blob: c72b940797fc0321fcb6043610bc50138a2325ed [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/smp_lock.h>
26#include <linux/personality.h>
27#include <linux/security.h>
28#include <linux/syscalls.h>
29#include <linux/mount.h>
30#include <linux/audit.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080031#include <linux/capability.h>
Trond Myklebust834f2a42005-10-18 14:20:16 -070032#include <linux/file.h>
Ulrich Drepper5590ff02006-01-18 17:43:53 -080033#include <linux/fcntl.h>
34#include <linux/namei.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <asm/namei.h>
36#include <asm/uaccess.h>
37
38#define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
39
40/* [Feb-1997 T. Schoebel-Theuer]
41 * Fundamental changes in the pathname lookup mechanisms (namei)
42 * were necessary because of omirr. The reason is that omirr needs
43 * to know the _real_ pathname, not the user-supplied one, in case
44 * of symlinks (and also when transname replacements occur).
45 *
46 * The new code replaces the old recursive symlink resolution with
47 * an iterative one (in case of non-nested symlink chains). It does
48 * this with calls to <fs>_follow_link().
49 * As a side effect, dir_namei(), _namei() and follow_link() are now
50 * replaced with a single function lookup_dentry() that can handle all
51 * the special cases of the former code.
52 *
53 * With the new dcache, the pathname is stored at each inode, at least as
54 * long as the refcount of the inode is positive. As a side effect, the
55 * size of the dcache depends on the inode cache and thus is dynamic.
56 *
57 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
58 * resolution to correspond with current state of the code.
59 *
60 * Note that the symlink resolution is not *completely* iterative.
61 * There is still a significant amount of tail- and mid- recursion in
62 * the algorithm. Also, note that <fs>_readlink() is not used in
63 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
64 * may return different results than <fs>_follow_link(). Many virtual
65 * filesystems (including /proc) exhibit this behavior.
66 */
67
68/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
69 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
70 * and the name already exists in form of a symlink, try to create the new
71 * name indicated by the symlink. The old code always complained that the
72 * name already exists, due to not following the symlink even if its target
73 * is nonexistent. The new semantics affects also mknod() and link() when
74 * the name is a symlink pointing to a non-existant name.
75 *
76 * I don't know which semantics is the right one, since I have no access
77 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
78 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
79 * "old" one. Personally, I think the new semantics is much more logical.
80 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
81 * file does succeed in both HP-UX and SunOs, but not in Solaris
82 * and in the old Linux semantics.
83 */
84
85/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
86 * semantics. See the comments in "open_namei" and "do_link" below.
87 *
88 * [10-Sep-98 Alan Modra] Another symlink change.
89 */
90
91/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
92 * inside the path - always follow.
93 * in the last component in creation/removal/renaming - never follow.
94 * if LOOKUP_FOLLOW passed - follow.
95 * if the pathname has trailing slashes - follow.
96 * otherwise - don't follow.
97 * (applied in that order).
98 *
99 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
100 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
101 * During the 2.4 we need to fix the userland stuff depending on it -
102 * hopefully we will be able to get rid of that wart in 2.5. So far only
103 * XEmacs seems to be relying on it...
104 */
105/*
106 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800107 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 * any extra contention...
109 */
110
111/* 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{
162 if (unlikely(current->audit_context))
163 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{
230 int retval, submask;
231
232 if (mask & MAY_WRITE) {
233 umode_t mode = inode->i_mode;
234
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
250 /* Ordinary permission routines do not understand MAY_APPEND. */
251 submask = mask & ~MAY_APPEND;
252 if (inode->i_op && inode->i_op->permission)
253 retval = inode->i_op->permission(inode, submask, nd);
254 else
255 retval = generic_permission(inode, submask, NULL);
256 if (retval)
257 return retval;
258
259 return security_inode_permission(inode, mask, nd);
260}
261
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800262/**
263 * vfs_permission - check for access rights to a given path
264 * @nd: lookup result that describes the path
265 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
266 *
267 * Used to check for read/write/execute permissions on a path.
268 * We use "fsuid" for this, letting us set arbitrary permissions
269 * for filesystem access without changing the "normal" uids which
270 * are used for other things.
271 */
272int vfs_permission(struct nameidata *nd, int mask)
273{
274 return permission(nd->dentry->d_inode, mask, nd);
275}
276
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800277/**
278 * file_permission - check for additional access rights to a given file
279 * @file: file to check access rights for
280 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
281 *
282 * Used to check for read/write/execute permissions on an already opened
283 * file.
284 *
285 * Note:
286 * Do not use this function in new code. All access checks should
287 * be done using vfs_permission().
288 */
289int file_permission(struct file *file, int mask)
290{
291 return permission(file->f_dentry->d_inode, mask, NULL);
292}
293
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294/*
295 * get_write_access() gets write permission for a file.
296 * put_write_access() releases this write permission.
297 * This is used for regular files.
298 * We cannot support write (and maybe mmap read-write shared) accesses and
299 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
300 * can have the following values:
301 * 0: no writers, no VM_DENYWRITE mappings
302 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
303 * > 0: (i_writecount) users are writing to the file.
304 *
305 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
306 * except for the cases where we don't hold i_writecount yet. Then we need to
307 * use {get,deny}_write_access() - these functions check the sign and refuse
308 * to do the change if sign is wrong. Exclusion between them is provided by
309 * the inode->i_lock spinlock.
310 */
311
312int get_write_access(struct inode * inode)
313{
314 spin_lock(&inode->i_lock);
315 if (atomic_read(&inode->i_writecount) < 0) {
316 spin_unlock(&inode->i_lock);
317 return -ETXTBSY;
318 }
319 atomic_inc(&inode->i_writecount);
320 spin_unlock(&inode->i_lock);
321
322 return 0;
323}
324
325int deny_write_access(struct file * file)
326{
327 struct inode *inode = file->f_dentry->d_inode;
328
329 spin_lock(&inode->i_lock);
330 if (atomic_read(&inode->i_writecount) > 0) {
331 spin_unlock(&inode->i_lock);
332 return -ETXTBSY;
333 }
334 atomic_dec(&inode->i_writecount);
335 spin_unlock(&inode->i_lock);
336
337 return 0;
338}
339
340void path_release(struct nameidata *nd)
341{
342 dput(nd->dentry);
343 mntput(nd->mnt);
344}
345
346/*
347 * umount() mustn't call path_release()/mntput() as that would clear
348 * mnt_expiry_mark
349 */
350void path_release_on_umount(struct nameidata *nd)
351{
352 dput(nd->dentry);
Miklos Szeredi751c4042005-07-07 17:57:30 -0700353 mntput_no_expire(nd->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354}
355
Trond Myklebust834f2a42005-10-18 14:20:16 -0700356/**
357 * release_open_intent - free up open intent resources
358 * @nd: pointer to nameidata
359 */
360void release_open_intent(struct nameidata *nd)
361{
362 if (nd->intent.open.file->f_dentry == NULL)
363 put_filp(nd->intent.open.file);
364 else
365 fput(nd->intent.open.file);
366}
367
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368/*
369 * Internal lookup() using the new generic dcache.
370 * SMP-safe
371 */
372static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
373{
374 struct dentry * dentry = __d_lookup(parent, name);
375
376 /* lockess __d_lookup may fail due to concurrent d_move()
377 * in some unrelated directory, so try with d_lookup
378 */
379 if (!dentry)
380 dentry = d_lookup(parent, name);
381
382 if (dentry && dentry->d_op && dentry->d_op->d_revalidate) {
383 if (!dentry->d_op->d_revalidate(dentry, nd) && !d_invalidate(dentry)) {
384 dput(dentry);
385 dentry = NULL;
386 }
387 }
388 return dentry;
389}
390
391/*
392 * Short-cut version of permission(), for calling by
393 * path_walk(), when dcache lock is held. Combines parts
394 * of permission() and generic_permission(), and tests ONLY for
395 * MAY_EXEC permission.
396 *
397 * If appropriate, check DAC only. If not appropriate, or
398 * short-cut DAC fails, then call permission() to do more
399 * complete permission check.
400 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800401static int exec_permission_lite(struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 struct nameidata *nd)
403{
404 umode_t mode = inode->i_mode;
405
406 if (inode->i_op && inode->i_op->permission)
407 return -EAGAIN;
408
409 if (current->fsuid == inode->i_uid)
410 mode >>= 6;
411 else if (in_group_p(inode->i_gid))
412 mode >>= 3;
413
414 if (mode & MAY_EXEC)
415 goto ok;
416
417 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
418 goto ok;
419
420 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
421 goto ok;
422
423 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
424 goto ok;
425
426 return -EACCES;
427ok:
428 return security_inode_permission(inode, MAY_EXEC, nd);
429}
430
431/*
432 * This is called when everything else fails, and we actually have
433 * to go to the low-level filesystem to find out what we should do..
434 *
435 * We get the directory semaphore, and after getting that we also
436 * make sure that nobody added the entry to the dcache in the meantime..
437 * SMP-safe
438 */
439static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
440{
441 struct dentry * result;
442 struct inode *dir = parent->d_inode;
443
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800444 mutex_lock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 /*
446 * First re-do the cached lookup just in case it was created
447 * while we waited for the directory semaphore..
448 *
449 * FIXME! This could use version numbering or similar to
450 * avoid unnecessary cache lookups.
451 *
452 * The "dcache_lock" is purely to protect the RCU list walker
453 * from concurrent renames at this point (we mustn't get false
454 * negatives from the RCU list walk here, unlike the optimistic
455 * fast walk).
456 *
457 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
458 */
459 result = d_lookup(parent, name);
460 if (!result) {
461 struct dentry * dentry = d_alloc(parent, name);
462 result = ERR_PTR(-ENOMEM);
463 if (dentry) {
464 result = dir->i_op->lookup(dir, dentry, nd);
465 if (result)
466 dput(dentry);
467 else
468 result = dentry;
469 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800470 mutex_unlock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 return result;
472 }
473
474 /*
475 * Uhhuh! Nasty case: the cache was re-populated while
476 * we waited on the semaphore. Need to revalidate.
477 */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800478 mutex_unlock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 if (result->d_op && result->d_op->d_revalidate) {
480 if (!result->d_op->d_revalidate(result, nd) && !d_invalidate(result)) {
481 dput(result);
482 result = ERR_PTR(-ENOENT);
483 }
484 }
485 return result;
486}
487
488static int __emul_lookup_dentry(const char *, struct nameidata *);
489
490/* SMP-safe */
Arjan van de Venf1662352006-01-14 13:21:31 -0800491static __always_inline int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492walk_init_root(const char *name, struct nameidata *nd)
493{
494 read_lock(&current->fs->lock);
495 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
496 nd->mnt = mntget(current->fs->altrootmnt);
497 nd->dentry = dget(current->fs->altroot);
498 read_unlock(&current->fs->lock);
499 if (__emul_lookup_dentry(name,nd))
500 return 0;
501 read_lock(&current->fs->lock);
502 }
503 nd->mnt = mntget(current->fs->rootmnt);
504 nd->dentry = dget(current->fs->root);
505 read_unlock(&current->fs->lock);
506 return 1;
507}
508
Arjan van de Venf1662352006-01-14 13:21:31 -0800509static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510{
511 int res = 0;
512 char *name;
513 if (IS_ERR(link))
514 goto fail;
515
516 if (*link == '/') {
517 path_release(nd);
518 if (!walk_init_root(link, nd))
519 /* weird __emul_prefix() stuff did it */
520 goto out;
521 }
522 res = link_path_walk(link, nd);
523out:
524 if (nd->depth || res || nd->last_type!=LAST_NORM)
525 return res;
526 /*
527 * If it is an iterative symlinks resolution in open_namei() we
528 * have to copy the last component. And all that crap because of
529 * bloody create() on broken symlinks. Furrfu...
530 */
531 name = __getname();
532 if (unlikely(!name)) {
533 path_release(nd);
534 return -ENOMEM;
535 }
536 strcpy(name, nd->last.name);
537 nd->last.name = name;
538 return 0;
539fail:
540 path_release(nd);
541 return PTR_ERR(link);
542}
543
Al Viro90ebe562005-06-06 13:35:58 -0700544struct path {
545 struct vfsmount *mnt;
546 struct dentry *dentry;
547};
548
Arjan van de Venf1662352006-01-14 13:21:31 -0800549static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550{
551 int error;
Linus Torvaldscc314ee2005-08-19 18:02:56 -0700552 void *cookie;
Al Virocd4e91d2005-06-06 13:36:03 -0700553 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554
Al Virod671a1c2005-06-06 13:36:14 -0700555 touch_atime(path->mnt, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 nd_set_link(nd, NULL);
Al Virocd4e91d2005-06-06 13:36:03 -0700557
Al Viro39ca6d42005-06-06 13:36:12 -0700558 if (path->mnt == nd->mnt)
559 mntget(path->mnt);
Linus Torvaldscc314ee2005-08-19 18:02:56 -0700560 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
561 error = PTR_ERR(cookie);
562 if (!IS_ERR(cookie)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 char *s = nd_get_link(nd);
Linus Torvaldscc314ee2005-08-19 18:02:56 -0700564 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 if (s)
566 error = __vfs_follow_link(nd, s);
567 if (dentry->d_inode->i_op->put_link)
Linus Torvaldscc314ee2005-08-19 18:02:56 -0700568 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 }
Al Virocd4e91d2005-06-06 13:36:03 -0700570 dput(dentry);
571 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572
573 return error;
574}
575
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700576static inline void dput_path(struct path *path, struct nameidata *nd)
577{
578 dput(path->dentry);
579 if (path->mnt != nd->mnt)
580 mntput(path->mnt);
581}
582
583static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
584{
585 dput(nd->dentry);
586 if (nd->mnt != path->mnt)
587 mntput(nd->mnt);
588 nd->mnt = path->mnt;
589 nd->dentry = path->dentry;
590}
591
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592/*
593 * This limits recursive symlink follows to 8, while
594 * limiting consecutive symlinks to 40.
595 *
596 * Without that kind of total limit, nasty chains of consecutive
597 * symlinks can cause almost arbitrarily long lookups.
598 */
Al Viro90ebe562005-06-06 13:35:58 -0700599static inline int do_follow_link(struct path *path, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600{
601 int err = -ELOOP;
602 if (current->link_count >= MAX_NESTED_LINKS)
603 goto loop;
604 if (current->total_link_count >= 40)
605 goto loop;
606 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
607 cond_resched();
Al Viro90ebe562005-06-06 13:35:58 -0700608 err = security_inode_follow_link(path->dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 if (err)
610 goto loop;
611 current->link_count++;
612 current->total_link_count++;
613 nd->depth++;
Al Virocd4e91d2005-06-06 13:36:03 -0700614 err = __do_follow_link(path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 current->link_count--;
616 nd->depth--;
617 return err;
618loop:
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700619 dput_path(path, nd);
Al Viro839d9f92005-06-06 13:36:02 -0700620 path_release(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 return err;
622}
623
624int follow_up(struct vfsmount **mnt, struct dentry **dentry)
625{
626 struct vfsmount *parent;
627 struct dentry *mountpoint;
628 spin_lock(&vfsmount_lock);
629 parent=(*mnt)->mnt_parent;
630 if (parent == *mnt) {
631 spin_unlock(&vfsmount_lock);
632 return 0;
633 }
634 mntget(parent);
635 mountpoint=dget((*mnt)->mnt_mountpoint);
636 spin_unlock(&vfsmount_lock);
637 dput(*dentry);
638 *dentry = mountpoint;
639 mntput(*mnt);
640 *mnt = parent;
641 return 1;
642}
643
644/* no need for dcache_lock, as serialization is taken care in
645 * namespace.c
646 */
Al Viro463ffb22005-06-06 13:36:05 -0700647static int __follow_mount(struct path *path)
648{
649 int res = 0;
650 while (d_mountpoint(path->dentry)) {
651 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
652 if (!mounted)
653 break;
654 dput(path->dentry);
655 if (res)
656 mntput(path->mnt);
657 path->mnt = mounted;
658 path->dentry = dget(mounted->mnt_root);
659 res = 1;
660 }
661 return res;
662}
663
Al Viro58c465e2005-06-06 13:36:13 -0700664static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 while (d_mountpoint(*dentry)) {
667 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
668 if (!mounted)
669 break;
Al Viro58c465e2005-06-06 13:36:13 -0700670 dput(*dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 mntput(*mnt);
672 *mnt = mounted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 *dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675}
676
677/* no need for dcache_lock, as serialization is taken care in
678 * namespace.c
679 */
Al Viroe13b2102005-06-06 13:36:06 -0700680int follow_down(struct vfsmount **mnt, struct dentry **dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681{
682 struct vfsmount *mounted;
683
684 mounted = lookup_mnt(*mnt, *dentry);
685 if (mounted) {
Al Viroe13b2102005-06-06 13:36:06 -0700686 dput(*dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 mntput(*mnt);
688 *mnt = mounted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 *dentry = dget(mounted->mnt_root);
690 return 1;
691 }
692 return 0;
693}
694
Arjan van de Venf1662352006-01-14 13:21:31 -0800695static __always_inline void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696{
697 while(1) {
698 struct vfsmount *parent;
Al Viro58c465e2005-06-06 13:36:13 -0700699 struct dentry *old = nd->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700
701 read_lock(&current->fs->lock);
Al Viro58c465e2005-06-06 13:36:13 -0700702 if (nd->dentry == current->fs->root &&
703 nd->mnt == current->fs->rootmnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 read_unlock(&current->fs->lock);
705 break;
706 }
707 read_unlock(&current->fs->lock);
708 spin_lock(&dcache_lock);
Al Viro58c465e2005-06-06 13:36:13 -0700709 if (nd->dentry != nd->mnt->mnt_root) {
710 nd->dentry = dget(nd->dentry->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 spin_unlock(&dcache_lock);
712 dput(old);
713 break;
714 }
715 spin_unlock(&dcache_lock);
716 spin_lock(&vfsmount_lock);
Al Viro58c465e2005-06-06 13:36:13 -0700717 parent = nd->mnt->mnt_parent;
718 if (parent == nd->mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 spin_unlock(&vfsmount_lock);
720 break;
721 }
722 mntget(parent);
Al Viro58c465e2005-06-06 13:36:13 -0700723 nd->dentry = dget(nd->mnt->mnt_mountpoint);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 spin_unlock(&vfsmount_lock);
725 dput(old);
Al Viro58c465e2005-06-06 13:36:13 -0700726 mntput(nd->mnt);
727 nd->mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 }
Al Viro58c465e2005-06-06 13:36:13 -0700729 follow_mount(&nd->mnt, &nd->dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730}
731
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732/*
733 * It's more convoluted than I'd like it to be, but... it's still fairly
734 * small and for now I'd prefer to have fast path as straight as possible.
735 * It _is_ time-critical.
736 */
737static int do_lookup(struct nameidata *nd, struct qstr *name,
738 struct path *path)
739{
740 struct vfsmount *mnt = nd->mnt;
741 struct dentry *dentry = __d_lookup(nd->dentry, name);
742
743 if (!dentry)
744 goto need_lookup;
745 if (dentry->d_op && dentry->d_op->d_revalidate)
746 goto need_revalidate;
747done:
748 path->mnt = mnt;
749 path->dentry = dentry;
Al Viro634ee702005-06-06 13:36:13 -0700750 __follow_mount(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 return 0;
752
753need_lookup:
754 dentry = real_lookup(nd->dentry, name, nd);
755 if (IS_ERR(dentry))
756 goto fail;
757 goto done;
758
759need_revalidate:
760 if (dentry->d_op->d_revalidate(dentry, nd))
761 goto done;
762 if (d_invalidate(dentry))
763 goto done;
764 dput(dentry);
765 goto need_lookup;
766
767fail:
768 return PTR_ERR(dentry);
769}
770
771/*
772 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +0100773 * This is the basic name resolution function, turning a pathname into
774 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 *
Prasanna Medaea3834d2005-04-29 16:00:17 +0100776 * Returns 0 and nd will have valid dentry and mnt on success.
777 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 */
779static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
780{
781 struct path next;
782 struct inode *inode;
783 int err;
784 unsigned int lookup_flags = nd->flags;
785
786 while (*name=='/')
787 name++;
788 if (!*name)
789 goto return_reval;
790
791 inode = nd->dentry->d_inode;
792 if (nd->depth)
Trond Myklebustf55eab82006-02-04 23:28:01 -0800793 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794
795 /* At this point we know we have a real path component. */
796 for(;;) {
797 unsigned long hash;
798 struct qstr this;
799 unsigned int c;
800
Trond Myklebustcdce5d62005-10-18 14:20:18 -0700801 nd->flags |= LOOKUP_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 err = exec_permission_lite(inode, nd);
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800803 if (err == -EAGAIN)
804 err = vfs_permission(nd, MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 if (err)
806 break;
807
808 this.name = name;
809 c = *(const unsigned char *)name;
810
811 hash = init_name_hash();
812 do {
813 name++;
814 hash = partial_name_hash(c, hash);
815 c = *(const unsigned char *)name;
816 } while (c && (c != '/'));
817 this.len = name - (const char *) this.name;
818 this.hash = end_name_hash(hash);
819
820 /* remove trailing slashes? */
821 if (!c)
822 goto last_component;
823 while (*++name == '/');
824 if (!*name)
825 goto last_with_slashes;
826
827 /*
828 * "." and ".." are special - ".." especially so because it has
829 * to be able to know about the current root directory and
830 * parent relationships.
831 */
832 if (this.name[0] == '.') switch (this.len) {
833 default:
834 break;
835 case 2:
836 if (this.name[1] != '.')
837 break;
Al Viro58c465e2005-06-06 13:36:13 -0700838 follow_dotdot(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 inode = nd->dentry->d_inode;
840 /* fallthrough */
841 case 1:
842 continue;
843 }
844 /*
845 * See if the low-level filesystem might want
846 * to use its own hash..
847 */
848 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
849 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
850 if (err < 0)
851 break;
852 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 /* This does the actual lookups.. */
854 err = do_lookup(nd, &this, &next);
855 if (err)
856 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857
858 err = -ENOENT;
859 inode = next.dentry->d_inode;
860 if (!inode)
861 goto out_dput;
862 err = -ENOTDIR;
863 if (!inode->i_op)
864 goto out_dput;
865
866 if (inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700867 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 if (err)
869 goto return_err;
870 err = -ENOENT;
871 inode = nd->dentry->d_inode;
872 if (!inode)
873 break;
874 err = -ENOTDIR;
875 if (!inode->i_op)
876 break;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700877 } else
878 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 err = -ENOTDIR;
880 if (!inode->i_op->lookup)
881 break;
882 continue;
883 /* here ends the main loop */
884
885last_with_slashes:
886 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
887last_component:
Trond Myklebustf55eab82006-02-04 23:28:01 -0800888 /* Clear LOOKUP_CONTINUE iff it was previously unset */
889 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 if (lookup_flags & LOOKUP_PARENT)
891 goto lookup_parent;
892 if (this.name[0] == '.') switch (this.len) {
893 default:
894 break;
895 case 2:
896 if (this.name[1] != '.')
897 break;
Al Viro58c465e2005-06-06 13:36:13 -0700898 follow_dotdot(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 inode = nd->dentry->d_inode;
900 /* fallthrough */
901 case 1:
902 goto return_reval;
903 }
904 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
905 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
906 if (err < 0)
907 break;
908 }
909 err = do_lookup(nd, &this, &next);
910 if (err)
911 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 inode = next.dentry->d_inode;
913 if ((lookup_flags & LOOKUP_FOLLOW)
914 && inode && inode->i_op && inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700915 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 if (err)
917 goto return_err;
918 inode = nd->dentry->d_inode;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700919 } else
920 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 err = -ENOENT;
922 if (!inode)
923 break;
924 if (lookup_flags & LOOKUP_DIRECTORY) {
925 err = -ENOTDIR;
926 if (!inode->i_op || !inode->i_op->lookup)
927 break;
928 }
929 goto return_base;
930lookup_parent:
931 nd->last = this;
932 nd->last_type = LAST_NORM;
933 if (this.name[0] != '.')
934 goto return_base;
935 if (this.len == 1)
936 nd->last_type = LAST_DOT;
937 else if (this.len == 2 && this.name[1] == '.')
938 nd->last_type = LAST_DOTDOT;
939 else
940 goto return_base;
941return_reval:
942 /*
943 * We bypassed the ordinary revalidation routines.
944 * We may need to check the cached dentry for staleness.
945 */
946 if (nd->dentry && nd->dentry->d_sb &&
947 (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
948 err = -ESTALE;
949 /* Note: we do not d_invalidate() */
950 if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
951 break;
952 }
953return_base:
954 return 0;
955out_dput:
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700956 dput_path(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 break;
958 }
959 path_release(nd);
960return_err:
961 return err;
962}
963
964/*
965 * Wrapper to retry pathname resolution whenever the underlying
966 * file system returns an ESTALE.
967 *
968 * Retry the whole path once, forcing real lookup requests
969 * instead of relying on the dcache.
970 */
971int fastcall link_path_walk(const char *name, struct nameidata *nd)
972{
973 struct nameidata save = *nd;
974 int result;
975
976 /* make sure the stuff we saved doesn't go away */
977 dget(save.dentry);
978 mntget(save.mnt);
979
980 result = __link_path_walk(name, nd);
981 if (result == -ESTALE) {
982 *nd = save;
983 dget(nd->dentry);
984 mntget(nd->mnt);
985 nd->flags |= LOOKUP_REVAL;
986 result = __link_path_walk(name, nd);
987 }
988
989 dput(save.dentry);
990 mntput(save.mnt);
991
992 return result;
993}
994
995int fastcall path_walk(const char * name, struct nameidata *nd)
996{
997 current->total_link_count = 0;
998 return link_path_walk(name, nd);
999}
1000
Prasanna Medaea3834d2005-04-29 16:00:17 +01001001/*
1002 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
1003 * everything is done. Returns 0 and drops input nd, if lookup failed;
1004 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
1006{
1007 if (path_walk(name, nd))
1008 return 0; /* something went wrong... */
1009
1010 if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
1011 struct dentry *old_dentry = nd->dentry;
1012 struct vfsmount *old_mnt = nd->mnt;
1013 struct qstr last = nd->last;
1014 int last_type = nd->last_type;
1015 /*
1016 * NAME was not found in alternate root or it's a directory. Try to find
1017 * it in the normal root:
1018 */
1019 nd->last_type = LAST_ROOT;
1020 read_lock(&current->fs->lock);
1021 nd->mnt = mntget(current->fs->rootmnt);
1022 nd->dentry = dget(current->fs->root);
1023 read_unlock(&current->fs->lock);
1024 if (path_walk(name, nd) == 0) {
1025 if (nd->dentry->d_inode) {
1026 dput(old_dentry);
1027 mntput(old_mnt);
1028 return 1;
1029 }
1030 path_release(nd);
1031 }
1032 nd->dentry = old_dentry;
1033 nd->mnt = old_mnt;
1034 nd->last = last;
1035 nd->last_type = last_type;
1036 }
1037 return 1;
1038}
1039
1040void set_fs_altroot(void)
1041{
1042 char *emul = __emul_prefix();
1043 struct nameidata nd;
1044 struct vfsmount *mnt = NULL, *oldmnt;
1045 struct dentry *dentry = NULL, *olddentry;
1046 int err;
1047
1048 if (!emul)
1049 goto set_it;
1050 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
1051 if (!err) {
1052 mnt = nd.mnt;
1053 dentry = nd.dentry;
1054 }
1055set_it:
1056 write_lock(&current->fs->lock);
1057 oldmnt = current->fs->altrootmnt;
1058 olddentry = current->fs->altroot;
1059 current->fs->altrootmnt = mnt;
1060 current->fs->altroot = dentry;
1061 write_unlock(&current->fs->lock);
1062 if (olddentry) {
1063 dput(olddentry);
1064 mntput(oldmnt);
1065 }
1066}
1067
Prasanna Medaea3834d2005-04-29 16:00:17 +01001068/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001069static int fastcall do_path_lookup(int dfd, const char *name,
1070 unsigned int flags, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001072 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001073 int fput_needed;
1074 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075
1076 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1077 nd->flags = flags;
1078 nd->depth = 0;
1079
1080 read_lock(&current->fs->lock);
1081 if (*name=='/') {
1082 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
1083 nd->mnt = mntget(current->fs->altrootmnt);
1084 nd->dentry = dget(current->fs->altroot);
1085 read_unlock(&current->fs->lock);
1086 if (__emul_lookup_dentry(name,nd))
Prasanna Medaea3834d2005-04-29 16:00:17 +01001087 goto out; /* found in altroot */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 read_lock(&current->fs->lock);
1089 }
1090 nd->mnt = mntget(current->fs->rootmnt);
1091 nd->dentry = dget(current->fs->root);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001092 } else if (dfd == AT_FDCWD) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 nd->mnt = mntget(current->fs->pwdmnt);
1094 nd->dentry = dget(current->fs->pwd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001095 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001096 struct dentry *dentry;
1097
1098 file = fget_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001099 retval = -EBADF;
1100 if (!file)
1101 goto unlock_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001102
1103 dentry = file->f_dentry;
1104
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001105 retval = -ENOTDIR;
1106 if (!S_ISDIR(dentry->d_inode->i_mode))
1107 goto fput_unlock_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001108
1109 retval = file_permission(file, MAY_EXEC);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001110 if (retval)
1111 goto fput_unlock_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001112
1113 nd->mnt = mntget(file->f_vfsmnt);
1114 nd->dentry = dget(dentry);
1115
1116 fput_light(file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 }
1118 read_unlock(&current->fs->lock);
1119 current->total_link_count = 0;
1120 retval = link_path_walk(name, nd);
Prasanna Medaea3834d2005-04-29 16:00:17 +01001121out:
Suzuki3bc84142006-02-07 12:58:36 -08001122 if (likely(retval == 0)) {
1123 if (unlikely(current->audit_context && nd && nd->dentry &&
1124 nd->dentry->d_inode))
David Woodhouseae7b9612005-06-20 16:11:05 +01001125 audit_inode(name, nd->dentry->d_inode, flags);
Suzuki3bc84142006-02-07 12:58:36 -08001126 }
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001127 return retval;
1128
1129fput_unlock_fail:
1130 fput_light(file, fput_needed);
1131unlock_fail:
1132 read_unlock(&current->fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 return retval;
1134}
1135
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001136int fastcall path_lookup(const char *name, unsigned int flags,
1137 struct nameidata *nd)
1138{
1139 return do_path_lookup(AT_FDCWD, name, flags, nd);
1140}
1141
1142static int __path_lookup_intent_open(int dfd, const char *name,
1143 unsigned int lookup_flags, struct nameidata *nd,
1144 int open_flags, int create_mode)
Trond Myklebust834f2a42005-10-18 14:20:16 -07001145{
1146 struct file *filp = get_empty_filp();
1147 int err;
1148
1149 if (filp == NULL)
1150 return -ENFILE;
1151 nd->intent.open.file = filp;
1152 nd->intent.open.flags = open_flags;
1153 nd->intent.open.create_mode = create_mode;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001154 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
Trond Myklebust834f2a42005-10-18 14:20:16 -07001155 if (IS_ERR(nd->intent.open.file)) {
1156 if (err == 0) {
1157 err = PTR_ERR(nd->intent.open.file);
1158 path_release(nd);
1159 }
1160 } else if (err != 0)
1161 release_open_intent(nd);
1162 return err;
1163}
1164
1165/**
1166 * path_lookup_open - lookup a file path with open intent
Martin Waitz7045f372006-02-01 03:06:57 -08001167 * @dfd: the directory to use as base, or AT_FDCWD
Trond Myklebust834f2a42005-10-18 14:20:16 -07001168 * @name: pointer to file name
1169 * @lookup_flags: lookup intent flags
1170 * @nd: pointer to nameidata
1171 * @open_flags: open intent flags
1172 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001173int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
Trond Myklebust834f2a42005-10-18 14:20:16 -07001174 struct nameidata *nd, int open_flags)
1175{
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001176 return __path_lookup_intent_open(dfd, name, lookup_flags, nd,
Trond Myklebust834f2a42005-10-18 14:20:16 -07001177 open_flags, 0);
1178}
1179
1180/**
1181 * path_lookup_create - lookup a file path with open + create intent
Martin Waitz7045f372006-02-01 03:06:57 -08001182 * @dfd: the directory to use as base, or AT_FDCWD
Trond Myklebust834f2a42005-10-18 14:20:16 -07001183 * @name: pointer to file name
1184 * @lookup_flags: lookup intent flags
1185 * @nd: pointer to nameidata
1186 * @open_flags: open intent flags
1187 * @create_mode: create intent flags
1188 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001189static int path_lookup_create(int dfd, const char *name,
1190 unsigned int lookup_flags, struct nameidata *nd,
1191 int open_flags, int create_mode)
Trond Myklebust834f2a42005-10-18 14:20:16 -07001192{
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001193 return __path_lookup_intent_open(dfd, name, lookup_flags|LOOKUP_CREATE,
1194 nd, open_flags, create_mode);
Trond Myklebust834f2a42005-10-18 14:20:16 -07001195}
1196
1197int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags,
1198 struct nameidata *nd, int open_flags)
1199{
1200 char *tmp = getname(name);
1201 int err = PTR_ERR(tmp);
1202
1203 if (!IS_ERR(tmp)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001204 err = __path_lookup_intent_open(AT_FDCWD, tmp, lookup_flags, nd, open_flags, 0);
Trond Myklebust834f2a42005-10-18 14:20:16 -07001205 putname(tmp);
1206 }
1207 return err;
1208}
1209
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210/*
1211 * Restricted form of lookup. Doesn't follow links, single-component only,
1212 * needs parent already locked. Doesn't follow mounts.
1213 * SMP-safe.
1214 */
1215static struct dentry * __lookup_hash(struct qstr *name, struct dentry * base, struct nameidata *nd)
1216{
1217 struct dentry * dentry;
1218 struct inode *inode;
1219 int err;
1220
1221 inode = base->d_inode;
1222 err = permission(inode, MAY_EXEC, nd);
1223 dentry = ERR_PTR(err);
1224 if (err)
1225 goto out;
1226
1227 /*
1228 * See if the low-level filesystem might want
1229 * to use its own hash..
1230 */
1231 if (base->d_op && base->d_op->d_hash) {
1232 err = base->d_op->d_hash(base, name);
1233 dentry = ERR_PTR(err);
1234 if (err < 0)
1235 goto out;
1236 }
1237
1238 dentry = cached_lookup(base, name, nd);
1239 if (!dentry) {
1240 struct dentry *new = d_alloc(base, name);
1241 dentry = ERR_PTR(-ENOMEM);
1242 if (!new)
1243 goto out;
1244 dentry = inode->i_op->lookup(inode, new, nd);
1245 if (!dentry)
1246 dentry = new;
1247 else
1248 dput(new);
1249 }
1250out:
1251 return dentry;
1252}
1253
Christoph Hellwig49705b72005-11-08 21:35:06 -08001254struct dentry * lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255{
Christoph Hellwig49705b72005-11-08 21:35:06 -08001256 return __lookup_hash(&nd->last, nd->dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257}
1258
1259/* SMP-safe */
1260struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
1261{
1262 unsigned long hash;
1263 struct qstr this;
1264 unsigned int c;
1265
1266 this.name = name;
1267 this.len = len;
1268 if (!len)
1269 goto access;
1270
1271 hash = init_name_hash();
1272 while (len--) {
1273 c = *(const unsigned char *)name++;
1274 if (c == '/' || c == '\0')
1275 goto access;
1276 hash = partial_name_hash(c, hash);
1277 }
1278 this.hash = end_name_hash(hash);
1279
Christoph Hellwig49705b72005-11-08 21:35:06 -08001280 return __lookup_hash(&this, base, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281access:
1282 return ERR_PTR(-EACCES);
1283}
1284
1285/*
1286 * namei()
1287 *
1288 * is used by most simple commands to get the inode of a specified name.
1289 * Open, link etc use their own routines, but this is enough for things
1290 * like 'chmod' etc.
1291 *
1292 * namei exists in two versions: namei/lnamei. The only difference is
1293 * that namei follows links, while lnamei does not.
1294 * SMP-safe
1295 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001296int fastcall __user_walk_fd(int dfd, const char __user *name, unsigned flags,
1297 struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298{
1299 char *tmp = getname(name);
1300 int err = PTR_ERR(tmp);
1301
1302 if (!IS_ERR(tmp)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001303 err = do_path_lookup(dfd, tmp, flags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 putname(tmp);
1305 }
1306 return err;
1307}
1308
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001309int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1310{
1311 return __user_walk_fd(AT_FDCWD, name, flags, nd);
1312}
1313
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314/*
1315 * It's inline, so penalty for filesystems that don't use sticky bit is
1316 * minimal.
1317 */
1318static inline int check_sticky(struct inode *dir, struct inode *inode)
1319{
1320 if (!(dir->i_mode & S_ISVTX))
1321 return 0;
1322 if (inode->i_uid == current->fsuid)
1323 return 0;
1324 if (dir->i_uid == current->fsuid)
1325 return 0;
1326 return !capable(CAP_FOWNER);
1327}
1328
1329/*
1330 * Check whether we can remove a link victim from directory dir, check
1331 * whether the type of victim is right.
1332 * 1. We can't do it if dir is read-only (done in permission())
1333 * 2. We should have write and exec permissions on dir
1334 * 3. We can't remove anything from append-only dir
1335 * 4. We can't do anything with immutable dir (done in permission())
1336 * 5. If the sticky bit on dir is set we should either
1337 * a. be owner of dir, or
1338 * b. be owner of victim, or
1339 * c. have CAP_FOWNER capability
1340 * 6. If the victim is append-only or immutable we can't do antyhing with
1341 * links pointing to it.
1342 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1343 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1344 * 9. We can't remove a root or mountpoint.
1345 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1346 * nfs_async_unlink().
1347 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001348static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349{
1350 int error;
1351
1352 if (!victim->d_inode)
1353 return -ENOENT;
1354
1355 BUG_ON(victim->d_parent->d_inode != dir);
1356
1357 error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
1358 if (error)
1359 return error;
1360 if (IS_APPEND(dir))
1361 return -EPERM;
1362 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1363 IS_IMMUTABLE(victim->d_inode))
1364 return -EPERM;
1365 if (isdir) {
1366 if (!S_ISDIR(victim->d_inode->i_mode))
1367 return -ENOTDIR;
1368 if (IS_ROOT(victim))
1369 return -EBUSY;
1370 } else if (S_ISDIR(victim->d_inode->i_mode))
1371 return -EISDIR;
1372 if (IS_DEADDIR(dir))
1373 return -ENOENT;
1374 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1375 return -EBUSY;
1376 return 0;
1377}
1378
1379/* Check whether we can create an object with dentry child in directory
1380 * dir.
1381 * 1. We can't do it if child already exists (open has special treatment for
1382 * this case, but since we are inlined it's OK)
1383 * 2. We can't do it if dir is read-only (done in permission())
1384 * 3. We should have write and exec permissions on dir
1385 * 4. We can't do it if dir is immutable (done in permission())
1386 */
1387static inline int may_create(struct inode *dir, struct dentry *child,
1388 struct nameidata *nd)
1389{
1390 if (child->d_inode)
1391 return -EEXIST;
1392 if (IS_DEADDIR(dir))
1393 return -ENOENT;
1394 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1395}
1396
1397/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 * O_DIRECTORY translates into forcing a directory lookup.
1399 */
1400static inline int lookup_flags(unsigned int f)
1401{
1402 unsigned long retval = LOOKUP_FOLLOW;
1403
1404 if (f & O_NOFOLLOW)
1405 retval &= ~LOOKUP_FOLLOW;
1406
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 if (f & O_DIRECTORY)
1408 retval |= LOOKUP_DIRECTORY;
1409
1410 return retval;
1411}
1412
1413/*
1414 * p1 and p2 should be directories on the same fs.
1415 */
1416struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1417{
1418 struct dentry *p;
1419
1420 if (p1 == p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001421 mutex_lock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 return NULL;
1423 }
1424
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001425 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426
1427 for (p = p1; p->d_parent != p; p = p->d_parent) {
1428 if (p->d_parent == p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001429 mutex_lock(&p2->d_inode->i_mutex);
1430 mutex_lock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 return p;
1432 }
1433 }
1434
1435 for (p = p2; p->d_parent != p; p = p->d_parent) {
1436 if (p->d_parent == p1) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001437 mutex_lock(&p1->d_inode->i_mutex);
1438 mutex_lock(&p2->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 return p;
1440 }
1441 }
1442
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001443 mutex_lock(&p1->d_inode->i_mutex);
1444 mutex_lock(&p2->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 return NULL;
1446}
1447
1448void unlock_rename(struct dentry *p1, struct dentry *p2)
1449{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001450 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001452 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001453 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 }
1455}
1456
1457int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1458 struct nameidata *nd)
1459{
1460 int error = may_create(dir, dentry, nd);
1461
1462 if (error)
1463 return error;
1464
1465 if (!dir->i_op || !dir->i_op->create)
1466 return -EACCES; /* shouldn't it be ENOSYS? */
1467 mode &= S_IALLUGO;
1468 mode |= S_IFREG;
1469 error = security_inode_create(dir, dentry, mode);
1470 if (error)
1471 return error;
1472 DQUOT_INIT(dir);
1473 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001474 if (!error)
Robert Love0eeca282005-07-12 17:06:03 -04001475 fsnotify_create(dir, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 return error;
1477}
1478
1479int may_open(struct nameidata *nd, int acc_mode, int flag)
1480{
1481 struct dentry *dentry = nd->dentry;
1482 struct inode *inode = dentry->d_inode;
1483 int error;
1484
1485 if (!inode)
1486 return -ENOENT;
1487
1488 if (S_ISLNK(inode->i_mode))
1489 return -ELOOP;
1490
1491 if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1492 return -EISDIR;
1493
Christoph Hellwige4543ed2005-11-08 21:35:04 -08001494 error = vfs_permission(nd, acc_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 if (error)
1496 return error;
1497
1498 /*
1499 * FIFO's, sockets and device files are special: they don't
1500 * actually live on the filesystem itself, and as such you
1501 * can write to them even if the filesystem is read-only.
1502 */
1503 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1504 flag &= ~O_TRUNC;
1505 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1506 if (nd->mnt->mnt_flags & MNT_NODEV)
1507 return -EACCES;
1508
1509 flag &= ~O_TRUNC;
1510 } else if (IS_RDONLY(inode) && (flag & FMODE_WRITE))
1511 return -EROFS;
1512 /*
1513 * An append-only file must be opened in append mode for writing.
1514 */
1515 if (IS_APPEND(inode)) {
1516 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1517 return -EPERM;
1518 if (flag & O_TRUNC)
1519 return -EPERM;
1520 }
1521
1522 /* O_NOATIME can only be set by the owner or superuser */
1523 if (flag & O_NOATIME)
1524 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1525 return -EPERM;
1526
1527 /*
1528 * Ensure there are no outstanding leases on the file.
1529 */
1530 error = break_lease(inode, flag);
1531 if (error)
1532 return error;
1533
1534 if (flag & O_TRUNC) {
1535 error = get_write_access(inode);
1536 if (error)
1537 return error;
1538
1539 /*
1540 * Refuse to truncate files with mandatory locks held on them.
1541 */
1542 error = locks_verify_locked(inode);
1543 if (!error) {
1544 DQUOT_INIT(inode);
1545
NeilBrown4a301312006-01-08 01:02:39 -08001546 error = do_truncate(dentry, 0, ATTR_MTIME|ATTR_CTIME, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 }
1548 put_write_access(inode);
1549 if (error)
1550 return error;
1551 } else
1552 if (flag & FMODE_WRITE)
1553 DQUOT_INIT(inode);
1554
1555 return 0;
1556}
1557
1558/*
1559 * open_namei()
1560 *
1561 * namei for open - this is in fact almost the whole open-routine.
1562 *
1563 * Note that the low bits of "flag" aren't the same as in the open
1564 * system call - they are 00 - no permissions needed
1565 * 01 - read permission needed
1566 * 10 - write permission needed
1567 * 11 - read/write permissions needed
1568 * which is a lot more logical, and also allows the "no perm" needed
1569 * for symlinks (where the permissions are checked later).
1570 * SMP-safe
1571 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001572int open_namei(int dfd, const char *pathname, int flag,
1573 int mode, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574{
Trond Myklebust834f2a42005-10-18 14:20:16 -07001575 int acc_mode, error;
Al Viro4e7506e2005-06-06 13:36:00 -07001576 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 struct dentry *dir;
1578 int count = 0;
1579
1580 acc_mode = ACC_MODE(flag);
1581
Trond Myklebust834f2a42005-10-18 14:20:16 -07001582 /* O_TRUNC implies we need access checks for write permissions */
1583 if (flag & O_TRUNC)
1584 acc_mode |= MAY_WRITE;
1585
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 /* Allow the LSM permission hook to distinguish append
1587 access from general write access. */
1588 if (flag & O_APPEND)
1589 acc_mode |= MAY_APPEND;
1590
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 /*
1592 * The simplest case - just a plain lookup.
1593 */
1594 if (!(flag & O_CREAT)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001595 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1596 nd, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597 if (error)
1598 return error;
1599 goto ok;
1600 }
1601
1602 /*
1603 * Create - we need to know the parent.
1604 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001605 error = path_lookup_create(dfd,pathname,LOOKUP_PARENT,nd,flag,mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 if (error)
1607 return error;
1608
1609 /*
1610 * We have the parent and last component. First of all, check
1611 * that we are not asked to creat(2) an obvious directory - that
1612 * will not do.
1613 */
1614 error = -EISDIR;
1615 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1616 goto exit;
1617
1618 dir = nd->dentry;
1619 nd->flags &= ~LOOKUP_PARENT;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001620 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001621 path.dentry = lookup_hash(nd);
Al Virod73ffe12005-06-06 13:36:01 -07001622 path.mnt = nd->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623
1624do_last:
Al Viro4e7506e2005-06-06 13:36:00 -07001625 error = PTR_ERR(path.dentry);
1626 if (IS_ERR(path.dentry)) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001627 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 goto exit;
1629 }
1630
1631 /* Negative dentry, just create the file */
Al Viro4e7506e2005-06-06 13:36:00 -07001632 if (!path.dentry->d_inode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 if (!IS_POSIXACL(dir->d_inode))
1634 mode &= ~current->fs->umask;
Al Viro4e7506e2005-06-06 13:36:00 -07001635 error = vfs_create(dir->d_inode, path.dentry, mode, nd);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001636 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 dput(nd->dentry);
Al Viro4e7506e2005-06-06 13:36:00 -07001638 nd->dentry = path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 if (error)
1640 goto exit;
1641 /* Don't check for write permission, don't truncate */
1642 acc_mode = 0;
1643 flag &= ~O_TRUNC;
1644 goto ok;
1645 }
1646
1647 /*
1648 * It already exists.
1649 */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001650 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651
1652 error = -EEXIST;
1653 if (flag & O_EXCL)
1654 goto exit_dput;
1655
Al Viroe13b2102005-06-06 13:36:06 -07001656 if (__follow_mount(&path)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 error = -ELOOP;
Al Viroba7a4c12005-06-06 13:36:08 -07001658 if (flag & O_NOFOLLOW)
1659 goto exit_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 }
1661 error = -ENOENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001662 if (!path.dentry->d_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 goto exit_dput;
Al Viro4e7506e2005-06-06 13:36:00 -07001664 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 goto do_link;
1666
Miklos Szeredi09dd17d2005-09-06 15:18:21 -07001667 path_to_nameidata(&path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 error = -EISDIR;
Al Viro4e7506e2005-06-06 13:36:00 -07001669 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 goto exit;
1671ok:
1672 error = may_open(nd, acc_mode, flag);
1673 if (error)
1674 goto exit;
1675 return 0;
1676
1677exit_dput:
Miklos Szeredi09dd17d2005-09-06 15:18:21 -07001678 dput_path(&path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679exit:
Trond Myklebust834f2a42005-10-18 14:20:16 -07001680 if (!IS_ERR(nd->intent.open.file))
1681 release_open_intent(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 path_release(nd);
1683 return error;
1684
1685do_link:
1686 error = -ELOOP;
1687 if (flag & O_NOFOLLOW)
1688 goto exit_dput;
1689 /*
1690 * This is subtle. Instead of calling do_follow_link() we do the
1691 * thing by hands. The reason is that this way we have zero link_count
1692 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1693 * After that we have the parent and last component, i.e.
1694 * we are in the same situation as after the first path_walk().
1695 * Well, almost - if the last component is normal we get its copy
1696 * stored in nd->last.name and we will have to putname() it when we
1697 * are done. Procfs-like symlinks just set LAST_BIND.
1698 */
1699 nd->flags |= LOOKUP_PARENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001700 error = security_inode_follow_link(path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 if (error)
1702 goto exit_dput;
Al Virocd4e91d2005-06-06 13:36:03 -07001703 error = __do_follow_link(&path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 if (error)
1705 return error;
1706 nd->flags &= ~LOOKUP_PARENT;
Al Virod671d5e2005-06-06 13:36:04 -07001707 if (nd->last_type == LAST_BIND)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 goto ok;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 error = -EISDIR;
1710 if (nd->last_type != LAST_NORM)
1711 goto exit;
1712 if (nd->last.name[nd->last.len]) {
Linus Torvalds82984112005-10-06 21:54:21 -07001713 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 goto exit;
1715 }
1716 error = -ELOOP;
1717 if (count++==32) {
Linus Torvalds82984112005-10-06 21:54:21 -07001718 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 goto exit;
1720 }
1721 dir = nd->dentry;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001722 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001723 path.dentry = lookup_hash(nd);
Al Virod671d5e2005-06-06 13:36:04 -07001724 path.mnt = nd->mnt;
Linus Torvalds82984112005-10-06 21:54:21 -07001725 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 goto do_last;
1727}
1728
1729/**
1730 * lookup_create - lookup a dentry, creating it if it doesn't exist
1731 * @nd: nameidata info
1732 * @is_dir: directory flag
1733 *
1734 * Simple function to lookup and return a dentry and create it
1735 * if it doesn't exist. Is SMP-safe.
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001736 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001737 * Returns with nd->dentry->d_inode->i_mutex locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738 */
1739struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1740{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001741 struct dentry *dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001743 mutex_lock(&nd->dentry->d_inode->i_mutex);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001744 /*
1745 * Yucky last component or no last component at all?
1746 * (foo/., foo/.., /////)
1747 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 if (nd->last_type != LAST_NORM)
1749 goto fail;
1750 nd->flags &= ~LOOKUP_PARENT;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001751
1752 /*
1753 * Do the final lookup.
1754 */
Christoph Hellwig49705b72005-11-08 21:35:06 -08001755 dentry = lookup_hash(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 if (IS_ERR(dentry))
1757 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001758
1759 /*
1760 * Special case - lookup gave negative, but... we had foo/bar/
1761 * From the vfs_mknod() POV we just have a negative dentry -
1762 * all is fine. Let's be bastards - you had / on the end, you've
1763 * been asking for (non-existent) directory. -ENOENT for you.
1764 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1766 goto enoent;
1767 return dentry;
1768enoent:
1769 dput(dentry);
1770 dentry = ERR_PTR(-ENOENT);
1771fail:
1772 return dentry;
1773}
Christoph Hellwigf81a0bf2005-05-19 12:26:43 -07001774EXPORT_SYMBOL_GPL(lookup_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775
1776int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1777{
1778 int error = may_create(dir, dentry, NULL);
1779
1780 if (error)
1781 return error;
1782
1783 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1784 return -EPERM;
1785
1786 if (!dir->i_op || !dir->i_op->mknod)
1787 return -EPERM;
1788
1789 error = security_inode_mknod(dir, dentry, mode, dev);
1790 if (error)
1791 return error;
1792
1793 DQUOT_INIT(dir);
1794 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001795 if (!error)
Robert Love0eeca282005-07-12 17:06:03 -04001796 fsnotify_create(dir, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797 return error;
1798}
1799
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001800asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
1801 unsigned dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802{
1803 int error = 0;
1804 char * tmp;
1805 struct dentry * dentry;
1806 struct nameidata nd;
1807
1808 if (S_ISDIR(mode))
1809 return -EPERM;
1810 tmp = getname(filename);
1811 if (IS_ERR(tmp))
1812 return PTR_ERR(tmp);
1813
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001814 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 if (error)
1816 goto out;
1817 dentry = lookup_create(&nd, 0);
1818 error = PTR_ERR(dentry);
1819
1820 if (!IS_POSIXACL(nd.dentry->d_inode))
1821 mode &= ~current->fs->umask;
1822 if (!IS_ERR(dentry)) {
1823 switch (mode & S_IFMT) {
1824 case 0: case S_IFREG:
1825 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1826 break;
1827 case S_IFCHR: case S_IFBLK:
1828 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,
1829 new_decode_dev(dev));
1830 break;
1831 case S_IFIFO: case S_IFSOCK:
1832 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,0);
1833 break;
1834 case S_IFDIR:
1835 error = -EPERM;
1836 break;
1837 default:
1838 error = -EINVAL;
1839 }
1840 dput(dentry);
1841 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001842 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843 path_release(&nd);
1844out:
1845 putname(tmp);
1846
1847 return error;
1848}
1849
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001850asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
1851{
1852 return sys_mknodat(AT_FDCWD, filename, mode, dev);
1853}
1854
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1856{
1857 int error = may_create(dir, dentry, NULL);
1858
1859 if (error)
1860 return error;
1861
1862 if (!dir->i_op || !dir->i_op->mkdir)
1863 return -EPERM;
1864
1865 mode &= (S_IRWXUGO|S_ISVTX);
1866 error = security_inode_mkdir(dir, dentry, mode);
1867 if (error)
1868 return error;
1869
1870 DQUOT_INIT(dir);
1871 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001872 if (!error)
Robert Love0eeca282005-07-12 17:06:03 -04001873 fsnotify_mkdir(dir, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 return error;
1875}
1876
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001877asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878{
1879 int error = 0;
1880 char * tmp;
1881
1882 tmp = getname(pathname);
1883 error = PTR_ERR(tmp);
1884 if (!IS_ERR(tmp)) {
1885 struct dentry *dentry;
1886 struct nameidata nd;
1887
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001888 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 if (error)
1890 goto out;
1891 dentry = lookup_create(&nd, 1);
1892 error = PTR_ERR(dentry);
1893 if (!IS_ERR(dentry)) {
1894 if (!IS_POSIXACL(nd.dentry->d_inode))
1895 mode &= ~current->fs->umask;
1896 error = vfs_mkdir(nd.dentry->d_inode, dentry, mode);
1897 dput(dentry);
1898 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001899 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900 path_release(&nd);
1901out:
1902 putname(tmp);
1903 }
1904
1905 return error;
1906}
1907
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001908asmlinkage long sys_mkdir(const char __user *pathname, int mode)
1909{
1910 return sys_mkdirat(AT_FDCWD, pathname, mode);
1911}
1912
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913/*
1914 * We try to drop the dentry early: we should have
1915 * a usage count of 2 if we're the only user of this
1916 * dentry, and if that is true (possibly after pruning
1917 * the dcache), then we drop the dentry now.
1918 *
1919 * A low-level filesystem can, if it choses, legally
1920 * do a
1921 *
1922 * if (!d_unhashed(dentry))
1923 * return -EBUSY;
1924 *
1925 * if it cannot handle the case of removing a directory
1926 * that is still in use by something else..
1927 */
1928void dentry_unhash(struct dentry *dentry)
1929{
1930 dget(dentry);
1931 if (atomic_read(&dentry->d_count))
1932 shrink_dcache_parent(dentry);
1933 spin_lock(&dcache_lock);
1934 spin_lock(&dentry->d_lock);
1935 if (atomic_read(&dentry->d_count) == 2)
1936 __d_drop(dentry);
1937 spin_unlock(&dentry->d_lock);
1938 spin_unlock(&dcache_lock);
1939}
1940
1941int vfs_rmdir(struct inode *dir, struct dentry *dentry)
1942{
1943 int error = may_delete(dir, dentry, 1);
1944
1945 if (error)
1946 return error;
1947
1948 if (!dir->i_op || !dir->i_op->rmdir)
1949 return -EPERM;
1950
1951 DQUOT_INIT(dir);
1952
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001953 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 dentry_unhash(dentry);
1955 if (d_mountpoint(dentry))
1956 error = -EBUSY;
1957 else {
1958 error = security_inode_rmdir(dir, dentry);
1959 if (!error) {
1960 error = dir->i_op->rmdir(dir, dentry);
1961 if (!error)
1962 dentry->d_inode->i_flags |= S_DEAD;
1963 }
1964 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001965 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 d_delete(dentry);
1968 }
1969 dput(dentry);
1970
1971 return error;
1972}
1973
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001974static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975{
1976 int error = 0;
1977 char * name;
1978 struct dentry *dentry;
1979 struct nameidata nd;
1980
1981 name = getname(pathname);
1982 if(IS_ERR(name))
1983 return PTR_ERR(name);
1984
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001985 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 if (error)
1987 goto exit;
1988
1989 switch(nd.last_type) {
1990 case LAST_DOTDOT:
1991 error = -ENOTEMPTY;
1992 goto exit1;
1993 case LAST_DOT:
1994 error = -EINVAL;
1995 goto exit1;
1996 case LAST_ROOT:
1997 error = -EBUSY;
1998 goto exit1;
1999 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002000 mutex_lock(&nd.dentry->d_inode->i_mutex);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002001 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 error = PTR_ERR(dentry);
2003 if (!IS_ERR(dentry)) {
2004 error = vfs_rmdir(nd.dentry->d_inode, dentry);
2005 dput(dentry);
2006 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002007 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008exit1:
2009 path_release(&nd);
2010exit:
2011 putname(name);
2012 return error;
2013}
2014
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002015asmlinkage long sys_rmdir(const char __user *pathname)
2016{
2017 return do_rmdir(AT_FDCWD, pathname);
2018}
2019
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020int vfs_unlink(struct inode *dir, struct dentry *dentry)
2021{
2022 int error = may_delete(dir, dentry, 0);
2023
2024 if (error)
2025 return error;
2026
2027 if (!dir->i_op || !dir->i_op->unlink)
2028 return -EPERM;
2029
2030 DQUOT_INIT(dir);
2031
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002032 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 if (d_mountpoint(dentry))
2034 error = -EBUSY;
2035 else {
2036 error = security_inode_unlink(dir, dentry);
2037 if (!error)
2038 error = dir->i_op->unlink(dir, dentry);
2039 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002040 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041
2042 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2043 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2044 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 }
Robert Love0eeca282005-07-12 17:06:03 -04002046
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 return error;
2048}
2049
2050/*
2051 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002052 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 * writeout happening, and we don't want to prevent access to the directory
2054 * while waiting on the I/O.
2055 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002056static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057{
2058 int error = 0;
2059 char * name;
2060 struct dentry *dentry;
2061 struct nameidata nd;
2062 struct inode *inode = NULL;
2063
2064 name = getname(pathname);
2065 if(IS_ERR(name))
2066 return PTR_ERR(name);
2067
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002068 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069 if (error)
2070 goto exit;
2071 error = -EISDIR;
2072 if (nd.last_type != LAST_NORM)
2073 goto exit1;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002074 mutex_lock(&nd.dentry->d_inode->i_mutex);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002075 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 error = PTR_ERR(dentry);
2077 if (!IS_ERR(dentry)) {
2078 /* Why not before? Because we want correct error value */
2079 if (nd.last.name[nd.last.len])
2080 goto slashes;
2081 inode = dentry->d_inode;
2082 if (inode)
2083 atomic_inc(&inode->i_count);
2084 error = vfs_unlink(nd.dentry->d_inode, dentry);
2085 exit2:
2086 dput(dentry);
2087 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002088 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089 if (inode)
2090 iput(inode); /* truncate the inode here */
2091exit1:
2092 path_release(&nd);
2093exit:
2094 putname(name);
2095 return error;
2096
2097slashes:
2098 error = !dentry->d_inode ? -ENOENT :
2099 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2100 goto exit2;
2101}
2102
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002103asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2104{
2105 if ((flag & ~AT_REMOVEDIR) != 0)
2106 return -EINVAL;
2107
2108 if (flag & AT_REMOVEDIR)
2109 return do_rmdir(dfd, pathname);
2110
2111 return do_unlinkat(dfd, pathname);
2112}
2113
2114asmlinkage long sys_unlink(const char __user *pathname)
2115{
2116 return do_unlinkat(AT_FDCWD, pathname);
2117}
2118
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname, int mode)
2120{
2121 int error = may_create(dir, dentry, NULL);
2122
2123 if (error)
2124 return error;
2125
2126 if (!dir->i_op || !dir->i_op->symlink)
2127 return -EPERM;
2128
2129 error = security_inode_symlink(dir, dentry, oldname);
2130 if (error)
2131 return error;
2132
2133 DQUOT_INIT(dir);
2134 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002135 if (!error)
Robert Love0eeca282005-07-12 17:06:03 -04002136 fsnotify_create(dir, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 return error;
2138}
2139
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002140asmlinkage long sys_symlinkat(const char __user *oldname,
2141 int newdfd, const char __user *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142{
2143 int error = 0;
2144 char * from;
2145 char * to;
2146
2147 from = getname(oldname);
2148 if(IS_ERR(from))
2149 return PTR_ERR(from);
2150 to = getname(newname);
2151 error = PTR_ERR(to);
2152 if (!IS_ERR(to)) {
2153 struct dentry *dentry;
2154 struct nameidata nd;
2155
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002156 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157 if (error)
2158 goto out;
2159 dentry = lookup_create(&nd, 0);
2160 error = PTR_ERR(dentry);
2161 if (!IS_ERR(dentry)) {
2162 error = vfs_symlink(nd.dentry->d_inode, dentry, from, S_IALLUGO);
2163 dput(dentry);
2164 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002165 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166 path_release(&nd);
2167out:
2168 putname(to);
2169 }
2170 putname(from);
2171 return error;
2172}
2173
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002174asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2175{
2176 return sys_symlinkat(oldname, AT_FDCWD, newname);
2177}
2178
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2180{
2181 struct inode *inode = old_dentry->d_inode;
2182 int error;
2183
2184 if (!inode)
2185 return -ENOENT;
2186
2187 error = may_create(dir, new_dentry, NULL);
2188 if (error)
2189 return error;
2190
2191 if (dir->i_sb != inode->i_sb)
2192 return -EXDEV;
2193
2194 /*
2195 * A link to an append-only or immutable file cannot be created.
2196 */
2197 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2198 return -EPERM;
2199 if (!dir->i_op || !dir->i_op->link)
2200 return -EPERM;
2201 if (S_ISDIR(old_dentry->d_inode->i_mode))
2202 return -EPERM;
2203
2204 error = security_inode_link(old_dentry, dir, new_dentry);
2205 if (error)
2206 return error;
2207
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002208 mutex_lock(&old_dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 DQUOT_INIT(dir);
2210 error = dir->i_op->link(old_dentry, dir, new_dentry);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002211 mutex_unlock(&old_dentry->d_inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002212 if (!error)
Robert Love0eeca282005-07-12 17:06:03 -04002213 fsnotify_create(dir, new_dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 return error;
2215}
2216
2217/*
2218 * Hardlinks are often used in delicate situations. We avoid
2219 * security-related surprises by not following symlinks on the
2220 * newname. --KAB
2221 *
2222 * We don't follow them on the oldname either to be compatible
2223 * with linux 2.0, and to avoid hard-linking to directories
2224 * and other special files. --ADM
2225 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002226asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002227 int newdfd, const char __user *newname,
2228 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229{
2230 struct dentry *new_dentry;
2231 struct nameidata nd, old_nd;
2232 int error;
2233 char * to;
2234
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002235 if (flags != 0)
2236 return -EINVAL;
2237
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238 to = getname(newname);
2239 if (IS_ERR(to))
2240 return PTR_ERR(to);
2241
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002242 error = __user_walk_fd(olddfd, oldname, 0, &old_nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243 if (error)
2244 goto exit;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002245 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 if (error)
2247 goto out;
2248 error = -EXDEV;
2249 if (old_nd.mnt != nd.mnt)
2250 goto out_release;
2251 new_dentry = lookup_create(&nd, 0);
2252 error = PTR_ERR(new_dentry);
2253 if (!IS_ERR(new_dentry)) {
2254 error = vfs_link(old_nd.dentry, nd.dentry->d_inode, new_dentry);
2255 dput(new_dentry);
2256 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002257 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258out_release:
2259 path_release(&nd);
2260out:
2261 path_release(&old_nd);
2262exit:
2263 putname(to);
2264
2265 return error;
2266}
2267
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002268asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2269{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002270 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002271}
2272
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273/*
2274 * The worst of all namespace operations - renaming directory. "Perverted"
2275 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2276 * Problems:
2277 * a) we can get into loop creation. Check is done in is_subdir().
2278 * b) race potential - two innocent renames can create a loop together.
2279 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002280 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 * story.
2282 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002283 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284 * whether the target exists). Solution: try to be smart with locking
2285 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002286 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 * move will be locked. Thus we can rank directories by the tree
2288 * (ancestors first) and rank all non-directories after them.
2289 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002290 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291 * HOWEVER, it relies on the assumption that any object with ->lookup()
2292 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2293 * we'd better make sure that there's no link(2) for them.
2294 * d) some filesystems don't support opened-but-unlinked directories,
2295 * either because of layout or because they are not ready to deal with
2296 * all cases correctly. The latter will be fixed (taking this sort of
2297 * stuff into VFS), but the former is not going away. Solution: the same
2298 * trick as in rmdir().
2299 * e) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002300 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002302 * ->i_mutex on parents, which works but leads to some truely excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303 * locking].
2304 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07002305static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2306 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307{
2308 int error = 0;
2309 struct inode *target;
2310
2311 /*
2312 * If we are going to change the parent - check write permissions,
2313 * we'll need to flip '..'.
2314 */
2315 if (new_dir != old_dir) {
2316 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2317 if (error)
2318 return error;
2319 }
2320
2321 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2322 if (error)
2323 return error;
2324
2325 target = new_dentry->d_inode;
2326 if (target) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002327 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328 dentry_unhash(new_dentry);
2329 }
2330 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2331 error = -EBUSY;
2332 else
2333 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2334 if (target) {
2335 if (!error)
2336 target->i_flags |= S_DEAD;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002337 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338 if (d_unhashed(new_dentry))
2339 d_rehash(new_dentry);
2340 dput(new_dentry);
2341 }
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002342 if (!error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344 return error;
2345}
2346
Adrian Bunk75c96f82005-05-05 16:16:09 -07002347static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2348 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349{
2350 struct inode *target;
2351 int error;
2352
2353 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2354 if (error)
2355 return error;
2356
2357 dget(new_dentry);
2358 target = new_dentry->d_inode;
2359 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002360 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2362 error = -EBUSY;
2363 else
2364 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2365 if (!error) {
2366 /* The following d_move() should become unconditional */
2367 if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
2368 d_move(old_dentry, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369 }
2370 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002371 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372 dput(new_dentry);
2373 return error;
2374}
2375
2376int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2377 struct inode *new_dir, struct dentry *new_dentry)
2378{
2379 int error;
2380 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Robert Love0eeca282005-07-12 17:06:03 -04002381 const char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382
2383 if (old_dentry->d_inode == new_dentry->d_inode)
2384 return 0;
2385
2386 error = may_delete(old_dir, old_dentry, is_dir);
2387 if (error)
2388 return error;
2389
2390 if (!new_dentry->d_inode)
2391 error = may_create(new_dir, new_dentry, NULL);
2392 else
2393 error = may_delete(new_dir, new_dentry, is_dir);
2394 if (error)
2395 return error;
2396
2397 if (!old_dir->i_op || !old_dir->i_op->rename)
2398 return -EPERM;
2399
2400 DQUOT_INIT(old_dir);
2401 DQUOT_INIT(new_dir);
2402
Robert Love0eeca282005-07-12 17:06:03 -04002403 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2404
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405 if (is_dir)
2406 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2407 else
2408 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2409 if (!error) {
Robert Love0eeca282005-07-12 17:06:03 -04002410 const char *new_name = old_dentry->d_name.name;
John McCutchan89204c42005-08-15 12:13:28 -04002411 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2412 new_dentry->d_inode, old_dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 }
Robert Love0eeca282005-07-12 17:06:03 -04002414 fsnotify_oldname_free(old_name);
2415
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416 return error;
2417}
2418
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002419static int do_rename(int olddfd, const char *oldname,
2420 int newdfd, const char *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421{
2422 int error = 0;
2423 struct dentry * old_dir, * new_dir;
2424 struct dentry * old_dentry, *new_dentry;
2425 struct dentry * trap;
2426 struct nameidata oldnd, newnd;
2427
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002428 error = do_path_lookup(olddfd, oldname, LOOKUP_PARENT, &oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429 if (error)
2430 goto exit;
2431
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002432 error = do_path_lookup(newdfd, newname, LOOKUP_PARENT, &newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433 if (error)
2434 goto exit1;
2435
2436 error = -EXDEV;
2437 if (oldnd.mnt != newnd.mnt)
2438 goto exit2;
2439
2440 old_dir = oldnd.dentry;
2441 error = -EBUSY;
2442 if (oldnd.last_type != LAST_NORM)
2443 goto exit2;
2444
2445 new_dir = newnd.dentry;
2446 if (newnd.last_type != LAST_NORM)
2447 goto exit2;
2448
2449 trap = lock_rename(new_dir, old_dir);
2450
Christoph Hellwig49705b72005-11-08 21:35:06 -08002451 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452 error = PTR_ERR(old_dentry);
2453 if (IS_ERR(old_dentry))
2454 goto exit3;
2455 /* source must exist */
2456 error = -ENOENT;
2457 if (!old_dentry->d_inode)
2458 goto exit4;
2459 /* unless the source is a directory trailing slashes give -ENOTDIR */
2460 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2461 error = -ENOTDIR;
2462 if (oldnd.last.name[oldnd.last.len])
2463 goto exit4;
2464 if (newnd.last.name[newnd.last.len])
2465 goto exit4;
2466 }
2467 /* source should not be ancestor of target */
2468 error = -EINVAL;
2469 if (old_dentry == trap)
2470 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08002471 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472 error = PTR_ERR(new_dentry);
2473 if (IS_ERR(new_dentry))
2474 goto exit4;
2475 /* target should not be an ancestor of source */
2476 error = -ENOTEMPTY;
2477 if (new_dentry == trap)
2478 goto exit5;
2479
2480 error = vfs_rename(old_dir->d_inode, old_dentry,
2481 new_dir->d_inode, new_dentry);
2482exit5:
2483 dput(new_dentry);
2484exit4:
2485 dput(old_dentry);
2486exit3:
2487 unlock_rename(new_dir, old_dir);
2488exit2:
2489 path_release(&newnd);
2490exit1:
2491 path_release(&oldnd);
2492exit:
2493 return error;
2494}
2495
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002496asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2497 int newdfd, const char __user *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498{
2499 int error;
2500 char * from;
2501 char * to;
2502
2503 from = getname(oldname);
2504 if(IS_ERR(from))
2505 return PTR_ERR(from);
2506 to = getname(newname);
2507 error = PTR_ERR(to);
2508 if (!IS_ERR(to)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002509 error = do_rename(olddfd, from, newdfd, to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510 putname(to);
2511 }
2512 putname(from);
2513 return error;
2514}
2515
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002516asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2517{
2518 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2519}
2520
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2522{
2523 int len;
2524
2525 len = PTR_ERR(link);
2526 if (IS_ERR(link))
2527 goto out;
2528
2529 len = strlen(link);
2530 if (len > (unsigned) buflen)
2531 len = buflen;
2532 if (copy_to_user(buffer, link, len))
2533 len = -EFAULT;
2534out:
2535 return len;
2536}
2537
2538/*
2539 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2540 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2541 * using) it for any given inode is up to filesystem.
2542 */
2543int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2544{
2545 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002546 void *cookie;
2547
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002549 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2550 if (!IS_ERR(cookie)) {
2551 int res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552 if (dentry->d_inode->i_op->put_link)
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002553 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2554 cookie = ERR_PTR(res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 }
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002556 return PTR_ERR(cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557}
2558
2559int vfs_follow_link(struct nameidata *nd, const char *link)
2560{
2561 return __vfs_follow_link(nd, link);
2562}
2563
2564/* get the link contents into pagecache */
2565static char *page_getlink(struct dentry * dentry, struct page **ppage)
2566{
2567 struct page * page;
2568 struct address_space *mapping = dentry->d_inode->i_mapping;
2569 page = read_cache_page(mapping, 0, (filler_t *)mapping->a_ops->readpage,
2570 NULL);
2571 if (IS_ERR(page))
2572 goto sync_fail;
2573 wait_on_page_locked(page);
2574 if (!PageUptodate(page))
2575 goto async_fail;
2576 *ppage = page;
2577 return kmap(page);
2578
2579async_fail:
2580 page_cache_release(page);
2581 return ERR_PTR(-EIO);
2582
2583sync_fail:
2584 return (char*)page;
2585}
2586
2587int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2588{
2589 struct page *page = NULL;
2590 char *s = page_getlink(dentry, &page);
2591 int res = vfs_readlink(dentry,buffer,buflen,s);
2592 if (page) {
2593 kunmap(page);
2594 page_cache_release(page);
2595 }
2596 return res;
2597}
2598
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002599void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002601 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002603 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002604}
2605
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002606void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002608 struct page *page = cookie;
2609
2610 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611 kunmap(page);
2612 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613 }
2614}
2615
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002616int __page_symlink(struct inode *inode, const char *symname, int len,
2617 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618{
2619 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002620 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621 int err = -ENOMEM;
2622 char *kaddr;
2623
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002624 page = find_or_create_page(mapping, 0, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625 if (!page)
2626 goto fail;
2627 err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
2628 if (err)
2629 goto fail_map;
2630 kaddr = kmap_atomic(page, KM_USER0);
2631 memcpy(kaddr, symname, len-1);
2632 kunmap_atomic(kaddr, KM_USER0);
2633 mapping->a_ops->commit_write(NULL, page, 0, len-1);
2634 /*
2635 * Notice that we are _not_ going to block here - end of page is
2636 * unmapped, so this will only try to map the rest of page, see
2637 * that it is unmapped (typically even will not look into inode -
2638 * ->i_size will be enough for everything) and zero it out.
2639 * OTOH it's obviously correct and should make the page up-to-date.
2640 */
2641 if (!PageUptodate(page)) {
2642 err = mapping->a_ops->readpage(NULL, page);
2643 wait_on_page_locked(page);
2644 } else {
2645 unlock_page(page);
2646 }
2647 page_cache_release(page);
2648 if (err < 0)
2649 goto fail;
2650 mark_inode_dirty(inode);
2651 return 0;
2652fail_map:
2653 unlock_page(page);
2654 page_cache_release(page);
2655fail:
2656 return err;
2657}
2658
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002659int page_symlink(struct inode *inode, const char *symname, int len)
2660{
2661 return __page_symlink(inode, symname, len,
2662 mapping_gfp_mask(inode->i_mapping));
2663}
2664
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665struct inode_operations page_symlink_inode_operations = {
2666 .readlink = generic_readlink,
2667 .follow_link = page_follow_link_light,
2668 .put_link = page_put_link,
2669};
2670
2671EXPORT_SYMBOL(__user_walk);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002672EXPORT_SYMBOL(__user_walk_fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673EXPORT_SYMBOL(follow_down);
2674EXPORT_SYMBOL(follow_up);
2675EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2676EXPORT_SYMBOL(getname);
2677EXPORT_SYMBOL(lock_rename);
2678EXPORT_SYMBOL(lookup_hash);
2679EXPORT_SYMBOL(lookup_one_len);
2680EXPORT_SYMBOL(page_follow_link_light);
2681EXPORT_SYMBOL(page_put_link);
2682EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002683EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002684EXPORT_SYMBOL(page_symlink);
2685EXPORT_SYMBOL(page_symlink_inode_operations);
2686EXPORT_SYMBOL(path_lookup);
2687EXPORT_SYMBOL(path_release);
2688EXPORT_SYMBOL(path_walk);
2689EXPORT_SYMBOL(permission);
Christoph Hellwige4543ed2005-11-08 21:35:04 -08002690EXPORT_SYMBOL(vfs_permission);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -08002691EXPORT_SYMBOL(file_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692EXPORT_SYMBOL(unlock_rename);
2693EXPORT_SYMBOL(vfs_create);
2694EXPORT_SYMBOL(vfs_follow_link);
2695EXPORT_SYMBOL(vfs_link);
2696EXPORT_SYMBOL(vfs_mkdir);
2697EXPORT_SYMBOL(vfs_mknod);
2698EXPORT_SYMBOL(generic_permission);
2699EXPORT_SYMBOL(vfs_readlink);
2700EXPORT_SYMBOL(vfs_rename);
2701EXPORT_SYMBOL(vfs_rmdir);
2702EXPORT_SYMBOL(vfs_symlink);
2703EXPORT_SYMBOL(vfs_unlink);
2704EXPORT_SYMBOL(dentry_unhash);
2705EXPORT_SYMBOL(generic_readlink);