projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
[PATCH] Update some VFS documentation
[GitHub/mt8127/android_kernel_alcatel_ttab.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>
16f7e0fe 31#include <linux/capability.h>
834f2a4a 32#include <linux/file.h>
5590ff0d
UD		 33#include <linux/fcntl.h>
34#include <linux/namei.h>
1da177e4
LT		 35#include <asm/namei.h>
36#include <asm/uaccess.h>
37
38#define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
39
40/* [Feb-1997 T. Schoebel-Theuer]
41 * Fundamental changes in the pathname lookup mechanisms (namei)
42 * were necessary because of omirr. The reason is that omirr needs
43 * to know the _real_ pathname, not the user-supplied one, in case
44 * of symlinks (and also when transname replacements occur).
45 *
46 * The new code replaces the old recursive symlink resolution with
47 * an iterative one (in case of non-nested symlink chains). It does
48 * this with calls to <fs>_follow_link().
49 * As a side effect, dir_namei(), _namei() and follow_link() are now
50 * replaced with a single function lookup_dentry() that can handle all
51 * the special cases of the former code.
52 *
53 * With the new dcache, the pathname is stored at each inode, at least as
54 * long as the refcount of the inode is positive. As a side effect, the
55 * size of the dcache depends on the inode cache and thus is dynamic.
56 *
57 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
58 * resolution to correspond with current state of the code.
59 *
60 * Note that the symlink resolution is not *completely* iterative.
61 * There is still a significant amount of tail- and mid- recursion in
62 * the algorithm. Also, note that <fs>_readlink() is not used in
63 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
64 * may return different results than <fs>_follow_link(). Many virtual
65 * filesystems (including /proc) exhibit this behavior.
66 */
67
68/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
69 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
70 * and the name already exists in form of a symlink, try to create the new
71 * name indicated by the symlink. The old code always complained that the
72 * name already exists, due to not following the symlink even if its target
73 * is nonexistent. The new semantics affects also mknod() and link() when
74 * the name is a symlink pointing to a non-existant name.
75 *
76 * I don't know which semantics is the right one, since I have no access
77 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
78 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
79 * "old" one. Personally, I think the new semantics is much more logical.
80 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
81 * file does succeed in both HP-UX and SunOs, but not in Solaris
82 * and in the old Linux semantics.
83 */
84
85/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
86 * semantics. See the comments in "open_namei" and "do_link" below.
87 *
88 * [10-Sep-98 Alan Modra] Another symlink change.
89 */
90
91/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
92 * inside the path - always follow.
93 * in the last component in creation/removal/renaming - never follow.
94 * if LOOKUP_FOLLOW passed - follow.
95 * if the pathname has trailing slashes - follow.
96 * otherwise - don't follow.
97 * (applied in that order).
98 *
99 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
100 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
101 * During the 2.4 we need to fix the userland stuff depending on it -
102 * hopefully we will be able to get rid of that wart in 2.5. So far only
103 * XEmacs seems to be relying on it...
104 */
105/*
106 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
a11f3a05 107 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
1da177e4
LT		 108 * any extra contention...
109 */
110
111/* In order to reduce some races, while at the same time doing additional
112 * checking and hopefully speeding things up, we copy filenames to the
113 * kernel data space before using them..
114 *
115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
116 * PATH_MAX includes the nul terminator --RR.
117 */
858119e1 118static int do_getname(const char __user *filename, char *page)
1da177e4
LT		 119{
120 int retval;
121 unsigned long len = PATH_MAX;
122
123 if (!segment_eq(get_fs(), KERNEL_DS)) {
124 if ((unsigned long) filename >= TASK_SIZE)
125 return -EFAULT;
126 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
127 len = TASK_SIZE - (unsigned long) filename;
128 }
129
130 retval = strncpy_from_user(page, filename, len);
131 if (retval > 0) {
132 if (retval < len)
133 return 0;
134 return -ENAMETOOLONG;
135 } else if (!retval)
136 retval = -ENOENT;
137 return retval;
138}
139
140char * getname(const char __user * filename)
141{
142 char *tmp, *result;
143
144 result = ERR_PTR(-ENOMEM);
145 tmp = __getname();
146 if (tmp) {
147 int retval = do_getname(filename, tmp);
148
149 result = tmp;
150 if (retval < 0) {
151 __putname(tmp);
152 result = ERR_PTR(retval);
153 }
154 }
155 audit_getname(result);
156 return result;
157}
158
159#ifdef CONFIG_AUDITSYSCALL
160void putname(const char *name)
161{
162 if (unlikely(current->audit_context))
163 audit_putname(name);
164 else
165 __putname(name);
166}
167EXPORT_SYMBOL(putname);
168#endif
169
170
171/**
172 * generic_permission - check for access rights on a Posix-like filesystem
173 * @inode: inode to check access rights for
174 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
175 * @check_acl: optional callback to check for Posix ACLs
176 *
177 * Used to check for read/write/execute permissions on a file.
178 * We use "fsuid" for this, letting us set arbitrary permissions
179 * for filesystem access without changing the "normal" uids which
180 * are used for other things..
181 */
182int generic_permission(struct inode *inode, int mask,
183 int (*check_acl)(struct inode *inode, int mask))
184{
185 umode_t mode = inode->i_mode;
186
187 if (current->fsuid == inode->i_uid)
188 mode >>= 6;
189 else {
190 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
191 int error = check_acl(inode, mask);
192 if (error == -EACCES)
193 goto check_capabilities;
194 else if (error != -EAGAIN)
195 return error;
196 }
197
198 if (in_group_p(inode->i_gid))
199 mode >>= 3;
200 }
201
202 /*
203 * If the DACs are ok we don't need any capability check.
204 */
205 if (((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask))
206 return 0;
207
208 check_capabilities:
209 /*
210 * Read/write DACs are always overridable.
211 * Executable DACs are overridable if at least one exec bit is set.
212 */
213 if (!(mask & MAY_EXEC) ||
214 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
215 if (capable(CAP_DAC_OVERRIDE))
216 return 0;
217
218 /*
219 * Searching includes executable on directories, else just read.
220 */
221 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
222 if (capable(CAP_DAC_READ_SEARCH))
223 return 0;
224
225 return -EACCES;
226}
227
228int permission(struct inode *inode, int mask, struct nameidata *nd)
229{
230 int retval, submask;
231
232 if (mask & MAY_WRITE) {
233 umode_t mode = inode->i_mode;
234
235 /*
236 * Nobody gets write access to a read-only fs.
237 */
238 if (IS_RDONLY(inode) &&
239 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
240 return -EROFS;
241
242 /*
243 * Nobody gets write access to an immutable file.
244 */
245 if (IS_IMMUTABLE(inode))
246 return -EACCES;
247 }
248
249
250 /* Ordinary permission routines do not understand MAY_APPEND. */
251 submask = mask & ~MAY_APPEND;
252 if (inode->i_op && inode->i_op->permission)
253 retval = inode->i_op->permission(inode, submask, nd);
254 else
255 retval = generic_permission(inode, submask, NULL);
256 if (retval)
257 return retval;
258
259 return security_inode_permission(inode, mask, nd);
260}
261
e4543edd
CH		 262/**
263 * vfs_permission - check for access rights to a given path
264 * @nd: lookup result that describes the path
265 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
266 *
267 * Used to check for read/write/execute permissions on a path.
268 * We use "fsuid" for this, letting us set arbitrary permissions
269 * for filesystem access without changing the "normal" uids which
270 * are used for other things.
271 */
272int vfs_permission(struct nameidata *nd, int mask)
273{
274 return permission(nd->dentry->d_inode, mask, nd);
275}
276
8c744fb8
CH		 277/**
278 * file_permission - check for additional access rights to a given file
279 * @file: file to check access rights for
280 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
281 *
282 * Used to check for read/write/execute permissions on an already opened
283 * file.
284 *
285 * Note:
286 * Do not use this function in new code. All access checks should
287 * be done using vfs_permission().
288 */
289int file_permission(struct file *file, int mask)
290{
291 return permission(file->f_dentry->d_inode, mask, NULL);
292}
293
1da177e4
LT		 294/*
295 * get_write_access() gets write permission for a file.
296 * put_write_access() releases this write permission.
297 * This is used for regular files.
298 * We cannot support write (and maybe mmap read-write shared) accesses and
299 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
300 * can have the following values:
301 * 0: no writers, no VM_DENYWRITE mappings
302 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
303 * > 0: (i_writecount) users are writing to the file.
304 *
305 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
306 * except for the cases where we don't hold i_writecount yet. Then we need to
307 * use {get,deny}_write_access() - these functions check the sign and refuse
308 * to do the change if sign is wrong. Exclusion between them is provided by
309 * the inode->i_lock spinlock.
310 */
311
312int get_write_access(struct inode * inode)
313{
314 spin_lock(&inode->i_lock);
315 if (atomic_read(&inode->i_writecount) < 0) {
316 spin_unlock(&inode->i_lock);
317 return -ETXTBSY;
318 }
319 atomic_inc(&inode->i_writecount);
320 spin_unlock(&inode->i_lock);
321
322 return 0;
323}
324
325int deny_write_access(struct file * file)
326{
327 struct inode *inode = file->f_dentry->d_inode;
328
329 spin_lock(&inode->i_lock);
330 if (atomic_read(&inode->i_writecount) > 0) {
331 spin_unlock(&inode->i_lock);
332 return -ETXTBSY;
333 }
334 atomic_dec(&inode->i_writecount);
335 spin_unlock(&inode->i_lock);
336
337 return 0;
338}
339
340void path_release(struct nameidata *nd)
341{
342 dput(nd->dentry);
343 mntput(nd->mnt);
344}
345
346/*
347 * umount() mustn't call path_release()/mntput() as that would clear
348 * mnt_expiry_mark
349 */
350void path_release_on_umount(struct nameidata *nd)
351{
352 dput(nd->dentry);
751c404b 353 mntput_no_expire(nd->mnt);
1da177e4
LT		 354}
355
834f2a4a
TM		 356/**
357 * release_open_intent - free up open intent resources
358 * @nd: pointer to nameidata
359 */
360void release_open_intent(struct nameidata *nd)
361{
362 if (nd->intent.open.file->f_dentry == NULL)
363 put_filp(nd->intent.open.file);
364 else
365 fput(nd->intent.open.file);
366}
367
1da177e4
LT		 368/*
369 * Internal lookup() using the new generic dcache.
370 * SMP-safe
371 */
372static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
373{
374 struct dentry * dentry = __d_lookup(parent, name);
375
376 /* lockess __d_lookup may fail due to concurrent d_move()
377 * in some unrelated directory, so try with d_lookup
378 */
379 if (!dentry)
380 dentry = d_lookup(parent, name);
381
382 if (dentry && dentry->d_op && dentry->d_op->d_revalidate) {
383 if (!dentry->d_op->d_revalidate(dentry, nd) && !d_invalidate(dentry)) {
384 dput(dentry);
385 dentry = NULL;
386 }
387 }
388 return dentry;
389}
390
391/*
392 * Short-cut version of permission(), for calling by
393 * path_walk(), when dcache lock is held. Combines parts
394 * of permission() and generic_permission(), and tests ONLY for
395 * MAY_EXEC permission.
396 *
397 * If appropriate, check DAC only. If not appropriate, or
398 * short-cut DAC fails, then call permission() to do more
399 * complete permission check.
400 */
858119e1 401static int exec_permission_lite(struct inode *inode,
1da177e4
LT		 402 struct nameidata *nd)
403{
404 umode_t mode = inode->i_mode;
405
406 if (inode->i_op && inode->i_op->permission)
407 return -EAGAIN;
408
409 if (current->fsuid == inode->i_uid)
410 mode >>= 6;
411 else if (in_group_p(inode->i_gid))
412 mode >>= 3;
413
414 if (mode & MAY_EXEC)
415 goto ok;
416
417 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
418 goto ok;
419
420 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
421 goto ok;
422
423 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
424 goto ok;
425
426 return -EACCES;
427ok:
428 return security_inode_permission(inode, MAY_EXEC, nd);
429}
430
431/*
432 * This is called when everything else fails, and we actually have
433 * to go to the low-level filesystem to find out what we should do..
434 *
435 * We get the directory semaphore, and after getting that we also
436 * make sure that nobody added the entry to the dcache in the meantime..
437 * SMP-safe
438 */
439static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
440{
441 struct dentry * result;
442 struct inode *dir = parent->d_inode;
443
1b1dcc1b 444 mutex_lock(&dir->i_mutex);
1da177e4
LT		 445 /*
446 * First re-do the cached lookup just in case it was created
447 * while we waited for the directory semaphore..
448 *
449 * FIXME! This could use version numbering or similar to
450 * avoid unnecessary cache lookups.
451 *
452 * The "dcache_lock" is purely to protect the RCU list walker
453 * from concurrent renames at this point (we mustn't get false
454 * negatives from the RCU list walk here, unlike the optimistic
455 * fast walk).
456 *
457 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
458 */
459 result = d_lookup(parent, name);
460 if (!result) {
461 struct dentry * dentry = d_alloc(parent, name);
462 result = ERR_PTR(-ENOMEM);
463 if (dentry) {
464 result = dir->i_op->lookup(dir, dentry, nd);
465 if (result)
466 dput(dentry);
467 else
468 result = dentry;
469 }
1b1dcc1b 470 mutex_unlock(&dir->i_mutex);
1da177e4
LT		 471 return result;
472 }
473
474 /*
475 * Uhhuh! Nasty case: the cache was re-populated while
476 * we waited on the semaphore. Need to revalidate.
477 */
1b1dcc1b 478 mutex_unlock(&dir->i_mutex);
1da177e4
LT		 479 if (result->d_op && result->d_op->d_revalidate) {
480 if (!result->d_op->d_revalidate(result, nd) && !d_invalidate(result)) {
481 dput(result);
482 result = ERR_PTR(-ENOENT);
483 }
484 }
485 return result;
486}
487
488static int __emul_lookup_dentry(const char *, struct nameidata *);
489
490/* SMP-safe */
f1662356 491static __always_inline int
1da177e4
LT		 492walk_init_root(const char *name, struct nameidata *nd)
493{
494 read_lock(&current->fs->lock);
495 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
496 nd->mnt = mntget(current->fs->altrootmnt);
497 nd->dentry = dget(current->fs->altroot);
498 read_unlock(&current->fs->lock);
499 if (__emul_lookup_dentry(name,nd))
500 return 0;
501 read_lock(&current->fs->lock);
502 }
503 nd->mnt = mntget(current->fs->rootmnt);
504 nd->dentry = dget(current->fs->root);
505 read_unlock(&current->fs->lock);
506 return 1;
507}
508
f1662356 509static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
1da177e4
LT		 510{
511 int res = 0;
512 char *name;
513 if (IS_ERR(link))
514 goto fail;
515
516 if (*link == '/') {
517 path_release(nd);
518 if (!walk_init_root(link, nd))
519 /* weird __emul_prefix() stuff did it */
520 goto out;
521 }
522 res = link_path_walk(link, nd);
523out:
524 if (nd->depth || res || nd->last_type!=LAST_NORM)
525 return res;
526 /*
527 * If it is an iterative symlinks resolution in open_namei() we
528 * have to copy the last component. And all that crap because of
529 * bloody create() on broken symlinks. Furrfu...
530 */
531 name = __getname();
532 if (unlikely(!name)) {
533 path_release(nd);
534 return -ENOMEM;
535 }
536 strcpy(name, nd->last.name);
537 nd->last.name = name;
538 return 0;
539fail:
540 path_release(nd);
541 return PTR_ERR(link);
542}
543
90ebe565
AV		 544struct path {
545 struct vfsmount *mnt;
546 struct dentry *dentry;
547};
548
f1662356 549static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
1da177e4
LT		 550{
551 int error;
cc314eef 552 void *cookie;
cd4e91d3 553 struct dentry *dentry = path->dentry;
1da177e4 554
d671a1cb 555 touch_atime(path->mnt, dentry);
1da177e4 556 nd_set_link(nd, NULL);
cd4e91d3 557
39ca6d49
AV		 558 if (path->mnt == nd->mnt)
559 mntget(path->mnt);
cc314eef
LT		 560 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
561 error = PTR_ERR(cookie);
562 if (!IS_ERR(cookie)) {
1da177e4 563 char *s = nd_get_link(nd);
cc314eef 564 error = 0;
1da177e4
LT		 565 if (s)
566 error = __vfs_follow_link(nd, s);
567 if (dentry->d_inode->i_op->put_link)
cc314eef 568 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
1da177e4 569 }
cd4e91d3
AV		 570 dput(dentry);
571 mntput(path->mnt);
1da177e4
LT		 572
573 return error;
574}
575
09dd17d3
MS		 576static inline void dput_path(struct path *path, struct nameidata *nd)
577{
578 dput(path->dentry);
579 if (path->mnt != nd->mnt)
580 mntput(path->mnt);
581}
582
583static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
584{
585 dput(nd->dentry);
586 if (nd->mnt != path->mnt)
587 mntput(nd->mnt);
588 nd->mnt = path->mnt;
589 nd->dentry = path->dentry;
590}
591
1da177e4
LT		 592/*
593 * This limits recursive symlink follows to 8, while
594 * limiting consecutive symlinks to 40.
595 *
596 * Without that kind of total limit, nasty chains of consecutive
597 * symlinks can cause almost arbitrarily long lookups.
598 */
90ebe565 599static inline int do_follow_link(struct path *path, struct nameidata *nd)
1da177e4
LT		 600{
601 int err = -ELOOP;
602 if (current->link_count >= MAX_NESTED_LINKS)
603 goto loop;
604 if (current->total_link_count >= 40)
605 goto loop;
606 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
607 cond_resched();
90ebe565 608 err = security_inode_follow_link(path->dentry, nd);
1da177e4
LT		 609 if (err)
610 goto loop;
611 current->link_count++;
612 current->total_link_count++;
613 nd->depth++;
cd4e91d3 614 err = __do_follow_link(path, nd);
839d9f93
AV		 615 current->link_count--;
616 nd->depth--;
1da177e4
LT		 617 return err;
618loop:
09dd17d3 619 dput_path(path, nd);
839d9f93 620 path_release(nd);
1da177e4
LT		 621 return err;
622}
623
624int follow_up(struct vfsmount **mnt, struct dentry **dentry)
625{
626 struct vfsmount *parent;
627 struct dentry *mountpoint;
628 spin_lock(&vfsmount_lock);
629 parent=(*mnt)->mnt_parent;
630 if (parent == *mnt) {
631 spin_unlock(&vfsmount_lock);
632 return 0;
633 }
634 mntget(parent);
635 mountpoint=dget((*mnt)->mnt_mountpoint);
636 spin_unlock(&vfsmount_lock);
637 dput(*dentry);
638 *dentry = mountpoint;
639 mntput(*mnt);
640 *mnt = parent;
641 return 1;
642}
643
644/* no need for dcache_lock, as serialization is taken care in
645 * namespace.c
646 */
463ffb2e
AV		 647static int __follow_mount(struct path *path)
648{
649 int res = 0;
650 while (d_mountpoint(path->dentry)) {
651 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
652 if (!mounted)
653 break;
654 dput(path->dentry);
655 if (res)
656 mntput(path->mnt);
657 path->mnt = mounted;
658 path->dentry = dget(mounted->mnt_root);
659 res = 1;
660 }
661 return res;
662}
663
58c465eb 664static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
1da177e4 665{
1da177e4
LT		 666 while (d_mountpoint(*dentry)) {
667 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
668 if (!mounted)
669 break;
58c465eb 670 dput(*dentry);
1da177e4
LT		 671 mntput(*mnt);
672 *mnt = mounted;
1da177e4 673 *dentry = dget(mounted->mnt_root);
1da177e4 674 }
1da177e4
LT		 675}
676
677/* no need for dcache_lock, as serialization is taken care in
678 * namespace.c
679 */
e13b210f 680int follow_down(struct vfsmount **mnt, struct dentry **dentry)
1da177e4
LT		 681{
682 struct vfsmount *mounted;
683
684 mounted = lookup_mnt(*mnt, *dentry);
685 if (mounted) {
e13b210f 686 dput(*dentry);
1da177e4
LT		 687 mntput(*mnt);
688 *mnt = mounted;
1da177e4
LT		 689 *dentry = dget(mounted->mnt_root);
690 return 1;
691 }
692 return 0;
693}
694
f1662356 695static __always_inline void follow_dotdot(struct nameidata *nd)
1da177e4
LT		 696{
697 while(1) {
698 struct vfsmount *parent;
58c465eb 699 struct dentry *old = nd->dentry;
1da177e4
LT		 700
701 read_lock(&current->fs->lock);
58c465eb
AV		 702 if (nd->dentry == current->fs->root &&
703 nd->mnt == current->fs->rootmnt) {
1da177e4
LT		 704 read_unlock(&current->fs->lock);
705 break;
706 }
707 read_unlock(&current->fs->lock);
708 spin_lock(&dcache_lock);
58c465eb
AV		 709 if (nd->dentry != nd->mnt->mnt_root) {
710 nd->dentry = dget(nd->dentry->d_parent);
1da177e4
LT		 711 spin_unlock(&dcache_lock);
712 dput(old);
713 break;
714 }
715 spin_unlock(&dcache_lock);
716 spin_lock(&vfsmount_lock);
58c465eb
AV		 717 parent = nd->mnt->mnt_parent;
718 if (parent == nd->mnt) {
1da177e4
LT		 719 spin_unlock(&vfsmount_lock);
720 break;
721 }
722 mntget(parent);
58c465eb 723 nd->dentry = dget(nd->mnt->mnt_mountpoint);
1da177e4
LT		 724 spin_unlock(&vfsmount_lock);
725 dput(old);
58c465eb
AV		 726 mntput(nd->mnt);
727 nd->mnt = parent;
1da177e4 728 }
58c465eb 729 follow_mount(&nd->mnt, &nd->dentry);
1da177e4
LT		 730}
731
1da177e4
LT		 732/*
733 * It's more convoluted than I'd like it to be, but... it's still fairly
734 * small and for now I'd prefer to have fast path as straight as possible.
735 * It _is_ time-critical.
736 */
737static int do_lookup(struct nameidata *nd, struct qstr *name,
738 struct path *path)
739{
740 struct vfsmount *mnt = nd->mnt;
741 struct dentry *dentry = __d_lookup(nd->dentry, name);
742
743 if (!dentry)
744 goto need_lookup;
745 if (dentry->d_op && dentry->d_op->d_revalidate)
746 goto need_revalidate;
747done:
748 path->mnt = mnt;
749 path->dentry = dentry;
634ee701 750 __follow_mount(path);
1da177e4
LT		 751 return 0;
752
753need_lookup:
754 dentry = real_lookup(nd->dentry, name, nd);
755 if (IS_ERR(dentry))
756 goto fail;
757 goto done;
758
759need_revalidate:
760 if (dentry->d_op->d_revalidate(dentry, nd))
761 goto done;
762 if (d_invalidate(dentry))
763 goto done;
764 dput(dentry);
765 goto need_lookup;
766
767fail:
768 return PTR_ERR(dentry);
769}
770
771/*
772 * Name resolution.
ea3834d9
PM		 773 * This is the basic name resolution function, turning a pathname into
774 * the final dentry. We expect 'base' to be positive and a directory.
1da177e4 775 *
ea3834d9
PM		 776 * Returns 0 and nd will have valid dentry and mnt on success.
777 * Returns error and drops reference to input namei data on failure.
1da177e4
LT		 778 */
779static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
780{
781 struct path next;
782 struct inode *inode;
783 int err;
784 unsigned int lookup_flags = nd->flags;
785
786 while (*name=='/')
787 name++;
788 if (!*name)
789 goto return_reval;
790
791 inode = nd->dentry->d_inode;
792 if (nd->depth)
f55eab82 793 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
1da177e4
LT		 794
795 /* At this point we know we have a real path component. */
796 for(;;) {
797 unsigned long hash;
798 struct qstr this;
799 unsigned int c;
800
cdce5d6b 801 nd->flags |= LOOKUP_CONTINUE;
1da177e4 802 err = exec_permission_lite(inode, nd);
e4543edd
CH		 803 if (err == -EAGAIN)
804 err = vfs_permission(nd, MAY_EXEC);
1da177e4
LT		 805 if (err)
806 break;
807
808 this.name = name;
809 c = *(const unsigned char *)name;
810
811 hash = init_name_hash();
812 do {
813 name++;
814 hash = partial_name_hash(c, hash);
815 c = *(const unsigned char *)name;
816 } while (c && (c != '/'));
817 this.len = name - (const char *) this.name;
818 this.hash = end_name_hash(hash);
819
820 /* remove trailing slashes? */
821 if (!c)
822 goto last_component;
823 while (*++name == '/');
824 if (!*name)
825 goto last_with_slashes;
826
827 /*
828 * "." and ".." are special - ".." especially so because it has
829 * to be able to know about the current root directory and
830 * parent relationships.
831 */
832 if (this.name[0] == '.') switch (this.len) {
833 default:
834 break;
835 case 2:
836 if (this.name[1] != '.')
837 break;
58c465eb 838 follow_dotdot(nd);
1da177e4
LT		 839 inode = nd->dentry->d_inode;
840 /* fallthrough */
841 case 1:
842 continue;
843 }
844 /*
845 * See if the low-level filesystem might want
846 * to use its own hash..
847 */
848 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
849 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
850 if (err < 0)
851 break;
852 }
1da177e4
LT		 853 /* This does the actual lookups.. */
854 err = do_lookup(nd, &this, &next);
855 if (err)
856 break;
1da177e4
LT		 857
858 err = -ENOENT;
859 inode = next.dentry->d_inode;
860 if (!inode)
861 goto out_dput;
862 err = -ENOTDIR;
863 if (!inode->i_op)
864 goto out_dput;
865
866 if (inode->i_op->follow_link) {
90ebe565 867 err = do_follow_link(&next, nd);
1da177e4
LT		 868 if (err)
869 goto return_err;
870 err = -ENOENT;
871 inode = nd->dentry->d_inode;
872 if (!inode)
873 break;
874 err = -ENOTDIR;
875 if (!inode->i_op)
876 break;
09dd17d3
MS		 877 } else
878 path_to_nameidata(&next, nd);
1da177e4
LT		 879 err = -ENOTDIR;
880 if (!inode->i_op->lookup)
881 break;
882 continue;
883 /* here ends the main loop */
884
885last_with_slashes:
886 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
887last_component:
f55eab82
TM		 888 /* Clear LOOKUP_CONTINUE iff it was previously unset */
889 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
1da177e4
LT		 890 if (lookup_flags & LOOKUP_PARENT)
891 goto lookup_parent;
892 if (this.name[0] == '.') switch (this.len) {
893 default:
894 break;
895 case 2:
896 if (this.name[1] != '.')
897 break;
58c465eb 898 follow_dotdot(nd);
1da177e4
LT		 899 inode = nd->dentry->d_inode;
900 /* fallthrough */
901 case 1:
902 goto return_reval;
903 }
904 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
905 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
906 if (err < 0)
907 break;
908 }
909 err = do_lookup(nd, &this, &next);
910 if (err)
911 break;
1da177e4
LT		 912 inode = next.dentry->d_inode;
913 if ((lookup_flags & LOOKUP_FOLLOW)
914 && inode && inode->i_op && inode->i_op->follow_link) {
90ebe565 915 err = do_follow_link(&next, nd);
1da177e4
LT		 916 if (err)
917 goto return_err;
918 inode = nd->dentry->d_inode;
09dd17d3
MS		 919 } else
920 path_to_nameidata(&next, nd);
1da177e4
LT		 921 err = -ENOENT;
922 if (!inode)
923 break;
924 if (lookup_flags & LOOKUP_DIRECTORY) {
925 err = -ENOTDIR;
926 if (!inode->i_op || !inode->i_op->lookup)
927 break;
928 }
929 goto return_base;
930lookup_parent:
931 nd->last = this;
932 nd->last_type = LAST_NORM;
933 if (this.name[0] != '.')
934 goto return_base;
935 if (this.len == 1)
936 nd->last_type = LAST_DOT;
937 else if (this.len == 2 && this.name[1] == '.')
938 nd->last_type = LAST_DOTDOT;
939 else
940 goto return_base;
941return_reval:
942 /*
943 * We bypassed the ordinary revalidation routines.
944 * We may need to check the cached dentry for staleness.
945 */
946 if (nd->dentry && nd->dentry->d_sb &&
947 (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
948 err = -ESTALE;
949 /* Note: we do not d_invalidate() */
950 if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
951 break;
952 }
953return_base:
954 return 0;
955out_dput:
09dd17d3 956 dput_path(&next, nd);
1da177e4
LT		 957 break;
958 }
959 path_release(nd);
960return_err:
961 return err;
962}
963
964/*
965 * Wrapper to retry pathname resolution whenever the underlying
966 * file system returns an ESTALE.
967 *
968 * Retry the whole path once, forcing real lookup requests
969 * instead of relying on the dcache.
970 */
971int fastcall link_path_walk(const char *name, struct nameidata *nd)
972{
973 struct nameidata save = *nd;
974 int result;
975
976 /* make sure the stuff we saved doesn't go away */
977 dget(save.dentry);
978 mntget(save.mnt);
979
980 result = __link_path_walk(name, nd);
981 if (result == -ESTALE) {
982 *nd = save;
983 dget(nd->dentry);
984 mntget(nd->mnt);
985 nd->flags |= LOOKUP_REVAL;
986 result = __link_path_walk(name, nd);
987 }
988
989 dput(save.dentry);
990 mntput(save.mnt);
991
992 return result;
993}
994
995int fastcall path_walk(const char * name, struct nameidata *nd)
996{
997 current->total_link_count = 0;
998 return link_path_walk(name, nd);
999}
1000
ea3834d9
PM		 1001/*
1002 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
1003 * everything is done. Returns 0 and drops input nd, if lookup failed;
1004 */
1da177e4
LT		 1005static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
1006{
1007 if (path_walk(name, nd))
1008 return 0; /* something went wrong... */
1009
1010 if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
1011 struct dentry *old_dentry = nd->dentry;
1012 struct vfsmount *old_mnt = nd->mnt;
1013 struct qstr last = nd->last;
1014 int last_type = nd->last_type;
1015 /*
1016 * NAME was not found in alternate root or it's a directory. Try to find
1017 * it in the normal root:
1018 */
1019 nd->last_type = LAST_ROOT;
1020 read_lock(&current->fs->lock);
1021 nd->mnt = mntget(current->fs->rootmnt);
1022 nd->dentry = dget(current->fs->root);
1023 read_unlock(&current->fs->lock);
1024 if (path_walk(name, nd) == 0) {
1025 if (nd->dentry->d_inode) {
1026 dput(old_dentry);
1027 mntput(old_mnt);
1028 return 1;
1029 }
1030 path_release(nd);
1031 }
1032 nd->dentry = old_dentry;
1033 nd->mnt = old_mnt;
1034 nd->last = last;
1035 nd->last_type = last_type;
1036 }
1037 return 1;
1038}
1039
1040void set_fs_altroot(void)
1041{
1042 char *emul = __emul_prefix();
1043 struct nameidata nd;
1044 struct vfsmount *mnt = NULL, *oldmnt;
1045 struct dentry *dentry = NULL, *olddentry;
1046 int err;
1047
1048 if (!emul)
1049 goto set_it;
1050 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
1051 if (!err) {
1052 mnt = nd.mnt;
1053 dentry = nd.dentry;
1054 }
1055set_it:
1056 write_lock(&current->fs->lock);
1057 oldmnt = current->fs->altrootmnt;
1058 olddentry = current->fs->altroot;
1059 current->fs->altrootmnt = mnt;
1060 current->fs->altroot = dentry;
1061 write_unlock(&current->fs->lock);
1062 if (olddentry) {
1063 dput(olddentry);
1064 mntput(oldmnt);
1065 }
1066}
1067
ea3834d9 1068/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
5590ff0d
UD		 1069static int fastcall do_path_lookup(int dfd, const char *name,
1070 unsigned int flags, struct nameidata *nd)
1da177e4 1071{
ea3834d9 1072 int retval = 0;
170aa3d0
UD		 1073 int fput_needed;
1074 struct file *file;
1da177e4
LT		 1075
1076 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1077 nd->flags = flags;
1078 nd->depth = 0;
1079
1080 read_lock(&current->fs->lock);
1081 if (*name=='/') {
1082 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
1083 nd->mnt = mntget(current->fs->altrootmnt);
1084 nd->dentry = dget(current->fs->altroot);
1085 read_unlock(&current->fs->lock);
1086 if (__emul_lookup_dentry(name,nd))
ea3834d9 1087 goto out; /* found in altroot */
1da177e4
LT		 1088 read_lock(&current->fs->lock);
1089 }
1090 nd->mnt = mntget(current->fs->rootmnt);
1091 nd->dentry = dget(current->fs->root);
5590ff0d 1092 } else if (dfd == AT_FDCWD) {
1da177e4
LT		 1093 nd->mnt = mntget(current->fs->pwdmnt);
1094 nd->dentry = dget(current->fs->pwd);
5590ff0d 1095 } else {
5590ff0d
UD		 1096 struct dentry *dentry;
1097
1098 file = fget_light(dfd, &fput_needed);
170aa3d0
UD		 1099 retval = -EBADF;
1100 if (!file)
1101 goto unlock_fail;
5590ff0d
UD		 1102
1103 dentry = file->f_dentry;
1104
170aa3d0
UD		 1105 retval = -ENOTDIR;
1106 if (!S_ISDIR(dentry->d_inode->i_mode))
1107 goto fput_unlock_fail;
5590ff0d
UD		 1108
1109 retval = file_permission(file, MAY_EXEC);
170aa3d0
UD		 1110 if (retval)
1111 goto fput_unlock_fail;
5590ff0d
UD		 1112
1113 nd->mnt = mntget(file->f_vfsmnt);
1114 nd->dentry = dget(dentry);
1115
1116 fput_light(file, fput_needed);
1da177e4
LT		 1117 }
1118 read_unlock(&current->fs->lock);
1119 current->total_link_count = 0;
1120 retval = link_path_walk(name, nd);
ea3834d9 1121out:
3bc8414b
S		 1122 if (likely(retval == 0)) {
1123 if (unlikely(current->audit_context && nd && nd->dentry &&
1124 nd->dentry->d_inode))
ae7b961b 1125 audit_inode(name, nd->dentry->d_inode, flags);
3bc8414b 1126 }
170aa3d0
UD		 1127 return retval;
1128
1129fput_unlock_fail:
1130 fput_light(file, fput_needed);
1131unlock_fail:
1132 read_unlock(&current->fs->lock);
1da177e4
LT		 1133 return retval;
1134}
1135
5590ff0d
UD		 1136int fastcall path_lookup(const char *name, unsigned int flags,
1137 struct nameidata *nd)
1138{
1139 return do_path_lookup(AT_FDCWD, name, flags, nd);
1140}
1141
1142static int __path_lookup_intent_open(int dfd, const char *name,
1143 unsigned int lookup_flags, struct nameidata *nd,
1144 int open_flags, int create_mode)
834f2a4a
TM		 1145{
1146 struct file *filp = get_empty_filp();
1147 int err;
1148
1149 if (filp == NULL)
1150 return -ENFILE;
1151 nd->intent.open.file = filp;
1152 nd->intent.open.flags = open_flags;
1153 nd->intent.open.create_mode = create_mode;
5590ff0d 1154 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
834f2a4a
TM		 1155 if (IS_ERR(nd->intent.open.file)) {
1156 if (err == 0) {
1157 err = PTR_ERR(nd->intent.open.file);
1158 path_release(nd);
1159 }
1160 } else if (err != 0)
1161 release_open_intent(nd);
1162 return err;
1163}
1164
1165/**
1166 * path_lookup_open - lookup a file path with open intent
7045f37b 1167 * @dfd: the directory to use as base, or AT_FDCWD
834f2a4a
TM		 1168 * @name: pointer to file name
1169 * @lookup_flags: lookup intent flags
1170 * @nd: pointer to nameidata
1171 * @open_flags: open intent flags
1172 */
5590ff0d 1173int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
834f2a4a
TM		 1174 struct nameidata *nd, int open_flags)
1175{
5590ff0d 1176 return __path_lookup_intent_open(dfd, name, lookup_flags, nd,
834f2a4a
TM		 1177 open_flags, 0);
1178}
1179
1180/**
1181 * path_lookup_create - lookup a file path with open + create intent
7045f37b 1182 * @dfd: the directory to use as base, or AT_FDCWD
834f2a4a
TM		 1183 * @name: pointer to file name
1184 * @lookup_flags: lookup intent flags
1185 * @nd: pointer to nameidata
1186 * @open_flags: open intent flags
1187 * @create_mode: create intent flags
1188 */
5590ff0d
UD		 1189static int path_lookup_create(int dfd, const char *name,
1190 unsigned int lookup_flags, struct nameidata *nd,
1191 int open_flags, int create_mode)
834f2a4a 1192{
5590ff0d
UD		 1193 return __path_lookup_intent_open(dfd, name, lookup_flags|LOOKUP_CREATE,
1194 nd, open_flags, create_mode);
834f2a4a
TM		 1195}
1196
1197int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags,
1198 struct nameidata *nd, int open_flags)
1199{
1200 char *tmp = getname(name);
1201 int err = PTR_ERR(tmp);
1202
1203 if (!IS_ERR(tmp)) {
5590ff0d 1204 err = __path_lookup_intent_open(AT_FDCWD, tmp, lookup_flags, nd, open_flags, 0);
834f2a4a
TM		 1205 putname(tmp);
1206 }
1207 return err;
1208}
1209
1da177e4
LT		 1210/*
1211 * Restricted form of lookup. Doesn't follow links, single-component only,
1212 * needs parent already locked. Doesn't follow mounts.
1213 * SMP-safe.
1214 */
1215static struct dentry * __lookup_hash(struct qstr *name, struct dentry * base, struct nameidata *nd)
1216{
1217 struct dentry * dentry;
1218 struct inode *inode;
1219 int err;
1220
1221 inode = base->d_inode;
1222 err = permission(inode, MAY_EXEC, nd);
1223 dentry = ERR_PTR(err);
1224 if (err)
1225 goto out;
1226
1227 /*
1228 * See if the low-level filesystem might want
1229 * to use its own hash..
1230 */
1231 if (base->d_op && base->d_op->d_hash) {
1232 err = base->d_op->d_hash(base, name);
1233 dentry = ERR_PTR(err);
1234 if (err < 0)
1235 goto out;
1236 }
1237
1238 dentry = cached_lookup(base, name, nd);
1239 if (!dentry) {
1240 struct dentry *new = d_alloc(base, name);
1241 dentry = ERR_PTR(-ENOMEM);
1242 if (!new)
1243 goto out;
1244 dentry = inode->i_op->lookup(inode, new, nd);
1245 if (!dentry)
1246 dentry = new;
1247 else
1248 dput(new);
1249 }
1250out:
1251 return dentry;
1252}
1253
49705b77 1254struct dentry * lookup_hash(struct nameidata *nd)
1da177e4 1255{
49705b77 1256 return __lookup_hash(&nd->last, nd->dentry, nd);
1da177e4
LT		 1257}
1258
1259/* SMP-safe */
1260struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
1261{
1262 unsigned long hash;
1263 struct qstr this;
1264 unsigned int c;
1265
1266 this.name = name;
1267 this.len = len;
1268 if (!len)
1269 goto access;
1270
1271 hash = init_name_hash();
1272 while (len--) {
1273 c = *(const unsigned char *)name++;
1274 if (c == '/' || c == '\0')
1275 goto access;
1276 hash = partial_name_hash(c, hash);
1277 }
1278 this.hash = end_name_hash(hash);
1279
49705b77 1280 return __lookup_hash(&this, base, NULL);
1da177e4
LT		 1281access:
1282 return ERR_PTR(-EACCES);
1283}
1284
1285/*
1286 * namei()
1287 *
1288 * is used by most simple commands to get the inode of a specified name.
1289 * Open, link etc use their own routines, but this is enough for things
1290 * like 'chmod' etc.
1291 *
1292 * namei exists in two versions: namei/lnamei. The only difference is
1293 * that namei follows links, while lnamei does not.
1294 * SMP-safe
1295 */
5590ff0d
UD		 1296int fastcall __user_walk_fd(int dfd, const char __user *name, unsigned flags,
1297 struct nameidata *nd)
1da177e4
LT		 1298{
1299 char *tmp = getname(name);
1300 int err = PTR_ERR(tmp);
1301
1302 if (!IS_ERR(tmp)) {
5590ff0d 1303 err = do_path_lookup(dfd, tmp, flags, nd);
1da177e4
LT		 1304 putname(tmp);
1305 }
1306 return err;
1307}
1308
5590ff0d
UD		 1309int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1310{
1311 return __user_walk_fd(AT_FDCWD, name, flags, nd);
1312}
1313
1da177e4
LT		 1314/*
1315 * It's inline, so penalty for filesystems that don't use sticky bit is
1316 * minimal.
1317 */
1318static inline int check_sticky(struct inode *dir, struct inode *inode)
1319{
1320 if (!(dir->i_mode & S_ISVTX))
1321 return 0;
1322 if (inode->i_uid == current->fsuid)
1323 return 0;
1324 if (dir->i_uid == current->fsuid)
1325 return 0;
1326 return !capable(CAP_FOWNER);
1327}
1328
1329/*
1330 * Check whether we can remove a link victim from directory dir, check
1331 * whether the type of victim is right.
1332 * 1. We can't do it if dir is read-only (done in permission())
1333 * 2. We should have write and exec permissions on dir
1334 * 3. We can't remove anything from append-only dir
1335 * 4. We can't do anything with immutable dir (done in permission())
1336 * 5. If the sticky bit on dir is set we should either
1337 * a. be owner of dir, or
1338 * b. be owner of victim, or
1339 * c. have CAP_FOWNER capability
1340 * 6. If the victim is append-only or immutable we can't do antyhing with
1341 * links pointing to it.
1342 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1343 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1344 * 9. We can't remove a root or mountpoint.
1345 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1346 * nfs_async_unlink().
1347 */
858119e1 1348static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1da177e4
LT		 1349{
1350 int error;
1351
1352 if (!victim->d_inode)
1353 return -ENOENT;
1354
1355 BUG_ON(victim->d_parent->d_inode != dir);
1356
1357 error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
1358 if (error)
1359 return error;
1360 if (IS_APPEND(dir))
1361 return -EPERM;
1362 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1363 IS_IMMUTABLE(victim->d_inode))
1364 return -EPERM;
1365 if (isdir) {
1366 if (!S_ISDIR(victim->d_inode->i_mode))
1367 return -ENOTDIR;
1368 if (IS_ROOT(victim))
1369 return -EBUSY;
1370 } else if (S_ISDIR(victim->d_inode->i_mode))
1371 return -EISDIR;
1372 if (IS_DEADDIR(dir))
1373 return -ENOENT;
1374 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1375 return -EBUSY;
1376 return 0;
1377}
1378
1379/* Check whether we can create an object with dentry child in directory
1380 * dir.
1381 * 1. We can't do it if child already exists (open has special treatment for
1382 * this case, but since we are inlined it's OK)
1383 * 2. We can't do it if dir is read-only (done in permission())
1384 * 3. We should have write and exec permissions on dir
1385 * 4. We can't do it if dir is immutable (done in permission())
1386 */
1387static inline int may_create(struct inode *dir, struct dentry *child,
1388 struct nameidata *nd)
1389{
1390 if (child->d_inode)
1391 return -EEXIST;
1392 if (IS_DEADDIR(dir))
1393 return -ENOENT;
1394 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1395}
1396
1397/*
1da177e4
LT		 1398 * O_DIRECTORY translates into forcing a directory lookup.
1399 */
1400static inline int lookup_flags(unsigned int f)
1401{
1402 unsigned long retval = LOOKUP_FOLLOW;
1403
1404 if (f & O_NOFOLLOW)
1405 retval &= ~LOOKUP_FOLLOW;
1406
1da177e4
LT		 1407 if (f & O_DIRECTORY)
1408 retval |= LOOKUP_DIRECTORY;
1409
1410 return retval;
1411}
1412
1413/*
1414 * p1 and p2 should be directories on the same fs.
1415 */
1416struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1417{
1418 struct dentry *p;
1419
1420 if (p1 == p2) {
1b1dcc1b 1421 mutex_lock(&p1->d_inode->i_mutex);
1da177e4
LT		 1422 return NULL;
1423 }
1424
a11f3a05 1425 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4
LT		 1426
1427 for (p = p1; p->d_parent != p; p = p->d_parent) {
1428 if (p->d_parent == p2) {
1b1dcc1b
JS		 1429 mutex_lock(&p2->d_inode->i_mutex);
1430 mutex_lock(&p1->d_inode->i_mutex);
1da177e4
LT		 1431 return p;
1432 }
1433 }
1434
1435 for (p = p2; p->d_parent != p; p = p->d_parent) {
1436 if (p->d_parent == p1) {
1b1dcc1b
JS		 1437 mutex_lock(&p1->d_inode->i_mutex);
1438 mutex_lock(&p2->d_inode->i_mutex);
1da177e4
LT		 1439 return p;
1440 }
1441 }
1442
1b1dcc1b
JS		 1443 mutex_lock(&p1->d_inode->i_mutex);
1444 mutex_lock(&p2->d_inode->i_mutex);
1da177e4
LT		 1445 return NULL;
1446}
1447
1448void unlock_rename(struct dentry *p1, struct dentry *p2)
1449{
1b1dcc1b 1450 mutex_unlock(&p1->d_inode->i_mutex);
1da177e4 1451 if (p1 != p2) {
1b1dcc1b 1452 mutex_unlock(&p2->d_inode->i_mutex);
a11f3a05 1453 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4
LT		 1454 }
1455}
1456
1457int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1458 struct nameidata *nd)
1459{
1460 int error = may_create(dir, dentry, nd);
1461
1462 if (error)
1463 return error;
1464
1465 if (!dir->i_op || !dir->i_op->create)
1466 return -EACCES; /* shouldn't it be ENOSYS? */
1467 mode &= S_IALLUGO;
1468 mode |= S_IFREG;
1469 error = security_inode_create(dir, dentry, mode);
1470 if (error)
1471 return error;
1472 DQUOT_INIT(dir);
1473 error = dir->i_op->create(dir, dentry, mode, nd);
a74574aa 1474 if (!error)
0eeca283 1475 fsnotify_create(dir, dentry->d_name.name);
1da177e4
LT		 1476 return error;
1477}
1478
1479int may_open(struct nameidata *nd, int acc_mode, int flag)
1480{
1481 struct dentry *dentry = nd->dentry;
1482 struct inode *inode = dentry->d_inode;
1483 int error;
1484
1485 if (!inode)
1486 return -ENOENT;
1487
1488 if (S_ISLNK(inode->i_mode))
1489 return -ELOOP;
1490
1491 if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1492 return -EISDIR;
1493
e4543edd 1494 error = vfs_permission(nd, acc_mode);
1da177e4
LT		 1495 if (error)
1496 return error;
1497
1498 /*
1499 * FIFO's, sockets and device files are special: they don't
1500 * actually live on the filesystem itself, and as such you
1501 * can write to them even if the filesystem is read-only.
1502 */
1503 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1504 flag &= ~O_TRUNC;
1505 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1506 if (nd->mnt->mnt_flags & MNT_NODEV)
1507 return -EACCES;
1508
1509 flag &= ~O_TRUNC;
1510 } else if (IS_RDONLY(inode) && (flag & FMODE_WRITE))
1511 return -EROFS;
1512 /*
1513 * An append-only file must be opened in append mode for writing.
1514 */
1515 if (IS_APPEND(inode)) {
1516 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1517 return -EPERM;
1518 if (flag & O_TRUNC)
1519 return -EPERM;
1520 }
1521
1522 /* O_NOATIME can only be set by the owner or superuser */
1523 if (flag & O_NOATIME)
1524 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1525 return -EPERM;
1526
1527 /*
1528 * Ensure there are no outstanding leases on the file.
1529 */
1530 error = break_lease(inode, flag);
1531 if (error)
1532 return error;
1533
1534 if (flag & O_TRUNC) {
1535 error = get_write_access(inode);
1536 if (error)
1537 return error;
1538
1539 /*
1540 * Refuse to truncate files with mandatory locks held on them.
1541 */
1542 error = locks_verify_locked(inode);
1543 if (!error) {
1544 DQUOT_INIT(inode);
1545
4a30131e 1546 error = do_truncate(dentry, 0, ATTR_MTIME|ATTR_CTIME, NULL);
1da177e4
LT		 1547 }
1548 put_write_access(inode);
1549 if (error)
1550 return error;
1551 } else
1552 if (flag & FMODE_WRITE)
1553 DQUOT_INIT(inode);
1554
1555 return 0;
1556}
1557
1558/*
1559 * open_namei()
1560 *
1561 * namei for open - this is in fact almost the whole open-routine.
1562 *
1563 * Note that the low bits of "flag" aren't the same as in the open
1564 * system call - they are 00 - no permissions needed
1565 * 01 - read permission needed
1566 * 10 - write permission needed
1567 * 11 - read/write permissions needed
1568 * which is a lot more logical, and also allows the "no perm" needed
1569 * for symlinks (where the permissions are checked later).
1570 * SMP-safe
1571 */
5590ff0d
UD		 1572int open_namei(int dfd, const char *pathname, int flag,
1573 int mode, struct nameidata *nd)
1da177e4 1574{
834f2a4a 1575 int acc_mode, error;
4e7506e4 1576 struct path path;
1da177e4
LT		 1577 struct dentry *dir;
1578 int count = 0;
1579
1580 acc_mode = ACC_MODE(flag);
1581
834f2a4a
TM		 1582 /* O_TRUNC implies we need access checks for write permissions */
1583 if (flag & O_TRUNC)
1584 acc_mode |= MAY_WRITE;
1585
1da177e4
LT		 1586 /* Allow the LSM permission hook to distinguish append
1587 access from general write access. */
1588 if (flag & O_APPEND)
1589 acc_mode |= MAY_APPEND;
1590
1da177e4
LT		 1591 /*
1592 * The simplest case - just a plain lookup.
1593 */
1594 if (!(flag & O_CREAT)) {
5590ff0d
UD		 1595 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1596 nd, flag);
1da177e4
LT		 1597 if (error)
1598 return error;
1599 goto ok;
1600 }
1601
1602 /*
1603 * Create - we need to know the parent.
1604 */
5590ff0d 1605 error = path_lookup_create(dfd,pathname,LOOKUP_PARENT,nd,flag,mode);
1da177e4
LT		 1606 if (error)
1607 return error;
1608
1609 /*
1610 * We have the parent and last component. First of all, check
1611 * that we are not asked to creat(2) an obvious directory - that
1612 * will not do.
1613 */
1614 error = -EISDIR;
1615 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1616 goto exit;
1617
1618 dir = nd->dentry;
1619 nd->flags &= ~LOOKUP_PARENT;
1b1dcc1b 1620 mutex_lock(&dir->d_inode->i_mutex);
49705b77 1621 path.dentry = lookup_hash(nd);
d73ffe16 1622 path.mnt = nd->mnt;
1da177e4
LT		 1623
1624do_last:
4e7506e4
AV		 1625 error = PTR_ERR(path.dentry);
1626 if (IS_ERR(path.dentry)) {
1b1dcc1b 1627 mutex_unlock(&dir->d_inode->i_mutex);
1da177e4
LT		 1628 goto exit;
1629 }
1630
4af4c52f
OD		 1631 if (IS_ERR(nd->intent.open.file)) {
1632 mutex_unlock(&dir->d_inode->i_mutex);
1633 error = PTR_ERR(nd->intent.open.file);
1634 goto exit_dput;
1635 }
1636
1da177e4 1637 /* Negative dentry, just create the file */
4e7506e4 1638 if (!path.dentry->d_inode) {
1da177e4
LT		 1639 if (!IS_POSIXACL(dir->d_inode))
1640 mode &= ~current->fs->umask;
4e7506e4 1641 error = vfs_create(dir->d_inode, path.dentry, mode, nd);
1b1dcc1b 1642 mutex_unlock(&dir->d_inode->i_mutex);
1da177e4 1643 dput(nd->dentry);
4e7506e4 1644 nd->dentry = path.dentry;
1da177e4
LT		 1645 if (error)
1646 goto exit;
1647 /* Don't check for write permission, don't truncate */
1648 acc_mode = 0;
1649 flag &= ~O_TRUNC;
1650 goto ok;
1651 }
1652
1653 /*
1654 * It already exists.
1655 */
1b1dcc1b 1656 mutex_unlock(&dir->d_inode->i_mutex);
1da177e4
LT		 1657
1658 error = -EEXIST;
1659 if (flag & O_EXCL)
1660 goto exit_dput;
1661
e13b210f 1662 if (__follow_mount(&path)) {
1da177e4 1663 error = -ELOOP;
ba7a4c1a
AV		 1664 if (flag & O_NOFOLLOW)
1665 goto exit_dput;
1da177e4
LT		 1666 }
1667 error = -ENOENT;
4e7506e4 1668 if (!path.dentry->d_inode)
1da177e4 1669 goto exit_dput;
4e7506e4 1670 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
1da177e4
LT		 1671 goto do_link;
1672
09dd17d3 1673 path_to_nameidata(&path, nd);
1da177e4 1674 error = -EISDIR;
4e7506e4 1675 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1da177e4
LT		 1676 goto exit;
1677ok:
1678 error = may_open(nd, acc_mode, flag);
1679 if (error)
1680 goto exit;
1681 return 0;
1682
1683exit_dput:
09dd17d3 1684 dput_path(&path, nd);
1da177e4 1685exit:
834f2a4a
TM		 1686 if (!IS_ERR(nd->intent.open.file))
1687 release_open_intent(nd);
1da177e4
LT		 1688 path_release(nd);
1689 return error;
1690
1691do_link:
1692 error = -ELOOP;
1693 if (flag & O_NOFOLLOW)
1694 goto exit_dput;
1695 /*
1696 * This is subtle. Instead of calling do_follow_link() we do the
1697 * thing by hands. The reason is that this way we have zero link_count
1698 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1699 * After that we have the parent and last component, i.e.
1700 * we are in the same situation as after the first path_walk().
1701 * Well, almost - if the last component is normal we get its copy
1702 * stored in nd->last.name and we will have to putname() it when we
1703 * are done. Procfs-like symlinks just set LAST_BIND.
1704 */
1705 nd->flags |= LOOKUP_PARENT;
4e7506e4 1706 error = security_inode_follow_link(path.dentry, nd);
1da177e4
LT		 1707 if (error)
1708 goto exit_dput;
cd4e91d3 1709 error = __do_follow_link(&path, nd);
1da177e4
LT		 1710 if (error)
1711 return error;
1712 nd->flags &= ~LOOKUP_PARENT;
d671d5e5 1713 if (nd->last_type == LAST_BIND)
1da177e4 1714 goto ok;
1da177e4
LT		 1715 error = -EISDIR;
1716 if (nd->last_type != LAST_NORM)
1717 goto exit;
1718 if (nd->last.name[nd->last.len]) {
82984114 1719 __putname(nd->last.name);
1da177e4
LT		 1720 goto exit;
1721 }
1722 error = -ELOOP;
1723 if (count++==32) {
82984114 1724 __putname(nd->last.name);
1da177e4
LT		 1725 goto exit;
1726 }
1727 dir = nd->dentry;
1b1dcc1b 1728 mutex_lock(&dir->d_inode->i_mutex);
49705b77 1729 path.dentry = lookup_hash(nd);
d671d5e5 1730 path.mnt = nd->mnt;
82984114 1731 __putname(nd->last.name);
1da177e4
LT		 1732 goto do_last;
1733}
1734
1735/**
1736 * lookup_create - lookup a dentry, creating it if it doesn't exist
1737 * @nd: nameidata info
1738 * @is_dir: directory flag
1739 *
1740 * Simple function to lookup and return a dentry and create it
1741 * if it doesn't exist. Is SMP-safe.
c663e5d8 1742 *
1b1dcc1b 1743 * Returns with nd->dentry->d_inode->i_mutex locked.
1da177e4
LT		 1744 */
1745struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1746{
c663e5d8 1747 struct dentry *dentry = ERR_PTR(-EEXIST);
1da177e4 1748
1b1dcc1b 1749 mutex_lock(&nd->dentry->d_inode->i_mutex);
c663e5d8
CH		 1750 /*
1751 * Yucky last component or no last component at all?
1752 * (foo/., foo/.., /////)
1753 */
1da177e4
LT		 1754 if (nd->last_type != LAST_NORM)
1755 goto fail;
1756 nd->flags &= ~LOOKUP_PARENT;
c663e5d8
CH		 1757
1758 /*
1759 * Do the final lookup.
1760 */
49705b77 1761 dentry = lookup_hash(nd);
1da177e4
LT		 1762 if (IS_ERR(dentry))
1763 goto fail;
c663e5d8
CH		 1764
1765 /*
1766 * Special case - lookup gave negative, but... we had foo/bar/
1767 * From the vfs_mknod() POV we just have a negative dentry -
1768 * all is fine. Let's be bastards - you had / on the end, you've
1769 * been asking for (non-existent) directory. -ENOENT for you.
1770 */
1da177e4
LT		 1771 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1772 goto enoent;
1773 return dentry;
1774enoent:
1775 dput(dentry);
1776 dentry = ERR_PTR(-ENOENT);
1777fail:
1778 return dentry;
1779}
f81a0bff 1780EXPORT_SYMBOL_GPL(lookup_create);
1da177e4
LT		 1781
1782int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1783{
1784 int error = may_create(dir, dentry, NULL);
1785
1786 if (error)
1787 return error;
1788
1789 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1790 return -EPERM;
1791
1792 if (!dir->i_op || !dir->i_op->mknod)
1793 return -EPERM;
1794
1795 error = security_inode_mknod(dir, dentry, mode, dev);
1796 if (error)
1797 return error;
1798
1799 DQUOT_INIT(dir);
1800 error = dir->i_op->mknod(dir, dentry, mode, dev);
a74574aa 1801 if (!error)
0eeca283 1802 fsnotify_create(dir, dentry->d_name.name);
1da177e4
LT		 1803 return error;
1804}
1805
5590ff0d
UD		 1806asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
1807 unsigned dev)
1da177e4
LT		 1808{
1809 int error = 0;
1810 char * tmp;
1811 struct dentry * dentry;
1812 struct nameidata nd;
1813
1814 if (S_ISDIR(mode))
1815 return -EPERM;
1816 tmp = getname(filename);
1817 if (IS_ERR(tmp))
1818 return PTR_ERR(tmp);
1819
5590ff0d 1820 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
1da177e4
LT		 1821 if (error)
1822 goto out;
1823 dentry = lookup_create(&nd, 0);
1824 error = PTR_ERR(dentry);
1825
1826 if (!IS_POSIXACL(nd.dentry->d_inode))
1827 mode &= ~current->fs->umask;
1828 if (!IS_ERR(dentry)) {
1829 switch (mode & S_IFMT) {
1830 case 0: case S_IFREG:
1831 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1832 break;
1833 case S_IFCHR: case S_IFBLK:
1834 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,
1835 new_decode_dev(dev));
1836 break;
1837 case S_IFIFO: case S_IFSOCK:
1838 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,0);
1839 break;
1840 case S_IFDIR:
1841 error = -EPERM;
1842 break;
1843 default:
1844 error = -EINVAL;
1845 }
1846 dput(dentry);
1847 }
1b1dcc1b 1848 mutex_unlock(&nd.dentry->d_inode->i_mutex);
1da177e4
LT		 1849 path_release(&nd);
1850out:
1851 putname(tmp);
1852
1853 return error;
1854}
1855
5590ff0d
UD		 1856asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
1857{
1858 return sys_mknodat(AT_FDCWD, filename, mode, dev);
1859}
1860
1da177e4
LT		 1861int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1862{
1863 int error = may_create(dir, dentry, NULL);
1864
1865 if (error)
1866 return error;
1867
1868 if (!dir->i_op || !dir->i_op->mkdir)
1869 return -EPERM;
1870
1871 mode &= (S_IRWXUGO|S_ISVTX);
1872 error = security_inode_mkdir(dir, dentry, mode);
1873 if (error)
1874 return error;
1875
1876 DQUOT_INIT(dir);
1877 error = dir->i_op->mkdir(dir, dentry, mode);
a74574aa 1878 if (!error)
0eeca283 1879 fsnotify_mkdir(dir, dentry->d_name.name);
1da177e4
LT		 1880 return error;
1881}
1882
5590ff0d 1883asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
1da177e4
LT		 1884{
1885 int error = 0;
1886 char * tmp;
1887
1888 tmp = getname(pathname);
1889 error = PTR_ERR(tmp);
1890 if (!IS_ERR(tmp)) {
1891 struct dentry *dentry;
1892 struct nameidata nd;
1893
5590ff0d 1894 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
1da177e4
LT		 1895 if (error)
1896 goto out;
1897 dentry = lookup_create(&nd, 1);
1898 error = PTR_ERR(dentry);
1899 if (!IS_ERR(dentry)) {
1900 if (!IS_POSIXACL(nd.dentry->d_inode))
1901 mode &= ~current->fs->umask;
1902 error = vfs_mkdir(nd.dentry->d_inode, dentry, mode);
1903 dput(dentry);
1904 }
1b1dcc1b 1905 mutex_unlock(&nd.dentry->d_inode->i_mutex);
1da177e4
LT		 1906 path_release(&nd);
1907out:
1908 putname(tmp);
1909 }
1910
1911 return error;
1912}
1913
5590ff0d
UD		 1914asmlinkage long sys_mkdir(const char __user *pathname, int mode)
1915{
1916 return sys_mkdirat(AT_FDCWD, pathname, mode);
1917}
1918
1da177e4
LT		 1919/*
1920 * We try to drop the dentry early: we should have
1921 * a usage count of 2 if we're the only user of this
1922 * dentry, and if that is true (possibly after pruning
1923 * the dcache), then we drop the dentry now.
1924 *
1925 * A low-level filesystem can, if it choses, legally
1926 * do a
1927 *
1928 * if (!d_unhashed(dentry))
1929 * return -EBUSY;
1930 *
1931 * if it cannot handle the case of removing a directory
1932 * that is still in use by something else..
1933 */
1934void dentry_unhash(struct dentry *dentry)
1935{
1936 dget(dentry);
1937 if (atomic_read(&dentry->d_count))
1938 shrink_dcache_parent(dentry);
1939 spin_lock(&dcache_lock);
1940 spin_lock(&dentry->d_lock);
1941 if (atomic_read(&dentry->d_count) == 2)
1942 __d_drop(dentry);
1943 spin_unlock(&dentry->d_lock);
1944 spin_unlock(&dcache_lock);
1945}
1946
1947int vfs_rmdir(struct inode *dir, struct dentry *dentry)
1948{
1949 int error = may_delete(dir, dentry, 1);
1950
1951 if (error)
1952 return error;
1953
1954 if (!dir->i_op || !dir->i_op->rmdir)
1955 return -EPERM;
1956
1957 DQUOT_INIT(dir);
1958
1b1dcc1b 1959 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 1960 dentry_unhash(dentry);
1961 if (d_mountpoint(dentry))
1962 error = -EBUSY;
1963 else {
1964 error = security_inode_rmdir(dir, dentry);
1965 if (!error) {
1966 error = dir->i_op->rmdir(dir, dentry);
1967 if (!error)
1968 dentry->d_inode->i_flags |= S_DEAD;
1969 }
1970 }
1b1dcc1b 1971 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4 1972 if (!error) {
1da177e4
LT		 1973 d_delete(dentry);
1974 }
1975 dput(dentry);
1976
1977 return error;
1978}
1979
5590ff0d 1980static long do_rmdir(int dfd, const char __user *pathname)
1da177e4
LT		 1981{
1982 int error = 0;
1983 char * name;
1984 struct dentry *dentry;
1985 struct nameidata nd;
1986
1987 name = getname(pathname);
1988 if(IS_ERR(name))
1989 return PTR_ERR(name);
1990
5590ff0d 1991 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
1da177e4
LT		 1992 if (error)
1993 goto exit;
1994
1995 switch(nd.last_type) {
1996 case LAST_DOTDOT:
1997 error = -ENOTEMPTY;
1998 goto exit1;
1999 case LAST_DOT:
2000 error = -EINVAL;
2001 goto exit1;
2002 case LAST_ROOT:
2003 error = -EBUSY;
2004 goto exit1;
2005 }
1b1dcc1b 2006 mutex_lock(&nd.dentry->d_inode->i_mutex);
49705b77 2007 dentry = lookup_hash(&nd);
1da177e4
LT		 2008 error = PTR_ERR(dentry);
2009 if (!IS_ERR(dentry)) {
2010 error = vfs_rmdir(nd.dentry->d_inode, dentry);
2011 dput(dentry);
2012 }
1b1dcc1b 2013 mutex_unlock(&nd.dentry->d_inode->i_mutex);
1da177e4
LT		 2014exit1:
2015 path_release(&nd);
2016exit:
2017 putname(name);
2018 return error;
2019}
2020
5590ff0d
UD		 2021asmlinkage long sys_rmdir(const char __user *pathname)
2022{
2023 return do_rmdir(AT_FDCWD, pathname);
2024}
2025
1da177e4
LT		 2026int vfs_unlink(struct inode *dir, struct dentry *dentry)
2027{
2028 int error = may_delete(dir, dentry, 0);
2029
2030 if (error)
2031 return error;
2032
2033 if (!dir->i_op || !dir->i_op->unlink)
2034 return -EPERM;
2035
2036 DQUOT_INIT(dir);
2037
1b1dcc1b 2038 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2039 if (d_mountpoint(dentry))
2040 error = -EBUSY;
2041 else {
2042 error = security_inode_unlink(dir, dentry);
2043 if (!error)
2044 error = dir->i_op->unlink(dir, dentry);
2045 }
1b1dcc1b 2046 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2047
2048 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2049 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
e234f35c 2050 d_delete(dentry);
1da177e4 2051 }
0eeca283 2052
1da177e4
LT		 2053 return error;
2054}
2055
2056/*
2057 * Make sure that the actual truncation of the file will occur outside its
1b1dcc1b 2058 * directory's i_mutex. Truncate can take a long time if there is a lot of
1da177e4
LT		 2059 * writeout happening, and we don't want to prevent access to the directory
2060 * while waiting on the I/O.
2061 */
5590ff0d 2062static long do_unlinkat(int dfd, const char __user *pathname)
1da177e4
LT		 2063{
2064 int error = 0;
2065 char * name;
2066 struct dentry *dentry;
2067 struct nameidata nd;
2068 struct inode *inode = NULL;
2069
2070 name = getname(pathname);
2071 if(IS_ERR(name))
2072 return PTR_ERR(name);
2073
5590ff0d 2074 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
1da177e4
LT		 2075 if (error)
2076 goto exit;
2077 error = -EISDIR;
2078 if (nd.last_type != LAST_NORM)
2079 goto exit1;
1b1dcc1b 2080 mutex_lock(&nd.dentry->d_inode->i_mutex);
49705b77 2081 dentry = lookup_hash(&nd);
1da177e4
LT		 2082 error = PTR_ERR(dentry);
2083 if (!IS_ERR(dentry)) {
2084 /* Why not before? Because we want correct error value */
2085 if (nd.last.name[nd.last.len])
2086 goto slashes;
2087 inode = dentry->d_inode;
2088 if (inode)
2089 atomic_inc(&inode->i_count);
2090 error = vfs_unlink(nd.dentry->d_inode, dentry);
2091 exit2:
2092 dput(dentry);
2093 }
1b1dcc1b 2094 mutex_unlock(&nd.dentry->d_inode->i_mutex);
1da177e4
LT		 2095 if (inode)
2096 iput(inode); /* truncate the inode here */
2097exit1:
2098 path_release(&nd);
2099exit:
2100 putname(name);
2101 return error;
2102
2103slashes:
2104 error = !dentry->d_inode ? -ENOENT :
2105 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2106 goto exit2;
2107}
2108
5590ff0d
UD		 2109asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2110{
2111 if ((flag & ~AT_REMOVEDIR) != 0)
2112 return -EINVAL;
2113
2114 if (flag & AT_REMOVEDIR)
2115 return do_rmdir(dfd, pathname);
2116
2117 return do_unlinkat(dfd, pathname);
2118}
2119
2120asmlinkage long sys_unlink(const char __user *pathname)
2121{
2122 return do_unlinkat(AT_FDCWD, pathname);
2123}
2124
1da177e4
LT		 2125int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname, int mode)
2126{
2127 int error = may_create(dir, dentry, NULL);
2128
2129 if (error)
2130 return error;
2131
2132 if (!dir->i_op || !dir->i_op->symlink)
2133 return -EPERM;
2134
2135 error = security_inode_symlink(dir, dentry, oldname);
2136 if (error)
2137 return error;
2138
2139 DQUOT_INIT(dir);
2140 error = dir->i_op->symlink(dir, dentry, oldname);
a74574aa 2141 if (!error)
0eeca283 2142 fsnotify_create(dir, dentry->d_name.name);
1da177e4
LT		 2143 return error;
2144}
2145
5590ff0d
UD		 2146asmlinkage long sys_symlinkat(const char __user *oldname,
2147 int newdfd, const char __user *newname)
1da177e4
LT		 2148{
2149 int error = 0;
2150 char * from;
2151 char * to;
2152
2153 from = getname(oldname);
2154 if(IS_ERR(from))
2155 return PTR_ERR(from);
2156 to = getname(newname);
2157 error = PTR_ERR(to);
2158 if (!IS_ERR(to)) {
2159 struct dentry *dentry;
2160 struct nameidata nd;
2161
5590ff0d 2162 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
1da177e4
LT		 2163 if (error)
2164 goto out;
2165 dentry = lookup_create(&nd, 0);
2166 error = PTR_ERR(dentry);
2167 if (!IS_ERR(dentry)) {
2168 error = vfs_symlink(nd.dentry->d_inode, dentry, from, S_IALLUGO);
2169 dput(dentry);
2170 }
1b1dcc1b 2171 mutex_unlock(&nd.dentry->d_inode->i_mutex);
1da177e4
LT		 2172 path_release(&nd);
2173out:
2174 putname(to);
2175 }
2176 putname(from);
2177 return error;
2178}
2179
5590ff0d
UD		 2180asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2181{
2182 return sys_symlinkat(oldname, AT_FDCWD, newname);
2183}
2184
1da177e4
LT		 2185int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2186{
2187 struct inode *inode = old_dentry->d_inode;
2188 int error;
2189
2190 if (!inode)
2191 return -ENOENT;
2192
2193 error = may_create(dir, new_dentry, NULL);
2194 if (error)
2195 return error;
2196
2197 if (dir->i_sb != inode->i_sb)
2198 return -EXDEV;
2199
2200 /*
2201 * A link to an append-only or immutable file cannot be created.
2202 */
2203 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2204 return -EPERM;
2205 if (!dir->i_op || !dir->i_op->link)
2206 return -EPERM;
2207 if (S_ISDIR(old_dentry->d_inode->i_mode))
2208 return -EPERM;
2209
2210 error = security_inode_link(old_dentry, dir, new_dentry);
2211 if (error)
2212 return error;
2213
1b1dcc1b 2214 mutex_lock(&old_dentry->d_inode->i_mutex);
1da177e4
LT		 2215 DQUOT_INIT(dir);
2216 error = dir->i_op->link(old_dentry, dir, new_dentry);
1b1dcc1b 2217 mutex_unlock(&old_dentry->d_inode->i_mutex);
e31e14ec 2218 if (!error)
0eeca283 2219 fsnotify_create(dir, new_dentry->d_name.name);
1da177e4
LT		 2220 return error;
2221}
2222
2223/*
2224 * Hardlinks are often used in delicate situations. We avoid
2225 * security-related surprises by not following symlinks on the
2226 * newname. --KAB
2227 *
2228 * We don't follow them on the oldname either to be compatible
2229 * with linux 2.0, and to avoid hard-linking to directories
2230 * and other special files. --ADM
2231 */
5590ff0d 2232asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
c04030e1
UD		 2233 int newdfd, const char __user *newname,
2234 int flags)
1da177e4
LT		 2235{
2236 struct dentry *new_dentry;
2237 struct nameidata nd, old_nd;
2238 int error;
2239 char * to;
2240
c04030e1
UD		 2241 if (flags != 0)
2242 return -EINVAL;
2243
1da177e4
LT		 2244 to = getname(newname);
2245 if (IS_ERR(to))
2246 return PTR_ERR(to);
2247
5590ff0d 2248 error = __user_walk_fd(olddfd, oldname, 0, &old_nd);
1da177e4
LT		 2249 if (error)
2250 goto exit;
5590ff0d 2251 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
1da177e4
LT		 2252 if (error)
2253 goto out;
2254 error = -EXDEV;
2255 if (old_nd.mnt != nd.mnt)
2256 goto out_release;
2257 new_dentry = lookup_create(&nd, 0);
2258 error = PTR_ERR(new_dentry);
2259 if (!IS_ERR(new_dentry)) {
2260 error = vfs_link(old_nd.dentry, nd.dentry->d_inode, new_dentry);
2261 dput(new_dentry);
2262 }
1b1dcc1b 2263 mutex_unlock(&nd.dentry->d_inode->i_mutex);
1da177e4
LT		 2264out_release:
2265 path_release(&nd);
2266out:
2267 path_release(&old_nd);
2268exit:
2269 putname(to);
2270
2271 return error;
2272}
2273
5590ff0d
UD		 2274asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2275{
c04030e1 2276 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
5590ff0d
UD		 2277}
2278
1da177e4
LT		 2279/*
2280 * The worst of all namespace operations - renaming directory. "Perverted"
2281 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2282 * Problems:
2283 * a) we can get into loop creation. Check is done in is_subdir().
2284 * b) race potential - two innocent renames can create a loop together.
2285 * That's where 4.4 screws up. Current fix: serialization on
a11f3a05 2286 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
1da177e4
LT		 2287 * story.
2288 * c) we have to lock _three_ objects - parents and victim (if it exists).
1b1dcc1b 2289 * And that - after we got ->i_mutex on parents (until then we don't know
1da177e4
LT		 2290 * whether the target exists). Solution: try to be smart with locking
2291 * order for inodes. We rely on the fact that tree topology may change
a11f3a05 2292 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
1da177e4
LT		 2293 * move will be locked. Thus we can rank directories by the tree
2294 * (ancestors first) and rank all non-directories after them.
2295 * That works since everybody except rename does "lock parent, lookup,
a11f3a05 2296 * lock child" and rename is under ->s_vfs_rename_mutex.
1da177e4
LT		 2297 * HOWEVER, it relies on the assumption that any object with ->lookup()
2298 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2299 * we'd better make sure that there's no link(2) for them.
2300 * d) some filesystems don't support opened-but-unlinked directories,
2301 * either because of layout or because they are not ready to deal with
2302 * all cases correctly. The latter will be fixed (taking this sort of
2303 * stuff into VFS), but the former is not going away. Solution: the same
2304 * trick as in rmdir().
2305 * e) conversion from fhandle to dentry may come in the wrong moment - when
1b1dcc1b 2306 * we are removing the target. Solution: we will have to grab ->i_mutex
1da177e4 2307 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
1b1dcc1b 2308 * ->i_mutex on parents, which works but leads to some truely excessive
1da177e4
LT		 2309 * locking].
2310 */
75c96f85
AB		 2311static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2312 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 2313{
2314 int error = 0;
2315 struct inode *target;
2316
2317 /*
2318 * If we are going to change the parent - check write permissions,
2319 * we'll need to flip '..'.
2320 */
2321 if (new_dir != old_dir) {
2322 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2323 if (error)
2324 return error;
2325 }
2326
2327 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2328 if (error)
2329 return error;
2330
2331 target = new_dentry->d_inode;
2332 if (target) {
1b1dcc1b 2333 mutex_lock(&target->i_mutex);
1da177e4
LT		 2334 dentry_unhash(new_dentry);
2335 }
2336 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2337 error = -EBUSY;
2338 else
2339 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2340 if (target) {
2341 if (!error)
2342 target->i_flags |= S_DEAD;
1b1dcc1b 2343 mutex_unlock(&target->i_mutex);
1da177e4
LT		 2344 if (d_unhashed(new_dentry))
2345 d_rehash(new_dentry);
2346 dput(new_dentry);
2347 }
e31e14ec 2348 if (!error)
1da177e4 2349 d_move(old_dentry,new_dentry);
1da177e4
LT		 2350 return error;
2351}
2352
75c96f85
AB		 2353static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2354 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 2355{
2356 struct inode *target;
2357 int error;
2358
2359 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2360 if (error)
2361 return error;
2362
2363 dget(new_dentry);
2364 target = new_dentry->d_inode;
2365 if (target)
1b1dcc1b 2366 mutex_lock(&target->i_mutex);
1da177e4
LT		 2367 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2368 error = -EBUSY;
2369 else
2370 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2371 if (!error) {
2372 /* The following d_move() should become unconditional */
2373 if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
2374 d_move(old_dentry, new_dentry);
1da177e4
LT		 2375 }
2376 if (target)
1b1dcc1b 2377 mutex_unlock(&target->i_mutex);
1da177e4
LT		 2378 dput(new_dentry);
2379 return error;
2380}
2381
2382int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2383 struct inode *new_dir, struct dentry *new_dentry)
2384{
2385 int error;
2386 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
0eeca283 2387 const char *old_name;
1da177e4
LT		 2388
2389 if (old_dentry->d_inode == new_dentry->d_inode)
2390 return 0;
2391
2392 error = may_delete(old_dir, old_dentry, is_dir);
2393 if (error)
2394 return error;
2395
2396 if (!new_dentry->d_inode)
2397 error = may_create(new_dir, new_dentry, NULL);
2398 else
2399 error = may_delete(new_dir, new_dentry, is_dir);
2400 if (error)
2401 return error;
2402
2403 if (!old_dir->i_op || !old_dir->i_op->rename)
2404 return -EPERM;
2405
2406 DQUOT_INIT(old_dir);
2407 DQUOT_INIT(new_dir);
2408
0eeca283
RL		 2409 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2410
1da177e4
LT		 2411 if (is_dir)
2412 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2413 else
2414 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2415 if (!error) {
0eeca283 2416 const char *new_name = old_dentry->d_name.name;
89204c40
JM		 2417 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2418 new_dentry->d_inode, old_dentry->d_inode);
1da177e4 2419 }
0eeca283
RL		 2420 fsnotify_oldname_free(old_name);
2421
1da177e4
LT		 2422 return error;
2423}
2424
5590ff0d
UD		 2425static int do_rename(int olddfd, const char *oldname,
2426 int newdfd, const char *newname)
1da177e4
LT		 2427{
2428 int error = 0;
2429 struct dentry * old_dir, * new_dir;
2430 struct dentry * old_dentry, *new_dentry;
2431 struct dentry * trap;
2432 struct nameidata oldnd, newnd;
2433
5590ff0d 2434 error = do_path_lookup(olddfd, oldname, LOOKUP_PARENT, &oldnd);
1da177e4
LT		 2435 if (error)
2436 goto exit;
2437
5590ff0d 2438 error = do_path_lookup(newdfd, newname, LOOKUP_PARENT, &newnd);
1da177e4
LT		 2439 if (error)
2440 goto exit1;
2441
2442 error = -EXDEV;
2443 if (oldnd.mnt != newnd.mnt)
2444 goto exit2;
2445
2446 old_dir = oldnd.dentry;
2447 error = -EBUSY;
2448 if (oldnd.last_type != LAST_NORM)
2449 goto exit2;
2450
2451 new_dir = newnd.dentry;
2452 if (newnd.last_type != LAST_NORM)
2453 goto exit2;
2454
2455 trap = lock_rename(new_dir, old_dir);
2456
49705b77 2457 old_dentry = lookup_hash(&oldnd);
1da177e4
LT		 2458 error = PTR_ERR(old_dentry);
2459 if (IS_ERR(old_dentry))
2460 goto exit3;
2461 /* source must exist */
2462 error = -ENOENT;
2463 if (!old_dentry->d_inode)
2464 goto exit4;
2465 /* unless the source is a directory trailing slashes give -ENOTDIR */
2466 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2467 error = -ENOTDIR;
2468 if (oldnd.last.name[oldnd.last.len])
2469 goto exit4;
2470 if (newnd.last.name[newnd.last.len])
2471 goto exit4;
2472 }
2473 /* source should not be ancestor of target */
2474 error = -EINVAL;
2475 if (old_dentry == trap)
2476 goto exit4;
49705b77 2477 new_dentry = lookup_hash(&newnd);
1da177e4
LT		 2478 error = PTR_ERR(new_dentry);
2479 if (IS_ERR(new_dentry))
2480 goto exit4;
2481 /* target should not be an ancestor of source */
2482 error = -ENOTEMPTY;
2483 if (new_dentry == trap)
2484 goto exit5;
2485
2486 error = vfs_rename(old_dir->d_inode, old_dentry,
2487 new_dir->d_inode, new_dentry);
2488exit5:
2489 dput(new_dentry);
2490exit4:
2491 dput(old_dentry);
2492exit3:
2493 unlock_rename(new_dir, old_dir);
2494exit2:
2495 path_release(&newnd);
2496exit1:
2497 path_release(&oldnd);
2498exit:
2499 return error;
2500}
2501
5590ff0d
UD		 2502asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2503 int newdfd, const char __user *newname)
1da177e4
LT		 2504{
2505 int error;
2506 char * from;
2507 char * to;
2508
2509 from = getname(oldname);
2510 if(IS_ERR(from))
2511 return PTR_ERR(from);
2512 to = getname(newname);
2513 error = PTR_ERR(to);
2514 if (!IS_ERR(to)) {
5590ff0d 2515 error = do_rename(olddfd, from, newdfd, to);
1da177e4
LT		 2516 putname(to);
2517 }
2518 putname(from);
2519 return error;
2520}
2521
5590ff0d
UD		 2522asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2523{
2524 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2525}
2526
1da177e4
LT		 2527int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2528{
2529 int len;
2530
2531 len = PTR_ERR(link);
2532 if (IS_ERR(link))
2533 goto out;
2534
2535 len = strlen(link);
2536 if (len > (unsigned) buflen)
2537 len = buflen;
2538 if (copy_to_user(buffer, link, len))
2539 len = -EFAULT;
2540out:
2541 return len;
2542}
2543
2544/*
2545 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2546 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2547 * using) it for any given inode is up to filesystem.
2548 */
2549int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2550{
2551 struct nameidata nd;
cc314eef
LT		 2552 void *cookie;
2553
1da177e4 2554 nd.depth = 0;
cc314eef
LT		 2555 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2556 if (!IS_ERR(cookie)) {
2557 int res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
1da177e4 2558 if (dentry->d_inode->i_op->put_link)
cc314eef
LT		 2559 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2560 cookie = ERR_PTR(res);
1da177e4 2561 }
cc314eef 2562 return PTR_ERR(cookie);
1da177e4
LT		 2563}
2564
2565int vfs_follow_link(struct nameidata *nd, const char *link)
2566{
2567 return __vfs_follow_link(nd, link);
2568}
2569
2570/* get the link contents into pagecache */
2571static char *page_getlink(struct dentry * dentry, struct page **ppage)
2572{
2573 struct page * page;
2574 struct address_space *mapping = dentry->d_inode->i_mapping;
2575 page = read_cache_page(mapping, 0, (filler_t *)mapping->a_ops->readpage,
2576 NULL);
2577 if (IS_ERR(page))
2578 goto sync_fail;
2579 wait_on_page_locked(page);
2580 if (!PageUptodate(page))
2581 goto async_fail;
2582 *ppage = page;
2583 return kmap(page);
2584
2585async_fail:
2586 page_cache_release(page);
2587 return ERR_PTR(-EIO);
2588
2589sync_fail:
2590 return (char*)page;
2591}
2592
2593int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2594{
2595 struct page *page = NULL;
2596 char *s = page_getlink(dentry, &page);
2597 int res = vfs_readlink(dentry,buffer,buflen,s);
2598 if (page) {
2599 kunmap(page);
2600 page_cache_release(page);
2601 }
2602 return res;
2603}
2604
cc314eef 2605void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
1da177e4 2606{
cc314eef 2607 struct page *page = NULL;
1da177e4 2608 nd_set_link(nd, page_getlink(dentry, &page));
cc314eef 2609 return page;
1da177e4
LT		 2610}
2611
cc314eef 2612void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1da177e4 2613{
cc314eef
LT		 2614 struct page *page = cookie;
2615
2616 if (page) {
1da177e4
LT		 2617 kunmap(page);
2618 page_cache_release(page);
1da177e4
LT		 2619 }
2620}
2621
0adb25d2
KK		 2622int __page_symlink(struct inode *inode, const char *symname, int len,
2623 gfp_t gfp_mask)
1da177e4
LT		 2624{
2625 struct address_space *mapping = inode->i_mapping;
0adb25d2 2626 struct page *page;
1da177e4
LT		 2627 int err = -ENOMEM;
2628 char *kaddr;
2629
0adb25d2 2630 page = find_or_create_page(mapping, 0, gfp_mask);
1da177e4
LT		 2631 if (!page)
2632 goto fail;
2633 err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
2634 if (err)
2635 goto fail_map;
2636 kaddr = kmap_atomic(page, KM_USER0);
2637 memcpy(kaddr, symname, len-1);
2638 kunmap_atomic(kaddr, KM_USER0);
2639 mapping->a_ops->commit_write(NULL, page, 0, len-1);
2640 /*
2641 * Notice that we are _not_ going to block here - end of page is
2642 * unmapped, so this will only try to map the rest of page, see
2643 * that it is unmapped (typically even will not look into inode -
2644 * ->i_size will be enough for everything) and zero it out.
2645 * OTOH it's obviously correct and should make the page up-to-date.
2646 */
2647 if (!PageUptodate(page)) {
2648 err = mapping->a_ops->readpage(NULL, page);
2649 wait_on_page_locked(page);
2650 } else {
2651 unlock_page(page);
2652 }
2653 page_cache_release(page);
2654 if (err < 0)
2655 goto fail;
2656 mark_inode_dirty(inode);
2657 return 0;
2658fail_map:
2659 unlock_page(page);
2660 page_cache_release(page);
2661fail:
2662 return err;
2663}
2664
0adb25d2
KK		 2665int page_symlink(struct inode *inode, const char *symname, int len)
2666{
2667 return __page_symlink(inode, symname, len,
2668 mapping_gfp_mask(inode->i_mapping));
2669}
2670
1da177e4
LT		 2671struct inode_operations page_symlink_inode_operations = {
2672 .readlink = generic_readlink,
2673 .follow_link = page_follow_link_light,
2674 .put_link = page_put_link,
2675};
2676
2677EXPORT_SYMBOL(__user_walk);
5590ff0d 2678EXPORT_SYMBOL(__user_walk_fd);
1da177e4
LT		 2679EXPORT_SYMBOL(follow_down);
2680EXPORT_SYMBOL(follow_up);
2681EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2682EXPORT_SYMBOL(getname);
2683EXPORT_SYMBOL(lock_rename);
2684EXPORT_SYMBOL(lookup_hash);
2685EXPORT_SYMBOL(lookup_one_len);
2686EXPORT_SYMBOL(page_follow_link_light);
2687EXPORT_SYMBOL(page_put_link);
2688EXPORT_SYMBOL(page_readlink);
0adb25d2 2689EXPORT_SYMBOL(__page_symlink);
1da177e4
LT		 2690EXPORT_SYMBOL(page_symlink);
2691EXPORT_SYMBOL(page_symlink_inode_operations);
2692EXPORT_SYMBOL(path_lookup);
2693EXPORT_SYMBOL(path_release);
2694EXPORT_SYMBOL(path_walk);
2695EXPORT_SYMBOL(permission);
e4543edd 2696EXPORT_SYMBOL(vfs_permission);
8c744fb8 2697EXPORT_SYMBOL(file_permission);
1da177e4
LT		 2698EXPORT_SYMBOL(unlock_rename);
2699EXPORT_SYMBOL(vfs_create);
2700EXPORT_SYMBOL(vfs_follow_link);
2701EXPORT_SYMBOL(vfs_link);
2702EXPORT_SYMBOL(vfs_mkdir);
2703EXPORT_SYMBOL(vfs_mknod);
2704EXPORT_SYMBOL(generic_permission);
2705EXPORT_SYMBOL(vfs_readlink);
2706EXPORT_SYMBOL(vfs_rename);
2707EXPORT_SYMBOL(vfs_rmdir);
2708EXPORT_SYMBOL(vfs_symlink);
2709EXPORT_SYMBOL(vfs_unlink);
2710EXPORT_SYMBOL(dentry_unhash);
2711EXPORT_SYMBOL(generic_readlink);
kernel source for telekom puls (alcatel/mediatek)