static struct file *path_openat(struct nameidata *nd,
3472 const struct open_flags *op, unsigned flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003473{
Al Viro368ee9b2015-05-08 17:19:59 -04003474 const char *s;
Al Viro30d90492012-06-22 12:40:19 +04003475 struct file *file;
Al Viro47237682012-06-10 05:01:45 -04003476 int opened = 0;
Al Viro13aab422011-02-23 17:54:08 -05003477 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11003478
Al Viro30d90492012-06-22 12:40:19 +04003479 file = get_empty_filp();
Al Viro1afc99b2013-02-14 20:41:04 -05003480 if (IS_ERR(file))
3481 return file;
Nick Piggin31e6b012011-01-07 17:49:52 +11003482
Al Viro30d90492012-06-22 12:40:19 +04003483 file->f_flags = op->open_flag;
Nick Piggin31e6b012011-01-07 17:49:52 +11003484
Al Virobb458c62013-07-13 13:26:37 +04003485 if (unlikely(file->f_flags & __O_TMPFILE)) {
Al Viroc8a53ee2015-05-12 18:43:07 -04003486 error = do_tmpfile(nd, flags, op, file, &opened);
Al Virof15133d2015-05-08 22:53:15 -04003487 goto out2;
Al Viro60545d02013-06-07 01:20:27 -04003488 }
3489
Al Viro6ac08702016-04-26 00:02:50 -04003490 if (unlikely(file->f_flags & O_PATH)) {
3491 error = do_o_path(nd, flags, file);
3492 if (!error)
3493 opened |= FILE_OPENED;
3494 goto out2;
3495 }
3496
Al Viroc8a53ee2015-05-12 18:43:07 -04003497 s = path_init(nd, flags);
Al Viro368ee9b2015-05-08 17:19:59 -04003498 if (IS_ERR(s)) {
3499 put_filp(file);
3500 return ERR_CAST(s);
3501 }
Al Viro3bdba282015-05-08 17:37:07 -04003502 while (!(error = link_path_walk(s, nd)) &&
Al Viro76ae2a52015-05-12 18:44:32 -04003503 (error = do_last(nd, file, op, &opened)) > 0) {
Al Viro73d049a2011-03-11 12:08:24 -05003504 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
Al Viro3bdba282015-05-08 17:37:07 -04003505 s = trailing_symlink(nd);
3506 if (IS_ERR(s)) {
3507 error = PTR_ERR(s);
Al Viro2675a4e2012-06-22 12:41:10 +04003508 break;
Al Viro3bdba282015-05-08 17:37:07 -04003509 }
Al Viro806b6812009-12-26 07:16:40 -05003510 }
Al Virodeb106c2015-05-08 18:05:21 -04003511 terminate_walk(nd);
Al Virof15133d2015-05-08 22:53:15 -04003512out2:
Al Viro2675a4e2012-06-22 12:41:10 +04003513 if (!(opened & FILE_OPENED)) {
3514 BUG_ON(!error);
Al Viro30d90492012-06-22 12:40:19 +04003515 put_filp(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003516 }
Al Viro2675a4e2012-06-22 12:41:10 +04003517 if (unlikely(error)) {
3518 if (error == -EOPENSTALE) {
3519 if (flags & LOOKUP_RCU)
3520 error = -ECHILD;
3521 else
3522 error = -ESTALE;
3523 }
3524 file = ERR_PTR(error);
3525 }
3526 return file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003527}
3528
Jeff Layton669abf42012-10-10 16:43:10 -04003529struct file *do_filp_open(int dfd, struct filename *pathname,
Al Virof9652e12013-06-11 08:23:01 +04003530 const struct open_flags *op)
Al Viro13aab422011-02-23 17:54:08 -05003531{
Al Viro9883d182015-05-13 07:28:08 -04003532 struct nameidata nd;
Al Virof9652e12013-06-11 08:23:01 +04003533 int flags = op->lookup_flags;
Al Viro13aab422011-02-23 17:54:08 -05003534 struct file *filp;
3535
Al Viro9883d182015-05-13 07:28:08 -04003536 set_nameidata(&nd, dfd, pathname);
Al Viroc8a53ee2015-05-12 18:43:07 -04003537 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
Al Viro13aab422011-02-23 17:54:08 -05003538 if (unlikely(filp == ERR_PTR(-ECHILD)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003539 filp = path_openat(&nd, op, flags);
Al Viro13aab422011-02-23 17:54:08 -05003540 if (unlikely(filp == ERR_PTR(-ESTALE)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003541 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
Al Viro9883d182015-05-13 07:28:08 -04003542 restore_nameidata();
Al Viro13aab422011-02-23 17:54:08 -05003543 return filp;
3544}
3545
Al Viro73d049a2011-03-11 12:08:24 -05003546struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
Al Virof9652e12013-06-11 08:23:01 +04003547 const char *name, const struct open_flags *op)
Al Viro73d049a2011-03-11 12:08:24 -05003548{
Al Viro9883d182015-05-13 07:28:08 -04003549 struct nameidata nd;
Al Viro73d049a2011-03-11 12:08:24 -05003550 struct file *file;
Paul Moore51689102015-01-22 00:00:03 -05003551 struct filename *filename;
Al Virof9652e12013-06-11 08:23:01 +04003552 int flags = op->lookup_flags | LOOKUP_ROOT;
Al Viro73d049a2011-03-11 12:08:24 -05003553
3554 nd.root.mnt = mnt;
3555 nd.root.dentry = dentry;
3556
David Howellsb18825a2013-09-12 19:22:53 +01003557 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
Al Viro73d049a2011-03-11 12:08:24 -05003558 return ERR_PTR(-ELOOP);
3559
Paul Moore51689102015-01-22 00:00:03 -05003560 filename = getname_kernel(name);
Viresh Kumara1c83682015-08-12 15:59:44 +05303561 if (IS_ERR(filename))
Paul Moore51689102015-01-22 00:00:03 -05003562 return ERR_CAST(filename);
3563
Al Viro9883d182015-05-13 07:28:08 -04003564 set_nameidata(&nd, -1, filename);
Al Viroc8a53ee2015-05-12 18:43:07 -04003565 file = path_openat(&nd, op, flags | LOOKUP_RCU);
Al Viro73d049a2011-03-11 12:08:24 -05003566 if (unlikely(file == ERR_PTR(-ECHILD)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003567 file = path_openat(&nd, op, flags);
Al Viro73d049a2011-03-11 12:08:24 -05003568 if (unlikely(file == ERR_PTR(-ESTALE)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003569 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
Al Viro9883d182015-05-13 07:28:08 -04003570 restore_nameidata();
Paul Moore51689102015-01-22 00:00:03 -05003571 putname(filename);
Al Viro73d049a2011-03-11 12:08:24 -05003572 return file;
3573}
3574
Al Virofa14a0b2015-01-22 02:16:49 -05003575static struct dentry *filename_create(int dfd, struct filename *name,
Jeff Layton1ac12b42012-12-11 12:10:06 -05003576 struct path *path, unsigned int lookup_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003577{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003578 struct dentry *dentry = ERR_PTR(-EEXIST);
Al Viro391172c2015-05-09 11:19:16 -04003579 struct qstr last;
3580 int type;
Jan Karac30dabf2012-06-12 16:20:30 +02003581 int err2;
Jeff Layton1ac12b42012-12-11 12:10:06 -05003582 int error;
3583 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3584
3585 /*
3586 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3587 * other flags passed in are ignored!
3588 */
3589 lookup_flags &= LOOKUP_REVAL;
3590
Al Viro5c31b6c2015-05-12 17:32:54 -04003591 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3592 if (IS_ERR(name))
3593 return ERR_CAST(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003594
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003595 /*
3596 * Yucky last component or no last component at all?
3597 * (foo/., foo/.., /////)
3598 */
Al Viro5c31b6c2015-05-12 17:32:54 -04003599 if (unlikely(type != LAST_NORM))
Al Viroed75e952011-06-27 16:53:43 -04003600 goto out;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003601
Jan Karac30dabf2012-06-12 16:20:30 +02003602 /* don't fail immediately if it's r/o, at least try to report other errors */
Al Viro391172c2015-05-09 11:19:16 -04003603 err2 = mnt_want_write(path->mnt);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003604 /*
3605 * Do the final lookup.
3606 */
Al Viro391172c2015-05-09 11:19:16 -04003607 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
Al Viro59551022016-01-22 15:40:57 -05003608 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
Al Viro391172c2015-05-09 11:19:16 -04003609 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003610 if (IS_ERR(dentry))
Al Viroa8104a92012-07-20 02:25:00 +04003611 goto unlock;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003612
Al Viroa8104a92012-07-20 02:25:00 +04003613 error = -EEXIST;
David Howellsb18825a2013-09-12 19:22:53 +01003614 if (d_is_positive(dentry))
Al Viroa8104a92012-07-20 02:25:00 +04003615 goto fail;
David Howellsb18825a2013-09-12 19:22:53 +01003616
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003617 /*
3618 * Special case - lookup gave negative, but... we had foo/bar/
3619 * From the vfs_mknod() POV we just have a negative dentry -
3620 * all is fine. Let's be bastards - you had / on the end, you've
3621 * been asking for (non-existent) directory. -ENOENT for you.
3622 */
Al Viro391172c2015-05-09 11:19:16 -04003623 if (unlikely(!is_dir && last.name[last.len])) {
Al Viroa8104a92012-07-20 02:25:00 +04003624 error = -ENOENT;
Al Viroed75e952011-06-27 16:53:43 -04003625 goto fail;
Al Viroe9baf6e2008-05-15 04:49:12 -04003626 }
Jan Karac30dabf2012-06-12 16:20:30 +02003627 if (unlikely(err2)) {
3628 error = err2;
Al Viroa8104a92012-07-20 02:25:00 +04003629 goto fail;
Jan Karac30dabf2012-06-12 16:20:30 +02003630 }
Al Viro181c37b2015-05-12 17:21:25 -04003631 putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003632 return dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003633fail:
Al Viroa8104a92012-07-20 02:25:00 +04003634 dput(dentry);
3635 dentry = ERR_PTR(error);
3636unlock:
Al Viro59551022016-01-22 15:40:57 -05003637 inode_unlock(path->dentry->d_inode);
Jan Karac30dabf2012-06-12 16:20:30 +02003638 if (!err2)
Al Viro391172c2015-05-09 11:19:16 -04003639 mnt_drop_write(path->mnt);
Al Viroed75e952011-06-27 16:53:43 -04003640out:
Al Viro391172c2015-05-09 11:19:16 -04003641 path_put(path);
Al Viro181c37b2015-05-12 17:21:25 -04003642 putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003643 return dentry;
3644}
Al Virofa14a0b2015-01-22 02:16:49 -05003645
3646struct dentry *kern_path_create(int dfd, const char *pathname,
3647 struct path *path, unsigned int lookup_flags)
3648{
Al Viro181c37b2015-05-12 17:21:25 -04003649 return filename_create(dfd, getname_kernel(pathname),
3650 path, lookup_flags);
Al Virofa14a0b2015-01-22 02:16:49 -05003651}
Al Virodae6ad82011-06-26 11:50:15 -04003652EXPORT_SYMBOL(kern_path_create);
3653
Al Viro921a1652012-07-20 01:15:31 +04003654void done_path_create(struct path *path, struct dentry *dentry)
3655{
3656 dput(dentry);
Al Viro59551022016-01-22 15:40:57 -05003657 inode_unlock(path->dentry->d_inode);
Al Viroa8104a92012-07-20 02:25:00 +04003658 mnt_drop_write(path->mnt);
Al Viro921a1652012-07-20 01:15:31 +04003659 path_put(path);
3660}
3661EXPORT_SYMBOL(done_path_create);
3662
Al Viro520ae682015-05-13 07:00:28 -04003663inline struct dentry *user_path_create(int dfd, const char __user *pathname,
Jeff Layton1ac12b42012-12-11 12:10:06 -05003664 struct path *path, unsigned int lookup_flags)
Al Virodae6ad82011-06-26 11:50:15 -04003665{
Al Viro181c37b2015-05-12 17:21:25 -04003666 return filename_create(dfd, getname(pathname), path, lookup_flags);
Al Virodae6ad82011-06-26 11:50:15 -04003667}
3668EXPORT_SYMBOL(user_path_create);
3669
Al Viro1a67aaf2011-07-26 01:52:52 -04003670int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003671{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003672 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003673
3674 if (error)
3675 return error;
3676
Eric W. Biederman975d6b32011-11-13 12:16:43 -08003677 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003678 return -EPERM;
3679
Al Viroacfa4382008-12-04 10:06:33 -05003680 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003681 return -EPERM;
3682
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07003683 error = devcgroup_inode_mknod(mode, dev);
3684 if (error)
3685 return error;
3686
Linus Torvalds1da177e2005-04-16 15:20:36 -07003687 error = security_inode_mknod(dir, dentry, mode, dev);
3688 if (error)
3689 return error;
3690
Linus Torvalds1da177e2005-04-16 15:20:36 -07003691 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003692 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003693 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003694 return error;
3695}
Al Viro4d359502014-03-14 12:20:17 -04003696EXPORT_SYMBOL(vfs_mknod);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003697
Al Virof69aac02011-07-26 04:31:40 -04003698static int may_mknod(umode_t mode)
Dave Hansen463c3192008-02-15 14:37:57 -08003699{
3700 switch (mode & S_IFMT) {
3701 case S_IFREG:
3702 case S_IFCHR:
3703 case S_IFBLK:
3704 case S_IFIFO:
3705 case S_IFSOCK:
3706 case 0: /* zero mode translates to S_IFREG */
3707 return 0;
3708 case S_IFDIR:
3709 return -EPERM;
3710 default:
3711 return -EINVAL;
3712 }
3713}
3714
Al Viro8208a222011-07-25 17:32:17 -04003715SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003716 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003717{
Al Viro2ad94ae2008-07-21 09:32:51 -04003718 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003719 struct path path;
3720 int error;
Jeff Layton972567f2012-12-20 16:00:10 -05003721 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003722
Al Viro8e4bfca2012-07-20 01:17:26 +04003723 error = may_mknod(mode);
3724 if (error)
3725 return error;
Jeff Layton972567f2012-12-20 16:00:10 -05003726retry:
3727 dentry = user_path_create(dfd, filename, &path, lookup_flags);
Al Virodae6ad82011-06-26 11:50:15 -04003728 if (IS_ERR(dentry))
3729 return PTR_ERR(dentry);
Al Viro2ad94ae2008-07-21 09:32:51 -04003730
Al Virodae6ad82011-06-26 11:50:15 -04003731 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003732 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003733 error = security_path_mknod(&path, dentry, mode, dev);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003734 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04003735 goto out;
Dave Hansen463c3192008-02-15 14:37:57 -08003736 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003737 case 0: case S_IFREG:
Al Viro312b63f2012-06-10 18:09:36 -04003738 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
Mimi Zohar05d1a712016-02-29 19:52:05 -05003739 if (!error)
3740 ima_post_path_mknod(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003741 break;
3742 case S_IFCHR: case S_IFBLK:
Al Virodae6ad82011-06-26 11:50:15 -04003743 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003744 new_decode_dev(dev));
3745 break;
3746 case S_IFIFO: case S_IFSOCK:
Al Virodae6ad82011-06-26 11:50:15 -04003747 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003748 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003749 }
Al Viroa8104a92012-07-20 02:25:00 +04003750out:
Al Viro921a1652012-07-20 01:15:31 +04003751 done_path_create(&path, dentry);
Jeff Layton972567f2012-12-20 16:00:10 -05003752 if (retry_estale(error, lookup_flags)) {
3753 lookup_flags |= LOOKUP_REVAL;
3754 goto retry;
3755 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003756 return error;
3757}
3758
Al Viro8208a222011-07-25 17:32:17 -04003759SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003760{
3761 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3762}
3763
Al Viro18bb1db2011-07-26 01:41:39 -04003764int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003765{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003766 int error = may_create(dir, dentry);
Al Viro8de52772012-02-06 12:45:27 -05003767 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003768
3769 if (error)
3770 return error;
3771
Al Viroacfa4382008-12-04 10:06:33 -05003772 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003773 return -EPERM;
3774
3775 mode &= (S_IRWXUGO|S_ISVTX);
3776 error = security_inode_mkdir(dir, dentry, mode);
3777 if (error)
3778 return error;
3779
Al Viro8de52772012-02-06 12:45:27 -05003780 if (max_links && dir->i_nlink >= max_links)
3781 return -EMLINK;
3782
Linus Torvalds1da177e2005-04-16 15:20:36 -07003783 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003784 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003785 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003786 return error;
3787}
Al Viro4d359502014-03-14 12:20:17 -04003788EXPORT_SYMBOL(vfs_mkdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003789
Al Viroa218d0f2011-11-21 14:59:34 -05003790SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003791{
Dave Hansen6902d922006-09-30 23:29:01 -07003792 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003793 struct path path;
3794 int error;
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003795 unsigned int lookup_flags = LOOKUP_DIRECTORY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003796
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003797retry:
3798 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07003799 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003800 return PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003801
Al Virodae6ad82011-06-26 11:50:15 -04003802 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003803 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003804 error = security_path_mkdir(&path, dentry, mode);
Al Viroa8104a92012-07-20 02:25:00 +04003805 if (!error)
3806 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
Al Viro921a1652012-07-20 01:15:31 +04003807 done_path_create(&path, dentry);
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003808 if (retry_estale(error, lookup_flags)) {
3809 lookup_flags |= LOOKUP_REVAL;
3810 goto retry;
3811 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003812 return error;
3813}
3814
Al Viroa218d0f2011-11-21 14:59:34 -05003815SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003816{
3817 return sys_mkdirat(AT_FDCWD, pathname, mode);
3818}
3819
Linus Torvalds1da177e2005-04-16 15:20:36 -07003820int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3821{
3822 int error = may_delete(dir, dentry, 1);
3823
3824 if (error)
3825 return error;
3826
Al Viroacfa4382008-12-04 10:06:33 -05003827 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003828 return -EPERM;
3829
Al Viro1d2ef592011-09-14 18:55:41 +01003830 dget(dentry);
Al Viro59551022016-01-22 15:40:57 -05003831 inode_lock(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003832
Sage Weil912dbc12011-05-24 13:06:11 -07003833 error = -EBUSY;
Eric W. Biederman7af13642013-10-04 19:15:13 -07003834 if (is_local_mountpoint(dentry))
Sage Weil912dbc12011-05-24 13:06:11 -07003835 goto out;
3836
3837 error = security_inode_rmdir(dir, dentry);
3838 if (error)
3839 goto out;
3840
Sage Weil3cebde22011-05-29 21:20:59 -07003841 shrink_dcache_parent(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003842 error = dir->i_op->rmdir(dir, dentry);
3843 if (error)
3844 goto out;
3845
3846 dentry->d_inode->i_flags |= S_DEAD;
3847 dont_mount(dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003848 detach_mounts(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003849
3850out:
Al Viro59551022016-01-22 15:40:57 -05003851 inode_unlock(dentry->d_inode);
Al Viro1d2ef592011-09-14 18:55:41 +01003852 dput(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003853 if (!error)
3854 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003855 return error;
3856}
Al Viro4d359502014-03-14 12:20:17 -04003857EXPORT_SYMBOL(vfs_rmdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003858
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003859static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003860{
3861 int error = 0;
Jeff Layton91a27b22012-10-10 15:25:28 -04003862 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003863 struct dentry *dentry;
Al Virof5beed72015-04-30 16:09:11 -04003864 struct path path;
3865 struct qstr last;
3866 int type;
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003867 unsigned int lookup_flags = 0;
3868retry:
Al Virof5beed72015-04-30 16:09:11 -04003869 name = user_path_parent(dfd, pathname,
3870 &path, &last, &type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04003871 if (IS_ERR(name))
3872 return PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003873
Al Virof5beed72015-04-30 16:09:11 -04003874 switch (type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003875 case LAST_DOTDOT:
3876 error = -ENOTEMPTY;
3877 goto exit1;
3878 case LAST_DOT:
3879 error = -EINVAL;
3880 goto exit1;
3881 case LAST_ROOT:
3882 error = -EBUSY;
3883 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003884 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003885
Al Virof5beed72015-04-30 16:09:11 -04003886 error = mnt_want_write(path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02003887 if (error)
3888 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003889
Al Viro59551022016-01-22 15:40:57 -05003890 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
Al Virof5beed72015-04-30 16:09:11 -04003891 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003892 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003893 if (IS_ERR(dentry))
3894 goto exit2;
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003895 if (!dentry->d_inode) {
3896 error = -ENOENT;
3897 goto exit3;
3898 }
Al Virof5beed72015-04-30 16:09:11 -04003899 error = security_path_rmdir(&path, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003900 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003901 goto exit3;
Al Virof5beed72015-04-30 16:09:11 -04003902 error = vfs_rmdir(path.dentry->d_inode, dentry);
Dave Hansen06227532008-02-15 14:37:34 -08003903exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07003904 dput(dentry);
3905exit2:
Al Viro59551022016-01-22 15:40:57 -05003906 inode_unlock(path.dentry->d_inode);
Al Virof5beed72015-04-30 16:09:11 -04003907 mnt_drop_write(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003908exit1:
Al Virof5beed72015-04-30 16:09:11 -04003909 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003910 putname(name);
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003911 if (retry_estale(error, lookup_flags)) {
3912 lookup_flags |= LOOKUP_REVAL;
3913 goto retry;
3914 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003915 return error;
3916}
3917
Heiko Carstens3cdad422009-01-14 14:14:22 +01003918SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003919{
3920 return do_rmdir(AT_FDCWD, pathname);
3921}
3922
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003923/**
3924 * vfs_unlink - unlink a filesystem object
3925 * @dir: parent directory
3926 * @dentry: victim
3927 * @delegated_inode: returns victim inode, if the inode is delegated.
3928 *
3929 * The caller must hold dir->i_mutex.
3930 *
3931 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3932 * return a reference to the inode in delegated_inode. The caller
3933 * should then break the delegation on that inode and retry. Because
3934 * breaking a delegation may take a long time, the caller should drop
3935 * dir->i_mutex before doing so.
3936 *
3937 * Alternatively, a caller may pass NULL for delegated_inode. This may
3938 * be appropriate for callers that expect the underlying filesystem not
3939 * to be NFS exported.
3940 */
3941int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003942{
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003943 struct inode *target = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003944 int error = may_delete(dir, dentry, 0);
3945
3946 if (error)
3947 return error;
3948
Al Viroacfa4382008-12-04 10:06:33 -05003949 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003950 return -EPERM;
3951
Al Viro59551022016-01-22 15:40:57 -05003952 inode_lock(target);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003953 if (is_local_mountpoint(dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003954 error = -EBUSY;
3955 else {
3956 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05003957 if (!error) {
J. Bruce Fields5a146962012-08-28 07:50:40 -07003958 error = try_break_deleg(target, delegated_inode);
3959 if (error)
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003960 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003961 error = dir->i_op->unlink(dir, dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003962 if (!error) {
Al Virod83c49f2010-04-30 17:17:09 -04003963 dont_mount(dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003964 detach_mounts(dentry);
3965 }
Al Virobec10522010-03-03 14:12:08 -05003966 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003967 }
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003968out:
Al Viro59551022016-01-22 15:40:57 -05003969 inode_unlock(target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003970
3971 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3972 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003973 fsnotify_link_count(target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003974 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003975 }
Robert Love0eeca282005-07-12 17:06:03 -04003976
Linus Torvalds1da177e2005-04-16 15:20:36 -07003977 return error;
3978}
Al Viro4d359502014-03-14 12:20:17 -04003979EXPORT_SYMBOL(vfs_unlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003980
3981/*
3982 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003983 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07003984 * writeout happening, and we don't want to prevent access to the directory
3985 * while waiting on the I/O.
3986 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003987static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003988{
Al Viro2ad94ae2008-07-21 09:32:51 -04003989 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04003990 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003991 struct dentry *dentry;
Al Virof5beed72015-04-30 16:09:11 -04003992 struct path path;
3993 struct qstr last;
3994 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003995 struct inode *inode = NULL;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003996 struct inode *delegated_inode = NULL;
Jeff Layton5d18f812012-12-20 16:38:04 -05003997 unsigned int lookup_flags = 0;
3998retry:
Al Virof5beed72015-04-30 16:09:11 -04003999 name = user_path_parent(dfd, pathname,
4000 &path, &last, &type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04004001 if (IS_ERR(name))
4002 return PTR_ERR(name);
Al Viro2ad94ae2008-07-21 09:32:51 -04004003
Linus Torvalds1da177e2005-04-16 15:20:36 -07004004 error = -EISDIR;
Al Virof5beed72015-04-30 16:09:11 -04004005 if (type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004006 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09004007
Al Virof5beed72015-04-30 16:09:11 -04004008 error = mnt_want_write(path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02004009 if (error)
4010 goto exit1;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04004011retry_deleg:
Al Viro59551022016-01-22 15:40:57 -05004012 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
Al Virof5beed72015-04-30 16:09:11 -04004013 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004014 error = PTR_ERR(dentry);
4015 if (!IS_ERR(dentry)) {
4016 /* Why not before? Because we want correct error value */
Al Virof5beed72015-04-30 16:09:11 -04004017 if (last.name[last.len])
Török Edwin50338b82011-06-16 00:06:14 +03004018 goto slashes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004019 inode = dentry->d_inode;
David Howellsb18825a2013-09-12 19:22:53 +01004020 if (d_is_negative(dentry))
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04004021 goto slashes;
4022 ihold(inode);
Al Virof5beed72015-04-30 16:09:11 -04004023 error = security_path_unlink(&path, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004024 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02004025 goto exit2;
Al Virof5beed72015-04-30 16:09:11 -04004026 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
Jan Karac30dabf2012-06-12 16:20:30 +02004027exit2:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004028 dput(dentry);
4029 }
Al Viro59551022016-01-22 15:40:57 -05004030 inode_unlock(path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004031 if (inode)
4032 iput(inode); /* truncate the inode here */
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04004033 inode = NULL;
4034 if (delegated_inode) {
J. Bruce Fields5a146962012-08-28 07:50:40 -07004035 error = break_deleg_wait(&delegated_inode);
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04004036 if (!error)
4037 goto retry_deleg;
4038 }
Al Virof5beed72015-04-30 16:09:11 -04004039 mnt_drop_write(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004040exit1:
Al Virof5beed72015-04-30 16:09:11 -04004041 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004042 putname(name);
Jeff Layton5d18f812012-12-20 16:38:04 -05004043 if (retry_estale(error, lookup_flags)) {
4044 lookup_flags |= LOOKUP_REVAL;
4045 inode = NULL;
4046 goto retry;
4047 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004048 return error;
4049
4050slashes:
David Howellsb18825a2013-09-12 19:22:53 +01004051 if (d_is_negative(dentry))
4052 error = -ENOENT;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02004053 else if (d_is_dir(dentry))
David Howellsb18825a2013-09-12 19:22:53 +01004054 error = -EISDIR;
4055 else
4056 error = -ENOTDIR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004057 goto exit2;
4058}
4059
Heiko Carstens2e4d0922009-01-14 14:14:31 +01004060SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004061{
4062 if ((flag & ~AT_REMOVEDIR) != 0)
4063 return -EINVAL;
4064
4065 if (flag & AT_REMOVEDIR)
4066 return do_rmdir(dfd, pathname);
4067
4068 return do_unlinkat(dfd, pathname);
4069}
4070
Heiko Carstens3480b252009-01-14 14:14:16 +01004071SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004072{
4073 return do_unlinkat(AT_FDCWD, pathname);
4074}
4075
Miklos Szeredidb2e7472008-06-24 16:50:16 +02004076int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004077{
Miklos Szeredia95164d2008-07-30 15:08:48 +02004078 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004079
4080 if (error)
4081 return error;
4082
Al Viroacfa4382008-12-04 10:06:33 -05004083 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004084 return -EPERM;
4085
4086 error = security_inode_symlink(dir, dentry, oldname);
4087 if (error)
4088 return error;
4089
Linus Torvalds1da177e2005-04-16 15:20:36 -07004090 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07004091 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00004092 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004093 return error;
4094}
Al Viro4d359502014-03-14 12:20:17 -04004095EXPORT_SYMBOL(vfs_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004096
Heiko Carstens2e4d0922009-01-14 14:14:31 +01004097SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4098 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004099{
Al Viro2ad94ae2008-07-21 09:32:51 -04004100 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04004101 struct filename *from;
Dave Hansen6902d922006-09-30 23:29:01 -07004102 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04004103 struct path path;
Jeff Laytonf46d3562012-12-11 12:10:08 -05004104 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004105
4106 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04004107 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004108 return PTR_ERR(from);
Jeff Laytonf46d3562012-12-11 12:10:08 -05004109retry:
4110 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07004111 error = PTR_ERR(dentry);
4112 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04004113 goto out_putname;
Dave Hansen6902d922006-09-30 23:29:01 -07004114
Jeff Layton91a27b22012-10-10 15:25:28 -04004115 error = security_path_symlink(&path, dentry, from->name);
Al Viroa8104a92012-07-20 02:25:00 +04004116 if (!error)
Jeff Layton91a27b22012-10-10 15:25:28 -04004117 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
Al Viro921a1652012-07-20 01:15:31 +04004118 done_path_create(&path, dentry);
Jeff Laytonf46d3562012-12-11 12:10:08 -05004119 if (retry_estale(error, lookup_flags)) {
4120 lookup_flags |= LOOKUP_REVAL;
4121 goto retry;
4122 }
Dave Hansen6902d922006-09-30 23:29:01 -07004123out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004124 putname(from);
4125 return error;
4126}
4127
Heiko Carstens3480b252009-01-14 14:14:16 +01004128SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004129{
4130 return sys_symlinkat(oldname, AT_FDCWD, newname);
4131}
4132
J. Bruce Fields146a8592011-09-20 17:14:31 -04004133/**
4134 * vfs_link - create a new link
4135 * @old_dentry: object to be linked
4136 * @dir: new parent
4137 * @new_dentry: where to create the new link
4138 * @delegated_inode: returns inode needing a delegation break
4139 *
4140 * The caller must hold dir->i_mutex
4141 *
4142 * If vfs_link discovers a delegation on the to-be-linked file in need
4143 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4144 * inode in delegated_inode. The caller should then break the delegation
4145 * and retry. Because breaking a delegation may take a long time, the
4146 * caller should drop the i_mutex before doing so.
4147 *
4148 * Alternatively, a caller may pass NULL for delegated_inode. This may
4149 * be appropriate for callers that expect the underlying filesystem not
4150 * to be NFS exported.
4151 */
4152int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004153{
4154 struct inode *inode = old_dentry->d_inode;
Al Viro8de52772012-02-06 12:45:27 -05004155 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004156 int error;
4157
4158 if (!inode)
4159 return -ENOENT;
4160
Miklos Szeredia95164d2008-07-30 15:08:48 +02004161 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004162 if (error)
4163 return error;
4164
4165 if (dir->i_sb != inode->i_sb)
4166 return -EXDEV;
4167
4168 /*
4169 * A link to an append-only or immutable file cannot be created.
4170 */
4171 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4172 return -EPERM;
Eric W. Biederman0bd23d092016-06-29 14:54:46 -05004173 /*
4174 * Updating the link count will likely cause i_uid and i_gid to
4175 * be writen back improperly if their true value is unknown to
4176 * the vfs.
4177 */
4178 if (HAS_UNMAPPED_ID(inode))
4179 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05004180 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004181 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02004182 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004183 return -EPERM;
4184
4185 error = security_inode_link(old_dentry, dir, new_dentry);
4186 if (error)
4187 return error;
4188
Al Viro59551022016-01-22 15:40:57 -05004189 inode_lock(inode);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05304190 /* Make sure we don't allow creating hardlink to an unlinked file */
Al Virof4e0c302013-06-11 08:34:36 +04004191 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05304192 error = -ENOENT;
Al Viro8de52772012-02-06 12:45:27 -05004193 else if (max_links && inode->i_nlink >= max_links)
4194 error = -EMLINK;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004195 else {
4196 error = try_break_deleg(inode, delegated_inode);
4197 if (!error)
4198 error = dir->i_op->link(old_dentry, dir, new_dentry);
4199 }
Al Virof4e0c302013-06-11 08:34:36 +04004200
4201 if (!error && (inode->i_state & I_LINKABLE)) {
4202 spin_lock(&inode->i_lock);
4203 inode->i_state &= ~I_LINKABLE;
4204 spin_unlock(&inode->i_lock);
4205 }
Al Viro59551022016-01-22 15:40:57 -05004206 inode_unlock(inode);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07004207 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02004208 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004209 return error;
4210}
Al Viro4d359502014-03-14 12:20:17 -04004211EXPORT_SYMBOL(vfs_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004212
4213/*
4214 * Hardlinks are often used in delicate situations. We avoid
4215 * security-related surprises by not following symlinks on the
4216 * newname. --KAB
4217 *
4218 * We don't follow them on the oldname either to be compatible
4219 * with linux 2.0, and to avoid hard-linking to directories
4220 * and other special files. --ADM
4221 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01004222SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4223 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004224{
4225 struct dentry *new_dentry;
Al Virodae6ad82011-06-26 11:50:15 -04004226 struct path old_path, new_path;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004227 struct inode *delegated_inode = NULL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304228 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004229 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004230
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304231 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004232 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304233 /*
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004234 * To use null names we require CAP_DAC_READ_SEARCH
4235 * This ensures that not everyone will be able to create
4236 * handlink using the passed filedescriptor.
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304237 */
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004238 if (flags & AT_EMPTY_PATH) {
4239 if (!capable(CAP_DAC_READ_SEARCH))
4240 return -ENOENT;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304241 how = LOOKUP_EMPTY;
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004242 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004243
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304244 if (flags & AT_SYMLINK_FOLLOW)
4245 how |= LOOKUP_FOLLOW;
Jeff Layton442e31c2012-12-20 16:15:38 -05004246retry:
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304247 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004248 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04004249 return error;
4250
Jeff Layton442e31c2012-12-20 16:15:38 -05004251 new_dentry = user_path_create(newdfd, newname, &new_path,
4252 (how & LOOKUP_REVAL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004253 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07004254 if (IS_ERR(new_dentry))
Al Virodae6ad82011-06-26 11:50:15 -04004255 goto out;
4256
4257 error = -EXDEV;
4258 if (old_path.mnt != new_path.mnt)
4259 goto out_dput;
Kees Cook800179c2012-07-25 17:29:07 -07004260 error = may_linkat(&old_path);
4261 if (unlikely(error))
4262 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04004263 error = security_path_link(old_path.dentry, &new_path, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004264 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04004265 goto out_dput;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004266 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
Dave Hansen75c3f292008-02-15 14:37:45 -08004267out_dput:
Al Viro921a1652012-07-20 01:15:31 +04004268 done_path_create(&new_path, new_dentry);
J. Bruce Fields146a8592011-09-20 17:14:31 -04004269 if (delegated_inode) {
4270 error = break_deleg_wait(&delegated_inode);
Oleg Drokind22e6332014-01-31 15:41:58 -05004271 if (!error) {
4272 path_put(&old_path);
J. Bruce Fields146a8592011-09-20 17:14:31 -04004273 goto retry;
Oleg Drokind22e6332014-01-31 15:41:58 -05004274 }
J. Bruce Fields146a8592011-09-20 17:14:31 -04004275 }
Jeff Layton442e31c2012-12-20 16:15:38 -05004276 if (retry_estale(error, how)) {
Oleg Drokind22e6332014-01-31 15:41:58 -05004277 path_put(&old_path);
Jeff Layton442e31c2012-12-20 16:15:38 -05004278 how |= LOOKUP_REVAL;
4279 goto retry;
4280 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004281out:
Al Viro2d8f3032008-07-22 09:59:21 -04004282 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004283
4284 return error;
4285}
4286
Heiko Carstens3480b252009-01-14 14:14:16 +01004287SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004288{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004289 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004290}
4291
Miklos Szeredibc270272014-04-01 17:08:42 +02004292/**
4293 * vfs_rename - rename a filesystem object
4294 * @old_dir: parent of source
4295 * @old_dentry: source
4296 * @new_dir: parent of destination
4297 * @new_dentry: destination
4298 * @delegated_inode: returns an inode needing a delegation break
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004299 * @flags: rename flags
Miklos Szeredibc270272014-04-01 17:08:42 +02004300 *
4301 * The caller must hold multiple mutexes--see lock_rename()).
4302 *
4303 * If vfs_rename discovers a delegation in need of breaking at either
4304 * the source or destination, it will return -EWOULDBLOCK and return a
4305 * reference to the inode in delegated_inode. The caller should then
4306 * break the delegation and retry. Because breaking a delegation may
4307 * take a long time, the caller should drop all locks before doing
4308 * so.
4309 *
4310 * Alternatively, a caller may pass NULL for delegated_inode. This may
4311 * be appropriate for callers that expect the underlying filesystem not
4312 * to be NFS exported.
4313 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07004314 * The worst of all namespace operations - renaming directory. "Perverted"
4315 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4316 * Problems:
J. Bruce Fieldsd03b29a2014-02-17 16:52:33 -05004317 * a) we can get into loop creation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004318 * b) race potential - two innocent renames can create a loop together.
4319 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004320 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07004321 * story.
J. Bruce Fields6cedba82012-03-05 11:40:41 -05004322 * c) we have to lock _four_ objects - parents and victim (if it exists),
4323 * and source (if it is not a directory).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004324 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07004325 * whether the target exists). Solution: try to be smart with locking
4326 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004327 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07004328 * move will be locked. Thus we can rank directories by the tree
4329 * (ancestors first) and rank all non-directories after them.
4330 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004331 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004332 * HOWEVER, it relies on the assumption that any object with ->lookup()
4333 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4334 * we'd better make sure that there's no link(2) for them.
Sage Weile4eaac02011-05-24 13:06:07 -07004335 * d) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004336 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07004337 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05004338 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07004339 * locking].
4340 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004341int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004342 struct inode *new_dir, struct dentry *new_dentry,
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004343 struct inode **delegated_inode, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004344{
4345 int error;
Miklos Szeredibc270272014-04-01 17:08:42 +02004346 bool is_dir = d_is_dir(old_dentry);
Miklos Szeredibc270272014-04-01 17:08:42 +02004347 struct inode *source = old_dentry->d_inode;
4348 struct inode *target = new_dentry->d_inode;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004349 bool new_is_dir = false;
4350 unsigned max_links = new_dir->i_sb->s_max_links;
Al Viroad25f112017-07-07 14:51:19 -04004351 struct name_snapshot old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004352
Miklos Szeredi9409e222016-05-11 01:16:37 +02004353 /*
4354 * Check source == target.
4355 * On overlayfs need to look at underlying inodes.
4356 */
Miklos Szeredi2d902672016-06-30 08:53:27 +02004357 if (d_real_inode(old_dentry) == d_real_inode(new_dentry))
Miklos Szeredibc270272014-04-01 17:08:42 +02004358 return 0;
4359
Linus Torvalds1da177e2005-04-16 15:20:36 -07004360 error = may_delete(old_dir, old_dentry, is_dir);
4361 if (error)
4362 return error;
4363
Miklos Szeredida1ce062014-04-01 17:08:43 +02004364 if (!target) {
Miklos Szeredia95164d2008-07-30 15:08:48 +02004365 error = may_create(new_dir, new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004366 } else {
4367 new_is_dir = d_is_dir(new_dentry);
4368
4369 if (!(flags & RENAME_EXCHANGE))
4370 error = may_delete(new_dir, new_dentry, is_dir);
4371 else
4372 error = may_delete(new_dir, new_dentry, new_is_dir);
4373 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004374 if (error)
4375 return error;
4376
Miklos Szeredi2773bf02016-09-27 11:03:58 +02004377 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004378 return -EPERM;
4379
Miklos Szeredibc270272014-04-01 17:08:42 +02004380 /*
4381 * If we are going to change the parent - check write permissions,
4382 * we'll need to flip '..'.
4383 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02004384 if (new_dir != old_dir) {
4385 if (is_dir) {
4386 error = inode_permission(source, MAY_WRITE);
4387 if (error)
4388 return error;
4389 }
4390 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4391 error = inode_permission(target, MAY_WRITE);
4392 if (error)
4393 return error;
4394 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004395 }
Robert Love0eeca282005-07-12 17:06:03 -04004396
Miklos Szeredi0b3974e2014-04-01 17:08:43 +02004397 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4398 flags);
Miklos Szeredibc270272014-04-01 17:08:42 +02004399 if (error)
4400 return error;
4401
Al Viroad25f112017-07-07 14:51:19 -04004402 take_dentry_name_snapshot(&old_name, old_dentry);
Miklos Szeredibc270272014-04-01 17:08:42 +02004403 dget(new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004404 if (!is_dir || (flags & RENAME_EXCHANGE))
Miklos Szeredibc270272014-04-01 17:08:42 +02004405 lock_two_nondirectories(source, target);
4406 else if (target)
Al Viro59551022016-01-22 15:40:57 -05004407 inode_lock(target);
Miklos Szeredibc270272014-04-01 17:08:42 +02004408
4409 error = -EBUSY;
Eric W. Biederman7af13642013-10-04 19:15:13 -07004410 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
Miklos Szeredibc270272014-04-01 17:08:42 +02004411 goto out;
4412
Miklos Szeredida1ce062014-04-01 17:08:43 +02004413 if (max_links && new_dir != old_dir) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004414 error = -EMLINK;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004415 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
Miklos Szeredibc270272014-04-01 17:08:42 +02004416 goto out;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004417 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4418 old_dir->i_nlink >= max_links)
4419 goto out;
4420 }
4421 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4422 shrink_dcache_parent(new_dentry);
4423 if (!is_dir) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004424 error = try_break_deleg(source, delegated_inode);
4425 if (error)
4426 goto out;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004427 }
4428 if (target && !new_is_dir) {
4429 error = try_break_deleg(target, delegated_inode);
4430 if (error)
4431 goto out;
Miklos Szeredibc270272014-04-01 17:08:42 +02004432 }
Miklos Szeredi2773bf02016-09-27 11:03:58 +02004433 error = old_dir->i_op->rename(old_dir, old_dentry,
Miklos Szeredi18fc84d2016-09-27 11:03:58 +02004434 new_dir, new_dentry, flags);
Miklos Szeredibc270272014-04-01 17:08:42 +02004435 if (error)
4436 goto out;
4437
Miklos Szeredida1ce062014-04-01 17:08:43 +02004438 if (!(flags & RENAME_EXCHANGE) && target) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004439 if (is_dir)
4440 target->i_flags |= S_DEAD;
4441 dont_mount(new_dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07004442 detach_mounts(new_dentry);
Miklos Szeredibc270272014-04-01 17:08:42 +02004443 }
Miklos Szeredida1ce062014-04-01 17:08:43 +02004444 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4445 if (!(flags & RENAME_EXCHANGE))
4446 d_move(old_dentry, new_dentry);
4447 else
4448 d_exchange(old_dentry, new_dentry);
4449 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004450out:
Miklos Szeredida1ce062014-04-01 17:08:43 +02004451 if (!is_dir || (flags & RENAME_EXCHANGE))
Miklos Szeredibc270272014-04-01 17:08:42 +02004452 unlock_two_nondirectories(source, target);
4453 else if (target)
Al Viro59551022016-01-22 15:40:57 -05004454 inode_unlock(target);
Miklos Szeredibc270272014-04-01 17:08:42 +02004455 dput(new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004456 if (!error) {
Al Viroad25f112017-07-07 14:51:19 -04004457 fsnotify_move(old_dir, new_dir, old_name.name, is_dir,
Miklos Szeredida1ce062014-04-01 17:08:43 +02004458 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4459 if (flags & RENAME_EXCHANGE) {
4460 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4461 new_is_dir, NULL, new_dentry);
4462 }
4463 }
Al Viroad25f112017-07-07 14:51:19 -04004464 release_dentry_name_snapshot(&old_name);
Robert Love0eeca282005-07-12 17:06:03 -04004465
Linus Torvalds1da177e2005-04-16 15:20:36 -07004466 return error;
4467}
Al Viro4d359502014-03-14 12:20:17 -04004468EXPORT_SYMBOL(vfs_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004469
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004470SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4471 int, newdfd, const char __user *, newname, unsigned int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004472{
Al Viro2ad94ae2008-07-21 09:32:51 -04004473 struct dentry *old_dentry, *new_dentry;
4474 struct dentry *trap;
Al Virof5beed72015-04-30 16:09:11 -04004475 struct path old_path, new_path;
4476 struct qstr old_last, new_last;
4477 int old_type, new_type;
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004478 struct inode *delegated_inode = NULL;
Jeff Layton91a27b22012-10-10 15:25:28 -04004479 struct filename *from;
4480 struct filename *to;
Al Virof5beed72015-04-30 16:09:11 -04004481 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
Jeff Laytonc6a94282012-12-11 12:10:10 -05004482 bool should_retry = false;
Al Viro2ad94ae2008-07-21 09:32:51 -04004483 int error;
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004484
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004485 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredida1ce062014-04-01 17:08:43 +02004486 return -EINVAL;
4487
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004488 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4489 (flags & RENAME_EXCHANGE))
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004490 return -EINVAL;
4491
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004492 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4493 return -EPERM;
4494
Al Virof5beed72015-04-30 16:09:11 -04004495 if (flags & RENAME_EXCHANGE)
4496 target_flags = 0;
4497
Jeff Laytonc6a94282012-12-11 12:10:10 -05004498retry:
Al Virof5beed72015-04-30 16:09:11 -04004499 from = user_path_parent(olddfd, oldname,
4500 &old_path, &old_last, &old_type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04004501 if (IS_ERR(from)) {
4502 error = PTR_ERR(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004503 goto exit;
Jeff Layton91a27b22012-10-10 15:25:28 -04004504 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004505
Al Virof5beed72015-04-30 16:09:11 -04004506 to = user_path_parent(newdfd, newname,
4507 &new_path, &new_last, &new_type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04004508 if (IS_ERR(to)) {
4509 error = PTR_ERR(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004510 goto exit1;
Jeff Layton91a27b22012-10-10 15:25:28 -04004511 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004512
4513 error = -EXDEV;
Al Virof5beed72015-04-30 16:09:11 -04004514 if (old_path.mnt != new_path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004515 goto exit2;
4516
Linus Torvalds1da177e2005-04-16 15:20:36 -07004517 error = -EBUSY;
Al Virof5beed72015-04-30 16:09:11 -04004518 if (old_type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004519 goto exit2;
4520
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004521 if (flags & RENAME_NOREPLACE)
4522 error = -EEXIST;
Al Virof5beed72015-04-30 16:09:11 -04004523 if (new_type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004524 goto exit2;
4525
Al Virof5beed72015-04-30 16:09:11 -04004526 error = mnt_want_write(old_path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02004527 if (error)
4528 goto exit2;
4529
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004530retry_deleg:
Al Virof5beed72015-04-30 16:09:11 -04004531 trap = lock_rename(new_path.dentry, old_path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004532
Al Virof5beed72015-04-30 16:09:11 -04004533 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004534 error = PTR_ERR(old_dentry);
4535 if (IS_ERR(old_dentry))
4536 goto exit3;
4537 /* source must exist */
4538 error = -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01004539 if (d_is_negative(old_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004540 goto exit4;
Al Virof5beed72015-04-30 16:09:11 -04004541 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004542 error = PTR_ERR(new_dentry);
4543 if (IS_ERR(new_dentry))
4544 goto exit4;
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004545 error = -EEXIST;
4546 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4547 goto exit5;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004548 if (flags & RENAME_EXCHANGE) {
4549 error = -ENOENT;
4550 if (d_is_negative(new_dentry))
4551 goto exit5;
4552
4553 if (!d_is_dir(new_dentry)) {
4554 error = -ENOTDIR;
Al Virof5beed72015-04-30 16:09:11 -04004555 if (new_last.name[new_last.len])
Miklos Szeredida1ce062014-04-01 17:08:43 +02004556 goto exit5;
4557 }
4558 }
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004559 /* unless the source is a directory trailing slashes give -ENOTDIR */
4560 if (!d_is_dir(old_dentry)) {
4561 error = -ENOTDIR;
Al Virof5beed72015-04-30 16:09:11 -04004562 if (old_last.name[old_last.len])
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004563 goto exit5;
Al Virof5beed72015-04-30 16:09:11 -04004564 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004565 goto exit5;
4566 }
4567 /* source should not be ancestor of target */
4568 error = -EINVAL;
4569 if (old_dentry == trap)
4570 goto exit5;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004571 /* target should not be an ancestor of source */
Miklos Szeredida1ce062014-04-01 17:08:43 +02004572 if (!(flags & RENAME_EXCHANGE))
4573 error = -ENOTEMPTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004574 if (new_dentry == trap)
4575 goto exit5;
4576
Al Virof5beed72015-04-30 16:09:11 -04004577 error = security_path_rename(&old_path, old_dentry,
4578 &new_path, new_dentry, flags);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004579 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02004580 goto exit5;
Al Virof5beed72015-04-30 16:09:11 -04004581 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4582 new_path.dentry->d_inode, new_dentry,
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004583 &delegated_inode, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004584exit5:
4585 dput(new_dentry);
4586exit4:
4587 dput(old_dentry);
4588exit3:
Al Virof5beed72015-04-30 16:09:11 -04004589 unlock_rename(new_path.dentry, old_path.dentry);
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004590 if (delegated_inode) {
4591 error = break_deleg_wait(&delegated_inode);
4592 if (!error)
4593 goto retry_deleg;
4594 }
Al Virof5beed72015-04-30 16:09:11 -04004595 mnt_drop_write(old_path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004596exit2:
Jeff Laytonc6a94282012-12-11 12:10:10 -05004597 if (retry_estale(error, lookup_flags))
4598 should_retry = true;
Al Virof5beed72015-04-30 16:09:11 -04004599 path_put(&new_path);
Al Viro2ad94ae2008-07-21 09:32:51 -04004600 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004601exit1:
Al Virof5beed72015-04-30 16:09:11 -04004602 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004603 putname(from);
Jeff Laytonc6a94282012-12-11 12:10:10 -05004604 if (should_retry) {
4605 should_retry = false;
4606 lookup_flags |= LOOKUP_REVAL;
4607 goto retry;
4608 }
Al Viro2ad94ae2008-07-21 09:32:51 -04004609exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004610 return error;
4611}
4612
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004613SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4614 int, newdfd, const char __user *, newname)
4615{
4616 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4617}
4618
Heiko Carstensa26eab22009-01-14 14:14:17 +01004619SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004620{
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004621 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004622}
4623
Miklos Szeredi787fb6b2014-10-24 00:14:36 +02004624int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4625{
4626 int error = may_create(dir, dentry);
4627 if (error)
4628 return error;
4629
4630 if (!dir->i_op->mknod)
4631 return -EPERM;
4632
4633 return dir->i_op->mknod(dir, dentry,
4634 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4635}
4636EXPORT_SYMBOL(vfs_whiteout);
4637
Al Viro5d826c82014-03-14 13:42:45 -04004638int readlink_copy(char __user *buffer, int buflen, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004639{
Al Viro5d826c82014-03-14 13:42:45 -04004640 int len = PTR_ERR(link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004641 if (IS_ERR(link))
4642 goto out;
4643
4644 len = strlen(link);
4645 if (len > (unsigned) buflen)
4646 len = buflen;
4647 if (copy_to_user(buffer, link, len))
4648 len = -EFAULT;
4649out:
4650 return len;
4651}
4652
4653/*
4654 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
Al Viro6b255392015-11-17 10:20:54 -05004655 * have ->get_link() not calling nd_jump_link(). Using (or not using) it
4656 * for any given inode is up to filesystem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004657 */
4658int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4659{
Al Virofceef392015-12-29 15:58:39 -05004660 DEFINE_DELAYED_CALL(done);
Al Viro5f2c4172015-05-07 11:14:26 -04004661 struct inode *inode = d_inode(dentry);
4662 const char *link = inode->i_link;
Marcin Slusarz694a1762008-06-09 16:40:37 -07004663 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004664
Al Virod4dee482015-04-30 20:08:02 -04004665 if (!link) {
Al Virofceef392015-12-29 15:58:39 -05004666 link = inode->i_op->get_link(dentry, inode, &done);
Al Virod4dee482015-04-30 20:08:02 -04004667 if (IS_ERR(link))
4668 return PTR_ERR(link);
4669 }
Al Viro680baac2015-05-02 13:32:22 -04004670 res = readlink_copy(buffer, buflen, link);
Al Virofceef392015-12-29 15:58:39 -05004671 do_delayed_call(&done);
Marcin Slusarz694a1762008-06-09 16:40:37 -07004672 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004673}
Al Viro4d359502014-03-14 12:20:17 -04004674EXPORT_SYMBOL(generic_readlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004675
Miklos Szeredid60874c2016-10-04 14:40:45 +02004676/**
4677 * vfs_get_link - get symlink body
4678 * @dentry: dentry on which to get symbolic link
4679 * @done: caller needs to free returned data with this
4680 *
4681 * Calls security hook and i_op->get_link() on the supplied inode.
4682 *
4683 * It does not touch atime. That's up to the caller if necessary.
4684 *
4685 * Does not work on "special" symlinks like /proc/$$/fd/N
4686 */
4687const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4688{
4689 const char *res = ERR_PTR(-EINVAL);
4690 struct inode *inode = d_inode(dentry);
4691
4692 if (d_is_symlink(dentry)) {
4693 res = ERR_PTR(security_inode_readlink(dentry));
4694 if (!res)
4695 res = inode->i_op->get_link(dentry, inode, done);
4696 }
4697 return res;
4698}
4699EXPORT_SYMBOL(vfs_get_link);
4700
Linus Torvalds1da177e2005-04-16 15:20:36 -07004701/* get the link contents into pagecache */
Al Viro6b255392015-11-17 10:20:54 -05004702const char *page_get_link(struct dentry *dentry, struct inode *inode,
Al Virofceef392015-12-29 15:58:39 -05004703 struct delayed_call *callback)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004704{
Duane Griffinebd09ab2008-12-19 20:47:12 +00004705 char *kaddr;
4706 struct page *page;
Al Viro6b255392015-11-17 10:20:54 -05004707 struct address_space *mapping = inode->i_mapping;
4708
Al Virod3883d42015-11-17 10:41:04 -05004709 if (!dentry) {
4710 page = find_get_page(mapping, 0);
4711 if (!page)
4712 return ERR_PTR(-ECHILD);
4713 if (!PageUptodate(page)) {
4714 put_page(page);
4715 return ERR_PTR(-ECHILD);
4716 }
4717 } else {
4718 page = read_mapping_page(mapping, 0, NULL);
4719 if (IS_ERR(page))
4720 return (char*)page;
4721 }
Al Virofceef392015-12-29 15:58:39 -05004722 set_delayed_call(callback, page_put_link, page);
Al Viro21fc61c2015-11-17 01:07:57 -05004723 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4724 kaddr = page_address(page);
Al Viro6b255392015-11-17 10:20:54 -05004725 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
Duane Griffinebd09ab2008-12-19 20:47:12 +00004726 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004727}
4728
Al Viro6b255392015-11-17 10:20:54 -05004729EXPORT_SYMBOL(page_get_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004730
Al Virofceef392015-12-29 15:58:39 -05004731void page_put_link(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004732{
Al Virofceef392015-12-29 15:58:39 -05004733 put_page(arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004734}
Al Viro4d359502014-03-14 12:20:17 -04004735EXPORT_SYMBOL(page_put_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004736
Al Viroaa80dea2015-11-16 18:26:34 -05004737int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4738{
Al Virofceef392015-12-29 15:58:39 -05004739 DEFINE_DELAYED_CALL(done);
Al Viro6b255392015-11-17 10:20:54 -05004740 int res = readlink_copy(buffer, buflen,
4741 page_get_link(dentry, d_inode(dentry),
Al Virofceef392015-12-29 15:58:39 -05004742 &done));
4743 do_delayed_call(&done);
Al Viroaa80dea2015-11-16 18:26:34 -05004744 return res;
4745}
4746EXPORT_SYMBOL(page_readlink);
4747
Nick Piggin54566b22009-01-04 12:00:53 -08004748/*
4749 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4750 */
4751int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004752{
4753 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004754 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07004755 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08004756 int err;
Nick Piggin54566b22009-01-04 12:00:53 -08004757 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4758 if (nofs)
4759 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004760
NeilBrown7e53cac2006-03-25 03:07:57 -08004761retry:
Nick Pigginafddba42007-10-16 01:25:01 -07004762 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08004763 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004764 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07004765 goto fail;
4766
Al Viro21fc61c2015-11-17 01:07:57 -05004767 memcpy(page_address(page), symname, len-1);
Nick Pigginafddba42007-10-16 01:25:01 -07004768
4769 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4770 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004771 if (err < 0)
4772 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07004773 if (err < len-1)
4774 goto retry;
4775
Linus Torvalds1da177e2005-04-16 15:20:36 -07004776 mark_inode_dirty(inode);
4777 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004778fail:
4779 return err;
4780}
Al Viro4d359502014-03-14 12:20:17 -04004781EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004782
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004783int page_symlink(struct inode *inode, const char *symname, int len)
4784{
4785 return __page_symlink(inode, symname, len,
Michal Hockoc62d2552015-11-06 16:28:49 -08004786 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004787}
Al Viro4d359502014-03-14 12:20:17 -04004788EXPORT_SYMBOL(page_symlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004789
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08004790const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004791 .readlink = generic_readlink,
Al Viro6b255392015-11-17 10:20:54 -05004792 .get_link = page_get_link,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004793};
Linus Torvalds1da177e2005-04-16 15:20:36 -07004794EXPORT_SYMBOL(page_symlink_inode_operations);