blob: 4acdac043b6bb0a1d42b66b7a601513b2049f06b [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)
107 * implemented. Let's see if raised priority of ->s_vfs_rename_sem gives
108 * 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)
793 lookup_flags = LOOKUP_FOLLOW;
794
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:
888 nd->flags &= ~LOOKUP_CONTINUE;
889 if (lookup_flags & LOOKUP_PARENT)
890 goto lookup_parent;
891 if (this.name[0] == '.') switch (this.len) {
892 default:
893 break;
894 case 2:
895 if (this.name[1] != '.')
896 break;
Al Viro58c465e2005-06-06 13:36:13 -0700897 follow_dotdot(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 inode = nd->dentry->d_inode;
899 /* fallthrough */
900 case 1:
901 goto return_reval;
902 }
903 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
904 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
905 if (err < 0)
906 break;
907 }
908 err = do_lookup(nd, &this, &next);
909 if (err)
910 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 inode = next.dentry->d_inode;
912 if ((lookup_flags & LOOKUP_FOLLOW)
913 && inode && inode->i_op && inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700914 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 if (err)
916 goto return_err;
917 inode = nd->dentry->d_inode;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700918 } else
919 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 err = -ENOENT;
921 if (!inode)
922 break;
923 if (lookup_flags & LOOKUP_DIRECTORY) {
924 err = -ENOTDIR;
925 if (!inode->i_op || !inode->i_op->lookup)
926 break;
927 }
928 goto return_base;
929lookup_parent:
930 nd->last = this;
931 nd->last_type = LAST_NORM;
932 if (this.name[0] != '.')
933 goto return_base;
934 if (this.len == 1)
935 nd->last_type = LAST_DOT;
936 else if (this.len == 2 && this.name[1] == '.')
937 nd->last_type = LAST_DOTDOT;
938 else
939 goto return_base;
940return_reval:
941 /*
942 * We bypassed the ordinary revalidation routines.
943 * We may need to check the cached dentry for staleness.
944 */
945 if (nd->dentry && nd->dentry->d_sb &&
946 (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
947 err = -ESTALE;
948 /* Note: we do not d_invalidate() */
949 if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
950 break;
951 }
952return_base:
953 return 0;
954out_dput:
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700955 dput_path(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 break;
957 }
958 path_release(nd);
959return_err:
960 return err;
961}
962
963/*
964 * Wrapper to retry pathname resolution whenever the underlying
965 * file system returns an ESTALE.
966 *
967 * Retry the whole path once, forcing real lookup requests
968 * instead of relying on the dcache.
969 */
970int fastcall link_path_walk(const char *name, struct nameidata *nd)
971{
972 struct nameidata save = *nd;
973 int result;
974
975 /* make sure the stuff we saved doesn't go away */
976 dget(save.dentry);
977 mntget(save.mnt);
978
979 result = __link_path_walk(name, nd);
980 if (result == -ESTALE) {
981 *nd = save;
982 dget(nd->dentry);
983 mntget(nd->mnt);
984 nd->flags |= LOOKUP_REVAL;
985 result = __link_path_walk(name, nd);
986 }
987
988 dput(save.dentry);
989 mntput(save.mnt);
990
991 return result;
992}
993
994int fastcall path_walk(const char * name, struct nameidata *nd)
995{
996 current->total_link_count = 0;
997 return link_path_walk(name, nd);
998}
999
Prasanna Medaea3834d2005-04-29 16:00:17 +01001000/*
1001 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
1002 * everything is done. Returns 0 and drops input nd, if lookup failed;
1003 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
1005{
1006 if (path_walk(name, nd))
1007 return 0; /* something went wrong... */
1008
1009 if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
1010 struct dentry *old_dentry = nd->dentry;
1011 struct vfsmount *old_mnt = nd->mnt;
1012 struct qstr last = nd->last;
1013 int last_type = nd->last_type;
1014 /*
1015 * NAME was not found in alternate root or it's a directory. Try to find
1016 * it in the normal root:
1017 */
1018 nd->last_type = LAST_ROOT;
1019 read_lock(&current->fs->lock);
1020 nd->mnt = mntget(current->fs->rootmnt);
1021 nd->dentry = dget(current->fs->root);
1022 read_unlock(&current->fs->lock);
1023 if (path_walk(name, nd) == 0) {
1024 if (nd->dentry->d_inode) {
1025 dput(old_dentry);
1026 mntput(old_mnt);
1027 return 1;
1028 }
1029 path_release(nd);
1030 }
1031 nd->dentry = old_dentry;
1032 nd->mnt = old_mnt;
1033 nd->last = last;
1034 nd->last_type = last_type;
1035 }
1036 return 1;
1037}
1038
1039void set_fs_altroot(void)
1040{
1041 char *emul = __emul_prefix();
1042 struct nameidata nd;
1043 struct vfsmount *mnt = NULL, *oldmnt;
1044 struct dentry *dentry = NULL, *olddentry;
1045 int err;
1046
1047 if (!emul)
1048 goto set_it;
1049 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
1050 if (!err) {
1051 mnt = nd.mnt;
1052 dentry = nd.dentry;
1053 }
1054set_it:
1055 write_lock(&current->fs->lock);
1056 oldmnt = current->fs->altrootmnt;
1057 olddentry = current->fs->altroot;
1058 current->fs->altrootmnt = mnt;
1059 current->fs->altroot = dentry;
1060 write_unlock(&current->fs->lock);
1061 if (olddentry) {
1062 dput(olddentry);
1063 mntput(oldmnt);
1064 }
1065}
1066
Prasanna Medaea3834d2005-04-29 16:00:17 +01001067/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001068static int fastcall do_path_lookup(int dfd, const char *name,
1069 unsigned int flags, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001071 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072
1073 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1074 nd->flags = flags;
1075 nd->depth = 0;
1076
1077 read_lock(&current->fs->lock);
1078 if (*name=='/') {
1079 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
1080 nd->mnt = mntget(current->fs->altrootmnt);
1081 nd->dentry = dget(current->fs->altroot);
1082 read_unlock(&current->fs->lock);
1083 if (__emul_lookup_dentry(name,nd))
Prasanna Medaea3834d2005-04-29 16:00:17 +01001084 goto out; /* found in altroot */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 read_lock(&current->fs->lock);
1086 }
1087 nd->mnt = mntget(current->fs->rootmnt);
1088 nd->dentry = dget(current->fs->root);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001089 } else if (dfd == AT_FDCWD) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 nd->mnt = mntget(current->fs->pwdmnt);
1091 nd->dentry = dget(current->fs->pwd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001092 } else {
1093 struct file *file;
1094 int fput_needed;
1095 struct dentry *dentry;
1096
1097 file = fget_light(dfd, &fput_needed);
1098 if (!file) {
1099 retval = -EBADF;
1100 goto out_fail;
1101 }
1102
1103 dentry = file->f_dentry;
1104
1105 if (!S_ISDIR(dentry->d_inode->i_mode)) {
1106 retval = -ENOTDIR;
1107 fput_light(file, fput_needed);
1108 goto out_fail;
1109 }
1110
1111 retval = file_permission(file, MAY_EXEC);
1112 if (retval) {
1113 fput_light(file, fput_needed);
1114 goto out_fail;
1115 }
1116
1117 nd->mnt = mntget(file->f_vfsmnt);
1118 nd->dentry = dget(dentry);
1119
1120 fput_light(file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 }
1122 read_unlock(&current->fs->lock);
1123 current->total_link_count = 0;
1124 retval = link_path_walk(name, nd);
Prasanna Medaea3834d2005-04-29 16:00:17 +01001125out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 if (unlikely(current->audit_context
1127 && nd && nd->dentry && nd->dentry->d_inode))
David Woodhouseae7b9612005-06-20 16:11:05 +01001128 audit_inode(name, nd->dentry->d_inode, flags);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001129out_fail:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 return retval;
1131}
1132
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001133int fastcall path_lookup(const char *name, unsigned int flags,
1134 struct nameidata *nd)
1135{
1136 return do_path_lookup(AT_FDCWD, name, flags, nd);
1137}
1138
1139static int __path_lookup_intent_open(int dfd, const char *name,
1140 unsigned int lookup_flags, struct nameidata *nd,
1141 int open_flags, int create_mode)
Trond Myklebust834f2a42005-10-18 14:20:16 -07001142{
1143 struct file *filp = get_empty_filp();
1144 int err;
1145
1146 if (filp == NULL)
1147 return -ENFILE;
1148 nd->intent.open.file = filp;
1149 nd->intent.open.flags = open_flags;
1150 nd->intent.open.create_mode = create_mode;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001151 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
Trond Myklebust834f2a42005-10-18 14:20:16 -07001152 if (IS_ERR(nd->intent.open.file)) {
1153 if (err == 0) {
1154 err = PTR_ERR(nd->intent.open.file);
1155 path_release(nd);
1156 }
1157 } else if (err != 0)
1158 release_open_intent(nd);
1159 return err;
1160}
1161
1162/**
1163 * path_lookup_open - lookup a file path with open intent
1164 * @name: pointer to file name
1165 * @lookup_flags: lookup intent flags
1166 * @nd: pointer to nameidata
1167 * @open_flags: open intent flags
1168 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001169int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
Trond Myklebust834f2a42005-10-18 14:20:16 -07001170 struct nameidata *nd, int open_flags)
1171{
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001172 return __path_lookup_intent_open(dfd, name, lookup_flags, nd,
Trond Myklebust834f2a42005-10-18 14:20:16 -07001173 open_flags, 0);
1174}
1175
1176/**
1177 * path_lookup_create - lookup a file path with open + create intent
1178 * @name: pointer to file name
1179 * @lookup_flags: lookup intent flags
1180 * @nd: pointer to nameidata
1181 * @open_flags: open intent flags
1182 * @create_mode: create intent flags
1183 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001184static int path_lookup_create(int dfd, const char *name,
1185 unsigned int lookup_flags, struct nameidata *nd,
1186 int open_flags, int create_mode)
Trond Myklebust834f2a42005-10-18 14:20:16 -07001187{
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001188 return __path_lookup_intent_open(dfd, name, lookup_flags|LOOKUP_CREATE,
1189 nd, open_flags, create_mode);
Trond Myklebust834f2a42005-10-18 14:20:16 -07001190}
1191
1192int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags,
1193 struct nameidata *nd, int open_flags)
1194{
1195 char *tmp = getname(name);
1196 int err = PTR_ERR(tmp);
1197
1198 if (!IS_ERR(tmp)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001199 err = __path_lookup_intent_open(AT_FDCWD, tmp, lookup_flags, nd, open_flags, 0);
Trond Myklebust834f2a42005-10-18 14:20:16 -07001200 putname(tmp);
1201 }
1202 return err;
1203}
1204
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205/*
1206 * Restricted form of lookup. Doesn't follow links, single-component only,
1207 * needs parent already locked. Doesn't follow mounts.
1208 * SMP-safe.
1209 */
1210static struct dentry * __lookup_hash(struct qstr *name, struct dentry * base, struct nameidata *nd)
1211{
1212 struct dentry * dentry;
1213 struct inode *inode;
1214 int err;
1215
1216 inode = base->d_inode;
1217 err = permission(inode, MAY_EXEC, nd);
1218 dentry = ERR_PTR(err);
1219 if (err)
1220 goto out;
1221
1222 /*
1223 * See if the low-level filesystem might want
1224 * to use its own hash..
1225 */
1226 if (base->d_op && base->d_op->d_hash) {
1227 err = base->d_op->d_hash(base, name);
1228 dentry = ERR_PTR(err);
1229 if (err < 0)
1230 goto out;
1231 }
1232
1233 dentry = cached_lookup(base, name, nd);
1234 if (!dentry) {
1235 struct dentry *new = d_alloc(base, name);
1236 dentry = ERR_PTR(-ENOMEM);
1237 if (!new)
1238 goto out;
1239 dentry = inode->i_op->lookup(inode, new, nd);
1240 if (!dentry)
1241 dentry = new;
1242 else
1243 dput(new);
1244 }
1245out:
1246 return dentry;
1247}
1248
Christoph Hellwig49705b72005-11-08 21:35:06 -08001249struct dentry * lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250{
Christoph Hellwig49705b72005-11-08 21:35:06 -08001251 return __lookup_hash(&nd->last, nd->dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252}
1253
1254/* SMP-safe */
1255struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
1256{
1257 unsigned long hash;
1258 struct qstr this;
1259 unsigned int c;
1260
1261 this.name = name;
1262 this.len = len;
1263 if (!len)
1264 goto access;
1265
1266 hash = init_name_hash();
1267 while (len--) {
1268 c = *(const unsigned char *)name++;
1269 if (c == '/' || c == '\0')
1270 goto access;
1271 hash = partial_name_hash(c, hash);
1272 }
1273 this.hash = end_name_hash(hash);
1274
Christoph Hellwig49705b72005-11-08 21:35:06 -08001275 return __lookup_hash(&this, base, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276access:
1277 return ERR_PTR(-EACCES);
1278}
1279
1280/*
1281 * namei()
1282 *
1283 * is used by most simple commands to get the inode of a specified name.
1284 * Open, link etc use their own routines, but this is enough for things
1285 * like 'chmod' etc.
1286 *
1287 * namei exists in two versions: namei/lnamei. The only difference is
1288 * that namei follows links, while lnamei does not.
1289 * SMP-safe
1290 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001291int fastcall __user_walk_fd(int dfd, const char __user *name, unsigned flags,
1292 struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293{
1294 char *tmp = getname(name);
1295 int err = PTR_ERR(tmp);
1296
1297 if (!IS_ERR(tmp)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001298 err = do_path_lookup(dfd, tmp, flags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 putname(tmp);
1300 }
1301 return err;
1302}
1303
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001304int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1305{
1306 return __user_walk_fd(AT_FDCWD, name, flags, nd);
1307}
1308
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309/*
1310 * It's inline, so penalty for filesystems that don't use sticky bit is
1311 * minimal.
1312 */
1313static inline int check_sticky(struct inode *dir, struct inode *inode)
1314{
1315 if (!(dir->i_mode & S_ISVTX))
1316 return 0;
1317 if (inode->i_uid == current->fsuid)
1318 return 0;
1319 if (dir->i_uid == current->fsuid)
1320 return 0;
1321 return !capable(CAP_FOWNER);
1322}
1323
1324/*
1325 * Check whether we can remove a link victim from directory dir, check
1326 * whether the type of victim is right.
1327 * 1. We can't do it if dir is read-only (done in permission())
1328 * 2. We should have write and exec permissions on dir
1329 * 3. We can't remove anything from append-only dir
1330 * 4. We can't do anything with immutable dir (done in permission())
1331 * 5. If the sticky bit on dir is set we should either
1332 * a. be owner of dir, or
1333 * b. be owner of victim, or
1334 * c. have CAP_FOWNER capability
1335 * 6. If the victim is append-only or immutable we can't do antyhing with
1336 * links pointing to it.
1337 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1338 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1339 * 9. We can't remove a root or mountpoint.
1340 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1341 * nfs_async_unlink().
1342 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001343static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344{
1345 int error;
1346
1347 if (!victim->d_inode)
1348 return -ENOENT;
1349
1350 BUG_ON(victim->d_parent->d_inode != dir);
1351
1352 error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
1353 if (error)
1354 return error;
1355 if (IS_APPEND(dir))
1356 return -EPERM;
1357 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1358 IS_IMMUTABLE(victim->d_inode))
1359 return -EPERM;
1360 if (isdir) {
1361 if (!S_ISDIR(victim->d_inode->i_mode))
1362 return -ENOTDIR;
1363 if (IS_ROOT(victim))
1364 return -EBUSY;
1365 } else if (S_ISDIR(victim->d_inode->i_mode))
1366 return -EISDIR;
1367 if (IS_DEADDIR(dir))
1368 return -ENOENT;
1369 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1370 return -EBUSY;
1371 return 0;
1372}
1373
1374/* Check whether we can create an object with dentry child in directory
1375 * dir.
1376 * 1. We can't do it if child already exists (open has special treatment for
1377 * this case, but since we are inlined it's OK)
1378 * 2. We can't do it if dir is read-only (done in permission())
1379 * 3. We should have write and exec permissions on dir
1380 * 4. We can't do it if dir is immutable (done in permission())
1381 */
1382static inline int may_create(struct inode *dir, struct dentry *child,
1383 struct nameidata *nd)
1384{
1385 if (child->d_inode)
1386 return -EEXIST;
1387 if (IS_DEADDIR(dir))
1388 return -ENOENT;
1389 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1390}
1391
1392/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 * O_DIRECTORY translates into forcing a directory lookup.
1394 */
1395static inline int lookup_flags(unsigned int f)
1396{
1397 unsigned long retval = LOOKUP_FOLLOW;
1398
1399 if (f & O_NOFOLLOW)
1400 retval &= ~LOOKUP_FOLLOW;
1401
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 if (f & O_DIRECTORY)
1403 retval |= LOOKUP_DIRECTORY;
1404
1405 return retval;
1406}
1407
1408/*
1409 * p1 and p2 should be directories on the same fs.
1410 */
1411struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1412{
1413 struct dentry *p;
1414
1415 if (p1 == p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001416 mutex_lock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 return NULL;
1418 }
1419
1420 down(&p1->d_inode->i_sb->s_vfs_rename_sem);
1421
1422 for (p = p1; p->d_parent != p; p = p->d_parent) {
1423 if (p->d_parent == p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001424 mutex_lock(&p2->d_inode->i_mutex);
1425 mutex_lock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 return p;
1427 }
1428 }
1429
1430 for (p = p2; p->d_parent != p; p = p->d_parent) {
1431 if (p->d_parent == p1) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001432 mutex_lock(&p1->d_inode->i_mutex);
1433 mutex_lock(&p2->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 return p;
1435 }
1436 }
1437
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001438 mutex_lock(&p1->d_inode->i_mutex);
1439 mutex_lock(&p2->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 return NULL;
1441}
1442
1443void unlock_rename(struct dentry *p1, struct dentry *p2)
1444{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001445 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001447 mutex_unlock(&p2->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 up(&p1->d_inode->i_sb->s_vfs_rename_sem);
1449 }
1450}
1451
1452int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1453 struct nameidata *nd)
1454{
1455 int error = may_create(dir, dentry, nd);
1456
1457 if (error)
1458 return error;
1459
1460 if (!dir->i_op || !dir->i_op->create)
1461 return -EACCES; /* shouldn't it be ENOSYS? */
1462 mode &= S_IALLUGO;
1463 mode |= S_IFREG;
1464 error = security_inode_create(dir, dentry, mode);
1465 if (error)
1466 return error;
1467 DQUOT_INIT(dir);
1468 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001469 if (!error)
Robert Love0eeca282005-07-12 17:06:03 -04001470 fsnotify_create(dir, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 return error;
1472}
1473
1474int may_open(struct nameidata *nd, int acc_mode, int flag)
1475{
1476 struct dentry *dentry = nd->dentry;
1477 struct inode *inode = dentry->d_inode;
1478 int error;
1479
1480 if (!inode)
1481 return -ENOENT;
1482
1483 if (S_ISLNK(inode->i_mode))
1484 return -ELOOP;
1485
1486 if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1487 return -EISDIR;
1488
Christoph Hellwige4543ed2005-11-08 21:35:04 -08001489 error = vfs_permission(nd, acc_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 if (error)
1491 return error;
1492
1493 /*
1494 * FIFO's, sockets and device files are special: they don't
1495 * actually live on the filesystem itself, and as such you
1496 * can write to them even if the filesystem is read-only.
1497 */
1498 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1499 flag &= ~O_TRUNC;
1500 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1501 if (nd->mnt->mnt_flags & MNT_NODEV)
1502 return -EACCES;
1503
1504 flag &= ~O_TRUNC;
1505 } else if (IS_RDONLY(inode) && (flag & FMODE_WRITE))
1506 return -EROFS;
1507 /*
1508 * An append-only file must be opened in append mode for writing.
1509 */
1510 if (IS_APPEND(inode)) {
1511 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1512 return -EPERM;
1513 if (flag & O_TRUNC)
1514 return -EPERM;
1515 }
1516
1517 /* O_NOATIME can only be set by the owner or superuser */
1518 if (flag & O_NOATIME)
1519 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1520 return -EPERM;
1521
1522 /*
1523 * Ensure there are no outstanding leases on the file.
1524 */
1525 error = break_lease(inode, flag);
1526 if (error)
1527 return error;
1528
1529 if (flag & O_TRUNC) {
1530 error = get_write_access(inode);
1531 if (error)
1532 return error;
1533
1534 /*
1535 * Refuse to truncate files with mandatory locks held on them.
1536 */
1537 error = locks_verify_locked(inode);
1538 if (!error) {
1539 DQUOT_INIT(inode);
1540
NeilBrown4a301312006-01-08 01:02:39 -08001541 error = do_truncate(dentry, 0, ATTR_MTIME|ATTR_CTIME, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 }
1543 put_write_access(inode);
1544 if (error)
1545 return error;
1546 } else
1547 if (flag & FMODE_WRITE)
1548 DQUOT_INIT(inode);
1549
1550 return 0;
1551}
1552
1553/*
1554 * open_namei()
1555 *
1556 * namei for open - this is in fact almost the whole open-routine.
1557 *
1558 * Note that the low bits of "flag" aren't the same as in the open
1559 * system call - they are 00 - no permissions needed
1560 * 01 - read permission needed
1561 * 10 - write permission needed
1562 * 11 - read/write permissions needed
1563 * which is a lot more logical, and also allows the "no perm" needed
1564 * for symlinks (where the permissions are checked later).
1565 * SMP-safe
1566 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001567int open_namei(int dfd, const char *pathname, int flag,
1568 int mode, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569{
Trond Myklebust834f2a42005-10-18 14:20:16 -07001570 int acc_mode, error;
Al Viro4e7506e2005-06-06 13:36:00 -07001571 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 struct dentry *dir;
1573 int count = 0;
1574
1575 acc_mode = ACC_MODE(flag);
1576
Trond Myklebust834f2a42005-10-18 14:20:16 -07001577 /* O_TRUNC implies we need access checks for write permissions */
1578 if (flag & O_TRUNC)
1579 acc_mode |= MAY_WRITE;
1580
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 /* Allow the LSM permission hook to distinguish append
1582 access from general write access. */
1583 if (flag & O_APPEND)
1584 acc_mode |= MAY_APPEND;
1585
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 /*
1587 * The simplest case - just a plain lookup.
1588 */
1589 if (!(flag & O_CREAT)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001590 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1591 nd, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 if (error)
1593 return error;
1594 goto ok;
1595 }
1596
1597 /*
1598 * Create - we need to know the parent.
1599 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001600 error = path_lookup_create(dfd,pathname,LOOKUP_PARENT,nd,flag,mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 if (error)
1602 return error;
1603
1604 /*
1605 * We have the parent and last component. First of all, check
1606 * that we are not asked to creat(2) an obvious directory - that
1607 * will not do.
1608 */
1609 error = -EISDIR;
1610 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1611 goto exit;
1612
1613 dir = nd->dentry;
1614 nd->flags &= ~LOOKUP_PARENT;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001615 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001616 path.dentry = lookup_hash(nd);
Al Virod73ffe12005-06-06 13:36:01 -07001617 path.mnt = nd->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618
1619do_last:
Al Viro4e7506e2005-06-06 13:36:00 -07001620 error = PTR_ERR(path.dentry);
1621 if (IS_ERR(path.dentry)) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001622 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 goto exit;
1624 }
1625
1626 /* Negative dentry, just create the file */
Al Viro4e7506e2005-06-06 13:36:00 -07001627 if (!path.dentry->d_inode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 if (!IS_POSIXACL(dir->d_inode))
1629 mode &= ~current->fs->umask;
Al Viro4e7506e2005-06-06 13:36:00 -07001630 error = vfs_create(dir->d_inode, path.dentry, mode, nd);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001631 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 dput(nd->dentry);
Al Viro4e7506e2005-06-06 13:36:00 -07001633 nd->dentry = path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 if (error)
1635 goto exit;
1636 /* Don't check for write permission, don't truncate */
1637 acc_mode = 0;
1638 flag &= ~O_TRUNC;
1639 goto ok;
1640 }
1641
1642 /*
1643 * It already exists.
1644 */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001645 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646
1647 error = -EEXIST;
1648 if (flag & O_EXCL)
1649 goto exit_dput;
1650
Al Viroe13b2102005-06-06 13:36:06 -07001651 if (__follow_mount(&path)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652 error = -ELOOP;
Al Viroba7a4c12005-06-06 13:36:08 -07001653 if (flag & O_NOFOLLOW)
1654 goto exit_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 }
1656 error = -ENOENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001657 if (!path.dentry->d_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 goto exit_dput;
Al Viro4e7506e2005-06-06 13:36:00 -07001659 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 goto do_link;
1661
Miklos Szeredi09dd17d2005-09-06 15:18:21 -07001662 path_to_nameidata(&path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 error = -EISDIR;
Al Viro4e7506e2005-06-06 13:36:00 -07001664 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 goto exit;
1666ok:
1667 error = may_open(nd, acc_mode, flag);
1668 if (error)
1669 goto exit;
1670 return 0;
1671
1672exit_dput:
Miklos Szeredi09dd17d2005-09-06 15:18:21 -07001673 dput_path(&path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674exit:
Trond Myklebust834f2a42005-10-18 14:20:16 -07001675 if (!IS_ERR(nd->intent.open.file))
1676 release_open_intent(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 path_release(nd);
1678 return error;
1679
1680do_link:
1681 error = -ELOOP;
1682 if (flag & O_NOFOLLOW)
1683 goto exit_dput;
1684 /*
1685 * This is subtle. Instead of calling do_follow_link() we do the
1686 * thing by hands. The reason is that this way we have zero link_count
1687 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1688 * After that we have the parent and last component, i.e.
1689 * we are in the same situation as after the first path_walk().
1690 * Well, almost - if the last component is normal we get its copy
1691 * stored in nd->last.name and we will have to putname() it when we
1692 * are done. Procfs-like symlinks just set LAST_BIND.
1693 */
1694 nd->flags |= LOOKUP_PARENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001695 error = security_inode_follow_link(path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 if (error)
1697 goto exit_dput;
Al Virocd4e91d2005-06-06 13:36:03 -07001698 error = __do_follow_link(&path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 if (error)
1700 return error;
1701 nd->flags &= ~LOOKUP_PARENT;
Al Virod671d5e2005-06-06 13:36:04 -07001702 if (nd->last_type == LAST_BIND)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 goto ok;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 error = -EISDIR;
1705 if (nd->last_type != LAST_NORM)
1706 goto exit;
1707 if (nd->last.name[nd->last.len]) {
Linus Torvalds82984112005-10-06 21:54:21 -07001708 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 goto exit;
1710 }
1711 error = -ELOOP;
1712 if (count++==32) {
Linus Torvalds82984112005-10-06 21:54:21 -07001713 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 goto exit;
1715 }
1716 dir = nd->dentry;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001717 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001718 path.dentry = lookup_hash(nd);
Al Virod671d5e2005-06-06 13:36:04 -07001719 path.mnt = nd->mnt;
Linus Torvalds82984112005-10-06 21:54:21 -07001720 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 goto do_last;
1722}
1723
1724/**
1725 * lookup_create - lookup a dentry, creating it if it doesn't exist
1726 * @nd: nameidata info
1727 * @is_dir: directory flag
1728 *
1729 * Simple function to lookup and return a dentry and create it
1730 * if it doesn't exist. Is SMP-safe.
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001731 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001732 * Returns with nd->dentry->d_inode->i_mutex locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 */
1734struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1735{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001736 struct dentry *dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001738 mutex_lock(&nd->dentry->d_inode->i_mutex);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001739 /*
1740 * Yucky last component or no last component at all?
1741 * (foo/., foo/.., /////)
1742 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 if (nd->last_type != LAST_NORM)
1744 goto fail;
1745 nd->flags &= ~LOOKUP_PARENT;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001746
1747 /*
1748 * Do the final lookup.
1749 */
Christoph Hellwig49705b72005-11-08 21:35:06 -08001750 dentry = lookup_hash(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 if (IS_ERR(dentry))
1752 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001753
1754 /*
1755 * Special case - lookup gave negative, but... we had foo/bar/
1756 * From the vfs_mknod() POV we just have a negative dentry -
1757 * all is fine. Let's be bastards - you had / on the end, you've
1758 * been asking for (non-existent) directory. -ENOENT for you.
1759 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1761 goto enoent;
1762 return dentry;
1763enoent:
1764 dput(dentry);
1765 dentry = ERR_PTR(-ENOENT);
1766fail:
1767 return dentry;
1768}
Christoph Hellwigf81a0bf2005-05-19 12:26:43 -07001769EXPORT_SYMBOL_GPL(lookup_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770
1771int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1772{
1773 int error = may_create(dir, dentry, NULL);
1774
1775 if (error)
1776 return error;
1777
1778 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1779 return -EPERM;
1780
1781 if (!dir->i_op || !dir->i_op->mknod)
1782 return -EPERM;
1783
1784 error = security_inode_mknod(dir, dentry, mode, dev);
1785 if (error)
1786 return error;
1787
1788 DQUOT_INIT(dir);
1789 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001790 if (!error)
Robert Love0eeca282005-07-12 17:06:03 -04001791 fsnotify_create(dir, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 return error;
1793}
1794
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001795asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
1796 unsigned dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797{
1798 int error = 0;
1799 char * tmp;
1800 struct dentry * dentry;
1801 struct nameidata nd;
1802
1803 if (S_ISDIR(mode))
1804 return -EPERM;
1805 tmp = getname(filename);
1806 if (IS_ERR(tmp))
1807 return PTR_ERR(tmp);
1808
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001809 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 if (error)
1811 goto out;
1812 dentry = lookup_create(&nd, 0);
1813 error = PTR_ERR(dentry);
1814
1815 if (!IS_POSIXACL(nd.dentry->d_inode))
1816 mode &= ~current->fs->umask;
1817 if (!IS_ERR(dentry)) {
1818 switch (mode & S_IFMT) {
1819 case 0: case S_IFREG:
1820 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1821 break;
1822 case S_IFCHR: case S_IFBLK:
1823 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,
1824 new_decode_dev(dev));
1825 break;
1826 case S_IFIFO: case S_IFSOCK:
1827 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,0);
1828 break;
1829 case S_IFDIR:
1830 error = -EPERM;
1831 break;
1832 default:
1833 error = -EINVAL;
1834 }
1835 dput(dentry);
1836 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001837 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 path_release(&nd);
1839out:
1840 putname(tmp);
1841
1842 return error;
1843}
1844
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001845asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
1846{
1847 return sys_mknodat(AT_FDCWD, filename, mode, dev);
1848}
1849
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1851{
1852 int error = may_create(dir, dentry, NULL);
1853
1854 if (error)
1855 return error;
1856
1857 if (!dir->i_op || !dir->i_op->mkdir)
1858 return -EPERM;
1859
1860 mode &= (S_IRWXUGO|S_ISVTX);
1861 error = security_inode_mkdir(dir, dentry, mode);
1862 if (error)
1863 return error;
1864
1865 DQUOT_INIT(dir);
1866 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001867 if (!error)
Robert Love0eeca282005-07-12 17:06:03 -04001868 fsnotify_mkdir(dir, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869 return error;
1870}
1871
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001872asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873{
1874 int error = 0;
1875 char * tmp;
1876
1877 tmp = getname(pathname);
1878 error = PTR_ERR(tmp);
1879 if (!IS_ERR(tmp)) {
1880 struct dentry *dentry;
1881 struct nameidata nd;
1882
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001883 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884 if (error)
1885 goto out;
1886 dentry = lookup_create(&nd, 1);
1887 error = PTR_ERR(dentry);
1888 if (!IS_ERR(dentry)) {
1889 if (!IS_POSIXACL(nd.dentry->d_inode))
1890 mode &= ~current->fs->umask;
1891 error = vfs_mkdir(nd.dentry->d_inode, dentry, mode);
1892 dput(dentry);
1893 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001894 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895 path_release(&nd);
1896out:
1897 putname(tmp);
1898 }
1899
1900 return error;
1901}
1902
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001903asmlinkage long sys_mkdir(const char __user *pathname, int mode)
1904{
1905 return sys_mkdirat(AT_FDCWD, pathname, mode);
1906}
1907
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908/*
1909 * We try to drop the dentry early: we should have
1910 * a usage count of 2 if we're the only user of this
1911 * dentry, and if that is true (possibly after pruning
1912 * the dcache), then we drop the dentry now.
1913 *
1914 * A low-level filesystem can, if it choses, legally
1915 * do a
1916 *
1917 * if (!d_unhashed(dentry))
1918 * return -EBUSY;
1919 *
1920 * if it cannot handle the case of removing a directory
1921 * that is still in use by something else..
1922 */
1923void dentry_unhash(struct dentry *dentry)
1924{
1925 dget(dentry);
1926 if (atomic_read(&dentry->d_count))
1927 shrink_dcache_parent(dentry);
1928 spin_lock(&dcache_lock);
1929 spin_lock(&dentry->d_lock);
1930 if (atomic_read(&dentry->d_count) == 2)
1931 __d_drop(dentry);
1932 spin_unlock(&dentry->d_lock);
1933 spin_unlock(&dcache_lock);
1934}
1935
1936int vfs_rmdir(struct inode *dir, struct dentry *dentry)
1937{
1938 int error = may_delete(dir, dentry, 1);
1939
1940 if (error)
1941 return error;
1942
1943 if (!dir->i_op || !dir->i_op->rmdir)
1944 return -EPERM;
1945
1946 DQUOT_INIT(dir);
1947
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001948 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 dentry_unhash(dentry);
1950 if (d_mountpoint(dentry))
1951 error = -EBUSY;
1952 else {
1953 error = security_inode_rmdir(dir, dentry);
1954 if (!error) {
1955 error = dir->i_op->rmdir(dir, dentry);
1956 if (!error)
1957 dentry->d_inode->i_flags |= S_DEAD;
1958 }
1959 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001960 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 d_delete(dentry);
1963 }
1964 dput(dentry);
1965
1966 return error;
1967}
1968
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001969static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970{
1971 int error = 0;
1972 char * name;
1973 struct dentry *dentry;
1974 struct nameidata nd;
1975
1976 name = getname(pathname);
1977 if(IS_ERR(name))
1978 return PTR_ERR(name);
1979
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001980 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 if (error)
1982 goto exit;
1983
1984 switch(nd.last_type) {
1985 case LAST_DOTDOT:
1986 error = -ENOTEMPTY;
1987 goto exit1;
1988 case LAST_DOT:
1989 error = -EINVAL;
1990 goto exit1;
1991 case LAST_ROOT:
1992 error = -EBUSY;
1993 goto exit1;
1994 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001995 mutex_lock(&nd.dentry->d_inode->i_mutex);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001996 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997 error = PTR_ERR(dentry);
1998 if (!IS_ERR(dentry)) {
1999 error = vfs_rmdir(nd.dentry->d_inode, dentry);
2000 dput(dentry);
2001 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002002 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003exit1:
2004 path_release(&nd);
2005exit:
2006 putname(name);
2007 return error;
2008}
2009
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002010asmlinkage long sys_rmdir(const char __user *pathname)
2011{
2012 return do_rmdir(AT_FDCWD, pathname);
2013}
2014
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015int vfs_unlink(struct inode *dir, struct dentry *dentry)
2016{
2017 int error = may_delete(dir, dentry, 0);
2018
2019 if (error)
2020 return error;
2021
2022 if (!dir->i_op || !dir->i_op->unlink)
2023 return -EPERM;
2024
2025 DQUOT_INIT(dir);
2026
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002027 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028 if (d_mountpoint(dentry))
2029 error = -EBUSY;
2030 else {
2031 error = security_inode_unlink(dir, dentry);
2032 if (!error)
2033 error = dir->i_op->unlink(dir, dentry);
2034 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002035 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036
2037 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2038 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2039 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 }
Robert Love0eeca282005-07-12 17:06:03 -04002041
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 return error;
2043}
2044
2045/*
2046 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002047 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 * writeout happening, and we don't want to prevent access to the directory
2049 * while waiting on the I/O.
2050 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002051static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052{
2053 int error = 0;
2054 char * name;
2055 struct dentry *dentry;
2056 struct nameidata nd;
2057 struct inode *inode = NULL;
2058
2059 name = getname(pathname);
2060 if(IS_ERR(name))
2061 return PTR_ERR(name);
2062
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002063 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064 if (error)
2065 goto exit;
2066 error = -EISDIR;
2067 if (nd.last_type != LAST_NORM)
2068 goto exit1;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002069 mutex_lock(&nd.dentry->d_inode->i_mutex);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002070 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 error = PTR_ERR(dentry);
2072 if (!IS_ERR(dentry)) {
2073 /* Why not before? Because we want correct error value */
2074 if (nd.last.name[nd.last.len])
2075 goto slashes;
2076 inode = dentry->d_inode;
2077 if (inode)
2078 atomic_inc(&inode->i_count);
2079 error = vfs_unlink(nd.dentry->d_inode, dentry);
2080 exit2:
2081 dput(dentry);
2082 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002083 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084 if (inode)
2085 iput(inode); /* truncate the inode here */
2086exit1:
2087 path_release(&nd);
2088exit:
2089 putname(name);
2090 return error;
2091
2092slashes:
2093 error = !dentry->d_inode ? -ENOENT :
2094 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2095 goto exit2;
2096}
2097
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002098asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2099{
2100 if ((flag & ~AT_REMOVEDIR) != 0)
2101 return -EINVAL;
2102
2103 if (flag & AT_REMOVEDIR)
2104 return do_rmdir(dfd, pathname);
2105
2106 return do_unlinkat(dfd, pathname);
2107}
2108
2109asmlinkage long sys_unlink(const char __user *pathname)
2110{
2111 return do_unlinkat(AT_FDCWD, pathname);
2112}
2113
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname, int mode)
2115{
2116 int error = may_create(dir, dentry, NULL);
2117
2118 if (error)
2119 return error;
2120
2121 if (!dir->i_op || !dir->i_op->symlink)
2122 return -EPERM;
2123
2124 error = security_inode_symlink(dir, dentry, oldname);
2125 if (error)
2126 return error;
2127
2128 DQUOT_INIT(dir);
2129 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002130 if (!error)
Robert Love0eeca282005-07-12 17:06:03 -04002131 fsnotify_create(dir, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132 return error;
2133}
2134
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002135asmlinkage long sys_symlinkat(const char __user *oldname,
2136 int newdfd, const char __user *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137{
2138 int error = 0;
2139 char * from;
2140 char * to;
2141
2142 from = getname(oldname);
2143 if(IS_ERR(from))
2144 return PTR_ERR(from);
2145 to = getname(newname);
2146 error = PTR_ERR(to);
2147 if (!IS_ERR(to)) {
2148 struct dentry *dentry;
2149 struct nameidata nd;
2150
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002151 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 if (error)
2153 goto out;
2154 dentry = lookup_create(&nd, 0);
2155 error = PTR_ERR(dentry);
2156 if (!IS_ERR(dentry)) {
2157 error = vfs_symlink(nd.dentry->d_inode, dentry, from, S_IALLUGO);
2158 dput(dentry);
2159 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002160 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 path_release(&nd);
2162out:
2163 putname(to);
2164 }
2165 putname(from);
2166 return error;
2167}
2168
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002169asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2170{
2171 return sys_symlinkat(oldname, AT_FDCWD, newname);
2172}
2173
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2175{
2176 struct inode *inode = old_dentry->d_inode;
2177 int error;
2178
2179 if (!inode)
2180 return -ENOENT;
2181
2182 error = may_create(dir, new_dentry, NULL);
2183 if (error)
2184 return error;
2185
2186 if (dir->i_sb != inode->i_sb)
2187 return -EXDEV;
2188
2189 /*
2190 * A link to an append-only or immutable file cannot be created.
2191 */
2192 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2193 return -EPERM;
2194 if (!dir->i_op || !dir->i_op->link)
2195 return -EPERM;
2196 if (S_ISDIR(old_dentry->d_inode->i_mode))
2197 return -EPERM;
2198
2199 error = security_inode_link(old_dentry, dir, new_dentry);
2200 if (error)
2201 return error;
2202
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002203 mutex_lock(&old_dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 DQUOT_INIT(dir);
2205 error = dir->i_op->link(old_dentry, dir, new_dentry);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002206 mutex_unlock(&old_dentry->d_inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002207 if (!error)
Robert Love0eeca282005-07-12 17:06:03 -04002208 fsnotify_create(dir, new_dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 return error;
2210}
2211
2212/*
2213 * Hardlinks are often used in delicate situations. We avoid
2214 * security-related surprises by not following symlinks on the
2215 * newname. --KAB
2216 *
2217 * We don't follow them on the oldname either to be compatible
2218 * with linux 2.0, and to avoid hard-linking to directories
2219 * and other special files. --ADM
2220 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002221asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
2222 int newdfd, const char __user *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223{
2224 struct dentry *new_dentry;
2225 struct nameidata nd, old_nd;
2226 int error;
2227 char * to;
2228
2229 to = getname(newname);
2230 if (IS_ERR(to))
2231 return PTR_ERR(to);
2232
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002233 error = __user_walk_fd(olddfd, oldname, 0, &old_nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234 if (error)
2235 goto exit;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002236 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 if (error)
2238 goto out;
2239 error = -EXDEV;
2240 if (old_nd.mnt != nd.mnt)
2241 goto out_release;
2242 new_dentry = lookup_create(&nd, 0);
2243 error = PTR_ERR(new_dentry);
2244 if (!IS_ERR(new_dentry)) {
2245 error = vfs_link(old_nd.dentry, nd.dentry->d_inode, new_dentry);
2246 dput(new_dentry);
2247 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002248 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249out_release:
2250 path_release(&nd);
2251out:
2252 path_release(&old_nd);
2253exit:
2254 putname(to);
2255
2256 return error;
2257}
2258
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002259asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2260{
2261 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname);
2262}
2263
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264/*
2265 * The worst of all namespace operations - renaming directory. "Perverted"
2266 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2267 * Problems:
2268 * a) we can get into loop creation. Check is done in is_subdir().
2269 * b) race potential - two innocent renames can create a loop together.
2270 * That's where 4.4 screws up. Current fix: serialization on
2271 * sb->s_vfs_rename_sem. We might be more accurate, but that's another
2272 * story.
2273 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002274 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275 * whether the target exists). Solution: try to be smart with locking
2276 * order for inodes. We rely on the fact that tree topology may change
2277 * only under ->s_vfs_rename_sem _and_ that parent of the object we
2278 * move will be locked. Thus we can rank directories by the tree
2279 * (ancestors first) and rank all non-directories after them.
2280 * That works since everybody except rename does "lock parent, lookup,
2281 * lock child" and rename is under ->s_vfs_rename_sem.
2282 * HOWEVER, it relies on the assumption that any object with ->lookup()
2283 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2284 * we'd better make sure that there's no link(2) for them.
2285 * d) some filesystems don't support opened-but-unlinked directories,
2286 * either because of layout or because they are not ready to deal with
2287 * all cases correctly. The latter will be fixed (taking this sort of
2288 * stuff into VFS), but the former is not going away. Solution: the same
2289 * trick as in rmdir().
2290 * e) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002291 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002293 * ->i_mutex on parents, which works but leads to some truely excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 * locking].
2295 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07002296static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2297 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298{
2299 int error = 0;
2300 struct inode *target;
2301
2302 /*
2303 * If we are going to change the parent - check write permissions,
2304 * we'll need to flip '..'.
2305 */
2306 if (new_dir != old_dir) {
2307 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2308 if (error)
2309 return error;
2310 }
2311
2312 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2313 if (error)
2314 return error;
2315
2316 target = new_dentry->d_inode;
2317 if (target) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002318 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319 dentry_unhash(new_dentry);
2320 }
2321 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2322 error = -EBUSY;
2323 else
2324 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2325 if (target) {
2326 if (!error)
2327 target->i_flags |= S_DEAD;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002328 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329 if (d_unhashed(new_dentry))
2330 d_rehash(new_dentry);
2331 dput(new_dentry);
2332 }
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002333 if (!error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335 return error;
2336}
2337
Adrian Bunk75c96f82005-05-05 16:16:09 -07002338static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2339 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340{
2341 struct inode *target;
2342 int error;
2343
2344 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2345 if (error)
2346 return error;
2347
2348 dget(new_dentry);
2349 target = new_dentry->d_inode;
2350 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002351 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2353 error = -EBUSY;
2354 else
2355 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2356 if (!error) {
2357 /* The following d_move() should become unconditional */
2358 if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
2359 d_move(old_dentry, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360 }
2361 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002362 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363 dput(new_dentry);
2364 return error;
2365}
2366
2367int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2368 struct inode *new_dir, struct dentry *new_dentry)
2369{
2370 int error;
2371 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Robert Love0eeca282005-07-12 17:06:03 -04002372 const char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373
2374 if (old_dentry->d_inode == new_dentry->d_inode)
2375 return 0;
2376
2377 error = may_delete(old_dir, old_dentry, is_dir);
2378 if (error)
2379 return error;
2380
2381 if (!new_dentry->d_inode)
2382 error = may_create(new_dir, new_dentry, NULL);
2383 else
2384 error = may_delete(new_dir, new_dentry, is_dir);
2385 if (error)
2386 return error;
2387
2388 if (!old_dir->i_op || !old_dir->i_op->rename)
2389 return -EPERM;
2390
2391 DQUOT_INIT(old_dir);
2392 DQUOT_INIT(new_dir);
2393
Robert Love0eeca282005-07-12 17:06:03 -04002394 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2395
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396 if (is_dir)
2397 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2398 else
2399 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2400 if (!error) {
Robert Love0eeca282005-07-12 17:06:03 -04002401 const char *new_name = old_dentry->d_name.name;
John McCutchan89204c42005-08-15 12:13:28 -04002402 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2403 new_dentry->d_inode, old_dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404 }
Robert Love0eeca282005-07-12 17:06:03 -04002405 fsnotify_oldname_free(old_name);
2406
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407 return error;
2408}
2409
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002410static int do_rename(int olddfd, const char *oldname,
2411 int newdfd, const char *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412{
2413 int error = 0;
2414 struct dentry * old_dir, * new_dir;
2415 struct dentry * old_dentry, *new_dentry;
2416 struct dentry * trap;
2417 struct nameidata oldnd, newnd;
2418
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002419 error = do_path_lookup(olddfd, oldname, LOOKUP_PARENT, &oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420 if (error)
2421 goto exit;
2422
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002423 error = do_path_lookup(newdfd, newname, LOOKUP_PARENT, &newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 if (error)
2425 goto exit1;
2426
2427 error = -EXDEV;
2428 if (oldnd.mnt != newnd.mnt)
2429 goto exit2;
2430
2431 old_dir = oldnd.dentry;
2432 error = -EBUSY;
2433 if (oldnd.last_type != LAST_NORM)
2434 goto exit2;
2435
2436 new_dir = newnd.dentry;
2437 if (newnd.last_type != LAST_NORM)
2438 goto exit2;
2439
2440 trap = lock_rename(new_dir, old_dir);
2441
Christoph Hellwig49705b72005-11-08 21:35:06 -08002442 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443 error = PTR_ERR(old_dentry);
2444 if (IS_ERR(old_dentry))
2445 goto exit3;
2446 /* source must exist */
2447 error = -ENOENT;
2448 if (!old_dentry->d_inode)
2449 goto exit4;
2450 /* unless the source is a directory trailing slashes give -ENOTDIR */
2451 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2452 error = -ENOTDIR;
2453 if (oldnd.last.name[oldnd.last.len])
2454 goto exit4;
2455 if (newnd.last.name[newnd.last.len])
2456 goto exit4;
2457 }
2458 /* source should not be ancestor of target */
2459 error = -EINVAL;
2460 if (old_dentry == trap)
2461 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08002462 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463 error = PTR_ERR(new_dentry);
2464 if (IS_ERR(new_dentry))
2465 goto exit4;
2466 /* target should not be an ancestor of source */
2467 error = -ENOTEMPTY;
2468 if (new_dentry == trap)
2469 goto exit5;
2470
2471 error = vfs_rename(old_dir->d_inode, old_dentry,
2472 new_dir->d_inode, new_dentry);
2473exit5:
2474 dput(new_dentry);
2475exit4:
2476 dput(old_dentry);
2477exit3:
2478 unlock_rename(new_dir, old_dir);
2479exit2:
2480 path_release(&newnd);
2481exit1:
2482 path_release(&oldnd);
2483exit:
2484 return error;
2485}
2486
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002487asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2488 int newdfd, const char __user *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489{
2490 int error;
2491 char * from;
2492 char * to;
2493
2494 from = getname(oldname);
2495 if(IS_ERR(from))
2496 return PTR_ERR(from);
2497 to = getname(newname);
2498 error = PTR_ERR(to);
2499 if (!IS_ERR(to)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002500 error = do_rename(olddfd, from, newdfd, to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002501 putname(to);
2502 }
2503 putname(from);
2504 return error;
2505}
2506
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002507asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2508{
2509 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2510}
2511
Linus Torvalds1da177e2005-04-16 15:20:36 -07002512int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2513{
2514 int len;
2515
2516 len = PTR_ERR(link);
2517 if (IS_ERR(link))
2518 goto out;
2519
2520 len = strlen(link);
2521 if (len > (unsigned) buflen)
2522 len = buflen;
2523 if (copy_to_user(buffer, link, len))
2524 len = -EFAULT;
2525out:
2526 return len;
2527}
2528
2529/*
2530 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2531 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2532 * using) it for any given inode is up to filesystem.
2533 */
2534int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2535{
2536 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002537 void *cookie;
2538
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002540 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2541 if (!IS_ERR(cookie)) {
2542 int res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543 if (dentry->d_inode->i_op->put_link)
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002544 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2545 cookie = ERR_PTR(res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546 }
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002547 return PTR_ERR(cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548}
2549
2550int vfs_follow_link(struct nameidata *nd, const char *link)
2551{
2552 return __vfs_follow_link(nd, link);
2553}
2554
2555/* get the link contents into pagecache */
2556static char *page_getlink(struct dentry * dentry, struct page **ppage)
2557{
2558 struct page * page;
2559 struct address_space *mapping = dentry->d_inode->i_mapping;
2560 page = read_cache_page(mapping, 0, (filler_t *)mapping->a_ops->readpage,
2561 NULL);
2562 if (IS_ERR(page))
2563 goto sync_fail;
2564 wait_on_page_locked(page);
2565 if (!PageUptodate(page))
2566 goto async_fail;
2567 *ppage = page;
2568 return kmap(page);
2569
2570async_fail:
2571 page_cache_release(page);
2572 return ERR_PTR(-EIO);
2573
2574sync_fail:
2575 return (char*)page;
2576}
2577
2578int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2579{
2580 struct page *page = NULL;
2581 char *s = page_getlink(dentry, &page);
2582 int res = vfs_readlink(dentry,buffer,buflen,s);
2583 if (page) {
2584 kunmap(page);
2585 page_cache_release(page);
2586 }
2587 return res;
2588}
2589
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002590void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002592 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002594 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595}
2596
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002597void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002599 struct page *page = cookie;
2600
2601 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602 kunmap(page);
2603 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002604 }
2605}
2606
2607int page_symlink(struct inode *inode, const char *symname, int len)
2608{
2609 struct address_space *mapping = inode->i_mapping;
2610 struct page *page = grab_cache_page(mapping, 0);
2611 int err = -ENOMEM;
2612 char *kaddr;
2613
2614 if (!page)
2615 goto fail;
2616 err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
2617 if (err)
2618 goto fail_map;
2619 kaddr = kmap_atomic(page, KM_USER0);
2620 memcpy(kaddr, symname, len-1);
2621 kunmap_atomic(kaddr, KM_USER0);
2622 mapping->a_ops->commit_write(NULL, page, 0, len-1);
2623 /*
2624 * Notice that we are _not_ going to block here - end of page is
2625 * unmapped, so this will only try to map the rest of page, see
2626 * that it is unmapped (typically even will not look into inode -
2627 * ->i_size will be enough for everything) and zero it out.
2628 * OTOH it's obviously correct and should make the page up-to-date.
2629 */
2630 if (!PageUptodate(page)) {
2631 err = mapping->a_ops->readpage(NULL, page);
2632 wait_on_page_locked(page);
2633 } else {
2634 unlock_page(page);
2635 }
2636 page_cache_release(page);
2637 if (err < 0)
2638 goto fail;
2639 mark_inode_dirty(inode);
2640 return 0;
2641fail_map:
2642 unlock_page(page);
2643 page_cache_release(page);
2644fail:
2645 return err;
2646}
2647
2648struct inode_operations page_symlink_inode_operations = {
2649 .readlink = generic_readlink,
2650 .follow_link = page_follow_link_light,
2651 .put_link = page_put_link,
2652};
2653
2654EXPORT_SYMBOL(__user_walk);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002655EXPORT_SYMBOL(__user_walk_fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002656EXPORT_SYMBOL(follow_down);
2657EXPORT_SYMBOL(follow_up);
2658EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2659EXPORT_SYMBOL(getname);
2660EXPORT_SYMBOL(lock_rename);
2661EXPORT_SYMBOL(lookup_hash);
2662EXPORT_SYMBOL(lookup_one_len);
2663EXPORT_SYMBOL(page_follow_link_light);
2664EXPORT_SYMBOL(page_put_link);
2665EXPORT_SYMBOL(page_readlink);
2666EXPORT_SYMBOL(page_symlink);
2667EXPORT_SYMBOL(page_symlink_inode_operations);
2668EXPORT_SYMBOL(path_lookup);
2669EXPORT_SYMBOL(path_release);
2670EXPORT_SYMBOL(path_walk);
2671EXPORT_SYMBOL(permission);
Christoph Hellwige4543ed2005-11-08 21:35:04 -08002672EXPORT_SYMBOL(vfs_permission);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -08002673EXPORT_SYMBOL(file_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674EXPORT_SYMBOL(unlock_rename);
2675EXPORT_SYMBOL(vfs_create);
2676EXPORT_SYMBOL(vfs_follow_link);
2677EXPORT_SYMBOL(vfs_link);
2678EXPORT_SYMBOL(vfs_mkdir);
2679EXPORT_SYMBOL(vfs_mknod);
2680EXPORT_SYMBOL(generic_permission);
2681EXPORT_SYMBOL(vfs_readlink);
2682EXPORT_SYMBOL(vfs_rename);
2683EXPORT_SYMBOL(vfs_rmdir);
2684EXPORT_SYMBOL(vfs_symlink);
2685EXPORT_SYMBOL(vfs_unlink);
2686EXPORT_SYMBOL(dentry_unhash);
2687EXPORT_SYMBOL(generic_readlink);