blob: 671c3c1a3425bab83d07a6b1091360dc7d03b9f0 [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001// SPDX-License-Identifier: GPL-2.0
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
3 * linux/fs/namei.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8/*
9 * Some corrections by tytso.
10 */
11
12/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
13 * lookup logic.
14 */
15/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
16 */
17
18#include <linux/init.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050019#include <linux/export.h>
David S. Miller44696902012-05-23 20:12:50 -070020#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/slab.h>
22#include <linux/fs.h>
23#include <linux/namei.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/pagemap.h>
Robert Love0eeca282005-07-12 17:06:03 -040025#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/personality.h>
27#include <linux/security.h>
Mimi Zohar6146f0d2009-02-04 09:06:57 -050028#include <linux/ima.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/syscalls.h>
30#include <linux/mount.h>
31#include <linux/audit.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080032#include <linux/capability.h>
Trond Myklebust834f2a42005-10-18 14:20:16 -070033#include <linux/file.h>
Ulrich Drepper5590ff02006-01-18 17:43:53 -080034#include <linux/fcntl.h>
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -070035#include <linux/device_cgroup.h>
Al Viro5ad4e532009-03-29 19:50:06 -040036#include <linux/fs_struct.h>
Linus Torvaldse77819e2011-07-22 19:30:19 -070037#include <linux/posix_acl.h>
Linus Torvalds99d263d2014-09-13 11:30:10 -070038#include <linux/hash.h>
George Spelvin2a18da7a2016-05-23 07:43:58 -040039#include <linux/bitops.h>
Eric W. Biedermanaeaa4a72016-07-23 11:20:44 -050040#include <linux/init_task.h>
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080041#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
Eric Parise81e3f42009-12-04 15:47:36 -050043#include "internal.h"
Al Viroc7105362011-11-24 18:22:03 -050044#include "mount.h"
Eric Parise81e3f42009-12-04 15:47:36 -050045
Linus Torvalds1da177e2005-04-16 15:20:36 -070046/* [Feb-1997 T. Schoebel-Theuer]
47 * Fundamental changes in the pathname lookup mechanisms (namei)
48 * were necessary because of omirr. The reason is that omirr needs
49 * to know the _real_ pathname, not the user-supplied one, in case
50 * of symlinks (and also when transname replacements occur).
51 *
52 * The new code replaces the old recursive symlink resolution with
53 * an iterative one (in case of non-nested symlink chains). It does
54 * this with calls to <fs>_follow_link().
55 * As a side effect, dir_namei(), _namei() and follow_link() are now
56 * replaced with a single function lookup_dentry() that can handle all
57 * the special cases of the former code.
58 *
59 * With the new dcache, the pathname is stored at each inode, at least as
60 * long as the refcount of the inode is positive. As a side effect, the
61 * size of the dcache depends on the inode cache and thus is dynamic.
62 *
63 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
64 * resolution to correspond with current state of the code.
65 *
66 * Note that the symlink resolution is not *completely* iterative.
67 * There is still a significant amount of tail- and mid- recursion in
68 * the algorithm. Also, note that <fs>_readlink() is not used in
69 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
70 * may return different results than <fs>_follow_link(). Many virtual
71 * filesystems (including /proc) exhibit this behavior.
72 */
73
74/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
75 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
76 * and the name already exists in form of a symlink, try to create the new
77 * name indicated by the symlink. The old code always complained that the
78 * name already exists, due to not following the symlink even if its target
79 * is nonexistent. The new semantics affects also mknod() and link() when
Lucas De Marchi25985ed2011-03-30 22:57:33 -030080 * the name is a symlink pointing to a non-existent name.
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 *
82 * I don't know which semantics is the right one, since I have no access
83 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
84 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
85 * "old" one. Personally, I think the new semantics is much more logical.
86 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
87 * file does succeed in both HP-UX and SunOs, but not in Solaris
88 * and in the old Linux semantics.
89 */
90
91/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
92 * semantics. See the comments in "open_namei" and "do_link" below.
93 *
94 * [10-Sep-98 Alan Modra] Another symlink change.
95 */
96
97/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
98 * inside the path - always follow.
99 * in the last component in creation/removal/renaming - never follow.
100 * if LOOKUP_FOLLOW passed - follow.
101 * if the pathname has trailing slashes - follow.
102 * otherwise - don't follow.
103 * (applied in that order).
104 *
105 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
106 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
107 * During the 2.4 we need to fix the userland stuff depending on it -
108 * hopefully we will be able to get rid of that wart in 2.5. So far only
109 * XEmacs seems to be relying on it...
110 */
111/*
112 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800113 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 * any extra contention...
115 */
116
117/* In order to reduce some races, while at the same time doing additional
118 * checking and hopefully speeding things up, we copy filenames to the
119 * kernel data space before using them..
120 *
121 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
122 * PATH_MAX includes the nul terminator --RR.
123 */
Jeff Layton91a27b22012-10-10 15:25:28 -0400124
Al Virofd2f7cb2015-02-22 20:07:13 -0500125#define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
Jeff Layton7950e382012-10-10 16:43:13 -0400126
David Drysdale51f39a12014-12-12 16:57:29 -0800127struct filename *
Jeff Layton91a27b22012-10-10 15:25:28 -0400128getname_flags(const char __user *filename, int flags, int *empty)
129{
Al Viro94b5d262015-02-22 19:38:03 -0500130 struct filename *result;
Jeff Layton7950e382012-10-10 16:43:13 -0400131 char *kname;
Al Viro94b5d262015-02-22 19:38:03 -0500132 int len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133
Jeff Layton7ac86262012-10-10 15:25:28 -0400134 result = audit_reusename(filename);
135 if (result)
136 return result;
137
Jeff Layton7950e382012-10-10 16:43:13 -0400138 result = __getname();
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700139 if (unlikely(!result))
Eric Paris4043cde2012-01-03 14:23:08 -0500140 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141
Jeff Layton7950e382012-10-10 16:43:13 -0400142 /*
143 * First, try to embed the struct filename inside the names_cache
144 * allocation
145 */
Al Virofd2f7cb2015-02-22 20:07:13 -0500146 kname = (char *)result->iname;
Jeff Layton91a27b22012-10-10 15:25:28 -0400147 result->name = kname;
Jeff Layton7950e382012-10-10 16:43:13 -0400148
Al Viro94b5d262015-02-22 19:38:03 -0500149 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
Jeff Layton91a27b22012-10-10 15:25:28 -0400150 if (unlikely(len < 0)) {
Al Viro94b5d262015-02-22 19:38:03 -0500151 __putname(result);
152 return ERR_PTR(len);
Jeff Layton91a27b22012-10-10 15:25:28 -0400153 }
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700154
Jeff Layton7950e382012-10-10 16:43:13 -0400155 /*
156 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
157 * separate struct filename so we can dedicate the entire
158 * names_cache allocation for the pathname, and re-do the copy from
159 * userland.
160 */
Al Viro94b5d262015-02-22 19:38:03 -0500161 if (unlikely(len == EMBEDDED_NAME_MAX)) {
Al Virofd2f7cb2015-02-22 20:07:13 -0500162 const size_t size = offsetof(struct filename, iname[1]);
Jeff Layton7950e382012-10-10 16:43:13 -0400163 kname = (char *)result;
164
Al Virofd2f7cb2015-02-22 20:07:13 -0500165 /*
166 * size is chosen that way we to guarantee that
167 * result->iname[0] is within the same object and that
168 * kname can't be equal to result->iname, no matter what.
169 */
170 result = kzalloc(size, GFP_KERNEL);
Al Viro94b5d262015-02-22 19:38:03 -0500171 if (unlikely(!result)) {
172 __putname(kname);
173 return ERR_PTR(-ENOMEM);
Jeff Layton7950e382012-10-10 16:43:13 -0400174 }
175 result->name = kname;
Al Viro94b5d262015-02-22 19:38:03 -0500176 len = strncpy_from_user(kname, filename, PATH_MAX);
177 if (unlikely(len < 0)) {
178 __putname(kname);
179 kfree(result);
180 return ERR_PTR(len);
181 }
182 if (unlikely(len == PATH_MAX)) {
183 __putname(kname);
184 kfree(result);
185 return ERR_PTR(-ENAMETOOLONG);
186 }
Jeff Layton7950e382012-10-10 16:43:13 -0400187 }
188
Al Viro94b5d262015-02-22 19:38:03 -0500189 result->refcnt = 1;
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700190 /* The empty path is special. */
191 if (unlikely(!len)) {
192 if (empty)
Eric Paris4043cde2012-01-03 14:23:08 -0500193 *empty = 1;
Al Viro94b5d262015-02-22 19:38:03 -0500194 if (!(flags & LOOKUP_EMPTY)) {
195 putname(result);
196 return ERR_PTR(-ENOENT);
197 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 }
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700199
Jeff Layton7950e382012-10-10 16:43:13 -0400200 result->uptr = filename;
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800201 result->aname = NULL;
Jeff Layton7950e382012-10-10 16:43:13 -0400202 audit_getname(result);
203 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204}
205
Jeff Layton91a27b22012-10-10 15:25:28 -0400206struct filename *
207getname(const char __user * filename)
Al Virof52e0c12011-03-14 18:56:51 -0400208{
Linus Torvaldsf7493e52012-03-22 16:10:40 -0700209 return getname_flags(filename, 0, NULL);
Al Virof52e0c12011-03-14 18:56:51 -0400210}
211
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800212struct filename *
213getname_kernel(const char * filename)
214{
215 struct filename *result;
Paul Moore08518542015-01-21 23:59:56 -0500216 int len = strlen(filename) + 1;
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800217
218 result = __getname();
219 if (unlikely(!result))
220 return ERR_PTR(-ENOMEM);
221
Paul Moore08518542015-01-21 23:59:56 -0500222 if (len <= EMBEDDED_NAME_MAX) {
Al Virofd2f7cb2015-02-22 20:07:13 -0500223 result->name = (char *)result->iname;
Paul Moore08518542015-01-21 23:59:56 -0500224 } else if (len <= PATH_MAX) {
Al Viro30ce4d12018-04-08 11:57:10 -0400225 const size_t size = offsetof(struct filename, iname[1]);
Paul Moore08518542015-01-21 23:59:56 -0500226 struct filename *tmp;
227
Al Viro30ce4d12018-04-08 11:57:10 -0400228 tmp = kmalloc(size, GFP_KERNEL);
Paul Moore08518542015-01-21 23:59:56 -0500229 if (unlikely(!tmp)) {
230 __putname(result);
231 return ERR_PTR(-ENOMEM);
232 }
233 tmp->name = (char *)result;
Paul Moore08518542015-01-21 23:59:56 -0500234 result = tmp;
235 } else {
236 __putname(result);
237 return ERR_PTR(-ENAMETOOLONG);
238 }
239 memcpy((char *)result->name, filename, len);
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800240 result->uptr = NULL;
241 result->aname = NULL;
Paul Moore55422d02015-01-22 00:00:23 -0500242 result->refcnt = 1;
Paul Moorefd3522f2015-01-22 00:00:10 -0500243 audit_getname(result);
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800244
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800245 return result;
246}
247
Jeff Layton91a27b22012-10-10 15:25:28 -0400248void putname(struct filename *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249{
Paul Moore55422d02015-01-22 00:00:23 -0500250 BUG_ON(name->refcnt <= 0);
251
252 if (--name->refcnt > 0)
253 return;
254
Al Virofd2f7cb2015-02-22 20:07:13 -0500255 if (name->name != name->iname) {
Paul Moore55422d02015-01-22 00:00:23 -0500256 __putname(name->name);
257 kfree(name);
258 } else
259 __putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261
Linus Torvaldse77819e2011-07-22 19:30:19 -0700262static int check_acl(struct inode *inode, int mask)
263{
Linus Torvalds84635d62011-07-25 22:47:03 -0700264#ifdef CONFIG_FS_POSIX_ACL
Linus Torvaldse77819e2011-07-22 19:30:19 -0700265 struct posix_acl *acl;
266
Linus Torvaldse77819e2011-07-22 19:30:19 -0700267 if (mask & MAY_NOT_BLOCK) {
Al Viro35678662011-08-02 21:32:13 -0400268 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
269 if (!acl)
Linus Torvaldse77819e2011-07-22 19:30:19 -0700270 return -EAGAIN;
Al Viro35678662011-08-02 21:32:13 -0400271 /* no ->get_acl() calls in RCU mode... */
Andreas Gruenbacherb8a7a3a2016-03-24 14:38:37 +0100272 if (is_uncached_acl(acl))
Al Viro35678662011-08-02 21:32:13 -0400273 return -ECHILD;
Ari Savolainen206b1d02011-08-06 19:43:07 +0300274 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
Linus Torvaldse77819e2011-07-22 19:30:19 -0700275 }
276
Christoph Hellwig2982baa2013-12-20 05:16:38 -0800277 acl = get_acl(inode, ACL_TYPE_ACCESS);
278 if (IS_ERR(acl))
279 return PTR_ERR(acl);
Linus Torvaldse77819e2011-07-22 19:30:19 -0700280 if (acl) {
281 int error = posix_acl_permission(inode, acl, mask);
282 posix_acl_release(acl);
283 return error;
284 }
Linus Torvalds84635d62011-07-25 22:47:03 -0700285#endif
Linus Torvaldse77819e2011-07-22 19:30:19 -0700286
287 return -EAGAIN;
288}
289
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700290/*
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530291 * This does the basic permission checking
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700292 */
Al Viro7e401452011-06-20 19:12:17 -0400293static int acl_permission_check(struct inode *inode, int mask)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700294{
Linus Torvalds26cf46b2011-05-13 11:51:01 -0700295 unsigned int mode = inode->i_mode;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700296
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -0800297 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700298 mode >>= 6;
299 else {
Linus Torvaldse77819e2011-07-22 19:30:19 -0700300 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
Al Viro7e401452011-06-20 19:12:17 -0400301 int error = check_acl(inode, mask);
Nick Pigginb74c79e2011-01-07 17:49:58 +1100302 if (error != -EAGAIN)
303 return error;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700304 }
305
306 if (in_group_p(inode->i_gid))
307 mode >>= 3;
308 }
309
310 /*
311 * If the DACs are ok we don't need any capability check.
312 */
Al Viro9c2c7032011-06-20 19:06:22 -0400313 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700314 return 0;
315 return -EACCES;
316}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317
318/**
Nick Pigginb74c79e2011-01-07 17:49:58 +1100319 * generic_permission - check for access rights on a Posix-like filesystem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 * @inode: inode to check access rights for
Andreas Gruenbacher8fd90c82011-10-23 23:13:30 +0530321 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 *
323 * Used to check for read/write/execute permissions on a file.
324 * We use "fsuid" for this, letting us set arbitrary permissions
325 * for filesystem access without changing the "normal" uids which
Nick Pigginb74c79e2011-01-07 17:49:58 +1100326 * are used for other things.
327 *
328 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
329 * request cannot be satisfied (eg. requires blocking or too much complexity).
330 * It would then be called again in ref-walk mode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 */
Al Viro2830ba72011-06-20 19:16:29 -0400332int generic_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700334 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335
336 /*
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530337 * Do the basic permission checks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 */
Al Viro7e401452011-06-20 19:12:17 -0400339 ret = acl_permission_check(inode, mask);
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700340 if (ret != -EACCES)
341 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342
Al Virod594e7e2011-06-20 19:55:42 -0400343 if (S_ISDIR(inode->i_mode)) {
344 /* DACs are overridable for directories */
Al Virod594e7e2011-06-20 19:55:42 -0400345 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;
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700349 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 return 0;
Stephen Smalley2a4c2242017-03-10 12:14:18 -0500351 return -EACCES;
352 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353
354 /*
355 * Searching includes executable on directories, else just read.
356 */
Serge E. Hallyn7ea66002009-12-29 14:50:19 -0600357 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
Al Virod594e7e2011-06-20 19:55:42 -0400358 if (mask == MAY_READ)
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700359 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 return 0;
Stephen Smalley2a4c2242017-03-10 12:14:18 -0500361 /*
362 * Read/write DACs are always overridable.
363 * Executable DACs are overridable when there is
364 * at least one exec bit set.
365 */
366 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
367 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
368 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369
370 return -EACCES;
371}
Al Viro4d359502014-03-14 12:20:17 -0400372EXPORT_SYMBOL(generic_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700374/*
375 * We _really_ want to just do "generic_permission()" without
376 * even looking at the inode->i_op values. So we keep a cache
377 * flag in inode->i_opflags, that says "this has not special
378 * permission function, use the fast case".
379 */
380static inline int do_inode_permission(struct inode *inode, int mask)
381{
382 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
383 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 * sb_permission - Check superblock-level permissions
396 * @sb: Superblock of inode to check permission on
Randy Dunlap55852632012-08-18 17:39:25 -0700397 * @inode: Inode to check permission on
David Howells0bdaea92012-06-25 12:55:46 +0100398 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
399 *
400 * Separate out file-system wide checks from inode-specific permission checks.
401 */
402static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
403{
404 if (unlikely(mask & MAY_WRITE)) {
405 umode_t mode = inode->i_mode;
406
407 /* Nobody gets write access to a read-only fs. */
David Howellsbc98a422017-07-17 08:45:34 +0100408 if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
David Howells0bdaea92012-06-25 12:55:46 +0100409 return -EROFS;
410 }
411 return 0;
412}
413
414/**
415 * inode_permission - Check for access rights to a given inode
416 * @inode: Inode to check permission on
417 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
418 *
419 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
420 * this, letting us set arbitrary permissions for filesystem access without
421 * changing the "normal" UIDs which are used for other things.
422 *
423 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
424 */
425int inode_permission(struct inode *inode, int mask)
426{
427 int retval;
428
429 retval = sb_permission(inode->i_sb, inode, mask);
430 if (retval)
431 return retval;
Eric Biggers4bfd0542018-01-16 21:44:24 -0800432
433 if (unlikely(mask & MAY_WRITE)) {
434 /*
435 * Nobody gets write access to an immutable file.
436 */
437 if (IS_IMMUTABLE(inode))
438 return -EPERM;
439
440 /*
441 * Updating mtime will likely cause i_uid and i_gid to be
442 * written back improperly if their true value is unknown
443 * to the vfs.
444 */
445 if (HAS_UNMAPPED_ID(inode))
446 return -EACCES;
447 }
448
449 retval = do_inode_permission(inode, mask);
450 if (retval)
451 return retval;
452
453 retval = devcgroup_inode_permission(inode, mask);
454 if (retval)
455 return retval;
456
457 return security_inode_permission(inode, mask);
David Howells0bdaea92012-06-25 12:55:46 +0100458}
Al Viro4d359502014-03-14 12:20:17 -0400459EXPORT_SYMBOL(inode_permission);
David Howells0bdaea92012-06-25 12:55:46 +0100460
461/**
Jan Blunck5dd784d02008-02-14 19:34:38 -0800462 * path_get - get a reference to a path
463 * @path: path to get the reference to
464 *
465 * Given a path increment the reference count to the dentry and the vfsmount.
466 */
Al Virodcf787f2013-03-01 23:51:07 -0500467void path_get(const struct path *path)
Jan Blunck5dd784d02008-02-14 19:34:38 -0800468{
469 mntget(path->mnt);
470 dget(path->dentry);
471}
472EXPORT_SYMBOL(path_get);
473
474/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800475 * path_put - put a reference to a path
476 * @path: path to put the reference to
477 *
478 * Given a path decrement the reference count to the dentry and the vfsmount.
479 */
Al Virodcf787f2013-03-01 23:51:07 -0500480void path_put(const struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481{
Jan Blunck1d957f92008-02-14 19:34:35 -0800482 dput(path->dentry);
483 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484}
Jan Blunck1d957f92008-02-14 19:34:35 -0800485EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486
Al Viro894bc8c2015-05-02 07:16:16 -0400487#define EMBEDDED_LEVELS 2
Al Viro1f55a6e2014-11-01 19:30:41 -0400488struct nameidata {
489 struct path path;
Al Viro1cf26652015-05-06 16:01:56 -0400490 struct qstr last;
Al Viro1f55a6e2014-11-01 19:30:41 -0400491 struct path root;
492 struct inode *inode; /* path.dentry.d_inode */
493 unsigned int flags;
Al Viro9883d182015-05-13 07:28:08 -0400494 unsigned seq, m_seq;
Al Viro1f55a6e2014-11-01 19:30:41 -0400495 int last_type;
496 unsigned depth;
NeilBrown756daf22015-03-23 13:37:38 +1100497 int total_link_count;
Al Viro697fc6c2015-05-02 19:38:35 -0400498 struct saved {
499 struct path link;
Al Virofceef392015-12-29 15:58:39 -0500500 struct delayed_call done;
Al Viro697fc6c2015-05-02 19:38:35 -0400501 const char *name;
Al Viro0450b2d2015-05-08 13:23:53 -0400502 unsigned seq;
Al Viro894bc8c2015-05-02 07:16:16 -0400503 } *stack, internal[EMBEDDED_LEVELS];
Al Viro9883d182015-05-13 07:28:08 -0400504 struct filename *name;
505 struct nameidata *saved;
Al Virofceef392015-12-29 15:58:39 -0500506 struct inode *link_inode;
Al Viro9883d182015-05-13 07:28:08 -0400507 unsigned root_seq;
508 int dfd;
Kees Cook3859a272016-10-28 01:22:25 -0700509} __randomize_layout;
Al Viro1f55a6e2014-11-01 19:30:41 -0400510
Al Viro9883d182015-05-13 07:28:08 -0400511static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
Al Viro894bc8c2015-05-02 07:16:16 -0400512{
NeilBrown756daf22015-03-23 13:37:38 +1100513 struct nameidata *old = current->nameidata;
514 p->stack = p->internal;
Al Viroc8a53ee2015-05-12 18:43:07 -0400515 p->dfd = dfd;
516 p->name = name;
NeilBrown756daf22015-03-23 13:37:38 +1100517 p->total_link_count = old ? old->total_link_count : 0;
Al Viro9883d182015-05-13 07:28:08 -0400518 p->saved = old;
NeilBrown756daf22015-03-23 13:37:38 +1100519 current->nameidata = p;
Al Viro894bc8c2015-05-02 07:16:16 -0400520}
521
Al Viro9883d182015-05-13 07:28:08 -0400522static void restore_nameidata(void)
Al Viro894bc8c2015-05-02 07:16:16 -0400523{
Al Viro9883d182015-05-13 07:28:08 -0400524 struct nameidata *now = current->nameidata, *old = now->saved;
NeilBrown756daf22015-03-23 13:37:38 +1100525
526 current->nameidata = old;
527 if (old)
528 old->total_link_count = now->total_link_count;
Al Viroe1a63bb2015-12-05 21:06:33 -0500529 if (now->stack != now->internal)
NeilBrown756daf22015-03-23 13:37:38 +1100530 kfree(now->stack);
Al Viro894bc8c2015-05-02 07:16:16 -0400531}
532
533static int __nd_alloc_stack(struct nameidata *nd)
534{
Al Virobc40aee2015-05-09 13:04:24 -0400535 struct saved *p;
536
537 if (nd->flags & LOOKUP_RCU) {
Kees Cook6da2ec52018-06-12 13:55:00 -0700538 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
Al Virobc40aee2015-05-09 13:04:24 -0400539 GFP_ATOMIC);
540 if (unlikely(!p))
541 return -ECHILD;
542 } else {
Kees Cook6da2ec52018-06-12 13:55:00 -0700543 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
Al Viro894bc8c2015-05-02 07:16:16 -0400544 GFP_KERNEL);
Al Virobc40aee2015-05-09 13:04:24 -0400545 if (unlikely(!p))
546 return -ENOMEM;
547 }
Al Viro894bc8c2015-05-02 07:16:16 -0400548 memcpy(p, nd->internal, sizeof(nd->internal));
549 nd->stack = p;
550 return 0;
551}
552
Eric W. Biederman397d4252015-08-15 20:27:13 -0500553/**
554 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
555 * @path: nameidate to verify
556 *
557 * Rename can sometimes move a file or directory outside of a bind
558 * mount, path_connected allows those cases to be detected.
559 */
560static bool path_connected(const struct path *path)
561{
562 struct vfsmount *mnt = path->mnt;
Eric W. Biederman95dd7752018-03-14 18:20:29 -0500563 struct super_block *sb = mnt->mnt_sb;
Eric W. Biederman397d4252015-08-15 20:27:13 -0500564
Eric W. Biederman95dd7752018-03-14 18:20:29 -0500565 /* Bind mounts and multi-root filesystems can have disconnected paths */
566 if (!(sb->s_iflags & SB_I_MULTIROOT) && (mnt->mnt_root == sb->s_root))
Eric W. Biederman397d4252015-08-15 20:27:13 -0500567 return true;
568
569 return is_subdir(path->dentry, mnt->mnt_root);
570}
571
Al Viro894bc8c2015-05-02 07:16:16 -0400572static inline int nd_alloc_stack(struct nameidata *nd)
573{
Al Viroda4e0be2015-05-03 20:52:15 -0400574 if (likely(nd->depth != EMBEDDED_LEVELS))
Al Viro894bc8c2015-05-02 07:16:16 -0400575 return 0;
576 if (likely(nd->stack != nd->internal))
577 return 0;
578 return __nd_alloc_stack(nd);
579}
580
Al Viro7973387a2015-05-09 12:55:43 -0400581static void drop_links(struct nameidata *nd)
582{
583 int i = nd->depth;
584 while (i--) {
585 struct saved *last = nd->stack + i;
Al Virofceef392015-12-29 15:58:39 -0500586 do_delayed_call(&last->done);
587 clear_delayed_call(&last->done);
Al Viro7973387a2015-05-09 12:55:43 -0400588 }
589}
590
591static void terminate_walk(struct nameidata *nd)
592{
593 drop_links(nd);
594 if (!(nd->flags & LOOKUP_RCU)) {
595 int i;
596 path_put(&nd->path);
597 for (i = 0; i < nd->depth; i++)
598 path_put(&nd->stack[i].link);
Al Viro84a2bd32019-07-16 21:20:17 -0400599 if (nd->flags & LOOKUP_ROOT_GRABBED) {
Al Viro102b8af2015-05-12 17:35:52 -0400600 path_put(&nd->root);
Al Viro84a2bd32019-07-16 21:20:17 -0400601 nd->flags &= ~LOOKUP_ROOT_GRABBED;
Al Viro102b8af2015-05-12 17:35:52 -0400602 }
Al Viro7973387a2015-05-09 12:55:43 -0400603 } else {
604 nd->flags &= ~LOOKUP_RCU;
Al Viro7973387a2015-05-09 12:55:43 -0400605 rcu_read_unlock();
606 }
607 nd->depth = 0;
608}
609
610/* path_put is needed afterwards regardless of success or failure */
611static bool legitimize_path(struct nameidata *nd,
612 struct path *path, unsigned seq)
613{
614 int res = __legitimize_mnt(path->mnt, nd->m_seq);
615 if (unlikely(res)) {
616 if (res > 0)
617 path->mnt = NULL;
618 path->dentry = NULL;
619 return false;
620 }
621 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
622 path->dentry = NULL;
623 return false;
624 }
625 return !read_seqcount_retry(&path->dentry->d_seq, seq);
626}
627
628static bool legitimize_links(struct nameidata *nd)
629{
630 int i;
631 for (i = 0; i < nd->depth; i++) {
632 struct saved *last = nd->stack + i;
633 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
634 drop_links(nd);
635 nd->depth = i + 1;
636 return false;
637 }
638 }
639 return true;
640}
641
Al Viroee594bf2019-07-16 21:05:36 -0400642static bool legitimize_root(struct nameidata *nd)
643{
644 if (!nd->root.mnt || (nd->flags & LOOKUP_ROOT))
645 return true;
Al Viro84a2bd32019-07-16 21:20:17 -0400646 nd->flags |= LOOKUP_ROOT_GRABBED;
Al Viroee594bf2019-07-16 21:05:36 -0400647 return legitimize_path(nd, &nd->root, nd->root_seq);
648}
649
Al Viro19660af2011-03-25 10:32:48 -0400650/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100651 * Path walking has 2 modes, rcu-walk and ref-walk (see
Al Viro19660af2011-03-25 10:32:48 -0400652 * Documentation/filesystems/path-lookup.txt). In situations when we can't
653 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
Mike Marshall57e37152015-11-30 11:11:59 -0500654 * normal reference counts on dentries and vfsmounts to transition to ref-walk
Al Viro19660af2011-03-25 10:32:48 -0400655 * mode. Refcounts are grabbed at the last known good point before rcu-walk
656 * got stuck, so ref-walk may continue from there. If this is not successful
657 * (eg. a seqcount has changed), then failure is returned and it's up to caller
658 * to restart the path walk from the beginning in ref-walk mode.
Nick Piggin31e6b012011-01-07 17:49:52 +1100659 */
Nick Piggin31e6b012011-01-07 17:49:52 +1100660
661/**
Al Viro19660af2011-03-25 10:32:48 -0400662 * unlazy_walk - try to switch to ref-walk mode.
663 * @nd: nameidata pathwalk data
Randy Dunlap39191622011-01-08 19:36:21 -0800664 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100665 *
Al Viro4675ac32017-01-09 22:29:15 -0500666 * unlazy_walk attempts to legitimize the current nd->path and nd->root
667 * for ref-walk mode.
668 * Must be called from rcu-walk context.
Al Viro7973387a2015-05-09 12:55:43 -0400669 * Nothing should touch nameidata between unlazy_walk() failure and
670 * terminate_walk().
Nick Piggin31e6b012011-01-07 17:49:52 +1100671 */
Al Viro4675ac32017-01-09 22:29:15 -0500672static int unlazy_walk(struct nameidata *nd)
Nick Piggin31e6b012011-01-07 17:49:52 +1100673{
Nick Piggin31e6b012011-01-07 17:49:52 +1100674 struct dentry *parent = nd->path.dentry;
675
676 BUG_ON(!(nd->flags & LOOKUP_RCU));
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700677
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700678 nd->flags &= ~LOOKUP_RCU;
Al Viro7973387a2015-05-09 12:55:43 -0400679 if (unlikely(!legitimize_links(nd)))
Al Viro4675ac32017-01-09 22:29:15 -0500680 goto out1;
Al Viro84a2bd32019-07-16 21:20:17 -0400681 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
682 goto out;
Al Viroee594bf2019-07-16 21:05:36 -0400683 if (unlikely(!legitimize_root(nd)))
684 goto out;
Al Viro4675ac32017-01-09 22:29:15 -0500685 rcu_read_unlock();
686 BUG_ON(nd->inode != parent->d_inode);
687 return 0;
688
Al Viro84a2bd32019-07-16 21:20:17 -0400689out1:
Al Viro4675ac32017-01-09 22:29:15 -0500690 nd->path.mnt = NULL;
691 nd->path.dentry = NULL;
Al Viro4675ac32017-01-09 22:29:15 -0500692out:
693 rcu_read_unlock();
694 return -ECHILD;
695}
696
697/**
698 * unlazy_child - try to switch to ref-walk mode.
699 * @nd: nameidata pathwalk data
700 * @dentry: child of nd->path.dentry
701 * @seq: seq number to check dentry against
702 * Returns: 0 on success, -ECHILD on failure
703 *
704 * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry
705 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
706 * @nd. Must be called from rcu-walk context.
707 * Nothing should touch nameidata between unlazy_child() failure and
708 * terminate_walk().
709 */
710static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq)
711{
712 BUG_ON(!(nd->flags & LOOKUP_RCU));
713
714 nd->flags &= ~LOOKUP_RCU;
715 if (unlikely(!legitimize_links(nd)))
716 goto out2;
Al Viro7973387a2015-05-09 12:55:43 -0400717 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
718 goto out2;
Al Viro4675ac32017-01-09 22:29:15 -0500719 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
Al Viro7973387a2015-05-09 12:55:43 -0400720 goto out1;
Al Viro48a066e2013-09-29 22:06:07 -0400721
Linus Torvalds15570082013-09-02 11:38:06 -0700722 /*
Al Viro4675ac32017-01-09 22:29:15 -0500723 * We need to move both the parent and the dentry from the RCU domain
724 * to be properly refcounted. And the sequence number in the dentry
725 * validates *both* dentry counters, since we checked the sequence
726 * number of the parent after we got the child sequence number. So we
727 * know the parent must still be valid if the child sequence number is
Linus Torvalds15570082013-09-02 11:38:06 -0700728 */
Al Viro4675ac32017-01-09 22:29:15 -0500729 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
730 goto out;
Al Viro84a2bd32019-07-16 21:20:17 -0400731 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
732 goto out_dput;
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700733 /*
734 * Sequence counts matched. Now make sure that the root is
735 * still valid and get it if required.
736 */
Al Viro84a2bd32019-07-16 21:20:17 -0400737 if (unlikely(!legitimize_root(nd)))
738 goto out_dput;
Al Viro8b61e742013-11-08 12:45:01 -0500739 rcu_read_unlock();
Nick Piggin31e6b012011-01-07 17:49:52 +1100740 return 0;
Al Viro19660af2011-03-25 10:32:48 -0400741
Al Viro7973387a2015-05-09 12:55:43 -0400742out2:
743 nd->path.mnt = NULL;
744out1:
745 nd->path.dentry = NULL;
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700746out:
Al Viro8b61e742013-11-08 12:45:01 -0500747 rcu_read_unlock();
Al Viro84a2bd32019-07-16 21:20:17 -0400748 return -ECHILD;
749out_dput:
750 rcu_read_unlock();
751 dput(dentry);
Nick Piggin31e6b012011-01-07 17:49:52 +1100752 return -ECHILD;
753}
754
Al Viro4ce16ef32012-06-10 16:10:59 -0400755static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
Nick Piggin34286d62011-01-07 17:49:57 +1100756{
Al Viroa89f8332017-01-09 22:25:28 -0500757 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
758 return dentry->d_op->d_revalidate(dentry, flags);
759 else
760 return 1;
Nick Piggin34286d62011-01-07 17:49:57 +1100761}
762
Al Viro9f1fafe2011-03-25 11:00:12 -0400763/**
764 * complete_walk - successful completion of path walk
765 * @nd: pointer nameidata
Jeff Layton39159de2009-12-07 12:01:50 -0500766 *
Al Viro9f1fafe2011-03-25 11:00:12 -0400767 * If we had been in RCU mode, drop out of it and legitimize nd->path.
768 * Revalidate the final result, unless we'd already done that during
769 * the path walk or the filesystem doesn't ask for it. Return 0 on
770 * success, -error on failure. In case of failure caller does not
771 * need to drop nd->path.
Jeff Layton39159de2009-12-07 12:01:50 -0500772 */
Al Viro9f1fafe2011-03-25 11:00:12 -0400773static int complete_walk(struct nameidata *nd)
Jeff Layton39159de2009-12-07 12:01:50 -0500774{
Al Viro16c2cd72011-02-22 15:50:10 -0500775 struct dentry *dentry = nd->path.dentry;
Jeff Layton39159de2009-12-07 12:01:50 -0500776 int status;
Jeff Layton39159de2009-12-07 12:01:50 -0500777
Al Viro9f1fafe2011-03-25 11:00:12 -0400778 if (nd->flags & LOOKUP_RCU) {
Al Viro9f1fafe2011-03-25 11:00:12 -0400779 if (!(nd->flags & LOOKUP_ROOT))
780 nd->root.mnt = NULL;
Al Viro4675ac32017-01-09 22:29:15 -0500781 if (unlikely(unlazy_walk(nd)))
Al Viro48a066e2013-09-29 22:06:07 -0400782 return -ECHILD;
Al Viro9f1fafe2011-03-25 11:00:12 -0400783 }
784
Al Viro16c2cd72011-02-22 15:50:10 -0500785 if (likely(!(nd->flags & LOOKUP_JUMPED)))
Jeff Layton39159de2009-12-07 12:01:50 -0500786 return 0;
787
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500788 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
Al Viro16c2cd72011-02-22 15:50:10 -0500789 return 0;
790
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500791 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
Jeff Layton39159de2009-12-07 12:01:50 -0500792 if (status > 0)
793 return 0;
794
Al Viro16c2cd72011-02-22 15:50:10 -0500795 if (!status)
Jeff Layton39159de2009-12-07 12:01:50 -0500796 status = -ESTALE;
Al Viro16c2cd72011-02-22 15:50:10 -0500797
Jeff Layton39159de2009-12-07 12:01:50 -0500798 return status;
799}
800
Al Viro18d8c862015-05-12 16:32:34 -0400801static void set_root(struct nameidata *nd)
Al Viro2a737872009-04-07 11:49:53 -0400802{
Al Viro7bd88372014-09-13 21:55:46 -0400803 struct fs_struct *fs = current->fs;
Nick Pigginc28cc362011-01-07 17:49:53 +1100804
Al Viro9e6697e2015-12-05 20:07:21 -0500805 if (nd->flags & LOOKUP_RCU) {
806 unsigned seq;
807
808 do {
809 seq = read_seqcount_begin(&fs->seq);
810 nd->root = fs->root;
811 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
812 } while (read_seqcount_retry(&fs->seq, seq));
813 } else {
814 get_fs_root(fs, &nd->root);
Al Viro84a2bd32019-07-16 21:20:17 -0400815 nd->flags |= LOOKUP_ROOT_GRABBED;
Al Viro9e6697e2015-12-05 20:07:21 -0500816 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100817}
818
Jan Blunck1d957f92008-02-14 19:34:35 -0800819static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700820{
821 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800822 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700823 mntput(path->mnt);
824}
825
Nick Piggin7b9337a2011-01-14 08:42:43 +0000826static inline void path_to_nameidata(const struct path *path,
827 struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700828{
Nick Piggin31e6b012011-01-07 17:49:52 +1100829 if (!(nd->flags & LOOKUP_RCU)) {
830 dput(nd->path.dentry);
831 if (nd->path.mnt != path->mnt)
832 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800833 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100834 nd->path.mnt = path->mnt;
Jan Blunck4ac91372008-02-14 19:34:32 -0800835 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700836}
837
Al Viro248fb5b2015-12-05 20:51:58 -0500838static int nd_jump_root(struct nameidata *nd)
839{
840 if (nd->flags & LOOKUP_RCU) {
841 struct dentry *d;
842 nd->path = nd->root;
843 d = nd->path.dentry;
844 nd->inode = d->d_inode;
845 nd->seq = nd->root_seq;
846 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
847 return -ECHILD;
848 } else {
849 path_put(&nd->path);
850 nd->path = nd->root;
851 path_get(&nd->path);
852 nd->inode = nd->path.dentry->d_inode;
853 }
854 nd->flags |= LOOKUP_JUMPED;
855 return 0;
856}
857
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400858/*
Al Viro6b255392015-11-17 10:20:54 -0500859 * Helper to directly jump to a known parsed path from ->get_link,
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400860 * caller must have taken a reference to path beforehand.
861 */
Al Viro6e77137b2015-05-02 13:37:52 -0400862void nd_jump_link(struct path *path)
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400863{
Al Viro6e77137b2015-05-02 13:37:52 -0400864 struct nameidata *nd = current->nameidata;
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400865 path_put(&nd->path);
866
867 nd->path = *path;
868 nd->inode = nd->path.dentry->d_inode;
869 nd->flags |= LOOKUP_JUMPED;
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400870}
871
Al Virob9ff4422015-05-02 20:19:23 -0400872static inline void put_link(struct nameidata *nd)
Al Viro574197e2011-03-14 22:20:34 -0400873{
Al Viro21c30032015-05-03 21:06:24 -0400874 struct saved *last = nd->stack + --nd->depth;
Al Virofceef392015-12-29 15:58:39 -0500875 do_delayed_call(&last->done);
Al Viro6548fae2015-05-07 20:32:22 -0400876 if (!(nd->flags & LOOKUP_RCU))
877 path_put(&last->link);
Al Viro574197e2011-03-14 22:20:34 -0400878}
879
Linus Torvalds561ec642012-10-26 10:05:07 -0700880int sysctl_protected_symlinks __read_mostly = 0;
881int sysctl_protected_hardlinks __read_mostly = 0;
Salvatore Mesoraca30aba662018-08-23 17:00:35 -0700882int sysctl_protected_fifos __read_mostly;
883int sysctl_protected_regular __read_mostly;
Kees Cook800179c2012-07-25 17:29:07 -0700884
885/**
886 * may_follow_link - Check symlink following for unsafe situations
Randy Dunlap55852632012-08-18 17:39:25 -0700887 * @nd: nameidata pathwalk data
Kees Cook800179c2012-07-25 17:29:07 -0700888 *
889 * In the case of the sysctl_protected_symlinks sysctl being enabled,
890 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
891 * in a sticky world-writable directory. This is to protect privileged
892 * processes from failing races against path names that may change out
893 * from under them by way of other users creating malicious symlinks.
894 * It will permit symlinks to be followed only when outside a sticky
895 * world-writable directory, or when the uid of the symlink and follower
896 * match, or when the directory owner matches the symlink's owner.
897 *
898 * Returns 0 if following the symlink is allowed, -ve on error.
899 */
Al Virofec2fa22015-05-06 15:58:18 -0400900static inline int may_follow_link(struct nameidata *nd)
Kees Cook800179c2012-07-25 17:29:07 -0700901{
902 const struct inode *inode;
903 const struct inode *parent;
Seth Forshee2d7f9e22016-04-26 14:36:23 -0500904 kuid_t puid;
Kees Cook800179c2012-07-25 17:29:07 -0700905
906 if (!sysctl_protected_symlinks)
907 return 0;
908
909 /* Allowed if owner and follower match. */
Al Virofceef392015-12-29 15:58:39 -0500910 inode = nd->link_inode;
Eric W. Biederman81abe272012-08-03 09:38:08 -0700911 if (uid_eq(current_cred()->fsuid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700912 return 0;
913
914 /* Allowed if parent directory not sticky and world-writable. */
Al Viroaa65fa32015-08-04 23:23:50 -0400915 parent = nd->inode;
Kees Cook800179c2012-07-25 17:29:07 -0700916 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
917 return 0;
918
919 /* Allowed if parent directory and link owner match. */
Seth Forshee2d7f9e22016-04-26 14:36:23 -0500920 puid = parent->i_uid;
921 if (uid_valid(puid) && uid_eq(puid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700922 return 0;
923
Al Viro319565022015-05-07 20:37:40 -0400924 if (nd->flags & LOOKUP_RCU)
925 return -ECHILD;
926
Richard Guy Briggsea841ba2018-03-21 04:42:21 -0400927 audit_inode(nd->name, nd->stack[0].link.dentry, 0);
Richard Guy Briggs94b9d9b2018-03-21 04:42:20 -0400928 audit_log_link_denied("follow_link");
Kees Cook800179c2012-07-25 17:29:07 -0700929 return -EACCES;
930}
931
932/**
933 * safe_hardlink_source - Check for safe hardlink conditions
934 * @inode: the source inode to hardlink from
935 *
936 * Return false if at least one of the following conditions:
937 * - inode is not a regular file
938 * - inode is setuid
939 * - inode is setgid and group-exec
940 * - access failure for read and write
941 *
942 * Otherwise returns true.
943 */
944static bool safe_hardlink_source(struct inode *inode)
945{
946 umode_t mode = inode->i_mode;
947
948 /* Special files should not get pinned to the filesystem. */
949 if (!S_ISREG(mode))
950 return false;
951
952 /* Setuid files should not get pinned to the filesystem. */
953 if (mode & S_ISUID)
954 return false;
955
956 /* Executable setgid files should not get pinned to the filesystem. */
957 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
958 return false;
959
960 /* Hardlinking to unreadable or unwritable sources is dangerous. */
961 if (inode_permission(inode, MAY_READ | MAY_WRITE))
962 return false;
963
964 return true;
965}
966
967/**
968 * may_linkat - Check permissions for creating a hardlink
969 * @link: the source to hardlink from
970 *
971 * Block hardlink when all of:
972 * - sysctl_protected_hardlinks enabled
973 * - fsuid does not match inode
974 * - hardlink source is unsafe (see safe_hardlink_source() above)
Dirk Steinmetzf2ca3792015-10-20 16:09:19 +0200975 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
Kees Cook800179c2012-07-25 17:29:07 -0700976 *
977 * Returns 0 if successful, -ve on error.
978 */
979static int may_linkat(struct path *link)
980{
Eric W. Biederman593d1ce2017-09-14 12:07:32 -0500981 struct inode *inode = link->dentry->d_inode;
982
983 /* Inode writeback is not safe when the uid or gid are invalid. */
984 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
985 return -EOVERFLOW;
Kees Cook800179c2012-07-25 17:29:07 -0700986
987 if (!sysctl_protected_hardlinks)
988 return 0;
989
Kees Cook800179c2012-07-25 17:29:07 -0700990 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
991 * otherwise, it must be a safe source.
992 */
Kees Cookcc658db2017-06-21 09:53:06 -0700993 if (safe_hardlink_source(inode) || inode_owner_or_capable(inode))
Kees Cook800179c2012-07-25 17:29:07 -0700994 return 0;
995
Richard Guy Briggs94b9d9b2018-03-21 04:42:20 -0400996 audit_log_link_denied("linkat");
Kees Cook800179c2012-07-25 17:29:07 -0700997 return -EPERM;
998}
999
Salvatore Mesoraca30aba662018-08-23 17:00:35 -07001000/**
1001 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1002 * should be allowed, or not, on files that already
1003 * exist.
1004 * @dir: the sticky parent directory
1005 * @inode: the inode of the file to open
1006 *
1007 * Block an O_CREAT open of a FIFO (or a regular file) when:
1008 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1009 * - the file already exists
1010 * - we are in a sticky directory
1011 * - we don't own the file
1012 * - the owner of the directory doesn't own the file
1013 * - the directory is world writable
1014 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1015 * the directory doesn't have to be world writable: being group writable will
1016 * be enough.
1017 *
1018 * Returns 0 if the open is allowed, -ve on error.
1019 */
1020static int may_create_in_sticky(struct dentry * const dir,
1021 struct inode * const inode)
1022{
1023 if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1024 (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1025 likely(!(dir->d_inode->i_mode & S_ISVTX)) ||
1026 uid_eq(inode->i_uid, dir->d_inode->i_uid) ||
1027 uid_eq(current_fsuid(), inode->i_uid))
1028 return 0;
1029
1030 if (likely(dir->d_inode->i_mode & 0002) ||
1031 (dir->d_inode->i_mode & 0020 &&
1032 ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1033 (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1034 return -EACCES;
1035 }
1036 return 0;
1037}
1038
Al Viro3b2e7f72015-04-19 00:53:50 -04001039static __always_inline
1040const char *get_link(struct nameidata *nd)
Ian Kent051d3812006-03-27 01:14:53 -08001041{
Al Viroab104922015-05-10 11:50:01 -04001042 struct saved *last = nd->stack + nd->depth - 1;
Al Viro1cf26652015-05-06 16:01:56 -04001043 struct dentry *dentry = last->link.dentry;
Al Virofceef392015-12-29 15:58:39 -05001044 struct inode *inode = nd->link_inode;
Al Viro6d7b5aa2012-06-10 04:15:17 -04001045 int error;
Al Viro0a959df2015-04-18 18:23:41 -04001046 const char *res;
Ian Kent051d3812006-03-27 01:14:53 -08001047
NeilBrown8fa9dd22015-03-23 13:37:40 +11001048 if (!(nd->flags & LOOKUP_RCU)) {
1049 touch_atime(&last->link);
1050 cond_resched();
Miklos Szeredic6718542018-07-18 15:44:43 +02001051 } else if (atime_needs_update(&last->link, inode)) {
Al Viro4675ac32017-01-09 22:29:15 -05001052 if (unlikely(unlazy_walk(nd)))
NeilBrown8fa9dd22015-03-23 13:37:40 +11001053 return ERR_PTR(-ECHILD);
1054 touch_atime(&last->link);
1055 }
1056
NeilBrownbda0be72015-03-23 13:37:39 +11001057 error = security_inode_follow_link(dentry, inode,
1058 nd->flags & LOOKUP_RCU);
1059 if (unlikely(error))
Al Viro6920a442015-05-10 10:43:46 -04001060 return ERR_PTR(error);
Al Viro36f3b4f2011-02-22 21:24:38 -05001061
Al Viro86acdca12009-12-22 23:45:11 -05001062 nd->last_type = LAST_BIND;
Eric Biggers4c4f7c12019-04-10 13:21:14 -07001063 res = READ_ONCE(inode->i_link);
Al Virod4dee482015-04-30 20:08:02 -04001064 if (!res) {
Al Virofceef392015-12-29 15:58:39 -05001065 const char * (*get)(struct dentry *, struct inode *,
1066 struct delayed_call *);
1067 get = inode->i_op->get_link;
Al Viro8c1b4562015-05-09 18:15:21 -04001068 if (nd->flags & LOOKUP_RCU) {
Al Virofceef392015-12-29 15:58:39 -05001069 res = get(NULL, inode, &last->done);
Al Viro6b255392015-11-17 10:20:54 -05001070 if (res == ERR_PTR(-ECHILD)) {
Al Viro4675ac32017-01-09 22:29:15 -05001071 if (unlikely(unlazy_walk(nd)))
Al Viro6b255392015-11-17 10:20:54 -05001072 return ERR_PTR(-ECHILD);
Al Virofceef392015-12-29 15:58:39 -05001073 res = get(dentry, inode, &last->done);
Al Viro6b255392015-11-17 10:20:54 -05001074 }
1075 } else {
Al Virofceef392015-12-29 15:58:39 -05001076 res = get(dentry, inode, &last->done);
Al Viro8c1b4562015-05-09 18:15:21 -04001077 }
Al Virofceef392015-12-29 15:58:39 -05001078 if (IS_ERR_OR_NULL(res))
Al Virofab51e82015-05-10 11:01:00 -04001079 return res;
Ian Kent051d3812006-03-27 01:14:53 -08001080 }
Al Virofab51e82015-05-10 11:01:00 -04001081 if (*res == '/') {
Al Viro9e6697e2015-12-05 20:07:21 -05001082 if (!nd->root.mnt)
1083 set_root(nd);
Al Viro248fb5b2015-12-05 20:51:58 -05001084 if (unlikely(nd_jump_root(nd)))
1085 return ERR_PTR(-ECHILD);
Al Virofab51e82015-05-10 11:01:00 -04001086 while (unlikely(*++res == '/'))
1087 ;
1088 }
1089 if (!*res)
1090 res = NULL;
Al Viro0a959df2015-04-18 18:23:41 -04001091 return res;
1092}
Al Viro6d7b5aa2012-06-10 04:15:17 -04001093
David Howellsf015f1262012-06-25 12:55:28 +01001094/*
1095 * follow_up - Find the mountpoint of path's vfsmount
1096 *
1097 * Given a path, find the mountpoint of its source file system.
1098 * Replace @path with the path of the mountpoint in the parent mount.
1099 * Up is towards /.
1100 *
1101 * Return 1 if we went up a level and 0 if we were already at the
1102 * root.
1103 */
Al Virobab77eb2009-04-18 03:26:48 -04001104int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105{
Al Viro0714a532011-11-24 22:19:58 -05001106 struct mount *mnt = real_mount(path->mnt);
1107 struct mount *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +10001109
Al Viro48a066e2013-09-29 22:06:07 -04001110 read_seqlock_excl(&mount_lock);
Al Viro0714a532011-11-24 22:19:58 -05001111 parent = mnt->mnt_parent;
Al Viro3c0a6162012-07-18 17:32:50 +04001112 if (parent == mnt) {
Al Viro48a066e2013-09-29 22:06:07 -04001113 read_sequnlock_excl(&mount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 return 0;
1115 }
Al Viro0714a532011-11-24 22:19:58 -05001116 mntget(&parent->mnt);
Al Viroa73324d2011-11-24 22:25:07 -05001117 mountpoint = dget(mnt->mnt_mountpoint);
Al Viro48a066e2013-09-29 22:06:07 -04001118 read_sequnlock_excl(&mount_lock);
Al Virobab77eb2009-04-18 03:26:48 -04001119 dput(path->dentry);
1120 path->dentry = mountpoint;
1121 mntput(path->mnt);
Al Viro0714a532011-11-24 22:19:58 -05001122 path->mnt = &parent->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 return 1;
1124}
Al Viro4d359502014-03-14 12:20:17 -04001125EXPORT_SYMBOL(follow_up);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001127/*
David Howells9875cf82011-01-14 18:45:21 +00001128 * Perform an automount
1129 * - return -EISDIR to tell follow_managed() to stop and return the path we
1130 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 */
NeilBrown756daf22015-03-23 13:37:38 +11001132static int follow_automount(struct path *path, struct nameidata *nd,
David Howells9875cf82011-01-14 18:45:21 +00001133 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +11001134{
David Howells9875cf82011-01-14 18:45:21 +00001135 struct vfsmount *mnt;
David Howellsea5b7782011-01-14 19:10:03 +00001136 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001137
David Howells9875cf82011-01-14 18:45:21 +00001138 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1139 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -07001140
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +02001141 /* We don't want to mount if someone's just doing a stat -
1142 * unless they're stat'ing a directory and appended a '/' to
1143 * the name.
1144 *
1145 * We do, however, want to mount if someone wants to open or
1146 * create a file of any type under the mountpoint, wants to
1147 * traverse through the mountpoint or wants to open the
1148 * mounted directory. Also, autofs may mark negative dentries
1149 * as being automount points. These will need the attentions
1150 * of the daemon to instantiate them before they can be used.
David Howells9875cf82011-01-14 18:45:21 +00001151 */
NeilBrown756daf22015-03-23 13:37:38 +11001152 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
Ian Kent5d38f042017-11-29 16:11:26 -08001153 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1154 path->dentry->d_inode)
1155 return -EISDIR;
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +02001156
NeilBrown756daf22015-03-23 13:37:38 +11001157 nd->total_link_count++;
1158 if (nd->total_link_count >= 40)
David Howells9875cf82011-01-14 18:45:21 +00001159 return -ELOOP;
1160
1161 mnt = path->dentry->d_op->d_automount(path);
1162 if (IS_ERR(mnt)) {
1163 /*
1164 * The filesystem is allowed to return -EISDIR here to indicate
1165 * it doesn't want to automount. For instance, autofs would do
1166 * this so that its userspace daemon can mount on this dentry.
1167 *
1168 * However, we can only permit this if it's a terminal point in
1169 * the path being looked up; if it wasn't then the remainder of
1170 * the path is inaccessible and we should say so.
1171 */
NeilBrown756daf22015-03-23 13:37:38 +11001172 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
David Howells9875cf82011-01-14 18:45:21 +00001173 return -EREMOTE;
1174 return PTR_ERR(mnt);
1175 }
David Howellsea5b7782011-01-14 19:10:03 +00001176
David Howells9875cf82011-01-14 18:45:21 +00001177 if (!mnt) /* mount collision */
1178 return 0;
1179
Al Viro8aef1882011-06-16 15:10:06 +01001180 if (!*need_mntput) {
1181 /* lock_mount() may release path->mnt on error */
1182 mntget(path->mnt);
1183 *need_mntput = true;
1184 }
Al Viro19a167a2011-01-17 01:35:23 -05001185 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +00001186
David Howellsea5b7782011-01-14 19:10:03 +00001187 switch (err) {
1188 case -EBUSY:
1189 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -05001190 return 0;
David Howellsea5b7782011-01-14 19:10:03 +00001191 case 0:
Al Viro8aef1882011-06-16 15:10:06 +01001192 path_put(path);
David Howellsea5b7782011-01-14 19:10:03 +00001193 path->mnt = mnt;
1194 path->dentry = dget(mnt->mnt_root);
David Howellsea5b7782011-01-14 19:10:03 +00001195 return 0;
Al Viro19a167a2011-01-17 01:35:23 -05001196 default:
1197 return err;
David Howellsea5b7782011-01-14 19:10:03 +00001198 }
Al Viro19a167a2011-01-17 01:35:23 -05001199
David Howells9875cf82011-01-14 18:45:21 +00001200}
1201
1202/*
1203 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +00001204 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +00001205 * - Flagged as mountpoint
1206 * - Flagged as automount point
1207 *
1208 * This may only be called in refwalk mode.
1209 *
1210 * Serialization is taken care of in namespace.c
1211 */
NeilBrown756daf22015-03-23 13:37:38 +11001212static int follow_managed(struct path *path, struct nameidata *nd)
David Howells9875cf82011-01-14 18:45:21 +00001213{
Al Viro8aef1882011-06-16 15:10:06 +01001214 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
David Howells9875cf82011-01-14 18:45:21 +00001215 unsigned managed;
1216 bool need_mntput = false;
Al Viro8aef1882011-06-16 15:10:06 +01001217 int ret = 0;
David Howells9875cf82011-01-14 18:45:21 +00001218
1219 /* Given that we're not holding a lock here, we retain the value in a
1220 * local variable for each dentry as we look at it so that we don't see
1221 * the components of that value change under us */
Mark Rutland6aa7de02017-10-23 14:07:29 -07001222 while (managed = READ_ONCE(path->dentry->d_flags),
David Howells9875cf82011-01-14 18:45:21 +00001223 managed &= DCACHE_MANAGED_DENTRY,
1224 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +00001225 /* Allow the filesystem to manage the transit without i_mutex
1226 * being held. */
1227 if (managed & DCACHE_MANAGE_TRANSIT) {
1228 BUG_ON(!path->dentry->d_op);
1229 BUG_ON(!path->dentry->d_op->d_manage);
Ian Kentfb5f51c2016-11-24 08:03:41 +11001230 ret = path->dentry->d_op->d_manage(path, false);
David Howellscc53ce52011-01-14 18:45:26 +00001231 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001232 break;
David Howellscc53ce52011-01-14 18:45:26 +00001233 }
1234
David Howells9875cf82011-01-14 18:45:21 +00001235 /* Transit to a mounted filesystem. */
1236 if (managed & DCACHE_MOUNTED) {
1237 struct vfsmount *mounted = lookup_mnt(path);
1238 if (mounted) {
1239 dput(path->dentry);
1240 if (need_mntput)
1241 mntput(path->mnt);
1242 path->mnt = mounted;
1243 path->dentry = dget(mounted->mnt_root);
1244 need_mntput = true;
1245 continue;
1246 }
1247
1248 /* Something is mounted on this dentry in another
1249 * namespace and/or whatever was mounted there in this
Al Viro48a066e2013-09-29 22:06:07 -04001250 * namespace got unmounted before lookup_mnt() could
1251 * get it */
David Howells9875cf82011-01-14 18:45:21 +00001252 }
1253
1254 /* Handle an automount point */
1255 if (managed & DCACHE_NEED_AUTOMOUNT) {
NeilBrown756daf22015-03-23 13:37:38 +11001256 ret = follow_automount(path, nd, &need_mntput);
David Howells9875cf82011-01-14 18:45:21 +00001257 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001258 break;
David Howells9875cf82011-01-14 18:45:21 +00001259 continue;
1260 }
1261
1262 /* We didn't change the current path point */
1263 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 }
Al Viro8aef1882011-06-16 15:10:06 +01001265
1266 if (need_mntput && path->mnt == mnt)
1267 mntput(path->mnt);
Al Viroe9742b52016-03-05 22:04:59 -05001268 if (ret == -EISDIR || !ret)
1269 ret = 1;
Al Viro8402752e2015-04-22 10:30:08 -04001270 if (need_mntput)
1271 nd->flags |= LOOKUP_JUMPED;
1272 if (unlikely(ret < 0))
1273 path_put_conditional(path, nd);
1274 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275}
1276
David Howellscc53ce52011-01-14 18:45:26 +00001277int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278{
1279 struct vfsmount *mounted;
1280
Al Viro1c755af2009-04-18 14:06:57 -04001281 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -04001283 dput(path->dentry);
1284 mntput(path->mnt);
1285 path->mnt = mounted;
1286 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 return 1;
1288 }
1289 return 0;
1290}
Al Viro4d359502014-03-14 12:20:17 -04001291EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292
Ian Kentfb5f51c2016-11-24 08:03:41 +11001293static inline int managed_dentry_rcu(const struct path *path)
Ian Kent62a73752011-03-25 01:51:02 +08001294{
Ian Kentfb5f51c2016-11-24 08:03:41 +11001295 return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1296 path->dentry->d_op->d_manage(path, true) : 0;
Ian Kent62a73752011-03-25 01:51:02 +08001297}
1298
David Howells9875cf82011-01-14 18:45:21 +00001299/*
Al Viro287548e2011-05-27 06:50:06 -04001300 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1301 * we meet a managed dentry that would need blocking.
David Howells9875cf82011-01-14 18:45:21 +00001302 */
1303static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
Al Viro254cf582015-05-05 09:40:46 -04001304 struct inode **inode, unsigned *seqp)
David Howells9875cf82011-01-14 18:45:21 +00001305{
Ian Kent62a73752011-03-25 01:51:02 +08001306 for (;;) {
Al Viroc7105362011-11-24 18:22:03 -05001307 struct mount *mounted;
Ian Kent62a73752011-03-25 01:51:02 +08001308 /*
1309 * Don't forget we might have a non-mountpoint managed dentry
1310 * that wants to block transit.
1311 */
Ian Kentfb5f51c2016-11-24 08:03:41 +11001312 switch (managed_dentry_rcu(path)) {
NeilBrownb8faf032014-08-04 17:06:29 +10001313 case -ECHILD:
1314 default:
David Howellsab909112011-01-14 18:46:51 +00001315 return false;
NeilBrownb8faf032014-08-04 17:06:29 +10001316 case -EISDIR:
1317 return true;
1318 case 0:
1319 break;
1320 }
Ian Kent62a73752011-03-25 01:51:02 +08001321
1322 if (!d_mountpoint(path->dentry))
NeilBrownb8faf032014-08-04 17:06:29 +10001323 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
Ian Kent62a73752011-03-25 01:51:02 +08001324
Al Viro474279d2013-10-01 16:11:26 -04001325 mounted = __lookup_mnt(path->mnt, path->dentry);
David Howells9875cf82011-01-14 18:45:21 +00001326 if (!mounted)
1327 break;
Al Viroc7105362011-11-24 18:22:03 -05001328 path->mnt = &mounted->mnt;
1329 path->dentry = mounted->mnt.mnt_root;
Al Viroa3fbbde2011-11-07 21:21:26 +00001330 nd->flags |= LOOKUP_JUMPED;
Al Viro254cf582015-05-05 09:40:46 -04001331 *seqp = read_seqcount_begin(&path->dentry->d_seq);
Linus Torvalds59430262011-07-18 15:43:29 -07001332 /*
1333 * Update the inode too. We don't need to re-check the
1334 * dentry sequence number here after this d_inode read,
1335 * because a mount-point is always pinned.
1336 */
1337 *inode = path->dentry->d_inode;
David Howells9875cf82011-01-14 18:45:21 +00001338 }
Al Virof5be3e29122014-09-13 21:50:45 -04001339 return !read_seqretry(&mount_lock, nd->m_seq) &&
NeilBrownb8faf032014-08-04 17:06:29 +10001340 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
Al Viro287548e2011-05-27 06:50:06 -04001341}
1342
Nick Piggin31e6b012011-01-07 17:49:52 +11001343static int follow_dotdot_rcu(struct nameidata *nd)
1344{
Al Viro4023bfc2014-09-13 21:59:43 -04001345 struct inode *inode = nd->inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001346
David Howells9875cf82011-01-14 18:45:21 +00001347 while (1) {
Al Viroaed434a2015-05-12 12:22:47 -04001348 if (path_equal(&nd->path, &nd->root))
Nick Piggin31e6b012011-01-07 17:49:52 +11001349 break;
Nick Piggin31e6b012011-01-07 17:49:52 +11001350 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1351 struct dentry *old = nd->path.dentry;
1352 struct dentry *parent = old->d_parent;
1353 unsigned seq;
1354
Al Viro4023bfc2014-09-13 21:59:43 -04001355 inode = parent->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001356 seq = read_seqcount_begin(&parent->d_seq);
Al Viroaed434a2015-05-12 12:22:47 -04001357 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1358 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001359 nd->path.dentry = parent;
1360 nd->seq = seq;
Eric W. Biederman397d4252015-08-15 20:27:13 -05001361 if (unlikely(!path_connected(&nd->path)))
1362 return -ENOENT;
Nick Piggin31e6b012011-01-07 17:49:52 +11001363 break;
Al Viroaed434a2015-05-12 12:22:47 -04001364 } else {
1365 struct mount *mnt = real_mount(nd->path.mnt);
1366 struct mount *mparent = mnt->mnt_parent;
1367 struct dentry *mountpoint = mnt->mnt_mountpoint;
1368 struct inode *inode2 = mountpoint->d_inode;
1369 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1370 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1371 return -ECHILD;
1372 if (&mparent->mnt == nd->path.mnt)
1373 break;
1374 /* we know that mountpoint was pinned */
1375 nd->path.dentry = mountpoint;
1376 nd->path.mnt = &mparent->mnt;
1377 inode = inode2;
1378 nd->seq = seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11001379 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001380 }
Al Viroaed434a2015-05-12 12:22:47 -04001381 while (unlikely(d_mountpoint(nd->path.dentry))) {
Al Virob37199e2014-03-20 15:18:22 -04001382 struct mount *mounted;
1383 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
Al Viroaed434a2015-05-12 12:22:47 -04001384 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1385 return -ECHILD;
Al Virob37199e2014-03-20 15:18:22 -04001386 if (!mounted)
1387 break;
1388 nd->path.mnt = &mounted->mnt;
1389 nd->path.dentry = mounted->mnt.mnt_root;
Al Viro4023bfc2014-09-13 21:59:43 -04001390 inode = nd->path.dentry->d_inode;
Al Virob37199e2014-03-20 15:18:22 -04001391 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Virob37199e2014-03-20 15:18:22 -04001392 }
Al Viro4023bfc2014-09-13 21:59:43 -04001393 nd->inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001394 return 0;
1395}
1396
David Howells9875cf82011-01-14 18:45:21 +00001397/*
David Howellscc53ce52011-01-14 18:45:26 +00001398 * Follow down to the covering mount currently visible to userspace. At each
1399 * point, the filesystem owning that dentry may be queried as to whether the
1400 * caller is permitted to proceed or not.
David Howellscc53ce52011-01-14 18:45:26 +00001401 */
Al Viro7cc90cc2011-03-18 09:04:20 -04001402int follow_down(struct path *path)
David Howellscc53ce52011-01-14 18:45:26 +00001403{
1404 unsigned managed;
1405 int ret;
1406
Mark Rutland6aa7de02017-10-23 14:07:29 -07001407 while (managed = READ_ONCE(path->dentry->d_flags),
David Howellscc53ce52011-01-14 18:45:26 +00001408 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1409 /* Allow the filesystem to manage the transit without i_mutex
1410 * being held.
1411 *
1412 * We indicate to the filesystem if someone is trying to mount
1413 * something here. This gives autofs the chance to deny anyone
1414 * other than its daemon the right to mount on its
1415 * superstructure.
1416 *
1417 * The filesystem may sleep at this point.
1418 */
1419 if (managed & DCACHE_MANAGE_TRANSIT) {
1420 BUG_ON(!path->dentry->d_op);
1421 BUG_ON(!path->dentry->d_op->d_manage);
Ian Kentfb5f51c2016-11-24 08:03:41 +11001422 ret = path->dentry->d_op->d_manage(path, false);
David Howellscc53ce52011-01-14 18:45:26 +00001423 if (ret < 0)
1424 return ret == -EISDIR ? 0 : ret;
1425 }
1426
1427 /* Transit to a mounted filesystem. */
1428 if (managed & DCACHE_MOUNTED) {
1429 struct vfsmount *mounted = lookup_mnt(path);
1430 if (!mounted)
1431 break;
1432 dput(path->dentry);
1433 mntput(path->mnt);
1434 path->mnt = mounted;
1435 path->dentry = dget(mounted->mnt_root);
1436 continue;
1437 }
1438
1439 /* Don't handle automount points here */
1440 break;
1441 }
1442 return 0;
1443}
Al Viro4d359502014-03-14 12:20:17 -04001444EXPORT_SYMBOL(follow_down);
David Howellscc53ce52011-01-14 18:45:26 +00001445
1446/*
David Howells9875cf82011-01-14 18:45:21 +00001447 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1448 */
1449static void follow_mount(struct path *path)
1450{
1451 while (d_mountpoint(path->dentry)) {
1452 struct vfsmount *mounted = lookup_mnt(path);
1453 if (!mounted)
1454 break;
1455 dput(path->dentry);
1456 mntput(path->mnt);
1457 path->mnt = mounted;
1458 path->dentry = dget(mounted->mnt_root);
1459 }
1460}
1461
Eric W. Biedermaneedf2652016-06-02 10:29:47 -05001462static int path_parent_directory(struct path *path)
1463{
1464 struct dentry *old = path->dentry;
1465 /* rare case of legitimate dget_parent()... */
1466 path->dentry = dget_parent(path->dentry);
1467 dput(old);
1468 if (unlikely(!path_connected(path)))
1469 return -ENOENT;
1470 return 0;
1471}
1472
Eric W. Biederman397d4252015-08-15 20:27:13 -05001473static int follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474{
1475 while(1) {
Danilo Krummrich030c7e02018-04-23 10:30:59 +02001476 if (path_equal(&nd->path, &nd->root))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 break;
Jan Blunck4ac91372008-02-14 19:34:32 -08001478 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Eric W. Biedermaneedf2652016-06-02 10:29:47 -05001479 int ret = path_parent_directory(&nd->path);
1480 if (ret)
1481 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 break;
1483 }
Al Viro3088dd72010-01-30 15:47:29 -05001484 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 }
Al Viro79ed0222009-04-18 13:59:41 -04001487 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001488 nd->inode = nd->path.dentry->d_inode;
Eric W. Biederman397d4252015-08-15 20:27:13 -05001489 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490}
1491
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492/*
Oleg Drokinf4fdace2016-07-07 22:04:04 -04001493 * This looks up the name in dcache and possibly revalidates the found dentry.
1494 * NULL is returned if the dentry does not exist in the cache.
Nick Pigginbaa03892010-08-18 04:37:31 +10001495 */
Al Viroe3c13922016-03-06 14:03:27 -05001496static struct dentry *lookup_dcache(const struct qstr *name,
1497 struct dentry *dir,
Al Viro6c51e512016-03-05 20:09:32 -05001498 unsigned int flags)
Nick Pigginbaa03892010-08-18 04:37:31 +10001499{
Al Viroa89f8332017-01-09 22:25:28 -05001500 struct dentry *dentry = d_lookup(dir, name);
Miklos Szeredibad61182012-03-26 12:54:24 +02001501 if (dentry) {
Al Viroa89f8332017-01-09 22:25:28 -05001502 int error = d_revalidate(dentry, flags);
1503 if (unlikely(error <= 0)) {
1504 if (!error)
1505 d_invalidate(dentry);
1506 dput(dentry);
1507 return ERR_PTR(error);
Miklos Szeredibad61182012-03-26 12:54:24 +02001508 }
1509 }
Nick Pigginbaa03892010-08-18 04:37:31 +10001510 return dentry;
1511}
1512
1513/*
Al Viroa03ece52018-03-08 11:00:45 -05001514 * Parent directory has inode locked exclusive. This is one
1515 * and only case when ->lookup() gets called on non in-lookup
1516 * dentries - as the matter of fact, this only gets called
1517 * when directory is guaranteed to have no in-lookup children
1518 * at all.
Josef Bacik44396f42011-05-31 11:58:49 -04001519 */
Al Viroa03ece52018-03-08 11:00:45 -05001520static struct dentry *__lookup_hash(const struct qstr *name,
1521 struct dentry *base, unsigned int flags)
Josef Bacik44396f42011-05-31 11:58:49 -04001522{
Al Viroa03ece52018-03-08 11:00:45 -05001523 struct dentry *dentry = lookup_dcache(name, base, flags);
Josef Bacik44396f42011-05-31 11:58:49 -04001524 struct dentry *old;
Al Viroa03ece52018-03-08 11:00:45 -05001525 struct inode *dir = base->d_inode;
1526
1527 if (dentry)
1528 return dentry;
Josef Bacik44396f42011-05-31 11:58:49 -04001529
1530 /* Don't create child dentry for a dead directory. */
Al Viroa03ece52018-03-08 11:00:45 -05001531 if (unlikely(IS_DEADDIR(dir)))
Josef Bacik44396f42011-05-31 11:58:49 -04001532 return ERR_PTR(-ENOENT);
Al Viroa03ece52018-03-08 11:00:45 -05001533
1534 dentry = d_alloc(base, name);
1535 if (unlikely(!dentry))
1536 return ERR_PTR(-ENOMEM);
Josef Bacik44396f42011-05-31 11:58:49 -04001537
Al Viro72bd8662012-06-10 17:17:17 -04001538 old = dir->i_op->lookup(dir, dentry, flags);
Josef Bacik44396f42011-05-31 11:58:49 -04001539 if (unlikely(old)) {
1540 dput(dentry);
1541 dentry = old;
1542 }
1543 return dentry;
1544}
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)) {
Al Viro4675ac32017-01-09 22:29:15 -05001565 if (unlazy_walk(nd))
Al Viro5d0f49c2016-03-05 21:32:53 -05001566 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 Viroa89f8332017-01-09 22:25:28 -05001590 status = d_revalidate(dentry, nd->flags);
Al Viro209a7fb2017-01-09 01:35:39 -05001591 if (likely(status > 0)) {
Al Viro5d0f49c2016-03-05 21:32:53 -05001592 /*
1593 * Note: do negative dentry check after revalidation in
1594 * case that drops it.
1595 */
1596 if (unlikely(negative))
1597 return -ENOENT;
1598 path->mnt = mnt;
1599 path->dentry = dentry;
1600 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
Al Viroe9742b52016-03-05 22:04:59 -05001601 return 1;
Al Viro24643082011-02-15 01:26:22 -05001602 }
Al Viro4675ac32017-01-09 22:29:15 -05001603 if (unlazy_child(nd, dentry, seq))
Al Viro209a7fb2017-01-09 01:35:39 -05001604 return -ECHILD;
1605 if (unlikely(status == -ECHILD))
1606 /* we'd been told to redo it in non-rcu mode */
1607 status = d_revalidate(dentry, nd->flags);
Al Viro5a18fff2011-03-11 04:44:53 -05001608 } else {
Al Viroe97cdc82013-01-24 18:16:00 -05001609 dentry = __d_lookup(parent, &nd->last);
Al Viro5d0f49c2016-03-05 21:32:53 -05001610 if (unlikely(!dentry))
Al Viroe9742b52016-03-05 22:04:59 -05001611 return 0;
Al Viroa89f8332017-01-09 22:25:28 -05001612 status = d_revalidate(dentry, nd->flags);
Nick Piggin31e6b012011-01-07 17:49:52 +11001613 }
Al Viro5a18fff2011-03-11 04:44:53 -05001614 if (unlikely(status <= 0)) {
Al Viroe9742b52016-03-05 22:04:59 -05001615 if (!status)
Al Viro5d0f49c2016-03-05 21:32:53 -05001616 d_invalidate(dentry);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001617 dput(dentry);
Al Viro5d0f49c2016-03-05 21:32:53 -05001618 return status;
Al Viro5a18fff2011-03-11 04:44:53 -05001619 }
Al Viro766c4cb2015-05-07 19:24:57 -04001620 if (unlikely(d_is_negative(dentry))) {
1621 dput(dentry);
1622 return -ENOENT;
1623 }
Al Viro5d0f49c2016-03-05 21:32:53 -05001624
David Howells9875cf82011-01-14 18:45:21 +00001625 path->mnt = mnt;
1626 path->dentry = dentry;
NeilBrown756daf22015-03-23 13:37:38 +11001627 err = follow_managed(path, nd);
Al Viroe9742b52016-03-05 22:04:59 -05001628 if (likely(err > 0))
David Howells63afdfc2015-05-06 15:59:00 +01001629 *inode = d_backing_inode(path->dentry);
Al Viro8402752e2015-04-22 10:30:08 -04001630 return err;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001631}
1632
1633/* Fast lookup failed, do it the slow way */
Al Viro88d83312018-04-06 16:43:47 -04001634static struct dentry *__lookup_slow(const struct qstr *name,
1635 struct dentry *dir,
1636 unsigned int flags)
Miklos Szeredi697f5142012-05-21 17:30:05 +02001637{
Al Viro88d83312018-04-06 16:43:47 -04001638 struct dentry *dentry, *old;
Al Viro19363862016-04-14 19:33:34 -04001639 struct inode *inode = dir->d_inode;
Al Virod9171b92016-04-15 03:33:13 -04001640 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
Al Viro19363862016-04-14 19:33:34 -04001641
Al Viro19363862016-04-14 19:33:34 -04001642 /* Don't go there if it's already dead */
Al Viro94bdd652016-04-15 02:42:04 -04001643 if (unlikely(IS_DEADDIR(inode)))
Al Viro88d83312018-04-06 16:43:47 -04001644 return ERR_PTR(-ENOENT);
Al Viro94bdd652016-04-15 02:42:04 -04001645again:
Al Virod9171b92016-04-15 03:33:13 -04001646 dentry = d_alloc_parallel(dir, name, &wq);
Al Viro94bdd652016-04-15 02:42:04 -04001647 if (IS_ERR(dentry))
Al Viro88d83312018-04-06 16:43:47 -04001648 return dentry;
Al Viro94bdd652016-04-15 02:42:04 -04001649 if (unlikely(!d_in_lookup(dentry))) {
Al Viroa89f8332017-01-09 22:25:28 -05001650 if (!(flags & LOOKUP_NO_REVAL)) {
Al Viro949a8522016-03-06 14:20:52 -05001651 int error = d_revalidate(dentry, flags);
1652 if (unlikely(error <= 0)) {
Al Viro94bdd652016-04-15 02:42:04 -04001653 if (!error) {
Al Viro949a8522016-03-06 14:20:52 -05001654 d_invalidate(dentry);
Al Viro94bdd652016-04-15 02:42:04 -04001655 dput(dentry);
1656 goto again;
1657 }
Al Viro949a8522016-03-06 14:20:52 -05001658 dput(dentry);
1659 dentry = ERR_PTR(error);
1660 }
1661 }
Al Viro94bdd652016-04-15 02:42:04 -04001662 } else {
1663 old = inode->i_op->lookup(inode, dentry, flags);
1664 d_lookup_done(dentry);
1665 if (unlikely(old)) {
1666 dput(dentry);
1667 dentry = old;
Al Viro949a8522016-03-06 14:20:52 -05001668 }
1669 }
Al Viroe3c13922016-03-06 14:03:27 -05001670 return dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671}
1672
Al Viro88d83312018-04-06 16:43:47 -04001673static struct dentry *lookup_slow(const struct qstr *name,
1674 struct dentry *dir,
1675 unsigned int flags)
1676{
1677 struct inode *inode = dir->d_inode;
1678 struct dentry *res;
1679 inode_lock_shared(inode);
1680 res = __lookup_slow(name, dir, flags);
1681 inode_unlock_shared(inode);
1682 return res;
1683}
1684
Al Viro52094c82011-02-21 21:34:47 -05001685static inline int may_lookup(struct nameidata *nd)
1686{
1687 if (nd->flags & LOOKUP_RCU) {
Al Viro4ad5abb2011-06-20 19:57:03 -04001688 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
Al Viro52094c82011-02-21 21:34:47 -05001689 if (err != -ECHILD)
1690 return err;
Al Viro4675ac32017-01-09 22:29:15 -05001691 if (unlazy_walk(nd))
Al Viro52094c82011-02-21 21:34:47 -05001692 return -ECHILD;
1693 }
Al Viro4ad5abb2011-06-20 19:57:03 -04001694 return inode_permission(nd->inode, MAY_EXEC);
Al Viro52094c82011-02-21 21:34:47 -05001695}
1696
Al Viro9856fa12011-03-04 14:22:06 -05001697static inline int handle_dots(struct nameidata *nd, int type)
1698{
1699 if (type == LAST_DOTDOT) {
Al Viro9e6697e2015-12-05 20:07:21 -05001700 if (!nd->root.mnt)
1701 set_root(nd);
Al Viro9856fa12011-03-04 14:22:06 -05001702 if (nd->flags & LOOKUP_RCU) {
Al Viro70291ae2015-05-04 07:53:00 -04001703 return follow_dotdot_rcu(nd);
Al Viro9856fa12011-03-04 14:22:06 -05001704 } else
Eric W. Biederman397d4252015-08-15 20:27:13 -05001705 return follow_dotdot(nd);
Al Viro9856fa12011-03-04 14:22:06 -05001706 }
1707 return 0;
1708}
1709
Al Viro181548c2015-05-07 19:54:34 -04001710static int pick_link(struct nameidata *nd, struct path *link,
1711 struct inode *inode, unsigned seq)
Al Virod63ff282015-05-04 18:13:23 -04001712{
Al Viro626de992015-05-04 18:26:59 -04001713 int error;
Al Viro1cf26652015-05-06 16:01:56 -04001714 struct saved *last;
NeilBrown756daf22015-03-23 13:37:38 +11001715 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
Al Viro626de992015-05-04 18:26:59 -04001716 path_to_nameidata(link, nd);
1717 return -ELOOP;
1718 }
Al Virobc40aee2015-05-09 13:04:24 -04001719 if (!(nd->flags & LOOKUP_RCU)) {
Al Viro7973387a2015-05-09 12:55:43 -04001720 if (link->mnt == nd->path.mnt)
1721 mntget(link->mnt);
Al Virod63ff282015-05-04 18:13:23 -04001722 }
Al Viro626de992015-05-04 18:26:59 -04001723 error = nd_alloc_stack(nd);
1724 if (unlikely(error)) {
Al Virobc40aee2015-05-09 13:04:24 -04001725 if (error == -ECHILD) {
Al Viroad1633a2017-01-08 22:35:31 -05001726 if (unlikely(!legitimize_path(nd, link, seq))) {
1727 drop_links(nd);
1728 nd->depth = 0;
1729 nd->flags &= ~LOOKUP_RCU;
1730 nd->path.mnt = NULL;
1731 nd->path.dentry = NULL;
Al Viroad1633a2017-01-08 22:35:31 -05001732 rcu_read_unlock();
Al Viro4675ac32017-01-09 22:29:15 -05001733 } else if (likely(unlazy_walk(nd)) == 0)
Al Viroad1633a2017-01-08 22:35:31 -05001734 error = nd_alloc_stack(nd);
Al Virobc40aee2015-05-09 13:04:24 -04001735 }
1736 if (error) {
1737 path_put(link);
1738 return error;
1739 }
Al Viro626de992015-05-04 18:26:59 -04001740 }
1741
Al Viroab104922015-05-10 11:50:01 -04001742 last = nd->stack + nd->depth++;
Al Viro1cf26652015-05-06 16:01:56 -04001743 last->link = *link;
Al Virofceef392015-12-29 15:58:39 -05001744 clear_delayed_call(&last->done);
1745 nd->link_inode = inode;
Al Viro0450b2d2015-05-08 13:23:53 -04001746 last->seq = seq;
Al Virod63ff282015-05-04 18:13:23 -04001747 return 1;
1748}
1749
Al Viro8f64fb12016-11-14 01:50:26 -05001750enum {WALK_FOLLOW = 1, WALK_MORE = 2};
Al Viro31d66bc2016-11-14 01:43:34 -05001751
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001752/*
1753 * Do we need to follow links? We _really_ want to be able
1754 * to do this check without having to look at inode->i_op,
1755 * so we keep a cache of "no, this doesn't need follow_link"
1756 * for the common case.
1757 */
Al Viro8f64fb12016-11-14 01:50:26 -05001758static inline int step_into(struct nameidata *nd, struct path *path,
1759 int flags, struct inode *inode, unsigned seq)
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001760{
Al Viro31d66bc2016-11-14 01:43:34 -05001761 if (!(flags & WALK_MORE) && nd->depth)
1762 put_link(nd);
Al Viro8f64fb12016-11-14 01:50:26 -05001763 if (likely(!d_is_symlink(path->dentry)) ||
1764 !(flags & WALK_FOLLOW || nd->flags & LOOKUP_FOLLOW)) {
1765 /* not a symlink or should not follow */
1766 path_to_nameidata(path, nd);
1767 nd->inode = inode;
1768 nd->seq = seq;
Al Virod63ff282015-05-04 18:13:23 -04001769 return 0;
Al Viro8f64fb12016-11-14 01:50:26 -05001770 }
Al Viroa7f77542016-02-27 19:31:01 -05001771 /* make sure that d_is_symlink above matches inode */
1772 if (nd->flags & LOOKUP_RCU) {
Al Viro8f64fb12016-11-14 01:50:26 -05001773 if (read_seqcount_retry(&path->dentry->d_seq, seq))
Al Viroa7f77542016-02-27 19:31:01 -05001774 return -ECHILD;
1775 }
Al Viro8f64fb12016-11-14 01:50:26 -05001776 return pick_link(nd, path, inode, seq);
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001777}
1778
Al Viro4693a542015-05-04 17:47:11 -04001779static int walk_component(struct nameidata *nd, int flags)
Al Viroce57dfc2011-03-13 19:58:58 -04001780{
Al Virocaa856342015-04-22 17:52:47 -04001781 struct path path;
Al Viroce57dfc2011-03-13 19:58:58 -04001782 struct inode *inode;
Al Viro254cf582015-05-05 09:40:46 -04001783 unsigned seq;
Al Viroce57dfc2011-03-13 19:58:58 -04001784 int err;
1785 /*
1786 * "." and ".." are special - ".." especially so because it has
1787 * to be able to know about the current root directory and
1788 * parent relationships.
1789 */
Al Viro4693a542015-05-04 17:47:11 -04001790 if (unlikely(nd->last_type != LAST_NORM)) {
1791 err = handle_dots(nd, nd->last_type);
Al Viro1c4ff1a2016-11-14 01:39:36 -05001792 if (!(flags & WALK_MORE) && nd->depth)
Al Viro4693a542015-05-04 17:47:11 -04001793 put_link(nd);
1794 return err;
1795 }
Al Viro254cf582015-05-05 09:40:46 -04001796 err = lookup_fast(nd, &path, &inode, &seq);
Al Viroe9742b52016-03-05 22:04:59 -05001797 if (unlikely(err <= 0)) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001798 if (err < 0)
Al Virof0a9ba72015-05-04 07:59:30 -04001799 return err;
Al Viroe3c13922016-03-06 14:03:27 -05001800 path.dentry = lookup_slow(&nd->last, nd->path.dentry,
1801 nd->flags);
1802 if (IS_ERR(path.dentry))
1803 return PTR_ERR(path.dentry);
Al Viro7500c382016-03-31 00:23:05 -04001804
Al Viroe3c13922016-03-06 14:03:27 -05001805 path.mnt = nd->path.mnt;
1806 err = follow_managed(&path, nd);
1807 if (unlikely(err < 0))
Al Virof0a9ba72015-05-04 07:59:30 -04001808 return err;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001809
Al Viro7500c382016-03-31 00:23:05 -04001810 if (unlikely(d_is_negative(path.dentry))) {
1811 path_to_nameidata(&path, nd);
1812 return -ENOENT;
1813 }
1814
Al Viro254cf582015-05-05 09:40:46 -04001815 seq = 0; /* we are already out of RCU mode */
Al Virod4565642016-02-27 19:23:16 -05001816 inode = d_backing_inode(path.dentry);
Al Viroce57dfc2011-03-13 19:58:58 -04001817 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001818
Al Viro8f64fb12016-11-14 01:50:26 -05001819 return step_into(nd, &path, flags, inode, seq);
Al Viroce57dfc2011-03-13 19:58:58 -04001820}
1821
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822/*
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001823 * We can do the critical dentry name comparison and hashing
1824 * operations one word at a time, but we are limited to:
1825 *
1826 * - Architectures with fast unaligned word accesses. We could
1827 * do a "get_unaligned()" if this helps and is sufficiently
1828 * fast.
1829 *
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001830 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1831 * do not trap on the (extremely unlikely) case of a page
1832 * crossing operation.
1833 *
1834 * - Furthermore, we need an efficient 64-bit compile for the
1835 * 64-bit case in order to generate the "number of bytes in
1836 * the final mask". Again, that could be replaced with a
1837 * efficient population count instruction or similar.
1838 */
1839#ifdef CONFIG_DCACHE_WORD_ACCESS
1840
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001841#include <asm/word-at-a-time.h>
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001842
George Spelvin468a9422016-05-26 22:11:51 -04001843#ifdef HASH_MIX
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001844
George Spelvin468a9422016-05-26 22:11:51 -04001845/* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1846
1847#elif defined(CONFIG_64BIT)
George Spelvin2a18da7a2016-05-23 07:43:58 -04001848/*
1849 * Register pressure in the mixing function is an issue, particularly
1850 * on 32-bit x86, but almost any function requires one state value and
1851 * one temporary. Instead, use a function designed for two state values
1852 * and no temporaries.
1853 *
1854 * This function cannot create a collision in only two iterations, so
1855 * we have two iterations to achieve avalanche. In those two iterations,
1856 * we have six layers of mixing, which is enough to spread one bit's
1857 * influence out to 2^6 = 64 state bits.
1858 *
1859 * Rotate constants are scored by considering either 64 one-bit input
1860 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1861 * probability of that delta causing a change to each of the 128 output
1862 * bits, using a sample of random initial states.
1863 *
1864 * The Shannon entropy of the computed probabilities is then summed
1865 * to produce a score. Ideally, any input change has a 50% chance of
1866 * toggling any given output bit.
1867 *
1868 * Mixing scores (in bits) for (12,45):
1869 * Input delta: 1-bit 2-bit
1870 * 1 round: 713.3 42542.6
1871 * 2 rounds: 2753.7 140389.8
1872 * 3 rounds: 5954.1 233458.2
1873 * 4 rounds: 7862.6 256672.2
1874 * Perfect: 8192 258048
1875 * (64*128) (64*63/2 * 128)
1876 */
1877#define HASH_MIX(x, y, a) \
1878 ( x ^= (a), \
1879 y ^= x, x = rol64(x,12),\
1880 x += y, y = rol64(y,45),\
1881 y *= 9 )
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001882
George Spelvin0fed3ac2016-05-02 06:31:01 -04001883/*
George Spelvin2a18da7a2016-05-23 07:43:58 -04001884 * Fold two longs into one 32-bit hash value. This must be fast, but
1885 * latency isn't quite as critical, as there is a fair bit of additional
1886 * work done before the hash value is used.
George Spelvin0fed3ac2016-05-02 06:31:01 -04001887 */
George Spelvin2a18da7a2016-05-23 07:43:58 -04001888static inline unsigned int fold_hash(unsigned long x, unsigned long y)
George Spelvin0fed3ac2016-05-02 06:31:01 -04001889{
George Spelvin2a18da7a2016-05-23 07:43:58 -04001890 y ^= x * GOLDEN_RATIO_64;
1891 y *= GOLDEN_RATIO_64;
1892 return y >> 32;
George Spelvin0fed3ac2016-05-02 06:31:01 -04001893}
1894
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001895#else /* 32-bit case */
1896
George Spelvin2a18da7a2016-05-23 07:43:58 -04001897/*
1898 * Mixing scores (in bits) for (7,20):
1899 * Input delta: 1-bit 2-bit
1900 * 1 round: 330.3 9201.6
1901 * 2 rounds: 1246.4 25475.4
1902 * 3 rounds: 1907.1 31295.1
1903 * 4 rounds: 2042.3 31718.6
1904 * Perfect: 2048 31744
1905 * (32*64) (32*31/2 * 64)
1906 */
1907#define HASH_MIX(x, y, a) \
1908 ( x ^= (a), \
1909 y ^= x, x = rol32(x, 7),\
1910 x += y, y = rol32(y,20),\
1911 y *= 9 )
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001912
George Spelvin2a18da7a2016-05-23 07:43:58 -04001913static inline unsigned int fold_hash(unsigned long x, unsigned long y)
George Spelvin0fed3ac2016-05-02 06:31:01 -04001914{
George Spelvin2a18da7a2016-05-23 07:43:58 -04001915 /* Use arch-optimized multiply if one exists */
1916 return __hash_32(y ^ __hash_32(x));
George Spelvin0fed3ac2016-05-02 06:31:01 -04001917}
1918
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001919#endif
1920
George Spelvin2a18da7a2016-05-23 07:43:58 -04001921/*
1922 * Return the hash of a string of known length. This is carfully
1923 * designed to match hash_name(), which is the more critical function.
1924 * In particular, we must end by hashing a final word containing 0..7
1925 * payload bytes, to match the way that hash_name() iterates until it
1926 * finds the delimiter after the name.
1927 */
Linus Torvalds8387ff22016-06-10 07:51:30 -07001928unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001929{
Linus Torvalds8387ff22016-06-10 07:51:30 -07001930 unsigned long a, x = 0, y = (unsigned long)salt;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001931
1932 for (;;) {
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001933 if (!len)
1934 goto done;
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001935 a = load_unaligned_zeropad(name);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001936 if (len < sizeof(unsigned long))
1937 break;
George Spelvin2a18da7a2016-05-23 07:43:58 -04001938 HASH_MIX(x, y, a);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001939 name += sizeof(unsigned long);
1940 len -= sizeof(unsigned long);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001941 }
George Spelvin2a18da7a2016-05-23 07:43:58 -04001942 x ^= a & bytemask_from_count(len);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001943done:
George Spelvin2a18da7a2016-05-23 07:43:58 -04001944 return fold_hash(x, y);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001945}
1946EXPORT_SYMBOL(full_name_hash);
1947
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001948/* Return the "hash_len" (hash and length) of a null-terminated string */
Linus Torvalds8387ff22016-06-10 07:51:30 -07001949u64 hashlen_string(const void *salt, const char *name)
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001950{
Linus Torvalds8387ff22016-06-10 07:51:30 -07001951 unsigned long a = 0, x = 0, y = (unsigned long)salt;
1952 unsigned long adata, mask, len;
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001953 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1954
Linus Torvalds8387ff22016-06-10 07:51:30 -07001955 len = 0;
1956 goto inside;
1957
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001958 do {
George Spelvin2a18da7a2016-05-23 07:43:58 -04001959 HASH_MIX(x, y, a);
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001960 len += sizeof(unsigned long);
Linus Torvalds8387ff22016-06-10 07:51:30 -07001961inside:
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001962 a = load_unaligned_zeropad(name+len);
1963 } while (!has_zero(a, &adata, &constants));
1964
1965 adata = prep_zero_mask(a, adata, &constants);
1966 mask = create_zero_mask(adata);
George Spelvin2a18da7a2016-05-23 07:43:58 -04001967 x ^= a & zero_bytemask(mask);
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001968
George Spelvin2a18da7a2016-05-23 07:43:58 -04001969 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001970}
1971EXPORT_SYMBOL(hashlen_string);
1972
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001973/*
1974 * Calculate the length and hash of the path component, and
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001975 * return the "hash_len" as the result.
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001976 */
Linus Torvalds8387ff22016-06-10 07:51:30 -07001977static inline u64 hash_name(const void *salt, const char *name)
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001978{
Linus Torvalds8387ff22016-06-10 07:51:30 -07001979 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
1980 unsigned long adata, bdata, mask, len;
Linus Torvalds36126f82012-05-26 10:43:17 -07001981 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001982
Linus Torvalds8387ff22016-06-10 07:51:30 -07001983 len = 0;
1984 goto inside;
1985
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001986 do {
George Spelvin2a18da7a2016-05-23 07:43:58 -04001987 HASH_MIX(x, y, a);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001988 len += sizeof(unsigned long);
Linus Torvalds8387ff22016-06-10 07:51:30 -07001989inside:
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001990 a = load_unaligned_zeropad(name+len);
Linus Torvalds36126f82012-05-26 10:43:17 -07001991 b = a ^ REPEAT_BYTE('/');
1992 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001993
Linus Torvalds36126f82012-05-26 10:43:17 -07001994 adata = prep_zero_mask(a, adata, &constants);
1995 bdata = prep_zero_mask(b, bdata, &constants);
Linus Torvalds36126f82012-05-26 10:43:17 -07001996 mask = create_zero_mask(adata | bdata);
George Spelvin2a18da7a2016-05-23 07:43:58 -04001997 x ^= a & zero_bytemask(mask);
Linus Torvalds36126f82012-05-26 10:43:17 -07001998
George Spelvin2a18da7a2016-05-23 07:43:58 -04001999 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08002000}
2001
George Spelvin2a18da7a2016-05-23 07:43:58 -04002002#else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08002003
George Spelvinfcfd2fb2016-05-20 08:41:37 -04002004/* Return the hash of a string of known length */
Linus Torvalds8387ff22016-06-10 07:51:30 -07002005unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
Linus Torvalds0145acc2012-03-02 14:32:59 -08002006{
Linus Torvalds8387ff22016-06-10 07:51:30 -07002007 unsigned long hash = init_name_hash(salt);
Linus Torvalds0145acc2012-03-02 14:32:59 -08002008 while (len--)
George Spelvinfcfd2fb2016-05-20 08:41:37 -04002009 hash = partial_name_hash((unsigned char)*name++, hash);
Linus Torvalds0145acc2012-03-02 14:32:59 -08002010 return end_name_hash(hash);
2011}
Linus Torvaldsae942ae2012-03-02 19:40:57 -08002012EXPORT_SYMBOL(full_name_hash);
Linus Torvalds0145acc2012-03-02 14:32:59 -08002013
George Spelvinfcfd2fb2016-05-20 08:41:37 -04002014/* Return the "hash_len" (hash and length) of a null-terminated string */
Linus Torvalds8387ff22016-06-10 07:51:30 -07002015u64 hashlen_string(const void *salt, const char *name)
George Spelvinfcfd2fb2016-05-20 08:41:37 -04002016{
Linus Torvalds8387ff22016-06-10 07:51:30 -07002017 unsigned long hash = init_name_hash(salt);
George Spelvinfcfd2fb2016-05-20 08:41:37 -04002018 unsigned long len = 0, c;
2019
2020 c = (unsigned char)*name;
George Spelvine0ab7af2016-05-29 08:05:56 -04002021 while (c) {
George Spelvinfcfd2fb2016-05-20 08:41:37 -04002022 len++;
2023 hash = partial_name_hash(c, hash);
2024 c = (unsigned char)name[len];
George Spelvine0ab7af2016-05-29 08:05:56 -04002025 }
George Spelvinfcfd2fb2016-05-20 08:41:37 -04002026 return hashlen_create(end_name_hash(hash), len);
2027}
George Spelvinf2a031b2016-05-29 01:26:41 -04002028EXPORT_SYMBOL(hashlen_string);
George Spelvinfcfd2fb2016-05-20 08:41:37 -04002029
Linus Torvalds3ddcd052011-08-06 22:45:50 -07002030/*
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002031 * We know there's a real path component here of at least
2032 * one character.
2033 */
Linus Torvalds8387ff22016-06-10 07:51:30 -07002034static inline u64 hash_name(const void *salt, const char *name)
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002035{
Linus Torvalds8387ff22016-06-10 07:51:30 -07002036 unsigned long hash = init_name_hash(salt);
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002037 unsigned long len = 0, c;
2038
2039 c = (unsigned char)*name;
2040 do {
2041 len++;
2042 hash = partial_name_hash(c, hash);
2043 c = (unsigned char)name[len];
2044 } while (c && c != '/');
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002045 return hashlen_create(end_name_hash(hash), len);
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002046}
2047
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08002048#endif
2049
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002050/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01002052 * This is the basic name resolution function, turning a pathname into
2053 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01002055 * Returns 0 and nd will have valid dentry and mnt on success.
2056 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057 */
Al Viro6de88d72009-08-09 01:41:57 +04002058static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060 int err;
Al Viro32cd7462015-04-18 20:30:49 -04002061
Al Viro9b5858e2018-07-09 16:33:23 -04002062 if (IS_ERR(name))
2063 return PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064 while (*name=='/')
2065 name++;
2066 if (!*name)
Al Viro9e18f102015-04-18 20:44:34 -04002067 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069 /* At this point we know we have a real path component. */
2070 for(;;) {
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002071 u64 hash_len;
Al Virofe479a52011-02-22 15:10:03 -05002072 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073
Al Viro52094c82011-02-21 21:34:47 -05002074 err = may_lookup(nd);
George Spelvin2a18da7a2016-05-23 07:43:58 -04002075 if (err)
Al Viro3595e232015-05-09 16:54:45 -04002076 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077
Linus Torvalds8387ff22016-06-10 07:51:30 -07002078 hash_len = hash_name(nd->path.dentry, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079
Al Virofe479a52011-02-22 15:10:03 -05002080 type = LAST_NORM;
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002081 if (name[0] == '.') switch (hashlen_len(hash_len)) {
Al Virofe479a52011-02-22 15:10:03 -05002082 case 2:
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002083 if (name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05002084 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05002085 nd->flags |= LOOKUP_JUMPED;
2086 }
Al Virofe479a52011-02-22 15:10:03 -05002087 break;
2088 case 1:
2089 type = LAST_DOT;
2090 }
Al Viro5a202bc2011-03-08 14:17:44 -05002091 if (likely(type == LAST_NORM)) {
2092 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05002093 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05002094 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
James Hogana060dc52014-09-16 13:07:35 +01002095 struct qstr this = { { .hash_len = hash_len }, .name = name };
Linus Torvaldsda53be12013-05-21 15:22:44 -07002096 err = parent->d_op->d_hash(parent, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05002097 if (err < 0)
Al Viro3595e232015-05-09 16:54:45 -04002098 return err;
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002099 hash_len = this.hash_len;
2100 name = this.name;
Al Viro5a202bc2011-03-08 14:17:44 -05002101 }
2102 }
Al Virofe479a52011-02-22 15:10:03 -05002103
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002104 nd->last.hash_len = hash_len;
2105 nd->last.name = name;
Al Viro5f4a6a62013-01-24 18:04:22 -05002106 nd->last_type = type;
2107
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002108 name += hashlen_len(hash_len);
2109 if (!*name)
Al Virobdf6cbf2015-04-18 20:21:40 -04002110 goto OK;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002111 /*
2112 * If it wasn't NUL, we know it was '/'. Skip that
2113 * slash, and continue until no more slashes.
2114 */
2115 do {
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002116 name++;
2117 } while (unlikely(*name == '/'));
Al Viro8620c232015-05-04 08:58:35 -04002118 if (unlikely(!*name)) {
2119OK:
Al Viro368ee9b2015-05-08 17:19:59 -04002120 /* pathname body, done */
Al Viro8620c232015-05-04 08:58:35 -04002121 if (!nd->depth)
2122 return 0;
2123 name = nd->stack[nd->depth - 1].name;
Al Viro368ee9b2015-05-08 17:19:59 -04002124 /* trailing symlink, done */
Al Viro8620c232015-05-04 08:58:35 -04002125 if (!name)
2126 return 0;
2127 /* last component of nested symlink */
Al Viro8f64fb12016-11-14 01:50:26 -05002128 err = walk_component(nd, WALK_FOLLOW);
Al Viro1c4ff1a2016-11-14 01:39:36 -05002129 } else {
2130 /* not the last component */
Al Viro8f64fb12016-11-14 01:50:26 -05002131 err = walk_component(nd, WALK_FOLLOW | WALK_MORE);
Al Viro8620c232015-05-04 08:58:35 -04002132 }
Al Viroce57dfc2011-03-13 19:58:58 -04002133 if (err < 0)
Al Viro3595e232015-05-09 16:54:45 -04002134 return err;
Al Virofe479a52011-02-22 15:10:03 -05002135
Al Viroce57dfc2011-03-13 19:58:58 -04002136 if (err) {
Al Viro626de992015-05-04 18:26:59 -04002137 const char *s = get_link(nd);
Al Viro5a460272015-04-17 23:44:45 -04002138
Viresh Kumara1c83682015-08-12 15:59:44 +05302139 if (IS_ERR(s))
Al Viro3595e232015-05-09 16:54:45 -04002140 return PTR_ERR(s);
Al Virod40bcc02015-04-18 20:03:03 -04002141 err = 0;
2142 if (unlikely(!s)) {
2143 /* jumped */
Al Virob9ff4422015-05-02 20:19:23 -04002144 put_link(nd);
Al Virod40bcc02015-04-18 20:03:03 -04002145 } else {
Al Virofab51e82015-05-10 11:01:00 -04002146 nd->stack[nd->depth - 1].name = name;
2147 name = s;
2148 continue;
Al Virod40bcc02015-04-18 20:03:03 -04002149 }
Nick Piggin31e6b012011-01-07 17:49:52 +11002150 }
Al Viro97242f92015-08-01 19:59:28 -04002151 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2152 if (nd->flags & LOOKUP_RCU) {
Al Viro4675ac32017-01-09 22:29:15 -05002153 if (unlazy_walk(nd))
Al Viro97242f92015-08-01 19:59:28 -04002154 return -ECHILD;
2155 }
Al Viro3595e232015-05-09 16:54:45 -04002156 return -ENOTDIR;
Al Viro97242f92015-08-01 19:59:28 -04002157 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159}
2160
Al Viroedc2b1d2018-07-09 16:27:23 -04002161/* must be paired with terminate_walk() */
Al Viroc8a53ee2015-05-12 18:43:07 -04002162static const char *path_init(struct nameidata *nd, unsigned flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163{
Al Viroc8a53ee2015-05-12 18:43:07 -04002164 const char *s = nd->name->name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165
Linus Torvaldsc0eb0272017-04-02 17:10:08 -07002166 if (!*s)
2167 flags &= ~LOOKUP_RCU;
Al Viroedc2b1d2018-07-09 16:27:23 -04002168 if (flags & LOOKUP_RCU)
2169 rcu_read_lock();
Linus Torvaldsc0eb0272017-04-02 17:10:08 -07002170
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro980f3ea2014-11-20 14:20:24 -05002172 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173 nd->depth = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05002174 if (flags & LOOKUP_ROOT) {
David Howellsb18825a2013-09-12 19:22:53 +01002175 struct dentry *root = nd->root.dentry;
2176 struct inode *inode = root->d_inode;
Al Viro93893862017-04-15 17:29:14 -04002177 if (*s && unlikely(!d_can_lookup(root)))
2178 return ERR_PTR(-ENOTDIR);
Al Viro5b6ca022011-03-09 23:04:47 -05002179 nd->path = nd->root;
2180 nd->inode = inode;
2181 if (flags & LOOKUP_RCU) {
Al Viro5b6ca022011-03-09 23:04:47 -05002182 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro8f47a0162015-05-09 19:02:01 -04002183 nd->root_seq = nd->seq;
Al Viro48a066e2013-09-29 22:06:07 -04002184 nd->m_seq = read_seqbegin(&mount_lock);
Al Viro5b6ca022011-03-09 23:04:47 -05002185 } else {
2186 path_get(&nd->path);
2187 }
Al Viro368ee9b2015-05-08 17:19:59 -04002188 return s;
Al Viro5b6ca022011-03-09 23:04:47 -05002189 }
2190
Al Viro2a737872009-04-07 11:49:53 -04002191 nd->root.mnt = NULL;
Al Viro248fb5b2015-12-05 20:51:58 -05002192 nd->path.mnt = NULL;
2193 nd->path.dentry = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194
Al Viro48a066e2013-09-29 22:06:07 -04002195 nd->m_seq = read_seqbegin(&mount_lock);
Al Virofd2f7cb2015-02-22 20:07:13 -05002196 if (*s == '/') {
Al Viro9e6697e2015-12-05 20:07:21 -05002197 set_root(nd);
Al Viro248fb5b2015-12-05 20:51:58 -05002198 if (likely(!nd_jump_root(nd)))
Al Viroef55d912015-12-05 20:25:06 -05002199 return s;
Al Viroef55d912015-12-05 20:25:06 -05002200 return ERR_PTR(-ECHILD);
Al Viroc8a53ee2015-05-12 18:43:07 -04002201 } else if (nd->dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05002202 if (flags & LOOKUP_RCU) {
2203 struct fs_struct *fs = current->fs;
2204 unsigned seq;
2205
Al Viroe41f7d42011-02-22 14:02:58 -05002206 do {
2207 seq = read_seqcount_begin(&fs->seq);
2208 nd->path = fs->pwd;
Al Viroef55d912015-12-05 20:25:06 -05002209 nd->inode = nd->path.dentry->d_inode;
Al Viroe41f7d42011-02-22 14:02:58 -05002210 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2211 } while (read_seqcount_retry(&fs->seq, seq));
2212 } else {
2213 get_fs_pwd(current->fs, &nd->path);
Al Viroef55d912015-12-05 20:25:06 -05002214 nd->inode = nd->path.dentry->d_inode;
Al Viroe41f7d42011-02-22 14:02:58 -05002215 }
Al Viroef55d912015-12-05 20:25:06 -05002216 return s;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002217 } else {
Jeff Layton582aa642012-12-11 08:56:16 -05002218 /* Caller must check execute permissions on the starting path component */
Al Viroc8a53ee2015-05-12 18:43:07 -04002219 struct fd f = fdget_raw(nd->dfd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002220 struct dentry *dentry;
2221
Al Viro2903ff02012-08-28 12:52:22 -04002222 if (!f.file)
Al Viro368ee9b2015-05-08 17:19:59 -04002223 return ERR_PTR(-EBADF);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002224
Al Viro2903ff02012-08-28 12:52:22 -04002225 dentry = f.file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002226
Al Viroedc2b1d2018-07-09 16:27:23 -04002227 if (*s && unlikely(!d_can_lookup(dentry))) {
2228 fdput(f);
2229 return ERR_PTR(-ENOTDIR);
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 Viro34a26b92015-05-11 08:05:05 -04002234 nd->inode = nd->path.dentry->d_inode;
2235 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viroe41f7d42011-02-22 14:02:58 -05002236 } else {
Al Viro2903ff02012-08-28 12:52:22 -04002237 path_get(&nd->path);
Al Viro34a26b92015-05-11 08:05:05 -04002238 nd->inode = nd->path.dentry->d_inode;
Al Viroe41f7d42011-02-22 14:02:58 -05002239 }
Al Viro34a26b92015-05-11 08:05:05 -04002240 fdput(f);
Al Viro368ee9b2015-05-08 17:19:59 -04002241 return s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 }
Al Viro9b4a9b12009-04-07 11:44:16 -04002243}
2244
Al Viro3bdba282015-05-08 17:37:07 -04002245static const char *trailing_symlink(struct nameidata *nd)
Al Viro95fa25d2015-04-22 13:46:57 -04002246{
2247 const char *s;
Al Virofec2fa22015-05-06 15:58:18 -04002248 int error = may_follow_link(nd);
Al Virodeb106c2015-05-08 18:05:21 -04002249 if (unlikely(error))
Al Viro3bdba282015-05-08 17:37:07 -04002250 return ERR_PTR(error);
Al Viro95fa25d2015-04-22 13:46:57 -04002251 nd->flags |= LOOKUP_PARENT;
Al Virofab51e82015-05-10 11:01:00 -04002252 nd->stack[0].name = NULL;
Al Viro3b2e7f72015-04-19 00:53:50 -04002253 s = get_link(nd);
Al Virodeb106c2015-05-08 18:05:21 -04002254 return s ? s : "";
Al Viro95fa25d2015-04-22 13:46:57 -04002255}
2256
Al Virocaa856342015-04-22 17:52:47 -04002257static inline int lookup_last(struct nameidata *nd)
Al Virobd92d7f2011-03-14 19:54:59 -04002258{
2259 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2260 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2261
2262 nd->flags &= ~LOOKUP_PARENT;
Al Viro1c4ff1a2016-11-14 01:39:36 -05002263 return walk_component(nd, 0);
Al Virobd92d7f2011-03-14 19:54:59 -04002264}
2265
Al Viro4f757f32017-04-15 17:31:22 -04002266static int handle_lookup_down(struct nameidata *nd)
2267{
2268 struct path path = nd->path;
2269 struct inode *inode = nd->inode;
2270 unsigned seq = nd->seq;
2271 int err;
2272
2273 if (nd->flags & LOOKUP_RCU) {
2274 /*
2275 * don't bother with unlazy_walk on failure - we are
2276 * at the very beginning of walk, so we lose nothing
2277 * if we simply redo everything in non-RCU mode
2278 */
2279 if (unlikely(!__follow_mount_rcu(nd, &path, &inode, &seq)))
2280 return -ECHILD;
2281 } else {
2282 dget(path.dentry);
2283 err = follow_managed(&path, nd);
2284 if (unlikely(err < 0))
2285 return err;
2286 inode = d_backing_inode(path.dentry);
2287 seq = 0;
2288 }
2289 path_to_nameidata(&path, nd);
2290 nd->inode = inode;
2291 nd->seq = seq;
2292 return 0;
2293}
2294
Al Viro9b4a9b12009-04-07 11:44:16 -04002295/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroc8a53ee2015-05-12 18:43:07 -04002296static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
Al Viro9b4a9b12009-04-07 11:44:16 -04002297{
Al Viroc8a53ee2015-05-12 18:43:07 -04002298 const char *s = path_init(nd, flags);
Al Virobd92d7f2011-03-14 19:54:59 -04002299 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11002300
Al Viro9b5858e2018-07-09 16:33:23 -04002301 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
Al Viro4f757f32017-04-15 17:31:22 -04002302 err = handle_lookup_down(nd);
Al Viro5f336e72018-07-09 16:38:06 -04002303 if (unlikely(err < 0))
2304 s = ERR_PTR(err);
Al Viro4f757f32017-04-15 17:31:22 -04002305 }
2306
Al Viro3bdba282015-05-08 17:37:07 -04002307 while (!(err = link_path_walk(s, nd))
2308 && ((err = lookup_last(nd)) > 0)) {
2309 s = trailing_symlink(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04002310 }
Al Viro9f1fafe2011-03-25 11:00:12 -04002311 if (!err)
2312 err = complete_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04002313
Al Virodeb106c2015-05-08 18:05:21 -04002314 if (!err && nd->flags & LOOKUP_DIRECTORY)
2315 if (!d_can_lookup(nd->path.dentry))
Al Virobd23a532011-03-23 09:56:30 -04002316 err = -ENOTDIR;
Al Viro625b6d12015-05-12 16:36:12 -04002317 if (!err) {
2318 *path = nd->path;
2319 nd->path.mnt = NULL;
2320 nd->path.dentry = NULL;
2321 }
2322 terminate_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04002323 return err;
Al Viroee0827c2011-02-21 23:38:09 -05002324}
Nick Piggin31e6b012011-01-07 17:49:52 +11002325
David Howells31d921c2018-11-01 23:07:24 +00002326int filename_lookup(int dfd, struct filename *name, unsigned flags,
2327 struct path *path, struct path *root)
Jeff Layton873f1ee2012-10-10 15:25:29 -04002328{
Al Viro894bc8c2015-05-02 07:16:16 -04002329 int retval;
Al Viro9883d182015-05-13 07:28:08 -04002330 struct nameidata nd;
Al Viroabc9f5b2015-05-12 16:53:42 -04002331 if (IS_ERR(name))
2332 return PTR_ERR(name);
Al Viro9ad1aaa2015-05-12 16:44:39 -04002333 if (unlikely(root)) {
2334 nd.root = *root;
2335 flags |= LOOKUP_ROOT;
2336 }
Al Viro9883d182015-05-13 07:28:08 -04002337 set_nameidata(&nd, dfd, name);
Al Viroc8a53ee2015-05-12 18:43:07 -04002338 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002339 if (unlikely(retval == -ECHILD))
Al Viroc8a53ee2015-05-12 18:43:07 -04002340 retval = path_lookupat(&nd, flags, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002341 if (unlikely(retval == -ESTALE))
Al Viroc8a53ee2015-05-12 18:43:07 -04002342 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002343
2344 if (likely(!retval))
Al Viroff0ebee2019-07-14 12:18:05 -04002345 audit_inode(name, path->dentry, 0);
Al Viro9883d182015-05-13 07:28:08 -04002346 restore_nameidata();
Al Viroe4bd1c12015-05-12 16:40:39 -04002347 putname(name);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002348 return retval;
2349}
2350
Al Viro8bcb77f2015-05-08 16:59:20 -04002351/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroc8a53ee2015-05-12 18:43:07 -04002352static int path_parentat(struct nameidata *nd, unsigned flags,
Al Viro391172c2015-05-09 11:19:16 -04002353 struct path *parent)
Al Viro8bcb77f2015-05-08 16:59:20 -04002354{
Al Viroc8a53ee2015-05-12 18:43:07 -04002355 const char *s = path_init(nd, flags);
Al Viro9b5858e2018-07-09 16:33:23 -04002356 int err = link_path_walk(s, nd);
Al Viro8bcb77f2015-05-08 16:59:20 -04002357 if (!err)
2358 err = complete_walk(nd);
Al Viro391172c2015-05-09 11:19:16 -04002359 if (!err) {
2360 *parent = nd->path;
2361 nd->path.mnt = NULL;
2362 nd->path.dentry = NULL;
2363 }
2364 terminate_walk(nd);
Al Viro8bcb77f2015-05-08 16:59:20 -04002365 return err;
2366}
2367
Al Viro5c31b6c2015-05-12 17:32:54 -04002368static struct filename *filename_parentat(int dfd, struct filename *name,
Al Viro391172c2015-05-09 11:19:16 -04002369 unsigned int flags, struct path *parent,
2370 struct qstr *last, int *type)
Al Viro8bcb77f2015-05-08 16:59:20 -04002371{
2372 int retval;
Al Viro9883d182015-05-13 07:28:08 -04002373 struct nameidata nd;
Al Viro8bcb77f2015-05-08 16:59:20 -04002374
Al Viro5c31b6c2015-05-12 17:32:54 -04002375 if (IS_ERR(name))
2376 return name;
Al Viro9883d182015-05-13 07:28:08 -04002377 set_nameidata(&nd, dfd, name);
Al Viroc8a53ee2015-05-12 18:43:07 -04002378 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
Al Viro8bcb77f2015-05-08 16:59:20 -04002379 if (unlikely(retval == -ECHILD))
Al Viroc8a53ee2015-05-12 18:43:07 -04002380 retval = path_parentat(&nd, flags, parent);
Al Viro8bcb77f2015-05-08 16:59:20 -04002381 if (unlikely(retval == -ESTALE))
Al Viroc8a53ee2015-05-12 18:43:07 -04002382 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
Al Viro391172c2015-05-09 11:19:16 -04002383 if (likely(!retval)) {
2384 *last = nd.last;
2385 *type = nd.last_type;
Al Viroc9b07ea2019-07-14 13:22:27 -04002386 audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
Al Viro5c31b6c2015-05-12 17:32:54 -04002387 } else {
2388 putname(name);
2389 name = ERR_PTR(retval);
Al Viro391172c2015-05-09 11:19:16 -04002390 }
Al Viro9883d182015-05-13 07:28:08 -04002391 restore_nameidata();
Al Viro5c31b6c2015-05-12 17:32:54 -04002392 return name;
Al Viro8bcb77f2015-05-08 16:59:20 -04002393}
2394
Al Viro79714f72012-06-15 03:01:42 +04002395/* does lookup, returns the object with parent locked */
2396struct dentry *kern_path_locked(const char *name, struct path *path)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002397{
Al Viro5c31b6c2015-05-12 17:32:54 -04002398 struct filename *filename;
2399 struct dentry *d;
Al Viro391172c2015-05-09 11:19:16 -04002400 struct qstr last;
2401 int type;
Paul Moore51689102015-01-22 00:00:03 -05002402
Al Viro5c31b6c2015-05-12 17:32:54 -04002403 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2404 &last, &type);
Paul Moore51689102015-01-22 00:00:03 -05002405 if (IS_ERR(filename))
2406 return ERR_CAST(filename);
Al Viro5c31b6c2015-05-12 17:32:54 -04002407 if (unlikely(type != LAST_NORM)) {
Al Viro391172c2015-05-09 11:19:16 -04002408 path_put(path);
Al Viro5c31b6c2015-05-12 17:32:54 -04002409 putname(filename);
2410 return ERR_PTR(-EINVAL);
Al Viro79714f72012-06-15 03:01:42 +04002411 }
Al Viro59551022016-01-22 15:40:57 -05002412 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
Al Viro391172c2015-05-09 11:19:16 -04002413 d = __lookup_hash(&last, path->dentry, 0);
Al Viro79714f72012-06-15 03:01:42 +04002414 if (IS_ERR(d)) {
Al Viro59551022016-01-22 15:40:57 -05002415 inode_unlock(path->dentry->d_inode);
Al Viro391172c2015-05-09 11:19:16 -04002416 path_put(path);
Al Viro79714f72012-06-15 03:01:42 +04002417 }
Paul Moore51689102015-01-22 00:00:03 -05002418 putname(filename);
Al Viro79714f72012-06-15 03:01:42 +04002419 return d;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002420}
2421
Al Virod1811462008-08-02 00:49:18 -04002422int kern_path(const char *name, unsigned int flags, struct path *path)
2423{
Al Viroabc9f5b2015-05-12 16:53:42 -04002424 return filename_lookup(AT_FDCWD, getname_kernel(name),
2425 flags, path, NULL);
Al Virod1811462008-08-02 00:49:18 -04002426}
Al Viro4d359502014-03-14 12:20:17 -04002427EXPORT_SYMBOL(kern_path);
Al Virod1811462008-08-02 00:49:18 -04002428
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002429/**
2430 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2431 * @dentry: pointer to dentry of the base directory
2432 * @mnt: pointer to vfs mount of the base directory
2433 * @name: pointer to file name
2434 * @flags: lookup flags
Al Viroe0a01242011-06-27 17:00:37 -04002435 * @path: pointer to struct path to fill
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002436 */
2437int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2438 const char *name, unsigned int flags,
Al Viroe0a01242011-06-27 17:00:37 -04002439 struct path *path)
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002440{
Al Viro9ad1aaa2015-05-12 16:44:39 -04002441 struct path root = {.mnt = mnt, .dentry = dentry};
Al Viro9ad1aaa2015-05-12 16:44:39 -04002442 /* the first argument of filename_lookup() is ignored with root */
Al Viroabc9f5b2015-05-12 16:53:42 -04002443 return filename_lookup(AT_FDCWD, getname_kernel(name),
2444 flags , path, &root);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002445}
Al Viro4d359502014-03-14 12:20:17 -04002446EXPORT_SYMBOL(vfs_path_lookup);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002447
Al Viro3c95f0d2018-04-06 16:32:38 -04002448static int lookup_one_len_common(const char *name, struct dentry *base,
2449 int len, struct qstr *this)
2450{
2451 this->name = name;
2452 this->len = len;
2453 this->hash = full_name_hash(base, name, len);
2454 if (!len)
2455 return -EACCES;
2456
2457 if (unlikely(name[0] == '.')) {
2458 if (len < 2 || (len == 2 && name[1] == '.'))
2459 return -EACCES;
2460 }
2461
2462 while (len--) {
2463 unsigned int c = *(const unsigned char *)name++;
2464 if (c == '/' || c == '\0')
2465 return -EACCES;
2466 }
2467 /*
2468 * See if the low-level filesystem might want
2469 * to use its own hash..
2470 */
2471 if (base->d_flags & DCACHE_OP_HASH) {
2472 int err = base->d_op->d_hash(base, this);
2473 if (err < 0)
2474 return err;
2475 }
2476
2477 return inode_permission(base->d_inode, MAY_EXEC);
2478}
2479
Christoph Hellwigeead1912007-10-16 23:25:38 -07002480/**
David Howells0da0b7f2018-06-15 15:19:22 +01002481 * try_lookup_one_len - filesystem helper to lookup single pathname component
2482 * @name: pathname component to lookup
2483 * @base: base directory to lookup from
2484 * @len: maximum length @len should be interpreted to
2485 *
2486 * Look up a dentry by name in the dcache, returning NULL if it does not
2487 * currently exist. The function does not try to create a dentry.
2488 *
2489 * Note that this routine is purely a helper for filesystem usage and should
2490 * not be called by generic code.
2491 *
2492 * The caller must hold base->i_mutex.
2493 */
2494struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2495{
2496 struct qstr this;
2497 int err;
2498
2499 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2500
2501 err = lookup_one_len_common(name, base, len, &this);
2502 if (err)
2503 return ERR_PTR(err);
2504
2505 return lookup_dcache(&this, base, 0);
2506}
2507EXPORT_SYMBOL(try_lookup_one_len);
2508
2509/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07002510 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07002511 * @name: pathname component to lookup
2512 * @base: base directory to lookup from
2513 * @len: maximum length @len should be interpreted to
2514 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07002515 * Note that this routine is purely a helper for filesystem usage and should
Al Viro9e7543e2015-02-23 02:49:48 -05002516 * not be called by generic code.
NeilBrownbbddca82016-01-07 16:08:20 -05002517 *
2518 * The caller must hold base->i_mutex.
Christoph Hellwigeead1912007-10-16 23:25:38 -07002519 */
James Morris057f6c02007-04-26 00:12:05 -07002520struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2521{
Al Viro8613a202018-04-06 16:45:33 -04002522 struct dentry *dentry;
James Morris057f6c02007-04-26 00:12:05 -07002523 struct qstr this;
Miklos Szeredicda309d2012-03-26 12:54:21 +02002524 int err;
James Morris057f6c02007-04-26 00:12:05 -07002525
Al Viro59551022016-01-22 15:40:57 -05002526 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
David Woodhouse2f9092e2009-04-20 23:18:37 +01002527
Al Viro3c95f0d2018-04-06 16:32:38 -04002528 err = lookup_one_len_common(name, base, len, &this);
Miklos Szeredicda309d2012-03-26 12:54:21 +02002529 if (err)
2530 return ERR_PTR(err);
2531
Al Viro8613a202018-04-06 16:45:33 -04002532 dentry = lookup_dcache(&this, base, 0);
2533 return dentry ? dentry : __lookup_slow(&this, base, 0);
James Morris057f6c02007-04-26 00:12:05 -07002534}
Al Viro4d359502014-03-14 12:20:17 -04002535EXPORT_SYMBOL(lookup_one_len);
James Morris057f6c02007-04-26 00:12:05 -07002536
NeilBrownbbddca82016-01-07 16:08:20 -05002537/**
2538 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2539 * @name: pathname component to lookup
2540 * @base: base directory to lookup from
2541 * @len: maximum length @len should be interpreted to
2542 *
2543 * Note that this routine is purely a helper for filesystem usage and should
2544 * not be called by generic code.
2545 *
2546 * Unlike lookup_one_len, it should be called without the parent
2547 * i_mutex held, and will take the i_mutex itself if necessary.
2548 */
2549struct dentry *lookup_one_len_unlocked(const char *name,
2550 struct dentry *base, int len)
2551{
2552 struct qstr this;
NeilBrownbbddca82016-01-07 16:08:20 -05002553 int err;
Linus Torvalds20d00ee2016-07-29 12:17:52 -07002554 struct dentry *ret;
NeilBrownbbddca82016-01-07 16:08:20 -05002555
Al Viro3c95f0d2018-04-06 16:32:38 -04002556 err = lookup_one_len_common(name, base, len, &this);
NeilBrownbbddca82016-01-07 16:08:20 -05002557 if (err)
2558 return ERR_PTR(err);
2559
Linus Torvalds20d00ee2016-07-29 12:17:52 -07002560 ret = lookup_dcache(&this, base, 0);
2561 if (!ret)
2562 ret = lookup_slow(&this, base, 0);
2563 return ret;
NeilBrownbbddca82016-01-07 16:08:20 -05002564}
2565EXPORT_SYMBOL(lookup_one_len_unlocked);
2566
Eric W. Biedermaneedf2652016-06-02 10:29:47 -05002567#ifdef CONFIG_UNIX98_PTYS
2568int path_pts(struct path *path)
2569{
2570 /* Find something mounted on "pts" in the same directory as
2571 * the input path.
2572 */
2573 struct dentry *child, *parent;
2574 struct qstr this;
2575 int ret;
2576
2577 ret = path_parent_directory(path);
2578 if (ret)
2579 return ret;
2580
2581 parent = path->dentry;
2582 this.name = "pts";
2583 this.len = 3;
2584 child = d_hash_and_lookup(parent, &this);
2585 if (!child)
2586 return -ENOENT;
2587
2588 path->dentry = child;
2589 dput(parent);
2590 follow_mount(path);
2591 return 0;
2592}
2593#endif
2594
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002595int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2596 struct path *path, int *empty)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597{
Al Viroabc9f5b2015-05-12 16:53:42 -04002598 return filename_lookup(dfd, getname_flags(name, flags, empty),
2599 flags, path, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600}
Al Virob853a162015-05-13 09:12:02 -04002601EXPORT_SYMBOL(user_path_at_empty);
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002602
Jeff Layton80334262013-07-26 06:23:25 -04002603/**
Al Viro197df042013-09-08 14:03:27 -04002604 * mountpoint_last - look up last component for umount
Jeff Layton80334262013-07-26 06:23:25 -04002605 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
Jeff Layton80334262013-07-26 06:23:25 -04002606 *
2607 * This is a special lookup_last function just for umount. In this case, we
2608 * need to resolve the path without doing any revalidation.
2609 *
2610 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2611 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2612 * in almost all cases, this lookup will be served out of the dcache. The only
2613 * cases where it won't are if nd->last refers to a symlink or the path is
2614 * bogus and it doesn't exist.
2615 *
2616 * Returns:
2617 * -error: if there was an error during lookup. This includes -ENOENT if the
Al Viroba8f4612016-11-14 00:40:33 -05002618 * lookup found a negative dentry.
Jeff Layton80334262013-07-26 06:23:25 -04002619 *
Al Viroba8f4612016-11-14 00:40:33 -05002620 * 0: if we successfully resolved nd->last and found it to not to be a
2621 * symlink that needs to be followed.
Jeff Layton80334262013-07-26 06:23:25 -04002622 *
2623 * 1: if we successfully resolved nd->last and found it to be a symlink
Al Viroba8f4612016-11-14 00:40:33 -05002624 * that needs to be followed.
Jeff Layton80334262013-07-26 06:23:25 -04002625 */
2626static int
Al Viroba8f4612016-11-14 00:40:33 -05002627mountpoint_last(struct nameidata *nd)
Jeff Layton80334262013-07-26 06:23:25 -04002628{
2629 int error = 0;
Jeff Layton80334262013-07-26 06:23:25 -04002630 struct dentry *dir = nd->path.dentry;
Al Viroba8f4612016-11-14 00:40:33 -05002631 struct path path;
Jeff Layton80334262013-07-26 06:23:25 -04002632
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 Viro4675ac32017-01-09 22:29:15 -05002635 if (unlazy_walk(nd))
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 Viroba8f4612016-11-14 00:40:33 -05002645 path.dentry = dget(nd->path.dentry);
Al Viro949a8522016-03-06 14:20:52 -05002646 } else {
Al Viroba8f4612016-11-14 00:40:33 -05002647 path.dentry = d_lookup(dir, &nd->last);
2648 if (!path.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 */
Al Viroba8f4612016-11-14 00:40:33 -05002655 path.dentry = lookup_slow(&nd->last, dir,
Al Viro949a8522016-03-06 14:20:52 -05002656 nd->flags | LOOKUP_NO_REVAL);
Al Viroba8f4612016-11-14 00:40:33 -05002657 if (IS_ERR(path.dentry))
2658 return PTR_ERR(path.dentry);
Dave Jonesbcceeeb2013-09-10 17:04:25 -04002659 }
Jeff Layton80334262013-07-26 06:23:25 -04002660 }
Al Viroba8f4612016-11-14 00:40:33 -05002661 if (d_is_negative(path.dentry)) {
2662 dput(path.dentry);
Al Virodeb106c2015-05-08 18:05:21 -04002663 return -ENOENT;
Jeff Layton80334262013-07-26 06:23:25 -04002664 }
Al Viroba8f4612016-11-14 00:40:33 -05002665 path.mnt = nd->path.mnt;
Al Viro8f64fb12016-11-14 01:50:26 -05002666 return step_into(nd, &path, 0, d_backing_inode(path.dentry), 0);
Jeff Layton80334262013-07-26 06:23:25 -04002667}
2668
2669/**
Al Viro197df042013-09-08 14:03:27 -04002670 * path_mountpoint - look up a path to be umounted
Masanari Iida2a78b8572015-09-09 15:39:23 -07002671 * @nd: lookup context
Jeff Layton80334262013-07-26 06:23:25 -04002672 * @flags: lookup flags
Al Viroc8a53ee2015-05-12 18:43:07 -04002673 * @path: pointer to container for result
Jeff Layton80334262013-07-26 06:23:25 -04002674 *
2675 * Look up the given name, but don't attempt to revalidate the last component.
Randy Dunlap606d6fe2013-10-19 14:56:55 -07002676 * Returns 0 and "path" will be valid on success; Returns error otherwise.
Jeff Layton80334262013-07-26 06:23:25 -04002677 */
2678static int
Al Viroc8a53ee2015-05-12 18:43:07 -04002679path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
Jeff Layton80334262013-07-26 06:23:25 -04002680{
Al Viroc8a53ee2015-05-12 18:43:07 -04002681 const char *s = path_init(nd, flags);
Al Viro368ee9b2015-05-08 17:19:59 -04002682 int err;
Al Viro9b5858e2018-07-09 16:33:23 -04002683
Al Viro3bdba282015-05-08 17:37:07 -04002684 while (!(err = link_path_walk(s, nd)) &&
Al Viroba8f4612016-11-14 00:40:33 -05002685 (err = mountpoint_last(nd)) > 0) {
Al Viro3bdba282015-05-08 17:37:07 -04002686 s = trailing_symlink(nd);
Jeff Layton80334262013-07-26 06:23:25 -04002687 }
Al Viroba8f4612016-11-14 00:40:33 -05002688 if (!err) {
2689 *path = nd->path;
2690 nd->path.mnt = NULL;
2691 nd->path.dentry = NULL;
2692 follow_mount(path);
2693 }
Al Virodeb106c2015-05-08 18:05:21 -04002694 terminate_walk(nd);
Jeff Layton80334262013-07-26 06:23:25 -04002695 return err;
2696}
2697
Al Viro2d864652013-09-08 20:18:44 -04002698static int
Al Viro668696d2015-02-22 19:44:00 -05002699filename_mountpoint(int dfd, struct filename *name, struct path *path,
Al Viro2d864652013-09-08 20:18:44 -04002700 unsigned int flags)
2701{
Al Viro9883d182015-05-13 07:28:08 -04002702 struct nameidata nd;
Al Virocbaab2d2015-01-22 02:49:00 -05002703 int error;
Al Viro668696d2015-02-22 19:44:00 -05002704 if (IS_ERR(name))
2705 return PTR_ERR(name);
Al Viro9883d182015-05-13 07:28:08 -04002706 set_nameidata(&nd, dfd, name);
Al Viroc8a53ee2015-05-12 18:43:07 -04002707 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
Al Viro2d864652013-09-08 20:18:44 -04002708 if (unlikely(error == -ECHILD))
Al Viroc8a53ee2015-05-12 18:43:07 -04002709 error = path_mountpoint(&nd, flags, path);
Al Viro2d864652013-09-08 20:18:44 -04002710 if (unlikely(error == -ESTALE))
Al Viroc8a53ee2015-05-12 18:43:07 -04002711 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
Al Viro2d864652013-09-08 20:18:44 -04002712 if (likely(!error))
Al Viroc9b07ea2019-07-14 13:22:27 -04002713 audit_inode(name, path->dentry, AUDIT_INODE_NOEVAL);
Al Viro9883d182015-05-13 07:28:08 -04002714 restore_nameidata();
Al Viro668696d2015-02-22 19:44:00 -05002715 putname(name);
Al Viro2d864652013-09-08 20:18:44 -04002716 return error;
2717}
2718
Jeff Layton80334262013-07-26 06:23:25 -04002719/**
Al Viro197df042013-09-08 14:03:27 -04002720 * user_path_mountpoint_at - lookup a path from userland in order to umount it
Jeff Layton80334262013-07-26 06:23:25 -04002721 * @dfd: directory file descriptor
2722 * @name: pathname from userland
2723 * @flags: lookup flags
2724 * @path: pointer to container to hold result
2725 *
2726 * A umount is a special case for path walking. We're not actually interested
2727 * in the inode in this situation, and ESTALE errors can be a problem. We
2728 * simply want track down the dentry and vfsmount attached at the mountpoint
2729 * and avoid revalidating the last component.
2730 *
2731 * Returns 0 and populates "path" on success.
2732 */
2733int
Al Viro197df042013-09-08 14:03:27 -04002734user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
Jeff Layton80334262013-07-26 06:23:25 -04002735 struct path *path)
2736{
Al Virocbaab2d2015-01-22 02:49:00 -05002737 return filename_mountpoint(dfd, getname(name), path, flags);
Jeff Layton80334262013-07-26 06:23:25 -04002738}
2739
Al Viro2d864652013-09-08 20:18:44 -04002740int
2741kern_path_mountpoint(int dfd, const char *name, struct path *path,
2742 unsigned int flags)
2743{
Al Virocbaab2d2015-01-22 02:49:00 -05002744 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
Al Viro2d864652013-09-08 20:18:44 -04002745}
2746EXPORT_SYMBOL(kern_path_mountpoint);
2747
Miklos Szeredicbdf35b2014-10-24 00:14:36 +02002748int __check_sticky(struct inode *dir, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002749{
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002750 kuid_t fsuid = current_fsuid();
David Howellsda9592e2008-11-14 10:39:05 +11002751
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002752 if (uid_eq(inode->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002753 return 0;
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002754 if (uid_eq(dir->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002755 return 0;
Andy Lutomirski23adbe12014-06-10 12:45:42 -07002756 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002757}
Miklos Szeredicbdf35b2014-10-24 00:14:36 +02002758EXPORT_SYMBOL(__check_sticky);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759
2760/*
2761 * Check whether we can remove a link victim from directory dir, check
2762 * whether the type of victim is right.
2763 * 1. We can't do it if dir is read-only (done in permission())
2764 * 2. We should have write and exec permissions on dir
2765 * 3. We can't remove anything from append-only dir
2766 * 4. We can't do anything with immutable dir (done in permission())
2767 * 5. If the sticky bit on dir is set we should either
2768 * a. be owner of dir, or
2769 * b. be owner of victim, or
2770 * c. have CAP_FOWNER capability
2771 * 6. If the victim is append-only or immutable we can't do antyhing with
2772 * links pointing to it.
Eric W. Biederman0bd23d092016-06-29 14:54:46 -05002773 * 7. If the victim has an unknown uid or gid we can't change the inode.
2774 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2775 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2776 * 10. We can't remove a root or mountpoint.
2777 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
Linus Torvalds1da177e2005-04-16 15:20:36 -07002778 * nfs_async_unlink().
2779 */
David Howellsb18825a2013-09-12 19:22:53 +01002780static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002781{
David Howells63afdfc2015-05-06 15:59:00 +01002782 struct inode *inode = d_backing_inode(victim);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002783 int error;
2784
David Howellsb18825a2013-09-12 19:22:53 +01002785 if (d_is_negative(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786 return -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01002787 BUG_ON(!inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788
2789 BUG_ON(victim->d_parent->d_inode != dir);
Eric W. Biederman593d1ce2017-09-14 12:07:32 -05002790
2791 /* Inode writeback is not safe when the uid or gid are invalid. */
2792 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
2793 return -EOVERFLOW;
2794
Jeff Layton4fa6b5e2012-10-10 15:25:25 -04002795 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002796
Al Virof419a2e2008-07-22 00:07:17 -04002797 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798 if (error)
2799 return error;
2800 if (IS_APPEND(dir))
2801 return -EPERM;
David Howellsb18825a2013-09-12 19:22:53 +01002802
2803 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
Eric W. Biederman0bd23d092016-06-29 14:54:46 -05002804 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002805 return -EPERM;
2806 if (isdir) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002807 if (!d_is_dir(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002808 return -ENOTDIR;
2809 if (IS_ROOT(victim))
2810 return -EBUSY;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002811 } else if (d_is_dir(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812 return -EISDIR;
2813 if (IS_DEADDIR(dir))
2814 return -ENOENT;
2815 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2816 return -EBUSY;
2817 return 0;
2818}
2819
2820/* Check whether we can create an object with dentry child in directory
2821 * dir.
2822 * 1. We can't do it if child already exists (open has special treatment for
2823 * this case, but since we are inlined it's OK)
2824 * 2. We can't do it if dir is read-only (done in permission())
Eric W. Biederman036d5232016-07-01 12:52:06 -05002825 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2826 * 4. We should have write and exec permissions on dir
2827 * 5. We can't do it if dir is immutable (done in permission())
Linus Torvalds1da177e2005-04-16 15:20:36 -07002828 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02002829static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830{
Eric W. Biederman036d5232016-07-01 12:52:06 -05002831 struct user_namespace *s_user_ns;
Jeff Layton14e972b2013-05-08 10:25:58 -04002832 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002833 if (child->d_inode)
2834 return -EEXIST;
2835 if (IS_DEADDIR(dir))
2836 return -ENOENT;
Eric W. Biederman036d5232016-07-01 12:52:06 -05002837 s_user_ns = dir->i_sb->s_user_ns;
2838 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2839 !kgid_has_mapping(s_user_ns, current_fsgid()))
2840 return -EOVERFLOW;
Al Virof419a2e2008-07-22 00:07:17 -04002841 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002842}
2843
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844/*
2845 * p1 and p2 should be directories on the same fs.
2846 */
2847struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2848{
2849 struct dentry *p;
2850
2851 if (p1 == p2) {
Al Viro59551022016-01-22 15:40:57 -05002852 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002853 return NULL;
2854 }
2855
Al Virofc640052016-04-10 01:33:30 -04002856 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002858 p = d_ancestor(p2, p1);
2859 if (p) {
Al Viro59551022016-01-22 15:40:57 -05002860 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2861 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002862 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863 }
2864
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002865 p = d_ancestor(p1, p2);
2866 if (p) {
Al Viro59551022016-01-22 15:40:57 -05002867 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2868 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002869 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002870 }
2871
Al Viro59551022016-01-22 15:40:57 -05002872 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2873 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002874 return NULL;
2875}
Al Viro4d359502014-03-14 12:20:17 -04002876EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002877
2878void unlock_rename(struct dentry *p1, struct dentry *p2)
2879{
Al Viro59551022016-01-22 15:40:57 -05002880 inode_unlock(p1->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881 if (p1 != p2) {
Al Viro59551022016-01-22 15:40:57 -05002882 inode_unlock(p2->d_inode);
Al Virofc640052016-04-10 01:33:30 -04002883 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002884 }
2885}
Al Viro4d359502014-03-14 12:20:17 -04002886EXPORT_SYMBOL(unlock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002887
Al Viro4acdaf22011-07-26 01:42:34 -04002888int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viro312b63f2012-06-10 18:09:36 -04002889 bool want_excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002890{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002891 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002892 if (error)
2893 return error;
2894
Al Viroacfa4382008-12-04 10:06:33 -05002895 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002896 return -EACCES; /* shouldn't it be ENOSYS? */
2897 mode &= S_IALLUGO;
2898 mode |= S_IFREG;
2899 error = security_inode_create(dir, dentry, mode);
2900 if (error)
2901 return error;
Al Viro312b63f2012-06-10 18:09:36 -04002902 error = dir->i_op->create(dir, dentry, mode, want_excl);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002903 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002904 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002905 return error;
2906}
Al Viro4d359502014-03-14 12:20:17 -04002907EXPORT_SYMBOL(vfs_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002908
Al Viro8e6c8482017-12-01 17:12:45 -05002909int vfs_mkobj(struct dentry *dentry, umode_t mode,
2910 int (*f)(struct dentry *, umode_t, void *),
2911 void *arg)
2912{
2913 struct inode *dir = dentry->d_parent->d_inode;
2914 int error = may_create(dir, dentry);
2915 if (error)
2916 return error;
2917
2918 mode &= S_IALLUGO;
2919 mode |= S_IFREG;
2920 error = security_inode_create(dir, dentry, mode);
2921 if (error)
2922 return error;
2923 error = f(dentry, mode, arg);
2924 if (!error)
2925 fsnotify_create(dir, dentry);
2926 return error;
2927}
2928EXPORT_SYMBOL(vfs_mkobj);
2929
Eric W. Biedermana2982cc2016-06-09 15:34:02 -05002930bool may_open_dev(const struct path *path)
2931{
2932 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2933 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2934}
2935
Al Virof0bb5aa2016-11-20 20:27:12 -05002936static int may_open(const struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002937{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002938 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002939 struct inode *inode = dentry->d_inode;
2940 int error;
2941
2942 if (!inode)
2943 return -ENOENT;
2944
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002945 switch (inode->i_mode & S_IFMT) {
2946 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002947 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002948 case S_IFDIR:
2949 if (acc_mode & MAY_WRITE)
2950 return -EISDIR;
2951 break;
2952 case S_IFBLK:
2953 case S_IFCHR:
Eric W. Biedermana2982cc2016-06-09 15:34:02 -05002954 if (!may_open_dev(path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002955 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002956 /*FALLTHRU*/
2957 case S_IFIFO:
2958 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002959 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002960 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002961 }
Dave Hansenb41572e2007-10-16 23:31:14 -07002962
Al Viro62fb4a12015-12-26 22:33:24 -05002963 error = inode_permission(inode, MAY_OPEN | acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07002964 if (error)
2965 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05002966
Linus Torvalds1da177e2005-04-16 15:20:36 -07002967 /*
2968 * An append-only file must be opened in append mode for writing.
2969 */
2970 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05002971 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05002972 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002973 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05002974 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002975 }
2976
2977 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002978 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05002979 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002980
J. Bruce Fieldsf3c7691e2011-09-21 10:58:13 -04002981 return 0;
Al Viro7715b522009-12-16 03:54:00 -05002982}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002983
Jeff Laytone1181ee2010-12-07 16:19:50 -05002984static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002985{
Al Virof0bb5aa2016-11-20 20:27:12 -05002986 const struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002987 struct inode *inode = path->dentry->d_inode;
2988 int error = get_write_access(inode);
2989 if (error)
2990 return error;
2991 /*
2992 * Refuse to truncate files with mandatory locks held on them.
2993 */
Jeff Laytond7a06982014-03-10 09:54:15 -04002994 error = locks_verify_locked(filp);
Al Viro7715b522009-12-16 03:54:00 -05002995 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002996 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002997 if (!error) {
2998 error = do_truncate(path->dentry, 0,
2999 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05003000 filp);
Al Viro7715b522009-12-16 03:54:00 -05003001 }
3002 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04003003 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004}
3005
Dave Hansend57999e2008-02-15 14:37:27 -08003006static inline int open_to_namei_flags(int flag)
3007{
Al Viro8a5e9292011-06-25 19:15:54 -04003008 if ((flag & O_ACCMODE) == 3)
3009 flag--;
Dave Hansend57999e2008-02-15 14:37:27 -08003010 return flag;
3011}
3012
Al Virod3607752016-03-25 15:21:09 -04003013static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
Miklos Szeredid18e9002012-06-05 15:10:17 +02003014{
Seth Forshee1328c722017-01-26 14:33:46 -06003015 struct user_namespace *s_user_ns;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003016 int error = security_path_mknod(dir, dentry, mode, 0);
3017 if (error)
3018 return error;
3019
Seth Forshee1328c722017-01-26 14:33:46 -06003020 s_user_ns = dir->dentry->d_sb->s_user_ns;
3021 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
3022 !kgid_has_mapping(s_user_ns, current_fsgid()))
3023 return -EOVERFLOW;
3024
Miklos Szeredid18e9002012-06-05 15:10:17 +02003025 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
3026 if (error)
3027 return error;
3028
3029 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3030}
3031
David Howells1acf0af2012-06-14 16:13:46 +01003032/*
3033 * Attempt to atomically look up, create and open a file from a negative
3034 * dentry.
3035 *
3036 * Returns 0 if successful. The file will have been created and attached to
3037 * @file by the filesystem calling finish_open().
3038 *
Al Viro00a07c12018-07-09 19:30:20 -04003039 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3040 * be set. The caller will need to perform the open themselves. @path will
3041 * have been updated to point to the new dentry. This may be negative.
David Howells1acf0af2012-06-14 16:13:46 +01003042 *
3043 * Returns an error code otherwise.
3044 */
Al Viro2675a4e2012-06-22 12:41:10 +04003045static int atomic_open(struct nameidata *nd, struct dentry *dentry,
3046 struct path *path, struct file *file,
3047 const struct open_flags *op,
Al Viro3ec2eef2018-06-08 13:43:47 -04003048 int open_flag, umode_t mode)
Miklos Szeredid18e9002012-06-05 15:10:17 +02003049{
Miklos Szeredid18e9002012-06-05 15:10:17 +02003050 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003051 struct inode *dir = nd->path.dentry->d_inode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003052 int error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003053
Al Viro384f26e2016-04-28 02:03:55 -04003054 if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */
Miklos Szeredid18e9002012-06-05 15:10:17 +02003055 open_flag &= ~O_TRUNC;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003056
Miklos Szeredid18e9002012-06-05 15:10:17 +02003057 if (nd->flags & LOOKUP_DIRECTORY)
3058 open_flag |= O_DIRECTORY;
3059
Al Viro30d904942012-06-22 12:40:19 +04003060 file->f_path.dentry = DENTRY_NOT_SET;
3061 file->f_path.mnt = nd->path.mnt;
Al Viro0fb1ea02016-04-27 14:13:10 -04003062 error = dir->i_op->atomic_open(dir, dentry, file,
Al Viro44907d72018-06-08 13:32:02 -04003063 open_to_namei_flags(open_flag), mode);
Al Viro6fbd0712016-04-28 11:50:59 -04003064 d_lookup_done(dentry);
Al Viro384f26e2016-04-28 02:03:55 -04003065 if (!error) {
Al Viro64e1ac42018-07-09 19:17:52 -04003066 if (file->f_mode & FMODE_OPENED) {
3067 /*
3068 * We didn't have the inode before the open, so check open
3069 * permission here.
3070 */
3071 int acc_mode = op->acc_mode;
3072 if (file->f_mode & FMODE_CREATED) {
3073 WARN_ON(!(open_flag & O_CREAT));
3074 fsnotify_create(dir, dentry);
3075 acc_mode = 0;
3076 }
3077 error = may_open(&file->f_path, acc_mode, open_flag);
3078 if (WARN_ON(error > 0))
3079 error = -EINVAL;
3080 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
Al Viro2675a4e2012-06-22 12:41:10 +04003081 error = -EIO;
Al Viro03da6332013-09-16 19:22:33 -04003082 } else {
Al Viro384f26e2016-04-28 02:03:55 -04003083 if (file->f_path.dentry) {
3084 dput(dentry);
3085 dentry = file->f_path.dentry;
Al Viro03da6332013-09-16 19:22:33 -04003086 }
Al Viro73a09dd2018-06-08 13:22:02 -04003087 if (file->f_mode & FMODE_CREATED)
Al Viro384f26e2016-04-28 02:03:55 -04003088 fsnotify_create(dir, dentry);
Al Viroa01e7182016-06-07 21:53:51 -04003089 if (unlikely(d_is_negative(dentry))) {
3090 error = -ENOENT;
3091 } else {
3092 path->dentry = dentry;
3093 path->mnt = nd->path.mnt;
Al Viro00a07c12018-07-09 19:30:20 -04003094 return 0;
Al Viroa01e7182016-06-07 21:53:51 -04003095 }
Sage Weil62b2ce92012-08-15 13:30:12 -07003096 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02003097 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02003098 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04003099 return error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003100}
3101
Nick Piggin31e6b012011-01-07 17:49:52 +11003102/*
David Howells1acf0af2012-06-14 16:13:46 +01003103 * Look up and maybe create and open the last component.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003104 *
Al Viro00a07c12018-07-09 19:30:20 -04003105 * Must be called with parent locked (exclusive in O_CREAT case).
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003106 *
Al Viro00a07c12018-07-09 19:30:20 -04003107 * Returns 0 on success, that is, if
3108 * the file was successfully atomically created (if necessary) and opened, or
3109 * the file was not completely opened at this time, though lookups and
3110 * creations were performed.
3111 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3112 * In the latter case dentry returned in @path might be negative if O_CREAT
3113 * hadn't been specified.
David Howells1acf0af2012-06-14 16:13:46 +01003114 *
Al Viro00a07c12018-07-09 19:30:20 -04003115 * An error code is returned on failure.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003116 */
Al Viro2675a4e2012-06-22 12:41:10 +04003117static int lookup_open(struct nameidata *nd, struct path *path,
3118 struct file *file,
3119 const struct open_flags *op,
Al Viro3ec2eef2018-06-08 13:43:47 -04003120 bool got_write)
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003121{
3122 struct dentry *dir = nd->path.dentry;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02003123 struct inode *dir_inode = dir->d_inode;
Al Viro1643b432016-04-27 19:14:10 -04003124 int open_flag = op->open_flag;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003125 struct dentry *dentry;
Al Viro1643b432016-04-27 19:14:10 -04003126 int error, create_error = 0;
Al Viro1643b432016-04-27 19:14:10 -04003127 umode_t mode = op->mode;
Al Viro6fbd0712016-04-28 11:50:59 -04003128 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003129
Al Viroce8644f2016-04-26 14:17:56 -04003130 if (unlikely(IS_DEADDIR(dir_inode)))
3131 return -ENOENT;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003132
Al Viro73a09dd2018-06-08 13:22:02 -04003133 file->f_mode &= ~FMODE_CREATED;
Al Viro6fbd0712016-04-28 11:50:59 -04003134 dentry = d_lookup(dir, &nd->last);
3135 for (;;) {
3136 if (!dentry) {
3137 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3138 if (IS_ERR(dentry))
3139 return PTR_ERR(dentry);
3140 }
3141 if (d_in_lookup(dentry))
3142 break;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003143
Al Viro6fbd0712016-04-28 11:50:59 -04003144 error = d_revalidate(dentry, nd->flags);
3145 if (likely(error > 0))
3146 break;
3147 if (error)
3148 goto out_dput;
3149 d_invalidate(dentry);
3150 dput(dentry);
3151 dentry = NULL;
3152 }
3153 if (dentry->d_inode) {
Al Viro6c51e512016-03-05 20:09:32 -05003154 /* Cached positive dentry: will open in f_op->open */
Miklos Szeredid18e9002012-06-05 15:10:17 +02003155 goto out_no_open;
Al Viro6c51e512016-03-05 20:09:32 -05003156 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02003157
Al Viro1643b432016-04-27 19:14:10 -04003158 /*
3159 * Checking write permission is tricky, bacuse we don't know if we are
3160 * going to actually need it: O_CREAT opens should work as long as the
3161 * file exists. But checking existence breaks atomicity. The trick is
3162 * to check access and if not granted clear O_CREAT from the flags.
3163 *
3164 * Another problem is returing the "right" error value (e.g. for an
3165 * O_EXCL open we want to return EEXIST not EROFS).
3166 */
3167 if (open_flag & O_CREAT) {
3168 if (!IS_POSIXACL(dir->d_inode))
3169 mode &= ~current_umask();
3170 if (unlikely(!got_write)) {
3171 create_error = -EROFS;
3172 open_flag &= ~O_CREAT;
3173 if (open_flag & (O_EXCL | O_TRUNC))
3174 goto no_open;
3175 /* No side effects, safe to clear O_CREAT */
3176 } else {
3177 create_error = may_o_create(&nd->path, dentry, mode);
3178 if (create_error) {
3179 open_flag &= ~O_CREAT;
3180 if (open_flag & O_EXCL)
3181 goto no_open;
3182 }
3183 }
3184 } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3185 unlikely(!got_write)) {
3186 /*
3187 * No O_CREATE -> atomicity not a requirement -> fall
3188 * back to lookup + open
3189 */
3190 goto no_open;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003191 }
3192
Al Viro1643b432016-04-27 19:14:10 -04003193 if (dir_inode->i_op->atomic_open) {
3194 error = atomic_open(nd, dentry, path, file, op, open_flag,
Al Viro3ec2eef2018-06-08 13:43:47 -04003195 mode);
Al Viro1643b432016-04-27 19:14:10 -04003196 if (unlikely(error == -ENOENT) && create_error)
3197 error = create_error;
3198 return error;
3199 }
Miklos Szeredi54ef4872012-06-05 15:10:16 +02003200
Al Viro1643b432016-04-27 19:14:10 -04003201no_open:
Al Viro6fbd0712016-04-28 11:50:59 -04003202 if (d_in_lookup(dentry)) {
Al Viro12fa5e22016-04-28 11:19:43 -04003203 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3204 nd->flags);
Al Viro6fbd0712016-04-28 11:50:59 -04003205 d_lookup_done(dentry);
Al Viro12fa5e22016-04-28 11:19:43 -04003206 if (unlikely(res)) {
3207 if (IS_ERR(res)) {
3208 error = PTR_ERR(res);
3209 goto out_dput;
3210 }
3211 dput(dentry);
3212 dentry = res;
3213 }
Miklos Szeredi54ef4872012-06-05 15:10:16 +02003214 }
3215
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003216 /* Negative dentry, just create the file */
Al Viro1643b432016-04-27 19:14:10 -04003217 if (!dentry->d_inode && (open_flag & O_CREAT)) {
Al Viro73a09dd2018-06-08 13:22:02 -04003218 file->f_mode |= FMODE_CREATED;
Al Viroce8644f2016-04-26 14:17:56 -04003219 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
Al Viroce8644f2016-04-26 14:17:56 -04003220 if (!dir_inode->i_op->create) {
3221 error = -EACCES;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003222 goto out_dput;
Al Viro64894cf2012-07-31 00:53:35 +04003223 }
Al Viroce8644f2016-04-26 14:17:56 -04003224 error = dir_inode->i_op->create(dir_inode, dentry, mode,
Al Viro1643b432016-04-27 19:14:10 -04003225 open_flag & O_EXCL);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003226 if (error)
3227 goto out_dput;
Al Viroce8644f2016-04-26 14:17:56 -04003228 fsnotify_create(dir_inode, dentry);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003229 }
Al Viro1643b432016-04-27 19:14:10 -04003230 if (unlikely(create_error) && !dentry->d_inode) {
3231 error = create_error;
3232 goto out_dput;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003233 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02003234out_no_open:
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003235 path->dentry = dentry;
3236 path->mnt = nd->path.mnt;
Al Viro00a07c12018-07-09 19:30:20 -04003237 return 0;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003238
3239out_dput:
3240 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04003241 return error;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003242}
3243
3244/*
Al Virofe2d35f2011-03-05 22:58:25 -05003245 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11003246 */
Al Viro896475d2015-04-22 18:02:17 -04003247static int do_last(struct nameidata *nd,
Al Viro3ec2eef2018-06-08 13:43:47 -04003248 struct file *file, const struct open_flags *op)
Al Virofb1cc552009-12-24 01:58:28 -05003249{
Al Viroa1e28032009-12-24 02:12:06 -05003250 struct dentry *dir = nd->path.dentry;
Al Viroca344a892011-03-09 00:36:45 -05003251 int open_flag = op->open_flag;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003252 bool will_truncate = (open_flag & O_TRUNC) != 0;
Al Viro64894cf2012-07-31 00:53:35 +04003253 bool got_write = false;
Al Virobcda7652011-03-13 16:42:14 -04003254 int acc_mode = op->acc_mode;
Al Viro254cf582015-05-05 09:40:46 -04003255 unsigned seq;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02003256 struct inode *inode;
Al Viro896475d2015-04-22 18:02:17 -04003257 struct path path;
Al Viro16c2cd72011-02-22 15:50:10 -05003258 int error;
Al Virofb1cc552009-12-24 01:58:28 -05003259
Al Viroc3e380b2011-02-23 13:39:45 -05003260 nd->flags &= ~LOOKUP_PARENT;
3261 nd->flags |= op->intent;
3262
Al Virobc77daa2013-06-06 09:12:33 -04003263 if (nd->last_type != LAST_NORM) {
Al Virofe2d35f2011-03-05 22:58:25 -05003264 error = handle_dots(nd, nd->last_type);
Al Virodeb106c2015-05-08 18:05:21 -04003265 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04003266 return error;
Miklos Szeredie83db162012-06-05 15:10:29 +02003267 goto finish_open;
Al Viro1f36f772009-12-26 10:56:19 -05003268 }
Al Viro67ee3ad2009-12-26 07:01:01 -05003269
Al Viroca344a892011-03-09 00:36:45 -05003270 if (!(open_flag & O_CREAT)) {
Al Virofe2d35f2011-03-05 22:58:25 -05003271 if (nd->last.name[nd->last.len])
3272 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3273 /* we _can_ be in RCU mode here */
Al Viro254cf582015-05-05 09:40:46 -04003274 error = lookup_fast(nd, &path, &inode, &seq);
Al Viroe9742b52016-03-05 22:04:59 -05003275 if (likely(error > 0))
Miklos Szeredi71574862012-06-05 15:10:14 +02003276 goto finish_lookup;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02003277
Miklos Szeredi71574862012-06-05 15:10:14 +02003278 if (error < 0)
Al Virodeb106c2015-05-08 18:05:21 -04003279 return error;
Miklos Szeredi37d7fff2012-06-05 15:10:12 +02003280
Miklos Szeredi71574862012-06-05 15:10:14 +02003281 BUG_ON(nd->inode != dir->d_inode);
Al Viro6583fe22016-03-05 18:14:03 -05003282 BUG_ON(nd->flags & LOOKUP_RCU);
Miklos Szeredib6183df2012-06-05 15:10:13 +02003283 } else {
3284 /* create side of things */
3285 /*
3286 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3287 * has been cleared when we got to the last component we are
3288 * about to look up
3289 */
3290 error = complete_walk(nd);
Al Viroe8bb73d2015-05-08 16:28:42 -04003291 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003292 return error;
Miklos Szeredib6183df2012-06-05 15:10:13 +02003293
Al Viroc9b07ea2019-07-14 13:22:27 -04003294 audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
Miklos Szeredib6183df2012-06-05 15:10:13 +02003295 /* trailing slashes? */
Al Virodeb106c2015-05-08 18:05:21 -04003296 if (unlikely(nd->last.name[nd->last.len]))
3297 return -EISDIR;
Al Virofe2d35f2011-03-05 22:58:25 -05003298 }
3299
Al Viro9cf843e2016-04-28 19:35:16 -04003300 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
Al Viro64894cf2012-07-31 00:53:35 +04003301 error = mnt_want_write(nd->path.mnt);
3302 if (!error)
3303 got_write = true;
3304 /*
3305 * do _not_ fail yet - we might not need that or fail with
3306 * a different error; let lookup_open() decide; we'll be
3307 * dropping this one anyway.
3308 */
3309 }
Al Viro9cf843e2016-04-28 19:35:16 -04003310 if (open_flag & O_CREAT)
3311 inode_lock(dir->d_inode);
3312 else
3313 inode_lock_shared(dir->d_inode);
Al Viro3ec2eef2018-06-08 13:43:47 -04003314 error = lookup_open(nd, &path, file, op, got_write);
Al Viro9cf843e2016-04-28 19:35:16 -04003315 if (open_flag & O_CREAT)
3316 inode_unlock(dir->d_inode);
3317 else
3318 inode_unlock_shared(dir->d_inode);
Al Viroa1e28032009-12-24 02:12:06 -05003319
Al Viro00a07c12018-07-09 19:30:20 -04003320 if (error)
3321 goto out;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003322
Al Viro00a07c12018-07-09 19:30:20 -04003323 if (file->f_mode & FMODE_OPENED) {
Al Viro73a09dd2018-06-08 13:22:02 -04003324 if ((file->f_mode & FMODE_CREATED) ||
Al Viro496ad9a2013-01-23 17:07:38 -05003325 !S_ISREG(file_inode(file)->i_mode))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003326 will_truncate = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003327
Al Viro76ae2a52015-05-12 18:44:32 -04003328 audit_inode(nd->name, file->f_path.dentry, 0);
Miklos Szeredid18e9002012-06-05 15:10:17 +02003329 goto opened;
3330 }
Al Virofb1cc552009-12-24 01:58:28 -05003331
Al Viro73a09dd2018-06-08 13:22:02 -04003332 if (file->f_mode & FMODE_CREATED) {
Al Viro9b44f1b2011-03-09 00:17:27 -05003333 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05003334 open_flag &= ~O_TRUNC;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003335 will_truncate = false;
Al Viro62fb4a12015-12-26 22:33:24 -05003336 acc_mode = 0;
Al Viro896475d2015-04-22 18:02:17 -04003337 path_to_nameidata(&path, nd);
Miklos Szeredie83db162012-06-05 15:10:29 +02003338 goto finish_open_created;
Al Virofb1cc552009-12-24 01:58:28 -05003339 }
3340
3341 /*
Miklos Szeredid18e9002012-06-05 15:10:17 +02003342 * If atomic_open() acquired write access it is dropped now due to
3343 * possible mount and symlink following (this might be optimized away if
3344 * necessary...)
3345 */
Al Viro64894cf2012-07-31 00:53:35 +04003346 if (got_write) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02003347 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04003348 got_write = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003349 }
3350
Al Viroe6ec03a2016-06-05 00:23:09 -04003351 error = follow_managed(&path, nd);
3352 if (unlikely(error < 0))
3353 return error;
3354
Al Viro6583fe22016-03-05 18:14:03 -05003355 if (unlikely(d_is_negative(path.dentry))) {
3356 path_to_nameidata(&path, nd);
3357 return -ENOENT;
3358 }
3359
3360 /*
3361 * create/update audit record if it already exists.
3362 */
3363 audit_inode(nd->name, path.dentry, 0);
3364
Al Virodeb106c2015-05-08 18:05:21 -04003365 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3366 path_to_nameidata(&path, nd);
3367 return -EEXIST;
3368 }
Al Virofb1cc552009-12-24 01:58:28 -05003369
Al Viro254cf582015-05-05 09:40:46 -04003370 seq = 0; /* out of RCU mode, so the value doesn't matter */
Al Virod4565642016-02-27 19:23:16 -05003371 inode = d_backing_inode(path.dentry);
Al Viro766c4cb2015-05-07 19:24:57 -04003372finish_lookup:
Al Viro8f64fb12016-11-14 01:50:26 -05003373 error = step_into(nd, &path, 0, inode, seq);
Al Virodeb106c2015-05-08 18:05:21 -04003374 if (unlikely(error))
Al Virod63ff282015-05-04 18:13:23 -04003375 return error;
Al Virobc77daa2013-06-06 09:12:33 -04003376finish_open:
Al Viro8f64fb12016-11-14 01:50:26 -05003377 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
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);
Salvatore Mesoraca30aba662018-08-23 17:00:35 -07003382 if (open_flag & O_CREAT) {
3383 error = -EISDIR;
3384 if (d_is_dir(nd->path.dentry))
3385 goto out;
3386 error = may_create_in_sticky(dir,
3387 d_backing_inode(nd->path.dentry));
3388 if (unlikely(error))
3389 goto out;
3390 }
Miklos Szerediaf2f5542012-05-21 17:30:11 +02003391 error = -ENOTDIR;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003392 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04003393 goto out;
David Howells4bbcbd3b2015-03-17 22:16:40 +00003394 if (!d_is_reg(nd->path.dentry))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003395 will_truncate = false;
Al Viro6c0d46c2011-03-09 00:59:59 -05003396
Al Viro0f9d1a12011-03-09 00:13:14 -05003397 if (will_truncate) {
3398 error = mnt_want_write(nd->path.mnt);
3399 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003400 goto out;
Al Viro64894cf2012-07-31 00:53:35 +04003401 got_write = true;
Al Viro0f9d1a12011-03-09 00:13:14 -05003402 }
Miklos Szeredie83db162012-06-05 15:10:29 +02003403finish_open_created:
Al Viro6ac08702016-04-26 00:02:50 -04003404 error = may_open(&nd->path, acc_mode, open_flag);
3405 if (error)
3406 goto out;
Al Viroaad888f2018-06-08 12:58:04 -04003407 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
Al Viroae2bb292018-07-10 13:22:28 -04003408 error = vfs_open(&nd->path, file);
Al Virofac7d192016-06-04 11:41:49 -04003409 if (error)
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003410 goto out;
Miklos Szeredia8277b92012-06-05 15:10:32 +02003411opened:
Al Viro6035a272018-06-08 13:40:10 -04003412 error = ima_file_check(file, op->acc_mode);
Al Virofe9ec822016-04-27 03:14:20 -04003413 if (!error && will_truncate)
Al Viro2675a4e2012-06-22 12:41:10 +04003414 error = handle_truncate(file);
Al Viroca344a892011-03-09 00:36:45 -05003415out:
Al Viroc80567c2016-02-27 19:17:33 -05003416 if (unlikely(error > 0)) {
3417 WARN_ON(1);
3418 error = -EINVAL;
3419 }
Al Viro64894cf2012-07-31 00:53:35 +04003420 if (got_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05003421 mnt_drop_write(nd->path.mnt);
Al Viro2675a4e2012-06-22 12:41:10 +04003422 return error;
Al Virofb1cc552009-12-24 01:58:28 -05003423}
3424
Amir Goldsteinaf7bd4d2017-01-17 06:34:52 +02003425struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag)
3426{
Amir Goldsteinaf7bd4d2017-01-17 06:34:52 +02003427 struct dentry *child = NULL;
3428 struct inode *dir = dentry->d_inode;
3429 struct inode *inode;
3430 int error;
3431
3432 /* we want directory to be writable */
3433 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3434 if (error)
3435 goto out_err;
3436 error = -EOPNOTSUPP;
3437 if (!dir->i_op->tmpfile)
3438 goto out_err;
3439 error = -ENOMEM;
David Howellscdf01222017-07-04 17:25:22 +01003440 child = d_alloc(dentry, &slash_name);
Amir Goldsteinaf7bd4d2017-01-17 06:34:52 +02003441 if (unlikely(!child))
3442 goto out_err;
3443 error = dir->i_op->tmpfile(dir, child, mode);
3444 if (error)
3445 goto out_err;
3446 error = -ENOENT;
3447 inode = child->d_inode;
3448 if (unlikely(!inode))
3449 goto out_err;
3450 if (!(open_flag & O_EXCL)) {
3451 spin_lock(&inode->i_lock);
3452 inode->i_state |= I_LINKABLE;
3453 spin_unlock(&inode->i_lock);
3454 }
Mimi Zoharfdb24102019-01-22 14:06:49 -06003455 ima_post_create_tmpfile(inode);
Amir Goldsteinaf7bd4d2017-01-17 06:34:52 +02003456 return child;
3457
3458out_err:
3459 dput(child);
3460 return ERR_PTR(error);
3461}
3462EXPORT_SYMBOL(vfs_tmpfile);
3463
Al Viroc8a53ee2015-05-12 18:43:07 -04003464static int do_tmpfile(struct nameidata *nd, unsigned flags,
Al Viro60545d02013-06-07 01:20:27 -04003465 const struct open_flags *op,
Al Viro3ec2eef2018-06-08 13:43:47 -04003466 struct file *file)
Al Viro60545d02013-06-07 01:20:27 -04003467{
Al Viro625b6d12015-05-12 16:36:12 -04003468 struct dentry *child;
Al Viro625b6d12015-05-12 16:36:12 -04003469 struct path path;
Al Viroc8a53ee2015-05-12 18:43:07 -04003470 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
Al Viro60545d02013-06-07 01:20:27 -04003471 if (unlikely(error))
3472 return error;
Al Viro625b6d12015-05-12 16:36:12 -04003473 error = mnt_want_write(path.mnt);
Al Viro60545d02013-06-07 01:20:27 -04003474 if (unlikely(error))
3475 goto out;
Amir Goldsteinaf7bd4d2017-01-17 06:34:52 +02003476 child = vfs_tmpfile(path.dentry, op->mode, op->open_flag);
3477 error = PTR_ERR(child);
Hirofumi Nakagawa684e73b2017-09-26 03:21:26 +09003478 if (IS_ERR(child))
Al Viro60545d02013-06-07 01:20:27 -04003479 goto out2;
Al Viro625b6d12015-05-12 16:36:12 -04003480 dput(path.dentry);
3481 path.dentry = child;
Al Viroc8a53ee2015-05-12 18:43:07 -04003482 audit_inode(nd->name, child, 0);
Eric Rannaud69a91c22014-10-30 01:51:01 -07003483 /* Don't check for other permissions, the inode was just created */
Al Viro62fb4a12015-12-26 22:33:24 -05003484 error = may_open(&path, 0, op->open_flag);
Al Viro60545d02013-06-07 01:20:27 -04003485 if (error)
3486 goto out2;
Al Viro625b6d12015-05-12 16:36:12 -04003487 file->f_path.mnt = path.mnt;
Al Virobe12af32018-06-08 11:44:56 -04003488 error = finish_open(file, child, NULL);
Al Viro60545d02013-06-07 01:20:27 -04003489out2:
Al Viro625b6d12015-05-12 16:36:12 -04003490 mnt_drop_write(path.mnt);
Al Viro60545d02013-06-07 01:20:27 -04003491out:
Al Viro625b6d12015-05-12 16:36:12 -04003492 path_put(&path);
Al Viro60545d02013-06-07 01:20:27 -04003493 return error;
3494}
3495
Al Viro6ac08702016-04-26 00:02:50 -04003496static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3497{
3498 struct path path;
3499 int error = path_lookupat(nd, flags, &path);
3500 if (!error) {
3501 audit_inode(nd->name, path.dentry, 0);
Al Viroae2bb292018-07-10 13:22:28 -04003502 error = vfs_open(&path, file);
Al Viro6ac08702016-04-26 00:02:50 -04003503 path_put(&path);
3504 }
3505 return error;
3506}
3507
Al Viroc8a53ee2015-05-12 18:43:07 -04003508static struct file *path_openat(struct nameidata *nd,
3509 const struct open_flags *op, unsigned flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003510{
Al Viro30d904942012-06-22 12:40:19 +04003511 struct file *file;
Al Viro13aab422011-02-23 17:54:08 -05003512 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11003513
Al Viroea73ea72018-07-11 15:00:04 -04003514 file = alloc_empty_file(op->open_flag, current_cred());
Al Viro1afc99b2013-02-14 20:41:04 -05003515 if (IS_ERR(file))
3516 return file;
Nick Piggin31e6b012011-01-07 17:49:52 +11003517
Al Virobb458c62013-07-13 13:26:37 +04003518 if (unlikely(file->f_flags & __O_TMPFILE)) {
Al Viro3ec2eef2018-06-08 13:43:47 -04003519 error = do_tmpfile(nd, flags, op, file);
Al Viro5f336e72018-07-09 16:38:06 -04003520 } else if (unlikely(file->f_flags & O_PATH)) {
Al Viro6ac08702016-04-26 00:02:50 -04003521 error = do_o_path(nd, flags, file);
Al Viro5f336e72018-07-09 16:38:06 -04003522 } else {
3523 const char *s = path_init(nd, flags);
3524 while (!(error = link_path_walk(s, nd)) &&
3525 (error = do_last(nd, file, op)) > 0) {
3526 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3527 s = trailing_symlink(nd);
3528 }
3529 terminate_walk(nd);
Al Viro6ac08702016-04-26 00:02:50 -04003530 }
Al Viro7c1c01e2018-06-08 12:56:55 -04003531 if (likely(!error)) {
Al Viroaad888f2018-06-08 12:58:04 -04003532 if (likely(file->f_mode & FMODE_OPENED))
Al Viro7c1c01e2018-06-08 12:56:55 -04003533 return file;
3534 WARN_ON(1);
3535 error = -EINVAL;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003536 }
Al Viro7c1c01e2018-06-08 12:56:55 -04003537 fput(file);
3538 if (error == -EOPENSTALE) {
3539 if (flags & LOOKUP_RCU)
3540 error = -ECHILD;
3541 else
3542 error = -ESTALE;
Al Viro2675a4e2012-06-22 12:41:10 +04003543 }
Al Viro7c1c01e2018-06-08 12:56:55 -04003544 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003545}
3546
Jeff Layton669abf42012-10-10 16:43:10 -04003547struct file *do_filp_open(int dfd, struct filename *pathname,
Al Virof9652e12013-06-11 08:23:01 +04003548 const struct open_flags *op)
Al Viro13aab422011-02-23 17:54:08 -05003549{
Al Viro9883d182015-05-13 07:28:08 -04003550 struct nameidata nd;
Al Virof9652e12013-06-11 08:23:01 +04003551 int flags = op->lookup_flags;
Al Viro13aab422011-02-23 17:54:08 -05003552 struct file *filp;
3553
Al Viro9883d182015-05-13 07:28:08 -04003554 set_nameidata(&nd, dfd, pathname);
Al Viroc8a53ee2015-05-12 18:43:07 -04003555 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
Al Viro13aab422011-02-23 17:54:08 -05003556 if (unlikely(filp == ERR_PTR(-ECHILD)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003557 filp = path_openat(&nd, op, flags);
Al Viro13aab422011-02-23 17:54:08 -05003558 if (unlikely(filp == ERR_PTR(-ESTALE)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003559 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
Al Viro9883d182015-05-13 07:28:08 -04003560 restore_nameidata();
Al Viro13aab422011-02-23 17:54:08 -05003561 return filp;
3562}
3563
Al Viro73d049a2011-03-11 12:08:24 -05003564struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
Al Virof9652e12013-06-11 08:23:01 +04003565 const char *name, const struct open_flags *op)
Al Viro73d049a2011-03-11 12:08:24 -05003566{
Al Viro9883d182015-05-13 07:28:08 -04003567 struct nameidata nd;
Al Viro73d049a2011-03-11 12:08:24 -05003568 struct file *file;
Paul Moore51689102015-01-22 00:00:03 -05003569 struct filename *filename;
Al Virof9652e12013-06-11 08:23:01 +04003570 int flags = op->lookup_flags | LOOKUP_ROOT;
Al Viro73d049a2011-03-11 12:08:24 -05003571
3572 nd.root.mnt = mnt;
3573 nd.root.dentry = dentry;
3574
David Howellsb18825a2013-09-12 19:22:53 +01003575 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
Al Viro73d049a2011-03-11 12:08:24 -05003576 return ERR_PTR(-ELOOP);
3577
Paul Moore51689102015-01-22 00:00:03 -05003578 filename = getname_kernel(name);
Viresh Kumara1c83682015-08-12 15:59:44 +05303579 if (IS_ERR(filename))
Paul Moore51689102015-01-22 00:00:03 -05003580 return ERR_CAST(filename);
3581
Al Viro9883d182015-05-13 07:28:08 -04003582 set_nameidata(&nd, -1, filename);
Al Viroc8a53ee2015-05-12 18:43:07 -04003583 file = path_openat(&nd, op, flags | LOOKUP_RCU);
Al Viro73d049a2011-03-11 12:08:24 -05003584 if (unlikely(file == ERR_PTR(-ECHILD)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003585 file = path_openat(&nd, op, flags);
Al Viro73d049a2011-03-11 12:08:24 -05003586 if (unlikely(file == ERR_PTR(-ESTALE)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003587 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
Al Viro9883d182015-05-13 07:28:08 -04003588 restore_nameidata();
Paul Moore51689102015-01-22 00:00:03 -05003589 putname(filename);
Al Viro73d049a2011-03-11 12:08:24 -05003590 return file;
3591}
3592
Al Virofa14a0b2015-01-22 02:16:49 -05003593static struct dentry *filename_create(int dfd, struct filename *name,
Jeff Layton1ac12b42012-12-11 12:10:06 -05003594 struct path *path, unsigned int lookup_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003595{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003596 struct dentry *dentry = ERR_PTR(-EEXIST);
Al Viro391172c2015-05-09 11:19:16 -04003597 struct qstr last;
3598 int type;
Jan Karac30dabf2012-06-12 16:20:30 +02003599 int err2;
Jeff Layton1ac12b42012-12-11 12:10:06 -05003600 int error;
3601 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3602
3603 /*
3604 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3605 * other flags passed in are ignored!
3606 */
3607 lookup_flags &= LOOKUP_REVAL;
3608
Al Viro5c31b6c2015-05-12 17:32:54 -04003609 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3610 if (IS_ERR(name))
3611 return ERR_CAST(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003612
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003613 /*
3614 * Yucky last component or no last component at all?
3615 * (foo/., foo/.., /////)
3616 */
Al Viro5c31b6c2015-05-12 17:32:54 -04003617 if (unlikely(type != LAST_NORM))
Al Viroed75e952011-06-27 16:53:43 -04003618 goto out;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003619
Jan Karac30dabf2012-06-12 16:20:30 +02003620 /* don't fail immediately if it's r/o, at least try to report other errors */
Al Viro391172c2015-05-09 11:19:16 -04003621 err2 = mnt_want_write(path->mnt);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003622 /*
3623 * Do the final lookup.
3624 */
Al Viro391172c2015-05-09 11:19:16 -04003625 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
Al Viro59551022016-01-22 15:40:57 -05003626 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
Al Viro391172c2015-05-09 11:19:16 -04003627 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003628 if (IS_ERR(dentry))
Al Viroa8104a92012-07-20 02:25:00 +04003629 goto unlock;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003630
Al Viroa8104a92012-07-20 02:25:00 +04003631 error = -EEXIST;
David Howellsb18825a2013-09-12 19:22:53 +01003632 if (d_is_positive(dentry))
Al Viroa8104a92012-07-20 02:25:00 +04003633 goto fail;
David Howellsb18825a2013-09-12 19:22:53 +01003634
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003635 /*
3636 * Special case - lookup gave negative, but... we had foo/bar/
3637 * From the vfs_mknod() POV we just have a negative dentry -
3638 * all is fine. Let's be bastards - you had / on the end, you've
3639 * been asking for (non-existent) directory. -ENOENT for you.
3640 */
Al Viro391172c2015-05-09 11:19:16 -04003641 if (unlikely(!is_dir && last.name[last.len])) {
Al Viroa8104a92012-07-20 02:25:00 +04003642 error = -ENOENT;
Al Viroed75e952011-06-27 16:53:43 -04003643 goto fail;
Al Viroe9baf6e2008-05-15 04:49:12 -04003644 }
Jan Karac30dabf2012-06-12 16:20:30 +02003645 if (unlikely(err2)) {
3646 error = err2;
Al Viroa8104a92012-07-20 02:25:00 +04003647 goto fail;
Jan Karac30dabf2012-06-12 16:20:30 +02003648 }
Al Viro181c37b2015-05-12 17:21:25 -04003649 putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003650 return dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003651fail:
Al Viroa8104a92012-07-20 02:25:00 +04003652 dput(dentry);
3653 dentry = ERR_PTR(error);
3654unlock:
Al Viro59551022016-01-22 15:40:57 -05003655 inode_unlock(path->dentry->d_inode);
Jan Karac30dabf2012-06-12 16:20:30 +02003656 if (!err2)
Al Viro391172c2015-05-09 11:19:16 -04003657 mnt_drop_write(path->mnt);
Al Viroed75e952011-06-27 16:53:43 -04003658out:
Al Viro391172c2015-05-09 11:19:16 -04003659 path_put(path);
Al Viro181c37b2015-05-12 17:21:25 -04003660 putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003661 return dentry;
3662}
Al Virofa14a0b2015-01-22 02:16:49 -05003663
3664struct dentry *kern_path_create(int dfd, const char *pathname,
3665 struct path *path, unsigned int lookup_flags)
3666{
Al Viro181c37b2015-05-12 17:21:25 -04003667 return filename_create(dfd, getname_kernel(pathname),
3668 path, lookup_flags);
Al Virofa14a0b2015-01-22 02:16:49 -05003669}
Al Virodae6ad82011-06-26 11:50:15 -04003670EXPORT_SYMBOL(kern_path_create);
3671
Al Viro921a1652012-07-20 01:15:31 +04003672void done_path_create(struct path *path, struct dentry *dentry)
3673{
3674 dput(dentry);
Al Viro59551022016-01-22 15:40:57 -05003675 inode_unlock(path->dentry->d_inode);
Al Viroa8104a92012-07-20 02:25:00 +04003676 mnt_drop_write(path->mnt);
Al Viro921a1652012-07-20 01:15:31 +04003677 path_put(path);
3678}
3679EXPORT_SYMBOL(done_path_create);
3680
Al Viro520ae682015-05-13 07:00:28 -04003681inline struct dentry *user_path_create(int dfd, const char __user *pathname,
Jeff Layton1ac12b42012-12-11 12:10:06 -05003682 struct path *path, unsigned int lookup_flags)
Al Virodae6ad82011-06-26 11:50:15 -04003683{
Al Viro181c37b2015-05-12 17:21:25 -04003684 return filename_create(dfd, getname(pathname), path, lookup_flags);
Al Virodae6ad82011-06-26 11:50:15 -04003685}
3686EXPORT_SYMBOL(user_path_create);
3687
Al Viro1a67aaf2011-07-26 01:52:52 -04003688int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003689{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003690 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003691
3692 if (error)
3693 return error;
3694
Christian Brauner94f82002018-07-05 17:51:20 +02003695 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003696 return -EPERM;
3697
Al Viroacfa4382008-12-04 10:06:33 -05003698 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003699 return -EPERM;
3700
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07003701 error = devcgroup_inode_mknod(mode, dev);
3702 if (error)
3703 return error;
3704
Linus Torvalds1da177e2005-04-16 15:20:36 -07003705 error = security_inode_mknod(dir, dentry, mode, dev);
3706 if (error)
3707 return error;
3708
Linus Torvalds1da177e2005-04-16 15:20:36 -07003709 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003710 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003711 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003712 return error;
3713}
Al Viro4d359502014-03-14 12:20:17 -04003714EXPORT_SYMBOL(vfs_mknod);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003715
Al Virof69aac02011-07-26 04:31:40 -04003716static int may_mknod(umode_t mode)
Dave Hansen463c3192008-02-15 14:37:57 -08003717{
3718 switch (mode & S_IFMT) {
3719 case S_IFREG:
3720 case S_IFCHR:
3721 case S_IFBLK:
3722 case S_IFIFO:
3723 case S_IFSOCK:
3724 case 0: /* zero mode translates to S_IFREG */
3725 return 0;
3726 case S_IFDIR:
3727 return -EPERM;
3728 default:
3729 return -EINVAL;
3730 }
3731}
3732
Dominik Brodowski87c4e192018-03-11 11:34:50 +01003733long do_mknodat(int dfd, const char __user *filename, umode_t mode,
3734 unsigned int dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003735{
Al Viro2ad94ae2008-07-21 09:32:51 -04003736 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003737 struct path path;
3738 int error;
Jeff Layton972567f2012-12-20 16:00:10 -05003739 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003740
Al Viro8e4bfca2012-07-20 01:17:26 +04003741 error = may_mknod(mode);
3742 if (error)
3743 return error;
Jeff Layton972567f2012-12-20 16:00:10 -05003744retry:
3745 dentry = user_path_create(dfd, filename, &path, lookup_flags);
Al Virodae6ad82011-06-26 11:50:15 -04003746 if (IS_ERR(dentry))
3747 return PTR_ERR(dentry);
Al Viro2ad94ae2008-07-21 09:32:51 -04003748
Al Virodae6ad82011-06-26 11:50:15 -04003749 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003750 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003751 error = security_path_mknod(&path, dentry, mode, dev);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003752 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04003753 goto out;
Dave Hansen463c3192008-02-15 14:37:57 -08003754 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003755 case 0: case S_IFREG:
Al Viro312b63f2012-06-10 18:09:36 -04003756 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
Mimi Zohar05d1a712016-02-29 19:52:05 -05003757 if (!error)
3758 ima_post_path_mknod(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003759 break;
3760 case S_IFCHR: case S_IFBLK:
Al Virodae6ad82011-06-26 11:50:15 -04003761 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003762 new_decode_dev(dev));
3763 break;
3764 case S_IFIFO: case S_IFSOCK:
Al Virodae6ad82011-06-26 11:50:15 -04003765 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003766 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003767 }
Al Viroa8104a92012-07-20 02:25:00 +04003768out:
Al Viro921a1652012-07-20 01:15:31 +04003769 done_path_create(&path, dentry);
Jeff Layton972567f2012-12-20 16:00:10 -05003770 if (retry_estale(error, lookup_flags)) {
3771 lookup_flags |= LOOKUP_REVAL;
3772 goto retry;
3773 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003774 return error;
3775}
3776
Dominik Brodowski87c4e192018-03-11 11:34:50 +01003777SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3778 unsigned int, dev)
3779{
3780 return do_mknodat(dfd, filename, mode, dev);
3781}
3782
Al Viro8208a222011-07-25 17:32:17 -04003783SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003784{
Dominik Brodowski87c4e192018-03-11 11:34:50 +01003785 return do_mknodat(AT_FDCWD, filename, mode, dev);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003786}
3787
Al Viro18bb1db2011-07-26 01:41:39 -04003788int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003789{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003790 int error = may_create(dir, dentry);
Al Viro8de52772012-02-06 12:45:27 -05003791 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003792
3793 if (error)
3794 return error;
3795
Al Viroacfa4382008-12-04 10:06:33 -05003796 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003797 return -EPERM;
3798
3799 mode &= (S_IRWXUGO|S_ISVTX);
3800 error = security_inode_mkdir(dir, dentry, mode);
3801 if (error)
3802 return error;
3803
Al Viro8de52772012-02-06 12:45:27 -05003804 if (max_links && dir->i_nlink >= max_links)
3805 return -EMLINK;
3806
Linus Torvalds1da177e2005-04-16 15:20:36 -07003807 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003808 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003809 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003810 return error;
3811}
Al Viro4d359502014-03-14 12:20:17 -04003812EXPORT_SYMBOL(vfs_mkdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003813
Dominik Brodowski0101db72018-03-11 11:34:49 +01003814long do_mkdirat(int dfd, const char __user *pathname, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003815{
Dave Hansen6902d922006-09-30 23:29:01 -07003816 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003817 struct path path;
3818 int error;
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003819 unsigned int lookup_flags = LOOKUP_DIRECTORY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003820
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003821retry:
3822 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07003823 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003824 return PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003825
Al Virodae6ad82011-06-26 11:50:15 -04003826 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003827 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003828 error = security_path_mkdir(&path, dentry, mode);
Al Viroa8104a92012-07-20 02:25:00 +04003829 if (!error)
3830 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
Al Viro921a1652012-07-20 01:15:31 +04003831 done_path_create(&path, dentry);
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003832 if (retry_estale(error, lookup_flags)) {
3833 lookup_flags |= LOOKUP_REVAL;
3834 goto retry;
3835 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003836 return error;
3837}
3838
Dominik Brodowski0101db72018-03-11 11:34:49 +01003839SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3840{
3841 return do_mkdirat(dfd, pathname, mode);
3842}
3843
Al Viroa218d0f2011-11-21 14:59:34 -05003844SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003845{
Dominik Brodowski0101db72018-03-11 11:34:49 +01003846 return do_mkdirat(AT_FDCWD, pathname, mode);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003847}
3848
Linus Torvalds1da177e2005-04-16 15:20:36 -07003849int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3850{
3851 int error = may_delete(dir, dentry, 1);
3852
3853 if (error)
3854 return error;
3855
Al Viroacfa4382008-12-04 10:06:33 -05003856 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003857 return -EPERM;
3858
Al Viro1d2ef592011-09-14 18:55:41 +01003859 dget(dentry);
Al Viro59551022016-01-22 15:40:57 -05003860 inode_lock(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003861
Sage Weil912dbc12011-05-24 13:06:11 -07003862 error = -EBUSY;
Eric W. Biederman7af13642013-10-04 19:15:13 -07003863 if (is_local_mountpoint(dentry))
Sage Weil912dbc12011-05-24 13:06:11 -07003864 goto out;
3865
3866 error = security_inode_rmdir(dir, dentry);
3867 if (error)
3868 goto out;
3869
3870 error = dir->i_op->rmdir(dir, dentry);
3871 if (error)
3872 goto out;
3873
Al Viro87677122018-05-27 16:23:51 -04003874 shrink_dcache_parent(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003875 dentry->d_inode->i_flags |= S_DEAD;
3876 dont_mount(dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003877 detach_mounts(dentry);
Amir Goldstein116b9732019-05-26 17:34:02 +03003878 fsnotify_rmdir(dir, dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003879
3880out:
Al Viro59551022016-01-22 15:40:57 -05003881 inode_unlock(dentry->d_inode);
Al Viro1d2ef592011-09-14 18:55:41 +01003882 dput(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003883 if (!error)
3884 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003885 return error;
3886}
Al Viro4d359502014-03-14 12:20:17 -04003887EXPORT_SYMBOL(vfs_rmdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003888
Dominik Brodowskif459dffa2018-03-11 11:34:48 +01003889long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003890{
3891 int error = 0;
Jeff Layton91a27b22012-10-10 15:25:28 -04003892 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003893 struct dentry *dentry;
Al Virof5beed72015-04-30 16:09:11 -04003894 struct path path;
3895 struct qstr last;
3896 int type;
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003897 unsigned int lookup_flags = 0;
3898retry:
Al Viroc1d4dd22016-06-05 16:38:18 -04003899 name = filename_parentat(dfd, getname(pathname), lookup_flags,
3900 &path, &last, &type);
Jeff Layton91a27b22012-10-10 15:25:28 -04003901 if (IS_ERR(name))
3902 return PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003903
Al Virof5beed72015-04-30 16:09:11 -04003904 switch (type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003905 case LAST_DOTDOT:
3906 error = -ENOTEMPTY;
3907 goto exit1;
3908 case LAST_DOT:
3909 error = -EINVAL;
3910 goto exit1;
3911 case LAST_ROOT:
3912 error = -EBUSY;
3913 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003914 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003915
Al Virof5beed72015-04-30 16:09:11 -04003916 error = mnt_want_write(path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02003917 if (error)
3918 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003919
Al Viro59551022016-01-22 15:40:57 -05003920 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
Al Virof5beed72015-04-30 16:09:11 -04003921 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003922 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003923 if (IS_ERR(dentry))
3924 goto exit2;
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003925 if (!dentry->d_inode) {
3926 error = -ENOENT;
3927 goto exit3;
3928 }
Al Virof5beed72015-04-30 16:09:11 -04003929 error = security_path_rmdir(&path, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003930 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003931 goto exit3;
Al Virof5beed72015-04-30 16:09:11 -04003932 error = vfs_rmdir(path.dentry->d_inode, dentry);
Dave Hansen06227532008-02-15 14:37:34 -08003933exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07003934 dput(dentry);
3935exit2:
Al Viro59551022016-01-22 15:40:57 -05003936 inode_unlock(path.dentry->d_inode);
Al Virof5beed72015-04-30 16:09:11 -04003937 mnt_drop_write(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003938exit1:
Al Virof5beed72015-04-30 16:09:11 -04003939 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003940 putname(name);
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003941 if (retry_estale(error, lookup_flags)) {
3942 lookup_flags |= LOOKUP_REVAL;
3943 goto retry;
3944 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003945 return error;
3946}
3947
Heiko Carstens3cdad422009-01-14 14:14:22 +01003948SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003949{
3950 return do_rmdir(AT_FDCWD, pathname);
3951}
3952
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003953/**
3954 * vfs_unlink - unlink a filesystem object
3955 * @dir: parent directory
3956 * @dentry: victim
3957 * @delegated_inode: returns victim inode, if the inode is delegated.
3958 *
3959 * The caller must hold dir->i_mutex.
3960 *
3961 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3962 * return a reference to the inode in delegated_inode. The caller
3963 * should then break the delegation on that inode and retry. Because
3964 * breaking a delegation may take a long time, the caller should drop
3965 * dir->i_mutex before doing so.
3966 *
3967 * Alternatively, a caller may pass NULL for delegated_inode. This may
3968 * be appropriate for callers that expect the underlying filesystem not
3969 * to be NFS exported.
3970 */
3971int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003972{
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003973 struct inode *target = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003974 int error = may_delete(dir, dentry, 0);
3975
3976 if (error)
3977 return error;
3978
Al Viroacfa4382008-12-04 10:06:33 -05003979 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003980 return -EPERM;
3981
Al Viro59551022016-01-22 15:40:57 -05003982 inode_lock(target);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003983 if (is_local_mountpoint(dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003984 error = -EBUSY;
3985 else {
3986 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05003987 if (!error) {
J. Bruce Fields5a146962012-08-28 07:50:40 -07003988 error = try_break_deleg(target, delegated_inode);
3989 if (error)
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003990 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003991 error = dir->i_op->unlink(dir, dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003992 if (!error) {
Al Virod83c49f2010-04-30 17:17:09 -04003993 dont_mount(dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003994 detach_mounts(dentry);
Amir Goldstein116b9732019-05-26 17:34:02 +03003995 fsnotify_unlink(dir, dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003996 }
Al Virobec10522010-03-03 14:12:08 -05003997 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003998 }
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003999out:
Al Viro59551022016-01-22 15:40:57 -05004000 inode_unlock(target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004001
4002 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
4003 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
J. Bruce Fields9accbb92012-08-28 07:03:24 -04004004 fsnotify_link_count(target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004005 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004006 }
Robert Love0eeca282005-07-12 17:06:03 -04004007
Linus Torvalds1da177e2005-04-16 15:20:36 -07004008 return error;
4009}
Al Viro4d359502014-03-14 12:20:17 -04004010EXPORT_SYMBOL(vfs_unlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004011
4012/*
4013 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004014 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07004015 * writeout happening, and we don't want to prevent access to the directory
4016 * while waiting on the I/O.
4017 */
Christoph Hellwigda2f1362017-11-04 13:44:45 +03004018long do_unlinkat(int dfd, struct filename *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004019{
Al Viro2ad94ae2008-07-21 09:32:51 -04004020 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004021 struct dentry *dentry;
Al Virof5beed72015-04-30 16:09:11 -04004022 struct path path;
4023 struct qstr last;
4024 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004025 struct inode *inode = NULL;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04004026 struct inode *delegated_inode = NULL;
Jeff Layton5d18f812012-12-20 16:38:04 -05004027 unsigned int lookup_flags = 0;
4028retry:
Christoph Hellwigda2f1362017-11-04 13:44:45 +03004029 name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
Jeff Layton91a27b22012-10-10 15:25:28 -04004030 if (IS_ERR(name))
4031 return PTR_ERR(name);
Al Viro2ad94ae2008-07-21 09:32:51 -04004032
Linus Torvalds1da177e2005-04-16 15:20:36 -07004033 error = -EISDIR;
Al Virof5beed72015-04-30 16:09:11 -04004034 if (type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004035 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09004036
Al Virof5beed72015-04-30 16:09:11 -04004037 error = mnt_want_write(path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02004038 if (error)
4039 goto exit1;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04004040retry_deleg:
Al Viro59551022016-01-22 15:40:57 -05004041 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
Al Virof5beed72015-04-30 16:09:11 -04004042 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004043 error = PTR_ERR(dentry);
4044 if (!IS_ERR(dentry)) {
4045 /* Why not before? Because we want correct error value */
Al Virof5beed72015-04-30 16:09:11 -04004046 if (last.name[last.len])
Török Edwin50338b82011-06-16 00:06:14 +03004047 goto slashes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004048 inode = dentry->d_inode;
David Howellsb18825a2013-09-12 19:22:53 +01004049 if (d_is_negative(dentry))
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04004050 goto slashes;
4051 ihold(inode);
Al Virof5beed72015-04-30 16:09:11 -04004052 error = security_path_unlink(&path, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004053 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02004054 goto exit2;
Al Virof5beed72015-04-30 16:09:11 -04004055 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
Jan Karac30dabf2012-06-12 16:20:30 +02004056exit2:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004057 dput(dentry);
4058 }
Al Viro59551022016-01-22 15:40:57 -05004059 inode_unlock(path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004060 if (inode)
4061 iput(inode); /* truncate the inode here */
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04004062 inode = NULL;
4063 if (delegated_inode) {
J. Bruce Fields5a146962012-08-28 07:50:40 -07004064 error = break_deleg_wait(&delegated_inode);
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04004065 if (!error)
4066 goto retry_deleg;
4067 }
Al Virof5beed72015-04-30 16:09:11 -04004068 mnt_drop_write(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004069exit1:
Al Virof5beed72015-04-30 16:09:11 -04004070 path_put(&path);
Jeff Layton5d18f812012-12-20 16:38:04 -05004071 if (retry_estale(error, lookup_flags)) {
4072 lookup_flags |= LOOKUP_REVAL;
4073 inode = NULL;
4074 goto retry;
4075 }
Christoph Hellwigda2f1362017-11-04 13:44:45 +03004076 putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004077 return error;
4078
4079slashes:
David Howellsb18825a2013-09-12 19:22:53 +01004080 if (d_is_negative(dentry))
4081 error = -ENOENT;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02004082 else if (d_is_dir(dentry))
David Howellsb18825a2013-09-12 19:22:53 +01004083 error = -EISDIR;
4084 else
4085 error = -ENOTDIR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004086 goto exit2;
4087}
4088
Heiko Carstens2e4d0922009-01-14 14:14:31 +01004089SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004090{
4091 if ((flag & ~AT_REMOVEDIR) != 0)
4092 return -EINVAL;
4093
4094 if (flag & AT_REMOVEDIR)
4095 return do_rmdir(dfd, pathname);
4096
Christoph Hellwigda2f1362017-11-04 13:44:45 +03004097 return do_unlinkat(dfd, getname(pathname));
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004098}
4099
Heiko Carstens3480b252009-01-14 14:14:16 +01004100SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004101{
Christoph Hellwigda2f1362017-11-04 13:44:45 +03004102 return do_unlinkat(AT_FDCWD, getname(pathname));
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004103}
4104
Miklos Szeredidb2e7472008-06-24 16:50:16 +02004105int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004106{
Miklos Szeredia95164d2008-07-30 15:08:48 +02004107 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004108
4109 if (error)
4110 return error;
4111
Al Viroacfa4382008-12-04 10:06:33 -05004112 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004113 return -EPERM;
4114
4115 error = security_inode_symlink(dir, dentry, oldname);
4116 if (error)
4117 return error;
4118
Linus Torvalds1da177e2005-04-16 15:20:36 -07004119 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07004120 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00004121 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004122 return error;
4123}
Al Viro4d359502014-03-14 12:20:17 -04004124EXPORT_SYMBOL(vfs_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004125
Dominik Brodowskib724e842018-03-11 11:34:49 +01004126long do_symlinkat(const char __user *oldname, int newdfd,
4127 const char __user *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004128{
Al Viro2ad94ae2008-07-21 09:32:51 -04004129 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04004130 struct filename *from;
Dave Hansen6902d922006-09-30 23:29:01 -07004131 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04004132 struct path path;
Jeff Laytonf46d3562012-12-11 12:10:08 -05004133 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004134
4135 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04004136 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004137 return PTR_ERR(from);
Jeff Laytonf46d3562012-12-11 12:10:08 -05004138retry:
4139 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07004140 error = PTR_ERR(dentry);
4141 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04004142 goto out_putname;
Dave Hansen6902d922006-09-30 23:29:01 -07004143
Jeff Layton91a27b22012-10-10 15:25:28 -04004144 error = security_path_symlink(&path, dentry, from->name);
Al Viroa8104a92012-07-20 02:25:00 +04004145 if (!error)
Jeff Layton91a27b22012-10-10 15:25:28 -04004146 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
Al Viro921a1652012-07-20 01:15:31 +04004147 done_path_create(&path, dentry);
Jeff Laytonf46d3562012-12-11 12:10:08 -05004148 if (retry_estale(error, lookup_flags)) {
4149 lookup_flags |= LOOKUP_REVAL;
4150 goto retry;
4151 }
Dave Hansen6902d922006-09-30 23:29:01 -07004152out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004153 putname(from);
4154 return error;
4155}
4156
Dominik Brodowskib724e842018-03-11 11:34:49 +01004157SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4158 int, newdfd, const char __user *, newname)
4159{
4160 return do_symlinkat(oldname, newdfd, newname);
4161}
4162
Heiko Carstens3480b252009-01-14 14:14:16 +01004163SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004164{
Dominik Brodowskib724e842018-03-11 11:34:49 +01004165 return do_symlinkat(oldname, AT_FDCWD, newname);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004166}
4167
J. Bruce Fields146a8592011-09-20 17:14:31 -04004168/**
4169 * vfs_link - create a new link
4170 * @old_dentry: object to be linked
4171 * @dir: new parent
4172 * @new_dentry: where to create the new link
4173 * @delegated_inode: returns inode needing a delegation break
4174 *
4175 * The caller must hold dir->i_mutex
4176 *
4177 * If vfs_link discovers a delegation on the to-be-linked file in need
4178 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4179 * inode in delegated_inode. The caller should then break the delegation
4180 * and retry. Because breaking a delegation may take a long time, the
4181 * caller should drop the i_mutex before doing so.
4182 *
4183 * Alternatively, a caller may pass NULL for delegated_inode. This may
4184 * be appropriate for callers that expect the underlying filesystem not
4185 * to be NFS exported.
4186 */
4187int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004188{
4189 struct inode *inode = old_dentry->d_inode;
Al Viro8de52772012-02-06 12:45:27 -05004190 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004191 int error;
4192
4193 if (!inode)
4194 return -ENOENT;
4195
Miklos Szeredia95164d2008-07-30 15:08:48 +02004196 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004197 if (error)
4198 return error;
4199
4200 if (dir->i_sb != inode->i_sb)
4201 return -EXDEV;
4202
4203 /*
4204 * A link to an append-only or immutable file cannot be created.
4205 */
4206 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4207 return -EPERM;
Eric W. Biederman0bd23d092016-06-29 14:54:46 -05004208 /*
4209 * Updating the link count will likely cause i_uid and i_gid to
4210 * be writen back improperly if their true value is unknown to
4211 * the vfs.
4212 */
4213 if (HAS_UNMAPPED_ID(inode))
4214 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05004215 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004216 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02004217 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004218 return -EPERM;
4219
4220 error = security_inode_link(old_dentry, dir, new_dentry);
4221 if (error)
4222 return error;
4223
Al Viro59551022016-01-22 15:40:57 -05004224 inode_lock(inode);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05304225 /* Make sure we don't allow creating hardlink to an unlinked file */
Al Virof4e0c302013-06-11 08:34:36 +04004226 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05304227 error = -ENOENT;
Al Viro8de52772012-02-06 12:45:27 -05004228 else if (max_links && inode->i_nlink >= max_links)
4229 error = -EMLINK;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004230 else {
4231 error = try_break_deleg(inode, delegated_inode);
4232 if (!error)
4233 error = dir->i_op->link(old_dentry, dir, new_dentry);
4234 }
Al Virof4e0c302013-06-11 08:34:36 +04004235
4236 if (!error && (inode->i_state & I_LINKABLE)) {
4237 spin_lock(&inode->i_lock);
4238 inode->i_state &= ~I_LINKABLE;
4239 spin_unlock(&inode->i_lock);
4240 }
Al Viro59551022016-01-22 15:40:57 -05004241 inode_unlock(inode);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07004242 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02004243 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004244 return error;
4245}
Al Viro4d359502014-03-14 12:20:17 -04004246EXPORT_SYMBOL(vfs_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004247
4248/*
4249 * Hardlinks are often used in delicate situations. We avoid
4250 * security-related surprises by not following symlinks on the
4251 * newname. --KAB
4252 *
4253 * We don't follow them on the oldname either to be compatible
4254 * with linux 2.0, and to avoid hard-linking to directories
4255 * and other special files. --ADM
4256 */
Dominik Brodowski46ea89e2018-03-11 11:34:53 +01004257int do_linkat(int olddfd, const char __user *oldname, int newdfd,
4258 const char __user *newname, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004259{
4260 struct dentry *new_dentry;
Al Virodae6ad82011-06-26 11:50:15 -04004261 struct path old_path, new_path;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004262 struct inode *delegated_inode = NULL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304263 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004264 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004265
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304266 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004267 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304268 /*
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004269 * To use null names we require CAP_DAC_READ_SEARCH
4270 * This ensures that not everyone will be able to create
4271 * handlink using the passed filedescriptor.
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304272 */
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004273 if (flags & AT_EMPTY_PATH) {
4274 if (!capable(CAP_DAC_READ_SEARCH))
4275 return -ENOENT;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304276 how = LOOKUP_EMPTY;
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004277 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004278
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304279 if (flags & AT_SYMLINK_FOLLOW)
4280 how |= LOOKUP_FOLLOW;
Jeff Layton442e31c2012-12-20 16:15:38 -05004281retry:
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304282 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004283 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04004284 return error;
4285
Jeff Layton442e31c2012-12-20 16:15:38 -05004286 new_dentry = user_path_create(newdfd, newname, &new_path,
4287 (how & LOOKUP_REVAL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004288 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07004289 if (IS_ERR(new_dentry))
Al Virodae6ad82011-06-26 11:50:15 -04004290 goto out;
4291
4292 error = -EXDEV;
4293 if (old_path.mnt != new_path.mnt)
4294 goto out_dput;
Kees Cook800179c2012-07-25 17:29:07 -07004295 error = may_linkat(&old_path);
4296 if (unlikely(error))
4297 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04004298 error = security_path_link(old_path.dentry, &new_path, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004299 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04004300 goto out_dput;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004301 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
Dave Hansen75c3f292008-02-15 14:37:45 -08004302out_dput:
Al Viro921a1652012-07-20 01:15:31 +04004303 done_path_create(&new_path, new_dentry);
J. Bruce Fields146a8592011-09-20 17:14:31 -04004304 if (delegated_inode) {
4305 error = break_deleg_wait(&delegated_inode);
Oleg Drokind22e6332014-01-31 15:41:58 -05004306 if (!error) {
4307 path_put(&old_path);
J. Bruce Fields146a8592011-09-20 17:14:31 -04004308 goto retry;
Oleg Drokind22e6332014-01-31 15:41:58 -05004309 }
J. Bruce Fields146a8592011-09-20 17:14:31 -04004310 }
Jeff Layton442e31c2012-12-20 16:15:38 -05004311 if (retry_estale(error, how)) {
Oleg Drokind22e6332014-01-31 15:41:58 -05004312 path_put(&old_path);
Jeff Layton442e31c2012-12-20 16:15:38 -05004313 how |= LOOKUP_REVAL;
4314 goto retry;
4315 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004316out:
Al Viro2d8f3032008-07-22 09:59:21 -04004317 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004318
4319 return error;
4320}
4321
Dominik Brodowski46ea89e2018-03-11 11:34:53 +01004322SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4323 int, newdfd, const char __user *, newname, int, flags)
4324{
4325 return do_linkat(olddfd, oldname, newdfd, newname, flags);
4326}
4327
Heiko Carstens3480b252009-01-14 14:14:16 +01004328SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004329{
Dominik Brodowski46ea89e2018-03-11 11:34:53 +01004330 return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004331}
4332
Miklos Szeredibc270272014-04-01 17:08:42 +02004333/**
4334 * vfs_rename - rename a filesystem object
4335 * @old_dir: parent of source
4336 * @old_dentry: source
4337 * @new_dir: parent of destination
4338 * @new_dentry: destination
4339 * @delegated_inode: returns an inode needing a delegation break
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004340 * @flags: rename flags
Miklos Szeredibc270272014-04-01 17:08:42 +02004341 *
4342 * The caller must hold multiple mutexes--see lock_rename()).
4343 *
4344 * If vfs_rename discovers a delegation in need of breaking at either
4345 * the source or destination, it will return -EWOULDBLOCK and return a
4346 * reference to the inode in delegated_inode. The caller should then
4347 * break the delegation and retry. Because breaking a delegation may
4348 * take a long time, the caller should drop all locks before doing
4349 * so.
4350 *
4351 * Alternatively, a caller may pass NULL for delegated_inode. This may
4352 * be appropriate for callers that expect the underlying filesystem not
4353 * to be NFS exported.
4354 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07004355 * The worst of all namespace operations - renaming directory. "Perverted"
4356 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4357 * Problems:
Mauro Carvalho Chehab0117d422017-05-12 07:45:42 -03004358 *
J. Bruce Fieldsd03b29a22014-02-17 16:52:33 -05004359 * a) we can get into loop creation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004360 * b) race potential - two innocent renames can create a loop together.
4361 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004362 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07004363 * story.
J. Bruce Fields6cedba82012-03-05 11:40:41 -05004364 * c) we have to lock _four_ objects - parents and victim (if it exists),
4365 * and source (if it is not a directory).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004366 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07004367 * whether the target exists). Solution: try to be smart with locking
4368 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004369 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07004370 * move will be locked. Thus we can rank directories by the tree
4371 * (ancestors first) and rank all non-directories after them.
4372 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004373 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004374 * HOWEVER, it relies on the assumption that any object with ->lookup()
4375 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4376 * we'd better make sure that there's no link(2) for them.
Sage Weile4eaac02011-05-24 13:06:07 -07004377 * d) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004378 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07004379 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05004380 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07004381 * locking].
4382 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004383int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004384 struct inode *new_dir, struct dentry *new_dentry,
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004385 struct inode **delegated_inode, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004386{
4387 int error;
Miklos Szeredibc270272014-04-01 17:08:42 +02004388 bool is_dir = d_is_dir(old_dentry);
Miklos Szeredibc270272014-04-01 17:08:42 +02004389 struct inode *source = old_dentry->d_inode;
4390 struct inode *target = new_dentry->d_inode;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004391 bool new_is_dir = false;
4392 unsigned max_links = new_dir->i_sb->s_max_links;
Al Viro49d31c22017-07-07 14:51:19 -04004393 struct name_snapshot old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004394
Miklos Szeredi8d3e2932016-12-16 11:02:54 +01004395 if (source == target)
Miklos Szeredibc270272014-04-01 17:08:42 +02004396 return 0;
4397
Linus Torvalds1da177e2005-04-16 15:20:36 -07004398 error = may_delete(old_dir, old_dentry, is_dir);
4399 if (error)
4400 return error;
4401
Miklos Szeredida1ce062014-04-01 17:08:43 +02004402 if (!target) {
Miklos Szeredia95164d2008-07-30 15:08:48 +02004403 error = may_create(new_dir, new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004404 } else {
4405 new_is_dir = d_is_dir(new_dentry);
4406
4407 if (!(flags & RENAME_EXCHANGE))
4408 error = may_delete(new_dir, new_dentry, is_dir);
4409 else
4410 error = may_delete(new_dir, new_dentry, new_is_dir);
4411 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004412 if (error)
4413 return error;
4414
Miklos Szeredi2773bf02016-09-27 11:03:58 +02004415 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004416 return -EPERM;
4417
Miklos Szeredibc270272014-04-01 17:08:42 +02004418 /*
4419 * If we are going to change the parent - check write permissions,
4420 * we'll need to flip '..'.
4421 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02004422 if (new_dir != old_dir) {
4423 if (is_dir) {
4424 error = inode_permission(source, MAY_WRITE);
4425 if (error)
4426 return error;
4427 }
4428 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4429 error = inode_permission(target, MAY_WRITE);
4430 if (error)
4431 return error;
4432 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004433 }
Robert Love0eeca282005-07-12 17:06:03 -04004434
Miklos Szeredi0b3974e2014-04-01 17:08:43 +02004435 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4436 flags);
Miklos Szeredibc270272014-04-01 17:08:42 +02004437 if (error)
4438 return error;
4439
Al Viro49d31c22017-07-07 14:51:19 -04004440 take_dentry_name_snapshot(&old_name, old_dentry);
Miklos Szeredibc270272014-04-01 17:08:42 +02004441 dget(new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004442 if (!is_dir || (flags & RENAME_EXCHANGE))
Miklos Szeredibc270272014-04-01 17:08:42 +02004443 lock_two_nondirectories(source, target);
4444 else if (target)
Al Viro59551022016-01-22 15:40:57 -05004445 inode_lock(target);
Miklos Szeredibc270272014-04-01 17:08:42 +02004446
4447 error = -EBUSY;
Eric W. Biederman7af13642013-10-04 19:15:13 -07004448 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
Miklos Szeredibc270272014-04-01 17:08:42 +02004449 goto out;
4450
Miklos Szeredida1ce062014-04-01 17:08:43 +02004451 if (max_links && new_dir != old_dir) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004452 error = -EMLINK;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004453 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
Miklos Szeredibc270272014-04-01 17:08:42 +02004454 goto out;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004455 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4456 old_dir->i_nlink >= max_links)
4457 goto out;
4458 }
Miklos Szeredida1ce062014-04-01 17:08:43 +02004459 if (!is_dir) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004460 error = try_break_deleg(source, delegated_inode);
4461 if (error)
4462 goto out;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004463 }
4464 if (target && !new_is_dir) {
4465 error = try_break_deleg(target, delegated_inode);
4466 if (error)
4467 goto out;
Miklos Szeredibc270272014-04-01 17:08:42 +02004468 }
Miklos Szeredi2773bf02016-09-27 11:03:58 +02004469 error = old_dir->i_op->rename(old_dir, old_dentry,
Miklos Szeredi18fc84d2016-09-27 11:03:58 +02004470 new_dir, new_dentry, flags);
Miklos Szeredibc270272014-04-01 17:08:42 +02004471 if (error)
4472 goto out;
4473
Miklos Szeredida1ce062014-04-01 17:08:43 +02004474 if (!(flags & RENAME_EXCHANGE) && target) {
Al Viro87677122018-05-27 16:23:51 -04004475 if (is_dir) {
4476 shrink_dcache_parent(new_dentry);
Miklos Szeredibc270272014-04-01 17:08:42 +02004477 target->i_flags |= S_DEAD;
Al Viro87677122018-05-27 16:23:51 -04004478 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004479 dont_mount(new_dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07004480 detach_mounts(new_dentry);
Miklos Szeredibc270272014-04-01 17:08:42 +02004481 }
Miklos Szeredida1ce062014-04-01 17:08:43 +02004482 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4483 if (!(flags & RENAME_EXCHANGE))
4484 d_move(old_dentry, new_dentry);
4485 else
4486 d_exchange(old_dentry, new_dentry);
4487 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004488out:
Miklos Szeredida1ce062014-04-01 17:08:43 +02004489 if (!is_dir || (flags & RENAME_EXCHANGE))
Miklos Szeredibc270272014-04-01 17:08:42 +02004490 unlock_two_nondirectories(source, target);
4491 else if (target)
Al Viro59551022016-01-22 15:40:57 -05004492 inode_unlock(target);
Miklos Szeredibc270272014-04-01 17:08:42 +02004493 dput(new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004494 if (!error) {
Al Virof4ec3a32019-04-26 13:21:24 -04004495 fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
Miklos Szeredida1ce062014-04-01 17:08:43 +02004496 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4497 if (flags & RENAME_EXCHANGE) {
Al Virof4ec3a32019-04-26 13:21:24 -04004498 fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
Miklos Szeredida1ce062014-04-01 17:08:43 +02004499 new_is_dir, NULL, new_dentry);
4500 }
4501 }
Al Viro49d31c22017-07-07 14:51:19 -04004502 release_dentry_name_snapshot(&old_name);
Robert Love0eeca282005-07-12 17:06:03 -04004503
Linus Torvalds1da177e2005-04-16 15:20:36 -07004504 return error;
4505}
Al Viro4d359502014-03-14 12:20:17 -04004506EXPORT_SYMBOL(vfs_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004507
Dominik Brodowskiee81feb2018-03-11 11:34:28 +01004508static int do_renameat2(int olddfd, const char __user *oldname, int newdfd,
4509 const char __user *newname, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004510{
Al Viro2ad94ae2008-07-21 09:32:51 -04004511 struct dentry *old_dentry, *new_dentry;
4512 struct dentry *trap;
Al Virof5beed72015-04-30 16:09:11 -04004513 struct path old_path, new_path;
4514 struct qstr old_last, new_last;
4515 int old_type, new_type;
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004516 struct inode *delegated_inode = NULL;
Jeff Layton91a27b22012-10-10 15:25:28 -04004517 struct filename *from;
4518 struct filename *to;
Al Virof5beed72015-04-30 16:09:11 -04004519 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
Jeff Laytonc6a94282012-12-11 12:10:10 -05004520 bool should_retry = false;
Al Viro2ad94ae2008-07-21 09:32:51 -04004521 int error;
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004522
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004523 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredida1ce062014-04-01 17:08:43 +02004524 return -EINVAL;
4525
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004526 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4527 (flags & RENAME_EXCHANGE))
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004528 return -EINVAL;
4529
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004530 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4531 return -EPERM;
4532
Al Virof5beed72015-04-30 16:09:11 -04004533 if (flags & RENAME_EXCHANGE)
4534 target_flags = 0;
4535
Jeff Laytonc6a94282012-12-11 12:10:10 -05004536retry:
Al Viroc1d4dd22016-06-05 16:38:18 -04004537 from = filename_parentat(olddfd, getname(oldname), lookup_flags,
4538 &old_path, &old_last, &old_type);
Jeff Layton91a27b22012-10-10 15:25:28 -04004539 if (IS_ERR(from)) {
4540 error = PTR_ERR(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004541 goto exit;
Jeff Layton91a27b22012-10-10 15:25:28 -04004542 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004543
Al Viroc1d4dd22016-06-05 16:38:18 -04004544 to = filename_parentat(newdfd, getname(newname), lookup_flags,
4545 &new_path, &new_last, &new_type);
Jeff Layton91a27b22012-10-10 15:25:28 -04004546 if (IS_ERR(to)) {
4547 error = PTR_ERR(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004548 goto exit1;
Jeff Layton91a27b22012-10-10 15:25:28 -04004549 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004550
4551 error = -EXDEV;
Al Virof5beed72015-04-30 16:09:11 -04004552 if (old_path.mnt != new_path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004553 goto exit2;
4554
Linus Torvalds1da177e2005-04-16 15:20:36 -07004555 error = -EBUSY;
Al Virof5beed72015-04-30 16:09:11 -04004556 if (old_type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004557 goto exit2;
4558
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004559 if (flags & RENAME_NOREPLACE)
4560 error = -EEXIST;
Al Virof5beed72015-04-30 16:09:11 -04004561 if (new_type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004562 goto exit2;
4563
Al Virof5beed72015-04-30 16:09:11 -04004564 error = mnt_want_write(old_path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02004565 if (error)
4566 goto exit2;
4567
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004568retry_deleg:
Al Virof5beed72015-04-30 16:09:11 -04004569 trap = lock_rename(new_path.dentry, old_path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004570
Al Virof5beed72015-04-30 16:09:11 -04004571 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004572 error = PTR_ERR(old_dentry);
4573 if (IS_ERR(old_dentry))
4574 goto exit3;
4575 /* source must exist */
4576 error = -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01004577 if (d_is_negative(old_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004578 goto exit4;
Al Virof5beed72015-04-30 16:09:11 -04004579 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004580 error = PTR_ERR(new_dentry);
4581 if (IS_ERR(new_dentry))
4582 goto exit4;
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004583 error = -EEXIST;
4584 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4585 goto exit5;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004586 if (flags & RENAME_EXCHANGE) {
4587 error = -ENOENT;
4588 if (d_is_negative(new_dentry))
4589 goto exit5;
4590
4591 if (!d_is_dir(new_dentry)) {
4592 error = -ENOTDIR;
Al Virof5beed72015-04-30 16:09:11 -04004593 if (new_last.name[new_last.len])
Miklos Szeredida1ce062014-04-01 17:08:43 +02004594 goto exit5;
4595 }
4596 }
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004597 /* unless the source is a directory trailing slashes give -ENOTDIR */
4598 if (!d_is_dir(old_dentry)) {
4599 error = -ENOTDIR;
Al Virof5beed72015-04-30 16:09:11 -04004600 if (old_last.name[old_last.len])
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004601 goto exit5;
Al Virof5beed72015-04-30 16:09:11 -04004602 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004603 goto exit5;
4604 }
4605 /* source should not be ancestor of target */
4606 error = -EINVAL;
4607 if (old_dentry == trap)
4608 goto exit5;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004609 /* target should not be an ancestor of source */
Miklos Szeredida1ce062014-04-01 17:08:43 +02004610 if (!(flags & RENAME_EXCHANGE))
4611 error = -ENOTEMPTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004612 if (new_dentry == trap)
4613 goto exit5;
4614
Al Virof5beed72015-04-30 16:09:11 -04004615 error = security_path_rename(&old_path, old_dentry,
4616 &new_path, new_dentry, flags);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004617 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02004618 goto exit5;
Al Virof5beed72015-04-30 16:09:11 -04004619 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4620 new_path.dentry->d_inode, new_dentry,
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004621 &delegated_inode, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004622exit5:
4623 dput(new_dentry);
4624exit4:
4625 dput(old_dentry);
4626exit3:
Al Virof5beed72015-04-30 16:09:11 -04004627 unlock_rename(new_path.dentry, old_path.dentry);
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004628 if (delegated_inode) {
4629 error = break_deleg_wait(&delegated_inode);
4630 if (!error)
4631 goto retry_deleg;
4632 }
Al Virof5beed72015-04-30 16:09:11 -04004633 mnt_drop_write(old_path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004634exit2:
Jeff Laytonc6a94282012-12-11 12:10:10 -05004635 if (retry_estale(error, lookup_flags))
4636 should_retry = true;
Al Virof5beed72015-04-30 16:09:11 -04004637 path_put(&new_path);
Al Viro2ad94ae2008-07-21 09:32:51 -04004638 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004639exit1:
Al Virof5beed72015-04-30 16:09:11 -04004640 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004641 putname(from);
Jeff Laytonc6a94282012-12-11 12:10:10 -05004642 if (should_retry) {
4643 should_retry = false;
4644 lookup_flags |= LOOKUP_REVAL;
4645 goto retry;
4646 }
Al Viro2ad94ae2008-07-21 09:32:51 -04004647exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004648 return error;
4649}
4650
Dominik Brodowskiee81feb2018-03-11 11:34:28 +01004651SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4652 int, newdfd, const char __user *, newname, unsigned int, flags)
4653{
4654 return do_renameat2(olddfd, oldname, newdfd, newname, flags);
4655}
4656
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004657SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4658 int, newdfd, const char __user *, newname)
4659{
Dominik Brodowskiee81feb2018-03-11 11:34:28 +01004660 return do_renameat2(olddfd, oldname, newdfd, newname, 0);
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004661}
4662
Heiko Carstensa26eab22009-01-14 14:14:17 +01004663SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004664{
Dominik Brodowskiee81feb2018-03-11 11:34:28 +01004665 return do_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004666}
4667
Miklos Szeredi787fb6b2014-10-24 00:14:36 +02004668int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4669{
4670 int error = may_create(dir, dentry);
4671 if (error)
4672 return error;
4673
4674 if (!dir->i_op->mknod)
4675 return -EPERM;
4676
4677 return dir->i_op->mknod(dir, dentry,
4678 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4679}
4680EXPORT_SYMBOL(vfs_whiteout);
4681
Al Viro5d826c82014-03-14 13:42:45 -04004682int readlink_copy(char __user *buffer, int buflen, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004683{
Al Viro5d826c82014-03-14 13:42:45 -04004684 int len = PTR_ERR(link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004685 if (IS_ERR(link))
4686 goto out;
4687
4688 len = strlen(link);
4689 if (len > (unsigned) buflen)
4690 len = buflen;
4691 if (copy_to_user(buffer, link, len))
4692 len = -EFAULT;
4693out:
4694 return len;
4695}
4696
Miklos Szeredid60874c2016-10-04 14:40:45 +02004697/**
Miklos Szeredifd4a0edf2016-12-09 16:45:04 +01004698 * vfs_readlink - copy symlink body into userspace buffer
4699 * @dentry: dentry on which to get symbolic link
4700 * @buffer: user memory pointer
4701 * @buflen: size of buffer
4702 *
4703 * Does not touch atime. That's up to the caller if necessary
4704 *
4705 * Does not call security hook.
4706 */
4707int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4708{
4709 struct inode *inode = d_inode(dentry);
Al Virof2df5da2018-07-19 17:35:51 -04004710 DEFINE_DELAYED_CALL(done);
4711 const char *link;
4712 int res;
Miklos Szeredifd4a0edf2016-12-09 16:45:04 +01004713
Miklos Szeredi76fca902016-12-09 16:45:04 +01004714 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
4715 if (unlikely(inode->i_op->readlink))
4716 return inode->i_op->readlink(dentry, buffer, buflen);
Miklos Szeredifd4a0edf2016-12-09 16:45:04 +01004717
Miklos Szeredi76fca902016-12-09 16:45:04 +01004718 if (!d_is_symlink(dentry))
4719 return -EINVAL;
4720
4721 spin_lock(&inode->i_lock);
4722 inode->i_opflags |= IOP_DEFAULT_READLINK;
4723 spin_unlock(&inode->i_lock);
4724 }
4725
Eric Biggers4c4f7c12019-04-10 13:21:14 -07004726 link = READ_ONCE(inode->i_link);
Al Virof2df5da2018-07-19 17:35:51 -04004727 if (!link) {
4728 link = inode->i_op->get_link(dentry, inode, &done);
4729 if (IS_ERR(link))
4730 return PTR_ERR(link);
4731 }
4732 res = readlink_copy(buffer, buflen, link);
4733 do_delayed_call(&done);
4734 return res;
Miklos Szeredifd4a0edf2016-12-09 16:45:04 +01004735}
4736EXPORT_SYMBOL(vfs_readlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004737
Miklos Szeredid60874c2016-10-04 14:40:45 +02004738/**
4739 * vfs_get_link - get symlink body
4740 * @dentry: dentry on which to get symbolic link
4741 * @done: caller needs to free returned data with this
4742 *
4743 * Calls security hook and i_op->get_link() on the supplied inode.
4744 *
4745 * It does not touch atime. That's up to the caller if necessary.
4746 *
4747 * Does not work on "special" symlinks like /proc/$$/fd/N
4748 */
4749const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4750{
4751 const char *res = ERR_PTR(-EINVAL);
4752 struct inode *inode = d_inode(dentry);
4753
4754 if (d_is_symlink(dentry)) {
4755 res = ERR_PTR(security_inode_readlink(dentry));
4756 if (!res)
4757 res = inode->i_op->get_link(dentry, inode, done);
4758 }
4759 return res;
4760}
4761EXPORT_SYMBOL(vfs_get_link);
4762
Linus Torvalds1da177e2005-04-16 15:20:36 -07004763/* get the link contents into pagecache */
Al Viro6b255392015-11-17 10:20:54 -05004764const char *page_get_link(struct dentry *dentry, struct inode *inode,
Al Virofceef392015-12-29 15:58:39 -05004765 struct delayed_call *callback)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004766{
Duane Griffinebd09ab2008-12-19 20:47:12 +00004767 char *kaddr;
4768 struct page *page;
Al Viro6b255392015-11-17 10:20:54 -05004769 struct address_space *mapping = inode->i_mapping;
4770
Al Virod3883d42015-11-17 10:41:04 -05004771 if (!dentry) {
4772 page = find_get_page(mapping, 0);
4773 if (!page)
4774 return ERR_PTR(-ECHILD);
4775 if (!PageUptodate(page)) {
4776 put_page(page);
4777 return ERR_PTR(-ECHILD);
4778 }
4779 } else {
4780 page = read_mapping_page(mapping, 0, NULL);
4781 if (IS_ERR(page))
4782 return (char*)page;
4783 }
Al Virofceef392015-12-29 15:58:39 -05004784 set_delayed_call(callback, page_put_link, page);
Al Viro21fc61c2015-11-17 01:07:57 -05004785 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4786 kaddr = page_address(page);
Al Viro6b255392015-11-17 10:20:54 -05004787 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
Duane Griffinebd09ab2008-12-19 20:47:12 +00004788 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004789}
4790
Al Viro6b255392015-11-17 10:20:54 -05004791EXPORT_SYMBOL(page_get_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004792
Al Virofceef392015-12-29 15:58:39 -05004793void page_put_link(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004794{
Al Virofceef392015-12-29 15:58:39 -05004795 put_page(arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004796}
Al Viro4d359502014-03-14 12:20:17 -04004797EXPORT_SYMBOL(page_put_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004798
Al Viroaa80dea2015-11-16 18:26:34 -05004799int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4800{
Al Virofceef392015-12-29 15:58:39 -05004801 DEFINE_DELAYED_CALL(done);
Al Viro6b255392015-11-17 10:20:54 -05004802 int res = readlink_copy(buffer, buflen,
4803 page_get_link(dentry, d_inode(dentry),
Al Virofceef392015-12-29 15:58:39 -05004804 &done));
4805 do_delayed_call(&done);
Al Viroaa80dea2015-11-16 18:26:34 -05004806 return res;
4807}
4808EXPORT_SYMBOL(page_readlink);
4809
Nick Piggin54566b22009-01-04 12:00:53 -08004810/*
4811 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4812 */
4813int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004814{
4815 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004816 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07004817 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08004818 int err;
Tetsuo Handac718a972017-05-08 15:58:59 -07004819 unsigned int flags = 0;
Nick Piggin54566b22009-01-04 12:00:53 -08004820 if (nofs)
4821 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004822
NeilBrown7e53cac2006-03-25 03:07:57 -08004823retry:
Nick Pigginafddba42007-10-16 01:25:01 -07004824 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08004825 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004826 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07004827 goto fail;
4828
Al Viro21fc61c2015-11-17 01:07:57 -05004829 memcpy(page_address(page), symname, len-1);
Nick Pigginafddba42007-10-16 01:25:01 -07004830
4831 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4832 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004833 if (err < 0)
4834 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07004835 if (err < len-1)
4836 goto retry;
4837
Linus Torvalds1da177e2005-04-16 15:20:36 -07004838 mark_inode_dirty(inode);
4839 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004840fail:
4841 return err;
4842}
Al Viro4d359502014-03-14 12:20:17 -04004843EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004844
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004845int page_symlink(struct inode *inode, const char *symname, int len)
4846{
4847 return __page_symlink(inode, symname, len,
Michal Hockoc62d2552015-11-06 16:28:49 -08004848 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004849}
Al Viro4d359502014-03-14 12:20:17 -04004850EXPORT_SYMBOL(page_symlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004851
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08004852const struct inode_operations page_symlink_inode_operations = {
Al Viro6b255392015-11-17 10:20:54 -05004853 .get_link = page_get_link,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004854};
Linus Torvalds1da177e2005-04-16 15:20:36 -07004855EXPORT_SYMBOL(page_symlink_inode_operations);