blob: 5b4eed2215304a14ac2614058ae2b2002a3f2ae9 [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) {
224 struct filename *tmp;
225
226 tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
227 if (unlikely(!tmp)) {
228 __putname(result);
229 return ERR_PTR(-ENOMEM);
230 }
231 tmp->name = (char *)result;
Paul Moore08518542015-01-21 23:59:56 -0500232 result = tmp;
233 } else {
234 __putname(result);
235 return ERR_PTR(-ENAMETOOLONG);
236 }
237 memcpy((char *)result->name, filename, len);
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800238 result->uptr = NULL;
239 result->aname = NULL;
Paul Moore55422d02015-01-22 00:00:23 -0500240 result->refcnt = 1;
Paul Moorefd3522f2015-01-22 00:00:10 -0500241 audit_getname(result);
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800242
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800243 return result;
244}
245
Jeff Layton91a27b22012-10-10 15:25:28 -0400246void putname(struct filename *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247{
Paul Moore55422d02015-01-22 00:00:23 -0500248 BUG_ON(name->refcnt <= 0);
249
250 if (--name->refcnt > 0)
251 return;
252
Al Virofd2f7cb2015-02-22 20:07:13 -0500253 if (name->name != name->iname) {
Paul Moore55422d02015-01-22 00:00:23 -0500254 __putname(name->name);
255 kfree(name);
256 } else
257 __putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259
Linus Torvaldse77819e2011-07-22 19:30:19 -0700260static int check_acl(struct inode *inode, int mask)
261{
Linus Torvalds84635d62011-07-25 22:47:03 -0700262#ifdef CONFIG_FS_POSIX_ACL
Linus Torvaldse77819e2011-07-22 19:30:19 -0700263 struct posix_acl *acl;
264
Linus Torvaldse77819e2011-07-22 19:30:19 -0700265 if (mask & MAY_NOT_BLOCK) {
Al Viro35678662011-08-02 21:32:13 -0400266 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
267 if (!acl)
Linus Torvaldse77819e2011-07-22 19:30:19 -0700268 return -EAGAIN;
Al Viro35678662011-08-02 21:32:13 -0400269 /* no ->get_acl() calls in RCU mode... */
Andreas Gruenbacherb8a7a3a2016-03-24 14:38:37 +0100270 if (is_uncached_acl(acl))
Al Viro35678662011-08-02 21:32:13 -0400271 return -ECHILD;
Ari Savolainen206b1d02011-08-06 19:43:07 +0300272 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
Linus Torvaldse77819e2011-07-22 19:30:19 -0700273 }
274
Christoph Hellwig2982baa2013-12-20 05:16:38 -0800275 acl = get_acl(inode, ACL_TYPE_ACCESS);
276 if (IS_ERR(acl))
277 return PTR_ERR(acl);
Linus Torvaldse77819e2011-07-22 19:30:19 -0700278 if (acl) {
279 int error = posix_acl_permission(inode, acl, mask);
280 posix_acl_release(acl);
281 return error;
282 }
Linus Torvalds84635d62011-07-25 22:47:03 -0700283#endif
Linus Torvaldse77819e2011-07-22 19:30:19 -0700284
285 return -EAGAIN;
286}
287
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700288/*
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530289 * This does the basic permission checking
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700290 */
Al Viro7e401452011-06-20 19:12:17 -0400291static int acl_permission_check(struct inode *inode, int mask)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700292{
Linus Torvalds26cf46b2011-05-13 11:51:01 -0700293 unsigned int mode = inode->i_mode;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700294
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -0800295 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700296 mode >>= 6;
297 else {
Linus Torvaldse77819e2011-07-22 19:30:19 -0700298 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
Al Viro7e401452011-06-20 19:12:17 -0400299 int error = check_acl(inode, mask);
Nick Pigginb74c79e2011-01-07 17:49:58 +1100300 if (error != -EAGAIN)
301 return error;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700302 }
303
304 if (in_group_p(inode->i_gid))
305 mode >>= 3;
306 }
307
308 /*
309 * If the DACs are ok we don't need any capability check.
310 */
Al Viro9c2c7032011-06-20 19:06:22 -0400311 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700312 return 0;
313 return -EACCES;
314}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315
316/**
Nick Pigginb74c79e2011-01-07 17:49:58 +1100317 * generic_permission - check for access rights on a Posix-like filesystem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 * @inode: inode to check access rights for
Andreas Gruenbacher8fd90c82011-10-23 23:13:30 +0530319 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 *
321 * Used to check for read/write/execute permissions on a file.
322 * We use "fsuid" for this, letting us set arbitrary permissions
323 * for filesystem access without changing the "normal" uids which
Nick Pigginb74c79e2011-01-07 17:49:58 +1100324 * are used for other things.
325 *
326 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
327 * request cannot be satisfied (eg. requires blocking or too much complexity).
328 * It would then be called again in ref-walk mode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 */
Al Viro2830ba72011-06-20 19:16:29 -0400330int generic_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700332 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333
334 /*
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530335 * Do the basic permission checks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 */
Al Viro7e401452011-06-20 19:12:17 -0400337 ret = acl_permission_check(inode, mask);
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700338 if (ret != -EACCES)
339 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340
Al Virod594e7e2011-06-20 19:55:42 -0400341 if (S_ISDIR(inode->i_mode)) {
342 /* DACs are overridable for directories */
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700343 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
Al Virod594e7e2011-06-20 19:55:42 -0400344 return 0;
345 if (!(mask & MAY_WRITE))
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700346 if (capable_wrt_inode_uidgid(inode,
347 CAP_DAC_READ_SEARCH))
Al Virod594e7e2011-06-20 19:55:42 -0400348 return 0;
349 return -EACCES;
350 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 /*
352 * Read/write DACs are always overridable.
Al Virod594e7e2011-06-20 19:55:42 -0400353 * Executable DACs are overridable when there is
354 * at least one exec bit set.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 */
Al Virod594e7e2011-06-20 19:55:42 -0400356 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700357 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 return 0;
359
360 /*
361 * Searching includes executable on directories, else just read.
362 */
Serge E. Hallyn7ea66002009-12-29 14:50:19 -0600363 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
Al Virod594e7e2011-06-20 19:55:42 -0400364 if (mask == MAY_READ)
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700365 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 return 0;
367
368 return -EACCES;
369}
Al Viro4d359502014-03-14 12:20:17 -0400370EXPORT_SYMBOL(generic_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700372/*
373 * We _really_ want to just do "generic_permission()" without
374 * even looking at the inode->i_op values. So we keep a cache
375 * flag in inode->i_opflags, that says "this has not special
376 * permission function, use the fast case".
377 */
378static inline int do_inode_permission(struct inode *inode, int mask)
379{
380 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
381 if (likely(inode->i_op->permission))
382 return inode->i_op->permission(inode, mask);
383
384 /* This gets set once for the inode lifetime */
385 spin_lock(&inode->i_lock);
386 inode->i_opflags |= IOP_FASTPERM;
387 spin_unlock(&inode->i_lock);
388 }
389 return generic_permission(inode, mask);
390}
391
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200392/**
David Howells0bdaea92012-06-25 12:55:46 +0100393 * __inode_permission - Check for access rights to a given inode
394 * @inode: Inode to check permission on
395 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200396 *
David Howells0bdaea92012-06-25 12:55:46 +0100397 * Check for read/write/execute permissions on an inode.
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530398 *
399 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
David Howells0bdaea92012-06-25 12:55:46 +0100400 *
401 * This does not check for a read-only file system. You probably want
402 * inode_permission().
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200403 */
David Howells0bdaea92012-06-25 12:55:46 +0100404int __inode_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405{
Al Viroe6305c42008-07-15 21:03:57 -0400406 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700408 if (unlikely(mask & MAY_WRITE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 /*
410 * Nobody gets write access to an immutable file.
411 */
412 if (IS_IMMUTABLE(inode))
Eryu Guan337684a2016-08-02 19:58:28 +0800413 return -EPERM;
Eric W. Biederman0bd23d092016-06-29 14:54:46 -0500414
415 /*
416 * Updating mtime will likely cause i_uid and i_gid to be
417 * written back improperly if their true value is unknown
418 * to the vfs.
419 */
420 if (HAS_UNMAPPED_ID(inode))
421 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 }
423
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700424 retval = do_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 if (retval)
426 return retval;
427
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -0700428 retval = devcgroup_inode_permission(inode, mask);
429 if (retval)
430 return retval;
431
Eric Parisd09ca732010-07-23 11:43:57 -0400432 return security_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433}
Miklos Szeredibd5d0852014-10-24 00:14:35 +0200434EXPORT_SYMBOL(__inode_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435
Al Virof4d6ff82011-06-19 13:14:21 -0400436/**
David Howells0bdaea92012-06-25 12:55:46 +0100437 * sb_permission - Check superblock-level permissions
438 * @sb: Superblock of inode to check permission on
Randy Dunlap55852632012-08-18 17:39:25 -0700439 * @inode: Inode to check permission on
David Howells0bdaea92012-06-25 12:55:46 +0100440 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
441 *
442 * Separate out file-system wide checks from inode-specific permission checks.
443 */
444static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
445{
446 if (unlikely(mask & MAY_WRITE)) {
447 umode_t mode = inode->i_mode;
448
449 /* Nobody gets write access to a read-only fs. */
450 if ((sb->s_flags & MS_RDONLY) &&
451 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
452 return -EROFS;
453 }
454 return 0;
455}
456
457/**
458 * inode_permission - Check for access rights to a given inode
459 * @inode: Inode to check permission on
460 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
461 *
462 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
463 * this, letting us set arbitrary permissions for filesystem access without
464 * changing the "normal" UIDs which are used for other things.
465 *
466 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
467 */
468int inode_permission(struct inode *inode, int mask)
469{
470 int retval;
471
472 retval = sb_permission(inode->i_sb, inode, mask);
473 if (retval)
474 return retval;
475 return __inode_permission(inode, mask);
476}
Al Viro4d359502014-03-14 12:20:17 -0400477EXPORT_SYMBOL(inode_permission);
David Howells0bdaea92012-06-25 12:55:46 +0100478
479/**
Jan Blunck5dd784d02008-02-14 19:34:38 -0800480 * path_get - get a reference to a path
481 * @path: path to get the reference to
482 *
483 * Given a path increment the reference count to the dentry and the vfsmount.
484 */
Al Virodcf787f2013-03-01 23:51:07 -0500485void path_get(const struct path *path)
Jan Blunck5dd784d02008-02-14 19:34:38 -0800486{
487 mntget(path->mnt);
488 dget(path->dentry);
489}
490EXPORT_SYMBOL(path_get);
491
492/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800493 * path_put - put a reference to a path
494 * @path: path to put the reference to
495 *
496 * Given a path decrement the reference count to the dentry and the vfsmount.
497 */
Al Virodcf787f2013-03-01 23:51:07 -0500498void path_put(const struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499{
Jan Blunck1d957f92008-02-14 19:34:35 -0800500 dput(path->dentry);
501 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502}
Jan Blunck1d957f92008-02-14 19:34:35 -0800503EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504
Al Viro894bc8c2015-05-02 07:16:16 -0400505#define EMBEDDED_LEVELS 2
Al Viro1f55a6e2014-11-01 19:30:41 -0400506struct nameidata {
507 struct path path;
Al Viro1cf26652015-05-06 16:01:56 -0400508 struct qstr last;
Al Viro1f55a6e2014-11-01 19:30:41 -0400509 struct path root;
510 struct inode *inode; /* path.dentry.d_inode */
511 unsigned int flags;
Al Viro9883d182015-05-13 07:28:08 -0400512 unsigned seq, m_seq;
Al Viro1f55a6e2014-11-01 19:30:41 -0400513 int last_type;
514 unsigned depth;
NeilBrown756daf22015-03-23 13:37:38 +1100515 int total_link_count;
Al Viro697fc6c2015-05-02 19:38:35 -0400516 struct saved {
517 struct path link;
Al Virofceef392015-12-29 15:58:39 -0500518 struct delayed_call done;
Al Viro697fc6c2015-05-02 19:38:35 -0400519 const char *name;
Al Viro0450b2d2015-05-08 13:23:53 -0400520 unsigned seq;
Al Viro894bc8c2015-05-02 07:16:16 -0400521 } *stack, internal[EMBEDDED_LEVELS];
Al Viro9883d182015-05-13 07:28:08 -0400522 struct filename *name;
523 struct nameidata *saved;
Al Virofceef392015-12-29 15:58:39 -0500524 struct inode *link_inode;
Al Viro9883d182015-05-13 07:28:08 -0400525 unsigned root_seq;
526 int dfd;
Al Viro1f55a6e2014-11-01 19:30:41 -0400527};
528
Al Viro9883d182015-05-13 07:28:08 -0400529static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
Al Viro894bc8c2015-05-02 07:16:16 -0400530{
NeilBrown756daf22015-03-23 13:37:38 +1100531 struct nameidata *old = current->nameidata;
532 p->stack = p->internal;
Al Viroc8a53ee2015-05-12 18:43:07 -0400533 p->dfd = dfd;
534 p->name = name;
NeilBrown756daf22015-03-23 13:37:38 +1100535 p->total_link_count = old ? old->total_link_count : 0;
Al Viro9883d182015-05-13 07:28:08 -0400536 p->saved = old;
NeilBrown756daf22015-03-23 13:37:38 +1100537 current->nameidata = p;
Al Viro894bc8c2015-05-02 07:16:16 -0400538}
539
Al Viro9883d182015-05-13 07:28:08 -0400540static void restore_nameidata(void)
Al Viro894bc8c2015-05-02 07:16:16 -0400541{
Al Viro9883d182015-05-13 07:28:08 -0400542 struct nameidata *now = current->nameidata, *old = now->saved;
NeilBrown756daf22015-03-23 13:37:38 +1100543
544 current->nameidata = old;
545 if (old)
546 old->total_link_count = now->total_link_count;
Al Viroe1a63bb2015-12-05 21:06:33 -0500547 if (now->stack != now->internal)
NeilBrown756daf22015-03-23 13:37:38 +1100548 kfree(now->stack);
Al Viro894bc8c2015-05-02 07:16:16 -0400549}
550
551static int __nd_alloc_stack(struct nameidata *nd)
552{
Al Virobc40aee2015-05-09 13:04:24 -0400553 struct saved *p;
554
555 if (nd->flags & LOOKUP_RCU) {
556 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
557 GFP_ATOMIC);
558 if (unlikely(!p))
559 return -ECHILD;
560 } else {
561 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
Al Viro894bc8c2015-05-02 07:16:16 -0400562 GFP_KERNEL);
Al Virobc40aee2015-05-09 13:04:24 -0400563 if (unlikely(!p))
564 return -ENOMEM;
565 }
Al Viro894bc8c2015-05-02 07:16:16 -0400566 memcpy(p, nd->internal, sizeof(nd->internal));
567 nd->stack = p;
568 return 0;
569}
570
Eric W. Biederman397d4252015-08-15 20:27:13 -0500571/**
572 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
573 * @path: nameidate to verify
574 *
575 * Rename can sometimes move a file or directory outside of a bind
576 * mount, path_connected allows those cases to be detected.
577 */
578static bool path_connected(const struct path *path)
579{
580 struct vfsmount *mnt = path->mnt;
581
582 /* Only bind mounts can have disconnected paths */
583 if (mnt->mnt_root == mnt->mnt_sb->s_root)
584 return true;
585
586 return is_subdir(path->dentry, mnt->mnt_root);
587}
588
Al Viro894bc8c2015-05-02 07:16:16 -0400589static inline int nd_alloc_stack(struct nameidata *nd)
590{
Al Viroda4e0be2015-05-03 20:52:15 -0400591 if (likely(nd->depth != EMBEDDED_LEVELS))
Al Viro894bc8c2015-05-02 07:16:16 -0400592 return 0;
593 if (likely(nd->stack != nd->internal))
594 return 0;
595 return __nd_alloc_stack(nd);
596}
597
Al Viro79733872015-05-09 12:55:43 -0400598static void drop_links(struct nameidata *nd)
599{
600 int i = nd->depth;
601 while (i--) {
602 struct saved *last = nd->stack + i;
Al Virofceef392015-12-29 15:58:39 -0500603 do_delayed_call(&last->done);
604 clear_delayed_call(&last->done);
Al Viro79733872015-05-09 12:55:43 -0400605 }
606}
607
608static void terminate_walk(struct nameidata *nd)
609{
610 drop_links(nd);
611 if (!(nd->flags & LOOKUP_RCU)) {
612 int i;
613 path_put(&nd->path);
614 for (i = 0; i < nd->depth; i++)
615 path_put(&nd->stack[i].link);
Al Viro102b8af2015-05-12 17:35:52 -0400616 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
617 path_put(&nd->root);
618 nd->root.mnt = NULL;
619 }
Al Viro79733872015-05-09 12:55:43 -0400620 } else {
621 nd->flags &= ~LOOKUP_RCU;
622 if (!(nd->flags & LOOKUP_ROOT))
623 nd->root.mnt = NULL;
624 rcu_read_unlock();
625 }
626 nd->depth = 0;
627}
628
629/* path_put is needed afterwards regardless of success or failure */
630static bool legitimize_path(struct nameidata *nd,
631 struct path *path, unsigned seq)
632{
633 int res = __legitimize_mnt(path->mnt, nd->m_seq);
634 if (unlikely(res)) {
635 if (res > 0)
636 path->mnt = NULL;
637 path->dentry = NULL;
638 return false;
639 }
640 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
641 path->dentry = NULL;
642 return false;
643 }
644 return !read_seqcount_retry(&path->dentry->d_seq, seq);
645}
646
647static bool legitimize_links(struct nameidata *nd)
648{
649 int i;
650 for (i = 0; i < nd->depth; i++) {
651 struct saved *last = nd->stack + i;
652 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
653 drop_links(nd);
654 nd->depth = i + 1;
655 return false;
656 }
657 }
658 return true;
659}
660
Al Viro19660af2011-03-25 10:32:48 -0400661/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100662 * Path walking has 2 modes, rcu-walk and ref-walk (see
Al Viro19660af2011-03-25 10:32:48 -0400663 * Documentation/filesystems/path-lookup.txt). In situations when we can't
664 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
Mike Marshall57e37152015-11-30 11:11:59 -0500665 * normal reference counts on dentries and vfsmounts to transition to ref-walk
Al Viro19660af2011-03-25 10:32:48 -0400666 * mode. Refcounts are grabbed at the last known good point before rcu-walk
667 * got stuck, so ref-walk may continue from there. If this is not successful
668 * (eg. a seqcount has changed), then failure is returned and it's up to caller
669 * to restart the path walk from the beginning in ref-walk mode.
Nick Piggin31e6b012011-01-07 17:49:52 +1100670 */
Nick Piggin31e6b012011-01-07 17:49:52 +1100671
672/**
Al Viro19660af2011-03-25 10:32:48 -0400673 * unlazy_walk - try to switch to ref-walk mode.
674 * @nd: nameidata pathwalk data
675 * @dentry: child of nd->path.dentry or NULL
Al Viro6e9918b2015-05-05 09:26:05 -0400676 * @seq: seq number to check dentry against
Randy Dunlap39191622011-01-08 19:36:21 -0800677 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100678 *
Al Viro19660af2011-03-25 10:32:48 -0400679 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
680 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
681 * @nd or NULL. Must be called from rcu-walk context.
Al Viro79733872015-05-09 12:55:43 -0400682 * Nothing should touch nameidata between unlazy_walk() failure and
683 * terminate_walk().
Nick Piggin31e6b012011-01-07 17:49:52 +1100684 */
Al Viro6e9918b2015-05-05 09:26:05 -0400685static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq)
Nick Piggin31e6b012011-01-07 17:49:52 +1100686{
Nick Piggin31e6b012011-01-07 17:49:52 +1100687 struct dentry *parent = nd->path.dentry;
688
689 BUG_ON(!(nd->flags & LOOKUP_RCU));
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700690
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700691 nd->flags &= ~LOOKUP_RCU;
Al Viro79733872015-05-09 12:55:43 -0400692 if (unlikely(!legitimize_links(nd)))
693 goto out2;
694 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
695 goto out2;
696 if (unlikely(!lockref_get_not_dead(&parent->d_lockref)))
697 goto out1;
Al Viro48a066e2013-09-29 22:06:07 -0400698
Linus Torvalds15570082013-09-02 11:38:06 -0700699 /*
700 * For a negative lookup, the lookup sequence point is the parents
701 * sequence point, and it only needs to revalidate the parent dentry.
702 *
703 * For a positive lookup, we need to move both the parent and the
704 * dentry from the RCU domain to be properly refcounted. And the
705 * sequence number in the dentry validates *both* dentry counters,
706 * since we checked the sequence number of the parent after we got
707 * the child sequence number. So we know the parent must still
708 * be valid if the child sequence number is still valid.
709 */
Al Viro19660af2011-03-25 10:32:48 -0400710 if (!dentry) {
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700711 if (read_seqcount_retry(&parent->d_seq, nd->seq))
712 goto out;
Al Viro19660af2011-03-25 10:32:48 -0400713 BUG_ON(nd->inode != parent->d_inode);
714 } else {
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700715 if (!lockref_get_not_dead(&dentry->d_lockref))
716 goto out;
Al Viro6e9918b2015-05-05 09:26:05 -0400717 if (read_seqcount_retry(&dentry->d_seq, seq))
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700718 goto drop_dentry;
Al Viro19660af2011-03-25 10:32:48 -0400719 }
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700720
721 /*
722 * Sequence counts matched. Now make sure that the root is
723 * still valid and get it if required.
724 */
725 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
Al Viro5a8d87e2015-05-12 00:10:18 -0400726 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
727 rcu_read_unlock();
728 dput(dentry);
729 return -ECHILD;
Al Viro79733872015-05-09 12:55:43 -0400730 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100731 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100732
Al Viro8b61e742013-11-08 12:45:01 -0500733 rcu_read_unlock();
Nick Piggin31e6b012011-01-07 17:49:52 +1100734 return 0;
Al Viro19660af2011-03-25 10:32:48 -0400735
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700736drop_dentry:
Al Viro8b61e742013-11-08 12:45:01 -0500737 rcu_read_unlock();
Linus Torvalds15570082013-09-02 11:38:06 -0700738 dput(dentry);
Linus Torvaldsd0d27272013-09-10 12:17:49 -0700739 goto drop_root_mnt;
Al Viro79733872015-05-09 12:55:43 -0400740out2:
741 nd->path.mnt = NULL;
742out1:
743 nd->path.dentry = NULL;
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700744out:
Al Viro8b61e742013-11-08 12:45:01 -0500745 rcu_read_unlock();
Linus Torvaldsd0d27272013-09-10 12:17:49 -0700746drop_root_mnt:
747 if (!(nd->flags & LOOKUP_ROOT))
748 nd->root.mnt = NULL;
Nick Piggin31e6b012011-01-07 17:49:52 +1100749 return -ECHILD;
750}
751
Al Viro79733872015-05-09 12:55:43 -0400752static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq)
753{
754 if (unlikely(!legitimize_path(nd, link, seq))) {
755 drop_links(nd);
756 nd->depth = 0;
757 nd->flags &= ~LOOKUP_RCU;
758 nd->path.mnt = NULL;
759 nd->path.dentry = NULL;
760 if (!(nd->flags & LOOKUP_ROOT))
761 nd->root.mnt = NULL;
762 rcu_read_unlock();
763 } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) {
764 return 0;
765 }
766 path_put(link);
767 return -ECHILD;
768}
769
Al Viro4ce16ef32012-06-10 16:10:59 -0400770static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
Nick Piggin34286d62011-01-07 17:49:57 +1100771{
Al Viro4ce16ef32012-06-10 16:10:59 -0400772 return dentry->d_op->d_revalidate(dentry, flags);
Nick Piggin34286d62011-01-07 17:49:57 +1100773}
774
Al Viro9f1fafe2011-03-25 11:00:12 -0400775/**
776 * complete_walk - successful completion of path walk
777 * @nd: pointer nameidata
Jeff Layton39159de2009-12-07 12:01:50 -0500778 *
Al Viro9f1fafe2011-03-25 11:00:12 -0400779 * If we had been in RCU mode, drop out of it and legitimize nd->path.
780 * Revalidate the final result, unless we'd already done that during
781 * the path walk or the filesystem doesn't ask for it. Return 0 on
782 * success, -error on failure. In case of failure caller does not
783 * need to drop nd->path.
Jeff Layton39159de2009-12-07 12:01:50 -0500784 */
Al Viro9f1fafe2011-03-25 11:00:12 -0400785static int complete_walk(struct nameidata *nd)
Jeff Layton39159de2009-12-07 12:01:50 -0500786{
Al Viro16c2cd72011-02-22 15:50:10 -0500787 struct dentry *dentry = nd->path.dentry;
Jeff Layton39159de2009-12-07 12:01:50 -0500788 int status;
Jeff Layton39159de2009-12-07 12:01:50 -0500789
Al Viro9f1fafe2011-03-25 11:00:12 -0400790 if (nd->flags & LOOKUP_RCU) {
Al Viro9f1fafe2011-03-25 11:00:12 -0400791 if (!(nd->flags & LOOKUP_ROOT))
792 nd->root.mnt = NULL;
Al Viro6e9918b2015-05-05 09:26:05 -0400793 if (unlikely(unlazy_walk(nd, NULL, 0)))
Al Viro48a066e2013-09-29 22:06:07 -0400794 return -ECHILD;
Al Viro9f1fafe2011-03-25 11:00:12 -0400795 }
796
Al Viro16c2cd72011-02-22 15:50:10 -0500797 if (likely(!(nd->flags & LOOKUP_JUMPED)))
Jeff Layton39159de2009-12-07 12:01:50 -0500798 return 0;
799
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500800 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
Al Viro16c2cd72011-02-22 15:50:10 -0500801 return 0;
802
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500803 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
Jeff Layton39159de2009-12-07 12:01:50 -0500804 if (status > 0)
805 return 0;
806
Al Viro16c2cd72011-02-22 15:50:10 -0500807 if (!status)
Jeff Layton39159de2009-12-07 12:01:50 -0500808 status = -ESTALE;
Al Viro16c2cd72011-02-22 15:50:10 -0500809
Jeff Layton39159de2009-12-07 12:01:50 -0500810 return status;
811}
812
Al Viro18d8c862015-05-12 16:32:34 -0400813static void set_root(struct nameidata *nd)
Al Viro2a737872009-04-07 11:49:53 -0400814{
Al Viro7bd88372014-09-13 21:55:46 -0400815 struct fs_struct *fs = current->fs;
Nick Pigginc28cc362011-01-07 17:49:53 +1100816
Al Viro9e6697e2015-12-05 20:07:21 -0500817 if (nd->flags & LOOKUP_RCU) {
818 unsigned seq;
819
820 do {
821 seq = read_seqcount_begin(&fs->seq);
822 nd->root = fs->root;
823 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
824 } while (read_seqcount_retry(&fs->seq, seq));
825 } else {
826 get_fs_root(fs, &nd->root);
827 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100828}
829
Jan Blunck1d957f92008-02-14 19:34:35 -0800830static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700831{
832 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800833 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700834 mntput(path->mnt);
835}
836
Nick Piggin7b9337a2011-01-14 08:42:43 +0000837static inline void path_to_nameidata(const struct path *path,
838 struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700839{
Nick Piggin31e6b012011-01-07 17:49:52 +1100840 if (!(nd->flags & LOOKUP_RCU)) {
841 dput(nd->path.dentry);
842 if (nd->path.mnt != path->mnt)
843 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800844 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100845 nd->path.mnt = path->mnt;
Jan Blunck4ac91372008-02-14 19:34:32 -0800846 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700847}
848
Al Viro248fb5b2015-12-05 20:51:58 -0500849static int nd_jump_root(struct nameidata *nd)
850{
851 if (nd->flags & LOOKUP_RCU) {
852 struct dentry *d;
853 nd->path = nd->root;
854 d = nd->path.dentry;
855 nd->inode = d->d_inode;
856 nd->seq = nd->root_seq;
857 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
858 return -ECHILD;
859 } else {
860 path_put(&nd->path);
861 nd->path = nd->root;
862 path_get(&nd->path);
863 nd->inode = nd->path.dentry->d_inode;
864 }
865 nd->flags |= LOOKUP_JUMPED;
866 return 0;
867}
868
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400869/*
Al Viro6b255392015-11-17 10:20:54 -0500870 * Helper to directly jump to a known parsed path from ->get_link,
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400871 * caller must have taken a reference to path beforehand.
872 */
Al Viro6e771372015-05-02 13:37:52 -0400873void nd_jump_link(struct path *path)
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400874{
Al Viro6e771372015-05-02 13:37:52 -0400875 struct nameidata *nd = current->nameidata;
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400876 path_put(&nd->path);
877
878 nd->path = *path;
879 nd->inode = nd->path.dentry->d_inode;
880 nd->flags |= LOOKUP_JUMPED;
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400881}
882
Al Virob9ff4422015-05-02 20:19:23 -0400883static inline void put_link(struct nameidata *nd)
Al Viro574197e2011-03-14 22:20:34 -0400884{
Al Viro21c30032015-05-03 21:06:24 -0400885 struct saved *last = nd->stack + --nd->depth;
Al Virofceef392015-12-29 15:58:39 -0500886 do_delayed_call(&last->done);
Al Viro6548fae2015-05-07 20:32:22 -0400887 if (!(nd->flags & LOOKUP_RCU))
888 path_put(&last->link);
Al Viro574197e2011-03-14 22:20:34 -0400889}
890
Linus Torvalds561ec642012-10-26 10:05:07 -0700891int sysctl_protected_symlinks __read_mostly = 0;
892int sysctl_protected_hardlinks __read_mostly = 0;
Kees Cook800179c2012-07-25 17:29:07 -0700893
894/**
895 * may_follow_link - Check symlink following for unsafe situations
Randy Dunlap55852632012-08-18 17:39:25 -0700896 * @nd: nameidata pathwalk data
Kees Cook800179c2012-07-25 17:29:07 -0700897 *
898 * In the case of the sysctl_protected_symlinks sysctl being enabled,
899 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
900 * in a sticky world-writable directory. This is to protect privileged
901 * processes from failing races against path names that may change out
902 * from under them by way of other users creating malicious symlinks.
903 * It will permit symlinks to be followed only when outside a sticky
904 * world-writable directory, or when the uid of the symlink and follower
905 * match, or when the directory owner matches the symlink's owner.
906 *
907 * Returns 0 if following the symlink is allowed, -ve on error.
908 */
Al Virofec2fa22015-05-06 15:58:18 -0400909static inline int may_follow_link(struct nameidata *nd)
Kees Cook800179c2012-07-25 17:29:07 -0700910{
911 const struct inode *inode;
912 const struct inode *parent;
Seth Forshee2d7f9e22016-04-26 14:36:23 -0500913 kuid_t puid;
Kees Cook800179c2012-07-25 17:29:07 -0700914
915 if (!sysctl_protected_symlinks)
916 return 0;
917
918 /* Allowed if owner and follower match. */
Al Virofceef392015-12-29 15:58:39 -0500919 inode = nd->link_inode;
Eric W. Biederman81abe272012-08-03 09:38:08 -0700920 if (uid_eq(current_cred()->fsuid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700921 return 0;
922
923 /* Allowed if parent directory not sticky and world-writable. */
Al Viroaa65fa32015-08-04 23:23:50 -0400924 parent = nd->inode;
Kees Cook800179c2012-07-25 17:29:07 -0700925 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
926 return 0;
927
928 /* Allowed if parent directory and link owner match. */
Seth Forshee2d7f9e22016-04-26 14:36:23 -0500929 puid = parent->i_uid;
930 if (uid_valid(puid) && uid_eq(puid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700931 return 0;
932
Al Viro319565022015-05-07 20:37:40 -0400933 if (nd->flags & LOOKUP_RCU)
934 return -ECHILD;
935
Al Viro1cf26652015-05-06 16:01:56 -0400936 audit_log_link_denied("follow_link", &nd->stack[0].link);
Kees Cook800179c2012-07-25 17:29:07 -0700937 return -EACCES;
938}
939
940/**
941 * safe_hardlink_source - Check for safe hardlink conditions
942 * @inode: the source inode to hardlink from
943 *
944 * Return false if at least one of the following conditions:
945 * - inode is not a regular file
946 * - inode is setuid
947 * - inode is setgid and group-exec
948 * - access failure for read and write
949 *
950 * Otherwise returns true.
951 */
952static bool safe_hardlink_source(struct inode *inode)
953{
954 umode_t mode = inode->i_mode;
955
956 /* Special files should not get pinned to the filesystem. */
957 if (!S_ISREG(mode))
958 return false;
959
960 /* Setuid files should not get pinned to the filesystem. */
961 if (mode & S_ISUID)
962 return false;
963
964 /* Executable setgid files should not get pinned to the filesystem. */
965 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
966 return false;
967
968 /* Hardlinking to unreadable or unwritable sources is dangerous. */
969 if (inode_permission(inode, MAY_READ | MAY_WRITE))
970 return false;
971
972 return true;
973}
974
975/**
976 * may_linkat - Check permissions for creating a hardlink
977 * @link: the source to hardlink from
978 *
979 * Block hardlink when all of:
980 * - sysctl_protected_hardlinks enabled
981 * - fsuid does not match inode
982 * - hardlink source is unsafe (see safe_hardlink_source() above)
Dirk Steinmetzf2ca3792015-10-20 16:09:19 +0200983 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
Kees Cook800179c2012-07-25 17:29:07 -0700984 *
985 * Returns 0 if successful, -ve on error.
986 */
987static int may_linkat(struct path *link)
988{
Kees Cook800179c2012-07-25 17:29:07 -0700989 struct inode *inode;
990
991 if (!sysctl_protected_hardlinks)
992 return 0;
993
Kees Cook800179c2012-07-25 17:29:07 -0700994 inode = link->dentry->d_inode;
995
996 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
997 * otherwise, it must be a safe source.
998 */
Dirk Steinmetzf2ca3792015-10-20 16:09:19 +0200999 if (inode_owner_or_capable(inode) || safe_hardlink_source(inode))
Kees Cook800179c2012-07-25 17:29:07 -07001000 return 0;
1001
Kees Cooka51d9ea2012-07-25 17:29:08 -07001002 audit_log_link_denied("linkat", link);
Kees Cook800179c2012-07-25 17:29:07 -07001003 return -EPERM;
1004}
1005
Al Viro3b2e7f72015-04-19 00:53:50 -04001006static __always_inline
1007const char *get_link(struct nameidata *nd)
Ian Kent051d3812006-03-27 01:14:53 -08001008{
Al Viroab104922015-05-10 11:50:01 -04001009 struct saved *last = nd->stack + nd->depth - 1;
Al Viro1cf26652015-05-06 16:01:56 -04001010 struct dentry *dentry = last->link.dentry;
Al Virofceef392015-12-29 15:58:39 -05001011 struct inode *inode = nd->link_inode;
Al Viro6d7b5aa2012-06-10 04:15:17 -04001012 int error;
Al Viro0a959df2015-04-18 18:23:41 -04001013 const char *res;
Ian Kent051d3812006-03-27 01:14:53 -08001014
NeilBrown8fa9dd22015-03-23 13:37:40 +11001015 if (!(nd->flags & LOOKUP_RCU)) {
1016 touch_atime(&last->link);
1017 cond_resched();
Miklos Szeredi598e3c82016-09-16 12:44:20 +02001018 } else if (atime_needs_update_rcu(&last->link, inode)) {
NeilBrown8fa9dd22015-03-23 13:37:40 +11001019 if (unlikely(unlazy_walk(nd, NULL, 0)))
1020 return ERR_PTR(-ECHILD);
1021 touch_atime(&last->link);
1022 }
1023
NeilBrownbda0be72015-03-23 13:37:39 +11001024 error = security_inode_follow_link(dentry, inode,
1025 nd->flags & LOOKUP_RCU);
1026 if (unlikely(error))
Al Viro6920a442015-05-10 10:43:46 -04001027 return ERR_PTR(error);
Al Viro36f3b4f2011-02-22 21:24:38 -05001028
Al Viro86acdca12009-12-22 23:45:11 -05001029 nd->last_type = LAST_BIND;
Al Virod4dee482015-04-30 20:08:02 -04001030 res = inode->i_link;
1031 if (!res) {
Al Virofceef392015-12-29 15:58:39 -05001032 const char * (*get)(struct dentry *, struct inode *,
1033 struct delayed_call *);
1034 get = inode->i_op->get_link;
Al Viro8c1b4562015-05-09 18:15:21 -04001035 if (nd->flags & LOOKUP_RCU) {
Al Virofceef392015-12-29 15:58:39 -05001036 res = get(NULL, inode, &last->done);
Al Viro6b255392015-11-17 10:20:54 -05001037 if (res == ERR_PTR(-ECHILD)) {
1038 if (unlikely(unlazy_walk(nd, NULL, 0)))
1039 return ERR_PTR(-ECHILD);
Al Virofceef392015-12-29 15:58:39 -05001040 res = get(dentry, inode, &last->done);
Al Viro6b255392015-11-17 10:20:54 -05001041 }
1042 } else {
Al Virofceef392015-12-29 15:58:39 -05001043 res = get(dentry, inode, &last->done);
Al Viro8c1b4562015-05-09 18:15:21 -04001044 }
Al Virofceef392015-12-29 15:58:39 -05001045 if (IS_ERR_OR_NULL(res))
Al Virofab51e82015-05-10 11:01:00 -04001046 return res;
Ian Kent051d3812006-03-27 01:14:53 -08001047 }
Al Virofab51e82015-05-10 11:01:00 -04001048 if (*res == '/') {
Al Viro9e6697e2015-12-05 20:07:21 -05001049 if (!nd->root.mnt)
1050 set_root(nd);
Al Viro248fb5b2015-12-05 20:51:58 -05001051 if (unlikely(nd_jump_root(nd)))
1052 return ERR_PTR(-ECHILD);
Al Virofab51e82015-05-10 11:01:00 -04001053 while (unlikely(*++res == '/'))
1054 ;
1055 }
1056 if (!*res)
1057 res = NULL;
Al Viro0a959df2015-04-18 18:23:41 -04001058 return res;
1059}
Al Viro6d7b5aa2012-06-10 04:15:17 -04001060
David Howellsf015f1262012-06-25 12:55:28 +01001061/*
1062 * follow_up - Find the mountpoint of path's vfsmount
1063 *
1064 * Given a path, find the mountpoint of its source file system.
1065 * Replace @path with the path of the mountpoint in the parent mount.
1066 * Up is towards /.
1067 *
1068 * Return 1 if we went up a level and 0 if we were already at the
1069 * root.
1070 */
Al Virobab77eb2009-04-18 03:26:48 -04001071int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072{
Al Viro0714a532011-11-24 22:19:58 -05001073 struct mount *mnt = real_mount(path->mnt);
1074 struct mount *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +10001076
Al Viro48a066e2013-09-29 22:06:07 -04001077 read_seqlock_excl(&mount_lock);
Al Viro0714a532011-11-24 22:19:58 -05001078 parent = mnt->mnt_parent;
Al Viro3c0a6162012-07-18 17:32:50 +04001079 if (parent == mnt) {
Al Viro48a066e2013-09-29 22:06:07 -04001080 read_sequnlock_excl(&mount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 return 0;
1082 }
Al Viro0714a532011-11-24 22:19:58 -05001083 mntget(&parent->mnt);
Al Viroa73324d2011-11-24 22:25:07 -05001084 mountpoint = dget(mnt->mnt_mountpoint);
Al Viro48a066e2013-09-29 22:06:07 -04001085 read_sequnlock_excl(&mount_lock);
Al Virobab77eb2009-04-18 03:26:48 -04001086 dput(path->dentry);
1087 path->dentry = mountpoint;
1088 mntput(path->mnt);
Al Viro0714a532011-11-24 22:19:58 -05001089 path->mnt = &parent->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 return 1;
1091}
Al Viro4d359502014-03-14 12:20:17 -04001092EXPORT_SYMBOL(follow_up);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001094/*
David Howells9875cf82011-01-14 18:45:21 +00001095 * Perform an automount
1096 * - return -EISDIR to tell follow_managed() to stop and return the path we
1097 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 */
NeilBrown756daf22015-03-23 13:37:38 +11001099static int follow_automount(struct path *path, struct nameidata *nd,
David Howells9875cf82011-01-14 18:45:21 +00001100 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +11001101{
David Howells9875cf82011-01-14 18:45:21 +00001102 struct vfsmount *mnt;
Eric W. Biedermanaeaa4a72016-07-23 11:20:44 -05001103 const struct cred *old_cred;
David Howellsea5b7782011-01-14 19:10:03 +00001104 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001105
David Howells9875cf82011-01-14 18:45:21 +00001106 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1107 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -07001108
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +02001109 /* We don't want to mount if someone's just doing a stat -
1110 * unless they're stat'ing a directory and appended a '/' to
1111 * the name.
1112 *
1113 * We do, however, want to mount if someone wants to open or
1114 * create a file of any type under the mountpoint, wants to
1115 * traverse through the mountpoint or wants to open the
1116 * mounted directory. Also, autofs may mark negative dentries
1117 * as being automount points. These will need the attentions
1118 * of the daemon to instantiate them before they can be used.
David Howells9875cf82011-01-14 18:45:21 +00001119 */
NeilBrown756daf22015-03-23 13:37:38 +11001120 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1121 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +02001122 path->dentry->d_inode)
1123 return -EISDIR;
1124
Eric W. Biedermanaeaa4a72016-07-23 11:20:44 -05001125 if (path->dentry->d_sb->s_user_ns != &init_user_ns)
1126 return -EACCES;
1127
NeilBrown756daf22015-03-23 13:37:38 +11001128 nd->total_link_count++;
1129 if (nd->total_link_count >= 40)
David Howells9875cf82011-01-14 18:45:21 +00001130 return -ELOOP;
1131
Eric W. Biedermanaeaa4a72016-07-23 11:20:44 -05001132 old_cred = override_creds(&init_cred);
David Howells9875cf82011-01-14 18:45:21 +00001133 mnt = path->dentry->d_op->d_automount(path);
Eric W. Biedermanaeaa4a72016-07-23 11:20:44 -05001134 revert_creds(old_cred);
David Howells9875cf82011-01-14 18:45:21 +00001135 if (IS_ERR(mnt)) {
1136 /*
1137 * The filesystem is allowed to return -EISDIR here to indicate
1138 * it doesn't want to automount. For instance, autofs would do
1139 * this so that its userspace daemon can mount on this dentry.
1140 *
1141 * However, we can only permit this if it's a terminal point in
1142 * the path being looked up; if it wasn't then the remainder of
1143 * the path is inaccessible and we should say so.
1144 */
NeilBrown756daf22015-03-23 13:37:38 +11001145 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
David Howells9875cf82011-01-14 18:45:21 +00001146 return -EREMOTE;
1147 return PTR_ERR(mnt);
1148 }
David Howellsea5b7782011-01-14 19:10:03 +00001149
David Howells9875cf82011-01-14 18:45:21 +00001150 if (!mnt) /* mount collision */
1151 return 0;
1152
Al Viro8aef1882011-06-16 15:10:06 +01001153 if (!*need_mntput) {
1154 /* lock_mount() may release path->mnt on error */
1155 mntget(path->mnt);
1156 *need_mntput = true;
1157 }
Al Viro19a167a2011-01-17 01:35:23 -05001158 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +00001159
David Howellsea5b7782011-01-14 19:10:03 +00001160 switch (err) {
1161 case -EBUSY:
1162 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -05001163 return 0;
David Howellsea5b7782011-01-14 19:10:03 +00001164 case 0:
Al Viro8aef1882011-06-16 15:10:06 +01001165 path_put(path);
David Howellsea5b7782011-01-14 19:10:03 +00001166 path->mnt = mnt;
1167 path->dentry = dget(mnt->mnt_root);
David Howellsea5b7782011-01-14 19:10:03 +00001168 return 0;
Al Viro19a167a2011-01-17 01:35:23 -05001169 default:
1170 return err;
David Howellsea5b7782011-01-14 19:10:03 +00001171 }
Al Viro19a167a2011-01-17 01:35:23 -05001172
David Howells9875cf82011-01-14 18:45:21 +00001173}
1174
1175/*
1176 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +00001177 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +00001178 * - Flagged as mountpoint
1179 * - Flagged as automount point
1180 *
1181 * This may only be called in refwalk mode.
1182 *
1183 * Serialization is taken care of in namespace.c
1184 */
NeilBrown756daf22015-03-23 13:37:38 +11001185static int follow_managed(struct path *path, struct nameidata *nd)
David Howells9875cf82011-01-14 18:45:21 +00001186{
Al Viro8aef1882011-06-16 15:10:06 +01001187 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
David Howells9875cf82011-01-14 18:45:21 +00001188 unsigned managed;
1189 bool need_mntput = false;
Al Viro8aef1882011-06-16 15:10:06 +01001190 int ret = 0;
David Howells9875cf82011-01-14 18:45:21 +00001191
1192 /* Given that we're not holding a lock here, we retain the value in a
1193 * local variable for each dentry as we look at it so that we don't see
1194 * the components of that value change under us */
1195 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1196 managed &= DCACHE_MANAGED_DENTRY,
1197 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +00001198 /* Allow the filesystem to manage the transit without i_mutex
1199 * being held. */
1200 if (managed & DCACHE_MANAGE_TRANSIT) {
1201 BUG_ON(!path->dentry->d_op);
1202 BUG_ON(!path->dentry->d_op->d_manage);
Al Viro1aed3e42011-03-18 09:09:02 -04001203 ret = path->dentry->d_op->d_manage(path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001204 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001205 break;
David Howellscc53ce52011-01-14 18:45:26 +00001206 }
1207
David Howells9875cf82011-01-14 18:45:21 +00001208 /* Transit to a mounted filesystem. */
1209 if (managed & DCACHE_MOUNTED) {
1210 struct vfsmount *mounted = lookup_mnt(path);
1211 if (mounted) {
1212 dput(path->dentry);
1213 if (need_mntput)
1214 mntput(path->mnt);
1215 path->mnt = mounted;
1216 path->dentry = dget(mounted->mnt_root);
1217 need_mntput = true;
1218 continue;
1219 }
1220
1221 /* Something is mounted on this dentry in another
1222 * namespace and/or whatever was mounted there in this
Al Viro48a066e2013-09-29 22:06:07 -04001223 * namespace got unmounted before lookup_mnt() could
1224 * get it */
David Howells9875cf82011-01-14 18:45:21 +00001225 }
1226
1227 /* Handle an automount point */
1228 if (managed & DCACHE_NEED_AUTOMOUNT) {
NeilBrown756daf22015-03-23 13:37:38 +11001229 ret = follow_automount(path, nd, &need_mntput);
David Howells9875cf82011-01-14 18:45:21 +00001230 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001231 break;
David Howells9875cf82011-01-14 18:45:21 +00001232 continue;
1233 }
1234
1235 /* We didn't change the current path point */
1236 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 }
Al Viro8aef1882011-06-16 15:10:06 +01001238
1239 if (need_mntput && path->mnt == mnt)
1240 mntput(path->mnt);
Al Viroe9742b52016-03-05 22:04:59 -05001241 if (ret == -EISDIR || !ret)
1242 ret = 1;
Al Viro84027522015-04-22 10:30:08 -04001243 if (need_mntput)
1244 nd->flags |= LOOKUP_JUMPED;
1245 if (unlikely(ret < 0))
1246 path_put_conditional(path, nd);
1247 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248}
1249
David Howellscc53ce52011-01-14 18:45:26 +00001250int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251{
1252 struct vfsmount *mounted;
1253
Al Viro1c755af2009-04-18 14:06:57 -04001254 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -04001256 dput(path->dentry);
1257 mntput(path->mnt);
1258 path->mnt = mounted;
1259 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 return 1;
1261 }
1262 return 0;
1263}
Al Viro4d359502014-03-14 12:20:17 -04001264EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265
NeilBrownb8faf032014-08-04 17:06:29 +10001266static inline int managed_dentry_rcu(struct dentry *dentry)
Ian Kent62a73752011-03-25 01:51:02 +08001267{
NeilBrownb8faf032014-08-04 17:06:29 +10001268 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1269 dentry->d_op->d_manage(dentry, true) : 0;
Ian Kent62a73752011-03-25 01:51:02 +08001270}
1271
David Howells9875cf82011-01-14 18:45:21 +00001272/*
Al Viro287548e2011-05-27 06:50:06 -04001273 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1274 * we meet a managed dentry that would need blocking.
David Howells9875cf82011-01-14 18:45:21 +00001275 */
1276static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
Al Viro254cf582015-05-05 09:40:46 -04001277 struct inode **inode, unsigned *seqp)
David Howells9875cf82011-01-14 18:45:21 +00001278{
Ian Kent62a73752011-03-25 01:51:02 +08001279 for (;;) {
Al Viroc7105362011-11-24 18:22:03 -05001280 struct mount *mounted;
Ian Kent62a73752011-03-25 01:51:02 +08001281 /*
1282 * Don't forget we might have a non-mountpoint managed dentry
1283 * that wants to block transit.
1284 */
NeilBrownb8faf032014-08-04 17:06:29 +10001285 switch (managed_dentry_rcu(path->dentry)) {
1286 case -ECHILD:
1287 default:
David Howellsab909112011-01-14 18:46:51 +00001288 return false;
NeilBrownb8faf032014-08-04 17:06:29 +10001289 case -EISDIR:
1290 return true;
1291 case 0:
1292 break;
1293 }
Ian Kent62a73752011-03-25 01:51:02 +08001294
1295 if (!d_mountpoint(path->dentry))
NeilBrownb8faf032014-08-04 17:06:29 +10001296 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
Ian Kent62a73752011-03-25 01:51:02 +08001297
Al Viro474279d2013-10-01 16:11:26 -04001298 mounted = __lookup_mnt(path->mnt, path->dentry);
David Howells9875cf82011-01-14 18:45:21 +00001299 if (!mounted)
1300 break;
Al Viroc7105362011-11-24 18:22:03 -05001301 path->mnt = &mounted->mnt;
1302 path->dentry = mounted->mnt.mnt_root;
Al Viroa3fbbde2011-11-07 21:21:26 +00001303 nd->flags |= LOOKUP_JUMPED;
Al Viro254cf582015-05-05 09:40:46 -04001304 *seqp = read_seqcount_begin(&path->dentry->d_seq);
Linus Torvalds59430262011-07-18 15:43:29 -07001305 /*
1306 * Update the inode too. We don't need to re-check the
1307 * dentry sequence number here after this d_inode read,
1308 * because a mount-point is always pinned.
1309 */
1310 *inode = path->dentry->d_inode;
David Howells9875cf82011-01-14 18:45:21 +00001311 }
Al Virof5be3e29122014-09-13 21:50:45 -04001312 return !read_seqretry(&mount_lock, nd->m_seq) &&
NeilBrownb8faf032014-08-04 17:06:29 +10001313 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
Al Viro287548e2011-05-27 06:50:06 -04001314}
1315
Nick Piggin31e6b012011-01-07 17:49:52 +11001316static int follow_dotdot_rcu(struct nameidata *nd)
1317{
Al Viro4023bfc2014-09-13 21:59:43 -04001318 struct inode *inode = nd->inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001319
David Howells9875cf82011-01-14 18:45:21 +00001320 while (1) {
Al Viroaed434a2015-05-12 12:22:47 -04001321 if (path_equal(&nd->path, &nd->root))
Nick Piggin31e6b012011-01-07 17:49:52 +11001322 break;
Nick Piggin31e6b012011-01-07 17:49:52 +11001323 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1324 struct dentry *old = nd->path.dentry;
1325 struct dentry *parent = old->d_parent;
1326 unsigned seq;
1327
Al Viro4023bfc2014-09-13 21:59:43 -04001328 inode = parent->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001329 seq = read_seqcount_begin(&parent->d_seq);
Al Viroaed434a2015-05-12 12:22:47 -04001330 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1331 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001332 nd->path.dentry = parent;
1333 nd->seq = seq;
Eric W. Biederman397d4252015-08-15 20:27:13 -05001334 if (unlikely(!path_connected(&nd->path)))
1335 return -ENOENT;
Nick Piggin31e6b012011-01-07 17:49:52 +11001336 break;
Al Viroaed434a2015-05-12 12:22:47 -04001337 } else {
1338 struct mount *mnt = real_mount(nd->path.mnt);
1339 struct mount *mparent = mnt->mnt_parent;
1340 struct dentry *mountpoint = mnt->mnt_mountpoint;
1341 struct inode *inode2 = mountpoint->d_inode;
1342 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1343 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1344 return -ECHILD;
1345 if (&mparent->mnt == nd->path.mnt)
1346 break;
1347 /* we know that mountpoint was pinned */
1348 nd->path.dentry = mountpoint;
1349 nd->path.mnt = &mparent->mnt;
1350 inode = inode2;
1351 nd->seq = seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11001352 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001353 }
Al Viroaed434a2015-05-12 12:22:47 -04001354 while (unlikely(d_mountpoint(nd->path.dentry))) {
Al Virob37199e2014-03-20 15:18:22 -04001355 struct mount *mounted;
1356 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
Al Viroaed434a2015-05-12 12:22:47 -04001357 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1358 return -ECHILD;
Al Virob37199e2014-03-20 15:18:22 -04001359 if (!mounted)
1360 break;
1361 nd->path.mnt = &mounted->mnt;
1362 nd->path.dentry = mounted->mnt.mnt_root;
Al Viro4023bfc2014-09-13 21:59:43 -04001363 inode = nd->path.dentry->d_inode;
Al Virob37199e2014-03-20 15:18:22 -04001364 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Virob37199e2014-03-20 15:18:22 -04001365 }
Al Viro4023bfc2014-09-13 21:59:43 -04001366 nd->inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001367 return 0;
1368}
1369
David Howells9875cf82011-01-14 18:45:21 +00001370/*
David Howellscc53ce52011-01-14 18:45:26 +00001371 * Follow down to the covering mount currently visible to userspace. At each
1372 * point, the filesystem owning that dentry may be queried as to whether the
1373 * caller is permitted to proceed or not.
David Howellscc53ce52011-01-14 18:45:26 +00001374 */
Al Viro7cc90cc2011-03-18 09:04:20 -04001375int follow_down(struct path *path)
David Howellscc53ce52011-01-14 18:45:26 +00001376{
1377 unsigned managed;
1378 int ret;
1379
1380 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1381 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1382 /* Allow the filesystem to manage the transit without i_mutex
1383 * being held.
1384 *
1385 * We indicate to the filesystem if someone is trying to mount
1386 * something here. This gives autofs the chance to deny anyone
1387 * other than its daemon the right to mount on its
1388 * superstructure.
1389 *
1390 * The filesystem may sleep at this point.
1391 */
1392 if (managed & DCACHE_MANAGE_TRANSIT) {
1393 BUG_ON(!path->dentry->d_op);
1394 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +00001395 ret = path->dentry->d_op->d_manage(
Al Viro1aed3e42011-03-18 09:09:02 -04001396 path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001397 if (ret < 0)
1398 return ret == -EISDIR ? 0 : ret;
1399 }
1400
1401 /* Transit to a mounted filesystem. */
1402 if (managed & DCACHE_MOUNTED) {
1403 struct vfsmount *mounted = lookup_mnt(path);
1404 if (!mounted)
1405 break;
1406 dput(path->dentry);
1407 mntput(path->mnt);
1408 path->mnt = mounted;
1409 path->dentry = dget(mounted->mnt_root);
1410 continue;
1411 }
1412
1413 /* Don't handle automount points here */
1414 break;
1415 }
1416 return 0;
1417}
Al Viro4d359502014-03-14 12:20:17 -04001418EXPORT_SYMBOL(follow_down);
David Howellscc53ce52011-01-14 18:45:26 +00001419
1420/*
David Howells9875cf82011-01-14 18:45:21 +00001421 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1422 */
1423static void follow_mount(struct path *path)
1424{
1425 while (d_mountpoint(path->dentry)) {
1426 struct vfsmount *mounted = lookup_mnt(path);
1427 if (!mounted)
1428 break;
1429 dput(path->dentry);
1430 mntput(path->mnt);
1431 path->mnt = mounted;
1432 path->dentry = dget(mounted->mnt_root);
1433 }
1434}
1435
Eric W. Biedermaneedf2652016-06-02 10:29:47 -05001436static int path_parent_directory(struct path *path)
1437{
1438 struct dentry *old = path->dentry;
1439 /* rare case of legitimate dget_parent()... */
1440 path->dentry = dget_parent(path->dentry);
1441 dput(old);
1442 if (unlikely(!path_connected(path)))
1443 return -ENOENT;
1444 return 0;
1445}
1446
Eric W. Biederman397d4252015-08-15 20:27:13 -05001447static int follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448{
1449 while(1) {
Al Viro2a737872009-04-07 11:49:53 -04001450 if (nd->path.dentry == nd->root.dentry &&
1451 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 break;
1453 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001454 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Eric W. Biedermaneedf2652016-06-02 10:29:47 -05001455 int ret = path_parent_directory(&nd->path);
1456 if (ret)
1457 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 break;
1459 }
Al Viro3088dd72010-01-30 15:47:29 -05001460 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 }
Al Viro79ed0222009-04-18 13:59:41 -04001463 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001464 nd->inode = nd->path.dentry->d_inode;
Eric W. Biederman397d4252015-08-15 20:27:13 -05001465 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466}
1467
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468/*
Oleg Drokinf4fdace2016-07-07 22:04:04 -04001469 * This looks up the name in dcache and possibly revalidates the found dentry.
1470 * NULL is returned if the dentry does not exist in the cache.
Nick Pigginbaa03892010-08-18 04:37:31 +10001471 */
Al Viroe3c13922016-03-06 14:03:27 -05001472static struct dentry *lookup_dcache(const struct qstr *name,
1473 struct dentry *dir,
Al Viro6c51e512016-03-05 20:09:32 -05001474 unsigned int flags)
Nick Pigginbaa03892010-08-18 04:37:31 +10001475{
Nick Pigginbaa03892010-08-18 04:37:31 +10001476 struct dentry *dentry;
Miklos Szeredibad61182012-03-26 12:54:24 +02001477 int error;
Nick Pigginbaa03892010-08-18 04:37:31 +10001478
Miklos Szeredibad61182012-03-26 12:54:24 +02001479 dentry = d_lookup(dir, name);
1480 if (dentry) {
Jeff Layton39e3c952012-11-28 11:30:53 -05001481 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
Al Viro201f9562012-06-22 12:42:10 +04001482 error = d_revalidate(dentry, flags);
Miklos Szeredibad61182012-03-26 12:54:24 +02001483 if (unlikely(error <= 0)) {
Al Viro74ff0ff2016-03-05 22:37:46 -05001484 if (!error)
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001485 d_invalidate(dentry);
Al Viro74ff0ff2016-03-05 22:37:46 -05001486 dput(dentry);
1487 return ERR_PTR(error);
Miklos Szeredibad61182012-03-26 12:54:24 +02001488 }
1489 }
1490 }
Nick Pigginbaa03892010-08-18 04:37:31 +10001491 return dentry;
1492}
1493
1494/*
J. Bruce Fields13a2c3b2013-10-23 16:09:16 -04001495 * Call i_op->lookup on the dentry. The dentry must be negative and
1496 * unhashed.
Miklos Szeredibad61182012-03-26 12:54:24 +02001497 *
1498 * dir->d_inode->i_mutex must be held
Josef Bacik44396f42011-05-31 11:58:49 -04001499 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001500static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
Al Viro72bd8662012-06-10 17:17:17 -04001501 unsigned int flags)
Josef Bacik44396f42011-05-31 11:58:49 -04001502{
Josef Bacik44396f42011-05-31 11:58:49 -04001503 struct dentry *old;
1504
1505 /* Don't create child dentry for a dead directory. */
Miklos Szeredibad61182012-03-26 12:54:24 +02001506 if (unlikely(IS_DEADDIR(dir))) {
Miklos Szeredie188dc02012-02-03 14:25:18 +01001507 dput(dentry);
Josef Bacik44396f42011-05-31 11:58:49 -04001508 return ERR_PTR(-ENOENT);
Miklos Szeredie188dc02012-02-03 14:25:18 +01001509 }
Josef Bacik44396f42011-05-31 11:58:49 -04001510
Al Viro72bd8662012-06-10 17:17:17 -04001511 old = dir->i_op->lookup(dir, dentry, flags);
Josef Bacik44396f42011-05-31 11:58:49 -04001512 if (unlikely(old)) {
1513 dput(dentry);
1514 dentry = old;
1515 }
1516 return dentry;
1517}
1518
Al Viroe3c13922016-03-06 14:03:27 -05001519static struct dentry *__lookup_hash(const struct qstr *name,
Al Viro72bd8662012-06-10 17:17:17 -04001520 struct dentry *base, unsigned int flags)
Al Viroa3255542012-03-30 14:41:51 -04001521{
Al Viro6c51e512016-03-05 20:09:32 -05001522 struct dentry *dentry = lookup_dcache(name, base, flags);
Al Viroa3255542012-03-30 14:41:51 -04001523
Al Viro6c51e512016-03-05 20:09:32 -05001524 if (dentry)
Miklos Szeredibad61182012-03-26 12:54:24 +02001525 return dentry;
Al Viroa3255542012-03-30 14:41:51 -04001526
Al Viro6c51e512016-03-05 20:09:32 -05001527 dentry = d_alloc(base, name);
1528 if (unlikely(!dentry))
1529 return ERR_PTR(-ENOMEM);
1530
Al Viro72bd8662012-06-10 17:17:17 -04001531 return lookup_real(base->d_inode, dentry, flags);
Al Viroa3255542012-03-30 14:41:51 -04001532}
1533
Al Viroe97cdc82013-01-24 18:16:00 -05001534static int lookup_fast(struct nameidata *nd,
Al Viro254cf582015-05-05 09:40:46 -04001535 struct path *path, struct inode **inode,
1536 unsigned *seqp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537{
Jan Blunck4ac91372008-02-14 19:34:32 -08001538 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001539 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001540 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001541 int err;
1542
Al Viro3cac2602009-08-13 18:27:43 +04001543 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001544 * Rename seqlock is not required here because in the off chance
Al Viro5d0f49c2016-03-05 21:32:53 -05001545 * of a false negative due to a concurrent rename, the caller is
1546 * going to fall back to non-racy lookup.
Nick Pigginb04f7842010-08-18 04:37:34 +10001547 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001548 if (nd->flags & LOOKUP_RCU) {
1549 unsigned seq;
Al Viro766c4cb2015-05-07 19:24:57 -04001550 bool negative;
Linus Torvaldsda53be12013-05-21 15:22:44 -07001551 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
Al Viro5d0f49c2016-03-05 21:32:53 -05001552 if (unlikely(!dentry)) {
1553 if (unlazy_walk(nd, NULL, 0))
1554 return -ECHILD;
Al Viroe9742b52016-03-05 22:04:59 -05001555 return 0;
Al Viro5d0f49c2016-03-05 21:32:53 -05001556 }
Al Viro5a18fff2011-03-11 04:44:53 -05001557
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001558 /*
1559 * This sequence count validates that the inode matches
1560 * the dentry name information from lookup.
1561 */
David Howells63afdfc2015-05-06 15:59:00 +01001562 *inode = d_backing_inode(dentry);
Al Viro766c4cb2015-05-07 19:24:57 -04001563 negative = d_is_negative(dentry);
Al Viro5d0f49c2016-03-05 21:32:53 -05001564 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001565 return -ECHILD;
1566
1567 /*
1568 * This sequence count validates that the parent had no
1569 * changes while we did the lookup of the dentry above.
1570 *
1571 * The memory barrier in read_seqcount_begin of child is
1572 * enough, we can use __read_seqcount_retry here.
1573 */
Al Viro5d0f49c2016-03-05 21:32:53 -05001574 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
Nick Piggin31e6b012011-01-07 17:49:52 +11001575 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001576
Al Viro254cf582015-05-05 09:40:46 -04001577 *seqp = seq;
Al Viro5d0f49c2016-03-05 21:32:53 -05001578 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
Al Viro4ce16ef32012-06-10 16:10:59 -04001579 status = d_revalidate(dentry, nd->flags);
Al Viro5d0f49c2016-03-05 21:32:53 -05001580 if (unlikely(status <= 0)) {
1581 if (unlazy_walk(nd, dentry, seq))
1582 return -ECHILD;
1583 if (status == -ECHILD)
1584 status = d_revalidate(dentry, nd->flags);
1585 } else {
1586 /*
1587 * Note: do negative dentry check after revalidation in
1588 * case that drops it.
1589 */
1590 if (unlikely(negative))
1591 return -ENOENT;
1592 path->mnt = mnt;
1593 path->dentry = dentry;
1594 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
Al Viroe9742b52016-03-05 22:04:59 -05001595 return 1;
Al Viro5d0f49c2016-03-05 21:32:53 -05001596 if (unlazy_walk(nd, dentry, seq))
1597 return -ECHILD;
Al Viro24643082011-02-15 01:26:22 -05001598 }
Al Viro5a18fff2011-03-11 04:44:53 -05001599 } else {
Al Viroe97cdc82013-01-24 18:16:00 -05001600 dentry = __d_lookup(parent, &nd->last);
Al Viro5d0f49c2016-03-05 21:32:53 -05001601 if (unlikely(!dentry))
Al Viroe9742b52016-03-05 22:04:59 -05001602 return 0;
Al Viro5d0f49c2016-03-05 21:32:53 -05001603 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
1604 status = d_revalidate(dentry, nd->flags);
Nick Piggin31e6b012011-01-07 17:49:52 +11001605 }
Al Viro5a18fff2011-03-11 04:44:53 -05001606 if (unlikely(status <= 0)) {
Al Viroe9742b52016-03-05 22:04:59 -05001607 if (!status)
Al Viro5d0f49c2016-03-05 21:32:53 -05001608 d_invalidate(dentry);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001609 dput(dentry);
Al Viro5d0f49c2016-03-05 21:32:53 -05001610 return status;
Al Viro5a18fff2011-03-11 04:44:53 -05001611 }
Al Viro766c4cb2015-05-07 19:24:57 -04001612 if (unlikely(d_is_negative(dentry))) {
1613 dput(dentry);
1614 return -ENOENT;
1615 }
Al Viro5d0f49c2016-03-05 21:32:53 -05001616
David Howells9875cf82011-01-14 18:45:21 +00001617 path->mnt = mnt;
1618 path->dentry = dentry;
NeilBrown756daf22015-03-23 13:37:38 +11001619 err = follow_managed(path, nd);
Al Viroe9742b52016-03-05 22:04:59 -05001620 if (likely(err > 0))
David Howells63afdfc2015-05-06 15:59:00 +01001621 *inode = d_backing_inode(path->dentry);
Al Viro84027522015-04-22 10:30:08 -04001622 return err;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001623}
1624
1625/* Fast lookup failed, do it the slow way */
Al Viroe3c13922016-03-06 14:03:27 -05001626static struct dentry *lookup_slow(const struct qstr *name,
1627 struct dentry *dir,
1628 unsigned int flags)
Miklos Szeredi697f5142012-05-21 17:30:05 +02001629{
Al Viro94bdd652016-04-15 02:42:04 -04001630 struct dentry *dentry = ERR_PTR(-ENOENT), *old;
Al Viro19363862016-04-14 19:33:34 -04001631 struct inode *inode = dir->d_inode;
Al Virod9171b92016-04-15 03:33:13 -04001632 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
Al Viro19363862016-04-14 19:33:34 -04001633
Al Viro9902af72016-04-15 15:08:36 -04001634 inode_lock_shared(inode);
Al Viro19363862016-04-14 19:33:34 -04001635 /* Don't go there if it's already dead */
Al Viro94bdd652016-04-15 02:42:04 -04001636 if (unlikely(IS_DEADDIR(inode)))
1637 goto out;
1638again:
Al Virod9171b92016-04-15 03:33:13 -04001639 dentry = d_alloc_parallel(dir, name, &wq);
Al Viro94bdd652016-04-15 02:42:04 -04001640 if (IS_ERR(dentry))
1641 goto out;
1642 if (unlikely(!d_in_lookup(dentry))) {
Al Viro949a8522016-03-06 14:20:52 -05001643 if ((dentry->d_flags & DCACHE_OP_REVALIDATE) &&
1644 !(flags & LOOKUP_NO_REVAL)) {
1645 int error = d_revalidate(dentry, flags);
1646 if (unlikely(error <= 0)) {
Al Viro94bdd652016-04-15 02:42:04 -04001647 if (!error) {
Al Viro949a8522016-03-06 14:20:52 -05001648 d_invalidate(dentry);
Al Viro94bdd652016-04-15 02:42:04 -04001649 dput(dentry);
1650 goto again;
1651 }
Al Viro949a8522016-03-06 14:20:52 -05001652 dput(dentry);
1653 dentry = ERR_PTR(error);
1654 }
1655 }
Al Viro94bdd652016-04-15 02:42:04 -04001656 } else {
1657 old = inode->i_op->lookup(inode, dentry, flags);
1658 d_lookup_done(dentry);
1659 if (unlikely(old)) {
1660 dput(dentry);
1661 dentry = old;
Al Viro949a8522016-03-06 14:20:52 -05001662 }
1663 }
Al Viro94bdd652016-04-15 02:42:04 -04001664out:
Al Viro9902af72016-04-15 15:08:36 -04001665 inode_unlock_shared(inode);
Al Viroe3c13922016-03-06 14:03:27 -05001666 return dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667}
1668
Al Viro52094c82011-02-21 21:34:47 -05001669static inline int may_lookup(struct nameidata *nd)
1670{
1671 if (nd->flags & LOOKUP_RCU) {
Al Viro4ad5abb2011-06-20 19:57:03 -04001672 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
Al Viro52094c82011-02-21 21:34:47 -05001673 if (err != -ECHILD)
1674 return err;
Al Viro6e9918b2015-05-05 09:26:05 -04001675 if (unlazy_walk(nd, NULL, 0))
Al Viro52094c82011-02-21 21:34:47 -05001676 return -ECHILD;
1677 }
Al Viro4ad5abb2011-06-20 19:57:03 -04001678 return inode_permission(nd->inode, MAY_EXEC);
Al Viro52094c82011-02-21 21:34:47 -05001679}
1680
Al Viro9856fa12011-03-04 14:22:06 -05001681static inline int handle_dots(struct nameidata *nd, int type)
1682{
1683 if (type == LAST_DOTDOT) {
Al Viro9e6697e2015-12-05 20:07:21 -05001684 if (!nd->root.mnt)
1685 set_root(nd);
Al Viro9856fa12011-03-04 14:22:06 -05001686 if (nd->flags & LOOKUP_RCU) {
Al Viro70291ae2015-05-04 07:53:00 -04001687 return follow_dotdot_rcu(nd);
Al Viro9856fa12011-03-04 14:22:06 -05001688 } else
Eric W. Biederman397d4252015-08-15 20:27:13 -05001689 return follow_dotdot(nd);
Al Viro9856fa12011-03-04 14:22:06 -05001690 }
1691 return 0;
1692}
1693
Al Viro181548c2015-05-07 19:54:34 -04001694static int pick_link(struct nameidata *nd, struct path *link,
1695 struct inode *inode, unsigned seq)
Al Virod63ff282015-05-04 18:13:23 -04001696{
Al Viro626de992015-05-04 18:26:59 -04001697 int error;
Al Viro1cf26652015-05-06 16:01:56 -04001698 struct saved *last;
NeilBrown756daf22015-03-23 13:37:38 +11001699 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
Al Viro626de992015-05-04 18:26:59 -04001700 path_to_nameidata(link, nd);
1701 return -ELOOP;
1702 }
Al Virobc40aee2015-05-09 13:04:24 -04001703 if (!(nd->flags & LOOKUP_RCU)) {
Al Viro79733872015-05-09 12:55:43 -04001704 if (link->mnt == nd->path.mnt)
1705 mntget(link->mnt);
Al Virod63ff282015-05-04 18:13:23 -04001706 }
Al Viro626de992015-05-04 18:26:59 -04001707 error = nd_alloc_stack(nd);
1708 if (unlikely(error)) {
Al Virobc40aee2015-05-09 13:04:24 -04001709 if (error == -ECHILD) {
1710 if (unlikely(unlazy_link(nd, link, seq)))
1711 return -ECHILD;
1712 error = nd_alloc_stack(nd);
1713 }
1714 if (error) {
1715 path_put(link);
1716 return error;
1717 }
Al Viro626de992015-05-04 18:26:59 -04001718 }
1719
Al Viroab104922015-05-10 11:50:01 -04001720 last = nd->stack + nd->depth++;
Al Viro1cf26652015-05-06 16:01:56 -04001721 last->link = *link;
Al Virofceef392015-12-29 15:58:39 -05001722 clear_delayed_call(&last->done);
1723 nd->link_inode = inode;
Al Viro0450b2d2015-05-08 13:23:53 -04001724 last->seq = seq;
Al Virod63ff282015-05-04 18:13:23 -04001725 return 1;
1726}
1727
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001728/*
1729 * Do we need to follow links? We _really_ want to be able
1730 * to do this check without having to look at inode->i_op,
1731 * so we keep a cache of "no, this doesn't need follow_link"
1732 * for the common case.
1733 */
Al Viro254cf582015-05-05 09:40:46 -04001734static inline int should_follow_link(struct nameidata *nd, struct path *link,
Al Viro181548c2015-05-07 19:54:34 -04001735 int follow,
1736 struct inode *inode, unsigned seq)
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001737{
Al Virod63ff282015-05-04 18:13:23 -04001738 if (likely(!d_is_symlink(link->dentry)))
1739 return 0;
1740 if (!follow)
1741 return 0;
Al Viroa7f77542016-02-27 19:31:01 -05001742 /* make sure that d_is_symlink above matches inode */
1743 if (nd->flags & LOOKUP_RCU) {
1744 if (read_seqcount_retry(&link->dentry->d_seq, seq))
1745 return -ECHILD;
1746 }
Al Viro181548c2015-05-07 19:54:34 -04001747 return pick_link(nd, link, inode, seq);
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001748}
1749
Al Viro4693a542015-05-04 17:47:11 -04001750enum {WALK_GET = 1, WALK_PUT = 2};
1751
1752static int walk_component(struct nameidata *nd, int flags)
Al Viroce57dfc2011-03-13 19:58:58 -04001753{
Al Virocaa85632015-04-22 17:52:47 -04001754 struct path path;
Al Viroce57dfc2011-03-13 19:58:58 -04001755 struct inode *inode;
Al Viro254cf582015-05-05 09:40:46 -04001756 unsigned seq;
Al Viroce57dfc2011-03-13 19:58:58 -04001757 int err;
1758 /*
1759 * "." and ".." are special - ".." especially so because it has
1760 * to be able to know about the current root directory and
1761 * parent relationships.
1762 */
Al Viro4693a542015-05-04 17:47:11 -04001763 if (unlikely(nd->last_type != LAST_NORM)) {
1764 err = handle_dots(nd, nd->last_type);
1765 if (flags & WALK_PUT)
1766 put_link(nd);
1767 return err;
1768 }
Al Viro254cf582015-05-05 09:40:46 -04001769 err = lookup_fast(nd, &path, &inode, &seq);
Al Viroe9742b52016-03-05 22:04:59 -05001770 if (unlikely(err <= 0)) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001771 if (err < 0)
Al Virof0a9ba72015-05-04 07:59:30 -04001772 return err;
Al Viroe3c13922016-03-06 14:03:27 -05001773 path.dentry = lookup_slow(&nd->last, nd->path.dentry,
1774 nd->flags);
1775 if (IS_ERR(path.dentry))
1776 return PTR_ERR(path.dentry);
Al Viro7500c382016-03-31 00:23:05 -04001777
Al Viroe3c13922016-03-06 14:03:27 -05001778 path.mnt = nd->path.mnt;
1779 err = follow_managed(&path, nd);
1780 if (unlikely(err < 0))
Al Virof0a9ba72015-05-04 07:59:30 -04001781 return err;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001782
Al Viro7500c382016-03-31 00:23:05 -04001783 if (unlikely(d_is_negative(path.dentry))) {
1784 path_to_nameidata(&path, nd);
1785 return -ENOENT;
1786 }
1787
Al Viro254cf582015-05-05 09:40:46 -04001788 seq = 0; /* we are already out of RCU mode */
Al Virod4565642016-02-27 19:23:16 -05001789 inode = d_backing_inode(path.dentry);
Al Viroce57dfc2011-03-13 19:58:58 -04001790 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001791
Al Viro4693a542015-05-04 17:47:11 -04001792 if (flags & WALK_PUT)
1793 put_link(nd);
Al Viro181548c2015-05-07 19:54:34 -04001794 err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq);
Al Virod63ff282015-05-04 18:13:23 -04001795 if (unlikely(err))
1796 return err;
Al Virocaa85632015-04-22 17:52:47 -04001797 path_to_nameidata(&path, nd);
Al Viroce57dfc2011-03-13 19:58:58 -04001798 nd->inode = inode;
Al Viro254cf582015-05-05 09:40:46 -04001799 nd->seq = seq;
Al Viroce57dfc2011-03-13 19:58:58 -04001800 return 0;
1801}
1802
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803/*
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001804 * We can do the critical dentry name comparison and hashing
1805 * operations one word at a time, but we are limited to:
1806 *
1807 * - Architectures with fast unaligned word accesses. We could
1808 * do a "get_unaligned()" if this helps and is sufficiently
1809 * fast.
1810 *
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001811 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1812 * do not trap on the (extremely unlikely) case of a page
1813 * crossing operation.
1814 *
1815 * - Furthermore, we need an efficient 64-bit compile for the
1816 * 64-bit case in order to generate the "number of bytes in
1817 * the final mask". Again, that could be replaced with a
1818 * efficient population count instruction or similar.
1819 */
1820#ifdef CONFIG_DCACHE_WORD_ACCESS
1821
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001822#include <asm/word-at-a-time.h>
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001823
George Spelvin468a9422016-05-26 22:11:51 -04001824#ifdef HASH_MIX
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001825
George Spelvin468a9422016-05-26 22:11:51 -04001826/* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1827
1828#elif defined(CONFIG_64BIT)
George Spelvin2a18da7a2016-05-23 07:43:58 -04001829/*
1830 * Register pressure in the mixing function is an issue, particularly
1831 * on 32-bit x86, but almost any function requires one state value and
1832 * one temporary. Instead, use a function designed for two state values
1833 * and no temporaries.
1834 *
1835 * This function cannot create a collision in only two iterations, so
1836 * we have two iterations to achieve avalanche. In those two iterations,
1837 * we have six layers of mixing, which is enough to spread one bit's
1838 * influence out to 2^6 = 64 state bits.
1839 *
1840 * Rotate constants are scored by considering either 64 one-bit input
1841 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1842 * probability of that delta causing a change to each of the 128 output
1843 * bits, using a sample of random initial states.
1844 *
1845 * The Shannon entropy of the computed probabilities is then summed
1846 * to produce a score. Ideally, any input change has a 50% chance of
1847 * toggling any given output bit.
1848 *
1849 * Mixing scores (in bits) for (12,45):
1850 * Input delta: 1-bit 2-bit
1851 * 1 round: 713.3 42542.6
1852 * 2 rounds: 2753.7 140389.8
1853 * 3 rounds: 5954.1 233458.2
1854 * 4 rounds: 7862.6 256672.2
1855 * Perfect: 8192 258048
1856 * (64*128) (64*63/2 * 128)
1857 */
1858#define HASH_MIX(x, y, a) \
1859 ( x ^= (a), \
1860 y ^= x, x = rol64(x,12),\
1861 x += y, y = rol64(y,45),\
1862 y *= 9 )
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001863
George Spelvin0fed3ac2016-05-02 06:31:01 -04001864/*
George Spelvin2a18da7a2016-05-23 07:43:58 -04001865 * Fold two longs into one 32-bit hash value. This must be fast, but
1866 * latency isn't quite as critical, as there is a fair bit of additional
1867 * work done before the hash value is used.
George Spelvin0fed3ac2016-05-02 06:31:01 -04001868 */
George Spelvin2a18da7a2016-05-23 07:43:58 -04001869static inline unsigned int fold_hash(unsigned long x, unsigned long y)
George Spelvin0fed3ac2016-05-02 06:31:01 -04001870{
George Spelvin2a18da7a2016-05-23 07:43:58 -04001871 y ^= x * GOLDEN_RATIO_64;
1872 y *= GOLDEN_RATIO_64;
1873 return y >> 32;
George Spelvin0fed3ac2016-05-02 06:31:01 -04001874}
1875
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001876#else /* 32-bit case */
1877
George Spelvin2a18da7a2016-05-23 07:43:58 -04001878/*
1879 * Mixing scores (in bits) for (7,20):
1880 * Input delta: 1-bit 2-bit
1881 * 1 round: 330.3 9201.6
1882 * 2 rounds: 1246.4 25475.4
1883 * 3 rounds: 1907.1 31295.1
1884 * 4 rounds: 2042.3 31718.6
1885 * Perfect: 2048 31744
1886 * (32*64) (32*31/2 * 64)
1887 */
1888#define HASH_MIX(x, y, a) \
1889 ( x ^= (a), \
1890 y ^= x, x = rol32(x, 7),\
1891 x += y, y = rol32(y,20),\
1892 y *= 9 )
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001893
George Spelvin2a18da7a2016-05-23 07:43:58 -04001894static inline unsigned int fold_hash(unsigned long x, unsigned long y)
George Spelvin0fed3ac2016-05-02 06:31:01 -04001895{
George Spelvin2a18da7a2016-05-23 07:43:58 -04001896 /* Use arch-optimized multiply if one exists */
1897 return __hash_32(y ^ __hash_32(x));
George Spelvin0fed3ac2016-05-02 06:31:01 -04001898}
1899
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001900#endif
1901
George Spelvin2a18da7a2016-05-23 07:43:58 -04001902/*
1903 * Return the hash of a string of known length. This is carfully
1904 * designed to match hash_name(), which is the more critical function.
1905 * In particular, we must end by hashing a final word containing 0..7
1906 * payload bytes, to match the way that hash_name() iterates until it
1907 * finds the delimiter after the name.
1908 */
Linus Torvalds8387ff22016-06-10 07:51:30 -07001909unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001910{
Linus Torvalds8387ff22016-06-10 07:51:30 -07001911 unsigned long a, x = 0, y = (unsigned long)salt;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001912
1913 for (;;) {
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001914 if (!len)
1915 goto done;
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001916 a = load_unaligned_zeropad(name);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001917 if (len < sizeof(unsigned long))
1918 break;
George Spelvin2a18da7a2016-05-23 07:43:58 -04001919 HASH_MIX(x, y, a);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001920 name += sizeof(unsigned long);
1921 len -= sizeof(unsigned long);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001922 }
George Spelvin2a18da7a2016-05-23 07:43:58 -04001923 x ^= a & bytemask_from_count(len);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001924done:
George Spelvin2a18da7a2016-05-23 07:43:58 -04001925 return fold_hash(x, y);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001926}
1927EXPORT_SYMBOL(full_name_hash);
1928
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001929/* Return the "hash_len" (hash and length) of a null-terminated string */
Linus Torvalds8387ff22016-06-10 07:51:30 -07001930u64 hashlen_string(const void *salt, const char *name)
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001931{
Linus Torvalds8387ff22016-06-10 07:51:30 -07001932 unsigned long a = 0, x = 0, y = (unsigned long)salt;
1933 unsigned long adata, mask, len;
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001934 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1935
Linus Torvalds8387ff22016-06-10 07:51:30 -07001936 len = 0;
1937 goto inside;
1938
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001939 do {
George Spelvin2a18da7a2016-05-23 07:43:58 -04001940 HASH_MIX(x, y, a);
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001941 len += sizeof(unsigned long);
Linus Torvalds8387ff22016-06-10 07:51:30 -07001942inside:
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001943 a = load_unaligned_zeropad(name+len);
1944 } while (!has_zero(a, &adata, &constants));
1945
1946 adata = prep_zero_mask(a, adata, &constants);
1947 mask = create_zero_mask(adata);
George Spelvin2a18da7a2016-05-23 07:43:58 -04001948 x ^= a & zero_bytemask(mask);
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001949
George Spelvin2a18da7a2016-05-23 07:43:58 -04001950 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001951}
1952EXPORT_SYMBOL(hashlen_string);
1953
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001954/*
1955 * Calculate the length and hash of the path component, and
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001956 * return the "hash_len" as the result.
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001957 */
Linus Torvalds8387ff22016-06-10 07:51:30 -07001958static inline u64 hash_name(const void *salt, const char *name)
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001959{
Linus Torvalds8387ff22016-06-10 07:51:30 -07001960 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
1961 unsigned long adata, bdata, mask, len;
Linus Torvalds36126f82012-05-26 10:43:17 -07001962 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001963
Linus Torvalds8387ff22016-06-10 07:51:30 -07001964 len = 0;
1965 goto inside;
1966
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001967 do {
George Spelvin2a18da7a2016-05-23 07:43:58 -04001968 HASH_MIX(x, y, a);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001969 len += sizeof(unsigned long);
Linus Torvalds8387ff22016-06-10 07:51:30 -07001970inside:
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001971 a = load_unaligned_zeropad(name+len);
Linus Torvalds36126f82012-05-26 10:43:17 -07001972 b = a ^ REPEAT_BYTE('/');
1973 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001974
Linus Torvalds36126f82012-05-26 10:43:17 -07001975 adata = prep_zero_mask(a, adata, &constants);
1976 bdata = prep_zero_mask(b, bdata, &constants);
Linus Torvalds36126f82012-05-26 10:43:17 -07001977 mask = create_zero_mask(adata | bdata);
George Spelvin2a18da7a2016-05-23 07:43:58 -04001978 x ^= a & zero_bytemask(mask);
Linus Torvalds36126f82012-05-26 10:43:17 -07001979
George Spelvin2a18da7a2016-05-23 07:43:58 -04001980 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001981}
1982
George Spelvin2a18da7a2016-05-23 07:43:58 -04001983#else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001984
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001985/* Return the hash of a string of known length */
Linus Torvalds8387ff22016-06-10 07:51:30 -07001986unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
Linus Torvalds0145acc2012-03-02 14:32:59 -08001987{
Linus Torvalds8387ff22016-06-10 07:51:30 -07001988 unsigned long hash = init_name_hash(salt);
Linus Torvalds0145acc2012-03-02 14:32:59 -08001989 while (len--)
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001990 hash = partial_name_hash((unsigned char)*name++, hash);
Linus Torvalds0145acc2012-03-02 14:32:59 -08001991 return end_name_hash(hash);
1992}
Linus Torvaldsae942ae2012-03-02 19:40:57 -08001993EXPORT_SYMBOL(full_name_hash);
Linus Torvalds0145acc2012-03-02 14:32:59 -08001994
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001995/* Return the "hash_len" (hash and length) of a null-terminated string */
Linus Torvalds8387ff22016-06-10 07:51:30 -07001996u64 hashlen_string(const void *salt, const char *name)
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001997{
Linus Torvalds8387ff22016-06-10 07:51:30 -07001998 unsigned long hash = init_name_hash(salt);
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001999 unsigned long len = 0, c;
2000
2001 c = (unsigned char)*name;
George Spelvine0ab7af2016-05-29 08:05:56 -04002002 while (c) {
George Spelvinfcfd2fb2016-05-20 08:41:37 -04002003 len++;
2004 hash = partial_name_hash(c, hash);
2005 c = (unsigned char)name[len];
George Spelvine0ab7af2016-05-29 08:05:56 -04002006 }
George Spelvinfcfd2fb2016-05-20 08:41:37 -04002007 return hashlen_create(end_name_hash(hash), len);
2008}
George Spelvinf2a031b2016-05-29 01:26:41 -04002009EXPORT_SYMBOL(hashlen_string);
George Spelvinfcfd2fb2016-05-20 08:41:37 -04002010
Linus Torvalds3ddcd052011-08-06 22:45:50 -07002011/*
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002012 * We know there's a real path component here of at least
2013 * one character.
2014 */
Linus Torvalds8387ff22016-06-10 07:51:30 -07002015static inline u64 hash_name(const void *salt, const char *name)
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002016{
Linus Torvalds8387ff22016-06-10 07:51:30 -07002017 unsigned long hash = init_name_hash(salt);
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002018 unsigned long len = 0, c;
2019
2020 c = (unsigned char)*name;
2021 do {
2022 len++;
2023 hash = partial_name_hash(c, hash);
2024 c = (unsigned char)name[len];
2025 } while (c && c != '/');
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002026 return hashlen_create(end_name_hash(hash), len);
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002027}
2028
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08002029#endif
2030
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002031/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01002033 * This is the basic name resolution function, turning a pathname into
2034 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01002036 * Returns 0 and nd will have valid dentry and mnt on success.
2037 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 */
Al Viro6de88d72009-08-09 01:41:57 +04002039static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 int err;
Al Viro32cd7462015-04-18 20:30:49 -04002042
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 while (*name=='/')
2044 name++;
2045 if (!*name)
Al Viro9e18f102015-04-18 20:44:34 -04002046 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 /* At this point we know we have a real path component. */
2049 for(;;) {
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002050 u64 hash_len;
Al Virofe479a52011-02-22 15:10:03 -05002051 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052
Al Viro52094c82011-02-21 21:34:47 -05002053 err = may_lookup(nd);
George Spelvin2a18da7a2016-05-23 07:43:58 -04002054 if (err)
Al Viro3595e232015-05-09 16:54:45 -04002055 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056
Linus Torvalds8387ff22016-06-10 07:51:30 -07002057 hash_len = hash_name(nd->path.dentry, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058
Al Virofe479a52011-02-22 15:10:03 -05002059 type = LAST_NORM;
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002060 if (name[0] == '.') switch (hashlen_len(hash_len)) {
Al Virofe479a52011-02-22 15:10:03 -05002061 case 2:
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002062 if (name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05002063 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05002064 nd->flags |= LOOKUP_JUMPED;
2065 }
Al Virofe479a52011-02-22 15:10:03 -05002066 break;
2067 case 1:
2068 type = LAST_DOT;
2069 }
Al Viro5a202bc2011-03-08 14:17:44 -05002070 if (likely(type == LAST_NORM)) {
2071 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05002072 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05002073 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
James Hogana060dc52014-09-16 13:07:35 +01002074 struct qstr this = { { .hash_len = hash_len }, .name = name };
Linus Torvaldsda53be12013-05-21 15:22:44 -07002075 err = parent->d_op->d_hash(parent, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05002076 if (err < 0)
Al Viro3595e232015-05-09 16:54:45 -04002077 return err;
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002078 hash_len = this.hash_len;
2079 name = this.name;
Al Viro5a202bc2011-03-08 14:17:44 -05002080 }
2081 }
Al Virofe479a52011-02-22 15:10:03 -05002082
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002083 nd->last.hash_len = hash_len;
2084 nd->last.name = name;
Al Viro5f4a6a62013-01-24 18:04:22 -05002085 nd->last_type = type;
2086
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002087 name += hashlen_len(hash_len);
2088 if (!*name)
Al Virobdf6cbf2015-04-18 20:21:40 -04002089 goto OK;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002090 /*
2091 * If it wasn't NUL, we know it was '/'. Skip that
2092 * slash, and continue until no more slashes.
2093 */
2094 do {
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002095 name++;
2096 } while (unlikely(*name == '/'));
Al Viro8620c232015-05-04 08:58:35 -04002097 if (unlikely(!*name)) {
2098OK:
Al Viro368ee9b2015-05-08 17:19:59 -04002099 /* pathname body, done */
Al Viro8620c232015-05-04 08:58:35 -04002100 if (!nd->depth)
2101 return 0;
2102 name = nd->stack[nd->depth - 1].name;
Al Viro368ee9b2015-05-08 17:19:59 -04002103 /* trailing symlink, done */
Al Viro8620c232015-05-04 08:58:35 -04002104 if (!name)
2105 return 0;
2106 /* last component of nested symlink */
Al Viro4693a542015-05-04 17:47:11 -04002107 err = walk_component(nd, WALK_GET | WALK_PUT);
Al Viro8620c232015-05-04 08:58:35 -04002108 } else {
Al Viro4693a542015-05-04 17:47:11 -04002109 err = walk_component(nd, WALK_GET);
Al Viro8620c232015-05-04 08:58:35 -04002110 }
Al Viroce57dfc2011-03-13 19:58:58 -04002111 if (err < 0)
Al Viro3595e232015-05-09 16:54:45 -04002112 return err;
Al Virofe479a52011-02-22 15:10:03 -05002113
Al Viroce57dfc2011-03-13 19:58:58 -04002114 if (err) {
Al Viro626de992015-05-04 18:26:59 -04002115 const char *s = get_link(nd);
Al Viro5a460272015-04-17 23:44:45 -04002116
Viresh Kumara1c83682015-08-12 15:59:44 +05302117 if (IS_ERR(s))
Al Viro3595e232015-05-09 16:54:45 -04002118 return PTR_ERR(s);
Al Virod40bcc02015-04-18 20:03:03 -04002119 err = 0;
2120 if (unlikely(!s)) {
2121 /* jumped */
Al Virob9ff4422015-05-02 20:19:23 -04002122 put_link(nd);
Al Virod40bcc02015-04-18 20:03:03 -04002123 } else {
Al Virofab51e82015-05-10 11:01:00 -04002124 nd->stack[nd->depth - 1].name = name;
2125 name = s;
2126 continue;
Al Virod40bcc02015-04-18 20:03:03 -04002127 }
Nick Piggin31e6b012011-01-07 17:49:52 +11002128 }
Al Viro97242f92015-08-01 19:59:28 -04002129 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2130 if (nd->flags & LOOKUP_RCU) {
2131 if (unlazy_walk(nd, NULL, 0))
2132 return -ECHILD;
2133 }
Al Viro3595e232015-05-09 16:54:45 -04002134 return -ENOTDIR;
Al Viro97242f92015-08-01 19:59:28 -04002135 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137}
2138
Al Viroc8a53ee2015-05-12 18:43:07 -04002139static const char *path_init(struct nameidata *nd, unsigned flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140{
Prasanna Medaea3834d2005-04-29 16:00:17 +01002141 int retval = 0;
Al Viroc8a53ee2015-05-12 18:43:07 -04002142 const char *s = nd->name->name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143
2144 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro980f3ea2014-11-20 14:20:24 -05002145 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146 nd->depth = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05002147 if (flags & LOOKUP_ROOT) {
David Howellsb18825a2013-09-12 19:22:53 +01002148 struct dentry *root = nd->root.dentry;
2149 struct inode *inode = root->d_inode;
Al Virofd2f7cb2015-02-22 20:07:13 -05002150 if (*s) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002151 if (!d_can_lookup(root))
Al Viro368ee9b2015-05-08 17:19:59 -04002152 return ERR_PTR(-ENOTDIR);
Al Viro73d049a2011-03-11 12:08:24 -05002153 retval = inode_permission(inode, MAY_EXEC);
2154 if (retval)
Al Viro368ee9b2015-05-08 17:19:59 -04002155 return ERR_PTR(retval);
Al Viro73d049a2011-03-11 12:08:24 -05002156 }
Al Viro5b6ca022011-03-09 23:04:47 -05002157 nd->path = nd->root;
2158 nd->inode = inode;
2159 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05002160 rcu_read_lock();
Al Viro5b6ca022011-03-09 23:04:47 -05002161 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro8f47a0162015-05-09 19:02:01 -04002162 nd->root_seq = nd->seq;
Al Viro48a066e2013-09-29 22:06:07 -04002163 nd->m_seq = read_seqbegin(&mount_lock);
Al Viro5b6ca022011-03-09 23:04:47 -05002164 } else {
2165 path_get(&nd->path);
2166 }
Al Viro368ee9b2015-05-08 17:19:59 -04002167 return s;
Al Viro5b6ca022011-03-09 23:04:47 -05002168 }
2169
Al Viro2a737872009-04-07 11:49:53 -04002170 nd->root.mnt = NULL;
Al Viro248fb5b2015-12-05 20:51:58 -05002171 nd->path.mnt = NULL;
2172 nd->path.dentry = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173
Al Viro48a066e2013-09-29 22:06:07 -04002174 nd->m_seq = read_seqbegin(&mount_lock);
Al Virofd2f7cb2015-02-22 20:07:13 -05002175 if (*s == '/') {
Al Viro9e6697e2015-12-05 20:07:21 -05002176 if (flags & LOOKUP_RCU)
Al Viro8b61e742013-11-08 12:45:01 -05002177 rcu_read_lock();
Al Viro9e6697e2015-12-05 20:07:21 -05002178 set_root(nd);
Al Viro248fb5b2015-12-05 20:51:58 -05002179 if (likely(!nd_jump_root(nd)))
Al Viroef55d912015-12-05 20:25:06 -05002180 return s;
Al Viro248fb5b2015-12-05 20:51:58 -05002181 nd->root.mnt = NULL;
Al Viroef55d912015-12-05 20:25:06 -05002182 rcu_read_unlock();
2183 return ERR_PTR(-ECHILD);
Al Viroc8a53ee2015-05-12 18:43:07 -04002184 } else if (nd->dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05002185 if (flags & LOOKUP_RCU) {
2186 struct fs_struct *fs = current->fs;
2187 unsigned seq;
2188
Al Viro8b61e742013-11-08 12:45:01 -05002189 rcu_read_lock();
Al Viroe41f7d42011-02-22 14:02:58 -05002190
2191 do {
2192 seq = read_seqcount_begin(&fs->seq);
2193 nd->path = fs->pwd;
Al Viroef55d912015-12-05 20:25:06 -05002194 nd->inode = nd->path.dentry->d_inode;
Al Viroe41f7d42011-02-22 14:02:58 -05002195 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2196 } while (read_seqcount_retry(&fs->seq, seq));
2197 } else {
2198 get_fs_pwd(current->fs, &nd->path);
Al Viroef55d912015-12-05 20:25:06 -05002199 nd->inode = nd->path.dentry->d_inode;
Al Viroe41f7d42011-02-22 14:02:58 -05002200 }
Al Viroef55d912015-12-05 20:25:06 -05002201 return s;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002202 } else {
Jeff Layton582aa642012-12-11 08:56:16 -05002203 /* Caller must check execute permissions on the starting path component */
Al Viroc8a53ee2015-05-12 18:43:07 -04002204 struct fd f = fdget_raw(nd->dfd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002205 struct dentry *dentry;
2206
Al Viro2903ff02012-08-28 12:52:22 -04002207 if (!f.file)
Al Viro368ee9b2015-05-08 17:19:59 -04002208 return ERR_PTR(-EBADF);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002209
Al Viro2903ff02012-08-28 12:52:22 -04002210 dentry = f.file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002211
Al Virofd2f7cb2015-02-22 20:07:13 -05002212 if (*s) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002213 if (!d_can_lookup(dentry)) {
Al Viro2903ff02012-08-28 12:52:22 -04002214 fdput(f);
Al Viro368ee9b2015-05-08 17:19:59 -04002215 return ERR_PTR(-ENOTDIR);
Al Viro2903ff02012-08-28 12:52:22 -04002216 }
Al Virof52e0c12011-03-14 18:56:51 -04002217 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002218
Al Viro2903ff02012-08-28 12:52:22 -04002219 nd->path = f.file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05002220 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05002221 rcu_read_lock();
Al Viro34a26b92015-05-11 08:05:05 -04002222 nd->inode = nd->path.dentry->d_inode;
2223 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viroe41f7d42011-02-22 14:02:58 -05002224 } else {
Al Viro2903ff02012-08-28 12:52:22 -04002225 path_get(&nd->path);
Al Viro34a26b92015-05-11 08:05:05 -04002226 nd->inode = nd->path.dentry->d_inode;
Al Viroe41f7d42011-02-22 14:02:58 -05002227 }
Al Viro34a26b92015-05-11 08:05:05 -04002228 fdput(f);
Al Viro368ee9b2015-05-08 17:19:59 -04002229 return s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230 }
Al Viro9b4a9b12009-04-07 11:44:16 -04002231}
2232
Al Viro3bdba282015-05-08 17:37:07 -04002233static const char *trailing_symlink(struct nameidata *nd)
Al Viro95fa25d2015-04-22 13:46:57 -04002234{
2235 const char *s;
Al Virofec2fa22015-05-06 15:58:18 -04002236 int error = may_follow_link(nd);
Al Virodeb106c2015-05-08 18:05:21 -04002237 if (unlikely(error))
Al Viro3bdba282015-05-08 17:37:07 -04002238 return ERR_PTR(error);
Al Viro95fa25d2015-04-22 13:46:57 -04002239 nd->flags |= LOOKUP_PARENT;
Al Virofab51e82015-05-10 11:01:00 -04002240 nd->stack[0].name = NULL;
Al Viro3b2e7f72015-04-19 00:53:50 -04002241 s = get_link(nd);
Al Virodeb106c2015-05-08 18:05:21 -04002242 return s ? s : "";
Al Viro95fa25d2015-04-22 13:46:57 -04002243}
2244
Al Virocaa85632015-04-22 17:52:47 -04002245static inline int lookup_last(struct nameidata *nd)
Al Virobd92d7f2011-03-14 19:54:59 -04002246{
2247 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2248 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2249
2250 nd->flags &= ~LOOKUP_PARENT;
Al Virodeb106c2015-05-08 18:05:21 -04002251 return walk_component(nd,
Al Viro4693a542015-05-04 17:47:11 -04002252 nd->flags & LOOKUP_FOLLOW
2253 ? nd->depth
2254 ? WALK_PUT | WALK_GET
2255 : WALK_GET
2256 : 0);
Al Virobd92d7f2011-03-14 19:54:59 -04002257}
2258
Al Viro9b4a9b12009-04-07 11:44:16 -04002259/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroc8a53ee2015-05-12 18:43:07 -04002260static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
Al Viro9b4a9b12009-04-07 11:44:16 -04002261{
Al Viroc8a53ee2015-05-12 18:43:07 -04002262 const char *s = path_init(nd, flags);
Al Virobd92d7f2011-03-14 19:54:59 -04002263 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11002264
Al Viro368ee9b2015-05-08 17:19:59 -04002265 if (IS_ERR(s))
2266 return PTR_ERR(s);
Al Viro3bdba282015-05-08 17:37:07 -04002267 while (!(err = link_path_walk(s, nd))
2268 && ((err = lookup_last(nd)) > 0)) {
2269 s = trailing_symlink(nd);
2270 if (IS_ERR(s)) {
2271 err = PTR_ERR(s);
2272 break;
Al Virobd92d7f2011-03-14 19:54:59 -04002273 }
2274 }
Al Viro9f1fafe2011-03-25 11:00:12 -04002275 if (!err)
2276 err = complete_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04002277
Al Virodeb106c2015-05-08 18:05:21 -04002278 if (!err && nd->flags & LOOKUP_DIRECTORY)
2279 if (!d_can_lookup(nd->path.dentry))
Al Virobd23a532011-03-23 09:56:30 -04002280 err = -ENOTDIR;
Al Viro625b6d12015-05-12 16:36:12 -04002281 if (!err) {
2282 *path = nd->path;
2283 nd->path.mnt = NULL;
2284 nd->path.dentry = NULL;
2285 }
2286 terminate_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04002287 return err;
Al Viroee0827c2011-02-21 23:38:09 -05002288}
Nick Piggin31e6b012011-01-07 17:49:52 +11002289
Al Viro625b6d12015-05-12 16:36:12 -04002290static int filename_lookup(int dfd, struct filename *name, unsigned flags,
Al Viro9ad1aaa2015-05-12 16:44:39 -04002291 struct path *path, struct path *root)
Jeff Layton873f1ee2012-10-10 15:25:29 -04002292{
Al Viro894bc8c2015-05-02 07:16:16 -04002293 int retval;
Al Viro9883d182015-05-13 07:28:08 -04002294 struct nameidata nd;
Al Viroabc9f5b2015-05-12 16:53:42 -04002295 if (IS_ERR(name))
2296 return PTR_ERR(name);
Al Viro9ad1aaa2015-05-12 16:44:39 -04002297 if (unlikely(root)) {
2298 nd.root = *root;
2299 flags |= LOOKUP_ROOT;
2300 }
Al Viro9883d182015-05-13 07:28:08 -04002301 set_nameidata(&nd, dfd, name);
Al Viroc8a53ee2015-05-12 18:43:07 -04002302 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002303 if (unlikely(retval == -ECHILD))
Al Viroc8a53ee2015-05-12 18:43:07 -04002304 retval = path_lookupat(&nd, flags, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002305 if (unlikely(retval == -ESTALE))
Al Viroc8a53ee2015-05-12 18:43:07 -04002306 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002307
2308 if (likely(!retval))
Al Viro625b6d12015-05-12 16:36:12 -04002309 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
Al Viro9883d182015-05-13 07:28:08 -04002310 restore_nameidata();
Al Viroe4bd1c12015-05-12 16:40:39 -04002311 putname(name);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002312 return retval;
2313}
2314
Al Viro8bcb77f2015-05-08 16:59:20 -04002315/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroc8a53ee2015-05-12 18:43:07 -04002316static int path_parentat(struct nameidata *nd, unsigned flags,
Al Viro391172c2015-05-09 11:19:16 -04002317 struct path *parent)
Al Viro8bcb77f2015-05-08 16:59:20 -04002318{
Al Viroc8a53ee2015-05-12 18:43:07 -04002319 const char *s = path_init(nd, flags);
Al Viro368ee9b2015-05-08 17:19:59 -04002320 int err;
2321 if (IS_ERR(s))
2322 return PTR_ERR(s);
2323 err = link_path_walk(s, nd);
Al Viro8bcb77f2015-05-08 16:59:20 -04002324 if (!err)
2325 err = complete_walk(nd);
Al Viro391172c2015-05-09 11:19:16 -04002326 if (!err) {
2327 *parent = nd->path;
2328 nd->path.mnt = NULL;
2329 nd->path.dentry = NULL;
2330 }
2331 terminate_walk(nd);
Al Viro8bcb77f2015-05-08 16:59:20 -04002332 return err;
2333}
2334
Al Viro5c31b6c2015-05-12 17:32:54 -04002335static struct filename *filename_parentat(int dfd, struct filename *name,
Al Viro391172c2015-05-09 11:19:16 -04002336 unsigned int flags, struct path *parent,
2337 struct qstr *last, int *type)
Al Viro8bcb77f2015-05-08 16:59:20 -04002338{
2339 int retval;
Al Viro9883d182015-05-13 07:28:08 -04002340 struct nameidata nd;
Al Viro8bcb77f2015-05-08 16:59:20 -04002341
Al Viro5c31b6c2015-05-12 17:32:54 -04002342 if (IS_ERR(name))
2343 return name;
Al Viro9883d182015-05-13 07:28:08 -04002344 set_nameidata(&nd, dfd, name);
Al Viroc8a53ee2015-05-12 18:43:07 -04002345 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
Al Viro8bcb77f2015-05-08 16:59:20 -04002346 if (unlikely(retval == -ECHILD))
Al Viroc8a53ee2015-05-12 18:43:07 -04002347 retval = path_parentat(&nd, flags, parent);
Al Viro8bcb77f2015-05-08 16:59:20 -04002348 if (unlikely(retval == -ESTALE))
Al Viroc8a53ee2015-05-12 18:43:07 -04002349 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
Al Viro391172c2015-05-09 11:19:16 -04002350 if (likely(!retval)) {
2351 *last = nd.last;
2352 *type = nd.last_type;
2353 audit_inode(name, parent->dentry, LOOKUP_PARENT);
Al Viro5c31b6c2015-05-12 17:32:54 -04002354 } else {
2355 putname(name);
2356 name = ERR_PTR(retval);
Al Viro391172c2015-05-09 11:19:16 -04002357 }
Al Viro9883d182015-05-13 07:28:08 -04002358 restore_nameidata();
Al Viro5c31b6c2015-05-12 17:32:54 -04002359 return name;
Al Viro8bcb77f2015-05-08 16:59:20 -04002360}
2361
Al Viro79714f72012-06-15 03:01:42 +04002362/* does lookup, returns the object with parent locked */
2363struct dentry *kern_path_locked(const char *name, struct path *path)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002364{
Al Viro5c31b6c2015-05-12 17:32:54 -04002365 struct filename *filename;
2366 struct dentry *d;
Al Viro391172c2015-05-09 11:19:16 -04002367 struct qstr last;
2368 int type;
Paul Moore51689102015-01-22 00:00:03 -05002369
Al Viro5c31b6c2015-05-12 17:32:54 -04002370 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2371 &last, &type);
Paul Moore51689102015-01-22 00:00:03 -05002372 if (IS_ERR(filename))
2373 return ERR_CAST(filename);
Al Viro5c31b6c2015-05-12 17:32:54 -04002374 if (unlikely(type != LAST_NORM)) {
Al Viro391172c2015-05-09 11:19:16 -04002375 path_put(path);
Al Viro5c31b6c2015-05-12 17:32:54 -04002376 putname(filename);
2377 return ERR_PTR(-EINVAL);
Al Viro79714f72012-06-15 03:01:42 +04002378 }
Al Viro59551022016-01-22 15:40:57 -05002379 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
Al Viro391172c2015-05-09 11:19:16 -04002380 d = __lookup_hash(&last, path->dentry, 0);
Al Viro79714f72012-06-15 03:01:42 +04002381 if (IS_ERR(d)) {
Al Viro59551022016-01-22 15:40:57 -05002382 inode_unlock(path->dentry->d_inode);
Al Viro391172c2015-05-09 11:19:16 -04002383 path_put(path);
Al Viro79714f72012-06-15 03:01:42 +04002384 }
Paul Moore51689102015-01-22 00:00:03 -05002385 putname(filename);
Al Viro79714f72012-06-15 03:01:42 +04002386 return d;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002387}
2388
Al Virod1811462008-08-02 00:49:18 -04002389int kern_path(const char *name, unsigned int flags, struct path *path)
2390{
Al Viroabc9f5b2015-05-12 16:53:42 -04002391 return filename_lookup(AT_FDCWD, getname_kernel(name),
2392 flags, path, NULL);
Al Virod1811462008-08-02 00:49:18 -04002393}
Al Viro4d359502014-03-14 12:20:17 -04002394EXPORT_SYMBOL(kern_path);
Al Virod1811462008-08-02 00:49:18 -04002395
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002396/**
2397 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2398 * @dentry: pointer to dentry of the base directory
2399 * @mnt: pointer to vfs mount of the base directory
2400 * @name: pointer to file name
2401 * @flags: lookup flags
Al Viroe0a01242011-06-27 17:00:37 -04002402 * @path: pointer to struct path to fill
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002403 */
2404int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2405 const char *name, unsigned int flags,
Al Viroe0a01242011-06-27 17:00:37 -04002406 struct path *path)
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002407{
Al Viro9ad1aaa2015-05-12 16:44:39 -04002408 struct path root = {.mnt = mnt, .dentry = dentry};
Al Viro9ad1aaa2015-05-12 16:44:39 -04002409 /* the first argument of filename_lookup() is ignored with root */
Al Viroabc9f5b2015-05-12 16:53:42 -04002410 return filename_lookup(AT_FDCWD, getname_kernel(name),
2411 flags , path, &root);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002412}
Al Viro4d359502014-03-14 12:20:17 -04002413EXPORT_SYMBOL(vfs_path_lookup);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002414
Christoph Hellwigeead1912007-10-16 23:25:38 -07002415/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07002416 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07002417 * @name: pathname component to lookup
2418 * @base: base directory to lookup from
2419 * @len: maximum length @len should be interpreted to
2420 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07002421 * Note that this routine is purely a helper for filesystem usage and should
Al Viro9e7543e2015-02-23 02:49:48 -05002422 * not be called by generic code.
NeilBrownbbddca82016-01-07 16:08:20 -05002423 *
2424 * The caller must hold base->i_mutex.
Christoph Hellwigeead1912007-10-16 23:25:38 -07002425 */
James Morris057f6c02007-04-26 00:12:05 -07002426struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2427{
James Morris057f6c02007-04-26 00:12:05 -07002428 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05002429 unsigned int c;
Miklos Szeredicda309d2012-03-26 12:54:21 +02002430 int err;
James Morris057f6c02007-04-26 00:12:05 -07002431
Al Viro59551022016-01-22 15:40:57 -05002432 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
David Woodhouse2f9092e2009-04-20 23:18:37 +01002433
Al Viro6a96ba52011-03-07 23:49:20 -05002434 this.name = name;
2435 this.len = len;
Linus Torvalds8387ff22016-06-10 07:51:30 -07002436 this.hash = full_name_hash(base, name, len);
Al Viro6a96ba52011-03-07 23:49:20 -05002437 if (!len)
2438 return ERR_PTR(-EACCES);
2439
Al Viro21d8a152012-11-29 22:17:21 -05002440 if (unlikely(name[0] == '.')) {
2441 if (len < 2 || (len == 2 && name[1] == '.'))
2442 return ERR_PTR(-EACCES);
2443 }
2444
Al Viro6a96ba52011-03-07 23:49:20 -05002445 while (len--) {
2446 c = *(const unsigned char *)name++;
2447 if (c == '/' || c == '\0')
2448 return ERR_PTR(-EACCES);
Al Viro6a96ba52011-03-07 23:49:20 -05002449 }
Al Viro5a202bc2011-03-08 14:17:44 -05002450 /*
2451 * See if the low-level filesystem might want
2452 * to use its own hash..
2453 */
2454 if (base->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002455 int err = base->d_op->d_hash(base, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05002456 if (err < 0)
2457 return ERR_PTR(err);
2458 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07002459
Miklos Szeredicda309d2012-03-26 12:54:21 +02002460 err = inode_permission(base->d_inode, MAY_EXEC);
2461 if (err)
2462 return ERR_PTR(err);
2463
Al Viro72bd8662012-06-10 17:17:17 -04002464 return __lookup_hash(&this, base, 0);
James Morris057f6c02007-04-26 00:12:05 -07002465}
Al Viro4d359502014-03-14 12:20:17 -04002466EXPORT_SYMBOL(lookup_one_len);
James Morris057f6c02007-04-26 00:12:05 -07002467
NeilBrownbbddca82016-01-07 16:08:20 -05002468/**
2469 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2470 * @name: pathname component to lookup
2471 * @base: base directory to lookup from
2472 * @len: maximum length @len should be interpreted to
2473 *
2474 * Note that this routine is purely a helper for filesystem usage and should
2475 * not be called by generic code.
2476 *
2477 * Unlike lookup_one_len, it should be called without the parent
2478 * i_mutex held, and will take the i_mutex itself if necessary.
2479 */
2480struct dentry *lookup_one_len_unlocked(const char *name,
2481 struct dentry *base, int len)
2482{
2483 struct qstr this;
2484 unsigned int c;
2485 int err;
Linus Torvalds20d00ee2016-07-29 12:17:52 -07002486 struct dentry *ret;
NeilBrownbbddca82016-01-07 16:08:20 -05002487
2488 this.name = name;
2489 this.len = len;
Linus Torvalds8387ff22016-06-10 07:51:30 -07002490 this.hash = full_name_hash(base, name, len);
NeilBrownbbddca82016-01-07 16:08:20 -05002491 if (!len)
2492 return ERR_PTR(-EACCES);
2493
2494 if (unlikely(name[0] == '.')) {
2495 if (len < 2 || (len == 2 && name[1] == '.'))
2496 return ERR_PTR(-EACCES);
2497 }
2498
2499 while (len--) {
2500 c = *(const unsigned char *)name++;
2501 if (c == '/' || c == '\0')
2502 return ERR_PTR(-EACCES);
2503 }
2504 /*
2505 * See if the low-level filesystem might want
2506 * to use its own hash..
2507 */
2508 if (base->d_flags & DCACHE_OP_HASH) {
2509 int err = base->d_op->d_hash(base, &this);
2510 if (err < 0)
2511 return ERR_PTR(err);
2512 }
2513
2514 err = inode_permission(base->d_inode, MAY_EXEC);
2515 if (err)
2516 return ERR_PTR(err);
2517
Linus Torvalds20d00ee2016-07-29 12:17:52 -07002518 ret = lookup_dcache(&this, base, 0);
2519 if (!ret)
2520 ret = lookup_slow(&this, base, 0);
2521 return ret;
NeilBrownbbddca82016-01-07 16:08:20 -05002522}
2523EXPORT_SYMBOL(lookup_one_len_unlocked);
2524
Eric W. Biedermaneedf2652016-06-02 10:29:47 -05002525#ifdef CONFIG_UNIX98_PTYS
2526int path_pts(struct path *path)
2527{
2528 /* Find something mounted on "pts" in the same directory as
2529 * the input path.
2530 */
2531 struct dentry *child, *parent;
2532 struct qstr this;
2533 int ret;
2534
2535 ret = path_parent_directory(path);
2536 if (ret)
2537 return ret;
2538
2539 parent = path->dentry;
2540 this.name = "pts";
2541 this.len = 3;
2542 child = d_hash_and_lookup(parent, &this);
2543 if (!child)
2544 return -ENOENT;
2545
2546 path->dentry = child;
2547 dput(parent);
2548 follow_mount(path);
2549 return 0;
2550}
2551#endif
2552
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002553int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2554 struct path *path, int *empty)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555{
Al Viroabc9f5b2015-05-12 16:53:42 -04002556 return filename_lookup(dfd, getname_flags(name, flags, empty),
2557 flags, path, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558}
Al Virob853a162015-05-13 09:12:02 -04002559EXPORT_SYMBOL(user_path_at_empty);
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002560
Jeff Layton873f1ee2012-10-10 15:25:29 -04002561/*
2562 * NB: most callers don't do anything directly with the reference to the
2563 * to struct filename, but the nd->last pointer points into the name string
2564 * allocated by getname. So we must hold the reference to it until all
2565 * path-walking is complete.
2566 */
Al Viroa2ec4a22015-05-13 06:57:49 -04002567static inline struct filename *
Al Virof5beed72015-04-30 16:09:11 -04002568user_path_parent(int dfd, const char __user *path,
2569 struct path *parent,
2570 struct qstr *last,
2571 int *type,
Jeff Layton9e790bd2012-12-11 12:10:09 -05002572 unsigned int flags)
Al Viro2ad94ae2008-07-21 09:32:51 -04002573{
Jeff Layton9e790bd2012-12-11 12:10:09 -05002574 /* only LOOKUP_REVAL is allowed in extra flags */
Al Viro5c31b6c2015-05-12 17:32:54 -04002575 return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL,
2576 parent, last, type);
Al Viro2ad94ae2008-07-21 09:32:51 -04002577}
2578
Jeff Layton80334262013-07-26 06:23:25 -04002579/**
Al Viro197df042013-09-08 14:03:27 -04002580 * mountpoint_last - look up last component for umount
Jeff Layton80334262013-07-26 06:23:25 -04002581 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2582 * @path: pointer to container for result
2583 *
2584 * This is a special lookup_last function just for umount. In this case, we
2585 * need to resolve the path without doing any revalidation.
2586 *
2587 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2588 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2589 * in almost all cases, this lookup will be served out of the dcache. The only
2590 * cases where it won't are if nd->last refers to a symlink or the path is
2591 * bogus and it doesn't exist.
2592 *
2593 * Returns:
2594 * -error: if there was an error during lookup. This includes -ENOENT if the
2595 * lookup found a negative dentry. The nd->path reference will also be
2596 * put in this case.
2597 *
2598 * 0: if we successfully resolved nd->path and found it to not to be a
2599 * symlink that needs to be followed. "path" will also be populated.
2600 * The nd->path reference will also be put.
2601 *
2602 * 1: if we successfully resolved nd->last and found it to be a symlink
2603 * that needs to be followed. "path" will be populated with the path
2604 * to the link, and nd->path will *not* be put.
2605 */
2606static int
Al Viro197df042013-09-08 14:03:27 -04002607mountpoint_last(struct nameidata *nd, struct path *path)
Jeff Layton80334262013-07-26 06:23:25 -04002608{
2609 int error = 0;
2610 struct dentry *dentry;
2611 struct dentry *dir = nd->path.dentry;
2612
Al Viro35759522013-09-08 13:41:33 -04002613 /* If we're in rcuwalk, drop out of it to handle last component */
2614 if (nd->flags & LOOKUP_RCU) {
Al Viro6e9918b2015-05-05 09:26:05 -04002615 if (unlazy_walk(nd, NULL, 0))
Al Virodeb106c2015-05-08 18:05:21 -04002616 return -ECHILD;
Jeff Layton80334262013-07-26 06:23:25 -04002617 }
2618
2619 nd->flags &= ~LOOKUP_PARENT;
2620
2621 if (unlikely(nd->last_type != LAST_NORM)) {
2622 error = handle_dots(nd, nd->last_type);
Al Viro35759522013-09-08 13:41:33 -04002623 if (error)
Al Virodeb106c2015-05-08 18:05:21 -04002624 return error;
Al Viro35759522013-09-08 13:41:33 -04002625 dentry = dget(nd->path.dentry);
Al Viro949a8522016-03-06 14:20:52 -05002626 } else {
2627 dentry = d_lookup(dir, &nd->last);
Jeff Layton80334262013-07-26 06:23:25 -04002628 if (!dentry) {
Al Viro949a8522016-03-06 14:20:52 -05002629 /*
2630 * No cached dentry. Mounted dentries are pinned in the
2631 * cache, so that means that this dentry is probably
2632 * a symlink or the path doesn't actually point
2633 * to a mounted dentry.
2634 */
2635 dentry = lookup_slow(&nd->last, dir,
2636 nd->flags | LOOKUP_NO_REVAL);
2637 if (IS_ERR(dentry))
2638 return PTR_ERR(dentry);
Dave Jonesbcceeeb2013-09-10 17:04:25 -04002639 }
Jeff Layton80334262013-07-26 06:23:25 -04002640 }
David Howells698934d2015-03-17 17:33:52 +00002641 if (d_is_negative(dentry)) {
Al Viro35759522013-09-08 13:41:33 -04002642 dput(dentry);
Al Virodeb106c2015-05-08 18:05:21 -04002643 return -ENOENT;
Jeff Layton80334262013-07-26 06:23:25 -04002644 }
Al Viro191d7f72015-05-04 08:26:45 -04002645 if (nd->depth)
2646 put_link(nd);
Al Viro35759522013-09-08 13:41:33 -04002647 path->dentry = dentry;
Vasily Averin295dc392014-07-21 12:30:23 +04002648 path->mnt = nd->path.mnt;
Al Viro181548c2015-05-07 19:54:34 -04002649 error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW,
2650 d_backing_inode(dentry), 0);
Al Virodeb106c2015-05-08 18:05:21 -04002651 if (unlikely(error))
Al Virod63ff282015-05-04 18:13:23 -04002652 return error;
Vasily Averin295dc392014-07-21 12:30:23 +04002653 mntget(path->mnt);
Al Viro35759522013-09-08 13:41:33 -04002654 follow_mount(path);
Al Virodeb106c2015-05-08 18:05:21 -04002655 return 0;
Jeff Layton80334262013-07-26 06:23:25 -04002656}
2657
2658/**
Al Viro197df042013-09-08 14:03:27 -04002659 * path_mountpoint - look up a path to be umounted
Masanari Iida2a78b852015-09-09 15:39:23 -07002660 * @nd: lookup context
Jeff Layton80334262013-07-26 06:23:25 -04002661 * @flags: lookup flags
Al Viroc8a53ee2015-05-12 18:43:07 -04002662 * @path: pointer to container for result
Jeff Layton80334262013-07-26 06:23:25 -04002663 *
2664 * Look up the given name, but don't attempt to revalidate the last component.
Randy Dunlap606d6fe2013-10-19 14:56:55 -07002665 * Returns 0 and "path" will be valid on success; Returns error otherwise.
Jeff Layton80334262013-07-26 06:23:25 -04002666 */
2667static int
Al Viroc8a53ee2015-05-12 18:43:07 -04002668path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
Jeff Layton80334262013-07-26 06:23:25 -04002669{
Al Viroc8a53ee2015-05-12 18:43:07 -04002670 const char *s = path_init(nd, flags);
Al Viro368ee9b2015-05-08 17:19:59 -04002671 int err;
2672 if (IS_ERR(s))
2673 return PTR_ERR(s);
Al Viro3bdba282015-05-08 17:37:07 -04002674 while (!(err = link_path_walk(s, nd)) &&
2675 (err = mountpoint_last(nd, path)) > 0) {
2676 s = trailing_symlink(nd);
2677 if (IS_ERR(s)) {
2678 err = PTR_ERR(s);
Jeff Layton80334262013-07-26 06:23:25 -04002679 break;
Al Viro3bdba282015-05-08 17:37:07 -04002680 }
Jeff Layton80334262013-07-26 06:23:25 -04002681 }
Al Virodeb106c2015-05-08 18:05:21 -04002682 terminate_walk(nd);
Jeff Layton80334262013-07-26 06:23:25 -04002683 return err;
2684}
2685
Al Viro2d864652013-09-08 20:18:44 -04002686static int
Al Viro668696d2015-02-22 19:44:00 -05002687filename_mountpoint(int dfd, struct filename *name, struct path *path,
Al Viro2d864652013-09-08 20:18:44 -04002688 unsigned int flags)
2689{
Al Viro9883d182015-05-13 07:28:08 -04002690 struct nameidata nd;
Al Virocbaab2d2015-01-22 02:49:00 -05002691 int error;
Al Viro668696d2015-02-22 19:44:00 -05002692 if (IS_ERR(name))
2693 return PTR_ERR(name);
Al Viro9883d182015-05-13 07:28:08 -04002694 set_nameidata(&nd, dfd, name);
Al Viroc8a53ee2015-05-12 18:43:07 -04002695 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
Al Viro2d864652013-09-08 20:18:44 -04002696 if (unlikely(error == -ECHILD))
Al Viroc8a53ee2015-05-12 18:43:07 -04002697 error = path_mountpoint(&nd, flags, path);
Al Viro2d864652013-09-08 20:18:44 -04002698 if (unlikely(error == -ESTALE))
Al Viroc8a53ee2015-05-12 18:43:07 -04002699 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
Al Viro2d864652013-09-08 20:18:44 -04002700 if (likely(!error))
Al Viro668696d2015-02-22 19:44:00 -05002701 audit_inode(name, path->dentry, 0);
Al Viro9883d182015-05-13 07:28:08 -04002702 restore_nameidata();
Al Viro668696d2015-02-22 19:44:00 -05002703 putname(name);
Al Viro2d864652013-09-08 20:18:44 -04002704 return error;
2705}
2706
Jeff Layton80334262013-07-26 06:23:25 -04002707/**
Al Viro197df042013-09-08 14:03:27 -04002708 * user_path_mountpoint_at - lookup a path from userland in order to umount it
Jeff Layton80334262013-07-26 06:23:25 -04002709 * @dfd: directory file descriptor
2710 * @name: pathname from userland
2711 * @flags: lookup flags
2712 * @path: pointer to container to hold result
2713 *
2714 * A umount is a special case for path walking. We're not actually interested
2715 * in the inode in this situation, and ESTALE errors can be a problem. We
2716 * simply want track down the dentry and vfsmount attached at the mountpoint
2717 * and avoid revalidating the last component.
2718 *
2719 * Returns 0 and populates "path" on success.
2720 */
2721int
Al Viro197df042013-09-08 14:03:27 -04002722user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
Jeff Layton80334262013-07-26 06:23:25 -04002723 struct path *path)
2724{
Al Virocbaab2d2015-01-22 02:49:00 -05002725 return filename_mountpoint(dfd, getname(name), path, flags);
Jeff Layton80334262013-07-26 06:23:25 -04002726}
2727
Al Viro2d864652013-09-08 20:18:44 -04002728int
2729kern_path_mountpoint(int dfd, const char *name, struct path *path,
2730 unsigned int flags)
2731{
Al Virocbaab2d2015-01-22 02:49:00 -05002732 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
Al Viro2d864652013-09-08 20:18:44 -04002733}
2734EXPORT_SYMBOL(kern_path_mountpoint);
2735
Miklos Szeredicbdf35b2014-10-24 00:14:36 +02002736int __check_sticky(struct inode *dir, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737{
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002738 kuid_t fsuid = current_fsuid();
David Howellsda9592e2008-11-14 10:39:05 +11002739
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002740 if (uid_eq(inode->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002741 return 0;
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002742 if (uid_eq(dir->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002743 return 0;
Andy Lutomirski23adbe12014-06-10 12:45:42 -07002744 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002745}
Miklos Szeredicbdf35b2014-10-24 00:14:36 +02002746EXPORT_SYMBOL(__check_sticky);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002747
2748/*
2749 * Check whether we can remove a link victim from directory dir, check
2750 * whether the type of victim is right.
2751 * 1. We can't do it if dir is read-only (done in permission())
2752 * 2. We should have write and exec permissions on dir
2753 * 3. We can't remove anything from append-only dir
2754 * 4. We can't do anything with immutable dir (done in permission())
2755 * 5. If the sticky bit on dir is set we should either
2756 * a. be owner of dir, or
2757 * b. be owner of victim, or
2758 * c. have CAP_FOWNER capability
2759 * 6. If the victim is append-only or immutable we can't do antyhing with
2760 * links pointing to it.
Eric W. Biederman0bd23d092016-06-29 14:54:46 -05002761 * 7. If the victim has an unknown uid or gid we can't change the inode.
2762 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2763 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2764 * 10. We can't remove a root or mountpoint.
2765 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
Linus Torvalds1da177e2005-04-16 15:20:36 -07002766 * nfs_async_unlink().
2767 */
David Howellsb18825a2013-09-12 19:22:53 +01002768static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002769{
David Howells63afdfc2015-05-06 15:59:00 +01002770 struct inode *inode = d_backing_inode(victim);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002771 int error;
2772
David Howellsb18825a2013-09-12 19:22:53 +01002773 if (d_is_negative(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774 return -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01002775 BUG_ON(!inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776
2777 BUG_ON(victim->d_parent->d_inode != dir);
Jeff Layton4fa6b5e2012-10-10 15:25:25 -04002778 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779
Al Virof419a2e2008-07-22 00:07:17 -04002780 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002781 if (error)
2782 return error;
2783 if (IS_APPEND(dir))
2784 return -EPERM;
David Howellsb18825a2013-09-12 19:22:53 +01002785
2786 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
Eric W. Biederman0bd23d092016-06-29 14:54:46 -05002787 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788 return -EPERM;
2789 if (isdir) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002790 if (!d_is_dir(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791 return -ENOTDIR;
2792 if (IS_ROOT(victim))
2793 return -EBUSY;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002794 } else if (d_is_dir(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002795 return -EISDIR;
2796 if (IS_DEADDIR(dir))
2797 return -ENOENT;
2798 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2799 return -EBUSY;
2800 return 0;
2801}
2802
2803/* Check whether we can create an object with dentry child in directory
2804 * dir.
2805 * 1. We can't do it if child already exists (open has special treatment for
2806 * this case, but since we are inlined it's OK)
2807 * 2. We can't do it if dir is read-only (done in permission())
Eric W. Biederman036d5232016-07-01 12:52:06 -05002808 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2809 * 4. We should have write and exec permissions on dir
2810 * 5. We can't do it if dir is immutable (done in permission())
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02002812static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002813{
Eric W. Biederman036d5232016-07-01 12:52:06 -05002814 struct user_namespace *s_user_ns;
Jeff Layton14e972b2013-05-08 10:25:58 -04002815 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002816 if (child->d_inode)
2817 return -EEXIST;
2818 if (IS_DEADDIR(dir))
2819 return -ENOENT;
Eric W. Biederman036d5232016-07-01 12:52:06 -05002820 s_user_ns = dir->i_sb->s_user_ns;
2821 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2822 !kgid_has_mapping(s_user_ns, current_fsgid()))
2823 return -EOVERFLOW;
Al Virof419a2e2008-07-22 00:07:17 -04002824 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002825}
2826
Linus Torvalds1da177e2005-04-16 15:20:36 -07002827/*
2828 * p1 and p2 should be directories on the same fs.
2829 */
2830struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2831{
2832 struct dentry *p;
2833
2834 if (p1 == p2) {
Al Viro59551022016-01-22 15:40:57 -05002835 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836 return NULL;
2837 }
2838
Al Virofc640052016-04-10 01:33:30 -04002839 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002840
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002841 p = d_ancestor(p2, p1);
2842 if (p) {
Al Viro59551022016-01-22 15:40:57 -05002843 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2844 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002845 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002846 }
2847
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002848 p = d_ancestor(p1, p2);
2849 if (p) {
Al Viro59551022016-01-22 15:40:57 -05002850 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2851 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002852 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002853 }
2854
Al Viro59551022016-01-22 15:40:57 -05002855 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2856 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857 return NULL;
2858}
Al Viro4d359502014-03-14 12:20:17 -04002859EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002860
2861void unlock_rename(struct dentry *p1, struct dentry *p2)
2862{
Al Viro59551022016-01-22 15:40:57 -05002863 inode_unlock(p1->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864 if (p1 != p2) {
Al Viro59551022016-01-22 15:40:57 -05002865 inode_unlock(p2->d_inode);
Al Virofc640052016-04-10 01:33:30 -04002866 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867 }
2868}
Al Viro4d359502014-03-14 12:20:17 -04002869EXPORT_SYMBOL(unlock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002870
Al Viro4acdaf22011-07-26 01:42:34 -04002871int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viro312b63f2012-06-10 18:09:36 -04002872 bool want_excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002873{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002874 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002875 if (error)
2876 return error;
2877
Al Viroacfa4382008-12-04 10:06:33 -05002878 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879 return -EACCES; /* shouldn't it be ENOSYS? */
2880 mode &= S_IALLUGO;
2881 mode |= S_IFREG;
2882 error = security_inode_create(dir, dentry, mode);
2883 if (error)
2884 return error;
Al Viro312b63f2012-06-10 18:09:36 -04002885 error = dir->i_op->create(dir, dentry, mode, want_excl);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002886 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002887 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002888 return error;
2889}
Al Viro4d359502014-03-14 12:20:17 -04002890EXPORT_SYMBOL(vfs_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891
Eric W. Biedermana2982cc2016-06-09 15:34:02 -05002892bool may_open_dev(const struct path *path)
2893{
2894 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2895 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2896}
2897
Al Viro73d049a2011-03-11 12:08:24 -05002898static int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002899{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002900 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002901 struct inode *inode = dentry->d_inode;
2902 int error;
2903
2904 if (!inode)
2905 return -ENOENT;
2906
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002907 switch (inode->i_mode & S_IFMT) {
2908 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002910 case S_IFDIR:
2911 if (acc_mode & MAY_WRITE)
2912 return -EISDIR;
2913 break;
2914 case S_IFBLK:
2915 case S_IFCHR:
Eric W. Biedermana2982cc2016-06-09 15:34:02 -05002916 if (!may_open_dev(path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002917 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002918 /*FALLTHRU*/
2919 case S_IFIFO:
2920 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002921 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002922 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002923 }
Dave Hansenb41572e2007-10-16 23:31:14 -07002924
Al Viro62fb4a12015-12-26 22:33:24 -05002925 error = inode_permission(inode, MAY_OPEN | acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07002926 if (error)
2927 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05002928
Linus Torvalds1da177e2005-04-16 15:20:36 -07002929 /*
2930 * An append-only file must be opened in append mode for writing.
2931 */
2932 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05002933 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05002934 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002935 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05002936 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002937 }
2938
2939 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002940 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05002941 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002942
J. Bruce Fieldsf3c7691e2011-09-21 10:58:13 -04002943 return 0;
Al Viro7715b522009-12-16 03:54:00 -05002944}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002945
Jeff Laytone1181ee2010-12-07 16:19:50 -05002946static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002947{
Jeff Laytone1181ee2010-12-07 16:19:50 -05002948 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002949 struct inode *inode = path->dentry->d_inode;
2950 int error = get_write_access(inode);
2951 if (error)
2952 return error;
2953 /*
2954 * Refuse to truncate files with mandatory locks held on them.
2955 */
Jeff Laytond7a06982014-03-10 09:54:15 -04002956 error = locks_verify_locked(filp);
Al Viro7715b522009-12-16 03:54:00 -05002957 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002958 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002959 if (!error) {
2960 error = do_truncate(path->dentry, 0,
2961 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002962 filp);
Al Viro7715b522009-12-16 03:54:00 -05002963 }
2964 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002965 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002966}
2967
Dave Hansend57999e2008-02-15 14:37:27 -08002968static inline int open_to_namei_flags(int flag)
2969{
Al Viro8a5e9292011-06-25 19:15:54 -04002970 if ((flag & O_ACCMODE) == 3)
2971 flag--;
Dave Hansend57999e2008-02-15 14:37:27 -08002972 return flag;
2973}
2974
Al Virod3607752016-03-25 15:21:09 -04002975static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002976{
2977 int error = security_path_mknod(dir, dentry, mode, 0);
2978 if (error)
2979 return error;
2980
2981 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2982 if (error)
2983 return error;
2984
2985 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2986}
2987
David Howells1acf0af2012-06-14 16:13:46 +01002988/*
2989 * Attempt to atomically look up, create and open a file from a negative
2990 * dentry.
2991 *
2992 * Returns 0 if successful. The file will have been created and attached to
2993 * @file by the filesystem calling finish_open().
2994 *
2995 * Returns 1 if the file was looked up only or didn't need creating. The
2996 * caller will need to perform the open themselves. @path will have been
2997 * updated to point to the new dentry. This may be negative.
2998 *
2999 * Returns an error code otherwise.
3000 */
Al Viro2675a4e2012-06-22 12:41:10 +04003001static int atomic_open(struct nameidata *nd, struct dentry *dentry,
3002 struct path *path, struct file *file,
3003 const struct open_flags *op,
Al Viro1643b432016-04-27 19:14:10 -04003004 int open_flag, umode_t mode,
Al Viro2675a4e2012-06-22 12:41:10 +04003005 int *opened)
Miklos Szeredid18e9002012-06-05 15:10:17 +02003006{
Miklos Szeredid18e9002012-06-05 15:10:17 +02003007 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003008 struct inode *dir = nd->path.dentry->d_inode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003009 int error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003010
Al Viro384f26e2016-04-28 02:03:55 -04003011 if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */
Miklos Szeredid18e9002012-06-05 15:10:17 +02003012 open_flag &= ~O_TRUNC;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003013
Miklos Szeredid18e9002012-06-05 15:10:17 +02003014 if (nd->flags & LOOKUP_DIRECTORY)
3015 open_flag |= O_DIRECTORY;
3016
Al Viro30d90492012-06-22 12:40:19 +04003017 file->f_path.dentry = DENTRY_NOT_SET;
3018 file->f_path.mnt = nd->path.mnt;
Al Viro0fb1ea02016-04-27 14:13:10 -04003019 error = dir->i_op->atomic_open(dir, dentry, file,
3020 open_to_namei_flags(open_flag),
3021 mode, opened);
Al Viro6fbd0712016-04-28 11:50:59 -04003022 d_lookup_done(dentry);
Al Viro384f26e2016-04-28 02:03:55 -04003023 if (!error) {
3024 /*
3025 * We didn't have the inode before the open, so check open
3026 * permission here.
3027 */
3028 int acc_mode = op->acc_mode;
3029 if (*opened & FILE_CREATED) {
3030 WARN_ON(!(open_flag & O_CREAT));
3031 fsnotify_create(dir, dentry);
3032 acc_mode = 0;
3033 }
3034 error = may_open(&file->f_path, acc_mode, open_flag);
3035 if (WARN_ON(error > 0))
3036 error = -EINVAL;
3037 } else if (error > 0) {
Al Viro30d90492012-06-22 12:40:19 +04003038 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
Al Viro2675a4e2012-06-22 12:41:10 +04003039 error = -EIO;
Al Viro03da6332013-09-16 19:22:33 -04003040 } else {
Al Viro384f26e2016-04-28 02:03:55 -04003041 if (file->f_path.dentry) {
3042 dput(dentry);
3043 dentry = file->f_path.dentry;
Al Viro03da6332013-09-16 19:22:33 -04003044 }
Al Viro384f26e2016-04-28 02:03:55 -04003045 if (*opened & FILE_CREATED)
3046 fsnotify_create(dir, dentry);
Al Viroa01e7182016-06-07 21:53:51 -04003047 if (unlikely(d_is_negative(dentry))) {
3048 error = -ENOENT;
3049 } else {
3050 path->dentry = dentry;
3051 path->mnt = nd->path.mnt;
3052 return 1;
3053 }
Sage Weil62b2ce92012-08-15 13:30:12 -07003054 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02003055 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02003056 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04003057 return error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003058}
3059
Nick Piggin31e6b012011-01-07 17:49:52 +11003060/*
David Howells1acf0af2012-06-14 16:13:46 +01003061 * Look up and maybe create and open the last component.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003062 *
3063 * Must be called with i_mutex held on parent.
3064 *
David Howells1acf0af2012-06-14 16:13:46 +01003065 * Returns 0 if the file was successfully atomically created (if necessary) and
3066 * opened. In this case the file will be returned attached to @file.
3067 *
3068 * Returns 1 if the file was not completely opened at this time, though lookups
3069 * and creations will have been performed and the dentry returned in @path will
3070 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
3071 * specified then a negative dentry may be returned.
3072 *
3073 * An error code is returned otherwise.
3074 *
3075 * FILE_CREATE will be set in @*opened if the dentry was created and will be
3076 * cleared otherwise prior to returning.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003077 */
Al Viro2675a4e2012-06-22 12:41:10 +04003078static int lookup_open(struct nameidata *nd, struct path *path,
3079 struct file *file,
3080 const struct open_flags *op,
Al Viro64894cf2012-07-31 00:53:35 +04003081 bool got_write, int *opened)
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003082{
3083 struct dentry *dir = nd->path.dentry;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02003084 struct inode *dir_inode = dir->d_inode;
Al Viro1643b432016-04-27 19:14:10 -04003085 int open_flag = op->open_flag;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003086 struct dentry *dentry;
Al Viro1643b432016-04-27 19:14:10 -04003087 int error, create_error = 0;
Al Viro1643b432016-04-27 19:14:10 -04003088 umode_t mode = op->mode;
Al Viro6fbd0712016-04-28 11:50:59 -04003089 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003090
Al Viroce8644f2016-04-26 14:17:56 -04003091 if (unlikely(IS_DEADDIR(dir_inode)))
3092 return -ENOENT;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003093
Al Viro47237682012-06-10 05:01:45 -04003094 *opened &= ~FILE_CREATED;
Al Viro6fbd0712016-04-28 11:50:59 -04003095 dentry = d_lookup(dir, &nd->last);
3096 for (;;) {
3097 if (!dentry) {
3098 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3099 if (IS_ERR(dentry))
3100 return PTR_ERR(dentry);
3101 }
3102 if (d_in_lookup(dentry))
3103 break;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003104
Al Viro6fbd0712016-04-28 11:50:59 -04003105 if (!(dentry->d_flags & DCACHE_OP_REVALIDATE))
3106 break;
3107
3108 error = d_revalidate(dentry, nd->flags);
3109 if (likely(error > 0))
3110 break;
3111 if (error)
3112 goto out_dput;
3113 d_invalidate(dentry);
3114 dput(dentry);
3115 dentry = NULL;
3116 }
3117 if (dentry->d_inode) {
Al Viro6c51e512016-03-05 20:09:32 -05003118 /* Cached positive dentry: will open in f_op->open */
Miklos Szeredid18e9002012-06-05 15:10:17 +02003119 goto out_no_open;
Al Viro6c51e512016-03-05 20:09:32 -05003120 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02003121
Al Viro1643b432016-04-27 19:14:10 -04003122 /*
3123 * Checking write permission is tricky, bacuse we don't know if we are
3124 * going to actually need it: O_CREAT opens should work as long as the
3125 * file exists. But checking existence breaks atomicity. The trick is
3126 * to check access and if not granted clear O_CREAT from the flags.
3127 *
3128 * Another problem is returing the "right" error value (e.g. for an
3129 * O_EXCL open we want to return EEXIST not EROFS).
3130 */
3131 if (open_flag & O_CREAT) {
3132 if (!IS_POSIXACL(dir->d_inode))
3133 mode &= ~current_umask();
3134 if (unlikely(!got_write)) {
3135 create_error = -EROFS;
3136 open_flag &= ~O_CREAT;
3137 if (open_flag & (O_EXCL | O_TRUNC))
3138 goto no_open;
3139 /* No side effects, safe to clear O_CREAT */
3140 } else {
3141 create_error = may_o_create(&nd->path, dentry, mode);
3142 if (create_error) {
3143 open_flag &= ~O_CREAT;
3144 if (open_flag & O_EXCL)
3145 goto no_open;
3146 }
3147 }
3148 } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3149 unlikely(!got_write)) {
3150 /*
3151 * No O_CREATE -> atomicity not a requirement -> fall
3152 * back to lookup + open
3153 */
3154 goto no_open;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003155 }
3156
Al Viro1643b432016-04-27 19:14:10 -04003157 if (dir_inode->i_op->atomic_open) {
3158 error = atomic_open(nd, dentry, path, file, op, open_flag,
3159 mode, opened);
3160 if (unlikely(error == -ENOENT) && create_error)
3161 error = create_error;
3162 return error;
3163 }
Miklos Szeredi54ef4872012-06-05 15:10:16 +02003164
Al Viro1643b432016-04-27 19:14:10 -04003165no_open:
Al Viro6fbd0712016-04-28 11:50:59 -04003166 if (d_in_lookup(dentry)) {
Al Viro12fa5e22016-04-28 11:19:43 -04003167 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3168 nd->flags);
Al Viro6fbd0712016-04-28 11:50:59 -04003169 d_lookup_done(dentry);
Al Viro12fa5e22016-04-28 11:19:43 -04003170 if (unlikely(res)) {
3171 if (IS_ERR(res)) {
3172 error = PTR_ERR(res);
3173 goto out_dput;
3174 }
3175 dput(dentry);
3176 dentry = res;
3177 }
Miklos Szeredi54ef4872012-06-05 15:10:16 +02003178 }
3179
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003180 /* Negative dentry, just create the file */
Al Viro1643b432016-04-27 19:14:10 -04003181 if (!dentry->d_inode && (open_flag & O_CREAT)) {
Al Viro47237682012-06-10 05:01:45 -04003182 *opened |= FILE_CREATED;
Al Viroce8644f2016-04-26 14:17:56 -04003183 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
Al Viroce8644f2016-04-26 14:17:56 -04003184 if (!dir_inode->i_op->create) {
3185 error = -EACCES;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003186 goto out_dput;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003187 }
Al Viroce8644f2016-04-26 14:17:56 -04003188 error = dir_inode->i_op->create(dir_inode, dentry, mode,
Al Viro1643b432016-04-27 19:14:10 -04003189 open_flag & O_EXCL);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003190 if (error)
3191 goto out_dput;
Al Viroce8644f2016-04-26 14:17:56 -04003192 fsnotify_create(dir_inode, dentry);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003193 }
Al Viro1643b432016-04-27 19:14:10 -04003194 if (unlikely(create_error) && !dentry->d_inode) {
3195 error = create_error;
3196 goto out_dput;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003197 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02003198out_no_open:
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003199 path->dentry = dentry;
3200 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04003201 return 1;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003202
3203out_dput:
3204 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04003205 return error;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003206}
3207
3208/*
Al Virofe2d35f2011-03-05 22:58:25 -05003209 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11003210 */
Al Viro896475d2015-04-22 18:02:17 -04003211static int do_last(struct nameidata *nd,
Al Viro2675a4e2012-06-22 12:41:10 +04003212 struct file *file, const struct open_flags *op,
Al Viro76ae2a52015-05-12 18:44:32 -04003213 int *opened)
Al Virofb1cc552009-12-24 01:58:28 -05003214{
Al Viroa1e28032009-12-24 02:12:06 -05003215 struct dentry *dir = nd->path.dentry;
Al Viroca344a892011-03-09 00:36:45 -05003216 int open_flag = op->open_flag;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003217 bool will_truncate = (open_flag & O_TRUNC) != 0;
Al Viro64894cf2012-07-31 00:53:35 +04003218 bool got_write = false;
Al Virobcda7652011-03-13 16:42:14 -04003219 int acc_mode = op->acc_mode;
Al Viro254cf582015-05-05 09:40:46 -04003220 unsigned seq;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02003221 struct inode *inode;
Al Viro896475d2015-04-22 18:02:17 -04003222 struct path path;
Al Viro16c2cd72011-02-22 15:50:10 -05003223 int error;
Al Virofb1cc552009-12-24 01:58:28 -05003224
Al Viroc3e380b2011-02-23 13:39:45 -05003225 nd->flags &= ~LOOKUP_PARENT;
3226 nd->flags |= op->intent;
3227
Al Virobc77daa2013-06-06 09:12:33 -04003228 if (nd->last_type != LAST_NORM) {
Al Virofe2d35f2011-03-05 22:58:25 -05003229 error = handle_dots(nd, nd->last_type);
Al Virodeb106c2015-05-08 18:05:21 -04003230 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04003231 return error;
Miklos Szeredie83db162012-06-05 15:10:29 +02003232 goto finish_open;
Al Viro1f36f772009-12-26 10:56:19 -05003233 }
Al Viro67ee3ad2009-12-26 07:01:01 -05003234
Al Viroca344a892011-03-09 00:36:45 -05003235 if (!(open_flag & O_CREAT)) {
Al Virofe2d35f2011-03-05 22:58:25 -05003236 if (nd->last.name[nd->last.len])
3237 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3238 /* we _can_ be in RCU mode here */
Al Viro254cf582015-05-05 09:40:46 -04003239 error = lookup_fast(nd, &path, &inode, &seq);
Al Viroe9742b52016-03-05 22:04:59 -05003240 if (likely(error > 0))
Miklos Szeredi71574862012-06-05 15:10:14 +02003241 goto finish_lookup;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02003242
Miklos Szeredi71574862012-06-05 15:10:14 +02003243 if (error < 0)
Al Virodeb106c2015-05-08 18:05:21 -04003244 return error;
Miklos Szeredi37d7fff2012-06-05 15:10:12 +02003245
Miklos Szeredi71574862012-06-05 15:10:14 +02003246 BUG_ON(nd->inode != dir->d_inode);
Al Viro6583fe22016-03-05 18:14:03 -05003247 BUG_ON(nd->flags & LOOKUP_RCU);
Miklos Szeredib6183df2012-06-05 15:10:13 +02003248 } else {
3249 /* create side of things */
3250 /*
3251 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3252 * has been cleared when we got to the last component we are
3253 * about to look up
3254 */
3255 error = complete_walk(nd);
Al Viroe8bb73d2015-05-08 16:28:42 -04003256 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003257 return error;
Miklos Szeredib6183df2012-06-05 15:10:13 +02003258
Al Viro76ae2a52015-05-12 18:44:32 -04003259 audit_inode(nd->name, dir, LOOKUP_PARENT);
Miklos Szeredib6183df2012-06-05 15:10:13 +02003260 /* trailing slashes? */
Al Virodeb106c2015-05-08 18:05:21 -04003261 if (unlikely(nd->last.name[nd->last.len]))
3262 return -EISDIR;
Al Virofe2d35f2011-03-05 22:58:25 -05003263 }
3264
Al Viro9cf843e2016-04-28 19:35:16 -04003265 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
Al Viro64894cf2012-07-31 00:53:35 +04003266 error = mnt_want_write(nd->path.mnt);
3267 if (!error)
3268 got_write = true;
3269 /*
3270 * do _not_ fail yet - we might not need that or fail with
3271 * a different error; let lookup_open() decide; we'll be
3272 * dropping this one anyway.
3273 */
3274 }
Al Viro9cf843e2016-04-28 19:35:16 -04003275 if (open_flag & O_CREAT)
3276 inode_lock(dir->d_inode);
3277 else
3278 inode_lock_shared(dir->d_inode);
Al Viro896475d2015-04-22 18:02:17 -04003279 error = lookup_open(nd, &path, file, op, got_write, opened);
Al Viro9cf843e2016-04-28 19:35:16 -04003280 if (open_flag & O_CREAT)
3281 inode_unlock(dir->d_inode);
3282 else
3283 inode_unlock_shared(dir->d_inode);
Al Viroa1e28032009-12-24 02:12:06 -05003284
Al Viro2675a4e2012-06-22 12:41:10 +04003285 if (error <= 0) {
3286 if (error)
Miklos Szeredid18e9002012-06-05 15:10:17 +02003287 goto out;
3288
Al Viro47237682012-06-10 05:01:45 -04003289 if ((*opened & FILE_CREATED) ||
Al Viro496ad9a2013-01-23 17:07:38 -05003290 !S_ISREG(file_inode(file)->i_mode))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003291 will_truncate = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003292
Al Viro76ae2a52015-05-12 18:44:32 -04003293 audit_inode(nd->name, file->f_path.dentry, 0);
Miklos Szeredid18e9002012-06-05 15:10:17 +02003294 goto opened;
3295 }
Al Virofb1cc552009-12-24 01:58:28 -05003296
Al Viro47237682012-06-10 05:01:45 -04003297 if (*opened & FILE_CREATED) {
Al Viro9b44f1b2011-03-09 00:17:27 -05003298 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05003299 open_flag &= ~O_TRUNC;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003300 will_truncate = false;
Al Viro62fb4a12015-12-26 22:33:24 -05003301 acc_mode = 0;
Al Viro896475d2015-04-22 18:02:17 -04003302 path_to_nameidata(&path, nd);
Miklos Szeredie83db162012-06-05 15:10:29 +02003303 goto finish_open_created;
Al Virofb1cc552009-12-24 01:58:28 -05003304 }
3305
3306 /*
Miklos Szeredid18e9002012-06-05 15:10:17 +02003307 * If atomic_open() acquired write access it is dropped now due to
3308 * possible mount and symlink following (this might be optimized away if
3309 * necessary...)
3310 */
Al Viro64894cf2012-07-31 00:53:35 +04003311 if (got_write) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02003312 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04003313 got_write = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003314 }
3315
Al Viroe6ec03a2016-06-05 00:23:09 -04003316 error = follow_managed(&path, nd);
3317 if (unlikely(error < 0))
3318 return error;
3319
Al Viro6583fe22016-03-05 18:14:03 -05003320 if (unlikely(d_is_negative(path.dentry))) {
3321 path_to_nameidata(&path, nd);
3322 return -ENOENT;
3323 }
3324
3325 /*
3326 * create/update audit record if it already exists.
3327 */
3328 audit_inode(nd->name, path.dentry, 0);
3329
Al Virodeb106c2015-05-08 18:05:21 -04003330 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3331 path_to_nameidata(&path, nd);
3332 return -EEXIST;
3333 }
Al Virofb1cc552009-12-24 01:58:28 -05003334
Al Viro254cf582015-05-05 09:40:46 -04003335 seq = 0; /* out of RCU mode, so the value doesn't matter */
Al Virod4565642016-02-27 19:23:16 -05003336 inode = d_backing_inode(path.dentry);
Al Viro766c4cb2015-05-07 19:24:57 -04003337finish_lookup:
Al Virod63ff282015-05-04 18:13:23 -04003338 if (nd->depth)
3339 put_link(nd);
Al Viro181548c2015-05-07 19:54:34 -04003340 error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW,
3341 inode, seq);
Al Virodeb106c2015-05-08 18:05:21 -04003342 if (unlikely(error))
Al Virod63ff282015-05-04 18:13:23 -04003343 return error;
Al Virofb1cc552009-12-24 01:58:28 -05003344
Al Virofac7d192016-06-04 11:41:49 -04003345 path_to_nameidata(&path, nd);
Miklos Szeredidecf3402012-05-21 17:30:08 +02003346 nd->inode = inode;
Al Viro254cf582015-05-05 09:40:46 -04003347 nd->seq = seq;
Al Viroa3fbbde2011-11-07 21:21:26 +00003348 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
Al Virobc77daa2013-06-06 09:12:33 -04003349finish_open:
Al Viroa3fbbde2011-11-07 21:21:26 +00003350 error = complete_walk(nd);
Al Virofac7d192016-06-04 11:41:49 -04003351 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003352 return error;
Al Viro76ae2a52015-05-12 18:44:32 -04003353 audit_inode(nd->name, nd->path.dentry, 0);
Al Virofb1cc552009-12-24 01:58:28 -05003354 error = -EISDIR;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003355 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04003356 goto out;
Miklos Szerediaf2f5542012-05-21 17:30:11 +02003357 error = -ENOTDIR;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003358 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04003359 goto out;
David Howells4bbcbd3b2015-03-17 22:16:40 +00003360 if (!d_is_reg(nd->path.dentry))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003361 will_truncate = false;
Al Viro6c0d46c2011-03-09 00:59:59 -05003362
Al Viro0f9d1a12011-03-09 00:13:14 -05003363 if (will_truncate) {
3364 error = mnt_want_write(nd->path.mnt);
3365 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003366 goto out;
Al Viro64894cf2012-07-31 00:53:35 +04003367 got_write = true;
Al Viro0f9d1a12011-03-09 00:13:14 -05003368 }
Miklos Szeredie83db162012-06-05 15:10:29 +02003369finish_open_created:
Al Viro6ac08702016-04-26 00:02:50 -04003370 error = may_open(&nd->path, acc_mode, open_flag);
3371 if (error)
3372 goto out;
Miklos Szeredi4aa7c632014-10-24 00:14:35 +02003373 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3374 error = vfs_open(&nd->path, file, current_cred());
Al Virofac7d192016-06-04 11:41:49 -04003375 if (error)
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003376 goto out;
Al Virofac7d192016-06-04 11:41:49 -04003377 *opened |= FILE_OPENED;
Miklos Szeredia8277b92012-06-05 15:10:32 +02003378opened:
Al Viro2675a4e2012-06-22 12:41:10 +04003379 error = open_check_o_direct(file);
Al Virofe9ec822016-04-27 03:14:20 -04003380 if (!error)
3381 error = ima_file_check(file, op->acc_mode, *opened);
3382 if (!error && will_truncate)
Al Viro2675a4e2012-06-22 12:41:10 +04003383 error = handle_truncate(file);
Al Viroca344a892011-03-09 00:36:45 -05003384out:
Al Virofe9ec822016-04-27 03:14:20 -04003385 if (unlikely(error) && (*opened & FILE_OPENED))
3386 fput(file);
Al Viroc80567c2016-02-27 19:17:33 -05003387 if (unlikely(error > 0)) {
3388 WARN_ON(1);
3389 error = -EINVAL;
3390 }
Al Viro64894cf2012-07-31 00:53:35 +04003391 if (got_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05003392 mnt_drop_write(nd->path.mnt);
Al Viro2675a4e2012-06-22 12:41:10 +04003393 return error;
Al Virofb1cc552009-12-24 01:58:28 -05003394}
3395
Al Viroc8a53ee2015-05-12 18:43:07 -04003396static int do_tmpfile(struct nameidata *nd, unsigned flags,
Al Viro60545d02013-06-07 01:20:27 -04003397 const struct open_flags *op,
3398 struct file *file, int *opened)
3399{
3400 static const struct qstr name = QSTR_INIT("/", 1);
Al Viro625b6d12015-05-12 16:36:12 -04003401 struct dentry *child;
Al Viro60545d02013-06-07 01:20:27 -04003402 struct inode *dir;
Al Viro625b6d12015-05-12 16:36:12 -04003403 struct path path;
Al Viroc8a53ee2015-05-12 18:43:07 -04003404 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
Al Viro60545d02013-06-07 01:20:27 -04003405 if (unlikely(error))
3406 return error;
Al Viro625b6d12015-05-12 16:36:12 -04003407 error = mnt_want_write(path.mnt);
Al Viro60545d02013-06-07 01:20:27 -04003408 if (unlikely(error))
3409 goto out;
Al Viro625b6d12015-05-12 16:36:12 -04003410 dir = path.dentry->d_inode;
Al Viro60545d02013-06-07 01:20:27 -04003411 /* we want directory to be writable */
Al Viro625b6d12015-05-12 16:36:12 -04003412 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Al Viro60545d02013-06-07 01:20:27 -04003413 if (error)
3414 goto out2;
Al Viro60545d02013-06-07 01:20:27 -04003415 if (!dir->i_op->tmpfile) {
3416 error = -EOPNOTSUPP;
3417 goto out2;
3418 }
Al Viro625b6d12015-05-12 16:36:12 -04003419 child = d_alloc(path.dentry, &name);
Al Viro60545d02013-06-07 01:20:27 -04003420 if (unlikely(!child)) {
3421 error = -ENOMEM;
3422 goto out2;
3423 }
Al Viro625b6d12015-05-12 16:36:12 -04003424 dput(path.dentry);
3425 path.dentry = child;
3426 error = dir->i_op->tmpfile(dir, child, op->mode);
Al Viro60545d02013-06-07 01:20:27 -04003427 if (error)
3428 goto out2;
Al Viroc8a53ee2015-05-12 18:43:07 -04003429 audit_inode(nd->name, child, 0);
Eric Rannaud69a91c22014-10-30 01:51:01 -07003430 /* Don't check for other permissions, the inode was just created */
Al Viro62fb4a12015-12-26 22:33:24 -05003431 error = may_open(&path, 0, op->open_flag);
Al Viro60545d02013-06-07 01:20:27 -04003432 if (error)
3433 goto out2;
Al Viro625b6d12015-05-12 16:36:12 -04003434 file->f_path.mnt = path.mnt;
3435 error = finish_open(file, child, NULL, opened);
Al Viro60545d02013-06-07 01:20:27 -04003436 if (error)
3437 goto out2;
3438 error = open_check_o_direct(file);
Al Virof4e0c302013-06-11 08:34:36 +04003439 if (error) {
Al Viro60545d02013-06-07 01:20:27 -04003440 fput(file);
Al Virof4e0c302013-06-11 08:34:36 +04003441 } else if (!(op->open_flag & O_EXCL)) {
3442 struct inode *inode = file_inode(file);
3443 spin_lock(&inode->i_lock);
3444 inode->i_state |= I_LINKABLE;
3445 spin_unlock(&inode->i_lock);
3446 }
Al Viro60545d02013-06-07 01:20:27 -04003447out2:
Al Viro625b6d12015-05-12 16:36:12 -04003448 mnt_drop_write(path.mnt);
Al Viro60545d02013-06-07 01:20:27 -04003449out:
Al Viro625b6d12015-05-12 16:36:12 -04003450 path_put(&path);
Al Viro60545d02013-06-07 01:20:27 -04003451 return error;
3452}
3453
Al Viro6ac08702016-04-26 00:02:50 -04003454static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3455{
3456 struct path path;
3457 int error = path_lookupat(nd, flags, &path);
3458 if (!error) {
3459 audit_inode(nd->name, path.dentry, 0);
3460 error = vfs_open(&path, file, current_cred());
3461 path_put(&path);
3462 }
3463 return error;
3464}
3465
Al Viroc8a53ee2015-05-12 18:43:07 -04003466static struct file *path_openat(struct nameidata *nd,
3467 const struct open_flags *op, unsigned flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003468{
Al Viro368ee9b2015-05-08 17:19:59 -04003469 const char *s;
Al Viro30d90492012-06-22 12:40:19 +04003470 struct file *file;
Al Viro47237682012-06-10 05:01:45 -04003471 int opened = 0;
Al Viro13aab422011-02-23 17:54:08 -05003472 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11003473
Al Viro30d90492012-06-22 12:40:19 +04003474 file = get_empty_filp();
Al Viro1afc99b2013-02-14 20:41:04 -05003475 if (IS_ERR(file))
3476 return file;
Nick Piggin31e6b012011-01-07 17:49:52 +11003477
Al Viro30d90492012-06-22 12:40:19 +04003478 file->f_flags = op->open_flag;
Nick Piggin31e6b012011-01-07 17:49:52 +11003479
Al Virobb458c62013-07-13 13:26:37 +04003480 if (unlikely(file->f_flags & __O_TMPFILE)) {
Al Viroc8a53ee2015-05-12 18:43:07 -04003481 error = do_tmpfile(nd, flags, op, file, &opened);
Al Virof15133d2015-05-08 22:53:15 -04003482 goto out2;
Al Viro60545d02013-06-07 01:20:27 -04003483 }
3484
Al Viro6ac08702016-04-26 00:02:50 -04003485 if (unlikely(file->f_flags & O_PATH)) {
3486 error = do_o_path(nd, flags, file);
3487 if (!error)
3488 opened |= FILE_OPENED;
3489 goto out2;
3490 }
3491
Al Viroc8a53ee2015-05-12 18:43:07 -04003492 s = path_init(nd, flags);
Al Viro368ee9b2015-05-08 17:19:59 -04003493 if (IS_ERR(s)) {
3494 put_filp(file);
3495 return ERR_CAST(s);
3496 }
Al Viro3bdba282015-05-08 17:37:07 -04003497 while (!(error = link_path_walk(s, nd)) &&
Al Viro76ae2a52015-05-12 18:44:32 -04003498 (error = do_last(nd, file, op, &opened)) > 0) {
Al Viro73d049a2011-03-11 12:08:24 -05003499 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
Al Viro3bdba282015-05-08 17:37:07 -04003500 s = trailing_symlink(nd);
3501 if (IS_ERR(s)) {
3502 error = PTR_ERR(s);
Al Viro2675a4e2012-06-22 12:41:10 +04003503 break;
Al Viro3bdba282015-05-08 17:37:07 -04003504 }
Al Viro806b6812009-12-26 07:16:40 -05003505 }
Al Virodeb106c2015-05-08 18:05:21 -04003506 terminate_walk(nd);
Al Virof15133d2015-05-08 22:53:15 -04003507out2:
Al Viro2675a4e2012-06-22 12:41:10 +04003508 if (!(opened & FILE_OPENED)) {
3509 BUG_ON(!error);
Al Viro30d90492012-06-22 12:40:19 +04003510 put_filp(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003511 }
Al Viro2675a4e2012-06-22 12:41:10 +04003512 if (unlikely(error)) {
3513 if (error == -EOPENSTALE) {
3514 if (flags & LOOKUP_RCU)
3515 error = -ECHILD;
3516 else
3517 error = -ESTALE;
3518 }
3519 file = ERR_PTR(error);
3520 }
3521 return file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003522}
3523
Jeff Layton669abf42012-10-10 16:43:10 -04003524struct file *do_filp_open(int dfd, struct filename *pathname,
Al Virof9652e12013-06-11 08:23:01 +04003525 const struct open_flags *op)
Al Viro13aab422011-02-23 17:54:08 -05003526{
Al Viro9883d182015-05-13 07:28:08 -04003527 struct nameidata nd;
Al Virof9652e12013-06-11 08:23:01 +04003528 int flags = op->lookup_flags;
Al Viro13aab422011-02-23 17:54:08 -05003529 struct file *filp;
3530
Al Viro9883d182015-05-13 07:28:08 -04003531 set_nameidata(&nd, dfd, pathname);
Al Viroc8a53ee2015-05-12 18:43:07 -04003532 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
Al Viro13aab422011-02-23 17:54:08 -05003533 if (unlikely(filp == ERR_PTR(-ECHILD)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003534 filp = path_openat(&nd, op, flags);
Al Viro13aab422011-02-23 17:54:08 -05003535 if (unlikely(filp == ERR_PTR(-ESTALE)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003536 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
Al Viro9883d182015-05-13 07:28:08 -04003537 restore_nameidata();
Al Viro13aab422011-02-23 17:54:08 -05003538 return filp;
3539}
3540
Al Viro73d049a2011-03-11 12:08:24 -05003541struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
Al Virof9652e12013-06-11 08:23:01 +04003542 const char *name, const struct open_flags *op)
Al Viro73d049a2011-03-11 12:08:24 -05003543{
Al Viro9883d182015-05-13 07:28:08 -04003544 struct nameidata nd;
Al Viro73d049a2011-03-11 12:08:24 -05003545 struct file *file;
Paul Moore51689102015-01-22 00:00:03 -05003546 struct filename *filename;
Al Virof9652e12013-06-11 08:23:01 +04003547 int flags = op->lookup_flags | LOOKUP_ROOT;
Al Viro73d049a2011-03-11 12:08:24 -05003548
3549 nd.root.mnt = mnt;
3550 nd.root.dentry = dentry;
3551
David Howellsb18825a2013-09-12 19:22:53 +01003552 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
Al Viro73d049a2011-03-11 12:08:24 -05003553 return ERR_PTR(-ELOOP);
3554
Paul Moore51689102015-01-22 00:00:03 -05003555 filename = getname_kernel(name);
Viresh Kumara1c83682015-08-12 15:59:44 +05303556 if (IS_ERR(filename))
Paul Moore51689102015-01-22 00:00:03 -05003557 return ERR_CAST(filename);
3558
Al Viro9883d182015-05-13 07:28:08 -04003559 set_nameidata(&nd, -1, filename);
Al Viroc8a53ee2015-05-12 18:43:07 -04003560 file = path_openat(&nd, op, flags | LOOKUP_RCU);
Al Viro73d049a2011-03-11 12:08:24 -05003561 if (unlikely(file == ERR_PTR(-ECHILD)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003562 file = path_openat(&nd, op, flags);
Al Viro73d049a2011-03-11 12:08:24 -05003563 if (unlikely(file == ERR_PTR(-ESTALE)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003564 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
Al Viro9883d182015-05-13 07:28:08 -04003565 restore_nameidata();
Paul Moore51689102015-01-22 00:00:03 -05003566 putname(filename);
Al Viro73d049a2011-03-11 12:08:24 -05003567 return file;
3568}
3569
Al Virofa14a0b2015-01-22 02:16:49 -05003570static struct dentry *filename_create(int dfd, struct filename *name,
Jeff Layton1ac12b42012-12-11 12:10:06 -05003571 struct path *path, unsigned int lookup_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003572{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003573 struct dentry *dentry = ERR_PTR(-EEXIST);
Al Viro391172c2015-05-09 11:19:16 -04003574 struct qstr last;
3575 int type;
Jan Karac30dabf2012-06-12 16:20:30 +02003576 int err2;
Jeff Layton1ac12b42012-12-11 12:10:06 -05003577 int error;
3578 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3579
3580 /*
3581 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3582 * other flags passed in are ignored!
3583 */
3584 lookup_flags &= LOOKUP_REVAL;
3585
Al Viro5c31b6c2015-05-12 17:32:54 -04003586 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3587 if (IS_ERR(name))
3588 return ERR_CAST(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003589
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003590 /*
3591 * Yucky last component or no last component at all?
3592 * (foo/., foo/.., /////)
3593 */
Al Viro5c31b6c2015-05-12 17:32:54 -04003594 if (unlikely(type != LAST_NORM))
Al Viroed75e952011-06-27 16:53:43 -04003595 goto out;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003596
Jan Karac30dabf2012-06-12 16:20:30 +02003597 /* don't fail immediately if it's r/o, at least try to report other errors */
Al Viro391172c2015-05-09 11:19:16 -04003598 err2 = mnt_want_write(path->mnt);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003599 /*
3600 * Do the final lookup.
3601 */
Al Viro391172c2015-05-09 11:19:16 -04003602 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
Al Viro59551022016-01-22 15:40:57 -05003603 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
Al Viro391172c2015-05-09 11:19:16 -04003604 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003605 if (IS_ERR(dentry))
Al Viroa8104a92012-07-20 02:25:00 +04003606 goto unlock;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003607
Al Viroa8104a92012-07-20 02:25:00 +04003608 error = -EEXIST;
David Howellsb18825a2013-09-12 19:22:53 +01003609 if (d_is_positive(dentry))
Al Viroa8104a92012-07-20 02:25:00 +04003610 goto fail;
David Howellsb18825a2013-09-12 19:22:53 +01003611
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003612 /*
3613 * Special case - lookup gave negative, but... we had foo/bar/
3614 * From the vfs_mknod() POV we just have a negative dentry -
3615 * all is fine. Let's be bastards - you had / on the end, you've
3616 * been asking for (non-existent) directory. -ENOENT for you.
3617 */
Al Viro391172c2015-05-09 11:19:16 -04003618 if (unlikely(!is_dir && last.name[last.len])) {
Al Viroa8104a92012-07-20 02:25:00 +04003619 error = -ENOENT;
Al Viroed75e952011-06-27 16:53:43 -04003620 goto fail;
Al Viroe9baf6e2008-05-15 04:49:12 -04003621 }
Jan Karac30dabf2012-06-12 16:20:30 +02003622 if (unlikely(err2)) {
3623 error = err2;
Al Viroa8104a92012-07-20 02:25:00 +04003624 goto fail;
Jan Karac30dabf2012-06-12 16:20:30 +02003625 }
Al Viro181c37b2015-05-12 17:21:25 -04003626 putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003627 return dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003628fail:
Al Viroa8104a92012-07-20 02:25:00 +04003629 dput(dentry);
3630 dentry = ERR_PTR(error);
3631unlock:
Al Viro59551022016-01-22 15:40:57 -05003632 inode_unlock(path->dentry->d_inode);
Jan Karac30dabf2012-06-12 16:20:30 +02003633 if (!err2)
Al Viro391172c2015-05-09 11:19:16 -04003634 mnt_drop_write(path->mnt);
Al Viroed75e952011-06-27 16:53:43 -04003635out:
Al Viro391172c2015-05-09 11:19:16 -04003636 path_put(path);
Al Viro181c37b2015-05-12 17:21:25 -04003637 putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003638 return dentry;
3639}
Al Virofa14a0b2015-01-22 02:16:49 -05003640
3641struct dentry *kern_path_create(int dfd, const char *pathname,
3642 struct path *path, unsigned int lookup_flags)
3643{
Al Viro181c37b2015-05-12 17:21:25 -04003644 return filename_create(dfd, getname_kernel(pathname),
3645 path, lookup_flags);
Al Virofa14a0b2015-01-22 02:16:49 -05003646}
Al Virodae6ad82011-06-26 11:50:15 -04003647EXPORT_SYMBOL(kern_path_create);
3648
Al Viro921a1652012-07-20 01:15:31 +04003649void done_path_create(struct path *path, struct dentry *dentry)
3650{
3651 dput(dentry);
Al Viro59551022016-01-22 15:40:57 -05003652 inode_unlock(path->dentry->d_inode);
Al Viroa8104a92012-07-20 02:25:00 +04003653 mnt_drop_write(path->mnt);
Al Viro921a1652012-07-20 01:15:31 +04003654 path_put(path);
3655}
3656EXPORT_SYMBOL(done_path_create);
3657
Al Viro520ae682015-05-13 07:00:28 -04003658inline struct dentry *user_path_create(int dfd, const char __user *pathname,
Jeff Layton1ac12b42012-12-11 12:10:06 -05003659 struct path *path, unsigned int lookup_flags)
Al Virodae6ad82011-06-26 11:50:15 -04003660{
Al Viro181c37b2015-05-12 17:21:25 -04003661 return filename_create(dfd, getname(pathname), path, lookup_flags);
Al Virodae6ad82011-06-26 11:50:15 -04003662}
3663EXPORT_SYMBOL(user_path_create);
3664
Al Viro1a67aaf2011-07-26 01:52:52 -04003665int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003666{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003667 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003668
3669 if (error)
3670 return error;
3671
Eric W. Biederman975d6b32011-11-13 12:16:43 -08003672 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003673 return -EPERM;
3674
Al Viroacfa4382008-12-04 10:06:33 -05003675 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003676 return -EPERM;
3677
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07003678 error = devcgroup_inode_mknod(mode, dev);
3679 if (error)
3680 return error;
3681
Linus Torvalds1da177e2005-04-16 15:20:36 -07003682 error = security_inode_mknod(dir, dentry, mode, dev);
3683 if (error)
3684 return error;
3685
Linus Torvalds1da177e2005-04-16 15:20:36 -07003686 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003687 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003688 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003689 return error;
3690}
Al Viro4d359502014-03-14 12:20:17 -04003691EXPORT_SYMBOL(vfs_mknod);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003692
Al Virof69aac02011-07-26 04:31:40 -04003693static int may_mknod(umode_t mode)
Dave Hansen463c3192008-02-15 14:37:57 -08003694{
3695 switch (mode & S_IFMT) {
3696 case S_IFREG:
3697 case S_IFCHR:
3698 case S_IFBLK:
3699 case S_IFIFO:
3700 case S_IFSOCK:
3701 case 0: /* zero mode translates to S_IFREG */
3702 return 0;
3703 case S_IFDIR:
3704 return -EPERM;
3705 default:
3706 return -EINVAL;
3707 }
3708}
3709
Al Viro8208a222011-07-25 17:32:17 -04003710SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003711 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003712{
Al Viro2ad94ae2008-07-21 09:32:51 -04003713 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003714 struct path path;
3715 int error;
Jeff Layton972567f2012-12-20 16:00:10 -05003716 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003717
Al Viro8e4bfca2012-07-20 01:17:26 +04003718 error = may_mknod(mode);
3719 if (error)
3720 return error;
Jeff Layton972567f2012-12-20 16:00:10 -05003721retry:
3722 dentry = user_path_create(dfd, filename, &path, lookup_flags);
Al Virodae6ad82011-06-26 11:50:15 -04003723 if (IS_ERR(dentry))
3724 return PTR_ERR(dentry);
Al Viro2ad94ae2008-07-21 09:32:51 -04003725
Al Virodae6ad82011-06-26 11:50:15 -04003726 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003727 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003728 error = security_path_mknod(&path, dentry, mode, dev);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003729 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04003730 goto out;
Dave Hansen463c3192008-02-15 14:37:57 -08003731 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003732 case 0: case S_IFREG:
Al Viro312b63f2012-06-10 18:09:36 -04003733 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
Mimi Zohar05d1a712016-02-29 19:52:05 -05003734 if (!error)
3735 ima_post_path_mknod(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003736 break;
3737 case S_IFCHR: case S_IFBLK:
Al Virodae6ad82011-06-26 11:50:15 -04003738 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003739 new_decode_dev(dev));
3740 break;
3741 case S_IFIFO: case S_IFSOCK:
Al Virodae6ad82011-06-26 11:50:15 -04003742 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003743 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003744 }
Al Viroa8104a92012-07-20 02:25:00 +04003745out:
Al Viro921a1652012-07-20 01:15:31 +04003746 done_path_create(&path, dentry);
Jeff Layton972567f2012-12-20 16:00:10 -05003747 if (retry_estale(error, lookup_flags)) {
3748 lookup_flags |= LOOKUP_REVAL;
3749 goto retry;
3750 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003751 return error;
3752}
3753
Al Viro8208a222011-07-25 17:32:17 -04003754SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003755{
3756 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3757}
3758
Al Viro18bb1db2011-07-26 01:41:39 -04003759int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003760{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003761 int error = may_create(dir, dentry);
Al Viro8de52772012-02-06 12:45:27 -05003762 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003763
3764 if (error)
3765 return error;
3766
Al Viroacfa4382008-12-04 10:06:33 -05003767 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003768 return -EPERM;
3769
3770 mode &= (S_IRWXUGO|S_ISVTX);
3771 error = security_inode_mkdir(dir, dentry, mode);
3772 if (error)
3773 return error;
3774
Al Viro8de52772012-02-06 12:45:27 -05003775 if (max_links && dir->i_nlink >= max_links)
3776 return -EMLINK;
3777
Linus Torvalds1da177e2005-04-16 15:20:36 -07003778 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003779 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003780 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003781 return error;
3782}
Al Viro4d359502014-03-14 12:20:17 -04003783EXPORT_SYMBOL(vfs_mkdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003784
Al Viroa218d0f2011-11-21 14:59:34 -05003785SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003786{
Dave Hansen6902d922006-09-30 23:29:01 -07003787 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003788 struct path path;
3789 int error;
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003790 unsigned int lookup_flags = LOOKUP_DIRECTORY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003791
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003792retry:
3793 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07003794 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003795 return PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003796
Al Virodae6ad82011-06-26 11:50:15 -04003797 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003798 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003799 error = security_path_mkdir(&path, dentry, mode);
Al Viroa8104a92012-07-20 02:25:00 +04003800 if (!error)
3801 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
Al Viro921a1652012-07-20 01:15:31 +04003802 done_path_create(&path, dentry);
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003803 if (retry_estale(error, lookup_flags)) {
3804 lookup_flags |= LOOKUP_REVAL;
3805 goto retry;
3806 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003807 return error;
3808}
3809
Al Viroa218d0f2011-11-21 14:59:34 -05003810SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003811{
3812 return sys_mkdirat(AT_FDCWD, pathname, mode);
3813}
3814
Linus Torvalds1da177e2005-04-16 15:20:36 -07003815int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3816{
3817 int error = may_delete(dir, dentry, 1);
3818
3819 if (error)
3820 return error;
3821
Al Viroacfa4382008-12-04 10:06:33 -05003822 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003823 return -EPERM;
3824
Al Viro1d2ef592011-09-14 18:55:41 +01003825 dget(dentry);
Al Viro59551022016-01-22 15:40:57 -05003826 inode_lock(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003827
Sage Weil912dbc12011-05-24 13:06:11 -07003828 error = -EBUSY;
Eric W. Biederman7af13642013-10-04 19:15:13 -07003829 if (is_local_mountpoint(dentry))
Sage Weil912dbc12011-05-24 13:06:11 -07003830 goto out;
3831
3832 error = security_inode_rmdir(dir, dentry);
3833 if (error)
3834 goto out;
3835
Sage Weil3cebde22011-05-29 21:20:59 -07003836 shrink_dcache_parent(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003837 error = dir->i_op->rmdir(dir, dentry);
3838 if (error)
3839 goto out;
3840
3841 dentry->d_inode->i_flags |= S_DEAD;
3842 dont_mount(dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003843 detach_mounts(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003844
3845out:
Al Viro59551022016-01-22 15:40:57 -05003846 inode_unlock(dentry->d_inode);
Al Viro1d2ef592011-09-14 18:55:41 +01003847 dput(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003848 if (!error)
3849 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003850 return error;
3851}
Al Viro4d359502014-03-14 12:20:17 -04003852EXPORT_SYMBOL(vfs_rmdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003853
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003854static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003855{
3856 int error = 0;
Jeff Layton91a27b22012-10-10 15:25:28 -04003857 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003858 struct dentry *dentry;
Al Virof5beed72015-04-30 16:09:11 -04003859 struct path path;
3860 struct qstr last;
3861 int type;
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003862 unsigned int lookup_flags = 0;
3863retry:
Al Virof5beed72015-04-30 16:09:11 -04003864 name = user_path_parent(dfd, pathname,
3865 &path, &last, &type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04003866 if (IS_ERR(name))
3867 return PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003868
Al Virof5beed72015-04-30 16:09:11 -04003869 switch (type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003870 case LAST_DOTDOT:
3871 error = -ENOTEMPTY;
3872 goto exit1;
3873 case LAST_DOT:
3874 error = -EINVAL;
3875 goto exit1;
3876 case LAST_ROOT:
3877 error = -EBUSY;
3878 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003879 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003880
Al Virof5beed72015-04-30 16:09:11 -04003881 error = mnt_want_write(path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02003882 if (error)
3883 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003884
Al Viro59551022016-01-22 15:40:57 -05003885 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
Al Virof5beed72015-04-30 16:09:11 -04003886 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003887 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003888 if (IS_ERR(dentry))
3889 goto exit2;
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003890 if (!dentry->d_inode) {
3891 error = -ENOENT;
3892 goto exit3;
3893 }
Al Virof5beed72015-04-30 16:09:11 -04003894 error = security_path_rmdir(&path, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003895 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003896 goto exit3;
Al Virof5beed72015-04-30 16:09:11 -04003897 error = vfs_rmdir(path.dentry->d_inode, dentry);
Dave Hansen06227532008-02-15 14:37:34 -08003898exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07003899 dput(dentry);
3900exit2:
Al Viro59551022016-01-22 15:40:57 -05003901 inode_unlock(path.dentry->d_inode);
Al Virof5beed72015-04-30 16:09:11 -04003902 mnt_drop_write(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003903exit1:
Al Virof5beed72015-04-30 16:09:11 -04003904 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003905 putname(name);
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003906 if (retry_estale(error, lookup_flags)) {
3907 lookup_flags |= LOOKUP_REVAL;
3908 goto retry;
3909 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003910 return error;
3911}
3912
Heiko Carstens3cdad422009-01-14 14:14:22 +01003913SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003914{
3915 return do_rmdir(AT_FDCWD, pathname);
3916}
3917
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003918/**
3919 * vfs_unlink - unlink a filesystem object
3920 * @dir: parent directory
3921 * @dentry: victim
3922 * @delegated_inode: returns victim inode, if the inode is delegated.
3923 *
3924 * The caller must hold dir->i_mutex.
3925 *
3926 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3927 * return a reference to the inode in delegated_inode. The caller
3928 * should then break the delegation on that inode and retry. Because
3929 * breaking a delegation may take a long time, the caller should drop
3930 * dir->i_mutex before doing so.
3931 *
3932 * Alternatively, a caller may pass NULL for delegated_inode. This may
3933 * be appropriate for callers that expect the underlying filesystem not
3934 * to be NFS exported.
3935 */
3936int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003937{
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003938 struct inode *target = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003939 int error = may_delete(dir, dentry, 0);
3940
3941 if (error)
3942 return error;
3943
Al Viroacfa4382008-12-04 10:06:33 -05003944 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003945 return -EPERM;
3946
Al Viro59551022016-01-22 15:40:57 -05003947 inode_lock(target);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003948 if (is_local_mountpoint(dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003949 error = -EBUSY;
3950 else {
3951 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05003952 if (!error) {
J. Bruce Fields5a146962012-08-28 07:50:40 -07003953 error = try_break_deleg(target, delegated_inode);
3954 if (error)
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003955 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003956 error = dir->i_op->unlink(dir, dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003957 if (!error) {
Al Virod83c49f2010-04-30 17:17:09 -04003958 dont_mount(dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003959 detach_mounts(dentry);
3960 }
Al Virobec10522010-03-03 14:12:08 -05003961 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003962 }
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003963out:
Al Viro59551022016-01-22 15:40:57 -05003964 inode_unlock(target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003965
3966 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3967 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003968 fsnotify_link_count(target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003969 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003970 }
Robert Love0eeca282005-07-12 17:06:03 -04003971
Linus Torvalds1da177e2005-04-16 15:20:36 -07003972 return error;
3973}
Al Viro4d359502014-03-14 12:20:17 -04003974EXPORT_SYMBOL(vfs_unlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003975
3976/*
3977 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003978 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07003979 * writeout happening, and we don't want to prevent access to the directory
3980 * while waiting on the I/O.
3981 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003982static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003983{
Al Viro2ad94ae2008-07-21 09:32:51 -04003984 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04003985 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003986 struct dentry *dentry;
Al Virof5beed72015-04-30 16:09:11 -04003987 struct path path;
3988 struct qstr last;
3989 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003990 struct inode *inode = NULL;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003991 struct inode *delegated_inode = NULL;
Jeff Layton5d18f812012-12-20 16:38:04 -05003992 unsigned int lookup_flags = 0;
3993retry:
Al Virof5beed72015-04-30 16:09:11 -04003994 name = user_path_parent(dfd, pathname,
3995 &path, &last, &type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04003996 if (IS_ERR(name))
3997 return PTR_ERR(name);
Al Viro2ad94ae2008-07-21 09:32:51 -04003998
Linus Torvalds1da177e2005-04-16 15:20:36 -07003999 error = -EISDIR;
Al Virof5beed72015-04-30 16:09:11 -04004000 if (type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004001 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09004002
Al Virof5beed72015-04-30 16:09:11 -04004003 error = mnt_want_write(path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02004004 if (error)
4005 goto exit1;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04004006retry_deleg:
Al Viro59551022016-01-22 15:40:57 -05004007 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
Al Virof5beed72015-04-30 16:09:11 -04004008 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004009 error = PTR_ERR(dentry);
4010 if (!IS_ERR(dentry)) {
4011 /* Why not before? Because we want correct error value */
Al Virof5beed72015-04-30 16:09:11 -04004012 if (last.name[last.len])
Török Edwin50338b82011-06-16 00:06:14 +03004013 goto slashes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004014 inode = dentry->d_inode;
David Howellsb18825a2013-09-12 19:22:53 +01004015 if (d_is_negative(dentry))
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04004016 goto slashes;
4017 ihold(inode);
Al Virof5beed72015-04-30 16:09:11 -04004018 error = security_path_unlink(&path, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004019 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02004020 goto exit2;
Al Virof5beed72015-04-30 16:09:11 -04004021 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
Jan Karac30dabf2012-06-12 16:20:30 +02004022exit2:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004023 dput(dentry);
4024 }
Al Viro59551022016-01-22 15:40:57 -05004025 inode_unlock(path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004026 if (inode)
4027 iput(inode); /* truncate the inode here */
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04004028 inode = NULL;
4029 if (delegated_inode) {
J. Bruce Fields5a146962012-08-28 07:50:40 -07004030 error = break_deleg_wait(&delegated_inode);
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04004031 if (!error)
4032 goto retry_deleg;
4033 }
Al Virof5beed72015-04-30 16:09:11 -04004034 mnt_drop_write(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004035exit1:
Al Virof5beed72015-04-30 16:09:11 -04004036 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004037 putname(name);
Jeff Layton5d18f812012-12-20 16:38:04 -05004038 if (retry_estale(error, lookup_flags)) {
4039 lookup_flags |= LOOKUP_REVAL;
4040 inode = NULL;
4041 goto retry;
4042 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004043 return error;
4044
4045slashes:
David Howellsb18825a2013-09-12 19:22:53 +01004046 if (d_is_negative(dentry))
4047 error = -ENOENT;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02004048 else if (d_is_dir(dentry))
David Howellsb18825a2013-09-12 19:22:53 +01004049 error = -EISDIR;
4050 else
4051 error = -ENOTDIR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004052 goto exit2;
4053}
4054
Heiko Carstens2e4d0922009-01-14 14:14:31 +01004055SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004056{
4057 if ((flag & ~AT_REMOVEDIR) != 0)
4058 return -EINVAL;
4059
4060 if (flag & AT_REMOVEDIR)
4061 return do_rmdir(dfd, pathname);
4062
4063 return do_unlinkat(dfd, pathname);
4064}
4065
Heiko Carstens3480b252009-01-14 14:14:16 +01004066SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004067{
4068 return do_unlinkat(AT_FDCWD, pathname);
4069}
4070
Miklos Szeredidb2e7472008-06-24 16:50:16 +02004071int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004072{
Miklos Szeredia95164d2008-07-30 15:08:48 +02004073 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004074
4075 if (error)
4076 return error;
4077
Al Viroacfa4382008-12-04 10:06:33 -05004078 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004079 return -EPERM;
4080
4081 error = security_inode_symlink(dir, dentry, oldname);
4082 if (error)
4083 return error;
4084
Linus Torvalds1da177e2005-04-16 15:20:36 -07004085 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07004086 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00004087 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004088 return error;
4089}
Al Viro4d359502014-03-14 12:20:17 -04004090EXPORT_SYMBOL(vfs_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004091
Heiko Carstens2e4d0922009-01-14 14:14:31 +01004092SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4093 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004094{
Al Viro2ad94ae2008-07-21 09:32:51 -04004095 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04004096 struct filename *from;
Dave Hansen6902d922006-09-30 23:29:01 -07004097 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04004098 struct path path;
Jeff Laytonf46d3562012-12-11 12:10:08 -05004099 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004100
4101 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04004102 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004103 return PTR_ERR(from);
Jeff Laytonf46d3562012-12-11 12:10:08 -05004104retry:
4105 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07004106 error = PTR_ERR(dentry);
4107 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04004108 goto out_putname;
Dave Hansen6902d922006-09-30 23:29:01 -07004109
Jeff Layton91a27b22012-10-10 15:25:28 -04004110 error = security_path_symlink(&path, dentry, from->name);
Al Viroa8104a92012-07-20 02:25:00 +04004111 if (!error)
Jeff Layton91a27b22012-10-10 15:25:28 -04004112 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
Al Viro921a1652012-07-20 01:15:31 +04004113 done_path_create(&path, dentry);
Jeff Laytonf46d3562012-12-11 12:10:08 -05004114 if (retry_estale(error, lookup_flags)) {
4115 lookup_flags |= LOOKUP_REVAL;
4116 goto retry;
4117 }
Dave Hansen6902d922006-09-30 23:29:01 -07004118out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004119 putname(from);
4120 return error;
4121}
4122
Heiko Carstens3480b252009-01-14 14:14:16 +01004123SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004124{
4125 return sys_symlinkat(oldname, AT_FDCWD, newname);
4126}
4127
J. Bruce Fields146a8592011-09-20 17:14:31 -04004128/**
4129 * vfs_link - create a new link
4130 * @old_dentry: object to be linked
4131 * @dir: new parent
4132 * @new_dentry: where to create the new link
4133 * @delegated_inode: returns inode needing a delegation break
4134 *
4135 * The caller must hold dir->i_mutex
4136 *
4137 * If vfs_link discovers a delegation on the to-be-linked file in need
4138 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4139 * inode in delegated_inode. The caller should then break the delegation
4140 * and retry. Because breaking a delegation may take a long time, the
4141 * caller should drop the i_mutex before doing so.
4142 *
4143 * Alternatively, a caller may pass NULL for delegated_inode. This may
4144 * be appropriate for callers that expect the underlying filesystem not
4145 * to be NFS exported.
4146 */
4147int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004148{
4149 struct inode *inode = old_dentry->d_inode;
Al Viro8de52772012-02-06 12:45:27 -05004150 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004151 int error;
4152
4153 if (!inode)
4154 return -ENOENT;
4155
Miklos Szeredia95164d2008-07-30 15:08:48 +02004156 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004157 if (error)
4158 return error;
4159
4160 if (dir->i_sb != inode->i_sb)
4161 return -EXDEV;
4162
4163 /*
4164 * A link to an append-only or immutable file cannot be created.
4165 */
4166 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4167 return -EPERM;
Eric W. Biederman0bd23d092016-06-29 14:54:46 -05004168 /*
4169 * Updating the link count will likely cause i_uid and i_gid to
4170 * be writen back improperly if their true value is unknown to
4171 * the vfs.
4172 */
4173 if (HAS_UNMAPPED_ID(inode))
4174 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05004175 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004176 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02004177 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004178 return -EPERM;
4179
4180 error = security_inode_link(old_dentry, dir, new_dentry);
4181 if (error)
4182 return error;
4183
Al Viro59551022016-01-22 15:40:57 -05004184 inode_lock(inode);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05304185 /* Make sure we don't allow creating hardlink to an unlinked file */
Al Virof4e0c302013-06-11 08:34:36 +04004186 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05304187 error = -ENOENT;
Al Viro8de52772012-02-06 12:45:27 -05004188 else if (max_links && inode->i_nlink >= max_links)
4189 error = -EMLINK;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004190 else {
4191 error = try_break_deleg(inode, delegated_inode);
4192 if (!error)
4193 error = dir->i_op->link(old_dentry, dir, new_dentry);
4194 }
Al Virof4e0c302013-06-11 08:34:36 +04004195
4196 if (!error && (inode->i_state & I_LINKABLE)) {
4197 spin_lock(&inode->i_lock);
4198 inode->i_state &= ~I_LINKABLE;
4199 spin_unlock(&inode->i_lock);
4200 }
Al Viro59551022016-01-22 15:40:57 -05004201 inode_unlock(inode);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07004202 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02004203 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004204 return error;
4205}
Al Viro4d359502014-03-14 12:20:17 -04004206EXPORT_SYMBOL(vfs_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004207
4208/*
4209 * Hardlinks are often used in delicate situations. We avoid
4210 * security-related surprises by not following symlinks on the
4211 * newname. --KAB
4212 *
4213 * We don't follow them on the oldname either to be compatible
4214 * with linux 2.0, and to avoid hard-linking to directories
4215 * and other special files. --ADM
4216 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01004217SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4218 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004219{
4220 struct dentry *new_dentry;
Al Virodae6ad82011-06-26 11:50:15 -04004221 struct path old_path, new_path;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004222 struct inode *delegated_inode = NULL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304223 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004224 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004225
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304226 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004227 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304228 /*
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004229 * To use null names we require CAP_DAC_READ_SEARCH
4230 * This ensures that not everyone will be able to create
4231 * handlink using the passed filedescriptor.
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304232 */
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004233 if (flags & AT_EMPTY_PATH) {
4234 if (!capable(CAP_DAC_READ_SEARCH))
4235 return -ENOENT;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304236 how = LOOKUP_EMPTY;
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004237 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004238
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304239 if (flags & AT_SYMLINK_FOLLOW)
4240 how |= LOOKUP_FOLLOW;
Jeff Layton442e31c2012-12-20 16:15:38 -05004241retry:
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304242 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004243 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04004244 return error;
4245
Jeff Layton442e31c2012-12-20 16:15:38 -05004246 new_dentry = user_path_create(newdfd, newname, &new_path,
4247 (how & LOOKUP_REVAL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004248 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07004249 if (IS_ERR(new_dentry))
Al Virodae6ad82011-06-26 11:50:15 -04004250 goto out;
4251
4252 error = -EXDEV;
4253 if (old_path.mnt != new_path.mnt)
4254 goto out_dput;
Kees Cook800179c2012-07-25 17:29:07 -07004255 error = may_linkat(&old_path);
4256 if (unlikely(error))
4257 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04004258 error = security_path_link(old_path.dentry, &new_path, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004259 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04004260 goto out_dput;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004261 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
Dave Hansen75c3f292008-02-15 14:37:45 -08004262out_dput:
Al Viro921a1652012-07-20 01:15:31 +04004263 done_path_create(&new_path, new_dentry);
J. Bruce Fields146a8592011-09-20 17:14:31 -04004264 if (delegated_inode) {
4265 error = break_deleg_wait(&delegated_inode);
Oleg Drokind22e6332014-01-31 15:41:58 -05004266 if (!error) {
4267 path_put(&old_path);
J. Bruce Fields146a8592011-09-20 17:14:31 -04004268 goto retry;
Oleg Drokind22e6332014-01-31 15:41:58 -05004269 }
J. Bruce Fields146a8592011-09-20 17:14:31 -04004270 }
Jeff Layton442e31c2012-12-20 16:15:38 -05004271 if (retry_estale(error, how)) {
Oleg Drokind22e6332014-01-31 15:41:58 -05004272 path_put(&old_path);
Jeff Layton442e31c2012-12-20 16:15:38 -05004273 how |= LOOKUP_REVAL;
4274 goto retry;
4275 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004276out:
Al Viro2d8f3032008-07-22 09:59:21 -04004277 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004278
4279 return error;
4280}
4281
Heiko Carstens3480b252009-01-14 14:14:16 +01004282SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004283{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004284 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004285}
4286
Miklos Szeredibc270272014-04-01 17:08:42 +02004287/**
4288 * vfs_rename - rename a filesystem object
4289 * @old_dir: parent of source
4290 * @old_dentry: source
4291 * @new_dir: parent of destination
4292 * @new_dentry: destination
4293 * @delegated_inode: returns an inode needing a delegation break
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004294 * @flags: rename flags
Miklos Szeredibc270272014-04-01 17:08:42 +02004295 *
4296 * The caller must hold multiple mutexes--see lock_rename()).
4297 *
4298 * If vfs_rename discovers a delegation in need of breaking at either
4299 * the source or destination, it will return -EWOULDBLOCK and return a
4300 * reference to the inode in delegated_inode. The caller should then
4301 * break the delegation and retry. Because breaking a delegation may
4302 * take a long time, the caller should drop all locks before doing
4303 * so.
4304 *
4305 * Alternatively, a caller may pass NULL for delegated_inode. This may
4306 * be appropriate for callers that expect the underlying filesystem not
4307 * to be NFS exported.
4308 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07004309 * The worst of all namespace operations - renaming directory. "Perverted"
4310 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4311 * Problems:
J. Bruce Fieldsd03b29a2014-02-17 16:52:33 -05004312 * a) we can get into loop creation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004313 * b) race potential - two innocent renames can create a loop together.
4314 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004315 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07004316 * story.
J. Bruce Fields6cedba82012-03-05 11:40:41 -05004317 * c) we have to lock _four_ objects - parents and victim (if it exists),
4318 * and source (if it is not a directory).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004319 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07004320 * whether the target exists). Solution: try to be smart with locking
4321 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004322 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07004323 * move will be locked. Thus we can rank directories by the tree
4324 * (ancestors first) and rank all non-directories after them.
4325 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004326 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004327 * HOWEVER, it relies on the assumption that any object with ->lookup()
4328 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4329 * we'd better make sure that there's no link(2) for them.
Sage Weile4eaac02011-05-24 13:06:07 -07004330 * d) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004331 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07004332 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05004333 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07004334 * locking].
4335 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004336int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004337 struct inode *new_dir, struct dentry *new_dentry,
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004338 struct inode **delegated_inode, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004339{
4340 int error;
Miklos Szeredibc270272014-04-01 17:08:42 +02004341 bool is_dir = d_is_dir(old_dentry);
Eric Paris59b0df22010-02-08 12:53:52 -05004342 const unsigned char *old_name;
Miklos Szeredibc270272014-04-01 17:08:42 +02004343 struct inode *source = old_dentry->d_inode;
4344 struct inode *target = new_dentry->d_inode;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004345 bool new_is_dir = false;
4346 unsigned max_links = new_dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004347
Miklos Szeredi9409e222016-05-11 01:16:37 +02004348 /*
4349 * Check source == target.
4350 * On overlayfs need to look at underlying inodes.
4351 */
Miklos Szeredi2d902672016-06-30 08:53:27 +02004352 if (d_real_inode(old_dentry) == d_real_inode(new_dentry))
Miklos Szeredibc270272014-04-01 17:08:42 +02004353 return 0;
4354
Linus Torvalds1da177e2005-04-16 15:20:36 -07004355 error = may_delete(old_dir, old_dentry, is_dir);
4356 if (error)
4357 return error;
4358
Miklos Szeredida1ce062014-04-01 17:08:43 +02004359 if (!target) {
Miklos Szeredia95164d2008-07-30 15:08:48 +02004360 error = may_create(new_dir, new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004361 } else {
4362 new_is_dir = d_is_dir(new_dentry);
4363
4364 if (!(flags & RENAME_EXCHANGE))
4365 error = may_delete(new_dir, new_dentry, is_dir);
4366 else
4367 error = may_delete(new_dir, new_dentry, new_is_dir);
4368 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004369 if (error)
4370 return error;
4371
Miklos Szeredi2773bf02016-09-27 11:03:58 +02004372 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004373 return -EPERM;
4374
Miklos Szeredibc270272014-04-01 17:08:42 +02004375 /*
4376 * If we are going to change the parent - check write permissions,
4377 * we'll need to flip '..'.
4378 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02004379 if (new_dir != old_dir) {
4380 if (is_dir) {
4381 error = inode_permission(source, MAY_WRITE);
4382 if (error)
4383 return error;
4384 }
4385 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4386 error = inode_permission(target, MAY_WRITE);
4387 if (error)
4388 return error;
4389 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004390 }
Robert Love0eeca282005-07-12 17:06:03 -04004391
Miklos Szeredi0b3974e2014-04-01 17:08:43 +02004392 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4393 flags);
Miklos Szeredibc270272014-04-01 17:08:42 +02004394 if (error)
4395 return error;
4396
4397 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4398 dget(new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004399 if (!is_dir || (flags & RENAME_EXCHANGE))
Miklos Szeredibc270272014-04-01 17:08:42 +02004400 lock_two_nondirectories(source, target);
4401 else if (target)
Al Viro59551022016-01-22 15:40:57 -05004402 inode_lock(target);
Miklos Szeredibc270272014-04-01 17:08:42 +02004403
4404 error = -EBUSY;
Eric W. Biederman7af13642013-10-04 19:15:13 -07004405 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
Miklos Szeredibc270272014-04-01 17:08:42 +02004406 goto out;
4407
Miklos Szeredida1ce062014-04-01 17:08:43 +02004408 if (max_links && new_dir != old_dir) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004409 error = -EMLINK;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004410 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
Miklos Szeredibc270272014-04-01 17:08:42 +02004411 goto out;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004412 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4413 old_dir->i_nlink >= max_links)
4414 goto out;
4415 }
4416 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4417 shrink_dcache_parent(new_dentry);
4418 if (!is_dir) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004419 error = try_break_deleg(source, delegated_inode);
4420 if (error)
4421 goto out;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004422 }
4423 if (target && !new_is_dir) {
4424 error = try_break_deleg(target, delegated_inode);
4425 if (error)
4426 goto out;
Miklos Szeredibc270272014-04-01 17:08:42 +02004427 }
Miklos Szeredi2773bf02016-09-27 11:03:58 +02004428 error = old_dir->i_op->rename(old_dir, old_dentry,
Miklos Szeredi18fc84d2016-09-27 11:03:58 +02004429 new_dir, new_dentry, flags);
Miklos Szeredibc270272014-04-01 17:08:42 +02004430 if (error)
4431 goto out;
4432
Miklos Szeredida1ce062014-04-01 17:08:43 +02004433 if (!(flags & RENAME_EXCHANGE) && target) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004434 if (is_dir)
4435 target->i_flags |= S_DEAD;
4436 dont_mount(new_dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07004437 detach_mounts(new_dentry);
Miklos Szeredibc270272014-04-01 17:08:42 +02004438 }
Miklos Szeredida1ce062014-04-01 17:08:43 +02004439 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4440 if (!(flags & RENAME_EXCHANGE))
4441 d_move(old_dentry, new_dentry);
4442 else
4443 d_exchange(old_dentry, new_dentry);
4444 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004445out:
Miklos Szeredida1ce062014-04-01 17:08:43 +02004446 if (!is_dir || (flags & RENAME_EXCHANGE))
Miklos Szeredibc270272014-04-01 17:08:42 +02004447 unlock_two_nondirectories(source, target);
4448 else if (target)
Al Viro59551022016-01-22 15:40:57 -05004449 inode_unlock(target);
Miklos Szeredibc270272014-04-01 17:08:42 +02004450 dput(new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004451 if (!error) {
Al Viro123df292009-12-25 04:57:57 -05004452 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Miklos Szeredida1ce062014-04-01 17:08:43 +02004453 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4454 if (flags & RENAME_EXCHANGE) {
4455 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4456 new_is_dir, NULL, new_dentry);
4457 }
4458 }
Robert Love0eeca282005-07-12 17:06:03 -04004459 fsnotify_oldname_free(old_name);
4460
Linus Torvalds1da177e2005-04-16 15:20:36 -07004461 return error;
4462}
Al Viro4d359502014-03-14 12:20:17 -04004463EXPORT_SYMBOL(vfs_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004464
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004465SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4466 int, newdfd, const char __user *, newname, unsigned int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004467{
Al Viro2ad94ae2008-07-21 09:32:51 -04004468 struct dentry *old_dentry, *new_dentry;
4469 struct dentry *trap;
Al Virof5beed72015-04-30 16:09:11 -04004470 struct path old_path, new_path;
4471 struct qstr old_last, new_last;
4472 int old_type, new_type;
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004473 struct inode *delegated_inode = NULL;
Jeff Layton91a27b22012-10-10 15:25:28 -04004474 struct filename *from;
4475 struct filename *to;
Al Virof5beed72015-04-30 16:09:11 -04004476 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
Jeff Laytonc6a94282012-12-11 12:10:10 -05004477 bool should_retry = false;
Al Viro2ad94ae2008-07-21 09:32:51 -04004478 int error;
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004479
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004480 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredida1ce062014-04-01 17:08:43 +02004481 return -EINVAL;
4482
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004483 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4484 (flags & RENAME_EXCHANGE))
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004485 return -EINVAL;
4486
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004487 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4488 return -EPERM;
4489
Al Virof5beed72015-04-30 16:09:11 -04004490 if (flags & RENAME_EXCHANGE)
4491 target_flags = 0;
4492
Jeff Laytonc6a94282012-12-11 12:10:10 -05004493retry:
Al Virof5beed72015-04-30 16:09:11 -04004494 from = user_path_parent(olddfd, oldname,
4495 &old_path, &old_last, &old_type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04004496 if (IS_ERR(from)) {
4497 error = PTR_ERR(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004498 goto exit;
Jeff Layton91a27b22012-10-10 15:25:28 -04004499 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004500
Al Virof5beed72015-04-30 16:09:11 -04004501 to = user_path_parent(newdfd, newname,
4502 &new_path, &new_last, &new_type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04004503 if (IS_ERR(to)) {
4504 error = PTR_ERR(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004505 goto exit1;
Jeff Layton91a27b22012-10-10 15:25:28 -04004506 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004507
4508 error = -EXDEV;
Al Virof5beed72015-04-30 16:09:11 -04004509 if (old_path.mnt != new_path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004510 goto exit2;
4511
Linus Torvalds1da177e2005-04-16 15:20:36 -07004512 error = -EBUSY;
Al Virof5beed72015-04-30 16:09:11 -04004513 if (old_type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004514 goto exit2;
4515
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004516 if (flags & RENAME_NOREPLACE)
4517 error = -EEXIST;
Al Virof5beed72015-04-30 16:09:11 -04004518 if (new_type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004519 goto exit2;
4520
Al Virof5beed72015-04-30 16:09:11 -04004521 error = mnt_want_write(old_path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02004522 if (error)
4523 goto exit2;
4524
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004525retry_deleg:
Al Virof5beed72015-04-30 16:09:11 -04004526 trap = lock_rename(new_path.dentry, old_path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004527
Al Virof5beed72015-04-30 16:09:11 -04004528 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004529 error = PTR_ERR(old_dentry);
4530 if (IS_ERR(old_dentry))
4531 goto exit3;
4532 /* source must exist */
4533 error = -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01004534 if (d_is_negative(old_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004535 goto exit4;
Al Virof5beed72015-04-30 16:09:11 -04004536 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004537 error = PTR_ERR(new_dentry);
4538 if (IS_ERR(new_dentry))
4539 goto exit4;
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004540 error = -EEXIST;
4541 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4542 goto exit5;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004543 if (flags & RENAME_EXCHANGE) {
4544 error = -ENOENT;
4545 if (d_is_negative(new_dentry))
4546 goto exit5;
4547
4548 if (!d_is_dir(new_dentry)) {
4549 error = -ENOTDIR;
Al Virof5beed72015-04-30 16:09:11 -04004550 if (new_last.name[new_last.len])
Miklos Szeredida1ce062014-04-01 17:08:43 +02004551 goto exit5;
4552 }
4553 }
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004554 /* unless the source is a directory trailing slashes give -ENOTDIR */
4555 if (!d_is_dir(old_dentry)) {
4556 error = -ENOTDIR;
Al Virof5beed72015-04-30 16:09:11 -04004557 if (old_last.name[old_last.len])
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004558 goto exit5;
Al Virof5beed72015-04-30 16:09:11 -04004559 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004560 goto exit5;
4561 }
4562 /* source should not be ancestor of target */
4563 error = -EINVAL;
4564 if (old_dentry == trap)
4565 goto exit5;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004566 /* target should not be an ancestor of source */
Miklos Szeredida1ce062014-04-01 17:08:43 +02004567 if (!(flags & RENAME_EXCHANGE))
4568 error = -ENOTEMPTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004569 if (new_dentry == trap)
4570 goto exit5;
4571
Al Virof5beed72015-04-30 16:09:11 -04004572 error = security_path_rename(&old_path, old_dentry,
4573 &new_path, new_dentry, flags);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004574 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02004575 goto exit5;
Al Virof5beed72015-04-30 16:09:11 -04004576 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4577 new_path.dentry->d_inode, new_dentry,
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004578 &delegated_inode, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004579exit5:
4580 dput(new_dentry);
4581exit4:
4582 dput(old_dentry);
4583exit3:
Al Virof5beed72015-04-30 16:09:11 -04004584 unlock_rename(new_path.dentry, old_path.dentry);
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004585 if (delegated_inode) {
4586 error = break_deleg_wait(&delegated_inode);
4587 if (!error)
4588 goto retry_deleg;
4589 }
Al Virof5beed72015-04-30 16:09:11 -04004590 mnt_drop_write(old_path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004591exit2:
Jeff Laytonc6a94282012-12-11 12:10:10 -05004592 if (retry_estale(error, lookup_flags))
4593 should_retry = true;
Al Virof5beed72015-04-30 16:09:11 -04004594 path_put(&new_path);
Al Viro2ad94ae2008-07-21 09:32:51 -04004595 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004596exit1:
Al Virof5beed72015-04-30 16:09:11 -04004597 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004598 putname(from);
Jeff Laytonc6a94282012-12-11 12:10:10 -05004599 if (should_retry) {
4600 should_retry = false;
4601 lookup_flags |= LOOKUP_REVAL;
4602 goto retry;
4603 }
Al Viro2ad94ae2008-07-21 09:32:51 -04004604exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004605 return error;
4606}
4607
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004608SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4609 int, newdfd, const char __user *, newname)
4610{
4611 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4612}
4613
Heiko Carstensa26eab22009-01-14 14:14:17 +01004614SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004615{
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004616 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004617}
4618
Miklos Szeredi787fb6b2014-10-24 00:14:36 +02004619int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4620{
4621 int error = may_create(dir, dentry);
4622 if (error)
4623 return error;
4624
4625 if (!dir->i_op->mknod)
4626 return -EPERM;
4627
4628 return dir->i_op->mknod(dir, dentry,
4629 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4630}
4631EXPORT_SYMBOL(vfs_whiteout);
4632
Al Viro5d826c82014-03-14 13:42:45 -04004633int readlink_copy(char __user *buffer, int buflen, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004634{
Al Viro5d826c82014-03-14 13:42:45 -04004635 int len = PTR_ERR(link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004636 if (IS_ERR(link))
4637 goto out;
4638
4639 len = strlen(link);
4640 if (len > (unsigned) buflen)
4641 len = buflen;
4642 if (copy_to_user(buffer, link, len))
4643 len = -EFAULT;
4644out:
4645 return len;
4646}
4647
4648/*
4649 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
Al Viro6b255392015-11-17 10:20:54 -05004650 * have ->get_link() not calling nd_jump_link(). Using (or not using) it
4651 * for any given inode is up to filesystem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004652 */
4653int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4654{
Al Virofceef392015-12-29 15:58:39 -05004655 DEFINE_DELAYED_CALL(done);
Al Viro5f2c4172015-05-07 11:14:26 -04004656 struct inode *inode = d_inode(dentry);
4657 const char *link = inode->i_link;
Marcin Slusarz694a1762008-06-09 16:40:37 -07004658 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004659
Al Virod4dee482015-04-30 20:08:02 -04004660 if (!link) {
Al Virofceef392015-12-29 15:58:39 -05004661 link = inode->i_op->get_link(dentry, inode, &done);
Al Virod4dee482015-04-30 20:08:02 -04004662 if (IS_ERR(link))
4663 return PTR_ERR(link);
4664 }
Al Viro680baac2015-05-02 13:32:22 -04004665 res = readlink_copy(buffer, buflen, link);
Al Virofceef392015-12-29 15:58:39 -05004666 do_delayed_call(&done);
Marcin Slusarz694a1762008-06-09 16:40:37 -07004667 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004668}
Al Viro4d359502014-03-14 12:20:17 -04004669EXPORT_SYMBOL(generic_readlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004670
Miklos Szeredid60874c2016-10-04 14:40:45 +02004671/**
4672 * vfs_get_link - get symlink body
4673 * @dentry: dentry on which to get symbolic link
4674 * @done: caller needs to free returned data with this
4675 *
4676 * Calls security hook and i_op->get_link() on the supplied inode.
4677 *
4678 * It does not touch atime. That's up to the caller if necessary.
4679 *
4680 * Does not work on "special" symlinks like /proc/$$/fd/N
4681 */
4682const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4683{
4684 const char *res = ERR_PTR(-EINVAL);
4685 struct inode *inode = d_inode(dentry);
4686
4687 if (d_is_symlink(dentry)) {
4688 res = ERR_PTR(security_inode_readlink(dentry));
4689 if (!res)
4690 res = inode->i_op->get_link(dentry, inode, done);
4691 }
4692 return res;
4693}
4694EXPORT_SYMBOL(vfs_get_link);
4695
Linus Torvalds1da177e2005-04-16 15:20:36 -07004696/* get the link contents into pagecache */
Al Viro6b255392015-11-17 10:20:54 -05004697const char *page_get_link(struct dentry *dentry, struct inode *inode,
Al Virofceef392015-12-29 15:58:39 -05004698 struct delayed_call *callback)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004699{
Duane Griffinebd09ab2008-12-19 20:47:12 +00004700 char *kaddr;
4701 struct page *page;
Al Viro6b255392015-11-17 10:20:54 -05004702 struct address_space *mapping = inode->i_mapping;
4703
Al Virod3883d42015-11-17 10:41:04 -05004704 if (!dentry) {
4705 page = find_get_page(mapping, 0);
4706 if (!page)
4707 return ERR_PTR(-ECHILD);
4708 if (!PageUptodate(page)) {
4709 put_page(page);
4710 return ERR_PTR(-ECHILD);
4711 }
4712 } else {
4713 page = read_mapping_page(mapping, 0, NULL);
4714 if (IS_ERR(page))
4715 return (char*)page;
4716 }
Al Virofceef392015-12-29 15:58:39 -05004717 set_delayed_call(callback, page_put_link, page);
Al Viro21fc61c2015-11-17 01:07:57 -05004718 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4719 kaddr = page_address(page);
Al Viro6b255392015-11-17 10:20:54 -05004720 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
Duane Griffinebd09ab2008-12-19 20:47:12 +00004721 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004722}
4723
Al Viro6b255392015-11-17 10:20:54 -05004724EXPORT_SYMBOL(page_get_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004725
Al Virofceef392015-12-29 15:58:39 -05004726void page_put_link(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004727{
Al Virofceef392015-12-29 15:58:39 -05004728 put_page(arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004729}
Al Viro4d359502014-03-14 12:20:17 -04004730EXPORT_SYMBOL(page_put_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004731
Al Viroaa80dea2015-11-16 18:26:34 -05004732int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4733{
Al Virofceef392015-12-29 15:58:39 -05004734 DEFINE_DELAYED_CALL(done);
Al Viro6b255392015-11-17 10:20:54 -05004735 int res = readlink_copy(buffer, buflen,
4736 page_get_link(dentry, d_inode(dentry),
Al Virofceef392015-12-29 15:58:39 -05004737 &done));
4738 do_delayed_call(&done);
Al Viroaa80dea2015-11-16 18:26:34 -05004739 return res;
4740}
4741EXPORT_SYMBOL(page_readlink);
4742
Nick Piggin54566b22009-01-04 12:00:53 -08004743/*
4744 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4745 */
4746int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004747{
4748 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004749 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07004750 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08004751 int err;
Nick Piggin54566b22009-01-04 12:00:53 -08004752 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4753 if (nofs)
4754 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004755
NeilBrown7e53cac2006-03-25 03:07:57 -08004756retry:
Nick Pigginafddba42007-10-16 01:25:01 -07004757 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08004758 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004759 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07004760 goto fail;
4761
Al Viro21fc61c2015-11-17 01:07:57 -05004762 memcpy(page_address(page), symname, len-1);
Nick Pigginafddba42007-10-16 01:25:01 -07004763
4764 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4765 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004766 if (err < 0)
4767 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07004768 if (err < len-1)
4769 goto retry;
4770
Linus Torvalds1da177e2005-04-16 15:20:36 -07004771 mark_inode_dirty(inode);
4772 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004773fail:
4774 return err;
4775}
Al Viro4d359502014-03-14 12:20:17 -04004776EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004777
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004778int page_symlink(struct inode *inode, const char *symname, int len)
4779{
4780 return __page_symlink(inode, symname, len,
Michal Hockoc62d2552015-11-06 16:28:49 -08004781 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004782}
Al Viro4d359502014-03-14 12:20:17 -04004783EXPORT_SYMBOL(page_symlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004784
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08004785const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004786 .readlink = generic_readlink,
Al Viro6b255392015-11-17 10:20:54 -05004787 .get_link = page_get_link,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004788};
Linus Torvalds1da177e2005-04-16 15:20:36 -07004789EXPORT_SYMBOL(page_symlink_inode_operations);