blob: 1c2105ed20c5ef4fb390878fb5442943ceec29ae [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include <asm/uaccess.h>
39
Eric Parise81e3f42009-12-04 15:47:36 -050040#include "internal.h"
Al Viroc7105362011-11-24 18:22:03 -050041#include "mount.h"
Eric Parise81e3f42009-12-04 15:47:36 -050042
Linus Torvalds1da177e2005-04-16 15:20:36 -070043/* [Feb-1997 T. Schoebel-Theuer]
44 * Fundamental changes in the pathname lookup mechanisms (namei)
45 * were necessary because of omirr. The reason is that omirr needs
46 * to know the _real_ pathname, not the user-supplied one, in case
47 * of symlinks (and also when transname replacements occur).
48 *
49 * The new code replaces the old recursive symlink resolution with
50 * an iterative one (in case of non-nested symlink chains). It does
51 * this with calls to <fs>_follow_link().
52 * As a side effect, dir_namei(), _namei() and follow_link() are now
53 * replaced with a single function lookup_dentry() that can handle all
54 * the special cases of the former code.
55 *
56 * With the new dcache, the pathname is stored at each inode, at least as
57 * long as the refcount of the inode is positive. As a side effect, the
58 * size of the dcache depends on the inode cache and thus is dynamic.
59 *
60 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
61 * resolution to correspond with current state of the code.
62 *
63 * Note that the symlink resolution is not *completely* iterative.
64 * There is still a significant amount of tail- and mid- recursion in
65 * the algorithm. Also, note that <fs>_readlink() is not used in
66 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
67 * may return different results than <fs>_follow_link(). Many virtual
68 * filesystems (including /proc) exhibit this behavior.
69 */
70
71/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
72 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
73 * and the name already exists in form of a symlink, try to create the new
74 * name indicated by the symlink. The old code always complained that the
75 * name already exists, due to not following the symlink even if its target
76 * is nonexistent. The new semantics affects also mknod() and link() when
Lucas De Marchi25985ed2011-03-30 22:57:33 -030077 * the name is a symlink pointing to a non-existent name.
Linus Torvalds1da177e2005-04-16 15:20:36 -070078 *
79 * I don't know which semantics is the right one, since I have no access
80 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
81 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
82 * "old" one. Personally, I think the new semantics is much more logical.
83 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
84 * file does succeed in both HP-UX and SunOs, but not in Solaris
85 * and in the old Linux semantics.
86 */
87
88/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
89 * semantics. See the comments in "open_namei" and "do_link" below.
90 *
91 * [10-Sep-98 Alan Modra] Another symlink change.
92 */
93
94/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
95 * inside the path - always follow.
96 * in the last component in creation/removal/renaming - never follow.
97 * if LOOKUP_FOLLOW passed - follow.
98 * if the pathname has trailing slashes - follow.
99 * otherwise - don't follow.
100 * (applied in that order).
101 *
102 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
103 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
104 * During the 2.4 we need to fix the userland stuff depending on it -
105 * hopefully we will be able to get rid of that wart in 2.5. So far only
106 * XEmacs seems to be relying on it...
107 */
108/*
109 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800110 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 * any extra contention...
112 */
113
114/* In order to reduce some races, while at the same time doing additional
115 * checking and hopefully speeding things up, we copy filenames to the
116 * kernel data space before using them..
117 *
118 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
119 * PATH_MAX includes the nul terminator --RR.
120 */
Jeff Layton91a27b22012-10-10 15:25:28 -0400121
Al Virofd2f7cb2015-02-22 20:07:13 -0500122#define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
Jeff Layton7950e382012-10-10 16:43:13 -0400123
David Drysdale51f39a12014-12-12 16:57:29 -0800124struct filename *
Jeff Layton91a27b22012-10-10 15:25:28 -0400125getname_flags(const char __user *filename, int flags, int *empty)
126{
Al Viro94b5d262015-02-22 19:38:03 -0500127 struct filename *result;
Jeff Layton7950e382012-10-10 16:43:13 -0400128 char *kname;
Al Viro94b5d262015-02-22 19:38:03 -0500129 int len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130
Jeff Layton7ac86262012-10-10 15:25:28 -0400131 result = audit_reusename(filename);
132 if (result)
133 return result;
134
Jeff Layton7950e382012-10-10 16:43:13 -0400135 result = __getname();
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700136 if (unlikely(!result))
Eric Paris4043cde2012-01-03 14:23:08 -0500137 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138
Jeff Layton7950e382012-10-10 16:43:13 -0400139 /*
140 * First, try to embed the struct filename inside the names_cache
141 * allocation
142 */
Al Virofd2f7cb2015-02-22 20:07:13 -0500143 kname = (char *)result->iname;
Jeff Layton91a27b22012-10-10 15:25:28 -0400144 result->name = kname;
Jeff Layton7950e382012-10-10 16:43:13 -0400145
Al Viro94b5d262015-02-22 19:38:03 -0500146 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
Jeff Layton91a27b22012-10-10 15:25:28 -0400147 if (unlikely(len < 0)) {
Al Viro94b5d262015-02-22 19:38:03 -0500148 __putname(result);
149 return ERR_PTR(len);
Jeff Layton91a27b22012-10-10 15:25:28 -0400150 }
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700151
Jeff Layton7950e382012-10-10 16:43:13 -0400152 /*
153 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
154 * separate struct filename so we can dedicate the entire
155 * names_cache allocation for the pathname, and re-do the copy from
156 * userland.
157 */
Al Viro94b5d262015-02-22 19:38:03 -0500158 if (unlikely(len == EMBEDDED_NAME_MAX)) {
Al Virofd2f7cb2015-02-22 20:07:13 -0500159 const size_t size = offsetof(struct filename, iname[1]);
Jeff Layton7950e382012-10-10 16:43:13 -0400160 kname = (char *)result;
161
Al Virofd2f7cb2015-02-22 20:07:13 -0500162 /*
163 * size is chosen that way we to guarantee that
164 * result->iname[0] is within the same object and that
165 * kname can't be equal to result->iname, no matter what.
166 */
167 result = kzalloc(size, GFP_KERNEL);
Al Viro94b5d262015-02-22 19:38:03 -0500168 if (unlikely(!result)) {
169 __putname(kname);
170 return ERR_PTR(-ENOMEM);
Jeff Layton7950e382012-10-10 16:43:13 -0400171 }
172 result->name = kname;
Al Viro94b5d262015-02-22 19:38:03 -0500173 len = strncpy_from_user(kname, filename, PATH_MAX);
174 if (unlikely(len < 0)) {
175 __putname(kname);
176 kfree(result);
177 return ERR_PTR(len);
178 }
179 if (unlikely(len == PATH_MAX)) {
180 __putname(kname);
181 kfree(result);
182 return ERR_PTR(-ENAMETOOLONG);
183 }
Jeff Layton7950e382012-10-10 16:43:13 -0400184 }
185
Al Viro94b5d262015-02-22 19:38:03 -0500186 result->refcnt = 1;
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700187 /* The empty path is special. */
188 if (unlikely(!len)) {
189 if (empty)
Eric Paris4043cde2012-01-03 14:23:08 -0500190 *empty = 1;
Al Viro94b5d262015-02-22 19:38:03 -0500191 if (!(flags & LOOKUP_EMPTY)) {
192 putname(result);
193 return ERR_PTR(-ENOENT);
194 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 }
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700196
Jeff Layton7950e382012-10-10 16:43:13 -0400197 result->uptr = filename;
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800198 result->aname = NULL;
Jeff Layton7950e382012-10-10 16:43:13 -0400199 audit_getname(result);
200 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201}
202
Jeff Layton91a27b22012-10-10 15:25:28 -0400203struct filename *
204getname(const char __user * filename)
Al Virof52e0c12011-03-14 18:56:51 -0400205{
Linus Torvaldsf7493e52012-03-22 16:10:40 -0700206 return getname_flags(filename, 0, NULL);
Al Virof52e0c12011-03-14 18:56:51 -0400207}
208
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800209struct filename *
210getname_kernel(const char * filename)
211{
212 struct filename *result;
Paul Moore08518542015-01-21 23:59:56 -0500213 int len = strlen(filename) + 1;
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800214
215 result = __getname();
216 if (unlikely(!result))
217 return ERR_PTR(-ENOMEM);
218
Paul Moore08518542015-01-21 23:59:56 -0500219 if (len <= EMBEDDED_NAME_MAX) {
Al Virofd2f7cb2015-02-22 20:07:13 -0500220 result->name = (char *)result->iname;
Paul Moore08518542015-01-21 23:59:56 -0500221 } else if (len <= PATH_MAX) {
222 struct filename *tmp;
223
224 tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
225 if (unlikely(!tmp)) {
226 __putname(result);
227 return ERR_PTR(-ENOMEM);
228 }
229 tmp->name = (char *)result;
Paul Moore08518542015-01-21 23:59:56 -0500230 result = tmp;
231 } else {
232 __putname(result);
233 return ERR_PTR(-ENAMETOOLONG);
234 }
235 memcpy((char *)result->name, filename, len);
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800236 result->uptr = NULL;
237 result->aname = NULL;
Paul Moore55422d02015-01-22 00:00:23 -0500238 result->refcnt = 1;
Paul Moorefd3522f2015-01-22 00:00:10 -0500239 audit_getname(result);
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800240
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800241 return result;
242}
243
Jeff Layton91a27b22012-10-10 15:25:28 -0400244void putname(struct filename *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245{
Paul Moore55422d02015-01-22 00:00:23 -0500246 BUG_ON(name->refcnt <= 0);
247
248 if (--name->refcnt > 0)
249 return;
250
Al Virofd2f7cb2015-02-22 20:07:13 -0500251 if (name->name != name->iname) {
Paul Moore55422d02015-01-22 00:00:23 -0500252 __putname(name->name);
253 kfree(name);
254 } else
255 __putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257
Linus Torvaldse77819e2011-07-22 19:30:19 -0700258static int check_acl(struct inode *inode, int mask)
259{
Linus Torvalds84635d62011-07-25 22:47:03 -0700260#ifdef CONFIG_FS_POSIX_ACL
Linus Torvaldse77819e2011-07-22 19:30:19 -0700261 struct posix_acl *acl;
262
Linus Torvaldse77819e2011-07-22 19:30:19 -0700263 if (mask & MAY_NOT_BLOCK) {
Al Viro35678662011-08-02 21:32:13 -0400264 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
265 if (!acl)
Linus Torvaldse77819e2011-07-22 19:30:19 -0700266 return -EAGAIN;
Al Viro35678662011-08-02 21:32:13 -0400267 /* no ->get_acl() calls in RCU mode... */
268 if (acl == ACL_NOT_CACHED)
269 return -ECHILD;
Ari Savolainen206b1d02011-08-06 19:43:07 +0300270 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
Linus Torvaldse77819e2011-07-22 19:30:19 -0700271 }
272
Christoph Hellwig2982baa2013-12-20 05:16:38 -0800273 acl = get_acl(inode, ACL_TYPE_ACCESS);
274 if (IS_ERR(acl))
275 return PTR_ERR(acl);
Linus Torvaldse77819e2011-07-22 19:30:19 -0700276 if (acl) {
277 int error = posix_acl_permission(inode, acl, mask);
278 posix_acl_release(acl);
279 return error;
280 }
Linus Torvalds84635d62011-07-25 22:47:03 -0700281#endif
Linus Torvaldse77819e2011-07-22 19:30:19 -0700282
283 return -EAGAIN;
284}
285
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700286/*
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530287 * This does the basic permission checking
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700288 */
Al Viro7e401452011-06-20 19:12:17 -0400289static int acl_permission_check(struct inode *inode, int mask)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700290{
Linus Torvalds26cf46b2011-05-13 11:51:01 -0700291 unsigned int mode = inode->i_mode;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700292
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -0800293 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700294 mode >>= 6;
295 else {
Linus Torvaldse77819e2011-07-22 19:30:19 -0700296 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
Al Viro7e401452011-06-20 19:12:17 -0400297 int error = check_acl(inode, mask);
Nick Pigginb74c79e2011-01-07 17:49:58 +1100298 if (error != -EAGAIN)
299 return error;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700300 }
301
302 if (in_group_p(inode->i_gid))
303 mode >>= 3;
304 }
305
306 /*
307 * If the DACs are ok we don't need any capability check.
308 */
Al Viro9c2c7032011-06-20 19:06:22 -0400309 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700310 return 0;
311 return -EACCES;
312}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313
314/**
Nick Pigginb74c79e2011-01-07 17:49:58 +1100315 * generic_permission - check for access rights on a Posix-like filesystem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 * @inode: inode to check access rights for
Andreas Gruenbacher8fd90c82011-10-23 23:13:30 +0530317 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 *
319 * Used to check for read/write/execute permissions on a file.
320 * We use "fsuid" for this, letting us set arbitrary permissions
321 * for filesystem access without changing the "normal" uids which
Nick Pigginb74c79e2011-01-07 17:49:58 +1100322 * are used for other things.
323 *
324 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
325 * request cannot be satisfied (eg. requires blocking or too much complexity).
326 * It would then be called again in ref-walk mode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 */
Al Viro2830ba72011-06-20 19:16:29 -0400328int generic_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700330 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331
332 /*
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530333 * Do the basic permission checks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 */
Al Viro7e401452011-06-20 19:12:17 -0400335 ret = acl_permission_check(inode, mask);
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700336 if (ret != -EACCES)
337 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338
Al Virod594e7e2011-06-20 19:55:42 -0400339 if (S_ISDIR(inode->i_mode)) {
340 /* DACs are overridable for directories */
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700341 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
Al Virod594e7e2011-06-20 19:55:42 -0400342 return 0;
343 if (!(mask & MAY_WRITE))
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700344 if (capable_wrt_inode_uidgid(inode,
345 CAP_DAC_READ_SEARCH))
Al Virod594e7e2011-06-20 19:55:42 -0400346 return 0;
347 return -EACCES;
348 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 /*
350 * Read/write DACs are always overridable.
Al Virod594e7e2011-06-20 19:55:42 -0400351 * Executable DACs are overridable when there is
352 * at least one exec bit set.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 */
Al Virod594e7e2011-06-20 19:55:42 -0400354 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700355 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 return 0;
357
358 /*
359 * Searching includes executable on directories, else just read.
360 */
Serge E. Hallyn7ea66002009-12-29 14:50:19 -0600361 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
Al Virod594e7e2011-06-20 19:55:42 -0400362 if (mask == MAY_READ)
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700363 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 return 0;
365
366 return -EACCES;
367}
Al Viro4d359502014-03-14 12:20:17 -0400368EXPORT_SYMBOL(generic_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700370/*
371 * We _really_ want to just do "generic_permission()" without
372 * even looking at the inode->i_op values. So we keep a cache
373 * flag in inode->i_opflags, that says "this has not special
374 * permission function, use the fast case".
375 */
376static inline int do_inode_permission(struct inode *inode, int mask)
377{
378 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
379 if (likely(inode->i_op->permission))
380 return inode->i_op->permission(inode, mask);
381
382 /* This gets set once for the inode lifetime */
383 spin_lock(&inode->i_lock);
384 inode->i_opflags |= IOP_FASTPERM;
385 spin_unlock(&inode->i_lock);
386 }
387 return generic_permission(inode, mask);
388}
389
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200390/**
David Howells0bdaea92012-06-25 12:55:46 +0100391 * __inode_permission - Check for access rights to a given inode
392 * @inode: Inode to check permission on
393 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200394 *
David Howells0bdaea92012-06-25 12:55:46 +0100395 * Check for read/write/execute permissions on an inode.
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530396 *
397 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
David Howells0bdaea92012-06-25 12:55:46 +0100398 *
399 * This does not check for a read-only file system. You probably want
400 * inode_permission().
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200401 */
David Howells0bdaea92012-06-25 12:55:46 +0100402int __inode_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403{
Al Viroe6305c42008-07-15 21:03:57 -0400404 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700406 if (unlikely(mask & MAY_WRITE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 /*
408 * Nobody gets write access to an immutable file.
409 */
410 if (IS_IMMUTABLE(inode))
411 return -EACCES;
412 }
413
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700414 retval = do_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 if (retval)
416 return retval;
417
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -0700418 retval = devcgroup_inode_permission(inode, mask);
419 if (retval)
420 return retval;
421
Eric Parisd09ca732010-07-23 11:43:57 -0400422 return security_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423}
Miklos Szeredibd5d0852014-10-24 00:14:35 +0200424EXPORT_SYMBOL(__inode_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425
Al Virof4d6ff82011-06-19 13:14:21 -0400426/**
David Howells0bdaea92012-06-25 12:55:46 +0100427 * sb_permission - Check superblock-level permissions
428 * @sb: Superblock of inode to check permission on
Randy Dunlap55852632012-08-18 17:39:25 -0700429 * @inode: Inode to check permission on
David Howells0bdaea92012-06-25 12:55:46 +0100430 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
431 *
432 * Separate out file-system wide checks from inode-specific permission checks.
433 */
434static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
435{
436 if (unlikely(mask & MAY_WRITE)) {
437 umode_t mode = inode->i_mode;
438
439 /* Nobody gets write access to a read-only fs. */
440 if ((sb->s_flags & MS_RDONLY) &&
441 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
442 return -EROFS;
443 }
444 return 0;
445}
446
447/**
448 * inode_permission - Check for access rights to a given inode
449 * @inode: Inode to check permission on
450 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
451 *
452 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
453 * this, letting us set arbitrary permissions for filesystem access without
454 * changing the "normal" UIDs which are used for other things.
455 *
456 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
457 */
458int inode_permission(struct inode *inode, int mask)
459{
460 int retval;
461
462 retval = sb_permission(inode->i_sb, inode, mask);
463 if (retval)
464 return retval;
465 return __inode_permission(inode, mask);
466}
Al Viro4d359502014-03-14 12:20:17 -0400467EXPORT_SYMBOL(inode_permission);
David Howells0bdaea92012-06-25 12:55:46 +0100468
469/**
Jan Blunck5dd784d02008-02-14 19:34:38 -0800470 * path_get - get a reference to a path
471 * @path: path to get the reference to
472 *
473 * Given a path increment the reference count to the dentry and the vfsmount.
474 */
Al Virodcf787f2013-03-01 23:51:07 -0500475void path_get(const struct path *path)
Jan Blunck5dd784d02008-02-14 19:34:38 -0800476{
477 mntget(path->mnt);
478 dget(path->dentry);
479}
480EXPORT_SYMBOL(path_get);
481
482/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800483 * path_put - put a reference to a path
484 * @path: path to put the reference to
485 *
486 * Given a path decrement the reference count to the dentry and the vfsmount.
487 */
Al Virodcf787f2013-03-01 23:51:07 -0500488void path_put(const struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489{
Jan Blunck1d957f92008-02-14 19:34:35 -0800490 dput(path->dentry);
491 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492}
Jan Blunck1d957f92008-02-14 19:34:35 -0800493EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494
Al Viro894bc8c2015-05-02 07:16:16 -0400495#define EMBEDDED_LEVELS 2
Al Viro1f55a6e2014-11-01 19:30:41 -0400496struct nameidata {
497 struct path path;
Al Viro1cf26652015-05-06 16:01:56 -0400498 struct qstr last;
Al Viro1f55a6e2014-11-01 19:30:41 -0400499 struct path root;
500 struct inode *inode; /* path.dentry.d_inode */
501 unsigned int flags;
Al Viro9883d182015-05-13 07:28:08 -0400502 unsigned seq, m_seq;
Al Viro1f55a6e2014-11-01 19:30:41 -0400503 int last_type;
504 unsigned depth;
NeilBrown756daf22015-03-23 13:37:38 +1100505 int total_link_count;
Al Viro697fc6c2015-05-02 19:38:35 -0400506 struct saved {
507 struct path link;
508 void *cookie;
509 const char *name;
Al Viro237d8b32015-05-07 09:21:14 -0400510 struct inode *inode;
Al Viro0450b2d2015-05-08 13:23:53 -0400511 unsigned seq;
Al Viro894bc8c2015-05-02 07:16:16 -0400512 } *stack, internal[EMBEDDED_LEVELS];
Al Viro9883d182015-05-13 07:28:08 -0400513 struct filename *name;
514 struct nameidata *saved;
515 unsigned root_seq;
516 int dfd;
Al Viro1f55a6e2014-11-01 19:30:41 -0400517};
518
Al Viro9883d182015-05-13 07:28:08 -0400519static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
Al Viro894bc8c2015-05-02 07:16:16 -0400520{
NeilBrown756daf22015-03-23 13:37:38 +1100521 struct nameidata *old = current->nameidata;
522 p->stack = p->internal;
Al Viroc8a53ee2015-05-12 18:43:07 -0400523 p->dfd = dfd;
524 p->name = name;
NeilBrown756daf22015-03-23 13:37:38 +1100525 p->total_link_count = old ? old->total_link_count : 0;
Al Viro9883d182015-05-13 07:28:08 -0400526 p->saved = old;
NeilBrown756daf22015-03-23 13:37:38 +1100527 current->nameidata = p;
Al Viro894bc8c2015-05-02 07:16:16 -0400528}
529
Al Viro9883d182015-05-13 07:28:08 -0400530static void restore_nameidata(void)
Al Viro894bc8c2015-05-02 07:16:16 -0400531{
Al Viro9883d182015-05-13 07:28:08 -0400532 struct nameidata *now = current->nameidata, *old = now->saved;
NeilBrown756daf22015-03-23 13:37:38 +1100533
534 current->nameidata = old;
535 if (old)
536 old->total_link_count = now->total_link_count;
537 if (now->stack != now->internal) {
538 kfree(now->stack);
539 now->stack = now->internal;
Al Viro894bc8c2015-05-02 07:16:16 -0400540 }
541}
542
543static int __nd_alloc_stack(struct nameidata *nd)
544{
Al Virobc40aee2015-05-09 13:04:24 -0400545 struct saved *p;
546
547 if (nd->flags & LOOKUP_RCU) {
548 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
549 GFP_ATOMIC);
550 if (unlikely(!p))
551 return -ECHILD;
552 } else {
553 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
Al Viro894bc8c2015-05-02 07:16:16 -0400554 GFP_KERNEL);
Al Virobc40aee2015-05-09 13:04:24 -0400555 if (unlikely(!p))
556 return -ENOMEM;
557 }
Al Viro894bc8c2015-05-02 07:16:16 -0400558 memcpy(p, nd->internal, sizeof(nd->internal));
559 nd->stack = p;
560 return 0;
561}
562
563static inline int nd_alloc_stack(struct nameidata *nd)
564{
Al Viroda4e0be2015-05-03 20:52:15 -0400565 if (likely(nd->depth != EMBEDDED_LEVELS))
Al Viro894bc8c2015-05-02 07:16:16 -0400566 return 0;
567 if (likely(nd->stack != nd->internal))
568 return 0;
569 return __nd_alloc_stack(nd);
570}
571
Al Viro79733872015-05-09 12:55:43 -0400572static void drop_links(struct nameidata *nd)
573{
574 int i = nd->depth;
575 while (i--) {
576 struct saved *last = nd->stack + i;
577 struct inode *inode = last->inode;
578 if (last->cookie && inode->i_op->put_link) {
579 inode->i_op->put_link(inode, last->cookie);
580 last->cookie = NULL;
581 }
582 }
583}
584
585static void terminate_walk(struct nameidata *nd)
586{
587 drop_links(nd);
588 if (!(nd->flags & LOOKUP_RCU)) {
589 int i;
590 path_put(&nd->path);
591 for (i = 0; i < nd->depth; i++)
592 path_put(&nd->stack[i].link);
Al Viro102b8af2015-05-12 17:35:52 -0400593 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
594 path_put(&nd->root);
595 nd->root.mnt = NULL;
596 }
Al Viro79733872015-05-09 12:55:43 -0400597 } else {
598 nd->flags &= ~LOOKUP_RCU;
599 if (!(nd->flags & LOOKUP_ROOT))
600 nd->root.mnt = NULL;
601 rcu_read_unlock();
602 }
603 nd->depth = 0;
604}
605
606/* path_put is needed afterwards regardless of success or failure */
607static bool legitimize_path(struct nameidata *nd,
608 struct path *path, unsigned seq)
609{
610 int res = __legitimize_mnt(path->mnt, nd->m_seq);
611 if (unlikely(res)) {
612 if (res > 0)
613 path->mnt = NULL;
614 path->dentry = NULL;
615 return false;
616 }
617 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
618 path->dentry = NULL;
619 return false;
620 }
621 return !read_seqcount_retry(&path->dentry->d_seq, seq);
622}
623
624static bool legitimize_links(struct nameidata *nd)
625{
626 int i;
627 for (i = 0; i < nd->depth; i++) {
628 struct saved *last = nd->stack + i;
629 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
630 drop_links(nd);
631 nd->depth = i + 1;
632 return false;
633 }
634 }
635 return true;
636}
637
Al Viro19660af2011-03-25 10:32:48 -0400638/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100639 * Path walking has 2 modes, rcu-walk and ref-walk (see
Al Viro19660af2011-03-25 10:32:48 -0400640 * Documentation/filesystems/path-lookup.txt). In situations when we can't
641 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
642 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
643 * mode. Refcounts are grabbed at the last known good point before rcu-walk
644 * got stuck, so ref-walk may continue from there. If this is not successful
645 * (eg. a seqcount has changed), then failure is returned and it's up to caller
646 * to restart the path walk from the beginning in ref-walk mode.
Nick Piggin31e6b012011-01-07 17:49:52 +1100647 */
Nick Piggin31e6b012011-01-07 17:49:52 +1100648
649/**
Al Viro19660af2011-03-25 10:32:48 -0400650 * unlazy_walk - try to switch to ref-walk mode.
651 * @nd: nameidata pathwalk data
652 * @dentry: child of nd->path.dentry or NULL
Al Viro6e9918b2015-05-05 09:26:05 -0400653 * @seq: seq number to check dentry against
Randy Dunlap39191622011-01-08 19:36:21 -0800654 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100655 *
Al Viro19660af2011-03-25 10:32:48 -0400656 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
657 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
658 * @nd or NULL. Must be called from rcu-walk context.
Al Viro79733872015-05-09 12:55:43 -0400659 * Nothing should touch nameidata between unlazy_walk() failure and
660 * terminate_walk().
Nick Piggin31e6b012011-01-07 17:49:52 +1100661 */
Al Viro6e9918b2015-05-05 09:26:05 -0400662static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq)
Nick Piggin31e6b012011-01-07 17:49:52 +1100663{
Nick Piggin31e6b012011-01-07 17:49:52 +1100664 struct dentry *parent = nd->path.dentry;
665
666 BUG_ON(!(nd->flags & LOOKUP_RCU));
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700667
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700668 nd->flags &= ~LOOKUP_RCU;
Al Viro79733872015-05-09 12:55:43 -0400669 if (unlikely(!legitimize_links(nd)))
670 goto out2;
671 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
672 goto out2;
673 if (unlikely(!lockref_get_not_dead(&parent->d_lockref)))
674 goto out1;
Al Viro48a066e2013-09-29 22:06:07 -0400675
Linus Torvalds15570082013-09-02 11:38:06 -0700676 /*
677 * For a negative lookup, the lookup sequence point is the parents
678 * sequence point, and it only needs to revalidate the parent dentry.
679 *
680 * For a positive lookup, we need to move both the parent and the
681 * dentry from the RCU domain to be properly refcounted. And the
682 * sequence number in the dentry validates *both* dentry counters,
683 * since we checked the sequence number of the parent after we got
684 * the child sequence number. So we know the parent must still
685 * be valid if the child sequence number is still valid.
686 */
Al Viro19660af2011-03-25 10:32:48 -0400687 if (!dentry) {
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700688 if (read_seqcount_retry(&parent->d_seq, nd->seq))
689 goto out;
Al Viro19660af2011-03-25 10:32:48 -0400690 BUG_ON(nd->inode != parent->d_inode);
691 } else {
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700692 if (!lockref_get_not_dead(&dentry->d_lockref))
693 goto out;
Al Viro6e9918b2015-05-05 09:26:05 -0400694 if (read_seqcount_retry(&dentry->d_seq, seq))
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700695 goto drop_dentry;
Al Viro19660af2011-03-25 10:32:48 -0400696 }
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700697
698 /*
699 * Sequence counts matched. Now make sure that the root is
700 * still valid and get it if required.
701 */
702 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
Al Viro5a8d87e2015-05-12 00:10:18 -0400703 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
704 rcu_read_unlock();
705 dput(dentry);
706 return -ECHILD;
Al Viro79733872015-05-09 12:55:43 -0400707 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100708 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100709
Al Viro8b61e742013-11-08 12:45:01 -0500710 rcu_read_unlock();
Nick Piggin31e6b012011-01-07 17:49:52 +1100711 return 0;
Al Viro19660af2011-03-25 10:32:48 -0400712
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700713drop_dentry:
Al Viro8b61e742013-11-08 12:45:01 -0500714 rcu_read_unlock();
Linus Torvalds15570082013-09-02 11:38:06 -0700715 dput(dentry);
Linus Torvaldsd0d27272013-09-10 12:17:49 -0700716 goto drop_root_mnt;
Al Viro79733872015-05-09 12:55:43 -0400717out2:
718 nd->path.mnt = NULL;
719out1:
720 nd->path.dentry = NULL;
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700721out:
Al Viro8b61e742013-11-08 12:45:01 -0500722 rcu_read_unlock();
Linus Torvaldsd0d27272013-09-10 12:17:49 -0700723drop_root_mnt:
724 if (!(nd->flags & LOOKUP_ROOT))
725 nd->root.mnt = NULL;
Nick Piggin31e6b012011-01-07 17:49:52 +1100726 return -ECHILD;
727}
728
Al Viro79733872015-05-09 12:55:43 -0400729static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq)
730{
731 if (unlikely(!legitimize_path(nd, link, seq))) {
732 drop_links(nd);
733 nd->depth = 0;
734 nd->flags &= ~LOOKUP_RCU;
735 nd->path.mnt = NULL;
736 nd->path.dentry = NULL;
737 if (!(nd->flags & LOOKUP_ROOT))
738 nd->root.mnt = NULL;
739 rcu_read_unlock();
740 } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) {
741 return 0;
742 }
743 path_put(link);
744 return -ECHILD;
745}
746
Al Viro4ce16ef32012-06-10 16:10:59 -0400747static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
Nick Piggin34286d62011-01-07 17:49:57 +1100748{
Al Viro4ce16ef32012-06-10 16:10:59 -0400749 return dentry->d_op->d_revalidate(dentry, flags);
Nick Piggin34286d62011-01-07 17:49:57 +1100750}
751
Al Viro9f1fafe2011-03-25 11:00:12 -0400752/**
753 * complete_walk - successful completion of path walk
754 * @nd: pointer nameidata
Jeff Layton39159de2009-12-07 12:01:50 -0500755 *
Al Viro9f1fafe2011-03-25 11:00:12 -0400756 * If we had been in RCU mode, drop out of it and legitimize nd->path.
757 * Revalidate the final result, unless we'd already done that during
758 * the path walk or the filesystem doesn't ask for it. Return 0 on
759 * success, -error on failure. In case of failure caller does not
760 * need to drop nd->path.
Jeff Layton39159de2009-12-07 12:01:50 -0500761 */
Al Viro9f1fafe2011-03-25 11:00:12 -0400762static int complete_walk(struct nameidata *nd)
Jeff Layton39159de2009-12-07 12:01:50 -0500763{
Al Viro16c2cd72011-02-22 15:50:10 -0500764 struct dentry *dentry = nd->path.dentry;
Jeff Layton39159de2009-12-07 12:01:50 -0500765 int status;
Jeff Layton39159de2009-12-07 12:01:50 -0500766
Al Viro9f1fafe2011-03-25 11:00:12 -0400767 if (nd->flags & LOOKUP_RCU) {
Al Viro9f1fafe2011-03-25 11:00:12 -0400768 if (!(nd->flags & LOOKUP_ROOT))
769 nd->root.mnt = NULL;
Al Viro6e9918b2015-05-05 09:26:05 -0400770 if (unlikely(unlazy_walk(nd, NULL, 0)))
Al Viro48a066e2013-09-29 22:06:07 -0400771 return -ECHILD;
Al Viro9f1fafe2011-03-25 11:00:12 -0400772 }
773
Al Viro16c2cd72011-02-22 15:50:10 -0500774 if (likely(!(nd->flags & LOOKUP_JUMPED)))
Jeff Layton39159de2009-12-07 12:01:50 -0500775 return 0;
776
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500777 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
Al Viro16c2cd72011-02-22 15:50:10 -0500778 return 0;
779
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500780 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
Jeff Layton39159de2009-12-07 12:01:50 -0500781 if (status > 0)
782 return 0;
783
Al Viro16c2cd72011-02-22 15:50:10 -0500784 if (!status)
Jeff Layton39159de2009-12-07 12:01:50 -0500785 status = -ESTALE;
Al Viro16c2cd72011-02-22 15:50:10 -0500786
Jeff Layton39159de2009-12-07 12:01:50 -0500787 return status;
788}
789
Al Viro18d8c862015-05-12 16:32:34 -0400790static void set_root(struct nameidata *nd)
Al Viro2a737872009-04-07 11:49:53 -0400791{
Al Viro7bd88372014-09-13 21:55:46 -0400792 get_fs_root(current->fs, &nd->root);
Al Viro2a737872009-04-07 11:49:53 -0400793}
794
Al Viro06d71372015-06-29 12:07:04 -0400795static void set_root_rcu(struct nameidata *nd)
Nick Piggin31e6b012011-01-07 17:49:52 +1100796{
Al Viro7bd88372014-09-13 21:55:46 -0400797 struct fs_struct *fs = current->fs;
Al Viro8f47a0162015-05-09 19:02:01 -0400798 unsigned seq;
Nick Pigginc28cc362011-01-07 17:49:53 +1100799
Al Viro7bd88372014-09-13 21:55:46 -0400800 do {
801 seq = read_seqcount_begin(&fs->seq);
802 nd->root = fs->root;
Al Viro8f47a0162015-05-09 19:02:01 -0400803 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
Al Viro7bd88372014-09-13 21:55:46 -0400804 } while (read_seqcount_retry(&fs->seq, seq));
Nick Piggin31e6b012011-01-07 17:49:52 +1100805}
806
Jan Blunck1d957f92008-02-14 19:34:35 -0800807static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700808{
809 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800810 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700811 mntput(path->mnt);
812}
813
Nick Piggin7b9337a2011-01-14 08:42:43 +0000814static inline void path_to_nameidata(const struct path *path,
815 struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700816{
Nick Piggin31e6b012011-01-07 17:49:52 +1100817 if (!(nd->flags & LOOKUP_RCU)) {
818 dput(nd->path.dentry);
819 if (nd->path.mnt != path->mnt)
820 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800821 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100822 nd->path.mnt = path->mnt;
Jan Blunck4ac91372008-02-14 19:34:32 -0800823 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700824}
825
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400826/*
827 * Helper to directly jump to a known parsed path from ->follow_link,
828 * caller must have taken a reference to path beforehand.
829 */
Al Viro6e771372015-05-02 13:37:52 -0400830void nd_jump_link(struct path *path)
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400831{
Al Viro6e771372015-05-02 13:37:52 -0400832 struct nameidata *nd = current->nameidata;
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400833 path_put(&nd->path);
834
835 nd->path = *path;
836 nd->inode = nd->path.dentry->d_inode;
837 nd->flags |= LOOKUP_JUMPED;
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400838}
839
Al Virob9ff4422015-05-02 20:19:23 -0400840static inline void put_link(struct nameidata *nd)
Al Viro574197e2011-03-14 22:20:34 -0400841{
Al Viro21c30032015-05-03 21:06:24 -0400842 struct saved *last = nd->stack + --nd->depth;
Al Viro237d8b32015-05-07 09:21:14 -0400843 struct inode *inode = last->inode;
Al Virob9ff4422015-05-02 20:19:23 -0400844 if (last->cookie && inode->i_op->put_link)
Al Viro5f2c4172015-05-07 11:14:26 -0400845 inode->i_op->put_link(inode, last->cookie);
Al Viro6548fae2015-05-07 20:32:22 -0400846 if (!(nd->flags & LOOKUP_RCU))
847 path_put(&last->link);
Al Viro574197e2011-03-14 22:20:34 -0400848}
849
Linus Torvalds561ec642012-10-26 10:05:07 -0700850int sysctl_protected_symlinks __read_mostly = 0;
851int sysctl_protected_hardlinks __read_mostly = 0;
Kees Cook800179c2012-07-25 17:29:07 -0700852
853/**
854 * may_follow_link - Check symlink following for unsafe situations
Randy Dunlap55852632012-08-18 17:39:25 -0700855 * @nd: nameidata pathwalk data
Kees Cook800179c2012-07-25 17:29:07 -0700856 *
857 * In the case of the sysctl_protected_symlinks sysctl being enabled,
858 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
859 * in a sticky world-writable directory. This is to protect privileged
860 * processes from failing races against path names that may change out
861 * from under them by way of other users creating malicious symlinks.
862 * It will permit symlinks to be followed only when outside a sticky
863 * world-writable directory, or when the uid of the symlink and follower
864 * match, or when the directory owner matches the symlink's owner.
865 *
866 * Returns 0 if following the symlink is allowed, -ve on error.
867 */
Al Virofec2fa22015-05-06 15:58:18 -0400868static inline int may_follow_link(struct nameidata *nd)
Kees Cook800179c2012-07-25 17:29:07 -0700869{
870 const struct inode *inode;
871 const struct inode *parent;
872
873 if (!sysctl_protected_symlinks)
874 return 0;
875
876 /* Allowed if owner and follower match. */
Al Viro237d8b32015-05-07 09:21:14 -0400877 inode = nd->stack[0].inode;
Eric W. Biederman81abe272012-08-03 09:38:08 -0700878 if (uid_eq(current_cred()->fsuid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700879 return 0;
880
881 /* Allowed if parent directory not sticky and world-writable. */
Al Viroaa65fa32015-08-04 23:23:50 -0400882 parent = nd->inode;
Kees Cook800179c2012-07-25 17:29:07 -0700883 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
884 return 0;
885
886 /* Allowed if parent directory and link owner match. */
Eric W. Biederman81abe272012-08-03 09:38:08 -0700887 if (uid_eq(parent->i_uid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700888 return 0;
889
Al Viro319565022015-05-07 20:37:40 -0400890 if (nd->flags & LOOKUP_RCU)
891 return -ECHILD;
892
Al Viro1cf26652015-05-06 16:01:56 -0400893 audit_log_link_denied("follow_link", &nd->stack[0].link);
Kees Cook800179c2012-07-25 17:29:07 -0700894 return -EACCES;
895}
896
897/**
898 * safe_hardlink_source - Check for safe hardlink conditions
899 * @inode: the source inode to hardlink from
900 *
901 * Return false if at least one of the following conditions:
902 * - inode is not a regular file
903 * - inode is setuid
904 * - inode is setgid and group-exec
905 * - access failure for read and write
906 *
907 * Otherwise returns true.
908 */
909static bool safe_hardlink_source(struct inode *inode)
910{
911 umode_t mode = inode->i_mode;
912
913 /* Special files should not get pinned to the filesystem. */
914 if (!S_ISREG(mode))
915 return false;
916
917 /* Setuid files should not get pinned to the filesystem. */
918 if (mode & S_ISUID)
919 return false;
920
921 /* Executable setgid files should not get pinned to the filesystem. */
922 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
923 return false;
924
925 /* Hardlinking to unreadable or unwritable sources is dangerous. */
926 if (inode_permission(inode, MAY_READ | MAY_WRITE))
927 return false;
928
929 return true;
930}
931
932/**
933 * may_linkat - Check permissions for creating a hardlink
934 * @link: the source to hardlink from
935 *
936 * Block hardlink when all of:
937 * - sysctl_protected_hardlinks enabled
938 * - fsuid does not match inode
939 * - hardlink source is unsafe (see safe_hardlink_source() above)
940 * - not CAP_FOWNER
941 *
942 * Returns 0 if successful, -ve on error.
943 */
944static int may_linkat(struct path *link)
945{
946 const struct cred *cred;
947 struct inode *inode;
948
949 if (!sysctl_protected_hardlinks)
950 return 0;
951
952 cred = current_cred();
953 inode = link->dentry->d_inode;
954
955 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
956 * otherwise, it must be a safe source.
957 */
Eric W. Biederman81abe272012-08-03 09:38:08 -0700958 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
Kees Cook800179c2012-07-25 17:29:07 -0700959 capable(CAP_FOWNER))
960 return 0;
961
Kees Cooka51d9ea2012-07-25 17:29:08 -0700962 audit_log_link_denied("linkat", link);
Kees Cook800179c2012-07-25 17:29:07 -0700963 return -EPERM;
964}
965
Al Viro3b2e7f72015-04-19 00:53:50 -0400966static __always_inline
967const char *get_link(struct nameidata *nd)
Ian Kent051d3812006-03-27 01:14:53 -0800968{
Al Viroab104922015-05-10 11:50:01 -0400969 struct saved *last = nd->stack + nd->depth - 1;
Al Viro1cf26652015-05-06 16:01:56 -0400970 struct dentry *dentry = last->link.dentry;
Al Viro237d8b32015-05-07 09:21:14 -0400971 struct inode *inode = last->inode;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400972 int error;
Al Viro0a959df2015-04-18 18:23:41 -0400973 const char *res;
Ian Kent051d3812006-03-27 01:14:53 -0800974
NeilBrown8fa9dd22015-03-23 13:37:40 +1100975 if (!(nd->flags & LOOKUP_RCU)) {
976 touch_atime(&last->link);
977 cond_resched();
978 } else if (atime_needs_update(&last->link, inode)) {
979 if (unlikely(unlazy_walk(nd, NULL, 0)))
980 return ERR_PTR(-ECHILD);
981 touch_atime(&last->link);
982 }
983
NeilBrownbda0be72015-03-23 13:37:39 +1100984 error = security_inode_follow_link(dentry, inode,
985 nd->flags & LOOKUP_RCU);
986 if (unlikely(error))
Al Viro6920a442015-05-10 10:43:46 -0400987 return ERR_PTR(error);
Al Viro36f3b4f2011-02-22 21:24:38 -0500988
Al Viro86acdca12009-12-22 23:45:11 -0500989 nd->last_type = LAST_BIND;
Al Virod4dee482015-04-30 20:08:02 -0400990 res = inode->i_link;
991 if (!res) {
Al Viro8c1b4562015-05-09 18:15:21 -0400992 if (nd->flags & LOOKUP_RCU) {
993 if (unlikely(unlazy_walk(nd, NULL, 0)))
994 return ERR_PTR(-ECHILD);
995 }
Al Viro6e771372015-05-02 13:37:52 -0400996 res = inode->i_op->follow_link(dentry, &last->cookie);
Al Virofab51e82015-05-10 11:01:00 -0400997 if (IS_ERR_OR_NULL(res)) {
Al Viro6920a442015-05-10 10:43:46 -0400998 last->cookie = NULL;
Al Virofab51e82015-05-10 11:01:00 -0400999 return res;
1000 }
Ian Kent051d3812006-03-27 01:14:53 -08001001 }
Al Virofab51e82015-05-10 11:01:00 -04001002 if (*res == '/') {
Al Viro8c1b4562015-05-09 18:15:21 -04001003 if (nd->flags & LOOKUP_RCU) {
Al Viro8f47a0162015-05-09 19:02:01 -04001004 struct dentry *d;
1005 if (!nd->root.mnt)
1006 set_root_rcu(nd);
1007 nd->path = nd->root;
1008 d = nd->path.dentry;
1009 nd->inode = d->d_inode;
1010 nd->seq = nd->root_seq;
1011 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
Al Viro8c1b4562015-05-09 18:15:21 -04001012 return ERR_PTR(-ECHILD);
Al Viro8f47a0162015-05-09 19:02:01 -04001013 } else {
1014 if (!nd->root.mnt)
1015 set_root(nd);
1016 path_put(&nd->path);
1017 nd->path = nd->root;
1018 path_get(&nd->root);
1019 nd->inode = nd->path.dentry->d_inode;
Al Viro8c1b4562015-05-09 18:15:21 -04001020 }
Al Virofab51e82015-05-10 11:01:00 -04001021 nd->flags |= LOOKUP_JUMPED;
1022 while (unlikely(*++res == '/'))
1023 ;
1024 }
1025 if (!*res)
1026 res = NULL;
Al Viro0a959df2015-04-18 18:23:41 -04001027 return res;
1028}
Al Viro6d7b5aa2012-06-10 04:15:17 -04001029
David Howellsf015f1262012-06-25 12:55:28 +01001030/*
1031 * follow_up - Find the mountpoint of path's vfsmount
1032 *
1033 * Given a path, find the mountpoint of its source file system.
1034 * Replace @path with the path of the mountpoint in the parent mount.
1035 * Up is towards /.
1036 *
1037 * Return 1 if we went up a level and 0 if we were already at the
1038 * root.
1039 */
Al Virobab77eb2009-04-18 03:26:48 -04001040int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041{
Al Viro0714a532011-11-24 22:19:58 -05001042 struct mount *mnt = real_mount(path->mnt);
1043 struct mount *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +10001045
Al Viro48a066e2013-09-29 22:06:07 -04001046 read_seqlock_excl(&mount_lock);
Al Viro0714a532011-11-24 22:19:58 -05001047 parent = mnt->mnt_parent;
Al Viro3c0a6162012-07-18 17:32:50 +04001048 if (parent == mnt) {
Al Viro48a066e2013-09-29 22:06:07 -04001049 read_sequnlock_excl(&mount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 return 0;
1051 }
Al Viro0714a532011-11-24 22:19:58 -05001052 mntget(&parent->mnt);
Al Viroa73324d2011-11-24 22:25:07 -05001053 mountpoint = dget(mnt->mnt_mountpoint);
Al Viro48a066e2013-09-29 22:06:07 -04001054 read_sequnlock_excl(&mount_lock);
Al Virobab77eb2009-04-18 03:26:48 -04001055 dput(path->dentry);
1056 path->dentry = mountpoint;
1057 mntput(path->mnt);
Al Viro0714a532011-11-24 22:19:58 -05001058 path->mnt = &parent->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 return 1;
1060}
Al Viro4d359502014-03-14 12:20:17 -04001061EXPORT_SYMBOL(follow_up);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001063/*
David Howells9875cf82011-01-14 18:45:21 +00001064 * Perform an automount
1065 * - return -EISDIR to tell follow_managed() to stop and return the path we
1066 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 */
NeilBrown756daf22015-03-23 13:37:38 +11001068static int follow_automount(struct path *path, struct nameidata *nd,
David Howells9875cf82011-01-14 18:45:21 +00001069 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +11001070{
David Howells9875cf82011-01-14 18:45:21 +00001071 struct vfsmount *mnt;
David Howellsea5b7782011-01-14 19:10:03 +00001072 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001073
David Howells9875cf82011-01-14 18:45:21 +00001074 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1075 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -07001076
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +02001077 /* We don't want to mount if someone's just doing a stat -
1078 * unless they're stat'ing a directory and appended a '/' to
1079 * the name.
1080 *
1081 * We do, however, want to mount if someone wants to open or
1082 * create a file of any type under the mountpoint, wants to
1083 * traverse through the mountpoint or wants to open the
1084 * mounted directory. Also, autofs may mark negative dentries
1085 * as being automount points. These will need the attentions
1086 * of the daemon to instantiate them before they can be used.
David Howells9875cf82011-01-14 18:45:21 +00001087 */
NeilBrown756daf22015-03-23 13:37:38 +11001088 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1089 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +02001090 path->dentry->d_inode)
1091 return -EISDIR;
1092
NeilBrown756daf22015-03-23 13:37:38 +11001093 nd->total_link_count++;
1094 if (nd->total_link_count >= 40)
David Howells9875cf82011-01-14 18:45:21 +00001095 return -ELOOP;
1096
1097 mnt = path->dentry->d_op->d_automount(path);
1098 if (IS_ERR(mnt)) {
1099 /*
1100 * The filesystem is allowed to return -EISDIR here to indicate
1101 * it doesn't want to automount. For instance, autofs would do
1102 * this so that its userspace daemon can mount on this dentry.
1103 *
1104 * However, we can only permit this if it's a terminal point in
1105 * the path being looked up; if it wasn't then the remainder of
1106 * the path is inaccessible and we should say so.
1107 */
NeilBrown756daf22015-03-23 13:37:38 +11001108 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
David Howells9875cf82011-01-14 18:45:21 +00001109 return -EREMOTE;
1110 return PTR_ERR(mnt);
1111 }
David Howellsea5b7782011-01-14 19:10:03 +00001112
David Howells9875cf82011-01-14 18:45:21 +00001113 if (!mnt) /* mount collision */
1114 return 0;
1115
Al Viro8aef1882011-06-16 15:10:06 +01001116 if (!*need_mntput) {
1117 /* lock_mount() may release path->mnt on error */
1118 mntget(path->mnt);
1119 *need_mntput = true;
1120 }
Al Viro19a167a2011-01-17 01:35:23 -05001121 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +00001122
David Howellsea5b7782011-01-14 19:10:03 +00001123 switch (err) {
1124 case -EBUSY:
1125 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -05001126 return 0;
David Howellsea5b7782011-01-14 19:10:03 +00001127 case 0:
Al Viro8aef1882011-06-16 15:10:06 +01001128 path_put(path);
David Howellsea5b7782011-01-14 19:10:03 +00001129 path->mnt = mnt;
1130 path->dentry = dget(mnt->mnt_root);
David Howellsea5b7782011-01-14 19:10:03 +00001131 return 0;
Al Viro19a167a2011-01-17 01:35:23 -05001132 default:
1133 return err;
David Howellsea5b7782011-01-14 19:10:03 +00001134 }
Al Viro19a167a2011-01-17 01:35:23 -05001135
David Howells9875cf82011-01-14 18:45:21 +00001136}
1137
1138/*
1139 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +00001140 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +00001141 * - Flagged as mountpoint
1142 * - Flagged as automount point
1143 *
1144 * This may only be called in refwalk mode.
1145 *
1146 * Serialization is taken care of in namespace.c
1147 */
NeilBrown756daf22015-03-23 13:37:38 +11001148static int follow_managed(struct path *path, struct nameidata *nd)
David Howells9875cf82011-01-14 18:45:21 +00001149{
Al Viro8aef1882011-06-16 15:10:06 +01001150 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
David Howells9875cf82011-01-14 18:45:21 +00001151 unsigned managed;
1152 bool need_mntput = false;
Al Viro8aef1882011-06-16 15:10:06 +01001153 int ret = 0;
David Howells9875cf82011-01-14 18:45:21 +00001154
1155 /* Given that we're not holding a lock here, we retain the value in a
1156 * local variable for each dentry as we look at it so that we don't see
1157 * the components of that value change under us */
1158 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1159 managed &= DCACHE_MANAGED_DENTRY,
1160 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +00001161 /* Allow the filesystem to manage the transit without i_mutex
1162 * being held. */
1163 if (managed & DCACHE_MANAGE_TRANSIT) {
1164 BUG_ON(!path->dentry->d_op);
1165 BUG_ON(!path->dentry->d_op->d_manage);
Al Viro1aed3e42011-03-18 09:09:02 -04001166 ret = path->dentry->d_op->d_manage(path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001167 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001168 break;
David Howellscc53ce52011-01-14 18:45:26 +00001169 }
1170
David Howells9875cf82011-01-14 18:45:21 +00001171 /* Transit to a mounted filesystem. */
1172 if (managed & DCACHE_MOUNTED) {
1173 struct vfsmount *mounted = lookup_mnt(path);
1174 if (mounted) {
1175 dput(path->dentry);
1176 if (need_mntput)
1177 mntput(path->mnt);
1178 path->mnt = mounted;
1179 path->dentry = dget(mounted->mnt_root);
1180 need_mntput = true;
1181 continue;
1182 }
1183
1184 /* Something is mounted on this dentry in another
1185 * namespace and/or whatever was mounted there in this
Al Viro48a066e2013-09-29 22:06:07 -04001186 * namespace got unmounted before lookup_mnt() could
1187 * get it */
David Howells9875cf82011-01-14 18:45:21 +00001188 }
1189
1190 /* Handle an automount point */
1191 if (managed & DCACHE_NEED_AUTOMOUNT) {
NeilBrown756daf22015-03-23 13:37:38 +11001192 ret = follow_automount(path, nd, &need_mntput);
David Howells9875cf82011-01-14 18:45:21 +00001193 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001194 break;
David Howells9875cf82011-01-14 18:45:21 +00001195 continue;
1196 }
1197
1198 /* We didn't change the current path point */
1199 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 }
Al Viro8aef1882011-06-16 15:10:06 +01001201
1202 if (need_mntput && path->mnt == mnt)
1203 mntput(path->mnt);
1204 if (ret == -EISDIR)
1205 ret = 0;
Al Viro84027522015-04-22 10:30:08 -04001206 if (need_mntput)
1207 nd->flags |= LOOKUP_JUMPED;
1208 if (unlikely(ret < 0))
1209 path_put_conditional(path, nd);
1210 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211}
1212
David Howellscc53ce52011-01-14 18:45:26 +00001213int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214{
1215 struct vfsmount *mounted;
1216
Al Viro1c755af2009-04-18 14:06:57 -04001217 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -04001219 dput(path->dentry);
1220 mntput(path->mnt);
1221 path->mnt = mounted;
1222 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 return 1;
1224 }
1225 return 0;
1226}
Al Viro4d359502014-03-14 12:20:17 -04001227EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228
NeilBrownb8faf032014-08-04 17:06:29 +10001229static inline int managed_dentry_rcu(struct dentry *dentry)
Ian Kent62a73752011-03-25 01:51:02 +08001230{
NeilBrownb8faf032014-08-04 17:06:29 +10001231 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1232 dentry->d_op->d_manage(dentry, true) : 0;
Ian Kent62a73752011-03-25 01:51:02 +08001233}
1234
David Howells9875cf82011-01-14 18:45:21 +00001235/*
Al Viro287548e2011-05-27 06:50:06 -04001236 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1237 * we meet a managed dentry that would need blocking.
David Howells9875cf82011-01-14 18:45:21 +00001238 */
1239static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
Al Viro254cf582015-05-05 09:40:46 -04001240 struct inode **inode, unsigned *seqp)
David Howells9875cf82011-01-14 18:45:21 +00001241{
Ian Kent62a73752011-03-25 01:51:02 +08001242 for (;;) {
Al Viroc7105362011-11-24 18:22:03 -05001243 struct mount *mounted;
Ian Kent62a73752011-03-25 01:51:02 +08001244 /*
1245 * Don't forget we might have a non-mountpoint managed dentry
1246 * that wants to block transit.
1247 */
NeilBrownb8faf032014-08-04 17:06:29 +10001248 switch (managed_dentry_rcu(path->dentry)) {
1249 case -ECHILD:
1250 default:
David Howellsab909112011-01-14 18:46:51 +00001251 return false;
NeilBrownb8faf032014-08-04 17:06:29 +10001252 case -EISDIR:
1253 return true;
1254 case 0:
1255 break;
1256 }
Ian Kent62a73752011-03-25 01:51:02 +08001257
1258 if (!d_mountpoint(path->dentry))
NeilBrownb8faf032014-08-04 17:06:29 +10001259 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
Ian Kent62a73752011-03-25 01:51:02 +08001260
Al Viro474279d2013-10-01 16:11:26 -04001261 mounted = __lookup_mnt(path->mnt, path->dentry);
David Howells9875cf82011-01-14 18:45:21 +00001262 if (!mounted)
1263 break;
Al Viroc7105362011-11-24 18:22:03 -05001264 path->mnt = &mounted->mnt;
1265 path->dentry = mounted->mnt.mnt_root;
Al Viroa3fbbde2011-11-07 21:21:26 +00001266 nd->flags |= LOOKUP_JUMPED;
Al Viro254cf582015-05-05 09:40:46 -04001267 *seqp = read_seqcount_begin(&path->dentry->d_seq);
Linus Torvalds59430262011-07-18 15:43:29 -07001268 /*
1269 * Update the inode too. We don't need to re-check the
1270 * dentry sequence number here after this d_inode read,
1271 * because a mount-point is always pinned.
1272 */
1273 *inode = path->dentry->d_inode;
David Howells9875cf82011-01-14 18:45:21 +00001274 }
Al Virof5be3e29122014-09-13 21:50:45 -04001275 return !read_seqretry(&mount_lock, nd->m_seq) &&
NeilBrownb8faf032014-08-04 17:06:29 +10001276 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
Al Viro287548e2011-05-27 06:50:06 -04001277}
1278
Nick Piggin31e6b012011-01-07 17:49:52 +11001279static int follow_dotdot_rcu(struct nameidata *nd)
1280{
Al Viro4023bfc2014-09-13 21:59:43 -04001281 struct inode *inode = nd->inode;
Al Viro7bd88372014-09-13 21:55:46 -04001282 if (!nd->root.mnt)
1283 set_root_rcu(nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11001284
David Howells9875cf82011-01-14 18:45:21 +00001285 while (1) {
Al Viroaed434a2015-05-12 12:22:47 -04001286 if (path_equal(&nd->path, &nd->root))
Nick Piggin31e6b012011-01-07 17:49:52 +11001287 break;
Nick Piggin31e6b012011-01-07 17:49:52 +11001288 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1289 struct dentry *old = nd->path.dentry;
1290 struct dentry *parent = old->d_parent;
1291 unsigned seq;
1292
Al Viro4023bfc2014-09-13 21:59:43 -04001293 inode = parent->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001294 seq = read_seqcount_begin(&parent->d_seq);
Al Viroaed434a2015-05-12 12:22:47 -04001295 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1296 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001297 nd->path.dentry = parent;
1298 nd->seq = seq;
1299 break;
Al Viroaed434a2015-05-12 12:22:47 -04001300 } else {
1301 struct mount *mnt = real_mount(nd->path.mnt);
1302 struct mount *mparent = mnt->mnt_parent;
1303 struct dentry *mountpoint = mnt->mnt_mountpoint;
1304 struct inode *inode2 = mountpoint->d_inode;
1305 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1306 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1307 return -ECHILD;
1308 if (&mparent->mnt == nd->path.mnt)
1309 break;
1310 /* we know that mountpoint was pinned */
1311 nd->path.dentry = mountpoint;
1312 nd->path.mnt = &mparent->mnt;
1313 inode = inode2;
1314 nd->seq = seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11001315 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001316 }
Al Viroaed434a2015-05-12 12:22:47 -04001317 while (unlikely(d_mountpoint(nd->path.dentry))) {
Al Virob37199e2014-03-20 15:18:22 -04001318 struct mount *mounted;
1319 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
Al Viroaed434a2015-05-12 12:22:47 -04001320 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1321 return -ECHILD;
Al Virob37199e2014-03-20 15:18:22 -04001322 if (!mounted)
1323 break;
1324 nd->path.mnt = &mounted->mnt;
1325 nd->path.dentry = mounted->mnt.mnt_root;
Al Viro4023bfc2014-09-13 21:59:43 -04001326 inode = nd->path.dentry->d_inode;
Al Virob37199e2014-03-20 15:18:22 -04001327 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Virob37199e2014-03-20 15:18:22 -04001328 }
Al Viro4023bfc2014-09-13 21:59:43 -04001329 nd->inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001330 return 0;
1331}
1332
David Howells9875cf82011-01-14 18:45:21 +00001333/*
David Howellscc53ce52011-01-14 18:45:26 +00001334 * Follow down to the covering mount currently visible to userspace. At each
1335 * point, the filesystem owning that dentry may be queried as to whether the
1336 * caller is permitted to proceed or not.
David Howellscc53ce52011-01-14 18:45:26 +00001337 */
Al Viro7cc90cc2011-03-18 09:04:20 -04001338int follow_down(struct path *path)
David Howellscc53ce52011-01-14 18:45:26 +00001339{
1340 unsigned managed;
1341 int ret;
1342
1343 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1344 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1345 /* Allow the filesystem to manage the transit without i_mutex
1346 * being held.
1347 *
1348 * We indicate to the filesystem if someone is trying to mount
1349 * something here. This gives autofs the chance to deny anyone
1350 * other than its daemon the right to mount on its
1351 * superstructure.
1352 *
1353 * The filesystem may sleep at this point.
1354 */
1355 if (managed & DCACHE_MANAGE_TRANSIT) {
1356 BUG_ON(!path->dentry->d_op);
1357 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +00001358 ret = path->dentry->d_op->d_manage(
Al Viro1aed3e42011-03-18 09:09:02 -04001359 path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001360 if (ret < 0)
1361 return ret == -EISDIR ? 0 : ret;
1362 }
1363
1364 /* Transit to a mounted filesystem. */
1365 if (managed & DCACHE_MOUNTED) {
1366 struct vfsmount *mounted = lookup_mnt(path);
1367 if (!mounted)
1368 break;
1369 dput(path->dentry);
1370 mntput(path->mnt);
1371 path->mnt = mounted;
1372 path->dentry = dget(mounted->mnt_root);
1373 continue;
1374 }
1375
1376 /* Don't handle automount points here */
1377 break;
1378 }
1379 return 0;
1380}
Al Viro4d359502014-03-14 12:20:17 -04001381EXPORT_SYMBOL(follow_down);
David Howellscc53ce52011-01-14 18:45:26 +00001382
1383/*
David Howells9875cf82011-01-14 18:45:21 +00001384 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1385 */
1386static void follow_mount(struct path *path)
1387{
1388 while (d_mountpoint(path->dentry)) {
1389 struct vfsmount *mounted = lookup_mnt(path);
1390 if (!mounted)
1391 break;
1392 dput(path->dentry);
1393 mntput(path->mnt);
1394 path->mnt = mounted;
1395 path->dentry = dget(mounted->mnt_root);
1396 }
1397}
1398
Nick Piggin31e6b012011-01-07 17:49:52 +11001399static void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400{
Al Viro7bd88372014-09-13 21:55:46 -04001401 if (!nd->root.mnt)
1402 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001403
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 while(1) {
Jan Blunck4ac91372008-02-14 19:34:32 -08001405 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406
Al Viro2a737872009-04-07 11:49:53 -04001407 if (nd->path.dentry == nd->root.dentry &&
1408 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 break;
1410 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001411 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Al Viro3088dd72010-01-30 15:47:29 -05001412 /* rare case of legitimate dget_parent()... */
1413 nd->path.dentry = dget_parent(nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 dput(old);
1415 break;
1416 }
Al Viro3088dd72010-01-30 15:47:29 -05001417 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 }
Al Viro79ed0222009-04-18 13:59:41 -04001420 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001421 nd->inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422}
1423
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424/*
Miklos Szeredibad61182012-03-26 12:54:24 +02001425 * This looks up the name in dcache, possibly revalidates the old dentry and
1426 * allocates a new one if not found or not valid. In the need_lookup argument
1427 * returns whether i_op->lookup is necessary.
1428 *
1429 * dir->d_inode->i_mutex must be held
Nick Pigginbaa03892010-08-18 04:37:31 +10001430 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001431static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
Al Viro201f9562012-06-22 12:42:10 +04001432 unsigned int flags, bool *need_lookup)
Nick Pigginbaa03892010-08-18 04:37:31 +10001433{
Nick Pigginbaa03892010-08-18 04:37:31 +10001434 struct dentry *dentry;
Miklos Szeredibad61182012-03-26 12:54:24 +02001435 int error;
Nick Pigginbaa03892010-08-18 04:37:31 +10001436
Miklos Szeredibad61182012-03-26 12:54:24 +02001437 *need_lookup = false;
1438 dentry = d_lookup(dir, name);
1439 if (dentry) {
Jeff Layton39e3c952012-11-28 11:30:53 -05001440 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
Al Viro201f9562012-06-22 12:42:10 +04001441 error = d_revalidate(dentry, flags);
Miklos Szeredibad61182012-03-26 12:54:24 +02001442 if (unlikely(error <= 0)) {
1443 if (error < 0) {
1444 dput(dentry);
1445 return ERR_PTR(error);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001446 } else {
1447 d_invalidate(dentry);
Miklos Szeredibad61182012-03-26 12:54:24 +02001448 dput(dentry);
1449 dentry = NULL;
1450 }
1451 }
1452 }
1453 }
Nick Pigginbaa03892010-08-18 04:37:31 +10001454
Miklos Szeredibad61182012-03-26 12:54:24 +02001455 if (!dentry) {
1456 dentry = d_alloc(dir, name);
1457 if (unlikely(!dentry))
1458 return ERR_PTR(-ENOMEM);
Nick Pigginbaa03892010-08-18 04:37:31 +10001459
Miklos Szeredibad61182012-03-26 12:54:24 +02001460 *need_lookup = true;
Nick Pigginbaa03892010-08-18 04:37:31 +10001461 }
1462 return dentry;
1463}
1464
1465/*
J. Bruce Fields13a2c3b2013-10-23 16:09:16 -04001466 * Call i_op->lookup on the dentry. The dentry must be negative and
1467 * unhashed.
Miklos Szeredibad61182012-03-26 12:54:24 +02001468 *
1469 * dir->d_inode->i_mutex must be held
Josef Bacik44396f42011-05-31 11:58:49 -04001470 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001471static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
Al Viro72bd8662012-06-10 17:17:17 -04001472 unsigned int flags)
Josef Bacik44396f42011-05-31 11:58:49 -04001473{
Josef Bacik44396f42011-05-31 11:58:49 -04001474 struct dentry *old;
1475
1476 /* Don't create child dentry for a dead directory. */
Miklos Szeredibad61182012-03-26 12:54:24 +02001477 if (unlikely(IS_DEADDIR(dir))) {
Miklos Szeredie188dc02012-02-03 14:25:18 +01001478 dput(dentry);
Josef Bacik44396f42011-05-31 11:58:49 -04001479 return ERR_PTR(-ENOENT);
Miklos Szeredie188dc02012-02-03 14:25:18 +01001480 }
Josef Bacik44396f42011-05-31 11:58:49 -04001481
Al Viro72bd8662012-06-10 17:17:17 -04001482 old = dir->i_op->lookup(dir, dentry, flags);
Josef Bacik44396f42011-05-31 11:58:49 -04001483 if (unlikely(old)) {
1484 dput(dentry);
1485 dentry = old;
1486 }
1487 return dentry;
1488}
1489
Al Viroa3255542012-03-30 14:41:51 -04001490static struct dentry *__lookup_hash(struct qstr *name,
Al Viro72bd8662012-06-10 17:17:17 -04001491 struct dentry *base, unsigned int flags)
Al Viroa3255542012-03-30 14:41:51 -04001492{
Miklos Szeredibad61182012-03-26 12:54:24 +02001493 bool need_lookup;
Al Viroa3255542012-03-30 14:41:51 -04001494 struct dentry *dentry;
1495
Al Viro72bd8662012-06-10 17:17:17 -04001496 dentry = lookup_dcache(name, base, flags, &need_lookup);
Miklos Szeredibad61182012-03-26 12:54:24 +02001497 if (!need_lookup)
1498 return dentry;
Al Viroa3255542012-03-30 14:41:51 -04001499
Al Viro72bd8662012-06-10 17:17:17 -04001500 return lookup_real(base->d_inode, dentry, flags);
Al Viroa3255542012-03-30 14:41:51 -04001501}
1502
Josef Bacik44396f42011-05-31 11:58:49 -04001503/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 * It's more convoluted than I'd like it to be, but... it's still fairly
1505 * small and for now I'd prefer to have fast path as straight as possible.
1506 * It _is_ time-critical.
1507 */
Al Viroe97cdc82013-01-24 18:16:00 -05001508static int lookup_fast(struct nameidata *nd,
Al Viro254cf582015-05-05 09:40:46 -04001509 struct path *path, struct inode **inode,
1510 unsigned *seqp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511{
Jan Blunck4ac91372008-02-14 19:34:32 -08001512 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001513 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001514 int need_reval = 1;
1515 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001516 int err;
1517
Al Viro3cac2602009-08-13 18:27:43 +04001518 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001519 * Rename seqlock is not required here because in the off chance
1520 * of a false negative due to a concurrent rename, we're going to
1521 * do the non-racy lookup, below.
1522 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001523 if (nd->flags & LOOKUP_RCU) {
1524 unsigned seq;
Al Viro766c4cb2015-05-07 19:24:57 -04001525 bool negative;
Linus Torvaldsda53be12013-05-21 15:22:44 -07001526 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
Al Viro5a18fff2011-03-11 04:44:53 -05001527 if (!dentry)
1528 goto unlazy;
1529
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001530 /*
1531 * This sequence count validates that the inode matches
1532 * the dentry name information from lookup.
1533 */
David Howells63afdfc2015-05-06 15:59:00 +01001534 *inode = d_backing_inode(dentry);
Al Viro766c4cb2015-05-07 19:24:57 -04001535 negative = d_is_negative(dentry);
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001536 if (read_seqcount_retry(&dentry->d_seq, seq))
1537 return -ECHILD;
Al Viro766c4cb2015-05-07 19:24:57 -04001538 if (negative)
1539 return -ENOENT;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001540
1541 /*
1542 * This sequence count validates that the parent had no
1543 * changes while we did the lookup of the dentry above.
1544 *
1545 * The memory barrier in read_seqcount_begin of child is
1546 * enough, we can use __read_seqcount_retry here.
1547 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001548 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1549 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001550
Al Viro254cf582015-05-05 09:40:46 -04001551 *seqp = seq;
Al Viro24643082011-02-15 01:26:22 -05001552 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
Al Viro4ce16ef32012-06-10 16:10:59 -04001553 status = d_revalidate(dentry, nd->flags);
Al Viro5a18fff2011-03-11 04:44:53 -05001554 if (unlikely(status <= 0)) {
1555 if (status != -ECHILD)
1556 need_reval = 0;
1557 goto unlazy;
1558 }
Al Viro24643082011-02-15 01:26:22 -05001559 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001560 path->mnt = mnt;
1561 path->dentry = dentry;
Al Viro254cf582015-05-05 09:40:46 -04001562 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
NeilBrownb8faf032014-08-04 17:06:29 +10001563 return 0;
Al Viro5a18fff2011-03-11 04:44:53 -05001564unlazy:
Al Viro254cf582015-05-05 09:40:46 -04001565 if (unlazy_walk(nd, dentry, seq))
Al Viro19660af2011-03-25 10:32:48 -04001566 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001567 } else {
Al Viroe97cdc82013-01-24 18:16:00 -05001568 dentry = __d_lookup(parent, &nd->last);
Nick Piggin31e6b012011-01-07 17:49:52 +11001569 }
Al Viro5a18fff2011-03-11 04:44:53 -05001570
Al Viro81e6f522012-03-30 14:48:04 -04001571 if (unlikely(!dentry))
1572 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001573
Al Viro5a18fff2011-03-11 04:44:53 -05001574 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
Al Viro4ce16ef32012-06-10 16:10:59 -04001575 status = d_revalidate(dentry, nd->flags);
Al Viro5a18fff2011-03-11 04:44:53 -05001576 if (unlikely(status <= 0)) {
1577 if (status < 0) {
1578 dput(dentry);
1579 return status;
1580 }
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001581 d_invalidate(dentry);
1582 dput(dentry);
1583 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001584 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001585
Al Viro766c4cb2015-05-07 19:24:57 -04001586 if (unlikely(d_is_negative(dentry))) {
1587 dput(dentry);
1588 return -ENOENT;
1589 }
David Howells9875cf82011-01-14 18:45:21 +00001590 path->mnt = mnt;
1591 path->dentry = dentry;
NeilBrown756daf22015-03-23 13:37:38 +11001592 err = follow_managed(path, nd);
Al Viro84027522015-04-22 10:30:08 -04001593 if (likely(!err))
David Howells63afdfc2015-05-06 15:59:00 +01001594 *inode = d_backing_inode(path->dentry);
Al Viro84027522015-04-22 10:30:08 -04001595 return err;
Al Viro81e6f522012-03-30 14:48:04 -04001596
1597need_lookup:
Miklos Szeredi697f5142012-05-21 17:30:05 +02001598 return 1;
1599}
1600
1601/* Fast lookup failed, do it the slow way */
Al Virocc2a5272013-01-24 18:19:49 -05001602static int lookup_slow(struct nameidata *nd, struct path *path)
Miklos Szeredi697f5142012-05-21 17:30:05 +02001603{
1604 struct dentry *dentry, *parent;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001605
1606 parent = nd->path.dentry;
Al Viro81e6f522012-03-30 14:48:04 -04001607 BUG_ON(nd->inode != parent->d_inode);
1608
1609 mutex_lock(&parent->d_inode->i_mutex);
Al Virocc2a5272013-01-24 18:19:49 -05001610 dentry = __lookup_hash(&nd->last, parent, nd->flags);
Al Viro81e6f522012-03-30 14:48:04 -04001611 mutex_unlock(&parent->d_inode->i_mutex);
1612 if (IS_ERR(dentry))
1613 return PTR_ERR(dentry);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001614 path->mnt = nd->path.mnt;
1615 path->dentry = dentry;
Al Viro84027522015-04-22 10:30:08 -04001616 return follow_managed(path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617}
1618
Al Viro52094c82011-02-21 21:34:47 -05001619static inline int may_lookup(struct nameidata *nd)
1620{
1621 if (nd->flags & LOOKUP_RCU) {
Al Viro4ad5abb2011-06-20 19:57:03 -04001622 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
Al Viro52094c82011-02-21 21:34:47 -05001623 if (err != -ECHILD)
1624 return err;
Al Viro6e9918b2015-05-05 09:26:05 -04001625 if (unlazy_walk(nd, NULL, 0))
Al Viro52094c82011-02-21 21:34:47 -05001626 return -ECHILD;
1627 }
Al Viro4ad5abb2011-06-20 19:57:03 -04001628 return inode_permission(nd->inode, MAY_EXEC);
Al Viro52094c82011-02-21 21:34:47 -05001629}
1630
Al Viro9856fa12011-03-04 14:22:06 -05001631static inline int handle_dots(struct nameidata *nd, int type)
1632{
1633 if (type == LAST_DOTDOT) {
1634 if (nd->flags & LOOKUP_RCU) {
Al Viro70291ae2015-05-04 07:53:00 -04001635 return follow_dotdot_rcu(nd);
Al Viro9856fa12011-03-04 14:22:06 -05001636 } else
1637 follow_dotdot(nd);
1638 }
1639 return 0;
1640}
1641
Al Viro181548c2015-05-07 19:54:34 -04001642static int pick_link(struct nameidata *nd, struct path *link,
1643 struct inode *inode, unsigned seq)
Al Virod63ff282015-05-04 18:13:23 -04001644{
Al Viro626de992015-05-04 18:26:59 -04001645 int error;
Al Viro1cf26652015-05-06 16:01:56 -04001646 struct saved *last;
NeilBrown756daf22015-03-23 13:37:38 +11001647 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
Al Viro626de992015-05-04 18:26:59 -04001648 path_to_nameidata(link, nd);
1649 return -ELOOP;
1650 }
Al Virobc40aee2015-05-09 13:04:24 -04001651 if (!(nd->flags & LOOKUP_RCU)) {
Al Viro79733872015-05-09 12:55:43 -04001652 if (link->mnt == nd->path.mnt)
1653 mntget(link->mnt);
Al Virod63ff282015-05-04 18:13:23 -04001654 }
Al Viro626de992015-05-04 18:26:59 -04001655 error = nd_alloc_stack(nd);
1656 if (unlikely(error)) {
Al Virobc40aee2015-05-09 13:04:24 -04001657 if (error == -ECHILD) {
1658 if (unlikely(unlazy_link(nd, link, seq)))
1659 return -ECHILD;
1660 error = nd_alloc_stack(nd);
1661 }
1662 if (error) {
1663 path_put(link);
1664 return error;
1665 }
Al Viro626de992015-05-04 18:26:59 -04001666 }
1667
Al Viroab104922015-05-10 11:50:01 -04001668 last = nd->stack + nd->depth++;
Al Viro1cf26652015-05-06 16:01:56 -04001669 last->link = *link;
Al Viroab104922015-05-10 11:50:01 -04001670 last->cookie = NULL;
Al Viro181548c2015-05-07 19:54:34 -04001671 last->inode = inode;
Al Viro0450b2d2015-05-08 13:23:53 -04001672 last->seq = seq;
Al Virod63ff282015-05-04 18:13:23 -04001673 return 1;
1674}
1675
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001676/*
1677 * Do we need to follow links? We _really_ want to be able
1678 * to do this check without having to look at inode->i_op,
1679 * so we keep a cache of "no, this doesn't need follow_link"
1680 * for the common case.
1681 */
Al Viro254cf582015-05-05 09:40:46 -04001682static inline int should_follow_link(struct nameidata *nd, struct path *link,
Al Viro181548c2015-05-07 19:54:34 -04001683 int follow,
1684 struct inode *inode, unsigned seq)
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001685{
Al Virod63ff282015-05-04 18:13:23 -04001686 if (likely(!d_is_symlink(link->dentry)))
1687 return 0;
1688 if (!follow)
1689 return 0;
Al Viro181548c2015-05-07 19:54:34 -04001690 return pick_link(nd, link, inode, seq);
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001691}
1692
Al Viro4693a542015-05-04 17:47:11 -04001693enum {WALK_GET = 1, WALK_PUT = 2};
1694
1695static int walk_component(struct nameidata *nd, int flags)
Al Viroce57dfc2011-03-13 19:58:58 -04001696{
Al Virocaa85632015-04-22 17:52:47 -04001697 struct path path;
Al Viroce57dfc2011-03-13 19:58:58 -04001698 struct inode *inode;
Al Viro254cf582015-05-05 09:40:46 -04001699 unsigned seq;
Al Viroce57dfc2011-03-13 19:58:58 -04001700 int err;
1701 /*
1702 * "." and ".." are special - ".." especially so because it has
1703 * to be able to know about the current root directory and
1704 * parent relationships.
1705 */
Al Viro4693a542015-05-04 17:47:11 -04001706 if (unlikely(nd->last_type != LAST_NORM)) {
1707 err = handle_dots(nd, nd->last_type);
1708 if (flags & WALK_PUT)
1709 put_link(nd);
1710 return err;
1711 }
Al Viro254cf582015-05-05 09:40:46 -04001712 err = lookup_fast(nd, &path, &inode, &seq);
Al Viroce57dfc2011-03-13 19:58:58 -04001713 if (unlikely(err)) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001714 if (err < 0)
Al Virof0a9ba72015-05-04 07:59:30 -04001715 return err;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001716
Al Virocaa85632015-04-22 17:52:47 -04001717 err = lookup_slow(nd, &path);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001718 if (err < 0)
Al Virof0a9ba72015-05-04 07:59:30 -04001719 return err;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001720
David Howells63afdfc2015-05-06 15:59:00 +01001721 inode = d_backing_inode(path.dentry);
Al Viro254cf582015-05-05 09:40:46 -04001722 seq = 0; /* we are already out of RCU mode */
Al Viro766c4cb2015-05-07 19:24:57 -04001723 err = -ENOENT;
Al Virocaa85632015-04-22 17:52:47 -04001724 if (d_is_negative(path.dentry))
Al Viro766c4cb2015-05-07 19:24:57 -04001725 goto out_path_put;
Al Viroce57dfc2011-03-13 19:58:58 -04001726 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001727
Al Viro4693a542015-05-04 17:47:11 -04001728 if (flags & WALK_PUT)
1729 put_link(nd);
Al Viro181548c2015-05-07 19:54:34 -04001730 err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq);
Al Virod63ff282015-05-04 18:13:23 -04001731 if (unlikely(err))
1732 return err;
Al Virocaa85632015-04-22 17:52:47 -04001733 path_to_nameidata(&path, nd);
Al Viroce57dfc2011-03-13 19:58:58 -04001734 nd->inode = inode;
Al Viro254cf582015-05-05 09:40:46 -04001735 nd->seq = seq;
Al Viroce57dfc2011-03-13 19:58:58 -04001736 return 0;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001737
1738out_path_put:
Al Virocaa85632015-04-22 17:52:47 -04001739 path_to_nameidata(&path, nd);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001740 return err;
Al Viroce57dfc2011-03-13 19:58:58 -04001741}
1742
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743/*
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001744 * We can do the critical dentry name comparison and hashing
1745 * operations one word at a time, but we are limited to:
1746 *
1747 * - Architectures with fast unaligned word accesses. We could
1748 * do a "get_unaligned()" if this helps and is sufficiently
1749 * fast.
1750 *
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001751 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1752 * do not trap on the (extremely unlikely) case of a page
1753 * crossing operation.
1754 *
1755 * - Furthermore, we need an efficient 64-bit compile for the
1756 * 64-bit case in order to generate the "number of bytes in
1757 * the final mask". Again, that could be replaced with a
1758 * efficient population count instruction or similar.
1759 */
1760#ifdef CONFIG_DCACHE_WORD_ACCESS
1761
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001762#include <asm/word-at-a-time.h>
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001763
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001764#ifdef CONFIG_64BIT
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001765
1766static inline unsigned int fold_hash(unsigned long hash)
1767{
Linus Torvalds99d263d2014-09-13 11:30:10 -07001768 return hash_64(hash, 32);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001769}
1770
1771#else /* 32-bit case */
1772
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001773#define fold_hash(x) (x)
1774
1775#endif
1776
1777unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1778{
1779 unsigned long a, mask;
1780 unsigned long hash = 0;
1781
1782 for (;;) {
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001783 a = load_unaligned_zeropad(name);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001784 if (len < sizeof(unsigned long))
1785 break;
1786 hash += a;
Al Virof132c5b2012-03-22 21:59:52 +00001787 hash *= 9;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001788 name += sizeof(unsigned long);
1789 len -= sizeof(unsigned long);
1790 if (!len)
1791 goto done;
1792 }
Will Deacona5c21dc2013-12-12 17:40:21 +00001793 mask = bytemask_from_count(len);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001794 hash += mask & a;
1795done:
1796 return fold_hash(hash);
1797}
1798EXPORT_SYMBOL(full_name_hash);
1799
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001800/*
1801 * Calculate the length and hash of the path component, and
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001802 * return the "hash_len" as the result.
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001803 */
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001804static inline u64 hash_name(const char *name)
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001805{
Linus Torvalds36126f82012-05-26 10:43:17 -07001806 unsigned long a, b, adata, bdata, mask, hash, len;
1807 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001808
1809 hash = a = 0;
1810 len = -sizeof(unsigned long);
1811 do {
1812 hash = (hash + a) * 9;
1813 len += sizeof(unsigned long);
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001814 a = load_unaligned_zeropad(name+len);
Linus Torvalds36126f82012-05-26 10:43:17 -07001815 b = a ^ REPEAT_BYTE('/');
1816 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001817
Linus Torvalds36126f82012-05-26 10:43:17 -07001818 adata = prep_zero_mask(a, adata, &constants);
1819 bdata = prep_zero_mask(b, bdata, &constants);
1820
1821 mask = create_zero_mask(adata | bdata);
1822
1823 hash += a & zero_bytemask(mask);
Linus Torvalds9226b5b2014-09-14 17:28:32 -07001824 len += find_zero(mask);
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001825 return hashlen_create(fold_hash(hash), len);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001826}
1827
1828#else
1829
Linus Torvalds0145acc2012-03-02 14:32:59 -08001830unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1831{
1832 unsigned long hash = init_name_hash();
1833 while (len--)
1834 hash = partial_name_hash(*name++, hash);
1835 return end_name_hash(hash);
1836}
Linus Torvaldsae942ae2012-03-02 19:40:57 -08001837EXPORT_SYMBOL(full_name_hash);
Linus Torvalds0145acc2012-03-02 14:32:59 -08001838
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001839/*
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001840 * We know there's a real path component here of at least
1841 * one character.
1842 */
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001843static inline u64 hash_name(const char *name)
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001844{
1845 unsigned long hash = init_name_hash();
1846 unsigned long len = 0, c;
1847
1848 c = (unsigned char)*name;
1849 do {
1850 len++;
1851 hash = partial_name_hash(c, hash);
1852 c = (unsigned char)name[len];
1853 } while (c && c != '/');
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001854 return hashlen_create(end_name_hash(hash), len);
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001855}
1856
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001857#endif
1858
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001859/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01001861 * This is the basic name resolution function, turning a pathname into
1862 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01001864 * Returns 0 and nd will have valid dentry and mnt on success.
1865 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 */
Al Viro6de88d72009-08-09 01:41:57 +04001867static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869 int err;
Al Viro32cd7462015-04-18 20:30:49 -04001870
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 while (*name=='/')
1872 name++;
1873 if (!*name)
Al Viro9e18f102015-04-18 20:44:34 -04001874 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876 /* At this point we know we have a real path component. */
1877 for(;;) {
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001878 u64 hash_len;
Al Virofe479a52011-02-22 15:10:03 -05001879 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880
Al Viro52094c82011-02-21 21:34:47 -05001881 err = may_lookup(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 if (err)
Al Viro3595e232015-05-09 16:54:45 -04001883 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001885 hash_len = hash_name(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886
Al Virofe479a52011-02-22 15:10:03 -05001887 type = LAST_NORM;
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001888 if (name[0] == '.') switch (hashlen_len(hash_len)) {
Al Virofe479a52011-02-22 15:10:03 -05001889 case 2:
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001890 if (name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05001891 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05001892 nd->flags |= LOOKUP_JUMPED;
1893 }
Al Virofe479a52011-02-22 15:10:03 -05001894 break;
1895 case 1:
1896 type = LAST_DOT;
1897 }
Al Viro5a202bc2011-03-08 14:17:44 -05001898 if (likely(type == LAST_NORM)) {
1899 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05001900 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05001901 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
James Hogana060dc52014-09-16 13:07:35 +01001902 struct qstr this = { { .hash_len = hash_len }, .name = name };
Linus Torvaldsda53be12013-05-21 15:22:44 -07001903 err = parent->d_op->d_hash(parent, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05001904 if (err < 0)
Al Viro3595e232015-05-09 16:54:45 -04001905 return err;
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001906 hash_len = this.hash_len;
1907 name = this.name;
Al Viro5a202bc2011-03-08 14:17:44 -05001908 }
1909 }
Al Virofe479a52011-02-22 15:10:03 -05001910
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001911 nd->last.hash_len = hash_len;
1912 nd->last.name = name;
Al Viro5f4a6a62013-01-24 18:04:22 -05001913 nd->last_type = type;
1914
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001915 name += hashlen_len(hash_len);
1916 if (!*name)
Al Virobdf6cbf2015-04-18 20:21:40 -04001917 goto OK;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001918 /*
1919 * If it wasn't NUL, we know it was '/'. Skip that
1920 * slash, and continue until no more slashes.
1921 */
1922 do {
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001923 name++;
1924 } while (unlikely(*name == '/'));
Al Viro8620c232015-05-04 08:58:35 -04001925 if (unlikely(!*name)) {
1926OK:
Al Viro368ee9b2015-05-08 17:19:59 -04001927 /* pathname body, done */
Al Viro8620c232015-05-04 08:58:35 -04001928 if (!nd->depth)
1929 return 0;
1930 name = nd->stack[nd->depth - 1].name;
Al Viro368ee9b2015-05-08 17:19:59 -04001931 /* trailing symlink, done */
Al Viro8620c232015-05-04 08:58:35 -04001932 if (!name)
1933 return 0;
1934 /* last component of nested symlink */
Al Viro4693a542015-05-04 17:47:11 -04001935 err = walk_component(nd, WALK_GET | WALK_PUT);
Al Viro8620c232015-05-04 08:58:35 -04001936 } else {
Al Viro4693a542015-05-04 17:47:11 -04001937 err = walk_component(nd, WALK_GET);
Al Viro8620c232015-05-04 08:58:35 -04001938 }
Al Viroce57dfc2011-03-13 19:58:58 -04001939 if (err < 0)
Al Viro3595e232015-05-09 16:54:45 -04001940 return err;
Al Virofe479a52011-02-22 15:10:03 -05001941
Al Viroce57dfc2011-03-13 19:58:58 -04001942 if (err) {
Al Viro626de992015-05-04 18:26:59 -04001943 const char *s = get_link(nd);
Al Viro5a460272015-04-17 23:44:45 -04001944
Al Viro3595e232015-05-09 16:54:45 -04001945 if (unlikely(IS_ERR(s)))
1946 return PTR_ERR(s);
Al Virod40bcc02015-04-18 20:03:03 -04001947 err = 0;
1948 if (unlikely(!s)) {
1949 /* jumped */
Al Virob9ff4422015-05-02 20:19:23 -04001950 put_link(nd);
Al Virod40bcc02015-04-18 20:03:03 -04001951 } else {
Al Virofab51e82015-05-10 11:01:00 -04001952 nd->stack[nd->depth - 1].name = name;
1953 name = s;
1954 continue;
Al Virod40bcc02015-04-18 20:03:03 -04001955 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001956 }
Al Viro97242f92015-08-01 19:59:28 -04001957 if (unlikely(!d_can_lookup(nd->path.dentry))) {
1958 if (nd->flags & LOOKUP_RCU) {
1959 if (unlazy_walk(nd, NULL, 0))
1960 return -ECHILD;
1961 }
Al Viro3595e232015-05-09 16:54:45 -04001962 return -ENOTDIR;
Al Viro97242f92015-08-01 19:59:28 -04001963 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965}
1966
Al Viroc8a53ee2015-05-12 18:43:07 -04001967static const char *path_init(struct nameidata *nd, unsigned flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001969 int retval = 0;
Al Viroc8a53ee2015-05-12 18:43:07 -04001970 const char *s = nd->name->name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971
1972 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro980f3ea2014-11-20 14:20:24 -05001973 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 nd->depth = 0;
Al Viro368ee9b2015-05-08 17:19:59 -04001975 nd->total_link_count = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05001976 if (flags & LOOKUP_ROOT) {
David Howellsb18825a2013-09-12 19:22:53 +01001977 struct dentry *root = nd->root.dentry;
1978 struct inode *inode = root->d_inode;
Al Virofd2f7cb2015-02-22 20:07:13 -05001979 if (*s) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02001980 if (!d_can_lookup(root))
Al Viro368ee9b2015-05-08 17:19:59 -04001981 return ERR_PTR(-ENOTDIR);
Al Viro73d049a2011-03-11 12:08:24 -05001982 retval = inode_permission(inode, MAY_EXEC);
1983 if (retval)
Al Viro368ee9b2015-05-08 17:19:59 -04001984 return ERR_PTR(retval);
Al Viro73d049a2011-03-11 12:08:24 -05001985 }
Al Viro5b6ca022011-03-09 23:04:47 -05001986 nd->path = nd->root;
1987 nd->inode = inode;
1988 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05001989 rcu_read_lock();
Al Viro5b6ca022011-03-09 23:04:47 -05001990 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro8f47a0162015-05-09 19:02:01 -04001991 nd->root_seq = nd->seq;
Al Viro48a066e2013-09-29 22:06:07 -04001992 nd->m_seq = read_seqbegin(&mount_lock);
Al Viro5b6ca022011-03-09 23:04:47 -05001993 } else {
1994 path_get(&nd->path);
1995 }
Al Viro368ee9b2015-05-08 17:19:59 -04001996 return s;
Al Viro5b6ca022011-03-09 23:04:47 -05001997 }
1998
Al Viro2a737872009-04-07 11:49:53 -04001999 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000
Al Viro48a066e2013-09-29 22:06:07 -04002001 nd->m_seq = read_seqbegin(&mount_lock);
Al Virofd2f7cb2015-02-22 20:07:13 -05002002 if (*s == '/') {
Al Viroe41f7d42011-02-22 14:02:58 -05002003 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05002004 rcu_read_lock();
Al Viro06d71372015-06-29 12:07:04 -04002005 set_root_rcu(nd);
2006 nd->seq = nd->root_seq;
Al Viroe41f7d42011-02-22 14:02:58 -05002007 } else {
2008 set_root(nd);
2009 path_get(&nd->root);
2010 }
Al Viro2a737872009-04-07 11:49:53 -04002011 nd->path = nd->root;
Al Viroc8a53ee2015-05-12 18:43:07 -04002012 } else if (nd->dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05002013 if (flags & LOOKUP_RCU) {
2014 struct fs_struct *fs = current->fs;
2015 unsigned seq;
2016
Al Viro8b61e742013-11-08 12:45:01 -05002017 rcu_read_lock();
Al Viroe41f7d42011-02-22 14:02:58 -05002018
2019 do {
2020 seq = read_seqcount_begin(&fs->seq);
2021 nd->path = fs->pwd;
2022 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2023 } while (read_seqcount_retry(&fs->seq, seq));
2024 } else {
2025 get_fs_pwd(current->fs, &nd->path);
2026 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002027 } else {
Jeff Layton582aa642012-12-11 08:56:16 -05002028 /* Caller must check execute permissions on the starting path component */
Al Viroc8a53ee2015-05-12 18:43:07 -04002029 struct fd f = fdget_raw(nd->dfd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002030 struct dentry *dentry;
2031
Al Viro2903ff02012-08-28 12:52:22 -04002032 if (!f.file)
Al Viro368ee9b2015-05-08 17:19:59 -04002033 return ERR_PTR(-EBADF);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002034
Al Viro2903ff02012-08-28 12:52:22 -04002035 dentry = f.file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002036
Al Virofd2f7cb2015-02-22 20:07:13 -05002037 if (*s) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002038 if (!d_can_lookup(dentry)) {
Al Viro2903ff02012-08-28 12:52:22 -04002039 fdput(f);
Al Viro368ee9b2015-05-08 17:19:59 -04002040 return ERR_PTR(-ENOTDIR);
Al Viro2903ff02012-08-28 12:52:22 -04002041 }
Al Virof52e0c12011-03-14 18:56:51 -04002042 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002043
Al Viro2903ff02012-08-28 12:52:22 -04002044 nd->path = f.file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05002045 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05002046 rcu_read_lock();
Al Viro34a26b92015-05-11 08:05:05 -04002047 nd->inode = nd->path.dentry->d_inode;
2048 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viroe41f7d42011-02-22 14:02:58 -05002049 } else {
Al Viro2903ff02012-08-28 12:52:22 -04002050 path_get(&nd->path);
Al Viro34a26b92015-05-11 08:05:05 -04002051 nd->inode = nd->path.dentry->d_inode;
Al Viroe41f7d42011-02-22 14:02:58 -05002052 }
Al Viro34a26b92015-05-11 08:05:05 -04002053 fdput(f);
Al Viro368ee9b2015-05-08 17:19:59 -04002054 return s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 }
Al Viroe41f7d42011-02-22 14:02:58 -05002056
Nick Piggin31e6b012011-01-07 17:49:52 +11002057 nd->inode = nd->path.dentry->d_inode;
Al Viro4023bfc2014-09-13 21:59:43 -04002058 if (!(flags & LOOKUP_RCU))
Al Viro368ee9b2015-05-08 17:19:59 -04002059 return s;
Al Viro4023bfc2014-09-13 21:59:43 -04002060 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
Al Viro368ee9b2015-05-08 17:19:59 -04002061 return s;
Al Viro4023bfc2014-09-13 21:59:43 -04002062 if (!(nd->flags & LOOKUP_ROOT))
2063 nd->root.mnt = NULL;
2064 rcu_read_unlock();
Al Viro368ee9b2015-05-08 17:19:59 -04002065 return ERR_PTR(-ECHILD);
Al Viro9b4a9b12009-04-07 11:44:16 -04002066}
2067
Al Viro3bdba282015-05-08 17:37:07 -04002068static const char *trailing_symlink(struct nameidata *nd)
Al Viro95fa25d2015-04-22 13:46:57 -04002069{
2070 const char *s;
Al Virofec2fa22015-05-06 15:58:18 -04002071 int error = may_follow_link(nd);
Al Virodeb106c2015-05-08 18:05:21 -04002072 if (unlikely(error))
Al Viro3bdba282015-05-08 17:37:07 -04002073 return ERR_PTR(error);
Al Viro95fa25d2015-04-22 13:46:57 -04002074 nd->flags |= LOOKUP_PARENT;
Al Virofab51e82015-05-10 11:01:00 -04002075 nd->stack[0].name = NULL;
Al Viro3b2e7f72015-04-19 00:53:50 -04002076 s = get_link(nd);
Al Virodeb106c2015-05-08 18:05:21 -04002077 return s ? s : "";
Al Viro95fa25d2015-04-22 13:46:57 -04002078}
2079
Al Virocaa85632015-04-22 17:52:47 -04002080static inline int lookup_last(struct nameidata *nd)
Al Virobd92d7f2011-03-14 19:54:59 -04002081{
2082 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2083 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2084
2085 nd->flags &= ~LOOKUP_PARENT;
Al Virodeb106c2015-05-08 18:05:21 -04002086 return walk_component(nd,
Al Viro4693a542015-05-04 17:47:11 -04002087 nd->flags & LOOKUP_FOLLOW
2088 ? nd->depth
2089 ? WALK_PUT | WALK_GET
2090 : WALK_GET
2091 : 0);
Al Virobd92d7f2011-03-14 19:54:59 -04002092}
2093
Al Viro9b4a9b12009-04-07 11:44:16 -04002094/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroc8a53ee2015-05-12 18:43:07 -04002095static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
Al Viro9b4a9b12009-04-07 11:44:16 -04002096{
Al Viroc8a53ee2015-05-12 18:43:07 -04002097 const char *s = path_init(nd, flags);
Al Virobd92d7f2011-03-14 19:54:59 -04002098 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11002099
Al Viro368ee9b2015-05-08 17:19:59 -04002100 if (IS_ERR(s))
2101 return PTR_ERR(s);
Al Viro3bdba282015-05-08 17:37:07 -04002102 while (!(err = link_path_walk(s, nd))
2103 && ((err = lookup_last(nd)) > 0)) {
2104 s = trailing_symlink(nd);
2105 if (IS_ERR(s)) {
2106 err = PTR_ERR(s);
2107 break;
Al Virobd92d7f2011-03-14 19:54:59 -04002108 }
2109 }
Al Viro9f1fafe2011-03-25 11:00:12 -04002110 if (!err)
2111 err = complete_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04002112
Al Virodeb106c2015-05-08 18:05:21 -04002113 if (!err && nd->flags & LOOKUP_DIRECTORY)
2114 if (!d_can_lookup(nd->path.dentry))
Al Virobd23a532011-03-23 09:56:30 -04002115 err = -ENOTDIR;
Al Viro625b6d12015-05-12 16:36:12 -04002116 if (!err) {
2117 *path = nd->path;
2118 nd->path.mnt = NULL;
2119 nd->path.dentry = NULL;
2120 }
2121 terminate_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04002122 return err;
Al Viroee0827c2011-02-21 23:38:09 -05002123}
Nick Piggin31e6b012011-01-07 17:49:52 +11002124
Al Viro625b6d12015-05-12 16:36:12 -04002125static int filename_lookup(int dfd, struct filename *name, unsigned flags,
Al Viro9ad1aaa2015-05-12 16:44:39 -04002126 struct path *path, struct path *root)
Jeff Layton873f1ee2012-10-10 15:25:29 -04002127{
Al Viro894bc8c2015-05-02 07:16:16 -04002128 int retval;
Al Viro9883d182015-05-13 07:28:08 -04002129 struct nameidata nd;
Al Viroabc9f5b2015-05-12 16:53:42 -04002130 if (IS_ERR(name))
2131 return PTR_ERR(name);
Al Viro9ad1aaa2015-05-12 16:44:39 -04002132 if (unlikely(root)) {
2133 nd.root = *root;
2134 flags |= LOOKUP_ROOT;
2135 }
Al Viro9883d182015-05-13 07:28:08 -04002136 set_nameidata(&nd, dfd, name);
Al Viroc8a53ee2015-05-12 18:43:07 -04002137 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002138 if (unlikely(retval == -ECHILD))
Al Viroc8a53ee2015-05-12 18:43:07 -04002139 retval = path_lookupat(&nd, flags, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002140 if (unlikely(retval == -ESTALE))
Al Viroc8a53ee2015-05-12 18:43:07 -04002141 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002142
2143 if (likely(!retval))
Al Viro625b6d12015-05-12 16:36:12 -04002144 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
Al Viro9883d182015-05-13 07:28:08 -04002145 restore_nameidata();
Al Viroe4bd1c12015-05-12 16:40:39 -04002146 putname(name);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002147 return retval;
2148}
2149
Al Viro8bcb77f2015-05-08 16:59:20 -04002150/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroc8a53ee2015-05-12 18:43:07 -04002151static int path_parentat(struct nameidata *nd, unsigned flags,
Al Viro391172c2015-05-09 11:19:16 -04002152 struct path *parent)
Al Viro8bcb77f2015-05-08 16:59:20 -04002153{
Al Viroc8a53ee2015-05-12 18:43:07 -04002154 const char *s = path_init(nd, flags);
Al Viro368ee9b2015-05-08 17:19:59 -04002155 int err;
2156 if (IS_ERR(s))
2157 return PTR_ERR(s);
2158 err = link_path_walk(s, nd);
Al Viro8bcb77f2015-05-08 16:59:20 -04002159 if (!err)
2160 err = complete_walk(nd);
Al Viro391172c2015-05-09 11:19:16 -04002161 if (!err) {
2162 *parent = nd->path;
2163 nd->path.mnt = NULL;
2164 nd->path.dentry = NULL;
2165 }
2166 terminate_walk(nd);
Al Viro8bcb77f2015-05-08 16:59:20 -04002167 return err;
2168}
2169
Al Viro5c31b6c2015-05-12 17:32:54 -04002170static struct filename *filename_parentat(int dfd, struct filename *name,
Al Viro391172c2015-05-09 11:19:16 -04002171 unsigned int flags, struct path *parent,
2172 struct qstr *last, int *type)
Al Viro8bcb77f2015-05-08 16:59:20 -04002173{
2174 int retval;
Al Viro9883d182015-05-13 07:28:08 -04002175 struct nameidata nd;
Al Viro8bcb77f2015-05-08 16:59:20 -04002176
Al Viro5c31b6c2015-05-12 17:32:54 -04002177 if (IS_ERR(name))
2178 return name;
Al Viro9883d182015-05-13 07:28:08 -04002179 set_nameidata(&nd, dfd, name);
Al Viroc8a53ee2015-05-12 18:43:07 -04002180 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
Al Viro8bcb77f2015-05-08 16:59:20 -04002181 if (unlikely(retval == -ECHILD))
Al Viroc8a53ee2015-05-12 18:43:07 -04002182 retval = path_parentat(&nd, flags, parent);
Al Viro8bcb77f2015-05-08 16:59:20 -04002183 if (unlikely(retval == -ESTALE))
Al Viroc8a53ee2015-05-12 18:43:07 -04002184 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
Al Viro391172c2015-05-09 11:19:16 -04002185 if (likely(!retval)) {
2186 *last = nd.last;
2187 *type = nd.last_type;
2188 audit_inode(name, parent->dentry, LOOKUP_PARENT);
Al Viro5c31b6c2015-05-12 17:32:54 -04002189 } else {
2190 putname(name);
2191 name = ERR_PTR(retval);
Al Viro391172c2015-05-09 11:19:16 -04002192 }
Al Viro9883d182015-05-13 07:28:08 -04002193 restore_nameidata();
Al Viro5c31b6c2015-05-12 17:32:54 -04002194 return name;
Al Viro8bcb77f2015-05-08 16:59:20 -04002195}
2196
Al Viro79714f72012-06-15 03:01:42 +04002197/* does lookup, returns the object with parent locked */
2198struct dentry *kern_path_locked(const char *name, struct path *path)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002199{
Al Viro5c31b6c2015-05-12 17:32:54 -04002200 struct filename *filename;
2201 struct dentry *d;
Al Viro391172c2015-05-09 11:19:16 -04002202 struct qstr last;
2203 int type;
Paul Moore51689102015-01-22 00:00:03 -05002204
Al Viro5c31b6c2015-05-12 17:32:54 -04002205 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2206 &last, &type);
Paul Moore51689102015-01-22 00:00:03 -05002207 if (IS_ERR(filename))
2208 return ERR_CAST(filename);
Al Viro5c31b6c2015-05-12 17:32:54 -04002209 if (unlikely(type != LAST_NORM)) {
Al Viro391172c2015-05-09 11:19:16 -04002210 path_put(path);
Al Viro5c31b6c2015-05-12 17:32:54 -04002211 putname(filename);
2212 return ERR_PTR(-EINVAL);
Al Viro79714f72012-06-15 03:01:42 +04002213 }
Al Viro391172c2015-05-09 11:19:16 -04002214 mutex_lock_nested(&path->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2215 d = __lookup_hash(&last, path->dentry, 0);
Al Viro79714f72012-06-15 03:01:42 +04002216 if (IS_ERR(d)) {
Al Viro391172c2015-05-09 11:19:16 -04002217 mutex_unlock(&path->dentry->d_inode->i_mutex);
2218 path_put(path);
Al Viro79714f72012-06-15 03:01:42 +04002219 }
Paul Moore51689102015-01-22 00:00:03 -05002220 putname(filename);
Al Viro79714f72012-06-15 03:01:42 +04002221 return d;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002222}
2223
Al Virod1811462008-08-02 00:49:18 -04002224int kern_path(const char *name, unsigned int flags, struct path *path)
2225{
Al Viroabc9f5b2015-05-12 16:53:42 -04002226 return filename_lookup(AT_FDCWD, getname_kernel(name),
2227 flags, path, NULL);
Al Virod1811462008-08-02 00:49:18 -04002228}
Al Viro4d359502014-03-14 12:20:17 -04002229EXPORT_SYMBOL(kern_path);
Al Virod1811462008-08-02 00:49:18 -04002230
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002231/**
2232 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2233 * @dentry: pointer to dentry of the base directory
2234 * @mnt: pointer to vfs mount of the base directory
2235 * @name: pointer to file name
2236 * @flags: lookup flags
Al Viroe0a01242011-06-27 17:00:37 -04002237 * @path: pointer to struct path to fill
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002238 */
2239int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2240 const char *name, unsigned int flags,
Al Viroe0a01242011-06-27 17:00:37 -04002241 struct path *path)
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002242{
Al Viro9ad1aaa2015-05-12 16:44:39 -04002243 struct path root = {.mnt = mnt, .dentry = dentry};
Al Viro9ad1aaa2015-05-12 16:44:39 -04002244 /* the first argument of filename_lookup() is ignored with root */
Al Viroabc9f5b2015-05-12 16:53:42 -04002245 return filename_lookup(AT_FDCWD, getname_kernel(name),
2246 flags , path, &root);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002247}
Al Viro4d359502014-03-14 12:20:17 -04002248EXPORT_SYMBOL(vfs_path_lookup);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002249
Christoph Hellwigeead1912007-10-16 23:25:38 -07002250/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07002251 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07002252 * @name: pathname component to lookup
2253 * @base: base directory to lookup from
2254 * @len: maximum length @len should be interpreted to
2255 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07002256 * Note that this routine is purely a helper for filesystem usage and should
Al Viro9e7543e2015-02-23 02:49:48 -05002257 * not be called by generic code.
Christoph Hellwigeead1912007-10-16 23:25:38 -07002258 */
James Morris057f6c02007-04-26 00:12:05 -07002259struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2260{
James Morris057f6c02007-04-26 00:12:05 -07002261 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05002262 unsigned int c;
Miklos Szeredicda309d2012-03-26 12:54:21 +02002263 int err;
James Morris057f6c02007-04-26 00:12:05 -07002264
David Woodhouse2f9092e2009-04-20 23:18:37 +01002265 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2266
Al Viro6a96ba52011-03-07 23:49:20 -05002267 this.name = name;
2268 this.len = len;
Linus Torvalds0145acc2012-03-02 14:32:59 -08002269 this.hash = full_name_hash(name, len);
Al Viro6a96ba52011-03-07 23:49:20 -05002270 if (!len)
2271 return ERR_PTR(-EACCES);
2272
Al Viro21d8a152012-11-29 22:17:21 -05002273 if (unlikely(name[0] == '.')) {
2274 if (len < 2 || (len == 2 && name[1] == '.'))
2275 return ERR_PTR(-EACCES);
2276 }
2277
Al Viro6a96ba52011-03-07 23:49:20 -05002278 while (len--) {
2279 c = *(const unsigned char *)name++;
2280 if (c == '/' || c == '\0')
2281 return ERR_PTR(-EACCES);
Al Viro6a96ba52011-03-07 23:49:20 -05002282 }
Al Viro5a202bc2011-03-08 14:17:44 -05002283 /*
2284 * See if the low-level filesystem might want
2285 * to use its own hash..
2286 */
2287 if (base->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002288 int err = base->d_op->d_hash(base, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05002289 if (err < 0)
2290 return ERR_PTR(err);
2291 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07002292
Miklos Szeredicda309d2012-03-26 12:54:21 +02002293 err = inode_permission(base->d_inode, MAY_EXEC);
2294 if (err)
2295 return ERR_PTR(err);
2296
Al Viro72bd8662012-06-10 17:17:17 -04002297 return __lookup_hash(&this, base, 0);
James Morris057f6c02007-04-26 00:12:05 -07002298}
Al Viro4d359502014-03-14 12:20:17 -04002299EXPORT_SYMBOL(lookup_one_len);
James Morris057f6c02007-04-26 00:12:05 -07002300
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002301int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2302 struct path *path, int *empty)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303{
Al Viroabc9f5b2015-05-12 16:53:42 -04002304 return filename_lookup(dfd, getname_flags(name, flags, empty),
2305 flags, path, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306}
Al Virob853a162015-05-13 09:12:02 -04002307EXPORT_SYMBOL(user_path_at_empty);
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002308
Jeff Layton873f1ee2012-10-10 15:25:29 -04002309/*
2310 * NB: most callers don't do anything directly with the reference to the
2311 * to struct filename, but the nd->last pointer points into the name string
2312 * allocated by getname. So we must hold the reference to it until all
2313 * path-walking is complete.
2314 */
Al Viroa2ec4a22015-05-13 06:57:49 -04002315static inline struct filename *
Al Virof5beed72015-04-30 16:09:11 -04002316user_path_parent(int dfd, const char __user *path,
2317 struct path *parent,
2318 struct qstr *last,
2319 int *type,
Jeff Layton9e790bd2012-12-11 12:10:09 -05002320 unsigned int flags)
Al Viro2ad94ae2008-07-21 09:32:51 -04002321{
Jeff Layton9e790bd2012-12-11 12:10:09 -05002322 /* only LOOKUP_REVAL is allowed in extra flags */
Al Viro5c31b6c2015-05-12 17:32:54 -04002323 return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL,
2324 parent, last, type);
Al Viro2ad94ae2008-07-21 09:32:51 -04002325}
2326
Jeff Layton80334262013-07-26 06:23:25 -04002327/**
Al Viro197df042013-09-08 14:03:27 -04002328 * mountpoint_last - look up last component for umount
Jeff Layton80334262013-07-26 06:23:25 -04002329 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2330 * @path: pointer to container for result
2331 *
2332 * This is a special lookup_last function just for umount. In this case, we
2333 * need to resolve the path without doing any revalidation.
2334 *
2335 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2336 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2337 * in almost all cases, this lookup will be served out of the dcache. The only
2338 * cases where it won't are if nd->last refers to a symlink or the path is
2339 * bogus and it doesn't exist.
2340 *
2341 * Returns:
2342 * -error: if there was an error during lookup. This includes -ENOENT if the
2343 * lookup found a negative dentry. The nd->path reference will also be
2344 * put in this case.
2345 *
2346 * 0: if we successfully resolved nd->path and found it to not to be a
2347 * symlink that needs to be followed. "path" will also be populated.
2348 * The nd->path reference will also be put.
2349 *
2350 * 1: if we successfully resolved nd->last and found it to be a symlink
2351 * that needs to be followed. "path" will be populated with the path
2352 * to the link, and nd->path will *not* be put.
2353 */
2354static int
Al Viro197df042013-09-08 14:03:27 -04002355mountpoint_last(struct nameidata *nd, struct path *path)
Jeff Layton80334262013-07-26 06:23:25 -04002356{
2357 int error = 0;
2358 struct dentry *dentry;
2359 struct dentry *dir = nd->path.dentry;
2360
Al Viro35759522013-09-08 13:41:33 -04002361 /* If we're in rcuwalk, drop out of it to handle last component */
2362 if (nd->flags & LOOKUP_RCU) {
Al Viro6e9918b2015-05-05 09:26:05 -04002363 if (unlazy_walk(nd, NULL, 0))
Al Virodeb106c2015-05-08 18:05:21 -04002364 return -ECHILD;
Jeff Layton80334262013-07-26 06:23:25 -04002365 }
2366
2367 nd->flags &= ~LOOKUP_PARENT;
2368
2369 if (unlikely(nd->last_type != LAST_NORM)) {
2370 error = handle_dots(nd, nd->last_type);
Al Viro35759522013-09-08 13:41:33 -04002371 if (error)
Al Virodeb106c2015-05-08 18:05:21 -04002372 return error;
Al Viro35759522013-09-08 13:41:33 -04002373 dentry = dget(nd->path.dentry);
2374 goto done;
Jeff Layton80334262013-07-26 06:23:25 -04002375 }
2376
2377 mutex_lock(&dir->d_inode->i_mutex);
2378 dentry = d_lookup(dir, &nd->last);
2379 if (!dentry) {
2380 /*
2381 * No cached dentry. Mounted dentries are pinned in the cache,
2382 * so that means that this dentry is probably a symlink or the
2383 * path doesn't actually point to a mounted dentry.
2384 */
2385 dentry = d_alloc(dir, &nd->last);
2386 if (!dentry) {
Dave Jonesbcceeeb2013-09-10 17:04:25 -04002387 mutex_unlock(&dir->d_inode->i_mutex);
Al Virodeb106c2015-05-08 18:05:21 -04002388 return -ENOMEM;
Jeff Layton80334262013-07-26 06:23:25 -04002389 }
Al Viro35759522013-09-08 13:41:33 -04002390 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
Dave Jonesbcceeeb2013-09-10 17:04:25 -04002391 if (IS_ERR(dentry)) {
2392 mutex_unlock(&dir->d_inode->i_mutex);
Al Virodeb106c2015-05-08 18:05:21 -04002393 return PTR_ERR(dentry);
Dave Jonesbcceeeb2013-09-10 17:04:25 -04002394 }
Jeff Layton80334262013-07-26 06:23:25 -04002395 }
2396 mutex_unlock(&dir->d_inode->i_mutex);
2397
Al Viro35759522013-09-08 13:41:33 -04002398done:
David Howells698934d2015-03-17 17:33:52 +00002399 if (d_is_negative(dentry)) {
Al Viro35759522013-09-08 13:41:33 -04002400 dput(dentry);
Al Virodeb106c2015-05-08 18:05:21 -04002401 return -ENOENT;
Jeff Layton80334262013-07-26 06:23:25 -04002402 }
Al Viro191d7f72015-05-04 08:26:45 -04002403 if (nd->depth)
2404 put_link(nd);
Al Viro35759522013-09-08 13:41:33 -04002405 path->dentry = dentry;
Vasily Averin295dc392014-07-21 12:30:23 +04002406 path->mnt = nd->path.mnt;
Al Viro181548c2015-05-07 19:54:34 -04002407 error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW,
2408 d_backing_inode(dentry), 0);
Al Virodeb106c2015-05-08 18:05:21 -04002409 if (unlikely(error))
Al Virod63ff282015-05-04 18:13:23 -04002410 return error;
Vasily Averin295dc392014-07-21 12:30:23 +04002411 mntget(path->mnt);
Al Viro35759522013-09-08 13:41:33 -04002412 follow_mount(path);
Al Virodeb106c2015-05-08 18:05:21 -04002413 return 0;
Jeff Layton80334262013-07-26 06:23:25 -04002414}
2415
2416/**
Al Viro197df042013-09-08 14:03:27 -04002417 * path_mountpoint - look up a path to be umounted
Al Viroc8a53ee2015-05-12 18:43:07 -04002418 * @nameidata: lookup context
Jeff Layton80334262013-07-26 06:23:25 -04002419 * @flags: lookup flags
Al Viroc8a53ee2015-05-12 18:43:07 -04002420 * @path: pointer to container for result
Jeff Layton80334262013-07-26 06:23:25 -04002421 *
2422 * Look up the given name, but don't attempt to revalidate the last component.
Randy Dunlap606d6fe2013-10-19 14:56:55 -07002423 * Returns 0 and "path" will be valid on success; Returns error otherwise.
Jeff Layton80334262013-07-26 06:23:25 -04002424 */
2425static int
Al Viroc8a53ee2015-05-12 18:43:07 -04002426path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
Jeff Layton80334262013-07-26 06:23:25 -04002427{
Al Viroc8a53ee2015-05-12 18:43:07 -04002428 const char *s = path_init(nd, flags);
Al Viro368ee9b2015-05-08 17:19:59 -04002429 int err;
2430 if (IS_ERR(s))
2431 return PTR_ERR(s);
Al Viro3bdba282015-05-08 17:37:07 -04002432 while (!(err = link_path_walk(s, nd)) &&
2433 (err = mountpoint_last(nd, path)) > 0) {
2434 s = trailing_symlink(nd);
2435 if (IS_ERR(s)) {
2436 err = PTR_ERR(s);
Jeff Layton80334262013-07-26 06:23:25 -04002437 break;
Al Viro3bdba282015-05-08 17:37:07 -04002438 }
Jeff Layton80334262013-07-26 06:23:25 -04002439 }
Al Virodeb106c2015-05-08 18:05:21 -04002440 terminate_walk(nd);
Jeff Layton80334262013-07-26 06:23:25 -04002441 return err;
2442}
2443
Al Viro2d864652013-09-08 20:18:44 -04002444static int
Al Viro668696d2015-02-22 19:44:00 -05002445filename_mountpoint(int dfd, struct filename *name, struct path *path,
Al Viro2d864652013-09-08 20:18:44 -04002446 unsigned int flags)
2447{
Al Viro9883d182015-05-13 07:28:08 -04002448 struct nameidata nd;
Al Virocbaab2d2015-01-22 02:49:00 -05002449 int error;
Al Viro668696d2015-02-22 19:44:00 -05002450 if (IS_ERR(name))
2451 return PTR_ERR(name);
Al Viro9883d182015-05-13 07:28:08 -04002452 set_nameidata(&nd, dfd, name);
Al Viroc8a53ee2015-05-12 18:43:07 -04002453 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
Al Viro2d864652013-09-08 20:18:44 -04002454 if (unlikely(error == -ECHILD))
Al Viroc8a53ee2015-05-12 18:43:07 -04002455 error = path_mountpoint(&nd, flags, path);
Al Viro2d864652013-09-08 20:18:44 -04002456 if (unlikely(error == -ESTALE))
Al Viroc8a53ee2015-05-12 18:43:07 -04002457 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
Al Viro2d864652013-09-08 20:18:44 -04002458 if (likely(!error))
Al Viro668696d2015-02-22 19:44:00 -05002459 audit_inode(name, path->dentry, 0);
Al Viro9883d182015-05-13 07:28:08 -04002460 restore_nameidata();
Al Viro668696d2015-02-22 19:44:00 -05002461 putname(name);
Al Viro2d864652013-09-08 20:18:44 -04002462 return error;
2463}
2464
Jeff Layton80334262013-07-26 06:23:25 -04002465/**
Al Viro197df042013-09-08 14:03:27 -04002466 * user_path_mountpoint_at - lookup a path from userland in order to umount it
Jeff Layton80334262013-07-26 06:23:25 -04002467 * @dfd: directory file descriptor
2468 * @name: pathname from userland
2469 * @flags: lookup flags
2470 * @path: pointer to container to hold result
2471 *
2472 * A umount is a special case for path walking. We're not actually interested
2473 * in the inode in this situation, and ESTALE errors can be a problem. We
2474 * simply want track down the dentry and vfsmount attached at the mountpoint
2475 * and avoid revalidating the last component.
2476 *
2477 * Returns 0 and populates "path" on success.
2478 */
2479int
Al Viro197df042013-09-08 14:03:27 -04002480user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
Jeff Layton80334262013-07-26 06:23:25 -04002481 struct path *path)
2482{
Al Virocbaab2d2015-01-22 02:49:00 -05002483 return filename_mountpoint(dfd, getname(name), path, flags);
Jeff Layton80334262013-07-26 06:23:25 -04002484}
2485
Al Viro2d864652013-09-08 20:18:44 -04002486int
2487kern_path_mountpoint(int dfd, const char *name, struct path *path,
2488 unsigned int flags)
2489{
Al Virocbaab2d2015-01-22 02:49:00 -05002490 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
Al Viro2d864652013-09-08 20:18:44 -04002491}
2492EXPORT_SYMBOL(kern_path_mountpoint);
2493
Miklos Szeredicbdf35b2014-10-24 00:14:36 +02002494int __check_sticky(struct inode *dir, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495{
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002496 kuid_t fsuid = current_fsuid();
David Howellsda9592e2008-11-14 10:39:05 +11002497
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002498 if (uid_eq(inode->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002499 return 0;
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002500 if (uid_eq(dir->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002501 return 0;
Andy Lutomirski23adbe12014-06-10 12:45:42 -07002502 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503}
Miklos Szeredicbdf35b2014-10-24 00:14:36 +02002504EXPORT_SYMBOL(__check_sticky);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505
2506/*
2507 * Check whether we can remove a link victim from directory dir, check
2508 * whether the type of victim is right.
2509 * 1. We can't do it if dir is read-only (done in permission())
2510 * 2. We should have write and exec permissions on dir
2511 * 3. We can't remove anything from append-only dir
2512 * 4. We can't do anything with immutable dir (done in permission())
2513 * 5. If the sticky bit on dir is set we should either
2514 * a. be owner of dir, or
2515 * b. be owner of victim, or
2516 * c. have CAP_FOWNER capability
2517 * 6. If the victim is append-only or immutable we can't do antyhing with
2518 * links pointing to it.
2519 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2520 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2521 * 9. We can't remove a root or mountpoint.
2522 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2523 * nfs_async_unlink().
2524 */
David Howellsb18825a2013-09-12 19:22:53 +01002525static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526{
David Howells63afdfc2015-05-06 15:59:00 +01002527 struct inode *inode = d_backing_inode(victim);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528 int error;
2529
David Howellsb18825a2013-09-12 19:22:53 +01002530 if (d_is_negative(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531 return -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01002532 BUG_ON(!inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533
2534 BUG_ON(victim->d_parent->d_inode != dir);
Jeff Layton4fa6b5e2012-10-10 15:25:25 -04002535 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536
Al Virof419a2e2008-07-22 00:07:17 -04002537 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 if (error)
2539 return error;
2540 if (IS_APPEND(dir))
2541 return -EPERM;
David Howellsb18825a2013-09-12 19:22:53 +01002542
2543 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2544 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545 return -EPERM;
2546 if (isdir) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002547 if (!d_is_dir(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548 return -ENOTDIR;
2549 if (IS_ROOT(victim))
2550 return -EBUSY;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002551 } else if (d_is_dir(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552 return -EISDIR;
2553 if (IS_DEADDIR(dir))
2554 return -ENOENT;
2555 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2556 return -EBUSY;
2557 return 0;
2558}
2559
2560/* Check whether we can create an object with dentry child in directory
2561 * dir.
2562 * 1. We can't do it if child already exists (open has special treatment for
2563 * this case, but since we are inlined it's OK)
2564 * 2. We can't do it if dir is read-only (done in permission())
2565 * 3. We should have write and exec permissions on dir
2566 * 4. We can't do it if dir is immutable (done in permission())
2567 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02002568static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569{
Jeff Layton14e972b2013-05-08 10:25:58 -04002570 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571 if (child->d_inode)
2572 return -EEXIST;
2573 if (IS_DEADDIR(dir))
2574 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04002575 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002576}
2577
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578/*
2579 * p1 and p2 should be directories on the same fs.
2580 */
2581struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2582{
2583 struct dentry *p;
2584
2585 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07002586 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587 return NULL;
2588 }
2589
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002590 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002592 p = d_ancestor(p2, p1);
2593 if (p) {
2594 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2595 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2596 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597 }
2598
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002599 p = d_ancestor(p1, p2);
2600 if (p) {
2601 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2602 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2603 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002604 }
2605
Ingo Molnarf2eace22006-07-03 00:25:05 -07002606 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Miklos Szeredid1b72cc2014-10-27 15:42:01 +01002607 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608 return NULL;
2609}
Al Viro4d359502014-03-14 12:20:17 -04002610EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611
2612void unlock_rename(struct dentry *p1, struct dentry *p2)
2613{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002614 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002616 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002617 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618 }
2619}
Al Viro4d359502014-03-14 12:20:17 -04002620EXPORT_SYMBOL(unlock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621
Al Viro4acdaf22011-07-26 01:42:34 -04002622int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viro312b63f2012-06-10 18:09:36 -04002623 bool want_excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002624{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002625 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626 if (error)
2627 return error;
2628
Al Viroacfa4382008-12-04 10:06:33 -05002629 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630 return -EACCES; /* shouldn't it be ENOSYS? */
2631 mode &= S_IALLUGO;
2632 mode |= S_IFREG;
2633 error = security_inode_create(dir, dentry, mode);
2634 if (error)
2635 return error;
Al Viro312b63f2012-06-10 18:09:36 -04002636 error = dir->i_op->create(dir, dentry, mode, want_excl);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002637 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002638 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639 return error;
2640}
Al Viro4d359502014-03-14 12:20:17 -04002641EXPORT_SYMBOL(vfs_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642
Al Viro73d049a2011-03-11 12:08:24 -05002643static int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002645 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002646 struct inode *inode = dentry->d_inode;
2647 int error;
2648
Al Virobcda7652011-03-13 16:42:14 -04002649 /* O_PATH? */
2650 if (!acc_mode)
2651 return 0;
2652
Linus Torvalds1da177e2005-04-16 15:20:36 -07002653 if (!inode)
2654 return -ENOENT;
2655
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002656 switch (inode->i_mode & S_IFMT) {
2657 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002659 case S_IFDIR:
2660 if (acc_mode & MAY_WRITE)
2661 return -EISDIR;
2662 break;
2663 case S_IFBLK:
2664 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002665 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002667 /*FALLTHRU*/
2668 case S_IFIFO:
2669 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002671 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002672 }
Dave Hansenb41572e2007-10-16 23:31:14 -07002673
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002674 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07002675 if (error)
2676 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05002677
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678 /*
2679 * An append-only file must be opened in append mode for writing.
2680 */
2681 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05002682 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05002683 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002684 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05002685 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686 }
2687
2688 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002689 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05002690 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691
J. Bruce Fieldsf3c7691e2011-09-21 10:58:13 -04002692 return 0;
Al Viro7715b522009-12-16 03:54:00 -05002693}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002694
Jeff Laytone1181ee2010-12-07 16:19:50 -05002695static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002696{
Jeff Laytone1181ee2010-12-07 16:19:50 -05002697 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002698 struct inode *inode = path->dentry->d_inode;
2699 int error = get_write_access(inode);
2700 if (error)
2701 return error;
2702 /*
2703 * Refuse to truncate files with mandatory locks held on them.
2704 */
Jeff Laytond7a06982014-03-10 09:54:15 -04002705 error = locks_verify_locked(filp);
Al Viro7715b522009-12-16 03:54:00 -05002706 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002707 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002708 if (!error) {
2709 error = do_truncate(path->dentry, 0,
2710 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002711 filp);
Al Viro7715b522009-12-16 03:54:00 -05002712 }
2713 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002714 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715}
2716
Dave Hansend57999e2008-02-15 14:37:27 -08002717static inline int open_to_namei_flags(int flag)
2718{
Al Viro8a5e9292011-06-25 19:15:54 -04002719 if ((flag & O_ACCMODE) == 3)
2720 flag--;
Dave Hansend57999e2008-02-15 14:37:27 -08002721 return flag;
2722}
2723
Miklos Szeredid18e9002012-06-05 15:10:17 +02002724static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2725{
2726 int error = security_path_mknod(dir, dentry, mode, 0);
2727 if (error)
2728 return error;
2729
2730 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2731 if (error)
2732 return error;
2733
2734 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2735}
2736
David Howells1acf0af2012-06-14 16:13:46 +01002737/*
2738 * Attempt to atomically look up, create and open a file from a negative
2739 * dentry.
2740 *
2741 * Returns 0 if successful. The file will have been created and attached to
2742 * @file by the filesystem calling finish_open().
2743 *
2744 * Returns 1 if the file was looked up only or didn't need creating. The
2745 * caller will need to perform the open themselves. @path will have been
2746 * updated to point to the new dentry. This may be negative.
2747 *
2748 * Returns an error code otherwise.
2749 */
Al Viro2675a4e2012-06-22 12:41:10 +04002750static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2751 struct path *path, struct file *file,
2752 const struct open_flags *op,
Al Viro64894cf2012-07-31 00:53:35 +04002753 bool got_write, bool need_lookup,
Al Viro2675a4e2012-06-22 12:41:10 +04002754 int *opened)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002755{
2756 struct inode *dir = nd->path.dentry->d_inode;
2757 unsigned open_flag = open_to_namei_flags(op->open_flag);
2758 umode_t mode;
2759 int error;
2760 int acc_mode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002761 int create_error = 0;
2762 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
Miklos Szeredi116cc022013-09-16 14:52:05 +02002763 bool excl;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002764
2765 BUG_ON(dentry->d_inode);
2766
2767 /* Don't create child dentry for a dead directory. */
2768 if (unlikely(IS_DEADDIR(dir))) {
Al Viro2675a4e2012-06-22 12:41:10 +04002769 error = -ENOENT;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002770 goto out;
2771 }
2772
Miklos Szeredi62b259d2012-08-15 13:01:24 +02002773 mode = op->mode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002774 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2775 mode &= ~current_umask();
2776
Miklos Szeredi116cc022013-09-16 14:52:05 +02002777 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2778 if (excl)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002779 open_flag &= ~O_TRUNC;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002780
2781 /*
2782 * Checking write permission is tricky, bacuse we don't know if we are
2783 * going to actually need it: O_CREAT opens should work as long as the
2784 * file exists. But checking existence breaks atomicity. The trick is
2785 * to check access and if not granted clear O_CREAT from the flags.
2786 *
2787 * Another problem is returing the "right" error value (e.g. for an
2788 * O_EXCL open we want to return EEXIST not EROFS).
2789 */
Al Viro64894cf2012-07-31 00:53:35 +04002790 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2791 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2792 if (!(open_flag & O_CREAT)) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002793 /*
2794 * No O_CREATE -> atomicity not a requirement -> fall
2795 * back to lookup + open
2796 */
2797 goto no_open;
2798 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2799 /* Fall back and fail with the right error */
Al Viro64894cf2012-07-31 00:53:35 +04002800 create_error = -EROFS;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002801 goto no_open;
2802 } else {
2803 /* No side effects, safe to clear O_CREAT */
Al Viro64894cf2012-07-31 00:53:35 +04002804 create_error = -EROFS;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002805 open_flag &= ~O_CREAT;
2806 }
2807 }
2808
2809 if (open_flag & O_CREAT) {
Miklos Szeredi38227f72012-08-15 13:01:24 +02002810 error = may_o_create(&nd->path, dentry, mode);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002811 if (error) {
2812 create_error = error;
2813 if (open_flag & O_EXCL)
2814 goto no_open;
2815 open_flag &= ~O_CREAT;
2816 }
2817 }
2818
2819 if (nd->flags & LOOKUP_DIRECTORY)
2820 open_flag |= O_DIRECTORY;
2821
Al Viro30d90492012-06-22 12:40:19 +04002822 file->f_path.dentry = DENTRY_NOT_SET;
2823 file->f_path.mnt = nd->path.mnt;
2824 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
Al Viro47237682012-06-10 05:01:45 -04002825 opened);
Al Virod9585272012-06-22 12:39:14 +04002826 if (error < 0) {
Al Virod9585272012-06-22 12:39:14 +04002827 if (create_error && error == -ENOENT)
2828 error = create_error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002829 goto out;
2830 }
2831
Al Virod9585272012-06-22 12:39:14 +04002832 if (error) { /* returned 1, that is */
Al Viro30d90492012-06-22 12:40:19 +04002833 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
Al Viro2675a4e2012-06-22 12:41:10 +04002834 error = -EIO;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002835 goto out;
2836 }
Al Viro30d90492012-06-22 12:40:19 +04002837 if (file->f_path.dentry) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002838 dput(dentry);
Al Viro30d90492012-06-22 12:40:19 +04002839 dentry = file->f_path.dentry;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002840 }
Al Viro03da6332013-09-16 19:22:33 -04002841 if (*opened & FILE_CREATED)
2842 fsnotify_create(dir, dentry);
2843 if (!dentry->d_inode) {
2844 WARN_ON(*opened & FILE_CREATED);
2845 if (create_error) {
2846 error = create_error;
2847 goto out;
2848 }
2849 } else {
2850 if (excl && !(*opened & FILE_CREATED)) {
2851 error = -EEXIST;
2852 goto out;
2853 }
Sage Weil62b2ce92012-08-15 13:30:12 -07002854 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02002855 goto looked_up;
2856 }
2857
2858 /*
2859 * We didn't have the inode before the open, so check open permission
2860 * here.
2861 */
Al Viro03da6332013-09-16 19:22:33 -04002862 acc_mode = op->acc_mode;
2863 if (*opened & FILE_CREATED) {
2864 WARN_ON(!(open_flag & O_CREAT));
2865 fsnotify_create(dir, dentry);
2866 acc_mode = MAY_OPEN;
2867 }
Al Viro2675a4e2012-06-22 12:41:10 +04002868 error = may_open(&file->f_path, acc_mode, open_flag);
2869 if (error)
2870 fput(file);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002871
2872out:
2873 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04002874 return error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002875
Miklos Szeredid18e9002012-06-05 15:10:17 +02002876no_open:
2877 if (need_lookup) {
Al Viro72bd8662012-06-10 17:17:17 -04002878 dentry = lookup_real(dir, dentry, nd->flags);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002879 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002880 return PTR_ERR(dentry);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002881
2882 if (create_error) {
2883 int open_flag = op->open_flag;
2884
Al Viro2675a4e2012-06-22 12:41:10 +04002885 error = create_error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002886 if ((open_flag & O_EXCL)) {
2887 if (!dentry->d_inode)
2888 goto out;
2889 } else if (!dentry->d_inode) {
2890 goto out;
2891 } else if ((open_flag & O_TRUNC) &&
David Howellse36cb0b2015-01-29 12:02:35 +00002892 d_is_reg(dentry)) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002893 goto out;
2894 }
2895 /* will fail later, go on to get the right error */
2896 }
2897 }
2898looked_up:
2899 path->dentry = dentry;
2900 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04002901 return 1;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002902}
2903
Nick Piggin31e6b012011-01-07 17:49:52 +11002904/*
David Howells1acf0af2012-06-14 16:13:46 +01002905 * Look up and maybe create and open the last component.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002906 *
2907 * Must be called with i_mutex held on parent.
2908 *
David Howells1acf0af2012-06-14 16:13:46 +01002909 * Returns 0 if the file was successfully atomically created (if necessary) and
2910 * opened. In this case the file will be returned attached to @file.
2911 *
2912 * Returns 1 if the file was not completely opened at this time, though lookups
2913 * and creations will have been performed and the dentry returned in @path will
2914 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2915 * specified then a negative dentry may be returned.
2916 *
2917 * An error code is returned otherwise.
2918 *
2919 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2920 * cleared otherwise prior to returning.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002921 */
Al Viro2675a4e2012-06-22 12:41:10 +04002922static int lookup_open(struct nameidata *nd, struct path *path,
2923 struct file *file,
2924 const struct open_flags *op,
Al Viro64894cf2012-07-31 00:53:35 +04002925 bool got_write, int *opened)
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002926{
2927 struct dentry *dir = nd->path.dentry;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002928 struct inode *dir_inode = dir->d_inode;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002929 struct dentry *dentry;
2930 int error;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002931 bool need_lookup;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002932
Al Viro47237682012-06-10 05:01:45 -04002933 *opened &= ~FILE_CREATED;
Al Viro201f9562012-06-22 12:42:10 +04002934 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002935 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002936 return PTR_ERR(dentry);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002937
Miklos Szeredid18e9002012-06-05 15:10:17 +02002938 /* Cached positive dentry: will open in f_op->open */
2939 if (!need_lookup && dentry->d_inode)
2940 goto out_no_open;
2941
2942 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
Al Viro64894cf2012-07-31 00:53:35 +04002943 return atomic_open(nd, dentry, path, file, op, got_write,
Al Viro47237682012-06-10 05:01:45 -04002944 need_lookup, opened);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002945 }
2946
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002947 if (need_lookup) {
2948 BUG_ON(dentry->d_inode);
2949
Al Viro72bd8662012-06-10 17:17:17 -04002950 dentry = lookup_real(dir_inode, dentry, nd->flags);
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002951 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002952 return PTR_ERR(dentry);
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002953 }
2954
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002955 /* Negative dentry, just create the file */
2956 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2957 umode_t mode = op->mode;
2958 if (!IS_POSIXACL(dir->d_inode))
2959 mode &= ~current_umask();
2960 /*
2961 * This write is needed to ensure that a
2962 * rw->ro transition does not occur between
2963 * the time when the file is created and when
2964 * a permanent write count is taken through
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002965 * the 'struct file' in finish_open().
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002966 */
Al Viro64894cf2012-07-31 00:53:35 +04002967 if (!got_write) {
2968 error = -EROFS;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002969 goto out_dput;
Al Viro64894cf2012-07-31 00:53:35 +04002970 }
Al Viro47237682012-06-10 05:01:45 -04002971 *opened |= FILE_CREATED;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002972 error = security_path_mknod(&nd->path, dentry, mode, 0);
2973 if (error)
2974 goto out_dput;
Al Viro312b63f2012-06-10 18:09:36 -04002975 error = vfs_create(dir->d_inode, dentry, mode,
2976 nd->flags & LOOKUP_EXCL);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002977 if (error)
2978 goto out_dput;
2979 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02002980out_no_open:
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002981 path->dentry = dentry;
2982 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04002983 return 1;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002984
2985out_dput:
2986 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04002987 return error;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002988}
2989
2990/*
Al Virofe2d35f2011-03-05 22:58:25 -05002991 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11002992 */
Al Viro896475d2015-04-22 18:02:17 -04002993static int do_last(struct nameidata *nd,
Al Viro2675a4e2012-06-22 12:41:10 +04002994 struct file *file, const struct open_flags *op,
Al Viro76ae2a52015-05-12 18:44:32 -04002995 int *opened)
Al Virofb1cc552009-12-24 01:58:28 -05002996{
Al Viroa1e28032009-12-24 02:12:06 -05002997 struct dentry *dir = nd->path.dentry;
Al Viroca344a892011-03-09 00:36:45 -05002998 int open_flag = op->open_flag;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002999 bool will_truncate = (open_flag & O_TRUNC) != 0;
Al Viro64894cf2012-07-31 00:53:35 +04003000 bool got_write = false;
Al Virobcda7652011-03-13 16:42:14 -04003001 int acc_mode = op->acc_mode;
Al Viro254cf582015-05-05 09:40:46 -04003002 unsigned seq;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02003003 struct inode *inode;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003004 struct path save_parent = { .dentry = NULL, .mnt = NULL };
Al Viro896475d2015-04-22 18:02:17 -04003005 struct path path;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003006 bool retried = false;
Al Viro16c2cd72011-02-22 15:50:10 -05003007 int error;
Al Virofb1cc552009-12-24 01:58:28 -05003008
Al Viroc3e380b2011-02-23 13:39:45 -05003009 nd->flags &= ~LOOKUP_PARENT;
3010 nd->flags |= op->intent;
3011
Al Virobc77daa2013-06-06 09:12:33 -04003012 if (nd->last_type != LAST_NORM) {
Al Virofe2d35f2011-03-05 22:58:25 -05003013 error = handle_dots(nd, nd->last_type);
Al Virodeb106c2015-05-08 18:05:21 -04003014 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04003015 return error;
Miklos Szeredie83db162012-06-05 15:10:29 +02003016 goto finish_open;
Al Viro1f36f772009-12-26 10:56:19 -05003017 }
Al Viro67ee3ad2009-12-26 07:01:01 -05003018
Al Viroca344a892011-03-09 00:36:45 -05003019 if (!(open_flag & O_CREAT)) {
Al Virofe2d35f2011-03-05 22:58:25 -05003020 if (nd->last.name[nd->last.len])
3021 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3022 /* we _can_ be in RCU mode here */
Al Viro254cf582015-05-05 09:40:46 -04003023 error = lookup_fast(nd, &path, &inode, &seq);
Miklos Szeredi71574862012-06-05 15:10:14 +02003024 if (likely(!error))
3025 goto finish_lookup;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02003026
Miklos Szeredi71574862012-06-05 15:10:14 +02003027 if (error < 0)
Al Virodeb106c2015-05-08 18:05:21 -04003028 return error;
Miklos Szeredi37d7fff2012-06-05 15:10:12 +02003029
Miklos Szeredi71574862012-06-05 15:10:14 +02003030 BUG_ON(nd->inode != dir->d_inode);
Miklos Szeredib6183df2012-06-05 15:10:13 +02003031 } else {
3032 /* create side of things */
3033 /*
3034 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3035 * has been cleared when we got to the last component we are
3036 * about to look up
3037 */
3038 error = complete_walk(nd);
Al Viroe8bb73d2015-05-08 16:28:42 -04003039 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003040 return error;
Miklos Szeredib6183df2012-06-05 15:10:13 +02003041
Al Viro76ae2a52015-05-12 18:44:32 -04003042 audit_inode(nd->name, dir, LOOKUP_PARENT);
Miklos Szeredib6183df2012-06-05 15:10:13 +02003043 /* trailing slashes? */
Al Virodeb106c2015-05-08 18:05:21 -04003044 if (unlikely(nd->last.name[nd->last.len]))
3045 return -EISDIR;
Al Virofe2d35f2011-03-05 22:58:25 -05003046 }
3047
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003048retry_lookup:
Al Viro64894cf2012-07-31 00:53:35 +04003049 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3050 error = mnt_want_write(nd->path.mnt);
3051 if (!error)
3052 got_write = true;
3053 /*
3054 * do _not_ fail yet - we might not need that or fail with
3055 * a different error; let lookup_open() decide; we'll be
3056 * dropping this one anyway.
3057 */
3058 }
Al Viroa1e28032009-12-24 02:12:06 -05003059 mutex_lock(&dir->d_inode->i_mutex);
Al Viro896475d2015-04-22 18:02:17 -04003060 error = lookup_open(nd, &path, file, op, got_write, opened);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003061 mutex_unlock(&dir->d_inode->i_mutex);
Al Viroa1e28032009-12-24 02:12:06 -05003062
Al Viro2675a4e2012-06-22 12:41:10 +04003063 if (error <= 0) {
3064 if (error)
Miklos Szeredid18e9002012-06-05 15:10:17 +02003065 goto out;
3066
Al Viro47237682012-06-10 05:01:45 -04003067 if ((*opened & FILE_CREATED) ||
Al Viro496ad9a2013-01-23 17:07:38 -05003068 !S_ISREG(file_inode(file)->i_mode))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003069 will_truncate = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003070
Al Viro76ae2a52015-05-12 18:44:32 -04003071 audit_inode(nd->name, file->f_path.dentry, 0);
Miklos Szeredid18e9002012-06-05 15:10:17 +02003072 goto opened;
3073 }
Al Virofb1cc552009-12-24 01:58:28 -05003074
Al Viro47237682012-06-10 05:01:45 -04003075 if (*opened & FILE_CREATED) {
Al Viro9b44f1b2011-03-09 00:17:27 -05003076 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05003077 open_flag &= ~O_TRUNC;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003078 will_truncate = false;
Al Virobcda7652011-03-13 16:42:14 -04003079 acc_mode = MAY_OPEN;
Al Viro896475d2015-04-22 18:02:17 -04003080 path_to_nameidata(&path, nd);
Miklos Szeredie83db162012-06-05 15:10:29 +02003081 goto finish_open_created;
Al Virofb1cc552009-12-24 01:58:28 -05003082 }
3083
3084 /*
Jeff Layton3134f372012-07-25 10:19:47 -04003085 * create/update audit record if it already exists.
Al Virofb1cc552009-12-24 01:58:28 -05003086 */
Al Viro896475d2015-04-22 18:02:17 -04003087 if (d_is_positive(path.dentry))
Al Viro76ae2a52015-05-12 18:44:32 -04003088 audit_inode(nd->name, path.dentry, 0);
Al Virofb1cc552009-12-24 01:58:28 -05003089
Miklos Szeredid18e9002012-06-05 15:10:17 +02003090 /*
3091 * If atomic_open() acquired write access it is dropped now due to
3092 * possible mount and symlink following (this might be optimized away if
3093 * necessary...)
3094 */
Al Viro64894cf2012-07-31 00:53:35 +04003095 if (got_write) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02003096 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04003097 got_write = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003098 }
3099
Al Virodeb106c2015-05-08 18:05:21 -04003100 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3101 path_to_nameidata(&path, nd);
3102 return -EEXIST;
3103 }
Al Virofb1cc552009-12-24 01:58:28 -05003104
NeilBrown756daf22015-03-23 13:37:38 +11003105 error = follow_managed(&path, nd);
Al Virodeb106c2015-05-08 18:05:21 -04003106 if (unlikely(error < 0))
3107 return error;
Al Viroa3fbbde2011-11-07 21:21:26 +00003108
Miklos Szeredidecf3402012-05-21 17:30:08 +02003109 BUG_ON(nd->flags & LOOKUP_RCU);
David Howells63afdfc2015-05-06 15:59:00 +01003110 inode = d_backing_inode(path.dentry);
Al Viro254cf582015-05-05 09:40:46 -04003111 seq = 0; /* out of RCU mode, so the value doesn't matter */
Al Virodeb106c2015-05-08 18:05:21 -04003112 if (unlikely(d_is_negative(path.dentry))) {
Al Viro896475d2015-04-22 18:02:17 -04003113 path_to_nameidata(&path, nd);
Al Virodeb106c2015-05-08 18:05:21 -04003114 return -ENOENT;
Miklos Szeredi54c33e72012-05-21 17:30:10 +02003115 }
Al Viro766c4cb2015-05-07 19:24:57 -04003116finish_lookup:
Al Virod63ff282015-05-04 18:13:23 -04003117 if (nd->depth)
3118 put_link(nd);
Al Viro181548c2015-05-07 19:54:34 -04003119 error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW,
3120 inode, seq);
Al Virodeb106c2015-05-08 18:05:21 -04003121 if (unlikely(error))
Al Virod63ff282015-05-04 18:13:23 -04003122 return error;
Al Virofb1cc552009-12-24 01:58:28 -05003123
Al Viro896475d2015-04-22 18:02:17 -04003124 if (unlikely(d_is_symlink(path.dentry)) && !(open_flag & O_PATH)) {
3125 path_to_nameidata(&path, nd);
Al Virodeb106c2015-05-08 18:05:21 -04003126 return -ELOOP;
Al Viroa5cfe2d2015-04-22 12:10:45 -04003127 }
3128
Al Viro896475d2015-04-22 18:02:17 -04003129 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) {
3130 path_to_nameidata(&path, nd);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003131 } else {
3132 save_parent.dentry = nd->path.dentry;
Al Viro896475d2015-04-22 18:02:17 -04003133 save_parent.mnt = mntget(path.mnt);
3134 nd->path.dentry = path.dentry;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003135
3136 }
Miklos Szeredidecf3402012-05-21 17:30:08 +02003137 nd->inode = inode;
Al Viro254cf582015-05-05 09:40:46 -04003138 nd->seq = seq;
Al Viroa3fbbde2011-11-07 21:21:26 +00003139 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
Al Virobc77daa2013-06-06 09:12:33 -04003140finish_open:
Al Viroa3fbbde2011-11-07 21:21:26 +00003141 error = complete_walk(nd);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003142 if (error) {
3143 path_put(&save_parent);
Al Viro2675a4e2012-06-22 12:41:10 +04003144 return error;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003145 }
Al Viro76ae2a52015-05-12 18:44:32 -04003146 audit_inode(nd->name, nd->path.dentry, 0);
Al Virofb1cc552009-12-24 01:58:28 -05003147 error = -EISDIR;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003148 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04003149 goto out;
Miklos Szerediaf2f5542012-05-21 17:30:11 +02003150 error = -ENOTDIR;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003151 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04003152 goto out;
David Howells4bbcbd32015-03-17 22:16:40 +00003153 if (!d_is_reg(nd->path.dentry))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003154 will_truncate = false;
Al Viro6c0d46c2011-03-09 00:59:59 -05003155
Al Viro0f9d1a12011-03-09 00:13:14 -05003156 if (will_truncate) {
3157 error = mnt_want_write(nd->path.mnt);
3158 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003159 goto out;
Al Viro64894cf2012-07-31 00:53:35 +04003160 got_write = true;
Al Viro0f9d1a12011-03-09 00:13:14 -05003161 }
Miklos Szeredie83db162012-06-05 15:10:29 +02003162finish_open_created:
Al Virobcda7652011-03-13 16:42:14 -04003163 error = may_open(&nd->path, acc_mode, open_flag);
Al Viroca344a892011-03-09 00:36:45 -05003164 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003165 goto out;
Miklos Szeredi4aa7c632014-10-24 00:14:35 +02003166
3167 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3168 error = vfs_open(&nd->path, file, current_cred());
3169 if (!error) {
3170 *opened |= FILE_OPENED;
3171 } else {
Al Viro30d90492012-06-22 12:40:19 +04003172 if (error == -EOPENSTALE)
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003173 goto stale_open;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003174 goto out;
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003175 }
Miklos Szeredia8277b92012-06-05 15:10:32 +02003176opened:
Al Viro2675a4e2012-06-22 12:41:10 +04003177 error = open_check_o_direct(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003178 if (error)
3179 goto exit_fput;
Dmitry Kasatkin3034a142014-06-27 18:15:44 +03003180 error = ima_file_check(file, op->acc_mode, *opened);
Miklos Szerediaa4caad2012-06-05 15:10:28 +02003181 if (error)
3182 goto exit_fput;
3183
3184 if (will_truncate) {
Al Viro2675a4e2012-06-22 12:41:10 +04003185 error = handle_truncate(file);
Miklos Szerediaa4caad2012-06-05 15:10:28 +02003186 if (error)
3187 goto exit_fput;
Al Viro0f9d1a12011-03-09 00:13:14 -05003188 }
Al Viroca344a892011-03-09 00:36:45 -05003189out:
Al Viro64894cf2012-07-31 00:53:35 +04003190 if (got_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05003191 mnt_drop_write(nd->path.mnt);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003192 path_put(&save_parent);
Al Viro2675a4e2012-06-22 12:41:10 +04003193 return error;
Al Virofb1cc552009-12-24 01:58:28 -05003194
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003195exit_fput:
Al Viro2675a4e2012-06-22 12:41:10 +04003196 fput(file);
3197 goto out;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003198
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003199stale_open:
3200 /* If no saved parent or already retried then can't retry */
3201 if (!save_parent.dentry || retried)
3202 goto out;
3203
3204 BUG_ON(save_parent.dentry != dir);
3205 path_put(&nd->path);
3206 nd->path = save_parent;
3207 nd->inode = dir->d_inode;
3208 save_parent.mnt = NULL;
3209 save_parent.dentry = NULL;
Al Viro64894cf2012-07-31 00:53:35 +04003210 if (got_write) {
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003211 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04003212 got_write = false;
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003213 }
3214 retried = true;
3215 goto retry_lookup;
Al Virofb1cc552009-12-24 01:58:28 -05003216}
3217
Al Viroc8a53ee2015-05-12 18:43:07 -04003218static int do_tmpfile(struct nameidata *nd, unsigned flags,
Al Viro60545d02013-06-07 01:20:27 -04003219 const struct open_flags *op,
3220 struct file *file, int *opened)
3221{
3222 static const struct qstr name = QSTR_INIT("/", 1);
Al Viro625b6d12015-05-12 16:36:12 -04003223 struct dentry *child;
Al Viro60545d02013-06-07 01:20:27 -04003224 struct inode *dir;
Al Viro625b6d12015-05-12 16:36:12 -04003225 struct path path;
Al Viroc8a53ee2015-05-12 18:43:07 -04003226 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
Al Viro60545d02013-06-07 01:20:27 -04003227 if (unlikely(error))
3228 return error;
Al Viro625b6d12015-05-12 16:36:12 -04003229 error = mnt_want_write(path.mnt);
Al Viro60545d02013-06-07 01:20:27 -04003230 if (unlikely(error))
3231 goto out;
Al Viro625b6d12015-05-12 16:36:12 -04003232 dir = path.dentry->d_inode;
Al Viro60545d02013-06-07 01:20:27 -04003233 /* we want directory to be writable */
Al Viro625b6d12015-05-12 16:36:12 -04003234 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Al Viro60545d02013-06-07 01:20:27 -04003235 if (error)
3236 goto out2;
Al Viro60545d02013-06-07 01:20:27 -04003237 if (!dir->i_op->tmpfile) {
3238 error = -EOPNOTSUPP;
3239 goto out2;
3240 }
Al Viro625b6d12015-05-12 16:36:12 -04003241 child = d_alloc(path.dentry, &name);
Al Viro60545d02013-06-07 01:20:27 -04003242 if (unlikely(!child)) {
3243 error = -ENOMEM;
3244 goto out2;
3245 }
Al Viro625b6d12015-05-12 16:36:12 -04003246 dput(path.dentry);
3247 path.dentry = child;
3248 error = dir->i_op->tmpfile(dir, child, op->mode);
Al Viro60545d02013-06-07 01:20:27 -04003249 if (error)
3250 goto out2;
Al Viroc8a53ee2015-05-12 18:43:07 -04003251 audit_inode(nd->name, child, 0);
Eric Rannaud69a91c22014-10-30 01:51:01 -07003252 /* Don't check for other permissions, the inode was just created */
Al Viro625b6d12015-05-12 16:36:12 -04003253 error = may_open(&path, MAY_OPEN, op->open_flag);
Al Viro60545d02013-06-07 01:20:27 -04003254 if (error)
3255 goto out2;
Al Viro625b6d12015-05-12 16:36:12 -04003256 file->f_path.mnt = path.mnt;
3257 error = finish_open(file, child, NULL, opened);
Al Viro60545d02013-06-07 01:20:27 -04003258 if (error)
3259 goto out2;
3260 error = open_check_o_direct(file);
Al Virof4e0c302013-06-11 08:34:36 +04003261 if (error) {
Al Viro60545d02013-06-07 01:20:27 -04003262 fput(file);
Al Virof4e0c302013-06-11 08:34:36 +04003263 } else if (!(op->open_flag & O_EXCL)) {
3264 struct inode *inode = file_inode(file);
3265 spin_lock(&inode->i_lock);
3266 inode->i_state |= I_LINKABLE;
3267 spin_unlock(&inode->i_lock);
3268 }
Al Viro60545d02013-06-07 01:20:27 -04003269out2:
Al Viro625b6d12015-05-12 16:36:12 -04003270 mnt_drop_write(path.mnt);
Al Viro60545d02013-06-07 01:20:27 -04003271out:
Al Viro625b6d12015-05-12 16:36:12 -04003272 path_put(&path);
Al Viro60545d02013-06-07 01:20:27 -04003273 return error;
3274}
3275
Al Viroc8a53ee2015-05-12 18:43:07 -04003276static struct file *path_openat(struct nameidata *nd,
3277 const struct open_flags *op, unsigned flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003278{
Al Viro368ee9b2015-05-08 17:19:59 -04003279 const char *s;
Al Viro30d90492012-06-22 12:40:19 +04003280 struct file *file;
Al Viro47237682012-06-10 05:01:45 -04003281 int opened = 0;
Al Viro13aab422011-02-23 17:54:08 -05003282 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11003283
Al Viro30d90492012-06-22 12:40:19 +04003284 file = get_empty_filp();
Al Viro1afc99b2013-02-14 20:41:04 -05003285 if (IS_ERR(file))
3286 return file;
Nick Piggin31e6b012011-01-07 17:49:52 +11003287
Al Viro30d90492012-06-22 12:40:19 +04003288 file->f_flags = op->open_flag;
Nick Piggin31e6b012011-01-07 17:49:52 +11003289
Al Virobb458c62013-07-13 13:26:37 +04003290 if (unlikely(file->f_flags & __O_TMPFILE)) {
Al Viroc8a53ee2015-05-12 18:43:07 -04003291 error = do_tmpfile(nd, flags, op, file, &opened);
Al Virof15133d2015-05-08 22:53:15 -04003292 goto out2;
Al Viro60545d02013-06-07 01:20:27 -04003293 }
3294
Al Viroc8a53ee2015-05-12 18:43:07 -04003295 s = path_init(nd, flags);
Al Viro368ee9b2015-05-08 17:19:59 -04003296 if (IS_ERR(s)) {
3297 put_filp(file);
3298 return ERR_CAST(s);
3299 }
Al Viro3bdba282015-05-08 17:37:07 -04003300 while (!(error = link_path_walk(s, nd)) &&
Al Viro76ae2a52015-05-12 18:44:32 -04003301 (error = do_last(nd, file, op, &opened)) > 0) {
Al Viro73d049a2011-03-11 12:08:24 -05003302 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
Al Viro3bdba282015-05-08 17:37:07 -04003303 s = trailing_symlink(nd);
3304 if (IS_ERR(s)) {
3305 error = PTR_ERR(s);
Al Viro2675a4e2012-06-22 12:41:10 +04003306 break;
Al Viro3bdba282015-05-08 17:37:07 -04003307 }
Al Viro806b6812009-12-26 07:16:40 -05003308 }
Al Virodeb106c2015-05-08 18:05:21 -04003309 terminate_walk(nd);
Al Virof15133d2015-05-08 22:53:15 -04003310out2:
Al Viro2675a4e2012-06-22 12:41:10 +04003311 if (!(opened & FILE_OPENED)) {
3312 BUG_ON(!error);
Al Viro30d90492012-06-22 12:40:19 +04003313 put_filp(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003314 }
Al Viro2675a4e2012-06-22 12:41:10 +04003315 if (unlikely(error)) {
3316 if (error == -EOPENSTALE) {
3317 if (flags & LOOKUP_RCU)
3318 error = -ECHILD;
3319 else
3320 error = -ESTALE;
3321 }
3322 file = ERR_PTR(error);
3323 }
3324 return file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003325}
3326
Jeff Layton669abf42012-10-10 16:43:10 -04003327struct file *do_filp_open(int dfd, struct filename *pathname,
Al Virof9652e12013-06-11 08:23:01 +04003328 const struct open_flags *op)
Al Viro13aab422011-02-23 17:54:08 -05003329{
Al Viro9883d182015-05-13 07:28:08 -04003330 struct nameidata nd;
Al Virof9652e12013-06-11 08:23:01 +04003331 int flags = op->lookup_flags;
Al Viro13aab422011-02-23 17:54:08 -05003332 struct file *filp;
3333
Al Viro9883d182015-05-13 07:28:08 -04003334 set_nameidata(&nd, dfd, pathname);
Al Viroc8a53ee2015-05-12 18:43:07 -04003335 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
Al Viro13aab422011-02-23 17:54:08 -05003336 if (unlikely(filp == ERR_PTR(-ECHILD)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003337 filp = path_openat(&nd, op, flags);
Al Viro13aab422011-02-23 17:54:08 -05003338 if (unlikely(filp == ERR_PTR(-ESTALE)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003339 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
Al Viro9883d182015-05-13 07:28:08 -04003340 restore_nameidata();
Al Viro13aab422011-02-23 17:54:08 -05003341 return filp;
3342}
3343
Al Viro73d049a2011-03-11 12:08:24 -05003344struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
Al Virof9652e12013-06-11 08:23:01 +04003345 const char *name, const struct open_flags *op)
Al Viro73d049a2011-03-11 12:08:24 -05003346{
Al Viro9883d182015-05-13 07:28:08 -04003347 struct nameidata nd;
Al Viro73d049a2011-03-11 12:08:24 -05003348 struct file *file;
Paul Moore51689102015-01-22 00:00:03 -05003349 struct filename *filename;
Al Virof9652e12013-06-11 08:23:01 +04003350 int flags = op->lookup_flags | LOOKUP_ROOT;
Al Viro73d049a2011-03-11 12:08:24 -05003351
3352 nd.root.mnt = mnt;
3353 nd.root.dentry = dentry;
3354
David Howellsb18825a2013-09-12 19:22:53 +01003355 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
Al Viro73d049a2011-03-11 12:08:24 -05003356 return ERR_PTR(-ELOOP);
3357
Paul Moore51689102015-01-22 00:00:03 -05003358 filename = getname_kernel(name);
3359 if (unlikely(IS_ERR(filename)))
3360 return ERR_CAST(filename);
3361
Al Viro9883d182015-05-13 07:28:08 -04003362 set_nameidata(&nd, -1, filename);
Al Viroc8a53ee2015-05-12 18:43:07 -04003363 file = path_openat(&nd, op, flags | LOOKUP_RCU);
Al Viro73d049a2011-03-11 12:08:24 -05003364 if (unlikely(file == ERR_PTR(-ECHILD)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003365 file = path_openat(&nd, op, flags);
Al Viro73d049a2011-03-11 12:08:24 -05003366 if (unlikely(file == ERR_PTR(-ESTALE)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003367 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
Al Viro9883d182015-05-13 07:28:08 -04003368 restore_nameidata();
Paul Moore51689102015-01-22 00:00:03 -05003369 putname(filename);
Al Viro73d049a2011-03-11 12:08:24 -05003370 return file;
3371}
3372
Al Virofa14a0b2015-01-22 02:16:49 -05003373static struct dentry *filename_create(int dfd, struct filename *name,
Jeff Layton1ac12b42012-12-11 12:10:06 -05003374 struct path *path, unsigned int lookup_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003375{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003376 struct dentry *dentry = ERR_PTR(-EEXIST);
Al Viro391172c2015-05-09 11:19:16 -04003377 struct qstr last;
3378 int type;
Jan Karac30dabf2012-06-12 16:20:30 +02003379 int err2;
Jeff Layton1ac12b42012-12-11 12:10:06 -05003380 int error;
3381 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3382
3383 /*
3384 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3385 * other flags passed in are ignored!
3386 */
3387 lookup_flags &= LOOKUP_REVAL;
3388
Al Viro5c31b6c2015-05-12 17:32:54 -04003389 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3390 if (IS_ERR(name))
3391 return ERR_CAST(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003392
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003393 /*
3394 * Yucky last component or no last component at all?
3395 * (foo/., foo/.., /////)
3396 */
Al Viro5c31b6c2015-05-12 17:32:54 -04003397 if (unlikely(type != LAST_NORM))
Al Viroed75e952011-06-27 16:53:43 -04003398 goto out;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003399
Jan Karac30dabf2012-06-12 16:20:30 +02003400 /* don't fail immediately if it's r/o, at least try to report other errors */
Al Viro391172c2015-05-09 11:19:16 -04003401 err2 = mnt_want_write(path->mnt);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003402 /*
3403 * Do the final lookup.
3404 */
Al Viro391172c2015-05-09 11:19:16 -04003405 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3406 mutex_lock_nested(&path->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3407 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003408 if (IS_ERR(dentry))
Al Viroa8104a92012-07-20 02:25:00 +04003409 goto unlock;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003410
Al Viroa8104a92012-07-20 02:25:00 +04003411 error = -EEXIST;
David Howellsb18825a2013-09-12 19:22:53 +01003412 if (d_is_positive(dentry))
Al Viroa8104a92012-07-20 02:25:00 +04003413 goto fail;
David Howellsb18825a2013-09-12 19:22:53 +01003414
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003415 /*
3416 * Special case - lookup gave negative, but... we had foo/bar/
3417 * From the vfs_mknod() POV we just have a negative dentry -
3418 * all is fine. Let's be bastards - you had / on the end, you've
3419 * been asking for (non-existent) directory. -ENOENT for you.
3420 */
Al Viro391172c2015-05-09 11:19:16 -04003421 if (unlikely(!is_dir && last.name[last.len])) {
Al Viroa8104a92012-07-20 02:25:00 +04003422 error = -ENOENT;
Al Viroed75e952011-06-27 16:53:43 -04003423 goto fail;
Al Viroe9baf6e2008-05-15 04:49:12 -04003424 }
Jan Karac30dabf2012-06-12 16:20:30 +02003425 if (unlikely(err2)) {
3426 error = err2;
Al Viroa8104a92012-07-20 02:25:00 +04003427 goto fail;
Jan Karac30dabf2012-06-12 16:20:30 +02003428 }
Al Viro181c37b2015-05-12 17:21:25 -04003429 putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003430 return dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003431fail:
Al Viroa8104a92012-07-20 02:25:00 +04003432 dput(dentry);
3433 dentry = ERR_PTR(error);
3434unlock:
Al Viro391172c2015-05-09 11:19:16 -04003435 mutex_unlock(&path->dentry->d_inode->i_mutex);
Jan Karac30dabf2012-06-12 16:20:30 +02003436 if (!err2)
Al Viro391172c2015-05-09 11:19:16 -04003437 mnt_drop_write(path->mnt);
Al Viroed75e952011-06-27 16:53:43 -04003438out:
Al Viro391172c2015-05-09 11:19:16 -04003439 path_put(path);
Al Viro181c37b2015-05-12 17:21:25 -04003440 putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003441 return dentry;
3442}
Al Virofa14a0b2015-01-22 02:16:49 -05003443
3444struct dentry *kern_path_create(int dfd, const char *pathname,
3445 struct path *path, unsigned int lookup_flags)
3446{
Al Viro181c37b2015-05-12 17:21:25 -04003447 return filename_create(dfd, getname_kernel(pathname),
3448 path, lookup_flags);
Al Virofa14a0b2015-01-22 02:16:49 -05003449}
Al Virodae6ad82011-06-26 11:50:15 -04003450EXPORT_SYMBOL(kern_path_create);
3451
Al Viro921a1652012-07-20 01:15:31 +04003452void done_path_create(struct path *path, struct dentry *dentry)
3453{
3454 dput(dentry);
3455 mutex_unlock(&path->dentry->d_inode->i_mutex);
Al Viroa8104a92012-07-20 02:25:00 +04003456 mnt_drop_write(path->mnt);
Al Viro921a1652012-07-20 01:15:31 +04003457 path_put(path);
3458}
3459EXPORT_SYMBOL(done_path_create);
3460
Al Viro520ae682015-05-13 07:00:28 -04003461inline struct dentry *user_path_create(int dfd, const char __user *pathname,
Jeff Layton1ac12b42012-12-11 12:10:06 -05003462 struct path *path, unsigned int lookup_flags)
Al Virodae6ad82011-06-26 11:50:15 -04003463{
Al Viro181c37b2015-05-12 17:21:25 -04003464 return filename_create(dfd, getname(pathname), path, lookup_flags);
Al Virodae6ad82011-06-26 11:50:15 -04003465}
3466EXPORT_SYMBOL(user_path_create);
3467
Al Viro1a67aaf2011-07-26 01:52:52 -04003468int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003469{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003470 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003471
3472 if (error)
3473 return error;
3474
Eric W. Biederman975d6b32011-11-13 12:16:43 -08003475 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003476 return -EPERM;
3477
Al Viroacfa4382008-12-04 10:06:33 -05003478 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003479 return -EPERM;
3480
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07003481 error = devcgroup_inode_mknod(mode, dev);
3482 if (error)
3483 return error;
3484
Linus Torvalds1da177e2005-04-16 15:20:36 -07003485 error = security_inode_mknod(dir, dentry, mode, dev);
3486 if (error)
3487 return error;
3488
Linus Torvalds1da177e2005-04-16 15:20:36 -07003489 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003490 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003491 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003492 return error;
3493}
Al Viro4d359502014-03-14 12:20:17 -04003494EXPORT_SYMBOL(vfs_mknod);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003495
Al Virof69aac02011-07-26 04:31:40 -04003496static int may_mknod(umode_t mode)
Dave Hansen463c3192008-02-15 14:37:57 -08003497{
3498 switch (mode & S_IFMT) {
3499 case S_IFREG:
3500 case S_IFCHR:
3501 case S_IFBLK:
3502 case S_IFIFO:
3503 case S_IFSOCK:
3504 case 0: /* zero mode translates to S_IFREG */
3505 return 0;
3506 case S_IFDIR:
3507 return -EPERM;
3508 default:
3509 return -EINVAL;
3510 }
3511}
3512
Al Viro8208a222011-07-25 17:32:17 -04003513SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003514 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003515{
Al Viro2ad94ae2008-07-21 09:32:51 -04003516 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003517 struct path path;
3518 int error;
Jeff Layton972567f2012-12-20 16:00:10 -05003519 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003520
Al Viro8e4bfca2012-07-20 01:17:26 +04003521 error = may_mknod(mode);
3522 if (error)
3523 return error;
Jeff Layton972567f2012-12-20 16:00:10 -05003524retry:
3525 dentry = user_path_create(dfd, filename, &path, lookup_flags);
Al Virodae6ad82011-06-26 11:50:15 -04003526 if (IS_ERR(dentry))
3527 return PTR_ERR(dentry);
Al Viro2ad94ae2008-07-21 09:32:51 -04003528
Al Virodae6ad82011-06-26 11:50:15 -04003529 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003530 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003531 error = security_path_mknod(&path, dentry, mode, dev);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003532 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04003533 goto out;
Dave Hansen463c3192008-02-15 14:37:57 -08003534 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003535 case 0: case S_IFREG:
Al Viro312b63f2012-06-10 18:09:36 -04003536 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003537 break;
3538 case S_IFCHR: case S_IFBLK:
Al Virodae6ad82011-06-26 11:50:15 -04003539 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003540 new_decode_dev(dev));
3541 break;
3542 case S_IFIFO: case S_IFSOCK:
Al Virodae6ad82011-06-26 11:50:15 -04003543 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003544 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003545 }
Al Viroa8104a92012-07-20 02:25:00 +04003546out:
Al Viro921a1652012-07-20 01:15:31 +04003547 done_path_create(&path, dentry);
Jeff Layton972567f2012-12-20 16:00:10 -05003548 if (retry_estale(error, lookup_flags)) {
3549 lookup_flags |= LOOKUP_REVAL;
3550 goto retry;
3551 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003552 return error;
3553}
3554
Al Viro8208a222011-07-25 17:32:17 -04003555SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003556{
3557 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3558}
3559
Al Viro18bb1db2011-07-26 01:41:39 -04003560int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003561{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003562 int error = may_create(dir, dentry);
Al Viro8de52772012-02-06 12:45:27 -05003563 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003564
3565 if (error)
3566 return error;
3567
Al Viroacfa4382008-12-04 10:06:33 -05003568 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003569 return -EPERM;
3570
3571 mode &= (S_IRWXUGO|S_ISVTX);
3572 error = security_inode_mkdir(dir, dentry, mode);
3573 if (error)
3574 return error;
3575
Al Viro8de52772012-02-06 12:45:27 -05003576 if (max_links && dir->i_nlink >= max_links)
3577 return -EMLINK;
3578
Linus Torvalds1da177e2005-04-16 15:20:36 -07003579 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003580 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003581 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003582 return error;
3583}
Al Viro4d359502014-03-14 12:20:17 -04003584EXPORT_SYMBOL(vfs_mkdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003585
Al Viroa218d0f2011-11-21 14:59:34 -05003586SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003587{
Dave Hansen6902d922006-09-30 23:29:01 -07003588 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003589 struct path path;
3590 int error;
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003591 unsigned int lookup_flags = LOOKUP_DIRECTORY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003592
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003593retry:
3594 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07003595 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003596 return PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003597
Al Virodae6ad82011-06-26 11:50:15 -04003598 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003599 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003600 error = security_path_mkdir(&path, dentry, mode);
Al Viroa8104a92012-07-20 02:25:00 +04003601 if (!error)
3602 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
Al Viro921a1652012-07-20 01:15:31 +04003603 done_path_create(&path, dentry);
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003604 if (retry_estale(error, lookup_flags)) {
3605 lookup_flags |= LOOKUP_REVAL;
3606 goto retry;
3607 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003608 return error;
3609}
3610
Al Viroa218d0f2011-11-21 14:59:34 -05003611SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003612{
3613 return sys_mkdirat(AT_FDCWD, pathname, mode);
3614}
3615
Linus Torvalds1da177e2005-04-16 15:20:36 -07003616/*
Sage Weila71905f2011-05-24 13:06:08 -07003617 * The dentry_unhash() helper will try to drop the dentry early: we
J. Bruce Fieldsc0d02592012-02-15 11:48:40 -05003618 * should have a usage count of 1 if we're the only user of this
Sage Weila71905f2011-05-24 13:06:08 -07003619 * dentry, and if that is true (possibly after pruning the dcache),
3620 * then we drop the dentry now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003621 *
3622 * A low-level filesystem can, if it choses, legally
3623 * do a
3624 *
3625 * if (!d_unhashed(dentry))
3626 * return -EBUSY;
3627 *
3628 * if it cannot handle the case of removing a directory
3629 * that is still in use by something else..
3630 */
3631void dentry_unhash(struct dentry *dentry)
3632{
Vasily Averindc168422006-12-06 20:37:07 -08003633 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003634 spin_lock(&dentry->d_lock);
Waiman Long98474232013-08-28 18:24:59 -07003635 if (dentry->d_lockref.count == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003636 __d_drop(dentry);
3637 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003638}
Al Viro4d359502014-03-14 12:20:17 -04003639EXPORT_SYMBOL(dentry_unhash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003640
3641int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3642{
3643 int error = may_delete(dir, dentry, 1);
3644
3645 if (error)
3646 return error;
3647
Al Viroacfa4382008-12-04 10:06:33 -05003648 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003649 return -EPERM;
3650
Al Viro1d2ef592011-09-14 18:55:41 +01003651 dget(dentry);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003652 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003653
Sage Weil912dbc12011-05-24 13:06:11 -07003654 error = -EBUSY;
Eric W. Biederman7af13642013-10-04 19:15:13 -07003655 if (is_local_mountpoint(dentry))
Sage Weil912dbc12011-05-24 13:06:11 -07003656 goto out;
3657
3658 error = security_inode_rmdir(dir, dentry);
3659 if (error)
3660 goto out;
3661
Sage Weil3cebde22011-05-29 21:20:59 -07003662 shrink_dcache_parent(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003663 error = dir->i_op->rmdir(dir, dentry);
3664 if (error)
3665 goto out;
3666
3667 dentry->d_inode->i_flags |= S_DEAD;
3668 dont_mount(dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003669 detach_mounts(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003670
3671out:
3672 mutex_unlock(&dentry->d_inode->i_mutex);
Al Viro1d2ef592011-09-14 18:55:41 +01003673 dput(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003674 if (!error)
3675 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003676 return error;
3677}
Al Viro4d359502014-03-14 12:20:17 -04003678EXPORT_SYMBOL(vfs_rmdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003679
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003680static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003681{
3682 int error = 0;
Jeff Layton91a27b22012-10-10 15:25:28 -04003683 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003684 struct dentry *dentry;
Al Virof5beed72015-04-30 16:09:11 -04003685 struct path path;
3686 struct qstr last;
3687 int type;
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003688 unsigned int lookup_flags = 0;
3689retry:
Al Virof5beed72015-04-30 16:09:11 -04003690 name = user_path_parent(dfd, pathname,
3691 &path, &last, &type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04003692 if (IS_ERR(name))
3693 return PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003694
Al Virof5beed72015-04-30 16:09:11 -04003695 switch (type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003696 case LAST_DOTDOT:
3697 error = -ENOTEMPTY;
3698 goto exit1;
3699 case LAST_DOT:
3700 error = -EINVAL;
3701 goto exit1;
3702 case LAST_ROOT:
3703 error = -EBUSY;
3704 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003705 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003706
Al Virof5beed72015-04-30 16:09:11 -04003707 error = mnt_want_write(path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02003708 if (error)
3709 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003710
Al Virof5beed72015-04-30 16:09:11 -04003711 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3712 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003713 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003714 if (IS_ERR(dentry))
3715 goto exit2;
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003716 if (!dentry->d_inode) {
3717 error = -ENOENT;
3718 goto exit3;
3719 }
Al Virof5beed72015-04-30 16:09:11 -04003720 error = security_path_rmdir(&path, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003721 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003722 goto exit3;
Al Virof5beed72015-04-30 16:09:11 -04003723 error = vfs_rmdir(path.dentry->d_inode, dentry);
Dave Hansen06227532008-02-15 14:37:34 -08003724exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07003725 dput(dentry);
3726exit2:
Al Virof5beed72015-04-30 16:09:11 -04003727 mutex_unlock(&path.dentry->d_inode->i_mutex);
3728 mnt_drop_write(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003729exit1:
Al Virof5beed72015-04-30 16:09:11 -04003730 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003731 putname(name);
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003732 if (retry_estale(error, lookup_flags)) {
3733 lookup_flags |= LOOKUP_REVAL;
3734 goto retry;
3735 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003736 return error;
3737}
3738
Heiko Carstens3cdad422009-01-14 14:14:22 +01003739SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003740{
3741 return do_rmdir(AT_FDCWD, pathname);
3742}
3743
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003744/**
3745 * vfs_unlink - unlink a filesystem object
3746 * @dir: parent directory
3747 * @dentry: victim
3748 * @delegated_inode: returns victim inode, if the inode is delegated.
3749 *
3750 * The caller must hold dir->i_mutex.
3751 *
3752 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3753 * return a reference to the inode in delegated_inode. The caller
3754 * should then break the delegation on that inode and retry. Because
3755 * breaking a delegation may take a long time, the caller should drop
3756 * dir->i_mutex before doing so.
3757 *
3758 * Alternatively, a caller may pass NULL for delegated_inode. This may
3759 * be appropriate for callers that expect the underlying filesystem not
3760 * to be NFS exported.
3761 */
3762int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003763{
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003764 struct inode *target = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003765 int error = may_delete(dir, dentry, 0);
3766
3767 if (error)
3768 return error;
3769
Al Viroacfa4382008-12-04 10:06:33 -05003770 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003771 return -EPERM;
3772
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003773 mutex_lock(&target->i_mutex);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003774 if (is_local_mountpoint(dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003775 error = -EBUSY;
3776 else {
3777 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05003778 if (!error) {
J. Bruce Fields5a146962012-08-28 07:50:40 -07003779 error = try_break_deleg(target, delegated_inode);
3780 if (error)
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003781 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003782 error = dir->i_op->unlink(dir, dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003783 if (!error) {
Al Virod83c49f2010-04-30 17:17:09 -04003784 dont_mount(dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003785 detach_mounts(dentry);
3786 }
Al Virobec10522010-03-03 14:12:08 -05003787 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003788 }
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003789out:
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003790 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003791
3792 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3793 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003794 fsnotify_link_count(target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003795 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003796 }
Robert Love0eeca282005-07-12 17:06:03 -04003797
Linus Torvalds1da177e2005-04-16 15:20:36 -07003798 return error;
3799}
Al Viro4d359502014-03-14 12:20:17 -04003800EXPORT_SYMBOL(vfs_unlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003801
3802/*
3803 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003804 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07003805 * writeout happening, and we don't want to prevent access to the directory
3806 * while waiting on the I/O.
3807 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003808static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003809{
Al Viro2ad94ae2008-07-21 09:32:51 -04003810 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04003811 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003812 struct dentry *dentry;
Al Virof5beed72015-04-30 16:09:11 -04003813 struct path path;
3814 struct qstr last;
3815 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003816 struct inode *inode = NULL;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003817 struct inode *delegated_inode = NULL;
Jeff Layton5d18f812012-12-20 16:38:04 -05003818 unsigned int lookup_flags = 0;
3819retry:
Al Virof5beed72015-04-30 16:09:11 -04003820 name = user_path_parent(dfd, pathname,
3821 &path, &last, &type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04003822 if (IS_ERR(name))
3823 return PTR_ERR(name);
Al Viro2ad94ae2008-07-21 09:32:51 -04003824
Linus Torvalds1da177e2005-04-16 15:20:36 -07003825 error = -EISDIR;
Al Virof5beed72015-04-30 16:09:11 -04003826 if (type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003827 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003828
Al Virof5beed72015-04-30 16:09:11 -04003829 error = mnt_want_write(path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02003830 if (error)
3831 goto exit1;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003832retry_deleg:
Al Virof5beed72015-04-30 16:09:11 -04003833 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3834 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003835 error = PTR_ERR(dentry);
3836 if (!IS_ERR(dentry)) {
3837 /* Why not before? Because we want correct error value */
Al Virof5beed72015-04-30 16:09:11 -04003838 if (last.name[last.len])
Török Edwin50338b82011-06-16 00:06:14 +03003839 goto slashes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003840 inode = dentry->d_inode;
David Howellsb18825a2013-09-12 19:22:53 +01003841 if (d_is_negative(dentry))
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003842 goto slashes;
3843 ihold(inode);
Al Virof5beed72015-04-30 16:09:11 -04003844 error = security_path_unlink(&path, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003845 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003846 goto exit2;
Al Virof5beed72015-04-30 16:09:11 -04003847 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
Jan Karac30dabf2012-06-12 16:20:30 +02003848exit2:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003849 dput(dentry);
3850 }
Al Virof5beed72015-04-30 16:09:11 -04003851 mutex_unlock(&path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003852 if (inode)
3853 iput(inode); /* truncate the inode here */
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003854 inode = NULL;
3855 if (delegated_inode) {
J. Bruce Fields5a146962012-08-28 07:50:40 -07003856 error = break_deleg_wait(&delegated_inode);
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003857 if (!error)
3858 goto retry_deleg;
3859 }
Al Virof5beed72015-04-30 16:09:11 -04003860 mnt_drop_write(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003861exit1:
Al Virof5beed72015-04-30 16:09:11 -04003862 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003863 putname(name);
Jeff Layton5d18f812012-12-20 16:38:04 -05003864 if (retry_estale(error, lookup_flags)) {
3865 lookup_flags |= LOOKUP_REVAL;
3866 inode = NULL;
3867 goto retry;
3868 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003869 return error;
3870
3871slashes:
David Howellsb18825a2013-09-12 19:22:53 +01003872 if (d_is_negative(dentry))
3873 error = -ENOENT;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003874 else if (d_is_dir(dentry))
David Howellsb18825a2013-09-12 19:22:53 +01003875 error = -EISDIR;
3876 else
3877 error = -ENOTDIR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003878 goto exit2;
3879}
3880
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003881SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003882{
3883 if ((flag & ~AT_REMOVEDIR) != 0)
3884 return -EINVAL;
3885
3886 if (flag & AT_REMOVEDIR)
3887 return do_rmdir(dfd, pathname);
3888
3889 return do_unlinkat(dfd, pathname);
3890}
3891
Heiko Carstens3480b252009-01-14 14:14:16 +01003892SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003893{
3894 return do_unlinkat(AT_FDCWD, pathname);
3895}
3896
Miklos Szeredidb2e7472008-06-24 16:50:16 +02003897int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003898{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003899 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003900
3901 if (error)
3902 return error;
3903
Al Viroacfa4382008-12-04 10:06:33 -05003904 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003905 return -EPERM;
3906
3907 error = security_inode_symlink(dir, dentry, oldname);
3908 if (error)
3909 return error;
3910
Linus Torvalds1da177e2005-04-16 15:20:36 -07003911 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003912 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003913 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003914 return error;
3915}
Al Viro4d359502014-03-14 12:20:17 -04003916EXPORT_SYMBOL(vfs_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003917
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003918SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3919 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003920{
Al Viro2ad94ae2008-07-21 09:32:51 -04003921 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04003922 struct filename *from;
Dave Hansen6902d922006-09-30 23:29:01 -07003923 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003924 struct path path;
Jeff Laytonf46d3562012-12-11 12:10:08 -05003925 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003926
3927 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04003928 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003929 return PTR_ERR(from);
Jeff Laytonf46d3562012-12-11 12:10:08 -05003930retry:
3931 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07003932 error = PTR_ERR(dentry);
3933 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003934 goto out_putname;
Dave Hansen6902d922006-09-30 23:29:01 -07003935
Jeff Layton91a27b22012-10-10 15:25:28 -04003936 error = security_path_symlink(&path, dentry, from->name);
Al Viroa8104a92012-07-20 02:25:00 +04003937 if (!error)
Jeff Layton91a27b22012-10-10 15:25:28 -04003938 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
Al Viro921a1652012-07-20 01:15:31 +04003939 done_path_create(&path, dentry);
Jeff Laytonf46d3562012-12-11 12:10:08 -05003940 if (retry_estale(error, lookup_flags)) {
3941 lookup_flags |= LOOKUP_REVAL;
3942 goto retry;
3943 }
Dave Hansen6902d922006-09-30 23:29:01 -07003944out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003945 putname(from);
3946 return error;
3947}
3948
Heiko Carstens3480b252009-01-14 14:14:16 +01003949SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003950{
3951 return sys_symlinkat(oldname, AT_FDCWD, newname);
3952}
3953
J. Bruce Fields146a8592011-09-20 17:14:31 -04003954/**
3955 * vfs_link - create a new link
3956 * @old_dentry: object to be linked
3957 * @dir: new parent
3958 * @new_dentry: where to create the new link
3959 * @delegated_inode: returns inode needing a delegation break
3960 *
3961 * The caller must hold dir->i_mutex
3962 *
3963 * If vfs_link discovers a delegation on the to-be-linked file in need
3964 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3965 * inode in delegated_inode. The caller should then break the delegation
3966 * and retry. Because breaking a delegation may take a long time, the
3967 * caller should drop the i_mutex before doing so.
3968 *
3969 * Alternatively, a caller may pass NULL for delegated_inode. This may
3970 * be appropriate for callers that expect the underlying filesystem not
3971 * to be NFS exported.
3972 */
3973int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003974{
3975 struct inode *inode = old_dentry->d_inode;
Al Viro8de52772012-02-06 12:45:27 -05003976 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003977 int error;
3978
3979 if (!inode)
3980 return -ENOENT;
3981
Miklos Szeredia95164d2008-07-30 15:08:48 +02003982 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003983 if (error)
3984 return error;
3985
3986 if (dir->i_sb != inode->i_sb)
3987 return -EXDEV;
3988
3989 /*
3990 * A link to an append-only or immutable file cannot be created.
3991 */
3992 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3993 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05003994 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003995 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003996 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003997 return -EPERM;
3998
3999 error = security_inode_link(old_dentry, dir, new_dentry);
4000 if (error)
4001 return error;
4002
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02004003 mutex_lock(&inode->i_mutex);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05304004 /* Make sure we don't allow creating hardlink to an unlinked file */
Al Virof4e0c302013-06-11 08:34:36 +04004005 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05304006 error = -ENOENT;
Al Viro8de52772012-02-06 12:45:27 -05004007 else if (max_links && inode->i_nlink >= max_links)
4008 error = -EMLINK;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004009 else {
4010 error = try_break_deleg(inode, delegated_inode);
4011 if (!error)
4012 error = dir->i_op->link(old_dentry, dir, new_dentry);
4013 }
Al Virof4e0c302013-06-11 08:34:36 +04004014
4015 if (!error && (inode->i_state & I_LINKABLE)) {
4016 spin_lock(&inode->i_lock);
4017 inode->i_state &= ~I_LINKABLE;
4018 spin_unlock(&inode->i_lock);
4019 }
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02004020 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07004021 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02004022 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004023 return error;
4024}
Al Viro4d359502014-03-14 12:20:17 -04004025EXPORT_SYMBOL(vfs_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004026
4027/*
4028 * Hardlinks are often used in delicate situations. We avoid
4029 * security-related surprises by not following symlinks on the
4030 * newname. --KAB
4031 *
4032 * We don't follow them on the oldname either to be compatible
4033 * with linux 2.0, and to avoid hard-linking to directories
4034 * and other special files. --ADM
4035 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01004036SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4037 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004038{
4039 struct dentry *new_dentry;
Al Virodae6ad82011-06-26 11:50:15 -04004040 struct path old_path, new_path;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004041 struct inode *delegated_inode = NULL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304042 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004043 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004044
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304045 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004046 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304047 /*
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004048 * To use null names we require CAP_DAC_READ_SEARCH
4049 * This ensures that not everyone will be able to create
4050 * handlink using the passed filedescriptor.
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304051 */
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004052 if (flags & AT_EMPTY_PATH) {
4053 if (!capable(CAP_DAC_READ_SEARCH))
4054 return -ENOENT;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304055 how = LOOKUP_EMPTY;
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004056 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004057
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304058 if (flags & AT_SYMLINK_FOLLOW)
4059 how |= LOOKUP_FOLLOW;
Jeff Layton442e31c2012-12-20 16:15:38 -05004060retry:
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304061 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004062 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04004063 return error;
4064
Jeff Layton442e31c2012-12-20 16:15:38 -05004065 new_dentry = user_path_create(newdfd, newname, &new_path,
4066 (how & LOOKUP_REVAL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004067 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07004068 if (IS_ERR(new_dentry))
Al Virodae6ad82011-06-26 11:50:15 -04004069 goto out;
4070
4071 error = -EXDEV;
4072 if (old_path.mnt != new_path.mnt)
4073 goto out_dput;
Kees Cook800179c2012-07-25 17:29:07 -07004074 error = may_linkat(&old_path);
4075 if (unlikely(error))
4076 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04004077 error = security_path_link(old_path.dentry, &new_path, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004078 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04004079 goto out_dput;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004080 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
Dave Hansen75c3f292008-02-15 14:37:45 -08004081out_dput:
Al Viro921a1652012-07-20 01:15:31 +04004082 done_path_create(&new_path, new_dentry);
J. Bruce Fields146a8592011-09-20 17:14:31 -04004083 if (delegated_inode) {
4084 error = break_deleg_wait(&delegated_inode);
Oleg Drokind22e6332014-01-31 15:41:58 -05004085 if (!error) {
4086 path_put(&old_path);
J. Bruce Fields146a8592011-09-20 17:14:31 -04004087 goto retry;
Oleg Drokind22e6332014-01-31 15:41:58 -05004088 }
J. Bruce Fields146a8592011-09-20 17:14:31 -04004089 }
Jeff Layton442e31c2012-12-20 16:15:38 -05004090 if (retry_estale(error, how)) {
Oleg Drokind22e6332014-01-31 15:41:58 -05004091 path_put(&old_path);
Jeff Layton442e31c2012-12-20 16:15:38 -05004092 how |= LOOKUP_REVAL;
4093 goto retry;
4094 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004095out:
Al Viro2d8f3032008-07-22 09:59:21 -04004096 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004097
4098 return error;
4099}
4100
Heiko Carstens3480b252009-01-14 14:14:16 +01004101SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004102{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004103 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004104}
4105
Miklos Szeredibc270272014-04-01 17:08:42 +02004106/**
4107 * vfs_rename - rename a filesystem object
4108 * @old_dir: parent of source
4109 * @old_dentry: source
4110 * @new_dir: parent of destination
4111 * @new_dentry: destination
4112 * @delegated_inode: returns an inode needing a delegation break
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004113 * @flags: rename flags
Miklos Szeredibc270272014-04-01 17:08:42 +02004114 *
4115 * The caller must hold multiple mutexes--see lock_rename()).
4116 *
4117 * If vfs_rename discovers a delegation in need of breaking at either
4118 * the source or destination, it will return -EWOULDBLOCK and return a
4119 * reference to the inode in delegated_inode. The caller should then
4120 * break the delegation and retry. Because breaking a delegation may
4121 * take a long time, the caller should drop all locks before doing
4122 * so.
4123 *
4124 * Alternatively, a caller may pass NULL for delegated_inode. This may
4125 * be appropriate for callers that expect the underlying filesystem not
4126 * to be NFS exported.
4127 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07004128 * The worst of all namespace operations - renaming directory. "Perverted"
4129 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4130 * Problems:
J. Bruce Fieldsd03b29a2014-02-17 16:52:33 -05004131 * a) we can get into loop creation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004132 * b) race potential - two innocent renames can create a loop together.
4133 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004134 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07004135 * story.
J. Bruce Fields6cedba82012-03-05 11:40:41 -05004136 * c) we have to lock _four_ objects - parents and victim (if it exists),
4137 * and source (if it is not a directory).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004138 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07004139 * whether the target exists). Solution: try to be smart with locking
4140 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004141 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07004142 * move will be locked. Thus we can rank directories by the tree
4143 * (ancestors first) and rank all non-directories after them.
4144 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004145 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004146 * HOWEVER, it relies on the assumption that any object with ->lookup()
4147 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4148 * we'd better make sure that there's no link(2) for them.
Sage Weile4eaac02011-05-24 13:06:07 -07004149 * d) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004150 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07004151 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05004152 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07004153 * locking].
4154 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004155int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004156 struct inode *new_dir, struct dentry *new_dentry,
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004157 struct inode **delegated_inode, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004158{
4159 int error;
Miklos Szeredibc270272014-04-01 17:08:42 +02004160 bool is_dir = d_is_dir(old_dentry);
Eric Paris59b0df22010-02-08 12:53:52 -05004161 const unsigned char *old_name;
Miklos Szeredibc270272014-04-01 17:08:42 +02004162 struct inode *source = old_dentry->d_inode;
4163 struct inode *target = new_dentry->d_inode;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004164 bool new_is_dir = false;
4165 unsigned max_links = new_dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004166
Miklos Szeredibc270272014-04-01 17:08:42 +02004167 if (source == target)
4168 return 0;
4169
Linus Torvalds1da177e2005-04-16 15:20:36 -07004170 error = may_delete(old_dir, old_dentry, is_dir);
4171 if (error)
4172 return error;
4173
Miklos Szeredida1ce062014-04-01 17:08:43 +02004174 if (!target) {
Miklos Szeredia95164d2008-07-30 15:08:48 +02004175 error = may_create(new_dir, new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004176 } else {
4177 new_is_dir = d_is_dir(new_dentry);
4178
4179 if (!(flags & RENAME_EXCHANGE))
4180 error = may_delete(new_dir, new_dentry, is_dir);
4181 else
4182 error = may_delete(new_dir, new_dentry, new_is_dir);
4183 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004184 if (error)
4185 return error;
4186
Miklos Szeredi7177a9c2014-07-23 15:15:30 +02004187 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004188 return -EPERM;
4189
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004190 if (flags && !old_dir->i_op->rename2)
4191 return -EINVAL;
4192
Miklos Szeredibc270272014-04-01 17:08:42 +02004193 /*
4194 * If we are going to change the parent - check write permissions,
4195 * we'll need to flip '..'.
4196 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02004197 if (new_dir != old_dir) {
4198 if (is_dir) {
4199 error = inode_permission(source, MAY_WRITE);
4200 if (error)
4201 return error;
4202 }
4203 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4204 error = inode_permission(target, MAY_WRITE);
4205 if (error)
4206 return error;
4207 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004208 }
Robert Love0eeca282005-07-12 17:06:03 -04004209
Miklos Szeredi0b3974e2014-04-01 17:08:43 +02004210 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4211 flags);
Miklos Szeredibc270272014-04-01 17:08:42 +02004212 if (error)
4213 return error;
4214
4215 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4216 dget(new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004217 if (!is_dir || (flags & RENAME_EXCHANGE))
Miklos Szeredibc270272014-04-01 17:08:42 +02004218 lock_two_nondirectories(source, target);
4219 else if (target)
4220 mutex_lock(&target->i_mutex);
4221
4222 error = -EBUSY;
Eric W. Biederman7af13642013-10-04 19:15:13 -07004223 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
Miklos Szeredibc270272014-04-01 17:08:42 +02004224 goto out;
4225
Miklos Szeredida1ce062014-04-01 17:08:43 +02004226 if (max_links && new_dir != old_dir) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004227 error = -EMLINK;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004228 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
Miklos Szeredibc270272014-04-01 17:08:42 +02004229 goto out;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004230 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4231 old_dir->i_nlink >= max_links)
4232 goto out;
4233 }
4234 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4235 shrink_dcache_parent(new_dentry);
4236 if (!is_dir) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004237 error = try_break_deleg(source, delegated_inode);
4238 if (error)
4239 goto out;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004240 }
4241 if (target && !new_is_dir) {
4242 error = try_break_deleg(target, delegated_inode);
4243 if (error)
4244 goto out;
Miklos Szeredibc270272014-04-01 17:08:42 +02004245 }
Miklos Szeredi7177a9c2014-07-23 15:15:30 +02004246 if (!old_dir->i_op->rename2) {
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004247 error = old_dir->i_op->rename(old_dir, old_dentry,
4248 new_dir, new_dentry);
4249 } else {
Miklos Szeredi7177a9c2014-07-23 15:15:30 +02004250 WARN_ON(old_dir->i_op->rename != NULL);
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004251 error = old_dir->i_op->rename2(old_dir, old_dentry,
4252 new_dir, new_dentry, flags);
4253 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004254 if (error)
4255 goto out;
4256
Miklos Szeredida1ce062014-04-01 17:08:43 +02004257 if (!(flags & RENAME_EXCHANGE) && target) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004258 if (is_dir)
4259 target->i_flags |= S_DEAD;
4260 dont_mount(new_dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07004261 detach_mounts(new_dentry);
Miklos Szeredibc270272014-04-01 17:08:42 +02004262 }
Miklos Szeredida1ce062014-04-01 17:08:43 +02004263 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4264 if (!(flags & RENAME_EXCHANGE))
4265 d_move(old_dentry, new_dentry);
4266 else
4267 d_exchange(old_dentry, new_dentry);
4268 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004269out:
Miklos Szeredida1ce062014-04-01 17:08:43 +02004270 if (!is_dir || (flags & RENAME_EXCHANGE))
Miklos Szeredibc270272014-04-01 17:08:42 +02004271 unlock_two_nondirectories(source, target);
4272 else if (target)
4273 mutex_unlock(&target->i_mutex);
4274 dput(new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004275 if (!error) {
Al Viro123df292009-12-25 04:57:57 -05004276 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Miklos Szeredida1ce062014-04-01 17:08:43 +02004277 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4278 if (flags & RENAME_EXCHANGE) {
4279 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4280 new_is_dir, NULL, new_dentry);
4281 }
4282 }
Robert Love0eeca282005-07-12 17:06:03 -04004283 fsnotify_oldname_free(old_name);
4284
Linus Torvalds1da177e2005-04-16 15:20:36 -07004285 return error;
4286}
Al Viro4d359502014-03-14 12:20:17 -04004287EXPORT_SYMBOL(vfs_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004288
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004289SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4290 int, newdfd, const char __user *, newname, unsigned int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004291{
Al Viro2ad94ae2008-07-21 09:32:51 -04004292 struct dentry *old_dentry, *new_dentry;
4293 struct dentry *trap;
Al Virof5beed72015-04-30 16:09:11 -04004294 struct path old_path, new_path;
4295 struct qstr old_last, new_last;
4296 int old_type, new_type;
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004297 struct inode *delegated_inode = NULL;
Jeff Layton91a27b22012-10-10 15:25:28 -04004298 struct filename *from;
4299 struct filename *to;
Al Virof5beed72015-04-30 16:09:11 -04004300 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
Jeff Laytonc6a94282012-12-11 12:10:10 -05004301 bool should_retry = false;
Al Viro2ad94ae2008-07-21 09:32:51 -04004302 int error;
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004303
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004304 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredida1ce062014-04-01 17:08:43 +02004305 return -EINVAL;
4306
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004307 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4308 (flags & RENAME_EXCHANGE))
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004309 return -EINVAL;
4310
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004311 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4312 return -EPERM;
4313
Al Virof5beed72015-04-30 16:09:11 -04004314 if (flags & RENAME_EXCHANGE)
4315 target_flags = 0;
4316
Jeff Laytonc6a94282012-12-11 12:10:10 -05004317retry:
Al Virof5beed72015-04-30 16:09:11 -04004318 from = user_path_parent(olddfd, oldname,
4319 &old_path, &old_last, &old_type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04004320 if (IS_ERR(from)) {
4321 error = PTR_ERR(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004322 goto exit;
Jeff Layton91a27b22012-10-10 15:25:28 -04004323 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004324
Al Virof5beed72015-04-30 16:09:11 -04004325 to = user_path_parent(newdfd, newname,
4326 &new_path, &new_last, &new_type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04004327 if (IS_ERR(to)) {
4328 error = PTR_ERR(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004329 goto exit1;
Jeff Layton91a27b22012-10-10 15:25:28 -04004330 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004331
4332 error = -EXDEV;
Al Virof5beed72015-04-30 16:09:11 -04004333 if (old_path.mnt != new_path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004334 goto exit2;
4335
Linus Torvalds1da177e2005-04-16 15:20:36 -07004336 error = -EBUSY;
Al Virof5beed72015-04-30 16:09:11 -04004337 if (old_type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004338 goto exit2;
4339
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004340 if (flags & RENAME_NOREPLACE)
4341 error = -EEXIST;
Al Virof5beed72015-04-30 16:09:11 -04004342 if (new_type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004343 goto exit2;
4344
Al Virof5beed72015-04-30 16:09:11 -04004345 error = mnt_want_write(old_path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02004346 if (error)
4347 goto exit2;
4348
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004349retry_deleg:
Al Virof5beed72015-04-30 16:09:11 -04004350 trap = lock_rename(new_path.dentry, old_path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004351
Al Virof5beed72015-04-30 16:09:11 -04004352 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004353 error = PTR_ERR(old_dentry);
4354 if (IS_ERR(old_dentry))
4355 goto exit3;
4356 /* source must exist */
4357 error = -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01004358 if (d_is_negative(old_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004359 goto exit4;
Al Virof5beed72015-04-30 16:09:11 -04004360 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004361 error = PTR_ERR(new_dentry);
4362 if (IS_ERR(new_dentry))
4363 goto exit4;
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004364 error = -EEXIST;
4365 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4366 goto exit5;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004367 if (flags & RENAME_EXCHANGE) {
4368 error = -ENOENT;
4369 if (d_is_negative(new_dentry))
4370 goto exit5;
4371
4372 if (!d_is_dir(new_dentry)) {
4373 error = -ENOTDIR;
Al Virof5beed72015-04-30 16:09:11 -04004374 if (new_last.name[new_last.len])
Miklos Szeredida1ce062014-04-01 17:08:43 +02004375 goto exit5;
4376 }
4377 }
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004378 /* unless the source is a directory trailing slashes give -ENOTDIR */
4379 if (!d_is_dir(old_dentry)) {
4380 error = -ENOTDIR;
Al Virof5beed72015-04-30 16:09:11 -04004381 if (old_last.name[old_last.len])
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004382 goto exit5;
Al Virof5beed72015-04-30 16:09:11 -04004383 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004384 goto exit5;
4385 }
4386 /* source should not be ancestor of target */
4387 error = -EINVAL;
4388 if (old_dentry == trap)
4389 goto exit5;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004390 /* target should not be an ancestor of source */
Miklos Szeredida1ce062014-04-01 17:08:43 +02004391 if (!(flags & RENAME_EXCHANGE))
4392 error = -ENOTEMPTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004393 if (new_dentry == trap)
4394 goto exit5;
4395
Al Virof5beed72015-04-30 16:09:11 -04004396 error = security_path_rename(&old_path, old_dentry,
4397 &new_path, new_dentry, flags);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004398 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02004399 goto exit5;
Al Virof5beed72015-04-30 16:09:11 -04004400 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4401 new_path.dentry->d_inode, new_dentry,
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004402 &delegated_inode, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004403exit5:
4404 dput(new_dentry);
4405exit4:
4406 dput(old_dentry);
4407exit3:
Al Virof5beed72015-04-30 16:09:11 -04004408 unlock_rename(new_path.dentry, old_path.dentry);
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004409 if (delegated_inode) {
4410 error = break_deleg_wait(&delegated_inode);
4411 if (!error)
4412 goto retry_deleg;
4413 }
Al Virof5beed72015-04-30 16:09:11 -04004414 mnt_drop_write(old_path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004415exit2:
Jeff Laytonc6a94282012-12-11 12:10:10 -05004416 if (retry_estale(error, lookup_flags))
4417 should_retry = true;
Al Virof5beed72015-04-30 16:09:11 -04004418 path_put(&new_path);
Al Viro2ad94ae2008-07-21 09:32:51 -04004419 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004420exit1:
Al Virof5beed72015-04-30 16:09:11 -04004421 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004422 putname(from);
Jeff Laytonc6a94282012-12-11 12:10:10 -05004423 if (should_retry) {
4424 should_retry = false;
4425 lookup_flags |= LOOKUP_REVAL;
4426 goto retry;
4427 }
Al Viro2ad94ae2008-07-21 09:32:51 -04004428exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004429 return error;
4430}
4431
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004432SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4433 int, newdfd, const char __user *, newname)
4434{
4435 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4436}
4437
Heiko Carstensa26eab22009-01-14 14:14:17 +01004438SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004439{
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004440 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004441}
4442
Miklos Szeredi787fb6b2014-10-24 00:14:36 +02004443int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4444{
4445 int error = may_create(dir, dentry);
4446 if (error)
4447 return error;
4448
4449 if (!dir->i_op->mknod)
4450 return -EPERM;
4451
4452 return dir->i_op->mknod(dir, dentry,
4453 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4454}
4455EXPORT_SYMBOL(vfs_whiteout);
4456
Al Viro5d826c82014-03-14 13:42:45 -04004457int readlink_copy(char __user *buffer, int buflen, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004458{
Al Viro5d826c82014-03-14 13:42:45 -04004459 int len = PTR_ERR(link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004460 if (IS_ERR(link))
4461 goto out;
4462
4463 len = strlen(link);
4464 if (len > (unsigned) buflen)
4465 len = buflen;
4466 if (copy_to_user(buffer, link, len))
4467 len = -EFAULT;
4468out:
4469 return len;
4470}
Al Viro5d826c82014-03-14 13:42:45 -04004471EXPORT_SYMBOL(readlink_copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004472
4473/*
4474 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4475 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4476 * using) it for any given inode is up to filesystem.
4477 */
4478int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4479{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004480 void *cookie;
Al Viro5f2c4172015-05-07 11:14:26 -04004481 struct inode *inode = d_inode(dentry);
4482 const char *link = inode->i_link;
Marcin Slusarz694a1762008-06-09 16:40:37 -07004483 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004484
Al Virod4dee482015-04-30 20:08:02 -04004485 if (!link) {
Al Viro5f2c4172015-05-07 11:14:26 -04004486 link = inode->i_op->follow_link(dentry, &cookie);
Al Virod4dee482015-04-30 20:08:02 -04004487 if (IS_ERR(link))
4488 return PTR_ERR(link);
4489 }
Al Viro680baac2015-05-02 13:32:22 -04004490 res = readlink_copy(buffer, buflen, link);
Al Viro5f2c4172015-05-07 11:14:26 -04004491 if (inode->i_op->put_link)
4492 inode->i_op->put_link(inode, cookie);
Marcin Slusarz694a1762008-06-09 16:40:37 -07004493 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004494}
Al Viro4d359502014-03-14 12:20:17 -04004495EXPORT_SYMBOL(generic_readlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004496
Linus Torvalds1da177e2005-04-16 15:20:36 -07004497/* get the link contents into pagecache */
4498static char *page_getlink(struct dentry * dentry, struct page **ppage)
4499{
Duane Griffinebd09ab2008-12-19 20:47:12 +00004500 char *kaddr;
4501 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004502 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07004503 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004504 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07004505 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004506 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00004507 kaddr = kmap(page);
4508 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4509 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004510}
4511
4512int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4513{
4514 struct page *page = NULL;
Al Viro5d826c82014-03-14 13:42:45 -04004515 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004516 if (page) {
4517 kunmap(page);
4518 page_cache_release(page);
4519 }
4520 return res;
4521}
Al Viro4d359502014-03-14 12:20:17 -04004522EXPORT_SYMBOL(page_readlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004523
Al Viro6e771372015-05-02 13:37:52 -04004524const char *page_follow_link_light(struct dentry *dentry, void **cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004525{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004526 struct page *page = NULL;
Al Viro680baac2015-05-02 13:32:22 -04004527 char *res = page_getlink(dentry, &page);
4528 if (!IS_ERR(res))
4529 *cookie = page;
4530 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004531}
Al Viro4d359502014-03-14 12:20:17 -04004532EXPORT_SYMBOL(page_follow_link_light);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004533
Al Viro5f2c4172015-05-07 11:14:26 -04004534void page_put_link(struct inode *unused, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004535{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004536 struct page *page = cookie;
Al Viro680baac2015-05-02 13:32:22 -04004537 kunmap(page);
4538 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004539}
Al Viro4d359502014-03-14 12:20:17 -04004540EXPORT_SYMBOL(page_put_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004541
Nick Piggin54566b22009-01-04 12:00:53 -08004542/*
4543 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4544 */
4545int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004546{
4547 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004548 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07004549 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08004550 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004551 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08004552 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4553 if (nofs)
4554 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004555
NeilBrown7e53cac2006-03-25 03:07:57 -08004556retry:
Nick Pigginafddba42007-10-16 01:25:01 -07004557 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08004558 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004559 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07004560 goto fail;
4561
Cong Wange8e3c3d2011-11-25 23:14:27 +08004562 kaddr = kmap_atomic(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004563 memcpy(kaddr, symname, len-1);
Cong Wange8e3c3d2011-11-25 23:14:27 +08004564 kunmap_atomic(kaddr);
Nick Pigginafddba42007-10-16 01:25:01 -07004565
4566 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4567 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004568 if (err < 0)
4569 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07004570 if (err < len-1)
4571 goto retry;
4572
Linus Torvalds1da177e2005-04-16 15:20:36 -07004573 mark_inode_dirty(inode);
4574 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004575fail:
4576 return err;
4577}
Al Viro4d359502014-03-14 12:20:17 -04004578EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004579
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004580int page_symlink(struct inode *inode, const char *symname, int len)
4581{
4582 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08004583 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004584}
Al Viro4d359502014-03-14 12:20:17 -04004585EXPORT_SYMBOL(page_symlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004586
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08004587const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004588 .readlink = generic_readlink,
4589 .follow_link = page_follow_link_light,
4590 .put_link = page_put_link,
4591};
Linus Torvalds1da177e2005-04-16 15:20:36 -07004592EXPORT_SYMBOL(page_symlink_inode_operations);