blob: 7cba63295a82f4aec85e0ce1f52c9f0fb9bbc8d7 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
18#include <linux/module.h>
19#include <linux/slab.h>
20#include <linux/fs.h>
21#include <linux/namei.h>
22#include <linux/quotaops.h>
23#include <linux/pagemap.h>
Robert Love0eeca282005-07-12 17:06:03 -040024#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/personality.h>
26#include <linux/security.h>
27#include <linux/syscalls.h>
28#include <linux/mount.h>
29#include <linux/audit.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080030#include <linux/capability.h>
Trond Myklebust834f2a42005-10-18 14:20:16 -070031#include <linux/file.h>
Ulrich Drepper5590ff02006-01-18 17:43:53 -080032#include <linux/fcntl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include <asm/namei.h>
34#include <asm/uaccess.h>
35
36#define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
37
38/* [Feb-1997 T. Schoebel-Theuer]
39 * Fundamental changes in the pathname lookup mechanisms (namei)
40 * were necessary because of omirr. The reason is that omirr needs
41 * to know the _real_ pathname, not the user-supplied one, in case
42 * of symlinks (and also when transname replacements occur).
43 *
44 * The new code replaces the old recursive symlink resolution with
45 * an iterative one (in case of non-nested symlink chains). It does
46 * this with calls to <fs>_follow_link().
47 * As a side effect, dir_namei(), _namei() and follow_link() are now
48 * replaced with a single function lookup_dentry() that can handle all
49 * the special cases of the former code.
50 *
51 * With the new dcache, the pathname is stored at each inode, at least as
52 * long as the refcount of the inode is positive. As a side effect, the
53 * size of the dcache depends on the inode cache and thus is dynamic.
54 *
55 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
56 * resolution to correspond with current state of the code.
57 *
58 * Note that the symlink resolution is not *completely* iterative.
59 * There is still a significant amount of tail- and mid- recursion in
60 * the algorithm. Also, note that <fs>_readlink() is not used in
61 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
62 * may return different results than <fs>_follow_link(). Many virtual
63 * filesystems (including /proc) exhibit this behavior.
64 */
65
66/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
67 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
68 * and the name already exists in form of a symlink, try to create the new
69 * name indicated by the symlink. The old code always complained that the
70 * name already exists, due to not following the symlink even if its target
71 * is nonexistent. The new semantics affects also mknod() and link() when
72 * the name is a symlink pointing to a non-existant name.
73 *
74 * I don't know which semantics is the right one, since I have no access
75 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
76 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
77 * "old" one. Personally, I think the new semantics is much more logical.
78 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
79 * file does succeed in both HP-UX and SunOs, but not in Solaris
80 * and in the old Linux semantics.
81 */
82
83/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
84 * semantics. See the comments in "open_namei" and "do_link" below.
85 *
86 * [10-Sep-98 Alan Modra] Another symlink change.
87 */
88
89/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
90 * inside the path - always follow.
91 * in the last component in creation/removal/renaming - never follow.
92 * if LOOKUP_FOLLOW passed - follow.
93 * if the pathname has trailing slashes - follow.
94 * otherwise - don't follow.
95 * (applied in that order).
96 *
97 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
98 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
99 * During the 2.4 we need to fix the userland stuff depending on it -
100 * hopefully we will be able to get rid of that wart in 2.5. So far only
101 * XEmacs seems to be relying on it...
102 */
103/*
104 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800105 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 * any extra contention...
107 */
108
Josef 'Jeff' Sipekc4a78082007-07-19 01:48:22 -0700109static int fastcall link_path_walk(const char *name, struct nameidata *nd);
110
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111/* In order to reduce some races, while at the same time doing additional
112 * checking and hopefully speeding things up, we copy filenames to the
113 * kernel data space before using them..
114 *
115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
116 * PATH_MAX includes the nul terminator --RR.
117 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800118static int do_getname(const char __user *filename, char *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119{
120 int retval;
121 unsigned long len = PATH_MAX;
122
123 if (!segment_eq(get_fs(), KERNEL_DS)) {
124 if ((unsigned long) filename >= TASK_SIZE)
125 return -EFAULT;
126 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
127 len = TASK_SIZE - (unsigned long) filename;
128 }
129
130 retval = strncpy_from_user(page, filename, len);
131 if (retval > 0) {
132 if (retval < len)
133 return 0;
134 return -ENAMETOOLONG;
135 } else if (!retval)
136 retval = -ENOENT;
137 return retval;
138}
139
140char * getname(const char __user * filename)
141{
142 char *tmp, *result;
143
144 result = ERR_PTR(-ENOMEM);
145 tmp = __getname();
146 if (tmp) {
147 int retval = do_getname(filename, tmp);
148
149 result = tmp;
150 if (retval < 0) {
151 __putname(tmp);
152 result = ERR_PTR(retval);
153 }
154 }
155 audit_getname(result);
156 return result;
157}
158
159#ifdef CONFIG_AUDITSYSCALL
160void putname(const char *name)
161{
Al Viro5ac3a9c2006-07-16 06:38:45 -0400162 if (unlikely(!audit_dummy_context()))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 audit_putname(name);
164 else
165 __putname(name);
166}
167EXPORT_SYMBOL(putname);
168#endif
169
170
171/**
172 * generic_permission - check for access rights on a Posix-like filesystem
173 * @inode: inode to check access rights for
174 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
175 * @check_acl: optional callback to check for Posix ACLs
176 *
177 * Used to check for read/write/execute permissions on a file.
178 * We use "fsuid" for this, letting us set arbitrary permissions
179 * for filesystem access without changing the "normal" uids which
180 * are used for other things..
181 */
182int generic_permission(struct inode *inode, int mask,
183 int (*check_acl)(struct inode *inode, int mask))
184{
185 umode_t mode = inode->i_mode;
186
187 if (current->fsuid == inode->i_uid)
188 mode >>= 6;
189 else {
190 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
191 int error = check_acl(inode, mask);
192 if (error == -EACCES)
193 goto check_capabilities;
194 else if (error != -EAGAIN)
195 return error;
196 }
197
198 if (in_group_p(inode->i_gid))
199 mode >>= 3;
200 }
201
202 /*
203 * If the DACs are ok we don't need any capability check.
204 */
205 if (((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask))
206 return 0;
207
208 check_capabilities:
209 /*
210 * Read/write DACs are always overridable.
211 * Executable DACs are overridable if at least one exec bit is set.
212 */
213 if (!(mask & MAY_EXEC) ||
214 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
215 if (capable(CAP_DAC_OVERRIDE))
216 return 0;
217
218 /*
219 * Searching includes executable on directories, else just read.
220 */
221 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
222 if (capable(CAP_DAC_READ_SEARCH))
223 return 0;
224
225 return -EACCES;
226}
227
228int permission(struct inode *inode, int mask, struct nameidata *nd)
229{
Trond Myklebusta343bb72006-08-22 20:06:03 -0400230 umode_t mode = inode->i_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 int retval, submask;
232
233 if (mask & MAY_WRITE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234
235 /*
236 * Nobody gets write access to a read-only fs.
237 */
238 if (IS_RDONLY(inode) &&
239 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
240 return -EROFS;
241
242 /*
243 * Nobody gets write access to an immutable file.
244 */
245 if (IS_IMMUTABLE(inode))
246 return -EACCES;
247 }
248
249
Trond Myklebusta343bb72006-08-22 20:06:03 -0400250 /*
251 * MAY_EXEC on regular files requires special handling: We override
Stas Sergeev317a40a2006-12-06 20:35:25 -0800252 * filesystem execute permissions if the mode bits aren't set or
253 * the fs is mounted with the "noexec" flag.
Trond Myklebusta343bb72006-08-22 20:06:03 -0400254 */
Stas Sergeev317a40a2006-12-06 20:35:25 -0800255 if ((mask & MAY_EXEC) && S_ISREG(mode) && (!(mode & S_IXUGO) ||
256 (nd && nd->mnt && (nd->mnt->mnt_flags & MNT_NOEXEC))))
Trond Myklebusta343bb72006-08-22 20:06:03 -0400257 return -EACCES;
258
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 /* Ordinary permission routines do not understand MAY_APPEND. */
260 submask = mask & ~MAY_APPEND;
261 if (inode->i_op && inode->i_op->permission)
262 retval = inode->i_op->permission(inode, submask, nd);
263 else
264 retval = generic_permission(inode, submask, NULL);
265 if (retval)
266 return retval;
267
268 return security_inode_permission(inode, mask, nd);
269}
270
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800271/**
272 * vfs_permission - check for access rights to a given path
273 * @nd: lookup result that describes the path
274 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
275 *
276 * Used to check for read/write/execute permissions on a path.
277 * We use "fsuid" for this, letting us set arbitrary permissions
278 * for filesystem access without changing the "normal" uids which
279 * are used for other things.
280 */
281int vfs_permission(struct nameidata *nd, int mask)
282{
283 return permission(nd->dentry->d_inode, mask, nd);
284}
285
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800286/**
287 * file_permission - check for additional access rights to a given file
288 * @file: file to check access rights for
289 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
290 *
291 * Used to check for read/write/execute permissions on an already opened
292 * file.
293 *
294 * Note:
295 * Do not use this function in new code. All access checks should
296 * be done using vfs_permission().
297 */
298int file_permission(struct file *file, int mask)
299{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800300 return permission(file->f_path.dentry->d_inode, mask, NULL);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800301}
302
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303/*
304 * get_write_access() gets write permission for a file.
305 * put_write_access() releases this write permission.
306 * This is used for regular files.
307 * We cannot support write (and maybe mmap read-write shared) accesses and
308 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
309 * can have the following values:
310 * 0: no writers, no VM_DENYWRITE mappings
311 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
312 * > 0: (i_writecount) users are writing to the file.
313 *
314 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
315 * except for the cases where we don't hold i_writecount yet. Then we need to
316 * use {get,deny}_write_access() - these functions check the sign and refuse
317 * to do the change if sign is wrong. Exclusion between them is provided by
318 * the inode->i_lock spinlock.
319 */
320
321int get_write_access(struct inode * inode)
322{
323 spin_lock(&inode->i_lock);
324 if (atomic_read(&inode->i_writecount) < 0) {
325 spin_unlock(&inode->i_lock);
326 return -ETXTBSY;
327 }
328 atomic_inc(&inode->i_writecount);
329 spin_unlock(&inode->i_lock);
330
331 return 0;
332}
333
334int deny_write_access(struct file * file)
335{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800336 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337
338 spin_lock(&inode->i_lock);
339 if (atomic_read(&inode->i_writecount) > 0) {
340 spin_unlock(&inode->i_lock);
341 return -ETXTBSY;
342 }
343 atomic_dec(&inode->i_writecount);
344 spin_unlock(&inode->i_lock);
345
346 return 0;
347}
348
349void path_release(struct nameidata *nd)
350{
351 dput(nd->dentry);
352 mntput(nd->mnt);
353}
354
355/*
356 * umount() mustn't call path_release()/mntput() as that would clear
357 * mnt_expiry_mark
358 */
359void path_release_on_umount(struct nameidata *nd)
360{
361 dput(nd->dentry);
Miklos Szeredi751c4042005-07-07 17:57:30 -0700362 mntput_no_expire(nd->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363}
364
Trond Myklebust834f2a42005-10-18 14:20:16 -0700365/**
366 * release_open_intent - free up open intent resources
367 * @nd: pointer to nameidata
368 */
369void release_open_intent(struct nameidata *nd)
370{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800371 if (nd->intent.open.file->f_path.dentry == NULL)
Trond Myklebust834f2a42005-10-18 14:20:16 -0700372 put_filp(nd->intent.open.file);
373 else
374 fput(nd->intent.open.file);
375}
376
Ian Kentbcdc5e02006-09-27 01:50:44 -0700377static inline struct dentry *
378do_revalidate(struct dentry *dentry, struct nameidata *nd)
379{
380 int status = dentry->d_op->d_revalidate(dentry, nd);
381 if (unlikely(status <= 0)) {
382 /*
383 * The dentry failed validation.
384 * If d_revalidate returned 0 attempt to invalidate
385 * the dentry otherwise d_revalidate is asking us
386 * to return a fail status.
387 */
388 if (!status) {
389 if (!d_invalidate(dentry)) {
390 dput(dentry);
391 dentry = NULL;
392 }
393 } else {
394 dput(dentry);
395 dentry = ERR_PTR(status);
396 }
397 }
398 return dentry;
399}
400
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401/*
402 * Internal lookup() using the new generic dcache.
403 * SMP-safe
404 */
405static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
406{
407 struct dentry * dentry = __d_lookup(parent, name);
408
409 /* lockess __d_lookup may fail due to concurrent d_move()
410 * in some unrelated directory, so try with d_lookup
411 */
412 if (!dentry)
413 dentry = d_lookup(parent, name);
414
Ian Kentbcdc5e02006-09-27 01:50:44 -0700415 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
416 dentry = do_revalidate(dentry, nd);
417
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 return dentry;
419}
420
421/*
422 * Short-cut version of permission(), for calling by
423 * path_walk(), when dcache lock is held. Combines parts
424 * of permission() and generic_permission(), and tests ONLY for
425 * MAY_EXEC permission.
426 *
427 * If appropriate, check DAC only. If not appropriate, or
428 * short-cut DAC fails, then call permission() to do more
429 * complete permission check.
430 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800431static int exec_permission_lite(struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 struct nameidata *nd)
433{
434 umode_t mode = inode->i_mode;
435
436 if (inode->i_op && inode->i_op->permission)
437 return -EAGAIN;
438
439 if (current->fsuid == inode->i_uid)
440 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:
458 return security_inode_permission(inode, MAY_EXEC, nd);
459}
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) {
491 struct dentry * dentry = d_alloc(parent, name);
492 result = ERR_PTR(-ENOMEM);
493 if (dentry) {
494 result = dir->i_op->lookup(dir, dentry, nd);
495 if (result)
496 dput(dentry);
497 else
498 result = dentry;
499 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800500 mutex_unlock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 return result;
502 }
503
504 /*
505 * Uhhuh! Nasty case: the cache was re-populated while
506 * we waited on the semaphore. Need to revalidate.
507 */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800508 mutex_unlock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 if (result->d_op && result->d_op->d_revalidate) {
Ian Kentbcdc5e02006-09-27 01:50:44 -0700510 result = do_revalidate(result, nd);
511 if (!result)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 result = ERR_PTR(-ENOENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 }
514 return result;
515}
516
517static int __emul_lookup_dentry(const char *, struct nameidata *);
518
519/* SMP-safe */
Arjan van de Venf1662352006-01-14 13:21:31 -0800520static __always_inline int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521walk_init_root(const char *name, struct nameidata *nd)
522{
Andreas Mohre518ddb2006-09-29 02:01:22 -0700523 struct fs_struct *fs = current->fs;
524
525 read_lock(&fs->lock);
526 if (fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
527 nd->mnt = mntget(fs->altrootmnt);
528 nd->dentry = dget(fs->altroot);
529 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 if (__emul_lookup_dentry(name,nd))
531 return 0;
Andreas Mohre518ddb2006-09-29 02:01:22 -0700532 read_lock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 }
Andreas Mohre518ddb2006-09-29 02:01:22 -0700534 nd->mnt = mntget(fs->rootmnt);
535 nd->dentry = dget(fs->root);
536 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 return 1;
538}
539
Arjan van de Venf1662352006-01-14 13:21:31 -0800540static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541{
542 int res = 0;
543 char *name;
544 if (IS_ERR(link))
545 goto fail;
546
547 if (*link == '/') {
548 path_release(nd);
549 if (!walk_init_root(link, nd))
550 /* weird __emul_prefix() stuff did it */
551 goto out;
552 }
553 res = link_path_walk(link, nd);
554out:
555 if (nd->depth || res || nd->last_type!=LAST_NORM)
556 return res;
557 /*
558 * If it is an iterative symlinks resolution in open_namei() we
559 * have to copy the last component. And all that crap because of
560 * bloody create() on broken symlinks. Furrfu...
561 */
562 name = __getname();
563 if (unlikely(!name)) {
564 path_release(nd);
565 return -ENOMEM;
566 }
567 strcpy(name, nd->last.name);
568 nd->last.name = name;
569 return 0;
570fail:
571 path_release(nd);
572 return PTR_ERR(link);
573}
574
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700575static inline void dput_path(struct path *path, struct nameidata *nd)
576{
577 dput(path->dentry);
578 if (path->mnt != nd->mnt)
579 mntput(path->mnt);
580}
581
582static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
583{
584 dput(nd->dentry);
585 if (nd->mnt != path->mnt)
586 mntput(nd->mnt);
587 nd->mnt = path->mnt;
588 nd->dentry = path->dentry;
589}
590
Ian Kent051d3812006-03-27 01:14:53 -0800591static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
592{
593 int error;
594 void *cookie;
595 struct dentry *dentry = path->dentry;
596
597 touch_atime(path->mnt, dentry);
598 nd_set_link(nd, NULL);
599
600 if (path->mnt != nd->mnt) {
601 path_to_nameidata(path, nd);
602 dget(dentry);
603 }
604 mntget(path->mnt);
605 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
606 error = PTR_ERR(cookie);
607 if (!IS_ERR(cookie)) {
608 char *s = nd_get_link(nd);
609 error = 0;
610 if (s)
611 error = __vfs_follow_link(nd, s);
612 if (dentry->d_inode->i_op->put_link)
613 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
614 }
615 dput(dentry);
616 mntput(path->mnt);
617
618 return error;
619}
620
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621/*
622 * This limits recursive symlink follows to 8, while
623 * limiting consecutive symlinks to 40.
624 *
625 * Without that kind of total limit, nasty chains of consecutive
626 * symlinks can cause almost arbitrarily long lookups.
627 */
Al Viro90ebe562005-06-06 13:35:58 -0700628static inline int do_follow_link(struct path *path, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629{
630 int err = -ELOOP;
631 if (current->link_count >= MAX_NESTED_LINKS)
632 goto loop;
633 if (current->total_link_count >= 40)
634 goto loop;
635 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
636 cond_resched();
Al Viro90ebe562005-06-06 13:35:58 -0700637 err = security_inode_follow_link(path->dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 if (err)
639 goto loop;
640 current->link_count++;
641 current->total_link_count++;
642 nd->depth++;
Al Virocd4e91d2005-06-06 13:36:03 -0700643 err = __do_follow_link(path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 current->link_count--;
645 nd->depth--;
646 return err;
647loop:
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700648 dput_path(path, nd);
Al Viro839d9f92005-06-06 13:36:02 -0700649 path_release(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 return err;
651}
652
653int follow_up(struct vfsmount **mnt, struct dentry **dentry)
654{
655 struct vfsmount *parent;
656 struct dentry *mountpoint;
657 spin_lock(&vfsmount_lock);
658 parent=(*mnt)->mnt_parent;
659 if (parent == *mnt) {
660 spin_unlock(&vfsmount_lock);
661 return 0;
662 }
663 mntget(parent);
664 mountpoint=dget((*mnt)->mnt_mountpoint);
665 spin_unlock(&vfsmount_lock);
666 dput(*dentry);
667 *dentry = mountpoint;
668 mntput(*mnt);
669 *mnt = parent;
670 return 1;
671}
672
673/* no need for dcache_lock, as serialization is taken care in
674 * namespace.c
675 */
Al Viro463ffb22005-06-06 13:36:05 -0700676static int __follow_mount(struct path *path)
677{
678 int res = 0;
679 while (d_mountpoint(path->dentry)) {
680 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
681 if (!mounted)
682 break;
683 dput(path->dentry);
684 if (res)
685 mntput(path->mnt);
686 path->mnt = mounted;
687 path->dentry = dget(mounted->mnt_root);
688 res = 1;
689 }
690 return res;
691}
692
Al Viro58c465e2005-06-06 13:36:13 -0700693static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 while (d_mountpoint(*dentry)) {
696 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
697 if (!mounted)
698 break;
Al Viro58c465e2005-06-06 13:36:13 -0700699 dput(*dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 mntput(*mnt);
701 *mnt = mounted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 *dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704}
705
706/* no need for dcache_lock, as serialization is taken care in
707 * namespace.c
708 */
Al Viroe13b2102005-06-06 13:36:06 -0700709int follow_down(struct vfsmount **mnt, struct dentry **dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710{
711 struct vfsmount *mounted;
712
713 mounted = lookup_mnt(*mnt, *dentry);
714 if (mounted) {
Al Viroe13b2102005-06-06 13:36:06 -0700715 dput(*dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 mntput(*mnt);
717 *mnt = mounted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 *dentry = dget(mounted->mnt_root);
719 return 1;
720 }
721 return 0;
722}
723
Arjan van de Venf1662352006-01-14 13:21:31 -0800724static __always_inline void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725{
Andreas Mohre518ddb2006-09-29 02:01:22 -0700726 struct fs_struct *fs = current->fs;
727
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 while(1) {
729 struct vfsmount *parent;
Al Viro58c465e2005-06-06 13:36:13 -0700730 struct dentry *old = nd->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731
Andreas Mohre518ddb2006-09-29 02:01:22 -0700732 read_lock(&fs->lock);
733 if (nd->dentry == fs->root &&
734 nd->mnt == fs->rootmnt) {
735 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 break;
737 }
Andreas Mohre518ddb2006-09-29 02:01:22 -0700738 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 spin_lock(&dcache_lock);
Al Viro58c465e2005-06-06 13:36:13 -0700740 if (nd->dentry != nd->mnt->mnt_root) {
741 nd->dentry = dget(nd->dentry->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 spin_unlock(&dcache_lock);
743 dput(old);
744 break;
745 }
746 spin_unlock(&dcache_lock);
747 spin_lock(&vfsmount_lock);
Al Viro58c465e2005-06-06 13:36:13 -0700748 parent = nd->mnt->mnt_parent;
749 if (parent == nd->mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 spin_unlock(&vfsmount_lock);
751 break;
752 }
753 mntget(parent);
Al Viro58c465e2005-06-06 13:36:13 -0700754 nd->dentry = dget(nd->mnt->mnt_mountpoint);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 spin_unlock(&vfsmount_lock);
756 dput(old);
Al Viro58c465e2005-06-06 13:36:13 -0700757 mntput(nd->mnt);
758 nd->mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 }
Al Viro58c465e2005-06-06 13:36:13 -0700760 follow_mount(&nd->mnt, &nd->dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761}
762
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763/*
764 * It's more convoluted than I'd like it to be, but... it's still fairly
765 * small and for now I'd prefer to have fast path as straight as possible.
766 * It _is_ time-critical.
767 */
768static int do_lookup(struct nameidata *nd, struct qstr *name,
769 struct path *path)
770{
771 struct vfsmount *mnt = nd->mnt;
772 struct dentry *dentry = __d_lookup(nd->dentry, name);
773
774 if (!dentry)
775 goto need_lookup;
776 if (dentry->d_op && dentry->d_op->d_revalidate)
777 goto need_revalidate;
778done:
779 path->mnt = mnt;
780 path->dentry = dentry;
Al Viro634ee702005-06-06 13:36:13 -0700781 __follow_mount(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 return 0;
783
784need_lookup:
785 dentry = real_lookup(nd->dentry, name, nd);
786 if (IS_ERR(dentry))
787 goto fail;
788 goto done;
789
790need_revalidate:
Ian Kentbcdc5e02006-09-27 01:50:44 -0700791 dentry = do_revalidate(dentry, nd);
792 if (!dentry)
793 goto need_lookup;
794 if (IS_ERR(dentry))
795 goto fail;
796 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797
798fail:
799 return PTR_ERR(dentry);
800}
801
802/*
803 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +0100804 * This is the basic name resolution function, turning a pathname into
805 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 *
Prasanna Medaea3834d2005-04-29 16:00:17 +0100807 * Returns 0 and nd will have valid dentry and mnt on success.
808 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 */
810static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
811{
812 struct path next;
813 struct inode *inode;
814 int err;
815 unsigned int lookup_flags = nd->flags;
816
817 while (*name=='/')
818 name++;
819 if (!*name)
820 goto return_reval;
821
822 inode = nd->dentry->d_inode;
823 if (nd->depth)
Trond Myklebustf55eab82006-02-04 23:28:01 -0800824 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825
826 /* At this point we know we have a real path component. */
827 for(;;) {
828 unsigned long hash;
829 struct qstr this;
830 unsigned int c;
831
Trond Myklebustcdce5d62005-10-18 14:20:18 -0700832 nd->flags |= LOOKUP_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 err = exec_permission_lite(inode, nd);
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800834 if (err == -EAGAIN)
835 err = vfs_permission(nd, MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 if (err)
837 break;
838
839 this.name = name;
840 c = *(const unsigned char *)name;
841
842 hash = init_name_hash();
843 do {
844 name++;
845 hash = partial_name_hash(c, hash);
846 c = *(const unsigned char *)name;
847 } while (c && (c != '/'));
848 this.len = name - (const char *) this.name;
849 this.hash = end_name_hash(hash);
850
851 /* remove trailing slashes? */
852 if (!c)
853 goto last_component;
854 while (*++name == '/');
855 if (!*name)
856 goto last_with_slashes;
857
858 /*
859 * "." and ".." are special - ".." especially so because it has
860 * to be able to know about the current root directory and
861 * parent relationships.
862 */
863 if (this.name[0] == '.') switch (this.len) {
864 default:
865 break;
866 case 2:
867 if (this.name[1] != '.')
868 break;
Al Viro58c465e2005-06-06 13:36:13 -0700869 follow_dotdot(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 inode = nd->dentry->d_inode;
871 /* fallthrough */
872 case 1:
873 continue;
874 }
875 /*
876 * See if the low-level filesystem might want
877 * to use its own hash..
878 */
879 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
880 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
881 if (err < 0)
882 break;
883 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 /* This does the actual lookups.. */
885 err = do_lookup(nd, &this, &next);
886 if (err)
887 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888
889 err = -ENOENT;
890 inode = next.dentry->d_inode;
891 if (!inode)
892 goto out_dput;
893 err = -ENOTDIR;
894 if (!inode->i_op)
895 goto out_dput;
896
897 if (inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700898 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 if (err)
900 goto return_err;
901 err = -ENOENT;
902 inode = nd->dentry->d_inode;
903 if (!inode)
904 break;
905 err = -ENOTDIR;
906 if (!inode->i_op)
907 break;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700908 } else
909 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 err = -ENOTDIR;
911 if (!inode->i_op->lookup)
912 break;
913 continue;
914 /* here ends the main loop */
915
916last_with_slashes:
917 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
918last_component:
Trond Myklebustf55eab82006-02-04 23:28:01 -0800919 /* Clear LOOKUP_CONTINUE iff it was previously unset */
920 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 if (lookup_flags & LOOKUP_PARENT)
922 goto lookup_parent;
923 if (this.name[0] == '.') switch (this.len) {
924 default:
925 break;
926 case 2:
927 if (this.name[1] != '.')
928 break;
Al Viro58c465e2005-06-06 13:36:13 -0700929 follow_dotdot(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 inode = nd->dentry->d_inode;
931 /* fallthrough */
932 case 1:
933 goto return_reval;
934 }
935 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
936 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
937 if (err < 0)
938 break;
939 }
940 err = do_lookup(nd, &this, &next);
941 if (err)
942 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 inode = next.dentry->d_inode;
944 if ((lookup_flags & LOOKUP_FOLLOW)
945 && inode && inode->i_op && inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700946 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 if (err)
948 goto return_err;
949 inode = nd->dentry->d_inode;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700950 } else
951 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 err = -ENOENT;
953 if (!inode)
954 break;
955 if (lookup_flags & LOOKUP_DIRECTORY) {
956 err = -ENOTDIR;
957 if (!inode->i_op || !inode->i_op->lookup)
958 break;
959 }
960 goto return_base;
961lookup_parent:
962 nd->last = this;
963 nd->last_type = LAST_NORM;
964 if (this.name[0] != '.')
965 goto return_base;
966 if (this.len == 1)
967 nd->last_type = LAST_DOT;
968 else if (this.len == 2 && this.name[1] == '.')
969 nd->last_type = LAST_DOTDOT;
970 else
971 goto return_base;
972return_reval:
973 /*
974 * We bypassed the ordinary revalidation routines.
975 * We may need to check the cached dentry for staleness.
976 */
977 if (nd->dentry && nd->dentry->d_sb &&
978 (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
979 err = -ESTALE;
980 /* Note: we do not d_invalidate() */
981 if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
982 break;
983 }
984return_base:
985 return 0;
986out_dput:
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700987 dput_path(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 break;
989 }
990 path_release(nd);
991return_err:
992 return err;
993}
994
995/*
996 * Wrapper to retry pathname resolution whenever the underlying
997 * file system returns an ESTALE.
998 *
999 * Retry the whole path once, forcing real lookup requests
1000 * instead of relying on the dcache.
1001 */
Josef 'Jeff' Sipekc4a78082007-07-19 01:48:22 -07001002static int fastcall link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003{
1004 struct nameidata save = *nd;
1005 int result;
1006
1007 /* make sure the stuff we saved doesn't go away */
1008 dget(save.dentry);
1009 mntget(save.mnt);
1010
1011 result = __link_path_walk(name, nd);
1012 if (result == -ESTALE) {
1013 *nd = save;
1014 dget(nd->dentry);
1015 mntget(nd->mnt);
1016 nd->flags |= LOOKUP_REVAL;
1017 result = __link_path_walk(name, nd);
1018 }
1019
1020 dput(save.dentry);
1021 mntput(save.mnt);
1022
1023 return result;
1024}
1025
Josef 'Jeff' Sipekf79c20f2007-07-19 01:48:22 -07001026static int fastcall path_walk(const char * name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027{
1028 current->total_link_count = 0;
1029 return link_path_walk(name, nd);
1030}
1031
Prasanna Medaea3834d2005-04-29 16:00:17 +01001032/*
1033 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
1034 * everything is done. Returns 0 and drops input nd, if lookup failed;
1035 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
1037{
1038 if (path_walk(name, nd))
1039 return 0; /* something went wrong... */
1040
1041 if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
1042 struct dentry *old_dentry = nd->dentry;
1043 struct vfsmount *old_mnt = nd->mnt;
1044 struct qstr last = nd->last;
1045 int last_type = nd->last_type;
Andreas Mohre518ddb2006-09-29 02:01:22 -07001046 struct fs_struct *fs = current->fs;
1047
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 /*
Andreas Mohre518ddb2006-09-29 02:01:22 -07001049 * NAME was not found in alternate root or it's a directory.
1050 * Try to find it in the normal root:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 */
1052 nd->last_type = LAST_ROOT;
Andreas Mohre518ddb2006-09-29 02:01:22 -07001053 read_lock(&fs->lock);
1054 nd->mnt = mntget(fs->rootmnt);
1055 nd->dentry = dget(fs->root);
1056 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 if (path_walk(name, nd) == 0) {
1058 if (nd->dentry->d_inode) {
1059 dput(old_dentry);
1060 mntput(old_mnt);
1061 return 1;
1062 }
1063 path_release(nd);
1064 }
1065 nd->dentry = old_dentry;
1066 nd->mnt = old_mnt;
1067 nd->last = last;
1068 nd->last_type = last_type;
1069 }
1070 return 1;
1071}
1072
1073void set_fs_altroot(void)
1074{
1075 char *emul = __emul_prefix();
1076 struct nameidata nd;
1077 struct vfsmount *mnt = NULL, *oldmnt;
1078 struct dentry *dentry = NULL, *olddentry;
1079 int err;
Andreas Mohre518ddb2006-09-29 02:01:22 -07001080 struct fs_struct *fs = current->fs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081
1082 if (!emul)
1083 goto set_it;
1084 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
1085 if (!err) {
1086 mnt = nd.mnt;
1087 dentry = nd.dentry;
1088 }
1089set_it:
Andreas Mohre518ddb2006-09-29 02:01:22 -07001090 write_lock(&fs->lock);
1091 oldmnt = fs->altrootmnt;
1092 olddentry = fs->altroot;
1093 fs->altrootmnt = mnt;
1094 fs->altroot = dentry;
1095 write_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 if (olddentry) {
1097 dput(olddentry);
1098 mntput(oldmnt);
1099 }
1100}
1101
Prasanna Medaea3834d2005-04-29 16:00:17 +01001102/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001103static int fastcall do_path_lookup(int dfd, const char *name,
1104 unsigned int flags, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001106 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001107 int fput_needed;
1108 struct file *file;
Andreas Mohre518ddb2006-09-29 02:01:22 -07001109 struct fs_struct *fs = current->fs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110
1111 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1112 nd->flags = flags;
1113 nd->depth = 0;
1114
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 if (*name=='/') {
Andreas Mohre518ddb2006-09-29 02:01:22 -07001116 read_lock(&fs->lock);
1117 if (fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
1118 nd->mnt = mntget(fs->altrootmnt);
1119 nd->dentry = dget(fs->altroot);
1120 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 if (__emul_lookup_dentry(name,nd))
Prasanna Medaea3834d2005-04-29 16:00:17 +01001122 goto out; /* found in altroot */
Andreas Mohre518ddb2006-09-29 02:01:22 -07001123 read_lock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 }
Andreas Mohre518ddb2006-09-29 02:01:22 -07001125 nd->mnt = mntget(fs->rootmnt);
1126 nd->dentry = dget(fs->root);
1127 read_unlock(&fs->lock);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001128 } else if (dfd == AT_FDCWD) {
Andreas Mohre518ddb2006-09-29 02:01:22 -07001129 read_lock(&fs->lock);
1130 nd->mnt = mntget(fs->pwdmnt);
1131 nd->dentry = dget(fs->pwd);
1132 read_unlock(&fs->lock);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001133 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001134 struct dentry *dentry;
1135
1136 file = fget_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001137 retval = -EBADF;
1138 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001139 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001140
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001141 dentry = file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001142
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001143 retval = -ENOTDIR;
1144 if (!S_ISDIR(dentry->d_inode->i_mode))
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001145 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001146
1147 retval = file_permission(file, MAY_EXEC);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001148 if (retval)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001149 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001150
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001151 nd->mnt = mntget(file->f_path.mnt);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001152 nd->dentry = dget(dentry);
1153
1154 fput_light(file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 }
Josef 'Jeff' Sipek2dfdd262007-05-09 02:33:41 -07001156
1157 retval = path_walk(name, nd);
Prasanna Medaea3834d2005-04-29 16:00:17 +01001158out:
Josef 'Jeff' Sipek62ce39c2007-05-09 02:33:41 -07001159 if (unlikely(!retval && !audit_dummy_context() && nd->dentry &&
Suzuki3bc8414b2006-02-07 12:58:36 -08001160 nd->dentry->d_inode))
Amy Griffis9c937dc2006-06-08 23:19:31 -04001161 audit_inode(name, nd->dentry->d_inode);
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001162out_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001163 return retval;
1164
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001165fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001166 fput_light(file, fput_needed);
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001167 goto out_fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168}
1169
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001170int fastcall path_lookup(const char *name, unsigned int flags,
1171 struct nameidata *nd)
1172{
1173 return do_path_lookup(AT_FDCWD, name, flags, nd);
1174}
1175
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001176/**
1177 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1178 * @dentry: pointer to dentry of the base directory
1179 * @mnt: pointer to vfs mount of the base directory
1180 * @name: pointer to file name
1181 * @flags: lookup flags
1182 * @nd: pointer to nameidata
1183 */
1184int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1185 const char *name, unsigned int flags,
1186 struct nameidata *nd)
1187{
1188 int retval;
1189
1190 /* same as do_path_lookup */
1191 nd->last_type = LAST_ROOT;
1192 nd->flags = flags;
1193 nd->depth = 0;
1194
1195 nd->mnt = mntget(mnt);
1196 nd->dentry = dget(dentry);
1197
1198 retval = path_walk(name, nd);
1199 if (unlikely(!retval && !audit_dummy_context() && nd->dentry &&
1200 nd->dentry->d_inode))
1201 audit_inode(name, nd->dentry->d_inode);
1202
1203 return retval;
1204
1205}
1206
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001207static int __path_lookup_intent_open(int dfd, const char *name,
1208 unsigned int lookup_flags, struct nameidata *nd,
1209 int open_flags, int create_mode)
Trond Myklebust834f2a42005-10-18 14:20:16 -07001210{
1211 struct file *filp = get_empty_filp();
1212 int err;
1213
1214 if (filp == NULL)
1215 return -ENFILE;
1216 nd->intent.open.file = filp;
1217 nd->intent.open.flags = open_flags;
1218 nd->intent.open.create_mode = create_mode;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001219 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
Trond Myklebust834f2a42005-10-18 14:20:16 -07001220 if (IS_ERR(nd->intent.open.file)) {
1221 if (err == 0) {
1222 err = PTR_ERR(nd->intent.open.file);
1223 path_release(nd);
1224 }
1225 } else if (err != 0)
1226 release_open_intent(nd);
1227 return err;
1228}
1229
1230/**
1231 * path_lookup_open - lookup a file path with open intent
Martin Waitz7045f372006-02-01 03:06:57 -08001232 * @dfd: the directory to use as base, or AT_FDCWD
Trond Myklebust834f2a42005-10-18 14:20:16 -07001233 * @name: pointer to file name
1234 * @lookup_flags: lookup intent flags
1235 * @nd: pointer to nameidata
1236 * @open_flags: open intent flags
1237 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001238int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
Trond Myklebust834f2a42005-10-18 14:20:16 -07001239 struct nameidata *nd, int open_flags)
1240{
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001241 return __path_lookup_intent_open(dfd, name, lookup_flags, nd,
Trond Myklebust834f2a42005-10-18 14:20:16 -07001242 open_flags, 0);
1243}
1244
1245/**
1246 * path_lookup_create - lookup a file path with open + create intent
Martin Waitz7045f372006-02-01 03:06:57 -08001247 * @dfd: the directory to use as base, or AT_FDCWD
Trond Myklebust834f2a42005-10-18 14:20:16 -07001248 * @name: pointer to file name
1249 * @lookup_flags: lookup intent flags
1250 * @nd: pointer to nameidata
1251 * @open_flags: open intent flags
1252 * @create_mode: create intent flags
1253 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001254static int path_lookup_create(int dfd, const char *name,
1255 unsigned int lookup_flags, struct nameidata *nd,
1256 int open_flags, int create_mode)
Trond Myklebust834f2a42005-10-18 14:20:16 -07001257{
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001258 return __path_lookup_intent_open(dfd, name, lookup_flags|LOOKUP_CREATE,
1259 nd, open_flags, create_mode);
Trond Myklebust834f2a42005-10-18 14:20:16 -07001260}
1261
1262int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags,
1263 struct nameidata *nd, int open_flags)
1264{
1265 char *tmp = getname(name);
1266 int err = PTR_ERR(tmp);
1267
1268 if (!IS_ERR(tmp)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001269 err = __path_lookup_intent_open(AT_FDCWD, tmp, lookup_flags, nd, open_flags, 0);
Trond Myklebust834f2a42005-10-18 14:20:16 -07001270 putname(tmp);
1271 }
1272 return err;
1273}
1274
Christoph Hellwigeead1912007-10-16 23:25:38 -07001275static struct dentry *__lookup_hash(struct qstr *name,
1276 struct dentry *base, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277{
James Morris057f6c02007-04-26 00:12:05 -07001278 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 struct inode *inode;
1280 int err;
1281
1282 inode = base->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283
1284 /*
1285 * See if the low-level filesystem might want
1286 * to use its own hash..
1287 */
1288 if (base->d_op && base->d_op->d_hash) {
1289 err = base->d_op->d_hash(base, name);
1290 dentry = ERR_PTR(err);
1291 if (err < 0)
1292 goto out;
1293 }
1294
1295 dentry = cached_lookup(base, name, nd);
1296 if (!dentry) {
1297 struct dentry *new = d_alloc(base, name);
1298 dentry = ERR_PTR(-ENOMEM);
1299 if (!new)
1300 goto out;
1301 dentry = inode->i_op->lookup(inode, new, nd);
1302 if (!dentry)
1303 dentry = new;
1304 else
1305 dput(new);
1306 }
1307out:
1308 return dentry;
1309}
1310
James Morris057f6c02007-04-26 00:12:05 -07001311/*
1312 * Restricted form of lookup. Doesn't follow links, single-component only,
1313 * needs parent already locked. Doesn't follow mounts.
1314 * SMP-safe.
1315 */
Adrian Bunka244e162006-03-31 02:32:11 -08001316static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317{
Christoph Hellwigeead1912007-10-16 23:25:38 -07001318 int err;
1319
1320 err = permission(nd->dentry->d_inode, MAY_EXEC, nd);
1321 if (err)
1322 return ERR_PTR(err);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001323 return __lookup_hash(&nd->last, nd->dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324}
1325
Christoph Hellwigeead1912007-10-16 23:25:38 -07001326static int __lookup_one_len(const char *name, struct qstr *this,
1327 struct dentry *base, int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328{
1329 unsigned long hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 unsigned int c;
1331
James Morris057f6c02007-04-26 00:12:05 -07001332 this->name = name;
1333 this->len = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 if (!len)
James Morris057f6c02007-04-26 00:12:05 -07001335 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336
1337 hash = init_name_hash();
1338 while (len--) {
1339 c = *(const unsigned char *)name++;
1340 if (c == '/' || c == '\0')
James Morris057f6c02007-04-26 00:12:05 -07001341 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 hash = partial_name_hash(c, hash);
1343 }
James Morris057f6c02007-04-26 00:12:05 -07001344 this->hash = end_name_hash(hash);
1345 return 0;
1346}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347
Christoph Hellwigeead1912007-10-16 23:25:38 -07001348/**
1349 * lookup_one_len: filesystem helper to lookup single pathname component
1350 * @name: pathname component to lookup
1351 * @base: base directory to lookup from
1352 * @len: maximum length @len should be interpreted to
1353 *
1354 * Note that this routine is purely a helper for filesystem useage and should
1355 * not be called by generic code. Also note that by using this function to
1356 * nameidata argument is passed to the filesystem methods and a filesystem
1357 * using this helper needs to be prepared for that.
1358 */
James Morris057f6c02007-04-26 00:12:05 -07001359struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1360{
1361 int err;
1362 struct qstr this;
1363
1364 err = __lookup_one_len(name, &this, base, len);
1365 if (err)
1366 return ERR_PTR(err);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001367
1368 err = permission(base->d_inode, MAY_EXEC, NULL);
1369 if (err)
1370 return ERR_PTR(err);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001371 return __lookup_hash(&this, base, NULL);
James Morris057f6c02007-04-26 00:12:05 -07001372}
1373
Christoph Hellwigeead1912007-10-16 23:25:38 -07001374/**
1375 * lookup_one_noperm - bad hack for sysfs
1376 * @name: pathname component to lookup
1377 * @base: base directory to lookup from
1378 *
1379 * This is a variant of lookup_one_len that doesn't perform any permission
1380 * checks. It's a horrible hack to work around the braindead sysfs
1381 * architecture and should not be used anywhere else.
1382 *
1383 * DON'T USE THIS FUNCTION EVER, thanks.
1384 */
1385struct dentry *lookup_one_noperm(const char *name, struct dentry *base)
James Morris057f6c02007-04-26 00:12:05 -07001386{
1387 int err;
1388 struct qstr this;
1389
Christoph Hellwigeead1912007-10-16 23:25:38 -07001390 err = __lookup_one_len(name, &this, base, strlen(name));
James Morris057f6c02007-04-26 00:12:05 -07001391 if (err)
1392 return ERR_PTR(err);
Christoph Hellwigeead1912007-10-16 23:25:38 -07001393 return __lookup_hash(&this, base, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394}
1395
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001396int fastcall __user_walk_fd(int dfd, const char __user *name, unsigned flags,
1397 struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398{
1399 char *tmp = getname(name);
1400 int err = PTR_ERR(tmp);
1401
1402 if (!IS_ERR(tmp)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001403 err = do_path_lookup(dfd, tmp, flags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 putname(tmp);
1405 }
1406 return err;
1407}
1408
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001409int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1410{
1411 return __user_walk_fd(AT_FDCWD, name, flags, nd);
1412}
1413
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414/*
1415 * It's inline, so penalty for filesystems that don't use sticky bit is
1416 * minimal.
1417 */
1418static inline int check_sticky(struct inode *dir, struct inode *inode)
1419{
1420 if (!(dir->i_mode & S_ISVTX))
1421 return 0;
1422 if (inode->i_uid == current->fsuid)
1423 return 0;
1424 if (dir->i_uid == current->fsuid)
1425 return 0;
1426 return !capable(CAP_FOWNER);
1427}
1428
1429/*
1430 * Check whether we can remove a link victim from directory dir, check
1431 * whether the type of victim is right.
1432 * 1. We can't do it if dir is read-only (done in permission())
1433 * 2. We should have write and exec permissions on dir
1434 * 3. We can't remove anything from append-only dir
1435 * 4. We can't do anything with immutable dir (done in permission())
1436 * 5. If the sticky bit on dir is set we should either
1437 * a. be owner of dir, or
1438 * b. be owner of victim, or
1439 * c. have CAP_FOWNER capability
1440 * 6. If the victim is append-only or immutable we can't do antyhing with
1441 * links pointing to it.
1442 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1443 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1444 * 9. We can't remove a root or mountpoint.
1445 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1446 * nfs_async_unlink().
1447 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001448static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449{
1450 int error;
1451
1452 if (!victim->d_inode)
1453 return -ENOENT;
1454
1455 BUG_ON(victim->d_parent->d_inode != dir);
Amy Griffis73d3ec52006-07-13 13:16:39 -04001456 audit_inode_child(victim->d_name.name, victim->d_inode, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457
1458 error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
1459 if (error)
1460 return error;
1461 if (IS_APPEND(dir))
1462 return -EPERM;
1463 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1464 IS_IMMUTABLE(victim->d_inode))
1465 return -EPERM;
1466 if (isdir) {
1467 if (!S_ISDIR(victim->d_inode->i_mode))
1468 return -ENOTDIR;
1469 if (IS_ROOT(victim))
1470 return -EBUSY;
1471 } else if (S_ISDIR(victim->d_inode->i_mode))
1472 return -EISDIR;
1473 if (IS_DEADDIR(dir))
1474 return -ENOENT;
1475 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1476 return -EBUSY;
1477 return 0;
1478}
1479
1480/* Check whether we can create an object with dentry child in directory
1481 * dir.
1482 * 1. We can't do it if child already exists (open has special treatment for
1483 * this case, but since we are inlined it's OK)
1484 * 2. We can't do it if dir is read-only (done in permission())
1485 * 3. We should have write and exec permissions on dir
1486 * 4. We can't do it if dir is immutable (done in permission())
1487 */
1488static inline int may_create(struct inode *dir, struct dentry *child,
1489 struct nameidata *nd)
1490{
1491 if (child->d_inode)
1492 return -EEXIST;
1493 if (IS_DEADDIR(dir))
1494 return -ENOENT;
1495 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1496}
1497
1498/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 * O_DIRECTORY translates into forcing a directory lookup.
1500 */
1501static inline int lookup_flags(unsigned int f)
1502{
1503 unsigned long retval = LOOKUP_FOLLOW;
1504
1505 if (f & O_NOFOLLOW)
1506 retval &= ~LOOKUP_FOLLOW;
1507
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 if (f & O_DIRECTORY)
1509 retval |= LOOKUP_DIRECTORY;
1510
1511 return retval;
1512}
1513
1514/*
1515 * p1 and p2 should be directories on the same fs.
1516 */
1517struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1518{
1519 struct dentry *p;
1520
1521 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001522 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 return NULL;
1524 }
1525
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001526 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527
1528 for (p = p1; p->d_parent != p; p = p->d_parent) {
1529 if (p->d_parent == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001530 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1531 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 return p;
1533 }
1534 }
1535
1536 for (p = p2; p->d_parent != p; p = p->d_parent) {
1537 if (p->d_parent == p1) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001538 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1539 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 return p;
1541 }
1542 }
1543
Ingo Molnarf2eace22006-07-03 00:25:05 -07001544 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1545 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 return NULL;
1547}
1548
1549void unlock_rename(struct dentry *p1, struct dentry *p2)
1550{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001551 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001553 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001554 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 }
1556}
1557
1558int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1559 struct nameidata *nd)
1560{
1561 int error = may_create(dir, dentry, nd);
1562
1563 if (error)
1564 return error;
1565
1566 if (!dir->i_op || !dir->i_op->create)
1567 return -EACCES; /* shouldn't it be ENOSYS? */
1568 mode &= S_IALLUGO;
1569 mode |= S_IFREG;
1570 error = security_inode_create(dir, dentry, mode);
1571 if (error)
1572 return error;
1573 DQUOT_INIT(dir);
1574 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001575 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001576 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 return error;
1578}
1579
1580int may_open(struct nameidata *nd, int acc_mode, int flag)
1581{
1582 struct dentry *dentry = nd->dentry;
1583 struct inode *inode = dentry->d_inode;
1584 int error;
1585
1586 if (!inode)
1587 return -ENOENT;
1588
1589 if (S_ISLNK(inode->i_mode))
1590 return -ELOOP;
1591
1592 if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1593 return -EISDIR;
1594
Christoph Hellwige4543ed2005-11-08 21:35:04 -08001595 error = vfs_permission(nd, acc_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 if (error)
1597 return error;
1598
1599 /*
1600 * FIFO's, sockets and device files are special: they don't
1601 * actually live on the filesystem itself, and as such you
1602 * can write to them even if the filesystem is read-only.
1603 */
1604 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1605 flag &= ~O_TRUNC;
1606 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1607 if (nd->mnt->mnt_flags & MNT_NODEV)
1608 return -EACCES;
1609
1610 flag &= ~O_TRUNC;
1611 } else if (IS_RDONLY(inode) && (flag & FMODE_WRITE))
1612 return -EROFS;
1613 /*
1614 * An append-only file must be opened in append mode for writing.
1615 */
1616 if (IS_APPEND(inode)) {
1617 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1618 return -EPERM;
1619 if (flag & O_TRUNC)
1620 return -EPERM;
1621 }
1622
1623 /* O_NOATIME can only be set by the owner or superuser */
1624 if (flag & O_NOATIME)
Satyam Sharma3bd858a2007-07-17 15:00:08 +05301625 if (!is_owner_or_cap(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 return -EPERM;
1627
1628 /*
1629 * Ensure there are no outstanding leases on the file.
1630 */
1631 error = break_lease(inode, flag);
1632 if (error)
1633 return error;
1634
1635 if (flag & O_TRUNC) {
1636 error = get_write_access(inode);
1637 if (error)
1638 return error;
1639
1640 /*
1641 * Refuse to truncate files with mandatory locks held on them.
1642 */
1643 error = locks_verify_locked(inode);
1644 if (!error) {
1645 DQUOT_INIT(inode);
1646
NeilBrown4a301312006-01-08 01:02:39 -08001647 error = do_truncate(dentry, 0, ATTR_MTIME|ATTR_CTIME, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 }
1649 put_write_access(inode);
1650 if (error)
1651 return error;
1652 } else
1653 if (flag & FMODE_WRITE)
1654 DQUOT_INIT(inode);
1655
1656 return 0;
1657}
1658
Dave Hansenaab520e2006-09-30 23:29:02 -07001659static int open_namei_create(struct nameidata *nd, struct path *path,
1660 int flag, int mode)
1661{
1662 int error;
1663 struct dentry *dir = nd->dentry;
1664
1665 if (!IS_POSIXACL(dir->d_inode))
1666 mode &= ~current->fs->umask;
1667 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1668 mutex_unlock(&dir->d_inode->i_mutex);
1669 dput(nd->dentry);
1670 nd->dentry = path->dentry;
1671 if (error)
1672 return error;
1673 /* Don't check for write permission, don't truncate */
1674 return may_open(nd, 0, flag & ~O_TRUNC);
1675}
1676
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677/*
1678 * open_namei()
1679 *
1680 * namei for open - this is in fact almost the whole open-routine.
1681 *
1682 * Note that the low bits of "flag" aren't the same as in the open
1683 * system call - they are 00 - no permissions needed
1684 * 01 - read permission needed
1685 * 10 - write permission needed
1686 * 11 - read/write permissions needed
1687 * which is a lot more logical, and also allows the "no perm" needed
1688 * for symlinks (where the permissions are checked later).
1689 * SMP-safe
1690 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001691int open_namei(int dfd, const char *pathname, int flag,
1692 int mode, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693{
Trond Myklebust834f2a42005-10-18 14:20:16 -07001694 int acc_mode, error;
Al Viro4e7506e2005-06-06 13:36:00 -07001695 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 struct dentry *dir;
1697 int count = 0;
1698
1699 acc_mode = ACC_MODE(flag);
1700
Trond Myklebust834f2a42005-10-18 14:20:16 -07001701 /* O_TRUNC implies we need access checks for write permissions */
1702 if (flag & O_TRUNC)
1703 acc_mode |= MAY_WRITE;
1704
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 /* Allow the LSM permission hook to distinguish append
1706 access from general write access. */
1707 if (flag & O_APPEND)
1708 acc_mode |= MAY_APPEND;
1709
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 /*
1711 * The simplest case - just a plain lookup.
1712 */
1713 if (!(flag & O_CREAT)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001714 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1715 nd, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 if (error)
1717 return error;
1718 goto ok;
1719 }
1720
1721 /*
1722 * Create - we need to know the parent.
1723 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001724 error = path_lookup_create(dfd,pathname,LOOKUP_PARENT,nd,flag,mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 if (error)
1726 return error;
1727
1728 /*
1729 * We have the parent and last component. First of all, check
1730 * that we are not asked to creat(2) an obvious directory - that
1731 * will not do.
1732 */
1733 error = -EISDIR;
1734 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1735 goto exit;
1736
1737 dir = nd->dentry;
1738 nd->flags &= ~LOOKUP_PARENT;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001739 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001740 path.dentry = lookup_hash(nd);
Al Virod73ffe12005-06-06 13:36:01 -07001741 path.mnt = nd->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742
1743do_last:
Al Viro4e7506e2005-06-06 13:36:00 -07001744 error = PTR_ERR(path.dentry);
1745 if (IS_ERR(path.dentry)) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001746 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 goto exit;
1748 }
1749
Oleg Drokin4af4c522006-03-25 03:06:54 -08001750 if (IS_ERR(nd->intent.open.file)) {
1751 mutex_unlock(&dir->d_inode->i_mutex);
1752 error = PTR_ERR(nd->intent.open.file);
1753 goto exit_dput;
1754 }
1755
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 /* Negative dentry, just create the file */
Al Viro4e7506e2005-06-06 13:36:00 -07001757 if (!path.dentry->d_inode) {
Dave Hansenaab520e2006-09-30 23:29:02 -07001758 error = open_namei_create(nd, &path, flag, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 if (error)
1760 goto exit;
Dave Hansenaab520e2006-09-30 23:29:02 -07001761 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 }
1763
1764 /*
1765 * It already exists.
1766 */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001767 mutex_unlock(&dir->d_inode->i_mutex);
Amy Griffis4fc03b92007-02-13 14:15:01 -05001768 audit_inode(pathname, path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769
1770 error = -EEXIST;
1771 if (flag & O_EXCL)
1772 goto exit_dput;
1773
Al Viroe13b2102005-06-06 13:36:06 -07001774 if (__follow_mount(&path)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 error = -ELOOP;
Al Viroba7a4c12005-06-06 13:36:08 -07001776 if (flag & O_NOFOLLOW)
1777 goto exit_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 }
Amy Griffis3e2efce2006-07-13 13:16:02 -04001779
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 error = -ENOENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001781 if (!path.dentry->d_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 goto exit_dput;
Al Viro4e7506e2005-06-06 13:36:00 -07001783 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 goto do_link;
1785
Miklos Szeredi09dd17d2005-09-06 15:18:21 -07001786 path_to_nameidata(&path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787 error = -EISDIR;
Al Viro4e7506e2005-06-06 13:36:00 -07001788 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 goto exit;
1790ok:
1791 error = may_open(nd, acc_mode, flag);
1792 if (error)
1793 goto exit;
1794 return 0;
1795
1796exit_dput:
Miklos Szeredi09dd17d2005-09-06 15:18:21 -07001797 dput_path(&path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798exit:
Trond Myklebust834f2a42005-10-18 14:20:16 -07001799 if (!IS_ERR(nd->intent.open.file))
1800 release_open_intent(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 path_release(nd);
1802 return error;
1803
1804do_link:
1805 error = -ELOOP;
1806 if (flag & O_NOFOLLOW)
1807 goto exit_dput;
1808 /*
1809 * This is subtle. Instead of calling do_follow_link() we do the
1810 * thing by hands. The reason is that this way we have zero link_count
1811 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1812 * After that we have the parent and last component, i.e.
1813 * we are in the same situation as after the first path_walk().
1814 * Well, almost - if the last component is normal we get its copy
1815 * stored in nd->last.name and we will have to putname() it when we
1816 * are done. Procfs-like symlinks just set LAST_BIND.
1817 */
1818 nd->flags |= LOOKUP_PARENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001819 error = security_inode_follow_link(path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 if (error)
1821 goto exit_dput;
Al Virocd4e91d2005-06-06 13:36:03 -07001822 error = __do_follow_link(&path, nd);
Kirill Korotaevde459212006-07-14 00:23:49 -07001823 if (error) {
1824 /* Does someone understand code flow here? Or it is only
1825 * me so stupid? Anathema to whoever designed this non-sense
1826 * with "intent.open".
1827 */
1828 release_open_intent(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829 return error;
Kirill Korotaevde459212006-07-14 00:23:49 -07001830 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 nd->flags &= ~LOOKUP_PARENT;
Al Virod671d5e2005-06-06 13:36:04 -07001832 if (nd->last_type == LAST_BIND)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833 goto ok;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 error = -EISDIR;
1835 if (nd->last_type != LAST_NORM)
1836 goto exit;
1837 if (nd->last.name[nd->last.len]) {
Linus Torvalds82984112005-10-06 21:54:21 -07001838 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 goto exit;
1840 }
1841 error = -ELOOP;
1842 if (count++==32) {
Linus Torvalds82984112005-10-06 21:54:21 -07001843 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844 goto exit;
1845 }
1846 dir = nd->dentry;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001847 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001848 path.dentry = lookup_hash(nd);
Al Virod671d5e2005-06-06 13:36:04 -07001849 path.mnt = nd->mnt;
Linus Torvalds82984112005-10-06 21:54:21 -07001850 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851 goto do_last;
1852}
1853
1854/**
1855 * lookup_create - lookup a dentry, creating it if it doesn't exist
1856 * @nd: nameidata info
1857 * @is_dir: directory flag
1858 *
1859 * Simple function to lookup and return a dentry and create it
1860 * if it doesn't exist. Is SMP-safe.
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001861 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001862 * Returns with nd->dentry->d_inode->i_mutex locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 */
1864struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1865{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001866 struct dentry *dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867
Ingo Molnarf2eace22006-07-03 00:25:05 -07001868 mutex_lock_nested(&nd->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001869 /*
1870 * Yucky last component or no last component at all?
1871 * (foo/., foo/.., /////)
1872 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873 if (nd->last_type != LAST_NORM)
1874 goto fail;
1875 nd->flags &= ~LOOKUP_PARENT;
ASANO Masahiroa6349042006-08-22 20:06:02 -04001876 nd->flags |= LOOKUP_CREATE;
1877 nd->intent.open.flags = O_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001878
1879 /*
1880 * Do the final lookup.
1881 */
Christoph Hellwig49705b72005-11-08 21:35:06 -08001882 dentry = lookup_hash(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883 if (IS_ERR(dentry))
1884 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001885
1886 /*
1887 * Special case - lookup gave negative, but... we had foo/bar/
1888 * From the vfs_mknod() POV we just have a negative dentry -
1889 * all is fine. Let's be bastards - you had / on the end, you've
1890 * been asking for (non-existent) directory. -ENOENT for you.
1891 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1893 goto enoent;
1894 return dentry;
1895enoent:
1896 dput(dentry);
1897 dentry = ERR_PTR(-ENOENT);
1898fail:
1899 return dentry;
1900}
Christoph Hellwigf81a0bf2005-05-19 12:26:43 -07001901EXPORT_SYMBOL_GPL(lookup_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902
1903int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1904{
1905 int error = may_create(dir, dentry, NULL);
1906
1907 if (error)
1908 return error;
1909
1910 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1911 return -EPERM;
1912
1913 if (!dir->i_op || !dir->i_op->mknod)
1914 return -EPERM;
1915
1916 error = security_inode_mknod(dir, dentry, mode, dev);
1917 if (error)
1918 return error;
1919
1920 DQUOT_INIT(dir);
1921 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001922 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001923 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 return error;
1925}
1926
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001927asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
1928 unsigned dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929{
1930 int error = 0;
1931 char * tmp;
1932 struct dentry * dentry;
1933 struct nameidata nd;
1934
1935 if (S_ISDIR(mode))
1936 return -EPERM;
1937 tmp = getname(filename);
1938 if (IS_ERR(tmp))
1939 return PTR_ERR(tmp);
1940
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001941 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942 if (error)
1943 goto out;
1944 dentry = lookup_create(&nd, 0);
1945 error = PTR_ERR(dentry);
1946
1947 if (!IS_POSIXACL(nd.dentry->d_inode))
1948 mode &= ~current->fs->umask;
1949 if (!IS_ERR(dentry)) {
1950 switch (mode & S_IFMT) {
1951 case 0: case S_IFREG:
1952 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1953 break;
1954 case S_IFCHR: case S_IFBLK:
1955 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,
1956 new_decode_dev(dev));
1957 break;
1958 case S_IFIFO: case S_IFSOCK:
1959 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,0);
1960 break;
1961 case S_IFDIR:
1962 error = -EPERM;
1963 break;
1964 default:
1965 error = -EINVAL;
1966 }
1967 dput(dentry);
1968 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001969 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970 path_release(&nd);
1971out:
1972 putname(tmp);
1973
1974 return error;
1975}
1976
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001977asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
1978{
1979 return sys_mknodat(AT_FDCWD, filename, mode, dev);
1980}
1981
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1983{
1984 int error = may_create(dir, dentry, NULL);
1985
1986 if (error)
1987 return error;
1988
1989 if (!dir->i_op || !dir->i_op->mkdir)
1990 return -EPERM;
1991
1992 mode &= (S_IRWXUGO|S_ISVTX);
1993 error = security_inode_mkdir(dir, dentry, mode);
1994 if (error)
1995 return error;
1996
1997 DQUOT_INIT(dir);
1998 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001999 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002000 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 return error;
2002}
2003
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002004asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005{
2006 int error = 0;
2007 char * tmp;
Dave Hansen6902d922006-09-30 23:29:01 -07002008 struct dentry *dentry;
2009 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010
2011 tmp = getname(pathname);
2012 error = PTR_ERR(tmp);
Dave Hansen6902d922006-09-30 23:29:01 -07002013 if (IS_ERR(tmp))
2014 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015
Dave Hansen6902d922006-09-30 23:29:01 -07002016 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
2017 if (error)
2018 goto out;
2019 dentry = lookup_create(&nd, 1);
2020 error = PTR_ERR(dentry);
2021 if (IS_ERR(dentry))
2022 goto out_unlock;
2023
2024 if (!IS_POSIXACL(nd.dentry->d_inode))
2025 mode &= ~current->fs->umask;
2026 error = vfs_mkdir(nd.dentry->d_inode, dentry, mode);
2027 dput(dentry);
2028out_unlock:
2029 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2030 path_release(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031out:
Dave Hansen6902d922006-09-30 23:29:01 -07002032 putname(tmp);
2033out_err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 return error;
2035}
2036
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002037asmlinkage long sys_mkdir(const char __user *pathname, int mode)
2038{
2039 return sys_mkdirat(AT_FDCWD, pathname, mode);
2040}
2041
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042/*
2043 * We try to drop the dentry early: we should have
2044 * a usage count of 2 if we're the only user of this
2045 * dentry, and if that is true (possibly after pruning
2046 * the dcache), then we drop the dentry now.
2047 *
2048 * A low-level filesystem can, if it choses, legally
2049 * do a
2050 *
2051 * if (!d_unhashed(dentry))
2052 * return -EBUSY;
2053 *
2054 * if it cannot handle the case of removing a directory
2055 * that is still in use by something else..
2056 */
2057void dentry_unhash(struct dentry *dentry)
2058{
2059 dget(dentry);
Vasily Averindc168422006-12-06 20:37:07 -08002060 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 spin_lock(&dcache_lock);
2062 spin_lock(&dentry->d_lock);
2063 if (atomic_read(&dentry->d_count) == 2)
2064 __d_drop(dentry);
2065 spin_unlock(&dentry->d_lock);
2066 spin_unlock(&dcache_lock);
2067}
2068
2069int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2070{
2071 int error = may_delete(dir, dentry, 1);
2072
2073 if (error)
2074 return error;
2075
2076 if (!dir->i_op || !dir->i_op->rmdir)
2077 return -EPERM;
2078
2079 DQUOT_INIT(dir);
2080
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002081 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082 dentry_unhash(dentry);
2083 if (d_mountpoint(dentry))
2084 error = -EBUSY;
2085 else {
2086 error = security_inode_rmdir(dir, dentry);
2087 if (!error) {
2088 error = dir->i_op->rmdir(dir, dentry);
2089 if (!error)
2090 dentry->d_inode->i_flags |= S_DEAD;
2091 }
2092 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002093 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 d_delete(dentry);
2096 }
2097 dput(dentry);
2098
2099 return error;
2100}
2101
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002102static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103{
2104 int error = 0;
2105 char * name;
2106 struct dentry *dentry;
2107 struct nameidata nd;
2108
2109 name = getname(pathname);
2110 if(IS_ERR(name))
2111 return PTR_ERR(name);
2112
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002113 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114 if (error)
2115 goto exit;
2116
2117 switch(nd.last_type) {
2118 case LAST_DOTDOT:
2119 error = -ENOTEMPTY;
2120 goto exit1;
2121 case LAST_DOT:
2122 error = -EINVAL;
2123 goto exit1;
2124 case LAST_ROOT:
2125 error = -EBUSY;
2126 goto exit1;
2127 }
Ingo Molnarf2eace22006-07-03 00:25:05 -07002128 mutex_lock_nested(&nd.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002129 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002131 if (IS_ERR(dentry))
2132 goto exit2;
2133 error = vfs_rmdir(nd.dentry->d_inode, dentry);
2134 dput(dentry);
2135exit2:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002136 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137exit1:
2138 path_release(&nd);
2139exit:
2140 putname(name);
2141 return error;
2142}
2143
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002144asmlinkage long sys_rmdir(const char __user *pathname)
2145{
2146 return do_rmdir(AT_FDCWD, pathname);
2147}
2148
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149int vfs_unlink(struct inode *dir, struct dentry *dentry)
2150{
2151 int error = may_delete(dir, dentry, 0);
2152
2153 if (error)
2154 return error;
2155
2156 if (!dir->i_op || !dir->i_op->unlink)
2157 return -EPERM;
2158
2159 DQUOT_INIT(dir);
2160
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002161 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 if (d_mountpoint(dentry))
2163 error = -EBUSY;
2164 else {
2165 error = security_inode_unlink(dir, dentry);
2166 if (!error)
2167 error = dir->i_op->unlink(dir, dentry);
2168 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002169 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170
2171 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2172 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2173 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 }
Robert Love0eeca282005-07-12 17:06:03 -04002175
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 return error;
2177}
2178
2179/*
2180 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002181 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 * writeout happening, and we don't want to prevent access to the directory
2183 * while waiting on the I/O.
2184 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002185static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186{
2187 int error = 0;
2188 char * name;
2189 struct dentry *dentry;
2190 struct nameidata nd;
2191 struct inode *inode = NULL;
2192
2193 name = getname(pathname);
2194 if(IS_ERR(name))
2195 return PTR_ERR(name);
2196
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002197 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198 if (error)
2199 goto exit;
2200 error = -EISDIR;
2201 if (nd.last_type != LAST_NORM)
2202 goto exit1;
Ingo Molnarf2eace22006-07-03 00:25:05 -07002203 mutex_lock_nested(&nd.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002204 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 error = PTR_ERR(dentry);
2206 if (!IS_ERR(dentry)) {
2207 /* Why not before? Because we want correct error value */
2208 if (nd.last.name[nd.last.len])
2209 goto slashes;
2210 inode = dentry->d_inode;
2211 if (inode)
2212 atomic_inc(&inode->i_count);
2213 error = vfs_unlink(nd.dentry->d_inode, dentry);
2214 exit2:
2215 dput(dentry);
2216 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002217 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 if (inode)
2219 iput(inode); /* truncate the inode here */
2220exit1:
2221 path_release(&nd);
2222exit:
2223 putname(name);
2224 return error;
2225
2226slashes:
2227 error = !dentry->d_inode ? -ENOENT :
2228 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2229 goto exit2;
2230}
2231
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002232asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2233{
2234 if ((flag & ~AT_REMOVEDIR) != 0)
2235 return -EINVAL;
2236
2237 if (flag & AT_REMOVEDIR)
2238 return do_rmdir(dfd, pathname);
2239
2240 return do_unlinkat(dfd, pathname);
2241}
2242
2243asmlinkage long sys_unlink(const char __user *pathname)
2244{
2245 return do_unlinkat(AT_FDCWD, pathname);
2246}
2247
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname, int mode)
2249{
2250 int error = may_create(dir, dentry, NULL);
2251
2252 if (error)
2253 return error;
2254
2255 if (!dir->i_op || !dir->i_op->symlink)
2256 return -EPERM;
2257
2258 error = security_inode_symlink(dir, dentry, oldname);
2259 if (error)
2260 return error;
2261
2262 DQUOT_INIT(dir);
2263 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002264 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002265 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266 return error;
2267}
2268
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002269asmlinkage long sys_symlinkat(const char __user *oldname,
2270 int newdfd, const char __user *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271{
2272 int error = 0;
2273 char * from;
2274 char * to;
Dave Hansen6902d922006-09-30 23:29:01 -07002275 struct dentry *dentry;
2276 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277
2278 from = getname(oldname);
2279 if(IS_ERR(from))
2280 return PTR_ERR(from);
2281 to = getname(newname);
2282 error = PTR_ERR(to);
Dave Hansen6902d922006-09-30 23:29:01 -07002283 if (IS_ERR(to))
2284 goto out_putname;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285
Dave Hansen6902d922006-09-30 23:29:01 -07002286 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
2287 if (error)
2288 goto out;
2289 dentry = lookup_create(&nd, 0);
2290 error = PTR_ERR(dentry);
2291 if (IS_ERR(dentry))
2292 goto out_unlock;
2293
2294 error = vfs_symlink(nd.dentry->d_inode, dentry, from, S_IALLUGO);
2295 dput(dentry);
2296out_unlock:
2297 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2298 path_release(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299out:
Dave Hansen6902d922006-09-30 23:29:01 -07002300 putname(to);
2301out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302 putname(from);
2303 return error;
2304}
2305
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002306asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2307{
2308 return sys_symlinkat(oldname, AT_FDCWD, newname);
2309}
2310
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2312{
2313 struct inode *inode = old_dentry->d_inode;
2314 int error;
2315
2316 if (!inode)
2317 return -ENOENT;
2318
2319 error = may_create(dir, new_dentry, NULL);
2320 if (error)
2321 return error;
2322
2323 if (dir->i_sb != inode->i_sb)
2324 return -EXDEV;
2325
2326 /*
2327 * A link to an append-only or immutable file cannot be created.
2328 */
2329 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2330 return -EPERM;
2331 if (!dir->i_op || !dir->i_op->link)
2332 return -EPERM;
2333 if (S_ISDIR(old_dentry->d_inode->i_mode))
2334 return -EPERM;
2335
2336 error = security_inode_link(old_dentry, dir, new_dentry);
2337 if (error)
2338 return error;
2339
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002340 mutex_lock(&old_dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341 DQUOT_INIT(dir);
2342 error = dir->i_op->link(old_dentry, dir, new_dentry);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002343 mutex_unlock(&old_dentry->d_inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002344 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002345 fsnotify_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346 return error;
2347}
2348
2349/*
2350 * Hardlinks are often used in delicate situations. We avoid
2351 * security-related surprises by not following symlinks on the
2352 * newname. --KAB
2353 *
2354 * We don't follow them on the oldname either to be compatible
2355 * with linux 2.0, and to avoid hard-linking to directories
2356 * and other special files. --ADM
2357 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002358asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002359 int newdfd, const char __user *newname,
2360 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361{
2362 struct dentry *new_dentry;
2363 struct nameidata nd, old_nd;
2364 int error;
2365 char * to;
2366
Ulrich Drepper45c9b112006-06-25 05:49:11 -07002367 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002368 return -EINVAL;
2369
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370 to = getname(newname);
2371 if (IS_ERR(to))
2372 return PTR_ERR(to);
2373
Ulrich Drepper45c9b112006-06-25 05:49:11 -07002374 error = __user_walk_fd(olddfd, oldname,
2375 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2376 &old_nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377 if (error)
2378 goto exit;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002379 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380 if (error)
2381 goto out;
2382 error = -EXDEV;
2383 if (old_nd.mnt != nd.mnt)
2384 goto out_release;
2385 new_dentry = lookup_create(&nd, 0);
2386 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002387 if (IS_ERR(new_dentry))
2388 goto out_unlock;
2389 error = vfs_link(old_nd.dentry, nd.dentry->d_inode, new_dentry);
2390 dput(new_dentry);
2391out_unlock:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002392 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393out_release:
2394 path_release(&nd);
2395out:
2396 path_release(&old_nd);
2397exit:
2398 putname(to);
2399
2400 return error;
2401}
2402
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002403asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2404{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002405 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002406}
2407
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408/*
2409 * The worst of all namespace operations - renaming directory. "Perverted"
2410 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2411 * Problems:
2412 * a) we can get into loop creation. Check is done in is_subdir().
2413 * b) race potential - two innocent renames can create a loop together.
2414 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002415 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416 * story.
2417 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002418 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419 * whether the target exists). Solution: try to be smart with locking
2420 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002421 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422 * move will be locked. Thus we can rank directories by the tree
2423 * (ancestors first) and rank all non-directories after them.
2424 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002425 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426 * HOWEVER, it relies on the assumption that any object with ->lookup()
2427 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2428 * we'd better make sure that there's no link(2) for them.
2429 * d) some filesystems don't support opened-but-unlinked directories,
2430 * either because of layout or because they are not ready to deal with
2431 * all cases correctly. The latter will be fixed (taking this sort of
2432 * stuff into VFS), but the former is not going away. Solution: the same
2433 * trick as in rmdir().
2434 * e) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002435 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002437 * ->i_mutex on parents, which works but leads to some truely excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438 * locking].
2439 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07002440static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2441 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442{
2443 int error = 0;
2444 struct inode *target;
2445
2446 /*
2447 * If we are going to change the parent - check write permissions,
2448 * we'll need to flip '..'.
2449 */
2450 if (new_dir != old_dir) {
2451 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2452 if (error)
2453 return error;
2454 }
2455
2456 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2457 if (error)
2458 return error;
2459
2460 target = new_dentry->d_inode;
2461 if (target) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002462 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463 dentry_unhash(new_dentry);
2464 }
2465 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2466 error = -EBUSY;
2467 else
2468 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2469 if (target) {
2470 if (!error)
2471 target->i_flags |= S_DEAD;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002472 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002473 if (d_unhashed(new_dentry))
2474 d_rehash(new_dentry);
2475 dput(new_dentry);
2476 }
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002477 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07002478 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2479 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002480 return error;
2481}
2482
Adrian Bunk75c96f82005-05-05 16:16:09 -07002483static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2484 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485{
2486 struct inode *target;
2487 int error;
2488
2489 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2490 if (error)
2491 return error;
2492
2493 dget(new_dentry);
2494 target = new_dentry->d_inode;
2495 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002496 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2498 error = -EBUSY;
2499 else
2500 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2501 if (!error) {
Mark Fasheh349457c2006-09-08 14:22:21 -07002502 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503 d_move(old_dentry, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504 }
2505 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002506 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507 dput(new_dentry);
2508 return error;
2509}
2510
2511int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2512 struct inode *new_dir, struct dentry *new_dentry)
2513{
2514 int error;
2515 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Robert Love0eeca282005-07-12 17:06:03 -04002516 const char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517
2518 if (old_dentry->d_inode == new_dentry->d_inode)
2519 return 0;
2520
2521 error = may_delete(old_dir, old_dentry, is_dir);
2522 if (error)
2523 return error;
2524
2525 if (!new_dentry->d_inode)
2526 error = may_create(new_dir, new_dentry, NULL);
2527 else
2528 error = may_delete(new_dir, new_dentry, is_dir);
2529 if (error)
2530 return error;
2531
2532 if (!old_dir->i_op || !old_dir->i_op->rename)
2533 return -EPERM;
2534
2535 DQUOT_INIT(old_dir);
2536 DQUOT_INIT(new_dir);
2537
Robert Love0eeca282005-07-12 17:06:03 -04002538 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2539
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540 if (is_dir)
2541 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2542 else
2543 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2544 if (!error) {
Robert Love0eeca282005-07-12 17:06:03 -04002545 const char *new_name = old_dentry->d_name.name;
John McCutchan89204c42005-08-15 12:13:28 -04002546 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2547 new_dentry->d_inode, old_dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548 }
Robert Love0eeca282005-07-12 17:06:03 -04002549 fsnotify_oldname_free(old_name);
2550
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 return error;
2552}
2553
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002554static int do_rename(int olddfd, const char *oldname,
2555 int newdfd, const char *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556{
2557 int error = 0;
2558 struct dentry * old_dir, * new_dir;
2559 struct dentry * old_dentry, *new_dentry;
2560 struct dentry * trap;
2561 struct nameidata oldnd, newnd;
2562
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002563 error = do_path_lookup(olddfd, oldname, LOOKUP_PARENT, &oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564 if (error)
2565 goto exit;
2566
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002567 error = do_path_lookup(newdfd, newname, LOOKUP_PARENT, &newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568 if (error)
2569 goto exit1;
2570
2571 error = -EXDEV;
2572 if (oldnd.mnt != newnd.mnt)
2573 goto exit2;
2574
2575 old_dir = oldnd.dentry;
2576 error = -EBUSY;
2577 if (oldnd.last_type != LAST_NORM)
2578 goto exit2;
2579
2580 new_dir = newnd.dentry;
2581 if (newnd.last_type != LAST_NORM)
2582 goto exit2;
2583
2584 trap = lock_rename(new_dir, old_dir);
2585
Christoph Hellwig49705b72005-11-08 21:35:06 -08002586 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587 error = PTR_ERR(old_dentry);
2588 if (IS_ERR(old_dentry))
2589 goto exit3;
2590 /* source must exist */
2591 error = -ENOENT;
2592 if (!old_dentry->d_inode)
2593 goto exit4;
2594 /* unless the source is a directory trailing slashes give -ENOTDIR */
2595 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2596 error = -ENOTDIR;
2597 if (oldnd.last.name[oldnd.last.len])
2598 goto exit4;
2599 if (newnd.last.name[newnd.last.len])
2600 goto exit4;
2601 }
2602 /* source should not be ancestor of target */
2603 error = -EINVAL;
2604 if (old_dentry == trap)
2605 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08002606 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607 error = PTR_ERR(new_dentry);
2608 if (IS_ERR(new_dentry))
2609 goto exit4;
2610 /* target should not be an ancestor of source */
2611 error = -ENOTEMPTY;
2612 if (new_dentry == trap)
2613 goto exit5;
2614
2615 error = vfs_rename(old_dir->d_inode, old_dentry,
2616 new_dir->d_inode, new_dentry);
2617exit5:
2618 dput(new_dentry);
2619exit4:
2620 dput(old_dentry);
2621exit3:
2622 unlock_rename(new_dir, old_dir);
2623exit2:
2624 path_release(&newnd);
2625exit1:
2626 path_release(&oldnd);
2627exit:
2628 return error;
2629}
2630
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002631asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2632 int newdfd, const char __user *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002633{
2634 int error;
2635 char * from;
2636 char * to;
2637
2638 from = getname(oldname);
2639 if(IS_ERR(from))
2640 return PTR_ERR(from);
2641 to = getname(newname);
2642 error = PTR_ERR(to);
2643 if (!IS_ERR(to)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002644 error = do_rename(olddfd, from, newdfd, to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645 putname(to);
2646 }
2647 putname(from);
2648 return error;
2649}
2650
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002651asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2652{
2653 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2654}
2655
Linus Torvalds1da177e2005-04-16 15:20:36 -07002656int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2657{
2658 int len;
2659
2660 len = PTR_ERR(link);
2661 if (IS_ERR(link))
2662 goto out;
2663
2664 len = strlen(link);
2665 if (len > (unsigned) buflen)
2666 len = buflen;
2667 if (copy_to_user(buffer, link, len))
2668 len = -EFAULT;
2669out:
2670 return len;
2671}
2672
2673/*
2674 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2675 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2676 * using) it for any given inode is up to filesystem.
2677 */
2678int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2679{
2680 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002681 void *cookie;
2682
Linus Torvalds1da177e2005-04-16 15:20:36 -07002683 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002684 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2685 if (!IS_ERR(cookie)) {
2686 int res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687 if (dentry->d_inode->i_op->put_link)
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002688 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2689 cookie = ERR_PTR(res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690 }
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002691 return PTR_ERR(cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692}
2693
2694int vfs_follow_link(struct nameidata *nd, const char *link)
2695{
2696 return __vfs_follow_link(nd, link);
2697}
2698
2699/* get the link contents into pagecache */
2700static char *page_getlink(struct dentry * dentry, struct page **ppage)
2701{
2702 struct page * page;
2703 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07002704 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07002706 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707 *ppage = page;
2708 return kmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709}
2710
2711int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2712{
2713 struct page *page = NULL;
2714 char *s = page_getlink(dentry, &page);
2715 int res = vfs_readlink(dentry,buffer,buflen,s);
2716 if (page) {
2717 kunmap(page);
2718 page_cache_release(page);
2719 }
2720 return res;
2721}
2722
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002723void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002724{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002725 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002727 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728}
2729
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002730void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002731{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002732 struct page *page = cookie;
2733
2734 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735 kunmap(page);
2736 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737 }
2738}
2739
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002740int __page_symlink(struct inode *inode, const char *symname, int len,
2741 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742{
2743 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002744 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002745 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08002746 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002747 char *kaddr;
2748
NeilBrown7e53cac2006-03-25 03:07:57 -08002749retry:
Nick Pigginafddba42007-10-16 01:25:01 -07002750 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2751 AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002752 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07002753 goto fail;
2754
Linus Torvalds1da177e2005-04-16 15:20:36 -07002755 kaddr = kmap_atomic(page, KM_USER0);
2756 memcpy(kaddr, symname, len-1);
2757 kunmap_atomic(kaddr, KM_USER0);
Nick Pigginafddba42007-10-16 01:25:01 -07002758
2759 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2760 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761 if (err < 0)
2762 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07002763 if (err < len-1)
2764 goto retry;
2765
Linus Torvalds1da177e2005-04-16 15:20:36 -07002766 mark_inode_dirty(inode);
2767 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002768fail:
2769 return err;
2770}
2771
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002772int page_symlink(struct inode *inode, const char *symname, int len)
2773{
2774 return __page_symlink(inode, symname, len,
2775 mapping_gfp_mask(inode->i_mapping));
2776}
2777
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002778const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779 .readlink = generic_readlink,
2780 .follow_link = page_follow_link_light,
2781 .put_link = page_put_link,
2782};
2783
2784EXPORT_SYMBOL(__user_walk);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002785EXPORT_SYMBOL(__user_walk_fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786EXPORT_SYMBOL(follow_down);
2787EXPORT_SYMBOL(follow_up);
2788EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2789EXPORT_SYMBOL(getname);
2790EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791EXPORT_SYMBOL(lookup_one_len);
2792EXPORT_SYMBOL(page_follow_link_light);
2793EXPORT_SYMBOL(page_put_link);
2794EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002795EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002796EXPORT_SYMBOL(page_symlink);
2797EXPORT_SYMBOL(page_symlink_inode_operations);
2798EXPORT_SYMBOL(path_lookup);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002799EXPORT_SYMBOL(vfs_path_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800EXPORT_SYMBOL(path_release);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002801EXPORT_SYMBOL(permission);
Christoph Hellwige4543ed2005-11-08 21:35:04 -08002802EXPORT_SYMBOL(vfs_permission);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -08002803EXPORT_SYMBOL(file_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002804EXPORT_SYMBOL(unlock_rename);
2805EXPORT_SYMBOL(vfs_create);
2806EXPORT_SYMBOL(vfs_follow_link);
2807EXPORT_SYMBOL(vfs_link);
2808EXPORT_SYMBOL(vfs_mkdir);
2809EXPORT_SYMBOL(vfs_mknod);
2810EXPORT_SYMBOL(generic_permission);
2811EXPORT_SYMBOL(vfs_readlink);
2812EXPORT_SYMBOL(vfs_rename);
2813EXPORT_SYMBOL(vfs_rmdir);
2814EXPORT_SYMBOL(vfs_symlink);
2815EXPORT_SYMBOL(vfs_unlink);
2816EXPORT_SYMBOL(dentry_unhash);
2817EXPORT_SYMBOL(generic_readlink);