blob: bbc15c237558f82c7fbe2b661e55ee8fa1b9849d [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
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200229/**
230 * inode_permission - check for access rights to a given inode
231 * @inode: inode to check permission on
232 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
233 *
234 * Used to check for read/write/execute permissions on an inode.
235 * We use "fsuid" for this, letting us set arbitrary permissions
236 * for filesystem access without changing the "normal" uids which
237 * are used for other things.
238 */
Al Virof419a2e2008-07-22 00:07:17 -0400239int inode_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240{
Al Viroe6305c42008-07-15 21:03:57 -0400241 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242
243 if (mask & MAY_WRITE) {
Miklos Szeredi22590e42007-10-16 23:27:08 -0700244 umode_t mode = inode->i_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245
246 /*
247 * Nobody gets write access to a read-only fs.
248 */
249 if (IS_RDONLY(inode) &&
250 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
251 return -EROFS;
252
253 /*
254 * Nobody gets write access to an immutable file.
255 */
256 if (IS_IMMUTABLE(inode))
257 return -EACCES;
258 }
259
Al Viroacfa4382008-12-04 10:06:33 -0500260 if (inode->i_op->permission)
Al Virob77b0642008-07-17 09:37:02 -0400261 retval = inode->i_op->permission(inode, mask);
Miklos Szeredif696a362008-07-31 13:41:58 +0200262 else
Al Viroe6305c42008-07-15 21:03:57 -0400263 retval = generic_permission(inode, mask, NULL);
Miklos Szeredif696a362008-07-31 13:41:58 +0200264
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 if (retval)
266 return retval;
267
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -0700268 retval = devcgroup_inode_permission(inode, mask);
269 if (retval)
270 return retval;
271
Al Viroe6305c42008-07-15 21:03:57 -0400272 return security_inode_permission(inode,
Stephen Smalleyf418b002008-07-28 13:32:38 -0400273 mask & (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274}
275
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800276/**
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800277 * file_permission - check for additional access rights to a given file
278 * @file: file to check access rights for
279 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
280 *
281 * Used to check for read/write/execute permissions on an already opened
282 * file.
283 *
284 * Note:
285 * Do not use this function in new code. All access checks should
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200286 * be done using inode_permission().
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800287 */
288int file_permission(struct file *file, int mask)
289{
Al Virof419a2e2008-07-22 00:07:17 -0400290 return inode_permission(file->f_path.dentry->d_inode, mask);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800291}
292
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293/*
294 * get_write_access() gets write permission for a file.
295 * put_write_access() releases this write permission.
296 * This is used for regular files.
297 * We cannot support write (and maybe mmap read-write shared) accesses and
298 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
299 * can have the following values:
300 * 0: no writers, no VM_DENYWRITE mappings
301 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
302 * > 0: (i_writecount) users are writing to the file.
303 *
304 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
305 * except for the cases where we don't hold i_writecount yet. Then we need to
306 * use {get,deny}_write_access() - these functions check the sign and refuse
307 * to do the change if sign is wrong. Exclusion between them is provided by
308 * the inode->i_lock spinlock.
309 */
310
311int get_write_access(struct inode * inode)
312{
313 spin_lock(&inode->i_lock);
314 if (atomic_read(&inode->i_writecount) < 0) {
315 spin_unlock(&inode->i_lock);
316 return -ETXTBSY;
317 }
318 atomic_inc(&inode->i_writecount);
319 spin_unlock(&inode->i_lock);
320
321 return 0;
322}
323
324int deny_write_access(struct file * file)
325{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800326 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327
328 spin_lock(&inode->i_lock);
329 if (atomic_read(&inode->i_writecount) > 0) {
330 spin_unlock(&inode->i_lock);
331 return -ETXTBSY;
332 }
333 atomic_dec(&inode->i_writecount);
334 spin_unlock(&inode->i_lock);
335
336 return 0;
337}
338
Jan Blunck1d957f92008-02-14 19:34:35 -0800339/**
Jan Blunck5dd784d02008-02-14 19:34:38 -0800340 * path_get - get a reference to a path
341 * @path: path to get the reference to
342 *
343 * Given a path increment the reference count to the dentry and the vfsmount.
344 */
345void path_get(struct path *path)
346{
347 mntget(path->mnt);
348 dget(path->dentry);
349}
350EXPORT_SYMBOL(path_get);
351
352/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800353 * path_put - put a reference to a path
354 * @path: path to put the reference to
355 *
356 * Given a path decrement the reference count to the dentry and the vfsmount.
357 */
358void path_put(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359{
Jan Blunck1d957f92008-02-14 19:34:35 -0800360 dput(path->dentry);
361 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362}
Jan Blunck1d957f92008-02-14 19:34:35 -0800363EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364
Trond Myklebust834f2a42005-10-18 14:20:16 -0700365/**
366 * release_open_intent - free up open intent resources
367 * @nd: pointer to nameidata
368 */
369void release_open_intent(struct nameidata *nd)
370{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800371 if (nd->intent.open.file->f_path.dentry == NULL)
Trond Myklebust834f2a42005-10-18 14:20:16 -0700372 put_filp(nd->intent.open.file);
373 else
374 fput(nd->intent.open.file);
375}
376
Ian Kentbcdc5e02006-09-27 01:50:44 -0700377static inline struct dentry *
378do_revalidate(struct dentry *dentry, struct nameidata *nd)
379{
380 int status = dentry->d_op->d_revalidate(dentry, nd);
381 if (unlikely(status <= 0)) {
382 /*
383 * The dentry failed validation.
384 * If d_revalidate returned 0 attempt to invalidate
385 * the dentry otherwise d_revalidate is asking us
386 * to return a fail status.
387 */
388 if (!status) {
389 if (!d_invalidate(dentry)) {
390 dput(dentry);
391 dentry = NULL;
392 }
393 } else {
394 dput(dentry);
395 dentry = ERR_PTR(status);
396 }
397 }
398 return dentry;
399}
400
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401/*
402 * Internal lookup() using the new generic dcache.
403 * SMP-safe
404 */
405static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
406{
407 struct dentry * dentry = __d_lookup(parent, name);
408
409 /* lockess __d_lookup may fail due to concurrent d_move()
410 * in some unrelated directory, so try with d_lookup
411 */
412 if (!dentry)
413 dentry = d_lookup(parent, name);
414
Ian Kentbcdc5e02006-09-27 01:50:44 -0700415 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
416 dentry = do_revalidate(dentry, nd);
417
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 return dentry;
419}
420
421/*
422 * Short-cut version of permission(), for calling by
423 * path_walk(), when dcache lock is held. Combines parts
424 * of permission() and generic_permission(), and tests ONLY for
425 * MAY_EXEC permission.
426 *
427 * If appropriate, check DAC only. If not appropriate, or
428 * short-cut DAC fails, then call permission() to do more
429 * complete permission check.
430 */
Al Viro672b16b2008-07-17 09:45:01 -0400431static int exec_permission_lite(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432{
433 umode_t mode = inode->i_mode;
434
Al Viroacfa4382008-12-04 10:06:33 -0500435 if (inode->i_op->permission)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 return -EAGAIN;
437
David Howellsda9592e2008-11-14 10:39:05 +1100438 if (current_fsuid() == inode->i_uid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 mode >>= 6;
440 else if (in_group_p(inode->i_gid))
441 mode >>= 3;
442
443 if (mode & MAY_EXEC)
444 goto ok;
445
446 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
447 goto ok;
448
449 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
450 goto ok;
451
452 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
453 goto ok;
454
455 return -EACCES;
456ok:
Al Virob77b0642008-07-17 09:37:02 -0400457 return security_inode_permission(inode, MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458}
459
460/*
461 * This is called when everything else fails, and we actually have
462 * to go to the low-level filesystem to find out what we should do..
463 *
464 * We get the directory semaphore, and after getting that we also
465 * make sure that nobody added the entry to the dcache in the meantime..
466 * SMP-safe
467 */
468static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
469{
470 struct dentry * result;
471 struct inode *dir = parent->d_inode;
472
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800473 mutex_lock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 /*
475 * First re-do the cached lookup just in case it was created
476 * while we waited for the directory semaphore..
477 *
478 * FIXME! This could use version numbering or similar to
479 * avoid unnecessary cache lookups.
480 *
481 * The "dcache_lock" is purely to protect the RCU list walker
482 * from concurrent renames at this point (we mustn't get false
483 * negatives from the RCU list walk here, unlike the optimistic
484 * fast walk).
485 *
486 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
487 */
488 result = d_lookup(parent, name);
489 if (!result) {
Miklos Szeredid70b67c2008-07-02 21:30:15 +0200490 struct dentry *dentry;
491
492 /* Don't create child dentry for a dead directory. */
493 result = ERR_PTR(-ENOENT);
494 if (IS_DEADDIR(dir))
495 goto out_unlock;
496
497 dentry = d_alloc(parent, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 result = ERR_PTR(-ENOMEM);
499 if (dentry) {
500 result = dir->i_op->lookup(dir, dentry, nd);
501 if (result)
502 dput(dentry);
503 else
504 result = dentry;
505 }
Miklos Szeredid70b67c2008-07-02 21:30:15 +0200506out_unlock:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800507 mutex_unlock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 return result;
509 }
510
511 /*
512 * Uhhuh! Nasty case: the cache was re-populated while
513 * we waited on the semaphore. Need to revalidate.
514 */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800515 mutex_unlock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 if (result->d_op && result->d_op->d_revalidate) {
Ian Kentbcdc5e02006-09-27 01:50:44 -0700517 result = do_revalidate(result, nd);
518 if (!result)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 result = ERR_PTR(-ENOENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 }
521 return result;
522}
523
Al Viroa02f76c2008-02-23 15:14:28 +0000524/*
525 * Wrapper to retry pathname resolution whenever the underlying
526 * file system returns an ESTALE.
527 *
528 * Retry the whole path once, forcing real lookup requests
529 * instead of relying on the dcache.
530 */
531static __always_inline int link_path_walk(const char *name, struct nameidata *nd)
532{
533 struct path save = nd->path;
534 int result;
535
536 /* make sure the stuff we saved doesn't go away */
Jan Blunckc8e7f442008-06-09 16:40:35 -0700537 path_get(&save);
Al Viroa02f76c2008-02-23 15:14:28 +0000538
539 result = __link_path_walk(name, nd);
540 if (result == -ESTALE) {
541 /* nd->path had been dropped */
542 nd->path = save;
Jan Blunckc8e7f442008-06-09 16:40:35 -0700543 path_get(&nd->path);
Al Viroa02f76c2008-02-23 15:14:28 +0000544 nd->flags |= LOOKUP_REVAL;
545 result = __link_path_walk(name, nd);
546 }
547
548 path_put(&save);
549
550 return result;
551}
552
Arjan van de Venf1662352006-01-14 13:21:31 -0800553static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554{
555 int res = 0;
556 char *name;
557 if (IS_ERR(link))
558 goto fail;
559
560 if (*link == '/') {
Christoph Hellwigb4091d52008-11-05 15:07:21 +0100561 struct fs_struct *fs = current->fs;
562
Jan Blunck1d957f92008-02-14 19:34:35 -0800563 path_put(&nd->path);
Christoph Hellwigb4091d52008-11-05 15:07:21 +0100564
565 read_lock(&fs->lock);
566 nd->path = fs->root;
567 path_get(&fs->root);
568 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 }
Christoph Hellwigb4091d52008-11-05 15:07:21 +0100570
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 res = link_path_walk(link, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 if (nd->depth || res || nd->last_type!=LAST_NORM)
573 return res;
574 /*
575 * If it is an iterative symlinks resolution in open_namei() we
576 * have to copy the last component. And all that crap because of
577 * bloody create() on broken symlinks. Furrfu...
578 */
579 name = __getname();
580 if (unlikely(!name)) {
Jan Blunck1d957f92008-02-14 19:34:35 -0800581 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 return -ENOMEM;
583 }
584 strcpy(name, nd->last.name);
585 nd->last.name = name;
586 return 0;
587fail:
Jan Blunck1d957f92008-02-14 19:34:35 -0800588 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 return PTR_ERR(link);
590}
591
Jan Blunck1d957f92008-02-14 19:34:35 -0800592static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700593{
594 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800595 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700596 mntput(path->mnt);
597}
598
599static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
600{
Jan Blunck4ac91372008-02-14 19:34:32 -0800601 dput(nd->path.dentry);
602 if (nd->path.mnt != path->mnt)
603 mntput(nd->path.mnt);
604 nd->path.mnt = path->mnt;
605 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700606}
607
Ian Kent051d3812006-03-27 01:14:53 -0800608static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
609{
610 int error;
611 void *cookie;
612 struct dentry *dentry = path->dentry;
613
614 touch_atime(path->mnt, dentry);
615 nd_set_link(nd, NULL);
616
Jan Blunck4ac91372008-02-14 19:34:32 -0800617 if (path->mnt != nd->path.mnt) {
Ian Kent051d3812006-03-27 01:14:53 -0800618 path_to_nameidata(path, nd);
619 dget(dentry);
620 }
621 mntget(path->mnt);
622 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
623 error = PTR_ERR(cookie);
624 if (!IS_ERR(cookie)) {
625 char *s = nd_get_link(nd);
626 error = 0;
627 if (s)
628 error = __vfs_follow_link(nd, s);
629 if (dentry->d_inode->i_op->put_link)
630 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
631 }
Jan Blunck09da59162008-02-14 19:34:37 -0800632 path_put(path);
Ian Kent051d3812006-03-27 01:14:53 -0800633
634 return error;
635}
636
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637/*
638 * This limits recursive symlink follows to 8, while
639 * limiting consecutive symlinks to 40.
640 *
641 * Without that kind of total limit, nasty chains of consecutive
642 * symlinks can cause almost arbitrarily long lookups.
643 */
Al Viro90ebe562005-06-06 13:35:58 -0700644static inline int do_follow_link(struct path *path, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645{
646 int err = -ELOOP;
647 if (current->link_count >= MAX_NESTED_LINKS)
648 goto loop;
649 if (current->total_link_count >= 40)
650 goto loop;
651 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
652 cond_resched();
Al Viro90ebe562005-06-06 13:35:58 -0700653 err = security_inode_follow_link(path->dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 if (err)
655 goto loop;
656 current->link_count++;
657 current->total_link_count++;
658 nd->depth++;
Al Virocd4e91d2005-06-06 13:36:03 -0700659 err = __do_follow_link(path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 current->link_count--;
661 nd->depth--;
662 return err;
663loop:
Jan Blunck1d957f92008-02-14 19:34:35 -0800664 path_put_conditional(path, nd);
665 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 return err;
667}
668
669int follow_up(struct vfsmount **mnt, struct dentry **dentry)
670{
671 struct vfsmount *parent;
672 struct dentry *mountpoint;
673 spin_lock(&vfsmount_lock);
674 parent=(*mnt)->mnt_parent;
675 if (parent == *mnt) {
676 spin_unlock(&vfsmount_lock);
677 return 0;
678 }
679 mntget(parent);
680 mountpoint=dget((*mnt)->mnt_mountpoint);
681 spin_unlock(&vfsmount_lock);
682 dput(*dentry);
683 *dentry = mountpoint;
684 mntput(*mnt);
685 *mnt = parent;
686 return 1;
687}
688
689/* no need for dcache_lock, as serialization is taken care in
690 * namespace.c
691 */
Al Viro463ffb22005-06-06 13:36:05 -0700692static int __follow_mount(struct path *path)
693{
694 int res = 0;
695 while (d_mountpoint(path->dentry)) {
696 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
697 if (!mounted)
698 break;
699 dput(path->dentry);
700 if (res)
701 mntput(path->mnt);
702 path->mnt = mounted;
703 path->dentry = dget(mounted->mnt_root);
704 res = 1;
705 }
706 return res;
707}
708
Al Viro58c465e2005-06-06 13:36:13 -0700709static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 while (d_mountpoint(*dentry)) {
712 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
713 if (!mounted)
714 break;
Al Viro58c465e2005-06-06 13:36:13 -0700715 dput(*dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 mntput(*mnt);
717 *mnt = mounted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 *dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720}
721
722/* no need for dcache_lock, as serialization is taken care in
723 * namespace.c
724 */
Al Viroe13b2102005-06-06 13:36:06 -0700725int follow_down(struct vfsmount **mnt, struct dentry **dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726{
727 struct vfsmount *mounted;
728
729 mounted = lookup_mnt(*mnt, *dentry);
730 if (mounted) {
Al Viroe13b2102005-06-06 13:36:06 -0700731 dput(*dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 mntput(*mnt);
733 *mnt = mounted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 *dentry = dget(mounted->mnt_root);
735 return 1;
736 }
737 return 0;
738}
739
Arjan van de Venf1662352006-01-14 13:21:31 -0800740static __always_inline void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741{
Andreas Mohre518ddb2006-09-29 02:01:22 -0700742 struct fs_struct *fs = current->fs;
743
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 while(1) {
745 struct vfsmount *parent;
Jan Blunck4ac91372008-02-14 19:34:32 -0800746 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747
Andreas Mohre518ddb2006-09-29 02:01:22 -0700748 read_lock(&fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -0800749 if (nd->path.dentry == fs->root.dentry &&
750 nd->path.mnt == fs->root.mnt) {
Andreas Mohre518ddb2006-09-29 02:01:22 -0700751 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 break;
753 }
Andreas Mohre518ddb2006-09-29 02:01:22 -0700754 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 spin_lock(&dcache_lock);
Jan Blunck4ac91372008-02-14 19:34:32 -0800756 if (nd->path.dentry != nd->path.mnt->mnt_root) {
757 nd->path.dentry = dget(nd->path.dentry->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 spin_unlock(&dcache_lock);
759 dput(old);
760 break;
761 }
762 spin_unlock(&dcache_lock);
763 spin_lock(&vfsmount_lock);
Jan Blunck4ac91372008-02-14 19:34:32 -0800764 parent = nd->path.mnt->mnt_parent;
765 if (parent == nd->path.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 spin_unlock(&vfsmount_lock);
767 break;
768 }
769 mntget(parent);
Jan Blunck4ac91372008-02-14 19:34:32 -0800770 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771 spin_unlock(&vfsmount_lock);
772 dput(old);
Jan Blunck4ac91372008-02-14 19:34:32 -0800773 mntput(nd->path.mnt);
774 nd->path.mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 }
Jan Blunck4ac91372008-02-14 19:34:32 -0800776 follow_mount(&nd->path.mnt, &nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777}
778
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779/*
780 * It's more convoluted than I'd like it to be, but... it's still fairly
781 * small and for now I'd prefer to have fast path as straight as possible.
782 * It _is_ time-critical.
783 */
784static int do_lookup(struct nameidata *nd, struct qstr *name,
785 struct path *path)
786{
Jan Blunck4ac91372008-02-14 19:34:32 -0800787 struct vfsmount *mnt = nd->path.mnt;
788 struct dentry *dentry = __d_lookup(nd->path.dentry, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789
790 if (!dentry)
791 goto need_lookup;
792 if (dentry->d_op && dentry->d_op->d_revalidate)
793 goto need_revalidate;
794done:
795 path->mnt = mnt;
796 path->dentry = dentry;
Al Viro634ee702005-06-06 13:36:13 -0700797 __follow_mount(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 return 0;
799
800need_lookup:
Jan Blunck4ac91372008-02-14 19:34:32 -0800801 dentry = real_lookup(nd->path.dentry, name, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 if (IS_ERR(dentry))
803 goto fail;
804 goto done;
805
806need_revalidate:
Ian Kentbcdc5e02006-09-27 01:50:44 -0700807 dentry = do_revalidate(dentry, nd);
808 if (!dentry)
809 goto need_lookup;
810 if (IS_ERR(dentry))
811 goto fail;
812 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813
814fail:
815 return PTR_ERR(dentry);
816}
817
818/*
819 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +0100820 * This is the basic name resolution function, turning a pathname into
821 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 *
Prasanna Medaea3834d2005-04-29 16:00:17 +0100823 * Returns 0 and nd will have valid dentry and mnt on success.
824 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800826static int __link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827{
828 struct path next;
829 struct inode *inode;
830 int err;
831 unsigned int lookup_flags = nd->flags;
832
833 while (*name=='/')
834 name++;
835 if (!*name)
836 goto return_reval;
837
Jan Blunck4ac91372008-02-14 19:34:32 -0800838 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 if (nd->depth)
Trond Myklebustf55eab82006-02-04 23:28:01 -0800840 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841
842 /* At this point we know we have a real path component. */
843 for(;;) {
844 unsigned long hash;
845 struct qstr this;
846 unsigned int c;
847
Trond Myklebustcdce5d62005-10-18 14:20:18 -0700848 nd->flags |= LOOKUP_CONTINUE;
Al Viro672b16b2008-07-17 09:45:01 -0400849 err = exec_permission_lite(inode);
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800850 if (err == -EAGAIN)
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200851 err = inode_permission(nd->path.dentry->d_inode,
852 MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 if (err)
854 break;
855
856 this.name = name;
857 c = *(const unsigned char *)name;
858
859 hash = init_name_hash();
860 do {
861 name++;
862 hash = partial_name_hash(c, hash);
863 c = *(const unsigned char *)name;
864 } while (c && (c != '/'));
865 this.len = name - (const char *) this.name;
866 this.hash = end_name_hash(hash);
867
868 /* remove trailing slashes? */
869 if (!c)
870 goto last_component;
871 while (*++name == '/');
872 if (!*name)
873 goto last_with_slashes;
874
875 /*
876 * "." and ".." are special - ".." especially so because it has
877 * to be able to know about the current root directory and
878 * parent relationships.
879 */
880 if (this.name[0] == '.') switch (this.len) {
881 default:
882 break;
883 case 2:
884 if (this.name[1] != '.')
885 break;
Al Viro58c465e2005-06-06 13:36:13 -0700886 follow_dotdot(nd);
Jan Blunck4ac91372008-02-14 19:34:32 -0800887 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 /* fallthrough */
889 case 1:
890 continue;
891 }
892 /*
893 * See if the low-level filesystem might want
894 * to use its own hash..
895 */
Jan Blunck4ac91372008-02-14 19:34:32 -0800896 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
897 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
898 &this);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 if (err < 0)
900 break;
901 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 /* This does the actual lookups.. */
903 err = do_lookup(nd, &this, &next);
904 if (err)
905 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906
907 err = -ENOENT;
908 inode = next.dentry->d_inode;
909 if (!inode)
910 goto out_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911
912 if (inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700913 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 if (err)
915 goto return_err;
916 err = -ENOENT;
Jan Blunck4ac91372008-02-14 19:34:32 -0800917 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 if (!inode)
919 break;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700920 } else
921 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 err = -ENOTDIR;
923 if (!inode->i_op->lookup)
924 break;
925 continue;
926 /* here ends the main loop */
927
928last_with_slashes:
929 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
930last_component:
Trond Myklebustf55eab82006-02-04 23:28:01 -0800931 /* Clear LOOKUP_CONTINUE iff it was previously unset */
932 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 if (lookup_flags & LOOKUP_PARENT)
934 goto lookup_parent;
935 if (this.name[0] == '.') switch (this.len) {
936 default:
937 break;
938 case 2:
939 if (this.name[1] != '.')
940 break;
Al Viro58c465e2005-06-06 13:36:13 -0700941 follow_dotdot(nd);
Jan Blunck4ac91372008-02-14 19:34:32 -0800942 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 /* fallthrough */
944 case 1:
945 goto return_reval;
946 }
Jan Blunck4ac91372008-02-14 19:34:32 -0800947 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
948 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
949 &this);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 if (err < 0)
951 break;
952 }
953 err = do_lookup(nd, &this, &next);
954 if (err)
955 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 inode = next.dentry->d_inode;
957 if ((lookup_flags & LOOKUP_FOLLOW)
Al Viroacfa4382008-12-04 10:06:33 -0500958 && inode && inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700959 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 if (err)
961 goto return_err;
Jan Blunck4ac91372008-02-14 19:34:32 -0800962 inode = nd->path.dentry->d_inode;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700963 } else
964 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 err = -ENOENT;
966 if (!inode)
967 break;
968 if (lookup_flags & LOOKUP_DIRECTORY) {
969 err = -ENOTDIR;
Al Viroacfa4382008-12-04 10:06:33 -0500970 if (!inode->i_op->lookup)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 break;
972 }
973 goto return_base;
974lookup_parent:
975 nd->last = this;
976 nd->last_type = LAST_NORM;
977 if (this.name[0] != '.')
978 goto return_base;
979 if (this.len == 1)
980 nd->last_type = LAST_DOT;
981 else if (this.len == 2 && this.name[1] == '.')
982 nd->last_type = LAST_DOTDOT;
983 else
984 goto return_base;
985return_reval:
986 /*
987 * We bypassed the ordinary revalidation routines.
988 * We may need to check the cached dentry for staleness.
989 */
Jan Blunck4ac91372008-02-14 19:34:32 -0800990 if (nd->path.dentry && nd->path.dentry->d_sb &&
991 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 err = -ESTALE;
993 /* Note: we do not d_invalidate() */
Jan Blunck4ac91372008-02-14 19:34:32 -0800994 if (!nd->path.dentry->d_op->d_revalidate(
995 nd->path.dentry, nd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 break;
997 }
998return_base:
999 return 0;
1000out_dput:
Jan Blunck1d957f92008-02-14 19:34:35 -08001001 path_put_conditional(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 break;
1003 }
Jan Blunck1d957f92008-02-14 19:34:35 -08001004 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005return_err:
1006 return err;
1007}
1008
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -08001009static int path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010{
1011 current->total_link_count = 0;
1012 return link_path_walk(name, nd);
1013}
1014
Prasanna Medaea3834d2005-04-29 16:00:17 +01001015/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -08001016static int do_path_lookup(int dfd, const char *name,
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001017 unsigned int flags, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001019 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001020 int fput_needed;
1021 struct file *file;
Andreas Mohre518ddb2006-09-29 02:01:22 -07001022 struct fs_struct *fs = current->fs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023
1024 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1025 nd->flags = flags;
1026 nd->depth = 0;
1027
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 if (*name=='/') {
Andreas Mohre518ddb2006-09-29 02:01:22 -07001029 read_lock(&fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08001030 nd->path = fs->root;
1031 path_get(&fs->root);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001032 read_unlock(&fs->lock);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001033 } else if (dfd == AT_FDCWD) {
Andreas Mohre518ddb2006-09-29 02:01:22 -07001034 read_lock(&fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08001035 nd->path = fs->pwd;
1036 path_get(&fs->pwd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001037 read_unlock(&fs->lock);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001038 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001039 struct dentry *dentry;
1040
1041 file = fget_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001042 retval = -EBADF;
1043 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001044 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001045
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001046 dentry = file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001047
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001048 retval = -ENOTDIR;
1049 if (!S_ISDIR(dentry->d_inode->i_mode))
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001050 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001051
1052 retval = file_permission(file, MAY_EXEC);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001053 if (retval)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001054 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001055
Jan Blunck5dd784d02008-02-14 19:34:38 -08001056 nd->path = file->f_path;
1057 path_get(&file->f_path);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001058
1059 fput_light(file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 }
Josef 'Jeff' Sipek2dfdd262007-05-09 02:33:41 -07001061
1062 retval = path_walk(name, nd);
Jan Blunck4ac91372008-02-14 19:34:32 -08001063 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1064 nd->path.dentry->d_inode))
1065 audit_inode(name, nd->path.dentry);
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001066out_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001067 return retval;
1068
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001069fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001070 fput_light(file, fput_needed);
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001071 goto out_fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072}
1073
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -08001074int path_lookup(const char *name, unsigned int flags,
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001075 struct nameidata *nd)
1076{
1077 return do_path_lookup(AT_FDCWD, name, flags, nd);
1078}
1079
Al Virod1811462008-08-02 00:49:18 -04001080int kern_path(const char *name, unsigned int flags, struct path *path)
1081{
1082 struct nameidata nd;
1083 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1084 if (!res)
1085 *path = nd.path;
1086 return res;
1087}
1088
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001089/**
1090 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1091 * @dentry: pointer to dentry of the base directory
1092 * @mnt: pointer to vfs mount of the base directory
1093 * @name: pointer to file name
1094 * @flags: lookup flags
1095 * @nd: pointer to nameidata
1096 */
1097int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1098 const char *name, unsigned int flags,
1099 struct nameidata *nd)
1100{
1101 int retval;
1102
1103 /* same as do_path_lookup */
1104 nd->last_type = LAST_ROOT;
1105 nd->flags = flags;
1106 nd->depth = 0;
1107
Jan Blunckc8e7f442008-06-09 16:40:35 -07001108 nd->path.dentry = dentry;
1109 nd->path.mnt = mnt;
1110 path_get(&nd->path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001111
1112 retval = path_walk(name, nd);
Jan Blunck4ac91372008-02-14 19:34:32 -08001113 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1114 nd->path.dentry->d_inode))
1115 audit_inode(name, nd->path.dentry);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001116
1117 return retval;
1118
1119}
1120
Trond Myklebust834f2a42005-10-18 14:20:16 -07001121/**
1122 * path_lookup_open - lookup a file path with open intent
Martin Waitz7045f372006-02-01 03:06:57 -08001123 * @dfd: the directory to use as base, or AT_FDCWD
Trond Myklebust834f2a42005-10-18 14:20:16 -07001124 * @name: pointer to file name
1125 * @lookup_flags: lookup intent flags
1126 * @nd: pointer to nameidata
1127 * @open_flags: open intent flags
1128 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001129int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
Trond Myklebust834f2a42005-10-18 14:20:16 -07001130 struct nameidata *nd, int open_flags)
1131{
Al Viro8737f3a2008-08-02 22:36:57 -04001132 struct file *filp = get_empty_filp();
1133 int err;
Trond Myklebust834f2a42005-10-18 14:20:16 -07001134
Al Viro8737f3a2008-08-02 22:36:57 -04001135 if (filp == NULL)
1136 return -ENFILE;
1137 nd->intent.open.file = filp;
1138 nd->intent.open.flags = open_flags;
1139 nd->intent.open.create_mode = 0;
1140 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1141 if (IS_ERR(nd->intent.open.file)) {
1142 if (err == 0) {
1143 err = PTR_ERR(nd->intent.open.file);
1144 path_put(&nd->path);
1145 }
1146 } else if (err != 0)
1147 release_open_intent(nd);
1148 return err;
Trond Myklebust834f2a42005-10-18 14:20:16 -07001149}
1150
Christoph Hellwigeead1912007-10-16 23:25:38 -07001151static struct dentry *__lookup_hash(struct qstr *name,
1152 struct dentry *base, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153{
James Morris057f6c02007-04-26 00:12:05 -07001154 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 struct inode *inode;
1156 int err;
1157
1158 inode = base->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159
1160 /*
1161 * See if the low-level filesystem might want
1162 * to use its own hash..
1163 */
1164 if (base->d_op && base->d_op->d_hash) {
1165 err = base->d_op->d_hash(base, name);
1166 dentry = ERR_PTR(err);
1167 if (err < 0)
1168 goto out;
1169 }
1170
1171 dentry = cached_lookup(base, name, nd);
1172 if (!dentry) {
Miklos Szeredid70b67c2008-07-02 21:30:15 +02001173 struct dentry *new;
1174
1175 /* Don't create child dentry for a dead directory. */
1176 dentry = ERR_PTR(-ENOENT);
1177 if (IS_DEADDIR(inode))
1178 goto out;
1179
1180 new = d_alloc(base, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 dentry = ERR_PTR(-ENOMEM);
1182 if (!new)
1183 goto out;
1184 dentry = inode->i_op->lookup(inode, new, nd);
1185 if (!dentry)
1186 dentry = new;
1187 else
1188 dput(new);
1189 }
1190out:
1191 return dentry;
1192}
1193
James Morris057f6c02007-04-26 00:12:05 -07001194/*
1195 * Restricted form of lookup. Doesn't follow links, single-component only,
1196 * needs parent already locked. Doesn't follow mounts.
1197 * SMP-safe.
1198 */
Adrian Bunka244e162006-03-31 02:32:11 -08001199static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200{
Christoph Hellwigeead1912007-10-16 23:25:38 -07001201 int err;
1202
Al Virof419a2e2008-07-22 00:07:17 -04001203 err = inode_permission(nd->path.dentry->d_inode, MAY_EXEC);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001204 if (err)
1205 return ERR_PTR(err);
Jan Blunck4ac91372008-02-14 19:34:32 -08001206 return __lookup_hash(&nd->last, nd->path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207}
1208
Christoph Hellwigeead1912007-10-16 23:25:38 -07001209static int __lookup_one_len(const char *name, struct qstr *this,
1210 struct dentry *base, int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211{
1212 unsigned long hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 unsigned int c;
1214
James Morris057f6c02007-04-26 00:12:05 -07001215 this->name = name;
1216 this->len = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 if (!len)
James Morris057f6c02007-04-26 00:12:05 -07001218 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219
1220 hash = init_name_hash();
1221 while (len--) {
1222 c = *(const unsigned char *)name++;
1223 if (c == '/' || c == '\0')
James Morris057f6c02007-04-26 00:12:05 -07001224 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 hash = partial_name_hash(c, hash);
1226 }
James Morris057f6c02007-04-26 00:12:05 -07001227 this->hash = end_name_hash(hash);
1228 return 0;
1229}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230
Christoph Hellwigeead1912007-10-16 23:25:38 -07001231/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001232 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07001233 * @name: pathname component to lookup
1234 * @base: base directory to lookup from
1235 * @len: maximum length @len should be interpreted to
1236 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07001237 * Note that this routine is purely a helper for filesystem usage and should
1238 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07001239 * nameidata argument is passed to the filesystem methods and a filesystem
1240 * using this helper needs to be prepared for that.
1241 */
James Morris057f6c02007-04-26 00:12:05 -07001242struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1243{
1244 int err;
1245 struct qstr this;
1246
1247 err = __lookup_one_len(name, &this, base, len);
1248 if (err)
1249 return ERR_PTR(err);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001250
Al Virof419a2e2008-07-22 00:07:17 -04001251 err = inode_permission(base->d_inode, MAY_EXEC);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001252 if (err)
1253 return ERR_PTR(err);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001254 return __lookup_hash(&this, base, NULL);
James Morris057f6c02007-04-26 00:12:05 -07001255}
1256
Christoph Hellwigeead1912007-10-16 23:25:38 -07001257/**
1258 * lookup_one_noperm - bad hack for sysfs
1259 * @name: pathname component to lookup
1260 * @base: base directory to lookup from
1261 *
1262 * This is a variant of lookup_one_len that doesn't perform any permission
1263 * checks. It's a horrible hack to work around the braindead sysfs
1264 * architecture and should not be used anywhere else.
1265 *
1266 * DON'T USE THIS FUNCTION EVER, thanks.
1267 */
1268struct dentry *lookup_one_noperm(const char *name, struct dentry *base)
James Morris057f6c02007-04-26 00:12:05 -07001269{
1270 int err;
1271 struct qstr this;
1272
Christoph Hellwigeead1912007-10-16 23:25:38 -07001273 err = __lookup_one_len(name, &this, base, strlen(name));
James Morris057f6c02007-04-26 00:12:05 -07001274 if (err)
1275 return ERR_PTR(err);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001276 return __lookup_hash(&this, base, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277}
1278
Al Viro2d8f3032008-07-22 09:59:21 -04001279int user_path_at(int dfd, const char __user *name, unsigned flags,
1280 struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281{
Al Viro2d8f3032008-07-22 09:59:21 -04001282 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 char *tmp = getname(name);
1284 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04001286
1287 BUG_ON(flags & LOOKUP_PARENT);
1288
1289 err = do_path_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04001291 if (!err)
1292 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 }
1294 return err;
1295}
1296
Al Viro2ad94ae2008-07-21 09:32:51 -04001297static int user_path_parent(int dfd, const char __user *path,
1298 struct nameidata *nd, char **name)
1299{
1300 char *s = getname(path);
1301 int error;
1302
1303 if (IS_ERR(s))
1304 return PTR_ERR(s);
1305
1306 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1307 if (error)
1308 putname(s);
1309 else
1310 *name = s;
1311
1312 return error;
1313}
1314
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315/*
1316 * It's inline, so penalty for filesystems that don't use sticky bit is
1317 * minimal.
1318 */
1319static inline int check_sticky(struct inode *dir, struct inode *inode)
1320{
David Howellsda9592e2008-11-14 10:39:05 +11001321 uid_t fsuid = current_fsuid();
1322
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 if (!(dir->i_mode & S_ISVTX))
1324 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001325 if (inode->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001327 if (dir->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 return 0;
1329 return !capable(CAP_FOWNER);
1330}
1331
1332/*
1333 * Check whether we can remove a link victim from directory dir, check
1334 * whether the type of victim is right.
1335 * 1. We can't do it if dir is read-only (done in permission())
1336 * 2. We should have write and exec permissions on dir
1337 * 3. We can't remove anything from append-only dir
1338 * 4. We can't do anything with immutable dir (done in permission())
1339 * 5. If the sticky bit on dir is set we should either
1340 * a. be owner of dir, or
1341 * b. be owner of victim, or
1342 * c. have CAP_FOWNER capability
1343 * 6. If the victim is append-only or immutable we can't do antyhing with
1344 * links pointing to it.
1345 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1346 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1347 * 9. We can't remove a root or mountpoint.
1348 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1349 * nfs_async_unlink().
1350 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001351static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352{
1353 int error;
1354
1355 if (!victim->d_inode)
1356 return -ENOENT;
1357
1358 BUG_ON(victim->d_parent->d_inode != dir);
Al Viro5a190ae2007-06-07 12:19:32 -04001359 audit_inode_child(victim->d_name.name, victim, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360
Al Virof419a2e2008-07-22 00:07:17 -04001361 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 if (error)
1363 return error;
1364 if (IS_APPEND(dir))
1365 return -EPERM;
1366 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
Hugh Dickinsf9454542008-11-19 15:36:38 -08001367 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 return -EPERM;
1369 if (isdir) {
1370 if (!S_ISDIR(victim->d_inode->i_mode))
1371 return -ENOTDIR;
1372 if (IS_ROOT(victim))
1373 return -EBUSY;
1374 } else if (S_ISDIR(victim->d_inode->i_mode))
1375 return -EISDIR;
1376 if (IS_DEADDIR(dir))
1377 return -ENOENT;
1378 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1379 return -EBUSY;
1380 return 0;
1381}
1382
1383/* Check whether we can create an object with dentry child in directory
1384 * dir.
1385 * 1. We can't do it if child already exists (open has special treatment for
1386 * this case, but since we are inlined it's OK)
1387 * 2. We can't do it if dir is read-only (done in permission())
1388 * 3. We should have write and exec permissions on dir
1389 * 4. We can't do it if dir is immutable (done in permission())
1390 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02001391static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392{
1393 if (child->d_inode)
1394 return -EEXIST;
1395 if (IS_DEADDIR(dir))
1396 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04001397 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398}
1399
1400/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 * O_DIRECTORY translates into forcing a directory lookup.
1402 */
1403static inline int lookup_flags(unsigned int f)
1404{
1405 unsigned long retval = LOOKUP_FOLLOW;
1406
1407 if (f & O_NOFOLLOW)
1408 retval &= ~LOOKUP_FOLLOW;
1409
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 if (f & O_DIRECTORY)
1411 retval |= LOOKUP_DIRECTORY;
1412
1413 return retval;
1414}
1415
1416/*
1417 * p1 and p2 should be directories on the same fs.
1418 */
1419struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1420{
1421 struct dentry *p;
1422
1423 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001424 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 return NULL;
1426 }
1427
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001428 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001430 p = d_ancestor(p2, p1);
1431 if (p) {
1432 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1433 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1434 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 }
1436
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001437 p = d_ancestor(p1, p2);
1438 if (p) {
1439 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1440 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1441 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 }
1443
Ingo Molnarf2eace22006-07-03 00:25:05 -07001444 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1445 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 return NULL;
1447}
1448
1449void unlock_rename(struct dentry *p1, struct dentry *p2)
1450{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001451 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001453 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001454 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 }
1456}
1457
1458int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1459 struct nameidata *nd)
1460{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001461 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462
1463 if (error)
1464 return error;
1465
Al Viroacfa4382008-12-04 10:06:33 -05001466 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 return -EACCES; /* shouldn't it be ENOSYS? */
1468 mode &= S_IALLUGO;
1469 mode |= S_IFREG;
1470 error = security_inode_create(dir, dentry, mode);
1471 if (error)
1472 return error;
1473 DQUOT_INIT(dir);
1474 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001475 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001476 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 return error;
1478}
1479
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001480int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001482 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 struct inode *inode = dentry->d_inode;
1484 int error;
1485
1486 if (!inode)
1487 return -ENOENT;
1488
1489 if (S_ISLNK(inode->i_mode))
1490 return -ELOOP;
1491
Linus Torvalds974a9f02008-01-12 14:06:34 -08001492 if (S_ISDIR(inode->i_mode) && (acc_mode & MAY_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 return -EISDIR;
1494
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 /*
1496 * FIFO's, sockets and device files are special: they don't
1497 * actually live on the filesystem itself, and as such you
1498 * can write to them even if the filesystem is read-only.
1499 */
1500 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1501 flag &= ~O_TRUNC;
1502 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001503 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 return -EACCES;
1505
1506 flag &= ~O_TRUNC;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001507 }
Dave Hansenb41572e2007-10-16 23:31:14 -07001508
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001509 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07001510 if (error)
1511 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 /*
1513 * An append-only file must be opened in append mode for writing.
1514 */
1515 if (IS_APPEND(inode)) {
1516 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1517 return -EPERM;
1518 if (flag & O_TRUNC)
1519 return -EPERM;
1520 }
1521
1522 /* O_NOATIME can only be set by the owner or superuser */
1523 if (flag & O_NOATIME)
Satyam Sharma3bd858a2007-07-17 15:00:08 +05301524 if (!is_owner_or_cap(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 return -EPERM;
1526
1527 /*
1528 * Ensure there are no outstanding leases on the file.
1529 */
1530 error = break_lease(inode, flag);
1531 if (error)
1532 return error;
1533
1534 if (flag & O_TRUNC) {
1535 error = get_write_access(inode);
1536 if (error)
1537 return error;
1538
1539 /*
1540 * Refuse to truncate files with mandatory locks held on them.
1541 */
1542 error = locks_verify_locked(inode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001543 if (!error)
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001544 error = security_path_truncate(path, 0,
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001545 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 if (!error) {
1547 DQUOT_INIT(inode);
Miklos Szeredid139d7f2007-10-18 03:07:00 -07001548
1549 error = do_truncate(dentry, 0,
1550 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1551 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 }
1553 put_write_access(inode);
1554 if (error)
1555 return error;
1556 } else
1557 if (flag & FMODE_WRITE)
1558 DQUOT_INIT(inode);
1559
1560 return 0;
1561}
1562
Dave Hansend57999e2008-02-15 14:37:27 -08001563/*
1564 * Be careful about ever adding any more callers of this
1565 * function. Its flags must be in the namei format, not
1566 * what get passed to sys_open().
1567 */
1568static int __open_namei_create(struct nameidata *nd, struct path *path,
Dave Hansenaab520e2006-09-30 23:29:02 -07001569 int flag, int mode)
1570{
1571 int error;
Jan Blunck4ac91372008-02-14 19:34:32 -08001572 struct dentry *dir = nd->path.dentry;
Dave Hansenaab520e2006-09-30 23:29:02 -07001573
1574 if (!IS_POSIXACL(dir->d_inode))
1575 mode &= ~current->fs->umask;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001576 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1577 if (error)
1578 goto out_unlock;
Dave Hansenaab520e2006-09-30 23:29:02 -07001579 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001580out_unlock:
Dave Hansenaab520e2006-09-30 23:29:02 -07001581 mutex_unlock(&dir->d_inode->i_mutex);
Jan Blunck4ac91372008-02-14 19:34:32 -08001582 dput(nd->path.dentry);
1583 nd->path.dentry = path->dentry;
Dave Hansenaab520e2006-09-30 23:29:02 -07001584 if (error)
1585 return error;
1586 /* Don't check for write permission, don't truncate */
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001587 return may_open(&nd->path, 0, flag & ~O_TRUNC);
Dave Hansenaab520e2006-09-30 23:29:02 -07001588}
1589
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590/*
Dave Hansend57999e2008-02-15 14:37:27 -08001591 * Note that while the flag value (low two bits) for sys_open means:
1592 * 00 - read-only
1593 * 01 - write-only
1594 * 10 - read-write
1595 * 11 - special
1596 * it is changed into
1597 * 00 - no permissions needed
1598 * 01 - read-permission
1599 * 10 - write-permission
1600 * 11 - read-write
1601 * for the internal routines (ie open_namei()/follow_link() etc)
1602 * This is more logical, and also allows the 00 "no perm needed"
1603 * to be used for symlinks (where the permissions are checked
1604 * later).
1605 *
1606*/
1607static inline int open_to_namei_flags(int flag)
1608{
1609 if ((flag+1) & O_ACCMODE)
1610 flag++;
1611 return flag;
1612}
1613
Dave Hansen4a3fd212008-02-15 14:37:48 -08001614static int open_will_write_to_fs(int flag, struct inode *inode)
1615{
1616 /*
1617 * We'll never write to the fs underlying
1618 * a device file.
1619 */
1620 if (special_file(inode->i_mode))
1621 return 0;
1622 return (flag & O_TRUNC);
1623}
1624
Dave Hansend57999e2008-02-15 14:37:27 -08001625/*
Dave Hansen4a3fd212008-02-15 14:37:48 -08001626 * Note that the low bits of the passed in "open_flag"
1627 * are not the same as in the local variable "flag". See
1628 * open_to_namei_flags() for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001630struct file *do_filp_open(int dfd, const char *pathname,
1631 int open_flag, int mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632{
Dave Hansen4a3fd212008-02-15 14:37:48 -08001633 struct file *filp;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001634 struct nameidata nd;
Trond Myklebust834f2a42005-10-18 14:20:16 -07001635 int acc_mode, error;
Al Viro4e7506e2005-06-06 13:36:00 -07001636 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 struct dentry *dir;
1638 int count = 0;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001639 int will_write;
Dave Hansend57999e2008-02-15 14:37:27 -08001640 int flag = open_to_namei_flags(open_flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641
Al Virob77b0642008-07-17 09:37:02 -04001642 acc_mode = MAY_OPEN | ACC_MODE(flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643
Trond Myklebust834f2a42005-10-18 14:20:16 -07001644 /* O_TRUNC implies we need access checks for write permissions */
1645 if (flag & O_TRUNC)
1646 acc_mode |= MAY_WRITE;
1647
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 /* Allow the LSM permission hook to distinguish append
1649 access from general write access. */
1650 if (flag & O_APPEND)
1651 acc_mode |= MAY_APPEND;
1652
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 /*
1654 * The simplest case - just a plain lookup.
1655 */
1656 if (!(flag & O_CREAT)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001657 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001658 &nd, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 if (error)
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001660 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 goto ok;
1662 }
1663
1664 /*
1665 * Create - we need to know the parent.
1666 */
Al Viro8737f3a2008-08-02 22:36:57 -04001667 error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
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
1671 /*
1672 * We have the parent and last component. First of all, check
1673 * that we are not asked to creat(2) an obvious directory - that
1674 * will not do.
1675 */
1676 error = -EISDIR;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001677 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
Al Viro8737f3a2008-08-02 22:36:57 -04001678 goto exit_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679
Al Viro8737f3a2008-08-02 22:36:57 -04001680 error = -ENFILE;
1681 filp = get_empty_filp();
1682 if (filp == NULL)
1683 goto exit_parent;
1684 nd.intent.open.file = filp;
1685 nd.intent.open.flags = flag;
1686 nd.intent.open.create_mode = mode;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001687 dir = nd.path.dentry;
1688 nd.flags &= ~LOOKUP_PARENT;
Al Viro8737f3a2008-08-02 22:36:57 -04001689 nd.flags |= LOOKUP_CREATE | LOOKUP_OPEN;
Al Viro35165862008-08-05 03:00:49 -04001690 if (flag & O_EXCL)
1691 nd.flags |= LOOKUP_EXCL;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001692 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001693 path.dentry = lookup_hash(&nd);
1694 path.mnt = nd.path.mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695
1696do_last:
Al Viro4e7506e2005-06-06 13:36:00 -07001697 error = PTR_ERR(path.dentry);
1698 if (IS_ERR(path.dentry)) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001699 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 goto exit;
1701 }
1702
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001703 if (IS_ERR(nd.intent.open.file)) {
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001704 error = PTR_ERR(nd.intent.open.file);
Dave Hansen4a3fd212008-02-15 14:37:48 -08001705 goto exit_mutex_unlock;
Oleg Drokin4af4c522006-03-25 03:06:54 -08001706 }
1707
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 /* Negative dentry, just create the file */
Al Viro4e7506e2005-06-06 13:36:00 -07001709 if (!path.dentry->d_inode) {
Dave Hansen4a3fd212008-02-15 14:37:48 -08001710 /*
1711 * This write is needed to ensure that a
1712 * ro->rw transition does not occur between
1713 * the time when the file is created and when
1714 * a permanent write count is taken through
1715 * the 'struct file' in nameidata_to_filp().
1716 */
1717 error = mnt_want_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 if (error)
Dave Hansen4a3fd212008-02-15 14:37:48 -08001719 goto exit_mutex_unlock;
1720 error = __open_namei_create(&nd, &path, flag, mode);
1721 if (error) {
1722 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 goto exit;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001724 }
1725 filp = nameidata_to_filp(&nd, open_flag);
1726 mnt_drop_write(nd.path.mnt);
1727 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 }
1729
1730 /*
1731 * It already exists.
1732 */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001733 mutex_unlock(&dir->d_inode->i_mutex);
Al Viro5a190ae2007-06-07 12:19:32 -04001734 audit_inode(pathname, path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735
1736 error = -EEXIST;
1737 if (flag & O_EXCL)
1738 goto exit_dput;
1739
Al Viroe13b2102005-06-06 13:36:06 -07001740 if (__follow_mount(&path)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 error = -ELOOP;
Al Viroba7a4c12005-06-06 13:36:08 -07001742 if (flag & O_NOFOLLOW)
1743 goto exit_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 }
Amy Griffis3e2efce2006-07-13 13:16:02 -04001745
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 error = -ENOENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001747 if (!path.dentry->d_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 goto exit_dput;
Al Viroacfa4382008-12-04 10:06:33 -05001749 if (path.dentry->d_inode->i_op->follow_link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 goto do_link;
1751
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001752 path_to_nameidata(&path, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 error = -EISDIR;
Al Viro4e7506e2005-06-06 13:36:00 -07001754 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 goto exit;
1756ok:
Dave Hansen4a3fd212008-02-15 14:37:48 -08001757 /*
1758 * Consider:
1759 * 1. may_open() truncates a file
1760 * 2. a rw->ro mount transition occurs
1761 * 3. nameidata_to_filp() fails due to
1762 * the ro mount.
1763 * That would be inconsistent, and should
1764 * be avoided. Taking this mnt write here
1765 * ensures that (2) can not occur.
1766 */
1767 will_write = open_will_write_to_fs(flag, nd.path.dentry->d_inode);
1768 if (will_write) {
1769 error = mnt_want_write(nd.path.mnt);
1770 if (error)
1771 goto exit;
1772 }
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001773 error = may_open(&nd.path, acc_mode, flag);
Dave Hansen4a3fd212008-02-15 14:37:48 -08001774 if (error) {
1775 if (will_write)
1776 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 goto exit;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001778 }
1779 filp = nameidata_to_filp(&nd, open_flag);
1780 /*
1781 * It is now safe to drop the mnt write
1782 * because the filp has had a write taken
1783 * on its behalf.
1784 */
1785 if (will_write)
1786 mnt_drop_write(nd.path.mnt);
1787 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788
Dave Hansen4a3fd212008-02-15 14:37:48 -08001789exit_mutex_unlock:
1790 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791exit_dput:
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001792 path_put_conditional(&path, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793exit:
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001794 if (!IS_ERR(nd.intent.open.file))
1795 release_open_intent(&nd);
Al Viro8737f3a2008-08-02 22:36:57 -04001796exit_parent:
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001797 path_put(&nd.path);
1798 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799
1800do_link:
1801 error = -ELOOP;
1802 if (flag & O_NOFOLLOW)
1803 goto exit_dput;
1804 /*
1805 * This is subtle. Instead of calling do_follow_link() we do the
1806 * thing by hands. The reason is that this way we have zero link_count
1807 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1808 * After that we have the parent and last component, i.e.
1809 * we are in the same situation as after the first path_walk().
1810 * Well, almost - if the last component is normal we get its copy
1811 * stored in nd->last.name and we will have to putname() it when we
1812 * are done. Procfs-like symlinks just set LAST_BIND.
1813 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001814 nd.flags |= LOOKUP_PARENT;
1815 error = security_inode_follow_link(path.dentry, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816 if (error)
1817 goto exit_dput;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001818 error = __do_follow_link(&path, &nd);
Kirill Korotaevde459212006-07-14 00:23:49 -07001819 if (error) {
1820 /* Does someone understand code flow here? Or it is only
1821 * me so stupid? Anathema to whoever designed this non-sense
1822 * with "intent.open".
1823 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001824 release_open_intent(&nd);
1825 return ERR_PTR(error);
Kirill Korotaevde459212006-07-14 00:23:49 -07001826 }
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001827 nd.flags &= ~LOOKUP_PARENT;
1828 if (nd.last_type == LAST_BIND)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829 goto ok;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830 error = -EISDIR;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001831 if (nd.last_type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 goto exit;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001833 if (nd.last.name[nd.last.len]) {
1834 __putname(nd.last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 goto exit;
1836 }
1837 error = -ELOOP;
1838 if (count++==32) {
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001839 __putname(nd.last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840 goto exit;
1841 }
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001842 dir = nd.path.dentry;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001843 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001844 path.dentry = lookup_hash(&nd);
1845 path.mnt = nd.path.mnt;
1846 __putname(nd.last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 goto do_last;
1848}
1849
1850/**
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001851 * filp_open - open file and return file pointer
1852 *
1853 * @filename: path to open
1854 * @flags: open flags as per the open(2) second argument
1855 * @mode: mode for the new file if O_CREAT is set, else ignored
1856 *
1857 * This is the helper to open a file from kernelspace if you really
1858 * have to. But in generally you should not do this, so please move
1859 * along, nothing to see here..
1860 */
1861struct file *filp_open(const char *filename, int flags, int mode)
1862{
1863 return do_filp_open(AT_FDCWD, filename, flags, mode);
1864}
1865EXPORT_SYMBOL(filp_open);
1866
1867/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868 * lookup_create - lookup a dentry, creating it if it doesn't exist
1869 * @nd: nameidata info
1870 * @is_dir: directory flag
1871 *
1872 * Simple function to lookup and return a dentry and create it
1873 * if it doesn't exist. Is SMP-safe.
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001874 *
Jan Blunck4ac91372008-02-14 19:34:32 -08001875 * Returns with nd->path.dentry->d_inode->i_mutex locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876 */
1877struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1878{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001879 struct dentry *dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880
Jan Blunck4ac91372008-02-14 19:34:32 -08001881 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001882 /*
1883 * Yucky last component or no last component at all?
1884 * (foo/., foo/.., /////)
1885 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 if (nd->last_type != LAST_NORM)
1887 goto fail;
1888 nd->flags &= ~LOOKUP_PARENT;
Al Viro35165862008-08-05 03:00:49 -04001889 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
ASANO Masahiroa6349042006-08-22 20:06:02 -04001890 nd->intent.open.flags = O_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001891
1892 /*
1893 * Do the final lookup.
1894 */
Christoph Hellwig49705b72005-11-08 21:35:06 -08001895 dentry = lookup_hash(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896 if (IS_ERR(dentry))
1897 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001898
Al Viroe9baf6e2008-05-15 04:49:12 -04001899 if (dentry->d_inode)
1900 goto eexist;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001901 /*
1902 * Special case - lookup gave negative, but... we had foo/bar/
1903 * From the vfs_mknod() POV we just have a negative dentry -
1904 * all is fine. Let's be bastards - you had / on the end, you've
1905 * been asking for (non-existent) directory. -ENOENT for you.
1906 */
Al Viroe9baf6e2008-05-15 04:49:12 -04001907 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1908 dput(dentry);
1909 dentry = ERR_PTR(-ENOENT);
1910 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911 return dentry;
Al Viroe9baf6e2008-05-15 04:49:12 -04001912eexist:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 dput(dentry);
Al Viroe9baf6e2008-05-15 04:49:12 -04001914 dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915fail:
1916 return dentry;
1917}
Christoph Hellwigf81a0bf2005-05-19 12:26:43 -07001918EXPORT_SYMBOL_GPL(lookup_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919
1920int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1921{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001922 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923
1924 if (error)
1925 return error;
1926
1927 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1928 return -EPERM;
1929
Al Viroacfa4382008-12-04 10:06:33 -05001930 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 return -EPERM;
1932
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07001933 error = devcgroup_inode_mknod(mode, dev);
1934 if (error)
1935 return error;
1936
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 error = security_inode_mknod(dir, dentry, mode, dev);
1938 if (error)
1939 return error;
1940
1941 DQUOT_INIT(dir);
1942 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001943 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001944 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 return error;
1946}
1947
Dave Hansen463c3192008-02-15 14:37:57 -08001948static int may_mknod(mode_t mode)
1949{
1950 switch (mode & S_IFMT) {
1951 case S_IFREG:
1952 case S_IFCHR:
1953 case S_IFBLK:
1954 case S_IFIFO:
1955 case S_IFSOCK:
1956 case 0: /* zero mode translates to S_IFREG */
1957 return 0;
1958 case S_IFDIR:
1959 return -EPERM;
1960 default:
1961 return -EINVAL;
1962 }
1963}
1964
Heiko Carstens2e4d0922009-01-14 14:14:31 +01001965SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
1966 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967{
Al Viro2ad94ae2008-07-21 09:32:51 -04001968 int error;
1969 char *tmp;
1970 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 struct nameidata nd;
1972
1973 if (S_ISDIR(mode))
1974 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975
Al Viro2ad94ae2008-07-21 09:32:51 -04001976 error = user_path_parent(dfd, filename, &nd, &tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04001978 return error;
1979
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 dentry = lookup_create(&nd, 0);
Dave Hansen463c3192008-02-15 14:37:57 -08001981 if (IS_ERR(dentry)) {
1982 error = PTR_ERR(dentry);
1983 goto out_unlock;
1984 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001985 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 mode &= ~current->fs->umask;
Dave Hansen463c3192008-02-15 14:37:57 -08001987 error = may_mknod(mode);
1988 if (error)
1989 goto out_dput;
1990 error = mnt_want_write(nd.path.mnt);
1991 if (error)
1992 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001993 error = security_path_mknod(&nd.path, dentry, mode, dev);
1994 if (error)
1995 goto out_drop_write;
Dave Hansen463c3192008-02-15 14:37:57 -08001996 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997 case 0: case S_IFREG:
Jan Blunck4ac91372008-02-14 19:34:32 -08001998 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999 break;
2000 case S_IFCHR: case S_IFBLK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002001 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 new_decode_dev(dev));
2003 break;
2004 case S_IFIFO: case S_IFSOCK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002005 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002008out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002009 mnt_drop_write(nd.path.mnt);
2010out_dput:
2011 dput(dentry);
2012out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002013 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002014 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 putname(tmp);
2016
2017 return error;
2018}
2019
Heiko Carstens3480b252009-01-14 14:14:16 +01002020SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002021{
2022 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2023}
2024
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2026{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002027 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028
2029 if (error)
2030 return error;
2031
Al Viroacfa4382008-12-04 10:06:33 -05002032 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 return -EPERM;
2034
2035 mode &= (S_IRWXUGO|S_ISVTX);
2036 error = security_inode_mkdir(dir, dentry, mode);
2037 if (error)
2038 return error;
2039
2040 DQUOT_INIT(dir);
2041 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002042 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002043 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 return error;
2045}
2046
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002047SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048{
2049 int error = 0;
2050 char * tmp;
Dave Hansen6902d922006-09-30 23:29:01 -07002051 struct dentry *dentry;
2052 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053
Al Viro2ad94ae2008-07-21 09:32:51 -04002054 error = user_path_parent(dfd, pathname, &nd, &tmp);
2055 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002056 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057
Dave Hansen6902d922006-09-30 23:29:01 -07002058 dentry = lookup_create(&nd, 1);
2059 error = PTR_ERR(dentry);
2060 if (IS_ERR(dentry))
2061 goto out_unlock;
2062
Jan Blunck4ac91372008-02-14 19:34:32 -08002063 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Dave Hansen6902d922006-09-30 23:29:01 -07002064 mode &= ~current->fs->umask;
Dave Hansen463c3192008-02-15 14:37:57 -08002065 error = mnt_want_write(nd.path.mnt);
2066 if (error)
2067 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002068 error = security_path_mkdir(&nd.path, dentry, mode);
2069 if (error)
2070 goto out_drop_write;
Jan Blunck4ac91372008-02-14 19:34:32 -08002071 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002072out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002073 mnt_drop_write(nd.path.mnt);
2074out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002075 dput(dentry);
2076out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002077 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002078 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002079 putname(tmp);
2080out_err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081 return error;
2082}
2083
Heiko Carstens3cdad422009-01-14 14:14:22 +01002084SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002085{
2086 return sys_mkdirat(AT_FDCWD, pathname, mode);
2087}
2088
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089/*
2090 * We try to drop the dentry early: we should have
2091 * a usage count of 2 if we're the only user of this
2092 * dentry, and if that is true (possibly after pruning
2093 * the dcache), then we drop the dentry now.
2094 *
2095 * A low-level filesystem can, if it choses, legally
2096 * do a
2097 *
2098 * if (!d_unhashed(dentry))
2099 * return -EBUSY;
2100 *
2101 * if it cannot handle the case of removing a directory
2102 * that is still in use by something else..
2103 */
2104void dentry_unhash(struct dentry *dentry)
2105{
2106 dget(dentry);
Vasily Averindc168422006-12-06 20:37:07 -08002107 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108 spin_lock(&dcache_lock);
2109 spin_lock(&dentry->d_lock);
2110 if (atomic_read(&dentry->d_count) == 2)
2111 __d_drop(dentry);
2112 spin_unlock(&dentry->d_lock);
2113 spin_unlock(&dcache_lock);
2114}
2115
2116int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2117{
2118 int error = may_delete(dir, dentry, 1);
2119
2120 if (error)
2121 return error;
2122
Al Viroacfa4382008-12-04 10:06:33 -05002123 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 return -EPERM;
2125
2126 DQUOT_INIT(dir);
2127
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002128 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 dentry_unhash(dentry);
2130 if (d_mountpoint(dentry))
2131 error = -EBUSY;
2132 else {
2133 error = security_inode_rmdir(dir, dentry);
2134 if (!error) {
2135 error = dir->i_op->rmdir(dir, dentry);
2136 if (!error)
2137 dentry->d_inode->i_flags |= S_DEAD;
2138 }
2139 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002140 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142 d_delete(dentry);
2143 }
2144 dput(dentry);
2145
2146 return error;
2147}
2148
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002149static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150{
2151 int error = 0;
2152 char * name;
2153 struct dentry *dentry;
2154 struct nameidata nd;
2155
Al Viro2ad94ae2008-07-21 09:32:51 -04002156 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002158 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159
2160 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002161 case LAST_DOTDOT:
2162 error = -ENOTEMPTY;
2163 goto exit1;
2164 case LAST_DOT:
2165 error = -EINVAL;
2166 goto exit1;
2167 case LAST_ROOT:
2168 error = -EBUSY;
2169 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002171
2172 nd.flags &= ~LOOKUP_PARENT;
2173
Jan Blunck4ac91372008-02-14 19:34:32 -08002174 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002175 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002177 if (IS_ERR(dentry))
2178 goto exit2;
Dave Hansen06227532008-02-15 14:37:34 -08002179 error = mnt_want_write(nd.path.mnt);
2180 if (error)
2181 goto exit3;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002182 error = security_path_rmdir(&nd.path, dentry);
2183 if (error)
2184 goto exit4;
Jan Blunck4ac91372008-02-14 19:34:32 -08002185 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002186exit4:
Dave Hansen06227532008-02-15 14:37:34 -08002187 mnt_drop_write(nd.path.mnt);
2188exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07002189 dput(dentry);
2190exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08002191 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002193 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 putname(name);
2195 return error;
2196}
2197
Heiko Carstens3cdad422009-01-14 14:14:22 +01002198SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002199{
2200 return do_rmdir(AT_FDCWD, pathname);
2201}
2202
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203int vfs_unlink(struct inode *dir, struct dentry *dentry)
2204{
2205 int error = may_delete(dir, dentry, 0);
2206
2207 if (error)
2208 return error;
2209
Al Viroacfa4382008-12-04 10:06:33 -05002210 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211 return -EPERM;
2212
2213 DQUOT_INIT(dir);
2214
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002215 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216 if (d_mountpoint(dentry))
2217 error = -EBUSY;
2218 else {
2219 error = security_inode_unlink(dir, dentry);
2220 if (!error)
2221 error = dir->i_op->unlink(dir, dentry);
2222 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002223 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224
2225 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2226 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08002227 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229 }
Robert Love0eeca282005-07-12 17:06:03 -04002230
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231 return error;
2232}
2233
2234/*
2235 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002236 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 * writeout happening, and we don't want to prevent access to the directory
2238 * while waiting on the I/O.
2239 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002240static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241{
Al Viro2ad94ae2008-07-21 09:32:51 -04002242 int error;
2243 char *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244 struct dentry *dentry;
2245 struct nameidata nd;
2246 struct inode *inode = NULL;
2247
Al Viro2ad94ae2008-07-21 09:32:51 -04002248 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002250 return error;
2251
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252 error = -EISDIR;
2253 if (nd.last_type != LAST_NORM)
2254 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002255
2256 nd.flags &= ~LOOKUP_PARENT;
2257
Jan Blunck4ac91372008-02-14 19:34:32 -08002258 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002259 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260 error = PTR_ERR(dentry);
2261 if (!IS_ERR(dentry)) {
2262 /* Why not before? Because we want correct error value */
2263 if (nd.last.name[nd.last.len])
2264 goto slashes;
2265 inode = dentry->d_inode;
2266 if (inode)
2267 atomic_inc(&inode->i_count);
Dave Hansen06227532008-02-15 14:37:34 -08002268 error = mnt_want_write(nd.path.mnt);
2269 if (error)
2270 goto exit2;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002271 error = security_path_unlink(&nd.path, dentry);
2272 if (error)
2273 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08002274 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002275exit3:
Dave Hansen06227532008-02-15 14:37:34 -08002276 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277 exit2:
2278 dput(dentry);
2279 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002280 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 if (inode)
2282 iput(inode); /* truncate the inode here */
2283exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002284 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285 putname(name);
2286 return error;
2287
2288slashes:
2289 error = !dentry->d_inode ? -ENOENT :
2290 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2291 goto exit2;
2292}
2293
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002294SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002295{
2296 if ((flag & ~AT_REMOVEDIR) != 0)
2297 return -EINVAL;
2298
2299 if (flag & AT_REMOVEDIR)
2300 return do_rmdir(dfd, pathname);
2301
2302 return do_unlinkat(dfd, pathname);
2303}
2304
Heiko Carstens3480b252009-01-14 14:14:16 +01002305SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002306{
2307 return do_unlinkat(AT_FDCWD, pathname);
2308}
2309
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002310int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002312 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313
2314 if (error)
2315 return error;
2316
Al Viroacfa4382008-12-04 10:06:33 -05002317 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318 return -EPERM;
2319
2320 error = security_inode_symlink(dir, dentry, oldname);
2321 if (error)
2322 return error;
2323
2324 DQUOT_INIT(dir);
2325 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002326 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002327 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328 return error;
2329}
2330
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002331SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2332 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333{
Al Viro2ad94ae2008-07-21 09:32:51 -04002334 int error;
2335 char *from;
2336 char *to;
Dave Hansen6902d922006-09-30 23:29:01 -07002337 struct dentry *dentry;
2338 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339
2340 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04002341 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342 return PTR_ERR(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002343
2344 error = user_path_parent(newdfd, newname, &nd, &to);
2345 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002346 goto out_putname;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347
Dave Hansen6902d922006-09-30 23:29:01 -07002348 dentry = lookup_create(&nd, 0);
2349 error = PTR_ERR(dentry);
2350 if (IS_ERR(dentry))
2351 goto out_unlock;
2352
Dave Hansen75c3f292008-02-15 14:37:45 -08002353 error = mnt_want_write(nd.path.mnt);
2354 if (error)
2355 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002356 error = security_path_symlink(&nd.path, dentry, from);
2357 if (error)
2358 goto out_drop_write;
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002359 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002360out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002361 mnt_drop_write(nd.path.mnt);
2362out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002363 dput(dentry);
2364out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002365 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002366 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002367 putname(to);
2368out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369 putname(from);
2370 return error;
2371}
2372
Heiko Carstens3480b252009-01-14 14:14:16 +01002373SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002374{
2375 return sys_symlinkat(oldname, AT_FDCWD, newname);
2376}
2377
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2379{
2380 struct inode *inode = old_dentry->d_inode;
2381 int error;
2382
2383 if (!inode)
2384 return -ENOENT;
2385
Miklos Szeredia95164d2008-07-30 15:08:48 +02002386 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387 if (error)
2388 return error;
2389
2390 if (dir->i_sb != inode->i_sb)
2391 return -EXDEV;
2392
2393 /*
2394 * A link to an append-only or immutable file cannot be created.
2395 */
2396 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2397 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05002398 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002400 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401 return -EPERM;
2402
2403 error = security_inode_link(old_dentry, dir, new_dentry);
2404 if (error)
2405 return error;
2406
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002407 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408 DQUOT_INIT(dir);
2409 error = dir->i_op->link(old_dentry, dir, new_dentry);
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002410 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002411 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002412 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 return error;
2414}
2415
2416/*
2417 * Hardlinks are often used in delicate situations. We avoid
2418 * security-related surprises by not following symlinks on the
2419 * newname. --KAB
2420 *
2421 * We don't follow them on the oldname either to be compatible
2422 * with linux 2.0, and to avoid hard-linking to directories
2423 * and other special files. --ADM
2424 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002425SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2426 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427{
2428 struct dentry *new_dentry;
Al Viro2d8f3032008-07-22 09:59:21 -04002429 struct nameidata nd;
2430 struct path old_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431 int error;
Al Viro2ad94ae2008-07-21 09:32:51 -04002432 char *to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433
Ulrich Drepper45c9b112006-06-25 05:49:11 -07002434 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002435 return -EINVAL;
2436
Al Viro2d8f3032008-07-22 09:59:21 -04002437 error = user_path_at(olddfd, oldname,
2438 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2439 &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002441 return error;
2442
2443 error = user_path_parent(newdfd, newname, &nd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444 if (error)
2445 goto out;
2446 error = -EXDEV;
Al Viro2d8f3032008-07-22 09:59:21 -04002447 if (old_path.mnt != nd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448 goto out_release;
2449 new_dentry = lookup_create(&nd, 0);
2450 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002451 if (IS_ERR(new_dentry))
2452 goto out_unlock;
Dave Hansen75c3f292008-02-15 14:37:45 -08002453 error = mnt_want_write(nd.path.mnt);
2454 if (error)
2455 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002456 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2457 if (error)
2458 goto out_drop_write;
Al Viro2d8f3032008-07-22 09:59:21 -04002459 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002460out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002461 mnt_drop_write(nd.path.mnt);
2462out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002463 dput(new_dentry);
2464out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002465 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466out_release:
Jan Blunck1d957f92008-02-14 19:34:35 -08002467 path_put(&nd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002468 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469out:
Al Viro2d8f3032008-07-22 09:59:21 -04002470 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471
2472 return error;
2473}
2474
Heiko Carstens3480b252009-01-14 14:14:16 +01002475SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002476{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002477 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002478}
2479
Linus Torvalds1da177e2005-04-16 15:20:36 -07002480/*
2481 * The worst of all namespace operations - renaming directory. "Perverted"
2482 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2483 * Problems:
2484 * a) we can get into loop creation. Check is done in is_subdir().
2485 * b) race potential - two innocent renames can create a loop together.
2486 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002487 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488 * story.
2489 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002490 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491 * whether the target exists). Solution: try to be smart with locking
2492 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002493 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494 * move will be locked. Thus we can rank directories by the tree
2495 * (ancestors first) and rank all non-directories after them.
2496 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002497 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498 * HOWEVER, it relies on the assumption that any object with ->lookup()
2499 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2500 * we'd better make sure that there's no link(2) for them.
2501 * d) some filesystems don't support opened-but-unlinked directories,
2502 * either because of layout or because they are not ready to deal with
2503 * all cases correctly. The latter will be fixed (taking this sort of
2504 * stuff into VFS), but the former is not going away. Solution: the same
2505 * trick as in rmdir().
2506 * e) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002507 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002509 * ->i_mutex on parents, which works but leads to some truely excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510 * locking].
2511 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07002512static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2513 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514{
2515 int error = 0;
2516 struct inode *target;
2517
2518 /*
2519 * If we are going to change the parent - check write permissions,
2520 * we'll need to flip '..'.
2521 */
2522 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04002523 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 if (error)
2525 return error;
2526 }
2527
2528 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2529 if (error)
2530 return error;
2531
2532 target = new_dentry->d_inode;
2533 if (target) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002534 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535 dentry_unhash(new_dentry);
2536 }
2537 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2538 error = -EBUSY;
2539 else
2540 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2541 if (target) {
2542 if (!error)
2543 target->i_flags |= S_DEAD;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002544 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545 if (d_unhashed(new_dentry))
2546 d_rehash(new_dentry);
2547 dput(new_dentry);
2548 }
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002549 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07002550 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2551 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552 return error;
2553}
2554
Adrian Bunk75c96f82005-05-05 16:16:09 -07002555static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2556 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557{
2558 struct inode *target;
2559 int error;
2560
2561 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2562 if (error)
2563 return error;
2564
2565 dget(new_dentry);
2566 target = new_dentry->d_inode;
2567 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002568 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2570 error = -EBUSY;
2571 else
2572 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2573 if (!error) {
Mark Fasheh349457c2006-09-08 14:22:21 -07002574 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575 d_move(old_dentry, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002576 }
2577 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002578 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 dput(new_dentry);
2580 return error;
2581}
2582
2583int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2584 struct inode *new_dir, struct dentry *new_dentry)
2585{
2586 int error;
2587 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Robert Love0eeca282005-07-12 17:06:03 -04002588 const char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589
2590 if (old_dentry->d_inode == new_dentry->d_inode)
2591 return 0;
2592
2593 error = may_delete(old_dir, old_dentry, is_dir);
2594 if (error)
2595 return error;
2596
2597 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02002598 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599 else
2600 error = may_delete(new_dir, new_dentry, is_dir);
2601 if (error)
2602 return error;
2603
Al Viroacfa4382008-12-04 10:06:33 -05002604 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605 return -EPERM;
2606
2607 DQUOT_INIT(old_dir);
2608 DQUOT_INIT(new_dir);
2609
Robert Love0eeca282005-07-12 17:06:03 -04002610 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2611
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612 if (is_dir)
2613 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2614 else
2615 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2616 if (!error) {
Robert Love0eeca282005-07-12 17:06:03 -04002617 const char *new_name = old_dentry->d_name.name;
John McCutchan89204c42005-08-15 12:13:28 -04002618 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04002619 new_dentry->d_inode, old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620 }
Robert Love0eeca282005-07-12 17:06:03 -04002621 fsnotify_oldname_free(old_name);
2622
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623 return error;
2624}
2625
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002626SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2627 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628{
Al Viro2ad94ae2008-07-21 09:32:51 -04002629 struct dentry *old_dir, *new_dir;
2630 struct dentry *old_dentry, *new_dentry;
2631 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632 struct nameidata oldnd, newnd;
Al Viro2ad94ae2008-07-21 09:32:51 -04002633 char *from;
2634 char *to;
2635 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636
Al Viro2ad94ae2008-07-21 09:32:51 -04002637 error = user_path_parent(olddfd, oldname, &oldnd, &from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638 if (error)
2639 goto exit;
2640
Al Viro2ad94ae2008-07-21 09:32:51 -04002641 error = user_path_parent(newdfd, newname, &newnd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642 if (error)
2643 goto exit1;
2644
2645 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08002646 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647 goto exit2;
2648
Jan Blunck4ac91372008-02-14 19:34:32 -08002649 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650 error = -EBUSY;
2651 if (oldnd.last_type != LAST_NORM)
2652 goto exit2;
2653
Jan Blunck4ac91372008-02-14 19:34:32 -08002654 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655 if (newnd.last_type != LAST_NORM)
2656 goto exit2;
2657
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002658 oldnd.flags &= ~LOOKUP_PARENT;
2659 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09002660 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002661
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662 trap = lock_rename(new_dir, old_dir);
2663
Christoph Hellwig49705b72005-11-08 21:35:06 -08002664 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665 error = PTR_ERR(old_dentry);
2666 if (IS_ERR(old_dentry))
2667 goto exit3;
2668 /* source must exist */
2669 error = -ENOENT;
2670 if (!old_dentry->d_inode)
2671 goto exit4;
2672 /* unless the source is a directory trailing slashes give -ENOTDIR */
2673 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2674 error = -ENOTDIR;
2675 if (oldnd.last.name[oldnd.last.len])
2676 goto exit4;
2677 if (newnd.last.name[newnd.last.len])
2678 goto exit4;
2679 }
2680 /* source should not be ancestor of target */
2681 error = -EINVAL;
2682 if (old_dentry == trap)
2683 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08002684 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002685 error = PTR_ERR(new_dentry);
2686 if (IS_ERR(new_dentry))
2687 goto exit4;
2688 /* target should not be an ancestor of source */
2689 error = -ENOTEMPTY;
2690 if (new_dentry == trap)
2691 goto exit5;
2692
Dave Hansen9079b1e2008-02-15 14:37:49 -08002693 error = mnt_want_write(oldnd.path.mnt);
2694 if (error)
2695 goto exit5;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002696 error = security_path_rename(&oldnd.path, old_dentry,
2697 &newnd.path, new_dentry);
2698 if (error)
2699 goto exit6;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700 error = vfs_rename(old_dir->d_inode, old_dentry,
2701 new_dir->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002702exit6:
Dave Hansen9079b1e2008-02-15 14:37:49 -08002703 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002704exit5:
2705 dput(new_dentry);
2706exit4:
2707 dput(old_dentry);
2708exit3:
2709 unlock_rename(new_dir, old_dir);
2710exit2:
Jan Blunck1d957f92008-02-14 19:34:35 -08002711 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002712 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002714 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715 putname(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002716exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002717 return error;
2718}
2719
Heiko Carstensa26eab22009-01-14 14:14:17 +01002720SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002721{
2722 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2723}
2724
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2726{
2727 int len;
2728
2729 len = PTR_ERR(link);
2730 if (IS_ERR(link))
2731 goto out;
2732
2733 len = strlen(link);
2734 if (len > (unsigned) buflen)
2735 len = buflen;
2736 if (copy_to_user(buffer, link, len))
2737 len = -EFAULT;
2738out:
2739 return len;
2740}
2741
2742/*
2743 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2744 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2745 * using) it for any given inode is up to filesystem.
2746 */
2747int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2748{
2749 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002750 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07002751 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002752
Linus Torvalds1da177e2005-04-16 15:20:36 -07002753 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002754 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07002755 if (IS_ERR(cookie))
2756 return PTR_ERR(cookie);
2757
2758 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2759 if (dentry->d_inode->i_op->put_link)
2760 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2761 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002762}
2763
2764int vfs_follow_link(struct nameidata *nd, const char *link)
2765{
2766 return __vfs_follow_link(nd, link);
2767}
2768
2769/* get the link contents into pagecache */
2770static char *page_getlink(struct dentry * dentry, struct page **ppage)
2771{
Duane Griffinebd09ab2008-12-19 20:47:12 +00002772 char *kaddr;
2773 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07002775 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07002777 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002778 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00002779 kaddr = kmap(page);
2780 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2781 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782}
2783
2784int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2785{
2786 struct page *page = NULL;
2787 char *s = page_getlink(dentry, &page);
2788 int res = vfs_readlink(dentry,buffer,buflen,s);
2789 if (page) {
2790 kunmap(page);
2791 page_cache_release(page);
2792 }
2793 return res;
2794}
2795
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002796void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002798 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002799 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002800 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002801}
2802
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002803void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002804{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002805 struct page *page = cookie;
2806
2807 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002808 kunmap(page);
2809 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002810 }
2811}
2812
Nick Piggin54566b22009-01-04 12:00:53 -08002813/*
2814 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2815 */
2816int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817{
2818 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002819 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002820 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08002821 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08002823 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2824 if (nofs)
2825 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826
NeilBrown7e53cac2006-03-25 03:07:57 -08002827retry:
Nick Pigginafddba42007-10-16 01:25:01 -07002828 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08002829 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07002831 goto fail;
2832
Linus Torvalds1da177e2005-04-16 15:20:36 -07002833 kaddr = kmap_atomic(page, KM_USER0);
2834 memcpy(kaddr, symname, len-1);
2835 kunmap_atomic(kaddr, KM_USER0);
Nick Pigginafddba42007-10-16 01:25:01 -07002836
2837 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2838 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839 if (err < 0)
2840 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07002841 if (err < len-1)
2842 goto retry;
2843
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844 mark_inode_dirty(inode);
2845 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002846fail:
2847 return err;
2848}
2849
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002850int page_symlink(struct inode *inode, const char *symname, int len)
2851{
2852 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08002853 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002854}
2855
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002856const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857 .readlink = generic_readlink,
2858 .follow_link = page_follow_link_light,
2859 .put_link = page_put_link,
2860};
2861
Al Viro2d8f3032008-07-22 09:59:21 -04002862EXPORT_SYMBOL(user_path_at);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863EXPORT_SYMBOL(follow_down);
2864EXPORT_SYMBOL(follow_up);
2865EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2866EXPORT_SYMBOL(getname);
2867EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868EXPORT_SYMBOL(lookup_one_len);
2869EXPORT_SYMBOL(page_follow_link_light);
2870EXPORT_SYMBOL(page_put_link);
2871EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002872EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002873EXPORT_SYMBOL(page_symlink);
2874EXPORT_SYMBOL(page_symlink_inode_operations);
2875EXPORT_SYMBOL(path_lookup);
Al Virod1811462008-08-02 00:49:18 -04002876EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002877EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04002878EXPORT_SYMBOL(inode_permission);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -08002879EXPORT_SYMBOL(file_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002880EXPORT_SYMBOL(unlock_rename);
2881EXPORT_SYMBOL(vfs_create);
2882EXPORT_SYMBOL(vfs_follow_link);
2883EXPORT_SYMBOL(vfs_link);
2884EXPORT_SYMBOL(vfs_mkdir);
2885EXPORT_SYMBOL(vfs_mknod);
2886EXPORT_SYMBOL(generic_permission);
2887EXPORT_SYMBOL(vfs_readlink);
2888EXPORT_SYMBOL(vfs_rename);
2889EXPORT_SYMBOL(vfs_rmdir);
2890EXPORT_SYMBOL(vfs_symlink);
2891EXPORT_SYMBOL(vfs_unlink);
2892EXPORT_SYMBOL(dentry_unhash);
2893EXPORT_SYMBOL(generic_readlink);
Al Viro18d8fda2008-12-26 00:35:37 -05002894
2895/* to be mentioned only in INIT_TASK */
2896struct fs_struct init_fs = {
2897 .count = ATOMIC_INIT(1),
Steven Rostedt1239f262008-12-10 18:37:28 -05002898 .lock = __RW_LOCK_UNLOCKED(init_fs.lock),
Al Viro18d8fda2008-12-26 00:35:37 -05002899 .umask = 0022,
2900};