blob: 4b491b4319905b2dd6698d3f58aae1d0d0141499 [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 Torvalds1da177e2005-04-16 15:20:36 -070037#include <asm/uaccess.h>
38
Eric Parise81e3f42009-12-04 15:47:36 -050039#include "internal.h"
Al Viroc7105362011-11-24 18:22:03 -050040#include "mount.h"
Eric Parise81e3f42009-12-04 15:47:36 -050041
Linus Torvalds1da177e2005-04-16 15:20:36 -070042/* [Feb-1997 T. Schoebel-Theuer]
43 * Fundamental changes in the pathname lookup mechanisms (namei)
44 * were necessary because of omirr. The reason is that omirr needs
45 * to know the _real_ pathname, not the user-supplied one, in case
46 * of symlinks (and also when transname replacements occur).
47 *
48 * The new code replaces the old recursive symlink resolution with
49 * an iterative one (in case of non-nested symlink chains). It does
50 * this with calls to <fs>_follow_link().
51 * As a side effect, dir_namei(), _namei() and follow_link() are now
52 * replaced with a single function lookup_dentry() that can handle all
53 * the special cases of the former code.
54 *
55 * With the new dcache, the pathname is stored at each inode, at least as
56 * long as the refcount of the inode is positive. As a side effect, the
57 * size of the dcache depends on the inode cache and thus is dynamic.
58 *
59 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
60 * resolution to correspond with current state of the code.
61 *
62 * Note that the symlink resolution is not *completely* iterative.
63 * There is still a significant amount of tail- and mid- recursion in
64 * the algorithm. Also, note that <fs>_readlink() is not used in
65 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
66 * may return different results than <fs>_follow_link(). Many virtual
67 * filesystems (including /proc) exhibit this behavior.
68 */
69
70/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
71 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
72 * and the name already exists in form of a symlink, try to create the new
73 * name indicated by the symlink. The old code always complained that the
74 * name already exists, due to not following the symlink even if its target
75 * is nonexistent. The new semantics affects also mknod() and link() when
Lucas De Marchi25985ed2011-03-30 22:57:33 -030076 * the name is a symlink pointing to a non-existent name.
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 *
78 * I don't know which semantics is the right one, since I have no access
79 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
80 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
81 * "old" one. Personally, I think the new semantics is much more logical.
82 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
83 * file does succeed in both HP-UX and SunOs, but not in Solaris
84 * and in the old Linux semantics.
85 */
86
87/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
88 * semantics. See the comments in "open_namei" and "do_link" below.
89 *
90 * [10-Sep-98 Alan Modra] Another symlink change.
91 */
92
93/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
94 * inside the path - always follow.
95 * in the last component in creation/removal/renaming - never follow.
96 * if LOOKUP_FOLLOW passed - follow.
97 * if the pathname has trailing slashes - follow.
98 * otherwise - don't follow.
99 * (applied in that order).
100 *
101 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
102 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
103 * During the 2.4 we need to fix the userland stuff depending on it -
104 * hopefully we will be able to get rid of that wart in 2.5. So far only
105 * XEmacs seems to be relying on it...
106 */
107/*
108 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800109 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 * any extra contention...
111 */
112
113/* In order to reduce some races, while at the same time doing additional
114 * checking and hopefully speeding things up, we copy filenames to the
115 * kernel data space before using them..
116 *
117 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
118 * PATH_MAX includes the nul terminator --RR.
119 */
Jeff Layton91a27b22012-10-10 15:25:28 -0400120void final_putname(struct filename *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121{
Jeff Layton7950e382012-10-10 16:43:13 -0400122 if (name->separate) {
123 __putname(name->name);
124 kfree(name);
125 } else {
126 __putname(name);
127 }
Jeff Layton91a27b22012-10-10 15:25:28 -0400128}
129
Jeff Layton7950e382012-10-10 16:43:13 -0400130#define EMBEDDED_NAME_MAX (PATH_MAX - sizeof(struct filename))
131
Jeff Layton91a27b22012-10-10 15:25:28 -0400132static struct filename *
133getname_flags(const char __user *filename, int flags, int *empty)
134{
135 struct filename *result, *err;
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700136 int len;
Jeff Layton7950e382012-10-10 16:43:13 -0400137 long max;
138 char *kname;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139
Jeff Layton7ac86262012-10-10 15:25:28 -0400140 result = audit_reusename(filename);
141 if (result)
142 return result;
143
Jeff Layton7950e382012-10-10 16:43:13 -0400144 result = __getname();
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700145 if (unlikely(!result))
Eric Paris4043cde2012-01-03 14:23:08 -0500146 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147
Jeff Layton7950e382012-10-10 16:43:13 -0400148 /*
149 * First, try to embed the struct filename inside the names_cache
150 * allocation
151 */
152 kname = (char *)result + sizeof(*result);
Jeff Layton91a27b22012-10-10 15:25:28 -0400153 result->name = kname;
Jeff Layton7950e382012-10-10 16:43:13 -0400154 result->separate = false;
155 max = EMBEDDED_NAME_MAX;
156
157recopy:
158 len = strncpy_from_user(kname, filename, max);
Jeff Layton91a27b22012-10-10 15:25:28 -0400159 if (unlikely(len < 0)) {
160 err = ERR_PTR(len);
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700161 goto error;
Jeff Layton91a27b22012-10-10 15:25:28 -0400162 }
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700163
Jeff Layton7950e382012-10-10 16:43:13 -0400164 /*
165 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
166 * separate struct filename so we can dedicate the entire
167 * names_cache allocation for the pathname, and re-do the copy from
168 * userland.
169 */
170 if (len == EMBEDDED_NAME_MAX && max == EMBEDDED_NAME_MAX) {
171 kname = (char *)result;
172
173 result = kzalloc(sizeof(*result), GFP_KERNEL);
174 if (!result) {
175 err = ERR_PTR(-ENOMEM);
176 result = (struct filename *)kname;
177 goto error;
178 }
179 result->name = kname;
180 result->separate = true;
181 max = PATH_MAX;
182 goto recopy;
183 }
184
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700185 /* The empty path is special. */
186 if (unlikely(!len)) {
187 if (empty)
Eric Paris4043cde2012-01-03 14:23:08 -0500188 *empty = 1;
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700189 err = ERR_PTR(-ENOENT);
190 if (!(flags & LOOKUP_EMPTY))
191 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 }
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700193
194 err = ERR_PTR(-ENAMETOOLONG);
Jeff Layton7950e382012-10-10 16:43:13 -0400195 if (unlikely(len >= PATH_MAX))
196 goto error;
197
198 result->uptr = filename;
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800199 result->aname = NULL;
Jeff Layton7950e382012-10-10 16:43:13 -0400200 audit_getname(result);
201 return result;
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700202
203error:
Jeff Layton7950e382012-10-10 16:43:13 -0400204 final_putname(result);
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700205 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206}
207
Jeff Layton91a27b22012-10-10 15:25:28 -0400208struct filename *
209getname(const char __user * filename)
Al Virof52e0c12011-03-14 18:56:51 -0400210{
Linus Torvaldsf7493e52012-03-22 16:10:40 -0700211 return getname_flags(filename, 0, NULL);
Al Virof52e0c12011-03-14 18:56:51 -0400212}
213
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800214/*
215 * The "getname_kernel()" interface doesn't do pathnames longer
216 * than EMBEDDED_NAME_MAX. Deal with it - you're a kernel user.
217 */
218struct filename *
219getname_kernel(const char * filename)
220{
221 struct filename *result;
222 char *kname;
223 int len;
224
225 len = strlen(filename);
226 if (len >= EMBEDDED_NAME_MAX)
227 return ERR_PTR(-ENAMETOOLONG);
228
229 result = __getname();
230 if (unlikely(!result))
231 return ERR_PTR(-ENOMEM);
232
233 kname = (char *)result + sizeof(*result);
234 result->name = kname;
235 result->uptr = NULL;
236 result->aname = NULL;
237 result->separate = false;
238
239 strlcpy(kname, filename, EMBEDDED_NAME_MAX);
240 return result;
241}
242
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243#ifdef CONFIG_AUDITSYSCALL
Jeff Layton91a27b22012-10-10 15:25:28 -0400244void putname(struct filename *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245{
Al Viro5ac3a9c2006-07-16 06:38:45 -0400246 if (unlikely(!audit_dummy_context()))
Jeff Layton91a27b22012-10-10 15:25:28 -0400247 return audit_putname(name);
248 final_putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250#endif
251
Linus Torvaldse77819e2011-07-22 19:30:19 -0700252static int check_acl(struct inode *inode, int mask)
253{
Linus Torvalds84635d62011-07-25 22:47:03 -0700254#ifdef CONFIG_FS_POSIX_ACL
Linus Torvaldse77819e2011-07-22 19:30:19 -0700255 struct posix_acl *acl;
256
Linus Torvaldse77819e2011-07-22 19:30:19 -0700257 if (mask & MAY_NOT_BLOCK) {
Al Viro35678662011-08-02 21:32:13 -0400258 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
259 if (!acl)
Linus Torvaldse77819e2011-07-22 19:30:19 -0700260 return -EAGAIN;
Al Viro35678662011-08-02 21:32:13 -0400261 /* no ->get_acl() calls in RCU mode... */
262 if (acl == ACL_NOT_CACHED)
263 return -ECHILD;
Ari Savolainen206b1d02011-08-06 19:43:07 +0300264 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
Linus Torvaldse77819e2011-07-22 19:30:19 -0700265 }
266
Christoph Hellwig2982baa2013-12-20 05:16:38 -0800267 acl = get_acl(inode, ACL_TYPE_ACCESS);
268 if (IS_ERR(acl))
269 return PTR_ERR(acl);
Linus Torvaldse77819e2011-07-22 19:30:19 -0700270 if (acl) {
271 int error = posix_acl_permission(inode, acl, mask);
272 posix_acl_release(acl);
273 return error;
274 }
Linus Torvalds84635d62011-07-25 22:47:03 -0700275#endif
Linus Torvaldse77819e2011-07-22 19:30:19 -0700276
277 return -EAGAIN;
278}
279
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700280/*
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530281 * This does the basic permission checking
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700282 */
Al Viro7e401452011-06-20 19:12:17 -0400283static int acl_permission_check(struct inode *inode, int mask)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700284{
Linus Torvalds26cf46b2011-05-13 11:51:01 -0700285 unsigned int mode = inode->i_mode;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700286
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -0800287 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700288 mode >>= 6;
289 else {
Linus Torvaldse77819e2011-07-22 19:30:19 -0700290 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
Al Viro7e401452011-06-20 19:12:17 -0400291 int error = check_acl(inode, mask);
Nick Pigginb74c79e2011-01-07 17:49:58 +1100292 if (error != -EAGAIN)
293 return error;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700294 }
295
296 if (in_group_p(inode->i_gid))
297 mode >>= 3;
298 }
299
300 /*
301 * If the DACs are ok we don't need any capability check.
302 */
Al Viro9c2c7032011-06-20 19:06:22 -0400303 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700304 return 0;
305 return -EACCES;
306}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307
308/**
Nick Pigginb74c79e2011-01-07 17:49:58 +1100309 * generic_permission - check for access rights on a Posix-like filesystem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 * @inode: inode to check access rights for
Andreas Gruenbacher8fd90c82011-10-23 23:13:30 +0530311 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 *
313 * Used to check for read/write/execute permissions on a file.
314 * We use "fsuid" for this, letting us set arbitrary permissions
315 * for filesystem access without changing the "normal" uids which
Nick Pigginb74c79e2011-01-07 17:49:58 +1100316 * are used for other things.
317 *
318 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
319 * request cannot be satisfied (eg. requires blocking or too much complexity).
320 * It would then be called again in ref-walk mode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 */
Al Viro2830ba72011-06-20 19:16:29 -0400322int generic_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700324 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325
326 /*
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530327 * Do the basic permission checks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 */
Al Viro7e401452011-06-20 19:12:17 -0400329 ret = acl_permission_check(inode, mask);
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700330 if (ret != -EACCES)
331 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332
Al Virod594e7e2011-06-20 19:55:42 -0400333 if (S_ISDIR(inode->i_mode)) {
334 /* DACs are overridable for directories */
Eric W. Biederman1a48e2a2011-11-14 16:24:06 -0800335 if (inode_capable(inode, CAP_DAC_OVERRIDE))
Al Virod594e7e2011-06-20 19:55:42 -0400336 return 0;
337 if (!(mask & MAY_WRITE))
Eric W. Biederman1a48e2a2011-11-14 16:24:06 -0800338 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
Al Virod594e7e2011-06-20 19:55:42 -0400339 return 0;
340 return -EACCES;
341 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 /*
343 * Read/write DACs are always overridable.
Al Virod594e7e2011-06-20 19:55:42 -0400344 * Executable DACs are overridable when there is
345 * at least one exec bit set.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 */
Al Virod594e7e2011-06-20 19:55:42 -0400347 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
Eric W. Biederman1a48e2a2011-11-14 16:24:06 -0800348 if (inode_capable(inode, CAP_DAC_OVERRIDE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 return 0;
350
351 /*
352 * Searching includes executable on directories, else just read.
353 */
Serge E. Hallyn7ea66002009-12-29 14:50:19 -0600354 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
Al Virod594e7e2011-06-20 19:55:42 -0400355 if (mask == MAY_READ)
Eric W. Biederman1a48e2a2011-11-14 16:24:06 -0800356 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 return 0;
358
359 return -EACCES;
360}
361
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700362/*
363 * We _really_ want to just do "generic_permission()" without
364 * even looking at the inode->i_op values. So we keep a cache
365 * flag in inode->i_opflags, that says "this has not special
366 * permission function, use the fast case".
367 */
368static inline int do_inode_permission(struct inode *inode, int mask)
369{
370 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
371 if (likely(inode->i_op->permission))
372 return inode->i_op->permission(inode, mask);
373
374 /* This gets set once for the inode lifetime */
375 spin_lock(&inode->i_lock);
376 inode->i_opflags |= IOP_FASTPERM;
377 spin_unlock(&inode->i_lock);
378 }
379 return generic_permission(inode, mask);
380}
381
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200382/**
David Howells0bdaea92012-06-25 12:55:46 +0100383 * __inode_permission - Check for access rights to a given inode
384 * @inode: Inode to check permission on
385 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200386 *
David Howells0bdaea92012-06-25 12:55:46 +0100387 * Check for read/write/execute permissions on an inode.
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530388 *
389 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
David Howells0bdaea92012-06-25 12:55:46 +0100390 *
391 * This does not check for a read-only file system. You probably want
392 * inode_permission().
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200393 */
David Howells0bdaea92012-06-25 12:55:46 +0100394int __inode_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395{
Al Viroe6305c42008-07-15 21:03:57 -0400396 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700398 if (unlikely(mask & MAY_WRITE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 /*
400 * Nobody gets write access to an immutable file.
401 */
402 if (IS_IMMUTABLE(inode))
403 return -EACCES;
404 }
405
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700406 retval = do_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 if (retval)
408 return retval;
409
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -0700410 retval = devcgroup_inode_permission(inode, mask);
411 if (retval)
412 return retval;
413
Eric Parisd09ca732010-07-23 11:43:57 -0400414 return security_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415}
416
Al Virof4d6ff82011-06-19 13:14:21 -0400417/**
David Howells0bdaea92012-06-25 12:55:46 +0100418 * sb_permission - Check superblock-level permissions
419 * @sb: Superblock of inode to check permission on
Randy Dunlap55852632012-08-18 17:39:25 -0700420 * @inode: Inode to check permission on
David Howells0bdaea92012-06-25 12:55:46 +0100421 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
422 *
423 * Separate out file-system wide checks from inode-specific permission checks.
424 */
425static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
426{
427 if (unlikely(mask & MAY_WRITE)) {
428 umode_t mode = inode->i_mode;
429
430 /* Nobody gets write access to a read-only fs. */
431 if ((sb->s_flags & MS_RDONLY) &&
432 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
433 return -EROFS;
434 }
435 return 0;
436}
437
438/**
439 * inode_permission - Check for access rights to a given inode
440 * @inode: Inode to check permission on
441 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
442 *
443 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
444 * this, letting us set arbitrary permissions for filesystem access without
445 * changing the "normal" UIDs which are used for other things.
446 *
447 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
448 */
449int inode_permission(struct inode *inode, int mask)
450{
451 int retval;
452
453 retval = sb_permission(inode->i_sb, inode, mask);
454 if (retval)
455 return retval;
456 return __inode_permission(inode, mask);
457}
458
459/**
Jan Blunck5dd784d02008-02-14 19:34:38 -0800460 * path_get - get a reference to a path
461 * @path: path to get the reference to
462 *
463 * Given a path increment the reference count to the dentry and the vfsmount.
464 */
Al Virodcf787f2013-03-01 23:51:07 -0500465void path_get(const struct path *path)
Jan Blunck5dd784d02008-02-14 19:34:38 -0800466{
467 mntget(path->mnt);
468 dget(path->dentry);
469}
470EXPORT_SYMBOL(path_get);
471
472/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800473 * path_put - put a reference to a path
474 * @path: path to put the reference to
475 *
476 * Given a path decrement the reference count to the dentry and the vfsmount.
477 */
Al Virodcf787f2013-03-01 23:51:07 -0500478void path_put(const struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479{
Jan Blunck1d957f92008-02-14 19:34:35 -0800480 dput(path->dentry);
481 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482}
Jan Blunck1d957f92008-02-14 19:34:35 -0800483EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484
Al Viro19660af2011-03-25 10:32:48 -0400485/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100486 * Path walking has 2 modes, rcu-walk and ref-walk (see
Al Viro19660af2011-03-25 10:32:48 -0400487 * Documentation/filesystems/path-lookup.txt). In situations when we can't
488 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
489 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
490 * mode. Refcounts are grabbed at the last known good point before rcu-walk
491 * got stuck, so ref-walk may continue from there. If this is not successful
492 * (eg. a seqcount has changed), then failure is returned and it's up to caller
493 * to restart the path walk from the beginning in ref-walk mode.
Nick Piggin31e6b012011-01-07 17:49:52 +1100494 */
Nick Piggin31e6b012011-01-07 17:49:52 +1100495
496/**
Al Viro19660af2011-03-25 10:32:48 -0400497 * unlazy_walk - try to switch to ref-walk mode.
498 * @nd: nameidata pathwalk data
499 * @dentry: child of nd->path.dentry or NULL
Randy Dunlap39191622011-01-08 19:36:21 -0800500 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100501 *
Al Viro19660af2011-03-25 10:32:48 -0400502 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
503 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
504 * @nd or NULL. Must be called from rcu-walk context.
Nick Piggin31e6b012011-01-07 17:49:52 +1100505 */
Al Viro19660af2011-03-25 10:32:48 -0400506static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
Nick Piggin31e6b012011-01-07 17:49:52 +1100507{
508 struct fs_struct *fs = current->fs;
509 struct dentry *parent = nd->path.dentry;
510
511 BUG_ON(!(nd->flags & LOOKUP_RCU));
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700512
513 /*
Al Viro48a066e2013-09-29 22:06:07 -0400514 * After legitimizing the bastards, terminate_walk()
515 * will do the right thing for non-RCU mode, and all our
516 * subsequent exit cases should rcu_read_unlock()
517 * before returning. Do vfsmount first; if dentry
518 * can't be legitimized, just set nd->path.dentry to NULL
519 * and rely on dput(NULL) being a no-op.
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700520 */
Al Viro48a066e2013-09-29 22:06:07 -0400521 if (!legitimize_mnt(nd->path.mnt, nd->m_seq))
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700522 return -ECHILD;
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700523 nd->flags &= ~LOOKUP_RCU;
Linus Torvalds15570082013-09-02 11:38:06 -0700524
Al Viro48a066e2013-09-29 22:06:07 -0400525 if (!lockref_get_not_dead(&parent->d_lockref)) {
526 nd->path.dentry = NULL;
Al Virod870b4a2013-11-29 01:48:32 -0500527 goto out;
Al Viro48a066e2013-09-29 22:06:07 -0400528 }
529
Linus Torvalds15570082013-09-02 11:38:06 -0700530 /*
531 * For a negative lookup, the lookup sequence point is the parents
532 * sequence point, and it only needs to revalidate the parent dentry.
533 *
534 * For a positive lookup, we need to move both the parent and the
535 * dentry from the RCU domain to be properly refcounted. And the
536 * sequence number in the dentry validates *both* dentry counters,
537 * since we checked the sequence number of the parent after we got
538 * the child sequence number. So we know the parent must still
539 * be valid if the child sequence number is still valid.
540 */
Al Viro19660af2011-03-25 10:32:48 -0400541 if (!dentry) {
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700542 if (read_seqcount_retry(&parent->d_seq, nd->seq))
543 goto out;
Al Viro19660af2011-03-25 10:32:48 -0400544 BUG_ON(nd->inode != parent->d_inode);
545 } else {
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700546 if (!lockref_get_not_dead(&dentry->d_lockref))
547 goto out;
548 if (read_seqcount_retry(&dentry->d_seq, nd->seq))
549 goto drop_dentry;
Al Viro19660af2011-03-25 10:32:48 -0400550 }
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700551
552 /*
553 * Sequence counts matched. Now make sure that the root is
554 * still valid and get it if required.
555 */
556 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
557 spin_lock(&fs->lock);
558 if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry)
559 goto unlock_and_drop_dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +1100560 path_get(&nd->root);
561 spin_unlock(&fs->lock);
562 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100563
Al Viro8b61e742013-11-08 12:45:01 -0500564 rcu_read_unlock();
Nick Piggin31e6b012011-01-07 17:49:52 +1100565 return 0;
Al Viro19660af2011-03-25 10:32:48 -0400566
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700567unlock_and_drop_dentry:
568 spin_unlock(&fs->lock);
569drop_dentry:
Al Viro8b61e742013-11-08 12:45:01 -0500570 rcu_read_unlock();
Linus Torvalds15570082013-09-02 11:38:06 -0700571 dput(dentry);
Linus Torvaldsd0d27272013-09-10 12:17:49 -0700572 goto drop_root_mnt;
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700573out:
Al Viro8b61e742013-11-08 12:45:01 -0500574 rcu_read_unlock();
Linus Torvaldsd0d27272013-09-10 12:17:49 -0700575drop_root_mnt:
576 if (!(nd->flags & LOOKUP_ROOT))
577 nd->root.mnt = NULL;
Nick Piggin31e6b012011-01-07 17:49:52 +1100578 return -ECHILD;
579}
580
Al Viro4ce16ef32012-06-10 16:10:59 -0400581static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
Nick Piggin34286d62011-01-07 17:49:57 +1100582{
Al Viro4ce16ef32012-06-10 16:10:59 -0400583 return dentry->d_op->d_revalidate(dentry, flags);
Nick Piggin34286d62011-01-07 17:49:57 +1100584}
585
Al Viro9f1fafe2011-03-25 11:00:12 -0400586/**
587 * complete_walk - successful completion of path walk
588 * @nd: pointer nameidata
Jeff Layton39159de2009-12-07 12:01:50 -0500589 *
Al Viro9f1fafe2011-03-25 11:00:12 -0400590 * If we had been in RCU mode, drop out of it and legitimize nd->path.
591 * Revalidate the final result, unless we'd already done that during
592 * the path walk or the filesystem doesn't ask for it. Return 0 on
593 * success, -error on failure. In case of failure caller does not
594 * need to drop nd->path.
Jeff Layton39159de2009-12-07 12:01:50 -0500595 */
Al Viro9f1fafe2011-03-25 11:00:12 -0400596static int complete_walk(struct nameidata *nd)
Jeff Layton39159de2009-12-07 12:01:50 -0500597{
Al Viro16c2cd72011-02-22 15:50:10 -0500598 struct dentry *dentry = nd->path.dentry;
Jeff Layton39159de2009-12-07 12:01:50 -0500599 int status;
Jeff Layton39159de2009-12-07 12:01:50 -0500600
Al Viro9f1fafe2011-03-25 11:00:12 -0400601 if (nd->flags & LOOKUP_RCU) {
602 nd->flags &= ~LOOKUP_RCU;
603 if (!(nd->flags & LOOKUP_ROOT))
604 nd->root.mnt = NULL;
Linus Torvalds15570082013-09-02 11:38:06 -0700605
Al Viro48a066e2013-09-29 22:06:07 -0400606 if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) {
Al Viro8b61e742013-11-08 12:45:01 -0500607 rcu_read_unlock();
Al Viro48a066e2013-09-29 22:06:07 -0400608 return -ECHILD;
609 }
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700610 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) {
Al Viro8b61e742013-11-08 12:45:01 -0500611 rcu_read_unlock();
Al Viro48a066e2013-09-29 22:06:07 -0400612 mntput(nd->path.mnt);
Al Viro9f1fafe2011-03-25 11:00:12 -0400613 return -ECHILD;
614 }
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700615 if (read_seqcount_retry(&dentry->d_seq, nd->seq)) {
Al Viro8b61e742013-11-08 12:45:01 -0500616 rcu_read_unlock();
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700617 dput(dentry);
Al Viro48a066e2013-09-29 22:06:07 -0400618 mntput(nd->path.mnt);
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700619 return -ECHILD;
620 }
Al Viro8b61e742013-11-08 12:45:01 -0500621 rcu_read_unlock();
Al Viro9f1fafe2011-03-25 11:00:12 -0400622 }
623
Al Viro16c2cd72011-02-22 15:50:10 -0500624 if (likely(!(nd->flags & LOOKUP_JUMPED)))
Jeff Layton39159de2009-12-07 12:01:50 -0500625 return 0;
626
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500627 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
Al Viro16c2cd72011-02-22 15:50:10 -0500628 return 0;
629
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500630 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
Jeff Layton39159de2009-12-07 12:01:50 -0500631 if (status > 0)
632 return 0;
633
Al Viro16c2cd72011-02-22 15:50:10 -0500634 if (!status)
Jeff Layton39159de2009-12-07 12:01:50 -0500635 status = -ESTALE;
Al Viro16c2cd72011-02-22 15:50:10 -0500636
Al Viro9f1fafe2011-03-25 11:00:12 -0400637 path_put(&nd->path);
Jeff Layton39159de2009-12-07 12:01:50 -0500638 return status;
639}
640
Al Viro2a737872009-04-07 11:49:53 -0400641static __always_inline void set_root(struct nameidata *nd)
642{
Miklos Szeredif7ad3c62010-08-10 11:41:36 +0200643 if (!nd->root.mnt)
644 get_fs_root(current->fs, &nd->root);
Al Viro2a737872009-04-07 11:49:53 -0400645}
646
Al Viro6de88d72009-08-09 01:41:57 +0400647static int link_path_walk(const char *, struct nameidata *);
648
Nick Piggin31e6b012011-01-07 17:49:52 +1100649static __always_inline void set_root_rcu(struct nameidata *nd)
650{
651 if (!nd->root.mnt) {
652 struct fs_struct *fs = current->fs;
Nick Pigginc28cc362011-01-07 17:49:53 +1100653 unsigned seq;
654
655 do {
656 seq = read_seqcount_begin(&fs->seq);
657 nd->root = fs->root;
Tim Chenc1530012011-04-15 11:39:29 -0700658 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
Nick Pigginc28cc362011-01-07 17:49:53 +1100659 } while (read_seqcount_retry(&fs->seq, seq));
Nick Piggin31e6b012011-01-07 17:49:52 +1100660 }
661}
662
Jan Blunck1d957f92008-02-14 19:34:35 -0800663static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700664{
665 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800666 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700667 mntput(path->mnt);
668}
669
Nick Piggin7b9337a2011-01-14 08:42:43 +0000670static inline void path_to_nameidata(const struct path *path,
671 struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700672{
Nick Piggin31e6b012011-01-07 17:49:52 +1100673 if (!(nd->flags & LOOKUP_RCU)) {
674 dput(nd->path.dentry);
675 if (nd->path.mnt != path->mnt)
676 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800677 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100678 nd->path.mnt = path->mnt;
Jan Blunck4ac91372008-02-14 19:34:32 -0800679 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700680}
681
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400682/*
683 * Helper to directly jump to a known parsed path from ->follow_link,
684 * caller must have taken a reference to path beforehand.
685 */
686void nd_jump_link(struct nameidata *nd, struct path *path)
687{
688 path_put(&nd->path);
689
690 nd->path = *path;
691 nd->inode = nd->path.dentry->d_inode;
692 nd->flags |= LOOKUP_JUMPED;
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400693}
694
Al Viro574197e2011-03-14 22:20:34 -0400695static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
696{
697 struct inode *inode = link->dentry->d_inode;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400698 if (inode->i_op->put_link)
Al Viro574197e2011-03-14 22:20:34 -0400699 inode->i_op->put_link(link->dentry, nd, cookie);
700 path_put(link);
701}
702
Linus Torvalds561ec642012-10-26 10:05:07 -0700703int sysctl_protected_symlinks __read_mostly = 0;
704int sysctl_protected_hardlinks __read_mostly = 0;
Kees Cook800179c2012-07-25 17:29:07 -0700705
706/**
707 * may_follow_link - Check symlink following for unsafe situations
708 * @link: The path of the symlink
Randy Dunlap55852632012-08-18 17:39:25 -0700709 * @nd: nameidata pathwalk data
Kees Cook800179c2012-07-25 17:29:07 -0700710 *
711 * In the case of the sysctl_protected_symlinks sysctl being enabled,
712 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
713 * in a sticky world-writable directory. This is to protect privileged
714 * processes from failing races against path names that may change out
715 * from under them by way of other users creating malicious symlinks.
716 * It will permit symlinks to be followed only when outside a sticky
717 * world-writable directory, or when the uid of the symlink and follower
718 * match, or when the directory owner matches the symlink's owner.
719 *
720 * Returns 0 if following the symlink is allowed, -ve on error.
721 */
722static inline int may_follow_link(struct path *link, struct nameidata *nd)
723{
724 const struct inode *inode;
725 const struct inode *parent;
726
727 if (!sysctl_protected_symlinks)
728 return 0;
729
730 /* Allowed if owner and follower match. */
731 inode = link->dentry->d_inode;
Eric W. Biederman81abe272012-08-03 09:38:08 -0700732 if (uid_eq(current_cred()->fsuid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700733 return 0;
734
735 /* Allowed if parent directory not sticky and world-writable. */
736 parent = nd->path.dentry->d_inode;
737 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
738 return 0;
739
740 /* Allowed if parent directory and link owner match. */
Eric W. Biederman81abe272012-08-03 09:38:08 -0700741 if (uid_eq(parent->i_uid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700742 return 0;
743
Sasha Levinffd8d102012-10-04 19:56:40 -0400744 audit_log_link_denied("follow_link", link);
Kees Cook800179c2012-07-25 17:29:07 -0700745 path_put_conditional(link, nd);
746 path_put(&nd->path);
747 return -EACCES;
748}
749
750/**
751 * safe_hardlink_source - Check for safe hardlink conditions
752 * @inode: the source inode to hardlink from
753 *
754 * Return false if at least one of the following conditions:
755 * - inode is not a regular file
756 * - inode is setuid
757 * - inode is setgid and group-exec
758 * - access failure for read and write
759 *
760 * Otherwise returns true.
761 */
762static bool safe_hardlink_source(struct inode *inode)
763{
764 umode_t mode = inode->i_mode;
765
766 /* Special files should not get pinned to the filesystem. */
767 if (!S_ISREG(mode))
768 return false;
769
770 /* Setuid files should not get pinned to the filesystem. */
771 if (mode & S_ISUID)
772 return false;
773
774 /* Executable setgid files should not get pinned to the filesystem. */
775 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
776 return false;
777
778 /* Hardlinking to unreadable or unwritable sources is dangerous. */
779 if (inode_permission(inode, MAY_READ | MAY_WRITE))
780 return false;
781
782 return true;
783}
784
785/**
786 * may_linkat - Check permissions for creating a hardlink
787 * @link: the source to hardlink from
788 *
789 * Block hardlink when all of:
790 * - sysctl_protected_hardlinks enabled
791 * - fsuid does not match inode
792 * - hardlink source is unsafe (see safe_hardlink_source() above)
793 * - not CAP_FOWNER
794 *
795 * Returns 0 if successful, -ve on error.
796 */
797static int may_linkat(struct path *link)
798{
799 const struct cred *cred;
800 struct inode *inode;
801
802 if (!sysctl_protected_hardlinks)
803 return 0;
804
805 cred = current_cred();
806 inode = link->dentry->d_inode;
807
808 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
809 * otherwise, it must be a safe source.
810 */
Eric W. Biederman81abe272012-08-03 09:38:08 -0700811 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
Kees Cook800179c2012-07-25 17:29:07 -0700812 capable(CAP_FOWNER))
813 return 0;
814
Kees Cooka51d9ea2012-07-25 17:29:08 -0700815 audit_log_link_denied("linkat", link);
Kees Cook800179c2012-07-25 17:29:07 -0700816 return -EPERM;
817}
818
Al Virodef4af32009-12-26 08:37:05 -0500819static __always_inline int
Al Viro574197e2011-03-14 22:20:34 -0400820follow_link(struct path *link, struct nameidata *nd, void **p)
Ian Kent051d3812006-03-27 01:14:53 -0800821{
Nick Piggin7b9337a2011-01-14 08:42:43 +0000822 struct dentry *dentry = link->dentry;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400823 int error;
824 char *s;
Ian Kent051d3812006-03-27 01:14:53 -0800825
Al Viro844a3912011-02-15 00:38:26 -0500826 BUG_ON(nd->flags & LOOKUP_RCU);
827
Al Viro0e794582011-03-16 02:45:02 -0400828 if (link->mnt == nd->path.mnt)
829 mntget(link->mnt);
830
Al Viro6d7b5aa2012-06-10 04:15:17 -0400831 error = -ELOOP;
832 if (unlikely(current->total_link_count >= 40))
833 goto out_put_nd_path;
834
Al Viro574197e2011-03-14 22:20:34 -0400835 cond_resched();
836 current->total_link_count++;
837
Al Viro68ac1232012-03-15 08:21:57 -0400838 touch_atime(link);
Ian Kent051d3812006-03-27 01:14:53 -0800839 nd_set_link(nd, NULL);
840
Al Viro36f3b4f2011-02-22 21:24:38 -0500841 error = security_inode_follow_link(link->dentry, nd);
Al Viro6d7b5aa2012-06-10 04:15:17 -0400842 if (error)
843 goto out_put_nd_path;
Al Viro36f3b4f2011-02-22 21:24:38 -0500844
Al Viro86acdca12009-12-22 23:45:11 -0500845 nd->last_type = LAST_BIND;
Al Virodef4af32009-12-26 08:37:05 -0500846 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
847 error = PTR_ERR(*p);
Al Viro6d7b5aa2012-06-10 04:15:17 -0400848 if (IS_ERR(*p))
Christoph Hellwig408ef012012-06-18 10:47:03 -0400849 goto out_put_nd_path;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400850
851 error = 0;
852 s = nd_get_link(nd);
853 if (s) {
Al Viro443ed252013-09-10 12:00:43 -0400854 if (unlikely(IS_ERR(s))) {
855 path_put(&nd->path);
856 put_link(nd, link, *p);
857 return PTR_ERR(s);
858 }
859 if (*s == '/') {
860 set_root(nd);
861 path_put(&nd->path);
862 nd->path = nd->root;
863 path_get(&nd->root);
864 nd->flags |= LOOKUP_JUMPED;
865 }
866 nd->inode = nd->path.dentry->d_inode;
867 error = link_path_walk(s, nd);
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400868 if (unlikely(error))
869 put_link(nd, link, *p);
Ian Kent051d3812006-03-27 01:14:53 -0800870 }
Al Viro6d7b5aa2012-06-10 04:15:17 -0400871
872 return error;
873
874out_put_nd_path:
Arnd Bergmann98f6ef62012-10-11 13:20:00 +0000875 *p = NULL;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400876 path_put(&nd->path);
Al Viro6d7b5aa2012-06-10 04:15:17 -0400877 path_put(link);
Ian Kent051d3812006-03-27 01:14:53 -0800878 return error;
879}
880
Nick Piggin31e6b012011-01-07 17:49:52 +1100881static int follow_up_rcu(struct path *path)
882{
Al Viro0714a532011-11-24 22:19:58 -0500883 struct mount *mnt = real_mount(path->mnt);
884 struct mount *parent;
Nick Piggin31e6b012011-01-07 17:49:52 +1100885 struct dentry *mountpoint;
886
Al Viro0714a532011-11-24 22:19:58 -0500887 parent = mnt->mnt_parent;
888 if (&parent->mnt == path->mnt)
Nick Piggin31e6b012011-01-07 17:49:52 +1100889 return 0;
Al Viroa73324d2011-11-24 22:25:07 -0500890 mountpoint = mnt->mnt_mountpoint;
Nick Piggin31e6b012011-01-07 17:49:52 +1100891 path->dentry = mountpoint;
Al Viro0714a532011-11-24 22:19:58 -0500892 path->mnt = &parent->mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +1100893 return 1;
894}
895
David Howellsf015f1262012-06-25 12:55:28 +0100896/*
897 * follow_up - Find the mountpoint of path's vfsmount
898 *
899 * Given a path, find the mountpoint of its source file system.
900 * Replace @path with the path of the mountpoint in the parent mount.
901 * Up is towards /.
902 *
903 * Return 1 if we went up a level and 0 if we were already at the
904 * root.
905 */
Al Virobab77eb2009-04-18 03:26:48 -0400906int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907{
Al Viro0714a532011-11-24 22:19:58 -0500908 struct mount *mnt = real_mount(path->mnt);
909 struct mount *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +1000911
Al Viro48a066e2013-09-29 22:06:07 -0400912 read_seqlock_excl(&mount_lock);
Al Viro0714a532011-11-24 22:19:58 -0500913 parent = mnt->mnt_parent;
Al Viro3c0a6162012-07-18 17:32:50 +0400914 if (parent == mnt) {
Al Viro48a066e2013-09-29 22:06:07 -0400915 read_sequnlock_excl(&mount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 return 0;
917 }
Al Viro0714a532011-11-24 22:19:58 -0500918 mntget(&parent->mnt);
Al Viroa73324d2011-11-24 22:25:07 -0500919 mountpoint = dget(mnt->mnt_mountpoint);
Al Viro48a066e2013-09-29 22:06:07 -0400920 read_sequnlock_excl(&mount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400921 dput(path->dentry);
922 path->dentry = mountpoint;
923 mntput(path->mnt);
Al Viro0714a532011-11-24 22:19:58 -0500924 path->mnt = &parent->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 return 1;
926}
927
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100928/*
David Howells9875cf82011-01-14 18:45:21 +0000929 * Perform an automount
930 * - return -EISDIR to tell follow_managed() to stop and return the path we
931 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 */
David Howells9875cf82011-01-14 18:45:21 +0000933static int follow_automount(struct path *path, unsigned flags,
934 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +1100935{
David Howells9875cf82011-01-14 18:45:21 +0000936 struct vfsmount *mnt;
David Howellsea5b7782011-01-14 19:10:03 +0000937 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +1100938
David Howells9875cf82011-01-14 18:45:21 +0000939 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
940 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -0700941
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200942 /* We don't want to mount if someone's just doing a stat -
943 * unless they're stat'ing a directory and appended a '/' to
944 * the name.
945 *
946 * We do, however, want to mount if someone wants to open or
947 * create a file of any type under the mountpoint, wants to
948 * traverse through the mountpoint or wants to open the
949 * mounted directory. Also, autofs may mark negative dentries
950 * as being automount points. These will need the attentions
951 * of the daemon to instantiate them before they can be used.
David Howells9875cf82011-01-14 18:45:21 +0000952 */
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200953 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
Linus Torvaldsd94c1772011-09-26 17:44:55 -0700954 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200955 path->dentry->d_inode)
956 return -EISDIR;
957
David Howells9875cf82011-01-14 18:45:21 +0000958 current->total_link_count++;
959 if (current->total_link_count >= 40)
960 return -ELOOP;
961
962 mnt = path->dentry->d_op->d_automount(path);
963 if (IS_ERR(mnt)) {
964 /*
965 * The filesystem is allowed to return -EISDIR here to indicate
966 * it doesn't want to automount. For instance, autofs would do
967 * this so that its userspace daemon can mount on this dentry.
968 *
969 * However, we can only permit this if it's a terminal point in
970 * the path being looked up; if it wasn't then the remainder of
971 * the path is inaccessible and we should say so.
972 */
Al Viro49084c32011-06-25 21:59:52 -0400973 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
David Howells9875cf82011-01-14 18:45:21 +0000974 return -EREMOTE;
975 return PTR_ERR(mnt);
976 }
David Howellsea5b7782011-01-14 19:10:03 +0000977
David Howells9875cf82011-01-14 18:45:21 +0000978 if (!mnt) /* mount collision */
979 return 0;
980
Al Viro8aef1882011-06-16 15:10:06 +0100981 if (!*need_mntput) {
982 /* lock_mount() may release path->mnt on error */
983 mntget(path->mnt);
984 *need_mntput = true;
985 }
Al Viro19a167a2011-01-17 01:35:23 -0500986 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +0000987
David Howellsea5b7782011-01-14 19:10:03 +0000988 switch (err) {
989 case -EBUSY:
990 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -0500991 return 0;
David Howellsea5b7782011-01-14 19:10:03 +0000992 case 0:
Al Viro8aef1882011-06-16 15:10:06 +0100993 path_put(path);
David Howellsea5b7782011-01-14 19:10:03 +0000994 path->mnt = mnt;
995 path->dentry = dget(mnt->mnt_root);
David Howellsea5b7782011-01-14 19:10:03 +0000996 return 0;
Al Viro19a167a2011-01-17 01:35:23 -0500997 default:
998 return err;
David Howellsea5b7782011-01-14 19:10:03 +0000999 }
Al Viro19a167a2011-01-17 01:35:23 -05001000
David Howells9875cf82011-01-14 18:45:21 +00001001}
1002
1003/*
1004 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +00001005 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +00001006 * - Flagged as mountpoint
1007 * - Flagged as automount point
1008 *
1009 * This may only be called in refwalk mode.
1010 *
1011 * Serialization is taken care of in namespace.c
1012 */
1013static int follow_managed(struct path *path, unsigned flags)
1014{
Al Viro8aef1882011-06-16 15:10:06 +01001015 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
David Howells9875cf82011-01-14 18:45:21 +00001016 unsigned managed;
1017 bool need_mntput = false;
Al Viro8aef1882011-06-16 15:10:06 +01001018 int ret = 0;
David Howells9875cf82011-01-14 18:45:21 +00001019
1020 /* Given that we're not holding a lock here, we retain the value in a
1021 * local variable for each dentry as we look at it so that we don't see
1022 * the components of that value change under us */
1023 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1024 managed &= DCACHE_MANAGED_DENTRY,
1025 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +00001026 /* Allow the filesystem to manage the transit without i_mutex
1027 * being held. */
1028 if (managed & DCACHE_MANAGE_TRANSIT) {
1029 BUG_ON(!path->dentry->d_op);
1030 BUG_ON(!path->dentry->d_op->d_manage);
Al Viro1aed3e42011-03-18 09:09:02 -04001031 ret = path->dentry->d_op->d_manage(path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001032 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001033 break;
David Howellscc53ce52011-01-14 18:45:26 +00001034 }
1035
David Howells9875cf82011-01-14 18:45:21 +00001036 /* Transit to a mounted filesystem. */
1037 if (managed & DCACHE_MOUNTED) {
1038 struct vfsmount *mounted = lookup_mnt(path);
1039 if (mounted) {
1040 dput(path->dentry);
1041 if (need_mntput)
1042 mntput(path->mnt);
1043 path->mnt = mounted;
1044 path->dentry = dget(mounted->mnt_root);
1045 need_mntput = true;
1046 continue;
1047 }
1048
1049 /* Something is mounted on this dentry in another
1050 * namespace and/or whatever was mounted there in this
Al Viro48a066e2013-09-29 22:06:07 -04001051 * namespace got unmounted before lookup_mnt() could
1052 * get it */
David Howells9875cf82011-01-14 18:45:21 +00001053 }
1054
1055 /* Handle an automount point */
1056 if (managed & DCACHE_NEED_AUTOMOUNT) {
1057 ret = follow_automount(path, flags, &need_mntput);
1058 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001059 break;
David Howells9875cf82011-01-14 18:45:21 +00001060 continue;
1061 }
1062
1063 /* We didn't change the current path point */
1064 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 }
Al Viro8aef1882011-06-16 15:10:06 +01001066
1067 if (need_mntput && path->mnt == mnt)
1068 mntput(path->mnt);
1069 if (ret == -EISDIR)
1070 ret = 0;
Al Viroa3fbbde2011-11-07 21:21:26 +00001071 return ret < 0 ? ret : need_mntput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072}
1073
David Howellscc53ce52011-01-14 18:45:26 +00001074int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075{
1076 struct vfsmount *mounted;
1077
Al Viro1c755af2009-04-18 14:06:57 -04001078 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -04001080 dput(path->dentry);
1081 mntput(path->mnt);
1082 path->mnt = mounted;
1083 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 return 1;
1085 }
1086 return 0;
1087}
1088
Ian Kent62a73752011-03-25 01:51:02 +08001089static inline bool managed_dentry_might_block(struct dentry *dentry)
1090{
1091 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
1092 dentry->d_op->d_manage(dentry, true) < 0);
1093}
1094
David Howells9875cf82011-01-14 18:45:21 +00001095/*
Al Viro287548e2011-05-27 06:50:06 -04001096 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1097 * we meet a managed dentry that would need blocking.
David Howells9875cf82011-01-14 18:45:21 +00001098 */
1099static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
Al Viro287548e2011-05-27 06:50:06 -04001100 struct inode **inode)
David Howells9875cf82011-01-14 18:45:21 +00001101{
Ian Kent62a73752011-03-25 01:51:02 +08001102 for (;;) {
Al Viroc7105362011-11-24 18:22:03 -05001103 struct mount *mounted;
Ian Kent62a73752011-03-25 01:51:02 +08001104 /*
1105 * Don't forget we might have a non-mountpoint managed dentry
1106 * that wants to block transit.
1107 */
Al Viro287548e2011-05-27 06:50:06 -04001108 if (unlikely(managed_dentry_might_block(path->dentry)))
David Howellsab909112011-01-14 18:46:51 +00001109 return false;
Ian Kent62a73752011-03-25 01:51:02 +08001110
1111 if (!d_mountpoint(path->dentry))
Al Virob37199e2014-03-20 15:18:22 -04001112 return true;
Ian Kent62a73752011-03-25 01:51:02 +08001113
Al Viro474279d2013-10-01 16:11:26 -04001114 mounted = __lookup_mnt(path->mnt, path->dentry);
David Howells9875cf82011-01-14 18:45:21 +00001115 if (!mounted)
1116 break;
Al Viroc7105362011-11-24 18:22:03 -05001117 path->mnt = &mounted->mnt;
1118 path->dentry = mounted->mnt.mnt_root;
Al Viroa3fbbde2011-11-07 21:21:26 +00001119 nd->flags |= LOOKUP_JUMPED;
David Howells9875cf82011-01-14 18:45:21 +00001120 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
Linus Torvalds59430262011-07-18 15:43:29 -07001121 /*
1122 * Update the inode too. We don't need to re-check the
1123 * dentry sequence number here after this d_inode read,
1124 * because a mount-point is always pinned.
1125 */
1126 *inode = path->dentry->d_inode;
David Howells9875cf82011-01-14 18:45:21 +00001127 }
Al Virob37199e2014-03-20 15:18:22 -04001128 return read_seqretry(&mount_lock, nd->m_seq);
Al Viro287548e2011-05-27 06:50:06 -04001129}
1130
Nick Piggin31e6b012011-01-07 17:49:52 +11001131static int follow_dotdot_rcu(struct nameidata *nd)
1132{
Nick Piggin31e6b012011-01-07 17:49:52 +11001133 set_root_rcu(nd);
1134
David Howells9875cf82011-01-14 18:45:21 +00001135 while (1) {
Nick Piggin31e6b012011-01-07 17:49:52 +11001136 if (nd->path.dentry == nd->root.dentry &&
1137 nd->path.mnt == nd->root.mnt) {
1138 break;
1139 }
1140 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1141 struct dentry *old = nd->path.dentry;
1142 struct dentry *parent = old->d_parent;
1143 unsigned seq;
1144
1145 seq = read_seqcount_begin(&parent->d_seq);
1146 if (read_seqcount_retry(&old->d_seq, nd->seq))
Al Viroef7562d2011-03-04 14:35:59 -05001147 goto failed;
Nick Piggin31e6b012011-01-07 17:49:52 +11001148 nd->path.dentry = parent;
1149 nd->seq = seq;
1150 break;
1151 }
1152 if (!follow_up_rcu(&nd->path))
1153 break;
1154 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11001155 }
Al Virob37199e2014-03-20 15:18:22 -04001156 while (d_mountpoint(nd->path.dentry)) {
1157 struct mount *mounted;
1158 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1159 if (!mounted)
1160 break;
1161 nd->path.mnt = &mounted->mnt;
1162 nd->path.dentry = mounted->mnt.mnt_root;
1163 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1164 if (!read_seqretry(&mount_lock, nd->m_seq))
1165 goto failed;
1166 }
Al Virodea39372011-05-27 06:53:39 -04001167 nd->inode = nd->path.dentry->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001168 return 0;
Al Viroef7562d2011-03-04 14:35:59 -05001169
1170failed:
1171 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001172 if (!(nd->flags & LOOKUP_ROOT))
1173 nd->root.mnt = NULL;
Al Viro8b61e742013-11-08 12:45:01 -05001174 rcu_read_unlock();
Al Viroef7562d2011-03-04 14:35:59 -05001175 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001176}
1177
David Howells9875cf82011-01-14 18:45:21 +00001178/*
David Howellscc53ce52011-01-14 18:45:26 +00001179 * Follow down to the covering mount currently visible to userspace. At each
1180 * point, the filesystem owning that dentry may be queried as to whether the
1181 * caller is permitted to proceed or not.
David Howellscc53ce52011-01-14 18:45:26 +00001182 */
Al Viro7cc90cc2011-03-18 09:04:20 -04001183int follow_down(struct path *path)
David Howellscc53ce52011-01-14 18:45:26 +00001184{
1185 unsigned managed;
1186 int ret;
1187
1188 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1189 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1190 /* Allow the filesystem to manage the transit without i_mutex
1191 * being held.
1192 *
1193 * We indicate to the filesystem if someone is trying to mount
1194 * something here. This gives autofs the chance to deny anyone
1195 * other than its daemon the right to mount on its
1196 * superstructure.
1197 *
1198 * The filesystem may sleep at this point.
1199 */
1200 if (managed & DCACHE_MANAGE_TRANSIT) {
1201 BUG_ON(!path->dentry->d_op);
1202 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +00001203 ret = path->dentry->d_op->d_manage(
Al Viro1aed3e42011-03-18 09:09:02 -04001204 path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001205 if (ret < 0)
1206 return ret == -EISDIR ? 0 : ret;
1207 }
1208
1209 /* Transit to a mounted filesystem. */
1210 if (managed & DCACHE_MOUNTED) {
1211 struct vfsmount *mounted = lookup_mnt(path);
1212 if (!mounted)
1213 break;
1214 dput(path->dentry);
1215 mntput(path->mnt);
1216 path->mnt = mounted;
1217 path->dentry = dget(mounted->mnt_root);
1218 continue;
1219 }
1220
1221 /* Don't handle automount points here */
1222 break;
1223 }
1224 return 0;
1225}
1226
1227/*
David Howells9875cf82011-01-14 18:45:21 +00001228 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1229 */
1230static void follow_mount(struct path *path)
1231{
1232 while (d_mountpoint(path->dentry)) {
1233 struct vfsmount *mounted = lookup_mnt(path);
1234 if (!mounted)
1235 break;
1236 dput(path->dentry);
1237 mntput(path->mnt);
1238 path->mnt = mounted;
1239 path->dentry = dget(mounted->mnt_root);
1240 }
1241}
1242
Nick Piggin31e6b012011-01-07 17:49:52 +11001243static void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244{
Al Viro2a737872009-04-07 11:49:53 -04001245 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001246
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 while(1) {
Jan Blunck4ac91372008-02-14 19:34:32 -08001248 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249
Al Viro2a737872009-04-07 11:49:53 -04001250 if (nd->path.dentry == nd->root.dentry &&
1251 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 break;
1253 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001254 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Al Viro3088dd72010-01-30 15:47:29 -05001255 /* rare case of legitimate dget_parent()... */
1256 nd->path.dentry = dget_parent(nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 dput(old);
1258 break;
1259 }
Al Viro3088dd72010-01-30 15:47:29 -05001260 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 }
Al Viro79ed0222009-04-18 13:59:41 -04001263 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001264 nd->inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265}
1266
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267/*
Miklos Szeredibad61182012-03-26 12:54:24 +02001268 * This looks up the name in dcache, possibly revalidates the old dentry and
1269 * allocates a new one if not found or not valid. In the need_lookup argument
1270 * returns whether i_op->lookup is necessary.
1271 *
1272 * dir->d_inode->i_mutex must be held
Nick Pigginbaa03892010-08-18 04:37:31 +10001273 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001274static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
Al Viro201f9562012-06-22 12:42:10 +04001275 unsigned int flags, bool *need_lookup)
Nick Pigginbaa03892010-08-18 04:37:31 +10001276{
Nick Pigginbaa03892010-08-18 04:37:31 +10001277 struct dentry *dentry;
Miklos Szeredibad61182012-03-26 12:54:24 +02001278 int error;
Nick Pigginbaa03892010-08-18 04:37:31 +10001279
Miklos Szeredibad61182012-03-26 12:54:24 +02001280 *need_lookup = false;
1281 dentry = d_lookup(dir, name);
1282 if (dentry) {
Jeff Layton39e3c952012-11-28 11:30:53 -05001283 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
Al Viro201f9562012-06-22 12:42:10 +04001284 error = d_revalidate(dentry, flags);
Miklos Szeredibad61182012-03-26 12:54:24 +02001285 if (unlikely(error <= 0)) {
1286 if (error < 0) {
1287 dput(dentry);
1288 return ERR_PTR(error);
1289 } else if (!d_invalidate(dentry)) {
1290 dput(dentry);
1291 dentry = NULL;
1292 }
1293 }
1294 }
1295 }
Nick Pigginbaa03892010-08-18 04:37:31 +10001296
Miklos Szeredibad61182012-03-26 12:54:24 +02001297 if (!dentry) {
1298 dentry = d_alloc(dir, name);
1299 if (unlikely(!dentry))
1300 return ERR_PTR(-ENOMEM);
Nick Pigginbaa03892010-08-18 04:37:31 +10001301
Miklos Szeredibad61182012-03-26 12:54:24 +02001302 *need_lookup = true;
Nick Pigginbaa03892010-08-18 04:37:31 +10001303 }
1304 return dentry;
1305}
1306
1307/*
J. Bruce Fields13a2c3b2013-10-23 16:09:16 -04001308 * Call i_op->lookup on the dentry. The dentry must be negative and
1309 * unhashed.
Miklos Szeredibad61182012-03-26 12:54:24 +02001310 *
1311 * dir->d_inode->i_mutex must be held
Josef Bacik44396f42011-05-31 11:58:49 -04001312 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001313static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
Al Viro72bd8662012-06-10 17:17:17 -04001314 unsigned int flags)
Josef Bacik44396f42011-05-31 11:58:49 -04001315{
Josef Bacik44396f42011-05-31 11:58:49 -04001316 struct dentry *old;
1317
1318 /* Don't create child dentry for a dead directory. */
Miklos Szeredibad61182012-03-26 12:54:24 +02001319 if (unlikely(IS_DEADDIR(dir))) {
Miklos Szeredie188dc02012-02-03 14:25:18 +01001320 dput(dentry);
Josef Bacik44396f42011-05-31 11:58:49 -04001321 return ERR_PTR(-ENOENT);
Miklos Szeredie188dc02012-02-03 14:25:18 +01001322 }
Josef Bacik44396f42011-05-31 11:58:49 -04001323
Al Viro72bd8662012-06-10 17:17:17 -04001324 old = dir->i_op->lookup(dir, dentry, flags);
Josef Bacik44396f42011-05-31 11:58:49 -04001325 if (unlikely(old)) {
1326 dput(dentry);
1327 dentry = old;
1328 }
1329 return dentry;
1330}
1331
Al Viroa3255542012-03-30 14:41:51 -04001332static struct dentry *__lookup_hash(struct qstr *name,
Al Viro72bd8662012-06-10 17:17:17 -04001333 struct dentry *base, unsigned int flags)
Al Viroa3255542012-03-30 14:41:51 -04001334{
Miklos Szeredibad61182012-03-26 12:54:24 +02001335 bool need_lookup;
Al Viroa3255542012-03-30 14:41:51 -04001336 struct dentry *dentry;
1337
Al Viro72bd8662012-06-10 17:17:17 -04001338 dentry = lookup_dcache(name, base, flags, &need_lookup);
Miklos Szeredibad61182012-03-26 12:54:24 +02001339 if (!need_lookup)
1340 return dentry;
Al Viroa3255542012-03-30 14:41:51 -04001341
Al Viro72bd8662012-06-10 17:17:17 -04001342 return lookup_real(base->d_inode, dentry, flags);
Al Viroa3255542012-03-30 14:41:51 -04001343}
1344
Josef Bacik44396f42011-05-31 11:58:49 -04001345/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 * It's more convoluted than I'd like it to be, but... it's still fairly
1347 * small and for now I'd prefer to have fast path as straight as possible.
1348 * It _is_ time-critical.
1349 */
Al Viroe97cdc82013-01-24 18:16:00 -05001350static int lookup_fast(struct nameidata *nd,
Miklos Szeredi697f5142012-05-21 17:30:05 +02001351 struct path *path, struct inode **inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352{
Jan Blunck4ac91372008-02-14 19:34:32 -08001353 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001354 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001355 int need_reval = 1;
1356 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001357 int err;
1358
Al Viro3cac2602009-08-13 18:27:43 +04001359 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001360 * Rename seqlock is not required here because in the off chance
1361 * of a false negative due to a concurrent rename, we're going to
1362 * do the non-racy lookup, below.
1363 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001364 if (nd->flags & LOOKUP_RCU) {
1365 unsigned seq;
Linus Torvaldsda53be12013-05-21 15:22:44 -07001366 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
Al Viro5a18fff2011-03-11 04:44:53 -05001367 if (!dentry)
1368 goto unlazy;
1369
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001370 /*
1371 * This sequence count validates that the inode matches
1372 * the dentry name information from lookup.
1373 */
1374 *inode = dentry->d_inode;
1375 if (read_seqcount_retry(&dentry->d_seq, seq))
1376 return -ECHILD;
1377
1378 /*
1379 * This sequence count validates that the parent had no
1380 * changes while we did the lookup of the dentry above.
1381 *
1382 * The memory barrier in read_seqcount_begin of child is
1383 * enough, we can use __read_seqcount_retry here.
1384 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001385 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1386 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001387 nd->seq = seq;
Al Viro5a18fff2011-03-11 04:44:53 -05001388
Al Viro24643082011-02-15 01:26:22 -05001389 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
Al Viro4ce16ef32012-06-10 16:10:59 -04001390 status = d_revalidate(dentry, nd->flags);
Al Viro5a18fff2011-03-11 04:44:53 -05001391 if (unlikely(status <= 0)) {
1392 if (status != -ECHILD)
1393 need_reval = 0;
1394 goto unlazy;
1395 }
Al Viro24643082011-02-15 01:26:22 -05001396 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001397 path->mnt = mnt;
1398 path->dentry = dentry;
Al Virod6e9bd22011-05-27 07:03:15 -04001399 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1400 goto unlazy;
1401 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1402 goto unlazy;
1403 return 0;
Al Viro5a18fff2011-03-11 04:44:53 -05001404unlazy:
Al Viro19660af2011-03-25 10:32:48 -04001405 if (unlazy_walk(nd, dentry))
1406 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001407 } else {
Al Viroe97cdc82013-01-24 18:16:00 -05001408 dentry = __d_lookup(parent, &nd->last);
Nick Piggin31e6b012011-01-07 17:49:52 +11001409 }
Al Viro5a18fff2011-03-11 04:44:53 -05001410
Al Viro81e6f522012-03-30 14:48:04 -04001411 if (unlikely(!dentry))
1412 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001413
Al Viro5a18fff2011-03-11 04:44:53 -05001414 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
Al Viro4ce16ef32012-06-10 16:10:59 -04001415 status = d_revalidate(dentry, nd->flags);
Al Viro5a18fff2011-03-11 04:44:53 -05001416 if (unlikely(status <= 0)) {
1417 if (status < 0) {
1418 dput(dentry);
1419 return status;
1420 }
1421 if (!d_invalidate(dentry)) {
1422 dput(dentry);
Al Viro81e6f522012-03-30 14:48:04 -04001423 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001424 }
1425 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001426
David Howells9875cf82011-01-14 18:45:21 +00001427 path->mnt = mnt;
1428 path->dentry = dentry;
1429 err = follow_managed(path, nd->flags);
Ian Kent89312212011-01-18 12:06:10 +08001430 if (unlikely(err < 0)) {
1431 path_put_conditional(path, nd);
David Howells9875cf82011-01-14 18:45:21 +00001432 return err;
Ian Kent89312212011-01-18 12:06:10 +08001433 }
Al Viroa3fbbde2011-11-07 21:21:26 +00001434 if (err)
1435 nd->flags |= LOOKUP_JUMPED;
David Howells9875cf82011-01-14 18:45:21 +00001436 *inode = path->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 return 0;
Al Viro81e6f522012-03-30 14:48:04 -04001438
1439need_lookup:
Miklos Szeredi697f5142012-05-21 17:30:05 +02001440 return 1;
1441}
1442
1443/* Fast lookup failed, do it the slow way */
Al Virocc2a5272013-01-24 18:19:49 -05001444static int lookup_slow(struct nameidata *nd, struct path *path)
Miklos Szeredi697f5142012-05-21 17:30:05 +02001445{
1446 struct dentry *dentry, *parent;
1447 int err;
1448
1449 parent = nd->path.dentry;
Al Viro81e6f522012-03-30 14:48:04 -04001450 BUG_ON(nd->inode != parent->d_inode);
1451
1452 mutex_lock(&parent->d_inode->i_mutex);
Al Virocc2a5272013-01-24 18:19:49 -05001453 dentry = __lookup_hash(&nd->last, parent, nd->flags);
Al Viro81e6f522012-03-30 14:48:04 -04001454 mutex_unlock(&parent->d_inode->i_mutex);
1455 if (IS_ERR(dentry))
1456 return PTR_ERR(dentry);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001457 path->mnt = nd->path.mnt;
1458 path->dentry = dentry;
1459 err = follow_managed(path, nd->flags);
1460 if (unlikely(err < 0)) {
1461 path_put_conditional(path, nd);
1462 return err;
1463 }
1464 if (err)
1465 nd->flags |= LOOKUP_JUMPED;
1466 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467}
1468
Al Viro52094c82011-02-21 21:34:47 -05001469static inline int may_lookup(struct nameidata *nd)
1470{
1471 if (nd->flags & LOOKUP_RCU) {
Al Viro4ad5abb2011-06-20 19:57:03 -04001472 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
Al Viro52094c82011-02-21 21:34:47 -05001473 if (err != -ECHILD)
1474 return err;
Al Viro19660af2011-03-25 10:32:48 -04001475 if (unlazy_walk(nd, NULL))
Al Viro52094c82011-02-21 21:34:47 -05001476 return -ECHILD;
1477 }
Al Viro4ad5abb2011-06-20 19:57:03 -04001478 return inode_permission(nd->inode, MAY_EXEC);
Al Viro52094c82011-02-21 21:34:47 -05001479}
1480
Al Viro9856fa12011-03-04 14:22:06 -05001481static inline int handle_dots(struct nameidata *nd, int type)
1482{
1483 if (type == LAST_DOTDOT) {
1484 if (nd->flags & LOOKUP_RCU) {
1485 if (follow_dotdot_rcu(nd))
1486 return -ECHILD;
1487 } else
1488 follow_dotdot(nd);
1489 }
1490 return 0;
1491}
1492
Al Viro951361f2011-03-04 14:44:37 -05001493static void terminate_walk(struct nameidata *nd)
1494{
1495 if (!(nd->flags & LOOKUP_RCU)) {
1496 path_put(&nd->path);
1497 } else {
1498 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001499 if (!(nd->flags & LOOKUP_ROOT))
1500 nd->root.mnt = NULL;
Al Viro8b61e742013-11-08 12:45:01 -05001501 rcu_read_unlock();
Al Viro951361f2011-03-04 14:44:37 -05001502 }
1503}
1504
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001505/*
1506 * Do we need to follow links? We _really_ want to be able
1507 * to do this check without having to look at inode->i_op,
1508 * so we keep a cache of "no, this doesn't need follow_link"
1509 * for the common case.
1510 */
David Howellsb18825a2013-09-12 19:22:53 +01001511static inline int should_follow_link(struct dentry *dentry, int follow)
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001512{
David Howellsb18825a2013-09-12 19:22:53 +01001513 return unlikely(d_is_symlink(dentry)) ? follow : 0;
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001514}
1515
Al Viroce57dfc2011-03-13 19:58:58 -04001516static inline int walk_component(struct nameidata *nd, struct path *path,
Al Viro21b9b072013-01-24 18:10:25 -05001517 int follow)
Al Viroce57dfc2011-03-13 19:58:58 -04001518{
1519 struct inode *inode;
1520 int err;
1521 /*
1522 * "." and ".." are special - ".." especially so because it has
1523 * to be able to know about the current root directory and
1524 * parent relationships.
1525 */
Al Viro21b9b072013-01-24 18:10:25 -05001526 if (unlikely(nd->last_type != LAST_NORM))
1527 return handle_dots(nd, nd->last_type);
Al Viroe97cdc82013-01-24 18:16:00 -05001528 err = lookup_fast(nd, path, &inode);
Al Viroce57dfc2011-03-13 19:58:58 -04001529 if (unlikely(err)) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001530 if (err < 0)
1531 goto out_err;
1532
Al Virocc2a5272013-01-24 18:19:49 -05001533 err = lookup_slow(nd, path);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001534 if (err < 0)
1535 goto out_err;
1536
1537 inode = path->dentry->d_inode;
Al Viroce57dfc2011-03-13 19:58:58 -04001538 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001539 err = -ENOENT;
1540 if (!inode)
1541 goto out_path_put;
1542
David Howellsb18825a2013-09-12 19:22:53 +01001543 if (should_follow_link(path->dentry, follow)) {
Al Viro19660af2011-03-25 10:32:48 -04001544 if (nd->flags & LOOKUP_RCU) {
1545 if (unlikely(unlazy_walk(nd, path->dentry))) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001546 err = -ECHILD;
1547 goto out_err;
Al Viro19660af2011-03-25 10:32:48 -04001548 }
1549 }
Al Viroce57dfc2011-03-13 19:58:58 -04001550 BUG_ON(inode != path->dentry->d_inode);
1551 return 1;
1552 }
1553 path_to_nameidata(path, nd);
1554 nd->inode = inode;
1555 return 0;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001556
1557out_path_put:
1558 path_to_nameidata(path, nd);
1559out_err:
1560 terminate_walk(nd);
1561 return err;
Al Viroce57dfc2011-03-13 19:58:58 -04001562}
1563
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564/*
Al Virob3563792011-03-14 21:54:55 -04001565 * This limits recursive symlink follows to 8, while
1566 * limiting consecutive symlinks to 40.
1567 *
1568 * Without that kind of total limit, nasty chains of consecutive
1569 * symlinks can cause almost arbitrarily long lookups.
1570 */
1571static inline int nested_symlink(struct path *path, struct nameidata *nd)
1572{
1573 int res;
1574
Al Virob3563792011-03-14 21:54:55 -04001575 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1576 path_put_conditional(path, nd);
1577 path_put(&nd->path);
1578 return -ELOOP;
1579 }
Erez Zadok1a4022f2011-05-21 01:19:59 -04001580 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
Al Virob3563792011-03-14 21:54:55 -04001581
1582 nd->depth++;
1583 current->link_count++;
1584
1585 do {
1586 struct path link = *path;
1587 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04001588
1589 res = follow_link(&link, nd, &cookie);
Al Viro6d7b5aa2012-06-10 04:15:17 -04001590 if (res)
1591 break;
Al Viro21b9b072013-01-24 18:10:25 -05001592 res = walk_component(nd, path, LOOKUP_FOLLOW);
Al Viro574197e2011-03-14 22:20:34 -04001593 put_link(nd, &link, cookie);
Al Virob3563792011-03-14 21:54:55 -04001594 } while (res > 0);
1595
1596 current->link_count--;
1597 nd->depth--;
1598 return res;
1599}
1600
1601/*
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001602 * We can do the critical dentry name comparison and hashing
1603 * operations one word at a time, but we are limited to:
1604 *
1605 * - Architectures with fast unaligned word accesses. We could
1606 * do a "get_unaligned()" if this helps and is sufficiently
1607 * fast.
1608 *
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001609 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1610 * do not trap on the (extremely unlikely) case of a page
1611 * crossing operation.
1612 *
1613 * - Furthermore, we need an efficient 64-bit compile for the
1614 * 64-bit case in order to generate the "number of bytes in
1615 * the final mask". Again, that could be replaced with a
1616 * efficient population count instruction or similar.
1617 */
1618#ifdef CONFIG_DCACHE_WORD_ACCESS
1619
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001620#include <asm/word-at-a-time.h>
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001621
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001622#ifdef CONFIG_64BIT
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001623
1624static inline unsigned int fold_hash(unsigned long hash)
1625{
1626 hash += hash >> (8*sizeof(int));
1627 return hash;
1628}
1629
1630#else /* 32-bit case */
1631
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001632#define fold_hash(x) (x)
1633
1634#endif
1635
1636unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1637{
1638 unsigned long a, mask;
1639 unsigned long hash = 0;
1640
1641 for (;;) {
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001642 a = load_unaligned_zeropad(name);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001643 if (len < sizeof(unsigned long))
1644 break;
1645 hash += a;
Al Virof132c5b2012-03-22 21:59:52 +00001646 hash *= 9;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001647 name += sizeof(unsigned long);
1648 len -= sizeof(unsigned long);
1649 if (!len)
1650 goto done;
1651 }
Will Deacona5c21dc2013-12-12 17:40:21 +00001652 mask = bytemask_from_count(len);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001653 hash += mask & a;
1654done:
1655 return fold_hash(hash);
1656}
1657EXPORT_SYMBOL(full_name_hash);
1658
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001659/*
1660 * Calculate the length and hash of the path component, and
1661 * return the length of the component;
1662 */
1663static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1664{
Linus Torvalds36126f82012-05-26 10:43:17 -07001665 unsigned long a, b, adata, bdata, mask, hash, len;
1666 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001667
1668 hash = a = 0;
1669 len = -sizeof(unsigned long);
1670 do {
1671 hash = (hash + a) * 9;
1672 len += sizeof(unsigned long);
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001673 a = load_unaligned_zeropad(name+len);
Linus Torvalds36126f82012-05-26 10:43:17 -07001674 b = a ^ REPEAT_BYTE('/');
1675 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001676
Linus Torvalds36126f82012-05-26 10:43:17 -07001677 adata = prep_zero_mask(a, adata, &constants);
1678 bdata = prep_zero_mask(b, bdata, &constants);
1679
1680 mask = create_zero_mask(adata | bdata);
1681
1682 hash += a & zero_bytemask(mask);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001683 *hashp = fold_hash(hash);
1684
Linus Torvalds36126f82012-05-26 10:43:17 -07001685 return len + find_zero(mask);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001686}
1687
1688#else
1689
Linus Torvalds0145acc2012-03-02 14:32:59 -08001690unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1691{
1692 unsigned long hash = init_name_hash();
1693 while (len--)
1694 hash = partial_name_hash(*name++, hash);
1695 return end_name_hash(hash);
1696}
Linus Torvaldsae942ae2012-03-02 19:40:57 -08001697EXPORT_SYMBOL(full_name_hash);
Linus Torvalds0145acc2012-03-02 14:32:59 -08001698
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001699/*
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001700 * We know there's a real path component here of at least
1701 * one character.
1702 */
1703static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1704{
1705 unsigned long hash = init_name_hash();
1706 unsigned long len = 0, c;
1707
1708 c = (unsigned char)*name;
1709 do {
1710 len++;
1711 hash = partial_name_hash(c, hash);
1712 c = (unsigned char)name[len];
1713 } while (c && c != '/');
1714 *hashp = end_name_hash(hash);
1715 return len;
1716}
1717
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001718#endif
1719
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001720/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01001722 * This is the basic name resolution function, turning a pathname into
1723 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01001725 * Returns 0 and nd will have valid dentry and mnt on success.
1726 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 */
Al Viro6de88d72009-08-09 01:41:57 +04001728static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729{
1730 struct path next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732
1733 while (*name=='/')
1734 name++;
1735 if (!*name)
Al Viro086e1832011-02-22 20:56:27 -05001736 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738 /* At this point we know we have a real path component. */
1739 for(;;) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740 struct qstr this;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001741 long len;
Al Virofe479a52011-02-22 15:10:03 -05001742 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743
Al Viro52094c82011-02-21 21:34:47 -05001744 err = may_lookup(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745 if (err)
1746 break;
1747
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001748 len = hash_name(name, &this.hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 this.name = name;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001750 this.len = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751
Al Virofe479a52011-02-22 15:10:03 -05001752 type = LAST_NORM;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001753 if (name[0] == '.') switch (len) {
Al Virofe479a52011-02-22 15:10:03 -05001754 case 2:
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001755 if (name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05001756 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05001757 nd->flags |= LOOKUP_JUMPED;
1758 }
Al Virofe479a52011-02-22 15:10:03 -05001759 break;
1760 case 1:
1761 type = LAST_DOT;
1762 }
Al Viro5a202bc2011-03-08 14:17:44 -05001763 if (likely(type == LAST_NORM)) {
1764 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05001765 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05001766 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07001767 err = parent->d_op->d_hash(parent, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05001768 if (err < 0)
1769 break;
1770 }
1771 }
Al Virofe479a52011-02-22 15:10:03 -05001772
Al Viro5f4a6a62013-01-24 18:04:22 -05001773 nd->last = this;
1774 nd->last_type = type;
1775
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001776 if (!name[len])
Al Viro5f4a6a62013-01-24 18:04:22 -05001777 return 0;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001778 /*
1779 * If it wasn't NUL, we know it was '/'. Skip that
1780 * slash, and continue until no more slashes.
1781 */
1782 do {
1783 len++;
1784 } while (unlikely(name[len] == '/'));
1785 if (!name[len])
Al Viro5f4a6a62013-01-24 18:04:22 -05001786 return 0;
1787
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001788 name += len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789
Al Viro21b9b072013-01-24 18:10:25 -05001790 err = walk_component(nd, &next, LOOKUP_FOLLOW);
Al Viroce57dfc2011-03-13 19:58:58 -04001791 if (err < 0)
1792 return err;
Al Virofe479a52011-02-22 15:10:03 -05001793
Al Viroce57dfc2011-03-13 19:58:58 -04001794 if (err) {
Al Virob3563792011-03-14 21:54:55 -04001795 err = nested_symlink(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796 if (err)
Al Viroa7472ba2011-03-04 14:39:30 -05001797 return err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001798 }
David Howellsb18825a2013-09-12 19:22:53 +01001799 if (!d_is_directory(nd->path.dentry)) {
Al Viro5f4a6a62013-01-24 18:04:22 -05001800 err = -ENOTDIR;
1801 break;
1802 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 }
Al Viro951361f2011-03-04 14:44:37 -05001804 terminate_walk(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 return err;
1806}
1807
Al Viro70e9b352011-03-05 21:12:22 -05001808static int path_init(int dfd, const char *name, unsigned int flags,
1809 struct nameidata *nd, struct file **fp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001811 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812
1813 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro16c2cd72011-02-22 15:50:10 -05001814 nd->flags = flags | LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 nd->depth = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05001816 if (flags & LOOKUP_ROOT) {
David Howellsb18825a2013-09-12 19:22:53 +01001817 struct dentry *root = nd->root.dentry;
1818 struct inode *inode = root->d_inode;
Al Viro73d049a2011-03-11 12:08:24 -05001819 if (*name) {
David Howellsb18825a2013-09-12 19:22:53 +01001820 if (!d_is_directory(root))
Al Viro73d049a2011-03-11 12:08:24 -05001821 return -ENOTDIR;
1822 retval = inode_permission(inode, MAY_EXEC);
1823 if (retval)
1824 return retval;
1825 }
Al Viro5b6ca022011-03-09 23:04:47 -05001826 nd->path = nd->root;
1827 nd->inode = inode;
1828 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05001829 rcu_read_lock();
Al Viro5b6ca022011-03-09 23:04:47 -05001830 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro48a066e2013-09-29 22:06:07 -04001831 nd->m_seq = read_seqbegin(&mount_lock);
Al Viro5b6ca022011-03-09 23:04:47 -05001832 } else {
1833 path_get(&nd->path);
1834 }
1835 return 0;
1836 }
1837
Al Viro2a737872009-04-07 11:49:53 -04001838 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839
Al Viro48a066e2013-09-29 22:06:07 -04001840 nd->m_seq = read_seqbegin(&mount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 if (*name=='/') {
Al Viroe41f7d42011-02-22 14:02:58 -05001842 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05001843 rcu_read_lock();
Al Viroe41f7d42011-02-22 14:02:58 -05001844 set_root_rcu(nd);
1845 } else {
1846 set_root(nd);
1847 path_get(&nd->root);
1848 }
Al Viro2a737872009-04-07 11:49:53 -04001849 nd->path = nd->root;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001850 } else if (dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05001851 if (flags & LOOKUP_RCU) {
1852 struct fs_struct *fs = current->fs;
1853 unsigned seq;
1854
Al Viro8b61e742013-11-08 12:45:01 -05001855 rcu_read_lock();
Al Viroe41f7d42011-02-22 14:02:58 -05001856
1857 do {
1858 seq = read_seqcount_begin(&fs->seq);
1859 nd->path = fs->pwd;
1860 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1861 } while (read_seqcount_retry(&fs->seq, seq));
1862 } else {
1863 get_fs_pwd(current->fs, &nd->path);
1864 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001865 } else {
Jeff Layton582aa642012-12-11 08:56:16 -05001866 /* Caller must check execute permissions on the starting path component */
Al Viro2903ff02012-08-28 12:52:22 -04001867 struct fd f = fdget_raw(dfd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001868 struct dentry *dentry;
1869
Al Viro2903ff02012-08-28 12:52:22 -04001870 if (!f.file)
1871 return -EBADF;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001872
Al Viro2903ff02012-08-28 12:52:22 -04001873 dentry = f.file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001874
Al Virof52e0c12011-03-14 18:56:51 -04001875 if (*name) {
David Howellsb18825a2013-09-12 19:22:53 +01001876 if (!d_is_directory(dentry)) {
Al Viro2903ff02012-08-28 12:52:22 -04001877 fdput(f);
1878 return -ENOTDIR;
1879 }
Al Virof52e0c12011-03-14 18:56:51 -04001880 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001881
Al Viro2903ff02012-08-28 12:52:22 -04001882 nd->path = f.file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05001883 if (flags & LOOKUP_RCU) {
Linus Torvalds9c225f22014-03-03 09:36:58 -08001884 if (f.flags & FDPUT_FPUT)
Al Viro2903ff02012-08-28 12:52:22 -04001885 *fp = f.file;
Al Viroe41f7d42011-02-22 14:02:58 -05001886 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro8b61e742013-11-08 12:45:01 -05001887 rcu_read_lock();
Al Viroe41f7d42011-02-22 14:02:58 -05001888 } else {
Al Viro2903ff02012-08-28 12:52:22 -04001889 path_get(&nd->path);
1890 fdput(f);
Al Viroe41f7d42011-02-22 14:02:58 -05001891 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 }
Al Viroe41f7d42011-02-22 14:02:58 -05001893
Nick Piggin31e6b012011-01-07 17:49:52 +11001894 nd->inode = nd->path.dentry->d_inode;
Al Viro9b4a9b12009-04-07 11:44:16 -04001895 return 0;
Al Viro9b4a9b12009-04-07 11:44:16 -04001896}
1897
Al Virobd92d7f2011-03-14 19:54:59 -04001898static inline int lookup_last(struct nameidata *nd, struct path *path)
1899{
1900 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1901 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1902
1903 nd->flags &= ~LOOKUP_PARENT;
Al Viro21b9b072013-01-24 18:10:25 -05001904 return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW);
Al Virobd92d7f2011-03-14 19:54:59 -04001905}
1906
Al Viro9b4a9b12009-04-07 11:44:16 -04001907/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroee0827c2011-02-21 23:38:09 -05001908static int path_lookupat(int dfd, const char *name,
Al Viro9b4a9b12009-04-07 11:44:16 -04001909 unsigned int flags, struct nameidata *nd)
1910{
Al Viro70e9b352011-03-05 21:12:22 -05001911 struct file *base = NULL;
Al Virobd92d7f2011-03-14 19:54:59 -04001912 struct path path;
1913 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001914
1915 /*
1916 * Path walking is largely split up into 2 different synchronisation
1917 * schemes, rcu-walk and ref-walk (explained in
1918 * Documentation/filesystems/path-lookup.txt). These share much of the
1919 * path walk code, but some things particularly setup, cleanup, and
1920 * following mounts are sufficiently divergent that functions are
1921 * duplicated. Typically there is a function foo(), and its RCU
1922 * analogue, foo_rcu().
1923 *
1924 * -ECHILD is the error number of choice (just to avoid clashes) that
1925 * is returned if some aspect of an rcu-walk fails. Such an error must
1926 * be handled by restarting a traditional ref-walk (which will always
1927 * be able to complete).
1928 */
Al Virobd92d7f2011-03-14 19:54:59 -04001929 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
Al Viroee0827c2011-02-21 23:38:09 -05001930
Al Virobd92d7f2011-03-14 19:54:59 -04001931 if (unlikely(err))
1932 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001933
1934 current->total_link_count = 0;
Al Virobd92d7f2011-03-14 19:54:59 -04001935 err = link_path_walk(name, nd);
1936
1937 if (!err && !(flags & LOOKUP_PARENT)) {
Al Virobd92d7f2011-03-14 19:54:59 -04001938 err = lookup_last(nd, &path);
1939 while (err > 0) {
1940 void *cookie;
1941 struct path link = path;
Kees Cook800179c2012-07-25 17:29:07 -07001942 err = may_follow_link(&link, nd);
1943 if (unlikely(err))
1944 break;
Al Virobd92d7f2011-03-14 19:54:59 -04001945 nd->flags |= LOOKUP_PARENT;
Al Viro574197e2011-03-14 22:20:34 -04001946 err = follow_link(&link, nd, &cookie);
Al Viro6d7b5aa2012-06-10 04:15:17 -04001947 if (err)
1948 break;
1949 err = lookup_last(nd, &path);
Al Viro574197e2011-03-14 22:20:34 -04001950 put_link(nd, &link, cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04001951 }
1952 }
Al Viroee0827c2011-02-21 23:38:09 -05001953
Al Viro9f1fafe2011-03-25 11:00:12 -04001954 if (!err)
1955 err = complete_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04001956
1957 if (!err && nd->flags & LOOKUP_DIRECTORY) {
David Howellsb18825a2013-09-12 19:22:53 +01001958 if (!d_is_directory(nd->path.dentry)) {
Al Virobd92d7f2011-03-14 19:54:59 -04001959 path_put(&nd->path);
Al Virobd23a532011-03-23 09:56:30 -04001960 err = -ENOTDIR;
Al Virobd92d7f2011-03-14 19:54:59 -04001961 }
1962 }
Al Viro16c2cd72011-02-22 15:50:10 -05001963
Al Viro70e9b352011-03-05 21:12:22 -05001964 if (base)
1965 fput(base);
Al Viroee0827c2011-02-21 23:38:09 -05001966
Al Viro5b6ca022011-03-09 23:04:47 -05001967 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
Al Viro2a737872009-04-07 11:49:53 -04001968 path_put(&nd->root);
1969 nd->root.mnt = NULL;
1970 }
Al Virobd92d7f2011-03-14 19:54:59 -04001971 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001972}
Nick Piggin31e6b012011-01-07 17:49:52 +11001973
Jeff Layton873f1ee2012-10-10 15:25:29 -04001974static int filename_lookup(int dfd, struct filename *name,
1975 unsigned int flags, struct nameidata *nd)
1976{
1977 int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd);
1978 if (unlikely(retval == -ECHILD))
1979 retval = path_lookupat(dfd, name->name, flags, nd);
1980 if (unlikely(retval == -ESTALE))
1981 retval = path_lookupat(dfd, name->name,
1982 flags | LOOKUP_REVAL, nd);
1983
1984 if (likely(!retval))
Jeff Laytonadb5c242012-10-10 16:43:13 -04001985 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
Jeff Layton873f1ee2012-10-10 15:25:29 -04001986 return retval;
1987}
1988
Al Viroee0827c2011-02-21 23:38:09 -05001989static int do_path_lookup(int dfd, const char *name,
1990 unsigned int flags, struct nameidata *nd)
1991{
Jeff Layton873f1ee2012-10-10 15:25:29 -04001992 struct filename filename = { .name = name };
Nick Piggin31e6b012011-01-07 17:49:52 +11001993
Jeff Layton873f1ee2012-10-10 15:25:29 -04001994 return filename_lookup(dfd, &filename, flags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995}
1996
Al Viro79714f72012-06-15 03:01:42 +04001997/* does lookup, returns the object with parent locked */
1998struct dentry *kern_path_locked(const char *name, struct path *path)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001999{
Al Viro79714f72012-06-15 03:01:42 +04002000 struct nameidata nd;
2001 struct dentry *d;
2002 int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd);
2003 if (err)
2004 return ERR_PTR(err);
2005 if (nd.last_type != LAST_NORM) {
2006 path_put(&nd.path);
2007 return ERR_PTR(-EINVAL);
2008 }
2009 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Al Viro1e0ea002012-07-22 23:46:21 +04002010 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
Al Viro79714f72012-06-15 03:01:42 +04002011 if (IS_ERR(d)) {
2012 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2013 path_put(&nd.path);
2014 return d;
2015 }
2016 *path = nd.path;
2017 return d;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002018}
2019
Al Virod1811462008-08-02 00:49:18 -04002020int kern_path(const char *name, unsigned int flags, struct path *path)
2021{
2022 struct nameidata nd;
2023 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
2024 if (!res)
2025 *path = nd.path;
2026 return res;
2027}
2028
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002029/**
2030 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2031 * @dentry: pointer to dentry of the base directory
2032 * @mnt: pointer to vfs mount of the base directory
2033 * @name: pointer to file name
2034 * @flags: lookup flags
Al Viroe0a01242011-06-27 17:00:37 -04002035 * @path: pointer to struct path to fill
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002036 */
2037int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2038 const char *name, unsigned int flags,
Al Viroe0a01242011-06-27 17:00:37 -04002039 struct path *path)
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002040{
Al Viroe0a01242011-06-27 17:00:37 -04002041 struct nameidata nd;
2042 int err;
2043 nd.root.dentry = dentry;
2044 nd.root.mnt = mnt;
2045 BUG_ON(flags & LOOKUP_PARENT);
Al Viro5b6ca022011-03-09 23:04:47 -05002046 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
Al Viroe0a01242011-06-27 17:00:37 -04002047 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
2048 if (!err)
2049 *path = nd.path;
2050 return err;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002051}
2052
James Morris057f6c02007-04-26 00:12:05 -07002053/*
2054 * Restricted form of lookup. Doesn't follow links, single-component only,
2055 * needs parent already locked. Doesn't follow mounts.
2056 * SMP-safe.
2057 */
Adrian Bunka244e162006-03-31 02:32:11 -08002058static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059{
Al Viro72bd8662012-06-10 17:17:17 -04002060 return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061}
2062
Christoph Hellwigeead1912007-10-16 23:25:38 -07002063/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07002064 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07002065 * @name: pathname component to lookup
2066 * @base: base directory to lookup from
2067 * @len: maximum length @len should be interpreted to
2068 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07002069 * Note that this routine is purely a helper for filesystem usage and should
2070 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07002071 * nameidata argument is passed to the filesystem methods and a filesystem
2072 * using this helper needs to be prepared for that.
2073 */
James Morris057f6c02007-04-26 00:12:05 -07002074struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2075{
James Morris057f6c02007-04-26 00:12:05 -07002076 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05002077 unsigned int c;
Miklos Szeredicda309d2012-03-26 12:54:21 +02002078 int err;
James Morris057f6c02007-04-26 00:12:05 -07002079
David Woodhouse2f9092e2009-04-20 23:18:37 +01002080 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2081
Al Viro6a96ba52011-03-07 23:49:20 -05002082 this.name = name;
2083 this.len = len;
Linus Torvalds0145acc2012-03-02 14:32:59 -08002084 this.hash = full_name_hash(name, len);
Al Viro6a96ba52011-03-07 23:49:20 -05002085 if (!len)
2086 return ERR_PTR(-EACCES);
2087
Al Viro21d8a152012-11-29 22:17:21 -05002088 if (unlikely(name[0] == '.')) {
2089 if (len < 2 || (len == 2 && name[1] == '.'))
2090 return ERR_PTR(-EACCES);
2091 }
2092
Al Viro6a96ba52011-03-07 23:49:20 -05002093 while (len--) {
2094 c = *(const unsigned char *)name++;
2095 if (c == '/' || c == '\0')
2096 return ERR_PTR(-EACCES);
Al Viro6a96ba52011-03-07 23:49:20 -05002097 }
Al Viro5a202bc2011-03-08 14:17:44 -05002098 /*
2099 * See if the low-level filesystem might want
2100 * to use its own hash..
2101 */
2102 if (base->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002103 int err = base->d_op->d_hash(base, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05002104 if (err < 0)
2105 return ERR_PTR(err);
2106 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07002107
Miklos Szeredicda309d2012-03-26 12:54:21 +02002108 err = inode_permission(base->d_inode, MAY_EXEC);
2109 if (err)
2110 return ERR_PTR(err);
2111
Al Viro72bd8662012-06-10 17:17:17 -04002112 return __lookup_hash(&this, base, 0);
James Morris057f6c02007-04-26 00:12:05 -07002113}
2114
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002115int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2116 struct path *path, int *empty)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117{
Al Viro2d8f3032008-07-22 09:59:21 -04002118 struct nameidata nd;
Jeff Layton91a27b22012-10-10 15:25:28 -04002119 struct filename *tmp = getname_flags(name, flags, empty);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04002122
2123 BUG_ON(flags & LOOKUP_PARENT);
2124
Jeff Layton873f1ee2012-10-10 15:25:29 -04002125 err = filename_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04002127 if (!err)
2128 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 }
2130 return err;
2131}
2132
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002133int user_path_at(int dfd, const char __user *name, unsigned flags,
2134 struct path *path)
2135{
Linus Torvaldsf7493e52012-03-22 16:10:40 -07002136 return user_path_at_empty(dfd, name, flags, path, NULL);
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002137}
2138
Jeff Layton873f1ee2012-10-10 15:25:29 -04002139/*
2140 * NB: most callers don't do anything directly with the reference to the
2141 * to struct filename, but the nd->last pointer points into the name string
2142 * allocated by getname. So we must hold the reference to it until all
2143 * path-walking is complete.
2144 */
Jeff Layton91a27b22012-10-10 15:25:28 -04002145static struct filename *
Jeff Layton9e790bd2012-12-11 12:10:09 -05002146user_path_parent(int dfd, const char __user *path, struct nameidata *nd,
2147 unsigned int flags)
Al Viro2ad94ae2008-07-21 09:32:51 -04002148{
Jeff Layton91a27b22012-10-10 15:25:28 -04002149 struct filename *s = getname(path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002150 int error;
2151
Jeff Layton9e790bd2012-12-11 12:10:09 -05002152 /* only LOOKUP_REVAL is allowed in extra flags */
2153 flags &= LOOKUP_REVAL;
2154
Al Viro2ad94ae2008-07-21 09:32:51 -04002155 if (IS_ERR(s))
Jeff Layton91a27b22012-10-10 15:25:28 -04002156 return s;
Al Viro2ad94ae2008-07-21 09:32:51 -04002157
Jeff Layton9e790bd2012-12-11 12:10:09 -05002158 error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd);
Jeff Layton91a27b22012-10-10 15:25:28 -04002159 if (error) {
Al Viro2ad94ae2008-07-21 09:32:51 -04002160 putname(s);
Jeff Layton91a27b22012-10-10 15:25:28 -04002161 return ERR_PTR(error);
2162 }
Al Viro2ad94ae2008-07-21 09:32:51 -04002163
Jeff Layton91a27b22012-10-10 15:25:28 -04002164 return s;
Al Viro2ad94ae2008-07-21 09:32:51 -04002165}
2166
Jeff Layton80334262013-07-26 06:23:25 -04002167/**
Al Viro197df042013-09-08 14:03:27 -04002168 * mountpoint_last - look up last component for umount
Jeff Layton80334262013-07-26 06:23:25 -04002169 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2170 * @path: pointer to container for result
2171 *
2172 * This is a special lookup_last function just for umount. In this case, we
2173 * need to resolve the path without doing any revalidation.
2174 *
2175 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2176 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2177 * in almost all cases, this lookup will be served out of the dcache. The only
2178 * cases where it won't are if nd->last refers to a symlink or the path is
2179 * bogus and it doesn't exist.
2180 *
2181 * Returns:
2182 * -error: if there was an error during lookup. This includes -ENOENT if the
2183 * lookup found a negative dentry. The nd->path reference will also be
2184 * put in this case.
2185 *
2186 * 0: if we successfully resolved nd->path and found it to not to be a
2187 * symlink that needs to be followed. "path" will also be populated.
2188 * The nd->path reference will also be put.
2189 *
2190 * 1: if we successfully resolved nd->last and found it to be a symlink
2191 * that needs to be followed. "path" will be populated with the path
2192 * to the link, and nd->path will *not* be put.
2193 */
2194static int
Al Viro197df042013-09-08 14:03:27 -04002195mountpoint_last(struct nameidata *nd, struct path *path)
Jeff Layton80334262013-07-26 06:23:25 -04002196{
2197 int error = 0;
2198 struct dentry *dentry;
2199 struct dentry *dir = nd->path.dentry;
2200
Al Viro35759522013-09-08 13:41:33 -04002201 /* If we're in rcuwalk, drop out of it to handle last component */
2202 if (nd->flags & LOOKUP_RCU) {
2203 if (unlazy_walk(nd, NULL)) {
2204 error = -ECHILD;
2205 goto out;
2206 }
Jeff Layton80334262013-07-26 06:23:25 -04002207 }
2208
2209 nd->flags &= ~LOOKUP_PARENT;
2210
2211 if (unlikely(nd->last_type != LAST_NORM)) {
2212 error = handle_dots(nd, nd->last_type);
Al Viro35759522013-09-08 13:41:33 -04002213 if (error)
2214 goto out;
2215 dentry = dget(nd->path.dentry);
2216 goto done;
Jeff Layton80334262013-07-26 06:23:25 -04002217 }
2218
2219 mutex_lock(&dir->d_inode->i_mutex);
2220 dentry = d_lookup(dir, &nd->last);
2221 if (!dentry) {
2222 /*
2223 * No cached dentry. Mounted dentries are pinned in the cache,
2224 * so that means that this dentry is probably a symlink or the
2225 * path doesn't actually point to a mounted dentry.
2226 */
2227 dentry = d_alloc(dir, &nd->last);
2228 if (!dentry) {
2229 error = -ENOMEM;
Dave Jonesbcceeeb2013-09-10 17:04:25 -04002230 mutex_unlock(&dir->d_inode->i_mutex);
Al Viro35759522013-09-08 13:41:33 -04002231 goto out;
Jeff Layton80334262013-07-26 06:23:25 -04002232 }
Al Viro35759522013-09-08 13:41:33 -04002233 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2234 error = PTR_ERR(dentry);
Dave Jonesbcceeeb2013-09-10 17:04:25 -04002235 if (IS_ERR(dentry)) {
2236 mutex_unlock(&dir->d_inode->i_mutex);
Al Viro35759522013-09-08 13:41:33 -04002237 goto out;
Dave Jonesbcceeeb2013-09-10 17:04:25 -04002238 }
Jeff Layton80334262013-07-26 06:23:25 -04002239 }
2240 mutex_unlock(&dir->d_inode->i_mutex);
2241
Al Viro35759522013-09-08 13:41:33 -04002242done:
2243 if (!dentry->d_inode) {
2244 error = -ENOENT;
2245 dput(dentry);
2246 goto out;
Jeff Layton80334262013-07-26 06:23:25 -04002247 }
Al Viro35759522013-09-08 13:41:33 -04002248 path->dentry = dentry;
2249 path->mnt = mntget(nd->path.mnt);
David Howellsb18825a2013-09-12 19:22:53 +01002250 if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW))
Al Viro35759522013-09-08 13:41:33 -04002251 return 1;
2252 follow_mount(path);
2253 error = 0;
2254out:
Jeff Layton80334262013-07-26 06:23:25 -04002255 terminate_walk(nd);
2256 return error;
2257}
2258
2259/**
Al Viro197df042013-09-08 14:03:27 -04002260 * path_mountpoint - look up a path to be umounted
Jeff Layton80334262013-07-26 06:23:25 -04002261 * @dfd: directory file descriptor to start walk from
2262 * @name: full pathname to walk
Randy Dunlap606d6fe2013-10-19 14:56:55 -07002263 * @path: pointer to container for result
Jeff Layton80334262013-07-26 06:23:25 -04002264 * @flags: lookup flags
Jeff Layton80334262013-07-26 06:23:25 -04002265 *
2266 * Look up the given name, but don't attempt to revalidate the last component.
Randy Dunlap606d6fe2013-10-19 14:56:55 -07002267 * Returns 0 and "path" will be valid on success; Returns error otherwise.
Jeff Layton80334262013-07-26 06:23:25 -04002268 */
2269static int
Al Viro197df042013-09-08 14:03:27 -04002270path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags)
Jeff Layton80334262013-07-26 06:23:25 -04002271{
2272 struct file *base = NULL;
2273 struct nameidata nd;
2274 int err;
2275
2276 err = path_init(dfd, name, flags | LOOKUP_PARENT, &nd, &base);
2277 if (unlikely(err))
2278 return err;
2279
2280 current->total_link_count = 0;
2281 err = link_path_walk(name, &nd);
2282 if (err)
2283 goto out;
2284
Al Viro197df042013-09-08 14:03:27 -04002285 err = mountpoint_last(&nd, path);
Jeff Layton80334262013-07-26 06:23:25 -04002286 while (err > 0) {
2287 void *cookie;
2288 struct path link = *path;
2289 err = may_follow_link(&link, &nd);
2290 if (unlikely(err))
2291 break;
2292 nd.flags |= LOOKUP_PARENT;
2293 err = follow_link(&link, &nd, &cookie);
2294 if (err)
2295 break;
Al Viro197df042013-09-08 14:03:27 -04002296 err = mountpoint_last(&nd, path);
Jeff Layton80334262013-07-26 06:23:25 -04002297 put_link(&nd, &link, cookie);
2298 }
2299out:
2300 if (base)
2301 fput(base);
2302
2303 if (nd.root.mnt && !(nd.flags & LOOKUP_ROOT))
2304 path_put(&nd.root);
2305
2306 return err;
2307}
2308
Al Viro2d864652013-09-08 20:18:44 -04002309static int
2310filename_mountpoint(int dfd, struct filename *s, struct path *path,
2311 unsigned int flags)
2312{
2313 int error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU);
2314 if (unlikely(error == -ECHILD))
2315 error = path_mountpoint(dfd, s->name, path, flags);
2316 if (unlikely(error == -ESTALE))
2317 error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL);
2318 if (likely(!error))
2319 audit_inode(s, path->dentry, 0);
2320 return error;
2321}
2322
Jeff Layton80334262013-07-26 06:23:25 -04002323/**
Al Viro197df042013-09-08 14:03:27 -04002324 * user_path_mountpoint_at - lookup a path from userland in order to umount it
Jeff Layton80334262013-07-26 06:23:25 -04002325 * @dfd: directory file descriptor
2326 * @name: pathname from userland
2327 * @flags: lookup flags
2328 * @path: pointer to container to hold result
2329 *
2330 * A umount is a special case for path walking. We're not actually interested
2331 * in the inode in this situation, and ESTALE errors can be a problem. We
2332 * simply want track down the dentry and vfsmount attached at the mountpoint
2333 * and avoid revalidating the last component.
2334 *
2335 * Returns 0 and populates "path" on success.
2336 */
2337int
Al Viro197df042013-09-08 14:03:27 -04002338user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
Jeff Layton80334262013-07-26 06:23:25 -04002339 struct path *path)
2340{
2341 struct filename *s = getname(name);
2342 int error;
Jeff Layton80334262013-07-26 06:23:25 -04002343 if (IS_ERR(s))
2344 return PTR_ERR(s);
Al Viro2d864652013-09-08 20:18:44 -04002345 error = filename_mountpoint(dfd, s, path, flags);
Jeff Layton80334262013-07-26 06:23:25 -04002346 putname(s);
2347 return error;
2348}
2349
Al Viro2d864652013-09-08 20:18:44 -04002350int
2351kern_path_mountpoint(int dfd, const char *name, struct path *path,
2352 unsigned int flags)
2353{
2354 struct filename s = {.name = name};
2355 return filename_mountpoint(dfd, &s, path, flags);
2356}
2357EXPORT_SYMBOL(kern_path_mountpoint);
2358
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359/*
2360 * It's inline, so penalty for filesystems that don't use sticky bit is
2361 * minimal.
2362 */
2363static inline int check_sticky(struct inode *dir, struct inode *inode)
2364{
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002365 kuid_t fsuid = current_fsuid();
David Howellsda9592e2008-11-14 10:39:05 +11002366
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367 if (!(dir->i_mode & S_ISVTX))
2368 return 0;
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002369 if (uid_eq(inode->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370 return 0;
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002371 if (uid_eq(dir->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372 return 0;
Eric W. Biederman1a48e2a2011-11-14 16:24:06 -08002373 return !inode_capable(inode, CAP_FOWNER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374}
2375
2376/*
2377 * Check whether we can remove a link victim from directory dir, check
2378 * whether the type of victim is right.
2379 * 1. We can't do it if dir is read-only (done in permission())
2380 * 2. We should have write and exec permissions on dir
2381 * 3. We can't remove anything from append-only dir
2382 * 4. We can't do anything with immutable dir (done in permission())
2383 * 5. If the sticky bit on dir is set we should either
2384 * a. be owner of dir, or
2385 * b. be owner of victim, or
2386 * c. have CAP_FOWNER capability
2387 * 6. If the victim is append-only or immutable we can't do antyhing with
2388 * links pointing to it.
2389 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2390 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2391 * 9. We can't remove a root or mountpoint.
2392 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2393 * nfs_async_unlink().
2394 */
David Howellsb18825a2013-09-12 19:22:53 +01002395static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396{
David Howellsb18825a2013-09-12 19:22:53 +01002397 struct inode *inode = victim->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398 int error;
2399
David Howellsb18825a2013-09-12 19:22:53 +01002400 if (d_is_negative(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401 return -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01002402 BUG_ON(!inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403
2404 BUG_ON(victim->d_parent->d_inode != dir);
Jeff Layton4fa6b5e2012-10-10 15:25:25 -04002405 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406
Al Virof419a2e2008-07-22 00:07:17 -04002407 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408 if (error)
2409 return error;
2410 if (IS_APPEND(dir))
2411 return -EPERM;
David Howellsb18825a2013-09-12 19:22:53 +01002412
2413 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2414 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415 return -EPERM;
2416 if (isdir) {
David Howellsb18825a2013-09-12 19:22:53 +01002417 if (!d_is_directory(victim) && !d_is_autodir(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418 return -ENOTDIR;
2419 if (IS_ROOT(victim))
2420 return -EBUSY;
David Howellsb18825a2013-09-12 19:22:53 +01002421 } else if (d_is_directory(victim) || d_is_autodir(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422 return -EISDIR;
2423 if (IS_DEADDIR(dir))
2424 return -ENOENT;
2425 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2426 return -EBUSY;
2427 return 0;
2428}
2429
2430/* Check whether we can create an object with dentry child in directory
2431 * dir.
2432 * 1. We can't do it if child already exists (open has special treatment for
2433 * this case, but since we are inlined it's OK)
2434 * 2. We can't do it if dir is read-only (done in permission())
2435 * 3. We should have write and exec permissions on dir
2436 * 4. We can't do it if dir is immutable (done in permission())
2437 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02002438static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439{
Jeff Layton14e972b2013-05-08 10:25:58 -04002440 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441 if (child->d_inode)
2442 return -EEXIST;
2443 if (IS_DEADDIR(dir))
2444 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04002445 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446}
2447
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448/*
2449 * p1 and p2 should be directories on the same fs.
2450 */
2451struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2452{
2453 struct dentry *p;
2454
2455 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07002456 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457 return NULL;
2458 }
2459
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002460 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002462 p = d_ancestor(p2, p1);
2463 if (p) {
2464 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2465 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2466 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467 }
2468
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002469 p = d_ancestor(p1, p2);
2470 if (p) {
2471 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2472 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2473 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474 }
2475
Ingo Molnarf2eace22006-07-03 00:25:05 -07002476 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2477 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478 return NULL;
2479}
2480
2481void unlock_rename(struct dentry *p1, struct dentry *p2)
2482{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002483 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002485 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002486 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487 }
2488}
2489
Al Viro4acdaf22011-07-26 01:42:34 -04002490int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viro312b63f2012-06-10 18:09:36 -04002491 bool want_excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002493 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494 if (error)
2495 return error;
2496
Al Viroacfa4382008-12-04 10:06:33 -05002497 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498 return -EACCES; /* shouldn't it be ENOSYS? */
2499 mode &= S_IALLUGO;
2500 mode |= S_IFREG;
2501 error = security_inode_create(dir, dentry, mode);
2502 if (error)
2503 return error;
Al Viro312b63f2012-06-10 18:09:36 -04002504 error = dir->i_op->create(dir, dentry, mode, want_excl);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002505 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002506 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507 return error;
2508}
2509
Al Viro73d049a2011-03-11 12:08:24 -05002510static int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002512 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513 struct inode *inode = dentry->d_inode;
2514 int error;
2515
Al Virobcda7652011-03-13 16:42:14 -04002516 /* O_PATH? */
2517 if (!acc_mode)
2518 return 0;
2519
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520 if (!inode)
2521 return -ENOENT;
2522
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002523 switch (inode->i_mode & S_IFMT) {
2524 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002526 case S_IFDIR:
2527 if (acc_mode & MAY_WRITE)
2528 return -EISDIR;
2529 break;
2530 case S_IFBLK:
2531 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002532 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002534 /*FALLTHRU*/
2535 case S_IFIFO:
2536 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002538 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002539 }
Dave Hansenb41572e2007-10-16 23:31:14 -07002540
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002541 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07002542 if (error)
2543 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05002544
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545 /*
2546 * An append-only file must be opened in append mode for writing.
2547 */
2548 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05002549 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05002550 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05002552 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553 }
2554
2555 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002556 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05002557 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558
J. Bruce Fieldsf3c7691e2011-09-21 10:58:13 -04002559 return 0;
Al Viro7715b522009-12-16 03:54:00 -05002560}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561
Jeff Laytone1181ee2010-12-07 16:19:50 -05002562static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002563{
Jeff Laytone1181ee2010-12-07 16:19:50 -05002564 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002565 struct inode *inode = path->dentry->d_inode;
2566 int error = get_write_access(inode);
2567 if (error)
2568 return error;
2569 /*
2570 * Refuse to truncate files with mandatory locks held on them.
2571 */
2572 error = locks_verify_locked(inode);
2573 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002574 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002575 if (!error) {
2576 error = do_truncate(path->dentry, 0,
2577 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002578 filp);
Al Viro7715b522009-12-16 03:54:00 -05002579 }
2580 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002581 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582}
2583
Dave Hansend57999e2008-02-15 14:37:27 -08002584static inline int open_to_namei_flags(int flag)
2585{
Al Viro8a5e9292011-06-25 19:15:54 -04002586 if ((flag & O_ACCMODE) == 3)
2587 flag--;
Dave Hansend57999e2008-02-15 14:37:27 -08002588 return flag;
2589}
2590
Miklos Szeredid18e9002012-06-05 15:10:17 +02002591static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2592{
2593 int error = security_path_mknod(dir, dentry, mode, 0);
2594 if (error)
2595 return error;
2596
2597 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2598 if (error)
2599 return error;
2600
2601 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2602}
2603
David Howells1acf0af2012-06-14 16:13:46 +01002604/*
2605 * Attempt to atomically look up, create and open a file from a negative
2606 * dentry.
2607 *
2608 * Returns 0 if successful. The file will have been created and attached to
2609 * @file by the filesystem calling finish_open().
2610 *
2611 * Returns 1 if the file was looked up only or didn't need creating. The
2612 * caller will need to perform the open themselves. @path will have been
2613 * updated to point to the new dentry. This may be negative.
2614 *
2615 * Returns an error code otherwise.
2616 */
Al Viro2675a4e2012-06-22 12:41:10 +04002617static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2618 struct path *path, struct file *file,
2619 const struct open_flags *op,
Al Viro64894cf2012-07-31 00:53:35 +04002620 bool got_write, bool need_lookup,
Al Viro2675a4e2012-06-22 12:41:10 +04002621 int *opened)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002622{
2623 struct inode *dir = nd->path.dentry->d_inode;
2624 unsigned open_flag = open_to_namei_flags(op->open_flag);
2625 umode_t mode;
2626 int error;
2627 int acc_mode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002628 int create_error = 0;
2629 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
Miklos Szeredi116cc022013-09-16 14:52:05 +02002630 bool excl;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002631
2632 BUG_ON(dentry->d_inode);
2633
2634 /* Don't create child dentry for a dead directory. */
2635 if (unlikely(IS_DEADDIR(dir))) {
Al Viro2675a4e2012-06-22 12:41:10 +04002636 error = -ENOENT;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002637 goto out;
2638 }
2639
Miklos Szeredi62b259d2012-08-15 13:01:24 +02002640 mode = op->mode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002641 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2642 mode &= ~current_umask();
2643
Miklos Szeredi116cc022013-09-16 14:52:05 +02002644 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2645 if (excl)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002646 open_flag &= ~O_TRUNC;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002647
2648 /*
2649 * Checking write permission is tricky, bacuse we don't know if we are
2650 * going to actually need it: O_CREAT opens should work as long as the
2651 * file exists. But checking existence breaks atomicity. The trick is
2652 * to check access and if not granted clear O_CREAT from the flags.
2653 *
2654 * Another problem is returing the "right" error value (e.g. for an
2655 * O_EXCL open we want to return EEXIST not EROFS).
2656 */
Al Viro64894cf2012-07-31 00:53:35 +04002657 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2658 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2659 if (!(open_flag & O_CREAT)) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002660 /*
2661 * No O_CREATE -> atomicity not a requirement -> fall
2662 * back to lookup + open
2663 */
2664 goto no_open;
2665 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2666 /* Fall back and fail with the right error */
Al Viro64894cf2012-07-31 00:53:35 +04002667 create_error = -EROFS;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002668 goto no_open;
2669 } else {
2670 /* No side effects, safe to clear O_CREAT */
Al Viro64894cf2012-07-31 00:53:35 +04002671 create_error = -EROFS;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002672 open_flag &= ~O_CREAT;
2673 }
2674 }
2675
2676 if (open_flag & O_CREAT) {
Miklos Szeredi38227f72012-08-15 13:01:24 +02002677 error = may_o_create(&nd->path, dentry, mode);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002678 if (error) {
2679 create_error = error;
2680 if (open_flag & O_EXCL)
2681 goto no_open;
2682 open_flag &= ~O_CREAT;
2683 }
2684 }
2685
2686 if (nd->flags & LOOKUP_DIRECTORY)
2687 open_flag |= O_DIRECTORY;
2688
Al Viro30d90492012-06-22 12:40:19 +04002689 file->f_path.dentry = DENTRY_NOT_SET;
2690 file->f_path.mnt = nd->path.mnt;
2691 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
Al Viro47237682012-06-10 05:01:45 -04002692 opened);
Al Virod9585272012-06-22 12:39:14 +04002693 if (error < 0) {
Al Virod9585272012-06-22 12:39:14 +04002694 if (create_error && error == -ENOENT)
2695 error = create_error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002696 goto out;
2697 }
2698
Al Virod9585272012-06-22 12:39:14 +04002699 if (error) { /* returned 1, that is */
Al Viro30d90492012-06-22 12:40:19 +04002700 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
Al Viro2675a4e2012-06-22 12:41:10 +04002701 error = -EIO;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002702 goto out;
2703 }
Al Viro30d90492012-06-22 12:40:19 +04002704 if (file->f_path.dentry) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002705 dput(dentry);
Al Viro30d90492012-06-22 12:40:19 +04002706 dentry = file->f_path.dentry;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002707 }
Al Viro03da6332013-09-16 19:22:33 -04002708 if (*opened & FILE_CREATED)
2709 fsnotify_create(dir, dentry);
2710 if (!dentry->d_inode) {
2711 WARN_ON(*opened & FILE_CREATED);
2712 if (create_error) {
2713 error = create_error;
2714 goto out;
2715 }
2716 } else {
2717 if (excl && !(*opened & FILE_CREATED)) {
2718 error = -EEXIST;
2719 goto out;
2720 }
Sage Weil62b2ce92012-08-15 13:30:12 -07002721 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02002722 goto looked_up;
2723 }
2724
2725 /*
2726 * We didn't have the inode before the open, so check open permission
2727 * here.
2728 */
Al Viro03da6332013-09-16 19:22:33 -04002729 acc_mode = op->acc_mode;
2730 if (*opened & FILE_CREATED) {
2731 WARN_ON(!(open_flag & O_CREAT));
2732 fsnotify_create(dir, dentry);
2733 acc_mode = MAY_OPEN;
2734 }
Al Viro2675a4e2012-06-22 12:41:10 +04002735 error = may_open(&file->f_path, acc_mode, open_flag);
2736 if (error)
2737 fput(file);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002738
2739out:
2740 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04002741 return error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002742
Miklos Szeredid18e9002012-06-05 15:10:17 +02002743no_open:
2744 if (need_lookup) {
Al Viro72bd8662012-06-10 17:17:17 -04002745 dentry = lookup_real(dir, dentry, nd->flags);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002746 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002747 return PTR_ERR(dentry);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002748
2749 if (create_error) {
2750 int open_flag = op->open_flag;
2751
Al Viro2675a4e2012-06-22 12:41:10 +04002752 error = create_error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002753 if ((open_flag & O_EXCL)) {
2754 if (!dentry->d_inode)
2755 goto out;
2756 } else if (!dentry->d_inode) {
2757 goto out;
2758 } else if ((open_flag & O_TRUNC) &&
2759 S_ISREG(dentry->d_inode->i_mode)) {
2760 goto out;
2761 }
2762 /* will fail later, go on to get the right error */
2763 }
2764 }
2765looked_up:
2766 path->dentry = dentry;
2767 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04002768 return 1;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002769}
2770
Nick Piggin31e6b012011-01-07 17:49:52 +11002771/*
David Howells1acf0af2012-06-14 16:13:46 +01002772 * Look up and maybe create and open the last component.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002773 *
2774 * Must be called with i_mutex held on parent.
2775 *
David Howells1acf0af2012-06-14 16:13:46 +01002776 * Returns 0 if the file was successfully atomically created (if necessary) and
2777 * opened. In this case the file will be returned attached to @file.
2778 *
2779 * Returns 1 if the file was not completely opened at this time, though lookups
2780 * and creations will have been performed and the dentry returned in @path will
2781 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2782 * specified then a negative dentry may be returned.
2783 *
2784 * An error code is returned otherwise.
2785 *
2786 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2787 * cleared otherwise prior to returning.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002788 */
Al Viro2675a4e2012-06-22 12:41:10 +04002789static int lookup_open(struct nameidata *nd, struct path *path,
2790 struct file *file,
2791 const struct open_flags *op,
Al Viro64894cf2012-07-31 00:53:35 +04002792 bool got_write, int *opened)
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002793{
2794 struct dentry *dir = nd->path.dentry;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002795 struct inode *dir_inode = dir->d_inode;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002796 struct dentry *dentry;
2797 int error;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002798 bool need_lookup;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002799
Al Viro47237682012-06-10 05:01:45 -04002800 *opened &= ~FILE_CREATED;
Al Viro201f9562012-06-22 12:42:10 +04002801 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002802 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002803 return PTR_ERR(dentry);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002804
Miklos Szeredid18e9002012-06-05 15:10:17 +02002805 /* Cached positive dentry: will open in f_op->open */
2806 if (!need_lookup && dentry->d_inode)
2807 goto out_no_open;
2808
2809 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
Al Viro64894cf2012-07-31 00:53:35 +04002810 return atomic_open(nd, dentry, path, file, op, got_write,
Al Viro47237682012-06-10 05:01:45 -04002811 need_lookup, opened);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002812 }
2813
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002814 if (need_lookup) {
2815 BUG_ON(dentry->d_inode);
2816
Al Viro72bd8662012-06-10 17:17:17 -04002817 dentry = lookup_real(dir_inode, dentry, nd->flags);
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002818 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002819 return PTR_ERR(dentry);
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002820 }
2821
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002822 /* Negative dentry, just create the file */
2823 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2824 umode_t mode = op->mode;
2825 if (!IS_POSIXACL(dir->d_inode))
2826 mode &= ~current_umask();
2827 /*
2828 * This write is needed to ensure that a
2829 * rw->ro transition does not occur between
2830 * the time when the file is created and when
2831 * a permanent write count is taken through
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002832 * the 'struct file' in finish_open().
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002833 */
Al Viro64894cf2012-07-31 00:53:35 +04002834 if (!got_write) {
2835 error = -EROFS;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002836 goto out_dput;
Al Viro64894cf2012-07-31 00:53:35 +04002837 }
Al Viro47237682012-06-10 05:01:45 -04002838 *opened |= FILE_CREATED;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002839 error = security_path_mknod(&nd->path, dentry, mode, 0);
2840 if (error)
2841 goto out_dput;
Al Viro312b63f2012-06-10 18:09:36 -04002842 error = vfs_create(dir->d_inode, dentry, mode,
2843 nd->flags & LOOKUP_EXCL);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002844 if (error)
2845 goto out_dput;
2846 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02002847out_no_open:
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002848 path->dentry = dentry;
2849 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04002850 return 1;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002851
2852out_dput:
2853 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04002854 return error;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002855}
2856
2857/*
Al Virofe2d35f2011-03-05 22:58:25 -05002858 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11002859 */
Al Viro2675a4e2012-06-22 12:41:10 +04002860static int do_last(struct nameidata *nd, struct path *path,
2861 struct file *file, const struct open_flags *op,
Jeff Layton669abf42012-10-10 16:43:10 -04002862 int *opened, struct filename *name)
Al Virofb1cc552009-12-24 01:58:28 -05002863{
Al Viroa1e28032009-12-24 02:12:06 -05002864 struct dentry *dir = nd->path.dentry;
Al Viroca344a892011-03-09 00:36:45 -05002865 int open_flag = op->open_flag;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002866 bool will_truncate = (open_flag & O_TRUNC) != 0;
Al Viro64894cf2012-07-31 00:53:35 +04002867 bool got_write = false;
Al Virobcda7652011-03-13 16:42:14 -04002868 int acc_mode = op->acc_mode;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02002869 struct inode *inode;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002870 bool symlink_ok = false;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002871 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2872 bool retried = false;
Al Viro16c2cd72011-02-22 15:50:10 -05002873 int error;
Al Virofb1cc552009-12-24 01:58:28 -05002874
Al Viroc3e380b2011-02-23 13:39:45 -05002875 nd->flags &= ~LOOKUP_PARENT;
2876 nd->flags |= op->intent;
2877
Al Virobc77daa2013-06-06 09:12:33 -04002878 if (nd->last_type != LAST_NORM) {
Al Virofe2d35f2011-03-05 22:58:25 -05002879 error = handle_dots(nd, nd->last_type);
2880 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002881 return error;
Miklos Szeredie83db162012-06-05 15:10:29 +02002882 goto finish_open;
Al Viro1f36f772009-12-26 10:56:19 -05002883 }
Al Viro67ee3ad2009-12-26 07:01:01 -05002884
Al Viroca344a892011-03-09 00:36:45 -05002885 if (!(open_flag & O_CREAT)) {
Al Virofe2d35f2011-03-05 22:58:25 -05002886 if (nd->last.name[nd->last.len])
2887 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
Al Virobcda7652011-03-13 16:42:14 -04002888 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002889 symlink_ok = true;
Al Virofe2d35f2011-03-05 22:58:25 -05002890 /* we _can_ be in RCU mode here */
Al Viroe97cdc82013-01-24 18:16:00 -05002891 error = lookup_fast(nd, path, &inode);
Miklos Szeredi71574862012-06-05 15:10:14 +02002892 if (likely(!error))
2893 goto finish_lookup;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02002894
Miklos Szeredi71574862012-06-05 15:10:14 +02002895 if (error < 0)
Al Viro2675a4e2012-06-22 12:41:10 +04002896 goto out;
Miklos Szeredi37d7fff2012-06-05 15:10:12 +02002897
Miklos Szeredi71574862012-06-05 15:10:14 +02002898 BUG_ON(nd->inode != dir->d_inode);
Miklos Szeredib6183df2012-06-05 15:10:13 +02002899 } else {
2900 /* create side of things */
2901 /*
2902 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2903 * has been cleared when we got to the last component we are
2904 * about to look up
2905 */
2906 error = complete_walk(nd);
2907 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002908 return error;
Miklos Szeredib6183df2012-06-05 15:10:13 +02002909
Jeff Layton33e22082013-04-12 15:16:32 -04002910 audit_inode(name, dir, LOOKUP_PARENT);
Miklos Szeredib6183df2012-06-05 15:10:13 +02002911 error = -EISDIR;
2912 /* trailing slashes? */
2913 if (nd->last.name[nd->last.len])
Al Viro2675a4e2012-06-22 12:41:10 +04002914 goto out;
Al Virofe2d35f2011-03-05 22:58:25 -05002915 }
2916
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002917retry_lookup:
Al Viro64894cf2012-07-31 00:53:35 +04002918 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2919 error = mnt_want_write(nd->path.mnt);
2920 if (!error)
2921 got_write = true;
2922 /*
2923 * do _not_ fail yet - we might not need that or fail with
2924 * a different error; let lookup_open() decide; we'll be
2925 * dropping this one anyway.
2926 */
2927 }
Al Viroa1e28032009-12-24 02:12:06 -05002928 mutex_lock(&dir->d_inode->i_mutex);
Al Viro64894cf2012-07-31 00:53:35 +04002929 error = lookup_open(nd, path, file, op, got_write, opened);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002930 mutex_unlock(&dir->d_inode->i_mutex);
Al Viroa1e28032009-12-24 02:12:06 -05002931
Al Viro2675a4e2012-06-22 12:41:10 +04002932 if (error <= 0) {
2933 if (error)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002934 goto out;
2935
Al Viro47237682012-06-10 05:01:45 -04002936 if ((*opened & FILE_CREATED) ||
Al Viro496ad9a2013-01-23 17:07:38 -05002937 !S_ISREG(file_inode(file)->i_mode))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002938 will_truncate = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002939
Jeff Laytonadb5c242012-10-10 16:43:13 -04002940 audit_inode(name, file->f_path.dentry, 0);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002941 goto opened;
2942 }
Al Virofb1cc552009-12-24 01:58:28 -05002943
Al Viro47237682012-06-10 05:01:45 -04002944 if (*opened & FILE_CREATED) {
Al Viro9b44f1b2011-03-09 00:17:27 -05002945 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05002946 open_flag &= ~O_TRUNC;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002947 will_truncate = false;
Al Virobcda7652011-03-13 16:42:14 -04002948 acc_mode = MAY_OPEN;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002949 path_to_nameidata(path, nd);
Miklos Szeredie83db162012-06-05 15:10:29 +02002950 goto finish_open_created;
Al Virofb1cc552009-12-24 01:58:28 -05002951 }
2952
2953 /*
Jeff Layton3134f372012-07-25 10:19:47 -04002954 * create/update audit record if it already exists.
Al Virofb1cc552009-12-24 01:58:28 -05002955 */
David Howellsb18825a2013-09-12 19:22:53 +01002956 if (d_is_positive(path->dentry))
Jeff Laytonadb5c242012-10-10 16:43:13 -04002957 audit_inode(name, path->dentry, 0);
Al Virofb1cc552009-12-24 01:58:28 -05002958
Miklos Szeredid18e9002012-06-05 15:10:17 +02002959 /*
2960 * If atomic_open() acquired write access it is dropped now due to
2961 * possible mount and symlink following (this might be optimized away if
2962 * necessary...)
2963 */
Al Viro64894cf2012-07-31 00:53:35 +04002964 if (got_write) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002965 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04002966 got_write = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002967 }
2968
Al Virofb1cc552009-12-24 01:58:28 -05002969 error = -EEXIST;
Al Virof8310c52012-07-30 11:50:30 +04002970 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
Al Virofb1cc552009-12-24 01:58:28 -05002971 goto exit_dput;
2972
David Howells9875cf82011-01-14 18:45:21 +00002973 error = follow_managed(path, nd->flags);
2974 if (error < 0)
2975 goto exit_dput;
Al Virofb1cc552009-12-24 01:58:28 -05002976
Al Viroa3fbbde2011-11-07 21:21:26 +00002977 if (error)
2978 nd->flags |= LOOKUP_JUMPED;
2979
Miklos Szeredidecf3402012-05-21 17:30:08 +02002980 BUG_ON(nd->flags & LOOKUP_RCU);
2981 inode = path->dentry->d_inode;
Miklos Szeredi5f5daac2012-05-21 17:30:14 +02002982finish_lookup:
2983 /* we _can_ be in RCU mode here */
Al Virofb1cc552009-12-24 01:58:28 -05002984 error = -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01002985 if (d_is_negative(path->dentry)) {
Miklos Szeredi54c33e72012-05-21 17:30:10 +02002986 path_to_nameidata(path, nd);
Al Viro2675a4e2012-06-22 12:41:10 +04002987 goto out;
Miklos Szeredi54c33e72012-05-21 17:30:10 +02002988 }
Al Viro9e67f362009-12-26 07:04:50 -05002989
David Howellsb18825a2013-09-12 19:22:53 +01002990 if (should_follow_link(path->dentry, !symlink_ok)) {
Miklos Szeredid45ea862012-05-21 17:30:09 +02002991 if (nd->flags & LOOKUP_RCU) {
2992 if (unlikely(unlazy_walk(nd, path->dentry))) {
2993 error = -ECHILD;
Al Viro2675a4e2012-06-22 12:41:10 +04002994 goto out;
Miklos Szeredid45ea862012-05-21 17:30:09 +02002995 }
2996 }
2997 BUG_ON(inode != path->dentry->d_inode);
Al Viro2675a4e2012-06-22 12:41:10 +04002998 return 1;
Miklos Szeredid45ea862012-05-21 17:30:09 +02002999 }
Al Virofb1cc552009-12-24 01:58:28 -05003000
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003001 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
3002 path_to_nameidata(path, nd);
3003 } else {
3004 save_parent.dentry = nd->path.dentry;
3005 save_parent.mnt = mntget(path->mnt);
3006 nd->path.dentry = path->dentry;
3007
3008 }
Miklos Szeredidecf3402012-05-21 17:30:08 +02003009 nd->inode = inode;
Al Viroa3fbbde2011-11-07 21:21:26 +00003010 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
Al Virobc77daa2013-06-06 09:12:33 -04003011finish_open:
Al Viroa3fbbde2011-11-07 21:21:26 +00003012 error = complete_walk(nd);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003013 if (error) {
3014 path_put(&save_parent);
Al Viro2675a4e2012-06-22 12:41:10 +04003015 return error;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003016 }
Al Virobc77daa2013-06-06 09:12:33 -04003017 audit_inode(name, nd->path.dentry, 0);
Al Virofb1cc552009-12-24 01:58:28 -05003018 error = -EISDIR;
David Howellsb18825a2013-09-12 19:22:53 +01003019 if ((open_flag & O_CREAT) &&
3020 (d_is_directory(nd->path.dentry) || d_is_autodir(nd->path.dentry)))
Al Viro2675a4e2012-06-22 12:41:10 +04003021 goto out;
Miklos Szerediaf2f5542012-05-21 17:30:11 +02003022 error = -ENOTDIR;
David Howellsb18825a2013-09-12 19:22:53 +01003023 if ((nd->flags & LOOKUP_DIRECTORY) && !d_is_directory(nd->path.dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04003024 goto out;
Al Viro6c0d46c2011-03-09 00:59:59 -05003025 if (!S_ISREG(nd->inode->i_mode))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003026 will_truncate = false;
Al Viro6c0d46c2011-03-09 00:59:59 -05003027
Al Viro0f9d1a12011-03-09 00:13:14 -05003028 if (will_truncate) {
3029 error = mnt_want_write(nd->path.mnt);
3030 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003031 goto out;
Al Viro64894cf2012-07-31 00:53:35 +04003032 got_write = true;
Al Viro0f9d1a12011-03-09 00:13:14 -05003033 }
Miklos Szeredie83db162012-06-05 15:10:29 +02003034finish_open_created:
Al Virobcda7652011-03-13 16:42:14 -04003035 error = may_open(&nd->path, acc_mode, open_flag);
Al Viroca344a892011-03-09 00:36:45 -05003036 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003037 goto out;
Al Viro30d90492012-06-22 12:40:19 +04003038 file->f_path.mnt = nd->path.mnt;
3039 error = finish_open(file, nd->path.dentry, NULL, opened);
3040 if (error) {
Al Viro30d90492012-06-22 12:40:19 +04003041 if (error == -EOPENSTALE)
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003042 goto stale_open;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003043 goto out;
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003044 }
Miklos Szeredia8277b92012-06-05 15:10:32 +02003045opened:
Al Viro2675a4e2012-06-22 12:41:10 +04003046 error = open_check_o_direct(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003047 if (error)
3048 goto exit_fput;
Al Viro2675a4e2012-06-22 12:41:10 +04003049 error = ima_file_check(file, op->acc_mode);
Miklos Szerediaa4caad2012-06-05 15:10:28 +02003050 if (error)
3051 goto exit_fput;
3052
3053 if (will_truncate) {
Al Viro2675a4e2012-06-22 12:41:10 +04003054 error = handle_truncate(file);
Miklos Szerediaa4caad2012-06-05 15:10:28 +02003055 if (error)
3056 goto exit_fput;
Al Viro0f9d1a12011-03-09 00:13:14 -05003057 }
Al Viroca344a892011-03-09 00:36:45 -05003058out:
Al Viro64894cf2012-07-31 00:53:35 +04003059 if (got_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05003060 mnt_drop_write(nd->path.mnt);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003061 path_put(&save_parent);
Miklos Szeredie276ae62012-05-21 17:30:06 +02003062 terminate_walk(nd);
Al Viro2675a4e2012-06-22 12:41:10 +04003063 return error;
Al Virofb1cc552009-12-24 01:58:28 -05003064
Al Virofb1cc552009-12-24 01:58:28 -05003065exit_dput:
3066 path_put_conditional(path, nd);
Al Viroca344a892011-03-09 00:36:45 -05003067 goto out;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003068exit_fput:
Al Viro2675a4e2012-06-22 12:41:10 +04003069 fput(file);
3070 goto out;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003071
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003072stale_open:
3073 /* If no saved parent or already retried then can't retry */
3074 if (!save_parent.dentry || retried)
3075 goto out;
3076
3077 BUG_ON(save_parent.dentry != dir);
3078 path_put(&nd->path);
3079 nd->path = save_parent;
3080 nd->inode = dir->d_inode;
3081 save_parent.mnt = NULL;
3082 save_parent.dentry = NULL;
Al Viro64894cf2012-07-31 00:53:35 +04003083 if (got_write) {
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003084 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04003085 got_write = false;
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003086 }
3087 retried = true;
3088 goto retry_lookup;
Al Virofb1cc552009-12-24 01:58:28 -05003089}
3090
Al Viro60545d02013-06-07 01:20:27 -04003091static int do_tmpfile(int dfd, struct filename *pathname,
3092 struct nameidata *nd, int flags,
3093 const struct open_flags *op,
3094 struct file *file, int *opened)
3095{
3096 static const struct qstr name = QSTR_INIT("/", 1);
3097 struct dentry *dentry, *child;
3098 struct inode *dir;
3099 int error = path_lookupat(dfd, pathname->name,
3100 flags | LOOKUP_DIRECTORY, nd);
3101 if (unlikely(error))
3102 return error;
3103 error = mnt_want_write(nd->path.mnt);
3104 if (unlikely(error))
3105 goto out;
3106 /* we want directory to be writable */
3107 error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3108 if (error)
3109 goto out2;
3110 dentry = nd->path.dentry;
3111 dir = dentry->d_inode;
3112 if (!dir->i_op->tmpfile) {
3113 error = -EOPNOTSUPP;
3114 goto out2;
3115 }
3116 child = d_alloc(dentry, &name);
3117 if (unlikely(!child)) {
3118 error = -ENOMEM;
3119 goto out2;
3120 }
3121 nd->flags &= ~LOOKUP_DIRECTORY;
3122 nd->flags |= op->intent;
3123 dput(nd->path.dentry);
3124 nd->path.dentry = child;
3125 error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3126 if (error)
3127 goto out2;
3128 audit_inode(pathname, nd->path.dentry, 0);
3129 error = may_open(&nd->path, op->acc_mode, op->open_flag);
3130 if (error)
3131 goto out2;
3132 file->f_path.mnt = nd->path.mnt;
3133 error = finish_open(file, nd->path.dentry, NULL, opened);
3134 if (error)
3135 goto out2;
3136 error = open_check_o_direct(file);
Al Virof4e0c302013-06-11 08:34:36 +04003137 if (error) {
Al Viro60545d02013-06-07 01:20:27 -04003138 fput(file);
Al Virof4e0c302013-06-11 08:34:36 +04003139 } else if (!(op->open_flag & O_EXCL)) {
3140 struct inode *inode = file_inode(file);
3141 spin_lock(&inode->i_lock);
3142 inode->i_state |= I_LINKABLE;
3143 spin_unlock(&inode->i_lock);
3144 }
Al Viro60545d02013-06-07 01:20:27 -04003145out2:
3146 mnt_drop_write(nd->path.mnt);
3147out:
3148 path_put(&nd->path);
3149 return error;
3150}
3151
Jeff Layton669abf42012-10-10 16:43:10 -04003152static struct file *path_openat(int dfd, struct filename *pathname,
Al Viro73d049a2011-03-11 12:08:24 -05003153 struct nameidata *nd, const struct open_flags *op, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003154{
Al Virofe2d35f2011-03-05 22:58:25 -05003155 struct file *base = NULL;
Al Viro30d90492012-06-22 12:40:19 +04003156 struct file *file;
Al Viro9850c052010-01-13 15:01:15 -05003157 struct path path;
Al Viro47237682012-06-10 05:01:45 -04003158 int opened = 0;
Al Viro13aab422011-02-23 17:54:08 -05003159 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11003160
Al Viro30d90492012-06-22 12:40:19 +04003161 file = get_empty_filp();
Al Viro1afc99b2013-02-14 20:41:04 -05003162 if (IS_ERR(file))
3163 return file;
Nick Piggin31e6b012011-01-07 17:49:52 +11003164
Al Viro30d90492012-06-22 12:40:19 +04003165 file->f_flags = op->open_flag;
Nick Piggin31e6b012011-01-07 17:49:52 +11003166
Al Virobb458c62013-07-13 13:26:37 +04003167 if (unlikely(file->f_flags & __O_TMPFILE)) {
Al Viro60545d02013-06-07 01:20:27 -04003168 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3169 goto out;
3170 }
3171
Jeff Layton669abf42012-10-10 16:43:10 -04003172 error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base);
Nick Piggin31e6b012011-01-07 17:49:52 +11003173 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04003174 goto out;
Nick Piggin31e6b012011-01-07 17:49:52 +11003175
Al Virofe2d35f2011-03-05 22:58:25 -05003176 current->total_link_count = 0;
Jeff Layton669abf42012-10-10 16:43:10 -04003177 error = link_path_walk(pathname->name, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11003178 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04003179 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003180
Al Viro2675a4e2012-06-22 12:41:10 +04003181 error = do_last(nd, &path, file, op, &opened, pathname);
3182 while (unlikely(error > 0)) { /* trailing symlink */
Nick Piggin7b9337a2011-01-14 08:42:43 +00003183 struct path link = path;
Al Virodef4af32009-12-26 08:37:05 -05003184 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04003185 if (!(nd->flags & LOOKUP_FOLLOW)) {
Al Viro73d049a2011-03-11 12:08:24 -05003186 path_put_conditional(&path, nd);
3187 path_put(&nd->path);
Al Viro2675a4e2012-06-22 12:41:10 +04003188 error = -ELOOP;
Al Viro40b39132011-03-09 16:22:18 -05003189 break;
3190 }
Kees Cook800179c2012-07-25 17:29:07 -07003191 error = may_follow_link(&link, nd);
3192 if (unlikely(error))
3193 break;
Al Viro73d049a2011-03-11 12:08:24 -05003194 nd->flags |= LOOKUP_PARENT;
3195 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
Al Viro574197e2011-03-14 22:20:34 -04003196 error = follow_link(&link, nd, &cookie);
Al Viroc3e380b2011-02-23 13:39:45 -05003197 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04003198 break;
3199 error = do_last(nd, &path, file, op, &opened, pathname);
Al Viro574197e2011-03-14 22:20:34 -04003200 put_link(nd, &link, cookie);
Al Viro806b6812009-12-26 07:16:40 -05003201 }
Al Viro10fa8e62009-12-26 07:09:49 -05003202out:
Al Viro73d049a2011-03-11 12:08:24 -05003203 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
3204 path_put(&nd->root);
Al Virofe2d35f2011-03-05 22:58:25 -05003205 if (base)
3206 fput(base);
Al Viro2675a4e2012-06-22 12:41:10 +04003207 if (!(opened & FILE_OPENED)) {
3208 BUG_ON(!error);
Al Viro30d90492012-06-22 12:40:19 +04003209 put_filp(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003210 }
Al Viro2675a4e2012-06-22 12:41:10 +04003211 if (unlikely(error)) {
3212 if (error == -EOPENSTALE) {
3213 if (flags & LOOKUP_RCU)
3214 error = -ECHILD;
3215 else
3216 error = -ESTALE;
3217 }
3218 file = ERR_PTR(error);
3219 }
3220 return file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003221}
3222
Jeff Layton669abf42012-10-10 16:43:10 -04003223struct file *do_filp_open(int dfd, struct filename *pathname,
Al Virof9652e12013-06-11 08:23:01 +04003224 const struct open_flags *op)
Al Viro13aab422011-02-23 17:54:08 -05003225{
Al Viro73d049a2011-03-11 12:08:24 -05003226 struct nameidata nd;
Al Virof9652e12013-06-11 08:23:01 +04003227 int flags = op->lookup_flags;
Al Viro13aab422011-02-23 17:54:08 -05003228 struct file *filp;
3229
Al Viro73d049a2011-03-11 12:08:24 -05003230 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
Al Viro13aab422011-02-23 17:54:08 -05003231 if (unlikely(filp == ERR_PTR(-ECHILD)))
Al Viro73d049a2011-03-11 12:08:24 -05003232 filp = path_openat(dfd, pathname, &nd, op, flags);
Al Viro13aab422011-02-23 17:54:08 -05003233 if (unlikely(filp == ERR_PTR(-ESTALE)))
Al Viro73d049a2011-03-11 12:08:24 -05003234 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
Al Viro13aab422011-02-23 17:54:08 -05003235 return filp;
3236}
3237
Al Viro73d049a2011-03-11 12:08:24 -05003238struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
Al Virof9652e12013-06-11 08:23:01 +04003239 const char *name, const struct open_flags *op)
Al Viro73d049a2011-03-11 12:08:24 -05003240{
3241 struct nameidata nd;
3242 struct file *file;
Jeff Layton669abf42012-10-10 16:43:10 -04003243 struct filename filename = { .name = name };
Al Virof9652e12013-06-11 08:23:01 +04003244 int flags = op->lookup_flags | LOOKUP_ROOT;
Al Viro73d049a2011-03-11 12:08:24 -05003245
3246 nd.root.mnt = mnt;
3247 nd.root.dentry = dentry;
3248
David Howellsb18825a2013-09-12 19:22:53 +01003249 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
Al Viro73d049a2011-03-11 12:08:24 -05003250 return ERR_PTR(-ELOOP);
3251
Jeff Layton669abf42012-10-10 16:43:10 -04003252 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU);
Al Viro73d049a2011-03-11 12:08:24 -05003253 if (unlikely(file == ERR_PTR(-ECHILD)))
Jeff Layton669abf42012-10-10 16:43:10 -04003254 file = path_openat(-1, &filename, &nd, op, flags);
Al Viro73d049a2011-03-11 12:08:24 -05003255 if (unlikely(file == ERR_PTR(-ESTALE)))
Jeff Layton669abf42012-10-10 16:43:10 -04003256 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL);
Al Viro73d049a2011-03-11 12:08:24 -05003257 return file;
3258}
3259
Jeff Layton1ac12b42012-12-11 12:10:06 -05003260struct dentry *kern_path_create(int dfd, const char *pathname,
3261 struct path *path, unsigned int lookup_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003262{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003263 struct dentry *dentry = ERR_PTR(-EEXIST);
Al Viroed75e952011-06-27 16:53:43 -04003264 struct nameidata nd;
Jan Karac30dabf2012-06-12 16:20:30 +02003265 int err2;
Jeff Layton1ac12b42012-12-11 12:10:06 -05003266 int error;
3267 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3268
3269 /*
3270 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3271 * other flags passed in are ignored!
3272 */
3273 lookup_flags &= LOOKUP_REVAL;
3274
3275 error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd);
Al Viroed75e952011-06-27 16:53:43 -04003276 if (error)
3277 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003278
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003279 /*
3280 * Yucky last component or no last component at all?
3281 * (foo/., foo/.., /////)
3282 */
Al Viroed75e952011-06-27 16:53:43 -04003283 if (nd.last_type != LAST_NORM)
3284 goto out;
3285 nd.flags &= ~LOOKUP_PARENT;
3286 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003287
Jan Karac30dabf2012-06-12 16:20:30 +02003288 /* don't fail immediately if it's r/o, at least try to report other errors */
3289 err2 = mnt_want_write(nd.path.mnt);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003290 /*
3291 * Do the final lookup.
3292 */
Al Viroed75e952011-06-27 16:53:43 -04003293 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3294 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003295 if (IS_ERR(dentry))
Al Viroa8104a92012-07-20 02:25:00 +04003296 goto unlock;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003297
Al Viroa8104a92012-07-20 02:25:00 +04003298 error = -EEXIST;
David Howellsb18825a2013-09-12 19:22:53 +01003299 if (d_is_positive(dentry))
Al Viroa8104a92012-07-20 02:25:00 +04003300 goto fail;
David Howellsb18825a2013-09-12 19:22:53 +01003301
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003302 /*
3303 * Special case - lookup gave negative, but... we had foo/bar/
3304 * From the vfs_mknod() POV we just have a negative dentry -
3305 * all is fine. Let's be bastards - you had / on the end, you've
3306 * been asking for (non-existent) directory. -ENOENT for you.
3307 */
Al Viroed75e952011-06-27 16:53:43 -04003308 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
Al Viroa8104a92012-07-20 02:25:00 +04003309 error = -ENOENT;
Al Viroed75e952011-06-27 16:53:43 -04003310 goto fail;
Al Viroe9baf6e2008-05-15 04:49:12 -04003311 }
Jan Karac30dabf2012-06-12 16:20:30 +02003312 if (unlikely(err2)) {
3313 error = err2;
Al Viroa8104a92012-07-20 02:25:00 +04003314 goto fail;
Jan Karac30dabf2012-06-12 16:20:30 +02003315 }
Al Viroed75e952011-06-27 16:53:43 -04003316 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003317 return dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003318fail:
Al Viroa8104a92012-07-20 02:25:00 +04003319 dput(dentry);
3320 dentry = ERR_PTR(error);
3321unlock:
Al Viroed75e952011-06-27 16:53:43 -04003322 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Karac30dabf2012-06-12 16:20:30 +02003323 if (!err2)
3324 mnt_drop_write(nd.path.mnt);
Al Viroed75e952011-06-27 16:53:43 -04003325out:
3326 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003327 return dentry;
3328}
Al Virodae6ad82011-06-26 11:50:15 -04003329EXPORT_SYMBOL(kern_path_create);
3330
Al Viro921a1652012-07-20 01:15:31 +04003331void done_path_create(struct path *path, struct dentry *dentry)
3332{
3333 dput(dentry);
3334 mutex_unlock(&path->dentry->d_inode->i_mutex);
Al Viroa8104a92012-07-20 02:25:00 +04003335 mnt_drop_write(path->mnt);
Al Viro921a1652012-07-20 01:15:31 +04003336 path_put(path);
3337}
3338EXPORT_SYMBOL(done_path_create);
3339
Jeff Layton1ac12b42012-12-11 12:10:06 -05003340struct dentry *user_path_create(int dfd, const char __user *pathname,
3341 struct path *path, unsigned int lookup_flags)
Al Virodae6ad82011-06-26 11:50:15 -04003342{
Jeff Layton91a27b22012-10-10 15:25:28 -04003343 struct filename *tmp = getname(pathname);
Al Virodae6ad82011-06-26 11:50:15 -04003344 struct dentry *res;
3345 if (IS_ERR(tmp))
3346 return ERR_CAST(tmp);
Jeff Layton1ac12b42012-12-11 12:10:06 -05003347 res = kern_path_create(dfd, tmp->name, path, lookup_flags);
Al Virodae6ad82011-06-26 11:50:15 -04003348 putname(tmp);
3349 return res;
3350}
3351EXPORT_SYMBOL(user_path_create);
3352
Al Viro1a67aaf2011-07-26 01:52:52 -04003353int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003354{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003355 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003356
3357 if (error)
3358 return error;
3359
Eric W. Biederman975d6b32011-11-13 12:16:43 -08003360 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003361 return -EPERM;
3362
Al Viroacfa4382008-12-04 10:06:33 -05003363 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003364 return -EPERM;
3365
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07003366 error = devcgroup_inode_mknod(mode, dev);
3367 if (error)
3368 return error;
3369
Linus Torvalds1da177e2005-04-16 15:20:36 -07003370 error = security_inode_mknod(dir, dentry, mode, dev);
3371 if (error)
3372 return error;
3373
Linus Torvalds1da177e2005-04-16 15:20:36 -07003374 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003375 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003376 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003377 return error;
3378}
3379
Al Virof69aac02011-07-26 04:31:40 -04003380static int may_mknod(umode_t mode)
Dave Hansen463c3192008-02-15 14:37:57 -08003381{
3382 switch (mode & S_IFMT) {
3383 case S_IFREG:
3384 case S_IFCHR:
3385 case S_IFBLK:
3386 case S_IFIFO:
3387 case S_IFSOCK:
3388 case 0: /* zero mode translates to S_IFREG */
3389 return 0;
3390 case S_IFDIR:
3391 return -EPERM;
3392 default:
3393 return -EINVAL;
3394 }
3395}
3396
Al Viro8208a222011-07-25 17:32:17 -04003397SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003398 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003399{
Al Viro2ad94ae2008-07-21 09:32:51 -04003400 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003401 struct path path;
3402 int error;
Jeff Layton972567f2012-12-20 16:00:10 -05003403 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003404
Al Viro8e4bfca2012-07-20 01:17:26 +04003405 error = may_mknod(mode);
3406 if (error)
3407 return error;
Jeff Layton972567f2012-12-20 16:00:10 -05003408retry:
3409 dentry = user_path_create(dfd, filename, &path, lookup_flags);
Al Virodae6ad82011-06-26 11:50:15 -04003410 if (IS_ERR(dentry))
3411 return PTR_ERR(dentry);
Al Viro2ad94ae2008-07-21 09:32:51 -04003412
Al Virodae6ad82011-06-26 11:50:15 -04003413 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003414 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003415 error = security_path_mknod(&path, dentry, mode, dev);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003416 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04003417 goto out;
Dave Hansen463c3192008-02-15 14:37:57 -08003418 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003419 case 0: case S_IFREG:
Al Viro312b63f2012-06-10 18:09:36 -04003420 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003421 break;
3422 case S_IFCHR: case S_IFBLK:
Al Virodae6ad82011-06-26 11:50:15 -04003423 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003424 new_decode_dev(dev));
3425 break;
3426 case S_IFIFO: case S_IFSOCK:
Al Virodae6ad82011-06-26 11:50:15 -04003427 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003428 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003429 }
Al Viroa8104a92012-07-20 02:25:00 +04003430out:
Al Viro921a1652012-07-20 01:15:31 +04003431 done_path_create(&path, dentry);
Jeff Layton972567f2012-12-20 16:00:10 -05003432 if (retry_estale(error, lookup_flags)) {
3433 lookup_flags |= LOOKUP_REVAL;
3434 goto retry;
3435 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003436 return error;
3437}
3438
Al Viro8208a222011-07-25 17:32:17 -04003439SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003440{
3441 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3442}
3443
Al Viro18bb1db2011-07-26 01:41:39 -04003444int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003445{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003446 int error = may_create(dir, dentry);
Al Viro8de52772012-02-06 12:45:27 -05003447 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003448
3449 if (error)
3450 return error;
3451
Al Viroacfa4382008-12-04 10:06:33 -05003452 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003453 return -EPERM;
3454
3455 mode &= (S_IRWXUGO|S_ISVTX);
3456 error = security_inode_mkdir(dir, dentry, mode);
3457 if (error)
3458 return error;
3459
Al Viro8de52772012-02-06 12:45:27 -05003460 if (max_links && dir->i_nlink >= max_links)
3461 return -EMLINK;
3462
Linus Torvalds1da177e2005-04-16 15:20:36 -07003463 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003464 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003465 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003466 return error;
3467}
3468
Al Viroa218d0f2011-11-21 14:59:34 -05003469SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003470{
Dave Hansen6902d922006-09-30 23:29:01 -07003471 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003472 struct path path;
3473 int error;
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003474 unsigned int lookup_flags = LOOKUP_DIRECTORY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003475
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003476retry:
3477 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07003478 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003479 return PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003480
Al Virodae6ad82011-06-26 11:50:15 -04003481 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003482 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003483 error = security_path_mkdir(&path, dentry, mode);
Al Viroa8104a92012-07-20 02:25:00 +04003484 if (!error)
3485 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
Al Viro921a1652012-07-20 01:15:31 +04003486 done_path_create(&path, dentry);
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003487 if (retry_estale(error, lookup_flags)) {
3488 lookup_flags |= LOOKUP_REVAL;
3489 goto retry;
3490 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003491 return error;
3492}
3493
Al Viroa218d0f2011-11-21 14:59:34 -05003494SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003495{
3496 return sys_mkdirat(AT_FDCWD, pathname, mode);
3497}
3498
Linus Torvalds1da177e2005-04-16 15:20:36 -07003499/*
Sage Weila71905f2011-05-24 13:06:08 -07003500 * The dentry_unhash() helper will try to drop the dentry early: we
J. Bruce Fieldsc0d02592012-02-15 11:48:40 -05003501 * should have a usage count of 1 if we're the only user of this
Sage Weila71905f2011-05-24 13:06:08 -07003502 * dentry, and if that is true (possibly after pruning the dcache),
3503 * then we drop the dentry now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003504 *
3505 * A low-level filesystem can, if it choses, legally
3506 * do a
3507 *
3508 * if (!d_unhashed(dentry))
3509 * return -EBUSY;
3510 *
3511 * if it cannot handle the case of removing a directory
3512 * that is still in use by something else..
3513 */
3514void dentry_unhash(struct dentry *dentry)
3515{
Vasily Averindc168422006-12-06 20:37:07 -08003516 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003517 spin_lock(&dentry->d_lock);
Waiman Long98474232013-08-28 18:24:59 -07003518 if (dentry->d_lockref.count == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003519 __d_drop(dentry);
3520 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003521}
3522
3523int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3524{
3525 int error = may_delete(dir, dentry, 1);
3526
3527 if (error)
3528 return error;
3529
Al Viroacfa4382008-12-04 10:06:33 -05003530 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003531 return -EPERM;
3532
Al Viro1d2ef592011-09-14 18:55:41 +01003533 dget(dentry);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003534 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003535
Sage Weil912dbc12011-05-24 13:06:11 -07003536 error = -EBUSY;
3537 if (d_mountpoint(dentry))
3538 goto out;
3539
3540 error = security_inode_rmdir(dir, dentry);
3541 if (error)
3542 goto out;
3543
Sage Weil3cebde22011-05-29 21:20:59 -07003544 shrink_dcache_parent(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003545 error = dir->i_op->rmdir(dir, dentry);
3546 if (error)
3547 goto out;
3548
3549 dentry->d_inode->i_flags |= S_DEAD;
3550 dont_mount(dentry);
3551
3552out:
3553 mutex_unlock(&dentry->d_inode->i_mutex);
Al Viro1d2ef592011-09-14 18:55:41 +01003554 dput(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003555 if (!error)
3556 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003557 return error;
3558}
3559
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003560static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003561{
3562 int error = 0;
Jeff Layton91a27b22012-10-10 15:25:28 -04003563 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003564 struct dentry *dentry;
3565 struct nameidata nd;
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003566 unsigned int lookup_flags = 0;
3567retry:
3568 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04003569 if (IS_ERR(name))
3570 return PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003571
3572 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003573 case LAST_DOTDOT:
3574 error = -ENOTEMPTY;
3575 goto exit1;
3576 case LAST_DOT:
3577 error = -EINVAL;
3578 goto exit1;
3579 case LAST_ROOT:
3580 error = -EBUSY;
3581 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003582 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003583
3584 nd.flags &= ~LOOKUP_PARENT;
Jan Karac30dabf2012-06-12 16:20:30 +02003585 error = mnt_want_write(nd.path.mnt);
3586 if (error)
3587 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003588
Jan Blunck4ac91372008-02-14 19:34:32 -08003589 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08003590 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003591 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003592 if (IS_ERR(dentry))
3593 goto exit2;
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003594 if (!dentry->d_inode) {
3595 error = -ENOENT;
3596 goto exit3;
3597 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003598 error = security_path_rmdir(&nd.path, dentry);
3599 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003600 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08003601 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Dave Hansen06227532008-02-15 14:37:34 -08003602exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07003603 dput(dentry);
3604exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08003605 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Karac30dabf2012-06-12 16:20:30 +02003606 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003607exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003608 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003609 putname(name);
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003610 if (retry_estale(error, lookup_flags)) {
3611 lookup_flags |= LOOKUP_REVAL;
3612 goto retry;
3613 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003614 return error;
3615}
3616
Heiko Carstens3cdad422009-01-14 14:14:22 +01003617SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003618{
3619 return do_rmdir(AT_FDCWD, pathname);
3620}
3621
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003622/**
3623 * vfs_unlink - unlink a filesystem object
3624 * @dir: parent directory
3625 * @dentry: victim
3626 * @delegated_inode: returns victim inode, if the inode is delegated.
3627 *
3628 * The caller must hold dir->i_mutex.
3629 *
3630 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3631 * return a reference to the inode in delegated_inode. The caller
3632 * should then break the delegation on that inode and retry. Because
3633 * breaking a delegation may take a long time, the caller should drop
3634 * dir->i_mutex before doing so.
3635 *
3636 * Alternatively, a caller may pass NULL for delegated_inode. This may
3637 * be appropriate for callers that expect the underlying filesystem not
3638 * to be NFS exported.
3639 */
3640int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003641{
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003642 struct inode *target = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003643 int error = may_delete(dir, dentry, 0);
3644
3645 if (error)
3646 return error;
3647
Al Viroacfa4382008-12-04 10:06:33 -05003648 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003649 return -EPERM;
3650
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003651 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003652 if (d_mountpoint(dentry))
3653 error = -EBUSY;
3654 else {
3655 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05003656 if (!error) {
J. Bruce Fields5a146962012-08-28 07:50:40 -07003657 error = try_break_deleg(target, delegated_inode);
3658 if (error)
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003659 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003660 error = dir->i_op->unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05003661 if (!error)
Al Virod83c49f2010-04-30 17:17:09 -04003662 dont_mount(dentry);
Al Virobec10522010-03-03 14:12:08 -05003663 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003664 }
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003665out:
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003666 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003667
3668 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3669 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003670 fsnotify_link_count(target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003671 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003672 }
Robert Love0eeca282005-07-12 17:06:03 -04003673
Linus Torvalds1da177e2005-04-16 15:20:36 -07003674 return error;
3675}
3676
3677/*
3678 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003679 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07003680 * writeout happening, and we don't want to prevent access to the directory
3681 * while waiting on the I/O.
3682 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003683static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003684{
Al Viro2ad94ae2008-07-21 09:32:51 -04003685 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04003686 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003687 struct dentry *dentry;
3688 struct nameidata nd;
3689 struct inode *inode = NULL;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003690 struct inode *delegated_inode = NULL;
Jeff Layton5d18f812012-12-20 16:38:04 -05003691 unsigned int lookup_flags = 0;
3692retry:
3693 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04003694 if (IS_ERR(name))
3695 return PTR_ERR(name);
Al Viro2ad94ae2008-07-21 09:32:51 -04003696
Linus Torvalds1da177e2005-04-16 15:20:36 -07003697 error = -EISDIR;
3698 if (nd.last_type != LAST_NORM)
3699 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003700
3701 nd.flags &= ~LOOKUP_PARENT;
Jan Karac30dabf2012-06-12 16:20:30 +02003702 error = mnt_want_write(nd.path.mnt);
3703 if (error)
3704 goto exit1;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003705retry_deleg:
Jan Blunck4ac91372008-02-14 19:34:32 -08003706 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08003707 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003708 error = PTR_ERR(dentry);
3709 if (!IS_ERR(dentry)) {
3710 /* Why not before? Because we want correct error value */
Török Edwin50338b82011-06-16 00:06:14 +03003711 if (nd.last.name[nd.last.len])
3712 goto slashes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003713 inode = dentry->d_inode;
David Howellsb18825a2013-09-12 19:22:53 +01003714 if (d_is_negative(dentry))
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003715 goto slashes;
3716 ihold(inode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003717 error = security_path_unlink(&nd.path, dentry);
3718 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003719 goto exit2;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003720 error = vfs_unlink(nd.path.dentry->d_inode, dentry, &delegated_inode);
Jan Karac30dabf2012-06-12 16:20:30 +02003721exit2:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003722 dput(dentry);
3723 }
Jan Blunck4ac91372008-02-14 19:34:32 -08003724 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003725 if (inode)
3726 iput(inode); /* truncate the inode here */
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003727 inode = NULL;
3728 if (delegated_inode) {
J. Bruce Fields5a146962012-08-28 07:50:40 -07003729 error = break_deleg_wait(&delegated_inode);
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003730 if (!error)
3731 goto retry_deleg;
3732 }
Jan Karac30dabf2012-06-12 16:20:30 +02003733 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003734exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003735 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003736 putname(name);
Jeff Layton5d18f812012-12-20 16:38:04 -05003737 if (retry_estale(error, lookup_flags)) {
3738 lookup_flags |= LOOKUP_REVAL;
3739 inode = NULL;
3740 goto retry;
3741 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003742 return error;
3743
3744slashes:
David Howellsb18825a2013-09-12 19:22:53 +01003745 if (d_is_negative(dentry))
3746 error = -ENOENT;
3747 else if (d_is_directory(dentry) || d_is_autodir(dentry))
3748 error = -EISDIR;
3749 else
3750 error = -ENOTDIR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003751 goto exit2;
3752}
3753
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003754SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003755{
3756 if ((flag & ~AT_REMOVEDIR) != 0)
3757 return -EINVAL;
3758
3759 if (flag & AT_REMOVEDIR)
3760 return do_rmdir(dfd, pathname);
3761
3762 return do_unlinkat(dfd, pathname);
3763}
3764
Heiko Carstens3480b252009-01-14 14:14:16 +01003765SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003766{
3767 return do_unlinkat(AT_FDCWD, pathname);
3768}
3769
Miklos Szeredidb2e7472008-06-24 16:50:16 +02003770int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003771{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003772 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003773
3774 if (error)
3775 return error;
3776
Al Viroacfa4382008-12-04 10:06:33 -05003777 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003778 return -EPERM;
3779
3780 error = security_inode_symlink(dir, dentry, oldname);
3781 if (error)
3782 return error;
3783
Linus Torvalds1da177e2005-04-16 15:20:36 -07003784 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003785 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003786 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003787 return error;
3788}
3789
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003790SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3791 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003792{
Al Viro2ad94ae2008-07-21 09:32:51 -04003793 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04003794 struct filename *from;
Dave Hansen6902d922006-09-30 23:29:01 -07003795 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003796 struct path path;
Jeff Laytonf46d3562012-12-11 12:10:08 -05003797 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003798
3799 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04003800 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003801 return PTR_ERR(from);
Jeff Laytonf46d3562012-12-11 12:10:08 -05003802retry:
3803 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07003804 error = PTR_ERR(dentry);
3805 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003806 goto out_putname;
Dave Hansen6902d922006-09-30 23:29:01 -07003807
Jeff Layton91a27b22012-10-10 15:25:28 -04003808 error = security_path_symlink(&path, dentry, from->name);
Al Viroa8104a92012-07-20 02:25:00 +04003809 if (!error)
Jeff Layton91a27b22012-10-10 15:25:28 -04003810 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
Al Viro921a1652012-07-20 01:15:31 +04003811 done_path_create(&path, dentry);
Jeff Laytonf46d3562012-12-11 12:10:08 -05003812 if (retry_estale(error, lookup_flags)) {
3813 lookup_flags |= LOOKUP_REVAL;
3814 goto retry;
3815 }
Dave Hansen6902d922006-09-30 23:29:01 -07003816out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003817 putname(from);
3818 return error;
3819}
3820
Heiko Carstens3480b252009-01-14 14:14:16 +01003821SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003822{
3823 return sys_symlinkat(oldname, AT_FDCWD, newname);
3824}
3825
J. Bruce Fields146a8592011-09-20 17:14:31 -04003826/**
3827 * vfs_link - create a new link
3828 * @old_dentry: object to be linked
3829 * @dir: new parent
3830 * @new_dentry: where to create the new link
3831 * @delegated_inode: returns inode needing a delegation break
3832 *
3833 * The caller must hold dir->i_mutex
3834 *
3835 * If vfs_link discovers a delegation on the to-be-linked file in need
3836 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3837 * inode in delegated_inode. The caller should then break the delegation
3838 * and retry. Because breaking a delegation may take a long time, the
3839 * caller should drop the i_mutex before doing so.
3840 *
3841 * Alternatively, a caller may pass NULL for delegated_inode. This may
3842 * be appropriate for callers that expect the underlying filesystem not
3843 * to be NFS exported.
3844 */
3845int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003846{
3847 struct inode *inode = old_dentry->d_inode;
Al Viro8de52772012-02-06 12:45:27 -05003848 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003849 int error;
3850
3851 if (!inode)
3852 return -ENOENT;
3853
Miklos Szeredia95164d2008-07-30 15:08:48 +02003854 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003855 if (error)
3856 return error;
3857
3858 if (dir->i_sb != inode->i_sb)
3859 return -EXDEV;
3860
3861 /*
3862 * A link to an append-only or immutable file cannot be created.
3863 */
3864 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3865 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05003866 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003867 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003868 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003869 return -EPERM;
3870
3871 error = security_inode_link(old_dentry, dir, new_dentry);
3872 if (error)
3873 return error;
3874
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003875 mutex_lock(&inode->i_mutex);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05303876 /* Make sure we don't allow creating hardlink to an unlinked file */
Al Virof4e0c302013-06-11 08:34:36 +04003877 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05303878 error = -ENOENT;
Al Viro8de52772012-02-06 12:45:27 -05003879 else if (max_links && inode->i_nlink >= max_links)
3880 error = -EMLINK;
J. Bruce Fields146a8592011-09-20 17:14:31 -04003881 else {
3882 error = try_break_deleg(inode, delegated_inode);
3883 if (!error)
3884 error = dir->i_op->link(old_dentry, dir, new_dentry);
3885 }
Al Virof4e0c302013-06-11 08:34:36 +04003886
3887 if (!error && (inode->i_state & I_LINKABLE)) {
3888 spin_lock(&inode->i_lock);
3889 inode->i_state &= ~I_LINKABLE;
3890 spin_unlock(&inode->i_lock);
3891 }
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003892 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07003893 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003894 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003895 return error;
3896}
3897
3898/*
3899 * Hardlinks are often used in delicate situations. We avoid
3900 * security-related surprises by not following symlinks on the
3901 * newname. --KAB
3902 *
3903 * We don't follow them on the oldname either to be compatible
3904 * with linux 2.0, and to avoid hard-linking to directories
3905 * and other special files. --ADM
3906 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003907SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3908 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003909{
3910 struct dentry *new_dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003911 struct path old_path, new_path;
J. Bruce Fields146a8592011-09-20 17:14:31 -04003912 struct inode *delegated_inode = NULL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303913 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003914 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003915
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303916 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003917 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303918 /*
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07003919 * To use null names we require CAP_DAC_READ_SEARCH
3920 * This ensures that not everyone will be able to create
3921 * handlink using the passed filedescriptor.
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303922 */
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07003923 if (flags & AT_EMPTY_PATH) {
3924 if (!capable(CAP_DAC_READ_SEARCH))
3925 return -ENOENT;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303926 how = LOOKUP_EMPTY;
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07003927 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003928
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303929 if (flags & AT_SYMLINK_FOLLOW)
3930 how |= LOOKUP_FOLLOW;
Jeff Layton442e31c2012-12-20 16:15:38 -05003931retry:
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303932 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003933 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04003934 return error;
3935
Jeff Layton442e31c2012-12-20 16:15:38 -05003936 new_dentry = user_path_create(newdfd, newname, &new_path,
3937 (how & LOOKUP_REVAL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003938 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003939 if (IS_ERR(new_dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003940 goto out;
3941
3942 error = -EXDEV;
3943 if (old_path.mnt != new_path.mnt)
3944 goto out_dput;
Kees Cook800179c2012-07-25 17:29:07 -07003945 error = may_linkat(&old_path);
3946 if (unlikely(error))
3947 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04003948 error = security_path_link(old_path.dentry, &new_path, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003949 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04003950 goto out_dput;
J. Bruce Fields146a8592011-09-20 17:14:31 -04003951 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
Dave Hansen75c3f292008-02-15 14:37:45 -08003952out_dput:
Al Viro921a1652012-07-20 01:15:31 +04003953 done_path_create(&new_path, new_dentry);
J. Bruce Fields146a8592011-09-20 17:14:31 -04003954 if (delegated_inode) {
3955 error = break_deleg_wait(&delegated_inode);
Oleg Drokind22e6332014-01-31 15:41:58 -05003956 if (!error) {
3957 path_put(&old_path);
J. Bruce Fields146a8592011-09-20 17:14:31 -04003958 goto retry;
Oleg Drokind22e6332014-01-31 15:41:58 -05003959 }
J. Bruce Fields146a8592011-09-20 17:14:31 -04003960 }
Jeff Layton442e31c2012-12-20 16:15:38 -05003961 if (retry_estale(error, how)) {
Oleg Drokind22e6332014-01-31 15:41:58 -05003962 path_put(&old_path);
Jeff Layton442e31c2012-12-20 16:15:38 -05003963 how |= LOOKUP_REVAL;
3964 goto retry;
3965 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003966out:
Al Viro2d8f3032008-07-22 09:59:21 -04003967 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003968
3969 return error;
3970}
3971
Heiko Carstens3480b252009-01-14 14:14:16 +01003972SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003973{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003974 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003975}
3976
Linus Torvalds1da177e2005-04-16 15:20:36 -07003977/*
3978 * The worst of all namespace operations - renaming directory. "Perverted"
3979 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3980 * Problems:
3981 * a) we can get into loop creation. Check is done in is_subdir().
3982 * b) race potential - two innocent renames can create a loop together.
3983 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003984 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07003985 * story.
J. Bruce Fields6cedba82012-03-05 11:40:41 -05003986 * c) we have to lock _four_ objects - parents and victim (if it exists),
3987 * and source (if it is not a directory).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003988 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07003989 * whether the target exists). Solution: try to be smart with locking
3990 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003991 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07003992 * move will be locked. Thus we can rank directories by the tree
3993 * (ancestors first) and rank all non-directories after them.
3994 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003995 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003996 * HOWEVER, it relies on the assumption that any object with ->lookup()
3997 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3998 * we'd better make sure that there's no link(2) for them.
Sage Weile4eaac02011-05-24 13:06:07 -07003999 * d) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004000 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07004001 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05004002 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07004003 * locking].
4004 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07004005static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
4006 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004007{
4008 int error = 0;
Sage Weil9055cba2011-05-24 13:06:12 -07004009 struct inode *target = new_dentry->d_inode;
Al Viro8de52772012-02-06 12:45:27 -05004010 unsigned max_links = new_dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004011
4012 /*
4013 * If we are going to change the parent - check write permissions,
4014 * we'll need to flip '..'.
4015 */
4016 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04004017 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004018 if (error)
4019 return error;
4020 }
4021
4022 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
4023 if (error)
4024 return error;
4025
Al Viro1d2ef592011-09-14 18:55:41 +01004026 dget(new_dentry);
Al Virod83c49f2010-04-30 17:17:09 -04004027 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004028 mutex_lock(&target->i_mutex);
Sage Weil9055cba2011-05-24 13:06:12 -07004029
4030 error = -EBUSY;
4031 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
4032 goto out;
4033
Al Viro8de52772012-02-06 12:45:27 -05004034 error = -EMLINK;
4035 if (max_links && !target && new_dir != old_dir &&
4036 new_dir->i_nlink >= max_links)
4037 goto out;
4038
Sage Weil3cebde22011-05-29 21:20:59 -07004039 if (target)
4040 shrink_dcache_parent(new_dentry);
Sage Weil9055cba2011-05-24 13:06:12 -07004041 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
4042 if (error)
4043 goto out;
4044
Linus Torvalds1da177e2005-04-16 15:20:36 -07004045 if (target) {
Sage Weil9055cba2011-05-24 13:06:12 -07004046 target->i_flags |= S_DEAD;
4047 dont_mount(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004048 }
Sage Weil9055cba2011-05-24 13:06:12 -07004049out:
4050 if (target)
4051 mutex_unlock(&target->i_mutex);
Al Viro1d2ef592011-09-14 18:55:41 +01004052 dput(new_dentry);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07004053 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07004054 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
4055 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004056 return error;
4057}
4058
Adrian Bunk75c96f82005-05-05 16:16:09 -07004059static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004060 struct inode *new_dir, struct dentry *new_dentry,
4061 struct inode **delegated_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004062{
Sage Weil51892bb2011-05-24 13:06:13 -07004063 struct inode *target = new_dentry->d_inode;
J. Bruce Fields6cedba82012-03-05 11:40:41 -05004064 struct inode *source = old_dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004065 int error;
4066
4067 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
4068 if (error)
4069 return error;
4070
4071 dget(new_dentry);
J. Bruce Fields6cedba82012-03-05 11:40:41 -05004072 lock_two_nondirectories(source, target);
Sage Weil51892bb2011-05-24 13:06:13 -07004073
4074 error = -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004075 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
Sage Weil51892bb2011-05-24 13:06:13 -07004076 goto out;
4077
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004078 error = try_break_deleg(source, delegated_inode);
4079 if (error)
4080 goto out;
4081 if (target) {
4082 error = try_break_deleg(target, delegated_inode);
4083 if (error)
4084 goto out;
4085 }
Sage Weil51892bb2011-05-24 13:06:13 -07004086 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
4087 if (error)
4088 goto out;
4089
4090 if (target)
4091 dont_mount(new_dentry);
4092 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
4093 d_move(old_dentry, new_dentry);
4094out:
J. Bruce Fields6cedba82012-03-05 11:40:41 -05004095 unlock_two_nondirectories(source, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004096 dput(new_dentry);
4097 return error;
4098}
4099
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004100/**
4101 * vfs_rename - rename a filesystem object
4102 * @old_dir: parent of source
4103 * @old_dentry: source
4104 * @new_dir: parent of destination
4105 * @new_dentry: destination
4106 * @delegated_inode: returns an inode needing a delegation break
4107 *
4108 * The caller must hold multiple mutexes--see lock_rename()).
4109 *
4110 * If vfs_rename discovers a delegation in need of breaking at either
4111 * the source or destination, it will return -EWOULDBLOCK and return a
4112 * reference to the inode in delegated_inode. The caller should then
4113 * break the delegation and retry. Because breaking a delegation may
4114 * take a long time, the caller should drop all locks before doing
4115 * so.
4116 *
4117 * Alternatively, a caller may pass NULL for delegated_inode. This may
4118 * be appropriate for callers that expect the underlying filesystem not
4119 * to be NFS exported.
4120 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004121int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004122 struct inode *new_dir, struct dentry *new_dentry,
4123 struct inode **delegated_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004124{
4125 int error;
David Howellsb18825a2013-09-12 19:22:53 +01004126 int is_dir = d_is_directory(old_dentry) || d_is_autodir(old_dentry);
Eric Paris59b0df22010-02-08 12:53:52 -05004127 const unsigned char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004128
4129 if (old_dentry->d_inode == new_dentry->d_inode)
4130 return 0;
4131
4132 error = may_delete(old_dir, old_dentry, is_dir);
4133 if (error)
4134 return error;
4135
4136 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02004137 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004138 else
4139 error = may_delete(new_dir, new_dentry, is_dir);
4140 if (error)
4141 return error;
4142
Al Viroacfa4382008-12-04 10:06:33 -05004143 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004144 return -EPERM;
4145
Robert Love0eeca282005-07-12 17:06:03 -04004146 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4147
Linus Torvalds1da177e2005-04-16 15:20:36 -07004148 if (is_dir)
4149 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
4150 else
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004151 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry,delegated_inode);
Al Viro123df292009-12-25 04:57:57 -05004152 if (!error)
4153 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04004154 new_dentry->d_inode, old_dentry);
Robert Love0eeca282005-07-12 17:06:03 -04004155 fsnotify_oldname_free(old_name);
4156
Linus Torvalds1da177e2005-04-16 15:20:36 -07004157 return error;
4158}
4159
Heiko Carstens2e4d0922009-01-14 14:14:31 +01004160SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4161 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004162{
Al Viro2ad94ae2008-07-21 09:32:51 -04004163 struct dentry *old_dir, *new_dir;
4164 struct dentry *old_dentry, *new_dentry;
4165 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004166 struct nameidata oldnd, newnd;
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004167 struct inode *delegated_inode = NULL;
Jeff Layton91a27b22012-10-10 15:25:28 -04004168 struct filename *from;
4169 struct filename *to;
Jeff Laytonc6a94282012-12-11 12:10:10 -05004170 unsigned int lookup_flags = 0;
4171 bool should_retry = false;
Al Viro2ad94ae2008-07-21 09:32:51 -04004172 int error;
Jeff Laytonc6a94282012-12-11 12:10:10 -05004173retry:
4174 from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04004175 if (IS_ERR(from)) {
4176 error = PTR_ERR(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004177 goto exit;
Jeff Layton91a27b22012-10-10 15:25:28 -04004178 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004179
Jeff Laytonc6a94282012-12-11 12:10:10 -05004180 to = user_path_parent(newdfd, newname, &newnd, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04004181 if (IS_ERR(to)) {
4182 error = PTR_ERR(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004183 goto exit1;
Jeff Layton91a27b22012-10-10 15:25:28 -04004184 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004185
4186 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08004187 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004188 goto exit2;
4189
Jan Blunck4ac91372008-02-14 19:34:32 -08004190 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004191 error = -EBUSY;
4192 if (oldnd.last_type != LAST_NORM)
4193 goto exit2;
4194
Jan Blunck4ac91372008-02-14 19:34:32 -08004195 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004196 if (newnd.last_type != LAST_NORM)
4197 goto exit2;
4198
Jan Karac30dabf2012-06-12 16:20:30 +02004199 error = mnt_want_write(oldnd.path.mnt);
4200 if (error)
4201 goto exit2;
4202
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09004203 oldnd.flags &= ~LOOKUP_PARENT;
4204 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09004205 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09004206
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004207retry_deleg:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004208 trap = lock_rename(new_dir, old_dir);
4209
Christoph Hellwig49705b72005-11-08 21:35:06 -08004210 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004211 error = PTR_ERR(old_dentry);
4212 if (IS_ERR(old_dentry))
4213 goto exit3;
4214 /* source must exist */
4215 error = -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01004216 if (d_is_negative(old_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004217 goto exit4;
4218 /* unless the source is a directory trailing slashes give -ENOTDIR */
David Howellsb18825a2013-09-12 19:22:53 +01004219 if (!d_is_directory(old_dentry) && !d_is_autodir(old_dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004220 error = -ENOTDIR;
4221 if (oldnd.last.name[oldnd.last.len])
4222 goto exit4;
4223 if (newnd.last.name[newnd.last.len])
4224 goto exit4;
4225 }
4226 /* source should not be ancestor of target */
4227 error = -EINVAL;
4228 if (old_dentry == trap)
4229 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08004230 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004231 error = PTR_ERR(new_dentry);
4232 if (IS_ERR(new_dentry))
4233 goto exit4;
4234 /* target should not be an ancestor of source */
4235 error = -ENOTEMPTY;
4236 if (new_dentry == trap)
4237 goto exit5;
4238
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004239 error = security_path_rename(&oldnd.path, old_dentry,
4240 &newnd.path, new_dentry);
4241 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02004242 goto exit5;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004243 error = vfs_rename(old_dir->d_inode, old_dentry,
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004244 new_dir->d_inode, new_dentry,
4245 &delegated_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004246exit5:
4247 dput(new_dentry);
4248exit4:
4249 dput(old_dentry);
4250exit3:
4251 unlock_rename(new_dir, old_dir);
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004252 if (delegated_inode) {
4253 error = break_deleg_wait(&delegated_inode);
4254 if (!error)
4255 goto retry_deleg;
4256 }
Jan Karac30dabf2012-06-12 16:20:30 +02004257 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004258exit2:
Jeff Laytonc6a94282012-12-11 12:10:10 -05004259 if (retry_estale(error, lookup_flags))
4260 should_retry = true;
Jan Blunck1d957f92008-02-14 19:34:35 -08004261 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04004262 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004263exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08004264 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004265 putname(from);
Jeff Laytonc6a94282012-12-11 12:10:10 -05004266 if (should_retry) {
4267 should_retry = false;
4268 lookup_flags |= LOOKUP_REVAL;
4269 goto retry;
4270 }
Al Viro2ad94ae2008-07-21 09:32:51 -04004271exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004272 return error;
4273}
4274
Heiko Carstensa26eab22009-01-14 14:14:17 +01004275SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004276{
4277 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
4278}
4279
Linus Torvalds1da177e2005-04-16 15:20:36 -07004280int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
4281{
4282 int len;
4283
4284 len = PTR_ERR(link);
4285 if (IS_ERR(link))
4286 goto out;
4287
4288 len = strlen(link);
4289 if (len > (unsigned) buflen)
4290 len = buflen;
4291 if (copy_to_user(buffer, link, len))
4292 len = -EFAULT;
4293out:
4294 return len;
4295}
4296
4297/*
4298 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4299 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4300 * using) it for any given inode is up to filesystem.
4301 */
4302int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4303{
4304 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004305 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07004306 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004307
Linus Torvalds1da177e2005-04-16 15:20:36 -07004308 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004309 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07004310 if (IS_ERR(cookie))
4311 return PTR_ERR(cookie);
4312
4313 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
4314 if (dentry->d_inode->i_op->put_link)
4315 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
4316 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004317}
4318
Linus Torvalds1da177e2005-04-16 15:20:36 -07004319/* get the link contents into pagecache */
4320static char *page_getlink(struct dentry * dentry, struct page **ppage)
4321{
Duane Griffinebd09ab2008-12-19 20:47:12 +00004322 char *kaddr;
4323 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004324 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07004325 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004326 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07004327 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004328 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00004329 kaddr = kmap(page);
4330 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4331 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004332}
4333
4334int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4335{
4336 struct page *page = NULL;
4337 char *s = page_getlink(dentry, &page);
4338 int res = vfs_readlink(dentry,buffer,buflen,s);
4339 if (page) {
4340 kunmap(page);
4341 page_cache_release(page);
4342 }
4343 return res;
4344}
4345
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004346void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004347{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004348 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004349 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004350 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004351}
4352
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004353void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004354{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004355 struct page *page = cookie;
4356
4357 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004358 kunmap(page);
4359 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004360 }
4361}
4362
Nick Piggin54566b22009-01-04 12:00:53 -08004363/*
4364 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4365 */
4366int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004367{
4368 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004369 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07004370 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08004371 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004372 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08004373 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4374 if (nofs)
4375 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004376
NeilBrown7e53cac2006-03-25 03:07:57 -08004377retry:
Nick Pigginafddba42007-10-16 01:25:01 -07004378 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08004379 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004380 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07004381 goto fail;
4382
Cong Wange8e3c3d2011-11-25 23:14:27 +08004383 kaddr = kmap_atomic(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004384 memcpy(kaddr, symname, len-1);
Cong Wange8e3c3d2011-11-25 23:14:27 +08004385 kunmap_atomic(kaddr);
Nick Pigginafddba42007-10-16 01:25:01 -07004386
4387 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4388 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004389 if (err < 0)
4390 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07004391 if (err < len-1)
4392 goto retry;
4393
Linus Torvalds1da177e2005-04-16 15:20:36 -07004394 mark_inode_dirty(inode);
4395 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004396fail:
4397 return err;
4398}
4399
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004400int page_symlink(struct inode *inode, const char *symname, int len)
4401{
4402 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08004403 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004404}
4405
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08004406const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004407 .readlink = generic_readlink,
4408 .follow_link = page_follow_link_light,
4409 .put_link = page_put_link,
4410};
4411
Al Viro2d8f3032008-07-22 09:59:21 -04004412EXPORT_SYMBOL(user_path_at);
David Howellscc53ce52011-01-14 18:45:26 +00004413EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004414EXPORT_SYMBOL(follow_down);
4415EXPORT_SYMBOL(follow_up);
Al Virof6d2ac52012-08-26 12:55:54 -04004416EXPORT_SYMBOL(get_write_access); /* nfsd */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004417EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004418EXPORT_SYMBOL(lookup_one_len);
4419EXPORT_SYMBOL(page_follow_link_light);
4420EXPORT_SYMBOL(page_put_link);
4421EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004422EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004423EXPORT_SYMBOL(page_symlink);
4424EXPORT_SYMBOL(page_symlink_inode_operations);
Al Virod1811462008-08-02 00:49:18 -04004425EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07004426EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04004427EXPORT_SYMBOL(inode_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004428EXPORT_SYMBOL(unlock_rename);
4429EXPORT_SYMBOL(vfs_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004430EXPORT_SYMBOL(vfs_link);
4431EXPORT_SYMBOL(vfs_mkdir);
4432EXPORT_SYMBOL(vfs_mknod);
4433EXPORT_SYMBOL(generic_permission);
4434EXPORT_SYMBOL(vfs_readlink);
4435EXPORT_SYMBOL(vfs_rename);
4436EXPORT_SYMBOL(vfs_rmdir);
4437EXPORT_SYMBOL(vfs_symlink);
4438EXPORT_SYMBOL(vfs_unlink);
4439EXPORT_SYMBOL(dentry_unhash);
4440EXPORT_SYMBOL(generic_readlink);