blob: db5fe86319e69d51e59ff6dd0053d199b406bdc9 [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}
Miklos Szeredibd5d0852014-10-24 00:14:35 +0200419EXPORT_SYMBOL(__inode_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420
Al Virof4d6ff82011-06-19 13:14:21 -0400421/**
David Howells0bdaea92012-06-25 12:55:46 +0100422 * sb_permission - Check superblock-level permissions
423 * @sb: Superblock of inode to check permission on
Randy Dunlap55852632012-08-18 17:39:25 -0700424 * @inode: Inode to check permission on
David Howells0bdaea92012-06-25 12:55:46 +0100425 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
426 *
427 * Separate out file-system wide checks from inode-specific permission checks.
428 */
429static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
430{
431 if (unlikely(mask & MAY_WRITE)) {
432 umode_t mode = inode->i_mode;
433
434 /* Nobody gets write access to a read-only fs. */
435 if ((sb->s_flags & MS_RDONLY) &&
436 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
437 return -EROFS;
438 }
439 return 0;
440}
441
442/**
443 * inode_permission - Check for access rights to a given inode
444 * @inode: Inode to check permission on
445 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
446 *
447 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
448 * this, letting us set arbitrary permissions for filesystem access without
449 * changing the "normal" UIDs which are used for other things.
450 *
451 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
452 */
453int inode_permission(struct inode *inode, int mask)
454{
455 int retval;
456
457 retval = sb_permission(inode->i_sb, inode, mask);
458 if (retval)
459 return retval;
460 return __inode_permission(inode, mask);
461}
Al Viro4d359502014-03-14 12:20:17 -0400462EXPORT_SYMBOL(inode_permission);
David Howells0bdaea92012-06-25 12:55:46 +0100463
464/**
Jan Blunck5dd784d02008-02-14 19:34:38 -0800465 * path_get - get a reference to a path
466 * @path: path to get the reference to
467 *
468 * Given a path increment the reference count to the dentry and the vfsmount.
469 */
Al Virodcf787f2013-03-01 23:51:07 -0500470void path_get(const struct path *path)
Jan Blunck5dd784d02008-02-14 19:34:38 -0800471{
472 mntget(path->mnt);
473 dget(path->dentry);
474}
475EXPORT_SYMBOL(path_get);
476
477/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800478 * path_put - put a reference to a path
479 * @path: path to put the reference to
480 *
481 * Given a path decrement the reference count to the dentry and the vfsmount.
482 */
Al Virodcf787f2013-03-01 23:51:07 -0500483void path_put(const struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484{
Jan Blunck1d957f92008-02-14 19:34:35 -0800485 dput(path->dentry);
486 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487}
Jan Blunck1d957f92008-02-14 19:34:35 -0800488EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489
Al Viro19660af2011-03-25 10:32:48 -0400490/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100491 * Path walking has 2 modes, rcu-walk and ref-walk (see
Al Viro19660af2011-03-25 10:32:48 -0400492 * Documentation/filesystems/path-lookup.txt). In situations when we can't
493 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
494 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
495 * mode. Refcounts are grabbed at the last known good point before rcu-walk
496 * got stuck, so ref-walk may continue from there. If this is not successful
497 * (eg. a seqcount has changed), then failure is returned and it's up to caller
498 * to restart the path walk from the beginning in ref-walk mode.
Nick Piggin31e6b012011-01-07 17:49:52 +1100499 */
Nick Piggin31e6b012011-01-07 17:49:52 +1100500
501/**
Al Viro19660af2011-03-25 10:32:48 -0400502 * unlazy_walk - try to switch to ref-walk mode.
503 * @nd: nameidata pathwalk data
504 * @dentry: child of nd->path.dentry or NULL
Randy Dunlap39191622011-01-08 19:36:21 -0800505 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100506 *
Al Viro19660af2011-03-25 10:32:48 -0400507 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
508 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
509 * @nd or NULL. Must be called from rcu-walk context.
Nick Piggin31e6b012011-01-07 17:49:52 +1100510 */
Al Viro19660af2011-03-25 10:32:48 -0400511static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
Nick Piggin31e6b012011-01-07 17:49:52 +1100512{
513 struct fs_struct *fs = current->fs;
514 struct dentry *parent = nd->path.dentry;
515
516 BUG_ON(!(nd->flags & LOOKUP_RCU));
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700517
518 /*
Al Viro48a066e2013-09-29 22:06:07 -0400519 * After legitimizing the bastards, terminate_walk()
520 * will do the right thing for non-RCU mode, and all our
521 * subsequent exit cases should rcu_read_unlock()
522 * before returning. Do vfsmount first; if dentry
523 * can't be legitimized, just set nd->path.dentry to NULL
524 * and rely on dput(NULL) being a no-op.
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700525 */
Al Viro48a066e2013-09-29 22:06:07 -0400526 if (!legitimize_mnt(nd->path.mnt, nd->m_seq))
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700527 return -ECHILD;
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700528 nd->flags &= ~LOOKUP_RCU;
Linus Torvalds15570082013-09-02 11:38:06 -0700529
Al Viro48a066e2013-09-29 22:06:07 -0400530 if (!lockref_get_not_dead(&parent->d_lockref)) {
531 nd->path.dentry = NULL;
Al Virod870b4a2013-11-29 01:48:32 -0500532 goto out;
Al Viro48a066e2013-09-29 22:06:07 -0400533 }
534
Linus Torvalds15570082013-09-02 11:38:06 -0700535 /*
536 * For a negative lookup, the lookup sequence point is the parents
537 * sequence point, and it only needs to revalidate the parent dentry.
538 *
539 * For a positive lookup, we need to move both the parent and the
540 * dentry from the RCU domain to be properly refcounted. And the
541 * sequence number in the dentry validates *both* dentry counters,
542 * since we checked the sequence number of the parent after we got
543 * the child sequence number. So we know the parent must still
544 * be valid if the child sequence number is still valid.
545 */
Al Viro19660af2011-03-25 10:32:48 -0400546 if (!dentry) {
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700547 if (read_seqcount_retry(&parent->d_seq, nd->seq))
548 goto out;
Al Viro19660af2011-03-25 10:32:48 -0400549 BUG_ON(nd->inode != parent->d_inode);
550 } else {
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700551 if (!lockref_get_not_dead(&dentry->d_lockref))
552 goto out;
553 if (read_seqcount_retry(&dentry->d_seq, nd->seq))
554 goto drop_dentry;
Al Viro19660af2011-03-25 10:32:48 -0400555 }
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700556
557 /*
558 * Sequence counts matched. Now make sure that the root is
559 * still valid and get it if required.
560 */
561 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
562 spin_lock(&fs->lock);
563 if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry)
564 goto unlock_and_drop_dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +1100565 path_get(&nd->root);
566 spin_unlock(&fs->lock);
567 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100568
Al Viro8b61e742013-11-08 12:45:01 -0500569 rcu_read_unlock();
Nick Piggin31e6b012011-01-07 17:49:52 +1100570 return 0;
Al Viro19660af2011-03-25 10:32:48 -0400571
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700572unlock_and_drop_dentry:
573 spin_unlock(&fs->lock);
574drop_dentry:
Al Viro8b61e742013-11-08 12:45:01 -0500575 rcu_read_unlock();
Linus Torvalds15570082013-09-02 11:38:06 -0700576 dput(dentry);
Linus Torvaldsd0d27272013-09-10 12:17:49 -0700577 goto drop_root_mnt;
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700578out:
Al Viro8b61e742013-11-08 12:45:01 -0500579 rcu_read_unlock();
Linus Torvaldsd0d27272013-09-10 12:17:49 -0700580drop_root_mnt:
581 if (!(nd->flags & LOOKUP_ROOT))
582 nd->root.mnt = NULL;
Nick Piggin31e6b012011-01-07 17:49:52 +1100583 return -ECHILD;
584}
585
Al Viro4ce16ef32012-06-10 16:10:59 -0400586static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
Nick Piggin34286d62011-01-07 17:49:57 +1100587{
Al Viro4ce16ef32012-06-10 16:10:59 -0400588 return dentry->d_op->d_revalidate(dentry, flags);
Nick Piggin34286d62011-01-07 17:49:57 +1100589}
590
Al Viro9f1fafe2011-03-25 11:00:12 -0400591/**
592 * complete_walk - successful completion of path walk
593 * @nd: pointer nameidata
Jeff Layton39159de2009-12-07 12:01:50 -0500594 *
Al Viro9f1fafe2011-03-25 11:00:12 -0400595 * If we had been in RCU mode, drop out of it and legitimize nd->path.
596 * Revalidate the final result, unless we'd already done that during
597 * the path walk or the filesystem doesn't ask for it. Return 0 on
598 * success, -error on failure. In case of failure caller does not
599 * need to drop nd->path.
Jeff Layton39159de2009-12-07 12:01:50 -0500600 */
Al Viro9f1fafe2011-03-25 11:00:12 -0400601static int complete_walk(struct nameidata *nd)
Jeff Layton39159de2009-12-07 12:01:50 -0500602{
Al Viro16c2cd72011-02-22 15:50:10 -0500603 struct dentry *dentry = nd->path.dentry;
Jeff Layton39159de2009-12-07 12:01:50 -0500604 int status;
Jeff Layton39159de2009-12-07 12:01:50 -0500605
Al Viro9f1fafe2011-03-25 11:00:12 -0400606 if (nd->flags & LOOKUP_RCU) {
607 nd->flags &= ~LOOKUP_RCU;
608 if (!(nd->flags & LOOKUP_ROOT))
609 nd->root.mnt = NULL;
Linus Torvalds15570082013-09-02 11:38:06 -0700610
Al Viro48a066e2013-09-29 22:06:07 -0400611 if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) {
Al Viro8b61e742013-11-08 12:45:01 -0500612 rcu_read_unlock();
Al Viro48a066e2013-09-29 22:06:07 -0400613 return -ECHILD;
614 }
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700615 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) {
Al Viro8b61e742013-11-08 12:45:01 -0500616 rcu_read_unlock();
Al Viro48a066e2013-09-29 22:06:07 -0400617 mntput(nd->path.mnt);
Al Viro9f1fafe2011-03-25 11:00:12 -0400618 return -ECHILD;
619 }
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700620 if (read_seqcount_retry(&dentry->d_seq, nd->seq)) {
Al Viro8b61e742013-11-08 12:45:01 -0500621 rcu_read_unlock();
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700622 dput(dentry);
Al Viro48a066e2013-09-29 22:06:07 -0400623 mntput(nd->path.mnt);
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700624 return -ECHILD;
625 }
Al Viro8b61e742013-11-08 12:45:01 -0500626 rcu_read_unlock();
Al Viro9f1fafe2011-03-25 11:00:12 -0400627 }
628
Al Viro16c2cd72011-02-22 15:50:10 -0500629 if (likely(!(nd->flags & LOOKUP_JUMPED)))
Jeff Layton39159de2009-12-07 12:01:50 -0500630 return 0;
631
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500632 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
Al Viro16c2cd72011-02-22 15:50:10 -0500633 return 0;
634
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500635 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
Jeff Layton39159de2009-12-07 12:01:50 -0500636 if (status > 0)
637 return 0;
638
Al Viro16c2cd72011-02-22 15:50:10 -0500639 if (!status)
Jeff Layton39159de2009-12-07 12:01:50 -0500640 status = -ESTALE;
Al Viro16c2cd72011-02-22 15:50:10 -0500641
Al Viro9f1fafe2011-03-25 11:00:12 -0400642 path_put(&nd->path);
Jeff Layton39159de2009-12-07 12:01:50 -0500643 return status;
644}
645
Al Viro2a737872009-04-07 11:49:53 -0400646static __always_inline void set_root(struct nameidata *nd)
647{
Al Viro7bd88372014-09-13 21:55:46 -0400648 get_fs_root(current->fs, &nd->root);
Al Viro2a737872009-04-07 11:49:53 -0400649}
650
Al Viro6de88d72009-08-09 01:41:57 +0400651static int link_path_walk(const char *, struct nameidata *);
652
Al Viro7bd88372014-09-13 21:55:46 -0400653static __always_inline unsigned set_root_rcu(struct nameidata *nd)
Nick Piggin31e6b012011-01-07 17:49:52 +1100654{
Al Viro7bd88372014-09-13 21:55:46 -0400655 struct fs_struct *fs = current->fs;
656 unsigned seq, res;
Nick Pigginc28cc362011-01-07 17:49:53 +1100657
Al Viro7bd88372014-09-13 21:55:46 -0400658 do {
659 seq = read_seqcount_begin(&fs->seq);
660 nd->root = fs->root;
661 res = __read_seqcount_begin(&nd->root.dentry->d_seq);
662 } while (read_seqcount_retry(&fs->seq, seq));
663 return res;
Nick Piggin31e6b012011-01-07 17:49:52 +1100664}
665
Jan Blunck1d957f92008-02-14 19:34:35 -0800666static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700667{
668 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800669 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700670 mntput(path->mnt);
671}
672
Nick Piggin7b9337a2011-01-14 08:42:43 +0000673static inline void path_to_nameidata(const struct path *path,
674 struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700675{
Nick Piggin31e6b012011-01-07 17:49:52 +1100676 if (!(nd->flags & LOOKUP_RCU)) {
677 dput(nd->path.dentry);
678 if (nd->path.mnt != path->mnt)
679 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800680 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100681 nd->path.mnt = path->mnt;
Jan Blunck4ac91372008-02-14 19:34:32 -0800682 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700683}
684
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400685/*
686 * Helper to directly jump to a known parsed path from ->follow_link,
687 * caller must have taken a reference to path beforehand.
688 */
689void nd_jump_link(struct nameidata *nd, struct path *path)
690{
691 path_put(&nd->path);
692
693 nd->path = *path;
694 nd->inode = nd->path.dentry->d_inode;
695 nd->flags |= LOOKUP_JUMPED;
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400696}
697
Al Viro574197e2011-03-14 22:20:34 -0400698static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
699{
700 struct inode *inode = link->dentry->d_inode;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400701 if (inode->i_op->put_link)
Al Viro574197e2011-03-14 22:20:34 -0400702 inode->i_op->put_link(link->dentry, nd, cookie);
703 path_put(link);
704}
705
Linus Torvalds561ec642012-10-26 10:05:07 -0700706int sysctl_protected_symlinks __read_mostly = 0;
707int sysctl_protected_hardlinks __read_mostly = 0;
Kees Cook800179c2012-07-25 17:29:07 -0700708
709/**
710 * may_follow_link - Check symlink following for unsafe situations
711 * @link: The path of the symlink
Randy Dunlap55852632012-08-18 17:39:25 -0700712 * @nd: nameidata pathwalk data
Kees Cook800179c2012-07-25 17:29:07 -0700713 *
714 * In the case of the sysctl_protected_symlinks sysctl being enabled,
715 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
716 * in a sticky world-writable directory. This is to protect privileged
717 * processes from failing races against path names that may change out
718 * from under them by way of other users creating malicious symlinks.
719 * It will permit symlinks to be followed only when outside a sticky
720 * world-writable directory, or when the uid of the symlink and follower
721 * match, or when the directory owner matches the symlink's owner.
722 *
723 * Returns 0 if following the symlink is allowed, -ve on error.
724 */
725static inline int may_follow_link(struct path *link, struct nameidata *nd)
726{
727 const struct inode *inode;
728 const struct inode *parent;
729
730 if (!sysctl_protected_symlinks)
731 return 0;
732
733 /* Allowed if owner and follower match. */
734 inode = link->dentry->d_inode;
Eric W. Biederman81abe272012-08-03 09:38:08 -0700735 if (uid_eq(current_cred()->fsuid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700736 return 0;
737
738 /* Allowed if parent directory not sticky and world-writable. */
739 parent = nd->path.dentry->d_inode;
740 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
741 return 0;
742
743 /* Allowed if parent directory and link owner match. */
Eric W. Biederman81abe272012-08-03 09:38:08 -0700744 if (uid_eq(parent->i_uid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700745 return 0;
746
Sasha Levinffd8d102012-10-04 19:56:40 -0400747 audit_log_link_denied("follow_link", link);
Kees Cook800179c2012-07-25 17:29:07 -0700748 path_put_conditional(link, nd);
749 path_put(&nd->path);
750 return -EACCES;
751}
752
753/**
754 * safe_hardlink_source - Check for safe hardlink conditions
755 * @inode: the source inode to hardlink from
756 *
757 * Return false if at least one of the following conditions:
758 * - inode is not a regular file
759 * - inode is setuid
760 * - inode is setgid and group-exec
761 * - access failure for read and write
762 *
763 * Otherwise returns true.
764 */
765static bool safe_hardlink_source(struct inode *inode)
766{
767 umode_t mode = inode->i_mode;
768
769 /* Special files should not get pinned to the filesystem. */
770 if (!S_ISREG(mode))
771 return false;
772
773 /* Setuid files should not get pinned to the filesystem. */
774 if (mode & S_ISUID)
775 return false;
776
777 /* Executable setgid files should not get pinned to the filesystem. */
778 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
779 return false;
780
781 /* Hardlinking to unreadable or unwritable sources is dangerous. */
782 if (inode_permission(inode, MAY_READ | MAY_WRITE))
783 return false;
784
785 return true;
786}
787
788/**
789 * may_linkat - Check permissions for creating a hardlink
790 * @link: the source to hardlink from
791 *
792 * Block hardlink when all of:
793 * - sysctl_protected_hardlinks enabled
794 * - fsuid does not match inode
795 * - hardlink source is unsafe (see safe_hardlink_source() above)
796 * - not CAP_FOWNER
797 *
798 * Returns 0 if successful, -ve on error.
799 */
800static int may_linkat(struct path *link)
801{
802 const struct cred *cred;
803 struct inode *inode;
804
805 if (!sysctl_protected_hardlinks)
806 return 0;
807
808 cred = current_cred();
809 inode = link->dentry->d_inode;
810
811 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
812 * otherwise, it must be a safe source.
813 */
Eric W. Biederman81abe272012-08-03 09:38:08 -0700814 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
Kees Cook800179c2012-07-25 17:29:07 -0700815 capable(CAP_FOWNER))
816 return 0;
817
Kees Cooka51d9ea2012-07-25 17:29:08 -0700818 audit_log_link_denied("linkat", link);
Kees Cook800179c2012-07-25 17:29:07 -0700819 return -EPERM;
820}
821
Al Virodef4af32009-12-26 08:37:05 -0500822static __always_inline int
Al Viro574197e2011-03-14 22:20:34 -0400823follow_link(struct path *link, struct nameidata *nd, void **p)
Ian Kent051d3812006-03-27 01:14:53 -0800824{
Nick Piggin7b9337a2011-01-14 08:42:43 +0000825 struct dentry *dentry = link->dentry;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400826 int error;
827 char *s;
Ian Kent051d3812006-03-27 01:14:53 -0800828
Al Viro844a3912011-02-15 00:38:26 -0500829 BUG_ON(nd->flags & LOOKUP_RCU);
830
Al Viro0e794582011-03-16 02:45:02 -0400831 if (link->mnt == nd->path.mnt)
832 mntget(link->mnt);
833
Al Viro6d7b5aa2012-06-10 04:15:17 -0400834 error = -ELOOP;
835 if (unlikely(current->total_link_count >= 40))
836 goto out_put_nd_path;
837
Al Viro574197e2011-03-14 22:20:34 -0400838 cond_resched();
839 current->total_link_count++;
840
Al Viro68ac1232012-03-15 08:21:57 -0400841 touch_atime(link);
Ian Kent051d3812006-03-27 01:14:53 -0800842 nd_set_link(nd, NULL);
843
Al Viro36f3b4f2011-02-22 21:24:38 -0500844 error = security_inode_follow_link(link->dentry, nd);
Al Viro6d7b5aa2012-06-10 04:15:17 -0400845 if (error)
846 goto out_put_nd_path;
Al Viro36f3b4f2011-02-22 21:24:38 -0500847
Al Viro86acdca12009-12-22 23:45:11 -0500848 nd->last_type = LAST_BIND;
Al Virodef4af32009-12-26 08:37:05 -0500849 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
850 error = PTR_ERR(*p);
Al Viro6d7b5aa2012-06-10 04:15:17 -0400851 if (IS_ERR(*p))
Christoph Hellwig408ef012012-06-18 10:47:03 -0400852 goto out_put_nd_path;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400853
854 error = 0;
855 s = nd_get_link(nd);
856 if (s) {
Al Viro443ed252013-09-10 12:00:43 -0400857 if (unlikely(IS_ERR(s))) {
858 path_put(&nd->path);
859 put_link(nd, link, *p);
860 return PTR_ERR(s);
861 }
862 if (*s == '/') {
Al Viro7bd88372014-09-13 21:55:46 -0400863 if (!nd->root.mnt)
864 set_root(nd);
Al Viro443ed252013-09-10 12:00:43 -0400865 path_put(&nd->path);
866 nd->path = nd->root;
867 path_get(&nd->root);
868 nd->flags |= LOOKUP_JUMPED;
869 }
870 nd->inode = nd->path.dentry->d_inode;
871 error = link_path_walk(s, nd);
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400872 if (unlikely(error))
873 put_link(nd, link, *p);
Ian Kent051d3812006-03-27 01:14:53 -0800874 }
Al Viro6d7b5aa2012-06-10 04:15:17 -0400875
876 return error;
877
878out_put_nd_path:
Arnd Bergmann98f6ef62012-10-11 13:20:00 +0000879 *p = NULL;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400880 path_put(&nd->path);
Al Viro6d7b5aa2012-06-10 04:15:17 -0400881 path_put(link);
Ian Kent051d3812006-03-27 01:14:53 -0800882 return error;
883}
884
Nick Piggin31e6b012011-01-07 17:49:52 +1100885static int follow_up_rcu(struct path *path)
886{
Al Viro0714a532011-11-24 22:19:58 -0500887 struct mount *mnt = real_mount(path->mnt);
888 struct mount *parent;
Nick Piggin31e6b012011-01-07 17:49:52 +1100889 struct dentry *mountpoint;
890
Al Viro0714a532011-11-24 22:19:58 -0500891 parent = mnt->mnt_parent;
892 if (&parent->mnt == path->mnt)
Nick Piggin31e6b012011-01-07 17:49:52 +1100893 return 0;
Al Viroa73324d2011-11-24 22:25:07 -0500894 mountpoint = mnt->mnt_mountpoint;
Nick Piggin31e6b012011-01-07 17:49:52 +1100895 path->dentry = mountpoint;
Al Viro0714a532011-11-24 22:19:58 -0500896 path->mnt = &parent->mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +1100897 return 1;
898}
899
David Howellsf015f1262012-06-25 12:55:28 +0100900/*
901 * follow_up - Find the mountpoint of path's vfsmount
902 *
903 * Given a path, find the mountpoint of its source file system.
904 * Replace @path with the path of the mountpoint in the parent mount.
905 * Up is towards /.
906 *
907 * Return 1 if we went up a level and 0 if we were already at the
908 * root.
909 */
Al Virobab77eb2009-04-18 03:26:48 -0400910int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911{
Al Viro0714a532011-11-24 22:19:58 -0500912 struct mount *mnt = real_mount(path->mnt);
913 struct mount *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +1000915
Al Viro48a066e2013-09-29 22:06:07 -0400916 read_seqlock_excl(&mount_lock);
Al Viro0714a532011-11-24 22:19:58 -0500917 parent = mnt->mnt_parent;
Al Viro3c0a6162012-07-18 17:32:50 +0400918 if (parent == mnt) {
Al Viro48a066e2013-09-29 22:06:07 -0400919 read_sequnlock_excl(&mount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 return 0;
921 }
Al Viro0714a532011-11-24 22:19:58 -0500922 mntget(&parent->mnt);
Al Viroa73324d2011-11-24 22:25:07 -0500923 mountpoint = dget(mnt->mnt_mountpoint);
Al Viro48a066e2013-09-29 22:06:07 -0400924 read_sequnlock_excl(&mount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400925 dput(path->dentry);
926 path->dentry = mountpoint;
927 mntput(path->mnt);
Al Viro0714a532011-11-24 22:19:58 -0500928 path->mnt = &parent->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 return 1;
930}
Al Viro4d359502014-03-14 12:20:17 -0400931EXPORT_SYMBOL(follow_up);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100933/*
David Howells9875cf82011-01-14 18:45:21 +0000934 * Perform an automount
935 * - return -EISDIR to tell follow_managed() to stop and return the path we
936 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 */
David Howells9875cf82011-01-14 18:45:21 +0000938static int follow_automount(struct path *path, unsigned flags,
939 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +1100940{
David Howells9875cf82011-01-14 18:45:21 +0000941 struct vfsmount *mnt;
David Howellsea5b7782011-01-14 19:10:03 +0000942 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +1100943
David Howells9875cf82011-01-14 18:45:21 +0000944 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
945 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -0700946
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200947 /* We don't want to mount if someone's just doing a stat -
948 * unless they're stat'ing a directory and appended a '/' to
949 * the name.
950 *
951 * We do, however, want to mount if someone wants to open or
952 * create a file of any type under the mountpoint, wants to
953 * traverse through the mountpoint or wants to open the
954 * mounted directory. Also, autofs may mark negative dentries
955 * as being automount points. These will need the attentions
956 * of the daemon to instantiate them before they can be used.
David Howells9875cf82011-01-14 18:45:21 +0000957 */
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200958 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
Linus Torvaldsd94c1772011-09-26 17:44:55 -0700959 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200960 path->dentry->d_inode)
961 return -EISDIR;
962
David Howells9875cf82011-01-14 18:45:21 +0000963 current->total_link_count++;
964 if (current->total_link_count >= 40)
965 return -ELOOP;
966
967 mnt = path->dentry->d_op->d_automount(path);
968 if (IS_ERR(mnt)) {
969 /*
970 * The filesystem is allowed to return -EISDIR here to indicate
971 * it doesn't want to automount. For instance, autofs would do
972 * this so that its userspace daemon can mount on this dentry.
973 *
974 * However, we can only permit this if it's a terminal point in
975 * the path being looked up; if it wasn't then the remainder of
976 * the path is inaccessible and we should say so.
977 */
Al Viro49084c32011-06-25 21:59:52 -0400978 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
David Howells9875cf82011-01-14 18:45:21 +0000979 return -EREMOTE;
980 return PTR_ERR(mnt);
981 }
David Howellsea5b7782011-01-14 19:10:03 +0000982
David Howells9875cf82011-01-14 18:45:21 +0000983 if (!mnt) /* mount collision */
984 return 0;
985
Al Viro8aef1882011-06-16 15:10:06 +0100986 if (!*need_mntput) {
987 /* lock_mount() may release path->mnt on error */
988 mntget(path->mnt);
989 *need_mntput = true;
990 }
Al Viro19a167a2011-01-17 01:35:23 -0500991 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +0000992
David Howellsea5b7782011-01-14 19:10:03 +0000993 switch (err) {
994 case -EBUSY:
995 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -0500996 return 0;
David Howellsea5b7782011-01-14 19:10:03 +0000997 case 0:
Al Viro8aef1882011-06-16 15:10:06 +0100998 path_put(path);
David Howellsea5b7782011-01-14 19:10:03 +0000999 path->mnt = mnt;
1000 path->dentry = dget(mnt->mnt_root);
David Howellsea5b7782011-01-14 19:10:03 +00001001 return 0;
Al Viro19a167a2011-01-17 01:35:23 -05001002 default:
1003 return err;
David Howellsea5b7782011-01-14 19:10:03 +00001004 }
Al Viro19a167a2011-01-17 01:35:23 -05001005
David Howells9875cf82011-01-14 18:45:21 +00001006}
1007
1008/*
1009 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +00001010 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +00001011 * - Flagged as mountpoint
1012 * - Flagged as automount point
1013 *
1014 * This may only be called in refwalk mode.
1015 *
1016 * Serialization is taken care of in namespace.c
1017 */
1018static int follow_managed(struct path *path, unsigned flags)
1019{
Al Viro8aef1882011-06-16 15:10:06 +01001020 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
David Howells9875cf82011-01-14 18:45:21 +00001021 unsigned managed;
1022 bool need_mntput = false;
Al Viro8aef1882011-06-16 15:10:06 +01001023 int ret = 0;
David Howells9875cf82011-01-14 18:45:21 +00001024
1025 /* Given that we're not holding a lock here, we retain the value in a
1026 * local variable for each dentry as we look at it so that we don't see
1027 * the components of that value change under us */
1028 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1029 managed &= DCACHE_MANAGED_DENTRY,
1030 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +00001031 /* Allow the filesystem to manage the transit without i_mutex
1032 * being held. */
1033 if (managed & DCACHE_MANAGE_TRANSIT) {
1034 BUG_ON(!path->dentry->d_op);
1035 BUG_ON(!path->dentry->d_op->d_manage);
Al Viro1aed3e42011-03-18 09:09:02 -04001036 ret = path->dentry->d_op->d_manage(path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001037 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001038 break;
David Howellscc53ce52011-01-14 18:45:26 +00001039 }
1040
David Howells9875cf82011-01-14 18:45:21 +00001041 /* Transit to a mounted filesystem. */
1042 if (managed & DCACHE_MOUNTED) {
1043 struct vfsmount *mounted = lookup_mnt(path);
1044 if (mounted) {
1045 dput(path->dentry);
1046 if (need_mntput)
1047 mntput(path->mnt);
1048 path->mnt = mounted;
1049 path->dentry = dget(mounted->mnt_root);
1050 need_mntput = true;
1051 continue;
1052 }
1053
1054 /* Something is mounted on this dentry in another
1055 * namespace and/or whatever was mounted there in this
Al Viro48a066e2013-09-29 22:06:07 -04001056 * namespace got unmounted before lookup_mnt() could
1057 * get it */
David Howells9875cf82011-01-14 18:45:21 +00001058 }
1059
1060 /* Handle an automount point */
1061 if (managed & DCACHE_NEED_AUTOMOUNT) {
1062 ret = follow_automount(path, flags, &need_mntput);
1063 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001064 break;
David Howells9875cf82011-01-14 18:45:21 +00001065 continue;
1066 }
1067
1068 /* We didn't change the current path point */
1069 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 }
Al Viro8aef1882011-06-16 15:10:06 +01001071
1072 if (need_mntput && path->mnt == mnt)
1073 mntput(path->mnt);
1074 if (ret == -EISDIR)
1075 ret = 0;
Al Viroa3fbbde2011-11-07 21:21:26 +00001076 return ret < 0 ? ret : need_mntput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077}
1078
David Howellscc53ce52011-01-14 18:45:26 +00001079int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080{
1081 struct vfsmount *mounted;
1082
Al Viro1c755af2009-04-18 14:06:57 -04001083 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -04001085 dput(path->dentry);
1086 mntput(path->mnt);
1087 path->mnt = mounted;
1088 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 return 1;
1090 }
1091 return 0;
1092}
Al Viro4d359502014-03-14 12:20:17 -04001093EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094
NeilBrownb8faf032014-08-04 17:06:29 +10001095static inline int managed_dentry_rcu(struct dentry *dentry)
Ian Kent62a73752011-03-25 01:51:02 +08001096{
NeilBrownb8faf032014-08-04 17:06:29 +10001097 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1098 dentry->d_op->d_manage(dentry, true) : 0;
Ian Kent62a73752011-03-25 01:51:02 +08001099}
1100
David Howells9875cf82011-01-14 18:45:21 +00001101/*
Al Viro287548e2011-05-27 06:50:06 -04001102 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1103 * we meet a managed dentry that would need blocking.
David Howells9875cf82011-01-14 18:45:21 +00001104 */
1105static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
Al Viro287548e2011-05-27 06:50:06 -04001106 struct inode **inode)
David Howells9875cf82011-01-14 18:45:21 +00001107{
Ian Kent62a73752011-03-25 01:51:02 +08001108 for (;;) {
Al Viroc7105362011-11-24 18:22:03 -05001109 struct mount *mounted;
Ian Kent62a73752011-03-25 01:51:02 +08001110 /*
1111 * Don't forget we might have a non-mountpoint managed dentry
1112 * that wants to block transit.
1113 */
NeilBrownb8faf032014-08-04 17:06:29 +10001114 switch (managed_dentry_rcu(path->dentry)) {
1115 case -ECHILD:
1116 default:
David Howellsab909112011-01-14 18:46:51 +00001117 return false;
NeilBrownb8faf032014-08-04 17:06:29 +10001118 case -EISDIR:
1119 return true;
1120 case 0:
1121 break;
1122 }
Ian Kent62a73752011-03-25 01:51:02 +08001123
1124 if (!d_mountpoint(path->dentry))
NeilBrownb8faf032014-08-04 17:06:29 +10001125 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
Ian Kent62a73752011-03-25 01:51:02 +08001126
Al Viro474279d2013-10-01 16:11:26 -04001127 mounted = __lookup_mnt(path->mnt, path->dentry);
David Howells9875cf82011-01-14 18:45:21 +00001128 if (!mounted)
1129 break;
Al Viroc7105362011-11-24 18:22:03 -05001130 path->mnt = &mounted->mnt;
1131 path->dentry = mounted->mnt.mnt_root;
Al Viroa3fbbde2011-11-07 21:21:26 +00001132 nd->flags |= LOOKUP_JUMPED;
David Howells9875cf82011-01-14 18:45:21 +00001133 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
Linus Torvalds59430262011-07-18 15:43:29 -07001134 /*
1135 * Update the inode too. We don't need to re-check the
1136 * dentry sequence number here after this d_inode read,
1137 * because a mount-point is always pinned.
1138 */
1139 *inode = path->dentry->d_inode;
David Howells9875cf82011-01-14 18:45:21 +00001140 }
Al Virof5be3e29122014-09-13 21:50:45 -04001141 return !read_seqretry(&mount_lock, nd->m_seq) &&
NeilBrownb8faf032014-08-04 17:06:29 +10001142 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
Al Viro287548e2011-05-27 06:50:06 -04001143}
1144
Nick Piggin31e6b012011-01-07 17:49:52 +11001145static int follow_dotdot_rcu(struct nameidata *nd)
1146{
Al Viro4023bfc2014-09-13 21:59:43 -04001147 struct inode *inode = nd->inode;
Al Viro7bd88372014-09-13 21:55:46 -04001148 if (!nd->root.mnt)
1149 set_root_rcu(nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11001150
David Howells9875cf82011-01-14 18:45:21 +00001151 while (1) {
Nick Piggin31e6b012011-01-07 17:49:52 +11001152 if (nd->path.dentry == nd->root.dentry &&
1153 nd->path.mnt == nd->root.mnt) {
1154 break;
1155 }
1156 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1157 struct dentry *old = nd->path.dentry;
1158 struct dentry *parent = old->d_parent;
1159 unsigned seq;
1160
Al Viro4023bfc2014-09-13 21:59:43 -04001161 inode = parent->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001162 seq = read_seqcount_begin(&parent->d_seq);
1163 if (read_seqcount_retry(&old->d_seq, nd->seq))
Al Viroef7562d2011-03-04 14:35:59 -05001164 goto failed;
Nick Piggin31e6b012011-01-07 17:49:52 +11001165 nd->path.dentry = parent;
1166 nd->seq = seq;
1167 break;
1168 }
1169 if (!follow_up_rcu(&nd->path))
1170 break;
Al Viro4023bfc2014-09-13 21:59:43 -04001171 inode = nd->path.dentry->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001172 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11001173 }
Al Virob37199e2014-03-20 15:18:22 -04001174 while (d_mountpoint(nd->path.dentry)) {
1175 struct mount *mounted;
1176 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1177 if (!mounted)
1178 break;
1179 nd->path.mnt = &mounted->mnt;
1180 nd->path.dentry = mounted->mnt.mnt_root;
Al Viro4023bfc2014-09-13 21:59:43 -04001181 inode = nd->path.dentry->d_inode;
Al Virob37199e2014-03-20 15:18:22 -04001182 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Virof5be3e29122014-09-13 21:50:45 -04001183 if (read_seqretry(&mount_lock, nd->m_seq))
Al Virob37199e2014-03-20 15:18:22 -04001184 goto failed;
1185 }
Al Viro4023bfc2014-09-13 21:59:43 -04001186 nd->inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001187 return 0;
Al Viroef7562d2011-03-04 14:35:59 -05001188
1189failed:
1190 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001191 if (!(nd->flags & LOOKUP_ROOT))
1192 nd->root.mnt = NULL;
Al Viro8b61e742013-11-08 12:45:01 -05001193 rcu_read_unlock();
Al Viroef7562d2011-03-04 14:35:59 -05001194 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001195}
1196
David Howells9875cf82011-01-14 18:45:21 +00001197/*
David Howellscc53ce52011-01-14 18:45:26 +00001198 * Follow down to the covering mount currently visible to userspace. At each
1199 * point, the filesystem owning that dentry may be queried as to whether the
1200 * caller is permitted to proceed or not.
David Howellscc53ce52011-01-14 18:45:26 +00001201 */
Al Viro7cc90cc2011-03-18 09:04:20 -04001202int follow_down(struct path *path)
David Howellscc53ce52011-01-14 18:45:26 +00001203{
1204 unsigned managed;
1205 int ret;
1206
1207 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1208 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1209 /* Allow the filesystem to manage the transit without i_mutex
1210 * being held.
1211 *
1212 * We indicate to the filesystem if someone is trying to mount
1213 * something here. This gives autofs the chance to deny anyone
1214 * other than its daemon the right to mount on its
1215 * superstructure.
1216 *
1217 * The filesystem may sleep at this point.
1218 */
1219 if (managed & DCACHE_MANAGE_TRANSIT) {
1220 BUG_ON(!path->dentry->d_op);
1221 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +00001222 ret = path->dentry->d_op->d_manage(
Al Viro1aed3e42011-03-18 09:09:02 -04001223 path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001224 if (ret < 0)
1225 return ret == -EISDIR ? 0 : ret;
1226 }
1227
1228 /* Transit to a mounted filesystem. */
1229 if (managed & DCACHE_MOUNTED) {
1230 struct vfsmount *mounted = lookup_mnt(path);
1231 if (!mounted)
1232 break;
1233 dput(path->dentry);
1234 mntput(path->mnt);
1235 path->mnt = mounted;
1236 path->dentry = dget(mounted->mnt_root);
1237 continue;
1238 }
1239
1240 /* Don't handle automount points here */
1241 break;
1242 }
1243 return 0;
1244}
Al Viro4d359502014-03-14 12:20:17 -04001245EXPORT_SYMBOL(follow_down);
David Howellscc53ce52011-01-14 18:45:26 +00001246
1247/*
David Howells9875cf82011-01-14 18:45:21 +00001248 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1249 */
1250static void follow_mount(struct path *path)
1251{
1252 while (d_mountpoint(path->dentry)) {
1253 struct vfsmount *mounted = lookup_mnt(path);
1254 if (!mounted)
1255 break;
1256 dput(path->dentry);
1257 mntput(path->mnt);
1258 path->mnt = mounted;
1259 path->dentry = dget(mounted->mnt_root);
1260 }
1261}
1262
Nick Piggin31e6b012011-01-07 17:49:52 +11001263static void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264{
Al Viro7bd88372014-09-13 21:55:46 -04001265 if (!nd->root.mnt)
1266 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001267
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 while(1) {
Jan Blunck4ac91372008-02-14 19:34:32 -08001269 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270
Al Viro2a737872009-04-07 11:49:53 -04001271 if (nd->path.dentry == nd->root.dentry &&
1272 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 break;
1274 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001275 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Al Viro3088dd72010-01-30 15:47:29 -05001276 /* rare case of legitimate dget_parent()... */
1277 nd->path.dentry = dget_parent(nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 dput(old);
1279 break;
1280 }
Al Viro3088dd72010-01-30 15:47:29 -05001281 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 }
Al Viro79ed0222009-04-18 13:59:41 -04001284 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001285 nd->inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286}
1287
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288/*
Miklos Szeredibad61182012-03-26 12:54:24 +02001289 * This looks up the name in dcache, possibly revalidates the old dentry and
1290 * allocates a new one if not found or not valid. In the need_lookup argument
1291 * returns whether i_op->lookup is necessary.
1292 *
1293 * dir->d_inode->i_mutex must be held
Nick Pigginbaa03892010-08-18 04:37:31 +10001294 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001295static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
Al Viro201f9562012-06-22 12:42:10 +04001296 unsigned int flags, bool *need_lookup)
Nick Pigginbaa03892010-08-18 04:37:31 +10001297{
Nick Pigginbaa03892010-08-18 04:37:31 +10001298 struct dentry *dentry;
Miklos Szeredibad61182012-03-26 12:54:24 +02001299 int error;
Nick Pigginbaa03892010-08-18 04:37:31 +10001300
Miklos Szeredibad61182012-03-26 12:54:24 +02001301 *need_lookup = false;
1302 dentry = d_lookup(dir, name);
1303 if (dentry) {
Jeff Layton39e3c952012-11-28 11:30:53 -05001304 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
Al Viro201f9562012-06-22 12:42:10 +04001305 error = d_revalidate(dentry, flags);
Miklos Szeredibad61182012-03-26 12:54:24 +02001306 if (unlikely(error <= 0)) {
1307 if (error < 0) {
1308 dput(dentry);
1309 return ERR_PTR(error);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001310 } else {
1311 d_invalidate(dentry);
Miklos Szeredibad61182012-03-26 12:54:24 +02001312 dput(dentry);
1313 dentry = NULL;
1314 }
1315 }
1316 }
1317 }
Nick Pigginbaa03892010-08-18 04:37:31 +10001318
Miklos Szeredibad61182012-03-26 12:54:24 +02001319 if (!dentry) {
1320 dentry = d_alloc(dir, name);
1321 if (unlikely(!dentry))
1322 return ERR_PTR(-ENOMEM);
Nick Pigginbaa03892010-08-18 04:37:31 +10001323
Miklos Szeredibad61182012-03-26 12:54:24 +02001324 *need_lookup = true;
Nick Pigginbaa03892010-08-18 04:37:31 +10001325 }
1326 return dentry;
1327}
1328
1329/*
J. Bruce Fields13a2c3b2013-10-23 16:09:16 -04001330 * Call i_op->lookup on the dentry. The dentry must be negative and
1331 * unhashed.
Miklos Szeredibad61182012-03-26 12:54:24 +02001332 *
1333 * dir->d_inode->i_mutex must be held
Josef Bacik44396f42011-05-31 11:58:49 -04001334 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001335static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
Al Viro72bd8662012-06-10 17:17:17 -04001336 unsigned int flags)
Josef Bacik44396f42011-05-31 11:58:49 -04001337{
Josef Bacik44396f42011-05-31 11:58:49 -04001338 struct dentry *old;
1339
1340 /* Don't create child dentry for a dead directory. */
Miklos Szeredibad61182012-03-26 12:54:24 +02001341 if (unlikely(IS_DEADDIR(dir))) {
Miklos Szeredie188dc02012-02-03 14:25:18 +01001342 dput(dentry);
Josef Bacik44396f42011-05-31 11:58:49 -04001343 return ERR_PTR(-ENOENT);
Miklos Szeredie188dc02012-02-03 14:25:18 +01001344 }
Josef Bacik44396f42011-05-31 11:58:49 -04001345
Al Viro72bd8662012-06-10 17:17:17 -04001346 old = dir->i_op->lookup(dir, dentry, flags);
Josef Bacik44396f42011-05-31 11:58:49 -04001347 if (unlikely(old)) {
1348 dput(dentry);
1349 dentry = old;
1350 }
1351 return dentry;
1352}
1353
Al Viroa3255542012-03-30 14:41:51 -04001354static struct dentry *__lookup_hash(struct qstr *name,
Al Viro72bd8662012-06-10 17:17:17 -04001355 struct dentry *base, unsigned int flags)
Al Viroa3255542012-03-30 14:41:51 -04001356{
Miklos Szeredibad61182012-03-26 12:54:24 +02001357 bool need_lookup;
Al Viroa3255542012-03-30 14:41:51 -04001358 struct dentry *dentry;
1359
Al Viro72bd8662012-06-10 17:17:17 -04001360 dentry = lookup_dcache(name, base, flags, &need_lookup);
Miklos Szeredibad61182012-03-26 12:54:24 +02001361 if (!need_lookup)
1362 return dentry;
Al Viroa3255542012-03-30 14:41:51 -04001363
Al Viro72bd8662012-06-10 17:17:17 -04001364 return lookup_real(base->d_inode, dentry, flags);
Al Viroa3255542012-03-30 14:41:51 -04001365}
1366
Josef Bacik44396f42011-05-31 11:58:49 -04001367/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 * It's more convoluted than I'd like it to be, but... it's still fairly
1369 * small and for now I'd prefer to have fast path as straight as possible.
1370 * It _is_ time-critical.
1371 */
Al Viroe97cdc82013-01-24 18:16:00 -05001372static int lookup_fast(struct nameidata *nd,
Miklos Szeredi697f5142012-05-21 17:30:05 +02001373 struct path *path, struct inode **inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374{
Jan Blunck4ac91372008-02-14 19:34:32 -08001375 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001376 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001377 int need_reval = 1;
1378 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001379 int err;
1380
Al Viro3cac2602009-08-13 18:27:43 +04001381 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001382 * Rename seqlock is not required here because in the off chance
1383 * of a false negative due to a concurrent rename, we're going to
1384 * do the non-racy lookup, below.
1385 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001386 if (nd->flags & LOOKUP_RCU) {
1387 unsigned seq;
Linus Torvaldsda53be12013-05-21 15:22:44 -07001388 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
Al Viro5a18fff2011-03-11 04:44:53 -05001389 if (!dentry)
1390 goto unlazy;
1391
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001392 /*
1393 * This sequence count validates that the inode matches
1394 * the dentry name information from lookup.
1395 */
1396 *inode = dentry->d_inode;
1397 if (read_seqcount_retry(&dentry->d_seq, seq))
1398 return -ECHILD;
1399
1400 /*
1401 * This sequence count validates that the parent had no
1402 * changes while we did the lookup of the dentry above.
1403 *
1404 * The memory barrier in read_seqcount_begin of child is
1405 * enough, we can use __read_seqcount_retry here.
1406 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001407 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1408 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001409 nd->seq = seq;
Al Viro5a18fff2011-03-11 04:44:53 -05001410
Al Viro24643082011-02-15 01:26:22 -05001411 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
Al Viro4ce16ef32012-06-10 16:10:59 -04001412 status = d_revalidate(dentry, nd->flags);
Al Viro5a18fff2011-03-11 04:44:53 -05001413 if (unlikely(status <= 0)) {
1414 if (status != -ECHILD)
1415 need_reval = 0;
1416 goto unlazy;
1417 }
Al Viro24643082011-02-15 01:26:22 -05001418 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001419 path->mnt = mnt;
1420 path->dentry = dentry;
NeilBrownb8faf032014-08-04 17:06:29 +10001421 if (likely(__follow_mount_rcu(nd, path, inode)))
1422 return 0;
Al Viro5a18fff2011-03-11 04:44:53 -05001423unlazy:
Al Viro19660af2011-03-25 10:32:48 -04001424 if (unlazy_walk(nd, dentry))
1425 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001426 } else {
Al Viroe97cdc82013-01-24 18:16:00 -05001427 dentry = __d_lookup(parent, &nd->last);
Nick Piggin31e6b012011-01-07 17:49:52 +11001428 }
Al Viro5a18fff2011-03-11 04:44:53 -05001429
Al Viro81e6f522012-03-30 14:48:04 -04001430 if (unlikely(!dentry))
1431 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001432
Al Viro5a18fff2011-03-11 04:44:53 -05001433 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
Al Viro4ce16ef32012-06-10 16:10:59 -04001434 status = d_revalidate(dentry, nd->flags);
Al Viro5a18fff2011-03-11 04:44:53 -05001435 if (unlikely(status <= 0)) {
1436 if (status < 0) {
1437 dput(dentry);
1438 return status;
1439 }
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001440 d_invalidate(dentry);
1441 dput(dentry);
1442 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001443 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001444
David Howells9875cf82011-01-14 18:45:21 +00001445 path->mnt = mnt;
1446 path->dentry = dentry;
1447 err = follow_managed(path, nd->flags);
Ian Kent89312212011-01-18 12:06:10 +08001448 if (unlikely(err < 0)) {
1449 path_put_conditional(path, nd);
David Howells9875cf82011-01-14 18:45:21 +00001450 return err;
Ian Kent89312212011-01-18 12:06:10 +08001451 }
Al Viroa3fbbde2011-11-07 21:21:26 +00001452 if (err)
1453 nd->flags |= LOOKUP_JUMPED;
David Howells9875cf82011-01-14 18:45:21 +00001454 *inode = path->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 return 0;
Al Viro81e6f522012-03-30 14:48:04 -04001456
1457need_lookup:
Miklos Szeredi697f5142012-05-21 17:30:05 +02001458 return 1;
1459}
1460
1461/* Fast lookup failed, do it the slow way */
Al Virocc2a5272013-01-24 18:19:49 -05001462static int lookup_slow(struct nameidata *nd, struct path *path)
Miklos Szeredi697f5142012-05-21 17:30:05 +02001463{
1464 struct dentry *dentry, *parent;
1465 int err;
1466
1467 parent = nd->path.dentry;
Al Viro81e6f522012-03-30 14:48:04 -04001468 BUG_ON(nd->inode != parent->d_inode);
1469
1470 mutex_lock(&parent->d_inode->i_mutex);
Al Virocc2a5272013-01-24 18:19:49 -05001471 dentry = __lookup_hash(&nd->last, parent, nd->flags);
Al Viro81e6f522012-03-30 14:48:04 -04001472 mutex_unlock(&parent->d_inode->i_mutex);
1473 if (IS_ERR(dentry))
1474 return PTR_ERR(dentry);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001475 path->mnt = nd->path.mnt;
1476 path->dentry = dentry;
1477 err = follow_managed(path, nd->flags);
1478 if (unlikely(err < 0)) {
1479 path_put_conditional(path, nd);
1480 return err;
1481 }
1482 if (err)
1483 nd->flags |= LOOKUP_JUMPED;
1484 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485}
1486
Al Viro52094c82011-02-21 21:34:47 -05001487static inline int may_lookup(struct nameidata *nd)
1488{
1489 if (nd->flags & LOOKUP_RCU) {
Al Viro4ad5abb2011-06-20 19:57:03 -04001490 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
Al Viro52094c82011-02-21 21:34:47 -05001491 if (err != -ECHILD)
1492 return err;
Al Viro19660af2011-03-25 10:32:48 -04001493 if (unlazy_walk(nd, NULL))
Al Viro52094c82011-02-21 21:34:47 -05001494 return -ECHILD;
1495 }
Al Viro4ad5abb2011-06-20 19:57:03 -04001496 return inode_permission(nd->inode, MAY_EXEC);
Al Viro52094c82011-02-21 21:34:47 -05001497}
1498
Al Viro9856fa12011-03-04 14:22:06 -05001499static inline int handle_dots(struct nameidata *nd, int type)
1500{
1501 if (type == LAST_DOTDOT) {
1502 if (nd->flags & LOOKUP_RCU) {
1503 if (follow_dotdot_rcu(nd))
1504 return -ECHILD;
1505 } else
1506 follow_dotdot(nd);
1507 }
1508 return 0;
1509}
1510
Al Viro951361f2011-03-04 14:44:37 -05001511static void terminate_walk(struct nameidata *nd)
1512{
1513 if (!(nd->flags & LOOKUP_RCU)) {
1514 path_put(&nd->path);
1515 } else {
1516 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001517 if (!(nd->flags & LOOKUP_ROOT))
1518 nd->root.mnt = NULL;
Al Viro8b61e742013-11-08 12:45:01 -05001519 rcu_read_unlock();
Al Viro951361f2011-03-04 14:44:37 -05001520 }
1521}
1522
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001523/*
1524 * Do we need to follow links? We _really_ want to be able
1525 * to do this check without having to look at inode->i_op,
1526 * so we keep a cache of "no, this doesn't need follow_link"
1527 * for the common case.
1528 */
David Howellsb18825a2013-09-12 19:22:53 +01001529static inline int should_follow_link(struct dentry *dentry, int follow)
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001530{
David Howellsb18825a2013-09-12 19:22:53 +01001531 return unlikely(d_is_symlink(dentry)) ? follow : 0;
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001532}
1533
Al Viroce57dfc2011-03-13 19:58:58 -04001534static inline int walk_component(struct nameidata *nd, struct path *path,
Al Viro21b9b072013-01-24 18:10:25 -05001535 int follow)
Al Viroce57dfc2011-03-13 19:58:58 -04001536{
1537 struct inode *inode;
1538 int err;
1539 /*
1540 * "." and ".." are special - ".." especially so because it has
1541 * to be able to know about the current root directory and
1542 * parent relationships.
1543 */
Al Viro21b9b072013-01-24 18:10:25 -05001544 if (unlikely(nd->last_type != LAST_NORM))
1545 return handle_dots(nd, nd->last_type);
Al Viroe97cdc82013-01-24 18:16:00 -05001546 err = lookup_fast(nd, path, &inode);
Al Viroce57dfc2011-03-13 19:58:58 -04001547 if (unlikely(err)) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001548 if (err < 0)
1549 goto out_err;
1550
Al Virocc2a5272013-01-24 18:19:49 -05001551 err = lookup_slow(nd, path);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001552 if (err < 0)
1553 goto out_err;
1554
1555 inode = path->dentry->d_inode;
Al Viroce57dfc2011-03-13 19:58:58 -04001556 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001557 err = -ENOENT;
Al Viro22213312014-04-19 12:30:58 -04001558 if (!inode || d_is_negative(path->dentry))
Miklos Szeredi697f5142012-05-21 17:30:05 +02001559 goto out_path_put;
1560
David Howellsb18825a2013-09-12 19:22:53 +01001561 if (should_follow_link(path->dentry, follow)) {
Al Viro19660af2011-03-25 10:32:48 -04001562 if (nd->flags & LOOKUP_RCU) {
1563 if (unlikely(unlazy_walk(nd, path->dentry))) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001564 err = -ECHILD;
1565 goto out_err;
Al Viro19660af2011-03-25 10:32:48 -04001566 }
1567 }
Al Viroce57dfc2011-03-13 19:58:58 -04001568 BUG_ON(inode != path->dentry->d_inode);
1569 return 1;
1570 }
1571 path_to_nameidata(path, nd);
1572 nd->inode = inode;
1573 return 0;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001574
1575out_path_put:
1576 path_to_nameidata(path, nd);
1577out_err:
1578 terminate_walk(nd);
1579 return err;
Al Viroce57dfc2011-03-13 19:58:58 -04001580}
1581
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582/*
Al Virob3563792011-03-14 21:54:55 -04001583 * This limits recursive symlink follows to 8, while
1584 * limiting consecutive symlinks to 40.
1585 *
1586 * Without that kind of total limit, nasty chains of consecutive
1587 * symlinks can cause almost arbitrarily long lookups.
1588 */
1589static inline int nested_symlink(struct path *path, struct nameidata *nd)
1590{
1591 int res;
1592
Al Virob3563792011-03-14 21:54:55 -04001593 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1594 path_put_conditional(path, nd);
1595 path_put(&nd->path);
1596 return -ELOOP;
1597 }
Erez Zadok1a4022f2011-05-21 01:19:59 -04001598 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
Al Virob3563792011-03-14 21:54:55 -04001599
1600 nd->depth++;
1601 current->link_count++;
1602
1603 do {
1604 struct path link = *path;
1605 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04001606
1607 res = follow_link(&link, nd, &cookie);
Al Viro6d7b5aa2012-06-10 04:15:17 -04001608 if (res)
1609 break;
Al Viro21b9b072013-01-24 18:10:25 -05001610 res = walk_component(nd, path, LOOKUP_FOLLOW);
Al Viro574197e2011-03-14 22:20:34 -04001611 put_link(nd, &link, cookie);
Al Virob3563792011-03-14 21:54:55 -04001612 } while (res > 0);
1613
1614 current->link_count--;
1615 nd->depth--;
1616 return res;
1617}
1618
1619/*
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001620 * We can do the critical dentry name comparison and hashing
1621 * operations one word at a time, but we are limited to:
1622 *
1623 * - Architectures with fast unaligned word accesses. We could
1624 * do a "get_unaligned()" if this helps and is sufficiently
1625 * fast.
1626 *
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001627 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1628 * do not trap on the (extremely unlikely) case of a page
1629 * crossing operation.
1630 *
1631 * - Furthermore, we need an efficient 64-bit compile for the
1632 * 64-bit case in order to generate the "number of bytes in
1633 * the final mask". Again, that could be replaced with a
1634 * efficient population count instruction or similar.
1635 */
1636#ifdef CONFIG_DCACHE_WORD_ACCESS
1637
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001638#include <asm/word-at-a-time.h>
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001639
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001640#ifdef CONFIG_64BIT
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001641
1642static inline unsigned int fold_hash(unsigned long hash)
1643{
Linus Torvalds99d263d2014-09-13 11:30:10 -07001644 return hash_64(hash, 32);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001645}
1646
1647#else /* 32-bit case */
1648
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001649#define fold_hash(x) (x)
1650
1651#endif
1652
1653unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1654{
1655 unsigned long a, mask;
1656 unsigned long hash = 0;
1657
1658 for (;;) {
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001659 a = load_unaligned_zeropad(name);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001660 if (len < sizeof(unsigned long))
1661 break;
1662 hash += a;
Al Virof132c5b2012-03-22 21:59:52 +00001663 hash *= 9;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001664 name += sizeof(unsigned long);
1665 len -= sizeof(unsigned long);
1666 if (!len)
1667 goto done;
1668 }
Will Deacona5c21dc2013-12-12 17:40:21 +00001669 mask = bytemask_from_count(len);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001670 hash += mask & a;
1671done:
1672 return fold_hash(hash);
1673}
1674EXPORT_SYMBOL(full_name_hash);
1675
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001676/*
1677 * Calculate the length and hash of the path component, and
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001678 * return the "hash_len" as the result.
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001679 */
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001680static inline u64 hash_name(const char *name)
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001681{
Linus Torvalds36126f82012-05-26 10:43:17 -07001682 unsigned long a, b, adata, bdata, mask, hash, len;
1683 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001684
1685 hash = a = 0;
1686 len = -sizeof(unsigned long);
1687 do {
1688 hash = (hash + a) * 9;
1689 len += sizeof(unsigned long);
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001690 a = load_unaligned_zeropad(name+len);
Linus Torvalds36126f82012-05-26 10:43:17 -07001691 b = a ^ REPEAT_BYTE('/');
1692 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001693
Linus Torvalds36126f82012-05-26 10:43:17 -07001694 adata = prep_zero_mask(a, adata, &constants);
1695 bdata = prep_zero_mask(b, bdata, &constants);
1696
1697 mask = create_zero_mask(adata | bdata);
1698
1699 hash += a & zero_bytemask(mask);
Linus Torvalds9226b5b2014-09-14 17:28:32 -07001700 len += find_zero(mask);
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001701 return hashlen_create(fold_hash(hash), len);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001702}
1703
1704#else
1705
Linus Torvalds0145acc2012-03-02 14:32:59 -08001706unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1707{
1708 unsigned long hash = init_name_hash();
1709 while (len--)
1710 hash = partial_name_hash(*name++, hash);
1711 return end_name_hash(hash);
1712}
Linus Torvaldsae942ae2012-03-02 19:40:57 -08001713EXPORT_SYMBOL(full_name_hash);
Linus Torvalds0145acc2012-03-02 14:32:59 -08001714
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001715/*
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001716 * We know there's a real path component here of at least
1717 * one character.
1718 */
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001719static inline u64 hash_name(const char *name)
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001720{
1721 unsigned long hash = init_name_hash();
1722 unsigned long len = 0, c;
1723
1724 c = (unsigned char)*name;
1725 do {
1726 len++;
1727 hash = partial_name_hash(c, hash);
1728 c = (unsigned char)name[len];
1729 } while (c && c != '/');
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001730 return hashlen_create(end_name_hash(hash), len);
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001731}
1732
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001733#endif
1734
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001735/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01001737 * This is the basic name resolution function, turning a pathname into
1738 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01001740 * Returns 0 and nd will have valid dentry and mnt on success.
1741 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 */
Al Viro6de88d72009-08-09 01:41:57 +04001743static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744{
1745 struct path next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747
1748 while (*name=='/')
1749 name++;
1750 if (!*name)
Al Viro086e1832011-02-22 20:56:27 -05001751 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 /* At this point we know we have a real path component. */
1754 for(;;) {
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001755 u64 hash_len;
Al Virofe479a52011-02-22 15:10:03 -05001756 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757
Al Viro52094c82011-02-21 21:34:47 -05001758 err = may_lookup(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 if (err)
1760 break;
1761
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001762 hash_len = hash_name(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763
Al Virofe479a52011-02-22 15:10:03 -05001764 type = LAST_NORM;
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001765 if (name[0] == '.') switch (hashlen_len(hash_len)) {
Al Virofe479a52011-02-22 15:10:03 -05001766 case 2:
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001767 if (name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05001768 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05001769 nd->flags |= LOOKUP_JUMPED;
1770 }
Al Virofe479a52011-02-22 15:10:03 -05001771 break;
1772 case 1:
1773 type = LAST_DOT;
1774 }
Al Viro5a202bc2011-03-08 14:17:44 -05001775 if (likely(type == LAST_NORM)) {
1776 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05001777 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05001778 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
James Hogana060dc52014-09-16 13:07:35 +01001779 struct qstr this = { { .hash_len = hash_len }, .name = name };
Linus Torvaldsda53be12013-05-21 15:22:44 -07001780 err = parent->d_op->d_hash(parent, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05001781 if (err < 0)
1782 break;
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001783 hash_len = this.hash_len;
1784 name = this.name;
Al Viro5a202bc2011-03-08 14:17:44 -05001785 }
1786 }
Al Virofe479a52011-02-22 15:10:03 -05001787
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001788 nd->last.hash_len = hash_len;
1789 nd->last.name = name;
Al Viro5f4a6a62013-01-24 18:04:22 -05001790 nd->last_type = type;
1791
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001792 name += hashlen_len(hash_len);
1793 if (!*name)
Al Viro5f4a6a62013-01-24 18:04:22 -05001794 return 0;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001795 /*
1796 * If it wasn't NUL, we know it was '/'. Skip that
1797 * slash, and continue until no more slashes.
1798 */
1799 do {
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001800 name++;
1801 } while (unlikely(*name == '/'));
1802 if (!*name)
Al Viro5f4a6a62013-01-24 18:04:22 -05001803 return 0;
1804
Al Viro21b9b072013-01-24 18:10:25 -05001805 err = walk_component(nd, &next, LOOKUP_FOLLOW);
Al Viroce57dfc2011-03-13 19:58:58 -04001806 if (err < 0)
1807 return err;
Al Virofe479a52011-02-22 15:10:03 -05001808
Al Viroce57dfc2011-03-13 19:58:58 -04001809 if (err) {
Al Virob3563792011-03-14 21:54:55 -04001810 err = nested_symlink(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 if (err)
Al Viroa7472ba2011-03-04 14:39:30 -05001812 return err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001813 }
Miklos Szeredi44b1d532014-04-01 17:08:41 +02001814 if (!d_can_lookup(nd->path.dentry)) {
Al Viro5f4a6a62013-01-24 18:04:22 -05001815 err = -ENOTDIR;
1816 break;
1817 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818 }
Al Viro951361f2011-03-04 14:44:37 -05001819 terminate_walk(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 return err;
1821}
1822
Al Viro70e9b352011-03-05 21:12:22 -05001823static int path_init(int dfd, const char *name, unsigned int flags,
1824 struct nameidata *nd, struct file **fp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001826 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827
1828 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro16c2cd72011-02-22 15:50:10 -05001829 nd->flags = flags | LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830 nd->depth = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05001831 if (flags & LOOKUP_ROOT) {
David Howellsb18825a2013-09-12 19:22:53 +01001832 struct dentry *root = nd->root.dentry;
1833 struct inode *inode = root->d_inode;
Al Viro73d049a2011-03-11 12:08:24 -05001834 if (*name) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02001835 if (!d_can_lookup(root))
Al Viro73d049a2011-03-11 12:08:24 -05001836 return -ENOTDIR;
1837 retval = inode_permission(inode, MAY_EXEC);
1838 if (retval)
1839 return retval;
1840 }
Al Viro5b6ca022011-03-09 23:04:47 -05001841 nd->path = nd->root;
1842 nd->inode = inode;
1843 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05001844 rcu_read_lock();
Al Viro5b6ca022011-03-09 23:04:47 -05001845 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro48a066e2013-09-29 22:06:07 -04001846 nd->m_seq = read_seqbegin(&mount_lock);
Al Viro5b6ca022011-03-09 23:04:47 -05001847 } else {
1848 path_get(&nd->path);
1849 }
1850 return 0;
1851 }
1852
Al Viro2a737872009-04-07 11:49:53 -04001853 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854
Al Viro48a066e2013-09-29 22:06:07 -04001855 nd->m_seq = read_seqbegin(&mount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 if (*name=='/') {
Al Viroe41f7d42011-02-22 14:02:58 -05001857 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05001858 rcu_read_lock();
Al Viro7bd88372014-09-13 21:55:46 -04001859 nd->seq = set_root_rcu(nd);
Al Viroe41f7d42011-02-22 14:02:58 -05001860 } else {
1861 set_root(nd);
1862 path_get(&nd->root);
1863 }
Al Viro2a737872009-04-07 11:49:53 -04001864 nd->path = nd->root;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001865 } else if (dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05001866 if (flags & LOOKUP_RCU) {
1867 struct fs_struct *fs = current->fs;
1868 unsigned seq;
1869
Al Viro8b61e742013-11-08 12:45:01 -05001870 rcu_read_lock();
Al Viroe41f7d42011-02-22 14:02:58 -05001871
1872 do {
1873 seq = read_seqcount_begin(&fs->seq);
1874 nd->path = fs->pwd;
1875 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1876 } while (read_seqcount_retry(&fs->seq, seq));
1877 } else {
1878 get_fs_pwd(current->fs, &nd->path);
1879 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001880 } else {
Jeff Layton582aa642012-12-11 08:56:16 -05001881 /* Caller must check execute permissions on the starting path component */
Al Viro2903ff02012-08-28 12:52:22 -04001882 struct fd f = fdget_raw(dfd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001883 struct dentry *dentry;
1884
Al Viro2903ff02012-08-28 12:52:22 -04001885 if (!f.file)
1886 return -EBADF;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001887
Al Viro2903ff02012-08-28 12:52:22 -04001888 dentry = f.file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001889
Al Virof52e0c12011-03-14 18:56:51 -04001890 if (*name) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02001891 if (!d_can_lookup(dentry)) {
Al Viro2903ff02012-08-28 12:52:22 -04001892 fdput(f);
1893 return -ENOTDIR;
1894 }
Al Virof52e0c12011-03-14 18:56:51 -04001895 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001896
Al Viro2903ff02012-08-28 12:52:22 -04001897 nd->path = f.file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05001898 if (flags & LOOKUP_RCU) {
Linus Torvalds9c225f22014-03-03 09:36:58 -08001899 if (f.flags & FDPUT_FPUT)
Al Viro2903ff02012-08-28 12:52:22 -04001900 *fp = f.file;
Al Viroe41f7d42011-02-22 14:02:58 -05001901 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro8b61e742013-11-08 12:45:01 -05001902 rcu_read_lock();
Al Viroe41f7d42011-02-22 14:02:58 -05001903 } else {
Al Viro2903ff02012-08-28 12:52:22 -04001904 path_get(&nd->path);
1905 fdput(f);
Al Viroe41f7d42011-02-22 14:02:58 -05001906 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907 }
Al Viroe41f7d42011-02-22 14:02:58 -05001908
Nick Piggin31e6b012011-01-07 17:49:52 +11001909 nd->inode = nd->path.dentry->d_inode;
Al Viro4023bfc2014-09-13 21:59:43 -04001910 if (!(flags & LOOKUP_RCU))
1911 return 0;
1912 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
1913 return 0;
1914 if (!(nd->flags & LOOKUP_ROOT))
1915 nd->root.mnt = NULL;
1916 rcu_read_unlock();
1917 return -ECHILD;
Al Viro9b4a9b12009-04-07 11:44:16 -04001918}
1919
Al Virobd92d7f2011-03-14 19:54:59 -04001920static inline int lookup_last(struct nameidata *nd, struct path *path)
1921{
1922 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1923 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1924
1925 nd->flags &= ~LOOKUP_PARENT;
Al Viro21b9b072013-01-24 18:10:25 -05001926 return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW);
Al Virobd92d7f2011-03-14 19:54:59 -04001927}
1928
Al Viro9b4a9b12009-04-07 11:44:16 -04001929/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroee0827c2011-02-21 23:38:09 -05001930static int path_lookupat(int dfd, const char *name,
Al Viro9b4a9b12009-04-07 11:44:16 -04001931 unsigned int flags, struct nameidata *nd)
1932{
Al Viro70e9b352011-03-05 21:12:22 -05001933 struct file *base = NULL;
Al Virobd92d7f2011-03-14 19:54:59 -04001934 struct path path;
1935 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001936
1937 /*
1938 * Path walking is largely split up into 2 different synchronisation
1939 * schemes, rcu-walk and ref-walk (explained in
1940 * Documentation/filesystems/path-lookup.txt). These share much of the
1941 * path walk code, but some things particularly setup, cleanup, and
1942 * following mounts are sufficiently divergent that functions are
1943 * duplicated. Typically there is a function foo(), and its RCU
1944 * analogue, foo_rcu().
1945 *
1946 * -ECHILD is the error number of choice (just to avoid clashes) that
1947 * is returned if some aspect of an rcu-walk fails. Such an error must
1948 * be handled by restarting a traditional ref-walk (which will always
1949 * be able to complete).
1950 */
Al Virobd92d7f2011-03-14 19:54:59 -04001951 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
Al Viroee0827c2011-02-21 23:38:09 -05001952
Al Virobd92d7f2011-03-14 19:54:59 -04001953 if (unlikely(err))
Al Viro115cbfd2014-10-11 23:05:52 -04001954 goto out;
Al Viroee0827c2011-02-21 23:38:09 -05001955
1956 current->total_link_count = 0;
Al Virobd92d7f2011-03-14 19:54:59 -04001957 err = link_path_walk(name, nd);
1958
1959 if (!err && !(flags & LOOKUP_PARENT)) {
Al Virobd92d7f2011-03-14 19:54:59 -04001960 err = lookup_last(nd, &path);
1961 while (err > 0) {
1962 void *cookie;
1963 struct path link = path;
Kees Cook800179c2012-07-25 17:29:07 -07001964 err = may_follow_link(&link, nd);
1965 if (unlikely(err))
1966 break;
Al Virobd92d7f2011-03-14 19:54:59 -04001967 nd->flags |= LOOKUP_PARENT;
Al Viro574197e2011-03-14 22:20:34 -04001968 err = follow_link(&link, nd, &cookie);
Al Viro6d7b5aa2012-06-10 04:15:17 -04001969 if (err)
1970 break;
1971 err = lookup_last(nd, &path);
Al Viro574197e2011-03-14 22:20:34 -04001972 put_link(nd, &link, cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04001973 }
1974 }
Al Viroee0827c2011-02-21 23:38:09 -05001975
Al Viro9f1fafe2011-03-25 11:00:12 -04001976 if (!err)
1977 err = complete_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04001978
1979 if (!err && nd->flags & LOOKUP_DIRECTORY) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02001980 if (!d_can_lookup(nd->path.dentry)) {
Al Virobd92d7f2011-03-14 19:54:59 -04001981 path_put(&nd->path);
Al Virobd23a532011-03-23 09:56:30 -04001982 err = -ENOTDIR;
Al Virobd92d7f2011-03-14 19:54:59 -04001983 }
1984 }
Al Viro16c2cd72011-02-22 15:50:10 -05001985
Al Viro115cbfd2014-10-11 23:05:52 -04001986out:
Al Viro70e9b352011-03-05 21:12:22 -05001987 if (base)
1988 fput(base);
Al Viroee0827c2011-02-21 23:38:09 -05001989
Al Viro5b6ca022011-03-09 23:04:47 -05001990 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
Al Viro2a737872009-04-07 11:49:53 -04001991 path_put(&nd->root);
1992 nd->root.mnt = NULL;
1993 }
Al Virobd92d7f2011-03-14 19:54:59 -04001994 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001995}
Nick Piggin31e6b012011-01-07 17:49:52 +11001996
Jeff Layton873f1ee2012-10-10 15:25:29 -04001997static int filename_lookup(int dfd, struct filename *name,
1998 unsigned int flags, struct nameidata *nd)
1999{
2000 int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd);
2001 if (unlikely(retval == -ECHILD))
2002 retval = path_lookupat(dfd, name->name, flags, nd);
2003 if (unlikely(retval == -ESTALE))
2004 retval = path_lookupat(dfd, name->name,
2005 flags | LOOKUP_REVAL, nd);
2006
2007 if (likely(!retval))
Jeff Laytonadb5c242012-10-10 16:43:13 -04002008 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002009 return retval;
2010}
2011
Al Viroee0827c2011-02-21 23:38:09 -05002012static int do_path_lookup(int dfd, const char *name,
2013 unsigned int flags, struct nameidata *nd)
2014{
Jeff Layton873f1ee2012-10-10 15:25:29 -04002015 struct filename filename = { .name = name };
Nick Piggin31e6b012011-01-07 17:49:52 +11002016
Jeff Layton873f1ee2012-10-10 15:25:29 -04002017 return filename_lookup(dfd, &filename, flags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018}
2019
Al Viro79714f72012-06-15 03:01:42 +04002020/* does lookup, returns the object with parent locked */
2021struct dentry *kern_path_locked(const char *name, struct path *path)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002022{
Al Viro79714f72012-06-15 03:01:42 +04002023 struct nameidata nd;
2024 struct dentry *d;
2025 int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd);
2026 if (err)
2027 return ERR_PTR(err);
2028 if (nd.last_type != LAST_NORM) {
2029 path_put(&nd.path);
2030 return ERR_PTR(-EINVAL);
2031 }
2032 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Al Viro1e0ea002012-07-22 23:46:21 +04002033 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
Al Viro79714f72012-06-15 03:01:42 +04002034 if (IS_ERR(d)) {
2035 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2036 path_put(&nd.path);
2037 return d;
2038 }
2039 *path = nd.path;
2040 return d;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002041}
2042
Al Virod1811462008-08-02 00:49:18 -04002043int kern_path(const char *name, unsigned int flags, struct path *path)
2044{
2045 struct nameidata nd;
2046 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
2047 if (!res)
2048 *path = nd.path;
2049 return res;
2050}
Al Viro4d359502014-03-14 12:20:17 -04002051EXPORT_SYMBOL(kern_path);
Al Virod1811462008-08-02 00:49:18 -04002052
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002053/**
2054 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2055 * @dentry: pointer to dentry of the base directory
2056 * @mnt: pointer to vfs mount of the base directory
2057 * @name: pointer to file name
2058 * @flags: lookup flags
Al Viroe0a01242011-06-27 17:00:37 -04002059 * @path: pointer to struct path to fill
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002060 */
2061int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2062 const char *name, unsigned int flags,
Al Viroe0a01242011-06-27 17:00:37 -04002063 struct path *path)
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002064{
Al Viroe0a01242011-06-27 17:00:37 -04002065 struct nameidata nd;
2066 int err;
2067 nd.root.dentry = dentry;
2068 nd.root.mnt = mnt;
2069 BUG_ON(flags & LOOKUP_PARENT);
Al Viro5b6ca022011-03-09 23:04:47 -05002070 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
Al Viroe0a01242011-06-27 17:00:37 -04002071 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
2072 if (!err)
2073 *path = nd.path;
2074 return err;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002075}
Al Viro4d359502014-03-14 12:20:17 -04002076EXPORT_SYMBOL(vfs_path_lookup);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002077
James Morris057f6c02007-04-26 00:12:05 -07002078/*
2079 * Restricted form of lookup. Doesn't follow links, single-component only,
2080 * needs parent already locked. Doesn't follow mounts.
2081 * SMP-safe.
2082 */
Adrian Bunka244e162006-03-31 02:32:11 -08002083static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084{
Al Viro72bd8662012-06-10 17:17:17 -04002085 return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086}
2087
Christoph Hellwigeead1912007-10-16 23:25:38 -07002088/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07002089 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07002090 * @name: pathname component to lookup
2091 * @base: base directory to lookup from
2092 * @len: maximum length @len should be interpreted to
2093 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07002094 * Note that this routine is purely a helper for filesystem usage and should
2095 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07002096 * nameidata argument is passed to the filesystem methods and a filesystem
2097 * using this helper needs to be prepared for that.
2098 */
James Morris057f6c02007-04-26 00:12:05 -07002099struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2100{
James Morris057f6c02007-04-26 00:12:05 -07002101 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05002102 unsigned int c;
Miklos Szeredicda309d2012-03-26 12:54:21 +02002103 int err;
James Morris057f6c02007-04-26 00:12:05 -07002104
David Woodhouse2f9092e2009-04-20 23:18:37 +01002105 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2106
Al Viro6a96ba52011-03-07 23:49:20 -05002107 this.name = name;
2108 this.len = len;
Linus Torvalds0145acc2012-03-02 14:32:59 -08002109 this.hash = full_name_hash(name, len);
Al Viro6a96ba52011-03-07 23:49:20 -05002110 if (!len)
2111 return ERR_PTR(-EACCES);
2112
Al Viro21d8a152012-11-29 22:17:21 -05002113 if (unlikely(name[0] == '.')) {
2114 if (len < 2 || (len == 2 && name[1] == '.'))
2115 return ERR_PTR(-EACCES);
2116 }
2117
Al Viro6a96ba52011-03-07 23:49:20 -05002118 while (len--) {
2119 c = *(const unsigned char *)name++;
2120 if (c == '/' || c == '\0')
2121 return ERR_PTR(-EACCES);
Al Viro6a96ba52011-03-07 23:49:20 -05002122 }
Al Viro5a202bc2011-03-08 14:17:44 -05002123 /*
2124 * See if the low-level filesystem might want
2125 * to use its own hash..
2126 */
2127 if (base->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002128 int err = base->d_op->d_hash(base, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05002129 if (err < 0)
2130 return ERR_PTR(err);
2131 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07002132
Miklos Szeredicda309d2012-03-26 12:54:21 +02002133 err = inode_permission(base->d_inode, MAY_EXEC);
2134 if (err)
2135 return ERR_PTR(err);
2136
Al Viro72bd8662012-06-10 17:17:17 -04002137 return __lookup_hash(&this, base, 0);
James Morris057f6c02007-04-26 00:12:05 -07002138}
Al Viro4d359502014-03-14 12:20:17 -04002139EXPORT_SYMBOL(lookup_one_len);
James Morris057f6c02007-04-26 00:12:05 -07002140
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002141int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2142 struct path *path, int *empty)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143{
Al Viro2d8f3032008-07-22 09:59:21 -04002144 struct nameidata nd;
Jeff Layton91a27b22012-10-10 15:25:28 -04002145 struct filename *tmp = getname_flags(name, flags, empty);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04002148
2149 BUG_ON(flags & LOOKUP_PARENT);
2150
Jeff Layton873f1ee2012-10-10 15:25:29 -04002151 err = filename_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04002153 if (!err)
2154 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155 }
2156 return err;
2157}
2158
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002159int user_path_at(int dfd, const char __user *name, unsigned flags,
2160 struct path *path)
2161{
Linus Torvaldsf7493e52012-03-22 16:10:40 -07002162 return user_path_at_empty(dfd, name, flags, path, NULL);
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002163}
Al Viro4d359502014-03-14 12:20:17 -04002164EXPORT_SYMBOL(user_path_at);
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002165
Jeff Layton873f1ee2012-10-10 15:25:29 -04002166/*
2167 * NB: most callers don't do anything directly with the reference to the
2168 * to struct filename, but the nd->last pointer points into the name string
2169 * allocated by getname. So we must hold the reference to it until all
2170 * path-walking is complete.
2171 */
Jeff Layton91a27b22012-10-10 15:25:28 -04002172static struct filename *
Jeff Layton9e790bd2012-12-11 12:10:09 -05002173user_path_parent(int dfd, const char __user *path, struct nameidata *nd,
2174 unsigned int flags)
Al Viro2ad94ae2008-07-21 09:32:51 -04002175{
Jeff Layton91a27b22012-10-10 15:25:28 -04002176 struct filename *s = getname(path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002177 int error;
2178
Jeff Layton9e790bd2012-12-11 12:10:09 -05002179 /* only LOOKUP_REVAL is allowed in extra flags */
2180 flags &= LOOKUP_REVAL;
2181
Al Viro2ad94ae2008-07-21 09:32:51 -04002182 if (IS_ERR(s))
Jeff Layton91a27b22012-10-10 15:25:28 -04002183 return s;
Al Viro2ad94ae2008-07-21 09:32:51 -04002184
Jeff Layton9e790bd2012-12-11 12:10:09 -05002185 error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd);
Jeff Layton91a27b22012-10-10 15:25:28 -04002186 if (error) {
Al Viro2ad94ae2008-07-21 09:32:51 -04002187 putname(s);
Jeff Layton91a27b22012-10-10 15:25:28 -04002188 return ERR_PTR(error);
2189 }
Al Viro2ad94ae2008-07-21 09:32:51 -04002190
Jeff Layton91a27b22012-10-10 15:25:28 -04002191 return s;
Al Viro2ad94ae2008-07-21 09:32:51 -04002192}
2193
Jeff Layton80334262013-07-26 06:23:25 -04002194/**
Al Viro197df042013-09-08 14:03:27 -04002195 * mountpoint_last - look up last component for umount
Jeff Layton80334262013-07-26 06:23:25 -04002196 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2197 * @path: pointer to container for result
2198 *
2199 * This is a special lookup_last function just for umount. In this case, we
2200 * need to resolve the path without doing any revalidation.
2201 *
2202 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2203 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2204 * in almost all cases, this lookup will be served out of the dcache. The only
2205 * cases where it won't are if nd->last refers to a symlink or the path is
2206 * bogus and it doesn't exist.
2207 *
2208 * Returns:
2209 * -error: if there was an error during lookup. This includes -ENOENT if the
2210 * lookup found a negative dentry. The nd->path reference will also be
2211 * put in this case.
2212 *
2213 * 0: if we successfully resolved nd->path and found it to not to be a
2214 * symlink that needs to be followed. "path" will also be populated.
2215 * The nd->path reference will also be put.
2216 *
2217 * 1: if we successfully resolved nd->last and found it to be a symlink
2218 * that needs to be followed. "path" will be populated with the path
2219 * to the link, and nd->path will *not* be put.
2220 */
2221static int
Al Viro197df042013-09-08 14:03:27 -04002222mountpoint_last(struct nameidata *nd, struct path *path)
Jeff Layton80334262013-07-26 06:23:25 -04002223{
2224 int error = 0;
2225 struct dentry *dentry;
2226 struct dentry *dir = nd->path.dentry;
2227
Al Viro35759522013-09-08 13:41:33 -04002228 /* If we're in rcuwalk, drop out of it to handle last component */
2229 if (nd->flags & LOOKUP_RCU) {
2230 if (unlazy_walk(nd, NULL)) {
2231 error = -ECHILD;
2232 goto out;
2233 }
Jeff Layton80334262013-07-26 06:23:25 -04002234 }
2235
2236 nd->flags &= ~LOOKUP_PARENT;
2237
2238 if (unlikely(nd->last_type != LAST_NORM)) {
2239 error = handle_dots(nd, nd->last_type);
Al Viro35759522013-09-08 13:41:33 -04002240 if (error)
2241 goto out;
2242 dentry = dget(nd->path.dentry);
2243 goto done;
Jeff Layton80334262013-07-26 06:23:25 -04002244 }
2245
2246 mutex_lock(&dir->d_inode->i_mutex);
2247 dentry = d_lookup(dir, &nd->last);
2248 if (!dentry) {
2249 /*
2250 * No cached dentry. Mounted dentries are pinned in the cache,
2251 * so that means that this dentry is probably a symlink or the
2252 * path doesn't actually point to a mounted dentry.
2253 */
2254 dentry = d_alloc(dir, &nd->last);
2255 if (!dentry) {
2256 error = -ENOMEM;
Dave Jonesbcceeeb2013-09-10 17:04:25 -04002257 mutex_unlock(&dir->d_inode->i_mutex);
Al Viro35759522013-09-08 13:41:33 -04002258 goto out;
Jeff Layton80334262013-07-26 06:23:25 -04002259 }
Al Viro35759522013-09-08 13:41:33 -04002260 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2261 error = PTR_ERR(dentry);
Dave Jonesbcceeeb2013-09-10 17:04:25 -04002262 if (IS_ERR(dentry)) {
2263 mutex_unlock(&dir->d_inode->i_mutex);
Al Viro35759522013-09-08 13:41:33 -04002264 goto out;
Dave Jonesbcceeeb2013-09-10 17:04:25 -04002265 }
Jeff Layton80334262013-07-26 06:23:25 -04002266 }
2267 mutex_unlock(&dir->d_inode->i_mutex);
2268
Al Viro35759522013-09-08 13:41:33 -04002269done:
Al Viro22213312014-04-19 12:30:58 -04002270 if (!dentry->d_inode || d_is_negative(dentry)) {
Al Viro35759522013-09-08 13:41:33 -04002271 error = -ENOENT;
2272 dput(dentry);
2273 goto out;
Jeff Layton80334262013-07-26 06:23:25 -04002274 }
Al Viro35759522013-09-08 13:41:33 -04002275 path->dentry = dentry;
Vasily Averin295dc392014-07-21 12:30:23 +04002276 path->mnt = nd->path.mnt;
David Howellsb18825a2013-09-12 19:22:53 +01002277 if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW))
Al Viro35759522013-09-08 13:41:33 -04002278 return 1;
Vasily Averin295dc392014-07-21 12:30:23 +04002279 mntget(path->mnt);
Al Viro35759522013-09-08 13:41:33 -04002280 follow_mount(path);
2281 error = 0;
2282out:
Jeff Layton80334262013-07-26 06:23:25 -04002283 terminate_walk(nd);
2284 return error;
2285}
2286
2287/**
Al Viro197df042013-09-08 14:03:27 -04002288 * path_mountpoint - look up a path to be umounted
Jeff Layton80334262013-07-26 06:23:25 -04002289 * @dfd: directory file descriptor to start walk from
2290 * @name: full pathname to walk
Randy Dunlap606d6fe2013-10-19 14:56:55 -07002291 * @path: pointer to container for result
Jeff Layton80334262013-07-26 06:23:25 -04002292 * @flags: lookup flags
Jeff Layton80334262013-07-26 06:23:25 -04002293 *
2294 * Look up the given name, but don't attempt to revalidate the last component.
Randy Dunlap606d6fe2013-10-19 14:56:55 -07002295 * Returns 0 and "path" will be valid on success; Returns error otherwise.
Jeff Layton80334262013-07-26 06:23:25 -04002296 */
2297static int
Al Viro197df042013-09-08 14:03:27 -04002298path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags)
Jeff Layton80334262013-07-26 06:23:25 -04002299{
2300 struct file *base = NULL;
2301 struct nameidata nd;
2302 int err;
2303
2304 err = path_init(dfd, name, flags | LOOKUP_PARENT, &nd, &base);
2305 if (unlikely(err))
Al Viro115cbfd2014-10-11 23:05:52 -04002306 goto out;
Jeff Layton80334262013-07-26 06:23:25 -04002307
2308 current->total_link_count = 0;
2309 err = link_path_walk(name, &nd);
2310 if (err)
2311 goto out;
2312
Al Viro197df042013-09-08 14:03:27 -04002313 err = mountpoint_last(&nd, path);
Jeff Layton80334262013-07-26 06:23:25 -04002314 while (err > 0) {
2315 void *cookie;
2316 struct path link = *path;
2317 err = may_follow_link(&link, &nd);
2318 if (unlikely(err))
2319 break;
2320 nd.flags |= LOOKUP_PARENT;
2321 err = follow_link(&link, &nd, &cookie);
2322 if (err)
2323 break;
Al Viro197df042013-09-08 14:03:27 -04002324 err = mountpoint_last(&nd, path);
Jeff Layton80334262013-07-26 06:23:25 -04002325 put_link(&nd, &link, cookie);
2326 }
2327out:
2328 if (base)
2329 fput(base);
2330
2331 if (nd.root.mnt && !(nd.flags & LOOKUP_ROOT))
2332 path_put(&nd.root);
2333
2334 return err;
2335}
2336
Al Viro2d864652013-09-08 20:18:44 -04002337static int
2338filename_mountpoint(int dfd, struct filename *s, struct path *path,
2339 unsigned int flags)
2340{
2341 int error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU);
2342 if (unlikely(error == -ECHILD))
2343 error = path_mountpoint(dfd, s->name, path, flags);
2344 if (unlikely(error == -ESTALE))
2345 error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL);
2346 if (likely(!error))
2347 audit_inode(s, path->dentry, 0);
2348 return error;
2349}
2350
Jeff Layton80334262013-07-26 06:23:25 -04002351/**
Al Viro197df042013-09-08 14:03:27 -04002352 * user_path_mountpoint_at - lookup a path from userland in order to umount it
Jeff Layton80334262013-07-26 06:23:25 -04002353 * @dfd: directory file descriptor
2354 * @name: pathname from userland
2355 * @flags: lookup flags
2356 * @path: pointer to container to hold result
2357 *
2358 * A umount is a special case for path walking. We're not actually interested
2359 * in the inode in this situation, and ESTALE errors can be a problem. We
2360 * simply want track down the dentry and vfsmount attached at the mountpoint
2361 * and avoid revalidating the last component.
2362 *
2363 * Returns 0 and populates "path" on success.
2364 */
2365int
Al Viro197df042013-09-08 14:03:27 -04002366user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
Jeff Layton80334262013-07-26 06:23:25 -04002367 struct path *path)
2368{
2369 struct filename *s = getname(name);
2370 int error;
Jeff Layton80334262013-07-26 06:23:25 -04002371 if (IS_ERR(s))
2372 return PTR_ERR(s);
Al Viro2d864652013-09-08 20:18:44 -04002373 error = filename_mountpoint(dfd, s, path, flags);
Jeff Layton80334262013-07-26 06:23:25 -04002374 putname(s);
2375 return error;
2376}
2377
Al Viro2d864652013-09-08 20:18:44 -04002378int
2379kern_path_mountpoint(int dfd, const char *name, struct path *path,
2380 unsigned int flags)
2381{
2382 struct filename s = {.name = name};
2383 return filename_mountpoint(dfd, &s, path, flags);
2384}
2385EXPORT_SYMBOL(kern_path_mountpoint);
2386
Miklos Szeredicbdf35b2014-10-24 00:14:36 +02002387int __check_sticky(struct inode *dir, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388{
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002389 kuid_t fsuid = current_fsuid();
David Howellsda9592e2008-11-14 10:39:05 +11002390
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002391 if (uid_eq(inode->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392 return 0;
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002393 if (uid_eq(dir->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394 return 0;
Andy Lutomirski23adbe12014-06-10 12:45:42 -07002395 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396}
Miklos Szeredicbdf35b2014-10-24 00:14:36 +02002397EXPORT_SYMBOL(__check_sticky);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398
2399/*
2400 * Check whether we can remove a link victim from directory dir, check
2401 * whether the type of victim is right.
2402 * 1. We can't do it if dir is read-only (done in permission())
2403 * 2. We should have write and exec permissions on dir
2404 * 3. We can't remove anything from append-only dir
2405 * 4. We can't do anything with immutable dir (done in permission())
2406 * 5. If the sticky bit on dir is set we should either
2407 * a. be owner of dir, or
2408 * b. be owner of victim, or
2409 * c. have CAP_FOWNER capability
2410 * 6. If the victim is append-only or immutable we can't do antyhing with
2411 * links pointing to it.
2412 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2413 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2414 * 9. We can't remove a root or mountpoint.
2415 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2416 * nfs_async_unlink().
2417 */
David Howellsb18825a2013-09-12 19:22:53 +01002418static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419{
David Howellsb18825a2013-09-12 19:22:53 +01002420 struct inode *inode = victim->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421 int error;
2422
David Howellsb18825a2013-09-12 19:22:53 +01002423 if (d_is_negative(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 return -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01002425 BUG_ON(!inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426
2427 BUG_ON(victim->d_parent->d_inode != dir);
Jeff Layton4fa6b5e2012-10-10 15:25:25 -04002428 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429
Al Virof419a2e2008-07-22 00:07:17 -04002430 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431 if (error)
2432 return error;
2433 if (IS_APPEND(dir))
2434 return -EPERM;
David Howellsb18825a2013-09-12 19:22:53 +01002435
2436 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2437 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438 return -EPERM;
2439 if (isdir) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002440 if (!d_is_dir(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441 return -ENOTDIR;
2442 if (IS_ROOT(victim))
2443 return -EBUSY;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002444 } else if (d_is_dir(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445 return -EISDIR;
2446 if (IS_DEADDIR(dir))
2447 return -ENOENT;
2448 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2449 return -EBUSY;
2450 return 0;
2451}
2452
2453/* Check whether we can create an object with dentry child in directory
2454 * dir.
2455 * 1. We can't do it if child already exists (open has special treatment for
2456 * this case, but since we are inlined it's OK)
2457 * 2. We can't do it if dir is read-only (done in permission())
2458 * 3. We should have write and exec permissions on dir
2459 * 4. We can't do it if dir is immutable (done in permission())
2460 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02002461static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462{
Jeff Layton14e972b2013-05-08 10:25:58 -04002463 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464 if (child->d_inode)
2465 return -EEXIST;
2466 if (IS_DEADDIR(dir))
2467 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04002468 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469}
2470
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471/*
2472 * p1 and p2 should be directories on the same fs.
2473 */
2474struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2475{
2476 struct dentry *p;
2477
2478 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07002479 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002480 return NULL;
2481 }
2482
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002483 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002485 p = d_ancestor(p2, p1);
2486 if (p) {
2487 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2488 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2489 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490 }
2491
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002492 p = d_ancestor(p1, p2);
2493 if (p) {
2494 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2495 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2496 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497 }
2498
Ingo Molnarf2eace22006-07-03 00:25:05 -07002499 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Miklos Szeredid1b72cc2014-10-27 15:42:01 +01002500 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002501 return NULL;
2502}
Al Viro4d359502014-03-14 12:20:17 -04002503EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504
2505void unlock_rename(struct dentry *p1, struct dentry *p2)
2506{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002507 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002509 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002510 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511 }
2512}
Al Viro4d359502014-03-14 12:20:17 -04002513EXPORT_SYMBOL(unlock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514
Al Viro4acdaf22011-07-26 01:42:34 -04002515int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viro312b63f2012-06-10 18:09:36 -04002516 bool want_excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002518 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519 if (error)
2520 return error;
2521
Al Viroacfa4382008-12-04 10:06:33 -05002522 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523 return -EACCES; /* shouldn't it be ENOSYS? */
2524 mode &= S_IALLUGO;
2525 mode |= S_IFREG;
2526 error = security_inode_create(dir, dentry, mode);
2527 if (error)
2528 return error;
Al Viro312b63f2012-06-10 18:09:36 -04002529 error = dir->i_op->create(dir, dentry, mode, want_excl);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002530 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002531 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532 return error;
2533}
Al Viro4d359502014-03-14 12:20:17 -04002534EXPORT_SYMBOL(vfs_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535
Al Viro73d049a2011-03-11 12:08:24 -05002536static int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002538 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539 struct inode *inode = dentry->d_inode;
2540 int error;
2541
Al Virobcda7652011-03-13 16:42:14 -04002542 /* O_PATH? */
2543 if (!acc_mode)
2544 return 0;
2545
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546 if (!inode)
2547 return -ENOENT;
2548
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002549 switch (inode->i_mode & S_IFMT) {
2550 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002552 case S_IFDIR:
2553 if (acc_mode & MAY_WRITE)
2554 return -EISDIR;
2555 break;
2556 case S_IFBLK:
2557 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002558 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002560 /*FALLTHRU*/
2561 case S_IFIFO:
2562 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002564 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002565 }
Dave Hansenb41572e2007-10-16 23:31:14 -07002566
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002567 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07002568 if (error)
2569 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05002570
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571 /*
2572 * An append-only file must be opened in append mode for writing.
2573 */
2574 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05002575 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05002576 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05002578 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 }
2580
2581 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002582 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05002583 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584
J. Bruce Fieldsf3c7691e2011-09-21 10:58:13 -04002585 return 0;
Al Viro7715b522009-12-16 03:54:00 -05002586}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587
Jeff Laytone1181ee2010-12-07 16:19:50 -05002588static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002589{
Jeff Laytone1181ee2010-12-07 16:19:50 -05002590 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002591 struct inode *inode = path->dentry->d_inode;
2592 int error = get_write_access(inode);
2593 if (error)
2594 return error;
2595 /*
2596 * Refuse to truncate files with mandatory locks held on them.
2597 */
Jeff Laytond7a06982014-03-10 09:54:15 -04002598 error = locks_verify_locked(filp);
Al Viro7715b522009-12-16 03:54:00 -05002599 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002600 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002601 if (!error) {
2602 error = do_truncate(path->dentry, 0,
2603 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002604 filp);
Al Viro7715b522009-12-16 03:54:00 -05002605 }
2606 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002607 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608}
2609
Dave Hansend57999e2008-02-15 14:37:27 -08002610static inline int open_to_namei_flags(int flag)
2611{
Al Viro8a5e9292011-06-25 19:15:54 -04002612 if ((flag & O_ACCMODE) == 3)
2613 flag--;
Dave Hansend57999e2008-02-15 14:37:27 -08002614 return flag;
2615}
2616
Miklos Szeredid18e9002012-06-05 15:10:17 +02002617static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2618{
2619 int error = security_path_mknod(dir, dentry, mode, 0);
2620 if (error)
2621 return error;
2622
2623 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2624 if (error)
2625 return error;
2626
2627 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2628}
2629
David Howells1acf0af2012-06-14 16:13:46 +01002630/*
2631 * Attempt to atomically look up, create and open a file from a negative
2632 * dentry.
2633 *
2634 * Returns 0 if successful. The file will have been created and attached to
2635 * @file by the filesystem calling finish_open().
2636 *
2637 * Returns 1 if the file was looked up only or didn't need creating. The
2638 * caller will need to perform the open themselves. @path will have been
2639 * updated to point to the new dentry. This may be negative.
2640 *
2641 * Returns an error code otherwise.
2642 */
Al Viro2675a4e2012-06-22 12:41:10 +04002643static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2644 struct path *path, struct file *file,
2645 const struct open_flags *op,
Al Viro64894cf2012-07-31 00:53:35 +04002646 bool got_write, bool need_lookup,
Al Viro2675a4e2012-06-22 12:41:10 +04002647 int *opened)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002648{
2649 struct inode *dir = nd->path.dentry->d_inode;
2650 unsigned open_flag = open_to_namei_flags(op->open_flag);
2651 umode_t mode;
2652 int error;
2653 int acc_mode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002654 int create_error = 0;
2655 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
Miklos Szeredi116cc022013-09-16 14:52:05 +02002656 bool excl;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002657
2658 BUG_ON(dentry->d_inode);
2659
2660 /* Don't create child dentry for a dead directory. */
2661 if (unlikely(IS_DEADDIR(dir))) {
Al Viro2675a4e2012-06-22 12:41:10 +04002662 error = -ENOENT;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002663 goto out;
2664 }
2665
Miklos Szeredi62b259d2012-08-15 13:01:24 +02002666 mode = op->mode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002667 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2668 mode &= ~current_umask();
2669
Miklos Szeredi116cc022013-09-16 14:52:05 +02002670 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2671 if (excl)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002672 open_flag &= ~O_TRUNC;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002673
2674 /*
2675 * Checking write permission is tricky, bacuse we don't know if we are
2676 * going to actually need it: O_CREAT opens should work as long as the
2677 * file exists. But checking existence breaks atomicity. The trick is
2678 * to check access and if not granted clear O_CREAT from the flags.
2679 *
2680 * Another problem is returing the "right" error value (e.g. for an
2681 * O_EXCL open we want to return EEXIST not EROFS).
2682 */
Al Viro64894cf2012-07-31 00:53:35 +04002683 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2684 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2685 if (!(open_flag & O_CREAT)) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002686 /*
2687 * No O_CREATE -> atomicity not a requirement -> fall
2688 * back to lookup + open
2689 */
2690 goto no_open;
2691 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2692 /* Fall back and fail with the right error */
Al Viro64894cf2012-07-31 00:53:35 +04002693 create_error = -EROFS;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002694 goto no_open;
2695 } else {
2696 /* No side effects, safe to clear O_CREAT */
Al Viro64894cf2012-07-31 00:53:35 +04002697 create_error = -EROFS;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002698 open_flag &= ~O_CREAT;
2699 }
2700 }
2701
2702 if (open_flag & O_CREAT) {
Miklos Szeredi38227f72012-08-15 13:01:24 +02002703 error = may_o_create(&nd->path, dentry, mode);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002704 if (error) {
2705 create_error = error;
2706 if (open_flag & O_EXCL)
2707 goto no_open;
2708 open_flag &= ~O_CREAT;
2709 }
2710 }
2711
2712 if (nd->flags & LOOKUP_DIRECTORY)
2713 open_flag |= O_DIRECTORY;
2714
Al Viro30d90492012-06-22 12:40:19 +04002715 file->f_path.dentry = DENTRY_NOT_SET;
2716 file->f_path.mnt = nd->path.mnt;
2717 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
Al Viro47237682012-06-10 05:01:45 -04002718 opened);
Al Virod9585272012-06-22 12:39:14 +04002719 if (error < 0) {
Al Virod9585272012-06-22 12:39:14 +04002720 if (create_error && error == -ENOENT)
2721 error = create_error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002722 goto out;
2723 }
2724
Al Virod9585272012-06-22 12:39:14 +04002725 if (error) { /* returned 1, that is */
Al Viro30d90492012-06-22 12:40:19 +04002726 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
Al Viro2675a4e2012-06-22 12:41:10 +04002727 error = -EIO;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002728 goto out;
2729 }
Al Viro30d90492012-06-22 12:40:19 +04002730 if (file->f_path.dentry) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002731 dput(dentry);
Al Viro30d90492012-06-22 12:40:19 +04002732 dentry = file->f_path.dentry;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002733 }
Al Viro03da6332013-09-16 19:22:33 -04002734 if (*opened & FILE_CREATED)
2735 fsnotify_create(dir, dentry);
2736 if (!dentry->d_inode) {
2737 WARN_ON(*opened & FILE_CREATED);
2738 if (create_error) {
2739 error = create_error;
2740 goto out;
2741 }
2742 } else {
2743 if (excl && !(*opened & FILE_CREATED)) {
2744 error = -EEXIST;
2745 goto out;
2746 }
Sage Weil62b2ce92012-08-15 13:30:12 -07002747 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02002748 goto looked_up;
2749 }
2750
2751 /*
2752 * We didn't have the inode before the open, so check open permission
2753 * here.
2754 */
Al Viro03da6332013-09-16 19:22:33 -04002755 acc_mode = op->acc_mode;
2756 if (*opened & FILE_CREATED) {
2757 WARN_ON(!(open_flag & O_CREAT));
2758 fsnotify_create(dir, dentry);
2759 acc_mode = MAY_OPEN;
2760 }
Al Viro2675a4e2012-06-22 12:41:10 +04002761 error = may_open(&file->f_path, acc_mode, open_flag);
2762 if (error)
2763 fput(file);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002764
2765out:
2766 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04002767 return error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002768
Miklos Szeredid18e9002012-06-05 15:10:17 +02002769no_open:
2770 if (need_lookup) {
Al Viro72bd8662012-06-10 17:17:17 -04002771 dentry = lookup_real(dir, dentry, nd->flags);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002772 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002773 return PTR_ERR(dentry);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002774
2775 if (create_error) {
2776 int open_flag = op->open_flag;
2777
Al Viro2675a4e2012-06-22 12:41:10 +04002778 error = create_error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002779 if ((open_flag & O_EXCL)) {
2780 if (!dentry->d_inode)
2781 goto out;
2782 } else if (!dentry->d_inode) {
2783 goto out;
2784 } else if ((open_flag & O_TRUNC) &&
2785 S_ISREG(dentry->d_inode->i_mode)) {
2786 goto out;
2787 }
2788 /* will fail later, go on to get the right error */
2789 }
2790 }
2791looked_up:
2792 path->dentry = dentry;
2793 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04002794 return 1;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002795}
2796
Nick Piggin31e6b012011-01-07 17:49:52 +11002797/*
David Howells1acf0af2012-06-14 16:13:46 +01002798 * Look up and maybe create and open the last component.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002799 *
2800 * Must be called with i_mutex held on parent.
2801 *
David Howells1acf0af2012-06-14 16:13:46 +01002802 * Returns 0 if the file was successfully atomically created (if necessary) and
2803 * opened. In this case the file will be returned attached to @file.
2804 *
2805 * Returns 1 if the file was not completely opened at this time, though lookups
2806 * and creations will have been performed and the dentry returned in @path will
2807 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2808 * specified then a negative dentry may be returned.
2809 *
2810 * An error code is returned otherwise.
2811 *
2812 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2813 * cleared otherwise prior to returning.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002814 */
Al Viro2675a4e2012-06-22 12:41:10 +04002815static int lookup_open(struct nameidata *nd, struct path *path,
2816 struct file *file,
2817 const struct open_flags *op,
Al Viro64894cf2012-07-31 00:53:35 +04002818 bool got_write, int *opened)
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002819{
2820 struct dentry *dir = nd->path.dentry;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002821 struct inode *dir_inode = dir->d_inode;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002822 struct dentry *dentry;
2823 int error;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002824 bool need_lookup;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002825
Al Viro47237682012-06-10 05:01:45 -04002826 *opened &= ~FILE_CREATED;
Al Viro201f9562012-06-22 12:42:10 +04002827 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002828 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002829 return PTR_ERR(dentry);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002830
Miklos Szeredid18e9002012-06-05 15:10:17 +02002831 /* Cached positive dentry: will open in f_op->open */
2832 if (!need_lookup && dentry->d_inode)
2833 goto out_no_open;
2834
2835 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
Al Viro64894cf2012-07-31 00:53:35 +04002836 return atomic_open(nd, dentry, path, file, op, got_write,
Al Viro47237682012-06-10 05:01:45 -04002837 need_lookup, opened);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002838 }
2839
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002840 if (need_lookup) {
2841 BUG_ON(dentry->d_inode);
2842
Al Viro72bd8662012-06-10 17:17:17 -04002843 dentry = lookup_real(dir_inode, dentry, nd->flags);
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002844 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002845 return PTR_ERR(dentry);
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002846 }
2847
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002848 /* Negative dentry, just create the file */
2849 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2850 umode_t mode = op->mode;
2851 if (!IS_POSIXACL(dir->d_inode))
2852 mode &= ~current_umask();
2853 /*
2854 * This write is needed to ensure that a
2855 * rw->ro transition does not occur between
2856 * the time when the file is created and when
2857 * a permanent write count is taken through
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002858 * the 'struct file' in finish_open().
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002859 */
Al Viro64894cf2012-07-31 00:53:35 +04002860 if (!got_write) {
2861 error = -EROFS;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002862 goto out_dput;
Al Viro64894cf2012-07-31 00:53:35 +04002863 }
Al Viro47237682012-06-10 05:01:45 -04002864 *opened |= FILE_CREATED;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002865 error = security_path_mknod(&nd->path, dentry, mode, 0);
2866 if (error)
2867 goto out_dput;
Al Viro312b63f2012-06-10 18:09:36 -04002868 error = vfs_create(dir->d_inode, dentry, mode,
2869 nd->flags & LOOKUP_EXCL);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002870 if (error)
2871 goto out_dput;
2872 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02002873out_no_open:
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002874 path->dentry = dentry;
2875 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04002876 return 1;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002877
2878out_dput:
2879 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04002880 return error;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002881}
2882
2883/*
Al Virofe2d35f2011-03-05 22:58:25 -05002884 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11002885 */
Al Viro2675a4e2012-06-22 12:41:10 +04002886static int do_last(struct nameidata *nd, struct path *path,
2887 struct file *file, const struct open_flags *op,
Jeff Layton669abf42012-10-10 16:43:10 -04002888 int *opened, struct filename *name)
Al Virofb1cc552009-12-24 01:58:28 -05002889{
Al Viroa1e28032009-12-24 02:12:06 -05002890 struct dentry *dir = nd->path.dentry;
Al Viroca344a892011-03-09 00:36:45 -05002891 int open_flag = op->open_flag;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002892 bool will_truncate = (open_flag & O_TRUNC) != 0;
Al Viro64894cf2012-07-31 00:53:35 +04002893 bool got_write = false;
Al Virobcda7652011-03-13 16:42:14 -04002894 int acc_mode = op->acc_mode;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02002895 struct inode *inode;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002896 bool symlink_ok = false;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002897 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2898 bool retried = false;
Al Viro16c2cd72011-02-22 15:50:10 -05002899 int error;
Al Virofb1cc552009-12-24 01:58:28 -05002900
Al Viroc3e380b2011-02-23 13:39:45 -05002901 nd->flags &= ~LOOKUP_PARENT;
2902 nd->flags |= op->intent;
2903
Al Virobc77daa2013-06-06 09:12:33 -04002904 if (nd->last_type != LAST_NORM) {
Al Virofe2d35f2011-03-05 22:58:25 -05002905 error = handle_dots(nd, nd->last_type);
2906 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002907 return error;
Miklos Szeredie83db162012-06-05 15:10:29 +02002908 goto finish_open;
Al Viro1f36f772009-12-26 10:56:19 -05002909 }
Al Viro67ee3ad2009-12-26 07:01:01 -05002910
Al Viroca344a892011-03-09 00:36:45 -05002911 if (!(open_flag & O_CREAT)) {
Al Virofe2d35f2011-03-05 22:58:25 -05002912 if (nd->last.name[nd->last.len])
2913 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
Al Virobcda7652011-03-13 16:42:14 -04002914 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002915 symlink_ok = true;
Al Virofe2d35f2011-03-05 22:58:25 -05002916 /* we _can_ be in RCU mode here */
Al Viroe97cdc82013-01-24 18:16:00 -05002917 error = lookup_fast(nd, path, &inode);
Miklos Szeredi71574862012-06-05 15:10:14 +02002918 if (likely(!error))
2919 goto finish_lookup;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02002920
Miklos Szeredi71574862012-06-05 15:10:14 +02002921 if (error < 0)
Al Viro2675a4e2012-06-22 12:41:10 +04002922 goto out;
Miklos Szeredi37d7fff2012-06-05 15:10:12 +02002923
Miklos Szeredi71574862012-06-05 15:10:14 +02002924 BUG_ON(nd->inode != dir->d_inode);
Miklos Szeredib6183df2012-06-05 15:10:13 +02002925 } else {
2926 /* create side of things */
2927 /*
2928 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2929 * has been cleared when we got to the last component we are
2930 * about to look up
2931 */
2932 error = complete_walk(nd);
2933 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002934 return error;
Miklos Szeredib6183df2012-06-05 15:10:13 +02002935
Jeff Layton33e22082013-04-12 15:16:32 -04002936 audit_inode(name, dir, LOOKUP_PARENT);
Miklos Szeredib6183df2012-06-05 15:10:13 +02002937 error = -EISDIR;
2938 /* trailing slashes? */
2939 if (nd->last.name[nd->last.len])
Al Viro2675a4e2012-06-22 12:41:10 +04002940 goto out;
Al Virofe2d35f2011-03-05 22:58:25 -05002941 }
2942
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002943retry_lookup:
Al Viro64894cf2012-07-31 00:53:35 +04002944 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2945 error = mnt_want_write(nd->path.mnt);
2946 if (!error)
2947 got_write = true;
2948 /*
2949 * do _not_ fail yet - we might not need that or fail with
2950 * a different error; let lookup_open() decide; we'll be
2951 * dropping this one anyway.
2952 */
2953 }
Al Viroa1e28032009-12-24 02:12:06 -05002954 mutex_lock(&dir->d_inode->i_mutex);
Al Viro64894cf2012-07-31 00:53:35 +04002955 error = lookup_open(nd, path, file, op, got_write, opened);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002956 mutex_unlock(&dir->d_inode->i_mutex);
Al Viroa1e28032009-12-24 02:12:06 -05002957
Al Viro2675a4e2012-06-22 12:41:10 +04002958 if (error <= 0) {
2959 if (error)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002960 goto out;
2961
Al Viro47237682012-06-10 05:01:45 -04002962 if ((*opened & FILE_CREATED) ||
Al Viro496ad9a2013-01-23 17:07:38 -05002963 !S_ISREG(file_inode(file)->i_mode))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002964 will_truncate = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002965
Jeff Laytonadb5c242012-10-10 16:43:13 -04002966 audit_inode(name, file->f_path.dentry, 0);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002967 goto opened;
2968 }
Al Virofb1cc552009-12-24 01:58:28 -05002969
Al Viro47237682012-06-10 05:01:45 -04002970 if (*opened & FILE_CREATED) {
Al Viro9b44f1b2011-03-09 00:17:27 -05002971 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05002972 open_flag &= ~O_TRUNC;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002973 will_truncate = false;
Al Virobcda7652011-03-13 16:42:14 -04002974 acc_mode = MAY_OPEN;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002975 path_to_nameidata(path, nd);
Miklos Szeredie83db162012-06-05 15:10:29 +02002976 goto finish_open_created;
Al Virofb1cc552009-12-24 01:58:28 -05002977 }
2978
2979 /*
Jeff Layton3134f372012-07-25 10:19:47 -04002980 * create/update audit record if it already exists.
Al Virofb1cc552009-12-24 01:58:28 -05002981 */
David Howellsb18825a2013-09-12 19:22:53 +01002982 if (d_is_positive(path->dentry))
Jeff Laytonadb5c242012-10-10 16:43:13 -04002983 audit_inode(name, path->dentry, 0);
Al Virofb1cc552009-12-24 01:58:28 -05002984
Miklos Szeredid18e9002012-06-05 15:10:17 +02002985 /*
2986 * If atomic_open() acquired write access it is dropped now due to
2987 * possible mount and symlink following (this might be optimized away if
2988 * necessary...)
2989 */
Al Viro64894cf2012-07-31 00:53:35 +04002990 if (got_write) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002991 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04002992 got_write = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002993 }
2994
Al Virofb1cc552009-12-24 01:58:28 -05002995 error = -EEXIST;
Al Virof8310c52012-07-30 11:50:30 +04002996 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
Al Virofb1cc552009-12-24 01:58:28 -05002997 goto exit_dput;
2998
David Howells9875cf82011-01-14 18:45:21 +00002999 error = follow_managed(path, nd->flags);
3000 if (error < 0)
3001 goto exit_dput;
Al Virofb1cc552009-12-24 01:58:28 -05003002
Al Viroa3fbbde2011-11-07 21:21:26 +00003003 if (error)
3004 nd->flags |= LOOKUP_JUMPED;
3005
Miklos Szeredidecf3402012-05-21 17:30:08 +02003006 BUG_ON(nd->flags & LOOKUP_RCU);
3007 inode = path->dentry->d_inode;
Miklos Szeredi5f5daac2012-05-21 17:30:14 +02003008finish_lookup:
3009 /* we _can_ be in RCU mode here */
Al Virofb1cc552009-12-24 01:58:28 -05003010 error = -ENOENT;
Al Viro22213312014-04-19 12:30:58 -04003011 if (!inode || d_is_negative(path->dentry)) {
Miklos Szeredi54c33e72012-05-21 17:30:10 +02003012 path_to_nameidata(path, nd);
Al Viro2675a4e2012-06-22 12:41:10 +04003013 goto out;
Miklos Szeredi54c33e72012-05-21 17:30:10 +02003014 }
Al Viro9e67f362009-12-26 07:04:50 -05003015
David Howellsb18825a2013-09-12 19:22:53 +01003016 if (should_follow_link(path->dentry, !symlink_ok)) {
Miklos Szeredid45ea862012-05-21 17:30:09 +02003017 if (nd->flags & LOOKUP_RCU) {
3018 if (unlikely(unlazy_walk(nd, path->dentry))) {
3019 error = -ECHILD;
Al Viro2675a4e2012-06-22 12:41:10 +04003020 goto out;
Miklos Szeredid45ea862012-05-21 17:30:09 +02003021 }
3022 }
3023 BUG_ON(inode != path->dentry->d_inode);
Al Viro2675a4e2012-06-22 12:41:10 +04003024 return 1;
Miklos Szeredid45ea862012-05-21 17:30:09 +02003025 }
Al Virofb1cc552009-12-24 01:58:28 -05003026
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003027 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
3028 path_to_nameidata(path, nd);
3029 } else {
3030 save_parent.dentry = nd->path.dentry;
3031 save_parent.mnt = mntget(path->mnt);
3032 nd->path.dentry = path->dentry;
3033
3034 }
Miklos Szeredidecf3402012-05-21 17:30:08 +02003035 nd->inode = inode;
Al Viroa3fbbde2011-11-07 21:21:26 +00003036 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
Al Virobc77daa2013-06-06 09:12:33 -04003037finish_open:
Al Viroa3fbbde2011-11-07 21:21:26 +00003038 error = complete_walk(nd);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003039 if (error) {
3040 path_put(&save_parent);
Al Viro2675a4e2012-06-22 12:41:10 +04003041 return error;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003042 }
Al Virobc77daa2013-06-06 09:12:33 -04003043 audit_inode(name, nd->path.dentry, 0);
Al Virofb1cc552009-12-24 01:58:28 -05003044 error = -EISDIR;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003045 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04003046 goto out;
Miklos Szerediaf2f5542012-05-21 17:30:11 +02003047 error = -ENOTDIR;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003048 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04003049 goto out;
Al Viro6c0d46c2011-03-09 00:59:59 -05003050 if (!S_ISREG(nd->inode->i_mode))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003051 will_truncate = false;
Al Viro6c0d46c2011-03-09 00:59:59 -05003052
Al Viro0f9d1a12011-03-09 00:13:14 -05003053 if (will_truncate) {
3054 error = mnt_want_write(nd->path.mnt);
3055 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003056 goto out;
Al Viro64894cf2012-07-31 00:53:35 +04003057 got_write = true;
Al Viro0f9d1a12011-03-09 00:13:14 -05003058 }
Miklos Szeredie83db162012-06-05 15:10:29 +02003059finish_open_created:
Al Virobcda7652011-03-13 16:42:14 -04003060 error = may_open(&nd->path, acc_mode, open_flag);
Al Viroca344a892011-03-09 00:36:45 -05003061 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003062 goto out;
Miklos Szeredi4aa7c632014-10-24 00:14:35 +02003063
3064 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3065 error = vfs_open(&nd->path, file, current_cred());
3066 if (!error) {
3067 *opened |= FILE_OPENED;
3068 } else {
Al Viro30d90492012-06-22 12:40:19 +04003069 if (error == -EOPENSTALE)
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003070 goto stale_open;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003071 goto out;
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003072 }
Miklos Szeredia8277b92012-06-05 15:10:32 +02003073opened:
Al Viro2675a4e2012-06-22 12:41:10 +04003074 error = open_check_o_direct(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003075 if (error)
3076 goto exit_fput;
Dmitry Kasatkin3034a142014-06-27 18:15:44 +03003077 error = ima_file_check(file, op->acc_mode, *opened);
Miklos Szerediaa4caad2012-06-05 15:10:28 +02003078 if (error)
3079 goto exit_fput;
3080
3081 if (will_truncate) {
Al Viro2675a4e2012-06-22 12:41:10 +04003082 error = handle_truncate(file);
Miklos Szerediaa4caad2012-06-05 15:10:28 +02003083 if (error)
3084 goto exit_fput;
Al Viro0f9d1a12011-03-09 00:13:14 -05003085 }
Al Viroca344a892011-03-09 00:36:45 -05003086out:
Al Viro64894cf2012-07-31 00:53:35 +04003087 if (got_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05003088 mnt_drop_write(nd->path.mnt);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003089 path_put(&save_parent);
Miklos Szeredie276ae62012-05-21 17:30:06 +02003090 terminate_walk(nd);
Al Viro2675a4e2012-06-22 12:41:10 +04003091 return error;
Al Virofb1cc552009-12-24 01:58:28 -05003092
Al Virofb1cc552009-12-24 01:58:28 -05003093exit_dput:
3094 path_put_conditional(path, nd);
Al Viroca344a892011-03-09 00:36:45 -05003095 goto out;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003096exit_fput:
Al Viro2675a4e2012-06-22 12:41:10 +04003097 fput(file);
3098 goto out;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003099
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003100stale_open:
3101 /* If no saved parent or already retried then can't retry */
3102 if (!save_parent.dentry || retried)
3103 goto out;
3104
3105 BUG_ON(save_parent.dentry != dir);
3106 path_put(&nd->path);
3107 nd->path = save_parent;
3108 nd->inode = dir->d_inode;
3109 save_parent.mnt = NULL;
3110 save_parent.dentry = NULL;
Al Viro64894cf2012-07-31 00:53:35 +04003111 if (got_write) {
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003112 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04003113 got_write = false;
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003114 }
3115 retried = true;
3116 goto retry_lookup;
Al Virofb1cc552009-12-24 01:58:28 -05003117}
3118
Al Viro60545d02013-06-07 01:20:27 -04003119static int do_tmpfile(int dfd, struct filename *pathname,
3120 struct nameidata *nd, int flags,
3121 const struct open_flags *op,
3122 struct file *file, int *opened)
3123{
3124 static const struct qstr name = QSTR_INIT("/", 1);
3125 struct dentry *dentry, *child;
3126 struct inode *dir;
3127 int error = path_lookupat(dfd, pathname->name,
3128 flags | LOOKUP_DIRECTORY, nd);
3129 if (unlikely(error))
3130 return error;
3131 error = mnt_want_write(nd->path.mnt);
3132 if (unlikely(error))
3133 goto out;
3134 /* we want directory to be writable */
3135 error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3136 if (error)
3137 goto out2;
3138 dentry = nd->path.dentry;
3139 dir = dentry->d_inode;
3140 if (!dir->i_op->tmpfile) {
3141 error = -EOPNOTSUPP;
3142 goto out2;
3143 }
3144 child = d_alloc(dentry, &name);
3145 if (unlikely(!child)) {
3146 error = -ENOMEM;
3147 goto out2;
3148 }
3149 nd->flags &= ~LOOKUP_DIRECTORY;
3150 nd->flags |= op->intent;
3151 dput(nd->path.dentry);
3152 nd->path.dentry = child;
3153 error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3154 if (error)
3155 goto out2;
3156 audit_inode(pathname, nd->path.dentry, 0);
Eric Rannaud69a91c22014-10-30 01:51:01 -07003157 /* Don't check for other permissions, the inode was just created */
3158 error = may_open(&nd->path, MAY_OPEN, op->open_flag);
Al Viro60545d02013-06-07 01:20:27 -04003159 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 Szeredi0d7a8552014-10-24 00:14:37 +02004213 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredida1ce062014-04-01 17:08:43 +02004214 return -EINVAL;
4215
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004216 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4217 (flags & RENAME_EXCHANGE))
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004218 return -EINVAL;
4219
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004220 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4221 return -EPERM;
4222
Jeff Laytonc6a94282012-12-11 12:10:10 -05004223retry:
4224 from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04004225 if (IS_ERR(from)) {
4226 error = PTR_ERR(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004227 goto exit;
Jeff Layton91a27b22012-10-10 15:25:28 -04004228 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004229
Jeff Laytonc6a94282012-12-11 12:10:10 -05004230 to = user_path_parent(newdfd, newname, &newnd, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04004231 if (IS_ERR(to)) {
4232 error = PTR_ERR(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004233 goto exit1;
Jeff Layton91a27b22012-10-10 15:25:28 -04004234 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004235
4236 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08004237 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004238 goto exit2;
4239
Jan Blunck4ac91372008-02-14 19:34:32 -08004240 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004241 error = -EBUSY;
4242 if (oldnd.last_type != LAST_NORM)
4243 goto exit2;
4244
Jan Blunck4ac91372008-02-14 19:34:32 -08004245 new_dir = newnd.path.dentry;
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004246 if (flags & RENAME_NOREPLACE)
4247 error = -EEXIST;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004248 if (newnd.last_type != LAST_NORM)
4249 goto exit2;
4250
Jan Karac30dabf2012-06-12 16:20:30 +02004251 error = mnt_want_write(oldnd.path.mnt);
4252 if (error)
4253 goto exit2;
4254
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09004255 oldnd.flags &= ~LOOKUP_PARENT;
4256 newnd.flags &= ~LOOKUP_PARENT;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004257 if (!(flags & RENAME_EXCHANGE))
4258 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09004259
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004260retry_deleg:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004261 trap = lock_rename(new_dir, old_dir);
4262
Christoph Hellwig49705b72005-11-08 21:35:06 -08004263 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004264 error = PTR_ERR(old_dentry);
4265 if (IS_ERR(old_dentry))
4266 goto exit3;
4267 /* source must exist */
4268 error = -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01004269 if (d_is_negative(old_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004270 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08004271 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004272 error = PTR_ERR(new_dentry);
4273 if (IS_ERR(new_dentry))
4274 goto exit4;
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004275 error = -EEXIST;
4276 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4277 goto exit5;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004278 if (flags & RENAME_EXCHANGE) {
4279 error = -ENOENT;
4280 if (d_is_negative(new_dentry))
4281 goto exit5;
4282
4283 if (!d_is_dir(new_dentry)) {
4284 error = -ENOTDIR;
4285 if (newnd.last.name[newnd.last.len])
4286 goto exit5;
4287 }
4288 }
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004289 /* unless the source is a directory trailing slashes give -ENOTDIR */
4290 if (!d_is_dir(old_dentry)) {
4291 error = -ENOTDIR;
4292 if (oldnd.last.name[oldnd.last.len])
4293 goto exit5;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004294 if (!(flags & RENAME_EXCHANGE) && newnd.last.name[newnd.last.len])
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004295 goto exit5;
4296 }
4297 /* source should not be ancestor of target */
4298 error = -EINVAL;
4299 if (old_dentry == trap)
4300 goto exit5;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004301 /* target should not be an ancestor of source */
Miklos Szeredida1ce062014-04-01 17:08:43 +02004302 if (!(flags & RENAME_EXCHANGE))
4303 error = -ENOTEMPTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004304 if (new_dentry == trap)
4305 goto exit5;
4306
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004307 error = security_path_rename(&oldnd.path, old_dentry,
Miklos Szeredi0b3974e2014-04-01 17:08:43 +02004308 &newnd.path, new_dentry, flags);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004309 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02004310 goto exit5;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004311 error = vfs_rename(old_dir->d_inode, old_dentry,
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004312 new_dir->d_inode, new_dentry,
4313 &delegated_inode, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004314exit5:
4315 dput(new_dentry);
4316exit4:
4317 dput(old_dentry);
4318exit3:
4319 unlock_rename(new_dir, old_dir);
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004320 if (delegated_inode) {
4321 error = break_deleg_wait(&delegated_inode);
4322 if (!error)
4323 goto retry_deleg;
4324 }
Jan Karac30dabf2012-06-12 16:20:30 +02004325 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004326exit2:
Jeff Laytonc6a94282012-12-11 12:10:10 -05004327 if (retry_estale(error, lookup_flags))
4328 should_retry = true;
Jan Blunck1d957f92008-02-14 19:34:35 -08004329 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04004330 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004331exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08004332 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004333 putname(from);
Jeff Laytonc6a94282012-12-11 12:10:10 -05004334 if (should_retry) {
4335 should_retry = false;
4336 lookup_flags |= LOOKUP_REVAL;
4337 goto retry;
4338 }
Al Viro2ad94ae2008-07-21 09:32:51 -04004339exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004340 return error;
4341}
4342
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004343SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4344 int, newdfd, const char __user *, newname)
4345{
4346 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4347}
4348
Heiko Carstensa26eab22009-01-14 14:14:17 +01004349SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004350{
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004351 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004352}
4353
Miklos Szeredi787fb6b2014-10-24 00:14:36 +02004354int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4355{
4356 int error = may_create(dir, dentry);
4357 if (error)
4358 return error;
4359
4360 if (!dir->i_op->mknod)
4361 return -EPERM;
4362
4363 return dir->i_op->mknod(dir, dentry,
4364 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4365}
4366EXPORT_SYMBOL(vfs_whiteout);
4367
Al Viro5d826c82014-03-14 13:42:45 -04004368int readlink_copy(char __user *buffer, int buflen, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004369{
Al Viro5d826c82014-03-14 13:42:45 -04004370 int len = PTR_ERR(link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004371 if (IS_ERR(link))
4372 goto out;
4373
4374 len = strlen(link);
4375 if (len > (unsigned) buflen)
4376 len = buflen;
4377 if (copy_to_user(buffer, link, len))
4378 len = -EFAULT;
4379out:
4380 return len;
4381}
Al Viro5d826c82014-03-14 13:42:45 -04004382EXPORT_SYMBOL(readlink_copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004383
4384/*
4385 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4386 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4387 * using) it for any given inode is up to filesystem.
4388 */
4389int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4390{
4391 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004392 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07004393 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004394
Linus Torvalds1da177e2005-04-16 15:20:36 -07004395 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004396 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07004397 if (IS_ERR(cookie))
4398 return PTR_ERR(cookie);
4399
Al Viro5d826c82014-03-14 13:42:45 -04004400 res = readlink_copy(buffer, buflen, nd_get_link(&nd));
Marcin Slusarz694a1762008-06-09 16:40:37 -07004401 if (dentry->d_inode->i_op->put_link)
4402 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
4403 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004404}
Al Viro4d359502014-03-14 12:20:17 -04004405EXPORT_SYMBOL(generic_readlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004406
Linus Torvalds1da177e2005-04-16 15:20:36 -07004407/* get the link contents into pagecache */
4408static char *page_getlink(struct dentry * dentry, struct page **ppage)
4409{
Duane Griffinebd09ab2008-12-19 20:47:12 +00004410 char *kaddr;
4411 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004412 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07004413 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004414 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07004415 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004416 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00004417 kaddr = kmap(page);
4418 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4419 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004420}
4421
4422int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4423{
4424 struct page *page = NULL;
Al Viro5d826c82014-03-14 13:42:45 -04004425 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004426 if (page) {
4427 kunmap(page);
4428 page_cache_release(page);
4429 }
4430 return res;
4431}
Al Viro4d359502014-03-14 12:20:17 -04004432EXPORT_SYMBOL(page_readlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004433
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004434void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004435{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004436 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004437 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004438 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004439}
Al Viro4d359502014-03-14 12:20:17 -04004440EXPORT_SYMBOL(page_follow_link_light);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004441
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004442void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004443{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004444 struct page *page = cookie;
4445
4446 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004447 kunmap(page);
4448 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004449 }
4450}
Al Viro4d359502014-03-14 12:20:17 -04004451EXPORT_SYMBOL(page_put_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004452
Nick Piggin54566b22009-01-04 12:00:53 -08004453/*
4454 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4455 */
4456int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004457{
4458 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004459 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07004460 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08004461 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004462 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08004463 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4464 if (nofs)
4465 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004466
NeilBrown7e53cac2006-03-25 03:07:57 -08004467retry:
Nick Pigginafddba42007-10-16 01:25:01 -07004468 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08004469 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004470 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07004471 goto fail;
4472
Cong Wange8e3c3d2011-11-25 23:14:27 +08004473 kaddr = kmap_atomic(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004474 memcpy(kaddr, symname, len-1);
Cong Wange8e3c3d2011-11-25 23:14:27 +08004475 kunmap_atomic(kaddr);
Nick Pigginafddba42007-10-16 01:25:01 -07004476
4477 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4478 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004479 if (err < 0)
4480 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07004481 if (err < len-1)
4482 goto retry;
4483
Linus Torvalds1da177e2005-04-16 15:20:36 -07004484 mark_inode_dirty(inode);
4485 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004486fail:
4487 return err;
4488}
Al Viro4d359502014-03-14 12:20:17 -04004489EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004490
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004491int page_symlink(struct inode *inode, const char *symname, int len)
4492{
4493 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08004494 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004495}
Al Viro4d359502014-03-14 12:20:17 -04004496EXPORT_SYMBOL(page_symlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004497
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08004498const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004499 .readlink = generic_readlink,
4500 .follow_link = page_follow_link_light,
4501 .put_link = page_put_link,
4502};
Linus Torvalds1da177e2005-04-16 15:20:36 -07004503EXPORT_SYMBOL(page_symlink_inode_operations);