blob: af3783fff1de66ee922411dac7d5ad92811291d7 [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>
18#include <linux/module.h>
19#include <linux/slab.h>
20#include <linux/fs.h>
21#include <linux/namei.h>
22#include <linux/quotaops.h>
23#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>
27#include <linux/syscalls.h>
28#include <linux/mount.h>
29#include <linux/audit.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080030#include <linux/capability.h>
Trond Myklebust834f2a42005-10-18 14:20:16 -070031#include <linux/file.h>
Ulrich Drepper5590ff02006-01-18 17:43:53 -080032#include <linux/fcntl.h>
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -070033#include <linux/device_cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <asm/uaccess.h>
35
36#define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
37
38/* [Feb-1997 T. Schoebel-Theuer]
39 * Fundamental changes in the pathname lookup mechanisms (namei)
40 * were necessary because of omirr. The reason is that omirr needs
41 * to know the _real_ pathname, not the user-supplied one, in case
42 * of symlinks (and also when transname replacements occur).
43 *
44 * The new code replaces the old recursive symlink resolution with
45 * an iterative one (in case of non-nested symlink chains). It does
46 * this with calls to <fs>_follow_link().
47 * As a side effect, dir_namei(), _namei() and follow_link() are now
48 * replaced with a single function lookup_dentry() that can handle all
49 * the special cases of the former code.
50 *
51 * With the new dcache, the pathname is stored at each inode, at least as
52 * long as the refcount of the inode is positive. As a side effect, the
53 * size of the dcache depends on the inode cache and thus is dynamic.
54 *
55 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
56 * resolution to correspond with current state of the code.
57 *
58 * Note that the symlink resolution is not *completely* iterative.
59 * There is still a significant amount of tail- and mid- recursion in
60 * the algorithm. Also, note that <fs>_readlink() is not used in
61 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
62 * may return different results than <fs>_follow_link(). Many virtual
63 * filesystems (including /proc) exhibit this behavior.
64 */
65
66/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
67 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
68 * and the name already exists in form of a symlink, try to create the new
69 * name indicated by the symlink. The old code always complained that the
70 * name already exists, due to not following the symlink even if its target
71 * is nonexistent. The new semantics affects also mknod() and link() when
72 * the name is a symlink pointing to a non-existant name.
73 *
74 * I don't know which semantics is the right one, since I have no access
75 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
76 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
77 * "old" one. Personally, I think the new semantics is much more logical.
78 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
79 * file does succeed in both HP-UX and SunOs, but not in Solaris
80 * and in the old Linux semantics.
81 */
82
83/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
84 * semantics. See the comments in "open_namei" and "do_link" below.
85 *
86 * [10-Sep-98 Alan Modra] Another symlink change.
87 */
88
89/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
90 * inside the path - always follow.
91 * in the last component in creation/removal/renaming - never follow.
92 * if LOOKUP_FOLLOW passed - follow.
93 * if the pathname has trailing slashes - follow.
94 * otherwise - don't follow.
95 * (applied in that order).
96 *
97 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
98 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
99 * During the 2.4 we need to fix the userland stuff depending on it -
100 * hopefully we will be able to get rid of that wart in 2.5. So far only
101 * XEmacs seems to be relying on it...
102 */
103/*
104 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800105 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 * any extra contention...
107 */
108
Al Viroa02f76c2008-02-23 15:14:28 +0000109static int __link_path_walk(const char *name, struct nameidata *nd);
Josef 'Jeff' Sipekc4a78082007-07-19 01:48:22 -0700110
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111/* In order to reduce some races, while at the same time doing additional
112 * checking and hopefully speeding things up, we copy filenames to the
113 * kernel data space before using them..
114 *
115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
116 * PATH_MAX includes the nul terminator --RR.
117 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800118static int do_getname(const char __user *filename, char *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119{
120 int retval;
121 unsigned long len = PATH_MAX;
122
123 if (!segment_eq(get_fs(), KERNEL_DS)) {
124 if ((unsigned long) filename >= TASK_SIZE)
125 return -EFAULT;
126 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
127 len = TASK_SIZE - (unsigned long) filename;
128 }
129
130 retval = strncpy_from_user(page, filename, len);
131 if (retval > 0) {
132 if (retval < len)
133 return 0;
134 return -ENAMETOOLONG;
135 } else if (!retval)
136 retval = -ENOENT;
137 return retval;
138}
139
140char * getname(const char __user * filename)
141{
142 char *tmp, *result;
143
144 result = ERR_PTR(-ENOMEM);
145 tmp = __getname();
146 if (tmp) {
147 int retval = do_getname(filename, tmp);
148
149 result = tmp;
150 if (retval < 0) {
151 __putname(tmp);
152 result = ERR_PTR(retval);
153 }
154 }
155 audit_getname(result);
156 return result;
157}
158
159#ifdef CONFIG_AUDITSYSCALL
160void putname(const char *name)
161{
Al Viro5ac3a9c2006-07-16 06:38:45 -0400162 if (unlikely(!audit_dummy_context()))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 audit_putname(name);
164 else
165 __putname(name);
166}
167EXPORT_SYMBOL(putname);
168#endif
169
170
171/**
172 * generic_permission - check for access rights on a Posix-like filesystem
173 * @inode: inode to check access rights for
174 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
175 * @check_acl: optional callback to check for Posix ACLs
176 *
177 * Used to check for read/write/execute permissions on a file.
178 * We use "fsuid" for this, letting us set arbitrary permissions
179 * for filesystem access without changing the "normal" uids which
180 * are used for other things..
181 */
182int generic_permission(struct inode *inode, int mask,
183 int (*check_acl)(struct inode *inode, int mask))
184{
185 umode_t mode = inode->i_mode;
186
Al Viroe6305c42008-07-15 21:03:57 -0400187 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
188
David Howellsda9592e2008-11-14 10:39:05 +1100189 if (current_fsuid() == inode->i_uid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 mode >>= 6;
191 else {
192 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
193 int error = check_acl(inode, mask);
194 if (error == -EACCES)
195 goto check_capabilities;
196 else if (error != -EAGAIN)
197 return error;
198 }
199
200 if (in_group_p(inode->i_gid))
201 mode >>= 3;
202 }
203
204 /*
205 * If the DACs are ok we don't need any capability check.
206 */
Al Viroe6305c42008-07-15 21:03:57 -0400207 if ((mask & ~mode) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 return 0;
209
210 check_capabilities:
211 /*
212 * Read/write DACs are always overridable.
213 * Executable DACs are overridable if at least one exec bit is set.
214 */
Miklos Szeredif696a362008-07-31 13:41:58 +0200215 if (!(mask & MAY_EXEC) || execute_ok(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 if (capable(CAP_DAC_OVERRIDE))
217 return 0;
218
219 /*
220 * Searching includes executable on directories, else just read.
221 */
222 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
223 if (capable(CAP_DAC_READ_SEARCH))
224 return 0;
225
226 return -EACCES;
227}
228
Al Virof419a2e2008-07-22 00:07:17 -0400229int inode_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230{
Al Viroe6305c42008-07-15 21:03:57 -0400231 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232
233 if (mask & MAY_WRITE) {
Miklos Szeredi22590e42007-10-16 23:27:08 -0700234 umode_t mode = inode->i_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235
236 /*
237 * Nobody gets write access to a read-only fs.
238 */
239 if (IS_RDONLY(inode) &&
240 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
241 return -EROFS;
242
243 /*
244 * Nobody gets write access to an immutable file.
245 */
246 if (IS_IMMUTABLE(inode))
247 return -EACCES;
248 }
249
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 /* Ordinary permission routines do not understand MAY_APPEND. */
Miklos Szeredif696a362008-07-31 13:41:58 +0200251 if (inode->i_op && inode->i_op->permission)
Al Virob77b0642008-07-17 09:37:02 -0400252 retval = inode->i_op->permission(inode, mask);
Miklos Szeredif696a362008-07-31 13:41:58 +0200253 else
Al Viroe6305c42008-07-15 21:03:57 -0400254 retval = generic_permission(inode, mask, NULL);
Miklos Szeredif696a362008-07-31 13:41:58 +0200255
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 if (retval)
257 return retval;
258
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -0700259 retval = devcgroup_inode_permission(inode, mask);
260 if (retval)
261 return retval;
262
Al Viroe6305c42008-07-15 21:03:57 -0400263 return security_inode_permission(inode,
Stephen Smalleyf418b002008-07-28 13:32:38 -0400264 mask & (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265}
266
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800267/**
268 * vfs_permission - check for access rights to a given path
269 * @nd: lookup result that describes the path
270 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
271 *
272 * Used to check for read/write/execute permissions on a path.
273 * We use "fsuid" for this, letting us set arbitrary permissions
274 * for filesystem access without changing the "normal" uids which
275 * are used for other things.
276 */
277int vfs_permission(struct nameidata *nd, int mask)
278{
Al Virof419a2e2008-07-22 00:07:17 -0400279 return inode_permission(nd->path.dentry->d_inode, mask);
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800280}
281
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800282/**
283 * file_permission - check for additional access rights to a given file
284 * @file: file to check access rights for
285 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
286 *
287 * Used to check for read/write/execute permissions on an already opened
288 * file.
289 *
290 * Note:
291 * Do not use this function in new code. All access checks should
292 * be done using vfs_permission().
293 */
294int file_permission(struct file *file, int mask)
295{
Al Virof419a2e2008-07-22 00:07:17 -0400296 return inode_permission(file->f_path.dentry->d_inode, mask);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800297}
298
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299/*
300 * get_write_access() gets write permission for a file.
301 * put_write_access() releases this write permission.
302 * This is used for regular files.
303 * We cannot support write (and maybe mmap read-write shared) accesses and
304 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
305 * can have the following values:
306 * 0: no writers, no VM_DENYWRITE mappings
307 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
308 * > 0: (i_writecount) users are writing to the file.
309 *
310 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
311 * except for the cases where we don't hold i_writecount yet. Then we need to
312 * use {get,deny}_write_access() - these functions check the sign and refuse
313 * to do the change if sign is wrong. Exclusion between them is provided by
314 * the inode->i_lock spinlock.
315 */
316
317int get_write_access(struct inode * inode)
318{
319 spin_lock(&inode->i_lock);
320 if (atomic_read(&inode->i_writecount) < 0) {
321 spin_unlock(&inode->i_lock);
322 return -ETXTBSY;
323 }
324 atomic_inc(&inode->i_writecount);
325 spin_unlock(&inode->i_lock);
326
327 return 0;
328}
329
330int deny_write_access(struct file * file)
331{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800332 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333
334 spin_lock(&inode->i_lock);
335 if (atomic_read(&inode->i_writecount) > 0) {
336 spin_unlock(&inode->i_lock);
337 return -ETXTBSY;
338 }
339 atomic_dec(&inode->i_writecount);
340 spin_unlock(&inode->i_lock);
341
342 return 0;
343}
344
Jan Blunck1d957f92008-02-14 19:34:35 -0800345/**
Jan Blunck5dd784d2008-02-14 19:34:38 -0800346 * path_get - get a reference to a path
347 * @path: path to get the reference to
348 *
349 * Given a path increment the reference count to the dentry and the vfsmount.
350 */
351void path_get(struct path *path)
352{
353 mntget(path->mnt);
354 dget(path->dentry);
355}
356EXPORT_SYMBOL(path_get);
357
358/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800359 * path_put - put a reference to a path
360 * @path: path to put the reference to
361 *
362 * Given a path decrement the reference count to the dentry and the vfsmount.
363 */
364void path_put(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365{
Jan Blunck1d957f92008-02-14 19:34:35 -0800366 dput(path->dentry);
367 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368}
Jan Blunck1d957f92008-02-14 19:34:35 -0800369EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370
Trond Myklebust834f2a42005-10-18 14:20:16 -0700371/**
372 * release_open_intent - free up open intent resources
373 * @nd: pointer to nameidata
374 */
375void release_open_intent(struct nameidata *nd)
376{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800377 if (nd->intent.open.file->f_path.dentry == NULL)
Trond Myklebust834f2a42005-10-18 14:20:16 -0700378 put_filp(nd->intent.open.file);
379 else
380 fput(nd->intent.open.file);
381}
382
Ian Kentbcdc5e02006-09-27 01:50:44 -0700383static inline struct dentry *
384do_revalidate(struct dentry *dentry, struct nameidata *nd)
385{
386 int status = dentry->d_op->d_revalidate(dentry, nd);
387 if (unlikely(status <= 0)) {
388 /*
389 * The dentry failed validation.
390 * If d_revalidate returned 0 attempt to invalidate
391 * the dentry otherwise d_revalidate is asking us
392 * to return a fail status.
393 */
394 if (!status) {
395 if (!d_invalidate(dentry)) {
396 dput(dentry);
397 dentry = NULL;
398 }
399 } else {
400 dput(dentry);
401 dentry = ERR_PTR(status);
402 }
403 }
404 return dentry;
405}
406
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407/*
408 * Internal lookup() using the new generic dcache.
409 * SMP-safe
410 */
411static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
412{
413 struct dentry * dentry = __d_lookup(parent, name);
414
415 /* lockess __d_lookup may fail due to concurrent d_move()
416 * in some unrelated directory, so try with d_lookup
417 */
418 if (!dentry)
419 dentry = d_lookup(parent, name);
420
Ian Kentbcdc5e02006-09-27 01:50:44 -0700421 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
422 dentry = do_revalidate(dentry, nd);
423
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 return dentry;
425}
426
427/*
428 * Short-cut version of permission(), for calling by
429 * path_walk(), when dcache lock is held. Combines parts
430 * of permission() and generic_permission(), and tests ONLY for
431 * MAY_EXEC permission.
432 *
433 * If appropriate, check DAC only. If not appropriate, or
434 * short-cut DAC fails, then call permission() to do more
435 * complete permission check.
436 */
Al Viro672b16b2008-07-17 09:45:01 -0400437static int exec_permission_lite(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438{
439 umode_t mode = inode->i_mode;
440
441 if (inode->i_op && inode->i_op->permission)
442 return -EAGAIN;
443
David Howellsda9592e2008-11-14 10:39:05 +1100444 if (current_fsuid() == inode->i_uid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 mode >>= 6;
446 else if (in_group_p(inode->i_gid))
447 mode >>= 3;
448
449 if (mode & MAY_EXEC)
450 goto ok;
451
452 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
453 goto ok;
454
455 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
456 goto ok;
457
458 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
459 goto ok;
460
461 return -EACCES;
462ok:
Al Virob77b0642008-07-17 09:37:02 -0400463 return security_inode_permission(inode, MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464}
465
466/*
467 * This is called when everything else fails, and we actually have
468 * to go to the low-level filesystem to find out what we should do..
469 *
470 * We get the directory semaphore, and after getting that we also
471 * make sure that nobody added the entry to the dcache in the meantime..
472 * SMP-safe
473 */
474static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
475{
476 struct dentry * result;
477 struct inode *dir = parent->d_inode;
478
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800479 mutex_lock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 /*
481 * First re-do the cached lookup just in case it was created
482 * while we waited for the directory semaphore..
483 *
484 * FIXME! This could use version numbering or similar to
485 * avoid unnecessary cache lookups.
486 *
487 * The "dcache_lock" is purely to protect the RCU list walker
488 * from concurrent renames at this point (we mustn't get false
489 * negatives from the RCU list walk here, unlike the optimistic
490 * fast walk).
491 *
492 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
493 */
494 result = d_lookup(parent, name);
495 if (!result) {
Miklos Szeredid70b67c2008-07-02 21:30:15 +0200496 struct dentry *dentry;
497
498 /* Don't create child dentry for a dead directory. */
499 result = ERR_PTR(-ENOENT);
500 if (IS_DEADDIR(dir))
501 goto out_unlock;
502
503 dentry = d_alloc(parent, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 result = ERR_PTR(-ENOMEM);
505 if (dentry) {
506 result = dir->i_op->lookup(dir, dentry, nd);
507 if (result)
508 dput(dentry);
509 else
510 result = dentry;
511 }
Miklos Szeredid70b67c2008-07-02 21:30:15 +0200512out_unlock:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800513 mutex_unlock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 return result;
515 }
516
517 /*
518 * Uhhuh! Nasty case: the cache was re-populated while
519 * we waited on the semaphore. Need to revalidate.
520 */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800521 mutex_unlock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 if (result->d_op && result->d_op->d_revalidate) {
Ian Kentbcdc5e02006-09-27 01:50:44 -0700523 result = do_revalidate(result, nd);
524 if (!result)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 result = ERR_PTR(-ENOENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 }
527 return result;
528}
529
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530/* SMP-safe */
Al Viro7f2da1e2008-05-10 20:44:54 -0400531static __always_inline void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532walk_init_root(const char *name, struct nameidata *nd)
533{
Andreas Mohre518ddb2006-09-29 02:01:22 -0700534 struct fs_struct *fs = current->fs;
535
536 read_lock(&fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -0800537 nd->path = fs->root;
538 path_get(&fs->root);
Andreas Mohre518ddb2006-09-29 02:01:22 -0700539 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540}
541
Al Viroa02f76c2008-02-23 15:14:28 +0000542/*
543 * Wrapper to retry pathname resolution whenever the underlying
544 * file system returns an ESTALE.
545 *
546 * Retry the whole path once, forcing real lookup requests
547 * instead of relying on the dcache.
548 */
549static __always_inline int link_path_walk(const char *name, struct nameidata *nd)
550{
551 struct path save = nd->path;
552 int result;
553
554 /* make sure the stuff we saved doesn't go away */
Jan Blunckc8e7f442008-06-09 16:40:35 -0700555 path_get(&save);
Al Viroa02f76c2008-02-23 15:14:28 +0000556
557 result = __link_path_walk(name, nd);
558 if (result == -ESTALE) {
559 /* nd->path had been dropped */
560 nd->path = save;
Jan Blunckc8e7f442008-06-09 16:40:35 -0700561 path_get(&nd->path);
Al Viroa02f76c2008-02-23 15:14:28 +0000562 nd->flags |= LOOKUP_REVAL;
563 result = __link_path_walk(name, nd);
564 }
565
566 path_put(&save);
567
568 return result;
569}
570
Arjan van de Venf1662352006-01-14 13:21:31 -0800571static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572{
573 int res = 0;
574 char *name;
575 if (IS_ERR(link))
576 goto fail;
577
578 if (*link == '/') {
Jan Blunck1d957f92008-02-14 19:34:35 -0800579 path_put(&nd->path);
Al Viro7f2da1e2008-05-10 20:44:54 -0400580 walk_init_root(link, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 }
582 res = link_path_walk(link, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 if (nd->depth || res || nd->last_type!=LAST_NORM)
584 return res;
585 /*
586 * If it is an iterative symlinks resolution in open_namei() we
587 * have to copy the last component. And all that crap because of
588 * bloody create() on broken symlinks. Furrfu...
589 */
590 name = __getname();
591 if (unlikely(!name)) {
Jan Blunck1d957f92008-02-14 19:34:35 -0800592 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 return -ENOMEM;
594 }
595 strcpy(name, nd->last.name);
596 nd->last.name = name;
597 return 0;
598fail:
Jan Blunck1d957f92008-02-14 19:34:35 -0800599 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 return PTR_ERR(link);
601}
602
Jan Blunck1d957f92008-02-14 19:34:35 -0800603static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700604{
605 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800606 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700607 mntput(path->mnt);
608}
609
610static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
611{
Jan Blunck4ac91372008-02-14 19:34:32 -0800612 dput(nd->path.dentry);
613 if (nd->path.mnt != path->mnt)
614 mntput(nd->path.mnt);
615 nd->path.mnt = path->mnt;
616 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700617}
618
Ian Kent051d3812006-03-27 01:14:53 -0800619static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
620{
621 int error;
622 void *cookie;
623 struct dentry *dentry = path->dentry;
624
625 touch_atime(path->mnt, dentry);
626 nd_set_link(nd, NULL);
627
Jan Blunck4ac91372008-02-14 19:34:32 -0800628 if (path->mnt != nd->path.mnt) {
Ian Kent051d3812006-03-27 01:14:53 -0800629 path_to_nameidata(path, nd);
630 dget(dentry);
631 }
632 mntget(path->mnt);
633 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
634 error = PTR_ERR(cookie);
635 if (!IS_ERR(cookie)) {
636 char *s = nd_get_link(nd);
637 error = 0;
638 if (s)
639 error = __vfs_follow_link(nd, s);
640 if (dentry->d_inode->i_op->put_link)
641 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
642 }
Jan Blunck09da59162008-02-14 19:34:37 -0800643 path_put(path);
Ian Kent051d3812006-03-27 01:14:53 -0800644
645 return error;
646}
647
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648/*
649 * This limits recursive symlink follows to 8, while
650 * limiting consecutive symlinks to 40.
651 *
652 * Without that kind of total limit, nasty chains of consecutive
653 * symlinks can cause almost arbitrarily long lookups.
654 */
Al Viro90ebe562005-06-06 13:35:58 -0700655static inline int do_follow_link(struct path *path, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656{
657 int err = -ELOOP;
658 if (current->link_count >= MAX_NESTED_LINKS)
659 goto loop;
660 if (current->total_link_count >= 40)
661 goto loop;
662 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
663 cond_resched();
Al Viro90ebe562005-06-06 13:35:58 -0700664 err = security_inode_follow_link(path->dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 if (err)
666 goto loop;
667 current->link_count++;
668 current->total_link_count++;
669 nd->depth++;
Al Virocd4e91d2005-06-06 13:36:03 -0700670 err = __do_follow_link(path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 current->link_count--;
672 nd->depth--;
673 return err;
674loop:
Jan Blunck1d957f92008-02-14 19:34:35 -0800675 path_put_conditional(path, nd);
676 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 return err;
678}
679
680int follow_up(struct vfsmount **mnt, struct dentry **dentry)
681{
682 struct vfsmount *parent;
683 struct dentry *mountpoint;
684 spin_lock(&vfsmount_lock);
685 parent=(*mnt)->mnt_parent;
686 if (parent == *mnt) {
687 spin_unlock(&vfsmount_lock);
688 return 0;
689 }
690 mntget(parent);
691 mountpoint=dget((*mnt)->mnt_mountpoint);
692 spin_unlock(&vfsmount_lock);
693 dput(*dentry);
694 *dentry = mountpoint;
695 mntput(*mnt);
696 *mnt = parent;
697 return 1;
698}
699
700/* no need for dcache_lock, as serialization is taken care in
701 * namespace.c
702 */
Al Viro463ffb22005-06-06 13:36:05 -0700703static int __follow_mount(struct path *path)
704{
705 int res = 0;
706 while (d_mountpoint(path->dentry)) {
707 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
708 if (!mounted)
709 break;
710 dput(path->dentry);
711 if (res)
712 mntput(path->mnt);
713 path->mnt = mounted;
714 path->dentry = dget(mounted->mnt_root);
715 res = 1;
716 }
717 return res;
718}
719
Al Viro58c465e2005-06-06 13:36:13 -0700720static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 while (d_mountpoint(*dentry)) {
723 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
724 if (!mounted)
725 break;
Al Viro58c465e2005-06-06 13:36:13 -0700726 dput(*dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 mntput(*mnt);
728 *mnt = mounted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 *dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731}
732
733/* no need for dcache_lock, as serialization is taken care in
734 * namespace.c
735 */
Al Viroe13b2102005-06-06 13:36:06 -0700736int follow_down(struct vfsmount **mnt, struct dentry **dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737{
738 struct vfsmount *mounted;
739
740 mounted = lookup_mnt(*mnt, *dentry);
741 if (mounted) {
Al Viroe13b2102005-06-06 13:36:06 -0700742 dput(*dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 mntput(*mnt);
744 *mnt = mounted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 *dentry = dget(mounted->mnt_root);
746 return 1;
747 }
748 return 0;
749}
750
Arjan van de Venf1662352006-01-14 13:21:31 -0800751static __always_inline void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752{
Andreas Mohre518ddb2006-09-29 02:01:22 -0700753 struct fs_struct *fs = current->fs;
754
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 while(1) {
756 struct vfsmount *parent;
Jan Blunck4ac91372008-02-14 19:34:32 -0800757 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758
Andreas Mohre518ddb2006-09-29 02:01:22 -0700759 read_lock(&fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -0800760 if (nd->path.dentry == fs->root.dentry &&
761 nd->path.mnt == fs->root.mnt) {
Andreas Mohre518ddb2006-09-29 02:01:22 -0700762 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 break;
764 }
Andreas Mohre518ddb2006-09-29 02:01:22 -0700765 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 spin_lock(&dcache_lock);
Jan Blunck4ac91372008-02-14 19:34:32 -0800767 if (nd->path.dentry != nd->path.mnt->mnt_root) {
768 nd->path.dentry = dget(nd->path.dentry->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 spin_unlock(&dcache_lock);
770 dput(old);
771 break;
772 }
773 spin_unlock(&dcache_lock);
774 spin_lock(&vfsmount_lock);
Jan Blunck4ac91372008-02-14 19:34:32 -0800775 parent = nd->path.mnt->mnt_parent;
776 if (parent == nd->path.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 spin_unlock(&vfsmount_lock);
778 break;
779 }
780 mntget(parent);
Jan Blunck4ac91372008-02-14 19:34:32 -0800781 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 spin_unlock(&vfsmount_lock);
783 dput(old);
Jan Blunck4ac91372008-02-14 19:34:32 -0800784 mntput(nd->path.mnt);
785 nd->path.mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 }
Jan Blunck4ac91372008-02-14 19:34:32 -0800787 follow_mount(&nd->path.mnt, &nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788}
789
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790/*
791 * It's more convoluted than I'd like it to be, but... it's still fairly
792 * small and for now I'd prefer to have fast path as straight as possible.
793 * It _is_ time-critical.
794 */
795static int do_lookup(struct nameidata *nd, struct qstr *name,
796 struct path *path)
797{
Jan Blunck4ac91372008-02-14 19:34:32 -0800798 struct vfsmount *mnt = nd->path.mnt;
799 struct dentry *dentry = __d_lookup(nd->path.dentry, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800
801 if (!dentry)
802 goto need_lookup;
803 if (dentry->d_op && dentry->d_op->d_revalidate)
804 goto need_revalidate;
805done:
806 path->mnt = mnt;
807 path->dentry = dentry;
Al Viro634ee702005-06-06 13:36:13 -0700808 __follow_mount(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 return 0;
810
811need_lookup:
Jan Blunck4ac91372008-02-14 19:34:32 -0800812 dentry = real_lookup(nd->path.dentry, name, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 if (IS_ERR(dentry))
814 goto fail;
815 goto done;
816
817need_revalidate:
Ian Kentbcdc5e02006-09-27 01:50:44 -0700818 dentry = do_revalidate(dentry, nd);
819 if (!dentry)
820 goto need_lookup;
821 if (IS_ERR(dentry))
822 goto fail;
823 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824
825fail:
826 return PTR_ERR(dentry);
827}
828
829/*
830 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +0100831 * This is the basic name resolution function, turning a pathname into
832 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 *
Prasanna Medaea3834d2005-04-29 16:00:17 +0100834 * Returns 0 and nd will have valid dentry and mnt on success.
835 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800837static int __link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838{
839 struct path next;
840 struct inode *inode;
841 int err;
842 unsigned int lookup_flags = nd->flags;
843
844 while (*name=='/')
845 name++;
846 if (!*name)
847 goto return_reval;
848
Jan Blunck4ac91372008-02-14 19:34:32 -0800849 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 if (nd->depth)
Trond Myklebustf55eab82006-02-04 23:28:01 -0800851 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852
853 /* At this point we know we have a real path component. */
854 for(;;) {
855 unsigned long hash;
856 struct qstr this;
857 unsigned int c;
858
Trond Myklebustcdce5d62005-10-18 14:20:18 -0700859 nd->flags |= LOOKUP_CONTINUE;
Al Viro672b16b2008-07-17 09:45:01 -0400860 err = exec_permission_lite(inode);
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800861 if (err == -EAGAIN)
862 err = vfs_permission(nd, MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 if (err)
864 break;
865
866 this.name = name;
867 c = *(const unsigned char *)name;
868
869 hash = init_name_hash();
870 do {
871 name++;
872 hash = partial_name_hash(c, hash);
873 c = *(const unsigned char *)name;
874 } while (c && (c != '/'));
875 this.len = name - (const char *) this.name;
876 this.hash = end_name_hash(hash);
877
878 /* remove trailing slashes? */
879 if (!c)
880 goto last_component;
881 while (*++name == '/');
882 if (!*name)
883 goto last_with_slashes;
884
885 /*
886 * "." and ".." are special - ".." especially so because it has
887 * to be able to know about the current root directory and
888 * parent relationships.
889 */
890 if (this.name[0] == '.') switch (this.len) {
891 default:
892 break;
893 case 2:
894 if (this.name[1] != '.')
895 break;
Al Viro58c465e2005-06-06 13:36:13 -0700896 follow_dotdot(nd);
Jan Blunck4ac91372008-02-14 19:34:32 -0800897 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 /* fallthrough */
899 case 1:
900 continue;
901 }
902 /*
903 * See if the low-level filesystem might want
904 * to use its own hash..
905 */
Jan Blunck4ac91372008-02-14 19:34:32 -0800906 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
907 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
908 &this);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 if (err < 0)
910 break;
911 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 /* This does the actual lookups.. */
913 err = do_lookup(nd, &this, &next);
914 if (err)
915 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916
917 err = -ENOENT;
918 inode = next.dentry->d_inode;
919 if (!inode)
920 goto out_dput;
921 err = -ENOTDIR;
922 if (!inode->i_op)
923 goto out_dput;
924
925 if (inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700926 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 if (err)
928 goto return_err;
929 err = -ENOENT;
Jan Blunck4ac91372008-02-14 19:34:32 -0800930 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 if (!inode)
932 break;
933 err = -ENOTDIR;
934 if (!inode->i_op)
935 break;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700936 } else
937 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 err = -ENOTDIR;
939 if (!inode->i_op->lookup)
940 break;
941 continue;
942 /* here ends the main loop */
943
944last_with_slashes:
945 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
946last_component:
Trond Myklebustf55eab82006-02-04 23:28:01 -0800947 /* Clear LOOKUP_CONTINUE iff it was previously unset */
948 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 if (lookup_flags & LOOKUP_PARENT)
950 goto lookup_parent;
951 if (this.name[0] == '.') switch (this.len) {
952 default:
953 break;
954 case 2:
955 if (this.name[1] != '.')
956 break;
Al Viro58c465e2005-06-06 13:36:13 -0700957 follow_dotdot(nd);
Jan Blunck4ac91372008-02-14 19:34:32 -0800958 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 /* fallthrough */
960 case 1:
961 goto return_reval;
962 }
Jan Blunck4ac91372008-02-14 19:34:32 -0800963 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
964 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
965 &this);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 if (err < 0)
967 break;
968 }
969 err = do_lookup(nd, &this, &next);
970 if (err)
971 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 inode = next.dentry->d_inode;
973 if ((lookup_flags & LOOKUP_FOLLOW)
974 && inode && inode->i_op && inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700975 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 if (err)
977 goto return_err;
Jan Blunck4ac91372008-02-14 19:34:32 -0800978 inode = nd->path.dentry->d_inode;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700979 } else
980 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 err = -ENOENT;
982 if (!inode)
983 break;
984 if (lookup_flags & LOOKUP_DIRECTORY) {
985 err = -ENOTDIR;
986 if (!inode->i_op || !inode->i_op->lookup)
987 break;
988 }
989 goto return_base;
990lookup_parent:
991 nd->last = this;
992 nd->last_type = LAST_NORM;
993 if (this.name[0] != '.')
994 goto return_base;
995 if (this.len == 1)
996 nd->last_type = LAST_DOT;
997 else if (this.len == 2 && this.name[1] == '.')
998 nd->last_type = LAST_DOTDOT;
999 else
1000 goto return_base;
1001return_reval:
1002 /*
1003 * We bypassed the ordinary revalidation routines.
1004 * We may need to check the cached dentry for staleness.
1005 */
Jan Blunck4ac91372008-02-14 19:34:32 -08001006 if (nd->path.dentry && nd->path.dentry->d_sb &&
1007 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 err = -ESTALE;
1009 /* Note: we do not d_invalidate() */
Jan Blunck4ac91372008-02-14 19:34:32 -08001010 if (!nd->path.dentry->d_op->d_revalidate(
1011 nd->path.dentry, nd))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 break;
1013 }
1014return_base:
1015 return 0;
1016out_dput:
Jan Blunck1d957f92008-02-14 19:34:35 -08001017 path_put_conditional(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 break;
1019 }
Jan Blunck1d957f92008-02-14 19:34:35 -08001020 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021return_err:
1022 return err;
1023}
1024
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -08001025static int path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026{
1027 current->total_link_count = 0;
1028 return link_path_walk(name, nd);
1029}
1030
Prasanna Medaea3834d2005-04-29 16:00:17 +01001031/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -08001032static int do_path_lookup(int dfd, const char *name,
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001033 unsigned int flags, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001035 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001036 int fput_needed;
1037 struct file *file;
Andreas Mohre518ddb2006-09-29 02:01:22 -07001038 struct fs_struct *fs = current->fs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039
1040 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1041 nd->flags = flags;
1042 nd->depth = 0;
1043
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 if (*name=='/') {
Andreas Mohre518ddb2006-09-29 02:01:22 -07001045 read_lock(&fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08001046 nd->path = fs->root;
1047 path_get(&fs->root);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001048 read_unlock(&fs->lock);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001049 } else if (dfd == AT_FDCWD) {
Andreas Mohre518ddb2006-09-29 02:01:22 -07001050 read_lock(&fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08001051 nd->path = fs->pwd;
1052 path_get(&fs->pwd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001053 read_unlock(&fs->lock);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001054 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001055 struct dentry *dentry;
1056
1057 file = fget_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001058 retval = -EBADF;
1059 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001060 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001061
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001062 dentry = file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001063
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001064 retval = -ENOTDIR;
1065 if (!S_ISDIR(dentry->d_inode->i_mode))
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001066 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001067
1068 retval = file_permission(file, MAY_EXEC);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001069 if (retval)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001070 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001071
Jan Blunck5dd784d2008-02-14 19:34:38 -08001072 nd->path = file->f_path;
1073 path_get(&file->f_path);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001074
1075 fput_light(file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 }
Josef 'Jeff' Sipek2dfdd262007-05-09 02:33:41 -07001077
1078 retval = path_walk(name, nd);
Jan Blunck4ac91372008-02-14 19:34:32 -08001079 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1080 nd->path.dentry->d_inode))
1081 audit_inode(name, nd->path.dentry);
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001082out_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001083 return retval;
1084
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001085fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001086 fput_light(file, fput_needed);
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001087 goto out_fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088}
1089
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -08001090int path_lookup(const char *name, unsigned int flags,
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001091 struct nameidata *nd)
1092{
1093 return do_path_lookup(AT_FDCWD, name, flags, nd);
1094}
1095
Al Virod1811462008-08-02 00:49:18 -04001096int kern_path(const char *name, unsigned int flags, struct path *path)
1097{
1098 struct nameidata nd;
1099 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1100 if (!res)
1101 *path = nd.path;
1102 return res;
1103}
1104
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001105/**
1106 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1107 * @dentry: pointer to dentry of the base directory
1108 * @mnt: pointer to vfs mount of the base directory
1109 * @name: pointer to file name
1110 * @flags: lookup flags
1111 * @nd: pointer to nameidata
1112 */
1113int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1114 const char *name, unsigned int flags,
1115 struct nameidata *nd)
1116{
1117 int retval;
1118
1119 /* same as do_path_lookup */
1120 nd->last_type = LAST_ROOT;
1121 nd->flags = flags;
1122 nd->depth = 0;
1123
Jan Blunckc8e7f442008-06-09 16:40:35 -07001124 nd->path.dentry = dentry;
1125 nd->path.mnt = mnt;
1126 path_get(&nd->path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001127
1128 retval = path_walk(name, nd);
Jan Blunck4ac91372008-02-14 19:34:32 -08001129 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1130 nd->path.dentry->d_inode))
1131 audit_inode(name, nd->path.dentry);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001132
1133 return retval;
1134
1135}
1136
Trond Myklebust834f2a42005-10-18 14:20:16 -07001137/**
1138 * path_lookup_open - lookup a file path with open intent
Martin Waitz7045f372006-02-01 03:06:57 -08001139 * @dfd: the directory to use as base, or AT_FDCWD
Trond Myklebust834f2a42005-10-18 14:20:16 -07001140 * @name: pointer to file name
1141 * @lookup_flags: lookup intent flags
1142 * @nd: pointer to nameidata
1143 * @open_flags: open intent flags
1144 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001145int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
Trond Myklebust834f2a42005-10-18 14:20:16 -07001146 struct nameidata *nd, int open_flags)
1147{
Al Viro8737f3a2008-08-02 22:36:57 -04001148 struct file *filp = get_empty_filp();
1149 int err;
Trond Myklebust834f2a42005-10-18 14:20:16 -07001150
Al Viro8737f3a2008-08-02 22:36:57 -04001151 if (filp == NULL)
1152 return -ENFILE;
1153 nd->intent.open.file = filp;
1154 nd->intent.open.flags = open_flags;
1155 nd->intent.open.create_mode = 0;
1156 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1157 if (IS_ERR(nd->intent.open.file)) {
1158 if (err == 0) {
1159 err = PTR_ERR(nd->intent.open.file);
1160 path_put(&nd->path);
1161 }
1162 } else if (err != 0)
1163 release_open_intent(nd);
1164 return err;
Trond Myklebust834f2a42005-10-18 14:20:16 -07001165}
1166
Christoph Hellwigeead1912007-10-16 23:25:38 -07001167static struct dentry *__lookup_hash(struct qstr *name,
1168 struct dentry *base, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169{
James Morris057f6c02007-04-26 00:12:05 -07001170 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 struct inode *inode;
1172 int err;
1173
1174 inode = base->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175
1176 /*
1177 * See if the low-level filesystem might want
1178 * to use its own hash..
1179 */
1180 if (base->d_op && base->d_op->d_hash) {
1181 err = base->d_op->d_hash(base, name);
1182 dentry = ERR_PTR(err);
1183 if (err < 0)
1184 goto out;
1185 }
1186
1187 dentry = cached_lookup(base, name, nd);
1188 if (!dentry) {
Miklos Szeredid70b67c2008-07-02 21:30:15 +02001189 struct dentry *new;
1190
1191 /* Don't create child dentry for a dead directory. */
1192 dentry = ERR_PTR(-ENOENT);
1193 if (IS_DEADDIR(inode))
1194 goto out;
1195
1196 new = d_alloc(base, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 dentry = ERR_PTR(-ENOMEM);
1198 if (!new)
1199 goto out;
1200 dentry = inode->i_op->lookup(inode, new, nd);
1201 if (!dentry)
1202 dentry = new;
1203 else
1204 dput(new);
1205 }
1206out:
1207 return dentry;
1208}
1209
James Morris057f6c02007-04-26 00:12:05 -07001210/*
1211 * Restricted form of lookup. Doesn't follow links, single-component only,
1212 * needs parent already locked. Doesn't follow mounts.
1213 * SMP-safe.
1214 */
Adrian Bunka244e162006-03-31 02:32:11 -08001215static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216{
Christoph Hellwigeead1912007-10-16 23:25:38 -07001217 int err;
1218
Al Virof419a2e2008-07-22 00:07:17 -04001219 err = inode_permission(nd->path.dentry->d_inode, MAY_EXEC);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001220 if (err)
1221 return ERR_PTR(err);
Jan Blunck4ac91372008-02-14 19:34:32 -08001222 return __lookup_hash(&nd->last, nd->path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223}
1224
Christoph Hellwigeead1912007-10-16 23:25:38 -07001225static int __lookup_one_len(const char *name, struct qstr *this,
1226 struct dentry *base, int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227{
1228 unsigned long hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 unsigned int c;
1230
James Morris057f6c02007-04-26 00:12:05 -07001231 this->name = name;
1232 this->len = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 if (!len)
James Morris057f6c02007-04-26 00:12:05 -07001234 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235
1236 hash = init_name_hash();
1237 while (len--) {
1238 c = *(const unsigned char *)name++;
1239 if (c == '/' || c == '\0')
James Morris057f6c02007-04-26 00:12:05 -07001240 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 hash = partial_name_hash(c, hash);
1242 }
James Morris057f6c02007-04-26 00:12:05 -07001243 this->hash = end_name_hash(hash);
1244 return 0;
1245}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246
Christoph Hellwigeead1912007-10-16 23:25:38 -07001247/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001248 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07001249 * @name: pathname component to lookup
1250 * @base: base directory to lookup from
1251 * @len: maximum length @len should be interpreted to
1252 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07001253 * Note that this routine is purely a helper for filesystem usage and should
1254 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07001255 * nameidata argument is passed to the filesystem methods and a filesystem
1256 * using this helper needs to be prepared for that.
1257 */
James Morris057f6c02007-04-26 00:12:05 -07001258struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1259{
1260 int err;
1261 struct qstr this;
1262
1263 err = __lookup_one_len(name, &this, base, len);
1264 if (err)
1265 return ERR_PTR(err);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001266
Al Virof419a2e2008-07-22 00:07:17 -04001267 err = inode_permission(base->d_inode, MAY_EXEC);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001268 if (err)
1269 return ERR_PTR(err);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001270 return __lookup_hash(&this, base, NULL);
James Morris057f6c02007-04-26 00:12:05 -07001271}
1272
Christoph Hellwigeead1912007-10-16 23:25:38 -07001273/**
1274 * lookup_one_noperm - bad hack for sysfs
1275 * @name: pathname component to lookup
1276 * @base: base directory to lookup from
1277 *
1278 * This is a variant of lookup_one_len that doesn't perform any permission
1279 * checks. It's a horrible hack to work around the braindead sysfs
1280 * architecture and should not be used anywhere else.
1281 *
1282 * DON'T USE THIS FUNCTION EVER, thanks.
1283 */
1284struct dentry *lookup_one_noperm(const char *name, struct dentry *base)
James Morris057f6c02007-04-26 00:12:05 -07001285{
1286 int err;
1287 struct qstr this;
1288
Christoph Hellwigeead1912007-10-16 23:25:38 -07001289 err = __lookup_one_len(name, &this, base, strlen(name));
James Morris057f6c02007-04-26 00:12:05 -07001290 if (err)
1291 return ERR_PTR(err);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001292 return __lookup_hash(&this, base, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293}
1294
Al Viro2d8f3032008-07-22 09:59:21 -04001295int user_path_at(int dfd, const char __user *name, unsigned flags,
1296 struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297{
Al Viro2d8f3032008-07-22 09:59:21 -04001298 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 char *tmp = getname(name);
1300 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04001302
1303 BUG_ON(flags & LOOKUP_PARENT);
1304
1305 err = do_path_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04001307 if (!err)
1308 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 }
1310 return err;
1311}
1312
Al Viro2ad94ae2008-07-21 09:32:51 -04001313static int user_path_parent(int dfd, const char __user *path,
1314 struct nameidata *nd, char **name)
1315{
1316 char *s = getname(path);
1317 int error;
1318
1319 if (IS_ERR(s))
1320 return PTR_ERR(s);
1321
1322 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1323 if (error)
1324 putname(s);
1325 else
1326 *name = s;
1327
1328 return error;
1329}
1330
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331/*
1332 * It's inline, so penalty for filesystems that don't use sticky bit is
1333 * minimal.
1334 */
1335static inline int check_sticky(struct inode *dir, struct inode *inode)
1336{
David Howellsda9592e2008-11-14 10:39:05 +11001337 uid_t fsuid = current_fsuid();
1338
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 if (!(dir->i_mode & S_ISVTX))
1340 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001341 if (inode->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001343 if (dir->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 return 0;
1345 return !capable(CAP_FOWNER);
1346}
1347
1348/*
1349 * Check whether we can remove a link victim from directory dir, check
1350 * whether the type of victim is right.
1351 * 1. We can't do it if dir is read-only (done in permission())
1352 * 2. We should have write and exec permissions on dir
1353 * 3. We can't remove anything from append-only dir
1354 * 4. We can't do anything with immutable dir (done in permission())
1355 * 5. If the sticky bit on dir is set we should either
1356 * a. be owner of dir, or
1357 * b. be owner of victim, or
1358 * c. have CAP_FOWNER capability
1359 * 6. If the victim is append-only or immutable we can't do antyhing with
1360 * links pointing to it.
1361 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1362 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1363 * 9. We can't remove a root or mountpoint.
1364 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1365 * nfs_async_unlink().
1366 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001367static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368{
1369 int error;
1370
1371 if (!victim->d_inode)
1372 return -ENOENT;
1373
1374 BUG_ON(victim->d_parent->d_inode != dir);
Al Viro5a190ae2007-06-07 12:19:32 -04001375 audit_inode_child(victim->d_name.name, victim, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376
Al Virof419a2e2008-07-22 00:07:17 -04001377 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 if (error)
1379 return error;
1380 if (IS_APPEND(dir))
1381 return -EPERM;
1382 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
Hugh Dickinsf9454542008-11-19 15:36:38 -08001383 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 return -EPERM;
1385 if (isdir) {
1386 if (!S_ISDIR(victim->d_inode->i_mode))
1387 return -ENOTDIR;
1388 if (IS_ROOT(victim))
1389 return -EBUSY;
1390 } else if (S_ISDIR(victim->d_inode->i_mode))
1391 return -EISDIR;
1392 if (IS_DEADDIR(dir))
1393 return -ENOENT;
1394 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1395 return -EBUSY;
1396 return 0;
1397}
1398
1399/* Check whether we can create an object with dentry child in directory
1400 * dir.
1401 * 1. We can't do it if child already exists (open has special treatment for
1402 * this case, but since we are inlined it's OK)
1403 * 2. We can't do it if dir is read-only (done in permission())
1404 * 3. We should have write and exec permissions on dir
1405 * 4. We can't do it if dir is immutable (done in permission())
1406 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02001407static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408{
1409 if (child->d_inode)
1410 return -EEXIST;
1411 if (IS_DEADDIR(dir))
1412 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04001413 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414}
1415
1416/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 * O_DIRECTORY translates into forcing a directory lookup.
1418 */
1419static inline int lookup_flags(unsigned int f)
1420{
1421 unsigned long retval = LOOKUP_FOLLOW;
1422
1423 if (f & O_NOFOLLOW)
1424 retval &= ~LOOKUP_FOLLOW;
1425
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 if (f & O_DIRECTORY)
1427 retval |= LOOKUP_DIRECTORY;
1428
1429 return retval;
1430}
1431
1432/*
1433 * p1 and p2 should be directories on the same fs.
1434 */
1435struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1436{
1437 struct dentry *p;
1438
1439 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001440 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 return NULL;
1442 }
1443
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001444 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001446 p = d_ancestor(p2, p1);
1447 if (p) {
1448 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1449 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1450 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 }
1452
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001453 p = d_ancestor(p1, p2);
1454 if (p) {
1455 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1456 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1457 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 }
1459
Ingo Molnarf2eace22006-07-03 00:25:05 -07001460 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1461 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 return NULL;
1463}
1464
1465void unlock_rename(struct dentry *p1, struct dentry *p2)
1466{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001467 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001469 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001470 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 }
1472}
1473
1474int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1475 struct nameidata *nd)
1476{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001477 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478
1479 if (error)
1480 return error;
1481
1482 if (!dir->i_op || !dir->i_op->create)
1483 return -EACCES; /* shouldn't it be ENOSYS? */
1484 mode &= S_IALLUGO;
1485 mode |= S_IFREG;
1486 error = security_inode_create(dir, dentry, mode);
1487 if (error)
1488 return error;
1489 DQUOT_INIT(dir);
1490 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001491 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001492 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 return error;
1494}
1495
1496int may_open(struct nameidata *nd, int acc_mode, int flag)
1497{
Jan Blunck4ac91372008-02-14 19:34:32 -08001498 struct dentry *dentry = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 struct inode *inode = dentry->d_inode;
1500 int error;
1501
1502 if (!inode)
1503 return -ENOENT;
1504
1505 if (S_ISLNK(inode->i_mode))
1506 return -ELOOP;
1507
Linus Torvalds974a9f02008-01-12 14:06:34 -08001508 if (S_ISDIR(inode->i_mode) && (acc_mode & MAY_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 return -EISDIR;
1510
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 /*
1512 * FIFO's, sockets and device files are special: they don't
1513 * actually live on the filesystem itself, and as such you
1514 * can write to them even if the filesystem is read-only.
1515 */
1516 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1517 flag &= ~O_TRUNC;
1518 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
Jan Blunck4ac91372008-02-14 19:34:32 -08001519 if (nd->path.mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 return -EACCES;
1521
1522 flag &= ~O_TRUNC;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001523 }
Dave Hansenb41572e2007-10-16 23:31:14 -07001524
1525 error = vfs_permission(nd, acc_mode);
1526 if (error)
1527 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 /*
1529 * An append-only file must be opened in append mode for writing.
1530 */
1531 if (IS_APPEND(inode)) {
1532 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1533 return -EPERM;
1534 if (flag & O_TRUNC)
1535 return -EPERM;
1536 }
1537
1538 /* O_NOATIME can only be set by the owner or superuser */
1539 if (flag & O_NOATIME)
Satyam Sharma3bd858a2007-07-17 15:00:08 +05301540 if (!is_owner_or_cap(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 return -EPERM;
1542
1543 /*
1544 * Ensure there are no outstanding leases on the file.
1545 */
1546 error = break_lease(inode, flag);
1547 if (error)
1548 return error;
1549
1550 if (flag & O_TRUNC) {
1551 error = get_write_access(inode);
1552 if (error)
1553 return error;
1554
1555 /*
1556 * Refuse to truncate files with mandatory locks held on them.
1557 */
1558 error = locks_verify_locked(inode);
1559 if (!error) {
1560 DQUOT_INIT(inode);
Miklos Szeredid139d7f2007-10-18 03:07:00 -07001561
1562 error = do_truncate(dentry, 0,
1563 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1564 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 }
1566 put_write_access(inode);
1567 if (error)
1568 return error;
1569 } else
1570 if (flag & FMODE_WRITE)
1571 DQUOT_INIT(inode);
1572
1573 return 0;
1574}
1575
Dave Hansend57999e2008-02-15 14:37:27 -08001576/*
1577 * Be careful about ever adding any more callers of this
1578 * function. Its flags must be in the namei format, not
1579 * what get passed to sys_open().
1580 */
1581static int __open_namei_create(struct nameidata *nd, struct path *path,
Dave Hansenaab520e2006-09-30 23:29:02 -07001582 int flag, int mode)
1583{
1584 int error;
Jan Blunck4ac91372008-02-14 19:34:32 -08001585 struct dentry *dir = nd->path.dentry;
Dave Hansenaab520e2006-09-30 23:29:02 -07001586
1587 if (!IS_POSIXACL(dir->d_inode))
1588 mode &= ~current->fs->umask;
1589 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1590 mutex_unlock(&dir->d_inode->i_mutex);
Jan Blunck4ac91372008-02-14 19:34:32 -08001591 dput(nd->path.dentry);
1592 nd->path.dentry = path->dentry;
Dave Hansenaab520e2006-09-30 23:29:02 -07001593 if (error)
1594 return error;
1595 /* Don't check for write permission, don't truncate */
1596 return may_open(nd, 0, flag & ~O_TRUNC);
1597}
1598
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599/*
Dave Hansend57999e2008-02-15 14:37:27 -08001600 * Note that while the flag value (low two bits) for sys_open means:
1601 * 00 - read-only
1602 * 01 - write-only
1603 * 10 - read-write
1604 * 11 - special
1605 * it is changed into
1606 * 00 - no permissions needed
1607 * 01 - read-permission
1608 * 10 - write-permission
1609 * 11 - read-write
1610 * for the internal routines (ie open_namei()/follow_link() etc)
1611 * This is more logical, and also allows the 00 "no perm needed"
1612 * to be used for symlinks (where the permissions are checked
1613 * later).
1614 *
1615*/
1616static inline int open_to_namei_flags(int flag)
1617{
1618 if ((flag+1) & O_ACCMODE)
1619 flag++;
1620 return flag;
1621}
1622
Dave Hansen4a3fd212008-02-15 14:37:48 -08001623static int open_will_write_to_fs(int flag, struct inode *inode)
1624{
1625 /*
1626 * We'll never write to the fs underlying
1627 * a device file.
1628 */
1629 if (special_file(inode->i_mode))
1630 return 0;
1631 return (flag & O_TRUNC);
1632}
1633
Dave Hansend57999e2008-02-15 14:37:27 -08001634/*
Dave Hansen4a3fd212008-02-15 14:37:48 -08001635 * Note that the low bits of the passed in "open_flag"
1636 * are not the same as in the local variable "flag". See
1637 * open_to_namei_flags() for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001639struct file *do_filp_open(int dfd, const char *pathname,
1640 int open_flag, int mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641{
Dave Hansen4a3fd212008-02-15 14:37:48 -08001642 struct file *filp;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001643 struct nameidata nd;
Trond Myklebust834f2a42005-10-18 14:20:16 -07001644 int acc_mode, error;
Al Viro4e7506e2005-06-06 13:36:00 -07001645 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 struct dentry *dir;
1647 int count = 0;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001648 int will_write;
Dave Hansend57999e2008-02-15 14:37:27 -08001649 int flag = open_to_namei_flags(open_flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650
Al Virob77b0642008-07-17 09:37:02 -04001651 acc_mode = MAY_OPEN | ACC_MODE(flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652
Trond Myklebust834f2a42005-10-18 14:20:16 -07001653 /* O_TRUNC implies we need access checks for write permissions */
1654 if (flag & O_TRUNC)
1655 acc_mode |= MAY_WRITE;
1656
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 /* Allow the LSM permission hook to distinguish append
1658 access from general write access. */
1659 if (flag & O_APPEND)
1660 acc_mode |= MAY_APPEND;
1661
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 /*
1663 * The simplest case - just a plain lookup.
1664 */
1665 if (!(flag & O_CREAT)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001666 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001667 &nd, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 if (error)
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001669 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 goto ok;
1671 }
1672
1673 /*
1674 * Create - we need to know the parent.
1675 */
Al Viro8737f3a2008-08-02 22:36:57 -04001676 error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 if (error)
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001678 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679
1680 /*
1681 * We have the parent and last component. First of all, check
1682 * that we are not asked to creat(2) an obvious directory - that
1683 * will not do.
1684 */
1685 error = -EISDIR;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001686 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
Al Viro8737f3a2008-08-02 22:36:57 -04001687 goto exit_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688
Al Viro8737f3a2008-08-02 22:36:57 -04001689 error = -ENFILE;
1690 filp = get_empty_filp();
1691 if (filp == NULL)
1692 goto exit_parent;
1693 nd.intent.open.file = filp;
1694 nd.intent.open.flags = flag;
1695 nd.intent.open.create_mode = mode;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001696 dir = nd.path.dentry;
1697 nd.flags &= ~LOOKUP_PARENT;
Al Viro8737f3a2008-08-02 22:36:57 -04001698 nd.flags |= LOOKUP_CREATE | LOOKUP_OPEN;
Al Viro35165862008-08-05 03:00:49 -04001699 if (flag & O_EXCL)
1700 nd.flags |= LOOKUP_EXCL;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001701 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001702 path.dentry = lookup_hash(&nd);
1703 path.mnt = nd.path.mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704
1705do_last:
Al Viro4e7506e2005-06-06 13:36:00 -07001706 error = PTR_ERR(path.dentry);
1707 if (IS_ERR(path.dentry)) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001708 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 goto exit;
1710 }
1711
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001712 if (IS_ERR(nd.intent.open.file)) {
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001713 error = PTR_ERR(nd.intent.open.file);
Dave Hansen4a3fd212008-02-15 14:37:48 -08001714 goto exit_mutex_unlock;
Oleg Drokin4af4c522006-03-25 03:06:54 -08001715 }
1716
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 /* Negative dentry, just create the file */
Al Viro4e7506e2005-06-06 13:36:00 -07001718 if (!path.dentry->d_inode) {
Dave Hansen4a3fd212008-02-15 14:37:48 -08001719 /*
1720 * This write is needed to ensure that a
1721 * ro->rw transition does not occur between
1722 * the time when the file is created and when
1723 * a permanent write count is taken through
1724 * the 'struct file' in nameidata_to_filp().
1725 */
1726 error = mnt_want_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 if (error)
Dave Hansen4a3fd212008-02-15 14:37:48 -08001728 goto exit_mutex_unlock;
1729 error = __open_namei_create(&nd, &path, flag, mode);
1730 if (error) {
1731 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732 goto exit;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001733 }
1734 filp = nameidata_to_filp(&nd, open_flag);
1735 mnt_drop_write(nd.path.mnt);
1736 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 }
1738
1739 /*
1740 * It already exists.
1741 */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001742 mutex_unlock(&dir->d_inode->i_mutex);
Al Viro5a190ae2007-06-07 12:19:32 -04001743 audit_inode(pathname, path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744
1745 error = -EEXIST;
1746 if (flag & O_EXCL)
1747 goto exit_dput;
1748
Al Viroe13b2102005-06-06 13:36:06 -07001749 if (__follow_mount(&path)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 error = -ELOOP;
Al Viroba7a4c12005-06-06 13:36:08 -07001751 if (flag & O_NOFOLLOW)
1752 goto exit_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 }
Amy Griffis3e2efce2006-07-13 13:16:02 -04001754
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 error = -ENOENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001756 if (!path.dentry->d_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 goto exit_dput;
Al Viro4e7506e2005-06-06 13:36:00 -07001758 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 goto do_link;
1760
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001761 path_to_nameidata(&path, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 error = -EISDIR;
Al Viro4e7506e2005-06-06 13:36:00 -07001763 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 goto exit;
1765ok:
Dave Hansen4a3fd212008-02-15 14:37:48 -08001766 /*
1767 * Consider:
1768 * 1. may_open() truncates a file
1769 * 2. a rw->ro mount transition occurs
1770 * 3. nameidata_to_filp() fails due to
1771 * the ro mount.
1772 * That would be inconsistent, and should
1773 * be avoided. Taking this mnt write here
1774 * ensures that (2) can not occur.
1775 */
1776 will_write = open_will_write_to_fs(flag, nd.path.dentry->d_inode);
1777 if (will_write) {
1778 error = mnt_want_write(nd.path.mnt);
1779 if (error)
1780 goto exit;
1781 }
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001782 error = may_open(&nd, acc_mode, flag);
Dave Hansen4a3fd212008-02-15 14:37:48 -08001783 if (error) {
1784 if (will_write)
1785 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 goto exit;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001787 }
1788 filp = nameidata_to_filp(&nd, open_flag);
1789 /*
1790 * It is now safe to drop the mnt write
1791 * because the filp has had a write taken
1792 * on its behalf.
1793 */
1794 if (will_write)
1795 mnt_drop_write(nd.path.mnt);
1796 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797
Dave Hansen4a3fd212008-02-15 14:37:48 -08001798exit_mutex_unlock:
1799 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800exit_dput:
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001801 path_put_conditional(&path, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802exit:
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001803 if (!IS_ERR(nd.intent.open.file))
1804 release_open_intent(&nd);
Al Viro8737f3a2008-08-02 22:36:57 -04001805exit_parent:
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001806 path_put(&nd.path);
1807 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808
1809do_link:
1810 error = -ELOOP;
1811 if (flag & O_NOFOLLOW)
1812 goto exit_dput;
1813 /*
1814 * This is subtle. Instead of calling do_follow_link() we do the
1815 * thing by hands. The reason is that this way we have zero link_count
1816 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1817 * After that we have the parent and last component, i.e.
1818 * we are in the same situation as after the first path_walk().
1819 * Well, almost - if the last component is normal we get its copy
1820 * stored in nd->last.name and we will have to putname() it when we
1821 * are done. Procfs-like symlinks just set LAST_BIND.
1822 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001823 nd.flags |= LOOKUP_PARENT;
1824 error = security_inode_follow_link(path.dentry, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 if (error)
1826 goto exit_dput;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001827 error = __do_follow_link(&path, &nd);
Kirill Korotaevde459212006-07-14 00:23:49 -07001828 if (error) {
1829 /* Does someone understand code flow here? Or it is only
1830 * me so stupid? Anathema to whoever designed this non-sense
1831 * with "intent.open".
1832 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001833 release_open_intent(&nd);
1834 return ERR_PTR(error);
Kirill Korotaevde459212006-07-14 00:23:49 -07001835 }
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001836 nd.flags &= ~LOOKUP_PARENT;
1837 if (nd.last_type == LAST_BIND)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 goto ok;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 error = -EISDIR;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001840 if (nd.last_type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 goto exit;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001842 if (nd.last.name[nd.last.len]) {
1843 __putname(nd.last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844 goto exit;
1845 }
1846 error = -ELOOP;
1847 if (count++==32) {
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001848 __putname(nd.last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 goto exit;
1850 }
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001851 dir = nd.path.dentry;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001852 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001853 path.dentry = lookup_hash(&nd);
1854 path.mnt = nd.path.mnt;
1855 __putname(nd.last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 goto do_last;
1857}
1858
1859/**
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001860 * filp_open - open file and return file pointer
1861 *
1862 * @filename: path to open
1863 * @flags: open flags as per the open(2) second argument
1864 * @mode: mode for the new file if O_CREAT is set, else ignored
1865 *
1866 * This is the helper to open a file from kernelspace if you really
1867 * have to. But in generally you should not do this, so please move
1868 * along, nothing to see here..
1869 */
1870struct file *filp_open(const char *filename, int flags, int mode)
1871{
1872 return do_filp_open(AT_FDCWD, filename, flags, mode);
1873}
1874EXPORT_SYMBOL(filp_open);
1875
1876/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 * lookup_create - lookup a dentry, creating it if it doesn't exist
1878 * @nd: nameidata info
1879 * @is_dir: directory flag
1880 *
1881 * Simple function to lookup and return a dentry and create it
1882 * if it doesn't exist. Is SMP-safe.
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001883 *
Jan Blunck4ac91372008-02-14 19:34:32 -08001884 * Returns with nd->path.dentry->d_inode->i_mutex locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885 */
1886struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1887{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001888 struct dentry *dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889
Jan Blunck4ac91372008-02-14 19:34:32 -08001890 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001891 /*
1892 * Yucky last component or no last component at all?
1893 * (foo/., foo/.., /////)
1894 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895 if (nd->last_type != LAST_NORM)
1896 goto fail;
1897 nd->flags &= ~LOOKUP_PARENT;
Al Viro35165862008-08-05 03:00:49 -04001898 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
ASANO Masahiroa6349042006-08-22 20:06:02 -04001899 nd->intent.open.flags = O_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001900
1901 /*
1902 * Do the final lookup.
1903 */
Christoph Hellwig49705b72005-11-08 21:35:06 -08001904 dentry = lookup_hash(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905 if (IS_ERR(dentry))
1906 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001907
Al Viroe9baf6e2008-05-15 04:49:12 -04001908 if (dentry->d_inode)
1909 goto eexist;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001910 /*
1911 * Special case - lookup gave negative, but... we had foo/bar/
1912 * From the vfs_mknod() POV we just have a negative dentry -
1913 * all is fine. Let's be bastards - you had / on the end, you've
1914 * been asking for (non-existent) directory. -ENOENT for you.
1915 */
Al Viroe9baf6e2008-05-15 04:49:12 -04001916 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1917 dput(dentry);
1918 dentry = ERR_PTR(-ENOENT);
1919 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920 return dentry;
Al Viroe9baf6e2008-05-15 04:49:12 -04001921eexist:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 dput(dentry);
Al Viroe9baf6e2008-05-15 04:49:12 -04001923 dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924fail:
1925 return dentry;
1926}
Christoph Hellwigf81a0bf2005-05-19 12:26:43 -07001927EXPORT_SYMBOL_GPL(lookup_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928
1929int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1930{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001931 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932
1933 if (error)
1934 return error;
1935
1936 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1937 return -EPERM;
1938
1939 if (!dir->i_op || !dir->i_op->mknod)
1940 return -EPERM;
1941
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07001942 error = devcgroup_inode_mknod(mode, dev);
1943 if (error)
1944 return error;
1945
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 error = security_inode_mknod(dir, dentry, mode, dev);
1947 if (error)
1948 return error;
1949
1950 DQUOT_INIT(dir);
1951 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001952 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001953 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 return error;
1955}
1956
Dave Hansen463c3192008-02-15 14:37:57 -08001957static int may_mknod(mode_t mode)
1958{
1959 switch (mode & S_IFMT) {
1960 case S_IFREG:
1961 case S_IFCHR:
1962 case S_IFBLK:
1963 case S_IFIFO:
1964 case S_IFSOCK:
1965 case 0: /* zero mode translates to S_IFREG */
1966 return 0;
1967 case S_IFDIR:
1968 return -EPERM;
1969 default:
1970 return -EINVAL;
1971 }
1972}
1973
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001974asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
1975 unsigned dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976{
Al Viro2ad94ae2008-07-21 09:32:51 -04001977 int error;
1978 char *tmp;
1979 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 struct nameidata nd;
1981
1982 if (S_ISDIR(mode))
1983 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984
Al Viro2ad94ae2008-07-21 09:32:51 -04001985 error = user_path_parent(dfd, filename, &nd, &tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04001987 return error;
1988
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989 dentry = lookup_create(&nd, 0);
Dave Hansen463c3192008-02-15 14:37:57 -08001990 if (IS_ERR(dentry)) {
1991 error = PTR_ERR(dentry);
1992 goto out_unlock;
1993 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001994 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 mode &= ~current->fs->umask;
Dave Hansen463c3192008-02-15 14:37:57 -08001996 error = may_mknod(mode);
1997 if (error)
1998 goto out_dput;
1999 error = mnt_want_write(nd.path.mnt);
2000 if (error)
2001 goto out_dput;
2002 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003 case 0: case S_IFREG:
Jan Blunck4ac91372008-02-14 19:34:32 -08002004 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005 break;
2006 case S_IFCHR: case S_IFBLK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002007 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008 new_decode_dev(dev));
2009 break;
2010 case S_IFIFO: case S_IFSOCK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002011 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 }
Dave Hansen463c3192008-02-15 14:37:57 -08002014 mnt_drop_write(nd.path.mnt);
2015out_dput:
2016 dput(dentry);
2017out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002018 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002019 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 putname(tmp);
2021
2022 return error;
2023}
2024
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002025asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
2026{
2027 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2028}
2029
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2031{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002032 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033
2034 if (error)
2035 return error;
2036
2037 if (!dir->i_op || !dir->i_op->mkdir)
2038 return -EPERM;
2039
2040 mode &= (S_IRWXUGO|S_ISVTX);
2041 error = security_inode_mkdir(dir, dentry, mode);
2042 if (error)
2043 return error;
2044
2045 DQUOT_INIT(dir);
2046 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002047 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002048 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 return error;
2050}
2051
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002052asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053{
2054 int error = 0;
2055 char * tmp;
Dave Hansen6902d922006-09-30 23:29:01 -07002056 struct dentry *dentry;
2057 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058
Al Viro2ad94ae2008-07-21 09:32:51 -04002059 error = user_path_parent(dfd, pathname, &nd, &tmp);
2060 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002061 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062
Dave Hansen6902d922006-09-30 23:29:01 -07002063 dentry = lookup_create(&nd, 1);
2064 error = PTR_ERR(dentry);
2065 if (IS_ERR(dentry))
2066 goto out_unlock;
2067
Jan Blunck4ac91372008-02-14 19:34:32 -08002068 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Dave Hansen6902d922006-09-30 23:29:01 -07002069 mode &= ~current->fs->umask;
Dave Hansen463c3192008-02-15 14:37:57 -08002070 error = mnt_want_write(nd.path.mnt);
2071 if (error)
2072 goto out_dput;
Jan Blunck4ac91372008-02-14 19:34:32 -08002073 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
Dave Hansen463c3192008-02-15 14:37:57 -08002074 mnt_drop_write(nd.path.mnt);
2075out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002076 dput(dentry);
2077out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002078 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002079 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002080 putname(tmp);
2081out_err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082 return error;
2083}
2084
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002085asmlinkage long sys_mkdir(const char __user *pathname, int mode)
2086{
2087 return sys_mkdirat(AT_FDCWD, pathname, mode);
2088}
2089
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090/*
2091 * We try to drop the dentry early: we should have
2092 * a usage count of 2 if we're the only user of this
2093 * dentry, and if that is true (possibly after pruning
2094 * the dcache), then we drop the dentry now.
2095 *
2096 * A low-level filesystem can, if it choses, legally
2097 * do a
2098 *
2099 * if (!d_unhashed(dentry))
2100 * return -EBUSY;
2101 *
2102 * if it cannot handle the case of removing a directory
2103 * that is still in use by something else..
2104 */
2105void dentry_unhash(struct dentry *dentry)
2106{
2107 dget(dentry);
Vasily Averindc168422006-12-06 20:37:07 -08002108 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 spin_lock(&dcache_lock);
2110 spin_lock(&dentry->d_lock);
2111 if (atomic_read(&dentry->d_count) == 2)
2112 __d_drop(dentry);
2113 spin_unlock(&dentry->d_lock);
2114 spin_unlock(&dcache_lock);
2115}
2116
2117int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2118{
2119 int error = may_delete(dir, dentry, 1);
2120
2121 if (error)
2122 return error;
2123
2124 if (!dir->i_op || !dir->i_op->rmdir)
2125 return -EPERM;
2126
2127 DQUOT_INIT(dir);
2128
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002129 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 dentry_unhash(dentry);
2131 if (d_mountpoint(dentry))
2132 error = -EBUSY;
2133 else {
2134 error = security_inode_rmdir(dir, dentry);
2135 if (!error) {
2136 error = dir->i_op->rmdir(dir, dentry);
2137 if (!error)
2138 dentry->d_inode->i_flags |= S_DEAD;
2139 }
2140 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002141 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143 d_delete(dentry);
2144 }
2145 dput(dentry);
2146
2147 return error;
2148}
2149
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002150static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151{
2152 int error = 0;
2153 char * name;
2154 struct dentry *dentry;
2155 struct nameidata nd;
2156
Al Viro2ad94ae2008-07-21 09:32:51 -04002157 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002159 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160
2161 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002162 case LAST_DOTDOT:
2163 error = -ENOTEMPTY;
2164 goto exit1;
2165 case LAST_DOT:
2166 error = -EINVAL;
2167 goto exit1;
2168 case LAST_ROOT:
2169 error = -EBUSY;
2170 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002172
2173 nd.flags &= ~LOOKUP_PARENT;
2174
Jan Blunck4ac91372008-02-14 19:34:32 -08002175 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002176 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002178 if (IS_ERR(dentry))
2179 goto exit2;
Dave Hansen06227532008-02-15 14:37:34 -08002180 error = mnt_want_write(nd.path.mnt);
2181 if (error)
2182 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08002183 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Dave Hansen06227532008-02-15 14:37:34 -08002184 mnt_drop_write(nd.path.mnt);
2185exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07002186 dput(dentry);
2187exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08002188 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002190 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191 putname(name);
2192 return error;
2193}
2194
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002195asmlinkage long sys_rmdir(const char __user *pathname)
2196{
2197 return do_rmdir(AT_FDCWD, pathname);
2198}
2199
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200int vfs_unlink(struct inode *dir, struct dentry *dentry)
2201{
2202 int error = may_delete(dir, dentry, 0);
2203
2204 if (error)
2205 return error;
2206
2207 if (!dir->i_op || !dir->i_op->unlink)
2208 return -EPERM;
2209
2210 DQUOT_INIT(dir);
2211
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002212 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213 if (d_mountpoint(dentry))
2214 error = -EBUSY;
2215 else {
2216 error = security_inode_unlink(dir, dentry);
2217 if (!error)
2218 error = dir->i_op->unlink(dir, dentry);
2219 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002220 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221
2222 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2223 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08002224 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 }
Robert Love0eeca282005-07-12 17:06:03 -04002227
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228 return error;
2229}
2230
2231/*
2232 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002233 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234 * writeout happening, and we don't want to prevent access to the directory
2235 * while waiting on the I/O.
2236 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002237static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238{
Al Viro2ad94ae2008-07-21 09:32:51 -04002239 int error;
2240 char *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241 struct dentry *dentry;
2242 struct nameidata nd;
2243 struct inode *inode = NULL;
2244
Al Viro2ad94ae2008-07-21 09:32:51 -04002245 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002247 return error;
2248
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249 error = -EISDIR;
2250 if (nd.last_type != LAST_NORM)
2251 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002252
2253 nd.flags &= ~LOOKUP_PARENT;
2254
Jan Blunck4ac91372008-02-14 19:34:32 -08002255 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002256 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257 error = PTR_ERR(dentry);
2258 if (!IS_ERR(dentry)) {
2259 /* Why not before? Because we want correct error value */
2260 if (nd.last.name[nd.last.len])
2261 goto slashes;
2262 inode = dentry->d_inode;
2263 if (inode)
2264 atomic_inc(&inode->i_count);
Dave Hansen06227532008-02-15 14:37:34 -08002265 error = mnt_want_write(nd.path.mnt);
2266 if (error)
2267 goto exit2;
Jan Blunck4ac91372008-02-14 19:34:32 -08002268 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Dave Hansen06227532008-02-15 14:37:34 -08002269 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 exit2:
2271 dput(dentry);
2272 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002273 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274 if (inode)
2275 iput(inode); /* truncate the inode here */
2276exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002277 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278 putname(name);
2279 return error;
2280
2281slashes:
2282 error = !dentry->d_inode ? -ENOENT :
2283 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2284 goto exit2;
2285}
2286
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002287asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2288{
2289 if ((flag & ~AT_REMOVEDIR) != 0)
2290 return -EINVAL;
2291
2292 if (flag & AT_REMOVEDIR)
2293 return do_rmdir(dfd, pathname);
2294
2295 return do_unlinkat(dfd, pathname);
2296}
2297
2298asmlinkage long sys_unlink(const char __user *pathname)
2299{
2300 return do_unlinkat(AT_FDCWD, pathname);
2301}
2302
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002303int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002305 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306
2307 if (error)
2308 return error;
2309
2310 if (!dir->i_op || !dir->i_op->symlink)
2311 return -EPERM;
2312
2313 error = security_inode_symlink(dir, dentry, oldname);
2314 if (error)
2315 return error;
2316
2317 DQUOT_INIT(dir);
2318 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002319 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002320 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321 return error;
2322}
2323
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002324asmlinkage long sys_symlinkat(const char __user *oldname,
2325 int newdfd, const char __user *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326{
Al Viro2ad94ae2008-07-21 09:32:51 -04002327 int error;
2328 char *from;
2329 char *to;
Dave Hansen6902d922006-09-30 23:29:01 -07002330 struct dentry *dentry;
2331 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332
2333 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04002334 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335 return PTR_ERR(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002336
2337 error = user_path_parent(newdfd, newname, &nd, &to);
2338 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002339 goto out_putname;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340
Dave Hansen6902d922006-09-30 23:29:01 -07002341 dentry = lookup_create(&nd, 0);
2342 error = PTR_ERR(dentry);
2343 if (IS_ERR(dentry))
2344 goto out_unlock;
2345
Dave Hansen75c3f292008-02-15 14:37:45 -08002346 error = mnt_want_write(nd.path.mnt);
2347 if (error)
2348 goto out_dput;
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002349 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
Dave Hansen75c3f292008-02-15 14:37:45 -08002350 mnt_drop_write(nd.path.mnt);
2351out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002352 dput(dentry);
2353out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002354 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002355 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002356 putname(to);
2357out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358 putname(from);
2359 return error;
2360}
2361
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002362asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2363{
2364 return sys_symlinkat(oldname, AT_FDCWD, newname);
2365}
2366
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2368{
2369 struct inode *inode = old_dentry->d_inode;
2370 int error;
2371
2372 if (!inode)
2373 return -ENOENT;
2374
Miklos Szeredia95164d2008-07-30 15:08:48 +02002375 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376 if (error)
2377 return error;
2378
2379 if (dir->i_sb != inode->i_sb)
2380 return -EXDEV;
2381
2382 /*
2383 * A link to an append-only or immutable file cannot be created.
2384 */
2385 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2386 return -EPERM;
2387 if (!dir->i_op || !dir->i_op->link)
2388 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002389 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390 return -EPERM;
2391
2392 error = security_inode_link(old_dentry, dir, new_dentry);
2393 if (error)
2394 return error;
2395
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002396 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002397 DQUOT_INIT(dir);
2398 error = dir->i_op->link(old_dentry, dir, new_dentry);
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002399 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002400 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002401 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402 return error;
2403}
2404
2405/*
2406 * Hardlinks are often used in delicate situations. We avoid
2407 * security-related surprises by not following symlinks on the
2408 * newname. --KAB
2409 *
2410 * We don't follow them on the oldname either to be compatible
2411 * with linux 2.0, and to avoid hard-linking to directories
2412 * and other special files. --ADM
2413 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002414asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002415 int newdfd, const char __user *newname,
2416 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417{
2418 struct dentry *new_dentry;
Al Viro2d8f3032008-07-22 09:59:21 -04002419 struct nameidata nd;
2420 struct path old_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421 int error;
Al Viro2ad94ae2008-07-21 09:32:51 -04002422 char *to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423
Ulrich Drepper45c9b112006-06-25 05:49:11 -07002424 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002425 return -EINVAL;
2426
Al Viro2d8f3032008-07-22 09:59:21 -04002427 error = user_path_at(olddfd, oldname,
2428 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2429 &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002431 return error;
2432
2433 error = user_path_parent(newdfd, newname, &nd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434 if (error)
2435 goto out;
2436 error = -EXDEV;
Al Viro2d8f3032008-07-22 09:59:21 -04002437 if (old_path.mnt != nd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438 goto out_release;
2439 new_dentry = lookup_create(&nd, 0);
2440 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002441 if (IS_ERR(new_dentry))
2442 goto out_unlock;
Dave Hansen75c3f292008-02-15 14:37:45 -08002443 error = mnt_want_write(nd.path.mnt);
2444 if (error)
2445 goto out_dput;
Al Viro2d8f3032008-07-22 09:59:21 -04002446 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
Dave Hansen75c3f292008-02-15 14:37:45 -08002447 mnt_drop_write(nd.path.mnt);
2448out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002449 dput(new_dentry);
2450out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002451 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452out_release:
Jan Blunck1d957f92008-02-14 19:34:35 -08002453 path_put(&nd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002454 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455out:
Al Viro2d8f3032008-07-22 09:59:21 -04002456 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457
2458 return error;
2459}
2460
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002461asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2462{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002463 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002464}
2465
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466/*
2467 * The worst of all namespace operations - renaming directory. "Perverted"
2468 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2469 * Problems:
2470 * a) we can get into loop creation. Check is done in is_subdir().
2471 * b) race potential - two innocent renames can create a loop together.
2472 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002473 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474 * story.
2475 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002476 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477 * whether the target exists). Solution: try to be smart with locking
2478 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002479 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07002480 * move will be locked. Thus we can rank directories by the tree
2481 * (ancestors first) and rank all non-directories after them.
2482 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002483 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484 * HOWEVER, it relies on the assumption that any object with ->lookup()
2485 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2486 * we'd better make sure that there's no link(2) for them.
2487 * d) some filesystems don't support opened-but-unlinked directories,
2488 * either because of layout or because they are not ready to deal with
2489 * all cases correctly. The latter will be fixed (taking this sort of
2490 * stuff into VFS), but the former is not going away. Solution: the same
2491 * trick as in rmdir().
2492 * e) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002493 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002495 * ->i_mutex on parents, which works but leads to some truely excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496 * locking].
2497 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07002498static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2499 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002500{
2501 int error = 0;
2502 struct inode *target;
2503
2504 /*
2505 * If we are going to change the parent - check write permissions,
2506 * we'll need to flip '..'.
2507 */
2508 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04002509 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510 if (error)
2511 return error;
2512 }
2513
2514 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2515 if (error)
2516 return error;
2517
2518 target = new_dentry->d_inode;
2519 if (target) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002520 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521 dentry_unhash(new_dentry);
2522 }
2523 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2524 error = -EBUSY;
2525 else
2526 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2527 if (target) {
2528 if (!error)
2529 target->i_flags |= S_DEAD;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002530 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531 if (d_unhashed(new_dentry))
2532 d_rehash(new_dentry);
2533 dput(new_dentry);
2534 }
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002535 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07002536 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2537 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 return error;
2539}
2540
Adrian Bunk75c96f82005-05-05 16:16:09 -07002541static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2542 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543{
2544 struct inode *target;
2545 int error;
2546
2547 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2548 if (error)
2549 return error;
2550
2551 dget(new_dentry);
2552 target = new_dentry->d_inode;
2553 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002554 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2556 error = -EBUSY;
2557 else
2558 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2559 if (!error) {
Mark Fasheh349457c2006-09-08 14:22:21 -07002560 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561 d_move(old_dentry, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562 }
2563 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002564 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565 dput(new_dentry);
2566 return error;
2567}
2568
2569int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2570 struct inode *new_dir, struct dentry *new_dentry)
2571{
2572 int error;
2573 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Robert Love0eeca282005-07-12 17:06:03 -04002574 const char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575
2576 if (old_dentry->d_inode == new_dentry->d_inode)
2577 return 0;
2578
2579 error = may_delete(old_dir, old_dentry, is_dir);
2580 if (error)
2581 return error;
2582
2583 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02002584 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585 else
2586 error = may_delete(new_dir, new_dentry, is_dir);
2587 if (error)
2588 return error;
2589
2590 if (!old_dir->i_op || !old_dir->i_op->rename)
2591 return -EPERM;
2592
2593 DQUOT_INIT(old_dir);
2594 DQUOT_INIT(new_dir);
2595
Robert Love0eeca282005-07-12 17:06:03 -04002596 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2597
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598 if (is_dir)
2599 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2600 else
2601 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2602 if (!error) {
Robert Love0eeca282005-07-12 17:06:03 -04002603 const char *new_name = old_dentry->d_name.name;
John McCutchan89204c42005-08-15 12:13:28 -04002604 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04002605 new_dentry->d_inode, old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002606 }
Robert Love0eeca282005-07-12 17:06:03 -04002607 fsnotify_oldname_free(old_name);
2608
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609 return error;
2610}
2611
Al Viro2ad94ae2008-07-21 09:32:51 -04002612asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2613 int newdfd, const char __user *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002614{
Al Viro2ad94ae2008-07-21 09:32:51 -04002615 struct dentry *old_dir, *new_dir;
2616 struct dentry *old_dentry, *new_dentry;
2617 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618 struct nameidata oldnd, newnd;
Al Viro2ad94ae2008-07-21 09:32:51 -04002619 char *from;
2620 char *to;
2621 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622
Al Viro2ad94ae2008-07-21 09:32:51 -04002623 error = user_path_parent(olddfd, oldname, &oldnd, &from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002624 if (error)
2625 goto exit;
2626
Al Viro2ad94ae2008-07-21 09:32:51 -04002627 error = user_path_parent(newdfd, newname, &newnd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628 if (error)
2629 goto exit1;
2630
2631 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08002632 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002633 goto exit2;
2634
Jan Blunck4ac91372008-02-14 19:34:32 -08002635 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636 error = -EBUSY;
2637 if (oldnd.last_type != LAST_NORM)
2638 goto exit2;
2639
Jan Blunck4ac91372008-02-14 19:34:32 -08002640 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641 if (newnd.last_type != LAST_NORM)
2642 goto exit2;
2643
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002644 oldnd.flags &= ~LOOKUP_PARENT;
2645 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09002646 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002647
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648 trap = lock_rename(new_dir, old_dir);
2649
Christoph Hellwig49705b72005-11-08 21:35:06 -08002650 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651 error = PTR_ERR(old_dentry);
2652 if (IS_ERR(old_dentry))
2653 goto exit3;
2654 /* source must exist */
2655 error = -ENOENT;
2656 if (!old_dentry->d_inode)
2657 goto exit4;
2658 /* unless the source is a directory trailing slashes give -ENOTDIR */
2659 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2660 error = -ENOTDIR;
2661 if (oldnd.last.name[oldnd.last.len])
2662 goto exit4;
2663 if (newnd.last.name[newnd.last.len])
2664 goto exit4;
2665 }
2666 /* source should not be ancestor of target */
2667 error = -EINVAL;
2668 if (old_dentry == trap)
2669 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08002670 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671 error = PTR_ERR(new_dentry);
2672 if (IS_ERR(new_dentry))
2673 goto exit4;
2674 /* target should not be an ancestor of source */
2675 error = -ENOTEMPTY;
2676 if (new_dentry == trap)
2677 goto exit5;
2678
Dave Hansen9079b1e2008-02-15 14:37:49 -08002679 error = mnt_want_write(oldnd.path.mnt);
2680 if (error)
2681 goto exit5;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682 error = vfs_rename(old_dir->d_inode, old_dentry,
2683 new_dir->d_inode, new_dentry);
Dave Hansen9079b1e2008-02-15 14:37:49 -08002684 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002685exit5:
2686 dput(new_dentry);
2687exit4:
2688 dput(old_dentry);
2689exit3:
2690 unlock_rename(new_dir, old_dir);
2691exit2:
Jan Blunck1d957f92008-02-14 19:34:35 -08002692 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002693 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002694exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002695 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002696 putname(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002697exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002698 return error;
2699}
2700
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002701asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2702{
2703 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2704}
2705
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2707{
2708 int len;
2709
2710 len = PTR_ERR(link);
2711 if (IS_ERR(link))
2712 goto out;
2713
2714 len = strlen(link);
2715 if (len > (unsigned) buflen)
2716 len = buflen;
2717 if (copy_to_user(buffer, link, len))
2718 len = -EFAULT;
2719out:
2720 return len;
2721}
2722
2723/*
2724 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2725 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2726 * using) it for any given inode is up to filesystem.
2727 */
2728int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2729{
2730 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002731 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07002732 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002733
Linus Torvalds1da177e2005-04-16 15:20:36 -07002734 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002735 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07002736 if (IS_ERR(cookie))
2737 return PTR_ERR(cookie);
2738
2739 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2740 if (dentry->d_inode->i_op->put_link)
2741 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2742 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002743}
2744
2745int vfs_follow_link(struct nameidata *nd, const char *link)
2746{
2747 return __vfs_follow_link(nd, link);
2748}
2749
2750/* get the link contents into pagecache */
2751static char *page_getlink(struct dentry * dentry, struct page **ppage)
2752{
2753 struct page * page;
2754 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07002755 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002756 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07002757 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002758 *ppage = page;
2759 return kmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760}
2761
2762int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2763{
2764 struct page *page = NULL;
2765 char *s = page_getlink(dentry, &page);
2766 int res = vfs_readlink(dentry,buffer,buflen,s);
2767 if (page) {
2768 kunmap(page);
2769 page_cache_release(page);
2770 }
2771 return res;
2772}
2773
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002774void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002775{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002776 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002778 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779}
2780
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002781void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002783 struct page *page = cookie;
2784
2785 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786 kunmap(page);
2787 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788 }
2789}
2790
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002791int __page_symlink(struct inode *inode, const char *symname, int len,
2792 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002793{
2794 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002795 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002796 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08002797 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798 char *kaddr;
2799
NeilBrown7e53cac2006-03-25 03:07:57 -08002800retry:
Nick Pigginafddba42007-10-16 01:25:01 -07002801 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2802 AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002803 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07002804 goto fail;
2805
Linus Torvalds1da177e2005-04-16 15:20:36 -07002806 kaddr = kmap_atomic(page, KM_USER0);
2807 memcpy(kaddr, symname, len-1);
2808 kunmap_atomic(kaddr, KM_USER0);
Nick Pigginafddba42007-10-16 01:25:01 -07002809
2810 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2811 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812 if (err < 0)
2813 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07002814 if (err < len-1)
2815 goto retry;
2816
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817 mark_inode_dirty(inode);
2818 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002819fail:
2820 return err;
2821}
2822
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002823int page_symlink(struct inode *inode, const char *symname, int len)
2824{
2825 return __page_symlink(inode, symname, len,
2826 mapping_gfp_mask(inode->i_mapping));
2827}
2828
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002829const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830 .readlink = generic_readlink,
2831 .follow_link = page_follow_link_light,
2832 .put_link = page_put_link,
2833};
2834
Al Viro2d8f3032008-07-22 09:59:21 -04002835EXPORT_SYMBOL(user_path_at);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836EXPORT_SYMBOL(follow_down);
2837EXPORT_SYMBOL(follow_up);
2838EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2839EXPORT_SYMBOL(getname);
2840EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841EXPORT_SYMBOL(lookup_one_len);
2842EXPORT_SYMBOL(page_follow_link_light);
2843EXPORT_SYMBOL(page_put_link);
2844EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002845EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002846EXPORT_SYMBOL(page_symlink);
2847EXPORT_SYMBOL(page_symlink_inode_operations);
2848EXPORT_SYMBOL(path_lookup);
Al Virod1811462008-08-02 00:49:18 -04002849EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002850EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04002851EXPORT_SYMBOL(inode_permission);
Christoph Hellwige4543ed2005-11-08 21:35:04 -08002852EXPORT_SYMBOL(vfs_permission);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -08002853EXPORT_SYMBOL(file_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002854EXPORT_SYMBOL(unlock_rename);
2855EXPORT_SYMBOL(vfs_create);
2856EXPORT_SYMBOL(vfs_follow_link);
2857EXPORT_SYMBOL(vfs_link);
2858EXPORT_SYMBOL(vfs_mkdir);
2859EXPORT_SYMBOL(vfs_mknod);
2860EXPORT_SYMBOL(generic_permission);
2861EXPORT_SYMBOL(vfs_readlink);
2862EXPORT_SYMBOL(vfs_rename);
2863EXPORT_SYMBOL(vfs_rmdir);
2864EXPORT_SYMBOL(vfs_symlink);
2865EXPORT_SYMBOL(vfs_unlink);
2866EXPORT_SYMBOL(dentry_unhash);
2867EXPORT_SYMBOL(generic_readlink);