blob: 96723ae83c89d5283b335320b7b4315145a02d6b [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
18#include <linux/module.h>
19#include <linux/slab.h>
20#include <linux/fs.h>
21#include <linux/namei.h>
22#include <linux/quotaops.h>
23#include <linux/pagemap.h>
Robert Love0eeca282005-07-12 17:06:03 -040024#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/smp_lock.h>
26#include <linux/personality.h>
27#include <linux/security.h>
28#include <linux/syscalls.h>
29#include <linux/mount.h>
30#include <linux/audit.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080031#include <linux/capability.h>
Trond Myklebust834f2a42005-10-18 14:20:16 -070032#include <linux/file.h>
Ulrich Drepper5590ff02006-01-18 17:43:53 -080033#include <linux/fcntl.h>
34#include <linux/namei.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <asm/namei.h>
36#include <asm/uaccess.h>
37
38#define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
39
40/* [Feb-1997 T. Schoebel-Theuer]
41 * Fundamental changes in the pathname lookup mechanisms (namei)
42 * were necessary because of omirr. The reason is that omirr needs
43 * to know the _real_ pathname, not the user-supplied one, in case
44 * of symlinks (and also when transname replacements occur).
45 *
46 * The new code replaces the old recursive symlink resolution with
47 * an iterative one (in case of non-nested symlink chains). It does
48 * this with calls to <fs>_follow_link().
49 * As a side effect, dir_namei(), _namei() and follow_link() are now
50 * replaced with a single function lookup_dentry() that can handle all
51 * the special cases of the former code.
52 *
53 * With the new dcache, the pathname is stored at each inode, at least as
54 * long as the refcount of the inode is positive. As a side effect, the
55 * size of the dcache depends on the inode cache and thus is dynamic.
56 *
57 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
58 * resolution to correspond with current state of the code.
59 *
60 * Note that the symlink resolution is not *completely* iterative.
61 * There is still a significant amount of tail- and mid- recursion in
62 * the algorithm. Also, note that <fs>_readlink() is not used in
63 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
64 * may return different results than <fs>_follow_link(). Many virtual
65 * filesystems (including /proc) exhibit this behavior.
66 */
67
68/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
69 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
70 * and the name already exists in form of a symlink, try to create the new
71 * name indicated by the symlink. The old code always complained that the
72 * name already exists, due to not following the symlink even if its target
73 * is nonexistent. The new semantics affects also mknod() and link() when
74 * the name is a symlink pointing to a non-existant name.
75 *
76 * I don't know which semantics is the right one, since I have no access
77 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
78 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
79 * "old" one. Personally, I think the new semantics is much more logical.
80 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
81 * file does succeed in both HP-UX and SunOs, but not in Solaris
82 * and in the old Linux semantics.
83 */
84
85/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
86 * semantics. See the comments in "open_namei" and "do_link" below.
87 *
88 * [10-Sep-98 Alan Modra] Another symlink change.
89 */
90
91/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
92 * inside the path - always follow.
93 * in the last component in creation/removal/renaming - never follow.
94 * if LOOKUP_FOLLOW passed - follow.
95 * if the pathname has trailing slashes - follow.
96 * otherwise - don't follow.
97 * (applied in that order).
98 *
99 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
100 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
101 * During the 2.4 we need to fix the userland stuff depending on it -
102 * hopefully we will be able to get rid of that wart in 2.5. So far only
103 * XEmacs seems to be relying on it...
104 */
105/*
106 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800107 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 * any extra contention...
109 */
110
111/* In order to reduce some races, while at the same time doing additional
112 * checking and hopefully speeding things up, we copy filenames to the
113 * kernel data space before using them..
114 *
115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
116 * PATH_MAX includes the nul terminator --RR.
117 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800118static int do_getname(const char __user *filename, char *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119{
120 int retval;
121 unsigned long len = PATH_MAX;
122
123 if (!segment_eq(get_fs(), KERNEL_DS)) {
124 if ((unsigned long) filename >= TASK_SIZE)
125 return -EFAULT;
126 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
127 len = TASK_SIZE - (unsigned long) filename;
128 }
129
130 retval = strncpy_from_user(page, filename, len);
131 if (retval > 0) {
132 if (retval < len)
133 return 0;
134 return -ENAMETOOLONG;
135 } else if (!retval)
136 retval = -ENOENT;
137 return retval;
138}
139
140char * getname(const char __user * filename)
141{
142 char *tmp, *result;
143
144 result = ERR_PTR(-ENOMEM);
145 tmp = __getname();
146 if (tmp) {
147 int retval = do_getname(filename, tmp);
148
149 result = tmp;
150 if (retval < 0) {
151 __putname(tmp);
152 result = ERR_PTR(retval);
153 }
154 }
155 audit_getname(result);
156 return result;
157}
158
159#ifdef CONFIG_AUDITSYSCALL
160void putname(const char *name)
161{
162 if (unlikely(current->audit_context))
163 audit_putname(name);
164 else
165 __putname(name);
166}
167EXPORT_SYMBOL(putname);
168#endif
169
170
171/**
172 * generic_permission - check for access rights on a Posix-like filesystem
173 * @inode: inode to check access rights for
174 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
175 * @check_acl: optional callback to check for Posix ACLs
176 *
177 * Used to check for read/write/execute permissions on a file.
178 * We use "fsuid" for this, letting us set arbitrary permissions
179 * for filesystem access without changing the "normal" uids which
180 * are used for other things..
181 */
182int generic_permission(struct inode *inode, int mask,
183 int (*check_acl)(struct inode *inode, int mask))
184{
185 umode_t mode = inode->i_mode;
186
187 if (current->fsuid == inode->i_uid)
188 mode >>= 6;
189 else {
190 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
191 int error = check_acl(inode, mask);
192 if (error == -EACCES)
193 goto check_capabilities;
194 else if (error != -EAGAIN)
195 return error;
196 }
197
198 if (in_group_p(inode->i_gid))
199 mode >>= 3;
200 }
201
202 /*
203 * If the DACs are ok we don't need any capability check.
204 */
205 if (((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask))
206 return 0;
207
208 check_capabilities:
209 /*
210 * Read/write DACs are always overridable.
211 * Executable DACs are overridable if at least one exec bit is set.
212 */
213 if (!(mask & MAY_EXEC) ||
214 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
215 if (capable(CAP_DAC_OVERRIDE))
216 return 0;
217
218 /*
219 * Searching includes executable on directories, else just read.
220 */
221 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
222 if (capable(CAP_DAC_READ_SEARCH))
223 return 0;
224
225 return -EACCES;
226}
227
228int permission(struct inode *inode, int mask, struct nameidata *nd)
229{
230 int retval, submask;
231
232 if (mask & MAY_WRITE) {
233 umode_t mode = inode->i_mode;
234
235 /*
236 * Nobody gets write access to a read-only fs.
237 */
238 if (IS_RDONLY(inode) &&
239 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
240 return -EROFS;
241
242 /*
243 * Nobody gets write access to an immutable file.
244 */
245 if (IS_IMMUTABLE(inode))
246 return -EACCES;
247 }
248
249
250 /* Ordinary permission routines do not understand MAY_APPEND. */
251 submask = mask & ~MAY_APPEND;
252 if (inode->i_op && inode->i_op->permission)
253 retval = inode->i_op->permission(inode, submask, nd);
254 else
255 retval = generic_permission(inode, submask, NULL);
256 if (retval)
257 return retval;
258
259 return security_inode_permission(inode, mask, nd);
260}
261
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800262/**
263 * vfs_permission - check for access rights to a given path
264 * @nd: lookup result that describes the path
265 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
266 *
267 * Used to check for read/write/execute permissions on a path.
268 * We use "fsuid" for this, letting us set arbitrary permissions
269 * for filesystem access without changing the "normal" uids which
270 * are used for other things.
271 */
272int vfs_permission(struct nameidata *nd, int mask)
273{
274 return permission(nd->dentry->d_inode, mask, nd);
275}
276
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800277/**
278 * file_permission - check for additional access rights to a given file
279 * @file: file to check access rights for
280 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
281 *
282 * Used to check for read/write/execute permissions on an already opened
283 * file.
284 *
285 * Note:
286 * Do not use this function in new code. All access checks should
287 * be done using vfs_permission().
288 */
289int file_permission(struct file *file, int mask)
290{
291 return permission(file->f_dentry->d_inode, mask, NULL);
292}
293
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294/*
295 * get_write_access() gets write permission for a file.
296 * put_write_access() releases this write permission.
297 * This is used for regular files.
298 * We cannot support write (and maybe mmap read-write shared) accesses and
299 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
300 * can have the following values:
301 * 0: no writers, no VM_DENYWRITE mappings
302 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
303 * > 0: (i_writecount) users are writing to the file.
304 *
305 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
306 * except for the cases where we don't hold i_writecount yet. Then we need to
307 * use {get,deny}_write_access() - these functions check the sign and refuse
308 * to do the change if sign is wrong. Exclusion between them is provided by
309 * the inode->i_lock spinlock.
310 */
311
312int get_write_access(struct inode * inode)
313{
314 spin_lock(&inode->i_lock);
315 if (atomic_read(&inode->i_writecount) < 0) {
316 spin_unlock(&inode->i_lock);
317 return -ETXTBSY;
318 }
319 atomic_inc(&inode->i_writecount);
320 spin_unlock(&inode->i_lock);
321
322 return 0;
323}
324
325int deny_write_access(struct file * file)
326{
327 struct inode *inode = file->f_dentry->d_inode;
328
329 spin_lock(&inode->i_lock);
330 if (atomic_read(&inode->i_writecount) > 0) {
331 spin_unlock(&inode->i_lock);
332 return -ETXTBSY;
333 }
334 atomic_dec(&inode->i_writecount);
335 spin_unlock(&inode->i_lock);
336
337 return 0;
338}
339
340void path_release(struct nameidata *nd)
341{
342 dput(nd->dentry);
343 mntput(nd->mnt);
344}
345
346/*
347 * umount() mustn't call path_release()/mntput() as that would clear
348 * mnt_expiry_mark
349 */
350void path_release_on_umount(struct nameidata *nd)
351{
352 dput(nd->dentry);
Miklos Szeredi751c4042005-07-07 17:57:30 -0700353 mntput_no_expire(nd->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354}
355
Trond Myklebust834f2a42005-10-18 14:20:16 -0700356/**
357 * release_open_intent - free up open intent resources
358 * @nd: pointer to nameidata
359 */
360void release_open_intent(struct nameidata *nd)
361{
362 if (nd->intent.open.file->f_dentry == NULL)
363 put_filp(nd->intent.open.file);
364 else
365 fput(nd->intent.open.file);
366}
367
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368/*
369 * Internal lookup() using the new generic dcache.
370 * SMP-safe
371 */
372static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
373{
374 struct dentry * dentry = __d_lookup(parent, name);
375
376 /* lockess __d_lookup may fail due to concurrent d_move()
377 * in some unrelated directory, so try with d_lookup
378 */
379 if (!dentry)
380 dentry = d_lookup(parent, name);
381
382 if (dentry && dentry->d_op && dentry->d_op->d_revalidate) {
383 if (!dentry->d_op->d_revalidate(dentry, nd) && !d_invalidate(dentry)) {
384 dput(dentry);
385 dentry = NULL;
386 }
387 }
388 return dentry;
389}
390
391/*
392 * Short-cut version of permission(), for calling by
393 * path_walk(), when dcache lock is held. Combines parts
394 * of permission() and generic_permission(), and tests ONLY for
395 * MAY_EXEC permission.
396 *
397 * If appropriate, check DAC only. If not appropriate, or
398 * short-cut DAC fails, then call permission() to do more
399 * complete permission check.
400 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800401static int exec_permission_lite(struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 struct nameidata *nd)
403{
404 umode_t mode = inode->i_mode;
405
406 if (inode->i_op && inode->i_op->permission)
407 return -EAGAIN;
408
409 if (current->fsuid == inode->i_uid)
410 mode >>= 6;
411 else if (in_group_p(inode->i_gid))
412 mode >>= 3;
413
414 if (mode & MAY_EXEC)
415 goto ok;
416
417 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
418 goto ok;
419
420 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
421 goto ok;
422
423 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
424 goto ok;
425
426 return -EACCES;
427ok:
428 return security_inode_permission(inode, MAY_EXEC, nd);
429}
430
431/*
432 * This is called when everything else fails, and we actually have
433 * to go to the low-level filesystem to find out what we should do..
434 *
435 * We get the directory semaphore, and after getting that we also
436 * make sure that nobody added the entry to the dcache in the meantime..
437 * SMP-safe
438 */
439static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
440{
441 struct dentry * result;
442 struct inode *dir = parent->d_inode;
443
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800444 mutex_lock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 /*
446 * First re-do the cached lookup just in case it was created
447 * while we waited for the directory semaphore..
448 *
449 * FIXME! This could use version numbering or similar to
450 * avoid unnecessary cache lookups.
451 *
452 * The "dcache_lock" is purely to protect the RCU list walker
453 * from concurrent renames at this point (we mustn't get false
454 * negatives from the RCU list walk here, unlike the optimistic
455 * fast walk).
456 *
457 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
458 */
459 result = d_lookup(parent, name);
460 if (!result) {
461 struct dentry * dentry = d_alloc(parent, name);
462 result = ERR_PTR(-ENOMEM);
463 if (dentry) {
464 result = dir->i_op->lookup(dir, dentry, nd);
465 if (result)
466 dput(dentry);
467 else
468 result = dentry;
469 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800470 mutex_unlock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 return result;
472 }
473
474 /*
475 * Uhhuh! Nasty case: the cache was re-populated while
476 * we waited on the semaphore. Need to revalidate.
477 */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800478 mutex_unlock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 if (result->d_op && result->d_op->d_revalidate) {
480 if (!result->d_op->d_revalidate(result, nd) && !d_invalidate(result)) {
481 dput(result);
482 result = ERR_PTR(-ENOENT);
483 }
484 }
485 return result;
486}
487
488static int __emul_lookup_dentry(const char *, struct nameidata *);
489
490/* SMP-safe */
Arjan van de Venf1662352006-01-14 13:21:31 -0800491static __always_inline int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492walk_init_root(const char *name, struct nameidata *nd)
493{
494 read_lock(&current->fs->lock);
495 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
496 nd->mnt = mntget(current->fs->altrootmnt);
497 nd->dentry = dget(current->fs->altroot);
498 read_unlock(&current->fs->lock);
499 if (__emul_lookup_dentry(name,nd))
500 return 0;
501 read_lock(&current->fs->lock);
502 }
503 nd->mnt = mntget(current->fs->rootmnt);
504 nd->dentry = dget(current->fs->root);
505 read_unlock(&current->fs->lock);
506 return 1;
507}
508
Arjan van de Venf1662352006-01-14 13:21:31 -0800509static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510{
511 int res = 0;
512 char *name;
513 if (IS_ERR(link))
514 goto fail;
515
516 if (*link == '/') {
517 path_release(nd);
518 if (!walk_init_root(link, nd))
519 /* weird __emul_prefix() stuff did it */
520 goto out;
521 }
522 res = link_path_walk(link, nd);
523out:
524 if (nd->depth || res || nd->last_type!=LAST_NORM)
525 return res;
526 /*
527 * If it is an iterative symlinks resolution in open_namei() we
528 * have to copy the last component. And all that crap because of
529 * bloody create() on broken symlinks. Furrfu...
530 */
531 name = __getname();
532 if (unlikely(!name)) {
533 path_release(nd);
534 return -ENOMEM;
535 }
536 strcpy(name, nd->last.name);
537 nd->last.name = name;
538 return 0;
539fail:
540 path_release(nd);
541 return PTR_ERR(link);
542}
543
Al Viro90ebe562005-06-06 13:35:58 -0700544struct path {
545 struct vfsmount *mnt;
546 struct dentry *dentry;
547};
548
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700549static inline void dput_path(struct path *path, struct nameidata *nd)
550{
551 dput(path->dentry);
552 if (path->mnt != nd->mnt)
553 mntput(path->mnt);
554}
555
556static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
557{
558 dput(nd->dentry);
559 if (nd->mnt != path->mnt)
560 mntput(nd->mnt);
561 nd->mnt = path->mnt;
562 nd->dentry = path->dentry;
563}
564
Ian Kent051d3812006-03-27 01:14:53 -0800565static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
566{
567 int error;
568 void *cookie;
569 struct dentry *dentry = path->dentry;
570
571 touch_atime(path->mnt, dentry);
572 nd_set_link(nd, NULL);
573
574 if (path->mnt != nd->mnt) {
575 path_to_nameidata(path, nd);
576 dget(dentry);
577 }
578 mntget(path->mnt);
579 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
580 error = PTR_ERR(cookie);
581 if (!IS_ERR(cookie)) {
582 char *s = nd_get_link(nd);
583 error = 0;
584 if (s)
585 error = __vfs_follow_link(nd, s);
586 if (dentry->d_inode->i_op->put_link)
587 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
588 }
589 dput(dentry);
590 mntput(path->mnt);
591
592 return error;
593}
594
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595/*
596 * This limits recursive symlink follows to 8, while
597 * limiting consecutive symlinks to 40.
598 *
599 * Without that kind of total limit, nasty chains of consecutive
600 * symlinks can cause almost arbitrarily long lookups.
601 */
Al Viro90ebe562005-06-06 13:35:58 -0700602static inline int do_follow_link(struct path *path, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603{
604 int err = -ELOOP;
605 if (current->link_count >= MAX_NESTED_LINKS)
606 goto loop;
607 if (current->total_link_count >= 40)
608 goto loop;
609 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
610 cond_resched();
Al Viro90ebe562005-06-06 13:35:58 -0700611 err = security_inode_follow_link(path->dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 if (err)
613 goto loop;
614 current->link_count++;
615 current->total_link_count++;
616 nd->depth++;
Al Virocd4e91d2005-06-06 13:36:03 -0700617 err = __do_follow_link(path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 current->link_count--;
619 nd->depth--;
620 return err;
621loop:
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700622 dput_path(path, nd);
Al Viro839d9f92005-06-06 13:36:02 -0700623 path_release(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 return err;
625}
626
627int follow_up(struct vfsmount **mnt, struct dentry **dentry)
628{
629 struct vfsmount *parent;
630 struct dentry *mountpoint;
631 spin_lock(&vfsmount_lock);
632 parent=(*mnt)->mnt_parent;
633 if (parent == *mnt) {
634 spin_unlock(&vfsmount_lock);
635 return 0;
636 }
637 mntget(parent);
638 mountpoint=dget((*mnt)->mnt_mountpoint);
639 spin_unlock(&vfsmount_lock);
640 dput(*dentry);
641 *dentry = mountpoint;
642 mntput(*mnt);
643 *mnt = parent;
644 return 1;
645}
646
647/* no need for dcache_lock, as serialization is taken care in
648 * namespace.c
649 */
Al Viro463ffb22005-06-06 13:36:05 -0700650static int __follow_mount(struct path *path)
651{
652 int res = 0;
653 while (d_mountpoint(path->dentry)) {
654 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
655 if (!mounted)
656 break;
657 dput(path->dentry);
658 if (res)
659 mntput(path->mnt);
660 path->mnt = mounted;
661 path->dentry = dget(mounted->mnt_root);
662 res = 1;
663 }
664 return res;
665}
666
Al Viro58c465e2005-06-06 13:36:13 -0700667static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 while (d_mountpoint(*dentry)) {
670 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
671 if (!mounted)
672 break;
Al Viro58c465e2005-06-06 13:36:13 -0700673 dput(*dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 mntput(*mnt);
675 *mnt = mounted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 *dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678}
679
680/* no need for dcache_lock, as serialization is taken care in
681 * namespace.c
682 */
Al Viroe13b2102005-06-06 13:36:06 -0700683int follow_down(struct vfsmount **mnt, struct dentry **dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684{
685 struct vfsmount *mounted;
686
687 mounted = lookup_mnt(*mnt, *dentry);
688 if (mounted) {
Al Viroe13b2102005-06-06 13:36:06 -0700689 dput(*dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 mntput(*mnt);
691 *mnt = mounted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 *dentry = dget(mounted->mnt_root);
693 return 1;
694 }
695 return 0;
696}
697
Arjan van de Venf1662352006-01-14 13:21:31 -0800698static __always_inline void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699{
700 while(1) {
701 struct vfsmount *parent;
Al Viro58c465e2005-06-06 13:36:13 -0700702 struct dentry *old = nd->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703
704 read_lock(&current->fs->lock);
Al Viro58c465e2005-06-06 13:36:13 -0700705 if (nd->dentry == current->fs->root &&
706 nd->mnt == current->fs->rootmnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 read_unlock(&current->fs->lock);
708 break;
709 }
710 read_unlock(&current->fs->lock);
711 spin_lock(&dcache_lock);
Al Viro58c465e2005-06-06 13:36:13 -0700712 if (nd->dentry != nd->mnt->mnt_root) {
713 nd->dentry = dget(nd->dentry->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 spin_unlock(&dcache_lock);
715 dput(old);
716 break;
717 }
718 spin_unlock(&dcache_lock);
719 spin_lock(&vfsmount_lock);
Al Viro58c465e2005-06-06 13:36:13 -0700720 parent = nd->mnt->mnt_parent;
721 if (parent == nd->mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 spin_unlock(&vfsmount_lock);
723 break;
724 }
725 mntget(parent);
Al Viro58c465e2005-06-06 13:36:13 -0700726 nd->dentry = dget(nd->mnt->mnt_mountpoint);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 spin_unlock(&vfsmount_lock);
728 dput(old);
Al Viro58c465e2005-06-06 13:36:13 -0700729 mntput(nd->mnt);
730 nd->mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 }
Al Viro58c465e2005-06-06 13:36:13 -0700732 follow_mount(&nd->mnt, &nd->dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733}
734
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735/*
736 * It's more convoluted than I'd like it to be, but... it's still fairly
737 * small and for now I'd prefer to have fast path as straight as possible.
738 * It _is_ time-critical.
739 */
740static int do_lookup(struct nameidata *nd, struct qstr *name,
741 struct path *path)
742{
743 struct vfsmount *mnt = nd->mnt;
744 struct dentry *dentry = __d_lookup(nd->dentry, name);
745
746 if (!dentry)
747 goto need_lookup;
748 if (dentry->d_op && dentry->d_op->d_revalidate)
749 goto need_revalidate;
750done:
751 path->mnt = mnt;
752 path->dentry = dentry;
Al Viro634ee702005-06-06 13:36:13 -0700753 __follow_mount(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 return 0;
755
756need_lookup:
757 dentry = real_lookup(nd->dentry, name, nd);
758 if (IS_ERR(dentry))
759 goto fail;
760 goto done;
761
762need_revalidate:
763 if (dentry->d_op->d_revalidate(dentry, nd))
764 goto done;
765 if (d_invalidate(dentry))
766 goto done;
767 dput(dentry);
768 goto need_lookup;
769
770fail:
771 return PTR_ERR(dentry);
772}
773
774/*
775 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +0100776 * This is the basic name resolution function, turning a pathname into
777 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 *
Prasanna Medaea3834d2005-04-29 16:00:17 +0100779 * Returns 0 and nd will have valid dentry and mnt on success.
780 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 */
782static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
783{
784 struct path next;
785 struct inode *inode;
786 int err;
787 unsigned int lookup_flags = nd->flags;
788
789 while (*name=='/')
790 name++;
791 if (!*name)
792 goto return_reval;
793
794 inode = nd->dentry->d_inode;
795 if (nd->depth)
Trond Myklebustf55eab82006-02-04 23:28:01 -0800796 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797
798 /* At this point we know we have a real path component. */
799 for(;;) {
800 unsigned long hash;
801 struct qstr this;
802 unsigned int c;
803
Trond Myklebustcdce5d62005-10-18 14:20:18 -0700804 nd->flags |= LOOKUP_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 err = exec_permission_lite(inode, nd);
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800806 if (err == -EAGAIN)
807 err = vfs_permission(nd, MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 if (err)
809 break;
810
811 this.name = name;
812 c = *(const unsigned char *)name;
813
814 hash = init_name_hash();
815 do {
816 name++;
817 hash = partial_name_hash(c, hash);
818 c = *(const unsigned char *)name;
819 } while (c && (c != '/'));
820 this.len = name - (const char *) this.name;
821 this.hash = end_name_hash(hash);
822
823 /* remove trailing slashes? */
824 if (!c)
825 goto last_component;
826 while (*++name == '/');
827 if (!*name)
828 goto last_with_slashes;
829
830 /*
831 * "." and ".." are special - ".." especially so because it has
832 * to be able to know about the current root directory and
833 * parent relationships.
834 */
835 if (this.name[0] == '.') switch (this.len) {
836 default:
837 break;
838 case 2:
839 if (this.name[1] != '.')
840 break;
Al Viro58c465e2005-06-06 13:36:13 -0700841 follow_dotdot(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 inode = nd->dentry->d_inode;
843 /* fallthrough */
844 case 1:
845 continue;
846 }
847 /*
848 * See if the low-level filesystem might want
849 * to use its own hash..
850 */
851 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
852 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
853 if (err < 0)
854 break;
855 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 /* This does the actual lookups.. */
857 err = do_lookup(nd, &this, &next);
858 if (err)
859 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860
861 err = -ENOENT;
862 inode = next.dentry->d_inode;
863 if (!inode)
864 goto out_dput;
865 err = -ENOTDIR;
866 if (!inode->i_op)
867 goto out_dput;
868
869 if (inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700870 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 if (err)
872 goto return_err;
873 err = -ENOENT;
874 inode = nd->dentry->d_inode;
875 if (!inode)
876 break;
877 err = -ENOTDIR;
878 if (!inode->i_op)
879 break;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700880 } else
881 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 err = -ENOTDIR;
883 if (!inode->i_op->lookup)
884 break;
885 continue;
886 /* here ends the main loop */
887
888last_with_slashes:
889 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
890last_component:
Trond Myklebustf55eab82006-02-04 23:28:01 -0800891 /* Clear LOOKUP_CONTINUE iff it was previously unset */
892 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 if (lookup_flags & LOOKUP_PARENT)
894 goto lookup_parent;
895 if (this.name[0] == '.') switch (this.len) {
896 default:
897 break;
898 case 2:
899 if (this.name[1] != '.')
900 break;
Al Viro58c465e2005-06-06 13:36:13 -0700901 follow_dotdot(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 inode = nd->dentry->d_inode;
903 /* fallthrough */
904 case 1:
905 goto return_reval;
906 }
907 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
908 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
909 if (err < 0)
910 break;
911 }
912 err = do_lookup(nd, &this, &next);
913 if (err)
914 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 inode = next.dentry->d_inode;
916 if ((lookup_flags & LOOKUP_FOLLOW)
917 && inode && inode->i_op && inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700918 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 if (err)
920 goto return_err;
921 inode = nd->dentry->d_inode;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700922 } else
923 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 err = -ENOENT;
925 if (!inode)
926 break;
927 if (lookup_flags & LOOKUP_DIRECTORY) {
928 err = -ENOTDIR;
929 if (!inode->i_op || !inode->i_op->lookup)
930 break;
931 }
932 goto return_base;
933lookup_parent:
934 nd->last = this;
935 nd->last_type = LAST_NORM;
936 if (this.name[0] != '.')
937 goto return_base;
938 if (this.len == 1)
939 nd->last_type = LAST_DOT;
940 else if (this.len == 2 && this.name[1] == '.')
941 nd->last_type = LAST_DOTDOT;
942 else
943 goto return_base;
944return_reval:
945 /*
946 * We bypassed the ordinary revalidation routines.
947 * We may need to check the cached dentry for staleness.
948 */
949 if (nd->dentry && nd->dentry->d_sb &&
950 (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
951 err = -ESTALE;
952 /* Note: we do not d_invalidate() */
953 if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
954 break;
955 }
956return_base:
957 return 0;
958out_dput:
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700959 dput_path(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 break;
961 }
962 path_release(nd);
963return_err:
964 return err;
965}
966
967/*
968 * Wrapper to retry pathname resolution whenever the underlying
969 * file system returns an ESTALE.
970 *
971 * Retry the whole path once, forcing real lookup requests
972 * instead of relying on the dcache.
973 */
974int fastcall link_path_walk(const char *name, struct nameidata *nd)
975{
976 struct nameidata save = *nd;
977 int result;
978
979 /* make sure the stuff we saved doesn't go away */
980 dget(save.dentry);
981 mntget(save.mnt);
982
983 result = __link_path_walk(name, nd);
984 if (result == -ESTALE) {
985 *nd = save;
986 dget(nd->dentry);
987 mntget(nd->mnt);
988 nd->flags |= LOOKUP_REVAL;
989 result = __link_path_walk(name, nd);
990 }
991
992 dput(save.dentry);
993 mntput(save.mnt);
994
995 return result;
996}
997
998int fastcall path_walk(const char * name, struct nameidata *nd)
999{
1000 current->total_link_count = 0;
1001 return link_path_walk(name, nd);
1002}
1003
Prasanna Medaea3834d2005-04-29 16:00:17 +01001004/*
1005 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
1006 * everything is done. Returns 0 and drops input nd, if lookup failed;
1007 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
1009{
1010 if (path_walk(name, nd))
1011 return 0; /* something went wrong... */
1012
1013 if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
1014 struct dentry *old_dentry = nd->dentry;
1015 struct vfsmount *old_mnt = nd->mnt;
1016 struct qstr last = nd->last;
1017 int last_type = nd->last_type;
1018 /*
1019 * NAME was not found in alternate root or it's a directory. Try to find
1020 * it in the normal root:
1021 */
1022 nd->last_type = LAST_ROOT;
1023 read_lock(&current->fs->lock);
1024 nd->mnt = mntget(current->fs->rootmnt);
1025 nd->dentry = dget(current->fs->root);
1026 read_unlock(&current->fs->lock);
1027 if (path_walk(name, nd) == 0) {
1028 if (nd->dentry->d_inode) {
1029 dput(old_dentry);
1030 mntput(old_mnt);
1031 return 1;
1032 }
1033 path_release(nd);
1034 }
1035 nd->dentry = old_dentry;
1036 nd->mnt = old_mnt;
1037 nd->last = last;
1038 nd->last_type = last_type;
1039 }
1040 return 1;
1041}
1042
1043void set_fs_altroot(void)
1044{
1045 char *emul = __emul_prefix();
1046 struct nameidata nd;
1047 struct vfsmount *mnt = NULL, *oldmnt;
1048 struct dentry *dentry = NULL, *olddentry;
1049 int err;
1050
1051 if (!emul)
1052 goto set_it;
1053 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
1054 if (!err) {
1055 mnt = nd.mnt;
1056 dentry = nd.dentry;
1057 }
1058set_it:
1059 write_lock(&current->fs->lock);
1060 oldmnt = current->fs->altrootmnt;
1061 olddentry = current->fs->altroot;
1062 current->fs->altrootmnt = mnt;
1063 current->fs->altroot = dentry;
1064 write_unlock(&current->fs->lock);
1065 if (olddentry) {
1066 dput(olddentry);
1067 mntput(oldmnt);
1068 }
1069}
1070
Prasanna Medaea3834d2005-04-29 16:00:17 +01001071/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001072static int fastcall do_path_lookup(int dfd, const char *name,
1073 unsigned int flags, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001075 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001076 int fput_needed;
1077 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078
1079 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1080 nd->flags = flags;
1081 nd->depth = 0;
1082
1083 read_lock(&current->fs->lock);
1084 if (*name=='/') {
1085 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
1086 nd->mnt = mntget(current->fs->altrootmnt);
1087 nd->dentry = dget(current->fs->altroot);
1088 read_unlock(&current->fs->lock);
1089 if (__emul_lookup_dentry(name,nd))
Prasanna Medaea3834d2005-04-29 16:00:17 +01001090 goto out; /* found in altroot */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 read_lock(&current->fs->lock);
1092 }
1093 nd->mnt = mntget(current->fs->rootmnt);
1094 nd->dentry = dget(current->fs->root);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001095 } else if (dfd == AT_FDCWD) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 nd->mnt = mntget(current->fs->pwdmnt);
1097 nd->dentry = dget(current->fs->pwd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001098 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001099 struct dentry *dentry;
1100
1101 file = fget_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001102 retval = -EBADF;
1103 if (!file)
1104 goto unlock_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001105
1106 dentry = file->f_dentry;
1107
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001108 retval = -ENOTDIR;
1109 if (!S_ISDIR(dentry->d_inode->i_mode))
1110 goto fput_unlock_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001111
1112 retval = file_permission(file, MAY_EXEC);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001113 if (retval)
1114 goto fput_unlock_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001115
1116 nd->mnt = mntget(file->f_vfsmnt);
1117 nd->dentry = dget(dentry);
1118
1119 fput_light(file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 }
1121 read_unlock(&current->fs->lock);
1122 current->total_link_count = 0;
1123 retval = link_path_walk(name, nd);
Prasanna Medaea3834d2005-04-29 16:00:17 +01001124out:
Suzuki3bc84142006-02-07 12:58:36 -08001125 if (likely(retval == 0)) {
1126 if (unlikely(current->audit_context && nd && nd->dentry &&
1127 nd->dentry->d_inode))
David Woodhouseae7b9612005-06-20 16:11:05 +01001128 audit_inode(name, nd->dentry->d_inode, flags);
Suzuki3bc84142006-02-07 12:58:36 -08001129 }
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001130 return retval;
1131
1132fput_unlock_fail:
1133 fput_light(file, fput_needed);
1134unlock_fail:
1135 read_unlock(&current->fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 return retval;
1137}
1138
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001139int fastcall path_lookup(const char *name, unsigned int flags,
1140 struct nameidata *nd)
1141{
1142 return do_path_lookup(AT_FDCWD, name, flags, nd);
1143}
1144
1145static int __path_lookup_intent_open(int dfd, const char *name,
1146 unsigned int lookup_flags, struct nameidata *nd,
1147 int open_flags, int create_mode)
Trond Myklebust834f2a42005-10-18 14:20:16 -07001148{
1149 struct file *filp = get_empty_filp();
1150 int err;
1151
1152 if (filp == NULL)
1153 return -ENFILE;
1154 nd->intent.open.file = filp;
1155 nd->intent.open.flags = open_flags;
1156 nd->intent.open.create_mode = create_mode;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001157 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
Trond Myklebust834f2a42005-10-18 14:20:16 -07001158 if (IS_ERR(nd->intent.open.file)) {
1159 if (err == 0) {
1160 err = PTR_ERR(nd->intent.open.file);
1161 path_release(nd);
1162 }
1163 } else if (err != 0)
1164 release_open_intent(nd);
1165 return err;
1166}
1167
1168/**
1169 * path_lookup_open - lookup a file path with open intent
Martin Waitz7045f372006-02-01 03:06:57 -08001170 * @dfd: the directory to use as base, or AT_FDCWD
Trond Myklebust834f2a42005-10-18 14:20:16 -07001171 * @name: pointer to file name
1172 * @lookup_flags: lookup intent flags
1173 * @nd: pointer to nameidata
1174 * @open_flags: open intent flags
1175 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001176int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
Trond Myklebust834f2a42005-10-18 14:20:16 -07001177 struct nameidata *nd, int open_flags)
1178{
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001179 return __path_lookup_intent_open(dfd, name, lookup_flags, nd,
Trond Myklebust834f2a42005-10-18 14:20:16 -07001180 open_flags, 0);
1181}
1182
1183/**
1184 * path_lookup_create - lookup a file path with open + create intent
Martin Waitz7045f372006-02-01 03:06:57 -08001185 * @dfd: the directory to use as base, or AT_FDCWD
Trond Myklebust834f2a42005-10-18 14:20:16 -07001186 * @name: pointer to file name
1187 * @lookup_flags: lookup intent flags
1188 * @nd: pointer to nameidata
1189 * @open_flags: open intent flags
1190 * @create_mode: create intent flags
1191 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001192static int path_lookup_create(int dfd, const char *name,
1193 unsigned int lookup_flags, struct nameidata *nd,
1194 int open_flags, int create_mode)
Trond Myklebust834f2a42005-10-18 14:20:16 -07001195{
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001196 return __path_lookup_intent_open(dfd, name, lookup_flags|LOOKUP_CREATE,
1197 nd, open_flags, create_mode);
Trond Myklebust834f2a42005-10-18 14:20:16 -07001198}
1199
1200int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags,
1201 struct nameidata *nd, int open_flags)
1202{
1203 char *tmp = getname(name);
1204 int err = PTR_ERR(tmp);
1205
1206 if (!IS_ERR(tmp)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001207 err = __path_lookup_intent_open(AT_FDCWD, tmp, lookup_flags, nd, open_flags, 0);
Trond Myklebust834f2a42005-10-18 14:20:16 -07001208 putname(tmp);
1209 }
1210 return err;
1211}
1212
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213/*
1214 * Restricted form of lookup. Doesn't follow links, single-component only,
1215 * needs parent already locked. Doesn't follow mounts.
1216 * SMP-safe.
1217 */
1218static struct dentry * __lookup_hash(struct qstr *name, struct dentry * base, struct nameidata *nd)
1219{
1220 struct dentry * dentry;
1221 struct inode *inode;
1222 int err;
1223
1224 inode = base->d_inode;
1225 err = permission(inode, MAY_EXEC, nd);
1226 dentry = ERR_PTR(err);
1227 if (err)
1228 goto out;
1229
1230 /*
1231 * See if the low-level filesystem might want
1232 * to use its own hash..
1233 */
1234 if (base->d_op && base->d_op->d_hash) {
1235 err = base->d_op->d_hash(base, name);
1236 dentry = ERR_PTR(err);
1237 if (err < 0)
1238 goto out;
1239 }
1240
1241 dentry = cached_lookup(base, name, nd);
1242 if (!dentry) {
1243 struct dentry *new = d_alloc(base, name);
1244 dentry = ERR_PTR(-ENOMEM);
1245 if (!new)
1246 goto out;
1247 dentry = inode->i_op->lookup(inode, new, nd);
1248 if (!dentry)
1249 dentry = new;
1250 else
1251 dput(new);
1252 }
1253out:
1254 return dentry;
1255}
1256
Adrian Bunka244e162006-03-31 02:32:11 -08001257static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258{
Christoph Hellwig49705b72005-11-08 21:35:06 -08001259 return __lookup_hash(&nd->last, nd->dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260}
1261
1262/* SMP-safe */
1263struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
1264{
1265 unsigned long hash;
1266 struct qstr this;
1267 unsigned int c;
1268
1269 this.name = name;
1270 this.len = len;
1271 if (!len)
1272 goto access;
1273
1274 hash = init_name_hash();
1275 while (len--) {
1276 c = *(const unsigned char *)name++;
1277 if (c == '/' || c == '\0')
1278 goto access;
1279 hash = partial_name_hash(c, hash);
1280 }
1281 this.hash = end_name_hash(hash);
1282
Christoph Hellwig49705b72005-11-08 21:35:06 -08001283 return __lookup_hash(&this, base, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284access:
1285 return ERR_PTR(-EACCES);
1286}
1287
1288/*
1289 * namei()
1290 *
1291 * is used by most simple commands to get the inode of a specified name.
1292 * Open, link etc use their own routines, but this is enough for things
1293 * like 'chmod' etc.
1294 *
1295 * namei exists in two versions: namei/lnamei. The only difference is
1296 * that namei follows links, while lnamei does not.
1297 * SMP-safe
1298 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001299int fastcall __user_walk_fd(int dfd, const char __user *name, unsigned flags,
1300 struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301{
1302 char *tmp = getname(name);
1303 int err = PTR_ERR(tmp);
1304
1305 if (!IS_ERR(tmp)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001306 err = do_path_lookup(dfd, tmp, flags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 putname(tmp);
1308 }
1309 return err;
1310}
1311
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001312int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1313{
1314 return __user_walk_fd(AT_FDCWD, name, flags, nd);
1315}
1316
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317/*
1318 * It's inline, so penalty for filesystems that don't use sticky bit is
1319 * minimal.
1320 */
1321static inline int check_sticky(struct inode *dir, struct inode *inode)
1322{
1323 if (!(dir->i_mode & S_ISVTX))
1324 return 0;
1325 if (inode->i_uid == current->fsuid)
1326 return 0;
1327 if (dir->i_uid == current->fsuid)
1328 return 0;
1329 return !capable(CAP_FOWNER);
1330}
1331
1332/*
1333 * Check whether we can remove a link victim from directory dir, check
1334 * whether the type of victim is right.
1335 * 1. We can't do it if dir is read-only (done in permission())
1336 * 2. We should have write and exec permissions on dir
1337 * 3. We can't remove anything from append-only dir
1338 * 4. We can't do anything with immutable dir (done in permission())
1339 * 5. If the sticky bit on dir is set we should either
1340 * a. be owner of dir, or
1341 * b. be owner of victim, or
1342 * c. have CAP_FOWNER capability
1343 * 6. If the victim is append-only or immutable we can't do antyhing with
1344 * links pointing to it.
1345 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1346 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1347 * 9. We can't remove a root or mountpoint.
1348 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1349 * nfs_async_unlink().
1350 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001351static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352{
1353 int error;
1354
1355 if (!victim->d_inode)
1356 return -ENOENT;
1357
1358 BUG_ON(victim->d_parent->d_inode != dir);
Amy Griffis73241cc2005-11-03 16:00:25 +00001359 audit_inode_child(victim->d_name.name, victim->d_inode, dir->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360
1361 error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
1362 if (error)
1363 return error;
1364 if (IS_APPEND(dir))
1365 return -EPERM;
1366 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1367 IS_IMMUTABLE(victim->d_inode))
1368 return -EPERM;
1369 if (isdir) {
1370 if (!S_ISDIR(victim->d_inode->i_mode))
1371 return -ENOTDIR;
1372 if (IS_ROOT(victim))
1373 return -EBUSY;
1374 } else if (S_ISDIR(victim->d_inode->i_mode))
1375 return -EISDIR;
1376 if (IS_DEADDIR(dir))
1377 return -ENOENT;
1378 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1379 return -EBUSY;
1380 return 0;
1381}
1382
1383/* Check whether we can create an object with dentry child in directory
1384 * dir.
1385 * 1. We can't do it if child already exists (open has special treatment for
1386 * this case, but since we are inlined it's OK)
1387 * 2. We can't do it if dir is read-only (done in permission())
1388 * 3. We should have write and exec permissions on dir
1389 * 4. We can't do it if dir is immutable (done in permission())
1390 */
1391static inline int may_create(struct inode *dir, struct dentry *child,
1392 struct nameidata *nd)
1393{
1394 if (child->d_inode)
1395 return -EEXIST;
1396 if (IS_DEADDIR(dir))
1397 return -ENOENT;
1398 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1399}
1400
1401/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 * O_DIRECTORY translates into forcing a directory lookup.
1403 */
1404static inline int lookup_flags(unsigned int f)
1405{
1406 unsigned long retval = LOOKUP_FOLLOW;
1407
1408 if (f & O_NOFOLLOW)
1409 retval &= ~LOOKUP_FOLLOW;
1410
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 if (f & O_DIRECTORY)
1412 retval |= LOOKUP_DIRECTORY;
1413
1414 return retval;
1415}
1416
1417/*
1418 * p1 and p2 should be directories on the same fs.
1419 */
1420struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1421{
1422 struct dentry *p;
1423
1424 if (p1 == p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001425 mutex_lock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 return NULL;
1427 }
1428
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001429 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430
1431 for (p = p1; p->d_parent != p; p = p->d_parent) {
1432 if (p->d_parent == p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001433 mutex_lock(&p2->d_inode->i_mutex);
1434 mutex_lock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 return p;
1436 }
1437 }
1438
1439 for (p = p2; p->d_parent != p; p = p->d_parent) {
1440 if (p->d_parent == p1) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001441 mutex_lock(&p1->d_inode->i_mutex);
1442 mutex_lock(&p2->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 return p;
1444 }
1445 }
1446
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001447 mutex_lock(&p1->d_inode->i_mutex);
1448 mutex_lock(&p2->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 return NULL;
1450}
1451
1452void unlock_rename(struct dentry *p1, struct dentry *p2)
1453{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001454 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001456 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001457 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 }
1459}
1460
1461int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1462 struct nameidata *nd)
1463{
1464 int error = may_create(dir, dentry, nd);
1465
1466 if (error)
1467 return error;
1468
1469 if (!dir->i_op || !dir->i_op->create)
1470 return -EACCES; /* shouldn't it be ENOSYS? */
1471 mode &= S_IALLUGO;
1472 mode |= S_IFREG;
1473 error = security_inode_create(dir, dentry, mode);
1474 if (error)
1475 return error;
1476 DQUOT_INIT(dir);
1477 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001478 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001479 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 return error;
1481}
1482
1483int may_open(struct nameidata *nd, int acc_mode, int flag)
1484{
1485 struct dentry *dentry = nd->dentry;
1486 struct inode *inode = dentry->d_inode;
1487 int error;
1488
1489 if (!inode)
1490 return -ENOENT;
1491
1492 if (S_ISLNK(inode->i_mode))
1493 return -ELOOP;
1494
1495 if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1496 return -EISDIR;
1497
Christoph Hellwige4543ed2005-11-08 21:35:04 -08001498 error = vfs_permission(nd, acc_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 if (error)
1500 return error;
1501
1502 /*
1503 * FIFO's, sockets and device files are special: they don't
1504 * actually live on the filesystem itself, and as such you
1505 * can write to them even if the filesystem is read-only.
1506 */
1507 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1508 flag &= ~O_TRUNC;
1509 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1510 if (nd->mnt->mnt_flags & MNT_NODEV)
1511 return -EACCES;
1512
1513 flag &= ~O_TRUNC;
1514 } else if (IS_RDONLY(inode) && (flag & FMODE_WRITE))
1515 return -EROFS;
1516 /*
1517 * An append-only file must be opened in append mode for writing.
1518 */
1519 if (IS_APPEND(inode)) {
1520 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1521 return -EPERM;
1522 if (flag & O_TRUNC)
1523 return -EPERM;
1524 }
1525
1526 /* O_NOATIME can only be set by the owner or superuser */
1527 if (flag & O_NOATIME)
1528 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1529 return -EPERM;
1530
1531 /*
1532 * Ensure there are no outstanding leases on the file.
1533 */
1534 error = break_lease(inode, flag);
1535 if (error)
1536 return error;
1537
1538 if (flag & O_TRUNC) {
1539 error = get_write_access(inode);
1540 if (error)
1541 return error;
1542
1543 /*
1544 * Refuse to truncate files with mandatory locks held on them.
1545 */
1546 error = locks_verify_locked(inode);
1547 if (!error) {
1548 DQUOT_INIT(inode);
1549
NeilBrown4a301312006-01-08 01:02:39 -08001550 error = do_truncate(dentry, 0, ATTR_MTIME|ATTR_CTIME, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 }
1552 put_write_access(inode);
1553 if (error)
1554 return error;
1555 } else
1556 if (flag & FMODE_WRITE)
1557 DQUOT_INIT(inode);
1558
1559 return 0;
1560}
1561
1562/*
1563 * open_namei()
1564 *
1565 * namei for open - this is in fact almost the whole open-routine.
1566 *
1567 * Note that the low bits of "flag" aren't the same as in the open
1568 * system call - they are 00 - no permissions needed
1569 * 01 - read permission needed
1570 * 10 - write permission needed
1571 * 11 - read/write permissions needed
1572 * which is a lot more logical, and also allows the "no perm" needed
1573 * for symlinks (where the permissions are checked later).
1574 * SMP-safe
1575 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001576int open_namei(int dfd, const char *pathname, int flag,
1577 int mode, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578{
Trond Myklebust834f2a42005-10-18 14:20:16 -07001579 int acc_mode, error;
Al Viro4e7506e2005-06-06 13:36:00 -07001580 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 struct dentry *dir;
1582 int count = 0;
1583
1584 acc_mode = ACC_MODE(flag);
1585
Trond Myklebust834f2a42005-10-18 14:20:16 -07001586 /* O_TRUNC implies we need access checks for write permissions */
1587 if (flag & O_TRUNC)
1588 acc_mode |= MAY_WRITE;
1589
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 /* Allow the LSM permission hook to distinguish append
1591 access from general write access. */
1592 if (flag & O_APPEND)
1593 acc_mode |= MAY_APPEND;
1594
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 /*
1596 * The simplest case - just a plain lookup.
1597 */
1598 if (!(flag & O_CREAT)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001599 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1600 nd, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 if (error)
1602 return error;
1603 goto ok;
1604 }
1605
1606 /*
1607 * Create - we need to know the parent.
1608 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001609 error = path_lookup_create(dfd,pathname,LOOKUP_PARENT,nd,flag,mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 if (error)
1611 return error;
1612
1613 /*
1614 * We have the parent and last component. First of all, check
1615 * that we are not asked to creat(2) an obvious directory - that
1616 * will not do.
1617 */
1618 error = -EISDIR;
1619 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1620 goto exit;
1621
1622 dir = nd->dentry;
1623 nd->flags &= ~LOOKUP_PARENT;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001624 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001625 path.dentry = lookup_hash(nd);
Al Virod73ffe12005-06-06 13:36:01 -07001626 path.mnt = nd->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627
1628do_last:
Al Viro4e7506e2005-06-06 13:36:00 -07001629 error = PTR_ERR(path.dentry);
1630 if (IS_ERR(path.dentry)) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001631 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 goto exit;
1633 }
1634
Oleg Drokin4af4c522006-03-25 03:06:54 -08001635 if (IS_ERR(nd->intent.open.file)) {
1636 mutex_unlock(&dir->d_inode->i_mutex);
1637 error = PTR_ERR(nd->intent.open.file);
1638 goto exit_dput;
1639 }
1640
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 /* Negative dentry, just create the file */
Al Viro4e7506e2005-06-06 13:36:00 -07001642 if (!path.dentry->d_inode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 if (!IS_POSIXACL(dir->d_inode))
1644 mode &= ~current->fs->umask;
Al Viro4e7506e2005-06-06 13:36:00 -07001645 error = vfs_create(dir->d_inode, path.dentry, mode, nd);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001646 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 dput(nd->dentry);
Al Viro4e7506e2005-06-06 13:36:00 -07001648 nd->dentry = path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 if (error)
1650 goto exit;
1651 /* Don't check for write permission, don't truncate */
1652 acc_mode = 0;
1653 flag &= ~O_TRUNC;
1654 goto ok;
1655 }
1656
1657 /*
1658 * It already exists.
1659 */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001660 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661
1662 error = -EEXIST;
1663 if (flag & O_EXCL)
1664 goto exit_dput;
1665
Al Viroe13b2102005-06-06 13:36:06 -07001666 if (__follow_mount(&path)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 error = -ELOOP;
Al Viroba7a4c12005-06-06 13:36:08 -07001668 if (flag & O_NOFOLLOW)
1669 goto exit_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 }
1671 error = -ENOENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001672 if (!path.dentry->d_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 goto exit_dput;
Al Viro4e7506e2005-06-06 13:36:00 -07001674 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 goto do_link;
1676
Miklos Szeredi09dd17d2005-09-06 15:18:21 -07001677 path_to_nameidata(&path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 error = -EISDIR;
Al Viro4e7506e2005-06-06 13:36:00 -07001679 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 goto exit;
1681ok:
1682 error = may_open(nd, acc_mode, flag);
1683 if (error)
1684 goto exit;
1685 return 0;
1686
1687exit_dput:
Miklos Szeredi09dd17d2005-09-06 15:18:21 -07001688 dput_path(&path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689exit:
Trond Myklebust834f2a42005-10-18 14:20:16 -07001690 if (!IS_ERR(nd->intent.open.file))
1691 release_open_intent(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 path_release(nd);
1693 return error;
1694
1695do_link:
1696 error = -ELOOP;
1697 if (flag & O_NOFOLLOW)
1698 goto exit_dput;
1699 /*
1700 * This is subtle. Instead of calling do_follow_link() we do the
1701 * thing by hands. The reason is that this way we have zero link_count
1702 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1703 * After that we have the parent and last component, i.e.
1704 * we are in the same situation as after the first path_walk().
1705 * Well, almost - if the last component is normal we get its copy
1706 * stored in nd->last.name and we will have to putname() it when we
1707 * are done. Procfs-like symlinks just set LAST_BIND.
1708 */
1709 nd->flags |= LOOKUP_PARENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001710 error = security_inode_follow_link(path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 if (error)
1712 goto exit_dput;
Al Virocd4e91d2005-06-06 13:36:03 -07001713 error = __do_follow_link(&path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 if (error)
1715 return error;
1716 nd->flags &= ~LOOKUP_PARENT;
Al Virod671d5e2005-06-06 13:36:04 -07001717 if (nd->last_type == LAST_BIND)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 goto ok;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 error = -EISDIR;
1720 if (nd->last_type != LAST_NORM)
1721 goto exit;
1722 if (nd->last.name[nd->last.len]) {
Linus Torvalds82984112005-10-06 21:54:21 -07001723 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 goto exit;
1725 }
1726 error = -ELOOP;
1727 if (count++==32) {
Linus Torvalds82984112005-10-06 21:54:21 -07001728 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 goto exit;
1730 }
1731 dir = nd->dentry;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001732 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001733 path.dentry = lookup_hash(nd);
Al Virod671d5e2005-06-06 13:36:04 -07001734 path.mnt = nd->mnt;
Linus Torvalds82984112005-10-06 21:54:21 -07001735 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 goto do_last;
1737}
1738
1739/**
1740 * lookup_create - lookup a dentry, creating it if it doesn't exist
1741 * @nd: nameidata info
1742 * @is_dir: directory flag
1743 *
1744 * Simple function to lookup and return a dentry and create it
1745 * if it doesn't exist. Is SMP-safe.
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001746 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001747 * Returns with nd->dentry->d_inode->i_mutex locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 */
1749struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1750{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001751 struct dentry *dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001753 mutex_lock(&nd->dentry->d_inode->i_mutex);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001754 /*
1755 * Yucky last component or no last component at all?
1756 * (foo/., foo/.., /////)
1757 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 if (nd->last_type != LAST_NORM)
1759 goto fail;
1760 nd->flags &= ~LOOKUP_PARENT;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001761
1762 /*
1763 * Do the final lookup.
1764 */
Christoph Hellwig49705b72005-11-08 21:35:06 -08001765 dentry = lookup_hash(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 if (IS_ERR(dentry))
1767 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001768
1769 /*
1770 * Special case - lookup gave negative, but... we had foo/bar/
1771 * From the vfs_mknod() POV we just have a negative dentry -
1772 * all is fine. Let's be bastards - you had / on the end, you've
1773 * been asking for (non-existent) directory. -ENOENT for you.
1774 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1776 goto enoent;
1777 return dentry;
1778enoent:
1779 dput(dentry);
1780 dentry = ERR_PTR(-ENOENT);
1781fail:
1782 return dentry;
1783}
Christoph Hellwigf81a0bf2005-05-19 12:26:43 -07001784EXPORT_SYMBOL_GPL(lookup_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785
1786int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1787{
1788 int error = may_create(dir, dentry, NULL);
1789
1790 if (error)
1791 return error;
1792
1793 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1794 return -EPERM;
1795
1796 if (!dir->i_op || !dir->i_op->mknod)
1797 return -EPERM;
1798
1799 error = security_inode_mknod(dir, dentry, mode, dev);
1800 if (error)
1801 return error;
1802
1803 DQUOT_INIT(dir);
1804 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001805 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001806 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807 return error;
1808}
1809
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001810asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
1811 unsigned dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812{
1813 int error = 0;
1814 char * tmp;
1815 struct dentry * dentry;
1816 struct nameidata nd;
1817
1818 if (S_ISDIR(mode))
1819 return -EPERM;
1820 tmp = getname(filename);
1821 if (IS_ERR(tmp))
1822 return PTR_ERR(tmp);
1823
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001824 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 if (error)
1826 goto out;
1827 dentry = lookup_create(&nd, 0);
1828 error = PTR_ERR(dentry);
1829
1830 if (!IS_POSIXACL(nd.dentry->d_inode))
1831 mode &= ~current->fs->umask;
1832 if (!IS_ERR(dentry)) {
1833 switch (mode & S_IFMT) {
1834 case 0: case S_IFREG:
1835 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1836 break;
1837 case S_IFCHR: case S_IFBLK:
1838 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,
1839 new_decode_dev(dev));
1840 break;
1841 case S_IFIFO: case S_IFSOCK:
1842 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,0);
1843 break;
1844 case S_IFDIR:
1845 error = -EPERM;
1846 break;
1847 default:
1848 error = -EINVAL;
1849 }
1850 dput(dentry);
1851 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001852 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 path_release(&nd);
1854out:
1855 putname(tmp);
1856
1857 return error;
1858}
1859
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001860asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
1861{
1862 return sys_mknodat(AT_FDCWD, filename, mode, dev);
1863}
1864
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1866{
1867 int error = may_create(dir, dentry, NULL);
1868
1869 if (error)
1870 return error;
1871
1872 if (!dir->i_op || !dir->i_op->mkdir)
1873 return -EPERM;
1874
1875 mode &= (S_IRWXUGO|S_ISVTX);
1876 error = security_inode_mkdir(dir, dentry, mode);
1877 if (error)
1878 return error;
1879
1880 DQUOT_INIT(dir);
1881 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001882 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001883 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884 return error;
1885}
1886
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001887asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888{
1889 int error = 0;
1890 char * tmp;
1891
1892 tmp = getname(pathname);
1893 error = PTR_ERR(tmp);
1894 if (!IS_ERR(tmp)) {
1895 struct dentry *dentry;
1896 struct nameidata nd;
1897
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001898 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 if (error)
1900 goto out;
1901 dentry = lookup_create(&nd, 1);
1902 error = PTR_ERR(dentry);
1903 if (!IS_ERR(dentry)) {
1904 if (!IS_POSIXACL(nd.dentry->d_inode))
1905 mode &= ~current->fs->umask;
1906 error = vfs_mkdir(nd.dentry->d_inode, dentry, mode);
1907 dput(dentry);
1908 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001909 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910 path_release(&nd);
1911out:
1912 putname(tmp);
1913 }
1914
1915 return error;
1916}
1917
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001918asmlinkage long sys_mkdir(const char __user *pathname, int mode)
1919{
1920 return sys_mkdirat(AT_FDCWD, pathname, mode);
1921}
1922
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923/*
1924 * We try to drop the dentry early: we should have
1925 * a usage count of 2 if we're the only user of this
1926 * dentry, and if that is true (possibly after pruning
1927 * the dcache), then we drop the dentry now.
1928 *
1929 * A low-level filesystem can, if it choses, legally
1930 * do a
1931 *
1932 * if (!d_unhashed(dentry))
1933 * return -EBUSY;
1934 *
1935 * if it cannot handle the case of removing a directory
1936 * that is still in use by something else..
1937 */
1938void dentry_unhash(struct dentry *dentry)
1939{
1940 dget(dentry);
1941 if (atomic_read(&dentry->d_count))
1942 shrink_dcache_parent(dentry);
1943 spin_lock(&dcache_lock);
1944 spin_lock(&dentry->d_lock);
1945 if (atomic_read(&dentry->d_count) == 2)
1946 __d_drop(dentry);
1947 spin_unlock(&dentry->d_lock);
1948 spin_unlock(&dcache_lock);
1949}
1950
1951int vfs_rmdir(struct inode *dir, struct dentry *dentry)
1952{
1953 int error = may_delete(dir, dentry, 1);
1954
1955 if (error)
1956 return error;
1957
1958 if (!dir->i_op || !dir->i_op->rmdir)
1959 return -EPERM;
1960
1961 DQUOT_INIT(dir);
1962
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001963 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964 dentry_unhash(dentry);
1965 if (d_mountpoint(dentry))
1966 error = -EBUSY;
1967 else {
1968 error = security_inode_rmdir(dir, dentry);
1969 if (!error) {
1970 error = dir->i_op->rmdir(dir, dentry);
1971 if (!error)
1972 dentry->d_inode->i_flags |= S_DEAD;
1973 }
1974 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001975 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 d_delete(dentry);
1978 }
1979 dput(dentry);
1980
1981 return error;
1982}
1983
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001984static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985{
1986 int error = 0;
1987 char * name;
1988 struct dentry *dentry;
1989 struct nameidata nd;
1990
1991 name = getname(pathname);
1992 if(IS_ERR(name))
1993 return PTR_ERR(name);
1994
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001995 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 if (error)
1997 goto exit;
1998
1999 switch(nd.last_type) {
2000 case LAST_DOTDOT:
2001 error = -ENOTEMPTY;
2002 goto exit1;
2003 case LAST_DOT:
2004 error = -EINVAL;
2005 goto exit1;
2006 case LAST_ROOT:
2007 error = -EBUSY;
2008 goto exit1;
2009 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002010 mutex_lock(&nd.dentry->d_inode->i_mutex);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002011 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 error = PTR_ERR(dentry);
2013 if (!IS_ERR(dentry)) {
2014 error = vfs_rmdir(nd.dentry->d_inode, dentry);
2015 dput(dentry);
2016 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002017 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018exit1:
2019 path_release(&nd);
2020exit:
2021 putname(name);
2022 return error;
2023}
2024
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002025asmlinkage long sys_rmdir(const char __user *pathname)
2026{
2027 return do_rmdir(AT_FDCWD, pathname);
2028}
2029
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030int vfs_unlink(struct inode *dir, struct dentry *dentry)
2031{
2032 int error = may_delete(dir, dentry, 0);
2033
2034 if (error)
2035 return error;
2036
2037 if (!dir->i_op || !dir->i_op->unlink)
2038 return -EPERM;
2039
2040 DQUOT_INIT(dir);
2041
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002042 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 if (d_mountpoint(dentry))
2044 error = -EBUSY;
2045 else {
2046 error = security_inode_unlink(dir, dentry);
2047 if (!error)
2048 error = dir->i_op->unlink(dir, dentry);
2049 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002050 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051
2052 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2053 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2054 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 }
Robert Love0eeca282005-07-12 17:06:03 -04002056
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057 return error;
2058}
2059
2060/*
2061 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002062 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 * writeout happening, and we don't want to prevent access to the directory
2064 * while waiting on the I/O.
2065 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002066static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067{
2068 int error = 0;
2069 char * name;
2070 struct dentry *dentry;
2071 struct nameidata nd;
2072 struct inode *inode = NULL;
2073
2074 name = getname(pathname);
2075 if(IS_ERR(name))
2076 return PTR_ERR(name);
2077
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002078 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079 if (error)
2080 goto exit;
2081 error = -EISDIR;
2082 if (nd.last_type != LAST_NORM)
2083 goto exit1;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002084 mutex_lock(&nd.dentry->d_inode->i_mutex);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002085 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086 error = PTR_ERR(dentry);
2087 if (!IS_ERR(dentry)) {
2088 /* Why not before? Because we want correct error value */
2089 if (nd.last.name[nd.last.len])
2090 goto slashes;
2091 inode = dentry->d_inode;
2092 if (inode)
2093 atomic_inc(&inode->i_count);
2094 error = vfs_unlink(nd.dentry->d_inode, dentry);
2095 exit2:
2096 dput(dentry);
2097 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002098 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099 if (inode)
2100 iput(inode); /* truncate the inode here */
2101exit1:
2102 path_release(&nd);
2103exit:
2104 putname(name);
2105 return error;
2106
2107slashes:
2108 error = !dentry->d_inode ? -ENOENT :
2109 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2110 goto exit2;
2111}
2112
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002113asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2114{
2115 if ((flag & ~AT_REMOVEDIR) != 0)
2116 return -EINVAL;
2117
2118 if (flag & AT_REMOVEDIR)
2119 return do_rmdir(dfd, pathname);
2120
2121 return do_unlinkat(dfd, pathname);
2122}
2123
2124asmlinkage long sys_unlink(const char __user *pathname)
2125{
2126 return do_unlinkat(AT_FDCWD, pathname);
2127}
2128
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname, int mode)
2130{
2131 int error = may_create(dir, dentry, NULL);
2132
2133 if (error)
2134 return error;
2135
2136 if (!dir->i_op || !dir->i_op->symlink)
2137 return -EPERM;
2138
2139 error = security_inode_symlink(dir, dentry, oldname);
2140 if (error)
2141 return error;
2142
2143 DQUOT_INIT(dir);
2144 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002145 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002146 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147 return error;
2148}
2149
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002150asmlinkage long sys_symlinkat(const char __user *oldname,
2151 int newdfd, const char __user *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152{
2153 int error = 0;
2154 char * from;
2155 char * to;
2156
2157 from = getname(oldname);
2158 if(IS_ERR(from))
2159 return PTR_ERR(from);
2160 to = getname(newname);
2161 error = PTR_ERR(to);
2162 if (!IS_ERR(to)) {
2163 struct dentry *dentry;
2164 struct nameidata nd;
2165
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002166 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167 if (error)
2168 goto out;
2169 dentry = lookup_create(&nd, 0);
2170 error = PTR_ERR(dentry);
2171 if (!IS_ERR(dentry)) {
2172 error = vfs_symlink(nd.dentry->d_inode, dentry, from, S_IALLUGO);
2173 dput(dentry);
2174 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002175 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 path_release(&nd);
2177out:
2178 putname(to);
2179 }
2180 putname(from);
2181 return error;
2182}
2183
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002184asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2185{
2186 return sys_symlinkat(oldname, AT_FDCWD, newname);
2187}
2188
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2190{
2191 struct inode *inode = old_dentry->d_inode;
2192 int error;
2193
2194 if (!inode)
2195 return -ENOENT;
2196
2197 error = may_create(dir, new_dentry, NULL);
2198 if (error)
2199 return error;
2200
2201 if (dir->i_sb != inode->i_sb)
2202 return -EXDEV;
2203
2204 /*
2205 * A link to an append-only or immutable file cannot be created.
2206 */
2207 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2208 return -EPERM;
2209 if (!dir->i_op || !dir->i_op->link)
2210 return -EPERM;
2211 if (S_ISDIR(old_dentry->d_inode->i_mode))
2212 return -EPERM;
2213
2214 error = security_inode_link(old_dentry, dir, new_dentry);
2215 if (error)
2216 return error;
2217
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002218 mutex_lock(&old_dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219 DQUOT_INIT(dir);
2220 error = dir->i_op->link(old_dentry, dir, new_dentry);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002221 mutex_unlock(&old_dentry->d_inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002222 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002223 fsnotify_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224 return error;
2225}
2226
2227/*
2228 * Hardlinks are often used in delicate situations. We avoid
2229 * security-related surprises by not following symlinks on the
2230 * newname. --KAB
2231 *
2232 * We don't follow them on the oldname either to be compatible
2233 * with linux 2.0, and to avoid hard-linking to directories
2234 * and other special files. --ADM
2235 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002236asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002237 int newdfd, const char __user *newname,
2238 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239{
2240 struct dentry *new_dentry;
2241 struct nameidata nd, old_nd;
2242 int error;
2243 char * to;
2244
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002245 if (flags != 0)
2246 return -EINVAL;
2247
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248 to = getname(newname);
2249 if (IS_ERR(to))
2250 return PTR_ERR(to);
2251
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002252 error = __user_walk_fd(olddfd, oldname, 0, &old_nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 if (error)
2254 goto exit;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002255 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256 if (error)
2257 goto out;
2258 error = -EXDEV;
2259 if (old_nd.mnt != nd.mnt)
2260 goto out_release;
2261 new_dentry = lookup_create(&nd, 0);
2262 error = PTR_ERR(new_dentry);
2263 if (!IS_ERR(new_dentry)) {
2264 error = vfs_link(old_nd.dentry, nd.dentry->d_inode, new_dentry);
2265 dput(new_dentry);
2266 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002267 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268out_release:
2269 path_release(&nd);
2270out:
2271 path_release(&old_nd);
2272exit:
2273 putname(to);
2274
2275 return error;
2276}
2277
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002278asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2279{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002280 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002281}
2282
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283/*
2284 * The worst of all namespace operations - renaming directory. "Perverted"
2285 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2286 * Problems:
2287 * a) we can get into loop creation. Check is done in is_subdir().
2288 * b) race potential - two innocent renames can create a loop together.
2289 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002290 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291 * story.
2292 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002293 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 * whether the target exists). Solution: try to be smart with locking
2295 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002296 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297 * move will be locked. Thus we can rank directories by the tree
2298 * (ancestors first) and rank all non-directories after them.
2299 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002300 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301 * HOWEVER, it relies on the assumption that any object with ->lookup()
2302 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2303 * we'd better make sure that there's no link(2) for them.
2304 * d) some filesystems don't support opened-but-unlinked directories,
2305 * either because of layout or because they are not ready to deal with
2306 * all cases correctly. The latter will be fixed (taking this sort of
2307 * stuff into VFS), but the former is not going away. Solution: the same
2308 * trick as in rmdir().
2309 * e) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002310 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002312 * ->i_mutex on parents, which works but leads to some truely excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313 * locking].
2314 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07002315static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2316 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317{
2318 int error = 0;
2319 struct inode *target;
2320
2321 /*
2322 * If we are going to change the parent - check write permissions,
2323 * we'll need to flip '..'.
2324 */
2325 if (new_dir != old_dir) {
2326 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2327 if (error)
2328 return error;
2329 }
2330
2331 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2332 if (error)
2333 return error;
2334
2335 target = new_dentry->d_inode;
2336 if (target) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002337 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338 dentry_unhash(new_dentry);
2339 }
2340 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2341 error = -EBUSY;
2342 else
2343 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2344 if (target) {
2345 if (!error)
2346 target->i_flags |= S_DEAD;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002347 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348 if (d_unhashed(new_dentry))
2349 d_rehash(new_dentry);
2350 dput(new_dentry);
2351 }
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002352 if (!error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354 return error;
2355}
2356
Adrian Bunk75c96f82005-05-05 16:16:09 -07002357static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2358 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359{
2360 struct inode *target;
2361 int error;
2362
2363 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2364 if (error)
2365 return error;
2366
2367 dget(new_dentry);
2368 target = new_dentry->d_inode;
2369 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002370 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2372 error = -EBUSY;
2373 else
2374 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2375 if (!error) {
2376 /* The following d_move() should become unconditional */
2377 if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
2378 d_move(old_dentry, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379 }
2380 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002381 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 dput(new_dentry);
2383 return error;
2384}
2385
2386int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2387 struct inode *new_dir, struct dentry *new_dentry)
2388{
2389 int error;
2390 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Robert Love0eeca282005-07-12 17:06:03 -04002391 const char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392
2393 if (old_dentry->d_inode == new_dentry->d_inode)
2394 return 0;
2395
2396 error = may_delete(old_dir, old_dentry, is_dir);
2397 if (error)
2398 return error;
2399
2400 if (!new_dentry->d_inode)
2401 error = may_create(new_dir, new_dentry, NULL);
2402 else
2403 error = may_delete(new_dir, new_dentry, is_dir);
2404 if (error)
2405 return error;
2406
2407 if (!old_dir->i_op || !old_dir->i_op->rename)
2408 return -EPERM;
2409
2410 DQUOT_INIT(old_dir);
2411 DQUOT_INIT(new_dir);
2412
Robert Love0eeca282005-07-12 17:06:03 -04002413 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2414
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415 if (is_dir)
2416 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2417 else
2418 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2419 if (!error) {
Robert Love0eeca282005-07-12 17:06:03 -04002420 const char *new_name = old_dentry->d_name.name;
John McCutchan89204c42005-08-15 12:13:28 -04002421 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2422 new_dentry->d_inode, old_dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 }
Robert Love0eeca282005-07-12 17:06:03 -04002424 fsnotify_oldname_free(old_name);
2425
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426 return error;
2427}
2428
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002429static int do_rename(int olddfd, const char *oldname,
2430 int newdfd, const char *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431{
2432 int error = 0;
2433 struct dentry * old_dir, * new_dir;
2434 struct dentry * old_dentry, *new_dentry;
2435 struct dentry * trap;
2436 struct nameidata oldnd, newnd;
2437
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002438 error = do_path_lookup(olddfd, oldname, LOOKUP_PARENT, &oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439 if (error)
2440 goto exit;
2441
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002442 error = do_path_lookup(newdfd, newname, LOOKUP_PARENT, &newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443 if (error)
2444 goto exit1;
2445
2446 error = -EXDEV;
2447 if (oldnd.mnt != newnd.mnt)
2448 goto exit2;
2449
2450 old_dir = oldnd.dentry;
2451 error = -EBUSY;
2452 if (oldnd.last_type != LAST_NORM)
2453 goto exit2;
2454
2455 new_dir = newnd.dentry;
2456 if (newnd.last_type != LAST_NORM)
2457 goto exit2;
2458
2459 trap = lock_rename(new_dir, old_dir);
2460
Christoph Hellwig49705b72005-11-08 21:35:06 -08002461 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462 error = PTR_ERR(old_dentry);
2463 if (IS_ERR(old_dentry))
2464 goto exit3;
2465 /* source must exist */
2466 error = -ENOENT;
2467 if (!old_dentry->d_inode)
2468 goto exit4;
2469 /* unless the source is a directory trailing slashes give -ENOTDIR */
2470 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2471 error = -ENOTDIR;
2472 if (oldnd.last.name[oldnd.last.len])
2473 goto exit4;
2474 if (newnd.last.name[newnd.last.len])
2475 goto exit4;
2476 }
2477 /* source should not be ancestor of target */
2478 error = -EINVAL;
2479 if (old_dentry == trap)
2480 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08002481 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482 error = PTR_ERR(new_dentry);
2483 if (IS_ERR(new_dentry))
2484 goto exit4;
2485 /* target should not be an ancestor of source */
2486 error = -ENOTEMPTY;
2487 if (new_dentry == trap)
2488 goto exit5;
2489
2490 error = vfs_rename(old_dir->d_inode, old_dentry,
2491 new_dir->d_inode, new_dentry);
2492exit5:
2493 dput(new_dentry);
2494exit4:
2495 dput(old_dentry);
2496exit3:
2497 unlock_rename(new_dir, old_dir);
2498exit2:
2499 path_release(&newnd);
2500exit1:
2501 path_release(&oldnd);
2502exit:
2503 return error;
2504}
2505
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002506asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2507 int newdfd, const char __user *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508{
2509 int error;
2510 char * from;
2511 char * to;
2512
2513 from = getname(oldname);
2514 if(IS_ERR(from))
2515 return PTR_ERR(from);
2516 to = getname(newname);
2517 error = PTR_ERR(to);
2518 if (!IS_ERR(to)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002519 error = do_rename(olddfd, from, newdfd, to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520 putname(to);
2521 }
2522 putname(from);
2523 return error;
2524}
2525
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002526asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2527{
2528 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2529}
2530
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2532{
2533 int len;
2534
2535 len = PTR_ERR(link);
2536 if (IS_ERR(link))
2537 goto out;
2538
2539 len = strlen(link);
2540 if (len > (unsigned) buflen)
2541 len = buflen;
2542 if (copy_to_user(buffer, link, len))
2543 len = -EFAULT;
2544out:
2545 return len;
2546}
2547
2548/*
2549 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2550 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2551 * using) it for any given inode is up to filesystem.
2552 */
2553int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2554{
2555 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002556 void *cookie;
2557
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002559 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2560 if (!IS_ERR(cookie)) {
2561 int res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562 if (dentry->d_inode->i_op->put_link)
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002563 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2564 cookie = ERR_PTR(res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565 }
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002566 return PTR_ERR(cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567}
2568
2569int vfs_follow_link(struct nameidata *nd, const char *link)
2570{
2571 return __vfs_follow_link(nd, link);
2572}
2573
2574/* get the link contents into pagecache */
2575static char *page_getlink(struct dentry * dentry, struct page **ppage)
2576{
2577 struct page * page;
2578 struct address_space *mapping = dentry->d_inode->i_mapping;
2579 page = read_cache_page(mapping, 0, (filler_t *)mapping->a_ops->readpage,
2580 NULL);
2581 if (IS_ERR(page))
2582 goto sync_fail;
2583 wait_on_page_locked(page);
2584 if (!PageUptodate(page))
2585 goto async_fail;
2586 *ppage = page;
2587 return kmap(page);
2588
2589async_fail:
2590 page_cache_release(page);
2591 return ERR_PTR(-EIO);
2592
2593sync_fail:
2594 return (char*)page;
2595}
2596
2597int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2598{
2599 struct page *page = NULL;
2600 char *s = page_getlink(dentry, &page);
2601 int res = vfs_readlink(dentry,buffer,buflen,s);
2602 if (page) {
2603 kunmap(page);
2604 page_cache_release(page);
2605 }
2606 return res;
2607}
2608
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002609void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002611 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002613 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002614}
2615
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002616void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002618 struct page *page = cookie;
2619
2620 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621 kunmap(page);
2622 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623 }
2624}
2625
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002626int __page_symlink(struct inode *inode, const char *symname, int len,
2627 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628{
2629 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002630 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631 int err = -ENOMEM;
2632 char *kaddr;
2633
NeilBrown7e53cac2006-03-25 03:07:57 -08002634retry:
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002635 page = find_or_create_page(mapping, 0, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636 if (!page)
2637 goto fail;
2638 err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
NeilBrown7e53cac2006-03-25 03:07:57 -08002639 if (err == AOP_TRUNCATED_PAGE) {
2640 page_cache_release(page);
2641 goto retry;
2642 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002643 if (err)
2644 goto fail_map;
2645 kaddr = kmap_atomic(page, KM_USER0);
2646 memcpy(kaddr, symname, len-1);
2647 kunmap_atomic(kaddr, KM_USER0);
NeilBrown7e53cac2006-03-25 03:07:57 -08002648 err = mapping->a_ops->commit_write(NULL, page, 0, len-1);
2649 if (err == AOP_TRUNCATED_PAGE) {
2650 page_cache_release(page);
2651 goto retry;
2652 }
2653 if (err)
2654 goto fail_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655 /*
2656 * Notice that we are _not_ going to block here - end of page is
2657 * unmapped, so this will only try to map the rest of page, see
2658 * that it is unmapped (typically even will not look into inode -
2659 * ->i_size will be enough for everything) and zero it out.
2660 * OTOH it's obviously correct and should make the page up-to-date.
2661 */
2662 if (!PageUptodate(page)) {
2663 err = mapping->a_ops->readpage(NULL, page);
NeilBrown7e53cac2006-03-25 03:07:57 -08002664 if (err != AOP_TRUNCATED_PAGE)
2665 wait_on_page_locked(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666 } else {
2667 unlock_page(page);
2668 }
2669 page_cache_release(page);
2670 if (err < 0)
2671 goto fail;
2672 mark_inode_dirty(inode);
2673 return 0;
2674fail_map:
2675 unlock_page(page);
2676 page_cache_release(page);
2677fail:
2678 return err;
2679}
2680
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002681int page_symlink(struct inode *inode, const char *symname, int len)
2682{
2683 return __page_symlink(inode, symname, len,
2684 mapping_gfp_mask(inode->i_mapping));
2685}
2686
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687struct inode_operations page_symlink_inode_operations = {
2688 .readlink = generic_readlink,
2689 .follow_link = page_follow_link_light,
2690 .put_link = page_put_link,
2691};
2692
2693EXPORT_SYMBOL(__user_walk);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002694EXPORT_SYMBOL(__user_walk_fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695EXPORT_SYMBOL(follow_down);
2696EXPORT_SYMBOL(follow_up);
2697EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2698EXPORT_SYMBOL(getname);
2699EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700EXPORT_SYMBOL(lookup_one_len);
2701EXPORT_SYMBOL(page_follow_link_light);
2702EXPORT_SYMBOL(page_put_link);
2703EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002704EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705EXPORT_SYMBOL(page_symlink);
2706EXPORT_SYMBOL(page_symlink_inode_operations);
2707EXPORT_SYMBOL(path_lookup);
2708EXPORT_SYMBOL(path_release);
2709EXPORT_SYMBOL(path_walk);
2710EXPORT_SYMBOL(permission);
Christoph Hellwige4543ed2005-11-08 21:35:04 -08002711EXPORT_SYMBOL(vfs_permission);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -08002712EXPORT_SYMBOL(file_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713EXPORT_SYMBOL(unlock_rename);
2714EXPORT_SYMBOL(vfs_create);
2715EXPORT_SYMBOL(vfs_follow_link);
2716EXPORT_SYMBOL(vfs_link);
2717EXPORT_SYMBOL(vfs_mkdir);
2718EXPORT_SYMBOL(vfs_mknod);
2719EXPORT_SYMBOL(generic_permission);
2720EXPORT_SYMBOL(vfs_readlink);
2721EXPORT_SYMBOL(vfs_rename);
2722EXPORT_SYMBOL(vfs_rmdir);
2723EXPORT_SYMBOL(vfs_symlink);
2724EXPORT_SYMBOL(vfs_unlink);
2725EXPORT_SYMBOL(dentry_unhash);
2726EXPORT_SYMBOL(generic_readlink);