blob: 0c3974cd3ecd55670ccab51c596b0b141cbdd721 [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)
Dirk Steinmetzf2ca3792015-10-20 16:09:19 +0200958 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
Kees Cook800179c2012-07-25 17:29:07 -0700959 *
960 * Returns 0 if successful, -ve on error.
961 */
962static int may_linkat(struct path *link)
963{
Kees Cook800179c2012-07-25 17:29:07 -0700964 struct inode *inode;
965
966 if (!sysctl_protected_hardlinks)
967 return 0;
968
Kees Cook800179c2012-07-25 17:29:07 -0700969 inode = link->dentry->d_inode;
970
971 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
972 * otherwise, it must be a safe source.
973 */
Dirk Steinmetzf2ca3792015-10-20 16:09:19 +0200974 if (inode_owner_or_capable(inode) || safe_hardlink_source(inode))
Kees Cook800179c2012-07-25 17:29:07 -0700975 return 0;
976
Kees Cooka51d9ea2012-07-25 17:29:08 -0700977 audit_log_link_denied("linkat", link);
Kees Cook800179c2012-07-25 17:29:07 -0700978 return -EPERM;
979}
980
Al Viro3b2e7f72015-04-19 00:53:50 -0400981static __always_inline
982const char *get_link(struct nameidata *nd)
Ian Kent051d3812006-03-27 01:14:53 -0800983{
Al Viroab104922015-05-10 11:50:01 -0400984 struct saved *last = nd->stack + nd->depth - 1;
Al Viro1cf26652015-05-06 16:01:56 -0400985 struct dentry *dentry = last->link.dentry;
Al Viro237d8b32015-05-07 09:21:14 -0400986 struct inode *inode = last->inode;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400987 int error;
Al Viro0a959df2015-04-18 18:23:41 -0400988 const char *res;
Ian Kent051d3812006-03-27 01:14:53 -0800989
NeilBrown8fa9dd22015-03-23 13:37:40 +1100990 if (!(nd->flags & LOOKUP_RCU)) {
991 touch_atime(&last->link);
992 cond_resched();
993 } else if (atime_needs_update(&last->link, inode)) {
994 if (unlikely(unlazy_walk(nd, NULL, 0)))
995 return ERR_PTR(-ECHILD);
996 touch_atime(&last->link);
997 }
998
NeilBrownbda0be72015-03-23 13:37:39 +1100999 error = security_inode_follow_link(dentry, inode,
1000 nd->flags & LOOKUP_RCU);
1001 if (unlikely(error))
Al Viro6920a442015-05-10 10:43:46 -04001002 return ERR_PTR(error);
Al Viro36f3b4f2011-02-22 21:24:38 -05001003
Al Viro86acdca12009-12-22 23:45:11 -05001004 nd->last_type = LAST_BIND;
Al Virod4dee482015-04-30 20:08:02 -04001005 res = inode->i_link;
1006 if (!res) {
Al Viro8c1b4562015-05-09 18:15:21 -04001007 if (nd->flags & LOOKUP_RCU) {
1008 if (unlikely(unlazy_walk(nd, NULL, 0)))
1009 return ERR_PTR(-ECHILD);
1010 }
Al Viro6e771372015-05-02 13:37:52 -04001011 res = inode->i_op->follow_link(dentry, &last->cookie);
Al Virofab51e82015-05-10 11:01:00 -04001012 if (IS_ERR_OR_NULL(res)) {
Al Viro6920a442015-05-10 10:43:46 -04001013 last->cookie = NULL;
Al Virofab51e82015-05-10 11:01:00 -04001014 return res;
1015 }
Ian Kent051d3812006-03-27 01:14:53 -08001016 }
Al Virofab51e82015-05-10 11:01:00 -04001017 if (*res == '/') {
Al Viro8c1b4562015-05-09 18:15:21 -04001018 if (nd->flags & LOOKUP_RCU) {
Al Viro8f47a0162015-05-09 19:02:01 -04001019 struct dentry *d;
1020 if (!nd->root.mnt)
1021 set_root_rcu(nd);
1022 nd->path = nd->root;
1023 d = nd->path.dentry;
1024 nd->inode = d->d_inode;
1025 nd->seq = nd->root_seq;
1026 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
Al Viro8c1b4562015-05-09 18:15:21 -04001027 return ERR_PTR(-ECHILD);
Al Viro8f47a0162015-05-09 19:02:01 -04001028 } else {
1029 if (!nd->root.mnt)
1030 set_root(nd);
1031 path_put(&nd->path);
1032 nd->path = nd->root;
1033 path_get(&nd->root);
1034 nd->inode = nd->path.dentry->d_inode;
Al Viro8c1b4562015-05-09 18:15:21 -04001035 }
Al Virofab51e82015-05-10 11:01:00 -04001036 nd->flags |= LOOKUP_JUMPED;
1037 while (unlikely(*++res == '/'))
1038 ;
1039 }
1040 if (!*res)
1041 res = NULL;
Al Viro0a959df2015-04-18 18:23:41 -04001042 return res;
1043}
Al Viro6d7b5aa2012-06-10 04:15:17 -04001044
David Howellsf015f1262012-06-25 12:55:28 +01001045/*
1046 * follow_up - Find the mountpoint of path's vfsmount
1047 *
1048 * Given a path, find the mountpoint of its source file system.
1049 * Replace @path with the path of the mountpoint in the parent mount.
1050 * Up is towards /.
1051 *
1052 * Return 1 if we went up a level and 0 if we were already at the
1053 * root.
1054 */
Al Virobab77eb2009-04-18 03:26:48 -04001055int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056{
Al Viro0714a532011-11-24 22:19:58 -05001057 struct mount *mnt = real_mount(path->mnt);
1058 struct mount *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +10001060
Al Viro48a066e2013-09-29 22:06:07 -04001061 read_seqlock_excl(&mount_lock);
Al Viro0714a532011-11-24 22:19:58 -05001062 parent = mnt->mnt_parent;
Al Viro3c0a6162012-07-18 17:32:50 +04001063 if (parent == mnt) {
Al Viro48a066e2013-09-29 22:06:07 -04001064 read_sequnlock_excl(&mount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 return 0;
1066 }
Al Viro0714a532011-11-24 22:19:58 -05001067 mntget(&parent->mnt);
Al Viroa73324d2011-11-24 22:25:07 -05001068 mountpoint = dget(mnt->mnt_mountpoint);
Al Viro48a066e2013-09-29 22:06:07 -04001069 read_sequnlock_excl(&mount_lock);
Al Virobab77eb2009-04-18 03:26:48 -04001070 dput(path->dentry);
1071 path->dentry = mountpoint;
1072 mntput(path->mnt);
Al Viro0714a532011-11-24 22:19:58 -05001073 path->mnt = &parent->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 return 1;
1075}
Al Viro4d359502014-03-14 12:20:17 -04001076EXPORT_SYMBOL(follow_up);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001078/*
David Howells9875cf82011-01-14 18:45:21 +00001079 * Perform an automount
1080 * - return -EISDIR to tell follow_managed() to stop and return the path we
1081 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 */
NeilBrown756daf22015-03-23 13:37:38 +11001083static int follow_automount(struct path *path, struct nameidata *nd,
David Howells9875cf82011-01-14 18:45:21 +00001084 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +11001085{
David Howells9875cf82011-01-14 18:45:21 +00001086 struct vfsmount *mnt;
David Howellsea5b7782011-01-14 19:10:03 +00001087 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001088
David Howells9875cf82011-01-14 18:45:21 +00001089 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1090 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -07001091
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +02001092 /* We don't want to mount if someone's just doing a stat -
1093 * unless they're stat'ing a directory and appended a '/' to
1094 * the name.
1095 *
1096 * We do, however, want to mount if someone wants to open or
1097 * create a file of any type under the mountpoint, wants to
1098 * traverse through the mountpoint or wants to open the
1099 * mounted directory. Also, autofs may mark negative dentries
1100 * as being automount points. These will need the attentions
1101 * of the daemon to instantiate them before they can be used.
David Howells9875cf82011-01-14 18:45:21 +00001102 */
NeilBrown756daf22015-03-23 13:37:38 +11001103 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1104 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +02001105 path->dentry->d_inode)
1106 return -EISDIR;
1107
NeilBrown756daf22015-03-23 13:37:38 +11001108 nd->total_link_count++;
1109 if (nd->total_link_count >= 40)
David Howells9875cf82011-01-14 18:45:21 +00001110 return -ELOOP;
1111
1112 mnt = path->dentry->d_op->d_automount(path);
1113 if (IS_ERR(mnt)) {
1114 /*
1115 * The filesystem is allowed to return -EISDIR here to indicate
1116 * it doesn't want to automount. For instance, autofs would do
1117 * this so that its userspace daemon can mount on this dentry.
1118 *
1119 * However, we can only permit this if it's a terminal point in
1120 * the path being looked up; if it wasn't then the remainder of
1121 * the path is inaccessible and we should say so.
1122 */
NeilBrown756daf22015-03-23 13:37:38 +11001123 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
David Howells9875cf82011-01-14 18:45:21 +00001124 return -EREMOTE;
1125 return PTR_ERR(mnt);
1126 }
David Howellsea5b7782011-01-14 19:10:03 +00001127
David Howells9875cf82011-01-14 18:45:21 +00001128 if (!mnt) /* mount collision */
1129 return 0;
1130
Al Viro8aef1882011-06-16 15:10:06 +01001131 if (!*need_mntput) {
1132 /* lock_mount() may release path->mnt on error */
1133 mntget(path->mnt);
1134 *need_mntput = true;
1135 }
Al Viro19a167a2011-01-17 01:35:23 -05001136 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +00001137
David Howellsea5b7782011-01-14 19:10:03 +00001138 switch (err) {
1139 case -EBUSY:
1140 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -05001141 return 0;
David Howellsea5b7782011-01-14 19:10:03 +00001142 case 0:
Al Viro8aef1882011-06-16 15:10:06 +01001143 path_put(path);
David Howellsea5b7782011-01-14 19:10:03 +00001144 path->mnt = mnt;
1145 path->dentry = dget(mnt->mnt_root);
David Howellsea5b7782011-01-14 19:10:03 +00001146 return 0;
Al Viro19a167a2011-01-17 01:35:23 -05001147 default:
1148 return err;
David Howellsea5b7782011-01-14 19:10:03 +00001149 }
Al Viro19a167a2011-01-17 01:35:23 -05001150
David Howells9875cf82011-01-14 18:45:21 +00001151}
1152
1153/*
1154 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +00001155 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +00001156 * - Flagged as mountpoint
1157 * - Flagged as automount point
1158 *
1159 * This may only be called in refwalk mode.
1160 *
1161 * Serialization is taken care of in namespace.c
1162 */
NeilBrown756daf22015-03-23 13:37:38 +11001163static int follow_managed(struct path *path, struct nameidata *nd)
David Howells9875cf82011-01-14 18:45:21 +00001164{
Al Viro8aef1882011-06-16 15:10:06 +01001165 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
David Howells9875cf82011-01-14 18:45:21 +00001166 unsigned managed;
1167 bool need_mntput = false;
Al Viro8aef1882011-06-16 15:10:06 +01001168 int ret = 0;
David Howells9875cf82011-01-14 18:45:21 +00001169
1170 /* Given that we're not holding a lock here, we retain the value in a
1171 * local variable for each dentry as we look at it so that we don't see
1172 * the components of that value change under us */
1173 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1174 managed &= DCACHE_MANAGED_DENTRY,
1175 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +00001176 /* Allow the filesystem to manage the transit without i_mutex
1177 * being held. */
1178 if (managed & DCACHE_MANAGE_TRANSIT) {
1179 BUG_ON(!path->dentry->d_op);
1180 BUG_ON(!path->dentry->d_op->d_manage);
Al Viro1aed3e42011-03-18 09:09:02 -04001181 ret = path->dentry->d_op->d_manage(path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001182 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001183 break;
David Howellscc53ce52011-01-14 18:45:26 +00001184 }
1185
David Howells9875cf82011-01-14 18:45:21 +00001186 /* Transit to a mounted filesystem. */
1187 if (managed & DCACHE_MOUNTED) {
1188 struct vfsmount *mounted = lookup_mnt(path);
1189 if (mounted) {
1190 dput(path->dentry);
1191 if (need_mntput)
1192 mntput(path->mnt);
1193 path->mnt = mounted;
1194 path->dentry = dget(mounted->mnt_root);
1195 need_mntput = true;
1196 continue;
1197 }
1198
1199 /* Something is mounted on this dentry in another
1200 * namespace and/or whatever was mounted there in this
Al Viro48a066e2013-09-29 22:06:07 -04001201 * namespace got unmounted before lookup_mnt() could
1202 * get it */
David Howells9875cf82011-01-14 18:45:21 +00001203 }
1204
1205 /* Handle an automount point */
1206 if (managed & DCACHE_NEED_AUTOMOUNT) {
NeilBrown756daf22015-03-23 13:37:38 +11001207 ret = follow_automount(path, nd, &need_mntput);
David Howells9875cf82011-01-14 18:45:21 +00001208 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001209 break;
David Howells9875cf82011-01-14 18:45:21 +00001210 continue;
1211 }
1212
1213 /* We didn't change the current path point */
1214 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 }
Al Viro8aef1882011-06-16 15:10:06 +01001216
1217 if (need_mntput && path->mnt == mnt)
1218 mntput(path->mnt);
1219 if (ret == -EISDIR)
1220 ret = 0;
Al Viro84027522015-04-22 10:30:08 -04001221 if (need_mntput)
1222 nd->flags |= LOOKUP_JUMPED;
1223 if (unlikely(ret < 0))
1224 path_put_conditional(path, nd);
1225 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226}
1227
David Howellscc53ce52011-01-14 18:45:26 +00001228int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229{
1230 struct vfsmount *mounted;
1231
Al Viro1c755af2009-04-18 14:06:57 -04001232 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -04001234 dput(path->dentry);
1235 mntput(path->mnt);
1236 path->mnt = mounted;
1237 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 return 1;
1239 }
1240 return 0;
1241}
Al Viro4d359502014-03-14 12:20:17 -04001242EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243
NeilBrownb8faf032014-08-04 17:06:29 +10001244static inline int managed_dentry_rcu(struct dentry *dentry)
Ian Kent62a73752011-03-25 01:51:02 +08001245{
NeilBrownb8faf032014-08-04 17:06:29 +10001246 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1247 dentry->d_op->d_manage(dentry, true) : 0;
Ian Kent62a73752011-03-25 01:51:02 +08001248}
1249
David Howells9875cf82011-01-14 18:45:21 +00001250/*
Al Viro287548e2011-05-27 06:50:06 -04001251 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1252 * we meet a managed dentry that would need blocking.
David Howells9875cf82011-01-14 18:45:21 +00001253 */
1254static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
Al Viro254cf582015-05-05 09:40:46 -04001255 struct inode **inode, unsigned *seqp)
David Howells9875cf82011-01-14 18:45:21 +00001256{
Ian Kent62a73752011-03-25 01:51:02 +08001257 for (;;) {
Al Viroc7105362011-11-24 18:22:03 -05001258 struct mount *mounted;
Ian Kent62a73752011-03-25 01:51:02 +08001259 /*
1260 * Don't forget we might have a non-mountpoint managed dentry
1261 * that wants to block transit.
1262 */
NeilBrownb8faf032014-08-04 17:06:29 +10001263 switch (managed_dentry_rcu(path->dentry)) {
1264 case -ECHILD:
1265 default:
David Howellsab909112011-01-14 18:46:51 +00001266 return false;
NeilBrownb8faf032014-08-04 17:06:29 +10001267 case -EISDIR:
1268 return true;
1269 case 0:
1270 break;
1271 }
Ian Kent62a73752011-03-25 01:51:02 +08001272
1273 if (!d_mountpoint(path->dentry))
NeilBrownb8faf032014-08-04 17:06:29 +10001274 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
Ian Kent62a73752011-03-25 01:51:02 +08001275
Al Viro474279d2013-10-01 16:11:26 -04001276 mounted = __lookup_mnt(path->mnt, path->dentry);
David Howells9875cf82011-01-14 18:45:21 +00001277 if (!mounted)
1278 break;
Al Viroc7105362011-11-24 18:22:03 -05001279 path->mnt = &mounted->mnt;
1280 path->dentry = mounted->mnt.mnt_root;
Al Viroa3fbbde2011-11-07 21:21:26 +00001281 nd->flags |= LOOKUP_JUMPED;
Al Viro254cf582015-05-05 09:40:46 -04001282 *seqp = read_seqcount_begin(&path->dentry->d_seq);
Linus Torvalds59430262011-07-18 15:43:29 -07001283 /*
1284 * Update the inode too. We don't need to re-check the
1285 * dentry sequence number here after this d_inode read,
1286 * because a mount-point is always pinned.
1287 */
1288 *inode = path->dentry->d_inode;
David Howells9875cf82011-01-14 18:45:21 +00001289 }
Al Virof5be3e29122014-09-13 21:50:45 -04001290 return !read_seqretry(&mount_lock, nd->m_seq) &&
NeilBrownb8faf032014-08-04 17:06:29 +10001291 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
Al Viro287548e2011-05-27 06:50:06 -04001292}
1293
Nick Piggin31e6b012011-01-07 17:49:52 +11001294static int follow_dotdot_rcu(struct nameidata *nd)
1295{
Al Viro4023bfc2014-09-13 21:59:43 -04001296 struct inode *inode = nd->inode;
Al Viro7bd88372014-09-13 21:55:46 -04001297 if (!nd->root.mnt)
1298 set_root_rcu(nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11001299
David Howells9875cf82011-01-14 18:45:21 +00001300 while (1) {
Al Viroaed434a2015-05-12 12:22:47 -04001301 if (path_equal(&nd->path, &nd->root))
Nick Piggin31e6b012011-01-07 17:49:52 +11001302 break;
Nick Piggin31e6b012011-01-07 17:49:52 +11001303 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1304 struct dentry *old = nd->path.dentry;
1305 struct dentry *parent = old->d_parent;
1306 unsigned seq;
1307
Al Viro4023bfc2014-09-13 21:59:43 -04001308 inode = parent->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001309 seq = read_seqcount_begin(&parent->d_seq);
Al Viroaed434a2015-05-12 12:22:47 -04001310 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1311 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001312 nd->path.dentry = parent;
1313 nd->seq = seq;
Eric W. Biederman397d4252015-08-15 20:27:13 -05001314 if (unlikely(!path_connected(&nd->path)))
1315 return -ENOENT;
Nick Piggin31e6b012011-01-07 17:49:52 +11001316 break;
Al Viroaed434a2015-05-12 12:22:47 -04001317 } else {
1318 struct mount *mnt = real_mount(nd->path.mnt);
1319 struct mount *mparent = mnt->mnt_parent;
1320 struct dentry *mountpoint = mnt->mnt_mountpoint;
1321 struct inode *inode2 = mountpoint->d_inode;
1322 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1323 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1324 return -ECHILD;
1325 if (&mparent->mnt == nd->path.mnt)
1326 break;
1327 /* we know that mountpoint was pinned */
1328 nd->path.dentry = mountpoint;
1329 nd->path.mnt = &mparent->mnt;
1330 inode = inode2;
1331 nd->seq = seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11001332 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001333 }
Al Viroaed434a2015-05-12 12:22:47 -04001334 while (unlikely(d_mountpoint(nd->path.dentry))) {
Al Virob37199e2014-03-20 15:18:22 -04001335 struct mount *mounted;
1336 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
Al Viroaed434a2015-05-12 12:22:47 -04001337 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1338 return -ECHILD;
Al Virob37199e2014-03-20 15:18:22 -04001339 if (!mounted)
1340 break;
1341 nd->path.mnt = &mounted->mnt;
1342 nd->path.dentry = mounted->mnt.mnt_root;
Al Viro4023bfc2014-09-13 21:59:43 -04001343 inode = nd->path.dentry->d_inode;
Al Virob37199e2014-03-20 15:18:22 -04001344 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Virob37199e2014-03-20 15:18:22 -04001345 }
Al Viro4023bfc2014-09-13 21:59:43 -04001346 nd->inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001347 return 0;
1348}
1349
David Howells9875cf82011-01-14 18:45:21 +00001350/*
David Howellscc53ce52011-01-14 18:45:26 +00001351 * Follow down to the covering mount currently visible to userspace. At each
1352 * point, the filesystem owning that dentry may be queried as to whether the
1353 * caller is permitted to proceed or not.
David Howellscc53ce52011-01-14 18:45:26 +00001354 */
Al Viro7cc90cc2011-03-18 09:04:20 -04001355int follow_down(struct path *path)
David Howellscc53ce52011-01-14 18:45:26 +00001356{
1357 unsigned managed;
1358 int ret;
1359
1360 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1361 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1362 /* Allow the filesystem to manage the transit without i_mutex
1363 * being held.
1364 *
1365 * We indicate to the filesystem if someone is trying to mount
1366 * something here. This gives autofs the chance to deny anyone
1367 * other than its daemon the right to mount on its
1368 * superstructure.
1369 *
1370 * The filesystem may sleep at this point.
1371 */
1372 if (managed & DCACHE_MANAGE_TRANSIT) {
1373 BUG_ON(!path->dentry->d_op);
1374 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +00001375 ret = path->dentry->d_op->d_manage(
Al Viro1aed3e42011-03-18 09:09:02 -04001376 path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001377 if (ret < 0)
1378 return ret == -EISDIR ? 0 : ret;
1379 }
1380
1381 /* Transit to a mounted filesystem. */
1382 if (managed & DCACHE_MOUNTED) {
1383 struct vfsmount *mounted = lookup_mnt(path);
1384 if (!mounted)
1385 break;
1386 dput(path->dentry);
1387 mntput(path->mnt);
1388 path->mnt = mounted;
1389 path->dentry = dget(mounted->mnt_root);
1390 continue;
1391 }
1392
1393 /* Don't handle automount points here */
1394 break;
1395 }
1396 return 0;
1397}
Al Viro4d359502014-03-14 12:20:17 -04001398EXPORT_SYMBOL(follow_down);
David Howellscc53ce52011-01-14 18:45:26 +00001399
1400/*
David Howells9875cf82011-01-14 18:45:21 +00001401 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1402 */
1403static void follow_mount(struct path *path)
1404{
1405 while (d_mountpoint(path->dentry)) {
1406 struct vfsmount *mounted = lookup_mnt(path);
1407 if (!mounted)
1408 break;
1409 dput(path->dentry);
1410 mntput(path->mnt);
1411 path->mnt = mounted;
1412 path->dentry = dget(mounted->mnt_root);
1413 }
1414}
1415
Eric W. Biederman397d4252015-08-15 20:27:13 -05001416static int follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417{
Al Viro7bd88372014-09-13 21:55:46 -04001418 if (!nd->root.mnt)
1419 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001420
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 while(1) {
Jan Blunck4ac91372008-02-14 19:34:32 -08001422 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423
Al Viro2a737872009-04-07 11:49:53 -04001424 if (nd->path.dentry == nd->root.dentry &&
1425 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 break;
1427 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001428 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Al Viro3088dd72010-01-30 15:47:29 -05001429 /* rare case of legitimate dget_parent()... */
1430 nd->path.dentry = dget_parent(nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 dput(old);
Eric W. Biederman397d4252015-08-15 20:27:13 -05001432 if (unlikely(!path_connected(&nd->path)))
1433 return -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 break;
1435 }
Al Viro3088dd72010-01-30 15:47:29 -05001436 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 }
Al Viro79ed0222009-04-18 13:59:41 -04001439 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001440 nd->inode = nd->path.dentry->d_inode;
Eric W. Biederman397d4252015-08-15 20:27:13 -05001441 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442}
1443
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444/*
Miklos Szeredibad61182012-03-26 12:54:24 +02001445 * This looks up the name in dcache, possibly revalidates the old dentry and
1446 * allocates a new one if not found or not valid. In the need_lookup argument
1447 * returns whether i_op->lookup is necessary.
1448 *
1449 * dir->d_inode->i_mutex must be held
Nick Pigginbaa03892010-08-18 04:37:31 +10001450 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001451static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
Al Viro201f9562012-06-22 12:42:10 +04001452 unsigned int flags, bool *need_lookup)
Nick Pigginbaa03892010-08-18 04:37:31 +10001453{
Nick Pigginbaa03892010-08-18 04:37:31 +10001454 struct dentry *dentry;
Miklos Szeredibad61182012-03-26 12:54:24 +02001455 int error;
Nick Pigginbaa03892010-08-18 04:37:31 +10001456
Miklos Szeredibad61182012-03-26 12:54:24 +02001457 *need_lookup = false;
1458 dentry = d_lookup(dir, name);
1459 if (dentry) {
Jeff Layton39e3c952012-11-28 11:30:53 -05001460 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
Al Viro201f9562012-06-22 12:42:10 +04001461 error = d_revalidate(dentry, flags);
Miklos Szeredibad61182012-03-26 12:54:24 +02001462 if (unlikely(error <= 0)) {
1463 if (error < 0) {
1464 dput(dentry);
1465 return ERR_PTR(error);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001466 } else {
1467 d_invalidate(dentry);
Miklos Szeredibad61182012-03-26 12:54:24 +02001468 dput(dentry);
1469 dentry = NULL;
1470 }
1471 }
1472 }
1473 }
Nick Pigginbaa03892010-08-18 04:37:31 +10001474
Miklos Szeredibad61182012-03-26 12:54:24 +02001475 if (!dentry) {
1476 dentry = d_alloc(dir, name);
1477 if (unlikely(!dentry))
1478 return ERR_PTR(-ENOMEM);
Nick Pigginbaa03892010-08-18 04:37:31 +10001479
Miklos Szeredibad61182012-03-26 12:54:24 +02001480 *need_lookup = true;
Nick Pigginbaa03892010-08-18 04:37:31 +10001481 }
1482 return dentry;
1483}
1484
1485/*
J. Bruce Fields13a2c3b2013-10-23 16:09:16 -04001486 * Call i_op->lookup on the dentry. The dentry must be negative and
1487 * unhashed.
Miklos Szeredibad61182012-03-26 12:54:24 +02001488 *
1489 * dir->d_inode->i_mutex must be held
Josef Bacik44396f42011-05-31 11:58:49 -04001490 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001491static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
Al Viro72bd8662012-06-10 17:17:17 -04001492 unsigned int flags)
Josef Bacik44396f42011-05-31 11:58:49 -04001493{
Josef Bacik44396f42011-05-31 11:58:49 -04001494 struct dentry *old;
1495
1496 /* Don't create child dentry for a dead directory. */
Miklos Szeredibad61182012-03-26 12:54:24 +02001497 if (unlikely(IS_DEADDIR(dir))) {
Miklos Szeredie188dc02012-02-03 14:25:18 +01001498 dput(dentry);
Josef Bacik44396f42011-05-31 11:58:49 -04001499 return ERR_PTR(-ENOENT);
Miklos Szeredie188dc02012-02-03 14:25:18 +01001500 }
Josef Bacik44396f42011-05-31 11:58:49 -04001501
Al Viro72bd8662012-06-10 17:17:17 -04001502 old = dir->i_op->lookup(dir, dentry, flags);
Josef Bacik44396f42011-05-31 11:58:49 -04001503 if (unlikely(old)) {
1504 dput(dentry);
1505 dentry = old;
1506 }
1507 return dentry;
1508}
1509
Al Viroa3255542012-03-30 14:41:51 -04001510static struct dentry *__lookup_hash(struct qstr *name,
Al Viro72bd8662012-06-10 17:17:17 -04001511 struct dentry *base, unsigned int flags)
Al Viroa3255542012-03-30 14:41:51 -04001512{
Miklos Szeredibad61182012-03-26 12:54:24 +02001513 bool need_lookup;
Al Viroa3255542012-03-30 14:41:51 -04001514 struct dentry *dentry;
1515
Al Viro72bd8662012-06-10 17:17:17 -04001516 dentry = lookup_dcache(name, base, flags, &need_lookup);
Miklos Szeredibad61182012-03-26 12:54:24 +02001517 if (!need_lookup)
1518 return dentry;
Al Viroa3255542012-03-30 14:41:51 -04001519
Al Viro72bd8662012-06-10 17:17:17 -04001520 return lookup_real(base->d_inode, dentry, flags);
Al Viroa3255542012-03-30 14:41:51 -04001521}
1522
Josef Bacik44396f42011-05-31 11:58:49 -04001523/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 * It's more convoluted than I'd like it to be, but... it's still fairly
1525 * small and for now I'd prefer to have fast path as straight as possible.
1526 * It _is_ time-critical.
1527 */
Al Viroe97cdc82013-01-24 18:16:00 -05001528static int lookup_fast(struct nameidata *nd,
Al Viro254cf582015-05-05 09:40:46 -04001529 struct path *path, struct inode **inode,
1530 unsigned *seqp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531{
Jan Blunck4ac91372008-02-14 19:34:32 -08001532 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001533 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001534 int need_reval = 1;
1535 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001536 int err;
1537
Al Viro3cac2602009-08-13 18:27:43 +04001538 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001539 * Rename seqlock is not required here because in the off chance
1540 * of a false negative due to a concurrent rename, we're going to
1541 * do the non-racy lookup, below.
1542 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001543 if (nd->flags & LOOKUP_RCU) {
1544 unsigned seq;
Al Viro766c4cb2015-05-07 19:24:57 -04001545 bool negative;
Linus Torvaldsda53be12013-05-21 15:22:44 -07001546 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
Al Viro5a18fff2011-03-11 04:44:53 -05001547 if (!dentry)
1548 goto unlazy;
1549
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001550 /*
1551 * This sequence count validates that the inode matches
1552 * the dentry name information from lookup.
1553 */
David Howells63afdfc2015-05-06 15:59:00 +01001554 *inode = d_backing_inode(dentry);
Al Viro766c4cb2015-05-07 19:24:57 -04001555 negative = d_is_negative(dentry);
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001556 if (read_seqcount_retry(&dentry->d_seq, seq))
1557 return -ECHILD;
1558
1559 /*
1560 * This sequence count validates that the parent had no
1561 * changes while we did the lookup of the dentry above.
1562 *
1563 * The memory barrier in read_seqcount_begin of child is
1564 * enough, we can use __read_seqcount_retry here.
1565 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001566 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1567 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001568
Al Viro254cf582015-05-05 09:40:46 -04001569 *seqp = seq;
Al Viro24643082011-02-15 01:26:22 -05001570 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
Al Viro4ce16ef32012-06-10 16:10:59 -04001571 status = d_revalidate(dentry, nd->flags);
Al Viro5a18fff2011-03-11 04:44:53 -05001572 if (unlikely(status <= 0)) {
1573 if (status != -ECHILD)
1574 need_reval = 0;
1575 goto unlazy;
1576 }
Al Viro24643082011-02-15 01:26:22 -05001577 }
Trond Myklebustdaf37612015-10-09 13:44:34 -04001578 /*
1579 * Note: do negative dentry check after revalidation in
1580 * case that drops it.
1581 */
1582 if (negative)
1583 return -ENOENT;
Nick Piggin31e6b012011-01-07 17:49:52 +11001584 path->mnt = mnt;
1585 path->dentry = dentry;
Al Viro254cf582015-05-05 09:40:46 -04001586 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
NeilBrownb8faf032014-08-04 17:06:29 +10001587 return 0;
Al Viro5a18fff2011-03-11 04:44:53 -05001588unlazy:
Al Viro254cf582015-05-05 09:40:46 -04001589 if (unlazy_walk(nd, dentry, seq))
Al Viro19660af2011-03-25 10:32:48 -04001590 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001591 } else {
Al Viroe97cdc82013-01-24 18:16:00 -05001592 dentry = __d_lookup(parent, &nd->last);
Nick Piggin31e6b012011-01-07 17:49:52 +11001593 }
Al Viro5a18fff2011-03-11 04:44:53 -05001594
Al Viro81e6f522012-03-30 14:48:04 -04001595 if (unlikely(!dentry))
1596 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001597
Al Viro5a18fff2011-03-11 04:44:53 -05001598 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
Al Viro4ce16ef32012-06-10 16:10:59 -04001599 status = d_revalidate(dentry, nd->flags);
Al Viro5a18fff2011-03-11 04:44:53 -05001600 if (unlikely(status <= 0)) {
1601 if (status < 0) {
1602 dput(dentry);
1603 return status;
1604 }
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001605 d_invalidate(dentry);
1606 dput(dentry);
1607 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001608 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001609
Al Viro766c4cb2015-05-07 19:24:57 -04001610 if (unlikely(d_is_negative(dentry))) {
1611 dput(dentry);
1612 return -ENOENT;
1613 }
David Howells9875cf82011-01-14 18:45:21 +00001614 path->mnt = mnt;
1615 path->dentry = dentry;
NeilBrown756daf22015-03-23 13:37:38 +11001616 err = follow_managed(path, nd);
Al Viro84027522015-04-22 10:30:08 -04001617 if (likely(!err))
David Howells63afdfc2015-05-06 15:59:00 +01001618 *inode = d_backing_inode(path->dentry);
Al Viro84027522015-04-22 10:30:08 -04001619 return err;
Al Viro81e6f522012-03-30 14:48:04 -04001620
1621need_lookup:
Miklos Szeredi697f5142012-05-21 17:30:05 +02001622 return 1;
1623}
1624
1625/* Fast lookup failed, do it the slow way */
Al Virocc2a5272013-01-24 18:19:49 -05001626static int lookup_slow(struct nameidata *nd, struct path *path)
Miklos Szeredi697f5142012-05-21 17:30:05 +02001627{
1628 struct dentry *dentry, *parent;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001629
1630 parent = nd->path.dentry;
Al Viro81e6f522012-03-30 14:48:04 -04001631 BUG_ON(nd->inode != parent->d_inode);
1632
1633 mutex_lock(&parent->d_inode->i_mutex);
Al Virocc2a5272013-01-24 18:19:49 -05001634 dentry = __lookup_hash(&nd->last, parent, nd->flags);
Al Viro81e6f522012-03-30 14:48:04 -04001635 mutex_unlock(&parent->d_inode->i_mutex);
1636 if (IS_ERR(dentry))
1637 return PTR_ERR(dentry);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001638 path->mnt = nd->path.mnt;
1639 path->dentry = dentry;
Al Viro84027522015-04-22 10:30:08 -04001640 return follow_managed(path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641}
1642
Al Viro52094c82011-02-21 21:34:47 -05001643static inline int may_lookup(struct nameidata *nd)
1644{
1645 if (nd->flags & LOOKUP_RCU) {
Al Viro4ad5abb2011-06-20 19:57:03 -04001646 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
Al Viro52094c82011-02-21 21:34:47 -05001647 if (err != -ECHILD)
1648 return err;
Al Viro6e9918b2015-05-05 09:26:05 -04001649 if (unlazy_walk(nd, NULL, 0))
Al Viro52094c82011-02-21 21:34:47 -05001650 return -ECHILD;
1651 }
Al Viro4ad5abb2011-06-20 19:57:03 -04001652 return inode_permission(nd->inode, MAY_EXEC);
Al Viro52094c82011-02-21 21:34:47 -05001653}
1654
Al Viro9856fa12011-03-04 14:22:06 -05001655static inline int handle_dots(struct nameidata *nd, int type)
1656{
1657 if (type == LAST_DOTDOT) {
1658 if (nd->flags & LOOKUP_RCU) {
Al Viro70291ae2015-05-04 07:53:00 -04001659 return follow_dotdot_rcu(nd);
Al Viro9856fa12011-03-04 14:22:06 -05001660 } else
Eric W. Biederman397d4252015-08-15 20:27:13 -05001661 return follow_dotdot(nd);
Al Viro9856fa12011-03-04 14:22:06 -05001662 }
1663 return 0;
1664}
1665
Al Viro181548c2015-05-07 19:54:34 -04001666static int pick_link(struct nameidata *nd, struct path *link,
1667 struct inode *inode, unsigned seq)
Al Virod63ff282015-05-04 18:13:23 -04001668{
Al Viro626de992015-05-04 18:26:59 -04001669 int error;
Al Viro1cf26652015-05-06 16:01:56 -04001670 struct saved *last;
NeilBrown756daf22015-03-23 13:37:38 +11001671 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
Al Viro626de992015-05-04 18:26:59 -04001672 path_to_nameidata(link, nd);
1673 return -ELOOP;
1674 }
Al Virobc40aee2015-05-09 13:04:24 -04001675 if (!(nd->flags & LOOKUP_RCU)) {
Al Viro79733872015-05-09 12:55:43 -04001676 if (link->mnt == nd->path.mnt)
1677 mntget(link->mnt);
Al Virod63ff282015-05-04 18:13:23 -04001678 }
Al Viro626de992015-05-04 18:26:59 -04001679 error = nd_alloc_stack(nd);
1680 if (unlikely(error)) {
Al Virobc40aee2015-05-09 13:04:24 -04001681 if (error == -ECHILD) {
1682 if (unlikely(unlazy_link(nd, link, seq)))
1683 return -ECHILD;
1684 error = nd_alloc_stack(nd);
1685 }
1686 if (error) {
1687 path_put(link);
1688 return error;
1689 }
Al Viro626de992015-05-04 18:26:59 -04001690 }
1691
Al Viroab104922015-05-10 11:50:01 -04001692 last = nd->stack + nd->depth++;
Al Viro1cf26652015-05-06 16:01:56 -04001693 last->link = *link;
Al Viroab104922015-05-10 11:50:01 -04001694 last->cookie = NULL;
Al Viro181548c2015-05-07 19:54:34 -04001695 last->inode = inode;
Al Viro0450b2d2015-05-08 13:23:53 -04001696 last->seq = seq;
Al Virod63ff282015-05-04 18:13:23 -04001697 return 1;
1698}
1699
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001700/*
1701 * Do we need to follow links? We _really_ want to be able
1702 * to do this check without having to look at inode->i_op,
1703 * so we keep a cache of "no, this doesn't need follow_link"
1704 * for the common case.
1705 */
Al Viro254cf582015-05-05 09:40:46 -04001706static inline int should_follow_link(struct nameidata *nd, struct path *link,
Al Viro181548c2015-05-07 19:54:34 -04001707 int follow,
1708 struct inode *inode, unsigned seq)
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001709{
Al Virod63ff282015-05-04 18:13:23 -04001710 if (likely(!d_is_symlink(link->dentry)))
1711 return 0;
1712 if (!follow)
1713 return 0;
Al Viro181548c2015-05-07 19:54:34 -04001714 return pick_link(nd, link, inode, seq);
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001715}
1716
Al Viro4693a542015-05-04 17:47:11 -04001717enum {WALK_GET = 1, WALK_PUT = 2};
1718
1719static int walk_component(struct nameidata *nd, int flags)
Al Viroce57dfc2011-03-13 19:58:58 -04001720{
Al Virocaa85632015-04-22 17:52:47 -04001721 struct path path;
Al Viroce57dfc2011-03-13 19:58:58 -04001722 struct inode *inode;
Al Viro254cf582015-05-05 09:40:46 -04001723 unsigned seq;
Al Viroce57dfc2011-03-13 19:58:58 -04001724 int err;
1725 /*
1726 * "." and ".." are special - ".." especially so because it has
1727 * to be able to know about the current root directory and
1728 * parent relationships.
1729 */
Al Viro4693a542015-05-04 17:47:11 -04001730 if (unlikely(nd->last_type != LAST_NORM)) {
1731 err = handle_dots(nd, nd->last_type);
1732 if (flags & WALK_PUT)
1733 put_link(nd);
1734 return err;
1735 }
Al Viro254cf582015-05-05 09:40:46 -04001736 err = lookup_fast(nd, &path, &inode, &seq);
Al Viroce57dfc2011-03-13 19:58:58 -04001737 if (unlikely(err)) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001738 if (err < 0)
Al Virof0a9ba72015-05-04 07:59:30 -04001739 return err;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001740
Al Virocaa85632015-04-22 17:52:47 -04001741 err = lookup_slow(nd, &path);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001742 if (err < 0)
Al Virof0a9ba72015-05-04 07:59:30 -04001743 return err;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001744
David Howells63afdfc2015-05-06 15:59:00 +01001745 inode = d_backing_inode(path.dentry);
Al Viro254cf582015-05-05 09:40:46 -04001746 seq = 0; /* we are already out of RCU mode */
Al Viro766c4cb2015-05-07 19:24:57 -04001747 err = -ENOENT;
Al Virocaa85632015-04-22 17:52:47 -04001748 if (d_is_negative(path.dentry))
Al Viro766c4cb2015-05-07 19:24:57 -04001749 goto out_path_put;
Al Viroce57dfc2011-03-13 19:58:58 -04001750 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001751
Al Viro4693a542015-05-04 17:47:11 -04001752 if (flags & WALK_PUT)
1753 put_link(nd);
Al Viro181548c2015-05-07 19:54:34 -04001754 err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq);
Al Virod63ff282015-05-04 18:13:23 -04001755 if (unlikely(err))
1756 return err;
Al Virocaa85632015-04-22 17:52:47 -04001757 path_to_nameidata(&path, nd);
Al Viroce57dfc2011-03-13 19:58:58 -04001758 nd->inode = inode;
Al Viro254cf582015-05-05 09:40:46 -04001759 nd->seq = seq;
Al Viroce57dfc2011-03-13 19:58:58 -04001760 return 0;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001761
1762out_path_put:
Al Virocaa85632015-04-22 17:52:47 -04001763 path_to_nameidata(&path, nd);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001764 return err;
Al Viroce57dfc2011-03-13 19:58:58 -04001765}
1766
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767/*
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001768 * We can do the critical dentry name comparison and hashing
1769 * operations one word at a time, but we are limited to:
1770 *
1771 * - Architectures with fast unaligned word accesses. We could
1772 * do a "get_unaligned()" if this helps and is sufficiently
1773 * fast.
1774 *
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001775 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1776 * do not trap on the (extremely unlikely) case of a page
1777 * crossing operation.
1778 *
1779 * - Furthermore, we need an efficient 64-bit compile for the
1780 * 64-bit case in order to generate the "number of bytes in
1781 * the final mask". Again, that could be replaced with a
1782 * efficient population count instruction or similar.
1783 */
1784#ifdef CONFIG_DCACHE_WORD_ACCESS
1785
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001786#include <asm/word-at-a-time.h>
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001787
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001788#ifdef CONFIG_64BIT
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001789
1790static inline unsigned int fold_hash(unsigned long hash)
1791{
Linus Torvalds99d263d2014-09-13 11:30:10 -07001792 return hash_64(hash, 32);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001793}
1794
1795#else /* 32-bit case */
1796
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001797#define fold_hash(x) (x)
1798
1799#endif
1800
1801unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1802{
1803 unsigned long a, mask;
1804 unsigned long hash = 0;
1805
1806 for (;;) {
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001807 a = load_unaligned_zeropad(name);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001808 if (len < sizeof(unsigned long))
1809 break;
1810 hash += a;
Al Virof132c5b2012-03-22 21:59:52 +00001811 hash *= 9;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001812 name += sizeof(unsigned long);
1813 len -= sizeof(unsigned long);
1814 if (!len)
1815 goto done;
1816 }
Will Deacona5c21dc2013-12-12 17:40:21 +00001817 mask = bytemask_from_count(len);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001818 hash += mask & a;
1819done:
1820 return fold_hash(hash);
1821}
1822EXPORT_SYMBOL(full_name_hash);
1823
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001824/*
1825 * Calculate the length and hash of the path component, and
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001826 * return the "hash_len" as the result.
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001827 */
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001828static inline u64 hash_name(const char *name)
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001829{
Linus Torvalds36126f82012-05-26 10:43:17 -07001830 unsigned long a, b, adata, bdata, mask, hash, len;
1831 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001832
1833 hash = a = 0;
1834 len = -sizeof(unsigned long);
1835 do {
1836 hash = (hash + a) * 9;
1837 len += sizeof(unsigned long);
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001838 a = load_unaligned_zeropad(name+len);
Linus Torvalds36126f82012-05-26 10:43:17 -07001839 b = a ^ REPEAT_BYTE('/');
1840 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001841
Linus Torvalds36126f82012-05-26 10:43:17 -07001842 adata = prep_zero_mask(a, adata, &constants);
1843 bdata = prep_zero_mask(b, bdata, &constants);
1844
1845 mask = create_zero_mask(adata | bdata);
1846
1847 hash += a & zero_bytemask(mask);
Linus Torvalds9226b5b2014-09-14 17:28:32 -07001848 len += find_zero(mask);
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001849 return hashlen_create(fold_hash(hash), len);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001850}
1851
1852#else
1853
Linus Torvalds0145acc2012-03-02 14:32:59 -08001854unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1855{
1856 unsigned long hash = init_name_hash();
1857 while (len--)
1858 hash = partial_name_hash(*name++, hash);
1859 return end_name_hash(hash);
1860}
Linus Torvaldsae942ae2012-03-02 19:40:57 -08001861EXPORT_SYMBOL(full_name_hash);
Linus Torvalds0145acc2012-03-02 14:32:59 -08001862
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001863/*
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001864 * We know there's a real path component here of at least
1865 * one character.
1866 */
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001867static inline u64 hash_name(const char *name)
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001868{
1869 unsigned long hash = init_name_hash();
1870 unsigned long len = 0, c;
1871
1872 c = (unsigned char)*name;
1873 do {
1874 len++;
1875 hash = partial_name_hash(c, hash);
1876 c = (unsigned char)name[len];
1877 } while (c && c != '/');
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001878 return hashlen_create(end_name_hash(hash), len);
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001879}
1880
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001881#endif
1882
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001883/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01001885 * This is the basic name resolution function, turning a pathname into
1886 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01001888 * Returns 0 and nd will have valid dentry and mnt on success.
1889 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 */
Al Viro6de88d72009-08-09 01:41:57 +04001891static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893 int err;
Al Viro32cd7462015-04-18 20:30:49 -04001894
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895 while (*name=='/')
1896 name++;
1897 if (!*name)
Al Viro9e18f102015-04-18 20:44:34 -04001898 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900 /* At this point we know we have a real path component. */
1901 for(;;) {
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001902 u64 hash_len;
Al Virofe479a52011-02-22 15:10:03 -05001903 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904
Al Viro52094c82011-02-21 21:34:47 -05001905 err = may_lookup(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906 if (err)
Al Viro3595e232015-05-09 16:54:45 -04001907 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001909 hash_len = hash_name(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910
Al Virofe479a52011-02-22 15:10:03 -05001911 type = LAST_NORM;
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001912 if (name[0] == '.') switch (hashlen_len(hash_len)) {
Al Virofe479a52011-02-22 15:10:03 -05001913 case 2:
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001914 if (name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05001915 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05001916 nd->flags |= LOOKUP_JUMPED;
1917 }
Al Virofe479a52011-02-22 15:10:03 -05001918 break;
1919 case 1:
1920 type = LAST_DOT;
1921 }
Al Viro5a202bc2011-03-08 14:17:44 -05001922 if (likely(type == LAST_NORM)) {
1923 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05001924 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05001925 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
James Hogana060dc52014-09-16 13:07:35 +01001926 struct qstr this = { { .hash_len = hash_len }, .name = name };
Linus Torvaldsda53be12013-05-21 15:22:44 -07001927 err = parent->d_op->d_hash(parent, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05001928 if (err < 0)
Al Viro3595e232015-05-09 16:54:45 -04001929 return err;
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001930 hash_len = this.hash_len;
1931 name = this.name;
Al Viro5a202bc2011-03-08 14:17:44 -05001932 }
1933 }
Al Virofe479a52011-02-22 15:10:03 -05001934
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001935 nd->last.hash_len = hash_len;
1936 nd->last.name = name;
Al Viro5f4a6a62013-01-24 18:04:22 -05001937 nd->last_type = type;
1938
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001939 name += hashlen_len(hash_len);
1940 if (!*name)
Al Virobdf6cbf2015-04-18 20:21:40 -04001941 goto OK;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001942 /*
1943 * If it wasn't NUL, we know it was '/'. Skip that
1944 * slash, and continue until no more slashes.
1945 */
1946 do {
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001947 name++;
1948 } while (unlikely(*name == '/'));
Al Viro8620c232015-05-04 08:58:35 -04001949 if (unlikely(!*name)) {
1950OK:
Al Viro368ee9b2015-05-08 17:19:59 -04001951 /* pathname body, done */
Al Viro8620c232015-05-04 08:58:35 -04001952 if (!nd->depth)
1953 return 0;
1954 name = nd->stack[nd->depth - 1].name;
Al Viro368ee9b2015-05-08 17:19:59 -04001955 /* trailing symlink, done */
Al Viro8620c232015-05-04 08:58:35 -04001956 if (!name)
1957 return 0;
1958 /* last component of nested symlink */
Al Viro4693a542015-05-04 17:47:11 -04001959 err = walk_component(nd, WALK_GET | WALK_PUT);
Al Viro8620c232015-05-04 08:58:35 -04001960 } else {
Al Viro4693a542015-05-04 17:47:11 -04001961 err = walk_component(nd, WALK_GET);
Al Viro8620c232015-05-04 08:58:35 -04001962 }
Al Viroce57dfc2011-03-13 19:58:58 -04001963 if (err < 0)
Al Viro3595e232015-05-09 16:54:45 -04001964 return err;
Al Virofe479a52011-02-22 15:10:03 -05001965
Al Viroce57dfc2011-03-13 19:58:58 -04001966 if (err) {
Al Viro626de992015-05-04 18:26:59 -04001967 const char *s = get_link(nd);
Al Viro5a460272015-04-17 23:44:45 -04001968
Viresh Kumara1c83682015-08-12 15:59:44 +05301969 if (IS_ERR(s))
Al Viro3595e232015-05-09 16:54:45 -04001970 return PTR_ERR(s);
Al Virod40bcc02015-04-18 20:03:03 -04001971 err = 0;
1972 if (unlikely(!s)) {
1973 /* jumped */
Al Virob9ff4422015-05-02 20:19:23 -04001974 put_link(nd);
Al Virod40bcc02015-04-18 20:03:03 -04001975 } else {
Al Virofab51e82015-05-10 11:01:00 -04001976 nd->stack[nd->depth - 1].name = name;
1977 name = s;
1978 continue;
Al Virod40bcc02015-04-18 20:03:03 -04001979 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001980 }
Al Viro97242f92015-08-01 19:59:28 -04001981 if (unlikely(!d_can_lookup(nd->path.dentry))) {
1982 if (nd->flags & LOOKUP_RCU) {
1983 if (unlazy_walk(nd, NULL, 0))
1984 return -ECHILD;
1985 }
Al Viro3595e232015-05-09 16:54:45 -04001986 return -ENOTDIR;
Al Viro97242f92015-08-01 19:59:28 -04001987 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989}
1990
Al Viroc8a53ee2015-05-12 18:43:07 -04001991static const char *path_init(struct nameidata *nd, unsigned flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001993 int retval = 0;
Al Viroc8a53ee2015-05-12 18:43:07 -04001994 const char *s = nd->name->name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995
1996 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro980f3ea2014-11-20 14:20:24 -05001997 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 nd->depth = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05001999 if (flags & LOOKUP_ROOT) {
David Howellsb18825a2013-09-12 19:22:53 +01002000 struct dentry *root = nd->root.dentry;
2001 struct inode *inode = root->d_inode;
Al Virofd2f7cb2015-02-22 20:07:13 -05002002 if (*s) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002003 if (!d_can_lookup(root))
Al Viro368ee9b2015-05-08 17:19:59 -04002004 return ERR_PTR(-ENOTDIR);
Al Viro73d049a2011-03-11 12:08:24 -05002005 retval = inode_permission(inode, MAY_EXEC);
2006 if (retval)
Al Viro368ee9b2015-05-08 17:19:59 -04002007 return ERR_PTR(retval);
Al Viro73d049a2011-03-11 12:08:24 -05002008 }
Al Viro5b6ca022011-03-09 23:04:47 -05002009 nd->path = nd->root;
2010 nd->inode = inode;
2011 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05002012 rcu_read_lock();
Al Viro5b6ca022011-03-09 23:04:47 -05002013 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro8f47a0162015-05-09 19:02:01 -04002014 nd->root_seq = nd->seq;
Al Viro48a066e2013-09-29 22:06:07 -04002015 nd->m_seq = read_seqbegin(&mount_lock);
Al Viro5b6ca022011-03-09 23:04:47 -05002016 } else {
2017 path_get(&nd->path);
2018 }
Al Viro368ee9b2015-05-08 17:19:59 -04002019 return s;
Al Viro5b6ca022011-03-09 23:04:47 -05002020 }
2021
Al Viro2a737872009-04-07 11:49:53 -04002022 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023
Al Viro48a066e2013-09-29 22:06:07 -04002024 nd->m_seq = read_seqbegin(&mount_lock);
Al Virofd2f7cb2015-02-22 20:07:13 -05002025 if (*s == '/') {
Al Viroe41f7d42011-02-22 14:02:58 -05002026 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05002027 rcu_read_lock();
Al Viro06d71372015-06-29 12:07:04 -04002028 set_root_rcu(nd);
2029 nd->seq = nd->root_seq;
Al Viroe41f7d42011-02-22 14:02:58 -05002030 } else {
2031 set_root(nd);
2032 path_get(&nd->root);
2033 }
Al Viro2a737872009-04-07 11:49:53 -04002034 nd->path = nd->root;
Al Viroc8a53ee2015-05-12 18:43:07 -04002035 } else if (nd->dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05002036 if (flags & LOOKUP_RCU) {
2037 struct fs_struct *fs = current->fs;
2038 unsigned seq;
2039
Al Viro8b61e742013-11-08 12:45:01 -05002040 rcu_read_lock();
Al Viroe41f7d42011-02-22 14:02:58 -05002041
2042 do {
2043 seq = read_seqcount_begin(&fs->seq);
2044 nd->path = fs->pwd;
2045 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2046 } while (read_seqcount_retry(&fs->seq, seq));
2047 } else {
2048 get_fs_pwd(current->fs, &nd->path);
2049 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002050 } else {
Jeff Layton582aa642012-12-11 08:56:16 -05002051 /* Caller must check execute permissions on the starting path component */
Al Viroc8a53ee2015-05-12 18:43:07 -04002052 struct fd f = fdget_raw(nd->dfd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002053 struct dentry *dentry;
2054
Al Viro2903ff02012-08-28 12:52:22 -04002055 if (!f.file)
Al Viro368ee9b2015-05-08 17:19:59 -04002056 return ERR_PTR(-EBADF);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002057
Al Viro2903ff02012-08-28 12:52:22 -04002058 dentry = f.file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002059
Al Virofd2f7cb2015-02-22 20:07:13 -05002060 if (*s) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002061 if (!d_can_lookup(dentry)) {
Al Viro2903ff02012-08-28 12:52:22 -04002062 fdput(f);
Al Viro368ee9b2015-05-08 17:19:59 -04002063 return ERR_PTR(-ENOTDIR);
Al Viro2903ff02012-08-28 12:52:22 -04002064 }
Al Virof52e0c12011-03-14 18:56:51 -04002065 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002066
Al Viro2903ff02012-08-28 12:52:22 -04002067 nd->path = f.file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05002068 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05002069 rcu_read_lock();
Al Viro34a26b92015-05-11 08:05:05 -04002070 nd->inode = nd->path.dentry->d_inode;
2071 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viroe41f7d42011-02-22 14:02:58 -05002072 } else {
Al Viro2903ff02012-08-28 12:52:22 -04002073 path_get(&nd->path);
Al Viro34a26b92015-05-11 08:05:05 -04002074 nd->inode = nd->path.dentry->d_inode;
Al Viroe41f7d42011-02-22 14:02:58 -05002075 }
Al Viro34a26b92015-05-11 08:05:05 -04002076 fdput(f);
Al Viro368ee9b2015-05-08 17:19:59 -04002077 return s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 }
Al Viroe41f7d42011-02-22 14:02:58 -05002079
Nick Piggin31e6b012011-01-07 17:49:52 +11002080 nd->inode = nd->path.dentry->d_inode;
Al Viro4023bfc2014-09-13 21:59:43 -04002081 if (!(flags & LOOKUP_RCU))
Al Viro368ee9b2015-05-08 17:19:59 -04002082 return s;
Al Viro4023bfc2014-09-13 21:59:43 -04002083 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
Al Viro368ee9b2015-05-08 17:19:59 -04002084 return s;
Al Viro4023bfc2014-09-13 21:59:43 -04002085 if (!(nd->flags & LOOKUP_ROOT))
2086 nd->root.mnt = NULL;
2087 rcu_read_unlock();
Al Viro368ee9b2015-05-08 17:19:59 -04002088 return ERR_PTR(-ECHILD);
Al Viro9b4a9b12009-04-07 11:44:16 -04002089}
2090
Al Viro3bdba282015-05-08 17:37:07 -04002091static const char *trailing_symlink(struct nameidata *nd)
Al Viro95fa25d2015-04-22 13:46:57 -04002092{
2093 const char *s;
Al Virofec2fa22015-05-06 15:58:18 -04002094 int error = may_follow_link(nd);
Al Virodeb106c2015-05-08 18:05:21 -04002095 if (unlikely(error))
Al Viro3bdba282015-05-08 17:37:07 -04002096 return ERR_PTR(error);
Al Viro95fa25d2015-04-22 13:46:57 -04002097 nd->flags |= LOOKUP_PARENT;
Al Virofab51e82015-05-10 11:01:00 -04002098 nd->stack[0].name = NULL;
Al Viro3b2e7f72015-04-19 00:53:50 -04002099 s = get_link(nd);
Al Virodeb106c2015-05-08 18:05:21 -04002100 return s ? s : "";
Al Viro95fa25d2015-04-22 13:46:57 -04002101}
2102
Al Virocaa85632015-04-22 17:52:47 -04002103static inline int lookup_last(struct nameidata *nd)
Al Virobd92d7f2011-03-14 19:54:59 -04002104{
2105 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2106 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2107
2108 nd->flags &= ~LOOKUP_PARENT;
Al Virodeb106c2015-05-08 18:05:21 -04002109 return walk_component(nd,
Al Viro4693a542015-05-04 17:47:11 -04002110 nd->flags & LOOKUP_FOLLOW
2111 ? nd->depth
2112 ? WALK_PUT | WALK_GET
2113 : WALK_GET
2114 : 0);
Al Virobd92d7f2011-03-14 19:54:59 -04002115}
2116
Al Viro9b4a9b12009-04-07 11:44:16 -04002117/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroc8a53ee2015-05-12 18:43:07 -04002118static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
Al Viro9b4a9b12009-04-07 11:44:16 -04002119{
Al Viroc8a53ee2015-05-12 18:43:07 -04002120 const char *s = path_init(nd, flags);
Al Virobd92d7f2011-03-14 19:54:59 -04002121 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11002122
Al Viro368ee9b2015-05-08 17:19:59 -04002123 if (IS_ERR(s))
2124 return PTR_ERR(s);
Al Viro3bdba282015-05-08 17:37:07 -04002125 while (!(err = link_path_walk(s, nd))
2126 && ((err = lookup_last(nd)) > 0)) {
2127 s = trailing_symlink(nd);
2128 if (IS_ERR(s)) {
2129 err = PTR_ERR(s);
2130 break;
Al Virobd92d7f2011-03-14 19:54:59 -04002131 }
2132 }
Al Viro9f1fafe2011-03-25 11:00:12 -04002133 if (!err)
2134 err = complete_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04002135
Al Virodeb106c2015-05-08 18:05:21 -04002136 if (!err && nd->flags & LOOKUP_DIRECTORY)
2137 if (!d_can_lookup(nd->path.dentry))
Al Virobd23a532011-03-23 09:56:30 -04002138 err = -ENOTDIR;
Al Viro625b6d12015-05-12 16:36:12 -04002139 if (!err) {
2140 *path = nd->path;
2141 nd->path.mnt = NULL;
2142 nd->path.dentry = NULL;
2143 }
2144 terminate_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04002145 return err;
Al Viroee0827c2011-02-21 23:38:09 -05002146}
Nick Piggin31e6b012011-01-07 17:49:52 +11002147
Al Viro625b6d12015-05-12 16:36:12 -04002148static int filename_lookup(int dfd, struct filename *name, unsigned flags,
Al Viro9ad1aaa2015-05-12 16:44:39 -04002149 struct path *path, struct path *root)
Jeff Layton873f1ee2012-10-10 15:25:29 -04002150{
Al Viro894bc8c2015-05-02 07:16:16 -04002151 int retval;
Al Viro9883d182015-05-13 07:28:08 -04002152 struct nameidata nd;
Al Viroabc9f5b2015-05-12 16:53:42 -04002153 if (IS_ERR(name))
2154 return PTR_ERR(name);
Al Viro9ad1aaa2015-05-12 16:44:39 -04002155 if (unlikely(root)) {
2156 nd.root = *root;
2157 flags |= LOOKUP_ROOT;
2158 }
Al Viro9883d182015-05-13 07:28:08 -04002159 set_nameidata(&nd, dfd, name);
Al Viroc8a53ee2015-05-12 18:43:07 -04002160 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002161 if (unlikely(retval == -ECHILD))
Al Viroc8a53ee2015-05-12 18:43:07 -04002162 retval = path_lookupat(&nd, flags, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002163 if (unlikely(retval == -ESTALE))
Al Viroc8a53ee2015-05-12 18:43:07 -04002164 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002165
2166 if (likely(!retval))
Al Viro625b6d12015-05-12 16:36:12 -04002167 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
Al Viro9883d182015-05-13 07:28:08 -04002168 restore_nameidata();
Al Viroe4bd1c12015-05-12 16:40:39 -04002169 putname(name);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002170 return retval;
2171}
2172
Al Viro8bcb77f2015-05-08 16:59:20 -04002173/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroc8a53ee2015-05-12 18:43:07 -04002174static int path_parentat(struct nameidata *nd, unsigned flags,
Al Viro391172c2015-05-09 11:19:16 -04002175 struct path *parent)
Al Viro8bcb77f2015-05-08 16:59:20 -04002176{
Al Viroc8a53ee2015-05-12 18:43:07 -04002177 const char *s = path_init(nd, flags);
Al Viro368ee9b2015-05-08 17:19:59 -04002178 int err;
2179 if (IS_ERR(s))
2180 return PTR_ERR(s);
2181 err = link_path_walk(s, nd);
Al Viro8bcb77f2015-05-08 16:59:20 -04002182 if (!err)
2183 err = complete_walk(nd);
Al Viro391172c2015-05-09 11:19:16 -04002184 if (!err) {
2185 *parent = nd->path;
2186 nd->path.mnt = NULL;
2187 nd->path.dentry = NULL;
2188 }
2189 terminate_walk(nd);
Al Viro8bcb77f2015-05-08 16:59:20 -04002190 return err;
2191}
2192
Al Viro5c31b6c2015-05-12 17:32:54 -04002193static struct filename *filename_parentat(int dfd, struct filename *name,
Al Viro391172c2015-05-09 11:19:16 -04002194 unsigned int flags, struct path *parent,
2195 struct qstr *last, int *type)
Al Viro8bcb77f2015-05-08 16:59:20 -04002196{
2197 int retval;
Al Viro9883d182015-05-13 07:28:08 -04002198 struct nameidata nd;
Al Viro8bcb77f2015-05-08 16:59:20 -04002199
Al Viro5c31b6c2015-05-12 17:32:54 -04002200 if (IS_ERR(name))
2201 return name;
Al Viro9883d182015-05-13 07:28:08 -04002202 set_nameidata(&nd, dfd, name);
Al Viroc8a53ee2015-05-12 18:43:07 -04002203 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
Al Viro8bcb77f2015-05-08 16:59:20 -04002204 if (unlikely(retval == -ECHILD))
Al Viroc8a53ee2015-05-12 18:43:07 -04002205 retval = path_parentat(&nd, flags, parent);
Al Viro8bcb77f2015-05-08 16:59:20 -04002206 if (unlikely(retval == -ESTALE))
Al Viroc8a53ee2015-05-12 18:43:07 -04002207 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
Al Viro391172c2015-05-09 11:19:16 -04002208 if (likely(!retval)) {
2209 *last = nd.last;
2210 *type = nd.last_type;
2211 audit_inode(name, parent->dentry, LOOKUP_PARENT);
Al Viro5c31b6c2015-05-12 17:32:54 -04002212 } else {
2213 putname(name);
2214 name = ERR_PTR(retval);
Al Viro391172c2015-05-09 11:19:16 -04002215 }
Al Viro9883d182015-05-13 07:28:08 -04002216 restore_nameidata();
Al Viro5c31b6c2015-05-12 17:32:54 -04002217 return name;
Al Viro8bcb77f2015-05-08 16:59:20 -04002218}
2219
Al Viro79714f72012-06-15 03:01:42 +04002220/* does lookup, returns the object with parent locked */
2221struct dentry *kern_path_locked(const char *name, struct path *path)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002222{
Al Viro5c31b6c2015-05-12 17:32:54 -04002223 struct filename *filename;
2224 struct dentry *d;
Al Viro391172c2015-05-09 11:19:16 -04002225 struct qstr last;
2226 int type;
Paul Moore51689102015-01-22 00:00:03 -05002227
Al Viro5c31b6c2015-05-12 17:32:54 -04002228 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2229 &last, &type);
Paul Moore51689102015-01-22 00:00:03 -05002230 if (IS_ERR(filename))
2231 return ERR_CAST(filename);
Al Viro5c31b6c2015-05-12 17:32:54 -04002232 if (unlikely(type != LAST_NORM)) {
Al Viro391172c2015-05-09 11:19:16 -04002233 path_put(path);
Al Viro5c31b6c2015-05-12 17:32:54 -04002234 putname(filename);
2235 return ERR_PTR(-EINVAL);
Al Viro79714f72012-06-15 03:01:42 +04002236 }
Al Viro391172c2015-05-09 11:19:16 -04002237 mutex_lock_nested(&path->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2238 d = __lookup_hash(&last, path->dentry, 0);
Al Viro79714f72012-06-15 03:01:42 +04002239 if (IS_ERR(d)) {
Al Viro391172c2015-05-09 11:19:16 -04002240 mutex_unlock(&path->dentry->d_inode->i_mutex);
2241 path_put(path);
Al Viro79714f72012-06-15 03:01:42 +04002242 }
Paul Moore51689102015-01-22 00:00:03 -05002243 putname(filename);
Al Viro79714f72012-06-15 03:01:42 +04002244 return d;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002245}
2246
Al Virod1811462008-08-02 00:49:18 -04002247int kern_path(const char *name, unsigned int flags, struct path *path)
2248{
Al Viroabc9f5b2015-05-12 16:53:42 -04002249 return filename_lookup(AT_FDCWD, getname_kernel(name),
2250 flags, path, NULL);
Al Virod1811462008-08-02 00:49:18 -04002251}
Al Viro4d359502014-03-14 12:20:17 -04002252EXPORT_SYMBOL(kern_path);
Al Virod1811462008-08-02 00:49:18 -04002253
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002254/**
2255 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2256 * @dentry: pointer to dentry of the base directory
2257 * @mnt: pointer to vfs mount of the base directory
2258 * @name: pointer to file name
2259 * @flags: lookup flags
Al Viroe0a01242011-06-27 17:00:37 -04002260 * @path: pointer to struct path to fill
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002261 */
2262int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2263 const char *name, unsigned int flags,
Al Viroe0a01242011-06-27 17:00:37 -04002264 struct path *path)
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002265{
Al Viro9ad1aaa2015-05-12 16:44:39 -04002266 struct path root = {.mnt = mnt, .dentry = dentry};
Al Viro9ad1aaa2015-05-12 16:44:39 -04002267 /* the first argument of filename_lookup() is ignored with root */
Al Viroabc9f5b2015-05-12 16:53:42 -04002268 return filename_lookup(AT_FDCWD, getname_kernel(name),
2269 flags , path, &root);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002270}
Al Viro4d359502014-03-14 12:20:17 -04002271EXPORT_SYMBOL(vfs_path_lookup);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002272
Christoph Hellwigeead1912007-10-16 23:25:38 -07002273/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07002274 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07002275 * @name: pathname component to lookup
2276 * @base: base directory to lookup from
2277 * @len: maximum length @len should be interpreted to
2278 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07002279 * Note that this routine is purely a helper for filesystem usage and should
Al Viro9e7543e2015-02-23 02:49:48 -05002280 * not be called by generic code.
Christoph Hellwigeead1912007-10-16 23:25:38 -07002281 */
James Morris057f6c02007-04-26 00:12:05 -07002282struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2283{
James Morris057f6c02007-04-26 00:12:05 -07002284 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05002285 unsigned int c;
Miklos Szeredicda309d2012-03-26 12:54:21 +02002286 int err;
James Morris057f6c02007-04-26 00:12:05 -07002287
David Woodhouse2f9092e2009-04-20 23:18:37 +01002288 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2289
Al Viro6a96ba52011-03-07 23:49:20 -05002290 this.name = name;
2291 this.len = len;
Linus Torvalds0145acc2012-03-02 14:32:59 -08002292 this.hash = full_name_hash(name, len);
Al Viro6a96ba52011-03-07 23:49:20 -05002293 if (!len)
2294 return ERR_PTR(-EACCES);
2295
Al Viro21d8a152012-11-29 22:17:21 -05002296 if (unlikely(name[0] == '.')) {
2297 if (len < 2 || (len == 2 && name[1] == '.'))
2298 return ERR_PTR(-EACCES);
2299 }
2300
Al Viro6a96ba52011-03-07 23:49:20 -05002301 while (len--) {
2302 c = *(const unsigned char *)name++;
2303 if (c == '/' || c == '\0')
2304 return ERR_PTR(-EACCES);
Al Viro6a96ba52011-03-07 23:49:20 -05002305 }
Al Viro5a202bc2011-03-08 14:17:44 -05002306 /*
2307 * See if the low-level filesystem might want
2308 * to use its own hash..
2309 */
2310 if (base->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002311 int err = base->d_op->d_hash(base, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05002312 if (err < 0)
2313 return ERR_PTR(err);
2314 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07002315
Miklos Szeredicda309d2012-03-26 12:54:21 +02002316 err = inode_permission(base->d_inode, MAY_EXEC);
2317 if (err)
2318 return ERR_PTR(err);
2319
Al Viro72bd8662012-06-10 17:17:17 -04002320 return __lookup_hash(&this, base, 0);
James Morris057f6c02007-04-26 00:12:05 -07002321}
Al Viro4d359502014-03-14 12:20:17 -04002322EXPORT_SYMBOL(lookup_one_len);
James Morris057f6c02007-04-26 00:12:05 -07002323
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002324int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2325 struct path *path, int *empty)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326{
Al Viroabc9f5b2015-05-12 16:53:42 -04002327 return filename_lookup(dfd, getname_flags(name, flags, empty),
2328 flags, path, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329}
Al Virob853a162015-05-13 09:12:02 -04002330EXPORT_SYMBOL(user_path_at_empty);
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002331
Jeff Layton873f1ee2012-10-10 15:25:29 -04002332/*
2333 * NB: most callers don't do anything directly with the reference to the
2334 * to struct filename, but the nd->last pointer points into the name string
2335 * allocated by getname. So we must hold the reference to it until all
2336 * path-walking is complete.
2337 */
Al Viroa2ec4a22015-05-13 06:57:49 -04002338static inline struct filename *
Al Virof5beed72015-04-30 16:09:11 -04002339user_path_parent(int dfd, const char __user *path,
2340 struct path *parent,
2341 struct qstr *last,
2342 int *type,
Jeff Layton9e790bd2012-12-11 12:10:09 -05002343 unsigned int flags)
Al Viro2ad94ae2008-07-21 09:32:51 -04002344{
Jeff Layton9e790bd2012-12-11 12:10:09 -05002345 /* only LOOKUP_REVAL is allowed in extra flags */
Al Viro5c31b6c2015-05-12 17:32:54 -04002346 return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL,
2347 parent, last, type);
Al Viro2ad94ae2008-07-21 09:32:51 -04002348}
2349
Jeff Layton80334262013-07-26 06:23:25 -04002350/**
Al Viro197df042013-09-08 14:03:27 -04002351 * mountpoint_last - look up last component for umount
Jeff Layton80334262013-07-26 06:23:25 -04002352 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2353 * @path: pointer to container for result
2354 *
2355 * This is a special lookup_last function just for umount. In this case, we
2356 * need to resolve the path without doing any revalidation.
2357 *
2358 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2359 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2360 * in almost all cases, this lookup will be served out of the dcache. The only
2361 * cases where it won't are if nd->last refers to a symlink or the path is
2362 * bogus and it doesn't exist.
2363 *
2364 * Returns:
2365 * -error: if there was an error during lookup. This includes -ENOENT if the
2366 * lookup found a negative dentry. The nd->path reference will also be
2367 * put in this case.
2368 *
2369 * 0: if we successfully resolved nd->path and found it to not to be a
2370 * symlink that needs to be followed. "path" will also be populated.
2371 * The nd->path reference will also be put.
2372 *
2373 * 1: if we successfully resolved nd->last and found it to be a symlink
2374 * that needs to be followed. "path" will be populated with the path
2375 * to the link, and nd->path will *not* be put.
2376 */
2377static int
Al Viro197df042013-09-08 14:03:27 -04002378mountpoint_last(struct nameidata *nd, struct path *path)
Jeff Layton80334262013-07-26 06:23:25 -04002379{
2380 int error = 0;
2381 struct dentry *dentry;
2382 struct dentry *dir = nd->path.dentry;
2383
Al Viro35759522013-09-08 13:41:33 -04002384 /* If we're in rcuwalk, drop out of it to handle last component */
2385 if (nd->flags & LOOKUP_RCU) {
Al Viro6e9918b2015-05-05 09:26:05 -04002386 if (unlazy_walk(nd, NULL, 0))
Al Virodeb106c2015-05-08 18:05:21 -04002387 return -ECHILD;
Jeff Layton80334262013-07-26 06:23:25 -04002388 }
2389
2390 nd->flags &= ~LOOKUP_PARENT;
2391
2392 if (unlikely(nd->last_type != LAST_NORM)) {
2393 error = handle_dots(nd, nd->last_type);
Al Viro35759522013-09-08 13:41:33 -04002394 if (error)
Al Virodeb106c2015-05-08 18:05:21 -04002395 return error;
Al Viro35759522013-09-08 13:41:33 -04002396 dentry = dget(nd->path.dentry);
2397 goto done;
Jeff Layton80334262013-07-26 06:23:25 -04002398 }
2399
2400 mutex_lock(&dir->d_inode->i_mutex);
2401 dentry = d_lookup(dir, &nd->last);
2402 if (!dentry) {
2403 /*
2404 * No cached dentry. Mounted dentries are pinned in the cache,
2405 * so that means that this dentry is probably a symlink or the
2406 * path doesn't actually point to a mounted dentry.
2407 */
2408 dentry = d_alloc(dir, &nd->last);
2409 if (!dentry) {
Dave Jonesbcceeeb2013-09-10 17:04:25 -04002410 mutex_unlock(&dir->d_inode->i_mutex);
Al Virodeb106c2015-05-08 18:05:21 -04002411 return -ENOMEM;
Jeff Layton80334262013-07-26 06:23:25 -04002412 }
Al Viro35759522013-09-08 13:41:33 -04002413 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
Dave Jonesbcceeeb2013-09-10 17:04:25 -04002414 if (IS_ERR(dentry)) {
2415 mutex_unlock(&dir->d_inode->i_mutex);
Al Virodeb106c2015-05-08 18:05:21 -04002416 return PTR_ERR(dentry);
Dave Jonesbcceeeb2013-09-10 17:04:25 -04002417 }
Jeff Layton80334262013-07-26 06:23:25 -04002418 }
2419 mutex_unlock(&dir->d_inode->i_mutex);
2420
Al Viro35759522013-09-08 13:41:33 -04002421done:
David Howells698934d2015-03-17 17:33:52 +00002422 if (d_is_negative(dentry)) {
Al Viro35759522013-09-08 13:41:33 -04002423 dput(dentry);
Al Virodeb106c2015-05-08 18:05:21 -04002424 return -ENOENT;
Jeff Layton80334262013-07-26 06:23:25 -04002425 }
Al Viro191d7f72015-05-04 08:26:45 -04002426 if (nd->depth)
2427 put_link(nd);
Al Viro35759522013-09-08 13:41:33 -04002428 path->dentry = dentry;
Vasily Averin295dc392014-07-21 12:30:23 +04002429 path->mnt = nd->path.mnt;
Al Viro181548c2015-05-07 19:54:34 -04002430 error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW,
2431 d_backing_inode(dentry), 0);
Al Virodeb106c2015-05-08 18:05:21 -04002432 if (unlikely(error))
Al Virod63ff282015-05-04 18:13:23 -04002433 return error;
Vasily Averin295dc392014-07-21 12:30:23 +04002434 mntget(path->mnt);
Al Viro35759522013-09-08 13:41:33 -04002435 follow_mount(path);
Al Virodeb106c2015-05-08 18:05:21 -04002436 return 0;
Jeff Layton80334262013-07-26 06:23:25 -04002437}
2438
2439/**
Al Viro197df042013-09-08 14:03:27 -04002440 * path_mountpoint - look up a path to be umounted
Masanari Iida2a78b852015-09-09 15:39:23 -07002441 * @nd: lookup context
Jeff Layton80334262013-07-26 06:23:25 -04002442 * @flags: lookup flags
Al Viroc8a53ee2015-05-12 18:43:07 -04002443 * @path: pointer to container for result
Jeff Layton80334262013-07-26 06:23:25 -04002444 *
2445 * Look up the given name, but don't attempt to revalidate the last component.
Randy Dunlap606d6fe2013-10-19 14:56:55 -07002446 * Returns 0 and "path" will be valid on success; Returns error otherwise.
Jeff Layton80334262013-07-26 06:23:25 -04002447 */
2448static int
Al Viroc8a53ee2015-05-12 18:43:07 -04002449path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
Jeff Layton80334262013-07-26 06:23:25 -04002450{
Al Viroc8a53ee2015-05-12 18:43:07 -04002451 const char *s = path_init(nd, flags);
Al Viro368ee9b2015-05-08 17:19:59 -04002452 int err;
2453 if (IS_ERR(s))
2454 return PTR_ERR(s);
Al Viro3bdba282015-05-08 17:37:07 -04002455 while (!(err = link_path_walk(s, nd)) &&
2456 (err = mountpoint_last(nd, path)) > 0) {
2457 s = trailing_symlink(nd);
2458 if (IS_ERR(s)) {
2459 err = PTR_ERR(s);
Jeff Layton80334262013-07-26 06:23:25 -04002460 break;
Al Viro3bdba282015-05-08 17:37:07 -04002461 }
Jeff Layton80334262013-07-26 06:23:25 -04002462 }
Al Virodeb106c2015-05-08 18:05:21 -04002463 terminate_walk(nd);
Jeff Layton80334262013-07-26 06:23:25 -04002464 return err;
2465}
2466
Al Viro2d864652013-09-08 20:18:44 -04002467static int
Al Viro668696d2015-02-22 19:44:00 -05002468filename_mountpoint(int dfd, struct filename *name, struct path *path,
Al Viro2d864652013-09-08 20:18:44 -04002469 unsigned int flags)
2470{
Al Viro9883d182015-05-13 07:28:08 -04002471 struct nameidata nd;
Al Virocbaab2d2015-01-22 02:49:00 -05002472 int error;
Al Viro668696d2015-02-22 19:44:00 -05002473 if (IS_ERR(name))
2474 return PTR_ERR(name);
Al Viro9883d182015-05-13 07:28:08 -04002475 set_nameidata(&nd, dfd, name);
Al Viroc8a53ee2015-05-12 18:43:07 -04002476 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
Al Viro2d864652013-09-08 20:18:44 -04002477 if (unlikely(error == -ECHILD))
Al Viroc8a53ee2015-05-12 18:43:07 -04002478 error = path_mountpoint(&nd, flags, path);
Al Viro2d864652013-09-08 20:18:44 -04002479 if (unlikely(error == -ESTALE))
Al Viroc8a53ee2015-05-12 18:43:07 -04002480 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
Al Viro2d864652013-09-08 20:18:44 -04002481 if (likely(!error))
Al Viro668696d2015-02-22 19:44:00 -05002482 audit_inode(name, path->dentry, 0);
Al Viro9883d182015-05-13 07:28:08 -04002483 restore_nameidata();
Al Viro668696d2015-02-22 19:44:00 -05002484 putname(name);
Al Viro2d864652013-09-08 20:18:44 -04002485 return error;
2486}
2487
Jeff Layton80334262013-07-26 06:23:25 -04002488/**
Al Viro197df042013-09-08 14:03:27 -04002489 * user_path_mountpoint_at - lookup a path from userland in order to umount it
Jeff Layton80334262013-07-26 06:23:25 -04002490 * @dfd: directory file descriptor
2491 * @name: pathname from userland
2492 * @flags: lookup flags
2493 * @path: pointer to container to hold result
2494 *
2495 * A umount is a special case for path walking. We're not actually interested
2496 * in the inode in this situation, and ESTALE errors can be a problem. We
2497 * simply want track down the dentry and vfsmount attached at the mountpoint
2498 * and avoid revalidating the last component.
2499 *
2500 * Returns 0 and populates "path" on success.
2501 */
2502int
Al Viro197df042013-09-08 14:03:27 -04002503user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
Jeff Layton80334262013-07-26 06:23:25 -04002504 struct path *path)
2505{
Al Virocbaab2d2015-01-22 02:49:00 -05002506 return filename_mountpoint(dfd, getname(name), path, flags);
Jeff Layton80334262013-07-26 06:23:25 -04002507}
2508
Al Viro2d864652013-09-08 20:18:44 -04002509int
2510kern_path_mountpoint(int dfd, const char *name, struct path *path,
2511 unsigned int flags)
2512{
Al Virocbaab2d2015-01-22 02:49:00 -05002513 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
Al Viro2d864652013-09-08 20:18:44 -04002514}
2515EXPORT_SYMBOL(kern_path_mountpoint);
2516
Miklos Szeredicbdf35b2014-10-24 00:14:36 +02002517int __check_sticky(struct inode *dir, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518{
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002519 kuid_t fsuid = current_fsuid();
David Howellsda9592e2008-11-14 10:39:05 +11002520
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002521 if (uid_eq(inode->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522 return 0;
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002523 if (uid_eq(dir->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 return 0;
Andy Lutomirski23adbe12014-06-10 12:45:42 -07002525 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526}
Miklos Szeredicbdf35b2014-10-24 00:14:36 +02002527EXPORT_SYMBOL(__check_sticky);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528
2529/*
2530 * Check whether we can remove a link victim from directory dir, check
2531 * whether the type of victim is right.
2532 * 1. We can't do it if dir is read-only (done in permission())
2533 * 2. We should have write and exec permissions on dir
2534 * 3. We can't remove anything from append-only dir
2535 * 4. We can't do anything with immutable dir (done in permission())
2536 * 5. If the sticky bit on dir is set we should either
2537 * a. be owner of dir, or
2538 * b. be owner of victim, or
2539 * c. have CAP_FOWNER capability
2540 * 6. If the victim is append-only or immutable we can't do antyhing with
2541 * links pointing to it.
2542 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2543 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2544 * 9. We can't remove a root or mountpoint.
2545 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2546 * nfs_async_unlink().
2547 */
David Howellsb18825a2013-09-12 19:22:53 +01002548static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549{
David Howells63afdfc2015-05-06 15:59:00 +01002550 struct inode *inode = d_backing_inode(victim);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 int error;
2552
David Howellsb18825a2013-09-12 19:22:53 +01002553 if (d_is_negative(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554 return -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01002555 BUG_ON(!inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556
2557 BUG_ON(victim->d_parent->d_inode != dir);
Jeff Layton4fa6b5e2012-10-10 15:25:25 -04002558 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559
Al Virof419a2e2008-07-22 00:07:17 -04002560 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561 if (error)
2562 return error;
2563 if (IS_APPEND(dir))
2564 return -EPERM;
David Howellsb18825a2013-09-12 19:22:53 +01002565
2566 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2567 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568 return -EPERM;
2569 if (isdir) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002570 if (!d_is_dir(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571 return -ENOTDIR;
2572 if (IS_ROOT(victim))
2573 return -EBUSY;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002574 } else if (d_is_dir(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575 return -EISDIR;
2576 if (IS_DEADDIR(dir))
2577 return -ENOENT;
2578 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2579 return -EBUSY;
2580 return 0;
2581}
2582
2583/* Check whether we can create an object with dentry child in directory
2584 * dir.
2585 * 1. We can't do it if child already exists (open has special treatment for
2586 * this case, but since we are inlined it's OK)
2587 * 2. We can't do it if dir is read-only (done in permission())
2588 * 3. We should have write and exec permissions on dir
2589 * 4. We can't do it if dir is immutable (done in permission())
2590 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02002591static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592{
Jeff Layton14e972b2013-05-08 10:25:58 -04002593 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594 if (child->d_inode)
2595 return -EEXIST;
2596 if (IS_DEADDIR(dir))
2597 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04002598 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599}
2600
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601/*
2602 * p1 and p2 should be directories on the same fs.
2603 */
2604struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2605{
2606 struct dentry *p;
2607
2608 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07002609 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610 return NULL;
2611 }
2612
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002613 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002614
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002615 p = d_ancestor(p2, p1);
2616 if (p) {
2617 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2618 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2619 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620 }
2621
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002622 p = d_ancestor(p1, p2);
2623 if (p) {
2624 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2625 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2626 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 }
2628
Ingo Molnarf2eace22006-07-03 00:25:05 -07002629 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Miklos Szeredid1b72cc2014-10-27 15:42:01 +01002630 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631 return NULL;
2632}
Al Viro4d359502014-03-14 12:20:17 -04002633EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634
2635void unlock_rename(struct dentry *p1, struct dentry *p2)
2636{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002637 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002639 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002640 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641 }
2642}
Al Viro4d359502014-03-14 12:20:17 -04002643EXPORT_SYMBOL(unlock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644
Al Viro4acdaf22011-07-26 01:42:34 -04002645int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viro312b63f2012-06-10 18:09:36 -04002646 bool want_excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002648 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649 if (error)
2650 return error;
2651
Al Viroacfa4382008-12-04 10:06:33 -05002652 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002653 return -EACCES; /* shouldn't it be ENOSYS? */
2654 mode &= S_IALLUGO;
2655 mode |= S_IFREG;
2656 error = security_inode_create(dir, dentry, mode);
2657 if (error)
2658 return error;
Al Viro312b63f2012-06-10 18:09:36 -04002659 error = dir->i_op->create(dir, dentry, mode, want_excl);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002660 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002661 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662 return error;
2663}
Al Viro4d359502014-03-14 12:20:17 -04002664EXPORT_SYMBOL(vfs_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665
Al Viro73d049a2011-03-11 12:08:24 -05002666static int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002668 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669 struct inode *inode = dentry->d_inode;
2670 int error;
2671
Al Virobcda7652011-03-13 16:42:14 -04002672 /* O_PATH? */
2673 if (!acc_mode)
2674 return 0;
2675
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676 if (!inode)
2677 return -ENOENT;
2678
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002679 switch (inode->i_mode & S_IFMT) {
2680 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002682 case S_IFDIR:
2683 if (acc_mode & MAY_WRITE)
2684 return -EISDIR;
2685 break;
2686 case S_IFBLK:
2687 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002688 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002689 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002690 /*FALLTHRU*/
2691 case S_IFIFO:
2692 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002694 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002695 }
Dave Hansenb41572e2007-10-16 23:31:14 -07002696
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002697 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07002698 if (error)
2699 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05002700
Linus Torvalds1da177e2005-04-16 15:20:36 -07002701 /*
2702 * An append-only file must be opened in append mode for writing.
2703 */
2704 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05002705 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05002706 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05002708 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709 }
2710
2711 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002712 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05002713 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714
J. Bruce Fieldsf3c7691e2011-09-21 10:58:13 -04002715 return 0;
Al Viro7715b522009-12-16 03:54:00 -05002716}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002717
Jeff Laytone1181ee2010-12-07 16:19:50 -05002718static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002719{
Jeff Laytone1181ee2010-12-07 16:19:50 -05002720 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002721 struct inode *inode = path->dentry->d_inode;
2722 int error = get_write_access(inode);
2723 if (error)
2724 return error;
2725 /*
2726 * Refuse to truncate files with mandatory locks held on them.
2727 */
Jeff Laytond7a06982014-03-10 09:54:15 -04002728 error = locks_verify_locked(filp);
Al Viro7715b522009-12-16 03:54:00 -05002729 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002730 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002731 if (!error) {
2732 error = do_truncate(path->dentry, 0,
2733 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002734 filp);
Al Viro7715b522009-12-16 03:54:00 -05002735 }
2736 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002737 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002738}
2739
Dave Hansend57999e2008-02-15 14:37:27 -08002740static inline int open_to_namei_flags(int flag)
2741{
Al Viro8a5e9292011-06-25 19:15:54 -04002742 if ((flag & O_ACCMODE) == 3)
2743 flag--;
Dave Hansend57999e2008-02-15 14:37:27 -08002744 return flag;
2745}
2746
Miklos Szeredid18e9002012-06-05 15:10:17 +02002747static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2748{
2749 int error = security_path_mknod(dir, dentry, mode, 0);
2750 if (error)
2751 return error;
2752
2753 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2754 if (error)
2755 return error;
2756
2757 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2758}
2759
David Howells1acf0af2012-06-14 16:13:46 +01002760/*
2761 * Attempt to atomically look up, create and open a file from a negative
2762 * dentry.
2763 *
2764 * Returns 0 if successful. The file will have been created and attached to
2765 * @file by the filesystem calling finish_open().
2766 *
2767 * Returns 1 if the file was looked up only or didn't need creating. The
2768 * caller will need to perform the open themselves. @path will have been
2769 * updated to point to the new dentry. This may be negative.
2770 *
2771 * Returns an error code otherwise.
2772 */
Al Viro2675a4e2012-06-22 12:41:10 +04002773static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2774 struct path *path, struct file *file,
2775 const struct open_flags *op,
Al Viro64894cf2012-07-31 00:53:35 +04002776 bool got_write, bool need_lookup,
Al Viro2675a4e2012-06-22 12:41:10 +04002777 int *opened)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002778{
2779 struct inode *dir = nd->path.dentry->d_inode;
2780 unsigned open_flag = open_to_namei_flags(op->open_flag);
2781 umode_t mode;
2782 int error;
2783 int acc_mode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002784 int create_error = 0;
2785 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
Miklos Szeredi116cc022013-09-16 14:52:05 +02002786 bool excl;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002787
2788 BUG_ON(dentry->d_inode);
2789
2790 /* Don't create child dentry for a dead directory. */
2791 if (unlikely(IS_DEADDIR(dir))) {
Al Viro2675a4e2012-06-22 12:41:10 +04002792 error = -ENOENT;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002793 goto out;
2794 }
2795
Miklos Szeredi62b259d2012-08-15 13:01:24 +02002796 mode = op->mode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002797 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2798 mode &= ~current_umask();
2799
Miklos Szeredi116cc022013-09-16 14:52:05 +02002800 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2801 if (excl)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002802 open_flag &= ~O_TRUNC;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002803
2804 /*
2805 * Checking write permission is tricky, bacuse we don't know if we are
2806 * going to actually need it: O_CREAT opens should work as long as the
2807 * file exists. But checking existence breaks atomicity. The trick is
2808 * to check access and if not granted clear O_CREAT from the flags.
2809 *
2810 * Another problem is returing the "right" error value (e.g. for an
2811 * O_EXCL open we want to return EEXIST not EROFS).
2812 */
Al Viro64894cf2012-07-31 00:53:35 +04002813 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2814 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2815 if (!(open_flag & O_CREAT)) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002816 /*
2817 * No O_CREATE -> atomicity not a requirement -> fall
2818 * back to lookup + open
2819 */
2820 goto no_open;
2821 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2822 /* Fall back and fail with the right error */
Al Viro64894cf2012-07-31 00:53:35 +04002823 create_error = -EROFS;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002824 goto no_open;
2825 } else {
2826 /* No side effects, safe to clear O_CREAT */
Al Viro64894cf2012-07-31 00:53:35 +04002827 create_error = -EROFS;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002828 open_flag &= ~O_CREAT;
2829 }
2830 }
2831
2832 if (open_flag & O_CREAT) {
Miklos Szeredi38227f72012-08-15 13:01:24 +02002833 error = may_o_create(&nd->path, dentry, mode);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002834 if (error) {
2835 create_error = error;
2836 if (open_flag & O_EXCL)
2837 goto no_open;
2838 open_flag &= ~O_CREAT;
2839 }
2840 }
2841
2842 if (nd->flags & LOOKUP_DIRECTORY)
2843 open_flag |= O_DIRECTORY;
2844
Al Viro30d90492012-06-22 12:40:19 +04002845 file->f_path.dentry = DENTRY_NOT_SET;
2846 file->f_path.mnt = nd->path.mnt;
2847 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
Al Viro47237682012-06-10 05:01:45 -04002848 opened);
Al Virod9585272012-06-22 12:39:14 +04002849 if (error < 0) {
Al Virod9585272012-06-22 12:39:14 +04002850 if (create_error && error == -ENOENT)
2851 error = create_error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002852 goto out;
2853 }
2854
Al Virod9585272012-06-22 12:39:14 +04002855 if (error) { /* returned 1, that is */
Al Viro30d90492012-06-22 12:40:19 +04002856 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
Al Viro2675a4e2012-06-22 12:41:10 +04002857 error = -EIO;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002858 goto out;
2859 }
Al Viro30d90492012-06-22 12:40:19 +04002860 if (file->f_path.dentry) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002861 dput(dentry);
Al Viro30d90492012-06-22 12:40:19 +04002862 dentry = file->f_path.dentry;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002863 }
Al Viro03da6332013-09-16 19:22:33 -04002864 if (*opened & FILE_CREATED)
2865 fsnotify_create(dir, dentry);
2866 if (!dentry->d_inode) {
2867 WARN_ON(*opened & FILE_CREATED);
2868 if (create_error) {
2869 error = create_error;
2870 goto out;
2871 }
2872 } else {
2873 if (excl && !(*opened & FILE_CREATED)) {
2874 error = -EEXIST;
2875 goto out;
2876 }
Sage Weil62b2ce92012-08-15 13:30:12 -07002877 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02002878 goto looked_up;
2879 }
2880
2881 /*
2882 * We didn't have the inode before the open, so check open permission
2883 * here.
2884 */
Al Viro03da6332013-09-16 19:22:33 -04002885 acc_mode = op->acc_mode;
2886 if (*opened & FILE_CREATED) {
2887 WARN_ON(!(open_flag & O_CREAT));
2888 fsnotify_create(dir, dentry);
2889 acc_mode = MAY_OPEN;
2890 }
Al Viro2675a4e2012-06-22 12:41:10 +04002891 error = may_open(&file->f_path, acc_mode, open_flag);
2892 if (error)
2893 fput(file);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002894
2895out:
2896 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04002897 return error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002898
Miklos Szeredid18e9002012-06-05 15:10:17 +02002899no_open:
2900 if (need_lookup) {
Al Viro72bd8662012-06-10 17:17:17 -04002901 dentry = lookup_real(dir, dentry, nd->flags);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002902 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002903 return PTR_ERR(dentry);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002904
2905 if (create_error) {
2906 int open_flag = op->open_flag;
2907
Al Viro2675a4e2012-06-22 12:41:10 +04002908 error = create_error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002909 if ((open_flag & O_EXCL)) {
2910 if (!dentry->d_inode)
2911 goto out;
2912 } else if (!dentry->d_inode) {
2913 goto out;
2914 } else if ((open_flag & O_TRUNC) &&
David Howellse36cb0b2015-01-29 12:02:35 +00002915 d_is_reg(dentry)) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002916 goto out;
2917 }
2918 /* will fail later, go on to get the right error */
2919 }
2920 }
2921looked_up:
2922 path->dentry = dentry;
2923 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04002924 return 1;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002925}
2926
Nick Piggin31e6b012011-01-07 17:49:52 +11002927/*
David Howells1acf0af2012-06-14 16:13:46 +01002928 * Look up and maybe create and open the last component.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002929 *
2930 * Must be called with i_mutex held on parent.
2931 *
David Howells1acf0af2012-06-14 16:13:46 +01002932 * Returns 0 if the file was successfully atomically created (if necessary) and
2933 * opened. In this case the file will be returned attached to @file.
2934 *
2935 * Returns 1 if the file was not completely opened at this time, though lookups
2936 * and creations will have been performed and the dentry returned in @path will
2937 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2938 * specified then a negative dentry may be returned.
2939 *
2940 * An error code is returned otherwise.
2941 *
2942 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2943 * cleared otherwise prior to returning.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002944 */
Al Viro2675a4e2012-06-22 12:41:10 +04002945static int lookup_open(struct nameidata *nd, struct path *path,
2946 struct file *file,
2947 const struct open_flags *op,
Al Viro64894cf2012-07-31 00:53:35 +04002948 bool got_write, int *opened)
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002949{
2950 struct dentry *dir = nd->path.dentry;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002951 struct inode *dir_inode = dir->d_inode;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002952 struct dentry *dentry;
2953 int error;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002954 bool need_lookup;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002955
Al Viro47237682012-06-10 05:01:45 -04002956 *opened &= ~FILE_CREATED;
Al Viro201f9562012-06-22 12:42:10 +04002957 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002958 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002959 return PTR_ERR(dentry);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002960
Miklos Szeredid18e9002012-06-05 15:10:17 +02002961 /* Cached positive dentry: will open in f_op->open */
2962 if (!need_lookup && dentry->d_inode)
2963 goto out_no_open;
2964
2965 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
Al Viro64894cf2012-07-31 00:53:35 +04002966 return atomic_open(nd, dentry, path, file, op, got_write,
Al Viro47237682012-06-10 05:01:45 -04002967 need_lookup, opened);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002968 }
2969
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002970 if (need_lookup) {
2971 BUG_ON(dentry->d_inode);
2972
Al Viro72bd8662012-06-10 17:17:17 -04002973 dentry = lookup_real(dir_inode, dentry, nd->flags);
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002974 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002975 return PTR_ERR(dentry);
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002976 }
2977
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002978 /* Negative dentry, just create the file */
2979 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2980 umode_t mode = op->mode;
2981 if (!IS_POSIXACL(dir->d_inode))
2982 mode &= ~current_umask();
2983 /*
2984 * This write is needed to ensure that a
2985 * rw->ro transition does not occur between
2986 * the time when the file is created and when
2987 * a permanent write count is taken through
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002988 * the 'struct file' in finish_open().
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002989 */
Al Viro64894cf2012-07-31 00:53:35 +04002990 if (!got_write) {
2991 error = -EROFS;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002992 goto out_dput;
Al Viro64894cf2012-07-31 00:53:35 +04002993 }
Al Viro47237682012-06-10 05:01:45 -04002994 *opened |= FILE_CREATED;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002995 error = security_path_mknod(&nd->path, dentry, mode, 0);
2996 if (error)
2997 goto out_dput;
Al Viro312b63f2012-06-10 18:09:36 -04002998 error = vfs_create(dir->d_inode, dentry, mode,
2999 nd->flags & LOOKUP_EXCL);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003000 if (error)
3001 goto out_dput;
3002 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02003003out_no_open:
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003004 path->dentry = dentry;
3005 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04003006 return 1;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003007
3008out_dput:
3009 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04003010 return error;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003011}
3012
3013/*
Al Virofe2d35f2011-03-05 22:58:25 -05003014 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11003015 */
Al Viro896475d2015-04-22 18:02:17 -04003016static int do_last(struct nameidata *nd,
Al Viro2675a4e2012-06-22 12:41:10 +04003017 struct file *file, const struct open_flags *op,
Al Viro76ae2a52015-05-12 18:44:32 -04003018 int *opened)
Al Virofb1cc552009-12-24 01:58:28 -05003019{
Al Viroa1e28032009-12-24 02:12:06 -05003020 struct dentry *dir = nd->path.dentry;
Al Viroca344a892011-03-09 00:36:45 -05003021 int open_flag = op->open_flag;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003022 bool will_truncate = (open_flag & O_TRUNC) != 0;
Al Viro64894cf2012-07-31 00:53:35 +04003023 bool got_write = false;
Al Virobcda7652011-03-13 16:42:14 -04003024 int acc_mode = op->acc_mode;
Al Viro254cf582015-05-05 09:40:46 -04003025 unsigned seq;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02003026 struct inode *inode;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003027 struct path save_parent = { .dentry = NULL, .mnt = NULL };
Al Viro896475d2015-04-22 18:02:17 -04003028 struct path path;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003029 bool retried = false;
Al Viro16c2cd72011-02-22 15:50:10 -05003030 int error;
Al Virofb1cc552009-12-24 01:58:28 -05003031
Al Viroc3e380b2011-02-23 13:39:45 -05003032 nd->flags &= ~LOOKUP_PARENT;
3033 nd->flags |= op->intent;
3034
Al Virobc77daa2013-06-06 09:12:33 -04003035 if (nd->last_type != LAST_NORM) {
Al Virofe2d35f2011-03-05 22:58:25 -05003036 error = handle_dots(nd, nd->last_type);
Al Virodeb106c2015-05-08 18:05:21 -04003037 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04003038 return error;
Miklos Szeredie83db162012-06-05 15:10:29 +02003039 goto finish_open;
Al Viro1f36f772009-12-26 10:56:19 -05003040 }
Al Viro67ee3ad2009-12-26 07:01:01 -05003041
Al Viroca344a892011-03-09 00:36:45 -05003042 if (!(open_flag & O_CREAT)) {
Al Virofe2d35f2011-03-05 22:58:25 -05003043 if (nd->last.name[nd->last.len])
3044 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3045 /* we _can_ be in RCU mode here */
Al Viro254cf582015-05-05 09:40:46 -04003046 error = lookup_fast(nd, &path, &inode, &seq);
Miklos Szeredi71574862012-06-05 15:10:14 +02003047 if (likely(!error))
3048 goto finish_lookup;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02003049
Miklos Szeredi71574862012-06-05 15:10:14 +02003050 if (error < 0)
Al Virodeb106c2015-05-08 18:05:21 -04003051 return error;
Miklos Szeredi37d7fff2012-06-05 15:10:12 +02003052
Miklos Szeredi71574862012-06-05 15:10:14 +02003053 BUG_ON(nd->inode != dir->d_inode);
Miklos Szeredib6183df2012-06-05 15:10:13 +02003054 } else {
3055 /* create side of things */
3056 /*
3057 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3058 * has been cleared when we got to the last component we are
3059 * about to look up
3060 */
3061 error = complete_walk(nd);
Al Viroe8bb73d2015-05-08 16:28:42 -04003062 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003063 return error;
Miklos Szeredib6183df2012-06-05 15:10:13 +02003064
Al Viro76ae2a52015-05-12 18:44:32 -04003065 audit_inode(nd->name, dir, LOOKUP_PARENT);
Miklos Szeredib6183df2012-06-05 15:10:13 +02003066 /* trailing slashes? */
Al Virodeb106c2015-05-08 18:05:21 -04003067 if (unlikely(nd->last.name[nd->last.len]))
3068 return -EISDIR;
Al Virofe2d35f2011-03-05 22:58:25 -05003069 }
3070
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003071retry_lookup:
Al Viro64894cf2012-07-31 00:53:35 +04003072 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3073 error = mnt_want_write(nd->path.mnt);
3074 if (!error)
3075 got_write = true;
3076 /*
3077 * do _not_ fail yet - we might not need that or fail with
3078 * a different error; let lookup_open() decide; we'll be
3079 * dropping this one anyway.
3080 */
3081 }
Al Viroa1e28032009-12-24 02:12:06 -05003082 mutex_lock(&dir->d_inode->i_mutex);
Al Viro896475d2015-04-22 18:02:17 -04003083 error = lookup_open(nd, &path, file, op, got_write, opened);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003084 mutex_unlock(&dir->d_inode->i_mutex);
Al Viroa1e28032009-12-24 02:12:06 -05003085
Al Viro2675a4e2012-06-22 12:41:10 +04003086 if (error <= 0) {
3087 if (error)
Miklos Szeredid18e9002012-06-05 15:10:17 +02003088 goto out;
3089
Al Viro47237682012-06-10 05:01:45 -04003090 if ((*opened & FILE_CREATED) ||
Al Viro496ad9a2013-01-23 17:07:38 -05003091 !S_ISREG(file_inode(file)->i_mode))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003092 will_truncate = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003093
Al Viro76ae2a52015-05-12 18:44:32 -04003094 audit_inode(nd->name, file->f_path.dentry, 0);
Miklos Szeredid18e9002012-06-05 15:10:17 +02003095 goto opened;
3096 }
Al Virofb1cc552009-12-24 01:58:28 -05003097
Al Viro47237682012-06-10 05:01:45 -04003098 if (*opened & FILE_CREATED) {
Al Viro9b44f1b2011-03-09 00:17:27 -05003099 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05003100 open_flag &= ~O_TRUNC;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003101 will_truncate = false;
Al Virobcda7652011-03-13 16:42:14 -04003102 acc_mode = MAY_OPEN;
Al Viro896475d2015-04-22 18:02:17 -04003103 path_to_nameidata(&path, nd);
Miklos Szeredie83db162012-06-05 15:10:29 +02003104 goto finish_open_created;
Al Virofb1cc552009-12-24 01:58:28 -05003105 }
3106
3107 /*
Jeff Layton3134f372012-07-25 10:19:47 -04003108 * create/update audit record if it already exists.
Al Virofb1cc552009-12-24 01:58:28 -05003109 */
Al Viro896475d2015-04-22 18:02:17 -04003110 if (d_is_positive(path.dentry))
Al Viro76ae2a52015-05-12 18:44:32 -04003111 audit_inode(nd->name, path.dentry, 0);
Al Virofb1cc552009-12-24 01:58:28 -05003112
Miklos Szeredid18e9002012-06-05 15:10:17 +02003113 /*
3114 * If atomic_open() acquired write access it is dropped now due to
3115 * possible mount and symlink following (this might be optimized away if
3116 * necessary...)
3117 */
Al Viro64894cf2012-07-31 00:53:35 +04003118 if (got_write) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02003119 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04003120 got_write = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003121 }
3122
Al Virodeb106c2015-05-08 18:05:21 -04003123 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3124 path_to_nameidata(&path, nd);
3125 return -EEXIST;
3126 }
Al Virofb1cc552009-12-24 01:58:28 -05003127
NeilBrown756daf22015-03-23 13:37:38 +11003128 error = follow_managed(&path, nd);
Al Virodeb106c2015-05-08 18:05:21 -04003129 if (unlikely(error < 0))
3130 return error;
Al Viroa3fbbde2011-11-07 21:21:26 +00003131
Miklos Szeredidecf3402012-05-21 17:30:08 +02003132 BUG_ON(nd->flags & LOOKUP_RCU);
David Howells63afdfc2015-05-06 15:59:00 +01003133 inode = d_backing_inode(path.dentry);
Al Viro254cf582015-05-05 09:40:46 -04003134 seq = 0; /* out of RCU mode, so the value doesn't matter */
Al Virodeb106c2015-05-08 18:05:21 -04003135 if (unlikely(d_is_negative(path.dentry))) {
Al Viro896475d2015-04-22 18:02:17 -04003136 path_to_nameidata(&path, nd);
Al Virodeb106c2015-05-08 18:05:21 -04003137 return -ENOENT;
Miklos Szeredi54c33e72012-05-21 17:30:10 +02003138 }
Al Viro766c4cb2015-05-07 19:24:57 -04003139finish_lookup:
Al Virod63ff282015-05-04 18:13:23 -04003140 if (nd->depth)
3141 put_link(nd);
Al Viro181548c2015-05-07 19:54:34 -04003142 error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW,
3143 inode, seq);
Al Virodeb106c2015-05-08 18:05:21 -04003144 if (unlikely(error))
Al Virod63ff282015-05-04 18:13:23 -04003145 return error;
Al Virofb1cc552009-12-24 01:58:28 -05003146
Al Viro896475d2015-04-22 18:02:17 -04003147 if (unlikely(d_is_symlink(path.dentry)) && !(open_flag & O_PATH)) {
3148 path_to_nameidata(&path, nd);
Al Virodeb106c2015-05-08 18:05:21 -04003149 return -ELOOP;
Al Viroa5cfe2d2015-04-22 12:10:45 -04003150 }
3151
Al Viro896475d2015-04-22 18:02:17 -04003152 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) {
3153 path_to_nameidata(&path, nd);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003154 } else {
3155 save_parent.dentry = nd->path.dentry;
Al Viro896475d2015-04-22 18:02:17 -04003156 save_parent.mnt = mntget(path.mnt);
3157 nd->path.dentry = path.dentry;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003158
3159 }
Miklos Szeredidecf3402012-05-21 17:30:08 +02003160 nd->inode = inode;
Al Viro254cf582015-05-05 09:40:46 -04003161 nd->seq = seq;
Al Viroa3fbbde2011-11-07 21:21:26 +00003162 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
Al Virobc77daa2013-06-06 09:12:33 -04003163finish_open:
Al Viroa3fbbde2011-11-07 21:21:26 +00003164 error = complete_walk(nd);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003165 if (error) {
3166 path_put(&save_parent);
Al Viro2675a4e2012-06-22 12:41:10 +04003167 return error;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003168 }
Al Viro76ae2a52015-05-12 18:44:32 -04003169 audit_inode(nd->name, nd->path.dentry, 0);
Al Virofb1cc552009-12-24 01:58:28 -05003170 error = -EISDIR;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003171 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04003172 goto out;
Miklos Szerediaf2f5542012-05-21 17:30:11 +02003173 error = -ENOTDIR;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003174 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04003175 goto out;
David Howells4bbcbd3b2015-03-17 22:16:40 +00003176 if (!d_is_reg(nd->path.dentry))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003177 will_truncate = false;
Al Viro6c0d46c2011-03-09 00:59:59 -05003178
Al Viro0f9d1a12011-03-09 00:13:14 -05003179 if (will_truncate) {
3180 error = mnt_want_write(nd->path.mnt);
3181 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003182 goto out;
Al Viro64894cf2012-07-31 00:53:35 +04003183 got_write = true;
Al Viro0f9d1a12011-03-09 00:13:14 -05003184 }
Miklos Szeredie83db162012-06-05 15:10:29 +02003185finish_open_created:
Al Virobcda7652011-03-13 16:42:14 -04003186 error = may_open(&nd->path, acc_mode, open_flag);
Al Viroca344a892011-03-09 00:36:45 -05003187 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003188 goto out;
Miklos Szeredi4aa7c632014-10-24 00:14:35 +02003189
3190 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3191 error = vfs_open(&nd->path, file, current_cred());
3192 if (!error) {
3193 *opened |= FILE_OPENED;
3194 } else {
Al Viro30d90492012-06-22 12:40:19 +04003195 if (error == -EOPENSTALE)
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003196 goto stale_open;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003197 goto out;
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003198 }
Miklos Szeredia8277b92012-06-05 15:10:32 +02003199opened:
Al Viro2675a4e2012-06-22 12:41:10 +04003200 error = open_check_o_direct(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003201 if (error)
3202 goto exit_fput;
Dmitry Kasatkin3034a142014-06-27 18:15:44 +03003203 error = ima_file_check(file, op->acc_mode, *opened);
Miklos Szerediaa4caad2012-06-05 15:10:28 +02003204 if (error)
3205 goto exit_fput;
3206
3207 if (will_truncate) {
Al Viro2675a4e2012-06-22 12:41:10 +04003208 error = handle_truncate(file);
Miklos Szerediaa4caad2012-06-05 15:10:28 +02003209 if (error)
3210 goto exit_fput;
Al Viro0f9d1a12011-03-09 00:13:14 -05003211 }
Al Viroca344a892011-03-09 00:36:45 -05003212out:
Al Viro64894cf2012-07-31 00:53:35 +04003213 if (got_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05003214 mnt_drop_write(nd->path.mnt);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003215 path_put(&save_parent);
Al Viro2675a4e2012-06-22 12:41:10 +04003216 return error;
Al Virofb1cc552009-12-24 01:58:28 -05003217
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003218exit_fput:
Al Viro2675a4e2012-06-22 12:41:10 +04003219 fput(file);
3220 goto out;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003221
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003222stale_open:
3223 /* If no saved parent or already retried then can't retry */
3224 if (!save_parent.dentry || retried)
3225 goto out;
3226
3227 BUG_ON(save_parent.dentry != dir);
3228 path_put(&nd->path);
3229 nd->path = save_parent;
3230 nd->inode = dir->d_inode;
3231 save_parent.mnt = NULL;
3232 save_parent.dentry = NULL;
Al Viro64894cf2012-07-31 00:53:35 +04003233 if (got_write) {
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003234 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04003235 got_write = false;
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003236 }
3237 retried = true;
3238 goto retry_lookup;
Al Virofb1cc552009-12-24 01:58:28 -05003239}
3240
Al Viroc8a53ee2015-05-12 18:43:07 -04003241static int do_tmpfile(struct nameidata *nd, unsigned flags,
Al Viro60545d02013-06-07 01:20:27 -04003242 const struct open_flags *op,
3243 struct file *file, int *opened)
3244{
3245 static const struct qstr name = QSTR_INIT("/", 1);
Al Viro625b6d12015-05-12 16:36:12 -04003246 struct dentry *child;
Al Viro60545d02013-06-07 01:20:27 -04003247 struct inode *dir;
Al Viro625b6d12015-05-12 16:36:12 -04003248 struct path path;
Al Viroc8a53ee2015-05-12 18:43:07 -04003249 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
Al Viro60545d02013-06-07 01:20:27 -04003250 if (unlikely(error))
3251 return error;
Al Viro625b6d12015-05-12 16:36:12 -04003252 error = mnt_want_write(path.mnt);
Al Viro60545d02013-06-07 01:20:27 -04003253 if (unlikely(error))
3254 goto out;
Al Viro625b6d12015-05-12 16:36:12 -04003255 dir = path.dentry->d_inode;
Al Viro60545d02013-06-07 01:20:27 -04003256 /* we want directory to be writable */
Al Viro625b6d12015-05-12 16:36:12 -04003257 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Al Viro60545d02013-06-07 01:20:27 -04003258 if (error)
3259 goto out2;
Al Viro60545d02013-06-07 01:20:27 -04003260 if (!dir->i_op->tmpfile) {
3261 error = -EOPNOTSUPP;
3262 goto out2;
3263 }
Al Viro625b6d12015-05-12 16:36:12 -04003264 child = d_alloc(path.dentry, &name);
Al Viro60545d02013-06-07 01:20:27 -04003265 if (unlikely(!child)) {
3266 error = -ENOMEM;
3267 goto out2;
3268 }
Al Viro625b6d12015-05-12 16:36:12 -04003269 dput(path.dentry);
3270 path.dentry = child;
3271 error = dir->i_op->tmpfile(dir, child, op->mode);
Al Viro60545d02013-06-07 01:20:27 -04003272 if (error)
3273 goto out2;
Al Viroc8a53ee2015-05-12 18:43:07 -04003274 audit_inode(nd->name, child, 0);
Eric Rannaud69a91c22014-10-30 01:51:01 -07003275 /* Don't check for other permissions, the inode was just created */
Al Viro625b6d12015-05-12 16:36:12 -04003276 error = may_open(&path, MAY_OPEN, op->open_flag);
Al Viro60545d02013-06-07 01:20:27 -04003277 if (error)
3278 goto out2;
Al Viro625b6d12015-05-12 16:36:12 -04003279 file->f_path.mnt = path.mnt;
3280 error = finish_open(file, child, NULL, opened);
Al Viro60545d02013-06-07 01:20:27 -04003281 if (error)
3282 goto out2;
3283 error = open_check_o_direct(file);
Al Virof4e0c302013-06-11 08:34:36 +04003284 if (error) {
Al Viro60545d02013-06-07 01:20:27 -04003285 fput(file);
Al Virof4e0c302013-06-11 08:34:36 +04003286 } else if (!(op->open_flag & O_EXCL)) {
3287 struct inode *inode = file_inode(file);
3288 spin_lock(&inode->i_lock);
3289 inode->i_state |= I_LINKABLE;
3290 spin_unlock(&inode->i_lock);
3291 }
Al Viro60545d02013-06-07 01:20:27 -04003292out2:
Al Viro625b6d12015-05-12 16:36:12 -04003293 mnt_drop_write(path.mnt);
Al Viro60545d02013-06-07 01:20:27 -04003294out:
Al Viro625b6d12015-05-12 16:36:12 -04003295 path_put(&path);
Al Viro60545d02013-06-07 01:20:27 -04003296 return error;
3297}
3298
Al Viroc8a53ee2015-05-12 18:43:07 -04003299static struct file *path_openat(struct nameidata *nd,
3300 const struct open_flags *op, unsigned flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003301{
Al Viro368ee9b2015-05-08 17:19:59 -04003302 const char *s;
Al Viro30d90492012-06-22 12:40:19 +04003303 struct file *file;
Al Viro47237682012-06-10 05:01:45 -04003304 int opened = 0;
Al Viro13aab422011-02-23 17:54:08 -05003305 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11003306
Al Viro30d90492012-06-22 12:40:19 +04003307 file = get_empty_filp();
Al Viro1afc99b2013-02-14 20:41:04 -05003308 if (IS_ERR(file))
3309 return file;
Nick Piggin31e6b012011-01-07 17:49:52 +11003310
Al Viro30d90492012-06-22 12:40:19 +04003311 file->f_flags = op->open_flag;
Nick Piggin31e6b012011-01-07 17:49:52 +11003312
Al Virobb458c62013-07-13 13:26:37 +04003313 if (unlikely(file->f_flags & __O_TMPFILE)) {
Al Viroc8a53ee2015-05-12 18:43:07 -04003314 error = do_tmpfile(nd, flags, op, file, &opened);
Al Virof15133d2015-05-08 22:53:15 -04003315 goto out2;
Al Viro60545d02013-06-07 01:20:27 -04003316 }
3317
Al Viroc8a53ee2015-05-12 18:43:07 -04003318 s = path_init(nd, flags);
Al Viro368ee9b2015-05-08 17:19:59 -04003319 if (IS_ERR(s)) {
3320 put_filp(file);
3321 return ERR_CAST(s);
3322 }
Al Viro3bdba282015-05-08 17:37:07 -04003323 while (!(error = link_path_walk(s, nd)) &&
Al Viro76ae2a52015-05-12 18:44:32 -04003324 (error = do_last(nd, file, op, &opened)) > 0) {
Al Viro73d049a2011-03-11 12:08:24 -05003325 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
Al Viro3bdba282015-05-08 17:37:07 -04003326 s = trailing_symlink(nd);
3327 if (IS_ERR(s)) {
3328 error = PTR_ERR(s);
Al Viro2675a4e2012-06-22 12:41:10 +04003329 break;
Al Viro3bdba282015-05-08 17:37:07 -04003330 }
Al Viro806b6812009-12-26 07:16:40 -05003331 }
Al Virodeb106c2015-05-08 18:05:21 -04003332 terminate_walk(nd);
Al Virof15133d2015-05-08 22:53:15 -04003333out2:
Al Viro2675a4e2012-06-22 12:41:10 +04003334 if (!(opened & FILE_OPENED)) {
3335 BUG_ON(!error);
Al Viro30d90492012-06-22 12:40:19 +04003336 put_filp(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003337 }
Al Viro2675a4e2012-06-22 12:41:10 +04003338 if (unlikely(error)) {
3339 if (error == -EOPENSTALE) {
3340 if (flags & LOOKUP_RCU)
3341 error = -ECHILD;
3342 else
3343 error = -ESTALE;
3344 }
3345 file = ERR_PTR(error);
3346 }
3347 return file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003348}
3349
Jeff Layton669abf42012-10-10 16:43:10 -04003350struct file *do_filp_open(int dfd, struct filename *pathname,
Al Virof9652e12013-06-11 08:23:01 +04003351 const struct open_flags *op)
Al Viro13aab422011-02-23 17:54:08 -05003352{
Al Viro9883d182015-05-13 07:28:08 -04003353 struct nameidata nd;
Al Virof9652e12013-06-11 08:23:01 +04003354 int flags = op->lookup_flags;
Al Viro13aab422011-02-23 17:54:08 -05003355 struct file *filp;
3356
Al Viro9883d182015-05-13 07:28:08 -04003357 set_nameidata(&nd, dfd, pathname);
Al Viroc8a53ee2015-05-12 18:43:07 -04003358 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
Al Viro13aab422011-02-23 17:54:08 -05003359 if (unlikely(filp == ERR_PTR(-ECHILD)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003360 filp = path_openat(&nd, op, flags);
Al Viro13aab422011-02-23 17:54:08 -05003361 if (unlikely(filp == ERR_PTR(-ESTALE)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003362 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
Al Viro9883d182015-05-13 07:28:08 -04003363 restore_nameidata();
Al Viro13aab422011-02-23 17:54:08 -05003364 return filp;
3365}
3366
Al Viro73d049a2011-03-11 12:08:24 -05003367struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
Al Virof9652e12013-06-11 08:23:01 +04003368 const char *name, const struct open_flags *op)
Al Viro73d049a2011-03-11 12:08:24 -05003369{
Al Viro9883d182015-05-13 07:28:08 -04003370 struct nameidata nd;
Al Viro73d049a2011-03-11 12:08:24 -05003371 struct file *file;
Paul Moore51689102015-01-22 00:00:03 -05003372 struct filename *filename;
Al Virof9652e12013-06-11 08:23:01 +04003373 int flags = op->lookup_flags | LOOKUP_ROOT;
Al Viro73d049a2011-03-11 12:08:24 -05003374
3375 nd.root.mnt = mnt;
3376 nd.root.dentry = dentry;
3377
David Howellsb18825a2013-09-12 19:22:53 +01003378 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
Al Viro73d049a2011-03-11 12:08:24 -05003379 return ERR_PTR(-ELOOP);
3380
Paul Moore51689102015-01-22 00:00:03 -05003381 filename = getname_kernel(name);
Viresh Kumara1c83682015-08-12 15:59:44 +05303382 if (IS_ERR(filename))
Paul Moore51689102015-01-22 00:00:03 -05003383 return ERR_CAST(filename);
3384
Al Viro9883d182015-05-13 07:28:08 -04003385 set_nameidata(&nd, -1, filename);
Al Viroc8a53ee2015-05-12 18:43:07 -04003386 file = path_openat(&nd, op, flags | LOOKUP_RCU);
Al Viro73d049a2011-03-11 12:08:24 -05003387 if (unlikely(file == ERR_PTR(-ECHILD)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003388 file = path_openat(&nd, op, flags);
Al Viro73d049a2011-03-11 12:08:24 -05003389 if (unlikely(file == ERR_PTR(-ESTALE)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003390 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
Al Viro9883d182015-05-13 07:28:08 -04003391 restore_nameidata();
Paul Moore51689102015-01-22 00:00:03 -05003392 putname(filename);
Al Viro73d049a2011-03-11 12:08:24 -05003393 return file;
3394}
3395
Al Virofa14a0b2015-01-22 02:16:49 -05003396static struct dentry *filename_create(int dfd, struct filename *name,
Jeff Layton1ac12b42012-12-11 12:10:06 -05003397 struct path *path, unsigned int lookup_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003398{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003399 struct dentry *dentry = ERR_PTR(-EEXIST);
Al Viro391172c2015-05-09 11:19:16 -04003400 struct qstr last;
3401 int type;
Jan Karac30dabf2012-06-12 16:20:30 +02003402 int err2;
Jeff Layton1ac12b42012-12-11 12:10:06 -05003403 int error;
3404 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3405
3406 /*
3407 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3408 * other flags passed in are ignored!
3409 */
3410 lookup_flags &= LOOKUP_REVAL;
3411
Al Viro5c31b6c2015-05-12 17:32:54 -04003412 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3413 if (IS_ERR(name))
3414 return ERR_CAST(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003415
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003416 /*
3417 * Yucky last component or no last component at all?
3418 * (foo/., foo/.., /////)
3419 */
Al Viro5c31b6c2015-05-12 17:32:54 -04003420 if (unlikely(type != LAST_NORM))
Al Viroed75e952011-06-27 16:53:43 -04003421 goto out;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003422
Jan Karac30dabf2012-06-12 16:20:30 +02003423 /* don't fail immediately if it's r/o, at least try to report other errors */
Al Viro391172c2015-05-09 11:19:16 -04003424 err2 = mnt_want_write(path->mnt);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003425 /*
3426 * Do the final lookup.
3427 */
Al Viro391172c2015-05-09 11:19:16 -04003428 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3429 mutex_lock_nested(&path->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3430 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003431 if (IS_ERR(dentry))
Al Viroa8104a92012-07-20 02:25:00 +04003432 goto unlock;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003433
Al Viroa8104a92012-07-20 02:25:00 +04003434 error = -EEXIST;
David Howellsb18825a2013-09-12 19:22:53 +01003435 if (d_is_positive(dentry))
Al Viroa8104a92012-07-20 02:25:00 +04003436 goto fail;
David Howellsb18825a2013-09-12 19:22:53 +01003437
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003438 /*
3439 * Special case - lookup gave negative, but... we had foo/bar/
3440 * From the vfs_mknod() POV we just have a negative dentry -
3441 * all is fine. Let's be bastards - you had / on the end, you've
3442 * been asking for (non-existent) directory. -ENOENT for you.
3443 */
Al Viro391172c2015-05-09 11:19:16 -04003444 if (unlikely(!is_dir && last.name[last.len])) {
Al Viroa8104a92012-07-20 02:25:00 +04003445 error = -ENOENT;
Al Viroed75e952011-06-27 16:53:43 -04003446 goto fail;
Al Viroe9baf6e2008-05-15 04:49:12 -04003447 }
Jan Karac30dabf2012-06-12 16:20:30 +02003448 if (unlikely(err2)) {
3449 error = err2;
Al Viroa8104a92012-07-20 02:25:00 +04003450 goto fail;
Jan Karac30dabf2012-06-12 16:20:30 +02003451 }
Al Viro181c37b2015-05-12 17:21:25 -04003452 putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003453 return dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003454fail:
Al Viroa8104a92012-07-20 02:25:00 +04003455 dput(dentry);
3456 dentry = ERR_PTR(error);
3457unlock:
Al Viro391172c2015-05-09 11:19:16 -04003458 mutex_unlock(&path->dentry->d_inode->i_mutex);
Jan Karac30dabf2012-06-12 16:20:30 +02003459 if (!err2)
Al Viro391172c2015-05-09 11:19:16 -04003460 mnt_drop_write(path->mnt);
Al Viroed75e952011-06-27 16:53:43 -04003461out:
Al Viro391172c2015-05-09 11:19:16 -04003462 path_put(path);
Al Viro181c37b2015-05-12 17:21:25 -04003463 putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003464 return dentry;
3465}
Al Virofa14a0b2015-01-22 02:16:49 -05003466
3467struct dentry *kern_path_create(int dfd, const char *pathname,
3468 struct path *path, unsigned int lookup_flags)
3469{
Al Viro181c37b2015-05-12 17:21:25 -04003470 return filename_create(dfd, getname_kernel(pathname),
3471 path, lookup_flags);
Al Virofa14a0b2015-01-22 02:16:49 -05003472}
Al Virodae6ad82011-06-26 11:50:15 -04003473EXPORT_SYMBOL(kern_path_create);
3474
Al Viro921a1652012-07-20 01:15:31 +04003475void done_path_create(struct path *path, struct dentry *dentry)
3476{
3477 dput(dentry);
3478 mutex_unlock(&path->dentry->d_inode->i_mutex);
Al Viroa8104a92012-07-20 02:25:00 +04003479 mnt_drop_write(path->mnt);
Al Viro921a1652012-07-20 01:15:31 +04003480 path_put(path);
3481}
3482EXPORT_SYMBOL(done_path_create);
3483
Al Viro520ae682015-05-13 07:00:28 -04003484inline struct dentry *user_path_create(int dfd, const char __user *pathname,
Jeff Layton1ac12b42012-12-11 12:10:06 -05003485 struct path *path, unsigned int lookup_flags)
Al Virodae6ad82011-06-26 11:50:15 -04003486{
Al Viro181c37b2015-05-12 17:21:25 -04003487 return filename_create(dfd, getname(pathname), path, lookup_flags);
Al Virodae6ad82011-06-26 11:50:15 -04003488}
3489EXPORT_SYMBOL(user_path_create);
3490
Al Viro1a67aaf2011-07-26 01:52:52 -04003491int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003492{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003493 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003494
3495 if (error)
3496 return error;
3497
Eric W. Biederman975d6b32011-11-13 12:16:43 -08003498 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003499 return -EPERM;
3500
Al Viroacfa4382008-12-04 10:06:33 -05003501 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003502 return -EPERM;
3503
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07003504 error = devcgroup_inode_mknod(mode, dev);
3505 if (error)
3506 return error;
3507
Linus Torvalds1da177e2005-04-16 15:20:36 -07003508 error = security_inode_mknod(dir, dentry, mode, dev);
3509 if (error)
3510 return error;
3511
Linus Torvalds1da177e2005-04-16 15:20:36 -07003512 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003513 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003514 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003515 return error;
3516}
Al Viro4d359502014-03-14 12:20:17 -04003517EXPORT_SYMBOL(vfs_mknod);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003518
Al Virof69aac02011-07-26 04:31:40 -04003519static int may_mknod(umode_t mode)
Dave Hansen463c3192008-02-15 14:37:57 -08003520{
3521 switch (mode & S_IFMT) {
3522 case S_IFREG:
3523 case S_IFCHR:
3524 case S_IFBLK:
3525 case S_IFIFO:
3526 case S_IFSOCK:
3527 case 0: /* zero mode translates to S_IFREG */
3528 return 0;
3529 case S_IFDIR:
3530 return -EPERM;
3531 default:
3532 return -EINVAL;
3533 }
3534}
3535
Al Viro8208a222011-07-25 17:32:17 -04003536SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003537 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003538{
Al Viro2ad94ae2008-07-21 09:32:51 -04003539 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003540 struct path path;
3541 int error;
Jeff Layton972567f2012-12-20 16:00:10 -05003542 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003543
Al Viro8e4bfca2012-07-20 01:17:26 +04003544 error = may_mknod(mode);
3545 if (error)
3546 return error;
Jeff Layton972567f2012-12-20 16:00:10 -05003547retry:
3548 dentry = user_path_create(dfd, filename, &path, lookup_flags);
Al Virodae6ad82011-06-26 11:50:15 -04003549 if (IS_ERR(dentry))
3550 return PTR_ERR(dentry);
Al Viro2ad94ae2008-07-21 09:32:51 -04003551
Al Virodae6ad82011-06-26 11:50:15 -04003552 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003553 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003554 error = security_path_mknod(&path, dentry, mode, dev);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003555 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04003556 goto out;
Dave Hansen463c3192008-02-15 14:37:57 -08003557 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003558 case 0: case S_IFREG:
Al Viro312b63f2012-06-10 18:09:36 -04003559 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003560 break;
3561 case S_IFCHR: case S_IFBLK:
Al Virodae6ad82011-06-26 11:50:15 -04003562 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003563 new_decode_dev(dev));
3564 break;
3565 case S_IFIFO: case S_IFSOCK:
Al Virodae6ad82011-06-26 11:50:15 -04003566 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003567 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003568 }
Al Viroa8104a92012-07-20 02:25:00 +04003569out:
Al Viro921a1652012-07-20 01:15:31 +04003570 done_path_create(&path, dentry);
Jeff Layton972567f2012-12-20 16:00:10 -05003571 if (retry_estale(error, lookup_flags)) {
3572 lookup_flags |= LOOKUP_REVAL;
3573 goto retry;
3574 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003575 return error;
3576}
3577
Al Viro8208a222011-07-25 17:32:17 -04003578SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003579{
3580 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3581}
3582
Al Viro18bb1db2011-07-26 01:41:39 -04003583int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003584{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003585 int error = may_create(dir, dentry);
Al Viro8de52772012-02-06 12:45:27 -05003586 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003587
3588 if (error)
3589 return error;
3590
Al Viroacfa4382008-12-04 10:06:33 -05003591 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003592 return -EPERM;
3593
3594 mode &= (S_IRWXUGO|S_ISVTX);
3595 error = security_inode_mkdir(dir, dentry, mode);
3596 if (error)
3597 return error;
3598
Al Viro8de52772012-02-06 12:45:27 -05003599 if (max_links && dir->i_nlink >= max_links)
3600 return -EMLINK;
3601
Linus Torvalds1da177e2005-04-16 15:20:36 -07003602 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003603 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003604 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003605 return error;
3606}
Al Viro4d359502014-03-14 12:20:17 -04003607EXPORT_SYMBOL(vfs_mkdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003608
Al Viroa218d0f2011-11-21 14:59:34 -05003609SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003610{
Dave Hansen6902d922006-09-30 23:29:01 -07003611 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003612 struct path path;
3613 int error;
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003614 unsigned int lookup_flags = LOOKUP_DIRECTORY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003615
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003616retry:
3617 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07003618 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003619 return PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003620
Al Virodae6ad82011-06-26 11:50:15 -04003621 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003622 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003623 error = security_path_mkdir(&path, dentry, mode);
Al Viroa8104a92012-07-20 02:25:00 +04003624 if (!error)
3625 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
Al Viro921a1652012-07-20 01:15:31 +04003626 done_path_create(&path, dentry);
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003627 if (retry_estale(error, lookup_flags)) {
3628 lookup_flags |= LOOKUP_REVAL;
3629 goto retry;
3630 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003631 return error;
3632}
3633
Al Viroa218d0f2011-11-21 14:59:34 -05003634SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003635{
3636 return sys_mkdirat(AT_FDCWD, pathname, mode);
3637}
3638
Linus Torvalds1da177e2005-04-16 15:20:36 -07003639/*
Sage Weila71905f2011-05-24 13:06:08 -07003640 * The dentry_unhash() helper will try to drop the dentry early: we
J. Bruce Fieldsc0d02592012-02-15 11:48:40 -05003641 * should have a usage count of 1 if we're the only user of this
Sage Weila71905f2011-05-24 13:06:08 -07003642 * dentry, and if that is true (possibly after pruning the dcache),
3643 * then we drop the dentry now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003644 *
3645 * A low-level filesystem can, if it choses, legally
3646 * do a
3647 *
3648 * if (!d_unhashed(dentry))
3649 * return -EBUSY;
3650 *
3651 * if it cannot handle the case of removing a directory
3652 * that is still in use by something else..
3653 */
3654void dentry_unhash(struct dentry *dentry)
3655{
Vasily Averindc168422006-12-06 20:37:07 -08003656 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003657 spin_lock(&dentry->d_lock);
Waiman Long98474232013-08-28 18:24:59 -07003658 if (dentry->d_lockref.count == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003659 __d_drop(dentry);
3660 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003661}
Al Viro4d359502014-03-14 12:20:17 -04003662EXPORT_SYMBOL(dentry_unhash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003663
3664int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3665{
3666 int error = may_delete(dir, dentry, 1);
3667
3668 if (error)
3669 return error;
3670
Al Viroacfa4382008-12-04 10:06:33 -05003671 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003672 return -EPERM;
3673
Al Viro1d2ef592011-09-14 18:55:41 +01003674 dget(dentry);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003675 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003676
Sage Weil912dbc12011-05-24 13:06:11 -07003677 error = -EBUSY;
Eric W. Biederman7af13642013-10-04 19:15:13 -07003678 if (is_local_mountpoint(dentry))
Sage Weil912dbc12011-05-24 13:06:11 -07003679 goto out;
3680
3681 error = security_inode_rmdir(dir, dentry);
3682 if (error)
3683 goto out;
3684
Sage Weil3cebde22011-05-29 21:20:59 -07003685 shrink_dcache_parent(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003686 error = dir->i_op->rmdir(dir, dentry);
3687 if (error)
3688 goto out;
3689
3690 dentry->d_inode->i_flags |= S_DEAD;
3691 dont_mount(dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003692 detach_mounts(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003693
3694out:
3695 mutex_unlock(&dentry->d_inode->i_mutex);
Al Viro1d2ef592011-09-14 18:55:41 +01003696 dput(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003697 if (!error)
3698 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003699 return error;
3700}
Al Viro4d359502014-03-14 12:20:17 -04003701EXPORT_SYMBOL(vfs_rmdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003702
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003703static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003704{
3705 int error = 0;
Jeff Layton91a27b22012-10-10 15:25:28 -04003706 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003707 struct dentry *dentry;
Al Virof5beed72015-04-30 16:09:11 -04003708 struct path path;
3709 struct qstr last;
3710 int type;
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003711 unsigned int lookup_flags = 0;
3712retry:
Al Virof5beed72015-04-30 16:09:11 -04003713 name = user_path_parent(dfd, pathname,
3714 &path, &last, &type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04003715 if (IS_ERR(name))
3716 return PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003717
Al Virof5beed72015-04-30 16:09:11 -04003718 switch (type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003719 case LAST_DOTDOT:
3720 error = -ENOTEMPTY;
3721 goto exit1;
3722 case LAST_DOT:
3723 error = -EINVAL;
3724 goto exit1;
3725 case LAST_ROOT:
3726 error = -EBUSY;
3727 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003728 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003729
Al Virof5beed72015-04-30 16:09:11 -04003730 error = mnt_want_write(path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02003731 if (error)
3732 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003733
Al Virof5beed72015-04-30 16:09:11 -04003734 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3735 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003736 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003737 if (IS_ERR(dentry))
3738 goto exit2;
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003739 if (!dentry->d_inode) {
3740 error = -ENOENT;
3741 goto exit3;
3742 }
Al Virof5beed72015-04-30 16:09:11 -04003743 error = security_path_rmdir(&path, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003744 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003745 goto exit3;
Al Virof5beed72015-04-30 16:09:11 -04003746 error = vfs_rmdir(path.dentry->d_inode, dentry);
Dave Hansen06227532008-02-15 14:37:34 -08003747exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07003748 dput(dentry);
3749exit2:
Al Virof5beed72015-04-30 16:09:11 -04003750 mutex_unlock(&path.dentry->d_inode->i_mutex);
3751 mnt_drop_write(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003752exit1:
Al Virof5beed72015-04-30 16:09:11 -04003753 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003754 putname(name);
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003755 if (retry_estale(error, lookup_flags)) {
3756 lookup_flags |= LOOKUP_REVAL;
3757 goto retry;
3758 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003759 return error;
3760}
3761
Heiko Carstens3cdad422009-01-14 14:14:22 +01003762SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003763{
3764 return do_rmdir(AT_FDCWD, pathname);
3765}
3766
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003767/**
3768 * vfs_unlink - unlink a filesystem object
3769 * @dir: parent directory
3770 * @dentry: victim
3771 * @delegated_inode: returns victim inode, if the inode is delegated.
3772 *
3773 * The caller must hold dir->i_mutex.
3774 *
3775 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3776 * return a reference to the inode in delegated_inode. The caller
3777 * should then break the delegation on that inode and retry. Because
3778 * breaking a delegation may take a long time, the caller should drop
3779 * dir->i_mutex before doing so.
3780 *
3781 * Alternatively, a caller may pass NULL for delegated_inode. This may
3782 * be appropriate for callers that expect the underlying filesystem not
3783 * to be NFS exported.
3784 */
3785int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003786{
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003787 struct inode *target = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003788 int error = may_delete(dir, dentry, 0);
3789
3790 if (error)
3791 return error;
3792
Al Viroacfa4382008-12-04 10:06:33 -05003793 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003794 return -EPERM;
3795
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003796 mutex_lock(&target->i_mutex);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003797 if (is_local_mountpoint(dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003798 error = -EBUSY;
3799 else {
3800 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05003801 if (!error) {
J. Bruce Fields5a146962012-08-28 07:50:40 -07003802 error = try_break_deleg(target, delegated_inode);
3803 if (error)
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003804 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003805 error = dir->i_op->unlink(dir, dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003806 if (!error) {
Al Virod83c49f2010-04-30 17:17:09 -04003807 dont_mount(dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003808 detach_mounts(dentry);
3809 }
Al Virobec10522010-03-03 14:12:08 -05003810 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003811 }
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003812out:
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003813 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003814
3815 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3816 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003817 fsnotify_link_count(target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003818 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003819 }
Robert Love0eeca282005-07-12 17:06:03 -04003820
Linus Torvalds1da177e2005-04-16 15:20:36 -07003821 return error;
3822}
Al Viro4d359502014-03-14 12:20:17 -04003823EXPORT_SYMBOL(vfs_unlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003824
3825/*
3826 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003827 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07003828 * writeout happening, and we don't want to prevent access to the directory
3829 * while waiting on the I/O.
3830 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003831static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003832{
Al Viro2ad94ae2008-07-21 09:32:51 -04003833 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04003834 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003835 struct dentry *dentry;
Al Virof5beed72015-04-30 16:09:11 -04003836 struct path path;
3837 struct qstr last;
3838 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003839 struct inode *inode = NULL;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003840 struct inode *delegated_inode = NULL;
Jeff Layton5d18f812012-12-20 16:38:04 -05003841 unsigned int lookup_flags = 0;
3842retry:
Al Virof5beed72015-04-30 16:09:11 -04003843 name = user_path_parent(dfd, pathname,
3844 &path, &last, &type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04003845 if (IS_ERR(name))
3846 return PTR_ERR(name);
Al Viro2ad94ae2008-07-21 09:32:51 -04003847
Linus Torvalds1da177e2005-04-16 15:20:36 -07003848 error = -EISDIR;
Al Virof5beed72015-04-30 16:09:11 -04003849 if (type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003850 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003851
Al Virof5beed72015-04-30 16:09:11 -04003852 error = mnt_want_write(path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02003853 if (error)
3854 goto exit1;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003855retry_deleg:
Al Virof5beed72015-04-30 16:09:11 -04003856 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3857 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003858 error = PTR_ERR(dentry);
3859 if (!IS_ERR(dentry)) {
3860 /* Why not before? Because we want correct error value */
Al Virof5beed72015-04-30 16:09:11 -04003861 if (last.name[last.len])
Török Edwin50338b82011-06-16 00:06:14 +03003862 goto slashes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003863 inode = dentry->d_inode;
David Howellsb18825a2013-09-12 19:22:53 +01003864 if (d_is_negative(dentry))
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003865 goto slashes;
3866 ihold(inode);
Al Virof5beed72015-04-30 16:09:11 -04003867 error = security_path_unlink(&path, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003868 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003869 goto exit2;
Al Virof5beed72015-04-30 16:09:11 -04003870 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
Jan Karac30dabf2012-06-12 16:20:30 +02003871exit2:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003872 dput(dentry);
3873 }
Al Virof5beed72015-04-30 16:09:11 -04003874 mutex_unlock(&path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003875 if (inode)
3876 iput(inode); /* truncate the inode here */
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003877 inode = NULL;
3878 if (delegated_inode) {
J. Bruce Fields5a146962012-08-28 07:50:40 -07003879 error = break_deleg_wait(&delegated_inode);
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003880 if (!error)
3881 goto retry_deleg;
3882 }
Al Virof5beed72015-04-30 16:09:11 -04003883 mnt_drop_write(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003884exit1:
Al Virof5beed72015-04-30 16:09:11 -04003885 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003886 putname(name);
Jeff Layton5d18f812012-12-20 16:38:04 -05003887 if (retry_estale(error, lookup_flags)) {
3888 lookup_flags |= LOOKUP_REVAL;
3889 inode = NULL;
3890 goto retry;
3891 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003892 return error;
3893
3894slashes:
David Howellsb18825a2013-09-12 19:22:53 +01003895 if (d_is_negative(dentry))
3896 error = -ENOENT;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003897 else if (d_is_dir(dentry))
David Howellsb18825a2013-09-12 19:22:53 +01003898 error = -EISDIR;
3899 else
3900 error = -ENOTDIR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003901 goto exit2;
3902}
3903
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003904SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003905{
3906 if ((flag & ~AT_REMOVEDIR) != 0)
3907 return -EINVAL;
3908
3909 if (flag & AT_REMOVEDIR)
3910 return do_rmdir(dfd, pathname);
3911
3912 return do_unlinkat(dfd, pathname);
3913}
3914
Heiko Carstens3480b252009-01-14 14:14:16 +01003915SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003916{
3917 return do_unlinkat(AT_FDCWD, pathname);
3918}
3919
Miklos Szeredidb2e7472008-06-24 16:50:16 +02003920int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003921{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003922 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003923
3924 if (error)
3925 return error;
3926
Al Viroacfa4382008-12-04 10:06:33 -05003927 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003928 return -EPERM;
3929
3930 error = security_inode_symlink(dir, dentry, oldname);
3931 if (error)
3932 return error;
3933
Linus Torvalds1da177e2005-04-16 15:20:36 -07003934 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003935 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003936 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003937 return error;
3938}
Al Viro4d359502014-03-14 12:20:17 -04003939EXPORT_SYMBOL(vfs_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003940
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003941SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3942 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003943{
Al Viro2ad94ae2008-07-21 09:32:51 -04003944 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04003945 struct filename *from;
Dave Hansen6902d922006-09-30 23:29:01 -07003946 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003947 struct path path;
Jeff Laytonf46d3562012-12-11 12:10:08 -05003948 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003949
3950 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04003951 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003952 return PTR_ERR(from);
Jeff Laytonf46d3562012-12-11 12:10:08 -05003953retry:
3954 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07003955 error = PTR_ERR(dentry);
3956 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003957 goto out_putname;
Dave Hansen6902d922006-09-30 23:29:01 -07003958
Jeff Layton91a27b22012-10-10 15:25:28 -04003959 error = security_path_symlink(&path, dentry, from->name);
Al Viroa8104a92012-07-20 02:25:00 +04003960 if (!error)
Jeff Layton91a27b22012-10-10 15:25:28 -04003961 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
Al Viro921a1652012-07-20 01:15:31 +04003962 done_path_create(&path, dentry);
Jeff Laytonf46d3562012-12-11 12:10:08 -05003963 if (retry_estale(error, lookup_flags)) {
3964 lookup_flags |= LOOKUP_REVAL;
3965 goto retry;
3966 }
Dave Hansen6902d922006-09-30 23:29:01 -07003967out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003968 putname(from);
3969 return error;
3970}
3971
Heiko Carstens3480b252009-01-14 14:14:16 +01003972SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003973{
3974 return sys_symlinkat(oldname, AT_FDCWD, newname);
3975}
3976
J. Bruce Fields146a8592011-09-20 17:14:31 -04003977/**
3978 * vfs_link - create a new link
3979 * @old_dentry: object to be linked
3980 * @dir: new parent
3981 * @new_dentry: where to create the new link
3982 * @delegated_inode: returns inode needing a delegation break
3983 *
3984 * The caller must hold dir->i_mutex
3985 *
3986 * If vfs_link discovers a delegation on the to-be-linked file in need
3987 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3988 * inode in delegated_inode. The caller should then break the delegation
3989 * and retry. Because breaking a delegation may take a long time, the
3990 * caller should drop the i_mutex before doing so.
3991 *
3992 * Alternatively, a caller may pass NULL for delegated_inode. This may
3993 * be appropriate for callers that expect the underlying filesystem not
3994 * to be NFS exported.
3995 */
3996int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003997{
3998 struct inode *inode = old_dentry->d_inode;
Al Viro8de52772012-02-06 12:45:27 -05003999 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004000 int error;
4001
4002 if (!inode)
4003 return -ENOENT;
4004
Miklos Szeredia95164d2008-07-30 15:08:48 +02004005 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004006 if (error)
4007 return error;
4008
4009 if (dir->i_sb != inode->i_sb)
4010 return -EXDEV;
4011
4012 /*
4013 * A link to an append-only or immutable file cannot be created.
4014 */
4015 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4016 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05004017 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004018 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02004019 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004020 return -EPERM;
4021
4022 error = security_inode_link(old_dentry, dir, new_dentry);
4023 if (error)
4024 return error;
4025
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02004026 mutex_lock(&inode->i_mutex);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05304027 /* Make sure we don't allow creating hardlink to an unlinked file */
Al Virof4e0c302013-06-11 08:34:36 +04004028 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05304029 error = -ENOENT;
Al Viro8de52772012-02-06 12:45:27 -05004030 else if (max_links && inode->i_nlink >= max_links)
4031 error = -EMLINK;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004032 else {
4033 error = try_break_deleg(inode, delegated_inode);
4034 if (!error)
4035 error = dir->i_op->link(old_dentry, dir, new_dentry);
4036 }
Al Virof4e0c302013-06-11 08:34:36 +04004037
4038 if (!error && (inode->i_state & I_LINKABLE)) {
4039 spin_lock(&inode->i_lock);
4040 inode->i_state &= ~I_LINKABLE;
4041 spin_unlock(&inode->i_lock);
4042 }
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02004043 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07004044 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02004045 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004046 return error;
4047}
Al Viro4d359502014-03-14 12:20:17 -04004048EXPORT_SYMBOL(vfs_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004049
4050/*
4051 * Hardlinks are often used in delicate situations. We avoid
4052 * security-related surprises by not following symlinks on the
4053 * newname. --KAB
4054 *
4055 * We don't follow them on the oldname either to be compatible
4056 * with linux 2.0, and to avoid hard-linking to directories
4057 * and other special files. --ADM
4058 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01004059SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4060 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004061{
4062 struct dentry *new_dentry;
Al Virodae6ad82011-06-26 11:50:15 -04004063 struct path old_path, new_path;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004064 struct inode *delegated_inode = NULL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304065 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004066 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004067
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304068 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004069 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304070 /*
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004071 * To use null names we require CAP_DAC_READ_SEARCH
4072 * This ensures that not everyone will be able to create
4073 * handlink using the passed filedescriptor.
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304074 */
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004075 if (flags & AT_EMPTY_PATH) {
4076 if (!capable(CAP_DAC_READ_SEARCH))
4077 return -ENOENT;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304078 how = LOOKUP_EMPTY;
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004079 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004080
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304081 if (flags & AT_SYMLINK_FOLLOW)
4082 how |= LOOKUP_FOLLOW;
Jeff Layton442e31c2012-12-20 16:15:38 -05004083retry:
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304084 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004085 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04004086 return error;
4087
Jeff Layton442e31c2012-12-20 16:15:38 -05004088 new_dentry = user_path_create(newdfd, newname, &new_path,
4089 (how & LOOKUP_REVAL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004090 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07004091 if (IS_ERR(new_dentry))
Al Virodae6ad82011-06-26 11:50:15 -04004092 goto out;
4093
4094 error = -EXDEV;
4095 if (old_path.mnt != new_path.mnt)
4096 goto out_dput;
Kees Cook800179c2012-07-25 17:29:07 -07004097 error = may_linkat(&old_path);
4098 if (unlikely(error))
4099 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04004100 error = security_path_link(old_path.dentry, &new_path, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004101 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04004102 goto out_dput;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004103 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
Dave Hansen75c3f292008-02-15 14:37:45 -08004104out_dput:
Al Viro921a1652012-07-20 01:15:31 +04004105 done_path_create(&new_path, new_dentry);
J. Bruce Fields146a8592011-09-20 17:14:31 -04004106 if (delegated_inode) {
4107 error = break_deleg_wait(&delegated_inode);
Oleg Drokind22e6332014-01-31 15:41:58 -05004108 if (!error) {
4109 path_put(&old_path);
J. Bruce Fields146a8592011-09-20 17:14:31 -04004110 goto retry;
Oleg Drokind22e6332014-01-31 15:41:58 -05004111 }
J. Bruce Fields146a8592011-09-20 17:14:31 -04004112 }
Jeff Layton442e31c2012-12-20 16:15:38 -05004113 if (retry_estale(error, how)) {
Oleg Drokind22e6332014-01-31 15:41:58 -05004114 path_put(&old_path);
Jeff Layton442e31c2012-12-20 16:15:38 -05004115 how |= LOOKUP_REVAL;
4116 goto retry;
4117 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004118out:
Al Viro2d8f3032008-07-22 09:59:21 -04004119 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004120
4121 return error;
4122}
4123
Heiko Carstens3480b252009-01-14 14:14:16 +01004124SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004125{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004126 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004127}
4128
Miklos Szeredibc270272014-04-01 17:08:42 +02004129/**
4130 * vfs_rename - rename a filesystem object
4131 * @old_dir: parent of source
4132 * @old_dentry: source
4133 * @new_dir: parent of destination
4134 * @new_dentry: destination
4135 * @delegated_inode: returns an inode needing a delegation break
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004136 * @flags: rename flags
Miklos Szeredibc270272014-04-01 17:08:42 +02004137 *
4138 * The caller must hold multiple mutexes--see lock_rename()).
4139 *
4140 * If vfs_rename discovers a delegation in need of breaking at either
4141 * the source or destination, it will return -EWOULDBLOCK and return a
4142 * reference to the inode in delegated_inode. The caller should then
4143 * break the delegation and retry. Because breaking a delegation may
4144 * take a long time, the caller should drop all locks before doing
4145 * so.
4146 *
4147 * Alternatively, a caller may pass NULL for delegated_inode. This may
4148 * be appropriate for callers that expect the underlying filesystem not
4149 * to be NFS exported.
4150 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07004151 * The worst of all namespace operations - renaming directory. "Perverted"
4152 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4153 * Problems:
J. Bruce Fieldsd03b29a2014-02-17 16:52:33 -05004154 * a) we can get into loop creation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004155 * b) race potential - two innocent renames can create a loop together.
4156 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004157 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07004158 * story.
J. Bruce Fields6cedba82012-03-05 11:40:41 -05004159 * c) we have to lock _four_ objects - parents and victim (if it exists),
4160 * and source (if it is not a directory).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004161 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07004162 * whether the target exists). Solution: try to be smart with locking
4163 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004164 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07004165 * move will be locked. Thus we can rank directories by the tree
4166 * (ancestors first) and rank all non-directories after them.
4167 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004168 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004169 * HOWEVER, it relies on the assumption that any object with ->lookup()
4170 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4171 * we'd better make sure that there's no link(2) for them.
Sage Weile4eaac02011-05-24 13:06:07 -07004172 * d) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004173 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07004174 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05004175 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07004176 * locking].
4177 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004178int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004179 struct inode *new_dir, struct dentry *new_dentry,
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004180 struct inode **delegated_inode, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004181{
4182 int error;
Miklos Szeredibc270272014-04-01 17:08:42 +02004183 bool is_dir = d_is_dir(old_dentry);
Eric Paris59b0df22010-02-08 12:53:52 -05004184 const unsigned char *old_name;
Miklos Szeredibc270272014-04-01 17:08:42 +02004185 struct inode *source = old_dentry->d_inode;
4186 struct inode *target = new_dentry->d_inode;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004187 bool new_is_dir = false;
4188 unsigned max_links = new_dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004189
Miklos Szeredibc270272014-04-01 17:08:42 +02004190 if (source == target)
4191 return 0;
4192
Linus Torvalds1da177e2005-04-16 15:20:36 -07004193 error = may_delete(old_dir, old_dentry, is_dir);
4194 if (error)
4195 return error;
4196
Miklos Szeredida1ce062014-04-01 17:08:43 +02004197 if (!target) {
Miklos Szeredia95164d2008-07-30 15:08:48 +02004198 error = may_create(new_dir, new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004199 } else {
4200 new_is_dir = d_is_dir(new_dentry);
4201
4202 if (!(flags & RENAME_EXCHANGE))
4203 error = may_delete(new_dir, new_dentry, is_dir);
4204 else
4205 error = may_delete(new_dir, new_dentry, new_is_dir);
4206 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004207 if (error)
4208 return error;
4209
Miklos Szeredi7177a9c2014-07-23 15:15:30 +02004210 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004211 return -EPERM;
4212
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004213 if (flags && !old_dir->i_op->rename2)
4214 return -EINVAL;
4215
Miklos Szeredibc270272014-04-01 17:08:42 +02004216 /*
4217 * If we are going to change the parent - check write permissions,
4218 * we'll need to flip '..'.
4219 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02004220 if (new_dir != old_dir) {
4221 if (is_dir) {
4222 error = inode_permission(source, MAY_WRITE);
4223 if (error)
4224 return error;
4225 }
4226 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4227 error = inode_permission(target, MAY_WRITE);
4228 if (error)
4229 return error;
4230 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004231 }
Robert Love0eeca282005-07-12 17:06:03 -04004232
Miklos Szeredi0b3974e2014-04-01 17:08:43 +02004233 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4234 flags);
Miklos Szeredibc270272014-04-01 17:08:42 +02004235 if (error)
4236 return error;
4237
4238 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4239 dget(new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004240 if (!is_dir || (flags & RENAME_EXCHANGE))
Miklos Szeredibc270272014-04-01 17:08:42 +02004241 lock_two_nondirectories(source, target);
4242 else if (target)
4243 mutex_lock(&target->i_mutex);
4244
4245 error = -EBUSY;
Eric W. Biederman7af13642013-10-04 19:15:13 -07004246 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
Miklos Szeredibc270272014-04-01 17:08:42 +02004247 goto out;
4248
Miklos Szeredida1ce062014-04-01 17:08:43 +02004249 if (max_links && new_dir != old_dir) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004250 error = -EMLINK;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004251 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
Miklos Szeredibc270272014-04-01 17:08:42 +02004252 goto out;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004253 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4254 old_dir->i_nlink >= max_links)
4255 goto out;
4256 }
4257 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4258 shrink_dcache_parent(new_dentry);
4259 if (!is_dir) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004260 error = try_break_deleg(source, delegated_inode);
4261 if (error)
4262 goto out;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004263 }
4264 if (target && !new_is_dir) {
4265 error = try_break_deleg(target, delegated_inode);
4266 if (error)
4267 goto out;
Miklos Szeredibc270272014-04-01 17:08:42 +02004268 }
Miklos Szeredi7177a9c2014-07-23 15:15:30 +02004269 if (!old_dir->i_op->rename2) {
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004270 error = old_dir->i_op->rename(old_dir, old_dentry,
4271 new_dir, new_dentry);
4272 } else {
Miklos Szeredi7177a9c2014-07-23 15:15:30 +02004273 WARN_ON(old_dir->i_op->rename != NULL);
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004274 error = old_dir->i_op->rename2(old_dir, old_dentry,
4275 new_dir, new_dentry, flags);
4276 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004277 if (error)
4278 goto out;
4279
Miklos Szeredida1ce062014-04-01 17:08:43 +02004280 if (!(flags & RENAME_EXCHANGE) && target) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004281 if (is_dir)
4282 target->i_flags |= S_DEAD;
4283 dont_mount(new_dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07004284 detach_mounts(new_dentry);
Miklos Szeredibc270272014-04-01 17:08:42 +02004285 }
Miklos Szeredida1ce062014-04-01 17:08:43 +02004286 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4287 if (!(flags & RENAME_EXCHANGE))
4288 d_move(old_dentry, new_dentry);
4289 else
4290 d_exchange(old_dentry, new_dentry);
4291 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004292out:
Miklos Szeredida1ce062014-04-01 17:08:43 +02004293 if (!is_dir || (flags & RENAME_EXCHANGE))
Miklos Szeredibc270272014-04-01 17:08:42 +02004294 unlock_two_nondirectories(source, target);
4295 else if (target)
4296 mutex_unlock(&target->i_mutex);
4297 dput(new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004298 if (!error) {
Al Viro123df292009-12-25 04:57:57 -05004299 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Miklos Szeredida1ce062014-04-01 17:08:43 +02004300 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4301 if (flags & RENAME_EXCHANGE) {
4302 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4303 new_is_dir, NULL, new_dentry);
4304 }
4305 }
Robert Love0eeca282005-07-12 17:06:03 -04004306 fsnotify_oldname_free(old_name);
4307
Linus Torvalds1da177e2005-04-16 15:20:36 -07004308 return error;
4309}
Al Viro4d359502014-03-14 12:20:17 -04004310EXPORT_SYMBOL(vfs_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004311
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004312SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4313 int, newdfd, const char __user *, newname, unsigned int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004314{
Al Viro2ad94ae2008-07-21 09:32:51 -04004315 struct dentry *old_dentry, *new_dentry;
4316 struct dentry *trap;
Al Virof5beed72015-04-30 16:09:11 -04004317 struct path old_path, new_path;
4318 struct qstr old_last, new_last;
4319 int old_type, new_type;
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004320 struct inode *delegated_inode = NULL;
Jeff Layton91a27b22012-10-10 15:25:28 -04004321 struct filename *from;
4322 struct filename *to;
Al Virof5beed72015-04-30 16:09:11 -04004323 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
Jeff Laytonc6a94282012-12-11 12:10:10 -05004324 bool should_retry = false;
Al Viro2ad94ae2008-07-21 09:32:51 -04004325 int error;
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004326
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004327 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredida1ce062014-04-01 17:08:43 +02004328 return -EINVAL;
4329
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004330 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4331 (flags & RENAME_EXCHANGE))
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004332 return -EINVAL;
4333
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004334 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4335 return -EPERM;
4336
Al Virof5beed72015-04-30 16:09:11 -04004337 if (flags & RENAME_EXCHANGE)
4338 target_flags = 0;
4339
Jeff Laytonc6a94282012-12-11 12:10:10 -05004340retry:
Al Virof5beed72015-04-30 16:09:11 -04004341 from = user_path_parent(olddfd, oldname,
4342 &old_path, &old_last, &old_type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04004343 if (IS_ERR(from)) {
4344 error = PTR_ERR(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004345 goto exit;
Jeff Layton91a27b22012-10-10 15:25:28 -04004346 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004347
Al Virof5beed72015-04-30 16:09:11 -04004348 to = user_path_parent(newdfd, newname,
4349 &new_path, &new_last, &new_type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04004350 if (IS_ERR(to)) {
4351 error = PTR_ERR(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004352 goto exit1;
Jeff Layton91a27b22012-10-10 15:25:28 -04004353 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004354
4355 error = -EXDEV;
Al Virof5beed72015-04-30 16:09:11 -04004356 if (old_path.mnt != new_path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004357 goto exit2;
4358
Linus Torvalds1da177e2005-04-16 15:20:36 -07004359 error = -EBUSY;
Al Virof5beed72015-04-30 16:09:11 -04004360 if (old_type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004361 goto exit2;
4362
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004363 if (flags & RENAME_NOREPLACE)
4364 error = -EEXIST;
Al Virof5beed72015-04-30 16:09:11 -04004365 if (new_type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004366 goto exit2;
4367
Al Virof5beed72015-04-30 16:09:11 -04004368 error = mnt_want_write(old_path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02004369 if (error)
4370 goto exit2;
4371
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004372retry_deleg:
Al Virof5beed72015-04-30 16:09:11 -04004373 trap = lock_rename(new_path.dentry, old_path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004374
Al Virof5beed72015-04-30 16:09:11 -04004375 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004376 error = PTR_ERR(old_dentry);
4377 if (IS_ERR(old_dentry))
4378 goto exit3;
4379 /* source must exist */
4380 error = -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01004381 if (d_is_negative(old_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004382 goto exit4;
Al Virof5beed72015-04-30 16:09:11 -04004383 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004384 error = PTR_ERR(new_dentry);
4385 if (IS_ERR(new_dentry))
4386 goto exit4;
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004387 error = -EEXIST;
4388 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4389 goto exit5;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004390 if (flags & RENAME_EXCHANGE) {
4391 error = -ENOENT;
4392 if (d_is_negative(new_dentry))
4393 goto exit5;
4394
4395 if (!d_is_dir(new_dentry)) {
4396 error = -ENOTDIR;
Al Virof5beed72015-04-30 16:09:11 -04004397 if (new_last.name[new_last.len])
Miklos Szeredida1ce062014-04-01 17:08:43 +02004398 goto exit5;
4399 }
4400 }
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004401 /* unless the source is a directory trailing slashes give -ENOTDIR */
4402 if (!d_is_dir(old_dentry)) {
4403 error = -ENOTDIR;
Al Virof5beed72015-04-30 16:09:11 -04004404 if (old_last.name[old_last.len])
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004405 goto exit5;
Al Virof5beed72015-04-30 16:09:11 -04004406 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004407 goto exit5;
4408 }
4409 /* source should not be ancestor of target */
4410 error = -EINVAL;
4411 if (old_dentry == trap)
4412 goto exit5;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004413 /* target should not be an ancestor of source */
Miklos Szeredida1ce062014-04-01 17:08:43 +02004414 if (!(flags & RENAME_EXCHANGE))
4415 error = -ENOTEMPTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004416 if (new_dentry == trap)
4417 goto exit5;
4418
Al Virof5beed72015-04-30 16:09:11 -04004419 error = security_path_rename(&old_path, old_dentry,
4420 &new_path, new_dentry, flags);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004421 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02004422 goto exit5;
Al Virof5beed72015-04-30 16:09:11 -04004423 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4424 new_path.dentry->d_inode, new_dentry,
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004425 &delegated_inode, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004426exit5:
4427 dput(new_dentry);
4428exit4:
4429 dput(old_dentry);
4430exit3:
Al Virof5beed72015-04-30 16:09:11 -04004431 unlock_rename(new_path.dentry, old_path.dentry);
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004432 if (delegated_inode) {
4433 error = break_deleg_wait(&delegated_inode);
4434 if (!error)
4435 goto retry_deleg;
4436 }
Al Virof5beed72015-04-30 16:09:11 -04004437 mnt_drop_write(old_path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004438exit2:
Jeff Laytonc6a94282012-12-11 12:10:10 -05004439 if (retry_estale(error, lookup_flags))
4440 should_retry = true;
Al Virof5beed72015-04-30 16:09:11 -04004441 path_put(&new_path);
Al Viro2ad94ae2008-07-21 09:32:51 -04004442 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004443exit1:
Al Virof5beed72015-04-30 16:09:11 -04004444 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004445 putname(from);
Jeff Laytonc6a94282012-12-11 12:10:10 -05004446 if (should_retry) {
4447 should_retry = false;
4448 lookup_flags |= LOOKUP_REVAL;
4449 goto retry;
4450 }
Al Viro2ad94ae2008-07-21 09:32:51 -04004451exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004452 return error;
4453}
4454
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004455SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4456 int, newdfd, const char __user *, newname)
4457{
4458 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4459}
4460
Heiko Carstensa26eab22009-01-14 14:14:17 +01004461SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004462{
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004463 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004464}
4465
Miklos Szeredi787fb6b2014-10-24 00:14:36 +02004466int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4467{
4468 int error = may_create(dir, dentry);
4469 if (error)
4470 return error;
4471
4472 if (!dir->i_op->mknod)
4473 return -EPERM;
4474
4475 return dir->i_op->mknod(dir, dentry,
4476 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4477}
4478EXPORT_SYMBOL(vfs_whiteout);
4479
Al Viro5d826c82014-03-14 13:42:45 -04004480int readlink_copy(char __user *buffer, int buflen, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004481{
Al Viro5d826c82014-03-14 13:42:45 -04004482 int len = PTR_ERR(link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004483 if (IS_ERR(link))
4484 goto out;
4485
4486 len = strlen(link);
4487 if (len > (unsigned) buflen)
4488 len = buflen;
4489 if (copy_to_user(buffer, link, len))
4490 len = -EFAULT;
4491out:
4492 return len;
4493}
Al Viro5d826c82014-03-14 13:42:45 -04004494EXPORT_SYMBOL(readlink_copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004495
4496/*
4497 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4498 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4499 * using) it for any given inode is up to filesystem.
4500 */
4501int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4502{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004503 void *cookie;
Al Viro5f2c4172015-05-07 11:14:26 -04004504 struct inode *inode = d_inode(dentry);
4505 const char *link = inode->i_link;
Marcin Slusarz694a1762008-06-09 16:40:37 -07004506 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004507
Al Virod4dee482015-04-30 20:08:02 -04004508 if (!link) {
Al Viro5f2c4172015-05-07 11:14:26 -04004509 link = inode->i_op->follow_link(dentry, &cookie);
Al Virod4dee482015-04-30 20:08:02 -04004510 if (IS_ERR(link))
4511 return PTR_ERR(link);
4512 }
Al Viro680baac2015-05-02 13:32:22 -04004513 res = readlink_copy(buffer, buflen, link);
Al Viro5f2c4172015-05-07 11:14:26 -04004514 if (inode->i_op->put_link)
4515 inode->i_op->put_link(inode, cookie);
Marcin Slusarz694a1762008-06-09 16:40:37 -07004516 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004517}
Al Viro4d359502014-03-14 12:20:17 -04004518EXPORT_SYMBOL(generic_readlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004519
Linus Torvalds1da177e2005-04-16 15:20:36 -07004520/* get the link contents into pagecache */
4521static char *page_getlink(struct dentry * dentry, struct page **ppage)
4522{
Duane Griffinebd09ab2008-12-19 20:47:12 +00004523 char *kaddr;
4524 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004525 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07004526 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004527 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07004528 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004529 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00004530 kaddr = kmap(page);
4531 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4532 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004533}
4534
4535int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4536{
4537 struct page *page = NULL;
Al Viro5d826c82014-03-14 13:42:45 -04004538 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004539 if (page) {
4540 kunmap(page);
4541 page_cache_release(page);
4542 }
4543 return res;
4544}
Al Viro4d359502014-03-14 12:20:17 -04004545EXPORT_SYMBOL(page_readlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004546
Al Viro6e771372015-05-02 13:37:52 -04004547const char *page_follow_link_light(struct dentry *dentry, void **cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004548{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004549 struct page *page = NULL;
Al Viro680baac2015-05-02 13:32:22 -04004550 char *res = page_getlink(dentry, &page);
4551 if (!IS_ERR(res))
4552 *cookie = page;
4553 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004554}
Al Viro4d359502014-03-14 12:20:17 -04004555EXPORT_SYMBOL(page_follow_link_light);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004556
Al Viro5f2c4172015-05-07 11:14:26 -04004557void page_put_link(struct inode *unused, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004558{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004559 struct page *page = cookie;
Al Viro680baac2015-05-02 13:32:22 -04004560 kunmap(page);
4561 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004562}
Al Viro4d359502014-03-14 12:20:17 -04004563EXPORT_SYMBOL(page_put_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004564
Nick Piggin54566b22009-01-04 12:00:53 -08004565/*
4566 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4567 */
4568int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004569{
4570 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004571 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07004572 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08004573 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004574 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08004575 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4576 if (nofs)
4577 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004578
NeilBrown7e53cac2006-03-25 03:07:57 -08004579retry:
Nick Pigginafddba42007-10-16 01:25:01 -07004580 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08004581 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004582 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07004583 goto fail;
4584
Cong Wange8e3c3d2011-11-25 23:14:27 +08004585 kaddr = kmap_atomic(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004586 memcpy(kaddr, symname, len-1);
Cong Wange8e3c3d2011-11-25 23:14:27 +08004587 kunmap_atomic(kaddr);
Nick Pigginafddba42007-10-16 01:25:01 -07004588
4589 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4590 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004591 if (err < 0)
4592 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07004593 if (err < len-1)
4594 goto retry;
4595
Linus Torvalds1da177e2005-04-16 15:20:36 -07004596 mark_inode_dirty(inode);
4597 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004598fail:
4599 return err;
4600}
Al Viro4d359502014-03-14 12:20:17 -04004601EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004602
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004603int page_symlink(struct inode *inode, const char *symname, int len)
4604{
4605 return __page_symlink(inode, symname, len,
Michal Hockoc62d2552015-11-06 16:28:49 -08004606 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004607}
Al Viro4d359502014-03-14 12:20:17 -04004608EXPORT_SYMBOL(page_symlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004609
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08004610const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004611 .readlink = generic_readlink,
4612 .follow_link = page_follow_link_light,
4613 .put_link = page_put_link,
4614};
Linus Torvalds1da177e2005-04-16 15:20:36 -07004615EXPORT_SYMBOL(page_symlink_inode_operations);