blob: 184fe4acf82449dae1c62eade2d4489b078f24aa [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
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 if (*name=='/') {
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001084 read_lock(&current->fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 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);
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001095 read_unlock(&current->fs->lock);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001096 } else if (dfd == AT_FDCWD) {
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001097 read_lock(&current->fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 nd->mnt = mntget(current->fs->pwdmnt);
1099 nd->dentry = dget(current->fs->pwd);
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001100 read_unlock(&current->fs->lock);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001101 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001102 struct dentry *dentry;
1103
1104 file = fget_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001105 retval = -EBADF;
1106 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001107 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001108
1109 dentry = file->f_dentry;
1110
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001111 retval = -ENOTDIR;
1112 if (!S_ISDIR(dentry->d_inode->i_mode))
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001113 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001114
1115 retval = file_permission(file, MAY_EXEC);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001116 if (retval)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001117 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001118
1119 nd->mnt = mntget(file->f_vfsmnt);
1120 nd->dentry = dget(dentry);
1121
1122 fput_light(file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 current->total_link_count = 0;
1125 retval = link_path_walk(name, nd);
Prasanna Medaea3834d2005-04-29 16:00:17 +01001126out:
Suzuki3bc8414b2006-02-07 12:58:36 -08001127 if (likely(retval == 0)) {
1128 if (unlikely(current->audit_context && nd && nd->dentry &&
1129 nd->dentry->d_inode))
Amy Griffis9c937dc2006-06-08 23:19:31 -04001130 audit_inode(name, nd->dentry->d_inode);
Suzuki3bc8414b2006-02-07 12:58:36 -08001131 }
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001132out_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001133 return retval;
1134
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001135fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001136 fput_light(file, fput_needed);
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001137 goto out_fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138}
1139
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001140int fastcall path_lookup(const char *name, unsigned int flags,
1141 struct nameidata *nd)
1142{
1143 return do_path_lookup(AT_FDCWD, name, flags, nd);
1144}
1145
1146static int __path_lookup_intent_open(int dfd, const char *name,
1147 unsigned int lookup_flags, struct nameidata *nd,
1148 int open_flags, int create_mode)
Trond Myklebust834f2a42005-10-18 14:20:16 -07001149{
1150 struct file *filp = get_empty_filp();
1151 int err;
1152
1153 if (filp == NULL)
1154 return -ENFILE;
1155 nd->intent.open.file = filp;
1156 nd->intent.open.flags = open_flags;
1157 nd->intent.open.create_mode = create_mode;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001158 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
Trond Myklebust834f2a42005-10-18 14:20:16 -07001159 if (IS_ERR(nd->intent.open.file)) {
1160 if (err == 0) {
1161 err = PTR_ERR(nd->intent.open.file);
1162 path_release(nd);
1163 }
1164 } else if (err != 0)
1165 release_open_intent(nd);
1166 return err;
1167}
1168
1169/**
1170 * path_lookup_open - lookup a file path with open intent
Martin Waitz7045f372006-02-01 03:06:57 -08001171 * @dfd: the directory to use as base, or AT_FDCWD
Trond Myklebust834f2a42005-10-18 14:20:16 -07001172 * @name: pointer to file name
1173 * @lookup_flags: lookup intent flags
1174 * @nd: pointer to nameidata
1175 * @open_flags: open intent flags
1176 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001177int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
Trond Myklebust834f2a42005-10-18 14:20:16 -07001178 struct nameidata *nd, int open_flags)
1179{
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001180 return __path_lookup_intent_open(dfd, name, lookup_flags, nd,
Trond Myklebust834f2a42005-10-18 14:20:16 -07001181 open_flags, 0);
1182}
1183
1184/**
1185 * path_lookup_create - lookup a file path with open + create intent
Martin Waitz7045f372006-02-01 03:06:57 -08001186 * @dfd: the directory to use as base, or AT_FDCWD
Trond Myklebust834f2a42005-10-18 14:20:16 -07001187 * @name: pointer to file name
1188 * @lookup_flags: lookup intent flags
1189 * @nd: pointer to nameidata
1190 * @open_flags: open intent flags
1191 * @create_mode: create intent flags
1192 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001193static int path_lookup_create(int dfd, const char *name,
1194 unsigned int lookup_flags, struct nameidata *nd,
1195 int open_flags, int create_mode)
Trond Myklebust834f2a42005-10-18 14:20:16 -07001196{
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001197 return __path_lookup_intent_open(dfd, name, lookup_flags|LOOKUP_CREATE,
1198 nd, open_flags, create_mode);
Trond Myklebust834f2a42005-10-18 14:20:16 -07001199}
1200
1201int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags,
1202 struct nameidata *nd, int open_flags)
1203{
1204 char *tmp = getname(name);
1205 int err = PTR_ERR(tmp);
1206
1207 if (!IS_ERR(tmp)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001208 err = __path_lookup_intent_open(AT_FDCWD, tmp, lookup_flags, nd, open_flags, 0);
Trond Myklebust834f2a42005-10-18 14:20:16 -07001209 putname(tmp);
1210 }
1211 return err;
1212}
1213
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214/*
1215 * Restricted form of lookup. Doesn't follow links, single-component only,
1216 * needs parent already locked. Doesn't follow mounts.
1217 * SMP-safe.
1218 */
1219static struct dentry * __lookup_hash(struct qstr *name, struct dentry * base, struct nameidata *nd)
1220{
1221 struct dentry * dentry;
1222 struct inode *inode;
1223 int err;
1224
1225 inode = base->d_inode;
1226 err = permission(inode, MAY_EXEC, nd);
1227 dentry = ERR_PTR(err);
1228 if (err)
1229 goto out;
1230
1231 /*
1232 * See if the low-level filesystem might want
1233 * to use its own hash..
1234 */
1235 if (base->d_op && base->d_op->d_hash) {
1236 err = base->d_op->d_hash(base, name);
1237 dentry = ERR_PTR(err);
1238 if (err < 0)
1239 goto out;
1240 }
1241
1242 dentry = cached_lookup(base, name, nd);
1243 if (!dentry) {
1244 struct dentry *new = d_alloc(base, name);
1245 dentry = ERR_PTR(-ENOMEM);
1246 if (!new)
1247 goto out;
1248 dentry = inode->i_op->lookup(inode, new, nd);
1249 if (!dentry)
1250 dentry = new;
1251 else
1252 dput(new);
1253 }
1254out:
1255 return dentry;
1256}
1257
Adrian Bunka244e162006-03-31 02:32:11 -08001258static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259{
Christoph Hellwig49705b72005-11-08 21:35:06 -08001260 return __lookup_hash(&nd->last, nd->dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261}
1262
1263/* SMP-safe */
1264struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
1265{
1266 unsigned long hash;
1267 struct qstr this;
1268 unsigned int c;
1269
1270 this.name = name;
1271 this.len = len;
1272 if (!len)
1273 goto access;
1274
1275 hash = init_name_hash();
1276 while (len--) {
1277 c = *(const unsigned char *)name++;
1278 if (c == '/' || c == '\0')
1279 goto access;
1280 hash = partial_name_hash(c, hash);
1281 }
1282 this.hash = end_name_hash(hash);
1283
Christoph Hellwig49705b72005-11-08 21:35:06 -08001284 return __lookup_hash(&this, base, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285access:
1286 return ERR_PTR(-EACCES);
1287}
1288
1289/*
1290 * namei()
1291 *
1292 * is used by most simple commands to get the inode of a specified name.
1293 * Open, link etc use their own routines, but this is enough for things
1294 * like 'chmod' etc.
1295 *
1296 * namei exists in two versions: namei/lnamei. The only difference is
1297 * that namei follows links, while lnamei does not.
1298 * SMP-safe
1299 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001300int fastcall __user_walk_fd(int dfd, const char __user *name, unsigned flags,
1301 struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302{
1303 char *tmp = getname(name);
1304 int err = PTR_ERR(tmp);
1305
1306 if (!IS_ERR(tmp)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001307 err = do_path_lookup(dfd, tmp, flags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 putname(tmp);
1309 }
1310 return err;
1311}
1312
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001313int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1314{
1315 return __user_walk_fd(AT_FDCWD, name, flags, nd);
1316}
1317
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318/*
1319 * It's inline, so penalty for filesystems that don't use sticky bit is
1320 * minimal.
1321 */
1322static inline int check_sticky(struct inode *dir, struct inode *inode)
1323{
1324 if (!(dir->i_mode & S_ISVTX))
1325 return 0;
1326 if (inode->i_uid == current->fsuid)
1327 return 0;
1328 if (dir->i_uid == current->fsuid)
1329 return 0;
1330 return !capable(CAP_FOWNER);
1331}
1332
1333/*
1334 * Check whether we can remove a link victim from directory dir, check
1335 * whether the type of victim is right.
1336 * 1. We can't do it if dir is read-only (done in permission())
1337 * 2. We should have write and exec permissions on dir
1338 * 3. We can't remove anything from append-only dir
1339 * 4. We can't do anything with immutable dir (done in permission())
1340 * 5. If the sticky bit on dir is set we should either
1341 * a. be owner of dir, or
1342 * b. be owner of victim, or
1343 * c. have CAP_FOWNER capability
1344 * 6. If the victim is append-only or immutable we can't do antyhing with
1345 * links pointing to it.
1346 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1347 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1348 * 9. We can't remove a root or mountpoint.
1349 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1350 * nfs_async_unlink().
1351 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001352static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353{
1354 int error;
1355
1356 if (!victim->d_inode)
1357 return -ENOENT;
1358
1359 BUG_ON(victim->d_parent->d_inode != dir);
Amy Griffis73241cc2005-11-03 16:00:25 +00001360 audit_inode_child(victim->d_name.name, victim->d_inode, dir->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361
1362 error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
1363 if (error)
1364 return error;
1365 if (IS_APPEND(dir))
1366 return -EPERM;
1367 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1368 IS_IMMUTABLE(victim->d_inode))
1369 return -EPERM;
1370 if (isdir) {
1371 if (!S_ISDIR(victim->d_inode->i_mode))
1372 return -ENOTDIR;
1373 if (IS_ROOT(victim))
1374 return -EBUSY;
1375 } else if (S_ISDIR(victim->d_inode->i_mode))
1376 return -EISDIR;
1377 if (IS_DEADDIR(dir))
1378 return -ENOENT;
1379 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1380 return -EBUSY;
1381 return 0;
1382}
1383
1384/* Check whether we can create an object with dentry child in directory
1385 * dir.
1386 * 1. We can't do it if child already exists (open has special treatment for
1387 * this case, but since we are inlined it's OK)
1388 * 2. We can't do it if dir is read-only (done in permission())
1389 * 3. We should have write and exec permissions on dir
1390 * 4. We can't do it if dir is immutable (done in permission())
1391 */
1392static inline int may_create(struct inode *dir, struct dentry *child,
1393 struct nameidata *nd)
1394{
1395 if (child->d_inode)
1396 return -EEXIST;
1397 if (IS_DEADDIR(dir))
1398 return -ENOENT;
1399 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1400}
1401
1402/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 * O_DIRECTORY translates into forcing a directory lookup.
1404 */
1405static inline int lookup_flags(unsigned int f)
1406{
1407 unsigned long retval = LOOKUP_FOLLOW;
1408
1409 if (f & O_NOFOLLOW)
1410 retval &= ~LOOKUP_FOLLOW;
1411
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 if (f & O_DIRECTORY)
1413 retval |= LOOKUP_DIRECTORY;
1414
1415 return retval;
1416}
1417
1418/*
1419 * p1 and p2 should be directories on the same fs.
1420 */
1421struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1422{
1423 struct dentry *p;
1424
1425 if (p1 == p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001426 mutex_lock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 return NULL;
1428 }
1429
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001430 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431
1432 for (p = p1; p->d_parent != p; p = p->d_parent) {
1433 if (p->d_parent == p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001434 mutex_lock(&p2->d_inode->i_mutex);
1435 mutex_lock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 return p;
1437 }
1438 }
1439
1440 for (p = p2; p->d_parent != p; p = p->d_parent) {
1441 if (p->d_parent == p1) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001442 mutex_lock(&p1->d_inode->i_mutex);
1443 mutex_lock(&p2->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 return p;
1445 }
1446 }
1447
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001448 mutex_lock(&p1->d_inode->i_mutex);
1449 mutex_lock(&p2->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 return NULL;
1451}
1452
1453void unlock_rename(struct dentry *p1, struct dentry *p2)
1454{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001455 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001457 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001458 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 }
1460}
1461
1462int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1463 struct nameidata *nd)
1464{
1465 int error = may_create(dir, dentry, nd);
1466
1467 if (error)
1468 return error;
1469
1470 if (!dir->i_op || !dir->i_op->create)
1471 return -EACCES; /* shouldn't it be ENOSYS? */
1472 mode &= S_IALLUGO;
1473 mode |= S_IFREG;
1474 error = security_inode_create(dir, dentry, mode);
1475 if (error)
1476 return error;
1477 DQUOT_INIT(dir);
1478 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001479 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001480 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 return error;
1482}
1483
1484int may_open(struct nameidata *nd, int acc_mode, int flag)
1485{
1486 struct dentry *dentry = nd->dentry;
1487 struct inode *inode = dentry->d_inode;
1488 int error;
1489
1490 if (!inode)
1491 return -ENOENT;
1492
1493 if (S_ISLNK(inode->i_mode))
1494 return -ELOOP;
1495
1496 if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1497 return -EISDIR;
1498
Christoph Hellwige4543ed2005-11-08 21:35:04 -08001499 error = vfs_permission(nd, acc_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 if (error)
1501 return error;
1502
1503 /*
1504 * FIFO's, sockets and device files are special: they don't
1505 * actually live on the filesystem itself, and as such you
1506 * can write to them even if the filesystem is read-only.
1507 */
1508 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1509 flag &= ~O_TRUNC;
1510 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1511 if (nd->mnt->mnt_flags & MNT_NODEV)
1512 return -EACCES;
1513
1514 flag &= ~O_TRUNC;
1515 } else if (IS_RDONLY(inode) && (flag & FMODE_WRITE))
1516 return -EROFS;
1517 /*
1518 * An append-only file must be opened in append mode for writing.
1519 */
1520 if (IS_APPEND(inode)) {
1521 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1522 return -EPERM;
1523 if (flag & O_TRUNC)
1524 return -EPERM;
1525 }
1526
1527 /* O_NOATIME can only be set by the owner or superuser */
1528 if (flag & O_NOATIME)
1529 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1530 return -EPERM;
1531
1532 /*
1533 * Ensure there are no outstanding leases on the file.
1534 */
1535 error = break_lease(inode, flag);
1536 if (error)
1537 return error;
1538
1539 if (flag & O_TRUNC) {
1540 error = get_write_access(inode);
1541 if (error)
1542 return error;
1543
1544 /*
1545 * Refuse to truncate files with mandatory locks held on them.
1546 */
1547 error = locks_verify_locked(inode);
1548 if (!error) {
1549 DQUOT_INIT(inode);
1550
NeilBrown4a301312006-01-08 01:02:39 -08001551 error = do_truncate(dentry, 0, ATTR_MTIME|ATTR_CTIME, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 }
1553 put_write_access(inode);
1554 if (error)
1555 return error;
1556 } else
1557 if (flag & FMODE_WRITE)
1558 DQUOT_INIT(inode);
1559
1560 return 0;
1561}
1562
1563/*
1564 * open_namei()
1565 *
1566 * namei for open - this is in fact almost the whole open-routine.
1567 *
1568 * Note that the low bits of "flag" aren't the same as in the open
1569 * system call - they are 00 - no permissions needed
1570 * 01 - read permission needed
1571 * 10 - write permission needed
1572 * 11 - read/write permissions needed
1573 * which is a lot more logical, and also allows the "no perm" needed
1574 * for symlinks (where the permissions are checked later).
1575 * SMP-safe
1576 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001577int open_namei(int dfd, const char *pathname, int flag,
1578 int mode, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579{
Trond Myklebust834f2a42005-10-18 14:20:16 -07001580 int acc_mode, error;
Al Viro4e7506e2005-06-06 13:36:00 -07001581 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 struct dentry *dir;
1583 int count = 0;
1584
1585 acc_mode = ACC_MODE(flag);
1586
Trond Myklebust834f2a42005-10-18 14:20:16 -07001587 /* O_TRUNC implies we need access checks for write permissions */
1588 if (flag & O_TRUNC)
1589 acc_mode |= MAY_WRITE;
1590
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 /* Allow the LSM permission hook to distinguish append
1592 access from general write access. */
1593 if (flag & O_APPEND)
1594 acc_mode |= MAY_APPEND;
1595
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 /*
1597 * The simplest case - just a plain lookup.
1598 */
1599 if (!(flag & O_CREAT)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001600 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1601 nd, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602 if (error)
1603 return error;
1604 goto ok;
1605 }
1606
1607 /*
1608 * Create - we need to know the parent.
1609 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001610 error = path_lookup_create(dfd,pathname,LOOKUP_PARENT,nd,flag,mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 if (error)
1612 return error;
1613
1614 /*
1615 * We have the parent and last component. First of all, check
1616 * that we are not asked to creat(2) an obvious directory - that
1617 * will not do.
1618 */
1619 error = -EISDIR;
1620 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1621 goto exit;
1622
1623 dir = nd->dentry;
1624 nd->flags &= ~LOOKUP_PARENT;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001625 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001626 path.dentry = lookup_hash(nd);
Al Virod73ffe12005-06-06 13:36:01 -07001627 path.mnt = nd->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628
1629do_last:
Al Viro4e7506e2005-06-06 13:36:00 -07001630 error = PTR_ERR(path.dentry);
1631 if (IS_ERR(path.dentry)) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001632 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 goto exit;
1634 }
1635
Oleg Drokin4af4c522006-03-25 03:06:54 -08001636 if (IS_ERR(nd->intent.open.file)) {
1637 mutex_unlock(&dir->d_inode->i_mutex);
1638 error = PTR_ERR(nd->intent.open.file);
1639 goto exit_dput;
1640 }
1641
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 /* Negative dentry, just create the file */
Al Viro4e7506e2005-06-06 13:36:00 -07001643 if (!path.dentry->d_inode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 if (!IS_POSIXACL(dir->d_inode))
1645 mode &= ~current->fs->umask;
Al Viro4e7506e2005-06-06 13:36:00 -07001646 error = vfs_create(dir->d_inode, path.dentry, mode, nd);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001647 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 dput(nd->dentry);
Al Viro4e7506e2005-06-06 13:36:00 -07001649 nd->dentry = path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 if (error)
1651 goto exit;
1652 /* Don't check for write permission, don't truncate */
1653 acc_mode = 0;
1654 flag &= ~O_TRUNC;
1655 goto ok;
1656 }
1657
1658 /*
1659 * It already exists.
1660 */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001661 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662
1663 error = -EEXIST;
1664 if (flag & O_EXCL)
1665 goto exit_dput;
1666
Al Viroe13b2102005-06-06 13:36:06 -07001667 if (__follow_mount(&path)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 error = -ELOOP;
Al Viroba7a4c12005-06-06 13:36:08 -07001669 if (flag & O_NOFOLLOW)
1670 goto exit_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 }
1672 error = -ENOENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001673 if (!path.dentry->d_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 goto exit_dput;
Al Viro4e7506e2005-06-06 13:36:00 -07001675 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 goto do_link;
1677
Miklos Szeredi09dd17d2005-09-06 15:18:21 -07001678 path_to_nameidata(&path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 error = -EISDIR;
Al Viro4e7506e2005-06-06 13:36:00 -07001680 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681 goto exit;
1682ok:
1683 error = may_open(nd, acc_mode, flag);
1684 if (error)
1685 goto exit;
1686 return 0;
1687
1688exit_dput:
Miklos Szeredi09dd17d2005-09-06 15:18:21 -07001689 dput_path(&path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690exit:
Trond Myklebust834f2a42005-10-18 14:20:16 -07001691 if (!IS_ERR(nd->intent.open.file))
1692 release_open_intent(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 path_release(nd);
1694 return error;
1695
1696do_link:
1697 error = -ELOOP;
1698 if (flag & O_NOFOLLOW)
1699 goto exit_dput;
1700 /*
1701 * This is subtle. Instead of calling do_follow_link() we do the
1702 * thing by hands. The reason is that this way we have zero link_count
1703 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1704 * After that we have the parent and last component, i.e.
1705 * we are in the same situation as after the first path_walk().
1706 * Well, almost - if the last component is normal we get its copy
1707 * stored in nd->last.name and we will have to putname() it when we
1708 * are done. Procfs-like symlinks just set LAST_BIND.
1709 */
1710 nd->flags |= LOOKUP_PARENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001711 error = security_inode_follow_link(path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 if (error)
1713 goto exit_dput;
Al Virocd4e91d2005-06-06 13:36:03 -07001714 error = __do_follow_link(&path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 if (error)
1716 return error;
1717 nd->flags &= ~LOOKUP_PARENT;
Al Virod671d5e2005-06-06 13:36:04 -07001718 if (nd->last_type == LAST_BIND)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 goto ok;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 error = -EISDIR;
1721 if (nd->last_type != LAST_NORM)
1722 goto exit;
1723 if (nd->last.name[nd->last.len]) {
Linus Torvalds82984112005-10-06 21:54:21 -07001724 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 goto exit;
1726 }
1727 error = -ELOOP;
1728 if (count++==32) {
Linus Torvalds82984112005-10-06 21:54:21 -07001729 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 goto exit;
1731 }
1732 dir = nd->dentry;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001733 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001734 path.dentry = lookup_hash(nd);
Al Virod671d5e2005-06-06 13:36:04 -07001735 path.mnt = nd->mnt;
Linus Torvalds82984112005-10-06 21:54:21 -07001736 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 goto do_last;
1738}
1739
1740/**
1741 * lookup_create - lookup a dentry, creating it if it doesn't exist
1742 * @nd: nameidata info
1743 * @is_dir: directory flag
1744 *
1745 * Simple function to lookup and return a dentry and create it
1746 * if it doesn't exist. Is SMP-safe.
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001747 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001748 * Returns with nd->dentry->d_inode->i_mutex locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 */
1750struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1751{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001752 struct dentry *dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001754 mutex_lock(&nd->dentry->d_inode->i_mutex);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001755 /*
1756 * Yucky last component or no last component at all?
1757 * (foo/., foo/.., /////)
1758 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 if (nd->last_type != LAST_NORM)
1760 goto fail;
1761 nd->flags &= ~LOOKUP_PARENT;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001762
1763 /*
1764 * Do the final lookup.
1765 */
Christoph Hellwig49705b72005-11-08 21:35:06 -08001766 dentry = lookup_hash(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767 if (IS_ERR(dentry))
1768 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001769
1770 /*
1771 * Special case - lookup gave negative, but... we had foo/bar/
1772 * From the vfs_mknod() POV we just have a negative dentry -
1773 * all is fine. Let's be bastards - you had / on the end, you've
1774 * been asking for (non-existent) directory. -ENOENT for you.
1775 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1777 goto enoent;
1778 return dentry;
1779enoent:
1780 dput(dentry);
1781 dentry = ERR_PTR(-ENOENT);
1782fail:
1783 return dentry;
1784}
Christoph Hellwigf81a0bf2005-05-19 12:26:43 -07001785EXPORT_SYMBOL_GPL(lookup_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786
1787int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1788{
1789 int error = may_create(dir, dentry, NULL);
1790
1791 if (error)
1792 return error;
1793
1794 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1795 return -EPERM;
1796
1797 if (!dir->i_op || !dir->i_op->mknod)
1798 return -EPERM;
1799
1800 error = security_inode_mknod(dir, dentry, mode, dev);
1801 if (error)
1802 return error;
1803
1804 DQUOT_INIT(dir);
1805 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001806 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001807 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 return error;
1809}
1810
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001811asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
1812 unsigned dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813{
1814 int error = 0;
1815 char * tmp;
1816 struct dentry * dentry;
1817 struct nameidata nd;
1818
1819 if (S_ISDIR(mode))
1820 return -EPERM;
1821 tmp = getname(filename);
1822 if (IS_ERR(tmp))
1823 return PTR_ERR(tmp);
1824
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001825 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 if (error)
1827 goto out;
1828 dentry = lookup_create(&nd, 0);
1829 error = PTR_ERR(dentry);
1830
1831 if (!IS_POSIXACL(nd.dentry->d_inode))
1832 mode &= ~current->fs->umask;
1833 if (!IS_ERR(dentry)) {
1834 switch (mode & S_IFMT) {
1835 case 0: case S_IFREG:
1836 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1837 break;
1838 case S_IFCHR: case S_IFBLK:
1839 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,
1840 new_decode_dev(dev));
1841 break;
1842 case S_IFIFO: case S_IFSOCK:
1843 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,0);
1844 break;
1845 case S_IFDIR:
1846 error = -EPERM;
1847 break;
1848 default:
1849 error = -EINVAL;
1850 }
1851 dput(dentry);
1852 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001853 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 path_release(&nd);
1855out:
1856 putname(tmp);
1857
1858 return error;
1859}
1860
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001861asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
1862{
1863 return sys_mknodat(AT_FDCWD, filename, mode, dev);
1864}
1865
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1867{
1868 int error = may_create(dir, dentry, NULL);
1869
1870 if (error)
1871 return error;
1872
1873 if (!dir->i_op || !dir->i_op->mkdir)
1874 return -EPERM;
1875
1876 mode &= (S_IRWXUGO|S_ISVTX);
1877 error = security_inode_mkdir(dir, dentry, mode);
1878 if (error)
1879 return error;
1880
1881 DQUOT_INIT(dir);
1882 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001883 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001884 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885 return error;
1886}
1887
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001888asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889{
1890 int error = 0;
1891 char * tmp;
1892
1893 tmp = getname(pathname);
1894 error = PTR_ERR(tmp);
1895 if (!IS_ERR(tmp)) {
1896 struct dentry *dentry;
1897 struct nameidata nd;
1898
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001899 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900 if (error)
1901 goto out;
1902 dentry = lookup_create(&nd, 1);
1903 error = PTR_ERR(dentry);
1904 if (!IS_ERR(dentry)) {
1905 if (!IS_POSIXACL(nd.dentry->d_inode))
1906 mode &= ~current->fs->umask;
1907 error = vfs_mkdir(nd.dentry->d_inode, dentry, mode);
1908 dput(dentry);
1909 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001910 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911 path_release(&nd);
1912out:
1913 putname(tmp);
1914 }
1915
1916 return error;
1917}
1918
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001919asmlinkage long sys_mkdir(const char __user *pathname, int mode)
1920{
1921 return sys_mkdirat(AT_FDCWD, pathname, mode);
1922}
1923
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924/*
1925 * We try to drop the dentry early: we should have
1926 * a usage count of 2 if we're the only user of this
1927 * dentry, and if that is true (possibly after pruning
1928 * the dcache), then we drop the dentry now.
1929 *
1930 * A low-level filesystem can, if it choses, legally
1931 * do a
1932 *
1933 * if (!d_unhashed(dentry))
1934 * return -EBUSY;
1935 *
1936 * if it cannot handle the case of removing a directory
1937 * that is still in use by something else..
1938 */
1939void dentry_unhash(struct dentry *dentry)
1940{
1941 dget(dentry);
1942 if (atomic_read(&dentry->d_count))
1943 shrink_dcache_parent(dentry);
1944 spin_lock(&dcache_lock);
1945 spin_lock(&dentry->d_lock);
1946 if (atomic_read(&dentry->d_count) == 2)
1947 __d_drop(dentry);
1948 spin_unlock(&dentry->d_lock);
1949 spin_unlock(&dcache_lock);
1950}
1951
1952int vfs_rmdir(struct inode *dir, struct dentry *dentry)
1953{
1954 int error = may_delete(dir, dentry, 1);
1955
1956 if (error)
1957 return error;
1958
1959 if (!dir->i_op || !dir->i_op->rmdir)
1960 return -EPERM;
1961
1962 DQUOT_INIT(dir);
1963
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001964 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 dentry_unhash(dentry);
1966 if (d_mountpoint(dentry))
1967 error = -EBUSY;
1968 else {
1969 error = security_inode_rmdir(dir, dentry);
1970 if (!error) {
1971 error = dir->i_op->rmdir(dir, dentry);
1972 if (!error)
1973 dentry->d_inode->i_flags |= S_DEAD;
1974 }
1975 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001976 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978 d_delete(dentry);
1979 }
1980 dput(dentry);
1981
1982 return error;
1983}
1984
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001985static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986{
1987 int error = 0;
1988 char * name;
1989 struct dentry *dentry;
1990 struct nameidata nd;
1991
1992 name = getname(pathname);
1993 if(IS_ERR(name))
1994 return PTR_ERR(name);
1995
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001996 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997 if (error)
1998 goto exit;
1999
2000 switch(nd.last_type) {
2001 case LAST_DOTDOT:
2002 error = -ENOTEMPTY;
2003 goto exit1;
2004 case LAST_DOT:
2005 error = -EINVAL;
2006 goto exit1;
2007 case LAST_ROOT:
2008 error = -EBUSY;
2009 goto exit1;
2010 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002011 mutex_lock(&nd.dentry->d_inode->i_mutex);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002012 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 error = PTR_ERR(dentry);
2014 if (!IS_ERR(dentry)) {
2015 error = vfs_rmdir(nd.dentry->d_inode, dentry);
2016 dput(dentry);
2017 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002018 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019exit1:
2020 path_release(&nd);
2021exit:
2022 putname(name);
2023 return error;
2024}
2025
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002026asmlinkage long sys_rmdir(const char __user *pathname)
2027{
2028 return do_rmdir(AT_FDCWD, pathname);
2029}
2030
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031int vfs_unlink(struct inode *dir, struct dentry *dentry)
2032{
2033 int error = may_delete(dir, dentry, 0);
2034
2035 if (error)
2036 return error;
2037
2038 if (!dir->i_op || !dir->i_op->unlink)
2039 return -EPERM;
2040
2041 DQUOT_INIT(dir);
2042
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002043 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 if (d_mountpoint(dentry))
2045 error = -EBUSY;
2046 else {
2047 error = security_inode_unlink(dir, dentry);
2048 if (!error)
2049 error = dir->i_op->unlink(dir, dentry);
2050 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002051 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052
2053 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2054 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2055 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056 }
Robert Love0eeca282005-07-12 17:06:03 -04002057
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058 return error;
2059}
2060
2061/*
2062 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002063 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064 * writeout happening, and we don't want to prevent access to the directory
2065 * while waiting on the I/O.
2066 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002067static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068{
2069 int error = 0;
2070 char * name;
2071 struct dentry *dentry;
2072 struct nameidata nd;
2073 struct inode *inode = NULL;
2074
2075 name = getname(pathname);
2076 if(IS_ERR(name))
2077 return PTR_ERR(name);
2078
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002079 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 if (error)
2081 goto exit;
2082 error = -EISDIR;
2083 if (nd.last_type != LAST_NORM)
2084 goto exit1;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002085 mutex_lock(&nd.dentry->d_inode->i_mutex);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002086 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087 error = PTR_ERR(dentry);
2088 if (!IS_ERR(dentry)) {
2089 /* Why not before? Because we want correct error value */
2090 if (nd.last.name[nd.last.len])
2091 goto slashes;
2092 inode = dentry->d_inode;
2093 if (inode)
2094 atomic_inc(&inode->i_count);
2095 error = vfs_unlink(nd.dentry->d_inode, dentry);
2096 exit2:
2097 dput(dentry);
2098 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002099 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 if (inode)
2101 iput(inode); /* truncate the inode here */
2102exit1:
2103 path_release(&nd);
2104exit:
2105 putname(name);
2106 return error;
2107
2108slashes:
2109 error = !dentry->d_inode ? -ENOENT :
2110 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2111 goto exit2;
2112}
2113
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002114asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2115{
2116 if ((flag & ~AT_REMOVEDIR) != 0)
2117 return -EINVAL;
2118
2119 if (flag & AT_REMOVEDIR)
2120 return do_rmdir(dfd, pathname);
2121
2122 return do_unlinkat(dfd, pathname);
2123}
2124
2125asmlinkage long sys_unlink(const char __user *pathname)
2126{
2127 return do_unlinkat(AT_FDCWD, pathname);
2128}
2129
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname, int mode)
2131{
2132 int error = may_create(dir, dentry, NULL);
2133
2134 if (error)
2135 return error;
2136
2137 if (!dir->i_op || !dir->i_op->symlink)
2138 return -EPERM;
2139
2140 error = security_inode_symlink(dir, dentry, oldname);
2141 if (error)
2142 return error;
2143
2144 DQUOT_INIT(dir);
2145 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002146 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002147 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148 return error;
2149}
2150
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002151asmlinkage long sys_symlinkat(const char __user *oldname,
2152 int newdfd, const char __user *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153{
2154 int error = 0;
2155 char * from;
2156 char * to;
2157
2158 from = getname(oldname);
2159 if(IS_ERR(from))
2160 return PTR_ERR(from);
2161 to = getname(newname);
2162 error = PTR_ERR(to);
2163 if (!IS_ERR(to)) {
2164 struct dentry *dentry;
2165 struct nameidata nd;
2166
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002167 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 if (error)
2169 goto out;
2170 dentry = lookup_create(&nd, 0);
2171 error = PTR_ERR(dentry);
2172 if (!IS_ERR(dentry)) {
2173 error = vfs_symlink(nd.dentry->d_inode, dentry, from, S_IALLUGO);
2174 dput(dentry);
2175 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002176 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177 path_release(&nd);
2178out:
2179 putname(to);
2180 }
2181 putname(from);
2182 return error;
2183}
2184
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002185asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2186{
2187 return sys_symlinkat(oldname, AT_FDCWD, newname);
2188}
2189
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2191{
2192 struct inode *inode = old_dentry->d_inode;
2193 int error;
2194
2195 if (!inode)
2196 return -ENOENT;
2197
2198 error = may_create(dir, new_dentry, NULL);
2199 if (error)
2200 return error;
2201
2202 if (dir->i_sb != inode->i_sb)
2203 return -EXDEV;
2204
2205 /*
2206 * A link to an append-only or immutable file cannot be created.
2207 */
2208 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2209 return -EPERM;
2210 if (!dir->i_op || !dir->i_op->link)
2211 return -EPERM;
2212 if (S_ISDIR(old_dentry->d_inode->i_mode))
2213 return -EPERM;
2214
2215 error = security_inode_link(old_dentry, dir, new_dentry);
2216 if (error)
2217 return error;
2218
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002219 mutex_lock(&old_dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220 DQUOT_INIT(dir);
2221 error = dir->i_op->link(old_dentry, dir, new_dentry);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002222 mutex_unlock(&old_dentry->d_inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002223 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002224 fsnotify_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225 return error;
2226}
2227
2228/*
2229 * Hardlinks are often used in delicate situations. We avoid
2230 * security-related surprises by not following symlinks on the
2231 * newname. --KAB
2232 *
2233 * We don't follow them on the oldname either to be compatible
2234 * with linux 2.0, and to avoid hard-linking to directories
2235 * and other special files. --ADM
2236 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002237asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002238 int newdfd, const char __user *newname,
2239 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240{
2241 struct dentry *new_dentry;
2242 struct nameidata nd, old_nd;
2243 int error;
2244 char * to;
2245
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002246 if (flags != 0)
2247 return -EINVAL;
2248
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249 to = getname(newname);
2250 if (IS_ERR(to))
2251 return PTR_ERR(to);
2252
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002253 error = __user_walk_fd(olddfd, oldname, 0, &old_nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254 if (error)
2255 goto exit;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002256 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257 if (error)
2258 goto out;
2259 error = -EXDEV;
2260 if (old_nd.mnt != nd.mnt)
2261 goto out_release;
2262 new_dentry = lookup_create(&nd, 0);
2263 error = PTR_ERR(new_dentry);
2264 if (!IS_ERR(new_dentry)) {
2265 error = vfs_link(old_nd.dentry, nd.dentry->d_inode, new_dentry);
2266 dput(new_dentry);
2267 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002268 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269out_release:
2270 path_release(&nd);
2271out:
2272 path_release(&old_nd);
2273exit:
2274 putname(to);
2275
2276 return error;
2277}
2278
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002279asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2280{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002281 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002282}
2283
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284/*
2285 * The worst of all namespace operations - renaming directory. "Perverted"
2286 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2287 * Problems:
2288 * a) we can get into loop creation. Check is done in is_subdir().
2289 * b) race potential - two innocent renames can create a loop together.
2290 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002291 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 * story.
2293 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002294 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 * whether the target exists). Solution: try to be smart with locking
2296 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002297 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298 * move will be locked. Thus we can rank directories by the tree
2299 * (ancestors first) and rank all non-directories after them.
2300 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002301 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302 * HOWEVER, it relies on the assumption that any object with ->lookup()
2303 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2304 * we'd better make sure that there's no link(2) for them.
2305 * d) some filesystems don't support opened-but-unlinked directories,
2306 * either because of layout or because they are not ready to deal with
2307 * all cases correctly. The latter will be fixed (taking this sort of
2308 * stuff into VFS), but the former is not going away. Solution: the same
2309 * trick as in rmdir().
2310 * e) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002311 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002313 * ->i_mutex on parents, which works but leads to some truely excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314 * locking].
2315 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07002316static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2317 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318{
2319 int error = 0;
2320 struct inode *target;
2321
2322 /*
2323 * If we are going to change the parent - check write permissions,
2324 * we'll need to flip '..'.
2325 */
2326 if (new_dir != old_dir) {
2327 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2328 if (error)
2329 return error;
2330 }
2331
2332 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2333 if (error)
2334 return error;
2335
2336 target = new_dentry->d_inode;
2337 if (target) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002338 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339 dentry_unhash(new_dentry);
2340 }
2341 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2342 error = -EBUSY;
2343 else
2344 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2345 if (target) {
2346 if (!error)
2347 target->i_flags |= S_DEAD;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002348 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 if (d_unhashed(new_dentry))
2350 d_rehash(new_dentry);
2351 dput(new_dentry);
2352 }
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002353 if (!error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355 return error;
2356}
2357
Adrian Bunk75c96f82005-05-05 16:16:09 -07002358static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2359 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360{
2361 struct inode *target;
2362 int error;
2363
2364 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2365 if (error)
2366 return error;
2367
2368 dget(new_dentry);
2369 target = new_dentry->d_inode;
2370 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002371 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2373 error = -EBUSY;
2374 else
2375 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2376 if (!error) {
2377 /* The following d_move() should become unconditional */
2378 if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
2379 d_move(old_dentry, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380 }
2381 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002382 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383 dput(new_dentry);
2384 return error;
2385}
2386
2387int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2388 struct inode *new_dir, struct dentry *new_dentry)
2389{
2390 int error;
2391 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Robert Love0eeca282005-07-12 17:06:03 -04002392 const char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393
2394 if (old_dentry->d_inode == new_dentry->d_inode)
2395 return 0;
2396
2397 error = may_delete(old_dir, old_dentry, is_dir);
2398 if (error)
2399 return error;
2400
2401 if (!new_dentry->d_inode)
2402 error = may_create(new_dir, new_dentry, NULL);
2403 else
2404 error = may_delete(new_dir, new_dentry, is_dir);
2405 if (error)
2406 return error;
2407
2408 if (!old_dir->i_op || !old_dir->i_op->rename)
2409 return -EPERM;
2410
2411 DQUOT_INIT(old_dir);
2412 DQUOT_INIT(new_dir);
2413
Robert Love0eeca282005-07-12 17:06:03 -04002414 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2415
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416 if (is_dir)
2417 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2418 else
2419 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2420 if (!error) {
Robert Love0eeca282005-07-12 17:06:03 -04002421 const char *new_name = old_dentry->d_name.name;
John McCutchan89204c42005-08-15 12:13:28 -04002422 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2423 new_dentry->d_inode, old_dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 }
Robert Love0eeca282005-07-12 17:06:03 -04002425 fsnotify_oldname_free(old_name);
2426
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 return error;
2428}
2429
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002430static int do_rename(int olddfd, const char *oldname,
2431 int newdfd, const char *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002432{
2433 int error = 0;
2434 struct dentry * old_dir, * new_dir;
2435 struct dentry * old_dentry, *new_dentry;
2436 struct dentry * trap;
2437 struct nameidata oldnd, newnd;
2438
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002439 error = do_path_lookup(olddfd, oldname, LOOKUP_PARENT, &oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440 if (error)
2441 goto exit;
2442
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002443 error = do_path_lookup(newdfd, newname, LOOKUP_PARENT, &newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444 if (error)
2445 goto exit1;
2446
2447 error = -EXDEV;
2448 if (oldnd.mnt != newnd.mnt)
2449 goto exit2;
2450
2451 old_dir = oldnd.dentry;
2452 error = -EBUSY;
2453 if (oldnd.last_type != LAST_NORM)
2454 goto exit2;
2455
2456 new_dir = newnd.dentry;
2457 if (newnd.last_type != LAST_NORM)
2458 goto exit2;
2459
2460 trap = lock_rename(new_dir, old_dir);
2461
Christoph Hellwig49705b72005-11-08 21:35:06 -08002462 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463 error = PTR_ERR(old_dentry);
2464 if (IS_ERR(old_dentry))
2465 goto exit3;
2466 /* source must exist */
2467 error = -ENOENT;
2468 if (!old_dentry->d_inode)
2469 goto exit4;
2470 /* unless the source is a directory trailing slashes give -ENOTDIR */
2471 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2472 error = -ENOTDIR;
2473 if (oldnd.last.name[oldnd.last.len])
2474 goto exit4;
2475 if (newnd.last.name[newnd.last.len])
2476 goto exit4;
2477 }
2478 /* source should not be ancestor of target */
2479 error = -EINVAL;
2480 if (old_dentry == trap)
2481 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08002482 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483 error = PTR_ERR(new_dentry);
2484 if (IS_ERR(new_dentry))
2485 goto exit4;
2486 /* target should not be an ancestor of source */
2487 error = -ENOTEMPTY;
2488 if (new_dentry == trap)
2489 goto exit5;
2490
2491 error = vfs_rename(old_dir->d_inode, old_dentry,
2492 new_dir->d_inode, new_dentry);
2493exit5:
2494 dput(new_dentry);
2495exit4:
2496 dput(old_dentry);
2497exit3:
2498 unlock_rename(new_dir, old_dir);
2499exit2:
2500 path_release(&newnd);
2501exit1:
2502 path_release(&oldnd);
2503exit:
2504 return error;
2505}
2506
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002507asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2508 int newdfd, const char __user *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509{
2510 int error;
2511 char * from;
2512 char * to;
2513
2514 from = getname(oldname);
2515 if(IS_ERR(from))
2516 return PTR_ERR(from);
2517 to = getname(newname);
2518 error = PTR_ERR(to);
2519 if (!IS_ERR(to)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002520 error = do_rename(olddfd, from, newdfd, to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521 putname(to);
2522 }
2523 putname(from);
2524 return error;
2525}
2526
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002527asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2528{
2529 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2530}
2531
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2533{
2534 int len;
2535
2536 len = PTR_ERR(link);
2537 if (IS_ERR(link))
2538 goto out;
2539
2540 len = strlen(link);
2541 if (len > (unsigned) buflen)
2542 len = buflen;
2543 if (copy_to_user(buffer, link, len))
2544 len = -EFAULT;
2545out:
2546 return len;
2547}
2548
2549/*
2550 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2551 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2552 * using) it for any given inode is up to filesystem.
2553 */
2554int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2555{
2556 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002557 void *cookie;
2558
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002560 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2561 if (!IS_ERR(cookie)) {
2562 int res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563 if (dentry->d_inode->i_op->put_link)
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002564 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2565 cookie = ERR_PTR(res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566 }
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002567 return PTR_ERR(cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568}
2569
2570int vfs_follow_link(struct nameidata *nd, const char *link)
2571{
2572 return __vfs_follow_link(nd, link);
2573}
2574
2575/* get the link contents into pagecache */
2576static char *page_getlink(struct dentry * dentry, struct page **ppage)
2577{
2578 struct page * page;
2579 struct address_space *mapping = dentry->d_inode->i_mapping;
2580 page = read_cache_page(mapping, 0, (filler_t *)mapping->a_ops->readpage,
2581 NULL);
2582 if (IS_ERR(page))
2583 goto sync_fail;
2584 wait_on_page_locked(page);
2585 if (!PageUptodate(page))
2586 goto async_fail;
2587 *ppage = page;
2588 return kmap(page);
2589
2590async_fail:
2591 page_cache_release(page);
2592 return ERR_PTR(-EIO);
2593
2594sync_fail:
2595 return (char*)page;
2596}
2597
2598int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2599{
2600 struct page *page = NULL;
2601 char *s = page_getlink(dentry, &page);
2602 int res = vfs_readlink(dentry,buffer,buflen,s);
2603 if (page) {
2604 kunmap(page);
2605 page_cache_release(page);
2606 }
2607 return res;
2608}
2609
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002610void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002612 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002614 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615}
2616
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002617void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002619 struct page *page = cookie;
2620
2621 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622 kunmap(page);
2623 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002624 }
2625}
2626
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002627int __page_symlink(struct inode *inode, const char *symname, int len,
2628 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629{
2630 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002631 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632 int err = -ENOMEM;
2633 char *kaddr;
2634
NeilBrown7e53cac2006-03-25 03:07:57 -08002635retry:
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002636 page = find_or_create_page(mapping, 0, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002637 if (!page)
2638 goto fail;
2639 err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
NeilBrown7e53cac2006-03-25 03:07:57 -08002640 if (err == AOP_TRUNCATED_PAGE) {
2641 page_cache_release(page);
2642 goto retry;
2643 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644 if (err)
2645 goto fail_map;
2646 kaddr = kmap_atomic(page, KM_USER0);
2647 memcpy(kaddr, symname, len-1);
2648 kunmap_atomic(kaddr, KM_USER0);
NeilBrown7e53cac2006-03-25 03:07:57 -08002649 err = mapping->a_ops->commit_write(NULL, page, 0, len-1);
2650 if (err == AOP_TRUNCATED_PAGE) {
2651 page_cache_release(page);
2652 goto retry;
2653 }
2654 if (err)
2655 goto fail_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002656 /*
2657 * Notice that we are _not_ going to block here - end of page is
2658 * unmapped, so this will only try to map the rest of page, see
2659 * that it is unmapped (typically even will not look into inode -
2660 * ->i_size will be enough for everything) and zero it out.
2661 * OTOH it's obviously correct and should make the page up-to-date.
2662 */
2663 if (!PageUptodate(page)) {
2664 err = mapping->a_ops->readpage(NULL, page);
NeilBrown7e53cac2006-03-25 03:07:57 -08002665 if (err != AOP_TRUNCATED_PAGE)
2666 wait_on_page_locked(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667 } else {
2668 unlock_page(page);
2669 }
2670 page_cache_release(page);
2671 if (err < 0)
2672 goto fail;
2673 mark_inode_dirty(inode);
2674 return 0;
2675fail_map:
2676 unlock_page(page);
2677 page_cache_release(page);
2678fail:
2679 return err;
2680}
2681
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002682int page_symlink(struct inode *inode, const char *symname, int len)
2683{
2684 return __page_symlink(inode, symname, len,
2685 mapping_gfp_mask(inode->i_mapping));
2686}
2687
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688struct inode_operations page_symlink_inode_operations = {
2689 .readlink = generic_readlink,
2690 .follow_link = page_follow_link_light,
2691 .put_link = page_put_link,
2692};
2693
2694EXPORT_SYMBOL(__user_walk);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002695EXPORT_SYMBOL(__user_walk_fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002696EXPORT_SYMBOL(follow_down);
2697EXPORT_SYMBOL(follow_up);
2698EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2699EXPORT_SYMBOL(getname);
2700EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002701EXPORT_SYMBOL(lookup_one_len);
2702EXPORT_SYMBOL(page_follow_link_light);
2703EXPORT_SYMBOL(page_put_link);
2704EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002705EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706EXPORT_SYMBOL(page_symlink);
2707EXPORT_SYMBOL(page_symlink_inode_operations);
2708EXPORT_SYMBOL(path_lookup);
2709EXPORT_SYMBOL(path_release);
2710EXPORT_SYMBOL(path_walk);
2711EXPORT_SYMBOL(permission);
Christoph Hellwige4543ed2005-11-08 21:35:04 -08002712EXPORT_SYMBOL(vfs_permission);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -08002713EXPORT_SYMBOL(file_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714EXPORT_SYMBOL(unlock_rename);
2715EXPORT_SYMBOL(vfs_create);
2716EXPORT_SYMBOL(vfs_follow_link);
2717EXPORT_SYMBOL(vfs_link);
2718EXPORT_SYMBOL(vfs_mkdir);
2719EXPORT_SYMBOL(vfs_mknod);
2720EXPORT_SYMBOL(generic_permission);
2721EXPORT_SYMBOL(vfs_readlink);
2722EXPORT_SYMBOL(vfs_rename);
2723EXPORT_SYMBOL(vfs_rmdir);
2724EXPORT_SYMBOL(vfs_symlink);
2725EXPORT_SYMBOL(vfs_unlink);
2726EXPORT_SYMBOL(dentry_unhash);
2727EXPORT_SYMBOL(generic_readlink);