[PATCH] drivers/block/pktcdvd.c: remove write-only variable in pkt_iosched_process_qu...
[linux-block.git] / fs / namei.c
CommitLineData
1da177e4
LT
1/*
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>
0eeca283 24#include <linux/fsnotify.h>
1da177e4
LT
25#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>
834f2a4a 31#include <linux/file.h>
1da177e4
LT
32#include <asm/namei.h>
33#include <asm/uaccess.h>
34
35#define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
36
37/* [Feb-1997 T. Schoebel-Theuer]
38 * Fundamental changes in the pathname lookup mechanisms (namei)
39 * were necessary because of omirr. The reason is that omirr needs
40 * to know the _real_ pathname, not the user-supplied one, in case
41 * of symlinks (and also when transname replacements occur).
42 *
43 * The new code replaces the old recursive symlink resolution with
44 * an iterative one (in case of non-nested symlink chains). It does
45 * this with calls to <fs>_follow_link().
46 * As a side effect, dir_namei(), _namei() and follow_link() are now
47 * replaced with a single function lookup_dentry() that can handle all
48 * the special cases of the former code.
49 *
50 * With the new dcache, the pathname is stored at each inode, at least as
51 * long as the refcount of the inode is positive. As a side effect, the
52 * size of the dcache depends on the inode cache and thus is dynamic.
53 *
54 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
55 * resolution to correspond with current state of the code.
56 *
57 * Note that the symlink resolution is not *completely* iterative.
58 * There is still a significant amount of tail- and mid- recursion in
59 * the algorithm. Also, note that <fs>_readlink() is not used in
60 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
61 * may return different results than <fs>_follow_link(). Many virtual
62 * filesystems (including /proc) exhibit this behavior.
63 */
64
65/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
66 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
67 * and the name already exists in form of a symlink, try to create the new
68 * name indicated by the symlink. The old code always complained that the
69 * name already exists, due to not following the symlink even if its target
70 * is nonexistent. The new semantics affects also mknod() and link() when
71 * the name is a symlink pointing to a non-existant name.
72 *
73 * I don't know which semantics is the right one, since I have no access
74 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
75 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
76 * "old" one. Personally, I think the new semantics is much more logical.
77 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
78 * file does succeed in both HP-UX and SunOs, but not in Solaris
79 * and in the old Linux semantics.
80 */
81
82/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
83 * semantics. See the comments in "open_namei" and "do_link" below.
84 *
85 * [10-Sep-98 Alan Modra] Another symlink change.
86 */
87
88/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
89 * inside the path - always follow.
90 * in the last component in creation/removal/renaming - never follow.
91 * if LOOKUP_FOLLOW passed - follow.
92 * if the pathname has trailing slashes - follow.
93 * otherwise - don't follow.
94 * (applied in that order).
95 *
96 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
97 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
98 * During the 2.4 we need to fix the userland stuff depending on it -
99 * hopefully we will be able to get rid of that wart in 2.5. So far only
100 * XEmacs seems to be relying on it...
101 */
102/*
103 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
104 * implemented. Let's see if raised priority of ->s_vfs_rename_sem gives
105 * any extra contention...
106 */
107
108/* In order to reduce some races, while at the same time doing additional
109 * checking and hopefully speeding things up, we copy filenames to the
110 * kernel data space before using them..
111 *
112 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
113 * PATH_MAX includes the nul terminator --RR.
114 */
115static inline int do_getname(const char __user *filename, char *page)
116{
117 int retval;
118 unsigned long len = PATH_MAX;
119
120 if (!segment_eq(get_fs(), KERNEL_DS)) {
121 if ((unsigned long) filename >= TASK_SIZE)
122 return -EFAULT;
123 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
124 len = TASK_SIZE - (unsigned long) filename;
125 }
126
127 retval = strncpy_from_user(page, filename, len);
128 if (retval > 0) {
129 if (retval < len)
130 return 0;
131 return -ENAMETOOLONG;
132 } else if (!retval)
133 retval = -ENOENT;
134 return retval;
135}
136
137char * getname(const char __user * filename)
138{
139 char *tmp, *result;
140
141 result = ERR_PTR(-ENOMEM);
142 tmp = __getname();
143 if (tmp) {
144 int retval = do_getname(filename, tmp);
145
146 result = tmp;
147 if (retval < 0) {
148 __putname(tmp);
149 result = ERR_PTR(retval);
150 }
151 }
152 audit_getname(result);
153 return result;
154}
155
156#ifdef CONFIG_AUDITSYSCALL
157void putname(const char *name)
158{
159 if (unlikely(current->audit_context))
160 audit_putname(name);
161 else
162 __putname(name);
163}
164EXPORT_SYMBOL(putname);
165#endif
166
167
168/**
169 * generic_permission - check for access rights on a Posix-like filesystem
170 * @inode: inode to check access rights for
171 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
172 * @check_acl: optional callback to check for Posix ACLs
173 *
174 * Used to check for read/write/execute permissions on a file.
175 * We use "fsuid" for this, letting us set arbitrary permissions
176 * for filesystem access without changing the "normal" uids which
177 * are used for other things..
178 */
179int generic_permission(struct inode *inode, int mask,
180 int (*check_acl)(struct inode *inode, int mask))
181{
182 umode_t mode = inode->i_mode;
183
184 if (current->fsuid == inode->i_uid)
185 mode >>= 6;
186 else {
187 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
188 int error = check_acl(inode, mask);
189 if (error == -EACCES)
190 goto check_capabilities;
191 else if (error != -EAGAIN)
192 return error;
193 }
194
195 if (in_group_p(inode->i_gid))
196 mode >>= 3;
197 }
198
199 /*
200 * If the DACs are ok we don't need any capability check.
201 */
202 if (((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask))
203 return 0;
204
205 check_capabilities:
206 /*
207 * Read/write DACs are always overridable.
208 * Executable DACs are overridable if at least one exec bit is set.
209 */
210 if (!(mask & MAY_EXEC) ||
211 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
212 if (capable(CAP_DAC_OVERRIDE))
213 return 0;
214
215 /*
216 * Searching includes executable on directories, else just read.
217 */
218 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
219 if (capable(CAP_DAC_READ_SEARCH))
220 return 0;
221
222 return -EACCES;
223}
224
225int permission(struct inode *inode, int mask, struct nameidata *nd)
226{
227 int retval, submask;
228
229 if (mask & MAY_WRITE) {
230 umode_t mode = inode->i_mode;
231
232 /*
233 * Nobody gets write access to a read-only fs.
234 */
235 if (IS_RDONLY(inode) &&
236 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
237 return -EROFS;
238
239 /*
240 * Nobody gets write access to an immutable file.
241 */
242 if (IS_IMMUTABLE(inode))
243 return -EACCES;
244 }
245
246
247 /* Ordinary permission routines do not understand MAY_APPEND. */
248 submask = mask & ~MAY_APPEND;
249 if (inode->i_op && inode->i_op->permission)
250 retval = inode->i_op->permission(inode, submask, nd);
251 else
252 retval = generic_permission(inode, submask, NULL);
253 if (retval)
254 return retval;
255
256 return security_inode_permission(inode, mask, nd);
257}
258
259/*
260 * get_write_access() gets write permission for a file.
261 * put_write_access() releases this write permission.
262 * This is used for regular files.
263 * We cannot support write (and maybe mmap read-write shared) accesses and
264 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
265 * can have the following values:
266 * 0: no writers, no VM_DENYWRITE mappings
267 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
268 * > 0: (i_writecount) users are writing to the file.
269 *
270 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
271 * except for the cases where we don't hold i_writecount yet. Then we need to
272 * use {get,deny}_write_access() - these functions check the sign and refuse
273 * to do the change if sign is wrong. Exclusion between them is provided by
274 * the inode->i_lock spinlock.
275 */
276
277int get_write_access(struct inode * inode)
278{
279 spin_lock(&inode->i_lock);
280 if (atomic_read(&inode->i_writecount) < 0) {
281 spin_unlock(&inode->i_lock);
282 return -ETXTBSY;
283 }
284 atomic_inc(&inode->i_writecount);
285 spin_unlock(&inode->i_lock);
286
287 return 0;
288}
289
290int deny_write_access(struct file * file)
291{
292 struct inode *inode = file->f_dentry->d_inode;
293
294 spin_lock(&inode->i_lock);
295 if (atomic_read(&inode->i_writecount) > 0) {
296 spin_unlock(&inode->i_lock);
297 return -ETXTBSY;
298 }
299 atomic_dec(&inode->i_writecount);
300 spin_unlock(&inode->i_lock);
301
302 return 0;
303}
304
305void path_release(struct nameidata *nd)
306{
307 dput(nd->dentry);
308 mntput(nd->mnt);
309}
310
311/*
312 * umount() mustn't call path_release()/mntput() as that would clear
313 * mnt_expiry_mark
314 */
315void path_release_on_umount(struct nameidata *nd)
316{
317 dput(nd->dentry);
751c404b 318 mntput_no_expire(nd->mnt);
1da177e4
LT
319}
320
834f2a4a
TM
321/**
322 * release_open_intent - free up open intent resources
323 * @nd: pointer to nameidata
324 */
325void release_open_intent(struct nameidata *nd)
326{
327 if (nd->intent.open.file->f_dentry == NULL)
328 put_filp(nd->intent.open.file);
329 else
330 fput(nd->intent.open.file);
331}
332
1da177e4
LT
333/*
334 * Internal lookup() using the new generic dcache.
335 * SMP-safe
336 */
337static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
338{
339 struct dentry * dentry = __d_lookup(parent, name);
340
341 /* lockess __d_lookup may fail due to concurrent d_move()
342 * in some unrelated directory, so try with d_lookup
343 */
344 if (!dentry)
345 dentry = d_lookup(parent, name);
346
347 if (dentry && dentry->d_op && dentry->d_op->d_revalidate) {
348 if (!dentry->d_op->d_revalidate(dentry, nd) && !d_invalidate(dentry)) {
349 dput(dentry);
350 dentry = NULL;
351 }
352 }
353 return dentry;
354}
355
356/*
357 * Short-cut version of permission(), for calling by
358 * path_walk(), when dcache lock is held. Combines parts
359 * of permission() and generic_permission(), and tests ONLY for
360 * MAY_EXEC permission.
361 *
362 * If appropriate, check DAC only. If not appropriate, or
363 * short-cut DAC fails, then call permission() to do more
364 * complete permission check.
365 */
366static inline int exec_permission_lite(struct inode *inode,
367 struct nameidata *nd)
368{
369 umode_t mode = inode->i_mode;
370
371 if (inode->i_op && inode->i_op->permission)
372 return -EAGAIN;
373
374 if (current->fsuid == inode->i_uid)
375 mode >>= 6;
376 else if (in_group_p(inode->i_gid))
377 mode >>= 3;
378
379 if (mode & MAY_EXEC)
380 goto ok;
381
382 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
383 goto ok;
384
385 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
386 goto ok;
387
388 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
389 goto ok;
390
391 return -EACCES;
392ok:
393 return security_inode_permission(inode, MAY_EXEC, nd);
394}
395
396/*
397 * This is called when everything else fails, and we actually have
398 * to go to the low-level filesystem to find out what we should do..
399 *
400 * We get the directory semaphore, and after getting that we also
401 * make sure that nobody added the entry to the dcache in the meantime..
402 * SMP-safe
403 */
404static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
405{
406 struct dentry * result;
407 struct inode *dir = parent->d_inode;
408
409 down(&dir->i_sem);
410 /*
411 * First re-do the cached lookup just in case it was created
412 * while we waited for the directory semaphore..
413 *
414 * FIXME! This could use version numbering or similar to
415 * avoid unnecessary cache lookups.
416 *
417 * The "dcache_lock" is purely to protect the RCU list walker
418 * from concurrent renames at this point (we mustn't get false
419 * negatives from the RCU list walk here, unlike the optimistic
420 * fast walk).
421 *
422 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
423 */
424 result = d_lookup(parent, name);
425 if (!result) {
426 struct dentry * dentry = d_alloc(parent, name);
427 result = ERR_PTR(-ENOMEM);
428 if (dentry) {
429 result = dir->i_op->lookup(dir, dentry, nd);
430 if (result)
431 dput(dentry);
432 else
433 result = dentry;
434 }
435 up(&dir->i_sem);
436 return result;
437 }
438
439 /*
440 * Uhhuh! Nasty case: the cache was re-populated while
441 * we waited on the semaphore. Need to revalidate.
442 */
443 up(&dir->i_sem);
444 if (result->d_op && result->d_op->d_revalidate) {
445 if (!result->d_op->d_revalidate(result, nd) && !d_invalidate(result)) {
446 dput(result);
447 result = ERR_PTR(-ENOENT);
448 }
449 }
450 return result;
451}
452
453static int __emul_lookup_dentry(const char *, struct nameidata *);
454
455/* SMP-safe */
456static inline int
457walk_init_root(const char *name, struct nameidata *nd)
458{
459 read_lock(&current->fs->lock);
460 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
461 nd->mnt = mntget(current->fs->altrootmnt);
462 nd->dentry = dget(current->fs->altroot);
463 read_unlock(&current->fs->lock);
464 if (__emul_lookup_dentry(name,nd))
465 return 0;
466 read_lock(&current->fs->lock);
467 }
468 nd->mnt = mntget(current->fs->rootmnt);
469 nd->dentry = dget(current->fs->root);
470 read_unlock(&current->fs->lock);
471 return 1;
472}
473
474static inline int __vfs_follow_link(struct nameidata *nd, const char *link)
475{
476 int res = 0;
477 char *name;
478 if (IS_ERR(link))
479 goto fail;
480
481 if (*link == '/') {
482 path_release(nd);
483 if (!walk_init_root(link, nd))
484 /* weird __emul_prefix() stuff did it */
485 goto out;
486 }
487 res = link_path_walk(link, nd);
488out:
489 if (nd->depth || res || nd->last_type!=LAST_NORM)
490 return res;
491 /*
492 * If it is an iterative symlinks resolution in open_namei() we
493 * have to copy the last component. And all that crap because of
494 * bloody create() on broken symlinks. Furrfu...
495 */
496 name = __getname();
497 if (unlikely(!name)) {
498 path_release(nd);
499 return -ENOMEM;
500 }
501 strcpy(name, nd->last.name);
502 nd->last.name = name;
503 return 0;
504fail:
505 path_release(nd);
506 return PTR_ERR(link);
507}
508
90ebe565
AV
509struct path {
510 struct vfsmount *mnt;
511 struct dentry *dentry;
512};
513
cd4e91d3 514static inline int __do_follow_link(struct path *path, struct nameidata *nd)
1da177e4
LT
515{
516 int error;
cc314eef 517 void *cookie;
cd4e91d3 518 struct dentry *dentry = path->dentry;
1da177e4 519
d671a1cb 520 touch_atime(path->mnt, dentry);
1da177e4 521 nd_set_link(nd, NULL);
cd4e91d3 522
39ca6d49
AV
523 if (path->mnt == nd->mnt)
524 mntget(path->mnt);
cc314eef
LT
525 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
526 error = PTR_ERR(cookie);
527 if (!IS_ERR(cookie)) {
1da177e4 528 char *s = nd_get_link(nd);
cc314eef 529 error = 0;
1da177e4
LT
530 if (s)
531 error = __vfs_follow_link(nd, s);
532 if (dentry->d_inode->i_op->put_link)
cc314eef 533 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
1da177e4 534 }
cd4e91d3
AV
535 dput(dentry);
536 mntput(path->mnt);
1da177e4
LT
537
538 return error;
539}
540
09dd17d3
MS
541static inline void dput_path(struct path *path, struct nameidata *nd)
542{
543 dput(path->dentry);
544 if (path->mnt != nd->mnt)
545 mntput(path->mnt);
546}
547
548static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
549{
550 dput(nd->dentry);
551 if (nd->mnt != path->mnt)
552 mntput(nd->mnt);
553 nd->mnt = path->mnt;
554 nd->dentry = path->dentry;
555}
556
1da177e4
LT
557/*
558 * This limits recursive symlink follows to 8, while
559 * limiting consecutive symlinks to 40.
560 *
561 * Without that kind of total limit, nasty chains of consecutive
562 * symlinks can cause almost arbitrarily long lookups.
563 */
90ebe565 564static inline int do_follow_link(struct path *path, struct nameidata *nd)
1da177e4
LT
565{
566 int err = -ELOOP;
567 if (current->link_count >= MAX_NESTED_LINKS)
568 goto loop;
569 if (current->total_link_count >= 40)
570 goto loop;
571 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
572 cond_resched();
90ebe565 573 err = security_inode_follow_link(path->dentry, nd);
1da177e4
LT
574 if (err)
575 goto loop;
576 current->link_count++;
577 current->total_link_count++;
578 nd->depth++;
cd4e91d3 579 err = __do_follow_link(path, nd);
839d9f93
AV
580 current->link_count--;
581 nd->depth--;
1da177e4
LT
582 return err;
583loop:
09dd17d3 584 dput_path(path, nd);
839d9f93 585 path_release(nd);
1da177e4
LT
586 return err;
587}
588
589int follow_up(struct vfsmount **mnt, struct dentry **dentry)
590{
591 struct vfsmount *parent;
592 struct dentry *mountpoint;
593 spin_lock(&vfsmount_lock);
594 parent=(*mnt)->mnt_parent;
595 if (parent == *mnt) {
596 spin_unlock(&vfsmount_lock);
597 return 0;
598 }
599 mntget(parent);
600 mountpoint=dget((*mnt)->mnt_mountpoint);
601 spin_unlock(&vfsmount_lock);
602 dput(*dentry);
603 *dentry = mountpoint;
604 mntput(*mnt);
605 *mnt = parent;
606 return 1;
607}
608
609/* no need for dcache_lock, as serialization is taken care in
610 * namespace.c
611 */
463ffb2e
AV
612static int __follow_mount(struct path *path)
613{
614 int res = 0;
615 while (d_mountpoint(path->dentry)) {
616 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
617 if (!mounted)
618 break;
619 dput(path->dentry);
620 if (res)
621 mntput(path->mnt);
622 path->mnt = mounted;
623 path->dentry = dget(mounted->mnt_root);
624 res = 1;
625 }
626 return res;
627}
628
58c465eb 629static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
1da177e4 630{
1da177e4
LT
631 while (d_mountpoint(*dentry)) {
632 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
633 if (!mounted)
634 break;
58c465eb 635 dput(*dentry);
1da177e4
LT
636 mntput(*mnt);
637 *mnt = mounted;
1da177e4 638 *dentry = dget(mounted->mnt_root);
1da177e4 639 }
1da177e4
LT
640}
641
642/* no need for dcache_lock, as serialization is taken care in
643 * namespace.c
644 */
e13b210f 645int follow_down(struct vfsmount **mnt, struct dentry **dentry)
1da177e4
LT
646{
647 struct vfsmount *mounted;
648
649 mounted = lookup_mnt(*mnt, *dentry);
650 if (mounted) {
e13b210f 651 dput(*dentry);
1da177e4
LT
652 mntput(*mnt);
653 *mnt = mounted;
1da177e4
LT
654 *dentry = dget(mounted->mnt_root);
655 return 1;
656 }
657 return 0;
658}
659
58c465eb 660static inline void follow_dotdot(struct nameidata *nd)
1da177e4
LT
661{
662 while(1) {
663 struct vfsmount *parent;
58c465eb 664 struct dentry *old = nd->dentry;
1da177e4
LT
665
666 read_lock(&current->fs->lock);
58c465eb
AV
667 if (nd->dentry == current->fs->root &&
668 nd->mnt == current->fs->rootmnt) {
1da177e4
LT
669 read_unlock(&current->fs->lock);
670 break;
671 }
672 read_unlock(&current->fs->lock);
673 spin_lock(&dcache_lock);
58c465eb
AV
674 if (nd->dentry != nd->mnt->mnt_root) {
675 nd->dentry = dget(nd->dentry->d_parent);
1da177e4
LT
676 spin_unlock(&dcache_lock);
677 dput(old);
678 break;
679 }
680 spin_unlock(&dcache_lock);
681 spin_lock(&vfsmount_lock);
58c465eb
AV
682 parent = nd->mnt->mnt_parent;
683 if (parent == nd->mnt) {
1da177e4
LT
684 spin_unlock(&vfsmount_lock);
685 break;
686 }
687 mntget(parent);
58c465eb 688 nd->dentry = dget(nd->mnt->mnt_mountpoint);
1da177e4
LT
689 spin_unlock(&vfsmount_lock);
690 dput(old);
58c465eb
AV
691 mntput(nd->mnt);
692 nd->mnt = parent;
1da177e4 693 }
58c465eb 694 follow_mount(&nd->mnt, &nd->dentry);
1da177e4
LT
695}
696
1da177e4
LT
697/*
698 * It's more convoluted than I'd like it to be, but... it's still fairly
699 * small and for now I'd prefer to have fast path as straight as possible.
700 * It _is_ time-critical.
701 */
702static int do_lookup(struct nameidata *nd, struct qstr *name,
703 struct path *path)
704{
705 struct vfsmount *mnt = nd->mnt;
706 struct dentry *dentry = __d_lookup(nd->dentry, name);
707
708 if (!dentry)
709 goto need_lookup;
710 if (dentry->d_op && dentry->d_op->d_revalidate)
711 goto need_revalidate;
712done:
713 path->mnt = mnt;
714 path->dentry = dentry;
634ee701 715 __follow_mount(path);
1da177e4
LT
716 return 0;
717
718need_lookup:
719 dentry = real_lookup(nd->dentry, name, nd);
720 if (IS_ERR(dentry))
721 goto fail;
722 goto done;
723
724need_revalidate:
725 if (dentry->d_op->d_revalidate(dentry, nd))
726 goto done;
727 if (d_invalidate(dentry))
728 goto done;
729 dput(dentry);
730 goto need_lookup;
731
732fail:
733 return PTR_ERR(dentry);
734}
735
736/*
737 * Name resolution.
ea3834d9
PM
738 * This is the basic name resolution function, turning a pathname into
739 * the final dentry. We expect 'base' to be positive and a directory.
1da177e4 740 *
ea3834d9
PM
741 * Returns 0 and nd will have valid dentry and mnt on success.
742 * Returns error and drops reference to input namei data on failure.
1da177e4
LT
743 */
744static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
745{
746 struct path next;
747 struct inode *inode;
748 int err;
749 unsigned int lookup_flags = nd->flags;
750
751 while (*name=='/')
752 name++;
753 if (!*name)
754 goto return_reval;
755
756 inode = nd->dentry->d_inode;
757 if (nd->depth)
758 lookup_flags = LOOKUP_FOLLOW;
759
760 /* At this point we know we have a real path component. */
761 for(;;) {
762 unsigned long hash;
763 struct qstr this;
764 unsigned int c;
765
cdce5d6b 766 nd->flags |= LOOKUP_CONTINUE;
1da177e4
LT
767 err = exec_permission_lite(inode, nd);
768 if (err == -EAGAIN) {
769 err = permission(inode, MAY_EXEC, nd);
770 }
771 if (err)
772 break;
773
774 this.name = name;
775 c = *(const unsigned char *)name;
776
777 hash = init_name_hash();
778 do {
779 name++;
780 hash = partial_name_hash(c, hash);
781 c = *(const unsigned char *)name;
782 } while (c && (c != '/'));
783 this.len = name - (const char *) this.name;
784 this.hash = end_name_hash(hash);
785
786 /* remove trailing slashes? */
787 if (!c)
788 goto last_component;
789 while (*++name == '/');
790 if (!*name)
791 goto last_with_slashes;
792
793 /*
794 * "." and ".." are special - ".." especially so because it has
795 * to be able to know about the current root directory and
796 * parent relationships.
797 */
798 if (this.name[0] == '.') switch (this.len) {
799 default:
800 break;
801 case 2:
802 if (this.name[1] != '.')
803 break;
58c465eb 804 follow_dotdot(nd);
1da177e4
LT
805 inode = nd->dentry->d_inode;
806 /* fallthrough */
807 case 1:
808 continue;
809 }
810 /*
811 * See if the low-level filesystem might want
812 * to use its own hash..
813 */
814 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
815 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
816 if (err < 0)
817 break;
818 }
1da177e4
LT
819 /* This does the actual lookups.. */
820 err = do_lookup(nd, &this, &next);
821 if (err)
822 break;
1da177e4
LT
823
824 err = -ENOENT;
825 inode = next.dentry->d_inode;
826 if (!inode)
827 goto out_dput;
828 err = -ENOTDIR;
829 if (!inode->i_op)
830 goto out_dput;
831
832 if (inode->i_op->follow_link) {
90ebe565 833 err = do_follow_link(&next, nd);
1da177e4
LT
834 if (err)
835 goto return_err;
836 err = -ENOENT;
837 inode = nd->dentry->d_inode;
838 if (!inode)
839 break;
840 err = -ENOTDIR;
841 if (!inode->i_op)
842 break;
09dd17d3
MS
843 } else
844 path_to_nameidata(&next, nd);
1da177e4
LT
845 err = -ENOTDIR;
846 if (!inode->i_op->lookup)
847 break;
848 continue;
849 /* here ends the main loop */
850
851last_with_slashes:
852 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
853last_component:
854 nd->flags &= ~LOOKUP_CONTINUE;
855 if (lookup_flags & LOOKUP_PARENT)
856 goto lookup_parent;
857 if (this.name[0] == '.') switch (this.len) {
858 default:
859 break;
860 case 2:
861 if (this.name[1] != '.')
862 break;
58c465eb 863 follow_dotdot(nd);
1da177e4
LT
864 inode = nd->dentry->d_inode;
865 /* fallthrough */
866 case 1:
867 goto return_reval;
868 }
869 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
870 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
871 if (err < 0)
872 break;
873 }
874 err = do_lookup(nd, &this, &next);
875 if (err)
876 break;
1da177e4
LT
877 inode = next.dentry->d_inode;
878 if ((lookup_flags & LOOKUP_FOLLOW)
879 && inode && inode->i_op && inode->i_op->follow_link) {
90ebe565 880 err = do_follow_link(&next, nd);
1da177e4
LT
881 if (err)
882 goto return_err;
883 inode = nd->dentry->d_inode;
09dd17d3
MS
884 } else
885 path_to_nameidata(&next, nd);
1da177e4
LT
886 err = -ENOENT;
887 if (!inode)
888 break;
889 if (lookup_flags & LOOKUP_DIRECTORY) {
890 err = -ENOTDIR;
891 if (!inode->i_op || !inode->i_op->lookup)
892 break;
893 }
894 goto return_base;
895lookup_parent:
896 nd->last = this;
897 nd->last_type = LAST_NORM;
898 if (this.name[0] != '.')
899 goto return_base;
900 if (this.len == 1)
901 nd->last_type = LAST_DOT;
902 else if (this.len == 2 && this.name[1] == '.')
903 nd->last_type = LAST_DOTDOT;
904 else
905 goto return_base;
906return_reval:
907 /*
908 * We bypassed the ordinary revalidation routines.
909 * We may need to check the cached dentry for staleness.
910 */
911 if (nd->dentry && nd->dentry->d_sb &&
912 (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
913 err = -ESTALE;
914 /* Note: we do not d_invalidate() */
915 if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
916 break;
917 }
918return_base:
919 return 0;
920out_dput:
09dd17d3 921 dput_path(&next, nd);
1da177e4
LT
922 break;
923 }
924 path_release(nd);
925return_err:
926 return err;
927}
928
929/*
930 * Wrapper to retry pathname resolution whenever the underlying
931 * file system returns an ESTALE.
932 *
933 * Retry the whole path once, forcing real lookup requests
934 * instead of relying on the dcache.
935 */
936int fastcall link_path_walk(const char *name, struct nameidata *nd)
937{
938 struct nameidata save = *nd;
939 int result;
940
941 /* make sure the stuff we saved doesn't go away */
942 dget(save.dentry);
943 mntget(save.mnt);
944
945 result = __link_path_walk(name, nd);
946 if (result == -ESTALE) {
947 *nd = save;
948 dget(nd->dentry);
949 mntget(nd->mnt);
950 nd->flags |= LOOKUP_REVAL;
951 result = __link_path_walk(name, nd);
952 }
953
954 dput(save.dentry);
955 mntput(save.mnt);
956
957 return result;
958}
959
960int fastcall path_walk(const char * name, struct nameidata *nd)
961{
962 current->total_link_count = 0;
963 return link_path_walk(name, nd);
964}
965
ea3834d9
PM
966/*
967 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
968 * everything is done. Returns 0 and drops input nd, if lookup failed;
969 */
1da177e4
LT
970static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
971{
972 if (path_walk(name, nd))
973 return 0; /* something went wrong... */
974
975 if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
976 struct dentry *old_dentry = nd->dentry;
977 struct vfsmount *old_mnt = nd->mnt;
978 struct qstr last = nd->last;
979 int last_type = nd->last_type;
980 /*
981 * NAME was not found in alternate root or it's a directory. Try to find
982 * it in the normal root:
983 */
984 nd->last_type = LAST_ROOT;
985 read_lock(&current->fs->lock);
986 nd->mnt = mntget(current->fs->rootmnt);
987 nd->dentry = dget(current->fs->root);
988 read_unlock(&current->fs->lock);
989 if (path_walk(name, nd) == 0) {
990 if (nd->dentry->d_inode) {
991 dput(old_dentry);
992 mntput(old_mnt);
993 return 1;
994 }
995 path_release(nd);
996 }
997 nd->dentry = old_dentry;
998 nd->mnt = old_mnt;
999 nd->last = last;
1000 nd->last_type = last_type;
1001 }
1002 return 1;
1003}
1004
1005void set_fs_altroot(void)
1006{
1007 char *emul = __emul_prefix();
1008 struct nameidata nd;
1009 struct vfsmount *mnt = NULL, *oldmnt;
1010 struct dentry *dentry = NULL, *olddentry;
1011 int err;
1012
1013 if (!emul)
1014 goto set_it;
1015 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
1016 if (!err) {
1017 mnt = nd.mnt;
1018 dentry = nd.dentry;
1019 }
1020set_it:
1021 write_lock(&current->fs->lock);
1022 oldmnt = current->fs->altrootmnt;
1023 olddentry = current->fs->altroot;
1024 current->fs->altrootmnt = mnt;
1025 current->fs->altroot = dentry;
1026 write_unlock(&current->fs->lock);
1027 if (olddentry) {
1028 dput(olddentry);
1029 mntput(oldmnt);
1030 }
1031}
1032
ea3834d9 1033/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1da177e4
LT
1034int fastcall path_lookup(const char *name, unsigned int flags, struct nameidata *nd)
1035{
ea3834d9 1036 int retval = 0;
1da177e4
LT
1037
1038 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1039 nd->flags = flags;
1040 nd->depth = 0;
1041
1042 read_lock(&current->fs->lock);
1043 if (*name=='/') {
1044 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
1045 nd->mnt = mntget(current->fs->altrootmnt);
1046 nd->dentry = dget(current->fs->altroot);
1047 read_unlock(&current->fs->lock);
1048 if (__emul_lookup_dentry(name,nd))
ea3834d9 1049 goto out; /* found in altroot */
1da177e4
LT
1050 read_lock(&current->fs->lock);
1051 }
1052 nd->mnt = mntget(current->fs->rootmnt);
1053 nd->dentry = dget(current->fs->root);
1054 } else {
1055 nd->mnt = mntget(current->fs->pwdmnt);
1056 nd->dentry = dget(current->fs->pwd);
1057 }
1058 read_unlock(&current->fs->lock);
1059 current->total_link_count = 0;
1060 retval = link_path_walk(name, nd);
ea3834d9 1061out:
1da177e4
LT
1062 if (unlikely(current->audit_context
1063 && nd && nd->dentry && nd->dentry->d_inode))
ae7b961b 1064 audit_inode(name, nd->dentry->d_inode, flags);
1da177e4
LT
1065 return retval;
1066}
1067
834f2a4a
TM
1068static int __path_lookup_intent_open(const char *name, unsigned int lookup_flags,
1069 struct nameidata *nd, int open_flags, int create_mode)
1070{
1071 struct file *filp = get_empty_filp();
1072 int err;
1073
1074 if (filp == NULL)
1075 return -ENFILE;
1076 nd->intent.open.file = filp;
1077 nd->intent.open.flags = open_flags;
1078 nd->intent.open.create_mode = create_mode;
1079 err = path_lookup(name, lookup_flags|LOOKUP_OPEN, nd);
1080 if (IS_ERR(nd->intent.open.file)) {
1081 if (err == 0) {
1082 err = PTR_ERR(nd->intent.open.file);
1083 path_release(nd);
1084 }
1085 } else if (err != 0)
1086 release_open_intent(nd);
1087 return err;
1088}
1089
1090/**
1091 * path_lookup_open - lookup a file path with open intent
1092 * @name: pointer to file name
1093 * @lookup_flags: lookup intent flags
1094 * @nd: pointer to nameidata
1095 * @open_flags: open intent flags
1096 */
1097int path_lookup_open(const char *name, unsigned int lookup_flags,
1098 struct nameidata *nd, int open_flags)
1099{
1100 return __path_lookup_intent_open(name, lookup_flags, nd,
1101 open_flags, 0);
1102}
1103
1104/**
1105 * path_lookup_create - lookup a file path with open + create intent
1106 * @name: pointer to file name
1107 * @lookup_flags: lookup intent flags
1108 * @nd: pointer to nameidata
1109 * @open_flags: open intent flags
1110 * @create_mode: create intent flags
1111 */
1112int path_lookup_create(const char *name, unsigned int lookup_flags,
1113 struct nameidata *nd, int open_flags, int create_mode)
1114{
1115 return __path_lookup_intent_open(name, lookup_flags|LOOKUP_CREATE, nd,
1116 open_flags, create_mode);
1117}
1118
1119int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags,
1120 struct nameidata *nd, int open_flags)
1121{
1122 char *tmp = getname(name);
1123 int err = PTR_ERR(tmp);
1124
1125 if (!IS_ERR(tmp)) {
1126 err = __path_lookup_intent_open(tmp, lookup_flags, nd, open_flags, 0);
1127 putname(tmp);
1128 }
1129 return err;
1130}
1131
1da177e4
LT
1132/*
1133 * Restricted form of lookup. Doesn't follow links, single-component only,
1134 * needs parent already locked. Doesn't follow mounts.
1135 * SMP-safe.
1136 */
1137static struct dentry * __lookup_hash(struct qstr *name, struct dentry * base, struct nameidata *nd)
1138{
1139 struct dentry * dentry;
1140 struct inode *inode;
1141 int err;
1142
1143 inode = base->d_inode;
1144 err = permission(inode, MAY_EXEC, nd);
1145 dentry = ERR_PTR(err);
1146 if (err)
1147 goto out;
1148
1149 /*
1150 * See if the low-level filesystem might want
1151 * to use its own hash..
1152 */
1153 if (base->d_op && base->d_op->d_hash) {
1154 err = base->d_op->d_hash(base, name);
1155 dentry = ERR_PTR(err);
1156 if (err < 0)
1157 goto out;
1158 }
1159
1160 dentry = cached_lookup(base, name, nd);
1161 if (!dentry) {
1162 struct dentry *new = d_alloc(base, name);
1163 dentry = ERR_PTR(-ENOMEM);
1164 if (!new)
1165 goto out;
1166 dentry = inode->i_op->lookup(inode, new, nd);
1167 if (!dentry)
1168 dentry = new;
1169 else
1170 dput(new);
1171 }
1172out:
1173 return dentry;
1174}
1175
1176struct dentry * lookup_hash(struct qstr *name, struct dentry * base)
1177{
1178 return __lookup_hash(name, base, NULL);
1179}
1180
1181/* SMP-safe */
1182struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
1183{
1184 unsigned long hash;
1185 struct qstr this;
1186 unsigned int c;
1187
1188 this.name = name;
1189 this.len = len;
1190 if (!len)
1191 goto access;
1192
1193 hash = init_name_hash();
1194 while (len--) {
1195 c = *(const unsigned char *)name++;
1196 if (c == '/' || c == '\0')
1197 goto access;
1198 hash = partial_name_hash(c, hash);
1199 }
1200 this.hash = end_name_hash(hash);
1201
1202 return lookup_hash(&this, base);
1203access:
1204 return ERR_PTR(-EACCES);
1205}
1206
1207/*
1208 * namei()
1209 *
1210 * is used by most simple commands to get the inode of a specified name.
1211 * Open, link etc use their own routines, but this is enough for things
1212 * like 'chmod' etc.
1213 *
1214 * namei exists in two versions: namei/lnamei. The only difference is
1215 * that namei follows links, while lnamei does not.
1216 * SMP-safe
1217 */
1218int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1219{
1220 char *tmp = getname(name);
1221 int err = PTR_ERR(tmp);
1222
1223 if (!IS_ERR(tmp)) {
1224 err = path_lookup(tmp, flags, nd);
1225 putname(tmp);
1226 }
1227 return err;
1228}
1229
1230/*
1231 * It's inline, so penalty for filesystems that don't use sticky bit is
1232 * minimal.
1233 */
1234static inline int check_sticky(struct inode *dir, struct inode *inode)
1235{
1236 if (!(dir->i_mode & S_ISVTX))
1237 return 0;
1238 if (inode->i_uid == current->fsuid)
1239 return 0;
1240 if (dir->i_uid == current->fsuid)
1241 return 0;
1242 return !capable(CAP_FOWNER);
1243}
1244
1245/*
1246 * Check whether we can remove a link victim from directory dir, check
1247 * whether the type of victim is right.
1248 * 1. We can't do it if dir is read-only (done in permission())
1249 * 2. We should have write and exec permissions on dir
1250 * 3. We can't remove anything from append-only dir
1251 * 4. We can't do anything with immutable dir (done in permission())
1252 * 5. If the sticky bit on dir is set we should either
1253 * a. be owner of dir, or
1254 * b. be owner of victim, or
1255 * c. have CAP_FOWNER capability
1256 * 6. If the victim is append-only or immutable we can't do antyhing with
1257 * links pointing to it.
1258 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1259 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1260 * 9. We can't remove a root or mountpoint.
1261 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1262 * nfs_async_unlink().
1263 */
1264static inline int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1265{
1266 int error;
1267
1268 if (!victim->d_inode)
1269 return -ENOENT;
1270
1271 BUG_ON(victim->d_parent->d_inode != dir);
1272
1273 error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
1274 if (error)
1275 return error;
1276 if (IS_APPEND(dir))
1277 return -EPERM;
1278 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1279 IS_IMMUTABLE(victim->d_inode))
1280 return -EPERM;
1281 if (isdir) {
1282 if (!S_ISDIR(victim->d_inode->i_mode))
1283 return -ENOTDIR;
1284 if (IS_ROOT(victim))
1285 return -EBUSY;
1286 } else if (S_ISDIR(victim->d_inode->i_mode))
1287 return -EISDIR;
1288 if (IS_DEADDIR(dir))
1289 return -ENOENT;
1290 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1291 return -EBUSY;
1292 return 0;
1293}
1294
1295/* Check whether we can create an object with dentry child in directory
1296 * dir.
1297 * 1. We can't do it if child already exists (open has special treatment for
1298 * this case, but since we are inlined it's OK)
1299 * 2. We can't do it if dir is read-only (done in permission())
1300 * 3. We should have write and exec permissions on dir
1301 * 4. We can't do it if dir is immutable (done in permission())
1302 */
1303static inline int may_create(struct inode *dir, struct dentry *child,
1304 struct nameidata *nd)
1305{
1306 if (child->d_inode)
1307 return -EEXIST;
1308 if (IS_DEADDIR(dir))
1309 return -ENOENT;
1310 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1311}
1312
1313/*
1da177e4
LT
1314 * O_DIRECTORY translates into forcing a directory lookup.
1315 */
1316static inline int lookup_flags(unsigned int f)
1317{
1318 unsigned long retval = LOOKUP_FOLLOW;
1319
1320 if (f & O_NOFOLLOW)
1321 retval &= ~LOOKUP_FOLLOW;
1322
1da177e4
LT
1323 if (f & O_DIRECTORY)
1324 retval |= LOOKUP_DIRECTORY;
1325
1326 return retval;
1327}
1328
1329/*
1330 * p1 and p2 should be directories on the same fs.
1331 */
1332struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1333{
1334 struct dentry *p;
1335
1336 if (p1 == p2) {
1337 down(&p1->d_inode->i_sem);
1338 return NULL;
1339 }
1340
1341 down(&p1->d_inode->i_sb->s_vfs_rename_sem);
1342
1343 for (p = p1; p->d_parent != p; p = p->d_parent) {
1344 if (p->d_parent == p2) {
1345 down(&p2->d_inode->i_sem);
1346 down(&p1->d_inode->i_sem);
1347 return p;
1348 }
1349 }
1350
1351 for (p = p2; p->d_parent != p; p = p->d_parent) {
1352 if (p->d_parent == p1) {
1353 down(&p1->d_inode->i_sem);
1354 down(&p2->d_inode->i_sem);
1355 return p;
1356 }
1357 }
1358
1359 down(&p1->d_inode->i_sem);
1360 down(&p2->d_inode->i_sem);
1361 return NULL;
1362}
1363
1364void unlock_rename(struct dentry *p1, struct dentry *p2)
1365{
1366 up(&p1->d_inode->i_sem);
1367 if (p1 != p2) {
1368 up(&p2->d_inode->i_sem);
1369 up(&p1->d_inode->i_sb->s_vfs_rename_sem);
1370 }
1371}
1372
1373int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1374 struct nameidata *nd)
1375{
1376 int error = may_create(dir, dentry, nd);
1377
1378 if (error)
1379 return error;
1380
1381 if (!dir->i_op || !dir->i_op->create)
1382 return -EACCES; /* shouldn't it be ENOSYS? */
1383 mode &= S_IALLUGO;
1384 mode |= S_IFREG;
1385 error = security_inode_create(dir, dentry, mode);
1386 if (error)
1387 return error;
1388 DQUOT_INIT(dir);
1389 error = dir->i_op->create(dir, dentry, mode, nd);
a74574aa 1390 if (!error)
0eeca283 1391 fsnotify_create(dir, dentry->d_name.name);
1da177e4
LT
1392 return error;
1393}
1394
1395int may_open(struct nameidata *nd, int acc_mode, int flag)
1396{
1397 struct dentry *dentry = nd->dentry;
1398 struct inode *inode = dentry->d_inode;
1399 int error;
1400
1401 if (!inode)
1402 return -ENOENT;
1403
1404 if (S_ISLNK(inode->i_mode))
1405 return -ELOOP;
1406
1407 if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1408 return -EISDIR;
1409
1410 error = permission(inode, acc_mode, nd);
1411 if (error)
1412 return error;
1413
1414 /*
1415 * FIFO's, sockets and device files are special: they don't
1416 * actually live on the filesystem itself, and as such you
1417 * can write to them even if the filesystem is read-only.
1418 */
1419 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1420 flag &= ~O_TRUNC;
1421 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1422 if (nd->mnt->mnt_flags & MNT_NODEV)
1423 return -EACCES;
1424
1425 flag &= ~O_TRUNC;
1426 } else if (IS_RDONLY(inode) && (flag & FMODE_WRITE))
1427 return -EROFS;
1428 /*
1429 * An append-only file must be opened in append mode for writing.
1430 */
1431 if (IS_APPEND(inode)) {
1432 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1433 return -EPERM;
1434 if (flag & O_TRUNC)
1435 return -EPERM;
1436 }
1437
1438 /* O_NOATIME can only be set by the owner or superuser */
1439 if (flag & O_NOATIME)
1440 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1441 return -EPERM;
1442
1443 /*
1444 * Ensure there are no outstanding leases on the file.
1445 */
1446 error = break_lease(inode, flag);
1447 if (error)
1448 return error;
1449
1450 if (flag & O_TRUNC) {
1451 error = get_write_access(inode);
1452 if (error)
1453 return error;
1454
1455 /*
1456 * Refuse to truncate files with mandatory locks held on them.
1457 */
1458 error = locks_verify_locked(inode);
1459 if (!error) {
1460 DQUOT_INIT(inode);
1461
cc4e69de 1462 error = do_truncate(dentry, 0, NULL);
1da177e4
LT
1463 }
1464 put_write_access(inode);
1465 if (error)
1466 return error;
1467 } else
1468 if (flag & FMODE_WRITE)
1469 DQUOT_INIT(inode);
1470
1471 return 0;
1472}
1473
1474/*
1475 * open_namei()
1476 *
1477 * namei for open - this is in fact almost the whole open-routine.
1478 *
1479 * Note that the low bits of "flag" aren't the same as in the open
1480 * system call - they are 00 - no permissions needed
1481 * 01 - read permission needed
1482 * 10 - write permission needed
1483 * 11 - read/write permissions needed
1484 * which is a lot more logical, and also allows the "no perm" needed
1485 * for symlinks (where the permissions are checked later).
1486 * SMP-safe
1487 */
1488int open_namei(const char * pathname, int flag, int mode, struct nameidata *nd)
1489{
834f2a4a 1490 int acc_mode, error;
4e7506e4 1491 struct path path;
1da177e4
LT
1492 struct dentry *dir;
1493 int count = 0;
1494
1495 acc_mode = ACC_MODE(flag);
1496
834f2a4a
TM
1497 /* O_TRUNC implies we need access checks for write permissions */
1498 if (flag & O_TRUNC)
1499 acc_mode |= MAY_WRITE;
1500
1da177e4
LT
1501 /* Allow the LSM permission hook to distinguish append
1502 access from general write access. */
1503 if (flag & O_APPEND)
1504 acc_mode |= MAY_APPEND;
1505
1da177e4
LT
1506 /*
1507 * The simplest case - just a plain lookup.
1508 */
1509 if (!(flag & O_CREAT)) {
834f2a4a 1510 error = path_lookup_open(pathname, lookup_flags(flag), nd, flag);
1da177e4
LT
1511 if (error)
1512 return error;
1513 goto ok;
1514 }
1515
1516 /*
1517 * Create - we need to know the parent.
1518 */
834f2a4a 1519 error = path_lookup_create(pathname, LOOKUP_PARENT, nd, flag, mode);
1da177e4
LT
1520 if (error)
1521 return error;
1522
1523 /*
1524 * We have the parent and last component. First of all, check
1525 * that we are not asked to creat(2) an obvious directory - that
1526 * will not do.
1527 */
1528 error = -EISDIR;
1529 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1530 goto exit;
1531
1532 dir = nd->dentry;
1533 nd->flags &= ~LOOKUP_PARENT;
1534 down(&dir->d_inode->i_sem);
4e7506e4 1535 path.dentry = __lookup_hash(&nd->last, nd->dentry, nd);
d73ffe16 1536 path.mnt = nd->mnt;
1da177e4
LT
1537
1538do_last:
4e7506e4
AV
1539 error = PTR_ERR(path.dentry);
1540 if (IS_ERR(path.dentry)) {
1da177e4
LT
1541 up(&dir->d_inode->i_sem);
1542 goto exit;
1543 }
1544
1545 /* Negative dentry, just create the file */
4e7506e4 1546 if (!path.dentry->d_inode) {
1da177e4
LT
1547 if (!IS_POSIXACL(dir->d_inode))
1548 mode &= ~current->fs->umask;
4e7506e4 1549 error = vfs_create(dir->d_inode, path.dentry, mode, nd);
1da177e4
LT
1550 up(&dir->d_inode->i_sem);
1551 dput(nd->dentry);
4e7506e4 1552 nd->dentry = path.dentry;
1da177e4
LT
1553 if (error)
1554 goto exit;
1555 /* Don't check for write permission, don't truncate */
1556 acc_mode = 0;
1557 flag &= ~O_TRUNC;
1558 goto ok;
1559 }
1560
1561 /*
1562 * It already exists.
1563 */
1564 up(&dir->d_inode->i_sem);
1565
1566 error = -EEXIST;
1567 if (flag & O_EXCL)
1568 goto exit_dput;
1569
e13b210f 1570 if (__follow_mount(&path)) {
1da177e4 1571 error = -ELOOP;
ba7a4c1a
AV
1572 if (flag & O_NOFOLLOW)
1573 goto exit_dput;
1da177e4
LT
1574 }
1575 error = -ENOENT;
4e7506e4 1576 if (!path.dentry->d_inode)
1da177e4 1577 goto exit_dput;
4e7506e4 1578 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
1da177e4
LT
1579 goto do_link;
1580
09dd17d3 1581 path_to_nameidata(&path, nd);
1da177e4 1582 error = -EISDIR;
4e7506e4 1583 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1da177e4
LT
1584 goto exit;
1585ok:
1586 error = may_open(nd, acc_mode, flag);
1587 if (error)
1588 goto exit;
1589 return 0;
1590
1591exit_dput:
09dd17d3 1592 dput_path(&path, nd);
1da177e4 1593exit:
834f2a4a
TM
1594 if (!IS_ERR(nd->intent.open.file))
1595 release_open_intent(nd);
1da177e4
LT
1596 path_release(nd);
1597 return error;
1598
1599do_link:
1600 error = -ELOOP;
1601 if (flag & O_NOFOLLOW)
1602 goto exit_dput;
1603 /*
1604 * This is subtle. Instead of calling do_follow_link() we do the
1605 * thing by hands. The reason is that this way we have zero link_count
1606 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1607 * After that we have the parent and last component, i.e.
1608 * we are in the same situation as after the first path_walk().
1609 * Well, almost - if the last component is normal we get its copy
1610 * stored in nd->last.name and we will have to putname() it when we
1611 * are done. Procfs-like symlinks just set LAST_BIND.
1612 */
1613 nd->flags |= LOOKUP_PARENT;
4e7506e4 1614 error = security_inode_follow_link(path.dentry, nd);
1da177e4
LT
1615 if (error)
1616 goto exit_dput;
cd4e91d3 1617 error = __do_follow_link(&path, nd);
1da177e4
LT
1618 if (error)
1619 return error;
1620 nd->flags &= ~LOOKUP_PARENT;
d671d5e5 1621 if (nd->last_type == LAST_BIND)
1da177e4 1622 goto ok;
1da177e4
LT
1623 error = -EISDIR;
1624 if (nd->last_type != LAST_NORM)
1625 goto exit;
1626 if (nd->last.name[nd->last.len]) {
82984114 1627 __putname(nd->last.name);
1da177e4
LT
1628 goto exit;
1629 }
1630 error = -ELOOP;
1631 if (count++==32) {
82984114 1632 __putname(nd->last.name);
1da177e4
LT
1633 goto exit;
1634 }
1635 dir = nd->dentry;
1636 down(&dir->d_inode->i_sem);
4e7506e4 1637 path.dentry = __lookup_hash(&nd->last, nd->dentry, nd);
d671d5e5 1638 path.mnt = nd->mnt;
82984114 1639 __putname(nd->last.name);
1da177e4
LT
1640 goto do_last;
1641}
1642
1643/**
1644 * lookup_create - lookup a dentry, creating it if it doesn't exist
1645 * @nd: nameidata info
1646 * @is_dir: directory flag
1647 *
1648 * Simple function to lookup and return a dentry and create it
1649 * if it doesn't exist. Is SMP-safe.
c663e5d8
CH
1650 *
1651 * Returns with nd->dentry->d_inode->i_sem locked.
1da177e4
LT
1652 */
1653struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1654{
c663e5d8 1655 struct dentry *dentry = ERR_PTR(-EEXIST);
1da177e4
LT
1656
1657 down(&nd->dentry->d_inode->i_sem);
c663e5d8
CH
1658 /*
1659 * Yucky last component or no last component at all?
1660 * (foo/., foo/.., /////)
1661 */
1da177e4
LT
1662 if (nd->last_type != LAST_NORM)
1663 goto fail;
1664 nd->flags &= ~LOOKUP_PARENT;
c663e5d8
CH
1665
1666 /*
1667 * Do the final lookup.
1668 */
1da177e4
LT
1669 dentry = lookup_hash(&nd->last, nd->dentry);
1670 if (IS_ERR(dentry))
1671 goto fail;
c663e5d8
CH
1672
1673 /*
1674 * Special case - lookup gave negative, but... we had foo/bar/
1675 * From the vfs_mknod() POV we just have a negative dentry -
1676 * all is fine. Let's be bastards - you had / on the end, you've
1677 * been asking for (non-existent) directory. -ENOENT for you.
1678 */
1da177e4
LT
1679 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1680 goto enoent;
1681 return dentry;
1682enoent:
1683 dput(dentry);
1684 dentry = ERR_PTR(-ENOENT);
1685fail:
1686 return dentry;
1687}
f81a0bff 1688EXPORT_SYMBOL_GPL(lookup_create);
1da177e4
LT
1689
1690int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1691{
1692 int error = may_create(dir, dentry, NULL);
1693
1694 if (error)
1695 return error;
1696
1697 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1698 return -EPERM;
1699
1700 if (!dir->i_op || !dir->i_op->mknod)
1701 return -EPERM;
1702
1703 error = security_inode_mknod(dir, dentry, mode, dev);
1704 if (error)
1705 return error;
1706
1707 DQUOT_INIT(dir);
1708 error = dir->i_op->mknod(dir, dentry, mode, dev);
a74574aa 1709 if (!error)
0eeca283 1710 fsnotify_create(dir, dentry->d_name.name);
1da177e4
LT
1711 return error;
1712}
1713
1714asmlinkage long sys_mknod(const char __user * filename, int mode, unsigned dev)
1715{
1716 int error = 0;
1717 char * tmp;
1718 struct dentry * dentry;
1719 struct nameidata nd;
1720
1721 if (S_ISDIR(mode))
1722 return -EPERM;
1723 tmp = getname(filename);
1724 if (IS_ERR(tmp))
1725 return PTR_ERR(tmp);
1726
1727 error = path_lookup(tmp, LOOKUP_PARENT, &nd);
1728 if (error)
1729 goto out;
1730 dentry = lookup_create(&nd, 0);
1731 error = PTR_ERR(dentry);
1732
1733 if (!IS_POSIXACL(nd.dentry->d_inode))
1734 mode &= ~current->fs->umask;
1735 if (!IS_ERR(dentry)) {
1736 switch (mode & S_IFMT) {
1737 case 0: case S_IFREG:
1738 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1739 break;
1740 case S_IFCHR: case S_IFBLK:
1741 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,
1742 new_decode_dev(dev));
1743 break;
1744 case S_IFIFO: case S_IFSOCK:
1745 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,0);
1746 break;
1747 case S_IFDIR:
1748 error = -EPERM;
1749 break;
1750 default:
1751 error = -EINVAL;
1752 }
1753 dput(dentry);
1754 }
1755 up(&nd.dentry->d_inode->i_sem);
1756 path_release(&nd);
1757out:
1758 putname(tmp);
1759
1760 return error;
1761}
1762
1763int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1764{
1765 int error = may_create(dir, dentry, NULL);
1766
1767 if (error)
1768 return error;
1769
1770 if (!dir->i_op || !dir->i_op->mkdir)
1771 return -EPERM;
1772
1773 mode &= (S_IRWXUGO|S_ISVTX);
1774 error = security_inode_mkdir(dir, dentry, mode);
1775 if (error)
1776 return error;
1777
1778 DQUOT_INIT(dir);
1779 error = dir->i_op->mkdir(dir, dentry, mode);
a74574aa 1780 if (!error)
0eeca283 1781 fsnotify_mkdir(dir, dentry->d_name.name);
1da177e4
LT
1782 return error;
1783}
1784
1785asmlinkage long sys_mkdir(const char __user * pathname, int mode)
1786{
1787 int error = 0;
1788 char * tmp;
1789
1790 tmp = getname(pathname);
1791 error = PTR_ERR(tmp);
1792 if (!IS_ERR(tmp)) {
1793 struct dentry *dentry;
1794 struct nameidata nd;
1795
1796 error = path_lookup(tmp, LOOKUP_PARENT, &nd);
1797 if (error)
1798 goto out;
1799 dentry = lookup_create(&nd, 1);
1800 error = PTR_ERR(dentry);
1801 if (!IS_ERR(dentry)) {
1802 if (!IS_POSIXACL(nd.dentry->d_inode))
1803 mode &= ~current->fs->umask;
1804 error = vfs_mkdir(nd.dentry->d_inode, dentry, mode);
1805 dput(dentry);
1806 }
1807 up(&nd.dentry->d_inode->i_sem);
1808 path_release(&nd);
1809out:
1810 putname(tmp);
1811 }
1812
1813 return error;
1814}
1815
1816/*
1817 * We try to drop the dentry early: we should have
1818 * a usage count of 2 if we're the only user of this
1819 * dentry, and if that is true (possibly after pruning
1820 * the dcache), then we drop the dentry now.
1821 *
1822 * A low-level filesystem can, if it choses, legally
1823 * do a
1824 *
1825 * if (!d_unhashed(dentry))
1826 * return -EBUSY;
1827 *
1828 * if it cannot handle the case of removing a directory
1829 * that is still in use by something else..
1830 */
1831void dentry_unhash(struct dentry *dentry)
1832{
1833 dget(dentry);
1834 if (atomic_read(&dentry->d_count))
1835 shrink_dcache_parent(dentry);
1836 spin_lock(&dcache_lock);
1837 spin_lock(&dentry->d_lock);
1838 if (atomic_read(&dentry->d_count) == 2)
1839 __d_drop(dentry);
1840 spin_unlock(&dentry->d_lock);
1841 spin_unlock(&dcache_lock);
1842}
1843
1844int vfs_rmdir(struct inode *dir, struct dentry *dentry)
1845{
1846 int error = may_delete(dir, dentry, 1);
1847
1848 if (error)
1849 return error;
1850
1851 if (!dir->i_op || !dir->i_op->rmdir)
1852 return -EPERM;
1853
1854 DQUOT_INIT(dir);
1855
1856 down(&dentry->d_inode->i_sem);
1857 dentry_unhash(dentry);
1858 if (d_mountpoint(dentry))
1859 error = -EBUSY;
1860 else {
1861 error = security_inode_rmdir(dir, dentry);
1862 if (!error) {
1863 error = dir->i_op->rmdir(dir, dentry);
1864 if (!error)
1865 dentry->d_inode->i_flags |= S_DEAD;
1866 }
1867 }
1868 up(&dentry->d_inode->i_sem);
1869 if (!error) {
1da177e4
LT
1870 d_delete(dentry);
1871 }
1872 dput(dentry);
1873
1874 return error;
1875}
1876
1877asmlinkage long sys_rmdir(const char __user * pathname)
1878{
1879 int error = 0;
1880 char * name;
1881 struct dentry *dentry;
1882 struct nameidata nd;
1883
1884 name = getname(pathname);
1885 if(IS_ERR(name))
1886 return PTR_ERR(name);
1887
1888 error = path_lookup(name, LOOKUP_PARENT, &nd);
1889 if (error)
1890 goto exit;
1891
1892 switch(nd.last_type) {
1893 case LAST_DOTDOT:
1894 error = -ENOTEMPTY;
1895 goto exit1;
1896 case LAST_DOT:
1897 error = -EINVAL;
1898 goto exit1;
1899 case LAST_ROOT:
1900 error = -EBUSY;
1901 goto exit1;
1902 }
1903 down(&nd.dentry->d_inode->i_sem);
1904 dentry = lookup_hash(&nd.last, nd.dentry);
1905 error = PTR_ERR(dentry);
1906 if (!IS_ERR(dentry)) {
1907 error = vfs_rmdir(nd.dentry->d_inode, dentry);
1908 dput(dentry);
1909 }
1910 up(&nd.dentry->d_inode->i_sem);
1911exit1:
1912 path_release(&nd);
1913exit:
1914 putname(name);
1915 return error;
1916}
1917
1918int vfs_unlink(struct inode *dir, struct dentry *dentry)
1919{
1920 int error = may_delete(dir, dentry, 0);
1921
1922 if (error)
1923 return error;
1924
1925 if (!dir->i_op || !dir->i_op->unlink)
1926 return -EPERM;
1927
1928 DQUOT_INIT(dir);
1929
1930 down(&dentry->d_inode->i_sem);
1931 if (d_mountpoint(dentry))
1932 error = -EBUSY;
1933 else {
1934 error = security_inode_unlink(dir, dentry);
1935 if (!error)
1936 error = dir->i_op->unlink(dir, dentry);
1937 }
1938 up(&dentry->d_inode->i_sem);
1939
1940 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
1941 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
e234f35c 1942 d_delete(dentry);
1da177e4 1943 }
0eeca283 1944
1da177e4
LT
1945 return error;
1946}
1947
1948/*
1949 * Make sure that the actual truncation of the file will occur outside its
1950 * directory's i_sem. Truncate can take a long time if there is a lot of
1951 * writeout happening, and we don't want to prevent access to the directory
1952 * while waiting on the I/O.
1953 */
1954asmlinkage long sys_unlink(const char __user * pathname)
1955{
1956 int error = 0;
1957 char * name;
1958 struct dentry *dentry;
1959 struct nameidata nd;
1960 struct inode *inode = NULL;
1961
1962 name = getname(pathname);
1963 if(IS_ERR(name))
1964 return PTR_ERR(name);
1965
1966 error = path_lookup(name, LOOKUP_PARENT, &nd);
1967 if (error)
1968 goto exit;
1969 error = -EISDIR;
1970 if (nd.last_type != LAST_NORM)
1971 goto exit1;
1972 down(&nd.dentry->d_inode->i_sem);
1973 dentry = lookup_hash(&nd.last, nd.dentry);
1974 error = PTR_ERR(dentry);
1975 if (!IS_ERR(dentry)) {
1976 /* Why not before? Because we want correct error value */
1977 if (nd.last.name[nd.last.len])
1978 goto slashes;
1979 inode = dentry->d_inode;
1980 if (inode)
1981 atomic_inc(&inode->i_count);
1982 error = vfs_unlink(nd.dentry->d_inode, dentry);
1983 exit2:
1984 dput(dentry);
1985 }
1986 up(&nd.dentry->d_inode->i_sem);
1987 if (inode)
1988 iput(inode); /* truncate the inode here */
1989exit1:
1990 path_release(&nd);
1991exit:
1992 putname(name);
1993 return error;
1994
1995slashes:
1996 error = !dentry->d_inode ? -ENOENT :
1997 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
1998 goto exit2;
1999}
2000
2001int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname, int mode)
2002{
2003 int error = may_create(dir, dentry, NULL);
2004
2005 if (error)
2006 return error;
2007
2008 if (!dir->i_op || !dir->i_op->symlink)
2009 return -EPERM;
2010
2011 error = security_inode_symlink(dir, dentry, oldname);
2012 if (error)
2013 return error;
2014
2015 DQUOT_INIT(dir);
2016 error = dir->i_op->symlink(dir, dentry, oldname);
a74574aa 2017 if (!error)
0eeca283 2018 fsnotify_create(dir, dentry->d_name.name);
1da177e4
LT
2019 return error;
2020}
2021
2022asmlinkage long sys_symlink(const char __user * oldname, const char __user * newname)
2023{
2024 int error = 0;
2025 char * from;
2026 char * to;
2027
2028 from = getname(oldname);
2029 if(IS_ERR(from))
2030 return PTR_ERR(from);
2031 to = getname(newname);
2032 error = PTR_ERR(to);
2033 if (!IS_ERR(to)) {
2034 struct dentry *dentry;
2035 struct nameidata nd;
2036
2037 error = path_lookup(to, LOOKUP_PARENT, &nd);
2038 if (error)
2039 goto out;
2040 dentry = lookup_create(&nd, 0);
2041 error = PTR_ERR(dentry);
2042 if (!IS_ERR(dentry)) {
2043 error = vfs_symlink(nd.dentry->d_inode, dentry, from, S_IALLUGO);
2044 dput(dentry);
2045 }
2046 up(&nd.dentry->d_inode->i_sem);
2047 path_release(&nd);
2048out:
2049 putname(to);
2050 }
2051 putname(from);
2052 return error;
2053}
2054
2055int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2056{
2057 struct inode *inode = old_dentry->d_inode;
2058 int error;
2059
2060 if (!inode)
2061 return -ENOENT;
2062
2063 error = may_create(dir, new_dentry, NULL);
2064 if (error)
2065 return error;
2066
2067 if (dir->i_sb != inode->i_sb)
2068 return -EXDEV;
2069
2070 /*
2071 * A link to an append-only or immutable file cannot be created.
2072 */
2073 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2074 return -EPERM;
2075 if (!dir->i_op || !dir->i_op->link)
2076 return -EPERM;
2077 if (S_ISDIR(old_dentry->d_inode->i_mode))
2078 return -EPERM;
2079
2080 error = security_inode_link(old_dentry, dir, new_dentry);
2081 if (error)
2082 return error;
2083
2084 down(&old_dentry->d_inode->i_sem);
2085 DQUOT_INIT(dir);
2086 error = dir->i_op->link(old_dentry, dir, new_dentry);
2087 up(&old_dentry->d_inode->i_sem);
e31e14ec 2088 if (!error)
0eeca283 2089 fsnotify_create(dir, new_dentry->d_name.name);
1da177e4
LT
2090 return error;
2091}
2092
2093/*
2094 * Hardlinks are often used in delicate situations. We avoid
2095 * security-related surprises by not following symlinks on the
2096 * newname. --KAB
2097 *
2098 * We don't follow them on the oldname either to be compatible
2099 * with linux 2.0, and to avoid hard-linking to directories
2100 * and other special files. --ADM
2101 */
2102asmlinkage long sys_link(const char __user * oldname, const char __user * newname)
2103{
2104 struct dentry *new_dentry;
2105 struct nameidata nd, old_nd;
2106 int error;
2107 char * to;
2108
2109 to = getname(newname);
2110 if (IS_ERR(to))
2111 return PTR_ERR(to);
2112
2113 error = __user_walk(oldname, 0, &old_nd);
2114 if (error)
2115 goto exit;
2116 error = path_lookup(to, LOOKUP_PARENT, &nd);
2117 if (error)
2118 goto out;
2119 error = -EXDEV;
2120 if (old_nd.mnt != nd.mnt)
2121 goto out_release;
2122 new_dentry = lookup_create(&nd, 0);
2123 error = PTR_ERR(new_dentry);
2124 if (!IS_ERR(new_dentry)) {
2125 error = vfs_link(old_nd.dentry, nd.dentry->d_inode, new_dentry);
2126 dput(new_dentry);
2127 }
2128 up(&nd.dentry->d_inode->i_sem);
2129out_release:
2130 path_release(&nd);
2131out:
2132 path_release(&old_nd);
2133exit:
2134 putname(to);
2135
2136 return error;
2137}
2138
2139/*
2140 * The worst of all namespace operations - renaming directory. "Perverted"
2141 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2142 * Problems:
2143 * a) we can get into loop creation. Check is done in is_subdir().
2144 * b) race potential - two innocent renames can create a loop together.
2145 * That's where 4.4 screws up. Current fix: serialization on
2146 * sb->s_vfs_rename_sem. We might be more accurate, but that's another
2147 * story.
2148 * c) we have to lock _three_ objects - parents and victim (if it exists).
2149 * And that - after we got ->i_sem on parents (until then we don't know
2150 * whether the target exists). Solution: try to be smart with locking
2151 * order for inodes. We rely on the fact that tree topology may change
2152 * only under ->s_vfs_rename_sem _and_ that parent of the object we
2153 * move will be locked. Thus we can rank directories by the tree
2154 * (ancestors first) and rank all non-directories after them.
2155 * That works since everybody except rename does "lock parent, lookup,
2156 * lock child" and rename is under ->s_vfs_rename_sem.
2157 * HOWEVER, it relies on the assumption that any object with ->lookup()
2158 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2159 * we'd better make sure that there's no link(2) for them.
2160 * d) some filesystems don't support opened-but-unlinked directories,
2161 * either because of layout or because they are not ready to deal with
2162 * all cases correctly. The latter will be fixed (taking this sort of
2163 * stuff into VFS), but the former is not going away. Solution: the same
2164 * trick as in rmdir().
2165 * e) conversion from fhandle to dentry may come in the wrong moment - when
2166 * we are removing the target. Solution: we will have to grab ->i_sem
2167 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2168 * ->i_sem on parents, which works but leads to some truely excessive
2169 * locking].
2170 */
75c96f85
AB
2171static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2172 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT
2173{
2174 int error = 0;
2175 struct inode *target;
2176
2177 /*
2178 * If we are going to change the parent - check write permissions,
2179 * we'll need to flip '..'.
2180 */
2181 if (new_dir != old_dir) {
2182 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2183 if (error)
2184 return error;
2185 }
2186
2187 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2188 if (error)
2189 return error;
2190
2191 target = new_dentry->d_inode;
2192 if (target) {
2193 down(&target->i_sem);
2194 dentry_unhash(new_dentry);
2195 }
2196 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2197 error = -EBUSY;
2198 else
2199 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2200 if (target) {
2201 if (!error)
2202 target->i_flags |= S_DEAD;
2203 up(&target->i_sem);
2204 if (d_unhashed(new_dentry))
2205 d_rehash(new_dentry);
2206 dput(new_dentry);
2207 }
e31e14ec 2208 if (!error)
1da177e4 2209 d_move(old_dentry,new_dentry);
1da177e4
LT
2210 return error;
2211}
2212
75c96f85
AB
2213static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2214 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT
2215{
2216 struct inode *target;
2217 int error;
2218
2219 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2220 if (error)
2221 return error;
2222
2223 dget(new_dentry);
2224 target = new_dentry->d_inode;
2225 if (target)
2226 down(&target->i_sem);
2227 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2228 error = -EBUSY;
2229 else
2230 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2231 if (!error) {
2232 /* The following d_move() should become unconditional */
2233 if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
2234 d_move(old_dentry, new_dentry);
1da177e4
LT
2235 }
2236 if (target)
2237 up(&target->i_sem);
2238 dput(new_dentry);
2239 return error;
2240}
2241
2242int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2243 struct inode *new_dir, struct dentry *new_dentry)
2244{
2245 int error;
2246 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
0eeca283 2247 const char *old_name;
1da177e4
LT
2248
2249 if (old_dentry->d_inode == new_dentry->d_inode)
2250 return 0;
2251
2252 error = may_delete(old_dir, old_dentry, is_dir);
2253 if (error)
2254 return error;
2255
2256 if (!new_dentry->d_inode)
2257 error = may_create(new_dir, new_dentry, NULL);
2258 else
2259 error = may_delete(new_dir, new_dentry, is_dir);
2260 if (error)
2261 return error;
2262
2263 if (!old_dir->i_op || !old_dir->i_op->rename)
2264 return -EPERM;
2265
2266 DQUOT_INIT(old_dir);
2267 DQUOT_INIT(new_dir);
2268
0eeca283
RL
2269 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2270
1da177e4
LT
2271 if (is_dir)
2272 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2273 else
2274 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2275 if (!error) {
0eeca283 2276 const char *new_name = old_dentry->d_name.name;
89204c40
JM
2277 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2278 new_dentry->d_inode, old_dentry->d_inode);
1da177e4 2279 }
0eeca283
RL
2280 fsnotify_oldname_free(old_name);
2281
1da177e4
LT
2282 return error;
2283}
2284
2285static inline int do_rename(const char * oldname, const char * newname)
2286{
2287 int error = 0;
2288 struct dentry * old_dir, * new_dir;
2289 struct dentry * old_dentry, *new_dentry;
2290 struct dentry * trap;
2291 struct nameidata oldnd, newnd;
2292
2293 error = path_lookup(oldname, LOOKUP_PARENT, &oldnd);
2294 if (error)
2295 goto exit;
2296
2297 error = path_lookup(newname, LOOKUP_PARENT, &newnd);
2298 if (error)
2299 goto exit1;
2300
2301 error = -EXDEV;
2302 if (oldnd.mnt != newnd.mnt)
2303 goto exit2;
2304
2305 old_dir = oldnd.dentry;
2306 error = -EBUSY;
2307 if (oldnd.last_type != LAST_NORM)
2308 goto exit2;
2309
2310 new_dir = newnd.dentry;
2311 if (newnd.last_type != LAST_NORM)
2312 goto exit2;
2313
2314 trap = lock_rename(new_dir, old_dir);
2315
2316 old_dentry = lookup_hash(&oldnd.last, old_dir);
2317 error = PTR_ERR(old_dentry);
2318 if (IS_ERR(old_dentry))
2319 goto exit3;
2320 /* source must exist */
2321 error = -ENOENT;
2322 if (!old_dentry->d_inode)
2323 goto exit4;
2324 /* unless the source is a directory trailing slashes give -ENOTDIR */
2325 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2326 error = -ENOTDIR;
2327 if (oldnd.last.name[oldnd.last.len])
2328 goto exit4;
2329 if (newnd.last.name[newnd.last.len])
2330 goto exit4;
2331 }
2332 /* source should not be ancestor of target */
2333 error = -EINVAL;
2334 if (old_dentry == trap)
2335 goto exit4;
2336 new_dentry = lookup_hash(&newnd.last, new_dir);
2337 error = PTR_ERR(new_dentry);
2338 if (IS_ERR(new_dentry))
2339 goto exit4;
2340 /* target should not be an ancestor of source */
2341 error = -ENOTEMPTY;
2342 if (new_dentry == trap)
2343 goto exit5;
2344
2345 error = vfs_rename(old_dir->d_inode, old_dentry,
2346 new_dir->d_inode, new_dentry);
2347exit5:
2348 dput(new_dentry);
2349exit4:
2350 dput(old_dentry);
2351exit3:
2352 unlock_rename(new_dir, old_dir);
2353exit2:
2354 path_release(&newnd);
2355exit1:
2356 path_release(&oldnd);
2357exit:
2358 return error;
2359}
2360
2361asmlinkage long sys_rename(const char __user * oldname, const char __user * newname)
2362{
2363 int error;
2364 char * from;
2365 char * to;
2366
2367 from = getname(oldname);
2368 if(IS_ERR(from))
2369 return PTR_ERR(from);
2370 to = getname(newname);
2371 error = PTR_ERR(to);
2372 if (!IS_ERR(to)) {
2373 error = do_rename(from,to);
2374 putname(to);
2375 }
2376 putname(from);
2377 return error;
2378}
2379
2380int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2381{
2382 int len;
2383
2384 len = PTR_ERR(link);
2385 if (IS_ERR(link))
2386 goto out;
2387
2388 len = strlen(link);
2389 if (len > (unsigned) buflen)
2390 len = buflen;
2391 if (copy_to_user(buffer, link, len))
2392 len = -EFAULT;
2393out:
2394 return len;
2395}
2396
2397/*
2398 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2399 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2400 * using) it for any given inode is up to filesystem.
2401 */
2402int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2403{
2404 struct nameidata nd;
cc314eef
LT
2405 void *cookie;
2406
1da177e4 2407 nd.depth = 0;
cc314eef
LT
2408 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2409 if (!IS_ERR(cookie)) {
2410 int res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
1da177e4 2411 if (dentry->d_inode->i_op->put_link)
cc314eef
LT
2412 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2413 cookie = ERR_PTR(res);
1da177e4 2414 }
cc314eef 2415 return PTR_ERR(cookie);
1da177e4
LT
2416}
2417
2418int vfs_follow_link(struct nameidata *nd, const char *link)
2419{
2420 return __vfs_follow_link(nd, link);
2421}
2422
2423/* get the link contents into pagecache */
2424static char *page_getlink(struct dentry * dentry, struct page **ppage)
2425{
2426 struct page * page;
2427 struct address_space *mapping = dentry->d_inode->i_mapping;
2428 page = read_cache_page(mapping, 0, (filler_t *)mapping->a_ops->readpage,
2429 NULL);
2430 if (IS_ERR(page))
2431 goto sync_fail;
2432 wait_on_page_locked(page);
2433 if (!PageUptodate(page))
2434 goto async_fail;
2435 *ppage = page;
2436 return kmap(page);
2437
2438async_fail:
2439 page_cache_release(page);
2440 return ERR_PTR(-EIO);
2441
2442sync_fail:
2443 return (char*)page;
2444}
2445
2446int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2447{
2448 struct page *page = NULL;
2449 char *s = page_getlink(dentry, &page);
2450 int res = vfs_readlink(dentry,buffer,buflen,s);
2451 if (page) {
2452 kunmap(page);
2453 page_cache_release(page);
2454 }
2455 return res;
2456}
2457
cc314eef 2458void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
1da177e4 2459{
cc314eef 2460 struct page *page = NULL;
1da177e4 2461 nd_set_link(nd, page_getlink(dentry, &page));
cc314eef 2462 return page;
1da177e4
LT
2463}
2464
cc314eef 2465void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1da177e4 2466{
cc314eef
LT
2467 struct page *page = cookie;
2468
2469 if (page) {
1da177e4
LT
2470 kunmap(page);
2471 page_cache_release(page);
1da177e4
LT
2472 }
2473}
2474
2475int page_symlink(struct inode *inode, const char *symname, int len)
2476{
2477 struct address_space *mapping = inode->i_mapping;
2478 struct page *page = grab_cache_page(mapping, 0);
2479 int err = -ENOMEM;
2480 char *kaddr;
2481
2482 if (!page)
2483 goto fail;
2484 err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
2485 if (err)
2486 goto fail_map;
2487 kaddr = kmap_atomic(page, KM_USER0);
2488 memcpy(kaddr, symname, len-1);
2489 kunmap_atomic(kaddr, KM_USER0);
2490 mapping->a_ops->commit_write(NULL, page, 0, len-1);
2491 /*
2492 * Notice that we are _not_ going to block here - end of page is
2493 * unmapped, so this will only try to map the rest of page, see
2494 * that it is unmapped (typically even will not look into inode -
2495 * ->i_size will be enough for everything) and zero it out.
2496 * OTOH it's obviously correct and should make the page up-to-date.
2497 */
2498 if (!PageUptodate(page)) {
2499 err = mapping->a_ops->readpage(NULL, page);
2500 wait_on_page_locked(page);
2501 } else {
2502 unlock_page(page);
2503 }
2504 page_cache_release(page);
2505 if (err < 0)
2506 goto fail;
2507 mark_inode_dirty(inode);
2508 return 0;
2509fail_map:
2510 unlock_page(page);
2511 page_cache_release(page);
2512fail:
2513 return err;
2514}
2515
2516struct inode_operations page_symlink_inode_operations = {
2517 .readlink = generic_readlink,
2518 .follow_link = page_follow_link_light,
2519 .put_link = page_put_link,
2520};
2521
2522EXPORT_SYMBOL(__user_walk);
2523EXPORT_SYMBOL(follow_down);
2524EXPORT_SYMBOL(follow_up);
2525EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2526EXPORT_SYMBOL(getname);
2527EXPORT_SYMBOL(lock_rename);
2528EXPORT_SYMBOL(lookup_hash);
2529EXPORT_SYMBOL(lookup_one_len);
2530EXPORT_SYMBOL(page_follow_link_light);
2531EXPORT_SYMBOL(page_put_link);
2532EXPORT_SYMBOL(page_readlink);
2533EXPORT_SYMBOL(page_symlink);
2534EXPORT_SYMBOL(page_symlink_inode_operations);
2535EXPORT_SYMBOL(path_lookup);
2536EXPORT_SYMBOL(path_release);
2537EXPORT_SYMBOL(path_walk);
2538EXPORT_SYMBOL(permission);
2539EXPORT_SYMBOL(unlock_rename);
2540EXPORT_SYMBOL(vfs_create);
2541EXPORT_SYMBOL(vfs_follow_link);
2542EXPORT_SYMBOL(vfs_link);
2543EXPORT_SYMBOL(vfs_mkdir);
2544EXPORT_SYMBOL(vfs_mknod);
2545EXPORT_SYMBOL(generic_permission);
2546EXPORT_SYMBOL(vfs_readlink);
2547EXPORT_SYMBOL(vfs_rename);
2548EXPORT_SYMBOL(vfs_rmdir);
2549EXPORT_SYMBOL(vfs_symlink);
2550EXPORT_SYMBOL(vfs_unlink);
2551EXPORT_SYMBOL(dentry_unhash);
2552EXPORT_SYMBOL(generic_readlink);