blob: d040ce11785d6acb115154da5e83a7500909df13 [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>
Mimi Zohar6146f0d2009-02-04 09:06:57 -050027#include <linux/ima.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/syscalls.h>
29#include <linux/mount.h>
30#include <linux/audit.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080031#include <linux/capability.h>
Trond Myklebust834f2a42005-10-18 14:20:16 -070032#include <linux/file.h>
Ulrich Drepper5590ff02006-01-18 17:43:53 -080033#include <linux/fcntl.h>
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -070034#include <linux/device_cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <asm/uaccess.h>
36
37#define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
38
39/* [Feb-1997 T. Schoebel-Theuer]
40 * Fundamental changes in the pathname lookup mechanisms (namei)
41 * were necessary because of omirr. The reason is that omirr needs
42 * to know the _real_ pathname, not the user-supplied one, in case
43 * of symlinks (and also when transname replacements occur).
44 *
45 * The new code replaces the old recursive symlink resolution with
46 * an iterative one (in case of non-nested symlink chains). It does
47 * this with calls to <fs>_follow_link().
48 * As a side effect, dir_namei(), _namei() and follow_link() are now
49 * replaced with a single function lookup_dentry() that can handle all
50 * the special cases of the former code.
51 *
52 * With the new dcache, the pathname is stored at each inode, at least as
53 * long as the refcount of the inode is positive. As a side effect, the
54 * size of the dcache depends on the inode cache and thus is dynamic.
55 *
56 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57 * resolution to correspond with current state of the code.
58 *
59 * Note that the symlink resolution is not *completely* iterative.
60 * There is still a significant amount of tail- and mid- recursion in
61 * the algorithm. Also, note that <fs>_readlink() is not used in
62 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63 * may return different results than <fs>_follow_link(). Many virtual
64 * filesystems (including /proc) exhibit this behavior.
65 */
66
67/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69 * and the name already exists in form of a symlink, try to create the new
70 * name indicated by the symlink. The old code always complained that the
71 * name already exists, due to not following the symlink even if its target
72 * is nonexistent. The new semantics affects also mknod() and link() when
73 * the name is a symlink pointing to a non-existant name.
74 *
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
82 */
83
84/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
86 *
87 * [10-Sep-98 Alan Modra] Another symlink change.
88 */
89
90/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
97 *
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
103 */
104/*
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800106 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 * any extra contention...
108 */
109
Al Viroa02f76c2008-02-23 15:14:28 +0000110static int __link_path_walk(const char *name, struct nameidata *nd);
Josef 'Jeff' Sipekc4a78082007-07-19 01:48:22 -0700111
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112/* In order to reduce some races, while at the same time doing additional
113 * checking and hopefully speeding things up, we copy filenames to the
114 * kernel data space before using them..
115 *
116 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
117 * PATH_MAX includes the nul terminator --RR.
118 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800119static int do_getname(const char __user *filename, char *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120{
121 int retval;
122 unsigned long len = PATH_MAX;
123
124 if (!segment_eq(get_fs(), KERNEL_DS)) {
125 if ((unsigned long) filename >= TASK_SIZE)
126 return -EFAULT;
127 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
128 len = TASK_SIZE - (unsigned long) filename;
129 }
130
131 retval = strncpy_from_user(page, filename, len);
132 if (retval > 0) {
133 if (retval < len)
134 return 0;
135 return -ENAMETOOLONG;
136 } else if (!retval)
137 retval = -ENOENT;
138 return retval;
139}
140
141char * getname(const char __user * filename)
142{
143 char *tmp, *result;
144
145 result = ERR_PTR(-ENOMEM);
146 tmp = __getname();
147 if (tmp) {
148 int retval = do_getname(filename, tmp);
149
150 result = tmp;
151 if (retval < 0) {
152 __putname(tmp);
153 result = ERR_PTR(retval);
154 }
155 }
156 audit_getname(result);
157 return result;
158}
159
160#ifdef CONFIG_AUDITSYSCALL
161void putname(const char *name)
162{
Al Viro5ac3a9c2006-07-16 06:38:45 -0400163 if (unlikely(!audit_dummy_context()))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 audit_putname(name);
165 else
166 __putname(name);
167}
168EXPORT_SYMBOL(putname);
169#endif
170
171
172/**
173 * generic_permission - check for access rights on a Posix-like filesystem
174 * @inode: inode to check access rights for
175 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
176 * @check_acl: optional callback to check for Posix ACLs
177 *
178 * Used to check for read/write/execute permissions on a file.
179 * We use "fsuid" for this, letting us set arbitrary permissions
180 * for filesystem access without changing the "normal" uids which
181 * are used for other things..
182 */
183int generic_permission(struct inode *inode, int mask,
184 int (*check_acl)(struct inode *inode, int mask))
185{
186 umode_t mode = inode->i_mode;
187
Al Viroe6305c42008-07-15 21:03:57 -0400188 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
189
David Howellsda9592e2008-11-14 10:39:05 +1100190 if (current_fsuid() == inode->i_uid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 mode >>= 6;
192 else {
193 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
194 int error = check_acl(inode, mask);
195 if (error == -EACCES)
196 goto check_capabilities;
197 else if (error != -EAGAIN)
198 return error;
199 }
200
201 if (in_group_p(inode->i_gid))
202 mode >>= 3;
203 }
204
205 /*
206 * If the DACs are ok we don't need any capability check.
207 */
Al Viroe6305c42008-07-15 21:03:57 -0400208 if ((mask & ~mode) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 return 0;
210
211 check_capabilities:
212 /*
213 * Read/write DACs are always overridable.
214 * Executable DACs are overridable if at least one exec bit is set.
215 */
Miklos Szeredif696a362008-07-31 13:41:58 +0200216 if (!(mask & MAY_EXEC) || execute_ok(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 if (capable(CAP_DAC_OVERRIDE))
218 return 0;
219
220 /*
221 * Searching includes executable on directories, else just read.
222 */
223 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
224 if (capable(CAP_DAC_READ_SEARCH))
225 return 0;
226
227 return -EACCES;
228}
229
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200230/**
231 * inode_permission - check for access rights to a given inode
232 * @inode: inode to check permission on
233 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
234 *
235 * Used to check for read/write/execute permissions on an inode.
236 * We use "fsuid" for this, letting us set arbitrary permissions
237 * for filesystem access without changing the "normal" uids which
238 * are used for other things.
239 */
Al Virof419a2e2008-07-22 00:07:17 -0400240int inode_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241{
Al Viroe6305c42008-07-15 21:03:57 -0400242 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243
244 if (mask & MAY_WRITE) {
Miklos Szeredi22590e42007-10-16 23:27:08 -0700245 umode_t mode = inode->i_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246
247 /*
248 * Nobody gets write access to a read-only fs.
249 */
250 if (IS_RDONLY(inode) &&
251 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
252 return -EROFS;
253
254 /*
255 * Nobody gets write access to an immutable file.
256 */
257 if (IS_IMMUTABLE(inode))
258 return -EACCES;
259 }
260
Al Viroacfa4382008-12-04 10:06:33 -0500261 if (inode->i_op->permission)
Al Virob77b0642008-07-17 09:37:02 -0400262 retval = inode->i_op->permission(inode, mask);
Miklos Szeredif696a362008-07-31 13:41:58 +0200263 else
Al Viroe6305c42008-07-15 21:03:57 -0400264 retval = generic_permission(inode, mask, NULL);
Miklos Szeredif696a362008-07-31 13:41:58 +0200265
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 if (retval)
267 return retval;
268
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -0700269 retval = devcgroup_inode_permission(inode, mask);
270 if (retval)
271 return retval;
272
Al Viroe6305c42008-07-15 21:03:57 -0400273 return security_inode_permission(inode,
Stephen Smalleyf418b002008-07-28 13:32:38 -0400274 mask & (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275}
276
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800277/**
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800278 * file_permission - check for additional access rights to a given file
279 * @file: file to check access rights for
280 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
281 *
282 * Used to check for read/write/execute permissions on an already opened
283 * file.
284 *
285 * Note:
286 * Do not use this function in new code. All access checks should
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200287 * be done using inode_permission().
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800288 */
289int file_permission(struct file *file, int mask)
290{
Al Virof419a2e2008-07-22 00:07:17 -0400291 return inode_permission(file->f_path.dentry->d_inode, mask);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800292}
293
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294/*
295 * get_write_access() gets write permission for a file.
296 * put_write_access() releases this write permission.
297 * This is used for regular files.
298 * We cannot support write (and maybe mmap read-write shared) accesses and
299 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
300 * can have the following values:
301 * 0: no writers, no VM_DENYWRITE mappings
302 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
303 * > 0: (i_writecount) users are writing to the file.
304 *
305 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
306 * except for the cases where we don't hold i_writecount yet. Then we need to
307 * use {get,deny}_write_access() - these functions check the sign and refuse
308 * to do the change if sign is wrong. Exclusion between them is provided by
309 * the inode->i_lock spinlock.
310 */
311
312int get_write_access(struct inode * inode)
313{
314 spin_lock(&inode->i_lock);
315 if (atomic_read(&inode->i_writecount) < 0) {
316 spin_unlock(&inode->i_lock);
317 return -ETXTBSY;
318 }
319 atomic_inc(&inode->i_writecount);
320 spin_unlock(&inode->i_lock);
321
322 return 0;
323}
324
325int deny_write_access(struct file * file)
326{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800327 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328
329 spin_lock(&inode->i_lock);
330 if (atomic_read(&inode->i_writecount) > 0) {
331 spin_unlock(&inode->i_lock);
332 return -ETXTBSY;
333 }
334 atomic_dec(&inode->i_writecount);
335 spin_unlock(&inode->i_lock);
336
337 return 0;
338}
339
Jan Blunck1d957f92008-02-14 19:34:35 -0800340/**
Jan Blunck5dd784d02008-02-14 19:34:38 -0800341 * path_get - get a reference to a path
342 * @path: path to get the reference to
343 *
344 * Given a path increment the reference count to the dentry and the vfsmount.
345 */
346void path_get(struct path *path)
347{
348 mntget(path->mnt);
349 dget(path->dentry);
350}
351EXPORT_SYMBOL(path_get);
352
353/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800354 * path_put - put a reference to a path
355 * @path: path to put the reference to
356 *
357 * Given a path decrement the reference count to the dentry and the vfsmount.
358 */
359void path_put(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360{
Jan Blunck1d957f92008-02-14 19:34:35 -0800361 dput(path->dentry);
362 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363}
Jan Blunck1d957f92008-02-14 19:34:35 -0800364EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365
Trond Myklebust834f2a42005-10-18 14:20:16 -0700366/**
367 * release_open_intent - free up open intent resources
368 * @nd: pointer to nameidata
369 */
370void release_open_intent(struct nameidata *nd)
371{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800372 if (nd->intent.open.file->f_path.dentry == NULL)
Trond Myklebust834f2a42005-10-18 14:20:16 -0700373 put_filp(nd->intent.open.file);
374 else
375 fput(nd->intent.open.file);
376}
377
Ian Kentbcdc5e02006-09-27 01:50:44 -0700378static inline struct dentry *
379do_revalidate(struct dentry *dentry, struct nameidata *nd)
380{
381 int status = dentry->d_op->d_revalidate(dentry, nd);
382 if (unlikely(status <= 0)) {
383 /*
384 * The dentry failed validation.
385 * If d_revalidate returned 0 attempt to invalidate
386 * the dentry otherwise d_revalidate is asking us
387 * to return a fail status.
388 */
389 if (!status) {
390 if (!d_invalidate(dentry)) {
391 dput(dentry);
392 dentry = NULL;
393 }
394 } else {
395 dput(dentry);
396 dentry = ERR_PTR(status);
397 }
398 }
399 return dentry;
400}
401
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402/*
403 * Internal lookup() using the new generic dcache.
404 * SMP-safe
405 */
406static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
407{
408 struct dentry * dentry = __d_lookup(parent, name);
409
410 /* lockess __d_lookup may fail due to concurrent d_move()
411 * in some unrelated directory, so try with d_lookup
412 */
413 if (!dentry)
414 dentry = d_lookup(parent, name);
415
Ian Kentbcdc5e02006-09-27 01:50:44 -0700416 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
417 dentry = do_revalidate(dentry, nd);
418
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 return dentry;
420}
421
422/*
423 * Short-cut version of permission(), for calling by
424 * path_walk(), when dcache lock is held. Combines parts
425 * of permission() and generic_permission(), and tests ONLY for
426 * MAY_EXEC permission.
427 *
428 * If appropriate, check DAC only. If not appropriate, or
429 * short-cut DAC fails, then call permission() to do more
430 * complete permission check.
431 */
Al Viro672b16b2008-07-17 09:45:01 -0400432static int exec_permission_lite(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433{
434 umode_t mode = inode->i_mode;
435
Al Viroacfa4382008-12-04 10:06:33 -0500436 if (inode->i_op->permission)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 return -EAGAIN;
438
David Howellsda9592e2008-11-14 10:39:05 +1100439 if (current_fsuid() == inode->i_uid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 mode >>= 6;
441 else if (in_group_p(inode->i_gid))
442 mode >>= 3;
443
444 if (mode & MAY_EXEC)
445 goto ok;
446
447 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
448 goto ok;
449
450 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
451 goto ok;
452
453 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
454 goto ok;
455
456 return -EACCES;
457ok:
Al Virob77b0642008-07-17 09:37:02 -0400458 return security_inode_permission(inode, MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459}
460
461/*
462 * This is called when everything else fails, and we actually have
463 * to go to the low-level filesystem to find out what we should do..
464 *
465 * We get the directory semaphore, and after getting that we also
466 * make sure that nobody added the entry to the dcache in the meantime..
467 * SMP-safe
468 */
469static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
470{
471 struct dentry * result;
472 struct inode *dir = parent->d_inode;
473
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800474 mutex_lock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 /*
476 * First re-do the cached lookup just in case it was created
477 * while we waited for the directory semaphore..
478 *
479 * FIXME! This could use version numbering or similar to
480 * avoid unnecessary cache lookups.
481 *
482 * The "dcache_lock" is purely to protect the RCU list walker
483 * from concurrent renames at this point (we mustn't get false
484 * negatives from the RCU list walk here, unlike the optimistic
485 * fast walk).
486 *
487 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
488 */
489 result = d_lookup(parent, name);
490 if (!result) {
Miklos Szeredid70b67c2008-07-02 21:30:15 +0200491 struct dentry *dentry;
492
493 /* Don't create child dentry for a dead directory. */
494 result = ERR_PTR(-ENOENT);
495 if (IS_DEADDIR(dir))
496 goto out_unlock;
497
498 dentry = d_alloc(parent, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 result = ERR_PTR(-ENOMEM);
500 if (dentry) {
501 result = dir->i_op->lookup(dir, dentry, nd);
502 if (result)
503 dput(dentry);
504 else
505 result = dentry;
506 }
Miklos Szeredid70b67c2008-07-02 21:30:15 +0200507out_unlock:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800508 mutex_unlock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 return result;
510 }
511
512 /*
513 * Uhhuh! Nasty case: the cache was re-populated while
514 * we waited on the semaphore. Need to revalidate.
515 */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800516 mutex_unlock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 if (result->d_op && result->d_op->d_revalidate) {
Ian Kentbcdc5e02006-09-27 01:50:44 -0700518 result = do_revalidate(result, nd);
519 if (!result)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 result = ERR_PTR(-ENOENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 }
522 return result;
523}
524
Al Viroa02f76c2008-02-23 15:14:28 +0000525/*
526 * Wrapper to retry pathname resolution whenever the underlying
527 * file system returns an ESTALE.
528 *
529 * Retry the whole path once, forcing real lookup requests
530 * instead of relying on the dcache.
531 */
532static __always_inline int link_path_walk(const char *name, struct nameidata *nd)
533{
534 struct path save = nd->path;
535 int result;
536
537 /* make sure the stuff we saved doesn't go away */
Jan Blunckc8e7f442008-06-09 16:40:35 -0700538 path_get(&save);
Al Viroa02f76c2008-02-23 15:14:28 +0000539
540 result = __link_path_walk(name, nd);
541 if (result == -ESTALE) {
542 /* nd->path had been dropped */
543 nd->path = save;
Jan Blunckc8e7f442008-06-09 16:40:35 -0700544 path_get(&nd->path);
Al Viroa02f76c2008-02-23 15:14:28 +0000545 nd->flags |= LOOKUP_REVAL;
546 result = __link_path_walk(name, nd);
547 }
548
549 path_put(&save);
550
551 return result;
552}
553
Arjan van de Venf1662352006-01-14 13:21:31 -0800554static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555{
556 int res = 0;
557 char *name;
558 if (IS_ERR(link))
559 goto fail;
560
561 if (*link == '/') {
Christoph Hellwigb4091d52008-11-05 15:07:21 +0100562 struct fs_struct *fs = current->fs;
563
Jan Blunck1d957f92008-02-14 19:34:35 -0800564 path_put(&nd->path);
Christoph Hellwigb4091d52008-11-05 15:07:21 +0100565
566 read_lock(&fs->lock);
567 nd->path = fs->root;
568 path_get(&fs->root);
569 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 }
Christoph Hellwigb4091d52008-11-05 15:07:21 +0100571
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 res = link_path_walk(link, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 if (nd->depth || res || nd->last_type!=LAST_NORM)
574 return res;
575 /*
576 * If it is an iterative symlinks resolution in open_namei() we
577 * have to copy the last component. And all that crap because of
578 * bloody create() on broken symlinks. Furrfu...
579 */
580 name = __getname();
581 if (unlikely(!name)) {
Jan Blunck1d957f92008-02-14 19:34:35 -0800582 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 return -ENOMEM;
584 }
585 strcpy(name, nd->last.name);
586 nd->last.name = name;
587 return 0;
588fail:
Jan Blunck1d957f92008-02-14 19:34:35 -0800589 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 return PTR_ERR(link);
591}
592
Jan Blunck1d957f92008-02-14 19:34:35 -0800593static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700594{
595 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800596 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700597 mntput(path->mnt);
598}
599
600static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
601{
Jan Blunck4ac91372008-02-14 19:34:32 -0800602 dput(nd->path.dentry);
603 if (nd->path.mnt != path->mnt)
604 mntput(nd->path.mnt);
605 nd->path.mnt = path->mnt;
606 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700607}
608
Ian Kent051d3812006-03-27 01:14:53 -0800609static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
610{
611 int error;
612 void *cookie;
613 struct dentry *dentry = path->dentry;
614
615 touch_atime(path->mnt, dentry);
616 nd_set_link(nd, NULL);
617
Jan Blunck4ac91372008-02-14 19:34:32 -0800618 if (path->mnt != nd->path.mnt) {
Ian Kent051d3812006-03-27 01:14:53 -0800619 path_to_nameidata(path, nd);
620 dget(dentry);
621 }
622 mntget(path->mnt);
623 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
624 error = PTR_ERR(cookie);
625 if (!IS_ERR(cookie)) {
626 char *s = nd_get_link(nd);
627 error = 0;
628 if (s)
629 error = __vfs_follow_link(nd, s);
630 if (dentry->d_inode->i_op->put_link)
631 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
632 }
Jan Blunck09da59162008-02-14 19:34:37 -0800633 path_put(path);
Ian Kent051d3812006-03-27 01:14:53 -0800634
635 return error;
636}
637
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638/*
639 * This limits recursive symlink follows to 8, while
640 * limiting consecutive symlinks to 40.
641 *
642 * Without that kind of total limit, nasty chains of consecutive
643 * symlinks can cause almost arbitrarily long lookups.
644 */
Al Viro90ebe562005-06-06 13:35:58 -0700645static inline int do_follow_link(struct path *path, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646{
647 int err = -ELOOP;
648 if (current->link_count >= MAX_NESTED_LINKS)
649 goto loop;
650 if (current->total_link_count >= 40)
651 goto loop;
652 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
653 cond_resched();
Al Viro90ebe562005-06-06 13:35:58 -0700654 err = security_inode_follow_link(path->dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 if (err)
656 goto loop;
657 current->link_count++;
658 current->total_link_count++;
659 nd->depth++;
Al Virocd4e91d2005-06-06 13:36:03 -0700660 err = __do_follow_link(path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 current->link_count--;
662 nd->depth--;
663 return err;
664loop:
Jan Blunck1d957f92008-02-14 19:34:35 -0800665 path_put_conditional(path, nd);
666 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 return err;
668}
669
670int follow_up(struct vfsmount **mnt, struct dentry **dentry)
671{
672 struct vfsmount *parent;
673 struct dentry *mountpoint;
674 spin_lock(&vfsmount_lock);
675 parent=(*mnt)->mnt_parent;
676 if (parent == *mnt) {
677 spin_unlock(&vfsmount_lock);
678 return 0;
679 }
680 mntget(parent);
681 mountpoint=dget((*mnt)->mnt_mountpoint);
682 spin_unlock(&vfsmount_lock);
683 dput(*dentry);
684 *dentry = mountpoint;
685 mntput(*mnt);
686 *mnt = parent;
687 return 1;
688}
689
690/* no need for dcache_lock, as serialization is taken care in
691 * namespace.c
692 */
Al Viro463ffb22005-06-06 13:36:05 -0700693static int __follow_mount(struct path *path)
694{
695 int res = 0;
696 while (d_mountpoint(path->dentry)) {
697 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
698 if (!mounted)
699 break;
700 dput(path->dentry);
701 if (res)
702 mntput(path->mnt);
703 path->mnt = mounted;
704 path->dentry = dget(mounted->mnt_root);
705 res = 1;
706 }
707 return res;
708}
709
Al Viro58c465e2005-06-06 13:36:13 -0700710static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 while (d_mountpoint(*dentry)) {
713 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
714 if (!mounted)
715 break;
Al Viro58c465e2005-06-06 13:36:13 -0700716 dput(*dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 mntput(*mnt);
718 *mnt = mounted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 *dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721}
722
723/* no need for dcache_lock, as serialization is taken care in
724 * namespace.c
725 */
Al Viroe13b2102005-06-06 13:36:06 -0700726int follow_down(struct vfsmount **mnt, struct dentry **dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727{
728 struct vfsmount *mounted;
729
730 mounted = lookup_mnt(*mnt, *dentry);
731 if (mounted) {
Al Viroe13b2102005-06-06 13:36:06 -0700732 dput(*dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 mntput(*mnt);
734 *mnt = mounted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 *dentry = dget(mounted->mnt_root);
736 return 1;
737 }
738 return 0;
739}
740
Arjan van de Venf1662352006-01-14 13:21:31 -0800741static __always_inline void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742{
Andreas Mohre518ddb2006-09-29 02:01:22 -0700743 struct fs_struct *fs = current->fs;
744
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 while(1) {
746 struct vfsmount *parent;
Jan Blunck4ac91372008-02-14 19:34:32 -0800747 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748
Andreas Mohre518ddb2006-09-29 02:01:22 -0700749 read_lock(&fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -0800750 if (nd->path.dentry == fs->root.dentry &&
751 nd->path.mnt == fs->root.mnt) {
Andreas Mohre518ddb2006-09-29 02:01:22 -0700752 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 break;
754 }
Andreas Mohre518ddb2006-09-29 02:01:22 -0700755 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 spin_lock(&dcache_lock);
Jan Blunck4ac91372008-02-14 19:34:32 -0800757 if (nd->path.dentry != nd->path.mnt->mnt_root) {
758 nd->path.dentry = dget(nd->path.dentry->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 spin_unlock(&dcache_lock);
760 dput(old);
761 break;
762 }
763 spin_unlock(&dcache_lock);
764 spin_lock(&vfsmount_lock);
Jan Blunck4ac91372008-02-14 19:34:32 -0800765 parent = nd->path.mnt->mnt_parent;
766 if (parent == nd->path.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 spin_unlock(&vfsmount_lock);
768 break;
769 }
770 mntget(parent);
Jan Blunck4ac91372008-02-14 19:34:32 -0800771 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 spin_unlock(&vfsmount_lock);
773 dput(old);
Jan Blunck4ac91372008-02-14 19:34:32 -0800774 mntput(nd->path.mnt);
775 nd->path.mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 }
Jan Blunck4ac91372008-02-14 19:34:32 -0800777 follow_mount(&nd->path.mnt, &nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778}
779
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780/*
781 * It's more convoluted than I'd like it to be, but... it's still fairly
782 * small and for now I'd prefer to have fast path as straight as possible.
783 * It _is_ time-critical.
784 */
785static int do_lookup(struct nameidata *nd, struct qstr *name,
786 struct path *path)
787{
Jan Blunck4ac91372008-02-14 19:34:32 -0800788 struct vfsmount *mnt = nd->path.mnt;
789 struct dentry *dentry = __d_lookup(nd->path.dentry, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790
791 if (!dentry)
792 goto need_lookup;
793 if (dentry->d_op && dentry->d_op->d_revalidate)
794 goto need_revalidate;
795done:
796 path->mnt = mnt;
797 path->dentry = dentry;
Al Viro634ee702005-06-06 13:36:13 -0700798 __follow_mount(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 return 0;
800
801need_lookup:
Jan Blunck4ac91372008-02-14 19:34:32 -0800802 dentry = real_lookup(nd->path.dentry, name, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 if (IS_ERR(dentry))
804 goto fail;
805 goto done;
806
807need_revalidate:
Ian Kentbcdc5e02006-09-27 01:50:44 -0700808 dentry = do_revalidate(dentry, nd);
809 if (!dentry)
810 goto need_lookup;
811 if (IS_ERR(dentry))
812 goto fail;
813 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814
815fail:
816 return PTR_ERR(dentry);
817}
818
819/*
820 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +0100821 * This is the basic name resolution function, turning a pathname into
822 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 *
Prasanna Medaea3834d2005-04-29 16:00:17 +0100824 * Returns 0 and nd will have valid dentry and mnt on success.
825 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800827static int __link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828{
829 struct path next;
830 struct inode *inode;
831 int err;
832 unsigned int lookup_flags = nd->flags;
833
834 while (*name=='/')
835 name++;
836 if (!*name)
837 goto return_reval;
838
Jan Blunck4ac91372008-02-14 19:34:32 -0800839 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 if (nd->depth)
Trond Myklebustf55eab82006-02-04 23:28:01 -0800841 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842
843 /* At this point we know we have a real path component. */
844 for(;;) {
845 unsigned long hash;
846 struct qstr this;
847 unsigned int c;
848
Trond Myklebustcdce5d62005-10-18 14:20:18 -0700849 nd->flags |= LOOKUP_CONTINUE;
Al Viro672b16b2008-07-17 09:45:01 -0400850 err = exec_permission_lite(inode);
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800851 if (err == -EAGAIN)
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200852 err = inode_permission(nd->path.dentry->d_inode,
853 MAY_EXEC);
Mimi Zohar6146f0d2009-02-04 09:06:57 -0500854 if (!err)
855 err = ima_path_check(&nd->path, MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 if (err)
857 break;
858
859 this.name = name;
860 c = *(const unsigned char *)name;
861
862 hash = init_name_hash();
863 do {
864 name++;
865 hash = partial_name_hash(c, hash);
866 c = *(const unsigned char *)name;
867 } while (c && (c != '/'));
868 this.len = name - (const char *) this.name;
869 this.hash = end_name_hash(hash);
870
871 /* remove trailing slashes? */
872 if (!c)
873 goto last_component;
874 while (*++name == '/');
875 if (!*name)
876 goto last_with_slashes;
877
878 /*
879 * "." and ".." are special - ".." especially so because it has
880 * to be able to know about the current root directory and
881 * parent relationships.
882 */
883 if (this.name[0] == '.') switch (this.len) {
884 default:
885 break;
886 case 2:
887 if (this.name[1] != '.')
888 break;
Al Viro58c465e2005-06-06 13:36:13 -0700889 follow_dotdot(nd);
Jan Blunck4ac91372008-02-14 19:34:32 -0800890 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 /* fallthrough */
892 case 1:
893 continue;
894 }
895 /*
896 * See if the low-level filesystem might want
897 * to use its own hash..
898 */
Jan Blunck4ac91372008-02-14 19:34:32 -0800899 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
900 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
901 &this);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 if (err < 0)
903 break;
904 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 /* This does the actual lookups.. */
906 err = do_lookup(nd, &this, &next);
907 if (err)
908 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909
910 err = -ENOENT;
911 inode = next.dentry->d_inode;
912 if (!inode)
913 goto out_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914
915 if (inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700916 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 if (err)
918 goto return_err;
919 err = -ENOENT;
Jan Blunck4ac91372008-02-14 19:34:32 -0800920 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 if (!inode)
922 break;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700923 } else
924 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 err = -ENOTDIR;
926 if (!inode->i_op->lookup)
927 break;
928 continue;
929 /* here ends the main loop */
930
931last_with_slashes:
932 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
933last_component:
Trond Myklebustf55eab82006-02-04 23:28:01 -0800934 /* Clear LOOKUP_CONTINUE iff it was previously unset */
935 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 if (lookup_flags & LOOKUP_PARENT)
937 goto lookup_parent;
938 if (this.name[0] == '.') switch (this.len) {
939 default:
940 break;
941 case 2:
942 if (this.name[1] != '.')
943 break;
Al Viro58c465e2005-06-06 13:36:13 -0700944 follow_dotdot(nd);
Jan Blunck4ac91372008-02-14 19:34:32 -0800945 inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 /* fallthrough */
947 case 1:
948 goto return_reval;
949 }
Jan Blunck4ac91372008-02-14 19:34:32 -0800950 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
951 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
952 &this);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 if (err < 0)
954 break;
955 }
956 err = do_lookup(nd, &this, &next);
957 if (err)
958 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 inode = next.dentry->d_inode;
960 if ((lookup_flags & LOOKUP_FOLLOW)
Al Viroacfa4382008-12-04 10:06:33 -0500961 && inode && inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700962 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 if (err)
964 goto return_err;
Jan Blunck4ac91372008-02-14 19:34:32 -0800965 inode = nd->path.dentry->d_inode;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700966 } else
967 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 err = -ENOENT;
969 if (!inode)
970 break;
971 if (lookup_flags & LOOKUP_DIRECTORY) {
972 err = -ENOTDIR;
Al Viroacfa4382008-12-04 10:06:33 -0500973 if (!inode->i_op->lookup)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 break;
975 }
976 goto return_base;
977lookup_parent:
978 nd->last = this;
979 nd->last_type = LAST_NORM;
980 if (this.name[0] != '.')
981 goto return_base;
982 if (this.len == 1)
983 nd->last_type = LAST_DOT;
984 else if (this.len == 2 && this.name[1] == '.')
985 nd->last_type = LAST_DOTDOT;
986 else
987 goto return_base;
988return_reval:
989 /*
990 * We bypassed the ordinary revalidation routines.
991 * We may need to check the cached dentry for staleness.
992 */
Jan Blunck4ac91372008-02-14 19:34:32 -0800993 if (nd->path.dentry && nd->path.dentry->d_sb &&
994 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 err = -ESTALE;
996 /* Note: we do not d_invalidate() */
Jan Blunck4ac91372008-02-14 19:34:32 -0800997 if (!nd->path.dentry->d_op->d_revalidate(
998 nd->path.dentry, nd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 break;
1000 }
1001return_base:
1002 return 0;
1003out_dput:
Jan Blunck1d957f92008-02-14 19:34:35 -08001004 path_put_conditional(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 break;
1006 }
Jan Blunck1d957f92008-02-14 19:34:35 -08001007 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008return_err:
1009 return err;
1010}
1011
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -08001012static int path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013{
1014 current->total_link_count = 0;
1015 return link_path_walk(name, nd);
1016}
1017
Prasanna Medaea3834d2005-04-29 16:00:17 +01001018/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -08001019static int do_path_lookup(int dfd, const char *name,
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001020 unsigned int flags, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001022 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001023 int fput_needed;
1024 struct file *file;
Andreas Mohre518ddb2006-09-29 02:01:22 -07001025 struct fs_struct *fs = current->fs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026
1027 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1028 nd->flags = flags;
1029 nd->depth = 0;
1030
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 if (*name=='/') {
Andreas Mohre518ddb2006-09-29 02:01:22 -07001032 read_lock(&fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08001033 nd->path = fs->root;
1034 path_get(&fs->root);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001035 read_unlock(&fs->lock);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001036 } else if (dfd == AT_FDCWD) {
Andreas Mohre518ddb2006-09-29 02:01:22 -07001037 read_lock(&fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08001038 nd->path = fs->pwd;
1039 path_get(&fs->pwd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001040 read_unlock(&fs->lock);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001041 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001042 struct dentry *dentry;
1043
1044 file = fget_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001045 retval = -EBADF;
1046 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001047 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001048
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001049 dentry = file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001050
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001051 retval = -ENOTDIR;
1052 if (!S_ISDIR(dentry->d_inode->i_mode))
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001053 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001054
1055 retval = file_permission(file, MAY_EXEC);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001056 if (retval)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001057 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001058
Jan Blunck5dd784d02008-02-14 19:34:38 -08001059 nd->path = file->f_path;
1060 path_get(&file->f_path);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001061
1062 fput_light(file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 }
Josef 'Jeff' Sipek2dfdd262007-05-09 02:33:41 -07001064
1065 retval = path_walk(name, nd);
Jan Blunck4ac91372008-02-14 19:34:32 -08001066 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1067 nd->path.dentry->d_inode))
1068 audit_inode(name, nd->path.dentry);
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001069out_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001070 return retval;
1071
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001072fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001073 fput_light(file, fput_needed);
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001074 goto out_fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075}
1076
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -08001077int path_lookup(const char *name, unsigned int flags,
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001078 struct nameidata *nd)
1079{
1080 return do_path_lookup(AT_FDCWD, name, flags, nd);
1081}
1082
Al Virod1811462008-08-02 00:49:18 -04001083int kern_path(const char *name, unsigned int flags, struct path *path)
1084{
1085 struct nameidata nd;
1086 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1087 if (!res)
1088 *path = nd.path;
1089 return res;
1090}
1091
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001092/**
1093 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1094 * @dentry: pointer to dentry of the base directory
1095 * @mnt: pointer to vfs mount of the base directory
1096 * @name: pointer to file name
1097 * @flags: lookup flags
1098 * @nd: pointer to nameidata
1099 */
1100int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1101 const char *name, unsigned int flags,
1102 struct nameidata *nd)
1103{
1104 int retval;
1105
1106 /* same as do_path_lookup */
1107 nd->last_type = LAST_ROOT;
1108 nd->flags = flags;
1109 nd->depth = 0;
1110
Jan Blunckc8e7f442008-06-09 16:40:35 -07001111 nd->path.dentry = dentry;
1112 nd->path.mnt = mnt;
1113 path_get(&nd->path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001114
1115 retval = path_walk(name, nd);
Jan Blunck4ac91372008-02-14 19:34:32 -08001116 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1117 nd->path.dentry->d_inode))
1118 audit_inode(name, nd->path.dentry);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001119
1120 return retval;
1121
1122}
1123
Trond Myklebust834f2a42005-10-18 14:20:16 -07001124/**
1125 * path_lookup_open - lookup a file path with open intent
Martin Waitz7045f372006-02-01 03:06:57 -08001126 * @dfd: the directory to use as base, or AT_FDCWD
Trond Myklebust834f2a42005-10-18 14:20:16 -07001127 * @name: pointer to file name
1128 * @lookup_flags: lookup intent flags
1129 * @nd: pointer to nameidata
1130 * @open_flags: open intent flags
1131 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001132int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
Trond Myklebust834f2a42005-10-18 14:20:16 -07001133 struct nameidata *nd, int open_flags)
1134{
Al Viro8737f3a2008-08-02 22:36:57 -04001135 struct file *filp = get_empty_filp();
1136 int err;
Trond Myklebust834f2a42005-10-18 14:20:16 -07001137
Al Viro8737f3a2008-08-02 22:36:57 -04001138 if (filp == NULL)
1139 return -ENFILE;
1140 nd->intent.open.file = filp;
1141 nd->intent.open.flags = open_flags;
1142 nd->intent.open.create_mode = 0;
1143 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1144 if (IS_ERR(nd->intent.open.file)) {
1145 if (err == 0) {
1146 err = PTR_ERR(nd->intent.open.file);
1147 path_put(&nd->path);
1148 }
1149 } else if (err != 0)
1150 release_open_intent(nd);
1151 return err;
Trond Myklebust834f2a42005-10-18 14:20:16 -07001152}
1153
Christoph Hellwigeead1912007-10-16 23:25:38 -07001154static struct dentry *__lookup_hash(struct qstr *name,
1155 struct dentry *base, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156{
James Morris057f6c02007-04-26 00:12:05 -07001157 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 struct inode *inode;
1159 int err;
1160
1161 inode = base->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162
1163 /*
1164 * See if the low-level filesystem might want
1165 * to use its own hash..
1166 */
1167 if (base->d_op && base->d_op->d_hash) {
1168 err = base->d_op->d_hash(base, name);
1169 dentry = ERR_PTR(err);
1170 if (err < 0)
1171 goto out;
1172 }
1173
1174 dentry = cached_lookup(base, name, nd);
1175 if (!dentry) {
Miklos Szeredid70b67c2008-07-02 21:30:15 +02001176 struct dentry *new;
1177
1178 /* Don't create child dentry for a dead directory. */
1179 dentry = ERR_PTR(-ENOENT);
1180 if (IS_DEADDIR(inode))
1181 goto out;
1182
1183 new = d_alloc(base, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 dentry = ERR_PTR(-ENOMEM);
1185 if (!new)
1186 goto out;
1187 dentry = inode->i_op->lookup(inode, new, nd);
1188 if (!dentry)
1189 dentry = new;
1190 else
1191 dput(new);
1192 }
1193out:
1194 return dentry;
1195}
1196
James Morris057f6c02007-04-26 00:12:05 -07001197/*
1198 * Restricted form of lookup. Doesn't follow links, single-component only,
1199 * needs parent already locked. Doesn't follow mounts.
1200 * SMP-safe.
1201 */
Adrian Bunka244e162006-03-31 02:32:11 -08001202static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203{
Christoph Hellwigeead1912007-10-16 23:25:38 -07001204 int err;
1205
Al Virof419a2e2008-07-22 00:07:17 -04001206 err = inode_permission(nd->path.dentry->d_inode, MAY_EXEC);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001207 if (err)
1208 return ERR_PTR(err);
Jan Blunck4ac91372008-02-14 19:34:32 -08001209 return __lookup_hash(&nd->last, nd->path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210}
1211
Christoph Hellwigeead1912007-10-16 23:25:38 -07001212static int __lookup_one_len(const char *name, struct qstr *this,
1213 struct dentry *base, int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214{
1215 unsigned long hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 unsigned int c;
1217
James Morris057f6c02007-04-26 00:12:05 -07001218 this->name = name;
1219 this->len = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 if (!len)
James Morris057f6c02007-04-26 00:12:05 -07001221 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222
1223 hash = init_name_hash();
1224 while (len--) {
1225 c = *(const unsigned char *)name++;
1226 if (c == '/' || c == '\0')
James Morris057f6c02007-04-26 00:12:05 -07001227 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 hash = partial_name_hash(c, hash);
1229 }
James Morris057f6c02007-04-26 00:12:05 -07001230 this->hash = end_name_hash(hash);
1231 return 0;
1232}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233
Christoph Hellwigeead1912007-10-16 23:25:38 -07001234/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001235 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07001236 * @name: pathname component to lookup
1237 * @base: base directory to lookup from
1238 * @len: maximum length @len should be interpreted to
1239 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07001240 * Note that this routine is purely a helper for filesystem usage and should
1241 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07001242 * nameidata argument is passed to the filesystem methods and a filesystem
1243 * using this helper needs to be prepared for that.
1244 */
James Morris057f6c02007-04-26 00:12:05 -07001245struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1246{
1247 int err;
1248 struct qstr this;
1249
1250 err = __lookup_one_len(name, &this, base, len);
1251 if (err)
1252 return ERR_PTR(err);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001253
Al Virof419a2e2008-07-22 00:07:17 -04001254 err = inode_permission(base->d_inode, MAY_EXEC);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001255 if (err)
1256 return ERR_PTR(err);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001257 return __lookup_hash(&this, base, NULL);
James Morris057f6c02007-04-26 00:12:05 -07001258}
1259
Christoph Hellwigeead1912007-10-16 23:25:38 -07001260/**
1261 * lookup_one_noperm - bad hack for sysfs
1262 * @name: pathname component to lookup
1263 * @base: base directory to lookup from
1264 *
1265 * This is a variant of lookup_one_len that doesn't perform any permission
1266 * checks. It's a horrible hack to work around the braindead sysfs
1267 * architecture and should not be used anywhere else.
1268 *
1269 * DON'T USE THIS FUNCTION EVER, thanks.
1270 */
1271struct dentry *lookup_one_noperm(const char *name, struct dentry *base)
James Morris057f6c02007-04-26 00:12:05 -07001272{
1273 int err;
1274 struct qstr this;
1275
Christoph Hellwigeead1912007-10-16 23:25:38 -07001276 err = __lookup_one_len(name, &this, base, strlen(name));
James Morris057f6c02007-04-26 00:12:05 -07001277 if (err)
1278 return ERR_PTR(err);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001279 return __lookup_hash(&this, base, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280}
1281
Al Viro2d8f3032008-07-22 09:59:21 -04001282int user_path_at(int dfd, const char __user *name, unsigned flags,
1283 struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284{
Al Viro2d8f3032008-07-22 09:59:21 -04001285 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 char *tmp = getname(name);
1287 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04001289
1290 BUG_ON(flags & LOOKUP_PARENT);
1291
1292 err = do_path_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04001294 if (!err)
1295 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 }
1297 return err;
1298}
1299
Al Viro2ad94ae2008-07-21 09:32:51 -04001300static int user_path_parent(int dfd, const char __user *path,
1301 struct nameidata *nd, char **name)
1302{
1303 char *s = getname(path);
1304 int error;
1305
1306 if (IS_ERR(s))
1307 return PTR_ERR(s);
1308
1309 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1310 if (error)
1311 putname(s);
1312 else
1313 *name = s;
1314
1315 return error;
1316}
1317
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318/*
1319 * It's inline, so penalty for filesystems that don't use sticky bit is
1320 * minimal.
1321 */
1322static inline int check_sticky(struct inode *dir, struct inode *inode)
1323{
David Howellsda9592e2008-11-14 10:39:05 +11001324 uid_t fsuid = current_fsuid();
1325
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 if (!(dir->i_mode & S_ISVTX))
1327 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001328 if (inode->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001330 if (dir->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 return 0;
1332 return !capable(CAP_FOWNER);
1333}
1334
1335/*
1336 * Check whether we can remove a link victim from directory dir, check
1337 * whether the type of victim is right.
1338 * 1. We can't do it if dir is read-only (done in permission())
1339 * 2. We should have write and exec permissions on dir
1340 * 3. We can't remove anything from append-only dir
1341 * 4. We can't do anything with immutable dir (done in permission())
1342 * 5. If the sticky bit on dir is set we should either
1343 * a. be owner of dir, or
1344 * b. be owner of victim, or
1345 * c. have CAP_FOWNER capability
1346 * 6. If the victim is append-only or immutable we can't do antyhing with
1347 * links pointing to it.
1348 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1349 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1350 * 9. We can't remove a root or mountpoint.
1351 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1352 * nfs_async_unlink().
1353 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001354static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355{
1356 int error;
1357
1358 if (!victim->d_inode)
1359 return -ENOENT;
1360
1361 BUG_ON(victim->d_parent->d_inode != dir);
Al Viro5a190ae2007-06-07 12:19:32 -04001362 audit_inode_child(victim->d_name.name, victim, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363
Al Virof419a2e2008-07-22 00:07:17 -04001364 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 if (error)
1366 return error;
1367 if (IS_APPEND(dir))
1368 return -EPERM;
1369 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
Hugh Dickinsf9454542008-11-19 15:36:38 -08001370 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 return -EPERM;
1372 if (isdir) {
1373 if (!S_ISDIR(victim->d_inode->i_mode))
1374 return -ENOTDIR;
1375 if (IS_ROOT(victim))
1376 return -EBUSY;
1377 } else if (S_ISDIR(victim->d_inode->i_mode))
1378 return -EISDIR;
1379 if (IS_DEADDIR(dir))
1380 return -ENOENT;
1381 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1382 return -EBUSY;
1383 return 0;
1384}
1385
1386/* Check whether we can create an object with dentry child in directory
1387 * dir.
1388 * 1. We can't do it if child already exists (open has special treatment for
1389 * this case, but since we are inlined it's OK)
1390 * 2. We can't do it if dir is read-only (done in permission())
1391 * 3. We should have write and exec permissions on dir
1392 * 4. We can't do it if dir is immutable (done in permission())
1393 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02001394static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395{
1396 if (child->d_inode)
1397 return -EEXIST;
1398 if (IS_DEADDIR(dir))
1399 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04001400 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401}
1402
1403/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 * O_DIRECTORY translates into forcing a directory lookup.
1405 */
1406static inline int lookup_flags(unsigned int f)
1407{
1408 unsigned long retval = LOOKUP_FOLLOW;
1409
1410 if (f & O_NOFOLLOW)
1411 retval &= ~LOOKUP_FOLLOW;
1412
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 if (f & O_DIRECTORY)
1414 retval |= LOOKUP_DIRECTORY;
1415
1416 return retval;
1417}
1418
1419/*
1420 * p1 and p2 should be directories on the same fs.
1421 */
1422struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1423{
1424 struct dentry *p;
1425
1426 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001427 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 return NULL;
1429 }
1430
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001431 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001433 p = d_ancestor(p2, p1);
1434 if (p) {
1435 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1436 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1437 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 }
1439
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001440 p = d_ancestor(p1, p2);
1441 if (p) {
1442 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1443 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1444 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 }
1446
Ingo Molnarf2eace22006-07-03 00:25:05 -07001447 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1448 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 return NULL;
1450}
1451
1452void unlock_rename(struct dentry *p1, struct dentry *p2)
1453{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001454 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001456 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001457 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 }
1459}
1460
1461int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1462 struct nameidata *nd)
1463{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001464 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465
1466 if (error)
1467 return error;
1468
Al Viroacfa4382008-12-04 10:06:33 -05001469 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 return -EACCES; /* shouldn't it be ENOSYS? */
1471 mode &= S_IALLUGO;
1472 mode |= S_IFREG;
1473 error = security_inode_create(dir, dentry, mode);
1474 if (error)
1475 return error;
Jan Kara9e3509e2009-01-26 16:45:12 +01001476 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001478 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001479 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 return error;
1481}
1482
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001483int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001485 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 struct inode *inode = dentry->d_inode;
1487 int error;
1488
1489 if (!inode)
1490 return -ENOENT;
1491
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001492 switch (inode->i_mode & S_IFMT) {
1493 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001495 case S_IFDIR:
1496 if (acc_mode & MAY_WRITE)
1497 return -EISDIR;
1498 break;
1499 case S_IFBLK:
1500 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001501 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001503 /*FALLTHRU*/
1504 case S_IFIFO:
1505 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001507 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001508 }
Dave Hansenb41572e2007-10-16 23:31:14 -07001509
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001510 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07001511 if (error)
1512 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05001513
James Morriscb5629b2009-02-06 11:01:45 +11001514 error = ima_path_check(path,
Mimi Zohar6146f0d2009-02-04 09:06:57 -05001515 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC));
1516 if (error)
1517 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 /*
1519 * An append-only file must be opened in append mode for writing.
1520 */
1521 if (IS_APPEND(inode)) {
1522 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1523 return -EPERM;
1524 if (flag & O_TRUNC)
1525 return -EPERM;
1526 }
1527
1528 /* O_NOATIME can only be set by the owner or superuser */
1529 if (flag & O_NOATIME)
Satyam Sharma3bd858a2007-07-17 15:00:08 +05301530 if (!is_owner_or_cap(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 return -EPERM;
1532
1533 /*
1534 * Ensure there are no outstanding leases on the file.
1535 */
1536 error = break_lease(inode, flag);
1537 if (error)
1538 return error;
1539
1540 if (flag & O_TRUNC) {
1541 error = get_write_access(inode);
1542 if (error)
1543 return error;
1544
1545 /*
1546 * Refuse to truncate files with mandatory locks held on them.
1547 */
1548 error = locks_verify_locked(inode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001549 if (!error)
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001550 error = security_path_truncate(path, 0,
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001551 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 if (!error) {
Jan Kara9e3509e2009-01-26 16:45:12 +01001553 vfs_dq_init(inode);
Miklos Szeredid139d7f2007-10-18 03:07:00 -07001554
1555 error = do_truncate(dentry, 0,
1556 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1557 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 }
1559 put_write_access(inode);
1560 if (error)
1561 return error;
1562 } else
1563 if (flag & FMODE_WRITE)
Jan Kara9e3509e2009-01-26 16:45:12 +01001564 vfs_dq_init(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565
1566 return 0;
1567}
1568
Dave Hansend57999e2008-02-15 14:37:27 -08001569/*
1570 * Be careful about ever adding any more callers of this
1571 * function. Its flags must be in the namei format, not
1572 * what get passed to sys_open().
1573 */
1574static int __open_namei_create(struct nameidata *nd, struct path *path,
Dave Hansenaab520e2006-09-30 23:29:02 -07001575 int flag, int mode)
1576{
1577 int error;
Jan Blunck4ac91372008-02-14 19:34:32 -08001578 struct dentry *dir = nd->path.dentry;
Dave Hansenaab520e2006-09-30 23:29:02 -07001579
1580 if (!IS_POSIXACL(dir->d_inode))
1581 mode &= ~current->fs->umask;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001582 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1583 if (error)
1584 goto out_unlock;
Dave Hansenaab520e2006-09-30 23:29:02 -07001585 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001586out_unlock:
Dave Hansenaab520e2006-09-30 23:29:02 -07001587 mutex_unlock(&dir->d_inode->i_mutex);
Jan Blunck4ac91372008-02-14 19:34:32 -08001588 dput(nd->path.dentry);
1589 nd->path.dentry = path->dentry;
Dave Hansenaab520e2006-09-30 23:29:02 -07001590 if (error)
1591 return error;
1592 /* Don't check for write permission, don't truncate */
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001593 return may_open(&nd->path, 0, flag & ~O_TRUNC);
Dave Hansenaab520e2006-09-30 23:29:02 -07001594}
1595
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596/*
Dave Hansend57999e2008-02-15 14:37:27 -08001597 * Note that while the flag value (low two bits) for sys_open means:
1598 * 00 - read-only
1599 * 01 - write-only
1600 * 10 - read-write
1601 * 11 - special
1602 * it is changed into
1603 * 00 - no permissions needed
1604 * 01 - read-permission
1605 * 10 - write-permission
1606 * 11 - read-write
1607 * for the internal routines (ie open_namei()/follow_link() etc)
1608 * This is more logical, and also allows the 00 "no perm needed"
1609 * to be used for symlinks (where the permissions are checked
1610 * later).
1611 *
1612*/
1613static inline int open_to_namei_flags(int flag)
1614{
1615 if ((flag+1) & O_ACCMODE)
1616 flag++;
1617 return flag;
1618}
1619
Dave Hansen4a3fd212008-02-15 14:37:48 -08001620static int open_will_write_to_fs(int flag, struct inode *inode)
1621{
1622 /*
1623 * We'll never write to the fs underlying
1624 * a device file.
1625 */
1626 if (special_file(inode->i_mode))
1627 return 0;
1628 return (flag & O_TRUNC);
1629}
1630
Dave Hansend57999e2008-02-15 14:37:27 -08001631/*
Dave Hansen4a3fd212008-02-15 14:37:48 -08001632 * Note that the low bits of the passed in "open_flag"
1633 * are not the same as in the local variable "flag". See
1634 * open_to_namei_flags() for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001636struct file *do_filp_open(int dfd, const char *pathname,
1637 int open_flag, int mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638{
Dave Hansen4a3fd212008-02-15 14:37:48 -08001639 struct file *filp;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001640 struct nameidata nd;
Trond Myklebust834f2a42005-10-18 14:20:16 -07001641 int acc_mode, error;
Al Viro4e7506e2005-06-06 13:36:00 -07001642 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 struct dentry *dir;
1644 int count = 0;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001645 int will_write;
Dave Hansend57999e2008-02-15 14:37:27 -08001646 int flag = open_to_namei_flags(open_flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647
Al Virob77b0642008-07-17 09:37:02 -04001648 acc_mode = MAY_OPEN | ACC_MODE(flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649
Trond Myklebust834f2a42005-10-18 14:20:16 -07001650 /* O_TRUNC implies we need access checks for write permissions */
1651 if (flag & O_TRUNC)
1652 acc_mode |= MAY_WRITE;
1653
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 /* Allow the LSM permission hook to distinguish append
1655 access from general write access. */
1656 if (flag & O_APPEND)
1657 acc_mode |= MAY_APPEND;
1658
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 /*
1660 * The simplest case - just a plain lookup.
1661 */
1662 if (!(flag & O_CREAT)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001663 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001664 &nd, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 if (error)
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001666 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 goto ok;
1668 }
1669
1670 /*
1671 * Create - we need to know the parent.
1672 */
Al Viro8737f3a2008-08-02 22:36:57 -04001673 error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 if (error)
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001675 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676
1677 /*
1678 * We have the parent and last component. First of all, check
1679 * that we are not asked to creat(2) an obvious directory - that
1680 * will not do.
1681 */
1682 error = -EISDIR;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001683 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
Al Viro8737f3a2008-08-02 22:36:57 -04001684 goto exit_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685
Al Viro8737f3a2008-08-02 22:36:57 -04001686 error = -ENFILE;
1687 filp = get_empty_filp();
1688 if (filp == NULL)
1689 goto exit_parent;
1690 nd.intent.open.file = filp;
1691 nd.intent.open.flags = flag;
1692 nd.intent.open.create_mode = mode;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001693 dir = nd.path.dentry;
1694 nd.flags &= ~LOOKUP_PARENT;
Al Viro8737f3a2008-08-02 22:36:57 -04001695 nd.flags |= LOOKUP_CREATE | LOOKUP_OPEN;
Al Viro35165862008-08-05 03:00:49 -04001696 if (flag & O_EXCL)
1697 nd.flags |= LOOKUP_EXCL;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001698 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001699 path.dentry = lookup_hash(&nd);
1700 path.mnt = nd.path.mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701
1702do_last:
Al Viro4e7506e2005-06-06 13:36:00 -07001703 error = PTR_ERR(path.dentry);
1704 if (IS_ERR(path.dentry)) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001705 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 goto exit;
1707 }
1708
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001709 if (IS_ERR(nd.intent.open.file)) {
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001710 error = PTR_ERR(nd.intent.open.file);
Dave Hansen4a3fd212008-02-15 14:37:48 -08001711 goto exit_mutex_unlock;
Oleg Drokin4af4c522006-03-25 03:06:54 -08001712 }
1713
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 /* Negative dentry, just create the file */
Al Viro4e7506e2005-06-06 13:36:00 -07001715 if (!path.dentry->d_inode) {
Dave Hansen4a3fd212008-02-15 14:37:48 -08001716 /*
1717 * This write is needed to ensure that a
1718 * ro->rw transition does not occur between
1719 * the time when the file is created and when
1720 * a permanent write count is taken through
1721 * the 'struct file' in nameidata_to_filp().
1722 */
1723 error = mnt_want_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 if (error)
Dave Hansen4a3fd212008-02-15 14:37:48 -08001725 goto exit_mutex_unlock;
1726 error = __open_namei_create(&nd, &path, flag, mode);
1727 if (error) {
1728 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 goto exit;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001730 }
1731 filp = nameidata_to_filp(&nd, open_flag);
1732 mnt_drop_write(nd.path.mnt);
1733 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 }
1735
1736 /*
1737 * It already exists.
1738 */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001739 mutex_unlock(&dir->d_inode->i_mutex);
Al Viro5a190ae2007-06-07 12:19:32 -04001740 audit_inode(pathname, path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741
1742 error = -EEXIST;
1743 if (flag & O_EXCL)
1744 goto exit_dput;
1745
Al Viroe13b2102005-06-06 13:36:06 -07001746 if (__follow_mount(&path)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 error = -ELOOP;
Al Viroba7a4c12005-06-06 13:36:08 -07001748 if (flag & O_NOFOLLOW)
1749 goto exit_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 }
Amy Griffis3e2efce2006-07-13 13:16:02 -04001751
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 error = -ENOENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001753 if (!path.dentry->d_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 goto exit_dput;
Al Viroacfa4382008-12-04 10:06:33 -05001755 if (path.dentry->d_inode->i_op->follow_link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 goto do_link;
1757
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001758 path_to_nameidata(&path, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 error = -EISDIR;
Al Viro4e7506e2005-06-06 13:36:00 -07001760 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 goto exit;
1762ok:
Dave Hansen4a3fd212008-02-15 14:37:48 -08001763 /*
1764 * Consider:
1765 * 1. may_open() truncates a file
1766 * 2. a rw->ro mount transition occurs
1767 * 3. nameidata_to_filp() fails due to
1768 * the ro mount.
1769 * That would be inconsistent, and should
1770 * be avoided. Taking this mnt write here
1771 * ensures that (2) can not occur.
1772 */
1773 will_write = open_will_write_to_fs(flag, nd.path.dentry->d_inode);
1774 if (will_write) {
1775 error = mnt_want_write(nd.path.mnt);
1776 if (error)
1777 goto exit;
1778 }
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001779 error = may_open(&nd.path, acc_mode, flag);
Dave Hansen4a3fd212008-02-15 14:37:48 -08001780 if (error) {
1781 if (will_write)
1782 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 goto exit;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001784 }
1785 filp = nameidata_to_filp(&nd, open_flag);
1786 /*
1787 * It is now safe to drop the mnt write
1788 * because the filp has had a write taken
1789 * on its behalf.
1790 */
1791 if (will_write)
1792 mnt_drop_write(nd.path.mnt);
1793 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794
Dave Hansen4a3fd212008-02-15 14:37:48 -08001795exit_mutex_unlock:
1796 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797exit_dput:
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001798 path_put_conditional(&path, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799exit:
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001800 if (!IS_ERR(nd.intent.open.file))
1801 release_open_intent(&nd);
Al Viro8737f3a2008-08-02 22:36:57 -04001802exit_parent:
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001803 path_put(&nd.path);
1804 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805
1806do_link:
1807 error = -ELOOP;
1808 if (flag & O_NOFOLLOW)
1809 goto exit_dput;
1810 /*
1811 * This is subtle. Instead of calling do_follow_link() we do the
1812 * thing by hands. The reason is that this way we have zero link_count
1813 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1814 * After that we have the parent and last component, i.e.
1815 * we are in the same situation as after the first path_walk().
1816 * Well, almost - if the last component is normal we get its copy
1817 * stored in nd->last.name and we will have to putname() it when we
1818 * are done. Procfs-like symlinks just set LAST_BIND.
1819 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001820 nd.flags |= LOOKUP_PARENT;
1821 error = security_inode_follow_link(path.dentry, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822 if (error)
1823 goto exit_dput;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001824 error = __do_follow_link(&path, &nd);
Kirill Korotaevde459212006-07-14 00:23:49 -07001825 if (error) {
1826 /* Does someone understand code flow here? Or it is only
1827 * me so stupid? Anathema to whoever designed this non-sense
1828 * with "intent.open".
1829 */
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001830 release_open_intent(&nd);
1831 return ERR_PTR(error);
Kirill Korotaevde459212006-07-14 00:23:49 -07001832 }
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001833 nd.flags &= ~LOOKUP_PARENT;
1834 if (nd.last_type == LAST_BIND)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 goto ok;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 error = -EISDIR;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001837 if (nd.last_type != LAST_NORM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 goto exit;
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001839 if (nd.last.name[nd.last.len]) {
1840 __putname(nd.last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 goto exit;
1842 }
1843 error = -ELOOP;
1844 if (count++==32) {
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001845 __putname(nd.last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846 goto exit;
1847 }
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001848 dir = nd.path.dentry;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001849 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001850 path.dentry = lookup_hash(&nd);
1851 path.mnt = nd.path.mnt;
1852 __putname(nd.last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 goto do_last;
1854}
1855
1856/**
Christoph Hellwiga70e65d2008-02-15 14:37:28 -08001857 * filp_open - open file and return file pointer
1858 *
1859 * @filename: path to open
1860 * @flags: open flags as per the open(2) second argument
1861 * @mode: mode for the new file if O_CREAT is set, else ignored
1862 *
1863 * This is the helper to open a file from kernelspace if you really
1864 * have to. But in generally you should not do this, so please move
1865 * along, nothing to see here..
1866 */
1867struct file *filp_open(const char *filename, int flags, int mode)
1868{
1869 return do_filp_open(AT_FDCWD, filename, flags, mode);
1870}
1871EXPORT_SYMBOL(filp_open);
1872
1873/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 * lookup_create - lookup a dentry, creating it if it doesn't exist
1875 * @nd: nameidata info
1876 * @is_dir: directory flag
1877 *
1878 * Simple function to lookup and return a dentry and create it
1879 * if it doesn't exist. Is SMP-safe.
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001880 *
Jan Blunck4ac91372008-02-14 19:34:32 -08001881 * Returns with nd->path.dentry->d_inode->i_mutex locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 */
1883struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1884{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001885 struct dentry *dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886
Jan Blunck4ac91372008-02-14 19:34:32 -08001887 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001888 /*
1889 * Yucky last component or no last component at all?
1890 * (foo/., foo/.., /////)
1891 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 if (nd->last_type != LAST_NORM)
1893 goto fail;
1894 nd->flags &= ~LOOKUP_PARENT;
Al Viro35165862008-08-05 03:00:49 -04001895 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
ASANO Masahiroa6349042006-08-22 20:06:02 -04001896 nd->intent.open.flags = O_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001897
1898 /*
1899 * Do the final lookup.
1900 */
Christoph Hellwig49705b72005-11-08 21:35:06 -08001901 dentry = lookup_hash(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902 if (IS_ERR(dentry))
1903 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001904
Al Viroe9baf6e2008-05-15 04:49:12 -04001905 if (dentry->d_inode)
1906 goto eexist;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001907 /*
1908 * Special case - lookup gave negative, but... we had foo/bar/
1909 * From the vfs_mknod() POV we just have a negative dentry -
1910 * all is fine. Let's be bastards - you had / on the end, you've
1911 * been asking for (non-existent) directory. -ENOENT for you.
1912 */
Al Viroe9baf6e2008-05-15 04:49:12 -04001913 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1914 dput(dentry);
1915 dentry = ERR_PTR(-ENOENT);
1916 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 return dentry;
Al Viroe9baf6e2008-05-15 04:49:12 -04001918eexist:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 dput(dentry);
Al Viroe9baf6e2008-05-15 04:49:12 -04001920 dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921fail:
1922 return dentry;
1923}
Christoph Hellwigf81a0bf2005-05-19 12:26:43 -07001924EXPORT_SYMBOL_GPL(lookup_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925
1926int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1927{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001928 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929
1930 if (error)
1931 return error;
1932
1933 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1934 return -EPERM;
1935
Al Viroacfa4382008-12-04 10:06:33 -05001936 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 return -EPERM;
1938
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07001939 error = devcgroup_inode_mknod(mode, dev);
1940 if (error)
1941 return error;
1942
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943 error = security_inode_mknod(dir, dentry, mode, dev);
1944 if (error)
1945 return error;
1946
Jan Kara9e3509e2009-01-26 16:45:12 +01001947 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001949 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001950 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 return error;
1952}
1953
Dave Hansen463c3192008-02-15 14:37:57 -08001954static int may_mknod(mode_t mode)
1955{
1956 switch (mode & S_IFMT) {
1957 case S_IFREG:
1958 case S_IFCHR:
1959 case S_IFBLK:
1960 case S_IFIFO:
1961 case S_IFSOCK:
1962 case 0: /* zero mode translates to S_IFREG */
1963 return 0;
1964 case S_IFDIR:
1965 return -EPERM;
1966 default:
1967 return -EINVAL;
1968 }
1969}
1970
Heiko Carstens2e4d0922009-01-14 14:14:31 +01001971SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
1972 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973{
Al Viro2ad94ae2008-07-21 09:32:51 -04001974 int error;
1975 char *tmp;
1976 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 struct nameidata nd;
1978
1979 if (S_ISDIR(mode))
1980 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981
Al Viro2ad94ae2008-07-21 09:32:51 -04001982 error = user_path_parent(dfd, filename, &nd, &tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04001984 return error;
1985
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 dentry = lookup_create(&nd, 0);
Dave Hansen463c3192008-02-15 14:37:57 -08001987 if (IS_ERR(dentry)) {
1988 error = PTR_ERR(dentry);
1989 goto out_unlock;
1990 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001991 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 mode &= ~current->fs->umask;
Dave Hansen463c3192008-02-15 14:37:57 -08001993 error = may_mknod(mode);
1994 if (error)
1995 goto out_dput;
1996 error = mnt_want_write(nd.path.mnt);
1997 if (error)
1998 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09001999 error = security_path_mknod(&nd.path, dentry, mode, dev);
2000 if (error)
2001 goto out_drop_write;
Dave Hansen463c3192008-02-15 14:37:57 -08002002 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003 case 0: case S_IFREG:
Jan Blunck4ac91372008-02-14 19:34:32 -08002004 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005 break;
2006 case S_IFCHR: case S_IFBLK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002007 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008 new_decode_dev(dev));
2009 break;
2010 case S_IFIFO: case S_IFSOCK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002011 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002014out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002015 mnt_drop_write(nd.path.mnt);
2016out_dput:
2017 dput(dentry);
2018out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002019 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002020 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 putname(tmp);
2022
2023 return error;
2024}
2025
Heiko Carstens3480b252009-01-14 14:14:16 +01002026SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002027{
2028 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2029}
2030
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2032{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002033 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034
2035 if (error)
2036 return error;
2037
Al Viroacfa4382008-12-04 10:06:33 -05002038 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039 return -EPERM;
2040
2041 mode &= (S_IRWXUGO|S_ISVTX);
2042 error = security_inode_mkdir(dir, dentry, mode);
2043 if (error)
2044 return error;
2045
Jan Kara9e3509e2009-01-26 16:45:12 +01002046 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002048 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002049 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 return error;
2051}
2052
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002053SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054{
2055 int error = 0;
2056 char * tmp;
Dave Hansen6902d922006-09-30 23:29:01 -07002057 struct dentry *dentry;
2058 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059
Al Viro2ad94ae2008-07-21 09:32:51 -04002060 error = user_path_parent(dfd, pathname, &nd, &tmp);
2061 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002062 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063
Dave Hansen6902d922006-09-30 23:29:01 -07002064 dentry = lookup_create(&nd, 1);
2065 error = PTR_ERR(dentry);
2066 if (IS_ERR(dentry))
2067 goto out_unlock;
2068
Jan Blunck4ac91372008-02-14 19:34:32 -08002069 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Dave Hansen6902d922006-09-30 23:29:01 -07002070 mode &= ~current->fs->umask;
Dave Hansen463c3192008-02-15 14:37:57 -08002071 error = mnt_want_write(nd.path.mnt);
2072 if (error)
2073 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002074 error = security_path_mkdir(&nd.path, dentry, mode);
2075 if (error)
2076 goto out_drop_write;
Jan Blunck4ac91372008-02-14 19:34:32 -08002077 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002078out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002079 mnt_drop_write(nd.path.mnt);
2080out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002081 dput(dentry);
2082out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002083 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002084 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002085 putname(tmp);
2086out_err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087 return error;
2088}
2089
Heiko Carstens3cdad422009-01-14 14:14:22 +01002090SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002091{
2092 return sys_mkdirat(AT_FDCWD, pathname, mode);
2093}
2094
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095/*
2096 * We try to drop the dentry early: we should have
2097 * a usage count of 2 if we're the only user of this
2098 * dentry, and if that is true (possibly after pruning
2099 * the dcache), then we drop the dentry now.
2100 *
2101 * A low-level filesystem can, if it choses, legally
2102 * do a
2103 *
2104 * if (!d_unhashed(dentry))
2105 * return -EBUSY;
2106 *
2107 * if it cannot handle the case of removing a directory
2108 * that is still in use by something else..
2109 */
2110void dentry_unhash(struct dentry *dentry)
2111{
2112 dget(dentry);
Vasily Averindc168422006-12-06 20:37:07 -08002113 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114 spin_lock(&dcache_lock);
2115 spin_lock(&dentry->d_lock);
2116 if (atomic_read(&dentry->d_count) == 2)
2117 __d_drop(dentry);
2118 spin_unlock(&dentry->d_lock);
2119 spin_unlock(&dcache_lock);
2120}
2121
2122int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2123{
2124 int error = may_delete(dir, dentry, 1);
2125
2126 if (error)
2127 return error;
2128
Al Viroacfa4382008-12-04 10:06:33 -05002129 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 return -EPERM;
2131
Jan Kara9e3509e2009-01-26 16:45:12 +01002132 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002134 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135 dentry_unhash(dentry);
2136 if (d_mountpoint(dentry))
2137 error = -EBUSY;
2138 else {
2139 error = security_inode_rmdir(dir, dentry);
2140 if (!error) {
2141 error = dir->i_op->rmdir(dir, dentry);
2142 if (!error)
2143 dentry->d_inode->i_flags |= S_DEAD;
2144 }
2145 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002146 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148 d_delete(dentry);
2149 }
2150 dput(dentry);
2151
2152 return error;
2153}
2154
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002155static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156{
2157 int error = 0;
2158 char * name;
2159 struct dentry *dentry;
2160 struct nameidata nd;
2161
Al Viro2ad94ae2008-07-21 09:32:51 -04002162 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002164 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165
2166 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002167 case LAST_DOTDOT:
2168 error = -ENOTEMPTY;
2169 goto exit1;
2170 case LAST_DOT:
2171 error = -EINVAL;
2172 goto exit1;
2173 case LAST_ROOT:
2174 error = -EBUSY;
2175 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002177
2178 nd.flags &= ~LOOKUP_PARENT;
2179
Jan Blunck4ac91372008-02-14 19:34:32 -08002180 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002181 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002183 if (IS_ERR(dentry))
2184 goto exit2;
Dave Hansen06227532008-02-15 14:37:34 -08002185 error = mnt_want_write(nd.path.mnt);
2186 if (error)
2187 goto exit3;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002188 error = security_path_rmdir(&nd.path, dentry);
2189 if (error)
2190 goto exit4;
Jan Blunck4ac91372008-02-14 19:34:32 -08002191 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002192exit4:
Dave Hansen06227532008-02-15 14:37:34 -08002193 mnt_drop_write(nd.path.mnt);
2194exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07002195 dput(dentry);
2196exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08002197 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002199 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 putname(name);
2201 return error;
2202}
2203
Heiko Carstens3cdad422009-01-14 14:14:22 +01002204SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002205{
2206 return do_rmdir(AT_FDCWD, pathname);
2207}
2208
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209int vfs_unlink(struct inode *dir, struct dentry *dentry)
2210{
2211 int error = may_delete(dir, dentry, 0);
2212
2213 if (error)
2214 return error;
2215
Al Viroacfa4382008-12-04 10:06:33 -05002216 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217 return -EPERM;
2218
Jan Kara9e3509e2009-01-26 16:45:12 +01002219 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002221 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 if (d_mountpoint(dentry))
2223 error = -EBUSY;
2224 else {
2225 error = security_inode_unlink(dir, dentry);
2226 if (!error)
2227 error = dir->i_op->unlink(dir, dentry);
2228 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002229 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230
2231 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2232 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08002233 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 }
Robert Love0eeca282005-07-12 17:06:03 -04002236
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 return error;
2238}
2239
2240/*
2241 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002242 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243 * writeout happening, and we don't want to prevent access to the directory
2244 * while waiting on the I/O.
2245 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002246static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247{
Al Viro2ad94ae2008-07-21 09:32:51 -04002248 int error;
2249 char *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 struct dentry *dentry;
2251 struct nameidata nd;
2252 struct inode *inode = NULL;
2253
Al Viro2ad94ae2008-07-21 09:32:51 -04002254 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002256 return error;
2257
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 error = -EISDIR;
2259 if (nd.last_type != LAST_NORM)
2260 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002261
2262 nd.flags &= ~LOOKUP_PARENT;
2263
Jan Blunck4ac91372008-02-14 19:34:32 -08002264 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002265 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266 error = PTR_ERR(dentry);
2267 if (!IS_ERR(dentry)) {
2268 /* Why not before? Because we want correct error value */
2269 if (nd.last.name[nd.last.len])
2270 goto slashes;
2271 inode = dentry->d_inode;
2272 if (inode)
2273 atomic_inc(&inode->i_count);
Dave Hansen06227532008-02-15 14:37:34 -08002274 error = mnt_want_write(nd.path.mnt);
2275 if (error)
2276 goto exit2;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002277 error = security_path_unlink(&nd.path, dentry);
2278 if (error)
2279 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08002280 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002281exit3:
Dave Hansen06227532008-02-15 14:37:34 -08002282 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283 exit2:
2284 dput(dentry);
2285 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002286 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 if (inode)
2288 iput(inode); /* truncate the inode here */
2289exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002290 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291 putname(name);
2292 return error;
2293
2294slashes:
2295 error = !dentry->d_inode ? -ENOENT :
2296 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2297 goto exit2;
2298}
2299
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002300SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002301{
2302 if ((flag & ~AT_REMOVEDIR) != 0)
2303 return -EINVAL;
2304
2305 if (flag & AT_REMOVEDIR)
2306 return do_rmdir(dfd, pathname);
2307
2308 return do_unlinkat(dfd, pathname);
2309}
2310
Heiko Carstens3480b252009-01-14 14:14:16 +01002311SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002312{
2313 return do_unlinkat(AT_FDCWD, pathname);
2314}
2315
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002316int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002318 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319
2320 if (error)
2321 return error;
2322
Al Viroacfa4382008-12-04 10:06:33 -05002323 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324 return -EPERM;
2325
2326 error = security_inode_symlink(dir, dentry, oldname);
2327 if (error)
2328 return error;
2329
Jan Kara9e3509e2009-01-26 16:45:12 +01002330 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002332 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002333 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334 return error;
2335}
2336
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002337SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2338 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339{
Al Viro2ad94ae2008-07-21 09:32:51 -04002340 int error;
2341 char *from;
2342 char *to;
Dave Hansen6902d922006-09-30 23:29:01 -07002343 struct dentry *dentry;
2344 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345
2346 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04002347 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348 return PTR_ERR(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002349
2350 error = user_path_parent(newdfd, newname, &nd, &to);
2351 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002352 goto out_putname;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353
Dave Hansen6902d922006-09-30 23:29:01 -07002354 dentry = lookup_create(&nd, 0);
2355 error = PTR_ERR(dentry);
2356 if (IS_ERR(dentry))
2357 goto out_unlock;
2358
Dave Hansen75c3f292008-02-15 14:37:45 -08002359 error = mnt_want_write(nd.path.mnt);
2360 if (error)
2361 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002362 error = security_path_symlink(&nd.path, dentry, from);
2363 if (error)
2364 goto out_drop_write;
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002365 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002366out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002367 mnt_drop_write(nd.path.mnt);
2368out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002369 dput(dentry);
2370out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002371 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002372 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002373 putname(to);
2374out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375 putname(from);
2376 return error;
2377}
2378
Heiko Carstens3480b252009-01-14 14:14:16 +01002379SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002380{
2381 return sys_symlinkat(oldname, AT_FDCWD, newname);
2382}
2383
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2385{
2386 struct inode *inode = old_dentry->d_inode;
2387 int error;
2388
2389 if (!inode)
2390 return -ENOENT;
2391
Miklos Szeredia95164d2008-07-30 15:08:48 +02002392 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393 if (error)
2394 return error;
2395
2396 if (dir->i_sb != inode->i_sb)
2397 return -EXDEV;
2398
2399 /*
2400 * A link to an append-only or immutable file cannot be created.
2401 */
2402 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2403 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05002404 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002406 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407 return -EPERM;
2408
2409 error = security_inode_link(old_dentry, dir, new_dentry);
2410 if (error)
2411 return error;
2412
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002413 mutex_lock(&inode->i_mutex);
Jan Kara9e3509e2009-01-26 16:45:12 +01002414 vfs_dq_init(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415 error = dir->i_op->link(old_dentry, dir, new_dentry);
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002416 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002417 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002418 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419 return error;
2420}
2421
2422/*
2423 * Hardlinks are often used in delicate situations. We avoid
2424 * security-related surprises by not following symlinks on the
2425 * newname. --KAB
2426 *
2427 * We don't follow them on the oldname either to be compatible
2428 * with linux 2.0, and to avoid hard-linking to directories
2429 * and other special files. --ADM
2430 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002431SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2432 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433{
2434 struct dentry *new_dentry;
Al Viro2d8f3032008-07-22 09:59:21 -04002435 struct nameidata nd;
2436 struct path old_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437 int error;
Al Viro2ad94ae2008-07-21 09:32:51 -04002438 char *to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439
Ulrich Drepper45c9b112006-06-25 05:49:11 -07002440 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002441 return -EINVAL;
2442
Al Viro2d8f3032008-07-22 09:59:21 -04002443 error = user_path_at(olddfd, oldname,
2444 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2445 &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002447 return error;
2448
2449 error = user_path_parent(newdfd, newname, &nd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450 if (error)
2451 goto out;
2452 error = -EXDEV;
Al Viro2d8f3032008-07-22 09:59:21 -04002453 if (old_path.mnt != nd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454 goto out_release;
2455 new_dentry = lookup_create(&nd, 0);
2456 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002457 if (IS_ERR(new_dentry))
2458 goto out_unlock;
Dave Hansen75c3f292008-02-15 14:37:45 -08002459 error = mnt_want_write(nd.path.mnt);
2460 if (error)
2461 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002462 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2463 if (error)
2464 goto out_drop_write;
Al Viro2d8f3032008-07-22 09:59:21 -04002465 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002466out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002467 mnt_drop_write(nd.path.mnt);
2468out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002469 dput(new_dentry);
2470out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002471 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472out_release:
Jan Blunck1d957f92008-02-14 19:34:35 -08002473 path_put(&nd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002474 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475out:
Al Viro2d8f3032008-07-22 09:59:21 -04002476 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477
2478 return error;
2479}
2480
Heiko Carstens3480b252009-01-14 14:14:16 +01002481SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002482{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002483 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002484}
2485
Linus Torvalds1da177e2005-04-16 15:20:36 -07002486/*
2487 * The worst of all namespace operations - renaming directory. "Perverted"
2488 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2489 * Problems:
2490 * a) we can get into loop creation. Check is done in is_subdir().
2491 * b) race potential - two innocent renames can create a loop together.
2492 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002493 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494 * story.
2495 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002496 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497 * whether the target exists). Solution: try to be smart with locking
2498 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002499 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07002500 * move will be locked. Thus we can rank directories by the tree
2501 * (ancestors first) and rank all non-directories after them.
2502 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002503 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504 * HOWEVER, it relies on the assumption that any object with ->lookup()
2505 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2506 * we'd better make sure that there's no link(2) for them.
2507 * d) some filesystems don't support opened-but-unlinked directories,
2508 * either because of layout or because they are not ready to deal with
2509 * all cases correctly. The latter will be fixed (taking this sort of
2510 * stuff into VFS), but the former is not going away. Solution: the same
2511 * trick as in rmdir().
2512 * e) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002513 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002515 * ->i_mutex on parents, which works but leads to some truely excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516 * locking].
2517 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07002518static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2519 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520{
2521 int error = 0;
2522 struct inode *target;
2523
2524 /*
2525 * If we are going to change the parent - check write permissions,
2526 * we'll need to flip '..'.
2527 */
2528 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04002529 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530 if (error)
2531 return error;
2532 }
2533
2534 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2535 if (error)
2536 return error;
2537
2538 target = new_dentry->d_inode;
2539 if (target) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002540 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002541 dentry_unhash(new_dentry);
2542 }
2543 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2544 error = -EBUSY;
2545 else
2546 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2547 if (target) {
2548 if (!error)
2549 target->i_flags |= S_DEAD;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002550 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 if (d_unhashed(new_dentry))
2552 d_rehash(new_dentry);
2553 dput(new_dentry);
2554 }
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002555 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07002556 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2557 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 return error;
2559}
2560
Adrian Bunk75c96f82005-05-05 16:16:09 -07002561static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2562 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563{
2564 struct inode *target;
2565 int error;
2566
2567 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2568 if (error)
2569 return error;
2570
2571 dget(new_dentry);
2572 target = new_dentry->d_inode;
2573 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002574 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2576 error = -EBUSY;
2577 else
2578 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2579 if (!error) {
Mark Fasheh349457c2006-09-08 14:22:21 -07002580 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581 d_move(old_dentry, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582 }
2583 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002584 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585 dput(new_dentry);
2586 return error;
2587}
2588
2589int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2590 struct inode *new_dir, struct dentry *new_dentry)
2591{
2592 int error;
2593 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Robert Love0eeca282005-07-12 17:06:03 -04002594 const char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595
2596 if (old_dentry->d_inode == new_dentry->d_inode)
2597 return 0;
2598
2599 error = may_delete(old_dir, old_dentry, is_dir);
2600 if (error)
2601 return error;
2602
2603 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02002604 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605 else
2606 error = may_delete(new_dir, new_dentry, is_dir);
2607 if (error)
2608 return error;
2609
Al Viroacfa4382008-12-04 10:06:33 -05002610 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611 return -EPERM;
2612
Jan Kara9e3509e2009-01-26 16:45:12 +01002613 vfs_dq_init(old_dir);
2614 vfs_dq_init(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615
Robert Love0eeca282005-07-12 17:06:03 -04002616 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2617
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618 if (is_dir)
2619 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2620 else
2621 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2622 if (!error) {
Robert Love0eeca282005-07-12 17:06:03 -04002623 const char *new_name = old_dentry->d_name.name;
John McCutchan89204c42005-08-15 12:13:28 -04002624 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04002625 new_dentry->d_inode, old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626 }
Robert Love0eeca282005-07-12 17:06:03 -04002627 fsnotify_oldname_free(old_name);
2628
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629 return error;
2630}
2631
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002632SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2633 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634{
Al Viro2ad94ae2008-07-21 09:32:51 -04002635 struct dentry *old_dir, *new_dir;
2636 struct dentry *old_dentry, *new_dentry;
2637 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638 struct nameidata oldnd, newnd;
Al Viro2ad94ae2008-07-21 09:32:51 -04002639 char *from;
2640 char *to;
2641 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642
Al Viro2ad94ae2008-07-21 09:32:51 -04002643 error = user_path_parent(olddfd, oldname, &oldnd, &from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644 if (error)
2645 goto exit;
2646
Al Viro2ad94ae2008-07-21 09:32:51 -04002647 error = user_path_parent(newdfd, newname, &newnd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648 if (error)
2649 goto exit1;
2650
2651 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08002652 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002653 goto exit2;
2654
Jan Blunck4ac91372008-02-14 19:34:32 -08002655 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002656 error = -EBUSY;
2657 if (oldnd.last_type != LAST_NORM)
2658 goto exit2;
2659
Jan Blunck4ac91372008-02-14 19:34:32 -08002660 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661 if (newnd.last_type != LAST_NORM)
2662 goto exit2;
2663
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002664 oldnd.flags &= ~LOOKUP_PARENT;
2665 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09002666 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002667
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668 trap = lock_rename(new_dir, old_dir);
2669
Christoph Hellwig49705b72005-11-08 21:35:06 -08002670 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671 error = PTR_ERR(old_dentry);
2672 if (IS_ERR(old_dentry))
2673 goto exit3;
2674 /* source must exist */
2675 error = -ENOENT;
2676 if (!old_dentry->d_inode)
2677 goto exit4;
2678 /* unless the source is a directory trailing slashes give -ENOTDIR */
2679 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2680 error = -ENOTDIR;
2681 if (oldnd.last.name[oldnd.last.len])
2682 goto exit4;
2683 if (newnd.last.name[newnd.last.len])
2684 goto exit4;
2685 }
2686 /* source should not be ancestor of target */
2687 error = -EINVAL;
2688 if (old_dentry == trap)
2689 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08002690 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691 error = PTR_ERR(new_dentry);
2692 if (IS_ERR(new_dentry))
2693 goto exit4;
2694 /* target should not be an ancestor of source */
2695 error = -ENOTEMPTY;
2696 if (new_dentry == trap)
2697 goto exit5;
2698
Dave Hansen9079b1e2008-02-15 14:37:49 -08002699 error = mnt_want_write(oldnd.path.mnt);
2700 if (error)
2701 goto exit5;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002702 error = security_path_rename(&oldnd.path, old_dentry,
2703 &newnd.path, new_dentry);
2704 if (error)
2705 goto exit6;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706 error = vfs_rename(old_dir->d_inode, old_dentry,
2707 new_dir->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002708exit6:
Dave Hansen9079b1e2008-02-15 14:37:49 -08002709 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710exit5:
2711 dput(new_dentry);
2712exit4:
2713 dput(old_dentry);
2714exit3:
2715 unlock_rename(new_dir, old_dir);
2716exit2:
Jan Blunck1d957f92008-02-14 19:34:35 -08002717 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002718 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002719exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002720 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002721 putname(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002722exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002723 return error;
2724}
2725
Heiko Carstensa26eab22009-01-14 14:14:17 +01002726SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002727{
2728 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2729}
2730
Linus Torvalds1da177e2005-04-16 15:20:36 -07002731int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2732{
2733 int len;
2734
2735 len = PTR_ERR(link);
2736 if (IS_ERR(link))
2737 goto out;
2738
2739 len = strlen(link);
2740 if (len > (unsigned) buflen)
2741 len = buflen;
2742 if (copy_to_user(buffer, link, len))
2743 len = -EFAULT;
2744out:
2745 return len;
2746}
2747
2748/*
2749 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2750 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2751 * using) it for any given inode is up to filesystem.
2752 */
2753int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2754{
2755 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002756 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07002757 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002758
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002760 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07002761 if (IS_ERR(cookie))
2762 return PTR_ERR(cookie);
2763
2764 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2765 if (dentry->d_inode->i_op->put_link)
2766 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2767 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002768}
2769
2770int vfs_follow_link(struct nameidata *nd, const char *link)
2771{
2772 return __vfs_follow_link(nd, link);
2773}
2774
2775/* get the link contents into pagecache */
2776static char *page_getlink(struct dentry * dentry, struct page **ppage)
2777{
Duane Griffinebd09ab2008-12-19 20:47:12 +00002778 char *kaddr;
2779 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07002781 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07002783 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00002785 kaddr = kmap(page);
2786 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2787 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788}
2789
2790int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2791{
2792 struct page *page = NULL;
2793 char *s = page_getlink(dentry, &page);
2794 int res = vfs_readlink(dentry,buffer,buflen,s);
2795 if (page) {
2796 kunmap(page);
2797 page_cache_release(page);
2798 }
2799 return res;
2800}
2801
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002802void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002803{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002804 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002805 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002806 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002807}
2808
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002809void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002810{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002811 struct page *page = cookie;
2812
2813 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814 kunmap(page);
2815 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002816 }
2817}
2818
Nick Piggin54566b22009-01-04 12:00:53 -08002819/*
2820 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2821 */
2822int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002823{
2824 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002825 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002826 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08002827 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002828 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08002829 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2830 if (nofs)
2831 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002832
NeilBrown7e53cac2006-03-25 03:07:57 -08002833retry:
Nick Pigginafddba42007-10-16 01:25:01 -07002834 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08002835 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07002837 goto fail;
2838
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839 kaddr = kmap_atomic(page, KM_USER0);
2840 memcpy(kaddr, symname, len-1);
2841 kunmap_atomic(kaddr, KM_USER0);
Nick Pigginafddba42007-10-16 01:25:01 -07002842
2843 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2844 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845 if (err < 0)
2846 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07002847 if (err < len-1)
2848 goto retry;
2849
Linus Torvalds1da177e2005-04-16 15:20:36 -07002850 mark_inode_dirty(inode);
2851 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852fail:
2853 return err;
2854}
2855
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002856int page_symlink(struct inode *inode, const char *symname, int len)
2857{
2858 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08002859 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002860}
2861
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002862const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863 .readlink = generic_readlink,
2864 .follow_link = page_follow_link_light,
2865 .put_link = page_put_link,
2866};
2867
Al Viro2d8f3032008-07-22 09:59:21 -04002868EXPORT_SYMBOL(user_path_at);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002869EXPORT_SYMBOL(follow_down);
2870EXPORT_SYMBOL(follow_up);
2871EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2872EXPORT_SYMBOL(getname);
2873EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002874EXPORT_SYMBOL(lookup_one_len);
2875EXPORT_SYMBOL(page_follow_link_light);
2876EXPORT_SYMBOL(page_put_link);
2877EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002878EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879EXPORT_SYMBOL(page_symlink);
2880EXPORT_SYMBOL(page_symlink_inode_operations);
2881EXPORT_SYMBOL(path_lookup);
Al Virod1811462008-08-02 00:49:18 -04002882EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002883EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04002884EXPORT_SYMBOL(inode_permission);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -08002885EXPORT_SYMBOL(file_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002886EXPORT_SYMBOL(unlock_rename);
2887EXPORT_SYMBOL(vfs_create);
2888EXPORT_SYMBOL(vfs_follow_link);
2889EXPORT_SYMBOL(vfs_link);
2890EXPORT_SYMBOL(vfs_mkdir);
2891EXPORT_SYMBOL(vfs_mknod);
2892EXPORT_SYMBOL(generic_permission);
2893EXPORT_SYMBOL(vfs_readlink);
2894EXPORT_SYMBOL(vfs_rename);
2895EXPORT_SYMBOL(vfs_rmdir);
2896EXPORT_SYMBOL(vfs_symlink);
2897EXPORT_SYMBOL(vfs_unlink);
2898EXPORT_SYMBOL(dentry_unhash);
2899EXPORT_SYMBOL(generic_readlink);
Al Viro18d8fda2008-12-26 00:35:37 -05002900
2901/* to be mentioned only in INIT_TASK */
2902struct fs_struct init_fs = {
2903 .count = ATOMIC_INIT(1),
Steven Rostedt1239f262008-12-10 18:37:28 -05002904 .lock = __RW_LOCK_UNLOCKED(init_fs.lock),
Al Viro18d8fda2008-12-26 00:35:37 -05002905 .umask = 0022,
2906};