blob: 02a824cd3c5c01b7af923ad81e96b940c2cd1a08 [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>
31#include <asm/namei.h>
32#include <asm/uaccess.h>
33
34#define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
35
36/* [Feb-1997 T. Schoebel-Theuer]
37 * Fundamental changes in the pathname lookup mechanisms (namei)
38 * were necessary because of omirr. The reason is that omirr needs
39 * to know the _real_ pathname, not the user-supplied one, in case
40 * of symlinks (and also when transname replacements occur).
41 *
42 * The new code replaces the old recursive symlink resolution with
43 * an iterative one (in case of non-nested symlink chains). It does
44 * this with calls to <fs>_follow_link().
45 * As a side effect, dir_namei(), _namei() and follow_link() are now
46 * replaced with a single function lookup_dentry() that can handle all
47 * the special cases of the former code.
48 *
49 * With the new dcache, the pathname is stored at each inode, at least as
50 * long as the refcount of the inode is positive. As a side effect, the
51 * size of the dcache depends on the inode cache and thus is dynamic.
52 *
53 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
54 * resolution to correspond with current state of the code.
55 *
56 * Note that the symlink resolution is not *completely* iterative.
57 * There is still a significant amount of tail- and mid- recursion in
58 * the algorithm. Also, note that <fs>_readlink() is not used in
59 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
60 * may return different results than <fs>_follow_link(). Many virtual
61 * filesystems (including /proc) exhibit this behavior.
62 */
63
64/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
65 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
66 * and the name already exists in form of a symlink, try to create the new
67 * name indicated by the symlink. The old code always complained that the
68 * name already exists, due to not following the symlink even if its target
69 * is nonexistent. The new semantics affects also mknod() and link() when
70 * the name is a symlink pointing to a non-existant name.
71 *
72 * I don't know which semantics is the right one, since I have no access
73 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
74 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
75 * "old" one. Personally, I think the new semantics is much more logical.
76 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
77 * file does succeed in both HP-UX and SunOs, but not in Solaris
78 * and in the old Linux semantics.
79 */
80
81/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
82 * semantics. See the comments in "open_namei" and "do_link" below.
83 *
84 * [10-Sep-98 Alan Modra] Another symlink change.
85 */
86
87/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
88 * inside the path - always follow.
89 * in the last component in creation/removal/renaming - never follow.
90 * if LOOKUP_FOLLOW passed - follow.
91 * if the pathname has trailing slashes - follow.
92 * otherwise - don't follow.
93 * (applied in that order).
94 *
95 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
96 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
97 * During the 2.4 we need to fix the userland stuff depending on it -
98 * hopefully we will be able to get rid of that wart in 2.5. So far only
99 * XEmacs seems to be relying on it...
100 */
101/*
102 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
103 * implemented. Let's see if raised priority of ->s_vfs_rename_sem gives
104 * any extra contention...
105 */
106
107/* In order to reduce some races, while at the same time doing additional
108 * checking and hopefully speeding things up, we copy filenames to the
109 * kernel data space before using them..
110 *
111 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
112 * PATH_MAX includes the nul terminator --RR.
113 */
114static inline int do_getname(const char __user *filename, char *page)
115{
116 int retval;
117 unsigned long len = PATH_MAX;
118
119 if (!segment_eq(get_fs(), KERNEL_DS)) {
120 if ((unsigned long) filename >= TASK_SIZE)
121 return -EFAULT;
122 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
123 len = TASK_SIZE - (unsigned long) filename;
124 }
125
126 retval = strncpy_from_user(page, filename, len);
127 if (retval > 0) {
128 if (retval < len)
129 return 0;
130 return -ENAMETOOLONG;
131 } else if (!retval)
132 retval = -ENOENT;
133 return retval;
134}
135
136char * getname(const char __user * filename)
137{
138 char *tmp, *result;
139
140 result = ERR_PTR(-ENOMEM);
141 tmp = __getname();
142 if (tmp) {
143 int retval = do_getname(filename, tmp);
144
145 result = tmp;
146 if (retval < 0) {
147 __putname(tmp);
148 result = ERR_PTR(retval);
149 }
150 }
151 audit_getname(result);
152 return result;
153}
154
155#ifdef CONFIG_AUDITSYSCALL
156void putname(const char *name)
157{
158 if (unlikely(current->audit_context))
159 audit_putname(name);
160 else
161 __putname(name);
162}
163EXPORT_SYMBOL(putname);
164#endif
165
166
167/**
168 * generic_permission - check for access rights on a Posix-like filesystem
169 * @inode: inode to check access rights for
170 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
171 * @check_acl: optional callback to check for Posix ACLs
172 *
173 * Used to check for read/write/execute permissions on a file.
174 * We use "fsuid" for this, letting us set arbitrary permissions
175 * for filesystem access without changing the "normal" uids which
176 * are used for other things..
177 */
178int generic_permission(struct inode *inode, int mask,
179 int (*check_acl)(struct inode *inode, int mask))
180{
181 umode_t mode = inode->i_mode;
182
183 if (current->fsuid == inode->i_uid)
184 mode >>= 6;
185 else {
186 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
187 int error = check_acl(inode, mask);
188 if (error == -EACCES)
189 goto check_capabilities;
190 else if (error != -EAGAIN)
191 return error;
192 }
193
194 if (in_group_p(inode->i_gid))
195 mode >>= 3;
196 }
197
198 /*
199 * If the DACs are ok we don't need any capability check.
200 */
201 if (((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask))
202 return 0;
203
204 check_capabilities:
205 /*
206 * Read/write DACs are always overridable.
207 * Executable DACs are overridable if at least one exec bit is set.
208 */
209 if (!(mask & MAY_EXEC) ||
210 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
211 if (capable(CAP_DAC_OVERRIDE))
212 return 0;
213
214 /*
215 * Searching includes executable on directories, else just read.
216 */
217 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
218 if (capable(CAP_DAC_READ_SEARCH))
219 return 0;
220
221 return -EACCES;
222}
223
224int permission(struct inode *inode, int mask, struct nameidata *nd)
225{
226 int retval, submask;
227
228 if (mask & MAY_WRITE) {
229 umode_t mode = inode->i_mode;
230
231 /*
232 * Nobody gets write access to a read-only fs.
233 */
234 if (IS_RDONLY(inode) &&
235 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
236 return -EROFS;
237
238 /*
239 * Nobody gets write access to an immutable file.
240 */
241 if (IS_IMMUTABLE(inode))
242 return -EACCES;
243 }
244
245
246 /* Ordinary permission routines do not understand MAY_APPEND. */
247 submask = mask & ~MAY_APPEND;
248 if (inode->i_op && inode->i_op->permission)
249 retval = inode->i_op->permission(inode, submask, nd);
250 else
251 retval = generic_permission(inode, submask, NULL);
252 if (retval)
253 return retval;
254
255 return security_inode_permission(inode, mask, nd);
256}
257
258/*
259 * get_write_access() gets write permission for a file.
260 * put_write_access() releases this write permission.
261 * This is used for regular files.
262 * We cannot support write (and maybe mmap read-write shared) accesses and
263 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
264 * can have the following values:
265 * 0: no writers, no VM_DENYWRITE mappings
266 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
267 * > 0: (i_writecount) users are writing to the file.
268 *
269 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
270 * except for the cases where we don't hold i_writecount yet. Then we need to
271 * use {get,deny}_write_access() - these functions check the sign and refuse
272 * to do the change if sign is wrong. Exclusion between them is provided by
273 * the inode->i_lock spinlock.
274 */
275
276int get_write_access(struct inode * inode)
277{
278 spin_lock(&inode->i_lock);
279 if (atomic_read(&inode->i_writecount) < 0) {
280 spin_unlock(&inode->i_lock);
281 return -ETXTBSY;
282 }
283 atomic_inc(&inode->i_writecount);
284 spin_unlock(&inode->i_lock);
285
286 return 0;
287}
288
289int deny_write_access(struct file * file)
290{
291 struct inode *inode = file->f_dentry->d_inode;
292
293 spin_lock(&inode->i_lock);
294 if (atomic_read(&inode->i_writecount) > 0) {
295 spin_unlock(&inode->i_lock);
296 return -ETXTBSY;
297 }
298 atomic_dec(&inode->i_writecount);
299 spin_unlock(&inode->i_lock);
300
301 return 0;
302}
303
304void path_release(struct nameidata *nd)
305{
306 dput(nd->dentry);
307 mntput(nd->mnt);
308}
309
310/*
311 * umount() mustn't call path_release()/mntput() as that would clear
312 * mnt_expiry_mark
313 */
314void path_release_on_umount(struct nameidata *nd)
315{
316 dput(nd->dentry);
Miklos Szeredi751c4042005-07-07 17:57:30 -0700317 mntput_no_expire(nd->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318}
319
320/*
321 * Internal lookup() using the new generic dcache.
322 * SMP-safe
323 */
324static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
325{
326 struct dentry * dentry = __d_lookup(parent, name);
327
328 /* lockess __d_lookup may fail due to concurrent d_move()
329 * in some unrelated directory, so try with d_lookup
330 */
331 if (!dentry)
332 dentry = d_lookup(parent, name);
333
334 if (dentry && dentry->d_op && dentry->d_op->d_revalidate) {
335 if (!dentry->d_op->d_revalidate(dentry, nd) && !d_invalidate(dentry)) {
336 dput(dentry);
337 dentry = NULL;
338 }
339 }
340 return dentry;
341}
342
343/*
344 * Short-cut version of permission(), for calling by
345 * path_walk(), when dcache lock is held. Combines parts
346 * of permission() and generic_permission(), and tests ONLY for
347 * MAY_EXEC permission.
348 *
349 * If appropriate, check DAC only. If not appropriate, or
350 * short-cut DAC fails, then call permission() to do more
351 * complete permission check.
352 */
353static inline int exec_permission_lite(struct inode *inode,
354 struct nameidata *nd)
355{
356 umode_t mode = inode->i_mode;
357
358 if (inode->i_op && inode->i_op->permission)
359 return -EAGAIN;
360
361 if (current->fsuid == inode->i_uid)
362 mode >>= 6;
363 else if (in_group_p(inode->i_gid))
364 mode >>= 3;
365
366 if (mode & MAY_EXEC)
367 goto ok;
368
369 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
370 goto ok;
371
372 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
373 goto ok;
374
375 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
376 goto ok;
377
378 return -EACCES;
379ok:
380 return security_inode_permission(inode, MAY_EXEC, nd);
381}
382
383/*
384 * This is called when everything else fails, and we actually have
385 * to go to the low-level filesystem to find out what we should do..
386 *
387 * We get the directory semaphore, and after getting that we also
388 * make sure that nobody added the entry to the dcache in the meantime..
389 * SMP-safe
390 */
391static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
392{
393 struct dentry * result;
394 struct inode *dir = parent->d_inode;
395
396 down(&dir->i_sem);
397 /*
398 * First re-do the cached lookup just in case it was created
399 * while we waited for the directory semaphore..
400 *
401 * FIXME! This could use version numbering or similar to
402 * avoid unnecessary cache lookups.
403 *
404 * The "dcache_lock" is purely to protect the RCU list walker
405 * from concurrent renames at this point (we mustn't get false
406 * negatives from the RCU list walk here, unlike the optimistic
407 * fast walk).
408 *
409 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
410 */
411 result = d_lookup(parent, name);
412 if (!result) {
413 struct dentry * dentry = d_alloc(parent, name);
414 result = ERR_PTR(-ENOMEM);
415 if (dentry) {
416 result = dir->i_op->lookup(dir, dentry, nd);
417 if (result)
418 dput(dentry);
419 else
420 result = dentry;
421 }
422 up(&dir->i_sem);
423 return result;
424 }
425
426 /*
427 * Uhhuh! Nasty case: the cache was re-populated while
428 * we waited on the semaphore. Need to revalidate.
429 */
430 up(&dir->i_sem);
431 if (result->d_op && result->d_op->d_revalidate) {
432 if (!result->d_op->d_revalidate(result, nd) && !d_invalidate(result)) {
433 dput(result);
434 result = ERR_PTR(-ENOENT);
435 }
436 }
437 return result;
438}
439
440static int __emul_lookup_dentry(const char *, struct nameidata *);
441
442/* SMP-safe */
443static inline int
444walk_init_root(const char *name, struct nameidata *nd)
445{
446 read_lock(&current->fs->lock);
447 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
448 nd->mnt = mntget(current->fs->altrootmnt);
449 nd->dentry = dget(current->fs->altroot);
450 read_unlock(&current->fs->lock);
451 if (__emul_lookup_dentry(name,nd))
452 return 0;
453 read_lock(&current->fs->lock);
454 }
455 nd->mnt = mntget(current->fs->rootmnt);
456 nd->dentry = dget(current->fs->root);
457 read_unlock(&current->fs->lock);
458 return 1;
459}
460
461static inline int __vfs_follow_link(struct nameidata *nd, const char *link)
462{
463 int res = 0;
464 char *name;
465 if (IS_ERR(link))
466 goto fail;
467
468 if (*link == '/') {
469 path_release(nd);
470 if (!walk_init_root(link, nd))
471 /* weird __emul_prefix() stuff did it */
472 goto out;
473 }
474 res = link_path_walk(link, nd);
475out:
476 if (nd->depth || res || nd->last_type!=LAST_NORM)
477 return res;
478 /*
479 * If it is an iterative symlinks resolution in open_namei() we
480 * have to copy the last component. And all that crap because of
481 * bloody create() on broken symlinks. Furrfu...
482 */
483 name = __getname();
484 if (unlikely(!name)) {
485 path_release(nd);
486 return -ENOMEM;
487 }
488 strcpy(name, nd->last.name);
489 nd->last.name = name;
490 return 0;
491fail:
492 path_release(nd);
493 return PTR_ERR(link);
494}
495
Al Viro90ebe562005-06-06 13:35:58 -0700496struct path {
497 struct vfsmount *mnt;
498 struct dentry *dentry;
499};
500
Al Virocd4e91d2005-06-06 13:36:03 -0700501static inline int __do_follow_link(struct path *path, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502{
503 int error;
Al Virocd4e91d2005-06-06 13:36:03 -0700504 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505
Al Virod671a1c2005-06-06 13:36:14 -0700506 touch_atime(path->mnt, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 nd_set_link(nd, NULL);
Al Virocd4e91d2005-06-06 13:36:03 -0700508
Al Viro39ca6d42005-06-06 13:36:12 -0700509 if (path->mnt == nd->mnt)
510 mntget(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 error = dentry->d_inode->i_op->follow_link(dentry, nd);
512 if (!error) {
513 char *s = nd_get_link(nd);
514 if (s)
515 error = __vfs_follow_link(nd, s);
516 if (dentry->d_inode->i_op->put_link)
517 dentry->d_inode->i_op->put_link(dentry, nd);
518 }
Al Virocd4e91d2005-06-06 13:36:03 -0700519 dput(dentry);
520 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521
522 return error;
523}
524
525/*
526 * This limits recursive symlink follows to 8, while
527 * limiting consecutive symlinks to 40.
528 *
529 * Without that kind of total limit, nasty chains of consecutive
530 * symlinks can cause almost arbitrarily long lookups.
531 */
Al Viro90ebe562005-06-06 13:35:58 -0700532static inline int do_follow_link(struct path *path, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533{
534 int err = -ELOOP;
535 if (current->link_count >= MAX_NESTED_LINKS)
536 goto loop;
537 if (current->total_link_count >= 40)
538 goto loop;
539 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
540 cond_resched();
Al Viro90ebe562005-06-06 13:35:58 -0700541 err = security_inode_follow_link(path->dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 if (err)
543 goto loop;
544 current->link_count++;
545 current->total_link_count++;
546 nd->depth++;
Al Virocd4e91d2005-06-06 13:36:03 -0700547 err = __do_follow_link(path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 current->link_count--;
549 nd->depth--;
550 return err;
551loop:
Al Viro5f92b3bc2005-06-06 13:35:59 -0700552 dput(path->dentry);
Al Virod9d29a22005-06-06 13:36:11 -0700553 if (path->mnt != nd->mnt)
554 mntput(path->mnt);
Al Viro839d9f92005-06-06 13:36:02 -0700555 path_release(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 return err;
557}
558
559int follow_up(struct vfsmount **mnt, struct dentry **dentry)
560{
561 struct vfsmount *parent;
562 struct dentry *mountpoint;
563 spin_lock(&vfsmount_lock);
564 parent=(*mnt)->mnt_parent;
565 if (parent == *mnt) {
566 spin_unlock(&vfsmount_lock);
567 return 0;
568 }
569 mntget(parent);
570 mountpoint=dget((*mnt)->mnt_mountpoint);
571 spin_unlock(&vfsmount_lock);
572 dput(*dentry);
573 *dentry = mountpoint;
574 mntput(*mnt);
575 *mnt = parent;
576 return 1;
577}
578
579/* no need for dcache_lock, as serialization is taken care in
580 * namespace.c
581 */
Al Viro463ffb22005-06-06 13:36:05 -0700582static int __follow_mount(struct path *path)
583{
584 int res = 0;
585 while (d_mountpoint(path->dentry)) {
586 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
587 if (!mounted)
588 break;
589 dput(path->dentry);
590 if (res)
591 mntput(path->mnt);
592 path->mnt = mounted;
593 path->dentry = dget(mounted->mnt_root);
594 res = 1;
595 }
596 return res;
597}
598
Al Viro58c465e2005-06-06 13:36:13 -0700599static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 while (d_mountpoint(*dentry)) {
602 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
603 if (!mounted)
604 break;
Al Viro58c465e2005-06-06 13:36:13 -0700605 dput(*dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 mntput(*mnt);
607 *mnt = mounted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 *dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610}
611
612/* no need for dcache_lock, as serialization is taken care in
613 * namespace.c
614 */
Al Viroe13b2102005-06-06 13:36:06 -0700615int follow_down(struct vfsmount **mnt, struct dentry **dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616{
617 struct vfsmount *mounted;
618
619 mounted = lookup_mnt(*mnt, *dentry);
620 if (mounted) {
Al Viroe13b2102005-06-06 13:36:06 -0700621 dput(*dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 mntput(*mnt);
623 *mnt = mounted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 *dentry = dget(mounted->mnt_root);
625 return 1;
626 }
627 return 0;
628}
629
Al Viro58c465e2005-06-06 13:36:13 -0700630static inline void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631{
632 while(1) {
633 struct vfsmount *parent;
Al Viro58c465e2005-06-06 13:36:13 -0700634 struct dentry *old = nd->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635
636 read_lock(&current->fs->lock);
Al Viro58c465e2005-06-06 13:36:13 -0700637 if (nd->dentry == current->fs->root &&
638 nd->mnt == current->fs->rootmnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 read_unlock(&current->fs->lock);
640 break;
641 }
642 read_unlock(&current->fs->lock);
643 spin_lock(&dcache_lock);
Al Viro58c465e2005-06-06 13:36:13 -0700644 if (nd->dentry != nd->mnt->mnt_root) {
645 nd->dentry = dget(nd->dentry->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 spin_unlock(&dcache_lock);
647 dput(old);
648 break;
649 }
650 spin_unlock(&dcache_lock);
651 spin_lock(&vfsmount_lock);
Al Viro58c465e2005-06-06 13:36:13 -0700652 parent = nd->mnt->mnt_parent;
653 if (parent == nd->mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 spin_unlock(&vfsmount_lock);
655 break;
656 }
657 mntget(parent);
Al Viro58c465e2005-06-06 13:36:13 -0700658 nd->dentry = dget(nd->mnt->mnt_mountpoint);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 spin_unlock(&vfsmount_lock);
660 dput(old);
Al Viro58c465e2005-06-06 13:36:13 -0700661 mntput(nd->mnt);
662 nd->mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 }
Al Viro58c465e2005-06-06 13:36:13 -0700664 follow_mount(&nd->mnt, &nd->dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665}
666
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667/*
668 * It's more convoluted than I'd like it to be, but... it's still fairly
669 * small and for now I'd prefer to have fast path as straight as possible.
670 * It _is_ time-critical.
671 */
672static int do_lookup(struct nameidata *nd, struct qstr *name,
673 struct path *path)
674{
675 struct vfsmount *mnt = nd->mnt;
676 struct dentry *dentry = __d_lookup(nd->dentry, name);
677
678 if (!dentry)
679 goto need_lookup;
680 if (dentry->d_op && dentry->d_op->d_revalidate)
681 goto need_revalidate;
682done:
683 path->mnt = mnt;
684 path->dentry = dentry;
Al Viro634ee702005-06-06 13:36:13 -0700685 __follow_mount(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 return 0;
687
688need_lookup:
689 dentry = real_lookup(nd->dentry, name, nd);
690 if (IS_ERR(dentry))
691 goto fail;
692 goto done;
693
694need_revalidate:
695 if (dentry->d_op->d_revalidate(dentry, nd))
696 goto done;
697 if (d_invalidate(dentry))
698 goto done;
699 dput(dentry);
700 goto need_lookup;
701
702fail:
703 return PTR_ERR(dentry);
704}
705
706/*
707 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +0100708 * This is the basic name resolution function, turning a pathname into
709 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 *
Prasanna Medaea3834d2005-04-29 16:00:17 +0100711 * Returns 0 and nd will have valid dentry and mnt on success.
712 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 */
714static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
715{
716 struct path next;
717 struct inode *inode;
718 int err;
719 unsigned int lookup_flags = nd->flags;
720
721 while (*name=='/')
722 name++;
723 if (!*name)
724 goto return_reval;
725
726 inode = nd->dentry->d_inode;
727 if (nd->depth)
728 lookup_flags = LOOKUP_FOLLOW;
729
730 /* At this point we know we have a real path component. */
731 for(;;) {
732 unsigned long hash;
733 struct qstr this;
734 unsigned int c;
735
736 err = exec_permission_lite(inode, nd);
737 if (err == -EAGAIN) {
738 err = permission(inode, MAY_EXEC, nd);
739 }
740 if (err)
741 break;
742
743 this.name = name;
744 c = *(const unsigned char *)name;
745
746 hash = init_name_hash();
747 do {
748 name++;
749 hash = partial_name_hash(c, hash);
750 c = *(const unsigned char *)name;
751 } while (c && (c != '/'));
752 this.len = name - (const char *) this.name;
753 this.hash = end_name_hash(hash);
754
755 /* remove trailing slashes? */
756 if (!c)
757 goto last_component;
758 while (*++name == '/');
759 if (!*name)
760 goto last_with_slashes;
761
762 /*
763 * "." and ".." are special - ".." especially so because it has
764 * to be able to know about the current root directory and
765 * parent relationships.
766 */
767 if (this.name[0] == '.') switch (this.len) {
768 default:
769 break;
770 case 2:
771 if (this.name[1] != '.')
772 break;
Al Viro58c465e2005-06-06 13:36:13 -0700773 follow_dotdot(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 inode = nd->dentry->d_inode;
775 /* fallthrough */
776 case 1:
777 continue;
778 }
779 /*
780 * See if the low-level filesystem might want
781 * to use its own hash..
782 */
783 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
784 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
785 if (err < 0)
786 break;
787 }
788 nd->flags |= LOOKUP_CONTINUE;
789 /* This does the actual lookups.. */
790 err = do_lookup(nd, &this, &next);
791 if (err)
792 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793
794 err = -ENOENT;
795 inode = next.dentry->d_inode;
796 if (!inode)
797 goto out_dput;
798 err = -ENOTDIR;
799 if (!inode->i_op)
800 goto out_dput;
801
802 if (inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700803 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 if (err)
805 goto return_err;
806 err = -ENOENT;
807 inode = nd->dentry->d_inode;
808 if (!inode)
809 break;
810 err = -ENOTDIR;
811 if (!inode->i_op)
812 break;
813 } else {
814 dput(nd->dentry);
Al Viro2f12dbf2005-06-06 13:36:07 -0700815 if (nd->mnt != next.mnt)
816 mntput(nd->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 nd->mnt = next.mnt;
818 nd->dentry = next.dentry;
819 }
820 err = -ENOTDIR;
821 if (!inode->i_op->lookup)
822 break;
823 continue;
824 /* here ends the main loop */
825
826last_with_slashes:
827 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
828last_component:
829 nd->flags &= ~LOOKUP_CONTINUE;
830 if (lookup_flags & LOOKUP_PARENT)
831 goto lookup_parent;
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 goto return_reval;
843 }
844 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
845 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
846 if (err < 0)
847 break;
848 }
849 err = do_lookup(nd, &this, &next);
850 if (err)
851 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 inode = next.dentry->d_inode;
853 if ((lookup_flags & LOOKUP_FOLLOW)
854 && inode && inode->i_op && inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700855 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 if (err)
857 goto return_err;
858 inode = nd->dentry->d_inode;
859 } else {
860 dput(nd->dentry);
Al Viro2f12dbf2005-06-06 13:36:07 -0700861 if (nd->mnt != next.mnt)
862 mntput(nd->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 nd->mnt = next.mnt;
864 nd->dentry = next.dentry;
865 }
866 err = -ENOENT;
867 if (!inode)
868 break;
869 if (lookup_flags & LOOKUP_DIRECTORY) {
870 err = -ENOTDIR;
871 if (!inode->i_op || !inode->i_op->lookup)
872 break;
873 }
874 goto return_base;
875lookup_parent:
876 nd->last = this;
877 nd->last_type = LAST_NORM;
878 if (this.name[0] != '.')
879 goto return_base;
880 if (this.len == 1)
881 nd->last_type = LAST_DOT;
882 else if (this.len == 2 && this.name[1] == '.')
883 nd->last_type = LAST_DOTDOT;
884 else
885 goto return_base;
886return_reval:
887 /*
888 * We bypassed the ordinary revalidation routines.
889 * We may need to check the cached dentry for staleness.
890 */
891 if (nd->dentry && nd->dentry->d_sb &&
892 (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
893 err = -ESTALE;
894 /* Note: we do not d_invalidate() */
895 if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
896 break;
897 }
898return_base:
899 return 0;
900out_dput:
901 dput(next.dentry);
Al Viro2f12dbf2005-06-06 13:36:07 -0700902 if (nd->mnt != next.mnt)
Al Virod9d29a22005-06-06 13:36:11 -0700903 mntput(next.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 break;
905 }
906 path_release(nd);
907return_err:
908 return err;
909}
910
911/*
912 * Wrapper to retry pathname resolution whenever the underlying
913 * file system returns an ESTALE.
914 *
915 * Retry the whole path once, forcing real lookup requests
916 * instead of relying on the dcache.
917 */
918int fastcall link_path_walk(const char *name, struct nameidata *nd)
919{
920 struct nameidata save = *nd;
921 int result;
922
923 /* make sure the stuff we saved doesn't go away */
924 dget(save.dentry);
925 mntget(save.mnt);
926
927 result = __link_path_walk(name, nd);
928 if (result == -ESTALE) {
929 *nd = save;
930 dget(nd->dentry);
931 mntget(nd->mnt);
932 nd->flags |= LOOKUP_REVAL;
933 result = __link_path_walk(name, nd);
934 }
935
936 dput(save.dentry);
937 mntput(save.mnt);
938
939 return result;
940}
941
942int fastcall path_walk(const char * name, struct nameidata *nd)
943{
944 current->total_link_count = 0;
945 return link_path_walk(name, nd);
946}
947
Prasanna Medaea3834d2005-04-29 16:00:17 +0100948/*
949 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
950 * everything is done. Returns 0 and drops input nd, if lookup failed;
951 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
953{
954 if (path_walk(name, nd))
955 return 0; /* something went wrong... */
956
957 if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
958 struct dentry *old_dentry = nd->dentry;
959 struct vfsmount *old_mnt = nd->mnt;
960 struct qstr last = nd->last;
961 int last_type = nd->last_type;
962 /*
963 * NAME was not found in alternate root or it's a directory. Try to find
964 * it in the normal root:
965 */
966 nd->last_type = LAST_ROOT;
967 read_lock(&current->fs->lock);
968 nd->mnt = mntget(current->fs->rootmnt);
969 nd->dentry = dget(current->fs->root);
970 read_unlock(&current->fs->lock);
971 if (path_walk(name, nd) == 0) {
972 if (nd->dentry->d_inode) {
973 dput(old_dentry);
974 mntput(old_mnt);
975 return 1;
976 }
977 path_release(nd);
978 }
979 nd->dentry = old_dentry;
980 nd->mnt = old_mnt;
981 nd->last = last;
982 nd->last_type = last_type;
983 }
984 return 1;
985}
986
987void set_fs_altroot(void)
988{
989 char *emul = __emul_prefix();
990 struct nameidata nd;
991 struct vfsmount *mnt = NULL, *oldmnt;
992 struct dentry *dentry = NULL, *olddentry;
993 int err;
994
995 if (!emul)
996 goto set_it;
997 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
998 if (!err) {
999 mnt = nd.mnt;
1000 dentry = nd.dentry;
1001 }
1002set_it:
1003 write_lock(&current->fs->lock);
1004 oldmnt = current->fs->altrootmnt;
1005 olddentry = current->fs->altroot;
1006 current->fs->altrootmnt = mnt;
1007 current->fs->altroot = dentry;
1008 write_unlock(&current->fs->lock);
1009 if (olddentry) {
1010 dput(olddentry);
1011 mntput(oldmnt);
1012 }
1013}
1014
Prasanna Medaea3834d2005-04-29 16:00:17 +01001015/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016int fastcall path_lookup(const char *name, unsigned int flags, struct nameidata *nd)
1017{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001018 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019
1020 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1021 nd->flags = flags;
1022 nd->depth = 0;
1023
1024 read_lock(&current->fs->lock);
1025 if (*name=='/') {
1026 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
1027 nd->mnt = mntget(current->fs->altrootmnt);
1028 nd->dentry = dget(current->fs->altroot);
1029 read_unlock(&current->fs->lock);
1030 if (__emul_lookup_dentry(name,nd))
Prasanna Medaea3834d2005-04-29 16:00:17 +01001031 goto out; /* found in altroot */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 read_lock(&current->fs->lock);
1033 }
1034 nd->mnt = mntget(current->fs->rootmnt);
1035 nd->dentry = dget(current->fs->root);
1036 } else {
1037 nd->mnt = mntget(current->fs->pwdmnt);
1038 nd->dentry = dget(current->fs->pwd);
1039 }
1040 read_unlock(&current->fs->lock);
1041 current->total_link_count = 0;
1042 retval = link_path_walk(name, nd);
Prasanna Medaea3834d2005-04-29 16:00:17 +01001043out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 if (unlikely(current->audit_context
1045 && nd && nd->dentry && nd->dentry->d_inode))
1046 audit_inode(name, nd->dentry->d_inode);
1047 return retval;
1048}
1049
1050/*
1051 * Restricted form of lookup. Doesn't follow links, single-component only,
1052 * needs parent already locked. Doesn't follow mounts.
1053 * SMP-safe.
1054 */
1055static struct dentry * __lookup_hash(struct qstr *name, struct dentry * base, struct nameidata *nd)
1056{
1057 struct dentry * dentry;
1058 struct inode *inode;
1059 int err;
1060
1061 inode = base->d_inode;
1062 err = permission(inode, MAY_EXEC, nd);
1063 dentry = ERR_PTR(err);
1064 if (err)
1065 goto out;
1066
1067 /*
1068 * See if the low-level filesystem might want
1069 * to use its own hash..
1070 */
1071 if (base->d_op && base->d_op->d_hash) {
1072 err = base->d_op->d_hash(base, name);
1073 dentry = ERR_PTR(err);
1074 if (err < 0)
1075 goto out;
1076 }
1077
1078 dentry = cached_lookup(base, name, nd);
1079 if (!dentry) {
1080 struct dentry *new = d_alloc(base, name);
1081 dentry = ERR_PTR(-ENOMEM);
1082 if (!new)
1083 goto out;
1084 dentry = inode->i_op->lookup(inode, new, nd);
1085 if (!dentry)
1086 dentry = new;
1087 else
1088 dput(new);
1089 }
1090out:
1091 return dentry;
1092}
1093
1094struct dentry * lookup_hash(struct qstr *name, struct dentry * base)
1095{
1096 return __lookup_hash(name, base, NULL);
1097}
1098
1099/* SMP-safe */
1100struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
1101{
1102 unsigned long hash;
1103 struct qstr this;
1104 unsigned int c;
1105
1106 this.name = name;
1107 this.len = len;
1108 if (!len)
1109 goto access;
1110
1111 hash = init_name_hash();
1112 while (len--) {
1113 c = *(const unsigned char *)name++;
1114 if (c == '/' || c == '\0')
1115 goto access;
1116 hash = partial_name_hash(c, hash);
1117 }
1118 this.hash = end_name_hash(hash);
1119
1120 return lookup_hash(&this, base);
1121access:
1122 return ERR_PTR(-EACCES);
1123}
1124
1125/*
1126 * namei()
1127 *
1128 * is used by most simple commands to get the inode of a specified name.
1129 * Open, link etc use their own routines, but this is enough for things
1130 * like 'chmod' etc.
1131 *
1132 * namei exists in two versions: namei/lnamei. The only difference is
1133 * that namei follows links, while lnamei does not.
1134 * SMP-safe
1135 */
1136int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1137{
1138 char *tmp = getname(name);
1139 int err = PTR_ERR(tmp);
1140
1141 if (!IS_ERR(tmp)) {
1142 err = path_lookup(tmp, flags, nd);
1143 putname(tmp);
1144 }
1145 return err;
1146}
1147
1148/*
1149 * It's inline, so penalty for filesystems that don't use sticky bit is
1150 * minimal.
1151 */
1152static inline int check_sticky(struct inode *dir, struct inode *inode)
1153{
1154 if (!(dir->i_mode & S_ISVTX))
1155 return 0;
1156 if (inode->i_uid == current->fsuid)
1157 return 0;
1158 if (dir->i_uid == current->fsuid)
1159 return 0;
1160 return !capable(CAP_FOWNER);
1161}
1162
1163/*
1164 * Check whether we can remove a link victim from directory dir, check
1165 * whether the type of victim is right.
1166 * 1. We can't do it if dir is read-only (done in permission())
1167 * 2. We should have write and exec permissions on dir
1168 * 3. We can't remove anything from append-only dir
1169 * 4. We can't do anything with immutable dir (done in permission())
1170 * 5. If the sticky bit on dir is set we should either
1171 * a. be owner of dir, or
1172 * b. be owner of victim, or
1173 * c. have CAP_FOWNER capability
1174 * 6. If the victim is append-only or immutable we can't do antyhing with
1175 * links pointing to it.
1176 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1177 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1178 * 9. We can't remove a root or mountpoint.
1179 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1180 * nfs_async_unlink().
1181 */
1182static inline int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1183{
1184 int error;
1185
1186 if (!victim->d_inode)
1187 return -ENOENT;
1188
1189 BUG_ON(victim->d_parent->d_inode != dir);
1190
1191 error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
1192 if (error)
1193 return error;
1194 if (IS_APPEND(dir))
1195 return -EPERM;
1196 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1197 IS_IMMUTABLE(victim->d_inode))
1198 return -EPERM;
1199 if (isdir) {
1200 if (!S_ISDIR(victim->d_inode->i_mode))
1201 return -ENOTDIR;
1202 if (IS_ROOT(victim))
1203 return -EBUSY;
1204 } else if (S_ISDIR(victim->d_inode->i_mode))
1205 return -EISDIR;
1206 if (IS_DEADDIR(dir))
1207 return -ENOENT;
1208 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1209 return -EBUSY;
1210 return 0;
1211}
1212
1213/* Check whether we can create an object with dentry child in directory
1214 * dir.
1215 * 1. We can't do it if child already exists (open has special treatment for
1216 * this case, but since we are inlined it's OK)
1217 * 2. We can't do it if dir is read-only (done in permission())
1218 * 3. We should have write and exec permissions on dir
1219 * 4. We can't do it if dir is immutable (done in permission())
1220 */
1221static inline int may_create(struct inode *dir, struct dentry *child,
1222 struct nameidata *nd)
1223{
1224 if (child->d_inode)
1225 return -EEXIST;
1226 if (IS_DEADDIR(dir))
1227 return -ENOENT;
1228 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1229}
1230
1231/*
1232 * Special case: O_CREAT|O_EXCL implies O_NOFOLLOW for security
1233 * reasons.
1234 *
1235 * O_DIRECTORY translates into forcing a directory lookup.
1236 */
1237static inline int lookup_flags(unsigned int f)
1238{
1239 unsigned long retval = LOOKUP_FOLLOW;
1240
1241 if (f & O_NOFOLLOW)
1242 retval &= ~LOOKUP_FOLLOW;
1243
1244 if ((f & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1245 retval &= ~LOOKUP_FOLLOW;
1246
1247 if (f & O_DIRECTORY)
1248 retval |= LOOKUP_DIRECTORY;
1249
1250 return retval;
1251}
1252
1253/*
1254 * p1 and p2 should be directories on the same fs.
1255 */
1256struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1257{
1258 struct dentry *p;
1259
1260 if (p1 == p2) {
1261 down(&p1->d_inode->i_sem);
1262 return NULL;
1263 }
1264
1265 down(&p1->d_inode->i_sb->s_vfs_rename_sem);
1266
1267 for (p = p1; p->d_parent != p; p = p->d_parent) {
1268 if (p->d_parent == p2) {
1269 down(&p2->d_inode->i_sem);
1270 down(&p1->d_inode->i_sem);
1271 return p;
1272 }
1273 }
1274
1275 for (p = p2; p->d_parent != p; p = p->d_parent) {
1276 if (p->d_parent == p1) {
1277 down(&p1->d_inode->i_sem);
1278 down(&p2->d_inode->i_sem);
1279 return p;
1280 }
1281 }
1282
1283 down(&p1->d_inode->i_sem);
1284 down(&p2->d_inode->i_sem);
1285 return NULL;
1286}
1287
1288void unlock_rename(struct dentry *p1, struct dentry *p2)
1289{
1290 up(&p1->d_inode->i_sem);
1291 if (p1 != p2) {
1292 up(&p2->d_inode->i_sem);
1293 up(&p1->d_inode->i_sb->s_vfs_rename_sem);
1294 }
1295}
1296
1297int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1298 struct nameidata *nd)
1299{
1300 int error = may_create(dir, dentry, nd);
1301
1302 if (error)
1303 return error;
1304
1305 if (!dir->i_op || !dir->i_op->create)
1306 return -EACCES; /* shouldn't it be ENOSYS? */
1307 mode &= S_IALLUGO;
1308 mode |= S_IFREG;
1309 error = security_inode_create(dir, dentry, mode);
1310 if (error)
1311 return error;
1312 DQUOT_INIT(dir);
1313 error = dir->i_op->create(dir, dentry, mode, nd);
1314 if (!error) {
Robert Love0eeca282005-07-12 17:06:03 -04001315 fsnotify_create(dir, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 security_inode_post_create(dir, dentry, mode);
1317 }
1318 return error;
1319}
1320
1321int may_open(struct nameidata *nd, int acc_mode, int flag)
1322{
1323 struct dentry *dentry = nd->dentry;
1324 struct inode *inode = dentry->d_inode;
1325 int error;
1326
1327 if (!inode)
1328 return -ENOENT;
1329
1330 if (S_ISLNK(inode->i_mode))
1331 return -ELOOP;
1332
1333 if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1334 return -EISDIR;
1335
1336 error = permission(inode, acc_mode, nd);
1337 if (error)
1338 return error;
1339
1340 /*
1341 * FIFO's, sockets and device files are special: they don't
1342 * actually live on the filesystem itself, and as such you
1343 * can write to them even if the filesystem is read-only.
1344 */
1345 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1346 flag &= ~O_TRUNC;
1347 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1348 if (nd->mnt->mnt_flags & MNT_NODEV)
1349 return -EACCES;
1350
1351 flag &= ~O_TRUNC;
1352 } else if (IS_RDONLY(inode) && (flag & FMODE_WRITE))
1353 return -EROFS;
1354 /*
1355 * An append-only file must be opened in append mode for writing.
1356 */
1357 if (IS_APPEND(inode)) {
1358 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1359 return -EPERM;
1360 if (flag & O_TRUNC)
1361 return -EPERM;
1362 }
1363
1364 /* O_NOATIME can only be set by the owner or superuser */
1365 if (flag & O_NOATIME)
1366 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1367 return -EPERM;
1368
1369 /*
1370 * Ensure there are no outstanding leases on the file.
1371 */
1372 error = break_lease(inode, flag);
1373 if (error)
1374 return error;
1375
1376 if (flag & O_TRUNC) {
1377 error = get_write_access(inode);
1378 if (error)
1379 return error;
1380
1381 /*
1382 * Refuse to truncate files with mandatory locks held on them.
1383 */
1384 error = locks_verify_locked(inode);
1385 if (!error) {
1386 DQUOT_INIT(inode);
1387
1388 error = do_truncate(dentry, 0);
1389 }
1390 put_write_access(inode);
1391 if (error)
1392 return error;
1393 } else
1394 if (flag & FMODE_WRITE)
1395 DQUOT_INIT(inode);
1396
1397 return 0;
1398}
1399
1400/*
1401 * open_namei()
1402 *
1403 * namei for open - this is in fact almost the whole open-routine.
1404 *
1405 * Note that the low bits of "flag" aren't the same as in the open
1406 * system call - they are 00 - no permissions needed
1407 * 01 - read permission needed
1408 * 10 - write permission needed
1409 * 11 - read/write permissions needed
1410 * which is a lot more logical, and also allows the "no perm" needed
1411 * for symlinks (where the permissions are checked later).
1412 * SMP-safe
1413 */
1414int open_namei(const char * pathname, int flag, int mode, struct nameidata *nd)
1415{
1416 int acc_mode, error = 0;
Al Viro4e7506e2005-06-06 13:36:00 -07001417 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 struct dentry *dir;
1419 int count = 0;
1420
1421 acc_mode = ACC_MODE(flag);
1422
1423 /* Allow the LSM permission hook to distinguish append
1424 access from general write access. */
1425 if (flag & O_APPEND)
1426 acc_mode |= MAY_APPEND;
1427
1428 /* Fill in the open() intent data */
1429 nd->intent.open.flags = flag;
1430 nd->intent.open.create_mode = mode;
1431
1432 /*
1433 * The simplest case - just a plain lookup.
1434 */
1435 if (!(flag & O_CREAT)) {
1436 error = path_lookup(pathname, lookup_flags(flag)|LOOKUP_OPEN, nd);
1437 if (error)
1438 return error;
1439 goto ok;
1440 }
1441
1442 /*
1443 * Create - we need to know the parent.
1444 */
1445 error = path_lookup(pathname, LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE, nd);
1446 if (error)
1447 return error;
1448
1449 /*
1450 * We have the parent and last component. First of all, check
1451 * that we are not asked to creat(2) an obvious directory - that
1452 * will not do.
1453 */
1454 error = -EISDIR;
1455 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1456 goto exit;
1457
1458 dir = nd->dentry;
1459 nd->flags &= ~LOOKUP_PARENT;
1460 down(&dir->d_inode->i_sem);
Al Viro4e7506e2005-06-06 13:36:00 -07001461 path.dentry = __lookup_hash(&nd->last, nd->dentry, nd);
Al Virod73ffe12005-06-06 13:36:01 -07001462 path.mnt = nd->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463
1464do_last:
Al Viro4e7506e2005-06-06 13:36:00 -07001465 error = PTR_ERR(path.dentry);
1466 if (IS_ERR(path.dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 up(&dir->d_inode->i_sem);
1468 goto exit;
1469 }
1470
1471 /* Negative dentry, just create the file */
Al Viro4e7506e2005-06-06 13:36:00 -07001472 if (!path.dentry->d_inode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 if (!IS_POSIXACL(dir->d_inode))
1474 mode &= ~current->fs->umask;
Al Viro4e7506e2005-06-06 13:36:00 -07001475 error = vfs_create(dir->d_inode, path.dentry, mode, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 up(&dir->d_inode->i_sem);
1477 dput(nd->dentry);
Al Viro4e7506e2005-06-06 13:36:00 -07001478 nd->dentry = path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 if (error)
1480 goto exit;
1481 /* Don't check for write permission, don't truncate */
1482 acc_mode = 0;
1483 flag &= ~O_TRUNC;
1484 goto ok;
1485 }
1486
1487 /*
1488 * It already exists.
1489 */
1490 up(&dir->d_inode->i_sem);
1491
1492 error = -EEXIST;
1493 if (flag & O_EXCL)
1494 goto exit_dput;
1495
Al Viroe13b2102005-06-06 13:36:06 -07001496 if (__follow_mount(&path)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 error = -ELOOP;
Al Viroba7a4c12005-06-06 13:36:08 -07001498 if (flag & O_NOFOLLOW)
1499 goto exit_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 }
1501 error = -ENOENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001502 if (!path.dentry->d_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 goto exit_dput;
Al Viro4e7506e2005-06-06 13:36:00 -07001504 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 goto do_link;
1506
1507 dput(nd->dentry);
Al Viro4e7506e2005-06-06 13:36:00 -07001508 nd->dentry = path.dentry;
Al Viroa15a3f62005-06-06 13:36:08 -07001509 if (nd->mnt != path.mnt)
1510 mntput(nd->mnt);
1511 nd->mnt = path.mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 error = -EISDIR;
Al Viro4e7506e2005-06-06 13:36:00 -07001513 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 goto exit;
1515ok:
1516 error = may_open(nd, acc_mode, flag);
1517 if (error)
1518 goto exit;
1519 return 0;
1520
1521exit_dput:
Al Viro4e7506e2005-06-06 13:36:00 -07001522 dput(path.dentry);
Al Viroa15a3f62005-06-06 13:36:08 -07001523 if (nd->mnt != path.mnt)
Al Viroba7a4c12005-06-06 13:36:08 -07001524 mntput(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525exit:
1526 path_release(nd);
1527 return error;
1528
1529do_link:
1530 error = -ELOOP;
1531 if (flag & O_NOFOLLOW)
1532 goto exit_dput;
1533 /*
1534 * This is subtle. Instead of calling do_follow_link() we do the
1535 * thing by hands. The reason is that this way we have zero link_count
1536 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1537 * After that we have the parent and last component, i.e.
1538 * we are in the same situation as after the first path_walk().
1539 * Well, almost - if the last component is normal we get its copy
1540 * stored in nd->last.name and we will have to putname() it when we
1541 * are done. Procfs-like symlinks just set LAST_BIND.
1542 */
1543 nd->flags |= LOOKUP_PARENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001544 error = security_inode_follow_link(path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 if (error)
1546 goto exit_dput;
Al Virocd4e91d2005-06-06 13:36:03 -07001547 error = __do_follow_link(&path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 if (error)
1549 return error;
1550 nd->flags &= ~LOOKUP_PARENT;
Al Virod671d5e2005-06-06 13:36:04 -07001551 if (nd->last_type == LAST_BIND)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 goto ok;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 error = -EISDIR;
1554 if (nd->last_type != LAST_NORM)
1555 goto exit;
1556 if (nd->last.name[nd->last.len]) {
1557 putname(nd->last.name);
1558 goto exit;
1559 }
1560 error = -ELOOP;
1561 if (count++==32) {
1562 putname(nd->last.name);
1563 goto exit;
1564 }
1565 dir = nd->dentry;
1566 down(&dir->d_inode->i_sem);
Al Viro4e7506e2005-06-06 13:36:00 -07001567 path.dentry = __lookup_hash(&nd->last, nd->dentry, nd);
Al Virod671d5e2005-06-06 13:36:04 -07001568 path.mnt = nd->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 putname(nd->last.name);
1570 goto do_last;
1571}
1572
1573/**
1574 * lookup_create - lookup a dentry, creating it if it doesn't exist
1575 * @nd: nameidata info
1576 * @is_dir: directory flag
1577 *
1578 * Simple function to lookup and return a dentry and create it
1579 * if it doesn't exist. Is SMP-safe.
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001580 *
1581 * Returns with nd->dentry->d_inode->i_sem locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 */
1583struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1584{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001585 struct dentry *dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586
1587 down(&nd->dentry->d_inode->i_sem);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001588 /*
1589 * Yucky last component or no last component at all?
1590 * (foo/., foo/.., /////)
1591 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 if (nd->last_type != LAST_NORM)
1593 goto fail;
1594 nd->flags &= ~LOOKUP_PARENT;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001595
1596 /*
1597 * Do the final lookup.
1598 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 dentry = lookup_hash(&nd->last, nd->dentry);
1600 if (IS_ERR(dentry))
1601 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001602
1603 /*
1604 * Special case - lookup gave negative, but... we had foo/bar/
1605 * From the vfs_mknod() POV we just have a negative dentry -
1606 * all is fine. Let's be bastards - you had / on the end, you've
1607 * been asking for (non-existent) directory. -ENOENT for you.
1608 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1610 goto enoent;
1611 return dentry;
1612enoent:
1613 dput(dentry);
1614 dentry = ERR_PTR(-ENOENT);
1615fail:
1616 return dentry;
1617}
Christoph Hellwigf81a0bf2005-05-19 12:26:43 -07001618EXPORT_SYMBOL_GPL(lookup_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619
1620int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1621{
1622 int error = may_create(dir, dentry, NULL);
1623
1624 if (error)
1625 return error;
1626
1627 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1628 return -EPERM;
1629
1630 if (!dir->i_op || !dir->i_op->mknod)
1631 return -EPERM;
1632
1633 error = security_inode_mknod(dir, dentry, mode, dev);
1634 if (error)
1635 return error;
1636
1637 DQUOT_INIT(dir);
1638 error = dir->i_op->mknod(dir, dentry, mode, dev);
1639 if (!error) {
Robert Love0eeca282005-07-12 17:06:03 -04001640 fsnotify_create(dir, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 security_inode_post_mknod(dir, dentry, mode, dev);
1642 }
1643 return error;
1644}
1645
1646asmlinkage long sys_mknod(const char __user * filename, int mode, unsigned dev)
1647{
1648 int error = 0;
1649 char * tmp;
1650 struct dentry * dentry;
1651 struct nameidata nd;
1652
1653 if (S_ISDIR(mode))
1654 return -EPERM;
1655 tmp = getname(filename);
1656 if (IS_ERR(tmp))
1657 return PTR_ERR(tmp);
1658
1659 error = path_lookup(tmp, LOOKUP_PARENT, &nd);
1660 if (error)
1661 goto out;
1662 dentry = lookup_create(&nd, 0);
1663 error = PTR_ERR(dentry);
1664
1665 if (!IS_POSIXACL(nd.dentry->d_inode))
1666 mode &= ~current->fs->umask;
1667 if (!IS_ERR(dentry)) {
1668 switch (mode & S_IFMT) {
1669 case 0: case S_IFREG:
1670 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1671 break;
1672 case S_IFCHR: case S_IFBLK:
1673 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,
1674 new_decode_dev(dev));
1675 break;
1676 case S_IFIFO: case S_IFSOCK:
1677 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,0);
1678 break;
1679 case S_IFDIR:
1680 error = -EPERM;
1681 break;
1682 default:
1683 error = -EINVAL;
1684 }
1685 dput(dentry);
1686 }
1687 up(&nd.dentry->d_inode->i_sem);
1688 path_release(&nd);
1689out:
1690 putname(tmp);
1691
1692 return error;
1693}
1694
1695int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1696{
1697 int error = may_create(dir, dentry, NULL);
1698
1699 if (error)
1700 return error;
1701
1702 if (!dir->i_op || !dir->i_op->mkdir)
1703 return -EPERM;
1704
1705 mode &= (S_IRWXUGO|S_ISVTX);
1706 error = security_inode_mkdir(dir, dentry, mode);
1707 if (error)
1708 return error;
1709
1710 DQUOT_INIT(dir);
1711 error = dir->i_op->mkdir(dir, dentry, mode);
1712 if (!error) {
Robert Love0eeca282005-07-12 17:06:03 -04001713 fsnotify_mkdir(dir, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 security_inode_post_mkdir(dir,dentry, mode);
1715 }
1716 return error;
1717}
1718
1719asmlinkage long sys_mkdir(const char __user * pathname, int mode)
1720{
1721 int error = 0;
1722 char * tmp;
1723
1724 tmp = getname(pathname);
1725 error = PTR_ERR(tmp);
1726 if (!IS_ERR(tmp)) {
1727 struct dentry *dentry;
1728 struct nameidata nd;
1729
1730 error = path_lookup(tmp, LOOKUP_PARENT, &nd);
1731 if (error)
1732 goto out;
1733 dentry = lookup_create(&nd, 1);
1734 error = PTR_ERR(dentry);
1735 if (!IS_ERR(dentry)) {
1736 if (!IS_POSIXACL(nd.dentry->d_inode))
1737 mode &= ~current->fs->umask;
1738 error = vfs_mkdir(nd.dentry->d_inode, dentry, mode);
1739 dput(dentry);
1740 }
1741 up(&nd.dentry->d_inode->i_sem);
1742 path_release(&nd);
1743out:
1744 putname(tmp);
1745 }
1746
1747 return error;
1748}
1749
1750/*
1751 * We try to drop the dentry early: we should have
1752 * a usage count of 2 if we're the only user of this
1753 * dentry, and if that is true (possibly after pruning
1754 * the dcache), then we drop the dentry now.
1755 *
1756 * A low-level filesystem can, if it choses, legally
1757 * do a
1758 *
1759 * if (!d_unhashed(dentry))
1760 * return -EBUSY;
1761 *
1762 * if it cannot handle the case of removing a directory
1763 * that is still in use by something else..
1764 */
1765void dentry_unhash(struct dentry *dentry)
1766{
1767 dget(dentry);
1768 if (atomic_read(&dentry->d_count))
1769 shrink_dcache_parent(dentry);
1770 spin_lock(&dcache_lock);
1771 spin_lock(&dentry->d_lock);
1772 if (atomic_read(&dentry->d_count) == 2)
1773 __d_drop(dentry);
1774 spin_unlock(&dentry->d_lock);
1775 spin_unlock(&dcache_lock);
1776}
1777
1778int vfs_rmdir(struct inode *dir, struct dentry *dentry)
1779{
1780 int error = may_delete(dir, dentry, 1);
1781
1782 if (error)
1783 return error;
1784
1785 if (!dir->i_op || !dir->i_op->rmdir)
1786 return -EPERM;
1787
1788 DQUOT_INIT(dir);
1789
1790 down(&dentry->d_inode->i_sem);
1791 dentry_unhash(dentry);
1792 if (d_mountpoint(dentry))
1793 error = -EBUSY;
1794 else {
1795 error = security_inode_rmdir(dir, dentry);
1796 if (!error) {
1797 error = dir->i_op->rmdir(dir, dentry);
1798 if (!error)
1799 dentry->d_inode->i_flags |= S_DEAD;
1800 }
1801 }
1802 up(&dentry->d_inode->i_sem);
1803 if (!error) {
Robert Love0eeca282005-07-12 17:06:03 -04001804 fsnotify_rmdir(dentry, dentry->d_inode, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 d_delete(dentry);
1806 }
1807 dput(dentry);
1808
1809 return error;
1810}
1811
1812asmlinkage long sys_rmdir(const char __user * pathname)
1813{
1814 int error = 0;
1815 char * name;
1816 struct dentry *dentry;
1817 struct nameidata nd;
1818
1819 name = getname(pathname);
1820 if(IS_ERR(name))
1821 return PTR_ERR(name);
1822
1823 error = path_lookup(name, LOOKUP_PARENT, &nd);
1824 if (error)
1825 goto exit;
1826
1827 switch(nd.last_type) {
1828 case LAST_DOTDOT:
1829 error = -ENOTEMPTY;
1830 goto exit1;
1831 case LAST_DOT:
1832 error = -EINVAL;
1833 goto exit1;
1834 case LAST_ROOT:
1835 error = -EBUSY;
1836 goto exit1;
1837 }
1838 down(&nd.dentry->d_inode->i_sem);
1839 dentry = lookup_hash(&nd.last, nd.dentry);
1840 error = PTR_ERR(dentry);
1841 if (!IS_ERR(dentry)) {
1842 error = vfs_rmdir(nd.dentry->d_inode, dentry);
1843 dput(dentry);
1844 }
1845 up(&nd.dentry->d_inode->i_sem);
1846exit1:
1847 path_release(&nd);
1848exit:
1849 putname(name);
1850 return error;
1851}
1852
1853int vfs_unlink(struct inode *dir, struct dentry *dentry)
1854{
1855 int error = may_delete(dir, dentry, 0);
1856
1857 if (error)
1858 return error;
1859
1860 if (!dir->i_op || !dir->i_op->unlink)
1861 return -EPERM;
1862
1863 DQUOT_INIT(dir);
1864
1865 down(&dentry->d_inode->i_sem);
1866 if (d_mountpoint(dentry))
1867 error = -EBUSY;
1868 else {
1869 error = security_inode_unlink(dir, dentry);
1870 if (!error)
1871 error = dir->i_op->unlink(dir, dentry);
1872 }
1873 up(&dentry->d_inode->i_sem);
1874
1875 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
1876 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Robert Love0eeca282005-07-12 17:06:03 -04001877 fsnotify_unlink(dentry, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 }
Robert Love0eeca282005-07-12 17:06:03 -04001880
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 return error;
1882}
1883
1884/*
1885 * Make sure that the actual truncation of the file will occur outside its
1886 * directory's i_sem. Truncate can take a long time if there is a lot of
1887 * writeout happening, and we don't want to prevent access to the directory
1888 * while waiting on the I/O.
1889 */
1890asmlinkage long sys_unlink(const char __user * pathname)
1891{
1892 int error = 0;
1893 char * name;
1894 struct dentry *dentry;
1895 struct nameidata nd;
1896 struct inode *inode = NULL;
1897
1898 name = getname(pathname);
1899 if(IS_ERR(name))
1900 return PTR_ERR(name);
1901
1902 error = path_lookup(name, LOOKUP_PARENT, &nd);
1903 if (error)
1904 goto exit;
1905 error = -EISDIR;
1906 if (nd.last_type != LAST_NORM)
1907 goto exit1;
1908 down(&nd.dentry->d_inode->i_sem);
1909 dentry = lookup_hash(&nd.last, nd.dentry);
1910 error = PTR_ERR(dentry);
1911 if (!IS_ERR(dentry)) {
1912 /* Why not before? Because we want correct error value */
1913 if (nd.last.name[nd.last.len])
1914 goto slashes;
1915 inode = dentry->d_inode;
1916 if (inode)
1917 atomic_inc(&inode->i_count);
1918 error = vfs_unlink(nd.dentry->d_inode, dentry);
1919 exit2:
1920 dput(dentry);
1921 }
1922 up(&nd.dentry->d_inode->i_sem);
1923 if (inode)
1924 iput(inode); /* truncate the inode here */
1925exit1:
1926 path_release(&nd);
1927exit:
1928 putname(name);
1929 return error;
1930
1931slashes:
1932 error = !dentry->d_inode ? -ENOENT :
1933 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
1934 goto exit2;
1935}
1936
1937int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname, int mode)
1938{
1939 int error = may_create(dir, dentry, NULL);
1940
1941 if (error)
1942 return error;
1943
1944 if (!dir->i_op || !dir->i_op->symlink)
1945 return -EPERM;
1946
1947 error = security_inode_symlink(dir, dentry, oldname);
1948 if (error)
1949 return error;
1950
1951 DQUOT_INIT(dir);
1952 error = dir->i_op->symlink(dir, dentry, oldname);
1953 if (!error) {
Robert Love0eeca282005-07-12 17:06:03 -04001954 fsnotify_create(dir, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 security_inode_post_symlink(dir, dentry, oldname);
1956 }
1957 return error;
1958}
1959
1960asmlinkage long sys_symlink(const char __user * oldname, const char __user * newname)
1961{
1962 int error = 0;
1963 char * from;
1964 char * to;
1965
1966 from = getname(oldname);
1967 if(IS_ERR(from))
1968 return PTR_ERR(from);
1969 to = getname(newname);
1970 error = PTR_ERR(to);
1971 if (!IS_ERR(to)) {
1972 struct dentry *dentry;
1973 struct nameidata nd;
1974
1975 error = path_lookup(to, LOOKUP_PARENT, &nd);
1976 if (error)
1977 goto out;
1978 dentry = lookup_create(&nd, 0);
1979 error = PTR_ERR(dentry);
1980 if (!IS_ERR(dentry)) {
1981 error = vfs_symlink(nd.dentry->d_inode, dentry, from, S_IALLUGO);
1982 dput(dentry);
1983 }
1984 up(&nd.dentry->d_inode->i_sem);
1985 path_release(&nd);
1986out:
1987 putname(to);
1988 }
1989 putname(from);
1990 return error;
1991}
1992
1993int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
1994{
1995 struct inode *inode = old_dentry->d_inode;
1996 int error;
1997
1998 if (!inode)
1999 return -ENOENT;
2000
2001 error = may_create(dir, new_dentry, NULL);
2002 if (error)
2003 return error;
2004
2005 if (dir->i_sb != inode->i_sb)
2006 return -EXDEV;
2007
2008 /*
2009 * A link to an append-only or immutable file cannot be created.
2010 */
2011 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2012 return -EPERM;
2013 if (!dir->i_op || !dir->i_op->link)
2014 return -EPERM;
2015 if (S_ISDIR(old_dentry->d_inode->i_mode))
2016 return -EPERM;
2017
2018 error = security_inode_link(old_dentry, dir, new_dentry);
2019 if (error)
2020 return error;
2021
2022 down(&old_dentry->d_inode->i_sem);
2023 DQUOT_INIT(dir);
2024 error = dir->i_op->link(old_dentry, dir, new_dentry);
2025 up(&old_dentry->d_inode->i_sem);
2026 if (!error) {
Robert Love0eeca282005-07-12 17:06:03 -04002027 fsnotify_create(dir, new_dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028 security_inode_post_link(old_dentry, dir, new_dentry);
2029 }
2030 return error;
2031}
2032
2033/*
2034 * Hardlinks are often used in delicate situations. We avoid
2035 * security-related surprises by not following symlinks on the
2036 * newname. --KAB
2037 *
2038 * We don't follow them on the oldname either to be compatible
2039 * with linux 2.0, and to avoid hard-linking to directories
2040 * and other special files. --ADM
2041 */
2042asmlinkage long sys_link(const char __user * oldname, const char __user * newname)
2043{
2044 struct dentry *new_dentry;
2045 struct nameidata nd, old_nd;
2046 int error;
2047 char * to;
2048
2049 to = getname(newname);
2050 if (IS_ERR(to))
2051 return PTR_ERR(to);
2052
2053 error = __user_walk(oldname, 0, &old_nd);
2054 if (error)
2055 goto exit;
2056 error = path_lookup(to, LOOKUP_PARENT, &nd);
2057 if (error)
2058 goto out;
2059 error = -EXDEV;
2060 if (old_nd.mnt != nd.mnt)
2061 goto out_release;
2062 new_dentry = lookup_create(&nd, 0);
2063 error = PTR_ERR(new_dentry);
2064 if (!IS_ERR(new_dentry)) {
2065 error = vfs_link(old_nd.dentry, nd.dentry->d_inode, new_dentry);
2066 dput(new_dentry);
2067 }
2068 up(&nd.dentry->d_inode->i_sem);
2069out_release:
2070 path_release(&nd);
2071out:
2072 path_release(&old_nd);
2073exit:
2074 putname(to);
2075
2076 return error;
2077}
2078
2079/*
2080 * The worst of all namespace operations - renaming directory. "Perverted"
2081 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2082 * Problems:
2083 * a) we can get into loop creation. Check is done in is_subdir().
2084 * b) race potential - two innocent renames can create a loop together.
2085 * That's where 4.4 screws up. Current fix: serialization on
2086 * sb->s_vfs_rename_sem. We might be more accurate, but that's another
2087 * story.
2088 * c) we have to lock _three_ objects - parents and victim (if it exists).
2089 * And that - after we got ->i_sem on parents (until then we don't know
2090 * whether the target exists). Solution: try to be smart with locking
2091 * order for inodes. We rely on the fact that tree topology may change
2092 * only under ->s_vfs_rename_sem _and_ that parent of the object we
2093 * move will be locked. Thus we can rank directories by the tree
2094 * (ancestors first) and rank all non-directories after them.
2095 * That works since everybody except rename does "lock parent, lookup,
2096 * lock child" and rename is under ->s_vfs_rename_sem.
2097 * HOWEVER, it relies on the assumption that any object with ->lookup()
2098 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2099 * we'd better make sure that there's no link(2) for them.
2100 * d) some filesystems don't support opened-but-unlinked directories,
2101 * either because of layout or because they are not ready to deal with
2102 * all cases correctly. The latter will be fixed (taking this sort of
2103 * stuff into VFS), but the former is not going away. Solution: the same
2104 * trick as in rmdir().
2105 * e) conversion from fhandle to dentry may come in the wrong moment - when
2106 * we are removing the target. Solution: we will have to grab ->i_sem
2107 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2108 * ->i_sem on parents, which works but leads to some truely excessive
2109 * locking].
2110 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07002111static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2112 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113{
2114 int error = 0;
2115 struct inode *target;
2116
2117 /*
2118 * If we are going to change the parent - check write permissions,
2119 * we'll need to flip '..'.
2120 */
2121 if (new_dir != old_dir) {
2122 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2123 if (error)
2124 return error;
2125 }
2126
2127 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2128 if (error)
2129 return error;
2130
2131 target = new_dentry->d_inode;
2132 if (target) {
2133 down(&target->i_sem);
2134 dentry_unhash(new_dentry);
2135 }
2136 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2137 error = -EBUSY;
2138 else
2139 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2140 if (target) {
2141 if (!error)
2142 target->i_flags |= S_DEAD;
2143 up(&target->i_sem);
2144 if (d_unhashed(new_dentry))
2145 d_rehash(new_dentry);
2146 dput(new_dentry);
2147 }
2148 if (!error) {
2149 d_move(old_dentry,new_dentry);
2150 security_inode_post_rename(old_dir, old_dentry,
2151 new_dir, new_dentry);
2152 }
2153 return error;
2154}
2155
Adrian Bunk75c96f82005-05-05 16:16:09 -07002156static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2157 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158{
2159 struct inode *target;
2160 int error;
2161
2162 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2163 if (error)
2164 return error;
2165
2166 dget(new_dentry);
2167 target = new_dentry->d_inode;
2168 if (target)
2169 down(&target->i_sem);
2170 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2171 error = -EBUSY;
2172 else
2173 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2174 if (!error) {
2175 /* The following d_move() should become unconditional */
2176 if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
2177 d_move(old_dentry, new_dentry);
2178 security_inode_post_rename(old_dir, old_dentry, new_dir, new_dentry);
2179 }
2180 if (target)
2181 up(&target->i_sem);
2182 dput(new_dentry);
2183 return error;
2184}
2185
2186int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2187 struct inode *new_dir, struct dentry *new_dentry)
2188{
2189 int error;
2190 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Robert Love0eeca282005-07-12 17:06:03 -04002191 const char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192
2193 if (old_dentry->d_inode == new_dentry->d_inode)
2194 return 0;
2195
2196 error = may_delete(old_dir, old_dentry, is_dir);
2197 if (error)
2198 return error;
2199
2200 if (!new_dentry->d_inode)
2201 error = may_create(new_dir, new_dentry, NULL);
2202 else
2203 error = may_delete(new_dir, new_dentry, is_dir);
2204 if (error)
2205 return error;
2206
2207 if (!old_dir->i_op || !old_dir->i_op->rename)
2208 return -EPERM;
2209
2210 DQUOT_INIT(old_dir);
2211 DQUOT_INIT(new_dir);
2212
Robert Love0eeca282005-07-12 17:06:03 -04002213 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2214
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215 if (is_dir)
2216 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2217 else
2218 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2219 if (!error) {
Robert Love0eeca282005-07-12 17:06:03 -04002220 const char *new_name = old_dentry->d_name.name;
2221 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 }
Robert Love0eeca282005-07-12 17:06:03 -04002223 fsnotify_oldname_free(old_name);
2224
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225 return error;
2226}
2227
2228static inline int do_rename(const char * oldname, const char * newname)
2229{
2230 int error = 0;
2231 struct dentry * old_dir, * new_dir;
2232 struct dentry * old_dentry, *new_dentry;
2233 struct dentry * trap;
2234 struct nameidata oldnd, newnd;
2235
2236 error = path_lookup(oldname, LOOKUP_PARENT, &oldnd);
2237 if (error)
2238 goto exit;
2239
2240 error = path_lookup(newname, LOOKUP_PARENT, &newnd);
2241 if (error)
2242 goto exit1;
2243
2244 error = -EXDEV;
2245 if (oldnd.mnt != newnd.mnt)
2246 goto exit2;
2247
2248 old_dir = oldnd.dentry;
2249 error = -EBUSY;
2250 if (oldnd.last_type != LAST_NORM)
2251 goto exit2;
2252
2253 new_dir = newnd.dentry;
2254 if (newnd.last_type != LAST_NORM)
2255 goto exit2;
2256
2257 trap = lock_rename(new_dir, old_dir);
2258
2259 old_dentry = lookup_hash(&oldnd.last, old_dir);
2260 error = PTR_ERR(old_dentry);
2261 if (IS_ERR(old_dentry))
2262 goto exit3;
2263 /* source must exist */
2264 error = -ENOENT;
2265 if (!old_dentry->d_inode)
2266 goto exit4;
2267 /* unless the source is a directory trailing slashes give -ENOTDIR */
2268 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2269 error = -ENOTDIR;
2270 if (oldnd.last.name[oldnd.last.len])
2271 goto exit4;
2272 if (newnd.last.name[newnd.last.len])
2273 goto exit4;
2274 }
2275 /* source should not be ancestor of target */
2276 error = -EINVAL;
2277 if (old_dentry == trap)
2278 goto exit4;
2279 new_dentry = lookup_hash(&newnd.last, new_dir);
2280 error = PTR_ERR(new_dentry);
2281 if (IS_ERR(new_dentry))
2282 goto exit4;
2283 /* target should not be an ancestor of source */
2284 error = -ENOTEMPTY;
2285 if (new_dentry == trap)
2286 goto exit5;
2287
2288 error = vfs_rename(old_dir->d_inode, old_dentry,
2289 new_dir->d_inode, new_dentry);
2290exit5:
2291 dput(new_dentry);
2292exit4:
2293 dput(old_dentry);
2294exit3:
2295 unlock_rename(new_dir, old_dir);
2296exit2:
2297 path_release(&newnd);
2298exit1:
2299 path_release(&oldnd);
2300exit:
2301 return error;
2302}
2303
2304asmlinkage long sys_rename(const char __user * oldname, const char __user * newname)
2305{
2306 int error;
2307 char * from;
2308 char * to;
2309
2310 from = getname(oldname);
2311 if(IS_ERR(from))
2312 return PTR_ERR(from);
2313 to = getname(newname);
2314 error = PTR_ERR(to);
2315 if (!IS_ERR(to)) {
2316 error = do_rename(from,to);
2317 putname(to);
2318 }
2319 putname(from);
2320 return error;
2321}
2322
2323int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2324{
2325 int len;
2326
2327 len = PTR_ERR(link);
2328 if (IS_ERR(link))
2329 goto out;
2330
2331 len = strlen(link);
2332 if (len > (unsigned) buflen)
2333 len = buflen;
2334 if (copy_to_user(buffer, link, len))
2335 len = -EFAULT;
2336out:
2337 return len;
2338}
2339
2340/*
2341 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2342 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2343 * using) it for any given inode is up to filesystem.
2344 */
2345int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2346{
2347 struct nameidata nd;
2348 int res;
2349 nd.depth = 0;
2350 res = dentry->d_inode->i_op->follow_link(dentry, &nd);
2351 if (!res) {
2352 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2353 if (dentry->d_inode->i_op->put_link)
2354 dentry->d_inode->i_op->put_link(dentry, &nd);
2355 }
2356 return res;
2357}
2358
2359int vfs_follow_link(struct nameidata *nd, const char *link)
2360{
2361 return __vfs_follow_link(nd, link);
2362}
2363
2364/* get the link contents into pagecache */
2365static char *page_getlink(struct dentry * dentry, struct page **ppage)
2366{
2367 struct page * page;
2368 struct address_space *mapping = dentry->d_inode->i_mapping;
2369 page = read_cache_page(mapping, 0, (filler_t *)mapping->a_ops->readpage,
2370 NULL);
2371 if (IS_ERR(page))
2372 goto sync_fail;
2373 wait_on_page_locked(page);
2374 if (!PageUptodate(page))
2375 goto async_fail;
2376 *ppage = page;
2377 return kmap(page);
2378
2379async_fail:
2380 page_cache_release(page);
2381 return ERR_PTR(-EIO);
2382
2383sync_fail:
2384 return (char*)page;
2385}
2386
2387int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2388{
2389 struct page *page = NULL;
2390 char *s = page_getlink(dentry, &page);
2391 int res = vfs_readlink(dentry,buffer,buflen,s);
2392 if (page) {
2393 kunmap(page);
2394 page_cache_release(page);
2395 }
2396 return res;
2397}
2398
2399int page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2400{
2401 struct page *page;
2402 nd_set_link(nd, page_getlink(dentry, &page));
2403 return 0;
2404}
2405
2406void page_put_link(struct dentry *dentry, struct nameidata *nd)
2407{
2408 if (!IS_ERR(nd_get_link(nd))) {
2409 struct page *page;
2410 page = find_get_page(dentry->d_inode->i_mapping, 0);
2411 if (!page)
2412 BUG();
2413 kunmap(page);
2414 page_cache_release(page);
2415 page_cache_release(page);
2416 }
2417}
2418
2419int page_symlink(struct inode *inode, const char *symname, int len)
2420{
2421 struct address_space *mapping = inode->i_mapping;
2422 struct page *page = grab_cache_page(mapping, 0);
2423 int err = -ENOMEM;
2424 char *kaddr;
2425
2426 if (!page)
2427 goto fail;
2428 err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
2429 if (err)
2430 goto fail_map;
2431 kaddr = kmap_atomic(page, KM_USER0);
2432 memcpy(kaddr, symname, len-1);
2433 kunmap_atomic(kaddr, KM_USER0);
2434 mapping->a_ops->commit_write(NULL, page, 0, len-1);
2435 /*
2436 * Notice that we are _not_ going to block here - end of page is
2437 * unmapped, so this will only try to map the rest of page, see
2438 * that it is unmapped (typically even will not look into inode -
2439 * ->i_size will be enough for everything) and zero it out.
2440 * OTOH it's obviously correct and should make the page up-to-date.
2441 */
2442 if (!PageUptodate(page)) {
2443 err = mapping->a_ops->readpage(NULL, page);
2444 wait_on_page_locked(page);
2445 } else {
2446 unlock_page(page);
2447 }
2448 page_cache_release(page);
2449 if (err < 0)
2450 goto fail;
2451 mark_inode_dirty(inode);
2452 return 0;
2453fail_map:
2454 unlock_page(page);
2455 page_cache_release(page);
2456fail:
2457 return err;
2458}
2459
2460struct inode_operations page_symlink_inode_operations = {
2461 .readlink = generic_readlink,
2462 .follow_link = page_follow_link_light,
2463 .put_link = page_put_link,
2464};
2465
2466EXPORT_SYMBOL(__user_walk);
2467EXPORT_SYMBOL(follow_down);
2468EXPORT_SYMBOL(follow_up);
2469EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2470EXPORT_SYMBOL(getname);
2471EXPORT_SYMBOL(lock_rename);
2472EXPORT_SYMBOL(lookup_hash);
2473EXPORT_SYMBOL(lookup_one_len);
2474EXPORT_SYMBOL(page_follow_link_light);
2475EXPORT_SYMBOL(page_put_link);
2476EXPORT_SYMBOL(page_readlink);
2477EXPORT_SYMBOL(page_symlink);
2478EXPORT_SYMBOL(page_symlink_inode_operations);
2479EXPORT_SYMBOL(path_lookup);
2480EXPORT_SYMBOL(path_release);
2481EXPORT_SYMBOL(path_walk);
2482EXPORT_SYMBOL(permission);
2483EXPORT_SYMBOL(unlock_rename);
2484EXPORT_SYMBOL(vfs_create);
2485EXPORT_SYMBOL(vfs_follow_link);
2486EXPORT_SYMBOL(vfs_link);
2487EXPORT_SYMBOL(vfs_mkdir);
2488EXPORT_SYMBOL(vfs_mknod);
2489EXPORT_SYMBOL(generic_permission);
2490EXPORT_SYMBOL(vfs_readlink);
2491EXPORT_SYMBOL(vfs_rename);
2492EXPORT_SYMBOL(vfs_rmdir);
2493EXPORT_SYMBOL(vfs_symlink);
2494EXPORT_SYMBOL(vfs_unlink);
2495EXPORT_SYMBOL(dentry_unhash);
2496EXPORT_SYMBOL(generic_readlink);