projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
vmscan: let GFP_NOFS go to swap again
[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/personality.h>
26#include <linux/security.h>
27#include <linux/syscalls.h>
28#include <linux/mount.h>
29#include <linux/audit.h>
16f7e0fe 30#include <linux/capability.h>
834f2a4a 31#include <linux/file.h>
5590ff0d 32#include <linux/fcntl.h>
08ce5f16 33#include <linux/device_cgroup.h>
1da177e4
LT		 34#include <asm/uaccess.h>
35
36#define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
37
38/* [Feb-1997 T. Schoebel-Theuer]
39 * Fundamental changes in the pathname lookup mechanisms (namei)
40 * were necessary because of omirr. The reason is that omirr needs
41 * to know the _real_ pathname, not the user-supplied one, in case
42 * of symlinks (and also when transname replacements occur).
43 *
44 * The new code replaces the old recursive symlink resolution with
45 * an iterative one (in case of non-nested symlink chains). It does
46 * this with calls to <fs>_follow_link().
47 * As a side effect, dir_namei(), _namei() and follow_link() are now
48 * replaced with a single function lookup_dentry() that can handle all
49 * the special cases of the former code.
50 *
51 * With the new dcache, the pathname is stored at each inode, at least as
52 * long as the refcount of the inode is positive. As a side effect, the
53 * size of the dcache depends on the inode cache and thus is dynamic.
54 *
55 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
56 * resolution to correspond with current state of the code.
57 *
58 * Note that the symlink resolution is not *completely* iterative.
59 * There is still a significant amount of tail- and mid- recursion in
60 * the algorithm. Also, note that <fs>_readlink() is not used in
61 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
62 * may return different results than <fs>_follow_link(). Many virtual
63 * filesystems (including /proc) exhibit this behavior.
64 */
65
66/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
67 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
68 * and the name already exists in form of a symlink, try to create the new
69 * name indicated by the symlink. The old code always complained that the
70 * name already exists, due to not following the symlink even if its target
71 * is nonexistent. The new semantics affects also mknod() and link() when
72 * the name is a symlink pointing to a non-existant name.
73 *
74 * I don't know which semantics is the right one, since I have no access
75 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
76 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
77 * "old" one. Personally, I think the new semantics is much more logical.
78 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
79 * file does succeed in both HP-UX and SunOs, but not in Solaris
80 * and in the old Linux semantics.
81 */
82
83/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
84 * semantics. See the comments in "open_namei" and "do_link" below.
85 *
86 * [10-Sep-98 Alan Modra] Another symlink change.
87 */
88
89/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
90 * inside the path - always follow.
91 * in the last component in creation/removal/renaming - never follow.
92 * if LOOKUP_FOLLOW passed - follow.
93 * if the pathname has trailing slashes - follow.
94 * otherwise - don't follow.
95 * (applied in that order).
96 *
97 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
98 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
99 * During the 2.4 we need to fix the userland stuff depending on it -
100 * hopefully we will be able to get rid of that wart in 2.5. So far only
101 * XEmacs seems to be relying on it...
102 */
103/*
104 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
a11f3a05 105 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
1da177e4
LT		 106 * any extra contention...
107 */
108
a02f76c3 109static int __link_path_walk(const char *name, struct nameidata *nd);
c4a7808f 110
1da177e4
LT		 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{
5ac3a9c2 162 if (unlikely(!audit_dummy_context()))
1da177e4
LT		 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
e6305c43
AV		 187 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
188
1da177e4
LT		 189 if (current->fsuid == inode->i_uid)
190 mode >>= 6;
191 else {
192 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
193 int error = check_acl(inode, mask);
194 if (error == -EACCES)
195 goto check_capabilities;
196 else if (error != -EAGAIN)
197 return error;
198 }
199
200 if (in_group_p(inode->i_gid))
201 mode >>= 3;
202 }
203
204 /*
205 * If the DACs are ok we don't need any capability check.
206 */
e6305c43 207 if ((mask & ~mode) == 0)
1da177e4
LT		 208 return 0;
209
210 check_capabilities:
211 /*
212 * Read/write DACs are always overridable.
213 * Executable DACs are overridable if at least one exec bit is set.
214 */
f696a365 215 if (!(mask & MAY_EXEC) || execute_ok(inode))
1da177e4
LT		 216 if (capable(CAP_DAC_OVERRIDE))
217 return 0;
218
219 /*
220 * Searching includes executable on directories, else just read.
221 */
222 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
223 if (capable(CAP_DAC_READ_SEARCH))
224 return 0;
225
226 return -EACCES;
227}
228
f419a2e3 229int inode_permission(struct inode *inode, int mask)
1da177e4 230{
e6305c43 231 int retval;
1da177e4
LT		 232
233 if (mask & MAY_WRITE) {
22590e41 234 umode_t mode = inode->i_mode;
1da177e4
LT		 235
236 /*
237 * Nobody gets write access to a read-only fs.
238 */
239 if (IS_RDONLY(inode) &&
240 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
241 return -EROFS;
242
243 /*
244 * Nobody gets write access to an immutable file.
245 */
246 if (IS_IMMUTABLE(inode))
247 return -EACCES;
248 }
249
1da177e4 250 /* Ordinary permission routines do not understand MAY_APPEND. */
f696a365 251 if (inode->i_op && inode->i_op->permission)
b77b0646 252 retval = inode->i_op->permission(inode, mask);
f696a365 253 else
e6305c43 254 retval = generic_permission(inode, mask, NULL);
f696a365 255
1da177e4
LT		 256 if (retval)
257 return retval;
258
08ce5f16
SH		 259 retval = devcgroup_inode_permission(inode, mask);
260 if (retval)
261 return retval;
262
e6305c43 263 return security_inode_permission(inode,
f418b006 264 mask & (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND));
1da177e4
LT		 265}
266
e4543edd
CH		 267/**
268 * vfs_permission - check for access rights to a given path
269 * @nd: lookup result that describes the path
270 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
271 *
272 * Used to check for read/write/execute permissions on a path.
273 * We use "fsuid" for this, letting us set arbitrary permissions
274 * for filesystem access without changing the "normal" uids which
275 * are used for other things.
276 */
277int vfs_permission(struct nameidata *nd, int mask)
278{
f419a2e3 279 return inode_permission(nd->path.dentry->d_inode, mask);
e4543edd
CH		 280}
281
8c744fb8
CH		 282/**
283 * file_permission - check for additional access rights to a given file
284 * @file: file to check access rights for
285 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
286 *
287 * Used to check for read/write/execute permissions on an already opened
288 * file.
289 *
290 * Note:
291 * Do not use this function in new code. All access checks should
292 * be done using vfs_permission().
293 */
294int file_permission(struct file *file, int mask)
295{
f419a2e3 296 return inode_permission(file->f_path.dentry->d_inode, mask);
8c744fb8
CH		 297}
298
1da177e4
LT		 299/*
300 * get_write_access() gets write permission for a file.
301 * put_write_access() releases this write permission.
302 * This is used for regular files.
303 * We cannot support write (and maybe mmap read-write shared) accesses and
304 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
305 * can have the following values:
306 * 0: no writers, no VM_DENYWRITE mappings
307 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
308 * > 0: (i_writecount) users are writing to the file.
309 *
310 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
311 * except for the cases where we don't hold i_writecount yet. Then we need to
312 * use {get,deny}_write_access() - these functions check the sign and refuse
313 * to do the change if sign is wrong. Exclusion between them is provided by
314 * the inode->i_lock spinlock.
315 */
316
317int get_write_access(struct inode * inode)
318{
319 spin_lock(&inode->i_lock);
320 if (atomic_read(&inode->i_writecount) < 0) {
321 spin_unlock(&inode->i_lock);
322 return -ETXTBSY;
323 }
324 atomic_inc(&inode->i_writecount);
325 spin_unlock(&inode->i_lock);
326
327 return 0;
328}
329
330int deny_write_access(struct file * file)
331{
0f7fc9e4 332 struct inode *inode = file->f_path.dentry->d_inode;
1da177e4
LT		 333
334 spin_lock(&inode->i_lock);
335 if (atomic_read(&inode->i_writecount) > 0) {
336 spin_unlock(&inode->i_lock);
337 return -ETXTBSY;
338 }
339 atomic_dec(&inode->i_writecount);
340 spin_unlock(&inode->i_lock);
341
342 return 0;
343}
344
5dd784d0
JB		 345/**
346 * path_get - get a reference to a path
347 * @path: path to get the reference to
348 *
349 * Given a path increment the reference count to the dentry and the vfsmount.
350 */
351void path_get(struct path *path)
352{
353 mntget(path->mnt);
354 dget(path->dentry);
355}
356EXPORT_SYMBOL(path_get);
357
1d957f9b
JB		 358/**
359 * path_put - put a reference to a path
360 * @path: path to put the reference to
361 *
362 * Given a path decrement the reference count to the dentry and the vfsmount.
363 */
364void path_put(struct path *path)
1da177e4 365{
1d957f9b
JB		 366 dput(path->dentry);
367 mntput(path->mnt);
1da177e4 368}
1d957f9b 369EXPORT_SYMBOL(path_put);
1da177e4 370
834f2a4a
TM		 371/**
372 * release_open_intent - free up open intent resources
373 * @nd: pointer to nameidata
374 */
375void release_open_intent(struct nameidata *nd)
376{
0f7fc9e4 377 if (nd->intent.open.file->f_path.dentry == NULL)
834f2a4a
TM		 378 put_filp(nd->intent.open.file);
379 else
380 fput(nd->intent.open.file);
381}
382
bcdc5e01
IK		 383static inline struct dentry *
384do_revalidate(struct dentry *dentry, struct nameidata *nd)
385{
386 int status = dentry->d_op->d_revalidate(dentry, nd);
387 if (unlikely(status <= 0)) {
388 /*
389 * The dentry failed validation.
390 * If d_revalidate returned 0 attempt to invalidate
391 * the dentry otherwise d_revalidate is asking us
392 * to return a fail status.
393 */
394 if (!status) {
395 if (!d_invalidate(dentry)) {
396 dput(dentry);
397 dentry = NULL;
398 }
399 } else {
400 dput(dentry);
401 dentry = ERR_PTR(status);
402 }
403 }
404 return dentry;
405}
406
1da177e4
LT		 407/*
408 * Internal lookup() using the new generic dcache.
409 * SMP-safe
410 */
411static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
412{
413 struct dentry * dentry = __d_lookup(parent, name);
414
415 /* lockess __d_lookup may fail due to concurrent d_move()
416 * in some unrelated directory, so try with d_lookup
417 */
418 if (!dentry)
419 dentry = d_lookup(parent, name);
420
bcdc5e01
IK		 421 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
422 dentry = do_revalidate(dentry, nd);
423
1da177e4
LT		 424 return dentry;
425}
426
427/*
428 * Short-cut version of permission(), for calling by
429 * path_walk(), when dcache lock is held. Combines parts
430 * of permission() and generic_permission(), and tests ONLY for
431 * MAY_EXEC permission.
432 *
433 * If appropriate, check DAC only. If not appropriate, or
434 * short-cut DAC fails, then call permission() to do more
435 * complete permission check.
436 */
672b16b2 437static int exec_permission_lite(struct inode *inode)
1da177e4
LT		 438{
439 umode_t mode = inode->i_mode;
440
441 if (inode->i_op && inode->i_op->permission)
442 return -EAGAIN;
443
444 if (current->fsuid == inode->i_uid)
445 mode >>= 6;
446 else if (in_group_p(inode->i_gid))
447 mode >>= 3;
448
449 if (mode & MAY_EXEC)
450 goto ok;
451
452 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
453 goto ok;
454
455 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
456 goto ok;
457
458 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
459 goto ok;
460
461 return -EACCES;
462ok:
b77b0646 463 return security_inode_permission(inode, MAY_EXEC);
1da177e4
LT		 464}
465
466/*
467 * This is called when everything else fails, and we actually have
468 * to go to the low-level filesystem to find out what we should do..
469 *
470 * We get the directory semaphore, and after getting that we also
471 * make sure that nobody added the entry to the dcache in the meantime..
472 * SMP-safe
473 */
474static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
475{
476 struct dentry * result;
477 struct inode *dir = parent->d_inode;
478
1b1dcc1b 479 mutex_lock(&dir->i_mutex);
1da177e4
LT		 480 /*
481 * First re-do the cached lookup just in case it was created
482 * while we waited for the directory semaphore..
483 *
484 * FIXME! This could use version numbering or similar to
485 * avoid unnecessary cache lookups.
486 *
487 * The "dcache_lock" is purely to protect the RCU list walker
488 * from concurrent renames at this point (we mustn't get false
489 * negatives from the RCU list walk here, unlike the optimistic
490 * fast walk).
491 *
492 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
493 */
494 result = d_lookup(parent, name);
495 if (!result) {
d70b67c8
MS		 496 struct dentry *dentry;
497
498 /* Don't create child dentry for a dead directory. */
499 result = ERR_PTR(-ENOENT);
500 if (IS_DEADDIR(dir))
501 goto out_unlock;
502
503 dentry = d_alloc(parent, name);
1da177e4
LT		 504 result = ERR_PTR(-ENOMEM);
505 if (dentry) {
506 result = dir->i_op->lookup(dir, dentry, nd);
507 if (result)
508 dput(dentry);
509 else
510 result = dentry;
511 }
d70b67c8 512out_unlock:
1b1dcc1b 513 mutex_unlock(&dir->i_mutex);
1da177e4
LT		 514 return result;
515 }
516
517 /*
518 * Uhhuh! Nasty case: the cache was re-populated while
519 * we waited on the semaphore. Need to revalidate.
520 */
1b1dcc1b 521 mutex_unlock(&dir->i_mutex);
1da177e4 522 if (result->d_op && result->d_op->d_revalidate) {
bcdc5e01
IK		 523 result = do_revalidate(result, nd);
524 if (!result)
1da177e4 525 result = ERR_PTR(-ENOENT);
1da177e4
LT		 526 }
527 return result;
528}
529
1da177e4 530/* SMP-safe */
7f2da1e7 531static __always_inline void
1da177e4
LT		 532walk_init_root(const char *name, struct nameidata *nd)
533{
e518ddb7
AM		 534 struct fs_struct *fs = current->fs;
535
536 read_lock(&fs->lock);
6ac08c39
JB		 537 nd->path = fs->root;
538 path_get(&fs->root);
e518ddb7 539 read_unlock(&fs->lock);
1da177e4
LT		 540}
541
a02f76c3
AV		 542/*
543 * Wrapper to retry pathname resolution whenever the underlying
544 * file system returns an ESTALE.
545 *
546 * Retry the whole path once, forcing real lookup requests
547 * instead of relying on the dcache.
548 */
549static __always_inline int link_path_walk(const char *name, struct nameidata *nd)
550{
551 struct path save = nd->path;
552 int result;
553
554 /* make sure the stuff we saved doesn't go away */
c8e7f449 555 path_get(&save);
a02f76c3
AV		 556
557 result = __link_path_walk(name, nd);
558 if (result == -ESTALE) {
559 /* nd->path had been dropped */
560 nd->path = save;
c8e7f449 561 path_get(&nd->path);
a02f76c3
AV		 562 nd->flags |= LOOKUP_REVAL;
563 result = __link_path_walk(name, nd);
564 }
565
566 path_put(&save);
567
568 return result;
569}
570
f1662356 571static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
1da177e4
LT		 572{
573 int res = 0;
574 char *name;
575 if (IS_ERR(link))
576 goto fail;
577
578 if (*link == '/') {
1d957f9b 579 path_put(&nd->path);
7f2da1e7 580 walk_init_root(link, nd);
1da177e4
LT		 581 }
582 res = link_path_walk(link, nd);
1da177e4
LT		 583 if (nd->depth || res || nd->last_type!=LAST_NORM)
584 return res;
585 /*
586 * If it is an iterative symlinks resolution in open_namei() we
587 * have to copy the last component. And all that crap because of
588 * bloody create() on broken symlinks. Furrfu...
589 */
590 name = __getname();
591 if (unlikely(!name)) {
1d957f9b 592 path_put(&nd->path);
1da177e4
LT		 593 return -ENOMEM;
594 }
595 strcpy(name, nd->last.name);
596 nd->last.name = name;
597 return 0;
598fail:
1d957f9b 599 path_put(&nd->path);
1da177e4
LT		 600 return PTR_ERR(link);
601}
602
1d957f9b 603static void path_put_conditional(struct path *path, struct nameidata *nd)
051d3812
IK		 604{
605 dput(path->dentry);
4ac91378 606 if (path->mnt != nd->path.mnt)
051d3812
IK		 607 mntput(path->mnt);
608}
609
610static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
611{
4ac91378
JB		 612 dput(nd->path.dentry);
613 if (nd->path.mnt != path->mnt)
614 mntput(nd->path.mnt);
615 nd->path.mnt = path->mnt;
616 nd->path.dentry = path->dentry;
051d3812
IK		 617}
618
f1662356 619static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
1da177e4
LT		 620{
621 int error;
cc314eef 622 void *cookie;
cd4e91d3 623 struct dentry *dentry = path->dentry;
1da177e4 624
d671a1cb 625 touch_atime(path->mnt, dentry);
1da177e4 626 nd_set_link(nd, NULL);
cd4e91d3 627
4ac91378 628 if (path->mnt != nd->path.mnt) {
051d3812
IK		 629 path_to_nameidata(path, nd);
630 dget(dentry);
631 }
632 mntget(path->mnt);
cc314eef
LT		 633 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
634 error = PTR_ERR(cookie);
635 if (!IS_ERR(cookie)) {
1da177e4 636 char *s = nd_get_link(nd);
cc314eef 637 error = 0;
1da177e4
LT		 638 if (s)
639 error = __vfs_follow_link(nd, s);
640 if (dentry->d_inode->i_op->put_link)
cc314eef 641 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
1da177e4 642 }
09da5916 643 path_put(path);
1da177e4
LT		 644
645 return error;
646}
647
648/*
649 * This limits recursive symlink follows to 8, while
650 * limiting consecutive symlinks to 40.
651 *
652 * Without that kind of total limit, nasty chains of consecutive
653 * symlinks can cause almost arbitrarily long lookups.
654 */
90ebe565 655static inline int do_follow_link(struct path *path, struct nameidata *nd)
1da177e4
LT		 656{
657 int err = -ELOOP;
658 if (current->link_count >= MAX_NESTED_LINKS)
659 goto loop;
660 if (current->total_link_count >= 40)
661 goto loop;
662 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
663 cond_resched();
90ebe565 664 err = security_inode_follow_link(path->dentry, nd);
1da177e4
LT		 665 if (err)
666 goto loop;
667 current->link_count++;
668 current->total_link_count++;
669 nd->depth++;
cd4e91d3 670 err = __do_follow_link(path, nd);
839d9f93
AV		 671 current->link_count--;
672 nd->depth--;
1da177e4
LT		 673 return err;
674loop:
1d957f9b
JB		 675 path_put_conditional(path, nd);
676 path_put(&nd->path);
1da177e4
LT		 677 return err;
678}
679
680int follow_up(struct vfsmount **mnt, struct dentry **dentry)
681{
682 struct vfsmount *parent;
683 struct dentry *mountpoint;
684 spin_lock(&vfsmount_lock);
685 parent=(*mnt)->mnt_parent;
686 if (parent == *mnt) {
687 spin_unlock(&vfsmount_lock);
688 return 0;
689 }
690 mntget(parent);
691 mountpoint=dget((*mnt)->mnt_mountpoint);
692 spin_unlock(&vfsmount_lock);
693 dput(*dentry);
694 *dentry = mountpoint;
695 mntput(*mnt);
696 *mnt = parent;
697 return 1;
698}
699
700/* no need for dcache_lock, as serialization is taken care in
701 * namespace.c
702 */
463ffb2e
AV		 703static int __follow_mount(struct path *path)
704{
705 int res = 0;
706 while (d_mountpoint(path->dentry)) {
707 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
708 if (!mounted)
709 break;
710 dput(path->dentry);
711 if (res)
712 mntput(path->mnt);
713 path->mnt = mounted;
714 path->dentry = dget(mounted->mnt_root);
715 res = 1;
716 }
717 return res;
718}
719
58c465eb 720static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
1da177e4 721{
1da177e4
LT		 722 while (d_mountpoint(*dentry)) {
723 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
724 if (!mounted)
725 break;
58c465eb 726 dput(*dentry);
1da177e4
LT		 727 mntput(*mnt);
728 *mnt = mounted;
1da177e4 729 *dentry = dget(mounted->mnt_root);
1da177e4 730 }
1da177e4
LT		 731}
732
733/* no need for dcache_lock, as serialization is taken care in
734 * namespace.c
735 */
e13b210f 736int follow_down(struct vfsmount **mnt, struct dentry **dentry)
1da177e4
LT		 737{
738 struct vfsmount *mounted;
739
740 mounted = lookup_mnt(*mnt, *dentry);
741 if (mounted) {
e13b210f 742 dput(*dentry);
1da177e4
LT		 743 mntput(*mnt);
744 *mnt = mounted;
1da177e4
LT		 745 *dentry = dget(mounted->mnt_root);
746 return 1;
747 }
748 return 0;
749}
750
f1662356 751static __always_inline void follow_dotdot(struct nameidata *nd)
1da177e4 752{
e518ddb7
AM		 753 struct fs_struct *fs = current->fs;
754
1da177e4
LT		 755 while(1) {
756 struct vfsmount *parent;
4ac91378 757 struct dentry *old = nd->path.dentry;
1da177e4 758
e518ddb7 759 read_lock(&fs->lock);
6ac08c39
JB		 760 if (nd->path.dentry == fs->root.dentry &&
761 nd->path.mnt == fs->root.mnt) {
e518ddb7 762 read_unlock(&fs->lock);
1da177e4
LT		 763 break;
764 }
e518ddb7 765 read_unlock(&fs->lock);
1da177e4 766 spin_lock(&dcache_lock);
4ac91378
JB		 767 if (nd->path.dentry != nd->path.mnt->mnt_root) {
768 nd->path.dentry = dget(nd->path.dentry->d_parent);
1da177e4
LT		 769 spin_unlock(&dcache_lock);
770 dput(old);
771 break;
772 }
773 spin_unlock(&dcache_lock);
774 spin_lock(&vfsmount_lock);
4ac91378
JB		 775 parent = nd->path.mnt->mnt_parent;
776 if (parent == nd->path.mnt) {
1da177e4
LT		 777 spin_unlock(&vfsmount_lock);
778 break;
779 }
780 mntget(parent);
4ac91378 781 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
1da177e4
LT		 782 spin_unlock(&vfsmount_lock);
783 dput(old);
4ac91378
JB		 784 mntput(nd->path.mnt);
785 nd->path.mnt = parent;
1da177e4 786 }
4ac91378 787 follow_mount(&nd->path.mnt, &nd->path.dentry);
1da177e4
LT		 788}
789
1da177e4
LT		 790/*
791 * It's more convoluted than I'd like it to be, but... it's still fairly
792 * small and for now I'd prefer to have fast path as straight as possible.
793 * It _is_ time-critical.
794 */
795static int do_lookup(struct nameidata *nd, struct qstr *name,
796 struct path *path)
797{
4ac91378
JB		 798 struct vfsmount *mnt = nd->path.mnt;
799 struct dentry *dentry = __d_lookup(nd->path.dentry, name);
1da177e4
LT		 800
801 if (!dentry)
802 goto need_lookup;
803 if (dentry->d_op && dentry->d_op->d_revalidate)
804 goto need_revalidate;
805done:
806 path->mnt = mnt;
807 path->dentry = dentry;
634ee701 808 __follow_mount(path);
1da177e4
LT		 809 return 0;
810
811need_lookup:
4ac91378 812 dentry = real_lookup(nd->path.dentry, name, nd);
1da177e4
LT		 813 if (IS_ERR(dentry))
814 goto fail;
815 goto done;
816
817need_revalidate:
bcdc5e01
IK		 818 dentry = do_revalidate(dentry, nd);
819 if (!dentry)
820 goto need_lookup;
821 if (IS_ERR(dentry))
822 goto fail;
823 goto done;
1da177e4
LT		 824
825fail:
826 return PTR_ERR(dentry);
827}
828
829/*
830 * Name resolution.
ea3834d9
PM		 831 * This is the basic name resolution function, turning a pathname into
832 * the final dentry. We expect 'base' to be positive and a directory.
1da177e4 833 *
ea3834d9
PM		 834 * Returns 0 and nd will have valid dentry and mnt on success.
835 * Returns error and drops reference to input namei data on failure.
1da177e4 836 */
fc9b52cd 837static int __link_path_walk(const char *name, struct nameidata *nd)
1da177e4
LT		 838{
839 struct path next;
840 struct inode *inode;
841 int err;
842 unsigned int lookup_flags = nd->flags;
843
844 while (*name=='/')
845 name++;
846 if (!*name)
847 goto return_reval;
848
4ac91378 849 inode = nd->path.dentry->d_inode;
1da177e4 850 if (nd->depth)
f55eab82 851 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
1da177e4
LT		 852
853 /* At this point we know we have a real path component. */
854 for(;;) {
855 unsigned long hash;
856 struct qstr this;
857 unsigned int c;
858
cdce5d6b 859 nd->flags |= LOOKUP_CONTINUE;
672b16b2 860 err = exec_permission_lite(inode);
e4543edd
CH		 861 if (err == -EAGAIN)
862 err = vfs_permission(nd, MAY_EXEC);
1da177e4
LT		 863 if (err)
864 break;
865
866 this.name = name;
867 c = *(const unsigned char *)name;
868
869 hash = init_name_hash();
870 do {
871 name++;
872 hash = partial_name_hash(c, hash);
873 c = *(const unsigned char *)name;
874 } while (c && (c != '/'));
875 this.len = name - (const char *) this.name;
876 this.hash = end_name_hash(hash);
877
878 /* remove trailing slashes? */
879 if (!c)
880 goto last_component;
881 while (*++name == '/');
882 if (!*name)
883 goto last_with_slashes;
884
885 /*
886 * "." and ".." are special - ".." especially so because it has
887 * to be able to know about the current root directory and
888 * parent relationships.
889 */
890 if (this.name[0] == '.') switch (this.len) {
891 default:
892 break;
893 case 2:
894 if (this.name[1] != '.')
895 break;
58c465eb 896 follow_dotdot(nd);
4ac91378 897 inode = nd->path.dentry->d_inode;
1da177e4
LT		 898 /* fallthrough */
899 case 1:
900 continue;
901 }
902 /*
903 * See if the low-level filesystem might want
904 * to use its own hash..
905 */
4ac91378
JB		 906 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
907 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
908 &this);
1da177e4
LT		 909 if (err < 0)
910 break;
911 }
1da177e4
LT		 912 /* This does the actual lookups.. */
913 err = do_lookup(nd, &this, &next);
914 if (err)
915 break;
1da177e4
LT		 916
917 err = -ENOENT;
918 inode = next.dentry->d_inode;
919 if (!inode)
920 goto out_dput;
921 err = -ENOTDIR;
922 if (!inode->i_op)
923 goto out_dput;
924
925 if (inode->i_op->follow_link) {
90ebe565 926 err = do_follow_link(&next, nd);
1da177e4
LT		 927 if (err)
928 goto return_err;
929 err = -ENOENT;
4ac91378 930 inode = nd->path.dentry->d_inode;
1da177e4
LT		 931 if (!inode)
932 break;
933 err = -ENOTDIR;
934 if (!inode->i_op)
935 break;
09dd17d3
MS		 936 } else
937 path_to_nameidata(&next, nd);
1da177e4
LT		 938 err = -ENOTDIR;
939 if (!inode->i_op->lookup)
940 break;
941 continue;
942 /* here ends the main loop */
943
944last_with_slashes:
945 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
946last_component:
f55eab82
TM		 947 /* Clear LOOKUP_CONTINUE iff it was previously unset */
948 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
1da177e4
LT		 949 if (lookup_flags & LOOKUP_PARENT)
950 goto lookup_parent;
951 if (this.name[0] == '.') switch (this.len) {
952 default:
953 break;
954 case 2:
955 if (this.name[1] != '.')
956 break;
58c465eb 957 follow_dotdot(nd);
4ac91378 958 inode = nd->path.dentry->d_inode;
1da177e4
LT		 959 /* fallthrough */
960 case 1:
961 goto return_reval;
962 }
4ac91378
JB		 963 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
964 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
965 &this);
1da177e4
LT		 966 if (err < 0)
967 break;
968 }
969 err = do_lookup(nd, &this, &next);
970 if (err)
971 break;
1da177e4
LT		 972 inode = next.dentry->d_inode;
973 if ((lookup_flags & LOOKUP_FOLLOW)
974 && inode && inode->i_op && inode->i_op->follow_link) {
90ebe565 975 err = do_follow_link(&next, nd);
1da177e4
LT		 976 if (err)
977 goto return_err;
4ac91378 978 inode = nd->path.dentry->d_inode;
09dd17d3
MS		 979 } else
980 path_to_nameidata(&next, nd);
1da177e4
LT		 981 err = -ENOENT;
982 if (!inode)
983 break;
984 if (lookup_flags & LOOKUP_DIRECTORY) {
985 err = -ENOTDIR;
986 if (!inode->i_op || !inode->i_op->lookup)
987 break;
988 }
989 goto return_base;
990lookup_parent:
991 nd->last = this;
992 nd->last_type = LAST_NORM;
993 if (this.name[0] != '.')
994 goto return_base;
995 if (this.len == 1)
996 nd->last_type = LAST_DOT;
997 else if (this.len == 2 && this.name[1] == '.')
998 nd->last_type = LAST_DOTDOT;
999 else
1000 goto return_base;
1001return_reval:
1002 /*
1003 * We bypassed the ordinary revalidation routines.
1004 * We may need to check the cached dentry for staleness.
1005 */
4ac91378
JB		 1006 if (nd->path.dentry && nd->path.dentry->d_sb &&
1007 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
1da177e4
LT		 1008 err = -ESTALE;
1009 /* Note: we do not d_invalidate() */
4ac91378
JB		 1010 if (!nd->path.dentry->d_op->d_revalidate(
1011 nd->path.dentry, nd))
1da177e4
LT		 1012 break;
1013 }
1014return_base:
1015 return 0;
1016out_dput:
1d957f9b 1017 path_put_conditional(&next, nd);
1da177e4
LT		 1018 break;
1019 }
1d957f9b 1020 path_put(&nd->path);
1da177e4
LT		 1021return_err:
1022 return err;
1023}
1024
fc9b52cd 1025static int path_walk(const char *name, struct nameidata *nd)
1da177e4
LT		 1026{
1027 current->total_link_count = 0;
1028 return link_path_walk(name, nd);
1029}
1030
ea3834d9 1031/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
fc9b52cd 1032static int do_path_lookup(int dfd, const char *name,
5590ff0d 1033 unsigned int flags, struct nameidata *nd)
1da177e4 1034{
ea3834d9 1035 int retval = 0;
170aa3d0
UD		 1036 int fput_needed;
1037 struct file *file;
e518ddb7 1038 struct fs_struct *fs = current->fs;
1da177e4
LT		 1039
1040 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1041 nd->flags = flags;
1042 nd->depth = 0;
1043
1da177e4 1044 if (*name=='/') {
e518ddb7 1045 read_lock(&fs->lock);
6ac08c39
JB		 1046 nd->path = fs->root;
1047 path_get(&fs->root);
e518ddb7 1048 read_unlock(&fs->lock);
5590ff0d 1049 } else if (dfd == AT_FDCWD) {
e518ddb7 1050 read_lock(&fs->lock);
6ac08c39
JB		 1051 nd->path = fs->pwd;
1052 path_get(&fs->pwd);
e518ddb7 1053 read_unlock(&fs->lock);
5590ff0d 1054 } else {
5590ff0d
UD		 1055 struct dentry *dentry;
1056
1057 file = fget_light(dfd, &fput_needed);
170aa3d0
UD		 1058 retval = -EBADF;
1059 if (!file)
6d09bb62 1060 goto out_fail;
5590ff0d 1061
0f7fc9e4 1062 dentry = file->f_path.dentry;
5590ff0d 1063
170aa3d0
UD		 1064 retval = -ENOTDIR;
1065 if (!S_ISDIR(dentry->d_inode->i_mode))
6d09bb62 1066 goto fput_fail;
5590ff0d
UD		 1067
1068 retval = file_permission(file, MAY_EXEC);
170aa3d0 1069 if (retval)
6d09bb62 1070 goto fput_fail;
5590ff0d 1071
5dd784d0
JB		 1072 nd->path = file->f_path;
1073 path_get(&file->f_path);
5590ff0d
UD		 1074
1075 fput_light(file, fput_needed);
1da177e4 1076 }
2dfdd266
JJS		 1077
1078 retval = path_walk(name, nd);
4ac91378
JB		 1079 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1080 nd->path.dentry->d_inode))
1081 audit_inode(name, nd->path.dentry);
6d09bb62 1082out_fail:
170aa3d0
UD		 1083 return retval;
1084
6d09bb62 1085fput_fail:
170aa3d0 1086 fput_light(file, fput_needed);
6d09bb62 1087 goto out_fail;
1da177e4
LT		 1088}
1089
fc9b52cd 1090int path_lookup(const char *name, unsigned int flags,
5590ff0d
UD		 1091 struct nameidata *nd)
1092{
1093 return do_path_lookup(AT_FDCWD, name, flags, nd);
1094}
1095
d1811465
AV		 1096int kern_path(const char *name, unsigned int flags, struct path *path)
1097{
1098 struct nameidata nd;
1099 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1100 if (!res)
1101 *path = nd.path;
1102 return res;
1103}
1104
16f18200
JJS		 1105/**
1106 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1107 * @dentry: pointer to dentry of the base directory
1108 * @mnt: pointer to vfs mount of the base directory
1109 * @name: pointer to file name
1110 * @flags: lookup flags
1111 * @nd: pointer to nameidata
1112 */
1113int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1114 const char *name, unsigned int flags,
1115 struct nameidata *nd)
1116{
1117 int retval;
1118
1119 /* same as do_path_lookup */
1120 nd->last_type = LAST_ROOT;
1121 nd->flags = flags;
1122 nd->depth = 0;
1123
c8e7f449
JB		 1124 nd->path.dentry = dentry;
1125 nd->path.mnt = mnt;
1126 path_get(&nd->path);
16f18200
JJS		 1127
1128 retval = path_walk(name, nd);
4ac91378
JB		 1129 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1130 nd->path.dentry->d_inode))
1131 audit_inode(name, nd->path.dentry);
16f18200
JJS		 1132
1133 return retval;
1134
1135}
1136
8737f3a1
AV		 1137/**
1138 * path_lookup_open - lookup a file path with open intent
1139 * @dfd: the directory to use as base, or AT_FDCWD
1140 * @name: pointer to file name
1141 * @lookup_flags: lookup intent flags
1142 * @nd: pointer to nameidata
1143 * @open_flags: open intent flags
1144 */
1145int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
1146 struct nameidata *nd, int open_flags)
834f2a4a
TM		 1147{
1148 struct file *filp = get_empty_filp();
1149 int err;
1150
1151 if (filp == NULL)
1152 return -ENFILE;
1153 nd->intent.open.file = filp;
1154 nd->intent.open.flags = open_flags;
8737f3a1 1155 nd->intent.open.create_mode = 0;
5590ff0d 1156 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
834f2a4a
TM		 1157 if (IS_ERR(nd->intent.open.file)) {
1158 if (err == 0) {
1159 err = PTR_ERR(nd->intent.open.file);
1d957f9b 1160 path_put(&nd->path);
834f2a4a
TM		 1161 }
1162 } else if (err != 0)
1163 release_open_intent(nd);
1164 return err;
1165}
1166
eead1911
CH		 1167static struct dentry *__lookup_hash(struct qstr *name,
1168 struct dentry *base, struct nameidata *nd)
1da177e4 1169{
057f6c01 1170 struct dentry *dentry;
1da177e4
LT		 1171 struct inode *inode;
1172 int err;
1173
1174 inode = base->d_inode;
1da177e4
LT		 1175
1176 /*
1177 * See if the low-level filesystem might want
1178 * to use its own hash..
1179 */
1180 if (base->d_op && base->d_op->d_hash) {
1181 err = base->d_op->d_hash(base, name);
1182 dentry = ERR_PTR(err);
1183 if (err < 0)
1184 goto out;
1185 }
1186
1187 dentry = cached_lookup(base, name, nd);
1188 if (!dentry) {
d70b67c8
MS		 1189 struct dentry *new;
1190
1191 /* Don't create child dentry for a dead directory. */
1192 dentry = ERR_PTR(-ENOENT);
1193 if (IS_DEADDIR(inode))
1194 goto out;
1195
1196 new = d_alloc(base, name);
1da177e4
LT		 1197 dentry = ERR_PTR(-ENOMEM);
1198 if (!new)
1199 goto out;
1200 dentry = inode->i_op->lookup(inode, new, nd);
1201 if (!dentry)
1202 dentry = new;
1203 else
1204 dput(new);
1205 }
1206out:
1207 return dentry;
1208}
1209
057f6c01
JM		 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 */
eead1911 1215static struct dentry *lookup_hash(struct nameidata *nd)
057f6c01 1216{
057f6c01
JM		 1217 int err;
1218
f419a2e3 1219 err = inode_permission(nd->path.dentry->d_inode, MAY_EXEC);
057f6c01 1220 if (err)
eead1911 1221 return ERR_PTR(err);
4ac91378 1222 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1da177e4
LT		 1223}
1224
eead1911
CH		 1225static int __lookup_one_len(const char *name, struct qstr *this,
1226 struct dentry *base, int len)
1da177e4
LT		 1227{
1228 unsigned long hash;
1da177e4
LT		 1229 unsigned int c;
1230
057f6c01
JM		 1231 this->name = name;
1232 this->len = len;
1da177e4 1233 if (!len)
057f6c01 1234 return -EACCES;
1da177e4
LT		 1235
1236 hash = init_name_hash();
1237 while (len--) {
1238 c = *(const unsigned char *)name++;
1239 if (c == '/' || c == '\0')
057f6c01 1240 return -EACCES;
1da177e4
LT		 1241 hash = partial_name_hash(c, hash);
1242 }
057f6c01
JM		 1243 this->hash = end_name_hash(hash);
1244 return 0;
1245}
1da177e4 1246
eead1911 1247/**
a6b91919 1248 * lookup_one_len - filesystem helper to lookup single pathname component
eead1911
CH		 1249 * @name: pathname component to lookup
1250 * @base: base directory to lookup from
1251 * @len: maximum length @len should be interpreted to
1252 *
a6b91919
RD		 1253 * Note that this routine is purely a helper for filesystem usage and should
1254 * not be called by generic code. Also note that by using this function the
eead1911
CH		 1255 * nameidata argument is passed to the filesystem methods and a filesystem
1256 * using this helper needs to be prepared for that.
1257 */
057f6c01
JM		 1258struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1259{
1260 int err;
1261 struct qstr this;
1262
1263 err = __lookup_one_len(name, &this, base, len);
eead1911
CH		 1264 if (err)
1265 return ERR_PTR(err);
1266
f419a2e3 1267 err = inode_permission(base->d_inode, MAY_EXEC);
057f6c01
JM		 1268 if (err)
1269 return ERR_PTR(err);
49705b77 1270 return __lookup_hash(&this, base, NULL);
057f6c01
JM		 1271}
1272
eead1911
CH		 1273/**
1274 * lookup_one_noperm - bad hack for sysfs
1275 * @name: pathname component to lookup
1276 * @base: base directory to lookup from
1277 *
1278 * This is a variant of lookup_one_len that doesn't perform any permission
1279 * checks. It's a horrible hack to work around the braindead sysfs
1280 * architecture and should not be used anywhere else.
1281 *
1282 * DON'T USE THIS FUNCTION EVER, thanks.
1283 */
1284struct dentry *lookup_one_noperm(const char *name, struct dentry *base)
057f6c01
JM		 1285{
1286 int err;
1287 struct qstr this;
1288
eead1911 1289 err = __lookup_one_len(name, &this, base, strlen(name));
057f6c01
JM		 1290 if (err)
1291 return ERR_PTR(err);
eead1911 1292 return __lookup_hash(&this, base, NULL);
1da177e4
LT		 1293}
1294
2d8f3038
AV		 1295int user_path_at(int dfd, const char __user *name, unsigned flags,
1296 struct path *path)
1da177e4 1297{
2d8f3038 1298 struct nameidata nd;
1da177e4
LT		 1299 char *tmp = getname(name);
1300 int err = PTR_ERR(tmp);
1da177e4 1301 if (!IS_ERR(tmp)) {
2d8f3038
AV		 1302
1303 BUG_ON(flags & LOOKUP_PARENT);
1304
1305 err = do_path_lookup(dfd, tmp, flags, &nd);
1da177e4 1306 putname(tmp);
2d8f3038
AV		 1307 if (!err)
1308 *path = nd.path;
1da177e4
LT		 1309 }
1310 return err;
1311}
1312
2ad94ae6
AV		 1313static int user_path_parent(int dfd, const char __user *path,
1314 struct nameidata *nd, char **name)
1315{
1316 char *s = getname(path);
1317 int error;
1318
1319 if (IS_ERR(s))
1320 return PTR_ERR(s);
1321
1322 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1323 if (error)
1324 putname(s);
1325 else
1326 *name = s;
1327
1328 return error;
1329}
1330
1da177e4
LT		 1331/*
1332 * It's inline, so penalty for filesystems that don't use sticky bit is
1333 * minimal.
1334 */
1335static inline int check_sticky(struct inode *dir, struct inode *inode)
1336{
1337 if (!(dir->i_mode & S_ISVTX))
1338 return 0;
1339 if (inode->i_uid == current->fsuid)
1340 return 0;
1341 if (dir->i_uid == current->fsuid)
1342 return 0;
1343 return !capable(CAP_FOWNER);
1344}
1345
1346/*
1347 * Check whether we can remove a link victim from directory dir, check
1348 * whether the type of victim is right.
1349 * 1. We can't do it if dir is read-only (done in permission())
1350 * 2. We should have write and exec permissions on dir
1351 * 3. We can't remove anything from append-only dir
1352 * 4. We can't do anything with immutable dir (done in permission())
1353 * 5. If the sticky bit on dir is set we should either
1354 * a. be owner of dir, or
1355 * b. be owner of victim, or
1356 * c. have CAP_FOWNER capability
1357 * 6. If the victim is append-only or immutable we can't do antyhing with
1358 * links pointing to it.
1359 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1360 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1361 * 9. We can't remove a root or mountpoint.
1362 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1363 * nfs_async_unlink().
1364 */
858119e1 1365static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1da177e4
LT		 1366{
1367 int error;
1368
1369 if (!victim->d_inode)
1370 return -ENOENT;
1371
1372 BUG_ON(victim->d_parent->d_inode != dir);
5a190ae6 1373 audit_inode_child(victim->d_name.name, victim, dir);
1da177e4 1374
f419a2e3 1375 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1376 if (error)
1377 return error;
1378 if (IS_APPEND(dir))
1379 return -EPERM;
1380 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1381 IS_IMMUTABLE(victim->d_inode))
1382 return -EPERM;
1383 if (isdir) {
1384 if (!S_ISDIR(victim->d_inode->i_mode))
1385 return -ENOTDIR;
1386 if (IS_ROOT(victim))
1387 return -EBUSY;
1388 } else if (S_ISDIR(victim->d_inode->i_mode))
1389 return -EISDIR;
1390 if (IS_DEADDIR(dir))
1391 return -ENOENT;
1392 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1393 return -EBUSY;
1394 return 0;
1395}
1396
1397/* Check whether we can create an object with dentry child in directory
1398 * dir.
1399 * 1. We can't do it if child already exists (open has special treatment for
1400 * this case, but since we are inlined it's OK)
1401 * 2. We can't do it if dir is read-only (done in permission())
1402 * 3. We should have write and exec permissions on dir
1403 * 4. We can't do it if dir is immutable (done in permission())
1404 */
a95164d9 1405static inline int may_create(struct inode *dir, struct dentry *child)
1da177e4
LT		 1406{
1407 if (child->d_inode)
1408 return -EEXIST;
1409 if (IS_DEADDIR(dir))
1410 return -ENOENT;
f419a2e3 1411 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1412}
1413
1414/*
1da177e4
LT		 1415 * O_DIRECTORY translates into forcing a directory lookup.
1416 */
1417static inline int lookup_flags(unsigned int f)
1418{
1419 unsigned long retval = LOOKUP_FOLLOW;
1420
1421 if (f & O_NOFOLLOW)
1422 retval &= ~LOOKUP_FOLLOW;
1423
1da177e4
LT		 1424 if (f & O_DIRECTORY)
1425 retval |= LOOKUP_DIRECTORY;
1426
1427 return retval;
1428}
1429
1430/*
1431 * p1 and p2 should be directories on the same fs.
1432 */
1433struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1434{
1435 struct dentry *p;
1436
1437 if (p1 == p2) {
f2eace23 1438 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1da177e4
LT		 1439 return NULL;
1440 }
1441
a11f3a05 1442 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4 1443
e2761a11
OH		 1444 p = d_ancestor(p2, p1);
1445 if (p) {
1446 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1447 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1448 return p;
1da177e4
LT		 1449 }
1450
e2761a11
OH		 1451 p = d_ancestor(p1, p2);
1452 if (p) {
1453 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1454 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1455 return p;
1da177e4
LT		 1456 }
1457
f2eace23
IM		 1458 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1459 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1da177e4
LT		 1460 return NULL;
1461}
1462
1463void unlock_rename(struct dentry *p1, struct dentry *p2)
1464{
1b1dcc1b 1465 mutex_unlock(&p1->d_inode->i_mutex);
1da177e4 1466 if (p1 != p2) {
1b1dcc1b 1467 mutex_unlock(&p2->d_inode->i_mutex);
a11f3a05 1468 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4
LT		 1469 }
1470}
1471
1472int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1473 struct nameidata *nd)
1474{
a95164d9 1475 int error = may_create(dir, dentry);
1da177e4
LT		 1476
1477 if (error)
1478 return error;
1479
1480 if (!dir->i_op || !dir->i_op->create)
1481 return -EACCES; /* shouldn't it be ENOSYS? */
1482 mode &= S_IALLUGO;
1483 mode |= S_IFREG;
1484 error = security_inode_create(dir, dentry, mode);
1485 if (error)
1486 return error;
1487 DQUOT_INIT(dir);
1488 error = dir->i_op->create(dir, dentry, mode, nd);
a74574aa 1489 if (!error)
f38aa942 1490 fsnotify_create(dir, dentry);
1da177e4
LT		 1491 return error;
1492}
1493
1494int may_open(struct nameidata *nd, int acc_mode, int flag)
1495{
4ac91378 1496 struct dentry *dentry = nd->path.dentry;
1da177e4
LT		 1497 struct inode *inode = dentry->d_inode;
1498 int error;
1499
1500 if (!inode)
1501 return -ENOENT;
1502
1503 if (S_ISLNK(inode->i_mode))
1504 return -ELOOP;
1505
974a9f0b 1506 if (S_ISDIR(inode->i_mode) && (acc_mode & MAY_WRITE))
1da177e4
LT		 1507 return -EISDIR;
1508
1da177e4
LT		 1509 /*
1510 * FIFO's, sockets and device files are special: they don't
1511 * actually live on the filesystem itself, and as such you
1512 * can write to them even if the filesystem is read-only.
1513 */
1514 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1515 flag &= ~O_TRUNC;
1516 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
4ac91378 1517 if (nd->path.mnt->mnt_flags & MNT_NODEV)
1da177e4
LT		 1518 return -EACCES;
1519
1520 flag &= ~O_TRUNC;
4a3fd211 1521 }
b41572e9
DH		 1522
1523 error = vfs_permission(nd, acc_mode);
1524 if (error)
1525 return error;
1da177e4
LT		 1526 /*
1527 * An append-only file must be opened in append mode for writing.
1528 */
1529 if (IS_APPEND(inode)) {
1530 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1531 return -EPERM;
1532 if (flag & O_TRUNC)
1533 return -EPERM;
1534 }
1535
1536 /* O_NOATIME can only be set by the owner or superuser */
1537 if (flag & O_NOATIME)
3bd858ab 1538 if (!is_owner_or_cap(inode))
1da177e4
LT		 1539 return -EPERM;
1540
1541 /*
1542 * Ensure there are no outstanding leases on the file.
1543 */
1544 error = break_lease(inode, flag);
1545 if (error)
1546 return error;
1547
1548 if (flag & O_TRUNC) {
1549 error = get_write_access(inode);
1550 if (error)
1551 return error;
1552
1553 /*
1554 * Refuse to truncate files with mandatory locks held on them.
1555 */
1556 error = locks_verify_locked(inode);
1557 if (!error) {
1558 DQUOT_INIT(inode);
d139d7ff
MS		 1559
1560 error = do_truncate(dentry, 0,
1561 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1562 NULL);
1da177e4
LT		 1563 }
1564 put_write_access(inode);
1565 if (error)
1566 return error;
1567 } else
1568 if (flag & FMODE_WRITE)
1569 DQUOT_INIT(inode);
1570
1571 return 0;
1572}
1573
d57999e1
DH		 1574/*
1575 * Be careful about ever adding any more callers of this
1576 * function. Its flags must be in the namei format, not
1577 * what get passed to sys_open().
1578 */
1579static int __open_namei_create(struct nameidata *nd, struct path *path,
aab520e2
DH		 1580 int flag, int mode)
1581{
1582 int error;
4ac91378 1583 struct dentry *dir = nd->path.dentry;
aab520e2
DH		 1584
1585 if (!IS_POSIXACL(dir->d_inode))
1586 mode &= ~current->fs->umask;
1587 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1588 mutex_unlock(&dir->d_inode->i_mutex);
4ac91378
JB		 1589 dput(nd->path.dentry);
1590 nd->path.dentry = path->dentry;
aab520e2
DH		 1591 if (error)
1592 return error;
1593 /* Don't check for write permission, don't truncate */
1594 return may_open(nd, 0, flag & ~O_TRUNC);
1595}
1596
d57999e1
DH		 1597/*
1598 * Note that while the flag value (low two bits) for sys_open means:
1599 * 00 - read-only
1600 * 01 - write-only
1601 * 10 - read-write
1602 * 11 - special
1603 * it is changed into
1604 * 00 - no permissions needed
1605 * 01 - read-permission
1606 * 10 - write-permission
1607 * 11 - read-write
1608 * for the internal routines (ie open_namei()/follow_link() etc)
1609 * This is more logical, and also allows the 00 "no perm needed"
1610 * to be used for symlinks (where the permissions are checked
1611 * later).
1612 *
1613*/
1614static inline int open_to_namei_flags(int flag)
1615{
1616 if ((flag+1) & O_ACCMODE)
1617 flag++;
1618 return flag;
1619}
1620
4a3fd211
DH		 1621static int open_will_write_to_fs(int flag, struct inode *inode)
1622{
1623 /*
1624 * We'll never write to the fs underlying
1625 * a device file.
1626 */
1627 if (special_file(inode->i_mode))
1628 return 0;
1629 return (flag & O_TRUNC);
1630}
1631
1da177e4 1632/*
4a3fd211
DH		 1633 * Note that the low bits of the passed in "open_flag"
1634 * are not the same as in the local variable "flag". See
1635 * open_to_namei_flags() for more details.
1da177e4 1636 */
a70e65df
CH		 1637struct file *do_filp_open(int dfd, const char *pathname,
1638 int open_flag, int mode)
1da177e4 1639{
4a3fd211 1640 struct file *filp;
a70e65df 1641 struct nameidata nd;
834f2a4a 1642 int acc_mode, error;
4e7506e4 1643 struct path path;
1da177e4
LT		 1644 struct dentry *dir;
1645 int count = 0;
4a3fd211 1646 int will_write;
d57999e1 1647 int flag = open_to_namei_flags(open_flag);
1da177e4 1648
b77b0646 1649 acc_mode = MAY_OPEN | ACC_MODE(flag);
1da177e4 1650
834f2a4a
TM		 1651 /* O_TRUNC implies we need access checks for write permissions */
1652 if (flag & O_TRUNC)
1653 acc_mode |= MAY_WRITE;
1654
1da177e4
LT		 1655 /* Allow the LSM permission hook to distinguish append
1656 access from general write access. */
1657 if (flag & O_APPEND)
1658 acc_mode |= MAY_APPEND;
1659
1da177e4
LT		 1660 /*
1661 * The simplest case - just a plain lookup.
1662 */
1663 if (!(flag & O_CREAT)) {
5590ff0d 1664 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
a70e65df 1665 &nd, flag);
1da177e4 1666 if (error)
a70e65df 1667 return ERR_PTR(error);
1da177e4
LT		 1668 goto ok;
1669 }
1670
1671 /*
1672 * Create - we need to know the parent.
1673 */
8737f3a1 1674 error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
1da177e4 1675 if (error)
a70e65df 1676 return ERR_PTR(error);
1da177e4
LT		 1677
1678 /*
1679 * We have the parent and last component. First of all, check
1680 * that we are not asked to creat(2) an obvious directory - that
1681 * will not do.
1682 */
1683 error = -EISDIR;
a70e65df 1684 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
8737f3a1 1685 goto exit_parent;
1da177e4 1686
8737f3a1
AV		 1687 error = -ENFILE;
1688 filp = get_empty_filp();
1689 if (filp == NULL)
1690 goto exit_parent;
1691 nd.intent.open.file = filp;
1692 nd.intent.open.flags = flag;
1693 nd.intent.open.create_mode = mode;
a70e65df
CH		 1694 dir = nd.path.dentry;
1695 nd.flags &= ~LOOKUP_PARENT;
8737f3a1 1696 nd.flags |= LOOKUP_CREATE | LOOKUP_OPEN;
3516586a
AV		 1697 if (flag & O_EXCL)
1698 nd.flags |= LOOKUP_EXCL;
1b1dcc1b 1699 mutex_lock(&dir->d_inode->i_mutex);
a70e65df
CH		 1700 path.dentry = lookup_hash(&nd);
1701 path.mnt = nd.path.mnt;
1da177e4
LT		 1702
1703do_last:
4e7506e4
AV		 1704 error = PTR_ERR(path.dentry);
1705 if (IS_ERR(path.dentry)) {
1b1dcc1b 1706 mutex_unlock(&dir->d_inode->i_mutex);
1da177e4
LT		 1707 goto exit;
1708 }
1709
a70e65df 1710 if (IS_ERR(nd.intent.open.file)) {
a70e65df 1711 error = PTR_ERR(nd.intent.open.file);
4a3fd211 1712 goto exit_mutex_unlock;
4af4c52f
OD		 1713 }
1714
1da177e4 1715 /* Negative dentry, just create the file */
4e7506e4 1716 if (!path.dentry->d_inode) {
4a3fd211
DH		 1717 /*
1718 * This write is needed to ensure that a
1719 * ro->rw transition does not occur between
1720 * the time when the file is created and when
1721 * a permanent write count is taken through
1722 * the 'struct file' in nameidata_to_filp().
1723 */
1724 error = mnt_want_write(nd.path.mnt);
1da177e4 1725 if (error)
4a3fd211
DH		 1726 goto exit_mutex_unlock;
1727 error = __open_namei_create(&nd, &path, flag, mode);
1728 if (error) {
1729 mnt_drop_write(nd.path.mnt);
1da177e4 1730 goto exit;
4a3fd211
DH		 1731 }
1732 filp = nameidata_to_filp(&nd, open_flag);
1733 mnt_drop_write(nd.path.mnt);
1734 return filp;
1da177e4
LT		 1735 }
1736
1737 /*
1738 * It already exists.
1739 */
1b1dcc1b 1740 mutex_unlock(&dir->d_inode->i_mutex);
5a190ae6 1741 audit_inode(pathname, path.dentry);
1da177e4
LT		 1742
1743 error = -EEXIST;
1744 if (flag & O_EXCL)
1745 goto exit_dput;
1746
e13b210f 1747 if (__follow_mount(&path)) {
1da177e4 1748 error = -ELOOP;
ba7a4c1a
AV		 1749 if (flag & O_NOFOLLOW)
1750 goto exit_dput;
1da177e4 1751 }
3e2efce0 1752
1da177e4 1753 error = -ENOENT;
4e7506e4 1754 if (!path.dentry->d_inode)
1da177e4 1755 goto exit_dput;
4e7506e4 1756 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
1da177e4
LT		 1757 goto do_link;
1758
a70e65df 1759 path_to_nameidata(&path, &nd);
1da177e4 1760 error = -EISDIR;
4e7506e4 1761 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1da177e4
LT		 1762 goto exit;
1763ok:
4a3fd211
DH		 1764 /*
1765 * Consider:
1766 * 1. may_open() truncates a file
1767 * 2. a rw->ro mount transition occurs
1768 * 3. nameidata_to_filp() fails due to
1769 * the ro mount.
1770 * That would be inconsistent, and should
1771 * be avoided. Taking this mnt write here
1772 * ensures that (2) can not occur.
1773 */
1774 will_write = open_will_write_to_fs(flag, nd.path.dentry->d_inode);
1775 if (will_write) {
1776 error = mnt_want_write(nd.path.mnt);
1777 if (error)
1778 goto exit;
1779 }
a70e65df 1780 error = may_open(&nd, acc_mode, flag);
4a3fd211
DH		 1781 if (error) {
1782 if (will_write)
1783 mnt_drop_write(nd.path.mnt);
1da177e4 1784 goto exit;
4a3fd211
DH		 1785 }
1786 filp = nameidata_to_filp(&nd, open_flag);
1787 /*
1788 * It is now safe to drop the mnt write
1789 * because the filp has had a write taken
1790 * on its behalf.
1791 */
1792 if (will_write)
1793 mnt_drop_write(nd.path.mnt);
1794 return filp;
1da177e4 1795
4a3fd211
DH		 1796exit_mutex_unlock:
1797 mutex_unlock(&dir->d_inode->i_mutex);
1da177e4 1798exit_dput:
a70e65df 1799 path_put_conditional(&path, &nd);
1da177e4 1800exit:
a70e65df
CH		 1801 if (!IS_ERR(nd.intent.open.file))
1802 release_open_intent(&nd);
8737f3a1 1803exit_parent:
a70e65df
CH		 1804 path_put(&nd.path);
1805 return ERR_PTR(error);
1da177e4
LT		 1806
1807do_link:
1808 error = -ELOOP;
1809 if (flag & O_NOFOLLOW)
1810 goto exit_dput;
1811 /*
1812 * This is subtle. Instead of calling do_follow_link() we do the
1813 * thing by hands. The reason is that this way we have zero link_count
1814 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1815 * After that we have the parent and last component, i.e.
1816 * we are in the same situation as after the first path_walk().
1817 * Well, almost - if the last component is normal we get its copy
1818 * stored in nd->last.name and we will have to putname() it when we
1819 * are done. Procfs-like symlinks just set LAST_BIND.
1820 */
a70e65df
CH		 1821 nd.flags |= LOOKUP_PARENT;
1822 error = security_inode_follow_link(path.dentry, &nd);
1da177e4
LT		 1823 if (error)
1824 goto exit_dput;
a70e65df 1825 error = __do_follow_link(&path, &nd);
de459215
KK		 1826 if (error) {
1827 /* Does someone understand code flow here? Or it is only
1828 * me so stupid? Anathema to whoever designed this non-sense
1829 * with "intent.open".
1830 */
a70e65df
CH		 1831 release_open_intent(&nd);
1832 return ERR_PTR(error);
de459215 1833 }
a70e65df
CH		 1834 nd.flags &= ~LOOKUP_PARENT;
1835 if (nd.last_type == LAST_BIND)
1da177e4 1836 goto ok;
1da177e4 1837 error = -EISDIR;
a70e65df 1838 if (nd.last_type != LAST_NORM)
1da177e4 1839 goto exit;
a70e65df
CH		 1840 if (nd.last.name[nd.last.len]) {
1841 __putname(nd.last.name);
1da177e4
LT		 1842 goto exit;
1843 }
1844 error = -ELOOP;
1845 if (count++==32) {
a70e65df 1846 __putname(nd.last.name);
1da177e4
LT		 1847 goto exit;
1848 }
a70e65df 1849 dir = nd.path.dentry;
1b1dcc1b 1850 mutex_lock(&dir->d_inode->i_mutex);
a70e65df
CH		 1851 path.dentry = lookup_hash(&nd);
1852 path.mnt = nd.path.mnt;
1853 __putname(nd.last.name);
1da177e4
LT		 1854 goto do_last;
1855}
1856
a70e65df
CH		 1857/**
1858 * filp_open - open file and return file pointer
1859 *
1860 * @filename: path to open
1861 * @flags: open flags as per the open(2) second argument
1862 * @mode: mode for the new file if O_CREAT is set, else ignored
1863 *
1864 * This is the helper to open a file from kernelspace if you really
1865 * have to. But in generally you should not do this, so please move
1866 * along, nothing to see here..
1867 */
1868struct file *filp_open(const char *filename, int flags, int mode)
1869{
1870 return do_filp_open(AT_FDCWD, filename, flags, mode);
1871}
1872EXPORT_SYMBOL(filp_open);
1873
1da177e4
LT		 1874/**
1875 * lookup_create - lookup a dentry, creating it if it doesn't exist
1876 * @nd: nameidata info
1877 * @is_dir: directory flag
1878 *
1879 * Simple function to lookup and return a dentry and create it
1880 * if it doesn't exist. Is SMP-safe.
c663e5d8 1881 *
4ac91378 1882 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1da177e4
LT		 1883 */
1884struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1885{
c663e5d8 1886 struct dentry *dentry = ERR_PTR(-EEXIST);
1da177e4 1887
4ac91378 1888 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
c663e5d8
CH		 1889 /*
1890 * Yucky last component or no last component at all?
1891 * (foo/., foo/.., /////)
1892 */
1da177e4
LT		 1893 if (nd->last_type != LAST_NORM)
1894 goto fail;
1895 nd->flags &= ~LOOKUP_PARENT;
3516586a 1896 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
a634904a 1897 nd->intent.open.flags = O_EXCL;
c663e5d8
CH		 1898
1899 /*
1900 * Do the final lookup.
1901 */
49705b77 1902 dentry = lookup_hash(nd);
1da177e4
LT		 1903 if (IS_ERR(dentry))
1904 goto fail;
c663e5d8 1905
e9baf6e5
AV		 1906 if (dentry->d_inode)
1907 goto eexist;
c663e5d8
CH		 1908 /*
1909 * Special case - lookup gave negative, but... we had foo/bar/
1910 * From the vfs_mknod() POV we just have a negative dentry -
1911 * all is fine. Let's be bastards - you had / on the end, you've
1912 * been asking for (non-existent) directory. -ENOENT for you.
1913 */
e9baf6e5
AV		 1914 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1915 dput(dentry);
1916 dentry = ERR_PTR(-ENOENT);
1917 }
1da177e4 1918 return dentry;
e9baf6e5 1919eexist:
1da177e4 1920 dput(dentry);
e9baf6e5 1921 dentry = ERR_PTR(-EEXIST);
1da177e4
LT		 1922fail:
1923 return dentry;
1924}
f81a0bff 1925EXPORT_SYMBOL_GPL(lookup_create);
1da177e4
LT		 1926
1927int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1928{
a95164d9 1929 int error = may_create(dir, dentry);
1da177e4
LT		 1930
1931 if (error)
1932 return error;
1933
1934 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1935 return -EPERM;
1936
1937 if (!dir->i_op || !dir->i_op->mknod)
1938 return -EPERM;
1939
08ce5f16
SH		 1940 error = devcgroup_inode_mknod(mode, dev);
1941 if (error)
1942 return error;
1943
1da177e4
LT		 1944 error = security_inode_mknod(dir, dentry, mode, dev);
1945 if (error)
1946 return error;
1947
1948 DQUOT_INIT(dir);
1949 error = dir->i_op->mknod(dir, dentry, mode, dev);
a74574aa 1950 if (!error)
f38aa942 1951 fsnotify_create(dir, dentry);
1da177e4
LT		 1952 return error;
1953}
1954
463c3197
DH		 1955static int may_mknod(mode_t mode)
1956{
1957 switch (mode & S_IFMT) {
1958 case S_IFREG:
1959 case S_IFCHR:
1960 case S_IFBLK:
1961 case S_IFIFO:
1962 case S_IFSOCK:
1963 case 0: /* zero mode translates to S_IFREG */
1964 return 0;
1965 case S_IFDIR:
1966 return -EPERM;
1967 default:
1968 return -EINVAL;
1969 }
1970}
1971
5590ff0d
UD		 1972asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
1973 unsigned dev)
1da177e4 1974{
2ad94ae6
AV		 1975 int error;
1976 char *tmp;
1977 struct dentry *dentry;
1da177e4
LT		 1978 struct nameidata nd;
1979
1980 if (S_ISDIR(mode))
1981 return -EPERM;
1da177e4 1982
2ad94ae6 1983 error = user_path_parent(dfd, filename, &nd, &tmp);
1da177e4 1984 if (error)
2ad94ae6
AV		 1985 return error;
1986
1da177e4 1987 dentry = lookup_create(&nd, 0);
463c3197
DH		 1988 if (IS_ERR(dentry)) {
1989 error = PTR_ERR(dentry);
1990 goto out_unlock;
1991 }
4ac91378 1992 if (!IS_POSIXACL(nd.path.dentry->d_inode))
1da177e4 1993 mode &= ~current->fs->umask;
463c3197
DH		 1994 error = may_mknod(mode);
1995 if (error)
1996 goto out_dput;
1997 error = mnt_want_write(nd.path.mnt);
1998 if (error)
1999 goto out_dput;
2000 switch (mode & S_IFMT) {
1da177e4 2001 case 0: case S_IFREG:
4ac91378 2002 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
1da177e4
LT		 2003 break;
2004 case S_IFCHR: case S_IFBLK:
4ac91378 2005 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
1da177e4
LT		 2006 new_decode_dev(dev));
2007 break;
2008 case S_IFIFO: case S_IFSOCK:
4ac91378 2009 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
1da177e4 2010 break;
1da177e4 2011 }
463c3197
DH		 2012 mnt_drop_write(nd.path.mnt);
2013out_dput:
2014 dput(dentry);
2015out_unlock:
4ac91378 2016 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2017 path_put(&nd.path);
1da177e4
LT		 2018 putname(tmp);
2019
2020 return error;
2021}
2022
5590ff0d
UD		 2023asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
2024{
2025 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2026}
2027
1da177e4
LT		 2028int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2029{
a95164d9 2030 int error = may_create(dir, dentry);
1da177e4
LT		 2031
2032 if (error)
2033 return error;
2034
2035 if (!dir->i_op || !dir->i_op->mkdir)
2036 return -EPERM;
2037
2038 mode &= (S_IRWXUGO|S_ISVTX);
2039 error = security_inode_mkdir(dir, dentry, mode);
2040 if (error)
2041 return error;
2042
2043 DQUOT_INIT(dir);
2044 error = dir->i_op->mkdir(dir, dentry, mode);
a74574aa 2045 if (!error)
f38aa942 2046 fsnotify_mkdir(dir, dentry);
1da177e4
LT		 2047 return error;
2048}
2049
5590ff0d 2050asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
1da177e4
LT		 2051{
2052 int error = 0;
2053 char * tmp;
6902d925
DH		 2054 struct dentry *dentry;
2055 struct nameidata nd;
1da177e4 2056
2ad94ae6
AV		 2057 error = user_path_parent(dfd, pathname, &nd, &tmp);
2058 if (error)
6902d925 2059 goto out_err;
1da177e4 2060
6902d925
DH		 2061 dentry = lookup_create(&nd, 1);
2062 error = PTR_ERR(dentry);
2063 if (IS_ERR(dentry))
2064 goto out_unlock;
1da177e4 2065
4ac91378 2066 if (!IS_POSIXACL(nd.path.dentry->d_inode))
6902d925 2067 mode &= ~current->fs->umask;
463c3197
DH		 2068 error = mnt_want_write(nd.path.mnt);
2069 if (error)
2070 goto out_dput;
4ac91378 2071 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
463c3197
DH		 2072 mnt_drop_write(nd.path.mnt);
2073out_dput:
6902d925
DH		 2074 dput(dentry);
2075out_unlock:
4ac91378 2076 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2077 path_put(&nd.path);
6902d925
DH		 2078 putname(tmp);
2079out_err:
1da177e4
LT		 2080 return error;
2081}
2082
5590ff0d
UD		 2083asmlinkage long sys_mkdir(const char __user *pathname, int mode)
2084{
2085 return sys_mkdirat(AT_FDCWD, pathname, mode);
2086}
2087
1da177e4
LT		 2088/*
2089 * We try to drop the dentry early: we should have
2090 * a usage count of 2 if we're the only user of this
2091 * dentry, and if that is true (possibly after pruning
2092 * the dcache), then we drop the dentry now.
2093 *
2094 * A low-level filesystem can, if it choses, legally
2095 * do a
2096 *
2097 * if (!d_unhashed(dentry))
2098 * return -EBUSY;
2099 *
2100 * if it cannot handle the case of removing a directory
2101 * that is still in use by something else..
2102 */
2103void dentry_unhash(struct dentry *dentry)
2104{
2105 dget(dentry);
dc168427 2106 shrink_dcache_parent(dentry);
1da177e4
LT		 2107 spin_lock(&dcache_lock);
2108 spin_lock(&dentry->d_lock);
2109 if (atomic_read(&dentry->d_count) == 2)
2110 __d_drop(dentry);
2111 spin_unlock(&dentry->d_lock);
2112 spin_unlock(&dcache_lock);
2113}
2114
2115int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2116{
2117 int error = may_delete(dir, dentry, 1);
2118
2119 if (error)
2120 return error;
2121
2122 if (!dir->i_op || !dir->i_op->rmdir)
2123 return -EPERM;
2124
2125 DQUOT_INIT(dir);
2126
1b1dcc1b 2127 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2128 dentry_unhash(dentry);
2129 if (d_mountpoint(dentry))
2130 error = -EBUSY;
2131 else {
2132 error = security_inode_rmdir(dir, dentry);
2133 if (!error) {
2134 error = dir->i_op->rmdir(dir, dentry);
2135 if (!error)
2136 dentry->d_inode->i_flags |= S_DEAD;
2137 }
2138 }
1b1dcc1b 2139 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4 2140 if (!error) {
1da177e4
LT		 2141 d_delete(dentry);
2142 }
2143 dput(dentry);
2144
2145 return error;
2146}
2147
5590ff0d 2148static long do_rmdir(int dfd, const char __user *pathname)
1da177e4
LT		 2149{
2150 int error = 0;
2151 char * name;
2152 struct dentry *dentry;
2153 struct nameidata nd;
2154
2ad94ae6 2155 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2156 if (error)
2ad94ae6 2157 return error;
1da177e4
LT		 2158
2159 switch(nd.last_type) {
0612d9fb
OH		 2160 case LAST_DOTDOT:
2161 error = -ENOTEMPTY;
2162 goto exit1;
2163 case LAST_DOT:
2164 error = -EINVAL;
2165 goto exit1;
2166 case LAST_ROOT:
2167 error = -EBUSY;
2168 goto exit1;
1da177e4 2169 }
0612d9fb
OH		 2170
2171 nd.flags &= ~LOOKUP_PARENT;
2172
4ac91378 2173 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2174 dentry = lookup_hash(&nd);
1da177e4 2175 error = PTR_ERR(dentry);
6902d925
DH		 2176 if (IS_ERR(dentry))
2177 goto exit2;
0622753b
DH		 2178 error = mnt_want_write(nd.path.mnt);
2179 if (error)
2180 goto exit3;
4ac91378 2181 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
0622753b
DH		 2182 mnt_drop_write(nd.path.mnt);
2183exit3:
6902d925
DH		 2184 dput(dentry);
2185exit2:
4ac91378 2186 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2187exit1:
1d957f9b 2188 path_put(&nd.path);
1da177e4
LT		 2189 putname(name);
2190 return error;
2191}
2192
5590ff0d
UD		 2193asmlinkage long sys_rmdir(const char __user *pathname)
2194{
2195 return do_rmdir(AT_FDCWD, pathname);
2196}
2197
1da177e4
LT		 2198int vfs_unlink(struct inode *dir, struct dentry *dentry)
2199{
2200 int error = may_delete(dir, dentry, 0);
2201
2202 if (error)
2203 return error;
2204
2205 if (!dir->i_op || !dir->i_op->unlink)
2206 return -EPERM;
2207
2208 DQUOT_INIT(dir);
2209
1b1dcc1b 2210 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2211 if (d_mountpoint(dentry))
2212 error = -EBUSY;
2213 else {
2214 error = security_inode_unlink(dir, dentry);
2215 if (!error)
2216 error = dir->i_op->unlink(dir, dentry);
2217 }
1b1dcc1b 2218 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2219
2220 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2221 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
ece95912 2222 fsnotify_link_count(dentry->d_inode);
e234f35c 2223 d_delete(dentry);
1da177e4 2224 }
0eeca283 2225
1da177e4
LT		 2226 return error;
2227}
2228
2229/*
2230 * Make sure that the actual truncation of the file will occur outside its
1b1dcc1b 2231 * directory's i_mutex. Truncate can take a long time if there is a lot of
1da177e4
LT		 2232 * writeout happening, and we don't want to prevent access to the directory
2233 * while waiting on the I/O.
2234 */
5590ff0d 2235static long do_unlinkat(int dfd, const char __user *pathname)
1da177e4 2236{
2ad94ae6
AV		 2237 int error;
2238 char *name;
1da177e4
LT		 2239 struct dentry *dentry;
2240 struct nameidata nd;
2241 struct inode *inode = NULL;
2242
2ad94ae6 2243 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2244 if (error)
2ad94ae6
AV		 2245 return error;
2246
1da177e4
LT		 2247 error = -EISDIR;
2248 if (nd.last_type != LAST_NORM)
2249 goto exit1;
0612d9fb
OH		 2250
2251 nd.flags &= ~LOOKUP_PARENT;
2252
4ac91378 2253 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2254 dentry = lookup_hash(&nd);
1da177e4
LT		 2255 error = PTR_ERR(dentry);
2256 if (!IS_ERR(dentry)) {
2257 /* Why not before? Because we want correct error value */
2258 if (nd.last.name[nd.last.len])
2259 goto slashes;
2260 inode = dentry->d_inode;
2261 if (inode)
2262 atomic_inc(&inode->i_count);
0622753b
DH		 2263 error = mnt_want_write(nd.path.mnt);
2264 if (error)
2265 goto exit2;
4ac91378 2266 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
0622753b 2267 mnt_drop_write(nd.path.mnt);
1da177e4
LT		 2268 exit2:
2269 dput(dentry);
2270 }
4ac91378 2271 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4
LT		 2272 if (inode)
2273 iput(inode); /* truncate the inode here */
2274exit1:
1d957f9b 2275 path_put(&nd.path);
1da177e4
LT		 2276 putname(name);
2277 return error;
2278
2279slashes:
2280 error = !dentry->d_inode ? -ENOENT :
2281 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2282 goto exit2;
2283}
2284
5590ff0d
UD		 2285asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2286{
2287 if ((flag & ~AT_REMOVEDIR) != 0)
2288 return -EINVAL;
2289
2290 if (flag & AT_REMOVEDIR)
2291 return do_rmdir(dfd, pathname);
2292
2293 return do_unlinkat(dfd, pathname);
2294}
2295
2296asmlinkage long sys_unlink(const char __user *pathname)
2297{
2298 return do_unlinkat(AT_FDCWD, pathname);
2299}
2300
db2e747b 2301int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
1da177e4 2302{
a95164d9 2303 int error = may_create(dir, dentry);
1da177e4
LT		 2304
2305 if (error)
2306 return error;
2307
2308 if (!dir->i_op || !dir->i_op->symlink)
2309 return -EPERM;
2310
2311 error = security_inode_symlink(dir, dentry, oldname);
2312 if (error)
2313 return error;
2314
2315 DQUOT_INIT(dir);
2316 error = dir->i_op->symlink(dir, dentry, oldname);
a74574aa 2317 if (!error)
f38aa942 2318 fsnotify_create(dir, dentry);
1da177e4
LT		 2319 return error;
2320}
2321
5590ff0d
UD		 2322asmlinkage long sys_symlinkat(const char __user *oldname,
2323 int newdfd, const char __user *newname)
1da177e4 2324{
2ad94ae6
AV		 2325 int error;
2326 char *from;
2327 char *to;
6902d925
DH		 2328 struct dentry *dentry;
2329 struct nameidata nd;
1da177e4
LT		 2330
2331 from = getname(oldname);
2ad94ae6 2332 if (IS_ERR(from))
1da177e4 2333 return PTR_ERR(from);
1da177e4 2334
2ad94ae6 2335 error = user_path_parent(newdfd, newname, &nd, &to);
6902d925 2336 if (error)
2ad94ae6
AV		 2337 goto out_putname;
2338
6902d925
DH		 2339 dentry = lookup_create(&nd, 0);
2340 error = PTR_ERR(dentry);
2341 if (IS_ERR(dentry))
2342 goto out_unlock;
2343
75c3f29d
DH		 2344 error = mnt_want_write(nd.path.mnt);
2345 if (error)
2346 goto out_dput;
db2e747b 2347 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
75c3f29d
DH		 2348 mnt_drop_write(nd.path.mnt);
2349out_dput:
6902d925
DH		 2350 dput(dentry);
2351out_unlock:
4ac91378 2352 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2353 path_put(&nd.path);
6902d925
DH		 2354 putname(to);
2355out_putname:
1da177e4
LT		 2356 putname(from);
2357 return error;
2358}
2359
5590ff0d
UD		 2360asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2361{
2362 return sys_symlinkat(oldname, AT_FDCWD, newname);
2363}
2364
1da177e4
LT		 2365int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2366{
2367 struct inode *inode = old_dentry->d_inode;
2368 int error;
2369
2370 if (!inode)
2371 return -ENOENT;
2372
a95164d9 2373 error = may_create(dir, new_dentry);
1da177e4
LT		 2374 if (error)
2375 return error;
2376
2377 if (dir->i_sb != inode->i_sb)
2378 return -EXDEV;
2379
2380 /*
2381 * A link to an append-only or immutable file cannot be created.
2382 */
2383 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2384 return -EPERM;
2385 if (!dir->i_op || !dir->i_op->link)
2386 return -EPERM;
7e79eedb 2387 if (S_ISDIR(inode->i_mode))
1da177e4
LT		 2388 return -EPERM;
2389
2390 error = security_inode_link(old_dentry, dir, new_dentry);
2391 if (error)
2392 return error;
2393
7e79eedb 2394 mutex_lock(&inode->i_mutex);
1da177e4
LT		 2395 DQUOT_INIT(dir);
2396 error = dir->i_op->link(old_dentry, dir, new_dentry);
7e79eedb 2397 mutex_unlock(&inode->i_mutex);
e31e14ec 2398 if (!error)
7e79eedb 2399 fsnotify_link(dir, inode, new_dentry);
1da177e4
LT		 2400 return error;
2401}
2402
2403/*
2404 * Hardlinks are often used in delicate situations. We avoid
2405 * security-related surprises by not following symlinks on the
2406 * newname. --KAB
2407 *
2408 * We don't follow them on the oldname either to be compatible
2409 * with linux 2.0, and to avoid hard-linking to directories
2410 * and other special files. --ADM
2411 */
5590ff0d 2412asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
c04030e1
UD		 2413 int newdfd, const char __user *newname,
2414 int flags)
1da177e4
LT		 2415{
2416 struct dentry *new_dentry;
2d8f3038
AV		 2417 struct nameidata nd;
2418 struct path old_path;
1da177e4 2419 int error;
2ad94ae6 2420 char *to;
1da177e4 2421
45c9b11a 2422 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
c04030e1
UD		 2423 return -EINVAL;
2424
2d8f3038
AV		 2425 error = user_path_at(olddfd, oldname,
2426 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2427 &old_path);
1da177e4 2428 if (error)
2ad94ae6
AV		 2429 return error;
2430
2431 error = user_path_parent(newdfd, newname, &nd, &to);
1da177e4
LT		 2432 if (error)
2433 goto out;
2434 error = -EXDEV;
2d8f3038 2435 if (old_path.mnt != nd.path.mnt)
1da177e4
LT		 2436 goto out_release;
2437 new_dentry = lookup_create(&nd, 0);
2438 error = PTR_ERR(new_dentry);
6902d925
DH		 2439 if (IS_ERR(new_dentry))
2440 goto out_unlock;
75c3f29d
DH		 2441 error = mnt_want_write(nd.path.mnt);
2442 if (error)
2443 goto out_dput;
2d8f3038 2444 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
75c3f29d
DH		 2445 mnt_drop_write(nd.path.mnt);
2446out_dput:
6902d925
DH		 2447 dput(new_dentry);
2448out_unlock:
4ac91378 2449 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2450out_release:
1d957f9b 2451 path_put(&nd.path);
2ad94ae6 2452 putname(to);
1da177e4 2453out:
2d8f3038 2454 path_put(&old_path);
1da177e4
LT		 2455
2456 return error;
2457}
2458
5590ff0d
UD		 2459asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2460{
c04030e1 2461 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
5590ff0d
UD		 2462}
2463
1da177e4
LT		 2464/*
2465 * The worst of all namespace operations - renaming directory. "Perverted"
2466 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2467 * Problems:
2468 * a) we can get into loop creation. Check is done in is_subdir().
2469 * b) race potential - two innocent renames can create a loop together.
2470 * That's where 4.4 screws up. Current fix: serialization on
a11f3a05 2471 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
1da177e4
LT		 2472 * story.
2473 * c) we have to lock _three_ objects - parents and victim (if it exists).
1b1dcc1b 2474 * And that - after we got ->i_mutex on parents (until then we don't know
1da177e4
LT		 2475 * whether the target exists). Solution: try to be smart with locking
2476 * order for inodes. We rely on the fact that tree topology may change
a11f3a05 2477 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
1da177e4
LT		 2478 * move will be locked. Thus we can rank directories by the tree
2479 * (ancestors first) and rank all non-directories after them.
2480 * That works since everybody except rename does "lock parent, lookup,
a11f3a05 2481 * lock child" and rename is under ->s_vfs_rename_mutex.
1da177e4
LT		 2482 * HOWEVER, it relies on the assumption that any object with ->lookup()
2483 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2484 * we'd better make sure that there's no link(2) for them.
2485 * d) some filesystems don't support opened-but-unlinked directories,
2486 * either because of layout or because they are not ready to deal with
2487 * all cases correctly. The latter will be fixed (taking this sort of
2488 * stuff into VFS), but the former is not going away. Solution: the same
2489 * trick as in rmdir().
2490 * e) conversion from fhandle to dentry may come in the wrong moment - when
1b1dcc1b 2491 * we are removing the target. Solution: we will have to grab ->i_mutex
1da177e4 2492 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
1b1dcc1b 2493 * ->i_mutex on parents, which works but leads to some truely excessive
1da177e4
LT		 2494 * locking].
2495 */
75c96f85
AB		 2496static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2497 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 2498{
2499 int error = 0;
2500 struct inode *target;
2501
2502 /*
2503 * If we are going to change the parent - check write permissions,
2504 * we'll need to flip '..'.
2505 */
2506 if (new_dir != old_dir) {
f419a2e3 2507 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
1da177e4
LT		 2508 if (error)
2509 return error;
2510 }
2511
2512 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2513 if (error)
2514 return error;
2515
2516 target = new_dentry->d_inode;
2517 if (target) {
1b1dcc1b 2518 mutex_lock(&target->i_mutex);
1da177e4
LT		 2519 dentry_unhash(new_dentry);
2520 }
2521 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2522 error = -EBUSY;
2523 else
2524 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2525 if (target) {
2526 if (!error)
2527 target->i_flags |= S_DEAD;
1b1dcc1b 2528 mutex_unlock(&target->i_mutex);
1da177e4
LT		 2529 if (d_unhashed(new_dentry))
2530 d_rehash(new_dentry);
2531 dput(new_dentry);
2532 }
e31e14ec 2533 if (!error)
349457cc
MF		 2534 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2535 d_move(old_dentry,new_dentry);
1da177e4
LT		 2536 return error;
2537}
2538
75c96f85
AB		 2539static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2540 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 2541{
2542 struct inode *target;
2543 int error;
2544
2545 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2546 if (error)
2547 return error;
2548
2549 dget(new_dentry);
2550 target = new_dentry->d_inode;
2551 if (target)
1b1dcc1b 2552 mutex_lock(&target->i_mutex);
1da177e4
LT		 2553 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2554 error = -EBUSY;
2555 else
2556 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2557 if (!error) {
349457cc 2558 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
1da177e4 2559 d_move(old_dentry, new_dentry);
1da177e4
LT		 2560 }
2561 if (target)
1b1dcc1b 2562 mutex_unlock(&target->i_mutex);
1da177e4
LT		 2563 dput(new_dentry);
2564 return error;
2565}
2566
2567int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2568 struct inode *new_dir, struct dentry *new_dentry)
2569{
2570 int error;
2571 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
0eeca283 2572 const char *old_name;
1da177e4
LT		 2573
2574 if (old_dentry->d_inode == new_dentry->d_inode)
2575 return 0;
2576
2577 error = may_delete(old_dir, old_dentry, is_dir);
2578 if (error)
2579 return error;
2580
2581 if (!new_dentry->d_inode)
a95164d9 2582 error = may_create(new_dir, new_dentry);
1da177e4
LT		 2583 else
2584 error = may_delete(new_dir, new_dentry, is_dir);
2585 if (error)
2586 return error;
2587
2588 if (!old_dir->i_op || !old_dir->i_op->rename)
2589 return -EPERM;
2590
2591 DQUOT_INIT(old_dir);
2592 DQUOT_INIT(new_dir);
2593
0eeca283
RL		 2594 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2595
1da177e4
LT		 2596 if (is_dir)
2597 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2598 else
2599 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2600 if (!error) {
0eeca283 2601 const char *new_name = old_dentry->d_name.name;
89204c40 2602 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
5a190ae6 2603 new_dentry->d_inode, old_dentry);
1da177e4 2604 }
0eeca283
RL		 2605 fsnotify_oldname_free(old_name);
2606
1da177e4
LT		 2607 return error;
2608}
2609
2ad94ae6
AV		 2610asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2611 int newdfd, const char __user *newname)
1da177e4 2612{
2ad94ae6
AV		 2613 struct dentry *old_dir, *new_dir;
2614 struct dentry *old_dentry, *new_dentry;
2615 struct dentry *trap;
1da177e4 2616 struct nameidata oldnd, newnd;
2ad94ae6
AV		 2617 char *from;
2618 char *to;
2619 int error;
1da177e4 2620
2ad94ae6 2621 error = user_path_parent(olddfd, oldname, &oldnd, &from);
1da177e4
LT		 2622 if (error)
2623 goto exit;
2624
2ad94ae6 2625 error = user_path_parent(newdfd, newname, &newnd, &to);
1da177e4
LT		 2626 if (error)
2627 goto exit1;
2628
2629 error = -EXDEV;
4ac91378 2630 if (oldnd.path.mnt != newnd.path.mnt)
1da177e4
LT		 2631 goto exit2;
2632
4ac91378 2633 old_dir = oldnd.path.dentry;
1da177e4
LT		 2634 error = -EBUSY;
2635 if (oldnd.last_type != LAST_NORM)
2636 goto exit2;
2637
4ac91378 2638 new_dir = newnd.path.dentry;
1da177e4
LT		 2639 if (newnd.last_type != LAST_NORM)
2640 goto exit2;
2641
0612d9fb
OH		 2642 oldnd.flags &= ~LOOKUP_PARENT;
2643 newnd.flags &= ~LOOKUP_PARENT;
4e9ed2f8 2644 newnd.flags |= LOOKUP_RENAME_TARGET;
0612d9fb 2645
1da177e4
LT		 2646 trap = lock_rename(new_dir, old_dir);
2647
49705b77 2648 old_dentry = lookup_hash(&oldnd);
1da177e4
LT		 2649 error = PTR_ERR(old_dentry);
2650 if (IS_ERR(old_dentry))
2651 goto exit3;
2652 /* source must exist */
2653 error = -ENOENT;
2654 if (!old_dentry->d_inode)
2655 goto exit4;
2656 /* unless the source is a directory trailing slashes give -ENOTDIR */
2657 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2658 error = -ENOTDIR;
2659 if (oldnd.last.name[oldnd.last.len])
2660 goto exit4;
2661 if (newnd.last.name[newnd.last.len])
2662 goto exit4;
2663 }
2664 /* source should not be ancestor of target */
2665 error = -EINVAL;
2666 if (old_dentry == trap)
2667 goto exit4;
49705b77 2668 new_dentry = lookup_hash(&newnd);
1da177e4
LT		 2669 error = PTR_ERR(new_dentry);
2670 if (IS_ERR(new_dentry))
2671 goto exit4;
2672 /* target should not be an ancestor of source */
2673 error = -ENOTEMPTY;
2674 if (new_dentry == trap)
2675 goto exit5;
2676
9079b1eb
DH		 2677 error = mnt_want_write(oldnd.path.mnt);
2678 if (error)
2679 goto exit5;
1da177e4
LT		 2680 error = vfs_rename(old_dir->d_inode, old_dentry,
2681 new_dir->d_inode, new_dentry);
9079b1eb 2682 mnt_drop_write(oldnd.path.mnt);
1da177e4
LT		 2683exit5:
2684 dput(new_dentry);
2685exit4:
2686 dput(old_dentry);
2687exit3:
2688 unlock_rename(new_dir, old_dir);
2689exit2:
1d957f9b 2690 path_put(&newnd.path);
2ad94ae6 2691 putname(to);
1da177e4 2692exit1:
1d957f9b 2693 path_put(&oldnd.path);
1da177e4 2694 putname(from);
2ad94ae6 2695exit:
1da177e4
LT		 2696 return error;
2697}
2698
5590ff0d
UD		 2699asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2700{
2701 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2702}
2703
1da177e4
LT		 2704int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2705{
2706 int len;
2707
2708 len = PTR_ERR(link);
2709 if (IS_ERR(link))
2710 goto out;
2711
2712 len = strlen(link);
2713 if (len > (unsigned) buflen)
2714 len = buflen;
2715 if (copy_to_user(buffer, link, len))
2716 len = -EFAULT;
2717out:
2718 return len;
2719}
2720
2721/*
2722 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2723 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2724 * using) it for any given inode is up to filesystem.
2725 */
2726int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2727{
2728 struct nameidata nd;
cc314eef 2729 void *cookie;
694a1764 2730 int res;
cc314eef 2731
1da177e4 2732 nd.depth = 0;
cc314eef 2733 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
694a1764
MS		 2734 if (IS_ERR(cookie))
2735 return PTR_ERR(cookie);
2736
2737 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2738 if (dentry->d_inode->i_op->put_link)
2739 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2740 return res;
1da177e4
LT		 2741}
2742
2743int vfs_follow_link(struct nameidata *nd, const char *link)
2744{
2745 return __vfs_follow_link(nd, link);
2746}
2747
2748/* get the link contents into pagecache */
2749static char *page_getlink(struct dentry * dentry, struct page **ppage)
2750{
2751 struct page * page;
2752 struct address_space *mapping = dentry->d_inode->i_mapping;
090d2b18 2753 page = read_mapping_page(mapping, 0, NULL);
1da177e4 2754 if (IS_ERR(page))
6fe6900e 2755 return (char*)page;
1da177e4
LT		 2756 *ppage = page;
2757 return kmap(page);
1da177e4
LT		 2758}
2759
2760int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2761{
2762 struct page *page = NULL;
2763 char *s = page_getlink(dentry, &page);
2764 int res = vfs_readlink(dentry,buffer,buflen,s);
2765 if (page) {
2766 kunmap(page);
2767 page_cache_release(page);
2768 }
2769 return res;
2770}
2771
cc314eef 2772void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
1da177e4 2773{
cc314eef 2774 struct page *page = NULL;
1da177e4 2775 nd_set_link(nd, page_getlink(dentry, &page));
cc314eef 2776 return page;
1da177e4
LT		 2777}
2778
cc314eef 2779void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1da177e4 2780{
cc314eef
LT		 2781 struct page *page = cookie;
2782
2783 if (page) {
1da177e4
LT		 2784 kunmap(page);
2785 page_cache_release(page);
1da177e4
LT		 2786 }
2787}
2788
0adb25d2
KK		 2789int __page_symlink(struct inode *inode, const char *symname, int len,
2790 gfp_t gfp_mask)
1da177e4
LT		 2791{
2792 struct address_space *mapping = inode->i_mapping;
0adb25d2 2793 struct page *page;
afddba49 2794 void *fsdata;
beb497ab 2795 int err;
1da177e4
LT		 2796 char *kaddr;
2797
7e53cac4 2798retry:
afddba49
NP		 2799 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2800 AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
1da177e4 2801 if (err)
afddba49
NP		 2802 goto fail;
2803
1da177e4
LT		 2804 kaddr = kmap_atomic(page, KM_USER0);
2805 memcpy(kaddr, symname, len-1);
2806 kunmap_atomic(kaddr, KM_USER0);
afddba49
NP		 2807
2808 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2809 page, fsdata);
1da177e4
LT		 2810 if (err < 0)
2811 goto fail;
afddba49
NP		 2812 if (err < len-1)
2813 goto retry;
2814
1da177e4
LT		 2815 mark_inode_dirty(inode);
2816 return 0;
1da177e4
LT		 2817fail:
2818 return err;
2819}
2820
0adb25d2
KK		 2821int page_symlink(struct inode *inode, const char *symname, int len)
2822{
2823 return __page_symlink(inode, symname, len,
2824 mapping_gfp_mask(inode->i_mapping));
2825}
2826
92e1d5be 2827const struct inode_operations page_symlink_inode_operations = {
1da177e4
LT		 2828 .readlink = generic_readlink,
2829 .follow_link = page_follow_link_light,
2830 .put_link = page_put_link,
2831};
2832
2d8f3038 2833EXPORT_SYMBOL(user_path_at);
1da177e4
LT		 2834EXPORT_SYMBOL(follow_down);
2835EXPORT_SYMBOL(follow_up);
2836EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2837EXPORT_SYMBOL(getname);
2838EXPORT_SYMBOL(lock_rename);
1da177e4
LT		 2839EXPORT_SYMBOL(lookup_one_len);
2840EXPORT_SYMBOL(page_follow_link_light);
2841EXPORT_SYMBOL(page_put_link);
2842EXPORT_SYMBOL(page_readlink);
0adb25d2 2843EXPORT_SYMBOL(__page_symlink);
1da177e4
LT		 2844EXPORT_SYMBOL(page_symlink);
2845EXPORT_SYMBOL(page_symlink_inode_operations);
2846EXPORT_SYMBOL(path_lookup);
d1811465 2847EXPORT_SYMBOL(kern_path);
16f18200 2848EXPORT_SYMBOL(vfs_path_lookup);
f419a2e3 2849EXPORT_SYMBOL(inode_permission);
e4543edd 2850EXPORT_SYMBOL(vfs_permission);
8c744fb8 2851EXPORT_SYMBOL(file_permission);
1da177e4
LT		 2852EXPORT_SYMBOL(unlock_rename);
2853EXPORT_SYMBOL(vfs_create);
2854EXPORT_SYMBOL(vfs_follow_link);
2855EXPORT_SYMBOL(vfs_link);
2856EXPORT_SYMBOL(vfs_mkdir);
2857EXPORT_SYMBOL(vfs_mknod);
2858EXPORT_SYMBOL(generic_permission);
2859EXPORT_SYMBOL(vfs_readlink);
2860EXPORT_SYMBOL(vfs_rename);
2861EXPORT_SYMBOL(vfs_rmdir);
2862EXPORT_SYMBOL(vfs_symlink);
2863EXPORT_SYMBOL(vfs_unlink);
2864EXPORT_SYMBOL(dentry_unhash);
2865EXPORT_SYMBOL(generic_readlink);
kernel source for telekom puls (alcatel/mediatek)