blob: 98dc2e1343621608a391ceb019a2956221bb6d84 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
18#include <linux/module.h>
19#include <linux/slab.h>
20#include <linux/fs.h>
21#include <linux/namei.h>
22#include <linux/quotaops.h>
23#include <linux/pagemap.h>
Robert Love0eeca282005-07-12 17:06:03 -040024#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/smp_lock.h>
26#include <linux/personality.h>
27#include <linux/security.h>
28#include <linux/syscalls.h>
29#include <linux/mount.h>
30#include <linux/audit.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080031#include <linux/capability.h>
Trond Myklebust834f2a42005-10-18 14:20:16 -070032#include <linux/file.h>
Ulrich Drepper5590ff02006-01-18 17:43:53 -080033#include <linux/fcntl.h>
34#include <linux/namei.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <asm/namei.h>
36#include <asm/uaccess.h>
37
38#define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
39
40/* [Feb-1997 T. Schoebel-Theuer]
41 * Fundamental changes in the pathname lookup mechanisms (namei)
42 * were necessary because of omirr. The reason is that omirr needs
43 * to know the _real_ pathname, not the user-supplied one, in case
44 * of symlinks (and also when transname replacements occur).
45 *
46 * The new code replaces the old recursive symlink resolution with
47 * an iterative one (in case of non-nested symlink chains). It does
48 * this with calls to <fs>_follow_link().
49 * As a side effect, dir_namei(), _namei() and follow_link() are now
50 * replaced with a single function lookup_dentry() that can handle all
51 * the special cases of the former code.
52 *
53 * With the new dcache, the pathname is stored at each inode, at least as
54 * long as the refcount of the inode is positive. As a side effect, the
55 * size of the dcache depends on the inode cache and thus is dynamic.
56 *
57 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
58 * resolution to correspond with current state of the code.
59 *
60 * Note that the symlink resolution is not *completely* iterative.
61 * There is still a significant amount of tail- and mid- recursion in
62 * the algorithm. Also, note that <fs>_readlink() is not used in
63 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
64 * may return different results than <fs>_follow_link(). Many virtual
65 * filesystems (including /proc) exhibit this behavior.
66 */
67
68/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
69 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
70 * and the name already exists in form of a symlink, try to create the new
71 * name indicated by the symlink. The old code always complained that the
72 * name already exists, due to not following the symlink even if its target
73 * is nonexistent. The new semantics affects also mknod() and link() when
74 * the name is a symlink pointing to a non-existant name.
75 *
76 * I don't know which semantics is the right one, since I have no access
77 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
78 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
79 * "old" one. Personally, I think the new semantics is much more logical.
80 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
81 * file does succeed in both HP-UX and SunOs, but not in Solaris
82 * and in the old Linux semantics.
83 */
84
85/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
86 * semantics. See the comments in "open_namei" and "do_link" below.
87 *
88 * [10-Sep-98 Alan Modra] Another symlink change.
89 */
90
91/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
92 * inside the path - always follow.
93 * in the last component in creation/removal/renaming - never follow.
94 * if LOOKUP_FOLLOW passed - follow.
95 * if the pathname has trailing slashes - follow.
96 * otherwise - don't follow.
97 * (applied in that order).
98 *
99 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
100 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
101 * During the 2.4 we need to fix the userland stuff depending on it -
102 * hopefully we will be able to get rid of that wart in 2.5. So far only
103 * XEmacs seems to be relying on it...
104 */
105/*
106 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800107 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 * any extra contention...
109 */
110
111/* In order to reduce some races, while at the same time doing additional
112 * checking and hopefully speeding things up, we copy filenames to the
113 * kernel data space before using them..
114 *
115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
116 * PATH_MAX includes the nul terminator --RR.
117 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800118static int do_getname(const char __user *filename, char *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119{
120 int retval;
121 unsigned long len = PATH_MAX;
122
123 if (!segment_eq(get_fs(), KERNEL_DS)) {
124 if ((unsigned long) filename >= TASK_SIZE)
125 return -EFAULT;
126 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
127 len = TASK_SIZE - (unsigned long) filename;
128 }
129
130 retval = strncpy_from_user(page, filename, len);
131 if (retval > 0) {
132 if (retval < len)
133 return 0;
134 return -ENAMETOOLONG;
135 } else if (!retval)
136 retval = -ENOENT;
137 return retval;
138}
139
140char * getname(const char __user * filename)
141{
142 char *tmp, *result;
143
144 result = ERR_PTR(-ENOMEM);
145 tmp = __getname();
146 if (tmp) {
147 int retval = do_getname(filename, tmp);
148
149 result = tmp;
150 if (retval < 0) {
151 __putname(tmp);
152 result = ERR_PTR(retval);
153 }
154 }
155 audit_getname(result);
156 return result;
157}
158
159#ifdef CONFIG_AUDITSYSCALL
160void putname(const char *name)
161{
162 if (unlikely(current->audit_context))
163 audit_putname(name);
164 else
165 __putname(name);
166}
167EXPORT_SYMBOL(putname);
168#endif
169
170
171/**
172 * generic_permission - check for access rights on a Posix-like filesystem
173 * @inode: inode to check access rights for
174 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
175 * @check_acl: optional callback to check for Posix ACLs
176 *
177 * Used to check for read/write/execute permissions on a file.
178 * We use "fsuid" for this, letting us set arbitrary permissions
179 * for filesystem access without changing the "normal" uids which
180 * are used for other things..
181 */
182int generic_permission(struct inode *inode, int mask,
183 int (*check_acl)(struct inode *inode, int mask))
184{
185 umode_t mode = inode->i_mode;
186
187 if (current->fsuid == inode->i_uid)
188 mode >>= 6;
189 else {
190 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
191 int error = check_acl(inode, mask);
192 if (error == -EACCES)
193 goto check_capabilities;
194 else if (error != -EAGAIN)
195 return error;
196 }
197
198 if (in_group_p(inode->i_gid))
199 mode >>= 3;
200 }
201
202 /*
203 * If the DACs are ok we don't need any capability check.
204 */
205 if (((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask))
206 return 0;
207
208 check_capabilities:
209 /*
210 * Read/write DACs are always overridable.
211 * Executable DACs are overridable if at least one exec bit is set.
212 */
213 if (!(mask & MAY_EXEC) ||
214 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
215 if (capable(CAP_DAC_OVERRIDE))
216 return 0;
217
218 /*
219 * Searching includes executable on directories, else just read.
220 */
221 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
222 if (capable(CAP_DAC_READ_SEARCH))
223 return 0;
224
225 return -EACCES;
226}
227
228int permission(struct inode *inode, int mask, struct nameidata *nd)
229{
230 int retval, submask;
231
232 if (mask & MAY_WRITE) {
233 umode_t mode = inode->i_mode;
234
235 /*
236 * Nobody gets write access to a read-only fs.
237 */
238 if (IS_RDONLY(inode) &&
239 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
240 return -EROFS;
241
242 /*
243 * Nobody gets write access to an immutable file.
244 */
245 if (IS_IMMUTABLE(inode))
246 return -EACCES;
247 }
248
249
250 /* Ordinary permission routines do not understand MAY_APPEND. */
251 submask = mask & ~MAY_APPEND;
252 if (inode->i_op && inode->i_op->permission)
253 retval = inode->i_op->permission(inode, submask, nd);
254 else
255 retval = generic_permission(inode, submask, NULL);
256 if (retval)
257 return retval;
258
259 return security_inode_permission(inode, mask, nd);
260}
261
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800262/**
263 * vfs_permission - check for access rights to a given path
264 * @nd: lookup result that describes the path
265 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
266 *
267 * Used to check for read/write/execute permissions on a path.
268 * We use "fsuid" for this, letting us set arbitrary permissions
269 * for filesystem access without changing the "normal" uids which
270 * are used for other things.
271 */
272int vfs_permission(struct nameidata *nd, int mask)
273{
274 return permission(nd->dentry->d_inode, mask, nd);
275}
276
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800277/**
278 * file_permission - check for additional access rights to a given file
279 * @file: file to check access rights for
280 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
281 *
282 * Used to check for read/write/execute permissions on an already opened
283 * file.
284 *
285 * Note:
286 * Do not use this function in new code. All access checks should
287 * be done using vfs_permission().
288 */
289int file_permission(struct file *file, int mask)
290{
291 return permission(file->f_dentry->d_inode, mask, NULL);
292}
293
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294/*
295 * get_write_access() gets write permission for a file.
296 * put_write_access() releases this write permission.
297 * This is used for regular files.
298 * We cannot support write (and maybe mmap read-write shared) accesses and
299 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
300 * can have the following values:
301 * 0: no writers, no VM_DENYWRITE mappings
302 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
303 * > 0: (i_writecount) users are writing to the file.
304 *
305 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
306 * except for the cases where we don't hold i_writecount yet. Then we need to
307 * use {get,deny}_write_access() - these functions check the sign and refuse
308 * to do the change if sign is wrong. Exclusion between them is provided by
309 * the inode->i_lock spinlock.
310 */
311
312int get_write_access(struct inode * inode)
313{
314 spin_lock(&inode->i_lock);
315 if (atomic_read(&inode->i_writecount) < 0) {
316 spin_unlock(&inode->i_lock);
317 return -ETXTBSY;
318 }
319 atomic_inc(&inode->i_writecount);
320 spin_unlock(&inode->i_lock);
321
322 return 0;
323}
324
325int deny_write_access(struct file * file)
326{
327 struct inode *inode = file->f_dentry->d_inode;
328
329 spin_lock(&inode->i_lock);
330 if (atomic_read(&inode->i_writecount) > 0) {
331 spin_unlock(&inode->i_lock);
332 return -ETXTBSY;
333 }
334 atomic_dec(&inode->i_writecount);
335 spin_unlock(&inode->i_lock);
336
337 return 0;
338}
339
340void path_release(struct nameidata *nd)
341{
342 dput(nd->dentry);
343 mntput(nd->mnt);
344}
345
346/*
347 * umount() mustn't call path_release()/mntput() as that would clear
348 * mnt_expiry_mark
349 */
350void path_release_on_umount(struct nameidata *nd)
351{
352 dput(nd->dentry);
Miklos Szeredi751c4042005-07-07 17:57:30 -0700353 mntput_no_expire(nd->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354}
355
Trond Myklebust834f2a42005-10-18 14:20:16 -0700356/**
357 * release_open_intent - free up open intent resources
358 * @nd: pointer to nameidata
359 */
360void release_open_intent(struct nameidata *nd)
361{
362 if (nd->intent.open.file->f_dentry == NULL)
363 put_filp(nd->intent.open.file);
364 else
365 fput(nd->intent.open.file);
366}
367
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368/*
369 * Internal lookup() using the new generic dcache.
370 * SMP-safe
371 */
372static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
373{
374 struct dentry * dentry = __d_lookup(parent, name);
375
376 /* lockess __d_lookup may fail due to concurrent d_move()
377 * in some unrelated directory, so try with d_lookup
378 */
379 if (!dentry)
380 dentry = d_lookup(parent, name);
381
382 if (dentry && dentry->d_op && dentry->d_op->d_revalidate) {
383 if (!dentry->d_op->d_revalidate(dentry, nd) && !d_invalidate(dentry)) {
384 dput(dentry);
385 dentry = NULL;
386 }
387 }
388 return dentry;
389}
390
391/*
392 * Short-cut version of permission(), for calling by
393 * path_walk(), when dcache lock is held. Combines parts
394 * of permission() and generic_permission(), and tests ONLY for
395 * MAY_EXEC permission.
396 *
397 * If appropriate, check DAC only. If not appropriate, or
398 * short-cut DAC fails, then call permission() to do more
399 * complete permission check.
400 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800401static int exec_permission_lite(struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 struct nameidata *nd)
403{
404 umode_t mode = inode->i_mode;
405
406 if (inode->i_op && inode->i_op->permission)
407 return -EAGAIN;
408
409 if (current->fsuid == inode->i_uid)
410 mode >>= 6;
411 else if (in_group_p(inode->i_gid))
412 mode >>= 3;
413
414 if (mode & MAY_EXEC)
415 goto ok;
416
417 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
418 goto ok;
419
420 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
421 goto ok;
422
423 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
424 goto ok;
425
426 return -EACCES;
427ok:
428 return security_inode_permission(inode, MAY_EXEC, nd);
429}
430
431/*
432 * This is called when everything else fails, and we actually have
433 * to go to the low-level filesystem to find out what we should do..
434 *
435 * We get the directory semaphore, and after getting that we also
436 * make sure that nobody added the entry to the dcache in the meantime..
437 * SMP-safe
438 */
439static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
440{
441 struct dentry * result;
442 struct inode *dir = parent->d_inode;
443
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800444 mutex_lock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 /*
446 * First re-do the cached lookup just in case it was created
447 * while we waited for the directory semaphore..
448 *
449 * FIXME! This could use version numbering or similar to
450 * avoid unnecessary cache lookups.
451 *
452 * The "dcache_lock" is purely to protect the RCU list walker
453 * from concurrent renames at this point (we mustn't get false
454 * negatives from the RCU list walk here, unlike the optimistic
455 * fast walk).
456 *
457 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
458 */
459 result = d_lookup(parent, name);
460 if (!result) {
461 struct dentry * dentry = d_alloc(parent, name);
462 result = ERR_PTR(-ENOMEM);
463 if (dentry) {
464 result = dir->i_op->lookup(dir, dentry, nd);
465 if (result)
466 dput(dentry);
467 else
468 result = dentry;
469 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800470 mutex_unlock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 return result;
472 }
473
474 /*
475 * Uhhuh! Nasty case: the cache was re-populated while
476 * we waited on the semaphore. Need to revalidate.
477 */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800478 mutex_unlock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 if (result->d_op && result->d_op->d_revalidate) {
480 if (!result->d_op->d_revalidate(result, nd) && !d_invalidate(result)) {
481 dput(result);
482 result = ERR_PTR(-ENOENT);
483 }
484 }
485 return result;
486}
487
488static int __emul_lookup_dentry(const char *, struct nameidata *);
489
490/* SMP-safe */
Arjan van de Venf1662352006-01-14 13:21:31 -0800491static __always_inline int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492walk_init_root(const char *name, struct nameidata *nd)
493{
494 read_lock(&current->fs->lock);
495 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
496 nd->mnt = mntget(current->fs->altrootmnt);
497 nd->dentry = dget(current->fs->altroot);
498 read_unlock(&current->fs->lock);
499 if (__emul_lookup_dentry(name,nd))
500 return 0;
501 read_lock(&current->fs->lock);
502 }
503 nd->mnt = mntget(current->fs->rootmnt);
504 nd->dentry = dget(current->fs->root);
505 read_unlock(&current->fs->lock);
506 return 1;
507}
508
Arjan van de Venf1662352006-01-14 13:21:31 -0800509static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510{
511 int res = 0;
512 char *name;
513 if (IS_ERR(link))
514 goto fail;
515
516 if (*link == '/') {
517 path_release(nd);
518 if (!walk_init_root(link, nd))
519 /* weird __emul_prefix() stuff did it */
520 goto out;
521 }
522 res = link_path_walk(link, nd);
523out:
524 if (nd->depth || res || nd->last_type!=LAST_NORM)
525 return res;
526 /*
527 * If it is an iterative symlinks resolution in open_namei() we
528 * have to copy the last component. And all that crap because of
529 * bloody create() on broken symlinks. Furrfu...
530 */
531 name = __getname();
532 if (unlikely(!name)) {
533 path_release(nd);
534 return -ENOMEM;
535 }
536 strcpy(name, nd->last.name);
537 nd->last.name = name;
538 return 0;
539fail:
540 path_release(nd);
541 return PTR_ERR(link);
542}
543
Al Viro90ebe562005-06-06 13:35:58 -0700544struct path {
545 struct vfsmount *mnt;
546 struct dentry *dentry;
547};
548
Arjan van de Venf1662352006-01-14 13:21:31 -0800549static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550{
551 int error;
Linus Torvaldscc314ee2005-08-19 18:02:56 -0700552 void *cookie;
Al Virocd4e91d2005-06-06 13:36:03 -0700553 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554
Al Virod671a1c2005-06-06 13:36:14 -0700555 touch_atime(path->mnt, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 nd_set_link(nd, NULL);
Al Virocd4e91d2005-06-06 13:36:03 -0700557
Al Viro39ca6d42005-06-06 13:36:12 -0700558 if (path->mnt == nd->mnt)
559 mntget(path->mnt);
Linus Torvaldscc314ee2005-08-19 18:02:56 -0700560 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
561 error = PTR_ERR(cookie);
562 if (!IS_ERR(cookie)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 char *s = nd_get_link(nd);
Linus Torvaldscc314ee2005-08-19 18:02:56 -0700564 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 if (s)
566 error = __vfs_follow_link(nd, s);
567 if (dentry->d_inode->i_op->put_link)
Linus Torvaldscc314ee2005-08-19 18:02:56 -0700568 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 }
Al Virocd4e91d2005-06-06 13:36:03 -0700570 dput(dentry);
571 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572
573 return error;
574}
575
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700576static inline void dput_path(struct path *path, struct nameidata *nd)
577{
578 dput(path->dentry);
579 if (path->mnt != nd->mnt)
580 mntput(path->mnt);
581}
582
583static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
584{
585 dput(nd->dentry);
586 if (nd->mnt != path->mnt)
587 mntput(nd->mnt);
588 nd->mnt = path->mnt;
589 nd->dentry = path->dentry;
590}
591
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592/*
593 * This limits recursive symlink follows to 8, while
594 * limiting consecutive symlinks to 40.
595 *
596 * Without that kind of total limit, nasty chains of consecutive
597 * symlinks can cause almost arbitrarily long lookups.
598 */
Al Viro90ebe562005-06-06 13:35:58 -0700599static inline int do_follow_link(struct path *path, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600{
601 int err = -ELOOP;
602 if (current->link_count >= MAX_NESTED_LINKS)
603 goto loop;
604 if (current->total_link_count >= 40)
605 goto loop;
606 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
607 cond_resched();
Al Viro90ebe562005-06-06 13:35:58 -0700608 err = security_inode_follow_link(path->dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 if (err)
610 goto loop;
611 current->link_count++;
612 current->total_link_count++;
613 nd->depth++;
Al Virocd4e91d2005-06-06 13:36:03 -0700614 err = __do_follow_link(path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 current->link_count--;
616 nd->depth--;
617 return err;
618loop:
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700619 dput_path(path, nd);
Al Viro839d9f92005-06-06 13:36:02 -0700620 path_release(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 return err;
622}
623
624int follow_up(struct vfsmount **mnt, struct dentry **dentry)
625{
626 struct vfsmount *parent;
627 struct dentry *mountpoint;
628 spin_lock(&vfsmount_lock);
629 parent=(*mnt)->mnt_parent;
630 if (parent == *mnt) {
631 spin_unlock(&vfsmount_lock);
632 return 0;
633 }
634 mntget(parent);
635 mountpoint=dget((*mnt)->mnt_mountpoint);
636 spin_unlock(&vfsmount_lock);
637 dput(*dentry);
638 *dentry = mountpoint;
639 mntput(*mnt);
640 *mnt = parent;
641 return 1;
642}
643
644/* no need for dcache_lock, as serialization is taken care in
645 * namespace.c
646 */
Al Viro463ffb22005-06-06 13:36:05 -0700647static int __follow_mount(struct path *path)
648{
649 int res = 0;
650 while (d_mountpoint(path->dentry)) {
651 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
652 if (!mounted)
653 break;
654 dput(path->dentry);
655 if (res)
656 mntput(path->mnt);
657 path->mnt = mounted;
658 path->dentry = dget(mounted->mnt_root);
659 res = 1;
660 }
661 return res;
662}
663
Al Viro58c465e2005-06-06 13:36:13 -0700664static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 while (d_mountpoint(*dentry)) {
667 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
668 if (!mounted)
669 break;
Al Viro58c465e2005-06-06 13:36:13 -0700670 dput(*dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 mntput(*mnt);
672 *mnt = mounted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 *dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675}
676
677/* no need for dcache_lock, as serialization is taken care in
678 * namespace.c
679 */
Al Viroe13b2102005-06-06 13:36:06 -0700680int follow_down(struct vfsmount **mnt, struct dentry **dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681{
682 struct vfsmount *mounted;
683
684 mounted = lookup_mnt(*mnt, *dentry);
685 if (mounted) {
Al Viroe13b2102005-06-06 13:36:06 -0700686 dput(*dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 mntput(*mnt);
688 *mnt = mounted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 *dentry = dget(mounted->mnt_root);
690 return 1;
691 }
692 return 0;
693}
694
Arjan van de Venf1662352006-01-14 13:21:31 -0800695static __always_inline void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696{
697 while(1) {
698 struct vfsmount *parent;
Al Viro58c465e2005-06-06 13:36:13 -0700699 struct dentry *old = nd->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700
701 read_lock(&current->fs->lock);
Al Viro58c465e2005-06-06 13:36:13 -0700702 if (nd->dentry == current->fs->root &&
703 nd->mnt == current->fs->rootmnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 read_unlock(&current->fs->lock);
705 break;
706 }
707 read_unlock(&current->fs->lock);
708 spin_lock(&dcache_lock);
Al Viro58c465e2005-06-06 13:36:13 -0700709 if (nd->dentry != nd->mnt->mnt_root) {
710 nd->dentry = dget(nd->dentry->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 spin_unlock(&dcache_lock);
712 dput(old);
713 break;
714 }
715 spin_unlock(&dcache_lock);
716 spin_lock(&vfsmount_lock);
Al Viro58c465e2005-06-06 13:36:13 -0700717 parent = nd->mnt->mnt_parent;
718 if (parent == nd->mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 spin_unlock(&vfsmount_lock);
720 break;
721 }
722 mntget(parent);
Al Viro58c465e2005-06-06 13:36:13 -0700723 nd->dentry = dget(nd->mnt->mnt_mountpoint);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 spin_unlock(&vfsmount_lock);
725 dput(old);
Al Viro58c465e2005-06-06 13:36:13 -0700726 mntput(nd->mnt);
727 nd->mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 }
Al Viro58c465e2005-06-06 13:36:13 -0700729 follow_mount(&nd->mnt, &nd->dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730}
731
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732/*
733 * It's more convoluted than I'd like it to be, but... it's still fairly
734 * small and for now I'd prefer to have fast path as straight as possible.
735 * It _is_ time-critical.
736 */
737static int do_lookup(struct nameidata *nd, struct qstr *name,
738 struct path *path)
739{
740 struct vfsmount *mnt = nd->mnt;
741 struct dentry *dentry = __d_lookup(nd->dentry, name);
742
743 if (!dentry)
744 goto need_lookup;
745 if (dentry->d_op && dentry->d_op->d_revalidate)
746 goto need_revalidate;
747done:
748 path->mnt = mnt;
749 path->dentry = dentry;
Al Viro634ee702005-06-06 13:36:13 -0700750 __follow_mount(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 return 0;
752
753need_lookup:
754 dentry = real_lookup(nd->dentry, name, nd);
755 if (IS_ERR(dentry))
756 goto fail;
757 goto done;
758
759need_revalidate:
760 if (dentry->d_op->d_revalidate(dentry, nd))
761 goto done;
762 if (d_invalidate(dentry))
763 goto done;
764 dput(dentry);
765 goto need_lookup;
766
767fail:
768 return PTR_ERR(dentry);
769}
770
771/*
772 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +0100773 * This is the basic name resolution function, turning a pathname into
774 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 *
Prasanna Medaea3834d2005-04-29 16:00:17 +0100776 * Returns 0 and nd will have valid dentry and mnt on success.
777 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 */
779static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
780{
781 struct path next;
782 struct inode *inode;
783 int err;
784 unsigned int lookup_flags = nd->flags;
785
786 while (*name=='/')
787 name++;
788 if (!*name)
789 goto return_reval;
790
791 inode = nd->dentry->d_inode;
792 if (nd->depth)
Trond Myklebustf55eab82006-02-04 23:28:01 -0800793 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794
795 /* At this point we know we have a real path component. */
796 for(;;) {
797 unsigned long hash;
798 struct qstr this;
799 unsigned int c;
800
Trond Myklebustcdce5d62005-10-18 14:20:18 -0700801 nd->flags |= LOOKUP_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 err = exec_permission_lite(inode, nd);
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800803 if (err == -EAGAIN)
804 err = vfs_permission(nd, MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 if (err)
806 break;
807
808 this.name = name;
809 c = *(const unsigned char *)name;
810
811 hash = init_name_hash();
812 do {
813 name++;
814 hash = partial_name_hash(c, hash);
815 c = *(const unsigned char *)name;
816 } while (c && (c != '/'));
817 this.len = name - (const char *) this.name;
818 this.hash = end_name_hash(hash);
819
820 /* remove trailing slashes? */
821 if (!c)
822 goto last_component;
823 while (*++name == '/');
824 if (!*name)
825 goto last_with_slashes;
826
827 /*
828 * "." and ".." are special - ".." especially so because it has
829 * to be able to know about the current root directory and
830 * parent relationships.
831 */
832 if (this.name[0] == '.') switch (this.len) {
833 default:
834 break;
835 case 2:
836 if (this.name[1] != '.')
837 break;
Al Viro58c465e2005-06-06 13:36:13 -0700838 follow_dotdot(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 inode = nd->dentry->d_inode;
840 /* fallthrough */
841 case 1:
842 continue;
843 }
844 /*
845 * See if the low-level filesystem might want
846 * to use its own hash..
847 */
848 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
849 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
850 if (err < 0)
851 break;
852 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 /* This does the actual lookups.. */
854 err = do_lookup(nd, &this, &next);
855 if (err)
856 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857
858 err = -ENOENT;
859 inode = next.dentry->d_inode;
860 if (!inode)
861 goto out_dput;
862 err = -ENOTDIR;
863 if (!inode->i_op)
864 goto out_dput;
865
866 if (inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700867 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 if (err)
869 goto return_err;
870 err = -ENOENT;
871 inode = nd->dentry->d_inode;
872 if (!inode)
873 break;
874 err = -ENOTDIR;
875 if (!inode->i_op)
876 break;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700877 } else
878 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 err = -ENOTDIR;
880 if (!inode->i_op->lookup)
881 break;
882 continue;
883 /* here ends the main loop */
884
885last_with_slashes:
886 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
887last_component:
Trond Myklebustf55eab82006-02-04 23:28:01 -0800888 /* Clear LOOKUP_CONTINUE iff it was previously unset */
889 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 if (lookup_flags & LOOKUP_PARENT)
891 goto lookup_parent;
892 if (this.name[0] == '.') switch (this.len) {
893 default:
894 break;
895 case 2:
896 if (this.name[1] != '.')
897 break;
Al Viro58c465e2005-06-06 13:36:13 -0700898 follow_dotdot(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 inode = nd->dentry->d_inode;
900 /* fallthrough */
901 case 1:
902 goto return_reval;
903 }
904 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
905 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
906 if (err < 0)
907 break;
908 }
909 err = do_lookup(nd, &this, &next);
910 if (err)
911 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 inode = next.dentry->d_inode;
913 if ((lookup_flags & LOOKUP_FOLLOW)
914 && inode && inode->i_op && inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700915 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 if (err)
917 goto return_err;
918 inode = nd->dentry->d_inode;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700919 } else
920 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 err = -ENOENT;
922 if (!inode)
923 break;
924 if (lookup_flags & LOOKUP_DIRECTORY) {
925 err = -ENOTDIR;
926 if (!inode->i_op || !inode->i_op->lookup)
927 break;
928 }
929 goto return_base;
930lookup_parent:
931 nd->last = this;
932 nd->last_type = LAST_NORM;
933 if (this.name[0] != '.')
934 goto return_base;
935 if (this.len == 1)
936 nd->last_type = LAST_DOT;
937 else if (this.len == 2 && this.name[1] == '.')
938 nd->last_type = LAST_DOTDOT;
939 else
940 goto return_base;
941return_reval:
942 /*
943 * We bypassed the ordinary revalidation routines.
944 * We may need to check the cached dentry for staleness.
945 */
946 if (nd->dentry && nd->dentry->d_sb &&
947 (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
948 err = -ESTALE;
949 /* Note: we do not d_invalidate() */
950 if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
951 break;
952 }
953return_base:
954 return 0;
955out_dput:
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700956 dput_path(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 break;
958 }
959 path_release(nd);
960return_err:
961 return err;
962}
963
964/*
965 * Wrapper to retry pathname resolution whenever the underlying
966 * file system returns an ESTALE.
967 *
968 * Retry the whole path once, forcing real lookup requests
969 * instead of relying on the dcache.
970 */
971int fastcall link_path_walk(const char *name, struct nameidata *nd)
972{
973 struct nameidata save = *nd;
974 int result;
975
976 /* make sure the stuff we saved doesn't go away */
977 dget(save.dentry);
978 mntget(save.mnt);
979
980 result = __link_path_walk(name, nd);
981 if (result == -ESTALE) {
982 *nd = save;
983 dget(nd->dentry);
984 mntget(nd->mnt);
985 nd->flags |= LOOKUP_REVAL;
986 result = __link_path_walk(name, nd);
987 }
988
989 dput(save.dentry);
990 mntput(save.mnt);
991
992 return result;
993}
994
995int fastcall path_walk(const char * name, struct nameidata *nd)
996{
997 current->total_link_count = 0;
998 return link_path_walk(name, nd);
999}
1000
Prasanna Medaea3834d2005-04-29 16:00:17 +01001001/*
1002 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
1003 * everything is done. Returns 0 and drops input nd, if lookup failed;
1004 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
1006{
1007 if (path_walk(name, nd))
1008 return 0; /* something went wrong... */
1009
1010 if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
1011 struct dentry *old_dentry = nd->dentry;
1012 struct vfsmount *old_mnt = nd->mnt;
1013 struct qstr last = nd->last;
1014 int last_type = nd->last_type;
1015 /*
1016 * NAME was not found in alternate root or it's a directory. Try to find
1017 * it in the normal root:
1018 */
1019 nd->last_type = LAST_ROOT;
1020 read_lock(&current->fs->lock);
1021 nd->mnt = mntget(current->fs->rootmnt);
1022 nd->dentry = dget(current->fs->root);
1023 read_unlock(&current->fs->lock);
1024 if (path_walk(name, nd) == 0) {
1025 if (nd->dentry->d_inode) {
1026 dput(old_dentry);
1027 mntput(old_mnt);
1028 return 1;
1029 }
1030 path_release(nd);
1031 }
1032 nd->dentry = old_dentry;
1033 nd->mnt = old_mnt;
1034 nd->last = last;
1035 nd->last_type = last_type;
1036 }
1037 return 1;
1038}
1039
1040void set_fs_altroot(void)
1041{
1042 char *emul = __emul_prefix();
1043 struct nameidata nd;
1044 struct vfsmount *mnt = NULL, *oldmnt;
1045 struct dentry *dentry = NULL, *olddentry;
1046 int err;
1047
1048 if (!emul)
1049 goto set_it;
1050 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
1051 if (!err) {
1052 mnt = nd.mnt;
1053 dentry = nd.dentry;
1054 }
1055set_it:
1056 write_lock(&current->fs->lock);
1057 oldmnt = current->fs->altrootmnt;
1058 olddentry = current->fs->altroot;
1059 current->fs->altrootmnt = mnt;
1060 current->fs->altroot = dentry;
1061 write_unlock(&current->fs->lock);
1062 if (olddentry) {
1063 dput(olddentry);
1064 mntput(oldmnt);
1065 }
1066}
1067
Prasanna Medaea3834d2005-04-29 16:00:17 +01001068/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001069static int fastcall do_path_lookup(int dfd, const char *name,
1070 unsigned int flags, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001072 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001073 int fput_needed;
1074 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075
1076 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1077 nd->flags = flags;
1078 nd->depth = 0;
1079
1080 read_lock(&current->fs->lock);
1081 if (*name=='/') {
1082 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
1083 nd->mnt = mntget(current->fs->altrootmnt);
1084 nd->dentry = dget(current->fs->altroot);
1085 read_unlock(&current->fs->lock);
1086 if (__emul_lookup_dentry(name,nd))
Prasanna Medaea3834d2005-04-29 16:00:17 +01001087 goto out; /* found in altroot */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 read_lock(&current->fs->lock);
1089 }
1090 nd->mnt = mntget(current->fs->rootmnt);
1091 nd->dentry = dget(current->fs->root);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001092 } else if (dfd == AT_FDCWD) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 nd->mnt = mntget(current->fs->pwdmnt);
1094 nd->dentry = dget(current->fs->pwd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001095 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001096 struct dentry *dentry;
1097
1098 file = fget_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001099 retval = -EBADF;
1100 if (!file)
1101 goto unlock_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001102
1103 dentry = file->f_dentry;
1104
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001105 retval = -ENOTDIR;
1106 if (!S_ISDIR(dentry->d_inode->i_mode))
1107 goto fput_unlock_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001108
1109 retval = file_permission(file, MAY_EXEC);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001110 if (retval)
1111 goto fput_unlock_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001112
1113 nd->mnt = mntget(file->f_vfsmnt);
1114 nd->dentry = dget(dentry);
1115
1116 fput_light(file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 }
1118 read_unlock(&current->fs->lock);
1119 current->total_link_count = 0;
1120 retval = link_path_walk(name, nd);
Prasanna Medaea3834d2005-04-29 16:00:17 +01001121out:
Suzuki3bc8414b2006-02-07 12:58:36 -08001122 if (likely(retval == 0)) {
1123 if (unlikely(current->audit_context && nd && nd->dentry &&
1124 nd->dentry->d_inode))
David Woodhouseae7b9612005-06-20 16:11:05 +01001125 audit_inode(name, nd->dentry->d_inode, flags);
Suzuki3bc8414b2006-02-07 12:58:36 -08001126 }
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001127 return retval;
1128
1129fput_unlock_fail:
1130 fput_light(file, fput_needed);
1131unlock_fail:
1132 read_unlock(&current->fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 return retval;
1134}
1135
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001136int fastcall path_lookup(const char *name, unsigned int flags,
1137 struct nameidata *nd)
1138{
1139 return do_path_lookup(AT_FDCWD, name, flags, nd);
1140}
1141
1142static int __path_lookup_intent_open(int dfd, const char *name,
1143 unsigned int lookup_flags, struct nameidata *nd,
1144 int open_flags, int create_mode)
Trond Myklebust834f2a42005-10-18 14:20:16 -07001145{
1146 struct file *filp = get_empty_filp();
1147 int err;
1148
1149 if (filp == NULL)
1150 return -ENFILE;
1151 nd->intent.open.file = filp;
1152 nd->intent.open.flags = open_flags;
1153 nd->intent.open.create_mode = create_mode;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001154 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
Trond Myklebust834f2a42005-10-18 14:20:16 -07001155 if (IS_ERR(nd->intent.open.file)) {
1156 if (err == 0) {
1157 err = PTR_ERR(nd->intent.open.file);
1158 path_release(nd);
1159 }
1160 } else if (err != 0)
1161 release_open_intent(nd);
1162 return err;
1163}
1164
1165/**
1166 * path_lookup_open - lookup a file path with open intent
Martin Waitz7045f372006-02-01 03:06:57 -08001167 * @dfd: the directory to use as base, or AT_FDCWD
Trond Myklebust834f2a42005-10-18 14:20:16 -07001168 * @name: pointer to file name
1169 * @lookup_flags: lookup intent flags
1170 * @nd: pointer to nameidata
1171 * @open_flags: open intent flags
1172 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001173int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
Trond Myklebust834f2a42005-10-18 14:20:16 -07001174 struct nameidata *nd, int open_flags)
1175{
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001176 return __path_lookup_intent_open(dfd, name, lookup_flags, nd,
Trond Myklebust834f2a42005-10-18 14:20:16 -07001177 open_flags, 0);
1178}
1179
1180/**
1181 * path_lookup_create - lookup a file path with open + create intent
Martin Waitz7045f372006-02-01 03:06:57 -08001182 * @dfd: the directory to use as base, or AT_FDCWD
Trond Myklebust834f2a42005-10-18 14:20:16 -07001183 * @name: pointer to file name
1184 * @lookup_flags: lookup intent flags
1185 * @nd: pointer to nameidata
1186 * @open_flags: open intent flags
1187 * @create_mode: create intent flags
1188 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001189static int path_lookup_create(int dfd, const char *name,
1190 unsigned int lookup_flags, struct nameidata *nd,
1191 int open_flags, int create_mode)
Trond Myklebust834f2a42005-10-18 14:20:16 -07001192{
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001193 return __path_lookup_intent_open(dfd, name, lookup_flags|LOOKUP_CREATE,
1194 nd, open_flags, create_mode);
Trond Myklebust834f2a42005-10-18 14:20:16 -07001195}
1196
1197int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags,
1198 struct nameidata *nd, int open_flags)
1199{
1200 char *tmp = getname(name);
1201 int err = PTR_ERR(tmp);
1202
1203 if (!IS_ERR(tmp)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001204 err = __path_lookup_intent_open(AT_FDCWD, tmp, lookup_flags, nd, open_flags, 0);
Trond Myklebust834f2a42005-10-18 14:20:16 -07001205 putname(tmp);
1206 }
1207 return err;
1208}
1209
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210/*
1211 * Restricted form of lookup. Doesn't follow links, single-component only,
1212 * needs parent already locked. Doesn't follow mounts.
1213 * SMP-safe.
1214 */
1215static struct dentry * __lookup_hash(struct qstr *name, struct dentry * base, struct nameidata *nd)
1216{
1217 struct dentry * dentry;
1218 struct inode *inode;
1219 int err;
1220
1221 inode = base->d_inode;
1222 err = permission(inode, MAY_EXEC, nd);
1223 dentry = ERR_PTR(err);
1224 if (err)
1225 goto out;
1226
1227 /*
1228 * See if the low-level filesystem might want
1229 * to use its own hash..
1230 */
1231 if (base->d_op && base->d_op->d_hash) {
1232 err = base->d_op->d_hash(base, name);
1233 dentry = ERR_PTR(err);
1234 if (err < 0)
1235 goto out;
1236 }
1237
1238 dentry = cached_lookup(base, name, nd);
1239 if (!dentry) {
1240 struct dentry *new = d_alloc(base, name);
1241 dentry = ERR_PTR(-ENOMEM);
1242 if (!new)
1243 goto out;
1244 dentry = inode->i_op->lookup(inode, new, nd);
1245 if (!dentry)
1246 dentry = new;
1247 else
1248 dput(new);
1249 }
1250out:
1251 return dentry;
1252}
1253
Christoph Hellwig49705b72005-11-08 21:35:06 -08001254struct dentry * lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255{
Christoph Hellwig49705b72005-11-08 21:35:06 -08001256 return __lookup_hash(&nd->last, nd->dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257}
1258
1259/* SMP-safe */
1260struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
1261{
1262 unsigned long hash;
1263 struct qstr this;
1264 unsigned int c;
1265
1266 this.name = name;
1267 this.len = len;
1268 if (!len)
1269 goto access;
1270
1271 hash = init_name_hash();
1272 while (len--) {
1273 c = *(const unsigned char *)name++;
1274 if (c == '/' || c == '\0')
1275 goto access;
1276 hash = partial_name_hash(c, hash);
1277 }
1278 this.hash = end_name_hash(hash);
1279
Christoph Hellwig49705b72005-11-08 21:35:06 -08001280 return __lookup_hash(&this, base, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281access:
1282 return ERR_PTR(-EACCES);
1283}
1284
1285/*
1286 * namei()
1287 *
1288 * is used by most simple commands to get the inode of a specified name.
1289 * Open, link etc use their own routines, but this is enough for things
1290 * like 'chmod' etc.
1291 *
1292 * namei exists in two versions: namei/lnamei. The only difference is
1293 * that namei follows links, while lnamei does not.
1294 * SMP-safe
1295 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001296int fastcall __user_walk_fd(int dfd, const char __user *name, unsigned flags,
1297 struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298{
1299 char *tmp = getname(name);
1300 int err = PTR_ERR(tmp);
1301
1302 if (!IS_ERR(tmp)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001303 err = do_path_lookup(dfd, tmp, flags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 putname(tmp);
1305 }
1306 return err;
1307}
1308
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001309int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1310{
1311 return __user_walk_fd(AT_FDCWD, name, flags, nd);
1312}
1313
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314/*
1315 * It's inline, so penalty for filesystems that don't use sticky bit is
1316 * minimal.
1317 */
1318static inline int check_sticky(struct inode *dir, struct inode *inode)
1319{
1320 if (!(dir->i_mode & S_ISVTX))
1321 return 0;
1322 if (inode->i_uid == current->fsuid)
1323 return 0;
1324 if (dir->i_uid == current->fsuid)
1325 return 0;
1326 return !capable(CAP_FOWNER);
1327}
1328
1329/*
1330 * Check whether we can remove a link victim from directory dir, check
1331 * whether the type of victim is right.
1332 * 1. We can't do it if dir is read-only (done in permission())
1333 * 2. We should have write and exec permissions on dir
1334 * 3. We can't remove anything from append-only dir
1335 * 4. We can't do anything with immutable dir (done in permission())
1336 * 5. If the sticky bit on dir is set we should either
1337 * a. be owner of dir, or
1338 * b. be owner of victim, or
1339 * c. have CAP_FOWNER capability
1340 * 6. If the victim is append-only or immutable we can't do antyhing with
1341 * links pointing to it.
1342 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1343 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1344 * 9. We can't remove a root or mountpoint.
1345 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1346 * nfs_async_unlink().
1347 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001348static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349{
1350 int error;
1351
1352 if (!victim->d_inode)
1353 return -ENOENT;
1354
1355 BUG_ON(victim->d_parent->d_inode != dir);
Amy Griffis73241cc2005-11-03 16:00:25 +00001356 audit_inode_child(victim->d_name.name, victim->d_inode, dir->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357
1358 error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
1359 if (error)
1360 return error;
1361 if (IS_APPEND(dir))
1362 return -EPERM;
1363 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1364 IS_IMMUTABLE(victim->d_inode))
1365 return -EPERM;
1366 if (isdir) {
1367 if (!S_ISDIR(victim->d_inode->i_mode))
1368 return -ENOTDIR;
1369 if (IS_ROOT(victim))
1370 return -EBUSY;
1371 } else if (S_ISDIR(victim->d_inode->i_mode))
1372 return -EISDIR;
1373 if (IS_DEADDIR(dir))
1374 return -ENOENT;
1375 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1376 return -EBUSY;
1377 return 0;
1378}
1379
1380/* Check whether we can create an object with dentry child in directory
1381 * dir.
1382 * 1. We can't do it if child already exists (open has special treatment for
1383 * this case, but since we are inlined it's OK)
1384 * 2. We can't do it if dir is read-only (done in permission())
1385 * 3. We should have write and exec permissions on dir
1386 * 4. We can't do it if dir is immutable (done in permission())
1387 */
1388static inline int may_create(struct inode *dir, struct dentry *child,
1389 struct nameidata *nd)
1390{
1391 if (child->d_inode)
1392 return -EEXIST;
1393 if (IS_DEADDIR(dir))
1394 return -ENOENT;
1395 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1396}
1397
1398/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 * O_DIRECTORY translates into forcing a directory lookup.
1400 */
1401static inline int lookup_flags(unsigned int f)
1402{
1403 unsigned long retval = LOOKUP_FOLLOW;
1404
1405 if (f & O_NOFOLLOW)
1406 retval &= ~LOOKUP_FOLLOW;
1407
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 if (f & O_DIRECTORY)
1409 retval |= LOOKUP_DIRECTORY;
1410
1411 return retval;
1412}
1413
1414/*
1415 * p1 and p2 should be directories on the same fs.
1416 */
1417struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1418{
1419 struct dentry *p;
1420
1421 if (p1 == p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001422 mutex_lock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 return NULL;
1424 }
1425
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001426 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427
1428 for (p = p1; p->d_parent != p; p = p->d_parent) {
1429 if (p->d_parent == p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001430 mutex_lock(&p2->d_inode->i_mutex);
1431 mutex_lock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 return p;
1433 }
1434 }
1435
1436 for (p = p2; p->d_parent != p; p = p->d_parent) {
1437 if (p->d_parent == p1) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001438 mutex_lock(&p1->d_inode->i_mutex);
1439 mutex_lock(&p2->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 return p;
1441 }
1442 }
1443
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001444 mutex_lock(&p1->d_inode->i_mutex);
1445 mutex_lock(&p2->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 return NULL;
1447}
1448
1449void unlock_rename(struct dentry *p1, struct dentry *p2)
1450{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001451 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001453 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001454 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 }
1456}
1457
1458int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1459 struct nameidata *nd)
1460{
1461 int error = may_create(dir, dentry, nd);
1462
1463 if (error)
1464 return error;
1465
1466 if (!dir->i_op || !dir->i_op->create)
1467 return -EACCES; /* shouldn't it be ENOSYS? */
1468 mode &= S_IALLUGO;
1469 mode |= S_IFREG;
1470 error = security_inode_create(dir, dentry, mode);
1471 if (error)
1472 return error;
1473 DQUOT_INIT(dir);
1474 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001475 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001476 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 return error;
1478}
1479
1480int may_open(struct nameidata *nd, int acc_mode, int flag)
1481{
1482 struct dentry *dentry = nd->dentry;
1483 struct inode *inode = dentry->d_inode;
1484 int error;
1485
1486 if (!inode)
1487 return -ENOENT;
1488
1489 if (S_ISLNK(inode->i_mode))
1490 return -ELOOP;
1491
1492 if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1493 return -EISDIR;
1494
Christoph Hellwige4543ed2005-11-08 21:35:04 -08001495 error = vfs_permission(nd, acc_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 if (error)
1497 return error;
1498
1499 /*
1500 * FIFO's, sockets and device files are special: they don't
1501 * actually live on the filesystem itself, and as such you
1502 * can write to them even if the filesystem is read-only.
1503 */
1504 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1505 flag &= ~O_TRUNC;
1506 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1507 if (nd->mnt->mnt_flags & MNT_NODEV)
1508 return -EACCES;
1509
1510 flag &= ~O_TRUNC;
1511 } else if (IS_RDONLY(inode) && (flag & FMODE_WRITE))
1512 return -EROFS;
1513 /*
1514 * An append-only file must be opened in append mode for writing.
1515 */
1516 if (IS_APPEND(inode)) {
1517 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1518 return -EPERM;
1519 if (flag & O_TRUNC)
1520 return -EPERM;
1521 }
1522
1523 /* O_NOATIME can only be set by the owner or superuser */
1524 if (flag & O_NOATIME)
1525 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1526 return -EPERM;
1527
1528 /*
1529 * Ensure there are no outstanding leases on the file.
1530 */
1531 error = break_lease(inode, flag);
1532 if (error)
1533 return error;
1534
1535 if (flag & O_TRUNC) {
1536 error = get_write_access(inode);
1537 if (error)
1538 return error;
1539
1540 /*
1541 * Refuse to truncate files with mandatory locks held on them.
1542 */
1543 error = locks_verify_locked(inode);
1544 if (!error) {
1545 DQUOT_INIT(inode);
1546
NeilBrown4a301312006-01-08 01:02:39 -08001547 error = do_truncate(dentry, 0, ATTR_MTIME|ATTR_CTIME, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 }
1549 put_write_access(inode);
1550 if (error)
1551 return error;
1552 } else
1553 if (flag & FMODE_WRITE)
1554 DQUOT_INIT(inode);
1555
1556 return 0;
1557}
1558
1559/*
1560 * open_namei()
1561 *
1562 * namei for open - this is in fact almost the whole open-routine.
1563 *
1564 * Note that the low bits of "flag" aren't the same as in the open
1565 * system call - they are 00 - no permissions needed
1566 * 01 - read permission needed
1567 * 10 - write permission needed
1568 * 11 - read/write permissions needed
1569 * which is a lot more logical, and also allows the "no perm" needed
1570 * for symlinks (where the permissions are checked later).
1571 * SMP-safe
1572 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001573int open_namei(int dfd, const char *pathname, int flag,
1574 int mode, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575{
Trond Myklebust834f2a42005-10-18 14:20:16 -07001576 int acc_mode, error;
Al Viro4e7506e2005-06-06 13:36:00 -07001577 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 struct dentry *dir;
1579 int count = 0;
1580
1581 acc_mode = ACC_MODE(flag);
1582
Trond Myklebust834f2a42005-10-18 14:20:16 -07001583 /* O_TRUNC implies we need access checks for write permissions */
1584 if (flag & O_TRUNC)
1585 acc_mode |= MAY_WRITE;
1586
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 /* Allow the LSM permission hook to distinguish append
1588 access from general write access. */
1589 if (flag & O_APPEND)
1590 acc_mode |= MAY_APPEND;
1591
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 /*
1593 * The simplest case - just a plain lookup.
1594 */
1595 if (!(flag & O_CREAT)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001596 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1597 nd, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 if (error)
1599 return error;
1600 goto ok;
1601 }
1602
1603 /*
1604 * Create - we need to know the parent.
1605 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001606 error = path_lookup_create(dfd,pathname,LOOKUP_PARENT,nd,flag,mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 if (error)
1608 return error;
1609
1610 /*
1611 * We have the parent and last component. First of all, check
1612 * that we are not asked to creat(2) an obvious directory - that
1613 * will not do.
1614 */
1615 error = -EISDIR;
1616 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1617 goto exit;
1618
1619 dir = nd->dentry;
1620 nd->flags &= ~LOOKUP_PARENT;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001621 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001622 path.dentry = lookup_hash(nd);
Al Virod73ffe12005-06-06 13:36:01 -07001623 path.mnt = nd->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624
1625do_last:
Al Viro4e7506e2005-06-06 13:36:00 -07001626 error = PTR_ERR(path.dentry);
1627 if (IS_ERR(path.dentry)) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001628 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 goto exit;
1630 }
1631
Oleg Drokin4af4c522006-03-25 03:06:54 -08001632 if (IS_ERR(nd->intent.open.file)) {
1633 mutex_unlock(&dir->d_inode->i_mutex);
1634 error = PTR_ERR(nd->intent.open.file);
1635 goto exit_dput;
1636 }
1637
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 /* Negative dentry, just create the file */
Al Viro4e7506e2005-06-06 13:36:00 -07001639 if (!path.dentry->d_inode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 if (!IS_POSIXACL(dir->d_inode))
1641 mode &= ~current->fs->umask;
Al Viro4e7506e2005-06-06 13:36:00 -07001642 error = vfs_create(dir->d_inode, path.dentry, mode, nd);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001643 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 dput(nd->dentry);
Al Viro4e7506e2005-06-06 13:36:00 -07001645 nd->dentry = path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 if (error)
1647 goto exit;
1648 /* Don't check for write permission, don't truncate */
1649 acc_mode = 0;
1650 flag &= ~O_TRUNC;
1651 goto ok;
1652 }
1653
1654 /*
1655 * It already exists.
1656 */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001657 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658
1659 error = -EEXIST;
1660 if (flag & O_EXCL)
1661 goto exit_dput;
1662
Al Viroe13b2102005-06-06 13:36:06 -07001663 if (__follow_mount(&path)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 error = -ELOOP;
Al Viroba7a4c12005-06-06 13:36:08 -07001665 if (flag & O_NOFOLLOW)
1666 goto exit_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 }
1668 error = -ENOENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001669 if (!path.dentry->d_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 goto exit_dput;
Al Viro4e7506e2005-06-06 13:36:00 -07001671 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 goto do_link;
1673
Miklos Szeredi09dd17d2005-09-06 15:18:21 -07001674 path_to_nameidata(&path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 error = -EISDIR;
Al Viro4e7506e2005-06-06 13:36:00 -07001676 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 goto exit;
1678ok:
1679 error = may_open(nd, acc_mode, flag);
1680 if (error)
1681 goto exit;
1682 return 0;
1683
1684exit_dput:
Miklos Szeredi09dd17d2005-09-06 15:18:21 -07001685 dput_path(&path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686exit:
Trond Myklebust834f2a42005-10-18 14:20:16 -07001687 if (!IS_ERR(nd->intent.open.file))
1688 release_open_intent(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 path_release(nd);
1690 return error;
1691
1692do_link:
1693 error = -ELOOP;
1694 if (flag & O_NOFOLLOW)
1695 goto exit_dput;
1696 /*
1697 * This is subtle. Instead of calling do_follow_link() we do the
1698 * thing by hands. The reason is that this way we have zero link_count
1699 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1700 * After that we have the parent and last component, i.e.
1701 * we are in the same situation as after the first path_walk().
1702 * Well, almost - if the last component is normal we get its copy
1703 * stored in nd->last.name and we will have to putname() it when we
1704 * are done. Procfs-like symlinks just set LAST_BIND.
1705 */
1706 nd->flags |= LOOKUP_PARENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001707 error = security_inode_follow_link(path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 if (error)
1709 goto exit_dput;
Al Virocd4e91d2005-06-06 13:36:03 -07001710 error = __do_follow_link(&path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 if (error)
1712 return error;
1713 nd->flags &= ~LOOKUP_PARENT;
Al Virod671d5e2005-06-06 13:36:04 -07001714 if (nd->last_type == LAST_BIND)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 goto ok;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 error = -EISDIR;
1717 if (nd->last_type != LAST_NORM)
1718 goto exit;
1719 if (nd->last.name[nd->last.len]) {
Linus Torvalds82984112005-10-06 21:54:21 -07001720 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 goto exit;
1722 }
1723 error = -ELOOP;
1724 if (count++==32) {
Linus Torvalds82984112005-10-06 21:54:21 -07001725 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 goto exit;
1727 }
1728 dir = nd->dentry;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001729 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001730 path.dentry = lookup_hash(nd);
Al Virod671d5e2005-06-06 13:36:04 -07001731 path.mnt = nd->mnt;
Linus Torvalds82984112005-10-06 21:54:21 -07001732 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 goto do_last;
1734}
1735
1736/**
1737 * lookup_create - lookup a dentry, creating it if it doesn't exist
1738 * @nd: nameidata info
1739 * @is_dir: directory flag
1740 *
1741 * Simple function to lookup and return a dentry and create it
1742 * if it doesn't exist. Is SMP-safe.
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001743 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001744 * Returns with nd->dentry->d_inode->i_mutex locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745 */
1746struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1747{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001748 struct dentry *dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001750 mutex_lock(&nd->dentry->d_inode->i_mutex);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001751 /*
1752 * Yucky last component or no last component at all?
1753 * (foo/., foo/.., /////)
1754 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 if (nd->last_type != LAST_NORM)
1756 goto fail;
1757 nd->flags &= ~LOOKUP_PARENT;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001758
1759 /*
1760 * Do the final lookup.
1761 */
Christoph Hellwig49705b72005-11-08 21:35:06 -08001762 dentry = lookup_hash(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 if (IS_ERR(dentry))
1764 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001765
1766 /*
1767 * Special case - lookup gave negative, but... we had foo/bar/
1768 * From the vfs_mknod() POV we just have a negative dentry -
1769 * all is fine. Let's be bastards - you had / on the end, you've
1770 * been asking for (non-existent) directory. -ENOENT for you.
1771 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1773 goto enoent;
1774 return dentry;
1775enoent:
1776 dput(dentry);
1777 dentry = ERR_PTR(-ENOENT);
1778fail:
1779 return dentry;
1780}
Christoph Hellwigf81a0bf2005-05-19 12:26:43 -07001781EXPORT_SYMBOL_GPL(lookup_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782
1783int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1784{
1785 int error = may_create(dir, dentry, NULL);
1786
1787 if (error)
1788 return error;
1789
1790 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1791 return -EPERM;
1792
1793 if (!dir->i_op || !dir->i_op->mknod)
1794 return -EPERM;
1795
1796 error = security_inode_mknod(dir, dentry, mode, dev);
1797 if (error)
1798 return error;
1799
1800 DQUOT_INIT(dir);
1801 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001802 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001803 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 return error;
1805}
1806
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001807asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
1808 unsigned dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809{
1810 int error = 0;
1811 char * tmp;
1812 struct dentry * dentry;
1813 struct nameidata nd;
1814
1815 if (S_ISDIR(mode))
1816 return -EPERM;
1817 tmp = getname(filename);
1818 if (IS_ERR(tmp))
1819 return PTR_ERR(tmp);
1820
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001821 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822 if (error)
1823 goto out;
1824 dentry = lookup_create(&nd, 0);
1825 error = PTR_ERR(dentry);
1826
1827 if (!IS_POSIXACL(nd.dentry->d_inode))
1828 mode &= ~current->fs->umask;
1829 if (!IS_ERR(dentry)) {
1830 switch (mode & S_IFMT) {
1831 case 0: case S_IFREG:
1832 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1833 break;
1834 case S_IFCHR: case S_IFBLK:
1835 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,
1836 new_decode_dev(dev));
1837 break;
1838 case S_IFIFO: case S_IFSOCK:
1839 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,0);
1840 break;
1841 case S_IFDIR:
1842 error = -EPERM;
1843 break;
1844 default:
1845 error = -EINVAL;
1846 }
1847 dput(dentry);
1848 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001849 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 path_release(&nd);
1851out:
1852 putname(tmp);
1853
1854 return error;
1855}
1856
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001857asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
1858{
1859 return sys_mknodat(AT_FDCWD, filename, mode, dev);
1860}
1861
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1863{
1864 int error = may_create(dir, dentry, NULL);
1865
1866 if (error)
1867 return error;
1868
1869 if (!dir->i_op || !dir->i_op->mkdir)
1870 return -EPERM;
1871
1872 mode &= (S_IRWXUGO|S_ISVTX);
1873 error = security_inode_mkdir(dir, dentry, mode);
1874 if (error)
1875 return error;
1876
1877 DQUOT_INIT(dir);
1878 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001879 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001880 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 return error;
1882}
1883
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001884asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885{
1886 int error = 0;
1887 char * tmp;
1888
1889 tmp = getname(pathname);
1890 error = PTR_ERR(tmp);
1891 if (!IS_ERR(tmp)) {
1892 struct dentry *dentry;
1893 struct nameidata nd;
1894
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001895 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896 if (error)
1897 goto out;
1898 dentry = lookup_create(&nd, 1);
1899 error = PTR_ERR(dentry);
1900 if (!IS_ERR(dentry)) {
1901 if (!IS_POSIXACL(nd.dentry->d_inode))
1902 mode &= ~current->fs->umask;
1903 error = vfs_mkdir(nd.dentry->d_inode, dentry, mode);
1904 dput(dentry);
1905 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001906 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907 path_release(&nd);
1908out:
1909 putname(tmp);
1910 }
1911
1912 return error;
1913}
1914
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001915asmlinkage long sys_mkdir(const char __user *pathname, int mode)
1916{
1917 return sys_mkdirat(AT_FDCWD, pathname, mode);
1918}
1919
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920/*
1921 * We try to drop the dentry early: we should have
1922 * a usage count of 2 if we're the only user of this
1923 * dentry, and if that is true (possibly after pruning
1924 * the dcache), then we drop the dentry now.
1925 *
1926 * A low-level filesystem can, if it choses, legally
1927 * do a
1928 *
1929 * if (!d_unhashed(dentry))
1930 * return -EBUSY;
1931 *
1932 * if it cannot handle the case of removing a directory
1933 * that is still in use by something else..
1934 */
1935void dentry_unhash(struct dentry *dentry)
1936{
1937 dget(dentry);
1938 if (atomic_read(&dentry->d_count))
1939 shrink_dcache_parent(dentry);
1940 spin_lock(&dcache_lock);
1941 spin_lock(&dentry->d_lock);
1942 if (atomic_read(&dentry->d_count) == 2)
1943 __d_drop(dentry);
1944 spin_unlock(&dentry->d_lock);
1945 spin_unlock(&dcache_lock);
1946}
1947
1948int vfs_rmdir(struct inode *dir, struct dentry *dentry)
1949{
1950 int error = may_delete(dir, dentry, 1);
1951
1952 if (error)
1953 return error;
1954
1955 if (!dir->i_op || !dir->i_op->rmdir)
1956 return -EPERM;
1957
1958 DQUOT_INIT(dir);
1959
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001960 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 dentry_unhash(dentry);
1962 if (d_mountpoint(dentry))
1963 error = -EBUSY;
1964 else {
1965 error = security_inode_rmdir(dir, dentry);
1966 if (!error) {
1967 error = dir->i_op->rmdir(dir, dentry);
1968 if (!error)
1969 dentry->d_inode->i_flags |= S_DEAD;
1970 }
1971 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001972 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 d_delete(dentry);
1975 }
1976 dput(dentry);
1977
1978 return error;
1979}
1980
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001981static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982{
1983 int error = 0;
1984 char * name;
1985 struct dentry *dentry;
1986 struct nameidata nd;
1987
1988 name = getname(pathname);
1989 if(IS_ERR(name))
1990 return PTR_ERR(name);
1991
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001992 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993 if (error)
1994 goto exit;
1995
1996 switch(nd.last_type) {
1997 case LAST_DOTDOT:
1998 error = -ENOTEMPTY;
1999 goto exit1;
2000 case LAST_DOT:
2001 error = -EINVAL;
2002 goto exit1;
2003 case LAST_ROOT:
2004 error = -EBUSY;
2005 goto exit1;
2006 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002007 mutex_lock(&nd.dentry->d_inode->i_mutex);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002008 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 error = PTR_ERR(dentry);
2010 if (!IS_ERR(dentry)) {
2011 error = vfs_rmdir(nd.dentry->d_inode, dentry);
2012 dput(dentry);
2013 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002014 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015exit1:
2016 path_release(&nd);
2017exit:
2018 putname(name);
2019 return error;
2020}
2021
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002022asmlinkage long sys_rmdir(const char __user *pathname)
2023{
2024 return do_rmdir(AT_FDCWD, pathname);
2025}
2026
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027int vfs_unlink(struct inode *dir, struct dentry *dentry)
2028{
2029 int error = may_delete(dir, dentry, 0);
2030
2031 if (error)
2032 return error;
2033
2034 if (!dir->i_op || !dir->i_op->unlink)
2035 return -EPERM;
2036
2037 DQUOT_INIT(dir);
2038
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002039 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 if (d_mountpoint(dentry))
2041 error = -EBUSY;
2042 else {
2043 error = security_inode_unlink(dir, dentry);
2044 if (!error)
2045 error = dir->i_op->unlink(dir, dentry);
2046 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002047 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048
2049 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2050 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2051 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 }
Robert Love0eeca282005-07-12 17:06:03 -04002053
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054 return error;
2055}
2056
2057/*
2058 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002059 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060 * writeout happening, and we don't want to prevent access to the directory
2061 * while waiting on the I/O.
2062 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002063static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064{
2065 int error = 0;
2066 char * name;
2067 struct dentry *dentry;
2068 struct nameidata nd;
2069 struct inode *inode = NULL;
2070
2071 name = getname(pathname);
2072 if(IS_ERR(name))
2073 return PTR_ERR(name);
2074
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002075 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 if (error)
2077 goto exit;
2078 error = -EISDIR;
2079 if (nd.last_type != LAST_NORM)
2080 goto exit1;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002081 mutex_lock(&nd.dentry->d_inode->i_mutex);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002082 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083 error = PTR_ERR(dentry);
2084 if (!IS_ERR(dentry)) {
2085 /* Why not before? Because we want correct error value */
2086 if (nd.last.name[nd.last.len])
2087 goto slashes;
2088 inode = dentry->d_inode;
2089 if (inode)
2090 atomic_inc(&inode->i_count);
2091 error = vfs_unlink(nd.dentry->d_inode, dentry);
2092 exit2:
2093 dput(dentry);
2094 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002095 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096 if (inode)
2097 iput(inode); /* truncate the inode here */
2098exit1:
2099 path_release(&nd);
2100exit:
2101 putname(name);
2102 return error;
2103
2104slashes:
2105 error = !dentry->d_inode ? -ENOENT :
2106 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2107 goto exit2;
2108}
2109
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002110asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2111{
2112 if ((flag & ~AT_REMOVEDIR) != 0)
2113 return -EINVAL;
2114
2115 if (flag & AT_REMOVEDIR)
2116 return do_rmdir(dfd, pathname);
2117
2118 return do_unlinkat(dfd, pathname);
2119}
2120
2121asmlinkage long sys_unlink(const char __user *pathname)
2122{
2123 return do_unlinkat(AT_FDCWD, pathname);
2124}
2125
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname, int mode)
2127{
2128 int error = may_create(dir, dentry, NULL);
2129
2130 if (error)
2131 return error;
2132
2133 if (!dir->i_op || !dir->i_op->symlink)
2134 return -EPERM;
2135
2136 error = security_inode_symlink(dir, dentry, oldname);
2137 if (error)
2138 return error;
2139
2140 DQUOT_INIT(dir);
2141 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002142 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002143 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 return error;
2145}
2146
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002147asmlinkage long sys_symlinkat(const char __user *oldname,
2148 int newdfd, const char __user *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149{
2150 int error = 0;
2151 char * from;
2152 char * to;
2153
2154 from = getname(oldname);
2155 if(IS_ERR(from))
2156 return PTR_ERR(from);
2157 to = getname(newname);
2158 error = PTR_ERR(to);
2159 if (!IS_ERR(to)) {
2160 struct dentry *dentry;
2161 struct nameidata nd;
2162
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002163 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 if (error)
2165 goto out;
2166 dentry = lookup_create(&nd, 0);
2167 error = PTR_ERR(dentry);
2168 if (!IS_ERR(dentry)) {
2169 error = vfs_symlink(nd.dentry->d_inode, dentry, from, S_IALLUGO);
2170 dput(dentry);
2171 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002172 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173 path_release(&nd);
2174out:
2175 putname(to);
2176 }
2177 putname(from);
2178 return error;
2179}
2180
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002181asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2182{
2183 return sys_symlinkat(oldname, AT_FDCWD, newname);
2184}
2185
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2187{
2188 struct inode *inode = old_dentry->d_inode;
2189 int error;
2190
2191 if (!inode)
2192 return -ENOENT;
2193
2194 error = may_create(dir, new_dentry, NULL);
2195 if (error)
2196 return error;
2197
2198 if (dir->i_sb != inode->i_sb)
2199 return -EXDEV;
2200
2201 /*
2202 * A link to an append-only or immutable file cannot be created.
2203 */
2204 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2205 return -EPERM;
2206 if (!dir->i_op || !dir->i_op->link)
2207 return -EPERM;
2208 if (S_ISDIR(old_dentry->d_inode->i_mode))
2209 return -EPERM;
2210
2211 error = security_inode_link(old_dentry, dir, new_dentry);
2212 if (error)
2213 return error;
2214
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002215 mutex_lock(&old_dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216 DQUOT_INIT(dir);
2217 error = dir->i_op->link(old_dentry, dir, new_dentry);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002218 mutex_unlock(&old_dentry->d_inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002219 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002220 fsnotify_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221 return error;
2222}
2223
2224/*
2225 * Hardlinks are often used in delicate situations. We avoid
2226 * security-related surprises by not following symlinks on the
2227 * newname. --KAB
2228 *
2229 * We don't follow them on the oldname either to be compatible
2230 * with linux 2.0, and to avoid hard-linking to directories
2231 * and other special files. --ADM
2232 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002233asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002234 int newdfd, const char __user *newname,
2235 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236{
2237 struct dentry *new_dentry;
2238 struct nameidata nd, old_nd;
2239 int error;
2240 char * to;
2241
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002242 if (flags != 0)
2243 return -EINVAL;
2244
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245 to = getname(newname);
2246 if (IS_ERR(to))
2247 return PTR_ERR(to);
2248
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002249 error = __user_walk_fd(olddfd, oldname, 0, &old_nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 if (error)
2251 goto exit;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002252 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 if (error)
2254 goto out;
2255 error = -EXDEV;
2256 if (old_nd.mnt != nd.mnt)
2257 goto out_release;
2258 new_dentry = lookup_create(&nd, 0);
2259 error = PTR_ERR(new_dentry);
2260 if (!IS_ERR(new_dentry)) {
2261 error = vfs_link(old_nd.dentry, nd.dentry->d_inode, new_dentry);
2262 dput(new_dentry);
2263 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002264 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265out_release:
2266 path_release(&nd);
2267out:
2268 path_release(&old_nd);
2269exit:
2270 putname(to);
2271
2272 return error;
2273}
2274
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002275asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2276{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002277 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002278}
2279
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280/*
2281 * The worst of all namespace operations - renaming directory. "Perverted"
2282 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2283 * Problems:
2284 * a) we can get into loop creation. Check is done in is_subdir().
2285 * b) race potential - two innocent renames can create a loop together.
2286 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002287 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288 * story.
2289 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002290 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291 * whether the target exists). Solution: try to be smart with locking
2292 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002293 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 * move will be locked. Thus we can rank directories by the tree
2295 * (ancestors first) and rank all non-directories after them.
2296 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002297 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298 * HOWEVER, it relies on the assumption that any object with ->lookup()
2299 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2300 * we'd better make sure that there's no link(2) for them.
2301 * d) some filesystems don't support opened-but-unlinked directories,
2302 * either because of layout or because they are not ready to deal with
2303 * all cases correctly. The latter will be fixed (taking this sort of
2304 * stuff into VFS), but the former is not going away. Solution: the same
2305 * trick as in rmdir().
2306 * e) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002307 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002309 * ->i_mutex on parents, which works but leads to some truely excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310 * locking].
2311 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07002312static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2313 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314{
2315 int error = 0;
2316 struct inode *target;
2317
2318 /*
2319 * If we are going to change the parent - check write permissions,
2320 * we'll need to flip '..'.
2321 */
2322 if (new_dir != old_dir) {
2323 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2324 if (error)
2325 return error;
2326 }
2327
2328 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2329 if (error)
2330 return error;
2331
2332 target = new_dentry->d_inode;
2333 if (target) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002334 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335 dentry_unhash(new_dentry);
2336 }
2337 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2338 error = -EBUSY;
2339 else
2340 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2341 if (target) {
2342 if (!error)
2343 target->i_flags |= S_DEAD;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002344 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345 if (d_unhashed(new_dentry))
2346 d_rehash(new_dentry);
2347 dput(new_dentry);
2348 }
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002349 if (!error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002351 return error;
2352}
2353
Adrian Bunk75c96f82005-05-05 16:16:09 -07002354static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2355 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356{
2357 struct inode *target;
2358 int error;
2359
2360 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2361 if (error)
2362 return error;
2363
2364 dget(new_dentry);
2365 target = new_dentry->d_inode;
2366 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002367 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2369 error = -EBUSY;
2370 else
2371 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2372 if (!error) {
2373 /* The following d_move() should become unconditional */
2374 if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
2375 d_move(old_dentry, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376 }
2377 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002378 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379 dput(new_dentry);
2380 return error;
2381}
2382
2383int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2384 struct inode *new_dir, struct dentry *new_dentry)
2385{
2386 int error;
2387 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Robert Love0eeca282005-07-12 17:06:03 -04002388 const char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389
2390 if (old_dentry->d_inode == new_dentry->d_inode)
2391 return 0;
2392
2393 error = may_delete(old_dir, old_dentry, is_dir);
2394 if (error)
2395 return error;
2396
2397 if (!new_dentry->d_inode)
2398 error = may_create(new_dir, new_dentry, NULL);
2399 else
2400 error = may_delete(new_dir, new_dentry, is_dir);
2401 if (error)
2402 return error;
2403
2404 if (!old_dir->i_op || !old_dir->i_op->rename)
2405 return -EPERM;
2406
2407 DQUOT_INIT(old_dir);
2408 DQUOT_INIT(new_dir);
2409
Robert Love0eeca282005-07-12 17:06:03 -04002410 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2411
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412 if (is_dir)
2413 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2414 else
2415 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2416 if (!error) {
Robert Love0eeca282005-07-12 17:06:03 -04002417 const char *new_name = old_dentry->d_name.name;
John McCutchan89204c42005-08-15 12:13:28 -04002418 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2419 new_dentry->d_inode, old_dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420 }
Robert Love0eeca282005-07-12 17:06:03 -04002421 fsnotify_oldname_free(old_name);
2422
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 return error;
2424}
2425
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002426static int do_rename(int olddfd, const char *oldname,
2427 int newdfd, const char *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428{
2429 int error = 0;
2430 struct dentry * old_dir, * new_dir;
2431 struct dentry * old_dentry, *new_dentry;
2432 struct dentry * trap;
2433 struct nameidata oldnd, newnd;
2434
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002435 error = do_path_lookup(olddfd, oldname, LOOKUP_PARENT, &oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436 if (error)
2437 goto exit;
2438
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002439 error = do_path_lookup(newdfd, newname, LOOKUP_PARENT, &newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440 if (error)
2441 goto exit1;
2442
2443 error = -EXDEV;
2444 if (oldnd.mnt != newnd.mnt)
2445 goto exit2;
2446
2447 old_dir = oldnd.dentry;
2448 error = -EBUSY;
2449 if (oldnd.last_type != LAST_NORM)
2450 goto exit2;
2451
2452 new_dir = newnd.dentry;
2453 if (newnd.last_type != LAST_NORM)
2454 goto exit2;
2455
2456 trap = lock_rename(new_dir, old_dir);
2457
Christoph Hellwig49705b72005-11-08 21:35:06 -08002458 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459 error = PTR_ERR(old_dentry);
2460 if (IS_ERR(old_dentry))
2461 goto exit3;
2462 /* source must exist */
2463 error = -ENOENT;
2464 if (!old_dentry->d_inode)
2465 goto exit4;
2466 /* unless the source is a directory trailing slashes give -ENOTDIR */
2467 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2468 error = -ENOTDIR;
2469 if (oldnd.last.name[oldnd.last.len])
2470 goto exit4;
2471 if (newnd.last.name[newnd.last.len])
2472 goto exit4;
2473 }
2474 /* source should not be ancestor of target */
2475 error = -EINVAL;
2476 if (old_dentry == trap)
2477 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08002478 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479 error = PTR_ERR(new_dentry);
2480 if (IS_ERR(new_dentry))
2481 goto exit4;
2482 /* target should not be an ancestor of source */
2483 error = -ENOTEMPTY;
2484 if (new_dentry == trap)
2485 goto exit5;
2486
2487 error = vfs_rename(old_dir->d_inode, old_dentry,
2488 new_dir->d_inode, new_dentry);
2489exit5:
2490 dput(new_dentry);
2491exit4:
2492 dput(old_dentry);
2493exit3:
2494 unlock_rename(new_dir, old_dir);
2495exit2:
2496 path_release(&newnd);
2497exit1:
2498 path_release(&oldnd);
2499exit:
2500 return error;
2501}
2502
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002503asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2504 int newdfd, const char __user *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505{
2506 int error;
2507 char * from;
2508 char * to;
2509
2510 from = getname(oldname);
2511 if(IS_ERR(from))
2512 return PTR_ERR(from);
2513 to = getname(newname);
2514 error = PTR_ERR(to);
2515 if (!IS_ERR(to)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002516 error = do_rename(olddfd, from, newdfd, to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517 putname(to);
2518 }
2519 putname(from);
2520 return error;
2521}
2522
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002523asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2524{
2525 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2526}
2527
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2529{
2530 int len;
2531
2532 len = PTR_ERR(link);
2533 if (IS_ERR(link))
2534 goto out;
2535
2536 len = strlen(link);
2537 if (len > (unsigned) buflen)
2538 len = buflen;
2539 if (copy_to_user(buffer, link, len))
2540 len = -EFAULT;
2541out:
2542 return len;
2543}
2544
2545/*
2546 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2547 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2548 * using) it for any given inode is up to filesystem.
2549 */
2550int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2551{
2552 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002553 void *cookie;
2554
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002556 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2557 if (!IS_ERR(cookie)) {
2558 int res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 if (dentry->d_inode->i_op->put_link)
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002560 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2561 cookie = ERR_PTR(res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562 }
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002563 return PTR_ERR(cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564}
2565
2566int vfs_follow_link(struct nameidata *nd, const char *link)
2567{
2568 return __vfs_follow_link(nd, link);
2569}
2570
2571/* get the link contents into pagecache */
2572static char *page_getlink(struct dentry * dentry, struct page **ppage)
2573{
2574 struct page * page;
2575 struct address_space *mapping = dentry->d_inode->i_mapping;
2576 page = read_cache_page(mapping, 0, (filler_t *)mapping->a_ops->readpage,
2577 NULL);
2578 if (IS_ERR(page))
2579 goto sync_fail;
2580 wait_on_page_locked(page);
2581 if (!PageUptodate(page))
2582 goto async_fail;
2583 *ppage = page;
2584 return kmap(page);
2585
2586async_fail:
2587 page_cache_release(page);
2588 return ERR_PTR(-EIO);
2589
2590sync_fail:
2591 return (char*)page;
2592}
2593
2594int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2595{
2596 struct page *page = NULL;
2597 char *s = page_getlink(dentry, &page);
2598 int res = vfs_readlink(dentry,buffer,buflen,s);
2599 if (page) {
2600 kunmap(page);
2601 page_cache_release(page);
2602 }
2603 return res;
2604}
2605
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002606void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002608 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002610 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611}
2612
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002613void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002614{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002615 struct page *page = cookie;
2616
2617 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618 kunmap(page);
2619 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620 }
2621}
2622
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002623int __page_symlink(struct inode *inode, const char *symname, int len,
2624 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625{
2626 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002627 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628 int err = -ENOMEM;
2629 char *kaddr;
2630
NeilBrown7e53cac2006-03-25 03:07:57 -08002631retry:
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002632 page = find_or_create_page(mapping, 0, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002633 if (!page)
2634 goto fail;
2635 err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
NeilBrown7e53cac2006-03-25 03:07:57 -08002636 if (err == AOP_TRUNCATED_PAGE) {
2637 page_cache_release(page);
2638 goto retry;
2639 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002640 if (err)
2641 goto fail_map;
2642 kaddr = kmap_atomic(page, KM_USER0);
2643 memcpy(kaddr, symname, len-1);
2644 kunmap_atomic(kaddr, KM_USER0);
NeilBrown7e53cac2006-03-25 03:07:57 -08002645 err = mapping->a_ops->commit_write(NULL, page, 0, len-1);
2646 if (err == AOP_TRUNCATED_PAGE) {
2647 page_cache_release(page);
2648 goto retry;
2649 }
2650 if (err)
2651 goto fail_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652 /*
2653 * Notice that we are _not_ going to block here - end of page is
2654 * unmapped, so this will only try to map the rest of page, see
2655 * that it is unmapped (typically even will not look into inode -
2656 * ->i_size will be enough for everything) and zero it out.
2657 * OTOH it's obviously correct and should make the page up-to-date.
2658 */
2659 if (!PageUptodate(page)) {
2660 err = mapping->a_ops->readpage(NULL, page);
NeilBrown7e53cac2006-03-25 03:07:57 -08002661 if (err != AOP_TRUNCATED_PAGE)
2662 wait_on_page_locked(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663 } else {
2664 unlock_page(page);
2665 }
2666 page_cache_release(page);
2667 if (err < 0)
2668 goto fail;
2669 mark_inode_dirty(inode);
2670 return 0;
2671fail_map:
2672 unlock_page(page);
2673 page_cache_release(page);
2674fail:
2675 return err;
2676}
2677
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002678int page_symlink(struct inode *inode, const char *symname, int len)
2679{
2680 return __page_symlink(inode, symname, len,
2681 mapping_gfp_mask(inode->i_mapping));
2682}
2683
Linus Torvalds1da177e2005-04-16 15:20:36 -07002684struct inode_operations page_symlink_inode_operations = {
2685 .readlink = generic_readlink,
2686 .follow_link = page_follow_link_light,
2687 .put_link = page_put_link,
2688};
2689
2690EXPORT_SYMBOL(__user_walk);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002691EXPORT_SYMBOL(__user_walk_fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692EXPORT_SYMBOL(follow_down);
2693EXPORT_SYMBOL(follow_up);
2694EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2695EXPORT_SYMBOL(getname);
2696EXPORT_SYMBOL(lock_rename);
2697EXPORT_SYMBOL(lookup_hash);
2698EXPORT_SYMBOL(lookup_one_len);
2699EXPORT_SYMBOL(page_follow_link_light);
2700EXPORT_SYMBOL(page_put_link);
2701EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002702EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002703EXPORT_SYMBOL(page_symlink);
2704EXPORT_SYMBOL(page_symlink_inode_operations);
2705EXPORT_SYMBOL(path_lookup);
2706EXPORT_SYMBOL(path_release);
2707EXPORT_SYMBOL(path_walk);
2708EXPORT_SYMBOL(permission);
Christoph Hellwige4543ed2005-11-08 21:35:04 -08002709EXPORT_SYMBOL(vfs_permission);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -08002710EXPORT_SYMBOL(file_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711EXPORT_SYMBOL(unlock_rename);
2712EXPORT_SYMBOL(vfs_create);
2713EXPORT_SYMBOL(vfs_follow_link);
2714EXPORT_SYMBOL(vfs_link);
2715EXPORT_SYMBOL(vfs_mkdir);
2716EXPORT_SYMBOL(vfs_mknod);
2717EXPORT_SYMBOL(generic_permission);
2718EXPORT_SYMBOL(vfs_readlink);
2719EXPORT_SYMBOL(vfs_rename);
2720EXPORT_SYMBOL(vfs_rmdir);
2721EXPORT_SYMBOL(vfs_symlink);
2722EXPORT_SYMBOL(vfs_unlink);
2723EXPORT_SYMBOL(dentry_unhash);
2724EXPORT_SYMBOL(generic_readlink);