blob: 43927d14db67c68706ef8dd208c714ec2ec6addb [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 -0400121void final_putname(struct filename *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122{
Jeff Layton7950e382012-10-10 16:43:13 -0400123 if (name->separate) {
124 __putname(name->name);
125 kfree(name);
126 } else {
127 __putname(name);
128 }
Jeff Layton91a27b22012-10-10 15:25:28 -0400129}
130
Jeff Layton7950e382012-10-10 16:43:13 -0400131#define EMBEDDED_NAME_MAX (PATH_MAX - sizeof(struct filename))
132
Jeff Layton91a27b22012-10-10 15:25:28 -0400133static struct filename *
134getname_flags(const char __user *filename, int flags, int *empty)
135{
136 struct filename *result, *err;
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700137 int len;
Jeff Layton7950e382012-10-10 16:43:13 -0400138 long max;
139 char *kname;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140
Jeff Layton7ac86262012-10-10 15:25:28 -0400141 result = audit_reusename(filename);
142 if (result)
143 return result;
144
Jeff Layton7950e382012-10-10 16:43:13 -0400145 result = __getname();
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700146 if (unlikely(!result))
Eric Paris4043cde2012-01-03 14:23:08 -0500147 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148
Jeff Layton7950e382012-10-10 16:43:13 -0400149 /*
150 * First, try to embed the struct filename inside the names_cache
151 * allocation
152 */
153 kname = (char *)result + sizeof(*result);
Jeff Layton91a27b22012-10-10 15:25:28 -0400154 result->name = kname;
Jeff Layton7950e382012-10-10 16:43:13 -0400155 result->separate = false;
156 max = EMBEDDED_NAME_MAX;
157
158recopy:
159 len = strncpy_from_user(kname, filename, max);
Jeff Layton91a27b22012-10-10 15:25:28 -0400160 if (unlikely(len < 0)) {
161 err = ERR_PTR(len);
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700162 goto error;
Jeff Layton91a27b22012-10-10 15:25:28 -0400163 }
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700164
Jeff Layton7950e382012-10-10 16:43:13 -0400165 /*
166 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
167 * separate struct filename so we can dedicate the entire
168 * names_cache allocation for the pathname, and re-do the copy from
169 * userland.
170 */
171 if (len == EMBEDDED_NAME_MAX && max == EMBEDDED_NAME_MAX) {
172 kname = (char *)result;
173
174 result = kzalloc(sizeof(*result), GFP_KERNEL);
175 if (!result) {
176 err = ERR_PTR(-ENOMEM);
177 result = (struct filename *)kname;
178 goto error;
179 }
180 result->name = kname;
181 result->separate = true;
182 max = PATH_MAX;
183 goto recopy;
184 }
185
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700186 /* The empty path is special. */
187 if (unlikely(!len)) {
188 if (empty)
Eric Paris4043cde2012-01-03 14:23:08 -0500189 *empty = 1;
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700190 err = ERR_PTR(-ENOENT);
191 if (!(flags & LOOKUP_EMPTY))
192 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 }
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700194
195 err = ERR_PTR(-ENAMETOOLONG);
Jeff Layton7950e382012-10-10 16:43:13 -0400196 if (unlikely(len >= PATH_MAX))
197 goto error;
198
199 result->uptr = filename;
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800200 result->aname = NULL;
Jeff Layton7950e382012-10-10 16:43:13 -0400201 audit_getname(result);
202 return result;
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700203
204error:
Jeff Layton7950e382012-10-10 16:43:13 -0400205 final_putname(result);
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700206 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207}
208
Jeff Layton91a27b22012-10-10 15:25:28 -0400209struct filename *
210getname(const char __user * filename)
Al Virof52e0c12011-03-14 18:56:51 -0400211{
Linus Torvaldsf7493e52012-03-22 16:10:40 -0700212 return getname_flags(filename, 0, NULL);
Al Virof52e0c12011-03-14 18:56:51 -0400213}
214
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800215/*
216 * The "getname_kernel()" interface doesn't do pathnames longer
217 * than EMBEDDED_NAME_MAX. Deal with it - you're a kernel user.
218 */
219struct filename *
220getname_kernel(const char * filename)
221{
222 struct filename *result;
223 char *kname;
224 int len;
225
226 len = strlen(filename);
227 if (len >= EMBEDDED_NAME_MAX)
228 return ERR_PTR(-ENAMETOOLONG);
229
230 result = __getname();
231 if (unlikely(!result))
232 return ERR_PTR(-ENOMEM);
233
234 kname = (char *)result + sizeof(*result);
235 result->name = kname;
236 result->uptr = NULL;
237 result->aname = NULL;
238 result->separate = false;
239
240 strlcpy(kname, filename, EMBEDDED_NAME_MAX);
241 return result;
242}
243
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244#ifdef CONFIG_AUDITSYSCALL
Jeff Layton91a27b22012-10-10 15:25:28 -0400245void putname(struct filename *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246{
Al Viro5ac3a9c2006-07-16 06:38:45 -0400247 if (unlikely(!audit_dummy_context()))
Jeff Layton91a27b22012-10-10 15:25:28 -0400248 return audit_putname(name);
249 final_putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251#endif
252
Linus Torvaldse77819e2011-07-22 19:30:19 -0700253static int check_acl(struct inode *inode, int mask)
254{
Linus Torvalds84635d62011-07-25 22:47:03 -0700255#ifdef CONFIG_FS_POSIX_ACL
Linus Torvaldse77819e2011-07-22 19:30:19 -0700256 struct posix_acl *acl;
257
Linus Torvaldse77819e2011-07-22 19:30:19 -0700258 if (mask & MAY_NOT_BLOCK) {
Al Viro35678662011-08-02 21:32:13 -0400259 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
260 if (!acl)
Linus Torvaldse77819e2011-07-22 19:30:19 -0700261 return -EAGAIN;
Al Viro35678662011-08-02 21:32:13 -0400262 /* no ->get_acl() calls in RCU mode... */
263 if (acl == ACL_NOT_CACHED)
264 return -ECHILD;
Ari Savolainen206b1d02011-08-06 19:43:07 +0300265 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
Linus Torvaldse77819e2011-07-22 19:30:19 -0700266 }
267
Christoph Hellwig2982baa2013-12-20 05:16:38 -0800268 acl = get_acl(inode, ACL_TYPE_ACCESS);
269 if (IS_ERR(acl))
270 return PTR_ERR(acl);
Linus Torvaldse77819e2011-07-22 19:30:19 -0700271 if (acl) {
272 int error = posix_acl_permission(inode, acl, mask);
273 posix_acl_release(acl);
274 return error;
275 }
Linus Torvalds84635d62011-07-25 22:47:03 -0700276#endif
Linus Torvaldse77819e2011-07-22 19:30:19 -0700277
278 return -EAGAIN;
279}
280
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700281/*
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530282 * This does the basic permission checking
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700283 */
Al Viro7e401452011-06-20 19:12:17 -0400284static int acl_permission_check(struct inode *inode, int mask)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700285{
Linus Torvalds26cf46b2011-05-13 11:51:01 -0700286 unsigned int mode = inode->i_mode;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700287
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -0800288 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700289 mode >>= 6;
290 else {
Linus Torvaldse77819e2011-07-22 19:30:19 -0700291 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
Al Viro7e401452011-06-20 19:12:17 -0400292 int error = check_acl(inode, mask);
Nick Pigginb74c79e2011-01-07 17:49:58 +1100293 if (error != -EAGAIN)
294 return error;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700295 }
296
297 if (in_group_p(inode->i_gid))
298 mode >>= 3;
299 }
300
301 /*
302 * If the DACs are ok we don't need any capability check.
303 */
Al Viro9c2c7032011-06-20 19:06:22 -0400304 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700305 return 0;
306 return -EACCES;
307}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308
309/**
Nick Pigginb74c79e2011-01-07 17:49:58 +1100310 * generic_permission - check for access rights on a Posix-like filesystem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 * @inode: inode to check access rights for
Andreas Gruenbacher8fd90c82011-10-23 23:13:30 +0530312 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 *
314 * Used to check for read/write/execute permissions on a file.
315 * We use "fsuid" for this, letting us set arbitrary permissions
316 * for filesystem access without changing the "normal" uids which
Nick Pigginb74c79e2011-01-07 17:49:58 +1100317 * are used for other things.
318 *
319 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
320 * request cannot be satisfied (eg. requires blocking or too much complexity).
321 * It would then be called again in ref-walk mode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 */
Al Viro2830ba72011-06-20 19:16:29 -0400323int generic_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700325 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326
327 /*
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530328 * Do the basic permission checks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 */
Al Viro7e401452011-06-20 19:12:17 -0400330 ret = acl_permission_check(inode, mask);
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700331 if (ret != -EACCES)
332 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333
Al Virod594e7e2011-06-20 19:55:42 -0400334 if (S_ISDIR(inode->i_mode)) {
335 /* DACs are overridable for directories */
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700336 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
Al Virod594e7e2011-06-20 19:55:42 -0400337 return 0;
338 if (!(mask & MAY_WRITE))
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700339 if (capable_wrt_inode_uidgid(inode,
340 CAP_DAC_READ_SEARCH))
Al Virod594e7e2011-06-20 19:55:42 -0400341 return 0;
342 return -EACCES;
343 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 /*
345 * Read/write DACs are always overridable.
Al Virod594e7e2011-06-20 19:55:42 -0400346 * Executable DACs are overridable when there is
347 * at least one exec bit set.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 */
Al Virod594e7e2011-06-20 19:55:42 -0400349 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700350 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 return 0;
352
353 /*
354 * Searching includes executable on directories, else just read.
355 */
Serge E. Hallyn7ea66002009-12-29 14:50:19 -0600356 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
Al Virod594e7e2011-06-20 19:55:42 -0400357 if (mask == MAY_READ)
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700358 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 return 0;
360
361 return -EACCES;
362}
Al Viro4d359502014-03-14 12:20:17 -0400363EXPORT_SYMBOL(generic_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700365/*
366 * We _really_ want to just do "generic_permission()" without
367 * even looking at the inode->i_op values. So we keep a cache
368 * flag in inode->i_opflags, that says "this has not special
369 * permission function, use the fast case".
370 */
371static inline int do_inode_permission(struct inode *inode, int mask)
372{
373 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
374 if (likely(inode->i_op->permission))
375 return inode->i_op->permission(inode, mask);
376
377 /* This gets set once for the inode lifetime */
378 spin_lock(&inode->i_lock);
379 inode->i_opflags |= IOP_FASTPERM;
380 spin_unlock(&inode->i_lock);
381 }
382 return generic_permission(inode, mask);
383}
384
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200385/**
David Howells0bdaea92012-06-25 12:55:46 +0100386 * __inode_permission - Check for access rights to a given inode
387 * @inode: Inode to check permission on
388 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200389 *
David Howells0bdaea92012-06-25 12:55:46 +0100390 * Check for read/write/execute permissions on an inode.
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530391 *
392 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
David Howells0bdaea92012-06-25 12:55:46 +0100393 *
394 * This does not check for a read-only file system. You probably want
395 * inode_permission().
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200396 */
David Howells0bdaea92012-06-25 12:55:46 +0100397int __inode_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398{
Al Viroe6305c42008-07-15 21:03:57 -0400399 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700401 if (unlikely(mask & MAY_WRITE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 /*
403 * Nobody gets write access to an immutable file.
404 */
405 if (IS_IMMUTABLE(inode))
406 return -EACCES;
407 }
408
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700409 retval = do_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 if (retval)
411 return retval;
412
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -0700413 retval = devcgroup_inode_permission(inode, mask);
414 if (retval)
415 return retval;
416
Eric Parisd09ca732010-07-23 11:43:57 -0400417 return security_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418}
419
Al Virof4d6ff82011-06-19 13:14:21 -0400420/**
David Howells0bdaea92012-06-25 12:55:46 +0100421 * sb_permission - Check superblock-level permissions
422 * @sb: Superblock of inode to check permission on
Randy Dunlap55852632012-08-18 17:39:25 -0700423 * @inode: Inode to check permission on
David Howells0bdaea92012-06-25 12:55:46 +0100424 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
425 *
426 * Separate out file-system wide checks from inode-specific permission checks.
427 */
428static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
429{
430 if (unlikely(mask & MAY_WRITE)) {
431 umode_t mode = inode->i_mode;
432
433 /* Nobody gets write access to a read-only fs. */
434 if ((sb->s_flags & MS_RDONLY) &&
435 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
436 return -EROFS;
437 }
438 return 0;
439}
440
441/**
442 * inode_permission - Check for access rights to a given inode
443 * @inode: Inode to check permission on
444 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
445 *
446 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
447 * this, letting us set arbitrary permissions for filesystem access without
448 * changing the "normal" UIDs which are used for other things.
449 *
450 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
451 */
452int inode_permission(struct inode *inode, int mask)
453{
454 int retval;
455
456 retval = sb_permission(inode->i_sb, inode, mask);
457 if (retval)
458 return retval;
459 return __inode_permission(inode, mask);
460}
Al Viro4d359502014-03-14 12:20:17 -0400461EXPORT_SYMBOL(inode_permission);
David Howells0bdaea92012-06-25 12:55:46 +0100462
463/**
Jan Blunck5dd784d02008-02-14 19:34:38 -0800464 * path_get - get a reference to a path
465 * @path: path to get the reference to
466 *
467 * Given a path increment the reference count to the dentry and the vfsmount.
468 */
Al Virodcf787f2013-03-01 23:51:07 -0500469void path_get(const struct path *path)
Jan Blunck5dd784d02008-02-14 19:34:38 -0800470{
471 mntget(path->mnt);
472 dget(path->dentry);
473}
474EXPORT_SYMBOL(path_get);
475
476/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800477 * path_put - put a reference to a path
478 * @path: path to put the reference to
479 *
480 * Given a path decrement the reference count to the dentry and the vfsmount.
481 */
Al Virodcf787f2013-03-01 23:51:07 -0500482void path_put(const struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483{
Jan Blunck1d957f92008-02-14 19:34:35 -0800484 dput(path->dentry);
485 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486}
Jan Blunck1d957f92008-02-14 19:34:35 -0800487EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488
Al Viro19660af2011-03-25 10:32:48 -0400489/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100490 * Path walking has 2 modes, rcu-walk and ref-walk (see
Al Viro19660af2011-03-25 10:32:48 -0400491 * Documentation/filesystems/path-lookup.txt). In situations when we can't
492 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
493 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
494 * mode. Refcounts are grabbed at the last known good point before rcu-walk
495 * got stuck, so ref-walk may continue from there. If this is not successful
496 * (eg. a seqcount has changed), then failure is returned and it's up to caller
497 * to restart the path walk from the beginning in ref-walk mode.
Nick Piggin31e6b012011-01-07 17:49:52 +1100498 */
Nick Piggin31e6b012011-01-07 17:49:52 +1100499
500/**
Al Viro19660af2011-03-25 10:32:48 -0400501 * unlazy_walk - try to switch to ref-walk mode.
502 * @nd: nameidata pathwalk data
503 * @dentry: child of nd->path.dentry or NULL
Randy Dunlap39191622011-01-08 19:36:21 -0800504 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100505 *
Al Viro19660af2011-03-25 10:32:48 -0400506 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
507 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
508 * @nd or NULL. Must be called from rcu-walk context.
Nick Piggin31e6b012011-01-07 17:49:52 +1100509 */
Al Viro19660af2011-03-25 10:32:48 -0400510static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
Nick Piggin31e6b012011-01-07 17:49:52 +1100511{
512 struct fs_struct *fs = current->fs;
513 struct dentry *parent = nd->path.dentry;
514
515 BUG_ON(!(nd->flags & LOOKUP_RCU));
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700516
517 /*
Al Viro48a066e2013-09-29 22:06:07 -0400518 * After legitimizing the bastards, terminate_walk()
519 * will do the right thing for non-RCU mode, and all our
520 * subsequent exit cases should rcu_read_unlock()
521 * before returning. Do vfsmount first; if dentry
522 * can't be legitimized, just set nd->path.dentry to NULL
523 * and rely on dput(NULL) being a no-op.
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700524 */
Al Viro48a066e2013-09-29 22:06:07 -0400525 if (!legitimize_mnt(nd->path.mnt, nd->m_seq))
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700526 return -ECHILD;
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700527 nd->flags &= ~LOOKUP_RCU;
Linus Torvalds15570082013-09-02 11:38:06 -0700528
Al Viro48a066e2013-09-29 22:06:07 -0400529 if (!lockref_get_not_dead(&parent->d_lockref)) {
530 nd->path.dentry = NULL;
Al Virod870b4a2013-11-29 01:48:32 -0500531 goto out;
Al Viro48a066e2013-09-29 22:06:07 -0400532 }
533
Linus Torvalds15570082013-09-02 11:38:06 -0700534 /*
535 * For a negative lookup, the lookup sequence point is the parents
536 * sequence point, and it only needs to revalidate the parent dentry.
537 *
538 * For a positive lookup, we need to move both the parent and the
539 * dentry from the RCU domain to be properly refcounted. And the
540 * sequence number in the dentry validates *both* dentry counters,
541 * since we checked the sequence number of the parent after we got
542 * the child sequence number. So we know the parent must still
543 * be valid if the child sequence number is still valid.
544 */
Al Viro19660af2011-03-25 10:32:48 -0400545 if (!dentry) {
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700546 if (read_seqcount_retry(&parent->d_seq, nd->seq))
547 goto out;
Al Viro19660af2011-03-25 10:32:48 -0400548 BUG_ON(nd->inode != parent->d_inode);
549 } else {
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700550 if (!lockref_get_not_dead(&dentry->d_lockref))
551 goto out;
552 if (read_seqcount_retry(&dentry->d_seq, nd->seq))
553 goto drop_dentry;
Al Viro19660af2011-03-25 10:32:48 -0400554 }
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700555
556 /*
557 * Sequence counts matched. Now make sure that the root is
558 * still valid and get it if required.
559 */
560 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
561 spin_lock(&fs->lock);
562 if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry)
563 goto unlock_and_drop_dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +1100564 path_get(&nd->root);
565 spin_unlock(&fs->lock);
566 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100567
Al Viro8b61e742013-11-08 12:45:01 -0500568 rcu_read_unlock();
Nick Piggin31e6b012011-01-07 17:49:52 +1100569 return 0;
Al Viro19660af2011-03-25 10:32:48 -0400570
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700571unlock_and_drop_dentry:
572 spin_unlock(&fs->lock);
573drop_dentry:
Al Viro8b61e742013-11-08 12:45:01 -0500574 rcu_read_unlock();
Linus Torvalds15570082013-09-02 11:38:06 -0700575 dput(dentry);
Linus Torvaldsd0d27272013-09-10 12:17:49 -0700576 goto drop_root_mnt;
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700577out:
Al Viro8b61e742013-11-08 12:45:01 -0500578 rcu_read_unlock();
Linus Torvaldsd0d27272013-09-10 12:17:49 -0700579drop_root_mnt:
580 if (!(nd->flags & LOOKUP_ROOT))
581 nd->root.mnt = NULL;
Nick Piggin31e6b012011-01-07 17:49:52 +1100582 return -ECHILD;
583}
584
Al Viro4ce16ef32012-06-10 16:10:59 -0400585static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
Nick Piggin34286d62011-01-07 17:49:57 +1100586{
Al Viro4ce16ef32012-06-10 16:10:59 -0400587 return dentry->d_op->d_revalidate(dentry, flags);
Nick Piggin34286d62011-01-07 17:49:57 +1100588}
589
Al Viro9f1fafe2011-03-25 11:00:12 -0400590/**
591 * complete_walk - successful completion of path walk
592 * @nd: pointer nameidata
Jeff Layton39159de2009-12-07 12:01:50 -0500593 *
Al Viro9f1fafe2011-03-25 11:00:12 -0400594 * If we had been in RCU mode, drop out of it and legitimize nd->path.
595 * Revalidate the final result, unless we'd already done that during
596 * the path walk or the filesystem doesn't ask for it. Return 0 on
597 * success, -error on failure. In case of failure caller does not
598 * need to drop nd->path.
Jeff Layton39159de2009-12-07 12:01:50 -0500599 */
Al Viro9f1fafe2011-03-25 11:00:12 -0400600static int complete_walk(struct nameidata *nd)
Jeff Layton39159de2009-12-07 12:01:50 -0500601{
Al Viro16c2cd72011-02-22 15:50:10 -0500602 struct dentry *dentry = nd->path.dentry;
Jeff Layton39159de2009-12-07 12:01:50 -0500603 int status;
Jeff Layton39159de2009-12-07 12:01:50 -0500604
Al Viro9f1fafe2011-03-25 11:00:12 -0400605 if (nd->flags & LOOKUP_RCU) {
606 nd->flags &= ~LOOKUP_RCU;
607 if (!(nd->flags & LOOKUP_ROOT))
608 nd->root.mnt = NULL;
Linus Torvalds15570082013-09-02 11:38:06 -0700609
Al Viro48a066e2013-09-29 22:06:07 -0400610 if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) {
Al Viro8b61e742013-11-08 12:45:01 -0500611 rcu_read_unlock();
Al Viro48a066e2013-09-29 22:06:07 -0400612 return -ECHILD;
613 }
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700614 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) {
Al Viro8b61e742013-11-08 12:45:01 -0500615 rcu_read_unlock();
Al Viro48a066e2013-09-29 22:06:07 -0400616 mntput(nd->path.mnt);
Al Viro9f1fafe2011-03-25 11:00:12 -0400617 return -ECHILD;
618 }
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700619 if (read_seqcount_retry(&dentry->d_seq, nd->seq)) {
Al Viro8b61e742013-11-08 12:45:01 -0500620 rcu_read_unlock();
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700621 dput(dentry);
Al Viro48a066e2013-09-29 22:06:07 -0400622 mntput(nd->path.mnt);
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700623 return -ECHILD;
624 }
Al Viro8b61e742013-11-08 12:45:01 -0500625 rcu_read_unlock();
Al Viro9f1fafe2011-03-25 11:00:12 -0400626 }
627
Al Viro16c2cd72011-02-22 15:50:10 -0500628 if (likely(!(nd->flags & LOOKUP_JUMPED)))
Jeff Layton39159de2009-12-07 12:01:50 -0500629 return 0;
630
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500631 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
Al Viro16c2cd72011-02-22 15:50:10 -0500632 return 0;
633
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500634 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
Jeff Layton39159de2009-12-07 12:01:50 -0500635 if (status > 0)
636 return 0;
637
Al Viro16c2cd72011-02-22 15:50:10 -0500638 if (!status)
Jeff Layton39159de2009-12-07 12:01:50 -0500639 status = -ESTALE;
Al Viro16c2cd72011-02-22 15:50:10 -0500640
Al Viro9f1fafe2011-03-25 11:00:12 -0400641 path_put(&nd->path);
Jeff Layton39159de2009-12-07 12:01:50 -0500642 return status;
643}
644
Al Viro2a737872009-04-07 11:49:53 -0400645static __always_inline void set_root(struct nameidata *nd)
646{
Al Viro7bd88372014-09-13 21:55:46 -0400647 get_fs_root(current->fs, &nd->root);
Al Viro2a737872009-04-07 11:49:53 -0400648}
649
Al Viro6de88d72009-08-09 01:41:57 +0400650static int link_path_walk(const char *, struct nameidata *);
651
Al Viro7bd88372014-09-13 21:55:46 -0400652static __always_inline unsigned set_root_rcu(struct nameidata *nd)
Nick Piggin31e6b012011-01-07 17:49:52 +1100653{
Al Viro7bd88372014-09-13 21:55:46 -0400654 struct fs_struct *fs = current->fs;
655 unsigned seq, res;
Nick Pigginc28cc362011-01-07 17:49:53 +1100656
Al Viro7bd88372014-09-13 21:55:46 -0400657 do {
658 seq = read_seqcount_begin(&fs->seq);
659 nd->root = fs->root;
660 res = __read_seqcount_begin(&nd->root.dentry->d_seq);
661 } while (read_seqcount_retry(&fs->seq, seq));
662 return res;
Nick Piggin31e6b012011-01-07 17:49:52 +1100663}
664
Jan Blunck1d957f92008-02-14 19:34:35 -0800665static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700666{
667 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800668 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700669 mntput(path->mnt);
670}
671
Nick Piggin7b9337a2011-01-14 08:42:43 +0000672static inline void path_to_nameidata(const struct path *path,
673 struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700674{
Nick Piggin31e6b012011-01-07 17:49:52 +1100675 if (!(nd->flags & LOOKUP_RCU)) {
676 dput(nd->path.dentry);
677 if (nd->path.mnt != path->mnt)
678 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800679 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100680 nd->path.mnt = path->mnt;
Jan Blunck4ac91372008-02-14 19:34:32 -0800681 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700682}
683
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400684/*
685 * Helper to directly jump to a known parsed path from ->follow_link,
686 * caller must have taken a reference to path beforehand.
687 */
688void nd_jump_link(struct nameidata *nd, struct path *path)
689{
690 path_put(&nd->path);
691
692 nd->path = *path;
693 nd->inode = nd->path.dentry->d_inode;
694 nd->flags |= LOOKUP_JUMPED;
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400695}
696
Al Viro574197e2011-03-14 22:20:34 -0400697static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
698{
699 struct inode *inode = link->dentry->d_inode;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400700 if (inode->i_op->put_link)
Al Viro574197e2011-03-14 22:20:34 -0400701 inode->i_op->put_link(link->dentry, nd, cookie);
702 path_put(link);
703}
704
Linus Torvalds561ec642012-10-26 10:05:07 -0700705int sysctl_protected_symlinks __read_mostly = 0;
706int sysctl_protected_hardlinks __read_mostly = 0;
Kees Cook800179c2012-07-25 17:29:07 -0700707
708/**
709 * may_follow_link - Check symlink following for unsafe situations
710 * @link: The path of the symlink
Randy Dunlap55852632012-08-18 17:39:25 -0700711 * @nd: nameidata pathwalk data
Kees Cook800179c2012-07-25 17:29:07 -0700712 *
713 * In the case of the sysctl_protected_symlinks sysctl being enabled,
714 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
715 * in a sticky world-writable directory. This is to protect privileged
716 * processes from failing races against path names that may change out
717 * from under them by way of other users creating malicious symlinks.
718 * It will permit symlinks to be followed only when outside a sticky
719 * world-writable directory, or when the uid of the symlink and follower
720 * match, or when the directory owner matches the symlink's owner.
721 *
722 * Returns 0 if following the symlink is allowed, -ve on error.
723 */
724static inline int may_follow_link(struct path *link, struct nameidata *nd)
725{
726 const struct inode *inode;
727 const struct inode *parent;
728
729 if (!sysctl_protected_symlinks)
730 return 0;
731
732 /* Allowed if owner and follower match. */
733 inode = link->dentry->d_inode;
Eric W. Biederman81abe272012-08-03 09:38:08 -0700734 if (uid_eq(current_cred()->fsuid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700735 return 0;
736
737 /* Allowed if parent directory not sticky and world-writable. */
738 parent = nd->path.dentry->d_inode;
739 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
740 return 0;
741
742 /* Allowed if parent directory and link owner match. */
Eric W. Biederman81abe272012-08-03 09:38:08 -0700743 if (uid_eq(parent->i_uid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700744 return 0;
745
Sasha Levinffd8d102012-10-04 19:56:40 -0400746 audit_log_link_denied("follow_link", link);
Kees Cook800179c2012-07-25 17:29:07 -0700747 path_put_conditional(link, nd);
748 path_put(&nd->path);
749 return -EACCES;
750}
751
752/**
753 * safe_hardlink_source - Check for safe hardlink conditions
754 * @inode: the source inode to hardlink from
755 *
756 * Return false if at least one of the following conditions:
757 * - inode is not a regular file
758 * - inode is setuid
759 * - inode is setgid and group-exec
760 * - access failure for read and write
761 *
762 * Otherwise returns true.
763 */
764static bool safe_hardlink_source(struct inode *inode)
765{
766 umode_t mode = inode->i_mode;
767
768 /* Special files should not get pinned to the filesystem. */
769 if (!S_ISREG(mode))
770 return false;
771
772 /* Setuid files should not get pinned to the filesystem. */
773 if (mode & S_ISUID)
774 return false;
775
776 /* Executable setgid files should not get pinned to the filesystem. */
777 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
778 return false;
779
780 /* Hardlinking to unreadable or unwritable sources is dangerous. */
781 if (inode_permission(inode, MAY_READ | MAY_WRITE))
782 return false;
783
784 return true;
785}
786
787/**
788 * may_linkat - Check permissions for creating a hardlink
789 * @link: the source to hardlink from
790 *
791 * Block hardlink when all of:
792 * - sysctl_protected_hardlinks enabled
793 * - fsuid does not match inode
794 * - hardlink source is unsafe (see safe_hardlink_source() above)
795 * - not CAP_FOWNER
796 *
797 * Returns 0 if successful, -ve on error.
798 */
799static int may_linkat(struct path *link)
800{
801 const struct cred *cred;
802 struct inode *inode;
803
804 if (!sysctl_protected_hardlinks)
805 return 0;
806
807 cred = current_cred();
808 inode = link->dentry->d_inode;
809
810 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
811 * otherwise, it must be a safe source.
812 */
Eric W. Biederman81abe272012-08-03 09:38:08 -0700813 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
Kees Cook800179c2012-07-25 17:29:07 -0700814 capable(CAP_FOWNER))
815 return 0;
816
Kees Cooka51d9ea2012-07-25 17:29:08 -0700817 audit_log_link_denied("linkat", link);
Kees Cook800179c2012-07-25 17:29:07 -0700818 return -EPERM;
819}
820
Al Virodef4af32009-12-26 08:37:05 -0500821static __always_inline int
Al Viro574197e2011-03-14 22:20:34 -0400822follow_link(struct path *link, struct nameidata *nd, void **p)
Ian Kent051d3812006-03-27 01:14:53 -0800823{
Nick Piggin7b9337a2011-01-14 08:42:43 +0000824 struct dentry *dentry = link->dentry;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400825 int error;
826 char *s;
Ian Kent051d3812006-03-27 01:14:53 -0800827
Al Viro844a3912011-02-15 00:38:26 -0500828 BUG_ON(nd->flags & LOOKUP_RCU);
829
Al Viro0e794582011-03-16 02:45:02 -0400830 if (link->mnt == nd->path.mnt)
831 mntget(link->mnt);
832
Al Viro6d7b5aa2012-06-10 04:15:17 -0400833 error = -ELOOP;
834 if (unlikely(current->total_link_count >= 40))
835 goto out_put_nd_path;
836
Al Viro574197e2011-03-14 22:20:34 -0400837 cond_resched();
838 current->total_link_count++;
839
Al Viro68ac1232012-03-15 08:21:57 -0400840 touch_atime(link);
Ian Kent051d3812006-03-27 01:14:53 -0800841 nd_set_link(nd, NULL);
842
Al Viro36f3b4f2011-02-22 21:24:38 -0500843 error = security_inode_follow_link(link->dentry, nd);
Al Viro6d7b5aa2012-06-10 04:15:17 -0400844 if (error)
845 goto out_put_nd_path;
Al Viro36f3b4f2011-02-22 21:24:38 -0500846
Al Viro86acdca12009-12-22 23:45:11 -0500847 nd->last_type = LAST_BIND;
Al Virodef4af32009-12-26 08:37:05 -0500848 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
849 error = PTR_ERR(*p);
Al Viro6d7b5aa2012-06-10 04:15:17 -0400850 if (IS_ERR(*p))
Christoph Hellwig408ef012012-06-18 10:47:03 -0400851 goto out_put_nd_path;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400852
853 error = 0;
854 s = nd_get_link(nd);
855 if (s) {
Al Viro443ed252013-09-10 12:00:43 -0400856 if (unlikely(IS_ERR(s))) {
857 path_put(&nd->path);
858 put_link(nd, link, *p);
859 return PTR_ERR(s);
860 }
861 if (*s == '/') {
Al Viro7bd88372014-09-13 21:55:46 -0400862 if (!nd->root.mnt)
863 set_root(nd);
Al Viro443ed252013-09-10 12:00:43 -0400864 path_put(&nd->path);
865 nd->path = nd->root;
866 path_get(&nd->root);
867 nd->flags |= LOOKUP_JUMPED;
868 }
869 nd->inode = nd->path.dentry->d_inode;
870 error = link_path_walk(s, nd);
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400871 if (unlikely(error))
872 put_link(nd, link, *p);
Ian Kent051d3812006-03-27 01:14:53 -0800873 }
Al Viro6d7b5aa2012-06-10 04:15:17 -0400874
875 return error;
876
877out_put_nd_path:
Arnd Bergmann98f6ef62012-10-11 13:20:00 +0000878 *p = NULL;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400879 path_put(&nd->path);
Al Viro6d7b5aa2012-06-10 04:15:17 -0400880 path_put(link);
Ian Kent051d3812006-03-27 01:14:53 -0800881 return error;
882}
883
Nick Piggin31e6b012011-01-07 17:49:52 +1100884static int follow_up_rcu(struct path *path)
885{
Al Viro0714a532011-11-24 22:19:58 -0500886 struct mount *mnt = real_mount(path->mnt);
887 struct mount *parent;
Nick Piggin31e6b012011-01-07 17:49:52 +1100888 struct dentry *mountpoint;
889
Al Viro0714a532011-11-24 22:19:58 -0500890 parent = mnt->mnt_parent;
891 if (&parent->mnt == path->mnt)
Nick Piggin31e6b012011-01-07 17:49:52 +1100892 return 0;
Al Viroa73324d2011-11-24 22:25:07 -0500893 mountpoint = mnt->mnt_mountpoint;
Nick Piggin31e6b012011-01-07 17:49:52 +1100894 path->dentry = mountpoint;
Al Viro0714a532011-11-24 22:19:58 -0500895 path->mnt = &parent->mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +1100896 return 1;
897}
898
David Howellsf015f1262012-06-25 12:55:28 +0100899/*
900 * follow_up - Find the mountpoint of path's vfsmount
901 *
902 * Given a path, find the mountpoint of its source file system.
903 * Replace @path with the path of the mountpoint in the parent mount.
904 * Up is towards /.
905 *
906 * Return 1 if we went up a level and 0 if we were already at the
907 * root.
908 */
Al Virobab77eb2009-04-18 03:26:48 -0400909int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910{
Al Viro0714a532011-11-24 22:19:58 -0500911 struct mount *mnt = real_mount(path->mnt);
912 struct mount *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +1000914
Al Viro48a066e2013-09-29 22:06:07 -0400915 read_seqlock_excl(&mount_lock);
Al Viro0714a532011-11-24 22:19:58 -0500916 parent = mnt->mnt_parent;
Al Viro3c0a6162012-07-18 17:32:50 +0400917 if (parent == mnt) {
Al Viro48a066e2013-09-29 22:06:07 -0400918 read_sequnlock_excl(&mount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 return 0;
920 }
Al Viro0714a532011-11-24 22:19:58 -0500921 mntget(&parent->mnt);
Al Viroa73324d2011-11-24 22:25:07 -0500922 mountpoint = dget(mnt->mnt_mountpoint);
Al Viro48a066e2013-09-29 22:06:07 -0400923 read_sequnlock_excl(&mount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400924 dput(path->dentry);
925 path->dentry = mountpoint;
926 mntput(path->mnt);
Al Viro0714a532011-11-24 22:19:58 -0500927 path->mnt = &parent->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 return 1;
929}
Al Viro4d359502014-03-14 12:20:17 -0400930EXPORT_SYMBOL(follow_up);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100932/*
David Howells9875cf82011-01-14 18:45:21 +0000933 * Perform an automount
934 * - return -EISDIR to tell follow_managed() to stop and return the path we
935 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 */
David Howells9875cf82011-01-14 18:45:21 +0000937static int follow_automount(struct path *path, unsigned flags,
938 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +1100939{
David Howells9875cf82011-01-14 18:45:21 +0000940 struct vfsmount *mnt;
David Howellsea5b7782011-01-14 19:10:03 +0000941 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +1100942
David Howells9875cf82011-01-14 18:45:21 +0000943 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
944 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -0700945
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200946 /* We don't want to mount if someone's just doing a stat -
947 * unless they're stat'ing a directory and appended a '/' to
948 * the name.
949 *
950 * We do, however, want to mount if someone wants to open or
951 * create a file of any type under the mountpoint, wants to
952 * traverse through the mountpoint or wants to open the
953 * mounted directory. Also, autofs may mark negative dentries
954 * as being automount points. These will need the attentions
955 * of the daemon to instantiate them before they can be used.
David Howells9875cf82011-01-14 18:45:21 +0000956 */
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200957 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
Linus Torvaldsd94c1772011-09-26 17:44:55 -0700958 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200959 path->dentry->d_inode)
960 return -EISDIR;
961
David Howells9875cf82011-01-14 18:45:21 +0000962 current->total_link_count++;
963 if (current->total_link_count >= 40)
964 return -ELOOP;
965
966 mnt = path->dentry->d_op->d_automount(path);
967 if (IS_ERR(mnt)) {
968 /*
969 * The filesystem is allowed to return -EISDIR here to indicate
970 * it doesn't want to automount. For instance, autofs would do
971 * this so that its userspace daemon can mount on this dentry.
972 *
973 * However, we can only permit this if it's a terminal point in
974 * the path being looked up; if it wasn't then the remainder of
975 * the path is inaccessible and we should say so.
976 */
Al Viro49084c32011-06-25 21:59:52 -0400977 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
David Howells9875cf82011-01-14 18:45:21 +0000978 return -EREMOTE;
979 return PTR_ERR(mnt);
980 }
David Howellsea5b7782011-01-14 19:10:03 +0000981
David Howells9875cf82011-01-14 18:45:21 +0000982 if (!mnt) /* mount collision */
983 return 0;
984
Al Viro8aef1882011-06-16 15:10:06 +0100985 if (!*need_mntput) {
986 /* lock_mount() may release path->mnt on error */
987 mntget(path->mnt);
988 *need_mntput = true;
989 }
Al Viro19a167a2011-01-17 01:35:23 -0500990 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +0000991
David Howellsea5b7782011-01-14 19:10:03 +0000992 switch (err) {
993 case -EBUSY:
994 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -0500995 return 0;
David Howellsea5b7782011-01-14 19:10:03 +0000996 case 0:
Al Viro8aef1882011-06-16 15:10:06 +0100997 path_put(path);
David Howellsea5b7782011-01-14 19:10:03 +0000998 path->mnt = mnt;
999 path->dentry = dget(mnt->mnt_root);
David Howellsea5b7782011-01-14 19:10:03 +00001000 return 0;
Al Viro19a167a2011-01-17 01:35:23 -05001001 default:
1002 return err;
David Howellsea5b7782011-01-14 19:10:03 +00001003 }
Al Viro19a167a2011-01-17 01:35:23 -05001004
David Howells9875cf82011-01-14 18:45:21 +00001005}
1006
1007/*
1008 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +00001009 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +00001010 * - Flagged as mountpoint
1011 * - Flagged as automount point
1012 *
1013 * This may only be called in refwalk mode.
1014 *
1015 * Serialization is taken care of in namespace.c
1016 */
1017static int follow_managed(struct path *path, unsigned flags)
1018{
Al Viro8aef1882011-06-16 15:10:06 +01001019 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
David Howells9875cf82011-01-14 18:45:21 +00001020 unsigned managed;
1021 bool need_mntput = false;
Al Viro8aef1882011-06-16 15:10:06 +01001022 int ret = 0;
David Howells9875cf82011-01-14 18:45:21 +00001023
1024 /* Given that we're not holding a lock here, we retain the value in a
1025 * local variable for each dentry as we look at it so that we don't see
1026 * the components of that value change under us */
1027 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1028 managed &= DCACHE_MANAGED_DENTRY,
1029 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +00001030 /* Allow the filesystem to manage the transit without i_mutex
1031 * being held. */
1032 if (managed & DCACHE_MANAGE_TRANSIT) {
1033 BUG_ON(!path->dentry->d_op);
1034 BUG_ON(!path->dentry->d_op->d_manage);
Al Viro1aed3e42011-03-18 09:09:02 -04001035 ret = path->dentry->d_op->d_manage(path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001036 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001037 break;
David Howellscc53ce52011-01-14 18:45:26 +00001038 }
1039
David Howells9875cf82011-01-14 18:45:21 +00001040 /* Transit to a mounted filesystem. */
1041 if (managed & DCACHE_MOUNTED) {
1042 struct vfsmount *mounted = lookup_mnt(path);
1043 if (mounted) {
1044 dput(path->dentry);
1045 if (need_mntput)
1046 mntput(path->mnt);
1047 path->mnt = mounted;
1048 path->dentry = dget(mounted->mnt_root);
1049 need_mntput = true;
1050 continue;
1051 }
1052
1053 /* Something is mounted on this dentry in another
1054 * namespace and/or whatever was mounted there in this
Al Viro48a066e2013-09-29 22:06:07 -04001055 * namespace got unmounted before lookup_mnt() could
1056 * get it */
David Howells9875cf82011-01-14 18:45:21 +00001057 }
1058
1059 /* Handle an automount point */
1060 if (managed & DCACHE_NEED_AUTOMOUNT) {
1061 ret = follow_automount(path, flags, &need_mntput);
1062 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001063 break;
David Howells9875cf82011-01-14 18:45:21 +00001064 continue;
1065 }
1066
1067 /* We didn't change the current path point */
1068 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 }
Al Viro8aef1882011-06-16 15:10:06 +01001070
1071 if (need_mntput && path->mnt == mnt)
1072 mntput(path->mnt);
1073 if (ret == -EISDIR)
1074 ret = 0;
Al Viroa3fbbde2011-11-07 21:21:26 +00001075 return ret < 0 ? ret : need_mntput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076}
1077
David Howellscc53ce52011-01-14 18:45:26 +00001078int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079{
1080 struct vfsmount *mounted;
1081
Al Viro1c755af2009-04-18 14:06:57 -04001082 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -04001084 dput(path->dentry);
1085 mntput(path->mnt);
1086 path->mnt = mounted;
1087 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 return 1;
1089 }
1090 return 0;
1091}
Al Viro4d359502014-03-14 12:20:17 -04001092EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093
NeilBrownb8faf032014-08-04 17:06:29 +10001094static inline int managed_dentry_rcu(struct dentry *dentry)
Ian Kent62a73752011-03-25 01:51:02 +08001095{
NeilBrownb8faf032014-08-04 17:06:29 +10001096 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1097 dentry->d_op->d_manage(dentry, true) : 0;
Ian Kent62a73752011-03-25 01:51:02 +08001098}
1099
David Howells9875cf82011-01-14 18:45:21 +00001100/*
Al Viro287548e2011-05-27 06:50:06 -04001101 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1102 * we meet a managed dentry that would need blocking.
David Howells9875cf82011-01-14 18:45:21 +00001103 */
1104static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
Al Viro287548e2011-05-27 06:50:06 -04001105 struct inode **inode)
David Howells9875cf82011-01-14 18:45:21 +00001106{
Ian Kent62a73752011-03-25 01:51:02 +08001107 for (;;) {
Al Viroc7105362011-11-24 18:22:03 -05001108 struct mount *mounted;
Ian Kent62a73752011-03-25 01:51:02 +08001109 /*
1110 * Don't forget we might have a non-mountpoint managed dentry
1111 * that wants to block transit.
1112 */
NeilBrownb8faf032014-08-04 17:06:29 +10001113 switch (managed_dentry_rcu(path->dentry)) {
1114 case -ECHILD:
1115 default:
David Howellsab909112011-01-14 18:46:51 +00001116 return false;
NeilBrownb8faf032014-08-04 17:06:29 +10001117 case -EISDIR:
1118 return true;
1119 case 0:
1120 break;
1121 }
Ian Kent62a73752011-03-25 01:51:02 +08001122
1123 if (!d_mountpoint(path->dentry))
NeilBrownb8faf032014-08-04 17:06:29 +10001124 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
Ian Kent62a73752011-03-25 01:51:02 +08001125
Al Viro474279d2013-10-01 16:11:26 -04001126 mounted = __lookup_mnt(path->mnt, path->dentry);
David Howells9875cf82011-01-14 18:45:21 +00001127 if (!mounted)
1128 break;
Al Viroc7105362011-11-24 18:22:03 -05001129 path->mnt = &mounted->mnt;
1130 path->dentry = mounted->mnt.mnt_root;
Al Viroa3fbbde2011-11-07 21:21:26 +00001131 nd->flags |= LOOKUP_JUMPED;
David Howells9875cf82011-01-14 18:45:21 +00001132 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
Linus Torvalds59430262011-07-18 15:43:29 -07001133 /*
1134 * Update the inode too. We don't need to re-check the
1135 * dentry sequence number here after this d_inode read,
1136 * because a mount-point is always pinned.
1137 */
1138 *inode = path->dentry->d_inode;
David Howells9875cf82011-01-14 18:45:21 +00001139 }
Al Virof5be3e29122014-09-13 21:50:45 -04001140 return !read_seqretry(&mount_lock, nd->m_seq) &&
NeilBrownb8faf032014-08-04 17:06:29 +10001141 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
Al Viro287548e2011-05-27 06:50:06 -04001142}
1143
Nick Piggin31e6b012011-01-07 17:49:52 +11001144static int follow_dotdot_rcu(struct nameidata *nd)
1145{
Al Viro4023bfc2014-09-13 21:59:43 -04001146 struct inode *inode = nd->inode;
Al Viro7bd88372014-09-13 21:55:46 -04001147 if (!nd->root.mnt)
1148 set_root_rcu(nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11001149
David Howells9875cf82011-01-14 18:45:21 +00001150 while (1) {
Nick Piggin31e6b012011-01-07 17:49:52 +11001151 if (nd->path.dentry == nd->root.dentry &&
1152 nd->path.mnt == nd->root.mnt) {
1153 break;
1154 }
1155 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1156 struct dentry *old = nd->path.dentry;
1157 struct dentry *parent = old->d_parent;
1158 unsigned seq;
1159
Al Viro4023bfc2014-09-13 21:59:43 -04001160 inode = parent->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001161 seq = read_seqcount_begin(&parent->d_seq);
1162 if (read_seqcount_retry(&old->d_seq, nd->seq))
Al Viroef7562d2011-03-04 14:35:59 -05001163 goto failed;
Nick Piggin31e6b012011-01-07 17:49:52 +11001164 nd->path.dentry = parent;
1165 nd->seq = seq;
1166 break;
1167 }
1168 if (!follow_up_rcu(&nd->path))
1169 break;
Al Viro4023bfc2014-09-13 21:59:43 -04001170 inode = nd->path.dentry->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001171 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11001172 }
Al Virob37199e2014-03-20 15:18:22 -04001173 while (d_mountpoint(nd->path.dentry)) {
1174 struct mount *mounted;
1175 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1176 if (!mounted)
1177 break;
1178 nd->path.mnt = &mounted->mnt;
1179 nd->path.dentry = mounted->mnt.mnt_root;
Al Viro4023bfc2014-09-13 21:59:43 -04001180 inode = nd->path.dentry->d_inode;
Al Virob37199e2014-03-20 15:18:22 -04001181 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Virof5be3e29122014-09-13 21:50:45 -04001182 if (read_seqretry(&mount_lock, nd->m_seq))
Al Virob37199e2014-03-20 15:18:22 -04001183 goto failed;
1184 }
Al Viro4023bfc2014-09-13 21:59:43 -04001185 nd->inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001186 return 0;
Al Viroef7562d2011-03-04 14:35:59 -05001187
1188failed:
1189 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001190 if (!(nd->flags & LOOKUP_ROOT))
1191 nd->root.mnt = NULL;
Al Viro8b61e742013-11-08 12:45:01 -05001192 rcu_read_unlock();
Al Viroef7562d2011-03-04 14:35:59 -05001193 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001194}
1195
David Howells9875cf82011-01-14 18:45:21 +00001196/*
David Howellscc53ce52011-01-14 18:45:26 +00001197 * Follow down to the covering mount currently visible to userspace. At each
1198 * point, the filesystem owning that dentry may be queried as to whether the
1199 * caller is permitted to proceed or not.
David Howellscc53ce52011-01-14 18:45:26 +00001200 */
Al Viro7cc90cc2011-03-18 09:04:20 -04001201int follow_down(struct path *path)
David Howellscc53ce52011-01-14 18:45:26 +00001202{
1203 unsigned managed;
1204 int ret;
1205
1206 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1207 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1208 /* Allow the filesystem to manage the transit without i_mutex
1209 * being held.
1210 *
1211 * We indicate to the filesystem if someone is trying to mount
1212 * something here. This gives autofs the chance to deny anyone
1213 * other than its daemon the right to mount on its
1214 * superstructure.
1215 *
1216 * The filesystem may sleep at this point.
1217 */
1218 if (managed & DCACHE_MANAGE_TRANSIT) {
1219 BUG_ON(!path->dentry->d_op);
1220 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +00001221 ret = path->dentry->d_op->d_manage(
Al Viro1aed3e42011-03-18 09:09:02 -04001222 path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001223 if (ret < 0)
1224 return ret == -EISDIR ? 0 : ret;
1225 }
1226
1227 /* Transit to a mounted filesystem. */
1228 if (managed & DCACHE_MOUNTED) {
1229 struct vfsmount *mounted = lookup_mnt(path);
1230 if (!mounted)
1231 break;
1232 dput(path->dentry);
1233 mntput(path->mnt);
1234 path->mnt = mounted;
1235 path->dentry = dget(mounted->mnt_root);
1236 continue;
1237 }
1238
1239 /* Don't handle automount points here */
1240 break;
1241 }
1242 return 0;
1243}
Al Viro4d359502014-03-14 12:20:17 -04001244EXPORT_SYMBOL(follow_down);
David Howellscc53ce52011-01-14 18:45:26 +00001245
1246/*
David Howells9875cf82011-01-14 18:45:21 +00001247 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1248 */
1249static void follow_mount(struct path *path)
1250{
1251 while (d_mountpoint(path->dentry)) {
1252 struct vfsmount *mounted = lookup_mnt(path);
1253 if (!mounted)
1254 break;
1255 dput(path->dentry);
1256 mntput(path->mnt);
1257 path->mnt = mounted;
1258 path->dentry = dget(mounted->mnt_root);
1259 }
1260}
1261
Nick Piggin31e6b012011-01-07 17:49:52 +11001262static void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263{
Al Viro7bd88372014-09-13 21:55:46 -04001264 if (!nd->root.mnt)
1265 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001266
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 while(1) {
Jan Blunck4ac91372008-02-14 19:34:32 -08001268 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269
Al Viro2a737872009-04-07 11:49:53 -04001270 if (nd->path.dentry == nd->root.dentry &&
1271 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 break;
1273 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001274 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Al Viro3088dd72010-01-30 15:47:29 -05001275 /* rare case of legitimate dget_parent()... */
1276 nd->path.dentry = dget_parent(nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 dput(old);
1278 break;
1279 }
Al Viro3088dd72010-01-30 15:47:29 -05001280 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 }
Al Viro79ed0222009-04-18 13:59:41 -04001283 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001284 nd->inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285}
1286
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287/*
Miklos Szeredibad61182012-03-26 12:54:24 +02001288 * This looks up the name in dcache, possibly revalidates the old dentry and
1289 * allocates a new one if not found or not valid. In the need_lookup argument
1290 * returns whether i_op->lookup is necessary.
1291 *
1292 * dir->d_inode->i_mutex must be held
Nick Pigginbaa03892010-08-18 04:37:31 +10001293 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001294static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
Al Viro201f9562012-06-22 12:42:10 +04001295 unsigned int flags, bool *need_lookup)
Nick Pigginbaa03892010-08-18 04:37:31 +10001296{
Nick Pigginbaa03892010-08-18 04:37:31 +10001297 struct dentry *dentry;
Miklos Szeredibad61182012-03-26 12:54:24 +02001298 int error;
Nick Pigginbaa03892010-08-18 04:37:31 +10001299
Miklos Szeredibad61182012-03-26 12:54:24 +02001300 *need_lookup = false;
1301 dentry = d_lookup(dir, name);
1302 if (dentry) {
Jeff Layton39e3c952012-11-28 11:30:53 -05001303 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
Al Viro201f9562012-06-22 12:42:10 +04001304 error = d_revalidate(dentry, flags);
Miklos Szeredibad61182012-03-26 12:54:24 +02001305 if (unlikely(error <= 0)) {
1306 if (error < 0) {
1307 dput(dentry);
1308 return ERR_PTR(error);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001309 } else {
1310 d_invalidate(dentry);
Miklos Szeredibad61182012-03-26 12:54:24 +02001311 dput(dentry);
1312 dentry = NULL;
1313 }
1314 }
1315 }
1316 }
Nick Pigginbaa03892010-08-18 04:37:31 +10001317
Miklos Szeredibad61182012-03-26 12:54:24 +02001318 if (!dentry) {
1319 dentry = d_alloc(dir, name);
1320 if (unlikely(!dentry))
1321 return ERR_PTR(-ENOMEM);
Nick Pigginbaa03892010-08-18 04:37:31 +10001322
Miklos Szeredibad61182012-03-26 12:54:24 +02001323 *need_lookup = true;
Nick Pigginbaa03892010-08-18 04:37:31 +10001324 }
1325 return dentry;
1326}
1327
1328/*
J. Bruce Fields13a2c3b2013-10-23 16:09:16 -04001329 * Call i_op->lookup on the dentry. The dentry must be negative and
1330 * unhashed.
Miklos Szeredibad61182012-03-26 12:54:24 +02001331 *
1332 * dir->d_inode->i_mutex must be held
Josef Bacik44396f42011-05-31 11:58:49 -04001333 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001334static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
Al Viro72bd8662012-06-10 17:17:17 -04001335 unsigned int flags)
Josef Bacik44396f42011-05-31 11:58:49 -04001336{
Josef Bacik44396f42011-05-31 11:58:49 -04001337 struct dentry *old;
1338
1339 /* Don't create child dentry for a dead directory. */
Miklos Szeredibad61182012-03-26 12:54:24 +02001340 if (unlikely(IS_DEADDIR(dir))) {
Miklos Szeredie188dc02012-02-03 14:25:18 +01001341 dput(dentry);
Josef Bacik44396f42011-05-31 11:58:49 -04001342 return ERR_PTR(-ENOENT);
Miklos Szeredie188dc02012-02-03 14:25:18 +01001343 }
Josef Bacik44396f42011-05-31 11:58:49 -04001344
Al Viro72bd8662012-06-10 17:17:17 -04001345 old = dir->i_op->lookup(dir, dentry, flags);
Josef Bacik44396f42011-05-31 11:58:49 -04001346 if (unlikely(old)) {
1347 dput(dentry);
1348 dentry = old;
1349 }
1350 return dentry;
1351}
1352
Al Viroa3255542012-03-30 14:41:51 -04001353static struct dentry *__lookup_hash(struct qstr *name,
Al Viro72bd8662012-06-10 17:17:17 -04001354 struct dentry *base, unsigned int flags)
Al Viroa3255542012-03-30 14:41:51 -04001355{
Miklos Szeredibad61182012-03-26 12:54:24 +02001356 bool need_lookup;
Al Viroa3255542012-03-30 14:41:51 -04001357 struct dentry *dentry;
1358
Al Viro72bd8662012-06-10 17:17:17 -04001359 dentry = lookup_dcache(name, base, flags, &need_lookup);
Miklos Szeredibad61182012-03-26 12:54:24 +02001360 if (!need_lookup)
1361 return dentry;
Al Viroa3255542012-03-30 14:41:51 -04001362
Al Viro72bd8662012-06-10 17:17:17 -04001363 return lookup_real(base->d_inode, dentry, flags);
Al Viroa3255542012-03-30 14:41:51 -04001364}
1365
Josef Bacik44396f42011-05-31 11:58:49 -04001366/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 * It's more convoluted than I'd like it to be, but... it's still fairly
1368 * small and for now I'd prefer to have fast path as straight as possible.
1369 * It _is_ time-critical.
1370 */
Al Viroe97cdc82013-01-24 18:16:00 -05001371static int lookup_fast(struct nameidata *nd,
Miklos Szeredi697f5142012-05-21 17:30:05 +02001372 struct path *path, struct inode **inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373{
Jan Blunck4ac91372008-02-14 19:34:32 -08001374 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001375 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001376 int need_reval = 1;
1377 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001378 int err;
1379
Al Viro3cac2602009-08-13 18:27:43 +04001380 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001381 * Rename seqlock is not required here because in the off chance
1382 * of a false negative due to a concurrent rename, we're going to
1383 * do the non-racy lookup, below.
1384 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001385 if (nd->flags & LOOKUP_RCU) {
1386 unsigned seq;
Linus Torvaldsda53be12013-05-21 15:22:44 -07001387 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
Al Viro5a18fff2011-03-11 04:44:53 -05001388 if (!dentry)
1389 goto unlazy;
1390
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001391 /*
1392 * This sequence count validates that the inode matches
1393 * the dentry name information from lookup.
1394 */
1395 *inode = dentry->d_inode;
1396 if (read_seqcount_retry(&dentry->d_seq, seq))
1397 return -ECHILD;
1398
1399 /*
1400 * This sequence count validates that the parent had no
1401 * changes while we did the lookup of the dentry above.
1402 *
1403 * The memory barrier in read_seqcount_begin of child is
1404 * enough, we can use __read_seqcount_retry here.
1405 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001406 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1407 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001408 nd->seq = seq;
Al Viro5a18fff2011-03-11 04:44:53 -05001409
Al Viro24643082011-02-15 01:26:22 -05001410 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
Al Viro4ce16ef32012-06-10 16:10:59 -04001411 status = d_revalidate(dentry, nd->flags);
Al Viro5a18fff2011-03-11 04:44:53 -05001412 if (unlikely(status <= 0)) {
1413 if (status != -ECHILD)
1414 need_reval = 0;
1415 goto unlazy;
1416 }
Al Viro24643082011-02-15 01:26:22 -05001417 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001418 path->mnt = mnt;
1419 path->dentry = dentry;
NeilBrownb8faf032014-08-04 17:06:29 +10001420 if (likely(__follow_mount_rcu(nd, path, inode)))
1421 return 0;
Al Viro5a18fff2011-03-11 04:44:53 -05001422unlazy:
Al Viro19660af2011-03-25 10:32:48 -04001423 if (unlazy_walk(nd, dentry))
1424 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001425 } else {
Al Viroe97cdc82013-01-24 18:16:00 -05001426 dentry = __d_lookup(parent, &nd->last);
Nick Piggin31e6b012011-01-07 17:49:52 +11001427 }
Al Viro5a18fff2011-03-11 04:44:53 -05001428
Al Viro81e6f522012-03-30 14:48:04 -04001429 if (unlikely(!dentry))
1430 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001431
Al Viro5a18fff2011-03-11 04:44:53 -05001432 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
Al Viro4ce16ef32012-06-10 16:10:59 -04001433 status = d_revalidate(dentry, nd->flags);
Al Viro5a18fff2011-03-11 04:44:53 -05001434 if (unlikely(status <= 0)) {
1435 if (status < 0) {
1436 dput(dentry);
1437 return status;
1438 }
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001439 d_invalidate(dentry);
1440 dput(dentry);
1441 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001442 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001443
David Howells9875cf82011-01-14 18:45:21 +00001444 path->mnt = mnt;
1445 path->dentry = dentry;
1446 err = follow_managed(path, nd->flags);
Ian Kent89312212011-01-18 12:06:10 +08001447 if (unlikely(err < 0)) {
1448 path_put_conditional(path, nd);
David Howells9875cf82011-01-14 18:45:21 +00001449 return err;
Ian Kent89312212011-01-18 12:06:10 +08001450 }
Al Viroa3fbbde2011-11-07 21:21:26 +00001451 if (err)
1452 nd->flags |= LOOKUP_JUMPED;
David Howells9875cf82011-01-14 18:45:21 +00001453 *inode = path->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 return 0;
Al Viro81e6f522012-03-30 14:48:04 -04001455
1456need_lookup:
Miklos Szeredi697f5142012-05-21 17:30:05 +02001457 return 1;
1458}
1459
1460/* Fast lookup failed, do it the slow way */
Al Virocc2a5272013-01-24 18:19:49 -05001461static int lookup_slow(struct nameidata *nd, struct path *path)
Miklos Szeredi697f5142012-05-21 17:30:05 +02001462{
1463 struct dentry *dentry, *parent;
1464 int err;
1465
1466 parent = nd->path.dentry;
Al Viro81e6f522012-03-30 14:48:04 -04001467 BUG_ON(nd->inode != parent->d_inode);
1468
1469 mutex_lock(&parent->d_inode->i_mutex);
Al Virocc2a5272013-01-24 18:19:49 -05001470 dentry = __lookup_hash(&nd->last, parent, nd->flags);
Al Viro81e6f522012-03-30 14:48:04 -04001471 mutex_unlock(&parent->d_inode->i_mutex);
1472 if (IS_ERR(dentry))
1473 return PTR_ERR(dentry);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001474 path->mnt = nd->path.mnt;
1475 path->dentry = dentry;
1476 err = follow_managed(path, nd->flags);
1477 if (unlikely(err < 0)) {
1478 path_put_conditional(path, nd);
1479 return err;
1480 }
1481 if (err)
1482 nd->flags |= LOOKUP_JUMPED;
1483 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484}
1485
Al Viro52094c82011-02-21 21:34:47 -05001486static inline int may_lookup(struct nameidata *nd)
1487{
1488 if (nd->flags & LOOKUP_RCU) {
Al Viro4ad5abb2011-06-20 19:57:03 -04001489 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
Al Viro52094c82011-02-21 21:34:47 -05001490 if (err != -ECHILD)
1491 return err;
Al Viro19660af2011-03-25 10:32:48 -04001492 if (unlazy_walk(nd, NULL))
Al Viro52094c82011-02-21 21:34:47 -05001493 return -ECHILD;
1494 }
Al Viro4ad5abb2011-06-20 19:57:03 -04001495 return inode_permission(nd->inode, MAY_EXEC);
Al Viro52094c82011-02-21 21:34:47 -05001496}
1497
Al Viro9856fa12011-03-04 14:22:06 -05001498static inline int handle_dots(struct nameidata *nd, int type)
1499{
1500 if (type == LAST_DOTDOT) {
1501 if (nd->flags & LOOKUP_RCU) {
1502 if (follow_dotdot_rcu(nd))
1503 return -ECHILD;
1504 } else
1505 follow_dotdot(nd);
1506 }
1507 return 0;
1508}
1509
Al Viro951361f2011-03-04 14:44:37 -05001510static void terminate_walk(struct nameidata *nd)
1511{
1512 if (!(nd->flags & LOOKUP_RCU)) {
1513 path_put(&nd->path);
1514 } else {
1515 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001516 if (!(nd->flags & LOOKUP_ROOT))
1517 nd->root.mnt = NULL;
Al Viro8b61e742013-11-08 12:45:01 -05001518 rcu_read_unlock();
Al Viro951361f2011-03-04 14:44:37 -05001519 }
1520}
1521
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001522/*
1523 * Do we need to follow links? We _really_ want to be able
1524 * to do this check without having to look at inode->i_op,
1525 * so we keep a cache of "no, this doesn't need follow_link"
1526 * for the common case.
1527 */
David Howellsb18825a2013-09-12 19:22:53 +01001528static inline int should_follow_link(struct dentry *dentry, int follow)
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001529{
David Howellsb18825a2013-09-12 19:22:53 +01001530 return unlikely(d_is_symlink(dentry)) ? follow : 0;
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001531}
1532
Al Viroce57dfc2011-03-13 19:58:58 -04001533static inline int walk_component(struct nameidata *nd, struct path *path,
Al Viro21b9b072013-01-24 18:10:25 -05001534 int follow)
Al Viroce57dfc2011-03-13 19:58:58 -04001535{
1536 struct inode *inode;
1537 int err;
1538 /*
1539 * "." and ".." are special - ".." especially so because it has
1540 * to be able to know about the current root directory and
1541 * parent relationships.
1542 */
Al Viro21b9b072013-01-24 18:10:25 -05001543 if (unlikely(nd->last_type != LAST_NORM))
1544 return handle_dots(nd, nd->last_type);
Al Viroe97cdc82013-01-24 18:16:00 -05001545 err = lookup_fast(nd, path, &inode);
Al Viroce57dfc2011-03-13 19:58:58 -04001546 if (unlikely(err)) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001547 if (err < 0)
1548 goto out_err;
1549
Al Virocc2a5272013-01-24 18:19:49 -05001550 err = lookup_slow(nd, path);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001551 if (err < 0)
1552 goto out_err;
1553
1554 inode = path->dentry->d_inode;
Al Viroce57dfc2011-03-13 19:58:58 -04001555 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001556 err = -ENOENT;
Al Viro22213312014-04-19 12:30:58 -04001557 if (!inode || d_is_negative(path->dentry))
Miklos Szeredi697f5142012-05-21 17:30:05 +02001558 goto out_path_put;
1559
David Howellsb18825a2013-09-12 19:22:53 +01001560 if (should_follow_link(path->dentry, follow)) {
Al Viro19660af2011-03-25 10:32:48 -04001561 if (nd->flags & LOOKUP_RCU) {
1562 if (unlikely(unlazy_walk(nd, path->dentry))) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001563 err = -ECHILD;
1564 goto out_err;
Al Viro19660af2011-03-25 10:32:48 -04001565 }
1566 }
Al Viroce57dfc2011-03-13 19:58:58 -04001567 BUG_ON(inode != path->dentry->d_inode);
1568 return 1;
1569 }
1570 path_to_nameidata(path, nd);
1571 nd->inode = inode;
1572 return 0;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001573
1574out_path_put:
1575 path_to_nameidata(path, nd);
1576out_err:
1577 terminate_walk(nd);
1578 return err;
Al Viroce57dfc2011-03-13 19:58:58 -04001579}
1580
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581/*
Al Virob3563792011-03-14 21:54:55 -04001582 * This limits recursive symlink follows to 8, while
1583 * limiting consecutive symlinks to 40.
1584 *
1585 * Without that kind of total limit, nasty chains of consecutive
1586 * symlinks can cause almost arbitrarily long lookups.
1587 */
1588static inline int nested_symlink(struct path *path, struct nameidata *nd)
1589{
1590 int res;
1591
Al Virob3563792011-03-14 21:54:55 -04001592 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1593 path_put_conditional(path, nd);
1594 path_put(&nd->path);
1595 return -ELOOP;
1596 }
Erez Zadok1a4022f2011-05-21 01:19:59 -04001597 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
Al Virob3563792011-03-14 21:54:55 -04001598
1599 nd->depth++;
1600 current->link_count++;
1601
1602 do {
1603 struct path link = *path;
1604 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04001605
1606 res = follow_link(&link, nd, &cookie);
Al Viro6d7b5aa2012-06-10 04:15:17 -04001607 if (res)
1608 break;
Al Viro21b9b072013-01-24 18:10:25 -05001609 res = walk_component(nd, path, LOOKUP_FOLLOW);
Al Viro574197e2011-03-14 22:20:34 -04001610 put_link(nd, &link, cookie);
Al Virob3563792011-03-14 21:54:55 -04001611 } while (res > 0);
1612
1613 current->link_count--;
1614 nd->depth--;
1615 return res;
1616}
1617
1618/*
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001619 * We can do the critical dentry name comparison and hashing
1620 * operations one word at a time, but we are limited to:
1621 *
1622 * - Architectures with fast unaligned word accesses. We could
1623 * do a "get_unaligned()" if this helps and is sufficiently
1624 * fast.
1625 *
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001626 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1627 * do not trap on the (extremely unlikely) case of a page
1628 * crossing operation.
1629 *
1630 * - Furthermore, we need an efficient 64-bit compile for the
1631 * 64-bit case in order to generate the "number of bytes in
1632 * the final mask". Again, that could be replaced with a
1633 * efficient population count instruction or similar.
1634 */
1635#ifdef CONFIG_DCACHE_WORD_ACCESS
1636
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001637#include <asm/word-at-a-time.h>
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001638
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001639#ifdef CONFIG_64BIT
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001640
1641static inline unsigned int fold_hash(unsigned long hash)
1642{
Linus Torvalds99d263d2014-09-13 11:30:10 -07001643 return hash_64(hash, 32);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001644}
1645
1646#else /* 32-bit case */
1647
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001648#define fold_hash(x) (x)
1649
1650#endif
1651
1652unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1653{
1654 unsigned long a, mask;
1655 unsigned long hash = 0;
1656
1657 for (;;) {
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001658 a = load_unaligned_zeropad(name);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001659 if (len < sizeof(unsigned long))
1660 break;
1661 hash += a;
Al Virof132c5b2012-03-22 21:59:52 +00001662 hash *= 9;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001663 name += sizeof(unsigned long);
1664 len -= sizeof(unsigned long);
1665 if (!len)
1666 goto done;
1667 }
Will Deacona5c21dc2013-12-12 17:40:21 +00001668 mask = bytemask_from_count(len);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001669 hash += mask & a;
1670done:
1671 return fold_hash(hash);
1672}
1673EXPORT_SYMBOL(full_name_hash);
1674
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001675/*
1676 * Calculate the length and hash of the path component, and
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001677 * return the "hash_len" as the result.
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001678 */
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001679static inline u64 hash_name(const char *name)
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001680{
Linus Torvalds36126f82012-05-26 10:43:17 -07001681 unsigned long a, b, adata, bdata, mask, hash, len;
1682 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001683
1684 hash = a = 0;
1685 len = -sizeof(unsigned long);
1686 do {
1687 hash = (hash + a) * 9;
1688 len += sizeof(unsigned long);
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001689 a = load_unaligned_zeropad(name+len);
Linus Torvalds36126f82012-05-26 10:43:17 -07001690 b = a ^ REPEAT_BYTE('/');
1691 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001692
Linus Torvalds36126f82012-05-26 10:43:17 -07001693 adata = prep_zero_mask(a, adata, &constants);
1694 bdata = prep_zero_mask(b, bdata, &constants);
1695
1696 mask = create_zero_mask(adata | bdata);
1697
1698 hash += a & zero_bytemask(mask);
Linus Torvalds9226b5b2014-09-14 17:28:32 -07001699 len += find_zero(mask);
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001700 return hashlen_create(fold_hash(hash), len);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001701}
1702
1703#else
1704
Linus Torvalds0145acc2012-03-02 14:32:59 -08001705unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1706{
1707 unsigned long hash = init_name_hash();
1708 while (len--)
1709 hash = partial_name_hash(*name++, hash);
1710 return end_name_hash(hash);
1711}
Linus Torvaldsae942ae2012-03-02 19:40:57 -08001712EXPORT_SYMBOL(full_name_hash);
Linus Torvalds0145acc2012-03-02 14:32:59 -08001713
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001714/*
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001715 * We know there's a real path component here of at least
1716 * one character.
1717 */
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001718static inline u64 hash_name(const char *name)
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001719{
1720 unsigned long hash = init_name_hash();
1721 unsigned long len = 0, c;
1722
1723 c = (unsigned char)*name;
1724 do {
1725 len++;
1726 hash = partial_name_hash(c, hash);
1727 c = (unsigned char)name[len];
1728 } while (c && c != '/');
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001729 return hashlen_create(end_name_hash(hash), len);
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001730}
1731
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001732#endif
1733
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001734/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01001736 * This is the basic name resolution function, turning a pathname into
1737 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01001739 * Returns 0 and nd will have valid dentry and mnt on success.
1740 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 */
Al Viro6de88d72009-08-09 01:41:57 +04001742static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743{
1744 struct path next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746
1747 while (*name=='/')
1748 name++;
1749 if (!*name)
Al Viro086e1832011-02-22 20:56:27 -05001750 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 /* At this point we know we have a real path component. */
1753 for(;;) {
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001754 u64 hash_len;
Al Virofe479a52011-02-22 15:10:03 -05001755 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756
Al Viro52094c82011-02-21 21:34:47 -05001757 err = may_lookup(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 if (err)
1759 break;
1760
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001761 hash_len = hash_name(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762
Al Virofe479a52011-02-22 15:10:03 -05001763 type = LAST_NORM;
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001764 if (name[0] == '.') switch (hashlen_len(hash_len)) {
Al Virofe479a52011-02-22 15:10:03 -05001765 case 2:
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001766 if (name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05001767 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05001768 nd->flags |= LOOKUP_JUMPED;
1769 }
Al Virofe479a52011-02-22 15:10:03 -05001770 break;
1771 case 1:
1772 type = LAST_DOT;
1773 }
Al Viro5a202bc2011-03-08 14:17:44 -05001774 if (likely(type == LAST_NORM)) {
1775 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05001776 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05001777 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
James Hogana060dc52014-09-16 13:07:35 +01001778 struct qstr this = { { .hash_len = hash_len }, .name = name };
Linus Torvaldsda53be12013-05-21 15:22:44 -07001779 err = parent->d_op->d_hash(parent, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05001780 if (err < 0)
1781 break;
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001782 hash_len = this.hash_len;
1783 name = this.name;
Al Viro5a202bc2011-03-08 14:17:44 -05001784 }
1785 }
Al Virofe479a52011-02-22 15:10:03 -05001786
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001787 nd->last.hash_len = hash_len;
1788 nd->last.name = name;
Al Viro5f4a6a62013-01-24 18:04:22 -05001789 nd->last_type = type;
1790
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001791 name += hashlen_len(hash_len);
1792 if (!*name)
Al Viro5f4a6a62013-01-24 18:04:22 -05001793 return 0;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001794 /*
1795 * If it wasn't NUL, we know it was '/'. Skip that
1796 * slash, and continue until no more slashes.
1797 */
1798 do {
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001799 name++;
1800 } while (unlikely(*name == '/'));
1801 if (!*name)
Al Viro5f4a6a62013-01-24 18:04:22 -05001802 return 0;
1803
Al Viro21b9b072013-01-24 18:10:25 -05001804 err = walk_component(nd, &next, LOOKUP_FOLLOW);
Al Viroce57dfc2011-03-13 19:58:58 -04001805 if (err < 0)
1806 return err;
Al Virofe479a52011-02-22 15:10:03 -05001807
Al Viroce57dfc2011-03-13 19:58:58 -04001808 if (err) {
Al Virob3563792011-03-14 21:54:55 -04001809 err = nested_symlink(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 if (err)
Al Viroa7472ba2011-03-04 14:39:30 -05001811 return err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001812 }
Miklos Szeredi44b1d532014-04-01 17:08:41 +02001813 if (!d_can_lookup(nd->path.dentry)) {
Al Viro5f4a6a62013-01-24 18:04:22 -05001814 err = -ENOTDIR;
1815 break;
1816 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 }
Al Viro951361f2011-03-04 14:44:37 -05001818 terminate_walk(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 return err;
1820}
1821
Al Viro70e9b352011-03-05 21:12:22 -05001822static int path_init(int dfd, const char *name, unsigned int flags,
1823 struct nameidata *nd, struct file **fp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001825 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826
1827 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro16c2cd72011-02-22 15:50:10 -05001828 nd->flags = flags | LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829 nd->depth = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05001830 if (flags & LOOKUP_ROOT) {
David Howellsb18825a2013-09-12 19:22:53 +01001831 struct dentry *root = nd->root.dentry;
1832 struct inode *inode = root->d_inode;
Al Viro73d049a2011-03-11 12:08:24 -05001833 if (*name) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02001834 if (!d_can_lookup(root))
Al Viro73d049a2011-03-11 12:08:24 -05001835 return -ENOTDIR;
1836 retval = inode_permission(inode, MAY_EXEC);
1837 if (retval)
1838 return retval;
1839 }
Al Viro5b6ca022011-03-09 23:04:47 -05001840 nd->path = nd->root;
1841 nd->inode = inode;
1842 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05001843 rcu_read_lock();
Al Viro5b6ca022011-03-09 23:04:47 -05001844 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro48a066e2013-09-29 22:06:07 -04001845 nd->m_seq = read_seqbegin(&mount_lock);
Al Viro5b6ca022011-03-09 23:04:47 -05001846 } else {
1847 path_get(&nd->path);
1848 }
1849 return 0;
1850 }
1851
Al Viro2a737872009-04-07 11:49:53 -04001852 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853
Al Viro48a066e2013-09-29 22:06:07 -04001854 nd->m_seq = read_seqbegin(&mount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 if (*name=='/') {
Al Viroe41f7d42011-02-22 14:02:58 -05001856 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05001857 rcu_read_lock();
Al Viro7bd88372014-09-13 21:55:46 -04001858 nd->seq = set_root_rcu(nd);
Al Viroe41f7d42011-02-22 14:02:58 -05001859 } else {
1860 set_root(nd);
1861 path_get(&nd->root);
1862 }
Al Viro2a737872009-04-07 11:49:53 -04001863 nd->path = nd->root;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001864 } else if (dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05001865 if (flags & LOOKUP_RCU) {
1866 struct fs_struct *fs = current->fs;
1867 unsigned seq;
1868
Al Viro8b61e742013-11-08 12:45:01 -05001869 rcu_read_lock();
Al Viroe41f7d42011-02-22 14:02:58 -05001870
1871 do {
1872 seq = read_seqcount_begin(&fs->seq);
1873 nd->path = fs->pwd;
1874 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1875 } while (read_seqcount_retry(&fs->seq, seq));
1876 } else {
1877 get_fs_pwd(current->fs, &nd->path);
1878 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001879 } else {
Jeff Layton582aa642012-12-11 08:56:16 -05001880 /* Caller must check execute permissions on the starting path component */
Al Viro2903ff02012-08-28 12:52:22 -04001881 struct fd f = fdget_raw(dfd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001882 struct dentry *dentry;
1883
Al Viro2903ff02012-08-28 12:52:22 -04001884 if (!f.file)
1885 return -EBADF;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001886
Al Viro2903ff02012-08-28 12:52:22 -04001887 dentry = f.file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001888
Al Virof52e0c12011-03-14 18:56:51 -04001889 if (*name) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02001890 if (!d_can_lookup(dentry)) {
Al Viro2903ff02012-08-28 12:52:22 -04001891 fdput(f);
1892 return -ENOTDIR;
1893 }
Al Virof52e0c12011-03-14 18:56:51 -04001894 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001895
Al Viro2903ff02012-08-28 12:52:22 -04001896 nd->path = f.file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05001897 if (flags & LOOKUP_RCU) {
Linus Torvalds9c225f22014-03-03 09:36:58 -08001898 if (f.flags & FDPUT_FPUT)
Al Viro2903ff02012-08-28 12:52:22 -04001899 *fp = f.file;
Al Viroe41f7d42011-02-22 14:02:58 -05001900 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro8b61e742013-11-08 12:45:01 -05001901 rcu_read_lock();
Al Viroe41f7d42011-02-22 14:02:58 -05001902 } else {
Al Viro2903ff02012-08-28 12:52:22 -04001903 path_get(&nd->path);
1904 fdput(f);
Al Viroe41f7d42011-02-22 14:02:58 -05001905 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906 }
Al Viroe41f7d42011-02-22 14:02:58 -05001907
Nick Piggin31e6b012011-01-07 17:49:52 +11001908 nd->inode = nd->path.dentry->d_inode;
Al Viro4023bfc2014-09-13 21:59:43 -04001909 if (!(flags & LOOKUP_RCU))
1910 return 0;
1911 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
1912 return 0;
1913 if (!(nd->flags & LOOKUP_ROOT))
1914 nd->root.mnt = NULL;
1915 rcu_read_unlock();
1916 return -ECHILD;
Al Viro9b4a9b12009-04-07 11:44:16 -04001917}
1918
Al Virobd92d7f2011-03-14 19:54:59 -04001919static inline int lookup_last(struct nameidata *nd, struct path *path)
1920{
1921 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1922 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1923
1924 nd->flags &= ~LOOKUP_PARENT;
Al Viro21b9b072013-01-24 18:10:25 -05001925 return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW);
Al Virobd92d7f2011-03-14 19:54:59 -04001926}
1927
Al Viro9b4a9b12009-04-07 11:44:16 -04001928/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroee0827c2011-02-21 23:38:09 -05001929static int path_lookupat(int dfd, const char *name,
Al Viro9b4a9b12009-04-07 11:44:16 -04001930 unsigned int flags, struct nameidata *nd)
1931{
Al Viro70e9b352011-03-05 21:12:22 -05001932 struct file *base = NULL;
Al Virobd92d7f2011-03-14 19:54:59 -04001933 struct path path;
1934 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001935
1936 /*
1937 * Path walking is largely split up into 2 different synchronisation
1938 * schemes, rcu-walk and ref-walk (explained in
1939 * Documentation/filesystems/path-lookup.txt). These share much of the
1940 * path walk code, but some things particularly setup, cleanup, and
1941 * following mounts are sufficiently divergent that functions are
1942 * duplicated. Typically there is a function foo(), and its RCU
1943 * analogue, foo_rcu().
1944 *
1945 * -ECHILD is the error number of choice (just to avoid clashes) that
1946 * is returned if some aspect of an rcu-walk fails. Such an error must
1947 * be handled by restarting a traditional ref-walk (which will always
1948 * be able to complete).
1949 */
Al Virobd92d7f2011-03-14 19:54:59 -04001950 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
Al Viroee0827c2011-02-21 23:38:09 -05001951
Al Virobd92d7f2011-03-14 19:54:59 -04001952 if (unlikely(err))
Al Viro115cbfd2014-10-11 23:05:52 -04001953 goto out;
Al Viroee0827c2011-02-21 23:38:09 -05001954
1955 current->total_link_count = 0;
Al Virobd92d7f2011-03-14 19:54:59 -04001956 err = link_path_walk(name, nd);
1957
1958 if (!err && !(flags & LOOKUP_PARENT)) {
Al Virobd92d7f2011-03-14 19:54:59 -04001959 err = lookup_last(nd, &path);
1960 while (err > 0) {
1961 void *cookie;
1962 struct path link = path;
Kees Cook800179c2012-07-25 17:29:07 -07001963 err = may_follow_link(&link, nd);
1964 if (unlikely(err))
1965 break;
Al Virobd92d7f2011-03-14 19:54:59 -04001966 nd->flags |= LOOKUP_PARENT;
Al Viro574197e2011-03-14 22:20:34 -04001967 err = follow_link(&link, nd, &cookie);
Al Viro6d7b5aa2012-06-10 04:15:17 -04001968 if (err)
1969 break;
1970 err = lookup_last(nd, &path);
Al Viro574197e2011-03-14 22:20:34 -04001971 put_link(nd, &link, cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04001972 }
1973 }
Al Viroee0827c2011-02-21 23:38:09 -05001974
Al Viro9f1fafe2011-03-25 11:00:12 -04001975 if (!err)
1976 err = complete_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04001977
1978 if (!err && nd->flags & LOOKUP_DIRECTORY) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02001979 if (!d_can_lookup(nd->path.dentry)) {
Al Virobd92d7f2011-03-14 19:54:59 -04001980 path_put(&nd->path);
Al Virobd23a532011-03-23 09:56:30 -04001981 err = -ENOTDIR;
Al Virobd92d7f2011-03-14 19:54:59 -04001982 }
1983 }
Al Viro16c2cd72011-02-22 15:50:10 -05001984
Al Viro115cbfd2014-10-11 23:05:52 -04001985out:
Al Viro70e9b352011-03-05 21:12:22 -05001986 if (base)
1987 fput(base);
Al Viroee0827c2011-02-21 23:38:09 -05001988
Al Viro5b6ca022011-03-09 23:04:47 -05001989 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
Al Viro2a737872009-04-07 11:49:53 -04001990 path_put(&nd->root);
1991 nd->root.mnt = NULL;
1992 }
Al Virobd92d7f2011-03-14 19:54:59 -04001993 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001994}
Nick Piggin31e6b012011-01-07 17:49:52 +11001995
Jeff Layton873f1ee2012-10-10 15:25:29 -04001996static int filename_lookup(int dfd, struct filename *name,
1997 unsigned int flags, struct nameidata *nd)
1998{
1999 int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd);
2000 if (unlikely(retval == -ECHILD))
2001 retval = path_lookupat(dfd, name->name, flags, nd);
2002 if (unlikely(retval == -ESTALE))
2003 retval = path_lookupat(dfd, name->name,
2004 flags | LOOKUP_REVAL, nd);
2005
2006 if (likely(!retval))
Jeff Laytonadb5c242012-10-10 16:43:13 -04002007 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002008 return retval;
2009}
2010
Al Viroee0827c2011-02-21 23:38:09 -05002011static int do_path_lookup(int dfd, const char *name,
2012 unsigned int flags, struct nameidata *nd)
2013{
Jeff Layton873f1ee2012-10-10 15:25:29 -04002014 struct filename filename = { .name = name };
Nick Piggin31e6b012011-01-07 17:49:52 +11002015
Jeff Layton873f1ee2012-10-10 15:25:29 -04002016 return filename_lookup(dfd, &filename, flags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017}
2018
Al Viro79714f72012-06-15 03:01:42 +04002019/* does lookup, returns the object with parent locked */
2020struct dentry *kern_path_locked(const char *name, struct path *path)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002021{
Al Viro79714f72012-06-15 03:01:42 +04002022 struct nameidata nd;
2023 struct dentry *d;
2024 int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd);
2025 if (err)
2026 return ERR_PTR(err);
2027 if (nd.last_type != LAST_NORM) {
2028 path_put(&nd.path);
2029 return ERR_PTR(-EINVAL);
2030 }
2031 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Al Viro1e0ea002012-07-22 23:46:21 +04002032 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
Al Viro79714f72012-06-15 03:01:42 +04002033 if (IS_ERR(d)) {
2034 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2035 path_put(&nd.path);
2036 return d;
2037 }
2038 *path = nd.path;
2039 return d;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002040}
2041
Al Virod1811462008-08-02 00:49:18 -04002042int kern_path(const char *name, unsigned int flags, struct path *path)
2043{
2044 struct nameidata nd;
2045 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
2046 if (!res)
2047 *path = nd.path;
2048 return res;
2049}
Al Viro4d359502014-03-14 12:20:17 -04002050EXPORT_SYMBOL(kern_path);
Al Virod1811462008-08-02 00:49:18 -04002051
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002052/**
2053 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2054 * @dentry: pointer to dentry of the base directory
2055 * @mnt: pointer to vfs mount of the base directory
2056 * @name: pointer to file name
2057 * @flags: lookup flags
Al Viroe0a01242011-06-27 17:00:37 -04002058 * @path: pointer to struct path to fill
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002059 */
2060int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2061 const char *name, unsigned int flags,
Al Viroe0a01242011-06-27 17:00:37 -04002062 struct path *path)
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002063{
Al Viroe0a01242011-06-27 17:00:37 -04002064 struct nameidata nd;
2065 int err;
2066 nd.root.dentry = dentry;
2067 nd.root.mnt = mnt;
2068 BUG_ON(flags & LOOKUP_PARENT);
Al Viro5b6ca022011-03-09 23:04:47 -05002069 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
Al Viroe0a01242011-06-27 17:00:37 -04002070 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
2071 if (!err)
2072 *path = nd.path;
2073 return err;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002074}
Al Viro4d359502014-03-14 12:20:17 -04002075EXPORT_SYMBOL(vfs_path_lookup);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002076
James Morris057f6c02007-04-26 00:12:05 -07002077/*
2078 * Restricted form of lookup. Doesn't follow links, single-component only,
2079 * needs parent already locked. Doesn't follow mounts.
2080 * SMP-safe.
2081 */
Adrian Bunka244e162006-03-31 02:32:11 -08002082static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083{
Al Viro72bd8662012-06-10 17:17:17 -04002084 return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085}
2086
Christoph Hellwigeead1912007-10-16 23:25:38 -07002087/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07002088 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07002089 * @name: pathname component to lookup
2090 * @base: base directory to lookup from
2091 * @len: maximum length @len should be interpreted to
2092 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07002093 * Note that this routine is purely a helper for filesystem usage and should
2094 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07002095 * nameidata argument is passed to the filesystem methods and a filesystem
2096 * using this helper needs to be prepared for that.
2097 */
James Morris057f6c02007-04-26 00:12:05 -07002098struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2099{
James Morris057f6c02007-04-26 00:12:05 -07002100 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05002101 unsigned int c;
Miklos Szeredicda309d2012-03-26 12:54:21 +02002102 int err;
James Morris057f6c02007-04-26 00:12:05 -07002103
David Woodhouse2f9092e2009-04-20 23:18:37 +01002104 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2105
Al Viro6a96ba52011-03-07 23:49:20 -05002106 this.name = name;
2107 this.len = len;
Linus Torvalds0145acc2012-03-02 14:32:59 -08002108 this.hash = full_name_hash(name, len);
Al Viro6a96ba52011-03-07 23:49:20 -05002109 if (!len)
2110 return ERR_PTR(-EACCES);
2111
Al Viro21d8a152012-11-29 22:17:21 -05002112 if (unlikely(name[0] == '.')) {
2113 if (len < 2 || (len == 2 && name[1] == '.'))
2114 return ERR_PTR(-EACCES);
2115 }
2116
Al Viro6a96ba52011-03-07 23:49:20 -05002117 while (len--) {
2118 c = *(const unsigned char *)name++;
2119 if (c == '/' || c == '\0')
2120 return ERR_PTR(-EACCES);
Al Viro6a96ba52011-03-07 23:49:20 -05002121 }
Al Viro5a202bc2011-03-08 14:17:44 -05002122 /*
2123 * See if the low-level filesystem might want
2124 * to use its own hash..
2125 */
2126 if (base->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002127 int err = base->d_op->d_hash(base, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05002128 if (err < 0)
2129 return ERR_PTR(err);
2130 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07002131
Miklos Szeredicda309d2012-03-26 12:54:21 +02002132 err = inode_permission(base->d_inode, MAY_EXEC);
2133 if (err)
2134 return ERR_PTR(err);
2135
Al Viro72bd8662012-06-10 17:17:17 -04002136 return __lookup_hash(&this, base, 0);
James Morris057f6c02007-04-26 00:12:05 -07002137}
Al Viro4d359502014-03-14 12:20:17 -04002138EXPORT_SYMBOL(lookup_one_len);
James Morris057f6c02007-04-26 00:12:05 -07002139
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002140int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2141 struct path *path, int *empty)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142{
Al Viro2d8f3032008-07-22 09:59:21 -04002143 struct nameidata nd;
Jeff Layton91a27b22012-10-10 15:25:28 -04002144 struct filename *tmp = getname_flags(name, flags, empty);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04002147
2148 BUG_ON(flags & LOOKUP_PARENT);
2149
Jeff Layton873f1ee2012-10-10 15:25:29 -04002150 err = filename_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04002152 if (!err)
2153 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154 }
2155 return err;
2156}
2157
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002158int user_path_at(int dfd, const char __user *name, unsigned flags,
2159 struct path *path)
2160{
Linus Torvaldsf7493e52012-03-22 16:10:40 -07002161 return user_path_at_empty(dfd, name, flags, path, NULL);
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002162}
Al Viro4d359502014-03-14 12:20:17 -04002163EXPORT_SYMBOL(user_path_at);
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002164
Jeff Layton873f1ee2012-10-10 15:25:29 -04002165/*
2166 * NB: most callers don't do anything directly with the reference to the
2167 * to struct filename, but the nd->last pointer points into the name string
2168 * allocated by getname. So we must hold the reference to it until all
2169 * path-walking is complete.
2170 */
Jeff Layton91a27b22012-10-10 15:25:28 -04002171static struct filename *
Jeff Layton9e790bd2012-12-11 12:10:09 -05002172user_path_parent(int dfd, const char __user *path, struct nameidata *nd,
2173 unsigned int flags)
Al Viro2ad94ae2008-07-21 09:32:51 -04002174{
Jeff Layton91a27b22012-10-10 15:25:28 -04002175 struct filename *s = getname(path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002176 int error;
2177
Jeff Layton9e790bd2012-12-11 12:10:09 -05002178 /* only LOOKUP_REVAL is allowed in extra flags */
2179 flags &= LOOKUP_REVAL;
2180
Al Viro2ad94ae2008-07-21 09:32:51 -04002181 if (IS_ERR(s))
Jeff Layton91a27b22012-10-10 15:25:28 -04002182 return s;
Al Viro2ad94ae2008-07-21 09:32:51 -04002183
Jeff Layton9e790bd2012-12-11 12:10:09 -05002184 error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd);
Jeff Layton91a27b22012-10-10 15:25:28 -04002185 if (error) {
Al Viro2ad94ae2008-07-21 09:32:51 -04002186 putname(s);
Jeff Layton91a27b22012-10-10 15:25:28 -04002187 return ERR_PTR(error);
2188 }
Al Viro2ad94ae2008-07-21 09:32:51 -04002189
Jeff Layton91a27b22012-10-10 15:25:28 -04002190 return s;
Al Viro2ad94ae2008-07-21 09:32:51 -04002191}
2192
Jeff Layton80334262013-07-26 06:23:25 -04002193/**
Al Viro197df042013-09-08 14:03:27 -04002194 * mountpoint_last - look up last component for umount
Jeff Layton80334262013-07-26 06:23:25 -04002195 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2196 * @path: pointer to container for result
2197 *
2198 * This is a special lookup_last function just for umount. In this case, we
2199 * need to resolve the path without doing any revalidation.
2200 *
2201 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2202 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2203 * in almost all cases, this lookup will be served out of the dcache. The only
2204 * cases where it won't are if nd->last refers to a symlink or the path is
2205 * bogus and it doesn't exist.
2206 *
2207 * Returns:
2208 * -error: if there was an error during lookup. This includes -ENOENT if the
2209 * lookup found a negative dentry. The nd->path reference will also be
2210 * put in this case.
2211 *
2212 * 0: if we successfully resolved nd->path and found it to not to be a
2213 * symlink that needs to be followed. "path" will also be populated.
2214 * The nd->path reference will also be put.
2215 *
2216 * 1: if we successfully resolved nd->last and found it to be a symlink
2217 * that needs to be followed. "path" will be populated with the path
2218 * to the link, and nd->path will *not* be put.
2219 */
2220static int
Al Viro197df042013-09-08 14:03:27 -04002221mountpoint_last(struct nameidata *nd, struct path *path)
Jeff Layton80334262013-07-26 06:23:25 -04002222{
2223 int error = 0;
2224 struct dentry *dentry;
2225 struct dentry *dir = nd->path.dentry;
2226
Al Viro35759522013-09-08 13:41:33 -04002227 /* If we're in rcuwalk, drop out of it to handle last component */
2228 if (nd->flags & LOOKUP_RCU) {
2229 if (unlazy_walk(nd, NULL)) {
2230 error = -ECHILD;
2231 goto out;
2232 }
Jeff Layton80334262013-07-26 06:23:25 -04002233 }
2234
2235 nd->flags &= ~LOOKUP_PARENT;
2236
2237 if (unlikely(nd->last_type != LAST_NORM)) {
2238 error = handle_dots(nd, nd->last_type);
Al Viro35759522013-09-08 13:41:33 -04002239 if (error)
2240 goto out;
2241 dentry = dget(nd->path.dentry);
2242 goto done;
Jeff Layton80334262013-07-26 06:23:25 -04002243 }
2244
2245 mutex_lock(&dir->d_inode->i_mutex);
2246 dentry = d_lookup(dir, &nd->last);
2247 if (!dentry) {
2248 /*
2249 * No cached dentry. Mounted dentries are pinned in the cache,
2250 * so that means that this dentry is probably a symlink or the
2251 * path doesn't actually point to a mounted dentry.
2252 */
2253 dentry = d_alloc(dir, &nd->last);
2254 if (!dentry) {
2255 error = -ENOMEM;
Dave Jonesbcceeeb2013-09-10 17:04:25 -04002256 mutex_unlock(&dir->d_inode->i_mutex);
Al Viro35759522013-09-08 13:41:33 -04002257 goto out;
Jeff Layton80334262013-07-26 06:23:25 -04002258 }
Al Viro35759522013-09-08 13:41:33 -04002259 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2260 error = PTR_ERR(dentry);
Dave Jonesbcceeeb2013-09-10 17:04:25 -04002261 if (IS_ERR(dentry)) {
2262 mutex_unlock(&dir->d_inode->i_mutex);
Al Viro35759522013-09-08 13:41:33 -04002263 goto out;
Dave Jonesbcceeeb2013-09-10 17:04:25 -04002264 }
Jeff Layton80334262013-07-26 06:23:25 -04002265 }
2266 mutex_unlock(&dir->d_inode->i_mutex);
2267
Al Viro35759522013-09-08 13:41:33 -04002268done:
Al Viro22213312014-04-19 12:30:58 -04002269 if (!dentry->d_inode || d_is_negative(dentry)) {
Al Viro35759522013-09-08 13:41:33 -04002270 error = -ENOENT;
2271 dput(dentry);
2272 goto out;
Jeff Layton80334262013-07-26 06:23:25 -04002273 }
Al Viro35759522013-09-08 13:41:33 -04002274 path->dentry = dentry;
Vasily Averin295dc392014-07-21 12:30:23 +04002275 path->mnt = nd->path.mnt;
David Howellsb18825a2013-09-12 19:22:53 +01002276 if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW))
Al Viro35759522013-09-08 13:41:33 -04002277 return 1;
Vasily Averin295dc392014-07-21 12:30:23 +04002278 mntget(path->mnt);
Al Viro35759522013-09-08 13:41:33 -04002279 follow_mount(path);
2280 error = 0;
2281out:
Jeff Layton80334262013-07-26 06:23:25 -04002282 terminate_walk(nd);
2283 return error;
2284}
2285
2286/**
Al Viro197df042013-09-08 14:03:27 -04002287 * path_mountpoint - look up a path to be umounted
Jeff Layton80334262013-07-26 06:23:25 -04002288 * @dfd: directory file descriptor to start walk from
2289 * @name: full pathname to walk
Randy Dunlap606d6fe2013-10-19 14:56:55 -07002290 * @path: pointer to container for result
Jeff Layton80334262013-07-26 06:23:25 -04002291 * @flags: lookup flags
Jeff Layton80334262013-07-26 06:23:25 -04002292 *
2293 * Look up the given name, but don't attempt to revalidate the last component.
Randy Dunlap606d6fe2013-10-19 14:56:55 -07002294 * Returns 0 and "path" will be valid on success; Returns error otherwise.
Jeff Layton80334262013-07-26 06:23:25 -04002295 */
2296static int
Al Viro197df042013-09-08 14:03:27 -04002297path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags)
Jeff Layton80334262013-07-26 06:23:25 -04002298{
2299 struct file *base = NULL;
2300 struct nameidata nd;
2301 int err;
2302
2303 err = path_init(dfd, name, flags | LOOKUP_PARENT, &nd, &base);
2304 if (unlikely(err))
Al Viro115cbfd2014-10-11 23:05:52 -04002305 goto out;
Jeff Layton80334262013-07-26 06:23:25 -04002306
2307 current->total_link_count = 0;
2308 err = link_path_walk(name, &nd);
2309 if (err)
2310 goto out;
2311
Al Viro197df042013-09-08 14:03:27 -04002312 err = mountpoint_last(&nd, path);
Jeff Layton80334262013-07-26 06:23:25 -04002313 while (err > 0) {
2314 void *cookie;
2315 struct path link = *path;
2316 err = may_follow_link(&link, &nd);
2317 if (unlikely(err))
2318 break;
2319 nd.flags |= LOOKUP_PARENT;
2320 err = follow_link(&link, &nd, &cookie);
2321 if (err)
2322 break;
Al Viro197df042013-09-08 14:03:27 -04002323 err = mountpoint_last(&nd, path);
Jeff Layton80334262013-07-26 06:23:25 -04002324 put_link(&nd, &link, cookie);
2325 }
2326out:
2327 if (base)
2328 fput(base);
2329
2330 if (nd.root.mnt && !(nd.flags & LOOKUP_ROOT))
2331 path_put(&nd.root);
2332
2333 return err;
2334}
2335
Al Viro2d864652013-09-08 20:18:44 -04002336static int
2337filename_mountpoint(int dfd, struct filename *s, struct path *path,
2338 unsigned int flags)
2339{
2340 int error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU);
2341 if (unlikely(error == -ECHILD))
2342 error = path_mountpoint(dfd, s->name, path, flags);
2343 if (unlikely(error == -ESTALE))
2344 error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL);
2345 if (likely(!error))
2346 audit_inode(s, path->dentry, 0);
2347 return error;
2348}
2349
Jeff Layton80334262013-07-26 06:23:25 -04002350/**
Al Viro197df042013-09-08 14:03:27 -04002351 * user_path_mountpoint_at - lookup a path from userland in order to umount it
Jeff Layton80334262013-07-26 06:23:25 -04002352 * @dfd: directory file descriptor
2353 * @name: pathname from userland
2354 * @flags: lookup flags
2355 * @path: pointer to container to hold result
2356 *
2357 * A umount is a special case for path walking. We're not actually interested
2358 * in the inode in this situation, and ESTALE errors can be a problem. We
2359 * simply want track down the dentry and vfsmount attached at the mountpoint
2360 * and avoid revalidating the last component.
2361 *
2362 * Returns 0 and populates "path" on success.
2363 */
2364int
Al Viro197df042013-09-08 14:03:27 -04002365user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
Jeff Layton80334262013-07-26 06:23:25 -04002366 struct path *path)
2367{
2368 struct filename *s = getname(name);
2369 int error;
Jeff Layton80334262013-07-26 06:23:25 -04002370 if (IS_ERR(s))
2371 return PTR_ERR(s);
Al Viro2d864652013-09-08 20:18:44 -04002372 error = filename_mountpoint(dfd, s, path, flags);
Jeff Layton80334262013-07-26 06:23:25 -04002373 putname(s);
2374 return error;
2375}
2376
Al Viro2d864652013-09-08 20:18:44 -04002377int
2378kern_path_mountpoint(int dfd, const char *name, struct path *path,
2379 unsigned int flags)
2380{
2381 struct filename s = {.name = name};
2382 return filename_mountpoint(dfd, &s, path, flags);
2383}
2384EXPORT_SYMBOL(kern_path_mountpoint);
2385
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386/*
2387 * It's inline, so penalty for filesystems that don't use sticky bit is
2388 * minimal.
2389 */
2390static inline int check_sticky(struct inode *dir, struct inode *inode)
2391{
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002392 kuid_t fsuid = current_fsuid();
David Howellsda9592e2008-11-14 10:39:05 +11002393
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394 if (!(dir->i_mode & S_ISVTX))
2395 return 0;
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002396 if (uid_eq(inode->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002397 return 0;
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002398 if (uid_eq(dir->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399 return 0;
Andy Lutomirski23adbe12014-06-10 12:45:42 -07002400 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401}
2402
2403/*
2404 * Check whether we can remove a link victim from directory dir, check
2405 * whether the type of victim is right.
2406 * 1. We can't do it if dir is read-only (done in permission())
2407 * 2. We should have write and exec permissions on dir
2408 * 3. We can't remove anything from append-only dir
2409 * 4. We can't do anything with immutable dir (done in permission())
2410 * 5. If the sticky bit on dir is set we should either
2411 * a. be owner of dir, or
2412 * b. be owner of victim, or
2413 * c. have CAP_FOWNER capability
2414 * 6. If the victim is append-only or immutable we can't do antyhing with
2415 * links pointing to it.
2416 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2417 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2418 * 9. We can't remove a root or mountpoint.
2419 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2420 * nfs_async_unlink().
2421 */
David Howellsb18825a2013-09-12 19:22:53 +01002422static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423{
David Howellsb18825a2013-09-12 19:22:53 +01002424 struct inode *inode = victim->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425 int error;
2426
David Howellsb18825a2013-09-12 19:22:53 +01002427 if (d_is_negative(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428 return -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01002429 BUG_ON(!inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430
2431 BUG_ON(victim->d_parent->d_inode != dir);
Jeff Layton4fa6b5e2012-10-10 15:25:25 -04002432 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433
Al Virof419a2e2008-07-22 00:07:17 -04002434 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002435 if (error)
2436 return error;
2437 if (IS_APPEND(dir))
2438 return -EPERM;
David Howellsb18825a2013-09-12 19:22:53 +01002439
2440 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2441 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442 return -EPERM;
2443 if (isdir) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002444 if (!d_is_dir(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445 return -ENOTDIR;
2446 if (IS_ROOT(victim))
2447 return -EBUSY;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002448 } else if (d_is_dir(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449 return -EISDIR;
2450 if (IS_DEADDIR(dir))
2451 return -ENOENT;
2452 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2453 return -EBUSY;
2454 return 0;
2455}
2456
2457/* Check whether we can create an object with dentry child in directory
2458 * dir.
2459 * 1. We can't do it if child already exists (open has special treatment for
2460 * this case, but since we are inlined it's OK)
2461 * 2. We can't do it if dir is read-only (done in permission())
2462 * 3. We should have write and exec permissions on dir
2463 * 4. We can't do it if dir is immutable (done in permission())
2464 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02002465static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466{
Jeff Layton14e972b2013-05-08 10:25:58 -04002467 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468 if (child->d_inode)
2469 return -EEXIST;
2470 if (IS_DEADDIR(dir))
2471 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04002472 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002473}
2474
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475/*
2476 * p1 and p2 should be directories on the same fs.
2477 */
2478struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2479{
2480 struct dentry *p;
2481
2482 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07002483 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484 return NULL;
2485 }
2486
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002487 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002489 p = d_ancestor(p2, p1);
2490 if (p) {
2491 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2492 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2493 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494 }
2495
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002496 p = d_ancestor(p1, p2);
2497 if (p) {
2498 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2499 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2500 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002501 }
2502
Ingo Molnarf2eace22006-07-03 00:25:05 -07002503 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2504 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505 return NULL;
2506}
Al Viro4d359502014-03-14 12:20:17 -04002507EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508
2509void unlock_rename(struct dentry *p1, struct dentry *p2)
2510{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002511 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002512 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002513 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002514 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515 }
2516}
Al Viro4d359502014-03-14 12:20:17 -04002517EXPORT_SYMBOL(unlock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518
Al Viro4acdaf22011-07-26 01:42:34 -04002519int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viro312b63f2012-06-10 18:09:36 -04002520 bool want_excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002522 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523 if (error)
2524 return error;
2525
Al Viroacfa4382008-12-04 10:06:33 -05002526 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527 return -EACCES; /* shouldn't it be ENOSYS? */
2528 mode &= S_IALLUGO;
2529 mode |= S_IFREG;
2530 error = security_inode_create(dir, dentry, mode);
2531 if (error)
2532 return error;
Al Viro312b63f2012-06-10 18:09:36 -04002533 error = dir->i_op->create(dir, dentry, mode, want_excl);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002534 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002535 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536 return error;
2537}
Al Viro4d359502014-03-14 12:20:17 -04002538EXPORT_SYMBOL(vfs_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539
Al Viro73d049a2011-03-11 12:08:24 -05002540static int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002541{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002542 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543 struct inode *inode = dentry->d_inode;
2544 int error;
2545
Al Virobcda7652011-03-13 16:42:14 -04002546 /* O_PATH? */
2547 if (!acc_mode)
2548 return 0;
2549
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550 if (!inode)
2551 return -ENOENT;
2552
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002553 switch (inode->i_mode & S_IFMT) {
2554 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002556 case S_IFDIR:
2557 if (acc_mode & MAY_WRITE)
2558 return -EISDIR;
2559 break;
2560 case S_IFBLK:
2561 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002562 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002564 /*FALLTHRU*/
2565 case S_IFIFO:
2566 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002568 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002569 }
Dave Hansenb41572e2007-10-16 23:31:14 -07002570
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002571 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07002572 if (error)
2573 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05002574
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575 /*
2576 * An append-only file must be opened in append mode for writing.
2577 */
2578 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05002579 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05002580 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05002582 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583 }
2584
2585 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002586 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05002587 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588
J. Bruce Fieldsf3c7691e2011-09-21 10:58:13 -04002589 return 0;
Al Viro7715b522009-12-16 03:54:00 -05002590}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591
Jeff Laytone1181ee2010-12-07 16:19:50 -05002592static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002593{
Jeff Laytone1181ee2010-12-07 16:19:50 -05002594 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002595 struct inode *inode = path->dentry->d_inode;
2596 int error = get_write_access(inode);
2597 if (error)
2598 return error;
2599 /*
2600 * Refuse to truncate files with mandatory locks held on them.
2601 */
Jeff Laytond7a06982014-03-10 09:54:15 -04002602 error = locks_verify_locked(filp);
Al Viro7715b522009-12-16 03:54:00 -05002603 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002604 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002605 if (!error) {
2606 error = do_truncate(path->dentry, 0,
2607 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002608 filp);
Al Viro7715b522009-12-16 03:54:00 -05002609 }
2610 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002611 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612}
2613
Dave Hansend57999e2008-02-15 14:37:27 -08002614static inline int open_to_namei_flags(int flag)
2615{
Al Viro8a5e9292011-06-25 19:15:54 -04002616 if ((flag & O_ACCMODE) == 3)
2617 flag--;
Dave Hansend57999e2008-02-15 14:37:27 -08002618 return flag;
2619}
2620
Miklos Szeredid18e9002012-06-05 15:10:17 +02002621static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2622{
2623 int error = security_path_mknod(dir, dentry, mode, 0);
2624 if (error)
2625 return error;
2626
2627 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2628 if (error)
2629 return error;
2630
2631 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2632}
2633
David Howells1acf0af2012-06-14 16:13:46 +01002634/*
2635 * Attempt to atomically look up, create and open a file from a negative
2636 * dentry.
2637 *
2638 * Returns 0 if successful. The file will have been created and attached to
2639 * @file by the filesystem calling finish_open().
2640 *
2641 * Returns 1 if the file was looked up only or didn't need creating. The
2642 * caller will need to perform the open themselves. @path will have been
2643 * updated to point to the new dentry. This may be negative.
2644 *
2645 * Returns an error code otherwise.
2646 */
Al Viro2675a4e2012-06-22 12:41:10 +04002647static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2648 struct path *path, struct file *file,
2649 const struct open_flags *op,
Al Viro64894cf2012-07-31 00:53:35 +04002650 bool got_write, bool need_lookup,
Al Viro2675a4e2012-06-22 12:41:10 +04002651 int *opened)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002652{
2653 struct inode *dir = nd->path.dentry->d_inode;
2654 unsigned open_flag = open_to_namei_flags(op->open_flag);
2655 umode_t mode;
2656 int error;
2657 int acc_mode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002658 int create_error = 0;
2659 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
Miklos Szeredi116cc022013-09-16 14:52:05 +02002660 bool excl;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002661
2662 BUG_ON(dentry->d_inode);
2663
2664 /* Don't create child dentry for a dead directory. */
2665 if (unlikely(IS_DEADDIR(dir))) {
Al Viro2675a4e2012-06-22 12:41:10 +04002666 error = -ENOENT;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002667 goto out;
2668 }
2669
Miklos Szeredi62b259d2012-08-15 13:01:24 +02002670 mode = op->mode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002671 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2672 mode &= ~current_umask();
2673
Miklos Szeredi116cc022013-09-16 14:52:05 +02002674 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2675 if (excl)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002676 open_flag &= ~O_TRUNC;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002677
2678 /*
2679 * Checking write permission is tricky, bacuse we don't know if we are
2680 * going to actually need it: O_CREAT opens should work as long as the
2681 * file exists. But checking existence breaks atomicity. The trick is
2682 * to check access and if not granted clear O_CREAT from the flags.
2683 *
2684 * Another problem is returing the "right" error value (e.g. for an
2685 * O_EXCL open we want to return EEXIST not EROFS).
2686 */
Al Viro64894cf2012-07-31 00:53:35 +04002687 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2688 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2689 if (!(open_flag & O_CREAT)) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002690 /*
2691 * No O_CREATE -> atomicity not a requirement -> fall
2692 * back to lookup + open
2693 */
2694 goto no_open;
2695 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2696 /* Fall back and fail with the right error */
Al Viro64894cf2012-07-31 00:53:35 +04002697 create_error = -EROFS;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002698 goto no_open;
2699 } else {
2700 /* No side effects, safe to clear O_CREAT */
Al Viro64894cf2012-07-31 00:53:35 +04002701 create_error = -EROFS;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002702 open_flag &= ~O_CREAT;
2703 }
2704 }
2705
2706 if (open_flag & O_CREAT) {
Miklos Szeredi38227f72012-08-15 13:01:24 +02002707 error = may_o_create(&nd->path, dentry, mode);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002708 if (error) {
2709 create_error = error;
2710 if (open_flag & O_EXCL)
2711 goto no_open;
2712 open_flag &= ~O_CREAT;
2713 }
2714 }
2715
2716 if (nd->flags & LOOKUP_DIRECTORY)
2717 open_flag |= O_DIRECTORY;
2718
Al Viro30d90492012-06-22 12:40:19 +04002719 file->f_path.dentry = DENTRY_NOT_SET;
2720 file->f_path.mnt = nd->path.mnt;
2721 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
Al Viro47237682012-06-10 05:01:45 -04002722 opened);
Al Virod9585272012-06-22 12:39:14 +04002723 if (error < 0) {
Al Virod9585272012-06-22 12:39:14 +04002724 if (create_error && error == -ENOENT)
2725 error = create_error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002726 goto out;
2727 }
2728
Al Virod9585272012-06-22 12:39:14 +04002729 if (error) { /* returned 1, that is */
Al Viro30d90492012-06-22 12:40:19 +04002730 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
Al Viro2675a4e2012-06-22 12:41:10 +04002731 error = -EIO;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002732 goto out;
2733 }
Al Viro30d90492012-06-22 12:40:19 +04002734 if (file->f_path.dentry) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002735 dput(dentry);
Al Viro30d90492012-06-22 12:40:19 +04002736 dentry = file->f_path.dentry;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002737 }
Al Viro03da6332013-09-16 19:22:33 -04002738 if (*opened & FILE_CREATED)
2739 fsnotify_create(dir, dentry);
2740 if (!dentry->d_inode) {
2741 WARN_ON(*opened & FILE_CREATED);
2742 if (create_error) {
2743 error = create_error;
2744 goto out;
2745 }
2746 } else {
2747 if (excl && !(*opened & FILE_CREATED)) {
2748 error = -EEXIST;
2749 goto out;
2750 }
Sage Weil62b2ce92012-08-15 13:30:12 -07002751 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02002752 goto looked_up;
2753 }
2754
2755 /*
2756 * We didn't have the inode before the open, so check open permission
2757 * here.
2758 */
Al Viro03da6332013-09-16 19:22:33 -04002759 acc_mode = op->acc_mode;
2760 if (*opened & FILE_CREATED) {
2761 WARN_ON(!(open_flag & O_CREAT));
2762 fsnotify_create(dir, dentry);
2763 acc_mode = MAY_OPEN;
2764 }
Al Viro2675a4e2012-06-22 12:41:10 +04002765 error = may_open(&file->f_path, acc_mode, open_flag);
2766 if (error)
2767 fput(file);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002768
2769out:
2770 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04002771 return error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002772
Miklos Szeredid18e9002012-06-05 15:10:17 +02002773no_open:
2774 if (need_lookup) {
Al Viro72bd8662012-06-10 17:17:17 -04002775 dentry = lookup_real(dir, dentry, nd->flags);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002776 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002777 return PTR_ERR(dentry);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002778
2779 if (create_error) {
2780 int open_flag = op->open_flag;
2781
Al Viro2675a4e2012-06-22 12:41:10 +04002782 error = create_error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002783 if ((open_flag & O_EXCL)) {
2784 if (!dentry->d_inode)
2785 goto out;
2786 } else if (!dentry->d_inode) {
2787 goto out;
2788 } else if ((open_flag & O_TRUNC) &&
2789 S_ISREG(dentry->d_inode->i_mode)) {
2790 goto out;
2791 }
2792 /* will fail later, go on to get the right error */
2793 }
2794 }
2795looked_up:
2796 path->dentry = dentry;
2797 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04002798 return 1;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002799}
2800
Nick Piggin31e6b012011-01-07 17:49:52 +11002801/*
David Howells1acf0af2012-06-14 16:13:46 +01002802 * Look up and maybe create and open the last component.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002803 *
2804 * Must be called with i_mutex held on parent.
2805 *
David Howells1acf0af2012-06-14 16:13:46 +01002806 * Returns 0 if the file was successfully atomically created (if necessary) and
2807 * opened. In this case the file will be returned attached to @file.
2808 *
2809 * Returns 1 if the file was not completely opened at this time, though lookups
2810 * and creations will have been performed and the dentry returned in @path will
2811 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2812 * specified then a negative dentry may be returned.
2813 *
2814 * An error code is returned otherwise.
2815 *
2816 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2817 * cleared otherwise prior to returning.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002818 */
Al Viro2675a4e2012-06-22 12:41:10 +04002819static int lookup_open(struct nameidata *nd, struct path *path,
2820 struct file *file,
2821 const struct open_flags *op,
Al Viro64894cf2012-07-31 00:53:35 +04002822 bool got_write, int *opened)
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002823{
2824 struct dentry *dir = nd->path.dentry;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002825 struct inode *dir_inode = dir->d_inode;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002826 struct dentry *dentry;
2827 int error;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002828 bool need_lookup;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002829
Al Viro47237682012-06-10 05:01:45 -04002830 *opened &= ~FILE_CREATED;
Al Viro201f9562012-06-22 12:42:10 +04002831 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002832 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002833 return PTR_ERR(dentry);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002834
Miklos Szeredid18e9002012-06-05 15:10:17 +02002835 /* Cached positive dentry: will open in f_op->open */
2836 if (!need_lookup && dentry->d_inode)
2837 goto out_no_open;
2838
2839 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
Al Viro64894cf2012-07-31 00:53:35 +04002840 return atomic_open(nd, dentry, path, file, op, got_write,
Al Viro47237682012-06-10 05:01:45 -04002841 need_lookup, opened);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002842 }
2843
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002844 if (need_lookup) {
2845 BUG_ON(dentry->d_inode);
2846
Al Viro72bd8662012-06-10 17:17:17 -04002847 dentry = lookup_real(dir_inode, dentry, nd->flags);
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002848 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002849 return PTR_ERR(dentry);
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002850 }
2851
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002852 /* Negative dentry, just create the file */
2853 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2854 umode_t mode = op->mode;
2855 if (!IS_POSIXACL(dir->d_inode))
2856 mode &= ~current_umask();
2857 /*
2858 * This write is needed to ensure that a
2859 * rw->ro transition does not occur between
2860 * the time when the file is created and when
2861 * a permanent write count is taken through
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002862 * the 'struct file' in finish_open().
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002863 */
Al Viro64894cf2012-07-31 00:53:35 +04002864 if (!got_write) {
2865 error = -EROFS;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002866 goto out_dput;
Al Viro64894cf2012-07-31 00:53:35 +04002867 }
Al Viro47237682012-06-10 05:01:45 -04002868 *opened |= FILE_CREATED;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002869 error = security_path_mknod(&nd->path, dentry, mode, 0);
2870 if (error)
2871 goto out_dput;
Al Viro312b63f2012-06-10 18:09:36 -04002872 error = vfs_create(dir->d_inode, dentry, mode,
2873 nd->flags & LOOKUP_EXCL);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002874 if (error)
2875 goto out_dput;
2876 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02002877out_no_open:
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002878 path->dentry = dentry;
2879 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04002880 return 1;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002881
2882out_dput:
2883 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04002884 return error;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002885}
2886
2887/*
Al Virofe2d35f2011-03-05 22:58:25 -05002888 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11002889 */
Al Viro2675a4e2012-06-22 12:41:10 +04002890static int do_last(struct nameidata *nd, struct path *path,
2891 struct file *file, const struct open_flags *op,
Jeff Layton669abf42012-10-10 16:43:10 -04002892 int *opened, struct filename *name)
Al Virofb1cc552009-12-24 01:58:28 -05002893{
Al Viroa1e28032009-12-24 02:12:06 -05002894 struct dentry *dir = nd->path.dentry;
Al Viroca344a892011-03-09 00:36:45 -05002895 int open_flag = op->open_flag;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002896 bool will_truncate = (open_flag & O_TRUNC) != 0;
Al Viro64894cf2012-07-31 00:53:35 +04002897 bool got_write = false;
Al Virobcda7652011-03-13 16:42:14 -04002898 int acc_mode = op->acc_mode;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02002899 struct inode *inode;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002900 bool symlink_ok = false;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002901 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2902 bool retried = false;
Al Viro16c2cd72011-02-22 15:50:10 -05002903 int error;
Al Virofb1cc552009-12-24 01:58:28 -05002904
Al Viroc3e380b2011-02-23 13:39:45 -05002905 nd->flags &= ~LOOKUP_PARENT;
2906 nd->flags |= op->intent;
2907
Al Virobc77daa2013-06-06 09:12:33 -04002908 if (nd->last_type != LAST_NORM) {
Al Virofe2d35f2011-03-05 22:58:25 -05002909 error = handle_dots(nd, nd->last_type);
2910 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002911 return error;
Miklos Szeredie83db162012-06-05 15:10:29 +02002912 goto finish_open;
Al Viro1f36f772009-12-26 10:56:19 -05002913 }
Al Viro67ee3ad2009-12-26 07:01:01 -05002914
Al Viroca344a892011-03-09 00:36:45 -05002915 if (!(open_flag & O_CREAT)) {
Al Virofe2d35f2011-03-05 22:58:25 -05002916 if (nd->last.name[nd->last.len])
2917 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
Al Virobcda7652011-03-13 16:42:14 -04002918 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002919 symlink_ok = true;
Al Virofe2d35f2011-03-05 22:58:25 -05002920 /* we _can_ be in RCU mode here */
Al Viroe97cdc82013-01-24 18:16:00 -05002921 error = lookup_fast(nd, path, &inode);
Miklos Szeredi71574862012-06-05 15:10:14 +02002922 if (likely(!error))
2923 goto finish_lookup;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02002924
Miklos Szeredi71574862012-06-05 15:10:14 +02002925 if (error < 0)
Al Viro2675a4e2012-06-22 12:41:10 +04002926 goto out;
Miklos Szeredi37d7fff2012-06-05 15:10:12 +02002927
Miklos Szeredi71574862012-06-05 15:10:14 +02002928 BUG_ON(nd->inode != dir->d_inode);
Miklos Szeredib6183df2012-06-05 15:10:13 +02002929 } else {
2930 /* create side of things */
2931 /*
2932 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2933 * has been cleared when we got to the last component we are
2934 * about to look up
2935 */
2936 error = complete_walk(nd);
2937 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002938 return error;
Miklos Szeredib6183df2012-06-05 15:10:13 +02002939
Jeff Layton33e22082013-04-12 15:16:32 -04002940 audit_inode(name, dir, LOOKUP_PARENT);
Miklos Szeredib6183df2012-06-05 15:10:13 +02002941 error = -EISDIR;
2942 /* trailing slashes? */
2943 if (nd->last.name[nd->last.len])
Al Viro2675a4e2012-06-22 12:41:10 +04002944 goto out;
Al Virofe2d35f2011-03-05 22:58:25 -05002945 }
2946
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002947retry_lookup:
Al Viro64894cf2012-07-31 00:53:35 +04002948 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2949 error = mnt_want_write(nd->path.mnt);
2950 if (!error)
2951 got_write = true;
2952 /*
2953 * do _not_ fail yet - we might not need that or fail with
2954 * a different error; let lookup_open() decide; we'll be
2955 * dropping this one anyway.
2956 */
2957 }
Al Viroa1e28032009-12-24 02:12:06 -05002958 mutex_lock(&dir->d_inode->i_mutex);
Al Viro64894cf2012-07-31 00:53:35 +04002959 error = lookup_open(nd, path, file, op, got_write, opened);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002960 mutex_unlock(&dir->d_inode->i_mutex);
Al Viroa1e28032009-12-24 02:12:06 -05002961
Al Viro2675a4e2012-06-22 12:41:10 +04002962 if (error <= 0) {
2963 if (error)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002964 goto out;
2965
Al Viro47237682012-06-10 05:01:45 -04002966 if ((*opened & FILE_CREATED) ||
Al Viro496ad9a2013-01-23 17:07:38 -05002967 !S_ISREG(file_inode(file)->i_mode))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002968 will_truncate = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002969
Jeff Laytonadb5c242012-10-10 16:43:13 -04002970 audit_inode(name, file->f_path.dentry, 0);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002971 goto opened;
2972 }
Al Virofb1cc552009-12-24 01:58:28 -05002973
Al Viro47237682012-06-10 05:01:45 -04002974 if (*opened & FILE_CREATED) {
Al Viro9b44f1b2011-03-09 00:17:27 -05002975 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05002976 open_flag &= ~O_TRUNC;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002977 will_truncate = false;
Al Virobcda7652011-03-13 16:42:14 -04002978 acc_mode = MAY_OPEN;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002979 path_to_nameidata(path, nd);
Miklos Szeredie83db162012-06-05 15:10:29 +02002980 goto finish_open_created;
Al Virofb1cc552009-12-24 01:58:28 -05002981 }
2982
2983 /*
Jeff Layton3134f372012-07-25 10:19:47 -04002984 * create/update audit record if it already exists.
Al Virofb1cc552009-12-24 01:58:28 -05002985 */
David Howellsb18825a2013-09-12 19:22:53 +01002986 if (d_is_positive(path->dentry))
Jeff Laytonadb5c242012-10-10 16:43:13 -04002987 audit_inode(name, path->dentry, 0);
Al Virofb1cc552009-12-24 01:58:28 -05002988
Miklos Szeredid18e9002012-06-05 15:10:17 +02002989 /*
2990 * If atomic_open() acquired write access it is dropped now due to
2991 * possible mount and symlink following (this might be optimized away if
2992 * necessary...)
2993 */
Al Viro64894cf2012-07-31 00:53:35 +04002994 if (got_write) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002995 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04002996 got_write = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002997 }
2998
Al Virofb1cc552009-12-24 01:58:28 -05002999 error = -EEXIST;
Al Virof8310c52012-07-30 11:50:30 +04003000 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
Al Virofb1cc552009-12-24 01:58:28 -05003001 goto exit_dput;
3002
David Howells9875cf82011-01-14 18:45:21 +00003003 error = follow_managed(path, nd->flags);
3004 if (error < 0)
3005 goto exit_dput;
Al Virofb1cc552009-12-24 01:58:28 -05003006
Al Viroa3fbbde2011-11-07 21:21:26 +00003007 if (error)
3008 nd->flags |= LOOKUP_JUMPED;
3009
Miklos Szeredidecf3402012-05-21 17:30:08 +02003010 BUG_ON(nd->flags & LOOKUP_RCU);
3011 inode = path->dentry->d_inode;
Miklos Szeredi5f5daac2012-05-21 17:30:14 +02003012finish_lookup:
3013 /* we _can_ be in RCU mode here */
Al Virofb1cc552009-12-24 01:58:28 -05003014 error = -ENOENT;
Al Viro22213312014-04-19 12:30:58 -04003015 if (!inode || d_is_negative(path->dentry)) {
Miklos Szeredi54c33e72012-05-21 17:30:10 +02003016 path_to_nameidata(path, nd);
Al Viro2675a4e2012-06-22 12:41:10 +04003017 goto out;
Miklos Szeredi54c33e72012-05-21 17:30:10 +02003018 }
Al Viro9e67f362009-12-26 07:04:50 -05003019
David Howellsb18825a2013-09-12 19:22:53 +01003020 if (should_follow_link(path->dentry, !symlink_ok)) {
Miklos Szeredid45ea862012-05-21 17:30:09 +02003021 if (nd->flags & LOOKUP_RCU) {
3022 if (unlikely(unlazy_walk(nd, path->dentry))) {
3023 error = -ECHILD;
Al Viro2675a4e2012-06-22 12:41:10 +04003024 goto out;
Miklos Szeredid45ea862012-05-21 17:30:09 +02003025 }
3026 }
3027 BUG_ON(inode != path->dentry->d_inode);
Al Viro2675a4e2012-06-22 12:41:10 +04003028 return 1;
Miklos Szeredid45ea862012-05-21 17:30:09 +02003029 }
Al Virofb1cc552009-12-24 01:58:28 -05003030
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003031 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
3032 path_to_nameidata(path, nd);
3033 } else {
3034 save_parent.dentry = nd->path.dentry;
3035 save_parent.mnt = mntget(path->mnt);
3036 nd->path.dentry = path->dentry;
3037
3038 }
Miklos Szeredidecf3402012-05-21 17:30:08 +02003039 nd->inode = inode;
Al Viroa3fbbde2011-11-07 21:21:26 +00003040 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
Al Virobc77daa2013-06-06 09:12:33 -04003041finish_open:
Al Viroa3fbbde2011-11-07 21:21:26 +00003042 error = complete_walk(nd);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003043 if (error) {
3044 path_put(&save_parent);
Al Viro2675a4e2012-06-22 12:41:10 +04003045 return error;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003046 }
Al Virobc77daa2013-06-06 09:12:33 -04003047 audit_inode(name, nd->path.dentry, 0);
Al Virofb1cc552009-12-24 01:58:28 -05003048 error = -EISDIR;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003049 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04003050 goto out;
Miklos Szerediaf2f5542012-05-21 17:30:11 +02003051 error = -ENOTDIR;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003052 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04003053 goto out;
Al Viro6c0d46c2011-03-09 00:59:59 -05003054 if (!S_ISREG(nd->inode->i_mode))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003055 will_truncate = false;
Al Viro6c0d46c2011-03-09 00:59:59 -05003056
Al Viro0f9d1a12011-03-09 00:13:14 -05003057 if (will_truncate) {
3058 error = mnt_want_write(nd->path.mnt);
3059 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003060 goto out;
Al Viro64894cf2012-07-31 00:53:35 +04003061 got_write = true;
Al Viro0f9d1a12011-03-09 00:13:14 -05003062 }
Miklos Szeredie83db162012-06-05 15:10:29 +02003063finish_open_created:
Al Virobcda7652011-03-13 16:42:14 -04003064 error = may_open(&nd->path, acc_mode, open_flag);
Al Viroca344a892011-03-09 00:36:45 -05003065 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003066 goto out;
Al Viro30d90492012-06-22 12:40:19 +04003067 file->f_path.mnt = nd->path.mnt;
3068 error = finish_open(file, nd->path.dentry, NULL, opened);
3069 if (error) {
Al Viro30d90492012-06-22 12:40:19 +04003070 if (error == -EOPENSTALE)
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003071 goto stale_open;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003072 goto out;
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003073 }
Miklos Szeredia8277b92012-06-05 15:10:32 +02003074opened:
Al Viro2675a4e2012-06-22 12:41:10 +04003075 error = open_check_o_direct(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003076 if (error)
3077 goto exit_fput;
Dmitry Kasatkin3034a142014-06-27 18:15:44 +03003078 error = ima_file_check(file, op->acc_mode, *opened);
Miklos Szerediaa4caad2012-06-05 15:10:28 +02003079 if (error)
3080 goto exit_fput;
3081
3082 if (will_truncate) {
Al Viro2675a4e2012-06-22 12:41:10 +04003083 error = handle_truncate(file);
Miklos Szerediaa4caad2012-06-05 15:10:28 +02003084 if (error)
3085 goto exit_fput;
Al Viro0f9d1a12011-03-09 00:13:14 -05003086 }
Al Viroca344a892011-03-09 00:36:45 -05003087out:
Al Viro64894cf2012-07-31 00:53:35 +04003088 if (got_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05003089 mnt_drop_write(nd->path.mnt);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003090 path_put(&save_parent);
Miklos Szeredie276ae62012-05-21 17:30:06 +02003091 terminate_walk(nd);
Al Viro2675a4e2012-06-22 12:41:10 +04003092 return error;
Al Virofb1cc552009-12-24 01:58:28 -05003093
Al Virofb1cc552009-12-24 01:58:28 -05003094exit_dput:
3095 path_put_conditional(path, nd);
Al Viroca344a892011-03-09 00:36:45 -05003096 goto out;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003097exit_fput:
Al Viro2675a4e2012-06-22 12:41:10 +04003098 fput(file);
3099 goto out;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003100
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003101stale_open:
3102 /* If no saved parent or already retried then can't retry */
3103 if (!save_parent.dentry || retried)
3104 goto out;
3105
3106 BUG_ON(save_parent.dentry != dir);
3107 path_put(&nd->path);
3108 nd->path = save_parent;
3109 nd->inode = dir->d_inode;
3110 save_parent.mnt = NULL;
3111 save_parent.dentry = NULL;
Al Viro64894cf2012-07-31 00:53:35 +04003112 if (got_write) {
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003113 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04003114 got_write = false;
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003115 }
3116 retried = true;
3117 goto retry_lookup;
Al Virofb1cc552009-12-24 01:58:28 -05003118}
3119
Al Viro60545d02013-06-07 01:20:27 -04003120static int do_tmpfile(int dfd, struct filename *pathname,
3121 struct nameidata *nd, int flags,
3122 const struct open_flags *op,
3123 struct file *file, int *opened)
3124{
3125 static const struct qstr name = QSTR_INIT("/", 1);
3126 struct dentry *dentry, *child;
3127 struct inode *dir;
3128 int error = path_lookupat(dfd, pathname->name,
3129 flags | LOOKUP_DIRECTORY, nd);
3130 if (unlikely(error))
3131 return error;
3132 error = mnt_want_write(nd->path.mnt);
3133 if (unlikely(error))
3134 goto out;
3135 /* we want directory to be writable */
3136 error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3137 if (error)
3138 goto out2;
3139 dentry = nd->path.dentry;
3140 dir = dentry->d_inode;
3141 if (!dir->i_op->tmpfile) {
3142 error = -EOPNOTSUPP;
3143 goto out2;
3144 }
3145 child = d_alloc(dentry, &name);
3146 if (unlikely(!child)) {
3147 error = -ENOMEM;
3148 goto out2;
3149 }
3150 nd->flags &= ~LOOKUP_DIRECTORY;
3151 nd->flags |= op->intent;
3152 dput(nd->path.dentry);
3153 nd->path.dentry = child;
3154 error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3155 if (error)
3156 goto out2;
3157 audit_inode(pathname, nd->path.dentry, 0);
3158 error = may_open(&nd->path, op->acc_mode, op->open_flag);
3159 if (error)
3160 goto out2;
3161 file->f_path.mnt = nd->path.mnt;
3162 error = finish_open(file, nd->path.dentry, NULL, opened);
3163 if (error)
3164 goto out2;
3165 error = open_check_o_direct(file);
Al Virof4e0c302013-06-11 08:34:36 +04003166 if (error) {
Al Viro60545d02013-06-07 01:20:27 -04003167 fput(file);
Al Virof4e0c302013-06-11 08:34:36 +04003168 } else if (!(op->open_flag & O_EXCL)) {
3169 struct inode *inode = file_inode(file);
3170 spin_lock(&inode->i_lock);
3171 inode->i_state |= I_LINKABLE;
3172 spin_unlock(&inode->i_lock);
3173 }
Al Viro60545d02013-06-07 01:20:27 -04003174out2:
3175 mnt_drop_write(nd->path.mnt);
3176out:
3177 path_put(&nd->path);
3178 return error;
3179}
3180
Jeff Layton669abf42012-10-10 16:43:10 -04003181static struct file *path_openat(int dfd, struct filename *pathname,
Al Viro73d049a2011-03-11 12:08:24 -05003182 struct nameidata *nd, const struct open_flags *op, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003183{
Al Virofe2d35f2011-03-05 22:58:25 -05003184 struct file *base = NULL;
Al Viro30d90492012-06-22 12:40:19 +04003185 struct file *file;
Al Viro9850c052010-01-13 15:01:15 -05003186 struct path path;
Al Viro47237682012-06-10 05:01:45 -04003187 int opened = 0;
Al Viro13aab422011-02-23 17:54:08 -05003188 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11003189
Al Viro30d90492012-06-22 12:40:19 +04003190 file = get_empty_filp();
Al Viro1afc99b2013-02-14 20:41:04 -05003191 if (IS_ERR(file))
3192 return file;
Nick Piggin31e6b012011-01-07 17:49:52 +11003193
Al Viro30d90492012-06-22 12:40:19 +04003194 file->f_flags = op->open_flag;
Nick Piggin31e6b012011-01-07 17:49:52 +11003195
Al Virobb458c62013-07-13 13:26:37 +04003196 if (unlikely(file->f_flags & __O_TMPFILE)) {
Al Viro60545d02013-06-07 01:20:27 -04003197 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3198 goto out;
3199 }
3200
Jeff Layton669abf42012-10-10 16:43:10 -04003201 error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base);
Nick Piggin31e6b012011-01-07 17:49:52 +11003202 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04003203 goto out;
Nick Piggin31e6b012011-01-07 17:49:52 +11003204
Al Virofe2d35f2011-03-05 22:58:25 -05003205 current->total_link_count = 0;
Jeff Layton669abf42012-10-10 16:43:10 -04003206 error = link_path_walk(pathname->name, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11003207 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04003208 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003209
Al Viro2675a4e2012-06-22 12:41:10 +04003210 error = do_last(nd, &path, file, op, &opened, pathname);
3211 while (unlikely(error > 0)) { /* trailing symlink */
Nick Piggin7b9337a2011-01-14 08:42:43 +00003212 struct path link = path;
Al Virodef4af32009-12-26 08:37:05 -05003213 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04003214 if (!(nd->flags & LOOKUP_FOLLOW)) {
Al Viro73d049a2011-03-11 12:08:24 -05003215 path_put_conditional(&path, nd);
3216 path_put(&nd->path);
Al Viro2675a4e2012-06-22 12:41:10 +04003217 error = -ELOOP;
Al Viro40b39132011-03-09 16:22:18 -05003218 break;
3219 }
Kees Cook800179c2012-07-25 17:29:07 -07003220 error = may_follow_link(&link, nd);
3221 if (unlikely(error))
3222 break;
Al Viro73d049a2011-03-11 12:08:24 -05003223 nd->flags |= LOOKUP_PARENT;
3224 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
Al Viro574197e2011-03-14 22:20:34 -04003225 error = follow_link(&link, nd, &cookie);
Al Viroc3e380b2011-02-23 13:39:45 -05003226 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04003227 break;
3228 error = do_last(nd, &path, file, op, &opened, pathname);
Al Viro574197e2011-03-14 22:20:34 -04003229 put_link(nd, &link, cookie);
Al Viro806b6812009-12-26 07:16:40 -05003230 }
Al Viro10fa8e62009-12-26 07:09:49 -05003231out:
Al Viro73d049a2011-03-11 12:08:24 -05003232 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
3233 path_put(&nd->root);
Al Virofe2d35f2011-03-05 22:58:25 -05003234 if (base)
3235 fput(base);
Al Viro2675a4e2012-06-22 12:41:10 +04003236 if (!(opened & FILE_OPENED)) {
3237 BUG_ON(!error);
Al Viro30d90492012-06-22 12:40:19 +04003238 put_filp(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003239 }
Al Viro2675a4e2012-06-22 12:41:10 +04003240 if (unlikely(error)) {
3241 if (error == -EOPENSTALE) {
3242 if (flags & LOOKUP_RCU)
3243 error = -ECHILD;
3244 else
3245 error = -ESTALE;
3246 }
3247 file = ERR_PTR(error);
3248 }
3249 return file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003250}
3251
Jeff Layton669abf42012-10-10 16:43:10 -04003252struct file *do_filp_open(int dfd, struct filename *pathname,
Al Virof9652e12013-06-11 08:23:01 +04003253 const struct open_flags *op)
Al Viro13aab422011-02-23 17:54:08 -05003254{
Al Viro73d049a2011-03-11 12:08:24 -05003255 struct nameidata nd;
Al Virof9652e12013-06-11 08:23:01 +04003256 int flags = op->lookup_flags;
Al Viro13aab422011-02-23 17:54:08 -05003257 struct file *filp;
3258
Al Viro73d049a2011-03-11 12:08:24 -05003259 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
Al Viro13aab422011-02-23 17:54:08 -05003260 if (unlikely(filp == ERR_PTR(-ECHILD)))
Al Viro73d049a2011-03-11 12:08:24 -05003261 filp = path_openat(dfd, pathname, &nd, op, flags);
Al Viro13aab422011-02-23 17:54:08 -05003262 if (unlikely(filp == ERR_PTR(-ESTALE)))
Al Viro73d049a2011-03-11 12:08:24 -05003263 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
Al Viro13aab422011-02-23 17:54:08 -05003264 return filp;
3265}
3266
Al Viro73d049a2011-03-11 12:08:24 -05003267struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
Al Virof9652e12013-06-11 08:23:01 +04003268 const char *name, const struct open_flags *op)
Al Viro73d049a2011-03-11 12:08:24 -05003269{
3270 struct nameidata nd;
3271 struct file *file;
Jeff Layton669abf42012-10-10 16:43:10 -04003272 struct filename filename = { .name = name };
Al Virof9652e12013-06-11 08:23:01 +04003273 int flags = op->lookup_flags | LOOKUP_ROOT;
Al Viro73d049a2011-03-11 12:08:24 -05003274
3275 nd.root.mnt = mnt;
3276 nd.root.dentry = dentry;
3277
David Howellsb18825a2013-09-12 19:22:53 +01003278 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
Al Viro73d049a2011-03-11 12:08:24 -05003279 return ERR_PTR(-ELOOP);
3280
Jeff Layton669abf42012-10-10 16:43:10 -04003281 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU);
Al Viro73d049a2011-03-11 12:08:24 -05003282 if (unlikely(file == ERR_PTR(-ECHILD)))
Jeff Layton669abf42012-10-10 16:43:10 -04003283 file = path_openat(-1, &filename, &nd, op, flags);
Al Viro73d049a2011-03-11 12:08:24 -05003284 if (unlikely(file == ERR_PTR(-ESTALE)))
Jeff Layton669abf42012-10-10 16:43:10 -04003285 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL);
Al Viro73d049a2011-03-11 12:08:24 -05003286 return file;
3287}
3288
Jeff Layton1ac12b42012-12-11 12:10:06 -05003289struct dentry *kern_path_create(int dfd, const char *pathname,
3290 struct path *path, unsigned int lookup_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003291{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003292 struct dentry *dentry = ERR_PTR(-EEXIST);
Al Viroed75e952011-06-27 16:53:43 -04003293 struct nameidata nd;
Jan Karac30dabf2012-06-12 16:20:30 +02003294 int err2;
Jeff Layton1ac12b42012-12-11 12:10:06 -05003295 int error;
3296 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3297
3298 /*
3299 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3300 * other flags passed in are ignored!
3301 */
3302 lookup_flags &= LOOKUP_REVAL;
3303
3304 error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd);
Al Viroed75e952011-06-27 16:53:43 -04003305 if (error)
3306 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003307
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003308 /*
3309 * Yucky last component or no last component at all?
3310 * (foo/., foo/.., /////)
3311 */
Al Viroed75e952011-06-27 16:53:43 -04003312 if (nd.last_type != LAST_NORM)
3313 goto out;
3314 nd.flags &= ~LOOKUP_PARENT;
3315 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003316
Jan Karac30dabf2012-06-12 16:20:30 +02003317 /* don't fail immediately if it's r/o, at least try to report other errors */
3318 err2 = mnt_want_write(nd.path.mnt);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003319 /*
3320 * Do the final lookup.
3321 */
Al Viroed75e952011-06-27 16:53:43 -04003322 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3323 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003324 if (IS_ERR(dentry))
Al Viroa8104a92012-07-20 02:25:00 +04003325 goto unlock;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003326
Al Viroa8104a92012-07-20 02:25:00 +04003327 error = -EEXIST;
David Howellsb18825a2013-09-12 19:22:53 +01003328 if (d_is_positive(dentry))
Al Viroa8104a92012-07-20 02:25:00 +04003329 goto fail;
David Howellsb18825a2013-09-12 19:22:53 +01003330
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003331 /*
3332 * Special case - lookup gave negative, but... we had foo/bar/
3333 * From the vfs_mknod() POV we just have a negative dentry -
3334 * all is fine. Let's be bastards - you had / on the end, you've
3335 * been asking for (non-existent) directory. -ENOENT for you.
3336 */
Al Viroed75e952011-06-27 16:53:43 -04003337 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
Al Viroa8104a92012-07-20 02:25:00 +04003338 error = -ENOENT;
Al Viroed75e952011-06-27 16:53:43 -04003339 goto fail;
Al Viroe9baf6e2008-05-15 04:49:12 -04003340 }
Jan Karac30dabf2012-06-12 16:20:30 +02003341 if (unlikely(err2)) {
3342 error = err2;
Al Viroa8104a92012-07-20 02:25:00 +04003343 goto fail;
Jan Karac30dabf2012-06-12 16:20:30 +02003344 }
Al Viroed75e952011-06-27 16:53:43 -04003345 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003346 return dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003347fail:
Al Viroa8104a92012-07-20 02:25:00 +04003348 dput(dentry);
3349 dentry = ERR_PTR(error);
3350unlock:
Al Viroed75e952011-06-27 16:53:43 -04003351 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Karac30dabf2012-06-12 16:20:30 +02003352 if (!err2)
3353 mnt_drop_write(nd.path.mnt);
Al Viroed75e952011-06-27 16:53:43 -04003354out:
3355 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003356 return dentry;
3357}
Al Virodae6ad82011-06-26 11:50:15 -04003358EXPORT_SYMBOL(kern_path_create);
3359
Al Viro921a1652012-07-20 01:15:31 +04003360void done_path_create(struct path *path, struct dentry *dentry)
3361{
3362 dput(dentry);
3363 mutex_unlock(&path->dentry->d_inode->i_mutex);
Al Viroa8104a92012-07-20 02:25:00 +04003364 mnt_drop_write(path->mnt);
Al Viro921a1652012-07-20 01:15:31 +04003365 path_put(path);
3366}
3367EXPORT_SYMBOL(done_path_create);
3368
Jeff Layton1ac12b42012-12-11 12:10:06 -05003369struct dentry *user_path_create(int dfd, const char __user *pathname,
3370 struct path *path, unsigned int lookup_flags)
Al Virodae6ad82011-06-26 11:50:15 -04003371{
Jeff Layton91a27b22012-10-10 15:25:28 -04003372 struct filename *tmp = getname(pathname);
Al Virodae6ad82011-06-26 11:50:15 -04003373 struct dentry *res;
3374 if (IS_ERR(tmp))
3375 return ERR_CAST(tmp);
Jeff Layton1ac12b42012-12-11 12:10:06 -05003376 res = kern_path_create(dfd, tmp->name, path, lookup_flags);
Al Virodae6ad82011-06-26 11:50:15 -04003377 putname(tmp);
3378 return res;
3379}
3380EXPORT_SYMBOL(user_path_create);
3381
Al Viro1a67aaf2011-07-26 01:52:52 -04003382int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003383{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003384 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003385
3386 if (error)
3387 return error;
3388
Eric W. Biederman975d6b32011-11-13 12:16:43 -08003389 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003390 return -EPERM;
3391
Al Viroacfa4382008-12-04 10:06:33 -05003392 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003393 return -EPERM;
3394
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07003395 error = devcgroup_inode_mknod(mode, dev);
3396 if (error)
3397 return error;
3398
Linus Torvalds1da177e2005-04-16 15:20:36 -07003399 error = security_inode_mknod(dir, dentry, mode, dev);
3400 if (error)
3401 return error;
3402
Linus Torvalds1da177e2005-04-16 15:20:36 -07003403 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003404 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003405 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003406 return error;
3407}
Al Viro4d359502014-03-14 12:20:17 -04003408EXPORT_SYMBOL(vfs_mknod);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003409
Al Virof69aac02011-07-26 04:31:40 -04003410static int may_mknod(umode_t mode)
Dave Hansen463c3192008-02-15 14:37:57 -08003411{
3412 switch (mode & S_IFMT) {
3413 case S_IFREG:
3414 case S_IFCHR:
3415 case S_IFBLK:
3416 case S_IFIFO:
3417 case S_IFSOCK:
3418 case 0: /* zero mode translates to S_IFREG */
3419 return 0;
3420 case S_IFDIR:
3421 return -EPERM;
3422 default:
3423 return -EINVAL;
3424 }
3425}
3426
Al Viro8208a222011-07-25 17:32:17 -04003427SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003428 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003429{
Al Viro2ad94ae2008-07-21 09:32:51 -04003430 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003431 struct path path;
3432 int error;
Jeff Layton972567f2012-12-20 16:00:10 -05003433 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003434
Al Viro8e4bfca2012-07-20 01:17:26 +04003435 error = may_mknod(mode);
3436 if (error)
3437 return error;
Jeff Layton972567f2012-12-20 16:00:10 -05003438retry:
3439 dentry = user_path_create(dfd, filename, &path, lookup_flags);
Al Virodae6ad82011-06-26 11:50:15 -04003440 if (IS_ERR(dentry))
3441 return PTR_ERR(dentry);
Al Viro2ad94ae2008-07-21 09:32:51 -04003442
Al Virodae6ad82011-06-26 11:50:15 -04003443 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003444 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003445 error = security_path_mknod(&path, dentry, mode, dev);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003446 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04003447 goto out;
Dave Hansen463c3192008-02-15 14:37:57 -08003448 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003449 case 0: case S_IFREG:
Al Viro312b63f2012-06-10 18:09:36 -04003450 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003451 break;
3452 case S_IFCHR: case S_IFBLK:
Al Virodae6ad82011-06-26 11:50:15 -04003453 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003454 new_decode_dev(dev));
3455 break;
3456 case S_IFIFO: case S_IFSOCK:
Al Virodae6ad82011-06-26 11:50:15 -04003457 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003458 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003459 }
Al Viroa8104a92012-07-20 02:25:00 +04003460out:
Al Viro921a1652012-07-20 01:15:31 +04003461 done_path_create(&path, dentry);
Jeff Layton972567f2012-12-20 16:00:10 -05003462 if (retry_estale(error, lookup_flags)) {
3463 lookup_flags |= LOOKUP_REVAL;
3464 goto retry;
3465 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003466 return error;
3467}
3468
Al Viro8208a222011-07-25 17:32:17 -04003469SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003470{
3471 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3472}
3473
Al Viro18bb1db2011-07-26 01:41:39 -04003474int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003475{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003476 int error = may_create(dir, dentry);
Al Viro8de52772012-02-06 12:45:27 -05003477 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003478
3479 if (error)
3480 return error;
3481
Al Viroacfa4382008-12-04 10:06:33 -05003482 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003483 return -EPERM;
3484
3485 mode &= (S_IRWXUGO|S_ISVTX);
3486 error = security_inode_mkdir(dir, dentry, mode);
3487 if (error)
3488 return error;
3489
Al Viro8de52772012-02-06 12:45:27 -05003490 if (max_links && dir->i_nlink >= max_links)
3491 return -EMLINK;
3492
Linus Torvalds1da177e2005-04-16 15:20:36 -07003493 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003494 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003495 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003496 return error;
3497}
Al Viro4d359502014-03-14 12:20:17 -04003498EXPORT_SYMBOL(vfs_mkdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003499
Al Viroa218d0f2011-11-21 14:59:34 -05003500SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003501{
Dave Hansen6902d922006-09-30 23:29:01 -07003502 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003503 struct path path;
3504 int error;
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003505 unsigned int lookup_flags = LOOKUP_DIRECTORY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003506
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003507retry:
3508 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07003509 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003510 return PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003511
Al Virodae6ad82011-06-26 11:50:15 -04003512 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003513 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003514 error = security_path_mkdir(&path, dentry, mode);
Al Viroa8104a92012-07-20 02:25:00 +04003515 if (!error)
3516 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
Al Viro921a1652012-07-20 01:15:31 +04003517 done_path_create(&path, dentry);
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003518 if (retry_estale(error, lookup_flags)) {
3519 lookup_flags |= LOOKUP_REVAL;
3520 goto retry;
3521 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003522 return error;
3523}
3524
Al Viroa218d0f2011-11-21 14:59:34 -05003525SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003526{
3527 return sys_mkdirat(AT_FDCWD, pathname, mode);
3528}
3529
Linus Torvalds1da177e2005-04-16 15:20:36 -07003530/*
Sage Weila71905f2011-05-24 13:06:08 -07003531 * The dentry_unhash() helper will try to drop the dentry early: we
J. Bruce Fieldsc0d02592012-02-15 11:48:40 -05003532 * should have a usage count of 1 if we're the only user of this
Sage Weila71905f2011-05-24 13:06:08 -07003533 * dentry, and if that is true (possibly after pruning the dcache),
3534 * then we drop the dentry now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003535 *
3536 * A low-level filesystem can, if it choses, legally
3537 * do a
3538 *
3539 * if (!d_unhashed(dentry))
3540 * return -EBUSY;
3541 *
3542 * if it cannot handle the case of removing a directory
3543 * that is still in use by something else..
3544 */
3545void dentry_unhash(struct dentry *dentry)
3546{
Vasily Averindc168422006-12-06 20:37:07 -08003547 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003548 spin_lock(&dentry->d_lock);
Waiman Long98474232013-08-28 18:24:59 -07003549 if (dentry->d_lockref.count == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003550 __d_drop(dentry);
3551 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003552}
Al Viro4d359502014-03-14 12:20:17 -04003553EXPORT_SYMBOL(dentry_unhash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003554
3555int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3556{
3557 int error = may_delete(dir, dentry, 1);
3558
3559 if (error)
3560 return error;
3561
Al Viroacfa4382008-12-04 10:06:33 -05003562 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003563 return -EPERM;
3564
Al Viro1d2ef592011-09-14 18:55:41 +01003565 dget(dentry);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003566 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003567
Sage Weil912dbc12011-05-24 13:06:11 -07003568 error = -EBUSY;
Eric W. Biederman7af13642013-10-04 19:15:13 -07003569 if (is_local_mountpoint(dentry))
Sage Weil912dbc12011-05-24 13:06:11 -07003570 goto out;
3571
3572 error = security_inode_rmdir(dir, dentry);
3573 if (error)
3574 goto out;
3575
Sage Weil3cebde22011-05-29 21:20:59 -07003576 shrink_dcache_parent(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003577 error = dir->i_op->rmdir(dir, dentry);
3578 if (error)
3579 goto out;
3580
3581 dentry->d_inode->i_flags |= S_DEAD;
3582 dont_mount(dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003583 detach_mounts(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003584
3585out:
3586 mutex_unlock(&dentry->d_inode->i_mutex);
Al Viro1d2ef592011-09-14 18:55:41 +01003587 dput(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003588 if (!error)
3589 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003590 return error;
3591}
Al Viro4d359502014-03-14 12:20:17 -04003592EXPORT_SYMBOL(vfs_rmdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003593
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003594static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003595{
3596 int error = 0;
Jeff Layton91a27b22012-10-10 15:25:28 -04003597 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003598 struct dentry *dentry;
3599 struct nameidata nd;
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003600 unsigned int lookup_flags = 0;
3601retry:
3602 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04003603 if (IS_ERR(name))
3604 return PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003605
3606 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003607 case LAST_DOTDOT:
3608 error = -ENOTEMPTY;
3609 goto exit1;
3610 case LAST_DOT:
3611 error = -EINVAL;
3612 goto exit1;
3613 case LAST_ROOT:
3614 error = -EBUSY;
3615 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003616 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003617
3618 nd.flags &= ~LOOKUP_PARENT;
Jan Karac30dabf2012-06-12 16:20:30 +02003619 error = mnt_want_write(nd.path.mnt);
3620 if (error)
3621 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003622
Jan Blunck4ac91372008-02-14 19:34:32 -08003623 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08003624 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003625 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003626 if (IS_ERR(dentry))
3627 goto exit2;
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003628 if (!dentry->d_inode) {
3629 error = -ENOENT;
3630 goto exit3;
3631 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003632 error = security_path_rmdir(&nd.path, dentry);
3633 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003634 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08003635 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Dave Hansen06227532008-02-15 14:37:34 -08003636exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07003637 dput(dentry);
3638exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08003639 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Karac30dabf2012-06-12 16:20:30 +02003640 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003641exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003642 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003643 putname(name);
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003644 if (retry_estale(error, lookup_flags)) {
3645 lookup_flags |= LOOKUP_REVAL;
3646 goto retry;
3647 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003648 return error;
3649}
3650
Heiko Carstens3cdad422009-01-14 14:14:22 +01003651SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003652{
3653 return do_rmdir(AT_FDCWD, pathname);
3654}
3655
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003656/**
3657 * vfs_unlink - unlink a filesystem object
3658 * @dir: parent directory
3659 * @dentry: victim
3660 * @delegated_inode: returns victim inode, if the inode is delegated.
3661 *
3662 * The caller must hold dir->i_mutex.
3663 *
3664 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3665 * return a reference to the inode in delegated_inode. The caller
3666 * should then break the delegation on that inode and retry. Because
3667 * breaking a delegation may take a long time, the caller should drop
3668 * dir->i_mutex before doing so.
3669 *
3670 * Alternatively, a caller may pass NULL for delegated_inode. This may
3671 * be appropriate for callers that expect the underlying filesystem not
3672 * to be NFS exported.
3673 */
3674int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003675{
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003676 struct inode *target = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003677 int error = may_delete(dir, dentry, 0);
3678
3679 if (error)
3680 return error;
3681
Al Viroacfa4382008-12-04 10:06:33 -05003682 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003683 return -EPERM;
3684
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003685 mutex_lock(&target->i_mutex);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003686 if (is_local_mountpoint(dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003687 error = -EBUSY;
3688 else {
3689 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05003690 if (!error) {
J. Bruce Fields5a146962012-08-28 07:50:40 -07003691 error = try_break_deleg(target, delegated_inode);
3692 if (error)
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003693 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003694 error = dir->i_op->unlink(dir, dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003695 if (!error) {
Al Virod83c49f2010-04-30 17:17:09 -04003696 dont_mount(dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003697 detach_mounts(dentry);
3698 }
Al Virobec10522010-03-03 14:12:08 -05003699 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003700 }
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003701out:
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003702 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003703
3704 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3705 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003706 fsnotify_link_count(target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003707 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003708 }
Robert Love0eeca282005-07-12 17:06:03 -04003709
Linus Torvalds1da177e2005-04-16 15:20:36 -07003710 return error;
3711}
Al Viro4d359502014-03-14 12:20:17 -04003712EXPORT_SYMBOL(vfs_unlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003713
3714/*
3715 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003716 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07003717 * writeout happening, and we don't want to prevent access to the directory
3718 * while waiting on the I/O.
3719 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003720static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003721{
Al Viro2ad94ae2008-07-21 09:32:51 -04003722 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04003723 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003724 struct dentry *dentry;
3725 struct nameidata nd;
3726 struct inode *inode = NULL;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003727 struct inode *delegated_inode = NULL;
Jeff Layton5d18f812012-12-20 16:38:04 -05003728 unsigned int lookup_flags = 0;
3729retry:
3730 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04003731 if (IS_ERR(name))
3732 return PTR_ERR(name);
Al Viro2ad94ae2008-07-21 09:32:51 -04003733
Linus Torvalds1da177e2005-04-16 15:20:36 -07003734 error = -EISDIR;
3735 if (nd.last_type != LAST_NORM)
3736 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003737
3738 nd.flags &= ~LOOKUP_PARENT;
Jan Karac30dabf2012-06-12 16:20:30 +02003739 error = mnt_want_write(nd.path.mnt);
3740 if (error)
3741 goto exit1;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003742retry_deleg:
Jan Blunck4ac91372008-02-14 19:34:32 -08003743 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08003744 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003745 error = PTR_ERR(dentry);
3746 if (!IS_ERR(dentry)) {
3747 /* Why not before? Because we want correct error value */
Török Edwin50338b82011-06-16 00:06:14 +03003748 if (nd.last.name[nd.last.len])
3749 goto slashes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003750 inode = dentry->d_inode;
David Howellsb18825a2013-09-12 19:22:53 +01003751 if (d_is_negative(dentry))
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003752 goto slashes;
3753 ihold(inode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003754 error = security_path_unlink(&nd.path, dentry);
3755 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003756 goto exit2;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003757 error = vfs_unlink(nd.path.dentry->d_inode, dentry, &delegated_inode);
Jan Karac30dabf2012-06-12 16:20:30 +02003758exit2:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003759 dput(dentry);
3760 }
Jan Blunck4ac91372008-02-14 19:34:32 -08003761 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003762 if (inode)
3763 iput(inode); /* truncate the inode here */
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003764 inode = NULL;
3765 if (delegated_inode) {
J. Bruce Fields5a146962012-08-28 07:50:40 -07003766 error = break_deleg_wait(&delegated_inode);
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003767 if (!error)
3768 goto retry_deleg;
3769 }
Jan Karac30dabf2012-06-12 16:20:30 +02003770 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003771exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003772 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003773 putname(name);
Jeff Layton5d18f812012-12-20 16:38:04 -05003774 if (retry_estale(error, lookup_flags)) {
3775 lookup_flags |= LOOKUP_REVAL;
3776 inode = NULL;
3777 goto retry;
3778 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003779 return error;
3780
3781slashes:
David Howellsb18825a2013-09-12 19:22:53 +01003782 if (d_is_negative(dentry))
3783 error = -ENOENT;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003784 else if (d_is_dir(dentry))
David Howellsb18825a2013-09-12 19:22:53 +01003785 error = -EISDIR;
3786 else
3787 error = -ENOTDIR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003788 goto exit2;
3789}
3790
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003791SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003792{
3793 if ((flag & ~AT_REMOVEDIR) != 0)
3794 return -EINVAL;
3795
3796 if (flag & AT_REMOVEDIR)
3797 return do_rmdir(dfd, pathname);
3798
3799 return do_unlinkat(dfd, pathname);
3800}
3801
Heiko Carstens3480b252009-01-14 14:14:16 +01003802SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003803{
3804 return do_unlinkat(AT_FDCWD, pathname);
3805}
3806
Miklos Szeredidb2e7472008-06-24 16:50:16 +02003807int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003808{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003809 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003810
3811 if (error)
3812 return error;
3813
Al Viroacfa4382008-12-04 10:06:33 -05003814 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003815 return -EPERM;
3816
3817 error = security_inode_symlink(dir, dentry, oldname);
3818 if (error)
3819 return error;
3820
Linus Torvalds1da177e2005-04-16 15:20:36 -07003821 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003822 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003823 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003824 return error;
3825}
Al Viro4d359502014-03-14 12:20:17 -04003826EXPORT_SYMBOL(vfs_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003827
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003828SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3829 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003830{
Al Viro2ad94ae2008-07-21 09:32:51 -04003831 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04003832 struct filename *from;
Dave Hansen6902d922006-09-30 23:29:01 -07003833 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003834 struct path path;
Jeff Laytonf46d3562012-12-11 12:10:08 -05003835 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003836
3837 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04003838 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003839 return PTR_ERR(from);
Jeff Laytonf46d3562012-12-11 12:10:08 -05003840retry:
3841 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07003842 error = PTR_ERR(dentry);
3843 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003844 goto out_putname;
Dave Hansen6902d922006-09-30 23:29:01 -07003845
Jeff Layton91a27b22012-10-10 15:25:28 -04003846 error = security_path_symlink(&path, dentry, from->name);
Al Viroa8104a92012-07-20 02:25:00 +04003847 if (!error)
Jeff Layton91a27b22012-10-10 15:25:28 -04003848 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
Al Viro921a1652012-07-20 01:15:31 +04003849 done_path_create(&path, dentry);
Jeff Laytonf46d3562012-12-11 12:10:08 -05003850 if (retry_estale(error, lookup_flags)) {
3851 lookup_flags |= LOOKUP_REVAL;
3852 goto retry;
3853 }
Dave Hansen6902d922006-09-30 23:29:01 -07003854out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003855 putname(from);
3856 return error;
3857}
3858
Heiko Carstens3480b252009-01-14 14:14:16 +01003859SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003860{
3861 return sys_symlinkat(oldname, AT_FDCWD, newname);
3862}
3863
J. Bruce Fields146a8592011-09-20 17:14:31 -04003864/**
3865 * vfs_link - create a new link
3866 * @old_dentry: object to be linked
3867 * @dir: new parent
3868 * @new_dentry: where to create the new link
3869 * @delegated_inode: returns inode needing a delegation break
3870 *
3871 * The caller must hold dir->i_mutex
3872 *
3873 * If vfs_link discovers a delegation on the to-be-linked file in need
3874 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3875 * inode in delegated_inode. The caller should then break the delegation
3876 * and retry. Because breaking a delegation may take a long time, the
3877 * caller should drop the i_mutex before doing so.
3878 *
3879 * Alternatively, a caller may pass NULL for delegated_inode. This may
3880 * be appropriate for callers that expect the underlying filesystem not
3881 * to be NFS exported.
3882 */
3883int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003884{
3885 struct inode *inode = old_dentry->d_inode;
Al Viro8de52772012-02-06 12:45:27 -05003886 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003887 int error;
3888
3889 if (!inode)
3890 return -ENOENT;
3891
Miklos Szeredia95164d2008-07-30 15:08:48 +02003892 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003893 if (error)
3894 return error;
3895
3896 if (dir->i_sb != inode->i_sb)
3897 return -EXDEV;
3898
3899 /*
3900 * A link to an append-only or immutable file cannot be created.
3901 */
3902 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3903 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05003904 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003905 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003906 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003907 return -EPERM;
3908
3909 error = security_inode_link(old_dentry, dir, new_dentry);
3910 if (error)
3911 return error;
3912
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003913 mutex_lock(&inode->i_mutex);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05303914 /* Make sure we don't allow creating hardlink to an unlinked file */
Al Virof4e0c302013-06-11 08:34:36 +04003915 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05303916 error = -ENOENT;
Al Viro8de52772012-02-06 12:45:27 -05003917 else if (max_links && inode->i_nlink >= max_links)
3918 error = -EMLINK;
J. Bruce Fields146a8592011-09-20 17:14:31 -04003919 else {
3920 error = try_break_deleg(inode, delegated_inode);
3921 if (!error)
3922 error = dir->i_op->link(old_dentry, dir, new_dentry);
3923 }
Al Virof4e0c302013-06-11 08:34:36 +04003924
3925 if (!error && (inode->i_state & I_LINKABLE)) {
3926 spin_lock(&inode->i_lock);
3927 inode->i_state &= ~I_LINKABLE;
3928 spin_unlock(&inode->i_lock);
3929 }
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003930 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07003931 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003932 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003933 return error;
3934}
Al Viro4d359502014-03-14 12:20:17 -04003935EXPORT_SYMBOL(vfs_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003936
3937/*
3938 * Hardlinks are often used in delicate situations. We avoid
3939 * security-related surprises by not following symlinks on the
3940 * newname. --KAB
3941 *
3942 * We don't follow them on the oldname either to be compatible
3943 * with linux 2.0, and to avoid hard-linking to directories
3944 * and other special files. --ADM
3945 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003946SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3947 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003948{
3949 struct dentry *new_dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003950 struct path old_path, new_path;
J. Bruce Fields146a8592011-09-20 17:14:31 -04003951 struct inode *delegated_inode = NULL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303952 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003953 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003954
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303955 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003956 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303957 /*
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07003958 * To use null names we require CAP_DAC_READ_SEARCH
3959 * This ensures that not everyone will be able to create
3960 * handlink using the passed filedescriptor.
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303961 */
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07003962 if (flags & AT_EMPTY_PATH) {
3963 if (!capable(CAP_DAC_READ_SEARCH))
3964 return -ENOENT;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303965 how = LOOKUP_EMPTY;
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07003966 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003967
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303968 if (flags & AT_SYMLINK_FOLLOW)
3969 how |= LOOKUP_FOLLOW;
Jeff Layton442e31c2012-12-20 16:15:38 -05003970retry:
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303971 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003972 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04003973 return error;
3974
Jeff Layton442e31c2012-12-20 16:15:38 -05003975 new_dentry = user_path_create(newdfd, newname, &new_path,
3976 (how & LOOKUP_REVAL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003977 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003978 if (IS_ERR(new_dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003979 goto out;
3980
3981 error = -EXDEV;
3982 if (old_path.mnt != new_path.mnt)
3983 goto out_dput;
Kees Cook800179c2012-07-25 17:29:07 -07003984 error = may_linkat(&old_path);
3985 if (unlikely(error))
3986 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04003987 error = security_path_link(old_path.dentry, &new_path, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003988 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04003989 goto out_dput;
J. Bruce Fields146a8592011-09-20 17:14:31 -04003990 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
Dave Hansen75c3f292008-02-15 14:37:45 -08003991out_dput:
Al Viro921a1652012-07-20 01:15:31 +04003992 done_path_create(&new_path, new_dentry);
J. Bruce Fields146a8592011-09-20 17:14:31 -04003993 if (delegated_inode) {
3994 error = break_deleg_wait(&delegated_inode);
Oleg Drokind22e6332014-01-31 15:41:58 -05003995 if (!error) {
3996 path_put(&old_path);
J. Bruce Fields146a8592011-09-20 17:14:31 -04003997 goto retry;
Oleg Drokind22e6332014-01-31 15:41:58 -05003998 }
J. Bruce Fields146a8592011-09-20 17:14:31 -04003999 }
Jeff Layton442e31c2012-12-20 16:15:38 -05004000 if (retry_estale(error, how)) {
Oleg Drokind22e6332014-01-31 15:41:58 -05004001 path_put(&old_path);
Jeff Layton442e31c2012-12-20 16:15:38 -05004002 how |= LOOKUP_REVAL;
4003 goto retry;
4004 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004005out:
Al Viro2d8f3032008-07-22 09:59:21 -04004006 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004007
4008 return error;
4009}
4010
Heiko Carstens3480b252009-01-14 14:14:16 +01004011SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004012{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004013 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004014}
4015
Miklos Szeredibc270272014-04-01 17:08:42 +02004016/**
4017 * vfs_rename - rename a filesystem object
4018 * @old_dir: parent of source
4019 * @old_dentry: source
4020 * @new_dir: parent of destination
4021 * @new_dentry: destination
4022 * @delegated_inode: returns an inode needing a delegation break
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004023 * @flags: rename flags
Miklos Szeredibc270272014-04-01 17:08:42 +02004024 *
4025 * The caller must hold multiple mutexes--see lock_rename()).
4026 *
4027 * If vfs_rename discovers a delegation in need of breaking at either
4028 * the source or destination, it will return -EWOULDBLOCK and return a
4029 * reference to the inode in delegated_inode. The caller should then
4030 * break the delegation and retry. Because breaking a delegation may
4031 * take a long time, the caller should drop all locks before doing
4032 * so.
4033 *
4034 * Alternatively, a caller may pass NULL for delegated_inode. This may
4035 * be appropriate for callers that expect the underlying filesystem not
4036 * to be NFS exported.
4037 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07004038 * The worst of all namespace operations - renaming directory. "Perverted"
4039 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4040 * Problems:
J. Bruce Fieldsd03b29a2014-02-17 16:52:33 -05004041 * a) we can get into loop creation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004042 * b) race potential - two innocent renames can create a loop together.
4043 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004044 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07004045 * story.
J. Bruce Fields6cedba82012-03-05 11:40:41 -05004046 * c) we have to lock _four_ objects - parents and victim (if it exists),
4047 * and source (if it is not a directory).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004048 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07004049 * whether the target exists). Solution: try to be smart with locking
4050 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004051 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07004052 * move will be locked. Thus we can rank directories by the tree
4053 * (ancestors first) and rank all non-directories after them.
4054 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004055 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004056 * HOWEVER, it relies on the assumption that any object with ->lookup()
4057 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4058 * we'd better make sure that there's no link(2) for them.
Sage Weile4eaac02011-05-24 13:06:07 -07004059 * d) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004060 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07004061 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05004062 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07004063 * locking].
4064 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004065int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004066 struct inode *new_dir, struct dentry *new_dentry,
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004067 struct inode **delegated_inode, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004068{
4069 int error;
Miklos Szeredibc270272014-04-01 17:08:42 +02004070 bool is_dir = d_is_dir(old_dentry);
Eric Paris59b0df22010-02-08 12:53:52 -05004071 const unsigned char *old_name;
Miklos Szeredibc270272014-04-01 17:08:42 +02004072 struct inode *source = old_dentry->d_inode;
4073 struct inode *target = new_dentry->d_inode;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004074 bool new_is_dir = false;
4075 unsigned max_links = new_dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004076
Miklos Szeredibc270272014-04-01 17:08:42 +02004077 if (source == target)
4078 return 0;
4079
Linus Torvalds1da177e2005-04-16 15:20:36 -07004080 error = may_delete(old_dir, old_dentry, is_dir);
4081 if (error)
4082 return error;
4083
Miklos Szeredida1ce062014-04-01 17:08:43 +02004084 if (!target) {
Miklos Szeredia95164d2008-07-30 15:08:48 +02004085 error = may_create(new_dir, new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004086 } else {
4087 new_is_dir = d_is_dir(new_dentry);
4088
4089 if (!(flags & RENAME_EXCHANGE))
4090 error = may_delete(new_dir, new_dentry, is_dir);
4091 else
4092 error = may_delete(new_dir, new_dentry, new_is_dir);
4093 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004094 if (error)
4095 return error;
4096
Miklos Szeredi7177a9c2014-07-23 15:15:30 +02004097 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004098 return -EPERM;
4099
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004100 if (flags && !old_dir->i_op->rename2)
4101 return -EINVAL;
4102
Miklos Szeredibc270272014-04-01 17:08:42 +02004103 /*
4104 * If we are going to change the parent - check write permissions,
4105 * we'll need to flip '..'.
4106 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02004107 if (new_dir != old_dir) {
4108 if (is_dir) {
4109 error = inode_permission(source, MAY_WRITE);
4110 if (error)
4111 return error;
4112 }
4113 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4114 error = inode_permission(target, MAY_WRITE);
4115 if (error)
4116 return error;
4117 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004118 }
Robert Love0eeca282005-07-12 17:06:03 -04004119
Miklos Szeredi0b3974e2014-04-01 17:08:43 +02004120 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4121 flags);
Miklos Szeredibc270272014-04-01 17:08:42 +02004122 if (error)
4123 return error;
4124
4125 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4126 dget(new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004127 if (!is_dir || (flags & RENAME_EXCHANGE))
Miklos Szeredibc270272014-04-01 17:08:42 +02004128 lock_two_nondirectories(source, target);
4129 else if (target)
4130 mutex_lock(&target->i_mutex);
4131
4132 error = -EBUSY;
Eric W. Biederman7af13642013-10-04 19:15:13 -07004133 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
Miklos Szeredibc270272014-04-01 17:08:42 +02004134 goto out;
4135
Miklos Szeredida1ce062014-04-01 17:08:43 +02004136 if (max_links && new_dir != old_dir) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004137 error = -EMLINK;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004138 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
Miklos Szeredibc270272014-04-01 17:08:42 +02004139 goto out;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004140 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4141 old_dir->i_nlink >= max_links)
4142 goto out;
4143 }
4144 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4145 shrink_dcache_parent(new_dentry);
4146 if (!is_dir) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004147 error = try_break_deleg(source, delegated_inode);
4148 if (error)
4149 goto out;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004150 }
4151 if (target && !new_is_dir) {
4152 error = try_break_deleg(target, delegated_inode);
4153 if (error)
4154 goto out;
Miklos Szeredibc270272014-04-01 17:08:42 +02004155 }
Miklos Szeredi7177a9c2014-07-23 15:15:30 +02004156 if (!old_dir->i_op->rename2) {
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004157 error = old_dir->i_op->rename(old_dir, old_dentry,
4158 new_dir, new_dentry);
4159 } else {
Miklos Szeredi7177a9c2014-07-23 15:15:30 +02004160 WARN_ON(old_dir->i_op->rename != NULL);
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004161 error = old_dir->i_op->rename2(old_dir, old_dentry,
4162 new_dir, new_dentry, flags);
4163 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004164 if (error)
4165 goto out;
4166
Miklos Szeredida1ce062014-04-01 17:08:43 +02004167 if (!(flags & RENAME_EXCHANGE) && target) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004168 if (is_dir)
4169 target->i_flags |= S_DEAD;
4170 dont_mount(new_dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07004171 detach_mounts(new_dentry);
Miklos Szeredibc270272014-04-01 17:08:42 +02004172 }
Miklos Szeredida1ce062014-04-01 17:08:43 +02004173 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4174 if (!(flags & RENAME_EXCHANGE))
4175 d_move(old_dentry, new_dentry);
4176 else
4177 d_exchange(old_dentry, new_dentry);
4178 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004179out:
Miklos Szeredida1ce062014-04-01 17:08:43 +02004180 if (!is_dir || (flags & RENAME_EXCHANGE))
Miklos Szeredibc270272014-04-01 17:08:42 +02004181 unlock_two_nondirectories(source, target);
4182 else if (target)
4183 mutex_unlock(&target->i_mutex);
4184 dput(new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004185 if (!error) {
Al Viro123df292009-12-25 04:57:57 -05004186 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Miklos Szeredida1ce062014-04-01 17:08:43 +02004187 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4188 if (flags & RENAME_EXCHANGE) {
4189 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4190 new_is_dir, NULL, new_dentry);
4191 }
4192 }
Robert Love0eeca282005-07-12 17:06:03 -04004193 fsnotify_oldname_free(old_name);
4194
Linus Torvalds1da177e2005-04-16 15:20:36 -07004195 return error;
4196}
Al Viro4d359502014-03-14 12:20:17 -04004197EXPORT_SYMBOL(vfs_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004198
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004199SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4200 int, newdfd, const char __user *, newname, unsigned int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004201{
Al Viro2ad94ae2008-07-21 09:32:51 -04004202 struct dentry *old_dir, *new_dir;
4203 struct dentry *old_dentry, *new_dentry;
4204 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004205 struct nameidata oldnd, newnd;
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004206 struct inode *delegated_inode = NULL;
Jeff Layton91a27b22012-10-10 15:25:28 -04004207 struct filename *from;
4208 struct filename *to;
Jeff Laytonc6a94282012-12-11 12:10:10 -05004209 unsigned int lookup_flags = 0;
4210 bool should_retry = false;
Al Viro2ad94ae2008-07-21 09:32:51 -04004211 int error;
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004212
Miklos Szeredida1ce062014-04-01 17:08:43 +02004213 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE))
4214 return -EINVAL;
4215
4216 if ((flags & RENAME_NOREPLACE) && (flags & RENAME_EXCHANGE))
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004217 return -EINVAL;
4218
Jeff Laytonc6a94282012-12-11 12:10:10 -05004219retry:
4220 from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04004221 if (IS_ERR(from)) {
4222 error = PTR_ERR(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004223 goto exit;
Jeff Layton91a27b22012-10-10 15:25:28 -04004224 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004225
Jeff Laytonc6a94282012-12-11 12:10:10 -05004226 to = user_path_parent(newdfd, newname, &newnd, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04004227 if (IS_ERR(to)) {
4228 error = PTR_ERR(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004229 goto exit1;
Jeff Layton91a27b22012-10-10 15:25:28 -04004230 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004231
4232 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08004233 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004234 goto exit2;
4235
Jan Blunck4ac91372008-02-14 19:34:32 -08004236 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004237 error = -EBUSY;
4238 if (oldnd.last_type != LAST_NORM)
4239 goto exit2;
4240
Jan Blunck4ac91372008-02-14 19:34:32 -08004241 new_dir = newnd.path.dentry;
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004242 if (flags & RENAME_NOREPLACE)
4243 error = -EEXIST;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004244 if (newnd.last_type != LAST_NORM)
4245 goto exit2;
4246
Jan Karac30dabf2012-06-12 16:20:30 +02004247 error = mnt_want_write(oldnd.path.mnt);
4248 if (error)
4249 goto exit2;
4250
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09004251 oldnd.flags &= ~LOOKUP_PARENT;
4252 newnd.flags &= ~LOOKUP_PARENT;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004253 if (!(flags & RENAME_EXCHANGE))
4254 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09004255
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004256retry_deleg:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004257 trap = lock_rename(new_dir, old_dir);
4258
Christoph Hellwig49705b72005-11-08 21:35:06 -08004259 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004260 error = PTR_ERR(old_dentry);
4261 if (IS_ERR(old_dentry))
4262 goto exit3;
4263 /* source must exist */
4264 error = -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01004265 if (d_is_negative(old_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004266 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08004267 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004268 error = PTR_ERR(new_dentry);
4269 if (IS_ERR(new_dentry))
4270 goto exit4;
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004271 error = -EEXIST;
4272 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4273 goto exit5;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004274 if (flags & RENAME_EXCHANGE) {
4275 error = -ENOENT;
4276 if (d_is_negative(new_dentry))
4277 goto exit5;
4278
4279 if (!d_is_dir(new_dentry)) {
4280 error = -ENOTDIR;
4281 if (newnd.last.name[newnd.last.len])
4282 goto exit5;
4283 }
4284 }
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004285 /* unless the source is a directory trailing slashes give -ENOTDIR */
4286 if (!d_is_dir(old_dentry)) {
4287 error = -ENOTDIR;
4288 if (oldnd.last.name[oldnd.last.len])
4289 goto exit5;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004290 if (!(flags & RENAME_EXCHANGE) && newnd.last.name[newnd.last.len])
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004291 goto exit5;
4292 }
4293 /* source should not be ancestor of target */
4294 error = -EINVAL;
4295 if (old_dentry == trap)
4296 goto exit5;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004297 /* target should not be an ancestor of source */
Miklos Szeredida1ce062014-04-01 17:08:43 +02004298 if (!(flags & RENAME_EXCHANGE))
4299 error = -ENOTEMPTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004300 if (new_dentry == trap)
4301 goto exit5;
4302
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004303 error = security_path_rename(&oldnd.path, old_dentry,
Miklos Szeredi0b3974e2014-04-01 17:08:43 +02004304 &newnd.path, new_dentry, flags);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004305 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02004306 goto exit5;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004307 error = vfs_rename(old_dir->d_inode, old_dentry,
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004308 new_dir->d_inode, new_dentry,
4309 &delegated_inode, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004310exit5:
4311 dput(new_dentry);
4312exit4:
4313 dput(old_dentry);
4314exit3:
4315 unlock_rename(new_dir, old_dir);
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004316 if (delegated_inode) {
4317 error = break_deleg_wait(&delegated_inode);
4318 if (!error)
4319 goto retry_deleg;
4320 }
Jan Karac30dabf2012-06-12 16:20:30 +02004321 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004322exit2:
Jeff Laytonc6a94282012-12-11 12:10:10 -05004323 if (retry_estale(error, lookup_flags))
4324 should_retry = true;
Jan Blunck1d957f92008-02-14 19:34:35 -08004325 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04004326 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004327exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08004328 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004329 putname(from);
Jeff Laytonc6a94282012-12-11 12:10:10 -05004330 if (should_retry) {
4331 should_retry = false;
4332 lookup_flags |= LOOKUP_REVAL;
4333 goto retry;
4334 }
Al Viro2ad94ae2008-07-21 09:32:51 -04004335exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004336 return error;
4337}
4338
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004339SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4340 int, newdfd, const char __user *, newname)
4341{
4342 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4343}
4344
Heiko Carstensa26eab22009-01-14 14:14:17 +01004345SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004346{
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004347 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004348}
4349
Al Viro5d826c82014-03-14 13:42:45 -04004350int readlink_copy(char __user *buffer, int buflen, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004351{
Al Viro5d826c82014-03-14 13:42:45 -04004352 int len = PTR_ERR(link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004353 if (IS_ERR(link))
4354 goto out;
4355
4356 len = strlen(link);
4357 if (len > (unsigned) buflen)
4358 len = buflen;
4359 if (copy_to_user(buffer, link, len))
4360 len = -EFAULT;
4361out:
4362 return len;
4363}
Al Viro5d826c82014-03-14 13:42:45 -04004364EXPORT_SYMBOL(readlink_copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004365
4366/*
4367 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4368 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4369 * using) it for any given inode is up to filesystem.
4370 */
4371int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4372{
4373 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004374 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07004375 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004376
Linus Torvalds1da177e2005-04-16 15:20:36 -07004377 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004378 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07004379 if (IS_ERR(cookie))
4380 return PTR_ERR(cookie);
4381
Al Viro5d826c82014-03-14 13:42:45 -04004382 res = readlink_copy(buffer, buflen, nd_get_link(&nd));
Marcin Slusarz694a1762008-06-09 16:40:37 -07004383 if (dentry->d_inode->i_op->put_link)
4384 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
4385 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004386}
Al Viro4d359502014-03-14 12:20:17 -04004387EXPORT_SYMBOL(generic_readlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004388
Linus Torvalds1da177e2005-04-16 15:20:36 -07004389/* get the link contents into pagecache */
4390static char *page_getlink(struct dentry * dentry, struct page **ppage)
4391{
Duane Griffinebd09ab2008-12-19 20:47:12 +00004392 char *kaddr;
4393 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004394 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07004395 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004396 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07004397 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004398 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00004399 kaddr = kmap(page);
4400 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4401 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004402}
4403
4404int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4405{
4406 struct page *page = NULL;
Al Viro5d826c82014-03-14 13:42:45 -04004407 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004408 if (page) {
4409 kunmap(page);
4410 page_cache_release(page);
4411 }
4412 return res;
4413}
Al Viro4d359502014-03-14 12:20:17 -04004414EXPORT_SYMBOL(page_readlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004415
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004416void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004417{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004418 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004419 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004420 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004421}
Al Viro4d359502014-03-14 12:20:17 -04004422EXPORT_SYMBOL(page_follow_link_light);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004423
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004424void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004425{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004426 struct page *page = cookie;
4427
4428 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004429 kunmap(page);
4430 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004431 }
4432}
Al Viro4d359502014-03-14 12:20:17 -04004433EXPORT_SYMBOL(page_put_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004434
Nick Piggin54566b22009-01-04 12:00:53 -08004435/*
4436 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4437 */
4438int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004439{
4440 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004441 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07004442 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08004443 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004444 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08004445 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4446 if (nofs)
4447 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004448
NeilBrown7e53cac2006-03-25 03:07:57 -08004449retry:
Nick Pigginafddba42007-10-16 01:25:01 -07004450 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08004451 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004452 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07004453 goto fail;
4454
Cong Wange8e3c3d2011-11-25 23:14:27 +08004455 kaddr = kmap_atomic(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004456 memcpy(kaddr, symname, len-1);
Cong Wange8e3c3d2011-11-25 23:14:27 +08004457 kunmap_atomic(kaddr);
Nick Pigginafddba42007-10-16 01:25:01 -07004458
4459 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4460 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004461 if (err < 0)
4462 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07004463 if (err < len-1)
4464 goto retry;
4465
Linus Torvalds1da177e2005-04-16 15:20:36 -07004466 mark_inode_dirty(inode);
4467 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004468fail:
4469 return err;
4470}
Al Viro4d359502014-03-14 12:20:17 -04004471EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004472
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004473int page_symlink(struct inode *inode, const char *symname, int len)
4474{
4475 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08004476 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004477}
Al Viro4d359502014-03-14 12:20:17 -04004478EXPORT_SYMBOL(page_symlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004479
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08004480const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004481 .readlink = generic_readlink,
4482 .follow_link = page_follow_link_light,
4483 .put_link = page_put_link,
4484};
Linus Torvalds1da177e2005-04-16 15:20:36 -07004485EXPORT_SYMBOL(page_symlink_inode_operations);