blob: 33e9495a31293e2c080b5b0bd2e50523a460ceee [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
Eric W. Biederman397d4252015-08-15 20:27:13 -0500563/**
564 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
565 * @path: nameidate to verify
566 *
567 * Rename can sometimes move a file or directory outside of a bind
568 * mount, path_connected allows those cases to be detected.
569 */
570static bool path_connected(const struct path *path)
571{
572 struct vfsmount *mnt = path->mnt;
573
574 /* Only bind mounts can have disconnected paths */
575 if (mnt->mnt_root == mnt->mnt_sb->s_root)
576 return true;
577
578 return is_subdir(path->dentry, mnt->mnt_root);
579}
580
Al Viro894bc8c2015-05-02 07:16:16 -0400581static inline int nd_alloc_stack(struct nameidata *nd)
582{
Al Viroda4e0be2015-05-03 20:52:15 -0400583 if (likely(nd->depth != EMBEDDED_LEVELS))
Al Viro894bc8c2015-05-02 07:16:16 -0400584 return 0;
585 if (likely(nd->stack != nd->internal))
586 return 0;
587 return __nd_alloc_stack(nd);
588}
589
Al Viro79733872015-05-09 12:55:43 -0400590static void drop_links(struct nameidata *nd)
591{
592 int i = nd->depth;
593 while (i--) {
594 struct saved *last = nd->stack + i;
595 struct inode *inode = last->inode;
596 if (last->cookie && inode->i_op->put_link) {
597 inode->i_op->put_link(inode, last->cookie);
598 last->cookie = NULL;
599 }
600 }
601}
602
603static void terminate_walk(struct nameidata *nd)
604{
605 drop_links(nd);
606 if (!(nd->flags & LOOKUP_RCU)) {
607 int i;
608 path_put(&nd->path);
609 for (i = 0; i < nd->depth; i++)
610 path_put(&nd->stack[i].link);
Al Viro102b8af2015-05-12 17:35:52 -0400611 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
612 path_put(&nd->root);
613 nd->root.mnt = NULL;
614 }
Al Viro79733872015-05-09 12:55:43 -0400615 } else {
616 nd->flags &= ~LOOKUP_RCU;
617 if (!(nd->flags & LOOKUP_ROOT))
618 nd->root.mnt = NULL;
619 rcu_read_unlock();
620 }
621 nd->depth = 0;
622}
623
624/* path_put is needed afterwards regardless of success or failure */
625static bool legitimize_path(struct nameidata *nd,
626 struct path *path, unsigned seq)
627{
628 int res = __legitimize_mnt(path->mnt, nd->m_seq);
629 if (unlikely(res)) {
630 if (res > 0)
631 path->mnt = NULL;
632 path->dentry = NULL;
633 return false;
634 }
635 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
636 path->dentry = NULL;
637 return false;
638 }
639 return !read_seqcount_retry(&path->dentry->d_seq, seq);
640}
641
642static bool legitimize_links(struct nameidata *nd)
643{
644 int i;
645 for (i = 0; i < nd->depth; i++) {
646 struct saved *last = nd->stack + i;
647 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
648 drop_links(nd);
649 nd->depth = i + 1;
650 return false;
651 }
652 }
653 return true;
654}
655
Al Viro19660af2011-03-25 10:32:48 -0400656/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100657 * Path walking has 2 modes, rcu-walk and ref-walk (see
Al Viro19660af2011-03-25 10:32:48 -0400658 * Documentation/filesystems/path-lookup.txt). In situations when we can't
659 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
660 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
661 * mode. Refcounts are grabbed at the last known good point before rcu-walk
662 * got stuck, so ref-walk may continue from there. If this is not successful
663 * (eg. a seqcount has changed), then failure is returned and it's up to caller
664 * to restart the path walk from the beginning in ref-walk mode.
Nick Piggin31e6b012011-01-07 17:49:52 +1100665 */
Nick Piggin31e6b012011-01-07 17:49:52 +1100666
667/**
Al Viro19660af2011-03-25 10:32:48 -0400668 * unlazy_walk - try to switch to ref-walk mode.
669 * @nd: nameidata pathwalk data
670 * @dentry: child of nd->path.dentry or NULL
Al Viro6e9918b2015-05-05 09:26:05 -0400671 * @seq: seq number to check dentry against
Randy Dunlap39191622011-01-08 19:36:21 -0800672 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100673 *
Al Viro19660af2011-03-25 10:32:48 -0400674 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
675 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
676 * @nd or NULL. Must be called from rcu-walk context.
Al Viro79733872015-05-09 12:55:43 -0400677 * Nothing should touch nameidata between unlazy_walk() failure and
678 * terminate_walk().
Nick Piggin31e6b012011-01-07 17:49:52 +1100679 */
Al Viro6e9918b2015-05-05 09:26:05 -0400680static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq)
Nick Piggin31e6b012011-01-07 17:49:52 +1100681{
Nick Piggin31e6b012011-01-07 17:49:52 +1100682 struct dentry *parent = nd->path.dentry;
683
684 BUG_ON(!(nd->flags & LOOKUP_RCU));
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700685
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700686 nd->flags &= ~LOOKUP_RCU;
Al Viro79733872015-05-09 12:55:43 -0400687 if (unlikely(!legitimize_links(nd)))
688 goto out2;
689 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
690 goto out2;
691 if (unlikely(!lockref_get_not_dead(&parent->d_lockref)))
692 goto out1;
Al Viro48a066e2013-09-29 22:06:07 -0400693
Linus Torvalds15570082013-09-02 11:38:06 -0700694 /*
695 * For a negative lookup, the lookup sequence point is the parents
696 * sequence point, and it only needs to revalidate the parent dentry.
697 *
698 * For a positive lookup, we need to move both the parent and the
699 * dentry from the RCU domain to be properly refcounted. And the
700 * sequence number in the dentry validates *both* dentry counters,
701 * since we checked the sequence number of the parent after we got
702 * the child sequence number. So we know the parent must still
703 * be valid if the child sequence number is still valid.
704 */
Al Viro19660af2011-03-25 10:32:48 -0400705 if (!dentry) {
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700706 if (read_seqcount_retry(&parent->d_seq, nd->seq))
707 goto out;
Al Viro19660af2011-03-25 10:32:48 -0400708 BUG_ON(nd->inode != parent->d_inode);
709 } else {
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700710 if (!lockref_get_not_dead(&dentry->d_lockref))
711 goto out;
Al Viro6e9918b2015-05-05 09:26:05 -0400712 if (read_seqcount_retry(&dentry->d_seq, seq))
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700713 goto drop_dentry;
Al Viro19660af2011-03-25 10:32:48 -0400714 }
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700715
716 /*
717 * Sequence counts matched. Now make sure that the root is
718 * still valid and get it if required.
719 */
720 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
Al Viro5a8d87e2015-05-12 00:10:18 -0400721 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
722 rcu_read_unlock();
723 dput(dentry);
724 return -ECHILD;
Al Viro79733872015-05-09 12:55:43 -0400725 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100726 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100727
Al Viro8b61e742013-11-08 12:45:01 -0500728 rcu_read_unlock();
Nick Piggin31e6b012011-01-07 17:49:52 +1100729 return 0;
Al Viro19660af2011-03-25 10:32:48 -0400730
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700731drop_dentry:
Al Viro8b61e742013-11-08 12:45:01 -0500732 rcu_read_unlock();
Linus Torvalds15570082013-09-02 11:38:06 -0700733 dput(dentry);
Linus Torvaldsd0d27272013-09-10 12:17:49 -0700734 goto drop_root_mnt;
Al Viro79733872015-05-09 12:55:43 -0400735out2:
736 nd->path.mnt = NULL;
737out1:
738 nd->path.dentry = NULL;
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700739out:
Al Viro8b61e742013-11-08 12:45:01 -0500740 rcu_read_unlock();
Linus Torvaldsd0d27272013-09-10 12:17:49 -0700741drop_root_mnt:
742 if (!(nd->flags & LOOKUP_ROOT))
743 nd->root.mnt = NULL;
Nick Piggin31e6b012011-01-07 17:49:52 +1100744 return -ECHILD;
745}
746
Al Viro79733872015-05-09 12:55:43 -0400747static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq)
748{
749 if (unlikely(!legitimize_path(nd, link, seq))) {
750 drop_links(nd);
751 nd->depth = 0;
752 nd->flags &= ~LOOKUP_RCU;
753 nd->path.mnt = NULL;
754 nd->path.dentry = NULL;
755 if (!(nd->flags & LOOKUP_ROOT))
756 nd->root.mnt = NULL;
757 rcu_read_unlock();
758 } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) {
759 return 0;
760 }
761 path_put(link);
762 return -ECHILD;
763}
764
Al Viro4ce16ef32012-06-10 16:10:59 -0400765static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
Nick Piggin34286d62011-01-07 17:49:57 +1100766{
Al Viro4ce16ef32012-06-10 16:10:59 -0400767 return dentry->d_op->d_revalidate(dentry, flags);
Nick Piggin34286d62011-01-07 17:49:57 +1100768}
769
Al Viro9f1fafe2011-03-25 11:00:12 -0400770/**
771 * complete_walk - successful completion of path walk
772 * @nd: pointer nameidata
Jeff Layton39159de2009-12-07 12:01:50 -0500773 *
Al Viro9f1fafe2011-03-25 11:00:12 -0400774 * If we had been in RCU mode, drop out of it and legitimize nd->path.
775 * Revalidate the final result, unless we'd already done that during
776 * the path walk or the filesystem doesn't ask for it. Return 0 on
777 * success, -error on failure. In case of failure caller does not
778 * need to drop nd->path.
Jeff Layton39159de2009-12-07 12:01:50 -0500779 */
Al Viro9f1fafe2011-03-25 11:00:12 -0400780static int complete_walk(struct nameidata *nd)
Jeff Layton39159de2009-12-07 12:01:50 -0500781{
Al Viro16c2cd72011-02-22 15:50:10 -0500782 struct dentry *dentry = nd->path.dentry;
Jeff Layton39159de2009-12-07 12:01:50 -0500783 int status;
Jeff Layton39159de2009-12-07 12:01:50 -0500784
Al Viro9f1fafe2011-03-25 11:00:12 -0400785 if (nd->flags & LOOKUP_RCU) {
Al Viro9f1fafe2011-03-25 11:00:12 -0400786 if (!(nd->flags & LOOKUP_ROOT))
787 nd->root.mnt = NULL;
Al Viro6e9918b2015-05-05 09:26:05 -0400788 if (unlikely(unlazy_walk(nd, NULL, 0)))
Al Viro48a066e2013-09-29 22:06:07 -0400789 return -ECHILD;
Al Viro9f1fafe2011-03-25 11:00:12 -0400790 }
791
Al Viro16c2cd72011-02-22 15:50:10 -0500792 if (likely(!(nd->flags & LOOKUP_JUMPED)))
Jeff Layton39159de2009-12-07 12:01:50 -0500793 return 0;
794
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500795 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
Al Viro16c2cd72011-02-22 15:50:10 -0500796 return 0;
797
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500798 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
Jeff Layton39159de2009-12-07 12:01:50 -0500799 if (status > 0)
800 return 0;
801
Al Viro16c2cd72011-02-22 15:50:10 -0500802 if (!status)
Jeff Layton39159de2009-12-07 12:01:50 -0500803 status = -ESTALE;
Al Viro16c2cd72011-02-22 15:50:10 -0500804
Jeff Layton39159de2009-12-07 12:01:50 -0500805 return status;
806}
807
Al Viro18d8c862015-05-12 16:32:34 -0400808static void set_root(struct nameidata *nd)
Al Viro2a737872009-04-07 11:49:53 -0400809{
Al Viro7bd88372014-09-13 21:55:46 -0400810 get_fs_root(current->fs, &nd->root);
Al Viro2a737872009-04-07 11:49:53 -0400811}
812
Al Viro06d71372015-06-29 12:07:04 -0400813static void set_root_rcu(struct nameidata *nd)
Nick Piggin31e6b012011-01-07 17:49:52 +1100814{
Al Viro7bd88372014-09-13 21:55:46 -0400815 struct fs_struct *fs = current->fs;
Al Viro8f47a0162015-05-09 19:02:01 -0400816 unsigned seq;
Nick Pigginc28cc362011-01-07 17:49:53 +1100817
Al Viro7bd88372014-09-13 21:55:46 -0400818 do {
819 seq = read_seqcount_begin(&fs->seq);
820 nd->root = fs->root;
Al Viro8f47a0162015-05-09 19:02:01 -0400821 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
Al Viro7bd88372014-09-13 21:55:46 -0400822 } while (read_seqcount_retry(&fs->seq, seq));
Nick Piggin31e6b012011-01-07 17:49:52 +1100823}
824
Jan Blunck1d957f92008-02-14 19:34:35 -0800825static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700826{
827 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800828 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700829 mntput(path->mnt);
830}
831
Nick Piggin7b9337a2011-01-14 08:42:43 +0000832static inline void path_to_nameidata(const struct path *path,
833 struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700834{
Nick Piggin31e6b012011-01-07 17:49:52 +1100835 if (!(nd->flags & LOOKUP_RCU)) {
836 dput(nd->path.dentry);
837 if (nd->path.mnt != path->mnt)
838 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800839 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100840 nd->path.mnt = path->mnt;
Jan Blunck4ac91372008-02-14 19:34:32 -0800841 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700842}
843
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400844/*
845 * Helper to directly jump to a known parsed path from ->follow_link,
846 * caller must have taken a reference to path beforehand.
847 */
Al Viro6e771372015-05-02 13:37:52 -0400848void nd_jump_link(struct path *path)
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400849{
Al Viro6e771372015-05-02 13:37:52 -0400850 struct nameidata *nd = current->nameidata;
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400851 path_put(&nd->path);
852
853 nd->path = *path;
854 nd->inode = nd->path.dentry->d_inode;
855 nd->flags |= LOOKUP_JUMPED;
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400856}
857
Al Virob9ff4422015-05-02 20:19:23 -0400858static inline void put_link(struct nameidata *nd)
Al Viro574197e2011-03-14 22:20:34 -0400859{
Al Viro21c30032015-05-03 21:06:24 -0400860 struct saved *last = nd->stack + --nd->depth;
Al Viro237d8b32015-05-07 09:21:14 -0400861 struct inode *inode = last->inode;
Al Virob9ff4422015-05-02 20:19:23 -0400862 if (last->cookie && inode->i_op->put_link)
Al Viro5f2c4172015-05-07 11:14:26 -0400863 inode->i_op->put_link(inode, last->cookie);
Al Viro6548fae2015-05-07 20:32:22 -0400864 if (!(nd->flags & LOOKUP_RCU))
865 path_put(&last->link);
Al Viro574197e2011-03-14 22:20:34 -0400866}
867
Linus Torvalds561ec642012-10-26 10:05:07 -0700868int sysctl_protected_symlinks __read_mostly = 0;
869int sysctl_protected_hardlinks __read_mostly = 0;
Kees Cook800179c2012-07-25 17:29:07 -0700870
871/**
872 * may_follow_link - Check symlink following for unsafe situations
Randy Dunlap55852632012-08-18 17:39:25 -0700873 * @nd: nameidata pathwalk data
Kees Cook800179c2012-07-25 17:29:07 -0700874 *
875 * In the case of the sysctl_protected_symlinks sysctl being enabled,
876 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
877 * in a sticky world-writable directory. This is to protect privileged
878 * processes from failing races against path names that may change out
879 * from under them by way of other users creating malicious symlinks.
880 * It will permit symlinks to be followed only when outside a sticky
881 * world-writable directory, or when the uid of the symlink and follower
882 * match, or when the directory owner matches the symlink's owner.
883 *
884 * Returns 0 if following the symlink is allowed, -ve on error.
885 */
Al Virofec2fa22015-05-06 15:58:18 -0400886static inline int may_follow_link(struct nameidata *nd)
Kees Cook800179c2012-07-25 17:29:07 -0700887{
888 const struct inode *inode;
889 const struct inode *parent;
890
891 if (!sysctl_protected_symlinks)
892 return 0;
893
894 /* Allowed if owner and follower match. */
Al Viro237d8b32015-05-07 09:21:14 -0400895 inode = nd->stack[0].inode;
Eric W. Biederman81abe272012-08-03 09:38:08 -0700896 if (uid_eq(current_cred()->fsuid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700897 return 0;
898
899 /* Allowed if parent directory not sticky and world-writable. */
Al Viroaa65fa32015-08-04 23:23:50 -0400900 parent = nd->inode;
Kees Cook800179c2012-07-25 17:29:07 -0700901 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
902 return 0;
903
904 /* Allowed if parent directory and link owner match. */
Eric W. Biederman81abe272012-08-03 09:38:08 -0700905 if (uid_eq(parent->i_uid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700906 return 0;
907
Al Viro319565022015-05-07 20:37:40 -0400908 if (nd->flags & LOOKUP_RCU)
909 return -ECHILD;
910
Al Viro1cf26652015-05-06 16:01:56 -0400911 audit_log_link_denied("follow_link", &nd->stack[0].link);
Kees Cook800179c2012-07-25 17:29:07 -0700912 return -EACCES;
913}
914
915/**
916 * safe_hardlink_source - Check for safe hardlink conditions
917 * @inode: the source inode to hardlink from
918 *
919 * Return false if at least one of the following conditions:
920 * - inode is not a regular file
921 * - inode is setuid
922 * - inode is setgid and group-exec
923 * - access failure for read and write
924 *
925 * Otherwise returns true.
926 */
927static bool safe_hardlink_source(struct inode *inode)
928{
929 umode_t mode = inode->i_mode;
930
931 /* Special files should not get pinned to the filesystem. */
932 if (!S_ISREG(mode))
933 return false;
934
935 /* Setuid files should not get pinned to the filesystem. */
936 if (mode & S_ISUID)
937 return false;
938
939 /* Executable setgid files should not get pinned to the filesystem. */
940 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
941 return false;
942
943 /* Hardlinking to unreadable or unwritable sources is dangerous. */
944 if (inode_permission(inode, MAY_READ | MAY_WRITE))
945 return false;
946
947 return true;
948}
949
950/**
951 * may_linkat - Check permissions for creating a hardlink
952 * @link: the source to hardlink from
953 *
954 * Block hardlink when all of:
955 * - sysctl_protected_hardlinks enabled
956 * - fsuid does not match inode
957 * - hardlink source is unsafe (see safe_hardlink_source() above)
958 * - not CAP_FOWNER
959 *
960 * Returns 0 if successful, -ve on error.
961 */
962static int may_linkat(struct path *link)
963{
964 const struct cred *cred;
965 struct inode *inode;
966
967 if (!sysctl_protected_hardlinks)
968 return 0;
969
970 cred = current_cred();
971 inode = link->dentry->d_inode;
972
973 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
974 * otherwise, it must be a safe source.
975 */
Eric W. Biederman81abe272012-08-03 09:38:08 -0700976 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
Kees Cook800179c2012-07-25 17:29:07 -0700977 capable(CAP_FOWNER))
978 return 0;
979
Kees Cooka51d9ea2012-07-25 17:29:08 -0700980 audit_log_link_denied("linkat", link);
Kees Cook800179c2012-07-25 17:29:07 -0700981 return -EPERM;
982}
983
Al Viro3b2e7f72015-04-19 00:53:50 -0400984static __always_inline
985const char *get_link(struct nameidata *nd)
Ian Kent051d3812006-03-27 01:14:53 -0800986{
Al Viroab104922015-05-10 11:50:01 -0400987 struct saved *last = nd->stack + nd->depth - 1;
Al Viro1cf26652015-05-06 16:01:56 -0400988 struct dentry *dentry = last->link.dentry;
Al Viro237d8b32015-05-07 09:21:14 -0400989 struct inode *inode = last->inode;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400990 int error;
Al Viro0a959df2015-04-18 18:23:41 -0400991 const char *res;
Ian Kent051d3812006-03-27 01:14:53 -0800992
NeilBrown8fa9dd22015-03-23 13:37:40 +1100993 if (!(nd->flags & LOOKUP_RCU)) {
994 touch_atime(&last->link);
995 cond_resched();
996 } else if (atime_needs_update(&last->link, inode)) {
997 if (unlikely(unlazy_walk(nd, NULL, 0)))
998 return ERR_PTR(-ECHILD);
999 touch_atime(&last->link);
1000 }
1001
NeilBrownbda0be72015-03-23 13:37:39 +11001002 error = security_inode_follow_link(dentry, inode,
1003 nd->flags & LOOKUP_RCU);
1004 if (unlikely(error))
Al Viro6920a442015-05-10 10:43:46 -04001005 return ERR_PTR(error);
Al Viro36f3b4f2011-02-22 21:24:38 -05001006
Al Viro86acdca12009-12-22 23:45:11 -05001007 nd->last_type = LAST_BIND;
Al Virod4dee482015-04-30 20:08:02 -04001008 res = inode->i_link;
1009 if (!res) {
Al Viro8c1b4562015-05-09 18:15:21 -04001010 if (nd->flags & LOOKUP_RCU) {
1011 if (unlikely(unlazy_walk(nd, NULL, 0)))
1012 return ERR_PTR(-ECHILD);
1013 }
Al Viro6e771372015-05-02 13:37:52 -04001014 res = inode->i_op->follow_link(dentry, &last->cookie);
Al Virofab51e82015-05-10 11:01:00 -04001015 if (IS_ERR_OR_NULL(res)) {
Al Viro6920a442015-05-10 10:43:46 -04001016 last->cookie = NULL;
Al Virofab51e82015-05-10 11:01:00 -04001017 return res;
1018 }
Ian Kent051d3812006-03-27 01:14:53 -08001019 }
Al Virofab51e82015-05-10 11:01:00 -04001020 if (*res == '/') {
Al Viro8c1b4562015-05-09 18:15:21 -04001021 if (nd->flags & LOOKUP_RCU) {
Al Viro8f47a0162015-05-09 19:02:01 -04001022 struct dentry *d;
1023 if (!nd->root.mnt)
1024 set_root_rcu(nd);
1025 nd->path = nd->root;
1026 d = nd->path.dentry;
1027 nd->inode = d->d_inode;
1028 nd->seq = nd->root_seq;
1029 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
Al Viro8c1b4562015-05-09 18:15:21 -04001030 return ERR_PTR(-ECHILD);
Al Viro8f47a0162015-05-09 19:02:01 -04001031 } else {
1032 if (!nd->root.mnt)
1033 set_root(nd);
1034 path_put(&nd->path);
1035 nd->path = nd->root;
1036 path_get(&nd->root);
1037 nd->inode = nd->path.dentry->d_inode;
Al Viro8c1b4562015-05-09 18:15:21 -04001038 }
Al Virofab51e82015-05-10 11:01:00 -04001039 nd->flags |= LOOKUP_JUMPED;
1040 while (unlikely(*++res == '/'))
1041 ;
1042 }
1043 if (!*res)
1044 res = NULL;
Al Viro0a959df2015-04-18 18:23:41 -04001045 return res;
1046}
Al Viro6d7b5aa2012-06-10 04:15:17 -04001047
David Howellsf015f1262012-06-25 12:55:28 +01001048/*
1049 * follow_up - Find the mountpoint of path's vfsmount
1050 *
1051 * Given a path, find the mountpoint of its source file system.
1052 * Replace @path with the path of the mountpoint in the parent mount.
1053 * Up is towards /.
1054 *
1055 * Return 1 if we went up a level and 0 if we were already at the
1056 * root.
1057 */
Al Virobab77eb2009-04-18 03:26:48 -04001058int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059{
Al Viro0714a532011-11-24 22:19:58 -05001060 struct mount *mnt = real_mount(path->mnt);
1061 struct mount *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +10001063
Al Viro48a066e2013-09-29 22:06:07 -04001064 read_seqlock_excl(&mount_lock);
Al Viro0714a532011-11-24 22:19:58 -05001065 parent = mnt->mnt_parent;
Al Viro3c0a6162012-07-18 17:32:50 +04001066 if (parent == mnt) {
Al Viro48a066e2013-09-29 22:06:07 -04001067 read_sequnlock_excl(&mount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 return 0;
1069 }
Al Viro0714a532011-11-24 22:19:58 -05001070 mntget(&parent->mnt);
Al Viroa73324d2011-11-24 22:25:07 -05001071 mountpoint = dget(mnt->mnt_mountpoint);
Al Viro48a066e2013-09-29 22:06:07 -04001072 read_sequnlock_excl(&mount_lock);
Al Virobab77eb2009-04-18 03:26:48 -04001073 dput(path->dentry);
1074 path->dentry = mountpoint;
1075 mntput(path->mnt);
Al Viro0714a532011-11-24 22:19:58 -05001076 path->mnt = &parent->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 return 1;
1078}
Al Viro4d359502014-03-14 12:20:17 -04001079EXPORT_SYMBOL(follow_up);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001081/*
David Howells9875cf82011-01-14 18:45:21 +00001082 * Perform an automount
1083 * - return -EISDIR to tell follow_managed() to stop and return the path we
1084 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 */
NeilBrown756daf22015-03-23 13:37:38 +11001086static int follow_automount(struct path *path, struct nameidata *nd,
David Howells9875cf82011-01-14 18:45:21 +00001087 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +11001088{
David Howells9875cf82011-01-14 18:45:21 +00001089 struct vfsmount *mnt;
David Howellsea5b7782011-01-14 19:10:03 +00001090 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001091
David Howells9875cf82011-01-14 18:45:21 +00001092 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1093 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -07001094
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +02001095 /* We don't want to mount if someone's just doing a stat -
1096 * unless they're stat'ing a directory and appended a '/' to
1097 * the name.
1098 *
1099 * We do, however, want to mount if someone wants to open or
1100 * create a file of any type under the mountpoint, wants to
1101 * traverse through the mountpoint or wants to open the
1102 * mounted directory. Also, autofs may mark negative dentries
1103 * as being automount points. These will need the attentions
1104 * of the daemon to instantiate them before they can be used.
David Howells9875cf82011-01-14 18:45:21 +00001105 */
NeilBrown756daf22015-03-23 13:37:38 +11001106 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1107 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +02001108 path->dentry->d_inode)
1109 return -EISDIR;
1110
NeilBrown756daf22015-03-23 13:37:38 +11001111 nd->total_link_count++;
1112 if (nd->total_link_count >= 40)
David Howells9875cf82011-01-14 18:45:21 +00001113 return -ELOOP;
1114
1115 mnt = path->dentry->d_op->d_automount(path);
1116 if (IS_ERR(mnt)) {
1117 /*
1118 * The filesystem is allowed to return -EISDIR here to indicate
1119 * it doesn't want to automount. For instance, autofs would do
1120 * this so that its userspace daemon can mount on this dentry.
1121 *
1122 * However, we can only permit this if it's a terminal point in
1123 * the path being looked up; if it wasn't then the remainder of
1124 * the path is inaccessible and we should say so.
1125 */
NeilBrown756daf22015-03-23 13:37:38 +11001126 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
David Howells9875cf82011-01-14 18:45:21 +00001127 return -EREMOTE;
1128 return PTR_ERR(mnt);
1129 }
David Howellsea5b7782011-01-14 19:10:03 +00001130
David Howells9875cf82011-01-14 18:45:21 +00001131 if (!mnt) /* mount collision */
1132 return 0;
1133
Al Viro8aef1882011-06-16 15:10:06 +01001134 if (!*need_mntput) {
1135 /* lock_mount() may release path->mnt on error */
1136 mntget(path->mnt);
1137 *need_mntput = true;
1138 }
Al Viro19a167a2011-01-17 01:35:23 -05001139 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +00001140
David Howellsea5b7782011-01-14 19:10:03 +00001141 switch (err) {
1142 case -EBUSY:
1143 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -05001144 return 0;
David Howellsea5b7782011-01-14 19:10:03 +00001145 case 0:
Al Viro8aef1882011-06-16 15:10:06 +01001146 path_put(path);
David Howellsea5b7782011-01-14 19:10:03 +00001147 path->mnt = mnt;
1148 path->dentry = dget(mnt->mnt_root);
David Howellsea5b7782011-01-14 19:10:03 +00001149 return 0;
Al Viro19a167a2011-01-17 01:35:23 -05001150 default:
1151 return err;
David Howellsea5b7782011-01-14 19:10:03 +00001152 }
Al Viro19a167a2011-01-17 01:35:23 -05001153
David Howells9875cf82011-01-14 18:45:21 +00001154}
1155
1156/*
1157 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +00001158 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +00001159 * - Flagged as mountpoint
1160 * - Flagged as automount point
1161 *
1162 * This may only be called in refwalk mode.
1163 *
1164 * Serialization is taken care of in namespace.c
1165 */
NeilBrown756daf22015-03-23 13:37:38 +11001166static int follow_managed(struct path *path, struct nameidata *nd)
David Howells9875cf82011-01-14 18:45:21 +00001167{
Al Viro8aef1882011-06-16 15:10:06 +01001168 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
David Howells9875cf82011-01-14 18:45:21 +00001169 unsigned managed;
1170 bool need_mntput = false;
Al Viro8aef1882011-06-16 15:10:06 +01001171 int ret = 0;
David Howells9875cf82011-01-14 18:45:21 +00001172
1173 /* Given that we're not holding a lock here, we retain the value in a
1174 * local variable for each dentry as we look at it so that we don't see
1175 * the components of that value change under us */
1176 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1177 managed &= DCACHE_MANAGED_DENTRY,
1178 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +00001179 /* Allow the filesystem to manage the transit without i_mutex
1180 * being held. */
1181 if (managed & DCACHE_MANAGE_TRANSIT) {
1182 BUG_ON(!path->dentry->d_op);
1183 BUG_ON(!path->dentry->d_op->d_manage);
Al Viro1aed3e42011-03-18 09:09:02 -04001184 ret = path->dentry->d_op->d_manage(path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001185 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001186 break;
David Howellscc53ce52011-01-14 18:45:26 +00001187 }
1188
David Howells9875cf82011-01-14 18:45:21 +00001189 /* Transit to a mounted filesystem. */
1190 if (managed & DCACHE_MOUNTED) {
1191 struct vfsmount *mounted = lookup_mnt(path);
1192 if (mounted) {
1193 dput(path->dentry);
1194 if (need_mntput)
1195 mntput(path->mnt);
1196 path->mnt = mounted;
1197 path->dentry = dget(mounted->mnt_root);
1198 need_mntput = true;
1199 continue;
1200 }
1201
1202 /* Something is mounted on this dentry in another
1203 * namespace and/or whatever was mounted there in this
Al Viro48a066e2013-09-29 22:06:07 -04001204 * namespace got unmounted before lookup_mnt() could
1205 * get it */
David Howells9875cf82011-01-14 18:45:21 +00001206 }
1207
1208 /* Handle an automount point */
1209 if (managed & DCACHE_NEED_AUTOMOUNT) {
NeilBrown756daf22015-03-23 13:37:38 +11001210 ret = follow_automount(path, nd, &need_mntput);
David Howells9875cf82011-01-14 18:45:21 +00001211 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001212 break;
David Howells9875cf82011-01-14 18:45:21 +00001213 continue;
1214 }
1215
1216 /* We didn't change the current path point */
1217 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 }
Al Viro8aef1882011-06-16 15:10:06 +01001219
1220 if (need_mntput && path->mnt == mnt)
1221 mntput(path->mnt);
1222 if (ret == -EISDIR)
1223 ret = 0;
Al Viro84027522015-04-22 10:30:08 -04001224 if (need_mntput)
1225 nd->flags |= LOOKUP_JUMPED;
1226 if (unlikely(ret < 0))
1227 path_put_conditional(path, nd);
1228 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229}
1230
David Howellscc53ce52011-01-14 18:45:26 +00001231int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232{
1233 struct vfsmount *mounted;
1234
Al Viro1c755af2009-04-18 14:06:57 -04001235 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -04001237 dput(path->dentry);
1238 mntput(path->mnt);
1239 path->mnt = mounted;
1240 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 return 1;
1242 }
1243 return 0;
1244}
Al Viro4d359502014-03-14 12:20:17 -04001245EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246
NeilBrownb8faf032014-08-04 17:06:29 +10001247static inline int managed_dentry_rcu(struct dentry *dentry)
Ian Kent62a73752011-03-25 01:51:02 +08001248{
NeilBrownb8faf032014-08-04 17:06:29 +10001249 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1250 dentry->d_op->d_manage(dentry, true) : 0;
Ian Kent62a73752011-03-25 01:51:02 +08001251}
1252
David Howells9875cf82011-01-14 18:45:21 +00001253/*
Al Viro287548e2011-05-27 06:50:06 -04001254 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1255 * we meet a managed dentry that would need blocking.
David Howells9875cf82011-01-14 18:45:21 +00001256 */
1257static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
Al Viro254cf582015-05-05 09:40:46 -04001258 struct inode **inode, unsigned *seqp)
David Howells9875cf82011-01-14 18:45:21 +00001259{
Ian Kent62a73752011-03-25 01:51:02 +08001260 for (;;) {
Al Viroc7105362011-11-24 18:22:03 -05001261 struct mount *mounted;
Ian Kent62a73752011-03-25 01:51:02 +08001262 /*
1263 * Don't forget we might have a non-mountpoint managed dentry
1264 * that wants to block transit.
1265 */
NeilBrownb8faf032014-08-04 17:06:29 +10001266 switch (managed_dentry_rcu(path->dentry)) {
1267 case -ECHILD:
1268 default:
David Howellsab909112011-01-14 18:46:51 +00001269 return false;
NeilBrownb8faf032014-08-04 17:06:29 +10001270 case -EISDIR:
1271 return true;
1272 case 0:
1273 break;
1274 }
Ian Kent62a73752011-03-25 01:51:02 +08001275
1276 if (!d_mountpoint(path->dentry))
NeilBrownb8faf032014-08-04 17:06:29 +10001277 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
Ian Kent62a73752011-03-25 01:51:02 +08001278
Al Viro474279d2013-10-01 16:11:26 -04001279 mounted = __lookup_mnt(path->mnt, path->dentry);
David Howells9875cf82011-01-14 18:45:21 +00001280 if (!mounted)
1281 break;
Al Viroc7105362011-11-24 18:22:03 -05001282 path->mnt = &mounted->mnt;
1283 path->dentry = mounted->mnt.mnt_root;
Al Viroa3fbbde2011-11-07 21:21:26 +00001284 nd->flags |= LOOKUP_JUMPED;
Al Viro254cf582015-05-05 09:40:46 -04001285 *seqp = read_seqcount_begin(&path->dentry->d_seq);
Linus Torvalds59430262011-07-18 15:43:29 -07001286 /*
1287 * Update the inode too. We don't need to re-check the
1288 * dentry sequence number here after this d_inode read,
1289 * because a mount-point is always pinned.
1290 */
1291 *inode = path->dentry->d_inode;
David Howells9875cf82011-01-14 18:45:21 +00001292 }
Al Virof5be3e29122014-09-13 21:50:45 -04001293 return !read_seqretry(&mount_lock, nd->m_seq) &&
NeilBrownb8faf032014-08-04 17:06:29 +10001294 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
Al Viro287548e2011-05-27 06:50:06 -04001295}
1296
Nick Piggin31e6b012011-01-07 17:49:52 +11001297static int follow_dotdot_rcu(struct nameidata *nd)
1298{
Al Viro4023bfc2014-09-13 21:59:43 -04001299 struct inode *inode = nd->inode;
Al Viro7bd88372014-09-13 21:55:46 -04001300 if (!nd->root.mnt)
1301 set_root_rcu(nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11001302
David Howells9875cf82011-01-14 18:45:21 +00001303 while (1) {
Al Viroaed434a2015-05-12 12:22:47 -04001304 if (path_equal(&nd->path, &nd->root))
Nick Piggin31e6b012011-01-07 17:49:52 +11001305 break;
Nick Piggin31e6b012011-01-07 17:49:52 +11001306 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1307 struct dentry *old = nd->path.dentry;
1308 struct dentry *parent = old->d_parent;
1309 unsigned seq;
1310
Al Viro4023bfc2014-09-13 21:59:43 -04001311 inode = parent->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001312 seq = read_seqcount_begin(&parent->d_seq);
Al Viroaed434a2015-05-12 12:22:47 -04001313 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1314 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001315 nd->path.dentry = parent;
1316 nd->seq = seq;
Eric W. Biederman397d4252015-08-15 20:27:13 -05001317 if (unlikely(!path_connected(&nd->path)))
1318 return -ENOENT;
Nick Piggin31e6b012011-01-07 17:49:52 +11001319 break;
Al Viroaed434a2015-05-12 12:22:47 -04001320 } else {
1321 struct mount *mnt = real_mount(nd->path.mnt);
1322 struct mount *mparent = mnt->mnt_parent;
1323 struct dentry *mountpoint = mnt->mnt_mountpoint;
1324 struct inode *inode2 = mountpoint->d_inode;
1325 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1326 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1327 return -ECHILD;
1328 if (&mparent->mnt == nd->path.mnt)
1329 break;
1330 /* we know that mountpoint was pinned */
1331 nd->path.dentry = mountpoint;
1332 nd->path.mnt = &mparent->mnt;
1333 inode = inode2;
1334 nd->seq = seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11001335 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001336 }
Al Viroaed434a2015-05-12 12:22:47 -04001337 while (unlikely(d_mountpoint(nd->path.dentry))) {
Al Virob37199e2014-03-20 15:18:22 -04001338 struct mount *mounted;
1339 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
Al Viroaed434a2015-05-12 12:22:47 -04001340 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1341 return -ECHILD;
Al Virob37199e2014-03-20 15:18:22 -04001342 if (!mounted)
1343 break;
1344 nd->path.mnt = &mounted->mnt;
1345 nd->path.dentry = mounted->mnt.mnt_root;
Al Viro4023bfc2014-09-13 21:59:43 -04001346 inode = nd->path.dentry->d_inode;
Al Virob37199e2014-03-20 15:18:22 -04001347 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Virob37199e2014-03-20 15:18:22 -04001348 }
Al Viro4023bfc2014-09-13 21:59:43 -04001349 nd->inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001350 return 0;
1351}
1352
David Howells9875cf82011-01-14 18:45:21 +00001353/*
David Howellscc53ce52011-01-14 18:45:26 +00001354 * Follow down to the covering mount currently visible to userspace. At each
1355 * point, the filesystem owning that dentry may be queried as to whether the
1356 * caller is permitted to proceed or not.
David Howellscc53ce52011-01-14 18:45:26 +00001357 */
Al Viro7cc90cc2011-03-18 09:04:20 -04001358int follow_down(struct path *path)
David Howellscc53ce52011-01-14 18:45:26 +00001359{
1360 unsigned managed;
1361 int ret;
1362
1363 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1364 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1365 /* Allow the filesystem to manage the transit without i_mutex
1366 * being held.
1367 *
1368 * We indicate to the filesystem if someone is trying to mount
1369 * something here. This gives autofs the chance to deny anyone
1370 * other than its daemon the right to mount on its
1371 * superstructure.
1372 *
1373 * The filesystem may sleep at this point.
1374 */
1375 if (managed & DCACHE_MANAGE_TRANSIT) {
1376 BUG_ON(!path->dentry->d_op);
1377 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +00001378 ret = path->dentry->d_op->d_manage(
Al Viro1aed3e42011-03-18 09:09:02 -04001379 path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001380 if (ret < 0)
1381 return ret == -EISDIR ? 0 : ret;
1382 }
1383
1384 /* Transit to a mounted filesystem. */
1385 if (managed & DCACHE_MOUNTED) {
1386 struct vfsmount *mounted = lookup_mnt(path);
1387 if (!mounted)
1388 break;
1389 dput(path->dentry);
1390 mntput(path->mnt);
1391 path->mnt = mounted;
1392 path->dentry = dget(mounted->mnt_root);
1393 continue;
1394 }
1395
1396 /* Don't handle automount points here */
1397 break;
1398 }
1399 return 0;
1400}
Al Viro4d359502014-03-14 12:20:17 -04001401EXPORT_SYMBOL(follow_down);
David Howellscc53ce52011-01-14 18:45:26 +00001402
1403/*
David Howells9875cf82011-01-14 18:45:21 +00001404 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1405 */
1406static void follow_mount(struct path *path)
1407{
1408 while (d_mountpoint(path->dentry)) {
1409 struct vfsmount *mounted = lookup_mnt(path);
1410 if (!mounted)
1411 break;
1412 dput(path->dentry);
1413 mntput(path->mnt);
1414 path->mnt = mounted;
1415 path->dentry = dget(mounted->mnt_root);
1416 }
1417}
1418
Eric W. Biederman397d4252015-08-15 20:27:13 -05001419static int follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420{
Al Viro7bd88372014-09-13 21:55:46 -04001421 if (!nd->root.mnt)
1422 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001423
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 while(1) {
Jan Blunck4ac91372008-02-14 19:34:32 -08001425 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426
Al Viro2a737872009-04-07 11:49:53 -04001427 if (nd->path.dentry == nd->root.dentry &&
1428 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 break;
1430 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001431 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Al Viro3088dd72010-01-30 15:47:29 -05001432 /* rare case of legitimate dget_parent()... */
1433 nd->path.dentry = dget_parent(nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 dput(old);
Eric W. Biederman397d4252015-08-15 20:27:13 -05001435 if (unlikely(!path_connected(&nd->path)))
1436 return -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 break;
1438 }
Al Viro3088dd72010-01-30 15:47:29 -05001439 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 }
Al Viro79ed0222009-04-18 13:59:41 -04001442 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001443 nd->inode = nd->path.dentry->d_inode;
Eric W. Biederman397d4252015-08-15 20:27:13 -05001444 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445}
1446
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447/*
Miklos Szeredibad61182012-03-26 12:54:24 +02001448 * This looks up the name in dcache, possibly revalidates the old dentry and
1449 * allocates a new one if not found or not valid. In the need_lookup argument
1450 * returns whether i_op->lookup is necessary.
1451 *
1452 * dir->d_inode->i_mutex must be held
Nick Pigginbaa03892010-08-18 04:37:31 +10001453 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001454static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
Al Viro201f9562012-06-22 12:42:10 +04001455 unsigned int flags, bool *need_lookup)
Nick Pigginbaa03892010-08-18 04:37:31 +10001456{
Nick Pigginbaa03892010-08-18 04:37:31 +10001457 struct dentry *dentry;
Miklos Szeredibad61182012-03-26 12:54:24 +02001458 int error;
Nick Pigginbaa03892010-08-18 04:37:31 +10001459
Miklos Szeredibad61182012-03-26 12:54:24 +02001460 *need_lookup = false;
1461 dentry = d_lookup(dir, name);
1462 if (dentry) {
Jeff Layton39e3c952012-11-28 11:30:53 -05001463 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
Al Viro201f9562012-06-22 12:42:10 +04001464 error = d_revalidate(dentry, flags);
Miklos Szeredibad61182012-03-26 12:54:24 +02001465 if (unlikely(error <= 0)) {
1466 if (error < 0) {
1467 dput(dentry);
1468 return ERR_PTR(error);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001469 } else {
1470 d_invalidate(dentry);
Miklos Szeredibad61182012-03-26 12:54:24 +02001471 dput(dentry);
1472 dentry = NULL;
1473 }
1474 }
1475 }
1476 }
Nick Pigginbaa03892010-08-18 04:37:31 +10001477
Miklos Szeredibad61182012-03-26 12:54:24 +02001478 if (!dentry) {
1479 dentry = d_alloc(dir, name);
1480 if (unlikely(!dentry))
1481 return ERR_PTR(-ENOMEM);
Nick Pigginbaa03892010-08-18 04:37:31 +10001482
Miklos Szeredibad61182012-03-26 12:54:24 +02001483 *need_lookup = true;
Nick Pigginbaa03892010-08-18 04:37:31 +10001484 }
1485 return dentry;
1486}
1487
1488/*
J. Bruce Fields13a2c3b2013-10-23 16:09:16 -04001489 * Call i_op->lookup on the dentry. The dentry must be negative and
1490 * unhashed.
Miklos Szeredibad61182012-03-26 12:54:24 +02001491 *
1492 * dir->d_inode->i_mutex must be held
Josef Bacik44396f42011-05-31 11:58:49 -04001493 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001494static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
Al Viro72bd8662012-06-10 17:17:17 -04001495 unsigned int flags)
Josef Bacik44396f42011-05-31 11:58:49 -04001496{
Josef Bacik44396f42011-05-31 11:58:49 -04001497 struct dentry *old;
1498
1499 /* Don't create child dentry for a dead directory. */
Miklos Szeredibad61182012-03-26 12:54:24 +02001500 if (unlikely(IS_DEADDIR(dir))) {
Miklos Szeredie188dc02012-02-03 14:25:18 +01001501 dput(dentry);
Josef Bacik44396f42011-05-31 11:58:49 -04001502 return ERR_PTR(-ENOENT);
Miklos Szeredie188dc02012-02-03 14:25:18 +01001503 }
Josef Bacik44396f42011-05-31 11:58:49 -04001504
Al Viro72bd8662012-06-10 17:17:17 -04001505 old = dir->i_op->lookup(dir, dentry, flags);
Josef Bacik44396f42011-05-31 11:58:49 -04001506 if (unlikely(old)) {
1507 dput(dentry);
1508 dentry = old;
1509 }
1510 return dentry;
1511}
1512
Al Viroa3255542012-03-30 14:41:51 -04001513static struct dentry *__lookup_hash(struct qstr *name,
Al Viro72bd8662012-06-10 17:17:17 -04001514 struct dentry *base, unsigned int flags)
Al Viroa3255542012-03-30 14:41:51 -04001515{
Miklos Szeredibad61182012-03-26 12:54:24 +02001516 bool need_lookup;
Al Viroa3255542012-03-30 14:41:51 -04001517 struct dentry *dentry;
1518
Al Viro72bd8662012-06-10 17:17:17 -04001519 dentry = lookup_dcache(name, base, flags, &need_lookup);
Miklos Szeredibad61182012-03-26 12:54:24 +02001520 if (!need_lookup)
1521 return dentry;
Al Viroa3255542012-03-30 14:41:51 -04001522
Al Viro72bd8662012-06-10 17:17:17 -04001523 return lookup_real(base->d_inode, dentry, flags);
Al Viroa3255542012-03-30 14:41:51 -04001524}
1525
Josef Bacik44396f42011-05-31 11:58:49 -04001526/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 * It's more convoluted than I'd like it to be, but... it's still fairly
1528 * small and for now I'd prefer to have fast path as straight as possible.
1529 * It _is_ time-critical.
1530 */
Al Viroe97cdc82013-01-24 18:16:00 -05001531static int lookup_fast(struct nameidata *nd,
Al Viro254cf582015-05-05 09:40:46 -04001532 struct path *path, struct inode **inode,
1533 unsigned *seqp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534{
Jan Blunck4ac91372008-02-14 19:34:32 -08001535 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001536 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001537 int need_reval = 1;
1538 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001539 int err;
1540
Al Viro3cac2602009-08-13 18:27:43 +04001541 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001542 * Rename seqlock is not required here because in the off chance
1543 * of a false negative due to a concurrent rename, we're going to
1544 * do the non-racy lookup, below.
1545 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001546 if (nd->flags & LOOKUP_RCU) {
1547 unsigned seq;
Al Viro766c4cb2015-05-07 19:24:57 -04001548 bool negative;
Linus Torvaldsda53be12013-05-21 15:22:44 -07001549 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
Al Viro5a18fff2011-03-11 04:44:53 -05001550 if (!dentry)
1551 goto unlazy;
1552
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001553 /*
1554 * This sequence count validates that the inode matches
1555 * the dentry name information from lookup.
1556 */
David Howells63afdfc2015-05-06 15:59:00 +01001557 *inode = d_backing_inode(dentry);
Al Viro766c4cb2015-05-07 19:24:57 -04001558 negative = d_is_negative(dentry);
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001559 if (read_seqcount_retry(&dentry->d_seq, seq))
1560 return -ECHILD;
1561
1562 /*
1563 * This sequence count validates that the parent had no
1564 * changes while we did the lookup of the dentry above.
1565 *
1566 * The memory barrier in read_seqcount_begin of child is
1567 * enough, we can use __read_seqcount_retry here.
1568 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001569 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1570 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001571
Al Viro254cf582015-05-05 09:40:46 -04001572 *seqp = seq;
Al Viro24643082011-02-15 01:26:22 -05001573 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
Al Viro4ce16ef32012-06-10 16:10:59 -04001574 status = d_revalidate(dentry, nd->flags);
Al Viro5a18fff2011-03-11 04:44:53 -05001575 if (unlikely(status <= 0)) {
1576 if (status != -ECHILD)
1577 need_reval = 0;
1578 goto unlazy;
1579 }
Al Viro24643082011-02-15 01:26:22 -05001580 }
Trond Myklebustdaf37612015-10-09 13:44:34 -04001581 /*
1582 * Note: do negative dentry check after revalidation in
1583 * case that drops it.
1584 */
1585 if (negative)
1586 return -ENOENT;
Nick Piggin31e6b012011-01-07 17:49:52 +11001587 path->mnt = mnt;
1588 path->dentry = dentry;
Al Viro254cf582015-05-05 09:40:46 -04001589 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
NeilBrownb8faf032014-08-04 17:06:29 +10001590 return 0;
Al Viro5a18fff2011-03-11 04:44:53 -05001591unlazy:
Al Viro254cf582015-05-05 09:40:46 -04001592 if (unlazy_walk(nd, dentry, seq))
Al Viro19660af2011-03-25 10:32:48 -04001593 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001594 } else {
Al Viroe97cdc82013-01-24 18:16:00 -05001595 dentry = __d_lookup(parent, &nd->last);
Nick Piggin31e6b012011-01-07 17:49:52 +11001596 }
Al Viro5a18fff2011-03-11 04:44:53 -05001597
Al Viro81e6f522012-03-30 14:48:04 -04001598 if (unlikely(!dentry))
1599 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001600
Al Viro5a18fff2011-03-11 04:44:53 -05001601 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
Al Viro4ce16ef32012-06-10 16:10:59 -04001602 status = d_revalidate(dentry, nd->flags);
Al Viro5a18fff2011-03-11 04:44:53 -05001603 if (unlikely(status <= 0)) {
1604 if (status < 0) {
1605 dput(dentry);
1606 return status;
1607 }
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001608 d_invalidate(dentry);
1609 dput(dentry);
1610 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001611 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001612
Al Viro766c4cb2015-05-07 19:24:57 -04001613 if (unlikely(d_is_negative(dentry))) {
1614 dput(dentry);
1615 return -ENOENT;
1616 }
David Howells9875cf82011-01-14 18:45:21 +00001617 path->mnt = mnt;
1618 path->dentry = dentry;
NeilBrown756daf22015-03-23 13:37:38 +11001619 err = follow_managed(path, nd);
Al Viro84027522015-04-22 10:30:08 -04001620 if (likely(!err))
David Howells63afdfc2015-05-06 15:59:00 +01001621 *inode = d_backing_inode(path->dentry);
Al Viro84027522015-04-22 10:30:08 -04001622 return err;
Al Viro81e6f522012-03-30 14:48:04 -04001623
1624need_lookup:
Miklos Szeredi697f5142012-05-21 17:30:05 +02001625 return 1;
1626}
1627
1628/* Fast lookup failed, do it the slow way */
Al Virocc2a5272013-01-24 18:19:49 -05001629static int lookup_slow(struct nameidata *nd, struct path *path)
Miklos Szeredi697f5142012-05-21 17:30:05 +02001630{
1631 struct dentry *dentry, *parent;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001632
1633 parent = nd->path.dentry;
Al Viro81e6f522012-03-30 14:48:04 -04001634 BUG_ON(nd->inode != parent->d_inode);
1635
1636 mutex_lock(&parent->d_inode->i_mutex);
Al Virocc2a5272013-01-24 18:19:49 -05001637 dentry = __lookup_hash(&nd->last, parent, nd->flags);
Al Viro81e6f522012-03-30 14:48:04 -04001638 mutex_unlock(&parent->d_inode->i_mutex);
1639 if (IS_ERR(dentry))
1640 return PTR_ERR(dentry);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001641 path->mnt = nd->path.mnt;
1642 path->dentry = dentry;
Al Viro84027522015-04-22 10:30:08 -04001643 return follow_managed(path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644}
1645
Al Viro52094c82011-02-21 21:34:47 -05001646static inline int may_lookup(struct nameidata *nd)
1647{
1648 if (nd->flags & LOOKUP_RCU) {
Al Viro4ad5abb2011-06-20 19:57:03 -04001649 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
Al Viro52094c82011-02-21 21:34:47 -05001650 if (err != -ECHILD)
1651 return err;
Al Viro6e9918b2015-05-05 09:26:05 -04001652 if (unlazy_walk(nd, NULL, 0))
Al Viro52094c82011-02-21 21:34:47 -05001653 return -ECHILD;
1654 }
Al Viro4ad5abb2011-06-20 19:57:03 -04001655 return inode_permission(nd->inode, MAY_EXEC);
Al Viro52094c82011-02-21 21:34:47 -05001656}
1657
Al Viro9856fa12011-03-04 14:22:06 -05001658static inline int handle_dots(struct nameidata *nd, int type)
1659{
1660 if (type == LAST_DOTDOT) {
1661 if (nd->flags & LOOKUP_RCU) {
Al Viro70291ae2015-05-04 07:53:00 -04001662 return follow_dotdot_rcu(nd);
Al Viro9856fa12011-03-04 14:22:06 -05001663 } else
Eric W. Biederman397d4252015-08-15 20:27:13 -05001664 return follow_dotdot(nd);
Al Viro9856fa12011-03-04 14:22:06 -05001665 }
1666 return 0;
1667}
1668
Al Viro181548c2015-05-07 19:54:34 -04001669static int pick_link(struct nameidata *nd, struct path *link,
1670 struct inode *inode, unsigned seq)
Al Virod63ff282015-05-04 18:13:23 -04001671{
Al Viro626de992015-05-04 18:26:59 -04001672 int error;
Al Viro1cf26652015-05-06 16:01:56 -04001673 struct saved *last;
NeilBrown756daf22015-03-23 13:37:38 +11001674 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
Al Viro626de992015-05-04 18:26:59 -04001675 path_to_nameidata(link, nd);
1676 return -ELOOP;
1677 }
Al Virobc40aee2015-05-09 13:04:24 -04001678 if (!(nd->flags & LOOKUP_RCU)) {
Al Viro79733872015-05-09 12:55:43 -04001679 if (link->mnt == nd->path.mnt)
1680 mntget(link->mnt);
Al Virod63ff282015-05-04 18:13:23 -04001681 }
Al Viro626de992015-05-04 18:26:59 -04001682 error = nd_alloc_stack(nd);
1683 if (unlikely(error)) {
Al Virobc40aee2015-05-09 13:04:24 -04001684 if (error == -ECHILD) {
1685 if (unlikely(unlazy_link(nd, link, seq)))
1686 return -ECHILD;
1687 error = nd_alloc_stack(nd);
1688 }
1689 if (error) {
1690 path_put(link);
1691 return error;
1692 }
Al Viro626de992015-05-04 18:26:59 -04001693 }
1694
Al Viroab104922015-05-10 11:50:01 -04001695 last = nd->stack + nd->depth++;
Al Viro1cf26652015-05-06 16:01:56 -04001696 last->link = *link;
Al Viroab104922015-05-10 11:50:01 -04001697 last->cookie = NULL;
Al Viro181548c2015-05-07 19:54:34 -04001698 last->inode = inode;
Al Viro0450b2d2015-05-08 13:23:53 -04001699 last->seq = seq;
Al Virod63ff282015-05-04 18:13:23 -04001700 return 1;
1701}
1702
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001703/*
1704 * Do we need to follow links? We _really_ want to be able
1705 * to do this check without having to look at inode->i_op,
1706 * so we keep a cache of "no, this doesn't need follow_link"
1707 * for the common case.
1708 */
Al Viro254cf582015-05-05 09:40:46 -04001709static inline int should_follow_link(struct nameidata *nd, struct path *link,
Al Viro181548c2015-05-07 19:54:34 -04001710 int follow,
1711 struct inode *inode, unsigned seq)
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001712{
Al Virod63ff282015-05-04 18:13:23 -04001713 if (likely(!d_is_symlink(link->dentry)))
1714 return 0;
1715 if (!follow)
1716 return 0;
Al Viro181548c2015-05-07 19:54:34 -04001717 return pick_link(nd, link, inode, seq);
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001718}
1719
Al Viro4693a542015-05-04 17:47:11 -04001720enum {WALK_GET = 1, WALK_PUT = 2};
1721
1722static int walk_component(struct nameidata *nd, int flags)
Al Viroce57dfc2011-03-13 19:58:58 -04001723{
Al Virocaa85632015-04-22 17:52:47 -04001724 struct path path;
Al Viroce57dfc2011-03-13 19:58:58 -04001725 struct inode *inode;
Al Viro254cf582015-05-05 09:40:46 -04001726 unsigned seq;
Al Viroce57dfc2011-03-13 19:58:58 -04001727 int err;
1728 /*
1729 * "." and ".." are special - ".." especially so because it has
1730 * to be able to know about the current root directory and
1731 * parent relationships.
1732 */
Al Viro4693a542015-05-04 17:47:11 -04001733 if (unlikely(nd->last_type != LAST_NORM)) {
1734 err = handle_dots(nd, nd->last_type);
1735 if (flags & WALK_PUT)
1736 put_link(nd);
1737 return err;
1738 }
Al Viro254cf582015-05-05 09:40:46 -04001739 err = lookup_fast(nd, &path, &inode, &seq);
Al Viroce57dfc2011-03-13 19:58:58 -04001740 if (unlikely(err)) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001741 if (err < 0)
Al Virof0a9ba72015-05-04 07:59:30 -04001742 return err;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001743
Al Virocaa85632015-04-22 17:52:47 -04001744 err = lookup_slow(nd, &path);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001745 if (err < 0)
Al Virof0a9ba72015-05-04 07:59:30 -04001746 return err;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001747
David Howells63afdfc2015-05-06 15:59:00 +01001748 inode = d_backing_inode(path.dentry);
Al Viro254cf582015-05-05 09:40:46 -04001749 seq = 0; /* we are already out of RCU mode */
Al Viro766c4cb2015-05-07 19:24:57 -04001750 err = -ENOENT;
Al Virocaa85632015-04-22 17:52:47 -04001751 if (d_is_negative(path.dentry))
Al Viro766c4cb2015-05-07 19:24:57 -04001752 goto out_path_put;
Al Viroce57dfc2011-03-13 19:58:58 -04001753 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001754
Al Viro4693a542015-05-04 17:47:11 -04001755 if (flags & WALK_PUT)
1756 put_link(nd);
Al Viro181548c2015-05-07 19:54:34 -04001757 err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq);
Al Virod63ff282015-05-04 18:13:23 -04001758 if (unlikely(err))
1759 return err;
Al Virocaa85632015-04-22 17:52:47 -04001760 path_to_nameidata(&path, nd);
Al Viroce57dfc2011-03-13 19:58:58 -04001761 nd->inode = inode;
Al Viro254cf582015-05-05 09:40:46 -04001762 nd->seq = seq;
Al Viroce57dfc2011-03-13 19:58:58 -04001763 return 0;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001764
1765out_path_put:
Al Virocaa85632015-04-22 17:52:47 -04001766 path_to_nameidata(&path, nd);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001767 return err;
Al Viroce57dfc2011-03-13 19:58:58 -04001768}
1769
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770/*
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001771 * We can do the critical dentry name comparison and hashing
1772 * operations one word at a time, but we are limited to:
1773 *
1774 * - Architectures with fast unaligned word accesses. We could
1775 * do a "get_unaligned()" if this helps and is sufficiently
1776 * fast.
1777 *
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001778 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1779 * do not trap on the (extremely unlikely) case of a page
1780 * crossing operation.
1781 *
1782 * - Furthermore, we need an efficient 64-bit compile for the
1783 * 64-bit case in order to generate the "number of bytes in
1784 * the final mask". Again, that could be replaced with a
1785 * efficient population count instruction or similar.
1786 */
1787#ifdef CONFIG_DCACHE_WORD_ACCESS
1788
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001789#include <asm/word-at-a-time.h>
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001790
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001791#ifdef CONFIG_64BIT
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001792
1793static inline unsigned int fold_hash(unsigned long hash)
1794{
Linus Torvalds99d263d2014-09-13 11:30:10 -07001795 return hash_64(hash, 32);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001796}
1797
1798#else /* 32-bit case */
1799
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001800#define fold_hash(x) (x)
1801
1802#endif
1803
1804unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1805{
1806 unsigned long a, mask;
1807 unsigned long hash = 0;
1808
1809 for (;;) {
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001810 a = load_unaligned_zeropad(name);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001811 if (len < sizeof(unsigned long))
1812 break;
1813 hash += a;
Al Virof132c5b2012-03-22 21:59:52 +00001814 hash *= 9;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001815 name += sizeof(unsigned long);
1816 len -= sizeof(unsigned long);
1817 if (!len)
1818 goto done;
1819 }
Will Deacona5c21dc2013-12-12 17:40:21 +00001820 mask = bytemask_from_count(len);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001821 hash += mask & a;
1822done:
1823 return fold_hash(hash);
1824}
1825EXPORT_SYMBOL(full_name_hash);
1826
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001827/*
1828 * Calculate the length and hash of the path component, and
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001829 * return the "hash_len" as the result.
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001830 */
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001831static inline u64 hash_name(const char *name)
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001832{
Linus Torvalds36126f82012-05-26 10:43:17 -07001833 unsigned long a, b, adata, bdata, mask, hash, len;
1834 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001835
1836 hash = a = 0;
1837 len = -sizeof(unsigned long);
1838 do {
1839 hash = (hash + a) * 9;
1840 len += sizeof(unsigned long);
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001841 a = load_unaligned_zeropad(name+len);
Linus Torvalds36126f82012-05-26 10:43:17 -07001842 b = a ^ REPEAT_BYTE('/');
1843 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001844
Linus Torvalds36126f82012-05-26 10:43:17 -07001845 adata = prep_zero_mask(a, adata, &constants);
1846 bdata = prep_zero_mask(b, bdata, &constants);
1847
1848 mask = create_zero_mask(adata | bdata);
1849
1850 hash += a & zero_bytemask(mask);
Linus Torvalds9226b5b2014-09-14 17:28:32 -07001851 len += find_zero(mask);
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001852 return hashlen_create(fold_hash(hash), len);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001853}
1854
1855#else
1856
Linus Torvalds0145acc2012-03-02 14:32:59 -08001857unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1858{
1859 unsigned long hash = init_name_hash();
1860 while (len--)
1861 hash = partial_name_hash(*name++, hash);
1862 return end_name_hash(hash);
1863}
Linus Torvaldsae942ae2012-03-02 19:40:57 -08001864EXPORT_SYMBOL(full_name_hash);
Linus Torvalds0145acc2012-03-02 14:32:59 -08001865
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001866/*
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001867 * We know there's a real path component here of at least
1868 * one character.
1869 */
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001870static inline u64 hash_name(const char *name)
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001871{
1872 unsigned long hash = init_name_hash();
1873 unsigned long len = 0, c;
1874
1875 c = (unsigned char)*name;
1876 do {
1877 len++;
1878 hash = partial_name_hash(c, hash);
1879 c = (unsigned char)name[len];
1880 } while (c && c != '/');
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001881 return hashlen_create(end_name_hash(hash), len);
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001882}
1883
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001884#endif
1885
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001886/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01001888 * This is the basic name resolution function, turning a pathname into
1889 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01001891 * Returns 0 and nd will have valid dentry and mnt on success.
1892 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893 */
Al Viro6de88d72009-08-09 01:41:57 +04001894static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896 int err;
Al Viro32cd7462015-04-18 20:30:49 -04001897
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898 while (*name=='/')
1899 name++;
1900 if (!*name)
Al Viro9e18f102015-04-18 20:44:34 -04001901 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 /* At this point we know we have a real path component. */
1904 for(;;) {
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001905 u64 hash_len;
Al Virofe479a52011-02-22 15:10:03 -05001906 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907
Al Viro52094c82011-02-21 21:34:47 -05001908 err = may_lookup(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 if (err)
Al Viro3595e232015-05-09 16:54:45 -04001910 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001912 hash_len = hash_name(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913
Al Virofe479a52011-02-22 15:10:03 -05001914 type = LAST_NORM;
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001915 if (name[0] == '.') switch (hashlen_len(hash_len)) {
Al Virofe479a52011-02-22 15:10:03 -05001916 case 2:
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001917 if (name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05001918 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05001919 nd->flags |= LOOKUP_JUMPED;
1920 }
Al Virofe479a52011-02-22 15:10:03 -05001921 break;
1922 case 1:
1923 type = LAST_DOT;
1924 }
Al Viro5a202bc2011-03-08 14:17:44 -05001925 if (likely(type == LAST_NORM)) {
1926 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05001927 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05001928 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
James Hogana060dc52014-09-16 13:07:35 +01001929 struct qstr this = { { .hash_len = hash_len }, .name = name };
Linus Torvaldsda53be12013-05-21 15:22:44 -07001930 err = parent->d_op->d_hash(parent, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05001931 if (err < 0)
Al Viro3595e232015-05-09 16:54:45 -04001932 return err;
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001933 hash_len = this.hash_len;
1934 name = this.name;
Al Viro5a202bc2011-03-08 14:17:44 -05001935 }
1936 }
Al Virofe479a52011-02-22 15:10:03 -05001937
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001938 nd->last.hash_len = hash_len;
1939 nd->last.name = name;
Al Viro5f4a6a62013-01-24 18:04:22 -05001940 nd->last_type = type;
1941
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001942 name += hashlen_len(hash_len);
1943 if (!*name)
Al Virobdf6cbf2015-04-18 20:21:40 -04001944 goto OK;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001945 /*
1946 * If it wasn't NUL, we know it was '/'. Skip that
1947 * slash, and continue until no more slashes.
1948 */
1949 do {
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001950 name++;
1951 } while (unlikely(*name == '/'));
Al Viro8620c232015-05-04 08:58:35 -04001952 if (unlikely(!*name)) {
1953OK:
Al Viro368ee9b2015-05-08 17:19:59 -04001954 /* pathname body, done */
Al Viro8620c232015-05-04 08:58:35 -04001955 if (!nd->depth)
1956 return 0;
1957 name = nd->stack[nd->depth - 1].name;
Al Viro368ee9b2015-05-08 17:19:59 -04001958 /* trailing symlink, done */
Al Viro8620c232015-05-04 08:58:35 -04001959 if (!name)
1960 return 0;
1961 /* last component of nested symlink */
Al Viro4693a542015-05-04 17:47:11 -04001962 err = walk_component(nd, WALK_GET | WALK_PUT);
Al Viro8620c232015-05-04 08:58:35 -04001963 } else {
Al Viro4693a542015-05-04 17:47:11 -04001964 err = walk_component(nd, WALK_GET);
Al Viro8620c232015-05-04 08:58:35 -04001965 }
Al Viroce57dfc2011-03-13 19:58:58 -04001966 if (err < 0)
Al Viro3595e232015-05-09 16:54:45 -04001967 return err;
Al Virofe479a52011-02-22 15:10:03 -05001968
Al Viroce57dfc2011-03-13 19:58:58 -04001969 if (err) {
Al Viro626de992015-05-04 18:26:59 -04001970 const char *s = get_link(nd);
Al Viro5a460272015-04-17 23:44:45 -04001971
Al Viro3595e232015-05-09 16:54:45 -04001972 if (unlikely(IS_ERR(s)))
1973 return PTR_ERR(s);
Al Virod40bcc02015-04-18 20:03:03 -04001974 err = 0;
1975 if (unlikely(!s)) {
1976 /* jumped */
Al Virob9ff4422015-05-02 20:19:23 -04001977 put_link(nd);
Al Virod40bcc02015-04-18 20:03:03 -04001978 } else {
Al Virofab51e82015-05-10 11:01:00 -04001979 nd->stack[nd->depth - 1].name = name;
1980 name = s;
1981 continue;
Al Virod40bcc02015-04-18 20:03:03 -04001982 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001983 }
Al Viro97242f92015-08-01 19:59:28 -04001984 if (unlikely(!d_can_lookup(nd->path.dentry))) {
1985 if (nd->flags & LOOKUP_RCU) {
1986 if (unlazy_walk(nd, NULL, 0))
1987 return -ECHILD;
1988 }
Al Viro3595e232015-05-09 16:54:45 -04001989 return -ENOTDIR;
Al Viro97242f92015-08-01 19:59:28 -04001990 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992}
1993
Al Viroc8a53ee2015-05-12 18:43:07 -04001994static const char *path_init(struct nameidata *nd, unsigned flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001996 int retval = 0;
Al Viroc8a53ee2015-05-12 18:43:07 -04001997 const char *s = nd->name->name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998
1999 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro980f3ea2014-11-20 14:20:24 -05002000 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 nd->depth = 0;
Al Viro368ee9b2015-05-08 17:19:59 -04002002 nd->total_link_count = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05002003 if (flags & LOOKUP_ROOT) {
David Howellsb18825a2013-09-12 19:22:53 +01002004 struct dentry *root = nd->root.dentry;
2005 struct inode *inode = root->d_inode;
Al Virofd2f7cb2015-02-22 20:07:13 -05002006 if (*s) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002007 if (!d_can_lookup(root))
Al Viro368ee9b2015-05-08 17:19:59 -04002008 return ERR_PTR(-ENOTDIR);
Al Viro73d049a2011-03-11 12:08:24 -05002009 retval = inode_permission(inode, MAY_EXEC);
2010 if (retval)
Al Viro368ee9b2015-05-08 17:19:59 -04002011 return ERR_PTR(retval);
Al Viro73d049a2011-03-11 12:08:24 -05002012 }
Al Viro5b6ca022011-03-09 23:04:47 -05002013 nd->path = nd->root;
2014 nd->inode = inode;
2015 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05002016 rcu_read_lock();
Al Viro5b6ca022011-03-09 23:04:47 -05002017 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro8f47a0162015-05-09 19:02:01 -04002018 nd->root_seq = nd->seq;
Al Viro48a066e2013-09-29 22:06:07 -04002019 nd->m_seq = read_seqbegin(&mount_lock);
Al Viro5b6ca022011-03-09 23:04:47 -05002020 } else {
2021 path_get(&nd->path);
2022 }
Al Viro368ee9b2015-05-08 17:19:59 -04002023 return s;
Al Viro5b6ca022011-03-09 23:04:47 -05002024 }
2025
Al Viro2a737872009-04-07 11:49:53 -04002026 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027
Al Viro48a066e2013-09-29 22:06:07 -04002028 nd->m_seq = read_seqbegin(&mount_lock);
Al Virofd2f7cb2015-02-22 20:07:13 -05002029 if (*s == '/') {
Al Viroe41f7d42011-02-22 14:02:58 -05002030 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05002031 rcu_read_lock();
Al Viro06d71372015-06-29 12:07:04 -04002032 set_root_rcu(nd);
2033 nd->seq = nd->root_seq;
Al Viroe41f7d42011-02-22 14:02:58 -05002034 } else {
2035 set_root(nd);
2036 path_get(&nd->root);
2037 }
Al Viro2a737872009-04-07 11:49:53 -04002038 nd->path = nd->root;
Al Viroc8a53ee2015-05-12 18:43:07 -04002039 } else if (nd->dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05002040 if (flags & LOOKUP_RCU) {
2041 struct fs_struct *fs = current->fs;
2042 unsigned seq;
2043
Al Viro8b61e742013-11-08 12:45:01 -05002044 rcu_read_lock();
Al Viroe41f7d42011-02-22 14:02:58 -05002045
2046 do {
2047 seq = read_seqcount_begin(&fs->seq);
2048 nd->path = fs->pwd;
2049 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2050 } while (read_seqcount_retry(&fs->seq, seq));
2051 } else {
2052 get_fs_pwd(current->fs, &nd->path);
2053 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002054 } else {
Jeff Layton582aa642012-12-11 08:56:16 -05002055 /* Caller must check execute permissions on the starting path component */
Al Viroc8a53ee2015-05-12 18:43:07 -04002056 struct fd f = fdget_raw(nd->dfd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002057 struct dentry *dentry;
2058
Al Viro2903ff02012-08-28 12:52:22 -04002059 if (!f.file)
Al Viro368ee9b2015-05-08 17:19:59 -04002060 return ERR_PTR(-EBADF);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002061
Al Viro2903ff02012-08-28 12:52:22 -04002062 dentry = f.file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002063
Al Virofd2f7cb2015-02-22 20:07:13 -05002064 if (*s) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002065 if (!d_can_lookup(dentry)) {
Al Viro2903ff02012-08-28 12:52:22 -04002066 fdput(f);
Al Viro368ee9b2015-05-08 17:19:59 -04002067 return ERR_PTR(-ENOTDIR);
Al Viro2903ff02012-08-28 12:52:22 -04002068 }
Al Virof52e0c12011-03-14 18:56:51 -04002069 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002070
Al Viro2903ff02012-08-28 12:52:22 -04002071 nd->path = f.file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05002072 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05002073 rcu_read_lock();
Al Viro34a26b92015-05-11 08:05:05 -04002074 nd->inode = nd->path.dentry->d_inode;
2075 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viroe41f7d42011-02-22 14:02:58 -05002076 } else {
Al Viro2903ff02012-08-28 12:52:22 -04002077 path_get(&nd->path);
Al Viro34a26b92015-05-11 08:05:05 -04002078 nd->inode = nd->path.dentry->d_inode;
Al Viroe41f7d42011-02-22 14:02:58 -05002079 }
Al Viro34a26b92015-05-11 08:05:05 -04002080 fdput(f);
Al Viro368ee9b2015-05-08 17:19:59 -04002081 return s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082 }
Al Viroe41f7d42011-02-22 14:02:58 -05002083
Nick Piggin31e6b012011-01-07 17:49:52 +11002084 nd->inode = nd->path.dentry->d_inode;
Al Viro4023bfc2014-09-13 21:59:43 -04002085 if (!(flags & LOOKUP_RCU))
Al Viro368ee9b2015-05-08 17:19:59 -04002086 return s;
Al Viro4023bfc2014-09-13 21:59:43 -04002087 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
Al Viro368ee9b2015-05-08 17:19:59 -04002088 return s;
Al Viro4023bfc2014-09-13 21:59:43 -04002089 if (!(nd->flags & LOOKUP_ROOT))
2090 nd->root.mnt = NULL;
2091 rcu_read_unlock();
Al Viro368ee9b2015-05-08 17:19:59 -04002092 return ERR_PTR(-ECHILD);
Al Viro9b4a9b12009-04-07 11:44:16 -04002093}
2094
Al Viro3bdba282015-05-08 17:37:07 -04002095static const char *trailing_symlink(struct nameidata *nd)
Al Viro95fa25d2015-04-22 13:46:57 -04002096{
2097 const char *s;
Al Virofec2fa22015-05-06 15:58:18 -04002098 int error = may_follow_link(nd);
Al Virodeb106c2015-05-08 18:05:21 -04002099 if (unlikely(error))
Al Viro3bdba282015-05-08 17:37:07 -04002100 return ERR_PTR(error);
Al Viro95fa25d2015-04-22 13:46:57 -04002101 nd->flags |= LOOKUP_PARENT;
Al Virofab51e82015-05-10 11:01:00 -04002102 nd->stack[0].name = NULL;
Al Viro3b2e7f72015-04-19 00:53:50 -04002103 s = get_link(nd);
Al Virodeb106c2015-05-08 18:05:21 -04002104 return s ? s : "";
Al Viro95fa25d2015-04-22 13:46:57 -04002105}
2106
Al Virocaa85632015-04-22 17:52:47 -04002107static inline int lookup_last(struct nameidata *nd)
Al Virobd92d7f2011-03-14 19:54:59 -04002108{
2109 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2110 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2111
2112 nd->flags &= ~LOOKUP_PARENT;
Al Virodeb106c2015-05-08 18:05:21 -04002113 return walk_component(nd,
Al Viro4693a542015-05-04 17:47:11 -04002114 nd->flags & LOOKUP_FOLLOW
2115 ? nd->depth
2116 ? WALK_PUT | WALK_GET
2117 : WALK_GET
2118 : 0);
Al Virobd92d7f2011-03-14 19:54:59 -04002119}
2120
Al Viro9b4a9b12009-04-07 11:44:16 -04002121/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroc8a53ee2015-05-12 18:43:07 -04002122static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
Al Viro9b4a9b12009-04-07 11:44:16 -04002123{
Al Viroc8a53ee2015-05-12 18:43:07 -04002124 const char *s = path_init(nd, flags);
Al Virobd92d7f2011-03-14 19:54:59 -04002125 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11002126
Al Viro368ee9b2015-05-08 17:19:59 -04002127 if (IS_ERR(s))
2128 return PTR_ERR(s);
Al Viro3bdba282015-05-08 17:37:07 -04002129 while (!(err = link_path_walk(s, nd))
2130 && ((err = lookup_last(nd)) > 0)) {
2131 s = trailing_symlink(nd);
2132 if (IS_ERR(s)) {
2133 err = PTR_ERR(s);
2134 break;
Al Virobd92d7f2011-03-14 19:54:59 -04002135 }
2136 }
Al Viro9f1fafe2011-03-25 11:00:12 -04002137 if (!err)
2138 err = complete_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04002139
Al Virodeb106c2015-05-08 18:05:21 -04002140 if (!err && nd->flags & LOOKUP_DIRECTORY)
2141 if (!d_can_lookup(nd->path.dentry))
Al Virobd23a532011-03-23 09:56:30 -04002142 err = -ENOTDIR;
Al Viro625b6d12015-05-12 16:36:12 -04002143 if (!err) {
2144 *path = nd->path;
2145 nd->path.mnt = NULL;
2146 nd->path.dentry = NULL;
2147 }
2148 terminate_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04002149 return err;
Al Viroee0827c2011-02-21 23:38:09 -05002150}
Nick Piggin31e6b012011-01-07 17:49:52 +11002151
Al Viro625b6d12015-05-12 16:36:12 -04002152static int filename_lookup(int dfd, struct filename *name, unsigned flags,
Al Viro9ad1aaa2015-05-12 16:44:39 -04002153 struct path *path, struct path *root)
Jeff Layton873f1ee2012-10-10 15:25:29 -04002154{
Al Viro894bc8c2015-05-02 07:16:16 -04002155 int retval;
Al Viro9883d182015-05-13 07:28:08 -04002156 struct nameidata nd;
Al Viroabc9f5b2015-05-12 16:53:42 -04002157 if (IS_ERR(name))
2158 return PTR_ERR(name);
Al Viro9ad1aaa2015-05-12 16:44:39 -04002159 if (unlikely(root)) {
2160 nd.root = *root;
2161 flags |= LOOKUP_ROOT;
2162 }
Al Viro9883d182015-05-13 07:28:08 -04002163 set_nameidata(&nd, dfd, name);
Al Viroc8a53ee2015-05-12 18:43:07 -04002164 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002165 if (unlikely(retval == -ECHILD))
Al Viroc8a53ee2015-05-12 18:43:07 -04002166 retval = path_lookupat(&nd, flags, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002167 if (unlikely(retval == -ESTALE))
Al Viroc8a53ee2015-05-12 18:43:07 -04002168 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002169
2170 if (likely(!retval))
Al Viro625b6d12015-05-12 16:36:12 -04002171 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
Al Viro9883d182015-05-13 07:28:08 -04002172 restore_nameidata();
Al Viroe4bd1c12015-05-12 16:40:39 -04002173 putname(name);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002174 return retval;
2175}
2176
Al Viro8bcb77f2015-05-08 16:59:20 -04002177/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroc8a53ee2015-05-12 18:43:07 -04002178static int path_parentat(struct nameidata *nd, unsigned flags,
Al Viro391172c2015-05-09 11:19:16 -04002179 struct path *parent)
Al Viro8bcb77f2015-05-08 16:59:20 -04002180{
Al Viroc8a53ee2015-05-12 18:43:07 -04002181 const char *s = path_init(nd, flags);
Al Viro368ee9b2015-05-08 17:19:59 -04002182 int err;
2183 if (IS_ERR(s))
2184 return PTR_ERR(s);
2185 err = link_path_walk(s, nd);
Al Viro8bcb77f2015-05-08 16:59:20 -04002186 if (!err)
2187 err = complete_walk(nd);
Al Viro391172c2015-05-09 11:19:16 -04002188 if (!err) {
2189 *parent = nd->path;
2190 nd->path.mnt = NULL;
2191 nd->path.dentry = NULL;
2192 }
2193 terminate_walk(nd);
Al Viro8bcb77f2015-05-08 16:59:20 -04002194 return err;
2195}
2196
Al Viro5c31b6c2015-05-12 17:32:54 -04002197static struct filename *filename_parentat(int dfd, struct filename *name,
Al Viro391172c2015-05-09 11:19:16 -04002198 unsigned int flags, struct path *parent,
2199 struct qstr *last, int *type)
Al Viro8bcb77f2015-05-08 16:59:20 -04002200{
2201 int retval;
Al Viro9883d182015-05-13 07:28:08 -04002202 struct nameidata nd;
Al Viro8bcb77f2015-05-08 16:59:20 -04002203
Al Viro5c31b6c2015-05-12 17:32:54 -04002204 if (IS_ERR(name))
2205 return name;
Al Viro9883d182015-05-13 07:28:08 -04002206 set_nameidata(&nd, dfd, name);
Al Viroc8a53ee2015-05-12 18:43:07 -04002207 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
Al Viro8bcb77f2015-05-08 16:59:20 -04002208 if (unlikely(retval == -ECHILD))
Al Viroc8a53ee2015-05-12 18:43:07 -04002209 retval = path_parentat(&nd, flags, parent);
Al Viro8bcb77f2015-05-08 16:59:20 -04002210 if (unlikely(retval == -ESTALE))
Al Viroc8a53ee2015-05-12 18:43:07 -04002211 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
Al Viro391172c2015-05-09 11:19:16 -04002212 if (likely(!retval)) {
2213 *last = nd.last;
2214 *type = nd.last_type;
2215 audit_inode(name, parent->dentry, LOOKUP_PARENT);
Al Viro5c31b6c2015-05-12 17:32:54 -04002216 } else {
2217 putname(name);
2218 name = ERR_PTR(retval);
Al Viro391172c2015-05-09 11:19:16 -04002219 }
Al Viro9883d182015-05-13 07:28:08 -04002220 restore_nameidata();
Al Viro5c31b6c2015-05-12 17:32:54 -04002221 return name;
Al Viro8bcb77f2015-05-08 16:59:20 -04002222}
2223
Al Viro79714f72012-06-15 03:01:42 +04002224/* does lookup, returns the object with parent locked */
2225struct dentry *kern_path_locked(const char *name, struct path *path)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002226{
Al Viro5c31b6c2015-05-12 17:32:54 -04002227 struct filename *filename;
2228 struct dentry *d;
Al Viro391172c2015-05-09 11:19:16 -04002229 struct qstr last;
2230 int type;
Paul Moore51689102015-01-22 00:00:03 -05002231
Al Viro5c31b6c2015-05-12 17:32:54 -04002232 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2233 &last, &type);
Paul Moore51689102015-01-22 00:00:03 -05002234 if (IS_ERR(filename))
2235 return ERR_CAST(filename);
Al Viro5c31b6c2015-05-12 17:32:54 -04002236 if (unlikely(type != LAST_NORM)) {
Al Viro391172c2015-05-09 11:19:16 -04002237 path_put(path);
Al Viro5c31b6c2015-05-12 17:32:54 -04002238 putname(filename);
2239 return ERR_PTR(-EINVAL);
Al Viro79714f72012-06-15 03:01:42 +04002240 }
Al Viro391172c2015-05-09 11:19:16 -04002241 mutex_lock_nested(&path->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2242 d = __lookup_hash(&last, path->dentry, 0);
Al Viro79714f72012-06-15 03:01:42 +04002243 if (IS_ERR(d)) {
Al Viro391172c2015-05-09 11:19:16 -04002244 mutex_unlock(&path->dentry->d_inode->i_mutex);
2245 path_put(path);
Al Viro79714f72012-06-15 03:01:42 +04002246 }
Paul Moore51689102015-01-22 00:00:03 -05002247 putname(filename);
Al Viro79714f72012-06-15 03:01:42 +04002248 return d;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002249}
2250
Al Virod1811462008-08-02 00:49:18 -04002251int kern_path(const char *name, unsigned int flags, struct path *path)
2252{
Al Viroabc9f5b2015-05-12 16:53:42 -04002253 return filename_lookup(AT_FDCWD, getname_kernel(name),
2254 flags, path, NULL);
Al Virod1811462008-08-02 00:49:18 -04002255}
Al Viro4d359502014-03-14 12:20:17 -04002256EXPORT_SYMBOL(kern_path);
Al Virod1811462008-08-02 00:49:18 -04002257
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002258/**
2259 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2260 * @dentry: pointer to dentry of the base directory
2261 * @mnt: pointer to vfs mount of the base directory
2262 * @name: pointer to file name
2263 * @flags: lookup flags
Al Viroe0a01242011-06-27 17:00:37 -04002264 * @path: pointer to struct path to fill
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002265 */
2266int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2267 const char *name, unsigned int flags,
Al Viroe0a01242011-06-27 17:00:37 -04002268 struct path *path)
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002269{
Al Viro9ad1aaa2015-05-12 16:44:39 -04002270 struct path root = {.mnt = mnt, .dentry = dentry};
Al Viro9ad1aaa2015-05-12 16:44:39 -04002271 /* the first argument of filename_lookup() is ignored with root */
Al Viroabc9f5b2015-05-12 16:53:42 -04002272 return filename_lookup(AT_FDCWD, getname_kernel(name),
2273 flags , path, &root);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002274}
Al Viro4d359502014-03-14 12:20:17 -04002275EXPORT_SYMBOL(vfs_path_lookup);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002276
Christoph Hellwigeead1912007-10-16 23:25:38 -07002277/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07002278 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07002279 * @name: pathname component to lookup
2280 * @base: base directory to lookup from
2281 * @len: maximum length @len should be interpreted to
2282 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07002283 * Note that this routine is purely a helper for filesystem usage and should
Al Viro9e7543e2015-02-23 02:49:48 -05002284 * not be called by generic code.
Christoph Hellwigeead1912007-10-16 23:25:38 -07002285 */
James Morris057f6c02007-04-26 00:12:05 -07002286struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2287{
James Morris057f6c02007-04-26 00:12:05 -07002288 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05002289 unsigned int c;
Miklos Szeredicda309d2012-03-26 12:54:21 +02002290 int err;
James Morris057f6c02007-04-26 00:12:05 -07002291
David Woodhouse2f9092e2009-04-20 23:18:37 +01002292 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2293
Al Viro6a96ba52011-03-07 23:49:20 -05002294 this.name = name;
2295 this.len = len;
Linus Torvalds0145acc2012-03-02 14:32:59 -08002296 this.hash = full_name_hash(name, len);
Al Viro6a96ba52011-03-07 23:49:20 -05002297 if (!len)
2298 return ERR_PTR(-EACCES);
2299
Al Viro21d8a152012-11-29 22:17:21 -05002300 if (unlikely(name[0] == '.')) {
2301 if (len < 2 || (len == 2 && name[1] == '.'))
2302 return ERR_PTR(-EACCES);
2303 }
2304
Al Viro6a96ba52011-03-07 23:49:20 -05002305 while (len--) {
2306 c = *(const unsigned char *)name++;
2307 if (c == '/' || c == '\0')
2308 return ERR_PTR(-EACCES);
Al Viro6a96ba52011-03-07 23:49:20 -05002309 }
Al Viro5a202bc2011-03-08 14:17:44 -05002310 /*
2311 * See if the low-level filesystem might want
2312 * to use its own hash..
2313 */
2314 if (base->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002315 int err = base->d_op->d_hash(base, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05002316 if (err < 0)
2317 return ERR_PTR(err);
2318 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07002319
Miklos Szeredicda309d2012-03-26 12:54:21 +02002320 err = inode_permission(base->d_inode, MAY_EXEC);
2321 if (err)
2322 return ERR_PTR(err);
2323
Al Viro72bd8662012-06-10 17:17:17 -04002324 return __lookup_hash(&this, base, 0);
James Morris057f6c02007-04-26 00:12:05 -07002325}
Al Viro4d359502014-03-14 12:20:17 -04002326EXPORT_SYMBOL(lookup_one_len);
James Morris057f6c02007-04-26 00:12:05 -07002327
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002328int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2329 struct path *path, int *empty)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330{
Al Viroabc9f5b2015-05-12 16:53:42 -04002331 return filename_lookup(dfd, getname_flags(name, flags, empty),
2332 flags, path, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333}
Al Virob853a162015-05-13 09:12:02 -04002334EXPORT_SYMBOL(user_path_at_empty);
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002335
Jeff Layton873f1ee2012-10-10 15:25:29 -04002336/*
2337 * NB: most callers don't do anything directly with the reference to the
2338 * to struct filename, but the nd->last pointer points into the name string
2339 * allocated by getname. So we must hold the reference to it until all
2340 * path-walking is complete.
2341 */
Al Viroa2ec4a22015-05-13 06:57:49 -04002342static inline struct filename *
Al Virof5beed72015-04-30 16:09:11 -04002343user_path_parent(int dfd, const char __user *path,
2344 struct path *parent,
2345 struct qstr *last,
2346 int *type,
Jeff Layton9e790bd2012-12-11 12:10:09 -05002347 unsigned int flags)
Al Viro2ad94ae2008-07-21 09:32:51 -04002348{
Jeff Layton9e790bd2012-12-11 12:10:09 -05002349 /* only LOOKUP_REVAL is allowed in extra flags */
Al Viro5c31b6c2015-05-12 17:32:54 -04002350 return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL,
2351 parent, last, type);
Al Viro2ad94ae2008-07-21 09:32:51 -04002352}
2353
Jeff Layton80334262013-07-26 06:23:25 -04002354/**
Al Viro197df042013-09-08 14:03:27 -04002355 * mountpoint_last - look up last component for umount
Jeff Layton80334262013-07-26 06:23:25 -04002356 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2357 * @path: pointer to container for result
2358 *
2359 * This is a special lookup_last function just for umount. In this case, we
2360 * need to resolve the path without doing any revalidation.
2361 *
2362 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2363 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2364 * in almost all cases, this lookup will be served out of the dcache. The only
2365 * cases where it won't are if nd->last refers to a symlink or the path is
2366 * bogus and it doesn't exist.
2367 *
2368 * Returns:
2369 * -error: if there was an error during lookup. This includes -ENOENT if the
2370 * lookup found a negative dentry. The nd->path reference will also be
2371 * put in this case.
2372 *
2373 * 0: if we successfully resolved nd->path and found it to not to be a
2374 * symlink that needs to be followed. "path" will also be populated.
2375 * The nd->path reference will also be put.
2376 *
2377 * 1: if we successfully resolved nd->last and found it to be a symlink
2378 * that needs to be followed. "path" will be populated with the path
2379 * to the link, and nd->path will *not* be put.
2380 */
2381static int
Al Viro197df042013-09-08 14:03:27 -04002382mountpoint_last(struct nameidata *nd, struct path *path)
Jeff Layton80334262013-07-26 06:23:25 -04002383{
2384 int error = 0;
2385 struct dentry *dentry;
2386 struct dentry *dir = nd->path.dentry;
2387
Al Viro35759522013-09-08 13:41:33 -04002388 /* If we're in rcuwalk, drop out of it to handle last component */
2389 if (nd->flags & LOOKUP_RCU) {
Al Viro6e9918b2015-05-05 09:26:05 -04002390 if (unlazy_walk(nd, NULL, 0))
Al Virodeb106c2015-05-08 18:05:21 -04002391 return -ECHILD;
Jeff Layton80334262013-07-26 06:23:25 -04002392 }
2393
2394 nd->flags &= ~LOOKUP_PARENT;
2395
2396 if (unlikely(nd->last_type != LAST_NORM)) {
2397 error = handle_dots(nd, nd->last_type);
Al Viro35759522013-09-08 13:41:33 -04002398 if (error)
Al Virodeb106c2015-05-08 18:05:21 -04002399 return error;
Al Viro35759522013-09-08 13:41:33 -04002400 dentry = dget(nd->path.dentry);
2401 goto done;
Jeff Layton80334262013-07-26 06:23:25 -04002402 }
2403
2404 mutex_lock(&dir->d_inode->i_mutex);
2405 dentry = d_lookup(dir, &nd->last);
2406 if (!dentry) {
2407 /*
2408 * No cached dentry. Mounted dentries are pinned in the cache,
2409 * so that means that this dentry is probably a symlink or the
2410 * path doesn't actually point to a mounted dentry.
2411 */
2412 dentry = d_alloc(dir, &nd->last);
2413 if (!dentry) {
Dave Jonesbcceeeb2013-09-10 17:04:25 -04002414 mutex_unlock(&dir->d_inode->i_mutex);
Al Virodeb106c2015-05-08 18:05:21 -04002415 return -ENOMEM;
Jeff Layton80334262013-07-26 06:23:25 -04002416 }
Al Viro35759522013-09-08 13:41:33 -04002417 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
Dave Jonesbcceeeb2013-09-10 17:04:25 -04002418 if (IS_ERR(dentry)) {
2419 mutex_unlock(&dir->d_inode->i_mutex);
Al Virodeb106c2015-05-08 18:05:21 -04002420 return PTR_ERR(dentry);
Dave Jonesbcceeeb2013-09-10 17:04:25 -04002421 }
Jeff Layton80334262013-07-26 06:23:25 -04002422 }
2423 mutex_unlock(&dir->d_inode->i_mutex);
2424
Al Viro35759522013-09-08 13:41:33 -04002425done:
David Howells698934d2015-03-17 17:33:52 +00002426 if (d_is_negative(dentry)) {
Al Viro35759522013-09-08 13:41:33 -04002427 dput(dentry);
Al Virodeb106c2015-05-08 18:05:21 -04002428 return -ENOENT;
Jeff Layton80334262013-07-26 06:23:25 -04002429 }
Al Viro191d7f72015-05-04 08:26:45 -04002430 if (nd->depth)
2431 put_link(nd);
Al Viro35759522013-09-08 13:41:33 -04002432 path->dentry = dentry;
Vasily Averin295dc392014-07-21 12:30:23 +04002433 path->mnt = nd->path.mnt;
Al Viro181548c2015-05-07 19:54:34 -04002434 error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW,
2435 d_backing_inode(dentry), 0);
Al Virodeb106c2015-05-08 18:05:21 -04002436 if (unlikely(error))
Al Virod63ff282015-05-04 18:13:23 -04002437 return error;
Vasily Averin295dc392014-07-21 12:30:23 +04002438 mntget(path->mnt);
Al Viro35759522013-09-08 13:41:33 -04002439 follow_mount(path);
Al Virodeb106c2015-05-08 18:05:21 -04002440 return 0;
Jeff Layton80334262013-07-26 06:23:25 -04002441}
2442
2443/**
Al Viro197df042013-09-08 14:03:27 -04002444 * path_mountpoint - look up a path to be umounted
Masanari Iida2a78b852015-09-09 15:39:23 -07002445 * @nd: lookup context
Jeff Layton80334262013-07-26 06:23:25 -04002446 * @flags: lookup flags
Al Viroc8a53ee2015-05-12 18:43:07 -04002447 * @path: pointer to container for result
Jeff Layton80334262013-07-26 06:23:25 -04002448 *
2449 * Look up the given name, but don't attempt to revalidate the last component.
Randy Dunlap606d6fe2013-10-19 14:56:55 -07002450 * Returns 0 and "path" will be valid on success; Returns error otherwise.
Jeff Layton80334262013-07-26 06:23:25 -04002451 */
2452static int
Al Viroc8a53ee2015-05-12 18:43:07 -04002453path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
Jeff Layton80334262013-07-26 06:23:25 -04002454{
Al Viroc8a53ee2015-05-12 18:43:07 -04002455 const char *s = path_init(nd, flags);
Al Viro368ee9b2015-05-08 17:19:59 -04002456 int err;
2457 if (IS_ERR(s))
2458 return PTR_ERR(s);
Al Viro3bdba282015-05-08 17:37:07 -04002459 while (!(err = link_path_walk(s, nd)) &&
2460 (err = mountpoint_last(nd, path)) > 0) {
2461 s = trailing_symlink(nd);
2462 if (IS_ERR(s)) {
2463 err = PTR_ERR(s);
Jeff Layton80334262013-07-26 06:23:25 -04002464 break;
Al Viro3bdba282015-05-08 17:37:07 -04002465 }
Jeff Layton80334262013-07-26 06:23:25 -04002466 }
Al Virodeb106c2015-05-08 18:05:21 -04002467 terminate_walk(nd);
Jeff Layton80334262013-07-26 06:23:25 -04002468 return err;
2469}
2470
Al Viro2d864652013-09-08 20:18:44 -04002471static int
Al Viro668696d2015-02-22 19:44:00 -05002472filename_mountpoint(int dfd, struct filename *name, struct path *path,
Al Viro2d864652013-09-08 20:18:44 -04002473 unsigned int flags)
2474{
Al Viro9883d182015-05-13 07:28:08 -04002475 struct nameidata nd;
Al Virocbaab2d2015-01-22 02:49:00 -05002476 int error;
Al Viro668696d2015-02-22 19:44:00 -05002477 if (IS_ERR(name))
2478 return PTR_ERR(name);
Al Viro9883d182015-05-13 07:28:08 -04002479 set_nameidata(&nd, dfd, name);
Al Viroc8a53ee2015-05-12 18:43:07 -04002480 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
Al Viro2d864652013-09-08 20:18:44 -04002481 if (unlikely(error == -ECHILD))
Al Viroc8a53ee2015-05-12 18:43:07 -04002482 error = path_mountpoint(&nd, flags, path);
Al Viro2d864652013-09-08 20:18:44 -04002483 if (unlikely(error == -ESTALE))
Al Viroc8a53ee2015-05-12 18:43:07 -04002484 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
Al Viro2d864652013-09-08 20:18:44 -04002485 if (likely(!error))
Al Viro668696d2015-02-22 19:44:00 -05002486 audit_inode(name, path->dentry, 0);
Al Viro9883d182015-05-13 07:28:08 -04002487 restore_nameidata();
Al Viro668696d2015-02-22 19:44:00 -05002488 putname(name);
Al Viro2d864652013-09-08 20:18:44 -04002489 return error;
2490}
2491
Jeff Layton80334262013-07-26 06:23:25 -04002492/**
Al Viro197df042013-09-08 14:03:27 -04002493 * user_path_mountpoint_at - lookup a path from userland in order to umount it
Jeff Layton80334262013-07-26 06:23:25 -04002494 * @dfd: directory file descriptor
2495 * @name: pathname from userland
2496 * @flags: lookup flags
2497 * @path: pointer to container to hold result
2498 *
2499 * A umount is a special case for path walking. We're not actually interested
2500 * in the inode in this situation, and ESTALE errors can be a problem. We
2501 * simply want track down the dentry and vfsmount attached at the mountpoint
2502 * and avoid revalidating the last component.
2503 *
2504 * Returns 0 and populates "path" on success.
2505 */
2506int
Al Viro197df042013-09-08 14:03:27 -04002507user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
Jeff Layton80334262013-07-26 06:23:25 -04002508 struct path *path)
2509{
Al Virocbaab2d2015-01-22 02:49:00 -05002510 return filename_mountpoint(dfd, getname(name), path, flags);
Jeff Layton80334262013-07-26 06:23:25 -04002511}
2512
Al Viro2d864652013-09-08 20:18:44 -04002513int
2514kern_path_mountpoint(int dfd, const char *name, struct path *path,
2515 unsigned int flags)
2516{
Al Virocbaab2d2015-01-22 02:49:00 -05002517 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
Al Viro2d864652013-09-08 20:18:44 -04002518}
2519EXPORT_SYMBOL(kern_path_mountpoint);
2520
Miklos Szeredicbdf35b2014-10-24 00:14:36 +02002521int __check_sticky(struct inode *dir, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522{
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002523 kuid_t fsuid = current_fsuid();
David Howellsda9592e2008-11-14 10:39:05 +11002524
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002525 if (uid_eq(inode->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526 return 0;
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002527 if (uid_eq(dir->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528 return 0;
Andy Lutomirski23adbe12014-06-10 12:45:42 -07002529 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530}
Miklos Szeredicbdf35b2014-10-24 00:14:36 +02002531EXPORT_SYMBOL(__check_sticky);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532
2533/*
2534 * Check whether we can remove a link victim from directory dir, check
2535 * whether the type of victim is right.
2536 * 1. We can't do it if dir is read-only (done in permission())
2537 * 2. We should have write and exec permissions on dir
2538 * 3. We can't remove anything from append-only dir
2539 * 4. We can't do anything with immutable dir (done in permission())
2540 * 5. If the sticky bit on dir is set we should either
2541 * a. be owner of dir, or
2542 * b. be owner of victim, or
2543 * c. have CAP_FOWNER capability
2544 * 6. If the victim is append-only or immutable we can't do antyhing with
2545 * links pointing to it.
2546 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2547 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2548 * 9. We can't remove a root or mountpoint.
2549 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2550 * nfs_async_unlink().
2551 */
David Howellsb18825a2013-09-12 19:22:53 +01002552static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553{
David Howells63afdfc2015-05-06 15:59:00 +01002554 struct inode *inode = d_backing_inode(victim);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 int error;
2556
David Howellsb18825a2013-09-12 19:22:53 +01002557 if (d_is_negative(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 return -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01002559 BUG_ON(!inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560
2561 BUG_ON(victim->d_parent->d_inode != dir);
Jeff Layton4fa6b5e2012-10-10 15:25:25 -04002562 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563
Al Virof419a2e2008-07-22 00:07:17 -04002564 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565 if (error)
2566 return error;
2567 if (IS_APPEND(dir))
2568 return -EPERM;
David Howellsb18825a2013-09-12 19:22:53 +01002569
2570 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2571 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572 return -EPERM;
2573 if (isdir) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002574 if (!d_is_dir(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575 return -ENOTDIR;
2576 if (IS_ROOT(victim))
2577 return -EBUSY;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002578 } else if (d_is_dir(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 return -EISDIR;
2580 if (IS_DEADDIR(dir))
2581 return -ENOENT;
2582 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2583 return -EBUSY;
2584 return 0;
2585}
2586
2587/* Check whether we can create an object with dentry child in directory
2588 * dir.
2589 * 1. We can't do it if child already exists (open has special treatment for
2590 * this case, but since we are inlined it's OK)
2591 * 2. We can't do it if dir is read-only (done in permission())
2592 * 3. We should have write and exec permissions on dir
2593 * 4. We can't do it if dir is immutable (done in permission())
2594 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02002595static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596{
Jeff Layton14e972b2013-05-08 10:25:58 -04002597 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598 if (child->d_inode)
2599 return -EEXIST;
2600 if (IS_DEADDIR(dir))
2601 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04002602 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603}
2604
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605/*
2606 * p1 and p2 should be directories on the same fs.
2607 */
2608struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2609{
2610 struct dentry *p;
2611
2612 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07002613 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002614 return NULL;
2615 }
2616
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002617 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002619 p = d_ancestor(p2, p1);
2620 if (p) {
2621 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2622 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2623 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002624 }
2625
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002626 p = d_ancestor(p1, p2);
2627 if (p) {
2628 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2629 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2630 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631 }
2632
Ingo Molnarf2eace22006-07-03 00:25:05 -07002633 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Miklos Szeredid1b72cc2014-10-27 15:42:01 +01002634 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635 return NULL;
2636}
Al Viro4d359502014-03-14 12:20:17 -04002637EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638
2639void unlock_rename(struct dentry *p1, struct dentry *p2)
2640{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002641 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002643 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002644 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645 }
2646}
Al Viro4d359502014-03-14 12:20:17 -04002647EXPORT_SYMBOL(unlock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648
Al Viro4acdaf22011-07-26 01:42:34 -04002649int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viro312b63f2012-06-10 18:09:36 -04002650 bool want_excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002652 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002653 if (error)
2654 return error;
2655
Al Viroacfa4382008-12-04 10:06:33 -05002656 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657 return -EACCES; /* shouldn't it be ENOSYS? */
2658 mode &= S_IALLUGO;
2659 mode |= S_IFREG;
2660 error = security_inode_create(dir, dentry, mode);
2661 if (error)
2662 return error;
Al Viro312b63f2012-06-10 18:09:36 -04002663 error = dir->i_op->create(dir, dentry, mode, want_excl);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002664 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002665 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666 return error;
2667}
Al Viro4d359502014-03-14 12:20:17 -04002668EXPORT_SYMBOL(vfs_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669
Al Viro73d049a2011-03-11 12:08:24 -05002670static int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002672 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673 struct inode *inode = dentry->d_inode;
2674 int error;
2675
Al Virobcda7652011-03-13 16:42:14 -04002676 /* O_PATH? */
2677 if (!acc_mode)
2678 return 0;
2679
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680 if (!inode)
2681 return -ENOENT;
2682
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002683 switch (inode->i_mode & S_IFMT) {
2684 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002685 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002686 case S_IFDIR:
2687 if (acc_mode & MAY_WRITE)
2688 return -EISDIR;
2689 break;
2690 case S_IFBLK:
2691 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002692 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002694 /*FALLTHRU*/
2695 case S_IFIFO:
2696 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002698 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002699 }
Dave Hansenb41572e2007-10-16 23:31:14 -07002700
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002701 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07002702 if (error)
2703 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05002704
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705 /*
2706 * An append-only file must be opened in append mode for writing.
2707 */
2708 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05002709 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05002710 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05002712 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713 }
2714
2715 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002716 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05002717 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718
J. Bruce Fieldsf3c7691e2011-09-21 10:58:13 -04002719 return 0;
Al Viro7715b522009-12-16 03:54:00 -05002720}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002721
Jeff Laytone1181ee2010-12-07 16:19:50 -05002722static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002723{
Jeff Laytone1181ee2010-12-07 16:19:50 -05002724 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002725 struct inode *inode = path->dentry->d_inode;
2726 int error = get_write_access(inode);
2727 if (error)
2728 return error;
2729 /*
2730 * Refuse to truncate files with mandatory locks held on them.
2731 */
Jeff Laytond7a06982014-03-10 09:54:15 -04002732 error = locks_verify_locked(filp);
Al Viro7715b522009-12-16 03:54:00 -05002733 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002734 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002735 if (!error) {
2736 error = do_truncate(path->dentry, 0,
2737 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002738 filp);
Al Viro7715b522009-12-16 03:54:00 -05002739 }
2740 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002741 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742}
2743
Dave Hansend57999e2008-02-15 14:37:27 -08002744static inline int open_to_namei_flags(int flag)
2745{
Al Viro8a5e9292011-06-25 19:15:54 -04002746 if ((flag & O_ACCMODE) == 3)
2747 flag--;
Dave Hansend57999e2008-02-15 14:37:27 -08002748 return flag;
2749}
2750
Miklos Szeredid18e9002012-06-05 15:10:17 +02002751static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2752{
2753 int error = security_path_mknod(dir, dentry, mode, 0);
2754 if (error)
2755 return error;
2756
2757 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2758 if (error)
2759 return error;
2760
2761 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2762}
2763
David Howells1acf0af2012-06-14 16:13:46 +01002764/*
2765 * Attempt to atomically look up, create and open a file from a negative
2766 * dentry.
2767 *
2768 * Returns 0 if successful. The file will have been created and attached to
2769 * @file by the filesystem calling finish_open().
2770 *
2771 * Returns 1 if the file was looked up only or didn't need creating. The
2772 * caller will need to perform the open themselves. @path will have been
2773 * updated to point to the new dentry. This may be negative.
2774 *
2775 * Returns an error code otherwise.
2776 */
Al Viro2675a4e2012-06-22 12:41:10 +04002777static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2778 struct path *path, struct file *file,
2779 const struct open_flags *op,
Al Viro64894cf2012-07-31 00:53:35 +04002780 bool got_write, bool need_lookup,
Al Viro2675a4e2012-06-22 12:41:10 +04002781 int *opened)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002782{
2783 struct inode *dir = nd->path.dentry->d_inode;
2784 unsigned open_flag = open_to_namei_flags(op->open_flag);
2785 umode_t mode;
2786 int error;
2787 int acc_mode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002788 int create_error = 0;
2789 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
Miklos Szeredi116cc022013-09-16 14:52:05 +02002790 bool excl;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002791
2792 BUG_ON(dentry->d_inode);
2793
2794 /* Don't create child dentry for a dead directory. */
2795 if (unlikely(IS_DEADDIR(dir))) {
Al Viro2675a4e2012-06-22 12:41:10 +04002796 error = -ENOENT;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002797 goto out;
2798 }
2799
Miklos Szeredi62b259d2012-08-15 13:01:24 +02002800 mode = op->mode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002801 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2802 mode &= ~current_umask();
2803
Miklos Szeredi116cc022013-09-16 14:52:05 +02002804 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2805 if (excl)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002806 open_flag &= ~O_TRUNC;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002807
2808 /*
2809 * Checking write permission is tricky, bacuse we don't know if we are
2810 * going to actually need it: O_CREAT opens should work as long as the
2811 * file exists. But checking existence breaks atomicity. The trick is
2812 * to check access and if not granted clear O_CREAT from the flags.
2813 *
2814 * Another problem is returing the "right" error value (e.g. for an
2815 * O_EXCL open we want to return EEXIST not EROFS).
2816 */
Al Viro64894cf2012-07-31 00:53:35 +04002817 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2818 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2819 if (!(open_flag & O_CREAT)) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002820 /*
2821 * No O_CREATE -> atomicity not a requirement -> fall
2822 * back to lookup + open
2823 */
2824 goto no_open;
2825 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2826 /* Fall back and fail with the right error */
Al Viro64894cf2012-07-31 00:53:35 +04002827 create_error = -EROFS;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002828 goto no_open;
2829 } else {
2830 /* No side effects, safe to clear O_CREAT */
Al Viro64894cf2012-07-31 00:53:35 +04002831 create_error = -EROFS;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002832 open_flag &= ~O_CREAT;
2833 }
2834 }
2835
2836 if (open_flag & O_CREAT) {
Miklos Szeredi38227f72012-08-15 13:01:24 +02002837 error = may_o_create(&nd->path, dentry, mode);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002838 if (error) {
2839 create_error = error;
2840 if (open_flag & O_EXCL)
2841 goto no_open;
2842 open_flag &= ~O_CREAT;
2843 }
2844 }
2845
2846 if (nd->flags & LOOKUP_DIRECTORY)
2847 open_flag |= O_DIRECTORY;
2848
Al Viro30d90492012-06-22 12:40:19 +04002849 file->f_path.dentry = DENTRY_NOT_SET;
2850 file->f_path.mnt = nd->path.mnt;
2851 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
Al Viro47237682012-06-10 05:01:45 -04002852 opened);
Al Virod9585272012-06-22 12:39:14 +04002853 if (error < 0) {
Al Virod9585272012-06-22 12:39:14 +04002854 if (create_error && error == -ENOENT)
2855 error = create_error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002856 goto out;
2857 }
2858
Al Virod9585272012-06-22 12:39:14 +04002859 if (error) { /* returned 1, that is */
Al Viro30d90492012-06-22 12:40:19 +04002860 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
Al Viro2675a4e2012-06-22 12:41:10 +04002861 error = -EIO;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002862 goto out;
2863 }
Al Viro30d90492012-06-22 12:40:19 +04002864 if (file->f_path.dentry) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002865 dput(dentry);
Al Viro30d90492012-06-22 12:40:19 +04002866 dentry = file->f_path.dentry;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002867 }
Al Viro03da6332013-09-16 19:22:33 -04002868 if (*opened & FILE_CREATED)
2869 fsnotify_create(dir, dentry);
2870 if (!dentry->d_inode) {
2871 WARN_ON(*opened & FILE_CREATED);
2872 if (create_error) {
2873 error = create_error;
2874 goto out;
2875 }
2876 } else {
2877 if (excl && !(*opened & FILE_CREATED)) {
2878 error = -EEXIST;
2879 goto out;
2880 }
Sage Weil62b2ce92012-08-15 13:30:12 -07002881 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02002882 goto looked_up;
2883 }
2884
2885 /*
2886 * We didn't have the inode before the open, so check open permission
2887 * here.
2888 */
Al Viro03da6332013-09-16 19:22:33 -04002889 acc_mode = op->acc_mode;
2890 if (*opened & FILE_CREATED) {
2891 WARN_ON(!(open_flag & O_CREAT));
2892 fsnotify_create(dir, dentry);
2893 acc_mode = MAY_OPEN;
2894 }
Al Viro2675a4e2012-06-22 12:41:10 +04002895 error = may_open(&file->f_path, acc_mode, open_flag);
2896 if (error)
2897 fput(file);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002898
2899out:
2900 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04002901 return error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002902
Miklos Szeredid18e9002012-06-05 15:10:17 +02002903no_open:
2904 if (need_lookup) {
Al Viro72bd8662012-06-10 17:17:17 -04002905 dentry = lookup_real(dir, dentry, nd->flags);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002906 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002907 return PTR_ERR(dentry);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002908
2909 if (create_error) {
2910 int open_flag = op->open_flag;
2911
Al Viro2675a4e2012-06-22 12:41:10 +04002912 error = create_error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002913 if ((open_flag & O_EXCL)) {
2914 if (!dentry->d_inode)
2915 goto out;
2916 } else if (!dentry->d_inode) {
2917 goto out;
2918 } else if ((open_flag & O_TRUNC) &&
David Howellse36cb0b2015-01-29 12:02:35 +00002919 d_is_reg(dentry)) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002920 goto out;
2921 }
2922 /* will fail later, go on to get the right error */
2923 }
2924 }
2925looked_up:
2926 path->dentry = dentry;
2927 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04002928 return 1;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002929}
2930
Nick Piggin31e6b012011-01-07 17:49:52 +11002931/*
David Howells1acf0af2012-06-14 16:13:46 +01002932 * Look up and maybe create and open the last component.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002933 *
2934 * Must be called with i_mutex held on parent.
2935 *
David Howells1acf0af2012-06-14 16:13:46 +01002936 * Returns 0 if the file was successfully atomically created (if necessary) and
2937 * opened. In this case the file will be returned attached to @file.
2938 *
2939 * Returns 1 if the file was not completely opened at this time, though lookups
2940 * and creations will have been performed and the dentry returned in @path will
2941 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2942 * specified then a negative dentry may be returned.
2943 *
2944 * An error code is returned otherwise.
2945 *
2946 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2947 * cleared otherwise prior to returning.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002948 */
Al Viro2675a4e2012-06-22 12:41:10 +04002949static int lookup_open(struct nameidata *nd, struct path *path,
2950 struct file *file,
2951 const struct open_flags *op,
Al Viro64894cf2012-07-31 00:53:35 +04002952 bool got_write, int *opened)
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002953{
2954 struct dentry *dir = nd->path.dentry;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002955 struct inode *dir_inode = dir->d_inode;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002956 struct dentry *dentry;
2957 int error;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002958 bool need_lookup;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002959
Al Viro47237682012-06-10 05:01:45 -04002960 *opened &= ~FILE_CREATED;
Al Viro201f9562012-06-22 12:42:10 +04002961 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002962 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002963 return PTR_ERR(dentry);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002964
Miklos Szeredid18e9002012-06-05 15:10:17 +02002965 /* Cached positive dentry: will open in f_op->open */
2966 if (!need_lookup && dentry->d_inode)
2967 goto out_no_open;
2968
2969 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
Al Viro64894cf2012-07-31 00:53:35 +04002970 return atomic_open(nd, dentry, path, file, op, got_write,
Al Viro47237682012-06-10 05:01:45 -04002971 need_lookup, opened);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002972 }
2973
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002974 if (need_lookup) {
2975 BUG_ON(dentry->d_inode);
2976
Al Viro72bd8662012-06-10 17:17:17 -04002977 dentry = lookup_real(dir_inode, dentry, nd->flags);
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002978 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002979 return PTR_ERR(dentry);
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002980 }
2981
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002982 /* Negative dentry, just create the file */
2983 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2984 umode_t mode = op->mode;
2985 if (!IS_POSIXACL(dir->d_inode))
2986 mode &= ~current_umask();
2987 /*
2988 * This write is needed to ensure that a
2989 * rw->ro transition does not occur between
2990 * the time when the file is created and when
2991 * a permanent write count is taken through
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002992 * the 'struct file' in finish_open().
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002993 */
Al Viro64894cf2012-07-31 00:53:35 +04002994 if (!got_write) {
2995 error = -EROFS;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002996 goto out_dput;
Al Viro64894cf2012-07-31 00:53:35 +04002997 }
Al Viro47237682012-06-10 05:01:45 -04002998 *opened |= FILE_CREATED;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002999 error = security_path_mknod(&nd->path, dentry, mode, 0);
3000 if (error)
3001 goto out_dput;
Al Viro312b63f2012-06-10 18:09:36 -04003002 error = vfs_create(dir->d_inode, dentry, mode,
3003 nd->flags & LOOKUP_EXCL);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003004 if (error)
3005 goto out_dput;
3006 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02003007out_no_open:
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003008 path->dentry = dentry;
3009 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04003010 return 1;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003011
3012out_dput:
3013 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04003014 return error;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003015}
3016
3017/*
Al Virofe2d35f2011-03-05 22:58:25 -05003018 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11003019 */
Al Viro896475d2015-04-22 18:02:17 -04003020static int do_last(struct nameidata *nd,
Al Viro2675a4e2012-06-22 12:41:10 +04003021 struct file *file, const struct open_flags *op,
Al Viro76ae2a52015-05-12 18:44:32 -04003022 int *opened)
Al Virofb1cc552009-12-24 01:58:28 -05003023{
Al Viroa1e28032009-12-24 02:12:06 -05003024 struct dentry *dir = nd->path.dentry;
Al Viroca344a892011-03-09 00:36:45 -05003025 int open_flag = op->open_flag;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003026 bool will_truncate = (open_flag & O_TRUNC) != 0;
Al Viro64894cf2012-07-31 00:53:35 +04003027 bool got_write = false;
Al Virobcda7652011-03-13 16:42:14 -04003028 int acc_mode = op->acc_mode;
Al Viro254cf582015-05-05 09:40:46 -04003029 unsigned seq;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02003030 struct inode *inode;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003031 struct path save_parent = { .dentry = NULL, .mnt = NULL };
Al Viro896475d2015-04-22 18:02:17 -04003032 struct path path;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003033 bool retried = false;
Al Viro16c2cd72011-02-22 15:50:10 -05003034 int error;
Al Virofb1cc552009-12-24 01:58:28 -05003035
Al Viroc3e380b2011-02-23 13:39:45 -05003036 nd->flags &= ~LOOKUP_PARENT;
3037 nd->flags |= op->intent;
3038
Al Virobc77daa2013-06-06 09:12:33 -04003039 if (nd->last_type != LAST_NORM) {
Al Virofe2d35f2011-03-05 22:58:25 -05003040 error = handle_dots(nd, nd->last_type);
Al Virodeb106c2015-05-08 18:05:21 -04003041 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04003042 return error;
Miklos Szeredie83db162012-06-05 15:10:29 +02003043 goto finish_open;
Al Viro1f36f772009-12-26 10:56:19 -05003044 }
Al Viro67ee3ad2009-12-26 07:01:01 -05003045
Al Viroca344a892011-03-09 00:36:45 -05003046 if (!(open_flag & O_CREAT)) {
Al Virofe2d35f2011-03-05 22:58:25 -05003047 if (nd->last.name[nd->last.len])
3048 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3049 /* we _can_ be in RCU mode here */
Al Viro254cf582015-05-05 09:40:46 -04003050 error = lookup_fast(nd, &path, &inode, &seq);
Miklos Szeredi71574862012-06-05 15:10:14 +02003051 if (likely(!error))
3052 goto finish_lookup;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02003053
Miklos Szeredi71574862012-06-05 15:10:14 +02003054 if (error < 0)
Al Virodeb106c2015-05-08 18:05:21 -04003055 return error;
Miklos Szeredi37d7fff2012-06-05 15:10:12 +02003056
Miklos Szeredi71574862012-06-05 15:10:14 +02003057 BUG_ON(nd->inode != dir->d_inode);
Miklos Szeredib6183df2012-06-05 15:10:13 +02003058 } else {
3059 /* create side of things */
3060 /*
3061 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3062 * has been cleared when we got to the last component we are
3063 * about to look up
3064 */
3065 error = complete_walk(nd);
Al Viroe8bb73d2015-05-08 16:28:42 -04003066 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003067 return error;
Miklos Szeredib6183df2012-06-05 15:10:13 +02003068
Al Viro76ae2a52015-05-12 18:44:32 -04003069 audit_inode(nd->name, dir, LOOKUP_PARENT);
Miklos Szeredib6183df2012-06-05 15:10:13 +02003070 /* trailing slashes? */
Al Virodeb106c2015-05-08 18:05:21 -04003071 if (unlikely(nd->last.name[nd->last.len]))
3072 return -EISDIR;
Al Virofe2d35f2011-03-05 22:58:25 -05003073 }
3074
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003075retry_lookup:
Al Viro64894cf2012-07-31 00:53:35 +04003076 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3077 error = mnt_want_write(nd->path.mnt);
3078 if (!error)
3079 got_write = true;
3080 /*
3081 * do _not_ fail yet - we might not need that or fail with
3082 * a different error; let lookup_open() decide; we'll be
3083 * dropping this one anyway.
3084 */
3085 }
Al Viroa1e28032009-12-24 02:12:06 -05003086 mutex_lock(&dir->d_inode->i_mutex);
Al Viro896475d2015-04-22 18:02:17 -04003087 error = lookup_open(nd, &path, file, op, got_write, opened);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003088 mutex_unlock(&dir->d_inode->i_mutex);
Al Viroa1e28032009-12-24 02:12:06 -05003089
Al Viro2675a4e2012-06-22 12:41:10 +04003090 if (error <= 0) {
3091 if (error)
Miklos Szeredid18e9002012-06-05 15:10:17 +02003092 goto out;
3093
Al Viro47237682012-06-10 05:01:45 -04003094 if ((*opened & FILE_CREATED) ||
Al Viro496ad9a2013-01-23 17:07:38 -05003095 !S_ISREG(file_inode(file)->i_mode))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003096 will_truncate = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003097
Al Viro76ae2a52015-05-12 18:44:32 -04003098 audit_inode(nd->name, file->f_path.dentry, 0);
Miklos Szeredid18e9002012-06-05 15:10:17 +02003099 goto opened;
3100 }
Al Virofb1cc552009-12-24 01:58:28 -05003101
Al Viro47237682012-06-10 05:01:45 -04003102 if (*opened & FILE_CREATED) {
Al Viro9b44f1b2011-03-09 00:17:27 -05003103 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05003104 open_flag &= ~O_TRUNC;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003105 will_truncate = false;
Al Virobcda7652011-03-13 16:42:14 -04003106 acc_mode = MAY_OPEN;
Al Viro896475d2015-04-22 18:02:17 -04003107 path_to_nameidata(&path, nd);
Miklos Szeredie83db162012-06-05 15:10:29 +02003108 goto finish_open_created;
Al Virofb1cc552009-12-24 01:58:28 -05003109 }
3110
3111 /*
Jeff Layton3134f372012-07-25 10:19:47 -04003112 * create/update audit record if it already exists.
Al Virofb1cc552009-12-24 01:58:28 -05003113 */
Al Viro896475d2015-04-22 18:02:17 -04003114 if (d_is_positive(path.dentry))
Al Viro76ae2a52015-05-12 18:44:32 -04003115 audit_inode(nd->name, path.dentry, 0);
Al Virofb1cc552009-12-24 01:58:28 -05003116
Miklos Szeredid18e9002012-06-05 15:10:17 +02003117 /*
3118 * If atomic_open() acquired write access it is dropped now due to
3119 * possible mount and symlink following (this might be optimized away if
3120 * necessary...)
3121 */
Al Viro64894cf2012-07-31 00:53:35 +04003122 if (got_write) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02003123 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04003124 got_write = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003125 }
3126
Al Virodeb106c2015-05-08 18:05:21 -04003127 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3128 path_to_nameidata(&path, nd);
3129 return -EEXIST;
3130 }
Al Virofb1cc552009-12-24 01:58:28 -05003131
NeilBrown756daf22015-03-23 13:37:38 +11003132 error = follow_managed(&path, nd);
Al Virodeb106c2015-05-08 18:05:21 -04003133 if (unlikely(error < 0))
3134 return error;
Al Viroa3fbbde2011-11-07 21:21:26 +00003135
Miklos Szeredidecf3402012-05-21 17:30:08 +02003136 BUG_ON(nd->flags & LOOKUP_RCU);
David Howells63afdfc2015-05-06 15:59:00 +01003137 inode = d_backing_inode(path.dentry);
Al Viro254cf582015-05-05 09:40:46 -04003138 seq = 0; /* out of RCU mode, so the value doesn't matter */
Al Virodeb106c2015-05-08 18:05:21 -04003139 if (unlikely(d_is_negative(path.dentry))) {
Al Viro896475d2015-04-22 18:02:17 -04003140 path_to_nameidata(&path, nd);
Al Virodeb106c2015-05-08 18:05:21 -04003141 return -ENOENT;
Miklos Szeredi54c33e72012-05-21 17:30:10 +02003142 }
Al Viro766c4cb2015-05-07 19:24:57 -04003143finish_lookup:
Al Virod63ff282015-05-04 18:13:23 -04003144 if (nd->depth)
3145 put_link(nd);
Al Viro181548c2015-05-07 19:54:34 -04003146 error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW,
3147 inode, seq);
Al Virodeb106c2015-05-08 18:05:21 -04003148 if (unlikely(error))
Al Virod63ff282015-05-04 18:13:23 -04003149 return error;
Al Virofb1cc552009-12-24 01:58:28 -05003150
Al Viro896475d2015-04-22 18:02:17 -04003151 if (unlikely(d_is_symlink(path.dentry)) && !(open_flag & O_PATH)) {
3152 path_to_nameidata(&path, nd);
Al Virodeb106c2015-05-08 18:05:21 -04003153 return -ELOOP;
Al Viroa5cfe2d2015-04-22 12:10:45 -04003154 }
3155
Al Viro896475d2015-04-22 18:02:17 -04003156 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) {
3157 path_to_nameidata(&path, nd);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003158 } else {
3159 save_parent.dentry = nd->path.dentry;
Al Viro896475d2015-04-22 18:02:17 -04003160 save_parent.mnt = mntget(path.mnt);
3161 nd->path.dentry = path.dentry;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003162
3163 }
Miklos Szeredidecf3402012-05-21 17:30:08 +02003164 nd->inode = inode;
Al Viro254cf582015-05-05 09:40:46 -04003165 nd->seq = seq;
Al Viroa3fbbde2011-11-07 21:21:26 +00003166 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
Al Virobc77daa2013-06-06 09:12:33 -04003167finish_open:
Al Viroa3fbbde2011-11-07 21:21:26 +00003168 error = complete_walk(nd);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003169 if (error) {
3170 path_put(&save_parent);
Al Viro2675a4e2012-06-22 12:41:10 +04003171 return error;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003172 }
Al Viro76ae2a52015-05-12 18:44:32 -04003173 audit_inode(nd->name, nd->path.dentry, 0);
Al Virofb1cc552009-12-24 01:58:28 -05003174 error = -EISDIR;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003175 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04003176 goto out;
Miklos Szerediaf2f5542012-05-21 17:30:11 +02003177 error = -ENOTDIR;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003178 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04003179 goto out;
David Howells4bbcbd32015-03-17 22:16:40 +00003180 if (!d_is_reg(nd->path.dentry))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003181 will_truncate = false;
Al Viro6c0d46c2011-03-09 00:59:59 -05003182
Al Viro0f9d1a12011-03-09 00:13:14 -05003183 if (will_truncate) {
3184 error = mnt_want_write(nd->path.mnt);
3185 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003186 goto out;
Al Viro64894cf2012-07-31 00:53:35 +04003187 got_write = true;
Al Viro0f9d1a12011-03-09 00:13:14 -05003188 }
Miklos Szeredie83db162012-06-05 15:10:29 +02003189finish_open_created:
Al Virobcda7652011-03-13 16:42:14 -04003190 error = may_open(&nd->path, acc_mode, open_flag);
Al Viroca344a892011-03-09 00:36:45 -05003191 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003192 goto out;
Miklos Szeredi4aa7c632014-10-24 00:14:35 +02003193
3194 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3195 error = vfs_open(&nd->path, file, current_cred());
3196 if (!error) {
3197 *opened |= FILE_OPENED;
3198 } else {
Al Viro30d90492012-06-22 12:40:19 +04003199 if (error == -EOPENSTALE)
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003200 goto stale_open;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003201 goto out;
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003202 }
Miklos Szeredia8277b92012-06-05 15:10:32 +02003203opened:
Al Viro2675a4e2012-06-22 12:41:10 +04003204 error = open_check_o_direct(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003205 if (error)
3206 goto exit_fput;
Dmitry Kasatkin3034a142014-06-27 18:15:44 +03003207 error = ima_file_check(file, op->acc_mode, *opened);
Miklos Szerediaa4caad2012-06-05 15:10:28 +02003208 if (error)
3209 goto exit_fput;
3210
3211 if (will_truncate) {
Al Viro2675a4e2012-06-22 12:41:10 +04003212 error = handle_truncate(file);
Miklos Szerediaa4caad2012-06-05 15:10:28 +02003213 if (error)
3214 goto exit_fput;
Al Viro0f9d1a12011-03-09 00:13:14 -05003215 }
Al Viroca344a892011-03-09 00:36:45 -05003216out:
Al Viro64894cf2012-07-31 00:53:35 +04003217 if (got_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05003218 mnt_drop_write(nd->path.mnt);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003219 path_put(&save_parent);
Al Viro2675a4e2012-06-22 12:41:10 +04003220 return error;
Al Virofb1cc552009-12-24 01:58:28 -05003221
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003222exit_fput:
Al Viro2675a4e2012-06-22 12:41:10 +04003223 fput(file);
3224 goto out;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003225
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003226stale_open:
3227 /* If no saved parent or already retried then can't retry */
3228 if (!save_parent.dentry || retried)
3229 goto out;
3230
3231 BUG_ON(save_parent.dentry != dir);
3232 path_put(&nd->path);
3233 nd->path = save_parent;
3234 nd->inode = dir->d_inode;
3235 save_parent.mnt = NULL;
3236 save_parent.dentry = NULL;
Al Viro64894cf2012-07-31 00:53:35 +04003237 if (got_write) {
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003238 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04003239 got_write = false;
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003240 }
3241 retried = true;
3242 goto retry_lookup;
Al Virofb1cc552009-12-24 01:58:28 -05003243}
3244
Al Viroc8a53ee2015-05-12 18:43:07 -04003245static int do_tmpfile(struct nameidata *nd, unsigned flags,
Al Viro60545d02013-06-07 01:20:27 -04003246 const struct open_flags *op,
3247 struct file *file, int *opened)
3248{
3249 static const struct qstr name = QSTR_INIT("/", 1);
Al Viro625b6d12015-05-12 16:36:12 -04003250 struct dentry *child;
Al Viro60545d02013-06-07 01:20:27 -04003251 struct inode *dir;
Al Viro625b6d12015-05-12 16:36:12 -04003252 struct path path;
Al Viroc8a53ee2015-05-12 18:43:07 -04003253 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
Al Viro60545d02013-06-07 01:20:27 -04003254 if (unlikely(error))
3255 return error;
Al Viro625b6d12015-05-12 16:36:12 -04003256 error = mnt_want_write(path.mnt);
Al Viro60545d02013-06-07 01:20:27 -04003257 if (unlikely(error))
3258 goto out;
Al Viro625b6d12015-05-12 16:36:12 -04003259 dir = path.dentry->d_inode;
Al Viro60545d02013-06-07 01:20:27 -04003260 /* we want directory to be writable */
Al Viro625b6d12015-05-12 16:36:12 -04003261 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Al Viro60545d02013-06-07 01:20:27 -04003262 if (error)
3263 goto out2;
Al Viro60545d02013-06-07 01:20:27 -04003264 if (!dir->i_op->tmpfile) {
3265 error = -EOPNOTSUPP;
3266 goto out2;
3267 }
Al Viro625b6d12015-05-12 16:36:12 -04003268 child = d_alloc(path.dentry, &name);
Al Viro60545d02013-06-07 01:20:27 -04003269 if (unlikely(!child)) {
3270 error = -ENOMEM;
3271 goto out2;
3272 }
Al Viro625b6d12015-05-12 16:36:12 -04003273 dput(path.dentry);
3274 path.dentry = child;
3275 error = dir->i_op->tmpfile(dir, child, op->mode);
Al Viro60545d02013-06-07 01:20:27 -04003276 if (error)
3277 goto out2;
Al Viroc8a53ee2015-05-12 18:43:07 -04003278 audit_inode(nd->name, child, 0);
Eric Rannaud69a91c22014-10-30 01:51:01 -07003279 /* Don't check for other permissions, the inode was just created */
Al Viro625b6d12015-05-12 16:36:12 -04003280 error = may_open(&path, MAY_OPEN, op->open_flag);
Al Viro60545d02013-06-07 01:20:27 -04003281 if (error)
3282 goto out2;
Al Viro625b6d12015-05-12 16:36:12 -04003283 file->f_path.mnt = path.mnt;
3284 error = finish_open(file, child, NULL, opened);
Al Viro60545d02013-06-07 01:20:27 -04003285 if (error)
3286 goto out2;
3287 error = open_check_o_direct(file);
Al Virof4e0c302013-06-11 08:34:36 +04003288 if (error) {
Al Viro60545d02013-06-07 01:20:27 -04003289 fput(file);
Al Virof4e0c302013-06-11 08:34:36 +04003290 } else if (!(op->open_flag & O_EXCL)) {
3291 struct inode *inode = file_inode(file);
3292 spin_lock(&inode->i_lock);
3293 inode->i_state |= I_LINKABLE;
3294 spin_unlock(&inode->i_lock);
3295 }
Al Viro60545d02013-06-07 01:20:27 -04003296out2:
Al Viro625b6d12015-05-12 16:36:12 -04003297 mnt_drop_write(path.mnt);
Al Viro60545d02013-06-07 01:20:27 -04003298out:
Al Viro625b6d12015-05-12 16:36:12 -04003299 path_put(&path);
Al Viro60545d02013-06-07 01:20:27 -04003300 return error;
3301}
3302
Al Viroc8a53ee2015-05-12 18:43:07 -04003303static struct file *path_openat(struct nameidata *nd,
3304 const struct open_flags *op, unsigned flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003305{
Al Viro368ee9b2015-05-08 17:19:59 -04003306 const char *s;
Al Viro30d90492012-06-22 12:40:19 +04003307 struct file *file;
Al Viro47237682012-06-10 05:01:45 -04003308 int opened = 0;
Al Viro13aab422011-02-23 17:54:08 -05003309 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11003310
Al Viro30d90492012-06-22 12:40:19 +04003311 file = get_empty_filp();
Al Viro1afc99b2013-02-14 20:41:04 -05003312 if (IS_ERR(file))
3313 return file;
Nick Piggin31e6b012011-01-07 17:49:52 +11003314
Al Viro30d90492012-06-22 12:40:19 +04003315 file->f_flags = op->open_flag;
Nick Piggin31e6b012011-01-07 17:49:52 +11003316
Al Virobb458c62013-07-13 13:26:37 +04003317 if (unlikely(file->f_flags & __O_TMPFILE)) {
Al Viroc8a53ee2015-05-12 18:43:07 -04003318 error = do_tmpfile(nd, flags, op, file, &opened);
Al Virof15133d2015-05-08 22:53:15 -04003319 goto out2;
Al Viro60545d02013-06-07 01:20:27 -04003320 }
3321
Al Viroc8a53ee2015-05-12 18:43:07 -04003322 s = path_init(nd, flags);
Al Viro368ee9b2015-05-08 17:19:59 -04003323 if (IS_ERR(s)) {
3324 put_filp(file);
3325 return ERR_CAST(s);
3326 }
Al Viro3bdba282015-05-08 17:37:07 -04003327 while (!(error = link_path_walk(s, nd)) &&
Al Viro76ae2a52015-05-12 18:44:32 -04003328 (error = do_last(nd, file, op, &opened)) > 0) {
Al Viro73d049a2011-03-11 12:08:24 -05003329 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
Al Viro3bdba282015-05-08 17:37:07 -04003330 s = trailing_symlink(nd);
3331 if (IS_ERR(s)) {
3332 error = PTR_ERR(s);
Al Viro2675a4e2012-06-22 12:41:10 +04003333 break;
Al Viro3bdba282015-05-08 17:37:07 -04003334 }
Al Viro806b6812009-12-26 07:16:40 -05003335 }
Al Virodeb106c2015-05-08 18:05:21 -04003336 terminate_walk(nd);
Al Virof15133d2015-05-08 22:53:15 -04003337out2:
Al Viro2675a4e2012-06-22 12:41:10 +04003338 if (!(opened & FILE_OPENED)) {
3339 BUG_ON(!error);
Al Viro30d90492012-06-22 12:40:19 +04003340 put_filp(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003341 }
Al Viro2675a4e2012-06-22 12:41:10 +04003342 if (unlikely(error)) {
3343 if (error == -EOPENSTALE) {
3344 if (flags & LOOKUP_RCU)
3345 error = -ECHILD;
3346 else
3347 error = -ESTALE;
3348 }
3349 file = ERR_PTR(error);
3350 }
3351 return file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003352}
3353
Jeff Layton669abf42012-10-10 16:43:10 -04003354struct file *do_filp_open(int dfd, struct filename *pathname,
Al Virof9652e12013-06-11 08:23:01 +04003355 const struct open_flags *op)
Al Viro13aab422011-02-23 17:54:08 -05003356{
Al Viro9883d182015-05-13 07:28:08 -04003357 struct nameidata nd;
Al Virof9652e12013-06-11 08:23:01 +04003358 int flags = op->lookup_flags;
Al Viro13aab422011-02-23 17:54:08 -05003359 struct file *filp;
3360
Al Viro9883d182015-05-13 07:28:08 -04003361 set_nameidata(&nd, dfd, pathname);
Al Viroc8a53ee2015-05-12 18:43:07 -04003362 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
Al Viro13aab422011-02-23 17:54:08 -05003363 if (unlikely(filp == ERR_PTR(-ECHILD)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003364 filp = path_openat(&nd, op, flags);
Al Viro13aab422011-02-23 17:54:08 -05003365 if (unlikely(filp == ERR_PTR(-ESTALE)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003366 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
Al Viro9883d182015-05-13 07:28:08 -04003367 restore_nameidata();
Al Viro13aab422011-02-23 17:54:08 -05003368 return filp;
3369}
3370
Al Viro73d049a2011-03-11 12:08:24 -05003371struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
Al Virof9652e12013-06-11 08:23:01 +04003372 const char *name, const struct open_flags *op)
Al Viro73d049a2011-03-11 12:08:24 -05003373{
Al Viro9883d182015-05-13 07:28:08 -04003374 struct nameidata nd;
Al Viro73d049a2011-03-11 12:08:24 -05003375 struct file *file;
Paul Moore51689102015-01-22 00:00:03 -05003376 struct filename *filename;
Al Virof9652e12013-06-11 08:23:01 +04003377 int flags = op->lookup_flags | LOOKUP_ROOT;
Al Viro73d049a2011-03-11 12:08:24 -05003378
3379 nd.root.mnt = mnt;
3380 nd.root.dentry = dentry;
3381
David Howellsb18825a2013-09-12 19:22:53 +01003382 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
Al Viro73d049a2011-03-11 12:08:24 -05003383 return ERR_PTR(-ELOOP);
3384
Paul Moore51689102015-01-22 00:00:03 -05003385 filename = getname_kernel(name);
3386 if (unlikely(IS_ERR(filename)))
3387 return ERR_CAST(filename);
3388
Al Viro9883d182015-05-13 07:28:08 -04003389 set_nameidata(&nd, -1, filename);
Al Viroc8a53ee2015-05-12 18:43:07 -04003390 file = path_openat(&nd, op, flags | LOOKUP_RCU);
Al Viro73d049a2011-03-11 12:08:24 -05003391 if (unlikely(file == ERR_PTR(-ECHILD)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003392 file = path_openat(&nd, op, flags);
Al Viro73d049a2011-03-11 12:08:24 -05003393 if (unlikely(file == ERR_PTR(-ESTALE)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003394 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
Al Viro9883d182015-05-13 07:28:08 -04003395 restore_nameidata();
Paul Moore51689102015-01-22 00:00:03 -05003396 putname(filename);
Al Viro73d049a2011-03-11 12:08:24 -05003397 return file;
3398}
3399
Al Virofa14a0b2015-01-22 02:16:49 -05003400static struct dentry *filename_create(int dfd, struct filename *name,
Jeff Layton1ac12b42012-12-11 12:10:06 -05003401 struct path *path, unsigned int lookup_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003402{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003403 struct dentry *dentry = ERR_PTR(-EEXIST);
Al Viro391172c2015-05-09 11:19:16 -04003404 struct qstr last;
3405 int type;
Jan Karac30dabf2012-06-12 16:20:30 +02003406 int err2;
Jeff Layton1ac12b42012-12-11 12:10:06 -05003407 int error;
3408 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3409
3410 /*
3411 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3412 * other flags passed in are ignored!
3413 */
3414 lookup_flags &= LOOKUP_REVAL;
3415
Al Viro5c31b6c2015-05-12 17:32:54 -04003416 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3417 if (IS_ERR(name))
3418 return ERR_CAST(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003419
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003420 /*
3421 * Yucky last component or no last component at all?
3422 * (foo/., foo/.., /////)
3423 */
Al Viro5c31b6c2015-05-12 17:32:54 -04003424 if (unlikely(type != LAST_NORM))
Al Viroed75e952011-06-27 16:53:43 -04003425 goto out;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003426
Jan Karac30dabf2012-06-12 16:20:30 +02003427 /* don't fail immediately if it's r/o, at least try to report other errors */
Al Viro391172c2015-05-09 11:19:16 -04003428 err2 = mnt_want_write(path->mnt);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003429 /*
3430 * Do the final lookup.
3431 */
Al Viro391172c2015-05-09 11:19:16 -04003432 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3433 mutex_lock_nested(&path->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3434 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003435 if (IS_ERR(dentry))
Al Viroa8104a92012-07-20 02:25:00 +04003436 goto unlock;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003437
Al Viroa8104a92012-07-20 02:25:00 +04003438 error = -EEXIST;
David Howellsb18825a2013-09-12 19:22:53 +01003439 if (d_is_positive(dentry))
Al Viroa8104a92012-07-20 02:25:00 +04003440 goto fail;
David Howellsb18825a2013-09-12 19:22:53 +01003441
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003442 /*
3443 * Special case - lookup gave negative, but... we had foo/bar/
3444 * From the vfs_mknod() POV we just have a negative dentry -
3445 * all is fine. Let's be bastards - you had / on the end, you've
3446 * been asking for (non-existent) directory. -ENOENT for you.
3447 */
Al Viro391172c2015-05-09 11:19:16 -04003448 if (unlikely(!is_dir && last.name[last.len])) {
Al Viroa8104a92012-07-20 02:25:00 +04003449 error = -ENOENT;
Al Viroed75e952011-06-27 16:53:43 -04003450 goto fail;
Al Viroe9baf6e2008-05-15 04:49:12 -04003451 }
Jan Karac30dabf2012-06-12 16:20:30 +02003452 if (unlikely(err2)) {
3453 error = err2;
Al Viroa8104a92012-07-20 02:25:00 +04003454 goto fail;
Jan Karac30dabf2012-06-12 16:20:30 +02003455 }
Al Viro181c37b2015-05-12 17:21:25 -04003456 putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003457 return dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003458fail:
Al Viroa8104a92012-07-20 02:25:00 +04003459 dput(dentry);
3460 dentry = ERR_PTR(error);
3461unlock:
Al Viro391172c2015-05-09 11:19:16 -04003462 mutex_unlock(&path->dentry->d_inode->i_mutex);
Jan Karac30dabf2012-06-12 16:20:30 +02003463 if (!err2)
Al Viro391172c2015-05-09 11:19:16 -04003464 mnt_drop_write(path->mnt);
Al Viroed75e952011-06-27 16:53:43 -04003465out:
Al Viro391172c2015-05-09 11:19:16 -04003466 path_put(path);
Al Viro181c37b2015-05-12 17:21:25 -04003467 putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003468 return dentry;
3469}
Al Virofa14a0b2015-01-22 02:16:49 -05003470
3471struct dentry *kern_path_create(int dfd, const char *pathname,
3472 struct path *path, unsigned int lookup_flags)
3473{
Al Viro181c37b2015-05-12 17:21:25 -04003474 return filename_create(dfd, getname_kernel(pathname),
3475 path, lookup_flags);
Al Virofa14a0b2015-01-22 02:16:49 -05003476}
Al Virodae6ad82011-06-26 11:50:15 -04003477EXPORT_SYMBOL(kern_path_create);
3478
Al Viro921a1652012-07-20 01:15:31 +04003479void done_path_create(struct path *path, struct dentry *dentry)
3480{
3481 dput(dentry);
3482 mutex_unlock(&path->dentry->d_inode->i_mutex);
Al Viroa8104a92012-07-20 02:25:00 +04003483 mnt_drop_write(path->mnt);
Al Viro921a1652012-07-20 01:15:31 +04003484 path_put(path);
3485}
3486EXPORT_SYMBOL(done_path_create);
3487
Al Viro520ae682015-05-13 07:00:28 -04003488inline struct dentry *user_path_create(int dfd, const char __user *pathname,
Jeff Layton1ac12b42012-12-11 12:10:06 -05003489 struct path *path, unsigned int lookup_flags)
Al Virodae6ad82011-06-26 11:50:15 -04003490{
Al Viro181c37b2015-05-12 17:21:25 -04003491 return filename_create(dfd, getname(pathname), path, lookup_flags);
Al Virodae6ad82011-06-26 11:50:15 -04003492}
3493EXPORT_SYMBOL(user_path_create);
3494
Al Viro1a67aaf2011-07-26 01:52:52 -04003495int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003496{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003497 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003498
3499 if (error)
3500 return error;
3501
Eric W. Biederman975d6b32011-11-13 12:16:43 -08003502 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003503 return -EPERM;
3504
Al Viroacfa4382008-12-04 10:06:33 -05003505 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003506 return -EPERM;
3507
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07003508 error = devcgroup_inode_mknod(mode, dev);
3509 if (error)
3510 return error;
3511
Linus Torvalds1da177e2005-04-16 15:20:36 -07003512 error = security_inode_mknod(dir, dentry, mode, dev);
3513 if (error)
3514 return error;
3515
Linus Torvalds1da177e2005-04-16 15:20:36 -07003516 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003517 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003518 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003519 return error;
3520}
Al Viro4d359502014-03-14 12:20:17 -04003521EXPORT_SYMBOL(vfs_mknod);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003522
Al Virof69aac02011-07-26 04:31:40 -04003523static int may_mknod(umode_t mode)
Dave Hansen463c3192008-02-15 14:37:57 -08003524{
3525 switch (mode & S_IFMT) {
3526 case S_IFREG:
3527 case S_IFCHR:
3528 case S_IFBLK:
3529 case S_IFIFO:
3530 case S_IFSOCK:
3531 case 0: /* zero mode translates to S_IFREG */
3532 return 0;
3533 case S_IFDIR:
3534 return -EPERM;
3535 default:
3536 return -EINVAL;
3537 }
3538}
3539
Al Viro8208a222011-07-25 17:32:17 -04003540SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003541 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003542{
Al Viro2ad94ae2008-07-21 09:32:51 -04003543 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003544 struct path path;
3545 int error;
Jeff Layton972567f2012-12-20 16:00:10 -05003546 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003547
Al Viro8e4bfca2012-07-20 01:17:26 +04003548 error = may_mknod(mode);
3549 if (error)
3550 return error;
Jeff Layton972567f2012-12-20 16:00:10 -05003551retry:
3552 dentry = user_path_create(dfd, filename, &path, lookup_flags);
Al Virodae6ad82011-06-26 11:50:15 -04003553 if (IS_ERR(dentry))
3554 return PTR_ERR(dentry);
Al Viro2ad94ae2008-07-21 09:32:51 -04003555
Al Virodae6ad82011-06-26 11:50:15 -04003556 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003557 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003558 error = security_path_mknod(&path, dentry, mode, dev);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003559 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04003560 goto out;
Dave Hansen463c3192008-02-15 14:37:57 -08003561 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003562 case 0: case S_IFREG:
Al Viro312b63f2012-06-10 18:09:36 -04003563 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003564 break;
3565 case S_IFCHR: case S_IFBLK:
Al Virodae6ad82011-06-26 11:50:15 -04003566 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003567 new_decode_dev(dev));
3568 break;
3569 case S_IFIFO: case S_IFSOCK:
Al Virodae6ad82011-06-26 11:50:15 -04003570 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003571 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003572 }
Al Viroa8104a92012-07-20 02:25:00 +04003573out:
Al Viro921a1652012-07-20 01:15:31 +04003574 done_path_create(&path, dentry);
Jeff Layton972567f2012-12-20 16:00:10 -05003575 if (retry_estale(error, lookup_flags)) {
3576 lookup_flags |= LOOKUP_REVAL;
3577 goto retry;
3578 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003579 return error;
3580}
3581
Al Viro8208a222011-07-25 17:32:17 -04003582SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003583{
3584 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3585}
3586
Al Viro18bb1db2011-07-26 01:41:39 -04003587int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003588{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003589 int error = may_create(dir, dentry);
Al Viro8de52772012-02-06 12:45:27 -05003590 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003591
3592 if (error)
3593 return error;
3594
Al Viroacfa4382008-12-04 10:06:33 -05003595 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003596 return -EPERM;
3597
3598 mode &= (S_IRWXUGO|S_ISVTX);
3599 error = security_inode_mkdir(dir, dentry, mode);
3600 if (error)
3601 return error;
3602
Al Viro8de52772012-02-06 12:45:27 -05003603 if (max_links && dir->i_nlink >= max_links)
3604 return -EMLINK;
3605
Linus Torvalds1da177e2005-04-16 15:20:36 -07003606 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003607 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003608 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003609 return error;
3610}
Al Viro4d359502014-03-14 12:20:17 -04003611EXPORT_SYMBOL(vfs_mkdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003612
Al Viroa218d0f2011-11-21 14:59:34 -05003613SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003614{
Dave Hansen6902d922006-09-30 23:29:01 -07003615 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003616 struct path path;
3617 int error;
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003618 unsigned int lookup_flags = LOOKUP_DIRECTORY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003619
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003620retry:
3621 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07003622 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003623 return PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003624
Al Virodae6ad82011-06-26 11:50:15 -04003625 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003626 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003627 error = security_path_mkdir(&path, dentry, mode);
Al Viroa8104a92012-07-20 02:25:00 +04003628 if (!error)
3629 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
Al Viro921a1652012-07-20 01:15:31 +04003630 done_path_create(&path, dentry);
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003631 if (retry_estale(error, lookup_flags)) {
3632 lookup_flags |= LOOKUP_REVAL;
3633 goto retry;
3634 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003635 return error;
3636}
3637
Al Viroa218d0f2011-11-21 14:59:34 -05003638SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003639{
3640 return sys_mkdirat(AT_FDCWD, pathname, mode);
3641}
3642
Linus Torvalds1da177e2005-04-16 15:20:36 -07003643/*
Sage Weila71905f2011-05-24 13:06:08 -07003644 * The dentry_unhash() helper will try to drop the dentry early: we
J. Bruce Fieldsc0d02592012-02-15 11:48:40 -05003645 * should have a usage count of 1 if we're the only user of this
Sage Weila71905f2011-05-24 13:06:08 -07003646 * dentry, and if that is true (possibly after pruning the dcache),
3647 * then we drop the dentry now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003648 *
3649 * A low-level filesystem can, if it choses, legally
3650 * do a
3651 *
3652 * if (!d_unhashed(dentry))
3653 * return -EBUSY;
3654 *
3655 * if it cannot handle the case of removing a directory
3656 * that is still in use by something else..
3657 */
3658void dentry_unhash(struct dentry *dentry)
3659{
Vasily Averindc168422006-12-06 20:37:07 -08003660 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003661 spin_lock(&dentry->d_lock);
Waiman Long98474232013-08-28 18:24:59 -07003662 if (dentry->d_lockref.count == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003663 __d_drop(dentry);
3664 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003665}
Al Viro4d359502014-03-14 12:20:17 -04003666EXPORT_SYMBOL(dentry_unhash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003667
3668int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3669{
3670 int error = may_delete(dir, dentry, 1);
3671
3672 if (error)
3673 return error;
3674
Al Viroacfa4382008-12-04 10:06:33 -05003675 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003676 return -EPERM;
3677
Al Viro1d2ef592011-09-14 18:55:41 +01003678 dget(dentry);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003679 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003680
Sage Weil912dbc12011-05-24 13:06:11 -07003681 error = -EBUSY;
Eric W. Biederman7af13642013-10-04 19:15:13 -07003682 if (is_local_mountpoint(dentry))
Sage Weil912dbc12011-05-24 13:06:11 -07003683 goto out;
3684
3685 error = security_inode_rmdir(dir, dentry);
3686 if (error)
3687 goto out;
3688
Sage Weil3cebde22011-05-29 21:20:59 -07003689 shrink_dcache_parent(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003690 error = dir->i_op->rmdir(dir, dentry);
3691 if (error)
3692 goto out;
3693
3694 dentry->d_inode->i_flags |= S_DEAD;
3695 dont_mount(dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003696 detach_mounts(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003697
3698out:
3699 mutex_unlock(&dentry->d_inode->i_mutex);
Al Viro1d2ef592011-09-14 18:55:41 +01003700 dput(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003701 if (!error)
3702 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003703 return error;
3704}
Al Viro4d359502014-03-14 12:20:17 -04003705EXPORT_SYMBOL(vfs_rmdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003706
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003707static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003708{
3709 int error = 0;
Jeff Layton91a27b22012-10-10 15:25:28 -04003710 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003711 struct dentry *dentry;
Al Virof5beed72015-04-30 16:09:11 -04003712 struct path path;
3713 struct qstr last;
3714 int type;
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003715 unsigned int lookup_flags = 0;
3716retry:
Al Virof5beed72015-04-30 16:09:11 -04003717 name = user_path_parent(dfd, pathname,
3718 &path, &last, &type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04003719 if (IS_ERR(name))
3720 return PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003721
Al Virof5beed72015-04-30 16:09:11 -04003722 switch (type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003723 case LAST_DOTDOT:
3724 error = -ENOTEMPTY;
3725 goto exit1;
3726 case LAST_DOT:
3727 error = -EINVAL;
3728 goto exit1;
3729 case LAST_ROOT:
3730 error = -EBUSY;
3731 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003732 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003733
Al Virof5beed72015-04-30 16:09:11 -04003734 error = mnt_want_write(path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02003735 if (error)
3736 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003737
Al Virof5beed72015-04-30 16:09:11 -04003738 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3739 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003740 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003741 if (IS_ERR(dentry))
3742 goto exit2;
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003743 if (!dentry->d_inode) {
3744 error = -ENOENT;
3745 goto exit3;
3746 }
Al Virof5beed72015-04-30 16:09:11 -04003747 error = security_path_rmdir(&path, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003748 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003749 goto exit3;
Al Virof5beed72015-04-30 16:09:11 -04003750 error = vfs_rmdir(path.dentry->d_inode, dentry);
Dave Hansen06227532008-02-15 14:37:34 -08003751exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07003752 dput(dentry);
3753exit2:
Al Virof5beed72015-04-30 16:09:11 -04003754 mutex_unlock(&path.dentry->d_inode->i_mutex);
3755 mnt_drop_write(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003756exit1:
Al Virof5beed72015-04-30 16:09:11 -04003757 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003758 putname(name);
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003759 if (retry_estale(error, lookup_flags)) {
3760 lookup_flags |= LOOKUP_REVAL;
3761 goto retry;
3762 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003763 return error;
3764}
3765
Heiko Carstens3cdad422009-01-14 14:14:22 +01003766SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003767{
3768 return do_rmdir(AT_FDCWD, pathname);
3769}
3770
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003771/**
3772 * vfs_unlink - unlink a filesystem object
3773 * @dir: parent directory
3774 * @dentry: victim
3775 * @delegated_inode: returns victim inode, if the inode is delegated.
3776 *
3777 * The caller must hold dir->i_mutex.
3778 *
3779 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3780 * return a reference to the inode in delegated_inode. The caller
3781 * should then break the delegation on that inode and retry. Because
3782 * breaking a delegation may take a long time, the caller should drop
3783 * dir->i_mutex before doing so.
3784 *
3785 * Alternatively, a caller may pass NULL for delegated_inode. This may
3786 * be appropriate for callers that expect the underlying filesystem not
3787 * to be NFS exported.
3788 */
3789int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003790{
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003791 struct inode *target = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003792 int error = may_delete(dir, dentry, 0);
3793
3794 if (error)
3795 return error;
3796
Al Viroacfa4382008-12-04 10:06:33 -05003797 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003798 return -EPERM;
3799
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003800 mutex_lock(&target->i_mutex);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003801 if (is_local_mountpoint(dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003802 error = -EBUSY;
3803 else {
3804 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05003805 if (!error) {
J. Bruce Fields5a146962012-08-28 07:50:40 -07003806 error = try_break_deleg(target, delegated_inode);
3807 if (error)
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003808 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003809 error = dir->i_op->unlink(dir, dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003810 if (!error) {
Al Virod83c49f2010-04-30 17:17:09 -04003811 dont_mount(dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003812 detach_mounts(dentry);
3813 }
Al Virobec10522010-03-03 14:12:08 -05003814 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003815 }
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003816out:
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003817 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003818
3819 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3820 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003821 fsnotify_link_count(target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003822 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003823 }
Robert Love0eeca282005-07-12 17:06:03 -04003824
Linus Torvalds1da177e2005-04-16 15:20:36 -07003825 return error;
3826}
Al Viro4d359502014-03-14 12:20:17 -04003827EXPORT_SYMBOL(vfs_unlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003828
3829/*
3830 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003831 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07003832 * writeout happening, and we don't want to prevent access to the directory
3833 * while waiting on the I/O.
3834 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003835static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003836{
Al Viro2ad94ae2008-07-21 09:32:51 -04003837 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04003838 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003839 struct dentry *dentry;
Al Virof5beed72015-04-30 16:09:11 -04003840 struct path path;
3841 struct qstr last;
3842 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003843 struct inode *inode = NULL;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003844 struct inode *delegated_inode = NULL;
Jeff Layton5d18f812012-12-20 16:38:04 -05003845 unsigned int lookup_flags = 0;
3846retry:
Al Virof5beed72015-04-30 16:09:11 -04003847 name = user_path_parent(dfd, pathname,
3848 &path, &last, &type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04003849 if (IS_ERR(name))
3850 return PTR_ERR(name);
Al Viro2ad94ae2008-07-21 09:32:51 -04003851
Linus Torvalds1da177e2005-04-16 15:20:36 -07003852 error = -EISDIR;
Al Virof5beed72015-04-30 16:09:11 -04003853 if (type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003854 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003855
Al Virof5beed72015-04-30 16:09:11 -04003856 error = mnt_want_write(path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02003857 if (error)
3858 goto exit1;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003859retry_deleg:
Al Virof5beed72015-04-30 16:09:11 -04003860 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3861 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003862 error = PTR_ERR(dentry);
3863 if (!IS_ERR(dentry)) {
3864 /* Why not before? Because we want correct error value */
Al Virof5beed72015-04-30 16:09:11 -04003865 if (last.name[last.len])
Török Edwin50338b82011-06-16 00:06:14 +03003866 goto slashes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003867 inode = dentry->d_inode;
David Howellsb18825a2013-09-12 19:22:53 +01003868 if (d_is_negative(dentry))
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003869 goto slashes;
3870 ihold(inode);
Al Virof5beed72015-04-30 16:09:11 -04003871 error = security_path_unlink(&path, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003872 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003873 goto exit2;
Al Virof5beed72015-04-30 16:09:11 -04003874 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
Jan Karac30dabf2012-06-12 16:20:30 +02003875exit2:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003876 dput(dentry);
3877 }
Al Virof5beed72015-04-30 16:09:11 -04003878 mutex_unlock(&path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003879 if (inode)
3880 iput(inode); /* truncate the inode here */
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003881 inode = NULL;
3882 if (delegated_inode) {
J. Bruce Fields5a146962012-08-28 07:50:40 -07003883 error = break_deleg_wait(&delegated_inode);
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003884 if (!error)
3885 goto retry_deleg;
3886 }
Al Virof5beed72015-04-30 16:09:11 -04003887 mnt_drop_write(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003888exit1:
Al Virof5beed72015-04-30 16:09:11 -04003889 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003890 putname(name);
Jeff Layton5d18f812012-12-20 16:38:04 -05003891 if (retry_estale(error, lookup_flags)) {
3892 lookup_flags |= LOOKUP_REVAL;
3893 inode = NULL;
3894 goto retry;
3895 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003896 return error;
3897
3898slashes:
David Howellsb18825a2013-09-12 19:22:53 +01003899 if (d_is_negative(dentry))
3900 error = -ENOENT;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003901 else if (d_is_dir(dentry))
David Howellsb18825a2013-09-12 19:22:53 +01003902 error = -EISDIR;
3903 else
3904 error = -ENOTDIR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003905 goto exit2;
3906}
3907
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003908SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003909{
3910 if ((flag & ~AT_REMOVEDIR) != 0)
3911 return -EINVAL;
3912
3913 if (flag & AT_REMOVEDIR)
3914 return do_rmdir(dfd, pathname);
3915
3916 return do_unlinkat(dfd, pathname);
3917}
3918
Heiko Carstens3480b252009-01-14 14:14:16 +01003919SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003920{
3921 return do_unlinkat(AT_FDCWD, pathname);
3922}
3923
Miklos Szeredidb2e7472008-06-24 16:50:16 +02003924int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003925{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003926 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003927
3928 if (error)
3929 return error;
3930
Al Viroacfa4382008-12-04 10:06:33 -05003931 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003932 return -EPERM;
3933
3934 error = security_inode_symlink(dir, dentry, oldname);
3935 if (error)
3936 return error;
3937
Linus Torvalds1da177e2005-04-16 15:20:36 -07003938 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003939 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003940 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003941 return error;
3942}
Al Viro4d359502014-03-14 12:20:17 -04003943EXPORT_SYMBOL(vfs_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003944
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003945SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3946 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003947{
Al Viro2ad94ae2008-07-21 09:32:51 -04003948 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04003949 struct filename *from;
Dave Hansen6902d922006-09-30 23:29:01 -07003950 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003951 struct path path;
Jeff Laytonf46d3562012-12-11 12:10:08 -05003952 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003953
3954 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04003955 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003956 return PTR_ERR(from);
Jeff Laytonf46d3562012-12-11 12:10:08 -05003957retry:
3958 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07003959 error = PTR_ERR(dentry);
3960 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003961 goto out_putname;
Dave Hansen6902d922006-09-30 23:29:01 -07003962
Jeff Layton91a27b22012-10-10 15:25:28 -04003963 error = security_path_symlink(&path, dentry, from->name);
Al Viroa8104a92012-07-20 02:25:00 +04003964 if (!error)
Jeff Layton91a27b22012-10-10 15:25:28 -04003965 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
Al Viro921a1652012-07-20 01:15:31 +04003966 done_path_create(&path, dentry);
Jeff Laytonf46d3562012-12-11 12:10:08 -05003967 if (retry_estale(error, lookup_flags)) {
3968 lookup_flags |= LOOKUP_REVAL;
3969 goto retry;
3970 }
Dave Hansen6902d922006-09-30 23:29:01 -07003971out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003972 putname(from);
3973 return error;
3974}
3975
Heiko Carstens3480b252009-01-14 14:14:16 +01003976SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003977{
3978 return sys_symlinkat(oldname, AT_FDCWD, newname);
3979}
3980
J. Bruce Fields146a8592011-09-20 17:14:31 -04003981/**
3982 * vfs_link - create a new link
3983 * @old_dentry: object to be linked
3984 * @dir: new parent
3985 * @new_dentry: where to create the new link
3986 * @delegated_inode: returns inode needing a delegation break
3987 *
3988 * The caller must hold dir->i_mutex
3989 *
3990 * If vfs_link discovers a delegation on the to-be-linked file in need
3991 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3992 * inode in delegated_inode. The caller should then break the delegation
3993 * and retry. Because breaking a delegation may take a long time, the
3994 * caller should drop the i_mutex before doing so.
3995 *
3996 * Alternatively, a caller may pass NULL for delegated_inode. This may
3997 * be appropriate for callers that expect the underlying filesystem not
3998 * to be NFS exported.
3999 */
4000int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004001{
4002 struct inode *inode = old_dentry->d_inode;
Al Viro8de52772012-02-06 12:45:27 -05004003 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004004 int error;
4005
4006 if (!inode)
4007 return -ENOENT;
4008
Miklos Szeredia95164d2008-07-30 15:08:48 +02004009 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004010 if (error)
4011 return error;
4012
4013 if (dir->i_sb != inode->i_sb)
4014 return -EXDEV;
4015
4016 /*
4017 * A link to an append-only or immutable file cannot be created.
4018 */
4019 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4020 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05004021 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004022 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02004023 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004024 return -EPERM;
4025
4026 error = security_inode_link(old_dentry, dir, new_dentry);
4027 if (error)
4028 return error;
4029
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02004030 mutex_lock(&inode->i_mutex);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05304031 /* Make sure we don't allow creating hardlink to an unlinked file */
Al Virof4e0c302013-06-11 08:34:36 +04004032 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05304033 error = -ENOENT;
Al Viro8de52772012-02-06 12:45:27 -05004034 else if (max_links && inode->i_nlink >= max_links)
4035 error = -EMLINK;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004036 else {
4037 error = try_break_deleg(inode, delegated_inode);
4038 if (!error)
4039 error = dir->i_op->link(old_dentry, dir, new_dentry);
4040 }
Al Virof4e0c302013-06-11 08:34:36 +04004041
4042 if (!error && (inode->i_state & I_LINKABLE)) {
4043 spin_lock(&inode->i_lock);
4044 inode->i_state &= ~I_LINKABLE;
4045 spin_unlock(&inode->i_lock);
4046 }
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02004047 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07004048 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02004049 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004050 return error;
4051}
Al Viro4d359502014-03-14 12:20:17 -04004052EXPORT_SYMBOL(vfs_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004053
4054/*
4055 * Hardlinks are often used in delicate situations. We avoid
4056 * security-related surprises by not following symlinks on the
4057 * newname. --KAB
4058 *
4059 * We don't follow them on the oldname either to be compatible
4060 * with linux 2.0, and to avoid hard-linking to directories
4061 * and other special files. --ADM
4062 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01004063SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4064 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004065{
4066 struct dentry *new_dentry;
Al Virodae6ad82011-06-26 11:50:15 -04004067 struct path old_path, new_path;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004068 struct inode *delegated_inode = NULL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304069 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004070 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004071
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304072 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004073 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304074 /*
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004075 * To use null names we require CAP_DAC_READ_SEARCH
4076 * This ensures that not everyone will be able to create
4077 * handlink using the passed filedescriptor.
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304078 */
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004079 if (flags & AT_EMPTY_PATH) {
4080 if (!capable(CAP_DAC_READ_SEARCH))
4081 return -ENOENT;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304082 how = LOOKUP_EMPTY;
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004083 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004084
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304085 if (flags & AT_SYMLINK_FOLLOW)
4086 how |= LOOKUP_FOLLOW;
Jeff Layton442e31c2012-12-20 16:15:38 -05004087retry:
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304088 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004089 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04004090 return error;
4091
Jeff Layton442e31c2012-12-20 16:15:38 -05004092 new_dentry = user_path_create(newdfd, newname, &new_path,
4093 (how & LOOKUP_REVAL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004094 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07004095 if (IS_ERR(new_dentry))
Al Virodae6ad82011-06-26 11:50:15 -04004096 goto out;
4097
4098 error = -EXDEV;
4099 if (old_path.mnt != new_path.mnt)
4100 goto out_dput;
Kees Cook800179c2012-07-25 17:29:07 -07004101 error = may_linkat(&old_path);
4102 if (unlikely(error))
4103 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04004104 error = security_path_link(old_path.dentry, &new_path, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004105 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04004106 goto out_dput;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004107 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
Dave Hansen75c3f292008-02-15 14:37:45 -08004108out_dput:
Al Viro921a1652012-07-20 01:15:31 +04004109 done_path_create(&new_path, new_dentry);
J. Bruce Fields146a8592011-09-20 17:14:31 -04004110 if (delegated_inode) {
4111 error = break_deleg_wait(&delegated_inode);
Oleg Drokind22e6332014-01-31 15:41:58 -05004112 if (!error) {
4113 path_put(&old_path);
J. Bruce Fields146a8592011-09-20 17:14:31 -04004114 goto retry;
Oleg Drokind22e6332014-01-31 15:41:58 -05004115 }
J. Bruce Fields146a8592011-09-20 17:14:31 -04004116 }
Jeff Layton442e31c2012-12-20 16:15:38 -05004117 if (retry_estale(error, how)) {
Oleg Drokind22e6332014-01-31 15:41:58 -05004118 path_put(&old_path);
Jeff Layton442e31c2012-12-20 16:15:38 -05004119 how |= LOOKUP_REVAL;
4120 goto retry;
4121 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004122out:
Al Viro2d8f3032008-07-22 09:59:21 -04004123 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004124
4125 return error;
4126}
4127
Heiko Carstens3480b252009-01-14 14:14:16 +01004128SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004129{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004130 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004131}
4132
Miklos Szeredibc270272014-04-01 17:08:42 +02004133/**
4134 * vfs_rename - rename a filesystem object
4135 * @old_dir: parent of source
4136 * @old_dentry: source
4137 * @new_dir: parent of destination
4138 * @new_dentry: destination
4139 * @delegated_inode: returns an inode needing a delegation break
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004140 * @flags: rename flags
Miklos Szeredibc270272014-04-01 17:08:42 +02004141 *
4142 * The caller must hold multiple mutexes--see lock_rename()).
4143 *
4144 * If vfs_rename discovers a delegation in need of breaking at either
4145 * the source or destination, it will return -EWOULDBLOCK and return a
4146 * reference to the inode in delegated_inode. The caller should then
4147 * break the delegation and retry. Because breaking a delegation may
4148 * take a long time, the caller should drop all locks before doing
4149 * so.
4150 *
4151 * Alternatively, a caller may pass NULL for delegated_inode. This may
4152 * be appropriate for callers that expect the underlying filesystem not
4153 * to be NFS exported.
4154 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07004155 * The worst of all namespace operations - renaming directory. "Perverted"
4156 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4157 * Problems:
J. Bruce Fieldsd03b29a2014-02-17 16:52:33 -05004158 * a) we can get into loop creation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004159 * b) race potential - two innocent renames can create a loop together.
4160 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004161 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07004162 * story.
J. Bruce Fields6cedba82012-03-05 11:40:41 -05004163 * c) we have to lock _four_ objects - parents and victim (if it exists),
4164 * and source (if it is not a directory).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004165 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07004166 * whether the target exists). Solution: try to be smart with locking
4167 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004168 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07004169 * move will be locked. Thus we can rank directories by the tree
4170 * (ancestors first) and rank all non-directories after them.
4171 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004172 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004173 * HOWEVER, it relies on the assumption that any object with ->lookup()
4174 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4175 * we'd better make sure that there's no link(2) for them.
Sage Weile4eaac02011-05-24 13:06:07 -07004176 * d) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004177 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07004178 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05004179 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07004180 * locking].
4181 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004182int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004183 struct inode *new_dir, struct dentry *new_dentry,
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004184 struct inode **delegated_inode, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004185{
4186 int error;
Miklos Szeredibc270272014-04-01 17:08:42 +02004187 bool is_dir = d_is_dir(old_dentry);
Eric Paris59b0df22010-02-08 12:53:52 -05004188 const unsigned char *old_name;
Miklos Szeredibc270272014-04-01 17:08:42 +02004189 struct inode *source = old_dentry->d_inode;
4190 struct inode *target = new_dentry->d_inode;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004191 bool new_is_dir = false;
4192 unsigned max_links = new_dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004193
Miklos Szeredibc270272014-04-01 17:08:42 +02004194 if (source == target)
4195 return 0;
4196
Linus Torvalds1da177e2005-04-16 15:20:36 -07004197 error = may_delete(old_dir, old_dentry, is_dir);
4198 if (error)
4199 return error;
4200
Miklos Szeredida1ce062014-04-01 17:08:43 +02004201 if (!target) {
Miklos Szeredia95164d2008-07-30 15:08:48 +02004202 error = may_create(new_dir, new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004203 } else {
4204 new_is_dir = d_is_dir(new_dentry);
4205
4206 if (!(flags & RENAME_EXCHANGE))
4207 error = may_delete(new_dir, new_dentry, is_dir);
4208 else
4209 error = may_delete(new_dir, new_dentry, new_is_dir);
4210 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004211 if (error)
4212 return error;
4213
Miklos Szeredi7177a9c2014-07-23 15:15:30 +02004214 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004215 return -EPERM;
4216
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004217 if (flags && !old_dir->i_op->rename2)
4218 return -EINVAL;
4219
Miklos Szeredibc270272014-04-01 17:08:42 +02004220 /*
4221 * If we are going to change the parent - check write permissions,
4222 * we'll need to flip '..'.
4223 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02004224 if (new_dir != old_dir) {
4225 if (is_dir) {
4226 error = inode_permission(source, MAY_WRITE);
4227 if (error)
4228 return error;
4229 }
4230 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4231 error = inode_permission(target, MAY_WRITE);
4232 if (error)
4233 return error;
4234 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004235 }
Robert Love0eeca282005-07-12 17:06:03 -04004236
Miklos Szeredi0b3974e2014-04-01 17:08:43 +02004237 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4238 flags);
Miklos Szeredibc270272014-04-01 17:08:42 +02004239 if (error)
4240 return error;
4241
4242 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4243 dget(new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004244 if (!is_dir || (flags & RENAME_EXCHANGE))
Miklos Szeredibc270272014-04-01 17:08:42 +02004245 lock_two_nondirectories(source, target);
4246 else if (target)
4247 mutex_lock(&target->i_mutex);
4248
4249 error = -EBUSY;
Eric W. Biederman7af13642013-10-04 19:15:13 -07004250 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
Miklos Szeredibc270272014-04-01 17:08:42 +02004251 goto out;
4252
Miklos Szeredida1ce062014-04-01 17:08:43 +02004253 if (max_links && new_dir != old_dir) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004254 error = -EMLINK;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004255 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
Miklos Szeredibc270272014-04-01 17:08:42 +02004256 goto out;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004257 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4258 old_dir->i_nlink >= max_links)
4259 goto out;
4260 }
4261 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4262 shrink_dcache_parent(new_dentry);
4263 if (!is_dir) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004264 error = try_break_deleg(source, delegated_inode);
4265 if (error)
4266 goto out;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004267 }
4268 if (target && !new_is_dir) {
4269 error = try_break_deleg(target, delegated_inode);
4270 if (error)
4271 goto out;
Miklos Szeredibc270272014-04-01 17:08:42 +02004272 }
Miklos Szeredi7177a9c2014-07-23 15:15:30 +02004273 if (!old_dir->i_op->rename2) {
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004274 error = old_dir->i_op->rename(old_dir, old_dentry,
4275 new_dir, new_dentry);
4276 } else {
Miklos Szeredi7177a9c2014-07-23 15:15:30 +02004277 WARN_ON(old_dir->i_op->rename != NULL);
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004278 error = old_dir->i_op->rename2(old_dir, old_dentry,
4279 new_dir, new_dentry, flags);
4280 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004281 if (error)
4282 goto out;
4283
Miklos Szeredida1ce062014-04-01 17:08:43 +02004284 if (!(flags & RENAME_EXCHANGE) && target) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004285 if (is_dir)
4286 target->i_flags |= S_DEAD;
4287 dont_mount(new_dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07004288 detach_mounts(new_dentry);
Miklos Szeredibc270272014-04-01 17:08:42 +02004289 }
Miklos Szeredida1ce062014-04-01 17:08:43 +02004290 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4291 if (!(flags & RENAME_EXCHANGE))
4292 d_move(old_dentry, new_dentry);
4293 else
4294 d_exchange(old_dentry, new_dentry);
4295 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004296out:
Miklos Szeredida1ce062014-04-01 17:08:43 +02004297 if (!is_dir || (flags & RENAME_EXCHANGE))
Miklos Szeredibc270272014-04-01 17:08:42 +02004298 unlock_two_nondirectories(source, target);
4299 else if (target)
4300 mutex_unlock(&target->i_mutex);
4301 dput(new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004302 if (!error) {
Al Viro123df292009-12-25 04:57:57 -05004303 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Miklos Szeredida1ce062014-04-01 17:08:43 +02004304 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4305 if (flags & RENAME_EXCHANGE) {
4306 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4307 new_is_dir, NULL, new_dentry);
4308 }
4309 }
Robert Love0eeca282005-07-12 17:06:03 -04004310 fsnotify_oldname_free(old_name);
4311
Linus Torvalds1da177e2005-04-16 15:20:36 -07004312 return error;
4313}
Al Viro4d359502014-03-14 12:20:17 -04004314EXPORT_SYMBOL(vfs_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004315
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004316SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4317 int, newdfd, const char __user *, newname, unsigned int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004318{
Al Viro2ad94ae2008-07-21 09:32:51 -04004319 struct dentry *old_dentry, *new_dentry;
4320 struct dentry *trap;
Al Virof5beed72015-04-30 16:09:11 -04004321 struct path old_path, new_path;
4322 struct qstr old_last, new_last;
4323 int old_type, new_type;
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004324 struct inode *delegated_inode = NULL;
Jeff Layton91a27b22012-10-10 15:25:28 -04004325 struct filename *from;
4326 struct filename *to;
Al Virof5beed72015-04-30 16:09:11 -04004327 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
Jeff Laytonc6a94282012-12-11 12:10:10 -05004328 bool should_retry = false;
Al Viro2ad94ae2008-07-21 09:32:51 -04004329 int error;
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004330
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004331 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredida1ce062014-04-01 17:08:43 +02004332 return -EINVAL;
4333
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004334 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4335 (flags & RENAME_EXCHANGE))
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004336 return -EINVAL;
4337
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004338 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4339 return -EPERM;
4340
Al Virof5beed72015-04-30 16:09:11 -04004341 if (flags & RENAME_EXCHANGE)
4342 target_flags = 0;
4343
Jeff Laytonc6a94282012-12-11 12:10:10 -05004344retry:
Al Virof5beed72015-04-30 16:09:11 -04004345 from = user_path_parent(olddfd, oldname,
4346 &old_path, &old_last, &old_type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04004347 if (IS_ERR(from)) {
4348 error = PTR_ERR(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004349 goto exit;
Jeff Layton91a27b22012-10-10 15:25:28 -04004350 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004351
Al Virof5beed72015-04-30 16:09:11 -04004352 to = user_path_parent(newdfd, newname,
4353 &new_path, &new_last, &new_type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04004354 if (IS_ERR(to)) {
4355 error = PTR_ERR(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004356 goto exit1;
Jeff Layton91a27b22012-10-10 15:25:28 -04004357 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004358
4359 error = -EXDEV;
Al Virof5beed72015-04-30 16:09:11 -04004360 if (old_path.mnt != new_path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004361 goto exit2;
4362
Linus Torvalds1da177e2005-04-16 15:20:36 -07004363 error = -EBUSY;
Al Virof5beed72015-04-30 16:09:11 -04004364 if (old_type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004365 goto exit2;
4366
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004367 if (flags & RENAME_NOREPLACE)
4368 error = -EEXIST;
Al Virof5beed72015-04-30 16:09:11 -04004369 if (new_type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004370 goto exit2;
4371
Al Virof5beed72015-04-30 16:09:11 -04004372 error = mnt_want_write(old_path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02004373 if (error)
4374 goto exit2;
4375
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004376retry_deleg:
Al Virof5beed72015-04-30 16:09:11 -04004377 trap = lock_rename(new_path.dentry, old_path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004378
Al Virof5beed72015-04-30 16:09:11 -04004379 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004380 error = PTR_ERR(old_dentry);
4381 if (IS_ERR(old_dentry))
4382 goto exit3;
4383 /* source must exist */
4384 error = -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01004385 if (d_is_negative(old_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004386 goto exit4;
Al Virof5beed72015-04-30 16:09:11 -04004387 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004388 error = PTR_ERR(new_dentry);
4389 if (IS_ERR(new_dentry))
4390 goto exit4;
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004391 error = -EEXIST;
4392 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4393 goto exit5;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004394 if (flags & RENAME_EXCHANGE) {
4395 error = -ENOENT;
4396 if (d_is_negative(new_dentry))
4397 goto exit5;
4398
4399 if (!d_is_dir(new_dentry)) {
4400 error = -ENOTDIR;
Al Virof5beed72015-04-30 16:09:11 -04004401 if (new_last.name[new_last.len])
Miklos Szeredida1ce062014-04-01 17:08:43 +02004402 goto exit5;
4403 }
4404 }
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004405 /* unless the source is a directory trailing slashes give -ENOTDIR */
4406 if (!d_is_dir(old_dentry)) {
4407 error = -ENOTDIR;
Al Virof5beed72015-04-30 16:09:11 -04004408 if (old_last.name[old_last.len])
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004409 goto exit5;
Al Virof5beed72015-04-30 16:09:11 -04004410 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004411 goto exit5;
4412 }
4413 /* source should not be ancestor of target */
4414 error = -EINVAL;
4415 if (old_dentry == trap)
4416 goto exit5;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004417 /* target should not be an ancestor of source */
Miklos Szeredida1ce062014-04-01 17:08:43 +02004418 if (!(flags & RENAME_EXCHANGE))
4419 error = -ENOTEMPTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004420 if (new_dentry == trap)
4421 goto exit5;
4422
Al Virof5beed72015-04-30 16:09:11 -04004423 error = security_path_rename(&old_path, old_dentry,
4424 &new_path, new_dentry, flags);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004425 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02004426 goto exit5;
Al Virof5beed72015-04-30 16:09:11 -04004427 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4428 new_path.dentry->d_inode, new_dentry,
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004429 &delegated_inode, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004430exit5:
4431 dput(new_dentry);
4432exit4:
4433 dput(old_dentry);
4434exit3:
Al Virof5beed72015-04-30 16:09:11 -04004435 unlock_rename(new_path.dentry, old_path.dentry);
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004436 if (delegated_inode) {
4437 error = break_deleg_wait(&delegated_inode);
4438 if (!error)
4439 goto retry_deleg;
4440 }
Al Virof5beed72015-04-30 16:09:11 -04004441 mnt_drop_write(old_path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004442exit2:
Jeff Laytonc6a94282012-12-11 12:10:10 -05004443 if (retry_estale(error, lookup_flags))
4444 should_retry = true;
Al Virof5beed72015-04-30 16:09:11 -04004445 path_put(&new_path);
Al Viro2ad94ae2008-07-21 09:32:51 -04004446 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004447exit1:
Al Virof5beed72015-04-30 16:09:11 -04004448 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004449 putname(from);
Jeff Laytonc6a94282012-12-11 12:10:10 -05004450 if (should_retry) {
4451 should_retry = false;
4452 lookup_flags |= LOOKUP_REVAL;
4453 goto retry;
4454 }
Al Viro2ad94ae2008-07-21 09:32:51 -04004455exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004456 return error;
4457}
4458
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004459SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4460 int, newdfd, const char __user *, newname)
4461{
4462 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4463}
4464
Heiko Carstensa26eab22009-01-14 14:14:17 +01004465SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004466{
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004467 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004468}
4469
Miklos Szeredi787fb6b2014-10-24 00:14:36 +02004470int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4471{
4472 int error = may_create(dir, dentry);
4473 if (error)
4474 return error;
4475
4476 if (!dir->i_op->mknod)
4477 return -EPERM;
4478
4479 return dir->i_op->mknod(dir, dentry,
4480 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4481}
4482EXPORT_SYMBOL(vfs_whiteout);
4483
Al Viro5d826c82014-03-14 13:42:45 -04004484int readlink_copy(char __user *buffer, int buflen, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004485{
Al Viro5d826c82014-03-14 13:42:45 -04004486 int len = PTR_ERR(link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004487 if (IS_ERR(link))
4488 goto out;
4489
4490 len = strlen(link);
4491 if (len > (unsigned) buflen)
4492 len = buflen;
4493 if (copy_to_user(buffer, link, len))
4494 len = -EFAULT;
4495out:
4496 return len;
4497}
Al Viro5d826c82014-03-14 13:42:45 -04004498EXPORT_SYMBOL(readlink_copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004499
4500/*
4501 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4502 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4503 * using) it for any given inode is up to filesystem.
4504 */
4505int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4506{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004507 void *cookie;
Al Viro5f2c4172015-05-07 11:14:26 -04004508 struct inode *inode = d_inode(dentry);
4509 const char *link = inode->i_link;
Marcin Slusarz694a1762008-06-09 16:40:37 -07004510 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004511
Al Virod4dee482015-04-30 20:08:02 -04004512 if (!link) {
Al Viro5f2c4172015-05-07 11:14:26 -04004513 link = inode->i_op->follow_link(dentry, &cookie);
Al Virod4dee482015-04-30 20:08:02 -04004514 if (IS_ERR(link))
4515 return PTR_ERR(link);
4516 }
Al Viro680baac2015-05-02 13:32:22 -04004517 res = readlink_copy(buffer, buflen, link);
Al Viro5f2c4172015-05-07 11:14:26 -04004518 if (inode->i_op->put_link)
4519 inode->i_op->put_link(inode, cookie);
Marcin Slusarz694a1762008-06-09 16:40:37 -07004520 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004521}
Al Viro4d359502014-03-14 12:20:17 -04004522EXPORT_SYMBOL(generic_readlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004523
Linus Torvalds1da177e2005-04-16 15:20:36 -07004524/* get the link contents into pagecache */
4525static char *page_getlink(struct dentry * dentry, struct page **ppage)
4526{
Duane Griffinebd09ab2008-12-19 20:47:12 +00004527 char *kaddr;
4528 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004529 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07004530 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004531 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07004532 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004533 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00004534 kaddr = kmap(page);
4535 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4536 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004537}
4538
4539int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4540{
4541 struct page *page = NULL;
Al Viro5d826c82014-03-14 13:42:45 -04004542 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004543 if (page) {
4544 kunmap(page);
4545 page_cache_release(page);
4546 }
4547 return res;
4548}
Al Viro4d359502014-03-14 12:20:17 -04004549EXPORT_SYMBOL(page_readlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004550
Al Viro6e771372015-05-02 13:37:52 -04004551const char *page_follow_link_light(struct dentry *dentry, void **cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004552{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004553 struct page *page = NULL;
Al Viro680baac2015-05-02 13:32:22 -04004554 char *res = page_getlink(dentry, &page);
4555 if (!IS_ERR(res))
4556 *cookie = page;
4557 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004558}
Al Viro4d359502014-03-14 12:20:17 -04004559EXPORT_SYMBOL(page_follow_link_light);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004560
Al Viro5f2c4172015-05-07 11:14:26 -04004561void page_put_link(struct inode *unused, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004562{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004563 struct page *page = cookie;
Al Viro680baac2015-05-02 13:32:22 -04004564 kunmap(page);
4565 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004566}
Al Viro4d359502014-03-14 12:20:17 -04004567EXPORT_SYMBOL(page_put_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004568
Nick Piggin54566b22009-01-04 12:00:53 -08004569/*
4570 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4571 */
4572int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004573{
4574 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004575 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07004576 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08004577 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004578 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08004579 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4580 if (nofs)
4581 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004582
NeilBrown7e53cac2006-03-25 03:07:57 -08004583retry:
Nick Pigginafddba42007-10-16 01:25:01 -07004584 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08004585 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004586 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07004587 goto fail;
4588
Cong Wange8e3c3d2011-11-25 23:14:27 +08004589 kaddr = kmap_atomic(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004590 memcpy(kaddr, symname, len-1);
Cong Wange8e3c3d2011-11-25 23:14:27 +08004591 kunmap_atomic(kaddr);
Nick Pigginafddba42007-10-16 01:25:01 -07004592
4593 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4594 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004595 if (err < 0)
4596 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07004597 if (err < len-1)
4598 goto retry;
4599
Linus Torvalds1da177e2005-04-16 15:20:36 -07004600 mark_inode_dirty(inode);
4601 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004602fail:
4603 return err;
4604}
Al Viro4d359502014-03-14 12:20:17 -04004605EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004606
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004607int page_symlink(struct inode *inode, const char *symname, int len)
4608{
4609 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08004610 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004611}
Al Viro4d359502014-03-14 12:20:17 -04004612EXPORT_SYMBOL(page_symlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004613
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08004614const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004615 .readlink = generic_readlink,
4616 .follow_link = page_follow_link_light,
4617 .put_link = page_put_link,
4618};
Linus Torvalds1da177e2005-04-16 15:20:36 -07004619EXPORT_SYMBOL(page_symlink_inode_operations);