blob: 79177c954c1bb85f2ee7deeb23c7575fa4513585 [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 */
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -0700378static inline int do_inode_permission(struct vfsmount *mnt, struct inode *inode, int mask)
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700379{
380 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -0700381 if (likely(mnt && inode->i_op->permission2))
382 return inode->i_op->permission2(mnt, inode, mask);
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700383 if (likely(inode->i_op->permission))
384 return inode->i_op->permission(inode, mask);
385
386 /* This gets set once for the inode lifetime */
387 spin_lock(&inode->i_lock);
388 inode->i_opflags |= IOP_FASTPERM;
389 spin_unlock(&inode->i_lock);
390 }
391 return generic_permission(inode, mask);
392}
393
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200394/**
David Howells0bdaea92012-06-25 12:55:46 +0100395 * __inode_permission - Check for access rights to a given inode
396 * @inode: Inode to check permission on
397 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200398 *
David Howells0bdaea92012-06-25 12:55:46 +0100399 * Check for read/write/execute permissions on an inode.
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530400 *
401 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
David Howells0bdaea92012-06-25 12:55:46 +0100402 *
403 * This does not check for a read-only file system. You probably want
404 * inode_permission().
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200405 */
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -0700406int __inode_permission2(struct vfsmount *mnt, struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407{
Al Viroe6305c42008-07-15 21:03:57 -0400408 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700410 if (unlikely(mask & MAY_WRITE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 /*
412 * Nobody gets write access to an immutable file.
413 */
414 if (IS_IMMUTABLE(inode))
Eryu Guan337684a2016-08-02 19:58:28 +0800415 return -EPERM;
Eric W. Biederman0bd23d092016-06-29 14:54:46 -0500416
417 /*
418 * Updating mtime will likely cause i_uid and i_gid to be
419 * written back improperly if their true value is unknown
420 * to the vfs.
421 */
422 if (HAS_UNMAPPED_ID(inode))
423 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 }
425
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -0700426 retval = do_inode_permission(mnt, inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 if (retval)
428 return retval;
429
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -0700430 retval = devcgroup_inode_permission(inode, mask);
431 if (retval)
432 return retval;
433
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -0700434 retval = security_inode_permission(inode, mask);
435 return retval;
436}
437EXPORT_SYMBOL(__inode_permission2);
438
439int __inode_permission(struct inode *inode, int mask)
440{
441 return __inode_permission2(NULL, inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442}
Miklos Szeredibd5d0852014-10-24 00:14:35 +0200443EXPORT_SYMBOL(__inode_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444
Al Virof4d6ff82011-06-19 13:14:21 -0400445/**
David Howells0bdaea92012-06-25 12:55:46 +0100446 * sb_permission - Check superblock-level permissions
447 * @sb: Superblock of inode to check permission on
Randy Dunlap55852632012-08-18 17:39:25 -0700448 * @inode: Inode to check permission on
David Howells0bdaea92012-06-25 12:55:46 +0100449 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
450 *
451 * Separate out file-system wide checks from inode-specific permission checks.
452 */
453static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
454{
455 if (unlikely(mask & MAY_WRITE)) {
456 umode_t mode = inode->i_mode;
457
458 /* Nobody gets write access to a read-only fs. */
459 if ((sb->s_flags & MS_RDONLY) &&
460 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
461 return -EROFS;
462 }
463 return 0;
464}
465
466/**
467 * inode_permission - Check for access rights to a given inode
468 * @inode: Inode to check permission on
469 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
470 *
471 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
472 * this, letting us set arbitrary permissions for filesystem access without
473 * changing the "normal" UIDs which are used for other things.
474 *
475 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
476 */
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -0700477int inode_permission2(struct vfsmount *mnt, struct inode *inode, int mask)
David Howells0bdaea92012-06-25 12:55:46 +0100478{
479 int retval;
480
481 retval = sb_permission(inode->i_sb, inode, mask);
482 if (retval)
483 return retval;
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -0700484 return __inode_permission2(mnt, inode, mask);
485}
486EXPORT_SYMBOL(inode_permission2);
487
488int inode_permission(struct inode *inode, int mask)
489{
490 return inode_permission2(NULL, inode, mask);
David Howells0bdaea92012-06-25 12:55:46 +0100491}
Al Viro4d359502014-03-14 12:20:17 -0400492EXPORT_SYMBOL(inode_permission);
David Howells0bdaea92012-06-25 12:55:46 +0100493
494/**
Jan Blunck5dd784d02008-02-14 19:34:38 -0800495 * path_get - get a reference to a path
496 * @path: path to get the reference to
497 *
498 * Given a path increment the reference count to the dentry and the vfsmount.
499 */
Al Virodcf787f2013-03-01 23:51:07 -0500500void path_get(const struct path *path)
Jan Blunck5dd784d02008-02-14 19:34:38 -0800501{
502 mntget(path->mnt);
503 dget(path->dentry);
504}
505EXPORT_SYMBOL(path_get);
506
507/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800508 * path_put - put a reference to a path
509 * @path: path to put the reference to
510 *
511 * Given a path decrement the reference count to the dentry and the vfsmount.
512 */
Al Virodcf787f2013-03-01 23:51:07 -0500513void path_put(const struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514{
Jan Blunck1d957f92008-02-14 19:34:35 -0800515 dput(path->dentry);
516 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517}
Jan Blunck1d957f92008-02-14 19:34:35 -0800518EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519
Al Viro894bc8c2015-05-02 07:16:16 -0400520#define EMBEDDED_LEVELS 2
Al Viro1f55a6e2014-11-01 19:30:41 -0400521struct nameidata {
522 struct path path;
Al Viro1cf26652015-05-06 16:01:56 -0400523 struct qstr last;
Al Viro1f55a6e2014-11-01 19:30:41 -0400524 struct path root;
525 struct inode *inode; /* path.dentry.d_inode */
526 unsigned int flags;
Al Viro9883d182015-05-13 07:28:08 -0400527 unsigned seq, m_seq;
Al Viro1f55a6e2014-11-01 19:30:41 -0400528 int last_type;
529 unsigned depth;
NeilBrown756daf22015-03-23 13:37:38 +1100530 int total_link_count;
Al Viro697fc6c2015-05-02 19:38:35 -0400531 struct saved {
532 struct path link;
Al Virofceef392015-12-29 15:58:39 -0500533 struct delayed_call done;
Al Viro697fc6c2015-05-02 19:38:35 -0400534 const char *name;
Al Viro0450b2d2015-05-08 13:23:53 -0400535 unsigned seq;
Al Viro894bc8c2015-05-02 07:16:16 -0400536 } *stack, internal[EMBEDDED_LEVELS];
Al Viro9883d182015-05-13 07:28:08 -0400537 struct filename *name;
538 struct nameidata *saved;
Al Virofceef392015-12-29 15:58:39 -0500539 struct inode *link_inode;
Al Viro9883d182015-05-13 07:28:08 -0400540 unsigned root_seq;
541 int dfd;
Al Viro1f55a6e2014-11-01 19:30:41 -0400542};
543
Al Viro9883d182015-05-13 07:28:08 -0400544static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
Al Viro894bc8c2015-05-02 07:16:16 -0400545{
NeilBrown756daf22015-03-23 13:37:38 +1100546 struct nameidata *old = current->nameidata;
547 p->stack = p->internal;
Al Viroc8a53ee2015-05-12 18:43:07 -0400548 p->dfd = dfd;
549 p->name = name;
NeilBrown756daf22015-03-23 13:37:38 +1100550 p->total_link_count = old ? old->total_link_count : 0;
Al Viro9883d182015-05-13 07:28:08 -0400551 p->saved = old;
NeilBrown756daf22015-03-23 13:37:38 +1100552 current->nameidata = p;
Al Viro894bc8c2015-05-02 07:16:16 -0400553}
554
Al Viro9883d182015-05-13 07:28:08 -0400555static void restore_nameidata(void)
Al Viro894bc8c2015-05-02 07:16:16 -0400556{
Al Viro9883d182015-05-13 07:28:08 -0400557 struct nameidata *now = current->nameidata, *old = now->saved;
NeilBrown756daf22015-03-23 13:37:38 +1100558
559 current->nameidata = old;
560 if (old)
561 old->total_link_count = now->total_link_count;
Al Viroe1a63bb2015-12-05 21:06:33 -0500562 if (now->stack != now->internal)
NeilBrown756daf22015-03-23 13:37:38 +1100563 kfree(now->stack);
Al Viro894bc8c2015-05-02 07:16:16 -0400564}
565
566static int __nd_alloc_stack(struct nameidata *nd)
567{
Al Virobc40aee2015-05-09 13:04:24 -0400568 struct saved *p;
569
570 if (nd->flags & LOOKUP_RCU) {
571 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
572 GFP_ATOMIC);
573 if (unlikely(!p))
574 return -ECHILD;
575 } else {
576 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
Al Viro894bc8c2015-05-02 07:16:16 -0400577 GFP_KERNEL);
Al Virobc40aee2015-05-09 13:04:24 -0400578 if (unlikely(!p))
579 return -ENOMEM;
580 }
Al Viro894bc8c2015-05-02 07:16:16 -0400581 memcpy(p, nd->internal, sizeof(nd->internal));
582 nd->stack = p;
583 return 0;
584}
585
Eric W. Biederman397d4252015-08-15 20:27:13 -0500586/**
587 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
588 * @path: nameidate to verify
589 *
590 * Rename can sometimes move a file or directory outside of a bind
591 * mount, path_connected allows those cases to be detected.
592 */
593static bool path_connected(const struct path *path)
594{
595 struct vfsmount *mnt = path->mnt;
596
597 /* Only bind mounts can have disconnected paths */
598 if (mnt->mnt_root == mnt->mnt_sb->s_root)
599 return true;
600
601 return is_subdir(path->dentry, mnt->mnt_root);
602}
603
Al Viro894bc8c2015-05-02 07:16:16 -0400604static inline int nd_alloc_stack(struct nameidata *nd)
605{
Al Viroda4e0be2015-05-03 20:52:15 -0400606 if (likely(nd->depth != EMBEDDED_LEVELS))
Al Viro894bc8c2015-05-02 07:16:16 -0400607 return 0;
608 if (likely(nd->stack != nd->internal))
609 return 0;
610 return __nd_alloc_stack(nd);
611}
612
Al Viro79733872015-05-09 12:55:43 -0400613static void drop_links(struct nameidata *nd)
614{
615 int i = nd->depth;
616 while (i--) {
617 struct saved *last = nd->stack + i;
Al Virofceef392015-12-29 15:58:39 -0500618 do_delayed_call(&last->done);
619 clear_delayed_call(&last->done);
Al Viro79733872015-05-09 12:55:43 -0400620 }
621}
622
623static void terminate_walk(struct nameidata *nd)
624{
625 drop_links(nd);
626 if (!(nd->flags & LOOKUP_RCU)) {
627 int i;
628 path_put(&nd->path);
629 for (i = 0; i < nd->depth; i++)
630 path_put(&nd->stack[i].link);
Al Viro102b8af2015-05-12 17:35:52 -0400631 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
632 path_put(&nd->root);
633 nd->root.mnt = NULL;
634 }
Al Viro79733872015-05-09 12:55:43 -0400635 } else {
636 nd->flags &= ~LOOKUP_RCU;
637 if (!(nd->flags & LOOKUP_ROOT))
638 nd->root.mnt = NULL;
639 rcu_read_unlock();
640 }
641 nd->depth = 0;
642}
643
644/* path_put is needed afterwards regardless of success or failure */
645static bool legitimize_path(struct nameidata *nd,
646 struct path *path, unsigned seq)
647{
648 int res = __legitimize_mnt(path->mnt, nd->m_seq);
649 if (unlikely(res)) {
650 if (res > 0)
651 path->mnt = NULL;
652 path->dentry = NULL;
653 return false;
654 }
655 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
656 path->dentry = NULL;
657 return false;
658 }
659 return !read_seqcount_retry(&path->dentry->d_seq, seq);
660}
661
662static bool legitimize_links(struct nameidata *nd)
663{
664 int i;
665 for (i = 0; i < nd->depth; i++) {
666 struct saved *last = nd->stack + i;
667 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
668 drop_links(nd);
669 nd->depth = i + 1;
670 return false;
671 }
672 }
673 return true;
674}
675
Al Viro19660af2011-03-25 10:32:48 -0400676/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100677 * Path walking has 2 modes, rcu-walk and ref-walk (see
Al Viro19660af2011-03-25 10:32:48 -0400678 * Documentation/filesystems/path-lookup.txt). In situations when we can't
679 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
Mike Marshall57e37152015-11-30 11:11:59 -0500680 * normal reference counts on dentries and vfsmounts to transition to ref-walk
Al Viro19660af2011-03-25 10:32:48 -0400681 * mode. Refcounts are grabbed at the last known good point before rcu-walk
682 * got stuck, so ref-walk may continue from there. If this is not successful
683 * (eg. a seqcount has changed), then failure is returned and it's up to caller
684 * to restart the path walk from the beginning in ref-walk mode.
Nick Piggin31e6b012011-01-07 17:49:52 +1100685 */
Nick Piggin31e6b012011-01-07 17:49:52 +1100686
687/**
Al Viro19660af2011-03-25 10:32:48 -0400688 * unlazy_walk - try to switch to ref-walk mode.
689 * @nd: nameidata pathwalk data
690 * @dentry: child of nd->path.dentry or NULL
Al Viro6e9918b2015-05-05 09:26:05 -0400691 * @seq: seq number to check dentry against
Randy Dunlap39191622011-01-08 19:36:21 -0800692 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100693 *
Al Viro19660af2011-03-25 10:32:48 -0400694 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
695 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
696 * @nd or NULL. Must be called from rcu-walk context.
Al Viro79733872015-05-09 12:55:43 -0400697 * Nothing should touch nameidata between unlazy_walk() failure and
698 * terminate_walk().
Nick Piggin31e6b012011-01-07 17:49:52 +1100699 */
Al Viro6e9918b2015-05-05 09:26:05 -0400700static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq)
Nick Piggin31e6b012011-01-07 17:49:52 +1100701{
Nick Piggin31e6b012011-01-07 17:49:52 +1100702 struct dentry *parent = nd->path.dentry;
703
704 BUG_ON(!(nd->flags & LOOKUP_RCU));
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700705
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700706 nd->flags &= ~LOOKUP_RCU;
Al Viro79733872015-05-09 12:55:43 -0400707 if (unlikely(!legitimize_links(nd)))
708 goto out2;
709 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
710 goto out2;
711 if (unlikely(!lockref_get_not_dead(&parent->d_lockref)))
712 goto out1;
Al Viro48a066e2013-09-29 22:06:07 -0400713
Linus Torvalds15570082013-09-02 11:38:06 -0700714 /*
715 * For a negative lookup, the lookup sequence point is the parents
716 * sequence point, and it only needs to revalidate the parent dentry.
717 *
718 * For a positive lookup, we need to move both the parent and the
719 * dentry from the RCU domain to be properly refcounted. And the
720 * sequence number in the dentry validates *both* dentry counters,
721 * since we checked the sequence number of the parent after we got
722 * the child sequence number. So we know the parent must still
723 * be valid if the child sequence number is still valid.
724 */
Al Viro19660af2011-03-25 10:32:48 -0400725 if (!dentry) {
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700726 if (read_seqcount_retry(&parent->d_seq, nd->seq))
727 goto out;
Al Viro19660af2011-03-25 10:32:48 -0400728 BUG_ON(nd->inode != parent->d_inode);
729 } else {
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700730 if (!lockref_get_not_dead(&dentry->d_lockref))
731 goto out;
Al Viro6e9918b2015-05-05 09:26:05 -0400732 if (read_seqcount_retry(&dentry->d_seq, seq))
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700733 goto drop_dentry;
Al Viro19660af2011-03-25 10:32:48 -0400734 }
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700735
736 /*
737 * Sequence counts matched. Now make sure that the root is
738 * still valid and get it if required.
739 */
740 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
Al Viro5a8d87e2015-05-12 00:10:18 -0400741 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
742 rcu_read_unlock();
743 dput(dentry);
744 return -ECHILD;
Al Viro79733872015-05-09 12:55:43 -0400745 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100746 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100747
Al Viro8b61e742013-11-08 12:45:01 -0500748 rcu_read_unlock();
Nick Piggin31e6b012011-01-07 17:49:52 +1100749 return 0;
Al Viro19660af2011-03-25 10:32:48 -0400750
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700751drop_dentry:
Al Viro8b61e742013-11-08 12:45:01 -0500752 rcu_read_unlock();
Linus Torvalds15570082013-09-02 11:38:06 -0700753 dput(dentry);
Linus Torvaldsd0d27272013-09-10 12:17:49 -0700754 goto drop_root_mnt;
Al Viro79733872015-05-09 12:55:43 -0400755out2:
756 nd->path.mnt = NULL;
757out1:
758 nd->path.dentry = NULL;
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700759out:
Al Viro8b61e742013-11-08 12:45:01 -0500760 rcu_read_unlock();
Linus Torvaldsd0d27272013-09-10 12:17:49 -0700761drop_root_mnt:
762 if (!(nd->flags & LOOKUP_ROOT))
763 nd->root.mnt = NULL;
Nick Piggin31e6b012011-01-07 17:49:52 +1100764 return -ECHILD;
765}
766
Al Viro79733872015-05-09 12:55:43 -0400767static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq)
768{
769 if (unlikely(!legitimize_path(nd, link, seq))) {
770 drop_links(nd);
771 nd->depth = 0;
772 nd->flags &= ~LOOKUP_RCU;
773 nd->path.mnt = NULL;
774 nd->path.dentry = NULL;
775 if (!(nd->flags & LOOKUP_ROOT))
776 nd->root.mnt = NULL;
777 rcu_read_unlock();
778 } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) {
779 return 0;
780 }
781 path_put(link);
782 return -ECHILD;
783}
784
Al Viro4ce16ef32012-06-10 16:10:59 -0400785static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
Nick Piggin34286d62011-01-07 17:49:57 +1100786{
Al Viro4ce16ef32012-06-10 16:10:59 -0400787 return dentry->d_op->d_revalidate(dentry, flags);
Nick Piggin34286d62011-01-07 17:49:57 +1100788}
789
Al Viro9f1fafe2011-03-25 11:00:12 -0400790/**
791 * complete_walk - successful completion of path walk
792 * @nd: pointer nameidata
Jeff Layton39159de2009-12-07 12:01:50 -0500793 *
Al Viro9f1fafe2011-03-25 11:00:12 -0400794 * If we had been in RCU mode, drop out of it and legitimize nd->path.
795 * Revalidate the final result, unless we'd already done that during
796 * the path walk or the filesystem doesn't ask for it. Return 0 on
797 * success, -error on failure. In case of failure caller does not
798 * need to drop nd->path.
Jeff Layton39159de2009-12-07 12:01:50 -0500799 */
Al Viro9f1fafe2011-03-25 11:00:12 -0400800static int complete_walk(struct nameidata *nd)
Jeff Layton39159de2009-12-07 12:01:50 -0500801{
Al Viro16c2cd72011-02-22 15:50:10 -0500802 struct dentry *dentry = nd->path.dentry;
Jeff Layton39159de2009-12-07 12:01:50 -0500803 int status;
Jeff Layton39159de2009-12-07 12:01:50 -0500804
Al Viro9f1fafe2011-03-25 11:00:12 -0400805 if (nd->flags & LOOKUP_RCU) {
Al Viro9f1fafe2011-03-25 11:00:12 -0400806 if (!(nd->flags & LOOKUP_ROOT))
807 nd->root.mnt = NULL;
Al Viro6e9918b2015-05-05 09:26:05 -0400808 if (unlikely(unlazy_walk(nd, NULL, 0)))
Al Viro48a066e2013-09-29 22:06:07 -0400809 return -ECHILD;
Al Viro9f1fafe2011-03-25 11:00:12 -0400810 }
811
Al Viro16c2cd72011-02-22 15:50:10 -0500812 if (likely(!(nd->flags & LOOKUP_JUMPED)))
Jeff Layton39159de2009-12-07 12:01:50 -0500813 return 0;
814
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500815 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
Al Viro16c2cd72011-02-22 15:50:10 -0500816 return 0;
817
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500818 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
Jeff Layton39159de2009-12-07 12:01:50 -0500819 if (status > 0)
820 return 0;
821
Al Viro16c2cd72011-02-22 15:50:10 -0500822 if (!status)
Jeff Layton39159de2009-12-07 12:01:50 -0500823 status = -ESTALE;
Al Viro16c2cd72011-02-22 15:50:10 -0500824
Jeff Layton39159de2009-12-07 12:01:50 -0500825 return status;
826}
827
Al Viro18d8c862015-05-12 16:32:34 -0400828static void set_root(struct nameidata *nd)
Al Viro2a737872009-04-07 11:49:53 -0400829{
Al Viro7bd88372014-09-13 21:55:46 -0400830 struct fs_struct *fs = current->fs;
Nick Pigginc28cc362011-01-07 17:49:53 +1100831
Al Viro9e6697e2015-12-05 20:07:21 -0500832 if (nd->flags & LOOKUP_RCU) {
833 unsigned seq;
834
835 do {
836 seq = read_seqcount_begin(&fs->seq);
837 nd->root = fs->root;
838 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
839 } while (read_seqcount_retry(&fs->seq, seq));
840 } else {
841 get_fs_root(fs, &nd->root);
842 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100843}
844
Jan Blunck1d957f92008-02-14 19:34:35 -0800845static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700846{
847 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800848 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700849 mntput(path->mnt);
850}
851
Nick Piggin7b9337a2011-01-14 08:42:43 +0000852static inline void path_to_nameidata(const struct path *path,
853 struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700854{
Nick Piggin31e6b012011-01-07 17:49:52 +1100855 if (!(nd->flags & LOOKUP_RCU)) {
856 dput(nd->path.dentry);
857 if (nd->path.mnt != path->mnt)
858 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800859 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100860 nd->path.mnt = path->mnt;
Jan Blunck4ac91372008-02-14 19:34:32 -0800861 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700862}
863
Al Viro248fb5b2015-12-05 20:51:58 -0500864static int nd_jump_root(struct nameidata *nd)
865{
866 if (nd->flags & LOOKUP_RCU) {
867 struct dentry *d;
868 nd->path = nd->root;
869 d = nd->path.dentry;
870 nd->inode = d->d_inode;
871 nd->seq = nd->root_seq;
872 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
873 return -ECHILD;
874 } else {
875 path_put(&nd->path);
876 nd->path = nd->root;
877 path_get(&nd->path);
878 nd->inode = nd->path.dentry->d_inode;
879 }
880 nd->flags |= LOOKUP_JUMPED;
881 return 0;
882}
883
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400884/*
Al Viro6b255392015-11-17 10:20:54 -0500885 * Helper to directly jump to a known parsed path from ->get_link,
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400886 * caller must have taken a reference to path beforehand.
887 */
Al Viro6e771372015-05-02 13:37:52 -0400888void nd_jump_link(struct path *path)
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400889{
Al Viro6e771372015-05-02 13:37:52 -0400890 struct nameidata *nd = current->nameidata;
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400891 path_put(&nd->path);
892
893 nd->path = *path;
894 nd->inode = nd->path.dentry->d_inode;
895 nd->flags |= LOOKUP_JUMPED;
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400896}
897
Al Virob9ff4422015-05-02 20:19:23 -0400898static inline void put_link(struct nameidata *nd)
Al Viro574197e2011-03-14 22:20:34 -0400899{
Al Viro21c30032015-05-03 21:06:24 -0400900 struct saved *last = nd->stack + --nd->depth;
Al Virofceef392015-12-29 15:58:39 -0500901 do_delayed_call(&last->done);
Al Viro6548fae2015-05-07 20:32:22 -0400902 if (!(nd->flags & LOOKUP_RCU))
903 path_put(&last->link);
Al Viro574197e2011-03-14 22:20:34 -0400904}
905
Linus Torvalds561ec642012-10-26 10:05:07 -0700906int sysctl_protected_symlinks __read_mostly = 0;
907int sysctl_protected_hardlinks __read_mostly = 0;
Kees Cook800179c2012-07-25 17:29:07 -0700908
909/**
910 * may_follow_link - Check symlink following for unsafe situations
Randy Dunlap55852632012-08-18 17:39:25 -0700911 * @nd: nameidata pathwalk data
Kees Cook800179c2012-07-25 17:29:07 -0700912 *
913 * In the case of the sysctl_protected_symlinks sysctl being enabled,
914 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
915 * in a sticky world-writable directory. This is to protect privileged
916 * processes from failing races against path names that may change out
917 * from under them by way of other users creating malicious symlinks.
918 * It will permit symlinks to be followed only when outside a sticky
919 * world-writable directory, or when the uid of the symlink and follower
920 * match, or when the directory owner matches the symlink's owner.
921 *
922 * Returns 0 if following the symlink is allowed, -ve on error.
923 */
Al Virofec2fa22015-05-06 15:58:18 -0400924static inline int may_follow_link(struct nameidata *nd)
Kees Cook800179c2012-07-25 17:29:07 -0700925{
926 const struct inode *inode;
927 const struct inode *parent;
Seth Forshee2d7f9e22016-04-26 14:36:23 -0500928 kuid_t puid;
Kees Cook800179c2012-07-25 17:29:07 -0700929
930 if (!sysctl_protected_symlinks)
931 return 0;
932
933 /* Allowed if owner and follower match. */
Al Virofceef392015-12-29 15:58:39 -0500934 inode = nd->link_inode;
Eric W. Biederman81abe272012-08-03 09:38:08 -0700935 if (uid_eq(current_cred()->fsuid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700936 return 0;
937
938 /* Allowed if parent directory not sticky and world-writable. */
Al Viroaa65fa32015-08-04 23:23:50 -0400939 parent = nd->inode;
Kees Cook800179c2012-07-25 17:29:07 -0700940 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
941 return 0;
942
943 /* Allowed if parent directory and link owner match. */
Seth Forshee2d7f9e22016-04-26 14:36:23 -0500944 puid = parent->i_uid;
945 if (uid_valid(puid) && uid_eq(puid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700946 return 0;
947
Al Viro319565022015-05-07 20:37:40 -0400948 if (nd->flags & LOOKUP_RCU)
949 return -ECHILD;
950
Al Viro1cf26652015-05-06 16:01:56 -0400951 audit_log_link_denied("follow_link", &nd->stack[0].link);
Kees Cook800179c2012-07-25 17:29:07 -0700952 return -EACCES;
953}
954
955/**
956 * safe_hardlink_source - Check for safe hardlink conditions
957 * @inode: the source inode to hardlink from
958 *
959 * Return false if at least one of the following conditions:
960 * - inode is not a regular file
961 * - inode is setuid
962 * - inode is setgid and group-exec
963 * - access failure for read and write
964 *
965 * Otherwise returns true.
966 */
967static bool safe_hardlink_source(struct inode *inode)
968{
969 umode_t mode = inode->i_mode;
970
971 /* Special files should not get pinned to the filesystem. */
972 if (!S_ISREG(mode))
973 return false;
974
975 /* Setuid files should not get pinned to the filesystem. */
976 if (mode & S_ISUID)
977 return false;
978
979 /* Executable setgid files should not get pinned to the filesystem. */
980 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
981 return false;
982
983 /* Hardlinking to unreadable or unwritable sources is dangerous. */
984 if (inode_permission(inode, MAY_READ | MAY_WRITE))
985 return false;
986
987 return true;
988}
989
990/**
991 * may_linkat - Check permissions for creating a hardlink
992 * @link: the source to hardlink from
993 *
994 * Block hardlink when all of:
995 * - sysctl_protected_hardlinks enabled
996 * - fsuid does not match inode
997 * - hardlink source is unsafe (see safe_hardlink_source() above)
Dirk Steinmetzf2ca3792015-10-20 16:09:19 +0200998 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
Kees Cook800179c2012-07-25 17:29:07 -0700999 *
1000 * Returns 0 if successful, -ve on error.
1001 */
1002static int may_linkat(struct path *link)
1003{
Kees Cook800179c2012-07-25 17:29:07 -07001004 struct inode *inode;
1005
1006 if (!sysctl_protected_hardlinks)
1007 return 0;
1008
Kees Cook800179c2012-07-25 17:29:07 -07001009 inode = link->dentry->d_inode;
1010
1011 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1012 * otherwise, it must be a safe source.
1013 */
Dirk Steinmetzf2ca3792015-10-20 16:09:19 +02001014 if (inode_owner_or_capable(inode) || safe_hardlink_source(inode))
Kees Cook800179c2012-07-25 17:29:07 -07001015 return 0;
1016
Kees Cooka51d9ea2012-07-25 17:29:08 -07001017 audit_log_link_denied("linkat", link);
Kees Cook800179c2012-07-25 17:29:07 -07001018 return -EPERM;
1019}
1020
Al Viro3b2e7f72015-04-19 00:53:50 -04001021static __always_inline
1022const char *get_link(struct nameidata *nd)
Ian Kent051d3812006-03-27 01:14:53 -08001023{
Al Viroab104922015-05-10 11:50:01 -04001024 struct saved *last = nd->stack + nd->depth - 1;
Al Viro1cf26652015-05-06 16:01:56 -04001025 struct dentry *dentry = last->link.dentry;
Al Virofceef392015-12-29 15:58:39 -05001026 struct inode *inode = nd->link_inode;
Al Viro6d7b5aa2012-06-10 04:15:17 -04001027 int error;
Al Viro0a959df2015-04-18 18:23:41 -04001028 const char *res;
Ian Kent051d3812006-03-27 01:14:53 -08001029
NeilBrown8fa9dd22015-03-23 13:37:40 +11001030 if (!(nd->flags & LOOKUP_RCU)) {
1031 touch_atime(&last->link);
1032 cond_resched();
Miklos Szeredi598e3c82016-09-16 12:44:20 +02001033 } else if (atime_needs_update_rcu(&last->link, inode)) {
NeilBrown8fa9dd22015-03-23 13:37:40 +11001034 if (unlikely(unlazy_walk(nd, NULL, 0)))
1035 return ERR_PTR(-ECHILD);
1036 touch_atime(&last->link);
1037 }
1038
NeilBrownbda0be72015-03-23 13:37:39 +11001039 error = security_inode_follow_link(dentry, inode,
1040 nd->flags & LOOKUP_RCU);
1041 if (unlikely(error))
Al Viro6920a442015-05-10 10:43:46 -04001042 return ERR_PTR(error);
Al Viro36f3b4f2011-02-22 21:24:38 -05001043
Al Viro86acdca12009-12-22 23:45:11 -05001044 nd->last_type = LAST_BIND;
Al Virod4dee482015-04-30 20:08:02 -04001045 res = inode->i_link;
1046 if (!res) {
Al Virofceef392015-12-29 15:58:39 -05001047 const char * (*get)(struct dentry *, struct inode *,
1048 struct delayed_call *);
1049 get = inode->i_op->get_link;
Al Viro8c1b4562015-05-09 18:15:21 -04001050 if (nd->flags & LOOKUP_RCU) {
Al Virofceef392015-12-29 15:58:39 -05001051 res = get(NULL, inode, &last->done);
Al Viro6b255392015-11-17 10:20:54 -05001052 if (res == ERR_PTR(-ECHILD)) {
1053 if (unlikely(unlazy_walk(nd, NULL, 0)))
1054 return ERR_PTR(-ECHILD);
Al Virofceef392015-12-29 15:58:39 -05001055 res = get(dentry, inode, &last->done);
Al Viro6b255392015-11-17 10:20:54 -05001056 }
1057 } else {
Al Virofceef392015-12-29 15:58:39 -05001058 res = get(dentry, inode, &last->done);
Al Viro8c1b4562015-05-09 18:15:21 -04001059 }
Al Virofceef392015-12-29 15:58:39 -05001060 if (IS_ERR_OR_NULL(res))
Al Virofab51e82015-05-10 11:01:00 -04001061 return res;
Ian Kent051d3812006-03-27 01:14:53 -08001062 }
Al Virofab51e82015-05-10 11:01:00 -04001063 if (*res == '/') {
Al Viro9e6697e2015-12-05 20:07:21 -05001064 if (!nd->root.mnt)
1065 set_root(nd);
Al Viro248fb5b2015-12-05 20:51:58 -05001066 if (unlikely(nd_jump_root(nd)))
1067 return ERR_PTR(-ECHILD);
Al Virofab51e82015-05-10 11:01:00 -04001068 while (unlikely(*++res == '/'))
1069 ;
1070 }
1071 if (!*res)
1072 res = NULL;
Al Viro0a959df2015-04-18 18:23:41 -04001073 return res;
1074}
Al Viro6d7b5aa2012-06-10 04:15:17 -04001075
David Howellsf015f1262012-06-25 12:55:28 +01001076/*
1077 * follow_up - Find the mountpoint of path's vfsmount
1078 *
1079 * Given a path, find the mountpoint of its source file system.
1080 * Replace @path with the path of the mountpoint in the parent mount.
1081 * Up is towards /.
1082 *
1083 * Return 1 if we went up a level and 0 if we were already at the
1084 * root.
1085 */
Al Virobab77eb2009-04-18 03:26:48 -04001086int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087{
Al Viro0714a532011-11-24 22:19:58 -05001088 struct mount *mnt = real_mount(path->mnt);
1089 struct mount *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +10001091
Al Viro48a066e2013-09-29 22:06:07 -04001092 read_seqlock_excl(&mount_lock);
Al Viro0714a532011-11-24 22:19:58 -05001093 parent = mnt->mnt_parent;
Al Viro3c0a6162012-07-18 17:32:50 +04001094 if (parent == mnt) {
Al Viro48a066e2013-09-29 22:06:07 -04001095 read_sequnlock_excl(&mount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 return 0;
1097 }
Al Viro0714a532011-11-24 22:19:58 -05001098 mntget(&parent->mnt);
Al Viroa73324d2011-11-24 22:25:07 -05001099 mountpoint = dget(mnt->mnt_mountpoint);
Al Viro48a066e2013-09-29 22:06:07 -04001100 read_sequnlock_excl(&mount_lock);
Al Virobab77eb2009-04-18 03:26:48 -04001101 dput(path->dentry);
1102 path->dentry = mountpoint;
1103 mntput(path->mnt);
Al Viro0714a532011-11-24 22:19:58 -05001104 path->mnt = &parent->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 return 1;
1106}
Al Viro4d359502014-03-14 12:20:17 -04001107EXPORT_SYMBOL(follow_up);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001109/*
David Howells9875cf82011-01-14 18:45:21 +00001110 * Perform an automount
1111 * - return -EISDIR to tell follow_managed() to stop and return the path we
1112 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 */
NeilBrown756daf22015-03-23 13:37:38 +11001114static int follow_automount(struct path *path, struct nameidata *nd,
David Howells9875cf82011-01-14 18:45:21 +00001115 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +11001116{
David Howells9875cf82011-01-14 18:45:21 +00001117 struct vfsmount *mnt;
David Howellsea5b7782011-01-14 19:10:03 +00001118 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001119
David Howells9875cf82011-01-14 18:45:21 +00001120 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1121 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -07001122
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +02001123 /* We don't want to mount if someone's just doing a stat -
1124 * unless they're stat'ing a directory and appended a '/' to
1125 * the name.
1126 *
1127 * We do, however, want to mount if someone wants to open or
1128 * create a file of any type under the mountpoint, wants to
1129 * traverse through the mountpoint or wants to open the
1130 * mounted directory. Also, autofs may mark negative dentries
1131 * as being automount points. These will need the attentions
1132 * of the daemon to instantiate them before they can be used.
David Howells9875cf82011-01-14 18:45:21 +00001133 */
NeilBrown756daf22015-03-23 13:37:38 +11001134 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1135 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +02001136 path->dentry->d_inode)
1137 return -EISDIR;
1138
Eric W. Biedermanaeaa4a72016-07-23 11:20:44 -05001139 if (path->dentry->d_sb->s_user_ns != &init_user_ns)
1140 return -EACCES;
1141
NeilBrown756daf22015-03-23 13:37:38 +11001142 nd->total_link_count++;
1143 if (nd->total_link_count >= 40)
David Howells9875cf82011-01-14 18:45:21 +00001144 return -ELOOP;
1145
1146 mnt = path->dentry->d_op->d_automount(path);
1147 if (IS_ERR(mnt)) {
1148 /*
1149 * The filesystem is allowed to return -EISDIR here to indicate
1150 * it doesn't want to automount. For instance, autofs would do
1151 * this so that its userspace daemon can mount on this dentry.
1152 *
1153 * However, we can only permit this if it's a terminal point in
1154 * the path being looked up; if it wasn't then the remainder of
1155 * the path is inaccessible and we should say so.
1156 */
NeilBrown756daf22015-03-23 13:37:38 +11001157 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
David Howells9875cf82011-01-14 18:45:21 +00001158 return -EREMOTE;
1159 return PTR_ERR(mnt);
1160 }
David Howellsea5b7782011-01-14 19:10:03 +00001161
David Howells9875cf82011-01-14 18:45:21 +00001162 if (!mnt) /* mount collision */
1163 return 0;
1164
Al Viro8aef1882011-06-16 15:10:06 +01001165 if (!*need_mntput) {
1166 /* lock_mount() may release path->mnt on error */
1167 mntget(path->mnt);
1168 *need_mntput = true;
1169 }
Al Viro19a167a2011-01-17 01:35:23 -05001170 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +00001171
David Howellsea5b7782011-01-14 19:10:03 +00001172 switch (err) {
1173 case -EBUSY:
1174 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -05001175 return 0;
David Howellsea5b7782011-01-14 19:10:03 +00001176 case 0:
Al Viro8aef1882011-06-16 15:10:06 +01001177 path_put(path);
David Howellsea5b7782011-01-14 19:10:03 +00001178 path->mnt = mnt;
1179 path->dentry = dget(mnt->mnt_root);
David Howellsea5b7782011-01-14 19:10:03 +00001180 return 0;
Al Viro19a167a2011-01-17 01:35:23 -05001181 default:
1182 return err;
David Howellsea5b7782011-01-14 19:10:03 +00001183 }
Al Viro19a167a2011-01-17 01:35:23 -05001184
David Howells9875cf82011-01-14 18:45:21 +00001185}
1186
1187/*
1188 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +00001189 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +00001190 * - Flagged as mountpoint
1191 * - Flagged as automount point
1192 *
1193 * This may only be called in refwalk mode.
1194 *
1195 * Serialization is taken care of in namespace.c
1196 */
NeilBrown756daf22015-03-23 13:37:38 +11001197static int follow_managed(struct path *path, struct nameidata *nd)
David Howells9875cf82011-01-14 18:45:21 +00001198{
Al Viro8aef1882011-06-16 15:10:06 +01001199 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
David Howells9875cf82011-01-14 18:45:21 +00001200 unsigned managed;
1201 bool need_mntput = false;
Al Viro8aef1882011-06-16 15:10:06 +01001202 int ret = 0;
David Howells9875cf82011-01-14 18:45:21 +00001203
1204 /* Given that we're not holding a lock here, we retain the value in a
1205 * local variable for each dentry as we look at it so that we don't see
1206 * the components of that value change under us */
1207 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1208 managed &= DCACHE_MANAGED_DENTRY,
1209 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +00001210 /* Allow the filesystem to manage the transit without i_mutex
1211 * being held. */
1212 if (managed & DCACHE_MANAGE_TRANSIT) {
1213 BUG_ON(!path->dentry->d_op);
1214 BUG_ON(!path->dentry->d_op->d_manage);
Al Viro1aed3e42011-03-18 09:09:02 -04001215 ret = path->dentry->d_op->d_manage(path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001216 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001217 break;
David Howellscc53ce52011-01-14 18:45:26 +00001218 }
1219
David Howells9875cf82011-01-14 18:45:21 +00001220 /* Transit to a mounted filesystem. */
1221 if (managed & DCACHE_MOUNTED) {
1222 struct vfsmount *mounted = lookup_mnt(path);
1223 if (mounted) {
1224 dput(path->dentry);
1225 if (need_mntput)
1226 mntput(path->mnt);
1227 path->mnt = mounted;
1228 path->dentry = dget(mounted->mnt_root);
1229 need_mntput = true;
1230 continue;
1231 }
1232
1233 /* Something is mounted on this dentry in another
1234 * namespace and/or whatever was mounted there in this
Al Viro48a066e2013-09-29 22:06:07 -04001235 * namespace got unmounted before lookup_mnt() could
1236 * get it */
David Howells9875cf82011-01-14 18:45:21 +00001237 }
1238
1239 /* Handle an automount point */
1240 if (managed & DCACHE_NEED_AUTOMOUNT) {
NeilBrown756daf22015-03-23 13:37:38 +11001241 ret = follow_automount(path, nd, &need_mntput);
David Howells9875cf82011-01-14 18:45:21 +00001242 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001243 break;
David Howells9875cf82011-01-14 18:45:21 +00001244 continue;
1245 }
1246
1247 /* We didn't change the current path point */
1248 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 }
Al Viro8aef1882011-06-16 15:10:06 +01001250
1251 if (need_mntput && path->mnt == mnt)
1252 mntput(path->mnt);
Al Viroe9742b52016-03-05 22:04:59 -05001253 if (ret == -EISDIR || !ret)
1254 ret = 1;
Al Viro84027522015-04-22 10:30:08 -04001255 if (need_mntput)
1256 nd->flags |= LOOKUP_JUMPED;
1257 if (unlikely(ret < 0))
1258 path_put_conditional(path, nd);
1259 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260}
1261
David Howellscc53ce52011-01-14 18:45:26 +00001262int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263{
1264 struct vfsmount *mounted;
1265
Al Viro1c755af2009-04-18 14:06:57 -04001266 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -04001268 dput(path->dentry);
1269 mntput(path->mnt);
1270 path->mnt = mounted;
1271 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 return 1;
1273 }
1274 return 0;
1275}
Al Viro4d359502014-03-14 12:20:17 -04001276EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277
NeilBrownb8faf032014-08-04 17:06:29 +10001278static inline int managed_dentry_rcu(struct dentry *dentry)
Ian Kent62a73752011-03-25 01:51:02 +08001279{
NeilBrownb8faf032014-08-04 17:06:29 +10001280 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1281 dentry->d_op->d_manage(dentry, true) : 0;
Ian Kent62a73752011-03-25 01:51:02 +08001282}
1283
David Howells9875cf82011-01-14 18:45:21 +00001284/*
Al Viro287548e2011-05-27 06:50:06 -04001285 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1286 * we meet a managed dentry that would need blocking.
David Howells9875cf82011-01-14 18:45:21 +00001287 */
1288static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
Al Viro254cf582015-05-05 09:40:46 -04001289 struct inode **inode, unsigned *seqp)
David Howells9875cf82011-01-14 18:45:21 +00001290{
Ian Kent62a73752011-03-25 01:51:02 +08001291 for (;;) {
Al Viroc7105362011-11-24 18:22:03 -05001292 struct mount *mounted;
Ian Kent62a73752011-03-25 01:51:02 +08001293 /*
1294 * Don't forget we might have a non-mountpoint managed dentry
1295 * that wants to block transit.
1296 */
NeilBrownb8faf032014-08-04 17:06:29 +10001297 switch (managed_dentry_rcu(path->dentry)) {
1298 case -ECHILD:
1299 default:
David Howellsab909112011-01-14 18:46:51 +00001300 return false;
NeilBrownb8faf032014-08-04 17:06:29 +10001301 case -EISDIR:
1302 return true;
1303 case 0:
1304 break;
1305 }
Ian Kent62a73752011-03-25 01:51:02 +08001306
1307 if (!d_mountpoint(path->dentry))
NeilBrownb8faf032014-08-04 17:06:29 +10001308 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
Ian Kent62a73752011-03-25 01:51:02 +08001309
Al Viro474279d2013-10-01 16:11:26 -04001310 mounted = __lookup_mnt(path->mnt, path->dentry);
David Howells9875cf82011-01-14 18:45:21 +00001311 if (!mounted)
1312 break;
Al Viroc7105362011-11-24 18:22:03 -05001313 path->mnt = &mounted->mnt;
1314 path->dentry = mounted->mnt.mnt_root;
Al Viroa3fbbde2011-11-07 21:21:26 +00001315 nd->flags |= LOOKUP_JUMPED;
Al Viro254cf582015-05-05 09:40:46 -04001316 *seqp = read_seqcount_begin(&path->dentry->d_seq);
Linus Torvalds59430262011-07-18 15:43:29 -07001317 /*
1318 * Update the inode too. We don't need to re-check the
1319 * dentry sequence number here after this d_inode read,
1320 * because a mount-point is always pinned.
1321 */
1322 *inode = path->dentry->d_inode;
David Howells9875cf82011-01-14 18:45:21 +00001323 }
Al Virof5be3e29122014-09-13 21:50:45 -04001324 return !read_seqretry(&mount_lock, nd->m_seq) &&
NeilBrownb8faf032014-08-04 17:06:29 +10001325 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
Al Viro287548e2011-05-27 06:50:06 -04001326}
1327
Nick Piggin31e6b012011-01-07 17:49:52 +11001328static int follow_dotdot_rcu(struct nameidata *nd)
1329{
Al Viro4023bfc2014-09-13 21:59:43 -04001330 struct inode *inode = nd->inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001331
David Howells9875cf82011-01-14 18:45:21 +00001332 while (1) {
Al Viroaed434a2015-05-12 12:22:47 -04001333 if (path_equal(&nd->path, &nd->root))
Nick Piggin31e6b012011-01-07 17:49:52 +11001334 break;
Nick Piggin31e6b012011-01-07 17:49:52 +11001335 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1336 struct dentry *old = nd->path.dentry;
1337 struct dentry *parent = old->d_parent;
1338 unsigned seq;
1339
Al Viro4023bfc2014-09-13 21:59:43 -04001340 inode = parent->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001341 seq = read_seqcount_begin(&parent->d_seq);
Al Viroaed434a2015-05-12 12:22:47 -04001342 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1343 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001344 nd->path.dentry = parent;
1345 nd->seq = seq;
Eric W. Biederman397d4252015-08-15 20:27:13 -05001346 if (unlikely(!path_connected(&nd->path)))
1347 return -ENOENT;
Nick Piggin31e6b012011-01-07 17:49:52 +11001348 break;
Al Viroaed434a2015-05-12 12:22:47 -04001349 } else {
1350 struct mount *mnt = real_mount(nd->path.mnt);
1351 struct mount *mparent = mnt->mnt_parent;
1352 struct dentry *mountpoint = mnt->mnt_mountpoint;
1353 struct inode *inode2 = mountpoint->d_inode;
1354 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1355 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1356 return -ECHILD;
1357 if (&mparent->mnt == nd->path.mnt)
1358 break;
1359 /* we know that mountpoint was pinned */
1360 nd->path.dentry = mountpoint;
1361 nd->path.mnt = &mparent->mnt;
1362 inode = inode2;
1363 nd->seq = seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11001364 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001365 }
Al Viroaed434a2015-05-12 12:22:47 -04001366 while (unlikely(d_mountpoint(nd->path.dentry))) {
Al Virob37199e2014-03-20 15:18:22 -04001367 struct mount *mounted;
1368 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
Al Viroaed434a2015-05-12 12:22:47 -04001369 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1370 return -ECHILD;
Al Virob37199e2014-03-20 15:18:22 -04001371 if (!mounted)
1372 break;
1373 nd->path.mnt = &mounted->mnt;
1374 nd->path.dentry = mounted->mnt.mnt_root;
Al Viro4023bfc2014-09-13 21:59:43 -04001375 inode = nd->path.dentry->d_inode;
Al Virob37199e2014-03-20 15:18:22 -04001376 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Virob37199e2014-03-20 15:18:22 -04001377 }
Al Viro4023bfc2014-09-13 21:59:43 -04001378 nd->inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001379 return 0;
1380}
1381
David Howells9875cf82011-01-14 18:45:21 +00001382/*
David Howellscc53ce52011-01-14 18:45:26 +00001383 * Follow down to the covering mount currently visible to userspace. At each
1384 * point, the filesystem owning that dentry may be queried as to whether the
1385 * caller is permitted to proceed or not.
David Howellscc53ce52011-01-14 18:45:26 +00001386 */
Al Viro7cc90cc2011-03-18 09:04:20 -04001387int follow_down(struct path *path)
David Howellscc53ce52011-01-14 18:45:26 +00001388{
1389 unsigned managed;
1390 int ret;
1391
1392 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1393 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1394 /* Allow the filesystem to manage the transit without i_mutex
1395 * being held.
1396 *
1397 * We indicate to the filesystem if someone is trying to mount
1398 * something here. This gives autofs the chance to deny anyone
1399 * other than its daemon the right to mount on its
1400 * superstructure.
1401 *
1402 * The filesystem may sleep at this point.
1403 */
1404 if (managed & DCACHE_MANAGE_TRANSIT) {
1405 BUG_ON(!path->dentry->d_op);
1406 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +00001407 ret = path->dentry->d_op->d_manage(
Al Viro1aed3e42011-03-18 09:09:02 -04001408 path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001409 if (ret < 0)
1410 return ret == -EISDIR ? 0 : ret;
1411 }
1412
1413 /* Transit to a mounted filesystem. */
1414 if (managed & DCACHE_MOUNTED) {
1415 struct vfsmount *mounted = lookup_mnt(path);
1416 if (!mounted)
1417 break;
1418 dput(path->dentry);
1419 mntput(path->mnt);
1420 path->mnt = mounted;
1421 path->dentry = dget(mounted->mnt_root);
1422 continue;
1423 }
1424
1425 /* Don't handle automount points here */
1426 break;
1427 }
1428 return 0;
1429}
Al Viro4d359502014-03-14 12:20:17 -04001430EXPORT_SYMBOL(follow_down);
David Howellscc53ce52011-01-14 18:45:26 +00001431
1432/*
David Howells9875cf82011-01-14 18:45:21 +00001433 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1434 */
1435static void follow_mount(struct path *path)
1436{
1437 while (d_mountpoint(path->dentry)) {
1438 struct vfsmount *mounted = lookup_mnt(path);
1439 if (!mounted)
1440 break;
1441 dput(path->dentry);
1442 mntput(path->mnt);
1443 path->mnt = mounted;
1444 path->dentry = dget(mounted->mnt_root);
1445 }
1446}
1447
Eric W. Biedermaneedf2652016-06-02 10:29:47 -05001448static int path_parent_directory(struct path *path)
1449{
1450 struct dentry *old = path->dentry;
1451 /* rare case of legitimate dget_parent()... */
1452 path->dentry = dget_parent(path->dentry);
1453 dput(old);
1454 if (unlikely(!path_connected(path)))
1455 return -ENOENT;
1456 return 0;
1457}
1458
Eric W. Biederman397d4252015-08-15 20:27:13 -05001459static int follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460{
1461 while(1) {
Al Viro2a737872009-04-07 11:49:53 -04001462 if (nd->path.dentry == nd->root.dentry &&
1463 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 break;
1465 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001466 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Eric W. Biedermaneedf2652016-06-02 10:29:47 -05001467 int ret = path_parent_directory(&nd->path);
1468 if (ret)
1469 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 break;
1471 }
Al Viro3088dd72010-01-30 15:47:29 -05001472 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 }
Al Viro79ed0222009-04-18 13:59:41 -04001475 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001476 nd->inode = nd->path.dentry->d_inode;
Eric W. Biederman397d4252015-08-15 20:27:13 -05001477 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478}
1479
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480/*
Oleg Drokinf4fdace2016-07-07 22:04:04 -04001481 * This looks up the name in dcache and possibly revalidates the found dentry.
1482 * NULL is returned if the dentry does not exist in the cache.
Nick Pigginbaa03892010-08-18 04:37:31 +10001483 */
Al Viroe3c13922016-03-06 14:03:27 -05001484static struct dentry *lookup_dcache(const struct qstr *name,
1485 struct dentry *dir,
Al Viro6c51e512016-03-05 20:09:32 -05001486 unsigned int flags)
Nick Pigginbaa03892010-08-18 04:37:31 +10001487{
Nick Pigginbaa03892010-08-18 04:37:31 +10001488 struct dentry *dentry;
Miklos Szeredibad61182012-03-26 12:54:24 +02001489 int error;
Nick Pigginbaa03892010-08-18 04:37:31 +10001490
Miklos Szeredibad61182012-03-26 12:54:24 +02001491 dentry = d_lookup(dir, name);
1492 if (dentry) {
Jeff Layton39e3c952012-11-28 11:30:53 -05001493 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
Al Viro201f9562012-06-22 12:42:10 +04001494 error = d_revalidate(dentry, flags);
Miklos Szeredibad61182012-03-26 12:54:24 +02001495 if (unlikely(error <= 0)) {
Al Viro74ff0ff2016-03-05 22:37:46 -05001496 if (!error)
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001497 d_invalidate(dentry);
Al Viro74ff0ff2016-03-05 22:37:46 -05001498 dput(dentry);
1499 return ERR_PTR(error);
Miklos Szeredibad61182012-03-26 12:54:24 +02001500 }
1501 }
1502 }
Nick Pigginbaa03892010-08-18 04:37:31 +10001503 return dentry;
1504}
1505
1506/*
J. Bruce Fields13a2c3b2013-10-23 16:09:16 -04001507 * Call i_op->lookup on the dentry. The dentry must be negative and
1508 * unhashed.
Miklos Szeredibad61182012-03-26 12:54:24 +02001509 *
1510 * dir->d_inode->i_mutex must be held
Josef Bacik44396f42011-05-31 11:58:49 -04001511 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001512static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
Al Viro72bd8662012-06-10 17:17:17 -04001513 unsigned int flags)
Josef Bacik44396f42011-05-31 11:58:49 -04001514{
Josef Bacik44396f42011-05-31 11:58:49 -04001515 struct dentry *old;
1516
1517 /* Don't create child dentry for a dead directory. */
Miklos Szeredibad61182012-03-26 12:54:24 +02001518 if (unlikely(IS_DEADDIR(dir))) {
Miklos Szeredie188dc02012-02-03 14:25:18 +01001519 dput(dentry);
Josef Bacik44396f42011-05-31 11:58:49 -04001520 return ERR_PTR(-ENOENT);
Miklos Szeredie188dc02012-02-03 14:25:18 +01001521 }
Josef Bacik44396f42011-05-31 11:58:49 -04001522
Al Viro72bd8662012-06-10 17:17:17 -04001523 old = dir->i_op->lookup(dir, dentry, flags);
Josef Bacik44396f42011-05-31 11:58:49 -04001524 if (unlikely(old)) {
1525 dput(dentry);
1526 dentry = old;
1527 }
1528 return dentry;
1529}
1530
Al Viroe3c13922016-03-06 14:03:27 -05001531static struct dentry *__lookup_hash(const struct qstr *name,
Al Viro72bd8662012-06-10 17:17:17 -04001532 struct dentry *base, unsigned int flags)
Al Viroa3255542012-03-30 14:41:51 -04001533{
Al Viro6c51e512016-03-05 20:09:32 -05001534 struct dentry *dentry = lookup_dcache(name, base, flags);
Al Viroa3255542012-03-30 14:41:51 -04001535
Al Viro6c51e512016-03-05 20:09:32 -05001536 if (dentry)
Miklos Szeredibad61182012-03-26 12:54:24 +02001537 return dentry;
Al Viroa3255542012-03-30 14:41:51 -04001538
Al Viro6c51e512016-03-05 20:09:32 -05001539 dentry = d_alloc(base, name);
1540 if (unlikely(!dentry))
1541 return ERR_PTR(-ENOMEM);
1542
Al Viro72bd8662012-06-10 17:17:17 -04001543 return lookup_real(base->d_inode, dentry, flags);
Al Viroa3255542012-03-30 14:41:51 -04001544}
1545
Al Viroe97cdc82013-01-24 18:16:00 -05001546static int lookup_fast(struct nameidata *nd,
Al Viro254cf582015-05-05 09:40:46 -04001547 struct path *path, struct inode **inode,
1548 unsigned *seqp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549{
Jan Blunck4ac91372008-02-14 19:34:32 -08001550 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001551 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001552 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001553 int err;
1554
Al Viro3cac2602009-08-13 18:27:43 +04001555 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001556 * Rename seqlock is not required here because in the off chance
Al Viro5d0f49c2016-03-05 21:32:53 -05001557 * of a false negative due to a concurrent rename, the caller is
1558 * going to fall back to non-racy lookup.
Nick Pigginb04f7842010-08-18 04:37:34 +10001559 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001560 if (nd->flags & LOOKUP_RCU) {
1561 unsigned seq;
Al Viro766c4cb2015-05-07 19:24:57 -04001562 bool negative;
Linus Torvaldsda53be12013-05-21 15:22:44 -07001563 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
Al Viro5d0f49c2016-03-05 21:32:53 -05001564 if (unlikely(!dentry)) {
1565 if (unlazy_walk(nd, NULL, 0))
1566 return -ECHILD;
Al Viroe9742b52016-03-05 22:04:59 -05001567 return 0;
Al Viro5d0f49c2016-03-05 21:32:53 -05001568 }
Al Viro5a18fff2011-03-11 04:44:53 -05001569
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001570 /*
1571 * This sequence count validates that the inode matches
1572 * the dentry name information from lookup.
1573 */
David Howells63afdfc2015-05-06 15:59:00 +01001574 *inode = d_backing_inode(dentry);
Al Viro766c4cb2015-05-07 19:24:57 -04001575 negative = d_is_negative(dentry);
Al Viro5d0f49c2016-03-05 21:32:53 -05001576 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001577 return -ECHILD;
1578
1579 /*
1580 * This sequence count validates that the parent had no
1581 * changes while we did the lookup of the dentry above.
1582 *
1583 * The memory barrier in read_seqcount_begin of child is
1584 * enough, we can use __read_seqcount_retry here.
1585 */
Al Viro5d0f49c2016-03-05 21:32:53 -05001586 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
Nick Piggin31e6b012011-01-07 17:49:52 +11001587 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001588
Al Viro254cf582015-05-05 09:40:46 -04001589 *seqp = seq;
Al Viro5d0f49c2016-03-05 21:32:53 -05001590 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
Al Viro4ce16ef32012-06-10 16:10:59 -04001591 status = d_revalidate(dentry, nd->flags);
Al Viro5d0f49c2016-03-05 21:32:53 -05001592 if (unlikely(status <= 0)) {
1593 if (unlazy_walk(nd, dentry, seq))
1594 return -ECHILD;
1595 if (status == -ECHILD)
1596 status = d_revalidate(dentry, nd->flags);
1597 } else {
1598 /*
1599 * Note: do negative dentry check after revalidation in
1600 * case that drops it.
1601 */
1602 if (unlikely(negative))
1603 return -ENOENT;
1604 path->mnt = mnt;
1605 path->dentry = dentry;
1606 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
Al Viroe9742b52016-03-05 22:04:59 -05001607 return 1;
Al Viro5d0f49c2016-03-05 21:32:53 -05001608 if (unlazy_walk(nd, dentry, seq))
1609 return -ECHILD;
Al Viro24643082011-02-15 01:26:22 -05001610 }
Al Viro5a18fff2011-03-11 04:44:53 -05001611 } else {
Al Viroe97cdc82013-01-24 18:16:00 -05001612 dentry = __d_lookup(parent, &nd->last);
Al Viro5d0f49c2016-03-05 21:32:53 -05001613 if (unlikely(!dentry))
Al Viroe9742b52016-03-05 22:04:59 -05001614 return 0;
Al Viro5d0f49c2016-03-05 21:32:53 -05001615 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
1616 status = d_revalidate(dentry, nd->flags);
Nick Piggin31e6b012011-01-07 17:49:52 +11001617 }
Al Viro5a18fff2011-03-11 04:44:53 -05001618 if (unlikely(status <= 0)) {
Al Viroe9742b52016-03-05 22:04:59 -05001619 if (!status)
Al Viro5d0f49c2016-03-05 21:32:53 -05001620 d_invalidate(dentry);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001621 dput(dentry);
Al Viro5d0f49c2016-03-05 21:32:53 -05001622 return status;
Al Viro5a18fff2011-03-11 04:44:53 -05001623 }
Al Viro766c4cb2015-05-07 19:24:57 -04001624 if (unlikely(d_is_negative(dentry))) {
1625 dput(dentry);
1626 return -ENOENT;
1627 }
Al Viro5d0f49c2016-03-05 21:32:53 -05001628
David Howells9875cf82011-01-14 18:45:21 +00001629 path->mnt = mnt;
1630 path->dentry = dentry;
NeilBrown756daf22015-03-23 13:37:38 +11001631 err = follow_managed(path, nd);
Al Viroe9742b52016-03-05 22:04:59 -05001632 if (likely(err > 0))
David Howells63afdfc2015-05-06 15:59:00 +01001633 *inode = d_backing_inode(path->dentry);
Al Viro84027522015-04-22 10:30:08 -04001634 return err;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001635}
1636
1637/* Fast lookup failed, do it the slow way */
Al Viroe3c13922016-03-06 14:03:27 -05001638static struct dentry *lookup_slow(const struct qstr *name,
1639 struct dentry *dir,
1640 unsigned int flags)
Miklos Szeredi697f5142012-05-21 17:30:05 +02001641{
Al Viro94bdd652016-04-15 02:42:04 -04001642 struct dentry *dentry = ERR_PTR(-ENOENT), *old;
Al Viro19363862016-04-14 19:33:34 -04001643 struct inode *inode = dir->d_inode;
Al Virod9171b92016-04-15 03:33:13 -04001644 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
Al Viro19363862016-04-14 19:33:34 -04001645
Al Viro9902af72016-04-15 15:08:36 -04001646 inode_lock_shared(inode);
Al Viro19363862016-04-14 19:33:34 -04001647 /* Don't go there if it's already dead */
Al Viro94bdd652016-04-15 02:42:04 -04001648 if (unlikely(IS_DEADDIR(inode)))
1649 goto out;
1650again:
Al Virod9171b92016-04-15 03:33:13 -04001651 dentry = d_alloc_parallel(dir, name, &wq);
Al Viro94bdd652016-04-15 02:42:04 -04001652 if (IS_ERR(dentry))
1653 goto out;
1654 if (unlikely(!d_in_lookup(dentry))) {
Al Viro949a8522016-03-06 14:20:52 -05001655 if ((dentry->d_flags & DCACHE_OP_REVALIDATE) &&
1656 !(flags & LOOKUP_NO_REVAL)) {
1657 int error = d_revalidate(dentry, flags);
1658 if (unlikely(error <= 0)) {
Al Viro94bdd652016-04-15 02:42:04 -04001659 if (!error) {
Al Viro949a8522016-03-06 14:20:52 -05001660 d_invalidate(dentry);
Al Viro94bdd652016-04-15 02:42:04 -04001661 dput(dentry);
1662 goto again;
1663 }
Al Viro949a8522016-03-06 14:20:52 -05001664 dput(dentry);
1665 dentry = ERR_PTR(error);
1666 }
1667 }
Al Viro94bdd652016-04-15 02:42:04 -04001668 } else {
1669 old = inode->i_op->lookup(inode, dentry, flags);
1670 d_lookup_done(dentry);
1671 if (unlikely(old)) {
1672 dput(dentry);
1673 dentry = old;
Al Viro949a8522016-03-06 14:20:52 -05001674 }
1675 }
Al Viro94bdd652016-04-15 02:42:04 -04001676out:
Al Viro9902af72016-04-15 15:08:36 -04001677 inode_unlock_shared(inode);
Al Viroe3c13922016-03-06 14:03:27 -05001678 return dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679}
1680
Al Viro52094c82011-02-21 21:34:47 -05001681static inline int may_lookup(struct nameidata *nd)
1682{
1683 if (nd->flags & LOOKUP_RCU) {
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07001684 int err = inode_permission2(nd->path.mnt, nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
Al Viro52094c82011-02-21 21:34:47 -05001685 if (err != -ECHILD)
1686 return err;
Al Viro6e9918b2015-05-05 09:26:05 -04001687 if (unlazy_walk(nd, NULL, 0))
Al Viro52094c82011-02-21 21:34:47 -05001688 return -ECHILD;
1689 }
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07001690 return inode_permission2(nd->path.mnt, nd->inode, MAY_EXEC);
Al Viro52094c82011-02-21 21:34:47 -05001691}
1692
Al Viro9856fa12011-03-04 14:22:06 -05001693static inline int handle_dots(struct nameidata *nd, int type)
1694{
1695 if (type == LAST_DOTDOT) {
Al Viro9e6697e2015-12-05 20:07:21 -05001696 if (!nd->root.mnt)
1697 set_root(nd);
Al Viro9856fa12011-03-04 14:22:06 -05001698 if (nd->flags & LOOKUP_RCU) {
Al Viro70291ae2015-05-04 07:53:00 -04001699 return follow_dotdot_rcu(nd);
Al Viro9856fa12011-03-04 14:22:06 -05001700 } else
Eric W. Biederman397d4252015-08-15 20:27:13 -05001701 return follow_dotdot(nd);
Al Viro9856fa12011-03-04 14:22:06 -05001702 }
1703 return 0;
1704}
1705
Al Viro181548c2015-05-07 19:54:34 -04001706static int pick_link(struct nameidata *nd, struct path *link,
1707 struct inode *inode, unsigned seq)
Al Virod63ff282015-05-04 18:13:23 -04001708{
Al Viro626de992015-05-04 18:26:59 -04001709 int error;
Al Viro1cf26652015-05-06 16:01:56 -04001710 struct saved *last;
NeilBrown756daf22015-03-23 13:37:38 +11001711 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
Al Viro626de992015-05-04 18:26:59 -04001712 path_to_nameidata(link, nd);
1713 return -ELOOP;
1714 }
Al Virobc40aee2015-05-09 13:04:24 -04001715 if (!(nd->flags & LOOKUP_RCU)) {
Al Viro79733872015-05-09 12:55:43 -04001716 if (link->mnt == nd->path.mnt)
1717 mntget(link->mnt);
Al Virod63ff282015-05-04 18:13:23 -04001718 }
Al Viro626de992015-05-04 18:26:59 -04001719 error = nd_alloc_stack(nd);
1720 if (unlikely(error)) {
Al Virobc40aee2015-05-09 13:04:24 -04001721 if (error == -ECHILD) {
1722 if (unlikely(unlazy_link(nd, link, seq)))
1723 return -ECHILD;
1724 error = nd_alloc_stack(nd);
1725 }
1726 if (error) {
1727 path_put(link);
1728 return error;
1729 }
Al Viro626de992015-05-04 18:26:59 -04001730 }
1731
Al Viroab104922015-05-10 11:50:01 -04001732 last = nd->stack + nd->depth++;
Al Viro1cf26652015-05-06 16:01:56 -04001733 last->link = *link;
Al Virofceef392015-12-29 15:58:39 -05001734 clear_delayed_call(&last->done);
1735 nd->link_inode = inode;
Al Viro0450b2d2015-05-08 13:23:53 -04001736 last->seq = seq;
Al Virod63ff282015-05-04 18:13:23 -04001737 return 1;
1738}
1739
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001740/*
1741 * Do we need to follow links? We _really_ want to be able
1742 * to do this check without having to look at inode->i_op,
1743 * so we keep a cache of "no, this doesn't need follow_link"
1744 * for the common case.
1745 */
Al Viro254cf582015-05-05 09:40:46 -04001746static inline int should_follow_link(struct nameidata *nd, struct path *link,
Al Viro181548c2015-05-07 19:54:34 -04001747 int follow,
1748 struct inode *inode, unsigned seq)
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001749{
Al Virod63ff282015-05-04 18:13:23 -04001750 if (likely(!d_is_symlink(link->dentry)))
1751 return 0;
1752 if (!follow)
1753 return 0;
Al Viroa7f77542016-02-27 19:31:01 -05001754 /* make sure that d_is_symlink above matches inode */
1755 if (nd->flags & LOOKUP_RCU) {
1756 if (read_seqcount_retry(&link->dentry->d_seq, seq))
1757 return -ECHILD;
1758 }
Al Viro181548c2015-05-07 19:54:34 -04001759 return pick_link(nd, link, inode, seq);
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001760}
1761
Al Viro4693a542015-05-04 17:47:11 -04001762enum {WALK_GET = 1, WALK_PUT = 2};
1763
1764static int walk_component(struct nameidata *nd, int flags)
Al Viroce57dfc2011-03-13 19:58:58 -04001765{
Al Virocaa85632015-04-22 17:52:47 -04001766 struct path path;
Al Viroce57dfc2011-03-13 19:58:58 -04001767 struct inode *inode;
Al Viro254cf582015-05-05 09:40:46 -04001768 unsigned seq;
Al Viroce57dfc2011-03-13 19:58:58 -04001769 int err;
1770 /*
1771 * "." and ".." are special - ".." especially so because it has
1772 * to be able to know about the current root directory and
1773 * parent relationships.
1774 */
Al Viro4693a542015-05-04 17:47:11 -04001775 if (unlikely(nd->last_type != LAST_NORM)) {
1776 err = handle_dots(nd, nd->last_type);
1777 if (flags & WALK_PUT)
1778 put_link(nd);
1779 return err;
1780 }
Al Viro254cf582015-05-05 09:40:46 -04001781 err = lookup_fast(nd, &path, &inode, &seq);
Al Viroe9742b52016-03-05 22:04:59 -05001782 if (unlikely(err <= 0)) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001783 if (err < 0)
Al Virof0a9ba72015-05-04 07:59:30 -04001784 return err;
Al Viroe3c13922016-03-06 14:03:27 -05001785 path.dentry = lookup_slow(&nd->last, nd->path.dentry,
1786 nd->flags);
1787 if (IS_ERR(path.dentry))
1788 return PTR_ERR(path.dentry);
Al Viro7500c382016-03-31 00:23:05 -04001789
Al Viroe3c13922016-03-06 14:03:27 -05001790 path.mnt = nd->path.mnt;
1791 err = follow_managed(&path, nd);
1792 if (unlikely(err < 0))
Al Virof0a9ba72015-05-04 07:59:30 -04001793 return err;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001794
Al Viro7500c382016-03-31 00:23:05 -04001795 if (unlikely(d_is_negative(path.dentry))) {
1796 path_to_nameidata(&path, nd);
1797 return -ENOENT;
1798 }
1799
Al Viro254cf582015-05-05 09:40:46 -04001800 seq = 0; /* we are already out of RCU mode */
Al Virod4565642016-02-27 19:23:16 -05001801 inode = d_backing_inode(path.dentry);
Al Viroce57dfc2011-03-13 19:58:58 -04001802 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001803
Al Viro4693a542015-05-04 17:47:11 -04001804 if (flags & WALK_PUT)
1805 put_link(nd);
Al Viro181548c2015-05-07 19:54:34 -04001806 err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq);
Al Virod63ff282015-05-04 18:13:23 -04001807 if (unlikely(err))
1808 return err;
Al Virocaa85632015-04-22 17:52:47 -04001809 path_to_nameidata(&path, nd);
Al Viroce57dfc2011-03-13 19:58:58 -04001810 nd->inode = inode;
Al Viro254cf582015-05-05 09:40:46 -04001811 nd->seq = seq;
Al Viroce57dfc2011-03-13 19:58:58 -04001812 return 0;
1813}
1814
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815/*
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001816 * We can do the critical dentry name comparison and hashing
1817 * operations one word at a time, but we are limited to:
1818 *
1819 * - Architectures with fast unaligned word accesses. We could
1820 * do a "get_unaligned()" if this helps and is sufficiently
1821 * fast.
1822 *
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001823 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1824 * do not trap on the (extremely unlikely) case of a page
1825 * crossing operation.
1826 *
1827 * - Furthermore, we need an efficient 64-bit compile for the
1828 * 64-bit case in order to generate the "number of bytes in
1829 * the final mask". Again, that could be replaced with a
1830 * efficient population count instruction or similar.
1831 */
1832#ifdef CONFIG_DCACHE_WORD_ACCESS
1833
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001834#include <asm/word-at-a-time.h>
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001835
George Spelvin468a9422016-05-26 22:11:51 -04001836#ifdef HASH_MIX
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001837
George Spelvin468a9422016-05-26 22:11:51 -04001838/* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1839
1840#elif defined(CONFIG_64BIT)
George Spelvin2a18da7a2016-05-23 07:43:58 -04001841/*
1842 * Register pressure in the mixing function is an issue, particularly
1843 * on 32-bit x86, but almost any function requires one state value and
1844 * one temporary. Instead, use a function designed for two state values
1845 * and no temporaries.
1846 *
1847 * This function cannot create a collision in only two iterations, so
1848 * we have two iterations to achieve avalanche. In those two iterations,
1849 * we have six layers of mixing, which is enough to spread one bit's
1850 * influence out to 2^6 = 64 state bits.
1851 *
1852 * Rotate constants are scored by considering either 64 one-bit input
1853 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1854 * probability of that delta causing a change to each of the 128 output
1855 * bits, using a sample of random initial states.
1856 *
1857 * The Shannon entropy of the computed probabilities is then summed
1858 * to produce a score. Ideally, any input change has a 50% chance of
1859 * toggling any given output bit.
1860 *
1861 * Mixing scores (in bits) for (12,45):
1862 * Input delta: 1-bit 2-bit
1863 * 1 round: 713.3 42542.6
1864 * 2 rounds: 2753.7 140389.8
1865 * 3 rounds: 5954.1 233458.2
1866 * 4 rounds: 7862.6 256672.2
1867 * Perfect: 8192 258048
1868 * (64*128) (64*63/2 * 128)
1869 */
1870#define HASH_MIX(x, y, a) \
1871 ( x ^= (a), \
1872 y ^= x, x = rol64(x,12),\
1873 x += y, y = rol64(y,45),\
1874 y *= 9 )
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001875
George Spelvin0fed3ac2016-05-02 06:31:01 -04001876/*
George Spelvin2a18da7a2016-05-23 07:43:58 -04001877 * Fold two longs into one 32-bit hash value. This must be fast, but
1878 * latency isn't quite as critical, as there is a fair bit of additional
1879 * work done before the hash value is used.
George Spelvin0fed3ac2016-05-02 06:31:01 -04001880 */
George Spelvin2a18da7a2016-05-23 07:43:58 -04001881static inline unsigned int fold_hash(unsigned long x, unsigned long y)
George Spelvin0fed3ac2016-05-02 06:31:01 -04001882{
George Spelvin2a18da7a2016-05-23 07:43:58 -04001883 y ^= x * GOLDEN_RATIO_64;
1884 y *= GOLDEN_RATIO_64;
1885 return y >> 32;
George Spelvin0fed3ac2016-05-02 06:31:01 -04001886}
1887
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001888#else /* 32-bit case */
1889
George Spelvin2a18da7a2016-05-23 07:43:58 -04001890/*
1891 * Mixing scores (in bits) for (7,20):
1892 * Input delta: 1-bit 2-bit
1893 * 1 round: 330.3 9201.6
1894 * 2 rounds: 1246.4 25475.4
1895 * 3 rounds: 1907.1 31295.1
1896 * 4 rounds: 2042.3 31718.6
1897 * Perfect: 2048 31744
1898 * (32*64) (32*31/2 * 64)
1899 */
1900#define HASH_MIX(x, y, a) \
1901 ( x ^= (a), \
1902 y ^= x, x = rol32(x, 7),\
1903 x += y, y = rol32(y,20),\
1904 y *= 9 )
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001905
George Spelvin2a18da7a2016-05-23 07:43:58 -04001906static inline unsigned int fold_hash(unsigned long x, unsigned long y)
George Spelvin0fed3ac2016-05-02 06:31:01 -04001907{
George Spelvin2a18da7a2016-05-23 07:43:58 -04001908 /* Use arch-optimized multiply if one exists */
1909 return __hash_32(y ^ __hash_32(x));
George Spelvin0fed3ac2016-05-02 06:31:01 -04001910}
1911
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001912#endif
1913
George Spelvin2a18da7a2016-05-23 07:43:58 -04001914/*
1915 * Return the hash of a string of known length. This is carfully
1916 * designed to match hash_name(), which is the more critical function.
1917 * In particular, we must end by hashing a final word containing 0..7
1918 * payload bytes, to match the way that hash_name() iterates until it
1919 * finds the delimiter after the name.
1920 */
Linus Torvalds8387ff22016-06-10 07:51:30 -07001921unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001922{
Linus Torvalds8387ff22016-06-10 07:51:30 -07001923 unsigned long a, x = 0, y = (unsigned long)salt;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001924
1925 for (;;) {
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001926 if (!len)
1927 goto done;
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001928 a = load_unaligned_zeropad(name);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001929 if (len < sizeof(unsigned long))
1930 break;
George Spelvin2a18da7a2016-05-23 07:43:58 -04001931 HASH_MIX(x, y, a);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001932 name += sizeof(unsigned long);
1933 len -= sizeof(unsigned long);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001934 }
George Spelvin2a18da7a2016-05-23 07:43:58 -04001935 x ^= a & bytemask_from_count(len);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001936done:
George Spelvin2a18da7a2016-05-23 07:43:58 -04001937 return fold_hash(x, y);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001938}
1939EXPORT_SYMBOL(full_name_hash);
1940
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001941/* Return the "hash_len" (hash and length) of a null-terminated string */
Linus Torvalds8387ff22016-06-10 07:51:30 -07001942u64 hashlen_string(const void *salt, const char *name)
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001943{
Linus Torvalds8387ff22016-06-10 07:51:30 -07001944 unsigned long a = 0, x = 0, y = (unsigned long)salt;
1945 unsigned long adata, mask, len;
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001946 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1947
Linus Torvalds8387ff22016-06-10 07:51:30 -07001948 len = 0;
1949 goto inside;
1950
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001951 do {
George Spelvin2a18da7a2016-05-23 07:43:58 -04001952 HASH_MIX(x, y, a);
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001953 len += sizeof(unsigned long);
Linus Torvalds8387ff22016-06-10 07:51:30 -07001954inside:
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001955 a = load_unaligned_zeropad(name+len);
1956 } while (!has_zero(a, &adata, &constants));
1957
1958 adata = prep_zero_mask(a, adata, &constants);
1959 mask = create_zero_mask(adata);
George Spelvin2a18da7a2016-05-23 07:43:58 -04001960 x ^= a & zero_bytemask(mask);
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001961
George Spelvin2a18da7a2016-05-23 07:43:58 -04001962 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001963}
1964EXPORT_SYMBOL(hashlen_string);
1965
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001966/*
1967 * Calculate the length and hash of the path component, and
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001968 * return the "hash_len" as the result.
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001969 */
Linus Torvalds8387ff22016-06-10 07:51:30 -07001970static inline u64 hash_name(const void *salt, const char *name)
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001971{
Linus Torvalds8387ff22016-06-10 07:51:30 -07001972 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
1973 unsigned long adata, bdata, mask, len;
Linus Torvalds36126f82012-05-26 10:43:17 -07001974 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001975
Linus Torvalds8387ff22016-06-10 07:51:30 -07001976 len = 0;
1977 goto inside;
1978
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001979 do {
George Spelvin2a18da7a2016-05-23 07:43:58 -04001980 HASH_MIX(x, y, a);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001981 len += sizeof(unsigned long);
Linus Torvalds8387ff22016-06-10 07:51:30 -07001982inside:
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001983 a = load_unaligned_zeropad(name+len);
Linus Torvalds36126f82012-05-26 10:43:17 -07001984 b = a ^ REPEAT_BYTE('/');
1985 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001986
Linus Torvalds36126f82012-05-26 10:43:17 -07001987 adata = prep_zero_mask(a, adata, &constants);
1988 bdata = prep_zero_mask(b, bdata, &constants);
Linus Torvalds36126f82012-05-26 10:43:17 -07001989 mask = create_zero_mask(adata | bdata);
George Spelvin2a18da7a2016-05-23 07:43:58 -04001990 x ^= a & zero_bytemask(mask);
Linus Torvalds36126f82012-05-26 10:43:17 -07001991
George Spelvin2a18da7a2016-05-23 07:43:58 -04001992 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001993}
1994
George Spelvin2a18da7a2016-05-23 07:43:58 -04001995#else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001996
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001997/* Return the hash of a string of known length */
Linus Torvalds8387ff22016-06-10 07:51:30 -07001998unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
Linus Torvalds0145acc2012-03-02 14:32:59 -08001999{
Linus Torvalds8387ff22016-06-10 07:51:30 -07002000 unsigned long hash = init_name_hash(salt);
Linus Torvalds0145acc2012-03-02 14:32:59 -08002001 while (len--)
George Spelvinfcfd2fb2016-05-20 08:41:37 -04002002 hash = partial_name_hash((unsigned char)*name++, hash);
Linus Torvalds0145acc2012-03-02 14:32:59 -08002003 return end_name_hash(hash);
2004}
Linus Torvaldsae942ae2012-03-02 19:40:57 -08002005EXPORT_SYMBOL(full_name_hash);
Linus Torvalds0145acc2012-03-02 14:32:59 -08002006
George Spelvinfcfd2fb2016-05-20 08:41:37 -04002007/* Return the "hash_len" (hash and length) of a null-terminated string */
Linus Torvalds8387ff22016-06-10 07:51:30 -07002008u64 hashlen_string(const void *salt, const char *name)
George Spelvinfcfd2fb2016-05-20 08:41:37 -04002009{
Linus Torvalds8387ff22016-06-10 07:51:30 -07002010 unsigned long hash = init_name_hash(salt);
George Spelvinfcfd2fb2016-05-20 08:41:37 -04002011 unsigned long len = 0, c;
2012
2013 c = (unsigned char)*name;
George Spelvine0ab7af2016-05-29 08:05:56 -04002014 while (c) {
George Spelvinfcfd2fb2016-05-20 08:41:37 -04002015 len++;
2016 hash = partial_name_hash(c, hash);
2017 c = (unsigned char)name[len];
George Spelvine0ab7af2016-05-29 08:05:56 -04002018 }
George Spelvinfcfd2fb2016-05-20 08:41:37 -04002019 return hashlen_create(end_name_hash(hash), len);
2020}
George Spelvinf2a031b2016-05-29 01:26:41 -04002021EXPORT_SYMBOL(hashlen_string);
George Spelvinfcfd2fb2016-05-20 08:41:37 -04002022
Linus Torvalds3ddcd052011-08-06 22:45:50 -07002023/*
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002024 * We know there's a real path component here of at least
2025 * one character.
2026 */
Linus Torvalds8387ff22016-06-10 07:51:30 -07002027static inline u64 hash_name(const void *salt, const char *name)
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002028{
Linus Torvalds8387ff22016-06-10 07:51:30 -07002029 unsigned long hash = init_name_hash(salt);
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002030 unsigned long len = 0, c;
2031
2032 c = (unsigned char)*name;
2033 do {
2034 len++;
2035 hash = partial_name_hash(c, hash);
2036 c = (unsigned char)name[len];
2037 } while (c && c != '/');
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002038 return hashlen_create(end_name_hash(hash), len);
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002039}
2040
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08002041#endif
2042
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002043/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01002045 * This is the basic name resolution function, turning a pathname into
2046 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01002048 * Returns 0 and nd will have valid dentry and mnt on success.
2049 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 */
Al Viro6de88d72009-08-09 01:41:57 +04002051static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 int err;
Al Viro32cd7462015-04-18 20:30:49 -04002054
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 while (*name=='/')
2056 name++;
2057 if (!*name)
Al Viro9e18f102015-04-18 20:44:34 -04002058 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060 /* At this point we know we have a real path component. */
2061 for(;;) {
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002062 u64 hash_len;
Al Virofe479a52011-02-22 15:10:03 -05002063 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064
Al Viro52094c82011-02-21 21:34:47 -05002065 err = may_lookup(nd);
George Spelvin2a18da7a2016-05-23 07:43:58 -04002066 if (err)
Al Viro3595e232015-05-09 16:54:45 -04002067 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068
Linus Torvalds8387ff22016-06-10 07:51:30 -07002069 hash_len = hash_name(nd->path.dentry, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070
Al Virofe479a52011-02-22 15:10:03 -05002071 type = LAST_NORM;
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002072 if (name[0] == '.') switch (hashlen_len(hash_len)) {
Al Virofe479a52011-02-22 15:10:03 -05002073 case 2:
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002074 if (name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05002075 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05002076 nd->flags |= LOOKUP_JUMPED;
2077 }
Al Virofe479a52011-02-22 15:10:03 -05002078 break;
2079 case 1:
2080 type = LAST_DOT;
2081 }
Al Viro5a202bc2011-03-08 14:17:44 -05002082 if (likely(type == LAST_NORM)) {
2083 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05002084 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05002085 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
James Hogana060dc52014-09-16 13:07:35 +01002086 struct qstr this = { { .hash_len = hash_len }, .name = name };
Linus Torvaldsda53be12013-05-21 15:22:44 -07002087 err = parent->d_op->d_hash(parent, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05002088 if (err < 0)
Al Viro3595e232015-05-09 16:54:45 -04002089 return err;
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002090 hash_len = this.hash_len;
2091 name = this.name;
Al Viro5a202bc2011-03-08 14:17:44 -05002092 }
2093 }
Al Virofe479a52011-02-22 15:10:03 -05002094
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002095 nd->last.hash_len = hash_len;
2096 nd->last.name = name;
Al Viro5f4a6a62013-01-24 18:04:22 -05002097 nd->last_type = type;
2098
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002099 name += hashlen_len(hash_len);
2100 if (!*name)
Al Virobdf6cbf2015-04-18 20:21:40 -04002101 goto OK;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002102 /*
2103 * If it wasn't NUL, we know it was '/'. Skip that
2104 * slash, and continue until no more slashes.
2105 */
2106 do {
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002107 name++;
2108 } while (unlikely(*name == '/'));
Al Viro8620c232015-05-04 08:58:35 -04002109 if (unlikely(!*name)) {
2110OK:
Al Viro368ee9b2015-05-08 17:19:59 -04002111 /* pathname body, done */
Al Viro8620c232015-05-04 08:58:35 -04002112 if (!nd->depth)
2113 return 0;
2114 name = nd->stack[nd->depth - 1].name;
Al Viro368ee9b2015-05-08 17:19:59 -04002115 /* trailing symlink, done */
Al Viro8620c232015-05-04 08:58:35 -04002116 if (!name)
2117 return 0;
2118 /* last component of nested symlink */
Al Viro4693a542015-05-04 17:47:11 -04002119 err = walk_component(nd, WALK_GET | WALK_PUT);
Al Viro8620c232015-05-04 08:58:35 -04002120 } else {
Al Viro4693a542015-05-04 17:47:11 -04002121 err = walk_component(nd, WALK_GET);
Al Viro8620c232015-05-04 08:58:35 -04002122 }
Al Viroce57dfc2011-03-13 19:58:58 -04002123 if (err < 0)
Al Viro3595e232015-05-09 16:54:45 -04002124 return err;
Al Virofe479a52011-02-22 15:10:03 -05002125
Al Viroce57dfc2011-03-13 19:58:58 -04002126 if (err) {
Al Viro626de992015-05-04 18:26:59 -04002127 const char *s = get_link(nd);
Al Viro5a460272015-04-17 23:44:45 -04002128
Viresh Kumara1c83682015-08-12 15:59:44 +05302129 if (IS_ERR(s))
Al Viro3595e232015-05-09 16:54:45 -04002130 return PTR_ERR(s);
Al Virod40bcc02015-04-18 20:03:03 -04002131 err = 0;
2132 if (unlikely(!s)) {
2133 /* jumped */
Al Virob9ff4422015-05-02 20:19:23 -04002134 put_link(nd);
Al Virod40bcc02015-04-18 20:03:03 -04002135 } else {
Al Virofab51e82015-05-10 11:01:00 -04002136 nd->stack[nd->depth - 1].name = name;
2137 name = s;
2138 continue;
Al Virod40bcc02015-04-18 20:03:03 -04002139 }
Nick Piggin31e6b012011-01-07 17:49:52 +11002140 }
Al Viro97242f92015-08-01 19:59:28 -04002141 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2142 if (nd->flags & LOOKUP_RCU) {
2143 if (unlazy_walk(nd, NULL, 0))
2144 return -ECHILD;
2145 }
Al Viro3595e232015-05-09 16:54:45 -04002146 return -ENOTDIR;
Al Viro97242f92015-08-01 19:59:28 -04002147 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149}
2150
Al Viroc8a53ee2015-05-12 18:43:07 -04002151static const char *path_init(struct nameidata *nd, unsigned flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152{
Prasanna Medaea3834d2005-04-29 16:00:17 +01002153 int retval = 0;
Al Viroc8a53ee2015-05-12 18:43:07 -04002154 const char *s = nd->name->name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155
2156 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro980f3ea2014-11-20 14:20:24 -05002157 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158 nd->depth = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05002159 if (flags & LOOKUP_ROOT) {
David Howellsb18825a2013-09-12 19:22:53 +01002160 struct dentry *root = nd->root.dentry;
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07002161 struct vfsmount *mnt = nd->root.mnt;
David Howellsb18825a2013-09-12 19:22:53 +01002162 struct inode *inode = root->d_inode;
Al Virofd2f7cb2015-02-22 20:07:13 -05002163 if (*s) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002164 if (!d_can_lookup(root))
Al Viro368ee9b2015-05-08 17:19:59 -04002165 return ERR_PTR(-ENOTDIR);
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07002166 retval = inode_permission2(mnt, inode, MAY_EXEC);
Al Viro73d049a2011-03-11 12:08:24 -05002167 if (retval)
Al Viro368ee9b2015-05-08 17:19:59 -04002168 return ERR_PTR(retval);
Al Viro73d049a2011-03-11 12:08:24 -05002169 }
Al Viro5b6ca022011-03-09 23:04:47 -05002170 nd->path = nd->root;
2171 nd->inode = inode;
2172 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05002173 rcu_read_lock();
Al Viro5b6ca022011-03-09 23:04:47 -05002174 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro8f47a0162015-05-09 19:02:01 -04002175 nd->root_seq = nd->seq;
Al Viro48a066e2013-09-29 22:06:07 -04002176 nd->m_seq = read_seqbegin(&mount_lock);
Al Viro5b6ca022011-03-09 23:04:47 -05002177 } else {
2178 path_get(&nd->path);
2179 }
Al Viro368ee9b2015-05-08 17:19:59 -04002180 return s;
Al Viro5b6ca022011-03-09 23:04:47 -05002181 }
2182
Al Viro2a737872009-04-07 11:49:53 -04002183 nd->root.mnt = NULL;
Al Viro248fb5b2015-12-05 20:51:58 -05002184 nd->path.mnt = NULL;
2185 nd->path.dentry = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186
Al Viro48a066e2013-09-29 22:06:07 -04002187 nd->m_seq = read_seqbegin(&mount_lock);
Al Virofd2f7cb2015-02-22 20:07:13 -05002188 if (*s == '/') {
Al Viro9e6697e2015-12-05 20:07:21 -05002189 if (flags & LOOKUP_RCU)
Al Viro8b61e742013-11-08 12:45:01 -05002190 rcu_read_lock();
Al Viro9e6697e2015-12-05 20:07:21 -05002191 set_root(nd);
Al Viro248fb5b2015-12-05 20:51:58 -05002192 if (likely(!nd_jump_root(nd)))
Al Viroef55d912015-12-05 20:25:06 -05002193 return s;
Al Viro248fb5b2015-12-05 20:51:58 -05002194 nd->root.mnt = NULL;
Al Viroef55d912015-12-05 20:25:06 -05002195 rcu_read_unlock();
2196 return ERR_PTR(-ECHILD);
Al Viroc8a53ee2015-05-12 18:43:07 -04002197 } else if (nd->dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05002198 if (flags & LOOKUP_RCU) {
2199 struct fs_struct *fs = current->fs;
2200 unsigned seq;
2201
Al Viro8b61e742013-11-08 12:45:01 -05002202 rcu_read_lock();
Al Viroe41f7d42011-02-22 14:02:58 -05002203
2204 do {
2205 seq = read_seqcount_begin(&fs->seq);
2206 nd->path = fs->pwd;
Al Viroef55d912015-12-05 20:25:06 -05002207 nd->inode = nd->path.dentry->d_inode;
Al Viroe41f7d42011-02-22 14:02:58 -05002208 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2209 } while (read_seqcount_retry(&fs->seq, seq));
2210 } else {
2211 get_fs_pwd(current->fs, &nd->path);
Al Viroef55d912015-12-05 20:25:06 -05002212 nd->inode = nd->path.dentry->d_inode;
Al Viroe41f7d42011-02-22 14:02:58 -05002213 }
Al Viroef55d912015-12-05 20:25:06 -05002214 return s;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002215 } else {
Jeff Layton582aa642012-12-11 08:56:16 -05002216 /* Caller must check execute permissions on the starting path component */
Al Viroc8a53ee2015-05-12 18:43:07 -04002217 struct fd f = fdget_raw(nd->dfd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002218 struct dentry *dentry;
2219
Al Viro2903ff02012-08-28 12:52:22 -04002220 if (!f.file)
Al Viro368ee9b2015-05-08 17:19:59 -04002221 return ERR_PTR(-EBADF);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002222
Al Viro2903ff02012-08-28 12:52:22 -04002223 dentry = f.file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002224
Al Virofd2f7cb2015-02-22 20:07:13 -05002225 if (*s) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002226 if (!d_can_lookup(dentry)) {
Al Viro2903ff02012-08-28 12:52:22 -04002227 fdput(f);
Al Viro368ee9b2015-05-08 17:19:59 -04002228 return ERR_PTR(-ENOTDIR);
Al Viro2903ff02012-08-28 12:52:22 -04002229 }
Al Virof52e0c12011-03-14 18:56:51 -04002230 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002231
Al Viro2903ff02012-08-28 12:52:22 -04002232 nd->path = f.file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05002233 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05002234 rcu_read_lock();
Al Viro34a26b92015-05-11 08:05:05 -04002235 nd->inode = nd->path.dentry->d_inode;
2236 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viroe41f7d42011-02-22 14:02:58 -05002237 } else {
Al Viro2903ff02012-08-28 12:52:22 -04002238 path_get(&nd->path);
Al Viro34a26b92015-05-11 08:05:05 -04002239 nd->inode = nd->path.dentry->d_inode;
Al Viroe41f7d42011-02-22 14:02:58 -05002240 }
Al Viro34a26b92015-05-11 08:05:05 -04002241 fdput(f);
Al Viro368ee9b2015-05-08 17:19:59 -04002242 return s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243 }
Al Viro9b4a9b12009-04-07 11:44:16 -04002244}
2245
Al Viro3bdba282015-05-08 17:37:07 -04002246static const char *trailing_symlink(struct nameidata *nd)
Al Viro95fa25d2015-04-22 13:46:57 -04002247{
2248 const char *s;
Al Virofec2fa22015-05-06 15:58:18 -04002249 int error = may_follow_link(nd);
Al Virodeb106c2015-05-08 18:05:21 -04002250 if (unlikely(error))
Al Viro3bdba282015-05-08 17:37:07 -04002251 return ERR_PTR(error);
Al Viro95fa25d2015-04-22 13:46:57 -04002252 nd->flags |= LOOKUP_PARENT;
Al Virofab51e82015-05-10 11:01:00 -04002253 nd->stack[0].name = NULL;
Al Viro3b2e7f72015-04-19 00:53:50 -04002254 s = get_link(nd);
Al Virodeb106c2015-05-08 18:05:21 -04002255 return s ? s : "";
Al Viro95fa25d2015-04-22 13:46:57 -04002256}
2257
Al Virocaa85632015-04-22 17:52:47 -04002258static inline int lookup_last(struct nameidata *nd)
Al Virobd92d7f2011-03-14 19:54:59 -04002259{
2260 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2261 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2262
2263 nd->flags &= ~LOOKUP_PARENT;
Al Virodeb106c2015-05-08 18:05:21 -04002264 return walk_component(nd,
Al Viro4693a542015-05-04 17:47:11 -04002265 nd->flags & LOOKUP_FOLLOW
2266 ? nd->depth
2267 ? WALK_PUT | WALK_GET
2268 : WALK_GET
2269 : 0);
Al Virobd92d7f2011-03-14 19:54:59 -04002270}
2271
Al Viro9b4a9b12009-04-07 11:44:16 -04002272/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroc8a53ee2015-05-12 18:43:07 -04002273static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
Al Viro9b4a9b12009-04-07 11:44:16 -04002274{
Al Viroc8a53ee2015-05-12 18:43:07 -04002275 const char *s = path_init(nd, flags);
Al Virobd92d7f2011-03-14 19:54:59 -04002276 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11002277
Al Viro368ee9b2015-05-08 17:19:59 -04002278 if (IS_ERR(s))
2279 return PTR_ERR(s);
Al Viro3bdba282015-05-08 17:37:07 -04002280 while (!(err = link_path_walk(s, nd))
2281 && ((err = lookup_last(nd)) > 0)) {
2282 s = trailing_symlink(nd);
2283 if (IS_ERR(s)) {
2284 err = PTR_ERR(s);
2285 break;
Al Virobd92d7f2011-03-14 19:54:59 -04002286 }
2287 }
Al Viro9f1fafe2011-03-25 11:00:12 -04002288 if (!err)
2289 err = complete_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04002290
Al Virodeb106c2015-05-08 18:05:21 -04002291 if (!err && nd->flags & LOOKUP_DIRECTORY)
2292 if (!d_can_lookup(nd->path.dentry))
Al Virobd23a532011-03-23 09:56:30 -04002293 err = -ENOTDIR;
Al Viro625b6d12015-05-12 16:36:12 -04002294 if (!err) {
2295 *path = nd->path;
2296 nd->path.mnt = NULL;
2297 nd->path.dentry = NULL;
2298 }
2299 terminate_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04002300 return err;
Al Viroee0827c2011-02-21 23:38:09 -05002301}
Nick Piggin31e6b012011-01-07 17:49:52 +11002302
Al Viro625b6d12015-05-12 16:36:12 -04002303static int filename_lookup(int dfd, struct filename *name, unsigned flags,
Al Viro9ad1aaa2015-05-12 16:44:39 -04002304 struct path *path, struct path *root)
Jeff Layton873f1ee2012-10-10 15:25:29 -04002305{
Al Viro894bc8c2015-05-02 07:16:16 -04002306 int retval;
Al Viro9883d182015-05-13 07:28:08 -04002307 struct nameidata nd;
Al Viroabc9f5b2015-05-12 16:53:42 -04002308 if (IS_ERR(name))
2309 return PTR_ERR(name);
Al Viro9ad1aaa2015-05-12 16:44:39 -04002310 if (unlikely(root)) {
2311 nd.root = *root;
2312 flags |= LOOKUP_ROOT;
2313 }
Al Viro9883d182015-05-13 07:28:08 -04002314 set_nameidata(&nd, dfd, name);
Al Viroc8a53ee2015-05-12 18:43:07 -04002315 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002316 if (unlikely(retval == -ECHILD))
Al Viroc8a53ee2015-05-12 18:43:07 -04002317 retval = path_lookupat(&nd, flags, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002318 if (unlikely(retval == -ESTALE))
Al Viroc8a53ee2015-05-12 18:43:07 -04002319 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002320
2321 if (likely(!retval))
Al Viro625b6d12015-05-12 16:36:12 -04002322 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
Al Viro9883d182015-05-13 07:28:08 -04002323 restore_nameidata();
Al Viroe4bd1c12015-05-12 16:40:39 -04002324 putname(name);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002325 return retval;
2326}
2327
Al Viro8bcb77f2015-05-08 16:59:20 -04002328/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroc8a53ee2015-05-12 18:43:07 -04002329static int path_parentat(struct nameidata *nd, unsigned flags,
Al Viro391172c2015-05-09 11:19:16 -04002330 struct path *parent)
Al Viro8bcb77f2015-05-08 16:59:20 -04002331{
Al Viroc8a53ee2015-05-12 18:43:07 -04002332 const char *s = path_init(nd, flags);
Al Viro368ee9b2015-05-08 17:19:59 -04002333 int err;
2334 if (IS_ERR(s))
2335 return PTR_ERR(s);
2336 err = link_path_walk(s, nd);
Al Viro8bcb77f2015-05-08 16:59:20 -04002337 if (!err)
2338 err = complete_walk(nd);
Al Viro391172c2015-05-09 11:19:16 -04002339 if (!err) {
2340 *parent = nd->path;
2341 nd->path.mnt = NULL;
2342 nd->path.dentry = NULL;
2343 }
2344 terminate_walk(nd);
Al Viro8bcb77f2015-05-08 16:59:20 -04002345 return err;
2346}
2347
Al Viro5c31b6c2015-05-12 17:32:54 -04002348static struct filename *filename_parentat(int dfd, struct filename *name,
Al Viro391172c2015-05-09 11:19:16 -04002349 unsigned int flags, struct path *parent,
2350 struct qstr *last, int *type)
Al Viro8bcb77f2015-05-08 16:59:20 -04002351{
2352 int retval;
Al Viro9883d182015-05-13 07:28:08 -04002353 struct nameidata nd;
Al Viro8bcb77f2015-05-08 16:59:20 -04002354
Al Viro5c31b6c2015-05-12 17:32:54 -04002355 if (IS_ERR(name))
2356 return name;
Al Viro9883d182015-05-13 07:28:08 -04002357 set_nameidata(&nd, dfd, name);
Al Viroc8a53ee2015-05-12 18:43:07 -04002358 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
Al Viro8bcb77f2015-05-08 16:59:20 -04002359 if (unlikely(retval == -ECHILD))
Al Viroc8a53ee2015-05-12 18:43:07 -04002360 retval = path_parentat(&nd, flags, parent);
Al Viro8bcb77f2015-05-08 16:59:20 -04002361 if (unlikely(retval == -ESTALE))
Al Viroc8a53ee2015-05-12 18:43:07 -04002362 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
Al Viro391172c2015-05-09 11:19:16 -04002363 if (likely(!retval)) {
2364 *last = nd.last;
2365 *type = nd.last_type;
2366 audit_inode(name, parent->dentry, LOOKUP_PARENT);
Al Viro5c31b6c2015-05-12 17:32:54 -04002367 } else {
2368 putname(name);
2369 name = ERR_PTR(retval);
Al Viro391172c2015-05-09 11:19:16 -04002370 }
Al Viro9883d182015-05-13 07:28:08 -04002371 restore_nameidata();
Al Viro5c31b6c2015-05-12 17:32:54 -04002372 return name;
Al Viro8bcb77f2015-05-08 16:59:20 -04002373}
2374
Al Viro79714f72012-06-15 03:01:42 +04002375/* does lookup, returns the object with parent locked */
2376struct dentry *kern_path_locked(const char *name, struct path *path)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002377{
Al Viro5c31b6c2015-05-12 17:32:54 -04002378 struct filename *filename;
2379 struct dentry *d;
Al Viro391172c2015-05-09 11:19:16 -04002380 struct qstr last;
2381 int type;
Paul Moore51689102015-01-22 00:00:03 -05002382
Al Viro5c31b6c2015-05-12 17:32:54 -04002383 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2384 &last, &type);
Paul Moore51689102015-01-22 00:00:03 -05002385 if (IS_ERR(filename))
2386 return ERR_CAST(filename);
Al Viro5c31b6c2015-05-12 17:32:54 -04002387 if (unlikely(type != LAST_NORM)) {
Al Viro391172c2015-05-09 11:19:16 -04002388 path_put(path);
Al Viro5c31b6c2015-05-12 17:32:54 -04002389 putname(filename);
2390 return ERR_PTR(-EINVAL);
Al Viro79714f72012-06-15 03:01:42 +04002391 }
Al Viro59551022016-01-22 15:40:57 -05002392 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
Al Viro391172c2015-05-09 11:19:16 -04002393 d = __lookup_hash(&last, path->dentry, 0);
Al Viro79714f72012-06-15 03:01:42 +04002394 if (IS_ERR(d)) {
Al Viro59551022016-01-22 15:40:57 -05002395 inode_unlock(path->dentry->d_inode);
Al Viro391172c2015-05-09 11:19:16 -04002396 path_put(path);
Al Viro79714f72012-06-15 03:01:42 +04002397 }
Paul Moore51689102015-01-22 00:00:03 -05002398 putname(filename);
Al Viro79714f72012-06-15 03:01:42 +04002399 return d;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002400}
2401
Al Virod1811462008-08-02 00:49:18 -04002402int kern_path(const char *name, unsigned int flags, struct path *path)
2403{
Al Viroabc9f5b2015-05-12 16:53:42 -04002404 return filename_lookup(AT_FDCWD, getname_kernel(name),
2405 flags, path, NULL);
Al Virod1811462008-08-02 00:49:18 -04002406}
Al Viro4d359502014-03-14 12:20:17 -04002407EXPORT_SYMBOL(kern_path);
Al Virod1811462008-08-02 00:49:18 -04002408
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002409/**
2410 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2411 * @dentry: pointer to dentry of the base directory
2412 * @mnt: pointer to vfs mount of the base directory
2413 * @name: pointer to file name
2414 * @flags: lookup flags
Al Viroe0a01242011-06-27 17:00:37 -04002415 * @path: pointer to struct path to fill
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002416 */
2417int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2418 const char *name, unsigned int flags,
Al Viroe0a01242011-06-27 17:00:37 -04002419 struct path *path)
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002420{
Al Viro9ad1aaa2015-05-12 16:44:39 -04002421 struct path root = {.mnt = mnt, .dentry = dentry};
Al Viro9ad1aaa2015-05-12 16:44:39 -04002422 /* the first argument of filename_lookup() is ignored with root */
Al Viroabc9f5b2015-05-12 16:53:42 -04002423 return filename_lookup(AT_FDCWD, getname_kernel(name),
2424 flags , path, &root);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002425}
Al Viro4d359502014-03-14 12:20:17 -04002426EXPORT_SYMBOL(vfs_path_lookup);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002427
Christoph Hellwigeead1912007-10-16 23:25:38 -07002428/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07002429 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07002430 * @name: pathname component to lookup
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07002431 * @mnt: mount we are looking up on
Christoph Hellwigeead1912007-10-16 23:25:38 -07002432 * @base: base directory to lookup from
2433 * @len: maximum length @len should be interpreted to
2434 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07002435 * Note that this routine is purely a helper for filesystem usage and should
Al Viro9e7543e2015-02-23 02:49:48 -05002436 * not be called by generic code.
NeilBrownbbddca82016-01-07 16:08:20 -05002437 *
2438 * The caller must hold base->i_mutex.
Christoph Hellwigeead1912007-10-16 23:25:38 -07002439 */
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07002440struct dentry *lookup_one_len2(const char *name, struct vfsmount *mnt, struct dentry *base, int len)
James Morris057f6c02007-04-26 00:12:05 -07002441{
James Morris057f6c02007-04-26 00:12:05 -07002442 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05002443 unsigned int c;
Miklos Szeredicda309d2012-03-26 12:54:21 +02002444 int err;
James Morris057f6c02007-04-26 00:12:05 -07002445
Al Viro59551022016-01-22 15:40:57 -05002446 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
David Woodhouse2f9092e2009-04-20 23:18:37 +01002447
Al Viro6a96ba52011-03-07 23:49:20 -05002448 this.name = name;
2449 this.len = len;
Linus Torvalds8387ff22016-06-10 07:51:30 -07002450 this.hash = full_name_hash(base, name, len);
Al Viro6a96ba52011-03-07 23:49:20 -05002451 if (!len)
2452 return ERR_PTR(-EACCES);
2453
Al Viro21d8a152012-11-29 22:17:21 -05002454 if (unlikely(name[0] == '.')) {
2455 if (len < 2 || (len == 2 && name[1] == '.'))
2456 return ERR_PTR(-EACCES);
2457 }
2458
Al Viro6a96ba52011-03-07 23:49:20 -05002459 while (len--) {
2460 c = *(const unsigned char *)name++;
2461 if (c == '/' || c == '\0')
2462 return ERR_PTR(-EACCES);
Al Viro6a96ba52011-03-07 23:49:20 -05002463 }
Al Viro5a202bc2011-03-08 14:17:44 -05002464 /*
2465 * See if the low-level filesystem might want
2466 * to use its own hash..
2467 */
2468 if (base->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002469 int err = base->d_op->d_hash(base, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05002470 if (err < 0)
2471 return ERR_PTR(err);
2472 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07002473
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07002474 err = inode_permission2(mnt, base->d_inode, MAY_EXEC);
Miklos Szeredicda309d2012-03-26 12:54:21 +02002475 if (err)
2476 return ERR_PTR(err);
2477
Al Viro72bd8662012-06-10 17:17:17 -04002478 return __lookup_hash(&this, base, 0);
James Morris057f6c02007-04-26 00:12:05 -07002479}
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07002480EXPORT_SYMBOL(lookup_one_len2);
2481
2482struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2483{
2484 return lookup_one_len2(name, NULL, base, len);
2485}
Al Viro4d359502014-03-14 12:20:17 -04002486EXPORT_SYMBOL(lookup_one_len);
James Morris057f6c02007-04-26 00:12:05 -07002487
NeilBrownbbddca82016-01-07 16:08:20 -05002488/**
2489 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2490 * @name: pathname component to lookup
2491 * @base: base directory to lookup from
2492 * @len: maximum length @len should be interpreted to
2493 *
2494 * Note that this routine is purely a helper for filesystem usage and should
2495 * not be called by generic code.
2496 *
2497 * Unlike lookup_one_len, it should be called without the parent
2498 * i_mutex held, and will take the i_mutex itself if necessary.
2499 */
2500struct dentry *lookup_one_len_unlocked(const char *name,
2501 struct dentry *base, int len)
2502{
2503 struct qstr this;
2504 unsigned int c;
2505 int err;
Linus Torvalds20d00ee2016-07-29 12:17:52 -07002506 struct dentry *ret;
NeilBrownbbddca82016-01-07 16:08:20 -05002507
2508 this.name = name;
2509 this.len = len;
Linus Torvalds8387ff22016-06-10 07:51:30 -07002510 this.hash = full_name_hash(base, name, len);
NeilBrownbbddca82016-01-07 16:08:20 -05002511 if (!len)
2512 return ERR_PTR(-EACCES);
2513
2514 if (unlikely(name[0] == '.')) {
2515 if (len < 2 || (len == 2 && name[1] == '.'))
2516 return ERR_PTR(-EACCES);
2517 }
2518
2519 while (len--) {
2520 c = *(const unsigned char *)name++;
2521 if (c == '/' || c == '\0')
2522 return ERR_PTR(-EACCES);
2523 }
2524 /*
2525 * See if the low-level filesystem might want
2526 * to use its own hash..
2527 */
2528 if (base->d_flags & DCACHE_OP_HASH) {
2529 int err = base->d_op->d_hash(base, &this);
2530 if (err < 0)
2531 return ERR_PTR(err);
2532 }
2533
2534 err = inode_permission(base->d_inode, MAY_EXEC);
2535 if (err)
2536 return ERR_PTR(err);
2537
Linus Torvalds20d00ee2016-07-29 12:17:52 -07002538 ret = lookup_dcache(&this, base, 0);
2539 if (!ret)
2540 ret = lookup_slow(&this, base, 0);
2541 return ret;
NeilBrownbbddca82016-01-07 16:08:20 -05002542}
2543EXPORT_SYMBOL(lookup_one_len_unlocked);
2544
Eric W. Biedermaneedf2652016-06-02 10:29:47 -05002545#ifdef CONFIG_UNIX98_PTYS
2546int path_pts(struct path *path)
2547{
2548 /* Find something mounted on "pts" in the same directory as
2549 * the input path.
2550 */
2551 struct dentry *child, *parent;
2552 struct qstr this;
2553 int ret;
2554
2555 ret = path_parent_directory(path);
2556 if (ret)
2557 return ret;
2558
2559 parent = path->dentry;
2560 this.name = "pts";
2561 this.len = 3;
2562 child = d_hash_and_lookup(parent, &this);
2563 if (!child)
2564 return -ENOENT;
2565
2566 path->dentry = child;
2567 dput(parent);
2568 follow_mount(path);
2569 return 0;
2570}
2571#endif
2572
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002573int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2574 struct path *path, int *empty)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575{
Al Viroabc9f5b2015-05-12 16:53:42 -04002576 return filename_lookup(dfd, getname_flags(name, flags, empty),
2577 flags, path, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578}
Al Virob853a162015-05-13 09:12:02 -04002579EXPORT_SYMBOL(user_path_at_empty);
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002580
Jeff Layton873f1ee2012-10-10 15:25:29 -04002581/*
2582 * NB: most callers don't do anything directly with the reference to the
2583 * to struct filename, but the nd->last pointer points into the name string
2584 * allocated by getname. So we must hold the reference to it until all
2585 * path-walking is complete.
2586 */
Al Viroa2ec4a22015-05-13 06:57:49 -04002587static inline struct filename *
Al Virof5beed72015-04-30 16:09:11 -04002588user_path_parent(int dfd, const char __user *path,
2589 struct path *parent,
2590 struct qstr *last,
2591 int *type,
Jeff Layton9e790bd2012-12-11 12:10:09 -05002592 unsigned int flags)
Al Viro2ad94ae2008-07-21 09:32:51 -04002593{
Jeff Layton9e790bd2012-12-11 12:10:09 -05002594 /* only LOOKUP_REVAL is allowed in extra flags */
Al Viro5c31b6c2015-05-12 17:32:54 -04002595 return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL,
2596 parent, last, type);
Al Viro2ad94ae2008-07-21 09:32:51 -04002597}
2598
Jeff Layton80334262013-07-26 06:23:25 -04002599/**
Al Viro197df042013-09-08 14:03:27 -04002600 * mountpoint_last - look up last component for umount
Jeff Layton80334262013-07-26 06:23:25 -04002601 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2602 * @path: pointer to container for result
2603 *
2604 * This is a special lookup_last function just for umount. In this case, we
2605 * need to resolve the path without doing any revalidation.
2606 *
2607 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2608 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2609 * in almost all cases, this lookup will be served out of the dcache. The only
2610 * cases where it won't are if nd->last refers to a symlink or the path is
2611 * bogus and it doesn't exist.
2612 *
2613 * Returns:
2614 * -error: if there was an error during lookup. This includes -ENOENT if the
2615 * lookup found a negative dentry. The nd->path reference will also be
2616 * put in this case.
2617 *
2618 * 0: if we successfully resolved nd->path and found it to not to be a
2619 * symlink that needs to be followed. "path" will also be populated.
2620 * The nd->path reference will also be put.
2621 *
2622 * 1: if we successfully resolved nd->last and found it to be a symlink
2623 * that needs to be followed. "path" will be populated with the path
2624 * to the link, and nd->path will *not* be put.
2625 */
2626static int
Al Viro197df042013-09-08 14:03:27 -04002627mountpoint_last(struct nameidata *nd, struct path *path)
Jeff Layton80334262013-07-26 06:23:25 -04002628{
2629 int error = 0;
2630 struct dentry *dentry;
2631 struct dentry *dir = nd->path.dentry;
2632
Al Viro35759522013-09-08 13:41:33 -04002633 /* If we're in rcuwalk, drop out of it to handle last component */
2634 if (nd->flags & LOOKUP_RCU) {
Al Viro6e9918b2015-05-05 09:26:05 -04002635 if (unlazy_walk(nd, NULL, 0))
Al Virodeb106c2015-05-08 18:05:21 -04002636 return -ECHILD;
Jeff Layton80334262013-07-26 06:23:25 -04002637 }
2638
2639 nd->flags &= ~LOOKUP_PARENT;
2640
2641 if (unlikely(nd->last_type != LAST_NORM)) {
2642 error = handle_dots(nd, nd->last_type);
Al Viro35759522013-09-08 13:41:33 -04002643 if (error)
Al Virodeb106c2015-05-08 18:05:21 -04002644 return error;
Al Viro35759522013-09-08 13:41:33 -04002645 dentry = dget(nd->path.dentry);
Al Viro949a8522016-03-06 14:20:52 -05002646 } else {
2647 dentry = d_lookup(dir, &nd->last);
Jeff Layton80334262013-07-26 06:23:25 -04002648 if (!dentry) {
Al Viro949a8522016-03-06 14:20:52 -05002649 /*
2650 * No cached dentry. Mounted dentries are pinned in the
2651 * cache, so that means that this dentry is probably
2652 * a symlink or the path doesn't actually point
2653 * to a mounted dentry.
2654 */
2655 dentry = lookup_slow(&nd->last, dir,
2656 nd->flags | LOOKUP_NO_REVAL);
2657 if (IS_ERR(dentry))
2658 return PTR_ERR(dentry);
Dave Jonesbcceeeb2013-09-10 17:04:25 -04002659 }
Jeff Layton80334262013-07-26 06:23:25 -04002660 }
David Howells698934d2015-03-17 17:33:52 +00002661 if (d_is_negative(dentry)) {
Al Viro35759522013-09-08 13:41:33 -04002662 dput(dentry);
Al Virodeb106c2015-05-08 18:05:21 -04002663 return -ENOENT;
Jeff Layton80334262013-07-26 06:23:25 -04002664 }
Al Viro191d7f72015-05-04 08:26:45 -04002665 if (nd->depth)
2666 put_link(nd);
Al Viro35759522013-09-08 13:41:33 -04002667 path->dentry = dentry;
Vasily Averin295dc392014-07-21 12:30:23 +04002668 path->mnt = nd->path.mnt;
Al Viro181548c2015-05-07 19:54:34 -04002669 error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW,
2670 d_backing_inode(dentry), 0);
Al Virodeb106c2015-05-08 18:05:21 -04002671 if (unlikely(error))
Al Virod63ff282015-05-04 18:13:23 -04002672 return error;
Vasily Averin295dc392014-07-21 12:30:23 +04002673 mntget(path->mnt);
Al Viro35759522013-09-08 13:41:33 -04002674 follow_mount(path);
Al Virodeb106c2015-05-08 18:05:21 -04002675 return 0;
Jeff Layton80334262013-07-26 06:23:25 -04002676}
2677
2678/**
Al Viro197df042013-09-08 14:03:27 -04002679 * path_mountpoint - look up a path to be umounted
Masanari Iida2a78b852015-09-09 15:39:23 -07002680 * @nd: lookup context
Jeff Layton80334262013-07-26 06:23:25 -04002681 * @flags: lookup flags
Al Viroc8a53ee2015-05-12 18:43:07 -04002682 * @path: pointer to container for result
Jeff Layton80334262013-07-26 06:23:25 -04002683 *
2684 * Look up the given name, but don't attempt to revalidate the last component.
Randy Dunlap606d6fe2013-10-19 14:56:55 -07002685 * Returns 0 and "path" will be valid on success; Returns error otherwise.
Jeff Layton80334262013-07-26 06:23:25 -04002686 */
2687static int
Al Viroc8a53ee2015-05-12 18:43:07 -04002688path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
Jeff Layton80334262013-07-26 06:23:25 -04002689{
Al Viroc8a53ee2015-05-12 18:43:07 -04002690 const char *s = path_init(nd, flags);
Al Viro368ee9b2015-05-08 17:19:59 -04002691 int err;
2692 if (IS_ERR(s))
2693 return PTR_ERR(s);
Al Viro3bdba282015-05-08 17:37:07 -04002694 while (!(err = link_path_walk(s, nd)) &&
2695 (err = mountpoint_last(nd, path)) > 0) {
2696 s = trailing_symlink(nd);
2697 if (IS_ERR(s)) {
2698 err = PTR_ERR(s);
Jeff Layton80334262013-07-26 06:23:25 -04002699 break;
Al Viro3bdba282015-05-08 17:37:07 -04002700 }
Jeff Layton80334262013-07-26 06:23:25 -04002701 }
Al Virodeb106c2015-05-08 18:05:21 -04002702 terminate_walk(nd);
Jeff Layton80334262013-07-26 06:23:25 -04002703 return err;
2704}
2705
Al Viro2d864652013-09-08 20:18:44 -04002706static int
Al Viro668696d2015-02-22 19:44:00 -05002707filename_mountpoint(int dfd, struct filename *name, struct path *path,
Al Viro2d864652013-09-08 20:18:44 -04002708 unsigned int flags)
2709{
Al Viro9883d182015-05-13 07:28:08 -04002710 struct nameidata nd;
Al Virocbaab2d2015-01-22 02:49:00 -05002711 int error;
Al Viro668696d2015-02-22 19:44:00 -05002712 if (IS_ERR(name))
2713 return PTR_ERR(name);
Al Viro9883d182015-05-13 07:28:08 -04002714 set_nameidata(&nd, dfd, name);
Al Viroc8a53ee2015-05-12 18:43:07 -04002715 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
Al Viro2d864652013-09-08 20:18:44 -04002716 if (unlikely(error == -ECHILD))
Al Viroc8a53ee2015-05-12 18:43:07 -04002717 error = path_mountpoint(&nd, flags, path);
Al Viro2d864652013-09-08 20:18:44 -04002718 if (unlikely(error == -ESTALE))
Al Viroc8a53ee2015-05-12 18:43:07 -04002719 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
Al Viro2d864652013-09-08 20:18:44 -04002720 if (likely(!error))
Al Viro668696d2015-02-22 19:44:00 -05002721 audit_inode(name, path->dentry, 0);
Al Viro9883d182015-05-13 07:28:08 -04002722 restore_nameidata();
Al Viro668696d2015-02-22 19:44:00 -05002723 putname(name);
Al Viro2d864652013-09-08 20:18:44 -04002724 return error;
2725}
2726
Jeff Layton80334262013-07-26 06:23:25 -04002727/**
Al Viro197df042013-09-08 14:03:27 -04002728 * user_path_mountpoint_at - lookup a path from userland in order to umount it
Jeff Layton80334262013-07-26 06:23:25 -04002729 * @dfd: directory file descriptor
2730 * @name: pathname from userland
2731 * @flags: lookup flags
2732 * @path: pointer to container to hold result
2733 *
2734 * A umount is a special case for path walking. We're not actually interested
2735 * in the inode in this situation, and ESTALE errors can be a problem. We
2736 * simply want track down the dentry and vfsmount attached at the mountpoint
2737 * and avoid revalidating the last component.
2738 *
2739 * Returns 0 and populates "path" on success.
2740 */
2741int
Al Viro197df042013-09-08 14:03:27 -04002742user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
Jeff Layton80334262013-07-26 06:23:25 -04002743 struct path *path)
2744{
Al Virocbaab2d2015-01-22 02:49:00 -05002745 return filename_mountpoint(dfd, getname(name), path, flags);
Jeff Layton80334262013-07-26 06:23:25 -04002746}
2747
Al Viro2d864652013-09-08 20:18:44 -04002748int
2749kern_path_mountpoint(int dfd, const char *name, struct path *path,
2750 unsigned int flags)
2751{
Al Virocbaab2d2015-01-22 02:49:00 -05002752 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
Al Viro2d864652013-09-08 20:18:44 -04002753}
2754EXPORT_SYMBOL(kern_path_mountpoint);
2755
Miklos Szeredicbdf35b2014-10-24 00:14:36 +02002756int __check_sticky(struct inode *dir, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002757{
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002758 kuid_t fsuid = current_fsuid();
David Howellsda9592e2008-11-14 10:39:05 +11002759
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002760 if (uid_eq(inode->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761 return 0;
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002762 if (uid_eq(dir->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002763 return 0;
Andy Lutomirski23adbe12014-06-10 12:45:42 -07002764 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002765}
Miklos Szeredicbdf35b2014-10-24 00:14:36 +02002766EXPORT_SYMBOL(__check_sticky);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002767
2768/*
2769 * Check whether we can remove a link victim from directory dir, check
2770 * whether the type of victim is right.
2771 * 1. We can't do it if dir is read-only (done in permission())
2772 * 2. We should have write and exec permissions on dir
2773 * 3. We can't remove anything from append-only dir
2774 * 4. We can't do anything with immutable dir (done in permission())
2775 * 5. If the sticky bit on dir is set we should either
2776 * a. be owner of dir, or
2777 * b. be owner of victim, or
2778 * c. have CAP_FOWNER capability
2779 * 6. If the victim is append-only or immutable we can't do antyhing with
2780 * links pointing to it.
Eric W. Biederman0bd23d092016-06-29 14:54:46 -05002781 * 7. If the victim has an unknown uid or gid we can't change the inode.
2782 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2783 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2784 * 10. We can't remove a root or mountpoint.
2785 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786 * nfs_async_unlink().
2787 */
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07002788static int may_delete(struct vfsmount *mnt, struct inode *dir, struct dentry *victim, bool isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002789{
David Howells63afdfc2015-05-06 15:59:00 +01002790 struct inode *inode = d_backing_inode(victim);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791 int error;
2792
David Howellsb18825a2013-09-12 19:22:53 +01002793 if (d_is_negative(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794 return -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01002795 BUG_ON(!inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002796
2797 BUG_ON(victim->d_parent->d_inode != dir);
Jeff Layton4fa6b5e2012-10-10 15:25:25 -04002798 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002799
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07002800 error = inode_permission2(mnt, dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002801 if (error)
2802 return error;
2803 if (IS_APPEND(dir))
2804 return -EPERM;
David Howellsb18825a2013-09-12 19:22:53 +01002805
2806 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
Eric W. Biederman0bd23d092016-06-29 14:54:46 -05002807 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002808 return -EPERM;
2809 if (isdir) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002810 if (!d_is_dir(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811 return -ENOTDIR;
2812 if (IS_ROOT(victim))
2813 return -EBUSY;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002814 } else if (d_is_dir(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815 return -EISDIR;
2816 if (IS_DEADDIR(dir))
2817 return -ENOENT;
2818 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2819 return -EBUSY;
2820 return 0;
2821}
2822
2823/* Check whether we can create an object with dentry child in directory
2824 * dir.
2825 * 1. We can't do it if child already exists (open has special treatment for
2826 * this case, but since we are inlined it's OK)
2827 * 2. We can't do it if dir is read-only (done in permission())
Eric W. Biederman036d5232016-07-01 12:52:06 -05002828 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2829 * 4. We should have write and exec permissions on dir
2830 * 5. We can't do it if dir is immutable (done in permission())
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831 */
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07002832static inline int may_create(struct vfsmount *mnt, struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002833{
Eric W. Biederman036d5232016-07-01 12:52:06 -05002834 struct user_namespace *s_user_ns;
Jeff Layton14e972b2013-05-08 10:25:58 -04002835 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836 if (child->d_inode)
2837 return -EEXIST;
2838 if (IS_DEADDIR(dir))
2839 return -ENOENT;
Eric W. Biederman036d5232016-07-01 12:52:06 -05002840 s_user_ns = dir->i_sb->s_user_ns;
2841 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2842 !kgid_has_mapping(s_user_ns, current_fsgid()))
2843 return -EOVERFLOW;
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07002844 return inode_permission2(mnt, dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845}
2846
Linus Torvalds1da177e2005-04-16 15:20:36 -07002847/*
2848 * p1 and p2 should be directories on the same fs.
2849 */
2850struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2851{
2852 struct dentry *p;
2853
2854 if (p1 == p2) {
Al Viro59551022016-01-22 15:40:57 -05002855 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002856 return NULL;
2857 }
2858
Al Virofc640052016-04-10 01:33:30 -04002859 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002860
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002861 p = d_ancestor(p2, p1);
2862 if (p) {
Al Viro59551022016-01-22 15:40:57 -05002863 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2864 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002865 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002866 }
2867
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002868 p = d_ancestor(p1, p2);
2869 if (p) {
Al Viro59551022016-01-22 15:40:57 -05002870 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2871 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002872 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002873 }
2874
Al Viro59551022016-01-22 15:40:57 -05002875 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2876 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002877 return NULL;
2878}
Al Viro4d359502014-03-14 12:20:17 -04002879EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002880
2881void unlock_rename(struct dentry *p1, struct dentry *p2)
2882{
Al Viro59551022016-01-22 15:40:57 -05002883 inode_unlock(p1->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002884 if (p1 != p2) {
Al Viro59551022016-01-22 15:40:57 -05002885 inode_unlock(p2->d_inode);
Al Virofc640052016-04-10 01:33:30 -04002886 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002887 }
2888}
Al Viro4d359502014-03-14 12:20:17 -04002889EXPORT_SYMBOL(unlock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002890
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07002891int vfs_create2(struct vfsmount *mnt, struct inode *dir, struct dentry *dentry,
2892 umode_t mode, bool want_excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002893{
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07002894 int error = may_create(mnt, dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002895 if (error)
2896 return error;
2897
Al Viroacfa4382008-12-04 10:06:33 -05002898 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002899 return -EACCES; /* shouldn't it be ENOSYS? */
2900 mode &= S_IALLUGO;
2901 mode |= S_IFREG;
2902 error = security_inode_create(dir, dentry, mode);
2903 if (error)
2904 return error;
Al Viro312b63f2012-06-10 18:09:36 -04002905 error = dir->i_op->create(dir, dentry, mode, want_excl);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002906 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002907 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002908 return error;
2909}
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07002910EXPORT_SYMBOL(vfs_create2);
2911
2912int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2913 bool want_excl)
2914{
2915 return vfs_create2(NULL, dir, dentry, mode, want_excl);
2916}
Al Viro4d359502014-03-14 12:20:17 -04002917EXPORT_SYMBOL(vfs_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002918
Eric W. Biedermana2982cc2016-06-09 15:34:02 -05002919bool may_open_dev(const struct path *path)
2920{
2921 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2922 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2923}
2924
Al Viro73d049a2011-03-11 12:08:24 -05002925static int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002926{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002927 struct dentry *dentry = path->dentry;
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07002928 struct vfsmount *mnt = path->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002929 struct inode *inode = dentry->d_inode;
2930 int error;
2931
2932 if (!inode)
2933 return -ENOENT;
2934
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002935 switch (inode->i_mode & S_IFMT) {
2936 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002937 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002938 case S_IFDIR:
2939 if (acc_mode & MAY_WRITE)
2940 return -EISDIR;
2941 break;
2942 case S_IFBLK:
2943 case S_IFCHR:
Eric W. Biedermana2982cc2016-06-09 15:34:02 -05002944 if (!may_open_dev(path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002945 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002946 /*FALLTHRU*/
2947 case S_IFIFO:
2948 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002950 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002951 }
Dave Hansenb41572e2007-10-16 23:31:14 -07002952
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07002953 error = inode_permission2(mnt, inode, MAY_OPEN | acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07002954 if (error)
2955 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05002956
Linus Torvalds1da177e2005-04-16 15:20:36 -07002957 /*
2958 * An append-only file must be opened in append mode for writing.
2959 */
2960 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05002961 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05002962 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002963 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05002964 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002965 }
2966
2967 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002968 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05002969 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002970
J. Bruce Fieldsf3c7691e2011-09-21 10:58:13 -04002971 return 0;
Al Viro7715b522009-12-16 03:54:00 -05002972}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002973
Jeff Laytone1181ee2010-12-07 16:19:50 -05002974static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002975{
Jeff Laytone1181ee2010-12-07 16:19:50 -05002976 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002977 struct inode *inode = path->dentry->d_inode;
2978 int error = get_write_access(inode);
2979 if (error)
2980 return error;
2981 /*
2982 * Refuse to truncate files with mandatory locks held on them.
2983 */
Jeff Laytond7a06982014-03-10 09:54:15 -04002984 error = locks_verify_locked(filp);
Al Viro7715b522009-12-16 03:54:00 -05002985 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002986 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002987 if (!error) {
Daniel Rosenberg2757e9b2016-10-26 16:33:11 -07002988 error = do_truncate2(path->mnt, path->dentry, 0,
Al Viro7715b522009-12-16 03:54:00 -05002989 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002990 filp);
Al Viro7715b522009-12-16 03:54:00 -05002991 }
2992 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002993 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002994}
2995
Dave Hansend57999e2008-02-15 14:37:27 -08002996static inline int open_to_namei_flags(int flag)
2997{
Al Viro8a5e9292011-06-25 19:15:54 -04002998 if ((flag & O_ACCMODE) == 3)
2999 flag--;
Dave Hansend57999e2008-02-15 14:37:27 -08003000 return flag;
3001}
3002
Al Virod3607752016-03-25 15:21:09 -04003003static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
Miklos Szeredid18e9002012-06-05 15:10:17 +02003004{
3005 int error = security_path_mknod(dir, dentry, mode, 0);
3006 if (error)
3007 return error;
3008
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07003009 error = inode_permission2(dir->mnt, dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
Miklos Szeredid18e9002012-06-05 15:10:17 +02003010 if (error)
3011 return error;
3012
3013 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3014}
3015
David Howells1acf0af2012-06-14 16:13:46 +01003016/*
3017 * Attempt to atomically look up, create and open a file from a negative
3018 * dentry.
3019 *
3020 * Returns 0 if successful. The file will have been created and attached to
3021 * @file by the filesystem calling finish_open().
3022 *
3023 * Returns 1 if the file was looked up only or didn't need creating. The
3024 * caller will need to perform the open themselves. @path will have been
3025 * updated to point to the new dentry. This may be negative.
3026 *
3027 * Returns an error code otherwise.
3028 */
Al Viro2675a4e2012-06-22 12:41:10 +04003029static int atomic_open(struct nameidata *nd, struct dentry *dentry,
3030 struct path *path, struct file *file,
3031 const struct open_flags *op,
Al Viro1643b432016-04-27 19:14:10 -04003032 int open_flag, umode_t mode,
Al Viro2675a4e2012-06-22 12:41:10 +04003033 int *opened)
Miklos Szeredid18e9002012-06-05 15:10:17 +02003034{
Miklos Szeredid18e9002012-06-05 15:10:17 +02003035 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003036 struct inode *dir = nd->path.dentry->d_inode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003037 int error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003038
Al Viro384f26e2016-04-28 02:03:55 -04003039 if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */
Miklos Szeredid18e9002012-06-05 15:10:17 +02003040 open_flag &= ~O_TRUNC;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003041
Miklos Szeredid18e9002012-06-05 15:10:17 +02003042 if (nd->flags & LOOKUP_DIRECTORY)
3043 open_flag |= O_DIRECTORY;
3044
Al Viro30d90492012-06-22 12:40:19 +04003045 file->f_path.dentry = DENTRY_NOT_SET;
3046 file->f_path.mnt = nd->path.mnt;
Al Viro0fb1ea02016-04-27 14:13:10 -04003047 error = dir->i_op->atomic_open(dir, dentry, file,
3048 open_to_namei_flags(open_flag),
3049 mode, opened);
Al Viro6fbd0712016-04-28 11:50:59 -04003050 d_lookup_done(dentry);
Al Viro384f26e2016-04-28 02:03:55 -04003051 if (!error) {
3052 /*
3053 * We didn't have the inode before the open, so check open
3054 * permission here.
3055 */
3056 int acc_mode = op->acc_mode;
3057 if (*opened & FILE_CREATED) {
3058 WARN_ON(!(open_flag & O_CREAT));
3059 fsnotify_create(dir, dentry);
3060 acc_mode = 0;
3061 }
3062 error = may_open(&file->f_path, acc_mode, open_flag);
3063 if (WARN_ON(error > 0))
3064 error = -EINVAL;
3065 } else if (error > 0) {
Al Viro30d90492012-06-22 12:40:19 +04003066 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
Al Viro2675a4e2012-06-22 12:41:10 +04003067 error = -EIO;
Al Viro03da6332013-09-16 19:22:33 -04003068 } else {
Al Viro384f26e2016-04-28 02:03:55 -04003069 if (file->f_path.dentry) {
3070 dput(dentry);
3071 dentry = file->f_path.dentry;
Al Viro03da6332013-09-16 19:22:33 -04003072 }
Al Viro384f26e2016-04-28 02:03:55 -04003073 if (*opened & FILE_CREATED)
3074 fsnotify_create(dir, dentry);
Al Viroa01e7182016-06-07 21:53:51 -04003075 if (unlikely(d_is_negative(dentry))) {
3076 error = -ENOENT;
3077 } else {
3078 path->dentry = dentry;
3079 path->mnt = nd->path.mnt;
3080 return 1;
3081 }
Sage Weil62b2ce92012-08-15 13:30:12 -07003082 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02003083 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02003084 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04003085 return error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003086}
3087
Nick Piggin31e6b012011-01-07 17:49:52 +11003088/*
David Howells1acf0af2012-06-14 16:13:46 +01003089 * Look up and maybe create and open the last component.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003090 *
3091 * Must be called with i_mutex held on parent.
3092 *
David Howells1acf0af2012-06-14 16:13:46 +01003093 * Returns 0 if the file was successfully atomically created (if necessary) and
3094 * opened. In this case the file will be returned attached to @file.
3095 *
3096 * Returns 1 if the file was not completely opened at this time, though lookups
3097 * and creations will have been performed and the dentry returned in @path will
3098 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
3099 * specified then a negative dentry may be returned.
3100 *
3101 * An error code is returned otherwise.
3102 *
3103 * FILE_CREATE will be set in @*opened if the dentry was created and will be
3104 * cleared otherwise prior to returning.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003105 */
Al Viro2675a4e2012-06-22 12:41:10 +04003106static int lookup_open(struct nameidata *nd, struct path *path,
3107 struct file *file,
3108 const struct open_flags *op,
Al Viro64894cf2012-07-31 00:53:35 +04003109 bool got_write, int *opened)
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003110{
3111 struct dentry *dir = nd->path.dentry;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02003112 struct inode *dir_inode = dir->d_inode;
Al Viro1643b432016-04-27 19:14:10 -04003113 int open_flag = op->open_flag;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003114 struct dentry *dentry;
Al Viro1643b432016-04-27 19:14:10 -04003115 int error, create_error = 0;
Al Viro1643b432016-04-27 19:14:10 -04003116 umode_t mode = op->mode;
Al Viro6fbd0712016-04-28 11:50:59 -04003117 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003118
Al Viroce8644f2016-04-26 14:17:56 -04003119 if (unlikely(IS_DEADDIR(dir_inode)))
3120 return -ENOENT;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003121
Al Viro47237682012-06-10 05:01:45 -04003122 *opened &= ~FILE_CREATED;
Al Viro6fbd0712016-04-28 11:50:59 -04003123 dentry = d_lookup(dir, &nd->last);
3124 for (;;) {
3125 if (!dentry) {
3126 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3127 if (IS_ERR(dentry))
3128 return PTR_ERR(dentry);
3129 }
3130 if (d_in_lookup(dentry))
3131 break;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003132
Al Viro6fbd0712016-04-28 11:50:59 -04003133 if (!(dentry->d_flags & DCACHE_OP_REVALIDATE))
3134 break;
3135
3136 error = d_revalidate(dentry, nd->flags);
3137 if (likely(error > 0))
3138 break;
3139 if (error)
3140 goto out_dput;
3141 d_invalidate(dentry);
3142 dput(dentry);
3143 dentry = NULL;
3144 }
3145 if (dentry->d_inode) {
Al Viro6c51e512016-03-05 20:09:32 -05003146 /* Cached positive dentry: will open in f_op->open */
Miklos Szeredid18e9002012-06-05 15:10:17 +02003147 goto out_no_open;
Al Viro6c51e512016-03-05 20:09:32 -05003148 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02003149
Al Viro1643b432016-04-27 19:14:10 -04003150 /*
3151 * Checking write permission is tricky, bacuse we don't know if we are
3152 * going to actually need it: O_CREAT opens should work as long as the
3153 * file exists. But checking existence breaks atomicity. The trick is
3154 * to check access and if not granted clear O_CREAT from the flags.
3155 *
3156 * Another problem is returing the "right" error value (e.g. for an
3157 * O_EXCL open we want to return EEXIST not EROFS).
3158 */
3159 if (open_flag & O_CREAT) {
3160 if (!IS_POSIXACL(dir->d_inode))
3161 mode &= ~current_umask();
3162 if (unlikely(!got_write)) {
3163 create_error = -EROFS;
3164 open_flag &= ~O_CREAT;
3165 if (open_flag & (O_EXCL | O_TRUNC))
3166 goto no_open;
3167 /* No side effects, safe to clear O_CREAT */
3168 } else {
3169 create_error = may_o_create(&nd->path, dentry, mode);
3170 if (create_error) {
3171 open_flag &= ~O_CREAT;
3172 if (open_flag & O_EXCL)
3173 goto no_open;
3174 }
3175 }
3176 } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3177 unlikely(!got_write)) {
3178 /*
3179 * No O_CREATE -> atomicity not a requirement -> fall
3180 * back to lookup + open
3181 */
3182 goto no_open;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003183 }
3184
Al Viro1643b432016-04-27 19:14:10 -04003185 if (dir_inode->i_op->atomic_open) {
3186 error = atomic_open(nd, dentry, path, file, op, open_flag,
3187 mode, opened);
3188 if (unlikely(error == -ENOENT) && create_error)
3189 error = create_error;
3190 return error;
3191 }
Miklos Szeredi54ef4872012-06-05 15:10:16 +02003192
Al Viro1643b432016-04-27 19:14:10 -04003193no_open:
Al Viro6fbd0712016-04-28 11:50:59 -04003194 if (d_in_lookup(dentry)) {
Al Viro12fa5e22016-04-28 11:19:43 -04003195 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3196 nd->flags);
Al Viro6fbd0712016-04-28 11:50:59 -04003197 d_lookup_done(dentry);
Al Viro12fa5e22016-04-28 11:19:43 -04003198 if (unlikely(res)) {
3199 if (IS_ERR(res)) {
3200 error = PTR_ERR(res);
3201 goto out_dput;
3202 }
3203 dput(dentry);
3204 dentry = res;
3205 }
Miklos Szeredi54ef4872012-06-05 15:10:16 +02003206 }
3207
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003208 /* Negative dentry, just create the file */
Al Viro1643b432016-04-27 19:14:10 -04003209 if (!dentry->d_inode && (open_flag & O_CREAT)) {
Al Viro47237682012-06-10 05:01:45 -04003210 *opened |= FILE_CREATED;
Al Viroce8644f2016-04-26 14:17:56 -04003211 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
Al Viroce8644f2016-04-26 14:17:56 -04003212 if (!dir_inode->i_op->create) {
3213 error = -EACCES;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003214 goto out_dput;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003215 }
Al Viroce8644f2016-04-26 14:17:56 -04003216 error = dir_inode->i_op->create(dir_inode, dentry, mode,
Al Viro1643b432016-04-27 19:14:10 -04003217 open_flag & O_EXCL);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003218 if (error)
3219 goto out_dput;
Al Viroce8644f2016-04-26 14:17:56 -04003220 fsnotify_create(dir_inode, dentry);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003221 }
Al Viro1643b432016-04-27 19:14:10 -04003222 if (unlikely(create_error) && !dentry->d_inode) {
3223 error = create_error;
3224 goto out_dput;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003225 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02003226out_no_open:
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003227 path->dentry = dentry;
3228 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04003229 return 1;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003230
3231out_dput:
3232 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04003233 return error;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003234}
3235
3236/*
Al Virofe2d35f2011-03-05 22:58:25 -05003237 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11003238 */
Al Viro896475d2015-04-22 18:02:17 -04003239static int do_last(struct nameidata *nd,
Al Viro2675a4e2012-06-22 12:41:10 +04003240 struct file *file, const struct open_flags *op,
Al Viro76ae2a52015-05-12 18:44:32 -04003241 int *opened)
Al Virofb1cc552009-12-24 01:58:28 -05003242{
Al Viroa1e28032009-12-24 02:12:06 -05003243 struct dentry *dir = nd->path.dentry;
Al Viroca344a892011-03-09 00:36:45 -05003244 int open_flag = op->open_flag;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003245 bool will_truncate = (open_flag & O_TRUNC) != 0;
Al Viro64894cf2012-07-31 00:53:35 +04003246 bool got_write = false;
Al Virobcda7652011-03-13 16:42:14 -04003247 int acc_mode = op->acc_mode;
Al Viro254cf582015-05-05 09:40:46 -04003248 unsigned seq;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02003249 struct inode *inode;
Al Viro896475d2015-04-22 18:02:17 -04003250 struct path path;
Al Viro16c2cd72011-02-22 15:50:10 -05003251 int error;
Al Virofb1cc552009-12-24 01:58:28 -05003252
Al Viroc3e380b2011-02-23 13:39:45 -05003253 nd->flags &= ~LOOKUP_PARENT;
3254 nd->flags |= op->intent;
3255
Al Virobc77daa2013-06-06 09:12:33 -04003256 if (nd->last_type != LAST_NORM) {
Al Virofe2d35f2011-03-05 22:58:25 -05003257 error = handle_dots(nd, nd->last_type);
Al Virodeb106c2015-05-08 18:05:21 -04003258 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04003259 return error;
Miklos Szeredie83db162012-06-05 15:10:29 +02003260 goto finish_open;
Al Viro1f36f772009-12-26 10:56:19 -05003261 }
Al Viro67ee3ad2009-12-26 07:01:01 -05003262
Al Viroca344a892011-03-09 00:36:45 -05003263 if (!(open_flag & O_CREAT)) {
Al Virofe2d35f2011-03-05 22:58:25 -05003264 if (nd->last.name[nd->last.len])
3265 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3266 /* we _can_ be in RCU mode here */
Al Viro254cf582015-05-05 09:40:46 -04003267 error = lookup_fast(nd, &path, &inode, &seq);
Al Viroe9742b52016-03-05 22:04:59 -05003268 if (likely(error > 0))
Miklos Szeredi71574862012-06-05 15:10:14 +02003269 goto finish_lookup;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02003270
Miklos Szeredi71574862012-06-05 15:10:14 +02003271 if (error < 0)
Al Virodeb106c2015-05-08 18:05:21 -04003272 return error;
Miklos Szeredi37d7fff2012-06-05 15:10:12 +02003273
Miklos Szeredi71574862012-06-05 15:10:14 +02003274 BUG_ON(nd->inode != dir->d_inode);
Al Viro6583fe22016-03-05 18:14:03 -05003275 BUG_ON(nd->flags & LOOKUP_RCU);
Miklos Szeredib6183df2012-06-05 15:10:13 +02003276 } else {
3277 /* create side of things */
3278 /*
3279 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3280 * has been cleared when we got to the last component we are
3281 * about to look up
3282 */
3283 error = complete_walk(nd);
Al Viroe8bb73d2015-05-08 16:28:42 -04003284 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003285 return error;
Miklos Szeredib6183df2012-06-05 15:10:13 +02003286
Al Viro76ae2a52015-05-12 18:44:32 -04003287 audit_inode(nd->name, dir, LOOKUP_PARENT);
Miklos Szeredib6183df2012-06-05 15:10:13 +02003288 /* trailing slashes? */
Al Virodeb106c2015-05-08 18:05:21 -04003289 if (unlikely(nd->last.name[nd->last.len]))
3290 return -EISDIR;
Al Virofe2d35f2011-03-05 22:58:25 -05003291 }
3292
Al Viro9cf843e2016-04-28 19:35:16 -04003293 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
Al Viro64894cf2012-07-31 00:53:35 +04003294 error = mnt_want_write(nd->path.mnt);
3295 if (!error)
3296 got_write = true;
3297 /*
3298 * do _not_ fail yet - we might not need that or fail with
3299 * a different error; let lookup_open() decide; we'll be
3300 * dropping this one anyway.
3301 */
3302 }
Al Viro9cf843e2016-04-28 19:35:16 -04003303 if (open_flag & O_CREAT)
3304 inode_lock(dir->d_inode);
3305 else
3306 inode_lock_shared(dir->d_inode);
Al Viro896475d2015-04-22 18:02:17 -04003307 error = lookup_open(nd, &path, file, op, got_write, opened);
Al Viro9cf843e2016-04-28 19:35:16 -04003308 if (open_flag & O_CREAT)
3309 inode_unlock(dir->d_inode);
3310 else
3311 inode_unlock_shared(dir->d_inode);
Al Viroa1e28032009-12-24 02:12:06 -05003312
Al Viro2675a4e2012-06-22 12:41:10 +04003313 if (error <= 0) {
3314 if (error)
Miklos Szeredid18e9002012-06-05 15:10:17 +02003315 goto out;
3316
Al Viro47237682012-06-10 05:01:45 -04003317 if ((*opened & FILE_CREATED) ||
Al Viro496ad9a2013-01-23 17:07:38 -05003318 !S_ISREG(file_inode(file)->i_mode))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003319 will_truncate = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003320
Al Viro76ae2a52015-05-12 18:44:32 -04003321 audit_inode(nd->name, file->f_path.dentry, 0);
Miklos Szeredid18e9002012-06-05 15:10:17 +02003322 goto opened;
3323 }
Al Virofb1cc552009-12-24 01:58:28 -05003324
Al Viro47237682012-06-10 05:01:45 -04003325 if (*opened & FILE_CREATED) {
Al Viro9b44f1b2011-03-09 00:17:27 -05003326 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05003327 open_flag &= ~O_TRUNC;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003328 will_truncate = false;
Al Viro62fb4a12015-12-26 22:33:24 -05003329 acc_mode = 0;
Al Viro896475d2015-04-22 18:02:17 -04003330 path_to_nameidata(&path, nd);
Miklos Szeredie83db162012-06-05 15:10:29 +02003331 goto finish_open_created;
Al Virofb1cc552009-12-24 01:58:28 -05003332 }
3333
3334 /*
Miklos Szeredid18e9002012-06-05 15:10:17 +02003335 * If atomic_open() acquired write access it is dropped now due to
3336 * possible mount and symlink following (this might be optimized away if
3337 * necessary...)
3338 */
Al Viro64894cf2012-07-31 00:53:35 +04003339 if (got_write) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02003340 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04003341 got_write = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003342 }
3343
Al Viroe6ec03a2016-06-05 00:23:09 -04003344 error = follow_managed(&path, nd);
3345 if (unlikely(error < 0))
3346 return error;
3347
Al Viro6583fe22016-03-05 18:14:03 -05003348 if (unlikely(d_is_negative(path.dentry))) {
3349 path_to_nameidata(&path, nd);
3350 return -ENOENT;
3351 }
3352
3353 /*
3354 * create/update audit record if it already exists.
3355 */
3356 audit_inode(nd->name, path.dentry, 0);
3357
Al Virodeb106c2015-05-08 18:05:21 -04003358 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3359 path_to_nameidata(&path, nd);
3360 return -EEXIST;
3361 }
Al Virofb1cc552009-12-24 01:58:28 -05003362
Al Viro254cf582015-05-05 09:40:46 -04003363 seq = 0; /* out of RCU mode, so the value doesn't matter */
Al Virod4565642016-02-27 19:23:16 -05003364 inode = d_backing_inode(path.dentry);
Al Viro766c4cb2015-05-07 19:24:57 -04003365finish_lookup:
Al Virod63ff282015-05-04 18:13:23 -04003366 if (nd->depth)
3367 put_link(nd);
Al Viro181548c2015-05-07 19:54:34 -04003368 error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW,
3369 inode, seq);
Al Virodeb106c2015-05-08 18:05:21 -04003370 if (unlikely(error))
Al Virod63ff282015-05-04 18:13:23 -04003371 return error;
Al Virofb1cc552009-12-24 01:58:28 -05003372
Al Virofac7d192016-06-04 11:41:49 -04003373 path_to_nameidata(&path, nd);
Miklos Szeredidecf3402012-05-21 17:30:08 +02003374 nd->inode = inode;
Al Viro254cf582015-05-05 09:40:46 -04003375 nd->seq = seq;
Al Viroa3fbbde2011-11-07 21:21:26 +00003376 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
Al Virobc77daa2013-06-06 09:12:33 -04003377finish_open:
Al Viroa3fbbde2011-11-07 21:21:26 +00003378 error = complete_walk(nd);
Al Virofac7d192016-06-04 11:41:49 -04003379 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003380 return error;
Al Viro76ae2a52015-05-12 18:44:32 -04003381 audit_inode(nd->name, nd->path.dentry, 0);
Al Virofb1cc552009-12-24 01:58:28 -05003382 error = -EISDIR;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003383 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04003384 goto out;
Miklos Szerediaf2f5542012-05-21 17:30:11 +02003385 error = -ENOTDIR;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003386 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04003387 goto out;
David Howells4bbcbd32015-03-17 22:16:40 +00003388 if (!d_is_reg(nd->path.dentry))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003389 will_truncate = false;
Al Viro6c0d46c2011-03-09 00:59:59 -05003390
Al Viro0f9d1a12011-03-09 00:13:14 -05003391 if (will_truncate) {
3392 error = mnt_want_write(nd->path.mnt);
3393 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003394 goto out;
Al Viro64894cf2012-07-31 00:53:35 +04003395 got_write = true;
Al Viro0f9d1a12011-03-09 00:13:14 -05003396 }
Miklos Szeredie83db162012-06-05 15:10:29 +02003397finish_open_created:
Al Viro6ac08702016-04-26 00:02:50 -04003398 error = may_open(&nd->path, acc_mode, open_flag);
3399 if (error)
3400 goto out;
Miklos Szeredi4aa7c632014-10-24 00:14:35 +02003401 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3402 error = vfs_open(&nd->path, file, current_cred());
Al Virofac7d192016-06-04 11:41:49 -04003403 if (error)
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003404 goto out;
Al Virofac7d192016-06-04 11:41:49 -04003405 *opened |= FILE_OPENED;
Miklos Szeredia8277b92012-06-05 15:10:32 +02003406opened:
Al Viro2675a4e2012-06-22 12:41:10 +04003407 error = open_check_o_direct(file);
Al Virofe9ec822016-04-27 03:14:20 -04003408 if (!error)
3409 error = ima_file_check(file, op->acc_mode, *opened);
3410 if (!error && will_truncate)
Al Viro2675a4e2012-06-22 12:41:10 +04003411 error = handle_truncate(file);
Al Viroca344a892011-03-09 00:36:45 -05003412out:
Al Virofe9ec822016-04-27 03:14:20 -04003413 if (unlikely(error) && (*opened & FILE_OPENED))
3414 fput(file);
Al Viroc80567c2016-02-27 19:17:33 -05003415 if (unlikely(error > 0)) {
3416 WARN_ON(1);
3417 error = -EINVAL;
3418 }
Al Viro64894cf2012-07-31 00:53:35 +04003419 if (got_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05003420 mnt_drop_write(nd->path.mnt);
Al Viro2675a4e2012-06-22 12:41:10 +04003421 return error;
Al Virofb1cc552009-12-24 01:58:28 -05003422}
3423
Al Viroc8a53ee2015-05-12 18:43:07 -04003424static int do_tmpfile(struct nameidata *nd, unsigned flags,
Al Viro60545d02013-06-07 01:20:27 -04003425 const struct open_flags *op,
3426 struct file *file, int *opened)
3427{
3428 static const struct qstr name = QSTR_INIT("/", 1);
Al Viro625b6d12015-05-12 16:36:12 -04003429 struct dentry *child;
Al Viro60545d02013-06-07 01:20:27 -04003430 struct inode *dir;
Al Viro625b6d12015-05-12 16:36:12 -04003431 struct path path;
Al Viroc8a53ee2015-05-12 18:43:07 -04003432 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
Al Viro60545d02013-06-07 01:20:27 -04003433 if (unlikely(error))
3434 return error;
Al Viro625b6d12015-05-12 16:36:12 -04003435 error = mnt_want_write(path.mnt);
Al Viro60545d02013-06-07 01:20:27 -04003436 if (unlikely(error))
3437 goto out;
Al Viro625b6d12015-05-12 16:36:12 -04003438 dir = path.dentry->d_inode;
Al Viro60545d02013-06-07 01:20:27 -04003439 /* we want directory to be writable */
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07003440 error = inode_permission2(nd->path.mnt, dir, MAY_WRITE | MAY_EXEC);
Al Viro60545d02013-06-07 01:20:27 -04003441 if (error)
3442 goto out2;
Al Viro60545d02013-06-07 01:20:27 -04003443 if (!dir->i_op->tmpfile) {
3444 error = -EOPNOTSUPP;
3445 goto out2;
3446 }
Al Viro625b6d12015-05-12 16:36:12 -04003447 child = d_alloc(path.dentry, &name);
Al Viro60545d02013-06-07 01:20:27 -04003448 if (unlikely(!child)) {
3449 error = -ENOMEM;
3450 goto out2;
3451 }
Al Viro625b6d12015-05-12 16:36:12 -04003452 dput(path.dentry);
3453 path.dentry = child;
3454 error = dir->i_op->tmpfile(dir, child, op->mode);
Al Viro60545d02013-06-07 01:20:27 -04003455 if (error)
3456 goto out2;
Al Viroc8a53ee2015-05-12 18:43:07 -04003457 audit_inode(nd->name, child, 0);
Eric Rannaud69a91c22014-10-30 01:51:01 -07003458 /* Don't check for other permissions, the inode was just created */
Al Viro62fb4a12015-12-26 22:33:24 -05003459 error = may_open(&path, 0, op->open_flag);
Al Viro60545d02013-06-07 01:20:27 -04003460 if (error)
3461 goto out2;
Al Viro625b6d12015-05-12 16:36:12 -04003462 file->f_path.mnt = path.mnt;
3463 error = finish_open(file, child, NULL, opened);
Al Viro60545d02013-06-07 01:20:27 -04003464 if (error)
3465 goto out2;
3466 error = open_check_o_direct(file);
Al Virof4e0c302013-06-11 08:34:36 +04003467 if (error) {
Al Viro60545d02013-06-07 01:20:27 -04003468 fput(file);
Al Virof4e0c302013-06-11 08:34:36 +04003469 } else if (!(op->open_flag & O_EXCL)) {
3470 struct inode *inode = file_inode(file);
3471 spin_lock(&inode->i_lock);
3472 inode->i_state |= I_LINKABLE;
3473 spin_unlock(&inode->i_lock);
3474 }
Al Viro60545d02013-06-07 01:20:27 -04003475out2:
Al Viro625b6d12015-05-12 16:36:12 -04003476 mnt_drop_write(path.mnt);
Al Viro60545d02013-06-07 01:20:27 -04003477out:
Al Viro625b6d12015-05-12 16:36:12 -04003478 path_put(&path);
Al Viro60545d02013-06-07 01:20:27 -04003479 return error;
3480}
3481
Al Viro6ac08702016-04-26 00:02:50 -04003482static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3483{
3484 struct path path;
3485 int error = path_lookupat(nd, flags, &path);
3486 if (!error) {
3487 audit_inode(nd->name, path.dentry, 0);
3488 error = vfs_open(&path, file, current_cred());
3489 path_put(&path);
3490 }
3491 return error;
3492}
3493
Al Viroc8a53ee2015-05-12 18:43:07 -04003494static struct file *path_openat(struct nameidata *nd,
3495 const struct open_flags *op, unsigned flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003496{
Al Viro368ee9b2015-05-08 17:19:59 -04003497 const char *s;
Al Viro30d90492012-06-22 12:40:19 +04003498 struct file *file;
Al Viro47237682012-06-10 05:01:45 -04003499 int opened = 0;
Al Viro13aab422011-02-23 17:54:08 -05003500 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11003501
Al Viro30d90492012-06-22 12:40:19 +04003502 file = get_empty_filp();
Al Viro1afc99b2013-02-14 20:41:04 -05003503 if (IS_ERR(file))
3504 return file;
Nick Piggin31e6b012011-01-07 17:49:52 +11003505
Al Viro30d90492012-06-22 12:40:19 +04003506 file->f_flags = op->open_flag;
Nick Piggin31e6b012011-01-07 17:49:52 +11003507
Al Virobb458c62013-07-13 13:26:37 +04003508 if (unlikely(file->f_flags & __O_TMPFILE)) {
Al Viroc8a53ee2015-05-12 18:43:07 -04003509 error = do_tmpfile(nd, flags, op, file, &opened);
Al Virof15133d2015-05-08 22:53:15 -04003510 goto out2;
Al Viro60545d02013-06-07 01:20:27 -04003511 }
3512
Al Viro6ac08702016-04-26 00:02:50 -04003513 if (unlikely(file->f_flags & O_PATH)) {
3514 error = do_o_path(nd, flags, file);
3515 if (!error)
3516 opened |= FILE_OPENED;
3517 goto out2;
3518 }
3519
Al Viroc8a53ee2015-05-12 18:43:07 -04003520 s = path_init(nd, flags);
Al Viro368ee9b2015-05-08 17:19:59 -04003521 if (IS_ERR(s)) {
3522 put_filp(file);
3523 return ERR_CAST(s);
3524 }
Al Viro3bdba282015-05-08 17:37:07 -04003525 while (!(error = link_path_walk(s, nd)) &&
Al Viro76ae2a52015-05-12 18:44:32 -04003526 (error = do_last(nd, file, op, &opened)) > 0) {
Al Viro73d049a2011-03-11 12:08:24 -05003527 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
Al Viro3bdba282015-05-08 17:37:07 -04003528 s = trailing_symlink(nd);
3529 if (IS_ERR(s)) {
3530 error = PTR_ERR(s);
Al Viro2675a4e2012-06-22 12:41:10 +04003531 break;
Al Viro3bdba282015-05-08 17:37:07 -04003532 }
Al Viro806b6812009-12-26 07:16:40 -05003533 }
Al Virodeb106c2015-05-08 18:05:21 -04003534 terminate_walk(nd);
Al Virof15133d2015-05-08 22:53:15 -04003535out2:
Al Viro2675a4e2012-06-22 12:41:10 +04003536 if (!(opened & FILE_OPENED)) {
3537 BUG_ON(!error);
Al Viro30d90492012-06-22 12:40:19 +04003538 put_filp(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003539 }
Al Viro2675a4e2012-06-22 12:41:10 +04003540 if (unlikely(error)) {
3541 if (error == -EOPENSTALE) {
3542 if (flags & LOOKUP_RCU)
3543 error = -ECHILD;
3544 else
3545 error = -ESTALE;
3546 }
3547 file = ERR_PTR(error);
3548 }
3549 return file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003550}
3551
Jeff Layton669abf42012-10-10 16:43:10 -04003552struct file *do_filp_open(int dfd, struct filename *pathname,
Al Virof9652e12013-06-11 08:23:01 +04003553 const struct open_flags *op)
Al Viro13aab422011-02-23 17:54:08 -05003554{
Al Viro9883d182015-05-13 07:28:08 -04003555 struct nameidata nd;
Al Virof9652e12013-06-11 08:23:01 +04003556 int flags = op->lookup_flags;
Al Viro13aab422011-02-23 17:54:08 -05003557 struct file *filp;
3558
Al Viro9883d182015-05-13 07:28:08 -04003559 set_nameidata(&nd, dfd, pathname);
Al Viroc8a53ee2015-05-12 18:43:07 -04003560 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
Al Viro13aab422011-02-23 17:54:08 -05003561 if (unlikely(filp == ERR_PTR(-ECHILD)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003562 filp = path_openat(&nd, op, flags);
Al Viro13aab422011-02-23 17:54:08 -05003563 if (unlikely(filp == ERR_PTR(-ESTALE)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003564 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
Al Viro9883d182015-05-13 07:28:08 -04003565 restore_nameidata();
Al Viro13aab422011-02-23 17:54:08 -05003566 return filp;
3567}
3568
Al Viro73d049a2011-03-11 12:08:24 -05003569struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
Al Virof9652e12013-06-11 08:23:01 +04003570 const char *name, const struct open_flags *op)
Al Viro73d049a2011-03-11 12:08:24 -05003571{
Al Viro9883d182015-05-13 07:28:08 -04003572 struct nameidata nd;
Al Viro73d049a2011-03-11 12:08:24 -05003573 struct file *file;
Paul Moore51689102015-01-22 00:00:03 -05003574 struct filename *filename;
Al Virof9652e12013-06-11 08:23:01 +04003575 int flags = op->lookup_flags | LOOKUP_ROOT;
Al Viro73d049a2011-03-11 12:08:24 -05003576
3577 nd.root.mnt = mnt;
3578 nd.root.dentry = dentry;
3579
David Howellsb18825a2013-09-12 19:22:53 +01003580 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
Al Viro73d049a2011-03-11 12:08:24 -05003581 return ERR_PTR(-ELOOP);
3582
Paul Moore51689102015-01-22 00:00:03 -05003583 filename = getname_kernel(name);
Viresh Kumara1c83682015-08-12 15:59:44 +05303584 if (IS_ERR(filename))
Paul Moore51689102015-01-22 00:00:03 -05003585 return ERR_CAST(filename);
3586
Al Viro9883d182015-05-13 07:28:08 -04003587 set_nameidata(&nd, -1, filename);
Al Viroc8a53ee2015-05-12 18:43:07 -04003588 file = path_openat(&nd, op, flags | LOOKUP_RCU);
Al Viro73d049a2011-03-11 12:08:24 -05003589 if (unlikely(file == ERR_PTR(-ECHILD)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003590 file = path_openat(&nd, op, flags);
Al Viro73d049a2011-03-11 12:08:24 -05003591 if (unlikely(file == ERR_PTR(-ESTALE)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003592 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
Al Viro9883d182015-05-13 07:28:08 -04003593 restore_nameidata();
Paul Moore51689102015-01-22 00:00:03 -05003594 putname(filename);
Al Viro73d049a2011-03-11 12:08:24 -05003595 return file;
3596}
3597
Al Virofa14a0b2015-01-22 02:16:49 -05003598static struct dentry *filename_create(int dfd, struct filename *name,
Jeff Layton1ac12b42012-12-11 12:10:06 -05003599 struct path *path, unsigned int lookup_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003600{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003601 struct dentry *dentry = ERR_PTR(-EEXIST);
Al Viro391172c2015-05-09 11:19:16 -04003602 struct qstr last;
3603 int type;
Jan Karac30dabf2012-06-12 16:20:30 +02003604 int err2;
Jeff Layton1ac12b42012-12-11 12:10:06 -05003605 int error;
3606 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3607
3608 /*
3609 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3610 * other flags passed in are ignored!
3611 */
3612 lookup_flags &= LOOKUP_REVAL;
3613
Al Viro5c31b6c2015-05-12 17:32:54 -04003614 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3615 if (IS_ERR(name))
3616 return ERR_CAST(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003617
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003618 /*
3619 * Yucky last component or no last component at all?
3620 * (foo/., foo/.., /////)
3621 */
Al Viro5c31b6c2015-05-12 17:32:54 -04003622 if (unlikely(type != LAST_NORM))
Al Viroed75e952011-06-27 16:53:43 -04003623 goto out;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003624
Jan Karac30dabf2012-06-12 16:20:30 +02003625 /* don't fail immediately if it's r/o, at least try to report other errors */
Al Viro391172c2015-05-09 11:19:16 -04003626 err2 = mnt_want_write(path->mnt);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003627 /*
3628 * Do the final lookup.
3629 */
Al Viro391172c2015-05-09 11:19:16 -04003630 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
Al Viro59551022016-01-22 15:40:57 -05003631 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
Al Viro391172c2015-05-09 11:19:16 -04003632 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003633 if (IS_ERR(dentry))
Al Viroa8104a92012-07-20 02:25:00 +04003634 goto unlock;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003635
Al Viroa8104a92012-07-20 02:25:00 +04003636 error = -EEXIST;
David Howellsb18825a2013-09-12 19:22:53 +01003637 if (d_is_positive(dentry))
Al Viroa8104a92012-07-20 02:25:00 +04003638 goto fail;
David Howellsb18825a2013-09-12 19:22:53 +01003639
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003640 /*
3641 * Special case - lookup gave negative, but... we had foo/bar/
3642 * From the vfs_mknod() POV we just have a negative dentry -
3643 * all is fine. Let's be bastards - you had / on the end, you've
3644 * been asking for (non-existent) directory. -ENOENT for you.
3645 */
Al Viro391172c2015-05-09 11:19:16 -04003646 if (unlikely(!is_dir && last.name[last.len])) {
Al Viroa8104a92012-07-20 02:25:00 +04003647 error = -ENOENT;
Al Viroed75e952011-06-27 16:53:43 -04003648 goto fail;
Al Viroe9baf6e2008-05-15 04:49:12 -04003649 }
Jan Karac30dabf2012-06-12 16:20:30 +02003650 if (unlikely(err2)) {
3651 error = err2;
Al Viroa8104a92012-07-20 02:25:00 +04003652 goto fail;
Jan Karac30dabf2012-06-12 16:20:30 +02003653 }
Al Viro181c37b2015-05-12 17:21:25 -04003654 putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003655 return dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003656fail:
Al Viroa8104a92012-07-20 02:25:00 +04003657 dput(dentry);
3658 dentry = ERR_PTR(error);
3659unlock:
Al Viro59551022016-01-22 15:40:57 -05003660 inode_unlock(path->dentry->d_inode);
Jan Karac30dabf2012-06-12 16:20:30 +02003661 if (!err2)
Al Viro391172c2015-05-09 11:19:16 -04003662 mnt_drop_write(path->mnt);
Al Viroed75e952011-06-27 16:53:43 -04003663out:
Al Viro391172c2015-05-09 11:19:16 -04003664 path_put(path);
Al Viro181c37b2015-05-12 17:21:25 -04003665 putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003666 return dentry;
3667}
Al Virofa14a0b2015-01-22 02:16:49 -05003668
3669struct dentry *kern_path_create(int dfd, const char *pathname,
3670 struct path *path, unsigned int lookup_flags)
3671{
Al Viro181c37b2015-05-12 17:21:25 -04003672 return filename_create(dfd, getname_kernel(pathname),
3673 path, lookup_flags);
Al Virofa14a0b2015-01-22 02:16:49 -05003674}
Al Virodae6ad82011-06-26 11:50:15 -04003675EXPORT_SYMBOL(kern_path_create);
3676
Al Viro921a1652012-07-20 01:15:31 +04003677void done_path_create(struct path *path, struct dentry *dentry)
3678{
3679 dput(dentry);
Al Viro59551022016-01-22 15:40:57 -05003680 inode_unlock(path->dentry->d_inode);
Al Viroa8104a92012-07-20 02:25:00 +04003681 mnt_drop_write(path->mnt);
Al Viro921a1652012-07-20 01:15:31 +04003682 path_put(path);
3683}
3684EXPORT_SYMBOL(done_path_create);
3685
Al Viro520ae682015-05-13 07:00:28 -04003686inline struct dentry *user_path_create(int dfd, const char __user *pathname,
Jeff Layton1ac12b42012-12-11 12:10:06 -05003687 struct path *path, unsigned int lookup_flags)
Al Virodae6ad82011-06-26 11:50:15 -04003688{
Al Viro181c37b2015-05-12 17:21:25 -04003689 return filename_create(dfd, getname(pathname), path, lookup_flags);
Al Virodae6ad82011-06-26 11:50:15 -04003690}
3691EXPORT_SYMBOL(user_path_create);
3692
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07003693int vfs_mknod2(struct vfsmount *mnt, struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003694{
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07003695 int error = may_create(mnt, dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003696
3697 if (error)
3698 return error;
3699
Eric W. Biederman975d6b32011-11-13 12:16:43 -08003700 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003701 return -EPERM;
3702
Al Viroacfa4382008-12-04 10:06:33 -05003703 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003704 return -EPERM;
3705
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07003706 error = devcgroup_inode_mknod(mode, dev);
3707 if (error)
3708 return error;
3709
Linus Torvalds1da177e2005-04-16 15:20:36 -07003710 error = security_inode_mknod(dir, dentry, mode, dev);
3711 if (error)
3712 return error;
3713
Linus Torvalds1da177e2005-04-16 15:20:36 -07003714 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003715 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003716 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003717 return error;
3718}
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07003719EXPORT_SYMBOL(vfs_mknod2);
3720
3721int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3722{
3723 return vfs_mknod2(NULL, dir, dentry, mode, dev);
3724}
Al Viro4d359502014-03-14 12:20:17 -04003725EXPORT_SYMBOL(vfs_mknod);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003726
Al Virof69aac02011-07-26 04:31:40 -04003727static int may_mknod(umode_t mode)
Dave Hansen463c3192008-02-15 14:37:57 -08003728{
3729 switch (mode & S_IFMT) {
3730 case S_IFREG:
3731 case S_IFCHR:
3732 case S_IFBLK:
3733 case S_IFIFO:
3734 case S_IFSOCK:
3735 case 0: /* zero mode translates to S_IFREG */
3736 return 0;
3737 case S_IFDIR:
3738 return -EPERM;
3739 default:
3740 return -EINVAL;
3741 }
3742}
3743
Al Viro8208a222011-07-25 17:32:17 -04003744SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003745 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003746{
Al Viro2ad94ae2008-07-21 09:32:51 -04003747 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003748 struct path path;
3749 int error;
Jeff Layton972567f2012-12-20 16:00:10 -05003750 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003751
Al Viro8e4bfca2012-07-20 01:17:26 +04003752 error = may_mknod(mode);
3753 if (error)
3754 return error;
Jeff Layton972567f2012-12-20 16:00:10 -05003755retry:
3756 dentry = user_path_create(dfd, filename, &path, lookup_flags);
Al Virodae6ad82011-06-26 11:50:15 -04003757 if (IS_ERR(dentry))
3758 return PTR_ERR(dentry);
Al Viro2ad94ae2008-07-21 09:32:51 -04003759
Al Virodae6ad82011-06-26 11:50:15 -04003760 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003761 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003762 error = security_path_mknod(&path, dentry, mode, dev);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003763 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04003764 goto out;
Dave Hansen463c3192008-02-15 14:37:57 -08003765 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003766 case 0: case S_IFREG:
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07003767 error = vfs_create2(path.mnt, path.dentry->d_inode,dentry,mode,true);
Mimi Zohar05d1a712016-02-29 19:52:05 -05003768 if (!error)
3769 ima_post_path_mknod(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003770 break;
3771 case S_IFCHR: case S_IFBLK:
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07003772 error = vfs_mknod2(path.mnt, path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003773 new_decode_dev(dev));
3774 break;
3775 case S_IFIFO: case S_IFSOCK:
Al Virodae6ad82011-06-26 11:50:15 -04003776 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003777 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003778 }
Al Viroa8104a92012-07-20 02:25:00 +04003779out:
Al Viro921a1652012-07-20 01:15:31 +04003780 done_path_create(&path, dentry);
Jeff Layton972567f2012-12-20 16:00:10 -05003781 if (retry_estale(error, lookup_flags)) {
3782 lookup_flags |= LOOKUP_REVAL;
3783 goto retry;
3784 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003785 return error;
3786}
3787
Al Viro8208a222011-07-25 17:32:17 -04003788SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003789{
3790 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3791}
3792
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07003793int vfs_mkdir2(struct vfsmount *mnt, struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003794{
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07003795 int error = may_create(mnt, dir, dentry);
Al Viro8de52772012-02-06 12:45:27 -05003796 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003797
3798 if (error)
3799 return error;
3800
Al Viroacfa4382008-12-04 10:06:33 -05003801 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003802 return -EPERM;
3803
3804 mode &= (S_IRWXUGO|S_ISVTX);
3805 error = security_inode_mkdir(dir, dentry, mode);
3806 if (error)
3807 return error;
3808
Al Viro8de52772012-02-06 12:45:27 -05003809 if (max_links && dir->i_nlink >= max_links)
3810 return -EMLINK;
3811
Linus Torvalds1da177e2005-04-16 15:20:36 -07003812 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003813 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003814 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003815 return error;
3816}
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07003817EXPORT_SYMBOL(vfs_mkdir2);
3818
3819int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3820{
3821 return vfs_mkdir2(NULL, dir, dentry, mode);
3822}
Al Viro4d359502014-03-14 12:20:17 -04003823EXPORT_SYMBOL(vfs_mkdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003824
Al Viroa218d0f2011-11-21 14:59:34 -05003825SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003826{
Dave Hansen6902d922006-09-30 23:29:01 -07003827 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003828 struct path path;
3829 int error;
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003830 unsigned int lookup_flags = LOOKUP_DIRECTORY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003831
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003832retry:
3833 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07003834 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003835 return PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003836
Al Virodae6ad82011-06-26 11:50:15 -04003837 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003838 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003839 error = security_path_mkdir(&path, dentry, mode);
Al Viroa8104a92012-07-20 02:25:00 +04003840 if (!error)
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07003841 error = vfs_mkdir2(path.mnt, path.dentry->d_inode, dentry, mode);
Al Viro921a1652012-07-20 01:15:31 +04003842 done_path_create(&path, dentry);
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003843 if (retry_estale(error, lookup_flags)) {
3844 lookup_flags |= LOOKUP_REVAL;
3845 goto retry;
3846 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003847 return error;
3848}
3849
Al Viroa218d0f2011-11-21 14:59:34 -05003850SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003851{
3852 return sys_mkdirat(AT_FDCWD, pathname, mode);
3853}
3854
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07003855int vfs_rmdir2(struct vfsmount *mnt, struct inode *dir, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003856{
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07003857 int error = may_delete(mnt, dir, dentry, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003858
3859 if (error)
3860 return error;
3861
Al Viroacfa4382008-12-04 10:06:33 -05003862 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003863 return -EPERM;
3864
Al Viro1d2ef592011-09-14 18:55:41 +01003865 dget(dentry);
Al Viro59551022016-01-22 15:40:57 -05003866 inode_lock(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003867
Sage Weil912dbc12011-05-24 13:06:11 -07003868 error = -EBUSY;
Eric W. Biederman7af13642013-10-04 19:15:13 -07003869 if (is_local_mountpoint(dentry))
Sage Weil912dbc12011-05-24 13:06:11 -07003870 goto out;
3871
3872 error = security_inode_rmdir(dir, dentry);
3873 if (error)
3874 goto out;
3875
Sage Weil3cebde22011-05-29 21:20:59 -07003876 shrink_dcache_parent(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003877 error = dir->i_op->rmdir(dir, dentry);
3878 if (error)
3879 goto out;
3880
3881 dentry->d_inode->i_flags |= S_DEAD;
3882 dont_mount(dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003883 detach_mounts(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003884
3885out:
Al Viro59551022016-01-22 15:40:57 -05003886 inode_unlock(dentry->d_inode);
Al Viro1d2ef592011-09-14 18:55:41 +01003887 dput(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003888 if (!error)
3889 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003890 return error;
3891}
Guenter Roeckf1222162017-01-30 12:29:00 -08003892EXPORT_SYMBOL(vfs_rmdir2);
3893
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07003894int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3895{
3896 return vfs_rmdir2(NULL, dir, dentry);
3897}
Al Viro4d359502014-03-14 12:20:17 -04003898EXPORT_SYMBOL(vfs_rmdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003899
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003900static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003901{
3902 int error = 0;
Jeff Layton91a27b22012-10-10 15:25:28 -04003903 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003904 struct dentry *dentry;
Al Virof5beed72015-04-30 16:09:11 -04003905 struct path path;
3906 struct qstr last;
3907 int type;
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003908 unsigned int lookup_flags = 0;
3909retry:
Al Virof5beed72015-04-30 16:09:11 -04003910 name = user_path_parent(dfd, pathname,
3911 &path, &last, &type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04003912 if (IS_ERR(name))
3913 return PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003914
Al Virof5beed72015-04-30 16:09:11 -04003915 switch (type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003916 case LAST_DOTDOT:
3917 error = -ENOTEMPTY;
3918 goto exit1;
3919 case LAST_DOT:
3920 error = -EINVAL;
3921 goto exit1;
3922 case LAST_ROOT:
3923 error = -EBUSY;
3924 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003925 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003926
Al Virof5beed72015-04-30 16:09:11 -04003927 error = mnt_want_write(path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02003928 if (error)
3929 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003930
Al Viro59551022016-01-22 15:40:57 -05003931 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
Al Virof5beed72015-04-30 16:09:11 -04003932 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003933 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003934 if (IS_ERR(dentry))
3935 goto exit2;
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003936 if (!dentry->d_inode) {
3937 error = -ENOENT;
3938 goto exit3;
3939 }
Al Virof5beed72015-04-30 16:09:11 -04003940 error = security_path_rmdir(&path, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003941 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003942 goto exit3;
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07003943 error = vfs_rmdir2(path.mnt, path.dentry->d_inode, dentry);
Dave Hansen06227532008-02-15 14:37:34 -08003944exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07003945 dput(dentry);
3946exit2:
Al Viro59551022016-01-22 15:40:57 -05003947 inode_unlock(path.dentry->d_inode);
Al Virof5beed72015-04-30 16:09:11 -04003948 mnt_drop_write(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003949exit1:
Al Virof5beed72015-04-30 16:09:11 -04003950 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003951 putname(name);
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003952 if (retry_estale(error, lookup_flags)) {
3953 lookup_flags |= LOOKUP_REVAL;
3954 goto retry;
3955 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003956 return error;
3957}
3958
Heiko Carstens3cdad422009-01-14 14:14:22 +01003959SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003960{
3961 return do_rmdir(AT_FDCWD, pathname);
3962}
3963
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003964/**
3965 * vfs_unlink - unlink a filesystem object
3966 * @dir: parent directory
3967 * @dentry: victim
3968 * @delegated_inode: returns victim inode, if the inode is delegated.
3969 *
3970 * The caller must hold dir->i_mutex.
3971 *
3972 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3973 * return a reference to the inode in delegated_inode. The caller
3974 * should then break the delegation on that inode and retry. Because
3975 * breaking a delegation may take a long time, the caller should drop
3976 * dir->i_mutex before doing so.
3977 *
3978 * Alternatively, a caller may pass NULL for delegated_inode. This may
3979 * be appropriate for callers that expect the underlying filesystem not
3980 * to be NFS exported.
3981 */
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07003982int vfs_unlink2(struct vfsmount *mnt, struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003983{
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003984 struct inode *target = dentry->d_inode;
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07003985 int error = may_delete(mnt, dir, dentry, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003986
3987 if (error)
3988 return error;
3989
Al Viroacfa4382008-12-04 10:06:33 -05003990 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003991 return -EPERM;
3992
Al Viro59551022016-01-22 15:40:57 -05003993 inode_lock(target);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003994 if (is_local_mountpoint(dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003995 error = -EBUSY;
3996 else {
3997 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05003998 if (!error) {
J. Bruce Fields5a146962012-08-28 07:50:40 -07003999 error = try_break_deleg(target, delegated_inode);
4000 if (error)
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04004001 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004002 error = dir->i_op->unlink(dir, dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07004003 if (!error) {
Al Virod83c49f2010-04-30 17:17:09 -04004004 dont_mount(dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07004005 detach_mounts(dentry);
4006 }
Al Virobec10522010-03-03 14:12:08 -05004007 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004008 }
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04004009out:
Al Viro59551022016-01-22 15:40:57 -05004010 inode_unlock(target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004011
4012 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
4013 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
J. Bruce Fields9accbb92012-08-28 07:03:24 -04004014 fsnotify_link_count(target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004015 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004016 }
Robert Love0eeca282005-07-12 17:06:03 -04004017
Linus Torvalds1da177e2005-04-16 15:20:36 -07004018 return error;
4019}
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07004020EXPORT_SYMBOL(vfs_unlink2);
4021
4022int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
4023{
4024 return vfs_unlink2(NULL, dir, dentry, delegated_inode);
4025}
Al Viro4d359502014-03-14 12:20:17 -04004026EXPORT_SYMBOL(vfs_unlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004027
4028/*
4029 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004030 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07004031 * writeout happening, and we don't want to prevent access to the directory
4032 * while waiting on the I/O.
4033 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004034static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004035{
Al Viro2ad94ae2008-07-21 09:32:51 -04004036 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04004037 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004038 struct dentry *dentry;
Al Virof5beed72015-04-30 16:09:11 -04004039 struct path path;
4040 struct qstr last;
4041 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004042 struct inode *inode = NULL;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04004043 struct inode *delegated_inode = NULL;
Jeff Layton5d18f812012-12-20 16:38:04 -05004044 unsigned int lookup_flags = 0;
4045retry:
Al Virof5beed72015-04-30 16:09:11 -04004046 name = user_path_parent(dfd, pathname,
4047 &path, &last, &type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04004048 if (IS_ERR(name))
4049 return PTR_ERR(name);
Al Viro2ad94ae2008-07-21 09:32:51 -04004050
Linus Torvalds1da177e2005-04-16 15:20:36 -07004051 error = -EISDIR;
Al Virof5beed72015-04-30 16:09:11 -04004052 if (type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004053 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09004054
Al Virof5beed72015-04-30 16:09:11 -04004055 error = mnt_want_write(path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02004056 if (error)
4057 goto exit1;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04004058retry_deleg:
Al Viro59551022016-01-22 15:40:57 -05004059 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
Al Virof5beed72015-04-30 16:09:11 -04004060 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004061 error = PTR_ERR(dentry);
4062 if (!IS_ERR(dentry)) {
4063 /* Why not before? Because we want correct error value */
Al Virof5beed72015-04-30 16:09:11 -04004064 if (last.name[last.len])
Török Edwin50338b82011-06-16 00:06:14 +03004065 goto slashes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004066 inode = dentry->d_inode;
David Howellsb18825a2013-09-12 19:22:53 +01004067 if (d_is_negative(dentry))
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04004068 goto slashes;
4069 ihold(inode);
Al Virof5beed72015-04-30 16:09:11 -04004070 error = security_path_unlink(&path, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004071 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02004072 goto exit2;
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07004073 error = vfs_unlink2(path.mnt, path.dentry->d_inode, dentry, &delegated_inode);
Jan Karac30dabf2012-06-12 16:20:30 +02004074exit2:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004075 dput(dentry);
4076 }
Al Viro59551022016-01-22 15:40:57 -05004077 inode_unlock(path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004078 if (inode)
4079 iput(inode); /* truncate the inode here */
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04004080 inode = NULL;
4081 if (delegated_inode) {
J. Bruce Fields5a146962012-08-28 07:50:40 -07004082 error = break_deleg_wait(&delegated_inode);
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04004083 if (!error)
4084 goto retry_deleg;
4085 }
Al Virof5beed72015-04-30 16:09:11 -04004086 mnt_drop_write(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004087exit1:
Al Virof5beed72015-04-30 16:09:11 -04004088 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004089 putname(name);
Jeff Layton5d18f812012-12-20 16:38:04 -05004090 if (retry_estale(error, lookup_flags)) {
4091 lookup_flags |= LOOKUP_REVAL;
4092 inode = NULL;
4093 goto retry;
4094 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004095 return error;
4096
4097slashes:
David Howellsb18825a2013-09-12 19:22:53 +01004098 if (d_is_negative(dentry))
4099 error = -ENOENT;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02004100 else if (d_is_dir(dentry))
David Howellsb18825a2013-09-12 19:22:53 +01004101 error = -EISDIR;
4102 else
4103 error = -ENOTDIR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004104 goto exit2;
4105}
4106
Heiko Carstens2e4d0922009-01-14 14:14:31 +01004107SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004108{
4109 if ((flag & ~AT_REMOVEDIR) != 0)
4110 return -EINVAL;
4111
4112 if (flag & AT_REMOVEDIR)
4113 return do_rmdir(dfd, pathname);
4114
4115 return do_unlinkat(dfd, pathname);
4116}
4117
Heiko Carstens3480b252009-01-14 14:14:16 +01004118SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004119{
4120 return do_unlinkat(AT_FDCWD, pathname);
4121}
4122
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07004123int vfs_symlink2(struct vfsmount *mnt, struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004124{
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07004125 int error = may_create(mnt, dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004126
4127 if (error)
4128 return error;
4129
Al Viroacfa4382008-12-04 10:06:33 -05004130 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004131 return -EPERM;
4132
4133 error = security_inode_symlink(dir, dentry, oldname);
4134 if (error)
4135 return error;
4136
Linus Torvalds1da177e2005-04-16 15:20:36 -07004137 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07004138 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00004139 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004140 return error;
4141}
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07004142EXPORT_SYMBOL(vfs_symlink2);
4143
4144int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
4145{
4146 return vfs_symlink2(NULL, dir, dentry, oldname);
4147}
Al Viro4d359502014-03-14 12:20:17 -04004148EXPORT_SYMBOL(vfs_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004149
Heiko Carstens2e4d0922009-01-14 14:14:31 +01004150SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4151 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004152{
Al Viro2ad94ae2008-07-21 09:32:51 -04004153 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04004154 struct filename *from;
Dave Hansen6902d922006-09-30 23:29:01 -07004155 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04004156 struct path path;
Jeff Laytonf46d3562012-12-11 12:10:08 -05004157 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004158
4159 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04004160 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004161 return PTR_ERR(from);
Jeff Laytonf46d3562012-12-11 12:10:08 -05004162retry:
4163 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07004164 error = PTR_ERR(dentry);
4165 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04004166 goto out_putname;
Dave Hansen6902d922006-09-30 23:29:01 -07004167
Jeff Layton91a27b22012-10-10 15:25:28 -04004168 error = security_path_symlink(&path, dentry, from->name);
Al Viroa8104a92012-07-20 02:25:00 +04004169 if (!error)
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07004170 error = vfs_symlink2(path.mnt, path.dentry->d_inode, dentry, from->name);
Al Viro921a1652012-07-20 01:15:31 +04004171 done_path_create(&path, dentry);
Jeff Laytonf46d3562012-12-11 12:10:08 -05004172 if (retry_estale(error, lookup_flags)) {
4173 lookup_flags |= LOOKUP_REVAL;
4174 goto retry;
4175 }
Dave Hansen6902d922006-09-30 23:29:01 -07004176out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004177 putname(from);
4178 return error;
4179}
4180
Heiko Carstens3480b252009-01-14 14:14:16 +01004181SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004182{
4183 return sys_symlinkat(oldname, AT_FDCWD, newname);
4184}
4185
J. Bruce Fields146a8592011-09-20 17:14:31 -04004186/**
4187 * vfs_link - create a new link
4188 * @old_dentry: object to be linked
4189 * @dir: new parent
4190 * @new_dentry: where to create the new link
4191 * @delegated_inode: returns inode needing a delegation break
4192 *
4193 * The caller must hold dir->i_mutex
4194 *
4195 * If vfs_link discovers a delegation on the to-be-linked file in need
4196 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4197 * inode in delegated_inode. The caller should then break the delegation
4198 * and retry. Because breaking a delegation may take a long time, the
4199 * caller should drop the i_mutex before doing so.
4200 *
4201 * Alternatively, a caller may pass NULL for delegated_inode. This may
4202 * be appropriate for callers that expect the underlying filesystem not
4203 * to be NFS exported.
4204 */
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07004205int vfs_link2(struct vfsmount *mnt, struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004206{
4207 struct inode *inode = old_dentry->d_inode;
Al Viro8de52772012-02-06 12:45:27 -05004208 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004209 int error;
4210
4211 if (!inode)
4212 return -ENOENT;
4213
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07004214 error = may_create(mnt, dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004215 if (error)
4216 return error;
4217
4218 if (dir->i_sb != inode->i_sb)
4219 return -EXDEV;
4220
4221 /*
4222 * A link to an append-only or immutable file cannot be created.
4223 */
4224 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4225 return -EPERM;
Eric W. Biederman0bd23d092016-06-29 14:54:46 -05004226 /*
4227 * Updating the link count will likely cause i_uid and i_gid to
4228 * be writen back improperly if their true value is unknown to
4229 * the vfs.
4230 */
4231 if (HAS_UNMAPPED_ID(inode))
4232 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05004233 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004234 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02004235 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004236 return -EPERM;
4237
4238 error = security_inode_link(old_dentry, dir, new_dentry);
4239 if (error)
4240 return error;
4241
Al Viro59551022016-01-22 15:40:57 -05004242 inode_lock(inode);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05304243 /* Make sure we don't allow creating hardlink to an unlinked file */
Al Virof4e0c302013-06-11 08:34:36 +04004244 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05304245 error = -ENOENT;
Al Viro8de52772012-02-06 12:45:27 -05004246 else if (max_links && inode->i_nlink >= max_links)
4247 error = -EMLINK;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004248 else {
4249 error = try_break_deleg(inode, delegated_inode);
4250 if (!error)
4251 error = dir->i_op->link(old_dentry, dir, new_dentry);
4252 }
Al Virof4e0c302013-06-11 08:34:36 +04004253
4254 if (!error && (inode->i_state & I_LINKABLE)) {
4255 spin_lock(&inode->i_lock);
4256 inode->i_state &= ~I_LINKABLE;
4257 spin_unlock(&inode->i_lock);
4258 }
Al Viro59551022016-01-22 15:40:57 -05004259 inode_unlock(inode);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07004260 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02004261 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004262 return error;
4263}
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07004264EXPORT_SYMBOL(vfs_link2);
4265
4266int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4267{
4268 return vfs_link2(NULL, old_dentry, dir, new_dentry, delegated_inode);
4269}
Al Viro4d359502014-03-14 12:20:17 -04004270EXPORT_SYMBOL(vfs_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004271
4272/*
4273 * Hardlinks are often used in delicate situations. We avoid
4274 * security-related surprises by not following symlinks on the
4275 * newname. --KAB
4276 *
4277 * We don't follow them on the oldname either to be compatible
4278 * with linux 2.0, and to avoid hard-linking to directories
4279 * and other special files. --ADM
4280 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01004281SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4282 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004283{
4284 struct dentry *new_dentry;
Al Virodae6ad82011-06-26 11:50:15 -04004285 struct path old_path, new_path;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004286 struct inode *delegated_inode = NULL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304287 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004288 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004289
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304290 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004291 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304292 /*
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004293 * To use null names we require CAP_DAC_READ_SEARCH
4294 * This ensures that not everyone will be able to create
4295 * handlink using the passed filedescriptor.
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304296 */
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004297 if (flags & AT_EMPTY_PATH) {
4298 if (!capable(CAP_DAC_READ_SEARCH))
4299 return -ENOENT;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304300 how = LOOKUP_EMPTY;
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004301 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004302
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304303 if (flags & AT_SYMLINK_FOLLOW)
4304 how |= LOOKUP_FOLLOW;
Jeff Layton442e31c2012-12-20 16:15:38 -05004305retry:
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304306 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004307 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04004308 return error;
4309
Jeff Layton442e31c2012-12-20 16:15:38 -05004310 new_dentry = user_path_create(newdfd, newname, &new_path,
4311 (how & LOOKUP_REVAL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004312 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07004313 if (IS_ERR(new_dentry))
Al Virodae6ad82011-06-26 11:50:15 -04004314 goto out;
4315
4316 error = -EXDEV;
4317 if (old_path.mnt != new_path.mnt)
4318 goto out_dput;
Kees Cook800179c2012-07-25 17:29:07 -07004319 error = may_linkat(&old_path);
4320 if (unlikely(error))
4321 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04004322 error = security_path_link(old_path.dentry, &new_path, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004323 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04004324 goto out_dput;
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07004325 error = vfs_link2(old_path.mnt, old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
Dave Hansen75c3f292008-02-15 14:37:45 -08004326out_dput:
Al Viro921a1652012-07-20 01:15:31 +04004327 done_path_create(&new_path, new_dentry);
J. Bruce Fields146a8592011-09-20 17:14:31 -04004328 if (delegated_inode) {
4329 error = break_deleg_wait(&delegated_inode);
Oleg Drokind22e6332014-01-31 15:41:58 -05004330 if (!error) {
4331 path_put(&old_path);
J. Bruce Fields146a8592011-09-20 17:14:31 -04004332 goto retry;
Oleg Drokind22e6332014-01-31 15:41:58 -05004333 }
J. Bruce Fields146a8592011-09-20 17:14:31 -04004334 }
Jeff Layton442e31c2012-12-20 16:15:38 -05004335 if (retry_estale(error, how)) {
Oleg Drokind22e6332014-01-31 15:41:58 -05004336 path_put(&old_path);
Jeff Layton442e31c2012-12-20 16:15:38 -05004337 how |= LOOKUP_REVAL;
4338 goto retry;
4339 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004340out:
Al Viro2d8f3032008-07-22 09:59:21 -04004341 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004342
4343 return error;
4344}
4345
Heiko Carstens3480b252009-01-14 14:14:16 +01004346SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004347{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004348 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004349}
4350
Miklos Szeredibc270272014-04-01 17:08:42 +02004351/**
4352 * vfs_rename - rename a filesystem object
4353 * @old_dir: parent of source
4354 * @old_dentry: source
4355 * @new_dir: parent of destination
4356 * @new_dentry: destination
4357 * @delegated_inode: returns an inode needing a delegation break
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004358 * @flags: rename flags
Miklos Szeredibc270272014-04-01 17:08:42 +02004359 *
4360 * The caller must hold multiple mutexes--see lock_rename()).
4361 *
4362 * If vfs_rename discovers a delegation in need of breaking at either
4363 * the source or destination, it will return -EWOULDBLOCK and return a
4364 * reference to the inode in delegated_inode. The caller should then
4365 * break the delegation and retry. Because breaking a delegation may
4366 * take a long time, the caller should drop all locks before doing
4367 * so.
4368 *
4369 * Alternatively, a caller may pass NULL for delegated_inode. This may
4370 * be appropriate for callers that expect the underlying filesystem not
4371 * to be NFS exported.
4372 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07004373 * The worst of all namespace operations - renaming directory. "Perverted"
4374 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4375 * Problems:
J. Bruce Fieldsd03b29a2014-02-17 16:52:33 -05004376 * a) we can get into loop creation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004377 * b) race potential - two innocent renames can create a loop together.
4378 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004379 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07004380 * story.
J. Bruce Fields6cedba82012-03-05 11:40:41 -05004381 * c) we have to lock _four_ objects - parents and victim (if it exists),
4382 * and source (if it is not a directory).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004383 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07004384 * whether the target exists). Solution: try to be smart with locking
4385 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004386 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07004387 * move will be locked. Thus we can rank directories by the tree
4388 * (ancestors first) and rank all non-directories after them.
4389 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004390 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004391 * HOWEVER, it relies on the assumption that any object with ->lookup()
4392 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4393 * we'd better make sure that there's no link(2) for them.
Sage Weile4eaac02011-05-24 13:06:07 -07004394 * d) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004395 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07004396 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05004397 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07004398 * locking].
4399 */
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07004400int vfs_rename2(struct vfsmount *mnt,
4401 struct inode *old_dir, struct dentry *old_dentry,
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004402 struct inode *new_dir, struct dentry *new_dentry,
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004403 struct inode **delegated_inode, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004404{
4405 int error;
Miklos Szeredibc270272014-04-01 17:08:42 +02004406 bool is_dir = d_is_dir(old_dentry);
Eric Paris59b0df22010-02-08 12:53:52 -05004407 const unsigned char *old_name;
Miklos Szeredibc270272014-04-01 17:08:42 +02004408 struct inode *source = old_dentry->d_inode;
4409 struct inode *target = new_dentry->d_inode;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004410 bool new_is_dir = false;
4411 unsigned max_links = new_dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004412
Miklos Szeredi9409e222016-05-11 01:16:37 +02004413 /*
4414 * Check source == target.
4415 * On overlayfs need to look at underlying inodes.
4416 */
Miklos Szeredi2d902672016-06-30 08:53:27 +02004417 if (d_real_inode(old_dentry) == d_real_inode(new_dentry))
Miklos Szeredibc270272014-04-01 17:08:42 +02004418 return 0;
4419
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07004420 error = may_delete(mnt, old_dir, old_dentry, is_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004421 if (error)
4422 return error;
4423
Miklos Szeredida1ce062014-04-01 17:08:43 +02004424 if (!target) {
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07004425 error = may_create(mnt, new_dir, new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004426 } else {
4427 new_is_dir = d_is_dir(new_dentry);
4428
4429 if (!(flags & RENAME_EXCHANGE))
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07004430 error = may_delete(mnt, new_dir, new_dentry, is_dir);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004431 else
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07004432 error = may_delete(mnt, new_dir, new_dentry, new_is_dir);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004433 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004434 if (error)
4435 return error;
4436
Miklos Szeredi2773bf02016-09-27 11:03:58 +02004437 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004438 return -EPERM;
4439
Miklos Szeredibc270272014-04-01 17:08:42 +02004440 /*
4441 * If we are going to change the parent - check write permissions,
4442 * we'll need to flip '..'.
4443 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02004444 if (new_dir != old_dir) {
4445 if (is_dir) {
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07004446 error = inode_permission2(mnt, source, MAY_WRITE);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004447 if (error)
4448 return error;
4449 }
4450 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07004451 error = inode_permission2(mnt, target, MAY_WRITE);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004452 if (error)
4453 return error;
4454 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004455 }
Robert Love0eeca282005-07-12 17:06:03 -04004456
Miklos Szeredi0b3974e2014-04-01 17:08:43 +02004457 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4458 flags);
Miklos Szeredibc270272014-04-01 17:08:42 +02004459 if (error)
4460 return error;
4461
4462 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4463 dget(new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004464 if (!is_dir || (flags & RENAME_EXCHANGE))
Miklos Szeredibc270272014-04-01 17:08:42 +02004465 lock_two_nondirectories(source, target);
4466 else if (target)
Al Viro59551022016-01-22 15:40:57 -05004467 inode_lock(target);
Miklos Szeredibc270272014-04-01 17:08:42 +02004468
4469 error = -EBUSY;
Eric W. Biederman7af13642013-10-04 19:15:13 -07004470 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
Miklos Szeredibc270272014-04-01 17:08:42 +02004471 goto out;
4472
Miklos Szeredida1ce062014-04-01 17:08:43 +02004473 if (max_links && new_dir != old_dir) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004474 error = -EMLINK;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004475 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
Miklos Szeredibc270272014-04-01 17:08:42 +02004476 goto out;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004477 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4478 old_dir->i_nlink >= max_links)
4479 goto out;
4480 }
4481 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4482 shrink_dcache_parent(new_dentry);
4483 if (!is_dir) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004484 error = try_break_deleg(source, delegated_inode);
4485 if (error)
4486 goto out;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004487 }
4488 if (target && !new_is_dir) {
4489 error = try_break_deleg(target, delegated_inode);
4490 if (error)
4491 goto out;
Miklos Szeredibc270272014-04-01 17:08:42 +02004492 }
Miklos Szeredi2773bf02016-09-27 11:03:58 +02004493 error = old_dir->i_op->rename(old_dir, old_dentry,
Miklos Szeredi18fc84d2016-09-27 11:03:58 +02004494 new_dir, new_dentry, flags);
Miklos Szeredibc270272014-04-01 17:08:42 +02004495 if (error)
4496 goto out;
4497
Miklos Szeredida1ce062014-04-01 17:08:43 +02004498 if (!(flags & RENAME_EXCHANGE) && target) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004499 if (is_dir)
4500 target->i_flags |= S_DEAD;
4501 dont_mount(new_dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07004502 detach_mounts(new_dentry);
Miklos Szeredibc270272014-04-01 17:08:42 +02004503 }
Miklos Szeredida1ce062014-04-01 17:08:43 +02004504 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4505 if (!(flags & RENAME_EXCHANGE))
4506 d_move(old_dentry, new_dentry);
4507 else
4508 d_exchange(old_dentry, new_dentry);
4509 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004510out:
Miklos Szeredida1ce062014-04-01 17:08:43 +02004511 if (!is_dir || (flags & RENAME_EXCHANGE))
Miklos Szeredibc270272014-04-01 17:08:42 +02004512 unlock_two_nondirectories(source, target);
4513 else if (target)
Al Viro59551022016-01-22 15:40:57 -05004514 inode_unlock(target);
Miklos Szeredibc270272014-04-01 17:08:42 +02004515 dput(new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004516 if (!error) {
Al Viro123df292009-12-25 04:57:57 -05004517 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Miklos Szeredida1ce062014-04-01 17:08:43 +02004518 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4519 if (flags & RENAME_EXCHANGE) {
4520 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4521 new_is_dir, NULL, new_dentry);
4522 }
4523 }
Robert Love0eeca282005-07-12 17:06:03 -04004524 fsnotify_oldname_free(old_name);
4525
Linus Torvalds1da177e2005-04-16 15:20:36 -07004526 return error;
4527}
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07004528EXPORT_SYMBOL(vfs_rename2);
4529
4530int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4531 struct inode *new_dir, struct dentry *new_dentry,
4532 struct inode **delegated_inode, unsigned int flags)
4533{
4534 return vfs_rename2(NULL, old_dir, old_dentry, new_dir, new_dentry, delegated_inode, flags);
4535}
Al Viro4d359502014-03-14 12:20:17 -04004536EXPORT_SYMBOL(vfs_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004537
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004538SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4539 int, newdfd, const char __user *, newname, unsigned int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004540{
Al Viro2ad94ae2008-07-21 09:32:51 -04004541 struct dentry *old_dentry, *new_dentry;
4542 struct dentry *trap;
Al Virof5beed72015-04-30 16:09:11 -04004543 struct path old_path, new_path;
4544 struct qstr old_last, new_last;
4545 int old_type, new_type;
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004546 struct inode *delegated_inode = NULL;
Jeff Layton91a27b22012-10-10 15:25:28 -04004547 struct filename *from;
4548 struct filename *to;
Al Virof5beed72015-04-30 16:09:11 -04004549 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
Jeff Laytonc6a94282012-12-11 12:10:10 -05004550 bool should_retry = false;
Al Viro2ad94ae2008-07-21 09:32:51 -04004551 int error;
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004552
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004553 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredida1ce062014-04-01 17:08:43 +02004554 return -EINVAL;
4555
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004556 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4557 (flags & RENAME_EXCHANGE))
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004558 return -EINVAL;
4559
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004560 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4561 return -EPERM;
4562
Al Virof5beed72015-04-30 16:09:11 -04004563 if (flags & RENAME_EXCHANGE)
4564 target_flags = 0;
4565
Jeff Laytonc6a94282012-12-11 12:10:10 -05004566retry:
Al Virof5beed72015-04-30 16:09:11 -04004567 from = user_path_parent(olddfd, oldname,
4568 &old_path, &old_last, &old_type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04004569 if (IS_ERR(from)) {
4570 error = PTR_ERR(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004571 goto exit;
Jeff Layton91a27b22012-10-10 15:25:28 -04004572 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004573
Al Virof5beed72015-04-30 16:09:11 -04004574 to = user_path_parent(newdfd, newname,
4575 &new_path, &new_last, &new_type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04004576 if (IS_ERR(to)) {
4577 error = PTR_ERR(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004578 goto exit1;
Jeff Layton91a27b22012-10-10 15:25:28 -04004579 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004580
4581 error = -EXDEV;
Al Virof5beed72015-04-30 16:09:11 -04004582 if (old_path.mnt != new_path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004583 goto exit2;
4584
Linus Torvalds1da177e2005-04-16 15:20:36 -07004585 error = -EBUSY;
Al Virof5beed72015-04-30 16:09:11 -04004586 if (old_type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004587 goto exit2;
4588
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004589 if (flags & RENAME_NOREPLACE)
4590 error = -EEXIST;
Al Virof5beed72015-04-30 16:09:11 -04004591 if (new_type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004592 goto exit2;
4593
Al Virof5beed72015-04-30 16:09:11 -04004594 error = mnt_want_write(old_path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02004595 if (error)
4596 goto exit2;
4597
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004598retry_deleg:
Al Virof5beed72015-04-30 16:09:11 -04004599 trap = lock_rename(new_path.dentry, old_path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004600
Al Virof5beed72015-04-30 16:09:11 -04004601 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004602 error = PTR_ERR(old_dentry);
4603 if (IS_ERR(old_dentry))
4604 goto exit3;
4605 /* source must exist */
4606 error = -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01004607 if (d_is_negative(old_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004608 goto exit4;
Al Virof5beed72015-04-30 16:09:11 -04004609 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004610 error = PTR_ERR(new_dentry);
4611 if (IS_ERR(new_dentry))
4612 goto exit4;
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004613 error = -EEXIST;
4614 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4615 goto exit5;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004616 if (flags & RENAME_EXCHANGE) {
4617 error = -ENOENT;
4618 if (d_is_negative(new_dentry))
4619 goto exit5;
4620
4621 if (!d_is_dir(new_dentry)) {
4622 error = -ENOTDIR;
Al Virof5beed72015-04-30 16:09:11 -04004623 if (new_last.name[new_last.len])
Miklos Szeredida1ce062014-04-01 17:08:43 +02004624 goto exit5;
4625 }
4626 }
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004627 /* unless the source is a directory trailing slashes give -ENOTDIR */
4628 if (!d_is_dir(old_dentry)) {
4629 error = -ENOTDIR;
Al Virof5beed72015-04-30 16:09:11 -04004630 if (old_last.name[old_last.len])
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004631 goto exit5;
Al Virof5beed72015-04-30 16:09:11 -04004632 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004633 goto exit5;
4634 }
4635 /* source should not be ancestor of target */
4636 error = -EINVAL;
4637 if (old_dentry == trap)
4638 goto exit5;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004639 /* target should not be an ancestor of source */
Miklos Szeredida1ce062014-04-01 17:08:43 +02004640 if (!(flags & RENAME_EXCHANGE))
4641 error = -ENOTEMPTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004642 if (new_dentry == trap)
4643 goto exit5;
4644
Al Virof5beed72015-04-30 16:09:11 -04004645 error = security_path_rename(&old_path, old_dentry,
4646 &new_path, new_dentry, flags);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004647 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02004648 goto exit5;
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07004649 error = vfs_rename2(old_path.mnt, old_path.dentry->d_inode, old_dentry,
Al Virof5beed72015-04-30 16:09:11 -04004650 new_path.dentry->d_inode, new_dentry,
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004651 &delegated_inode, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004652exit5:
4653 dput(new_dentry);
4654exit4:
4655 dput(old_dentry);
4656exit3:
Al Virof5beed72015-04-30 16:09:11 -04004657 unlock_rename(new_path.dentry, old_path.dentry);
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004658 if (delegated_inode) {
4659 error = break_deleg_wait(&delegated_inode);
4660 if (!error)
4661 goto retry_deleg;
4662 }
Al Virof5beed72015-04-30 16:09:11 -04004663 mnt_drop_write(old_path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004664exit2:
Jeff Laytonc6a94282012-12-11 12:10:10 -05004665 if (retry_estale(error, lookup_flags))
4666 should_retry = true;
Al Virof5beed72015-04-30 16:09:11 -04004667 path_put(&new_path);
Al Viro2ad94ae2008-07-21 09:32:51 -04004668 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004669exit1:
Al Virof5beed72015-04-30 16:09:11 -04004670 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004671 putname(from);
Jeff Laytonc6a94282012-12-11 12:10:10 -05004672 if (should_retry) {
4673 should_retry = false;
4674 lookup_flags |= LOOKUP_REVAL;
4675 goto retry;
4676 }
Al Viro2ad94ae2008-07-21 09:32:51 -04004677exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004678 return error;
4679}
4680
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004681SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4682 int, newdfd, const char __user *, newname)
4683{
4684 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4685}
4686
Heiko Carstensa26eab22009-01-14 14:14:17 +01004687SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004688{
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004689 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004690}
4691
Miklos Szeredi787fb6b2014-10-24 00:14:36 +02004692int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4693{
Daniel Rosenbergbbcd0ff2016-10-26 16:27:45 -07004694 int error = may_create(NULL, dir, dentry);
Miklos Szeredi787fb6b2014-10-24 00:14:36 +02004695 if (error)
4696 return error;
4697
4698 if (!dir->i_op->mknod)
4699 return -EPERM;
4700
4701 return dir->i_op->mknod(dir, dentry,
4702 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4703}
4704EXPORT_SYMBOL(vfs_whiteout);
4705
Al Viro5d826c82014-03-14 13:42:45 -04004706int readlink_copy(char __user *buffer, int buflen, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004707{
Al Viro5d826c82014-03-14 13:42:45 -04004708 int len = PTR_ERR(link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004709 if (IS_ERR(link))
4710 goto out;
4711
4712 len = strlen(link);
4713 if (len > (unsigned) buflen)
4714 len = buflen;
4715 if (copy_to_user(buffer, link, len))
4716 len = -EFAULT;
4717out:
4718 return len;
4719}
4720
4721/*
4722 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
Al Viro6b255392015-11-17 10:20:54 -05004723 * have ->get_link() not calling nd_jump_link(). Using (or not using) it
4724 * for any given inode is up to filesystem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004725 */
4726int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4727{
Al Virofceef392015-12-29 15:58:39 -05004728 DEFINE_DELAYED_CALL(done);
Al Viro5f2c4172015-05-07 11:14:26 -04004729 struct inode *inode = d_inode(dentry);
4730 const char *link = inode->i_link;
Marcin Slusarz694a1762008-06-09 16:40:37 -07004731 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004732
Al Virod4dee482015-04-30 20:08:02 -04004733 if (!link) {
Al Virofceef392015-12-29 15:58:39 -05004734 link = inode->i_op->get_link(dentry, inode, &done);
Al Virod4dee482015-04-30 20:08:02 -04004735 if (IS_ERR(link))
4736 return PTR_ERR(link);
4737 }
Al Viro680baac2015-05-02 13:32:22 -04004738 res = readlink_copy(buffer, buflen, link);
Al Virofceef392015-12-29 15:58:39 -05004739 do_delayed_call(&done);
Marcin Slusarz694a1762008-06-09 16:40:37 -07004740 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004741}
Al Viro4d359502014-03-14 12:20:17 -04004742EXPORT_SYMBOL(generic_readlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004743
Miklos Szeredid60874c2016-10-04 14:40:45 +02004744/**
4745 * vfs_get_link - get symlink body
4746 * @dentry: dentry on which to get symbolic link
4747 * @done: caller needs to free returned data with this
4748 *
4749 * Calls security hook and i_op->get_link() on the supplied inode.
4750 *
4751 * It does not touch atime. That's up to the caller if necessary.
4752 *
4753 * Does not work on "special" symlinks like /proc/$$/fd/N
4754 */
4755const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4756{
4757 const char *res = ERR_PTR(-EINVAL);
4758 struct inode *inode = d_inode(dentry);
4759
4760 if (d_is_symlink(dentry)) {
4761 res = ERR_PTR(security_inode_readlink(dentry));
4762 if (!res)
4763 res = inode->i_op->get_link(dentry, inode, done);
4764 }
4765 return res;
4766}
4767EXPORT_SYMBOL(vfs_get_link);
4768
Linus Torvalds1da177e2005-04-16 15:20:36 -07004769/* get the link contents into pagecache */
Al Viro6b255392015-11-17 10:20:54 -05004770const char *page_get_link(struct dentry *dentry, struct inode *inode,
Al Virofceef392015-12-29 15:58:39 -05004771 struct delayed_call *callback)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004772{
Duane Griffinebd09ab2008-12-19 20:47:12 +00004773 char *kaddr;
4774 struct page *page;
Al Viro6b255392015-11-17 10:20:54 -05004775 struct address_space *mapping = inode->i_mapping;
4776
Al Virod3883d42015-11-17 10:41:04 -05004777 if (!dentry) {
4778 page = find_get_page(mapping, 0);
4779 if (!page)
4780 return ERR_PTR(-ECHILD);
4781 if (!PageUptodate(page)) {
4782 put_page(page);
4783 return ERR_PTR(-ECHILD);
4784 }
4785 } else {
4786 page = read_mapping_page(mapping, 0, NULL);
4787 if (IS_ERR(page))
4788 return (char*)page;
4789 }
Al Virofceef392015-12-29 15:58:39 -05004790 set_delayed_call(callback, page_put_link, page);
Al Viro21fc61c2015-11-17 01:07:57 -05004791 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4792 kaddr = page_address(page);
Al Viro6b255392015-11-17 10:20:54 -05004793 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
Duane Griffinebd09ab2008-12-19 20:47:12 +00004794 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004795}
4796
Al Viro6b255392015-11-17 10:20:54 -05004797EXPORT_SYMBOL(page_get_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004798
Al Virofceef392015-12-29 15:58:39 -05004799void page_put_link(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004800{
Al Virofceef392015-12-29 15:58:39 -05004801 put_page(arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004802}
Al Viro4d359502014-03-14 12:20:17 -04004803EXPORT_SYMBOL(page_put_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004804
Al Viroaa80dea2015-11-16 18:26:34 -05004805int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4806{
Al Virofceef392015-12-29 15:58:39 -05004807 DEFINE_DELAYED_CALL(done);
Al Viro6b255392015-11-17 10:20:54 -05004808 int res = readlink_copy(buffer, buflen,
4809 page_get_link(dentry, d_inode(dentry),
Al Virofceef392015-12-29 15:58:39 -05004810 &done));
4811 do_delayed_call(&done);
Al Viroaa80dea2015-11-16 18:26:34 -05004812 return res;
4813}
4814EXPORT_SYMBOL(page_readlink);
4815
Nick Piggin54566b22009-01-04 12:00:53 -08004816/*
4817 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4818 */
4819int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004820{
4821 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004822 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07004823 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08004824 int err;
Nick Piggin54566b22009-01-04 12:00:53 -08004825 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4826 if (nofs)
4827 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004828
NeilBrown7e53cac2006-03-25 03:07:57 -08004829retry:
Nick Pigginafddba42007-10-16 01:25:01 -07004830 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08004831 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004832 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07004833 goto fail;
4834
Al Viro21fc61c2015-11-17 01:07:57 -05004835 memcpy(page_address(page), symname, len-1);
Nick Pigginafddba42007-10-16 01:25:01 -07004836
4837 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4838 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004839 if (err < 0)
4840 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07004841 if (err < len-1)
4842 goto retry;
4843
Linus Torvalds1da177e2005-04-16 15:20:36 -07004844 mark_inode_dirty(inode);
4845 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004846fail:
4847 return err;
4848}
Al Viro4d359502014-03-14 12:20:17 -04004849EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004850
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004851int page_symlink(struct inode *inode, const char *symname, int len)
4852{
4853 return __page_symlink(inode, symname, len,
Michal Hockoc62d2552015-11-06 16:28:49 -08004854 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004855}
Al Viro4d359502014-03-14 12:20:17 -04004856EXPORT_SYMBOL(page_symlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004857
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08004858const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004859 .readlink = generic_readlink,
Al Viro6b255392015-11-17 10:20:54 -05004860 .get_link = page_get_link,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004861};
Linus Torvalds1da177e2005-04-16 15:20:36 -07004862EXPORT_SYMBOL(page_symlink_inode_operations);