blob: e7d125c23aa66eed9df5eee379477825439101cb [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050018#include <linux/export.h>
David S. Miller44696902012-05-23 20:12:50 -070019#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/slab.h>
21#include <linux/fs.h>
22#include <linux/namei.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/pagemap.h>
Robert Love0eeca282005-07-12 17:06:03 -040024#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/personality.h>
26#include <linux/security.h>
Mimi Zohar6146f0d2009-02-04 09:06:57 -050027#include <linux/ima.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/syscalls.h>
29#include <linux/mount.h>
30#include <linux/audit.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080031#include <linux/capability.h>
Trond Myklebust834f2a42005-10-18 14:20:16 -070032#include <linux/file.h>
Ulrich Drepper5590ff02006-01-18 17:43:53 -080033#include <linux/fcntl.h>
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -070034#include <linux/device_cgroup.h>
Al Viro5ad4e532009-03-29 19:50:06 -040035#include <linux/fs_struct.h>
Linus Torvaldse77819e2011-07-22 19:30:19 -070036#include <linux/posix_acl.h>
Linus Torvalds99d263d2014-09-13 11:30:10 -070037#include <linux/hash.h>
George Spelvin2a18da7a2016-05-23 07:43:58 -040038#include <linux/bitops.h>
Eric W. Biedermanaeaa4a72016-07-23 11:20:44 -050039#include <linux/init_task.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include <asm/uaccess.h>
41
Eric Parise81e3f42009-12-04 15:47:36 -050042#include "internal.h"
Al Viroc7105362011-11-24 18:22:03 -050043#include "mount.h"
Eric Parise81e3f42009-12-04 15:47:36 -050044
Linus Torvalds1da177e2005-04-16 15:20:36 -070045/* [Feb-1997 T. Schoebel-Theuer]
46 * Fundamental changes in the pathname lookup mechanisms (namei)
47 * were necessary because of omirr. The reason is that omirr needs
48 * to know the _real_ pathname, not the user-supplied one, in case
49 * of symlinks (and also when transname replacements occur).
50 *
51 * The new code replaces the old recursive symlink resolution with
52 * an iterative one (in case of non-nested symlink chains). It does
53 * this with calls to <fs>_follow_link().
54 * As a side effect, dir_namei(), _namei() and follow_link() are now
55 * replaced with a single function lookup_dentry() that can handle all
56 * the special cases of the former code.
57 *
58 * With the new dcache, the pathname is stored at each inode, at least as
59 * long as the refcount of the inode is positive. As a side effect, the
60 * size of the dcache depends on the inode cache and thus is dynamic.
61 *
62 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
63 * resolution to correspond with current state of the code.
64 *
65 * Note that the symlink resolution is not *completely* iterative.
66 * There is still a significant amount of tail- and mid- recursion in
67 * the algorithm. Also, note that <fs>_readlink() is not used in
68 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
69 * may return different results than <fs>_follow_link(). Many virtual
70 * filesystems (including /proc) exhibit this behavior.
71 */
72
73/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
74 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
75 * and the name already exists in form of a symlink, try to create the new
76 * name indicated by the symlink. The old code always complained that the
77 * name already exists, due to not following the symlink even if its target
78 * is nonexistent. The new semantics affects also mknod() and link() when
Lucas De Marchi25985ed2011-03-30 22:57:33 -030079 * the name is a symlink pointing to a non-existent name.
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 *
81 * I don't know which semantics is the right one, since I have no access
82 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
83 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
84 * "old" one. Personally, I think the new semantics is much more logical.
85 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
86 * file does succeed in both HP-UX and SunOs, but not in Solaris
87 * and in the old Linux semantics.
88 */
89
90/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
91 * semantics. See the comments in "open_namei" and "do_link" below.
92 *
93 * [10-Sep-98 Alan Modra] Another symlink change.
94 */
95
96/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
97 * inside the path - always follow.
98 * in the last component in creation/removal/renaming - never follow.
99 * if LOOKUP_FOLLOW passed - follow.
100 * if the pathname has trailing slashes - follow.
101 * otherwise - don't follow.
102 * (applied in that order).
103 *
104 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
105 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
106 * During the 2.4 we need to fix the userland stuff depending on it -
107 * hopefully we will be able to get rid of that wart in 2.5. So far only
108 * XEmacs seems to be relying on it...
109 */
110/*
111 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800112 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 * any extra contention...
114 */
115
116/* In order to reduce some races, while at the same time doing additional
117 * checking and hopefully speeding things up, we copy filenames to the
118 * kernel data space before using them..
119 *
120 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
121 * PATH_MAX includes the nul terminator --RR.
122 */
Jeff Layton91a27b22012-10-10 15:25:28 -0400123
Al Virofd2f7cb2015-02-22 20:07:13 -0500124#define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
Jeff Layton7950e382012-10-10 16:43:13 -0400125
David Drysdale51f39a12014-12-12 16:57:29 -0800126struct filename *
Jeff Layton91a27b22012-10-10 15:25:28 -0400127getname_flags(const char __user *filename, int flags, int *empty)
128{
Al Viro94b5d262015-02-22 19:38:03 -0500129 struct filename *result;
Jeff Layton7950e382012-10-10 16:43:13 -0400130 char *kname;
Al Viro94b5d262015-02-22 19:38:03 -0500131 int len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132
Jeff Layton7ac86262012-10-10 15:25:28 -0400133 result = audit_reusename(filename);
134 if (result)
135 return result;
136
Jeff Layton7950e382012-10-10 16:43:13 -0400137 result = __getname();
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700138 if (unlikely(!result))
Eric Paris4043cde2012-01-03 14:23:08 -0500139 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140
Jeff Layton7950e382012-10-10 16:43:13 -0400141 /*
142 * First, try to embed the struct filename inside the names_cache
143 * allocation
144 */
Al Virofd2f7cb2015-02-22 20:07:13 -0500145 kname = (char *)result->iname;
Jeff Layton91a27b22012-10-10 15:25:28 -0400146 result->name = kname;
Jeff Layton7950e382012-10-10 16:43:13 -0400147
Al Viro94b5d262015-02-22 19:38:03 -0500148 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
Jeff Layton91a27b22012-10-10 15:25:28 -0400149 if (unlikely(len < 0)) {
Al Viro94b5d262015-02-22 19:38:03 -0500150 __putname(result);
151 return ERR_PTR(len);
Jeff Layton91a27b22012-10-10 15:25:28 -0400152 }
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700153
Jeff Layton7950e382012-10-10 16:43:13 -0400154 /*
155 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
156 * separate struct filename so we can dedicate the entire
157 * names_cache allocation for the pathname, and re-do the copy from
158 * userland.
159 */
Al Viro94b5d262015-02-22 19:38:03 -0500160 if (unlikely(len == EMBEDDED_NAME_MAX)) {
Al Virofd2f7cb2015-02-22 20:07:13 -0500161 const size_t size = offsetof(struct filename, iname[1]);
Jeff Layton7950e382012-10-10 16:43:13 -0400162 kname = (char *)result;
163
Al Virofd2f7cb2015-02-22 20:07:13 -0500164 /*
165 * size is chosen that way we to guarantee that
166 * result->iname[0] is within the same object and that
167 * kname can't be equal to result->iname, no matter what.
168 */
169 result = kzalloc(size, GFP_KERNEL);
Al Viro94b5d262015-02-22 19:38:03 -0500170 if (unlikely(!result)) {
171 __putname(kname);
172 return ERR_PTR(-ENOMEM);
Jeff Layton7950e382012-10-10 16:43:13 -0400173 }
174 result->name = kname;
Al Viro94b5d262015-02-22 19:38:03 -0500175 len = strncpy_from_user(kname, filename, PATH_MAX);
176 if (unlikely(len < 0)) {
177 __putname(kname);
178 kfree(result);
179 return ERR_PTR(len);
180 }
181 if (unlikely(len == PATH_MAX)) {
182 __putname(kname);
183 kfree(result);
184 return ERR_PTR(-ENAMETOOLONG);
185 }
Jeff Layton7950e382012-10-10 16:43:13 -0400186 }
187
Al Viro94b5d262015-02-22 19:38:03 -0500188 result->refcnt = 1;
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700189 /* The empty path is special. */
190 if (unlikely(!len)) {
191 if (empty)
Eric Paris4043cde2012-01-03 14:23:08 -0500192 *empty = 1;
Al Viro94b5d262015-02-22 19:38:03 -0500193 if (!(flags & LOOKUP_EMPTY)) {
194 putname(result);
195 return ERR_PTR(-ENOENT);
196 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 }
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700198
Jeff Layton7950e382012-10-10 16:43:13 -0400199 result->uptr = filename;
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800200 result->aname = NULL;
Jeff Layton7950e382012-10-10 16:43:13 -0400201 audit_getname(result);
202 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203}
204
Jeff Layton91a27b22012-10-10 15:25:28 -0400205struct filename *
206getname(const char __user * filename)
Al Virof52e0c12011-03-14 18:56:51 -0400207{
Linus Torvaldsf7493e52012-03-22 16:10:40 -0700208 return getname_flags(filename, 0, NULL);
Al Virof52e0c12011-03-14 18:56:51 -0400209}
210
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800211struct filename *
212getname_kernel(const char * filename)
213{
214 struct filename *result;
Paul Moore08518542015-01-21 23:59:56 -0500215 int len = strlen(filename) + 1;
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800216
217 result = __getname();
218 if (unlikely(!result))
219 return ERR_PTR(-ENOMEM);
220
Paul Moore08518542015-01-21 23:59:56 -0500221 if (len <= EMBEDDED_NAME_MAX) {
Al Virofd2f7cb2015-02-22 20:07:13 -0500222 result->name = (char *)result->iname;
Paul Moore08518542015-01-21 23:59:56 -0500223 } else if (len <= PATH_MAX) {
224 struct filename *tmp;
225
226 tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
227 if (unlikely(!tmp)) {
228 __putname(result);
229 return ERR_PTR(-ENOMEM);
230 }
231 tmp->name = (char *)result;
Paul Moore08518542015-01-21 23:59:56 -0500232 result = tmp;
233 } else {
234 __putname(result);
235 return ERR_PTR(-ENAMETOOLONG);
236 }
237 memcpy((char *)result->name, filename, len);
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800238 result->uptr = NULL;
239 result->aname = NULL;
Paul Moore55422d02015-01-22 00:00:23 -0500240 result->refcnt = 1;
Paul Moorefd3522f2015-01-22 00:00:10 -0500241 audit_getname(result);
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800242
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800243 return result;
244}
245
Jeff Layton91a27b22012-10-10 15:25:28 -0400246void putname(struct filename *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247{
Paul Moore55422d02015-01-22 00:00:23 -0500248 BUG_ON(name->refcnt <= 0);
249
250 if (--name->refcnt > 0)
251 return;
252
Al Virofd2f7cb2015-02-22 20:07:13 -0500253 if (name->name != name->iname) {
Paul Moore55422d02015-01-22 00:00:23 -0500254 __putname(name->name);
255 kfree(name);
256 } else
257 __putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259
Linus Torvaldse77819e2011-07-22 19:30:19 -0700260static int check_acl(struct inode *inode, int mask)
261{
Linus Torvalds84635d62011-07-25 22:47:03 -0700262#ifdef CONFIG_FS_POSIX_ACL
Linus Torvaldse77819e2011-07-22 19:30:19 -0700263 struct posix_acl *acl;
264
Linus Torvaldse77819e2011-07-22 19:30:19 -0700265 if (mask & MAY_NOT_BLOCK) {
Al Viro35678662011-08-02 21:32:13 -0400266 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
267 if (!acl)
Linus Torvaldse77819e2011-07-22 19:30:19 -0700268 return -EAGAIN;
Al Viro35678662011-08-02 21:32:13 -0400269 /* no ->get_acl() calls in RCU mode... */
Andreas Gruenbacherb8a7a3a2016-03-24 14:38:37 +0100270 if (is_uncached_acl(acl))
Al Viro35678662011-08-02 21:32:13 -0400271 return -ECHILD;
Ari Savolainen206b1d02011-08-06 19:43:07 +0300272 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
Linus Torvaldse77819e2011-07-22 19:30:19 -0700273 }
274
Christoph Hellwig2982baa2013-12-20 05:16:38 -0800275 acl = get_acl(inode, ACL_TYPE_ACCESS);
276 if (IS_ERR(acl))
277 return PTR_ERR(acl);
Linus Torvaldse77819e2011-07-22 19:30:19 -0700278 if (acl) {
279 int error = posix_acl_permission(inode, acl, mask);
280 posix_acl_release(acl);
281 return error;
282 }
Linus Torvalds84635d62011-07-25 22:47:03 -0700283#endif
Linus Torvaldse77819e2011-07-22 19:30:19 -0700284
285 return -EAGAIN;
286}
287
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700288/*
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530289 * This does the basic permission checking
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700290 */
Al Viro7e401452011-06-20 19:12:17 -0400291static int acl_permission_check(struct inode *inode, int mask)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700292{
Linus Torvalds26cf46b2011-05-13 11:51:01 -0700293 unsigned int mode = inode->i_mode;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700294
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -0800295 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700296 mode >>= 6;
297 else {
Linus Torvaldse77819e2011-07-22 19:30:19 -0700298 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
Al Viro7e401452011-06-20 19:12:17 -0400299 int error = check_acl(inode, mask);
Nick Pigginb74c79e2011-01-07 17:49:58 +1100300 if (error != -EAGAIN)
301 return error;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700302 }
303
304 if (in_group_p(inode->i_gid))
305 mode >>= 3;
306 }
307
308 /*
309 * If the DACs are ok we don't need any capability check.
310 */
Al Viro9c2c7032011-06-20 19:06:22 -0400311 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700312 return 0;
313 return -EACCES;
314}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315
316/**
Nick Pigginb74c79e2011-01-07 17:49:58 +1100317 * generic_permission - check for access rights on a Posix-like filesystem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 * @inode: inode to check access rights for
Andreas Gruenbacher8fd90c82011-10-23 23:13:30 +0530319 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 *
321 * Used to check for read/write/execute permissions on a file.
322 * We use "fsuid" for this, letting us set arbitrary permissions
323 * for filesystem access without changing the "normal" uids which
Nick Pigginb74c79e2011-01-07 17:49:58 +1100324 * are used for other things.
325 *
326 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
327 * request cannot be satisfied (eg. requires blocking or too much complexity).
328 * It would then be called again in ref-walk mode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 */
Al Viro2830ba72011-06-20 19:16:29 -0400330int generic_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700332 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333
334 /*
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530335 * Do the basic permission checks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 */
Al Viro7e401452011-06-20 19:12:17 -0400337 ret = acl_permission_check(inode, mask);
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700338 if (ret != -EACCES)
339 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340
Al Virod594e7e2011-06-20 19:55:42 -0400341 if (S_ISDIR(inode->i_mode)) {
342 /* DACs are overridable for directories */
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700343 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
Al Virod594e7e2011-06-20 19:55:42 -0400344 return 0;
345 if (!(mask & MAY_WRITE))
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700346 if (capable_wrt_inode_uidgid(inode,
347 CAP_DAC_READ_SEARCH))
Al Virod594e7e2011-06-20 19:55:42 -0400348 return 0;
349 return -EACCES;
350 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 /*
352 * Read/write DACs are always overridable.
Al Virod594e7e2011-06-20 19:55:42 -0400353 * Executable DACs are overridable when there is
354 * at least one exec bit set.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 */
Al Virod594e7e2011-06-20 19:55:42 -0400356 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700357 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 return 0;
359
360 /*
361 * Searching includes executable on directories, else just read.
362 */
Serge E. Hallyn7ea66002009-12-29 14:50:19 -0600363 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
Al Virod594e7e2011-06-20 19:55:42 -0400364 if (mask == MAY_READ)
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700365 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 return 0;
367
368 return -EACCES;
369}
Al Viro4d359502014-03-14 12:20:17 -0400370EXPORT_SYMBOL(generic_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700372/*
373 * We _really_ want to just do "generic_permission()" without
374 * even looking at the inode->i_op values. So we keep a cache
375 * flag in inode->i_opflags, that says "this has not special
376 * permission function, use the fast case".
377 */
378static inline int do_inode_permission(struct inode *inode, int mask)
379{
380 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
381 if (likely(inode->i_op->permission))
382 return inode->i_op->permission(inode, mask);
383
384 /* This gets set once for the inode lifetime */
385 spin_lock(&inode->i_lock);
386 inode->i_opflags |= IOP_FASTPERM;
387 spin_unlock(&inode->i_lock);
388 }
389 return generic_permission(inode, mask);
390}
391
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200392/**
David Howells0bdaea92012-06-25 12:55:46 +0100393 * __inode_permission - Check for access rights to a given inode
394 * @inode: Inode to check permission on
395 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200396 *
David Howells0bdaea92012-06-25 12:55:46 +0100397 * Check for read/write/execute permissions on an inode.
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530398 *
399 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
David Howells0bdaea92012-06-25 12:55:46 +0100400 *
401 * This does not check for a read-only file system. You probably want
402 * inode_permission().
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200403 */
David Howells0bdaea92012-06-25 12:55:46 +0100404int __inode_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405{
Al Viroe6305c42008-07-15 21:03:57 -0400406 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700408 if (unlikely(mask & MAY_WRITE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 /*
410 * Nobody gets write access to an immutable file.
411 */
412 if (IS_IMMUTABLE(inode))
Eryu Guan337684a2016-08-02 19:58:28 +0800413 return -EPERM;
Eric W. Biederman0bd23d092016-06-29 14:54:46 -0500414
415 /*
416 * Updating mtime will likely cause i_uid and i_gid to be
417 * written back improperly if their true value is unknown
418 * to the vfs.
419 */
420 if (HAS_UNMAPPED_ID(inode))
421 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 }
423
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700424 retval = do_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 if (retval)
426 return retval;
427
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -0700428 retval = devcgroup_inode_permission(inode, mask);
429 if (retval)
430 return retval;
431
Eric Parisd09ca732010-07-23 11:43:57 -0400432 return security_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433}
Miklos Szeredibd5d0852014-10-24 00:14:35 +0200434EXPORT_SYMBOL(__inode_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435
Al Virof4d6ff82011-06-19 13:14:21 -0400436/**
David Howells0bdaea92012-06-25 12:55:46 +0100437 * sb_permission - Check superblock-level permissions
438 * @sb: Superblock of inode to check permission on
Randy Dunlap55852632012-08-18 17:39:25 -0700439 * @inode: Inode to check permission on
David Howells0bdaea92012-06-25 12:55:46 +0100440 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
441 *
442 * Separate out file-system wide checks from inode-specific permission checks.
443 */
444static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
445{
446 if (unlikely(mask & MAY_WRITE)) {
447 umode_t mode = inode->i_mode;
448
449 /* Nobody gets write access to a read-only fs. */
450 if ((sb->s_flags & MS_RDONLY) &&
451 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
452 return -EROFS;
453 }
454 return 0;
455}
456
457/**
458 * inode_permission - Check for access rights to a given inode
459 * @inode: Inode to check permission on
460 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
461 *
462 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
463 * this, letting us set arbitrary permissions for filesystem access without
464 * changing the "normal" UIDs which are used for other things.
465 *
466 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
467 */
468int inode_permission(struct inode *inode, int mask)
469{
470 int retval;
471
472 retval = sb_permission(inode->i_sb, inode, mask);
473 if (retval)
474 return retval;
475 return __inode_permission(inode, mask);
476}
Al Viro4d359502014-03-14 12:20:17 -0400477EXPORT_SYMBOL(inode_permission);
David Howells0bdaea92012-06-25 12:55:46 +0100478
479/**
Jan Blunck5dd784d02008-02-14 19:34:38 -0800480 * path_get - get a reference to a path
481 * @path: path to get the reference to
482 *
483 * Given a path increment the reference count to the dentry and the vfsmount.
484 */
Al Virodcf787f2013-03-01 23:51:07 -0500485void path_get(const struct path *path)
Jan Blunck5dd784d02008-02-14 19:34:38 -0800486{
487 mntget(path->mnt);
488 dget(path->dentry);
489}
490EXPORT_SYMBOL(path_get);
491
492/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800493 * path_put - put a reference to a path
494 * @path: path to put the reference to
495 *
496 * Given a path decrement the reference count to the dentry and the vfsmount.
497 */
Al Virodcf787f2013-03-01 23:51:07 -0500498void path_put(const struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499{
Jan Blunck1d957f92008-02-14 19:34:35 -0800500 dput(path->dentry);
501 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502}
Jan Blunck1d957f92008-02-14 19:34:35 -0800503EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504
Al Viro894bc8c2015-05-02 07:16:16 -0400505#define EMBEDDED_LEVELS 2
Al Viro1f55a6e2014-11-01 19:30:41 -0400506struct nameidata {
507 struct path path;
Al Viro1cf26652015-05-06 16:01:56 -0400508 struct qstr last;
Al Viro1f55a6e2014-11-01 19:30:41 -0400509 struct path root;
510 struct inode *inode; /* path.dentry.d_inode */
511 unsigned int flags;
Al Viro9883d182015-05-13 07:28:08 -0400512 unsigned seq, m_seq;
Al Viro1f55a6e2014-11-01 19:30:41 -0400513 int last_type;
514 unsigned depth;
NeilBrown756daf22015-03-23 13:37:38 +1100515 int total_link_count;
Al Viro697fc6c2015-05-02 19:38:35 -0400516 struct saved {
517 struct path link;
Al Virofceef392015-12-29 15:58:39 -0500518 struct delayed_call done;
Al Viro697fc6c2015-05-02 19:38:35 -0400519 const char *name;
Al Viro0450b2d2015-05-08 13:23:53 -0400520 unsigned seq;
Al Viro894bc8c2015-05-02 07:16:16 -0400521 } *stack, internal[EMBEDDED_LEVELS];
Al Viro9883d182015-05-13 07:28:08 -0400522 struct filename *name;
523 struct nameidata *saved;
Al Virofceef392015-12-29 15:58:39 -0500524 struct inode *link_inode;
Al Viro9883d182015-05-13 07:28:08 -0400525 unsigned root_seq;
526 int dfd;
Al Viro1f55a6e2014-11-01 19:30:41 -0400527};
528
Al Viro9883d182015-05-13 07:28:08 -0400529static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
Al Viro894bc8c2015-05-02 07:16:16 -0400530{
NeilBrown756daf22015-03-23 13:37:38 +1100531 struct nameidata *old = current->nameidata;
532 p->stack = p->internal;
Al Viroc8a53ee2015-05-12 18:43:07 -0400533 p->dfd = dfd;
534 p->name = name;
NeilBrown756daf22015-03-23 13:37:38 +1100535 p->total_link_count = old ? old->total_link_count : 0;
Al Viro9883d182015-05-13 07:28:08 -0400536 p->saved = old;
NeilBrown756daf22015-03-23 13:37:38 +1100537 current->nameidata = p;
Al Viro894bc8c2015-05-02 07:16:16 -0400538}
539
Al Viro9883d182015-05-13 07:28:08 -0400540static void restore_nameidata(void)
Al Viro894bc8c2015-05-02 07:16:16 -0400541{
Al Viro9883d182015-05-13 07:28:08 -0400542 struct nameidata *now = current->nameidata, *old = now->saved;
NeilBrown756daf22015-03-23 13:37:38 +1100543
544 current->nameidata = old;
545 if (old)
546 old->total_link_count = now->total_link_count;
Al Viroe1a63bb2015-12-05 21:06:33 -0500547 if (now->stack != now->internal)
NeilBrown756daf22015-03-23 13:37:38 +1100548 kfree(now->stack);
Al Viro894bc8c2015-05-02 07:16:16 -0400549}
550
551static int __nd_alloc_stack(struct nameidata *nd)
552{
Al Virobc40aee2015-05-09 13:04:24 -0400553 struct saved *p;
554
555 if (nd->flags & LOOKUP_RCU) {
556 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
557 GFP_ATOMIC);
558 if (unlikely(!p))
559 return -ECHILD;
560 } else {
561 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
Al Viro894bc8c2015-05-02 07:16:16 -0400562 GFP_KERNEL);
Al Virobc40aee2015-05-09 13:04:24 -0400563 if (unlikely(!p))
564 return -ENOMEM;
565 }
Al Viro894bc8c2015-05-02 07:16:16 -0400566 memcpy(p, nd->internal, sizeof(nd->internal));
567 nd->stack = p;
568 return 0;
569}
570
Eric W. Biederman397d4252015-08-15 20:27:13 -0500571/**
572 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
573 * @path: nameidate to verify
574 *
575 * Rename can sometimes move a file or directory outside of a bind
576 * mount, path_connected allows those cases to be detected.
577 */
578static bool path_connected(const struct path *path)
579{
580 struct vfsmount *mnt = path->mnt;
581
582 /* Only bind mounts can have disconnected paths */
583 if (mnt->mnt_root == mnt->mnt_sb->s_root)
584 return true;
585
586 return is_subdir(path->dentry, mnt->mnt_root);
587}
588
Al Viro894bc8c2015-05-02 07:16:16 -0400589static inline int nd_alloc_stack(struct nameidata *nd)
590{
Al Viroda4e0be2015-05-03 20:52:15 -0400591 if (likely(nd->depth != EMBEDDED_LEVELS))
Al Viro894bc8c2015-05-02 07:16:16 -0400592 return 0;
593 if (likely(nd->stack != nd->internal))
594 return 0;
595 return __nd_alloc_stack(nd);
596}
597
Al Viro79733872015-05-09 12:55:43 -0400598static void drop_links(struct nameidata *nd)
599{
600 int i = nd->depth;
601 while (i--) {
602 struct saved *last = nd->stack + i;
Al Virofceef392015-12-29 15:58:39 -0500603 do_delayed_call(&last->done);
604 clear_delayed_call(&last->done);
Al Viro79733872015-05-09 12:55:43 -0400605 }
606}
607
608static void terminate_walk(struct nameidata *nd)
609{
610 drop_links(nd);
611 if (!(nd->flags & LOOKUP_RCU)) {
612 int i;
613 path_put(&nd->path);
614 for (i = 0; i < nd->depth; i++)
615 path_put(&nd->stack[i].link);
Al Viro102b8af2015-05-12 17:35:52 -0400616 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
617 path_put(&nd->root);
618 nd->root.mnt = NULL;
619 }
Al Viro79733872015-05-09 12:55:43 -0400620 } else {
621 nd->flags &= ~LOOKUP_RCU;
622 if (!(nd->flags & LOOKUP_ROOT))
623 nd->root.mnt = NULL;
624 rcu_read_unlock();
625 }
626 nd->depth = 0;
627}
628
629/* path_put is needed afterwards regardless of success or failure */
630static bool legitimize_path(struct nameidata *nd,
631 struct path *path, unsigned seq)
632{
633 int res = __legitimize_mnt(path->mnt, nd->m_seq);
634 if (unlikely(res)) {
635 if (res > 0)
636 path->mnt = NULL;
637 path->dentry = NULL;
638 return false;
639 }
640 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
641 path->dentry = NULL;
642 return false;
643 }
644 return !read_seqcount_retry(&path->dentry->d_seq, seq);
645}
646
647static bool legitimize_links(struct nameidata *nd)
648{
649 int i;
650 for (i = 0; i < nd->depth; i++) {
651 struct saved *last = nd->stack + i;
652 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
653 drop_links(nd);
654 nd->depth = i + 1;
655 return false;
656 }
657 }
658 return true;
659}
660
Al Viro19660af2011-03-25 10:32:48 -0400661/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100662 * Path walking has 2 modes, rcu-walk and ref-walk (see
Al Viro19660af2011-03-25 10:32:48 -0400663 * Documentation/filesystems/path-lookup.txt). In situations when we can't
664 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
Mike Marshall57e37152015-11-30 11:11:59 -0500665 * normal reference counts on dentries and vfsmounts to transition to ref-walk
Al Viro19660af2011-03-25 10:32:48 -0400666 * mode. Refcounts are grabbed at the last known good point before rcu-walk
667 * got stuck, so ref-walk may continue from there. If this is not successful
668 * (eg. a seqcount has changed), then failure is returned and it's up to caller
669 * to restart the path walk from the beginning in ref-walk mode.
Nick Piggin31e6b012011-01-07 17:49:52 +1100670 */
Nick Piggin31e6b012011-01-07 17:49:52 +1100671
672/**
Al Viro19660af2011-03-25 10:32:48 -0400673 * unlazy_walk - try to switch to ref-walk mode.
674 * @nd: nameidata pathwalk data
675 * @dentry: child of nd->path.dentry or NULL
Al Viro6e9918b2015-05-05 09:26:05 -0400676 * @seq: seq number to check dentry against
Randy Dunlap39191622011-01-08 19:36:21 -0800677 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100678 *
Al Viro19660af2011-03-25 10:32:48 -0400679 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
680 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
681 * @nd or NULL. Must be called from rcu-walk context.
Al Viro79733872015-05-09 12:55:43 -0400682 * Nothing should touch nameidata between unlazy_walk() failure and
683 * terminate_walk().
Nick Piggin31e6b012011-01-07 17:49:52 +1100684 */
Al Viro6e9918b2015-05-05 09:26:05 -0400685static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq)
Nick Piggin31e6b012011-01-07 17:49:52 +1100686{
Nick Piggin31e6b012011-01-07 17:49:52 +1100687 struct dentry *parent = nd->path.dentry;
688
689 BUG_ON(!(nd->flags & LOOKUP_RCU));
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700690
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700691 nd->flags &= ~LOOKUP_RCU;
Al Viro79733872015-05-09 12:55:43 -0400692 if (unlikely(!legitimize_links(nd)))
693 goto out2;
694 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
695 goto out2;
696 if (unlikely(!lockref_get_not_dead(&parent->d_lockref)))
697 goto out1;
Al Viro48a066e2013-09-29 22:06:07 -0400698
Linus Torvalds15570082013-09-02 11:38:06 -0700699 /*
700 * For a negative lookup, the lookup sequence point is the parents
701 * sequence point, and it only needs to revalidate the parent dentry.
702 *
703 * For a positive lookup, we need to move both the parent and the
704 * dentry from the RCU domain to be properly refcounted. And the
705 * sequence number in the dentry validates *both* dentry counters,
706 * since we checked the sequence number of the parent after we got
707 * the child sequence number. So we know the parent must still
708 * be valid if the child sequence number is still valid.
709 */
Al Viro19660af2011-03-25 10:32:48 -0400710 if (!dentry) {
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700711 if (read_seqcount_retry(&parent->d_seq, nd->seq))
712 goto out;
Al Viro19660af2011-03-25 10:32:48 -0400713 BUG_ON(nd->inode != parent->d_inode);
714 } else {
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700715 if (!lockref_get_not_dead(&dentry->d_lockref))
716 goto out;
Al Viro6e9918b2015-05-05 09:26:05 -0400717 if (read_seqcount_retry(&dentry->d_seq, seq))
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700718 goto drop_dentry;
Al Viro19660af2011-03-25 10:32:48 -0400719 }
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700720
721 /*
722 * Sequence counts matched. Now make sure that the root is
723 * still valid and get it if required.
724 */
725 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
Al Viro5a8d87e2015-05-12 00:10:18 -0400726 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
727 rcu_read_unlock();
728 dput(dentry);
729 return -ECHILD;
Al Viro79733872015-05-09 12:55:43 -0400730 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100731 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100732
Al Viro8b61e742013-11-08 12:45:01 -0500733 rcu_read_unlock();
Nick Piggin31e6b012011-01-07 17:49:52 +1100734 return 0;
Al Viro19660af2011-03-25 10:32:48 -0400735
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700736drop_dentry:
Al Viro8b61e742013-11-08 12:45:01 -0500737 rcu_read_unlock();
Linus Torvalds15570082013-09-02 11:38:06 -0700738 dput(dentry);
Linus Torvaldsd0d27272013-09-10 12:17:49 -0700739 goto drop_root_mnt;
Al Viro79733872015-05-09 12:55:43 -0400740out2:
741 nd->path.mnt = NULL;
742out1:
743 nd->path.dentry = NULL;
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700744out:
Al Viro8b61e742013-11-08 12:45:01 -0500745 rcu_read_unlock();
Linus Torvaldsd0d27272013-09-10 12:17:49 -0700746drop_root_mnt:
747 if (!(nd->flags & LOOKUP_ROOT))
748 nd->root.mnt = NULL;
Nick Piggin31e6b012011-01-07 17:49:52 +1100749 return -ECHILD;
750}
751
Al Viro79733872015-05-09 12:55:43 -0400752static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq)
753{
754 if (unlikely(!legitimize_path(nd, link, seq))) {
755 drop_links(nd);
756 nd->depth = 0;
757 nd->flags &= ~LOOKUP_RCU;
758 nd->path.mnt = NULL;
759 nd->path.dentry = NULL;
760 if (!(nd->flags & LOOKUP_ROOT))
761 nd->root.mnt = NULL;
762 rcu_read_unlock();
763 } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) {
764 return 0;
765 }
766 path_put(link);
767 return -ECHILD;
768}
769
Al Viro4ce16ef32012-06-10 16:10:59 -0400770static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
Nick Piggin34286d62011-01-07 17:49:57 +1100771{
Al Viro4ce16ef32012-06-10 16:10:59 -0400772 return dentry->d_op->d_revalidate(dentry, flags);
Nick Piggin34286d62011-01-07 17:49:57 +1100773}
774
Al Viro9f1fafe2011-03-25 11:00:12 -0400775/**
776 * complete_walk - successful completion of path walk
777 * @nd: pointer nameidata
Jeff Layton39159de2009-12-07 12:01:50 -0500778 *
Al Viro9f1fafe2011-03-25 11:00:12 -0400779 * If we had been in RCU mode, drop out of it and legitimize nd->path.
780 * Revalidate the final result, unless we'd already done that during
781 * the path walk or the filesystem doesn't ask for it. Return 0 on
782 * success, -error on failure. In case of failure caller does not
783 * need to drop nd->path.
Jeff Layton39159de2009-12-07 12:01:50 -0500784 */
Al Viro9f1fafe2011-03-25 11:00:12 -0400785static int complete_walk(struct nameidata *nd)
Jeff Layton39159de2009-12-07 12:01:50 -0500786{
Al Viro16c2cd72011-02-22 15:50:10 -0500787 struct dentry *dentry = nd->path.dentry;
Jeff Layton39159de2009-12-07 12:01:50 -0500788 int status;
Jeff Layton39159de2009-12-07 12:01:50 -0500789
Al Viro9f1fafe2011-03-25 11:00:12 -0400790 if (nd->flags & LOOKUP_RCU) {
Al Viro9f1fafe2011-03-25 11:00:12 -0400791 if (!(nd->flags & LOOKUP_ROOT))
792 nd->root.mnt = NULL;
Al Viro6e9918b2015-05-05 09:26:05 -0400793 if (unlikely(unlazy_walk(nd, NULL, 0)))
Al Viro48a066e2013-09-29 22:06:07 -0400794 return -ECHILD;
Al Viro9f1fafe2011-03-25 11:00:12 -0400795 }
796
Al Viro16c2cd72011-02-22 15:50:10 -0500797 if (likely(!(nd->flags & LOOKUP_JUMPED)))
Jeff Layton39159de2009-12-07 12:01:50 -0500798 return 0;
799
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500800 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
Al Viro16c2cd72011-02-22 15:50:10 -0500801 return 0;
802
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500803 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
Jeff Layton39159de2009-12-07 12:01:50 -0500804 if (status > 0)
805 return 0;
806
Al Viro16c2cd72011-02-22 15:50:10 -0500807 if (!status)
Jeff Layton39159de2009-12-07 12:01:50 -0500808 status = -ESTALE;
Al Viro16c2cd72011-02-22 15:50:10 -0500809
Jeff Layton39159de2009-12-07 12:01:50 -0500810 return status;
811}
812
Al Viro18d8c862015-05-12 16:32:34 -0400813static void set_root(struct nameidata *nd)
Al Viro2a737872009-04-07 11:49:53 -0400814{
Al Viro7bd88372014-09-13 21:55:46 -0400815 struct fs_struct *fs = current->fs;
Nick Pigginc28cc362011-01-07 17:49:53 +1100816
Al Viro9e6697e2015-12-05 20:07:21 -0500817 if (nd->flags & LOOKUP_RCU) {
818 unsigned seq;
819
820 do {
821 seq = read_seqcount_begin(&fs->seq);
822 nd->root = fs->root;
823 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
824 } while (read_seqcount_retry(&fs->seq, seq));
825 } else {
826 get_fs_root(fs, &nd->root);
827 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100828}
829
Jan Blunck1d957f92008-02-14 19:34:35 -0800830static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700831{
832 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800833 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700834 mntput(path->mnt);
835}
836
Nick Piggin7b9337a2011-01-14 08:42:43 +0000837static inline void path_to_nameidata(const struct path *path,
838 struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700839{
Nick Piggin31e6b012011-01-07 17:49:52 +1100840 if (!(nd->flags & LOOKUP_RCU)) {
841 dput(nd->path.dentry);
842 if (nd->path.mnt != path->mnt)
843 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800844 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100845 nd->path.mnt = path->mnt;
Jan Blunck4ac91372008-02-14 19:34:32 -0800846 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700847}
848
Al Viro248fb5b2015-12-05 20:51:58 -0500849static int nd_jump_root(struct nameidata *nd)
850{
851 if (nd->flags & LOOKUP_RCU) {
852 struct dentry *d;
853 nd->path = nd->root;
854 d = nd->path.dentry;
855 nd->inode = d->d_inode;
856 nd->seq = nd->root_seq;
857 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
858 return -ECHILD;
859 } else {
860 path_put(&nd->path);
861 nd->path = nd->root;
862 path_get(&nd->path);
863 nd->inode = nd->path.dentry->d_inode;
864 }
865 nd->flags |= LOOKUP_JUMPED;
866 return 0;
867}
868
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400869/*
Al Viro6b255392015-11-17 10:20:54 -0500870 * Helper to directly jump to a known parsed path from ->get_link,
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400871 * caller must have taken a reference to path beforehand.
872 */
Al Viro6e771372015-05-02 13:37:52 -0400873void nd_jump_link(struct path *path)
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400874{
Al Viro6e771372015-05-02 13:37:52 -0400875 struct nameidata *nd = current->nameidata;
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400876 path_put(&nd->path);
877
878 nd->path = *path;
879 nd->inode = nd->path.dentry->d_inode;
880 nd->flags |= LOOKUP_JUMPED;
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400881}
882
Al Virob9ff4422015-05-02 20:19:23 -0400883static inline void put_link(struct nameidata *nd)
Al Viro574197e2011-03-14 22:20:34 -0400884{
Al Viro21c30032015-05-03 21:06:24 -0400885 struct saved *last = nd->stack + --nd->depth;
Al Virofceef392015-12-29 15:58:39 -0500886 do_delayed_call(&last->done);
Al Viro6548fae2015-05-07 20:32:22 -0400887 if (!(nd->flags & LOOKUP_RCU))
888 path_put(&last->link);
Al Viro574197e2011-03-14 22:20:34 -0400889}
890
Linus Torvalds561ec642012-10-26 10:05:07 -0700891int sysctl_protected_symlinks __read_mostly = 0;
892int sysctl_protected_hardlinks __read_mostly = 0;
Kees Cook800179c2012-07-25 17:29:07 -0700893
894/**
895 * may_follow_link - Check symlink following for unsafe situations
Randy Dunlap55852632012-08-18 17:39:25 -0700896 * @nd: nameidata pathwalk data
Kees Cook800179c2012-07-25 17:29:07 -0700897 *
898 * In the case of the sysctl_protected_symlinks sysctl being enabled,
899 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
900 * in a sticky world-writable directory. This is to protect privileged
901 * processes from failing races against path names that may change out
902 * from under them by way of other users creating malicious symlinks.
903 * It will permit symlinks to be followed only when outside a sticky
904 * world-writable directory, or when the uid of the symlink and follower
905 * match, or when the directory owner matches the symlink's owner.
906 *
907 * Returns 0 if following the symlink is allowed, -ve on error.
908 */
Al Virofec2fa22015-05-06 15:58:18 -0400909static inline int may_follow_link(struct nameidata *nd)
Kees Cook800179c2012-07-25 17:29:07 -0700910{
911 const struct inode *inode;
912 const struct inode *parent;
Seth Forshee2d7f9e22016-04-26 14:36:23 -0500913 kuid_t puid;
Kees Cook800179c2012-07-25 17:29:07 -0700914
915 if (!sysctl_protected_symlinks)
916 return 0;
917
918 /* Allowed if owner and follower match. */
Al Virofceef392015-12-29 15:58:39 -0500919 inode = nd->link_inode;
Eric W. Biederman81abe272012-08-03 09:38:08 -0700920 if (uid_eq(current_cred()->fsuid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700921 return 0;
922
923 /* Allowed if parent directory not sticky and world-writable. */
Al Viroaa65fa32015-08-04 23:23:50 -0400924 parent = nd->inode;
Kees Cook800179c2012-07-25 17:29:07 -0700925 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
926 return 0;
927
928 /* Allowed if parent directory and link owner match. */
Seth Forshee2d7f9e22016-04-26 14:36:23 -0500929 puid = parent->i_uid;
930 if (uid_valid(puid) && uid_eq(puid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700931 return 0;
932
Al Viro319565022015-05-07 20:37:40 -0400933 if (nd->flags & LOOKUP_RCU)
934 return -ECHILD;
935
Al Viro1cf26652015-05-06 16:01:56 -0400936 audit_log_link_denied("follow_link", &nd->stack[0].link);
Kees Cook800179c2012-07-25 17:29:07 -0700937 return -EACCES;
938}
939
940/**
941 * safe_hardlink_source - Check for safe hardlink conditions
942 * @inode: the source inode to hardlink from
943 *
944 * Return false if at least one of the following conditions:
945 * - inode is not a regular file
946 * - inode is setuid
947 * - inode is setgid and group-exec
948 * - access failure for read and write
949 *
950 * Otherwise returns true.
951 */
952static bool safe_hardlink_source(struct inode *inode)
953{
954 umode_t mode = inode->i_mode;
955
956 /* Special files should not get pinned to the filesystem. */
957 if (!S_ISREG(mode))
958 return false;
959
960 /* Setuid files should not get pinned to the filesystem. */
961 if (mode & S_ISUID)
962 return false;
963
964 /* Executable setgid files should not get pinned to the filesystem. */
965 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
966 return false;
967
968 /* Hardlinking to unreadable or unwritable sources is dangerous. */
969 if (inode_permission(inode, MAY_READ | MAY_WRITE))
970 return false;
971
972 return true;
973}
974
975/**
976 * may_linkat - Check permissions for creating a hardlink
977 * @link: the source to hardlink from
978 *
979 * Block hardlink when all of:
980 * - sysctl_protected_hardlinks enabled
981 * - fsuid does not match inode
982 * - hardlink source is unsafe (see safe_hardlink_source() above)
Dirk Steinmetzf2ca3792015-10-20 16:09:19 +0200983 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
Kees Cook800179c2012-07-25 17:29:07 -0700984 *
985 * Returns 0 if successful, -ve on error.
986 */
987static int may_linkat(struct path *link)
988{
Kees Cook800179c2012-07-25 17:29:07 -0700989 struct inode *inode;
990
991 if (!sysctl_protected_hardlinks)
992 return 0;
993
Kees Cook800179c2012-07-25 17:29:07 -0700994 inode = link->dentry->d_inode;
995
996 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
997 * otherwise, it must be a safe source.
998 */
Dirk Steinmetzf2ca3792015-10-20 16:09:19 +0200999 if (inode_owner_or_capable(inode) || safe_hardlink_source(inode))
Kees Cook800179c2012-07-25 17:29:07 -07001000 return 0;
1001
Kees Cooka51d9ea2012-07-25 17:29:08 -07001002 audit_log_link_denied("linkat", link);
Kees Cook800179c2012-07-25 17:29:07 -07001003 return -EPERM;
1004}
1005
Al Viro3b2e7f72015-04-19 00:53:50 -04001006static __always_inline
1007const char *get_link(struct nameidata *nd)
Ian Kent051d3812006-03-27 01:14:53 -08001008{
Al Viroab104922015-05-10 11:50:01 -04001009 struct saved *last = nd->stack + nd->depth - 1;
Al Viro1cf26652015-05-06 16:01:56 -04001010 struct dentry *dentry = last->link.dentry;
Al Virofceef392015-12-29 15:58:39 -05001011 struct inode *inode = nd->link_inode;
Al Viro6d7b5aa2012-06-10 04:15:17 -04001012 int error;
Al Viro0a959df2015-04-18 18:23:41 -04001013 const char *res;
Ian Kent051d3812006-03-27 01:14:53 -08001014
NeilBrown8fa9dd22015-03-23 13:37:40 +11001015 if (!(nd->flags & LOOKUP_RCU)) {
1016 touch_atime(&last->link);
1017 cond_resched();
Miklos Szeredi598e3c82016-09-16 12:44:20 +02001018 } else if (atime_needs_update_rcu(&last->link, inode)) {
NeilBrown8fa9dd22015-03-23 13:37:40 +11001019 if (unlikely(unlazy_walk(nd, NULL, 0)))
1020 return ERR_PTR(-ECHILD);
1021 touch_atime(&last->link);
1022 }
1023
NeilBrownbda0be72015-03-23 13:37:39 +11001024 error = security_inode_follow_link(dentry, inode,
1025 nd->flags & LOOKUP_RCU);
1026 if (unlikely(error))
Al Viro6920a442015-05-10 10:43:46 -04001027 return ERR_PTR(error);
Al Viro36f3b4f2011-02-22 21:24:38 -05001028
Al Viro86acdca12009-12-22 23:45:11 -05001029 nd->last_type = LAST_BIND;
Al Virod4dee482015-04-30 20:08:02 -04001030 res = inode->i_link;
1031 if (!res) {
Al Virofceef392015-12-29 15:58:39 -05001032 const char * (*get)(struct dentry *, struct inode *,
1033 struct delayed_call *);
1034 get = inode->i_op->get_link;
Al Viro8c1b4562015-05-09 18:15:21 -04001035 if (nd->flags & LOOKUP_RCU) {
Al Virofceef392015-12-29 15:58:39 -05001036 res = get(NULL, inode, &last->done);
Al Viro6b255392015-11-17 10:20:54 -05001037 if (res == ERR_PTR(-ECHILD)) {
1038 if (unlikely(unlazy_walk(nd, NULL, 0)))
1039 return ERR_PTR(-ECHILD);
Al Virofceef392015-12-29 15:58:39 -05001040 res = get(dentry, inode, &last->done);
Al Viro6b255392015-11-17 10:20:54 -05001041 }
1042 } else {
Al Virofceef392015-12-29 15:58:39 -05001043 res = get(dentry, inode, &last->done);
Al Viro8c1b4562015-05-09 18:15:21 -04001044 }
Al Virofceef392015-12-29 15:58:39 -05001045 if (IS_ERR_OR_NULL(res))
Al Virofab51e82015-05-10 11:01:00 -04001046 return res;
Ian Kent051d3812006-03-27 01:14:53 -08001047 }
Al Virofab51e82015-05-10 11:01:00 -04001048 if (*res == '/') {
Al Viro9e6697e2015-12-05 20:07:21 -05001049 if (!nd->root.mnt)
1050 set_root(nd);
Al Viro248fb5b2015-12-05 20:51:58 -05001051 if (unlikely(nd_jump_root(nd)))
1052 return ERR_PTR(-ECHILD);
Al Virofab51e82015-05-10 11:01:00 -04001053 while (unlikely(*++res == '/'))
1054 ;
1055 }
1056 if (!*res)
1057 res = NULL;
Al Viro0a959df2015-04-18 18:23:41 -04001058 return res;
1059}
Al Viro6d7b5aa2012-06-10 04:15:17 -04001060
David Howellsf015f1262012-06-25 12:55:28 +01001061/*
1062 * follow_up - Find the mountpoint of path's vfsmount
1063 *
1064 * Given a path, find the mountpoint of its source file system.
1065 * Replace @path with the path of the mountpoint in the parent mount.
1066 * Up is towards /.
1067 *
1068 * Return 1 if we went up a level and 0 if we were already at the
1069 * root.
1070 */
Al Virobab77eb2009-04-18 03:26:48 -04001071int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072{
Al Viro0714a532011-11-24 22:19:58 -05001073 struct mount *mnt = real_mount(path->mnt);
1074 struct mount *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +10001076
Al Viro48a066e2013-09-29 22:06:07 -04001077 read_seqlock_excl(&mount_lock);
Al Viro0714a532011-11-24 22:19:58 -05001078 parent = mnt->mnt_parent;
Al Viro3c0a6162012-07-18 17:32:50 +04001079 if (parent == mnt) {
Al Viro48a066e2013-09-29 22:06:07 -04001080 read_sequnlock_excl(&mount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 return 0;
1082 }
Al Viro0714a532011-11-24 22:19:58 -05001083 mntget(&parent->mnt);
Al Viroa73324d2011-11-24 22:25:07 -05001084 mountpoint = dget(mnt->mnt_mountpoint);
Al Viro48a066e2013-09-29 22:06:07 -04001085 read_sequnlock_excl(&mount_lock);
Al Virobab77eb2009-04-18 03:26:48 -04001086 dput(path->dentry);
1087 path->dentry = mountpoint;
1088 mntput(path->mnt);
Al Viro0714a532011-11-24 22:19:58 -05001089 path->mnt = &parent->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 return 1;
1091}
Al Viro4d359502014-03-14 12:20:17 -04001092EXPORT_SYMBOL(follow_up);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001094/*
David Howells9875cf82011-01-14 18:45:21 +00001095 * Perform an automount
1096 * - return -EISDIR to tell follow_managed() to stop and return the path we
1097 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 */
NeilBrown756daf22015-03-23 13:37:38 +11001099static int follow_automount(struct path *path, struct nameidata *nd,
David Howells9875cf82011-01-14 18:45:21 +00001100 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +11001101{
David Howells9875cf82011-01-14 18:45:21 +00001102 struct vfsmount *mnt;
David Howellsea5b7782011-01-14 19:10:03 +00001103 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001104
David Howells9875cf82011-01-14 18:45:21 +00001105 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1106 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -07001107
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +02001108 /* We don't want to mount if someone's just doing a stat -
1109 * unless they're stat'ing a directory and appended a '/' to
1110 * the name.
1111 *
1112 * We do, however, want to mount if someone wants to open or
1113 * create a file of any type under the mountpoint, wants to
1114 * traverse through the mountpoint or wants to open the
1115 * mounted directory. Also, autofs may mark negative dentries
1116 * as being automount points. These will need the attentions
1117 * of the daemon to instantiate them before they can be used.
David Howells9875cf82011-01-14 18:45:21 +00001118 */
NeilBrown756daf22015-03-23 13:37:38 +11001119 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1120 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +02001121 path->dentry->d_inode)
1122 return -EISDIR;
1123
Eric W. Biedermanaeaa4a72016-07-23 11:20:44 -05001124 if (path->dentry->d_sb->s_user_ns != &init_user_ns)
1125 return -EACCES;
1126
NeilBrown756daf22015-03-23 13:37:38 +11001127 nd->total_link_count++;
1128 if (nd->total_link_count >= 40)
David Howells9875cf82011-01-14 18:45:21 +00001129 return -ELOOP;
1130
1131 mnt = path->dentry->d_op->d_automount(path);
1132 if (IS_ERR(mnt)) {
1133 /*
1134 * The filesystem is allowed to return -EISDIR here to indicate
1135 * it doesn't want to automount. For instance, autofs would do
1136 * this so that its userspace daemon can mount on this dentry.
1137 *
1138 * However, we can only permit this if it's a terminal point in
1139 * the path being looked up; if it wasn't then the remainder of
1140 * the path is inaccessible and we should say so.
1141 */
NeilBrown756daf22015-03-23 13:37:38 +11001142 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
David Howells9875cf82011-01-14 18:45:21 +00001143 return -EREMOTE;
1144 return PTR_ERR(mnt);
1145 }
David Howellsea5b7782011-01-14 19:10:03 +00001146
David Howells9875cf82011-01-14 18:45:21 +00001147 if (!mnt) /* mount collision */
1148 return 0;
1149
Al Viro8aef1882011-06-16 15:10:06 +01001150 if (!*need_mntput) {
1151 /* lock_mount() may release path->mnt on error */
1152 mntget(path->mnt);
1153 *need_mntput = true;
1154 }
Al Viro19a167a2011-01-17 01:35:23 -05001155 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +00001156
David Howellsea5b7782011-01-14 19:10:03 +00001157 switch (err) {
1158 case -EBUSY:
1159 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -05001160 return 0;
David Howellsea5b7782011-01-14 19:10:03 +00001161 case 0:
Al Viro8aef1882011-06-16 15:10:06 +01001162 path_put(path);
David Howellsea5b7782011-01-14 19:10:03 +00001163 path->mnt = mnt;
1164 path->dentry = dget(mnt->mnt_root);
David Howellsea5b7782011-01-14 19:10:03 +00001165 return 0;
Al Viro19a167a2011-01-17 01:35:23 -05001166 default:
1167 return err;
David Howellsea5b7782011-01-14 19:10:03 +00001168 }
Al Viro19a167a2011-01-17 01:35:23 -05001169
David Howells9875cf82011-01-14 18:45:21 +00001170}
1171
1172/*
1173 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +00001174 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +00001175 * - Flagged as mountpoint
1176 * - Flagged as automount point
1177 *
1178 * This may only be called in refwalk mode.
1179 *
1180 * Serialization is taken care of in namespace.c
1181 */
NeilBrown756daf22015-03-23 13:37:38 +11001182static int follow_managed(struct path *path, struct nameidata *nd)
David Howells9875cf82011-01-14 18:45:21 +00001183{
Al Viro8aef1882011-06-16 15:10:06 +01001184 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
David Howells9875cf82011-01-14 18:45:21 +00001185 unsigned managed;
1186 bool need_mntput = false;
Al Viro8aef1882011-06-16 15:10:06 +01001187 int ret = 0;
David Howells9875cf82011-01-14 18:45:21 +00001188
1189 /* Given that we're not holding a lock here, we retain the value in a
1190 * local variable for each dentry as we look at it so that we don't see
1191 * the components of that value change under us */
1192 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1193 managed &= DCACHE_MANAGED_DENTRY,
1194 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +00001195 /* Allow the filesystem to manage the transit without i_mutex
1196 * being held. */
1197 if (managed & DCACHE_MANAGE_TRANSIT) {
1198 BUG_ON(!path->dentry->d_op);
1199 BUG_ON(!path->dentry->d_op->d_manage);
Al Viro1aed3e42011-03-18 09:09:02 -04001200 ret = path->dentry->d_op->d_manage(path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001201 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001202 break;
David Howellscc53ce52011-01-14 18:45:26 +00001203 }
1204
David Howells9875cf82011-01-14 18:45:21 +00001205 /* Transit to a mounted filesystem. */
1206 if (managed & DCACHE_MOUNTED) {
1207 struct vfsmount *mounted = lookup_mnt(path);
1208 if (mounted) {
1209 dput(path->dentry);
1210 if (need_mntput)
1211 mntput(path->mnt);
1212 path->mnt = mounted;
1213 path->dentry = dget(mounted->mnt_root);
1214 need_mntput = true;
1215 continue;
1216 }
1217
1218 /* Something is mounted on this dentry in another
1219 * namespace and/or whatever was mounted there in this
Al Viro48a066e2013-09-29 22:06:07 -04001220 * namespace got unmounted before lookup_mnt() could
1221 * get it */
David Howells9875cf82011-01-14 18:45:21 +00001222 }
1223
1224 /* Handle an automount point */
1225 if (managed & DCACHE_NEED_AUTOMOUNT) {
NeilBrown756daf22015-03-23 13:37:38 +11001226 ret = follow_automount(path, nd, &need_mntput);
David Howells9875cf82011-01-14 18:45:21 +00001227 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001228 break;
David Howells9875cf82011-01-14 18:45:21 +00001229 continue;
1230 }
1231
1232 /* We didn't change the current path point */
1233 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 }
Al Viro8aef1882011-06-16 15:10:06 +01001235
1236 if (need_mntput && path->mnt == mnt)
1237 mntput(path->mnt);
Al Viroe9742b52016-03-05 22:04:59 -05001238 if (ret == -EISDIR || !ret)
1239 ret = 1;
Al Viro84027522015-04-22 10:30:08 -04001240 if (need_mntput)
1241 nd->flags |= LOOKUP_JUMPED;
1242 if (unlikely(ret < 0))
1243 path_put_conditional(path, nd);
1244 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245}
1246
David Howellscc53ce52011-01-14 18:45:26 +00001247int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248{
1249 struct vfsmount *mounted;
1250
Al Viro1c755af2009-04-18 14:06:57 -04001251 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -04001253 dput(path->dentry);
1254 mntput(path->mnt);
1255 path->mnt = mounted;
1256 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 return 1;
1258 }
1259 return 0;
1260}
Al Viro4d359502014-03-14 12:20:17 -04001261EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262
NeilBrownb8faf032014-08-04 17:06:29 +10001263static inline int managed_dentry_rcu(struct dentry *dentry)
Ian Kent62a73752011-03-25 01:51:02 +08001264{
NeilBrownb8faf032014-08-04 17:06:29 +10001265 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1266 dentry->d_op->d_manage(dentry, true) : 0;
Ian Kent62a73752011-03-25 01:51:02 +08001267}
1268
David Howells9875cf82011-01-14 18:45:21 +00001269/*
Al Viro287548e2011-05-27 06:50:06 -04001270 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1271 * we meet a managed dentry that would need blocking.
David Howells9875cf82011-01-14 18:45:21 +00001272 */
1273static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
Al Viro254cf582015-05-05 09:40:46 -04001274 struct inode **inode, unsigned *seqp)
David Howells9875cf82011-01-14 18:45:21 +00001275{
Ian Kent62a73752011-03-25 01:51:02 +08001276 for (;;) {
Al Viroc7105362011-11-24 18:22:03 -05001277 struct mount *mounted;
Ian Kent62a73752011-03-25 01:51:02 +08001278 /*
1279 * Don't forget we might have a non-mountpoint managed dentry
1280 * that wants to block transit.
1281 */
NeilBrownb8faf032014-08-04 17:06:29 +10001282 switch (managed_dentry_rcu(path->dentry)) {
1283 case -ECHILD:
1284 default:
David Howellsab909112011-01-14 18:46:51 +00001285 return false;
NeilBrownb8faf032014-08-04 17:06:29 +10001286 case -EISDIR:
1287 return true;
1288 case 0:
1289 break;
1290 }
Ian Kent62a73752011-03-25 01:51:02 +08001291
1292 if (!d_mountpoint(path->dentry))
NeilBrownb8faf032014-08-04 17:06:29 +10001293 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
Ian Kent62a73752011-03-25 01:51:02 +08001294
Al Viro474279d2013-10-01 16:11:26 -04001295 mounted = __lookup_mnt(path->mnt, path->dentry);
David Howells9875cf82011-01-14 18:45:21 +00001296 if (!mounted)
1297 break;
Al Viroc7105362011-11-24 18:22:03 -05001298 path->mnt = &mounted->mnt;
1299 path->dentry = mounted->mnt.mnt_root;
Al Viroa3fbbde2011-11-07 21:21:26 +00001300 nd->flags |= LOOKUP_JUMPED;
Al Viro254cf582015-05-05 09:40:46 -04001301 *seqp = read_seqcount_begin(&path->dentry->d_seq);
Linus Torvalds59430262011-07-18 15:43:29 -07001302 /*
1303 * Update the inode too. We don't need to re-check the
1304 * dentry sequence number here after this d_inode read,
1305 * because a mount-point is always pinned.
1306 */
1307 *inode = path->dentry->d_inode;
David Howells9875cf82011-01-14 18:45:21 +00001308 }
Al Virof5be3e29122014-09-13 21:50:45 -04001309 return !read_seqretry(&mount_lock, nd->m_seq) &&
NeilBrownb8faf032014-08-04 17:06:29 +10001310 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
Al Viro287548e2011-05-27 06:50:06 -04001311}
1312
Nick Piggin31e6b012011-01-07 17:49:52 +11001313static int follow_dotdot_rcu(struct nameidata *nd)
1314{
Al Viro4023bfc2014-09-13 21:59:43 -04001315 struct inode *inode = nd->inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001316
David Howells9875cf82011-01-14 18:45:21 +00001317 while (1) {
Al Viroaed434a2015-05-12 12:22:47 -04001318 if (path_equal(&nd->path, &nd->root))
Nick Piggin31e6b012011-01-07 17:49:52 +11001319 break;
Nick Piggin31e6b012011-01-07 17:49:52 +11001320 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1321 struct dentry *old = nd->path.dentry;
1322 struct dentry *parent = old->d_parent;
1323 unsigned seq;
1324
Al Viro4023bfc2014-09-13 21:59:43 -04001325 inode = parent->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001326 seq = read_seqcount_begin(&parent->d_seq);
Al Viroaed434a2015-05-12 12:22:47 -04001327 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1328 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001329 nd->path.dentry = parent;
1330 nd->seq = seq;
Eric W. Biederman397d4252015-08-15 20:27:13 -05001331 if (unlikely(!path_connected(&nd->path)))
1332 return -ENOENT;
Nick Piggin31e6b012011-01-07 17:49:52 +11001333 break;
Al Viroaed434a2015-05-12 12:22:47 -04001334 } else {
1335 struct mount *mnt = real_mount(nd->path.mnt);
1336 struct mount *mparent = mnt->mnt_parent;
1337 struct dentry *mountpoint = mnt->mnt_mountpoint;
1338 struct inode *inode2 = mountpoint->d_inode;
1339 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1340 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1341 return -ECHILD;
1342 if (&mparent->mnt == nd->path.mnt)
1343 break;
1344 /* we know that mountpoint was pinned */
1345 nd->path.dentry = mountpoint;
1346 nd->path.mnt = &mparent->mnt;
1347 inode = inode2;
1348 nd->seq = seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11001349 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001350 }
Al Viroaed434a2015-05-12 12:22:47 -04001351 while (unlikely(d_mountpoint(nd->path.dentry))) {
Al Virob37199e2014-03-20 15:18:22 -04001352 struct mount *mounted;
1353 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
Al Viroaed434a2015-05-12 12:22:47 -04001354 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1355 return -ECHILD;
Al Virob37199e2014-03-20 15:18:22 -04001356 if (!mounted)
1357 break;
1358 nd->path.mnt = &mounted->mnt;
1359 nd->path.dentry = mounted->mnt.mnt_root;
Al Viro4023bfc2014-09-13 21:59:43 -04001360 inode = nd->path.dentry->d_inode;
Al Virob37199e2014-03-20 15:18:22 -04001361 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Virob37199e2014-03-20 15:18:22 -04001362 }
Al Viro4023bfc2014-09-13 21:59:43 -04001363 nd->inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001364 return 0;
1365}
1366
David Howells9875cf82011-01-14 18:45:21 +00001367/*
David Howellscc53ce52011-01-14 18:45:26 +00001368 * Follow down to the covering mount currently visible to userspace. At each
1369 * point, the filesystem owning that dentry may be queried as to whether the
1370 * caller is permitted to proceed or not.
David Howellscc53ce52011-01-14 18:45:26 +00001371 */
Al Viro7cc90cc2011-03-18 09:04:20 -04001372int follow_down(struct path *path)
David Howellscc53ce52011-01-14 18:45:26 +00001373{
1374 unsigned managed;
1375 int ret;
1376
1377 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1378 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1379 /* Allow the filesystem to manage the transit without i_mutex
1380 * being held.
1381 *
1382 * We indicate to the filesystem if someone is trying to mount
1383 * something here. This gives autofs the chance to deny anyone
1384 * other than its daemon the right to mount on its
1385 * superstructure.
1386 *
1387 * The filesystem may sleep at this point.
1388 */
1389 if (managed & DCACHE_MANAGE_TRANSIT) {
1390 BUG_ON(!path->dentry->d_op);
1391 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +00001392 ret = path->dentry->d_op->d_manage(
Al Viro1aed3e42011-03-18 09:09:02 -04001393 path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001394 if (ret < 0)
1395 return ret == -EISDIR ? 0 : ret;
1396 }
1397
1398 /* Transit to a mounted filesystem. */
1399 if (managed & DCACHE_MOUNTED) {
1400 struct vfsmount *mounted = lookup_mnt(path);
1401 if (!mounted)
1402 break;
1403 dput(path->dentry);
1404 mntput(path->mnt);
1405 path->mnt = mounted;
1406 path->dentry = dget(mounted->mnt_root);
1407 continue;
1408 }
1409
1410 /* Don't handle automount points here */
1411 break;
1412 }
1413 return 0;
1414}
Al Viro4d359502014-03-14 12:20:17 -04001415EXPORT_SYMBOL(follow_down);
David Howellscc53ce52011-01-14 18:45:26 +00001416
1417/*
David Howells9875cf82011-01-14 18:45:21 +00001418 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1419 */
1420static void follow_mount(struct path *path)
1421{
1422 while (d_mountpoint(path->dentry)) {
1423 struct vfsmount *mounted = lookup_mnt(path);
1424 if (!mounted)
1425 break;
1426 dput(path->dentry);
1427 mntput(path->mnt);
1428 path->mnt = mounted;
1429 path->dentry = dget(mounted->mnt_root);
1430 }
1431}
1432
Eric W. Biedermaneedf2652016-06-02 10:29:47 -05001433static int path_parent_directory(struct path *path)
1434{
1435 struct dentry *old = path->dentry;
1436 /* rare case of legitimate dget_parent()... */
1437 path->dentry = dget_parent(path->dentry);
1438 dput(old);
1439 if (unlikely(!path_connected(path)))
1440 return -ENOENT;
1441 return 0;
1442}
1443
Eric W. Biederman397d4252015-08-15 20:27:13 -05001444static int follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445{
1446 while(1) {
Al Viro2a737872009-04-07 11:49:53 -04001447 if (nd->path.dentry == nd->root.dentry &&
1448 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 break;
1450 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001451 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Eric W. Biedermaneedf2652016-06-02 10:29:47 -05001452 int ret = path_parent_directory(&nd->path);
1453 if (ret)
1454 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 break;
1456 }
Al Viro3088dd72010-01-30 15:47:29 -05001457 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 }
Al Viro79ed0222009-04-18 13:59:41 -04001460 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001461 nd->inode = nd->path.dentry->d_inode;
Eric W. Biederman397d4252015-08-15 20:27:13 -05001462 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463}
1464
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465/*
Oleg Drokinf4fdace2016-07-07 22:04:04 -04001466 * This looks up the name in dcache and possibly revalidates the found dentry.
1467 * NULL is returned if the dentry does not exist in the cache.
Nick Pigginbaa03892010-08-18 04:37:31 +10001468 */
Al Viroe3c13922016-03-06 14:03:27 -05001469static struct dentry *lookup_dcache(const struct qstr *name,
1470 struct dentry *dir,
Al Viro6c51e512016-03-05 20:09:32 -05001471 unsigned int flags)
Nick Pigginbaa03892010-08-18 04:37:31 +10001472{
Nick Pigginbaa03892010-08-18 04:37:31 +10001473 struct dentry *dentry;
Miklos Szeredibad61182012-03-26 12:54:24 +02001474 int error;
Nick Pigginbaa03892010-08-18 04:37:31 +10001475
Miklos Szeredibad61182012-03-26 12:54:24 +02001476 dentry = d_lookup(dir, name);
1477 if (dentry) {
Jeff Layton39e3c952012-11-28 11:30:53 -05001478 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
Al Viro201f9562012-06-22 12:42:10 +04001479 error = d_revalidate(dentry, flags);
Miklos Szeredibad61182012-03-26 12:54:24 +02001480 if (unlikely(error <= 0)) {
Al Viro74ff0ff2016-03-05 22:37:46 -05001481 if (!error)
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001482 d_invalidate(dentry);
Al Viro74ff0ff2016-03-05 22:37:46 -05001483 dput(dentry);
1484 return ERR_PTR(error);
Miklos Szeredibad61182012-03-26 12:54:24 +02001485 }
1486 }
1487 }
Nick Pigginbaa03892010-08-18 04:37:31 +10001488 return dentry;
1489}
1490
1491/*
J. Bruce Fields13a2c3b2013-10-23 16:09:16 -04001492 * Call i_op->lookup on the dentry. The dentry must be negative and
1493 * unhashed.
Miklos Szeredibad61182012-03-26 12:54:24 +02001494 *
1495 * dir->d_inode->i_mutex must be held
Josef Bacik44396f42011-05-31 11:58:49 -04001496 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001497static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
Al Viro72bd8662012-06-10 17:17:17 -04001498 unsigned int flags)
Josef Bacik44396f42011-05-31 11:58:49 -04001499{
Josef Bacik44396f42011-05-31 11:58:49 -04001500 struct dentry *old;
1501
1502 /* Don't create child dentry for a dead directory. */
Miklos Szeredibad61182012-03-26 12:54:24 +02001503 if (unlikely(IS_DEADDIR(dir))) {
Miklos Szeredie188dc02012-02-03 14:25:18 +01001504 dput(dentry);
Josef Bacik44396f42011-05-31 11:58:49 -04001505 return ERR_PTR(-ENOENT);
Miklos Szeredie188dc02012-02-03 14:25:18 +01001506 }
Josef Bacik44396f42011-05-31 11:58:49 -04001507
Al Viro72bd8662012-06-10 17:17:17 -04001508 old = dir->i_op->lookup(dir, dentry, flags);
Josef Bacik44396f42011-05-31 11:58:49 -04001509 if (unlikely(old)) {
1510 dput(dentry);
1511 dentry = old;
1512 }
1513 return dentry;
1514}
1515
Al Viroe3c13922016-03-06 14:03:27 -05001516static struct dentry *__lookup_hash(const struct qstr *name,
Al Viro72bd8662012-06-10 17:17:17 -04001517 struct dentry *base, unsigned int flags)
Al Viroa3255542012-03-30 14:41:51 -04001518{
Al Viro6c51e512016-03-05 20:09:32 -05001519 struct dentry *dentry = lookup_dcache(name, base, flags);
Al Viroa3255542012-03-30 14:41:51 -04001520
Al Viro6c51e512016-03-05 20:09:32 -05001521 if (dentry)
Miklos Szeredibad61182012-03-26 12:54:24 +02001522 return dentry;
Al Viroa3255542012-03-30 14:41:51 -04001523
Al Viro6c51e512016-03-05 20:09:32 -05001524 dentry = d_alloc(base, name);
1525 if (unlikely(!dentry))
1526 return ERR_PTR(-ENOMEM);
1527
Al Viro72bd8662012-06-10 17:17:17 -04001528 return lookup_real(base->d_inode, dentry, flags);
Al Viroa3255542012-03-30 14:41:51 -04001529}
1530
Al Viroe97cdc82013-01-24 18:16:00 -05001531static int lookup_fast(struct nameidata *nd,
Al Viro254cf582015-05-05 09:40:46 -04001532 struct path *path, struct inode **inode,
1533 unsigned *seqp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534{
Jan Blunck4ac91372008-02-14 19:34:32 -08001535 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001536 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001537 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001538 int err;
1539
Al Viro3cac2602009-08-13 18:27:43 +04001540 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001541 * Rename seqlock is not required here because in the off chance
Al Viro5d0f49c2016-03-05 21:32:53 -05001542 * of a false negative due to a concurrent rename, the caller is
1543 * going to fall back to non-racy lookup.
Nick Pigginb04f7842010-08-18 04:37:34 +10001544 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001545 if (nd->flags & LOOKUP_RCU) {
1546 unsigned seq;
Al Viro766c4cb2015-05-07 19:24:57 -04001547 bool negative;
Linus Torvaldsda53be12013-05-21 15:22:44 -07001548 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
Al Viro5d0f49c2016-03-05 21:32:53 -05001549 if (unlikely(!dentry)) {
1550 if (unlazy_walk(nd, NULL, 0))
1551 return -ECHILD;
Al Viroe9742b52016-03-05 22:04:59 -05001552 return 0;
Al Viro5d0f49c2016-03-05 21:32:53 -05001553 }
Al Viro5a18fff2011-03-11 04:44:53 -05001554
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001555 /*
1556 * This sequence count validates that the inode matches
1557 * the dentry name information from lookup.
1558 */
David Howells63afdfc2015-05-06 15:59:00 +01001559 *inode = d_backing_inode(dentry);
Al Viro766c4cb2015-05-07 19:24:57 -04001560 negative = d_is_negative(dentry);
Al Viro5d0f49c2016-03-05 21:32:53 -05001561 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001562 return -ECHILD;
1563
1564 /*
1565 * This sequence count validates that the parent had no
1566 * changes while we did the lookup of the dentry above.
1567 *
1568 * The memory barrier in read_seqcount_begin of child is
1569 * enough, we can use __read_seqcount_retry here.
1570 */
Al Viro5d0f49c2016-03-05 21:32:53 -05001571 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
Nick Piggin31e6b012011-01-07 17:49:52 +11001572 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001573
Al Viro254cf582015-05-05 09:40:46 -04001574 *seqp = seq;
Al Viro5d0f49c2016-03-05 21:32:53 -05001575 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
Al Viro4ce16ef32012-06-10 16:10:59 -04001576 status = d_revalidate(dentry, nd->flags);
Al Viro5d0f49c2016-03-05 21:32:53 -05001577 if (unlikely(status <= 0)) {
1578 if (unlazy_walk(nd, dentry, seq))
1579 return -ECHILD;
1580 if (status == -ECHILD)
1581 status = d_revalidate(dentry, nd->flags);
1582 } else {
1583 /*
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 Viro5d0f49c2016-03-05 21:32:53 -05001593 if (unlazy_walk(nd, dentry, seq))
1594 return -ECHILD;
Al Viro24643082011-02-15 01:26:22 -05001595 }
Al Viro5a18fff2011-03-11 04:44:53 -05001596 } else {
Al Viroe97cdc82013-01-24 18:16:00 -05001597 dentry = __d_lookup(parent, &nd->last);
Al Viro5d0f49c2016-03-05 21:32:53 -05001598 if (unlikely(!dentry))
Al Viroe9742b52016-03-05 22:04:59 -05001599 return 0;
Al Viro5d0f49c2016-03-05 21:32:53 -05001600 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
1601 status = d_revalidate(dentry, nd->flags);
Nick Piggin31e6b012011-01-07 17:49:52 +11001602 }
Al Viro5a18fff2011-03-11 04:44:53 -05001603 if (unlikely(status <= 0)) {
Al Viroe9742b52016-03-05 22:04:59 -05001604 if (!status)
Al Viro5d0f49c2016-03-05 21:32:53 -05001605 d_invalidate(dentry);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001606 dput(dentry);
Al Viro5d0f49c2016-03-05 21:32:53 -05001607 return status;
Al Viro5a18fff2011-03-11 04:44:53 -05001608 }
Al Viro766c4cb2015-05-07 19:24:57 -04001609 if (unlikely(d_is_negative(dentry))) {
1610 dput(dentry);
1611 return -ENOENT;
1612 }
Al Viro5d0f49c2016-03-05 21:32:53 -05001613
David Howells9875cf82011-01-14 18:45:21 +00001614 path->mnt = mnt;
1615 path->dentry = dentry;
NeilBrown756daf22015-03-23 13:37:38 +11001616 err = follow_managed(path, nd);
Al Viroe9742b52016-03-05 22:04:59 -05001617 if (likely(err > 0))
David Howells63afdfc2015-05-06 15:59:00 +01001618 *inode = d_backing_inode(path->dentry);
Al Viro84027522015-04-22 10:30:08 -04001619 return err;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001620}
1621
1622/* Fast lookup failed, do it the slow way */
Al Viroe3c13922016-03-06 14:03:27 -05001623static struct dentry *lookup_slow(const struct qstr *name,
1624 struct dentry *dir,
1625 unsigned int flags)
Miklos Szeredi697f5142012-05-21 17:30:05 +02001626{
Al Viro94bdd652016-04-15 02:42:04 -04001627 struct dentry *dentry = ERR_PTR(-ENOENT), *old;
Al Viro19363862016-04-14 19:33:34 -04001628 struct inode *inode = dir->d_inode;
Al Virod9171b92016-04-15 03:33:13 -04001629 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
Al Viro19363862016-04-14 19:33:34 -04001630
Al Viro9902af72016-04-15 15:08:36 -04001631 inode_lock_shared(inode);
Al Viro19363862016-04-14 19:33:34 -04001632 /* Don't go there if it's already dead */
Al Viro94bdd652016-04-15 02:42:04 -04001633 if (unlikely(IS_DEADDIR(inode)))
1634 goto out;
1635again:
Al Virod9171b92016-04-15 03:33:13 -04001636 dentry = d_alloc_parallel(dir, name, &wq);
Al Viro94bdd652016-04-15 02:42:04 -04001637 if (IS_ERR(dentry))
1638 goto out;
1639 if (unlikely(!d_in_lookup(dentry))) {
Al Viro949a8522016-03-06 14:20:52 -05001640 if ((dentry->d_flags & DCACHE_OP_REVALIDATE) &&
1641 !(flags & LOOKUP_NO_REVAL)) {
1642 int error = d_revalidate(dentry, flags);
1643 if (unlikely(error <= 0)) {
Al Viro94bdd652016-04-15 02:42:04 -04001644 if (!error) {
Al Viro949a8522016-03-06 14:20:52 -05001645 d_invalidate(dentry);
Al Viro94bdd652016-04-15 02:42:04 -04001646 dput(dentry);
1647 goto again;
1648 }
Al Viro949a8522016-03-06 14:20:52 -05001649 dput(dentry);
1650 dentry = ERR_PTR(error);
1651 }
1652 }
Al Viro94bdd652016-04-15 02:42:04 -04001653 } else {
1654 old = inode->i_op->lookup(inode, dentry, flags);
1655 d_lookup_done(dentry);
1656 if (unlikely(old)) {
1657 dput(dentry);
1658 dentry = old;
Al Viro949a8522016-03-06 14:20:52 -05001659 }
1660 }
Al Viro94bdd652016-04-15 02:42:04 -04001661out:
Al Viro9902af72016-04-15 15:08:36 -04001662 inode_unlock_shared(inode);
Al Viroe3c13922016-03-06 14:03:27 -05001663 return dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664}
1665
Al Viro52094c82011-02-21 21:34:47 -05001666static inline int may_lookup(struct nameidata *nd)
1667{
1668 if (nd->flags & LOOKUP_RCU) {
Al Viro4ad5abb2011-06-20 19:57:03 -04001669 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
Al Viro52094c82011-02-21 21:34:47 -05001670 if (err != -ECHILD)
1671 return err;
Al Viro6e9918b2015-05-05 09:26:05 -04001672 if (unlazy_walk(nd, NULL, 0))
Al Viro52094c82011-02-21 21:34:47 -05001673 return -ECHILD;
1674 }
Al Viro4ad5abb2011-06-20 19:57:03 -04001675 return inode_permission(nd->inode, MAY_EXEC);
Al Viro52094c82011-02-21 21:34:47 -05001676}
1677
Al Viro9856fa12011-03-04 14:22:06 -05001678static inline int handle_dots(struct nameidata *nd, int type)
1679{
1680 if (type == LAST_DOTDOT) {
Al Viro9e6697e2015-12-05 20:07:21 -05001681 if (!nd->root.mnt)
1682 set_root(nd);
Al Viro9856fa12011-03-04 14:22:06 -05001683 if (nd->flags & LOOKUP_RCU) {
Al Viro70291ae2015-05-04 07:53:00 -04001684 return follow_dotdot_rcu(nd);
Al Viro9856fa12011-03-04 14:22:06 -05001685 } else
Eric W. Biederman397d4252015-08-15 20:27:13 -05001686 return follow_dotdot(nd);
Al Viro9856fa12011-03-04 14:22:06 -05001687 }
1688 return 0;
1689}
1690
Al Viro181548c2015-05-07 19:54:34 -04001691static int pick_link(struct nameidata *nd, struct path *link,
1692 struct inode *inode, unsigned seq)
Al Virod63ff282015-05-04 18:13:23 -04001693{
Al Viro626de992015-05-04 18:26:59 -04001694 int error;
Al Viro1cf26652015-05-06 16:01:56 -04001695 struct saved *last;
NeilBrown756daf22015-03-23 13:37:38 +11001696 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
Al Viro626de992015-05-04 18:26:59 -04001697 path_to_nameidata(link, nd);
1698 return -ELOOP;
1699 }
Al Virobc40aee2015-05-09 13:04:24 -04001700 if (!(nd->flags & LOOKUP_RCU)) {
Al Viro79733872015-05-09 12:55:43 -04001701 if (link->mnt == nd->path.mnt)
1702 mntget(link->mnt);
Al Virod63ff282015-05-04 18:13:23 -04001703 }
Al Viro626de992015-05-04 18:26:59 -04001704 error = nd_alloc_stack(nd);
1705 if (unlikely(error)) {
Al Virobc40aee2015-05-09 13:04:24 -04001706 if (error == -ECHILD) {
1707 if (unlikely(unlazy_link(nd, link, seq)))
1708 return -ECHILD;
1709 error = nd_alloc_stack(nd);
1710 }
1711 if (error) {
1712 path_put(link);
1713 return error;
1714 }
Al Viro626de992015-05-04 18:26:59 -04001715 }
1716
Al Viroab104922015-05-10 11:50:01 -04001717 last = nd->stack + nd->depth++;
Al Viro1cf26652015-05-06 16:01:56 -04001718 last->link = *link;
Al Virofceef392015-12-29 15:58:39 -05001719 clear_delayed_call(&last->done);
1720 nd->link_inode = inode;
Al Viro0450b2d2015-05-08 13:23:53 -04001721 last->seq = seq;
Al Virod63ff282015-05-04 18:13:23 -04001722 return 1;
1723}
1724
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001725/*
1726 * Do we need to follow links? We _really_ want to be able
1727 * to do this check without having to look at inode->i_op,
1728 * so we keep a cache of "no, this doesn't need follow_link"
1729 * for the common case.
1730 */
Al Viro254cf582015-05-05 09:40:46 -04001731static inline int should_follow_link(struct nameidata *nd, struct path *link,
Al Viro181548c2015-05-07 19:54:34 -04001732 int follow,
1733 struct inode *inode, unsigned seq)
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001734{
Al Virod63ff282015-05-04 18:13:23 -04001735 if (likely(!d_is_symlink(link->dentry)))
1736 return 0;
1737 if (!follow)
1738 return 0;
Al Viroa7f77542016-02-27 19:31:01 -05001739 /* make sure that d_is_symlink above matches inode */
1740 if (nd->flags & LOOKUP_RCU) {
1741 if (read_seqcount_retry(&link->dentry->d_seq, seq))
1742 return -ECHILD;
1743 }
Al Viro181548c2015-05-07 19:54:34 -04001744 return pick_link(nd, link, inode, seq);
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001745}
1746
Al Viro4693a542015-05-04 17:47:11 -04001747enum {WALK_GET = 1, WALK_PUT = 2};
1748
1749static int walk_component(struct nameidata *nd, int flags)
Al Viroce57dfc2011-03-13 19:58:58 -04001750{
Al Virocaa85632015-04-22 17:52:47 -04001751 struct path path;
Al Viroce57dfc2011-03-13 19:58:58 -04001752 struct inode *inode;
Al Viro254cf582015-05-05 09:40:46 -04001753 unsigned seq;
Al Viroce57dfc2011-03-13 19:58:58 -04001754 int err;
1755 /*
1756 * "." and ".." are special - ".." especially so because it has
1757 * to be able to know about the current root directory and
1758 * parent relationships.
1759 */
Al Viro4693a542015-05-04 17:47:11 -04001760 if (unlikely(nd->last_type != LAST_NORM)) {
1761 err = handle_dots(nd, nd->last_type);
1762 if (flags & WALK_PUT)
1763 put_link(nd);
1764 return err;
1765 }
Al Viro254cf582015-05-05 09:40:46 -04001766 err = lookup_fast(nd, &path, &inode, &seq);
Al Viroe9742b52016-03-05 22:04:59 -05001767 if (unlikely(err <= 0)) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001768 if (err < 0)
Al Virof0a9ba72015-05-04 07:59:30 -04001769 return err;
Al Viroe3c13922016-03-06 14:03:27 -05001770 path.dentry = lookup_slow(&nd->last, nd->path.dentry,
1771 nd->flags);
1772 if (IS_ERR(path.dentry))
1773 return PTR_ERR(path.dentry);
Al Viro7500c382016-03-31 00:23:05 -04001774
Al Viroe3c13922016-03-06 14:03:27 -05001775 path.mnt = nd->path.mnt;
1776 err = follow_managed(&path, nd);
1777 if (unlikely(err < 0))
Al Virof0a9ba72015-05-04 07:59:30 -04001778 return err;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001779
Al Viro7500c382016-03-31 00:23:05 -04001780 if (unlikely(d_is_negative(path.dentry))) {
1781 path_to_nameidata(&path, nd);
1782 return -ENOENT;
1783 }
1784
Al Viro254cf582015-05-05 09:40:46 -04001785 seq = 0; /* we are already out of RCU mode */
Al Virod4565642016-02-27 19:23:16 -05001786 inode = d_backing_inode(path.dentry);
Al Viroce57dfc2011-03-13 19:58:58 -04001787 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001788
Al Viro4693a542015-05-04 17:47:11 -04001789 if (flags & WALK_PUT)
1790 put_link(nd);
Al Viro181548c2015-05-07 19:54:34 -04001791 err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq);
Al Virod63ff282015-05-04 18:13:23 -04001792 if (unlikely(err))
1793 return err;
Al Virocaa85632015-04-22 17:52:47 -04001794 path_to_nameidata(&path, nd);
Al Viroce57dfc2011-03-13 19:58:58 -04001795 nd->inode = inode;
Al Viro254cf582015-05-05 09:40:46 -04001796 nd->seq = seq;
Al Viroce57dfc2011-03-13 19:58:58 -04001797 return 0;
1798}
1799
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800/*
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001801 * We can do the critical dentry name comparison and hashing
1802 * operations one word at a time, but we are limited to:
1803 *
1804 * - Architectures with fast unaligned word accesses. We could
1805 * do a "get_unaligned()" if this helps and is sufficiently
1806 * fast.
1807 *
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001808 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1809 * do not trap on the (extremely unlikely) case of a page
1810 * crossing operation.
1811 *
1812 * - Furthermore, we need an efficient 64-bit compile for the
1813 * 64-bit case in order to generate the "number of bytes in
1814 * the final mask". Again, that could be replaced with a
1815 * efficient population count instruction or similar.
1816 */
1817#ifdef CONFIG_DCACHE_WORD_ACCESS
1818
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001819#include <asm/word-at-a-time.h>
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001820
George Spelvin468a9422016-05-26 22:11:51 -04001821#ifdef HASH_MIX
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001822
George Spelvin468a9422016-05-26 22:11:51 -04001823/* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1824
1825#elif defined(CONFIG_64BIT)
George Spelvin2a18da7a2016-05-23 07:43:58 -04001826/*
1827 * Register pressure in the mixing function is an issue, particularly
1828 * on 32-bit x86, but almost any function requires one state value and
1829 * one temporary. Instead, use a function designed for two state values
1830 * and no temporaries.
1831 *
1832 * This function cannot create a collision in only two iterations, so
1833 * we have two iterations to achieve avalanche. In those two iterations,
1834 * we have six layers of mixing, which is enough to spread one bit's
1835 * influence out to 2^6 = 64 state bits.
1836 *
1837 * Rotate constants are scored by considering either 64 one-bit input
1838 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1839 * probability of that delta causing a change to each of the 128 output
1840 * bits, using a sample of random initial states.
1841 *
1842 * The Shannon entropy of the computed probabilities is then summed
1843 * to produce a score. Ideally, any input change has a 50% chance of
1844 * toggling any given output bit.
1845 *
1846 * Mixing scores (in bits) for (12,45):
1847 * Input delta: 1-bit 2-bit
1848 * 1 round: 713.3 42542.6
1849 * 2 rounds: 2753.7 140389.8
1850 * 3 rounds: 5954.1 233458.2
1851 * 4 rounds: 7862.6 256672.2
1852 * Perfect: 8192 258048
1853 * (64*128) (64*63/2 * 128)
1854 */
1855#define HASH_MIX(x, y, a) \
1856 ( x ^= (a), \
1857 y ^= x, x = rol64(x,12),\
1858 x += y, y = rol64(y,45),\
1859 y *= 9 )
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001860
George Spelvin0fed3ac2016-05-02 06:31:01 -04001861/*
George Spelvin2a18da7a2016-05-23 07:43:58 -04001862 * Fold two longs into one 32-bit hash value. This must be fast, but
1863 * latency isn't quite as critical, as there is a fair bit of additional
1864 * work done before the hash value is used.
George Spelvin0fed3ac2016-05-02 06:31:01 -04001865 */
George Spelvin2a18da7a2016-05-23 07:43:58 -04001866static inline unsigned int fold_hash(unsigned long x, unsigned long y)
George Spelvin0fed3ac2016-05-02 06:31:01 -04001867{
George Spelvin2a18da7a2016-05-23 07:43:58 -04001868 y ^= x * GOLDEN_RATIO_64;
1869 y *= GOLDEN_RATIO_64;
1870 return y >> 32;
George Spelvin0fed3ac2016-05-02 06:31:01 -04001871}
1872
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001873#else /* 32-bit case */
1874
George Spelvin2a18da7a2016-05-23 07:43:58 -04001875/*
1876 * Mixing scores (in bits) for (7,20):
1877 * Input delta: 1-bit 2-bit
1878 * 1 round: 330.3 9201.6
1879 * 2 rounds: 1246.4 25475.4
1880 * 3 rounds: 1907.1 31295.1
1881 * 4 rounds: 2042.3 31718.6
1882 * Perfect: 2048 31744
1883 * (32*64) (32*31/2 * 64)
1884 */
1885#define HASH_MIX(x, y, a) \
1886 ( x ^= (a), \
1887 y ^= x, x = rol32(x, 7),\
1888 x += y, y = rol32(y,20),\
1889 y *= 9 )
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001890
George Spelvin2a18da7a2016-05-23 07:43:58 -04001891static inline unsigned int fold_hash(unsigned long x, unsigned long y)
George Spelvin0fed3ac2016-05-02 06:31:01 -04001892{
George Spelvin2a18da7a2016-05-23 07:43:58 -04001893 /* Use arch-optimized multiply if one exists */
1894 return __hash_32(y ^ __hash_32(x));
George Spelvin0fed3ac2016-05-02 06:31:01 -04001895}
1896
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001897#endif
1898
George Spelvin2a18da7a2016-05-23 07:43:58 -04001899/*
1900 * Return the hash of a string of known length. This is carfully
1901 * designed to match hash_name(), which is the more critical function.
1902 * In particular, we must end by hashing a final word containing 0..7
1903 * payload bytes, to match the way that hash_name() iterates until it
1904 * finds the delimiter after the name.
1905 */
Linus Torvalds8387ff22016-06-10 07:51:30 -07001906unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001907{
Linus Torvalds8387ff22016-06-10 07:51:30 -07001908 unsigned long a, x = 0, y = (unsigned long)salt;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001909
1910 for (;;) {
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001911 if (!len)
1912 goto done;
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001913 a = load_unaligned_zeropad(name);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001914 if (len < sizeof(unsigned long))
1915 break;
George Spelvin2a18da7a2016-05-23 07:43:58 -04001916 HASH_MIX(x, y, a);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001917 name += sizeof(unsigned long);
1918 len -= sizeof(unsigned long);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001919 }
George Spelvin2a18da7a2016-05-23 07:43:58 -04001920 x ^= a & bytemask_from_count(len);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001921done:
George Spelvin2a18da7a2016-05-23 07:43:58 -04001922 return fold_hash(x, y);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001923}
1924EXPORT_SYMBOL(full_name_hash);
1925
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001926/* Return the "hash_len" (hash and length) of a null-terminated string */
Linus Torvalds8387ff22016-06-10 07:51:30 -07001927u64 hashlen_string(const void *salt, const char *name)
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001928{
Linus Torvalds8387ff22016-06-10 07:51:30 -07001929 unsigned long a = 0, x = 0, y = (unsigned long)salt;
1930 unsigned long adata, mask, len;
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001931 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1932
Linus Torvalds8387ff22016-06-10 07:51:30 -07001933 len = 0;
1934 goto inside;
1935
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001936 do {
George Spelvin2a18da7a2016-05-23 07:43:58 -04001937 HASH_MIX(x, y, a);
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001938 len += sizeof(unsigned long);
Linus Torvalds8387ff22016-06-10 07:51:30 -07001939inside:
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001940 a = load_unaligned_zeropad(name+len);
1941 } while (!has_zero(a, &adata, &constants));
1942
1943 adata = prep_zero_mask(a, adata, &constants);
1944 mask = create_zero_mask(adata);
George Spelvin2a18da7a2016-05-23 07:43:58 -04001945 x ^= a & zero_bytemask(mask);
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001946
George Spelvin2a18da7a2016-05-23 07:43:58 -04001947 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001948}
1949EXPORT_SYMBOL(hashlen_string);
1950
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001951/*
1952 * Calculate the length and hash of the path component, and
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001953 * return the "hash_len" as the result.
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001954 */
Linus Torvalds8387ff22016-06-10 07:51:30 -07001955static inline u64 hash_name(const void *salt, const char *name)
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001956{
Linus Torvalds8387ff22016-06-10 07:51:30 -07001957 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
1958 unsigned long adata, bdata, mask, len;
Linus Torvalds36126f82012-05-26 10:43:17 -07001959 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001960
Linus Torvalds8387ff22016-06-10 07:51:30 -07001961 len = 0;
1962 goto inside;
1963
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001964 do {
George Spelvin2a18da7a2016-05-23 07:43:58 -04001965 HASH_MIX(x, y, a);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001966 len += sizeof(unsigned long);
Linus Torvalds8387ff22016-06-10 07:51:30 -07001967inside:
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001968 a = load_unaligned_zeropad(name+len);
Linus Torvalds36126f82012-05-26 10:43:17 -07001969 b = a ^ REPEAT_BYTE('/');
1970 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001971
Linus Torvalds36126f82012-05-26 10:43:17 -07001972 adata = prep_zero_mask(a, adata, &constants);
1973 bdata = prep_zero_mask(b, bdata, &constants);
Linus Torvalds36126f82012-05-26 10:43:17 -07001974 mask = create_zero_mask(adata | bdata);
George Spelvin2a18da7a2016-05-23 07:43:58 -04001975 x ^= a & zero_bytemask(mask);
Linus Torvalds36126f82012-05-26 10:43:17 -07001976
George Spelvin2a18da7a2016-05-23 07:43:58 -04001977 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001978}
1979
George Spelvin2a18da7a2016-05-23 07:43:58 -04001980#else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001981
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001982/* Return the hash of a string of known length */
Linus Torvalds8387ff22016-06-10 07:51:30 -07001983unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
Linus Torvalds0145acc2012-03-02 14:32:59 -08001984{
Linus Torvalds8387ff22016-06-10 07:51:30 -07001985 unsigned long hash = init_name_hash(salt);
Linus Torvalds0145acc2012-03-02 14:32:59 -08001986 while (len--)
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001987 hash = partial_name_hash((unsigned char)*name++, hash);
Linus Torvalds0145acc2012-03-02 14:32:59 -08001988 return end_name_hash(hash);
1989}
Linus Torvaldsae942ae2012-03-02 19:40:57 -08001990EXPORT_SYMBOL(full_name_hash);
Linus Torvalds0145acc2012-03-02 14:32:59 -08001991
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001992/* Return the "hash_len" (hash and length) of a null-terminated string */
Linus Torvalds8387ff22016-06-10 07:51:30 -07001993u64 hashlen_string(const void *salt, const char *name)
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001994{
Linus Torvalds8387ff22016-06-10 07:51:30 -07001995 unsigned long hash = init_name_hash(salt);
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001996 unsigned long len = 0, c;
1997
1998 c = (unsigned char)*name;
George Spelvine0ab7af2016-05-29 08:05:56 -04001999 while (c) {
George Spelvinfcfd2fb2016-05-20 08:41:37 -04002000 len++;
2001 hash = partial_name_hash(c, hash);
2002 c = (unsigned char)name[len];
George Spelvine0ab7af2016-05-29 08:05:56 -04002003 }
George Spelvinfcfd2fb2016-05-20 08:41:37 -04002004 return hashlen_create(end_name_hash(hash), len);
2005}
George Spelvinf2a031b2016-05-29 01:26:41 -04002006EXPORT_SYMBOL(hashlen_string);
George Spelvinfcfd2fb2016-05-20 08:41:37 -04002007
Linus Torvalds3ddcd052011-08-06 22:45:50 -07002008/*
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002009 * We know there's a real path component here of at least
2010 * one character.
2011 */
Linus Torvalds8387ff22016-06-10 07:51:30 -07002012static inline u64 hash_name(const void *salt, const char *name)
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002013{
Linus Torvalds8387ff22016-06-10 07:51:30 -07002014 unsigned long hash = init_name_hash(salt);
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002015 unsigned long len = 0, c;
2016
2017 c = (unsigned char)*name;
2018 do {
2019 len++;
2020 hash = partial_name_hash(c, hash);
2021 c = (unsigned char)name[len];
2022 } while (c && c != '/');
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002023 return hashlen_create(end_name_hash(hash), len);
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002024}
2025
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08002026#endif
2027
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002028/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01002030 * This is the basic name resolution function, turning a pathname into
2031 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01002033 * Returns 0 and nd will have valid dentry and mnt on success.
2034 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 */
Al Viro6de88d72009-08-09 01:41:57 +04002036static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 int err;
Al Viro32cd7462015-04-18 20:30:49 -04002039
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 while (*name=='/')
2041 name++;
2042 if (!*name)
Al Viro9e18f102015-04-18 20:44:34 -04002043 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 /* At this point we know we have a real path component. */
2046 for(;;) {
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002047 u64 hash_len;
Al Virofe479a52011-02-22 15:10:03 -05002048 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049
Al Viro52094c82011-02-21 21:34:47 -05002050 err = may_lookup(nd);
George Spelvin2a18da7a2016-05-23 07:43:58 -04002051 if (err)
Al Viro3595e232015-05-09 16:54:45 -04002052 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053
Linus Torvalds8387ff22016-06-10 07:51:30 -07002054 hash_len = hash_name(nd->path.dentry, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055
Al Virofe479a52011-02-22 15:10:03 -05002056 type = LAST_NORM;
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002057 if (name[0] == '.') switch (hashlen_len(hash_len)) {
Al Virofe479a52011-02-22 15:10:03 -05002058 case 2:
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002059 if (name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05002060 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05002061 nd->flags |= LOOKUP_JUMPED;
2062 }
Al Virofe479a52011-02-22 15:10:03 -05002063 break;
2064 case 1:
2065 type = LAST_DOT;
2066 }
Al Viro5a202bc2011-03-08 14:17:44 -05002067 if (likely(type == LAST_NORM)) {
2068 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05002069 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05002070 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
James Hogana060dc52014-09-16 13:07:35 +01002071 struct qstr this = { { .hash_len = hash_len }, .name = name };
Linus Torvaldsda53be12013-05-21 15:22:44 -07002072 err = parent->d_op->d_hash(parent, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05002073 if (err < 0)
Al Viro3595e232015-05-09 16:54:45 -04002074 return err;
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002075 hash_len = this.hash_len;
2076 name = this.name;
Al Viro5a202bc2011-03-08 14:17:44 -05002077 }
2078 }
Al Virofe479a52011-02-22 15:10:03 -05002079
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002080 nd->last.hash_len = hash_len;
2081 nd->last.name = name;
Al Viro5f4a6a62013-01-24 18:04:22 -05002082 nd->last_type = type;
2083
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002084 name += hashlen_len(hash_len);
2085 if (!*name)
Al Virobdf6cbf2015-04-18 20:21:40 -04002086 goto OK;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08002087 /*
2088 * If it wasn't NUL, we know it was '/'. Skip that
2089 * slash, and continue until no more slashes.
2090 */
2091 do {
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07002092 name++;
2093 } while (unlikely(*name == '/'));
Al Viro8620c232015-05-04 08:58:35 -04002094 if (unlikely(!*name)) {
2095OK:
Al Viro368ee9b2015-05-08 17:19:59 -04002096 /* pathname body, done */
Al Viro8620c232015-05-04 08:58:35 -04002097 if (!nd->depth)
2098 return 0;
2099 name = nd->stack[nd->depth - 1].name;
Al Viro368ee9b2015-05-08 17:19:59 -04002100 /* trailing symlink, done */
Al Viro8620c232015-05-04 08:58:35 -04002101 if (!name)
2102 return 0;
2103 /* last component of nested symlink */
Al Viro4693a542015-05-04 17:47:11 -04002104 err = walk_component(nd, WALK_GET | WALK_PUT);
Al Viro8620c232015-05-04 08:58:35 -04002105 } else {
Al Viro4693a542015-05-04 17:47:11 -04002106 err = walk_component(nd, WALK_GET);
Al Viro8620c232015-05-04 08:58:35 -04002107 }
Al Viroce57dfc2011-03-13 19:58:58 -04002108 if (err < 0)
Al Viro3595e232015-05-09 16:54:45 -04002109 return err;
Al Virofe479a52011-02-22 15:10:03 -05002110
Al Viroce57dfc2011-03-13 19:58:58 -04002111 if (err) {
Al Viro626de992015-05-04 18:26:59 -04002112 const char *s = get_link(nd);
Al Viro5a460272015-04-17 23:44:45 -04002113
Viresh Kumara1c83682015-08-12 15:59:44 +05302114 if (IS_ERR(s))
Al Viro3595e232015-05-09 16:54:45 -04002115 return PTR_ERR(s);
Al Virod40bcc02015-04-18 20:03:03 -04002116 err = 0;
2117 if (unlikely(!s)) {
2118 /* jumped */
Al Virob9ff4422015-05-02 20:19:23 -04002119 put_link(nd);
Al Virod40bcc02015-04-18 20:03:03 -04002120 } else {
Al Virofab51e82015-05-10 11:01:00 -04002121 nd->stack[nd->depth - 1].name = name;
2122 name = s;
2123 continue;
Al Virod40bcc02015-04-18 20:03:03 -04002124 }
Nick Piggin31e6b012011-01-07 17:49:52 +11002125 }
Al Viro97242f92015-08-01 19:59:28 -04002126 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2127 if (nd->flags & LOOKUP_RCU) {
2128 if (unlazy_walk(nd, NULL, 0))
2129 return -ECHILD;
2130 }
Al Viro3595e232015-05-09 16:54:45 -04002131 return -ENOTDIR;
Al Viro97242f92015-08-01 19:59:28 -04002132 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134}
2135
Al Viroc8a53ee2015-05-12 18:43:07 -04002136static const char *path_init(struct nameidata *nd, unsigned flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137{
Prasanna Medaea3834d2005-04-29 16:00:17 +01002138 int retval = 0;
Al Viroc8a53ee2015-05-12 18:43:07 -04002139 const char *s = nd->name->name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140
2141 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro980f3ea2014-11-20 14:20:24 -05002142 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143 nd->depth = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05002144 if (flags & LOOKUP_ROOT) {
David Howellsb18825a2013-09-12 19:22:53 +01002145 struct dentry *root = nd->root.dentry;
2146 struct inode *inode = root->d_inode;
Al Virofd2f7cb2015-02-22 20:07:13 -05002147 if (*s) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002148 if (!d_can_lookup(root))
Al Viro368ee9b2015-05-08 17:19:59 -04002149 return ERR_PTR(-ENOTDIR);
Al Viro73d049a2011-03-11 12:08:24 -05002150 retval = inode_permission(inode, MAY_EXEC);
2151 if (retval)
Al Viro368ee9b2015-05-08 17:19:59 -04002152 return ERR_PTR(retval);
Al Viro73d049a2011-03-11 12:08:24 -05002153 }
Al Viro5b6ca022011-03-09 23:04:47 -05002154 nd->path = nd->root;
2155 nd->inode = inode;
2156 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05002157 rcu_read_lock();
Al Viro5b6ca022011-03-09 23:04:47 -05002158 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro8f47a0162015-05-09 19:02:01 -04002159 nd->root_seq = nd->seq;
Al Viro48a066e2013-09-29 22:06:07 -04002160 nd->m_seq = read_seqbegin(&mount_lock);
Al Viro5b6ca022011-03-09 23:04:47 -05002161 } else {
2162 path_get(&nd->path);
2163 }
Al Viro368ee9b2015-05-08 17:19:59 -04002164 return s;
Al Viro5b6ca022011-03-09 23:04:47 -05002165 }
2166
Al Viro2a737872009-04-07 11:49:53 -04002167 nd->root.mnt = NULL;
Al Viro248fb5b2015-12-05 20:51:58 -05002168 nd->path.mnt = NULL;
2169 nd->path.dentry = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170
Al Viro48a066e2013-09-29 22:06:07 -04002171 nd->m_seq = read_seqbegin(&mount_lock);
Al Virofd2f7cb2015-02-22 20:07:13 -05002172 if (*s == '/') {
Al Viro9e6697e2015-12-05 20:07:21 -05002173 if (flags & LOOKUP_RCU)
Al Viro8b61e742013-11-08 12:45:01 -05002174 rcu_read_lock();
Al Viro9e6697e2015-12-05 20:07:21 -05002175 set_root(nd);
Al Viro248fb5b2015-12-05 20:51:58 -05002176 if (likely(!nd_jump_root(nd)))
Al Viroef55d912015-12-05 20:25:06 -05002177 return s;
Al Viro248fb5b2015-12-05 20:51:58 -05002178 nd->root.mnt = NULL;
Al Viroef55d912015-12-05 20:25:06 -05002179 rcu_read_unlock();
2180 return ERR_PTR(-ECHILD);
Al Viroc8a53ee2015-05-12 18:43:07 -04002181 } else if (nd->dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05002182 if (flags & LOOKUP_RCU) {
2183 struct fs_struct *fs = current->fs;
2184 unsigned seq;
2185
Al Viro8b61e742013-11-08 12:45:01 -05002186 rcu_read_lock();
Al Viroe41f7d42011-02-22 14:02:58 -05002187
2188 do {
2189 seq = read_seqcount_begin(&fs->seq);
2190 nd->path = fs->pwd;
Al Viroef55d912015-12-05 20:25:06 -05002191 nd->inode = nd->path.dentry->d_inode;
Al Viroe41f7d42011-02-22 14:02:58 -05002192 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2193 } while (read_seqcount_retry(&fs->seq, seq));
2194 } else {
2195 get_fs_pwd(current->fs, &nd->path);
Al Viroef55d912015-12-05 20:25:06 -05002196 nd->inode = nd->path.dentry->d_inode;
Al Viroe41f7d42011-02-22 14:02:58 -05002197 }
Al Viroef55d912015-12-05 20:25:06 -05002198 return s;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002199 } else {
Jeff Layton582aa642012-12-11 08:56:16 -05002200 /* Caller must check execute permissions on the starting path component */
Al Viroc8a53ee2015-05-12 18:43:07 -04002201 struct fd f = fdget_raw(nd->dfd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002202 struct dentry *dentry;
2203
Al Viro2903ff02012-08-28 12:52:22 -04002204 if (!f.file)
Al Viro368ee9b2015-05-08 17:19:59 -04002205 return ERR_PTR(-EBADF);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002206
Al Viro2903ff02012-08-28 12:52:22 -04002207 dentry = f.file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002208
Al Virofd2f7cb2015-02-22 20:07:13 -05002209 if (*s) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002210 if (!d_can_lookup(dentry)) {
Al Viro2903ff02012-08-28 12:52:22 -04002211 fdput(f);
Al Viro368ee9b2015-05-08 17:19:59 -04002212 return ERR_PTR(-ENOTDIR);
Al Viro2903ff02012-08-28 12:52:22 -04002213 }
Al Virof52e0c12011-03-14 18:56:51 -04002214 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002215
Al Viro2903ff02012-08-28 12:52:22 -04002216 nd->path = f.file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05002217 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05002218 rcu_read_lock();
Al Viro34a26b92015-05-11 08:05:05 -04002219 nd->inode = nd->path.dentry->d_inode;
2220 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viroe41f7d42011-02-22 14:02:58 -05002221 } else {
Al Viro2903ff02012-08-28 12:52:22 -04002222 path_get(&nd->path);
Al Viro34a26b92015-05-11 08:05:05 -04002223 nd->inode = nd->path.dentry->d_inode;
Al Viroe41f7d42011-02-22 14:02:58 -05002224 }
Al Viro34a26b92015-05-11 08:05:05 -04002225 fdput(f);
Al Viro368ee9b2015-05-08 17:19:59 -04002226 return s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227 }
Al Viro9b4a9b12009-04-07 11:44:16 -04002228}
2229
Al Viro3bdba282015-05-08 17:37:07 -04002230static const char *trailing_symlink(struct nameidata *nd)
Al Viro95fa25d2015-04-22 13:46:57 -04002231{
2232 const char *s;
Al Virofec2fa22015-05-06 15:58:18 -04002233 int error = may_follow_link(nd);
Al Virodeb106c2015-05-08 18:05:21 -04002234 if (unlikely(error))
Al Viro3bdba282015-05-08 17:37:07 -04002235 return ERR_PTR(error);
Al Viro95fa25d2015-04-22 13:46:57 -04002236 nd->flags |= LOOKUP_PARENT;
Al Virofab51e82015-05-10 11:01:00 -04002237 nd->stack[0].name = NULL;
Al Viro3b2e7f72015-04-19 00:53:50 -04002238 s = get_link(nd);
Al Virodeb106c2015-05-08 18:05:21 -04002239 return s ? s : "";
Al Viro95fa25d2015-04-22 13:46:57 -04002240}
2241
Al Virocaa85632015-04-22 17:52:47 -04002242static inline int lookup_last(struct nameidata *nd)
Al Virobd92d7f2011-03-14 19:54:59 -04002243{
2244 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2245 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2246
2247 nd->flags &= ~LOOKUP_PARENT;
Al Virodeb106c2015-05-08 18:05:21 -04002248 return walk_component(nd,
Al Viro4693a542015-05-04 17:47:11 -04002249 nd->flags & LOOKUP_FOLLOW
2250 ? nd->depth
2251 ? WALK_PUT | WALK_GET
2252 : WALK_GET
2253 : 0);
Al Virobd92d7f2011-03-14 19:54:59 -04002254}
2255
Al Viro9b4a9b12009-04-07 11:44:16 -04002256/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroc8a53ee2015-05-12 18:43:07 -04002257static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
Al Viro9b4a9b12009-04-07 11:44:16 -04002258{
Al Viroc8a53ee2015-05-12 18:43:07 -04002259 const char *s = path_init(nd, flags);
Al Virobd92d7f2011-03-14 19:54:59 -04002260 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11002261
Al Viro368ee9b2015-05-08 17:19:59 -04002262 if (IS_ERR(s))
2263 return PTR_ERR(s);
Al Viro3bdba282015-05-08 17:37:07 -04002264 while (!(err = link_path_walk(s, nd))
2265 && ((err = lookup_last(nd)) > 0)) {
2266 s = trailing_symlink(nd);
2267 if (IS_ERR(s)) {
2268 err = PTR_ERR(s);
2269 break;
Al Virobd92d7f2011-03-14 19:54:59 -04002270 }
2271 }
Al Viro9f1fafe2011-03-25 11:00:12 -04002272 if (!err)
2273 err = complete_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04002274
Al Virodeb106c2015-05-08 18:05:21 -04002275 if (!err && nd->flags & LOOKUP_DIRECTORY)
2276 if (!d_can_lookup(nd->path.dentry))
Al Virobd23a532011-03-23 09:56:30 -04002277 err = -ENOTDIR;
Al Viro625b6d12015-05-12 16:36:12 -04002278 if (!err) {
2279 *path = nd->path;
2280 nd->path.mnt = NULL;
2281 nd->path.dentry = NULL;
2282 }
2283 terminate_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04002284 return err;
Al Viroee0827c2011-02-21 23:38:09 -05002285}
Nick Piggin31e6b012011-01-07 17:49:52 +11002286
Al Viro625b6d12015-05-12 16:36:12 -04002287static int filename_lookup(int dfd, struct filename *name, unsigned flags,
Al Viro9ad1aaa2015-05-12 16:44:39 -04002288 struct path *path, struct path *root)
Jeff Layton873f1ee2012-10-10 15:25:29 -04002289{
Al Viro894bc8c2015-05-02 07:16:16 -04002290 int retval;
Al Viro9883d182015-05-13 07:28:08 -04002291 struct nameidata nd;
Al Viroabc9f5b2015-05-12 16:53:42 -04002292 if (IS_ERR(name))
2293 return PTR_ERR(name);
Al Viro9ad1aaa2015-05-12 16:44:39 -04002294 if (unlikely(root)) {
2295 nd.root = *root;
2296 flags |= LOOKUP_ROOT;
2297 }
Al Viro9883d182015-05-13 07:28:08 -04002298 set_nameidata(&nd, dfd, name);
Al Viroc8a53ee2015-05-12 18:43:07 -04002299 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002300 if (unlikely(retval == -ECHILD))
Al Viroc8a53ee2015-05-12 18:43:07 -04002301 retval = path_lookupat(&nd, flags, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002302 if (unlikely(retval == -ESTALE))
Al Viroc8a53ee2015-05-12 18:43:07 -04002303 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002304
2305 if (likely(!retval))
Al Viro625b6d12015-05-12 16:36:12 -04002306 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
Al Viro9883d182015-05-13 07:28:08 -04002307 restore_nameidata();
Al Viroe4bd1c12015-05-12 16:40:39 -04002308 putname(name);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002309 return retval;
2310}
2311
Al Viro8bcb77f2015-05-08 16:59:20 -04002312/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroc8a53ee2015-05-12 18:43:07 -04002313static int path_parentat(struct nameidata *nd, unsigned flags,
Al Viro391172c2015-05-09 11:19:16 -04002314 struct path *parent)
Al Viro8bcb77f2015-05-08 16:59:20 -04002315{
Al Viroc8a53ee2015-05-12 18:43:07 -04002316 const char *s = path_init(nd, flags);
Al Viro368ee9b2015-05-08 17:19:59 -04002317 int err;
2318 if (IS_ERR(s))
2319 return PTR_ERR(s);
2320 err = link_path_walk(s, nd);
Al Viro8bcb77f2015-05-08 16:59:20 -04002321 if (!err)
2322 err = complete_walk(nd);
Al Viro391172c2015-05-09 11:19:16 -04002323 if (!err) {
2324 *parent = nd->path;
2325 nd->path.mnt = NULL;
2326 nd->path.dentry = NULL;
2327 }
2328 terminate_walk(nd);
Al Viro8bcb77f2015-05-08 16:59:20 -04002329 return err;
2330}
2331
Al Viro5c31b6c2015-05-12 17:32:54 -04002332static struct filename *filename_parentat(int dfd, struct filename *name,
Al Viro391172c2015-05-09 11:19:16 -04002333 unsigned int flags, struct path *parent,
2334 struct qstr *last, int *type)
Al Viro8bcb77f2015-05-08 16:59:20 -04002335{
2336 int retval;
Al Viro9883d182015-05-13 07:28:08 -04002337 struct nameidata nd;
Al Viro8bcb77f2015-05-08 16:59:20 -04002338
Al Viro5c31b6c2015-05-12 17:32:54 -04002339 if (IS_ERR(name))
2340 return name;
Al Viro9883d182015-05-13 07:28:08 -04002341 set_nameidata(&nd, dfd, name);
Al Viroc8a53ee2015-05-12 18:43:07 -04002342 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
Al Viro8bcb77f2015-05-08 16:59:20 -04002343 if (unlikely(retval == -ECHILD))
Al Viroc8a53ee2015-05-12 18:43:07 -04002344 retval = path_parentat(&nd, flags, parent);
Al Viro8bcb77f2015-05-08 16:59:20 -04002345 if (unlikely(retval == -ESTALE))
Al Viroc8a53ee2015-05-12 18:43:07 -04002346 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
Al Viro391172c2015-05-09 11:19:16 -04002347 if (likely(!retval)) {
2348 *last = nd.last;
2349 *type = nd.last_type;
2350 audit_inode(name, parent->dentry, LOOKUP_PARENT);
Al Viro5c31b6c2015-05-12 17:32:54 -04002351 } else {
2352 putname(name);
2353 name = ERR_PTR(retval);
Al Viro391172c2015-05-09 11:19:16 -04002354 }
Al Viro9883d182015-05-13 07:28:08 -04002355 restore_nameidata();
Al Viro5c31b6c2015-05-12 17:32:54 -04002356 return name;
Al Viro8bcb77f2015-05-08 16:59:20 -04002357}
2358
Al Viro79714f72012-06-15 03:01:42 +04002359/* does lookup, returns the object with parent locked */
2360struct dentry *kern_path_locked(const char *name, struct path *path)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002361{
Al Viro5c31b6c2015-05-12 17:32:54 -04002362 struct filename *filename;
2363 struct dentry *d;
Al Viro391172c2015-05-09 11:19:16 -04002364 struct qstr last;
2365 int type;
Paul Moore51689102015-01-22 00:00:03 -05002366
Al Viro5c31b6c2015-05-12 17:32:54 -04002367 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2368 &last, &type);
Paul Moore51689102015-01-22 00:00:03 -05002369 if (IS_ERR(filename))
2370 return ERR_CAST(filename);
Al Viro5c31b6c2015-05-12 17:32:54 -04002371 if (unlikely(type != LAST_NORM)) {
Al Viro391172c2015-05-09 11:19:16 -04002372 path_put(path);
Al Viro5c31b6c2015-05-12 17:32:54 -04002373 putname(filename);
2374 return ERR_PTR(-EINVAL);
Al Viro79714f72012-06-15 03:01:42 +04002375 }
Al Viro59551022016-01-22 15:40:57 -05002376 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
Al Viro391172c2015-05-09 11:19:16 -04002377 d = __lookup_hash(&last, path->dentry, 0);
Al Viro79714f72012-06-15 03:01:42 +04002378 if (IS_ERR(d)) {
Al Viro59551022016-01-22 15:40:57 -05002379 inode_unlock(path->dentry->d_inode);
Al Viro391172c2015-05-09 11:19:16 -04002380 path_put(path);
Al Viro79714f72012-06-15 03:01:42 +04002381 }
Paul Moore51689102015-01-22 00:00:03 -05002382 putname(filename);
Al Viro79714f72012-06-15 03:01:42 +04002383 return d;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002384}
2385
Al Virod1811462008-08-02 00:49:18 -04002386int kern_path(const char *name, unsigned int flags, struct path *path)
2387{
Al Viroabc9f5b2015-05-12 16:53:42 -04002388 return filename_lookup(AT_FDCWD, getname_kernel(name),
2389 flags, path, NULL);
Al Virod1811462008-08-02 00:49:18 -04002390}
Al Viro4d359502014-03-14 12:20:17 -04002391EXPORT_SYMBOL(kern_path);
Al Virod1811462008-08-02 00:49:18 -04002392
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002393/**
2394 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2395 * @dentry: pointer to dentry of the base directory
2396 * @mnt: pointer to vfs mount of the base directory
2397 * @name: pointer to file name
2398 * @flags: lookup flags
Al Viroe0a01242011-06-27 17:00:37 -04002399 * @path: pointer to struct path to fill
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002400 */
2401int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2402 const char *name, unsigned int flags,
Al Viroe0a01242011-06-27 17:00:37 -04002403 struct path *path)
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002404{
Al Viro9ad1aaa2015-05-12 16:44:39 -04002405 struct path root = {.mnt = mnt, .dentry = dentry};
Al Viro9ad1aaa2015-05-12 16:44:39 -04002406 /* the first argument of filename_lookup() is ignored with root */
Al Viroabc9f5b2015-05-12 16:53:42 -04002407 return filename_lookup(AT_FDCWD, getname_kernel(name),
2408 flags , path, &root);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002409}
Al Viro4d359502014-03-14 12:20:17 -04002410EXPORT_SYMBOL(vfs_path_lookup);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002411
Christoph Hellwigeead1912007-10-16 23:25:38 -07002412/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07002413 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07002414 * @name: pathname component to lookup
2415 * @base: base directory to lookup from
2416 * @len: maximum length @len should be interpreted to
2417 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07002418 * Note that this routine is purely a helper for filesystem usage and should
Al Viro9e7543e2015-02-23 02:49:48 -05002419 * not be called by generic code.
NeilBrownbbddca82016-01-07 16:08:20 -05002420 *
2421 * The caller must hold base->i_mutex.
Christoph Hellwigeead1912007-10-16 23:25:38 -07002422 */
James Morris057f6c02007-04-26 00:12:05 -07002423struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2424{
James Morris057f6c02007-04-26 00:12:05 -07002425 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05002426 unsigned int c;
Miklos Szeredicda309d2012-03-26 12:54:21 +02002427 int err;
James Morris057f6c02007-04-26 00:12:05 -07002428
Al Viro59551022016-01-22 15:40:57 -05002429 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
David Woodhouse2f9092e2009-04-20 23:18:37 +01002430
Al Viro6a96ba52011-03-07 23:49:20 -05002431 this.name = name;
2432 this.len = len;
Linus Torvalds8387ff22016-06-10 07:51:30 -07002433 this.hash = full_name_hash(base, name, len);
Al Viro6a96ba52011-03-07 23:49:20 -05002434 if (!len)
2435 return ERR_PTR(-EACCES);
2436
Al Viro21d8a152012-11-29 22:17:21 -05002437 if (unlikely(name[0] == '.')) {
2438 if (len < 2 || (len == 2 && name[1] == '.'))
2439 return ERR_PTR(-EACCES);
2440 }
2441
Al Viro6a96ba52011-03-07 23:49:20 -05002442 while (len--) {
2443 c = *(const unsigned char *)name++;
2444 if (c == '/' || c == '\0')
2445 return ERR_PTR(-EACCES);
Al Viro6a96ba52011-03-07 23:49:20 -05002446 }
Al Viro5a202bc2011-03-08 14:17:44 -05002447 /*
2448 * See if the low-level filesystem might want
2449 * to use its own hash..
2450 */
2451 if (base->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002452 int err = base->d_op->d_hash(base, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05002453 if (err < 0)
2454 return ERR_PTR(err);
2455 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07002456
Miklos Szeredicda309d2012-03-26 12:54:21 +02002457 err = inode_permission(base->d_inode, MAY_EXEC);
2458 if (err)
2459 return ERR_PTR(err);
2460
Al Viro72bd8662012-06-10 17:17:17 -04002461 return __lookup_hash(&this, base, 0);
James Morris057f6c02007-04-26 00:12:05 -07002462}
Al Viro4d359502014-03-14 12:20:17 -04002463EXPORT_SYMBOL(lookup_one_len);
James Morris057f6c02007-04-26 00:12:05 -07002464
NeilBrownbbddca82016-01-07 16:08:20 -05002465/**
2466 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2467 * @name: pathname component to lookup
2468 * @base: base directory to lookup from
2469 * @len: maximum length @len should be interpreted to
2470 *
2471 * Note that this routine is purely a helper for filesystem usage and should
2472 * not be called by generic code.
2473 *
2474 * Unlike lookup_one_len, it should be called without the parent
2475 * i_mutex held, and will take the i_mutex itself if necessary.
2476 */
2477struct dentry *lookup_one_len_unlocked(const char *name,
2478 struct dentry *base, int len)
2479{
2480 struct qstr this;
2481 unsigned int c;
2482 int err;
Linus Torvalds20d00ee2016-07-29 12:17:52 -07002483 struct dentry *ret;
NeilBrownbbddca82016-01-07 16:08:20 -05002484
2485 this.name = name;
2486 this.len = len;
Linus Torvalds8387ff22016-06-10 07:51:30 -07002487 this.hash = full_name_hash(base, name, len);
NeilBrownbbddca82016-01-07 16:08:20 -05002488 if (!len)
2489 return ERR_PTR(-EACCES);
2490
2491 if (unlikely(name[0] == '.')) {
2492 if (len < 2 || (len == 2 && name[1] == '.'))
2493 return ERR_PTR(-EACCES);
2494 }
2495
2496 while (len--) {
2497 c = *(const unsigned char *)name++;
2498 if (c == '/' || c == '\0')
2499 return ERR_PTR(-EACCES);
2500 }
2501 /*
2502 * See if the low-level filesystem might want
2503 * to use its own hash..
2504 */
2505 if (base->d_flags & DCACHE_OP_HASH) {
2506 int err = base->d_op->d_hash(base, &this);
2507 if (err < 0)
2508 return ERR_PTR(err);
2509 }
2510
2511 err = inode_permission(base->d_inode, MAY_EXEC);
2512 if (err)
2513 return ERR_PTR(err);
2514
Linus Torvalds20d00ee2016-07-29 12:17:52 -07002515 ret = lookup_dcache(&this, base, 0);
2516 if (!ret)
2517 ret = lookup_slow(&this, base, 0);
2518 return ret;
NeilBrownbbddca82016-01-07 16:08:20 -05002519}
2520EXPORT_SYMBOL(lookup_one_len_unlocked);
2521
Eric W. Biedermaneedf2652016-06-02 10:29:47 -05002522#ifdef CONFIG_UNIX98_PTYS
2523int path_pts(struct path *path)
2524{
2525 /* Find something mounted on "pts" in the same directory as
2526 * the input path.
2527 */
2528 struct dentry *child, *parent;
2529 struct qstr this;
2530 int ret;
2531
2532 ret = path_parent_directory(path);
2533 if (ret)
2534 return ret;
2535
2536 parent = path->dentry;
2537 this.name = "pts";
2538 this.len = 3;
2539 child = d_hash_and_lookup(parent, &this);
2540 if (!child)
2541 return -ENOENT;
2542
2543 path->dentry = child;
2544 dput(parent);
2545 follow_mount(path);
2546 return 0;
2547}
2548#endif
2549
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002550int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2551 struct path *path, int *empty)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552{
Al Viroabc9f5b2015-05-12 16:53:42 -04002553 return filename_lookup(dfd, getname_flags(name, flags, empty),
2554 flags, path, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555}
Al Virob853a162015-05-13 09:12:02 -04002556EXPORT_SYMBOL(user_path_at_empty);
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002557
Jeff Layton873f1ee2012-10-10 15:25:29 -04002558/*
2559 * NB: most callers don't do anything directly with the reference to the
2560 * to struct filename, but the nd->last pointer points into the name string
2561 * allocated by getname. So we must hold the reference to it until all
2562 * path-walking is complete.
2563 */
Al Viroa2ec4a22015-05-13 06:57:49 -04002564static inline struct filename *
Al Virof5beed72015-04-30 16:09:11 -04002565user_path_parent(int dfd, const char __user *path,
2566 struct path *parent,
2567 struct qstr *last,
2568 int *type,
Jeff Layton9e790bd2012-12-11 12:10:09 -05002569 unsigned int flags)
Al Viro2ad94ae2008-07-21 09:32:51 -04002570{
Jeff Layton9e790bd2012-12-11 12:10:09 -05002571 /* only LOOKUP_REVAL is allowed in extra flags */
Al Viro5c31b6c2015-05-12 17:32:54 -04002572 return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL,
2573 parent, last, type);
Al Viro2ad94ae2008-07-21 09:32:51 -04002574}
2575
Jeff Layton80334262013-07-26 06:23:25 -04002576/**
Al Viro197df042013-09-08 14:03:27 -04002577 * mountpoint_last - look up last component for umount
Jeff Layton80334262013-07-26 06:23:25 -04002578 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2579 * @path: pointer to container for result
2580 *
2581 * This is a special lookup_last function just for umount. In this case, we
2582 * need to resolve the path without doing any revalidation.
2583 *
2584 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2585 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2586 * in almost all cases, this lookup will be served out of the dcache. The only
2587 * cases where it won't are if nd->last refers to a symlink or the path is
2588 * bogus and it doesn't exist.
2589 *
2590 * Returns:
2591 * -error: if there was an error during lookup. This includes -ENOENT if the
2592 * lookup found a negative dentry. The nd->path reference will also be
2593 * put in this case.
2594 *
2595 * 0: if we successfully resolved nd->path and found it to not to be a
2596 * symlink that needs to be followed. "path" will also be populated.
2597 * The nd->path reference will also be put.
2598 *
2599 * 1: if we successfully resolved nd->last and found it to be a symlink
2600 * that needs to be followed. "path" will be populated with the path
2601 * to the link, and nd->path will *not* be put.
2602 */
2603static int
Al Viro197df042013-09-08 14:03:27 -04002604mountpoint_last(struct nameidata *nd, struct path *path)
Jeff Layton80334262013-07-26 06:23:25 -04002605{
2606 int error = 0;
2607 struct dentry *dentry;
2608 struct dentry *dir = nd->path.dentry;
2609
Al Viro35759522013-09-08 13:41:33 -04002610 /* If we're in rcuwalk, drop out of it to handle last component */
2611 if (nd->flags & LOOKUP_RCU) {
Al Viro6e9918b2015-05-05 09:26:05 -04002612 if (unlazy_walk(nd, NULL, 0))
Al Virodeb106c2015-05-08 18:05:21 -04002613 return -ECHILD;
Jeff Layton80334262013-07-26 06:23:25 -04002614 }
2615
2616 nd->flags &= ~LOOKUP_PARENT;
2617
2618 if (unlikely(nd->last_type != LAST_NORM)) {
2619 error = handle_dots(nd, nd->last_type);
Al Viro35759522013-09-08 13:41:33 -04002620 if (error)
Al Virodeb106c2015-05-08 18:05:21 -04002621 return error;
Al Viro35759522013-09-08 13:41:33 -04002622 dentry = dget(nd->path.dentry);
Al Viro949a8522016-03-06 14:20:52 -05002623 } else {
2624 dentry = d_lookup(dir, &nd->last);
Jeff Layton80334262013-07-26 06:23:25 -04002625 if (!dentry) {
Al Viro949a8522016-03-06 14:20:52 -05002626 /*
2627 * No cached dentry. Mounted dentries are pinned in the
2628 * cache, so that means that this dentry is probably
2629 * a symlink or the path doesn't actually point
2630 * to a mounted dentry.
2631 */
2632 dentry = lookup_slow(&nd->last, dir,
2633 nd->flags | LOOKUP_NO_REVAL);
2634 if (IS_ERR(dentry))
2635 return PTR_ERR(dentry);
Dave Jonesbcceeeb2013-09-10 17:04:25 -04002636 }
Jeff Layton80334262013-07-26 06:23:25 -04002637 }
David Howells698934d2015-03-17 17:33:52 +00002638 if (d_is_negative(dentry)) {
Al Viro35759522013-09-08 13:41:33 -04002639 dput(dentry);
Al Virodeb106c2015-05-08 18:05:21 -04002640 return -ENOENT;
Jeff Layton80334262013-07-26 06:23:25 -04002641 }
Al Viro191d7f72015-05-04 08:26:45 -04002642 if (nd->depth)
2643 put_link(nd);
Al Viro35759522013-09-08 13:41:33 -04002644 path->dentry = dentry;
Vasily Averin295dc392014-07-21 12:30:23 +04002645 path->mnt = nd->path.mnt;
Al Viro181548c2015-05-07 19:54:34 -04002646 error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW,
2647 d_backing_inode(dentry), 0);
Al Virodeb106c2015-05-08 18:05:21 -04002648 if (unlikely(error))
Al Virod63ff282015-05-04 18:13:23 -04002649 return error;
Vasily Averin295dc392014-07-21 12:30:23 +04002650 mntget(path->mnt);
Al Viro35759522013-09-08 13:41:33 -04002651 follow_mount(path);
Al Virodeb106c2015-05-08 18:05:21 -04002652 return 0;
Jeff Layton80334262013-07-26 06:23:25 -04002653}
2654
2655/**
Al Viro197df042013-09-08 14:03:27 -04002656 * path_mountpoint - look up a path to be umounted
Masanari Iida2a78b852015-09-09 15:39:23 -07002657 * @nd: lookup context
Jeff Layton80334262013-07-26 06:23:25 -04002658 * @flags: lookup flags
Al Viroc8a53ee2015-05-12 18:43:07 -04002659 * @path: pointer to container for result
Jeff Layton80334262013-07-26 06:23:25 -04002660 *
2661 * Look up the given name, but don't attempt to revalidate the last component.
Randy Dunlap606d6fe2013-10-19 14:56:55 -07002662 * Returns 0 and "path" will be valid on success; Returns error otherwise.
Jeff Layton80334262013-07-26 06:23:25 -04002663 */
2664static int
Al Viroc8a53ee2015-05-12 18:43:07 -04002665path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
Jeff Layton80334262013-07-26 06:23:25 -04002666{
Al Viroc8a53ee2015-05-12 18:43:07 -04002667 const char *s = path_init(nd, flags);
Al Viro368ee9b2015-05-08 17:19:59 -04002668 int err;
2669 if (IS_ERR(s))
2670 return PTR_ERR(s);
Al Viro3bdba282015-05-08 17:37:07 -04002671 while (!(err = link_path_walk(s, nd)) &&
2672 (err = mountpoint_last(nd, path)) > 0) {
2673 s = trailing_symlink(nd);
2674 if (IS_ERR(s)) {
2675 err = PTR_ERR(s);
Jeff Layton80334262013-07-26 06:23:25 -04002676 break;
Al Viro3bdba282015-05-08 17:37:07 -04002677 }
Jeff Layton80334262013-07-26 06:23:25 -04002678 }
Al Virodeb106c2015-05-08 18:05:21 -04002679 terminate_walk(nd);
Jeff Layton80334262013-07-26 06:23:25 -04002680 return err;
2681}
2682
Al Viro2d864652013-09-08 20:18:44 -04002683static int
Al Viro668696d2015-02-22 19:44:00 -05002684filename_mountpoint(int dfd, struct filename *name, struct path *path,
Al Viro2d864652013-09-08 20:18:44 -04002685 unsigned int flags)
2686{
Al Viro9883d182015-05-13 07:28:08 -04002687 struct nameidata nd;
Al Virocbaab2d2015-01-22 02:49:00 -05002688 int error;
Al Viro668696d2015-02-22 19:44:00 -05002689 if (IS_ERR(name))
2690 return PTR_ERR(name);
Al Viro9883d182015-05-13 07:28:08 -04002691 set_nameidata(&nd, dfd, name);
Al Viroc8a53ee2015-05-12 18:43:07 -04002692 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
Al Viro2d864652013-09-08 20:18:44 -04002693 if (unlikely(error == -ECHILD))
Al Viroc8a53ee2015-05-12 18:43:07 -04002694 error = path_mountpoint(&nd, flags, path);
Al Viro2d864652013-09-08 20:18:44 -04002695 if (unlikely(error == -ESTALE))
Al Viroc8a53ee2015-05-12 18:43:07 -04002696 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
Al Viro2d864652013-09-08 20:18:44 -04002697 if (likely(!error))
Al Viro668696d2015-02-22 19:44:00 -05002698 audit_inode(name, path->dentry, 0);
Al Viro9883d182015-05-13 07:28:08 -04002699 restore_nameidata();
Al Viro668696d2015-02-22 19:44:00 -05002700 putname(name);
Al Viro2d864652013-09-08 20:18:44 -04002701 return error;
2702}
2703
Jeff Layton80334262013-07-26 06:23:25 -04002704/**
Al Viro197df042013-09-08 14:03:27 -04002705 * user_path_mountpoint_at - lookup a path from userland in order to umount it
Jeff Layton80334262013-07-26 06:23:25 -04002706 * @dfd: directory file descriptor
2707 * @name: pathname from userland
2708 * @flags: lookup flags
2709 * @path: pointer to container to hold result
2710 *
2711 * A umount is a special case for path walking. We're not actually interested
2712 * in the inode in this situation, and ESTALE errors can be a problem. We
2713 * simply want track down the dentry and vfsmount attached at the mountpoint
2714 * and avoid revalidating the last component.
2715 *
2716 * Returns 0 and populates "path" on success.
2717 */
2718int
Al Viro197df042013-09-08 14:03:27 -04002719user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
Jeff Layton80334262013-07-26 06:23:25 -04002720 struct path *path)
2721{
Al Virocbaab2d2015-01-22 02:49:00 -05002722 return filename_mountpoint(dfd, getname(name), path, flags);
Jeff Layton80334262013-07-26 06:23:25 -04002723}
2724
Al Viro2d864652013-09-08 20:18:44 -04002725int
2726kern_path_mountpoint(int dfd, const char *name, struct path *path,
2727 unsigned int flags)
2728{
Al Virocbaab2d2015-01-22 02:49:00 -05002729 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
Al Viro2d864652013-09-08 20:18:44 -04002730}
2731EXPORT_SYMBOL(kern_path_mountpoint);
2732
Miklos Szeredicbdf35b2014-10-24 00:14:36 +02002733int __check_sticky(struct inode *dir, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002734{
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002735 kuid_t fsuid = current_fsuid();
David Howellsda9592e2008-11-14 10:39:05 +11002736
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002737 if (uid_eq(inode->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002738 return 0;
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002739 if (uid_eq(dir->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002740 return 0;
Andy Lutomirski23adbe12014-06-10 12:45:42 -07002741 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742}
Miklos Szeredicbdf35b2014-10-24 00:14:36 +02002743EXPORT_SYMBOL(__check_sticky);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744
2745/*
2746 * Check whether we can remove a link victim from directory dir, check
2747 * whether the type of victim is right.
2748 * 1. We can't do it if dir is read-only (done in permission())
2749 * 2. We should have write and exec permissions on dir
2750 * 3. We can't remove anything from append-only dir
2751 * 4. We can't do anything with immutable dir (done in permission())
2752 * 5. If the sticky bit on dir is set we should either
2753 * a. be owner of dir, or
2754 * b. be owner of victim, or
2755 * c. have CAP_FOWNER capability
2756 * 6. If the victim is append-only or immutable we can't do antyhing with
2757 * links pointing to it.
Eric W. Biederman0bd23d092016-06-29 14:54:46 -05002758 * 7. If the victim has an unknown uid or gid we can't change the inode.
2759 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2760 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2761 * 10. We can't remove a root or mountpoint.
2762 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
Linus Torvalds1da177e2005-04-16 15:20:36 -07002763 * nfs_async_unlink().
2764 */
David Howellsb18825a2013-09-12 19:22:53 +01002765static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002766{
David Howells63afdfc2015-05-06 15:59:00 +01002767 struct inode *inode = d_backing_inode(victim);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002768 int error;
2769
David Howellsb18825a2013-09-12 19:22:53 +01002770 if (d_is_negative(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002771 return -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01002772 BUG_ON(!inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773
2774 BUG_ON(victim->d_parent->d_inode != dir);
Jeff Layton4fa6b5e2012-10-10 15:25:25 -04002775 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776
Al Virof419a2e2008-07-22 00:07:17 -04002777 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002778 if (error)
2779 return error;
2780 if (IS_APPEND(dir))
2781 return -EPERM;
David Howellsb18825a2013-09-12 19:22:53 +01002782
2783 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
Eric W. Biederman0bd23d092016-06-29 14:54:46 -05002784 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785 return -EPERM;
2786 if (isdir) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002787 if (!d_is_dir(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788 return -ENOTDIR;
2789 if (IS_ROOT(victim))
2790 return -EBUSY;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002791 } else if (d_is_dir(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002792 return -EISDIR;
2793 if (IS_DEADDIR(dir))
2794 return -ENOENT;
2795 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2796 return -EBUSY;
2797 return 0;
2798}
2799
2800/* Check whether we can create an object with dentry child in directory
2801 * dir.
2802 * 1. We can't do it if child already exists (open has special treatment for
2803 * this case, but since we are inlined it's OK)
2804 * 2. We can't do it if dir is read-only (done in permission())
Eric W. Biederman036d5232016-07-01 12:52:06 -05002805 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2806 * 4. We should have write and exec permissions on dir
2807 * 5. We can't do it if dir is immutable (done in permission())
Linus Torvalds1da177e2005-04-16 15:20:36 -07002808 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02002809static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002810{
Eric W. Biederman036d5232016-07-01 12:52:06 -05002811 struct user_namespace *s_user_ns;
Jeff Layton14e972b2013-05-08 10:25:58 -04002812 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002813 if (child->d_inode)
2814 return -EEXIST;
2815 if (IS_DEADDIR(dir))
2816 return -ENOENT;
Eric W. Biederman036d5232016-07-01 12:52:06 -05002817 s_user_ns = dir->i_sb->s_user_ns;
2818 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2819 !kgid_has_mapping(s_user_ns, current_fsgid()))
2820 return -EOVERFLOW;
Al Virof419a2e2008-07-22 00:07:17 -04002821 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822}
2823
Linus Torvalds1da177e2005-04-16 15:20:36 -07002824/*
2825 * p1 and p2 should be directories on the same fs.
2826 */
2827struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2828{
2829 struct dentry *p;
2830
2831 if (p1 == p2) {
Al Viro59551022016-01-22 15:40:57 -05002832 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002833 return NULL;
2834 }
2835
Al Virofc640052016-04-10 01:33:30 -04002836 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002838 p = d_ancestor(p2, p1);
2839 if (p) {
Al Viro59551022016-01-22 15:40:57 -05002840 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2841 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002842 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843 }
2844
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002845 p = d_ancestor(p1, p2);
2846 if (p) {
Al Viro59551022016-01-22 15:40:57 -05002847 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2848 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002849 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002850 }
2851
Al Viro59551022016-01-22 15:40:57 -05002852 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2853 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002854 return NULL;
2855}
Al Viro4d359502014-03-14 12:20:17 -04002856EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857
2858void unlock_rename(struct dentry *p1, struct dentry *p2)
2859{
Al Viro59551022016-01-22 15:40:57 -05002860 inode_unlock(p1->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002861 if (p1 != p2) {
Al Viro59551022016-01-22 15:40:57 -05002862 inode_unlock(p2->d_inode);
Al Virofc640052016-04-10 01:33:30 -04002863 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864 }
2865}
Al Viro4d359502014-03-14 12:20:17 -04002866EXPORT_SYMBOL(unlock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867
Al Viro4acdaf22011-07-26 01:42:34 -04002868int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viro312b63f2012-06-10 18:09:36 -04002869 bool want_excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002870{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002871 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872 if (error)
2873 return error;
2874
Al Viroacfa4382008-12-04 10:06:33 -05002875 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002876 return -EACCES; /* shouldn't it be ENOSYS? */
2877 mode &= S_IALLUGO;
2878 mode |= S_IFREG;
2879 error = security_inode_create(dir, dentry, mode);
2880 if (error)
2881 return error;
Al Viro312b63f2012-06-10 18:09:36 -04002882 error = dir->i_op->create(dir, dentry, mode, want_excl);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002883 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002884 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002885 return error;
2886}
Al Viro4d359502014-03-14 12:20:17 -04002887EXPORT_SYMBOL(vfs_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002888
Eric W. Biedermana2982cc2016-06-09 15:34:02 -05002889bool may_open_dev(const struct path *path)
2890{
2891 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2892 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2893}
2894
Al Viro73d049a2011-03-11 12:08:24 -05002895static int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002896{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002897 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002898 struct inode *inode = dentry->d_inode;
2899 int error;
2900
2901 if (!inode)
2902 return -ENOENT;
2903
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002904 switch (inode->i_mode & S_IFMT) {
2905 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002906 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002907 case S_IFDIR:
2908 if (acc_mode & MAY_WRITE)
2909 return -EISDIR;
2910 break;
2911 case S_IFBLK:
2912 case S_IFCHR:
Eric W. Biedermana2982cc2016-06-09 15:34:02 -05002913 if (!may_open_dev(path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002914 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002915 /*FALLTHRU*/
2916 case S_IFIFO:
2917 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002918 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002919 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002920 }
Dave Hansenb41572e2007-10-16 23:31:14 -07002921
Al Viro62fb4a12015-12-26 22:33:24 -05002922 error = inode_permission(inode, MAY_OPEN | acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07002923 if (error)
2924 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05002925
Linus Torvalds1da177e2005-04-16 15:20:36 -07002926 /*
2927 * An append-only file must be opened in append mode for writing.
2928 */
2929 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05002930 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05002931 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002932 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05002933 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002934 }
2935
2936 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002937 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05002938 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002939
J. Bruce Fieldsf3c7691e2011-09-21 10:58:13 -04002940 return 0;
Al Viro7715b522009-12-16 03:54:00 -05002941}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002942
Jeff Laytone1181ee2010-12-07 16:19:50 -05002943static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002944{
Jeff Laytone1181ee2010-12-07 16:19:50 -05002945 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002946 struct inode *inode = path->dentry->d_inode;
2947 int error = get_write_access(inode);
2948 if (error)
2949 return error;
2950 /*
2951 * Refuse to truncate files with mandatory locks held on them.
2952 */
Jeff Laytond7a06982014-03-10 09:54:15 -04002953 error = locks_verify_locked(filp);
Al Viro7715b522009-12-16 03:54:00 -05002954 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002955 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002956 if (!error) {
2957 error = do_truncate(path->dentry, 0,
2958 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002959 filp);
Al Viro7715b522009-12-16 03:54:00 -05002960 }
2961 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002962 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002963}
2964
Dave Hansend57999e2008-02-15 14:37:27 -08002965static inline int open_to_namei_flags(int flag)
2966{
Al Viro8a5e9292011-06-25 19:15:54 -04002967 if ((flag & O_ACCMODE) == 3)
2968 flag--;
Dave Hansend57999e2008-02-15 14:37:27 -08002969 return flag;
2970}
2971
Al Virod3607752016-03-25 15:21:09 -04002972static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002973{
Seth Forshee2e727b82017-10-07 22:36:55 +00002974 struct user_namespace *s_user_ns;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002975 int error = security_path_mknod(dir, dentry, mode, 0);
2976 if (error)
2977 return error;
2978
Seth Forshee2e727b82017-10-07 22:36:55 +00002979 s_user_ns = dir->dentry->d_sb->s_user_ns;
2980 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2981 !kgid_has_mapping(s_user_ns, current_fsgid()))
2982 return -EOVERFLOW;
2983
Miklos Szeredid18e9002012-06-05 15:10:17 +02002984 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2985 if (error)
2986 return error;
2987
2988 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2989}
2990
David Howells1acf0af2012-06-14 16:13:46 +01002991/*
2992 * Attempt to atomically look up, create and open a file from a negative
2993 * dentry.
2994 *
2995 * Returns 0 if successful. The file will have been created and attached to
2996 * @file by the filesystem calling finish_open().
2997 *
2998 * Returns 1 if the file was looked up only or didn't need creating. The
2999 * caller will need to perform the open themselves. @path will have been
3000 * updated to point to the new dentry. This may be negative.
3001 *
3002 * Returns an error code otherwise.
3003 */
Al Viro2675a4e2012-06-22 12:41:10 +04003004static int atomic_open(struct nameidata *nd, struct dentry *dentry,
3005 struct path *path, struct file *file,
3006 const struct open_flags *op,
Al Viro1643b432016-04-27 19:14:10 -04003007 int open_flag, umode_t mode,
Al Viro2675a4e2012-06-22 12:41:10 +04003008 int *opened)
Miklos Szeredid18e9002012-06-05 15:10:17 +02003009{
Miklos Szeredid18e9002012-06-05 15:10:17 +02003010 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003011 struct inode *dir = nd->path.dentry->d_inode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003012 int error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003013
Al Viro384f26e2016-04-28 02:03:55 -04003014 if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */
Miklos Szeredid18e9002012-06-05 15:10:17 +02003015 open_flag &= ~O_TRUNC;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003016
Miklos Szeredid18e9002012-06-05 15:10:17 +02003017 if (nd->flags & LOOKUP_DIRECTORY)
3018 open_flag |= O_DIRECTORY;
3019
Al Viro30d90492012-06-22 12:40:19 +04003020 file->f_path.dentry = DENTRY_NOT_SET;
3021 file->f_path.mnt = nd->path.mnt;
Al Viro0fb1ea02016-04-27 14:13:10 -04003022 error = dir->i_op->atomic_open(dir, dentry, file,
3023 open_to_namei_flags(open_flag),
3024 mode, opened);
Al Viro6fbd0712016-04-28 11:50:59 -04003025 d_lookup_done(dentry);
Al Viro384f26e2016-04-28 02:03:55 -04003026 if (!error) {
3027 /*
3028 * We didn't have the inode before the open, so check open
3029 * permission here.
3030 */
3031 int acc_mode = op->acc_mode;
3032 if (*opened & FILE_CREATED) {
3033 WARN_ON(!(open_flag & O_CREAT));
3034 fsnotify_create(dir, dentry);
3035 acc_mode = 0;
3036 }
3037 error = may_open(&file->f_path, acc_mode, open_flag);
3038 if (WARN_ON(error > 0))
3039 error = -EINVAL;
3040 } else if (error > 0) {
Al Viro30d90492012-06-22 12:40:19 +04003041 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
Al Viro2675a4e2012-06-22 12:41:10 +04003042 error = -EIO;
Al Viro03da6332013-09-16 19:22:33 -04003043 } else {
Al Viro384f26e2016-04-28 02:03:55 -04003044 if (file->f_path.dentry) {
3045 dput(dentry);
3046 dentry = file->f_path.dentry;
Al Viro03da6332013-09-16 19:22:33 -04003047 }
Al Viro384f26e2016-04-28 02:03:55 -04003048 if (*opened & FILE_CREATED)
3049 fsnotify_create(dir, dentry);
Al Viroa01e7182016-06-07 21:53:51 -04003050 if (unlikely(d_is_negative(dentry))) {
3051 error = -ENOENT;
3052 } else {
3053 path->dentry = dentry;
3054 path->mnt = nd->path.mnt;
3055 return 1;
3056 }
Sage Weil62b2ce92012-08-15 13:30:12 -07003057 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02003058 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02003059 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04003060 return error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003061}
3062
Nick Piggin31e6b012011-01-07 17:49:52 +11003063/*
David Howells1acf0af2012-06-14 16:13:46 +01003064 * Look up and maybe create and open the last component.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003065 *
3066 * Must be called with i_mutex held on parent.
3067 *
David Howells1acf0af2012-06-14 16:13:46 +01003068 * Returns 0 if the file was successfully atomically created (if necessary) and
3069 * opened. In this case the file will be returned attached to @file.
3070 *
3071 * Returns 1 if the file was not completely opened at this time, though lookups
3072 * and creations will have been performed and the dentry returned in @path will
3073 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
3074 * specified then a negative dentry may be returned.
3075 *
3076 * An error code is returned otherwise.
3077 *
3078 * FILE_CREATE will be set in @*opened if the dentry was created and will be
3079 * cleared otherwise prior to returning.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003080 */
Al Viro2675a4e2012-06-22 12:41:10 +04003081static int lookup_open(struct nameidata *nd, struct path *path,
3082 struct file *file,
3083 const struct open_flags *op,
Al Viro64894cf2012-07-31 00:53:35 +04003084 bool got_write, int *opened)
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003085{
3086 struct dentry *dir = nd->path.dentry;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02003087 struct inode *dir_inode = dir->d_inode;
Al Viro1643b432016-04-27 19:14:10 -04003088 int open_flag = op->open_flag;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003089 struct dentry *dentry;
Al Viro1643b432016-04-27 19:14:10 -04003090 int error, create_error = 0;
Al Viro1643b432016-04-27 19:14:10 -04003091 umode_t mode = op->mode;
Al Viro6fbd0712016-04-28 11:50:59 -04003092 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003093
Al Viroce8644f2016-04-26 14:17:56 -04003094 if (unlikely(IS_DEADDIR(dir_inode)))
3095 return -ENOENT;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003096
Al Viro47237682012-06-10 05:01:45 -04003097 *opened &= ~FILE_CREATED;
Al Viro6fbd0712016-04-28 11:50:59 -04003098 dentry = d_lookup(dir, &nd->last);
3099 for (;;) {
3100 if (!dentry) {
3101 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3102 if (IS_ERR(dentry))
3103 return PTR_ERR(dentry);
3104 }
3105 if (d_in_lookup(dentry))
3106 break;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003107
Al Viro6fbd0712016-04-28 11:50:59 -04003108 if (!(dentry->d_flags & DCACHE_OP_REVALIDATE))
3109 break;
3110
3111 error = d_revalidate(dentry, nd->flags);
3112 if (likely(error > 0))
3113 break;
3114 if (error)
3115 goto out_dput;
3116 d_invalidate(dentry);
3117 dput(dentry);
3118 dentry = NULL;
3119 }
3120 if (dentry->d_inode) {
Al Viro6c51e512016-03-05 20:09:32 -05003121 /* Cached positive dentry: will open in f_op->open */
Miklos Szeredid18e9002012-06-05 15:10:17 +02003122 goto out_no_open;
Al Viro6c51e512016-03-05 20:09:32 -05003123 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02003124
Al Viro1643b432016-04-27 19:14:10 -04003125 /*
3126 * Checking write permission is tricky, bacuse we don't know if we are
3127 * going to actually need it: O_CREAT opens should work as long as the
3128 * file exists. But checking existence breaks atomicity. The trick is
3129 * to check access and if not granted clear O_CREAT from the flags.
3130 *
3131 * Another problem is returing the "right" error value (e.g. for an
3132 * O_EXCL open we want to return EEXIST not EROFS).
3133 */
3134 if (open_flag & O_CREAT) {
3135 if (!IS_POSIXACL(dir->d_inode))
3136 mode &= ~current_umask();
3137 if (unlikely(!got_write)) {
3138 create_error = -EROFS;
3139 open_flag &= ~O_CREAT;
3140 if (open_flag & (O_EXCL | O_TRUNC))
3141 goto no_open;
3142 /* No side effects, safe to clear O_CREAT */
3143 } else {
3144 create_error = may_o_create(&nd->path, dentry, mode);
3145 if (create_error) {
3146 open_flag &= ~O_CREAT;
3147 if (open_flag & O_EXCL)
3148 goto no_open;
3149 }
3150 }
3151 } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3152 unlikely(!got_write)) {
3153 /*
3154 * No O_CREATE -> atomicity not a requirement -> fall
3155 * back to lookup + open
3156 */
3157 goto no_open;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003158 }
3159
Al Viro1643b432016-04-27 19:14:10 -04003160 if (dir_inode->i_op->atomic_open) {
3161 error = atomic_open(nd, dentry, path, file, op, open_flag,
3162 mode, opened);
3163 if (unlikely(error == -ENOENT) && create_error)
3164 error = create_error;
3165 return error;
3166 }
Miklos Szeredi54ef4872012-06-05 15:10:16 +02003167
Al Viro1643b432016-04-27 19:14:10 -04003168no_open:
Al Viro6fbd0712016-04-28 11:50:59 -04003169 if (d_in_lookup(dentry)) {
Al Viro12fa5e22016-04-28 11:19:43 -04003170 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3171 nd->flags);
Al Viro6fbd0712016-04-28 11:50:59 -04003172 d_lookup_done(dentry);
Al Viro12fa5e22016-04-28 11:19:43 -04003173 if (unlikely(res)) {
3174 if (IS_ERR(res)) {
3175 error = PTR_ERR(res);
3176 goto out_dput;
3177 }
3178 dput(dentry);
3179 dentry = res;
3180 }
Miklos Szeredi54ef4872012-06-05 15:10:16 +02003181 }
3182
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003183 /* Negative dentry, just create the file */
Al Viro1643b432016-04-27 19:14:10 -04003184 if (!dentry->d_inode && (open_flag & O_CREAT)) {
Al Viro47237682012-06-10 05:01:45 -04003185 *opened |= FILE_CREATED;
Al Viroce8644f2016-04-26 14:17:56 -04003186 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
Al Viroce8644f2016-04-26 14:17:56 -04003187 if (!dir_inode->i_op->create) {
3188 error = -EACCES;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003189 goto out_dput;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003190 }
Al Viroce8644f2016-04-26 14:17:56 -04003191 error = dir_inode->i_op->create(dir_inode, dentry, mode,
Al Viro1643b432016-04-27 19:14:10 -04003192 open_flag & O_EXCL);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003193 if (error)
3194 goto out_dput;
Al Viroce8644f2016-04-26 14:17:56 -04003195 fsnotify_create(dir_inode, dentry);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003196 }
Al Viro1643b432016-04-27 19:14:10 -04003197 if (unlikely(create_error) && !dentry->d_inode) {
3198 error = create_error;
3199 goto out_dput;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003200 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02003201out_no_open:
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003202 path->dentry = dentry;
3203 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04003204 return 1;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003205
3206out_dput:
3207 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04003208 return error;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003209}
3210
3211/*
Al Virofe2d35f2011-03-05 22:58:25 -05003212 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11003213 */
Al Viro896475d2015-04-22 18:02:17 -04003214static int do_last(struct nameidata *nd,
Al Viro2675a4e2012-06-22 12:41:10 +04003215 struct file *file, const struct open_flags *op,
Al Viro76ae2a52015-05-12 18:44:32 -04003216 int *opened)
Al Virofb1cc552009-12-24 01:58:28 -05003217{
Al Viroa1e28032009-12-24 02:12:06 -05003218 struct dentry *dir = nd->path.dentry;
Al Viroca344a892011-03-09 00:36:45 -05003219 int open_flag = op->open_flag;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003220 bool will_truncate = (open_flag & O_TRUNC) != 0;
Al Viro64894cf2012-07-31 00:53:35 +04003221 bool got_write = false;
Al Virobcda7652011-03-13 16:42:14 -04003222 int acc_mode = op->acc_mode;
Al Viro254cf582015-05-05 09:40:46 -04003223 unsigned seq;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02003224 struct inode *inode;
Al Viro896475d2015-04-22 18:02:17 -04003225 struct path path;
Al Viro16c2cd72011-02-22 15:50:10 -05003226 int error;
Al Virofb1cc552009-12-24 01:58:28 -05003227
Al Viroc3e380b2011-02-23 13:39:45 -05003228 nd->flags &= ~LOOKUP_PARENT;
3229 nd->flags |= op->intent;
3230
Al Virobc77daa2013-06-06 09:12:33 -04003231 if (nd->last_type != LAST_NORM) {
Al Virofe2d35f2011-03-05 22:58:25 -05003232 error = handle_dots(nd, nd->last_type);
Al Virodeb106c2015-05-08 18:05:21 -04003233 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04003234 return error;
Miklos Szeredie83db162012-06-05 15:10:29 +02003235 goto finish_open;
Al Viro1f36f772009-12-26 10:56:19 -05003236 }
Al Viro67ee3ad2009-12-26 07:01:01 -05003237
Al Viroca344a892011-03-09 00:36:45 -05003238 if (!(open_flag & O_CREAT)) {
Al Virofe2d35f2011-03-05 22:58:25 -05003239 if (nd->last.name[nd->last.len])
3240 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3241 /* we _can_ be in RCU mode here */
Al Viro254cf582015-05-05 09:40:46 -04003242 error = lookup_fast(nd, &path, &inode, &seq);
Al Viroe9742b52016-03-05 22:04:59 -05003243 if (likely(error > 0))
Miklos Szeredi71574862012-06-05 15:10:14 +02003244 goto finish_lookup;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02003245
Miklos Szeredi71574862012-06-05 15:10:14 +02003246 if (error < 0)
Al Virodeb106c2015-05-08 18:05:21 -04003247 return error;
Miklos Szeredi37d7fff2012-06-05 15:10:12 +02003248
Miklos Szeredi71574862012-06-05 15:10:14 +02003249 BUG_ON(nd->inode != dir->d_inode);
Al Viro6583fe22016-03-05 18:14:03 -05003250 BUG_ON(nd->flags & LOOKUP_RCU);
Miklos Szeredib6183df2012-06-05 15:10:13 +02003251 } else {
3252 /* create side of things */
3253 /*
3254 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3255 * has been cleared when we got to the last component we are
3256 * about to look up
3257 */
3258 error = complete_walk(nd);
Al Viroe8bb73d2015-05-08 16:28:42 -04003259 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003260 return error;
Miklos Szeredib6183df2012-06-05 15:10:13 +02003261
Al Viro76ae2a52015-05-12 18:44:32 -04003262 audit_inode(nd->name, dir, LOOKUP_PARENT);
Miklos Szeredib6183df2012-06-05 15:10:13 +02003263 /* trailing slashes? */
Al Virodeb106c2015-05-08 18:05:21 -04003264 if (unlikely(nd->last.name[nd->last.len]))
3265 return -EISDIR;
Al Virofe2d35f2011-03-05 22:58:25 -05003266 }
3267
Al Viro9cf843e2016-04-28 19:35:16 -04003268 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
Al Viro64894cf2012-07-31 00:53:35 +04003269 error = mnt_want_write(nd->path.mnt);
3270 if (!error)
3271 got_write = true;
3272 /*
3273 * do _not_ fail yet - we might not need that or fail with
3274 * a different error; let lookup_open() decide; we'll be
3275 * dropping this one anyway.
3276 */
3277 }
Al Viro9cf843e2016-04-28 19:35:16 -04003278 if (open_flag & O_CREAT)
3279 inode_lock(dir->d_inode);
3280 else
3281 inode_lock_shared(dir->d_inode);
Al Viro896475d2015-04-22 18:02:17 -04003282 error = lookup_open(nd, &path, file, op, got_write, opened);
Al Viro9cf843e2016-04-28 19:35:16 -04003283 if (open_flag & O_CREAT)
3284 inode_unlock(dir->d_inode);
3285 else
3286 inode_unlock_shared(dir->d_inode);
Al Viroa1e28032009-12-24 02:12:06 -05003287
Al Viro2675a4e2012-06-22 12:41:10 +04003288 if (error <= 0) {
3289 if (error)
Miklos Szeredid18e9002012-06-05 15:10:17 +02003290 goto out;
3291
Al Viro47237682012-06-10 05:01:45 -04003292 if ((*opened & FILE_CREATED) ||
Al Viro496ad9a2013-01-23 17:07:38 -05003293 !S_ISREG(file_inode(file)->i_mode))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003294 will_truncate = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003295
Al Viro76ae2a52015-05-12 18:44:32 -04003296 audit_inode(nd->name, file->f_path.dentry, 0);
Miklos Szeredid18e9002012-06-05 15:10:17 +02003297 goto opened;
3298 }
Al Virofb1cc552009-12-24 01:58:28 -05003299
Al Viro47237682012-06-10 05:01:45 -04003300 if (*opened & FILE_CREATED) {
Al Viro9b44f1b2011-03-09 00:17:27 -05003301 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05003302 open_flag &= ~O_TRUNC;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003303 will_truncate = false;
Al Viro62fb4a12015-12-26 22:33:24 -05003304 acc_mode = 0;
Al Viro896475d2015-04-22 18:02:17 -04003305 path_to_nameidata(&path, nd);
Miklos Szeredie83db162012-06-05 15:10:29 +02003306 goto finish_open_created;
Al Virofb1cc552009-12-24 01:58:28 -05003307 }
3308
3309 /*
Miklos Szeredid18e9002012-06-05 15:10:17 +02003310 * If atomic_open() acquired write access it is dropped now due to
3311 * possible mount and symlink following (this might be optimized away if
3312 * necessary...)
3313 */
Al Viro64894cf2012-07-31 00:53:35 +04003314 if (got_write) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02003315 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04003316 got_write = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003317 }
3318
Al Viroe6ec03a2016-06-05 00:23:09 -04003319 error = follow_managed(&path, nd);
3320 if (unlikely(error < 0))
3321 return error;
3322
Al Viro6583fe22016-03-05 18:14:03 -05003323 if (unlikely(d_is_negative(path.dentry))) {
3324 path_to_nameidata(&path, nd);
3325 return -ENOENT;
3326 }
3327
3328 /*
3329 * create/update audit record if it already exists.
3330 */
3331 audit_inode(nd->name, path.dentry, 0);
3332
Al Virodeb106c2015-05-08 18:05:21 -04003333 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3334 path_to_nameidata(&path, nd);
3335 return -EEXIST;
3336 }
Al Virofb1cc552009-12-24 01:58:28 -05003337
Al Viro254cf582015-05-05 09:40:46 -04003338 seq = 0; /* out of RCU mode, so the value doesn't matter */
Al Virod4565642016-02-27 19:23:16 -05003339 inode = d_backing_inode(path.dentry);
Al Viro766c4cb2015-05-07 19:24:57 -04003340finish_lookup:
Al Virod63ff282015-05-04 18:13:23 -04003341 if (nd->depth)
3342 put_link(nd);
Al Viro181548c2015-05-07 19:54:34 -04003343 error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW,
3344 inode, seq);
Al Virodeb106c2015-05-08 18:05:21 -04003345 if (unlikely(error))
Al Virod63ff282015-05-04 18:13:23 -04003346 return error;
Al Virofb1cc552009-12-24 01:58:28 -05003347
Al Virofac7d192016-06-04 11:41:49 -04003348 path_to_nameidata(&path, nd);
Miklos Szeredidecf3402012-05-21 17:30:08 +02003349 nd->inode = inode;
Al Viro254cf582015-05-05 09:40:46 -04003350 nd->seq = seq;
Al Viroa3fbbde2011-11-07 21:21:26 +00003351 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
Al Virobc77daa2013-06-06 09:12:33 -04003352finish_open:
Al Viroa3fbbde2011-11-07 21:21:26 +00003353 error = complete_walk(nd);
Al Virofac7d192016-06-04 11:41:49 -04003354 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003355 return error;
Al Viro76ae2a52015-05-12 18:44:32 -04003356 audit_inode(nd->name, nd->path.dentry, 0);
Al Virofb1cc552009-12-24 01:58:28 -05003357 error = -EISDIR;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003358 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04003359 goto out;
Miklos Szerediaf2f5542012-05-21 17:30:11 +02003360 error = -ENOTDIR;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003361 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04003362 goto out;
David Howells4bbcbd32015-03-17 22:16:40 +00003363 if (!d_is_reg(nd->path.dentry))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003364 will_truncate = false;
Al Viro6c0d46c2011-03-09 00:59:59 -05003365
Al Viro0f9d1a12011-03-09 00:13:14 -05003366 if (will_truncate) {
3367 error = mnt_want_write(nd->path.mnt);
3368 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003369 goto out;
Al Viro64894cf2012-07-31 00:53:35 +04003370 got_write = true;
Al Viro0f9d1a12011-03-09 00:13:14 -05003371 }
Miklos Szeredie83db162012-06-05 15:10:29 +02003372finish_open_created:
Al Viro6ac08702016-04-26 00:02:50 -04003373 error = may_open(&nd->path, acc_mode, open_flag);
3374 if (error)
3375 goto out;
Miklos Szeredi4aa7c632014-10-24 00:14:35 +02003376 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3377 error = vfs_open(&nd->path, file, current_cred());
Al Virofac7d192016-06-04 11:41:49 -04003378 if (error)
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003379 goto out;
Al Virofac7d192016-06-04 11:41:49 -04003380 *opened |= FILE_OPENED;
Miklos Szeredia8277b92012-06-05 15:10:32 +02003381opened:
Al Viro2675a4e2012-06-22 12:41:10 +04003382 error = open_check_o_direct(file);
Al Virofe9ec822016-04-27 03:14:20 -04003383 if (!error)
3384 error = ima_file_check(file, op->acc_mode, *opened);
3385 if (!error && will_truncate)
Al Viro2675a4e2012-06-22 12:41:10 +04003386 error = handle_truncate(file);
Al Viroca344a892011-03-09 00:36:45 -05003387out:
Al Virofe9ec822016-04-27 03:14:20 -04003388 if (unlikely(error) && (*opened & FILE_OPENED))
3389 fput(file);
Al Viroc80567c2016-02-27 19:17:33 -05003390 if (unlikely(error > 0)) {
3391 WARN_ON(1);
3392 error = -EINVAL;
3393 }
Al Viro64894cf2012-07-31 00:53:35 +04003394 if (got_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05003395 mnt_drop_write(nd->path.mnt);
Al Viro2675a4e2012-06-22 12:41:10 +04003396 return error;
Al Virofb1cc552009-12-24 01:58:28 -05003397}
3398
Al Viroc8a53ee2015-05-12 18:43:07 -04003399static int do_tmpfile(struct nameidata *nd, unsigned flags,
Al Viro60545d02013-06-07 01:20:27 -04003400 const struct open_flags *op,
3401 struct file *file, int *opened)
3402{
3403 static const struct qstr name = QSTR_INIT("/", 1);
Al Viro625b6d12015-05-12 16:36:12 -04003404 struct dentry *child;
Al Viro60545d02013-06-07 01:20:27 -04003405 struct inode *dir;
Al Viro625b6d12015-05-12 16:36:12 -04003406 struct path path;
Al Viroc8a53ee2015-05-12 18:43:07 -04003407 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
Al Viro60545d02013-06-07 01:20:27 -04003408 if (unlikely(error))
3409 return error;
Al Viro625b6d12015-05-12 16:36:12 -04003410 error = mnt_want_write(path.mnt);
Al Viro60545d02013-06-07 01:20:27 -04003411 if (unlikely(error))
3412 goto out;
Al Viro625b6d12015-05-12 16:36:12 -04003413 dir = path.dentry->d_inode;
Al Viro60545d02013-06-07 01:20:27 -04003414 /* we want directory to be writable */
Al Viro625b6d12015-05-12 16:36:12 -04003415 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Al Viro60545d02013-06-07 01:20:27 -04003416 if (error)
3417 goto out2;
Al Viro60545d02013-06-07 01:20:27 -04003418 if (!dir->i_op->tmpfile) {
3419 error = -EOPNOTSUPP;
3420 goto out2;
3421 }
Al Viro625b6d12015-05-12 16:36:12 -04003422 child = d_alloc(path.dentry, &name);
Al Viro60545d02013-06-07 01:20:27 -04003423 if (unlikely(!child)) {
3424 error = -ENOMEM;
3425 goto out2;
3426 }
Al Viro625b6d12015-05-12 16:36:12 -04003427 dput(path.dentry);
3428 path.dentry = child;
3429 error = dir->i_op->tmpfile(dir, child, op->mode);
Al Viro60545d02013-06-07 01:20:27 -04003430 if (error)
3431 goto out2;
Al Viroc8a53ee2015-05-12 18:43:07 -04003432 audit_inode(nd->name, child, 0);
Eric Rannaud69a91c22014-10-30 01:51:01 -07003433 /* Don't check for other permissions, the inode was just created */
Al Viro62fb4a12015-12-26 22:33:24 -05003434 error = may_open(&path, 0, op->open_flag);
Al Viro60545d02013-06-07 01:20:27 -04003435 if (error)
3436 goto out2;
Al Viro625b6d12015-05-12 16:36:12 -04003437 file->f_path.mnt = path.mnt;
3438 error = finish_open(file, child, NULL, opened);
Al Viro60545d02013-06-07 01:20:27 -04003439 if (error)
3440 goto out2;
3441 error = open_check_o_direct(file);
Al Virof4e0c302013-06-11 08:34:36 +04003442 if (error) {
Al Viro60545d02013-06-07 01:20:27 -04003443 fput(file);
Al Virof4e0c302013-06-11 08:34:36 +04003444 } else if (!(op->open_flag & O_EXCL)) {
3445 struct inode *inode = file_inode(file);
3446 spin_lock(&inode->i_lock);
3447 inode->i_state |= I_LINKABLE;
3448 spin_unlock(&inode->i_lock);
3449 }
Al Viro60545d02013-06-07 01:20:27 -04003450out2:
Al Viro625b6d12015-05-12 16:36:12 -04003451 mnt_drop_write(path.mnt);
Al Viro60545d02013-06-07 01:20:27 -04003452out:
Al Viro625b6d12015-05-12 16:36:12 -04003453 path_put(&path);
Al Viro60545d02013-06-07 01:20:27 -04003454 return error;
3455}
3456
Al Viro6ac08702016-04-26 00:02:50 -04003457static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3458{
3459 struct path path;
3460 int error = path_lookupat(nd, flags, &path);
3461 if (!error) {
3462 audit_inode(nd->name, path.dentry, 0);
3463 error = vfs_open(&path, file, current_cred());
3464 path_put(&path);
3465 }
3466 return error;
3467}
3468
Al Viroc8a53ee2015-05-12 18:43:07 -04003469static struct file *path_openat(struct nameidata *nd,
3470 const struct open_flags *op, unsigned flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003471{
Al Viro368ee9b2015-05-08 17:19:59 -04003472 const char *s;
Al Viro30d90492012-06-22 12:40:19 +04003473 struct file *file;
Al Viro47237682012-06-10 05:01:45 -04003474 int opened = 0;
Al Viro13aab422011-02-23 17:54:08 -05003475 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11003476
Al Viro30d90492012-06-22 12:40:19 +04003477 file = get_empty_filp();
Al Viro1afc99b2013-02-14 20:41:04 -05003478 if (IS_ERR(file))
3479 return file;
Nick Piggin31e6b012011-01-07 17:49:52 +11003480
Al Viro30d90492012-06-22 12:40:19 +04003481 file->f_flags = op->open_flag;
Nick Piggin31e6b012011-01-07 17:49:52 +11003482
Al Virobb458c62013-07-13 13:26:37 +04003483 if (unlikely(file->f_flags & __O_TMPFILE)) {
Al Viroc8a53ee2015-05-12 18:43:07 -04003484 error = do_tmpfile(nd, flags, op, file, &opened);
Al Virof15133d2015-05-08 22:53:15 -04003485 goto out2;
Al Viro60545d02013-06-07 01:20:27 -04003486 }
3487
Al Viro6ac08702016-04-26 00:02:50 -04003488 if (unlikely(file->f_flags & O_PATH)) {
3489 error = do_o_path(nd, flags, file);
3490 if (!error)
3491 opened |= FILE_OPENED;
3492 goto out2;
3493 }
3494
Al Viroc8a53ee2015-05-12 18:43:07 -04003495 s = path_init(nd, flags);
Al Viro368ee9b2015-05-08 17:19:59 -04003496 if (IS_ERR(s)) {
3497 put_filp(file);
3498 return ERR_CAST(s);
3499 }
Al Viro3bdba282015-05-08 17:37:07 -04003500 while (!(error = link_path_walk(s, nd)) &&
Al Viro76ae2a52015-05-12 18:44:32 -04003501 (error = do_last(nd, file, op, &opened)) > 0) {
Al Viro73d049a2011-03-11 12:08:24 -05003502 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
Al Viro3bdba282015-05-08 17:37:07 -04003503 s = trailing_symlink(nd);
3504 if (IS_ERR(s)) {
3505 error = PTR_ERR(s);
Al Viro2675a4e2012-06-22 12:41:10 +04003506 break;
Al Viro3bdba282015-05-08 17:37:07 -04003507 }
Al Viro806b6812009-12-26 07:16:40 -05003508 }
Al Virodeb106c2015-05-08 18:05:21 -04003509 terminate_walk(nd);
Al Virof15133d2015-05-08 22:53:15 -04003510out2:
Al Viro2675a4e2012-06-22 12:41:10 +04003511 if (!(opened & FILE_OPENED)) {
3512 BUG_ON(!error);
Al Viro30d90492012-06-22 12:40:19 +04003513 put_filp(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003514 }
Al Viro2675a4e2012-06-22 12:41:10 +04003515 if (unlikely(error)) {
3516 if (error == -EOPENSTALE) {
3517 if (flags & LOOKUP_RCU)
3518 error = -ECHILD;
3519 else
3520 error = -ESTALE;
3521 }
3522 file = ERR_PTR(error);
3523 }
3524 return file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003525}
3526
Jeff Layton669abf42012-10-10 16:43:10 -04003527struct file *do_filp_open(int dfd, struct filename *pathname,
Al Virof9652e12013-06-11 08:23:01 +04003528 const struct open_flags *op)
Al Viro13aab422011-02-23 17:54:08 -05003529{
Al Viro9883d182015-05-13 07:28:08 -04003530 struct nameidata nd;
Al Virof9652e12013-06-11 08:23:01 +04003531 int flags = op->lookup_flags;
Al Viro13aab422011-02-23 17:54:08 -05003532 struct file *filp;
3533
Al Viro9883d182015-05-13 07:28:08 -04003534 set_nameidata(&nd, dfd, pathname);
Al Viroc8a53ee2015-05-12 18:43:07 -04003535 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
Al Viro13aab422011-02-23 17:54:08 -05003536 if (unlikely(filp == ERR_PTR(-ECHILD)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003537 filp = path_openat(&nd, op, flags);
Al Viro13aab422011-02-23 17:54:08 -05003538 if (unlikely(filp == ERR_PTR(-ESTALE)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003539 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
Al Viro9883d182015-05-13 07:28:08 -04003540 restore_nameidata();
Al Viro13aab422011-02-23 17:54:08 -05003541 return filp;
3542}
3543
Al Viro73d049a2011-03-11 12:08:24 -05003544struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
Al Virof9652e12013-06-11 08:23:01 +04003545 const char *name, const struct open_flags *op)
Al Viro73d049a2011-03-11 12:08:24 -05003546{
Al Viro9883d182015-05-13 07:28:08 -04003547 struct nameidata nd;
Al Viro73d049a2011-03-11 12:08:24 -05003548 struct file *file;
Paul Moore51689102015-01-22 00:00:03 -05003549 struct filename *filename;
Al Virof9652e12013-06-11 08:23:01 +04003550 int flags = op->lookup_flags | LOOKUP_ROOT;
Al Viro73d049a2011-03-11 12:08:24 -05003551
3552 nd.root.mnt = mnt;
3553 nd.root.dentry = dentry;
3554
David Howellsb18825a2013-09-12 19:22:53 +01003555 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
Al Viro73d049a2011-03-11 12:08:24 -05003556 return ERR_PTR(-ELOOP);
3557
Paul Moore51689102015-01-22 00:00:03 -05003558 filename = getname_kernel(name);
Viresh Kumara1c83682015-08-12 15:59:44 +05303559 if (IS_ERR(filename))
Paul Moore51689102015-01-22 00:00:03 -05003560 return ERR_CAST(filename);
3561
Al Viro9883d182015-05-13 07:28:08 -04003562 set_nameidata(&nd, -1, filename);
Al Viroc8a53ee2015-05-12 18:43:07 -04003563 file = path_openat(&nd, op, flags | LOOKUP_RCU);
Al Viro73d049a2011-03-11 12:08:24 -05003564 if (unlikely(file == ERR_PTR(-ECHILD)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003565 file = path_openat(&nd, op, flags);
Al Viro73d049a2011-03-11 12:08:24 -05003566 if (unlikely(file == ERR_PTR(-ESTALE)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003567 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
Al Viro9883d182015-05-13 07:28:08 -04003568 restore_nameidata();
Paul Moore51689102015-01-22 00:00:03 -05003569 putname(filename);
Al Viro73d049a2011-03-11 12:08:24 -05003570 return file;
3571}
3572
Al Virofa14a0b2015-01-22 02:16:49 -05003573static struct dentry *filename_create(int dfd, struct filename *name,
Jeff Layton1ac12b42012-12-11 12:10:06 -05003574 struct path *path, unsigned int lookup_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003575{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003576 struct dentry *dentry = ERR_PTR(-EEXIST);
Al Viro391172c2015-05-09 11:19:16 -04003577 struct qstr last;
3578 int type;
Jan Karac30dabf2012-06-12 16:20:30 +02003579 int err2;
Jeff Layton1ac12b42012-12-11 12:10:06 -05003580 int error;
3581 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3582
3583 /*
3584 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3585 * other flags passed in are ignored!
3586 */
3587 lookup_flags &= LOOKUP_REVAL;
3588
Al Viro5c31b6c2015-05-12 17:32:54 -04003589 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3590 if (IS_ERR(name))
3591 return ERR_CAST(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003592
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003593 /*
3594 * Yucky last component or no last component at all?
3595 * (foo/., foo/.., /////)
3596 */
Al Viro5c31b6c2015-05-12 17:32:54 -04003597 if (unlikely(type != LAST_NORM))
Al Viroed75e952011-06-27 16:53:43 -04003598 goto out;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003599
Jan Karac30dabf2012-06-12 16:20:30 +02003600 /* don't fail immediately if it's r/o, at least try to report other errors */
Al Viro391172c2015-05-09 11:19:16 -04003601 err2 = mnt_want_write(path->mnt);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003602 /*
3603 * Do the final lookup.
3604 */
Al Viro391172c2015-05-09 11:19:16 -04003605 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
Al Viro59551022016-01-22 15:40:57 -05003606 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
Al Viro391172c2015-05-09 11:19:16 -04003607 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003608 if (IS_ERR(dentry))
Al Viroa8104a92012-07-20 02:25:00 +04003609 goto unlock;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003610
Al Viroa8104a92012-07-20 02:25:00 +04003611 error = -EEXIST;
David Howellsb18825a2013-09-12 19:22:53 +01003612 if (d_is_positive(dentry))
Al Viroa8104a92012-07-20 02:25:00 +04003613 goto fail;
David Howellsb18825a2013-09-12 19:22:53 +01003614
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003615 /*
3616 * Special case - lookup gave negative, but... we had foo/bar/
3617 * From the vfs_mknod() POV we just have a negative dentry -
3618 * all is fine. Let's be bastards - you had / on the end, you've
3619 * been asking for (non-existent) directory. -ENOENT for you.
3620 */
Al Viro391172c2015-05-09 11:19:16 -04003621 if (unlikely(!is_dir && last.name[last.len])) {
Al Viroa8104a92012-07-20 02:25:00 +04003622 error = -ENOENT;
Al Viroed75e952011-06-27 16:53:43 -04003623 goto fail;
Al Viroe9baf6e2008-05-15 04:49:12 -04003624 }
Jan Karac30dabf2012-06-12 16:20:30 +02003625 if (unlikely(err2)) {
3626 error = err2;
Al Viroa8104a92012-07-20 02:25:00 +04003627 goto fail;
Jan Karac30dabf2012-06-12 16:20:30 +02003628 }
Al Viro181c37b2015-05-12 17:21:25 -04003629 putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003630 return dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003631fail:
Al Viroa8104a92012-07-20 02:25:00 +04003632 dput(dentry);
3633 dentry = ERR_PTR(error);
3634unlock:
Al Viro59551022016-01-22 15:40:57 -05003635 inode_unlock(path->dentry->d_inode);
Jan Karac30dabf2012-06-12 16:20:30 +02003636 if (!err2)
Al Viro391172c2015-05-09 11:19:16 -04003637 mnt_drop_write(path->mnt);
Al Viroed75e952011-06-27 16:53:43 -04003638out:
Al Viro391172c2015-05-09 11:19:16 -04003639 path_put(path);
Al Viro181c37b2015-05-12 17:21:25 -04003640 putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003641 return dentry;
3642}
Al Virofa14a0b2015-01-22 02:16:49 -05003643
3644struct dentry *kern_path_create(int dfd, const char *pathname,
3645 struct path *path, unsigned int lookup_flags)
3646{
Al Viro181c37b2015-05-12 17:21:25 -04003647 return filename_create(dfd, getname_kernel(pathname),
3648 path, lookup_flags);
Al Virofa14a0b2015-01-22 02:16:49 -05003649}
Al Virodae6ad82011-06-26 11:50:15 -04003650EXPORT_SYMBOL(kern_path_create);
3651
Al Viro921a1652012-07-20 01:15:31 +04003652void done_path_create(struct path *path, struct dentry *dentry)
3653{
3654 dput(dentry);
Al Viro59551022016-01-22 15:40:57 -05003655 inode_unlock(path->dentry->d_inode);
Al Viroa8104a92012-07-20 02:25:00 +04003656 mnt_drop_write(path->mnt);
Al Viro921a1652012-07-20 01:15:31 +04003657 path_put(path);
3658}
3659EXPORT_SYMBOL(done_path_create);
3660
Al Viro520ae682015-05-13 07:00:28 -04003661inline struct dentry *user_path_create(int dfd, const char __user *pathname,
Jeff Layton1ac12b42012-12-11 12:10:06 -05003662 struct path *path, unsigned int lookup_flags)
Al Virodae6ad82011-06-26 11:50:15 -04003663{
Al Viro181c37b2015-05-12 17:21:25 -04003664 return filename_create(dfd, getname(pathname), path, lookup_flags);
Al Virodae6ad82011-06-26 11:50:15 -04003665}
3666EXPORT_SYMBOL(user_path_create);
3667
Al Viro1a67aaf2011-07-26 01:52:52 -04003668int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003669{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003670 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003671
3672 if (error)
3673 return error;
3674
Eric W. Biederman975d6b32011-11-13 12:16:43 -08003675 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003676 return -EPERM;
3677
Al Viroacfa4382008-12-04 10:06:33 -05003678 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003679 return -EPERM;
3680
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07003681 error = devcgroup_inode_mknod(mode, dev);
3682 if (error)
3683 return error;
3684
Linus Torvalds1da177e2005-04-16 15:20:36 -07003685 error = security_inode_mknod(dir, dentry, mode, dev);
3686 if (error)
3687 return error;
3688
Linus Torvalds1da177e2005-04-16 15:20:36 -07003689 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003690 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003691 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003692 return error;
3693}
Al Viro4d359502014-03-14 12:20:17 -04003694EXPORT_SYMBOL(vfs_mknod);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003695
Al Virof69aac02011-07-26 04:31:40 -04003696static int may_mknod(umode_t mode)
Dave Hansen463c3192008-02-15 14:37:57 -08003697{
3698 switch (mode & S_IFMT) {
3699 case S_IFREG:
3700 case S_IFCHR:
3701 case S_IFBLK:
3702 case S_IFIFO:
3703 case S_IFSOCK:
3704 case 0: /* zero mode translates to S_IFREG */
3705 return 0;
3706 case S_IFDIR:
3707 return -EPERM;
3708 default:
3709 return -EINVAL;
3710 }
3711}
3712
Al Viro8208a222011-07-25 17:32:17 -04003713SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003714 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003715{
Al Viro2ad94ae2008-07-21 09:32:51 -04003716 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003717 struct path path;
3718 int error;
Jeff Layton972567f2012-12-20 16:00:10 -05003719 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003720
Al Viro8e4bfca2012-07-20 01:17:26 +04003721 error = may_mknod(mode);
3722 if (error)
3723 return error;
Jeff Layton972567f2012-12-20 16:00:10 -05003724retry:
3725 dentry = user_path_create(dfd, filename, &path, lookup_flags);
Al Virodae6ad82011-06-26 11:50:15 -04003726 if (IS_ERR(dentry))
3727 return PTR_ERR(dentry);
Al Viro2ad94ae2008-07-21 09:32:51 -04003728
Al Virodae6ad82011-06-26 11:50:15 -04003729 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003730 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003731 error = security_path_mknod(&path, dentry, mode, dev);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003732 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04003733 goto out;
Dave Hansen463c3192008-02-15 14:37:57 -08003734 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003735 case 0: case S_IFREG:
Al Viro312b63f2012-06-10 18:09:36 -04003736 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
Mimi Zohar05d1a712016-02-29 19:52:05 -05003737 if (!error)
3738 ima_post_path_mknod(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003739 break;
3740 case S_IFCHR: case S_IFBLK:
Al Virodae6ad82011-06-26 11:50:15 -04003741 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003742 new_decode_dev(dev));
3743 break;
3744 case S_IFIFO: case S_IFSOCK:
Al Virodae6ad82011-06-26 11:50:15 -04003745 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003746 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003747 }
Al Viroa8104a92012-07-20 02:25:00 +04003748out:
Al Viro921a1652012-07-20 01:15:31 +04003749 done_path_create(&path, dentry);
Jeff Layton972567f2012-12-20 16:00:10 -05003750 if (retry_estale(error, lookup_flags)) {
3751 lookup_flags |= LOOKUP_REVAL;
3752 goto retry;
3753 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003754 return error;
3755}
3756
Al Viro8208a222011-07-25 17:32:17 -04003757SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003758{
3759 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3760}
3761
Al Viro18bb1db2011-07-26 01:41:39 -04003762int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003763{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003764 int error = may_create(dir, dentry);
Al Viro8de52772012-02-06 12:45:27 -05003765 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003766
3767 if (error)
3768 return error;
3769
Al Viroacfa4382008-12-04 10:06:33 -05003770 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003771 return -EPERM;
3772
3773 mode &= (S_IRWXUGO|S_ISVTX);
3774 error = security_inode_mkdir(dir, dentry, mode);
3775 if (error)
3776 return error;
3777
Al Viro8de52772012-02-06 12:45:27 -05003778 if (max_links && dir->i_nlink >= max_links)
3779 return -EMLINK;
3780
Linus Torvalds1da177e2005-04-16 15:20:36 -07003781 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003782 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003783 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003784 return error;
3785}
Al Viro4d359502014-03-14 12:20:17 -04003786EXPORT_SYMBOL(vfs_mkdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003787
Al Viroa218d0f2011-11-21 14:59:34 -05003788SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003789{
Dave Hansen6902d922006-09-30 23:29:01 -07003790 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003791 struct path path;
3792 int error;
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003793 unsigned int lookup_flags = LOOKUP_DIRECTORY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003794
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003795retry:
3796 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07003797 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003798 return PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003799
Al Virodae6ad82011-06-26 11:50:15 -04003800 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003801 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003802 error = security_path_mkdir(&path, dentry, mode);
Al Viroa8104a92012-07-20 02:25:00 +04003803 if (!error)
3804 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
Al Viro921a1652012-07-20 01:15:31 +04003805 done_path_create(&path, dentry);
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003806 if (retry_estale(error, lookup_flags)) {
3807 lookup_flags |= LOOKUP_REVAL;
3808 goto retry;
3809 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003810 return error;
3811}
3812
Al Viroa218d0f2011-11-21 14:59:34 -05003813SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003814{
3815 return sys_mkdirat(AT_FDCWD, pathname, mode);
3816}
3817
Linus Torvalds1da177e2005-04-16 15:20:36 -07003818int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3819{
3820 int error = may_delete(dir, dentry, 1);
3821
3822 if (error)
3823 return error;
3824
Al Viroacfa4382008-12-04 10:06:33 -05003825 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003826 return -EPERM;
3827
Al Viro1d2ef592011-09-14 18:55:41 +01003828 dget(dentry);
Al Viro59551022016-01-22 15:40:57 -05003829 inode_lock(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003830
Sage Weil912dbc12011-05-24 13:06:11 -07003831 error = -EBUSY;
Eric W. Biederman7af13642013-10-04 19:15:13 -07003832 if (is_local_mountpoint(dentry))
Sage Weil912dbc12011-05-24 13:06:11 -07003833 goto out;
3834
3835 error = security_inode_rmdir(dir, dentry);
3836 if (error)
3837 goto out;
3838
Sage Weil3cebde22011-05-29 21:20:59 -07003839 shrink_dcache_parent(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003840 error = dir->i_op->rmdir(dir, dentry);
3841 if (error)
3842 goto out;
3843
3844 dentry->d_inode->i_flags |= S_DEAD;
3845 dont_mount(dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003846 detach_mounts(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003847
3848out:
Al Viro59551022016-01-22 15:40:57 -05003849 inode_unlock(dentry->d_inode);
Al Viro1d2ef592011-09-14 18:55:41 +01003850 dput(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003851 if (!error)
3852 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003853 return error;
3854}
Al Viro4d359502014-03-14 12:20:17 -04003855EXPORT_SYMBOL(vfs_rmdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003856
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003857static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003858{
3859 int error = 0;
Jeff Layton91a27b22012-10-10 15:25:28 -04003860 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003861 struct dentry *dentry;
Al Virof5beed72015-04-30 16:09:11 -04003862 struct path path;
3863 struct qstr last;
3864 int type;
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003865 unsigned int lookup_flags = 0;
3866retry:
Al Virof5beed72015-04-30 16:09:11 -04003867 name = user_path_parent(dfd, pathname,
3868 &path, &last, &type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04003869 if (IS_ERR(name))
3870 return PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003871
Al Virof5beed72015-04-30 16:09:11 -04003872 switch (type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003873 case LAST_DOTDOT:
3874 error = -ENOTEMPTY;
3875 goto exit1;
3876 case LAST_DOT:
3877 error = -EINVAL;
3878 goto exit1;
3879 case LAST_ROOT:
3880 error = -EBUSY;
3881 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003882 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003883
Al Virof5beed72015-04-30 16:09:11 -04003884 error = mnt_want_write(path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02003885 if (error)
3886 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003887
Al Viro59551022016-01-22 15:40:57 -05003888 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
Al Virof5beed72015-04-30 16:09:11 -04003889 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003890 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003891 if (IS_ERR(dentry))
3892 goto exit2;
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003893 if (!dentry->d_inode) {
3894 error = -ENOENT;
3895 goto exit3;
3896 }
Al Virof5beed72015-04-30 16:09:11 -04003897 error = security_path_rmdir(&path, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003898 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003899 goto exit3;
Al Virof5beed72015-04-30 16:09:11 -04003900 error = vfs_rmdir(path.dentry->d_inode, dentry);
Dave Hansen06227532008-02-15 14:37:34 -08003901exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07003902 dput(dentry);
3903exit2:
Al Viro59551022016-01-22 15:40:57 -05003904 inode_unlock(path.dentry->d_inode);
Al Virof5beed72015-04-30 16:09:11 -04003905 mnt_drop_write(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003906exit1:
Al Virof5beed72015-04-30 16:09:11 -04003907 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003908 putname(name);
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003909 if (retry_estale(error, lookup_flags)) {
3910 lookup_flags |= LOOKUP_REVAL;
3911 goto retry;
3912 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003913 return error;
3914}
3915
Heiko Carstens3cdad422009-01-14 14:14:22 +01003916SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003917{
3918 return do_rmdir(AT_FDCWD, pathname);
3919}
3920
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003921/**
3922 * vfs_unlink - unlink a filesystem object
3923 * @dir: parent directory
3924 * @dentry: victim
3925 * @delegated_inode: returns victim inode, if the inode is delegated.
3926 *
3927 * The caller must hold dir->i_mutex.
3928 *
3929 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3930 * return a reference to the inode in delegated_inode. The caller
3931 * should then break the delegation on that inode and retry. Because
3932 * breaking a delegation may take a long time, the caller should drop
3933 * dir->i_mutex before doing so.
3934 *
3935 * Alternatively, a caller may pass NULL for delegated_inode. This may
3936 * be appropriate for callers that expect the underlying filesystem not
3937 * to be NFS exported.
3938 */
3939int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003940{
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003941 struct inode *target = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003942 int error = may_delete(dir, dentry, 0);
3943
3944 if (error)
3945 return error;
3946
Al Viroacfa4382008-12-04 10:06:33 -05003947 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003948 return -EPERM;
3949
Al Viro59551022016-01-22 15:40:57 -05003950 inode_lock(target);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003951 if (is_local_mountpoint(dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003952 error = -EBUSY;
3953 else {
3954 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05003955 if (!error) {
J. Bruce Fields5a146962012-08-28 07:50:40 -07003956 error = try_break_deleg(target, delegated_inode);
3957 if (error)
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003958 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003959 error = dir->i_op->unlink(dir, dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003960 if (!error) {
Al Virod83c49f2010-04-30 17:17:09 -04003961 dont_mount(dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003962 detach_mounts(dentry);
3963 }
Al Virobec10522010-03-03 14:12:08 -05003964 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003965 }
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003966out:
Al Viro59551022016-01-22 15:40:57 -05003967 inode_unlock(target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003968
3969 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3970 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003971 fsnotify_link_count(target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003972 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003973 }
Robert Love0eeca282005-07-12 17:06:03 -04003974
Linus Torvalds1da177e2005-04-16 15:20:36 -07003975 return error;
3976}
Al Viro4d359502014-03-14 12:20:17 -04003977EXPORT_SYMBOL(vfs_unlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003978
3979/*
3980 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003981 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07003982 * writeout happening, and we don't want to prevent access to the directory
3983 * while waiting on the I/O.
3984 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003985static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003986{
Al Viro2ad94ae2008-07-21 09:32:51 -04003987 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04003988 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003989 struct dentry *dentry;
Al Virof5beed72015-04-30 16:09:11 -04003990 struct path path;
3991 struct qstr last;
3992 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003993 struct inode *inode = NULL;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003994 struct inode *delegated_inode = NULL;
Jeff Layton5d18f812012-12-20 16:38:04 -05003995 unsigned int lookup_flags = 0;
3996retry:
Al Virof5beed72015-04-30 16:09:11 -04003997 name = user_path_parent(dfd, pathname,
3998 &path, &last, &type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04003999 if (IS_ERR(name))
4000 return PTR_ERR(name);
Al Viro2ad94ae2008-07-21 09:32:51 -04004001
Linus Torvalds1da177e2005-04-16 15:20:36 -07004002 error = -EISDIR;
Al Virof5beed72015-04-30 16:09:11 -04004003 if (type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004004 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09004005
Al Virof5beed72015-04-30 16:09:11 -04004006 error = mnt_want_write(path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02004007 if (error)
4008 goto exit1;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04004009retry_deleg:
Al Viro59551022016-01-22 15:40:57 -05004010 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
Al Virof5beed72015-04-30 16:09:11 -04004011 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004012 error = PTR_ERR(dentry);
4013 if (!IS_ERR(dentry)) {
4014 /* Why not before? Because we want correct error value */
Al Virof5beed72015-04-30 16:09:11 -04004015 if (last.name[last.len])
Török Edwin50338b82011-06-16 00:06:14 +03004016 goto slashes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004017 inode = dentry->d_inode;
David Howellsb18825a2013-09-12 19:22:53 +01004018 if (d_is_negative(dentry))
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04004019 goto slashes;
4020 ihold(inode);
Al Virof5beed72015-04-30 16:09:11 -04004021 error = security_path_unlink(&path, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004022 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02004023 goto exit2;
Al Virof5beed72015-04-30 16:09:11 -04004024 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
Jan Karac30dabf2012-06-12 16:20:30 +02004025exit2:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004026 dput(dentry);
4027 }
Al Viro59551022016-01-22 15:40:57 -05004028 inode_unlock(path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004029 if (inode)
4030 iput(inode); /* truncate the inode here */
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04004031 inode = NULL;
4032 if (delegated_inode) {
J. Bruce Fields5a146962012-08-28 07:50:40 -07004033 error = break_deleg_wait(&delegated_inode);
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04004034 if (!error)
4035 goto retry_deleg;
4036 }
Al Virof5beed72015-04-30 16:09:11 -04004037 mnt_drop_write(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004038exit1:
Al Virof5beed72015-04-30 16:09:11 -04004039 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004040 putname(name);
Jeff Layton5d18f812012-12-20 16:38:04 -05004041 if (retry_estale(error, lookup_flags)) {
4042 lookup_flags |= LOOKUP_REVAL;
4043 inode = NULL;
4044 goto retry;
4045 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004046 return error;
4047
4048slashes:
David Howellsb18825a2013-09-12 19:22:53 +01004049 if (d_is_negative(dentry))
4050 error = -ENOENT;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02004051 else if (d_is_dir(dentry))
David Howellsb18825a2013-09-12 19:22:53 +01004052 error = -EISDIR;
4053 else
4054 error = -ENOTDIR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004055 goto exit2;
4056}
4057
Heiko Carstens2e4d0922009-01-14 14:14:31 +01004058SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004059{
4060 if ((flag & ~AT_REMOVEDIR) != 0)
4061 return -EINVAL;
4062
4063 if (flag & AT_REMOVEDIR)
4064 return do_rmdir(dfd, pathname);
4065
4066 return do_unlinkat(dfd, pathname);
4067}
4068
Heiko Carstens3480b252009-01-14 14:14:16 +01004069SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004070{
4071 return do_unlinkat(AT_FDCWD, pathname);
4072}
4073
Miklos Szeredidb2e7472008-06-24 16:50:16 +02004074int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004075{
Miklos Szeredia95164d2008-07-30 15:08:48 +02004076 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004077
4078 if (error)
4079 return error;
4080
Al Viroacfa4382008-12-04 10:06:33 -05004081 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004082 return -EPERM;
4083
4084 error = security_inode_symlink(dir, dentry, oldname);
4085 if (error)
4086 return error;
4087
Linus Torvalds1da177e2005-04-16 15:20:36 -07004088 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07004089 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00004090 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004091 return error;
4092}
Al Viro4d359502014-03-14 12:20:17 -04004093EXPORT_SYMBOL(vfs_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004094
Heiko Carstens2e4d0922009-01-14 14:14:31 +01004095SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4096 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004097{
Al Viro2ad94ae2008-07-21 09:32:51 -04004098 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04004099 struct filename *from;
Dave Hansen6902d922006-09-30 23:29:01 -07004100 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04004101 struct path path;
Jeff Laytonf46d3562012-12-11 12:10:08 -05004102 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004103
4104 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04004105 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004106 return PTR_ERR(from);
Jeff Laytonf46d3562012-12-11 12:10:08 -05004107retry:
4108 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07004109 error = PTR_ERR(dentry);
4110 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04004111 goto out_putname;
Dave Hansen6902d922006-09-30 23:29:01 -07004112
Jeff Layton91a27b22012-10-10 15:25:28 -04004113 error = security_path_symlink(&path, dentry, from->name);
Al Viroa8104a92012-07-20 02:25:00 +04004114 if (!error)
Jeff Layton91a27b22012-10-10 15:25:28 -04004115 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
Al Viro921a1652012-07-20 01:15:31 +04004116 done_path_create(&path, dentry);
Jeff Laytonf46d3562012-12-11 12:10:08 -05004117 if (retry_estale(error, lookup_flags)) {
4118 lookup_flags |= LOOKUP_REVAL;
4119 goto retry;
4120 }
Dave Hansen6902d922006-09-30 23:29:01 -07004121out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004122 putname(from);
4123 return error;
4124}
4125
Heiko Carstens3480b252009-01-14 14:14:16 +01004126SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004127{
4128 return sys_symlinkat(oldname, AT_FDCWD, newname);
4129}
4130
J. Bruce Fields146a8592011-09-20 17:14:31 -04004131/**
4132 * vfs_link - create a new link
4133 * @old_dentry: object to be linked
4134 * @dir: new parent
4135 * @new_dentry: where to create the new link
4136 * @delegated_inode: returns inode needing a delegation break
4137 *
4138 * The caller must hold dir->i_mutex
4139 *
4140 * If vfs_link discovers a delegation on the to-be-linked file in need
4141 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4142 * inode in delegated_inode. The caller should then break the delegation
4143 * and retry. Because breaking a delegation may take a long time, the
4144 * caller should drop the i_mutex before doing so.
4145 *
4146 * Alternatively, a caller may pass NULL for delegated_inode. This may
4147 * be appropriate for callers that expect the underlying filesystem not
4148 * to be NFS exported.
4149 */
4150int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004151{
4152 struct inode *inode = old_dentry->d_inode;
Al Viro8de52772012-02-06 12:45:27 -05004153 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004154 int error;
4155
4156 if (!inode)
4157 return -ENOENT;
4158
Miklos Szeredia95164d2008-07-30 15:08:48 +02004159 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004160 if (error)
4161 return error;
4162
4163 if (dir->i_sb != inode->i_sb)
4164 return -EXDEV;
4165
4166 /*
4167 * A link to an append-only or immutable file cannot be created.
4168 */
4169 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4170 return -EPERM;
Eric W. Biederman0bd23d092016-06-29 14:54:46 -05004171 /*
4172 * Updating the link count will likely cause i_uid and i_gid to
4173 * be writen back improperly if their true value is unknown to
4174 * the vfs.
4175 */
4176 if (HAS_UNMAPPED_ID(inode))
4177 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05004178 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004179 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02004180 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004181 return -EPERM;
4182
4183 error = security_inode_link(old_dentry, dir, new_dentry);
4184 if (error)
4185 return error;
4186
Al Viro59551022016-01-22 15:40:57 -05004187 inode_lock(inode);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05304188 /* Make sure we don't allow creating hardlink to an unlinked file */
Al Virof4e0c302013-06-11 08:34:36 +04004189 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05304190 error = -ENOENT;
Al Viro8de52772012-02-06 12:45:27 -05004191 else if (max_links && inode->i_nlink >= max_links)
4192 error = -EMLINK;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004193 else {
4194 error = try_break_deleg(inode, delegated_inode);
4195 if (!error)
4196 error = dir->i_op->link(old_dentry, dir, new_dentry);
4197 }
Al Virof4e0c302013-06-11 08:34:36 +04004198
4199 if (!error && (inode->i_state & I_LINKABLE)) {
4200 spin_lock(&inode->i_lock);
4201 inode->i_state &= ~I_LINKABLE;
4202 spin_unlock(&inode->i_lock);
4203 }
Al Viro59551022016-01-22 15:40:57 -05004204 inode_unlock(inode);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07004205 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02004206 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004207 return error;
4208}
Al Viro4d359502014-03-14 12:20:17 -04004209EXPORT_SYMBOL(vfs_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004210
4211/*
4212 * Hardlinks are often used in delicate situations. We avoid
4213 * security-related surprises by not following symlinks on the
4214 * newname. --KAB
4215 *
4216 * We don't follow them on the oldname either to be compatible
4217 * with linux 2.0, and to avoid hard-linking to directories
4218 * and other special files. --ADM
4219 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01004220SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4221 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004222{
4223 struct dentry *new_dentry;
Al Virodae6ad82011-06-26 11:50:15 -04004224 struct path old_path, new_path;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004225 struct inode *delegated_inode = NULL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304226 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004227 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004228
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304229 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004230 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304231 /*
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004232 * To use null names we require CAP_DAC_READ_SEARCH
4233 * This ensures that not everyone will be able to create
4234 * handlink using the passed filedescriptor.
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304235 */
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004236 if (flags & AT_EMPTY_PATH) {
4237 if (!capable(CAP_DAC_READ_SEARCH))
4238 return -ENOENT;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304239 how = LOOKUP_EMPTY;
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004240 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004241
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304242 if (flags & AT_SYMLINK_FOLLOW)
4243 how |= LOOKUP_FOLLOW;
Jeff Layton442e31c2012-12-20 16:15:38 -05004244retry:
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304245 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004246 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04004247 return error;
4248
Jeff Layton442e31c2012-12-20 16:15:38 -05004249 new_dentry = user_path_create(newdfd, newname, &new_path,
4250 (how & LOOKUP_REVAL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004251 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07004252 if (IS_ERR(new_dentry))
Al Virodae6ad82011-06-26 11:50:15 -04004253 goto out;
4254
4255 error = -EXDEV;
4256 if (old_path.mnt != new_path.mnt)
4257 goto out_dput;
Kees Cook800179c2012-07-25 17:29:07 -07004258 error = may_linkat(&old_path);
4259 if (unlikely(error))
4260 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04004261 error = security_path_link(old_path.dentry, &new_path, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004262 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04004263 goto out_dput;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004264 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
Dave Hansen75c3f292008-02-15 14:37:45 -08004265out_dput:
Al Viro921a1652012-07-20 01:15:31 +04004266 done_path_create(&new_path, new_dentry);
J. Bruce Fields146a8592011-09-20 17:14:31 -04004267 if (delegated_inode) {
4268 error = break_deleg_wait(&delegated_inode);
Oleg Drokind22e6332014-01-31 15:41:58 -05004269 if (!error) {
4270 path_put(&old_path);
J. Bruce Fields146a8592011-09-20 17:14:31 -04004271 goto retry;
Oleg Drokind22e6332014-01-31 15:41:58 -05004272 }
J. Bruce Fields146a8592011-09-20 17:14:31 -04004273 }
Jeff Layton442e31c2012-12-20 16:15:38 -05004274 if (retry_estale(error, how)) {
Oleg Drokind22e6332014-01-31 15:41:58 -05004275 path_put(&old_path);
Jeff Layton442e31c2012-12-20 16:15:38 -05004276 how |= LOOKUP_REVAL;
4277 goto retry;
4278 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004279out:
Al Viro2d8f3032008-07-22 09:59:21 -04004280 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004281
4282 return error;
4283}
4284
Heiko Carstens3480b252009-01-14 14:14:16 +01004285SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004286{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004287 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004288}
4289
Miklos Szeredibc270272014-04-01 17:08:42 +02004290/**
4291 * vfs_rename - rename a filesystem object
4292 * @old_dir: parent of source
4293 * @old_dentry: source
4294 * @new_dir: parent of destination
4295 * @new_dentry: destination
4296 * @delegated_inode: returns an inode needing a delegation break
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004297 * @flags: rename flags
Miklos Szeredibc270272014-04-01 17:08:42 +02004298 *
4299 * The caller must hold multiple mutexes--see lock_rename()).
4300 *
4301 * If vfs_rename discovers a delegation in need of breaking at either
4302 * the source or destination, it will return -EWOULDBLOCK and return a
4303 * reference to the inode in delegated_inode. The caller should then
4304 * break the delegation and retry. Because breaking a delegation may
4305 * take a long time, the caller should drop all locks before doing
4306 * so.
4307 *
4308 * Alternatively, a caller may pass NULL for delegated_inode. This may
4309 * be appropriate for callers that expect the underlying filesystem not
4310 * to be NFS exported.
4311 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07004312 * The worst of all namespace operations - renaming directory. "Perverted"
4313 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4314 * Problems:
J. Bruce Fieldsd03b29a2014-02-17 16:52:33 -05004315 * a) we can get into loop creation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004316 * b) race potential - two innocent renames can create a loop together.
4317 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004318 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07004319 * story.
J. Bruce Fields6cedba82012-03-05 11:40:41 -05004320 * c) we have to lock _four_ objects - parents and victim (if it exists),
4321 * and source (if it is not a directory).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004322 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07004323 * whether the target exists). Solution: try to be smart with locking
4324 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004325 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07004326 * move will be locked. Thus we can rank directories by the tree
4327 * (ancestors first) and rank all non-directories after them.
4328 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004329 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004330 * HOWEVER, it relies on the assumption that any object with ->lookup()
4331 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4332 * we'd better make sure that there's no link(2) for them.
Sage Weile4eaac02011-05-24 13:06:07 -07004333 * d) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004334 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07004335 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05004336 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07004337 * locking].
4338 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004339int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004340 struct inode *new_dir, struct dentry *new_dentry,
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004341 struct inode **delegated_inode, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004342{
4343 int error;
Miklos Szeredibc270272014-04-01 17:08:42 +02004344 bool is_dir = d_is_dir(old_dentry);
Miklos Szeredibc270272014-04-01 17:08:42 +02004345 struct inode *source = old_dentry->d_inode;
4346 struct inode *target = new_dentry->d_inode;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004347 bool new_is_dir = false;
4348 unsigned max_links = new_dir->i_sb->s_max_links;
Al Viroad25f112017-07-07 14:51:19 -04004349 struct name_snapshot old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004350
Miklos Szeredi9409e222016-05-11 01:16:37 +02004351 /*
4352 * Check source == target.
4353 * On overlayfs need to look at underlying inodes.
4354 */
Miklos Szeredi2d902672016-06-30 08:53:27 +02004355 if (d_real_inode(old_dentry) == d_real_inode(new_dentry))
Miklos Szeredibc270272014-04-01 17:08:42 +02004356 return 0;
4357
Linus Torvalds1da177e2005-04-16 15:20:36 -07004358 error = may_delete(old_dir, old_dentry, is_dir);
4359 if (error)
4360 return error;
4361
Miklos Szeredida1ce062014-04-01 17:08:43 +02004362 if (!target) {
Miklos Szeredia95164d2008-07-30 15:08:48 +02004363 error = may_create(new_dir, new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004364 } else {
4365 new_is_dir = d_is_dir(new_dentry);
4366
4367 if (!(flags & RENAME_EXCHANGE))
4368 error = may_delete(new_dir, new_dentry, is_dir);
4369 else
4370 error = may_delete(new_dir, new_dentry, new_is_dir);
4371 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004372 if (error)
4373 return error;
4374
Miklos Szeredi2773bf02016-09-27 11:03:58 +02004375 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004376 return -EPERM;
4377
Miklos Szeredibc270272014-04-01 17:08:42 +02004378 /*
4379 * If we are going to change the parent - check write permissions,
4380 * we'll need to flip '..'.
4381 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02004382 if (new_dir != old_dir) {
4383 if (is_dir) {
4384 error = inode_permission(source, MAY_WRITE);
4385 if (error)
4386 return error;
4387 }
4388 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4389 error = inode_permission(target, MAY_WRITE);
4390 if (error)
4391 return error;
4392 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004393 }
Robert Love0eeca282005-07-12 17:06:03 -04004394
Miklos Szeredi0b3974e2014-04-01 17:08:43 +02004395 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4396 flags);
Miklos Szeredibc270272014-04-01 17:08:42 +02004397 if (error)
4398 return error;
4399
Al Viroad25f112017-07-07 14:51:19 -04004400 take_dentry_name_snapshot(&old_name, old_dentry);
Miklos Szeredibc270272014-04-01 17:08:42 +02004401 dget(new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004402 if (!is_dir || (flags & RENAME_EXCHANGE))
Miklos Szeredibc270272014-04-01 17:08:42 +02004403 lock_two_nondirectories(source, target);
4404 else if (target)
Al Viro59551022016-01-22 15:40:57 -05004405 inode_lock(target);
Miklos Szeredibc270272014-04-01 17:08:42 +02004406
4407 error = -EBUSY;
Eric W. Biederman7af13642013-10-04 19:15:13 -07004408 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
Miklos Szeredibc270272014-04-01 17:08:42 +02004409 goto out;
4410
Miklos Szeredida1ce062014-04-01 17:08:43 +02004411 if (max_links && new_dir != old_dir) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004412 error = -EMLINK;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004413 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
Miklos Szeredibc270272014-04-01 17:08:42 +02004414 goto out;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004415 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4416 old_dir->i_nlink >= max_links)
4417 goto out;
4418 }
4419 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4420 shrink_dcache_parent(new_dentry);
4421 if (!is_dir) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004422 error = try_break_deleg(source, delegated_inode);
4423 if (error)
4424 goto out;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004425 }
4426 if (target && !new_is_dir) {
4427 error = try_break_deleg(target, delegated_inode);
4428 if (error)
4429 goto out;
Miklos Szeredibc270272014-04-01 17:08:42 +02004430 }
Miklos Szeredi2773bf02016-09-27 11:03:58 +02004431 error = old_dir->i_op->rename(old_dir, old_dentry,
Miklos Szeredi18fc84d2016-09-27 11:03:58 +02004432 new_dir, new_dentry, flags);
Miklos Szeredibc270272014-04-01 17:08:42 +02004433 if (error)
4434 goto out;
4435
Miklos Szeredida1ce062014-04-01 17:08:43 +02004436 if (!(flags & RENAME_EXCHANGE) && target) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004437 if (is_dir)
4438 target->i_flags |= S_DEAD;
4439 dont_mount(new_dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07004440 detach_mounts(new_dentry);
Miklos Szeredibc270272014-04-01 17:08:42 +02004441 }
Miklos Szeredida1ce062014-04-01 17:08:43 +02004442 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4443 if (!(flags & RENAME_EXCHANGE))
4444 d_move(old_dentry, new_dentry);
4445 else
4446 d_exchange(old_dentry, new_dentry);
4447 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004448out:
Miklos Szeredida1ce062014-04-01 17:08:43 +02004449 if (!is_dir || (flags & RENAME_EXCHANGE))
Miklos Szeredibc270272014-04-01 17:08:42 +02004450 unlock_two_nondirectories(source, target);
4451 else if (target)
Al Viro59551022016-01-22 15:40:57 -05004452 inode_unlock(target);
Miklos Szeredibc270272014-04-01 17:08:42 +02004453 dput(new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004454 if (!error) {
Al Viroad25f112017-07-07 14:51:19 -04004455 fsnotify_move(old_dir, new_dir, old_name.name, is_dir,
Miklos Szeredida1ce062014-04-01 17:08:43 +02004456 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4457 if (flags & RENAME_EXCHANGE) {
4458 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4459 new_is_dir, NULL, new_dentry);
4460 }
4461 }
Al Viroad25f112017-07-07 14:51:19 -04004462 release_dentry_name_snapshot(&old_name);
Robert Love0eeca282005-07-12 17:06:03 -04004463
Linus Torvalds1da177e2005-04-16 15:20:36 -07004464 return error;
4465}
Al Viro4d359502014-03-14 12:20:17 -04004466EXPORT_SYMBOL(vfs_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004467
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004468SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4469 int, newdfd, const char __user *, newname, unsigned int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004470{
Al Viro2ad94ae2008-07-21 09:32:51 -04004471 struct dentry *old_dentry, *new_dentry;
4472 struct dentry *trap;
Al Virof5beed72015-04-30 16:09:11 -04004473 struct path old_path, new_path;
4474 struct qstr old_last, new_last;
4475 int old_type, new_type;
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004476 struct inode *delegated_inode = NULL;
Jeff Layton91a27b22012-10-10 15:25:28 -04004477 struct filename *from;
4478 struct filename *to;
Al Virof5beed72015-04-30 16:09:11 -04004479 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
Jeff Laytonc6a94282012-12-11 12:10:10 -05004480 bool should_retry = false;
Al Viro2ad94ae2008-07-21 09:32:51 -04004481 int error;
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004482
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004483 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredida1ce062014-04-01 17:08:43 +02004484 return -EINVAL;
4485
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004486 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4487 (flags & RENAME_EXCHANGE))
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004488 return -EINVAL;
4489
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004490 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4491 return -EPERM;
4492
Al Virof5beed72015-04-30 16:09:11 -04004493 if (flags & RENAME_EXCHANGE)
4494 target_flags = 0;
4495
Jeff Laytonc6a94282012-12-11 12:10:10 -05004496retry:
Al Virof5beed72015-04-30 16:09:11 -04004497 from = user_path_parent(olddfd, oldname,
4498 &old_path, &old_last, &old_type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04004499 if (IS_ERR(from)) {
4500 error = PTR_ERR(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004501 goto exit;
Jeff Layton91a27b22012-10-10 15:25:28 -04004502 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004503
Al Virof5beed72015-04-30 16:09:11 -04004504 to = user_path_parent(newdfd, newname,
4505 &new_path, &new_last, &new_type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04004506 if (IS_ERR(to)) {
4507 error = PTR_ERR(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004508 goto exit1;
Jeff Layton91a27b22012-10-10 15:25:28 -04004509 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004510
4511 error = -EXDEV;
Al Virof5beed72015-04-30 16:09:11 -04004512 if (old_path.mnt != new_path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004513 goto exit2;
4514
Linus Torvalds1da177e2005-04-16 15:20:36 -07004515 error = -EBUSY;
Al Virof5beed72015-04-30 16:09:11 -04004516 if (old_type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004517 goto exit2;
4518
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004519 if (flags & RENAME_NOREPLACE)
4520 error = -EEXIST;
Al Virof5beed72015-04-30 16:09:11 -04004521 if (new_type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004522 goto exit2;
4523
Al Virof5beed72015-04-30 16:09:11 -04004524 error = mnt_want_write(old_path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02004525 if (error)
4526 goto exit2;
4527
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004528retry_deleg:
Al Virof5beed72015-04-30 16:09:11 -04004529 trap = lock_rename(new_path.dentry, old_path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004530
Al Virof5beed72015-04-30 16:09:11 -04004531 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004532 error = PTR_ERR(old_dentry);
4533 if (IS_ERR(old_dentry))
4534 goto exit3;
4535 /* source must exist */
4536 error = -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01004537 if (d_is_negative(old_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004538 goto exit4;
Al Virof5beed72015-04-30 16:09:11 -04004539 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004540 error = PTR_ERR(new_dentry);
4541 if (IS_ERR(new_dentry))
4542 goto exit4;
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004543 error = -EEXIST;
4544 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4545 goto exit5;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004546 if (flags & RENAME_EXCHANGE) {
4547 error = -ENOENT;
4548 if (d_is_negative(new_dentry))
4549 goto exit5;
4550
4551 if (!d_is_dir(new_dentry)) {
4552 error = -ENOTDIR;
Al Virof5beed72015-04-30 16:09:11 -04004553 if (new_last.name[new_last.len])
Miklos Szeredida1ce062014-04-01 17:08:43 +02004554 goto exit5;
4555 }
4556 }
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004557 /* unless the source is a directory trailing slashes give -ENOTDIR */
4558 if (!d_is_dir(old_dentry)) {
4559 error = -ENOTDIR;
Al Virof5beed72015-04-30 16:09:11 -04004560 if (old_last.name[old_last.len])
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004561 goto exit5;
Al Virof5beed72015-04-30 16:09:11 -04004562 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004563 goto exit5;
4564 }
4565 /* source should not be ancestor of target */
4566 error = -EINVAL;
4567 if (old_dentry == trap)
4568 goto exit5;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004569 /* target should not be an ancestor of source */
Miklos Szeredida1ce062014-04-01 17:08:43 +02004570 if (!(flags & RENAME_EXCHANGE))
4571 error = -ENOTEMPTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004572 if (new_dentry == trap)
4573 goto exit5;
4574
Al Virof5beed72015-04-30 16:09:11 -04004575 error = security_path_rename(&old_path, old_dentry,
4576 &new_path, new_dentry, flags);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004577 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02004578 goto exit5;
Al Virof5beed72015-04-30 16:09:11 -04004579 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4580 new_path.dentry->d_inode, new_dentry,
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004581 &delegated_inode, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004582exit5:
4583 dput(new_dentry);
4584exit4:
4585 dput(old_dentry);
4586exit3:
Al Virof5beed72015-04-30 16:09:11 -04004587 unlock_rename(new_path.dentry, old_path.dentry);
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004588 if (delegated_inode) {
4589 error = break_deleg_wait(&delegated_inode);
4590 if (!error)
4591 goto retry_deleg;
4592 }
Al Virof5beed72015-04-30 16:09:11 -04004593 mnt_drop_write(old_path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004594exit2:
Jeff Laytonc6a94282012-12-11 12:10:10 -05004595 if (retry_estale(error, lookup_flags))
4596 should_retry = true;
Al Virof5beed72015-04-30 16:09:11 -04004597 path_put(&new_path);
Al Viro2ad94ae2008-07-21 09:32:51 -04004598 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004599exit1:
Al Virof5beed72015-04-30 16:09:11 -04004600 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004601 putname(from);
Jeff Laytonc6a94282012-12-11 12:10:10 -05004602 if (should_retry) {
4603 should_retry = false;
4604 lookup_flags |= LOOKUP_REVAL;
4605 goto retry;
4606 }
Al Viro2ad94ae2008-07-21 09:32:51 -04004607exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004608 return error;
4609}
4610
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004611SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4612 int, newdfd, const char __user *, newname)
4613{
4614 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4615}
4616
Heiko Carstensa26eab22009-01-14 14:14:17 +01004617SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004618{
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004619 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004620}
4621
Miklos Szeredi787fb6b2014-10-24 00:14:36 +02004622int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4623{
4624 int error = may_create(dir, dentry);
4625 if (error)
4626 return error;
4627
4628 if (!dir->i_op->mknod)
4629 return -EPERM;
4630
4631 return dir->i_op->mknod(dir, dentry,
4632 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4633}
4634EXPORT_SYMBOL(vfs_whiteout);
4635
Al Viro5d826c82014-03-14 13:42:45 -04004636int readlink_copy(char __user *buffer, int buflen, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004637{
Al Viro5d826c82014-03-14 13:42:45 -04004638 int len = PTR_ERR(link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004639 if (IS_ERR(link))
4640 goto out;
4641
4642 len = strlen(link);
4643 if (len > (unsigned) buflen)
4644 len = buflen;
4645 if (copy_to_user(buffer, link, len))
4646 len = -EFAULT;
4647out:
4648 return len;
4649}
4650
4651/*
4652 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
Al Viro6b255392015-11-17 10:20:54 -05004653 * have ->get_link() not calling nd_jump_link(). Using (or not using) it
4654 * for any given inode is up to filesystem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004655 */
4656int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4657{
Al Virofceef392015-12-29 15:58:39 -05004658 DEFINE_DELAYED_CALL(done);
Al Viro5f2c4172015-05-07 11:14:26 -04004659 struct inode *inode = d_inode(dentry);
4660 const char *link = inode->i_link;
Marcin Slusarz694a1762008-06-09 16:40:37 -07004661 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004662
Al Virod4dee482015-04-30 20:08:02 -04004663 if (!link) {
Al Virofceef392015-12-29 15:58:39 -05004664 link = inode->i_op->get_link(dentry, inode, &done);
Al Virod4dee482015-04-30 20:08:02 -04004665 if (IS_ERR(link))
4666 return PTR_ERR(link);
4667 }
Al Viro680baac2015-05-02 13:32:22 -04004668 res = readlink_copy(buffer, buflen, link);
Al Virofceef392015-12-29 15:58:39 -05004669 do_delayed_call(&done);
Marcin Slusarz694a1762008-06-09 16:40:37 -07004670 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004671}
Al Viro4d359502014-03-14 12:20:17 -04004672EXPORT_SYMBOL(generic_readlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004673
Miklos Szeredid60874c2016-10-04 14:40:45 +02004674/**
4675 * vfs_get_link - get symlink body
4676 * @dentry: dentry on which to get symbolic link
4677 * @done: caller needs to free returned data with this
4678 *
4679 * Calls security hook and i_op->get_link() on the supplied inode.
4680 *
4681 * It does not touch atime. That's up to the caller if necessary.
4682 *
4683 * Does not work on "special" symlinks like /proc/$$/fd/N
4684 */
4685const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4686{
4687 const char *res = ERR_PTR(-EINVAL);
4688 struct inode *inode = d_inode(dentry);
4689
4690 if (d_is_symlink(dentry)) {
4691 res = ERR_PTR(security_inode_readlink(dentry));
4692 if (!res)
4693 res = inode->i_op->get_link(dentry, inode, done);
4694 }
4695 return res;
4696}
4697EXPORT_SYMBOL(vfs_get_link);
4698
Linus Torvalds1da177e2005-04-16 15:20:36 -07004699/* get the link contents into pagecache */
Al Viro6b255392015-11-17 10:20:54 -05004700const char *page_get_link(struct dentry *dentry, struct inode *inode,
Al Virofceef392015-12-29 15:58:39 -05004701 struct delayed_call *callback)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004702{
Duane Griffinebd09ab2008-12-19 20:47:12 +00004703 char *kaddr;
4704 struct page *page;
Al Viro6b255392015-11-17 10:20:54 -05004705 struct address_space *mapping = inode->i_mapping;
4706
Al Virod3883d42015-11-17 10:41:04 -05004707 if (!dentry) {
4708 page = find_get_page(mapping, 0);
4709 if (!page)
4710 return ERR_PTR(-ECHILD);
4711 if (!PageUptodate(page)) {
4712 put_page(page);
4713 return ERR_PTR(-ECHILD);
4714 }
4715 } else {
4716 page = read_mapping_page(mapping, 0, NULL);
4717 if (IS_ERR(page))
4718 return (char*)page;
4719 }
Al Virofceef392015-12-29 15:58:39 -05004720 set_delayed_call(callback, page_put_link, page);
Al Viro21fc61c2015-11-17 01:07:57 -05004721 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4722 kaddr = page_address(page);
Al Viro6b255392015-11-17 10:20:54 -05004723 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
Duane Griffinebd09ab2008-12-19 20:47:12 +00004724 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004725}
4726
Al Viro6b255392015-11-17 10:20:54 -05004727EXPORT_SYMBOL(page_get_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004728
Al Virofceef392015-12-29 15:58:39 -05004729void page_put_link(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004730{
Al Virofceef392015-12-29 15:58:39 -05004731 put_page(arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004732}
Al Viro4d359502014-03-14 12:20:17 -04004733EXPORT_SYMBOL(page_put_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004734
Al Viroaa80dea2015-11-16 18:26:34 -05004735int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4736{
Al Virofceef392015-12-29 15:58:39 -05004737 DEFINE_DELAYED_CALL(done);
Al Viro6b255392015-11-17 10:20:54 -05004738 int res = readlink_copy(buffer, buflen,
4739 page_get_link(dentry, d_inode(dentry),
Al Virofceef392015-12-29 15:58:39 -05004740 &done));
4741 do_delayed_call(&done);
Al Viroaa80dea2015-11-16 18:26:34 -05004742 return res;
4743}
4744EXPORT_SYMBOL(page_readlink);
4745
Nick Piggin54566b22009-01-04 12:00:53 -08004746/*
4747 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4748 */
4749int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004750{
4751 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004752 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07004753 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08004754 int err;
Nick Piggin54566b22009-01-04 12:00:53 -08004755 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4756 if (nofs)
4757 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004758
NeilBrown7e53cac2006-03-25 03:07:57 -08004759retry:
Nick Pigginafddba42007-10-16 01:25:01 -07004760 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08004761 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004762 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07004763 goto fail;
4764
Al Viro21fc61c2015-11-17 01:07:57 -05004765 memcpy(page_address(page), symname, len-1);
Nick Pigginafddba42007-10-16 01:25:01 -07004766
4767 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4768 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004769 if (err < 0)
4770 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07004771 if (err < len-1)
4772 goto retry;
4773
Linus Torvalds1da177e2005-04-16 15:20:36 -07004774 mark_inode_dirty(inode);
4775 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004776fail:
4777 return err;
4778}
Al Viro4d359502014-03-14 12:20:17 -04004779EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004780
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004781int page_symlink(struct inode *inode, const char *symname, int len)
4782{
4783 return __page_symlink(inode, symname, len,
Michal Hockoc62d2552015-11-06 16:28:49 -08004784 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004785}
Al Viro4d359502014-03-14 12:20:17 -04004786EXPORT_SYMBOL(page_symlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004787
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08004788const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004789 .readlink = generic_readlink,
Al Viro6b255392015-11-17 10:20:54 -05004790 .get_link = page_get_link,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004791};
Linus Torvalds1da177e2005-04-16 15:20:36 -07004792EXPORT_SYMBOL(page_symlink_inode_operations);