blob: f624d132e01e6f5488fd0d9af13007b015e1e178 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050018#include <linux/export.h>
David S. Miller44696902012-05-23 20:12:50 -070019#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/slab.h>
21#include <linux/fs.h>
22#include <linux/namei.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/pagemap.h>
Robert Love0eeca282005-07-12 17:06:03 -040024#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/personality.h>
26#include <linux/security.h>
Mimi Zohar6146f0d2009-02-04 09:06:57 -050027#include <linux/ima.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/syscalls.h>
29#include <linux/mount.h>
30#include <linux/audit.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080031#include <linux/capability.h>
Trond Myklebust834f2a42005-10-18 14:20:16 -070032#include <linux/file.h>
Ulrich Drepper5590ff02006-01-18 17:43:53 -080033#include <linux/fcntl.h>
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -070034#include <linux/device_cgroup.h>
Al Viro5ad4e532009-03-29 19:50:06 -040035#include <linux/fs_struct.h>
Linus Torvaldse77819e2011-07-22 19:30:19 -070036#include <linux/posix_acl.h>
Linus Torvalds99d263d2014-09-13 11:30:10 -070037#include <linux/hash.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include <asm/uaccess.h>
39
Eric Parise81e3f42009-12-04 15:47:36 -050040#include "internal.h"
Al Viroc7105362011-11-24 18:22:03 -050041#include "mount.h"
Eric Parise81e3f42009-12-04 15:47:36 -050042
Linus Torvalds1da177e2005-04-16 15:20:36 -070043/* [Feb-1997 T. Schoebel-Theuer]
44 * Fundamental changes in the pathname lookup mechanisms (namei)
45 * were necessary because of omirr. The reason is that omirr needs
46 * to know the _real_ pathname, not the user-supplied one, in case
47 * of symlinks (and also when transname replacements occur).
48 *
49 * The new code replaces the old recursive symlink resolution with
50 * an iterative one (in case of non-nested symlink chains). It does
51 * this with calls to <fs>_follow_link().
52 * As a side effect, dir_namei(), _namei() and follow_link() are now
53 * replaced with a single function lookup_dentry() that can handle all
54 * the special cases of the former code.
55 *
56 * With the new dcache, the pathname is stored at each inode, at least as
57 * long as the refcount of the inode is positive. As a side effect, the
58 * size of the dcache depends on the inode cache and thus is dynamic.
59 *
60 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
61 * resolution to correspond with current state of the code.
62 *
63 * Note that the symlink resolution is not *completely* iterative.
64 * There is still a significant amount of tail- and mid- recursion in
65 * the algorithm. Also, note that <fs>_readlink() is not used in
66 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
67 * may return different results than <fs>_follow_link(). Many virtual
68 * filesystems (including /proc) exhibit this behavior.
69 */
70
71/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
72 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
73 * and the name already exists in form of a symlink, try to create the new
74 * name indicated by the symlink. The old code always complained that the
75 * name already exists, due to not following the symlink even if its target
76 * is nonexistent. The new semantics affects also mknod() and link() when
Lucas De Marchi25985ed2011-03-30 22:57:33 -030077 * the name is a symlink pointing to a non-existent name.
Linus Torvalds1da177e2005-04-16 15:20:36 -070078 *
79 * I don't know which semantics is the right one, since I have no access
80 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
81 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
82 * "old" one. Personally, I think the new semantics is much more logical.
83 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
84 * file does succeed in both HP-UX and SunOs, but not in Solaris
85 * and in the old Linux semantics.
86 */
87
88/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
89 * semantics. See the comments in "open_namei" and "do_link" below.
90 *
91 * [10-Sep-98 Alan Modra] Another symlink change.
92 */
93
94/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
95 * inside the path - always follow.
96 * in the last component in creation/removal/renaming - never follow.
97 * if LOOKUP_FOLLOW passed - follow.
98 * if the pathname has trailing slashes - follow.
99 * otherwise - don't follow.
100 * (applied in that order).
101 *
102 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
103 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
104 * During the 2.4 we need to fix the userland stuff depending on it -
105 * hopefully we will be able to get rid of that wart in 2.5. So far only
106 * XEmacs seems to be relying on it...
107 */
108/*
109 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800110 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 * any extra contention...
112 */
113
114/* In order to reduce some races, while at the same time doing additional
115 * checking and hopefully speeding things up, we copy filenames to the
116 * kernel data space before using them..
117 *
118 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
119 * PATH_MAX includes the nul terminator --RR.
120 */
Jeff Layton91a27b22012-10-10 15:25:28 -0400121
Al Virofd2f7cb2015-02-22 20:07:13 -0500122#define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
Jeff Layton7950e382012-10-10 16:43:13 -0400123
David Drysdale51f39a12014-12-12 16:57:29 -0800124struct filename *
Jeff Layton91a27b22012-10-10 15:25:28 -0400125getname_flags(const char __user *filename, int flags, int *empty)
126{
Al Viro94b5d262015-02-22 19:38:03 -0500127 struct filename *result;
Jeff Layton7950e382012-10-10 16:43:13 -0400128 char *kname;
Al Viro94b5d262015-02-22 19:38:03 -0500129 int len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130
Jeff Layton7ac86262012-10-10 15:25:28 -0400131 result = audit_reusename(filename);
132 if (result)
133 return result;
134
Jeff Layton7950e382012-10-10 16:43:13 -0400135 result = __getname();
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700136 if (unlikely(!result))
Eric Paris4043cde2012-01-03 14:23:08 -0500137 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138
Jeff Layton7950e382012-10-10 16:43:13 -0400139 /*
140 * First, try to embed the struct filename inside the names_cache
141 * allocation
142 */
Al Virofd2f7cb2015-02-22 20:07:13 -0500143 kname = (char *)result->iname;
Jeff Layton91a27b22012-10-10 15:25:28 -0400144 result->name = kname;
Jeff Layton7950e382012-10-10 16:43:13 -0400145
Al Viro94b5d262015-02-22 19:38:03 -0500146 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
Jeff Layton91a27b22012-10-10 15:25:28 -0400147 if (unlikely(len < 0)) {
Al Viro94b5d262015-02-22 19:38:03 -0500148 __putname(result);
149 return ERR_PTR(len);
Jeff Layton91a27b22012-10-10 15:25:28 -0400150 }
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700151
Jeff Layton7950e382012-10-10 16:43:13 -0400152 /*
153 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
154 * separate struct filename so we can dedicate the entire
155 * names_cache allocation for the pathname, and re-do the copy from
156 * userland.
157 */
Al Viro94b5d262015-02-22 19:38:03 -0500158 if (unlikely(len == EMBEDDED_NAME_MAX)) {
Al Virofd2f7cb2015-02-22 20:07:13 -0500159 const size_t size = offsetof(struct filename, iname[1]);
Jeff Layton7950e382012-10-10 16:43:13 -0400160 kname = (char *)result;
161
Al Virofd2f7cb2015-02-22 20:07:13 -0500162 /*
163 * size is chosen that way we to guarantee that
164 * result->iname[0] is within the same object and that
165 * kname can't be equal to result->iname, no matter what.
166 */
167 result = kzalloc(size, GFP_KERNEL);
Al Viro94b5d262015-02-22 19:38:03 -0500168 if (unlikely(!result)) {
169 __putname(kname);
170 return ERR_PTR(-ENOMEM);
Jeff Layton7950e382012-10-10 16:43:13 -0400171 }
172 result->name = kname;
Al Viro94b5d262015-02-22 19:38:03 -0500173 len = strncpy_from_user(kname, filename, PATH_MAX);
174 if (unlikely(len < 0)) {
175 __putname(kname);
176 kfree(result);
177 return ERR_PTR(len);
178 }
179 if (unlikely(len == PATH_MAX)) {
180 __putname(kname);
181 kfree(result);
182 return ERR_PTR(-ENAMETOOLONG);
183 }
Jeff Layton7950e382012-10-10 16:43:13 -0400184 }
185
Al Viro94b5d262015-02-22 19:38:03 -0500186 result->refcnt = 1;
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700187 /* The empty path is special. */
188 if (unlikely(!len)) {
189 if (empty)
Eric Paris4043cde2012-01-03 14:23:08 -0500190 *empty = 1;
Al Viro94b5d262015-02-22 19:38:03 -0500191 if (!(flags & LOOKUP_EMPTY)) {
192 putname(result);
193 return ERR_PTR(-ENOENT);
194 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 }
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700196
Jeff Layton7950e382012-10-10 16:43:13 -0400197 result->uptr = filename;
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800198 result->aname = NULL;
Jeff Layton7950e382012-10-10 16:43:13 -0400199 audit_getname(result);
200 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201}
202
Jeff Layton91a27b22012-10-10 15:25:28 -0400203struct filename *
204getname(const char __user * filename)
Al Virof52e0c12011-03-14 18:56:51 -0400205{
Linus Torvaldsf7493e52012-03-22 16:10:40 -0700206 return getname_flags(filename, 0, NULL);
Al Virof52e0c12011-03-14 18:56:51 -0400207}
208
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800209struct filename *
210getname_kernel(const char * filename)
211{
212 struct filename *result;
Paul Moore08518542015-01-21 23:59:56 -0500213 int len = strlen(filename) + 1;
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800214
215 result = __getname();
216 if (unlikely(!result))
217 return ERR_PTR(-ENOMEM);
218
Paul Moore08518542015-01-21 23:59:56 -0500219 if (len <= EMBEDDED_NAME_MAX) {
Al Virofd2f7cb2015-02-22 20:07:13 -0500220 result->name = (char *)result->iname;
Paul Moore08518542015-01-21 23:59:56 -0500221 } else if (len <= PATH_MAX) {
222 struct filename *tmp;
223
224 tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
225 if (unlikely(!tmp)) {
226 __putname(result);
227 return ERR_PTR(-ENOMEM);
228 }
229 tmp->name = (char *)result;
Paul Moore08518542015-01-21 23:59:56 -0500230 result = tmp;
231 } else {
232 __putname(result);
233 return ERR_PTR(-ENAMETOOLONG);
234 }
235 memcpy((char *)result->name, filename, len);
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800236 result->uptr = NULL;
237 result->aname = NULL;
Paul Moore55422d02015-01-22 00:00:23 -0500238 result->refcnt = 1;
Paul Moorefd3522f2015-01-22 00:00:10 -0500239 audit_getname(result);
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800240
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800241 return result;
242}
243
Jeff Layton91a27b22012-10-10 15:25:28 -0400244void putname(struct filename *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245{
Paul Moore55422d02015-01-22 00:00:23 -0500246 BUG_ON(name->refcnt <= 0);
247
248 if (--name->refcnt > 0)
249 return;
250
Al Virofd2f7cb2015-02-22 20:07:13 -0500251 if (name->name != name->iname) {
Paul Moore55422d02015-01-22 00:00:23 -0500252 __putname(name->name);
253 kfree(name);
254 } else
255 __putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257
Linus Torvaldse77819e2011-07-22 19:30:19 -0700258static int check_acl(struct inode *inode, int mask)
259{
Linus Torvalds84635d62011-07-25 22:47:03 -0700260#ifdef CONFIG_FS_POSIX_ACL
Linus Torvaldse77819e2011-07-22 19:30:19 -0700261 struct posix_acl *acl;
262
Linus Torvaldse77819e2011-07-22 19:30:19 -0700263 if (mask & MAY_NOT_BLOCK) {
Al Viro35678662011-08-02 21:32:13 -0400264 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
265 if (!acl)
Linus Torvaldse77819e2011-07-22 19:30:19 -0700266 return -EAGAIN;
Al Viro35678662011-08-02 21:32:13 -0400267 /* no ->get_acl() calls in RCU mode... */
268 if (acl == ACL_NOT_CACHED)
269 return -ECHILD;
Ari Savolainen206b1d02011-08-06 19:43:07 +0300270 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
Linus Torvaldse77819e2011-07-22 19:30:19 -0700271 }
272
Christoph Hellwig2982baa2013-12-20 05:16:38 -0800273 acl = get_acl(inode, ACL_TYPE_ACCESS);
274 if (IS_ERR(acl))
275 return PTR_ERR(acl);
Linus Torvaldse77819e2011-07-22 19:30:19 -0700276 if (acl) {
277 int error = posix_acl_permission(inode, acl, mask);
278 posix_acl_release(acl);
279 return error;
280 }
Linus Torvalds84635d62011-07-25 22:47:03 -0700281#endif
Linus Torvaldse77819e2011-07-22 19:30:19 -0700282
283 return -EAGAIN;
284}
285
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700286/*
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530287 * This does the basic permission checking
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700288 */
Al Viro7e401452011-06-20 19:12:17 -0400289static int acl_permission_check(struct inode *inode, int mask)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700290{
Linus Torvalds26cf46b2011-05-13 11:51:01 -0700291 unsigned int mode = inode->i_mode;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700292
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -0800293 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700294 mode >>= 6;
295 else {
Linus Torvaldse77819e2011-07-22 19:30:19 -0700296 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
Al Viro7e401452011-06-20 19:12:17 -0400297 int error = check_acl(inode, mask);
Nick Pigginb74c79e2011-01-07 17:49:58 +1100298 if (error != -EAGAIN)
299 return error;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700300 }
301
302 if (in_group_p(inode->i_gid))
303 mode >>= 3;
304 }
305
306 /*
307 * If the DACs are ok we don't need any capability check.
308 */
Al Viro9c2c7032011-06-20 19:06:22 -0400309 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700310 return 0;
311 return -EACCES;
312}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313
314/**
Nick Pigginb74c79e2011-01-07 17:49:58 +1100315 * generic_permission - check for access rights on a Posix-like filesystem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 * @inode: inode to check access rights for
Andreas Gruenbacher8fd90c82011-10-23 23:13:30 +0530317 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 *
319 * Used to check for read/write/execute permissions on a file.
320 * We use "fsuid" for this, letting us set arbitrary permissions
321 * for filesystem access without changing the "normal" uids which
Nick Pigginb74c79e2011-01-07 17:49:58 +1100322 * are used for other things.
323 *
324 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
325 * request cannot be satisfied (eg. requires blocking or too much complexity).
326 * It would then be called again in ref-walk mode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 */
Al Viro2830ba72011-06-20 19:16:29 -0400328int generic_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700330 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331
332 /*
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530333 * Do the basic permission checks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 */
Al Viro7e401452011-06-20 19:12:17 -0400335 ret = acl_permission_check(inode, mask);
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700336 if (ret != -EACCES)
337 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338
Al Virod594e7e2011-06-20 19:55:42 -0400339 if (S_ISDIR(inode->i_mode)) {
340 /* DACs are overridable for directories */
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700341 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
Al Virod594e7e2011-06-20 19:55:42 -0400342 return 0;
343 if (!(mask & MAY_WRITE))
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700344 if (capable_wrt_inode_uidgid(inode,
345 CAP_DAC_READ_SEARCH))
Al Virod594e7e2011-06-20 19:55:42 -0400346 return 0;
347 return -EACCES;
348 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 /*
350 * Read/write DACs are always overridable.
Al Virod594e7e2011-06-20 19:55:42 -0400351 * Executable DACs are overridable when there is
352 * at least one exec bit set.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 */
Al Virod594e7e2011-06-20 19:55:42 -0400354 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700355 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 return 0;
357
358 /*
359 * Searching includes executable on directories, else just read.
360 */
Serge E. Hallyn7ea66002009-12-29 14:50:19 -0600361 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
Al Virod594e7e2011-06-20 19:55:42 -0400362 if (mask == MAY_READ)
Andy Lutomirski23adbe12014-06-10 12:45:42 -0700363 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 return 0;
365
366 return -EACCES;
367}
Al Viro4d359502014-03-14 12:20:17 -0400368EXPORT_SYMBOL(generic_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700370/*
371 * We _really_ want to just do "generic_permission()" without
372 * even looking at the inode->i_op values. So we keep a cache
373 * flag in inode->i_opflags, that says "this has not special
374 * permission function, use the fast case".
375 */
376static inline int do_inode_permission(struct inode *inode, int mask)
377{
378 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
379 if (likely(inode->i_op->permission))
380 return inode->i_op->permission(inode, mask);
381
382 /* This gets set once for the inode lifetime */
383 spin_lock(&inode->i_lock);
384 inode->i_opflags |= IOP_FASTPERM;
385 spin_unlock(&inode->i_lock);
386 }
387 return generic_permission(inode, mask);
388}
389
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200390/**
David Howells0bdaea92012-06-25 12:55:46 +0100391 * __inode_permission - Check for access rights to a given inode
392 * @inode: Inode to check permission on
393 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200394 *
David Howells0bdaea92012-06-25 12:55:46 +0100395 * Check for read/write/execute permissions on an inode.
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530396 *
397 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
David Howells0bdaea92012-06-25 12:55:46 +0100398 *
399 * This does not check for a read-only file system. You probably want
400 * inode_permission().
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200401 */
David Howells0bdaea92012-06-25 12:55:46 +0100402int __inode_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403{
Al Viroe6305c42008-07-15 21:03:57 -0400404 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700406 if (unlikely(mask & MAY_WRITE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 /*
408 * Nobody gets write access to an immutable file.
409 */
410 if (IS_IMMUTABLE(inode))
411 return -EACCES;
412 }
413
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700414 retval = do_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 if (retval)
416 return retval;
417
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -0700418 retval = devcgroup_inode_permission(inode, mask);
419 if (retval)
420 return retval;
421
Eric Parisd09ca732010-07-23 11:43:57 -0400422 return security_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423}
Miklos Szeredibd5d0852014-10-24 00:14:35 +0200424EXPORT_SYMBOL(__inode_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425
Al Virof4d6ff82011-06-19 13:14:21 -0400426/**
David Howells0bdaea92012-06-25 12:55:46 +0100427 * sb_permission - Check superblock-level permissions
428 * @sb: Superblock of inode to check permission on
Randy Dunlap55852632012-08-18 17:39:25 -0700429 * @inode: Inode to check permission on
David Howells0bdaea92012-06-25 12:55:46 +0100430 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
431 *
432 * Separate out file-system wide checks from inode-specific permission checks.
433 */
434static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
435{
436 if (unlikely(mask & MAY_WRITE)) {
437 umode_t mode = inode->i_mode;
438
439 /* Nobody gets write access to a read-only fs. */
440 if ((sb->s_flags & MS_RDONLY) &&
441 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
442 return -EROFS;
443 }
444 return 0;
445}
446
447/**
448 * inode_permission - Check for access rights to a given inode
449 * @inode: Inode to check permission on
450 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
451 *
452 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
453 * this, letting us set arbitrary permissions for filesystem access without
454 * changing the "normal" UIDs which are used for other things.
455 *
456 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
457 */
458int inode_permission(struct inode *inode, int mask)
459{
460 int retval;
461
462 retval = sb_permission(inode->i_sb, inode, mask);
463 if (retval)
464 return retval;
465 return __inode_permission(inode, mask);
466}
Al Viro4d359502014-03-14 12:20:17 -0400467EXPORT_SYMBOL(inode_permission);
David Howells0bdaea92012-06-25 12:55:46 +0100468
469/**
Jan Blunck5dd784d02008-02-14 19:34:38 -0800470 * path_get - get a reference to a path
471 * @path: path to get the reference to
472 *
473 * Given a path increment the reference count to the dentry and the vfsmount.
474 */
Al Virodcf787f2013-03-01 23:51:07 -0500475void path_get(const struct path *path)
Jan Blunck5dd784d02008-02-14 19:34:38 -0800476{
477 mntget(path->mnt);
478 dget(path->dentry);
479}
480EXPORT_SYMBOL(path_get);
481
482/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800483 * path_put - put a reference to a path
484 * @path: path to put the reference to
485 *
486 * Given a path decrement the reference count to the dentry and the vfsmount.
487 */
Al Virodcf787f2013-03-01 23:51:07 -0500488void path_put(const struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489{
Jan Blunck1d957f92008-02-14 19:34:35 -0800490 dput(path->dentry);
491 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492}
Jan Blunck1d957f92008-02-14 19:34:35 -0800493EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494
Al Viro894bc8c2015-05-02 07:16:16 -0400495#define EMBEDDED_LEVELS 2
Al Viro1f55a6e2014-11-01 19:30:41 -0400496struct nameidata {
497 struct path path;
Al Viro1cf26652015-05-06 16:01:56 -0400498 struct qstr last;
Al Viro1f55a6e2014-11-01 19:30:41 -0400499 struct path root;
500 struct inode *inode; /* path.dentry.d_inode */
501 unsigned int flags;
Al Viro9883d182015-05-13 07:28:08 -0400502 unsigned seq, m_seq;
Al Viro1f55a6e2014-11-01 19:30:41 -0400503 int last_type;
504 unsigned depth;
NeilBrown756daf22015-03-23 13:37:38 +1100505 int total_link_count;
Al Viro697fc6c2015-05-02 19:38:35 -0400506 struct saved {
507 struct path link;
Al Virofceef392015-12-29 15:58:39 -0500508 struct delayed_call done;
Al Viro697fc6c2015-05-02 19:38:35 -0400509 const char *name;
Al Viro0450b2d2015-05-08 13:23:53 -0400510 unsigned seq;
Al Viro894bc8c2015-05-02 07:16:16 -0400511 } *stack, internal[EMBEDDED_LEVELS];
Al Viro9883d182015-05-13 07:28:08 -0400512 struct filename *name;
513 struct nameidata *saved;
Al Virofceef392015-12-29 15:58:39 -0500514 struct inode *link_inode;
Al Viro9883d182015-05-13 07:28:08 -0400515 unsigned root_seq;
516 int dfd;
Al Viro1f55a6e2014-11-01 19:30:41 -0400517};
518
Al Viro9883d182015-05-13 07:28:08 -0400519static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
Al Viro894bc8c2015-05-02 07:16:16 -0400520{
NeilBrown756daf22015-03-23 13:37:38 +1100521 struct nameidata *old = current->nameidata;
522 p->stack = p->internal;
Al Viroc8a53ee2015-05-12 18:43:07 -0400523 p->dfd = dfd;
524 p->name = name;
NeilBrown756daf22015-03-23 13:37:38 +1100525 p->total_link_count = old ? old->total_link_count : 0;
Al Viro9883d182015-05-13 07:28:08 -0400526 p->saved = old;
NeilBrown756daf22015-03-23 13:37:38 +1100527 current->nameidata = p;
Al Viro894bc8c2015-05-02 07:16:16 -0400528}
529
Al Viro9883d182015-05-13 07:28:08 -0400530static void restore_nameidata(void)
Al Viro894bc8c2015-05-02 07:16:16 -0400531{
Al Viro9883d182015-05-13 07:28:08 -0400532 struct nameidata *now = current->nameidata, *old = now->saved;
NeilBrown756daf22015-03-23 13:37:38 +1100533
534 current->nameidata = old;
535 if (old)
536 old->total_link_count = now->total_link_count;
Al Viroe1a63bb2015-12-05 21:06:33 -0500537 if (now->stack != now->internal)
NeilBrown756daf22015-03-23 13:37:38 +1100538 kfree(now->stack);
Al Viro894bc8c2015-05-02 07:16:16 -0400539}
540
541static int __nd_alloc_stack(struct nameidata *nd)
542{
Al Virobc40aee2015-05-09 13:04:24 -0400543 struct saved *p;
544
545 if (nd->flags & LOOKUP_RCU) {
546 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
547 GFP_ATOMIC);
548 if (unlikely(!p))
549 return -ECHILD;
550 } else {
551 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
Al Viro894bc8c2015-05-02 07:16:16 -0400552 GFP_KERNEL);
Al Virobc40aee2015-05-09 13:04:24 -0400553 if (unlikely(!p))
554 return -ENOMEM;
555 }
Al Viro894bc8c2015-05-02 07:16:16 -0400556 memcpy(p, nd->internal, sizeof(nd->internal));
557 nd->stack = p;
558 return 0;
559}
560
Eric W. Biederman397d4252015-08-15 20:27:13 -0500561/**
562 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
563 * @path: nameidate to verify
564 *
565 * Rename can sometimes move a file or directory outside of a bind
566 * mount, path_connected allows those cases to be detected.
567 */
568static bool path_connected(const struct path *path)
569{
570 struct vfsmount *mnt = path->mnt;
571
572 /* Only bind mounts can have disconnected paths */
573 if (mnt->mnt_root == mnt->mnt_sb->s_root)
574 return true;
575
576 return is_subdir(path->dentry, mnt->mnt_root);
577}
578
Al Viro894bc8c2015-05-02 07:16:16 -0400579static inline int nd_alloc_stack(struct nameidata *nd)
580{
Al Viroda4e0be2015-05-03 20:52:15 -0400581 if (likely(nd->depth != EMBEDDED_LEVELS))
Al Viro894bc8c2015-05-02 07:16:16 -0400582 return 0;
583 if (likely(nd->stack != nd->internal))
584 return 0;
585 return __nd_alloc_stack(nd);
586}
587
Al Viro79733872015-05-09 12:55:43 -0400588static void drop_links(struct nameidata *nd)
589{
590 int i = nd->depth;
591 while (i--) {
592 struct saved *last = nd->stack + i;
Al Virofceef392015-12-29 15:58:39 -0500593 do_delayed_call(&last->done);
594 clear_delayed_call(&last->done);
Al Viro79733872015-05-09 12:55:43 -0400595 }
596}
597
598static void terminate_walk(struct nameidata *nd)
599{
600 drop_links(nd);
601 if (!(nd->flags & LOOKUP_RCU)) {
602 int i;
603 path_put(&nd->path);
604 for (i = 0; i < nd->depth; i++)
605 path_put(&nd->stack[i].link);
Al Viro102b8af2015-05-12 17:35:52 -0400606 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
607 path_put(&nd->root);
608 nd->root.mnt = NULL;
609 }
Al Viro79733872015-05-09 12:55:43 -0400610 } else {
611 nd->flags &= ~LOOKUP_RCU;
612 if (!(nd->flags & LOOKUP_ROOT))
613 nd->root.mnt = NULL;
614 rcu_read_unlock();
615 }
616 nd->depth = 0;
617}
618
619/* path_put is needed afterwards regardless of success or failure */
620static bool legitimize_path(struct nameidata *nd,
621 struct path *path, unsigned seq)
622{
623 int res = __legitimize_mnt(path->mnt, nd->m_seq);
624 if (unlikely(res)) {
625 if (res > 0)
626 path->mnt = NULL;
627 path->dentry = NULL;
628 return false;
629 }
630 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
631 path->dentry = NULL;
632 return false;
633 }
634 return !read_seqcount_retry(&path->dentry->d_seq, seq);
635}
636
637static bool legitimize_links(struct nameidata *nd)
638{
639 int i;
640 for (i = 0; i < nd->depth; i++) {
641 struct saved *last = nd->stack + i;
642 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
643 drop_links(nd);
644 nd->depth = i + 1;
645 return false;
646 }
647 }
648 return true;
649}
650
Al Viro19660af2011-03-25 10:32:48 -0400651/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100652 * Path walking has 2 modes, rcu-walk and ref-walk (see
Al Viro19660af2011-03-25 10:32:48 -0400653 * Documentation/filesystems/path-lookup.txt). In situations when we can't
654 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
Mike Marshall57e37152015-11-30 11:11:59 -0500655 * normal reference counts on dentries and vfsmounts to transition to ref-walk
Al Viro19660af2011-03-25 10:32:48 -0400656 * mode. Refcounts are grabbed at the last known good point before rcu-walk
657 * got stuck, so ref-walk may continue from there. If this is not successful
658 * (eg. a seqcount has changed), then failure is returned and it's up to caller
659 * to restart the path walk from the beginning in ref-walk mode.
Nick Piggin31e6b012011-01-07 17:49:52 +1100660 */
Nick Piggin31e6b012011-01-07 17:49:52 +1100661
662/**
Al Viro19660af2011-03-25 10:32:48 -0400663 * unlazy_walk - try to switch to ref-walk mode.
664 * @nd: nameidata pathwalk data
665 * @dentry: child of nd->path.dentry or NULL
Al Viro6e9918b2015-05-05 09:26:05 -0400666 * @seq: seq number to check dentry against
Randy Dunlap39191622011-01-08 19:36:21 -0800667 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100668 *
Al Viro19660af2011-03-25 10:32:48 -0400669 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
670 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
671 * @nd or NULL. Must be called from rcu-walk context.
Al Viro79733872015-05-09 12:55:43 -0400672 * Nothing should touch nameidata between unlazy_walk() failure and
673 * terminate_walk().
Nick Piggin31e6b012011-01-07 17:49:52 +1100674 */
Al Viro6e9918b2015-05-05 09:26:05 -0400675static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq)
Nick Piggin31e6b012011-01-07 17:49:52 +1100676{
Nick Piggin31e6b012011-01-07 17:49:52 +1100677 struct dentry *parent = nd->path.dentry;
678
679 BUG_ON(!(nd->flags & LOOKUP_RCU));
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700680
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700681 nd->flags &= ~LOOKUP_RCU;
Al Viro79733872015-05-09 12:55:43 -0400682 if (unlikely(!legitimize_links(nd)))
683 goto out2;
684 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
685 goto out2;
686 if (unlikely(!lockref_get_not_dead(&parent->d_lockref)))
687 goto out1;
Al Viro48a066e2013-09-29 22:06:07 -0400688
Linus Torvalds15570082013-09-02 11:38:06 -0700689 /*
690 * For a negative lookup, the lookup sequence point is the parents
691 * sequence point, and it only needs to revalidate the parent dentry.
692 *
693 * For a positive lookup, we need to move both the parent and the
694 * dentry from the RCU domain to be properly refcounted. And the
695 * sequence number in the dentry validates *both* dentry counters,
696 * since we checked the sequence number of the parent after we got
697 * the child sequence number. So we know the parent must still
698 * be valid if the child sequence number is still valid.
699 */
Al Viro19660af2011-03-25 10:32:48 -0400700 if (!dentry) {
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700701 if (read_seqcount_retry(&parent->d_seq, nd->seq))
702 goto out;
Al Viro19660af2011-03-25 10:32:48 -0400703 BUG_ON(nd->inode != parent->d_inode);
704 } else {
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700705 if (!lockref_get_not_dead(&dentry->d_lockref))
706 goto out;
Al Viro6e9918b2015-05-05 09:26:05 -0400707 if (read_seqcount_retry(&dentry->d_seq, seq))
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700708 goto drop_dentry;
Al Viro19660af2011-03-25 10:32:48 -0400709 }
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700710
711 /*
712 * Sequence counts matched. Now make sure that the root is
713 * still valid and get it if required.
714 */
715 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
Al Viro5a8d87e2015-05-12 00:10:18 -0400716 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
717 rcu_read_unlock();
718 dput(dentry);
719 return -ECHILD;
Al Viro79733872015-05-09 12:55:43 -0400720 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100721 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100722
Al Viro8b61e742013-11-08 12:45:01 -0500723 rcu_read_unlock();
Nick Piggin31e6b012011-01-07 17:49:52 +1100724 return 0;
Al Viro19660af2011-03-25 10:32:48 -0400725
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700726drop_dentry:
Al Viro8b61e742013-11-08 12:45:01 -0500727 rcu_read_unlock();
Linus Torvalds15570082013-09-02 11:38:06 -0700728 dput(dentry);
Linus Torvaldsd0d27272013-09-10 12:17:49 -0700729 goto drop_root_mnt;
Al Viro79733872015-05-09 12:55:43 -0400730out2:
731 nd->path.mnt = NULL;
732out1:
733 nd->path.dentry = NULL;
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700734out:
Al Viro8b61e742013-11-08 12:45:01 -0500735 rcu_read_unlock();
Linus Torvaldsd0d27272013-09-10 12:17:49 -0700736drop_root_mnt:
737 if (!(nd->flags & LOOKUP_ROOT))
738 nd->root.mnt = NULL;
Nick Piggin31e6b012011-01-07 17:49:52 +1100739 return -ECHILD;
740}
741
Al Viro79733872015-05-09 12:55:43 -0400742static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq)
743{
744 if (unlikely(!legitimize_path(nd, link, seq))) {
745 drop_links(nd);
746 nd->depth = 0;
747 nd->flags &= ~LOOKUP_RCU;
748 nd->path.mnt = NULL;
749 nd->path.dentry = NULL;
750 if (!(nd->flags & LOOKUP_ROOT))
751 nd->root.mnt = NULL;
752 rcu_read_unlock();
753 } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) {
754 return 0;
755 }
756 path_put(link);
757 return -ECHILD;
758}
759
Al Viro4ce16ef32012-06-10 16:10:59 -0400760static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
Nick Piggin34286d62011-01-07 17:49:57 +1100761{
Al Viro4ce16ef32012-06-10 16:10:59 -0400762 return dentry->d_op->d_revalidate(dentry, flags);
Nick Piggin34286d62011-01-07 17:49:57 +1100763}
764
Al Viro9f1fafe2011-03-25 11:00:12 -0400765/**
766 * complete_walk - successful completion of path walk
767 * @nd: pointer nameidata
Jeff Layton39159de2009-12-07 12:01:50 -0500768 *
Al Viro9f1fafe2011-03-25 11:00:12 -0400769 * If we had been in RCU mode, drop out of it and legitimize nd->path.
770 * Revalidate the final result, unless we'd already done that during
771 * the path walk or the filesystem doesn't ask for it. Return 0 on
772 * success, -error on failure. In case of failure caller does not
773 * need to drop nd->path.
Jeff Layton39159de2009-12-07 12:01:50 -0500774 */
Al Viro9f1fafe2011-03-25 11:00:12 -0400775static int complete_walk(struct nameidata *nd)
Jeff Layton39159de2009-12-07 12:01:50 -0500776{
Al Viro16c2cd72011-02-22 15:50:10 -0500777 struct dentry *dentry = nd->path.dentry;
Jeff Layton39159de2009-12-07 12:01:50 -0500778 int status;
Jeff Layton39159de2009-12-07 12:01:50 -0500779
Al Viro9f1fafe2011-03-25 11:00:12 -0400780 if (nd->flags & LOOKUP_RCU) {
Al Viro9f1fafe2011-03-25 11:00:12 -0400781 if (!(nd->flags & LOOKUP_ROOT))
782 nd->root.mnt = NULL;
Al Viro6e9918b2015-05-05 09:26:05 -0400783 if (unlikely(unlazy_walk(nd, NULL, 0)))
Al Viro48a066e2013-09-29 22:06:07 -0400784 return -ECHILD;
Al Viro9f1fafe2011-03-25 11:00:12 -0400785 }
786
Al Viro16c2cd72011-02-22 15:50:10 -0500787 if (likely(!(nd->flags & LOOKUP_JUMPED)))
Jeff Layton39159de2009-12-07 12:01:50 -0500788 return 0;
789
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500790 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
Al Viro16c2cd72011-02-22 15:50:10 -0500791 return 0;
792
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500793 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
Jeff Layton39159de2009-12-07 12:01:50 -0500794 if (status > 0)
795 return 0;
796
Al Viro16c2cd72011-02-22 15:50:10 -0500797 if (!status)
Jeff Layton39159de2009-12-07 12:01:50 -0500798 status = -ESTALE;
Al Viro16c2cd72011-02-22 15:50:10 -0500799
Jeff Layton39159de2009-12-07 12:01:50 -0500800 return status;
801}
802
Al Viro18d8c862015-05-12 16:32:34 -0400803static void set_root(struct nameidata *nd)
Al Viro2a737872009-04-07 11:49:53 -0400804{
Al Viro7bd88372014-09-13 21:55:46 -0400805 struct fs_struct *fs = current->fs;
Nick Pigginc28cc362011-01-07 17:49:53 +1100806
Al Viro9e6697e2015-12-05 20:07:21 -0500807 if (nd->flags & LOOKUP_RCU) {
808 unsigned seq;
809
810 do {
811 seq = read_seqcount_begin(&fs->seq);
812 nd->root = fs->root;
813 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
814 } while (read_seqcount_retry(&fs->seq, seq));
815 } else {
816 get_fs_root(fs, &nd->root);
817 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100818}
819
Jan Blunck1d957f92008-02-14 19:34:35 -0800820static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700821{
822 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800823 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700824 mntput(path->mnt);
825}
826
Nick Piggin7b9337a2011-01-14 08:42:43 +0000827static inline void path_to_nameidata(const struct path *path,
828 struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700829{
Nick Piggin31e6b012011-01-07 17:49:52 +1100830 if (!(nd->flags & LOOKUP_RCU)) {
831 dput(nd->path.dentry);
832 if (nd->path.mnt != path->mnt)
833 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800834 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100835 nd->path.mnt = path->mnt;
Jan Blunck4ac91372008-02-14 19:34:32 -0800836 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700837}
838
Al Viro248fb5b2015-12-05 20:51:58 -0500839static int nd_jump_root(struct nameidata *nd)
840{
841 if (nd->flags & LOOKUP_RCU) {
842 struct dentry *d;
843 nd->path = nd->root;
844 d = nd->path.dentry;
845 nd->inode = d->d_inode;
846 nd->seq = nd->root_seq;
847 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
848 return -ECHILD;
849 } else {
850 path_put(&nd->path);
851 nd->path = nd->root;
852 path_get(&nd->path);
853 nd->inode = nd->path.dentry->d_inode;
854 }
855 nd->flags |= LOOKUP_JUMPED;
856 return 0;
857}
858
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400859/*
Al Viro6b255392015-11-17 10:20:54 -0500860 * Helper to directly jump to a known parsed path from ->get_link,
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400861 * caller must have taken a reference to path beforehand.
862 */
Al Viro6e771372015-05-02 13:37:52 -0400863void nd_jump_link(struct path *path)
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400864{
Al Viro6e771372015-05-02 13:37:52 -0400865 struct nameidata *nd = current->nameidata;
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400866 path_put(&nd->path);
867
868 nd->path = *path;
869 nd->inode = nd->path.dentry->d_inode;
870 nd->flags |= LOOKUP_JUMPED;
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400871}
872
Al Virob9ff4422015-05-02 20:19:23 -0400873static inline void put_link(struct nameidata *nd)
Al Viro574197e2011-03-14 22:20:34 -0400874{
Al Viro21c30032015-05-03 21:06:24 -0400875 struct saved *last = nd->stack + --nd->depth;
Al Virofceef392015-12-29 15:58:39 -0500876 do_delayed_call(&last->done);
Al Viro6548fae2015-05-07 20:32:22 -0400877 if (!(nd->flags & LOOKUP_RCU))
878 path_put(&last->link);
Al Viro574197e2011-03-14 22:20:34 -0400879}
880
Linus Torvalds561ec642012-10-26 10:05:07 -0700881int sysctl_protected_symlinks __read_mostly = 0;
882int sysctl_protected_hardlinks __read_mostly = 0;
Kees Cook800179c2012-07-25 17:29:07 -0700883
884/**
885 * may_follow_link - Check symlink following for unsafe situations
Randy Dunlap55852632012-08-18 17:39:25 -0700886 * @nd: nameidata pathwalk data
Kees Cook800179c2012-07-25 17:29:07 -0700887 *
888 * In the case of the sysctl_protected_symlinks sysctl being enabled,
889 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
890 * in a sticky world-writable directory. This is to protect privileged
891 * processes from failing races against path names that may change out
892 * from under them by way of other users creating malicious symlinks.
893 * It will permit symlinks to be followed only when outside a sticky
894 * world-writable directory, or when the uid of the symlink and follower
895 * match, or when the directory owner matches the symlink's owner.
896 *
897 * Returns 0 if following the symlink is allowed, -ve on error.
898 */
Al Virofec2fa22015-05-06 15:58:18 -0400899static inline int may_follow_link(struct nameidata *nd)
Kees Cook800179c2012-07-25 17:29:07 -0700900{
901 const struct inode *inode;
902 const struct inode *parent;
903
904 if (!sysctl_protected_symlinks)
905 return 0;
906
907 /* Allowed if owner and follower match. */
Al Virofceef392015-12-29 15:58:39 -0500908 inode = nd->link_inode;
Eric W. Biederman81abe272012-08-03 09:38:08 -0700909 if (uid_eq(current_cred()->fsuid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700910 return 0;
911
912 /* Allowed if parent directory not sticky and world-writable. */
Al Viroaa65fa32015-08-04 23:23:50 -0400913 parent = nd->inode;
Kees Cook800179c2012-07-25 17:29:07 -0700914 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
915 return 0;
916
917 /* Allowed if parent directory and link owner match. */
Eric W. Biederman81abe272012-08-03 09:38:08 -0700918 if (uid_eq(parent->i_uid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700919 return 0;
920
Al Viro319565022015-05-07 20:37:40 -0400921 if (nd->flags & LOOKUP_RCU)
922 return -ECHILD;
923
Al Viro1cf26652015-05-06 16:01:56 -0400924 audit_log_link_denied("follow_link", &nd->stack[0].link);
Kees Cook800179c2012-07-25 17:29:07 -0700925 return -EACCES;
926}
927
928/**
929 * safe_hardlink_source - Check for safe hardlink conditions
930 * @inode: the source inode to hardlink from
931 *
932 * Return false if at least one of the following conditions:
933 * - inode is not a regular file
934 * - inode is setuid
935 * - inode is setgid and group-exec
936 * - access failure for read and write
937 *
938 * Otherwise returns true.
939 */
940static bool safe_hardlink_source(struct inode *inode)
941{
942 umode_t mode = inode->i_mode;
943
944 /* Special files should not get pinned to the filesystem. */
945 if (!S_ISREG(mode))
946 return false;
947
948 /* Setuid files should not get pinned to the filesystem. */
949 if (mode & S_ISUID)
950 return false;
951
952 /* Executable setgid files should not get pinned to the filesystem. */
953 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
954 return false;
955
956 /* Hardlinking to unreadable or unwritable sources is dangerous. */
957 if (inode_permission(inode, MAY_READ | MAY_WRITE))
958 return false;
959
960 return true;
961}
962
963/**
964 * may_linkat - Check permissions for creating a hardlink
965 * @link: the source to hardlink from
966 *
967 * Block hardlink when all of:
968 * - sysctl_protected_hardlinks enabled
969 * - fsuid does not match inode
970 * - hardlink source is unsafe (see safe_hardlink_source() above)
Dirk Steinmetzf2ca3792015-10-20 16:09:19 +0200971 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
Kees Cook800179c2012-07-25 17:29:07 -0700972 *
973 * Returns 0 if successful, -ve on error.
974 */
975static int may_linkat(struct path *link)
976{
Kees Cook800179c2012-07-25 17:29:07 -0700977 struct inode *inode;
978
979 if (!sysctl_protected_hardlinks)
980 return 0;
981
Kees Cook800179c2012-07-25 17:29:07 -0700982 inode = link->dentry->d_inode;
983
984 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
985 * otherwise, it must be a safe source.
986 */
Dirk Steinmetzf2ca3792015-10-20 16:09:19 +0200987 if (inode_owner_or_capable(inode) || safe_hardlink_source(inode))
Kees Cook800179c2012-07-25 17:29:07 -0700988 return 0;
989
Kees Cooka51d9ea2012-07-25 17:29:08 -0700990 audit_log_link_denied("linkat", link);
Kees Cook800179c2012-07-25 17:29:07 -0700991 return -EPERM;
992}
993
Al Viro3b2e7f72015-04-19 00:53:50 -0400994static __always_inline
995const char *get_link(struct nameidata *nd)
Ian Kent051d3812006-03-27 01:14:53 -0800996{
Al Viroab104922015-05-10 11:50:01 -0400997 struct saved *last = nd->stack + nd->depth - 1;
Al Viro1cf26652015-05-06 16:01:56 -0400998 struct dentry *dentry = last->link.dentry;
Al Virofceef392015-12-29 15:58:39 -0500999 struct inode *inode = nd->link_inode;
Al Viro6d7b5aa2012-06-10 04:15:17 -04001000 int error;
Al Viro0a959df2015-04-18 18:23:41 -04001001 const char *res;
Ian Kent051d3812006-03-27 01:14:53 -08001002
NeilBrown8fa9dd22015-03-23 13:37:40 +11001003 if (!(nd->flags & LOOKUP_RCU)) {
1004 touch_atime(&last->link);
1005 cond_resched();
1006 } else if (atime_needs_update(&last->link, inode)) {
1007 if (unlikely(unlazy_walk(nd, NULL, 0)))
1008 return ERR_PTR(-ECHILD);
1009 touch_atime(&last->link);
1010 }
1011
NeilBrownbda0be72015-03-23 13:37:39 +11001012 error = security_inode_follow_link(dentry, inode,
1013 nd->flags & LOOKUP_RCU);
1014 if (unlikely(error))
Al Viro6920a442015-05-10 10:43:46 -04001015 return ERR_PTR(error);
Al Viro36f3b4f2011-02-22 21:24:38 -05001016
Al Viro86acdca12009-12-22 23:45:11 -05001017 nd->last_type = LAST_BIND;
Al Virod4dee482015-04-30 20:08:02 -04001018 res = inode->i_link;
1019 if (!res) {
Al Virofceef392015-12-29 15:58:39 -05001020 const char * (*get)(struct dentry *, struct inode *,
1021 struct delayed_call *);
1022 get = inode->i_op->get_link;
Al Viro8c1b4562015-05-09 18:15:21 -04001023 if (nd->flags & LOOKUP_RCU) {
Al Virofceef392015-12-29 15:58:39 -05001024 res = get(NULL, inode, &last->done);
Al Viro6b255392015-11-17 10:20:54 -05001025 if (res == ERR_PTR(-ECHILD)) {
1026 if (unlikely(unlazy_walk(nd, NULL, 0)))
1027 return ERR_PTR(-ECHILD);
Al Virofceef392015-12-29 15:58:39 -05001028 res = get(dentry, inode, &last->done);
Al Viro6b255392015-11-17 10:20:54 -05001029 }
1030 } else {
Al Virofceef392015-12-29 15:58:39 -05001031 res = get(dentry, inode, &last->done);
Al Viro8c1b4562015-05-09 18:15:21 -04001032 }
Al Virofceef392015-12-29 15:58:39 -05001033 if (IS_ERR_OR_NULL(res))
Al Virofab51e82015-05-10 11:01:00 -04001034 return res;
Ian Kent051d3812006-03-27 01:14:53 -08001035 }
Al Virofab51e82015-05-10 11:01:00 -04001036 if (*res == '/') {
Al Viro9e6697e2015-12-05 20:07:21 -05001037 if (!nd->root.mnt)
1038 set_root(nd);
Al Viro248fb5b2015-12-05 20:51:58 -05001039 if (unlikely(nd_jump_root(nd)))
1040 return ERR_PTR(-ECHILD);
Al Virofab51e82015-05-10 11:01:00 -04001041 while (unlikely(*++res == '/'))
1042 ;
1043 }
1044 if (!*res)
1045 res = NULL;
Al Viro0a959df2015-04-18 18:23:41 -04001046 return res;
1047}
Al Viro6d7b5aa2012-06-10 04:15:17 -04001048
David Howellsf015f1262012-06-25 12:55:28 +01001049/*
1050 * follow_up - Find the mountpoint of path's vfsmount
1051 *
1052 * Given a path, find the mountpoint of its source file system.
1053 * Replace @path with the path of the mountpoint in the parent mount.
1054 * Up is towards /.
1055 *
1056 * Return 1 if we went up a level and 0 if we were already at the
1057 * root.
1058 */
Al Virobab77eb2009-04-18 03:26:48 -04001059int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060{
Al Viro0714a532011-11-24 22:19:58 -05001061 struct mount *mnt = real_mount(path->mnt);
1062 struct mount *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +10001064
Al Viro48a066e2013-09-29 22:06:07 -04001065 read_seqlock_excl(&mount_lock);
Al Viro0714a532011-11-24 22:19:58 -05001066 parent = mnt->mnt_parent;
Al Viro3c0a6162012-07-18 17:32:50 +04001067 if (parent == mnt) {
Al Viro48a066e2013-09-29 22:06:07 -04001068 read_sequnlock_excl(&mount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 return 0;
1070 }
Al Viro0714a532011-11-24 22:19:58 -05001071 mntget(&parent->mnt);
Al Viroa73324d2011-11-24 22:25:07 -05001072 mountpoint = dget(mnt->mnt_mountpoint);
Al Viro48a066e2013-09-29 22:06:07 -04001073 read_sequnlock_excl(&mount_lock);
Al Virobab77eb2009-04-18 03:26:48 -04001074 dput(path->dentry);
1075 path->dentry = mountpoint;
1076 mntput(path->mnt);
Al Viro0714a532011-11-24 22:19:58 -05001077 path->mnt = &parent->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 return 1;
1079}
Al Viro4d359502014-03-14 12:20:17 -04001080EXPORT_SYMBOL(follow_up);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001082/*
David Howells9875cf82011-01-14 18:45:21 +00001083 * Perform an automount
1084 * - return -EISDIR to tell follow_managed() to stop and return the path we
1085 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 */
NeilBrown756daf22015-03-23 13:37:38 +11001087static int follow_automount(struct path *path, struct nameidata *nd,
David Howells9875cf82011-01-14 18:45:21 +00001088 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +11001089{
David Howells9875cf82011-01-14 18:45:21 +00001090 struct vfsmount *mnt;
David Howellsea5b7782011-01-14 19:10:03 +00001091 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001092
David Howells9875cf82011-01-14 18:45:21 +00001093 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1094 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -07001095
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +02001096 /* We don't want to mount if someone's just doing a stat -
1097 * unless they're stat'ing a directory and appended a '/' to
1098 * the name.
1099 *
1100 * We do, however, want to mount if someone wants to open or
1101 * create a file of any type under the mountpoint, wants to
1102 * traverse through the mountpoint or wants to open the
1103 * mounted directory. Also, autofs may mark negative dentries
1104 * as being automount points. These will need the attentions
1105 * of the daemon to instantiate them before they can be used.
David Howells9875cf82011-01-14 18:45:21 +00001106 */
NeilBrown756daf22015-03-23 13:37:38 +11001107 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1108 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +02001109 path->dentry->d_inode)
1110 return -EISDIR;
1111
NeilBrown756daf22015-03-23 13:37:38 +11001112 nd->total_link_count++;
1113 if (nd->total_link_count >= 40)
David Howells9875cf82011-01-14 18:45:21 +00001114 return -ELOOP;
1115
1116 mnt = path->dentry->d_op->d_automount(path);
1117 if (IS_ERR(mnt)) {
1118 /*
1119 * The filesystem is allowed to return -EISDIR here to indicate
1120 * it doesn't want to automount. For instance, autofs would do
1121 * this so that its userspace daemon can mount on this dentry.
1122 *
1123 * However, we can only permit this if it's a terminal point in
1124 * the path being looked up; if it wasn't then the remainder of
1125 * the path is inaccessible and we should say so.
1126 */
NeilBrown756daf22015-03-23 13:37:38 +11001127 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
David Howells9875cf82011-01-14 18:45:21 +00001128 return -EREMOTE;
1129 return PTR_ERR(mnt);
1130 }
David Howellsea5b7782011-01-14 19:10:03 +00001131
David Howells9875cf82011-01-14 18:45:21 +00001132 if (!mnt) /* mount collision */
1133 return 0;
1134
Al Viro8aef1882011-06-16 15:10:06 +01001135 if (!*need_mntput) {
1136 /* lock_mount() may release path->mnt on error */
1137 mntget(path->mnt);
1138 *need_mntput = true;
1139 }
Al Viro19a167a2011-01-17 01:35:23 -05001140 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +00001141
David Howellsea5b7782011-01-14 19:10:03 +00001142 switch (err) {
1143 case -EBUSY:
1144 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -05001145 return 0;
David Howellsea5b7782011-01-14 19:10:03 +00001146 case 0:
Al Viro8aef1882011-06-16 15:10:06 +01001147 path_put(path);
David Howellsea5b7782011-01-14 19:10:03 +00001148 path->mnt = mnt;
1149 path->dentry = dget(mnt->mnt_root);
David Howellsea5b7782011-01-14 19:10:03 +00001150 return 0;
Al Viro19a167a2011-01-17 01:35:23 -05001151 default:
1152 return err;
David Howellsea5b7782011-01-14 19:10:03 +00001153 }
Al Viro19a167a2011-01-17 01:35:23 -05001154
David Howells9875cf82011-01-14 18:45:21 +00001155}
1156
1157/*
1158 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +00001159 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +00001160 * - Flagged as mountpoint
1161 * - Flagged as automount point
1162 *
1163 * This may only be called in refwalk mode.
1164 *
1165 * Serialization is taken care of in namespace.c
1166 */
NeilBrown756daf22015-03-23 13:37:38 +11001167static int follow_managed(struct path *path, struct nameidata *nd)
David Howells9875cf82011-01-14 18:45:21 +00001168{
Al Viro8aef1882011-06-16 15:10:06 +01001169 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
David Howells9875cf82011-01-14 18:45:21 +00001170 unsigned managed;
1171 bool need_mntput = false;
Al Viro8aef1882011-06-16 15:10:06 +01001172 int ret = 0;
David Howells9875cf82011-01-14 18:45:21 +00001173
1174 /* Given that we're not holding a lock here, we retain the value in a
1175 * local variable for each dentry as we look at it so that we don't see
1176 * the components of that value change under us */
1177 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1178 managed &= DCACHE_MANAGED_DENTRY,
1179 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +00001180 /* Allow the filesystem to manage the transit without i_mutex
1181 * being held. */
1182 if (managed & DCACHE_MANAGE_TRANSIT) {
1183 BUG_ON(!path->dentry->d_op);
1184 BUG_ON(!path->dentry->d_op->d_manage);
Al Viro1aed3e42011-03-18 09:09:02 -04001185 ret = path->dentry->d_op->d_manage(path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001186 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001187 break;
David Howellscc53ce52011-01-14 18:45:26 +00001188 }
1189
David Howells9875cf82011-01-14 18:45:21 +00001190 /* Transit to a mounted filesystem. */
1191 if (managed & DCACHE_MOUNTED) {
1192 struct vfsmount *mounted = lookup_mnt(path);
1193 if (mounted) {
1194 dput(path->dentry);
1195 if (need_mntput)
1196 mntput(path->mnt);
1197 path->mnt = mounted;
1198 path->dentry = dget(mounted->mnt_root);
1199 need_mntput = true;
1200 continue;
1201 }
1202
1203 /* Something is mounted on this dentry in another
1204 * namespace and/or whatever was mounted there in this
Al Viro48a066e2013-09-29 22:06:07 -04001205 * namespace got unmounted before lookup_mnt() could
1206 * get it */
David Howells9875cf82011-01-14 18:45:21 +00001207 }
1208
1209 /* Handle an automount point */
1210 if (managed & DCACHE_NEED_AUTOMOUNT) {
NeilBrown756daf22015-03-23 13:37:38 +11001211 ret = follow_automount(path, nd, &need_mntput);
David Howells9875cf82011-01-14 18:45:21 +00001212 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001213 break;
David Howells9875cf82011-01-14 18:45:21 +00001214 continue;
1215 }
1216
1217 /* We didn't change the current path point */
1218 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 }
Al Viro8aef1882011-06-16 15:10:06 +01001220
1221 if (need_mntput && path->mnt == mnt)
1222 mntput(path->mnt);
1223 if (ret == -EISDIR)
1224 ret = 0;
Al Viro84027522015-04-22 10:30:08 -04001225 if (need_mntput)
1226 nd->flags |= LOOKUP_JUMPED;
1227 if (unlikely(ret < 0))
1228 path_put_conditional(path, nd);
1229 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230}
1231
David Howellscc53ce52011-01-14 18:45:26 +00001232int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233{
1234 struct vfsmount *mounted;
1235
Al Viro1c755af2009-04-18 14:06:57 -04001236 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -04001238 dput(path->dentry);
1239 mntput(path->mnt);
1240 path->mnt = mounted;
1241 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 return 1;
1243 }
1244 return 0;
1245}
Al Viro4d359502014-03-14 12:20:17 -04001246EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247
NeilBrownb8faf032014-08-04 17:06:29 +10001248static inline int managed_dentry_rcu(struct dentry *dentry)
Ian Kent62a73752011-03-25 01:51:02 +08001249{
NeilBrownb8faf032014-08-04 17:06:29 +10001250 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1251 dentry->d_op->d_manage(dentry, true) : 0;
Ian Kent62a73752011-03-25 01:51:02 +08001252}
1253
David Howells9875cf82011-01-14 18:45:21 +00001254/*
Al Viro287548e2011-05-27 06:50:06 -04001255 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1256 * we meet a managed dentry that would need blocking.
David Howells9875cf82011-01-14 18:45:21 +00001257 */
1258static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
Al Viro254cf582015-05-05 09:40:46 -04001259 struct inode **inode, unsigned *seqp)
David Howells9875cf82011-01-14 18:45:21 +00001260{
Ian Kent62a73752011-03-25 01:51:02 +08001261 for (;;) {
Al Viroc7105362011-11-24 18:22:03 -05001262 struct mount *mounted;
Ian Kent62a73752011-03-25 01:51:02 +08001263 /*
1264 * Don't forget we might have a non-mountpoint managed dentry
1265 * that wants to block transit.
1266 */
NeilBrownb8faf032014-08-04 17:06:29 +10001267 switch (managed_dentry_rcu(path->dentry)) {
1268 case -ECHILD:
1269 default:
David Howellsab909112011-01-14 18:46:51 +00001270 return false;
NeilBrownb8faf032014-08-04 17:06:29 +10001271 case -EISDIR:
1272 return true;
1273 case 0:
1274 break;
1275 }
Ian Kent62a73752011-03-25 01:51:02 +08001276
1277 if (!d_mountpoint(path->dentry))
NeilBrownb8faf032014-08-04 17:06:29 +10001278 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
Ian Kent62a73752011-03-25 01:51:02 +08001279
Al Viro474279d2013-10-01 16:11:26 -04001280 mounted = __lookup_mnt(path->mnt, path->dentry);
David Howells9875cf82011-01-14 18:45:21 +00001281 if (!mounted)
1282 break;
Al Viroc7105362011-11-24 18:22:03 -05001283 path->mnt = &mounted->mnt;
1284 path->dentry = mounted->mnt.mnt_root;
Al Viroa3fbbde2011-11-07 21:21:26 +00001285 nd->flags |= LOOKUP_JUMPED;
Al Viro254cf582015-05-05 09:40:46 -04001286 *seqp = read_seqcount_begin(&path->dentry->d_seq);
Linus Torvalds59430262011-07-18 15:43:29 -07001287 /*
1288 * Update the inode too. We don't need to re-check the
1289 * dentry sequence number here after this d_inode read,
1290 * because a mount-point is always pinned.
1291 */
1292 *inode = path->dentry->d_inode;
David Howells9875cf82011-01-14 18:45:21 +00001293 }
Al Virof5be3e29122014-09-13 21:50:45 -04001294 return !read_seqretry(&mount_lock, nd->m_seq) &&
NeilBrownb8faf032014-08-04 17:06:29 +10001295 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
Al Viro287548e2011-05-27 06:50:06 -04001296}
1297
Nick Piggin31e6b012011-01-07 17:49:52 +11001298static int follow_dotdot_rcu(struct nameidata *nd)
1299{
Al Viro4023bfc2014-09-13 21:59:43 -04001300 struct inode *inode = nd->inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001301
David Howells9875cf82011-01-14 18:45:21 +00001302 while (1) {
Al Viroaed434a2015-05-12 12:22:47 -04001303 if (path_equal(&nd->path, &nd->root))
Nick Piggin31e6b012011-01-07 17:49:52 +11001304 break;
Nick Piggin31e6b012011-01-07 17:49:52 +11001305 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1306 struct dentry *old = nd->path.dentry;
1307 struct dentry *parent = old->d_parent;
1308 unsigned seq;
1309
Al Viro4023bfc2014-09-13 21:59:43 -04001310 inode = parent->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001311 seq = read_seqcount_begin(&parent->d_seq);
Al Viroaed434a2015-05-12 12:22:47 -04001312 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1313 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001314 nd->path.dentry = parent;
1315 nd->seq = seq;
Eric W. Biederman397d4252015-08-15 20:27:13 -05001316 if (unlikely(!path_connected(&nd->path)))
1317 return -ENOENT;
Nick Piggin31e6b012011-01-07 17:49:52 +11001318 break;
Al Viroaed434a2015-05-12 12:22:47 -04001319 } else {
1320 struct mount *mnt = real_mount(nd->path.mnt);
1321 struct mount *mparent = mnt->mnt_parent;
1322 struct dentry *mountpoint = mnt->mnt_mountpoint;
1323 struct inode *inode2 = mountpoint->d_inode;
1324 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1325 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1326 return -ECHILD;
1327 if (&mparent->mnt == nd->path.mnt)
1328 break;
1329 /* we know that mountpoint was pinned */
1330 nd->path.dentry = mountpoint;
1331 nd->path.mnt = &mparent->mnt;
1332 inode = inode2;
1333 nd->seq = seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11001334 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001335 }
Al Viroaed434a2015-05-12 12:22:47 -04001336 while (unlikely(d_mountpoint(nd->path.dentry))) {
Al Virob37199e2014-03-20 15:18:22 -04001337 struct mount *mounted;
1338 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
Al Viroaed434a2015-05-12 12:22:47 -04001339 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1340 return -ECHILD;
Al Virob37199e2014-03-20 15:18:22 -04001341 if (!mounted)
1342 break;
1343 nd->path.mnt = &mounted->mnt;
1344 nd->path.dentry = mounted->mnt.mnt_root;
Al Viro4023bfc2014-09-13 21:59:43 -04001345 inode = nd->path.dentry->d_inode;
Al Virob37199e2014-03-20 15:18:22 -04001346 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Virob37199e2014-03-20 15:18:22 -04001347 }
Al Viro4023bfc2014-09-13 21:59:43 -04001348 nd->inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001349 return 0;
1350}
1351
David Howells9875cf82011-01-14 18:45:21 +00001352/*
David Howellscc53ce52011-01-14 18:45:26 +00001353 * Follow down to the covering mount currently visible to userspace. At each
1354 * point, the filesystem owning that dentry may be queried as to whether the
1355 * caller is permitted to proceed or not.
David Howellscc53ce52011-01-14 18:45:26 +00001356 */
Al Viro7cc90cc2011-03-18 09:04:20 -04001357int follow_down(struct path *path)
David Howellscc53ce52011-01-14 18:45:26 +00001358{
1359 unsigned managed;
1360 int ret;
1361
1362 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1363 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1364 /* Allow the filesystem to manage the transit without i_mutex
1365 * being held.
1366 *
1367 * We indicate to the filesystem if someone is trying to mount
1368 * something here. This gives autofs the chance to deny anyone
1369 * other than its daemon the right to mount on its
1370 * superstructure.
1371 *
1372 * The filesystem may sleep at this point.
1373 */
1374 if (managed & DCACHE_MANAGE_TRANSIT) {
1375 BUG_ON(!path->dentry->d_op);
1376 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +00001377 ret = path->dentry->d_op->d_manage(
Al Viro1aed3e42011-03-18 09:09:02 -04001378 path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001379 if (ret < 0)
1380 return ret == -EISDIR ? 0 : ret;
1381 }
1382
1383 /* Transit to a mounted filesystem. */
1384 if (managed & DCACHE_MOUNTED) {
1385 struct vfsmount *mounted = lookup_mnt(path);
1386 if (!mounted)
1387 break;
1388 dput(path->dentry);
1389 mntput(path->mnt);
1390 path->mnt = mounted;
1391 path->dentry = dget(mounted->mnt_root);
1392 continue;
1393 }
1394
1395 /* Don't handle automount points here */
1396 break;
1397 }
1398 return 0;
1399}
Al Viro4d359502014-03-14 12:20:17 -04001400EXPORT_SYMBOL(follow_down);
David Howellscc53ce52011-01-14 18:45:26 +00001401
1402/*
David Howells9875cf82011-01-14 18:45:21 +00001403 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1404 */
1405static void follow_mount(struct path *path)
1406{
1407 while (d_mountpoint(path->dentry)) {
1408 struct vfsmount *mounted = lookup_mnt(path);
1409 if (!mounted)
1410 break;
1411 dput(path->dentry);
1412 mntput(path->mnt);
1413 path->mnt = mounted;
1414 path->dentry = dget(mounted->mnt_root);
1415 }
1416}
1417
Eric W. Biederman397d4252015-08-15 20:27:13 -05001418static int follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419{
1420 while(1) {
Jan Blunck4ac91372008-02-14 19:34:32 -08001421 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422
Al Viro2a737872009-04-07 11:49:53 -04001423 if (nd->path.dentry == nd->root.dentry &&
1424 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 break;
1426 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001427 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Al Viro3088dd72010-01-30 15:47:29 -05001428 /* rare case of legitimate dget_parent()... */
1429 nd->path.dentry = dget_parent(nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 dput(old);
Eric W. Biederman397d4252015-08-15 20:27:13 -05001431 if (unlikely(!path_connected(&nd->path)))
1432 return -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 break;
1434 }
Al Viro3088dd72010-01-30 15:47:29 -05001435 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 }
Al Viro79ed0222009-04-18 13:59:41 -04001438 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001439 nd->inode = nd->path.dentry->d_inode;
Eric W. Biederman397d4252015-08-15 20:27:13 -05001440 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441}
1442
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443/*
Miklos Szeredibad61182012-03-26 12:54:24 +02001444 * This looks up the name in dcache, possibly revalidates the old dentry and
1445 * allocates a new one if not found or not valid. In the need_lookup argument
1446 * returns whether i_op->lookup is necessary.
1447 *
1448 * dir->d_inode->i_mutex must be held
Nick Pigginbaa03892010-08-18 04:37:31 +10001449 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001450static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
Al Viro201f9562012-06-22 12:42:10 +04001451 unsigned int flags, bool *need_lookup)
Nick Pigginbaa03892010-08-18 04:37:31 +10001452{
Nick Pigginbaa03892010-08-18 04:37:31 +10001453 struct dentry *dentry;
Miklos Szeredibad61182012-03-26 12:54:24 +02001454 int error;
Nick Pigginbaa03892010-08-18 04:37:31 +10001455
Miklos Szeredibad61182012-03-26 12:54:24 +02001456 *need_lookup = false;
1457 dentry = d_lookup(dir, name);
1458 if (dentry) {
Jeff Layton39e3c952012-11-28 11:30:53 -05001459 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
Al Viro201f9562012-06-22 12:42:10 +04001460 error = d_revalidate(dentry, flags);
Miklos Szeredibad61182012-03-26 12:54:24 +02001461 if (unlikely(error <= 0)) {
1462 if (error < 0) {
1463 dput(dentry);
1464 return ERR_PTR(error);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001465 } else {
1466 d_invalidate(dentry);
Miklos Szeredibad61182012-03-26 12:54:24 +02001467 dput(dentry);
1468 dentry = NULL;
1469 }
1470 }
1471 }
1472 }
Nick Pigginbaa03892010-08-18 04:37:31 +10001473
Miklos Szeredibad61182012-03-26 12:54:24 +02001474 if (!dentry) {
1475 dentry = d_alloc(dir, name);
1476 if (unlikely(!dentry))
1477 return ERR_PTR(-ENOMEM);
Nick Pigginbaa03892010-08-18 04:37:31 +10001478
Miklos Szeredibad61182012-03-26 12:54:24 +02001479 *need_lookup = true;
Nick Pigginbaa03892010-08-18 04:37:31 +10001480 }
1481 return dentry;
1482}
1483
1484/*
J. Bruce Fields13a2c3b2013-10-23 16:09:16 -04001485 * Call i_op->lookup on the dentry. The dentry must be negative and
1486 * unhashed.
Miklos Szeredibad61182012-03-26 12:54:24 +02001487 *
1488 * dir->d_inode->i_mutex must be held
Josef Bacik44396f42011-05-31 11:58:49 -04001489 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001490static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
Al Viro72bd8662012-06-10 17:17:17 -04001491 unsigned int flags)
Josef Bacik44396f42011-05-31 11:58:49 -04001492{
Josef Bacik44396f42011-05-31 11:58:49 -04001493 struct dentry *old;
1494
1495 /* Don't create child dentry for a dead directory. */
Miklos Szeredibad61182012-03-26 12:54:24 +02001496 if (unlikely(IS_DEADDIR(dir))) {
Miklos Szeredie188dc02012-02-03 14:25:18 +01001497 dput(dentry);
Josef Bacik44396f42011-05-31 11:58:49 -04001498 return ERR_PTR(-ENOENT);
Miklos Szeredie188dc02012-02-03 14:25:18 +01001499 }
Josef Bacik44396f42011-05-31 11:58:49 -04001500
Al Viro72bd8662012-06-10 17:17:17 -04001501 old = dir->i_op->lookup(dir, dentry, flags);
Josef Bacik44396f42011-05-31 11:58:49 -04001502 if (unlikely(old)) {
1503 dput(dentry);
1504 dentry = old;
1505 }
1506 return dentry;
1507}
1508
Al Viroa3255542012-03-30 14:41:51 -04001509static struct dentry *__lookup_hash(struct qstr *name,
Al Viro72bd8662012-06-10 17:17:17 -04001510 struct dentry *base, unsigned int flags)
Al Viroa3255542012-03-30 14:41:51 -04001511{
Miklos Szeredibad61182012-03-26 12:54:24 +02001512 bool need_lookup;
Al Viroa3255542012-03-30 14:41:51 -04001513 struct dentry *dentry;
1514
Al Viro72bd8662012-06-10 17:17:17 -04001515 dentry = lookup_dcache(name, base, flags, &need_lookup);
Miklos Szeredibad61182012-03-26 12:54:24 +02001516 if (!need_lookup)
1517 return dentry;
Al Viroa3255542012-03-30 14:41:51 -04001518
Al Viro72bd8662012-06-10 17:17:17 -04001519 return lookup_real(base->d_inode, dentry, flags);
Al Viroa3255542012-03-30 14:41:51 -04001520}
1521
Josef Bacik44396f42011-05-31 11:58:49 -04001522/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 * It's more convoluted than I'd like it to be, but... it's still fairly
1524 * small and for now I'd prefer to have fast path as straight as possible.
1525 * It _is_ time-critical.
1526 */
Al Viroe97cdc82013-01-24 18:16:00 -05001527static int lookup_fast(struct nameidata *nd,
Al Viro254cf582015-05-05 09:40:46 -04001528 struct path *path, struct inode **inode,
1529 unsigned *seqp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530{
Jan Blunck4ac91372008-02-14 19:34:32 -08001531 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001532 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001533 int need_reval = 1;
1534 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001535 int err;
1536
Al Viro3cac2602009-08-13 18:27:43 +04001537 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001538 * Rename seqlock is not required here because in the off chance
1539 * of a false negative due to a concurrent rename, we're going to
1540 * do the non-racy lookup, below.
1541 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001542 if (nd->flags & LOOKUP_RCU) {
1543 unsigned seq;
Al Viro766c4cb2015-05-07 19:24:57 -04001544 bool negative;
Linus Torvaldsda53be12013-05-21 15:22:44 -07001545 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
Al Viro5a18fff2011-03-11 04:44:53 -05001546 if (!dentry)
1547 goto unlazy;
1548
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001549 /*
1550 * This sequence count validates that the inode matches
1551 * the dentry name information from lookup.
1552 */
David Howells63afdfc2015-05-06 15:59:00 +01001553 *inode = d_backing_inode(dentry);
Al Viro766c4cb2015-05-07 19:24:57 -04001554 negative = d_is_negative(dentry);
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001555 if (read_seqcount_retry(&dentry->d_seq, seq))
1556 return -ECHILD;
1557
1558 /*
1559 * This sequence count validates that the parent had no
1560 * changes while we did the lookup of the dentry above.
1561 *
1562 * The memory barrier in read_seqcount_begin of child is
1563 * enough, we can use __read_seqcount_retry here.
1564 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001565 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1566 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001567
Al Viro254cf582015-05-05 09:40:46 -04001568 *seqp = seq;
Al Viro24643082011-02-15 01:26:22 -05001569 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
Al Viro4ce16ef32012-06-10 16:10:59 -04001570 status = d_revalidate(dentry, nd->flags);
Al Viro5a18fff2011-03-11 04:44:53 -05001571 if (unlikely(status <= 0)) {
1572 if (status != -ECHILD)
1573 need_reval = 0;
1574 goto unlazy;
1575 }
Al Viro24643082011-02-15 01:26:22 -05001576 }
Trond Myklebustdaf37612015-10-09 13:44:34 -04001577 /*
1578 * Note: do negative dentry check after revalidation in
1579 * case that drops it.
1580 */
1581 if (negative)
1582 return -ENOENT;
Nick Piggin31e6b012011-01-07 17:49:52 +11001583 path->mnt = mnt;
1584 path->dentry = dentry;
Al Viro254cf582015-05-05 09:40:46 -04001585 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
NeilBrownb8faf032014-08-04 17:06:29 +10001586 return 0;
Al Viro5a18fff2011-03-11 04:44:53 -05001587unlazy:
Al Viro254cf582015-05-05 09:40:46 -04001588 if (unlazy_walk(nd, dentry, seq))
Al Viro19660af2011-03-25 10:32:48 -04001589 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001590 } else {
Al Viroe97cdc82013-01-24 18:16:00 -05001591 dentry = __d_lookup(parent, &nd->last);
Nick Piggin31e6b012011-01-07 17:49:52 +11001592 }
Al Viro5a18fff2011-03-11 04:44:53 -05001593
Al Viro81e6f522012-03-30 14:48:04 -04001594 if (unlikely(!dentry))
1595 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001596
Al Viro5a18fff2011-03-11 04:44:53 -05001597 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
Al Viro4ce16ef32012-06-10 16:10:59 -04001598 status = d_revalidate(dentry, nd->flags);
Al Viro5a18fff2011-03-11 04:44:53 -05001599 if (unlikely(status <= 0)) {
1600 if (status < 0) {
1601 dput(dentry);
1602 return status;
1603 }
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001604 d_invalidate(dentry);
1605 dput(dentry);
1606 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001607 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001608
Al Viro766c4cb2015-05-07 19:24:57 -04001609 if (unlikely(d_is_negative(dentry))) {
1610 dput(dentry);
1611 return -ENOENT;
1612 }
David Howells9875cf82011-01-14 18:45:21 +00001613 path->mnt = mnt;
1614 path->dentry = dentry;
NeilBrown756daf22015-03-23 13:37:38 +11001615 err = follow_managed(path, nd);
Al Viro84027522015-04-22 10:30:08 -04001616 if (likely(!err))
David Howells63afdfc2015-05-06 15:59:00 +01001617 *inode = d_backing_inode(path->dentry);
Al Viro84027522015-04-22 10:30:08 -04001618 return err;
Al Viro81e6f522012-03-30 14:48:04 -04001619
1620need_lookup:
Miklos Szeredi697f5142012-05-21 17:30:05 +02001621 return 1;
1622}
1623
1624/* Fast lookup failed, do it the slow way */
Al Virocc2a5272013-01-24 18:19:49 -05001625static int lookup_slow(struct nameidata *nd, struct path *path)
Miklos Szeredi697f5142012-05-21 17:30:05 +02001626{
1627 struct dentry *dentry, *parent;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001628
1629 parent = nd->path.dentry;
Al Viro81e6f522012-03-30 14:48:04 -04001630 BUG_ON(nd->inode != parent->d_inode);
1631
Al Viro59551022016-01-22 15:40:57 -05001632 inode_lock(parent->d_inode);
Al Virocc2a5272013-01-24 18:19:49 -05001633 dentry = __lookup_hash(&nd->last, parent, nd->flags);
Al Viro59551022016-01-22 15:40:57 -05001634 inode_unlock(parent->d_inode);
Al Viro81e6f522012-03-30 14:48:04 -04001635 if (IS_ERR(dentry))
1636 return PTR_ERR(dentry);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001637 path->mnt = nd->path.mnt;
1638 path->dentry = dentry;
Al Viro84027522015-04-22 10:30:08 -04001639 return follow_managed(path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640}
1641
Al Viro52094c82011-02-21 21:34:47 -05001642static inline int may_lookup(struct nameidata *nd)
1643{
1644 if (nd->flags & LOOKUP_RCU) {
Al Viro4ad5abb2011-06-20 19:57:03 -04001645 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
Al Viro52094c82011-02-21 21:34:47 -05001646 if (err != -ECHILD)
1647 return err;
Al Viro6e9918b2015-05-05 09:26:05 -04001648 if (unlazy_walk(nd, NULL, 0))
Al Viro52094c82011-02-21 21:34:47 -05001649 return -ECHILD;
1650 }
Al Viro4ad5abb2011-06-20 19:57:03 -04001651 return inode_permission(nd->inode, MAY_EXEC);
Al Viro52094c82011-02-21 21:34:47 -05001652}
1653
Al Viro9856fa12011-03-04 14:22:06 -05001654static inline int handle_dots(struct nameidata *nd, int type)
1655{
1656 if (type == LAST_DOTDOT) {
Al Viro9e6697e2015-12-05 20:07:21 -05001657 if (!nd->root.mnt)
1658 set_root(nd);
Al Viro9856fa12011-03-04 14:22:06 -05001659 if (nd->flags & LOOKUP_RCU) {
Al Viro70291ae2015-05-04 07:53:00 -04001660 return follow_dotdot_rcu(nd);
Al Viro9856fa12011-03-04 14:22:06 -05001661 } else
Eric W. Biederman397d4252015-08-15 20:27:13 -05001662 return follow_dotdot(nd);
Al Viro9856fa12011-03-04 14:22:06 -05001663 }
1664 return 0;
1665}
1666
Al Viro181548c2015-05-07 19:54:34 -04001667static int pick_link(struct nameidata *nd, struct path *link,
1668 struct inode *inode, unsigned seq)
Al Virod63ff282015-05-04 18:13:23 -04001669{
Al Viro626de992015-05-04 18:26:59 -04001670 int error;
Al Viro1cf26652015-05-06 16:01:56 -04001671 struct saved *last;
NeilBrown756daf22015-03-23 13:37:38 +11001672 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
Al Viro626de992015-05-04 18:26:59 -04001673 path_to_nameidata(link, nd);
1674 return -ELOOP;
1675 }
Al Virobc40aee2015-05-09 13:04:24 -04001676 if (!(nd->flags & LOOKUP_RCU)) {
Al Viro79733872015-05-09 12:55:43 -04001677 if (link->mnt == nd->path.mnt)
1678 mntget(link->mnt);
Al Virod63ff282015-05-04 18:13:23 -04001679 }
Al Viro626de992015-05-04 18:26:59 -04001680 error = nd_alloc_stack(nd);
1681 if (unlikely(error)) {
Al Virobc40aee2015-05-09 13:04:24 -04001682 if (error == -ECHILD) {
1683 if (unlikely(unlazy_link(nd, link, seq)))
1684 return -ECHILD;
1685 error = nd_alloc_stack(nd);
1686 }
1687 if (error) {
1688 path_put(link);
1689 return error;
1690 }
Al Viro626de992015-05-04 18:26:59 -04001691 }
1692
Al Viroab104922015-05-10 11:50:01 -04001693 last = nd->stack + nd->depth++;
Al Viro1cf26652015-05-06 16:01:56 -04001694 last->link = *link;
Al Virofceef392015-12-29 15:58:39 -05001695 clear_delayed_call(&last->done);
1696 nd->link_inode = inode;
Al Viro0450b2d2015-05-08 13:23:53 -04001697 last->seq = seq;
Al Virod63ff282015-05-04 18:13:23 -04001698 return 1;
1699}
1700
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001701/*
1702 * Do we need to follow links? We _really_ want to be able
1703 * to do this check without having to look at inode->i_op,
1704 * so we keep a cache of "no, this doesn't need follow_link"
1705 * for the common case.
1706 */
Al Viro254cf582015-05-05 09:40:46 -04001707static inline int should_follow_link(struct nameidata *nd, struct path *link,
Al Viro181548c2015-05-07 19:54:34 -04001708 int follow,
1709 struct inode *inode, unsigned seq)
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001710{
Al Virod63ff282015-05-04 18:13:23 -04001711 if (likely(!d_is_symlink(link->dentry)))
1712 return 0;
1713 if (!follow)
1714 return 0;
Al Viro181548c2015-05-07 19:54:34 -04001715 return pick_link(nd, link, inode, seq);
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001716}
1717
Al Viro4693a542015-05-04 17:47:11 -04001718enum {WALK_GET = 1, WALK_PUT = 2};
1719
1720static int walk_component(struct nameidata *nd, int flags)
Al Viroce57dfc2011-03-13 19:58:58 -04001721{
Al Virocaa85632015-04-22 17:52:47 -04001722 struct path path;
Al Viroce57dfc2011-03-13 19:58:58 -04001723 struct inode *inode;
Al Viro254cf582015-05-05 09:40:46 -04001724 unsigned seq;
Al Viroce57dfc2011-03-13 19:58:58 -04001725 int err;
1726 /*
1727 * "." and ".." are special - ".." especially so because it has
1728 * to be able to know about the current root directory and
1729 * parent relationships.
1730 */
Al Viro4693a542015-05-04 17:47:11 -04001731 if (unlikely(nd->last_type != LAST_NORM)) {
1732 err = handle_dots(nd, nd->last_type);
1733 if (flags & WALK_PUT)
1734 put_link(nd);
1735 return err;
1736 }
Al Viro254cf582015-05-05 09:40:46 -04001737 err = lookup_fast(nd, &path, &inode, &seq);
Al Viroce57dfc2011-03-13 19:58:58 -04001738 if (unlikely(err)) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001739 if (err < 0)
Al Virof0a9ba72015-05-04 07:59:30 -04001740 return err;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001741
Al Virocaa85632015-04-22 17:52:47 -04001742 err = lookup_slow(nd, &path);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001743 if (err < 0)
Al Virof0a9ba72015-05-04 07:59:30 -04001744 return err;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001745
David Howells63afdfc2015-05-06 15:59:00 +01001746 inode = d_backing_inode(path.dentry);
Al Viro254cf582015-05-05 09:40:46 -04001747 seq = 0; /* we are already out of RCU mode */
Al Viro766c4cb2015-05-07 19:24:57 -04001748 err = -ENOENT;
Al Virocaa85632015-04-22 17:52:47 -04001749 if (d_is_negative(path.dentry))
Al Viro766c4cb2015-05-07 19:24:57 -04001750 goto out_path_put;
Al Viroce57dfc2011-03-13 19:58:58 -04001751 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001752
Al Viro4693a542015-05-04 17:47:11 -04001753 if (flags & WALK_PUT)
1754 put_link(nd);
Al Viro181548c2015-05-07 19:54:34 -04001755 err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq);
Al Virod63ff282015-05-04 18:13:23 -04001756 if (unlikely(err))
1757 return err;
Al Virocaa85632015-04-22 17:52:47 -04001758 path_to_nameidata(&path, nd);
Al Viroce57dfc2011-03-13 19:58:58 -04001759 nd->inode = inode;
Al Viro254cf582015-05-05 09:40:46 -04001760 nd->seq = seq;
Al Viroce57dfc2011-03-13 19:58:58 -04001761 return 0;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001762
1763out_path_put:
Al Virocaa85632015-04-22 17:52:47 -04001764 path_to_nameidata(&path, nd);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001765 return err;
Al Viroce57dfc2011-03-13 19:58:58 -04001766}
1767
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768/*
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001769 * We can do the critical dentry name comparison and hashing
1770 * operations one word at a time, but we are limited to:
1771 *
1772 * - Architectures with fast unaligned word accesses. We could
1773 * do a "get_unaligned()" if this helps and is sufficiently
1774 * fast.
1775 *
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001776 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1777 * do not trap on the (extremely unlikely) case of a page
1778 * crossing operation.
1779 *
1780 * - Furthermore, we need an efficient 64-bit compile for the
1781 * 64-bit case in order to generate the "number of bytes in
1782 * the final mask". Again, that could be replaced with a
1783 * efficient population count instruction or similar.
1784 */
1785#ifdef CONFIG_DCACHE_WORD_ACCESS
1786
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001787#include <asm/word-at-a-time.h>
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001788
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001789#ifdef CONFIG_64BIT
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001790
1791static inline unsigned int fold_hash(unsigned long hash)
1792{
Linus Torvalds99d263d2014-09-13 11:30:10 -07001793 return hash_64(hash, 32);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001794}
1795
1796#else /* 32-bit case */
1797
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001798#define fold_hash(x) (x)
1799
1800#endif
1801
1802unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1803{
1804 unsigned long a, mask;
1805 unsigned long hash = 0;
1806
1807 for (;;) {
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001808 a = load_unaligned_zeropad(name);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001809 if (len < sizeof(unsigned long))
1810 break;
1811 hash += a;
Al Virof132c5b2012-03-22 21:59:52 +00001812 hash *= 9;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001813 name += sizeof(unsigned long);
1814 len -= sizeof(unsigned long);
1815 if (!len)
1816 goto done;
1817 }
Will Deacona5c21dc2013-12-12 17:40:21 +00001818 mask = bytemask_from_count(len);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001819 hash += mask & a;
1820done:
1821 return fold_hash(hash);
1822}
1823EXPORT_SYMBOL(full_name_hash);
1824
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001825/*
1826 * Calculate the length and hash of the path component, and
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001827 * return the "hash_len" as the result.
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001828 */
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001829static inline u64 hash_name(const char *name)
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001830{
Linus Torvalds36126f82012-05-26 10:43:17 -07001831 unsigned long a, b, adata, bdata, mask, hash, len;
1832 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001833
1834 hash = a = 0;
1835 len = -sizeof(unsigned long);
1836 do {
1837 hash = (hash + a) * 9;
1838 len += sizeof(unsigned long);
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001839 a = load_unaligned_zeropad(name+len);
Linus Torvalds36126f82012-05-26 10:43:17 -07001840 b = a ^ REPEAT_BYTE('/');
1841 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001842
Linus Torvalds36126f82012-05-26 10:43:17 -07001843 adata = prep_zero_mask(a, adata, &constants);
1844 bdata = prep_zero_mask(b, bdata, &constants);
1845
1846 mask = create_zero_mask(adata | bdata);
1847
1848 hash += a & zero_bytemask(mask);
Linus Torvalds9226b5b2014-09-14 17:28:32 -07001849 len += find_zero(mask);
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001850 return hashlen_create(fold_hash(hash), len);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001851}
1852
1853#else
1854
Linus Torvalds0145acc2012-03-02 14:32:59 -08001855unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1856{
1857 unsigned long hash = init_name_hash();
1858 while (len--)
1859 hash = partial_name_hash(*name++, hash);
1860 return end_name_hash(hash);
1861}
Linus Torvaldsae942ae2012-03-02 19:40:57 -08001862EXPORT_SYMBOL(full_name_hash);
Linus Torvalds0145acc2012-03-02 14:32:59 -08001863
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001864/*
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001865 * We know there's a real path component here of at least
1866 * one character.
1867 */
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001868static inline u64 hash_name(const char *name)
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001869{
1870 unsigned long hash = init_name_hash();
1871 unsigned long len = 0, c;
1872
1873 c = (unsigned char)*name;
1874 do {
1875 len++;
1876 hash = partial_name_hash(c, hash);
1877 c = (unsigned char)name[len];
1878 } while (c && c != '/');
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001879 return hashlen_create(end_name_hash(hash), len);
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001880}
1881
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001882#endif
1883
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001884/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01001886 * This is the basic name resolution function, turning a pathname into
1887 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01001889 * Returns 0 and nd will have valid dentry and mnt on success.
1890 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891 */
Al Viro6de88d72009-08-09 01:41:57 +04001892static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 int err;
Al Viro32cd7462015-04-18 20:30:49 -04001895
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896 while (*name=='/')
1897 name++;
1898 if (!*name)
Al Viro9e18f102015-04-18 20:44:34 -04001899 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901 /* At this point we know we have a real path component. */
1902 for(;;) {
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001903 u64 hash_len;
Al Virofe479a52011-02-22 15:10:03 -05001904 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905
Al Viro52094c82011-02-21 21:34:47 -05001906 err = may_lookup(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907 if (err)
Al Viro3595e232015-05-09 16:54:45 -04001908 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001910 hash_len = hash_name(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911
Al Virofe479a52011-02-22 15:10:03 -05001912 type = LAST_NORM;
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001913 if (name[0] == '.') switch (hashlen_len(hash_len)) {
Al Virofe479a52011-02-22 15:10:03 -05001914 case 2:
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001915 if (name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05001916 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05001917 nd->flags |= LOOKUP_JUMPED;
1918 }
Al Virofe479a52011-02-22 15:10:03 -05001919 break;
1920 case 1:
1921 type = LAST_DOT;
1922 }
Al Viro5a202bc2011-03-08 14:17:44 -05001923 if (likely(type == LAST_NORM)) {
1924 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05001925 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05001926 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
James Hogana060dc52014-09-16 13:07:35 +01001927 struct qstr this = { { .hash_len = hash_len }, .name = name };
Linus Torvaldsda53be12013-05-21 15:22:44 -07001928 err = parent->d_op->d_hash(parent, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05001929 if (err < 0)
Al Viro3595e232015-05-09 16:54:45 -04001930 return err;
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001931 hash_len = this.hash_len;
1932 name = this.name;
Al Viro5a202bc2011-03-08 14:17:44 -05001933 }
1934 }
Al Virofe479a52011-02-22 15:10:03 -05001935
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001936 nd->last.hash_len = hash_len;
1937 nd->last.name = name;
Al Viro5f4a6a62013-01-24 18:04:22 -05001938 nd->last_type = type;
1939
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001940 name += hashlen_len(hash_len);
1941 if (!*name)
Al Virobdf6cbf2015-04-18 20:21:40 -04001942 goto OK;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001943 /*
1944 * If it wasn't NUL, we know it was '/'. Skip that
1945 * slash, and continue until no more slashes.
1946 */
1947 do {
Linus Torvaldsd6bb3e92014-09-15 10:51:07 -07001948 name++;
1949 } while (unlikely(*name == '/'));
Al Viro8620c232015-05-04 08:58:35 -04001950 if (unlikely(!*name)) {
1951OK:
Al Viro368ee9b2015-05-08 17:19:59 -04001952 /* pathname body, done */
Al Viro8620c232015-05-04 08:58:35 -04001953 if (!nd->depth)
1954 return 0;
1955 name = nd->stack[nd->depth - 1].name;
Al Viro368ee9b2015-05-08 17:19:59 -04001956 /* trailing symlink, done */
Al Viro8620c232015-05-04 08:58:35 -04001957 if (!name)
1958 return 0;
1959 /* last component of nested symlink */
Al Viro4693a542015-05-04 17:47:11 -04001960 err = walk_component(nd, WALK_GET | WALK_PUT);
Al Viro8620c232015-05-04 08:58:35 -04001961 } else {
Al Viro4693a542015-05-04 17:47:11 -04001962 err = walk_component(nd, WALK_GET);
Al Viro8620c232015-05-04 08:58:35 -04001963 }
Al Viroce57dfc2011-03-13 19:58:58 -04001964 if (err < 0)
Al Viro3595e232015-05-09 16:54:45 -04001965 return err;
Al Virofe479a52011-02-22 15:10:03 -05001966
Al Viroce57dfc2011-03-13 19:58:58 -04001967 if (err) {
Al Viro626de992015-05-04 18:26:59 -04001968 const char *s = get_link(nd);
Al Viro5a460272015-04-17 23:44:45 -04001969
Viresh Kumara1c83682015-08-12 15:59:44 +05301970 if (IS_ERR(s))
Al Viro3595e232015-05-09 16:54:45 -04001971 return PTR_ERR(s);
Al Virod40bcc02015-04-18 20:03:03 -04001972 err = 0;
1973 if (unlikely(!s)) {
1974 /* jumped */
Al Virob9ff4422015-05-02 20:19:23 -04001975 put_link(nd);
Al Virod40bcc02015-04-18 20:03:03 -04001976 } else {
Al Virofab51e82015-05-10 11:01:00 -04001977 nd->stack[nd->depth - 1].name = name;
1978 name = s;
1979 continue;
Al Virod40bcc02015-04-18 20:03:03 -04001980 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001981 }
Al Viro97242f92015-08-01 19:59:28 -04001982 if (unlikely(!d_can_lookup(nd->path.dentry))) {
1983 if (nd->flags & LOOKUP_RCU) {
1984 if (unlazy_walk(nd, NULL, 0))
1985 return -ECHILD;
1986 }
Al Viro3595e232015-05-09 16:54:45 -04001987 return -ENOTDIR;
Al Viro97242f92015-08-01 19:59:28 -04001988 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990}
1991
Al Viroc8a53ee2015-05-12 18:43:07 -04001992static const char *path_init(struct nameidata *nd, unsigned flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001994 int retval = 0;
Al Viroc8a53ee2015-05-12 18:43:07 -04001995 const char *s = nd->name->name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996
1997 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro980f3ea2014-11-20 14:20:24 -05001998 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999 nd->depth = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05002000 if (flags & LOOKUP_ROOT) {
David Howellsb18825a2013-09-12 19:22:53 +01002001 struct dentry *root = nd->root.dentry;
2002 struct inode *inode = root->d_inode;
Al Virofd2f7cb2015-02-22 20:07:13 -05002003 if (*s) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002004 if (!d_can_lookup(root))
Al Viro368ee9b2015-05-08 17:19:59 -04002005 return ERR_PTR(-ENOTDIR);
Al Viro73d049a2011-03-11 12:08:24 -05002006 retval = inode_permission(inode, MAY_EXEC);
2007 if (retval)
Al Viro368ee9b2015-05-08 17:19:59 -04002008 return ERR_PTR(retval);
Al Viro73d049a2011-03-11 12:08:24 -05002009 }
Al Viro5b6ca022011-03-09 23:04:47 -05002010 nd->path = nd->root;
2011 nd->inode = inode;
2012 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05002013 rcu_read_lock();
Al Viro5b6ca022011-03-09 23:04:47 -05002014 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro8f47a0162015-05-09 19:02:01 -04002015 nd->root_seq = nd->seq;
Al Viro48a066e2013-09-29 22:06:07 -04002016 nd->m_seq = read_seqbegin(&mount_lock);
Al Viro5b6ca022011-03-09 23:04:47 -05002017 } else {
2018 path_get(&nd->path);
2019 }
Al Viro368ee9b2015-05-08 17:19:59 -04002020 return s;
Al Viro5b6ca022011-03-09 23:04:47 -05002021 }
2022
Al Viro2a737872009-04-07 11:49:53 -04002023 nd->root.mnt = NULL;
Al Viro248fb5b2015-12-05 20:51:58 -05002024 nd->path.mnt = NULL;
2025 nd->path.dentry = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026
Al Viro48a066e2013-09-29 22:06:07 -04002027 nd->m_seq = read_seqbegin(&mount_lock);
Al Virofd2f7cb2015-02-22 20:07:13 -05002028 if (*s == '/') {
Al Viro9e6697e2015-12-05 20:07:21 -05002029 if (flags & LOOKUP_RCU)
Al Viro8b61e742013-11-08 12:45:01 -05002030 rcu_read_lock();
Al Viro9e6697e2015-12-05 20:07:21 -05002031 set_root(nd);
Al Viro248fb5b2015-12-05 20:51:58 -05002032 if (likely(!nd_jump_root(nd)))
Al Viroef55d912015-12-05 20:25:06 -05002033 return s;
Al Viro248fb5b2015-12-05 20:51:58 -05002034 nd->root.mnt = NULL;
Al Viroef55d912015-12-05 20:25:06 -05002035 rcu_read_unlock();
2036 return ERR_PTR(-ECHILD);
Al Viroc8a53ee2015-05-12 18:43:07 -04002037 } else if (nd->dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05002038 if (flags & LOOKUP_RCU) {
2039 struct fs_struct *fs = current->fs;
2040 unsigned seq;
2041
Al Viro8b61e742013-11-08 12:45:01 -05002042 rcu_read_lock();
Al Viroe41f7d42011-02-22 14:02:58 -05002043
2044 do {
2045 seq = read_seqcount_begin(&fs->seq);
2046 nd->path = fs->pwd;
Al Viroef55d912015-12-05 20:25:06 -05002047 nd->inode = nd->path.dentry->d_inode;
Al Viroe41f7d42011-02-22 14:02:58 -05002048 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2049 } while (read_seqcount_retry(&fs->seq, seq));
2050 } else {
2051 get_fs_pwd(current->fs, &nd->path);
Al Viroef55d912015-12-05 20:25:06 -05002052 nd->inode = nd->path.dentry->d_inode;
Al Viroe41f7d42011-02-22 14:02:58 -05002053 }
Al Viroef55d912015-12-05 20:25:06 -05002054 return s;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002055 } else {
Jeff Layton582aa642012-12-11 08:56:16 -05002056 /* Caller must check execute permissions on the starting path component */
Al Viroc8a53ee2015-05-12 18:43:07 -04002057 struct fd f = fdget_raw(nd->dfd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002058 struct dentry *dentry;
2059
Al Viro2903ff02012-08-28 12:52:22 -04002060 if (!f.file)
Al Viro368ee9b2015-05-08 17:19:59 -04002061 return ERR_PTR(-EBADF);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002062
Al Viro2903ff02012-08-28 12:52:22 -04002063 dentry = f.file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002064
Al Virofd2f7cb2015-02-22 20:07:13 -05002065 if (*s) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002066 if (!d_can_lookup(dentry)) {
Al Viro2903ff02012-08-28 12:52:22 -04002067 fdput(f);
Al Viro368ee9b2015-05-08 17:19:59 -04002068 return ERR_PTR(-ENOTDIR);
Al Viro2903ff02012-08-28 12:52:22 -04002069 }
Al Virof52e0c12011-03-14 18:56:51 -04002070 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002071
Al Viro2903ff02012-08-28 12:52:22 -04002072 nd->path = f.file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05002073 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05002074 rcu_read_lock();
Al Viro34a26b92015-05-11 08:05:05 -04002075 nd->inode = nd->path.dentry->d_inode;
2076 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viroe41f7d42011-02-22 14:02:58 -05002077 } else {
Al Viro2903ff02012-08-28 12:52:22 -04002078 path_get(&nd->path);
Al Viro34a26b92015-05-11 08:05:05 -04002079 nd->inode = nd->path.dentry->d_inode;
Al Viroe41f7d42011-02-22 14:02:58 -05002080 }
Al Viro34a26b92015-05-11 08:05:05 -04002081 fdput(f);
Al Viro368ee9b2015-05-08 17:19:59 -04002082 return s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083 }
Al Viro9b4a9b12009-04-07 11:44:16 -04002084}
2085
Al Viro3bdba282015-05-08 17:37:07 -04002086static const char *trailing_symlink(struct nameidata *nd)
Al Viro95fa25d2015-04-22 13:46:57 -04002087{
2088 const char *s;
Al Virofec2fa22015-05-06 15:58:18 -04002089 int error = may_follow_link(nd);
Al Virodeb106c2015-05-08 18:05:21 -04002090 if (unlikely(error))
Al Viro3bdba282015-05-08 17:37:07 -04002091 return ERR_PTR(error);
Al Viro95fa25d2015-04-22 13:46:57 -04002092 nd->flags |= LOOKUP_PARENT;
Al Virofab51e82015-05-10 11:01:00 -04002093 nd->stack[0].name = NULL;
Al Viro3b2e7f72015-04-19 00:53:50 -04002094 s = get_link(nd);
Al Virodeb106c2015-05-08 18:05:21 -04002095 return s ? s : "";
Al Viro95fa25d2015-04-22 13:46:57 -04002096}
2097
Al Virocaa85632015-04-22 17:52:47 -04002098static inline int lookup_last(struct nameidata *nd)
Al Virobd92d7f2011-03-14 19:54:59 -04002099{
2100 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2101 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2102
2103 nd->flags &= ~LOOKUP_PARENT;
Al Virodeb106c2015-05-08 18:05:21 -04002104 return walk_component(nd,
Al Viro4693a542015-05-04 17:47:11 -04002105 nd->flags & LOOKUP_FOLLOW
2106 ? nd->depth
2107 ? WALK_PUT | WALK_GET
2108 : WALK_GET
2109 : 0);
Al Virobd92d7f2011-03-14 19:54:59 -04002110}
2111
Al Viro9b4a9b12009-04-07 11:44:16 -04002112/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroc8a53ee2015-05-12 18:43:07 -04002113static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
Al Viro9b4a9b12009-04-07 11:44:16 -04002114{
Al Viroc8a53ee2015-05-12 18:43:07 -04002115 const char *s = path_init(nd, flags);
Al Virobd92d7f2011-03-14 19:54:59 -04002116 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11002117
Al Viro368ee9b2015-05-08 17:19:59 -04002118 if (IS_ERR(s))
2119 return PTR_ERR(s);
Al Viro3bdba282015-05-08 17:37:07 -04002120 while (!(err = link_path_walk(s, nd))
2121 && ((err = lookup_last(nd)) > 0)) {
2122 s = trailing_symlink(nd);
2123 if (IS_ERR(s)) {
2124 err = PTR_ERR(s);
2125 break;
Al Virobd92d7f2011-03-14 19:54:59 -04002126 }
2127 }
Al Viro9f1fafe2011-03-25 11:00:12 -04002128 if (!err)
2129 err = complete_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04002130
Al Virodeb106c2015-05-08 18:05:21 -04002131 if (!err && nd->flags & LOOKUP_DIRECTORY)
2132 if (!d_can_lookup(nd->path.dentry))
Al Virobd23a532011-03-23 09:56:30 -04002133 err = -ENOTDIR;
Al Viro625b6d12015-05-12 16:36:12 -04002134 if (!err) {
2135 *path = nd->path;
2136 nd->path.mnt = NULL;
2137 nd->path.dentry = NULL;
2138 }
2139 terminate_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04002140 return err;
Al Viroee0827c2011-02-21 23:38:09 -05002141}
Nick Piggin31e6b012011-01-07 17:49:52 +11002142
Al Viro625b6d12015-05-12 16:36:12 -04002143static int filename_lookup(int dfd, struct filename *name, unsigned flags,
Al Viro9ad1aaa2015-05-12 16:44:39 -04002144 struct path *path, struct path *root)
Jeff Layton873f1ee2012-10-10 15:25:29 -04002145{
Al Viro894bc8c2015-05-02 07:16:16 -04002146 int retval;
Al Viro9883d182015-05-13 07:28:08 -04002147 struct nameidata nd;
Al Viroabc9f5b2015-05-12 16:53:42 -04002148 if (IS_ERR(name))
2149 return PTR_ERR(name);
Al Viro9ad1aaa2015-05-12 16:44:39 -04002150 if (unlikely(root)) {
2151 nd.root = *root;
2152 flags |= LOOKUP_ROOT;
2153 }
Al Viro9883d182015-05-13 07:28:08 -04002154 set_nameidata(&nd, dfd, name);
Al Viroc8a53ee2015-05-12 18:43:07 -04002155 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002156 if (unlikely(retval == -ECHILD))
Al Viroc8a53ee2015-05-12 18:43:07 -04002157 retval = path_lookupat(&nd, flags, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002158 if (unlikely(retval == -ESTALE))
Al Viroc8a53ee2015-05-12 18:43:07 -04002159 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002160
2161 if (likely(!retval))
Al Viro625b6d12015-05-12 16:36:12 -04002162 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
Al Viro9883d182015-05-13 07:28:08 -04002163 restore_nameidata();
Al Viroe4bd1c12015-05-12 16:40:39 -04002164 putname(name);
Jeff Layton873f1ee2012-10-10 15:25:29 -04002165 return retval;
2166}
2167
Al Viro8bcb77f2015-05-08 16:59:20 -04002168/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroc8a53ee2015-05-12 18:43:07 -04002169static int path_parentat(struct nameidata *nd, unsigned flags,
Al Viro391172c2015-05-09 11:19:16 -04002170 struct path *parent)
Al Viro8bcb77f2015-05-08 16:59:20 -04002171{
Al Viroc8a53ee2015-05-12 18:43:07 -04002172 const char *s = path_init(nd, flags);
Al Viro368ee9b2015-05-08 17:19:59 -04002173 int err;
2174 if (IS_ERR(s))
2175 return PTR_ERR(s);
2176 err = link_path_walk(s, nd);
Al Viro8bcb77f2015-05-08 16:59:20 -04002177 if (!err)
2178 err = complete_walk(nd);
Al Viro391172c2015-05-09 11:19:16 -04002179 if (!err) {
2180 *parent = nd->path;
2181 nd->path.mnt = NULL;
2182 nd->path.dentry = NULL;
2183 }
2184 terminate_walk(nd);
Al Viro8bcb77f2015-05-08 16:59:20 -04002185 return err;
2186}
2187
Al Viro5c31b6c2015-05-12 17:32:54 -04002188static struct filename *filename_parentat(int dfd, struct filename *name,
Al Viro391172c2015-05-09 11:19:16 -04002189 unsigned int flags, struct path *parent,
2190 struct qstr *last, int *type)
Al Viro8bcb77f2015-05-08 16:59:20 -04002191{
2192 int retval;
Al Viro9883d182015-05-13 07:28:08 -04002193 struct nameidata nd;
Al Viro8bcb77f2015-05-08 16:59:20 -04002194
Al Viro5c31b6c2015-05-12 17:32:54 -04002195 if (IS_ERR(name))
2196 return name;
Al Viro9883d182015-05-13 07:28:08 -04002197 set_nameidata(&nd, dfd, name);
Al Viroc8a53ee2015-05-12 18:43:07 -04002198 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
Al Viro8bcb77f2015-05-08 16:59:20 -04002199 if (unlikely(retval == -ECHILD))
Al Viroc8a53ee2015-05-12 18:43:07 -04002200 retval = path_parentat(&nd, flags, parent);
Al Viro8bcb77f2015-05-08 16:59:20 -04002201 if (unlikely(retval == -ESTALE))
Al Viroc8a53ee2015-05-12 18:43:07 -04002202 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
Al Viro391172c2015-05-09 11:19:16 -04002203 if (likely(!retval)) {
2204 *last = nd.last;
2205 *type = nd.last_type;
2206 audit_inode(name, parent->dentry, LOOKUP_PARENT);
Al Viro5c31b6c2015-05-12 17:32:54 -04002207 } else {
2208 putname(name);
2209 name = ERR_PTR(retval);
Al Viro391172c2015-05-09 11:19:16 -04002210 }
Al Viro9883d182015-05-13 07:28:08 -04002211 restore_nameidata();
Al Viro5c31b6c2015-05-12 17:32:54 -04002212 return name;
Al Viro8bcb77f2015-05-08 16:59:20 -04002213}
2214
Al Viro79714f72012-06-15 03:01:42 +04002215/* does lookup, returns the object with parent locked */
2216struct dentry *kern_path_locked(const char *name, struct path *path)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002217{
Al Viro5c31b6c2015-05-12 17:32:54 -04002218 struct filename *filename;
2219 struct dentry *d;
Al Viro391172c2015-05-09 11:19:16 -04002220 struct qstr last;
2221 int type;
Paul Moore51689102015-01-22 00:00:03 -05002222
Al Viro5c31b6c2015-05-12 17:32:54 -04002223 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2224 &last, &type);
Paul Moore51689102015-01-22 00:00:03 -05002225 if (IS_ERR(filename))
2226 return ERR_CAST(filename);
Al Viro5c31b6c2015-05-12 17:32:54 -04002227 if (unlikely(type != LAST_NORM)) {
Al Viro391172c2015-05-09 11:19:16 -04002228 path_put(path);
Al Viro5c31b6c2015-05-12 17:32:54 -04002229 putname(filename);
2230 return ERR_PTR(-EINVAL);
Al Viro79714f72012-06-15 03:01:42 +04002231 }
Al Viro59551022016-01-22 15:40:57 -05002232 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
Al Viro391172c2015-05-09 11:19:16 -04002233 d = __lookup_hash(&last, path->dentry, 0);
Al Viro79714f72012-06-15 03:01:42 +04002234 if (IS_ERR(d)) {
Al Viro59551022016-01-22 15:40:57 -05002235 inode_unlock(path->dentry->d_inode);
Al Viro391172c2015-05-09 11:19:16 -04002236 path_put(path);
Al Viro79714f72012-06-15 03:01:42 +04002237 }
Paul Moore51689102015-01-22 00:00:03 -05002238 putname(filename);
Al Viro79714f72012-06-15 03:01:42 +04002239 return d;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002240}
2241
Al Virod1811462008-08-02 00:49:18 -04002242int kern_path(const char *name, unsigned int flags, struct path *path)
2243{
Al Viroabc9f5b2015-05-12 16:53:42 -04002244 return filename_lookup(AT_FDCWD, getname_kernel(name),
2245 flags, path, NULL);
Al Virod1811462008-08-02 00:49:18 -04002246}
Al Viro4d359502014-03-14 12:20:17 -04002247EXPORT_SYMBOL(kern_path);
Al Virod1811462008-08-02 00:49:18 -04002248
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002249/**
2250 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2251 * @dentry: pointer to dentry of the base directory
2252 * @mnt: pointer to vfs mount of the base directory
2253 * @name: pointer to file name
2254 * @flags: lookup flags
Al Viroe0a01242011-06-27 17:00:37 -04002255 * @path: pointer to struct path to fill
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002256 */
2257int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2258 const char *name, unsigned int flags,
Al Viroe0a01242011-06-27 17:00:37 -04002259 struct path *path)
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002260{
Al Viro9ad1aaa2015-05-12 16:44:39 -04002261 struct path root = {.mnt = mnt, .dentry = dentry};
Al Viro9ad1aaa2015-05-12 16:44:39 -04002262 /* the first argument of filename_lookup() is ignored with root */
Al Viroabc9f5b2015-05-12 16:53:42 -04002263 return filename_lookup(AT_FDCWD, getname_kernel(name),
2264 flags , path, &root);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002265}
Al Viro4d359502014-03-14 12:20:17 -04002266EXPORT_SYMBOL(vfs_path_lookup);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002267
Christoph Hellwigeead1912007-10-16 23:25:38 -07002268/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07002269 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07002270 * @name: pathname component to lookup
2271 * @base: base directory to lookup from
2272 * @len: maximum length @len should be interpreted to
2273 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07002274 * Note that this routine is purely a helper for filesystem usage and should
Al Viro9e7543e2015-02-23 02:49:48 -05002275 * not be called by generic code.
NeilBrownbbddca82016-01-07 16:08:20 -05002276 *
2277 * The caller must hold base->i_mutex.
Christoph Hellwigeead1912007-10-16 23:25:38 -07002278 */
James Morris057f6c02007-04-26 00:12:05 -07002279struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2280{
James Morris057f6c02007-04-26 00:12:05 -07002281 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05002282 unsigned int c;
Miklos Szeredicda309d2012-03-26 12:54:21 +02002283 int err;
James Morris057f6c02007-04-26 00:12:05 -07002284
Al Viro59551022016-01-22 15:40:57 -05002285 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
David Woodhouse2f9092e2009-04-20 23:18:37 +01002286
Al Viro6a96ba52011-03-07 23:49:20 -05002287 this.name = name;
2288 this.len = len;
Linus Torvalds0145acc2012-03-02 14:32:59 -08002289 this.hash = full_name_hash(name, len);
Al Viro6a96ba52011-03-07 23:49:20 -05002290 if (!len)
2291 return ERR_PTR(-EACCES);
2292
Al Viro21d8a152012-11-29 22:17:21 -05002293 if (unlikely(name[0] == '.')) {
2294 if (len < 2 || (len == 2 && name[1] == '.'))
2295 return ERR_PTR(-EACCES);
2296 }
2297
Al Viro6a96ba52011-03-07 23:49:20 -05002298 while (len--) {
2299 c = *(const unsigned char *)name++;
2300 if (c == '/' || c == '\0')
2301 return ERR_PTR(-EACCES);
Al Viro6a96ba52011-03-07 23:49:20 -05002302 }
Al Viro5a202bc2011-03-08 14:17:44 -05002303 /*
2304 * See if the low-level filesystem might want
2305 * to use its own hash..
2306 */
2307 if (base->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002308 int err = base->d_op->d_hash(base, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05002309 if (err < 0)
2310 return ERR_PTR(err);
2311 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07002312
Miklos Szeredicda309d2012-03-26 12:54:21 +02002313 err = inode_permission(base->d_inode, MAY_EXEC);
2314 if (err)
2315 return ERR_PTR(err);
2316
Al Viro72bd8662012-06-10 17:17:17 -04002317 return __lookup_hash(&this, base, 0);
James Morris057f6c02007-04-26 00:12:05 -07002318}
Al Viro4d359502014-03-14 12:20:17 -04002319EXPORT_SYMBOL(lookup_one_len);
James Morris057f6c02007-04-26 00:12:05 -07002320
NeilBrownbbddca82016-01-07 16:08:20 -05002321/**
2322 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2323 * @name: pathname component to lookup
2324 * @base: base directory to lookup from
2325 * @len: maximum length @len should be interpreted to
2326 *
2327 * Note that this routine is purely a helper for filesystem usage and should
2328 * not be called by generic code.
2329 *
2330 * Unlike lookup_one_len, it should be called without the parent
2331 * i_mutex held, and will take the i_mutex itself if necessary.
2332 */
2333struct dentry *lookup_one_len_unlocked(const char *name,
2334 struct dentry *base, int len)
2335{
2336 struct qstr this;
2337 unsigned int c;
2338 int err;
2339 struct dentry *ret;
2340
2341 this.name = name;
2342 this.len = len;
2343 this.hash = full_name_hash(name, len);
2344 if (!len)
2345 return ERR_PTR(-EACCES);
2346
2347 if (unlikely(name[0] == '.')) {
2348 if (len < 2 || (len == 2 && name[1] == '.'))
2349 return ERR_PTR(-EACCES);
2350 }
2351
2352 while (len--) {
2353 c = *(const unsigned char *)name++;
2354 if (c == '/' || c == '\0')
2355 return ERR_PTR(-EACCES);
2356 }
2357 /*
2358 * See if the low-level filesystem might want
2359 * to use its own hash..
2360 */
2361 if (base->d_flags & DCACHE_OP_HASH) {
2362 int err = base->d_op->d_hash(base, &this);
2363 if (err < 0)
2364 return ERR_PTR(err);
2365 }
2366
2367 err = inode_permission(base->d_inode, MAY_EXEC);
2368 if (err)
2369 return ERR_PTR(err);
2370
2371 /*
2372 * __d_lookup() is used to try to get a quick answer and avoid the
2373 * mutex. A false-negative does no harm.
2374 */
2375 ret = __d_lookup(base, &this);
2376 if (ret && unlikely(ret->d_flags & DCACHE_OP_REVALIDATE)) {
2377 dput(ret);
2378 ret = NULL;
2379 }
2380 if (ret)
2381 return ret;
2382
Al Viro59551022016-01-22 15:40:57 -05002383 inode_lock(base->d_inode);
NeilBrownbbddca82016-01-07 16:08:20 -05002384 ret = __lookup_hash(&this, base, 0);
Al Viro59551022016-01-22 15:40:57 -05002385 inode_unlock(base->d_inode);
NeilBrownbbddca82016-01-07 16:08:20 -05002386 return ret;
2387}
2388EXPORT_SYMBOL(lookup_one_len_unlocked);
2389
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002390int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2391 struct path *path, int *empty)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392{
Al Viroabc9f5b2015-05-12 16:53:42 -04002393 return filename_lookup(dfd, getname_flags(name, flags, empty),
2394 flags, path, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395}
Al Virob853a162015-05-13 09:12:02 -04002396EXPORT_SYMBOL(user_path_at_empty);
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002397
Jeff Layton873f1ee2012-10-10 15:25:29 -04002398/*
2399 * NB: most callers don't do anything directly with the reference to the
2400 * to struct filename, but the nd->last pointer points into the name string
2401 * allocated by getname. So we must hold the reference to it until all
2402 * path-walking is complete.
2403 */
Al Viroa2ec4a22015-05-13 06:57:49 -04002404static inline struct filename *
Al Virof5beed72015-04-30 16:09:11 -04002405user_path_parent(int dfd, const char __user *path,
2406 struct path *parent,
2407 struct qstr *last,
2408 int *type,
Jeff Layton9e790bd2012-12-11 12:10:09 -05002409 unsigned int flags)
Al Viro2ad94ae2008-07-21 09:32:51 -04002410{
Jeff Layton9e790bd2012-12-11 12:10:09 -05002411 /* only LOOKUP_REVAL is allowed in extra flags */
Al Viro5c31b6c2015-05-12 17:32:54 -04002412 return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL,
2413 parent, last, type);
Al Viro2ad94ae2008-07-21 09:32:51 -04002414}
2415
Jeff Layton80334262013-07-26 06:23:25 -04002416/**
Al Viro197df042013-09-08 14:03:27 -04002417 * mountpoint_last - look up last component for umount
Jeff Layton80334262013-07-26 06:23:25 -04002418 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2419 * @path: pointer to container for result
2420 *
2421 * This is a special lookup_last function just for umount. In this case, we
2422 * need to resolve the path without doing any revalidation.
2423 *
2424 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2425 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2426 * in almost all cases, this lookup will be served out of the dcache. The only
2427 * cases where it won't are if nd->last refers to a symlink or the path is
2428 * bogus and it doesn't exist.
2429 *
2430 * Returns:
2431 * -error: if there was an error during lookup. This includes -ENOENT if the
2432 * lookup found a negative dentry. The nd->path reference will also be
2433 * put in this case.
2434 *
2435 * 0: if we successfully resolved nd->path and found it to not to be a
2436 * symlink that needs to be followed. "path" will also be populated.
2437 * The nd->path reference will also be put.
2438 *
2439 * 1: if we successfully resolved nd->last and found it to be a symlink
2440 * that needs to be followed. "path" will be populated with the path
2441 * to the link, and nd->path will *not* be put.
2442 */
2443static int
Al Viro197df042013-09-08 14:03:27 -04002444mountpoint_last(struct nameidata *nd, struct path *path)
Jeff Layton80334262013-07-26 06:23:25 -04002445{
2446 int error = 0;
2447 struct dentry *dentry;
2448 struct dentry *dir = nd->path.dentry;
2449
Al Viro35759522013-09-08 13:41:33 -04002450 /* If we're in rcuwalk, drop out of it to handle last component */
2451 if (nd->flags & LOOKUP_RCU) {
Al Viro6e9918b2015-05-05 09:26:05 -04002452 if (unlazy_walk(nd, NULL, 0))
Al Virodeb106c2015-05-08 18:05:21 -04002453 return -ECHILD;
Jeff Layton80334262013-07-26 06:23:25 -04002454 }
2455
2456 nd->flags &= ~LOOKUP_PARENT;
2457
2458 if (unlikely(nd->last_type != LAST_NORM)) {
2459 error = handle_dots(nd, nd->last_type);
Al Viro35759522013-09-08 13:41:33 -04002460 if (error)
Al Virodeb106c2015-05-08 18:05:21 -04002461 return error;
Al Viro35759522013-09-08 13:41:33 -04002462 dentry = dget(nd->path.dentry);
2463 goto done;
Jeff Layton80334262013-07-26 06:23:25 -04002464 }
2465
Al Viro59551022016-01-22 15:40:57 -05002466 inode_lock(dir->d_inode);
Jeff Layton80334262013-07-26 06:23:25 -04002467 dentry = d_lookup(dir, &nd->last);
2468 if (!dentry) {
2469 /*
2470 * No cached dentry. Mounted dentries are pinned in the cache,
2471 * so that means that this dentry is probably a symlink or the
2472 * path doesn't actually point to a mounted dentry.
2473 */
2474 dentry = d_alloc(dir, &nd->last);
2475 if (!dentry) {
Al Viro59551022016-01-22 15:40:57 -05002476 inode_unlock(dir->d_inode);
Al Virodeb106c2015-05-08 18:05:21 -04002477 return -ENOMEM;
Jeff Layton80334262013-07-26 06:23:25 -04002478 }
Al Viro35759522013-09-08 13:41:33 -04002479 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
Dave Jonesbcceeeb2013-09-10 17:04:25 -04002480 if (IS_ERR(dentry)) {
Al Viro59551022016-01-22 15:40:57 -05002481 inode_unlock(dir->d_inode);
Al Virodeb106c2015-05-08 18:05:21 -04002482 return PTR_ERR(dentry);
Dave Jonesbcceeeb2013-09-10 17:04:25 -04002483 }
Jeff Layton80334262013-07-26 06:23:25 -04002484 }
Al Viro59551022016-01-22 15:40:57 -05002485 inode_unlock(dir->d_inode);
Jeff Layton80334262013-07-26 06:23:25 -04002486
Al Viro35759522013-09-08 13:41:33 -04002487done:
David Howells698934d2015-03-17 17:33:52 +00002488 if (d_is_negative(dentry)) {
Al Viro35759522013-09-08 13:41:33 -04002489 dput(dentry);
Al Virodeb106c2015-05-08 18:05:21 -04002490 return -ENOENT;
Jeff Layton80334262013-07-26 06:23:25 -04002491 }
Al Viro191d7f72015-05-04 08:26:45 -04002492 if (nd->depth)
2493 put_link(nd);
Al Viro35759522013-09-08 13:41:33 -04002494 path->dentry = dentry;
Vasily Averin295dc392014-07-21 12:30:23 +04002495 path->mnt = nd->path.mnt;
Al Viro181548c2015-05-07 19:54:34 -04002496 error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW,
2497 d_backing_inode(dentry), 0);
Al Virodeb106c2015-05-08 18:05:21 -04002498 if (unlikely(error))
Al Virod63ff282015-05-04 18:13:23 -04002499 return error;
Vasily Averin295dc392014-07-21 12:30:23 +04002500 mntget(path->mnt);
Al Viro35759522013-09-08 13:41:33 -04002501 follow_mount(path);
Al Virodeb106c2015-05-08 18:05:21 -04002502 return 0;
Jeff Layton80334262013-07-26 06:23:25 -04002503}
2504
2505/**
Al Viro197df042013-09-08 14:03:27 -04002506 * path_mountpoint - look up a path to be umounted
Masanari Iida2a78b852015-09-09 15:39:23 -07002507 * @nd: lookup context
Jeff Layton80334262013-07-26 06:23:25 -04002508 * @flags: lookup flags
Al Viroc8a53ee2015-05-12 18:43:07 -04002509 * @path: pointer to container for result
Jeff Layton80334262013-07-26 06:23:25 -04002510 *
2511 * Look up the given name, but don't attempt to revalidate the last component.
Randy Dunlap606d6fe2013-10-19 14:56:55 -07002512 * Returns 0 and "path" will be valid on success; Returns error otherwise.
Jeff Layton80334262013-07-26 06:23:25 -04002513 */
2514static int
Al Viroc8a53ee2015-05-12 18:43:07 -04002515path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
Jeff Layton80334262013-07-26 06:23:25 -04002516{
Al Viroc8a53ee2015-05-12 18:43:07 -04002517 const char *s = path_init(nd, flags);
Al Viro368ee9b2015-05-08 17:19:59 -04002518 int err;
2519 if (IS_ERR(s))
2520 return PTR_ERR(s);
Al Viro3bdba282015-05-08 17:37:07 -04002521 while (!(err = link_path_walk(s, nd)) &&
2522 (err = mountpoint_last(nd, path)) > 0) {
2523 s = trailing_symlink(nd);
2524 if (IS_ERR(s)) {
2525 err = PTR_ERR(s);
Jeff Layton80334262013-07-26 06:23:25 -04002526 break;
Al Viro3bdba282015-05-08 17:37:07 -04002527 }
Jeff Layton80334262013-07-26 06:23:25 -04002528 }
Al Virodeb106c2015-05-08 18:05:21 -04002529 terminate_walk(nd);
Jeff Layton80334262013-07-26 06:23:25 -04002530 return err;
2531}
2532
Al Viro2d864652013-09-08 20:18:44 -04002533static int
Al Viro668696d2015-02-22 19:44:00 -05002534filename_mountpoint(int dfd, struct filename *name, struct path *path,
Al Viro2d864652013-09-08 20:18:44 -04002535 unsigned int flags)
2536{
Al Viro9883d182015-05-13 07:28:08 -04002537 struct nameidata nd;
Al Virocbaab2d2015-01-22 02:49:00 -05002538 int error;
Al Viro668696d2015-02-22 19:44:00 -05002539 if (IS_ERR(name))
2540 return PTR_ERR(name);
Al Viro9883d182015-05-13 07:28:08 -04002541 set_nameidata(&nd, dfd, name);
Al Viroc8a53ee2015-05-12 18:43:07 -04002542 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
Al Viro2d864652013-09-08 20:18:44 -04002543 if (unlikely(error == -ECHILD))
Al Viroc8a53ee2015-05-12 18:43:07 -04002544 error = path_mountpoint(&nd, flags, path);
Al Viro2d864652013-09-08 20:18:44 -04002545 if (unlikely(error == -ESTALE))
Al Viroc8a53ee2015-05-12 18:43:07 -04002546 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
Al Viro2d864652013-09-08 20:18:44 -04002547 if (likely(!error))
Al Viro668696d2015-02-22 19:44:00 -05002548 audit_inode(name, path->dentry, 0);
Al Viro9883d182015-05-13 07:28:08 -04002549 restore_nameidata();
Al Viro668696d2015-02-22 19:44:00 -05002550 putname(name);
Al Viro2d864652013-09-08 20:18:44 -04002551 return error;
2552}
2553
Jeff Layton80334262013-07-26 06:23:25 -04002554/**
Al Viro197df042013-09-08 14:03:27 -04002555 * user_path_mountpoint_at - lookup a path from userland in order to umount it
Jeff Layton80334262013-07-26 06:23:25 -04002556 * @dfd: directory file descriptor
2557 * @name: pathname from userland
2558 * @flags: lookup flags
2559 * @path: pointer to container to hold result
2560 *
2561 * A umount is a special case for path walking. We're not actually interested
2562 * in the inode in this situation, and ESTALE errors can be a problem. We
2563 * simply want track down the dentry and vfsmount attached at the mountpoint
2564 * and avoid revalidating the last component.
2565 *
2566 * Returns 0 and populates "path" on success.
2567 */
2568int
Al Viro197df042013-09-08 14:03:27 -04002569user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
Jeff Layton80334262013-07-26 06:23:25 -04002570 struct path *path)
2571{
Al Virocbaab2d2015-01-22 02:49:00 -05002572 return filename_mountpoint(dfd, getname(name), path, flags);
Jeff Layton80334262013-07-26 06:23:25 -04002573}
2574
Al Viro2d864652013-09-08 20:18:44 -04002575int
2576kern_path_mountpoint(int dfd, const char *name, struct path *path,
2577 unsigned int flags)
2578{
Al Virocbaab2d2015-01-22 02:49:00 -05002579 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
Al Viro2d864652013-09-08 20:18:44 -04002580}
2581EXPORT_SYMBOL(kern_path_mountpoint);
2582
Miklos Szeredicbdf35b2014-10-24 00:14:36 +02002583int __check_sticky(struct inode *dir, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584{
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002585 kuid_t fsuid = current_fsuid();
David Howellsda9592e2008-11-14 10:39:05 +11002586
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002587 if (uid_eq(inode->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588 return 0;
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002589 if (uid_eq(dir->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590 return 0;
Andy Lutomirski23adbe12014-06-10 12:45:42 -07002591 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592}
Miklos Szeredicbdf35b2014-10-24 00:14:36 +02002593EXPORT_SYMBOL(__check_sticky);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594
2595/*
2596 * Check whether we can remove a link victim from directory dir, check
2597 * whether the type of victim is right.
2598 * 1. We can't do it if dir is read-only (done in permission())
2599 * 2. We should have write and exec permissions on dir
2600 * 3. We can't remove anything from append-only dir
2601 * 4. We can't do anything with immutable dir (done in permission())
2602 * 5. If the sticky bit on dir is set we should either
2603 * a. be owner of dir, or
2604 * b. be owner of victim, or
2605 * c. have CAP_FOWNER capability
2606 * 6. If the victim is append-only or immutable we can't do antyhing with
2607 * links pointing to it.
2608 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2609 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2610 * 9. We can't remove a root or mountpoint.
2611 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2612 * nfs_async_unlink().
2613 */
David Howellsb18825a2013-09-12 19:22:53 +01002614static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615{
David Howells63afdfc2015-05-06 15:59:00 +01002616 struct inode *inode = d_backing_inode(victim);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617 int error;
2618
David Howellsb18825a2013-09-12 19:22:53 +01002619 if (d_is_negative(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620 return -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01002621 BUG_ON(!inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622
2623 BUG_ON(victim->d_parent->d_inode != dir);
Jeff Layton4fa6b5e2012-10-10 15:25:25 -04002624 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625
Al Virof419a2e2008-07-22 00:07:17 -04002626 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 if (error)
2628 return error;
2629 if (IS_APPEND(dir))
2630 return -EPERM;
David Howellsb18825a2013-09-12 19:22:53 +01002631
2632 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2633 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634 return -EPERM;
2635 if (isdir) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002636 if (!d_is_dir(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002637 return -ENOTDIR;
2638 if (IS_ROOT(victim))
2639 return -EBUSY;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002640 } else if (d_is_dir(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641 return -EISDIR;
2642 if (IS_DEADDIR(dir))
2643 return -ENOENT;
2644 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2645 return -EBUSY;
2646 return 0;
2647}
2648
2649/* Check whether we can create an object with dentry child in directory
2650 * dir.
2651 * 1. We can't do it if child already exists (open has special treatment for
2652 * this case, but since we are inlined it's OK)
2653 * 2. We can't do it if dir is read-only (done in permission())
2654 * 3. We should have write and exec permissions on dir
2655 * 4. We can't do it if dir is immutable (done in permission())
2656 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02002657static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658{
Jeff Layton14e972b2013-05-08 10:25:58 -04002659 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660 if (child->d_inode)
2661 return -EEXIST;
2662 if (IS_DEADDIR(dir))
2663 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04002664 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665}
2666
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667/*
2668 * p1 and p2 should be directories on the same fs.
2669 */
2670struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2671{
2672 struct dentry *p;
2673
2674 if (p1 == p2) {
Al Viro59551022016-01-22 15:40:57 -05002675 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676 return NULL;
2677 }
2678
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002679 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002681 p = d_ancestor(p2, p1);
2682 if (p) {
Al Viro59551022016-01-22 15:40:57 -05002683 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2684 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002685 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686 }
2687
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002688 p = d_ancestor(p1, p2);
2689 if (p) {
Al Viro59551022016-01-22 15:40:57 -05002690 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2691 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002692 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693 }
2694
Al Viro59551022016-01-22 15:40:57 -05002695 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2696 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697 return NULL;
2698}
Al Viro4d359502014-03-14 12:20:17 -04002699EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700
2701void unlock_rename(struct dentry *p1, struct dentry *p2)
2702{
Al Viro59551022016-01-22 15:40:57 -05002703 inode_unlock(p1->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002704 if (p1 != p2) {
Al Viro59551022016-01-22 15:40:57 -05002705 inode_unlock(p2->d_inode);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002706 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707 }
2708}
Al Viro4d359502014-03-14 12:20:17 -04002709EXPORT_SYMBOL(unlock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710
Al Viro4acdaf22011-07-26 01:42:34 -04002711int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viro312b63f2012-06-10 18:09:36 -04002712 bool want_excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002714 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715 if (error)
2716 return error;
2717
Al Viroacfa4382008-12-04 10:06:33 -05002718 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002719 return -EACCES; /* shouldn't it be ENOSYS? */
2720 mode &= S_IALLUGO;
2721 mode |= S_IFREG;
2722 error = security_inode_create(dir, dentry, mode);
2723 if (error)
2724 return error;
Al Viro312b63f2012-06-10 18:09:36 -04002725 error = dir->i_op->create(dir, dentry, mode, want_excl);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002726 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002727 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728 return error;
2729}
Al Viro4d359502014-03-14 12:20:17 -04002730EXPORT_SYMBOL(vfs_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002731
Al Viro73d049a2011-03-11 12:08:24 -05002732static int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002734 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735 struct inode *inode = dentry->d_inode;
2736 int error;
2737
2738 if (!inode)
2739 return -ENOENT;
2740
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002741 switch (inode->i_mode & S_IFMT) {
2742 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002743 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002744 case S_IFDIR:
2745 if (acc_mode & MAY_WRITE)
2746 return -EISDIR;
2747 break;
2748 case S_IFBLK:
2749 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002750 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002751 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002752 /*FALLTHRU*/
2753 case S_IFIFO:
2754 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002755 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002756 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002757 }
Dave Hansenb41572e2007-10-16 23:31:14 -07002758
Al Viro62fb4a12015-12-26 22:33:24 -05002759 error = inode_permission(inode, MAY_OPEN | acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07002760 if (error)
2761 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05002762
Linus Torvalds1da177e2005-04-16 15:20:36 -07002763 /*
2764 * An append-only file must be opened in append mode for writing.
2765 */
2766 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05002767 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05002768 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002769 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05002770 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002771 }
2772
2773 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002774 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05002775 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776
J. Bruce Fieldsf3c7691e2011-09-21 10:58:13 -04002777 return 0;
Al Viro7715b522009-12-16 03:54:00 -05002778}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779
Jeff Laytone1181ee2010-12-07 16:19:50 -05002780static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002781{
Jeff Laytone1181ee2010-12-07 16:19:50 -05002782 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002783 struct inode *inode = path->dentry->d_inode;
2784 int error = get_write_access(inode);
2785 if (error)
2786 return error;
2787 /*
2788 * Refuse to truncate files with mandatory locks held on them.
2789 */
Jeff Laytond7a06982014-03-10 09:54:15 -04002790 error = locks_verify_locked(filp);
Al Viro7715b522009-12-16 03:54:00 -05002791 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002792 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002793 if (!error) {
2794 error = do_truncate(path->dentry, 0,
2795 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002796 filp);
Al Viro7715b522009-12-16 03:54:00 -05002797 }
2798 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002799 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800}
2801
Dave Hansend57999e2008-02-15 14:37:27 -08002802static inline int open_to_namei_flags(int flag)
2803{
Al Viro8a5e9292011-06-25 19:15:54 -04002804 if ((flag & O_ACCMODE) == 3)
2805 flag--;
Dave Hansend57999e2008-02-15 14:37:27 -08002806 return flag;
2807}
2808
Miklos Szeredid18e9002012-06-05 15:10:17 +02002809static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2810{
2811 int error = security_path_mknod(dir, dentry, mode, 0);
2812 if (error)
2813 return error;
2814
2815 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2816 if (error)
2817 return error;
2818
2819 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2820}
2821
David Howells1acf0af2012-06-14 16:13:46 +01002822/*
2823 * Attempt to atomically look up, create and open a file from a negative
2824 * dentry.
2825 *
2826 * Returns 0 if successful. The file will have been created and attached to
2827 * @file by the filesystem calling finish_open().
2828 *
2829 * Returns 1 if the file was looked up only or didn't need creating. The
2830 * caller will need to perform the open themselves. @path will have been
2831 * updated to point to the new dentry. This may be negative.
2832 *
2833 * Returns an error code otherwise.
2834 */
Al Viro2675a4e2012-06-22 12:41:10 +04002835static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2836 struct path *path, struct file *file,
2837 const struct open_flags *op,
Al Viro64894cf2012-07-31 00:53:35 +04002838 bool got_write, bool need_lookup,
Al Viro2675a4e2012-06-22 12:41:10 +04002839 int *opened)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002840{
2841 struct inode *dir = nd->path.dentry->d_inode;
2842 unsigned open_flag = open_to_namei_flags(op->open_flag);
2843 umode_t mode;
2844 int error;
2845 int acc_mode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002846 int create_error = 0;
2847 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
Miklos Szeredi116cc022013-09-16 14:52:05 +02002848 bool excl;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002849
2850 BUG_ON(dentry->d_inode);
2851
2852 /* Don't create child dentry for a dead directory. */
2853 if (unlikely(IS_DEADDIR(dir))) {
Al Viro2675a4e2012-06-22 12:41:10 +04002854 error = -ENOENT;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002855 goto out;
2856 }
2857
Miklos Szeredi62b259d2012-08-15 13:01:24 +02002858 mode = op->mode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002859 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2860 mode &= ~current_umask();
2861
Miklos Szeredi116cc022013-09-16 14:52:05 +02002862 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2863 if (excl)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002864 open_flag &= ~O_TRUNC;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002865
2866 /*
2867 * Checking write permission is tricky, bacuse we don't know if we are
2868 * going to actually need it: O_CREAT opens should work as long as the
2869 * file exists. But checking existence breaks atomicity. The trick is
2870 * to check access and if not granted clear O_CREAT from the flags.
2871 *
2872 * Another problem is returing the "right" error value (e.g. for an
2873 * O_EXCL open we want to return EEXIST not EROFS).
2874 */
Al Viro64894cf2012-07-31 00:53:35 +04002875 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2876 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2877 if (!(open_flag & O_CREAT)) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002878 /*
2879 * No O_CREATE -> atomicity not a requirement -> fall
2880 * back to lookup + open
2881 */
2882 goto no_open;
2883 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2884 /* Fall back and fail with the right error */
Al Viro64894cf2012-07-31 00:53:35 +04002885 create_error = -EROFS;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002886 goto no_open;
2887 } else {
2888 /* No side effects, safe to clear O_CREAT */
Al Viro64894cf2012-07-31 00:53:35 +04002889 create_error = -EROFS;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002890 open_flag &= ~O_CREAT;
2891 }
2892 }
2893
2894 if (open_flag & O_CREAT) {
Miklos Szeredi38227f72012-08-15 13:01:24 +02002895 error = may_o_create(&nd->path, dentry, mode);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002896 if (error) {
2897 create_error = error;
2898 if (open_flag & O_EXCL)
2899 goto no_open;
2900 open_flag &= ~O_CREAT;
2901 }
2902 }
2903
2904 if (nd->flags & LOOKUP_DIRECTORY)
2905 open_flag |= O_DIRECTORY;
2906
Al Viro30d90492012-06-22 12:40:19 +04002907 file->f_path.dentry = DENTRY_NOT_SET;
2908 file->f_path.mnt = nd->path.mnt;
2909 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
Al Viro47237682012-06-10 05:01:45 -04002910 opened);
Al Virod9585272012-06-22 12:39:14 +04002911 if (error < 0) {
Al Virod9585272012-06-22 12:39:14 +04002912 if (create_error && error == -ENOENT)
2913 error = create_error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002914 goto out;
2915 }
2916
Al Virod9585272012-06-22 12:39:14 +04002917 if (error) { /* returned 1, that is */
Al Viro30d90492012-06-22 12:40:19 +04002918 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
Al Viro2675a4e2012-06-22 12:41:10 +04002919 error = -EIO;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002920 goto out;
2921 }
Al Viro30d90492012-06-22 12:40:19 +04002922 if (file->f_path.dentry) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002923 dput(dentry);
Al Viro30d90492012-06-22 12:40:19 +04002924 dentry = file->f_path.dentry;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002925 }
Al Viro03da6332013-09-16 19:22:33 -04002926 if (*opened & FILE_CREATED)
2927 fsnotify_create(dir, dentry);
2928 if (!dentry->d_inode) {
2929 WARN_ON(*opened & FILE_CREATED);
2930 if (create_error) {
2931 error = create_error;
2932 goto out;
2933 }
2934 } else {
2935 if (excl && !(*opened & FILE_CREATED)) {
2936 error = -EEXIST;
2937 goto out;
2938 }
Sage Weil62b2ce92012-08-15 13:30:12 -07002939 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02002940 goto looked_up;
2941 }
2942
2943 /*
2944 * We didn't have the inode before the open, so check open permission
2945 * here.
2946 */
Al Viro03da6332013-09-16 19:22:33 -04002947 acc_mode = op->acc_mode;
2948 if (*opened & FILE_CREATED) {
2949 WARN_ON(!(open_flag & O_CREAT));
2950 fsnotify_create(dir, dentry);
Al Viro62fb4a12015-12-26 22:33:24 -05002951 acc_mode = 0;
Al Viro03da6332013-09-16 19:22:33 -04002952 }
Al Viro2675a4e2012-06-22 12:41:10 +04002953 error = may_open(&file->f_path, acc_mode, open_flag);
2954 if (error)
2955 fput(file);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002956
2957out:
2958 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04002959 return error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002960
Miklos Szeredid18e9002012-06-05 15:10:17 +02002961no_open:
2962 if (need_lookup) {
Al Viro72bd8662012-06-10 17:17:17 -04002963 dentry = lookup_real(dir, dentry, nd->flags);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002964 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002965 return PTR_ERR(dentry);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002966
2967 if (create_error) {
2968 int open_flag = op->open_flag;
2969
Al Viro2675a4e2012-06-22 12:41:10 +04002970 error = create_error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002971 if ((open_flag & O_EXCL)) {
2972 if (!dentry->d_inode)
2973 goto out;
2974 } else if (!dentry->d_inode) {
2975 goto out;
2976 } else if ((open_flag & O_TRUNC) &&
David Howellse36cb0b2015-01-29 12:02:35 +00002977 d_is_reg(dentry)) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002978 goto out;
2979 }
2980 /* will fail later, go on to get the right error */
2981 }
2982 }
2983looked_up:
2984 path->dentry = dentry;
2985 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04002986 return 1;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002987}
2988
Nick Piggin31e6b012011-01-07 17:49:52 +11002989/*
David Howells1acf0af2012-06-14 16:13:46 +01002990 * Look up and maybe create and open the last component.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002991 *
2992 * Must be called with i_mutex held on parent.
2993 *
David Howells1acf0af2012-06-14 16:13:46 +01002994 * Returns 0 if the file was successfully atomically created (if necessary) and
2995 * opened. In this case the file will be returned attached to @file.
2996 *
2997 * Returns 1 if the file was not completely opened at this time, though lookups
2998 * and creations will have been performed and the dentry returned in @path will
2999 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
3000 * specified then a negative dentry may be returned.
3001 *
3002 * An error code is returned otherwise.
3003 *
3004 * FILE_CREATE will be set in @*opened if the dentry was created and will be
3005 * cleared otherwise prior to returning.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003006 */
Al Viro2675a4e2012-06-22 12:41:10 +04003007static int lookup_open(struct nameidata *nd, struct path *path,
3008 struct file *file,
3009 const struct open_flags *op,
Al Viro64894cf2012-07-31 00:53:35 +04003010 bool got_write, int *opened)
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003011{
3012 struct dentry *dir = nd->path.dentry;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02003013 struct inode *dir_inode = dir->d_inode;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003014 struct dentry *dentry;
3015 int error;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02003016 bool need_lookup;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003017
Al Viro47237682012-06-10 05:01:45 -04003018 *opened &= ~FILE_CREATED;
Al Viro201f9562012-06-22 12:42:10 +04003019 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003020 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04003021 return PTR_ERR(dentry);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003022
Miklos Szeredid18e9002012-06-05 15:10:17 +02003023 /* Cached positive dentry: will open in f_op->open */
3024 if (!need_lookup && dentry->d_inode)
3025 goto out_no_open;
3026
3027 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
Al Viro64894cf2012-07-31 00:53:35 +04003028 return atomic_open(nd, dentry, path, file, op, got_write,
Al Viro47237682012-06-10 05:01:45 -04003029 need_lookup, opened);
Miklos Szeredid18e9002012-06-05 15:10:17 +02003030 }
3031
Miklos Szeredi54ef4872012-06-05 15:10:16 +02003032 if (need_lookup) {
3033 BUG_ON(dentry->d_inode);
3034
Al Viro72bd8662012-06-10 17:17:17 -04003035 dentry = lookup_real(dir_inode, dentry, nd->flags);
Miklos Szeredi54ef4872012-06-05 15:10:16 +02003036 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04003037 return PTR_ERR(dentry);
Miklos Szeredi54ef4872012-06-05 15:10:16 +02003038 }
3039
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003040 /* Negative dentry, just create the file */
3041 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
3042 umode_t mode = op->mode;
3043 if (!IS_POSIXACL(dir->d_inode))
3044 mode &= ~current_umask();
3045 /*
3046 * This write is needed to ensure that a
3047 * rw->ro transition does not occur between
3048 * the time when the file is created and when
3049 * a permanent write count is taken through
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003050 * the 'struct file' in finish_open().
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003051 */
Al Viro64894cf2012-07-31 00:53:35 +04003052 if (!got_write) {
3053 error = -EROFS;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003054 goto out_dput;
Al Viro64894cf2012-07-31 00:53:35 +04003055 }
Al Viro47237682012-06-10 05:01:45 -04003056 *opened |= FILE_CREATED;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003057 error = security_path_mknod(&nd->path, dentry, mode, 0);
3058 if (error)
3059 goto out_dput;
Al Viro312b63f2012-06-10 18:09:36 -04003060 error = vfs_create(dir->d_inode, dentry, mode,
3061 nd->flags & LOOKUP_EXCL);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003062 if (error)
3063 goto out_dput;
3064 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02003065out_no_open:
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003066 path->dentry = dentry;
3067 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04003068 return 1;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003069
3070out_dput:
3071 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04003072 return error;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02003073}
3074
3075/*
Al Virofe2d35f2011-03-05 22:58:25 -05003076 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11003077 */
Al Viro896475d2015-04-22 18:02:17 -04003078static int do_last(struct nameidata *nd,
Al Viro2675a4e2012-06-22 12:41:10 +04003079 struct file *file, const struct open_flags *op,
Al Viro76ae2a52015-05-12 18:44:32 -04003080 int *opened)
Al Virofb1cc552009-12-24 01:58:28 -05003081{
Al Viroa1e28032009-12-24 02:12:06 -05003082 struct dentry *dir = nd->path.dentry;
Al Viroca344a892011-03-09 00:36:45 -05003083 int open_flag = op->open_flag;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003084 bool will_truncate = (open_flag & O_TRUNC) != 0;
Al Viro64894cf2012-07-31 00:53:35 +04003085 bool got_write = false;
Al Virobcda7652011-03-13 16:42:14 -04003086 int acc_mode = op->acc_mode;
Al Viro254cf582015-05-05 09:40:46 -04003087 unsigned seq;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02003088 struct inode *inode;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003089 struct path save_parent = { .dentry = NULL, .mnt = NULL };
Al Viro896475d2015-04-22 18:02:17 -04003090 struct path path;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003091 bool retried = false;
Al Viro16c2cd72011-02-22 15:50:10 -05003092 int error;
Al Virofb1cc552009-12-24 01:58:28 -05003093
Al Viroc3e380b2011-02-23 13:39:45 -05003094 nd->flags &= ~LOOKUP_PARENT;
3095 nd->flags |= op->intent;
3096
Al Virobc77daa2013-06-06 09:12:33 -04003097 if (nd->last_type != LAST_NORM) {
Al Virofe2d35f2011-03-05 22:58:25 -05003098 error = handle_dots(nd, nd->last_type);
Al Virodeb106c2015-05-08 18:05:21 -04003099 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04003100 return error;
Miklos Szeredie83db162012-06-05 15:10:29 +02003101 goto finish_open;
Al Viro1f36f772009-12-26 10:56:19 -05003102 }
Al Viro67ee3ad2009-12-26 07:01:01 -05003103
Al Viroca344a892011-03-09 00:36:45 -05003104 if (!(open_flag & O_CREAT)) {
Al Virofe2d35f2011-03-05 22:58:25 -05003105 if (nd->last.name[nd->last.len])
3106 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3107 /* we _can_ be in RCU mode here */
Al Viro254cf582015-05-05 09:40:46 -04003108 error = lookup_fast(nd, &path, &inode, &seq);
Miklos Szeredi71574862012-06-05 15:10:14 +02003109 if (likely(!error))
3110 goto finish_lookup;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02003111
Miklos Szeredi71574862012-06-05 15:10:14 +02003112 if (error < 0)
Al Virodeb106c2015-05-08 18:05:21 -04003113 return error;
Miklos Szeredi37d7fff2012-06-05 15:10:12 +02003114
Miklos Szeredi71574862012-06-05 15:10:14 +02003115 BUG_ON(nd->inode != dir->d_inode);
Miklos Szeredib6183df2012-06-05 15:10:13 +02003116 } else {
3117 /* create side of things */
3118 /*
3119 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3120 * has been cleared when we got to the last component we are
3121 * about to look up
3122 */
3123 error = complete_walk(nd);
Al Viroe8bb73d2015-05-08 16:28:42 -04003124 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003125 return error;
Miklos Szeredib6183df2012-06-05 15:10:13 +02003126
Al Viro76ae2a52015-05-12 18:44:32 -04003127 audit_inode(nd->name, dir, LOOKUP_PARENT);
Miklos Szeredib6183df2012-06-05 15:10:13 +02003128 /* trailing slashes? */
Al Virodeb106c2015-05-08 18:05:21 -04003129 if (unlikely(nd->last.name[nd->last.len]))
3130 return -EISDIR;
Al Virofe2d35f2011-03-05 22:58:25 -05003131 }
3132
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003133retry_lookup:
Al Viro64894cf2012-07-31 00:53:35 +04003134 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3135 error = mnt_want_write(nd->path.mnt);
3136 if (!error)
3137 got_write = true;
3138 /*
3139 * do _not_ fail yet - we might not need that or fail with
3140 * a different error; let lookup_open() decide; we'll be
3141 * dropping this one anyway.
3142 */
3143 }
Al Viro59551022016-01-22 15:40:57 -05003144 inode_lock(dir->d_inode);
Al Viro896475d2015-04-22 18:02:17 -04003145 error = lookup_open(nd, &path, file, op, got_write, opened);
Al Viro59551022016-01-22 15:40:57 -05003146 inode_unlock(dir->d_inode);
Al Viroa1e28032009-12-24 02:12:06 -05003147
Al Viro2675a4e2012-06-22 12:41:10 +04003148 if (error <= 0) {
3149 if (error)
Miklos Szeredid18e9002012-06-05 15:10:17 +02003150 goto out;
3151
Al Viro47237682012-06-10 05:01:45 -04003152 if ((*opened & FILE_CREATED) ||
Al Viro496ad9a2013-01-23 17:07:38 -05003153 !S_ISREG(file_inode(file)->i_mode))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003154 will_truncate = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003155
Al Viro76ae2a52015-05-12 18:44:32 -04003156 audit_inode(nd->name, file->f_path.dentry, 0);
Miklos Szeredid18e9002012-06-05 15:10:17 +02003157 goto opened;
3158 }
Al Virofb1cc552009-12-24 01:58:28 -05003159
Al Viro47237682012-06-10 05:01:45 -04003160 if (*opened & FILE_CREATED) {
Al Viro9b44f1b2011-03-09 00:17:27 -05003161 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05003162 open_flag &= ~O_TRUNC;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003163 will_truncate = false;
Al Viro62fb4a12015-12-26 22:33:24 -05003164 acc_mode = 0;
Al Viro896475d2015-04-22 18:02:17 -04003165 path_to_nameidata(&path, nd);
Miklos Szeredie83db162012-06-05 15:10:29 +02003166 goto finish_open_created;
Al Virofb1cc552009-12-24 01:58:28 -05003167 }
3168
3169 /*
Jeff Layton3134f372012-07-25 10:19:47 -04003170 * create/update audit record if it already exists.
Al Virofb1cc552009-12-24 01:58:28 -05003171 */
Al Viro896475d2015-04-22 18:02:17 -04003172 if (d_is_positive(path.dentry))
Al Viro76ae2a52015-05-12 18:44:32 -04003173 audit_inode(nd->name, path.dentry, 0);
Al Virofb1cc552009-12-24 01:58:28 -05003174
Miklos Szeredid18e9002012-06-05 15:10:17 +02003175 /*
3176 * If atomic_open() acquired write access it is dropped now due to
3177 * possible mount and symlink following (this might be optimized away if
3178 * necessary...)
3179 */
Al Viro64894cf2012-07-31 00:53:35 +04003180 if (got_write) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02003181 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04003182 got_write = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02003183 }
3184
Al Virodeb106c2015-05-08 18:05:21 -04003185 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3186 path_to_nameidata(&path, nd);
3187 return -EEXIST;
3188 }
Al Virofb1cc552009-12-24 01:58:28 -05003189
NeilBrown756daf22015-03-23 13:37:38 +11003190 error = follow_managed(&path, nd);
Al Virodeb106c2015-05-08 18:05:21 -04003191 if (unlikely(error < 0))
3192 return error;
Al Viroa3fbbde2011-11-07 21:21:26 +00003193
Miklos Szeredidecf3402012-05-21 17:30:08 +02003194 BUG_ON(nd->flags & LOOKUP_RCU);
David Howells63afdfc2015-05-06 15:59:00 +01003195 inode = d_backing_inode(path.dentry);
Al Viro254cf582015-05-05 09:40:46 -04003196 seq = 0; /* out of RCU mode, so the value doesn't matter */
Al Virodeb106c2015-05-08 18:05:21 -04003197 if (unlikely(d_is_negative(path.dentry))) {
Al Viro896475d2015-04-22 18:02:17 -04003198 path_to_nameidata(&path, nd);
Al Virodeb106c2015-05-08 18:05:21 -04003199 return -ENOENT;
Miklos Szeredi54c33e72012-05-21 17:30:10 +02003200 }
Al Viro766c4cb2015-05-07 19:24:57 -04003201finish_lookup:
Al Virod63ff282015-05-04 18:13:23 -04003202 if (nd->depth)
3203 put_link(nd);
Al Viro181548c2015-05-07 19:54:34 -04003204 error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW,
3205 inode, seq);
Al Virodeb106c2015-05-08 18:05:21 -04003206 if (unlikely(error))
Al Virod63ff282015-05-04 18:13:23 -04003207 return error;
Al Virofb1cc552009-12-24 01:58:28 -05003208
Al Viro896475d2015-04-22 18:02:17 -04003209 if (unlikely(d_is_symlink(path.dentry)) && !(open_flag & O_PATH)) {
3210 path_to_nameidata(&path, nd);
Al Virodeb106c2015-05-08 18:05:21 -04003211 return -ELOOP;
Al Viroa5cfe2d2015-04-22 12:10:45 -04003212 }
3213
Al Viro896475d2015-04-22 18:02:17 -04003214 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) {
3215 path_to_nameidata(&path, nd);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003216 } else {
3217 save_parent.dentry = nd->path.dentry;
Al Viro896475d2015-04-22 18:02:17 -04003218 save_parent.mnt = mntget(path.mnt);
3219 nd->path.dentry = path.dentry;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003220
3221 }
Miklos Szeredidecf3402012-05-21 17:30:08 +02003222 nd->inode = inode;
Al Viro254cf582015-05-05 09:40:46 -04003223 nd->seq = seq;
Al Viroa3fbbde2011-11-07 21:21:26 +00003224 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
Al Virobc77daa2013-06-06 09:12:33 -04003225finish_open:
Al Viroa3fbbde2011-11-07 21:21:26 +00003226 error = complete_walk(nd);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003227 if (error) {
3228 path_put(&save_parent);
Al Viro2675a4e2012-06-22 12:41:10 +04003229 return error;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003230 }
Al Viro76ae2a52015-05-12 18:44:32 -04003231 audit_inode(nd->name, nd->path.dentry, 0);
Al Virofb1cc552009-12-24 01:58:28 -05003232 error = -EISDIR;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003233 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04003234 goto out;
Miklos Szerediaf2f5542012-05-21 17:30:11 +02003235 error = -ENOTDIR;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003236 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04003237 goto out;
David Howells4bbcbd32015-03-17 22:16:40 +00003238 if (!d_is_reg(nd->path.dentry))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003239 will_truncate = false;
Al Viro6c0d46c2011-03-09 00:59:59 -05003240
Al Viro0f9d1a12011-03-09 00:13:14 -05003241 if (will_truncate) {
3242 error = mnt_want_write(nd->path.mnt);
3243 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003244 goto out;
Al Viro64894cf2012-07-31 00:53:35 +04003245 got_write = true;
Al Viro0f9d1a12011-03-09 00:13:14 -05003246 }
Miklos Szeredie83db162012-06-05 15:10:29 +02003247finish_open_created:
Al Viro62fb4a12015-12-26 22:33:24 -05003248 if (likely(!(open_flag & O_PATH))) {
3249 error = may_open(&nd->path, acc_mode, open_flag);
3250 if (error)
3251 goto out;
3252 }
Miklos Szeredi4aa7c632014-10-24 00:14:35 +02003253 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3254 error = vfs_open(&nd->path, file, current_cred());
3255 if (!error) {
3256 *opened |= FILE_OPENED;
3257 } else {
Al Viro30d90492012-06-22 12:40:19 +04003258 if (error == -EOPENSTALE)
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003259 goto stale_open;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003260 goto out;
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003261 }
Miklos Szeredia8277b92012-06-05 15:10:32 +02003262opened:
Al Viro2675a4e2012-06-22 12:41:10 +04003263 error = open_check_o_direct(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003264 if (error)
3265 goto exit_fput;
Dmitry Kasatkin3034a142014-06-27 18:15:44 +03003266 error = ima_file_check(file, op->acc_mode, *opened);
Miklos Szerediaa4caad2012-06-05 15:10:28 +02003267 if (error)
3268 goto exit_fput;
3269
3270 if (will_truncate) {
Al Viro2675a4e2012-06-22 12:41:10 +04003271 error = handle_truncate(file);
Miklos Szerediaa4caad2012-06-05 15:10:28 +02003272 if (error)
3273 goto exit_fput;
Al Viro0f9d1a12011-03-09 00:13:14 -05003274 }
Al Viroca344a892011-03-09 00:36:45 -05003275out:
Al Viro64894cf2012-07-31 00:53:35 +04003276 if (got_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05003277 mnt_drop_write(nd->path.mnt);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003278 path_put(&save_parent);
Al Viro2675a4e2012-06-22 12:41:10 +04003279 return error;
Al Virofb1cc552009-12-24 01:58:28 -05003280
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003281exit_fput:
Al Viro2675a4e2012-06-22 12:41:10 +04003282 fput(file);
3283 goto out;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003284
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003285stale_open:
3286 /* If no saved parent or already retried then can't retry */
3287 if (!save_parent.dentry || retried)
3288 goto out;
3289
3290 BUG_ON(save_parent.dentry != dir);
3291 path_put(&nd->path);
3292 nd->path = save_parent;
3293 nd->inode = dir->d_inode;
3294 save_parent.mnt = NULL;
3295 save_parent.dentry = NULL;
Al Viro64894cf2012-07-31 00:53:35 +04003296 if (got_write) {
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003297 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04003298 got_write = false;
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003299 }
3300 retried = true;
3301 goto retry_lookup;
Al Virofb1cc552009-12-24 01:58:28 -05003302}
3303
Al Viroc8a53ee2015-05-12 18:43:07 -04003304static int do_tmpfile(struct nameidata *nd, unsigned flags,
Al Viro60545d02013-06-07 01:20:27 -04003305 const struct open_flags *op,
3306 struct file *file, int *opened)
3307{
3308 static const struct qstr name = QSTR_INIT("/", 1);
Al Viro625b6d12015-05-12 16:36:12 -04003309 struct dentry *child;
Al Viro60545d02013-06-07 01:20:27 -04003310 struct inode *dir;
Al Viro625b6d12015-05-12 16:36:12 -04003311 struct path path;
Al Viroc8a53ee2015-05-12 18:43:07 -04003312 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
Al Viro60545d02013-06-07 01:20:27 -04003313 if (unlikely(error))
3314 return error;
Al Viro625b6d12015-05-12 16:36:12 -04003315 error = mnt_want_write(path.mnt);
Al Viro60545d02013-06-07 01:20:27 -04003316 if (unlikely(error))
3317 goto out;
Al Viro625b6d12015-05-12 16:36:12 -04003318 dir = path.dentry->d_inode;
Al Viro60545d02013-06-07 01:20:27 -04003319 /* we want directory to be writable */
Al Viro625b6d12015-05-12 16:36:12 -04003320 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Al Viro60545d02013-06-07 01:20:27 -04003321 if (error)
3322 goto out2;
Al Viro60545d02013-06-07 01:20:27 -04003323 if (!dir->i_op->tmpfile) {
3324 error = -EOPNOTSUPP;
3325 goto out2;
3326 }
Al Viro625b6d12015-05-12 16:36:12 -04003327 child = d_alloc(path.dentry, &name);
Al Viro60545d02013-06-07 01:20:27 -04003328 if (unlikely(!child)) {
3329 error = -ENOMEM;
3330 goto out2;
3331 }
Al Viro625b6d12015-05-12 16:36:12 -04003332 dput(path.dentry);
3333 path.dentry = child;
3334 error = dir->i_op->tmpfile(dir, child, op->mode);
Al Viro60545d02013-06-07 01:20:27 -04003335 if (error)
3336 goto out2;
Al Viroc8a53ee2015-05-12 18:43:07 -04003337 audit_inode(nd->name, child, 0);
Eric Rannaud69a91c22014-10-30 01:51:01 -07003338 /* Don't check for other permissions, the inode was just created */
Al Viro62fb4a12015-12-26 22:33:24 -05003339 error = may_open(&path, 0, op->open_flag);
Al Viro60545d02013-06-07 01:20:27 -04003340 if (error)
3341 goto out2;
Al Viro625b6d12015-05-12 16:36:12 -04003342 file->f_path.mnt = path.mnt;
3343 error = finish_open(file, child, NULL, opened);
Al Viro60545d02013-06-07 01:20:27 -04003344 if (error)
3345 goto out2;
3346 error = open_check_o_direct(file);
Al Virof4e0c302013-06-11 08:34:36 +04003347 if (error) {
Al Viro60545d02013-06-07 01:20:27 -04003348 fput(file);
Al Virof4e0c302013-06-11 08:34:36 +04003349 } else if (!(op->open_flag & O_EXCL)) {
3350 struct inode *inode = file_inode(file);
3351 spin_lock(&inode->i_lock);
3352 inode->i_state |= I_LINKABLE;
3353 spin_unlock(&inode->i_lock);
3354 }
Al Viro60545d02013-06-07 01:20:27 -04003355out2:
Al Viro625b6d12015-05-12 16:36:12 -04003356 mnt_drop_write(path.mnt);
Al Viro60545d02013-06-07 01:20:27 -04003357out:
Al Viro625b6d12015-05-12 16:36:12 -04003358 path_put(&path);
Al Viro60545d02013-06-07 01:20:27 -04003359 return error;
3360}
3361
Al Viroc8a53ee2015-05-12 18:43:07 -04003362static struct file *path_openat(struct nameidata *nd,
3363 const struct open_flags *op, unsigned flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003364{
Al Viro368ee9b2015-05-08 17:19:59 -04003365 const char *s;
Al Viro30d90492012-06-22 12:40:19 +04003366 struct file *file;
Al Viro47237682012-06-10 05:01:45 -04003367 int opened = 0;
Al Viro13aab422011-02-23 17:54:08 -05003368 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11003369
Al Viro30d90492012-06-22 12:40:19 +04003370 file = get_empty_filp();
Al Viro1afc99b2013-02-14 20:41:04 -05003371 if (IS_ERR(file))
3372 return file;
Nick Piggin31e6b012011-01-07 17:49:52 +11003373
Al Viro30d90492012-06-22 12:40:19 +04003374 file->f_flags = op->open_flag;
Nick Piggin31e6b012011-01-07 17:49:52 +11003375
Al Virobb458c62013-07-13 13:26:37 +04003376 if (unlikely(file->f_flags & __O_TMPFILE)) {
Al Viroc8a53ee2015-05-12 18:43:07 -04003377 error = do_tmpfile(nd, flags, op, file, &opened);
Al Virof15133d2015-05-08 22:53:15 -04003378 goto out2;
Al Viro60545d02013-06-07 01:20:27 -04003379 }
3380
Al Viroc8a53ee2015-05-12 18:43:07 -04003381 s = path_init(nd, flags);
Al Viro368ee9b2015-05-08 17:19:59 -04003382 if (IS_ERR(s)) {
3383 put_filp(file);
3384 return ERR_CAST(s);
3385 }
Al Viro3bdba282015-05-08 17:37:07 -04003386 while (!(error = link_path_walk(s, nd)) &&
Al Viro76ae2a52015-05-12 18:44:32 -04003387 (error = do_last(nd, file, op, &opened)) > 0) {
Al Viro73d049a2011-03-11 12:08:24 -05003388 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
Al Viro3bdba282015-05-08 17:37:07 -04003389 s = trailing_symlink(nd);
3390 if (IS_ERR(s)) {
3391 error = PTR_ERR(s);
Al Viro2675a4e2012-06-22 12:41:10 +04003392 break;
Al Viro3bdba282015-05-08 17:37:07 -04003393 }
Al Viro806b6812009-12-26 07:16:40 -05003394 }
Al Virodeb106c2015-05-08 18:05:21 -04003395 terminate_walk(nd);
Al Virof15133d2015-05-08 22:53:15 -04003396out2:
Al Viro2675a4e2012-06-22 12:41:10 +04003397 if (!(opened & FILE_OPENED)) {
3398 BUG_ON(!error);
Al Viro30d90492012-06-22 12:40:19 +04003399 put_filp(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003400 }
Al Viro2675a4e2012-06-22 12:41:10 +04003401 if (unlikely(error)) {
3402 if (error == -EOPENSTALE) {
3403 if (flags & LOOKUP_RCU)
3404 error = -ECHILD;
3405 else
3406 error = -ESTALE;
3407 }
3408 file = ERR_PTR(error);
3409 }
3410 return file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003411}
3412
Jeff Layton669abf42012-10-10 16:43:10 -04003413struct file *do_filp_open(int dfd, struct filename *pathname,
Al Virof9652e12013-06-11 08:23:01 +04003414 const struct open_flags *op)
Al Viro13aab422011-02-23 17:54:08 -05003415{
Al Viro9883d182015-05-13 07:28:08 -04003416 struct nameidata nd;
Al Virof9652e12013-06-11 08:23:01 +04003417 int flags = op->lookup_flags;
Al Viro13aab422011-02-23 17:54:08 -05003418 struct file *filp;
3419
Al Viro9883d182015-05-13 07:28:08 -04003420 set_nameidata(&nd, dfd, pathname);
Al Viroc8a53ee2015-05-12 18:43:07 -04003421 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
Al Viro13aab422011-02-23 17:54:08 -05003422 if (unlikely(filp == ERR_PTR(-ECHILD)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003423 filp = path_openat(&nd, op, flags);
Al Viro13aab422011-02-23 17:54:08 -05003424 if (unlikely(filp == ERR_PTR(-ESTALE)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003425 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
Al Viro9883d182015-05-13 07:28:08 -04003426 restore_nameidata();
Al Viro13aab422011-02-23 17:54:08 -05003427 return filp;
3428}
3429
Al Viro73d049a2011-03-11 12:08:24 -05003430struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
Al Virof9652e12013-06-11 08:23:01 +04003431 const char *name, const struct open_flags *op)
Al Viro73d049a2011-03-11 12:08:24 -05003432{
Al Viro9883d182015-05-13 07:28:08 -04003433 struct nameidata nd;
Al Viro73d049a2011-03-11 12:08:24 -05003434 struct file *file;
Paul Moore51689102015-01-22 00:00:03 -05003435 struct filename *filename;
Al Virof9652e12013-06-11 08:23:01 +04003436 int flags = op->lookup_flags | LOOKUP_ROOT;
Al Viro73d049a2011-03-11 12:08:24 -05003437
3438 nd.root.mnt = mnt;
3439 nd.root.dentry = dentry;
3440
David Howellsb18825a2013-09-12 19:22:53 +01003441 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
Al Viro73d049a2011-03-11 12:08:24 -05003442 return ERR_PTR(-ELOOP);
3443
Paul Moore51689102015-01-22 00:00:03 -05003444 filename = getname_kernel(name);
Viresh Kumara1c83682015-08-12 15:59:44 +05303445 if (IS_ERR(filename))
Paul Moore51689102015-01-22 00:00:03 -05003446 return ERR_CAST(filename);
3447
Al Viro9883d182015-05-13 07:28:08 -04003448 set_nameidata(&nd, -1, filename);
Al Viroc8a53ee2015-05-12 18:43:07 -04003449 file = path_openat(&nd, op, flags | LOOKUP_RCU);
Al Viro73d049a2011-03-11 12:08:24 -05003450 if (unlikely(file == ERR_PTR(-ECHILD)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003451 file = path_openat(&nd, op, flags);
Al Viro73d049a2011-03-11 12:08:24 -05003452 if (unlikely(file == ERR_PTR(-ESTALE)))
Al Viroc8a53ee2015-05-12 18:43:07 -04003453 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
Al Viro9883d182015-05-13 07:28:08 -04003454 restore_nameidata();
Paul Moore51689102015-01-22 00:00:03 -05003455 putname(filename);
Al Viro73d049a2011-03-11 12:08:24 -05003456 return file;
3457}
3458
Al Virofa14a0b2015-01-22 02:16:49 -05003459static struct dentry *filename_create(int dfd, struct filename *name,
Jeff Layton1ac12b42012-12-11 12:10:06 -05003460 struct path *path, unsigned int lookup_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003461{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003462 struct dentry *dentry = ERR_PTR(-EEXIST);
Al Viro391172c2015-05-09 11:19:16 -04003463 struct qstr last;
3464 int type;
Jan Karac30dabf2012-06-12 16:20:30 +02003465 int err2;
Jeff Layton1ac12b42012-12-11 12:10:06 -05003466 int error;
3467 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3468
3469 /*
3470 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3471 * other flags passed in are ignored!
3472 */
3473 lookup_flags &= LOOKUP_REVAL;
3474
Al Viro5c31b6c2015-05-12 17:32:54 -04003475 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3476 if (IS_ERR(name))
3477 return ERR_CAST(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003478
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003479 /*
3480 * Yucky last component or no last component at all?
3481 * (foo/., foo/.., /////)
3482 */
Al Viro5c31b6c2015-05-12 17:32:54 -04003483 if (unlikely(type != LAST_NORM))
Al Viroed75e952011-06-27 16:53:43 -04003484 goto out;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003485
Jan Karac30dabf2012-06-12 16:20:30 +02003486 /* don't fail immediately if it's r/o, at least try to report other errors */
Al Viro391172c2015-05-09 11:19:16 -04003487 err2 = mnt_want_write(path->mnt);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003488 /*
3489 * Do the final lookup.
3490 */
Al Viro391172c2015-05-09 11:19:16 -04003491 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
Al Viro59551022016-01-22 15:40:57 -05003492 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
Al Viro391172c2015-05-09 11:19:16 -04003493 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003494 if (IS_ERR(dentry))
Al Viroa8104a92012-07-20 02:25:00 +04003495 goto unlock;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003496
Al Viroa8104a92012-07-20 02:25:00 +04003497 error = -EEXIST;
David Howellsb18825a2013-09-12 19:22:53 +01003498 if (d_is_positive(dentry))
Al Viroa8104a92012-07-20 02:25:00 +04003499 goto fail;
David Howellsb18825a2013-09-12 19:22:53 +01003500
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003501 /*
3502 * Special case - lookup gave negative, but... we had foo/bar/
3503 * From the vfs_mknod() POV we just have a negative dentry -
3504 * all is fine. Let's be bastards - you had / on the end, you've
3505 * been asking for (non-existent) directory. -ENOENT for you.
3506 */
Al Viro391172c2015-05-09 11:19:16 -04003507 if (unlikely(!is_dir && last.name[last.len])) {
Al Viroa8104a92012-07-20 02:25:00 +04003508 error = -ENOENT;
Al Viroed75e952011-06-27 16:53:43 -04003509 goto fail;
Al Viroe9baf6e2008-05-15 04:49:12 -04003510 }
Jan Karac30dabf2012-06-12 16:20:30 +02003511 if (unlikely(err2)) {
3512 error = err2;
Al Viroa8104a92012-07-20 02:25:00 +04003513 goto fail;
Jan Karac30dabf2012-06-12 16:20:30 +02003514 }
Al Viro181c37b2015-05-12 17:21:25 -04003515 putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003516 return dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003517fail:
Al Viroa8104a92012-07-20 02:25:00 +04003518 dput(dentry);
3519 dentry = ERR_PTR(error);
3520unlock:
Al Viro59551022016-01-22 15:40:57 -05003521 inode_unlock(path->dentry->d_inode);
Jan Karac30dabf2012-06-12 16:20:30 +02003522 if (!err2)
Al Viro391172c2015-05-09 11:19:16 -04003523 mnt_drop_write(path->mnt);
Al Viroed75e952011-06-27 16:53:43 -04003524out:
Al Viro391172c2015-05-09 11:19:16 -04003525 path_put(path);
Al Viro181c37b2015-05-12 17:21:25 -04003526 putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003527 return dentry;
3528}
Al Virofa14a0b2015-01-22 02:16:49 -05003529
3530struct dentry *kern_path_create(int dfd, const char *pathname,
3531 struct path *path, unsigned int lookup_flags)
3532{
Al Viro181c37b2015-05-12 17:21:25 -04003533 return filename_create(dfd, getname_kernel(pathname),
3534 path, lookup_flags);
Al Virofa14a0b2015-01-22 02:16:49 -05003535}
Al Virodae6ad82011-06-26 11:50:15 -04003536EXPORT_SYMBOL(kern_path_create);
3537
Al Viro921a1652012-07-20 01:15:31 +04003538void done_path_create(struct path *path, struct dentry *dentry)
3539{
3540 dput(dentry);
Al Viro59551022016-01-22 15:40:57 -05003541 inode_unlock(path->dentry->d_inode);
Al Viroa8104a92012-07-20 02:25:00 +04003542 mnt_drop_write(path->mnt);
Al Viro921a1652012-07-20 01:15:31 +04003543 path_put(path);
3544}
3545EXPORT_SYMBOL(done_path_create);
3546
Al Viro520ae682015-05-13 07:00:28 -04003547inline struct dentry *user_path_create(int dfd, const char __user *pathname,
Jeff Layton1ac12b42012-12-11 12:10:06 -05003548 struct path *path, unsigned int lookup_flags)
Al Virodae6ad82011-06-26 11:50:15 -04003549{
Al Viro181c37b2015-05-12 17:21:25 -04003550 return filename_create(dfd, getname(pathname), path, lookup_flags);
Al Virodae6ad82011-06-26 11:50:15 -04003551}
3552EXPORT_SYMBOL(user_path_create);
3553
Al Viro1a67aaf2011-07-26 01:52:52 -04003554int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003555{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003556 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003557
3558 if (error)
3559 return error;
3560
Eric W. Biederman975d6b32011-11-13 12:16:43 -08003561 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003562 return -EPERM;
3563
Al Viroacfa4382008-12-04 10:06:33 -05003564 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003565 return -EPERM;
3566
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07003567 error = devcgroup_inode_mknod(mode, dev);
3568 if (error)
3569 return error;
3570
Linus Torvalds1da177e2005-04-16 15:20:36 -07003571 error = security_inode_mknod(dir, dentry, mode, dev);
3572 if (error)
3573 return error;
3574
Linus Torvalds1da177e2005-04-16 15:20:36 -07003575 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003576 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003577 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003578 return error;
3579}
Al Viro4d359502014-03-14 12:20:17 -04003580EXPORT_SYMBOL(vfs_mknod);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003581
Al Virof69aac02011-07-26 04:31:40 -04003582static int may_mknod(umode_t mode)
Dave Hansen463c3192008-02-15 14:37:57 -08003583{
3584 switch (mode & S_IFMT) {
3585 case S_IFREG:
3586 case S_IFCHR:
3587 case S_IFBLK:
3588 case S_IFIFO:
3589 case S_IFSOCK:
3590 case 0: /* zero mode translates to S_IFREG */
3591 return 0;
3592 case S_IFDIR:
3593 return -EPERM;
3594 default:
3595 return -EINVAL;
3596 }
3597}
3598
Al Viro8208a222011-07-25 17:32:17 -04003599SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003600 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003601{
Al Viro2ad94ae2008-07-21 09:32:51 -04003602 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003603 struct path path;
3604 int error;
Jeff Layton972567f2012-12-20 16:00:10 -05003605 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003606
Al Viro8e4bfca2012-07-20 01:17:26 +04003607 error = may_mknod(mode);
3608 if (error)
3609 return error;
Jeff Layton972567f2012-12-20 16:00:10 -05003610retry:
3611 dentry = user_path_create(dfd, filename, &path, lookup_flags);
Al Virodae6ad82011-06-26 11:50:15 -04003612 if (IS_ERR(dentry))
3613 return PTR_ERR(dentry);
Al Viro2ad94ae2008-07-21 09:32:51 -04003614
Al Virodae6ad82011-06-26 11:50:15 -04003615 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003616 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003617 error = security_path_mknod(&path, dentry, mode, dev);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003618 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04003619 goto out;
Dave Hansen463c3192008-02-15 14:37:57 -08003620 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003621 case 0: case S_IFREG:
Al Viro312b63f2012-06-10 18:09:36 -04003622 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003623 break;
3624 case S_IFCHR: case S_IFBLK:
Al Virodae6ad82011-06-26 11:50:15 -04003625 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003626 new_decode_dev(dev));
3627 break;
3628 case S_IFIFO: case S_IFSOCK:
Al Virodae6ad82011-06-26 11:50:15 -04003629 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003630 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003631 }
Al Viroa8104a92012-07-20 02:25:00 +04003632out:
Al Viro921a1652012-07-20 01:15:31 +04003633 done_path_create(&path, dentry);
Jeff Layton972567f2012-12-20 16:00:10 -05003634 if (retry_estale(error, lookup_flags)) {
3635 lookup_flags |= LOOKUP_REVAL;
3636 goto retry;
3637 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003638 return error;
3639}
3640
Al Viro8208a222011-07-25 17:32:17 -04003641SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003642{
3643 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3644}
3645
Al Viro18bb1db2011-07-26 01:41:39 -04003646int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003647{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003648 int error = may_create(dir, dentry);
Al Viro8de52772012-02-06 12:45:27 -05003649 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003650
3651 if (error)
3652 return error;
3653
Al Viroacfa4382008-12-04 10:06:33 -05003654 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003655 return -EPERM;
3656
3657 mode &= (S_IRWXUGO|S_ISVTX);
3658 error = security_inode_mkdir(dir, dentry, mode);
3659 if (error)
3660 return error;
3661
Al Viro8de52772012-02-06 12:45:27 -05003662 if (max_links && dir->i_nlink >= max_links)
3663 return -EMLINK;
3664
Linus Torvalds1da177e2005-04-16 15:20:36 -07003665 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003666 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003667 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003668 return error;
3669}
Al Viro4d359502014-03-14 12:20:17 -04003670EXPORT_SYMBOL(vfs_mkdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003671
Al Viroa218d0f2011-11-21 14:59:34 -05003672SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003673{
Dave Hansen6902d922006-09-30 23:29:01 -07003674 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003675 struct path path;
3676 int error;
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003677 unsigned int lookup_flags = LOOKUP_DIRECTORY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003678
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003679retry:
3680 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07003681 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003682 return PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003683
Al Virodae6ad82011-06-26 11:50:15 -04003684 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003685 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003686 error = security_path_mkdir(&path, dentry, mode);
Al Viroa8104a92012-07-20 02:25:00 +04003687 if (!error)
3688 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
Al Viro921a1652012-07-20 01:15:31 +04003689 done_path_create(&path, dentry);
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003690 if (retry_estale(error, lookup_flags)) {
3691 lookup_flags |= LOOKUP_REVAL;
3692 goto retry;
3693 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003694 return error;
3695}
3696
Al Viroa218d0f2011-11-21 14:59:34 -05003697SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003698{
3699 return sys_mkdirat(AT_FDCWD, pathname, mode);
3700}
3701
Linus Torvalds1da177e2005-04-16 15:20:36 -07003702/*
Sage Weila71905f2011-05-24 13:06:08 -07003703 * The dentry_unhash() helper will try to drop the dentry early: we
J. Bruce Fieldsc0d02592012-02-15 11:48:40 -05003704 * should have a usage count of 1 if we're the only user of this
Sage Weila71905f2011-05-24 13:06:08 -07003705 * dentry, and if that is true (possibly after pruning the dcache),
3706 * then we drop the dentry now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003707 *
3708 * A low-level filesystem can, if it choses, legally
3709 * do a
3710 *
3711 * if (!d_unhashed(dentry))
3712 * return -EBUSY;
3713 *
3714 * if it cannot handle the case of removing a directory
3715 * that is still in use by something else..
3716 */
3717void dentry_unhash(struct dentry *dentry)
3718{
Vasily Averindc168422006-12-06 20:37:07 -08003719 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003720 spin_lock(&dentry->d_lock);
Waiman Long98474232013-08-28 18:24:59 -07003721 if (dentry->d_lockref.count == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003722 __d_drop(dentry);
3723 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003724}
Al Viro4d359502014-03-14 12:20:17 -04003725EXPORT_SYMBOL(dentry_unhash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003726
3727int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3728{
3729 int error = may_delete(dir, dentry, 1);
3730
3731 if (error)
3732 return error;
3733
Al Viroacfa4382008-12-04 10:06:33 -05003734 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003735 return -EPERM;
3736
Al Viro1d2ef592011-09-14 18:55:41 +01003737 dget(dentry);
Al Viro59551022016-01-22 15:40:57 -05003738 inode_lock(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003739
Sage Weil912dbc12011-05-24 13:06:11 -07003740 error = -EBUSY;
Eric W. Biederman7af13642013-10-04 19:15:13 -07003741 if (is_local_mountpoint(dentry))
Sage Weil912dbc12011-05-24 13:06:11 -07003742 goto out;
3743
3744 error = security_inode_rmdir(dir, dentry);
3745 if (error)
3746 goto out;
3747
Sage Weil3cebde22011-05-29 21:20:59 -07003748 shrink_dcache_parent(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003749 error = dir->i_op->rmdir(dir, dentry);
3750 if (error)
3751 goto out;
3752
3753 dentry->d_inode->i_flags |= S_DEAD;
3754 dont_mount(dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003755 detach_mounts(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003756
3757out:
Al Viro59551022016-01-22 15:40:57 -05003758 inode_unlock(dentry->d_inode);
Al Viro1d2ef592011-09-14 18:55:41 +01003759 dput(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003760 if (!error)
3761 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003762 return error;
3763}
Al Viro4d359502014-03-14 12:20:17 -04003764EXPORT_SYMBOL(vfs_rmdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003765
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003766static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003767{
3768 int error = 0;
Jeff Layton91a27b22012-10-10 15:25:28 -04003769 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003770 struct dentry *dentry;
Al Virof5beed72015-04-30 16:09:11 -04003771 struct path path;
3772 struct qstr last;
3773 int type;
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003774 unsigned int lookup_flags = 0;
3775retry:
Al Virof5beed72015-04-30 16:09:11 -04003776 name = user_path_parent(dfd, pathname,
3777 &path, &last, &type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04003778 if (IS_ERR(name))
3779 return PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003780
Al Virof5beed72015-04-30 16:09:11 -04003781 switch (type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003782 case LAST_DOTDOT:
3783 error = -ENOTEMPTY;
3784 goto exit1;
3785 case LAST_DOT:
3786 error = -EINVAL;
3787 goto exit1;
3788 case LAST_ROOT:
3789 error = -EBUSY;
3790 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003791 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003792
Al Virof5beed72015-04-30 16:09:11 -04003793 error = mnt_want_write(path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02003794 if (error)
3795 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003796
Al Viro59551022016-01-22 15:40:57 -05003797 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
Al Virof5beed72015-04-30 16:09:11 -04003798 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003799 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003800 if (IS_ERR(dentry))
3801 goto exit2;
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003802 if (!dentry->d_inode) {
3803 error = -ENOENT;
3804 goto exit3;
3805 }
Al Virof5beed72015-04-30 16:09:11 -04003806 error = security_path_rmdir(&path, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003807 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003808 goto exit3;
Al Virof5beed72015-04-30 16:09:11 -04003809 error = vfs_rmdir(path.dentry->d_inode, dentry);
Dave Hansen06227532008-02-15 14:37:34 -08003810exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07003811 dput(dentry);
3812exit2:
Al Viro59551022016-01-22 15:40:57 -05003813 inode_unlock(path.dentry->d_inode);
Al Virof5beed72015-04-30 16:09:11 -04003814 mnt_drop_write(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003815exit1:
Al Virof5beed72015-04-30 16:09:11 -04003816 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003817 putname(name);
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003818 if (retry_estale(error, lookup_flags)) {
3819 lookup_flags |= LOOKUP_REVAL;
3820 goto retry;
3821 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003822 return error;
3823}
3824
Heiko Carstens3cdad422009-01-14 14:14:22 +01003825SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003826{
3827 return do_rmdir(AT_FDCWD, pathname);
3828}
3829
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003830/**
3831 * vfs_unlink - unlink a filesystem object
3832 * @dir: parent directory
3833 * @dentry: victim
3834 * @delegated_inode: returns victim inode, if the inode is delegated.
3835 *
3836 * The caller must hold dir->i_mutex.
3837 *
3838 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3839 * return a reference to the inode in delegated_inode. The caller
3840 * should then break the delegation on that inode and retry. Because
3841 * breaking a delegation may take a long time, the caller should drop
3842 * dir->i_mutex before doing so.
3843 *
3844 * Alternatively, a caller may pass NULL for delegated_inode. This may
3845 * be appropriate for callers that expect the underlying filesystem not
3846 * to be NFS exported.
3847 */
3848int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003849{
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003850 struct inode *target = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003851 int error = may_delete(dir, dentry, 0);
3852
3853 if (error)
3854 return error;
3855
Al Viroacfa4382008-12-04 10:06:33 -05003856 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003857 return -EPERM;
3858
Al Viro59551022016-01-22 15:40:57 -05003859 inode_lock(target);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003860 if (is_local_mountpoint(dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003861 error = -EBUSY;
3862 else {
3863 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05003864 if (!error) {
J. Bruce Fields5a146962012-08-28 07:50:40 -07003865 error = try_break_deleg(target, delegated_inode);
3866 if (error)
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003867 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003868 error = dir->i_op->unlink(dir, dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003869 if (!error) {
Al Virod83c49f2010-04-30 17:17:09 -04003870 dont_mount(dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003871 detach_mounts(dentry);
3872 }
Al Virobec10522010-03-03 14:12:08 -05003873 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003874 }
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003875out:
Al Viro59551022016-01-22 15:40:57 -05003876 inode_unlock(target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003877
3878 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3879 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003880 fsnotify_link_count(target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003881 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003882 }
Robert Love0eeca282005-07-12 17:06:03 -04003883
Linus Torvalds1da177e2005-04-16 15:20:36 -07003884 return error;
3885}
Al Viro4d359502014-03-14 12:20:17 -04003886EXPORT_SYMBOL(vfs_unlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003887
3888/*
3889 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003890 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07003891 * writeout happening, and we don't want to prevent access to the directory
3892 * while waiting on the I/O.
3893 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003894static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003895{
Al Viro2ad94ae2008-07-21 09:32:51 -04003896 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04003897 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003898 struct dentry *dentry;
Al Virof5beed72015-04-30 16:09:11 -04003899 struct path path;
3900 struct qstr last;
3901 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003902 struct inode *inode = NULL;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003903 struct inode *delegated_inode = NULL;
Jeff Layton5d18f812012-12-20 16:38:04 -05003904 unsigned int lookup_flags = 0;
3905retry:
Al Virof5beed72015-04-30 16:09:11 -04003906 name = user_path_parent(dfd, pathname,
3907 &path, &last, &type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04003908 if (IS_ERR(name))
3909 return PTR_ERR(name);
Al Viro2ad94ae2008-07-21 09:32:51 -04003910
Linus Torvalds1da177e2005-04-16 15:20:36 -07003911 error = -EISDIR;
Al Virof5beed72015-04-30 16:09:11 -04003912 if (type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003913 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003914
Al Virof5beed72015-04-30 16:09:11 -04003915 error = mnt_want_write(path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02003916 if (error)
3917 goto exit1;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003918retry_deleg:
Al Viro59551022016-01-22 15:40:57 -05003919 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
Al Virof5beed72015-04-30 16:09:11 -04003920 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003921 error = PTR_ERR(dentry);
3922 if (!IS_ERR(dentry)) {
3923 /* Why not before? Because we want correct error value */
Al Virof5beed72015-04-30 16:09:11 -04003924 if (last.name[last.len])
Török Edwin50338b82011-06-16 00:06:14 +03003925 goto slashes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003926 inode = dentry->d_inode;
David Howellsb18825a2013-09-12 19:22:53 +01003927 if (d_is_negative(dentry))
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003928 goto slashes;
3929 ihold(inode);
Al Virof5beed72015-04-30 16:09:11 -04003930 error = security_path_unlink(&path, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003931 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003932 goto exit2;
Al Virof5beed72015-04-30 16:09:11 -04003933 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
Jan Karac30dabf2012-06-12 16:20:30 +02003934exit2:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003935 dput(dentry);
3936 }
Al Viro59551022016-01-22 15:40:57 -05003937 inode_unlock(path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003938 if (inode)
3939 iput(inode); /* truncate the inode here */
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003940 inode = NULL;
3941 if (delegated_inode) {
J. Bruce Fields5a146962012-08-28 07:50:40 -07003942 error = break_deleg_wait(&delegated_inode);
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003943 if (!error)
3944 goto retry_deleg;
3945 }
Al Virof5beed72015-04-30 16:09:11 -04003946 mnt_drop_write(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003947exit1:
Al Virof5beed72015-04-30 16:09:11 -04003948 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003949 putname(name);
Jeff Layton5d18f812012-12-20 16:38:04 -05003950 if (retry_estale(error, lookup_flags)) {
3951 lookup_flags |= LOOKUP_REVAL;
3952 inode = NULL;
3953 goto retry;
3954 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003955 return error;
3956
3957slashes:
David Howellsb18825a2013-09-12 19:22:53 +01003958 if (d_is_negative(dentry))
3959 error = -ENOENT;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003960 else if (d_is_dir(dentry))
David Howellsb18825a2013-09-12 19:22:53 +01003961 error = -EISDIR;
3962 else
3963 error = -ENOTDIR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003964 goto exit2;
3965}
3966
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003967SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003968{
3969 if ((flag & ~AT_REMOVEDIR) != 0)
3970 return -EINVAL;
3971
3972 if (flag & AT_REMOVEDIR)
3973 return do_rmdir(dfd, pathname);
3974
3975 return do_unlinkat(dfd, pathname);
3976}
3977
Heiko Carstens3480b252009-01-14 14:14:16 +01003978SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003979{
3980 return do_unlinkat(AT_FDCWD, pathname);
3981}
3982
Miklos Szeredidb2e7472008-06-24 16:50:16 +02003983int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003984{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003985 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003986
3987 if (error)
3988 return error;
3989
Al Viroacfa4382008-12-04 10:06:33 -05003990 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003991 return -EPERM;
3992
3993 error = security_inode_symlink(dir, dentry, oldname);
3994 if (error)
3995 return error;
3996
Linus Torvalds1da177e2005-04-16 15:20:36 -07003997 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003998 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003999 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004000 return error;
4001}
Al Viro4d359502014-03-14 12:20:17 -04004002EXPORT_SYMBOL(vfs_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004003
Heiko Carstens2e4d0922009-01-14 14:14:31 +01004004SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4005 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004006{
Al Viro2ad94ae2008-07-21 09:32:51 -04004007 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04004008 struct filename *from;
Dave Hansen6902d922006-09-30 23:29:01 -07004009 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04004010 struct path path;
Jeff Laytonf46d3562012-12-11 12:10:08 -05004011 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004012
4013 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04004014 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004015 return PTR_ERR(from);
Jeff Laytonf46d3562012-12-11 12:10:08 -05004016retry:
4017 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07004018 error = PTR_ERR(dentry);
4019 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04004020 goto out_putname;
Dave Hansen6902d922006-09-30 23:29:01 -07004021
Jeff Layton91a27b22012-10-10 15:25:28 -04004022 error = security_path_symlink(&path, dentry, from->name);
Al Viroa8104a92012-07-20 02:25:00 +04004023 if (!error)
Jeff Layton91a27b22012-10-10 15:25:28 -04004024 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
Al Viro921a1652012-07-20 01:15:31 +04004025 done_path_create(&path, dentry);
Jeff Laytonf46d3562012-12-11 12:10:08 -05004026 if (retry_estale(error, lookup_flags)) {
4027 lookup_flags |= LOOKUP_REVAL;
4028 goto retry;
4029 }
Dave Hansen6902d922006-09-30 23:29:01 -07004030out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004031 putname(from);
4032 return error;
4033}
4034
Heiko Carstens3480b252009-01-14 14:14:16 +01004035SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004036{
4037 return sys_symlinkat(oldname, AT_FDCWD, newname);
4038}
4039
J. Bruce Fields146a8592011-09-20 17:14:31 -04004040/**
4041 * vfs_link - create a new link
4042 * @old_dentry: object to be linked
4043 * @dir: new parent
4044 * @new_dentry: where to create the new link
4045 * @delegated_inode: returns inode needing a delegation break
4046 *
4047 * The caller must hold dir->i_mutex
4048 *
4049 * If vfs_link discovers a delegation on the to-be-linked file in need
4050 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4051 * inode in delegated_inode. The caller should then break the delegation
4052 * and retry. Because breaking a delegation may take a long time, the
4053 * caller should drop the i_mutex before doing so.
4054 *
4055 * Alternatively, a caller may pass NULL for delegated_inode. This may
4056 * be appropriate for callers that expect the underlying filesystem not
4057 * to be NFS exported.
4058 */
4059int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004060{
4061 struct inode *inode = old_dentry->d_inode;
Al Viro8de52772012-02-06 12:45:27 -05004062 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004063 int error;
4064
4065 if (!inode)
4066 return -ENOENT;
4067
Miklos Szeredia95164d2008-07-30 15:08:48 +02004068 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004069 if (error)
4070 return error;
4071
4072 if (dir->i_sb != inode->i_sb)
4073 return -EXDEV;
4074
4075 /*
4076 * A link to an append-only or immutable file cannot be created.
4077 */
4078 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4079 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05004080 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004081 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02004082 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004083 return -EPERM;
4084
4085 error = security_inode_link(old_dentry, dir, new_dentry);
4086 if (error)
4087 return error;
4088
Al Viro59551022016-01-22 15:40:57 -05004089 inode_lock(inode);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05304090 /* Make sure we don't allow creating hardlink to an unlinked file */
Al Virof4e0c302013-06-11 08:34:36 +04004091 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05304092 error = -ENOENT;
Al Viro8de52772012-02-06 12:45:27 -05004093 else if (max_links && inode->i_nlink >= max_links)
4094 error = -EMLINK;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004095 else {
4096 error = try_break_deleg(inode, delegated_inode);
4097 if (!error)
4098 error = dir->i_op->link(old_dentry, dir, new_dentry);
4099 }
Al Virof4e0c302013-06-11 08:34:36 +04004100
4101 if (!error && (inode->i_state & I_LINKABLE)) {
4102 spin_lock(&inode->i_lock);
4103 inode->i_state &= ~I_LINKABLE;
4104 spin_unlock(&inode->i_lock);
4105 }
Al Viro59551022016-01-22 15:40:57 -05004106 inode_unlock(inode);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07004107 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02004108 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004109 return error;
4110}
Al Viro4d359502014-03-14 12:20:17 -04004111EXPORT_SYMBOL(vfs_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004112
4113/*
4114 * Hardlinks are often used in delicate situations. We avoid
4115 * security-related surprises by not following symlinks on the
4116 * newname. --KAB
4117 *
4118 * We don't follow them on the oldname either to be compatible
4119 * with linux 2.0, and to avoid hard-linking to directories
4120 * and other special files. --ADM
4121 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01004122SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4123 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004124{
4125 struct dentry *new_dentry;
Al Virodae6ad82011-06-26 11:50:15 -04004126 struct path old_path, new_path;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004127 struct inode *delegated_inode = NULL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304128 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004129 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004130
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304131 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004132 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304133 /*
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004134 * To use null names we require CAP_DAC_READ_SEARCH
4135 * This ensures that not everyone will be able to create
4136 * handlink using the passed filedescriptor.
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304137 */
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004138 if (flags & AT_EMPTY_PATH) {
4139 if (!capable(CAP_DAC_READ_SEARCH))
4140 return -ENOENT;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304141 how = LOOKUP_EMPTY;
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07004142 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004143
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304144 if (flags & AT_SYMLINK_FOLLOW)
4145 how |= LOOKUP_FOLLOW;
Jeff Layton442e31c2012-12-20 16:15:38 -05004146retry:
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05304147 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004148 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04004149 return error;
4150
Jeff Layton442e31c2012-12-20 16:15:38 -05004151 new_dentry = user_path_create(newdfd, newname, &new_path,
4152 (how & LOOKUP_REVAL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004153 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07004154 if (IS_ERR(new_dentry))
Al Virodae6ad82011-06-26 11:50:15 -04004155 goto out;
4156
4157 error = -EXDEV;
4158 if (old_path.mnt != new_path.mnt)
4159 goto out_dput;
Kees Cook800179c2012-07-25 17:29:07 -07004160 error = may_linkat(&old_path);
4161 if (unlikely(error))
4162 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04004163 error = security_path_link(old_path.dentry, &new_path, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004164 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04004165 goto out_dput;
J. Bruce Fields146a8592011-09-20 17:14:31 -04004166 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
Dave Hansen75c3f292008-02-15 14:37:45 -08004167out_dput:
Al Viro921a1652012-07-20 01:15:31 +04004168 done_path_create(&new_path, new_dentry);
J. Bruce Fields146a8592011-09-20 17:14:31 -04004169 if (delegated_inode) {
4170 error = break_deleg_wait(&delegated_inode);
Oleg Drokind22e6332014-01-31 15:41:58 -05004171 if (!error) {
4172 path_put(&old_path);
J. Bruce Fields146a8592011-09-20 17:14:31 -04004173 goto retry;
Oleg Drokind22e6332014-01-31 15:41:58 -05004174 }
J. Bruce Fields146a8592011-09-20 17:14:31 -04004175 }
Jeff Layton442e31c2012-12-20 16:15:38 -05004176 if (retry_estale(error, how)) {
Oleg Drokind22e6332014-01-31 15:41:58 -05004177 path_put(&old_path);
Jeff Layton442e31c2012-12-20 16:15:38 -05004178 how |= LOOKUP_REVAL;
4179 goto retry;
4180 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004181out:
Al Viro2d8f3032008-07-22 09:59:21 -04004182 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004183
4184 return error;
4185}
4186
Heiko Carstens3480b252009-01-14 14:14:16 +01004187SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004188{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08004189 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004190}
4191
Miklos Szeredibc270272014-04-01 17:08:42 +02004192/**
4193 * vfs_rename - rename a filesystem object
4194 * @old_dir: parent of source
4195 * @old_dentry: source
4196 * @new_dir: parent of destination
4197 * @new_dentry: destination
4198 * @delegated_inode: returns an inode needing a delegation break
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004199 * @flags: rename flags
Miklos Szeredibc270272014-04-01 17:08:42 +02004200 *
4201 * The caller must hold multiple mutexes--see lock_rename()).
4202 *
4203 * If vfs_rename discovers a delegation in need of breaking at either
4204 * the source or destination, it will return -EWOULDBLOCK and return a
4205 * reference to the inode in delegated_inode. The caller should then
4206 * break the delegation and retry. Because breaking a delegation may
4207 * take a long time, the caller should drop all locks before doing
4208 * so.
4209 *
4210 * Alternatively, a caller may pass NULL for delegated_inode. This may
4211 * be appropriate for callers that expect the underlying filesystem not
4212 * to be NFS exported.
4213 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07004214 * The worst of all namespace operations - renaming directory. "Perverted"
4215 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4216 * Problems:
J. Bruce Fieldsd03b29a2014-02-17 16:52:33 -05004217 * a) we can get into loop creation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004218 * b) race potential - two innocent renames can create a loop together.
4219 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004220 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07004221 * story.
J. Bruce Fields6cedba82012-03-05 11:40:41 -05004222 * c) we have to lock _four_ objects - parents and victim (if it exists),
4223 * and source (if it is not a directory).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004224 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07004225 * whether the target exists). Solution: try to be smart with locking
4226 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004227 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07004228 * move will be locked. Thus we can rank directories by the tree
4229 * (ancestors first) and rank all non-directories after them.
4230 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004231 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004232 * HOWEVER, it relies on the assumption that any object with ->lookup()
4233 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4234 * we'd better make sure that there's no link(2) for them.
Sage Weile4eaac02011-05-24 13:06:07 -07004235 * d) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004236 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07004237 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05004238 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07004239 * locking].
4240 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004241int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004242 struct inode *new_dir, struct dentry *new_dentry,
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004243 struct inode **delegated_inode, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004244{
4245 int error;
Miklos Szeredibc270272014-04-01 17:08:42 +02004246 bool is_dir = d_is_dir(old_dentry);
Eric Paris59b0df22010-02-08 12:53:52 -05004247 const unsigned char *old_name;
Miklos Szeredibc270272014-04-01 17:08:42 +02004248 struct inode *source = old_dentry->d_inode;
4249 struct inode *target = new_dentry->d_inode;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004250 bool new_is_dir = false;
4251 unsigned max_links = new_dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004252
Miklos Szeredibc270272014-04-01 17:08:42 +02004253 if (source == target)
4254 return 0;
4255
Linus Torvalds1da177e2005-04-16 15:20:36 -07004256 error = may_delete(old_dir, old_dentry, is_dir);
4257 if (error)
4258 return error;
4259
Miklos Szeredida1ce062014-04-01 17:08:43 +02004260 if (!target) {
Miklos Szeredia95164d2008-07-30 15:08:48 +02004261 error = may_create(new_dir, new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004262 } else {
4263 new_is_dir = d_is_dir(new_dentry);
4264
4265 if (!(flags & RENAME_EXCHANGE))
4266 error = may_delete(new_dir, new_dentry, is_dir);
4267 else
4268 error = may_delete(new_dir, new_dentry, new_is_dir);
4269 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004270 if (error)
4271 return error;
4272
Miklos Szeredi7177a9c2014-07-23 15:15:30 +02004273 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004274 return -EPERM;
4275
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004276 if (flags && !old_dir->i_op->rename2)
4277 return -EINVAL;
4278
Miklos Szeredibc270272014-04-01 17:08:42 +02004279 /*
4280 * If we are going to change the parent - check write permissions,
4281 * we'll need to flip '..'.
4282 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02004283 if (new_dir != old_dir) {
4284 if (is_dir) {
4285 error = inode_permission(source, MAY_WRITE);
4286 if (error)
4287 return error;
4288 }
4289 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4290 error = inode_permission(target, MAY_WRITE);
4291 if (error)
4292 return error;
4293 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004294 }
Robert Love0eeca282005-07-12 17:06:03 -04004295
Miklos Szeredi0b3974e2014-04-01 17:08:43 +02004296 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4297 flags);
Miklos Szeredibc270272014-04-01 17:08:42 +02004298 if (error)
4299 return error;
4300
4301 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4302 dget(new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004303 if (!is_dir || (flags & RENAME_EXCHANGE))
Miklos Szeredibc270272014-04-01 17:08:42 +02004304 lock_two_nondirectories(source, target);
4305 else if (target)
Al Viro59551022016-01-22 15:40:57 -05004306 inode_lock(target);
Miklos Szeredibc270272014-04-01 17:08:42 +02004307
4308 error = -EBUSY;
Eric W. Biederman7af13642013-10-04 19:15:13 -07004309 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
Miklos Szeredibc270272014-04-01 17:08:42 +02004310 goto out;
4311
Miklos Szeredida1ce062014-04-01 17:08:43 +02004312 if (max_links && new_dir != old_dir) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004313 error = -EMLINK;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004314 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
Miklos Szeredibc270272014-04-01 17:08:42 +02004315 goto out;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004316 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4317 old_dir->i_nlink >= max_links)
4318 goto out;
4319 }
4320 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4321 shrink_dcache_parent(new_dentry);
4322 if (!is_dir) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004323 error = try_break_deleg(source, delegated_inode);
4324 if (error)
4325 goto out;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004326 }
4327 if (target && !new_is_dir) {
4328 error = try_break_deleg(target, delegated_inode);
4329 if (error)
4330 goto out;
Miklos Szeredibc270272014-04-01 17:08:42 +02004331 }
Miklos Szeredi7177a9c2014-07-23 15:15:30 +02004332 if (!old_dir->i_op->rename2) {
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004333 error = old_dir->i_op->rename(old_dir, old_dentry,
4334 new_dir, new_dentry);
4335 } else {
Miklos Szeredi7177a9c2014-07-23 15:15:30 +02004336 WARN_ON(old_dir->i_op->rename != NULL);
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004337 error = old_dir->i_op->rename2(old_dir, old_dentry,
4338 new_dir, new_dentry, flags);
4339 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004340 if (error)
4341 goto out;
4342
Miklos Szeredida1ce062014-04-01 17:08:43 +02004343 if (!(flags & RENAME_EXCHANGE) && target) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004344 if (is_dir)
4345 target->i_flags |= S_DEAD;
4346 dont_mount(new_dentry);
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07004347 detach_mounts(new_dentry);
Miklos Szeredibc270272014-04-01 17:08:42 +02004348 }
Miklos Szeredida1ce062014-04-01 17:08:43 +02004349 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4350 if (!(flags & RENAME_EXCHANGE))
4351 d_move(old_dentry, new_dentry);
4352 else
4353 d_exchange(old_dentry, new_dentry);
4354 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004355out:
Miklos Szeredida1ce062014-04-01 17:08:43 +02004356 if (!is_dir || (flags & RENAME_EXCHANGE))
Miklos Szeredibc270272014-04-01 17:08:42 +02004357 unlock_two_nondirectories(source, target);
4358 else if (target)
Al Viro59551022016-01-22 15:40:57 -05004359 inode_unlock(target);
Miklos Szeredibc270272014-04-01 17:08:42 +02004360 dput(new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004361 if (!error) {
Al Viro123df292009-12-25 04:57:57 -05004362 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Miklos Szeredida1ce062014-04-01 17:08:43 +02004363 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4364 if (flags & RENAME_EXCHANGE) {
4365 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4366 new_is_dir, NULL, new_dentry);
4367 }
4368 }
Robert Love0eeca282005-07-12 17:06:03 -04004369 fsnotify_oldname_free(old_name);
4370
Linus Torvalds1da177e2005-04-16 15:20:36 -07004371 return error;
4372}
Al Viro4d359502014-03-14 12:20:17 -04004373EXPORT_SYMBOL(vfs_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004374
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004375SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4376 int, newdfd, const char __user *, newname, unsigned int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004377{
Al Viro2ad94ae2008-07-21 09:32:51 -04004378 struct dentry *old_dentry, *new_dentry;
4379 struct dentry *trap;
Al Virof5beed72015-04-30 16:09:11 -04004380 struct path old_path, new_path;
4381 struct qstr old_last, new_last;
4382 int old_type, new_type;
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004383 struct inode *delegated_inode = NULL;
Jeff Layton91a27b22012-10-10 15:25:28 -04004384 struct filename *from;
4385 struct filename *to;
Al Virof5beed72015-04-30 16:09:11 -04004386 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
Jeff Laytonc6a94282012-12-11 12:10:10 -05004387 bool should_retry = false;
Al Viro2ad94ae2008-07-21 09:32:51 -04004388 int error;
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004389
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004390 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredida1ce062014-04-01 17:08:43 +02004391 return -EINVAL;
4392
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004393 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4394 (flags & RENAME_EXCHANGE))
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004395 return -EINVAL;
4396
Miklos Szeredi0d7a8552014-10-24 00:14:37 +02004397 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4398 return -EPERM;
4399
Al Virof5beed72015-04-30 16:09:11 -04004400 if (flags & RENAME_EXCHANGE)
4401 target_flags = 0;
4402
Jeff Laytonc6a94282012-12-11 12:10:10 -05004403retry:
Al Virof5beed72015-04-30 16:09:11 -04004404 from = user_path_parent(olddfd, oldname,
4405 &old_path, &old_last, &old_type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04004406 if (IS_ERR(from)) {
4407 error = PTR_ERR(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004408 goto exit;
Jeff Layton91a27b22012-10-10 15:25:28 -04004409 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004410
Al Virof5beed72015-04-30 16:09:11 -04004411 to = user_path_parent(newdfd, newname,
4412 &new_path, &new_last, &new_type, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04004413 if (IS_ERR(to)) {
4414 error = PTR_ERR(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004415 goto exit1;
Jeff Layton91a27b22012-10-10 15:25:28 -04004416 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004417
4418 error = -EXDEV;
Al Virof5beed72015-04-30 16:09:11 -04004419 if (old_path.mnt != new_path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004420 goto exit2;
4421
Linus Torvalds1da177e2005-04-16 15:20:36 -07004422 error = -EBUSY;
Al Virof5beed72015-04-30 16:09:11 -04004423 if (old_type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004424 goto exit2;
4425
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004426 if (flags & RENAME_NOREPLACE)
4427 error = -EEXIST;
Al Virof5beed72015-04-30 16:09:11 -04004428 if (new_type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004429 goto exit2;
4430
Al Virof5beed72015-04-30 16:09:11 -04004431 error = mnt_want_write(old_path.mnt);
Jan Karac30dabf2012-06-12 16:20:30 +02004432 if (error)
4433 goto exit2;
4434
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004435retry_deleg:
Al Virof5beed72015-04-30 16:09:11 -04004436 trap = lock_rename(new_path.dentry, old_path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004437
Al Virof5beed72015-04-30 16:09:11 -04004438 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004439 error = PTR_ERR(old_dentry);
4440 if (IS_ERR(old_dentry))
4441 goto exit3;
4442 /* source must exist */
4443 error = -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01004444 if (d_is_negative(old_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004445 goto exit4;
Al Virof5beed72015-04-30 16:09:11 -04004446 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004447 error = PTR_ERR(new_dentry);
4448 if (IS_ERR(new_dentry))
4449 goto exit4;
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004450 error = -EEXIST;
4451 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4452 goto exit5;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004453 if (flags & RENAME_EXCHANGE) {
4454 error = -ENOENT;
4455 if (d_is_negative(new_dentry))
4456 goto exit5;
4457
4458 if (!d_is_dir(new_dentry)) {
4459 error = -ENOTDIR;
Al Virof5beed72015-04-30 16:09:11 -04004460 if (new_last.name[new_last.len])
Miklos Szeredida1ce062014-04-01 17:08:43 +02004461 goto exit5;
4462 }
4463 }
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004464 /* unless the source is a directory trailing slashes give -ENOTDIR */
4465 if (!d_is_dir(old_dentry)) {
4466 error = -ENOTDIR;
Al Virof5beed72015-04-30 16:09:11 -04004467 if (old_last.name[old_last.len])
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004468 goto exit5;
Al Virof5beed72015-04-30 16:09:11 -04004469 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004470 goto exit5;
4471 }
4472 /* source should not be ancestor of target */
4473 error = -EINVAL;
4474 if (old_dentry == trap)
4475 goto exit5;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004476 /* target should not be an ancestor of source */
Miklos Szeredida1ce062014-04-01 17:08:43 +02004477 if (!(flags & RENAME_EXCHANGE))
4478 error = -ENOTEMPTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004479 if (new_dentry == trap)
4480 goto exit5;
4481
Al Virof5beed72015-04-30 16:09:11 -04004482 error = security_path_rename(&old_path, old_dentry,
4483 &new_path, new_dentry, flags);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004484 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02004485 goto exit5;
Al Virof5beed72015-04-30 16:09:11 -04004486 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4487 new_path.dentry->d_inode, new_dentry,
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004488 &delegated_inode, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004489exit5:
4490 dput(new_dentry);
4491exit4:
4492 dput(old_dentry);
4493exit3:
Al Virof5beed72015-04-30 16:09:11 -04004494 unlock_rename(new_path.dentry, old_path.dentry);
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004495 if (delegated_inode) {
4496 error = break_deleg_wait(&delegated_inode);
4497 if (!error)
4498 goto retry_deleg;
4499 }
Al Virof5beed72015-04-30 16:09:11 -04004500 mnt_drop_write(old_path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004501exit2:
Jeff Laytonc6a94282012-12-11 12:10:10 -05004502 if (retry_estale(error, lookup_flags))
4503 should_retry = true;
Al Virof5beed72015-04-30 16:09:11 -04004504 path_put(&new_path);
Al Viro2ad94ae2008-07-21 09:32:51 -04004505 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004506exit1:
Al Virof5beed72015-04-30 16:09:11 -04004507 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004508 putname(from);
Jeff Laytonc6a94282012-12-11 12:10:10 -05004509 if (should_retry) {
4510 should_retry = false;
4511 lookup_flags |= LOOKUP_REVAL;
4512 goto retry;
4513 }
Al Viro2ad94ae2008-07-21 09:32:51 -04004514exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004515 return error;
4516}
4517
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004518SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4519 int, newdfd, const char __user *, newname)
4520{
4521 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4522}
4523
Heiko Carstensa26eab22009-01-14 14:14:17 +01004524SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004525{
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004526 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004527}
4528
Miklos Szeredi787fb6b2014-10-24 00:14:36 +02004529int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4530{
4531 int error = may_create(dir, dentry);
4532 if (error)
4533 return error;
4534
4535 if (!dir->i_op->mknod)
4536 return -EPERM;
4537
4538 return dir->i_op->mknod(dir, dentry,
4539 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4540}
4541EXPORT_SYMBOL(vfs_whiteout);
4542
Al Viro5d826c82014-03-14 13:42:45 -04004543int readlink_copy(char __user *buffer, int buflen, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004544{
Al Viro5d826c82014-03-14 13:42:45 -04004545 int len = PTR_ERR(link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004546 if (IS_ERR(link))
4547 goto out;
4548
4549 len = strlen(link);
4550 if (len > (unsigned) buflen)
4551 len = buflen;
4552 if (copy_to_user(buffer, link, len))
4553 len = -EFAULT;
4554out:
4555 return len;
4556}
Al Viro5d826c82014-03-14 13:42:45 -04004557EXPORT_SYMBOL(readlink_copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004558
4559/*
4560 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
Al Viro6b255392015-11-17 10:20:54 -05004561 * have ->get_link() not calling nd_jump_link(). Using (or not using) it
4562 * for any given inode is up to filesystem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004563 */
4564int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4565{
Al Virofceef392015-12-29 15:58:39 -05004566 DEFINE_DELAYED_CALL(done);
Al Viro5f2c4172015-05-07 11:14:26 -04004567 struct inode *inode = d_inode(dentry);
4568 const char *link = inode->i_link;
Marcin Slusarz694a1762008-06-09 16:40:37 -07004569 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004570
Al Virod4dee482015-04-30 20:08:02 -04004571 if (!link) {
Al Virofceef392015-12-29 15:58:39 -05004572 link = inode->i_op->get_link(dentry, inode, &done);
Al Virod4dee482015-04-30 20:08:02 -04004573 if (IS_ERR(link))
4574 return PTR_ERR(link);
4575 }
Al Viro680baac2015-05-02 13:32:22 -04004576 res = readlink_copy(buffer, buflen, link);
Al Virofceef392015-12-29 15:58:39 -05004577 do_delayed_call(&done);
Marcin Slusarz694a1762008-06-09 16:40:37 -07004578 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004579}
Al Viro4d359502014-03-14 12:20:17 -04004580EXPORT_SYMBOL(generic_readlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004581
Linus Torvalds1da177e2005-04-16 15:20:36 -07004582/* get the link contents into pagecache */
Al Viro6b255392015-11-17 10:20:54 -05004583const char *page_get_link(struct dentry *dentry, struct inode *inode,
Al Virofceef392015-12-29 15:58:39 -05004584 struct delayed_call *callback)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004585{
Duane Griffinebd09ab2008-12-19 20:47:12 +00004586 char *kaddr;
4587 struct page *page;
Al Viro6b255392015-11-17 10:20:54 -05004588 struct address_space *mapping = inode->i_mapping;
4589
Al Virod3883d42015-11-17 10:41:04 -05004590 if (!dentry) {
4591 page = find_get_page(mapping, 0);
4592 if (!page)
4593 return ERR_PTR(-ECHILD);
4594 if (!PageUptodate(page)) {
4595 put_page(page);
4596 return ERR_PTR(-ECHILD);
4597 }
4598 } else {
4599 page = read_mapping_page(mapping, 0, NULL);
4600 if (IS_ERR(page))
4601 return (char*)page;
4602 }
Al Virofceef392015-12-29 15:58:39 -05004603 set_delayed_call(callback, page_put_link, page);
Al Viro21fc61c2015-11-17 01:07:57 -05004604 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4605 kaddr = page_address(page);
Al Viro6b255392015-11-17 10:20:54 -05004606 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
Duane Griffinebd09ab2008-12-19 20:47:12 +00004607 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004608}
4609
Al Viro6b255392015-11-17 10:20:54 -05004610EXPORT_SYMBOL(page_get_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004611
Al Virofceef392015-12-29 15:58:39 -05004612void page_put_link(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004613{
Al Virofceef392015-12-29 15:58:39 -05004614 put_page(arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004615}
Al Viro4d359502014-03-14 12:20:17 -04004616EXPORT_SYMBOL(page_put_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004617
Al Viroaa80dea2015-11-16 18:26:34 -05004618int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4619{
Al Virofceef392015-12-29 15:58:39 -05004620 DEFINE_DELAYED_CALL(done);
Al Viro6b255392015-11-17 10:20:54 -05004621 int res = readlink_copy(buffer, buflen,
4622 page_get_link(dentry, d_inode(dentry),
Al Virofceef392015-12-29 15:58:39 -05004623 &done));
4624 do_delayed_call(&done);
Al Viroaa80dea2015-11-16 18:26:34 -05004625 return res;
4626}
4627EXPORT_SYMBOL(page_readlink);
4628
Nick Piggin54566b22009-01-04 12:00:53 -08004629/*
4630 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4631 */
4632int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004633{
4634 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004635 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07004636 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08004637 int err;
Nick Piggin54566b22009-01-04 12:00:53 -08004638 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4639 if (nofs)
4640 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004641
NeilBrown7e53cac2006-03-25 03:07:57 -08004642retry:
Nick Pigginafddba42007-10-16 01:25:01 -07004643 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08004644 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004645 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07004646 goto fail;
4647
Al Viro21fc61c2015-11-17 01:07:57 -05004648 memcpy(page_address(page), symname, len-1);
Nick Pigginafddba42007-10-16 01:25:01 -07004649
4650 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4651 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004652 if (err < 0)
4653 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07004654 if (err < len-1)
4655 goto retry;
4656
Linus Torvalds1da177e2005-04-16 15:20:36 -07004657 mark_inode_dirty(inode);
4658 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004659fail:
4660 return err;
4661}
Al Viro4d359502014-03-14 12:20:17 -04004662EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004663
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004664int page_symlink(struct inode *inode, const char *symname, int len)
4665{
4666 return __page_symlink(inode, symname, len,
Michal Hockoc62d2552015-11-06 16:28:49 -08004667 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004668}
Al Viro4d359502014-03-14 12:20:17 -04004669EXPORT_SYMBOL(page_symlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004670
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08004671const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004672 .readlink = generic_readlink,
Al Viro6b255392015-11-17 10:20:54 -05004673 .get_link = page_get_link,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004674};
Linus Torvalds1da177e2005-04-16 15:20:36 -07004675EXPORT_SYMBOL(page_symlink_inode_operations);