projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
ntfs: don't NULL i_op
[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
da9592ed 189 if (current_fsuid() == inode->i_uid)
1da177e4
LT		 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
cb23beb5
CH		 229/**
230 * inode_permission - check for access rights to a given inode
231 * @inode: inode to check permission on
232 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
233 *
234 * Used to check for read/write/execute permissions on an inode.
235 * We use "fsuid" for this, letting us set arbitrary permissions
236 * for filesystem access without changing the "normal" uids which
237 * are used for other things.
238 */
f419a2e3 239int inode_permission(struct inode *inode, int mask)
1da177e4 240{
e6305c43 241 int retval;
1da177e4
LT		 242
243 if (mask & MAY_WRITE) {
22590e41 244 umode_t mode = inode->i_mode;
1da177e4
LT		 245
246 /*
247 * Nobody gets write access to a read-only fs.
248 */
249 if (IS_RDONLY(inode) &&
250 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
251 return -EROFS;
252
253 /*
254 * Nobody gets write access to an immutable file.
255 */
256 if (IS_IMMUTABLE(inode))
257 return -EACCES;
258 }
259
f696a365 260 if (inode->i_op && inode->i_op->permission)
b77b0646 261 retval = inode->i_op->permission(inode, mask);
f696a365 262 else
e6305c43 263 retval = generic_permission(inode, mask, NULL);
f696a365 264
1da177e4
LT		 265 if (retval)
266 return retval;
267
08ce5f16
SH		 268 retval = devcgroup_inode_permission(inode, mask);
269 if (retval)
270 return retval;
271
e6305c43 272 return security_inode_permission(inode,
f418b006 273 mask & (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND));
1da177e4
LT		 274}
275
8c744fb8
CH		 276/**
277 * file_permission - check for additional access rights to a given file
278 * @file: file to check access rights for
279 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
280 *
281 * Used to check for read/write/execute permissions on an already opened
282 * file.
283 *
284 * Note:
285 * Do not use this function in new code. All access checks should
cb23beb5 286 * be done using inode_permission().
8c744fb8
CH		 287 */
288int file_permission(struct file *file, int mask)
289{
f419a2e3 290 return inode_permission(file->f_path.dentry->d_inode, mask);
8c744fb8
CH		 291}
292
1da177e4
LT		 293/*
294 * get_write_access() gets write permission for a file.
295 * put_write_access() releases this write permission.
296 * This is used for regular files.
297 * We cannot support write (and maybe mmap read-write shared) accesses and
298 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
299 * can have the following values:
300 * 0: no writers, no VM_DENYWRITE mappings
301 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
302 * > 0: (i_writecount) users are writing to the file.
303 *
304 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
305 * except for the cases where we don't hold i_writecount yet. Then we need to
306 * use {get,deny}_write_access() - these functions check the sign and refuse
307 * to do the change if sign is wrong. Exclusion between them is provided by
308 * the inode->i_lock spinlock.
309 */
310
311int get_write_access(struct inode * inode)
312{
313 spin_lock(&inode->i_lock);
314 if (atomic_read(&inode->i_writecount) < 0) {
315 spin_unlock(&inode->i_lock);
316 return -ETXTBSY;
317 }
318 atomic_inc(&inode->i_writecount);
319 spin_unlock(&inode->i_lock);
320
321 return 0;
322}
323
324int deny_write_access(struct file * file)
325{
0f7fc9e4 326 struct inode *inode = file->f_path.dentry->d_inode;
1da177e4
LT		 327
328 spin_lock(&inode->i_lock);
329 if (atomic_read(&inode->i_writecount) > 0) {
330 spin_unlock(&inode->i_lock);
331 return -ETXTBSY;
332 }
333 atomic_dec(&inode->i_writecount);
334 spin_unlock(&inode->i_lock);
335
336 return 0;
337}
338
5dd784d0
JB		 339/**
340 * path_get - get a reference to a path
341 * @path: path to get the reference to
342 *
343 * Given a path increment the reference count to the dentry and the vfsmount.
344 */
345void path_get(struct path *path)
346{
347 mntget(path->mnt);
348 dget(path->dentry);
349}
350EXPORT_SYMBOL(path_get);
351
1d957f9b
JB		 352/**
353 * path_put - put a reference to a path
354 * @path: path to put the reference to
355 *
356 * Given a path decrement the reference count to the dentry and the vfsmount.
357 */
358void path_put(struct path *path)
1da177e4 359{
1d957f9b
JB		 360 dput(path->dentry);
361 mntput(path->mnt);
1da177e4 362}
1d957f9b 363EXPORT_SYMBOL(path_put);
1da177e4 364
834f2a4a
TM		 365/**
366 * release_open_intent - free up open intent resources
367 * @nd: pointer to nameidata
368 */
369void release_open_intent(struct nameidata *nd)
370{
0f7fc9e4 371 if (nd->intent.open.file->f_path.dentry == NULL)
834f2a4a
TM		 372 put_filp(nd->intent.open.file);
373 else
374 fput(nd->intent.open.file);
375}
376
bcdc5e01
IK		 377static inline struct dentry *
378do_revalidate(struct dentry *dentry, struct nameidata *nd)
379{
380 int status = dentry->d_op->d_revalidate(dentry, nd);
381 if (unlikely(status <= 0)) {
382 /*
383 * The dentry failed validation.
384 * If d_revalidate returned 0 attempt to invalidate
385 * the dentry otherwise d_revalidate is asking us
386 * to return a fail status.
387 */
388 if (!status) {
389 if (!d_invalidate(dentry)) {
390 dput(dentry);
391 dentry = NULL;
392 }
393 } else {
394 dput(dentry);
395 dentry = ERR_PTR(status);
396 }
397 }
398 return dentry;
399}
400
1da177e4
LT		 401/*
402 * Internal lookup() using the new generic dcache.
403 * SMP-safe
404 */
405static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
406{
407 struct dentry * dentry = __d_lookup(parent, name);
408
409 /* lockess __d_lookup may fail due to concurrent d_move()
410 * in some unrelated directory, so try with d_lookup
411 */
412 if (!dentry)
413 dentry = d_lookup(parent, name);
414
bcdc5e01
IK		 415 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
416 dentry = do_revalidate(dentry, nd);
417
1da177e4
LT		 418 return dentry;
419}
420
421/*
422 * Short-cut version of permission(), for calling by
423 * path_walk(), when dcache lock is held. Combines parts
424 * of permission() and generic_permission(), and tests ONLY for
425 * MAY_EXEC permission.
426 *
427 * If appropriate, check DAC only. If not appropriate, or
428 * short-cut DAC fails, then call permission() to do more
429 * complete permission check.
430 */
672b16b2 431static int exec_permission_lite(struct inode *inode)
1da177e4
LT		 432{
433 umode_t mode = inode->i_mode;
434
435 if (inode->i_op && inode->i_op->permission)
436 return -EAGAIN;
437
da9592ed 438 if (current_fsuid() == inode->i_uid)
1da177e4
LT		 439 mode >>= 6;
440 else if (in_group_p(inode->i_gid))
441 mode >>= 3;
442
443 if (mode & MAY_EXEC)
444 goto ok;
445
446 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
447 goto ok;
448
449 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
450 goto ok;
451
452 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
453 goto ok;
454
455 return -EACCES;
456ok:
b77b0646 457 return security_inode_permission(inode, MAY_EXEC);
1da177e4
LT		 458}
459
460/*
461 * This is called when everything else fails, and we actually have
462 * to go to the low-level filesystem to find out what we should do..
463 *
464 * We get the directory semaphore, and after getting that we also
465 * make sure that nobody added the entry to the dcache in the meantime..
466 * SMP-safe
467 */
468static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
469{
470 struct dentry * result;
471 struct inode *dir = parent->d_inode;
472
1b1dcc1b 473 mutex_lock(&dir->i_mutex);
1da177e4
LT		 474 /*
475 * First re-do the cached lookup just in case it was created
476 * while we waited for the directory semaphore..
477 *
478 * FIXME! This could use version numbering or similar to
479 * avoid unnecessary cache lookups.
480 *
481 * The "dcache_lock" is purely to protect the RCU list walker
482 * from concurrent renames at this point (we mustn't get false
483 * negatives from the RCU list walk here, unlike the optimistic
484 * fast walk).
485 *
486 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
487 */
488 result = d_lookup(parent, name);
489 if (!result) {
d70b67c8
MS		 490 struct dentry *dentry;
491
492 /* Don't create child dentry for a dead directory. */
493 result = ERR_PTR(-ENOENT);
494 if (IS_DEADDIR(dir))
495 goto out_unlock;
496
497 dentry = d_alloc(parent, name);
1da177e4
LT		 498 result = ERR_PTR(-ENOMEM);
499 if (dentry) {
500 result = dir->i_op->lookup(dir, dentry, nd);
501 if (result)
502 dput(dentry);
503 else
504 result = dentry;
505 }
d70b67c8 506out_unlock:
1b1dcc1b 507 mutex_unlock(&dir->i_mutex);
1da177e4
LT		 508 return result;
509 }
510
511 /*
512 * Uhhuh! Nasty case: the cache was re-populated while
513 * we waited on the semaphore. Need to revalidate.
514 */
1b1dcc1b 515 mutex_unlock(&dir->i_mutex);
1da177e4 516 if (result->d_op && result->d_op->d_revalidate) {
bcdc5e01
IK		 517 result = do_revalidate(result, nd);
518 if (!result)
1da177e4 519 result = ERR_PTR(-ENOENT);
1da177e4
LT		 520 }
521 return result;
522}
523
a02f76c3
AV		 524/*
525 * Wrapper to retry pathname resolution whenever the underlying
526 * file system returns an ESTALE.
527 *
528 * Retry the whole path once, forcing real lookup requests
529 * instead of relying on the dcache.
530 */
531static __always_inline int link_path_walk(const char *name, struct nameidata *nd)
532{
533 struct path save = nd->path;
534 int result;
535
536 /* make sure the stuff we saved doesn't go away */
c8e7f449 537 path_get(&save);
a02f76c3
AV		 538
539 result = __link_path_walk(name, nd);
540 if (result == -ESTALE) {
541 /* nd->path had been dropped */
542 nd->path = save;
c8e7f449 543 path_get(&nd->path);
a02f76c3
AV		 544 nd->flags |= LOOKUP_REVAL;
545 result = __link_path_walk(name, nd);
546 }
547
548 path_put(&save);
549
550 return result;
551}
552
f1662356 553static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
1da177e4
LT		 554{
555 int res = 0;
556 char *name;
557 if (IS_ERR(link))
558 goto fail;
559
560 if (*link == '/') {
b4091d5f
CH		 561 struct fs_struct *fs = current->fs;
562
1d957f9b 563 path_put(&nd->path);
b4091d5f
CH		 564
565 read_lock(&fs->lock);
566 nd->path = fs->root;
567 path_get(&fs->root);
568 read_unlock(&fs->lock);
1da177e4 569 }
b4091d5f 570
1da177e4 571 res = link_path_walk(link, nd);
1da177e4
LT		 572 if (nd->depth || res || nd->last_type!=LAST_NORM)
573 return res;
574 /*
575 * If it is an iterative symlinks resolution in open_namei() we
576 * have to copy the last component. And all that crap because of
577 * bloody create() on broken symlinks. Furrfu...
578 */
579 name = __getname();
580 if (unlikely(!name)) {
1d957f9b 581 path_put(&nd->path);
1da177e4
LT		 582 return -ENOMEM;
583 }
584 strcpy(name, nd->last.name);
585 nd->last.name = name;
586 return 0;
587fail:
1d957f9b 588 path_put(&nd->path);
1da177e4
LT		 589 return PTR_ERR(link);
590}
591
1d957f9b 592static void path_put_conditional(struct path *path, struct nameidata *nd)
051d3812
IK		 593{
594 dput(path->dentry);
4ac91378 595 if (path->mnt != nd->path.mnt)
051d3812
IK		 596 mntput(path->mnt);
597}
598
599static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
600{
4ac91378
JB		 601 dput(nd->path.dentry);
602 if (nd->path.mnt != path->mnt)
603 mntput(nd->path.mnt);
604 nd->path.mnt = path->mnt;
605 nd->path.dentry = path->dentry;
051d3812
IK		 606}
607
f1662356 608static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
1da177e4
LT		 609{
610 int error;
cc314eef 611 void *cookie;
cd4e91d3 612 struct dentry *dentry = path->dentry;
1da177e4 613
d671a1cb 614 touch_atime(path->mnt, dentry);
1da177e4 615 nd_set_link(nd, NULL);
cd4e91d3 616
4ac91378 617 if (path->mnt != nd->path.mnt) {
051d3812
IK		 618 path_to_nameidata(path, nd);
619 dget(dentry);
620 }
621 mntget(path->mnt);
cc314eef
LT		 622 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
623 error = PTR_ERR(cookie);
624 if (!IS_ERR(cookie)) {
1da177e4 625 char *s = nd_get_link(nd);
cc314eef 626 error = 0;
1da177e4
LT		 627 if (s)
628 error = __vfs_follow_link(nd, s);
629 if (dentry->d_inode->i_op->put_link)
cc314eef 630 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
1da177e4 631 }
09da5916 632 path_put(path);
1da177e4
LT		 633
634 return error;
635}
636
637/*
638 * This limits recursive symlink follows to 8, while
639 * limiting consecutive symlinks to 40.
640 *
641 * Without that kind of total limit, nasty chains of consecutive
642 * symlinks can cause almost arbitrarily long lookups.
643 */
90ebe565 644static inline int do_follow_link(struct path *path, struct nameidata *nd)
1da177e4
LT		 645{
646 int err = -ELOOP;
647 if (current->link_count >= MAX_NESTED_LINKS)
648 goto loop;
649 if (current->total_link_count >= 40)
650 goto loop;
651 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
652 cond_resched();
90ebe565 653 err = security_inode_follow_link(path->dentry, nd);
1da177e4
LT		 654 if (err)
655 goto loop;
656 current->link_count++;
657 current->total_link_count++;
658 nd->depth++;
cd4e91d3 659 err = __do_follow_link(path, nd);
839d9f93
AV		 660 current->link_count--;
661 nd->depth--;
1da177e4
LT		 662 return err;
663loop:
1d957f9b
JB		 664 path_put_conditional(path, nd);
665 path_put(&nd->path);
1da177e4
LT		 666 return err;
667}
668
669int follow_up(struct vfsmount **mnt, struct dentry **dentry)
670{
671 struct vfsmount *parent;
672 struct dentry *mountpoint;
673 spin_lock(&vfsmount_lock);
674 parent=(*mnt)->mnt_parent;
675 if (parent == *mnt) {
676 spin_unlock(&vfsmount_lock);
677 return 0;
678 }
679 mntget(parent);
680 mountpoint=dget((*mnt)->mnt_mountpoint);
681 spin_unlock(&vfsmount_lock);
682 dput(*dentry);
683 *dentry = mountpoint;
684 mntput(*mnt);
685 *mnt = parent;
686 return 1;
687}
688
689/* no need for dcache_lock, as serialization is taken care in
690 * namespace.c
691 */
463ffb2e
AV		 692static int __follow_mount(struct path *path)
693{
694 int res = 0;
695 while (d_mountpoint(path->dentry)) {
696 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
697 if (!mounted)
698 break;
699 dput(path->dentry);
700 if (res)
701 mntput(path->mnt);
702 path->mnt = mounted;
703 path->dentry = dget(mounted->mnt_root);
704 res = 1;
705 }
706 return res;
707}
708
58c465eb 709static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
1da177e4 710{
1da177e4
LT		 711 while (d_mountpoint(*dentry)) {
712 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
713 if (!mounted)
714 break;
58c465eb 715 dput(*dentry);
1da177e4
LT		 716 mntput(*mnt);
717 *mnt = mounted;
1da177e4 718 *dentry = dget(mounted->mnt_root);
1da177e4 719 }
1da177e4
LT		 720}
721
722/* no need for dcache_lock, as serialization is taken care in
723 * namespace.c
724 */
e13b210f 725int follow_down(struct vfsmount **mnt, struct dentry **dentry)
1da177e4
LT		 726{
727 struct vfsmount *mounted;
728
729 mounted = lookup_mnt(*mnt, *dentry);
730 if (mounted) {
e13b210f 731 dput(*dentry);
1da177e4
LT		 732 mntput(*mnt);
733 *mnt = mounted;
1da177e4
LT		 734 *dentry = dget(mounted->mnt_root);
735 return 1;
736 }
737 return 0;
738}
739
f1662356 740static __always_inline void follow_dotdot(struct nameidata *nd)
1da177e4 741{
e518ddb7
AM		 742 struct fs_struct *fs = current->fs;
743
1da177e4
LT		 744 while(1) {
745 struct vfsmount *parent;
4ac91378 746 struct dentry *old = nd->path.dentry;
1da177e4 747
e518ddb7 748 read_lock(&fs->lock);
6ac08c39
JB		 749 if (nd->path.dentry == fs->root.dentry &&
750 nd->path.mnt == fs->root.mnt) {
e518ddb7 751 read_unlock(&fs->lock);
1da177e4
LT		 752 break;
753 }
e518ddb7 754 read_unlock(&fs->lock);
1da177e4 755 spin_lock(&dcache_lock);
4ac91378
JB		 756 if (nd->path.dentry != nd->path.mnt->mnt_root) {
757 nd->path.dentry = dget(nd->path.dentry->d_parent);
1da177e4
LT		 758 spin_unlock(&dcache_lock);
759 dput(old);
760 break;
761 }
762 spin_unlock(&dcache_lock);
763 spin_lock(&vfsmount_lock);
4ac91378
JB		 764 parent = nd->path.mnt->mnt_parent;
765 if (parent == nd->path.mnt) {
1da177e4
LT		 766 spin_unlock(&vfsmount_lock);
767 break;
768 }
769 mntget(parent);
4ac91378 770 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
1da177e4
LT		 771 spin_unlock(&vfsmount_lock);
772 dput(old);
4ac91378
JB		 773 mntput(nd->path.mnt);
774 nd->path.mnt = parent;
1da177e4 775 }
4ac91378 776 follow_mount(&nd->path.mnt, &nd->path.dentry);
1da177e4
LT		 777}
778
1da177e4
LT		 779/*
780 * It's more convoluted than I'd like it to be, but... it's still fairly
781 * small and for now I'd prefer to have fast path as straight as possible.
782 * It _is_ time-critical.
783 */
784static int do_lookup(struct nameidata *nd, struct qstr *name,
785 struct path *path)
786{
4ac91378
JB		 787 struct vfsmount *mnt = nd->path.mnt;
788 struct dentry *dentry = __d_lookup(nd->path.dentry, name);
1da177e4
LT		 789
790 if (!dentry)
791 goto need_lookup;
792 if (dentry->d_op && dentry->d_op->d_revalidate)
793 goto need_revalidate;
794done:
795 path->mnt = mnt;
796 path->dentry = dentry;
634ee701 797 __follow_mount(path);
1da177e4
LT		 798 return 0;
799
800need_lookup:
4ac91378 801 dentry = real_lookup(nd->path.dentry, name, nd);
1da177e4
LT		 802 if (IS_ERR(dentry))
803 goto fail;
804 goto done;
805
806need_revalidate:
bcdc5e01
IK		 807 dentry = do_revalidate(dentry, nd);
808 if (!dentry)
809 goto need_lookup;
810 if (IS_ERR(dentry))
811 goto fail;
812 goto done;
1da177e4
LT		 813
814fail:
815 return PTR_ERR(dentry);
816}
817
818/*
819 * Name resolution.
ea3834d9
PM		 820 * This is the basic name resolution function, turning a pathname into
821 * the final dentry. We expect 'base' to be positive and a directory.
1da177e4 822 *
ea3834d9
PM		 823 * Returns 0 and nd will have valid dentry and mnt on success.
824 * Returns error and drops reference to input namei data on failure.
1da177e4 825 */
fc9b52cd 826static int __link_path_walk(const char *name, struct nameidata *nd)
1da177e4
LT		 827{
828 struct path next;
829 struct inode *inode;
830 int err;
831 unsigned int lookup_flags = nd->flags;
832
833 while (*name=='/')
834 name++;
835 if (!*name)
836 goto return_reval;
837
4ac91378 838 inode = nd->path.dentry->d_inode;
1da177e4 839 if (nd->depth)
f55eab82 840 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
1da177e4
LT		 841
842 /* At this point we know we have a real path component. */
843 for(;;) {
844 unsigned long hash;
845 struct qstr this;
846 unsigned int c;
847
cdce5d6b 848 nd->flags |= LOOKUP_CONTINUE;
672b16b2 849 err = exec_permission_lite(inode);
e4543edd 850 if (err == -EAGAIN)
cb23beb5
CH		 851 err = inode_permission(nd->path.dentry->d_inode,
852 MAY_EXEC);
1da177e4
LT		 853 if (err)
854 break;
855
856 this.name = name;
857 c = *(const unsigned char *)name;
858
859 hash = init_name_hash();
860 do {
861 name++;
862 hash = partial_name_hash(c, hash);
863 c = *(const unsigned char *)name;
864 } while (c && (c != '/'));
865 this.len = name - (const char *) this.name;
866 this.hash = end_name_hash(hash);
867
868 /* remove trailing slashes? */
869 if (!c)
870 goto last_component;
871 while (*++name == '/');
872 if (!*name)
873 goto last_with_slashes;
874
875 /*
876 * "." and ".." are special - ".." especially so because it has
877 * to be able to know about the current root directory and
878 * parent relationships.
879 */
880 if (this.name[0] == '.') switch (this.len) {
881 default:
882 break;
883 case 2:
884 if (this.name[1] != '.')
885 break;
58c465eb 886 follow_dotdot(nd);
4ac91378 887 inode = nd->path.dentry->d_inode;
1da177e4
LT		 888 /* fallthrough */
889 case 1:
890 continue;
891 }
892 /*
893 * See if the low-level filesystem might want
894 * to use its own hash..
895 */
4ac91378
JB		 896 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
897 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
898 &this);
1da177e4
LT		 899 if (err < 0)
900 break;
901 }
1da177e4
LT		 902 /* This does the actual lookups.. */
903 err = do_lookup(nd, &this, &next);
904 if (err)
905 break;
1da177e4
LT		 906
907 err = -ENOENT;
908 inode = next.dentry->d_inode;
909 if (!inode)
910 goto out_dput;
911 err = -ENOTDIR;
912 if (!inode->i_op)
913 goto out_dput;
914
915 if (inode->i_op->follow_link) {
90ebe565 916 err = do_follow_link(&next, nd);
1da177e4
LT		 917 if (err)
918 goto return_err;
919 err = -ENOENT;
4ac91378 920 inode = nd->path.dentry->d_inode;
1da177e4
LT		 921 if (!inode)
922 break;
923 err = -ENOTDIR;
924 if (!inode->i_op)
925 break;
09dd17d3
MS		 926 } else
927 path_to_nameidata(&next, nd);
1da177e4
LT		 928 err = -ENOTDIR;
929 if (!inode->i_op->lookup)
930 break;
931 continue;
932 /* here ends the main loop */
933
934last_with_slashes:
935 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
936last_component:
f55eab82
TM		 937 /* Clear LOOKUP_CONTINUE iff it was previously unset */
938 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
1da177e4
LT		 939 if (lookup_flags & LOOKUP_PARENT)
940 goto lookup_parent;
941 if (this.name[0] == '.') switch (this.len) {
942 default:
943 break;
944 case 2:
945 if (this.name[1] != '.')
946 break;
58c465eb 947 follow_dotdot(nd);
4ac91378 948 inode = nd->path.dentry->d_inode;
1da177e4
LT		 949 /* fallthrough */
950 case 1:
951 goto return_reval;
952 }
4ac91378
JB		 953 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
954 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
955 &this);
1da177e4
LT		 956 if (err < 0)
957 break;
958 }
959 err = do_lookup(nd, &this, &next);
960 if (err)
961 break;
1da177e4
LT		 962 inode = next.dentry->d_inode;
963 if ((lookup_flags & LOOKUP_FOLLOW)
964 && inode && inode->i_op && inode->i_op->follow_link) {
90ebe565 965 err = do_follow_link(&next, nd);
1da177e4
LT		 966 if (err)
967 goto return_err;
4ac91378 968 inode = nd->path.dentry->d_inode;
09dd17d3
MS		 969 } else
970 path_to_nameidata(&next, nd);
1da177e4
LT		 971 err = -ENOENT;
972 if (!inode)
973 break;
974 if (lookup_flags & LOOKUP_DIRECTORY) {
975 err = -ENOTDIR;
976 if (!inode->i_op || !inode->i_op->lookup)
977 break;
978 }
979 goto return_base;
980lookup_parent:
981 nd->last = this;
982 nd->last_type = LAST_NORM;
983 if (this.name[0] != '.')
984 goto return_base;
985 if (this.len == 1)
986 nd->last_type = LAST_DOT;
987 else if (this.len == 2 && this.name[1] == '.')
988 nd->last_type = LAST_DOTDOT;
989 else
990 goto return_base;
991return_reval:
992 /*
993 * We bypassed the ordinary revalidation routines.
994 * We may need to check the cached dentry for staleness.
995 */
4ac91378
JB		 996 if (nd->path.dentry && nd->path.dentry->d_sb &&
997 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
1da177e4
LT		 998 err = -ESTALE;
999 /* Note: we do not d_invalidate() */
4ac91378
JB		 1000 if (!nd->path.dentry->d_op->d_revalidate(
1001 nd->path.dentry, nd))
1da177e4
LT		 1002 break;
1003 }
1004return_base:
1005 return 0;
1006out_dput:
1d957f9b 1007 path_put_conditional(&next, nd);
1da177e4
LT		 1008 break;
1009 }
1d957f9b 1010 path_put(&nd->path);
1da177e4
LT		 1011return_err:
1012 return err;
1013}
1014
fc9b52cd 1015static int path_walk(const char *name, struct nameidata *nd)
1da177e4
LT		 1016{
1017 current->total_link_count = 0;
1018 return link_path_walk(name, nd);
1019}
1020
ea3834d9 1021/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
fc9b52cd 1022static int do_path_lookup(int dfd, const char *name,
5590ff0d 1023 unsigned int flags, struct nameidata *nd)
1da177e4 1024{
ea3834d9 1025 int retval = 0;
170aa3d0
UD		 1026 int fput_needed;
1027 struct file *file;
e518ddb7 1028 struct fs_struct *fs = current->fs;
1da177e4
LT		 1029
1030 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1031 nd->flags = flags;
1032 nd->depth = 0;
1033
1da177e4 1034 if (*name=='/') {
e518ddb7 1035 read_lock(&fs->lock);
6ac08c39
JB		 1036 nd->path = fs->root;
1037 path_get(&fs->root);
e518ddb7 1038 read_unlock(&fs->lock);
5590ff0d 1039 } else if (dfd == AT_FDCWD) {
e518ddb7 1040 read_lock(&fs->lock);
6ac08c39
JB		 1041 nd->path = fs->pwd;
1042 path_get(&fs->pwd);
e518ddb7 1043 read_unlock(&fs->lock);
5590ff0d 1044 } else {
5590ff0d
UD		 1045 struct dentry *dentry;
1046
1047 file = fget_light(dfd, &fput_needed);
170aa3d0
UD		 1048 retval = -EBADF;
1049 if (!file)
6d09bb62 1050 goto out_fail;
5590ff0d 1051
0f7fc9e4 1052 dentry = file->f_path.dentry;
5590ff0d 1053
170aa3d0
UD		 1054 retval = -ENOTDIR;
1055 if (!S_ISDIR(dentry->d_inode->i_mode))
6d09bb62 1056 goto fput_fail;
5590ff0d
UD		 1057
1058 retval = file_permission(file, MAY_EXEC);
170aa3d0 1059 if (retval)
6d09bb62 1060 goto fput_fail;
5590ff0d 1061
5dd784d0
JB		 1062 nd->path = file->f_path;
1063 path_get(&file->f_path);
5590ff0d
UD		 1064
1065 fput_light(file, fput_needed);
1da177e4 1066 }
2dfdd266
JJS		 1067
1068 retval = path_walk(name, nd);
4ac91378
JB		 1069 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1070 nd->path.dentry->d_inode))
1071 audit_inode(name, nd->path.dentry);
6d09bb62 1072out_fail:
170aa3d0
UD		 1073 return retval;
1074
6d09bb62 1075fput_fail:
170aa3d0 1076 fput_light(file, fput_needed);
6d09bb62 1077 goto out_fail;
1da177e4
LT		 1078}
1079
fc9b52cd 1080int path_lookup(const char *name, unsigned int flags,
5590ff0d
UD		 1081 struct nameidata *nd)
1082{
1083 return do_path_lookup(AT_FDCWD, name, flags, nd);
1084}
1085
d1811465
AV		 1086int kern_path(const char *name, unsigned int flags, struct path *path)
1087{
1088 struct nameidata nd;
1089 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1090 if (!res)
1091 *path = nd.path;
1092 return res;
1093}
1094
16f18200
JJS		 1095/**
1096 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1097 * @dentry: pointer to dentry of the base directory
1098 * @mnt: pointer to vfs mount of the base directory
1099 * @name: pointer to file name
1100 * @flags: lookup flags
1101 * @nd: pointer to nameidata
1102 */
1103int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1104 const char *name, unsigned int flags,
1105 struct nameidata *nd)
1106{
1107 int retval;
1108
1109 /* same as do_path_lookup */
1110 nd->last_type = LAST_ROOT;
1111 nd->flags = flags;
1112 nd->depth = 0;
1113
c8e7f449
JB		 1114 nd->path.dentry = dentry;
1115 nd->path.mnt = mnt;
1116 path_get(&nd->path);
16f18200
JJS		 1117
1118 retval = path_walk(name, nd);
4ac91378
JB		 1119 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1120 nd->path.dentry->d_inode))
1121 audit_inode(name, nd->path.dentry);
16f18200
JJS		 1122
1123 return retval;
1124
1125}
1126
8737f3a1
AV		 1127/**
1128 * path_lookup_open - lookup a file path with open intent
1129 * @dfd: the directory to use as base, or AT_FDCWD
1130 * @name: pointer to file name
1131 * @lookup_flags: lookup intent flags
1132 * @nd: pointer to nameidata
1133 * @open_flags: open intent flags
1134 */
1135int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
1136 struct nameidata *nd, int open_flags)
834f2a4a
TM		 1137{
1138 struct file *filp = get_empty_filp();
1139 int err;
1140
1141 if (filp == NULL)
1142 return -ENFILE;
1143 nd->intent.open.file = filp;
1144 nd->intent.open.flags = open_flags;
8737f3a1 1145 nd->intent.open.create_mode = 0;
5590ff0d 1146 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
834f2a4a
TM		 1147 if (IS_ERR(nd->intent.open.file)) {
1148 if (err == 0) {
1149 err = PTR_ERR(nd->intent.open.file);
1d957f9b 1150 path_put(&nd->path);
834f2a4a
TM		 1151 }
1152 } else if (err != 0)
1153 release_open_intent(nd);
1154 return err;
1155}
1156
eead1911
CH		 1157static struct dentry *__lookup_hash(struct qstr *name,
1158 struct dentry *base, struct nameidata *nd)
1da177e4 1159{
057f6c01 1160 struct dentry *dentry;
1da177e4
LT		 1161 struct inode *inode;
1162 int err;
1163
1164 inode = base->d_inode;
1da177e4
LT		 1165
1166 /*
1167 * See if the low-level filesystem might want
1168 * to use its own hash..
1169 */
1170 if (base->d_op && base->d_op->d_hash) {
1171 err = base->d_op->d_hash(base, name);
1172 dentry = ERR_PTR(err);
1173 if (err < 0)
1174 goto out;
1175 }
1176
1177 dentry = cached_lookup(base, name, nd);
1178 if (!dentry) {
d70b67c8
MS		 1179 struct dentry *new;
1180
1181 /* Don't create child dentry for a dead directory. */
1182 dentry = ERR_PTR(-ENOENT);
1183 if (IS_DEADDIR(inode))
1184 goto out;
1185
1186 new = d_alloc(base, name);
1da177e4
LT		 1187 dentry = ERR_PTR(-ENOMEM);
1188 if (!new)
1189 goto out;
1190 dentry = inode->i_op->lookup(inode, new, nd);
1191 if (!dentry)
1192 dentry = new;
1193 else
1194 dput(new);
1195 }
1196out:
1197 return dentry;
1198}
1199
057f6c01
JM		 1200/*
1201 * Restricted form of lookup. Doesn't follow links, single-component only,
1202 * needs parent already locked. Doesn't follow mounts.
1203 * SMP-safe.
1204 */
eead1911 1205static struct dentry *lookup_hash(struct nameidata *nd)
057f6c01 1206{
057f6c01
JM		 1207 int err;
1208
f419a2e3 1209 err = inode_permission(nd->path.dentry->d_inode, MAY_EXEC);
057f6c01 1210 if (err)
eead1911 1211 return ERR_PTR(err);
4ac91378 1212 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1da177e4
LT		 1213}
1214
eead1911
CH		 1215static int __lookup_one_len(const char *name, struct qstr *this,
1216 struct dentry *base, int len)
1da177e4
LT		 1217{
1218 unsigned long hash;
1da177e4
LT		 1219 unsigned int c;
1220
057f6c01
JM		 1221 this->name = name;
1222 this->len = len;
1da177e4 1223 if (!len)
057f6c01 1224 return -EACCES;
1da177e4
LT		 1225
1226 hash = init_name_hash();
1227 while (len--) {
1228 c = *(const unsigned char *)name++;
1229 if (c == '/' || c == '\0')
057f6c01 1230 return -EACCES;
1da177e4
LT		 1231 hash = partial_name_hash(c, hash);
1232 }
057f6c01
JM		 1233 this->hash = end_name_hash(hash);
1234 return 0;
1235}
1da177e4 1236
eead1911 1237/**
a6b91919 1238 * lookup_one_len - filesystem helper to lookup single pathname component
eead1911
CH		 1239 * @name: pathname component to lookup
1240 * @base: base directory to lookup from
1241 * @len: maximum length @len should be interpreted to
1242 *
a6b91919
RD		 1243 * Note that this routine is purely a helper for filesystem usage and should
1244 * not be called by generic code. Also note that by using this function the
eead1911
CH		 1245 * nameidata argument is passed to the filesystem methods and a filesystem
1246 * using this helper needs to be prepared for that.
1247 */
057f6c01
JM		 1248struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1249{
1250 int err;
1251 struct qstr this;
1252
1253 err = __lookup_one_len(name, &this, base, len);
eead1911
CH		 1254 if (err)
1255 return ERR_PTR(err);
1256
f419a2e3 1257 err = inode_permission(base->d_inode, MAY_EXEC);
057f6c01
JM		 1258 if (err)
1259 return ERR_PTR(err);
49705b77 1260 return __lookup_hash(&this, base, NULL);
057f6c01
JM		 1261}
1262
eead1911
CH		 1263/**
1264 * lookup_one_noperm - bad hack for sysfs
1265 * @name: pathname component to lookup
1266 * @base: base directory to lookup from
1267 *
1268 * This is a variant of lookup_one_len that doesn't perform any permission
1269 * checks. It's a horrible hack to work around the braindead sysfs
1270 * architecture and should not be used anywhere else.
1271 *
1272 * DON'T USE THIS FUNCTION EVER, thanks.
1273 */
1274struct dentry *lookup_one_noperm(const char *name, struct dentry *base)
057f6c01
JM		 1275{
1276 int err;
1277 struct qstr this;
1278
eead1911 1279 err = __lookup_one_len(name, &this, base, strlen(name));
057f6c01
JM		 1280 if (err)
1281 return ERR_PTR(err);
eead1911 1282 return __lookup_hash(&this, base, NULL);
1da177e4
LT		 1283}
1284
2d8f3038
AV		 1285int user_path_at(int dfd, const char __user *name, unsigned flags,
1286 struct path *path)
1da177e4 1287{
2d8f3038 1288 struct nameidata nd;
1da177e4
LT		 1289 char *tmp = getname(name);
1290 int err = PTR_ERR(tmp);
1da177e4 1291 if (!IS_ERR(tmp)) {
2d8f3038
AV		 1292
1293 BUG_ON(flags & LOOKUP_PARENT);
1294
1295 err = do_path_lookup(dfd, tmp, flags, &nd);
1da177e4 1296 putname(tmp);
2d8f3038
AV		 1297 if (!err)
1298 *path = nd.path;
1da177e4
LT		 1299 }
1300 return err;
1301}
1302
2ad94ae6
AV		 1303static int user_path_parent(int dfd, const char __user *path,
1304 struct nameidata *nd, char **name)
1305{
1306 char *s = getname(path);
1307 int error;
1308
1309 if (IS_ERR(s))
1310 return PTR_ERR(s);
1311
1312 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1313 if (error)
1314 putname(s);
1315 else
1316 *name = s;
1317
1318 return error;
1319}
1320
1da177e4
LT		 1321/*
1322 * It's inline, so penalty for filesystems that don't use sticky bit is
1323 * minimal.
1324 */
1325static inline int check_sticky(struct inode *dir, struct inode *inode)
1326{
da9592ed
DH		 1327 uid_t fsuid = current_fsuid();
1328
1da177e4
LT		 1329 if (!(dir->i_mode & S_ISVTX))
1330 return 0;
da9592ed 1331 if (inode->i_uid == fsuid)
1da177e4 1332 return 0;
da9592ed 1333 if (dir->i_uid == fsuid)
1da177e4
LT		 1334 return 0;
1335 return !capable(CAP_FOWNER);
1336}
1337
1338/*
1339 * Check whether we can remove a link victim from directory dir, check
1340 * whether the type of victim is right.
1341 * 1. We can't do it if dir is read-only (done in permission())
1342 * 2. We should have write and exec permissions on dir
1343 * 3. We can't remove anything from append-only dir
1344 * 4. We can't do anything with immutable dir (done in permission())
1345 * 5. If the sticky bit on dir is set we should either
1346 * a. be owner of dir, or
1347 * b. be owner of victim, or
1348 * c. have CAP_FOWNER capability
1349 * 6. If the victim is append-only or immutable we can't do antyhing with
1350 * links pointing to it.
1351 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1352 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1353 * 9. We can't remove a root or mountpoint.
1354 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1355 * nfs_async_unlink().
1356 */
858119e1 1357static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1da177e4
LT		 1358{
1359 int error;
1360
1361 if (!victim->d_inode)
1362 return -ENOENT;
1363
1364 BUG_ON(victim->d_parent->d_inode != dir);
5a190ae6 1365 audit_inode_child(victim->d_name.name, victim, dir);
1da177e4 1366
f419a2e3 1367 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1368 if (error)
1369 return error;
1370 if (IS_APPEND(dir))
1371 return -EPERM;
1372 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
f9454548 1373 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1da177e4
LT		 1374 return -EPERM;
1375 if (isdir) {
1376 if (!S_ISDIR(victim->d_inode->i_mode))
1377 return -ENOTDIR;
1378 if (IS_ROOT(victim))
1379 return -EBUSY;
1380 } else if (S_ISDIR(victim->d_inode->i_mode))
1381 return -EISDIR;
1382 if (IS_DEADDIR(dir))
1383 return -ENOENT;
1384 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1385 return -EBUSY;
1386 return 0;
1387}
1388
1389/* Check whether we can create an object with dentry child in directory
1390 * dir.
1391 * 1. We can't do it if child already exists (open has special treatment for
1392 * this case, but since we are inlined it's OK)
1393 * 2. We can't do it if dir is read-only (done in permission())
1394 * 3. We should have write and exec permissions on dir
1395 * 4. We can't do it if dir is immutable (done in permission())
1396 */
a95164d9 1397static inline int may_create(struct inode *dir, struct dentry *child)
1da177e4
LT		 1398{
1399 if (child->d_inode)
1400 return -EEXIST;
1401 if (IS_DEADDIR(dir))
1402 return -ENOENT;
f419a2e3 1403 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1404}
1405
1406/*
1da177e4
LT		 1407 * O_DIRECTORY translates into forcing a directory lookup.
1408 */
1409static inline int lookup_flags(unsigned int f)
1410{
1411 unsigned long retval = LOOKUP_FOLLOW;
1412
1413 if (f & O_NOFOLLOW)
1414 retval &= ~LOOKUP_FOLLOW;
1415
1da177e4
LT		 1416 if (f & O_DIRECTORY)
1417 retval |= LOOKUP_DIRECTORY;
1418
1419 return retval;
1420}
1421
1422/*
1423 * p1 and p2 should be directories on the same fs.
1424 */
1425struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1426{
1427 struct dentry *p;
1428
1429 if (p1 == p2) {
f2eace23 1430 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1da177e4
LT		 1431 return NULL;
1432 }
1433
a11f3a05 1434 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4 1435
e2761a11
OH		 1436 p = d_ancestor(p2, p1);
1437 if (p) {
1438 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1439 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1440 return p;
1da177e4
LT		 1441 }
1442
e2761a11
OH		 1443 p = d_ancestor(p1, p2);
1444 if (p) {
1445 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1446 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1447 return p;
1da177e4
LT		 1448 }
1449
f2eace23
IM		 1450 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1451 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1da177e4
LT		 1452 return NULL;
1453}
1454
1455void unlock_rename(struct dentry *p1, struct dentry *p2)
1456{
1b1dcc1b 1457 mutex_unlock(&p1->d_inode->i_mutex);
1da177e4 1458 if (p1 != p2) {
1b1dcc1b 1459 mutex_unlock(&p2->d_inode->i_mutex);
a11f3a05 1460 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4
LT		 1461 }
1462}
1463
1464int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1465 struct nameidata *nd)
1466{
a95164d9 1467 int error = may_create(dir, dentry);
1da177e4
LT		 1468
1469 if (error)
1470 return error;
1471
1472 if (!dir->i_op || !dir->i_op->create)
1473 return -EACCES; /* shouldn't it be ENOSYS? */
1474 mode &= S_IALLUGO;
1475 mode |= S_IFREG;
1476 error = security_inode_create(dir, dentry, mode);
1477 if (error)
1478 return error;
1479 DQUOT_INIT(dir);
1480 error = dir->i_op->create(dir, dentry, mode, nd);
a74574aa 1481 if (!error)
f38aa942 1482 fsnotify_create(dir, dentry);
1da177e4
LT		 1483 return error;
1484}
1485
3fb64190 1486int may_open(struct path *path, int acc_mode, int flag)
1da177e4 1487{
3fb64190 1488 struct dentry *dentry = path->dentry;
1da177e4
LT		 1489 struct inode *inode = dentry->d_inode;
1490 int error;
1491
1492 if (!inode)
1493 return -ENOENT;
1494
1495 if (S_ISLNK(inode->i_mode))
1496 return -ELOOP;
1497
974a9f0b 1498 if (S_ISDIR(inode->i_mode) && (acc_mode & MAY_WRITE))
1da177e4
LT		 1499 return -EISDIR;
1500
1da177e4
LT		 1501 /*
1502 * FIFO's, sockets and device files are special: they don't
1503 * actually live on the filesystem itself, and as such you
1504 * can write to them even if the filesystem is read-only.
1505 */
1506 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1507 flag &= ~O_TRUNC;
1508 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
3fb64190 1509 if (path->mnt->mnt_flags & MNT_NODEV)
1da177e4
LT		 1510 return -EACCES;
1511
1512 flag &= ~O_TRUNC;
4a3fd211 1513 }
b41572e9 1514
3fb64190 1515 error = inode_permission(inode, acc_mode);
b41572e9
DH		 1516 if (error)
1517 return error;
1da177e4
LT		 1518 /*
1519 * An append-only file must be opened in append mode for writing.
1520 */
1521 if (IS_APPEND(inode)) {
1522 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1523 return -EPERM;
1524 if (flag & O_TRUNC)
1525 return -EPERM;
1526 }
1527
1528 /* O_NOATIME can only be set by the owner or superuser */
1529 if (flag & O_NOATIME)
3bd858ab 1530 if (!is_owner_or_cap(inode))
1da177e4
LT		 1531 return -EPERM;
1532
1533 /*
1534 * Ensure there are no outstanding leases on the file.
1535 */
1536 error = break_lease(inode, flag);
1537 if (error)
1538 return error;
1539
1540 if (flag & O_TRUNC) {
1541 error = get_write_access(inode);
1542 if (error)
1543 return error;
1544
1545 /*
1546 * Refuse to truncate files with mandatory locks held on them.
1547 */
1548 error = locks_verify_locked(inode);
be6d3e56 1549 if (!error)
3fb64190 1550 error = security_path_truncate(path, 0,
be6d3e56 1551 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN);
1da177e4
LT		 1552 if (!error) {
1553 DQUOT_INIT(inode);
d139d7ff
MS		 1554
1555 error = do_truncate(dentry, 0,
1556 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1557 NULL);
1da177e4
LT		 1558 }
1559 put_write_access(inode);
1560 if (error)
1561 return error;
1562 } else
1563 if (flag & FMODE_WRITE)
1564 DQUOT_INIT(inode);
1565
1566 return 0;
1567}
1568
d57999e1
DH		 1569/*
1570 * Be careful about ever adding any more callers of this
1571 * function. Its flags must be in the namei format, not
1572 * what get passed to sys_open().
1573 */
1574static int __open_namei_create(struct nameidata *nd, struct path *path,
aab520e2
DH		 1575 int flag, int mode)
1576{
1577 int error;
4ac91378 1578 struct dentry *dir = nd->path.dentry;
aab520e2
DH		 1579
1580 if (!IS_POSIXACL(dir->d_inode))
1581 mode &= ~current->fs->umask;
be6d3e56
KT		 1582 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1583 if (error)
1584 goto out_unlock;
aab520e2 1585 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
be6d3e56 1586out_unlock:
aab520e2 1587 mutex_unlock(&dir->d_inode->i_mutex);
4ac91378
JB		 1588 dput(nd->path.dentry);
1589 nd->path.dentry = path->dentry;
aab520e2
DH		 1590 if (error)
1591 return error;
1592 /* Don't check for write permission, don't truncate */
3fb64190 1593 return may_open(&nd->path, 0, flag & ~O_TRUNC);
aab520e2
DH		 1594}
1595
d57999e1
DH		 1596/*
1597 * Note that while the flag value (low two bits) for sys_open means:
1598 * 00 - read-only
1599 * 01 - write-only
1600 * 10 - read-write
1601 * 11 - special
1602 * it is changed into
1603 * 00 - no permissions needed
1604 * 01 - read-permission
1605 * 10 - write-permission
1606 * 11 - read-write
1607 * for the internal routines (ie open_namei()/follow_link() etc)
1608 * This is more logical, and also allows the 00 "no perm needed"
1609 * to be used for symlinks (where the permissions are checked
1610 * later).
1611 *
1612*/
1613static inline int open_to_namei_flags(int flag)
1614{
1615 if ((flag+1) & O_ACCMODE)
1616 flag++;
1617 return flag;
1618}
1619
4a3fd211
DH		 1620static int open_will_write_to_fs(int flag, struct inode *inode)
1621{
1622 /*
1623 * We'll never write to the fs underlying
1624 * a device file.
1625 */
1626 if (special_file(inode->i_mode))
1627 return 0;
1628 return (flag & O_TRUNC);
1629}
1630
1da177e4 1631/*
4a3fd211
DH		 1632 * Note that the low bits of the passed in "open_flag"
1633 * are not the same as in the local variable "flag". See
1634 * open_to_namei_flags() for more details.
1da177e4 1635 */
a70e65df
CH		 1636struct file *do_filp_open(int dfd, const char *pathname,
1637 int open_flag, int mode)
1da177e4 1638{
4a3fd211 1639 struct file *filp;
a70e65df 1640 struct nameidata nd;
834f2a4a 1641 int acc_mode, error;
4e7506e4 1642 struct path path;
1da177e4
LT		 1643 struct dentry *dir;
1644 int count = 0;
4a3fd211 1645 int will_write;
d57999e1 1646 int flag = open_to_namei_flags(open_flag);
1da177e4 1647
b77b0646 1648 acc_mode = MAY_OPEN | ACC_MODE(flag);
1da177e4 1649
834f2a4a
TM		 1650 /* O_TRUNC implies we need access checks for write permissions */
1651 if (flag & O_TRUNC)
1652 acc_mode |= MAY_WRITE;
1653
1da177e4
LT		 1654 /* Allow the LSM permission hook to distinguish append
1655 access from general write access. */
1656 if (flag & O_APPEND)
1657 acc_mode |= MAY_APPEND;
1658
1da177e4
LT		 1659 /*
1660 * The simplest case - just a plain lookup.
1661 */
1662 if (!(flag & O_CREAT)) {
5590ff0d 1663 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
a70e65df 1664 &nd, flag);
1da177e4 1665 if (error)
a70e65df 1666 return ERR_PTR(error);
1da177e4
LT		 1667 goto ok;
1668 }
1669
1670 /*
1671 * Create - we need to know the parent.
1672 */
8737f3a1 1673 error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
1da177e4 1674 if (error)
a70e65df 1675 return ERR_PTR(error);
1da177e4
LT		 1676
1677 /*
1678 * We have the parent and last component. First of all, check
1679 * that we are not asked to creat(2) an obvious directory - that
1680 * will not do.
1681 */
1682 error = -EISDIR;
a70e65df 1683 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
8737f3a1 1684 goto exit_parent;
1da177e4 1685
8737f3a1
AV		 1686 error = -ENFILE;
1687 filp = get_empty_filp();
1688 if (filp == NULL)
1689 goto exit_parent;
1690 nd.intent.open.file = filp;
1691 nd.intent.open.flags = flag;
1692 nd.intent.open.create_mode = mode;
a70e65df
CH		 1693 dir = nd.path.dentry;
1694 nd.flags &= ~LOOKUP_PARENT;
8737f3a1 1695 nd.flags |= LOOKUP_CREATE | LOOKUP_OPEN;
3516586a
AV		 1696 if (flag & O_EXCL)
1697 nd.flags |= LOOKUP_EXCL;
1b1dcc1b 1698 mutex_lock(&dir->d_inode->i_mutex);
a70e65df
CH		 1699 path.dentry = lookup_hash(&nd);
1700 path.mnt = nd.path.mnt;
1da177e4
LT		 1701
1702do_last:
4e7506e4
AV		 1703 error = PTR_ERR(path.dentry);
1704 if (IS_ERR(path.dentry)) {
1b1dcc1b 1705 mutex_unlock(&dir->d_inode->i_mutex);
1da177e4
LT		 1706 goto exit;
1707 }
1708
a70e65df 1709 if (IS_ERR(nd.intent.open.file)) {
a70e65df 1710 error = PTR_ERR(nd.intent.open.file);
4a3fd211 1711 goto exit_mutex_unlock;
4af4c52f
OD		 1712 }
1713
1da177e4 1714 /* Negative dentry, just create the file */
4e7506e4 1715 if (!path.dentry->d_inode) {
4a3fd211
DH		 1716 /*
1717 * This write is needed to ensure that a
1718 * ro->rw transition does not occur between
1719 * the time when the file is created and when
1720 * a permanent write count is taken through
1721 * the 'struct file' in nameidata_to_filp().
1722 */
1723 error = mnt_want_write(nd.path.mnt);
1da177e4 1724 if (error)
4a3fd211
DH		 1725 goto exit_mutex_unlock;
1726 error = __open_namei_create(&nd, &path, flag, mode);
1727 if (error) {
1728 mnt_drop_write(nd.path.mnt);
1da177e4 1729 goto exit;
4a3fd211
DH		 1730 }
1731 filp = nameidata_to_filp(&nd, open_flag);
1732 mnt_drop_write(nd.path.mnt);
1733 return filp;
1da177e4
LT		 1734 }
1735
1736 /*
1737 * It already exists.
1738 */
1b1dcc1b 1739 mutex_unlock(&dir->d_inode->i_mutex);
5a190ae6 1740 audit_inode(pathname, path.dentry);
1da177e4
LT		 1741
1742 error = -EEXIST;
1743 if (flag & O_EXCL)
1744 goto exit_dput;
1745
e13b210f 1746 if (__follow_mount(&path)) {
1da177e4 1747 error = -ELOOP;
ba7a4c1a
AV		 1748 if (flag & O_NOFOLLOW)
1749 goto exit_dput;
1da177e4 1750 }
3e2efce0 1751
1da177e4 1752 error = -ENOENT;
4e7506e4 1753 if (!path.dentry->d_inode)
1da177e4 1754 goto exit_dput;
4e7506e4 1755 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
1da177e4
LT		 1756 goto do_link;
1757
a70e65df 1758 path_to_nameidata(&path, &nd);
1da177e4 1759 error = -EISDIR;
4e7506e4 1760 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1da177e4
LT		 1761 goto exit;
1762ok:
4a3fd211
DH		 1763 /*
1764 * Consider:
1765 * 1. may_open() truncates a file
1766 * 2. a rw->ro mount transition occurs
1767 * 3. nameidata_to_filp() fails due to
1768 * the ro mount.
1769 * That would be inconsistent, and should
1770 * be avoided. Taking this mnt write here
1771 * ensures that (2) can not occur.
1772 */
1773 will_write = open_will_write_to_fs(flag, nd.path.dentry->d_inode);
1774 if (will_write) {
1775 error = mnt_want_write(nd.path.mnt);
1776 if (error)
1777 goto exit;
1778 }
3fb64190 1779 error = may_open(&nd.path, acc_mode, flag);
4a3fd211
DH		 1780 if (error) {
1781 if (will_write)
1782 mnt_drop_write(nd.path.mnt);
1da177e4 1783 goto exit;
4a3fd211
DH		 1784 }
1785 filp = nameidata_to_filp(&nd, open_flag);
1786 /*
1787 * It is now safe to drop the mnt write
1788 * because the filp has had a write taken
1789 * on its behalf.
1790 */
1791 if (will_write)
1792 mnt_drop_write(nd.path.mnt);
1793 return filp;
1da177e4 1794
4a3fd211
DH		 1795exit_mutex_unlock:
1796 mutex_unlock(&dir->d_inode->i_mutex);
1da177e4 1797exit_dput:
a70e65df 1798 path_put_conditional(&path, &nd);
1da177e4 1799exit:
a70e65df
CH		 1800 if (!IS_ERR(nd.intent.open.file))
1801 release_open_intent(&nd);
8737f3a1 1802exit_parent:
a70e65df
CH		 1803 path_put(&nd.path);
1804 return ERR_PTR(error);
1da177e4
LT		 1805
1806do_link:
1807 error = -ELOOP;
1808 if (flag & O_NOFOLLOW)
1809 goto exit_dput;
1810 /*
1811 * This is subtle. Instead of calling do_follow_link() we do the
1812 * thing by hands. The reason is that this way we have zero link_count
1813 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1814 * After that we have the parent and last component, i.e.
1815 * we are in the same situation as after the first path_walk().
1816 * Well, almost - if the last component is normal we get its copy
1817 * stored in nd->last.name and we will have to putname() it when we
1818 * are done. Procfs-like symlinks just set LAST_BIND.
1819 */
a70e65df
CH		 1820 nd.flags |= LOOKUP_PARENT;
1821 error = security_inode_follow_link(path.dentry, &nd);
1da177e4
LT		 1822 if (error)
1823 goto exit_dput;
a70e65df 1824 error = __do_follow_link(&path, &nd);
de459215
KK		 1825 if (error) {
1826 /* Does someone understand code flow here? Or it is only
1827 * me so stupid? Anathema to whoever designed this non-sense
1828 * with "intent.open".
1829 */
a70e65df
CH		 1830 release_open_intent(&nd);
1831 return ERR_PTR(error);
de459215 1832 }
a70e65df
CH		 1833 nd.flags &= ~LOOKUP_PARENT;
1834 if (nd.last_type == LAST_BIND)
1da177e4 1835 goto ok;
1da177e4 1836 error = -EISDIR;
a70e65df 1837 if (nd.last_type != LAST_NORM)
1da177e4 1838 goto exit;
a70e65df
CH		 1839 if (nd.last.name[nd.last.len]) {
1840 __putname(nd.last.name);
1da177e4
LT		 1841 goto exit;
1842 }
1843 error = -ELOOP;
1844 if (count++==32) {
a70e65df 1845 __putname(nd.last.name);
1da177e4
LT		 1846 goto exit;
1847 }
a70e65df 1848 dir = nd.path.dentry;
1b1dcc1b 1849 mutex_lock(&dir->d_inode->i_mutex);
a70e65df
CH		 1850 path.dentry = lookup_hash(&nd);
1851 path.mnt = nd.path.mnt;
1852 __putname(nd.last.name);
1da177e4
LT		 1853 goto do_last;
1854}
1855
a70e65df
CH		 1856/**
1857 * filp_open - open file and return file pointer
1858 *
1859 * @filename: path to open
1860 * @flags: open flags as per the open(2) second argument
1861 * @mode: mode for the new file if O_CREAT is set, else ignored
1862 *
1863 * This is the helper to open a file from kernelspace if you really
1864 * have to. But in generally you should not do this, so please move
1865 * along, nothing to see here..
1866 */
1867struct file *filp_open(const char *filename, int flags, int mode)
1868{
1869 return do_filp_open(AT_FDCWD, filename, flags, mode);
1870}
1871EXPORT_SYMBOL(filp_open);
1872
1da177e4
LT		 1873/**
1874 * lookup_create - lookup a dentry, creating it if it doesn't exist
1875 * @nd: nameidata info
1876 * @is_dir: directory flag
1877 *
1878 * Simple function to lookup and return a dentry and create it
1879 * if it doesn't exist. Is SMP-safe.
c663e5d8 1880 *
4ac91378 1881 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1da177e4
LT		 1882 */
1883struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1884{
c663e5d8 1885 struct dentry *dentry = ERR_PTR(-EEXIST);
1da177e4 1886
4ac91378 1887 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
c663e5d8
CH		 1888 /*
1889 * Yucky last component or no last component at all?
1890 * (foo/., foo/.., /////)
1891 */
1da177e4
LT		 1892 if (nd->last_type != LAST_NORM)
1893 goto fail;
1894 nd->flags &= ~LOOKUP_PARENT;
3516586a 1895 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
a634904a 1896 nd->intent.open.flags = O_EXCL;
c663e5d8
CH		 1897
1898 /*
1899 * Do the final lookup.
1900 */
49705b77 1901 dentry = lookup_hash(nd);
1da177e4
LT		 1902 if (IS_ERR(dentry))
1903 goto fail;
c663e5d8 1904
e9baf6e5
AV		 1905 if (dentry->d_inode)
1906 goto eexist;
c663e5d8
CH		 1907 /*
1908 * Special case - lookup gave negative, but... we had foo/bar/
1909 * From the vfs_mknod() POV we just have a negative dentry -
1910 * all is fine. Let's be bastards - you had / on the end, you've
1911 * been asking for (non-existent) directory. -ENOENT for you.
1912 */
e9baf6e5
AV		 1913 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1914 dput(dentry);
1915 dentry = ERR_PTR(-ENOENT);
1916 }
1da177e4 1917 return dentry;
e9baf6e5 1918eexist:
1da177e4 1919 dput(dentry);
e9baf6e5 1920 dentry = ERR_PTR(-EEXIST);
1da177e4
LT		 1921fail:
1922 return dentry;
1923}
f81a0bff 1924EXPORT_SYMBOL_GPL(lookup_create);
1da177e4
LT		 1925
1926int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1927{
a95164d9 1928 int error = may_create(dir, dentry);
1da177e4
LT		 1929
1930 if (error)
1931 return error;
1932
1933 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1934 return -EPERM;
1935
1936 if (!dir->i_op || !dir->i_op->mknod)
1937 return -EPERM;
1938
08ce5f16
SH		 1939 error = devcgroup_inode_mknod(mode, dev);
1940 if (error)
1941 return error;
1942
1da177e4
LT		 1943 error = security_inode_mknod(dir, dentry, mode, dev);
1944 if (error)
1945 return error;
1946
1947 DQUOT_INIT(dir);
1948 error = dir->i_op->mknod(dir, dentry, mode, dev);
a74574aa 1949 if (!error)
f38aa942 1950 fsnotify_create(dir, dentry);
1da177e4
LT		 1951 return error;
1952}
1953
463c3197
DH		 1954static int may_mknod(mode_t mode)
1955{
1956 switch (mode & S_IFMT) {
1957 case S_IFREG:
1958 case S_IFCHR:
1959 case S_IFBLK:
1960 case S_IFIFO:
1961 case S_IFSOCK:
1962 case 0: /* zero mode translates to S_IFREG */
1963 return 0;
1964 case S_IFDIR:
1965 return -EPERM;
1966 default:
1967 return -EINVAL;
1968 }
1969}
1970
5590ff0d
UD		 1971asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
1972 unsigned dev)
1da177e4 1973{
2ad94ae6
AV		 1974 int error;
1975 char *tmp;
1976 struct dentry *dentry;
1da177e4
LT		 1977 struct nameidata nd;
1978
1979 if (S_ISDIR(mode))
1980 return -EPERM;
1da177e4 1981
2ad94ae6 1982 error = user_path_parent(dfd, filename, &nd, &tmp);
1da177e4 1983 if (error)
2ad94ae6
AV		 1984 return error;
1985
1da177e4 1986 dentry = lookup_create(&nd, 0);
463c3197
DH		 1987 if (IS_ERR(dentry)) {
1988 error = PTR_ERR(dentry);
1989 goto out_unlock;
1990 }
4ac91378 1991 if (!IS_POSIXACL(nd.path.dentry->d_inode))
1da177e4 1992 mode &= ~current->fs->umask;
463c3197
DH		 1993 error = may_mknod(mode);
1994 if (error)
1995 goto out_dput;
1996 error = mnt_want_write(nd.path.mnt);
1997 if (error)
1998 goto out_dput;
be6d3e56
KT		 1999 error = security_path_mknod(&nd.path, dentry, mode, dev);
2000 if (error)
2001 goto out_drop_write;
463c3197 2002 switch (mode & S_IFMT) {
1da177e4 2003 case 0: case S_IFREG:
4ac91378 2004 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
1da177e4
LT		 2005 break;
2006 case S_IFCHR: case S_IFBLK:
4ac91378 2007 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
1da177e4
LT		 2008 new_decode_dev(dev));
2009 break;
2010 case S_IFIFO: case S_IFSOCK:
4ac91378 2011 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
1da177e4 2012 break;
1da177e4 2013 }
be6d3e56 2014out_drop_write:
463c3197
DH		 2015 mnt_drop_write(nd.path.mnt);
2016out_dput:
2017 dput(dentry);
2018out_unlock:
4ac91378 2019 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2020 path_put(&nd.path);
1da177e4
LT		 2021 putname(tmp);
2022
2023 return error;
2024}
2025
5590ff0d
UD		 2026asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
2027{
2028 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2029}
2030
1da177e4
LT		 2031int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2032{
a95164d9 2033 int error = may_create(dir, dentry);
1da177e4
LT		 2034
2035 if (error)
2036 return error;
2037
2038 if (!dir->i_op || !dir->i_op->mkdir)
2039 return -EPERM;
2040
2041 mode &= (S_IRWXUGO|S_ISVTX);
2042 error = security_inode_mkdir(dir, dentry, mode);
2043 if (error)
2044 return error;
2045
2046 DQUOT_INIT(dir);
2047 error = dir->i_op->mkdir(dir, dentry, mode);
a74574aa 2048 if (!error)
f38aa942 2049 fsnotify_mkdir(dir, dentry);
1da177e4
LT		 2050 return error;
2051}
2052
5590ff0d 2053asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
1da177e4
LT		 2054{
2055 int error = 0;
2056 char * tmp;
6902d925
DH		 2057 struct dentry *dentry;
2058 struct nameidata nd;
1da177e4 2059
2ad94ae6
AV		 2060 error = user_path_parent(dfd, pathname, &nd, &tmp);
2061 if (error)
6902d925 2062 goto out_err;
1da177e4 2063
6902d925
DH		 2064 dentry = lookup_create(&nd, 1);
2065 error = PTR_ERR(dentry);
2066 if (IS_ERR(dentry))
2067 goto out_unlock;
1da177e4 2068
4ac91378 2069 if (!IS_POSIXACL(nd.path.dentry->d_inode))
6902d925 2070 mode &= ~current->fs->umask;
463c3197
DH		 2071 error = mnt_want_write(nd.path.mnt);
2072 if (error)
2073 goto out_dput;
be6d3e56
KT		 2074 error = security_path_mkdir(&nd.path, dentry, mode);
2075 if (error)
2076 goto out_drop_write;
4ac91378 2077 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
be6d3e56 2078out_drop_write:
463c3197
DH		 2079 mnt_drop_write(nd.path.mnt);
2080out_dput:
6902d925
DH		 2081 dput(dentry);
2082out_unlock:
4ac91378 2083 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2084 path_put(&nd.path);
6902d925
DH		 2085 putname(tmp);
2086out_err:
1da177e4
LT		 2087 return error;
2088}
2089
5590ff0d
UD		 2090asmlinkage long sys_mkdir(const char __user *pathname, int mode)
2091{
2092 return sys_mkdirat(AT_FDCWD, pathname, mode);
2093}
2094
1da177e4
LT		 2095/*
2096 * We try to drop the dentry early: we should have
2097 * a usage count of 2 if we're the only user of this
2098 * dentry, and if that is true (possibly after pruning
2099 * the dcache), then we drop the dentry now.
2100 *
2101 * A low-level filesystem can, if it choses, legally
2102 * do a
2103 *
2104 * if (!d_unhashed(dentry))
2105 * return -EBUSY;
2106 *
2107 * if it cannot handle the case of removing a directory
2108 * that is still in use by something else..
2109 */
2110void dentry_unhash(struct dentry *dentry)
2111{
2112 dget(dentry);
dc168427 2113 shrink_dcache_parent(dentry);
1da177e4
LT		 2114 spin_lock(&dcache_lock);
2115 spin_lock(&dentry->d_lock);
2116 if (atomic_read(&dentry->d_count) == 2)
2117 __d_drop(dentry);
2118 spin_unlock(&dentry->d_lock);
2119 spin_unlock(&dcache_lock);
2120}
2121
2122int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2123{
2124 int error = may_delete(dir, dentry, 1);
2125
2126 if (error)
2127 return error;
2128
2129 if (!dir->i_op || !dir->i_op->rmdir)
2130 return -EPERM;
2131
2132 DQUOT_INIT(dir);
2133
1b1dcc1b 2134 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2135 dentry_unhash(dentry);
2136 if (d_mountpoint(dentry))
2137 error = -EBUSY;
2138 else {
2139 error = security_inode_rmdir(dir, dentry);
2140 if (!error) {
2141 error = dir->i_op->rmdir(dir, dentry);
2142 if (!error)
2143 dentry->d_inode->i_flags |= S_DEAD;
2144 }
2145 }
1b1dcc1b 2146 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4 2147 if (!error) {
1da177e4
LT		 2148 d_delete(dentry);
2149 }
2150 dput(dentry);
2151
2152 return error;
2153}
2154
5590ff0d 2155static long do_rmdir(int dfd, const char __user *pathname)
1da177e4
LT		 2156{
2157 int error = 0;
2158 char * name;
2159 struct dentry *dentry;
2160 struct nameidata nd;
2161
2ad94ae6 2162 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2163 if (error)
2ad94ae6 2164 return error;
1da177e4
LT		 2165
2166 switch(nd.last_type) {
0612d9fb
OH		 2167 case LAST_DOTDOT:
2168 error = -ENOTEMPTY;
2169 goto exit1;
2170 case LAST_DOT:
2171 error = -EINVAL;
2172 goto exit1;
2173 case LAST_ROOT:
2174 error = -EBUSY;
2175 goto exit1;
1da177e4 2176 }
0612d9fb
OH		 2177
2178 nd.flags &= ~LOOKUP_PARENT;
2179
4ac91378 2180 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2181 dentry = lookup_hash(&nd);
1da177e4 2182 error = PTR_ERR(dentry);
6902d925
DH		 2183 if (IS_ERR(dentry))
2184 goto exit2;
0622753b
DH		 2185 error = mnt_want_write(nd.path.mnt);
2186 if (error)
2187 goto exit3;
be6d3e56
KT		 2188 error = security_path_rmdir(&nd.path, dentry);
2189 if (error)
2190 goto exit4;
4ac91378 2191 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
be6d3e56 2192exit4:
0622753b
DH		 2193 mnt_drop_write(nd.path.mnt);
2194exit3:
6902d925
DH		 2195 dput(dentry);
2196exit2:
4ac91378 2197 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2198exit1:
1d957f9b 2199 path_put(&nd.path);
1da177e4
LT		 2200 putname(name);
2201 return error;
2202}
2203
5590ff0d
UD		 2204asmlinkage long sys_rmdir(const char __user *pathname)
2205{
2206 return do_rmdir(AT_FDCWD, pathname);
2207}
2208
1da177e4
LT		 2209int vfs_unlink(struct inode *dir, struct dentry *dentry)
2210{
2211 int error = may_delete(dir, dentry, 0);
2212
2213 if (error)
2214 return error;
2215
2216 if (!dir->i_op || !dir->i_op->unlink)
2217 return -EPERM;
2218
2219 DQUOT_INIT(dir);
2220
1b1dcc1b 2221 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2222 if (d_mountpoint(dentry))
2223 error = -EBUSY;
2224 else {
2225 error = security_inode_unlink(dir, dentry);
2226 if (!error)
2227 error = dir->i_op->unlink(dir, dentry);
2228 }
1b1dcc1b 2229 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2230
2231 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2232 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
ece95912 2233 fsnotify_link_count(dentry->d_inode);
e234f35c 2234 d_delete(dentry);
1da177e4 2235 }
0eeca283 2236
1da177e4
LT		 2237 return error;
2238}
2239
2240/*
2241 * Make sure that the actual truncation of the file will occur outside its
1b1dcc1b 2242 * directory's i_mutex. Truncate can take a long time if there is a lot of
1da177e4
LT		 2243 * writeout happening, and we don't want to prevent access to the directory
2244 * while waiting on the I/O.
2245 */
5590ff0d 2246static long do_unlinkat(int dfd, const char __user *pathname)
1da177e4 2247{
2ad94ae6
AV		 2248 int error;
2249 char *name;
1da177e4
LT		 2250 struct dentry *dentry;
2251 struct nameidata nd;
2252 struct inode *inode = NULL;
2253
2ad94ae6 2254 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2255 if (error)
2ad94ae6
AV		 2256 return error;
2257
1da177e4
LT		 2258 error = -EISDIR;
2259 if (nd.last_type != LAST_NORM)
2260 goto exit1;
0612d9fb
OH		 2261
2262 nd.flags &= ~LOOKUP_PARENT;
2263
4ac91378 2264 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2265 dentry = lookup_hash(&nd);
1da177e4
LT		 2266 error = PTR_ERR(dentry);
2267 if (!IS_ERR(dentry)) {
2268 /* Why not before? Because we want correct error value */
2269 if (nd.last.name[nd.last.len])
2270 goto slashes;
2271 inode = dentry->d_inode;
2272 if (inode)
2273 atomic_inc(&inode->i_count);
0622753b
DH		 2274 error = mnt_want_write(nd.path.mnt);
2275 if (error)
2276 goto exit2;
be6d3e56
KT		 2277 error = security_path_unlink(&nd.path, dentry);
2278 if (error)
2279 goto exit3;
4ac91378 2280 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
be6d3e56 2281exit3:
0622753b 2282 mnt_drop_write(nd.path.mnt);
1da177e4
LT		 2283 exit2:
2284 dput(dentry);
2285 }
4ac91378 2286 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4
LT		 2287 if (inode)
2288 iput(inode); /* truncate the inode here */
2289exit1:
1d957f9b 2290 path_put(&nd.path);
1da177e4
LT		 2291 putname(name);
2292 return error;
2293
2294slashes:
2295 error = !dentry->d_inode ? -ENOENT :
2296 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2297 goto exit2;
2298}
2299
5590ff0d
UD		 2300asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2301{
2302 if ((flag & ~AT_REMOVEDIR) != 0)
2303 return -EINVAL;
2304
2305 if (flag & AT_REMOVEDIR)
2306 return do_rmdir(dfd, pathname);
2307
2308 return do_unlinkat(dfd, pathname);
2309}
2310
2311asmlinkage long sys_unlink(const char __user *pathname)
2312{
2313 return do_unlinkat(AT_FDCWD, pathname);
2314}
2315
db2e747b 2316int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
1da177e4 2317{
a95164d9 2318 int error = may_create(dir, dentry);
1da177e4
LT		 2319
2320 if (error)
2321 return error;
2322
2323 if (!dir->i_op || !dir->i_op->symlink)
2324 return -EPERM;
2325
2326 error = security_inode_symlink(dir, dentry, oldname);
2327 if (error)
2328 return error;
2329
2330 DQUOT_INIT(dir);
2331 error = dir->i_op->symlink(dir, dentry, oldname);
a74574aa 2332 if (!error)
f38aa942 2333 fsnotify_create(dir, dentry);
1da177e4
LT		 2334 return error;
2335}
2336
5590ff0d
UD		 2337asmlinkage long sys_symlinkat(const char __user *oldname,
2338 int newdfd, const char __user *newname)
1da177e4 2339{
2ad94ae6
AV		 2340 int error;
2341 char *from;
2342 char *to;
6902d925
DH		 2343 struct dentry *dentry;
2344 struct nameidata nd;
1da177e4
LT		 2345
2346 from = getname(oldname);
2ad94ae6 2347 if (IS_ERR(from))
1da177e4 2348 return PTR_ERR(from);
1da177e4 2349
2ad94ae6 2350 error = user_path_parent(newdfd, newname, &nd, &to);
6902d925 2351 if (error)
2ad94ae6
AV		 2352 goto out_putname;
2353
6902d925
DH		 2354 dentry = lookup_create(&nd, 0);
2355 error = PTR_ERR(dentry);
2356 if (IS_ERR(dentry))
2357 goto out_unlock;
2358
75c3f29d
DH		 2359 error = mnt_want_write(nd.path.mnt);
2360 if (error)
2361 goto out_dput;
be6d3e56
KT		 2362 error = security_path_symlink(&nd.path, dentry, from);
2363 if (error)
2364 goto out_drop_write;
db2e747b 2365 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
be6d3e56 2366out_drop_write:
75c3f29d
DH		 2367 mnt_drop_write(nd.path.mnt);
2368out_dput:
6902d925
DH		 2369 dput(dentry);
2370out_unlock:
4ac91378 2371 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2372 path_put(&nd.path);
6902d925
DH		 2373 putname(to);
2374out_putname:
1da177e4
LT		 2375 putname(from);
2376 return error;
2377}
2378
5590ff0d
UD		 2379asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2380{
2381 return sys_symlinkat(oldname, AT_FDCWD, newname);
2382}
2383
1da177e4
LT		 2384int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2385{
2386 struct inode *inode = old_dentry->d_inode;
2387 int error;
2388
2389 if (!inode)
2390 return -ENOENT;
2391
a95164d9 2392 error = may_create(dir, new_dentry);
1da177e4
LT		 2393 if (error)
2394 return error;
2395
2396 if (dir->i_sb != inode->i_sb)
2397 return -EXDEV;
2398
2399 /*
2400 * A link to an append-only or immutable file cannot be created.
2401 */
2402 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2403 return -EPERM;
2404 if (!dir->i_op || !dir->i_op->link)
2405 return -EPERM;
7e79eedb 2406 if (S_ISDIR(inode->i_mode))
1da177e4
LT		 2407 return -EPERM;
2408
2409 error = security_inode_link(old_dentry, dir, new_dentry);
2410 if (error)
2411 return error;
2412
7e79eedb 2413 mutex_lock(&inode->i_mutex);
1da177e4
LT		 2414 DQUOT_INIT(dir);
2415 error = dir->i_op->link(old_dentry, dir, new_dentry);
7e79eedb 2416 mutex_unlock(&inode->i_mutex);
e31e14ec 2417 if (!error)
7e79eedb 2418 fsnotify_link(dir, inode, new_dentry);
1da177e4
LT		 2419 return error;
2420}
2421
2422/*
2423 * Hardlinks are often used in delicate situations. We avoid
2424 * security-related surprises by not following symlinks on the
2425 * newname. --KAB
2426 *
2427 * We don't follow them on the oldname either to be compatible
2428 * with linux 2.0, and to avoid hard-linking to directories
2429 * and other special files. --ADM
2430 */
5590ff0d 2431asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
c04030e1
UD		 2432 int newdfd, const char __user *newname,
2433 int flags)
1da177e4
LT		 2434{
2435 struct dentry *new_dentry;
2d8f3038
AV		 2436 struct nameidata nd;
2437 struct path old_path;
1da177e4 2438 int error;
2ad94ae6 2439 char *to;
1da177e4 2440
45c9b11a 2441 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
c04030e1
UD		 2442 return -EINVAL;
2443
2d8f3038
AV		 2444 error = user_path_at(olddfd, oldname,
2445 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2446 &old_path);
1da177e4 2447 if (error)
2ad94ae6
AV		 2448 return error;
2449
2450 error = user_path_parent(newdfd, newname, &nd, &to);
1da177e4
LT		 2451 if (error)
2452 goto out;
2453 error = -EXDEV;
2d8f3038 2454 if (old_path.mnt != nd.path.mnt)
1da177e4
LT		 2455 goto out_release;
2456 new_dentry = lookup_create(&nd, 0);
2457 error = PTR_ERR(new_dentry);
6902d925
DH		 2458 if (IS_ERR(new_dentry))
2459 goto out_unlock;
75c3f29d
DH		 2460 error = mnt_want_write(nd.path.mnt);
2461 if (error)
2462 goto out_dput;
be6d3e56
KT		 2463 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2464 if (error)
2465 goto out_drop_write;
2d8f3038 2466 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
be6d3e56 2467out_drop_write:
75c3f29d
DH		 2468 mnt_drop_write(nd.path.mnt);
2469out_dput:
6902d925
DH		 2470 dput(new_dentry);
2471out_unlock:
4ac91378 2472 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2473out_release:
1d957f9b 2474 path_put(&nd.path);
2ad94ae6 2475 putname(to);
1da177e4 2476out:
2d8f3038 2477 path_put(&old_path);
1da177e4
LT		 2478
2479 return error;
2480}
2481
5590ff0d
UD		 2482asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2483{
c04030e1 2484 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
5590ff0d
UD		 2485}
2486
1da177e4
LT		 2487/*
2488 * The worst of all namespace operations - renaming directory. "Perverted"
2489 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2490 * Problems:
2491 * a) we can get into loop creation. Check is done in is_subdir().
2492 * b) race potential - two innocent renames can create a loop together.
2493 * That's where 4.4 screws up. Current fix: serialization on
a11f3a05 2494 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
1da177e4
LT		 2495 * story.
2496 * c) we have to lock _three_ objects - parents and victim (if it exists).
1b1dcc1b 2497 * And that - after we got ->i_mutex on parents (until then we don't know
1da177e4
LT		 2498 * whether the target exists). Solution: try to be smart with locking
2499 * order for inodes. We rely on the fact that tree topology may change
a11f3a05 2500 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
1da177e4
LT		 2501 * move will be locked. Thus we can rank directories by the tree
2502 * (ancestors first) and rank all non-directories after them.
2503 * That works since everybody except rename does "lock parent, lookup,
a11f3a05 2504 * lock child" and rename is under ->s_vfs_rename_mutex.
1da177e4
LT		 2505 * HOWEVER, it relies on the assumption that any object with ->lookup()
2506 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2507 * we'd better make sure that there's no link(2) for them.
2508 * d) some filesystems don't support opened-but-unlinked directories,
2509 * either because of layout or because they are not ready to deal with
2510 * all cases correctly. The latter will be fixed (taking this sort of
2511 * stuff into VFS), but the former is not going away. Solution: the same
2512 * trick as in rmdir().
2513 * e) conversion from fhandle to dentry may come in the wrong moment - when
1b1dcc1b 2514 * we are removing the target. Solution: we will have to grab ->i_mutex
1da177e4 2515 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
1b1dcc1b 2516 * ->i_mutex on parents, which works but leads to some truely excessive
1da177e4
LT		 2517 * locking].
2518 */
75c96f85
AB		 2519static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2520 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 2521{
2522 int error = 0;
2523 struct inode *target;
2524
2525 /*
2526 * If we are going to change the parent - check write permissions,
2527 * we'll need to flip '..'.
2528 */
2529 if (new_dir != old_dir) {
f419a2e3 2530 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
1da177e4
LT		 2531 if (error)
2532 return error;
2533 }
2534
2535 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2536 if (error)
2537 return error;
2538
2539 target = new_dentry->d_inode;
2540 if (target) {
1b1dcc1b 2541 mutex_lock(&target->i_mutex);
1da177e4
LT		 2542 dentry_unhash(new_dentry);
2543 }
2544 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2545 error = -EBUSY;
2546 else
2547 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2548 if (target) {
2549 if (!error)
2550 target->i_flags |= S_DEAD;
1b1dcc1b 2551 mutex_unlock(&target->i_mutex);
1da177e4
LT		 2552 if (d_unhashed(new_dentry))
2553 d_rehash(new_dentry);
2554 dput(new_dentry);
2555 }
e31e14ec 2556 if (!error)
349457cc
MF		 2557 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2558 d_move(old_dentry,new_dentry);
1da177e4
LT		 2559 return error;
2560}
2561
75c96f85
AB		 2562static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2563 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 2564{
2565 struct inode *target;
2566 int error;
2567
2568 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2569 if (error)
2570 return error;
2571
2572 dget(new_dentry);
2573 target = new_dentry->d_inode;
2574 if (target)
1b1dcc1b 2575 mutex_lock(&target->i_mutex);
1da177e4
LT		 2576 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2577 error = -EBUSY;
2578 else
2579 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2580 if (!error) {
349457cc 2581 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
1da177e4 2582 d_move(old_dentry, new_dentry);
1da177e4
LT		 2583 }
2584 if (target)
1b1dcc1b 2585 mutex_unlock(&target->i_mutex);
1da177e4
LT		 2586 dput(new_dentry);
2587 return error;
2588}
2589
2590int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2591 struct inode *new_dir, struct dentry *new_dentry)
2592{
2593 int error;
2594 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
0eeca283 2595 const char *old_name;
1da177e4
LT		 2596
2597 if (old_dentry->d_inode == new_dentry->d_inode)
2598 return 0;
2599
2600 error = may_delete(old_dir, old_dentry, is_dir);
2601 if (error)
2602 return error;
2603
2604 if (!new_dentry->d_inode)
a95164d9 2605 error = may_create(new_dir, new_dentry);
1da177e4
LT		 2606 else
2607 error = may_delete(new_dir, new_dentry, is_dir);
2608 if (error)
2609 return error;
2610
2611 if (!old_dir->i_op || !old_dir->i_op->rename)
2612 return -EPERM;
2613
2614 DQUOT_INIT(old_dir);
2615 DQUOT_INIT(new_dir);
2616
0eeca283
RL		 2617 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2618
1da177e4
LT		 2619 if (is_dir)
2620 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2621 else
2622 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2623 if (!error) {
0eeca283 2624 const char *new_name = old_dentry->d_name.name;
89204c40 2625 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
5a190ae6 2626 new_dentry->d_inode, old_dentry);
1da177e4 2627 }
0eeca283
RL		 2628 fsnotify_oldname_free(old_name);
2629
1da177e4
LT		 2630 return error;
2631}
2632
2ad94ae6
AV		 2633asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2634 int newdfd, const char __user *newname)
1da177e4 2635{
2ad94ae6
AV		 2636 struct dentry *old_dir, *new_dir;
2637 struct dentry *old_dentry, *new_dentry;
2638 struct dentry *trap;
1da177e4 2639 struct nameidata oldnd, newnd;
2ad94ae6
AV		 2640 char *from;
2641 char *to;
2642 int error;
1da177e4 2643
2ad94ae6 2644 error = user_path_parent(olddfd, oldname, &oldnd, &from);
1da177e4
LT		 2645 if (error)
2646 goto exit;
2647
2ad94ae6 2648 error = user_path_parent(newdfd, newname, &newnd, &to);
1da177e4
LT		 2649 if (error)
2650 goto exit1;
2651
2652 error = -EXDEV;
4ac91378 2653 if (oldnd.path.mnt != newnd.path.mnt)
1da177e4
LT		 2654 goto exit2;
2655
4ac91378 2656 old_dir = oldnd.path.dentry;
1da177e4
LT		 2657 error = -EBUSY;
2658 if (oldnd.last_type != LAST_NORM)
2659 goto exit2;
2660
4ac91378 2661 new_dir = newnd.path.dentry;
1da177e4
LT		 2662 if (newnd.last_type != LAST_NORM)
2663 goto exit2;
2664
0612d9fb
OH		 2665 oldnd.flags &= ~LOOKUP_PARENT;
2666 newnd.flags &= ~LOOKUP_PARENT;
4e9ed2f8 2667 newnd.flags |= LOOKUP_RENAME_TARGET;
0612d9fb 2668
1da177e4
LT		 2669 trap = lock_rename(new_dir, old_dir);
2670
49705b77 2671 old_dentry = lookup_hash(&oldnd);
1da177e4
LT		 2672 error = PTR_ERR(old_dentry);
2673 if (IS_ERR(old_dentry))
2674 goto exit3;
2675 /* source must exist */
2676 error = -ENOENT;
2677 if (!old_dentry->d_inode)
2678 goto exit4;
2679 /* unless the source is a directory trailing slashes give -ENOTDIR */
2680 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2681 error = -ENOTDIR;
2682 if (oldnd.last.name[oldnd.last.len])
2683 goto exit4;
2684 if (newnd.last.name[newnd.last.len])
2685 goto exit4;
2686 }
2687 /* source should not be ancestor of target */
2688 error = -EINVAL;
2689 if (old_dentry == trap)
2690 goto exit4;
49705b77 2691 new_dentry = lookup_hash(&newnd);
1da177e4
LT		 2692 error = PTR_ERR(new_dentry);
2693 if (IS_ERR(new_dentry))
2694 goto exit4;
2695 /* target should not be an ancestor of source */
2696 error = -ENOTEMPTY;
2697 if (new_dentry == trap)
2698 goto exit5;
2699
9079b1eb
DH		 2700 error = mnt_want_write(oldnd.path.mnt);
2701 if (error)
2702 goto exit5;
be6d3e56
KT		 2703 error = security_path_rename(&oldnd.path, old_dentry,
2704 &newnd.path, new_dentry);
2705 if (error)
2706 goto exit6;
1da177e4
LT		 2707 error = vfs_rename(old_dir->d_inode, old_dentry,
2708 new_dir->d_inode, new_dentry);
be6d3e56 2709exit6:
9079b1eb 2710 mnt_drop_write(oldnd.path.mnt);
1da177e4
LT		 2711exit5:
2712 dput(new_dentry);
2713exit4:
2714 dput(old_dentry);
2715exit3:
2716 unlock_rename(new_dir, old_dir);
2717exit2:
1d957f9b 2718 path_put(&newnd.path);
2ad94ae6 2719 putname(to);
1da177e4 2720exit1:
1d957f9b 2721 path_put(&oldnd.path);
1da177e4 2722 putname(from);
2ad94ae6 2723exit:
1da177e4
LT		 2724 return error;
2725}
2726
5590ff0d
UD		 2727asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2728{
2729 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2730}
2731
1da177e4
LT		 2732int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2733{
2734 int len;
2735
2736 len = PTR_ERR(link);
2737 if (IS_ERR(link))
2738 goto out;
2739
2740 len = strlen(link);
2741 if (len > (unsigned) buflen)
2742 len = buflen;
2743 if (copy_to_user(buffer, link, len))
2744 len = -EFAULT;
2745out:
2746 return len;
2747}
2748
2749/*
2750 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2751 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2752 * using) it for any given inode is up to filesystem.
2753 */
2754int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2755{
2756 struct nameidata nd;
cc314eef 2757 void *cookie;
694a1764 2758 int res;
cc314eef 2759
1da177e4 2760 nd.depth = 0;
cc314eef 2761 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
694a1764
MS		 2762 if (IS_ERR(cookie))
2763 return PTR_ERR(cookie);
2764
2765 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2766 if (dentry->d_inode->i_op->put_link)
2767 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2768 return res;
1da177e4
LT		 2769}
2770
2771int vfs_follow_link(struct nameidata *nd, const char *link)
2772{
2773 return __vfs_follow_link(nd, link);
2774}
2775
2776/* get the link contents into pagecache */
2777static char *page_getlink(struct dentry * dentry, struct page **ppage)
2778{
ebd09abb
DG		 2779 char *kaddr;
2780 struct page *page;
1da177e4 2781 struct address_space *mapping = dentry->d_inode->i_mapping;
090d2b18 2782 page = read_mapping_page(mapping, 0, NULL);
1da177e4 2783 if (IS_ERR(page))
6fe6900e 2784 return (char*)page;
1da177e4 2785 *ppage = page;
ebd09abb
DG		 2786 kaddr = kmap(page);
2787 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2788 return kaddr;
1da177e4
LT		 2789}
2790
2791int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2792{
2793 struct page *page = NULL;
2794 char *s = page_getlink(dentry, &page);
2795 int res = vfs_readlink(dentry,buffer,buflen,s);
2796 if (page) {
2797 kunmap(page);
2798 page_cache_release(page);
2799 }
2800 return res;
2801}
2802
cc314eef 2803void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
1da177e4 2804{
cc314eef 2805 struct page *page = NULL;
1da177e4 2806 nd_set_link(nd, page_getlink(dentry, &page));
cc314eef 2807 return page;
1da177e4
LT		 2808}
2809
cc314eef 2810void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1da177e4 2811{
cc314eef
LT		 2812 struct page *page = cookie;
2813
2814 if (page) {
1da177e4
LT		 2815 kunmap(page);
2816 page_cache_release(page);
1da177e4
LT		 2817 }
2818}
2819
0adb25d2
KK		 2820int __page_symlink(struct inode *inode, const char *symname, int len,
2821 gfp_t gfp_mask)
1da177e4
LT		 2822{
2823 struct address_space *mapping = inode->i_mapping;
0adb25d2 2824 struct page *page;
afddba49 2825 void *fsdata;
beb497ab 2826 int err;
1da177e4
LT		 2827 char *kaddr;
2828
7e53cac4 2829retry:
afddba49
NP		 2830 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2831 AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
1da177e4 2832 if (err)
afddba49
NP		 2833 goto fail;
2834
1da177e4
LT		 2835 kaddr = kmap_atomic(page, KM_USER0);
2836 memcpy(kaddr, symname, len-1);
2837 kunmap_atomic(kaddr, KM_USER0);
afddba49
NP		 2838
2839 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2840 page, fsdata);
1da177e4
LT		 2841 if (err < 0)
2842 goto fail;
afddba49
NP		 2843 if (err < len-1)
2844 goto retry;
2845
1da177e4
LT		 2846 mark_inode_dirty(inode);
2847 return 0;
1da177e4
LT		 2848fail:
2849 return err;
2850}
2851
0adb25d2
KK		 2852int page_symlink(struct inode *inode, const char *symname, int len)
2853{
2854 return __page_symlink(inode, symname, len,
2855 mapping_gfp_mask(inode->i_mapping));
2856}
2857
92e1d5be 2858const struct inode_operations page_symlink_inode_operations = {
1da177e4
LT		 2859 .readlink = generic_readlink,
2860 .follow_link = page_follow_link_light,
2861 .put_link = page_put_link,
2862};
2863
2d8f3038 2864EXPORT_SYMBOL(user_path_at);
1da177e4
LT		 2865EXPORT_SYMBOL(follow_down);
2866EXPORT_SYMBOL(follow_up);
2867EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2868EXPORT_SYMBOL(getname);
2869EXPORT_SYMBOL(lock_rename);
1da177e4
LT		 2870EXPORT_SYMBOL(lookup_one_len);
2871EXPORT_SYMBOL(page_follow_link_light);
2872EXPORT_SYMBOL(page_put_link);
2873EXPORT_SYMBOL(page_readlink);
0adb25d2 2874EXPORT_SYMBOL(__page_symlink);
1da177e4
LT		 2875EXPORT_SYMBOL(page_symlink);
2876EXPORT_SYMBOL(page_symlink_inode_operations);
2877EXPORT_SYMBOL(path_lookup);
d1811465 2878EXPORT_SYMBOL(kern_path);
16f18200 2879EXPORT_SYMBOL(vfs_path_lookup);
f419a2e3 2880EXPORT_SYMBOL(inode_permission);
8c744fb8 2881EXPORT_SYMBOL(file_permission);
1da177e4
LT		 2882EXPORT_SYMBOL(unlock_rename);
2883EXPORT_SYMBOL(vfs_create);
2884EXPORT_SYMBOL(vfs_follow_link);
2885EXPORT_SYMBOL(vfs_link);
2886EXPORT_SYMBOL(vfs_mkdir);
2887EXPORT_SYMBOL(vfs_mknod);
2888EXPORT_SYMBOL(generic_permission);
2889EXPORT_SYMBOL(vfs_readlink);
2890EXPORT_SYMBOL(vfs_rename);
2891EXPORT_SYMBOL(vfs_rmdir);
2892EXPORT_SYMBOL(vfs_symlink);
2893EXPORT_SYMBOL(vfs_unlink);
2894EXPORT_SYMBOL(dentry_unhash);
2895EXPORT_SYMBOL(generic_readlink);
18d8fda7
AV		 2896
2897/* to be mentioned only in INIT_TASK */
2898struct fs_struct init_fs = {
2899 .count = ATOMIC_INIT(1),
1239f26c 2900 .lock = __RW_LOCK_UNLOCKED(init_fs.lock),
18d8fda7
AV		 2901 .umask = 0022,
2902};
kernel source for telekom puls (alcatel/mediatek)