blob: c6157c894fce234c333d5a2d787f81ee3e5e7ba9 [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}
Al Viro4d359502014-03-14 12:20:17 -0400361EXPORT_SYMBOL(generic_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700363/*
364 * We _really_ want to just do "generic_permission()" without
365 * even looking at the inode->i_op values. So we keep a cache
366 * flag in inode->i_opflags, that says "this has not special
367 * permission function, use the fast case".
368 */
369static inline int do_inode_permission(struct inode *inode, int mask)
370{
371 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
372 if (likely(inode->i_op->permission))
373 return inode->i_op->permission(inode, mask);
374
375 /* This gets set once for the inode lifetime */
376 spin_lock(&inode->i_lock);
377 inode->i_opflags |= IOP_FASTPERM;
378 spin_unlock(&inode->i_lock);
379 }
380 return generic_permission(inode, mask);
381}
382
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200383/**
David Howells0bdaea92012-06-25 12:55:46 +0100384 * __inode_permission - Check for access rights to a given inode
385 * @inode: Inode to check permission on
386 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200387 *
David Howells0bdaea92012-06-25 12:55:46 +0100388 * Check for read/write/execute permissions on an inode.
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530389 *
390 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
David Howells0bdaea92012-06-25 12:55:46 +0100391 *
392 * This does not check for a read-only file system. You probably want
393 * inode_permission().
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200394 */
David Howells0bdaea92012-06-25 12:55:46 +0100395int __inode_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396{
Al Viroe6305c42008-07-15 21:03:57 -0400397 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700399 if (unlikely(mask & MAY_WRITE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 /*
401 * Nobody gets write access to an immutable file.
402 */
403 if (IS_IMMUTABLE(inode))
404 return -EACCES;
405 }
406
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700407 retval = do_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 if (retval)
409 return retval;
410
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -0700411 retval = devcgroup_inode_permission(inode, mask);
412 if (retval)
413 return retval;
414
Eric Parisd09ca732010-07-23 11:43:57 -0400415 return security_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416}
417
Al Virof4d6ff82011-06-19 13:14:21 -0400418/**
David Howells0bdaea92012-06-25 12:55:46 +0100419 * sb_permission - Check superblock-level permissions
420 * @sb: Superblock of inode to check permission on
Randy Dunlap55852632012-08-18 17:39:25 -0700421 * @inode: Inode to check permission on
David Howells0bdaea92012-06-25 12:55:46 +0100422 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
423 *
424 * Separate out file-system wide checks from inode-specific permission checks.
425 */
426static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
427{
428 if (unlikely(mask & MAY_WRITE)) {
429 umode_t mode = inode->i_mode;
430
431 /* Nobody gets write access to a read-only fs. */
432 if ((sb->s_flags & MS_RDONLY) &&
433 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
434 return -EROFS;
435 }
436 return 0;
437}
438
439/**
440 * inode_permission - Check for access rights to a given inode
441 * @inode: Inode to check permission on
442 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
443 *
444 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
445 * this, letting us set arbitrary permissions for filesystem access without
446 * changing the "normal" UIDs which are used for other things.
447 *
448 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
449 */
450int inode_permission(struct inode *inode, int mask)
451{
452 int retval;
453
454 retval = sb_permission(inode->i_sb, inode, mask);
455 if (retval)
456 return retval;
457 return __inode_permission(inode, mask);
458}
Al Viro4d359502014-03-14 12:20:17 -0400459EXPORT_SYMBOL(inode_permission);
David Howells0bdaea92012-06-25 12:55:46 +0100460
461/**
Jan Blunck5dd784d02008-02-14 19:34:38 -0800462 * path_get - get a reference to a path
463 * @path: path to get the reference to
464 *
465 * Given a path increment the reference count to the dentry and the vfsmount.
466 */
Al Virodcf787f2013-03-01 23:51:07 -0500467void path_get(const struct path *path)
Jan Blunck5dd784d02008-02-14 19:34:38 -0800468{
469 mntget(path->mnt);
470 dget(path->dentry);
471}
472EXPORT_SYMBOL(path_get);
473
474/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800475 * path_put - put a reference to a path
476 * @path: path to put the reference to
477 *
478 * Given a path decrement the reference count to the dentry and the vfsmount.
479 */
Al Virodcf787f2013-03-01 23:51:07 -0500480void path_put(const struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481{
Jan Blunck1d957f92008-02-14 19:34:35 -0800482 dput(path->dentry);
483 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484}
Jan Blunck1d957f92008-02-14 19:34:35 -0800485EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486
Al Viro19660af2011-03-25 10:32:48 -0400487/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100488 * Path walking has 2 modes, rcu-walk and ref-walk (see
Al Viro19660af2011-03-25 10:32:48 -0400489 * Documentation/filesystems/path-lookup.txt). In situations when we can't
490 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
491 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
492 * mode. Refcounts are grabbed at the last known good point before rcu-walk
493 * got stuck, so ref-walk may continue from there. If this is not successful
494 * (eg. a seqcount has changed), then failure is returned and it's up to caller
495 * to restart the path walk from the beginning in ref-walk mode.
Nick Piggin31e6b012011-01-07 17:49:52 +1100496 */
Nick Piggin31e6b012011-01-07 17:49:52 +1100497
498/**
Al Viro19660af2011-03-25 10:32:48 -0400499 * unlazy_walk - try to switch to ref-walk mode.
500 * @nd: nameidata pathwalk data
501 * @dentry: child of nd->path.dentry or NULL
Randy Dunlap39191622011-01-08 19:36:21 -0800502 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100503 *
Al Viro19660af2011-03-25 10:32:48 -0400504 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
505 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
506 * @nd or NULL. Must be called from rcu-walk context.
Nick Piggin31e6b012011-01-07 17:49:52 +1100507 */
Al Viro19660af2011-03-25 10:32:48 -0400508static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
Nick Piggin31e6b012011-01-07 17:49:52 +1100509{
510 struct fs_struct *fs = current->fs;
511 struct dentry *parent = nd->path.dentry;
512
513 BUG_ON(!(nd->flags & LOOKUP_RCU));
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700514
515 /*
Al Viro48a066e2013-09-29 22:06:07 -0400516 * After legitimizing the bastards, terminate_walk()
517 * will do the right thing for non-RCU mode, and all our
518 * subsequent exit cases should rcu_read_unlock()
519 * before returning. Do vfsmount first; if dentry
520 * can't be legitimized, just set nd->path.dentry to NULL
521 * and rely on dput(NULL) being a no-op.
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700522 */
Al Viro48a066e2013-09-29 22:06:07 -0400523 if (!legitimize_mnt(nd->path.mnt, nd->m_seq))
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700524 return -ECHILD;
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700525 nd->flags &= ~LOOKUP_RCU;
Linus Torvalds15570082013-09-02 11:38:06 -0700526
Al Viro48a066e2013-09-29 22:06:07 -0400527 if (!lockref_get_not_dead(&parent->d_lockref)) {
528 nd->path.dentry = NULL;
Al Virod870b4a2013-11-29 01:48:32 -0500529 goto out;
Al Viro48a066e2013-09-29 22:06:07 -0400530 }
531
Linus Torvalds15570082013-09-02 11:38:06 -0700532 /*
533 * For a negative lookup, the lookup sequence point is the parents
534 * sequence point, and it only needs to revalidate the parent dentry.
535 *
536 * For a positive lookup, we need to move both the parent and the
537 * dentry from the RCU domain to be properly refcounted. And the
538 * sequence number in the dentry validates *both* dentry counters,
539 * since we checked the sequence number of the parent after we got
540 * the child sequence number. So we know the parent must still
541 * be valid if the child sequence number is still valid.
542 */
Al Viro19660af2011-03-25 10:32:48 -0400543 if (!dentry) {
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700544 if (read_seqcount_retry(&parent->d_seq, nd->seq))
545 goto out;
Al Viro19660af2011-03-25 10:32:48 -0400546 BUG_ON(nd->inode != parent->d_inode);
547 } else {
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700548 if (!lockref_get_not_dead(&dentry->d_lockref))
549 goto out;
550 if (read_seqcount_retry(&dentry->d_seq, nd->seq))
551 goto drop_dentry;
Al Viro19660af2011-03-25 10:32:48 -0400552 }
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700553
554 /*
555 * Sequence counts matched. Now make sure that the root is
556 * still valid and get it if required.
557 */
558 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
559 spin_lock(&fs->lock);
560 if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry)
561 goto unlock_and_drop_dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +1100562 path_get(&nd->root);
563 spin_unlock(&fs->lock);
564 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100565
Al Viro8b61e742013-11-08 12:45:01 -0500566 rcu_read_unlock();
Nick Piggin31e6b012011-01-07 17:49:52 +1100567 return 0;
Al Viro19660af2011-03-25 10:32:48 -0400568
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700569unlock_and_drop_dentry:
570 spin_unlock(&fs->lock);
571drop_dentry:
Al Viro8b61e742013-11-08 12:45:01 -0500572 rcu_read_unlock();
Linus Torvalds15570082013-09-02 11:38:06 -0700573 dput(dentry);
Linus Torvaldsd0d27272013-09-10 12:17:49 -0700574 goto drop_root_mnt;
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700575out:
Al Viro8b61e742013-11-08 12:45:01 -0500576 rcu_read_unlock();
Linus Torvaldsd0d27272013-09-10 12:17:49 -0700577drop_root_mnt:
578 if (!(nd->flags & LOOKUP_ROOT))
579 nd->root.mnt = NULL;
Nick Piggin31e6b012011-01-07 17:49:52 +1100580 return -ECHILD;
581}
582
Al Viro4ce16ef32012-06-10 16:10:59 -0400583static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
Nick Piggin34286d62011-01-07 17:49:57 +1100584{
Al Viro4ce16ef32012-06-10 16:10:59 -0400585 return dentry->d_op->d_revalidate(dentry, flags);
Nick Piggin34286d62011-01-07 17:49:57 +1100586}
587
Al Viro9f1fafe2011-03-25 11:00:12 -0400588/**
589 * complete_walk - successful completion of path walk
590 * @nd: pointer nameidata
Jeff Layton39159de2009-12-07 12:01:50 -0500591 *
Al Viro9f1fafe2011-03-25 11:00:12 -0400592 * If we had been in RCU mode, drop out of it and legitimize nd->path.
593 * Revalidate the final result, unless we'd already done that during
594 * the path walk or the filesystem doesn't ask for it. Return 0 on
595 * success, -error on failure. In case of failure caller does not
596 * need to drop nd->path.
Jeff Layton39159de2009-12-07 12:01:50 -0500597 */
Al Viro9f1fafe2011-03-25 11:00:12 -0400598static int complete_walk(struct nameidata *nd)
Jeff Layton39159de2009-12-07 12:01:50 -0500599{
Al Viro16c2cd72011-02-22 15:50:10 -0500600 struct dentry *dentry = nd->path.dentry;
Jeff Layton39159de2009-12-07 12:01:50 -0500601 int status;
Jeff Layton39159de2009-12-07 12:01:50 -0500602
Al Viro9f1fafe2011-03-25 11:00:12 -0400603 if (nd->flags & LOOKUP_RCU) {
604 nd->flags &= ~LOOKUP_RCU;
605 if (!(nd->flags & LOOKUP_ROOT))
606 nd->root.mnt = NULL;
Linus Torvalds15570082013-09-02 11:38:06 -0700607
Al Viro48a066e2013-09-29 22:06:07 -0400608 if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) {
Al Viro8b61e742013-11-08 12:45:01 -0500609 rcu_read_unlock();
Al Viro48a066e2013-09-29 22:06:07 -0400610 return -ECHILD;
611 }
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700612 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) {
Al Viro8b61e742013-11-08 12:45:01 -0500613 rcu_read_unlock();
Al Viro48a066e2013-09-29 22:06:07 -0400614 mntput(nd->path.mnt);
Al Viro9f1fafe2011-03-25 11:00:12 -0400615 return -ECHILD;
616 }
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700617 if (read_seqcount_retry(&dentry->d_seq, nd->seq)) {
Al Viro8b61e742013-11-08 12:45:01 -0500618 rcu_read_unlock();
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700619 dput(dentry);
Al Viro48a066e2013-09-29 22:06:07 -0400620 mntput(nd->path.mnt);
Linus Torvaldse5c832d2013-09-08 18:13:49 -0700621 return -ECHILD;
622 }
Al Viro8b61e742013-11-08 12:45:01 -0500623 rcu_read_unlock();
Al Viro9f1fafe2011-03-25 11:00:12 -0400624 }
625
Al Viro16c2cd72011-02-22 15:50:10 -0500626 if (likely(!(nd->flags & LOOKUP_JUMPED)))
Jeff Layton39159de2009-12-07 12:01:50 -0500627 return 0;
628
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500629 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
Al Viro16c2cd72011-02-22 15:50:10 -0500630 return 0;
631
Jeff Laytonecf3d1f2013-02-20 11:19:05 -0500632 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
Jeff Layton39159de2009-12-07 12:01:50 -0500633 if (status > 0)
634 return 0;
635
Al Viro16c2cd72011-02-22 15:50:10 -0500636 if (!status)
Jeff Layton39159de2009-12-07 12:01:50 -0500637 status = -ESTALE;
Al Viro16c2cd72011-02-22 15:50:10 -0500638
Al Viro9f1fafe2011-03-25 11:00:12 -0400639 path_put(&nd->path);
Jeff Layton39159de2009-12-07 12:01:50 -0500640 return status;
641}
642
Al Viro2a737872009-04-07 11:49:53 -0400643static __always_inline void set_root(struct nameidata *nd)
644{
Miklos Szeredif7ad3c62010-08-10 11:41:36 +0200645 if (!nd->root.mnt)
646 get_fs_root(current->fs, &nd->root);
Al Viro2a737872009-04-07 11:49:53 -0400647}
648
Al Viro6de88d72009-08-09 01:41:57 +0400649static int link_path_walk(const char *, struct nameidata *);
650
Nick Piggin31e6b012011-01-07 17:49:52 +1100651static __always_inline void set_root_rcu(struct nameidata *nd)
652{
653 if (!nd->root.mnt) {
654 struct fs_struct *fs = current->fs;
Nick Pigginc28cc362011-01-07 17:49:53 +1100655 unsigned seq;
656
657 do {
658 seq = read_seqcount_begin(&fs->seq);
659 nd->root = fs->root;
Tim Chenc1530012011-04-15 11:39:29 -0700660 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
Nick Pigginc28cc362011-01-07 17:49:53 +1100661 } while (read_seqcount_retry(&fs->seq, seq));
Nick Piggin31e6b012011-01-07 17:49:52 +1100662 }
663}
664
Jan Blunck1d957f92008-02-14 19:34:35 -0800665static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700666{
667 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800668 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700669 mntput(path->mnt);
670}
671
Nick Piggin7b9337a2011-01-14 08:42:43 +0000672static inline void path_to_nameidata(const struct path *path,
673 struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700674{
Nick Piggin31e6b012011-01-07 17:49:52 +1100675 if (!(nd->flags & LOOKUP_RCU)) {
676 dput(nd->path.dentry);
677 if (nd->path.mnt != path->mnt)
678 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800679 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100680 nd->path.mnt = path->mnt;
Jan Blunck4ac91372008-02-14 19:34:32 -0800681 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700682}
683
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400684/*
685 * Helper to directly jump to a known parsed path from ->follow_link,
686 * caller must have taken a reference to path beforehand.
687 */
688void nd_jump_link(struct nameidata *nd, struct path *path)
689{
690 path_put(&nd->path);
691
692 nd->path = *path;
693 nd->inode = nd->path.dentry->d_inode;
694 nd->flags |= LOOKUP_JUMPED;
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400695}
696
Al Viro574197e2011-03-14 22:20:34 -0400697static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
698{
699 struct inode *inode = link->dentry->d_inode;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400700 if (inode->i_op->put_link)
Al Viro574197e2011-03-14 22:20:34 -0400701 inode->i_op->put_link(link->dentry, nd, cookie);
702 path_put(link);
703}
704
Linus Torvalds561ec642012-10-26 10:05:07 -0700705int sysctl_protected_symlinks __read_mostly = 0;
706int sysctl_protected_hardlinks __read_mostly = 0;
Kees Cook800179c2012-07-25 17:29:07 -0700707
708/**
709 * may_follow_link - Check symlink following for unsafe situations
710 * @link: The path of the symlink
Randy Dunlap55852632012-08-18 17:39:25 -0700711 * @nd: nameidata pathwalk data
Kees Cook800179c2012-07-25 17:29:07 -0700712 *
713 * In the case of the sysctl_protected_symlinks sysctl being enabled,
714 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
715 * in a sticky world-writable directory. This is to protect privileged
716 * processes from failing races against path names that may change out
717 * from under them by way of other users creating malicious symlinks.
718 * It will permit symlinks to be followed only when outside a sticky
719 * world-writable directory, or when the uid of the symlink and follower
720 * match, or when the directory owner matches the symlink's owner.
721 *
722 * Returns 0 if following the symlink is allowed, -ve on error.
723 */
724static inline int may_follow_link(struct path *link, struct nameidata *nd)
725{
726 const struct inode *inode;
727 const struct inode *parent;
728
729 if (!sysctl_protected_symlinks)
730 return 0;
731
732 /* Allowed if owner and follower match. */
733 inode = link->dentry->d_inode;
Eric W. Biederman81abe272012-08-03 09:38:08 -0700734 if (uid_eq(current_cred()->fsuid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700735 return 0;
736
737 /* Allowed if parent directory not sticky and world-writable. */
738 parent = nd->path.dentry->d_inode;
739 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
740 return 0;
741
742 /* Allowed if parent directory and link owner match. */
Eric W. Biederman81abe272012-08-03 09:38:08 -0700743 if (uid_eq(parent->i_uid, inode->i_uid))
Kees Cook800179c2012-07-25 17:29:07 -0700744 return 0;
745
Sasha Levinffd8d102012-10-04 19:56:40 -0400746 audit_log_link_denied("follow_link", link);
Kees Cook800179c2012-07-25 17:29:07 -0700747 path_put_conditional(link, nd);
748 path_put(&nd->path);
749 return -EACCES;
750}
751
752/**
753 * safe_hardlink_source - Check for safe hardlink conditions
754 * @inode: the source inode to hardlink from
755 *
756 * Return false if at least one of the following conditions:
757 * - inode is not a regular file
758 * - inode is setuid
759 * - inode is setgid and group-exec
760 * - access failure for read and write
761 *
762 * Otherwise returns true.
763 */
764static bool safe_hardlink_source(struct inode *inode)
765{
766 umode_t mode = inode->i_mode;
767
768 /* Special files should not get pinned to the filesystem. */
769 if (!S_ISREG(mode))
770 return false;
771
772 /* Setuid files should not get pinned to the filesystem. */
773 if (mode & S_ISUID)
774 return false;
775
776 /* Executable setgid files should not get pinned to the filesystem. */
777 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
778 return false;
779
780 /* Hardlinking to unreadable or unwritable sources is dangerous. */
781 if (inode_permission(inode, MAY_READ | MAY_WRITE))
782 return false;
783
784 return true;
785}
786
787/**
788 * may_linkat - Check permissions for creating a hardlink
789 * @link: the source to hardlink from
790 *
791 * Block hardlink when all of:
792 * - sysctl_protected_hardlinks enabled
793 * - fsuid does not match inode
794 * - hardlink source is unsafe (see safe_hardlink_source() above)
795 * - not CAP_FOWNER
796 *
797 * Returns 0 if successful, -ve on error.
798 */
799static int may_linkat(struct path *link)
800{
801 const struct cred *cred;
802 struct inode *inode;
803
804 if (!sysctl_protected_hardlinks)
805 return 0;
806
807 cred = current_cred();
808 inode = link->dentry->d_inode;
809
810 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
811 * otherwise, it must be a safe source.
812 */
Eric W. Biederman81abe272012-08-03 09:38:08 -0700813 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
Kees Cook800179c2012-07-25 17:29:07 -0700814 capable(CAP_FOWNER))
815 return 0;
816
Kees Cooka51d9ea2012-07-25 17:29:08 -0700817 audit_log_link_denied("linkat", link);
Kees Cook800179c2012-07-25 17:29:07 -0700818 return -EPERM;
819}
820
Al Virodef4af32009-12-26 08:37:05 -0500821static __always_inline int
Al Viro574197e2011-03-14 22:20:34 -0400822follow_link(struct path *link, struct nameidata *nd, void **p)
Ian Kent051d3812006-03-27 01:14:53 -0800823{
Nick Piggin7b9337a2011-01-14 08:42:43 +0000824 struct dentry *dentry = link->dentry;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400825 int error;
826 char *s;
Ian Kent051d3812006-03-27 01:14:53 -0800827
Al Viro844a3912011-02-15 00:38:26 -0500828 BUG_ON(nd->flags & LOOKUP_RCU);
829
Al Viro0e794582011-03-16 02:45:02 -0400830 if (link->mnt == nd->path.mnt)
831 mntget(link->mnt);
832
Al Viro6d7b5aa2012-06-10 04:15:17 -0400833 error = -ELOOP;
834 if (unlikely(current->total_link_count >= 40))
835 goto out_put_nd_path;
836
Al Viro574197e2011-03-14 22:20:34 -0400837 cond_resched();
838 current->total_link_count++;
839
Al Viro68ac1232012-03-15 08:21:57 -0400840 touch_atime(link);
Ian Kent051d3812006-03-27 01:14:53 -0800841 nd_set_link(nd, NULL);
842
Al Viro36f3b4f2011-02-22 21:24:38 -0500843 error = security_inode_follow_link(link->dentry, nd);
Al Viro6d7b5aa2012-06-10 04:15:17 -0400844 if (error)
845 goto out_put_nd_path;
Al Viro36f3b4f2011-02-22 21:24:38 -0500846
Al Viro86acdca12009-12-22 23:45:11 -0500847 nd->last_type = LAST_BIND;
Al Virodef4af32009-12-26 08:37:05 -0500848 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
849 error = PTR_ERR(*p);
Al Viro6d7b5aa2012-06-10 04:15:17 -0400850 if (IS_ERR(*p))
Christoph Hellwig408ef012012-06-18 10:47:03 -0400851 goto out_put_nd_path;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400852
853 error = 0;
854 s = nd_get_link(nd);
855 if (s) {
Al Viro443ed252013-09-10 12:00:43 -0400856 if (unlikely(IS_ERR(s))) {
857 path_put(&nd->path);
858 put_link(nd, link, *p);
859 return PTR_ERR(s);
860 }
861 if (*s == '/') {
862 set_root(nd);
863 path_put(&nd->path);
864 nd->path = nd->root;
865 path_get(&nd->root);
866 nd->flags |= LOOKUP_JUMPED;
867 }
868 nd->inode = nd->path.dentry->d_inode;
869 error = link_path_walk(s, nd);
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400870 if (unlikely(error))
871 put_link(nd, link, *p);
Ian Kent051d3812006-03-27 01:14:53 -0800872 }
Al Viro6d7b5aa2012-06-10 04:15:17 -0400873
874 return error;
875
876out_put_nd_path:
Arnd Bergmann98f6ef62012-10-11 13:20:00 +0000877 *p = NULL;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400878 path_put(&nd->path);
Al Viro6d7b5aa2012-06-10 04:15:17 -0400879 path_put(link);
Ian Kent051d3812006-03-27 01:14:53 -0800880 return error;
881}
882
Nick Piggin31e6b012011-01-07 17:49:52 +1100883static int follow_up_rcu(struct path *path)
884{
Al Viro0714a532011-11-24 22:19:58 -0500885 struct mount *mnt = real_mount(path->mnt);
886 struct mount *parent;
Nick Piggin31e6b012011-01-07 17:49:52 +1100887 struct dentry *mountpoint;
888
Al Viro0714a532011-11-24 22:19:58 -0500889 parent = mnt->mnt_parent;
890 if (&parent->mnt == path->mnt)
Nick Piggin31e6b012011-01-07 17:49:52 +1100891 return 0;
Al Viroa73324d2011-11-24 22:25:07 -0500892 mountpoint = mnt->mnt_mountpoint;
Nick Piggin31e6b012011-01-07 17:49:52 +1100893 path->dentry = mountpoint;
Al Viro0714a532011-11-24 22:19:58 -0500894 path->mnt = &parent->mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +1100895 return 1;
896}
897
David Howellsf015f1262012-06-25 12:55:28 +0100898/*
899 * follow_up - Find the mountpoint of path's vfsmount
900 *
901 * Given a path, find the mountpoint of its source file system.
902 * Replace @path with the path of the mountpoint in the parent mount.
903 * Up is towards /.
904 *
905 * Return 1 if we went up a level and 0 if we were already at the
906 * root.
907 */
Al Virobab77eb2009-04-18 03:26:48 -0400908int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909{
Al Viro0714a532011-11-24 22:19:58 -0500910 struct mount *mnt = real_mount(path->mnt);
911 struct mount *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +1000913
Al Viro48a066e2013-09-29 22:06:07 -0400914 read_seqlock_excl(&mount_lock);
Al Viro0714a532011-11-24 22:19:58 -0500915 parent = mnt->mnt_parent;
Al Viro3c0a6162012-07-18 17:32:50 +0400916 if (parent == mnt) {
Al Viro48a066e2013-09-29 22:06:07 -0400917 read_sequnlock_excl(&mount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 return 0;
919 }
Al Viro0714a532011-11-24 22:19:58 -0500920 mntget(&parent->mnt);
Al Viroa73324d2011-11-24 22:25:07 -0500921 mountpoint = dget(mnt->mnt_mountpoint);
Al Viro48a066e2013-09-29 22:06:07 -0400922 read_sequnlock_excl(&mount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400923 dput(path->dentry);
924 path->dentry = mountpoint;
925 mntput(path->mnt);
Al Viro0714a532011-11-24 22:19:58 -0500926 path->mnt = &parent->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 return 1;
928}
Al Viro4d359502014-03-14 12:20:17 -0400929EXPORT_SYMBOL(follow_up);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100931/*
David Howells9875cf82011-01-14 18:45:21 +0000932 * Perform an automount
933 * - return -EISDIR to tell follow_managed() to stop and return the path we
934 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 */
David Howells9875cf82011-01-14 18:45:21 +0000936static int follow_automount(struct path *path, unsigned flags,
937 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +1100938{
David Howells9875cf82011-01-14 18:45:21 +0000939 struct vfsmount *mnt;
David Howellsea5b7782011-01-14 19:10:03 +0000940 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +1100941
David Howells9875cf82011-01-14 18:45:21 +0000942 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
943 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -0700944
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200945 /* We don't want to mount if someone's just doing a stat -
946 * unless they're stat'ing a directory and appended a '/' to
947 * the name.
948 *
949 * We do, however, want to mount if someone wants to open or
950 * create a file of any type under the mountpoint, wants to
951 * traverse through the mountpoint or wants to open the
952 * mounted directory. Also, autofs may mark negative dentries
953 * as being automount points. These will need the attentions
954 * of the daemon to instantiate them before they can be used.
David Howells9875cf82011-01-14 18:45:21 +0000955 */
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200956 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
Linus Torvaldsd94c1772011-09-26 17:44:55 -0700957 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200958 path->dentry->d_inode)
959 return -EISDIR;
960
David Howells9875cf82011-01-14 18:45:21 +0000961 current->total_link_count++;
962 if (current->total_link_count >= 40)
963 return -ELOOP;
964
965 mnt = path->dentry->d_op->d_automount(path);
966 if (IS_ERR(mnt)) {
967 /*
968 * The filesystem is allowed to return -EISDIR here to indicate
969 * it doesn't want to automount. For instance, autofs would do
970 * this so that its userspace daemon can mount on this dentry.
971 *
972 * However, we can only permit this if it's a terminal point in
973 * the path being looked up; if it wasn't then the remainder of
974 * the path is inaccessible and we should say so.
975 */
Al Viro49084c32011-06-25 21:59:52 -0400976 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
David Howells9875cf82011-01-14 18:45:21 +0000977 return -EREMOTE;
978 return PTR_ERR(mnt);
979 }
David Howellsea5b7782011-01-14 19:10:03 +0000980
David Howells9875cf82011-01-14 18:45:21 +0000981 if (!mnt) /* mount collision */
982 return 0;
983
Al Viro8aef1882011-06-16 15:10:06 +0100984 if (!*need_mntput) {
985 /* lock_mount() may release path->mnt on error */
986 mntget(path->mnt);
987 *need_mntput = true;
988 }
Al Viro19a167a2011-01-17 01:35:23 -0500989 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +0000990
David Howellsea5b7782011-01-14 19:10:03 +0000991 switch (err) {
992 case -EBUSY:
993 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -0500994 return 0;
David Howellsea5b7782011-01-14 19:10:03 +0000995 case 0:
Al Viro8aef1882011-06-16 15:10:06 +0100996 path_put(path);
David Howellsea5b7782011-01-14 19:10:03 +0000997 path->mnt = mnt;
998 path->dentry = dget(mnt->mnt_root);
David Howellsea5b7782011-01-14 19:10:03 +0000999 return 0;
Al Viro19a167a2011-01-17 01:35:23 -05001000 default:
1001 return err;
David Howellsea5b7782011-01-14 19:10:03 +00001002 }
Al Viro19a167a2011-01-17 01:35:23 -05001003
David Howells9875cf82011-01-14 18:45:21 +00001004}
1005
1006/*
1007 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +00001008 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +00001009 * - Flagged as mountpoint
1010 * - Flagged as automount point
1011 *
1012 * This may only be called in refwalk mode.
1013 *
1014 * Serialization is taken care of in namespace.c
1015 */
1016static int follow_managed(struct path *path, unsigned flags)
1017{
Al Viro8aef1882011-06-16 15:10:06 +01001018 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
David Howells9875cf82011-01-14 18:45:21 +00001019 unsigned managed;
1020 bool need_mntput = false;
Al Viro8aef1882011-06-16 15:10:06 +01001021 int ret = 0;
David Howells9875cf82011-01-14 18:45:21 +00001022
1023 /* Given that we're not holding a lock here, we retain the value in a
1024 * local variable for each dentry as we look at it so that we don't see
1025 * the components of that value change under us */
1026 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1027 managed &= DCACHE_MANAGED_DENTRY,
1028 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +00001029 /* Allow the filesystem to manage the transit without i_mutex
1030 * being held. */
1031 if (managed & DCACHE_MANAGE_TRANSIT) {
1032 BUG_ON(!path->dentry->d_op);
1033 BUG_ON(!path->dentry->d_op->d_manage);
Al Viro1aed3e42011-03-18 09:09:02 -04001034 ret = path->dentry->d_op->d_manage(path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001035 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001036 break;
David Howellscc53ce52011-01-14 18:45:26 +00001037 }
1038
David Howells9875cf82011-01-14 18:45:21 +00001039 /* Transit to a mounted filesystem. */
1040 if (managed & DCACHE_MOUNTED) {
1041 struct vfsmount *mounted = lookup_mnt(path);
1042 if (mounted) {
1043 dput(path->dentry);
1044 if (need_mntput)
1045 mntput(path->mnt);
1046 path->mnt = mounted;
1047 path->dentry = dget(mounted->mnt_root);
1048 need_mntput = true;
1049 continue;
1050 }
1051
1052 /* Something is mounted on this dentry in another
1053 * namespace and/or whatever was mounted there in this
Al Viro48a066e2013-09-29 22:06:07 -04001054 * namespace got unmounted before lookup_mnt() could
1055 * get it */
David Howells9875cf82011-01-14 18:45:21 +00001056 }
1057
1058 /* Handle an automount point */
1059 if (managed & DCACHE_NEED_AUTOMOUNT) {
1060 ret = follow_automount(path, flags, &need_mntput);
1061 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +01001062 break;
David Howells9875cf82011-01-14 18:45:21 +00001063 continue;
1064 }
1065
1066 /* We didn't change the current path point */
1067 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 }
Al Viro8aef1882011-06-16 15:10:06 +01001069
1070 if (need_mntput && path->mnt == mnt)
1071 mntput(path->mnt);
1072 if (ret == -EISDIR)
1073 ret = 0;
Al Viroa3fbbde2011-11-07 21:21:26 +00001074 return ret < 0 ? ret : need_mntput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075}
1076
David Howellscc53ce52011-01-14 18:45:26 +00001077int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078{
1079 struct vfsmount *mounted;
1080
Al Viro1c755af2009-04-18 14:06:57 -04001081 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -04001083 dput(path->dentry);
1084 mntput(path->mnt);
1085 path->mnt = mounted;
1086 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 return 1;
1088 }
1089 return 0;
1090}
Al Viro4d359502014-03-14 12:20:17 -04001091EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092
Ian Kent62a73752011-03-25 01:51:02 +08001093static inline bool managed_dentry_might_block(struct dentry *dentry)
1094{
1095 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
1096 dentry->d_op->d_manage(dentry, true) < 0);
1097}
1098
David Howells9875cf82011-01-14 18:45:21 +00001099/*
Al Viro287548e2011-05-27 06:50:06 -04001100 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1101 * we meet a managed dentry that would need blocking.
David Howells9875cf82011-01-14 18:45:21 +00001102 */
1103static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
Al Viro287548e2011-05-27 06:50:06 -04001104 struct inode **inode)
David Howells9875cf82011-01-14 18:45:21 +00001105{
Ian Kent62a73752011-03-25 01:51:02 +08001106 for (;;) {
Al Viroc7105362011-11-24 18:22:03 -05001107 struct mount *mounted;
Ian Kent62a73752011-03-25 01:51:02 +08001108 /*
1109 * Don't forget we might have a non-mountpoint managed dentry
1110 * that wants to block transit.
1111 */
Al Viro287548e2011-05-27 06:50:06 -04001112 if (unlikely(managed_dentry_might_block(path->dentry)))
David Howellsab909112011-01-14 18:46:51 +00001113 return false;
Ian Kent62a73752011-03-25 01:51:02 +08001114
1115 if (!d_mountpoint(path->dentry))
Al Virob37199e2014-03-20 15:18:22 -04001116 return true;
Ian Kent62a73752011-03-25 01:51:02 +08001117
Al Viro474279d2013-10-01 16:11:26 -04001118 mounted = __lookup_mnt(path->mnt, path->dentry);
David Howells9875cf82011-01-14 18:45:21 +00001119 if (!mounted)
1120 break;
Al Viroc7105362011-11-24 18:22:03 -05001121 path->mnt = &mounted->mnt;
1122 path->dentry = mounted->mnt.mnt_root;
Al Viroa3fbbde2011-11-07 21:21:26 +00001123 nd->flags |= LOOKUP_JUMPED;
David Howells9875cf82011-01-14 18:45:21 +00001124 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
Linus Torvalds59430262011-07-18 15:43:29 -07001125 /*
1126 * Update the inode too. We don't need to re-check the
1127 * dentry sequence number here after this d_inode read,
1128 * because a mount-point is always pinned.
1129 */
1130 *inode = path->dentry->d_inode;
David Howells9875cf82011-01-14 18:45:21 +00001131 }
Al Virob37199e2014-03-20 15:18:22 -04001132 return read_seqretry(&mount_lock, nd->m_seq);
Al Viro287548e2011-05-27 06:50:06 -04001133}
1134
Nick Piggin31e6b012011-01-07 17:49:52 +11001135static int follow_dotdot_rcu(struct nameidata *nd)
1136{
Nick Piggin31e6b012011-01-07 17:49:52 +11001137 set_root_rcu(nd);
1138
David Howells9875cf82011-01-14 18:45:21 +00001139 while (1) {
Nick Piggin31e6b012011-01-07 17:49:52 +11001140 if (nd->path.dentry == nd->root.dentry &&
1141 nd->path.mnt == nd->root.mnt) {
1142 break;
1143 }
1144 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1145 struct dentry *old = nd->path.dentry;
1146 struct dentry *parent = old->d_parent;
1147 unsigned seq;
1148
1149 seq = read_seqcount_begin(&parent->d_seq);
1150 if (read_seqcount_retry(&old->d_seq, nd->seq))
Al Viroef7562d2011-03-04 14:35:59 -05001151 goto failed;
Nick Piggin31e6b012011-01-07 17:49:52 +11001152 nd->path.dentry = parent;
1153 nd->seq = seq;
1154 break;
1155 }
1156 if (!follow_up_rcu(&nd->path))
1157 break;
1158 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11001159 }
Al Virob37199e2014-03-20 15:18:22 -04001160 while (d_mountpoint(nd->path.dentry)) {
1161 struct mount *mounted;
1162 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1163 if (!mounted)
1164 break;
1165 nd->path.mnt = &mounted->mnt;
1166 nd->path.dentry = mounted->mnt.mnt_root;
1167 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1168 if (!read_seqretry(&mount_lock, nd->m_seq))
1169 goto failed;
1170 }
Al Virodea39372011-05-27 06:53:39 -04001171 nd->inode = nd->path.dentry->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001172 return 0;
Al Viroef7562d2011-03-04 14:35:59 -05001173
1174failed:
1175 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001176 if (!(nd->flags & LOOKUP_ROOT))
1177 nd->root.mnt = NULL;
Al Viro8b61e742013-11-08 12:45:01 -05001178 rcu_read_unlock();
Al Viroef7562d2011-03-04 14:35:59 -05001179 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001180}
1181
David Howells9875cf82011-01-14 18:45:21 +00001182/*
David Howellscc53ce52011-01-14 18:45:26 +00001183 * Follow down to the covering mount currently visible to userspace. At each
1184 * point, the filesystem owning that dentry may be queried as to whether the
1185 * caller is permitted to proceed or not.
David Howellscc53ce52011-01-14 18:45:26 +00001186 */
Al Viro7cc90cc2011-03-18 09:04:20 -04001187int follow_down(struct path *path)
David Howellscc53ce52011-01-14 18:45:26 +00001188{
1189 unsigned managed;
1190 int ret;
1191
1192 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1193 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1194 /* Allow the filesystem to manage the transit without i_mutex
1195 * being held.
1196 *
1197 * We indicate to the filesystem if someone is trying to mount
1198 * something here. This gives autofs the chance to deny anyone
1199 * other than its daemon the right to mount on its
1200 * superstructure.
1201 *
1202 * The filesystem may sleep at this point.
1203 */
1204 if (managed & DCACHE_MANAGE_TRANSIT) {
1205 BUG_ON(!path->dentry->d_op);
1206 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +00001207 ret = path->dentry->d_op->d_manage(
Al Viro1aed3e42011-03-18 09:09:02 -04001208 path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001209 if (ret < 0)
1210 return ret == -EISDIR ? 0 : ret;
1211 }
1212
1213 /* Transit to a mounted filesystem. */
1214 if (managed & DCACHE_MOUNTED) {
1215 struct vfsmount *mounted = lookup_mnt(path);
1216 if (!mounted)
1217 break;
1218 dput(path->dentry);
1219 mntput(path->mnt);
1220 path->mnt = mounted;
1221 path->dentry = dget(mounted->mnt_root);
1222 continue;
1223 }
1224
1225 /* Don't handle automount points here */
1226 break;
1227 }
1228 return 0;
1229}
Al Viro4d359502014-03-14 12:20:17 -04001230EXPORT_SYMBOL(follow_down);
David Howellscc53ce52011-01-14 18:45:26 +00001231
1232/*
David Howells9875cf82011-01-14 18:45:21 +00001233 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1234 */
1235static void follow_mount(struct path *path)
1236{
1237 while (d_mountpoint(path->dentry)) {
1238 struct vfsmount *mounted = lookup_mnt(path);
1239 if (!mounted)
1240 break;
1241 dput(path->dentry);
1242 mntput(path->mnt);
1243 path->mnt = mounted;
1244 path->dentry = dget(mounted->mnt_root);
1245 }
1246}
1247
Nick Piggin31e6b012011-01-07 17:49:52 +11001248static void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249{
Al Viro2a737872009-04-07 11:49:53 -04001250 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001251
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 while(1) {
Jan Blunck4ac91372008-02-14 19:34:32 -08001253 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254
Al Viro2a737872009-04-07 11:49:53 -04001255 if (nd->path.dentry == nd->root.dentry &&
1256 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 break;
1258 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001259 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Al Viro3088dd72010-01-30 15:47:29 -05001260 /* rare case of legitimate dget_parent()... */
1261 nd->path.dentry = dget_parent(nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 dput(old);
1263 break;
1264 }
Al Viro3088dd72010-01-30 15:47:29 -05001265 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 }
Al Viro79ed0222009-04-18 13:59:41 -04001268 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001269 nd->inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270}
1271
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272/*
Miklos Szeredibad61182012-03-26 12:54:24 +02001273 * This looks up the name in dcache, possibly revalidates the old dentry and
1274 * allocates a new one if not found or not valid. In the need_lookup argument
1275 * returns whether i_op->lookup is necessary.
1276 *
1277 * dir->d_inode->i_mutex must be held
Nick Pigginbaa03892010-08-18 04:37:31 +10001278 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001279static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
Al Viro201f9562012-06-22 12:42:10 +04001280 unsigned int flags, bool *need_lookup)
Nick Pigginbaa03892010-08-18 04:37:31 +10001281{
Nick Pigginbaa03892010-08-18 04:37:31 +10001282 struct dentry *dentry;
Miklos Szeredibad61182012-03-26 12:54:24 +02001283 int error;
Nick Pigginbaa03892010-08-18 04:37:31 +10001284
Miklos Szeredibad61182012-03-26 12:54:24 +02001285 *need_lookup = false;
1286 dentry = d_lookup(dir, name);
1287 if (dentry) {
Jeff Layton39e3c952012-11-28 11:30:53 -05001288 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
Al Viro201f9562012-06-22 12:42:10 +04001289 error = d_revalidate(dentry, flags);
Miklos Szeredibad61182012-03-26 12:54:24 +02001290 if (unlikely(error <= 0)) {
1291 if (error < 0) {
1292 dput(dentry);
1293 return ERR_PTR(error);
1294 } else if (!d_invalidate(dentry)) {
1295 dput(dentry);
1296 dentry = NULL;
1297 }
1298 }
1299 }
1300 }
Nick Pigginbaa03892010-08-18 04:37:31 +10001301
Miklos Szeredibad61182012-03-26 12:54:24 +02001302 if (!dentry) {
1303 dentry = d_alloc(dir, name);
1304 if (unlikely(!dentry))
1305 return ERR_PTR(-ENOMEM);
Nick Pigginbaa03892010-08-18 04:37:31 +10001306
Miklos Szeredibad61182012-03-26 12:54:24 +02001307 *need_lookup = true;
Nick Pigginbaa03892010-08-18 04:37:31 +10001308 }
1309 return dentry;
1310}
1311
1312/*
J. Bruce Fields13a2c3b2013-10-23 16:09:16 -04001313 * Call i_op->lookup on the dentry. The dentry must be negative and
1314 * unhashed.
Miklos Szeredibad61182012-03-26 12:54:24 +02001315 *
1316 * dir->d_inode->i_mutex must be held
Josef Bacik44396f42011-05-31 11:58:49 -04001317 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001318static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
Al Viro72bd8662012-06-10 17:17:17 -04001319 unsigned int flags)
Josef Bacik44396f42011-05-31 11:58:49 -04001320{
Josef Bacik44396f42011-05-31 11:58:49 -04001321 struct dentry *old;
1322
1323 /* Don't create child dentry for a dead directory. */
Miklos Szeredibad61182012-03-26 12:54:24 +02001324 if (unlikely(IS_DEADDIR(dir))) {
Miklos Szeredie188dc02012-02-03 14:25:18 +01001325 dput(dentry);
Josef Bacik44396f42011-05-31 11:58:49 -04001326 return ERR_PTR(-ENOENT);
Miklos Szeredie188dc02012-02-03 14:25:18 +01001327 }
Josef Bacik44396f42011-05-31 11:58:49 -04001328
Al Viro72bd8662012-06-10 17:17:17 -04001329 old = dir->i_op->lookup(dir, dentry, flags);
Josef Bacik44396f42011-05-31 11:58:49 -04001330 if (unlikely(old)) {
1331 dput(dentry);
1332 dentry = old;
1333 }
1334 return dentry;
1335}
1336
Al Viroa3255542012-03-30 14:41:51 -04001337static struct dentry *__lookup_hash(struct qstr *name,
Al Viro72bd8662012-06-10 17:17:17 -04001338 struct dentry *base, unsigned int flags)
Al Viroa3255542012-03-30 14:41:51 -04001339{
Miklos Szeredibad61182012-03-26 12:54:24 +02001340 bool need_lookup;
Al Viroa3255542012-03-30 14:41:51 -04001341 struct dentry *dentry;
1342
Al Viro72bd8662012-06-10 17:17:17 -04001343 dentry = lookup_dcache(name, base, flags, &need_lookup);
Miklos Szeredibad61182012-03-26 12:54:24 +02001344 if (!need_lookup)
1345 return dentry;
Al Viroa3255542012-03-30 14:41:51 -04001346
Al Viro72bd8662012-06-10 17:17:17 -04001347 return lookup_real(base->d_inode, dentry, flags);
Al Viroa3255542012-03-30 14:41:51 -04001348}
1349
Josef Bacik44396f42011-05-31 11:58:49 -04001350/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 * It's more convoluted than I'd like it to be, but... it's still fairly
1352 * small and for now I'd prefer to have fast path as straight as possible.
1353 * It _is_ time-critical.
1354 */
Al Viroe97cdc82013-01-24 18:16:00 -05001355static int lookup_fast(struct nameidata *nd,
Miklos Szeredi697f5142012-05-21 17:30:05 +02001356 struct path *path, struct inode **inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357{
Jan Blunck4ac91372008-02-14 19:34:32 -08001358 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001359 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001360 int need_reval = 1;
1361 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001362 int err;
1363
Al Viro3cac2602009-08-13 18:27:43 +04001364 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001365 * Rename seqlock is not required here because in the off chance
1366 * of a false negative due to a concurrent rename, we're going to
1367 * do the non-racy lookup, below.
1368 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001369 if (nd->flags & LOOKUP_RCU) {
1370 unsigned seq;
Linus Torvaldsda53be12013-05-21 15:22:44 -07001371 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
Al Viro5a18fff2011-03-11 04:44:53 -05001372 if (!dentry)
1373 goto unlazy;
1374
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001375 /*
1376 * This sequence count validates that the inode matches
1377 * the dentry name information from lookup.
1378 */
1379 *inode = dentry->d_inode;
1380 if (read_seqcount_retry(&dentry->d_seq, seq))
1381 return -ECHILD;
1382
1383 /*
1384 * This sequence count validates that the parent had no
1385 * changes while we did the lookup of the dentry above.
1386 *
1387 * The memory barrier in read_seqcount_begin of child is
1388 * enough, we can use __read_seqcount_retry here.
1389 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001390 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1391 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001392 nd->seq = seq;
Al Viro5a18fff2011-03-11 04:44:53 -05001393
Al Viro24643082011-02-15 01:26:22 -05001394 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
Al Viro4ce16ef32012-06-10 16:10:59 -04001395 status = d_revalidate(dentry, nd->flags);
Al Viro5a18fff2011-03-11 04:44:53 -05001396 if (unlikely(status <= 0)) {
1397 if (status != -ECHILD)
1398 need_reval = 0;
1399 goto unlazy;
1400 }
Al Viro24643082011-02-15 01:26:22 -05001401 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001402 path->mnt = mnt;
1403 path->dentry = dentry;
Al Virod6e9bd22011-05-27 07:03:15 -04001404 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1405 goto unlazy;
1406 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1407 goto unlazy;
1408 return 0;
Al Viro5a18fff2011-03-11 04:44:53 -05001409unlazy:
Al Viro19660af2011-03-25 10:32:48 -04001410 if (unlazy_walk(nd, dentry))
1411 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001412 } else {
Al Viroe97cdc82013-01-24 18:16:00 -05001413 dentry = __d_lookup(parent, &nd->last);
Nick Piggin31e6b012011-01-07 17:49:52 +11001414 }
Al Viro5a18fff2011-03-11 04:44:53 -05001415
Al Viro81e6f522012-03-30 14:48:04 -04001416 if (unlikely(!dentry))
1417 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001418
Al Viro5a18fff2011-03-11 04:44:53 -05001419 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
Al Viro4ce16ef32012-06-10 16:10:59 -04001420 status = d_revalidate(dentry, nd->flags);
Al Viro5a18fff2011-03-11 04:44:53 -05001421 if (unlikely(status <= 0)) {
1422 if (status < 0) {
1423 dput(dentry);
1424 return status;
1425 }
1426 if (!d_invalidate(dentry)) {
1427 dput(dentry);
Al Viro81e6f522012-03-30 14:48:04 -04001428 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001429 }
1430 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001431
David Howells9875cf82011-01-14 18:45:21 +00001432 path->mnt = mnt;
1433 path->dentry = dentry;
1434 err = follow_managed(path, nd->flags);
Ian Kent89312212011-01-18 12:06:10 +08001435 if (unlikely(err < 0)) {
1436 path_put_conditional(path, nd);
David Howells9875cf82011-01-14 18:45:21 +00001437 return err;
Ian Kent89312212011-01-18 12:06:10 +08001438 }
Al Viroa3fbbde2011-11-07 21:21:26 +00001439 if (err)
1440 nd->flags |= LOOKUP_JUMPED;
David Howells9875cf82011-01-14 18:45:21 +00001441 *inode = path->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 return 0;
Al Viro81e6f522012-03-30 14:48:04 -04001443
1444need_lookup:
Miklos Szeredi697f5142012-05-21 17:30:05 +02001445 return 1;
1446}
1447
1448/* Fast lookup failed, do it the slow way */
Al Virocc2a5272013-01-24 18:19:49 -05001449static int lookup_slow(struct nameidata *nd, struct path *path)
Miklos Szeredi697f5142012-05-21 17:30:05 +02001450{
1451 struct dentry *dentry, *parent;
1452 int err;
1453
1454 parent = nd->path.dentry;
Al Viro81e6f522012-03-30 14:48:04 -04001455 BUG_ON(nd->inode != parent->d_inode);
1456
1457 mutex_lock(&parent->d_inode->i_mutex);
Al Virocc2a5272013-01-24 18:19:49 -05001458 dentry = __lookup_hash(&nd->last, parent, nd->flags);
Al Viro81e6f522012-03-30 14:48:04 -04001459 mutex_unlock(&parent->d_inode->i_mutex);
1460 if (IS_ERR(dentry))
1461 return PTR_ERR(dentry);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001462 path->mnt = nd->path.mnt;
1463 path->dentry = dentry;
1464 err = follow_managed(path, nd->flags);
1465 if (unlikely(err < 0)) {
1466 path_put_conditional(path, nd);
1467 return err;
1468 }
1469 if (err)
1470 nd->flags |= LOOKUP_JUMPED;
1471 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472}
1473
Al Viro52094c82011-02-21 21:34:47 -05001474static inline int may_lookup(struct nameidata *nd)
1475{
1476 if (nd->flags & LOOKUP_RCU) {
Al Viro4ad5abb2011-06-20 19:57:03 -04001477 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
Al Viro52094c82011-02-21 21:34:47 -05001478 if (err != -ECHILD)
1479 return err;
Al Viro19660af2011-03-25 10:32:48 -04001480 if (unlazy_walk(nd, NULL))
Al Viro52094c82011-02-21 21:34:47 -05001481 return -ECHILD;
1482 }
Al Viro4ad5abb2011-06-20 19:57:03 -04001483 return inode_permission(nd->inode, MAY_EXEC);
Al Viro52094c82011-02-21 21:34:47 -05001484}
1485
Al Viro9856fa12011-03-04 14:22:06 -05001486static inline int handle_dots(struct nameidata *nd, int type)
1487{
1488 if (type == LAST_DOTDOT) {
1489 if (nd->flags & LOOKUP_RCU) {
1490 if (follow_dotdot_rcu(nd))
1491 return -ECHILD;
1492 } else
1493 follow_dotdot(nd);
1494 }
1495 return 0;
1496}
1497
Al Viro951361f2011-03-04 14:44:37 -05001498static void terminate_walk(struct nameidata *nd)
1499{
1500 if (!(nd->flags & LOOKUP_RCU)) {
1501 path_put(&nd->path);
1502 } else {
1503 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001504 if (!(nd->flags & LOOKUP_ROOT))
1505 nd->root.mnt = NULL;
Al Viro8b61e742013-11-08 12:45:01 -05001506 rcu_read_unlock();
Al Viro951361f2011-03-04 14:44:37 -05001507 }
1508}
1509
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001510/*
1511 * Do we need to follow links? We _really_ want to be able
1512 * to do this check without having to look at inode->i_op,
1513 * so we keep a cache of "no, this doesn't need follow_link"
1514 * for the common case.
1515 */
David Howellsb18825a2013-09-12 19:22:53 +01001516static inline int should_follow_link(struct dentry *dentry, int follow)
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001517{
David Howellsb18825a2013-09-12 19:22:53 +01001518 return unlikely(d_is_symlink(dentry)) ? follow : 0;
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001519}
1520
Al Viroce57dfc2011-03-13 19:58:58 -04001521static inline int walk_component(struct nameidata *nd, struct path *path,
Al Viro21b9b072013-01-24 18:10:25 -05001522 int follow)
Al Viroce57dfc2011-03-13 19:58:58 -04001523{
1524 struct inode *inode;
1525 int err;
1526 /*
1527 * "." and ".." are special - ".." especially so because it has
1528 * to be able to know about the current root directory and
1529 * parent relationships.
1530 */
Al Viro21b9b072013-01-24 18:10:25 -05001531 if (unlikely(nd->last_type != LAST_NORM))
1532 return handle_dots(nd, nd->last_type);
Al Viroe97cdc82013-01-24 18:16:00 -05001533 err = lookup_fast(nd, path, &inode);
Al Viroce57dfc2011-03-13 19:58:58 -04001534 if (unlikely(err)) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001535 if (err < 0)
1536 goto out_err;
1537
Al Virocc2a5272013-01-24 18:19:49 -05001538 err = lookup_slow(nd, path);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001539 if (err < 0)
1540 goto out_err;
1541
1542 inode = path->dentry->d_inode;
Al Viroce57dfc2011-03-13 19:58:58 -04001543 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001544 err = -ENOENT;
1545 if (!inode)
1546 goto out_path_put;
1547
David Howellsb18825a2013-09-12 19:22:53 +01001548 if (should_follow_link(path->dentry, follow)) {
Al Viro19660af2011-03-25 10:32:48 -04001549 if (nd->flags & LOOKUP_RCU) {
1550 if (unlikely(unlazy_walk(nd, path->dentry))) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001551 err = -ECHILD;
1552 goto out_err;
Al Viro19660af2011-03-25 10:32:48 -04001553 }
1554 }
Al Viroce57dfc2011-03-13 19:58:58 -04001555 BUG_ON(inode != path->dentry->d_inode);
1556 return 1;
1557 }
1558 path_to_nameidata(path, nd);
1559 nd->inode = inode;
1560 return 0;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001561
1562out_path_put:
1563 path_to_nameidata(path, nd);
1564out_err:
1565 terminate_walk(nd);
1566 return err;
Al Viroce57dfc2011-03-13 19:58:58 -04001567}
1568
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569/*
Al Virob3563792011-03-14 21:54:55 -04001570 * This limits recursive symlink follows to 8, while
1571 * limiting consecutive symlinks to 40.
1572 *
1573 * Without that kind of total limit, nasty chains of consecutive
1574 * symlinks can cause almost arbitrarily long lookups.
1575 */
1576static inline int nested_symlink(struct path *path, struct nameidata *nd)
1577{
1578 int res;
1579
Al Virob3563792011-03-14 21:54:55 -04001580 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1581 path_put_conditional(path, nd);
1582 path_put(&nd->path);
1583 return -ELOOP;
1584 }
Erez Zadok1a4022f2011-05-21 01:19:59 -04001585 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
Al Virob3563792011-03-14 21:54:55 -04001586
1587 nd->depth++;
1588 current->link_count++;
1589
1590 do {
1591 struct path link = *path;
1592 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04001593
1594 res = follow_link(&link, nd, &cookie);
Al Viro6d7b5aa2012-06-10 04:15:17 -04001595 if (res)
1596 break;
Al Viro21b9b072013-01-24 18:10:25 -05001597 res = walk_component(nd, path, LOOKUP_FOLLOW);
Al Viro574197e2011-03-14 22:20:34 -04001598 put_link(nd, &link, cookie);
Al Virob3563792011-03-14 21:54:55 -04001599 } while (res > 0);
1600
1601 current->link_count--;
1602 nd->depth--;
1603 return res;
1604}
1605
1606/*
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001607 * We can do the critical dentry name comparison and hashing
1608 * operations one word at a time, but we are limited to:
1609 *
1610 * - Architectures with fast unaligned word accesses. We could
1611 * do a "get_unaligned()" if this helps and is sufficiently
1612 * fast.
1613 *
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001614 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1615 * do not trap on the (extremely unlikely) case of a page
1616 * crossing operation.
1617 *
1618 * - Furthermore, we need an efficient 64-bit compile for the
1619 * 64-bit case in order to generate the "number of bytes in
1620 * the final mask". Again, that could be replaced with a
1621 * efficient population count instruction or similar.
1622 */
1623#ifdef CONFIG_DCACHE_WORD_ACCESS
1624
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001625#include <asm/word-at-a-time.h>
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001626
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001627#ifdef CONFIG_64BIT
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001628
1629static inline unsigned int fold_hash(unsigned long hash)
1630{
1631 hash += hash >> (8*sizeof(int));
1632 return hash;
1633}
1634
1635#else /* 32-bit case */
1636
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001637#define fold_hash(x) (x)
1638
1639#endif
1640
1641unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1642{
1643 unsigned long a, mask;
1644 unsigned long hash = 0;
1645
1646 for (;;) {
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001647 a = load_unaligned_zeropad(name);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001648 if (len < sizeof(unsigned long))
1649 break;
1650 hash += a;
Al Virof132c5b2012-03-22 21:59:52 +00001651 hash *= 9;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001652 name += sizeof(unsigned long);
1653 len -= sizeof(unsigned long);
1654 if (!len)
1655 goto done;
1656 }
Will Deacona5c21dc2013-12-12 17:40:21 +00001657 mask = bytemask_from_count(len);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001658 hash += mask & a;
1659done:
1660 return fold_hash(hash);
1661}
1662EXPORT_SYMBOL(full_name_hash);
1663
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001664/*
1665 * Calculate the length and hash of the path component, and
1666 * return the length of the component;
1667 */
1668static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1669{
Linus Torvalds36126f82012-05-26 10:43:17 -07001670 unsigned long a, b, adata, bdata, mask, hash, len;
1671 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001672
1673 hash = a = 0;
1674 len = -sizeof(unsigned long);
1675 do {
1676 hash = (hash + a) * 9;
1677 len += sizeof(unsigned long);
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001678 a = load_unaligned_zeropad(name+len);
Linus Torvalds36126f82012-05-26 10:43:17 -07001679 b = a ^ REPEAT_BYTE('/');
1680 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001681
Linus Torvalds36126f82012-05-26 10:43:17 -07001682 adata = prep_zero_mask(a, adata, &constants);
1683 bdata = prep_zero_mask(b, bdata, &constants);
1684
1685 mask = create_zero_mask(adata | bdata);
1686
1687 hash += a & zero_bytemask(mask);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001688 *hashp = fold_hash(hash);
1689
Linus Torvalds36126f82012-05-26 10:43:17 -07001690 return len + find_zero(mask);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001691}
1692
1693#else
1694
Linus Torvalds0145acc2012-03-02 14:32:59 -08001695unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1696{
1697 unsigned long hash = init_name_hash();
1698 while (len--)
1699 hash = partial_name_hash(*name++, hash);
1700 return end_name_hash(hash);
1701}
Linus Torvaldsae942ae2012-03-02 19:40:57 -08001702EXPORT_SYMBOL(full_name_hash);
Linus Torvalds0145acc2012-03-02 14:32:59 -08001703
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001704/*
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001705 * We know there's a real path component here of at least
1706 * one character.
1707 */
1708static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1709{
1710 unsigned long hash = init_name_hash();
1711 unsigned long len = 0, c;
1712
1713 c = (unsigned char)*name;
1714 do {
1715 len++;
1716 hash = partial_name_hash(c, hash);
1717 c = (unsigned char)name[len];
1718 } while (c && c != '/');
1719 *hashp = end_name_hash(hash);
1720 return len;
1721}
1722
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001723#endif
1724
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001725/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01001727 * This is the basic name resolution function, turning a pathname into
1728 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01001730 * Returns 0 and nd will have valid dentry and mnt on success.
1731 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732 */
Al Viro6de88d72009-08-09 01:41:57 +04001733static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734{
1735 struct path next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737
1738 while (*name=='/')
1739 name++;
1740 if (!*name)
Al Viro086e1832011-02-22 20:56:27 -05001741 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 /* At this point we know we have a real path component. */
1744 for(;;) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745 struct qstr this;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001746 long len;
Al Virofe479a52011-02-22 15:10:03 -05001747 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748
Al Viro52094c82011-02-21 21:34:47 -05001749 err = may_lookup(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 if (err)
1751 break;
1752
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001753 len = hash_name(name, &this.hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 this.name = name;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001755 this.len = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756
Al Virofe479a52011-02-22 15:10:03 -05001757 type = LAST_NORM;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001758 if (name[0] == '.') switch (len) {
Al Virofe479a52011-02-22 15:10:03 -05001759 case 2:
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001760 if (name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05001761 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05001762 nd->flags |= LOOKUP_JUMPED;
1763 }
Al Virofe479a52011-02-22 15:10:03 -05001764 break;
1765 case 1:
1766 type = LAST_DOT;
1767 }
Al Viro5a202bc2011-03-08 14:17:44 -05001768 if (likely(type == LAST_NORM)) {
1769 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05001770 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05001771 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07001772 err = parent->d_op->d_hash(parent, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05001773 if (err < 0)
1774 break;
1775 }
1776 }
Al Virofe479a52011-02-22 15:10:03 -05001777
Al Viro5f4a6a62013-01-24 18:04:22 -05001778 nd->last = this;
1779 nd->last_type = type;
1780
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001781 if (!name[len])
Al Viro5f4a6a62013-01-24 18:04:22 -05001782 return 0;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001783 /*
1784 * If it wasn't NUL, we know it was '/'. Skip that
1785 * slash, and continue until no more slashes.
1786 */
1787 do {
1788 len++;
1789 } while (unlikely(name[len] == '/'));
1790 if (!name[len])
Al Viro5f4a6a62013-01-24 18:04:22 -05001791 return 0;
1792
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001793 name += len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794
Al Viro21b9b072013-01-24 18:10:25 -05001795 err = walk_component(nd, &next, LOOKUP_FOLLOW);
Al Viroce57dfc2011-03-13 19:58:58 -04001796 if (err < 0)
1797 return err;
Al Virofe479a52011-02-22 15:10:03 -05001798
Al Viroce57dfc2011-03-13 19:58:58 -04001799 if (err) {
Al Virob3563792011-03-14 21:54:55 -04001800 err = nested_symlink(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 if (err)
Al Viroa7472ba2011-03-04 14:39:30 -05001802 return err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001803 }
Miklos Szeredi44b1d532014-04-01 17:08:41 +02001804 if (!d_can_lookup(nd->path.dentry)) {
Al Viro5f4a6a62013-01-24 18:04:22 -05001805 err = -ENOTDIR;
1806 break;
1807 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 }
Al Viro951361f2011-03-04 14:44:37 -05001809 terminate_walk(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 return err;
1811}
1812
Al Viro70e9b352011-03-05 21:12:22 -05001813static int path_init(int dfd, const char *name, unsigned int flags,
1814 struct nameidata *nd, struct file **fp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001816 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817
1818 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro16c2cd72011-02-22 15:50:10 -05001819 nd->flags = flags | LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 nd->depth = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05001821 if (flags & LOOKUP_ROOT) {
David Howellsb18825a2013-09-12 19:22:53 +01001822 struct dentry *root = nd->root.dentry;
1823 struct inode *inode = root->d_inode;
Al Viro73d049a2011-03-11 12:08:24 -05001824 if (*name) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02001825 if (!d_can_lookup(root))
Al Viro73d049a2011-03-11 12:08:24 -05001826 return -ENOTDIR;
1827 retval = inode_permission(inode, MAY_EXEC);
1828 if (retval)
1829 return retval;
1830 }
Al Viro5b6ca022011-03-09 23:04:47 -05001831 nd->path = nd->root;
1832 nd->inode = inode;
1833 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05001834 rcu_read_lock();
Al Viro5b6ca022011-03-09 23:04:47 -05001835 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro48a066e2013-09-29 22:06:07 -04001836 nd->m_seq = read_seqbegin(&mount_lock);
Al Viro5b6ca022011-03-09 23:04:47 -05001837 } else {
1838 path_get(&nd->path);
1839 }
1840 return 0;
1841 }
1842
Al Viro2a737872009-04-07 11:49:53 -04001843 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844
Al Viro48a066e2013-09-29 22:06:07 -04001845 nd->m_seq = read_seqbegin(&mount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846 if (*name=='/') {
Al Viroe41f7d42011-02-22 14:02:58 -05001847 if (flags & LOOKUP_RCU) {
Al Viro8b61e742013-11-08 12:45:01 -05001848 rcu_read_lock();
Al Viroe41f7d42011-02-22 14:02:58 -05001849 set_root_rcu(nd);
1850 } else {
1851 set_root(nd);
1852 path_get(&nd->root);
1853 }
Al Viro2a737872009-04-07 11:49:53 -04001854 nd->path = nd->root;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001855 } else if (dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05001856 if (flags & LOOKUP_RCU) {
1857 struct fs_struct *fs = current->fs;
1858 unsigned seq;
1859
Al Viro8b61e742013-11-08 12:45:01 -05001860 rcu_read_lock();
Al Viroe41f7d42011-02-22 14:02:58 -05001861
1862 do {
1863 seq = read_seqcount_begin(&fs->seq);
1864 nd->path = fs->pwd;
1865 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1866 } while (read_seqcount_retry(&fs->seq, seq));
1867 } else {
1868 get_fs_pwd(current->fs, &nd->path);
1869 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001870 } else {
Jeff Layton582aa642012-12-11 08:56:16 -05001871 /* Caller must check execute permissions on the starting path component */
Al Viro2903ff02012-08-28 12:52:22 -04001872 struct fd f = fdget_raw(dfd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001873 struct dentry *dentry;
1874
Al Viro2903ff02012-08-28 12:52:22 -04001875 if (!f.file)
1876 return -EBADF;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001877
Al Viro2903ff02012-08-28 12:52:22 -04001878 dentry = f.file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001879
Al Virof52e0c12011-03-14 18:56:51 -04001880 if (*name) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02001881 if (!d_can_lookup(dentry)) {
Al Viro2903ff02012-08-28 12:52:22 -04001882 fdput(f);
1883 return -ENOTDIR;
1884 }
Al Virof52e0c12011-03-14 18:56:51 -04001885 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001886
Al Viro2903ff02012-08-28 12:52:22 -04001887 nd->path = f.file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05001888 if (flags & LOOKUP_RCU) {
Linus Torvalds9c225f22014-03-03 09:36:58 -08001889 if (f.flags & FDPUT_FPUT)
Al Viro2903ff02012-08-28 12:52:22 -04001890 *fp = f.file;
Al Viroe41f7d42011-02-22 14:02:58 -05001891 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro8b61e742013-11-08 12:45:01 -05001892 rcu_read_lock();
Al Viroe41f7d42011-02-22 14:02:58 -05001893 } else {
Al Viro2903ff02012-08-28 12:52:22 -04001894 path_get(&nd->path);
1895 fdput(f);
Al Viroe41f7d42011-02-22 14:02:58 -05001896 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897 }
Al Viroe41f7d42011-02-22 14:02:58 -05001898
Nick Piggin31e6b012011-01-07 17:49:52 +11001899 nd->inode = nd->path.dentry->d_inode;
Al Viro9b4a9b12009-04-07 11:44:16 -04001900 return 0;
Al Viro9b4a9b12009-04-07 11:44:16 -04001901}
1902
Al Virobd92d7f2011-03-14 19:54:59 -04001903static inline int lookup_last(struct nameidata *nd, struct path *path)
1904{
1905 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1906 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1907
1908 nd->flags &= ~LOOKUP_PARENT;
Al Viro21b9b072013-01-24 18:10:25 -05001909 return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW);
Al Virobd92d7f2011-03-14 19:54:59 -04001910}
1911
Al Viro9b4a9b12009-04-07 11:44:16 -04001912/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroee0827c2011-02-21 23:38:09 -05001913static int path_lookupat(int dfd, const char *name,
Al Viro9b4a9b12009-04-07 11:44:16 -04001914 unsigned int flags, struct nameidata *nd)
1915{
Al Viro70e9b352011-03-05 21:12:22 -05001916 struct file *base = NULL;
Al Virobd92d7f2011-03-14 19:54:59 -04001917 struct path path;
1918 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001919
1920 /*
1921 * Path walking is largely split up into 2 different synchronisation
1922 * schemes, rcu-walk and ref-walk (explained in
1923 * Documentation/filesystems/path-lookup.txt). These share much of the
1924 * path walk code, but some things particularly setup, cleanup, and
1925 * following mounts are sufficiently divergent that functions are
1926 * duplicated. Typically there is a function foo(), and its RCU
1927 * analogue, foo_rcu().
1928 *
1929 * -ECHILD is the error number of choice (just to avoid clashes) that
1930 * is returned if some aspect of an rcu-walk fails. Such an error must
1931 * be handled by restarting a traditional ref-walk (which will always
1932 * be able to complete).
1933 */
Al Virobd92d7f2011-03-14 19:54:59 -04001934 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
Al Viroee0827c2011-02-21 23:38:09 -05001935
Al Virobd92d7f2011-03-14 19:54:59 -04001936 if (unlikely(err))
1937 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001938
1939 current->total_link_count = 0;
Al Virobd92d7f2011-03-14 19:54:59 -04001940 err = link_path_walk(name, nd);
1941
1942 if (!err && !(flags & LOOKUP_PARENT)) {
Al Virobd92d7f2011-03-14 19:54:59 -04001943 err = lookup_last(nd, &path);
1944 while (err > 0) {
1945 void *cookie;
1946 struct path link = path;
Kees Cook800179c2012-07-25 17:29:07 -07001947 err = may_follow_link(&link, nd);
1948 if (unlikely(err))
1949 break;
Al Virobd92d7f2011-03-14 19:54:59 -04001950 nd->flags |= LOOKUP_PARENT;
Al Viro574197e2011-03-14 22:20:34 -04001951 err = follow_link(&link, nd, &cookie);
Al Viro6d7b5aa2012-06-10 04:15:17 -04001952 if (err)
1953 break;
1954 err = lookup_last(nd, &path);
Al Viro574197e2011-03-14 22:20:34 -04001955 put_link(nd, &link, cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04001956 }
1957 }
Al Viroee0827c2011-02-21 23:38:09 -05001958
Al Viro9f1fafe2011-03-25 11:00:12 -04001959 if (!err)
1960 err = complete_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04001961
1962 if (!err && nd->flags & LOOKUP_DIRECTORY) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02001963 if (!d_can_lookup(nd->path.dentry)) {
Al Virobd92d7f2011-03-14 19:54:59 -04001964 path_put(&nd->path);
Al Virobd23a532011-03-23 09:56:30 -04001965 err = -ENOTDIR;
Al Virobd92d7f2011-03-14 19:54:59 -04001966 }
1967 }
Al Viro16c2cd72011-02-22 15:50:10 -05001968
Al Viro70e9b352011-03-05 21:12:22 -05001969 if (base)
1970 fput(base);
Al Viroee0827c2011-02-21 23:38:09 -05001971
Al Viro5b6ca022011-03-09 23:04:47 -05001972 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
Al Viro2a737872009-04-07 11:49:53 -04001973 path_put(&nd->root);
1974 nd->root.mnt = NULL;
1975 }
Al Virobd92d7f2011-03-14 19:54:59 -04001976 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001977}
Nick Piggin31e6b012011-01-07 17:49:52 +11001978
Jeff Layton873f1ee2012-10-10 15:25:29 -04001979static int filename_lookup(int dfd, struct filename *name,
1980 unsigned int flags, struct nameidata *nd)
1981{
1982 int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd);
1983 if (unlikely(retval == -ECHILD))
1984 retval = path_lookupat(dfd, name->name, flags, nd);
1985 if (unlikely(retval == -ESTALE))
1986 retval = path_lookupat(dfd, name->name,
1987 flags | LOOKUP_REVAL, nd);
1988
1989 if (likely(!retval))
Jeff Laytonadb5c242012-10-10 16:43:13 -04001990 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
Jeff Layton873f1ee2012-10-10 15:25:29 -04001991 return retval;
1992}
1993
Al Viroee0827c2011-02-21 23:38:09 -05001994static int do_path_lookup(int dfd, const char *name,
1995 unsigned int flags, struct nameidata *nd)
1996{
Jeff Layton873f1ee2012-10-10 15:25:29 -04001997 struct filename filename = { .name = name };
Nick Piggin31e6b012011-01-07 17:49:52 +11001998
Jeff Layton873f1ee2012-10-10 15:25:29 -04001999 return filename_lookup(dfd, &filename, flags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000}
2001
Al Viro79714f72012-06-15 03:01:42 +04002002/* does lookup, returns the object with parent locked */
2003struct dentry *kern_path_locked(const char *name, struct path *path)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002004{
Al Viro79714f72012-06-15 03:01:42 +04002005 struct nameidata nd;
2006 struct dentry *d;
2007 int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd);
2008 if (err)
2009 return ERR_PTR(err);
2010 if (nd.last_type != LAST_NORM) {
2011 path_put(&nd.path);
2012 return ERR_PTR(-EINVAL);
2013 }
2014 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Al Viro1e0ea002012-07-22 23:46:21 +04002015 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
Al Viro79714f72012-06-15 03:01:42 +04002016 if (IS_ERR(d)) {
2017 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2018 path_put(&nd.path);
2019 return d;
2020 }
2021 *path = nd.path;
2022 return d;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002023}
2024
Al Virod1811462008-08-02 00:49:18 -04002025int kern_path(const char *name, unsigned int flags, struct path *path)
2026{
2027 struct nameidata nd;
2028 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
2029 if (!res)
2030 *path = nd.path;
2031 return res;
2032}
Al Viro4d359502014-03-14 12:20:17 -04002033EXPORT_SYMBOL(kern_path);
Al Virod1811462008-08-02 00:49:18 -04002034
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002035/**
2036 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2037 * @dentry: pointer to dentry of the base directory
2038 * @mnt: pointer to vfs mount of the base directory
2039 * @name: pointer to file name
2040 * @flags: lookup flags
Al Viroe0a01242011-06-27 17:00:37 -04002041 * @path: pointer to struct path to fill
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002042 */
2043int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2044 const char *name, unsigned int flags,
Al Viroe0a01242011-06-27 17:00:37 -04002045 struct path *path)
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002046{
Al Viroe0a01242011-06-27 17:00:37 -04002047 struct nameidata nd;
2048 int err;
2049 nd.root.dentry = dentry;
2050 nd.root.mnt = mnt;
2051 BUG_ON(flags & LOOKUP_PARENT);
Al Viro5b6ca022011-03-09 23:04:47 -05002052 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
Al Viroe0a01242011-06-27 17:00:37 -04002053 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
2054 if (!err)
2055 *path = nd.path;
2056 return err;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002057}
Al Viro4d359502014-03-14 12:20:17 -04002058EXPORT_SYMBOL(vfs_path_lookup);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002059
James Morris057f6c02007-04-26 00:12:05 -07002060/*
2061 * Restricted form of lookup. Doesn't follow links, single-component only,
2062 * needs parent already locked. Doesn't follow mounts.
2063 * SMP-safe.
2064 */
Adrian Bunka244e162006-03-31 02:32:11 -08002065static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066{
Al Viro72bd8662012-06-10 17:17:17 -04002067 return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068}
2069
Christoph Hellwigeead1912007-10-16 23:25:38 -07002070/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07002071 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07002072 * @name: pathname component to lookup
2073 * @base: base directory to lookup from
2074 * @len: maximum length @len should be interpreted to
2075 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07002076 * Note that this routine is purely a helper for filesystem usage and should
2077 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07002078 * nameidata argument is passed to the filesystem methods and a filesystem
2079 * using this helper needs to be prepared for that.
2080 */
James Morris057f6c02007-04-26 00:12:05 -07002081struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2082{
James Morris057f6c02007-04-26 00:12:05 -07002083 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05002084 unsigned int c;
Miklos Szeredicda309d2012-03-26 12:54:21 +02002085 int err;
James Morris057f6c02007-04-26 00:12:05 -07002086
David Woodhouse2f9092e2009-04-20 23:18:37 +01002087 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2088
Al Viro6a96ba52011-03-07 23:49:20 -05002089 this.name = name;
2090 this.len = len;
Linus Torvalds0145acc2012-03-02 14:32:59 -08002091 this.hash = full_name_hash(name, len);
Al Viro6a96ba52011-03-07 23:49:20 -05002092 if (!len)
2093 return ERR_PTR(-EACCES);
2094
Al Viro21d8a152012-11-29 22:17:21 -05002095 if (unlikely(name[0] == '.')) {
2096 if (len < 2 || (len == 2 && name[1] == '.'))
2097 return ERR_PTR(-EACCES);
2098 }
2099
Al Viro6a96ba52011-03-07 23:49:20 -05002100 while (len--) {
2101 c = *(const unsigned char *)name++;
2102 if (c == '/' || c == '\0')
2103 return ERR_PTR(-EACCES);
Al Viro6a96ba52011-03-07 23:49:20 -05002104 }
Al Viro5a202bc2011-03-08 14:17:44 -05002105 /*
2106 * See if the low-level filesystem might want
2107 * to use its own hash..
2108 */
2109 if (base->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002110 int err = base->d_op->d_hash(base, &this);
Al Viro5a202bc2011-03-08 14:17:44 -05002111 if (err < 0)
2112 return ERR_PTR(err);
2113 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07002114
Miklos Szeredicda309d2012-03-26 12:54:21 +02002115 err = inode_permission(base->d_inode, MAY_EXEC);
2116 if (err)
2117 return ERR_PTR(err);
2118
Al Viro72bd8662012-06-10 17:17:17 -04002119 return __lookup_hash(&this, base, 0);
James Morris057f6c02007-04-26 00:12:05 -07002120}
Al Viro4d359502014-03-14 12:20:17 -04002121EXPORT_SYMBOL(lookup_one_len);
James Morris057f6c02007-04-26 00:12:05 -07002122
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002123int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2124 struct path *path, int *empty)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125{
Al Viro2d8f3032008-07-22 09:59:21 -04002126 struct nameidata nd;
Jeff Layton91a27b22012-10-10 15:25:28 -04002127 struct filename *tmp = getname_flags(name, flags, empty);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04002130
2131 BUG_ON(flags & LOOKUP_PARENT);
2132
Jeff Layton873f1ee2012-10-10 15:25:29 -04002133 err = filename_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04002135 if (!err)
2136 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 }
2138 return err;
2139}
2140
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002141int user_path_at(int dfd, const char __user *name, unsigned flags,
2142 struct path *path)
2143{
Linus Torvaldsf7493e52012-03-22 16:10:40 -07002144 return user_path_at_empty(dfd, name, flags, path, NULL);
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002145}
Al Viro4d359502014-03-14 12:20:17 -04002146EXPORT_SYMBOL(user_path_at);
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002147
Jeff Layton873f1ee2012-10-10 15:25:29 -04002148/*
2149 * NB: most callers don't do anything directly with the reference to the
2150 * to struct filename, but the nd->last pointer points into the name string
2151 * allocated by getname. So we must hold the reference to it until all
2152 * path-walking is complete.
2153 */
Jeff Layton91a27b22012-10-10 15:25:28 -04002154static struct filename *
Jeff Layton9e790bd2012-12-11 12:10:09 -05002155user_path_parent(int dfd, const char __user *path, struct nameidata *nd,
2156 unsigned int flags)
Al Viro2ad94ae2008-07-21 09:32:51 -04002157{
Jeff Layton91a27b22012-10-10 15:25:28 -04002158 struct filename *s = getname(path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002159 int error;
2160
Jeff Layton9e790bd2012-12-11 12:10:09 -05002161 /* only LOOKUP_REVAL is allowed in extra flags */
2162 flags &= LOOKUP_REVAL;
2163
Al Viro2ad94ae2008-07-21 09:32:51 -04002164 if (IS_ERR(s))
Jeff Layton91a27b22012-10-10 15:25:28 -04002165 return s;
Al Viro2ad94ae2008-07-21 09:32:51 -04002166
Jeff Layton9e790bd2012-12-11 12:10:09 -05002167 error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd);
Jeff Layton91a27b22012-10-10 15:25:28 -04002168 if (error) {
Al Viro2ad94ae2008-07-21 09:32:51 -04002169 putname(s);
Jeff Layton91a27b22012-10-10 15:25:28 -04002170 return ERR_PTR(error);
2171 }
Al Viro2ad94ae2008-07-21 09:32:51 -04002172
Jeff Layton91a27b22012-10-10 15:25:28 -04002173 return s;
Al Viro2ad94ae2008-07-21 09:32:51 -04002174}
2175
Jeff Layton80334262013-07-26 06:23:25 -04002176/**
Al Viro197df042013-09-08 14:03:27 -04002177 * mountpoint_last - look up last component for umount
Jeff Layton80334262013-07-26 06:23:25 -04002178 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2179 * @path: pointer to container for result
2180 *
2181 * This is a special lookup_last function just for umount. In this case, we
2182 * need to resolve the path without doing any revalidation.
2183 *
2184 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2185 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2186 * in almost all cases, this lookup will be served out of the dcache. The only
2187 * cases where it won't are if nd->last refers to a symlink or the path is
2188 * bogus and it doesn't exist.
2189 *
2190 * Returns:
2191 * -error: if there was an error during lookup. This includes -ENOENT if the
2192 * lookup found a negative dentry. The nd->path reference will also be
2193 * put in this case.
2194 *
2195 * 0: if we successfully resolved nd->path and found it to not to be a
2196 * symlink that needs to be followed. "path" will also be populated.
2197 * The nd->path reference will also be put.
2198 *
2199 * 1: if we successfully resolved nd->last and found it to be a symlink
2200 * that needs to be followed. "path" will be populated with the path
2201 * to the link, and nd->path will *not* be put.
2202 */
2203static int
Al Viro197df042013-09-08 14:03:27 -04002204mountpoint_last(struct nameidata *nd, struct path *path)
Jeff Layton80334262013-07-26 06:23:25 -04002205{
2206 int error = 0;
2207 struct dentry *dentry;
2208 struct dentry *dir = nd->path.dentry;
2209
Al Viro35759522013-09-08 13:41:33 -04002210 /* If we're in rcuwalk, drop out of it to handle last component */
2211 if (nd->flags & LOOKUP_RCU) {
2212 if (unlazy_walk(nd, NULL)) {
2213 error = -ECHILD;
2214 goto out;
2215 }
Jeff Layton80334262013-07-26 06:23:25 -04002216 }
2217
2218 nd->flags &= ~LOOKUP_PARENT;
2219
2220 if (unlikely(nd->last_type != LAST_NORM)) {
2221 error = handle_dots(nd, nd->last_type);
Al Viro35759522013-09-08 13:41:33 -04002222 if (error)
2223 goto out;
2224 dentry = dget(nd->path.dentry);
2225 goto done;
Jeff Layton80334262013-07-26 06:23:25 -04002226 }
2227
2228 mutex_lock(&dir->d_inode->i_mutex);
2229 dentry = d_lookup(dir, &nd->last);
2230 if (!dentry) {
2231 /*
2232 * No cached dentry. Mounted dentries are pinned in the cache,
2233 * so that means that this dentry is probably a symlink or the
2234 * path doesn't actually point to a mounted dentry.
2235 */
2236 dentry = d_alloc(dir, &nd->last);
2237 if (!dentry) {
2238 error = -ENOMEM;
Dave Jonesbcceeeb2013-09-10 17:04:25 -04002239 mutex_unlock(&dir->d_inode->i_mutex);
Al Viro35759522013-09-08 13:41:33 -04002240 goto out;
Jeff Layton80334262013-07-26 06:23:25 -04002241 }
Al Viro35759522013-09-08 13:41:33 -04002242 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2243 error = PTR_ERR(dentry);
Dave Jonesbcceeeb2013-09-10 17:04:25 -04002244 if (IS_ERR(dentry)) {
2245 mutex_unlock(&dir->d_inode->i_mutex);
Al Viro35759522013-09-08 13:41:33 -04002246 goto out;
Dave Jonesbcceeeb2013-09-10 17:04:25 -04002247 }
Jeff Layton80334262013-07-26 06:23:25 -04002248 }
2249 mutex_unlock(&dir->d_inode->i_mutex);
2250
Al Viro35759522013-09-08 13:41:33 -04002251done:
2252 if (!dentry->d_inode) {
2253 error = -ENOENT;
2254 dput(dentry);
2255 goto out;
Jeff Layton80334262013-07-26 06:23:25 -04002256 }
Al Viro35759522013-09-08 13:41:33 -04002257 path->dentry = dentry;
2258 path->mnt = mntget(nd->path.mnt);
David Howellsb18825a2013-09-12 19:22:53 +01002259 if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW))
Al Viro35759522013-09-08 13:41:33 -04002260 return 1;
2261 follow_mount(path);
2262 error = 0;
2263out:
Jeff Layton80334262013-07-26 06:23:25 -04002264 terminate_walk(nd);
2265 return error;
2266}
2267
2268/**
Al Viro197df042013-09-08 14:03:27 -04002269 * path_mountpoint - look up a path to be umounted
Jeff Layton80334262013-07-26 06:23:25 -04002270 * @dfd: directory file descriptor to start walk from
2271 * @name: full pathname to walk
Randy Dunlap606d6fe2013-10-19 14:56:55 -07002272 * @path: pointer to container for result
Jeff Layton80334262013-07-26 06:23:25 -04002273 * @flags: lookup flags
Jeff Layton80334262013-07-26 06:23:25 -04002274 *
2275 * Look up the given name, but don't attempt to revalidate the last component.
Randy Dunlap606d6fe2013-10-19 14:56:55 -07002276 * Returns 0 and "path" will be valid on success; Returns error otherwise.
Jeff Layton80334262013-07-26 06:23:25 -04002277 */
2278static int
Al Viro197df042013-09-08 14:03:27 -04002279path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags)
Jeff Layton80334262013-07-26 06:23:25 -04002280{
2281 struct file *base = NULL;
2282 struct nameidata nd;
2283 int err;
2284
2285 err = path_init(dfd, name, flags | LOOKUP_PARENT, &nd, &base);
2286 if (unlikely(err))
2287 return err;
2288
2289 current->total_link_count = 0;
2290 err = link_path_walk(name, &nd);
2291 if (err)
2292 goto out;
2293
Al Viro197df042013-09-08 14:03:27 -04002294 err = mountpoint_last(&nd, path);
Jeff Layton80334262013-07-26 06:23:25 -04002295 while (err > 0) {
2296 void *cookie;
2297 struct path link = *path;
2298 err = may_follow_link(&link, &nd);
2299 if (unlikely(err))
2300 break;
2301 nd.flags |= LOOKUP_PARENT;
2302 err = follow_link(&link, &nd, &cookie);
2303 if (err)
2304 break;
Al Viro197df042013-09-08 14:03:27 -04002305 err = mountpoint_last(&nd, path);
Jeff Layton80334262013-07-26 06:23:25 -04002306 put_link(&nd, &link, cookie);
2307 }
2308out:
2309 if (base)
2310 fput(base);
2311
2312 if (nd.root.mnt && !(nd.flags & LOOKUP_ROOT))
2313 path_put(&nd.root);
2314
2315 return err;
2316}
2317
Al Viro2d864652013-09-08 20:18:44 -04002318static int
2319filename_mountpoint(int dfd, struct filename *s, struct path *path,
2320 unsigned int flags)
2321{
2322 int error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU);
2323 if (unlikely(error == -ECHILD))
2324 error = path_mountpoint(dfd, s->name, path, flags);
2325 if (unlikely(error == -ESTALE))
2326 error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL);
2327 if (likely(!error))
2328 audit_inode(s, path->dentry, 0);
2329 return error;
2330}
2331
Jeff Layton80334262013-07-26 06:23:25 -04002332/**
Al Viro197df042013-09-08 14:03:27 -04002333 * user_path_mountpoint_at - lookup a path from userland in order to umount it
Jeff Layton80334262013-07-26 06:23:25 -04002334 * @dfd: directory file descriptor
2335 * @name: pathname from userland
2336 * @flags: lookup flags
2337 * @path: pointer to container to hold result
2338 *
2339 * A umount is a special case for path walking. We're not actually interested
2340 * in the inode in this situation, and ESTALE errors can be a problem. We
2341 * simply want track down the dentry and vfsmount attached at the mountpoint
2342 * and avoid revalidating the last component.
2343 *
2344 * Returns 0 and populates "path" on success.
2345 */
2346int
Al Viro197df042013-09-08 14:03:27 -04002347user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
Jeff Layton80334262013-07-26 06:23:25 -04002348 struct path *path)
2349{
2350 struct filename *s = getname(name);
2351 int error;
Jeff Layton80334262013-07-26 06:23:25 -04002352 if (IS_ERR(s))
2353 return PTR_ERR(s);
Al Viro2d864652013-09-08 20:18:44 -04002354 error = filename_mountpoint(dfd, s, path, flags);
Jeff Layton80334262013-07-26 06:23:25 -04002355 putname(s);
2356 return error;
2357}
2358
Al Viro2d864652013-09-08 20:18:44 -04002359int
2360kern_path_mountpoint(int dfd, const char *name, struct path *path,
2361 unsigned int flags)
2362{
2363 struct filename s = {.name = name};
2364 return filename_mountpoint(dfd, &s, path, flags);
2365}
2366EXPORT_SYMBOL(kern_path_mountpoint);
2367
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368/*
2369 * It's inline, so penalty for filesystems that don't use sticky bit is
2370 * minimal.
2371 */
2372static inline int check_sticky(struct inode *dir, struct inode *inode)
2373{
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002374 kuid_t fsuid = current_fsuid();
David Howellsda9592e2008-11-14 10:39:05 +11002375
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376 if (!(dir->i_mode & S_ISVTX))
2377 return 0;
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002378 if (uid_eq(inode->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379 return 0;
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002380 if (uid_eq(dir->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381 return 0;
Eric W. Biederman1a48e2a2011-11-14 16:24:06 -08002382 return !inode_capable(inode, CAP_FOWNER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383}
2384
2385/*
2386 * Check whether we can remove a link victim from directory dir, check
2387 * whether the type of victim is right.
2388 * 1. We can't do it if dir is read-only (done in permission())
2389 * 2. We should have write and exec permissions on dir
2390 * 3. We can't remove anything from append-only dir
2391 * 4. We can't do anything with immutable dir (done in permission())
2392 * 5. If the sticky bit on dir is set we should either
2393 * a. be owner of dir, or
2394 * b. be owner of victim, or
2395 * c. have CAP_FOWNER capability
2396 * 6. If the victim is append-only or immutable we can't do antyhing with
2397 * links pointing to it.
2398 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2399 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2400 * 9. We can't remove a root or mountpoint.
2401 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2402 * nfs_async_unlink().
2403 */
David Howellsb18825a2013-09-12 19:22:53 +01002404static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405{
David Howellsb18825a2013-09-12 19:22:53 +01002406 struct inode *inode = victim->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407 int error;
2408
David Howellsb18825a2013-09-12 19:22:53 +01002409 if (d_is_negative(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410 return -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01002411 BUG_ON(!inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412
2413 BUG_ON(victim->d_parent->d_inode != dir);
Jeff Layton4fa6b5e2012-10-10 15:25:25 -04002414 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415
Al Virof419a2e2008-07-22 00:07:17 -04002416 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417 if (error)
2418 return error;
2419 if (IS_APPEND(dir))
2420 return -EPERM;
David Howellsb18825a2013-09-12 19:22:53 +01002421
2422 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2423 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 return -EPERM;
2425 if (isdir) {
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002426 if (!d_is_dir(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 return -ENOTDIR;
2428 if (IS_ROOT(victim))
2429 return -EBUSY;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02002430 } else if (d_is_dir(victim))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431 return -EISDIR;
2432 if (IS_DEADDIR(dir))
2433 return -ENOENT;
2434 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2435 return -EBUSY;
2436 return 0;
2437}
2438
2439/* Check whether we can create an object with dentry child in directory
2440 * dir.
2441 * 1. We can't do it if child already exists (open has special treatment for
2442 * this case, but since we are inlined it's OK)
2443 * 2. We can't do it if dir is read-only (done in permission())
2444 * 3. We should have write and exec permissions on dir
2445 * 4. We can't do it if dir is immutable (done in permission())
2446 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02002447static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448{
Jeff Layton14e972b2013-05-08 10:25:58 -04002449 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450 if (child->d_inode)
2451 return -EEXIST;
2452 if (IS_DEADDIR(dir))
2453 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04002454 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455}
2456
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457/*
2458 * p1 and p2 should be directories on the same fs.
2459 */
2460struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2461{
2462 struct dentry *p;
2463
2464 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07002465 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466 return NULL;
2467 }
2468
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002469 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002471 p = d_ancestor(p2, p1);
2472 if (p) {
2473 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2474 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2475 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476 }
2477
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002478 p = d_ancestor(p1, p2);
2479 if (p) {
2480 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2481 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2482 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483 }
2484
Ingo Molnarf2eace22006-07-03 00:25:05 -07002485 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2486 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487 return NULL;
2488}
Al Viro4d359502014-03-14 12:20:17 -04002489EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490
2491void unlock_rename(struct dentry *p1, struct dentry *p2)
2492{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002493 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002495 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002496 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497 }
2498}
Al Viro4d359502014-03-14 12:20:17 -04002499EXPORT_SYMBOL(unlock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002500
Al Viro4acdaf22011-07-26 01:42:34 -04002501int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viro312b63f2012-06-10 18:09:36 -04002502 bool want_excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002504 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505 if (error)
2506 return error;
2507
Al Viroacfa4382008-12-04 10:06:33 -05002508 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509 return -EACCES; /* shouldn't it be ENOSYS? */
2510 mode &= S_IALLUGO;
2511 mode |= S_IFREG;
2512 error = security_inode_create(dir, dentry, mode);
2513 if (error)
2514 return error;
Al Viro312b63f2012-06-10 18:09:36 -04002515 error = dir->i_op->create(dir, dentry, mode, want_excl);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002516 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002517 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518 return error;
2519}
Al Viro4d359502014-03-14 12:20:17 -04002520EXPORT_SYMBOL(vfs_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521
Al Viro73d049a2011-03-11 12:08:24 -05002522static int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002524 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525 struct inode *inode = dentry->d_inode;
2526 int error;
2527
Al Virobcda7652011-03-13 16:42:14 -04002528 /* O_PATH? */
2529 if (!acc_mode)
2530 return 0;
2531
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532 if (!inode)
2533 return -ENOENT;
2534
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002535 switch (inode->i_mode & S_IFMT) {
2536 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002538 case S_IFDIR:
2539 if (acc_mode & MAY_WRITE)
2540 return -EISDIR;
2541 break;
2542 case S_IFBLK:
2543 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002544 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002546 /*FALLTHRU*/
2547 case S_IFIFO:
2548 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002550 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002551 }
Dave Hansenb41572e2007-10-16 23:31:14 -07002552
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002553 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07002554 if (error)
2555 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05002556
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557 /*
2558 * An append-only file must be opened in append mode for writing.
2559 */
2560 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05002561 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05002562 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05002564 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565 }
2566
2567 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002568 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05002569 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570
J. Bruce Fieldsf3c7691e2011-09-21 10:58:13 -04002571 return 0;
Al Viro7715b522009-12-16 03:54:00 -05002572}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573
Jeff Laytone1181ee2010-12-07 16:19:50 -05002574static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002575{
Jeff Laytone1181ee2010-12-07 16:19:50 -05002576 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002577 struct inode *inode = path->dentry->d_inode;
2578 int error = get_write_access(inode);
2579 if (error)
2580 return error;
2581 /*
2582 * Refuse to truncate files with mandatory locks held on them.
2583 */
Jeff Laytond7a06982014-03-10 09:54:15 -04002584 error = locks_verify_locked(filp);
Al Viro7715b522009-12-16 03:54:00 -05002585 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002586 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002587 if (!error) {
2588 error = do_truncate(path->dentry, 0,
2589 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002590 filp);
Al Viro7715b522009-12-16 03:54:00 -05002591 }
2592 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002593 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594}
2595
Dave Hansend57999e2008-02-15 14:37:27 -08002596static inline int open_to_namei_flags(int flag)
2597{
Al Viro8a5e9292011-06-25 19:15:54 -04002598 if ((flag & O_ACCMODE) == 3)
2599 flag--;
Dave Hansend57999e2008-02-15 14:37:27 -08002600 return flag;
2601}
2602
Miklos Szeredid18e9002012-06-05 15:10:17 +02002603static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2604{
2605 int error = security_path_mknod(dir, dentry, mode, 0);
2606 if (error)
2607 return error;
2608
2609 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2610 if (error)
2611 return error;
2612
2613 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2614}
2615
David Howells1acf0af2012-06-14 16:13:46 +01002616/*
2617 * Attempt to atomically look up, create and open a file from a negative
2618 * dentry.
2619 *
2620 * Returns 0 if successful. The file will have been created and attached to
2621 * @file by the filesystem calling finish_open().
2622 *
2623 * Returns 1 if the file was looked up only or didn't need creating. The
2624 * caller will need to perform the open themselves. @path will have been
2625 * updated to point to the new dentry. This may be negative.
2626 *
2627 * Returns an error code otherwise.
2628 */
Al Viro2675a4e2012-06-22 12:41:10 +04002629static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2630 struct path *path, struct file *file,
2631 const struct open_flags *op,
Al Viro64894cf2012-07-31 00:53:35 +04002632 bool got_write, bool need_lookup,
Al Viro2675a4e2012-06-22 12:41:10 +04002633 int *opened)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002634{
2635 struct inode *dir = nd->path.dentry->d_inode;
2636 unsigned open_flag = open_to_namei_flags(op->open_flag);
2637 umode_t mode;
2638 int error;
2639 int acc_mode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002640 int create_error = 0;
2641 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
Miklos Szeredi116cc022013-09-16 14:52:05 +02002642 bool excl;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002643
2644 BUG_ON(dentry->d_inode);
2645
2646 /* Don't create child dentry for a dead directory. */
2647 if (unlikely(IS_DEADDIR(dir))) {
Al Viro2675a4e2012-06-22 12:41:10 +04002648 error = -ENOENT;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002649 goto out;
2650 }
2651
Miklos Szeredi62b259d2012-08-15 13:01:24 +02002652 mode = op->mode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002653 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2654 mode &= ~current_umask();
2655
Miklos Szeredi116cc022013-09-16 14:52:05 +02002656 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2657 if (excl)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002658 open_flag &= ~O_TRUNC;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002659
2660 /*
2661 * Checking write permission is tricky, bacuse we don't know if we are
2662 * going to actually need it: O_CREAT opens should work as long as the
2663 * file exists. But checking existence breaks atomicity. The trick is
2664 * to check access and if not granted clear O_CREAT from the flags.
2665 *
2666 * Another problem is returing the "right" error value (e.g. for an
2667 * O_EXCL open we want to return EEXIST not EROFS).
2668 */
Al Viro64894cf2012-07-31 00:53:35 +04002669 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2670 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2671 if (!(open_flag & O_CREAT)) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002672 /*
2673 * No O_CREATE -> atomicity not a requirement -> fall
2674 * back to lookup + open
2675 */
2676 goto no_open;
2677 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2678 /* Fall back and fail with the right error */
Al Viro64894cf2012-07-31 00:53:35 +04002679 create_error = -EROFS;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002680 goto no_open;
2681 } else {
2682 /* No side effects, safe to clear O_CREAT */
Al Viro64894cf2012-07-31 00:53:35 +04002683 create_error = -EROFS;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002684 open_flag &= ~O_CREAT;
2685 }
2686 }
2687
2688 if (open_flag & O_CREAT) {
Miklos Szeredi38227f72012-08-15 13:01:24 +02002689 error = may_o_create(&nd->path, dentry, mode);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002690 if (error) {
2691 create_error = error;
2692 if (open_flag & O_EXCL)
2693 goto no_open;
2694 open_flag &= ~O_CREAT;
2695 }
2696 }
2697
2698 if (nd->flags & LOOKUP_DIRECTORY)
2699 open_flag |= O_DIRECTORY;
2700
Al Viro30d90492012-06-22 12:40:19 +04002701 file->f_path.dentry = DENTRY_NOT_SET;
2702 file->f_path.mnt = nd->path.mnt;
2703 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
Al Viro47237682012-06-10 05:01:45 -04002704 opened);
Al Virod9585272012-06-22 12:39:14 +04002705 if (error < 0) {
Al Virod9585272012-06-22 12:39:14 +04002706 if (create_error && error == -ENOENT)
2707 error = create_error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002708 goto out;
2709 }
2710
Al Virod9585272012-06-22 12:39:14 +04002711 if (error) { /* returned 1, that is */
Al Viro30d90492012-06-22 12:40:19 +04002712 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
Al Viro2675a4e2012-06-22 12:41:10 +04002713 error = -EIO;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002714 goto out;
2715 }
Al Viro30d90492012-06-22 12:40:19 +04002716 if (file->f_path.dentry) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002717 dput(dentry);
Al Viro30d90492012-06-22 12:40:19 +04002718 dentry = file->f_path.dentry;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002719 }
Al Viro03da6332013-09-16 19:22:33 -04002720 if (*opened & FILE_CREATED)
2721 fsnotify_create(dir, dentry);
2722 if (!dentry->d_inode) {
2723 WARN_ON(*opened & FILE_CREATED);
2724 if (create_error) {
2725 error = create_error;
2726 goto out;
2727 }
2728 } else {
2729 if (excl && !(*opened & FILE_CREATED)) {
2730 error = -EEXIST;
2731 goto out;
2732 }
Sage Weil62b2ce92012-08-15 13:30:12 -07002733 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02002734 goto looked_up;
2735 }
2736
2737 /*
2738 * We didn't have the inode before the open, so check open permission
2739 * here.
2740 */
Al Viro03da6332013-09-16 19:22:33 -04002741 acc_mode = op->acc_mode;
2742 if (*opened & FILE_CREATED) {
2743 WARN_ON(!(open_flag & O_CREAT));
2744 fsnotify_create(dir, dentry);
2745 acc_mode = MAY_OPEN;
2746 }
Al Viro2675a4e2012-06-22 12:41:10 +04002747 error = may_open(&file->f_path, acc_mode, open_flag);
2748 if (error)
2749 fput(file);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002750
2751out:
2752 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04002753 return error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002754
Miklos Szeredid18e9002012-06-05 15:10:17 +02002755no_open:
2756 if (need_lookup) {
Al Viro72bd8662012-06-10 17:17:17 -04002757 dentry = lookup_real(dir, dentry, nd->flags);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002758 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002759 return PTR_ERR(dentry);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002760
2761 if (create_error) {
2762 int open_flag = op->open_flag;
2763
Al Viro2675a4e2012-06-22 12:41:10 +04002764 error = create_error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002765 if ((open_flag & O_EXCL)) {
2766 if (!dentry->d_inode)
2767 goto out;
2768 } else if (!dentry->d_inode) {
2769 goto out;
2770 } else if ((open_flag & O_TRUNC) &&
2771 S_ISREG(dentry->d_inode->i_mode)) {
2772 goto out;
2773 }
2774 /* will fail later, go on to get the right error */
2775 }
2776 }
2777looked_up:
2778 path->dentry = dentry;
2779 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04002780 return 1;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002781}
2782
Nick Piggin31e6b012011-01-07 17:49:52 +11002783/*
David Howells1acf0af2012-06-14 16:13:46 +01002784 * Look up and maybe create and open the last component.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002785 *
2786 * Must be called with i_mutex held on parent.
2787 *
David Howells1acf0af2012-06-14 16:13:46 +01002788 * Returns 0 if the file was successfully atomically created (if necessary) and
2789 * opened. In this case the file will be returned attached to @file.
2790 *
2791 * Returns 1 if the file was not completely opened at this time, though lookups
2792 * and creations will have been performed and the dentry returned in @path will
2793 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2794 * specified then a negative dentry may be returned.
2795 *
2796 * An error code is returned otherwise.
2797 *
2798 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2799 * cleared otherwise prior to returning.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002800 */
Al Viro2675a4e2012-06-22 12:41:10 +04002801static int lookup_open(struct nameidata *nd, struct path *path,
2802 struct file *file,
2803 const struct open_flags *op,
Al Viro64894cf2012-07-31 00:53:35 +04002804 bool got_write, int *opened)
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002805{
2806 struct dentry *dir = nd->path.dentry;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002807 struct inode *dir_inode = dir->d_inode;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002808 struct dentry *dentry;
2809 int error;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002810 bool need_lookup;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002811
Al Viro47237682012-06-10 05:01:45 -04002812 *opened &= ~FILE_CREATED;
Al Viro201f9562012-06-22 12:42:10 +04002813 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002814 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002815 return PTR_ERR(dentry);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002816
Miklos Szeredid18e9002012-06-05 15:10:17 +02002817 /* Cached positive dentry: will open in f_op->open */
2818 if (!need_lookup && dentry->d_inode)
2819 goto out_no_open;
2820
2821 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
Al Viro64894cf2012-07-31 00:53:35 +04002822 return atomic_open(nd, dentry, path, file, op, got_write,
Al Viro47237682012-06-10 05:01:45 -04002823 need_lookup, opened);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002824 }
2825
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002826 if (need_lookup) {
2827 BUG_ON(dentry->d_inode);
2828
Al Viro72bd8662012-06-10 17:17:17 -04002829 dentry = lookup_real(dir_inode, dentry, nd->flags);
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002830 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002831 return PTR_ERR(dentry);
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002832 }
2833
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002834 /* Negative dentry, just create the file */
2835 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2836 umode_t mode = op->mode;
2837 if (!IS_POSIXACL(dir->d_inode))
2838 mode &= ~current_umask();
2839 /*
2840 * This write is needed to ensure that a
2841 * rw->ro transition does not occur between
2842 * the time when the file is created and when
2843 * a permanent write count is taken through
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002844 * the 'struct file' in finish_open().
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002845 */
Al Viro64894cf2012-07-31 00:53:35 +04002846 if (!got_write) {
2847 error = -EROFS;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002848 goto out_dput;
Al Viro64894cf2012-07-31 00:53:35 +04002849 }
Al Viro47237682012-06-10 05:01:45 -04002850 *opened |= FILE_CREATED;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002851 error = security_path_mknod(&nd->path, dentry, mode, 0);
2852 if (error)
2853 goto out_dput;
Al Viro312b63f2012-06-10 18:09:36 -04002854 error = vfs_create(dir->d_inode, dentry, mode,
2855 nd->flags & LOOKUP_EXCL);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002856 if (error)
2857 goto out_dput;
2858 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02002859out_no_open:
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002860 path->dentry = dentry;
2861 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04002862 return 1;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002863
2864out_dput:
2865 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04002866 return error;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002867}
2868
2869/*
Al Virofe2d35f2011-03-05 22:58:25 -05002870 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11002871 */
Al Viro2675a4e2012-06-22 12:41:10 +04002872static int do_last(struct nameidata *nd, struct path *path,
2873 struct file *file, const struct open_flags *op,
Jeff Layton669abf42012-10-10 16:43:10 -04002874 int *opened, struct filename *name)
Al Virofb1cc552009-12-24 01:58:28 -05002875{
Al Viroa1e28032009-12-24 02:12:06 -05002876 struct dentry *dir = nd->path.dentry;
Al Viroca344a892011-03-09 00:36:45 -05002877 int open_flag = op->open_flag;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002878 bool will_truncate = (open_flag & O_TRUNC) != 0;
Al Viro64894cf2012-07-31 00:53:35 +04002879 bool got_write = false;
Al Virobcda7652011-03-13 16:42:14 -04002880 int acc_mode = op->acc_mode;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02002881 struct inode *inode;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002882 bool symlink_ok = false;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002883 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2884 bool retried = false;
Al Viro16c2cd72011-02-22 15:50:10 -05002885 int error;
Al Virofb1cc552009-12-24 01:58:28 -05002886
Al Viroc3e380b2011-02-23 13:39:45 -05002887 nd->flags &= ~LOOKUP_PARENT;
2888 nd->flags |= op->intent;
2889
Al Virobc77daa2013-06-06 09:12:33 -04002890 if (nd->last_type != LAST_NORM) {
Al Virofe2d35f2011-03-05 22:58:25 -05002891 error = handle_dots(nd, nd->last_type);
2892 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002893 return error;
Miklos Szeredie83db162012-06-05 15:10:29 +02002894 goto finish_open;
Al Viro1f36f772009-12-26 10:56:19 -05002895 }
Al Viro67ee3ad2009-12-26 07:01:01 -05002896
Al Viroca344a892011-03-09 00:36:45 -05002897 if (!(open_flag & O_CREAT)) {
Al Virofe2d35f2011-03-05 22:58:25 -05002898 if (nd->last.name[nd->last.len])
2899 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
Al Virobcda7652011-03-13 16:42:14 -04002900 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002901 symlink_ok = true;
Al Virofe2d35f2011-03-05 22:58:25 -05002902 /* we _can_ be in RCU mode here */
Al Viroe97cdc82013-01-24 18:16:00 -05002903 error = lookup_fast(nd, path, &inode);
Miklos Szeredi71574862012-06-05 15:10:14 +02002904 if (likely(!error))
2905 goto finish_lookup;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02002906
Miklos Szeredi71574862012-06-05 15:10:14 +02002907 if (error < 0)
Al Viro2675a4e2012-06-22 12:41:10 +04002908 goto out;
Miklos Szeredi37d7fff2012-06-05 15:10:12 +02002909
Miklos Szeredi71574862012-06-05 15:10:14 +02002910 BUG_ON(nd->inode != dir->d_inode);
Miklos Szeredib6183df2012-06-05 15:10:13 +02002911 } else {
2912 /* create side of things */
2913 /*
2914 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2915 * has been cleared when we got to the last component we are
2916 * about to look up
2917 */
2918 error = complete_walk(nd);
2919 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002920 return error;
Miklos Szeredib6183df2012-06-05 15:10:13 +02002921
Jeff Layton33e22082013-04-12 15:16:32 -04002922 audit_inode(name, dir, LOOKUP_PARENT);
Miklos Szeredib6183df2012-06-05 15:10:13 +02002923 error = -EISDIR;
2924 /* trailing slashes? */
2925 if (nd->last.name[nd->last.len])
Al Viro2675a4e2012-06-22 12:41:10 +04002926 goto out;
Al Virofe2d35f2011-03-05 22:58:25 -05002927 }
2928
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002929retry_lookup:
Al Viro64894cf2012-07-31 00:53:35 +04002930 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2931 error = mnt_want_write(nd->path.mnt);
2932 if (!error)
2933 got_write = true;
2934 /*
2935 * do _not_ fail yet - we might not need that or fail with
2936 * a different error; let lookup_open() decide; we'll be
2937 * dropping this one anyway.
2938 */
2939 }
Al Viroa1e28032009-12-24 02:12:06 -05002940 mutex_lock(&dir->d_inode->i_mutex);
Al Viro64894cf2012-07-31 00:53:35 +04002941 error = lookup_open(nd, path, file, op, got_write, opened);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002942 mutex_unlock(&dir->d_inode->i_mutex);
Al Viroa1e28032009-12-24 02:12:06 -05002943
Al Viro2675a4e2012-06-22 12:41:10 +04002944 if (error <= 0) {
2945 if (error)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002946 goto out;
2947
Al Viro47237682012-06-10 05:01:45 -04002948 if ((*opened & FILE_CREATED) ||
Al Viro496ad9a2013-01-23 17:07:38 -05002949 !S_ISREG(file_inode(file)->i_mode))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002950 will_truncate = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002951
Jeff Laytonadb5c242012-10-10 16:43:13 -04002952 audit_inode(name, file->f_path.dentry, 0);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002953 goto opened;
2954 }
Al Virofb1cc552009-12-24 01:58:28 -05002955
Al Viro47237682012-06-10 05:01:45 -04002956 if (*opened & FILE_CREATED) {
Al Viro9b44f1b2011-03-09 00:17:27 -05002957 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05002958 open_flag &= ~O_TRUNC;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002959 will_truncate = false;
Al Virobcda7652011-03-13 16:42:14 -04002960 acc_mode = MAY_OPEN;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002961 path_to_nameidata(path, nd);
Miklos Szeredie83db162012-06-05 15:10:29 +02002962 goto finish_open_created;
Al Virofb1cc552009-12-24 01:58:28 -05002963 }
2964
2965 /*
Jeff Layton3134f372012-07-25 10:19:47 -04002966 * create/update audit record if it already exists.
Al Virofb1cc552009-12-24 01:58:28 -05002967 */
David Howellsb18825a2013-09-12 19:22:53 +01002968 if (d_is_positive(path->dentry))
Jeff Laytonadb5c242012-10-10 16:43:13 -04002969 audit_inode(name, path->dentry, 0);
Al Virofb1cc552009-12-24 01:58:28 -05002970
Miklos Szeredid18e9002012-06-05 15:10:17 +02002971 /*
2972 * If atomic_open() acquired write access it is dropped now due to
2973 * possible mount and symlink following (this might be optimized away if
2974 * necessary...)
2975 */
Al Viro64894cf2012-07-31 00:53:35 +04002976 if (got_write) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002977 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04002978 got_write = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002979 }
2980
Al Virofb1cc552009-12-24 01:58:28 -05002981 error = -EEXIST;
Al Virof8310c52012-07-30 11:50:30 +04002982 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
Al Virofb1cc552009-12-24 01:58:28 -05002983 goto exit_dput;
2984
David Howells9875cf82011-01-14 18:45:21 +00002985 error = follow_managed(path, nd->flags);
2986 if (error < 0)
2987 goto exit_dput;
Al Virofb1cc552009-12-24 01:58:28 -05002988
Al Viroa3fbbde2011-11-07 21:21:26 +00002989 if (error)
2990 nd->flags |= LOOKUP_JUMPED;
2991
Miklos Szeredidecf3402012-05-21 17:30:08 +02002992 BUG_ON(nd->flags & LOOKUP_RCU);
2993 inode = path->dentry->d_inode;
Miklos Szeredi5f5daac2012-05-21 17:30:14 +02002994finish_lookup:
2995 /* we _can_ be in RCU mode here */
Al Virofb1cc552009-12-24 01:58:28 -05002996 error = -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01002997 if (d_is_negative(path->dentry)) {
Miklos Szeredi54c33e72012-05-21 17:30:10 +02002998 path_to_nameidata(path, nd);
Al Viro2675a4e2012-06-22 12:41:10 +04002999 goto out;
Miklos Szeredi54c33e72012-05-21 17:30:10 +02003000 }
Al Viro9e67f362009-12-26 07:04:50 -05003001
David Howellsb18825a2013-09-12 19:22:53 +01003002 if (should_follow_link(path->dentry, !symlink_ok)) {
Miklos Szeredid45ea862012-05-21 17:30:09 +02003003 if (nd->flags & LOOKUP_RCU) {
3004 if (unlikely(unlazy_walk(nd, path->dentry))) {
3005 error = -ECHILD;
Al Viro2675a4e2012-06-22 12:41:10 +04003006 goto out;
Miklos Szeredid45ea862012-05-21 17:30:09 +02003007 }
3008 }
3009 BUG_ON(inode != path->dentry->d_inode);
Al Viro2675a4e2012-06-22 12:41:10 +04003010 return 1;
Miklos Szeredid45ea862012-05-21 17:30:09 +02003011 }
Al Virofb1cc552009-12-24 01:58:28 -05003012
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003013 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
3014 path_to_nameidata(path, nd);
3015 } else {
3016 save_parent.dentry = nd->path.dentry;
3017 save_parent.mnt = mntget(path->mnt);
3018 nd->path.dentry = path->dentry;
3019
3020 }
Miklos Szeredidecf3402012-05-21 17:30:08 +02003021 nd->inode = inode;
Al Viroa3fbbde2011-11-07 21:21:26 +00003022 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
Al Virobc77daa2013-06-06 09:12:33 -04003023finish_open:
Al Viroa3fbbde2011-11-07 21:21:26 +00003024 error = complete_walk(nd);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003025 if (error) {
3026 path_put(&save_parent);
Al Viro2675a4e2012-06-22 12:41:10 +04003027 return error;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003028 }
Al Virobc77daa2013-06-06 09:12:33 -04003029 audit_inode(name, nd->path.dentry, 0);
Al Virofb1cc552009-12-24 01:58:28 -05003030 error = -EISDIR;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003031 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04003032 goto out;
Miklos Szerediaf2f5542012-05-21 17:30:11 +02003033 error = -ENOTDIR;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003034 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04003035 goto out;
Al Viro6c0d46c2011-03-09 00:59:59 -05003036 if (!S_ISREG(nd->inode->i_mode))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02003037 will_truncate = false;
Al Viro6c0d46c2011-03-09 00:59:59 -05003038
Al Viro0f9d1a12011-03-09 00:13:14 -05003039 if (will_truncate) {
3040 error = mnt_want_write(nd->path.mnt);
3041 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003042 goto out;
Al Viro64894cf2012-07-31 00:53:35 +04003043 got_write = true;
Al Viro0f9d1a12011-03-09 00:13:14 -05003044 }
Miklos Szeredie83db162012-06-05 15:10:29 +02003045finish_open_created:
Al Virobcda7652011-03-13 16:42:14 -04003046 error = may_open(&nd->path, acc_mode, open_flag);
Al Viroca344a892011-03-09 00:36:45 -05003047 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04003048 goto out;
Al Viro30d90492012-06-22 12:40:19 +04003049 file->f_path.mnt = nd->path.mnt;
3050 error = finish_open(file, nd->path.dentry, NULL, opened);
3051 if (error) {
Al Viro30d90492012-06-22 12:40:19 +04003052 if (error == -EOPENSTALE)
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003053 goto stale_open;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003054 goto out;
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003055 }
Miklos Szeredia8277b92012-06-05 15:10:32 +02003056opened:
Al Viro2675a4e2012-06-22 12:41:10 +04003057 error = open_check_o_direct(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003058 if (error)
3059 goto exit_fput;
Al Viro2675a4e2012-06-22 12:41:10 +04003060 error = ima_file_check(file, op->acc_mode);
Miklos Szerediaa4caad2012-06-05 15:10:28 +02003061 if (error)
3062 goto exit_fput;
3063
3064 if (will_truncate) {
Al Viro2675a4e2012-06-22 12:41:10 +04003065 error = handle_truncate(file);
Miklos Szerediaa4caad2012-06-05 15:10:28 +02003066 if (error)
3067 goto exit_fput;
Al Viro0f9d1a12011-03-09 00:13:14 -05003068 }
Al Viroca344a892011-03-09 00:36:45 -05003069out:
Al Viro64894cf2012-07-31 00:53:35 +04003070 if (got_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05003071 mnt_drop_write(nd->path.mnt);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02003072 path_put(&save_parent);
Miklos Szeredie276ae62012-05-21 17:30:06 +02003073 terminate_walk(nd);
Al Viro2675a4e2012-06-22 12:41:10 +04003074 return error;
Al Virofb1cc552009-12-24 01:58:28 -05003075
Al Virofb1cc552009-12-24 01:58:28 -05003076exit_dput:
3077 path_put_conditional(path, nd);
Al Viroca344a892011-03-09 00:36:45 -05003078 goto out;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003079exit_fput:
Al Viro2675a4e2012-06-22 12:41:10 +04003080 fput(file);
3081 goto out;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003082
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003083stale_open:
3084 /* If no saved parent or already retried then can't retry */
3085 if (!save_parent.dentry || retried)
3086 goto out;
3087
3088 BUG_ON(save_parent.dentry != dir);
3089 path_put(&nd->path);
3090 nd->path = save_parent;
3091 nd->inode = dir->d_inode;
3092 save_parent.mnt = NULL;
3093 save_parent.dentry = NULL;
Al Viro64894cf2012-07-31 00:53:35 +04003094 if (got_write) {
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003095 mnt_drop_write(nd->path.mnt);
Al Viro64894cf2012-07-31 00:53:35 +04003096 got_write = false;
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02003097 }
3098 retried = true;
3099 goto retry_lookup;
Al Virofb1cc552009-12-24 01:58:28 -05003100}
3101
Al Viro60545d02013-06-07 01:20:27 -04003102static int do_tmpfile(int dfd, struct filename *pathname,
3103 struct nameidata *nd, int flags,
3104 const struct open_flags *op,
3105 struct file *file, int *opened)
3106{
3107 static const struct qstr name = QSTR_INIT("/", 1);
3108 struct dentry *dentry, *child;
3109 struct inode *dir;
3110 int error = path_lookupat(dfd, pathname->name,
3111 flags | LOOKUP_DIRECTORY, nd);
3112 if (unlikely(error))
3113 return error;
3114 error = mnt_want_write(nd->path.mnt);
3115 if (unlikely(error))
3116 goto out;
3117 /* we want directory to be writable */
3118 error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3119 if (error)
3120 goto out2;
3121 dentry = nd->path.dentry;
3122 dir = dentry->d_inode;
3123 if (!dir->i_op->tmpfile) {
3124 error = -EOPNOTSUPP;
3125 goto out2;
3126 }
3127 child = d_alloc(dentry, &name);
3128 if (unlikely(!child)) {
3129 error = -ENOMEM;
3130 goto out2;
3131 }
3132 nd->flags &= ~LOOKUP_DIRECTORY;
3133 nd->flags |= op->intent;
3134 dput(nd->path.dentry);
3135 nd->path.dentry = child;
3136 error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3137 if (error)
3138 goto out2;
3139 audit_inode(pathname, nd->path.dentry, 0);
3140 error = may_open(&nd->path, op->acc_mode, op->open_flag);
3141 if (error)
3142 goto out2;
3143 file->f_path.mnt = nd->path.mnt;
3144 error = finish_open(file, nd->path.dentry, NULL, opened);
3145 if (error)
3146 goto out2;
3147 error = open_check_o_direct(file);
Al Virof4e0c302013-06-11 08:34:36 +04003148 if (error) {
Al Viro60545d02013-06-07 01:20:27 -04003149 fput(file);
Al Virof4e0c302013-06-11 08:34:36 +04003150 } else if (!(op->open_flag & O_EXCL)) {
3151 struct inode *inode = file_inode(file);
3152 spin_lock(&inode->i_lock);
3153 inode->i_state |= I_LINKABLE;
3154 spin_unlock(&inode->i_lock);
3155 }
Al Viro60545d02013-06-07 01:20:27 -04003156out2:
3157 mnt_drop_write(nd->path.mnt);
3158out:
3159 path_put(&nd->path);
3160 return error;
3161}
3162
Jeff Layton669abf42012-10-10 16:43:10 -04003163static struct file *path_openat(int dfd, struct filename *pathname,
Al Viro73d049a2011-03-11 12:08:24 -05003164 struct nameidata *nd, const struct open_flags *op, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003165{
Al Virofe2d35f2011-03-05 22:58:25 -05003166 struct file *base = NULL;
Al Viro30d90492012-06-22 12:40:19 +04003167 struct file *file;
Al Viro9850c052010-01-13 15:01:15 -05003168 struct path path;
Al Viro47237682012-06-10 05:01:45 -04003169 int opened = 0;
Al Viro13aab422011-02-23 17:54:08 -05003170 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11003171
Al Viro30d90492012-06-22 12:40:19 +04003172 file = get_empty_filp();
Al Viro1afc99b2013-02-14 20:41:04 -05003173 if (IS_ERR(file))
3174 return file;
Nick Piggin31e6b012011-01-07 17:49:52 +11003175
Al Viro30d90492012-06-22 12:40:19 +04003176 file->f_flags = op->open_flag;
Nick Piggin31e6b012011-01-07 17:49:52 +11003177
Al Virobb458c62013-07-13 13:26:37 +04003178 if (unlikely(file->f_flags & __O_TMPFILE)) {
Al Viro60545d02013-06-07 01:20:27 -04003179 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3180 goto out;
3181 }
3182
Jeff Layton669abf42012-10-10 16:43:10 -04003183 error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base);
Nick Piggin31e6b012011-01-07 17:49:52 +11003184 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04003185 goto out;
Nick Piggin31e6b012011-01-07 17:49:52 +11003186
Al Virofe2d35f2011-03-05 22:58:25 -05003187 current->total_link_count = 0;
Jeff Layton669abf42012-10-10 16:43:10 -04003188 error = link_path_walk(pathname->name, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11003189 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04003190 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003191
Al Viro2675a4e2012-06-22 12:41:10 +04003192 error = do_last(nd, &path, file, op, &opened, pathname);
3193 while (unlikely(error > 0)) { /* trailing symlink */
Nick Piggin7b9337a2011-01-14 08:42:43 +00003194 struct path link = path;
Al Virodef4af32009-12-26 08:37:05 -05003195 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04003196 if (!(nd->flags & LOOKUP_FOLLOW)) {
Al Viro73d049a2011-03-11 12:08:24 -05003197 path_put_conditional(&path, nd);
3198 path_put(&nd->path);
Al Viro2675a4e2012-06-22 12:41:10 +04003199 error = -ELOOP;
Al Viro40b39132011-03-09 16:22:18 -05003200 break;
3201 }
Kees Cook800179c2012-07-25 17:29:07 -07003202 error = may_follow_link(&link, nd);
3203 if (unlikely(error))
3204 break;
Al Viro73d049a2011-03-11 12:08:24 -05003205 nd->flags |= LOOKUP_PARENT;
3206 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
Al Viro574197e2011-03-14 22:20:34 -04003207 error = follow_link(&link, nd, &cookie);
Al Viroc3e380b2011-02-23 13:39:45 -05003208 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04003209 break;
3210 error = do_last(nd, &path, file, op, &opened, pathname);
Al Viro574197e2011-03-14 22:20:34 -04003211 put_link(nd, &link, cookie);
Al Viro806b6812009-12-26 07:16:40 -05003212 }
Al Viro10fa8e62009-12-26 07:09:49 -05003213out:
Al Viro73d049a2011-03-11 12:08:24 -05003214 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
3215 path_put(&nd->root);
Al Virofe2d35f2011-03-05 22:58:25 -05003216 if (base)
3217 fput(base);
Al Viro2675a4e2012-06-22 12:41:10 +04003218 if (!(opened & FILE_OPENED)) {
3219 BUG_ON(!error);
Al Viro30d90492012-06-22 12:40:19 +04003220 put_filp(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02003221 }
Al Viro2675a4e2012-06-22 12:41:10 +04003222 if (unlikely(error)) {
3223 if (error == -EOPENSTALE) {
3224 if (flags & LOOKUP_RCU)
3225 error = -ECHILD;
3226 else
3227 error = -ESTALE;
3228 }
3229 file = ERR_PTR(error);
3230 }
3231 return file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003232}
3233
Jeff Layton669abf42012-10-10 16:43:10 -04003234struct file *do_filp_open(int dfd, struct filename *pathname,
Al Virof9652e12013-06-11 08:23:01 +04003235 const struct open_flags *op)
Al Viro13aab422011-02-23 17:54:08 -05003236{
Al Viro73d049a2011-03-11 12:08:24 -05003237 struct nameidata nd;
Al Virof9652e12013-06-11 08:23:01 +04003238 int flags = op->lookup_flags;
Al Viro13aab422011-02-23 17:54:08 -05003239 struct file *filp;
3240
Al Viro73d049a2011-03-11 12:08:24 -05003241 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
Al Viro13aab422011-02-23 17:54:08 -05003242 if (unlikely(filp == ERR_PTR(-ECHILD)))
Al Viro73d049a2011-03-11 12:08:24 -05003243 filp = path_openat(dfd, pathname, &nd, op, flags);
Al Viro13aab422011-02-23 17:54:08 -05003244 if (unlikely(filp == ERR_PTR(-ESTALE)))
Al Viro73d049a2011-03-11 12:08:24 -05003245 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
Al Viro13aab422011-02-23 17:54:08 -05003246 return filp;
3247}
3248
Al Viro73d049a2011-03-11 12:08:24 -05003249struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
Al Virof9652e12013-06-11 08:23:01 +04003250 const char *name, const struct open_flags *op)
Al Viro73d049a2011-03-11 12:08:24 -05003251{
3252 struct nameidata nd;
3253 struct file *file;
Jeff Layton669abf42012-10-10 16:43:10 -04003254 struct filename filename = { .name = name };
Al Virof9652e12013-06-11 08:23:01 +04003255 int flags = op->lookup_flags | LOOKUP_ROOT;
Al Viro73d049a2011-03-11 12:08:24 -05003256
3257 nd.root.mnt = mnt;
3258 nd.root.dentry = dentry;
3259
David Howellsb18825a2013-09-12 19:22:53 +01003260 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
Al Viro73d049a2011-03-11 12:08:24 -05003261 return ERR_PTR(-ELOOP);
3262
Jeff Layton669abf42012-10-10 16:43:10 -04003263 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU);
Al Viro73d049a2011-03-11 12:08:24 -05003264 if (unlikely(file == ERR_PTR(-ECHILD)))
Jeff Layton669abf42012-10-10 16:43:10 -04003265 file = path_openat(-1, &filename, &nd, op, flags);
Al Viro73d049a2011-03-11 12:08:24 -05003266 if (unlikely(file == ERR_PTR(-ESTALE)))
Jeff Layton669abf42012-10-10 16:43:10 -04003267 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL);
Al Viro73d049a2011-03-11 12:08:24 -05003268 return file;
3269}
3270
Jeff Layton1ac12b42012-12-11 12:10:06 -05003271struct dentry *kern_path_create(int dfd, const char *pathname,
3272 struct path *path, unsigned int lookup_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003273{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003274 struct dentry *dentry = ERR_PTR(-EEXIST);
Al Viroed75e952011-06-27 16:53:43 -04003275 struct nameidata nd;
Jan Karac30dabf2012-06-12 16:20:30 +02003276 int err2;
Jeff Layton1ac12b42012-12-11 12:10:06 -05003277 int error;
3278 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3279
3280 /*
3281 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3282 * other flags passed in are ignored!
3283 */
3284 lookup_flags &= LOOKUP_REVAL;
3285
3286 error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd);
Al Viroed75e952011-06-27 16:53:43 -04003287 if (error)
3288 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003289
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003290 /*
3291 * Yucky last component or no last component at all?
3292 * (foo/., foo/.., /////)
3293 */
Al Viroed75e952011-06-27 16:53:43 -04003294 if (nd.last_type != LAST_NORM)
3295 goto out;
3296 nd.flags &= ~LOOKUP_PARENT;
3297 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003298
Jan Karac30dabf2012-06-12 16:20:30 +02003299 /* don't fail immediately if it's r/o, at least try to report other errors */
3300 err2 = mnt_want_write(nd.path.mnt);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003301 /*
3302 * Do the final lookup.
3303 */
Al Viroed75e952011-06-27 16:53:43 -04003304 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3305 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003306 if (IS_ERR(dentry))
Al Viroa8104a92012-07-20 02:25:00 +04003307 goto unlock;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003308
Al Viroa8104a92012-07-20 02:25:00 +04003309 error = -EEXIST;
David Howellsb18825a2013-09-12 19:22:53 +01003310 if (d_is_positive(dentry))
Al Viroa8104a92012-07-20 02:25:00 +04003311 goto fail;
David Howellsb18825a2013-09-12 19:22:53 +01003312
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07003313 /*
3314 * Special case - lookup gave negative, but... we had foo/bar/
3315 * From the vfs_mknod() POV we just have a negative dentry -
3316 * all is fine. Let's be bastards - you had / on the end, you've
3317 * been asking for (non-existent) directory. -ENOENT for you.
3318 */
Al Viroed75e952011-06-27 16:53:43 -04003319 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
Al Viroa8104a92012-07-20 02:25:00 +04003320 error = -ENOENT;
Al Viroed75e952011-06-27 16:53:43 -04003321 goto fail;
Al Viroe9baf6e2008-05-15 04:49:12 -04003322 }
Jan Karac30dabf2012-06-12 16:20:30 +02003323 if (unlikely(err2)) {
3324 error = err2;
Al Viroa8104a92012-07-20 02:25:00 +04003325 goto fail;
Jan Karac30dabf2012-06-12 16:20:30 +02003326 }
Al Viroed75e952011-06-27 16:53:43 -04003327 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003328 return dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003329fail:
Al Viroa8104a92012-07-20 02:25:00 +04003330 dput(dentry);
3331 dentry = ERR_PTR(error);
3332unlock:
Al Viroed75e952011-06-27 16:53:43 -04003333 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Karac30dabf2012-06-12 16:20:30 +02003334 if (!err2)
3335 mnt_drop_write(nd.path.mnt);
Al Viroed75e952011-06-27 16:53:43 -04003336out:
3337 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003338 return dentry;
3339}
Al Virodae6ad82011-06-26 11:50:15 -04003340EXPORT_SYMBOL(kern_path_create);
3341
Al Viro921a1652012-07-20 01:15:31 +04003342void done_path_create(struct path *path, struct dentry *dentry)
3343{
3344 dput(dentry);
3345 mutex_unlock(&path->dentry->d_inode->i_mutex);
Al Viroa8104a92012-07-20 02:25:00 +04003346 mnt_drop_write(path->mnt);
Al Viro921a1652012-07-20 01:15:31 +04003347 path_put(path);
3348}
3349EXPORT_SYMBOL(done_path_create);
3350
Jeff Layton1ac12b42012-12-11 12:10:06 -05003351struct dentry *user_path_create(int dfd, const char __user *pathname,
3352 struct path *path, unsigned int lookup_flags)
Al Virodae6ad82011-06-26 11:50:15 -04003353{
Jeff Layton91a27b22012-10-10 15:25:28 -04003354 struct filename *tmp = getname(pathname);
Al Virodae6ad82011-06-26 11:50:15 -04003355 struct dentry *res;
3356 if (IS_ERR(tmp))
3357 return ERR_CAST(tmp);
Jeff Layton1ac12b42012-12-11 12:10:06 -05003358 res = kern_path_create(dfd, tmp->name, path, lookup_flags);
Al Virodae6ad82011-06-26 11:50:15 -04003359 putname(tmp);
3360 return res;
3361}
3362EXPORT_SYMBOL(user_path_create);
3363
Al Viro1a67aaf2011-07-26 01:52:52 -04003364int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003365{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003366 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003367
3368 if (error)
3369 return error;
3370
Eric W. Biederman975d6b32011-11-13 12:16:43 -08003371 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003372 return -EPERM;
3373
Al Viroacfa4382008-12-04 10:06:33 -05003374 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003375 return -EPERM;
3376
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07003377 error = devcgroup_inode_mknod(mode, dev);
3378 if (error)
3379 return error;
3380
Linus Torvalds1da177e2005-04-16 15:20:36 -07003381 error = security_inode_mknod(dir, dentry, mode, dev);
3382 if (error)
3383 return error;
3384
Linus Torvalds1da177e2005-04-16 15:20:36 -07003385 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003386 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003387 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003388 return error;
3389}
Al Viro4d359502014-03-14 12:20:17 -04003390EXPORT_SYMBOL(vfs_mknod);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003391
Al Virof69aac02011-07-26 04:31:40 -04003392static int may_mknod(umode_t mode)
Dave Hansen463c3192008-02-15 14:37:57 -08003393{
3394 switch (mode & S_IFMT) {
3395 case S_IFREG:
3396 case S_IFCHR:
3397 case S_IFBLK:
3398 case S_IFIFO:
3399 case S_IFSOCK:
3400 case 0: /* zero mode translates to S_IFREG */
3401 return 0;
3402 case S_IFDIR:
3403 return -EPERM;
3404 default:
3405 return -EINVAL;
3406 }
3407}
3408
Al Viro8208a222011-07-25 17:32:17 -04003409SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003410 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003411{
Al Viro2ad94ae2008-07-21 09:32:51 -04003412 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003413 struct path path;
3414 int error;
Jeff Layton972567f2012-12-20 16:00:10 -05003415 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003416
Al Viro8e4bfca2012-07-20 01:17:26 +04003417 error = may_mknod(mode);
3418 if (error)
3419 return error;
Jeff Layton972567f2012-12-20 16:00:10 -05003420retry:
3421 dentry = user_path_create(dfd, filename, &path, lookup_flags);
Al Virodae6ad82011-06-26 11:50:15 -04003422 if (IS_ERR(dentry))
3423 return PTR_ERR(dentry);
Al Viro2ad94ae2008-07-21 09:32:51 -04003424
Al Virodae6ad82011-06-26 11:50:15 -04003425 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003426 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003427 error = security_path_mknod(&path, dentry, mode, dev);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003428 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04003429 goto out;
Dave Hansen463c3192008-02-15 14:37:57 -08003430 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003431 case 0: case S_IFREG:
Al Viro312b63f2012-06-10 18:09:36 -04003432 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003433 break;
3434 case S_IFCHR: case S_IFBLK:
Al Virodae6ad82011-06-26 11:50:15 -04003435 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003436 new_decode_dev(dev));
3437 break;
3438 case S_IFIFO: case S_IFSOCK:
Al Virodae6ad82011-06-26 11:50:15 -04003439 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003440 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003441 }
Al Viroa8104a92012-07-20 02:25:00 +04003442out:
Al Viro921a1652012-07-20 01:15:31 +04003443 done_path_create(&path, dentry);
Jeff Layton972567f2012-12-20 16:00:10 -05003444 if (retry_estale(error, lookup_flags)) {
3445 lookup_flags |= LOOKUP_REVAL;
3446 goto retry;
3447 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003448 return error;
3449}
3450
Al Viro8208a222011-07-25 17:32:17 -04003451SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003452{
3453 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3454}
3455
Al Viro18bb1db2011-07-26 01:41:39 -04003456int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003457{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003458 int error = may_create(dir, dentry);
Al Viro8de52772012-02-06 12:45:27 -05003459 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003460
3461 if (error)
3462 return error;
3463
Al Viroacfa4382008-12-04 10:06:33 -05003464 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003465 return -EPERM;
3466
3467 mode &= (S_IRWXUGO|S_ISVTX);
3468 error = security_inode_mkdir(dir, dentry, mode);
3469 if (error)
3470 return error;
3471
Al Viro8de52772012-02-06 12:45:27 -05003472 if (max_links && dir->i_nlink >= max_links)
3473 return -EMLINK;
3474
Linus Torvalds1da177e2005-04-16 15:20:36 -07003475 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003476 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003477 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003478 return error;
3479}
Al Viro4d359502014-03-14 12:20:17 -04003480EXPORT_SYMBOL(vfs_mkdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003481
Al Viroa218d0f2011-11-21 14:59:34 -05003482SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003483{
Dave Hansen6902d922006-09-30 23:29:01 -07003484 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003485 struct path path;
3486 int error;
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003487 unsigned int lookup_flags = LOOKUP_DIRECTORY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003488
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003489retry:
3490 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07003491 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003492 return PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003493
Al Virodae6ad82011-06-26 11:50:15 -04003494 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003495 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003496 error = security_path_mkdir(&path, dentry, mode);
Al Viroa8104a92012-07-20 02:25:00 +04003497 if (!error)
3498 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
Al Viro921a1652012-07-20 01:15:31 +04003499 done_path_create(&path, dentry);
Jeff Laytonb76d8b82012-12-20 16:04:09 -05003500 if (retry_estale(error, lookup_flags)) {
3501 lookup_flags |= LOOKUP_REVAL;
3502 goto retry;
3503 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003504 return error;
3505}
3506
Al Viroa218d0f2011-11-21 14:59:34 -05003507SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003508{
3509 return sys_mkdirat(AT_FDCWD, pathname, mode);
3510}
3511
Linus Torvalds1da177e2005-04-16 15:20:36 -07003512/*
Sage Weila71905f2011-05-24 13:06:08 -07003513 * The dentry_unhash() helper will try to drop the dentry early: we
J. Bruce Fieldsc0d02592012-02-15 11:48:40 -05003514 * should have a usage count of 1 if we're the only user of this
Sage Weila71905f2011-05-24 13:06:08 -07003515 * dentry, and if that is true (possibly after pruning the dcache),
3516 * then we drop the dentry now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003517 *
3518 * A low-level filesystem can, if it choses, legally
3519 * do a
3520 *
3521 * if (!d_unhashed(dentry))
3522 * return -EBUSY;
3523 *
3524 * if it cannot handle the case of removing a directory
3525 * that is still in use by something else..
3526 */
3527void dentry_unhash(struct dentry *dentry)
3528{
Vasily Averindc168422006-12-06 20:37:07 -08003529 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003530 spin_lock(&dentry->d_lock);
Waiman Long98474232013-08-28 18:24:59 -07003531 if (dentry->d_lockref.count == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003532 __d_drop(dentry);
3533 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003534}
Al Viro4d359502014-03-14 12:20:17 -04003535EXPORT_SYMBOL(dentry_unhash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003536
3537int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3538{
3539 int error = may_delete(dir, dentry, 1);
3540
3541 if (error)
3542 return error;
3543
Al Viroacfa4382008-12-04 10:06:33 -05003544 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003545 return -EPERM;
3546
Al Viro1d2ef592011-09-14 18:55:41 +01003547 dget(dentry);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003548 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003549
Sage Weil912dbc12011-05-24 13:06:11 -07003550 error = -EBUSY;
3551 if (d_mountpoint(dentry))
3552 goto out;
3553
3554 error = security_inode_rmdir(dir, dentry);
3555 if (error)
3556 goto out;
3557
Sage Weil3cebde22011-05-29 21:20:59 -07003558 shrink_dcache_parent(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003559 error = dir->i_op->rmdir(dir, dentry);
3560 if (error)
3561 goto out;
3562
3563 dentry->d_inode->i_flags |= S_DEAD;
3564 dont_mount(dentry);
3565
3566out:
3567 mutex_unlock(&dentry->d_inode->i_mutex);
Al Viro1d2ef592011-09-14 18:55:41 +01003568 dput(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003569 if (!error)
3570 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003571 return error;
3572}
Al Viro4d359502014-03-14 12:20:17 -04003573EXPORT_SYMBOL(vfs_rmdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003574
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003575static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003576{
3577 int error = 0;
Jeff Layton91a27b22012-10-10 15:25:28 -04003578 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003579 struct dentry *dentry;
3580 struct nameidata nd;
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003581 unsigned int lookup_flags = 0;
3582retry:
3583 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04003584 if (IS_ERR(name))
3585 return PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003586
3587 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003588 case LAST_DOTDOT:
3589 error = -ENOTEMPTY;
3590 goto exit1;
3591 case LAST_DOT:
3592 error = -EINVAL;
3593 goto exit1;
3594 case LAST_ROOT:
3595 error = -EBUSY;
3596 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003597 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003598
3599 nd.flags &= ~LOOKUP_PARENT;
Jan Karac30dabf2012-06-12 16:20:30 +02003600 error = mnt_want_write(nd.path.mnt);
3601 if (error)
3602 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003603
Jan Blunck4ac91372008-02-14 19:34:32 -08003604 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08003605 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003606 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003607 if (IS_ERR(dentry))
3608 goto exit2;
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003609 if (!dentry->d_inode) {
3610 error = -ENOENT;
3611 goto exit3;
3612 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003613 error = security_path_rmdir(&nd.path, dentry);
3614 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003615 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08003616 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Dave Hansen06227532008-02-15 14:37:34 -08003617exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07003618 dput(dentry);
3619exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08003620 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Karac30dabf2012-06-12 16:20:30 +02003621 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003622exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003623 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003624 putname(name);
Jeff Laytonc6ee9202012-12-20 16:28:33 -05003625 if (retry_estale(error, lookup_flags)) {
3626 lookup_flags |= LOOKUP_REVAL;
3627 goto retry;
3628 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003629 return error;
3630}
3631
Heiko Carstens3cdad422009-01-14 14:14:22 +01003632SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003633{
3634 return do_rmdir(AT_FDCWD, pathname);
3635}
3636
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003637/**
3638 * vfs_unlink - unlink a filesystem object
3639 * @dir: parent directory
3640 * @dentry: victim
3641 * @delegated_inode: returns victim inode, if the inode is delegated.
3642 *
3643 * The caller must hold dir->i_mutex.
3644 *
3645 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3646 * return a reference to the inode in delegated_inode. The caller
3647 * should then break the delegation on that inode and retry. Because
3648 * breaking a delegation may take a long time, the caller should drop
3649 * dir->i_mutex before doing so.
3650 *
3651 * Alternatively, a caller may pass NULL for delegated_inode. This may
3652 * be appropriate for callers that expect the underlying filesystem not
3653 * to be NFS exported.
3654 */
3655int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003656{
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003657 struct inode *target = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003658 int error = may_delete(dir, dentry, 0);
3659
3660 if (error)
3661 return error;
3662
Al Viroacfa4382008-12-04 10:06:33 -05003663 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003664 return -EPERM;
3665
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003666 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003667 if (d_mountpoint(dentry))
3668 error = -EBUSY;
3669 else {
3670 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05003671 if (!error) {
J. Bruce Fields5a146962012-08-28 07:50:40 -07003672 error = try_break_deleg(target, delegated_inode);
3673 if (error)
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003674 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003675 error = dir->i_op->unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05003676 if (!error)
Al Virod83c49f2010-04-30 17:17:09 -04003677 dont_mount(dentry);
Al Virobec10522010-03-03 14:12:08 -05003678 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003679 }
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003680out:
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003681 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003682
3683 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3684 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
J. Bruce Fields9accbb92012-08-28 07:03:24 -04003685 fsnotify_link_count(target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003686 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003687 }
Robert Love0eeca282005-07-12 17:06:03 -04003688
Linus Torvalds1da177e2005-04-16 15:20:36 -07003689 return error;
3690}
Al Viro4d359502014-03-14 12:20:17 -04003691EXPORT_SYMBOL(vfs_unlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003692
3693/*
3694 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003695 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07003696 * writeout happening, and we don't want to prevent access to the directory
3697 * while waiting on the I/O.
3698 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003699static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003700{
Al Viro2ad94ae2008-07-21 09:32:51 -04003701 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04003702 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003703 struct dentry *dentry;
3704 struct nameidata nd;
3705 struct inode *inode = NULL;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003706 struct inode *delegated_inode = NULL;
Jeff Layton5d18f812012-12-20 16:38:04 -05003707 unsigned int lookup_flags = 0;
3708retry:
3709 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04003710 if (IS_ERR(name))
3711 return PTR_ERR(name);
Al Viro2ad94ae2008-07-21 09:32:51 -04003712
Linus Torvalds1da177e2005-04-16 15:20:36 -07003713 error = -EISDIR;
3714 if (nd.last_type != LAST_NORM)
3715 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003716
3717 nd.flags &= ~LOOKUP_PARENT;
Jan Karac30dabf2012-06-12 16:20:30 +02003718 error = mnt_want_write(nd.path.mnt);
3719 if (error)
3720 goto exit1;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003721retry_deleg:
Jan Blunck4ac91372008-02-14 19:34:32 -08003722 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08003723 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003724 error = PTR_ERR(dentry);
3725 if (!IS_ERR(dentry)) {
3726 /* Why not before? Because we want correct error value */
Török Edwin50338b82011-06-16 00:06:14 +03003727 if (nd.last.name[nd.last.len])
3728 goto slashes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003729 inode = dentry->d_inode;
David Howellsb18825a2013-09-12 19:22:53 +01003730 if (d_is_negative(dentry))
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003731 goto slashes;
3732 ihold(inode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003733 error = security_path_unlink(&nd.path, dentry);
3734 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02003735 goto exit2;
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003736 error = vfs_unlink(nd.path.dentry->d_inode, dentry, &delegated_inode);
Jan Karac30dabf2012-06-12 16:20:30 +02003737exit2:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003738 dput(dentry);
3739 }
Jan Blunck4ac91372008-02-14 19:34:32 -08003740 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003741 if (inode)
3742 iput(inode); /* truncate the inode here */
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003743 inode = NULL;
3744 if (delegated_inode) {
J. Bruce Fields5a146962012-08-28 07:50:40 -07003745 error = break_deleg_wait(&delegated_inode);
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04003746 if (!error)
3747 goto retry_deleg;
3748 }
Jan Karac30dabf2012-06-12 16:20:30 +02003749 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003750exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003751 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003752 putname(name);
Jeff Layton5d18f812012-12-20 16:38:04 -05003753 if (retry_estale(error, lookup_flags)) {
3754 lookup_flags |= LOOKUP_REVAL;
3755 inode = NULL;
3756 goto retry;
3757 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003758 return error;
3759
3760slashes:
David Howellsb18825a2013-09-12 19:22:53 +01003761 if (d_is_negative(dentry))
3762 error = -ENOENT;
Miklos Szeredi44b1d532014-04-01 17:08:41 +02003763 else if (d_is_dir(dentry))
David Howellsb18825a2013-09-12 19:22:53 +01003764 error = -EISDIR;
3765 else
3766 error = -ENOTDIR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003767 goto exit2;
3768}
3769
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003770SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003771{
3772 if ((flag & ~AT_REMOVEDIR) != 0)
3773 return -EINVAL;
3774
3775 if (flag & AT_REMOVEDIR)
3776 return do_rmdir(dfd, pathname);
3777
3778 return do_unlinkat(dfd, pathname);
3779}
3780
Heiko Carstens3480b252009-01-14 14:14:16 +01003781SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003782{
3783 return do_unlinkat(AT_FDCWD, pathname);
3784}
3785
Miklos Szeredidb2e7472008-06-24 16:50:16 +02003786int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003787{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003788 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003789
3790 if (error)
3791 return error;
3792
Al Viroacfa4382008-12-04 10:06:33 -05003793 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003794 return -EPERM;
3795
3796 error = security_inode_symlink(dir, dentry, oldname);
3797 if (error)
3798 return error;
3799
Linus Torvalds1da177e2005-04-16 15:20:36 -07003800 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003801 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003802 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003803 return error;
3804}
Al Viro4d359502014-03-14 12:20:17 -04003805EXPORT_SYMBOL(vfs_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003806
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003807SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3808 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003809{
Al Viro2ad94ae2008-07-21 09:32:51 -04003810 int error;
Jeff Layton91a27b22012-10-10 15:25:28 -04003811 struct filename *from;
Dave Hansen6902d922006-09-30 23:29:01 -07003812 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003813 struct path path;
Jeff Laytonf46d3562012-12-11 12:10:08 -05003814 unsigned int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003815
3816 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04003817 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003818 return PTR_ERR(from);
Jeff Laytonf46d3562012-12-11 12:10:08 -05003819retry:
3820 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
Dave Hansen6902d922006-09-30 23:29:01 -07003821 error = PTR_ERR(dentry);
3822 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003823 goto out_putname;
Dave Hansen6902d922006-09-30 23:29:01 -07003824
Jeff Layton91a27b22012-10-10 15:25:28 -04003825 error = security_path_symlink(&path, dentry, from->name);
Al Viroa8104a92012-07-20 02:25:00 +04003826 if (!error)
Jeff Layton91a27b22012-10-10 15:25:28 -04003827 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
Al Viro921a1652012-07-20 01:15:31 +04003828 done_path_create(&path, dentry);
Jeff Laytonf46d3562012-12-11 12:10:08 -05003829 if (retry_estale(error, lookup_flags)) {
3830 lookup_flags |= LOOKUP_REVAL;
3831 goto retry;
3832 }
Dave Hansen6902d922006-09-30 23:29:01 -07003833out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003834 putname(from);
3835 return error;
3836}
3837
Heiko Carstens3480b252009-01-14 14:14:16 +01003838SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003839{
3840 return sys_symlinkat(oldname, AT_FDCWD, newname);
3841}
3842
J. Bruce Fields146a8592011-09-20 17:14:31 -04003843/**
3844 * vfs_link - create a new link
3845 * @old_dentry: object to be linked
3846 * @dir: new parent
3847 * @new_dentry: where to create the new link
3848 * @delegated_inode: returns inode needing a delegation break
3849 *
3850 * The caller must hold dir->i_mutex
3851 *
3852 * If vfs_link discovers a delegation on the to-be-linked file in need
3853 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3854 * inode in delegated_inode. The caller should then break the delegation
3855 * and retry. Because breaking a delegation may take a long time, the
3856 * caller should drop the i_mutex before doing so.
3857 *
3858 * Alternatively, a caller may pass NULL for delegated_inode. This may
3859 * be appropriate for callers that expect the underlying filesystem not
3860 * to be NFS exported.
3861 */
3862int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003863{
3864 struct inode *inode = old_dentry->d_inode;
Al Viro8de52772012-02-06 12:45:27 -05003865 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003866 int error;
3867
3868 if (!inode)
3869 return -ENOENT;
3870
Miklos Szeredia95164d2008-07-30 15:08:48 +02003871 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003872 if (error)
3873 return error;
3874
3875 if (dir->i_sb != inode->i_sb)
3876 return -EXDEV;
3877
3878 /*
3879 * A link to an append-only or immutable file cannot be created.
3880 */
3881 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3882 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05003883 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003884 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003885 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003886 return -EPERM;
3887
3888 error = security_inode_link(old_dentry, dir, new_dentry);
3889 if (error)
3890 return error;
3891
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003892 mutex_lock(&inode->i_mutex);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05303893 /* Make sure we don't allow creating hardlink to an unlinked file */
Al Virof4e0c302013-06-11 08:34:36 +04003894 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05303895 error = -ENOENT;
Al Viro8de52772012-02-06 12:45:27 -05003896 else if (max_links && inode->i_nlink >= max_links)
3897 error = -EMLINK;
J. Bruce Fields146a8592011-09-20 17:14:31 -04003898 else {
3899 error = try_break_deleg(inode, delegated_inode);
3900 if (!error)
3901 error = dir->i_op->link(old_dentry, dir, new_dentry);
3902 }
Al Virof4e0c302013-06-11 08:34:36 +04003903
3904 if (!error && (inode->i_state & I_LINKABLE)) {
3905 spin_lock(&inode->i_lock);
3906 inode->i_state &= ~I_LINKABLE;
3907 spin_unlock(&inode->i_lock);
3908 }
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003909 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07003910 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003911 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003912 return error;
3913}
Al Viro4d359502014-03-14 12:20:17 -04003914EXPORT_SYMBOL(vfs_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003915
3916/*
3917 * Hardlinks are often used in delicate situations. We avoid
3918 * security-related surprises by not following symlinks on the
3919 * newname. --KAB
3920 *
3921 * We don't follow them on the oldname either to be compatible
3922 * with linux 2.0, and to avoid hard-linking to directories
3923 * and other special files. --ADM
3924 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003925SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3926 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003927{
3928 struct dentry *new_dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003929 struct path old_path, new_path;
J. Bruce Fields146a8592011-09-20 17:14:31 -04003930 struct inode *delegated_inode = NULL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303931 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003932 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003933
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303934 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003935 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303936 /*
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07003937 * To use null names we require CAP_DAC_READ_SEARCH
3938 * This ensures that not everyone will be able to create
3939 * handlink using the passed filedescriptor.
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303940 */
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07003941 if (flags & AT_EMPTY_PATH) {
3942 if (!capable(CAP_DAC_READ_SEARCH))
3943 return -ENOENT;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303944 how = LOOKUP_EMPTY;
Linus Torvaldsf0cc6ff2013-08-28 09:18:05 -07003945 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003946
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303947 if (flags & AT_SYMLINK_FOLLOW)
3948 how |= LOOKUP_FOLLOW;
Jeff Layton442e31c2012-12-20 16:15:38 -05003949retry:
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303950 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003951 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04003952 return error;
3953
Jeff Layton442e31c2012-12-20 16:15:38 -05003954 new_dentry = user_path_create(newdfd, newname, &new_path,
3955 (how & LOOKUP_REVAL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003956 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003957 if (IS_ERR(new_dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003958 goto out;
3959
3960 error = -EXDEV;
3961 if (old_path.mnt != new_path.mnt)
3962 goto out_dput;
Kees Cook800179c2012-07-25 17:29:07 -07003963 error = may_linkat(&old_path);
3964 if (unlikely(error))
3965 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04003966 error = security_path_link(old_path.dentry, &new_path, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003967 if (error)
Al Viroa8104a92012-07-20 02:25:00 +04003968 goto out_dput;
J. Bruce Fields146a8592011-09-20 17:14:31 -04003969 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
Dave Hansen75c3f292008-02-15 14:37:45 -08003970out_dput:
Al Viro921a1652012-07-20 01:15:31 +04003971 done_path_create(&new_path, new_dentry);
J. Bruce Fields146a8592011-09-20 17:14:31 -04003972 if (delegated_inode) {
3973 error = break_deleg_wait(&delegated_inode);
Oleg Drokind22e6332014-01-31 15:41:58 -05003974 if (!error) {
3975 path_put(&old_path);
J. Bruce Fields146a8592011-09-20 17:14:31 -04003976 goto retry;
Oleg Drokind22e6332014-01-31 15:41:58 -05003977 }
J. Bruce Fields146a8592011-09-20 17:14:31 -04003978 }
Jeff Layton442e31c2012-12-20 16:15:38 -05003979 if (retry_estale(error, how)) {
Oleg Drokind22e6332014-01-31 15:41:58 -05003980 path_put(&old_path);
Jeff Layton442e31c2012-12-20 16:15:38 -05003981 how |= LOOKUP_REVAL;
3982 goto retry;
3983 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003984out:
Al Viro2d8f3032008-07-22 09:59:21 -04003985 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003986
3987 return error;
3988}
3989
Heiko Carstens3480b252009-01-14 14:14:16 +01003990SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003991{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003992 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003993}
3994
Miklos Szeredibc270272014-04-01 17:08:42 +02003995/**
3996 * vfs_rename - rename a filesystem object
3997 * @old_dir: parent of source
3998 * @old_dentry: source
3999 * @new_dir: parent of destination
4000 * @new_dentry: destination
4001 * @delegated_inode: returns an inode needing a delegation break
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004002 * @flags: rename flags
Miklos Szeredibc270272014-04-01 17:08:42 +02004003 *
4004 * The caller must hold multiple mutexes--see lock_rename()).
4005 *
4006 * If vfs_rename discovers a delegation in need of breaking at either
4007 * the source or destination, it will return -EWOULDBLOCK and return a
4008 * reference to the inode in delegated_inode. The caller should then
4009 * break the delegation and retry. Because breaking a delegation may
4010 * take a long time, the caller should drop all locks before doing
4011 * so.
4012 *
4013 * Alternatively, a caller may pass NULL for delegated_inode. This may
4014 * be appropriate for callers that expect the underlying filesystem not
4015 * to be NFS exported.
4016 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07004017 * The worst of all namespace operations - renaming directory. "Perverted"
4018 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4019 * Problems:
4020 * a) we can get into loop creation. Check is done in is_subdir().
4021 * b) race potential - two innocent renames can create a loop together.
4022 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004023 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07004024 * story.
J. Bruce Fields6cedba82012-03-05 11:40:41 -05004025 * c) we have to lock _four_ objects - parents and victim (if it exists),
4026 * and source (if it is not a directory).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004027 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07004028 * whether the target exists). Solution: try to be smart with locking
4029 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004030 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07004031 * move will be locked. Thus we can rank directories by the tree
4032 * (ancestors first) and rank all non-directories after them.
4033 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08004034 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004035 * HOWEVER, it relies on the assumption that any object with ->lookup()
4036 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4037 * we'd better make sure that there's no link(2) for them.
Sage Weile4eaac02011-05-24 13:06:07 -07004038 * d) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004039 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07004040 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05004041 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07004042 * locking].
4043 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004044int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004045 struct inode *new_dir, struct dentry *new_dentry,
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004046 struct inode **delegated_inode, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004047{
4048 int error;
Miklos Szeredibc270272014-04-01 17:08:42 +02004049 bool is_dir = d_is_dir(old_dentry);
Eric Paris59b0df22010-02-08 12:53:52 -05004050 const unsigned char *old_name;
Miklos Szeredibc270272014-04-01 17:08:42 +02004051 struct inode *source = old_dentry->d_inode;
4052 struct inode *target = new_dentry->d_inode;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004053 bool new_is_dir = false;
4054 unsigned max_links = new_dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004055
Miklos Szeredibc270272014-04-01 17:08:42 +02004056 if (source == target)
4057 return 0;
4058
Linus Torvalds1da177e2005-04-16 15:20:36 -07004059 error = may_delete(old_dir, old_dentry, is_dir);
4060 if (error)
4061 return error;
4062
Miklos Szeredida1ce062014-04-01 17:08:43 +02004063 if (!target) {
Miklos Szeredia95164d2008-07-30 15:08:48 +02004064 error = may_create(new_dir, new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004065 } else {
4066 new_is_dir = d_is_dir(new_dentry);
4067
4068 if (!(flags & RENAME_EXCHANGE))
4069 error = may_delete(new_dir, new_dentry, is_dir);
4070 else
4071 error = may_delete(new_dir, new_dentry, new_is_dir);
4072 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004073 if (error)
4074 return error;
4075
Al Viroacfa4382008-12-04 10:06:33 -05004076 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004077 return -EPERM;
4078
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004079 if (flags && !old_dir->i_op->rename2)
4080 return -EINVAL;
4081
Miklos Szeredibc270272014-04-01 17:08:42 +02004082 /*
4083 * If we are going to change the parent - check write permissions,
4084 * we'll need to flip '..'.
4085 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02004086 if (new_dir != old_dir) {
4087 if (is_dir) {
4088 error = inode_permission(source, MAY_WRITE);
4089 if (error)
4090 return error;
4091 }
4092 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4093 error = inode_permission(target, MAY_WRITE);
4094 if (error)
4095 return error;
4096 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004097 }
Robert Love0eeca282005-07-12 17:06:03 -04004098
Miklos Szeredi0b3974e2014-04-01 17:08:43 +02004099 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4100 flags);
Miklos Szeredibc270272014-04-01 17:08:42 +02004101 if (error)
4102 return error;
4103
4104 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4105 dget(new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004106 if (!is_dir || (flags & RENAME_EXCHANGE))
Miklos Szeredibc270272014-04-01 17:08:42 +02004107 lock_two_nondirectories(source, target);
4108 else if (target)
4109 mutex_lock(&target->i_mutex);
4110
4111 error = -EBUSY;
4112 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
4113 goto out;
4114
Miklos Szeredida1ce062014-04-01 17:08:43 +02004115 if (max_links && new_dir != old_dir) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004116 error = -EMLINK;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004117 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
Miklos Szeredibc270272014-04-01 17:08:42 +02004118 goto out;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004119 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4120 old_dir->i_nlink >= max_links)
4121 goto out;
4122 }
4123 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4124 shrink_dcache_parent(new_dentry);
4125 if (!is_dir) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004126 error = try_break_deleg(source, delegated_inode);
4127 if (error)
4128 goto out;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004129 }
4130 if (target && !new_is_dir) {
4131 error = try_break_deleg(target, delegated_inode);
4132 if (error)
4133 goto out;
Miklos Szeredibc270272014-04-01 17:08:42 +02004134 }
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004135 if (!flags) {
4136 error = old_dir->i_op->rename(old_dir, old_dentry,
4137 new_dir, new_dentry);
4138 } else {
4139 error = old_dir->i_op->rename2(old_dir, old_dentry,
4140 new_dir, new_dentry, flags);
4141 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004142 if (error)
4143 goto out;
4144
Miklos Szeredida1ce062014-04-01 17:08:43 +02004145 if (!(flags & RENAME_EXCHANGE) && target) {
Miklos Szeredibc270272014-04-01 17:08:42 +02004146 if (is_dir)
4147 target->i_flags |= S_DEAD;
4148 dont_mount(new_dentry);
4149 }
Miklos Szeredida1ce062014-04-01 17:08:43 +02004150 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4151 if (!(flags & RENAME_EXCHANGE))
4152 d_move(old_dentry, new_dentry);
4153 else
4154 d_exchange(old_dentry, new_dentry);
4155 }
Miklos Szeredibc270272014-04-01 17:08:42 +02004156out:
Miklos Szeredida1ce062014-04-01 17:08:43 +02004157 if (!is_dir || (flags & RENAME_EXCHANGE))
Miklos Szeredibc270272014-04-01 17:08:42 +02004158 unlock_two_nondirectories(source, target);
4159 else if (target)
4160 mutex_unlock(&target->i_mutex);
4161 dput(new_dentry);
Miklos Szeredida1ce062014-04-01 17:08:43 +02004162 if (!error) {
Al Viro123df292009-12-25 04:57:57 -05004163 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Miklos Szeredida1ce062014-04-01 17:08:43 +02004164 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4165 if (flags & RENAME_EXCHANGE) {
4166 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4167 new_is_dir, NULL, new_dentry);
4168 }
4169 }
Robert Love0eeca282005-07-12 17:06:03 -04004170 fsnotify_oldname_free(old_name);
4171
Linus Torvalds1da177e2005-04-16 15:20:36 -07004172 return error;
4173}
Al Viro4d359502014-03-14 12:20:17 -04004174EXPORT_SYMBOL(vfs_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004175
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004176SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4177 int, newdfd, const char __user *, newname, unsigned int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004178{
Al Viro2ad94ae2008-07-21 09:32:51 -04004179 struct dentry *old_dir, *new_dir;
4180 struct dentry *old_dentry, *new_dentry;
4181 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004182 struct nameidata oldnd, newnd;
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004183 struct inode *delegated_inode = NULL;
Jeff Layton91a27b22012-10-10 15:25:28 -04004184 struct filename *from;
4185 struct filename *to;
Jeff Laytonc6a94282012-12-11 12:10:10 -05004186 unsigned int lookup_flags = 0;
4187 bool should_retry = false;
Al Viro2ad94ae2008-07-21 09:32:51 -04004188 int error;
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004189
Miklos Szeredida1ce062014-04-01 17:08:43 +02004190 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE))
4191 return -EINVAL;
4192
4193 if ((flags & RENAME_NOREPLACE) && (flags & RENAME_EXCHANGE))
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004194 return -EINVAL;
4195
Jeff Laytonc6a94282012-12-11 12:10:10 -05004196retry:
4197 from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04004198 if (IS_ERR(from)) {
4199 error = PTR_ERR(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004200 goto exit;
Jeff Layton91a27b22012-10-10 15:25:28 -04004201 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004202
Jeff Laytonc6a94282012-12-11 12:10:10 -05004203 to = user_path_parent(newdfd, newname, &newnd, lookup_flags);
Jeff Layton91a27b22012-10-10 15:25:28 -04004204 if (IS_ERR(to)) {
4205 error = PTR_ERR(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004206 goto exit1;
Jeff Layton91a27b22012-10-10 15:25:28 -04004207 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004208
4209 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08004210 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004211 goto exit2;
4212
Jan Blunck4ac91372008-02-14 19:34:32 -08004213 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004214 error = -EBUSY;
4215 if (oldnd.last_type != LAST_NORM)
4216 goto exit2;
4217
Jan Blunck4ac91372008-02-14 19:34:32 -08004218 new_dir = newnd.path.dentry;
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004219 if (flags & RENAME_NOREPLACE)
4220 error = -EEXIST;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004221 if (newnd.last_type != LAST_NORM)
4222 goto exit2;
4223
Jan Karac30dabf2012-06-12 16:20:30 +02004224 error = mnt_want_write(oldnd.path.mnt);
4225 if (error)
4226 goto exit2;
4227
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09004228 oldnd.flags &= ~LOOKUP_PARENT;
4229 newnd.flags &= ~LOOKUP_PARENT;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004230 if (!(flags & RENAME_EXCHANGE))
4231 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09004232
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004233retry_deleg:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004234 trap = lock_rename(new_dir, old_dir);
4235
Christoph Hellwig49705b72005-11-08 21:35:06 -08004236 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004237 error = PTR_ERR(old_dentry);
4238 if (IS_ERR(old_dentry))
4239 goto exit3;
4240 /* source must exist */
4241 error = -ENOENT;
David Howellsb18825a2013-09-12 19:22:53 +01004242 if (d_is_negative(old_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004243 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08004244 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004245 error = PTR_ERR(new_dentry);
4246 if (IS_ERR(new_dentry))
4247 goto exit4;
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004248 error = -EEXIST;
4249 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4250 goto exit5;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004251 if (flags & RENAME_EXCHANGE) {
4252 error = -ENOENT;
4253 if (d_is_negative(new_dentry))
4254 goto exit5;
4255
4256 if (!d_is_dir(new_dentry)) {
4257 error = -ENOTDIR;
4258 if (newnd.last.name[newnd.last.len])
4259 goto exit5;
4260 }
4261 }
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004262 /* unless the source is a directory trailing slashes give -ENOTDIR */
4263 if (!d_is_dir(old_dentry)) {
4264 error = -ENOTDIR;
4265 if (oldnd.last.name[oldnd.last.len])
4266 goto exit5;
Miklos Szeredida1ce062014-04-01 17:08:43 +02004267 if (!(flags & RENAME_EXCHANGE) && newnd.last.name[newnd.last.len])
Miklos Szeredi0a7c3932014-04-01 17:08:43 +02004268 goto exit5;
4269 }
4270 /* source should not be ancestor of target */
4271 error = -EINVAL;
4272 if (old_dentry == trap)
4273 goto exit5;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004274 /* target should not be an ancestor of source */
Miklos Szeredida1ce062014-04-01 17:08:43 +02004275 if (!(flags & RENAME_EXCHANGE))
4276 error = -ENOTEMPTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004277 if (new_dentry == trap)
4278 goto exit5;
4279
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004280 error = security_path_rename(&oldnd.path, old_dentry,
Miklos Szeredi0b3974e2014-04-01 17:08:43 +02004281 &newnd.path, new_dentry, flags);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09004282 if (error)
Jan Karac30dabf2012-06-12 16:20:30 +02004283 goto exit5;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004284 error = vfs_rename(old_dir->d_inode, old_dentry,
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004285 new_dir->d_inode, new_dentry,
4286 &delegated_inode, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004287exit5:
4288 dput(new_dentry);
4289exit4:
4290 dput(old_dentry);
4291exit3:
4292 unlock_rename(new_dir, old_dir);
J. Bruce Fields8e6d7822011-09-20 16:59:58 -04004293 if (delegated_inode) {
4294 error = break_deleg_wait(&delegated_inode);
4295 if (!error)
4296 goto retry_deleg;
4297 }
Jan Karac30dabf2012-06-12 16:20:30 +02004298 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004299exit2:
Jeff Laytonc6a94282012-12-11 12:10:10 -05004300 if (retry_estale(error, lookup_flags))
4301 should_retry = true;
Jan Blunck1d957f92008-02-14 19:34:35 -08004302 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04004303 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004304exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08004305 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004306 putname(from);
Jeff Laytonc6a94282012-12-11 12:10:10 -05004307 if (should_retry) {
4308 should_retry = false;
4309 lookup_flags |= LOOKUP_REVAL;
4310 goto retry;
4311 }
Al Viro2ad94ae2008-07-21 09:32:51 -04004312exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004313 return error;
4314}
4315
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004316SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4317 int, newdfd, const char __user *, newname)
4318{
4319 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4320}
4321
Heiko Carstensa26eab22009-01-14 14:14:17 +01004322SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004323{
Miklos Szeredi520c8b12014-04-01 17:08:42 +02004324 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08004325}
4326
Al Viro5d826c82014-03-14 13:42:45 -04004327int readlink_copy(char __user *buffer, int buflen, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004328{
Al Viro5d826c82014-03-14 13:42:45 -04004329 int len = PTR_ERR(link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004330 if (IS_ERR(link))
4331 goto out;
4332
4333 len = strlen(link);
4334 if (len > (unsigned) buflen)
4335 len = buflen;
4336 if (copy_to_user(buffer, link, len))
4337 len = -EFAULT;
4338out:
4339 return len;
4340}
Al Viro5d826c82014-03-14 13:42:45 -04004341EXPORT_SYMBOL(readlink_copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004342
4343/*
4344 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4345 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4346 * using) it for any given inode is up to filesystem.
4347 */
4348int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4349{
4350 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004351 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07004352 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004353
Linus Torvalds1da177e2005-04-16 15:20:36 -07004354 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004355 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07004356 if (IS_ERR(cookie))
4357 return PTR_ERR(cookie);
4358
Al Viro5d826c82014-03-14 13:42:45 -04004359 res = readlink_copy(buffer, buflen, nd_get_link(&nd));
Marcin Slusarz694a1762008-06-09 16:40:37 -07004360 if (dentry->d_inode->i_op->put_link)
4361 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
4362 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004363}
Al Viro4d359502014-03-14 12:20:17 -04004364EXPORT_SYMBOL(generic_readlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004365
Linus Torvalds1da177e2005-04-16 15:20:36 -07004366/* get the link contents into pagecache */
4367static char *page_getlink(struct dentry * dentry, struct page **ppage)
4368{
Duane Griffinebd09ab2008-12-19 20:47:12 +00004369 char *kaddr;
4370 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004371 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07004372 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004373 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07004374 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004375 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00004376 kaddr = kmap(page);
4377 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4378 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004379}
4380
4381int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4382{
4383 struct page *page = NULL;
Al Viro5d826c82014-03-14 13:42:45 -04004384 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004385 if (page) {
4386 kunmap(page);
4387 page_cache_release(page);
4388 }
4389 return res;
4390}
Al Viro4d359502014-03-14 12:20:17 -04004391EXPORT_SYMBOL(page_readlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004392
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004393void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004394{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004395 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004396 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004397 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004398}
Al Viro4d359502014-03-14 12:20:17 -04004399EXPORT_SYMBOL(page_follow_link_light);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004400
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004401void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004402{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07004403 struct page *page = cookie;
4404
4405 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004406 kunmap(page);
4407 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004408 }
4409}
Al Viro4d359502014-03-14 12:20:17 -04004410EXPORT_SYMBOL(page_put_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004411
Nick Piggin54566b22009-01-04 12:00:53 -08004412/*
4413 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4414 */
4415int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004416{
4417 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004418 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07004419 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08004420 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004421 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08004422 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4423 if (nofs)
4424 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004425
NeilBrown7e53cac2006-03-25 03:07:57 -08004426retry:
Nick Pigginafddba42007-10-16 01:25:01 -07004427 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08004428 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004429 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07004430 goto fail;
4431
Cong Wange8e3c3d2011-11-25 23:14:27 +08004432 kaddr = kmap_atomic(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004433 memcpy(kaddr, symname, len-1);
Cong Wange8e3c3d2011-11-25 23:14:27 +08004434 kunmap_atomic(kaddr);
Nick Pigginafddba42007-10-16 01:25:01 -07004435
4436 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4437 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004438 if (err < 0)
4439 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07004440 if (err < len-1)
4441 goto retry;
4442
Linus Torvalds1da177e2005-04-16 15:20:36 -07004443 mark_inode_dirty(inode);
4444 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004445fail:
4446 return err;
4447}
Al Viro4d359502014-03-14 12:20:17 -04004448EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004449
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004450int page_symlink(struct inode *inode, const char *symname, int len)
4451{
4452 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08004453 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004454}
Al Viro4d359502014-03-14 12:20:17 -04004455EXPORT_SYMBOL(page_symlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08004456
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08004457const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004458 .readlink = generic_readlink,
4459 .follow_link = page_follow_link_light,
4460 .put_link = page_put_link,
4461};
Linus Torvalds1da177e2005-04-16 15:20:36 -07004462EXPORT_SYMBOL(page_symlink_inode_operations);