blob: 6571a5f5112ed82894fdbec1c9b6dfaf557d93b1 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050018#include <linux/export.h>
David S. Miller44696902012-05-23 20:12:50 -070019#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/slab.h>
21#include <linux/fs.h>
22#include <linux/namei.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/pagemap.h>
Robert Love0eeca282005-07-12 17:06:03 -040024#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/personality.h>
26#include <linux/security.h>
Mimi Zohar6146f0d2009-02-04 09:06:57 -050027#include <linux/ima.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/syscalls.h>
29#include <linux/mount.h>
30#include <linux/audit.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080031#include <linux/capability.h>
Trond Myklebust834f2a42005-10-18 14:20:16 -070032#include <linux/file.h>
Ulrich Drepper5590ff02006-01-18 17:43:53 -080033#include <linux/fcntl.h>
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -070034#include <linux/device_cgroup.h>
Al Viro5ad4e532009-03-29 19:50:06 -040035#include <linux/fs_struct.h>
Linus Torvaldse77819e2011-07-22 19:30:19 -070036#include <linux/posix_acl.h>
Linus Torvalds99d263d2014-09-13 11:30:10 -070037#include <linux/hash.h>
George Spelvin2a18da7a2016-05-23 07:43:58 -040038#include <linux/bitops.h>
Eric W. Biedermanaeaa4a72016-07-23 11:20:44 -050039#include <linux/init_task.h>
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080040#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Eric Parise81e3f42009-12-04 15:47:36 -050042#include "internal.h"
Al Viroc7105362011-11-24 18:22:03 -050043#include "mount.h"
Eric Parise81e3f42009-12-04 15:47:36 -050044
Linus Torvalds1da177e2005-04-16 15:20:36 -070045/* [Feb-1997 T. Schoebel-Theuer]
46 * Fundamental changes in the pathname lookup mechanisms (namei)
47 * were necessary because of omirr. The reason is that omirr needs
48 * to know the _real_ pathname, not the user-supplied one, in case
49 * of symlinks (and also when transname replacements occur).
50 *
51 * The new code replaces the old recursive symlink resolution with
52 * an iterative one (in case of non-nested symlink chains). It does
53 * this with calls to <fs>_follow_link().
54 * As a side effect, dir_namei(), _namei() and follow_link() are now
55 * replaced with a single function lookup_dentry() that can handle all
56 * the special cases of the former code.
57 *
58 * With the new dcache, the pathname is stored at each inode, at least as
59 * long as the refcount of the inode is positive. As a side effect, the
60 * size of the dcache depends on the inode cache and thus is dynamic.
61 *
62 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
63 * resolution to correspond with current state of the code.
64 *
65 * Note that the symlink resolution is not *completely* iterative.
66 * There is still a significant amount of tail- and mid- recursion in
67 * the algorithm. Also, note that <fs>_readlink() is not used in
68 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
69 * may return different results than <fs>_follow_link(). Many virtual
70 * filesystems (including /proc) exhibit this behavior.
71 */
72
73/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
74 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
75 * and the name already exists in form of a symlink, try to create the new
76 * name indicated by the symlink. The old code always complained that the
77 * name already exists, due to not following the symlink even if its target
78 * is nonexistent. The new semantics affects also mknod() and link() when
Lucas De Marchi25985ed2011-03-30 22:57:33 -030079 * the name is a symlink pointing to a non-existent name.
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 *
81 * I don't know which semantics is the right one, since I have no access
82 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
83 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
84 * "old" one. Personally, I think the new semantics is much more logical.
85 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
86 * file does succeed in both HP-UX and SunOs, but not in Solaris
87 * and in the old Linux semantics.
88 */
89
90/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
91 * semantics. See the comments in "open_namei" and "do_link" below.
92 *
93 * [10-Sep-98 Alan Modra] Another symlink change.
94 */
95
96/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
97 * inside the path - always follow.
98 * in the last component in creation/removal/renaming - never follow.
99 * if LOOKUP_FOLLOW passed - follow.
100 * if the pathname has trailing slashes - follow.
101 * otherwise - don't follow.
102 * (applied in that order).
103 *
104 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
105 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
106 * During the 2.4 we need to fix the userland stuff depending on it -
107 * hopefully we will be able to get rid of that wart in 2.5. So far only
108 * XEmacs seems to be relying on it...
109 */
110/*
111 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800112 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 * any extra contention...
114 */
115
116/* In order to reduce some races, while at the same time doing additional
117 * checking and hopefully speeding things up, we copy filenames to the
118 * kernel data space before using them..
119 *
120 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
121 * PATH_MAX includes the nul terminator --RR.
122 */
Jeff Layton91a27b22012-10-10 15:25:28 -0400123
Al Virofd2f7cb2015-02-22 20:07:13 -0500124#define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
Jeff Layton7950e382012-10-10 16:43:13 -0400125
David Drysdale51f39a12014-12-12 16:57:29 -0800126struct filename *
Jeff Layton91a27b22012-10-10 15:25:28 -0400127getname_flags(const char __user *filename, int flags, int *empty)
128{
Al Viro94b5d262015-02-22 19:38:03 -0500129 struct filename *result;
Jeff Layton7950e382012-10-10 16:43:13 -0400130 char *kname;
Al Viro94b5d262015-02-22 19:38:03 -0500131 int len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132
Jeff Layton7ac86262012-10-10 15:25:28 -0400133 result = audit_reusename(filename);
134 if (result)
135 return result;
136
Jeff Layton7950e382012-10-10 16:43:13 -0400137 result = __getname();
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700138 if (unlikely(!result))
Eric Paris4043cde2012-01-03 14:23:08 -0500139 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140
Jeff Layton7950e382012-10-10 16:43:13 -0400141 /*
142 * First, try to embed the struct filename inside the names_cache
143 * allocation
144 */
Al Virofd2f7cb2015-02-22 20:07:13 -0500145 kname = (char *)result->iname;
Jeff Layton91a27b22012-10-10 15:25:28 -0400146 result->name = kname;
Jeff Layton7950e382012-10-10 16:43:13 -0400147
Al Viro94b5d262015-02-22 19:38:03 -0500148 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
Jeff Layton91a27b22012-10-10 15:25:28 -0400149 if (unlikely(len < 0)) {
Al Viro94b5d262015-02-22 19:38:03 -0500150 __putname(result);
151 return ERR_PTR(len);
Jeff Layton91a27b22012-10-10 15:25:28 -0400152 }
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700153
Jeff Layton7950e382012-10-10 16:43:13 -0400154 /*
155 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
156 * separate struct filename so we can dedicate the entire
157 * names_cache allocation for the pathname, and re-do the copy from
158 * userland.
159 */
Al Viro94b5d262015-02-22 19:38:03 -0500160 if (unlikely(len == EMBEDDED_NAME_MAX)) {
Al Virofd2f7cb2015-02-22 20:07:13 -0500161 const size_t size = offsetof(struct filename, iname[1]);
Jeff Layton7950e382012-10-10 16:43:13 -0400162 kname = (char *)result;
163
Al Virofd2f7cb2015-02-22 20:07:13 -0500164 /*
165 * size is chosen that way we to guarantee that
166 * result->iname[0] is within the same object and that
167 * kname can't be equal to result->iname, no matter what.
168 */
169 result = kzalloc(size, GFP_KERNEL);
Al Viro94b5d262015-02-22 19:38:03 -0500170 if (unlikely(!result)) {
171 __putname(kname);
172 return ERR_PTR(-ENOMEM);
Jeff Layton7950e382012-10-10 16:43:13 -0400173 }
174 result->name = kname;
Al Viro94b5d262015-02-22 19:38:03 -0500175 len = strncpy_from_user(kname, filename, PATH_MAX);
176 if (unlikely(len < 0)) {
177 __putname(kname);
178 kfree(result);
179 return ERR_PTR(len);
180 }
181 if (unlikely(len == PATH_MAX)) {
182 __putname(kname);
183 kfree(result);
184 return ERR_PTR(-ENAMETOOLONG);
185 }
Jeff Layton7950e382012-10-10 16:43:13 -0400186 }
187
Al Viro94b5d262015-02-22 19:38:03 -0500188 result->refcnt = 1;
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700189 /* The empty path is special. */
190 if (unlikely(!len)) {
191 if (empty)
Eric Paris4043cde2012-01-03 14:23:08 -0500192 *empty = 1;
Al Viro94b5d262015-02-22 19:38:03 -0500193 if (!(flags & LOOKUP_EMPTY)) {
194 putname(result);
195 return ERR_PTR(-ENOENT);
196 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 }
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700198
Jeff Layton7950e382012-10-10 16:43:13 -0400199 result->uptr = filename;
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800200 result->aname = NULL;
Jeff Layton7950e382012-10-10 16:43:13 -0400201 audit_getname(result);
202 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203}
204
Jeff Layton91a27b22012-10-10 15:25:28 -0400205struct filename *
206getname(const char __user * filename)
Al Virof52e0c12011-03-14 18:56:51 -0400207{
Linus Torvaldsf7493e52012-03-22 16:10:40 -0700208 return getname_flags(filename, 0, NULL);
Al Virof52e0c12011-03-14 18:56:51 -0400209}
210
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800211struct filename *
212getname_kernel(const char * filename)
213{
214 struct filename *result;
Paul Moore08518542015-01-21 23:59:56 -0500215 int len = strlen(filename) + 1;
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800216
217 result = __getname();
218 if (unlikely(!result))
219 return ERR_PTR(-ENOMEM);
220
Paul Moore08518542015-01-21 23:59:56 -0500221 if (len <= EMBEDDED_NAME_MAX) {
Al Virofd2f7cb2015-02-22 20:07:13 -0500222 result->name = (char *)result->iname;
Paul Moore08518542015-01-21 23:59:56 -0500223 } else if (len <= PATH_MAX) {
224 struct filename *tmp;
225
226 tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
227 if (unlikely(!tmp)) {
228 __putname(result);
229 return ERR_PTR(-ENOMEM);
230 }
231 tmp->name = (char *)result;
Paul Moore08518542015-01-21 23:59:56 -0500232 result = tmp;
233 } else {
234 __putname(result);
235 return ERR_PTR(-ENAMETOOLONG);
236 }
237 memcpy((char *)result->name, filename, len);
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800238 result->uptr = NULL;
239 result->aname = NULL;
Paul Moore55422d02015-01-22 00:00:23 -0500240 result->refcnt = 1;
Paul Moorefd3522f2015-01-22 00:00:10 -0500241 audit_getname(result);
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800242
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800243 return result;
244}
245
Jeff Layton91a27b22012-10-10 15:25:28 -0400246void putname(struct filename *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247{
Paul Moore55422d02015-01-22 00:00:23 -0500248 BUG_ON(name->refcnt <= 0);
249
250 if (--name->refcnt > 0)
251 return;
252
Al Virofd2f7cb2015-02-22 20:07:13 -0500253 if (name->name != name->iname) {
Paul Moore55422d02015-01-22 00:00:23 -0500254 __putname(name->name);
255 kfree(name);
256 } else
257 __putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259
Linus Torvaldse77819e2011-07-22 19:30:19 -0700260static int check_acl(struct inode *inode, int mask)
261{
Linus Torvalds84635d62011-07-25 22:47:03 -0700262#ifdef CONFIG_FS_POSIX_ACL
Linus Torvaldse77819e2011-07-22 19:30:19 -0700263 struct posix_acl *acl;
264
Linus Torvaldse77819e2011-07-22 19:30:19 -0700265 if (mask & MAY_NOT_BLOCK) {
Al Viro35678662011-08-02 21:32:13 -0400266 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
267 if (!acl)
Linus Torvaldse77819e2011-07-22 19:30:19 -0700268 return -EAGAIN;
Al Viro35678662011-08-02 21:32:13 -0400269 /* no ->get_acl() calls in RCU mode... */
Andreas Gruenbacherb8a7a3a2016-03-24 14:38:37 +0100270 if (is_uncached_acl(acl))
Al Viro35678662011-08-02 21:32:13 -0400271 return -ECHILD;
Ari Savolainen206b1d02011-08-06 19:43:07 +0300272 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
Linus Torvaldse77819e2011-07-22 19:30:19 -0700273 }
274
Christoph Hellwig2982baa2013-12-20 05:16:38 -0800275 acl = get_acl(inode, ACL_TYPE_ACCESS);
276 if (IS_ERR(acl))
277 return PTR_ERR(acl);
Linus Torvaldse77819e2011-07-22 19:30:19 -0700278 if (acl) {
279 int error = posix_acl_permission(inode, acl, mask);
280 posix_acl_release(acl);
281 return error;
282 }
Linus Torvalds84635d62011-07-25 22:47:03 -0700283#endif
Linus Torvaldse77819e2011-07-22 19:30:19 -0700284
285 return -EAGAIN;
286}
287
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700288/*
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530289 * This does the basic permission checking
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700290 */
Al Viro7e401452011-06-20 19:12:17 -0400291static int acl_permission_check(struct inode *inode, int mask)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700292{
Linus Torvalds26cf46b2011-05-13 11:51:01 -0700293 unsigned int mode = inode->i_mode;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700294
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -0800295 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700296 mode >>= 6;
297 else {
Linus Torvaldse77819e2011-07-22 19:30:19 -0700298 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
Al Viro7e401452011-06-20 19:12:17 -0400299 int error = check_acl(inode, mask);
Nick Pigginb74c79e2011-01-07 17:49:58 +1100300 if (error != -EAGAIN)
301 return error;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700302 }
303
304 if (in_group_p(inode->i_gid))
305 mode >>= 3;
306 }
307
308 /*
309 * If the DACs are ok we don't need any capability check.
310 */
Al Viro9c2c7032011-06-20 19:06:22 -0400311 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700312 return 0;
313 return -EACCES;
314}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315
316/**
Nick Pigginb74c79e2011-01-07 17:49:58 +1100317 * generic_permission - check for access rights on a Posix-like filesystem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 * @inode: inode to check access rights for
Andreas Gruenbacher8fd90c82011-10-23 23:13:30 +0530319 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 *
321 * Used to check for read/write/execute permissions on a file.
322 * We use "fsuid" for this, letting us set arbitrary permissions
323 * for filesystem access without changing the "normal" uids which
Nick Pigginb74c79e2011-01-07 17:49:58 +1100324 * are used for other things.
325 *
326 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
327 * request cannot be satisfied (eg. requires blocking or too much complexity).
328 * It would then be called again in ref-walk mode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 */
Al Viro2830ba72011-06-20 19:16:29 -0400330int generic_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700332 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333
334 /*
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530335 * Do the basic permission checks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 */
Al Viro7e401452011-06-20 19:12:17 -0400337 ret = acl_permission_check(inode, mask);
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700338 if (ret != -EACCES)
339 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340
Al Virod594e7e2011-06-20 19:55:42 -0400341 if (S_ISDIR(inode->i_mode)) {
342 /* DACs are overridable for directories */
Al Virod594e7e2011-06-20 19:55:42 -0400343 if (!(mask & MAY_WRITE))
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700344 if (capable_wrt_inode_uidgid(inode,
345 CAP_DAC_READ_SEARCH))
Al Virod594e7e2011-06-20 19:55:42 -0400346 return 0;
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700347 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 return 0;
Stephen Smalley2a4c2242017-03-10 12:14:18 -0500349 return -EACCES;
350 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351
352 /*
353 * Searching includes executable on directories, else just read.
354 */
Serge E. Hallyn7ea66002009-12-29 14:50:19 -0600355 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
Al Virod594e7e2011-06-20 19:55:42 -0400356 if (mask == MAY_READ)
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700357 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 return 0;
Stephen Smalley2a4c2242017-03-10 12:14:18 -0500359 /*
360 * Read/write DACs are always overridable.
361 * Executable DACs are overridable when there is
362 * at least one exec bit set.
363 */
364 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
365 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
366 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367
368 return -EACCES;
369}
Al Viro4d359502014-03-14 12:20:17 -0400370EXPORT_SYMBOL(generic_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700372/*
373 * We _really_ want to just do "generic_permission()" without
374 * even looking at the inode->i_op values. So we keep a cache
375 * flag in inode->i_opflags, that says "this has not special
376 * permission function, use the fast case".
377 */
378static inline int do_inode_permission(struct inode *inode, int mask)
379{
380 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
381 if (likely(inode->i_op->permission))
382 return inode->i_op->permission(inode, mask);
383
384 /* This gets set once for the inode lifetime */
385 spin_lock(&inode->i_lock);
386 inode->i_opflags |= IOP_FASTPERM;
387 spin_unlock(&inode->i_lock);
388 }
389 return generic_permission(inode, mask);
390}
391
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200392/**
David Howells0bdaea92012-06-25 12:55:46 +0100393 * __inode_permission - Check for access rights to a given inode
394 * @inode: Inode to check permission on
395 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200396 *
David Howells0bdaea92012-06-25 12:55:46 +0100397 * Check for read/write/execute permissions on an inode.
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530398 *
399 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
David Howells0bdaea92012-06-25 12:55:46 +0100400 *
401 * This does not check for a read-only file system. You probably want
402 * inode_permission().
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200403 */
David Howells0bdaea92012-06-25 12:55:46 +0100404int __inode_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405{
Al Viroe6305c42008-07-15 21:03:57 -0400406 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700408 if (unlikely(mask & MAY_WRITE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 /*
410 * Nobody gets write access to an immutable file.
411 */
412 if (IS_IMMUTABLE(inode))
Eryu Guan337684a2016-08-02 19:58:28 +0800413 return -EPERM;
Eric W. Biederman0bd23d092016-06-29 14:54:46 -0500414
415 /*
416 * Updating mtime will likely cause i_uid and i_gid to be
417 * written back improperly if their true value is unknown
418 * to the vfs.
419 */
420 if (HAS_UNMAPPED_ID(inode))
421 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 }
423
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700424 retval = do_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 if (retval)
426 return retval;
427
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -0700428 retval = devcgroup_inode_permission(inode, mask);
429 if (retval)
430 return retval;
431
Eric Parisd09ca732010-07-23 11:43:57 -0400432 return security_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433}
Miklos Szeredibd5d0852014-10-24 00:14:35 +0200434EXPORT_SYMBOL(__inode_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435
Al Virof4d6ff82011-06-19 13:14:21 -0400436/**
David Howells0bdaea92012-06-25 12:55:46 +0100437 * sb_permission - Check superblock-level permissions
438 * @sb: Superblock of inode to check permission on
Randy Dunlap55852632012-08-18 17:39:25 -0700439 * @inode: Inode to check permission on
David Howells0bdaea92012-06-25 12:55:46 +0100440 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
441 *
442 * Separate out file-system wide checks from inode-specific permission checks.
443 */
444static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
445{
446 if (unlikely(mask & MAY_WRITE)) {
447 umode_t mode = inode->i_mode;
448
449 /* Nobody gets write access to a read-only fs. */
450 if ((sb->s_flags & MS_RDONLY) &&
451 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
452 return -EROFS;
453 }
454 return 0;
455}
456
457/**
458 * inode_permission - Check for access rights to a given inode
459 * @inode: Inode to check permission on
460 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
461 *
462 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
463 * this, letting us set arbitrary permissions for filesystem access without
464 * changing the "normal" UIDs which are used for other things.
465 *
466 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
467 */
468int inode_permission(struct inode *inode, int mask)
469{
470 int retval;
471
472 retval = sb_permission(inode->i_sb, inode, mask);
473 if (retval)
474 return retval;
475 return __inode_permission(inode, mask);
476}
Al Viro4d359502014-03-14 12:20:17 -0400477EXPORT_SYMBOL(inode_permission);
David Howells0bdaea92012-06-25 12:55:46 +0100478
479/**
Jan Blunck5dd784d02008-02-14 19:34:38 -0800480 * path_get - get a reference to a path
481 * @path: path to get the reference to
482 *
483 * Given a path increment the reference count to the dentry and the vfsmount.
484 */
Al Virodcf787f2013-03-01 23:51:07 -0500485void path_get(const struct path *path)
Jan Blunck5dd784d02008-02-14 19:34:38 -0800486{
487 mntget(path->mnt);
488 dget(path->dentry);
489}
490EXPORT_SYMBOL(path_get);
491
492/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800493 * path_put - put a reference to a path
494 * @path: path to put the reference to
495 *
496 * Given a path decrement the reference count to the dentry and the vfsmount.
497 */
Al Virodcf787f2013-03-01 23:51:07 -0500498void path_put(const struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499{
Jan Blunck1d957f92008-02-14 19:34:35 -0800500 dput(path->dentry);
501 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502}
Jan Blunck1d957f92008-02-14 19:34:35 -0800503EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504
Al Viro894bc8c2015-05-02 07:16:16 -0400505#define EMBEDDED_LEVELS 2
Al Viro1f55a6e2014-11-01 19:30:41 -0400506struct nameidata {
507 struct path path;
Al Viro1cf26652015-05-06 16:01:56 -0400508 struct qstr last;
Al Viro1f55a6e2014-11-01 19:30:41 -0400509 struct path root;
510 struct inode *inode; /* path.dentry.d_inode */
511 unsigned int flags;
Al Viro9883d182015-05-13 07:28:08 -0400512 unsigned seq, m_seq;
Al Viro1f55a6e2014-11-01 19:30:41 -0400513 int last_type;
514 unsigned depth;
NeilBrown756daf22015-03-23 13:37:38 +1100515 int total_link_count;
Al Viro697fc6c2015-05-02 19:38:35 -0400516 struct saved {
517 struct path link;
Al Virofceef392015-12-29 15:58:39 -0500518 struct delayed_call done;
Al Viro697fc6c2015-05-02 19:38:35 -0400519 const char *name;
Al Viro0450b2d2015-05-08 13:23:53 -0400520 unsigned seq;
Al Viro894bc8c2015-05-02 07:16:16 -0400521 } *stack, internal[EMBEDDED_LEVELS];
Al Viro9883d182015-05-13 07:28:08 -0400522 struct filename *name;
523 struct nameidata *saved;
Al Virofceef392015-12-29 15:58:39 -0500524 struct inode *link_inode;
Al Viro9883d182015-05-13 07:28:08 -0400525 unsigned root_seq;
526 int dfd;
Al Viro1f55a6e2014-11-01 19:30:41 -0400527};
528
Al Viro9883d182015-05-13 07:28:08 -0400529static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
Al Viro894bc8c2015-05-02 07:16:16 -0400530{
NeilBrown756daf22015-03-23 13:37:38 +1100531 struct nameidata *old = current->nameidata;
532 p->stack = p->internal;
Al Viroc8a53ee2015-05-12 18:43:07 -0400533 p->dfd = dfd;
534 p->name = name;
NeilBrown756daf22015-03-23 13:37:38 +1100535 p->total_link_count = old ? old->total_link_count : 0;
Al Viro9883d182015-05-13 07:28:08 -0400536 p->saved = old;
NeilBrown756daf22015-03-23 13:37:38 +1100537 current->nameidata = p;
Al Viro894bc8c2015-05-02 07:16:16 -0400538}
539
Al Viro9883d182015-05-13 07:28:08 -0400540static void restore_nameidata(void)
Al Viro894bc8c2015-05-02 07:16:16 -0400541{
Al Viro9883d182015-05-13 07:28:08 -0400542 struct nameidata *now = current->nameidata, *old = now->saved;
NeilBrown756daf22015-03-23 13:37:38 +1100543
544 current->nameidata = old;
545 if (old)
546 old->total_link_count = now->total_link_count;
Al Viroe1a63bb2015-12-05 21:06:33 -0500547 if (now->stack != now->internal)
NeilBrown756daf22015-03-23 13:37:38 +1100548 kfree(now->stack);
Al Viro894bc8c2015-05-02 07:16:16 -0400549}
550
551static int __nd_alloc_stack(struct nameidata *nd)
552{
Al Virobc40aee2015-05-09 13:04:24 -0400553 struct saved *p;
554
555 if (nd->flags & LOOKUP_RCU) {
556 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
557 GFP_ATOMIC);
558 if (unlikely(!p))
559 return -ECHILD;
560 } else {
561 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
Al Viro894bc8c2015-05-02 07:16:16 -0400562 GFP_KERNEL);
Al Virobc40aee2015-05-09 13:04:24 -0400563 if (unlikely(!p))
564 return -ENOMEM;
565 }
Al Viro894bc8c2015-05-02 07:16:16 -0400566 memcpy(p, nd->internal, sizeof(nd->internal));
567 nd->stack = p;
568 return 0;
569}
570
Eric W. Biederman397d4252015-08-15 20:27:13 -0500571/**
572 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
573 * @path: nameidate to verify
574 *
575 * Rename can sometimes move a file or directory outside of a bind
576 * mount, path_connected allows those cases to be detected.
577 */
578static bool path_connected(const struct path *path)
579{
580 struct vfsmount *mnt = path->mnt;
581
582 /* Only bind mounts can have disconnected paths */
583 if (mnt->mnt_root == mnt->mnt_sb->s_root)
584 return true;
585
586 return is_subdir(path->dentry, mnt->mnt_root);
587}
588
Al Viro894bc8c2015-05-02 07:16:16 -0400589static inline int nd_alloc_stack(struct nameidata *nd)
590{
Al Viroda4e0be2015-05-03 20:52:15 -0400591 if (likely(nd->depth != EMBEDDED_LEVELS))
Al Viro894bc8c2015-05-02 07:16:16 -0400592 return 0;
593 if (likely(nd->stack != nd->internal))
594 return 0;
595 return __nd_alloc_stack(nd);
596}
597
Al Viro79733872015-05-09 12:55:43 -0400598static void drop_links(struct nameidata *nd)
599{
600 int i = nd->depth;
601 while (i--) {
602 struct saved *last = nd->stack + i;
Al Virofceef392015-12-29 15:58:39 -0500603 do_delayed_call(&last->done);
604 clear_delayed_call(&last->done);
Al Viro79733872015-05-09 12:55:43 -0400605 }
606}
607
608static void terminate_walk(struct nameidata *nd)
609{
610 drop_links(nd);
611 if (!(nd->flags & LOOKUP_RCU)) {
612 int i;
613 path_put(&nd->path);
614 for (i = 0; i < nd->depth; i++)
615 path_put(&nd->stack[i].link);
Al Viro102b8af2015-05-12 17:35:52 -0400616 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
617 path_put(&nd->root);
618 nd->root.mnt = NULL;
619 }
Al Viro79733872015-05-09 12:55:43 -0400620 } else {
621 nd->flags &= ~LOOKUP_RCU;
622 if (!(nd->flags & LOOKUP_ROOT))
623 nd->root.mnt = NULL;
624 rcu_read_unlock();
625 }
626 nd->depth = 0;
627}
628
629/* path_put is needed afterwards regardless of success or failure */
630static bool legitimize_path(struct nameidata *nd,
631 struct path *path, unsigned seq)
632{
633 int res = __legitimize_mnt(path->mnt, nd->m_seq);
634 if (unlikely(res)) {
635 if (res > 0)
636 path->mnt = NULL;
637 path->dentry = NULL;
638 return false;
639 }
640 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
641 path->dentry = NULL;
642 return false;
643 }
644 return !read_seqcount_retry(&path->dentry->d_seq, seq);
645}
646
647static bool legitimize_links(struct nameidata *nd)
648{
649 int i;
650 for (i = 0; i < nd->depth; i++) {
651 struct saved *last = nd->stack + i;
652 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
653 drop_links(nd);
654 nd->depth = i + 1;
655 return false;
656 }
657 }
658 return true;
659}
660
Al Viro19660af2011-03-25 10:32:48 -0400661/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100662 * Path walking has 2 modes, rcu-walk and ref-walk (see
Al Viro19660af2011-03-25 10:32:48 -0400663 * Documentation/filesystems/path-lookup.txt). In situations when we can't
664 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
Mike Marshall57e37152015-11-30 11:11:59 -0500665 * normal reference counts on dentries and vfsmounts to transition to ref-walk
Al Viro19660af2011-03-25 10:32:48 -0400666 * mode. Refcounts are grabbed at the last known good point before rcu-walk
667 * got stuck, so ref-walk may continue from there. If this is not successful
668 * (eg. a seqcount has changed), then failure is returned and it's up to caller
669 * to restart the path walk from the beginning in ref-walk mode.
Nick Piggin31e6b012011-01-07 17:49:52 +1100670 */
Nick Piggin31e6b012011-01-07 17:49:52 +1100671
672/**
Al Viro19660af2011-03-25 10:32:48 -0400673 * unlazy_walk - try to switch to ref-walk mode.
674 * @nd: nameidata pathwalk data
Randy Dunlap39191622011-01-08 19:36:21 -0800675 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100676 *
Al Viro4675ac32017-01-09 22:29:15 -0500677 * unlazy_walk attempts to legitimize the current nd->path and nd->root
678 * for ref-walk mode.
679 * Must be called from rcu-walk context.
Al Viro79733872015-05-09 12:55:43 -0400680 * Nothing should touch nameidata between unlazy_walk() failure and
681 * terminate_walk().
Nick Piggin31e6b012011-01-07 17:49:52 +1100682 */
Al Viro4675ac32017-01-09 22:29:15 -0500683static int unlazy_walk(struct nameidata *nd)
Nick Piggin31e6b012011-01-07 17:49:52 +1100684{
Nick Piggin31e6b012011-01-07 17:49:52 +1100685 struct dentry *parent = nd->path.dentry;
686
687 BUG_ON(!(nd->flags & LOOKUP_RCU));
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700688
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700689 nd->flags &= ~LOOKUP_RCU;
Al Viro79733872015-05-09 12:55:43 -0400690 if (unlikely(!legitimize_links(nd)))
691 goto out2;
Al Viro4675ac32017-01-09 22:29:15 -0500692 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
693 goto out1;
694 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
695 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq)))
696 goto out;
697 }
698 rcu_read_unlock();
699 BUG_ON(nd->inode != parent->d_inode);
700 return 0;
701
702out2:
703 nd->path.mnt = NULL;
704 nd->path.dentry = NULL;
705out1:
706 if (!(nd->flags & LOOKUP_ROOT))
707 nd->root.mnt = NULL;
708out:
709 rcu_read_unlock();
710 return -ECHILD;
711}
712
713/**
714 * unlazy_child - try to switch to ref-walk mode.
715 * @nd: nameidata pathwalk data
716 * @dentry: child of nd->path.dentry
717 * @seq: seq number to check dentry against
718 * Returns: 0 on success, -ECHILD on failure
719 *
720 * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry
721 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
722 * @nd. Must be called from rcu-walk context.
723 * Nothing should touch nameidata between unlazy_child() failure and
724 * terminate_walk().
725 */
726static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq)
727{
728 BUG_ON(!(nd->flags & LOOKUP_RCU));
729
730 nd->flags &= ~LOOKUP_RCU;
731 if (unlikely(!legitimize_links(nd)))
732 goto out2;
Al Viro79733872015-05-09 12:55:43 -0400733 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
734 goto out2;
Al Viro4675ac32017-01-09 22:29:15 -0500735 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
Al Viro79733872015-05-09 12:55:43 -0400736 goto out1;
Al Viro48a066e2013-09-29 22:06:07 -0400737
Linus Torvalds15570082013-09-02 11:38:06 -0700738 /*
Al Viro4675ac32017-01-09 22:29:15 -0500739 * We need to move both the parent and the dentry from the RCU domain
740 * to be properly refcounted. And the sequence number in the dentry
741 * validates *both* dentry counters, since we checked the sequence
742 * number of the parent after we got the child sequence number. So we
743 * know the parent must still be valid if the child sequence number is
Linus Torvalds15570082013-09-02 11:38:06 -0700744 */
Al Viro4675ac32017-01-09 22:29:15 -0500745 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
746 goto out;
747 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) {
748 rcu_read_unlock();
749 dput(dentry);
750 goto drop_root_mnt;
Al Viro19660af2011-03-25 10:32:48 -0400751 }
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700752 /*
753 * Sequence counts matched. Now make sure that the root is
754 * still valid and get it if required.
755 */
756 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
Al Viro5a8d87e2015-05-12 00:10:18 -0400757 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
758 rcu_read_unlock();
759 dput(dentry);
760 return -ECHILD;
Al Viro79733872015-05-09 12:55:43 -0400761 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100762 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100763
Al Viro8b61e742013-11-08 12:45:01 -0500764 rcu_read_unlock();
Nick Piggin31e6b012011-01-07 17:49:52 +1100765 return 0;
Al Viro19660af2011-03-25 10:32:48 -0400766
Al Viro79733872015-05-09 12:55:43 -0400767out2:
768 nd->path.mnt = NULL;
769out1:
770 nd->path.dentry = NULL;
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700771out:
Al Viro8b61e742013-11-08 12:45:01 -0500772 rcu_read_unlock();
Linus Torvaldsd0d27272013-09-10 12:17:49 -0700773drop_root_mnt:
774 if (!(nd->flags & LOOKUP_ROOT))
775 nd->root.mnt = NULL;
Nick Piggin31e6b012011-01-07 17:49:52 +1100776 return -ECHILD;
777}
778
Al Viro4ce16ef32012-06-10 16:10:59 -0400779static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
Nick Piggin34286d62011-01-07 17:49:57 +1100780{
Al Viroa89f8332017-01-09 22:25:28 -0500781 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
782 return dentry->d_op->d_revalidate(dentry, flags);
783 else
784 return 1;
Nick Piggin34286d62011-01-07 17:49:57 +1100785}
786
Al Viro9f1fafe2011-03-25 11:00:12 -0400787/**
788 * complete_walk - successful completion of path walk
789 * @nd: pointer nameidata
Jeff Layton39159de2009-12-07 12:01:50 -0500790 *
Al Viro9f1fafe2011-03-25 11:00:12 -0400791 * If we had been in RCU mode, drop out of it and legitimize nd->path.
792 * Revalidate the final result, unless we'd already done that during
793 * the path walk or the filesystem doesn't ask for it. Return 0 on
794 * success, -error on failure. In case of failure caller does not
795 * need to drop nd->path.
Jeff Layton39159de2009-12-07 12:01:50 -0500796 */
Al Viro9f1fafe2011-03-25 11:00:12 -0400797static int complete_walk(struct nameidata *nd)
Jeff Layton39159de2009-12-07 12:01:50 -0500798{
Al Viro16c2cd72011-02-22 15:50:10 -0500799 struct dentry *dentry = nd->path.dentry;
Jeff Layton39159de2009-12-07 12:01:50 -0500800 int status;
Jeff Layton39159de2009-12-07 12:01:50 -0500801
Al Viro9f1fafe2011-03-25 11:00:12 -0400802 if (nd->flags & LOOKUP_RCU) {
Al Viro9f1fafe2011-03-25 11:00:12 -0400803 if (!(nd->flags & LOOKUP_ROOT))
804 nd->root.mnt = NULL;
Al Viro4675ac32017-01-09 22:29:15 -0500805 if (unlikely(unlazy_walk(nd)))
Al Viro48a066e2013-09-29 22:06:07 -0400806 return -ECHILD;
Al Viro9f1fafe2011-03-25 11:00:12 -0400807 }
808
Al Viro16c2cd72011-02-22 15:50:10 -0500809 if (likely(!(nd->flags & LOOKUP_JUMPED)))
Jeff Layton39159de2009-12-07 12:01:50 -0500810 return 0;
811
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500812 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
Al Viro16c2cd72011-02-22 15:50:10 -0500813 return 0;
814
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500815 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
Jeff Layton39159de2009-12-07 12:01:50 -0500816 if (status > 0)
817 return 0;
818
Al Viro16c2cd72011-02-22 15:50:10 -0500819 if (!status)
Jeff Layton39159de2009-12-07 12:01:50 -0500820 status = -ESTALE;
Al Viro16c2cd72011-02-22 15:50:10 -0500821
Jeff Layton39159de2009-12-07 12:01:50 -0500822 return status;
823}
824
Al Viro18d8c862015-05-12 16:32:34 -0400825static void set_root(struct nameidata *nd)
Al Viro2a737872009-04-07 11:49:53 -0400826{
Al Viro7bd88372014-09-13 21:55:46 -0400827 struct fs_struct *fs = current->fs;
Nick Pigginc28cc362011-01-07 17:49:53 +1100828
Al Viro9e6697e2015-12-05 20:07:21 -0500829 if (nd->flags & LOOKUP_RCU) {
830 unsigned seq;
831
832 do {
833 seq = read_seqcount_begin(&fs->seq);
834 nd->root = fs->root;
835 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
836 } while (read_seqcount_retry(&fs->seq, seq));
837 } else {
838 get_fs_root(fs, &nd->root);
839 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100840}
841
Jan Blunck1d957f92008-02-14 19:34:35 -0800842static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700843{
844 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800845 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700846 mntput(path->mnt);
847}
848
Nick Piggin7b9337a2011-01-14 08:42:43 +0000849static inline void path_to_nameidata(const struct path *path,
850 struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700851{
Nick Piggin31e6b012011-01-07 17:49:52 +1100852 if (!(nd->flags & LOOKUP_RCU)) {
853 dput(nd->path.dentry);
854 if (nd->path.mnt != path->mnt)
855 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800856 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100857 nd->path.mnt = path->mnt;
Jan Blunck4ac91372008-02-14 19:34:32 -0800858 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700859}
860
Al Viro248fb5b2015-12-05 20:51:58 -0500861static int nd_jump_root(struct nameidata *nd)
862{
863 if (nd->flags & LOOKUP_RCU) {
864 struct dentry *d;
865 nd->path = nd->root;
866 d = nd->path.dentry;
867 nd->inode = d->d_inode;
868 nd->seq = nd->root_seq;
869 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
870 return -ECHILD;
871 } else {
872 path_put(&nd->path);
873 nd->path = nd->root;
874 path_get(&nd->path);
875 nd->inode = nd->path.dentry->d_inode;
876 }
877 nd->flags |= LOOKUP_JUMPED;
878 return 0;
879}
880
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400881/*
Al Viro6b255392015-11-17 10:20:54 -0500882 * Helper to directly jump to a known parsed path from ->get_link,
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400883 * caller must have taken a reference to path beforehand.
884 */
Al Viro6e77137b2015-05-02 13:37:52 -0400885void nd_jump_link(struct path *path)
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400886{
Al Viro6e77137b2015-05-02 13:37:52 -0400887 struct nameidata *nd = current->nameidata;
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400888 path_put(&nd->path);
889
890 nd->path = *path;
891 nd->inode = nd->path.dentry->d_inode;
892 nd->flags |= LOOKUP_JUMPED;
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400893}
894
Al Virob9ff4422015-05-02 20:19:23 -0400895static inline void put_link(struct nameidata *nd)
Al Viro574197e2011-03-14 22:20:34 -0400896{
Al Viro21c30032015-05-03 21:06:24 -0400897 struct saved *last = nd->stack + --nd->depth;
Al Virofceef392015-12-29 15:58:39 -0500898 do_delayed_call(&last->done);
Al Viro6548fae2015-05-07 20:32:22 -0400899 if (!(nd->flags & LOOKUP_RCU))
900 path_put(&last->link);
Al Viro574197e2011-03-14 22:20:34 -0400901}
902
Linus Torvalds561ec642012-10-26 10:05:07 -0700903int sysctl_protected_symlinks __read_mostly = 0;
904int sysctl_protected_hardlinks __read_mostly = 0;
Kees Cook800179c2012-07-25 17:29:07 -0700905
906/**
907 * may_follow_link - Check symlink following for unsafe situations
Randy Dunlap55852632012-08-18 17:39:25 -0700908 * @nd: nameidata pathwalk data
Kees Cook800179c2012-07-25 17:29:07 -0700909 *
910 * In the case of the sysctl_protected_symlinks sysctl being enabled,
911 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
912 * in a sticky world-writable directory. This is to protect privileged
913 * processes from failing races against path names that may change out
914 * from under them by way of other users creating malicious symlinks.
915 * It will permit symlinks to be followed only when outside a sticky
916 * world-writable directory, or when the uid of the symlink and follower
917 * match, or when the directory owner matches the symlink's owner.
918 *
919 * Returns 0 if following the symlink is allowed, -ve on error.
920 */
Al Virofec2fa22015-05-06 15:58:18 -0400921static inline int may_follow_link(struct nameidata *nd)
Kees Cook800179c2012-07-25 17:29:07 -0700922{
923 const struct inode *inode;
924 const struct inode *parent;
Seth Forshee2d7f9e22016-04-26 14:36:23 -0500925 kuid_t puid;
Kees Cook800179c2012-07-25 17:29:07 -0700926
927 if (!sysctl_protected_symlinks)
928 return 0;
929
930 /* Allowed if owner and follower match. */
Al Virofceef392015-12-29 15:58:39 -0500931 inode = nd->link_inode;
Eric W. Biederman81abe272012-08-03 09:38:08 -0700932 if (uid_eq(current_cred()->fsuid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700933 return 0;
934
935 /* Allowed if parent directory not sticky and world-writable. */
Al Viroaa65fa32015-08-04 23:23:50 -0400936 parent = nd->inode;
Kees Cook800179c2012-07-25 17:29:07 -0700937 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
938 return 0;
939
940 /* Allowed if parent directory and link owner match. */
Seth Forshee2d7f9e22016-04-26 14:36:23 -0500941 puid = parent->i_uid;
942 if (uid_valid(puid) && uid_eq(puid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700943 return 0;
944
Al Viro319565022015-05-07 20:37:40 -0400945 if (nd->flags & LOOKUP_RCU)
946 return -ECHILD;
947
Al Viro1cf26652015-05-06 16:01:56 -0400948 audit_log_link_denied("follow_link", &nd->stack[0].link);
Kees Cook800179c2012-07-25 17:29:07 -0700949 return -EACCES;
950}
951
952/**
953 * safe_hardlink_source - Check for safe hardlink conditions
954 * @inode: the source inode to hardlink from
955 *
956 * Return false if at least one of the following conditions:
957 * - inode is not a regular file
958 * - inode is setuid
959 * - inode is setgid and group-exec
960 * - access failure for read and write
961 *
962 * Otherwise returns true.
963 */
964static bool safe_hardlink_source(struct inode *inode)
965{
966 umode_t mode = inode->i_mode;
967
968 /* Special files should not get pinned to the filesystem. */
969 if (!S_ISREG(mode))
970 return false;
971
972 /* Setuid files should not get pinned to the filesystem. */
973 if (mode & S_ISUID)
974 return false;
975
976 /* Executable setgid files should not get pinned to the filesystem. */
977 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
978 return false;
979
980 /* Hardlinking to unreadable or unwritable sources is dangerous. */
981 if (inode_permission(inode, MAY_READ | MAY_WRITE))
982 return false;
983
984 return true;
985}
986
987/**
988 * may_linkat - Check permissions for creating a hardlink
989 * @link: the source to hardlink from
990 *
991 * Block hardlink when all of:
992 * - sysctl_protected_hardlinks enabled
993 * - fsuid does not match inode
994 * - hardlink source is unsafe (see safe_hardlink_source() above)
Dirk Steinmetzf2ca3792015-10-20 16:09:19 +0200995 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
Kees Cook800179c2012-07-25 17:29:07 -0700996 *
997 * Returns 0 if successful, -ve on error.
998 */
999static int may_linkat(struct path *link)
1000{
Kees Cook800179c2012-07-25 17:29:07 -07001001 struct inode *inode;
1002
1003 if (!sysctl_protected_hardlinks)
1004 return 0;
1005
Kees Cook800179c2012-07-25 17:29:07 -07001006 inode = link->dentry->d_inode;
1007
1008 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1009 * otherwise, it must be a safe source.
1010 */
Dirk Steinmetzf2ca3792015-10-20 16:09:19 +02001011 if (inode_owner_or_capable(inode) || safe_hardlink_source(inode))
Kees Cook800179c2012-07-25 17:29:07 -07001012 return 0;
1013
Kees Cooka51d9ea2012-07-25 17:29:08 -07001014 audit_log_link_denied("linkat", link);
Kees Cook800179c2012-07-25 17:29:07 -07001015 return -EPERM;
1016}
1017
Al Viro3b2e7f72015-04-19 00:53:50 -04001018static __always_inline
1019const char *get_link(struct nameidata *nd)
Ian Kent051d3812006-03-27 01:14:53 -08001020{
Al Viroab104922015-05-10 11:50:01 -04001021 struct saved *last = nd->stack + nd->depth - 1;
Al Viro1cf26652015-05-06 16:01:56 -04001022 struct dentry *dentry = last->link.dentry;
Al Virofceef392015-12-29 15:58:39 -05001023 struct inode *inode = nd->link_inode;
Al Viro6d7b5aa2012-06-10 04:15:17 -04001024 int error;
Al Viro0a959df2015-04-18 18:23:41 -04001025 const char *res;
Ian Kent051d3812006-03-27 01:14:53 -08001026
NeilBrown8fa9dd22015-03-23 13:37:40 +11001027 if (!(nd->flags & LOOKUP_RCU)) {
1028 touch_atime(&last->link);
1029 cond_resched();
Miklos Szeredi598e3c82016-09-16 12:44:20 +02001030 } else if (atime_needs_update_rcu(&last->link, inode)) {
Al Viro4675ac32017-01-09 22:29:15 -05001031 if (unlikely(unlazy_walk(nd)))
NeilBrown8fa9dd22015-03-23 13:37:40 +11001032 return ERR_PTR(-ECHILD);
1033 touch_atime(&last->link);
1034 }
1035
NeilBrownbda0be72015-03-23 13:37:39 +11001036 error = security_inode_follow_link(dentry, inode,
1037 nd->flags & LOOKUP_RCU);
1038 if (unlikely(error))
Al Viro6920a442015-05-10 10:43:46 -04001039 return ERR_PTR(error);
Al Viro36f3b4f2011-02-22 21:24:38 -05001040
Al Viro86acdca12009-12-22 23:45:11 -05001041 nd->last_type = LAST_BIND;
Al Virod4dee482015-04-30 20:08:02 -04001042 res = inode->i_link;
1043 if (!res) {
Al Virofceef392015-12-29 15:58:39 -05001044 const char * (*get)(struct dentry *, struct inode *,
1045 struct delayed_call *);
1046 get = inode->i_op->get_link;
Al Viro8c1b4562015-05-09 18:15:21 -04001047 if (nd->flags & LOOKUP_RCU) {
Al Virofceef392015-12-29 15:58:39 -05001048 res = get(NULL, inode, &last->done);
Al Viro6b255392015-11-17 10:20:54 -05001049 if (res == ERR_PTR(-ECHILD)) {
Al Viro4675ac32017-01-09 22:29:15 -05001050 if (unlikely(unlazy_walk(nd)))
Al Viro6b255392015-11-17 10:20:54 -05001051 return ERR_PTR(-ECHILD);
Al Virofceef392015-12-29 15:58:39 -05001052 res = get(dentry, inode, &last->done);
Al Viro6b255392015-11-17 10:20:54 -05001053 }
1054 } else {
Al Virofceef392015-12-29 15:58:39 -05001055 res = get(dentry, inode, &last->done);
Al Viro8c1b4562015-05-09 18:15:21 -04001056 }
Al Virofceef392015-12-29 15:58:39 -05001057 if (IS_ERR_OR_NULL(res))
Al Virofab51e82015-05-10 11:01:00 -04001058 return res;
Ian Kent051d3812006-03-27 01:14:53 -08001059 }
Al Virofab51e82015-05-10 11:01:00 -04001060 if (*res == '/') {
Al Viro9e6697e2015-12-05 20:07:21 -05001061 if (!nd->root.mnt)
1062 set_root(nd);
Al Viro248fb5b2015-12-05 20:51:58 -05001063 if (unlikely(nd_jump_root(nd)))
1064 return ERR_PTR(-ECHILD);
Al Virofab51e82015-05-10 11:01:00 -04001065 while (unlikely(*++res == '/'))
1066 ;
1067 }
1068 if (!*res)
1069 res = NULL;
Al Viro0a959df2015-04-18 18:23:41 -04001070 return res;
1071}
Al Viro6d7b5aa2012-06-10 04:15:17 -04001072
David Howellsf015f1262012-06-25 12:55:28 +01001073/*
1074 * follow_up - Find the mountpoint of path's vfsmount
1075 *
1076 * Given a path, find the mountpoint of its source file system.
1077 * Replace @path with the path of the mountpoint in the parent mount.
1078 * Up is towards /.
1079 *
1080 * Return 1 if we went up a level and 0 if we were already at the
1081 * root.
1082 */
Al Virobab77eb2009-04-18 03:26:48 -04001083int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084{
Al Viro0714a532011-11-24 22:19:58 -05001085 struct mount *mnt = real_mount(path->mnt);
1086 struct mount *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +10001088
Al Viro48a066e2013-09-29 22:06:07 -04001089 read_seqlock_excl(&mount_lock);
Al Viro0714a532011-11-24 22:19:58 -05001090 parent = mnt->mnt_parent;
Al Viro3c0a6162012-07-18 17:32:50 +04001091 if (parent == mnt) {
Al Viro48a066e2013-09-29 22:06:07 -04001092 read_sequnlock_excl(&mount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 return 0;
1094 }
Al Viro0714a532011-11-24 22:19:58 -05001095 mntget(&parent->mnt);
Al Viroa73324d2011-11-24 22:25:07 -05001096 mountpoint = dget(mnt->mnt_mountpoint);
Al Viro48a066e2013-09-29 22:06:07 -04001097 read_sequnlock_excl(&mount_lock);
Al Virobab77eb2009-04-18 03:26:48 -04001098 dput(path->dentry);
1099 path->dentry = mountpoint;
1100 mntput(path->mnt);
Al Viro0714a532011-11-24 22:19:58 -05001101 path->mnt = &parent->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 return 1;
1103}
Al Viro4d359502014-03-14 12:20:17 -04001104EXPORT_SYMBOL(follow_up);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001106/*
David Howells9875cf82011-01-14 18:45:21 +00001107 * Perform an automount
1108 * - return -EISDIR to tell follow_managed() to stop and return the path we
1109 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 */
NeilBrown756daf22015-03-23 13:37:38 +11001111static int follow_automount(struct path *path, struct nameidata *nd,
David Howells9875cf82011-01-14 18:45:21 +00001112 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +11001113{
David Howells9875cf82011-01-14 18:45:21 +00001114 struct vfsmount *mnt;
David Howellsea5b7782011-01-14 19:10:03 +00001115 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001116
David Howells9875cf82011-01-14 18:45:21 +00001117 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1118 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -07001119
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +02001120 /* We don't want to mount if someone's just doing a stat -
1121 * unless they're stat'ing a directory and appended a '/' to
1122 * the name.
1123 *
1124 * We do, however, want to mount if someone wants to open or
1125 * create a file of any type under the mountpoint, wants to
1126 * traverse through the mountpoint or wants to open the
1127 * mounted directory. Also, autofs may mark negative dentries
1128 * as being automount points. These will need the attentions
1129 * of the daemon to instantiate them before they can be used.
David Howells9875cf82011-01-14 18:45:21 +00001130 */
NeilBrown756daf22015-03-23 13:37:38 +11001131 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1132 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +02001133 path->dentry->d_inode)
1134 return -EISDIR;
1135
Eric W. Biedermanaeaa4a72016-07-23 11:20:44 -05001136 if (path->dentry->d_sb->s_user_ns != &init_user_ns)
1137 return -EACCES;
1138
NeilBrown756daf22015-03-23 13:37:38 +11001139 nd->total_link_count++;
1140 if (nd->total_link_count >= 40)
David Howells9875cf82011-01-14 18:45:21 +00001141 return -ELOOP;
1142
1143 mnt = path->dentry->d_op->d_automount(path);
1144 if (IS_ERR(mnt)) {
1145 /*
1146 * The filesystem is allowed to return -EISDIR here to indicate
1147 * it doesn't want to automount. For instance, autofs would do
1148 * this so that its userspace daemon can mount on this dentry.
1149 *
1150 * However, we can only permit this if it's a terminal point in
1151 * the path being looked up; if it wasn't then the remainder of
1152 * the path is inaccessible and we should say so.
1153 */
NeilBrown756daf22015-03-23 13:37:38 +11001154 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
David Howells9875cf82011-01-14 18:45:21 +00001155 return -EREMOTE;
1156 return PTR_ERR(mnt);
1157 }
David Howellsea5b7782011-01-14 19:10:03 +00001158
David Howells9875cf82011-01-14 18:45:21 +00001159 if (!mnt) /* mount collision */
1160 return 0;
1161
Al Viro8aef1882011-06-16 15:10:06 +01001162 if (!*need_mntput) {
1163 /* lock_mount() may release path->mnt on error */
1164 mntget(path->mnt);
1165 *need_mntput = true;
1166 }
Al Viro19a167a2011-01-17 01:35:23 -05001167 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +00001168
David Howellsea5b7782011-01-14 19:10:03 +00001169 switch (err) {
1170 case -EBUSY:
1171 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -05001172 return 0;
David Howellsea5b7782011-01-14 19:10:03 +00001173 case 0:
Al Viro8aef1882011-06-16 15:10:06 +01001174 path_put(path);
David Howellsea5b7782011-01-14 19:10:03 +00001175 path->mnt = mnt;
1176 path->dentry = dget(mnt->mnt_root);
David Howellsea5b7782011-01-14 19:10:03 +00001177 return 0;
Al Viro19a167a2011-01-17 01:35:23 -05001178 default:
1179 return err;
David Howellsea5b7782011-01-14 19:10:03 +00001180 }
Al Viro19a167a2011-01-17 01:35:23 -05001181
David Howells9875cf82011-01-14 18:45:21 +00001182}
1183
1184/*
1185 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +00001186 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +00001187 * - Flagged as mountpoint
1188 * - Flagged as automount point
1189 *
1190 * This may only be called in refwalk mode.
1191 *
1192 * Serialization is taken care of in namespace.c
1193 */
NeilBrown756daf22015-03-23 13:37:38 +11001194static int follow_managed(struct path *path, struct nameidata *nd)
David Howells9875cf82011-01-14 18:45:21 +00001195{
Al Viro8aef1882011-06-16 15:10:06 +01001196 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
David Howells9875cf82011-01-14 18:45:21 +00001197 unsigned managed;
1198 bool need_mntput = false;
Al Viro8aef1882011-06-16 15:10:06 +01001199 int ret = 0;
David Howells9875cf82011-01-14 18:45:21 +00001200
1201 /* Given that we're not holding a lock here, we retain the value in a
1202 * local variable for each dentry as we look at it so that we don't see
1203 * the components of that value change under us */
1204 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1205 managed &= DCACHE_MANAGED_DENTRY,
1206 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +00001207 /* Allow the filesystem to manage the transit without i_mutex
1208 * being held. */
1209 if (managed & DCACHE_MANAGE_TRANSIT) {
1210 BUG_ON(!path->dentry->d_op);
1211 BUG_ON(!path->dentry->d_op->d_manage);
Ian Kentfb5f51c2016-11-24 08:03:41 +11001212 ret = path->dentry->d_op->d_manage(path, false);
David Howellscc53ce52011-01-14 18:45:26 +00001213 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001214 break;
David Howellscc53ce52011-01-14 18:45:26 +00001215 }
1216
David Howells9875cf82011-01-14 18:45:21 +00001217 /* Transit to a mounted filesystem. */
1218 if (managed & DCACHE_MOUNTED) {
1219 struct vfsmount *mounted = lookup_mnt(path);
1220 if (mounted) {
1221 dput(path->dentry);
1222 if (need_mntput)
1223 mntput(path->mnt);
1224 path->mnt = mounted;
1225 path->dentry = dget(mounted->mnt_root);
1226 need_mntput = true;
1227 continue;
1228 }
1229
1230 /* Something is mounted on this dentry in another
1231 * namespace and/or whatever was mounted there in this
Al Viro48a066e2013-09-29 22:06:07 -04001232 * namespace got unmounted before lookup_mnt() could
1233 * get it */
David Howells9875cf82011-01-14 18:45:21 +00001234 }
1235
1236 /* Handle an automount point */
1237 if (managed & DCACHE_NEED_AUTOMOUNT) {
NeilBrown756daf22015-03-23 13:37:38 +11001238 ret = follow_automount(path, nd, &need_mntput);
David Howells9875cf82011-01-14 18:45:21 +00001239 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001240 break;
David Howells9875cf82011-01-14 18:45:21 +00001241 continue;
1242 }
1243
1244 /* We didn't change the current path point */
1245 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 }
Al Viro8aef1882011-06-16 15:10:06 +01001247
1248 if (need_mntput && path->mnt == mnt)
1249 mntput(path->mnt);
Al Viroe9742b52016-03-05 22:04:59 -05001250 if (ret == -EISDIR || !ret)
1251 ret = 1;
Al Viro84027522015-04-22 10:30:08 -04001252 if (need_mntput)
1253 nd->flags |= LOOKUP_JUMPED;
1254 if (unlikely(ret < 0))
1255 path_put_conditional(path, nd);
1256 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257}
1258
David Howellscc53ce52011-01-14 18:45:26 +00001259int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260{
1261 struct vfsmount *mounted;
1262
Al Viro1c755af2009-04-18 14:06:57 -04001263 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -04001265 dput(path->dentry);
1266 mntput(path->mnt);
1267 path->mnt = mounted;
1268 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 return 1;
1270 }
1271 return 0;
1272}
Al Viro4d359502014-03-14 12:20:17 -04001273EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274
Ian Kentfb5f51c2016-11-24 08:03:41 +11001275static inline int managed_dentry_rcu(const struct path *path)
Ian Kent62a73752011-03-25 01:51:02 +08001276{
Ian Kentfb5f51c2016-11-24 08:03:41 +11001277 return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1278 path->dentry->d_op->d_manage(path, true) : 0;
Ian Kent62a73752011-03-25 01:51:02 +08001279}
1280
David Howells9875cf82011-01-14 18:45:21 +00001281/*
Al Viro287548e2011-05-27 06:50:06 -04001282 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1283 * we meet a managed dentry that would need blocking.
David Howells9875cf82011-01-14 18:45:21 +00001284 */
1285static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
Al Viro254cf582015-05-05 09:40:46 -04001286 struct inode **inode, unsigned *seqp)
David Howells9875cf82011-01-14 18:45:21 +00001287{
Ian Kent62a73752011-03-25 01:51:02 +08001288 for (;;) {
Al Viroc7105362011-11-24 18:22:03 -05001289 struct mount *mounted;
Ian Kent62a73752011-03-25 01:51:02 +08001290 /*
1291 * Don't forget we might have a non-mountpoint managed dentry
1292 * that wants to block transit.
1293 */
Ian Kentfb5f51c2016-11-24 08:03:41 +11001294 switch (managed_dentry_rcu(path)) {
NeilBrownb8faf032014-08-04 17:06:29 +10001295 case -ECHILD:
1296 default:
David Howellsab909112011-01-14 18:46:51 +00001297 return false;
NeilBrownb8faf032014-08-04 17:06:29 +10001298 case -EISDIR:
1299 return true;
1300 case 0:
1301 break;
1302 }
Ian Kent62a73752011-03-25 01:51:02 +08001303
1304 if (!d_mountpoint(path->dentry))
NeilBrownb8faf032014-08-04 17:06:29 +10001305 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
Ian Kent62a73752011-03-25 01:51:02 +08001306
Al Viro474279d2013-10-01 16:11:26 -04001307 mounted = __lookup_mnt(path->mnt, path->dentry);
David Howells9875cf82011-01-14 18:45:21 +00001308 if (!mounted)
1309 break;
Al Viroc7105362011-11-24 18:22:03 -05001310 path->mnt = &mounted->mnt;
1311 path->dentry = mounted->mnt.mnt_root;
Al Viroa3fbbde2011-11-07 21:21:26 +00001312 nd->flags |= LOOKUP_JUMPED;
Al Viro254cf582015-05-05 09:40:46 -04001313 *seqp = read_seqcount_begin(&path->dentry->d_seq);
Linus Torvalds59430262011-07-18 15:43:29 -07001314 /*
1315 * Update the inode too. We don't need to re-check the
1316 * dentry sequence number here after this d_inode read,
1317 * because a mount-point is always pinned.
1318 */
1319 *inode = path->dentry->d_inode;
David Howells9875cf82011-01-14 18:45:21 +00001320 }
Al Virof5be3e29122014-09-13 21:50:45 -04001321 return !read_seqretry(&mount_lock, nd->m_seq) &&
NeilBrownb8faf032014-08-04 17:06:29 +10001322 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
Al Viro287548e2011-05-27 06:50:06 -04001323}
1324
Nick Piggin31e6b012011-01-07 17:49:52 +11001325static int follow_dotdot_rcu(struct nameidata *nd)
1326{
Al Viro4023bfc2014-09-13 21:59:43 -04001327 struct inode *inode = nd->inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001328
David Howells9875cf82011-01-14 18:45:21 +00001329 while (1) {
Al Viroaed434a2015-05-12 12:22:47 -04001330 if (path_equal(&nd->path, &nd->root))
Nick Piggin31e6b012011-01-07 17:49:52 +11001331 break;
Nick Piggin31e6b012011-01-07 17:49:52 +11001332 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1333 struct dentry *old = nd->path.dentry;
1334 struct dentry *parent = old->d_parent;
1335 unsigned seq;
1336
Al Viro4023bfc2014-09-13 21:59:43 -04001337 inode = parent->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001338 seq = read_seqcount_begin(&parent->d_seq);
Al Viroaed434a2015-05-12 12:22:47 -04001339 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1340 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001341 nd->path.dentry = parent;
1342 nd->seq = seq;
Eric W. Biederman397d4252015-08-15 20:27:13 -05001343 if (unlikely(!path_connected(&nd->path)))
1344 return -ENOENT;
Nick Piggin31e6b012011-01-07 17:49:52 +11001345 break;
Al Viroaed434a2015-05-12 12:22:47 -04001346 } else {
1347 struct mount *mnt = real_mount(nd->path.mnt);
1348 struct mount *mparent = mnt->mnt_parent;
1349 struct dentry *mountpoint = mnt->mnt_mountpoint;
1350 struct inode *inode2 = mountpoint->d_inode;
1351 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1352 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1353 return -ECHILD;
1354 if (&mparent->mnt == nd->path.mnt)
1355 break;
1356 /* we know that mountpoint was pinned */
1357 nd->path.dentry = mountpoint;
1358 nd->path.mnt = &mparent->mnt;
1359 inode = inode2;
1360 nd->seq = seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11001361 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001362 }
Al Viroaed434a2015-05-12 12:22:47 -04001363 while (unlikely(d_mountpoint(nd->path.dentry))) {
Al Virob37199e2014-03-20 15:18:22 -04001364 struct mount *mounted;
1365 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
Al Viroaed434a2015-05-12 12:22:47 -04001366 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1367 return -ECHILD;
Al Virob37199e2014-03-20 15:18:22 -04001368 if (!mounted)
1369 break;
1370 nd->path.mnt = &mounted->mnt;
1371 nd->path.dentry = mounted->mnt.mnt_root;
Al Viro4023bfc2014-09-13 21:59:43 -04001372 inode = nd->path.dentry->d_inode;
Al Virob37199e2014-03-20 15:18:22 -04001373 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Virob37199e2014-03-20 15:18:22 -04001374 }
Al Viro4023bfc2014-09-13 21:59:43 -04001375 nd->inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001376 return 0;
1377}
1378
David Howells9875cf82011-01-14 18:45:21 +00001379/*
David Howellscc53ce52011-01-14 18:45:26 +00001380 * Follow down to the covering mount currently visible to userspace. At each
1381 * point, the filesystem owning that dentry may be queried as to whether the
1382 * caller is permitted to proceed or not.
David Howellscc53ce52011-01-14 18:45:26 +00001383 */
Al Viro7cc90cc2011-03-18 09:04:20 -04001384int follow_down(struct path *path)
David Howellscc53ce52011-01-14 18:45:26 +00001385{
1386 unsigned managed;
1387 int ret;
1388
1389 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1390 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1391 /* Allow the filesystem to manage the transit without i_mutex
1392 * being held.
1393 *
1394 * We indicate to the filesystem if someone is trying to mount
1395 * something here. This gives autofs the chance to deny anyone
1396 * other than its daemon the right to mount on its
1397 * superstructure.
1398 *
1399 * The filesystem may sleep at this point.
1400 */
1401 if (managed & DCACHE_MANAGE_TRANSIT) {
1402 BUG_ON(!path->dentry->d_op);
1403 BUG_ON(!path->dentry->d_op->d_manage);
Ian Kentfb5f51c2016-11-24 08:03:41 +11001404 ret = path->dentry->d_op->d_manage(path, false);
David Howellscc53ce52011-01-14 18:45:26 +00001405 if (ret < 0)
1406 return ret == -EISDIR ? 0 : ret;
1407 }
1408
1409 /* Transit to a mounted filesystem. */
1410 if (managed & DCACHE_MOUNTED) {
1411 struct vfsmount *mounted = lookup_mnt(path);
1412 if (!mounted)
1413 break;
1414 dput(path->dentry);
1415 mntput(path->mnt);
1416 path->mnt = mounted;
1417 path->dentry = dget(mounted->mnt_root);
1418 continue;
1419 }
1420
1421 /* Don't handle automount points here */
1422 break;
1423 }
1424 return 0;
1425}
Al Viro4d359502014-03-14 12:20:17 -04001426EXPORT_SYMBOL(follow_down);
David Howellscc53ce52011-01-14 18:45:26 +00001427
1428/*
David Howells9875cf82011-01-14 18:45:21 +00001429 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1430 */
1431static void follow_mount(struct path *path)
1432{
1433 while (d_mountpoint(path->dentry)) {
1434 struct vfsmount *mounted = lookup_mnt(path);
1435 if (!mounted)
1436 break;
1437 dput(path->dentry);
1438 mntput(path->mnt);
1439 path->mnt = mounted;
1440 path->dentry = dget(mounted->mnt_root);
1441 }
1442}
1443
Eric W. Biedermaneedf2652016-06-02 10:29:47 -05001444static int path_parent_directory(struct path *path)
1445{
1446 struct dentry *old = path->dentry;
1447 /* rare case of legitimate dget_parent()... */
1448 path->dentry = dget_parent(path->dentry);
1449 dput(old);
1450 if (unlikely(!path_connected(path)))
1451 return -ENOENT;
1452 return 0;
1453}
1454
Eric W. Biederman397d4252015-08-15 20:27:13 -05001455static int follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456{
1457 while(1) {
Al Viro2a737872009-04-07 11:49:53 -04001458 if (nd->path.dentry == nd->root.dentry &&
1459 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 break;
1461 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001462 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Eric W. Biedermaneedf2652016-06-02 10:29:47 -05001463 int ret = path_parent_directory(&nd->path);
1464 if (ret)
1465 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 break;
1467 }
Al Viro3088dd72010-01-30 15:47:29 -05001468 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 }
Al Viro79ed0222009-04-18 13:59:41 -04001471 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001472 nd->inode = nd->path.dentry->d_inode;
Eric W. Biederman397d4252015-08-15 20:27:13 -05001473 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474}
1475
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476/*
Oleg Drokinf4fdace2016-07-07 22:04:04 -04001477 * This looks up the name in dcache and possibly revalidates the found dentry.
1478 * NULL is returned if the dentry does not exist in the cache.
Nick Pigginbaa03892010-08-18 04:37:31 +10001479 */
Al Viroe3c13922016-03-06 14:03:27 -05001480static struct dentry *lookup_dcache(const struct qstr *name,
1481 struct dentry *dir,
Al Viro6c51e512016-03-05 20:09:32 -05001482 unsigned int flags)
Nick Pigginbaa03892010-08-18 04:37:31 +10001483{
Al Viroa89f8332017-01-09 22:25:28 -05001484 struct dentry *dentry = d_lookup(dir, name);
Miklos Szeredibad61182012-03-26 12:54:24 +02001485 if (dentry) {
Al Viroa89f8332017-01-09 22:25:28 -05001486 int error = d_revalidate(dentry, flags);
1487 if (unlikely(error <= 0)) {
1488 if (!error)
1489 d_invalidate(dentry);
1490 dput(dentry);
1491 return ERR_PTR(error);
Miklos Szeredibad61182012-03-26 12:54:24 +02001492 }
1493 }
Nick Pigginbaa03892010-08-18 04:37:31 +10001494 return dentry;
1495}
1496
1497/*
J. Bruce Fields13a2c3b2013-10-23 16:09:16 -04001498 * Call i_op->lookup on the dentry. The dentry must be negative and
1499 * unhashed.
Miklos Szeredibad61182012-03-26 12:54:24 +02001500 *
1501 * dir->d_inode->i_mutex must be held
Josef Bacik44396f42011-05-31 11:58:49 -04001502 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001503static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
Al Viro72bd8662012-06-10 17:17:17 -04001504 unsigned int flags)
Josef Bacik44396f42011-05-31 11:58:49 -04001505{
Josef Bacik44396f42011-05-31 11:58:49 -04001506 struct dentry *old;
1507
1508 /* Don't create child dentry for a dead directory. */
Miklos Szeredibad61182012-03-26 12:54:24 +02001509 if (unlikely(IS_DEADDIR(dir))) {
Miklos Szeredie188dc02012-02-03 14:25:18 +01001510 dput(dentry);
Josef Bacik44396f42011-05-31 11:58:49 -04001511 return ERR_PTR(-ENOENT);
Miklos Szeredie188dc02012-02-03 14:25:18 +01001512 }
Josef Bacik44396f42011-05-31 11:58:49 -04001513
Al Viro72bd8662012-06-10 17:17:17 -04001514 old = dir->i_op->lookup(dir, dentry, flags);
Josef Bacik44396f42011-05-31 11:58:49 -04001515 if (unlikely(old)) {
1516 dput(dentry);
1517 dentry = old;
1518 }
1519 return dentry;
1520}
1521
Al Viroe3c13922016-03-06 14:03:27 -05001522static struct dentry *__lookup_hash(const struct qstr *name,
Al Viro72bd8662012-06-10 17:17:17 -04001523 struct dentry *base, unsigned int flags)
Al Viroa3255542012-03-30 14:41:51 -04001524{
Al Viro6c51e512016-03-05 20:09:32 -05001525 struct dentry *dentry = lookup_dcache(name, base, flags);
Al Viroa3255542012-03-30 14:41:51 -04001526
Al Viro6c51e512016-03-05 20:09:32 -05001527 if (dentry)
Miklos Szeredibad61182012-03-26 12:54:24 +02001528 return dentry;
Al Viroa3255542012-03-30 14:41:51 -04001529
Al Viro6c51e512016-03-05 20:09:32 -05001530 dentry = d_alloc(base, name);
1531 if (unlikely(!dentry))
1532 return ERR_PTR(-ENOMEM);
1533
Al Viro72bd8662012-06-10 17:17:17 -04001534 return lookup_real(base->d_inode, dentry, flags);
Al Viroa3255542012-03-30 14:41:51 -04001535}
1536
Al Viroe97cdc82013-01-24 18:16:00 -05001537static int lookup_fast(struct nameidata *nd,
Al Viro254cf582015-05-05 09:40:46 -04001538 struct path *path, struct inode **inode,
1539 unsigned *seqp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540{
Jan Blunck4ac91372008-02-14 19:34:32 -08001541 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001542 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001543 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001544 int err;
1545
Al Viro3cac2602009-08-13 18:27:43 +04001546 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001547 * Rename seqlock is not required here because in the off chance
Al Viro5d0f49c2016-03-05 21:32:53 -05001548 * of a false negative due to a concurrent rename, the caller is
1549 * going to fall back to non-racy lookup.
Nick Pigginb04f7842010-08-18 04:37:34 +10001550 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001551 if (nd->flags & LOOKUP_RCU) {
1552 unsigned seq;
Al Viro766c4cb2015-05-07 19:24:57 -04001553 bool negative;
Linus Torvaldsda53be12013-05-21 15:22:44 -07001554 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
Al Viro5d0f49c2016-03-05 21:32:53 -05001555 if (unlikely(!dentry)) {
Al Viro4675ac32017-01-09 22:29:15 -05001556 if (unlazy_walk(nd))
Al Viro5d0f49c2016-03-05 21:32:53 -05001557 return -ECHILD;
Al Viroe9742b52016-03-05 22:04:59 -05001558 return 0;
Al Viro5d0f49c2016-03-05 21:32:53 -05001559 }
Al Viro5a18fff2011-03-11 04:44:53 -05001560
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001561 /*
1562 * This sequence count validates that the inode matches
1563 * the dentry name information from lookup.
1564 */
David Howells63afdfc2015-05-06 15:59:00 +01001565 *inode = d_backing_inode(dentry);
Al Viro766c4cb2015-05-07 19:24:57 -04001566 negative = d_is_negative(dentry);
Al Viro5d0f49c2016-03-05 21:32:53 -05001567 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001568 return -ECHILD;
1569
1570 /*
1571 * This sequence count validates that the parent had no
1572 * changes while we did the lookup of the dentry above.
1573 *
1574 * The memory barrier in read_seqcount_begin of child is
1575 * enough, we can use __read_seqcount_retry here.
1576 */
Al Viro5d0f49c2016-03-05 21:32:53 -05001577 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
Nick Piggin31e6b012011-01-07 17:49:52 +11001578 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001579
Al Viro254cf582015-05-05 09:40:46 -04001580 *seqp = seq;
Al Viroa89f8332017-01-09 22:25:28 -05001581 status = d_revalidate(dentry, nd->flags);
Al Viro209a7fb2017-01-09 01:35:39 -05001582 if (likely(status > 0)) {
Al Viro5d0f49c2016-03-05 21:32:53 -05001583 /*
1584 * Note: do negative dentry check after revalidation in
1585 * case that drops it.
1586 */
1587 if (unlikely(negative))
1588 return -ENOENT;
1589 path->mnt = mnt;
1590 path->dentry = dentry;
1591 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
Al Viroe9742b52016-03-05 22:04:59 -05001592 return 1;
Al Viro24643082011-02-15 01:26:22 -05001593 }
Al Viro4675ac32017-01-09 22:29:15 -05001594 if (unlazy_child(nd, dentry, seq))
Al Viro209a7fb2017-01-09 01:35:39 -05001595 return -ECHILD;
1596 if (unlikely(status == -ECHILD))
1597 /* we'd been told to redo it in non-rcu mode */
1598 status = d_revalidate(dentry, nd->flags);
Al Viro5a18fff2011-03-11 04:44:53 -05001599 } else {
Al Viroe97cdc82013-01-24 18:16:00 -05001600 dentry = __d_lookup(parent, &nd->last);
Al Viro5d0f49c2016-03-05 21:32:53 -05001601 if (unlikely(!dentry))
Al Viroe9742b52016-03-05 22:04:59 -05001602 return 0;
Al Viroa89f8332017-01-09 22:25:28 -05001603 status = d_revalidate(dentry, nd->flags);
Nick Piggin31e6b012011-01-07 17:49:52 +11001604 }
Al Viro5a18fff2011-03-11 04:44:53 -05001605 if (unlikely(status <= 0)) {
Al Viroe9742b52016-03-05 22:04:59 -05001606 if (!status)
Al Viro5d0f49c2016-03-05 21:32:53 -05001607 d_invalidate(dentry);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001608 dput(dentry);
Al Viro5d0f49c2016-03-05 21:32:53 -05001609 return status;
Al Viro5a18fff2011-03-11 04:44:53 -05001610 }
Al Viro766c4cb2015-05-07 19:24:57 -04001611 if (unlikely(d_is_negative(dentry))) {
1612 dput(dentry);
1613 return -ENOENT;
1614 }
Al Viro5d0f49c2016-03-05 21:32:53 -05001615
David Howells9875cf82011-01-14 18:45:21 +00001616 path->mnt = mnt;
1617 path->dentry = dentry;
NeilBrown756daf22015-03-23 13:37:38 +11001618 err = follow_managed(path, nd);
Al Viroe9742b52016-03-05 22:04:59 -05001619 if (likely(err > 0))
David Howells63afdfc2015-05-06 15:59:00 +01001620 *inode = d_backing_inode(path->dentry);
Al Viro84027522015-04-22 10:30:08 -04001621 return err;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001622}
1623
1624/* Fast lookup failed, do it the slow way */
Al Viroe3c13922016-03-06 14:03:27 -05001625static struct dentry *lookup_slow(const struct qstr *name,
1626 struct dentry *dir,
1627 unsigned int flags)
Miklos Szeredi697f5142012-05-21 17:30:05 +02001628{
Al Viro94bdd652016-04-15 02:42:04 -04001629 struct dentry *dentry = ERR_PTR(-ENOENT), *old;
Al Viro19363862016-04-14 19:33:34 -04001630 struct inode *inode = dir->d_inode;
Al Virod9171b92016-04-15 03:33:13 -04001631 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
Al Viro19363862016-04-14 19:33:34 -04001632
Al Viro9902af72016-04-15 15:08:36 -04001633 inode_lock_shared(inode);
Al Viro19363862016-04-14 19:33:34 -04001634 /* Don't go there if it's already dead */
Al Viro94bdd652016-04-15 02:42:04 -04001635 if (unlikely(IS_DEADDIR(inode)))
1636 goto out;
1637again:
Al Virod9171b92016-04-15 03:33:13 -04001638 dentry = d_alloc_parallel(dir, name, &wq);
Al Viro94bdd652016-04-15 02:42:04 -04001639 if (IS_ERR(dentry))
1640 goto out;
1641 if (unlikely(!d_in_lookup(dentry))) {
Al Viroa89f8332017-01-09 22:25:28 -05001642 if (!(flags & LOOKUP_NO_REVAL)) {
Al Viro949a8522016-03-06 14:20:52 -05001643 int error = d_revalidate(dentry, flags);
1644 if (unlikely(error <= 0)) {
Al Viro94bdd652016-04-15 02:42:04 -04001645 if (!error) {
Al Viro949a8522016-03-06 14:20:52 -05001646 d_invalidate(dentry);
Al Viro94bdd652016-04-15 02:42:04 -04001647 dput(dentry);
1648 goto again;
1649 }
Al Viro949a8522016-03-06 14:20:52 -05001650 dput(dentry);
1651 dentry = ERR_PTR(error);
1652 }
1653 }
Al Viro94bdd652016-04-15 02:42:04 -04001654 } else {
1655 old = inode->i_op->lookup(inode, dentry, flags);
1656 d_lookup_done(dentry);
1657 if (unlikely(old)) {
1658 dput(dentry);
1659 dentry = old;
Al Viro949a8522016-03-06 14:20:52 -05001660 }
1661 }
Al Viro94bdd652016-04-15 02:42:04 -04001662out:
Al Viro9902af72016-04-15 15:08:36 -04001663 inode_unlock_shared(inode);
Al Viroe3c13922016-03-06 14:03:27 -05001664 return dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665}
1666
Al Viro52094c82011-02-21 21:34:47 -05001667static inline int may_lookup(struct nameidata *nd)
1668{
1669 if (nd->flags & LOOKUP_RCU) {
Al Viro4ad5abb2011-06-20 19:57:03 -04001670 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
Al Viro52094c82011-02-21 21:34:47 -05001671 if (err != -ECHILD)
1672 return err;
Al Viro4675ac32017-01-09 22:29:15 -05001673 if (unlazy_walk(nd))
Al Viro52094c82011-02-21 21:34:47 -05001674 return -ECHILD;
1675 }
Al Viro4ad5abb2011-06-20 19:57:03 -04001676 return inode_permission(nd->inode, MAY_EXEC);
Al Viro52094c82011-02-21 21:34:47 -05001677}
1678
Al Viro9856fa12011-03-04 14:22:06 -05001679static inline int handle_dots(struct nameidata *nd, int type)
1680{
1681 if (type == LAST_DOTDOT) {
Al Viro9e6697e2015-12-05 20:07:21 -05001682 if (!nd->root.mnt)
1683 set_root(nd);
Al Viro9856fa12011-03-04 14:22:06 -05001684 if (nd->flags & LOOKUP_RCU) {
Al Viro70291ae2015-05-04 07:53:00 -04001685 return follow_dotdot_rcu(nd);
Al Viro9856fa12011-03-04 14:22:06 -05001686 } else
Eric W. Biederman397d4252015-08-15 20:27:13 -05001687 return follow_dotdot(nd);
Al Viro9856fa12011-03-04 14:22:06 -05001688 }
1689 return 0;
1690}
1691
Al Viro181548c2015-05-07 19:54:34 -04001692static int pick_link(struct nameidata *nd, struct path *link,
1693 struct inode *inode, unsigned seq)
Al Virod63ff282015-05-04 18:13:23 -04001694{
Al Viro626de992015-05-04 18:26:59 -04001695 int error;
Al Viro1cf26652015-05-06 16:01:56 -04001696 struct saved *last;
NeilBrown756daf22015-03-23 13:37:38 +11001697 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
Al Viro626de992015-05-04 18:26:59 -04001698 path_to_nameidata(link, nd);
1699 return -ELOOP;
1700 }
Al Virobc40aee2015-05-09 13:04:24 -04001701 if (!(nd->flags & LOOKUP_RCU)) {
Al Viro79733872015-05-09 12:55:43 -04001702 if (link->mnt == nd->path.mnt)
1703 mntget(link->mnt);
Al Virod63ff282015-05-04 18:13:23 -04001704 }
Al Viro626de992015-05-04 18:26:59 -04001705 error = nd_alloc_stack(nd);
1706 if (unlikely(error)) {
Al Virobc40aee2015-05-09 13:04:24 -04001707 if (error == -ECHILD) {
Al Viroad1633a2017-01-08 22:35:31 -05001708 if (unlikely(!legitimize_path(nd, link, seq))) {
1709 drop_links(nd);
1710 nd->depth = 0;
1711 nd->flags &= ~LOOKUP_RCU;
1712 nd->path.mnt = NULL;
1713 nd->path.dentry = NULL;
1714 if (!(nd->flags & LOOKUP_ROOT))
1715 nd->root.mnt = NULL;
1716 rcu_read_unlock();
Al Viro4675ac32017-01-09 22:29:15 -05001717 } else if (likely(unlazy_walk(nd)) == 0)
Al Viroad1633a2017-01-08 22:35:31 -05001718 error = nd_alloc_stack(nd);
Al Virobc40aee2015-05-09 13:04:24 -04001719 }
1720 if (error) {
1721 path_put(link);
1722 return error;
1723 }
Al Viro626de992015-05-04 18:26:59 -04001724 }
1725
Al Viroab104922015-05-10 11:50:01 -04001726 last = nd->stack + nd->depth++;
Al Viro1cf26652015-05-06 16:01:56 -04001727 last->link = *link;
Al Virofceef392015-12-29 15:58:39 -05001728 clear_delayed_call(&last->done);
1729 nd->link_inode = inode;
Al Viro0450b2d2015-05-08 13:23:53 -04001730 last->seq = seq;
Al Virod63ff282015-05-04 18:13:23 -04001731 return 1;
1732}
1733
Al Viro8f64fb12016-11-14 01:50:26 -05001734enum {WALK_FOLLOW = 1, WALK_MORE = 2};
Al Viro31d66bc2016-11-14 01:43:34 -05001735
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001736/*
1737 * Do we need to follow links? We _really_ want to be able
1738 * to do this check without having to look at inode->i_op,
1739 * so we keep a cache of "no, this doesn't need follow_link"
1740 * for the common case.
1741 */
Al Viro8f64fb12016-11-14 01:50:26 -05001742static inline int step_into(struct nameidata *nd, struct path *path,
1743 int flags, struct inode *inode, unsigned seq)
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001744{
Al Viro31d66bc2016-11-14 01:43:34 -05001745 if (!(flags & WALK_MORE) && nd->depth)
1746 put_link(nd);
Al Viro8f64fb12016-11-14 01:50:26 -05001747 if (likely(!d_is_symlink(path->dentry)) ||
1748 !(flags & WALK_FOLLOW || nd->flags & LOOKUP_FOLLOW)) {
1749 /* not a symlink or should not follow */
1750 path_to_nameidata(path, nd);
1751 nd->inode = inode;
1752 nd->seq = seq;
Al Virod63ff282015-05-04 18:13:23 -04001753 return 0;
Al Viro8f64fb12016-11-14 01:50:26 -05001754 }
Al Viroa7f77542016-02-27 19:31:01 -05001755 /* make sure that d_is_symlink above matches inode */
1756 if (nd->flags & LOOKUP_RCU) {
Al Viro8f64fb12016-11-14 01:50:26 -05001757 if (read_seqcount_retry(&path->dentry->d_seq, seq))
Al Viroa7f77542016-02-27 19:31:01 -05001758 return -ECHILD;
1759 }
Al Viro8f64fb12016-11-14 01:50:26 -05001760 return pick_link(nd, path, inode, seq);
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001761}
1762
Al Viro4693a542015-05-04 17:47:11 -04001763static int walk_component(struct nameidata *nd, int flags)
Al Viroce57dfc2011-03-13 19:58:58 -04001764{
Al Virocaa856342015-04-22 17:52:47 -04001765 struct path path;
Al Viroce57dfc2011-03-13 19:58:58 -04001766 struct inode *inode;
Al Viro254cf582015-05-05 09:40:46 -04001767 unsigned seq;
Al Viroce57dfc2011-03-13 19:58:58 -04001768 int err;
1769 /*
1770 * "." and ".." are special - ".." especially so because it has
1771 * to be able to know about the current root directory and
1772 * parent relationships.
1773 */
Al Viro4693a542015-05-04 17:47:11 -04001774 if (unlikely(nd->last_type != LAST_NORM)) {
1775 err = handle_dots(nd, nd->last_type);
Al Viro1c4ff1a2016-11-14 01:39:36 -05001776 if (!(flags & WALK_MORE) && nd->depth)
Al Viro4693a542015-05-04 17:47:11 -04001777 put_link(nd);
1778 return err;
1779 }
Al Viro254cf582015-05-05 09:40:46 -04001780 err = lookup_fast(nd, &path, &inode, &seq);
Al Viroe9742b52016-03-05 22:04:59 -05001781 if (unlikely(err <= 0)) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001782 if (err < 0)
Al Virof0a9ba72015-05-04 07:59:30 -04001783 return err;
Al Viroe3c13922016-03-06 14:03:27 -05001784 path.dentry = lookup_slow(&nd->last, nd->path.dentry,
1785 nd->flags);
1786 if (IS_ERR(path.dentry))
1787 return PTR_ERR(path.dentry);
Al Viro7500c382016-03-31 00:23:05 -04001788
Al Viroe3c13922016-03-06 14:03:27 -05001789 path.mnt = nd->path.mnt;
1790 err = follow_managed(&path, nd);
1791 if (unlikely(err < 0))
Al Virof0a9ba72015-05-04 07:59:30 -04001792 return err;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001793
Al Viro7500c382016-03-31 00:23:05 -04001794 if (unlikely(d_is_negative(path.dentry))) {
1795 path_to_nameidata(&path, nd);
1796 return -ENOENT;
1797 }
1798
Al Viro254cf582015-05-05 09:40:46 -04001799 seq = 0; /* we are already out of RCU mode */
Al Virod4565642016-02-27 19:23:16 -05001800 inode = d_backing_inode(path.dentry);
Al Viroce57dfc2011-03-13 19:58:58 -04001801 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001802
Al Viro8f64fb12016-11-14 01:50:26 -05001803 return step_into(nd, &path, flags, inode, seq);
Al Viroce57dfc2011-03-13 19:58:58 -04001804}
1805
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806/*
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001807 * We can do the critical dentry name comparison and hashing
1808 * operations one word at a time, but we are limited to:
1809 *
1810 * - Architectures with fast unaligned word accesses. We could
1811 * do a "get_unaligned()" if this helps and is sufficiently
1812 * fast.
1813 *
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001814 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1815 * do not trap on the (extremely unlikely) case of a page
1816 * crossing operation.
1817 *
1818 * - Furthermore, we need an efficient 64-bit compile for the
1819 * 64-bit case in order to generate the "number of bytes in
1820 * the final mask". Again, that could be replaced with a
1821 * efficient population count instruction or similar.
1822 */
1823#ifdef CONFIG_DCACHE_WORD_ACCESS
1824
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001825#include <asm/word-at-a-time.h>
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001826
George Spelvin468a9422016-05-26 22:11:51 -04001827#ifdef HASH_MIX
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001828
George Spelvin468a9422016-05-26 22:11:51 -04001829/* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1830
1831#elif defined(CONFIG_64BIT)
George Spelvin2a18da7a2016-05-23 07:43:58 -04001832/*
1833 * Register pressure in the mixing function is an issue, particularly
1834 * on 32-bit x86, but almost any function requires one state value and
1835 * one temporary. Instead, use a function designed for two state values
1836 * and no temporaries.
1837 *
1838 * This function cannot create a collision in only two iterations, so
1839 * we have two iterations to achieve avalanche. In those two iterations,
1840 * we have six layers of mixing, which is enough to spread one bit's
1841 * influence out to 2^6 = 64 state bits.
1842 *
1843 * Rotate constants are scored by considering either 64 one-bit input
1844 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1845 * probability of that delta causing a change to each of the 128 output
1846 * bits, using a sample of random initial states.
1847 *
1848 * The Shannon entropy of the computed probabilities is then summed
1849 * to produce a score. Ideally, any input change has a 50% chance of
1850 * toggling any given output bit.
1851 *
1852 * Mixing scores (in bits) for (12,45):
1853 * Input delta: 1-bit 2-bit
1854 * 1 round: 713.3 42542.6
1855 * 2 rounds: 2753.7 140389.8
1856 * 3 rounds: 5954.1 233458.2
1857 * 4 rounds: 7862.6 256672.2
1858 * Perfect: 8192 258048
1859 * (64*128) (64*63/2 * 128)
1860 */
1861#define HASH_MIX(x, y, a) \
1862 ( x ^= (a), \
1863 y ^= x, x = rol64(x,12),\
1864 x += y, y = rol64(y,45),\
1865 y *= 9 )
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001866
George Spelvin0fed3ac2016-05-02 06:31:01 -04001867/*
George Spelvin2a18da7a2016-05-23 07:43:58 -04001868 * Fold two longs into one 32-bit hash value. This must be fast, but
1869 * latency isn't quite as critical, as there is a fair bit of additional
1870 * work done before the hash value is used.
George Spelvin0fed3ac2016-05-02 06:31:01 -04001871 */
George Spelvin2a18da7a2016-05-23 07:43:58 -04001872static inline unsigned int fold_hash(unsigned long x, unsigned long y)
George Spelvin0fed3ac2016-05-02 06:31:01 -04001873{
George Spelvin2a18da7a2016-05-23 07:43:58 -04001874 y ^= x * GOLDEN_RATIO_64;
1875 y *= GOLDEN_RATIO_64;
1876 return y >> 32;
George Spelvin0fed3ac2016-05-02 06:31:01 -04001877}
1878
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001879#else /* 32-bit case */
1880
George Spelvin2a18da7a2016-05-23 07:43:58 -04001881/*
1882 * Mixing scores (in bits) for (7,20):
1883 * Input delta: 1-bit 2-bit
1884 * 1 round: 330.3 9201.6
1885 * 2 rounds: 1246.4 25475.4
1886 * 3 rounds: 1907.1 31295.1
1887 * 4 rounds: 2042.3 31718.6
1888 * Perfect: 2048 31744
1889 * (32*64) (32*31/2 * 64)
1890 */
1891#define HASH_MIX(x, y, a) \
1892 ( x ^= (a), \
1893 y ^= x, x = rol32(x, 7),\
1894 x += y, y = rol32(y,20),\
1895 y *= 9 )
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001896
George Spelvin2a18da7a2016-05-23 07:43:58 -04001897static inline unsigned int fold_hash(unsigned long x, unsigned long y)
George Spelvin0fed3ac2016-05-02 06:31:01 -04001898{
George Spelvin2a18da7a2016-05-23 07:43:58 -04001899 /* Use arch-optimized multiply if one exists */
1900 return __hash_32(y ^ __hash_32(x));
George Spelvin0fed3ac2016-05-02 06:31:01 -04001901}
1902
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001903#endif
1904
George Spelvin2a18da7a2016-05-23 07:43:58 -04001905/*
1906 * Return the hash of a string of known length. This is carfully
1907 * designed to match hash_name(), which is the more critical function.
1908 * In particular, we must end by hashing a final word containing 0..7
1909 * payload bytes, to match the way that hash_name() iterates until it
1910 * finds the delimiter after the name.
1911 */
Linus Torvalds8387ff22016-06-10 07:51:30 -07001912unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001913{
Linus Torvalds8387ff22016-06-10 07:51:30 -07001914 unsigned long a, x = 0, y = (unsigned long)salt;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001915
1916 for (;;) {
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001917 if (!len)
1918 goto done;
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001919 a = load_unaligned_zeropad(name);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001920 if (len < sizeof(unsigned long))
1921 break;
George Spelvin2a18da7a2016-05-23 07:43:58 -04001922 HASH_MIX(x, y, a);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001923 name += sizeof(unsigned long);
1924 len -= sizeof(unsigned long);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001925 }
George Spelvin2a18da7a2016-05-23 07:43:58 -04001926 x ^= a & bytemask_from_count(len);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001927done:
George Spelvin2a18da7a2016-05-23 07:43:58 -04001928 return fold_hash(x, y);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001929}
1930EXPORT_SYMBOL(full_name_hash);
1931
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001932/* Return the "hash_len" (hash and length) of a null-terminated string */
Linus Torvalds8387ff22016-06-10 07:51:30 -07001933u64 hashlen_string(const void *salt, const char *name)
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001934{
Linus Torvalds8387ff22016-06-10 07:51:30 -07001935 unsigned long a = 0, x = 0, y = (unsigned long)salt;
1936 unsigned long adata, mask, len;
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001937 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1938
Linus Torvalds8387ff22016-06-10 07:51:30 -07001939 len = 0;
1940 goto inside;
1941
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001942 do {
George Spelvin2a18da7a2016-05-23 07:43:58 -04001943 HASH_MIX(x, y, a);
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001944 len += sizeof(unsigned long);
Linus Torvalds8387ff22016-06-10 07:51:30 -07001945inside:
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001946 a = load_unaligned_zeropad(name+len);
1947 } while (!has_zero(a, &adata, &constants));
1948
1949 adata = prep_zero_mask(a, adata, &constants);
1950 mask = create_zero_mask(adata);
George Spelvin2a18da7a2016-05-23 07:43:58 -04001951 x ^= a & zero_bytemask(mask);
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001952
George Spelvin2a18da7a2016-05-23 07:43:58 -04001953 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001954}
1955EXPORT_SYMBOL(hashlen_string);
1956
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001957/*
1958 * Calculate the length and hash of the path component, and
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001959 * return the "hash_len" as the result.
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001960 */
Linus Torvalds8387ff22016-06-10 07:51:30 -07001961static inline u64 hash_name(const void *salt, const char *name)
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001962{
Linus Torvalds8387ff22016-06-10 07:51:30 -07001963 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
1964 unsigned long adata, bdata, mask, len;
Linus Torvalds36126f82012-05-26 10:43:17 -07001965 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001966
Linus Torvalds8387ff22016-06-10 07:51:30 -07001967 len = 0;
1968 goto inside;
1969
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001970 do {
George Spelvin2a18da7a2016-05-23 07:43:58 -04001971 HASH_MIX(x, y, a);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001972 len += sizeof(unsigned long);
Linus Torvalds8387ff22016-06-10 07:51:30 -07001973inside:
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001974 a = load_unaligned_zeropad(name+len);
Linus Torvalds36126f82012-05-26 10:43:17 -07001975 b = a ^ REPEAT_BYTE('/');
1976 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001977
Linus Torvalds36126f82012-05-26 10:43:17 -07001978 adata = prep_zero_mask(a, adata, &constants);
1979 bdata = prep_zero_mask(b, bdata, &constants);
Linus Torvalds36126f82012-05-26 10:43:17 -07001980 mask = create_zero_mask(adata | bdata);
George Spelvin2a18da7a2016-05-23 07:43:58 -04001981 x ^= a & zero_bytemask(mask);
Linus Torvalds36126f82012-05-26 10:43:17 -07001982
George Spelvin2a18da7a2016-05-23 07:43:58 -04001983 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001984}
1985
George Spelvin2a18da7a2016-05-23 07:43:58 -04001986#else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001987
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001988/* Return the hash of a string of known length */
Linus Torvalds8387ff22016-06-10 07:51:30 -07001989unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
Linus Torvalds0145acc2012-03-02 14:32:59 -08001990{
Linus Torvalds8387ff22016-06-10 07:51:30 -07001991 unsigned long hash = init_name_hash(salt);
Linus Torvalds0145acc2012-03-02 14:32:59 -08001992 while (len--)
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001993 hash = partial_name_hash((unsigned char)*name++, hash);
Linus Torvalds0145acc2012-03-02 14:32:59 -08001994 return end_name_hash(hash);
1995}
Linus Torvaldsae942ae2012-03-02 19:40:57 -08001996EXPORT_SYMBOL(full_name_hash);
Linus Torvalds0145acc2012-03-02 14:32:59 -08001997
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001998/* Return the "hash_len" (hash and length) of a null-terminated string */
Linus Torvalds8387ff22016-06-10 07:51:30 -07001999u64 hashlen_string(const void *salt, const char *name)
George Spelvinfcfd2fb2016-05-20 08:41:37 -04002000{
Linus Torvalds8387ff22016-06-10 07:51:30 -07002001 unsigned long hash = init_name_hash(salt);
George Spelvinfcfd2fb2016-05-20 08:41:37 -04002002 unsigned long len = 0, c;
2003
2004 c = (unsigned char)*name;
George Spelvine0ab7af2016-05-29 08:05:56 -04002005 while (c) {
George Spelvinfcfd2fb2016-05-20 08:41:37 -04002006 len++;
2007 hash = partial_name_hash(c, hash);
2008 c = (unsigned char)name[len];
George Spelvine0ab7af2016-05-29 08:05:56 -04002009 }
George Spelvinfcfd2fb2016-05-20 08:41:37 -04002010 return hashlen_create(end_name_hash(hash), len);
2011}
George Spelvinf2a031b2016-05-29 01:26:41 -04002012EXPORT_SYMBOL(hashlen_string);
George Spelvinfcfd2fb2016-05-20 08:41:37 -04002013
Linus Torvalds3ddcd052011-08-06 22:45:50 -07002014/*
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002015 * We know there's a real path component here of at least
2016 * one character.
2017 */
Linus Torvalds8387ff22016-06-10 07:51:30 -07002018static inline u64 hash_name(const void *salt, const char *name)
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002019{
Linus Torvalds8387ff22016-06-10 07:51:30 -07002020 unsigned long hash = init_name_hash(salt);
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002021 unsigned long len = 0, c;
2022
2023 c = (unsigned char)*name;
2024 do {
2025 len++;
2026 hash = partial_name_hash(c, hash);
2027 c = (unsigned char)name[len];
2028 } while (c && c != '/');
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002029 return hashlen_create(end_name_hash(hash), len);
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002030}
2031
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08002032#endif
2033
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002034/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01002036 * This is the basic name resolution function, turning a pathname into
2037 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01002039 * Returns 0 and nd will have valid dentry and mnt on success.
2040 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 */
Al Viro6de88d72009-08-09 01:41:57 +04002042static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 int err;
Al Viro32cd7462015-04-18 20:30:49 -04002045
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 while (*name=='/')
2047 name++;
2048 if (!*name)
Al Viro9e18f102015-04-18 20:44:34 -04002049 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 /* At this point we know we have a real path component. */
2052 for(;;) {
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002053 u64 hash_len;
Al Virofe479a52011-02-22 15:10:03 -05002054 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055
Al Viro52094c82011-02-21 21:34:47 -05002056 err = may_lookup(nd);
George Spelvin2a18da7a2016-05-23 07:43:58 -04002057 if (err)
Al Viro3595e232015-05-09 16:54:45 -04002058 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059
Linus Torvalds8387ff22016-06-10 07:51:30 -07002060 hash_len = hash_name(nd->path.dentry, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061
Al Virofe479a52011-02-22 15:10:03 -05002062 type = LAST_NORM;
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002063 if (name[0] == '.') switch (hashlen_len(hash_len)) {
Al Virofe479a52011-02-22 15:10:03 -05002064 case 2:
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002065 if (name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05002066 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05002067 nd->flags |= LOOKUP_JUMPED;
2068 }
Al Virofe479a52011-02-22 15:10:03 -05002069 break;
2070 case 1:
2071 type = LAST_DOT;
2072 }
Al Viro5a202bc2011-03-08 14:17:44 -05002073 if (likely(type == LAST_NORM)) {
2074 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05002075 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05002076 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
James Hogana060dc52014-09-16 13:07:35 +01002077 struct qstr this = { { .hash_len = hash_len }, .name = name };
Linus Torvaldsda53be12013-05-21 15:22:44 -07002078 err = parent->d_op->d_hash(parent, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05002079 if (err < 0)
Al Viro3595e232015-05-09 16:54:45 -04002080 return err;
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002081 hash_len = this.hash_len;
2082 name = this.name;
Al Viro5a202bc2011-03-08 14:17:44 -05002083 }
2084 }
Al Virofe479a52011-02-22 15:10:03 -05002085
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002086 nd->last.hash_len = hash_len;
2087 nd->last.name = name;
Al Viro5f4a6a62013-01-24 18:04:22 -05002088 nd->last_type = type;
2089
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002090 name += hashlen_len(hash_len);
2091 if (!*name)
Al Virobdf6cbf2015-04-18 20:21:40 -04002092 goto OK;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002093 /*
2094 * If it wasn't NUL, we know it was '/'. Skip that
2095 * slash, and continue until no more slashes.
2096 */
2097 do {
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002098 name++;
2099 } while (unlikely(*name == '/'));
Al Viro8620c232015-05-04 08:58:35 -04002100 if (unlikely(!*name)) {
2101OK:
Al Viro368ee9b2015-05-08 17:19:59 -04002102 /* pathname body, done */
Al Viro8620c232015-05-04 08:58:35 -04002103 if (!nd->depth)
2104 return 0;
2105 name = nd->stack[nd->depth - 1].name;
Al Viro368ee9b2015-05-08 17:19:59 -04002106 /* trailing symlink, done */
Al Viro8620c232015-05-04 08:58:35 -04002107 if (!name)
2108 return 0;
2109 /* last component of nested symlink */
Al Viro8f64fb12016-11-14 01:50:26 -05002110 err = walk_component(nd, WALK_FOLLOW);
Al Viro1c4ff1a2016-11-14 01:39:36 -05002111 } else {
2112 /* not the last component */
Al Viro8f64fb12016-11-14 01:50:26 -05002113 err = walk_component(nd, WALK_FOLLOW | WALK_MORE);
Al Viro8620c232015-05-04 08:58:35 -04002114 }
Al Viroce57dfc2011-03-13 19:58:58 -04002115 if (err < 0)
Al Viro3595e232015-05-09 16:54:45 -04002116 return err;
Al Virofe479a52011-02-22 15:10:03 -05002117
Al Viroce57dfc2011-03-13 19:58:58 -04002118 if (err) {
Al Viro626de992015-05-04 18:26:59 -04002119 const char *s = get_link(nd);
Al Viro5a460272015-04-17 23:44:45 -04002120
Viresh Kumara1c83682015-08-12 15:59:44 +05302121 if (IS_ERR(s))
Al Viro3595e232015-05-09 16:54:45 -04002122 return PTR_ERR(s);
Al Virod40bcc02015-04-18 20:03:03 -04002123 err = 0;
2124 if (unlikely(!s)) {
2125 /* jumped */
Al Virob9ff4422015-05-02 20:19:23 -04002126 put_link(nd);
Al Virod40bcc02015-04-18 20:03:03 -04002127 } else {
Al Virofab51e82015-05-10 11:01:00 -04002128 nd->stack[nd->depth - 1].name = name;
2129 name = s;
2130 continue;
Al Virod40bcc02015-04-18 20:03:03 -04002131 }
Nick Piggin31e6b012011-01-07 17:49:52 +11002132 }
Al Viro97242f92015-08-01 19:59:28 -04002133 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2134 if (nd->flags & LOOKUP_RCU) {
Al Viro4675ac32017-01-09 22:29:15 -05002135 if (unlazy_walk(nd))
Al Viro97242f92015-08-01 19:59:28 -04002136 return -ECHILD;
2137 }
Al Viro3595e232015-05-09 16:54:45 -04002138 return -ENOTDIR;
Al Viro97242f92015-08-01 19:59:28 -04002139 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141}
2142
Al Viroc8a53ee2015-05-12 18:43:07 -04002143static const char *path_init(struct nameidata *nd, unsigned flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144{
Al Viroc8a53ee2015-05-12 18:43:07 -04002145 const char *s = nd->name->name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146
Linus Torvaldsc0eb0272017-04-02 17:10:08 -07002147 if (!*s)
2148 flags &= ~LOOKUP_RCU;
2149
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro980f3ea2014-11-20 14:20:24 -05002151 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 nd->depth = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05002153 if (flags & LOOKUP_ROOT) {
David Howellsb18825a2013-09-12 19:22:53 +01002154 struct dentry *root = nd->root.dentry;
2155 struct inode *inode = root->d_inode;
Al Viro93893862017-04-15 17:29:14 -04002156 if (*s && unlikely(!d_can_lookup(root)))
2157 return ERR_PTR(-ENOTDIR);
Al Viro5b6ca022011-03-09 23:04:47 -05002158 nd->path = nd->root;
2159 nd->inode = inode;
2160 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05002161 rcu_read_lock();
Al Viro5b6ca022011-03-09 23:04:47 -05002162 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro8f47a0162015-05-09 19:02:01 -04002163 nd->root_seq = nd->seq;
Al Viro48a066e2013-09-29 22:06:07 -04002164 nd->m_seq = read_seqbegin(&mount_lock);
Al Viro5b6ca022011-03-09 23:04:47 -05002165 } else {
2166 path_get(&nd->path);
2167 }
Al Viro368ee9b2015-05-08 17:19:59 -04002168 return s;
Al Viro5b6ca022011-03-09 23:04:47 -05002169 }
2170
Al Viro2a737872009-04-07 11:49:53 -04002171 nd->root.mnt = NULL;
Al Viro248fb5b2015-12-05 20:51:58 -05002172 nd->path.mnt = NULL;
2173 nd->path.dentry = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174
Al Viro48a066e2013-09-29 22:06:07 -04002175 nd->m_seq = read_seqbegin(&mount_lock);
Al Virofd2f7cb2015-02-22 20:07:13 -05002176 if (*s == '/') {
Al Viro9e6697e2015-12-05 20:07:21 -05002177 if (flags & LOOKUP_RCU)
Al Viro8b61e742013-11-08 12:45:01 -05002178 rcu_read_lock();
Al Viro9e6697e2015-12-05 20:07:21 -05002179 set_root(nd);
Al Viro248fb5b2015-12-05 20:51:58 -05002180 if (likely(!nd_jump_root(nd)))
Al Viroef55d912015-12-05 20:25:06 -05002181 return s;
Al Viro248fb5b2015-12-05 20:51:58 -05002182 nd->root.mnt = NULL;
Al Viroef55d912015-12-05 20:25:06 -05002183 rcu_read_unlock();
2184 return ERR_PTR(-ECHILD);
Al Viroc8a53ee2015-05-12 18:43:07 -04002185 } else if (nd->dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05002186 if (flags & LOOKUP_RCU) {
2187 struct fs_struct *fs = current->fs;
2188 unsigned seq;
2189
Al Viro8b61e742013-11-08 12:45:01 -05002190 rcu_read_lock();
Al Viroe41f7d42011-02-22 14:02:58 -05002191
2192 do {
2193 seq = read_seqcount_begin(&fs->seq);
2194 nd->path = fs->pwd;
Al Viroef55d912015-12-05 20:25:06 -05002195 nd->inode = nd->path.dentry->d_inode;
Al Viroe41f7d42011-02-22 14:02:58 -05002196 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2197 } while (read_seqcount_retry(&fs->seq, seq));
2198 } else {
2199 get_fs_pwd(current->fs, &nd->path);
Al Viroef55d912015-12-05 20:25:06 -05002200 nd->inode = nd->path.dentry->d_inode;
Al Viroe41f7d42011-02-22 14:02:58 -05002201 }
Al Viroef55d912015-12-05 20:25:06 -05002202 return s;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002203 } else {
Jeff Layton582aa642012-12-11 08:56:16 -05002204 /* Caller must check execute permissions on the starting path component */
Al Viroc8a53ee2015-05-12 18:43:07 -04002205 struct fd f = fdget_raw(nd->dfd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002206 struct dentry *dentry;
2207
Al Viro2903ff02012-08-28 12:52:22 -04002208 if (!f.file)
Al Viro368ee9b2015-05-08 17:19:59 -04002209 return ERR_PTR(-EBADF);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002210
Al Viro2903ff02012-08-28 12:52:22 -04002211 dentry = f.file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002212
Al Virofd2f7cb2015-02-22 20:07:13 -05002213 if (*s) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002214 if (!d_can_lookup(dentry)) {
Al Viro2903ff02012-08-28 12:52:22 -04002215 fdput(f);
Al Viro368ee9b2015-05-08 17:19:59 -04002216 return ERR_PTR(-ENOTDIR);
Al Viro2903ff02012-08-28 12:52:22 -04002217 }
Al Virof52e0c12011-03-14 18:56:51 -04002218 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002219
Al Viro2903ff02012-08-28 12:52:22 -04002220 nd->path = f.file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05002221 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05002222 rcu_read_lock();
Al Viro34a26b92015-05-11 08:05:05 -04002223 nd->inode = nd->path.dentry->d_inode;
2224 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viroe41f7d42011-02-22 14:02:58 -05002225 } else {
Al Viro2903ff02012-08-28 12:52:22 -04002226 path_get(&nd->path);
Al Viro34a26b92015-05-11 08:05:05 -04002227 nd->inode = nd->path.dentry->d_inode;
Al Viroe41f7d42011-02-22 14:02:58 -05002228 }
Al Viro34a26b92015-05-11 08:05:05 -04002229 fdput(f);
Al Viro368ee9b2015-05-08 17:19:59 -04002230 return s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231 }
Al Viro9b4a9b12009-04-07 11:44:16 -04002232}
2233
Al Viro3bdba282015-05-08 17:37:07 -04002234static const char *trailing_symlink(struct nameidata *nd)
Al Viro95fa25d2015-04-22 13:46:57 -04002235{
2236 const char *s;
Al Virofec2fa22015-05-06 15:58:18 -04002237 int error = may_follow_link(nd);
Al Virodeb106c2015-05-08 18:05:21 -04002238 if (unlikely(error))
Al Viro3bdba282015-05-08 17:37:07 -04002239 return ERR_PTR(error);
Al Viro95fa25d2015-04-22 13:46:57 -04002240 nd->flags |= LOOKUP_PARENT;
Al Virofab51e82015-05-10 11:01:00 -04002241 nd->stack[0].name = NULL;
Al Viro3b2e7f72015-04-19 00:53:50 -04002242 s = get_link(nd);
Al Virodeb106c2015-05-08 18:05:21 -04002243 return s ? s : "";
Al Viro95fa25d2015-04-22 13:46:57 -04002244}
2245
Al Virocaa856342015-04-22 17:52:47 -04002246static inline int lookup_last(struct nameidata *nd)
Al Virobd92d7f2011-03-14 19:54:59 -04002247{
2248 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2249 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2250
2251 nd->flags &= ~LOOKUP_PARENT;
Al Viro1c4ff1a2016-11-14 01:39:36 -05002252 return walk_component(nd, 0);
Al Virobd92d7f2011-03-14 19:54:59 -04002253}
2254
Al Viro4f757f32017-04-15 17:31:22 -04002255static int handle_lookup_down(struct nameidata *nd)
2256{
2257 struct path path = nd->path;
2258 struct inode *inode = nd->inode;
2259 unsigned seq = nd->seq;
2260 int err;
2261
2262 if (nd->flags & LOOKUP_RCU) {
2263 /*
2264 * don't bother with unlazy_walk on failure - we are
2265 * at the very beginning of walk, so we lose nothing
2266 * if we simply redo everything in non-RCU mode
2267 */
2268 if (unlikely(!__follow_mount_rcu(nd, &path, &inode, &seq)))
2269 return -ECHILD;
2270 } else {
2271 dget(path.dentry);
2272 err = follow_managed(&path, nd);
2273 if (unlikely(err < 0))
2274 return err;
2275 inode = d_backing_inode(path.dentry);
2276 seq = 0;
2277 }
2278 path_to_nameidata(&path, nd);
2279 nd->inode = inode;
2280 nd->seq = seq;
2281 return 0;
2282}
2283
Al Viro9b4a9b12009-04-07 11:44:16 -04002284/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroc8a53ee2015-05-12 18:43:07 -04002285static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
Al Viro9b4a9b12009-04-07 11:44:16 -04002286{
Al Viroc8a53ee2015-05-12 18:43:07 -04002287 const char *s = path_init(nd, flags);
Al Virobd92d7f2011-03-14 19:54:59 -04002288 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11002289
Al Viro368ee9b2015-05-08 17:19:59 -04002290 if (IS_ERR(s))
2291 return PTR_ERR(s);
Al Viro4f757f32017-04-15 17:31:22 -04002292
2293 if (unlikely(flags & LOOKUP_DOWN)) {
2294 err = handle_lookup_down(nd);
2295 if (unlikely(err < 0)) {
2296 terminate_walk(nd);
2297 return err;
2298 }
2299 }
2300
Al Viro3bdba282015-05-08 17:37:07 -04002301 while (!(err = link_path_walk(s, nd))
2302 && ((err = lookup_last(nd)) > 0)) {
2303 s = trailing_symlink(nd);
2304 if (IS_ERR(s)) {
2305 err = PTR_ERR(s);
2306 break;
Al Virobd92d7f2011-03-14 19:54:59 -04002307 }
2308 }
Al Viro9f1fafe2011-03-25 11:00:12 -04002309 if (!err)
2310 err = complete_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04002311
Al Virodeb106c2015-05-08 18:05:21 -04002312 if (!err && nd->flags & LOOKUP_DIRECTORY)
2313 if (!d_can_lookup(nd->path.dentry))
Al Virobd23a532011-03-23 09:56:30 -04002314 err = -ENOTDIR;
Al Viro625b6d12015-05-12 16:36:12 -04002315 if (!err) {
2316 *path = nd->path;
2317 nd->path.mnt = NULL;
2318 nd->path.dentry = NULL;
2319 }
2320 terminate_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04002321 return err;
Al Viroee0827c2011-02-21 23:38:09 -05002322}
Nick Piggin31e6b012011-01-07 17:49:52 +11002323
Al Viro625b6d12015-05-12 16:36:12 -04002324static int filename_lookup(int dfd, struct filename *name, unsigned flags,
Al Viro9ad1aaa2015-05-12 16:44:39 -04002325 struct path *path, struct path *root)
Jeff Layton873f1ee2012-10-10 15:25:29 -04002326{
Al Viro894bc8c2015-05-02 07:16:16 -04002327 int retval;
Al Viro9883d182015-05-13 07:28:08 -04002328 struct nameidata nd;
Al Viroabc9f5b2015-05-12 16:53:42 -04002329 if (IS_ERR(name))
2330 return PTR_ERR(name);
Al Viro9ad1aaa2015-05-12 16:44:39 -04002331 if (unlikely(root)) {
2332 nd.root = *root;
2333 flags |= LOOKUP_ROOT;
2334 }
Al Viro9883d182015-05-13 07:28:08 -04002335 set_nameidata(&nd, dfd, name);
Al Viroc8a53ee2015-05-12 18:43:07 -04002336 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002337 if (unlikely(retval == -ECHILD))
Al Viroc8a53ee2015-05-12 18:43:07 -04002338 retval = path_lookupat(&nd, flags, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002339 if (unlikely(retval == -ESTALE))
Al Viroc8a53ee2015-05-12 18:43:07 -04002340 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002341
2342 if (likely(!retval))
Al Viro625b6d12015-05-12 16:36:12 -04002343 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
Al Viro9883d182015-05-13 07:28:08 -04002344 restore_nameidata();
Al Viroe4bd1c12015-05-12 16:40:39 -04002345 putname(name);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002346 return retval;
2347}
2348
Al Viro8bcb77f2015-05-08 16:59:20 -04002349/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroc8a53ee2015-05-12 18:43:07 -04002350static int path_parentat(struct nameidata *nd, unsigned flags,
Al Viro391172c2015-05-09 11:19:16 -04002351 struct path *parent)
Al Viro8bcb77f2015-05-08 16:59:20 -04002352{
Al Viroc8a53ee2015-05-12 18:43:07 -04002353 const char *s = path_init(nd, flags);
Al Viro368ee9b2015-05-08 17:19:59 -04002354 int err;
2355 if (IS_ERR(s))
2356 return PTR_ERR(s);
2357 err = link_path_walk(s, nd);
Al Viro8bcb77f2015-05-08 16:59:20 -04002358 if (!err)
2359 err = complete_walk(nd);
Al Viro391172c2015-05-09 11:19:16 -04002360 if (!err) {
2361 *parent = nd->path;
2362 nd->path.mnt = NULL;
2363 nd->path.dentry = NULL;
2364 }
2365 terminate_walk(nd);
Al Viro8bcb77f2015-05-08 16:59:20 -04002366 return err;
2367}
2368
Al Viro5c31b6c2015-05-12 17:32:54 -04002369static struct filename *filename_parentat(int dfd, struct filename *name,
Al Viro391172c2015-05-09 11:19:16 -04002370 unsigned int flags, struct path *parent,
2371 struct qstr *last, int *type)
Al Viro8bcb77f2015-05-08 16:59:20 -04002372{
2373 int retval;
Al Viro9883d182015-05-13 07:28:08 -04002374 struct nameidata nd;
Al Viro8bcb77f2015-05-08 16:59:20 -04002375
Al Viro5c31b6c2015-05-12 17:32:54 -04002376 if (IS_ERR(name))
2377 return name;
Al Viro9883d182015-05-13 07:28:08 -04002378 set_nameidata(&nd, dfd, name);
Al Viroc8a53ee2015-05-12 18:43:07 -04002379 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
Al Viro8bcb77f2015-05-08 16:59:20 -04002380 if (unlikely(retval == -ECHILD))
Al Viroc8a53ee2015-05-12 18:43:07 -04002381 retval = path_parentat(&nd, flags, parent);
Al Viro8bcb77f2015-05-08 16:59:20 -04002382 if (unlikely(retval == -ESTALE))
Al Viroc8a53ee2015-05-12 18:43:07 -04002383 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
Al Viro391172c2015-05-09 11:19:16 -04002384 if (likely(!retval)) {
2385 *last = nd.last;
2386 *type = nd.last_type;
2387 audit_inode(name, parent->dentry, LOOKUP_PARENT);
Al Viro5c31b6c2015-05-12 17:32:54 -04002388 } else {
2389 putname(name);
2390 name = ERR_PTR(retval);
Al Viro391172c2015-05-09 11:19:16 -04002391 }
Al Viro9883d182015-05-13 07:28:08 -04002392 restore_nameidata();
Al Viro5c31b6c2015-05-12 17:32:54 -04002393 return name;
Al Viro8bcb77f2015-05-08 16:59:20 -04002394}
2395
Al Viro79714f72012-06-15 03:01:42 +04002396/* does lookup, returns the object with parent locked */
2397struct dentry *kern_path_locked(const char *name, struct path *path)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002398{
Al Viro5c31b6c2015-05-12 17:32:54 -04002399 struct filename *filename;
2400 struct dentry *d;
Al Viro391172c2015-05-09 11:19:16 -04002401 struct qstr last;
2402 int type;
Paul Moore51689102015-01-22 00:00:03 -05002403
Al Viro5c31b6c2015-05-12 17:32:54 -04002404 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2405 &last, &type);
Paul Moore51689102015-01-22 00:00:03 -05002406 if (IS_ERR(filename))
2407 return ERR_CAST(filename);
Al Viro5c31b6c2015-05-12 17:32:54 -04002408 if (unlikely(type != LAST_NORM)) {
Al Viro391172c2015-05-09 11:19:16 -04002409 path_put(path);
Al Viro5c31b6c2015-05-12 17:32:54 -04002410 putname(filename);
2411 return ERR_PTR(-EINVAL);
Al Viro79714f72012-06-15 03:01:42 +04002412 }
Al Viro59551022016-01-22 15:40:57 -05002413 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
Al Viro391172c2015-05-09 11:19:16 -04002414 d = __lookup_hash(&last, path->dentry, 0);
Al Viro79714f72012-06-15 03:01:42 +04002415 if (IS_ERR(d)) {
Al Viro59551022016-01-22 15:40:57 -05002416 inode_unlock(path->dentry->d_inode);
Al Viro391172c2015-05-09 11:19:16 -04002417 path_put(path);
Al Viro79714f72012-06-15 03:01:42 +04002418 }
Paul Moore51689102015-01-22 00:00:03 -05002419 putname(filename);
Al Viro79714f72012-06-15 03:01:42 +04002420 return d;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002421}
2422
Al Virod1811462008-08-02 00:49:18 -04002423int kern_path(const char *name, unsigned int flags, struct path *path)
2424{
Al Viroabc9f5b2015-05-12 16:53:42 -04002425 return filename_lookup(AT_FDCWD, getname_kernel(name),
2426 flags, path, NULL);
Al Virod1811462008-08-02 00:49:18 -04002427}
Al Viro4d359502014-03-14 12:20:17 -04002428EXPORT_SYMBOL(kern_path);
Al Virod1811462008-08-02 00:49:18 -04002429
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002430/**
2431 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2432 * @dentry: pointer to dentry of the base directory
2433 * @mnt: pointer to vfs mount of the base directory
2434 * @name: pointer to file name
2435 * @flags: lookup flags
Al Viroe0a01242011-06-27 17:00:37 -04002436 * @path: pointer to struct path to fill
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002437 */
2438int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2439 const char *name, unsigned int flags,
Al Viroe0a01242011-06-27 17:00:37 -04002440 struct path *path)
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002441{
Al Viro9ad1aaa2015-05-12 16:44:39 -04002442 struct path root = {.mnt = mnt, .dentry = dentry};
Al Viro9ad1aaa2015-05-12 16:44:39 -04002443 /* the first argument of filename_lookup() is ignored with root */
Al Viroabc9f5b2015-05-12 16:53:42 -04002444 return filename_lookup(AT_FDCWD, getname_kernel(name),
2445 flags , path, &root);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002446}
Al Viro4d359502014-03-14 12:20:17 -04002447EXPORT_SYMBOL(vfs_path_lookup);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002448
Christoph Hellwigeead1912007-10-16 23:25:38 -07002449/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07002450 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07002451 * @name: pathname component to lookup
2452 * @base: base directory to lookup from
2453 * @len: maximum length @len should be interpreted to
2454 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07002455 * Note that this routine is purely a helper for filesystem usage and should
Al Viro9e7543e2015-02-23 02:49:48 -05002456 * not be called by generic code.
NeilBrownbbddca82016-01-07 16:08:20 -05002457 *
2458 * The caller must hold base->i_mutex.
Christoph Hellwigeead1912007-10-16 23:25:38 -07002459 */
James Morris057f6c02007-04-26 00:12:05 -07002460struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2461{
James Morris057f6c02007-04-26 00:12:05 -07002462 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05002463 unsigned int c;
Miklos Szeredicda309d2012-03-26 12:54:21 +02002464 int err;
James Morris057f6c02007-04-26 00:12:05 -07002465
Al Viro59551022016-01-22 15:40:57 -05002466 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
David Woodhouse2f9092e2009-04-20 23:18:37 +01002467
Al Viro6a96ba52011-03-07 23:49:20 -05002468 this.name = name;
2469 this.len = len;
Linus Torvalds8387ff22016-06-10 07:51:30 -07002470 this.hash = full_name_hash(base, name, len);
Al Viro6a96ba52011-03-07 23:49:20 -05002471 if (!len)
2472 return ERR_PTR(-EACCES);
2473
Al Viro21d8a152012-11-29 22:17:21 -05002474 if (unlikely(name[0] == '.')) {
2475 if (len < 2 || (len == 2 && name[1] == '.'))
2476 return ERR_PTR(-EACCES);
2477 }
2478
Al Viro6a96ba52011-03-07 23:49:20 -05002479 while (len--) {
2480 c = *(const unsigned char *)name++;
2481 if (c == '/' || c == '\0')
2482 return ERR_PTR(-EACCES);
Al Viro6a96ba52011-03-07 23:49:20 -05002483 }
Al Viro5a202bc2011-03-08 14:17:44 -05002484 /*
2485 * See if the low-level filesystem might want
2486 * to use its own hash..
2487 */
2488 if (base->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002489 int err = base->d_op->d_hash(base, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05002490 if (err < 0)
2491 return ERR_PTR(err);
2492 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07002493
Miklos Szeredicda309d2012-03-26 12:54:21 +02002494 err = inode_permission(base->d_inode, MAY_EXEC);
2495 if (err)
2496 return ERR_PTR(err);
2497
Al Viro72bd8662012-06-10 17:17:17 -04002498 return __lookup_hash(&this, base, 0);
James Morris057f6c02007-04-26 00:12:05 -07002499}
Al Viro4d359502014-03-14 12:20:17 -04002500EXPORT_SYMBOL(lookup_one_len);
James Morris057f6c02007-04-26 00:12:05 -07002501
NeilBrownbbddca82016-01-07 16:08:20 -05002502/**
2503 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2504 * @name: pathname component to lookup
2505 * @base: base directory to lookup from
2506 * @len: maximum length @len should be interpreted to
2507 *
2508 * Note that this routine is purely a helper for filesystem usage and should
2509 * not be called by generic code.
2510 *
2511 * Unlike lookup_one_len, it should be called without the parent
2512 * i_mutex held, and will take the i_mutex itself if necessary.
2513 */
2514struct dentry *lookup_one_len_unlocked(const char *name,
2515 struct dentry *base, int len)
2516{
2517 struct qstr this;
2518 unsigned int c;
2519 int err;
Linus Torvalds20d00ee2016-07-29 12:17:52 -07002520 struct dentry *ret;
NeilBrownbbddca82016-01-07 16:08:20 -05002521
2522 this.name = name;
2523 this.len = len;
Linus Torvalds8387ff22016-06-10 07:51:30 -07002524 this.hash = full_name_hash(base, name, len);
NeilBrownbbddca82016-01-07 16:08:20 -05002525 if (!len)
2526 return ERR_PTR(-EACCES);
2527
2528 if (unlikely(name[0] == '.')) {
2529 if (len < 2 || (len == 2 && name[1] == '.'))
2530 return ERR_PTR(-EACCES);
2531 }
2532
2533 while (len--) {
2534 c = *(const unsigned char *)name++;
2535 if (c == '/' || c == '\0')
2536 return ERR_PTR(-EACCES);
2537 }
2538 /*
2539 * See if the low-level filesystem might want
2540 * to use its own hash..
2541 */
2542 if (base->d_flags & DCACHE_OP_HASH) {
2543 int err = base->d_op->d_hash(base, &this);
2544 if (err < 0)
2545 return ERR_PTR(err);
2546 }
2547
2548 err = inode_permission(base->d_inode, MAY_EXEC);
2549 if (err)
2550 return ERR_PTR(err);
2551
Linus Torvalds20d00ee2016-07-29 12:17:52 -07002552 ret = lookup_dcache(&this, base, 0);
2553 if (!ret)
2554 ret = lookup_slow(&this, base, 0);
2555 return ret;
NeilBrownbbddca82016-01-07 16:08:20 -05002556}
2557EXPORT_SYMBOL(lookup_one_len_unlocked);
2558
Eric W. Biedermaneedf2652016-06-02 10:29:47 -05002559#ifdef CONFIG_UNIX98_PTYS
2560int path_pts(struct path *path)
2561{
2562 /* Find something mounted on "pts" in the same directory as
2563 * the input path.
2564 */
2565 struct dentry *child, *parent;
2566 struct qstr this;
2567 int ret;
2568
2569 ret = path_parent_directory(path);
2570 if (ret)
2571 return ret;
2572
2573 parent = path->dentry;
2574 this.name = "pts";
2575 this.len = 3;
2576 child = d_hash_and_lookup(parent, &this);
2577 if (!child)
2578 return -ENOENT;
2579
2580 path->dentry = child;
2581 dput(parent);
2582 follow_mount(path);
2583 return 0;
2584}
2585#endif
2586
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002587int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2588 struct path *path, int *empty)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589{
Al Viroabc9f5b2015-05-12 16:53:42 -04002590 return filename_lookup(dfd, getname_flags(name, flags, empty),
2591 flags, path, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592}
Al Virob853a162015-05-13 09:12:02 -04002593EXPORT_SYMBOL(user_path_at_empty);
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002594
Jeff Layton80334262013-07-26 06:23:25 -04002595/**
Al Viro197df042013-09-08 14:03:27 -04002596 * mountpoint_last - look up last component for umount
Jeff Layton80334262013-07-26 06:23:25 -04002597 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
Jeff Layton80334262013-07-26 06:23:25 -04002598 *
2599 * This is a special lookup_last function just for umount. In this case, we
2600 * need to resolve the path without doing any revalidation.
2601 *
2602 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2603 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2604 * in almost all cases, this lookup will be served out of the dcache. The only
2605 * cases where it won't are if nd->last refers to a symlink or the path is
2606 * bogus and it doesn't exist.
2607 *
2608 * Returns:
2609 * -error: if there was an error during lookup. This includes -ENOENT if the
Al Viroba8f4612016-11-14 00:40:33 -05002610 * lookup found a negative dentry.
Jeff Layton80334262013-07-26 06:23:25 -04002611 *
Al Viroba8f4612016-11-14 00:40:33 -05002612 * 0: if we successfully resolved nd->last and found it to not to be a
2613 * symlink that needs to be followed.
Jeff Layton80334262013-07-26 06:23:25 -04002614 *
2615 * 1: if we successfully resolved nd->last and found it to be a symlink
Al Viroba8f4612016-11-14 00:40:33 -05002616 * that needs to be followed.
Jeff Layton80334262013-07-26 06:23:25 -04002617 */
2618static int
Al Viroba8f4612016-11-14 00:40:33 -05002619mountpoint_last(struct nameidata *nd)
Jeff Layton80334262013-07-26 06:23:25 -04002620{
2621 int error = 0;
Jeff Layton80334262013-07-26 06:23:25 -04002622 struct dentry *dir = nd->path.dentry;
Al Viroba8f4612016-11-14 00:40:33 -05002623 struct path path;
Jeff Layton80334262013-07-26 06:23:25 -04002624
Al Viro35759522013-09-08 13:41:33 -04002625 /* If we're in rcuwalk, drop out of it to handle last component */
2626 if (nd->flags & LOOKUP_RCU) {
Al Viro4675ac32017-01-09 22:29:15 -05002627 if (unlazy_walk(nd))
Al Virodeb106c2015-05-08 18:05:21 -04002628 return -ECHILD;
Jeff Layton80334262013-07-26 06:23:25 -04002629 }
2630
2631 nd->flags &= ~LOOKUP_PARENT;
2632
2633 if (unlikely(nd->last_type != LAST_NORM)) {
2634 error = handle_dots(nd, nd->last_type);
Al Viro35759522013-09-08 13:41:33 -04002635 if (error)
Al Virodeb106c2015-05-08 18:05:21 -04002636 return error;
Al Viroba8f4612016-11-14 00:40:33 -05002637 path.dentry = dget(nd->path.dentry);
Al Viro949a8522016-03-06 14:20:52 -05002638 } else {
Al Viroba8f4612016-11-14 00:40:33 -05002639 path.dentry = d_lookup(dir, &nd->last);
2640 if (!path.dentry) {
Al Viro949a8522016-03-06 14:20:52 -05002641 /*
2642 * No cached dentry. Mounted dentries are pinned in the
2643 * cache, so that means that this dentry is probably
2644 * a symlink or the path doesn't actually point
2645 * to a mounted dentry.
2646 */
Al Viroba8f4612016-11-14 00:40:33 -05002647 path.dentry = lookup_slow(&nd->last, dir,
Al Viro949a8522016-03-06 14:20:52 -05002648 nd->flags | LOOKUP_NO_REVAL);
Al Viroba8f4612016-11-14 00:40:33 -05002649 if (IS_ERR(path.dentry))
2650 return PTR_ERR(path.dentry);
Dave Jonesbcceeeb2013-09-10 17:04:25 -04002651 }
Jeff Layton80334262013-07-26 06:23:25 -04002652 }
Al Viroba8f4612016-11-14 00:40:33 -05002653 if (d_is_negative(path.dentry)) {
2654 dput(path.dentry);
Al Virodeb106c2015-05-08 18:05:21 -04002655 return -ENOENT;
Jeff Layton80334262013-07-26 06:23:25 -04002656 }
Al Viroba8f4612016-11-14 00:40:33 -05002657 path.mnt = nd->path.mnt;
Al Viro8f64fb12016-11-14 01:50:26 -05002658 return step_into(nd, &path, 0, d_backing_inode(path.dentry), 0);
Jeff Layton80334262013-07-26 06:23:25 -04002659}
2660
2661/**
Al Viro197df042013-09-08 14:03:27 -04002662 * path_mountpoint - look up a path to be umounted
Masanari Iida2a78b8572015-09-09 15:39:23 -07002663 * @nd: lookup context
Jeff Layton80334262013-07-26 06:23:25 -04002664 * @flags: lookup flags
Al Viroc8a53ee2015-05-12 18:43:07 -04002665 * @path: pointer to container for result
Jeff Layton80334262013-07-26 06:23:25 -04002666 *
2667 * Look up the given name, but don't attempt to revalidate the last component.
Randy Dunlap606d6fe2013-10-19 14:56:55 -07002668 * Returns 0 and "path" will be valid on success; Returns error otherwise.
Jeff Layton80334262013-07-26 06:23:25 -04002669 */
2670static int
Al Viroc8a53ee2015-05-12 18:43:07 -04002671path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
Jeff Layton80334262013-07-26 06:23:25 -04002672{
Al Viroc8a53ee2015-05-12 18:43:07 -04002673 const char *s = path_init(nd, flags);
Al Viro368ee9b2015-05-08 17:19:59 -04002674 int err;
2675 if (IS_ERR(s))
2676 return PTR_ERR(s);
Al Viro3bdba282015-05-08 17:37:07 -04002677 while (!(err = link_path_walk(s, nd)) &&
Al Viroba8f4612016-11-14 00:40:33 -05002678 (err = mountpoint_last(nd)) > 0) {
Al Viro3bdba282015-05-08 17:37:07 -04002679 s = trailing_symlink(nd);
2680 if (IS_ERR(s)) {
2681 err = PTR_ERR(s);
Jeff Layton80334262013-07-26 06:23:25 -04002682 break;
Al Viro3bdba282015-05-08 17:37:07 -04002683 }
Jeff Layton80334262013-07-26 06:23:25 -04002684 }
Al Viroba8f4612016-11-14 00:40:33 -05002685 if (!err) {
2686 *path = nd->path;
2687 nd->path.mnt = NULL;
2688 nd->path.dentry = NULL;
2689 follow_mount(path);
2690 }
Al Virodeb106c2015-05-08 18:05:21 -04002691 terminate_walk(nd);
Jeff Layton80334262013-07-26 06:23:25 -04002692 return err;
2693}
2694
Al Viro2d864652013-09-08 20:18:44 -04002695static int
Al Viro668696d2015-02-22 19:44:00 -05002696filename_mountpoint(int dfd, struct filename *name, struct path *path,
Al Viro2d864652013-09-08 20:18:44 -04002697 unsigned int flags)
2698{
Al Viro9883d182015-05-13 07:28:08 -04002699 struct nameidata nd;
Al Virocbaab2d2015-01-22 02:49:00 -05002700 int error;
Al Viro668696d2015-02-22 19:44:00 -05002701 if (IS_ERR(name))
2702 return PTR_ERR(name);
Al Viro9883d182015-05-13 07:28:08 -04002703 set_nameidata(&nd, dfd, name);
Al Viroc8a53ee2015-05-12 18:43:07 -04002704 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
Al Viro2d864652013-09-08 20:18:44 -04002705 if (unlikely(error == -ECHILD))
Al Viroc8a53ee2015-05-12 18:43:07 -04002706 error = path_mountpoint(&nd, flags, path);
Al Viro2d864652013-09-08 20:18:44 -04002707 if (unlikely(error == -ESTALE))
Al Viroc8a53ee2015-05-12 18:43:07 -04002708 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
Al Viro2d864652013-09-08 20:18:44 -04002709 if (likely(!error))
Al Viro668696d2015-02-22 19:44:00 -05002710 audit_inode(name, path->dentry, 0);
Al Viro9883d182015-05-13 07:28:08 -04002711 restore_nameidata();
Al Viro668696d2015-02-22 19:44:00 -05002712 putname(name);
Al Viro2d864652013-09-08 20:18:44 -04002713 return error;
2714}
2715
Jeff Layton80334262013-07-26 06:23:25 -04002716/**
Al Viro197df042013-09-08 14:03:27 -04002717 * user_path_mountpoint_at - lookup a path from userland in order to umount it
Jeff Layton80334262013-07-26 06:23:25 -04002718 * @dfd: directory file descriptor
2719 * @name: pathname from userland
2720 * @flags: lookup flags
2721 * @path: pointer to container to hold result
2722 *
2723 * A umount is a special case for path walking. We're not actually interested
2724 * in the inode in this situation, and ESTALE errors can be a problem. We
2725 * simply want track down the dentry and vfsmount attached at the mountpoint
2726 * and avoid revalidating the last component.
2727 *
2728 * Returns 0 and populates "path" on success.
2729 */
2730int
Al Viro197df042013-09-08 14:03:27 -04002731user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
Jeff Layton80334262013-07-26 06:23:25 -04002732 struct path *path)
2733{
Al Virocbaab2d2015-01-22 02:49:00 -05002734 return filename_mountpoint(dfd, getname(name), path, flags);
Jeff Layton80334262013-07-26 06:23:25 -04002735}
2736
Al Viro2d864652013-09-08 20:18:44 -04002737int
2738kern_path_mountpoint(int dfd, const char *name, struct path *path,
2739 unsigned int flags)
2740{
Al Virocbaab2d2015-01-22 02:49:00 -05002741 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
Al Viro2d864652013-09-08 20:18:44 -04002742}
2743EXPORT_SYMBOL(kern_path_mountpoint);
2744
Miklos Szeredicbdf35b2014-10-24 00:14:36 +02002745int __check_sticky(struct inode *dir, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746{
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002747 kuid_t fsuid = current_fsuid();
David Howellsda9592e2008-11-14 10:39:05 +11002748
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002749 if (uid_eq(inode->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750 return 0;
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002751 if (uid_eq(dir->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002752 return 0;
Andy Lutomirski23adbe12014-06-10 12:45:42 -07002753 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002754}
Miklos Szeredicbdf35b2014-10-24 00:14:36 +02002755EXPORT_SYMBOL(__check_sticky);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002756
2757/*
2758 * Check whether we can remove a link victim from directory dir, check
2759 * whether the type of victim is right.
2760 * 1. We can't do it if dir is read-only (done in permission())
2761 * 2. We should have write and exec permissions on dir
2762 * 3. We can't remove anything from append-only dir
2763 * 4. We can't do anything with immutable dir (done in permission())
2764 * 5. If the sticky bit on dir is set we should either
2765 * a. be owner of dir, or
2766 * b. be owner of victim, or
2767 * c. have CAP_FOWNER capability
2768 * 6. If the victim is append-only or immutable we can't do antyhing with
2769 * links pointing to it.
Eric W. Biederman0bd23d092016-06-29 14:54:46 -05002770 * 7. If the victim has an unknown uid or gid we can't change the inode.
2771 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2772 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2773 * 10. We can't remove a root or mountpoint.
2774 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
Linus Torvalds1da177e2005-04-16 15:20:36 -07002775 * nfs_async_unlink().
2776 */
David Howellsb18825a2013-09-12 19:22:53 +01002777static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002778{
David Howells63afdfc2015-05-06 15:59:00 +01002779 struct inode *inode = d_backing_inode(victim);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780 int error;
2781
David Howellsb18825a2013-09-12 19:22:53 +01002782 if (d_is_negative(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002783 return -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01002784 BUG_ON(!inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785
2786 BUG_ON(victim->d_parent->d_inode != dir);
Jeff Layton4fa6b5e2012-10-10 15:25:25 -04002787 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788
Al Virof419a2e2008-07-22 00:07:17 -04002789 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002790 if (error)
2791 return error;
2792 if (IS_APPEND(dir))
2793 return -EPERM;
David Howellsb18825a2013-09-12 19:22:53 +01002794
2795 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
Eric W. Biederman0bd23d092016-06-29 14:54:46 -05002796 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797 return -EPERM;
2798 if (isdir) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002799 if (!d_is_dir(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800 return -ENOTDIR;
2801 if (IS_ROOT(victim))
2802 return -EBUSY;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002803 } else if (d_is_dir(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002804 return -EISDIR;
2805 if (IS_DEADDIR(dir))
2806 return -ENOENT;
2807 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2808 return -EBUSY;
2809 return 0;
2810}
2811
2812/* Check whether we can create an object with dentry child in directory
2813 * dir.
2814 * 1. We can't do it if child already exists (open has special treatment for
2815 * this case, but since we are inlined it's OK)
2816 * 2. We can't do it if dir is read-only (done in permission())
Eric W. Biederman036d5232016-07-01 12:52:06 -05002817 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2818 * 4. We should have write and exec permissions on dir
2819 * 5. We can't do it if dir is immutable (done in permission())
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02002821static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822{
Eric W. Biederman036d5232016-07-01 12:52:06 -05002823 struct user_namespace *s_user_ns;
Jeff Layton14e972b2013-05-08 10:25:58 -04002824 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002825 if (child->d_inode)
2826 return -EEXIST;
2827 if (IS_DEADDIR(dir))
2828 return -ENOENT;
Eric W. Biederman036d5232016-07-01 12:52:06 -05002829 s_user_ns = dir->i_sb->s_user_ns;
2830 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2831 !kgid_has_mapping(s_user_ns, current_fsgid()))
2832 return -EOVERFLOW;
Al Virof419a2e2008-07-22 00:07:17 -04002833 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002834}
2835
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836/*
2837 * p1 and p2 should be directories on the same fs.
2838 */
2839struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2840{
2841 struct dentry *p;
2842
2843 if (p1 == p2) {
Al Viro59551022016-01-22 15:40:57 -05002844 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845 return NULL;
2846 }
2847
Al Virofc640052016-04-10 01:33:30 -04002848 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002849
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002850 p = d_ancestor(p2, p1);
2851 if (p) {
Al Viro59551022016-01-22 15:40:57 -05002852 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2853 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002854 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855 }
2856
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002857 p = d_ancestor(p1, p2);
2858 if (p) {
Al Viro59551022016-01-22 15:40:57 -05002859 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2860 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002861 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862 }
2863
Al Viro59551022016-01-22 15:40:57 -05002864 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2865 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002866 return NULL;
2867}
Al Viro4d359502014-03-14 12:20:17 -04002868EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002869
2870void unlock_rename(struct dentry *p1, struct dentry *p2)
2871{
Al Viro59551022016-01-22 15:40:57 -05002872 inode_unlock(p1->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002873 if (p1 != p2) {
Al Viro59551022016-01-22 15:40:57 -05002874 inode_unlock(p2->d_inode);
Al Virofc640052016-04-10 01:33:30 -04002875 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002876 }
2877}
Al Viro4d359502014-03-14 12:20:17 -04002878EXPORT_SYMBOL(unlock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879
Al Viro4acdaf22011-07-26 01:42:34 -04002880int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viro312b63f2012-06-10 18:09:36 -04002881 bool want_excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002882{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002883 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002884 if (error)
2885 return error;
2886
Al Viroacfa4382008-12-04 10:06:33 -05002887 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002888 return -EACCES; /* shouldn't it be ENOSYS? */
2889 mode &= S_IALLUGO;
2890 mode |= S_IFREG;
2891 error = security_inode_create(dir, dentry, mode);
2892 if (error)
2893 return error;
Al Viro312b63f2012-06-10 18:09:36 -04002894 error = dir->i_op->create(dir, dentry, mode, want_excl);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002895 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002896 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002897 return error;
2898}
Al Viro4d359502014-03-14 12:20:17 -04002899EXPORT_SYMBOL(vfs_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002900
Eric W. Biedermana2982cc2016-06-09 15:34:02 -05002901bool may_open_dev(const struct path *path)
2902{
2903 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2904 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2905}
2906
Al Virof0bb5aa2016-11-20 20:27:12 -05002907static int may_open(const struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002908{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002909 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002910 struct inode *inode = dentry->d_inode;
2911 int error;
2912
2913 if (!inode)
2914 return -ENOENT;
2915
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002916 switch (inode->i_mode & S_IFMT) {
2917 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002918 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002919 case S_IFDIR:
2920 if (acc_mode & MAY_WRITE)
2921 return -EISDIR;
2922 break;
2923 case S_IFBLK:
2924 case S_IFCHR:
Eric W. Biedermana2982cc2016-06-09 15:34:02 -05002925 if (!may_open_dev(path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002926 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002927 /*FALLTHRU*/
2928 case S_IFIFO:
2929 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002930 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002931 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002932 }
Dave Hansenb41572e2007-10-16 23:31:14 -07002933
Al Viro62fb4a12015-12-26 22:33:24 -05002934 error = inode_permission(inode, MAY_OPEN | acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07002935 if (error)
2936 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05002937
Linus Torvalds1da177e2005-04-16 15:20:36 -07002938 /*
2939 * An append-only file must be opened in append mode for writing.
2940 */
2941 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05002942 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05002943 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002944 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05002945 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002946 }
2947
2948 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002949 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05002950 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002951
J. Bruce Fieldsf3c7691e2011-09-21 10:58:13 -04002952 return 0;
Al Viro7715b522009-12-16 03:54:00 -05002953}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002954
Jeff Laytone1181ee2010-12-07 16:19:50 -05002955static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002956{
Al Virof0bb5aa2016-11-20 20:27:12 -05002957 const struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002958 struct inode *inode = path->dentry->d_inode;
2959 int error = get_write_access(inode);
2960 if (error)
2961 return error;
2962 /*
2963 * Refuse to truncate files with mandatory locks held on them.
2964 */
Jeff Laytond7a06982014-03-10 09:54:15 -04002965 error = locks_verify_locked(filp);
Al Viro7715b522009-12-16 03:54:00 -05002966 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002967 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002968 if (!error) {
2969 error = do_truncate(path->dentry, 0,
2970 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002971 filp);
Al Viro7715b522009-12-16 03:54:00 -05002972 }
2973 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002974 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002975}
2976
Dave Hansend57999e2008-02-15 14:37:27 -08002977static inline int open_to_namei_flags(int flag)
2978{
Al Viro8a5e9292011-06-25 19:15:54 -04002979 if ((flag & O_ACCMODE) == 3)
2980 flag--;
Dave Hansend57999e2008-02-15 14:37:27 -08002981 return flag;
2982}
2983
Al Virod3607752016-03-25 15:21:09 -04002984static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002985{
Seth Forshee1328c722017-01-26 14:33:46 -06002986 struct user_namespace *s_user_ns;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002987 int error = security_path_mknod(dir, dentry, mode, 0);
2988 if (error)
2989 return error;
2990
Seth Forshee1328c722017-01-26 14:33:46 -06002991 s_user_ns = dir->dentry->d_sb->s_user_ns;
2992 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2993 !kgid_has_mapping(s_user_ns, current_fsgid()))
2994 return -EOVERFLOW;
2995
Miklos Szeredid18e9002012-06-05 15:10:17 +02002996 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2997 if (error)
2998 return error;
2999
3000 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3001}
3002
David Howells1acf0af2012-06-14 16:13:46 +01003003/*
3004 * Attempt to atomically look up, create and open a file from a negative
3005 * dentry.
3006 *
3007 * Returns 0 if successful. The file will have been created and attached to
3008 * @file by the filesystem calling finish_open().
3009 *
3010 * Returns 1 if the file was looked up only or didn't need creating. The
3011 * caller will need to perform the open themselves. @path will have been
3012 * updated to point to the new dentry. This may be negative.
3013 *
3014 * Returns an error code otherwise.
3015 */
Al Viro2675a4e2012-06-22 12:41:10 +04003016static int atomic_open(struct nameidata *nd, struct dentry *dentry,
3017 struct path *path, struct file *file,
3018 const struct open_flags *op,
Al Viro1643b432016-04-27 19:14:10 -04003019 int open_flag, umode_t mode,
Al Viro2675a4e2012-06-22 12:41:10 +04003020 int *opened)
Miklos Szeredid18e9002012-06-05 15:10:17 +02003021{
Miklos Szeredid18e9002012-06-05 15:10:17 +02003022 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003023 struct inode *dir = nd->path.dentry->d_inode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003024 int error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003025
Al Viro384f26e2016-04-28 02:03:55 -04003026 if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */
Miklos Szeredid18e9002012-06-05 15:10:17 +02003027 open_flag &= ~O_TRUNC;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003028
Miklos Szeredid18e9002012-06-05 15:10:17 +02003029 if (nd->flags & LOOKUP_DIRECTORY)
3030 open_flag |= O_DIRECTORY;
3031
Al Viro30d90492012-06-22 12:40:19 +04003032 file->f_path.dentry = DENTRY_NOT_SET;
3033 file->f_path.mnt = nd->path.mnt;
Al Viro0fb1ea02016-04-27 14:13:10 -04003034 error = dir->i_op->atomic_open(dir, dentry, file,
3035 open_to_namei_flags(open_flag),
3036 mode, opened);
Al Viro6fbd0712016-04-28 11:50:59 -04003037 d_lookup_done(dentry);
Al Viro384f26e2016-04-28 02:03:55 -04003038 if (!error) {
3039 /*
3040 * We didn't have the inode before the open, so check open
3041 * permission here.
3042 */
3043 int acc_mode = op->acc_mode;
3044 if (*opened & FILE_CREATED) {
3045 WARN_ON(!(open_flag & O_CREAT));
3046 fsnotify_create(dir, dentry);
3047 acc_mode = 0;
3048 }
3049 error = may_open(&file->f_path, acc_mode, open_flag);
3050 if (WARN_ON(error > 0))
3051 error = -EINVAL;
3052 } else if (error > 0) {
Al Viro30d90492012-06-22 12:40:19 +04003053 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
Al Viro2675a4e2012-06-22 12:41:10 +04003054 error = -EIO;
Al Viro03da6332013-09-16 19:22:33 -04003055 } else {
Al Viro384f26e2016-04-28 02:03:55 -04003056 if (file->f_path.dentry) {
3057 dput(dentry);
3058 dentry = file->f_path.dentry;
Al Viro03da6332013-09-16 19:22:33 -04003059 }
Al Viro384f26e2016-04-28 02:03:55 -04003060 if (*opened & FILE_CREATED)
3061 fsnotify_create(dir, dentry);
Al Viroa01e7182016-06-07 21:53:51 -04003062 if (unlikely(d_is_negative(dentry))) {
3063 error = -ENOENT;
3064 } else {
3065 path->dentry = dentry;
3066 path->mnt = nd->path.mnt;
3067 return 1;
3068 }
Sage Weil62b2ce92012-08-15 13:30:12 -07003069 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02003070 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02003071 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04003072 return error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003073}
3074
Nick Piggin31e6b012011-01-07 17:49:52 +11003075/*
David Howells1acf0af2012-06-14 16:13:46 +01003076 * Look up and maybe create and open the last component.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003077 *
3078 * Must be called with i_mutex held on parent.
3079 *
David Howells1acf0af2012-06-14 16:13:46 +01003080 * Returns 0 if the file was successfully atomically created (if necessary) and
3081 * opened. In this case the file will be returned attached to @file.
3082 *
3083 * Returns 1 if the file was not completely opened at this time, though lookups
3084 * and creations will have been performed and the dentry returned in @path will
3085 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
3086 * specified then a negative dentry may be returned.
3087 *
3088 * An error code is returned otherwise.
3089 *
3090 * FILE_CREATE will be set in @*opened if the dentry was created and will be
3091 * cleared otherwise prior to returning.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003092 */
Al Viro2675a4e2012-06-22 12:41:10 +04003093static int lookup_open(struct nameidata *nd, struct path *path,
3094 struct file *file,
3095 const struct open_flags *op,
Al Viro64894cf2012-07-31 00:53:35 +04003096 bool got_write, int *opened)
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003097{
3098 struct dentry *dir = nd->path.dentry;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02003099 struct inode *dir_inode = dir->d_inode;
Al Viro1643b432016-04-27 19:14:10 -04003100 int open_flag = op->open_flag;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003101 struct dentry *dentry;
Al Viro1643b432016-04-27 19:14:10 -04003102 int error, create_error = 0;
Al Viro1643b432016-04-27 19:14:10 -04003103 umode_t mode = op->mode;
Al Viro6fbd0712016-04-28 11:50:59 -04003104 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003105
Al Viroce8644f2016-04-26 14:17:56 -04003106 if (unlikely(IS_DEADDIR(dir_inode)))
3107 return -ENOENT;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003108
Al Viro47237682012-06-10 05:01:45 -04003109 *opened &= ~FILE_CREATED;
Al Viro6fbd0712016-04-28 11:50:59 -04003110 dentry = d_lookup(dir, &nd->last);
3111 for (;;) {
3112 if (!dentry) {
3113 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3114 if (IS_ERR(dentry))
3115 return PTR_ERR(dentry);
3116 }
3117 if (d_in_lookup(dentry))
3118 break;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003119
Al Viro6fbd0712016-04-28 11:50:59 -04003120 error = d_revalidate(dentry, nd->flags);
3121 if (likely(error > 0))
3122 break;
3123 if (error)
3124 goto out_dput;
3125 d_invalidate(dentry);
3126 dput(dentry);
3127 dentry = NULL;
3128 }
3129 if (dentry->d_inode) {
Al Viro6c51e512016-03-05 20:09:32 -05003130 /* Cached positive dentry: will open in f_op->open */
Miklos Szeredid18e9002012-06-05 15:10:17 +02003131 goto out_no_open;
Al Viro6c51e512016-03-05 20:09:32 -05003132 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02003133
Al Viro1643b432016-04-27 19:14:10 -04003134 /*
3135 * Checking write permission is tricky, bacuse we don't know if we are
3136 * going to actually need it: O_CREAT opens should work as long as the
3137 * file exists. But checking existence breaks atomicity. The trick is
3138 * to check access and if not granted clear O_CREAT from the flags.
3139 *
3140 * Another problem is returing the "right" error value (e.g. for an
3141 * O_EXCL open we want to return EEXIST not EROFS).
3142 */
3143 if (open_flag & O_CREAT) {
3144 if (!IS_POSIXACL(dir->d_inode))
3145 mode &= ~current_umask();
3146 if (unlikely(!got_write)) {
3147 create_error = -EROFS;
3148 open_flag &= ~O_CREAT;
3149 if (open_flag & (O_EXCL | O_TRUNC))
3150 goto no_open;
3151 /* No side effects, safe to clear O_CREAT */
3152 } else {
3153 create_error = may_o_create(&nd->path, dentry, mode);
3154 if (create_error) {
3155 open_flag &= ~O_CREAT;
3156 if (open_flag & O_EXCL)
3157 goto no_open;
3158 }
3159 }
3160 } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3161 unlikely(!got_write)) {
3162 /*
3163 * No O_CREATE -> atomicity not a requirement -> fall
3164 * back to lookup + open
3165 */
3166 goto no_open;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003167 }
3168
Al Viro1643b432016-04-27 19:14:10 -04003169 if (dir_inode->i_op->atomic_open) {
3170 error = atomic_open(nd, dentry, path, file, op, open_flag,
3171 mode, opened);
3172 if (unlikely(error == -ENOENT) && create_error)
3173 error = create_error;
3174 return error;
3175 }
Miklos Szeredi54ef4872012-06-05 15:10:16 +02003176
Al Viro1643b432016-04-27 19:14:10 -04003177no_open:
Al Viro6fbd0712016-04-28 11:50:59 -04003178 if (d_in_lookup(dentry)) {
Al Viro12fa5e22016-04-28 11:19:43 -04003179 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3180 nd->flags);
Al Viro6fbd0712016-04-28 11:50:59 -04003181 d_lookup_done(dentry);
Al Viro12fa5e22016-04-28 11:19:43 -04003182 if (unlikely(res)) {
3183 if (IS_ERR(res)) {
3184 error = PTR_ERR(res);
3185 goto out_dput;
3186 }
3187 dput(dentry);
3188 dentry = res;
3189 }
Miklos Szeredi54ef4872012-06-05 15:10:16 +02003190 }
3191
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003192 /* Negative dentry, just create the file */
Al Viro1643b432016-04-27 19:14:10 -04003193 if (!dentry->d_inode && (open_flag & O_CREAT)) {
Al Viro47237682012-06-10 05:01:45 -04003194 *opened |= FILE_CREATED;
Al Viroce8644f2016-04-26 14:17:56 -04003195 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
Al Viroce8644f2016-04-26 14:17:56 -04003196 if (!dir_inode->i_op->create) {
3197 error = -EACCES;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003198 goto out_dput;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003199 }
Al Viroce8644f2016-04-26 14:17:56 -04003200 error = dir_inode->i_op->create(dir_inode, dentry, mode,
Al Viro1643b432016-04-27 19:14:10 -04003201 open_flag & O_EXCL);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003202 if (error)
3203 goto out_dput;
Al Viroce8644f2016-04-26 14:17:56 -04003204 fsnotify_create(dir_inode, dentry);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003205 }
Al Viro1643b432016-04-27 19:14:10 -04003206 if (unlikely(create_error) && !dentry->d_inode) {
3207 error = create_error;
3208 goto out_dput;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003209 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02003210out_no_open:
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003211 path->dentry = dentry;
3212 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04003213 return 1;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003214
3215out_dput:
3216 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04003217 return error;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003218}
3219
3220/*
Al Virofe2d35f2011-03-05 22:58:25 -05003221 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11003222 */
Al Viro896475d2015-04-22 18:02:17 -04003223static int do_last(struct nameidata *nd,
Al Viro2675a4e2012-06-22 12:41:10 +04003224 struct file *file, const struct open_flags *op,
Al Viro76ae2a52015-05-12 18:44:32 -04003225 int *opened)
Al Virofb1cc552009-12-24 01:58:28 -05003226{
Al Viroa1e28032009-12-24 02:12:06 -05003227 struct dentry *dir = nd->path.dentry;
Al Viroca344a892011-03-09 00:36:45 -05003228 int open_flag = op->open_flag;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003229 bool will_truncate = (open_flag & O_TRUNC) != 0;
Al Viro64894cf2012-07-31 00:53:35 +04003230 bool got_write = false;
Al Virobcda7652011-03-13 16:42:14 -04003231 int acc_mode = op->acc_mode;
Al Viro254cf582015-05-05 09:40:46 -04003232 unsigned seq;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02003233 struct inode *inode;
Al Viro896475d2015-04-22 18:02:17 -04003234 struct path path;
Al Viro16c2cd72011-02-22 15:50:10 -05003235 int error;
Al Virofb1cc552009-12-24 01:58:28 -05003236
Al Viroc3e380b2011-02-23 13:39:45 -05003237 nd->flags &= ~LOOKUP_PARENT;
3238 nd->flags |= op->intent;
3239
Al Virobc77daa2013-06-06 09:12:33 -04003240 if (nd->last_type != LAST_NORM) {
Al Virofe2d35f2011-03-05 22:58:25 -05003241 error = handle_dots(nd, nd->last_type);
Al Virodeb106c2015-05-08 18:05:21 -04003242 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04003243 return error;
Miklos Szeredie83db162012-06-05 15:10:29 +02003244 goto finish_open;
Al Viro1f36f772009-12-26 10:56:19 -05003245 }
Al Viro67ee3ad2009-12-26 07:01:01 -05003246
Al Viroca344a892011-03-09 00:36:45 -05003247 if (!(open_flag & O_CREAT)) {
Al Virofe2d35f2011-03-05 22:58:25 -05003248 if (nd->last.name[nd->last.len])
3249 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3250 /* we _can_ be in RCU mode here */
Al Viro254cf582015-05-05 09:40:46 -04003251 error = lookup_fast(nd, &path, &inode, &seq);
Al Viroe9742b52016-03-05 22:04:59 -05003252 if (likely(error > 0))
Miklos Szeredi71574862012-06-05 15:10:14 +02003253 goto finish_lookup;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02003254
Miklos Szeredi71574862012-06-05 15:10:14 +02003255 if (error < 0)
Al Virodeb106c2015-05-08 18:05:21 -04003256 return error;
Miklos Szeredi37d7fff2012-06-05 15:10:12 +02003257
Miklos Szeredi71574862012-06-05 15:10:14 +02003258 BUG_ON(nd->inode != dir->d_inode);
Al Viro6583fe22016-03-05 18:14:03 -05003259 BUG_ON(nd->flags & LOOKUP_RCU);
Miklos Szeredib6183df2012-06-05 15:10:13 +02003260 } else {
3261 /* create side of things */
3262 /*
3263 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3264 * has been cleared when we got to the last component we are
3265 * about to look up
3266 */
3267 error = complete_walk(nd);
Al Viroe8bb73d2015-05-08 16:28:42 -04003268 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003269 return error;
Miklos Szeredib6183df2012-06-05 15:10:13 +02003270
Al Viro76ae2a52015-05-12 18:44:32 -04003271 audit_inode(nd->name, dir, LOOKUP_PARENT);
Miklos Szeredib6183df2012-06-05 15:10:13 +02003272 /* trailing slashes? */
Al Virodeb106c2015-05-08 18:05:21 -04003273 if (unlikely(nd->last.name[nd->last.len]))
3274 return -EISDIR;
Al Virofe2d35f2011-03-05 22:58:25 -05003275 }
3276
Al Viro9cf843e2016-04-28 19:35:16 -04003277 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
Al Viro64894cf2012-07-31 00:53:35 +04003278 error = mnt_want_write(nd->path.mnt);
3279 if (!error)
3280 got_write = true;
3281 /*
3282 * do _not_ fail yet - we might not need that or fail with
3283 * a different error; let lookup_open() decide; we'll be
3284 * dropping this one anyway.
3285 */
3286 }
Al Viro9cf843e2016-04-28 19:35:16 -04003287 if (open_flag & O_CREAT)
3288 inode_lock(dir->d_inode);
3289 else
3290 inode_lock_shared(dir->d_inode);
Al Viro896475d2015-04-22 18:02:17 -04003291 error = lookup_open(nd, &path, file, op, got_write, opened);
Al Viro9cf843e2016-04-28 19:35:16 -04003292 if (open_flag & O_CREAT)
3293 inode_unlock(dir->d_inode);
3294 else
3295 inode_unlock_shared(dir->d_inode);
Al Viroa1e28032009-12-24 02:12:06 -05003296
Al Viro2675a4e2012-06-22 12:41:10 +04003297 if (error <= 0) {
3298 if (error)
Miklos Szeredid18e9002012-06-05 15:10:17 +02003299 goto out;
3300
Al Viro47237682012-06-10 05:01:45 -04003301 if ((*opened & FILE_CREATED) ||
Al Viro496ad9a2013-01-23 17:07:38 -05003302 !S_ISREG(file_inode(file)->i_mode))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003303 will_truncate = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003304
Al Viro76ae2a52015-05-12 18:44:32 -04003305 audit_inode(nd->name, file->f_path.dentry, 0);
Miklos Szeredid18e9002012-06-05 15:10:17 +02003306 goto opened;
3307 }
Al Virofb1cc552009-12-24 01:58:28 -05003308
Al Viro47237682012-06-10 05:01:45 -04003309 if (*opened & FILE_CREATED) {
Al Viro9b44f1b2011-03-09 00:17:27 -05003310 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05003311 open_flag &= ~O_TRUNC;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003312 will_truncate = false;
Al Viro62fb4a12015-12-26 22:33:24 -05003313 acc_mode = 0;
Al Viro896475d2015-04-22 18:02:17 -04003314 path_to_nameidata(&path, nd);
Miklos Szeredie83db162012-06-05 15:10:29 +02003315 goto finish_open_created;
Al Virofb1cc552009-12-24 01:58:28 -05003316 }
3317
3318 /*
Miklos Szeredid18e9002012-06-05 15:10:17 +02003319 * If atomic_open() acquired write access it is dropped now due to
3320 * possible mount and symlink following (this might be optimized away if
3321 * necessary...)
3322 */
Al Viro64894cf2012-07-31 00:53:35 +04003323 if (got_write) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02003324 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04003325 got_write = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003326 }
3327
Al Viroe6ec03a2016-06-05 00:23:09 -04003328 error = follow_managed(&path, nd);
3329 if (unlikely(error < 0))
3330 return error;
3331
Al Viro6583fe22016-03-05 18:14:03 -05003332 if (unlikely(d_is_negative(path.dentry))) {
3333 path_to_nameidata(&path, nd);
3334 return -ENOENT;
3335 }
3336
3337 /*
3338 * create/update audit record if it already exists.
3339 */
3340 audit_inode(nd->name, path.dentry, 0);
3341
Al Virodeb106c2015-05-08 18:05:21 -04003342 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3343 path_to_nameidata(&path, nd);
3344 return -EEXIST;
3345 }
Al Virofb1cc552009-12-24 01:58:28 -05003346
Al Viro254cf582015-05-05 09:40:46 -04003347 seq = 0; /* out of RCU mode, so the value doesn't matter */
Al Virod4565642016-02-27 19:23:16 -05003348 inode = d_backing_inode(path.dentry);
Al Viro766c4cb2015-05-07 19:24:57 -04003349finish_lookup:
Al Viro8f64fb12016-11-14 01:50:26 -05003350 error = step_into(nd, &path, 0, inode, seq);
Al Virodeb106c2015-05-08 18:05:21 -04003351 if (unlikely(error))
Al Virod63ff282015-05-04 18:13:23 -04003352 return error;
Al Virobc77daa2013-06-06 09:12:33 -04003353finish_open:
Al Viro8f64fb12016-11-14 01:50:26 -05003354 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
Al Viroa3fbbde2011-11-07 21:21:26 +00003355 error = complete_walk(nd);
Al Virofac7d192016-06-04 11:41:49 -04003356 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003357 return error;
Al Viro76ae2a52015-05-12 18:44:32 -04003358 audit_inode(nd->name, nd->path.dentry, 0);
Al Virofb1cc552009-12-24 01:58:28 -05003359 error = -EISDIR;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003360 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04003361 goto out;
Miklos Szerediaf2f5542012-05-21 17:30:11 +02003362 error = -ENOTDIR;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003363 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04003364 goto out;
David Howells4bbcbd3b2015-03-17 22:16:40 +00003365 if (!d_is_reg(nd->path.dentry))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003366 will_truncate = false;
Al Viro6c0d46c2011-03-09 00:59:59 -05003367
Al Viro0f9d1a12011-03-09 00:13:14 -05003368 if (will_truncate) {
3369 error = mnt_want_write(nd->path.mnt);
3370 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003371 goto out;
Al Viro64894cf2012-07-31 00:53:35 +04003372 got_write = true;
Al Viro0f9d1a12011-03-09 00:13:14 -05003373 }
Miklos Szeredie83db162012-06-05 15:10:29 +02003374finish_open_created:
Al Viro6ac08702016-04-26 00:02:50 -04003375 error = may_open(&nd->path, acc_mode, open_flag);
3376 if (error)
3377 goto out;
Miklos Szeredi4aa7c632014-10-24 00:14:35 +02003378 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3379 error = vfs_open(&nd->path, file, current_cred());
Al Virofac7d192016-06-04 11:41:49 -04003380 if (error)
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003381 goto out;
Al Virofac7d192016-06-04 11:41:49 -04003382 *opened |= FILE_OPENED;
Miklos Szeredia8277b92012-06-05 15:10:32 +02003383opened:
Al Viro2675a4e2012-06-22 12:41:10 +04003384 error = open_check_o_direct(file);
Al Virofe9ec822016-04-27 03:14:20 -04003385 if (!error)
3386 error = ima_file_check(file, op->acc_mode, *opened);
3387 if (!error && will_truncate)
Al Viro2675a4e2012-06-22 12:41:10 +04003388 error = handle_truncate(file);
Al Viroca344a892011-03-09 00:36:45 -05003389out:
Al Virofe9ec822016-04-27 03:14:20 -04003390 if (unlikely(error) && (*opened & FILE_OPENED))
3391 fput(file);
Al Viroc80567c2016-02-27 19:17:33 -05003392 if (unlikely(error > 0)) {
3393 WARN_ON(1);
3394 error = -EINVAL;
3395 }
Al Viro64894cf2012-07-31 00:53:35 +04003396 if (got_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05003397 mnt_drop_write(nd->path.mnt);
Al Viro2675a4e2012-06-22 12:41:10 +04003398 return error;
Al Virofb1cc552009-12-24 01:58:28 -05003399}
3400
Amir Goldsteinaf7bd4d2017-01-17 06:34:52 +02003401struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag)
3402{
3403 static const struct qstr name = QSTR_INIT("/", 1);
3404 struct dentry *child = NULL;
3405 struct inode *dir = dentry->d_inode;
3406 struct inode *inode;
3407 int error;
3408
3409 /* we want directory to be writable */
3410 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3411 if (error)
3412 goto out_err;
3413 error = -EOPNOTSUPP;
3414 if (!dir->i_op->tmpfile)
3415 goto out_err;
3416 error = -ENOMEM;
3417 child = d_alloc(dentry, &name);
3418 if (unlikely(!child))
3419 goto out_err;
3420 error = dir->i_op->tmpfile(dir, child, mode);
3421 if (error)
3422 goto out_err;
3423 error = -ENOENT;
3424 inode = child->d_inode;
3425 if (unlikely(!inode))
3426 goto out_err;
3427 if (!(open_flag & O_EXCL)) {
3428 spin_lock(&inode->i_lock);
3429 inode->i_state |= I_LINKABLE;
3430 spin_unlock(&inode->i_lock);
3431 }
3432 return child;
3433
3434out_err:
3435 dput(child);
3436 return ERR_PTR(error);
3437}
3438EXPORT_SYMBOL(vfs_tmpfile);
3439
Al Viroc8a53ee2015-05-12 18:43:07 -04003440static int do_tmpfile(struct nameidata *nd, unsigned flags,
Al Viro60545d02013-06-07 01:20:27 -04003441 const struct open_flags *op,
3442 struct file *file, int *opened)
3443{
Al Viro625b6d12015-05-12 16:36:12 -04003444 struct dentry *child;
Al Viro625b6d12015-05-12 16:36:12 -04003445 struct path path;
Al Viroc8a53ee2015-05-12 18:43:07 -04003446 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
Al Viro60545d02013-06-07 01:20:27 -04003447 if (unlikely(error))
3448 return error;
Al Viro625b6d12015-05-12 16:36:12 -04003449 error = mnt_want_write(path.mnt);
Al Viro60545d02013-06-07 01:20:27 -04003450 if (unlikely(error))
3451 goto out;
Amir Goldsteinaf7bd4d2017-01-17 06:34:52 +02003452 child = vfs_tmpfile(path.dentry, op->mode, op->open_flag);
3453 error = PTR_ERR(child);
3454 if (unlikely(IS_ERR(child)))
Al Viro60545d02013-06-07 01:20:27 -04003455 goto out2;
Al Viro625b6d12015-05-12 16:36:12 -04003456 dput(path.dentry);
3457 path.dentry = child;
Al Viroc8a53ee2015-05-12 18:43:07 -04003458 audit_inode(nd->name, child, 0);
Eric Rannaud69a91c22014-10-30 01:51:01 -07003459 /* Don't check for other permissions, the inode was just created */
Al Viro62fb4a12015-12-26 22:33:24 -05003460 error = may_open(&path, 0, op->open_flag);
Al Viro60545d02013-06-07 01:20:27 -04003461 if (error)
3462 goto out2;
Al Viro625b6d12015-05-12 16:36:12 -04003463 file->f_path.mnt = path.mnt;
3464 error = finish_open(file, child, NULL, opened);
Al Viro60545d02013-06-07 01:20:27 -04003465 if (error)
3466 goto out2;
3467 error = open_check_o_direct(file);
Amir Goldsteinaf7bd4d2017-01-17 06:34:52 +02003468 if (error)
Al Viro60545d02013-06-07 01:20:27 -04003469 fput(file);
3470out2:
Al Viro625b6d12015-05-12 16:36:12 -04003471 mnt_drop_write(path.mnt);
Al Viro60545d02013-06-07 01:20:27 -04003472out:
Al Viro625b6d12015-05-12 16:36:12 -04003473 path_put(&path);
Al Viro60545d02013-06-07 01:20:27 -04003474 return error;
3475}
3476
Al Viro6ac08702016-04-26 00:02:50 -04003477static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3478{
3479 struct path path;
3480 int error = path_lookupat(nd, flags, &path);
3481 if (!error) {
3482 audit_inode(nd->name, path.dentry, 0);
3483 error = vfs_open(&path, file, current_cred());
3484 path_put(&path);
3485 }
3486 return error;
3487}
3488
Al Viroc8a53ee2015-05-12 18:43:07 -04003489static struct file *path_openat(struct nameidata *nd,
3490 const struct open_flags *op, unsigned flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003491{
Al Viro368ee9b2015-05-08 17:19:59 -04003492 const char *s;
Al Viro30d90492012-06-22 12:40:19 +04003493 struct file *file;
Al Viro47237682012-06-10 05:01:45 -04003494 int opened = 0;
Al Viro13aab422011-02-23 17:54:08 -05003495 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11003496
Al Viro30d90492012-06-22 12:40:19 +04003497 file = get_empty_filp();
Al Viro1afc99b2013-02-14 20:41:04 -05003498 if (IS_ERR(file))
3499 return file;
Nick Piggin31e6b012011-01-07 17:49:52 +11003500
Al Viro30d90492012-06-22 12:40:19 +04003501 file->f_flags = op->open_flag;
Nick Piggin31e6b012011-01-07 17:49:52 +11003502
Al Virobb458c62013-07-13 13:26:37 +04003503 if (unlikely(file->f_flags & __O_TMPFILE)) {
Al Viroc8a53ee2015-05-12 18:43:07 -04003504 error = do_tmpfile(nd, flags, op, file, &opened);
Al Virof15133d2015-05-08 22:53:15 -04003505 goto out2;
Al Viro60545d02013-06-07 01:20:27 -04003506 }
3507
Al Viro6ac08702016-04-26 00:02:50 -04003508 if (unlikely(file->f_flags & O_PATH)) {
3509 error = do_o_path(nd, flags, file);
3510 if (!error)
3511 opened |= FILE_OPENED;
3512 goto out2;
3513 }
3514
Al Viroc8a53ee2015-05-12 18:43:07 -04003515 s = path_init(nd, flags);
Al Viro368ee9b2015-05-08 17:19:59 -04003516 if (IS_ERR(s)) {
3517 put_filp(file);
3518 return ERR_CAST(s);
3519 }
Al Viro3bdba282015-05-08 17:37:07 -04003520 while (!(error = link_path_walk(s, nd)) &&
Al Viro76ae2a52015-05-12 18:44:32 -04003521 (error = do_last(nd, file, op, &opened)) > 0) {
Al Viro73d049a2011-03-11 12:08:24 -05003522 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
Al Viro3bdba282015-05-08 17:37:07 -04003523 s = trailing_symlink(nd);
3524 if (IS_ERR(s)) {
3525 error = PTR_ERR(s);
Al Viro2675a4e2012-06-22 12:41:10 +04003526 break;
Al Viro3bdba282015-05-08 17:37:07 -04003527 }
Al Viro806b6812009-12-26 07:16:40 -05003528 }
Al Virodeb106c2015-05-08 18:05:21 -04003529 terminate_walk(nd);
Al Virof15133d2015-05-08 22:53:15 -04003530out2:
Al Viro2675a4e2012-06-22 12:41:10 +04003531 if (!(opened & FILE_OPENED)) {
3532 BUG_ON(!error);
Al Viro30d90492012-06-22 12:40:19 +04003533 put_filp(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003534 }
Al Viro2675a4e2012-06-22 12:41:10 +04003535 if (unlikely(error)) {
3536 if (error == -EOPENSTALE) {
3537 if (flags & LOOKUP_RCU)
3538 error = -ECHILD;
3539 else
3540 error = -ESTALE;
3541 }
3542 file = ERR_PTR(error);
3543 }
3544 return file;
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
Eric W. Biederman975d6b32011-11-13 12:16:43 -08003695 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
Al Viro8208a222011-07-25 17:32:17 -04003733SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003734 unsigned, 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
Al Viro8208a222011-07-25 17:32:17 -04003777SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003778{
3779 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3780}
3781
Al Viro18bb1db2011-07-26 01:41:39 -04003782int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003783{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003784 int error = may_create(dir, dentry);
Al Viro8de52772012-02-06 12:45:27 -05003785 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003786
3787 if (error)
3788 return error;
3789
Al Viroacfa4382008-12-04 10:06:33 -05003790 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003791 return -EPERM;
3792
3793 mode &= (S_IRWXUGO|S_ISVTX);
3794 error = security_inode_mkdir(dir, dentry, mode);
3795 if (error)
3796 return error;
3797
Al Viro8de52772012-02-06 12:45:27 -05003798 if (max_links && dir->i_nlink >= max_links)
3799 return -EMLINK;
3800
Linus Torvalds1da177e2005-04-16 15:20:36 -07003801 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003802 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003803 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003804 return error;
3805}
Al Viro4d359502014-03-14 12:20:17 -04003806EXPORT_SYMBOL(vfs_mkdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003807
Al Viroa218d0f2011-11-21 14:59:34 -05003808SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003809{
Dave Hansen6902d922006-09-30 23:29:01 -07003810 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003811 struct path path;
3812 int error;
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003813 unsigned int lookup_flags = LOOKUP_DIRECTORY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003814
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003815retry:
3816 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07003817 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003818 return PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003819
Al Virodae6ad82011-06-26 11:50:15 -04003820 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003821 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003822 error = security_path_mkdir(&path, dentry, mode);
Al Viroa8104a92012-07-20 02:25:00 +04003823 if (!error)
3824 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
Al Viro921a1652012-07-20 01:15:31 +04003825 done_path_create(&path, dentry);
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003826 if (retry_estale(error, lookup_flags)) {
3827 lookup_flags |= LOOKUP_REVAL;
3828 goto retry;
3829 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003830 return error;
3831}
3832
Al Viroa218d0f2011-11-21 14:59:34 -05003833SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003834{
3835 return sys_mkdirat(AT_FDCWD, pathname, mode);
3836}
3837
Linus Torvalds1da177e2005-04-16 15:20:36 -07003838int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3839{
3840 int error = may_delete(dir, dentry, 1);
3841
3842 if (error)
3843 return error;
3844
Al Viroacfa4382008-12-04 10:06:33 -05003845 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003846 return -EPERM;
3847
Al Viro1d2ef592011-09-14 18:55:41 +01003848 dget(dentry);
Al Viro59551022016-01-22 15:40:57 -05003849 inode_lock(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003850
Sage Weil912dbc12011-05-24 13:06:11 -07003851 error = -EBUSY;
Eric W. Biederman7af13642013-10-04 19:15:13 -07003852 if (is_local_mountpoint(dentry))
Sage Weil912dbc12011-05-24 13:06:11 -07003853 goto out;
3854
3855 error = security_inode_rmdir(dir, dentry);
3856 if (error)
3857 goto out;
3858
Sage Weil3cebde22011-05-29 21:20:59 -07003859 shrink_dcache_parent(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003860 error = dir->i_op->rmdir(dir, dentry);
3861 if (error)
3862 goto out;
3863
3864 dentry->d_inode->i_flags |= S_DEAD;
3865 dont_mount(dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003866 detach_mounts(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003867
3868out:
Al Viro59551022016-01-22 15:40:57 -05003869 inode_unlock(dentry->d_inode);
Al Viro1d2ef592011-09-14 18:55:41 +01003870 dput(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003871 if (!error)
3872 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003873 return error;
3874}
Al Viro4d359502014-03-14 12:20:17 -04003875EXPORT_SYMBOL(vfs_rmdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003876
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003877static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003878{
3879 int error = 0;
Jeff Layton91a27b22012-10-10 15:25:28 -04003880 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003881 struct dentry *dentry;
Al Virof5beed72015-04-30 16:09:11 -04003882 struct path path;
3883 struct qstr last;
3884 int type;
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003885 unsigned int lookup_flags = 0;
3886retry:
Al Viroc1d4dd22016-06-05 16:38:18 -04003887 name = filename_parentat(dfd, getname(pathname), lookup_flags,
3888 &path, &last, &type);
Jeff Layton91a27b22012-10-10 15:25:28 -04003889 if (IS_ERR(name))
3890 return PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003891
Al Virof5beed72015-04-30 16:09:11 -04003892 switch (type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003893 case LAST_DOTDOT:
3894 error = -ENOTEMPTY;
3895 goto exit1;
3896 case LAST_DOT:
3897 error = -EINVAL;
3898 goto exit1;
3899 case LAST_ROOT:
3900 error = -EBUSY;
3901 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003902 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003903
Al Virof5beed72015-04-30 16:09:11 -04003904 error = mnt_want_write(path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02003905 if (error)
3906 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003907
Al Viro59551022016-01-22 15:40:57 -05003908 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
Al Virof5beed72015-04-30 16:09:11 -04003909 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003910 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003911 if (IS_ERR(dentry))
3912 goto exit2;
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003913 if (!dentry->d_inode) {
3914 error = -ENOENT;
3915 goto exit3;
3916 }
Al Virof5beed72015-04-30 16:09:11 -04003917 error = security_path_rmdir(&path, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003918 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003919 goto exit3;
Al Virof5beed72015-04-30 16:09:11 -04003920 error = vfs_rmdir(path.dentry->d_inode, dentry);
Dave Hansen06227532008-02-15 14:37:34 -08003921exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07003922 dput(dentry);
3923exit2:
Al Viro59551022016-01-22 15:40:57 -05003924 inode_unlock(path.dentry->d_inode);
Al Virof5beed72015-04-30 16:09:11 -04003925 mnt_drop_write(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003926exit1:
Al Virof5beed72015-04-30 16:09:11 -04003927 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003928 putname(name);
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003929 if (retry_estale(error, lookup_flags)) {
3930 lookup_flags |= LOOKUP_REVAL;
3931 goto retry;
3932 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003933 return error;
3934}
3935
Heiko Carstens3cdad422009-01-14 14:14:22 +01003936SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003937{
3938 return do_rmdir(AT_FDCWD, pathname);
3939}
3940
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003941/**
3942 * vfs_unlink - unlink a filesystem object
3943 * @dir: parent directory
3944 * @dentry: victim
3945 * @delegated_inode: returns victim inode, if the inode is delegated.
3946 *
3947 * The caller must hold dir->i_mutex.
3948 *
3949 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3950 * return a reference to the inode in delegated_inode. The caller
3951 * should then break the delegation on that inode and retry. Because
3952 * breaking a delegation may take a long time, the caller should drop
3953 * dir->i_mutex before doing so.
3954 *
3955 * Alternatively, a caller may pass NULL for delegated_inode. This may
3956 * be appropriate for callers that expect the underlying filesystem not
3957 * to be NFS exported.
3958 */
3959int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003960{
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003961 struct inode *target = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003962 int error = may_delete(dir, dentry, 0);
3963
3964 if (error)
3965 return error;
3966
Al Viroacfa4382008-12-04 10:06:33 -05003967 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003968 return -EPERM;
3969
Al Viro59551022016-01-22 15:40:57 -05003970 inode_lock(target);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003971 if (is_local_mountpoint(dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003972 error = -EBUSY;
3973 else {
3974 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05003975 if (!error) {
J. Bruce Fields5a146962012-08-28 07:50:40 -07003976 error = try_break_deleg(target, delegated_inode);
3977 if (error)
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003978 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003979 error = dir->i_op->unlink(dir, dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003980 if (!error) {
Al Virod83c49f2010-04-30 17:17:09 -04003981 dont_mount(dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003982 detach_mounts(dentry);
3983 }
Al Virobec10522010-03-03 14:12:08 -05003984 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003985 }
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003986out:
Al Viro59551022016-01-22 15:40:57 -05003987 inode_unlock(target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003988
3989 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3990 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003991 fsnotify_link_count(target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003992 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003993 }
Robert Love0eeca282005-07-12 17:06:03 -04003994
Linus Torvalds1da177e2005-04-16 15:20:36 -07003995 return error;
3996}
Al Viro4d359502014-03-14 12:20:17 -04003997EXPORT_SYMBOL(vfs_unlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003998
3999/*
4000 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004001 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07004002 * writeout happening, and we don't want to prevent access to the directory
4003 * while waiting on the I/O.
4004 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004005static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004006{
Al Viro2ad94ae2008-07-21 09:32:51 -04004007 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04004008 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004009 struct dentry *dentry;
Al Virof5beed72015-04-30 16:09:11 -04004010 struct path path;
4011 struct qstr last;
4012 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004013 struct inode *inode = NULL;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04004014 struct inode *delegated_inode = NULL;
Jeff Layton5d18f812012-12-20 16:38:04 -05004015 unsigned int lookup_flags = 0;
4016retry:
Al Viroc1d4dd22016-06-05 16:38:18 -04004017 name = filename_parentat(dfd, getname(pathname), lookup_flags,
4018 &path, &last, &type);
Jeff Layton91a27b22012-10-10 15:25:28 -04004019 if (IS_ERR(name))
4020 return PTR_ERR(name);
Al Viro2ad94ae2008-07-21 09:32:51 -04004021
Linus Torvalds1da177e2005-04-16 15:20:36 -07004022 error = -EISDIR;
Al Virof5beed72015-04-30 16:09:11 -04004023 if (type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004024 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09004025
Al Virof5beed72015-04-30 16:09:11 -04004026 error = mnt_want_write(path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02004027 if (error)
4028 goto exit1;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04004029retry_deleg:
Al Viro59551022016-01-22 15:40:57 -05004030 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
Al Virof5beed72015-04-30 16:09:11 -04004031 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004032 error = PTR_ERR(dentry);
4033 if (!IS_ERR(dentry)) {
4034 /* Why not before? Because we want correct error value */
Al Virof5beed72015-04-30 16:09:11 -04004035 if (last.name[last.len])
Török Edwin50338b82011-06-16 00:06:14 +03004036 goto slashes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004037 inode = dentry->d_inode;
David Howellsb18825a2013-09-12 19:22:53 +01004038 if (d_is_negative(dentry))
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04004039 goto slashes;
4040 ihold(inode);
Al Virof5beed72015-04-30 16:09:11 -04004041 error = security_path_unlink(&path, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004042 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02004043 goto exit2;
Al Virof5beed72015-04-30 16:09:11 -04004044 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
Jan Karac30dabf2012-06-12 16:20:30 +02004045exit2:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004046 dput(dentry);
4047 }
Al Viro59551022016-01-22 15:40:57 -05004048 inode_unlock(path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004049 if (inode)
4050 iput(inode); /* truncate the inode here */
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04004051 inode = NULL;
4052 if (delegated_inode) {
J. Bruce Fields5a146962012-08-28 07:50:40 -07004053 error = break_deleg_wait(&delegated_inode);
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04004054 if (!error)
4055 goto retry_deleg;
4056 }
Al Virof5beed72015-04-30 16:09:11 -04004057 mnt_drop_write(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004058exit1:
Al Virof5beed72015-04-30 16:09:11 -04004059 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004060 putname(name);
Jeff Layton5d18f812012-12-20 16:38:04 -05004061 if (retry_estale(error, lookup_flags)) {
4062 lookup_flags |= LOOKUP_REVAL;
4063 inode = NULL;
4064 goto retry;
4065 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004066 return error;
4067
4068slashes:
David Howellsb18825a2013-09-12 19:22:53 +01004069 if (d_is_negative(dentry))
4070 error = -ENOENT;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02004071 else if (d_is_dir(dentry))
David Howellsb18825a2013-09-12 19:22:53 +01004072 error = -EISDIR;
4073 else
4074 error = -ENOTDIR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004075 goto exit2;
4076}
4077
Heiko Carstens2e4d0922009-01-14 14:14:31 +01004078SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004079{
4080 if ((flag & ~AT_REMOVEDIR) != 0)
4081 return -EINVAL;
4082
4083 if (flag & AT_REMOVEDIR)
4084 return do_rmdir(dfd, pathname);
4085
4086 return do_unlinkat(dfd, pathname);
4087}
4088
Heiko Carstens3480b252009-01-14 14:14:16 +01004089SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004090{
4091 return do_unlinkat(AT_FDCWD, pathname);
4092}
4093
Miklos Szeredidb2e7472008-06-24 16:50:16 +02004094int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004095{
Miklos Szeredia95164d2008-07-30 15:08:48 +02004096 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004097
4098 if (error)
4099 return error;
4100
Al Viroacfa4382008-12-04 10:06:33 -05004101 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004102 return -EPERM;
4103
4104 error = security_inode_symlink(dir, dentry, oldname);
4105 if (error)
4106 return error;
4107
Linus Torvalds1da177e2005-04-16 15:20:36 -07004108 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07004109 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00004110 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004111 return error;
4112}
Al Viro4d359502014-03-14 12:20:17 -04004113EXPORT_SYMBOL(vfs_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004114
Heiko Carstens2e4d0922009-01-14 14:14:31 +01004115SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4116 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004117{
Al Viro2ad94ae2008-07-21 09:32:51 -04004118 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04004119 struct filename *from;
Dave Hansen6902d922006-09-30 23:29:01 -07004120 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04004121 struct path path;
Jeff Laytonf46d3562012-12-11 12:10:08 -05004122 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004123
4124 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04004125 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004126 return PTR_ERR(from);
Jeff Laytonf46d3562012-12-11 12:10:08 -05004127retry:
4128 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07004129 error = PTR_ERR(dentry);
4130 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04004131 goto out_putname;
Dave Hansen6902d922006-09-30 23:29:01 -07004132
Jeff Layton91a27b22012-10-10 15:25:28 -04004133 error = security_path_symlink(&path, dentry, from->name);
Al Viroa8104a92012-07-20 02:25:00 +04004134 if (!error)
Jeff Layton91a27b22012-10-10 15:25:28 -04004135 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
Al Viro921a1652012-07-20 01:15:31 +04004136 done_path_create(&path, dentry);
Jeff Laytonf46d3562012-12-11 12:10:08 -05004137 if (retry_estale(error, lookup_flags)) {
4138 lookup_flags |= LOOKUP_REVAL;
4139 goto retry;
4140 }
Dave Hansen6902d922006-09-30 23:29:01 -07004141out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004142 putname(from);
4143 return error;
4144}
4145
Heiko Carstens3480b252009-01-14 14:14:16 +01004146SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004147{
4148 return sys_symlinkat(oldname, AT_FDCWD, newname);
4149}
4150
J. Bruce Fields146a8592011-09-20 17:14:31 -04004151/**
4152 * vfs_link - create a new link
4153 * @old_dentry: object to be linked
4154 * @dir: new parent
4155 * @new_dentry: where to create the new link
4156 * @delegated_inode: returns inode needing a delegation break
4157 *
4158 * The caller must hold dir->i_mutex
4159 *
4160 * If vfs_link discovers a delegation on the to-be-linked file in need
4161 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4162 * inode in delegated_inode. The caller should then break the delegation
4163 * and retry. Because breaking a delegation may take a long time, the
4164 * caller should drop the i_mutex before doing so.
4165 *
4166 * Alternatively, a caller may pass NULL for delegated_inode. This may
4167 * be appropriate for callers that expect the underlying filesystem not
4168 * to be NFS exported.
4169 */
4170int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004171{
4172 struct inode *inode = old_dentry->d_inode;
Al Viro8de52772012-02-06 12:45:27 -05004173 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004174 int error;
4175
4176 if (!inode)
4177 return -ENOENT;
4178
Miklos Szeredia95164d2008-07-30 15:08:48 +02004179 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004180 if (error)
4181 return error;
4182
4183 if (dir->i_sb != inode->i_sb)
4184 return -EXDEV;
4185
4186 /*
4187 * A link to an append-only or immutable file cannot be created.
4188 */
4189 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4190 return -EPERM;
Eric W. Biederman0bd23d092016-06-29 14:54:46 -05004191 /*
4192 * Updating the link count will likely cause i_uid and i_gid to
4193 * be writen back improperly if their true value is unknown to
4194 * the vfs.
4195 */
4196 if (HAS_UNMAPPED_ID(inode))
4197 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05004198 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004199 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02004200 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004201 return -EPERM;
4202
4203 error = security_inode_link(old_dentry, dir, new_dentry);
4204 if (error)
4205 return error;
4206
Al Viro59551022016-01-22 15:40:57 -05004207 inode_lock(inode);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05304208 /* Make sure we don't allow creating hardlink to an unlinked file */
Al Virof4e0c302013-06-11 08:34:36 +04004209 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05304210 error = -ENOENT;
Al Viro8de52772012-02-06 12:45:27 -05004211 else if (max_links && inode->i_nlink >= max_links)
4212 error = -EMLINK;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004213 else {
4214 error = try_break_deleg(inode, delegated_inode);
4215 if (!error)
4216 error = dir->i_op->link(old_dentry, dir, new_dentry);
4217 }
Al Virof4e0c302013-06-11 08:34:36 +04004218
4219 if (!error && (inode->i_state & I_LINKABLE)) {
4220 spin_lock(&inode->i_lock);
4221 inode->i_state &= ~I_LINKABLE;
4222 spin_unlock(&inode->i_lock);
4223 }
Al Viro59551022016-01-22 15:40:57 -05004224 inode_unlock(inode);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07004225 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02004226 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004227 return error;
4228}
Al Viro4d359502014-03-14 12:20:17 -04004229EXPORT_SYMBOL(vfs_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004230
4231/*
4232 * Hardlinks are often used in delicate situations. We avoid
4233 * security-related surprises by not following symlinks on the
4234 * newname. --KAB
4235 *
4236 * We don't follow them on the oldname either to be compatible
4237 * with linux 2.0, and to avoid hard-linking to directories
4238 * and other special files. --ADM
4239 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01004240SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4241 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004242{
4243 struct dentry *new_dentry;
Al Virodae6ad82011-06-26 11:50:15 -04004244 struct path old_path, new_path;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004245 struct inode *delegated_inode = NULL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304246 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004247 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004248
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304249 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004250 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304251 /*
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004252 * To use null names we require CAP_DAC_READ_SEARCH
4253 * This ensures that not everyone will be able to create
4254 * handlink using the passed filedescriptor.
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304255 */
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004256 if (flags & AT_EMPTY_PATH) {
4257 if (!capable(CAP_DAC_READ_SEARCH))
4258 return -ENOENT;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304259 how = LOOKUP_EMPTY;
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004260 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004261
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304262 if (flags & AT_SYMLINK_FOLLOW)
4263 how |= LOOKUP_FOLLOW;
Jeff Layton442e31c2012-12-20 16:15:38 -05004264retry:
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304265 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004266 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04004267 return error;
4268
Jeff Layton442e31c2012-12-20 16:15:38 -05004269 new_dentry = user_path_create(newdfd, newname, &new_path,
4270 (how & LOOKUP_REVAL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004271 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07004272 if (IS_ERR(new_dentry))
Al Virodae6ad82011-06-26 11:50:15 -04004273 goto out;
4274
4275 error = -EXDEV;
4276 if (old_path.mnt != new_path.mnt)
4277 goto out_dput;
Kees Cook800179c2012-07-25 17:29:07 -07004278 error = may_linkat(&old_path);
4279 if (unlikely(error))
4280 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04004281 error = security_path_link(old_path.dentry, &new_path, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004282 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04004283 goto out_dput;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004284 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
Dave Hansen75c3f292008-02-15 14:37:45 -08004285out_dput:
Al Viro921a1652012-07-20 01:15:31 +04004286 done_path_create(&new_path, new_dentry);
J. Bruce Fields146a8592011-09-20 17:14:31 -04004287 if (delegated_inode) {
4288 error = break_deleg_wait(&delegated_inode);
Oleg Drokind22e6332014-01-31 15:41:58 -05004289 if (!error) {
4290 path_put(&old_path);
J. Bruce Fields146a8592011-09-20 17:14:31 -04004291 goto retry;
Oleg Drokind22e6332014-01-31 15:41:58 -05004292 }
J. Bruce Fields146a8592011-09-20 17:14:31 -04004293 }
Jeff Layton442e31c2012-12-20 16:15:38 -05004294 if (retry_estale(error, how)) {
Oleg Drokind22e6332014-01-31 15:41:58 -05004295 path_put(&old_path);
Jeff Layton442e31c2012-12-20 16:15:38 -05004296 how |= LOOKUP_REVAL;
4297 goto retry;
4298 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004299out:
Al Viro2d8f3032008-07-22 09:59:21 -04004300 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004301
4302 return error;
4303}
4304
Heiko Carstens3480b252009-01-14 14:14:16 +01004305SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004306{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004307 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004308}
4309
Miklos Szeredibc270272014-04-01 17:08:42 +02004310/**
4311 * vfs_rename - rename a filesystem object
4312 * @old_dir: parent of source
4313 * @old_dentry: source
4314 * @new_dir: parent of destination
4315 * @new_dentry: destination
4316 * @delegated_inode: returns an inode needing a delegation break
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004317 * @flags: rename flags
Miklos Szeredibc270272014-04-01 17:08:42 +02004318 *
4319 * The caller must hold multiple mutexes--see lock_rename()).
4320 *
4321 * If vfs_rename discovers a delegation in need of breaking at either
4322 * the source or destination, it will return -EWOULDBLOCK and return a
4323 * reference to the inode in delegated_inode. The caller should then
4324 * break the delegation and retry. Because breaking a delegation may
4325 * take a long time, the caller should drop all locks before doing
4326 * so.
4327 *
4328 * Alternatively, a caller may pass NULL for delegated_inode. This may
4329 * be appropriate for callers that expect the underlying filesystem not
4330 * to be NFS exported.
4331 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07004332 * The worst of all namespace operations - renaming directory. "Perverted"
4333 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4334 * Problems:
J. Bruce Fieldsd03b29a2014-02-17 16:52:33 -05004335 * a) we can get into loop creation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004336 * b) race potential - two innocent renames can create a loop together.
4337 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004338 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07004339 * story.
J. Bruce Fields6cedba82012-03-05 11:40:41 -05004340 * c) we have to lock _four_ objects - parents and victim (if it exists),
4341 * and source (if it is not a directory).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004342 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07004343 * whether the target exists). Solution: try to be smart with locking
4344 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004345 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07004346 * move will be locked. Thus we can rank directories by the tree
4347 * (ancestors first) and rank all non-directories after them.
4348 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004349 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004350 * HOWEVER, it relies on the assumption that any object with ->lookup()
4351 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4352 * we'd better make sure that there's no link(2) for them.
Sage Weile4eaac02011-05-24 13:06:07 -07004353 * d) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004354 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07004355 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05004356 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07004357 * locking].
4358 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004359int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004360 struct inode *new_dir, struct dentry *new_dentry,
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004361 struct inode **delegated_inode, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004362{
4363 int error;
Miklos Szeredibc270272014-04-01 17:08:42 +02004364 bool is_dir = d_is_dir(old_dentry);
Eric Paris59b0df22010-02-08 12:53:52 -05004365 const unsigned char *old_name;
Miklos Szeredibc270272014-04-01 17:08:42 +02004366 struct inode *source = old_dentry->d_inode;
4367 struct inode *target = new_dentry->d_inode;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004368 bool new_is_dir = false;
4369 unsigned max_links = new_dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004370
Miklos Szeredi8d3e2932016-12-16 11:02:54 +01004371 if (source == target)
Miklos Szeredibc270272014-04-01 17:08:42 +02004372 return 0;
4373
Linus Torvalds1da177e2005-04-16 15:20:36 -07004374 error = may_delete(old_dir, old_dentry, is_dir);
4375 if (error)
4376 return error;
4377
Miklos Szeredida1ce062014-04-01 17:08:43 +02004378 if (!target) {
Miklos Szeredia95164d2008-07-30 15:08:48 +02004379 error = may_create(new_dir, new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004380 } else {
4381 new_is_dir = d_is_dir(new_dentry);
4382
4383 if (!(flags & RENAME_EXCHANGE))
4384 error = may_delete(new_dir, new_dentry, is_dir);
4385 else
4386 error = may_delete(new_dir, new_dentry, new_is_dir);
4387 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004388 if (error)
4389 return error;
4390
Miklos Szeredi2773bf02016-09-27 11:03:58 +02004391 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004392 return -EPERM;
4393
Miklos Szeredibc270272014-04-01 17:08:42 +02004394 /*
4395 * If we are going to change the parent - check write permissions,
4396 * we'll need to flip '..'.
4397 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02004398 if (new_dir != old_dir) {
4399 if (is_dir) {
4400 error = inode_permission(source, MAY_WRITE);
4401 if (error)
4402 return error;
4403 }
4404 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4405 error = inode_permission(target, MAY_WRITE);
4406 if (error)
4407 return error;
4408 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004409 }
Robert Love0eeca282005-07-12 17:06:03 -04004410
Miklos Szeredi0b3974e2014-04-01 17:08:43 +02004411 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4412 flags);
Miklos Szeredibc270272014-04-01 17:08:42 +02004413 if (error)
4414 return error;
4415
4416 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4417 dget(new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004418 if (!is_dir || (flags & RENAME_EXCHANGE))
Miklos Szeredibc270272014-04-01 17:08:42 +02004419 lock_two_nondirectories(source, target);
4420 else if (target)
Al Viro59551022016-01-22 15:40:57 -05004421 inode_lock(target);
Miklos Szeredibc270272014-04-01 17:08:42 +02004422
4423 error = -EBUSY;
Eric W. Biederman7af13642013-10-04 19:15:13 -07004424 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
Miklos Szeredibc270272014-04-01 17:08:42 +02004425 goto out;
4426
Miklos Szeredida1ce062014-04-01 17:08:43 +02004427 if (max_links && new_dir != old_dir) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004428 error = -EMLINK;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004429 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
Miklos Szeredibc270272014-04-01 17:08:42 +02004430 goto out;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004431 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4432 old_dir->i_nlink >= max_links)
4433 goto out;
4434 }
4435 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4436 shrink_dcache_parent(new_dentry);
4437 if (!is_dir) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004438 error = try_break_deleg(source, delegated_inode);
4439 if (error)
4440 goto out;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004441 }
4442 if (target && !new_is_dir) {
4443 error = try_break_deleg(target, delegated_inode);
4444 if (error)
4445 goto out;
Miklos Szeredibc270272014-04-01 17:08:42 +02004446 }
Miklos Szeredi2773bf02016-09-27 11:03:58 +02004447 error = old_dir->i_op->rename(old_dir, old_dentry,
Miklos Szeredi18fc84d2016-09-27 11:03:58 +02004448 new_dir, new_dentry, flags);
Miklos Szeredibc270272014-04-01 17:08:42 +02004449 if (error)
4450 goto out;
4451
Miklos Szeredida1ce062014-04-01 17:08:43 +02004452 if (!(flags & RENAME_EXCHANGE) && target) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004453 if (is_dir)
4454 target->i_flags |= S_DEAD;
4455 dont_mount(new_dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07004456 detach_mounts(new_dentry);
Miklos Szeredibc270272014-04-01 17:08:42 +02004457 }
Miklos Szeredida1ce062014-04-01 17:08:43 +02004458 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4459 if (!(flags & RENAME_EXCHANGE))
4460 d_move(old_dentry, new_dentry);
4461 else
4462 d_exchange(old_dentry, new_dentry);
4463 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004464out:
Miklos Szeredida1ce062014-04-01 17:08:43 +02004465 if (!is_dir || (flags & RENAME_EXCHANGE))
Miklos Szeredibc270272014-04-01 17:08:42 +02004466 unlock_two_nondirectories(source, target);
4467 else if (target)
Al Viro59551022016-01-22 15:40:57 -05004468 inode_unlock(target);
Miklos Szeredibc270272014-04-01 17:08:42 +02004469 dput(new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004470 if (!error) {
Al Viro123df292009-12-25 04:57:57 -05004471 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Miklos Szeredida1ce062014-04-01 17:08:43 +02004472 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4473 if (flags & RENAME_EXCHANGE) {
4474 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4475 new_is_dir, NULL, new_dentry);
4476 }
4477 }
Robert Love0eeca282005-07-12 17:06:03 -04004478 fsnotify_oldname_free(old_name);
4479
Linus Torvalds1da177e2005-04-16 15:20:36 -07004480 return error;
4481}
Al Viro4d359502014-03-14 12:20:17 -04004482EXPORT_SYMBOL(vfs_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004483
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004484SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4485 int, newdfd, const char __user *, newname, unsigned int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004486{
Al Viro2ad94ae2008-07-21 09:32:51 -04004487 struct dentry *old_dentry, *new_dentry;
4488 struct dentry *trap;
Al Virof5beed72015-04-30 16:09:11 -04004489 struct path old_path, new_path;
4490 struct qstr old_last, new_last;
4491 int old_type, new_type;
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004492 struct inode *delegated_inode = NULL;
Jeff Layton91a27b22012-10-10 15:25:28 -04004493 struct filename *from;
4494 struct filename *to;
Al Virof5beed72015-04-30 16:09:11 -04004495 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
Jeff Laytonc6a94282012-12-11 12:10:10 -05004496 bool should_retry = false;
Al Viro2ad94ae2008-07-21 09:32:51 -04004497 int error;
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004498
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004499 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredida1ce062014-04-01 17:08:43 +02004500 return -EINVAL;
4501
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004502 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4503 (flags & RENAME_EXCHANGE))
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004504 return -EINVAL;
4505
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004506 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4507 return -EPERM;
4508
Al Virof5beed72015-04-30 16:09:11 -04004509 if (flags & RENAME_EXCHANGE)
4510 target_flags = 0;
4511
Jeff Laytonc6a94282012-12-11 12:10:10 -05004512retry:
Al Viroc1d4dd22016-06-05 16:38:18 -04004513 from = filename_parentat(olddfd, getname(oldname), lookup_flags,
4514 &old_path, &old_last, &old_type);
Jeff Layton91a27b22012-10-10 15:25:28 -04004515 if (IS_ERR(from)) {
4516 error = PTR_ERR(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004517 goto exit;
Jeff Layton91a27b22012-10-10 15:25:28 -04004518 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004519
Al Viroc1d4dd22016-06-05 16:38:18 -04004520 to = filename_parentat(newdfd, getname(newname), lookup_flags,
4521 &new_path, &new_last, &new_type);
Jeff Layton91a27b22012-10-10 15:25:28 -04004522 if (IS_ERR(to)) {
4523 error = PTR_ERR(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004524 goto exit1;
Jeff Layton91a27b22012-10-10 15:25:28 -04004525 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004526
4527 error = -EXDEV;
Al Virof5beed72015-04-30 16:09:11 -04004528 if (old_path.mnt != new_path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004529 goto exit2;
4530
Linus Torvalds1da177e2005-04-16 15:20:36 -07004531 error = -EBUSY;
Al Virof5beed72015-04-30 16:09:11 -04004532 if (old_type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004533 goto exit2;
4534
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004535 if (flags & RENAME_NOREPLACE)
4536 error = -EEXIST;
Al Virof5beed72015-04-30 16:09:11 -04004537 if (new_type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004538 goto exit2;
4539
Al Virof5beed72015-04-30 16:09:11 -04004540 error = mnt_want_write(old_path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02004541 if (error)
4542 goto exit2;
4543
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004544retry_deleg:
Al Virof5beed72015-04-30 16:09:11 -04004545 trap = lock_rename(new_path.dentry, old_path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004546
Al Virof5beed72015-04-30 16:09:11 -04004547 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004548 error = PTR_ERR(old_dentry);
4549 if (IS_ERR(old_dentry))
4550 goto exit3;
4551 /* source must exist */
4552 error = -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01004553 if (d_is_negative(old_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004554 goto exit4;
Al Virof5beed72015-04-30 16:09:11 -04004555 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004556 error = PTR_ERR(new_dentry);
4557 if (IS_ERR(new_dentry))
4558 goto exit4;
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004559 error = -EEXIST;
4560 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4561 goto exit5;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004562 if (flags & RENAME_EXCHANGE) {
4563 error = -ENOENT;
4564 if (d_is_negative(new_dentry))
4565 goto exit5;
4566
4567 if (!d_is_dir(new_dentry)) {
4568 error = -ENOTDIR;
Al Virof5beed72015-04-30 16:09:11 -04004569 if (new_last.name[new_last.len])
Miklos Szeredida1ce062014-04-01 17:08:43 +02004570 goto exit5;
4571 }
4572 }
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004573 /* unless the source is a directory trailing slashes give -ENOTDIR */
4574 if (!d_is_dir(old_dentry)) {
4575 error = -ENOTDIR;
Al Virof5beed72015-04-30 16:09:11 -04004576 if (old_last.name[old_last.len])
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004577 goto exit5;
Al Virof5beed72015-04-30 16:09:11 -04004578 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004579 goto exit5;
4580 }
4581 /* source should not be ancestor of target */
4582 error = -EINVAL;
4583 if (old_dentry == trap)
4584 goto exit5;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004585 /* target should not be an ancestor of source */
Miklos Szeredida1ce062014-04-01 17:08:43 +02004586 if (!(flags & RENAME_EXCHANGE))
4587 error = -ENOTEMPTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004588 if (new_dentry == trap)
4589 goto exit5;
4590
Al Virof5beed72015-04-30 16:09:11 -04004591 error = security_path_rename(&old_path, old_dentry,
4592 &new_path, new_dentry, flags);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004593 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02004594 goto exit5;
Al Virof5beed72015-04-30 16:09:11 -04004595 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4596 new_path.dentry->d_inode, new_dentry,
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004597 &delegated_inode, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004598exit5:
4599 dput(new_dentry);
4600exit4:
4601 dput(old_dentry);
4602exit3:
Al Virof5beed72015-04-30 16:09:11 -04004603 unlock_rename(new_path.dentry, old_path.dentry);
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004604 if (delegated_inode) {
4605 error = break_deleg_wait(&delegated_inode);
4606 if (!error)
4607 goto retry_deleg;
4608 }
Al Virof5beed72015-04-30 16:09:11 -04004609 mnt_drop_write(old_path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004610exit2:
Jeff Laytonc6a94282012-12-11 12:10:10 -05004611 if (retry_estale(error, lookup_flags))
4612 should_retry = true;
Al Virof5beed72015-04-30 16:09:11 -04004613 path_put(&new_path);
Al Viro2ad94ae2008-07-21 09:32:51 -04004614 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004615exit1:
Al Virof5beed72015-04-30 16:09:11 -04004616 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004617 putname(from);
Jeff Laytonc6a94282012-12-11 12:10:10 -05004618 if (should_retry) {
4619 should_retry = false;
4620 lookup_flags |= LOOKUP_REVAL;
4621 goto retry;
4622 }
Al Viro2ad94ae2008-07-21 09:32:51 -04004623exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004624 return error;
4625}
4626
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004627SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4628 int, newdfd, const char __user *, newname)
4629{
4630 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4631}
4632
Heiko Carstensa26eab22009-01-14 14:14:17 +01004633SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004634{
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004635 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004636}
4637
Miklos Szeredi787fb6b2014-10-24 00:14:36 +02004638int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4639{
4640 int error = may_create(dir, dentry);
4641 if (error)
4642 return error;
4643
4644 if (!dir->i_op->mknod)
4645 return -EPERM;
4646
4647 return dir->i_op->mknod(dir, dentry,
4648 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4649}
4650EXPORT_SYMBOL(vfs_whiteout);
4651
Al Viro5d826c82014-03-14 13:42:45 -04004652int readlink_copy(char __user *buffer, int buflen, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004653{
Al Viro5d826c82014-03-14 13:42:45 -04004654 int len = PTR_ERR(link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004655 if (IS_ERR(link))
4656 goto out;
4657
4658 len = strlen(link);
4659 if (len > (unsigned) buflen)
4660 len = buflen;
4661 if (copy_to_user(buffer, link, len))
4662 len = -EFAULT;
4663out:
4664 return len;
4665}
4666
4667/*
4668 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
Al Viro6b255392015-11-17 10:20:54 -05004669 * have ->get_link() not calling nd_jump_link(). Using (or not using) it
4670 * for any given inode is up to filesystem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004671 */
Miklos Szeredid16744e2016-12-09 16:45:04 +01004672static int generic_readlink(struct dentry *dentry, char __user *buffer,
4673 int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004674{
Al Virofceef392015-12-29 15:58:39 -05004675 DEFINE_DELAYED_CALL(done);
Al Viro5f2c4172015-05-07 11:14:26 -04004676 struct inode *inode = d_inode(dentry);
4677 const char *link = inode->i_link;
Marcin Slusarz694a1762008-06-09 16:40:37 -07004678 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004679
Al Virod4dee482015-04-30 20:08:02 -04004680 if (!link) {
Al Virofceef392015-12-29 15:58:39 -05004681 link = inode->i_op->get_link(dentry, inode, &done);
Al Virod4dee482015-04-30 20:08:02 -04004682 if (IS_ERR(link))
4683 return PTR_ERR(link);
4684 }
Al Viro680baac2015-05-02 13:32:22 -04004685 res = readlink_copy(buffer, buflen, link);
Al Virofceef392015-12-29 15:58:39 -05004686 do_delayed_call(&done);
Marcin Slusarz694a1762008-06-09 16:40:37 -07004687 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004688}
4689
Miklos Szeredid60874c2016-10-04 14:40:45 +02004690/**
Miklos Szeredifd4a0edf2016-12-09 16:45:04 +01004691 * vfs_readlink - copy symlink body into userspace buffer
4692 * @dentry: dentry on which to get symbolic link
4693 * @buffer: user memory pointer
4694 * @buflen: size of buffer
4695 *
4696 * Does not touch atime. That's up to the caller if necessary
4697 *
4698 * Does not call security hook.
4699 */
4700int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4701{
4702 struct inode *inode = d_inode(dentry);
4703
Miklos Szeredi76fca902016-12-09 16:45:04 +01004704 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
4705 if (unlikely(inode->i_op->readlink))
4706 return inode->i_op->readlink(dentry, buffer, buflen);
Miklos Szeredifd4a0edf2016-12-09 16:45:04 +01004707
Miklos Szeredi76fca902016-12-09 16:45:04 +01004708 if (!d_is_symlink(dentry))
4709 return -EINVAL;
4710
4711 spin_lock(&inode->i_lock);
4712 inode->i_opflags |= IOP_DEFAULT_READLINK;
4713 spin_unlock(&inode->i_lock);
4714 }
4715
4716 return generic_readlink(dentry, buffer, buflen);
Miklos Szeredifd4a0edf2016-12-09 16:45:04 +01004717}
4718EXPORT_SYMBOL(vfs_readlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004719
Miklos Szeredid60874c2016-10-04 14:40:45 +02004720/**
4721 * vfs_get_link - get symlink body
4722 * @dentry: dentry on which to get symbolic link
4723 * @done: caller needs to free returned data with this
4724 *
4725 * Calls security hook and i_op->get_link() on the supplied inode.
4726 *
4727 * It does not touch atime. That's up to the caller if necessary.
4728 *
4729 * Does not work on "special" symlinks like /proc/$$/fd/N
4730 */
4731const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4732{
4733 const char *res = ERR_PTR(-EINVAL);
4734 struct inode *inode = d_inode(dentry);
4735
4736 if (d_is_symlink(dentry)) {
4737 res = ERR_PTR(security_inode_readlink(dentry));
4738 if (!res)
4739 res = inode->i_op->get_link(dentry, inode, done);
4740 }
4741 return res;
4742}
4743EXPORT_SYMBOL(vfs_get_link);
4744
Linus Torvalds1da177e2005-04-16 15:20:36 -07004745/* get the link contents into pagecache */
Al Viro6b255392015-11-17 10:20:54 -05004746const char *page_get_link(struct dentry *dentry, struct inode *inode,
Al Virofceef392015-12-29 15:58:39 -05004747 struct delayed_call *callback)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004748{
Duane Griffinebd09ab2008-12-19 20:47:12 +00004749 char *kaddr;
4750 struct page *page;
Al Viro6b255392015-11-17 10:20:54 -05004751 struct address_space *mapping = inode->i_mapping;
4752
Al Virod3883d42015-11-17 10:41:04 -05004753 if (!dentry) {
4754 page = find_get_page(mapping, 0);
4755 if (!page)
4756 return ERR_PTR(-ECHILD);
4757 if (!PageUptodate(page)) {
4758 put_page(page);
4759 return ERR_PTR(-ECHILD);
4760 }
4761 } else {
4762 page = read_mapping_page(mapping, 0, NULL);
4763 if (IS_ERR(page))
4764 return (char*)page;
4765 }
Al Virofceef392015-12-29 15:58:39 -05004766 set_delayed_call(callback, page_put_link, page);
Al Viro21fc61c2015-11-17 01:07:57 -05004767 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4768 kaddr = page_address(page);
Al Viro6b255392015-11-17 10:20:54 -05004769 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
Duane Griffinebd09ab2008-12-19 20:47:12 +00004770 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004771}
4772
Al Viro6b255392015-11-17 10:20:54 -05004773EXPORT_SYMBOL(page_get_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004774
Al Virofceef392015-12-29 15:58:39 -05004775void page_put_link(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004776{
Al Virofceef392015-12-29 15:58:39 -05004777 put_page(arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004778}
Al Viro4d359502014-03-14 12:20:17 -04004779EXPORT_SYMBOL(page_put_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004780
Al Viroaa80dea2015-11-16 18:26:34 -05004781int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4782{
Al Virofceef392015-12-29 15:58:39 -05004783 DEFINE_DELAYED_CALL(done);
Al Viro6b255392015-11-17 10:20:54 -05004784 int res = readlink_copy(buffer, buflen,
4785 page_get_link(dentry, d_inode(dentry),
Al Virofceef392015-12-29 15:58:39 -05004786 &done));
4787 do_delayed_call(&done);
Al Viroaa80dea2015-11-16 18:26:34 -05004788 return res;
4789}
4790EXPORT_SYMBOL(page_readlink);
4791
Nick Piggin54566b22009-01-04 12:00:53 -08004792/*
4793 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4794 */
4795int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004796{
4797 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004798 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07004799 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08004800 int err;
Tetsuo Handac718a972017-05-08 15:58:59 -07004801 unsigned int flags = 0;
Nick Piggin54566b22009-01-04 12:00:53 -08004802 if (nofs)
4803 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004804
NeilBrown7e53cac2006-03-25 03:07:57 -08004805retry:
Nick Pigginafddba42007-10-16 01:25:01 -07004806 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08004807 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004808 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07004809 goto fail;
4810
Al Viro21fc61c2015-11-17 01:07:57 -05004811 memcpy(page_address(page), symname, len-1);
Nick Pigginafddba42007-10-16 01:25:01 -07004812
4813 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4814 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004815 if (err < 0)
4816 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07004817 if (err < len-1)
4818 goto retry;
4819
Linus Torvalds1da177e2005-04-16 15:20:36 -07004820 mark_inode_dirty(inode);
4821 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004822fail:
4823 return err;
4824}
Al Viro4d359502014-03-14 12:20:17 -04004825EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004826
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004827int page_symlink(struct inode *inode, const char *symname, int len)
4828{
4829 return __page_symlink(inode, symname, len,
Michal Hockoc62d2552015-11-06 16:28:49 -08004830 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004831}
Al Viro4d359502014-03-14 12:20:17 -04004832EXPORT_SYMBOL(page_symlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004833
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08004834const struct inode_operations page_symlink_inode_operations = {
Al Viro6b255392015-11-17 10:20:54 -05004835 .get_link = page_get_link,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004836};
Linus Torvalds1da177e2005-04-16 15:20:36 -07004837EXPORT_SYMBOL(page_symlink_inode_operations);