blob: faf61c35308cb023474f5859c606b8ef164d35c3 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
18#include <linux/module.h>
19#include <linux/slab.h>
20#include <linux/fs.h>
21#include <linux/namei.h>
22#include <linux/quotaops.h>
23#include <linux/pagemap.h>
Robert Love0eeca282005-07-12 17:06:03 -040024#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/smp_lock.h>
26#include <linux/personality.h>
27#include <linux/security.h>
28#include <linux/syscalls.h>
29#include <linux/mount.h>
30#include <linux/audit.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080031#include <linux/capability.h>
Trond Myklebust834f2a42005-10-18 14:20:16 -070032#include <linux/file.h>
Ulrich Drepper5590ff02006-01-18 17:43:53 -080033#include <linux/fcntl.h>
34#include <linux/namei.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <asm/namei.h>
36#include <asm/uaccess.h>
37
38#define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
39
40/* [Feb-1997 T. Schoebel-Theuer]
41 * Fundamental changes in the pathname lookup mechanisms (namei)
42 * were necessary because of omirr. The reason is that omirr needs
43 * to know the _real_ pathname, not the user-supplied one, in case
44 * of symlinks (and also when transname replacements occur).
45 *
46 * The new code replaces the old recursive symlink resolution with
47 * an iterative one (in case of non-nested symlink chains). It does
48 * this with calls to <fs>_follow_link().
49 * As a side effect, dir_namei(), _namei() and follow_link() are now
50 * replaced with a single function lookup_dentry() that can handle all
51 * the special cases of the former code.
52 *
53 * With the new dcache, the pathname is stored at each inode, at least as
54 * long as the refcount of the inode is positive. As a side effect, the
55 * size of the dcache depends on the inode cache and thus is dynamic.
56 *
57 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
58 * resolution to correspond with current state of the code.
59 *
60 * Note that the symlink resolution is not *completely* iterative.
61 * There is still a significant amount of tail- and mid- recursion in
62 * the algorithm. Also, note that <fs>_readlink() is not used in
63 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
64 * may return different results than <fs>_follow_link(). Many virtual
65 * filesystems (including /proc) exhibit this behavior.
66 */
67
68/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
69 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
70 * and the name already exists in form of a symlink, try to create the new
71 * name indicated by the symlink. The old code always complained that the
72 * name already exists, due to not following the symlink even if its target
73 * is nonexistent. The new semantics affects also mknod() and link() when
74 * the name is a symlink pointing to a non-existant name.
75 *
76 * I don't know which semantics is the right one, since I have no access
77 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
78 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
79 * "old" one. Personally, I think the new semantics is much more logical.
80 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
81 * file does succeed in both HP-UX and SunOs, but not in Solaris
82 * and in the old Linux semantics.
83 */
84
85/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
86 * semantics. See the comments in "open_namei" and "do_link" below.
87 *
88 * [10-Sep-98 Alan Modra] Another symlink change.
89 */
90
91/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
92 * inside the path - always follow.
93 * in the last component in creation/removal/renaming - never follow.
94 * if LOOKUP_FOLLOW passed - follow.
95 * if the pathname has trailing slashes - follow.
96 * otherwise - don't follow.
97 * (applied in that order).
98 *
99 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
100 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
101 * During the 2.4 we need to fix the userland stuff depending on it -
102 * hopefully we will be able to get rid of that wart in 2.5. So far only
103 * XEmacs seems to be relying on it...
104 */
105/*
106 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
107 * implemented. Let's see if raised priority of ->s_vfs_rename_sem gives
108 * any extra contention...
109 */
110
111/* In order to reduce some races, while at the same time doing additional
112 * checking and hopefully speeding things up, we copy filenames to the
113 * kernel data space before using them..
114 *
115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
116 * PATH_MAX includes the nul terminator --RR.
117 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800118static int do_getname(const char __user *filename, char *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119{
120 int retval;
121 unsigned long len = PATH_MAX;
122
123 if (!segment_eq(get_fs(), KERNEL_DS)) {
124 if ((unsigned long) filename >= TASK_SIZE)
125 return -EFAULT;
126 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
127 len = TASK_SIZE - (unsigned long) filename;
128 }
129
130 retval = strncpy_from_user(page, filename, len);
131 if (retval > 0) {
132 if (retval < len)
133 return 0;
134 return -ENAMETOOLONG;
135 } else if (!retval)
136 retval = -ENOENT;
137 return retval;
138}
139
140char * getname(const char __user * filename)
141{
142 char *tmp, *result;
143
144 result = ERR_PTR(-ENOMEM);
145 tmp = __getname();
146 if (tmp) {
147 int retval = do_getname(filename, tmp);
148
149 result = tmp;
150 if (retval < 0) {
151 __putname(tmp);
152 result = ERR_PTR(retval);
153 }
154 }
155 audit_getname(result);
156 return result;
157}
158
159#ifdef CONFIG_AUDITSYSCALL
160void putname(const char *name)
161{
162 if (unlikely(current->audit_context))
163 audit_putname(name);
164 else
165 __putname(name);
166}
167EXPORT_SYMBOL(putname);
168#endif
169
170
171/**
172 * generic_permission - check for access rights on a Posix-like filesystem
173 * @inode: inode to check access rights for
174 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
175 * @check_acl: optional callback to check for Posix ACLs
176 *
177 * Used to check for read/write/execute permissions on a file.
178 * We use "fsuid" for this, letting us set arbitrary permissions
179 * for filesystem access without changing the "normal" uids which
180 * are used for other things..
181 */
182int generic_permission(struct inode *inode, int mask,
183 int (*check_acl)(struct inode *inode, int mask))
184{
185 umode_t mode = inode->i_mode;
186
187 if (current->fsuid == inode->i_uid)
188 mode >>= 6;
189 else {
190 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
191 int error = check_acl(inode, mask);
192 if (error == -EACCES)
193 goto check_capabilities;
194 else if (error != -EAGAIN)
195 return error;
196 }
197
198 if (in_group_p(inode->i_gid))
199 mode >>= 3;
200 }
201
202 /*
203 * If the DACs are ok we don't need any capability check.
204 */
205 if (((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask))
206 return 0;
207
208 check_capabilities:
209 /*
210 * Read/write DACs are always overridable.
211 * Executable DACs are overridable if at least one exec bit is set.
212 */
213 if (!(mask & MAY_EXEC) ||
214 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
215 if (capable(CAP_DAC_OVERRIDE))
216 return 0;
217
218 /*
219 * Searching includes executable on directories, else just read.
220 */
221 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
222 if (capable(CAP_DAC_READ_SEARCH))
223 return 0;
224
225 return -EACCES;
226}
227
228int permission(struct inode *inode, int mask, struct nameidata *nd)
229{
230 int retval, submask;
231
232 if (mask & MAY_WRITE) {
233 umode_t mode = inode->i_mode;
234
235 /*
236 * Nobody gets write access to a read-only fs.
237 */
238 if (IS_RDONLY(inode) &&
239 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
240 return -EROFS;
241
242 /*
243 * Nobody gets write access to an immutable file.
244 */
245 if (IS_IMMUTABLE(inode))
246 return -EACCES;
247 }
248
249
250 /* Ordinary permission routines do not understand MAY_APPEND. */
251 submask = mask & ~MAY_APPEND;
252 if (inode->i_op && inode->i_op->permission)
253 retval = inode->i_op->permission(inode, submask, nd);
254 else
255 retval = generic_permission(inode, submask, NULL);
256 if (retval)
257 return retval;
258
259 return security_inode_permission(inode, mask, nd);
260}
261
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800262/**
263 * vfs_permission - check for access rights to a given path
264 * @nd: lookup result that describes the path
265 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
266 *
267 * Used to check for read/write/execute permissions on a path.
268 * We use "fsuid" for this, letting us set arbitrary permissions
269 * for filesystem access without changing the "normal" uids which
270 * are used for other things.
271 */
272int vfs_permission(struct nameidata *nd, int mask)
273{
274 return permission(nd->dentry->d_inode, mask, nd);
275}
276
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800277/**
278 * file_permission - check for additional access rights to a given file
279 * @file: file to check access rights for
280 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
281 *
282 * Used to check for read/write/execute permissions on an already opened
283 * file.
284 *
285 * Note:
286 * Do not use this function in new code. All access checks should
287 * be done using vfs_permission().
288 */
289int file_permission(struct file *file, int mask)
290{
291 return permission(file->f_dentry->d_inode, mask, NULL);
292}
293
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294/*
295 * get_write_access() gets write permission for a file.
296 * put_write_access() releases this write permission.
297 * This is used for regular files.
298 * We cannot support write (and maybe mmap read-write shared) accesses and
299 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
300 * can have the following values:
301 * 0: no writers, no VM_DENYWRITE mappings
302 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
303 * > 0: (i_writecount) users are writing to the file.
304 *
305 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
306 * except for the cases where we don't hold i_writecount yet. Then we need to
307 * use {get,deny}_write_access() - these functions check the sign and refuse
308 * to do the change if sign is wrong. Exclusion between them is provided by
309 * the inode->i_lock spinlock.
310 */
311
312int get_write_access(struct inode * inode)
313{
314 spin_lock(&inode->i_lock);
315 if (atomic_read(&inode->i_writecount) < 0) {
316 spin_unlock(&inode->i_lock);
317 return -ETXTBSY;
318 }
319 atomic_inc(&inode->i_writecount);
320 spin_unlock(&inode->i_lock);
321
322 return 0;
323}
324
325int deny_write_access(struct file * file)
326{
327 struct inode *inode = file->f_dentry->d_inode;
328
329 spin_lock(&inode->i_lock);
330 if (atomic_read(&inode->i_writecount) > 0) {
331 spin_unlock(&inode->i_lock);
332 return -ETXTBSY;
333 }
334 atomic_dec(&inode->i_writecount);
335 spin_unlock(&inode->i_lock);
336
337 return 0;
338}
339
340void path_release(struct nameidata *nd)
341{
342 dput(nd->dentry);
343 mntput(nd->mnt);
344}
345
346/*
347 * umount() mustn't call path_release()/mntput() as that would clear
348 * mnt_expiry_mark
349 */
350void path_release_on_umount(struct nameidata *nd)
351{
352 dput(nd->dentry);
Miklos Szeredi751c4042005-07-07 17:57:30 -0700353 mntput_no_expire(nd->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354}
355
Trond Myklebust834f2a42005-10-18 14:20:16 -0700356/**
357 * release_open_intent - free up open intent resources
358 * @nd: pointer to nameidata
359 */
360void release_open_intent(struct nameidata *nd)
361{
362 if (nd->intent.open.file->f_dentry == NULL)
363 put_filp(nd->intent.open.file);
364 else
365 fput(nd->intent.open.file);
366}
367
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368/*
369 * Internal lookup() using the new generic dcache.
370 * SMP-safe
371 */
372static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
373{
374 struct dentry * dentry = __d_lookup(parent, name);
375
376 /* lockess __d_lookup may fail due to concurrent d_move()
377 * in some unrelated directory, so try with d_lookup
378 */
379 if (!dentry)
380 dentry = d_lookup(parent, name);
381
382 if (dentry && dentry->d_op && dentry->d_op->d_revalidate) {
383 if (!dentry->d_op->d_revalidate(dentry, nd) && !d_invalidate(dentry)) {
384 dput(dentry);
385 dentry = NULL;
386 }
387 }
388 return dentry;
389}
390
391/*
392 * Short-cut version of permission(), for calling by
393 * path_walk(), when dcache lock is held. Combines parts
394 * of permission() and generic_permission(), and tests ONLY for
395 * MAY_EXEC permission.
396 *
397 * If appropriate, check DAC only. If not appropriate, or
398 * short-cut DAC fails, then call permission() to do more
399 * complete permission check.
400 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800401static int exec_permission_lite(struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 struct nameidata *nd)
403{
404 umode_t mode = inode->i_mode;
405
406 if (inode->i_op && inode->i_op->permission)
407 return -EAGAIN;
408
409 if (current->fsuid == inode->i_uid)
410 mode >>= 6;
411 else if (in_group_p(inode->i_gid))
412 mode >>= 3;
413
414 if (mode & MAY_EXEC)
415 goto ok;
416
417 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
418 goto ok;
419
420 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
421 goto ok;
422
423 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
424 goto ok;
425
426 return -EACCES;
427ok:
428 return security_inode_permission(inode, MAY_EXEC, nd);
429}
430
431/*
432 * This is called when everything else fails, and we actually have
433 * to go to the low-level filesystem to find out what we should do..
434 *
435 * We get the directory semaphore, and after getting that we also
436 * make sure that nobody added the entry to the dcache in the meantime..
437 * SMP-safe
438 */
439static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
440{
441 struct dentry * result;
442 struct inode *dir = parent->d_inode;
443
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800444 mutex_lock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 /*
446 * First re-do the cached lookup just in case it was created
447 * while we waited for the directory semaphore..
448 *
449 * FIXME! This could use version numbering or similar to
450 * avoid unnecessary cache lookups.
451 *
452 * The "dcache_lock" is purely to protect the RCU list walker
453 * from concurrent renames at this point (we mustn't get false
454 * negatives from the RCU list walk here, unlike the optimistic
455 * fast walk).
456 *
457 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
458 */
459 result = d_lookup(parent, name);
460 if (!result) {
461 struct dentry * dentry = d_alloc(parent, name);
462 result = ERR_PTR(-ENOMEM);
463 if (dentry) {
464 result = dir->i_op->lookup(dir, dentry, nd);
465 if (result)
466 dput(dentry);
467 else
468 result = dentry;
469 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800470 mutex_unlock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 return result;
472 }
473
474 /*
475 * Uhhuh! Nasty case: the cache was re-populated while
476 * we waited on the semaphore. Need to revalidate.
477 */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800478 mutex_unlock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 if (result->d_op && result->d_op->d_revalidate) {
480 if (!result->d_op->d_revalidate(result, nd) && !d_invalidate(result)) {
481 dput(result);
482 result = ERR_PTR(-ENOENT);
483 }
484 }
485 return result;
486}
487
488static int __emul_lookup_dentry(const char *, struct nameidata *);
489
490/* SMP-safe */
Arjan van de Venf1662352006-01-14 13:21:31 -0800491static __always_inline int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492walk_init_root(const char *name, struct nameidata *nd)
493{
494 read_lock(&current->fs->lock);
495 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
496 nd->mnt = mntget(current->fs->altrootmnt);
497 nd->dentry = dget(current->fs->altroot);
498 read_unlock(&current->fs->lock);
499 if (__emul_lookup_dentry(name,nd))
500 return 0;
501 read_lock(&current->fs->lock);
502 }
503 nd->mnt = mntget(current->fs->rootmnt);
504 nd->dentry = dget(current->fs->root);
505 read_unlock(&current->fs->lock);
506 return 1;
507}
508
Arjan van de Venf1662352006-01-14 13:21:31 -0800509static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510{
511 int res = 0;
512 char *name;
513 if (IS_ERR(link))
514 goto fail;
515
516 if (*link == '/') {
517 path_release(nd);
518 if (!walk_init_root(link, nd))
519 /* weird __emul_prefix() stuff did it */
520 goto out;
521 }
522 res = link_path_walk(link, nd);
523out:
524 if (nd->depth || res || nd->last_type!=LAST_NORM)
525 return res;
526 /*
527 * If it is an iterative symlinks resolution in open_namei() we
528 * have to copy the last component. And all that crap because of
529 * bloody create() on broken symlinks. Furrfu...
530 */
531 name = __getname();
532 if (unlikely(!name)) {
533 path_release(nd);
534 return -ENOMEM;
535 }
536 strcpy(name, nd->last.name);
537 nd->last.name = name;
538 return 0;
539fail:
540 path_release(nd);
541 return PTR_ERR(link);
542}
543
Al Viro90ebe562005-06-06 13:35:58 -0700544struct path {
545 struct vfsmount *mnt;
546 struct dentry *dentry;
547};
548
Arjan van de Venf1662352006-01-14 13:21:31 -0800549static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550{
551 int error;
Linus Torvaldscc314ee2005-08-19 18:02:56 -0700552 void *cookie;
Al Virocd4e91d2005-06-06 13:36:03 -0700553 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554
Al Virod671a1c2005-06-06 13:36:14 -0700555 touch_atime(path->mnt, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 nd_set_link(nd, NULL);
Al Virocd4e91d2005-06-06 13:36:03 -0700557
Al Viro39ca6d42005-06-06 13:36:12 -0700558 if (path->mnt == nd->mnt)
559 mntget(path->mnt);
Linus Torvaldscc314ee2005-08-19 18:02:56 -0700560 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
561 error = PTR_ERR(cookie);
562 if (!IS_ERR(cookie)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 char *s = nd_get_link(nd);
Linus Torvaldscc314ee2005-08-19 18:02:56 -0700564 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 if (s)
566 error = __vfs_follow_link(nd, s);
567 if (dentry->d_inode->i_op->put_link)
Linus Torvaldscc314ee2005-08-19 18:02:56 -0700568 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 }
Al Virocd4e91d2005-06-06 13:36:03 -0700570 dput(dentry);
571 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572
573 return error;
574}
575
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700576static inline void dput_path(struct path *path, struct nameidata *nd)
577{
578 dput(path->dentry);
579 if (path->mnt != nd->mnt)
580 mntput(path->mnt);
581}
582
583static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
584{
585 dput(nd->dentry);
586 if (nd->mnt != path->mnt)
587 mntput(nd->mnt);
588 nd->mnt = path->mnt;
589 nd->dentry = path->dentry;
590}
591
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592/*
593 * This limits recursive symlink follows to 8, while
594 * limiting consecutive symlinks to 40.
595 *
596 * Without that kind of total limit, nasty chains of consecutive
597 * symlinks can cause almost arbitrarily long lookups.
598 */
Al Viro90ebe562005-06-06 13:35:58 -0700599static inline int do_follow_link(struct path *path, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600{
601 int err = -ELOOP;
602 if (current->link_count >= MAX_NESTED_LINKS)
603 goto loop;
604 if (current->total_link_count >= 40)
605 goto loop;
606 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
607 cond_resched();
Al Viro90ebe562005-06-06 13:35:58 -0700608 err = security_inode_follow_link(path->dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 if (err)
610 goto loop;
611 current->link_count++;
612 current->total_link_count++;
613 nd->depth++;
Al Virocd4e91d2005-06-06 13:36:03 -0700614 err = __do_follow_link(path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 current->link_count--;
616 nd->depth--;
617 return err;
618loop:
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700619 dput_path(path, nd);
Al Viro839d9f92005-06-06 13:36:02 -0700620 path_release(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 return err;
622}
623
624int follow_up(struct vfsmount **mnt, struct dentry **dentry)
625{
626 struct vfsmount *parent;
627 struct dentry *mountpoint;
628 spin_lock(&vfsmount_lock);
629 parent=(*mnt)->mnt_parent;
630 if (parent == *mnt) {
631 spin_unlock(&vfsmount_lock);
632 return 0;
633 }
634 mntget(parent);
635 mountpoint=dget((*mnt)->mnt_mountpoint);
636 spin_unlock(&vfsmount_lock);
637 dput(*dentry);
638 *dentry = mountpoint;
639 mntput(*mnt);
640 *mnt = parent;
641 return 1;
642}
643
644/* no need for dcache_lock, as serialization is taken care in
645 * namespace.c
646 */
Al Viro463ffb22005-06-06 13:36:05 -0700647static int __follow_mount(struct path *path)
648{
649 int res = 0;
650 while (d_mountpoint(path->dentry)) {
651 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
652 if (!mounted)
653 break;
654 dput(path->dentry);
655 if (res)
656 mntput(path->mnt);
657 path->mnt = mounted;
658 path->dentry = dget(mounted->mnt_root);
659 res = 1;
660 }
661 return res;
662}
663
Al Viro58c465e2005-06-06 13:36:13 -0700664static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 while (d_mountpoint(*dentry)) {
667 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
668 if (!mounted)
669 break;
Al Viro58c465e2005-06-06 13:36:13 -0700670 dput(*dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 mntput(*mnt);
672 *mnt = mounted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 *dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675}
676
677/* no need for dcache_lock, as serialization is taken care in
678 * namespace.c
679 */
Al Viroe13b2102005-06-06 13:36:06 -0700680int follow_down(struct vfsmount **mnt, struct dentry **dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681{
682 struct vfsmount *mounted;
683
684 mounted = lookup_mnt(*mnt, *dentry);
685 if (mounted) {
Al Viroe13b2102005-06-06 13:36:06 -0700686 dput(*dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 mntput(*mnt);
688 *mnt = mounted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 *dentry = dget(mounted->mnt_root);
690 return 1;
691 }
692 return 0;
693}
694
Arjan van de Venf1662352006-01-14 13:21:31 -0800695static __always_inline void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696{
697 while(1) {
698 struct vfsmount *parent;
Al Viro58c465e2005-06-06 13:36:13 -0700699 struct dentry *old = nd->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700
701 read_lock(&current->fs->lock);
Al Viro58c465e2005-06-06 13:36:13 -0700702 if (nd->dentry == current->fs->root &&
703 nd->mnt == current->fs->rootmnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 read_unlock(&current->fs->lock);
705 break;
706 }
707 read_unlock(&current->fs->lock);
708 spin_lock(&dcache_lock);
Al Viro58c465e2005-06-06 13:36:13 -0700709 if (nd->dentry != nd->mnt->mnt_root) {
710 nd->dentry = dget(nd->dentry->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 spin_unlock(&dcache_lock);
712 dput(old);
713 break;
714 }
715 spin_unlock(&dcache_lock);
716 spin_lock(&vfsmount_lock);
Al Viro58c465e2005-06-06 13:36:13 -0700717 parent = nd->mnt->mnt_parent;
718 if (parent == nd->mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 spin_unlock(&vfsmount_lock);
720 break;
721 }
722 mntget(parent);
Al Viro58c465e2005-06-06 13:36:13 -0700723 nd->dentry = dget(nd->mnt->mnt_mountpoint);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 spin_unlock(&vfsmount_lock);
725 dput(old);
Al Viro58c465e2005-06-06 13:36:13 -0700726 mntput(nd->mnt);
727 nd->mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 }
Al Viro58c465e2005-06-06 13:36:13 -0700729 follow_mount(&nd->mnt, &nd->dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730}
731
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732/*
733 * It's more convoluted than I'd like it to be, but... it's still fairly
734 * small and for now I'd prefer to have fast path as straight as possible.
735 * It _is_ time-critical.
736 */
737static int do_lookup(struct nameidata *nd, struct qstr *name,
738 struct path *path)
739{
740 struct vfsmount *mnt = nd->mnt;
741 struct dentry *dentry = __d_lookup(nd->dentry, name);
742
743 if (!dentry)
744 goto need_lookup;
745 if (dentry->d_op && dentry->d_op->d_revalidate)
746 goto need_revalidate;
747done:
748 path->mnt = mnt;
749 path->dentry = dentry;
Al Viro634ee702005-06-06 13:36:13 -0700750 __follow_mount(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 return 0;
752
753need_lookup:
754 dentry = real_lookup(nd->dentry, name, nd);
755 if (IS_ERR(dentry))
756 goto fail;
757 goto done;
758
759need_revalidate:
760 if (dentry->d_op->d_revalidate(dentry, nd))
761 goto done;
762 if (d_invalidate(dentry))
763 goto done;
764 dput(dentry);
765 goto need_lookup;
766
767fail:
768 return PTR_ERR(dentry);
769}
770
771/*
772 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +0100773 * This is the basic name resolution function, turning a pathname into
774 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 *
Prasanna Medaea3834d2005-04-29 16:00:17 +0100776 * Returns 0 and nd will have valid dentry and mnt on success.
777 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 */
779static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
780{
781 struct path next;
782 struct inode *inode;
783 int err;
784 unsigned int lookup_flags = nd->flags;
785
786 while (*name=='/')
787 name++;
788 if (!*name)
789 goto return_reval;
790
791 inode = nd->dentry->d_inode;
792 if (nd->depth)
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:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 if (unlikely(current->audit_context
1123 && nd && nd->dentry && nd->dentry->d_inode))
David Woodhouseae7b9612005-06-20 16:11:05 +01001124 audit_inode(name, nd->dentry->d_inode, flags);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001125 return retval;
1126
1127fput_unlock_fail:
1128 fput_light(file, fput_needed);
1129unlock_fail:
1130 read_unlock(&current->fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 return retval;
1132}
1133
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001134int fastcall path_lookup(const char *name, unsigned int flags,
1135 struct nameidata *nd)
1136{
1137 return do_path_lookup(AT_FDCWD, name, flags, nd);
1138}
1139
1140static int __path_lookup_intent_open(int dfd, const char *name,
1141 unsigned int lookup_flags, struct nameidata *nd,
1142 int open_flags, int create_mode)
Trond Myklebust834f2a42005-10-18 14:20:16 -07001143{
1144 struct file *filp = get_empty_filp();
1145 int err;
1146
1147 if (filp == NULL)
1148 return -ENFILE;
1149 nd->intent.open.file = filp;
1150 nd->intent.open.flags = open_flags;
1151 nd->intent.open.create_mode = create_mode;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001152 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
Trond Myklebust834f2a42005-10-18 14:20:16 -07001153 if (IS_ERR(nd->intent.open.file)) {
1154 if (err == 0) {
1155 err = PTR_ERR(nd->intent.open.file);
1156 path_release(nd);
1157 }
1158 } else if (err != 0)
1159 release_open_intent(nd);
1160 return err;
1161}
1162
1163/**
1164 * path_lookup_open - lookup a file path with open intent
Martin Waitz7045f372006-02-01 03:06:57 -08001165 * @dfd: the directory to use as base, or AT_FDCWD
Trond Myklebust834f2a42005-10-18 14:20:16 -07001166 * @name: pointer to file name
1167 * @lookup_flags: lookup intent flags
1168 * @nd: pointer to nameidata
1169 * @open_flags: open intent flags
1170 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001171int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
Trond Myklebust834f2a42005-10-18 14:20:16 -07001172 struct nameidata *nd, int open_flags)
1173{
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001174 return __path_lookup_intent_open(dfd, name, lookup_flags, nd,
Trond Myklebust834f2a42005-10-18 14:20:16 -07001175 open_flags, 0);
1176}
1177
1178/**
1179 * path_lookup_create - lookup a file path with open + create intent
Martin Waitz7045f372006-02-01 03:06:57 -08001180 * @dfd: the directory to use as base, or AT_FDCWD
Trond Myklebust834f2a42005-10-18 14:20:16 -07001181 * @name: pointer to file name
1182 * @lookup_flags: lookup intent flags
1183 * @nd: pointer to nameidata
1184 * @open_flags: open intent flags
1185 * @create_mode: create intent flags
1186 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001187static int path_lookup_create(int dfd, const char *name,
1188 unsigned int lookup_flags, struct nameidata *nd,
1189 int open_flags, int create_mode)
Trond Myklebust834f2a42005-10-18 14:20:16 -07001190{
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001191 return __path_lookup_intent_open(dfd, name, lookup_flags|LOOKUP_CREATE,
1192 nd, open_flags, create_mode);
Trond Myklebust834f2a42005-10-18 14:20:16 -07001193}
1194
1195int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags,
1196 struct nameidata *nd, int open_flags)
1197{
1198 char *tmp = getname(name);
1199 int err = PTR_ERR(tmp);
1200
1201 if (!IS_ERR(tmp)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001202 err = __path_lookup_intent_open(AT_FDCWD, tmp, lookup_flags, nd, open_flags, 0);
Trond Myklebust834f2a42005-10-18 14:20:16 -07001203 putname(tmp);
1204 }
1205 return err;
1206}
1207
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208/*
1209 * Restricted form of lookup. Doesn't follow links, single-component only,
1210 * needs parent already locked. Doesn't follow mounts.
1211 * SMP-safe.
1212 */
1213static struct dentry * __lookup_hash(struct qstr *name, struct dentry * base, struct nameidata *nd)
1214{
1215 struct dentry * dentry;
1216 struct inode *inode;
1217 int err;
1218
1219 inode = base->d_inode;
1220 err = permission(inode, MAY_EXEC, nd);
1221 dentry = ERR_PTR(err);
1222 if (err)
1223 goto out;
1224
1225 /*
1226 * See if the low-level filesystem might want
1227 * to use its own hash..
1228 */
1229 if (base->d_op && base->d_op->d_hash) {
1230 err = base->d_op->d_hash(base, name);
1231 dentry = ERR_PTR(err);
1232 if (err < 0)
1233 goto out;
1234 }
1235
1236 dentry = cached_lookup(base, name, nd);
1237 if (!dentry) {
1238 struct dentry *new = d_alloc(base, name);
1239 dentry = ERR_PTR(-ENOMEM);
1240 if (!new)
1241 goto out;
1242 dentry = inode->i_op->lookup(inode, new, nd);
1243 if (!dentry)
1244 dentry = new;
1245 else
1246 dput(new);
1247 }
1248out:
1249 return dentry;
1250}
1251
Christoph Hellwig49705b72005-11-08 21:35:06 -08001252struct dentry * lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253{
Christoph Hellwig49705b72005-11-08 21:35:06 -08001254 return __lookup_hash(&nd->last, nd->dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255}
1256
1257/* SMP-safe */
1258struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
1259{
1260 unsigned long hash;
1261 struct qstr this;
1262 unsigned int c;
1263
1264 this.name = name;
1265 this.len = len;
1266 if (!len)
1267 goto access;
1268
1269 hash = init_name_hash();
1270 while (len--) {
1271 c = *(const unsigned char *)name++;
1272 if (c == '/' || c == '\0')
1273 goto access;
1274 hash = partial_name_hash(c, hash);
1275 }
1276 this.hash = end_name_hash(hash);
1277
Christoph Hellwig49705b72005-11-08 21:35:06 -08001278 return __lookup_hash(&this, base, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279access:
1280 return ERR_PTR(-EACCES);
1281}
1282
1283/*
1284 * namei()
1285 *
1286 * is used by most simple commands to get the inode of a specified name.
1287 * Open, link etc use their own routines, but this is enough for things
1288 * like 'chmod' etc.
1289 *
1290 * namei exists in two versions: namei/lnamei. The only difference is
1291 * that namei follows links, while lnamei does not.
1292 * SMP-safe
1293 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001294int fastcall __user_walk_fd(int dfd, const char __user *name, unsigned flags,
1295 struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296{
1297 char *tmp = getname(name);
1298 int err = PTR_ERR(tmp);
1299
1300 if (!IS_ERR(tmp)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001301 err = do_path_lookup(dfd, tmp, flags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 putname(tmp);
1303 }
1304 return err;
1305}
1306
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001307int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1308{
1309 return __user_walk_fd(AT_FDCWD, name, flags, nd);
1310}
1311
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312/*
1313 * It's inline, so penalty for filesystems that don't use sticky bit is
1314 * minimal.
1315 */
1316static inline int check_sticky(struct inode *dir, struct inode *inode)
1317{
1318 if (!(dir->i_mode & S_ISVTX))
1319 return 0;
1320 if (inode->i_uid == current->fsuid)
1321 return 0;
1322 if (dir->i_uid == current->fsuid)
1323 return 0;
1324 return !capable(CAP_FOWNER);
1325}
1326
1327/*
1328 * Check whether we can remove a link victim from directory dir, check
1329 * whether the type of victim is right.
1330 * 1. We can't do it if dir is read-only (done in permission())
1331 * 2. We should have write and exec permissions on dir
1332 * 3. We can't remove anything from append-only dir
1333 * 4. We can't do anything with immutable dir (done in permission())
1334 * 5. If the sticky bit on dir is set we should either
1335 * a. be owner of dir, or
1336 * b. be owner of victim, or
1337 * c. have CAP_FOWNER capability
1338 * 6. If the victim is append-only or immutable we can't do antyhing with
1339 * links pointing to it.
1340 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1341 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1342 * 9. We can't remove a root or mountpoint.
1343 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1344 * nfs_async_unlink().
1345 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001346static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347{
1348 int error;
1349
1350 if (!victim->d_inode)
1351 return -ENOENT;
1352
1353 BUG_ON(victim->d_parent->d_inode != dir);
1354
1355 error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
1356 if (error)
1357 return error;
1358 if (IS_APPEND(dir))
1359 return -EPERM;
1360 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1361 IS_IMMUTABLE(victim->d_inode))
1362 return -EPERM;
1363 if (isdir) {
1364 if (!S_ISDIR(victim->d_inode->i_mode))
1365 return -ENOTDIR;
1366 if (IS_ROOT(victim))
1367 return -EBUSY;
1368 } else if (S_ISDIR(victim->d_inode->i_mode))
1369 return -EISDIR;
1370 if (IS_DEADDIR(dir))
1371 return -ENOENT;
1372 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1373 return -EBUSY;
1374 return 0;
1375}
1376
1377/* Check whether we can create an object with dentry child in directory
1378 * dir.
1379 * 1. We can't do it if child already exists (open has special treatment for
1380 * this case, but since we are inlined it's OK)
1381 * 2. We can't do it if dir is read-only (done in permission())
1382 * 3. We should have write and exec permissions on dir
1383 * 4. We can't do it if dir is immutable (done in permission())
1384 */
1385static inline int may_create(struct inode *dir, struct dentry *child,
1386 struct nameidata *nd)
1387{
1388 if (child->d_inode)
1389 return -EEXIST;
1390 if (IS_DEADDIR(dir))
1391 return -ENOENT;
1392 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1393}
1394
1395/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 * O_DIRECTORY translates into forcing a directory lookup.
1397 */
1398static inline int lookup_flags(unsigned int f)
1399{
1400 unsigned long retval = LOOKUP_FOLLOW;
1401
1402 if (f & O_NOFOLLOW)
1403 retval &= ~LOOKUP_FOLLOW;
1404
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 if (f & O_DIRECTORY)
1406 retval |= LOOKUP_DIRECTORY;
1407
1408 return retval;
1409}
1410
1411/*
1412 * p1 and p2 should be directories on the same fs.
1413 */
1414struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1415{
1416 struct dentry *p;
1417
1418 if (p1 == p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001419 mutex_lock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 return NULL;
1421 }
1422
1423 down(&p1->d_inode->i_sb->s_vfs_rename_sem);
1424
1425 for (p = p1; p->d_parent != p; p = p->d_parent) {
1426 if (p->d_parent == p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001427 mutex_lock(&p2->d_inode->i_mutex);
1428 mutex_lock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 return p;
1430 }
1431 }
1432
1433 for (p = p2; p->d_parent != p; p = p->d_parent) {
1434 if (p->d_parent == p1) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001435 mutex_lock(&p1->d_inode->i_mutex);
1436 mutex_lock(&p2->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 return p;
1438 }
1439 }
1440
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001441 mutex_lock(&p1->d_inode->i_mutex);
1442 mutex_lock(&p2->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 return NULL;
1444}
1445
1446void unlock_rename(struct dentry *p1, struct dentry *p2)
1447{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001448 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001450 mutex_unlock(&p2->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 up(&p1->d_inode->i_sb->s_vfs_rename_sem);
1452 }
1453}
1454
1455int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1456 struct nameidata *nd)
1457{
1458 int error = may_create(dir, dentry, nd);
1459
1460 if (error)
1461 return error;
1462
1463 if (!dir->i_op || !dir->i_op->create)
1464 return -EACCES; /* shouldn't it be ENOSYS? */
1465 mode &= S_IALLUGO;
1466 mode |= S_IFREG;
1467 error = security_inode_create(dir, dentry, mode);
1468 if (error)
1469 return error;
1470 DQUOT_INIT(dir);
1471 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001472 if (!error)
Robert Love0eeca282005-07-12 17:06:03 -04001473 fsnotify_create(dir, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 return error;
1475}
1476
1477int may_open(struct nameidata *nd, int acc_mode, int flag)
1478{
1479 struct dentry *dentry = nd->dentry;
1480 struct inode *inode = dentry->d_inode;
1481 int error;
1482
1483 if (!inode)
1484 return -ENOENT;
1485
1486 if (S_ISLNK(inode->i_mode))
1487 return -ELOOP;
1488
1489 if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1490 return -EISDIR;
1491
Christoph Hellwige4543ed2005-11-08 21:35:04 -08001492 error = vfs_permission(nd, acc_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 if (error)
1494 return error;
1495
1496 /*
1497 * FIFO's, sockets and device files are special: they don't
1498 * actually live on the filesystem itself, and as such you
1499 * can write to them even if the filesystem is read-only.
1500 */
1501 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1502 flag &= ~O_TRUNC;
1503 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1504 if (nd->mnt->mnt_flags & MNT_NODEV)
1505 return -EACCES;
1506
1507 flag &= ~O_TRUNC;
1508 } else if (IS_RDONLY(inode) && (flag & FMODE_WRITE))
1509 return -EROFS;
1510 /*
1511 * An append-only file must be opened in append mode for writing.
1512 */
1513 if (IS_APPEND(inode)) {
1514 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1515 return -EPERM;
1516 if (flag & O_TRUNC)
1517 return -EPERM;
1518 }
1519
1520 /* O_NOATIME can only be set by the owner or superuser */
1521 if (flag & O_NOATIME)
1522 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1523 return -EPERM;
1524
1525 /*
1526 * Ensure there are no outstanding leases on the file.
1527 */
1528 error = break_lease(inode, flag);
1529 if (error)
1530 return error;
1531
1532 if (flag & O_TRUNC) {
1533 error = get_write_access(inode);
1534 if (error)
1535 return error;
1536
1537 /*
1538 * Refuse to truncate files with mandatory locks held on them.
1539 */
1540 error = locks_verify_locked(inode);
1541 if (!error) {
1542 DQUOT_INIT(inode);
1543
NeilBrown4a301312006-01-08 01:02:39 -08001544 error = do_truncate(dentry, 0, ATTR_MTIME|ATTR_CTIME, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 }
1546 put_write_access(inode);
1547 if (error)
1548 return error;
1549 } else
1550 if (flag & FMODE_WRITE)
1551 DQUOT_INIT(inode);
1552
1553 return 0;
1554}
1555
1556/*
1557 * open_namei()
1558 *
1559 * namei for open - this is in fact almost the whole open-routine.
1560 *
1561 * Note that the low bits of "flag" aren't the same as in the open
1562 * system call - they are 00 - no permissions needed
1563 * 01 - read permission needed
1564 * 10 - write permission needed
1565 * 11 - read/write permissions needed
1566 * which is a lot more logical, and also allows the "no perm" needed
1567 * for symlinks (where the permissions are checked later).
1568 * SMP-safe
1569 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001570int open_namei(int dfd, const char *pathname, int flag,
1571 int mode, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572{
Trond Myklebust834f2a42005-10-18 14:20:16 -07001573 int acc_mode, error;
Al Viro4e7506e2005-06-06 13:36:00 -07001574 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 struct dentry *dir;
1576 int count = 0;
1577
1578 acc_mode = ACC_MODE(flag);
1579
Trond Myklebust834f2a42005-10-18 14:20:16 -07001580 /* O_TRUNC implies we need access checks for write permissions */
1581 if (flag & O_TRUNC)
1582 acc_mode |= MAY_WRITE;
1583
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 /* Allow the LSM permission hook to distinguish append
1585 access from general write access. */
1586 if (flag & O_APPEND)
1587 acc_mode |= MAY_APPEND;
1588
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 /*
1590 * The simplest case - just a plain lookup.
1591 */
1592 if (!(flag & O_CREAT)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001593 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1594 nd, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 if (error)
1596 return error;
1597 goto ok;
1598 }
1599
1600 /*
1601 * Create - we need to know the parent.
1602 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001603 error = path_lookup_create(dfd,pathname,LOOKUP_PARENT,nd,flag,mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 if (error)
1605 return error;
1606
1607 /*
1608 * We have the parent and last component. First of all, check
1609 * that we are not asked to creat(2) an obvious directory - that
1610 * will not do.
1611 */
1612 error = -EISDIR;
1613 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1614 goto exit;
1615
1616 dir = nd->dentry;
1617 nd->flags &= ~LOOKUP_PARENT;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001618 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001619 path.dentry = lookup_hash(nd);
Al Virod73ffe12005-06-06 13:36:01 -07001620 path.mnt = nd->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621
1622do_last:
Al Viro4e7506e2005-06-06 13:36:00 -07001623 error = PTR_ERR(path.dentry);
1624 if (IS_ERR(path.dentry)) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001625 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 goto exit;
1627 }
1628
1629 /* Negative dentry, just create the file */
Al Viro4e7506e2005-06-06 13:36:00 -07001630 if (!path.dentry->d_inode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631 if (!IS_POSIXACL(dir->d_inode))
1632 mode &= ~current->fs->umask;
Al Viro4e7506e2005-06-06 13:36:00 -07001633 error = vfs_create(dir->d_inode, path.dentry, mode, nd);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001634 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 dput(nd->dentry);
Al Viro4e7506e2005-06-06 13:36:00 -07001636 nd->dentry = path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 if (error)
1638 goto exit;
1639 /* Don't check for write permission, don't truncate */
1640 acc_mode = 0;
1641 flag &= ~O_TRUNC;
1642 goto ok;
1643 }
1644
1645 /*
1646 * It already exists.
1647 */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001648 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649
1650 error = -EEXIST;
1651 if (flag & O_EXCL)
1652 goto exit_dput;
1653
Al Viroe13b2102005-06-06 13:36:06 -07001654 if (__follow_mount(&path)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 error = -ELOOP;
Al Viroba7a4c12005-06-06 13:36:08 -07001656 if (flag & O_NOFOLLOW)
1657 goto exit_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 }
1659 error = -ENOENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001660 if (!path.dentry->d_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 goto exit_dput;
Al Viro4e7506e2005-06-06 13:36:00 -07001662 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 goto do_link;
1664
Miklos Szeredi09dd17d2005-09-06 15:18:21 -07001665 path_to_nameidata(&path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 error = -EISDIR;
Al Viro4e7506e2005-06-06 13:36:00 -07001667 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 goto exit;
1669ok:
1670 error = may_open(nd, acc_mode, flag);
1671 if (error)
1672 goto exit;
1673 return 0;
1674
1675exit_dput:
Miklos Szeredi09dd17d2005-09-06 15:18:21 -07001676 dput_path(&path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677exit:
Trond Myklebust834f2a42005-10-18 14:20:16 -07001678 if (!IS_ERR(nd->intent.open.file))
1679 release_open_intent(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 path_release(nd);
1681 return error;
1682
1683do_link:
1684 error = -ELOOP;
1685 if (flag & O_NOFOLLOW)
1686 goto exit_dput;
1687 /*
1688 * This is subtle. Instead of calling do_follow_link() we do the
1689 * thing by hands. The reason is that this way we have zero link_count
1690 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1691 * After that we have the parent and last component, i.e.
1692 * we are in the same situation as after the first path_walk().
1693 * Well, almost - if the last component is normal we get its copy
1694 * stored in nd->last.name and we will have to putname() it when we
1695 * are done. Procfs-like symlinks just set LAST_BIND.
1696 */
1697 nd->flags |= LOOKUP_PARENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001698 error = security_inode_follow_link(path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 if (error)
1700 goto exit_dput;
Al Virocd4e91d2005-06-06 13:36:03 -07001701 error = __do_follow_link(&path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 if (error)
1703 return error;
1704 nd->flags &= ~LOOKUP_PARENT;
Al Virod671d5e2005-06-06 13:36:04 -07001705 if (nd->last_type == LAST_BIND)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 goto ok;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 error = -EISDIR;
1708 if (nd->last_type != LAST_NORM)
1709 goto exit;
1710 if (nd->last.name[nd->last.len]) {
Linus Torvalds82984112005-10-06 21:54:21 -07001711 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 goto exit;
1713 }
1714 error = -ELOOP;
1715 if (count++==32) {
Linus Torvalds82984112005-10-06 21:54:21 -07001716 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 goto exit;
1718 }
1719 dir = nd->dentry;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001720 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001721 path.dentry = lookup_hash(nd);
Al Virod671d5e2005-06-06 13:36:04 -07001722 path.mnt = nd->mnt;
Linus Torvalds82984112005-10-06 21:54:21 -07001723 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 goto do_last;
1725}
1726
1727/**
1728 * lookup_create - lookup a dentry, creating it if it doesn't exist
1729 * @nd: nameidata info
1730 * @is_dir: directory flag
1731 *
1732 * Simple function to lookup and return a dentry and create it
1733 * if it doesn't exist. Is SMP-safe.
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001734 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001735 * Returns with nd->dentry->d_inode->i_mutex locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 */
1737struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1738{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001739 struct dentry *dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001741 mutex_lock(&nd->dentry->d_inode->i_mutex);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001742 /*
1743 * Yucky last component or no last component at all?
1744 * (foo/., foo/.., /////)
1745 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 if (nd->last_type != LAST_NORM)
1747 goto fail;
1748 nd->flags &= ~LOOKUP_PARENT;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001749
1750 /*
1751 * Do the final lookup.
1752 */
Christoph Hellwig49705b72005-11-08 21:35:06 -08001753 dentry = lookup_hash(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 if (IS_ERR(dentry))
1755 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001756
1757 /*
1758 * Special case - lookup gave negative, but... we had foo/bar/
1759 * From the vfs_mknod() POV we just have a negative dentry -
1760 * all is fine. Let's be bastards - you had / on the end, you've
1761 * been asking for (non-existent) directory. -ENOENT for you.
1762 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1764 goto enoent;
1765 return dentry;
1766enoent:
1767 dput(dentry);
1768 dentry = ERR_PTR(-ENOENT);
1769fail:
1770 return dentry;
1771}
Christoph Hellwigf81a0bf2005-05-19 12:26:43 -07001772EXPORT_SYMBOL_GPL(lookup_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773
1774int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1775{
1776 int error = may_create(dir, dentry, NULL);
1777
1778 if (error)
1779 return error;
1780
1781 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1782 return -EPERM;
1783
1784 if (!dir->i_op || !dir->i_op->mknod)
1785 return -EPERM;
1786
1787 error = security_inode_mknod(dir, dentry, mode, dev);
1788 if (error)
1789 return error;
1790
1791 DQUOT_INIT(dir);
1792 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001793 if (!error)
Robert Love0eeca282005-07-12 17:06:03 -04001794 fsnotify_create(dir, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 return error;
1796}
1797
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001798asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
1799 unsigned dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800{
1801 int error = 0;
1802 char * tmp;
1803 struct dentry * dentry;
1804 struct nameidata nd;
1805
1806 if (S_ISDIR(mode))
1807 return -EPERM;
1808 tmp = getname(filename);
1809 if (IS_ERR(tmp))
1810 return PTR_ERR(tmp);
1811
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001812 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813 if (error)
1814 goto out;
1815 dentry = lookup_create(&nd, 0);
1816 error = PTR_ERR(dentry);
1817
1818 if (!IS_POSIXACL(nd.dentry->d_inode))
1819 mode &= ~current->fs->umask;
1820 if (!IS_ERR(dentry)) {
1821 switch (mode & S_IFMT) {
1822 case 0: case S_IFREG:
1823 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1824 break;
1825 case S_IFCHR: case S_IFBLK:
1826 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,
1827 new_decode_dev(dev));
1828 break;
1829 case S_IFIFO: case S_IFSOCK:
1830 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,0);
1831 break;
1832 case S_IFDIR:
1833 error = -EPERM;
1834 break;
1835 default:
1836 error = -EINVAL;
1837 }
1838 dput(dentry);
1839 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001840 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 path_release(&nd);
1842out:
1843 putname(tmp);
1844
1845 return error;
1846}
1847
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001848asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
1849{
1850 return sys_mknodat(AT_FDCWD, filename, mode, dev);
1851}
1852
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1854{
1855 int error = may_create(dir, dentry, NULL);
1856
1857 if (error)
1858 return error;
1859
1860 if (!dir->i_op || !dir->i_op->mkdir)
1861 return -EPERM;
1862
1863 mode &= (S_IRWXUGO|S_ISVTX);
1864 error = security_inode_mkdir(dir, dentry, mode);
1865 if (error)
1866 return error;
1867
1868 DQUOT_INIT(dir);
1869 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001870 if (!error)
Robert Love0eeca282005-07-12 17:06:03 -04001871 fsnotify_mkdir(dir, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 return error;
1873}
1874
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001875asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876{
1877 int error = 0;
1878 char * tmp;
1879
1880 tmp = getname(pathname);
1881 error = PTR_ERR(tmp);
1882 if (!IS_ERR(tmp)) {
1883 struct dentry *dentry;
1884 struct nameidata nd;
1885
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001886 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 if (error)
1888 goto out;
1889 dentry = lookup_create(&nd, 1);
1890 error = PTR_ERR(dentry);
1891 if (!IS_ERR(dentry)) {
1892 if (!IS_POSIXACL(nd.dentry->d_inode))
1893 mode &= ~current->fs->umask;
1894 error = vfs_mkdir(nd.dentry->d_inode, dentry, mode);
1895 dput(dentry);
1896 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001897 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898 path_release(&nd);
1899out:
1900 putname(tmp);
1901 }
1902
1903 return error;
1904}
1905
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001906asmlinkage long sys_mkdir(const char __user *pathname, int mode)
1907{
1908 return sys_mkdirat(AT_FDCWD, pathname, mode);
1909}
1910
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911/*
1912 * We try to drop the dentry early: we should have
1913 * a usage count of 2 if we're the only user of this
1914 * dentry, and if that is true (possibly after pruning
1915 * the dcache), then we drop the dentry now.
1916 *
1917 * A low-level filesystem can, if it choses, legally
1918 * do a
1919 *
1920 * if (!d_unhashed(dentry))
1921 * return -EBUSY;
1922 *
1923 * if it cannot handle the case of removing a directory
1924 * that is still in use by something else..
1925 */
1926void dentry_unhash(struct dentry *dentry)
1927{
1928 dget(dentry);
1929 if (atomic_read(&dentry->d_count))
1930 shrink_dcache_parent(dentry);
1931 spin_lock(&dcache_lock);
1932 spin_lock(&dentry->d_lock);
1933 if (atomic_read(&dentry->d_count) == 2)
1934 __d_drop(dentry);
1935 spin_unlock(&dentry->d_lock);
1936 spin_unlock(&dcache_lock);
1937}
1938
1939int vfs_rmdir(struct inode *dir, struct dentry *dentry)
1940{
1941 int error = may_delete(dir, dentry, 1);
1942
1943 if (error)
1944 return error;
1945
1946 if (!dir->i_op || !dir->i_op->rmdir)
1947 return -EPERM;
1948
1949 DQUOT_INIT(dir);
1950
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001951 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 dentry_unhash(dentry);
1953 if (d_mountpoint(dentry))
1954 error = -EBUSY;
1955 else {
1956 error = security_inode_rmdir(dir, dentry);
1957 if (!error) {
1958 error = dir->i_op->rmdir(dir, dentry);
1959 if (!error)
1960 dentry->d_inode->i_flags |= S_DEAD;
1961 }
1962 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001963 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 d_delete(dentry);
1966 }
1967 dput(dentry);
1968
1969 return error;
1970}
1971
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001972static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973{
1974 int error = 0;
1975 char * name;
1976 struct dentry *dentry;
1977 struct nameidata nd;
1978
1979 name = getname(pathname);
1980 if(IS_ERR(name))
1981 return PTR_ERR(name);
1982
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001983 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 if (error)
1985 goto exit;
1986
1987 switch(nd.last_type) {
1988 case LAST_DOTDOT:
1989 error = -ENOTEMPTY;
1990 goto exit1;
1991 case LAST_DOT:
1992 error = -EINVAL;
1993 goto exit1;
1994 case LAST_ROOT:
1995 error = -EBUSY;
1996 goto exit1;
1997 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001998 mutex_lock(&nd.dentry->d_inode->i_mutex);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001999 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 error = PTR_ERR(dentry);
2001 if (!IS_ERR(dentry)) {
2002 error = vfs_rmdir(nd.dentry->d_inode, dentry);
2003 dput(dentry);
2004 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002005 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006exit1:
2007 path_release(&nd);
2008exit:
2009 putname(name);
2010 return error;
2011}
2012
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002013asmlinkage long sys_rmdir(const char __user *pathname)
2014{
2015 return do_rmdir(AT_FDCWD, pathname);
2016}
2017
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018int vfs_unlink(struct inode *dir, struct dentry *dentry)
2019{
2020 int error = may_delete(dir, dentry, 0);
2021
2022 if (error)
2023 return error;
2024
2025 if (!dir->i_op || !dir->i_op->unlink)
2026 return -EPERM;
2027
2028 DQUOT_INIT(dir);
2029
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002030 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 if (d_mountpoint(dentry))
2032 error = -EBUSY;
2033 else {
2034 error = security_inode_unlink(dir, dentry);
2035 if (!error)
2036 error = dir->i_op->unlink(dir, dentry);
2037 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002038 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039
2040 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2041 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2042 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 }
Robert Love0eeca282005-07-12 17:06:03 -04002044
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 return error;
2046}
2047
2048/*
2049 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002050 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 * writeout happening, and we don't want to prevent access to the directory
2052 * while waiting on the I/O.
2053 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002054static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055{
2056 int error = 0;
2057 char * name;
2058 struct dentry *dentry;
2059 struct nameidata nd;
2060 struct inode *inode = NULL;
2061
2062 name = getname(pathname);
2063 if(IS_ERR(name))
2064 return PTR_ERR(name);
2065
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002066 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 if (error)
2068 goto exit;
2069 error = -EISDIR;
2070 if (nd.last_type != LAST_NORM)
2071 goto exit1;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002072 mutex_lock(&nd.dentry->d_inode->i_mutex);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002073 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074 error = PTR_ERR(dentry);
2075 if (!IS_ERR(dentry)) {
2076 /* Why not before? Because we want correct error value */
2077 if (nd.last.name[nd.last.len])
2078 goto slashes;
2079 inode = dentry->d_inode;
2080 if (inode)
2081 atomic_inc(&inode->i_count);
2082 error = vfs_unlink(nd.dentry->d_inode, dentry);
2083 exit2:
2084 dput(dentry);
2085 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002086 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087 if (inode)
2088 iput(inode); /* truncate the inode here */
2089exit1:
2090 path_release(&nd);
2091exit:
2092 putname(name);
2093 return error;
2094
2095slashes:
2096 error = !dentry->d_inode ? -ENOENT :
2097 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2098 goto exit2;
2099}
2100
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002101asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2102{
2103 if ((flag & ~AT_REMOVEDIR) != 0)
2104 return -EINVAL;
2105
2106 if (flag & AT_REMOVEDIR)
2107 return do_rmdir(dfd, pathname);
2108
2109 return do_unlinkat(dfd, pathname);
2110}
2111
2112asmlinkage long sys_unlink(const char __user *pathname)
2113{
2114 return do_unlinkat(AT_FDCWD, pathname);
2115}
2116
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname, int mode)
2118{
2119 int error = may_create(dir, dentry, NULL);
2120
2121 if (error)
2122 return error;
2123
2124 if (!dir->i_op || !dir->i_op->symlink)
2125 return -EPERM;
2126
2127 error = security_inode_symlink(dir, dentry, oldname);
2128 if (error)
2129 return error;
2130
2131 DQUOT_INIT(dir);
2132 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002133 if (!error)
Robert Love0eeca282005-07-12 17:06:03 -04002134 fsnotify_create(dir, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135 return error;
2136}
2137
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002138asmlinkage long sys_symlinkat(const char __user *oldname,
2139 int newdfd, const char __user *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140{
2141 int error = 0;
2142 char * from;
2143 char * to;
2144
2145 from = getname(oldname);
2146 if(IS_ERR(from))
2147 return PTR_ERR(from);
2148 to = getname(newname);
2149 error = PTR_ERR(to);
2150 if (!IS_ERR(to)) {
2151 struct dentry *dentry;
2152 struct nameidata nd;
2153
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002154 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155 if (error)
2156 goto out;
2157 dentry = lookup_create(&nd, 0);
2158 error = PTR_ERR(dentry);
2159 if (!IS_ERR(dentry)) {
2160 error = vfs_symlink(nd.dentry->d_inode, dentry, from, S_IALLUGO);
2161 dput(dentry);
2162 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002163 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 path_release(&nd);
2165out:
2166 putname(to);
2167 }
2168 putname(from);
2169 return error;
2170}
2171
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002172asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2173{
2174 return sys_symlinkat(oldname, AT_FDCWD, newname);
2175}
2176
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2178{
2179 struct inode *inode = old_dentry->d_inode;
2180 int error;
2181
2182 if (!inode)
2183 return -ENOENT;
2184
2185 error = may_create(dir, new_dentry, NULL);
2186 if (error)
2187 return error;
2188
2189 if (dir->i_sb != inode->i_sb)
2190 return -EXDEV;
2191
2192 /*
2193 * A link to an append-only or immutable file cannot be created.
2194 */
2195 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2196 return -EPERM;
2197 if (!dir->i_op || !dir->i_op->link)
2198 return -EPERM;
2199 if (S_ISDIR(old_dentry->d_inode->i_mode))
2200 return -EPERM;
2201
2202 error = security_inode_link(old_dentry, dir, new_dentry);
2203 if (error)
2204 return error;
2205
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002206 mutex_lock(&old_dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207 DQUOT_INIT(dir);
2208 error = dir->i_op->link(old_dentry, dir, new_dentry);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002209 mutex_unlock(&old_dentry->d_inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002210 if (!error)
Robert Love0eeca282005-07-12 17:06:03 -04002211 fsnotify_create(dir, new_dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 return error;
2213}
2214
2215/*
2216 * Hardlinks are often used in delicate situations. We avoid
2217 * security-related surprises by not following symlinks on the
2218 * newname. --KAB
2219 *
2220 * We don't follow them on the oldname either to be compatible
2221 * with linux 2.0, and to avoid hard-linking to directories
2222 * and other special files. --ADM
2223 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002224asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
2225 int newdfd, const char __user *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226{
2227 struct dentry *new_dentry;
2228 struct nameidata nd, old_nd;
2229 int error;
2230 char * to;
2231
2232 to = getname(newname);
2233 if (IS_ERR(to))
2234 return PTR_ERR(to);
2235
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002236 error = __user_walk_fd(olddfd, oldname, 0, &old_nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 if (error)
2238 goto exit;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002239 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 if (error)
2241 goto out;
2242 error = -EXDEV;
2243 if (old_nd.mnt != nd.mnt)
2244 goto out_release;
2245 new_dentry = lookup_create(&nd, 0);
2246 error = PTR_ERR(new_dentry);
2247 if (!IS_ERR(new_dentry)) {
2248 error = vfs_link(old_nd.dentry, nd.dentry->d_inode, new_dentry);
2249 dput(new_dentry);
2250 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002251 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252out_release:
2253 path_release(&nd);
2254out:
2255 path_release(&old_nd);
2256exit:
2257 putname(to);
2258
2259 return error;
2260}
2261
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002262asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2263{
2264 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname);
2265}
2266
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267/*
2268 * The worst of all namespace operations - renaming directory. "Perverted"
2269 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2270 * Problems:
2271 * a) we can get into loop creation. Check is done in is_subdir().
2272 * b) race potential - two innocent renames can create a loop together.
2273 * That's where 4.4 screws up. Current fix: serialization on
2274 * sb->s_vfs_rename_sem. We might be more accurate, but that's another
2275 * story.
2276 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002277 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278 * whether the target exists). Solution: try to be smart with locking
2279 * order for inodes. We rely on the fact that tree topology may change
2280 * only under ->s_vfs_rename_sem _and_ that parent of the object we
2281 * move will be locked. Thus we can rank directories by the tree
2282 * (ancestors first) and rank all non-directories after them.
2283 * That works since everybody except rename does "lock parent, lookup,
2284 * lock child" and rename is under ->s_vfs_rename_sem.
2285 * HOWEVER, it relies on the assumption that any object with ->lookup()
2286 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2287 * we'd better make sure that there's no link(2) for them.
2288 * d) some filesystems don't support opened-but-unlinked directories,
2289 * either because of layout or because they are not ready to deal with
2290 * all cases correctly. The latter will be fixed (taking this sort of
2291 * stuff into VFS), but the former is not going away. Solution: the same
2292 * trick as in rmdir().
2293 * e) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002294 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002296 * ->i_mutex on parents, which works but leads to some truely excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297 * locking].
2298 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07002299static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2300 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301{
2302 int error = 0;
2303 struct inode *target;
2304
2305 /*
2306 * If we are going to change the parent - check write permissions,
2307 * we'll need to flip '..'.
2308 */
2309 if (new_dir != old_dir) {
2310 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2311 if (error)
2312 return error;
2313 }
2314
2315 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2316 if (error)
2317 return error;
2318
2319 target = new_dentry->d_inode;
2320 if (target) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002321 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322 dentry_unhash(new_dentry);
2323 }
2324 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2325 error = -EBUSY;
2326 else
2327 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2328 if (target) {
2329 if (!error)
2330 target->i_flags |= S_DEAD;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002331 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332 if (d_unhashed(new_dentry))
2333 d_rehash(new_dentry);
2334 dput(new_dentry);
2335 }
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002336 if (!error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338 return error;
2339}
2340
Adrian Bunk75c96f82005-05-05 16:16:09 -07002341static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2342 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343{
2344 struct inode *target;
2345 int error;
2346
2347 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2348 if (error)
2349 return error;
2350
2351 dget(new_dentry);
2352 target = new_dentry->d_inode;
2353 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002354 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2356 error = -EBUSY;
2357 else
2358 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2359 if (!error) {
2360 /* The following d_move() should become unconditional */
2361 if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
2362 d_move(old_dentry, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363 }
2364 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002365 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366 dput(new_dentry);
2367 return error;
2368}
2369
2370int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2371 struct inode *new_dir, struct dentry *new_dentry)
2372{
2373 int error;
2374 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Robert Love0eeca282005-07-12 17:06:03 -04002375 const char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376
2377 if (old_dentry->d_inode == new_dentry->d_inode)
2378 return 0;
2379
2380 error = may_delete(old_dir, old_dentry, is_dir);
2381 if (error)
2382 return error;
2383
2384 if (!new_dentry->d_inode)
2385 error = may_create(new_dir, new_dentry, NULL);
2386 else
2387 error = may_delete(new_dir, new_dentry, is_dir);
2388 if (error)
2389 return error;
2390
2391 if (!old_dir->i_op || !old_dir->i_op->rename)
2392 return -EPERM;
2393
2394 DQUOT_INIT(old_dir);
2395 DQUOT_INIT(new_dir);
2396
Robert Love0eeca282005-07-12 17:06:03 -04002397 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2398
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399 if (is_dir)
2400 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2401 else
2402 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2403 if (!error) {
Robert Love0eeca282005-07-12 17:06:03 -04002404 const char *new_name = old_dentry->d_name.name;
John McCutchan89204c42005-08-15 12:13:28 -04002405 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2406 new_dentry->d_inode, old_dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407 }
Robert Love0eeca282005-07-12 17:06:03 -04002408 fsnotify_oldname_free(old_name);
2409
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410 return error;
2411}
2412
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002413static int do_rename(int olddfd, const char *oldname,
2414 int newdfd, const char *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415{
2416 int error = 0;
2417 struct dentry * old_dir, * new_dir;
2418 struct dentry * old_dentry, *new_dentry;
2419 struct dentry * trap;
2420 struct nameidata oldnd, newnd;
2421
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002422 error = do_path_lookup(olddfd, oldname, LOOKUP_PARENT, &oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 if (error)
2424 goto exit;
2425
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002426 error = do_path_lookup(newdfd, newname, LOOKUP_PARENT, &newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 if (error)
2428 goto exit1;
2429
2430 error = -EXDEV;
2431 if (oldnd.mnt != newnd.mnt)
2432 goto exit2;
2433
2434 old_dir = oldnd.dentry;
2435 error = -EBUSY;
2436 if (oldnd.last_type != LAST_NORM)
2437 goto exit2;
2438
2439 new_dir = newnd.dentry;
2440 if (newnd.last_type != LAST_NORM)
2441 goto exit2;
2442
2443 trap = lock_rename(new_dir, old_dir);
2444
Christoph Hellwig49705b72005-11-08 21:35:06 -08002445 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446 error = PTR_ERR(old_dentry);
2447 if (IS_ERR(old_dentry))
2448 goto exit3;
2449 /* source must exist */
2450 error = -ENOENT;
2451 if (!old_dentry->d_inode)
2452 goto exit4;
2453 /* unless the source is a directory trailing slashes give -ENOTDIR */
2454 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2455 error = -ENOTDIR;
2456 if (oldnd.last.name[oldnd.last.len])
2457 goto exit4;
2458 if (newnd.last.name[newnd.last.len])
2459 goto exit4;
2460 }
2461 /* source should not be ancestor of target */
2462 error = -EINVAL;
2463 if (old_dentry == trap)
2464 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08002465 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466 error = PTR_ERR(new_dentry);
2467 if (IS_ERR(new_dentry))
2468 goto exit4;
2469 /* target should not be an ancestor of source */
2470 error = -ENOTEMPTY;
2471 if (new_dentry == trap)
2472 goto exit5;
2473
2474 error = vfs_rename(old_dir->d_inode, old_dentry,
2475 new_dir->d_inode, new_dentry);
2476exit5:
2477 dput(new_dentry);
2478exit4:
2479 dput(old_dentry);
2480exit3:
2481 unlock_rename(new_dir, old_dir);
2482exit2:
2483 path_release(&newnd);
2484exit1:
2485 path_release(&oldnd);
2486exit:
2487 return error;
2488}
2489
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002490asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2491 int newdfd, const char __user *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492{
2493 int error;
2494 char * from;
2495 char * to;
2496
2497 from = getname(oldname);
2498 if(IS_ERR(from))
2499 return PTR_ERR(from);
2500 to = getname(newname);
2501 error = PTR_ERR(to);
2502 if (!IS_ERR(to)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002503 error = do_rename(olddfd, from, newdfd, to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504 putname(to);
2505 }
2506 putname(from);
2507 return error;
2508}
2509
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002510asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2511{
2512 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2513}
2514
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2516{
2517 int len;
2518
2519 len = PTR_ERR(link);
2520 if (IS_ERR(link))
2521 goto out;
2522
2523 len = strlen(link);
2524 if (len > (unsigned) buflen)
2525 len = buflen;
2526 if (copy_to_user(buffer, link, len))
2527 len = -EFAULT;
2528out:
2529 return len;
2530}
2531
2532/*
2533 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2534 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2535 * using) it for any given inode is up to filesystem.
2536 */
2537int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2538{
2539 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002540 void *cookie;
2541
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002543 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2544 if (!IS_ERR(cookie)) {
2545 int res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546 if (dentry->d_inode->i_op->put_link)
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002547 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2548 cookie = ERR_PTR(res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549 }
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002550 return PTR_ERR(cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551}
2552
2553int vfs_follow_link(struct nameidata *nd, const char *link)
2554{
2555 return __vfs_follow_link(nd, link);
2556}
2557
2558/* get the link contents into pagecache */
2559static char *page_getlink(struct dentry * dentry, struct page **ppage)
2560{
2561 struct page * page;
2562 struct address_space *mapping = dentry->d_inode->i_mapping;
2563 page = read_cache_page(mapping, 0, (filler_t *)mapping->a_ops->readpage,
2564 NULL);
2565 if (IS_ERR(page))
2566 goto sync_fail;
2567 wait_on_page_locked(page);
2568 if (!PageUptodate(page))
2569 goto async_fail;
2570 *ppage = page;
2571 return kmap(page);
2572
2573async_fail:
2574 page_cache_release(page);
2575 return ERR_PTR(-EIO);
2576
2577sync_fail:
2578 return (char*)page;
2579}
2580
2581int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2582{
2583 struct page *page = NULL;
2584 char *s = page_getlink(dentry, &page);
2585 int res = vfs_readlink(dentry,buffer,buflen,s);
2586 if (page) {
2587 kunmap(page);
2588 page_cache_release(page);
2589 }
2590 return res;
2591}
2592
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002593void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002595 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002597 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598}
2599
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002600void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002602 struct page *page = cookie;
2603
2604 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605 kunmap(page);
2606 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607 }
2608}
2609
2610int page_symlink(struct inode *inode, const char *symname, int len)
2611{
2612 struct address_space *mapping = inode->i_mapping;
2613 struct page *page = grab_cache_page(mapping, 0);
2614 int err = -ENOMEM;
2615 char *kaddr;
2616
2617 if (!page)
2618 goto fail;
2619 err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
2620 if (err)
2621 goto fail_map;
2622 kaddr = kmap_atomic(page, KM_USER0);
2623 memcpy(kaddr, symname, len-1);
2624 kunmap_atomic(kaddr, KM_USER0);
2625 mapping->a_ops->commit_write(NULL, page, 0, len-1);
2626 /*
2627 * Notice that we are _not_ going to block here - end of page is
2628 * unmapped, so this will only try to map the rest of page, see
2629 * that it is unmapped (typically even will not look into inode -
2630 * ->i_size will be enough for everything) and zero it out.
2631 * OTOH it's obviously correct and should make the page up-to-date.
2632 */
2633 if (!PageUptodate(page)) {
2634 err = mapping->a_ops->readpage(NULL, page);
2635 wait_on_page_locked(page);
2636 } else {
2637 unlock_page(page);
2638 }
2639 page_cache_release(page);
2640 if (err < 0)
2641 goto fail;
2642 mark_inode_dirty(inode);
2643 return 0;
2644fail_map:
2645 unlock_page(page);
2646 page_cache_release(page);
2647fail:
2648 return err;
2649}
2650
2651struct inode_operations page_symlink_inode_operations = {
2652 .readlink = generic_readlink,
2653 .follow_link = page_follow_link_light,
2654 .put_link = page_put_link,
2655};
2656
2657EXPORT_SYMBOL(__user_walk);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002658EXPORT_SYMBOL(__user_walk_fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659EXPORT_SYMBOL(follow_down);
2660EXPORT_SYMBOL(follow_up);
2661EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2662EXPORT_SYMBOL(getname);
2663EXPORT_SYMBOL(lock_rename);
2664EXPORT_SYMBOL(lookup_hash);
2665EXPORT_SYMBOL(lookup_one_len);
2666EXPORT_SYMBOL(page_follow_link_light);
2667EXPORT_SYMBOL(page_put_link);
2668EXPORT_SYMBOL(page_readlink);
2669EXPORT_SYMBOL(page_symlink);
2670EXPORT_SYMBOL(page_symlink_inode_operations);
2671EXPORT_SYMBOL(path_lookup);
2672EXPORT_SYMBOL(path_release);
2673EXPORT_SYMBOL(path_walk);
2674EXPORT_SYMBOL(permission);
Christoph Hellwige4543ed2005-11-08 21:35:04 -08002675EXPORT_SYMBOL(vfs_permission);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -08002676EXPORT_SYMBOL(file_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677EXPORT_SYMBOL(unlock_rename);
2678EXPORT_SYMBOL(vfs_create);
2679EXPORT_SYMBOL(vfs_follow_link);
2680EXPORT_SYMBOL(vfs_link);
2681EXPORT_SYMBOL(vfs_mkdir);
2682EXPORT_SYMBOL(vfs_mknod);
2683EXPORT_SYMBOL(generic_permission);
2684EXPORT_SYMBOL(vfs_readlink);
2685EXPORT_SYMBOL(vfs_rename);
2686EXPORT_SYMBOL(vfs_rmdir);
2687EXPORT_SYMBOL(vfs_symlink);
2688EXPORT_SYMBOL(vfs_unlink);
2689EXPORT_SYMBOL(dentry_unhash);
2690EXPORT_SYMBOL(generic_readlink);