projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
vfs: Implement proper O_SYNC semantics
[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>
6146f0d5 27#include <linux/ima.h>
1da177e4
LT		 28#include <linux/syscalls.h>
29#include <linux/mount.h>
30#include <linux/audit.h>
16f7e0fe 31#include <linux/capability.h>
834f2a4a 32#include <linux/file.h>
5590ff0d 33#include <linux/fcntl.h>
08ce5f16 34#include <linux/device_cgroup.h>
5ad4e53b 35#include <linux/fs_struct.h>
1da177e4
LT		 36#include <asm/uaccess.h>
37
38#define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
39
40/* [Feb-1997 T. Schoebel-Theuer]
41 * Fundamental changes in the pathname lookup mechanisms (namei)
42 * were necessary because of omirr. The reason is that omirr needs
43 * to know the _real_ pathname, not the user-supplied one, in case
44 * of symlinks (and also when transname replacements occur).
45 *
46 * The new code replaces the old recursive symlink resolution with
47 * an iterative one (in case of non-nested symlink chains). It does
48 * this with calls to <fs>_follow_link().
49 * As a side effect, dir_namei(), _namei() and follow_link() are now
50 * replaced with a single function lookup_dentry() that can handle all
51 * the special cases of the former code.
52 *
53 * With the new dcache, the pathname is stored at each inode, at least as
54 * long as the refcount of the inode is positive. As a side effect, the
55 * size of the dcache depends on the inode cache and thus is dynamic.
56 *
57 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
58 * resolution to correspond with current state of the code.
59 *
60 * Note that the symlink resolution is not *completely* iterative.
61 * There is still a significant amount of tail- and mid- recursion in
62 * the algorithm. Also, note that <fs>_readlink() is not used in
63 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
64 * may return different results than <fs>_follow_link(). Many virtual
65 * filesystems (including /proc) exhibit this behavior.
66 */
67
68/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
69 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
70 * and the name already exists in form of a symlink, try to create the new
71 * name indicated by the symlink. The old code always complained that the
72 * name already exists, due to not following the symlink even if its target
73 * is nonexistent. The new semantics affects also mknod() and link() when
74 * the name is a symlink pointing to a non-existant name.
75 *
76 * I don't know which semantics is the right one, since I have no access
77 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
78 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
79 * "old" one. Personally, I think the new semantics is much more logical.
80 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
81 * file does succeed in both HP-UX and SunOs, but not in Solaris
82 * and in the old Linux semantics.
83 */
84
85/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
86 * semantics. See the comments in "open_namei" and "do_link" below.
87 *
88 * [10-Sep-98 Alan Modra] Another symlink change.
89 */
90
91/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
92 * inside the path - always follow.
93 * in the last component in creation/removal/renaming - never follow.
94 * if LOOKUP_FOLLOW passed - follow.
95 * if the pathname has trailing slashes - follow.
96 * otherwise - don't follow.
97 * (applied in that order).
98 *
99 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
100 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
101 * During the 2.4 we need to fix the userland stuff depending on it -
102 * hopefully we will be able to get rid of that wart in 2.5. So far only
103 * XEmacs seems to be relying on it...
104 */
105/*
106 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
a11f3a05 107 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
1da177e4
LT		 108 * any extra contention...
109 */
110
a02f76c3 111static int __link_path_walk(const char *name, struct nameidata *nd);
c4a7808f 112
1da177e4
LT		 113/* In order to reduce some races, while at the same time doing additional
114 * checking and hopefully speeding things up, we copy filenames to the
115 * kernel data space before using them..
116 *
117 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
118 * PATH_MAX includes the nul terminator --RR.
119 */
858119e1 120static int do_getname(const char __user *filename, char *page)
1da177e4
LT		 121{
122 int retval;
123 unsigned long len = PATH_MAX;
124
125 if (!segment_eq(get_fs(), KERNEL_DS)) {
126 if ((unsigned long) filename >= TASK_SIZE)
127 return -EFAULT;
128 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
129 len = TASK_SIZE - (unsigned long) filename;
130 }
131
132 retval = strncpy_from_user(page, filename, len);
133 if (retval > 0) {
134 if (retval < len)
135 return 0;
136 return -ENAMETOOLONG;
137 } else if (!retval)
138 retval = -ENOENT;
139 return retval;
140}
141
142char * getname(const char __user * filename)
143{
144 char *tmp, *result;
145
146 result = ERR_PTR(-ENOMEM);
147 tmp = __getname();
148 if (tmp) {
149 int retval = do_getname(filename, tmp);
150
151 result = tmp;
152 if (retval < 0) {
153 __putname(tmp);
154 result = ERR_PTR(retval);
155 }
156 }
157 audit_getname(result);
158 return result;
159}
160
161#ifdef CONFIG_AUDITSYSCALL
162void putname(const char *name)
163{
5ac3a9c2 164 if (unlikely(!audit_dummy_context()))
1da177e4
LT		 165 audit_putname(name);
166 else
167 __putname(name);
168}
169EXPORT_SYMBOL(putname);
170#endif
171
5909ccaa
LT		 172/*
173 * This does basic POSIX ACL permission checking
1da177e4 174 */
5909ccaa 175static int acl_permission_check(struct inode *inode, int mask,
1da177e4
LT		 176 int (*check_acl)(struct inode *inode, int mask))
177{
178 umode_t mode = inode->i_mode;
179
e6305c43
AV		 180 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
181
da9592ed 182 if (current_fsuid() == inode->i_uid)
1da177e4
LT		 183 mode >>= 6;
184 else {
185 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
186 int error = check_acl(inode, mask);
5909ccaa 187 if (error != -EAGAIN)
1da177e4
LT		 188 return error;
189 }
190
191 if (in_group_p(inode->i_gid))
192 mode >>= 3;
193 }
194
195 /*
196 * If the DACs are ok we don't need any capability check.
197 */
e6305c43 198 if ((mask & ~mode) == 0)
1da177e4 199 return 0;
5909ccaa
LT		 200 return -EACCES;
201}
202
203/**
204 * generic_permission - check for access rights on a Posix-like filesystem
205 * @inode: inode to check access rights for
206 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
207 * @check_acl: optional callback to check for Posix ACLs
208 *
209 * Used to check for read/write/execute permissions on a file.
210 * We use "fsuid" for this, letting us set arbitrary permissions
211 * for filesystem access without changing the "normal" uids which
212 * are used for other things..
213 */
214int generic_permission(struct inode *inode, int mask,
215 int (*check_acl)(struct inode *inode, int mask))
216{
217 int ret;
218
219 /*
220 * Do the basic POSIX ACL permission checks.
221 */
222 ret = acl_permission_check(inode, mask, check_acl);
223 if (ret != -EACCES)
224 return ret;
1da177e4 225
1da177e4
LT		 226 /*
227 * Read/write DACs are always overridable.
228 * Executable DACs are overridable if at least one exec bit is set.
229 */
f696a365 230 if (!(mask & MAY_EXEC) || execute_ok(inode))
1da177e4
LT		 231 if (capable(CAP_DAC_OVERRIDE))
232 return 0;
233
234 /*
235 * Searching includes executable on directories, else just read.
236 */
237 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
238 if (capable(CAP_DAC_READ_SEARCH))
239 return 0;
240
241 return -EACCES;
242}
243
cb23beb5
CH		 244/**
245 * inode_permission - check for access rights to a given inode
246 * @inode: inode to check permission on
247 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
248 *
249 * Used to check for read/write/execute permissions on an inode.
250 * We use "fsuid" for this, letting us set arbitrary permissions
251 * for filesystem access without changing the "normal" uids which
252 * are used for other things.
253 */
f419a2e3 254int inode_permission(struct inode *inode, int mask)
1da177e4 255{
e6305c43 256 int retval;
1da177e4
LT		 257
258 if (mask & MAY_WRITE) {
22590e41 259 umode_t mode = inode->i_mode;
1da177e4
LT		 260
261 /*
262 * Nobody gets write access to a read-only fs.
263 */
264 if (IS_RDONLY(inode) &&
265 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
266 return -EROFS;
267
268 /*
269 * Nobody gets write access to an immutable file.
270 */
271 if (IS_IMMUTABLE(inode))
272 return -EACCES;
273 }
274
acfa4380 275 if (inode->i_op->permission)
b77b0646 276 retval = inode->i_op->permission(inode, mask);
f696a365 277 else
5909ccaa 278 retval = generic_permission(inode, mask, inode->i_op->check_acl);
f696a365 279
1da177e4
LT		 280 if (retval)
281 return retval;
282
08ce5f16
SH		 283 retval = devcgroup_inode_permission(inode, mask);
284 if (retval)
285 return retval;
286
e6305c43 287 return security_inode_permission(inode,
f418b006 288 mask & (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND));
1da177e4
LT		 289}
290
8c744fb8
CH		 291/**
292 * file_permission - check for additional access rights to a given file
293 * @file: file to check access rights for
294 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
295 *
296 * Used to check for read/write/execute permissions on an already opened
297 * file.
298 *
299 * Note:
300 * Do not use this function in new code. All access checks should
cb23beb5 301 * be done using inode_permission().
8c744fb8
CH		 302 */
303int file_permission(struct file *file, int mask)
304{
f419a2e3 305 return inode_permission(file->f_path.dentry->d_inode, mask);
8c744fb8
CH		 306}
307
1da177e4
LT		 308/*
309 * get_write_access() gets write permission for a file.
310 * put_write_access() releases this write permission.
311 * This is used for regular files.
312 * We cannot support write (and maybe mmap read-write shared) accesses and
313 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
314 * can have the following values:
315 * 0: no writers, no VM_DENYWRITE mappings
316 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
317 * > 0: (i_writecount) users are writing to the file.
318 *
319 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
320 * except for the cases where we don't hold i_writecount yet. Then we need to
321 * use {get,deny}_write_access() - these functions check the sign and refuse
322 * to do the change if sign is wrong. Exclusion between them is provided by
323 * the inode->i_lock spinlock.
324 */
325
326int get_write_access(struct inode * inode)
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_inc(&inode->i_writecount);
334 spin_unlock(&inode->i_lock);
335
336 return 0;
337}
338
339int deny_write_access(struct file * file)
340{
0f7fc9e4 341 struct inode *inode = file->f_path.dentry->d_inode;
1da177e4
LT		 342
343 spin_lock(&inode->i_lock);
344 if (atomic_read(&inode->i_writecount) > 0) {
345 spin_unlock(&inode->i_lock);
346 return -ETXTBSY;
347 }
348 atomic_dec(&inode->i_writecount);
349 spin_unlock(&inode->i_lock);
350
351 return 0;
352}
353
5dd784d0
JB		 354/**
355 * path_get - get a reference to a path
356 * @path: path to get the reference to
357 *
358 * Given a path increment the reference count to the dentry and the vfsmount.
359 */
360void path_get(struct path *path)
361{
362 mntget(path->mnt);
363 dget(path->dentry);
364}
365EXPORT_SYMBOL(path_get);
366
1d957f9b
JB		 367/**
368 * path_put - put a reference to a path
369 * @path: path to put the reference to
370 *
371 * Given a path decrement the reference count to the dentry and the vfsmount.
372 */
373void path_put(struct path *path)
1da177e4 374{
1d957f9b
JB		 375 dput(path->dentry);
376 mntput(path->mnt);
1da177e4 377}
1d957f9b 378EXPORT_SYMBOL(path_put);
1da177e4 379
834f2a4a
TM		 380/**
381 * release_open_intent - free up open intent resources
382 * @nd: pointer to nameidata
383 */
384void release_open_intent(struct nameidata *nd)
385{
0f7fc9e4 386 if (nd->intent.open.file->f_path.dentry == NULL)
834f2a4a
TM		 387 put_filp(nd->intent.open.file);
388 else
389 fput(nd->intent.open.file);
390}
391
bcdc5e01
IK		 392static inline struct dentry *
393do_revalidate(struct dentry *dentry, struct nameidata *nd)
394{
395 int status = dentry->d_op->d_revalidate(dentry, nd);
396 if (unlikely(status <= 0)) {
397 /*
398 * The dentry failed validation.
399 * If d_revalidate returned 0 attempt to invalidate
400 * the dentry otherwise d_revalidate is asking us
401 * to return a fail status.
402 */
403 if (!status) {
404 if (!d_invalidate(dentry)) {
405 dput(dentry);
406 dentry = NULL;
407 }
408 } else {
409 dput(dentry);
410 dentry = ERR_PTR(status);
411 }
412 }
413 return dentry;
414}
415
1da177e4
LT		 416/*
417 * Internal lookup() using the new generic dcache.
418 * SMP-safe
419 */
420static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
421{
422 struct dentry * dentry = __d_lookup(parent, name);
423
424 /* lockess __d_lookup may fail due to concurrent d_move()
425 * in some unrelated directory, so try with d_lookup
426 */
427 if (!dentry)
428 dentry = d_lookup(parent, name);
429
bcdc5e01
IK		 430 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
431 dentry = do_revalidate(dentry, nd);
432
1da177e4
LT		 433 return dentry;
434}
435
436/*
437 * Short-cut version of permission(), for calling by
438 * path_walk(), when dcache lock is held. Combines parts
439 * of permission() and generic_permission(), and tests ONLY for
440 * MAY_EXEC permission.
441 *
442 * If appropriate, check DAC only. If not appropriate, or
443 * short-cut DAC fails, then call permission() to do more
444 * complete permission check.
445 */
672b16b2 446static int exec_permission_lite(struct inode *inode)
1da177e4 447{
5909ccaa 448 int ret;
1da177e4 449
cb9179ea 450 if (inode->i_op->permission) {
5909ccaa 451 ret = inode->i_op->permission(inode, MAY_EXEC);
cb9179ea
LT		 452 if (!ret)
453 goto ok;
454 return ret;
455 }
5909ccaa
LT		 456 ret = acl_permission_check(inode, MAY_EXEC, inode->i_op->check_acl);
457 if (!ret)
1da177e4
LT		 458 goto ok;
459
f1ac9f6b 460 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
1da177e4
LT		 461 goto ok;
462
5909ccaa 463 return ret;
1da177e4 464ok:
b77b0646 465 return security_inode_permission(inode, MAY_EXEC);
1da177e4
LT		 466}
467
468/*
469 * This is called when everything else fails, and we actually have
470 * to go to the low-level filesystem to find out what we should do..
471 *
472 * We get the directory semaphore, and after getting that we also
473 * make sure that nobody added the entry to the dcache in the meantime..
474 * SMP-safe
475 */
476static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
477{
478 struct dentry * result;
479 struct inode *dir = parent->d_inode;
480
1b1dcc1b 481 mutex_lock(&dir->i_mutex);
1da177e4
LT		 482 /*
483 * First re-do the cached lookup just in case it was created
484 * while we waited for the directory semaphore..
485 *
486 * FIXME! This could use version numbering or similar to
487 * avoid unnecessary cache lookups.
488 *
489 * The "dcache_lock" is purely to protect the RCU list walker
490 * from concurrent renames at this point (we mustn't get false
491 * negatives from the RCU list walk here, unlike the optimistic
492 * fast walk).
493 *
494 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
495 */
496 result = d_lookup(parent, name);
497 if (!result) {
d70b67c8
MS		 498 struct dentry *dentry;
499
500 /* Don't create child dentry for a dead directory. */
501 result = ERR_PTR(-ENOENT);
502 if (IS_DEADDIR(dir))
503 goto out_unlock;
504
505 dentry = d_alloc(parent, name);
1da177e4
LT		 506 result = ERR_PTR(-ENOMEM);
507 if (dentry) {
508 result = dir->i_op->lookup(dir, dentry, nd);
509 if (result)
510 dput(dentry);
511 else
512 result = dentry;
513 }
d70b67c8 514out_unlock:
1b1dcc1b 515 mutex_unlock(&dir->i_mutex);
1da177e4
LT		 516 return result;
517 }
518
519 /*
520 * Uhhuh! Nasty case: the cache was re-populated while
521 * we waited on the semaphore. Need to revalidate.
522 */
1b1dcc1b 523 mutex_unlock(&dir->i_mutex);
1da177e4 524 if (result->d_op && result->d_op->d_revalidate) {
bcdc5e01
IK		 525 result = do_revalidate(result, nd);
526 if (!result)
1da177e4 527 result = ERR_PTR(-ENOENT);
1da177e4
LT		 528 }
529 return result;
530}
531
a02f76c3
AV		 532/*
533 * Wrapper to retry pathname resolution whenever the underlying
534 * file system returns an ESTALE.
535 *
536 * Retry the whole path once, forcing real lookup requests
537 * instead of relying on the dcache.
538 */
539static __always_inline int link_path_walk(const char *name, struct nameidata *nd)
540{
541 struct path save = nd->path;
542 int result;
543
544 /* make sure the stuff we saved doesn't go away */
c8e7f449 545 path_get(&save);
a02f76c3
AV		 546
547 result = __link_path_walk(name, nd);
548 if (result == -ESTALE) {
549 /* nd->path had been dropped */
550 nd->path = save;
c8e7f449 551 path_get(&nd->path);
a02f76c3
AV		 552 nd->flags |= LOOKUP_REVAL;
553 result = __link_path_walk(name, nd);
554 }
555
556 path_put(&save);
557
558 return result;
559}
560
2a737871
AV		 561static __always_inline void set_root(struct nameidata *nd)
562{
563 if (!nd->root.mnt) {
564 struct fs_struct *fs = current->fs;
565 read_lock(&fs->lock);
566 nd->root = fs->root;
567 path_get(&nd->root);
568 read_unlock(&fs->lock);
569 }
570}
571
f1662356 572static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
1da177e4
LT		 573{
574 int res = 0;
575 char *name;
576 if (IS_ERR(link))
577 goto fail;
578
579 if (*link == '/') {
2a737871 580 set_root(nd);
1d957f9b 581 path_put(&nd->path);
2a737871
AV		 582 nd->path = nd->root;
583 path_get(&nd->root);
1da177e4 584 }
b4091d5f 585
1da177e4 586 res = link_path_walk(link, nd);
1da177e4
LT		 587 if (nd->depth || res || nd->last_type!=LAST_NORM)
588 return res;
589 /*
590 * If it is an iterative symlinks resolution in open_namei() we
591 * have to copy the last component. And all that crap because of
592 * bloody create() on broken symlinks. Furrfu...
593 */
594 name = __getname();
595 if (unlikely(!name)) {
1d957f9b 596 path_put(&nd->path);
1da177e4
LT		 597 return -ENOMEM;
598 }
599 strcpy(name, nd->last.name);
600 nd->last.name = name;
601 return 0;
602fail:
1d957f9b 603 path_put(&nd->path);
1da177e4
LT		 604 return PTR_ERR(link);
605}
606
1d957f9b 607static void path_put_conditional(struct path *path, struct nameidata *nd)
051d3812
IK		 608{
609 dput(path->dentry);
4ac91378 610 if (path->mnt != nd->path.mnt)
051d3812
IK		 611 mntput(path->mnt);
612}
613
614static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
615{
4ac91378
JB		 616 dput(nd->path.dentry);
617 if (nd->path.mnt != path->mnt)
618 mntput(nd->path.mnt);
619 nd->path.mnt = path->mnt;
620 nd->path.dentry = path->dentry;
051d3812
IK		 621}
622
f1662356 623static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
1da177e4
LT		 624{
625 int error;
cc314eef 626 void *cookie;
cd4e91d3 627 struct dentry *dentry = path->dentry;
1da177e4 628
d671a1cb 629 touch_atime(path->mnt, dentry);
1da177e4 630 nd_set_link(nd, NULL);
cd4e91d3 631
4ac91378 632 if (path->mnt != nd->path.mnt) {
051d3812
IK		 633 path_to_nameidata(path, nd);
634 dget(dentry);
635 }
636 mntget(path->mnt);
cc314eef
LT		 637 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
638 error = PTR_ERR(cookie);
639 if (!IS_ERR(cookie)) {
1da177e4 640 char *s = nd_get_link(nd);
cc314eef 641 error = 0;
1da177e4
LT		 642 if (s)
643 error = __vfs_follow_link(nd, s);
644 if (dentry->d_inode->i_op->put_link)
cc314eef 645 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
1da177e4 646 }
09da5916 647 path_put(path);
1da177e4
LT		 648
649 return error;
650}
651
652/*
653 * This limits recursive symlink follows to 8, while
654 * limiting consecutive symlinks to 40.
655 *
656 * Without that kind of total limit, nasty chains of consecutive
657 * symlinks can cause almost arbitrarily long lookups.
658 */
90ebe565 659static inline int do_follow_link(struct path *path, struct nameidata *nd)
1da177e4
LT		 660{
661 int err = -ELOOP;
662 if (current->link_count >= MAX_NESTED_LINKS)
663 goto loop;
664 if (current->total_link_count >= 40)
665 goto loop;
666 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
667 cond_resched();
90ebe565 668 err = security_inode_follow_link(path->dentry, nd);
1da177e4
LT		 669 if (err)
670 goto loop;
671 current->link_count++;
672 current->total_link_count++;
673 nd->depth++;
cd4e91d3 674 err = __do_follow_link(path, nd);
839d9f93
AV		 675 current->link_count--;
676 nd->depth--;
1da177e4
LT		 677 return err;
678loop:
1d957f9b
JB		 679 path_put_conditional(path, nd);
680 path_put(&nd->path);
1da177e4
LT		 681 return err;
682}
683
bab77ebf 684int follow_up(struct path *path)
1da177e4
LT		 685{
686 struct vfsmount *parent;
687 struct dentry *mountpoint;
688 spin_lock(&vfsmount_lock);
bab77ebf
AV		 689 parent = path->mnt->mnt_parent;
690 if (parent == path->mnt) {
1da177e4
LT		 691 spin_unlock(&vfsmount_lock);
692 return 0;
693 }
694 mntget(parent);
bab77ebf 695 mountpoint = dget(path->mnt->mnt_mountpoint);
1da177e4 696 spin_unlock(&vfsmount_lock);
bab77ebf
AV		 697 dput(path->dentry);
698 path->dentry = mountpoint;
699 mntput(path->mnt);
700 path->mnt = parent;
1da177e4
LT		 701 return 1;
702}
703
704/* no need for dcache_lock, as serialization is taken care in
705 * namespace.c
706 */
463ffb2e
AV		 707static int __follow_mount(struct path *path)
708{
709 int res = 0;
710 while (d_mountpoint(path->dentry)) {
1c755af4 711 struct vfsmount *mounted = lookup_mnt(path);
463ffb2e
AV		 712 if (!mounted)
713 break;
714 dput(path->dentry);
715 if (res)
716 mntput(path->mnt);
717 path->mnt = mounted;
718 path->dentry = dget(mounted->mnt_root);
719 res = 1;
720 }
721 return res;
722}
723
79ed0226 724static void follow_mount(struct path *path)
1da177e4 725{
79ed0226 726 while (d_mountpoint(path->dentry)) {
1c755af4 727 struct vfsmount *mounted = lookup_mnt(path);
1da177e4
LT		 728 if (!mounted)
729 break;
79ed0226
AV		 730 dput(path->dentry);
731 mntput(path->mnt);
732 path->mnt = mounted;
733 path->dentry = dget(mounted->mnt_root);
1da177e4 734 }
1da177e4
LT		 735}
736
737/* no need for dcache_lock, as serialization is taken care in
738 * namespace.c
739 */
9393bd07 740int follow_down(struct path *path)
1da177e4
LT		 741{
742 struct vfsmount *mounted;
743
1c755af4 744 mounted = lookup_mnt(path);
1da177e4 745 if (mounted) {
9393bd07
AV		 746 dput(path->dentry);
747 mntput(path->mnt);
748 path->mnt = mounted;
749 path->dentry = dget(mounted->mnt_root);
1da177e4
LT		 750 return 1;
751 }
752 return 0;
753}
754
f1662356 755static __always_inline void follow_dotdot(struct nameidata *nd)
1da177e4 756{
2a737871 757 set_root(nd);
e518ddb7 758
1da177e4
LT		 759 while(1) {
760 struct vfsmount *parent;
4ac91378 761 struct dentry *old = nd->path.dentry;
1da177e4 762
2a737871
AV		 763 if (nd->path.dentry == nd->root.dentry &&
764 nd->path.mnt == nd->root.mnt) {
1da177e4
LT		 765 break;
766 }
1da177e4 767 spin_lock(&dcache_lock);
4ac91378
JB		 768 if (nd->path.dentry != nd->path.mnt->mnt_root) {
769 nd->path.dentry = dget(nd->path.dentry->d_parent);
1da177e4
LT		 770 spin_unlock(&dcache_lock);
771 dput(old);
772 break;
773 }
774 spin_unlock(&dcache_lock);
775 spin_lock(&vfsmount_lock);
4ac91378
JB		 776 parent = nd->path.mnt->mnt_parent;
777 if (parent == nd->path.mnt) {
1da177e4
LT		 778 spin_unlock(&vfsmount_lock);
779 break;
780 }
781 mntget(parent);
4ac91378 782 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
1da177e4
LT		 783 spin_unlock(&vfsmount_lock);
784 dput(old);
4ac91378
JB		 785 mntput(nd->path.mnt);
786 nd->path.mnt = parent;
1da177e4 787 }
79ed0226 788 follow_mount(&nd->path);
1da177e4
LT		 789}
790
1da177e4
LT		 791/*
792 * It's more convoluted than I'd like it to be, but... it's still fairly
793 * small and for now I'd prefer to have fast path as straight as possible.
794 * It _is_ time-critical.
795 */
796static int do_lookup(struct nameidata *nd, struct qstr *name,
797 struct path *path)
798{
4ac91378
JB		 799 struct vfsmount *mnt = nd->path.mnt;
800 struct dentry *dentry = __d_lookup(nd->path.dentry, name);
1da177e4
LT		 801
802 if (!dentry)
803 goto need_lookup;
804 if (dentry->d_op && dentry->d_op->d_revalidate)
805 goto need_revalidate;
806done:
807 path->mnt = mnt;
808 path->dentry = dentry;
634ee701 809 __follow_mount(path);
1da177e4
LT		 810 return 0;
811
812need_lookup:
4ac91378 813 dentry = real_lookup(nd->path.dentry, name, nd);
1da177e4
LT		 814 if (IS_ERR(dentry))
815 goto fail;
816 goto done;
817
818need_revalidate:
bcdc5e01
IK		 819 dentry = do_revalidate(dentry, nd);
820 if (!dentry)
821 goto need_lookup;
822 if (IS_ERR(dentry))
823 goto fail;
824 goto done;
1da177e4
LT		 825
826fail:
827 return PTR_ERR(dentry);
828}
829
830/*
831 * Name resolution.
ea3834d9
PM		 832 * This is the basic name resolution function, turning a pathname into
833 * the final dentry. We expect 'base' to be positive and a directory.
1da177e4 834 *
ea3834d9
PM		 835 * Returns 0 and nd will have valid dentry and mnt on success.
836 * Returns error and drops reference to input namei data on failure.
1da177e4 837 */
fc9b52cd 838static int __link_path_walk(const char *name, struct nameidata *nd)
1da177e4
LT		 839{
840 struct path next;
841 struct inode *inode;
842 int err;
843 unsigned int lookup_flags = nd->flags;
844
845 while (*name=='/')
846 name++;
847 if (!*name)
848 goto return_reval;
849
4ac91378 850 inode = nd->path.dentry->d_inode;
1da177e4 851 if (nd->depth)
f55eab82 852 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
1da177e4
LT		 853
854 /* At this point we know we have a real path component. */
855 for(;;) {
856 unsigned long hash;
857 struct qstr this;
858 unsigned int c;
859
cdce5d6b 860 nd->flags |= LOOKUP_CONTINUE;
672b16b2 861 err = exec_permission_lite(inode);
1da177e4
LT		 862 if (err)
863 break;
864
865 this.name = name;
866 c = *(const unsigned char *)name;
867
868 hash = init_name_hash();
869 do {
870 name++;
871 hash = partial_name_hash(c, hash);
872 c = *(const unsigned char *)name;
873 } while (c && (c != '/'));
874 this.len = name - (const char *) this.name;
875 this.hash = end_name_hash(hash);
876
877 /* remove trailing slashes? */
878 if (!c)
879 goto last_component;
880 while (*++name == '/');
881 if (!*name)
882 goto last_with_slashes;
883
884 /*
885 * "." and ".." are special - ".." especially so because it has
886 * to be able to know about the current root directory and
887 * parent relationships.
888 */
889 if (this.name[0] == '.') switch (this.len) {
890 default:
891 break;
892 case 2:
893 if (this.name[1] != '.')
894 break;
58c465eb 895 follow_dotdot(nd);
4ac91378 896 inode = nd->path.dentry->d_inode;
1da177e4
LT		 897 /* fallthrough */
898 case 1:
899 continue;
900 }
901 /*
902 * See if the low-level filesystem might want
903 * to use its own hash..
904 */
4ac91378
JB		 905 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
906 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
907 &this);
1da177e4
LT		 908 if (err < 0)
909 break;
910 }
1da177e4
LT		 911 /* This does the actual lookups.. */
912 err = do_lookup(nd, &this, &next);
913 if (err)
914 break;
1da177e4
LT		 915
916 err = -ENOENT;
917 inode = next.dentry->d_inode;
918 if (!inode)
919 goto out_dput;
1da177e4
LT		 920
921 if (inode->i_op->follow_link) {
90ebe565 922 err = do_follow_link(&next, nd);
1da177e4
LT		 923 if (err)
924 goto return_err;
925 err = -ENOENT;
4ac91378 926 inode = nd->path.dentry->d_inode;
1da177e4
LT		 927 if (!inode)
928 break;
09dd17d3
MS		 929 } else
930 path_to_nameidata(&next, nd);
1da177e4
LT		 931 err = -ENOTDIR;
932 if (!inode->i_op->lookup)
933 break;
934 continue;
935 /* here ends the main loop */
936
937last_with_slashes:
938 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
939last_component:
f55eab82
TM		 940 /* Clear LOOKUP_CONTINUE iff it was previously unset */
941 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
1da177e4
LT		 942 if (lookup_flags & LOOKUP_PARENT)
943 goto lookup_parent;
944 if (this.name[0] == '.') switch (this.len) {
945 default:
946 break;
947 case 2:
948 if (this.name[1] != '.')
949 break;
58c465eb 950 follow_dotdot(nd);
4ac91378 951 inode = nd->path.dentry->d_inode;
1da177e4
LT		 952 /* fallthrough */
953 case 1:
954 goto return_reval;
955 }
4ac91378
JB		 956 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
957 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
958 &this);
1da177e4
LT		 959 if (err < 0)
960 break;
961 }
962 err = do_lookup(nd, &this, &next);
963 if (err)
964 break;
1da177e4
LT		 965 inode = next.dentry->d_inode;
966 if ((lookup_flags & LOOKUP_FOLLOW)
acfa4380 967 && inode && inode->i_op->follow_link) {
90ebe565 968 err = do_follow_link(&next, nd);
1da177e4
LT		 969 if (err)
970 goto return_err;
4ac91378 971 inode = nd->path.dentry->d_inode;
09dd17d3
MS		 972 } else
973 path_to_nameidata(&next, nd);
1da177e4
LT		 974 err = -ENOENT;
975 if (!inode)
976 break;
977 if (lookup_flags & LOOKUP_DIRECTORY) {
978 err = -ENOTDIR;
acfa4380 979 if (!inode->i_op->lookup)
1da177e4
LT		 980 break;
981 }
982 goto return_base;
983lookup_parent:
984 nd->last = this;
985 nd->last_type = LAST_NORM;
986 if (this.name[0] != '.')
987 goto return_base;
988 if (this.len == 1)
989 nd->last_type = LAST_DOT;
990 else if (this.len == 2 && this.name[1] == '.')
991 nd->last_type = LAST_DOTDOT;
992 else
993 goto return_base;
994return_reval:
995 /*
996 * We bypassed the ordinary revalidation routines.
997 * We may need to check the cached dentry for staleness.
998 */
4ac91378
JB		 999 if (nd->path.dentry && nd->path.dentry->d_sb &&
1000 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
1da177e4
LT		 1001 err = -ESTALE;
1002 /* Note: we do not d_invalidate() */
4ac91378
JB		 1003 if (!nd->path.dentry->d_op->d_revalidate(
1004 nd->path.dentry, nd))
1da177e4
LT		 1005 break;
1006 }
1007return_base:
1008 return 0;
1009out_dput:
1d957f9b 1010 path_put_conditional(&next, nd);
1da177e4
LT		 1011 break;
1012 }
1d957f9b 1013 path_put(&nd->path);
1da177e4
LT		 1014return_err:
1015 return err;
1016}
1017
fc9b52cd 1018static int path_walk(const char *name, struct nameidata *nd)
1da177e4
LT		 1019{
1020 current->total_link_count = 0;
1021 return link_path_walk(name, nd);
1022}
1023
9b4a9b14 1024static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1da177e4 1025{
ea3834d9 1026 int retval = 0;
170aa3d0
UD		 1027 int fput_needed;
1028 struct file *file;
1da177e4
LT		 1029
1030 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1031 nd->flags = flags;
1032 nd->depth = 0;
2a737871 1033 nd->root.mnt = NULL;
1da177e4 1034
1da177e4 1035 if (*name=='/') {
2a737871
AV		 1036 set_root(nd);
1037 nd->path = nd->root;
1038 path_get(&nd->root);
5590ff0d 1039 } else if (dfd == AT_FDCWD) {
2a737871 1040 struct fs_struct *fs = current->fs;
e518ddb7 1041 read_lock(&fs->lock);
6ac08c39
JB		 1042 nd->path = fs->pwd;
1043 path_get(&fs->pwd);
e518ddb7 1044 read_unlock(&fs->lock);
5590ff0d 1045 } else {
5590ff0d
UD		 1046 struct dentry *dentry;
1047
1048 file = fget_light(dfd, &fput_needed);
170aa3d0
UD		 1049 retval = -EBADF;
1050 if (!file)
6d09bb62 1051 goto out_fail;
5590ff0d 1052
0f7fc9e4 1053 dentry = file->f_path.dentry;
5590ff0d 1054
170aa3d0
UD		 1055 retval = -ENOTDIR;
1056 if (!S_ISDIR(dentry->d_inode->i_mode))
6d09bb62 1057 goto fput_fail;
5590ff0d
UD		 1058
1059 retval = file_permission(file, MAY_EXEC);
170aa3d0 1060 if (retval)
6d09bb62 1061 goto fput_fail;
5590ff0d 1062
5dd784d0
JB		 1063 nd->path = file->f_path;
1064 path_get(&file->f_path);
5590ff0d
UD		 1065
1066 fput_light(file, fput_needed);
1da177e4 1067 }
9b4a9b14 1068 return 0;
2dfdd266 1069
9b4a9b14
AV		 1070fput_fail:
1071 fput_light(file, fput_needed);
1072out_fail:
1073 return retval;
1074}
1075
1076/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1077static int do_path_lookup(int dfd, const char *name,
1078 unsigned int flags, struct nameidata *nd)
1079{
1080 int retval = path_init(dfd, name, flags, nd);
1081 if (!retval)
1082 retval = path_walk(name, nd);
4ac91378
JB		 1083 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1084 nd->path.dentry->d_inode))
1085 audit_inode(name, nd->path.dentry);
2a737871
AV		 1086 if (nd->root.mnt) {
1087 path_put(&nd->root);
1088 nd->root.mnt = NULL;
1089 }
170aa3d0 1090 return retval;
1da177e4
LT		 1091}
1092
fc9b52cd 1093int path_lookup(const char *name, unsigned int flags,
5590ff0d
UD		 1094 struct nameidata *nd)
1095{
1096 return do_path_lookup(AT_FDCWD, name, flags, nd);
1097}
1098
d1811465
AV		 1099int kern_path(const char *name, unsigned int flags, struct path *path)
1100{
1101 struct nameidata nd;
1102 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1103 if (!res)
1104 *path = nd.path;
1105 return res;
1106}
1107
16f18200
JJS		 1108/**
1109 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1110 * @dentry: pointer to dentry of the base directory
1111 * @mnt: pointer to vfs mount of the base directory
1112 * @name: pointer to file name
1113 * @flags: lookup flags
1114 * @nd: pointer to nameidata
1115 */
1116int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1117 const char *name, unsigned int flags,
1118 struct nameidata *nd)
1119{
1120 int retval;
1121
1122 /* same as do_path_lookup */
1123 nd->last_type = LAST_ROOT;
1124 nd->flags = flags;
1125 nd->depth = 0;
1126
c8e7f449
JB		 1127 nd->path.dentry = dentry;
1128 nd->path.mnt = mnt;
1129 path_get(&nd->path);
5b857119
AV		 1130 nd->root = nd->path;
1131 path_get(&nd->root);
16f18200
JJS		 1132
1133 retval = path_walk(name, nd);
4ac91378
JB		 1134 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1135 nd->path.dentry->d_inode))
1136 audit_inode(name, nd->path.dentry);
16f18200 1137
5b857119
AV		 1138 path_put(&nd->root);
1139 nd->root.mnt = NULL;
16f18200 1140
2a737871 1141 return retval;
16f18200
JJS		 1142}
1143
8737f3a1
AV		 1144/**
1145 * path_lookup_open - lookup a file path with open intent
1146 * @dfd: the directory to use as base, or AT_FDCWD
1147 * @name: pointer to file name
1148 * @lookup_flags: lookup intent flags
1149 * @nd: pointer to nameidata
1150 * @open_flags: open intent flags
1151 */
6e8341a1
AV		 1152static int path_lookup_open(int dfd, const char *name,
1153 unsigned int lookup_flags, struct nameidata *nd, int open_flags)
834f2a4a
TM		 1154{
1155 struct file *filp = get_empty_filp();
1156 int err;
1157
1158 if (filp == NULL)
1159 return -ENFILE;
1160 nd->intent.open.file = filp;
1161 nd->intent.open.flags = open_flags;
8737f3a1 1162 nd->intent.open.create_mode = 0;
5590ff0d 1163 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
834f2a4a
TM		 1164 if (IS_ERR(nd->intent.open.file)) {
1165 if (err == 0) {
1166 err = PTR_ERR(nd->intent.open.file);
1d957f9b 1167 path_put(&nd->path);
834f2a4a
TM		 1168 }
1169 } else if (err != 0)
1170 release_open_intent(nd);
1171 return err;
1172}
1173
eead1911
CH		 1174static struct dentry *__lookup_hash(struct qstr *name,
1175 struct dentry *base, struct nameidata *nd)
1da177e4 1176{
057f6c01 1177 struct dentry *dentry;
1da177e4
LT		 1178 struct inode *inode;
1179 int err;
1180
1181 inode = base->d_inode;
1da177e4
LT		 1182
1183 /*
1184 * See if the low-level filesystem might want
1185 * to use its own hash..
1186 */
1187 if (base->d_op && base->d_op->d_hash) {
1188 err = base->d_op->d_hash(base, name);
1189 dentry = ERR_PTR(err);
1190 if (err < 0)
1191 goto out;
1192 }
1193
1194 dentry = cached_lookup(base, name, nd);
1195 if (!dentry) {
d70b67c8
MS		 1196 struct dentry *new;
1197
1198 /* Don't create child dentry for a dead directory. */
1199 dentry = ERR_PTR(-ENOENT);
1200 if (IS_DEADDIR(inode))
1201 goto out;
1202
1203 new = d_alloc(base, name);
1da177e4
LT		 1204 dentry = ERR_PTR(-ENOMEM);
1205 if (!new)
1206 goto out;
1207 dentry = inode->i_op->lookup(inode, new, nd);
1208 if (!dentry)
1209 dentry = new;
1210 else
1211 dput(new);
1212 }
1213out:
1214 return dentry;
1215}
1216
057f6c01
JM		 1217/*
1218 * Restricted form of lookup. Doesn't follow links, single-component only,
1219 * needs parent already locked. Doesn't follow mounts.
1220 * SMP-safe.
1221 */
eead1911 1222static struct dentry *lookup_hash(struct nameidata *nd)
057f6c01 1223{
057f6c01
JM		 1224 int err;
1225
f419a2e3 1226 err = inode_permission(nd->path.dentry->d_inode, MAY_EXEC);
057f6c01 1227 if (err)
eead1911 1228 return ERR_PTR(err);
4ac91378 1229 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1da177e4
LT		 1230}
1231
eead1911
CH		 1232static int __lookup_one_len(const char *name, struct qstr *this,
1233 struct dentry *base, int len)
1da177e4
LT		 1234{
1235 unsigned long hash;
1da177e4
LT		 1236 unsigned int c;
1237
057f6c01
JM		 1238 this->name = name;
1239 this->len = len;
1da177e4 1240 if (!len)
057f6c01 1241 return -EACCES;
1da177e4
LT		 1242
1243 hash = init_name_hash();
1244 while (len--) {
1245 c = *(const unsigned char *)name++;
1246 if (c == '/' || c == '\0')
057f6c01 1247 return -EACCES;
1da177e4
LT		 1248 hash = partial_name_hash(c, hash);
1249 }
057f6c01
JM		 1250 this->hash = end_name_hash(hash);
1251 return 0;
1252}
1da177e4 1253
eead1911 1254/**
a6b91919 1255 * lookup_one_len - filesystem helper to lookup single pathname component
eead1911
CH		 1256 * @name: pathname component to lookup
1257 * @base: base directory to lookup from
1258 * @len: maximum length @len should be interpreted to
1259 *
a6b91919
RD		 1260 * Note that this routine is purely a helper for filesystem usage and should
1261 * not be called by generic code. Also note that by using this function the
eead1911
CH		 1262 * nameidata argument is passed to the filesystem methods and a filesystem
1263 * using this helper needs to be prepared for that.
1264 */
057f6c01
JM		 1265struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1266{
1267 int err;
1268 struct qstr this;
1269
2f9092e1
DW		 1270 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1271
057f6c01 1272 err = __lookup_one_len(name, &this, base, len);
eead1911
CH		 1273 if (err)
1274 return ERR_PTR(err);
1275
f419a2e3 1276 err = inode_permission(base->d_inode, MAY_EXEC);
057f6c01
JM		 1277 if (err)
1278 return ERR_PTR(err);
49705b77 1279 return __lookup_hash(&this, base, NULL);
057f6c01
JM		 1280}
1281
eead1911
CH		 1282/**
1283 * lookup_one_noperm - bad hack for sysfs
1284 * @name: pathname component to lookup
1285 * @base: base directory to lookup from
1286 *
1287 * This is a variant of lookup_one_len that doesn't perform any permission
1288 * checks. It's a horrible hack to work around the braindead sysfs
1289 * architecture and should not be used anywhere else.
1290 *
1291 * DON'T USE THIS FUNCTION EVER, thanks.
1292 */
1293struct dentry *lookup_one_noperm(const char *name, struct dentry *base)
057f6c01
JM		 1294{
1295 int err;
1296 struct qstr this;
1297
eead1911 1298 err = __lookup_one_len(name, &this, base, strlen(name));
057f6c01
JM		 1299 if (err)
1300 return ERR_PTR(err);
eead1911 1301 return __lookup_hash(&this, base, NULL);
1da177e4
LT		 1302}
1303
2d8f3038
AV		 1304int user_path_at(int dfd, const char __user *name, unsigned flags,
1305 struct path *path)
1da177e4 1306{
2d8f3038 1307 struct nameidata nd;
1da177e4
LT		 1308 char *tmp = getname(name);
1309 int err = PTR_ERR(tmp);
1da177e4 1310 if (!IS_ERR(tmp)) {
2d8f3038
AV		 1311
1312 BUG_ON(flags & LOOKUP_PARENT);
1313
1314 err = do_path_lookup(dfd, tmp, flags, &nd);
1da177e4 1315 putname(tmp);
2d8f3038
AV		 1316 if (!err)
1317 *path = nd.path;
1da177e4
LT		 1318 }
1319 return err;
1320}
1321
2ad94ae6
AV		 1322static int user_path_parent(int dfd, const char __user *path,
1323 struct nameidata *nd, char **name)
1324{
1325 char *s = getname(path);
1326 int error;
1327
1328 if (IS_ERR(s))
1329 return PTR_ERR(s);
1330
1331 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1332 if (error)
1333 putname(s);
1334 else
1335 *name = s;
1336
1337 return error;
1338}
1339
1da177e4
LT		 1340/*
1341 * It's inline, so penalty for filesystems that don't use sticky bit is
1342 * minimal.
1343 */
1344static inline int check_sticky(struct inode *dir, struct inode *inode)
1345{
da9592ed
DH		 1346 uid_t fsuid = current_fsuid();
1347
1da177e4
LT		 1348 if (!(dir->i_mode & S_ISVTX))
1349 return 0;
da9592ed 1350 if (inode->i_uid == fsuid)
1da177e4 1351 return 0;
da9592ed 1352 if (dir->i_uid == fsuid)
1da177e4
LT		 1353 return 0;
1354 return !capable(CAP_FOWNER);
1355}
1356
1357/*
1358 * Check whether we can remove a link victim from directory dir, check
1359 * whether the type of victim is right.
1360 * 1. We can't do it if dir is read-only (done in permission())
1361 * 2. We should have write and exec permissions on dir
1362 * 3. We can't remove anything from append-only dir
1363 * 4. We can't do anything with immutable dir (done in permission())
1364 * 5. If the sticky bit on dir is set we should either
1365 * a. be owner of dir, or
1366 * b. be owner of victim, or
1367 * c. have CAP_FOWNER capability
1368 * 6. If the victim is append-only or immutable we can't do antyhing with
1369 * links pointing to it.
1370 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1371 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1372 * 9. We can't remove a root or mountpoint.
1373 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1374 * nfs_async_unlink().
1375 */
858119e1 1376static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1da177e4
LT		 1377{
1378 int error;
1379
1380 if (!victim->d_inode)
1381 return -ENOENT;
1382
1383 BUG_ON(victim->d_parent->d_inode != dir);
5a190ae6 1384 audit_inode_child(victim->d_name.name, victim, dir);
1da177e4 1385
f419a2e3 1386 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1387 if (error)
1388 return error;
1389 if (IS_APPEND(dir))
1390 return -EPERM;
1391 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
f9454548 1392 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1da177e4
LT		 1393 return -EPERM;
1394 if (isdir) {
1395 if (!S_ISDIR(victim->d_inode->i_mode))
1396 return -ENOTDIR;
1397 if (IS_ROOT(victim))
1398 return -EBUSY;
1399 } else if (S_ISDIR(victim->d_inode->i_mode))
1400 return -EISDIR;
1401 if (IS_DEADDIR(dir))
1402 return -ENOENT;
1403 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1404 return -EBUSY;
1405 return 0;
1406}
1407
1408/* Check whether we can create an object with dentry child in directory
1409 * dir.
1410 * 1. We can't do it if child already exists (open has special treatment for
1411 * this case, but since we are inlined it's OK)
1412 * 2. We can't do it if dir is read-only (done in permission())
1413 * 3. We should have write and exec permissions on dir
1414 * 4. We can't do it if dir is immutable (done in permission())
1415 */
a95164d9 1416static inline int may_create(struct inode *dir, struct dentry *child)
1da177e4
LT		 1417{
1418 if (child->d_inode)
1419 return -EEXIST;
1420 if (IS_DEADDIR(dir))
1421 return -ENOENT;
f419a2e3 1422 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1423}
1424
1425/*
1da177e4
LT		 1426 * O_DIRECTORY translates into forcing a directory lookup.
1427 */
1428static inline int lookup_flags(unsigned int f)
1429{
1430 unsigned long retval = LOOKUP_FOLLOW;
1431
1432 if (f & O_NOFOLLOW)
1433 retval &= ~LOOKUP_FOLLOW;
1434
1da177e4
LT		 1435 if (f & O_DIRECTORY)
1436 retval |= LOOKUP_DIRECTORY;
1437
1438 return retval;
1439}
1440
1441/*
1442 * p1 and p2 should be directories on the same fs.
1443 */
1444struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1445{
1446 struct dentry *p;
1447
1448 if (p1 == p2) {
f2eace23 1449 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1da177e4
LT		 1450 return NULL;
1451 }
1452
a11f3a05 1453 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4 1454
e2761a11
OH		 1455 p = d_ancestor(p2, p1);
1456 if (p) {
1457 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1458 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1459 return p;
1da177e4
LT		 1460 }
1461
e2761a11
OH		 1462 p = d_ancestor(p1, p2);
1463 if (p) {
1464 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1465 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1466 return p;
1da177e4
LT		 1467 }
1468
f2eace23
IM		 1469 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1470 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1da177e4
LT		 1471 return NULL;
1472}
1473
1474void unlock_rename(struct dentry *p1, struct dentry *p2)
1475{
1b1dcc1b 1476 mutex_unlock(&p1->d_inode->i_mutex);
1da177e4 1477 if (p1 != p2) {
1b1dcc1b 1478 mutex_unlock(&p2->d_inode->i_mutex);
a11f3a05 1479 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4
LT		 1480 }
1481}
1482
1483int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1484 struct nameidata *nd)
1485{
a95164d9 1486 int error = may_create(dir, dentry);
1da177e4
LT		 1487
1488 if (error)
1489 return error;
1490
acfa4380 1491 if (!dir->i_op->create)
1da177e4
LT		 1492 return -EACCES; /* shouldn't it be ENOSYS? */
1493 mode &= S_IALLUGO;
1494 mode |= S_IFREG;
1495 error = security_inode_create(dir, dentry, mode);
1496 if (error)
1497 return error;
9e3509e2 1498 vfs_dq_init(dir);
1da177e4 1499 error = dir->i_op->create(dir, dentry, mode, nd);
a74574aa 1500 if (!error)
f38aa942 1501 fsnotify_create(dir, dentry);
1da177e4
LT		 1502 return error;
1503}
1504
3fb64190 1505int may_open(struct path *path, int acc_mode, int flag)
1da177e4 1506{
3fb64190 1507 struct dentry *dentry = path->dentry;
1da177e4
LT		 1508 struct inode *inode = dentry->d_inode;
1509 int error;
1510
1511 if (!inode)
1512 return -ENOENT;
1513
c8fe8f30
CH		 1514 switch (inode->i_mode & S_IFMT) {
1515 case S_IFLNK:
1da177e4 1516 return -ELOOP;
c8fe8f30
CH		 1517 case S_IFDIR:
1518 if (acc_mode & MAY_WRITE)
1519 return -EISDIR;
1520 break;
1521 case S_IFBLK:
1522 case S_IFCHR:
3fb64190 1523 if (path->mnt->mnt_flags & MNT_NODEV)
1da177e4 1524 return -EACCES;
c8fe8f30
CH		 1525 /*FALLTHRU*/
1526 case S_IFIFO:
1527 case S_IFSOCK:
1da177e4 1528 flag &= ~O_TRUNC;
c8fe8f30 1529 break;
4a3fd211 1530 }
b41572e9 1531
3fb64190 1532 error = inode_permission(inode, acc_mode);
b41572e9
DH		 1533 if (error)
1534 return error;
6146f0d5 1535
6c1488fd
MZ		 1536 error = ima_path_check(path, acc_mode ?
1537 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC) :
1538 ACC_MODE(flag) & (MAY_READ | MAY_WRITE),
b9fc745d 1539 IMA_COUNT_UPDATE);
6c1488fd 1540
6146f0d5
MZ		 1541 if (error)
1542 return error;
1da177e4
LT		 1543 /*
1544 * An append-only file must be opened in append mode for writing.
1545 */
1546 if (IS_APPEND(inode)) {
acd0c935 1547 error = -EPERM;
1da177e4 1548 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
acd0c935 1549 goto err_out;
1da177e4 1550 if (flag & O_TRUNC)
acd0c935 1551 goto err_out;
1da177e4
LT		 1552 }
1553
1554 /* O_NOATIME can only be set by the owner or superuser */
1555 if (flag & O_NOATIME)
acd0c935
MZ		 1556 if (!is_owner_or_cap(inode)) {
1557 error = -EPERM;
1558 goto err_out;
1559 }
1da177e4
LT		 1560
1561 /*
1562 * Ensure there are no outstanding leases on the file.
1563 */
1564 error = break_lease(inode, flag);
1565 if (error)
acd0c935 1566 goto err_out;
1da177e4
LT		 1567
1568 if (flag & O_TRUNC) {
1569 error = get_write_access(inode);
1570 if (error)
acd0c935 1571 goto err_out;
1da177e4
LT		 1572
1573 /*
1574 * Refuse to truncate files with mandatory locks held on them.
1575 */
1576 error = locks_verify_locked(inode);
be6d3e56 1577 if (!error)
3fb64190 1578 error = security_path_truncate(path, 0,
be6d3e56 1579 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN);
1da177e4 1580 if (!error) {
9e3509e2 1581 vfs_dq_init(inode);
d139d7ff
MS		 1582
1583 error = do_truncate(dentry, 0,
1584 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1585 NULL);
1da177e4
LT		 1586 }
1587 put_write_access(inode);
1588 if (error)
acd0c935 1589 goto err_out;
1da177e4
LT		 1590 } else
1591 if (flag & FMODE_WRITE)
9e3509e2 1592 vfs_dq_init(inode);
1da177e4
LT		 1593
1594 return 0;
acd0c935
MZ		 1595err_out:
1596 ima_counts_put(path, acc_mode ?
1597 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC) :
1598 ACC_MODE(flag) & (MAY_READ | MAY_WRITE));
1599 return error;
1da177e4
LT		 1600}
1601
d57999e1
DH		 1602/*
1603 * Be careful about ever adding any more callers of this
1604 * function. Its flags must be in the namei format, not
1605 * what get passed to sys_open().
1606 */
1607static int __open_namei_create(struct nameidata *nd, struct path *path,
aab520e2
DH		 1608 int flag, int mode)
1609{
1610 int error;
4ac91378 1611 struct dentry *dir = nd->path.dentry;
aab520e2
DH		 1612
1613 if (!IS_POSIXACL(dir->d_inode))
ce3b0f8d 1614 mode &= ~current_umask();
be6d3e56
KT		 1615 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1616 if (error)
1617 goto out_unlock;
aab520e2 1618 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
be6d3e56 1619out_unlock:
aab520e2 1620 mutex_unlock(&dir->d_inode->i_mutex);
4ac91378
JB		 1621 dput(nd->path.dentry);
1622 nd->path.dentry = path->dentry;
aab520e2
DH		 1623 if (error)
1624 return error;
1625 /* Don't check for write permission, don't truncate */
3fb64190 1626 return may_open(&nd->path, 0, flag & ~O_TRUNC);
aab520e2
DH		 1627}
1628
d57999e1
DH		 1629/*
1630 * Note that while the flag value (low two bits) for sys_open means:
1631 * 00 - read-only
1632 * 01 - write-only
1633 * 10 - read-write
1634 * 11 - special
1635 * it is changed into
1636 * 00 - no permissions needed
1637 * 01 - read-permission
1638 * 10 - write-permission
1639 * 11 - read-write
1640 * for the internal routines (ie open_namei()/follow_link() etc)
1641 * This is more logical, and also allows the 00 "no perm needed"
1642 * to be used for symlinks (where the permissions are checked
1643 * later).
1644 *
1645*/
1646static inline int open_to_namei_flags(int flag)
1647{
1648 if ((flag+1) & O_ACCMODE)
1649 flag++;
1650 return flag;
1651}
1652
4a3fd211
DH		 1653static int open_will_write_to_fs(int flag, struct inode *inode)
1654{
1655 /*
1656 * We'll never write to the fs underlying
1657 * a device file.
1658 */
1659 if (special_file(inode->i_mode))
1660 return 0;
1661 return (flag & O_TRUNC);
1662}
1663
1da177e4 1664/*
4a3fd211
DH		 1665 * Note that the low bits of the passed in "open_flag"
1666 * are not the same as in the local variable "flag". See
1667 * open_to_namei_flags() for more details.
1da177e4 1668 */
a70e65df 1669struct file *do_filp_open(int dfd, const char *pathname,
6e8341a1 1670 int open_flag, int mode, int acc_mode)
1da177e4 1671{
4a3fd211 1672 struct file *filp;
a70e65df 1673 struct nameidata nd;
6e8341a1 1674 int error;
4e7506e4 1675 struct path path;
1da177e4
LT		 1676 struct dentry *dir;
1677 int count = 0;
4a3fd211 1678 int will_write;
d57999e1 1679 int flag = open_to_namei_flags(open_flag);
1da177e4 1680
6b2f3d1f
CH		 1681 /*
1682 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1683 * check for O_DSYNC if the need any syncing at all we enforce it's
1684 * always set instead of having to deal with possibly weird behaviour
1685 * for malicious applications setting only __O_SYNC.
1686 */
1687 if (open_flag & __O_SYNC)
1688 open_flag |= O_DSYNC;
1689
6e8341a1
AV		 1690 if (!acc_mode)
1691 acc_mode = MAY_OPEN | ACC_MODE(flag);
1da177e4 1692
834f2a4a
TM		 1693 /* O_TRUNC implies we need access checks for write permissions */
1694 if (flag & O_TRUNC)
1695 acc_mode |= MAY_WRITE;
1696
1da177e4
LT		 1697 /* Allow the LSM permission hook to distinguish append
1698 access from general write access. */
1699 if (flag & O_APPEND)
1700 acc_mode |= MAY_APPEND;
1701
1da177e4
LT		 1702 /*
1703 * The simplest case - just a plain lookup.
1704 */
1705 if (!(flag & O_CREAT)) {
5590ff0d 1706 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
a70e65df 1707 &nd, flag);
1da177e4 1708 if (error)
a70e65df 1709 return ERR_PTR(error);
1da177e4
LT		 1710 goto ok;
1711 }
1712
1713 /*
1714 * Create - we need to know the parent.
1715 */
9b4a9b14 1716 error = path_init(dfd, pathname, LOOKUP_PARENT, &nd);
1da177e4 1717 if (error)
a70e65df 1718 return ERR_PTR(error);
9b4a9b14 1719 error = path_walk(pathname, &nd);
654f562c
O		 1720 if (error) {
1721 if (nd.root.mnt)
1722 path_put(&nd.root);
9b4a9b14 1723 return ERR_PTR(error);
654f562c 1724 }
9b4a9b14
AV		 1725 if (unlikely(!audit_dummy_context()))
1726 audit_inode(pathname, nd.path.dentry);
1da177e4
LT		 1727
1728 /*
1729 * We have the parent and last component. First of all, check
1730 * that we are not asked to creat(2) an obvious directory - that
1731 * will not do.
1732 */
1733 error = -EISDIR;
a70e65df 1734 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
8737f3a1 1735 goto exit_parent;
1da177e4 1736
8737f3a1
AV		 1737 error = -ENFILE;
1738 filp = get_empty_filp();
1739 if (filp == NULL)
1740 goto exit_parent;
1741 nd.intent.open.file = filp;
1742 nd.intent.open.flags = flag;
1743 nd.intent.open.create_mode = mode;
a70e65df
CH		 1744 dir = nd.path.dentry;
1745 nd.flags &= ~LOOKUP_PARENT;
8737f3a1 1746 nd.flags |= LOOKUP_CREATE | LOOKUP_OPEN;
3516586a
AV		 1747 if (flag & O_EXCL)
1748 nd.flags |= LOOKUP_EXCL;
1b1dcc1b 1749 mutex_lock(&dir->d_inode->i_mutex);
a70e65df
CH		 1750 path.dentry = lookup_hash(&nd);
1751 path.mnt = nd.path.mnt;
1da177e4
LT		 1752
1753do_last:
4e7506e4
AV		 1754 error = PTR_ERR(path.dentry);
1755 if (IS_ERR(path.dentry)) {
1b1dcc1b 1756 mutex_unlock(&dir->d_inode->i_mutex);
1da177e4
LT		 1757 goto exit;
1758 }
1759
a70e65df 1760 if (IS_ERR(nd.intent.open.file)) {
a70e65df 1761 error = PTR_ERR(nd.intent.open.file);
4a3fd211 1762 goto exit_mutex_unlock;
4af4c52f
OD		 1763 }
1764
1da177e4 1765 /* Negative dentry, just create the file */
4e7506e4 1766 if (!path.dentry->d_inode) {
4a3fd211
DH		 1767 /*
1768 * This write is needed to ensure that a
1769 * ro->rw transition does not occur between
1770 * the time when the file is created and when
1771 * a permanent write count is taken through
1772 * the 'struct file' in nameidata_to_filp().
1773 */
1774 error = mnt_want_write(nd.path.mnt);
1da177e4 1775 if (error)
4a3fd211
DH		 1776 goto exit_mutex_unlock;
1777 error = __open_namei_create(&nd, &path, flag, mode);
1778 if (error) {
1779 mnt_drop_write(nd.path.mnt);
1da177e4 1780 goto exit;
4a3fd211
DH		 1781 }
1782 filp = nameidata_to_filp(&nd, open_flag);
94e5d714
MZ		 1783 if (IS_ERR(filp))
1784 ima_counts_put(&nd.path,
1785 acc_mode & (MAY_READ | MAY_WRITE |
1786 MAY_EXEC));
4a3fd211 1787 mnt_drop_write(nd.path.mnt);
654f562c
O		 1788 if (nd.root.mnt)
1789 path_put(&nd.root);
4a3fd211 1790 return filp;
1da177e4
LT		 1791 }
1792
1793 /*
1794 * It already exists.
1795 */
1b1dcc1b 1796 mutex_unlock(&dir->d_inode->i_mutex);
5a190ae6 1797 audit_inode(pathname, path.dentry);
1da177e4
LT		 1798
1799 error = -EEXIST;
1800 if (flag & O_EXCL)
1801 goto exit_dput;
1802
e13b210f 1803 if (__follow_mount(&path)) {
1da177e4 1804 error = -ELOOP;
ba7a4c1a
AV		 1805 if (flag & O_NOFOLLOW)
1806 goto exit_dput;
1da177e4 1807 }
3e2efce0 1808
1da177e4 1809 error = -ENOENT;
4e7506e4 1810 if (!path.dentry->d_inode)
1da177e4 1811 goto exit_dput;
acfa4380 1812 if (path.dentry->d_inode->i_op->follow_link)
1da177e4
LT		 1813 goto do_link;
1814
a70e65df 1815 path_to_nameidata(&path, &nd);
1da177e4 1816 error = -EISDIR;
4e7506e4 1817 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1da177e4
LT		 1818 goto exit;
1819ok:
4a3fd211
DH		 1820 /*
1821 * Consider:
1822 * 1. may_open() truncates a file
1823 * 2. a rw->ro mount transition occurs
1824 * 3. nameidata_to_filp() fails due to
1825 * the ro mount.
1826 * That would be inconsistent, and should
1827 * be avoided. Taking this mnt write here
1828 * ensures that (2) can not occur.
1829 */
1830 will_write = open_will_write_to_fs(flag, nd.path.dentry->d_inode);
1831 if (will_write) {
1832 error = mnt_want_write(nd.path.mnt);
1833 if (error)
1834 goto exit;
1835 }
3fb64190 1836 error = may_open(&nd.path, acc_mode, flag);
4a3fd211
DH		 1837 if (error) {
1838 if (will_write)
1839 mnt_drop_write(nd.path.mnt);
1da177e4 1840 goto exit;
4a3fd211
DH		 1841 }
1842 filp = nameidata_to_filp(&nd, open_flag);
94e5d714
MZ		 1843 if (IS_ERR(filp))
1844 ima_counts_put(&nd.path,
1845 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC));
4a3fd211
DH		 1846 /*
1847 * It is now safe to drop the mnt write
1848 * because the filp has had a write taken
1849 * on its behalf.
1850 */
1851 if (will_write)
1852 mnt_drop_write(nd.path.mnt);
654f562c
O		 1853 if (nd.root.mnt)
1854 path_put(&nd.root);
4a3fd211 1855 return filp;
1da177e4 1856
4a3fd211
DH		 1857exit_mutex_unlock:
1858 mutex_unlock(&dir->d_inode->i_mutex);
1da177e4 1859exit_dput:
a70e65df 1860 path_put_conditional(&path, &nd);
1da177e4 1861exit:
a70e65df
CH		 1862 if (!IS_ERR(nd.intent.open.file))
1863 release_open_intent(&nd);
8737f3a1 1864exit_parent:
2a737871
AV		 1865 if (nd.root.mnt)
1866 path_put(&nd.root);
a70e65df
CH		 1867 path_put(&nd.path);
1868 return ERR_PTR(error);
1da177e4
LT		 1869
1870do_link:
1871 error = -ELOOP;
1872 if (flag & O_NOFOLLOW)
1873 goto exit_dput;
1874 /*
1875 * This is subtle. Instead of calling do_follow_link() we do the
1876 * thing by hands. The reason is that this way we have zero link_count
1877 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1878 * After that we have the parent and last component, i.e.
1879 * we are in the same situation as after the first path_walk().
1880 * Well, almost - if the last component is normal we get its copy
1881 * stored in nd->last.name and we will have to putname() it when we
1882 * are done. Procfs-like symlinks just set LAST_BIND.
1883 */
a70e65df
CH		 1884 nd.flags |= LOOKUP_PARENT;
1885 error = security_inode_follow_link(path.dentry, &nd);
1da177e4
LT		 1886 if (error)
1887 goto exit_dput;
a70e65df 1888 error = __do_follow_link(&path, &nd);
de459215
KK		 1889 if (error) {
1890 /* Does someone understand code flow here? Or it is only
1891 * me so stupid? Anathema to whoever designed this non-sense
1892 * with "intent.open".
1893 */
a70e65df 1894 release_open_intent(&nd);
654f562c
O		 1895 if (nd.root.mnt)
1896 path_put(&nd.root);
a70e65df 1897 return ERR_PTR(error);
de459215 1898 }
a70e65df
CH		 1899 nd.flags &= ~LOOKUP_PARENT;
1900 if (nd.last_type == LAST_BIND)
1da177e4 1901 goto ok;
1da177e4 1902 error = -EISDIR;
a70e65df 1903 if (nd.last_type != LAST_NORM)
1da177e4 1904 goto exit;
a70e65df
CH		 1905 if (nd.last.name[nd.last.len]) {
1906 __putname(nd.last.name);
1da177e4
LT		 1907 goto exit;
1908 }
1909 error = -ELOOP;
1910 if (count++==32) {
a70e65df 1911 __putname(nd.last.name);
1da177e4
LT		 1912 goto exit;
1913 }
a70e65df 1914 dir = nd.path.dentry;
1b1dcc1b 1915 mutex_lock(&dir->d_inode->i_mutex);
a70e65df
CH		 1916 path.dentry = lookup_hash(&nd);
1917 path.mnt = nd.path.mnt;
1918 __putname(nd.last.name);
1da177e4
LT		 1919 goto do_last;
1920}
1921
a70e65df
CH		 1922/**
1923 * filp_open - open file and return file pointer
1924 *
1925 * @filename: path to open
1926 * @flags: open flags as per the open(2) second argument
1927 * @mode: mode for the new file if O_CREAT is set, else ignored
1928 *
1929 * This is the helper to open a file from kernelspace if you really
1930 * have to. But in generally you should not do this, so please move
1931 * along, nothing to see here..
1932 */
1933struct file *filp_open(const char *filename, int flags, int mode)
1934{
6e8341a1 1935 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
a70e65df
CH		 1936}
1937EXPORT_SYMBOL(filp_open);
1938
1da177e4
LT		 1939/**
1940 * lookup_create - lookup a dentry, creating it if it doesn't exist
1941 * @nd: nameidata info
1942 * @is_dir: directory flag
1943 *
1944 * Simple function to lookup and return a dentry and create it
1945 * if it doesn't exist. Is SMP-safe.
c663e5d8 1946 *
4ac91378 1947 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1da177e4
LT		 1948 */
1949struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1950{
c663e5d8 1951 struct dentry *dentry = ERR_PTR(-EEXIST);
1da177e4 1952
4ac91378 1953 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
c663e5d8
CH		 1954 /*
1955 * Yucky last component or no last component at all?
1956 * (foo/., foo/.., /////)
1957 */
1da177e4
LT		 1958 if (nd->last_type != LAST_NORM)
1959 goto fail;
1960 nd->flags &= ~LOOKUP_PARENT;
3516586a 1961 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
a634904a 1962 nd->intent.open.flags = O_EXCL;
c663e5d8
CH		 1963
1964 /*
1965 * Do the final lookup.
1966 */
49705b77 1967 dentry = lookup_hash(nd);
1da177e4
LT		 1968 if (IS_ERR(dentry))
1969 goto fail;
c663e5d8 1970
e9baf6e5
AV		 1971 if (dentry->d_inode)
1972 goto eexist;
c663e5d8
CH		 1973 /*
1974 * Special case - lookup gave negative, but... we had foo/bar/
1975 * From the vfs_mknod() POV we just have a negative dentry -
1976 * all is fine. Let's be bastards - you had / on the end, you've
1977 * been asking for (non-existent) directory. -ENOENT for you.
1978 */
e9baf6e5
AV		 1979 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1980 dput(dentry);
1981 dentry = ERR_PTR(-ENOENT);
1982 }
1da177e4 1983 return dentry;
e9baf6e5 1984eexist:
1da177e4 1985 dput(dentry);
e9baf6e5 1986 dentry = ERR_PTR(-EEXIST);
1da177e4
LT		 1987fail:
1988 return dentry;
1989}
f81a0bff 1990EXPORT_SYMBOL_GPL(lookup_create);
1da177e4
LT		 1991
1992int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1993{
a95164d9 1994 int error = may_create(dir, dentry);
1da177e4
LT		 1995
1996 if (error)
1997 return error;
1998
1999 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2000 return -EPERM;
2001
acfa4380 2002 if (!dir->i_op->mknod)
1da177e4
LT		 2003 return -EPERM;
2004
08ce5f16
SH		 2005 error = devcgroup_inode_mknod(mode, dev);
2006 if (error)
2007 return error;
2008
1da177e4
LT		 2009 error = security_inode_mknod(dir, dentry, mode, dev);
2010 if (error)
2011 return error;
2012
9e3509e2 2013 vfs_dq_init(dir);
1da177e4 2014 error = dir->i_op->mknod(dir, dentry, mode, dev);
a74574aa 2015 if (!error)
f38aa942 2016 fsnotify_create(dir, dentry);
1da177e4
LT		 2017 return error;
2018}
2019
463c3197
DH		 2020static int may_mknod(mode_t mode)
2021{
2022 switch (mode & S_IFMT) {
2023 case S_IFREG:
2024 case S_IFCHR:
2025 case S_IFBLK:
2026 case S_IFIFO:
2027 case S_IFSOCK:
2028 case 0: /* zero mode translates to S_IFREG */
2029 return 0;
2030 case S_IFDIR:
2031 return -EPERM;
2032 default:
2033 return -EINVAL;
2034 }
2035}
2036
2e4d0924
HC		 2037SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2038 unsigned, dev)
1da177e4 2039{
2ad94ae6
AV		 2040 int error;
2041 char *tmp;
2042 struct dentry *dentry;
1da177e4
LT		 2043 struct nameidata nd;
2044
2045 if (S_ISDIR(mode))
2046 return -EPERM;
1da177e4 2047
2ad94ae6 2048 error = user_path_parent(dfd, filename, &nd, &tmp);
1da177e4 2049 if (error)
2ad94ae6
AV		 2050 return error;
2051
1da177e4 2052 dentry = lookup_create(&nd, 0);
463c3197
DH		 2053 if (IS_ERR(dentry)) {
2054 error = PTR_ERR(dentry);
2055 goto out_unlock;
2056 }
4ac91378 2057 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2058 mode &= ~current_umask();
463c3197
DH		 2059 error = may_mknod(mode);
2060 if (error)
2061 goto out_dput;
2062 error = mnt_want_write(nd.path.mnt);
2063 if (error)
2064 goto out_dput;
be6d3e56
KT		 2065 error = security_path_mknod(&nd.path, dentry, mode, dev);
2066 if (error)
2067 goto out_drop_write;
463c3197 2068 switch (mode & S_IFMT) {
1da177e4 2069 case 0: case S_IFREG:
4ac91378 2070 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
1da177e4
LT		 2071 break;
2072 case S_IFCHR: case S_IFBLK:
4ac91378 2073 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
1da177e4
LT		 2074 new_decode_dev(dev));
2075 break;
2076 case S_IFIFO: case S_IFSOCK:
4ac91378 2077 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
1da177e4 2078 break;
1da177e4 2079 }
be6d3e56 2080out_drop_write:
463c3197
DH		 2081 mnt_drop_write(nd.path.mnt);
2082out_dput:
2083 dput(dentry);
2084out_unlock:
4ac91378 2085 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2086 path_put(&nd.path);
1da177e4
LT		 2087 putname(tmp);
2088
2089 return error;
2090}
2091
3480b257 2092SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
5590ff0d
UD		 2093{
2094 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2095}
2096
1da177e4
LT		 2097int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2098{
a95164d9 2099 int error = may_create(dir, dentry);
1da177e4
LT		 2100
2101 if (error)
2102 return error;
2103
acfa4380 2104 if (!dir->i_op->mkdir)
1da177e4
LT		 2105 return -EPERM;
2106
2107 mode &= (S_IRWXUGO|S_ISVTX);
2108 error = security_inode_mkdir(dir, dentry, mode);
2109 if (error)
2110 return error;
2111
9e3509e2 2112 vfs_dq_init(dir);
1da177e4 2113 error = dir->i_op->mkdir(dir, dentry, mode);
a74574aa 2114 if (!error)
f38aa942 2115 fsnotify_mkdir(dir, dentry);
1da177e4
LT		 2116 return error;
2117}
2118
2e4d0924 2119SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
1da177e4
LT		 2120{
2121 int error = 0;
2122 char * tmp;
6902d925
DH		 2123 struct dentry *dentry;
2124 struct nameidata nd;
1da177e4 2125
2ad94ae6
AV		 2126 error = user_path_parent(dfd, pathname, &nd, &tmp);
2127 if (error)
6902d925 2128 goto out_err;
1da177e4 2129
6902d925
DH		 2130 dentry = lookup_create(&nd, 1);
2131 error = PTR_ERR(dentry);
2132 if (IS_ERR(dentry))
2133 goto out_unlock;
1da177e4 2134
4ac91378 2135 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2136 mode &= ~current_umask();
463c3197
DH		 2137 error = mnt_want_write(nd.path.mnt);
2138 if (error)
2139 goto out_dput;
be6d3e56
KT		 2140 error = security_path_mkdir(&nd.path, dentry, mode);
2141 if (error)
2142 goto out_drop_write;
4ac91378 2143 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
be6d3e56 2144out_drop_write:
463c3197
DH		 2145 mnt_drop_write(nd.path.mnt);
2146out_dput:
6902d925
DH		 2147 dput(dentry);
2148out_unlock:
4ac91378 2149 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2150 path_put(&nd.path);
6902d925
DH		 2151 putname(tmp);
2152out_err:
1da177e4
LT		 2153 return error;
2154}
2155
3cdad428 2156SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
5590ff0d
UD		 2157{
2158 return sys_mkdirat(AT_FDCWD, pathname, mode);
2159}
2160
1da177e4
LT		 2161/*
2162 * We try to drop the dentry early: we should have
2163 * a usage count of 2 if we're the only user of this
2164 * dentry, and if that is true (possibly after pruning
2165 * the dcache), then we drop the dentry now.
2166 *
2167 * A low-level filesystem can, if it choses, legally
2168 * do a
2169 *
2170 * if (!d_unhashed(dentry))
2171 * return -EBUSY;
2172 *
2173 * if it cannot handle the case of removing a directory
2174 * that is still in use by something else..
2175 */
2176void dentry_unhash(struct dentry *dentry)
2177{
2178 dget(dentry);
dc168427 2179 shrink_dcache_parent(dentry);
1da177e4
LT		 2180 spin_lock(&dcache_lock);
2181 spin_lock(&dentry->d_lock);
2182 if (atomic_read(&dentry->d_count) == 2)
2183 __d_drop(dentry);
2184 spin_unlock(&dentry->d_lock);
2185 spin_unlock(&dcache_lock);
2186}
2187
2188int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2189{
2190 int error = may_delete(dir, dentry, 1);
2191
2192 if (error)
2193 return error;
2194
acfa4380 2195 if (!dir->i_op->rmdir)
1da177e4
LT		 2196 return -EPERM;
2197
9e3509e2 2198 vfs_dq_init(dir);
1da177e4 2199
1b1dcc1b 2200 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2201 dentry_unhash(dentry);
2202 if (d_mountpoint(dentry))
2203 error = -EBUSY;
2204 else {
2205 error = security_inode_rmdir(dir, dentry);
2206 if (!error) {
2207 error = dir->i_op->rmdir(dir, dentry);
2208 if (!error)
2209 dentry->d_inode->i_flags |= S_DEAD;
2210 }
2211 }
1b1dcc1b 2212 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4 2213 if (!error) {
1da177e4
LT		 2214 d_delete(dentry);
2215 }
2216 dput(dentry);
2217
2218 return error;
2219}
2220
5590ff0d 2221static long do_rmdir(int dfd, const char __user *pathname)
1da177e4
LT		 2222{
2223 int error = 0;
2224 char * name;
2225 struct dentry *dentry;
2226 struct nameidata nd;
2227
2ad94ae6 2228 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2229 if (error)
2ad94ae6 2230 return error;
1da177e4
LT		 2231
2232 switch(nd.last_type) {
0612d9fb
OH		 2233 case LAST_DOTDOT:
2234 error = -ENOTEMPTY;
2235 goto exit1;
2236 case LAST_DOT:
2237 error = -EINVAL;
2238 goto exit1;
2239 case LAST_ROOT:
2240 error = -EBUSY;
2241 goto exit1;
1da177e4 2242 }
0612d9fb
OH		 2243
2244 nd.flags &= ~LOOKUP_PARENT;
2245
4ac91378 2246 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2247 dentry = lookup_hash(&nd);
1da177e4 2248 error = PTR_ERR(dentry);
6902d925
DH		 2249 if (IS_ERR(dentry))
2250 goto exit2;
0622753b
DH		 2251 error = mnt_want_write(nd.path.mnt);
2252 if (error)
2253 goto exit3;
be6d3e56
KT		 2254 error = security_path_rmdir(&nd.path, dentry);
2255 if (error)
2256 goto exit4;
4ac91378 2257 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
be6d3e56 2258exit4:
0622753b
DH		 2259 mnt_drop_write(nd.path.mnt);
2260exit3:
6902d925
DH		 2261 dput(dentry);
2262exit2:
4ac91378 2263 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2264exit1:
1d957f9b 2265 path_put(&nd.path);
1da177e4
LT		 2266 putname(name);
2267 return error;
2268}
2269
3cdad428 2270SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
5590ff0d
UD		 2271{
2272 return do_rmdir(AT_FDCWD, pathname);
2273}
2274
1da177e4
LT		 2275int vfs_unlink(struct inode *dir, struct dentry *dentry)
2276{
2277 int error = may_delete(dir, dentry, 0);
2278
2279 if (error)
2280 return error;
2281
acfa4380 2282 if (!dir->i_op->unlink)
1da177e4
LT		 2283 return -EPERM;
2284
9e3509e2 2285 vfs_dq_init(dir);
1da177e4 2286
1b1dcc1b 2287 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2288 if (d_mountpoint(dentry))
2289 error = -EBUSY;
2290 else {
2291 error = security_inode_unlink(dir, dentry);
2292 if (!error)
2293 error = dir->i_op->unlink(dir, dentry);
2294 }
1b1dcc1b 2295 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2296
2297 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2298 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
ece95912 2299 fsnotify_link_count(dentry->d_inode);
e234f35c 2300 d_delete(dentry);
1da177e4 2301 }
0eeca283 2302
1da177e4
LT		 2303 return error;
2304}
2305
2306/*
2307 * Make sure that the actual truncation of the file will occur outside its
1b1dcc1b 2308 * directory's i_mutex. Truncate can take a long time if there is a lot of
1da177e4
LT		 2309 * writeout happening, and we don't want to prevent access to the directory
2310 * while waiting on the I/O.
2311 */
5590ff0d 2312static long do_unlinkat(int dfd, const char __user *pathname)
1da177e4 2313{
2ad94ae6
AV		 2314 int error;
2315 char *name;
1da177e4
LT		 2316 struct dentry *dentry;
2317 struct nameidata nd;
2318 struct inode *inode = NULL;
2319
2ad94ae6 2320 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2321 if (error)
2ad94ae6
AV		 2322 return error;
2323
1da177e4
LT		 2324 error = -EISDIR;
2325 if (nd.last_type != LAST_NORM)
2326 goto exit1;
0612d9fb
OH		 2327
2328 nd.flags &= ~LOOKUP_PARENT;
2329
4ac91378 2330 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2331 dentry = lookup_hash(&nd);
1da177e4
LT		 2332 error = PTR_ERR(dentry);
2333 if (!IS_ERR(dentry)) {
2334 /* Why not before? Because we want correct error value */
2335 if (nd.last.name[nd.last.len])
2336 goto slashes;
2337 inode = dentry->d_inode;
2338 if (inode)
2339 atomic_inc(&inode->i_count);
0622753b
DH		 2340 error = mnt_want_write(nd.path.mnt);
2341 if (error)
2342 goto exit2;
be6d3e56
KT		 2343 error = security_path_unlink(&nd.path, dentry);
2344 if (error)
2345 goto exit3;
4ac91378 2346 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
be6d3e56 2347exit3:
0622753b 2348 mnt_drop_write(nd.path.mnt);
1da177e4
LT		 2349 exit2:
2350 dput(dentry);
2351 }
4ac91378 2352 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4
LT		 2353 if (inode)
2354 iput(inode); /* truncate the inode here */
2355exit1:
1d957f9b 2356 path_put(&nd.path);
1da177e4
LT		 2357 putname(name);
2358 return error;
2359
2360slashes:
2361 error = !dentry->d_inode ? -ENOENT :
2362 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2363 goto exit2;
2364}
2365
2e4d0924 2366SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
5590ff0d
UD		 2367{
2368 if ((flag & ~AT_REMOVEDIR) != 0)
2369 return -EINVAL;
2370
2371 if (flag & AT_REMOVEDIR)
2372 return do_rmdir(dfd, pathname);
2373
2374 return do_unlinkat(dfd, pathname);
2375}
2376
3480b257 2377SYSCALL_DEFINE1(unlink, const char __user *, pathname)
5590ff0d
UD		 2378{
2379 return do_unlinkat(AT_FDCWD, pathname);
2380}
2381
db2e747b 2382int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
1da177e4 2383{
a95164d9 2384 int error = may_create(dir, dentry);
1da177e4
LT		 2385
2386 if (error)
2387 return error;
2388
acfa4380 2389 if (!dir->i_op->symlink)
1da177e4
LT		 2390 return -EPERM;
2391
2392 error = security_inode_symlink(dir, dentry, oldname);
2393 if (error)
2394 return error;
2395
9e3509e2 2396 vfs_dq_init(dir);
1da177e4 2397 error = dir->i_op->symlink(dir, dentry, oldname);
a74574aa 2398 if (!error)
f38aa942 2399 fsnotify_create(dir, dentry);
1da177e4
LT		 2400 return error;
2401}
2402
2e4d0924
HC		 2403SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2404 int, newdfd, const char __user *, newname)
1da177e4 2405{
2ad94ae6
AV		 2406 int error;
2407 char *from;
2408 char *to;
6902d925
DH		 2409 struct dentry *dentry;
2410 struct nameidata nd;
1da177e4
LT		 2411
2412 from = getname(oldname);
2ad94ae6 2413 if (IS_ERR(from))
1da177e4 2414 return PTR_ERR(from);
1da177e4 2415
2ad94ae6 2416 error = user_path_parent(newdfd, newname, &nd, &to);
6902d925 2417 if (error)
2ad94ae6
AV		 2418 goto out_putname;
2419
6902d925
DH		 2420 dentry = lookup_create(&nd, 0);
2421 error = PTR_ERR(dentry);
2422 if (IS_ERR(dentry))
2423 goto out_unlock;
2424
75c3f29d
DH		 2425 error = mnt_want_write(nd.path.mnt);
2426 if (error)
2427 goto out_dput;
be6d3e56
KT		 2428 error = security_path_symlink(&nd.path, dentry, from);
2429 if (error)
2430 goto out_drop_write;
db2e747b 2431 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
be6d3e56 2432out_drop_write:
75c3f29d
DH		 2433 mnt_drop_write(nd.path.mnt);
2434out_dput:
6902d925
DH		 2435 dput(dentry);
2436out_unlock:
4ac91378 2437 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2438 path_put(&nd.path);
6902d925
DH		 2439 putname(to);
2440out_putname:
1da177e4
LT		 2441 putname(from);
2442 return error;
2443}
2444
3480b257 2445SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 2446{
2447 return sys_symlinkat(oldname, AT_FDCWD, newname);
2448}
2449
1da177e4
LT		 2450int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2451{
2452 struct inode *inode = old_dentry->d_inode;
2453 int error;
2454
2455 if (!inode)
2456 return -ENOENT;
2457
a95164d9 2458 error = may_create(dir, new_dentry);
1da177e4
LT		 2459 if (error)
2460 return error;
2461
2462 if (dir->i_sb != inode->i_sb)
2463 return -EXDEV;
2464
2465 /*
2466 * A link to an append-only or immutable file cannot be created.
2467 */
2468 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2469 return -EPERM;
acfa4380 2470 if (!dir->i_op->link)
1da177e4 2471 return -EPERM;
7e79eedb 2472 if (S_ISDIR(inode->i_mode))
1da177e4
LT		 2473 return -EPERM;
2474
2475 error = security_inode_link(old_dentry, dir, new_dentry);
2476 if (error)
2477 return error;
2478
7e79eedb 2479 mutex_lock(&inode->i_mutex);
9e3509e2 2480 vfs_dq_init(dir);
1da177e4 2481 error = dir->i_op->link(old_dentry, dir, new_dentry);
7e79eedb 2482 mutex_unlock(&inode->i_mutex);
e31e14ec 2483 if (!error)
7e79eedb 2484 fsnotify_link(dir, inode, new_dentry);
1da177e4
LT		 2485 return error;
2486}
2487
2488/*
2489 * Hardlinks are often used in delicate situations. We avoid
2490 * security-related surprises by not following symlinks on the
2491 * newname. --KAB
2492 *
2493 * We don't follow them on the oldname either to be compatible
2494 * with linux 2.0, and to avoid hard-linking to directories
2495 * and other special files. --ADM
2496 */
2e4d0924
HC		 2497SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2498 int, newdfd, const char __user *, newname, int, flags)
1da177e4
LT		 2499{
2500 struct dentry *new_dentry;
2d8f3038
AV		 2501 struct nameidata nd;
2502 struct path old_path;
1da177e4 2503 int error;
2ad94ae6 2504 char *to;
1da177e4 2505
45c9b11a 2506 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
c04030e1
UD		 2507 return -EINVAL;
2508
2d8f3038
AV		 2509 error = user_path_at(olddfd, oldname,
2510 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2511 &old_path);
1da177e4 2512 if (error)
2ad94ae6
AV		 2513 return error;
2514
2515 error = user_path_parent(newdfd, newname, &nd, &to);
1da177e4
LT		 2516 if (error)
2517 goto out;
2518 error = -EXDEV;
2d8f3038 2519 if (old_path.mnt != nd.path.mnt)
1da177e4
LT		 2520 goto out_release;
2521 new_dentry = lookup_create(&nd, 0);
2522 error = PTR_ERR(new_dentry);
6902d925
DH		 2523 if (IS_ERR(new_dentry))
2524 goto out_unlock;
75c3f29d
DH		 2525 error = mnt_want_write(nd.path.mnt);
2526 if (error)
2527 goto out_dput;
be6d3e56
KT		 2528 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2529 if (error)
2530 goto out_drop_write;
2d8f3038 2531 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
be6d3e56 2532out_drop_write:
75c3f29d
DH		 2533 mnt_drop_write(nd.path.mnt);
2534out_dput:
6902d925
DH		 2535 dput(new_dentry);
2536out_unlock:
4ac91378 2537 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2538out_release:
1d957f9b 2539 path_put(&nd.path);
2ad94ae6 2540 putname(to);
1da177e4 2541out:
2d8f3038 2542 path_put(&old_path);
1da177e4
LT		 2543
2544 return error;
2545}
2546
3480b257 2547SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
5590ff0d 2548{
c04030e1 2549 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
5590ff0d
UD		 2550}
2551
1da177e4
LT		 2552/*
2553 * The worst of all namespace operations - renaming directory. "Perverted"
2554 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2555 * Problems:
2556 * a) we can get into loop creation. Check is done in is_subdir().
2557 * b) race potential - two innocent renames can create a loop together.
2558 * That's where 4.4 screws up. Current fix: serialization on
a11f3a05 2559 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
1da177e4
LT		 2560 * story.
2561 * c) we have to lock _three_ objects - parents and victim (if it exists).
1b1dcc1b 2562 * And that - after we got ->i_mutex on parents (until then we don't know
1da177e4
LT		 2563 * whether the target exists). Solution: try to be smart with locking
2564 * order for inodes. We rely on the fact that tree topology may change
a11f3a05 2565 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
1da177e4
LT		 2566 * move will be locked. Thus we can rank directories by the tree
2567 * (ancestors first) and rank all non-directories after them.
2568 * That works since everybody except rename does "lock parent, lookup,
a11f3a05 2569 * lock child" and rename is under ->s_vfs_rename_mutex.
1da177e4
LT		 2570 * HOWEVER, it relies on the assumption that any object with ->lookup()
2571 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2572 * we'd better make sure that there's no link(2) for them.
2573 * d) some filesystems don't support opened-but-unlinked directories,
2574 * either because of layout or because they are not ready to deal with
2575 * all cases correctly. The latter will be fixed (taking this sort of
2576 * stuff into VFS), but the former is not going away. Solution: the same
2577 * trick as in rmdir().
2578 * e) conversion from fhandle to dentry may come in the wrong moment - when
1b1dcc1b 2579 * we are removing the target. Solution: we will have to grab ->i_mutex
1da177e4 2580 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
1b1dcc1b 2581 * ->i_mutex on parents, which works but leads to some truely excessive
1da177e4
LT		 2582 * locking].
2583 */
75c96f85
AB		 2584static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2585 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 2586{
2587 int error = 0;
2588 struct inode *target;
2589
2590 /*
2591 * If we are going to change the parent - check write permissions,
2592 * we'll need to flip '..'.
2593 */
2594 if (new_dir != old_dir) {
f419a2e3 2595 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
1da177e4
LT		 2596 if (error)
2597 return error;
2598 }
2599
2600 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2601 if (error)
2602 return error;
2603
2604 target = new_dentry->d_inode;
2605 if (target) {
1b1dcc1b 2606 mutex_lock(&target->i_mutex);
1da177e4
LT		 2607 dentry_unhash(new_dentry);
2608 }
2609 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2610 error = -EBUSY;
2611 else
2612 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2613 if (target) {
2614 if (!error)
2615 target->i_flags |= S_DEAD;
1b1dcc1b 2616 mutex_unlock(&target->i_mutex);
1da177e4
LT		 2617 if (d_unhashed(new_dentry))
2618 d_rehash(new_dentry);
2619 dput(new_dentry);
2620 }
e31e14ec 2621 if (!error)
349457cc
MF		 2622 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2623 d_move(old_dentry,new_dentry);
1da177e4
LT		 2624 return error;
2625}
2626
75c96f85
AB		 2627static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2628 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 2629{
2630 struct inode *target;
2631 int error;
2632
2633 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2634 if (error)
2635 return error;
2636
2637 dget(new_dentry);
2638 target = new_dentry->d_inode;
2639 if (target)
1b1dcc1b 2640 mutex_lock(&target->i_mutex);
1da177e4
LT		 2641 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2642 error = -EBUSY;
2643 else
2644 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2645 if (!error) {
349457cc 2646 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
1da177e4 2647 d_move(old_dentry, new_dentry);
1da177e4
LT		 2648 }
2649 if (target)
1b1dcc1b 2650 mutex_unlock(&target->i_mutex);
1da177e4
LT		 2651 dput(new_dentry);
2652 return error;
2653}
2654
2655int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2656 struct inode *new_dir, struct dentry *new_dentry)
2657{
2658 int error;
2659 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
0eeca283 2660 const char *old_name;
1da177e4
LT		 2661
2662 if (old_dentry->d_inode == new_dentry->d_inode)
2663 return 0;
2664
2665 error = may_delete(old_dir, old_dentry, is_dir);
2666 if (error)
2667 return error;
2668
2669 if (!new_dentry->d_inode)
a95164d9 2670 error = may_create(new_dir, new_dentry);
1da177e4
LT		 2671 else
2672 error = may_delete(new_dir, new_dentry, is_dir);
2673 if (error)
2674 return error;
2675
acfa4380 2676 if (!old_dir->i_op->rename)
1da177e4
LT		 2677 return -EPERM;
2678
9e3509e2
JK		 2679 vfs_dq_init(old_dir);
2680 vfs_dq_init(new_dir);
1da177e4 2681
0eeca283
RL		 2682 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2683
1da177e4
LT		 2684 if (is_dir)
2685 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2686 else
2687 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2688 if (!error) {
0eeca283 2689 const char *new_name = old_dentry->d_name.name;
89204c40 2690 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
5a190ae6 2691 new_dentry->d_inode, old_dentry);
1da177e4 2692 }
0eeca283
RL		 2693 fsnotify_oldname_free(old_name);
2694
1da177e4
LT		 2695 return error;
2696}
2697
2e4d0924
HC		 2698SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2699 int, newdfd, const char __user *, newname)
1da177e4 2700{
2ad94ae6
AV		 2701 struct dentry *old_dir, *new_dir;
2702 struct dentry *old_dentry, *new_dentry;
2703 struct dentry *trap;
1da177e4 2704 struct nameidata oldnd, newnd;
2ad94ae6
AV		 2705 char *from;
2706 char *to;
2707 int error;
1da177e4 2708
2ad94ae6 2709 error = user_path_parent(olddfd, oldname, &oldnd, &from);
1da177e4
LT		 2710 if (error)
2711 goto exit;
2712
2ad94ae6 2713 error = user_path_parent(newdfd, newname, &newnd, &to);
1da177e4
LT		 2714 if (error)
2715 goto exit1;
2716
2717 error = -EXDEV;
4ac91378 2718 if (oldnd.path.mnt != newnd.path.mnt)
1da177e4
LT		 2719 goto exit2;
2720
4ac91378 2721 old_dir = oldnd.path.dentry;
1da177e4
LT		 2722 error = -EBUSY;
2723 if (oldnd.last_type != LAST_NORM)
2724 goto exit2;
2725
4ac91378 2726 new_dir = newnd.path.dentry;
1da177e4
LT		 2727 if (newnd.last_type != LAST_NORM)
2728 goto exit2;
2729
0612d9fb
OH		 2730 oldnd.flags &= ~LOOKUP_PARENT;
2731 newnd.flags &= ~LOOKUP_PARENT;
4e9ed2f8 2732 newnd.flags |= LOOKUP_RENAME_TARGET;
0612d9fb 2733
1da177e4
LT		 2734 trap = lock_rename(new_dir, old_dir);
2735
49705b77 2736 old_dentry = lookup_hash(&oldnd);
1da177e4
LT		 2737 error = PTR_ERR(old_dentry);
2738 if (IS_ERR(old_dentry))
2739 goto exit3;
2740 /* source must exist */
2741 error = -ENOENT;
2742 if (!old_dentry->d_inode)
2743 goto exit4;
2744 /* unless the source is a directory trailing slashes give -ENOTDIR */
2745 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2746 error = -ENOTDIR;
2747 if (oldnd.last.name[oldnd.last.len])
2748 goto exit4;
2749 if (newnd.last.name[newnd.last.len])
2750 goto exit4;
2751 }
2752 /* source should not be ancestor of target */
2753 error = -EINVAL;
2754 if (old_dentry == trap)
2755 goto exit4;
49705b77 2756 new_dentry = lookup_hash(&newnd);
1da177e4
LT		 2757 error = PTR_ERR(new_dentry);
2758 if (IS_ERR(new_dentry))
2759 goto exit4;
2760 /* target should not be an ancestor of source */
2761 error = -ENOTEMPTY;
2762 if (new_dentry == trap)
2763 goto exit5;
2764
9079b1eb
DH		 2765 error = mnt_want_write(oldnd.path.mnt);
2766 if (error)
2767 goto exit5;
be6d3e56
KT		 2768 error = security_path_rename(&oldnd.path, old_dentry,
2769 &newnd.path, new_dentry);
2770 if (error)
2771 goto exit6;
1da177e4
LT		 2772 error = vfs_rename(old_dir->d_inode, old_dentry,
2773 new_dir->d_inode, new_dentry);
be6d3e56 2774exit6:
9079b1eb 2775 mnt_drop_write(oldnd.path.mnt);
1da177e4
LT		 2776exit5:
2777 dput(new_dentry);
2778exit4:
2779 dput(old_dentry);
2780exit3:
2781 unlock_rename(new_dir, old_dir);
2782exit2:
1d957f9b 2783 path_put(&newnd.path);
2ad94ae6 2784 putname(to);
1da177e4 2785exit1:
1d957f9b 2786 path_put(&oldnd.path);
1da177e4 2787 putname(from);
2ad94ae6 2788exit:
1da177e4
LT		 2789 return error;
2790}
2791
a26eab24 2792SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 2793{
2794 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2795}
2796
1da177e4
LT		 2797int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2798{
2799 int len;
2800
2801 len = PTR_ERR(link);
2802 if (IS_ERR(link))
2803 goto out;
2804
2805 len = strlen(link);
2806 if (len > (unsigned) buflen)
2807 len = buflen;
2808 if (copy_to_user(buffer, link, len))
2809 len = -EFAULT;
2810out:
2811 return len;
2812}
2813
2814/*
2815 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2816 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2817 * using) it for any given inode is up to filesystem.
2818 */
2819int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2820{
2821 struct nameidata nd;
cc314eef 2822 void *cookie;
694a1764 2823 int res;
cc314eef 2824
1da177e4 2825 nd.depth = 0;
cc314eef 2826 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
694a1764
MS		 2827 if (IS_ERR(cookie))
2828 return PTR_ERR(cookie);
2829
2830 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2831 if (dentry->d_inode->i_op->put_link)
2832 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2833 return res;
1da177e4
LT		 2834}
2835
2836int vfs_follow_link(struct nameidata *nd, const char *link)
2837{
2838 return __vfs_follow_link(nd, link);
2839}
2840
2841/* get the link contents into pagecache */
2842static char *page_getlink(struct dentry * dentry, struct page **ppage)
2843{
ebd09abb
DG		 2844 char *kaddr;
2845 struct page *page;
1da177e4 2846 struct address_space *mapping = dentry->d_inode->i_mapping;
090d2b18 2847 page = read_mapping_page(mapping, 0, NULL);
1da177e4 2848 if (IS_ERR(page))
6fe6900e 2849 return (char*)page;
1da177e4 2850 *ppage = page;
ebd09abb
DG		 2851 kaddr = kmap(page);
2852 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2853 return kaddr;
1da177e4
LT		 2854}
2855
2856int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2857{
2858 struct page *page = NULL;
2859 char *s = page_getlink(dentry, &page);
2860 int res = vfs_readlink(dentry,buffer,buflen,s);
2861 if (page) {
2862 kunmap(page);
2863 page_cache_release(page);
2864 }
2865 return res;
2866}
2867
cc314eef 2868void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
1da177e4 2869{
cc314eef 2870 struct page *page = NULL;
1da177e4 2871 nd_set_link(nd, page_getlink(dentry, &page));
cc314eef 2872 return page;
1da177e4
LT		 2873}
2874
cc314eef 2875void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1da177e4 2876{
cc314eef
LT		 2877 struct page *page = cookie;
2878
2879 if (page) {
1da177e4
LT		 2880 kunmap(page);
2881 page_cache_release(page);
1da177e4
LT		 2882 }
2883}
2884
54566b2c
NP		 2885/*
2886 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2887 */
2888int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
1da177e4
LT		 2889{
2890 struct address_space *mapping = inode->i_mapping;
0adb25d2 2891 struct page *page;
afddba49 2892 void *fsdata;
beb497ab 2893 int err;
1da177e4 2894 char *kaddr;
54566b2c
NP		 2895 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2896 if (nofs)
2897 flags |= AOP_FLAG_NOFS;
1da177e4 2898
7e53cac4 2899retry:
afddba49 2900 err = pagecache_write_begin(NULL, mapping, 0, len-1,
54566b2c 2901 flags, &page, &fsdata);
1da177e4 2902 if (err)
afddba49
NP		 2903 goto fail;
2904
1da177e4
LT		 2905 kaddr = kmap_atomic(page, KM_USER0);
2906 memcpy(kaddr, symname, len-1);
2907 kunmap_atomic(kaddr, KM_USER0);
afddba49
NP		 2908
2909 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2910 page, fsdata);
1da177e4
LT		 2911 if (err < 0)
2912 goto fail;
afddba49
NP		 2913 if (err < len-1)
2914 goto retry;
2915
1da177e4
LT		 2916 mark_inode_dirty(inode);
2917 return 0;
1da177e4
LT		 2918fail:
2919 return err;
2920}
2921
0adb25d2
KK		 2922int page_symlink(struct inode *inode, const char *symname, int len)
2923{
2924 return __page_symlink(inode, symname, len,
54566b2c 2925 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
0adb25d2
KK		 2926}
2927
92e1d5be 2928const struct inode_operations page_symlink_inode_operations = {
1da177e4
LT		 2929 .readlink = generic_readlink,
2930 .follow_link = page_follow_link_light,
2931 .put_link = page_put_link,
2932};
2933
2d8f3038 2934EXPORT_SYMBOL(user_path_at);
1da177e4
LT		 2935EXPORT_SYMBOL(follow_down);
2936EXPORT_SYMBOL(follow_up);
2937EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2938EXPORT_SYMBOL(getname);
2939EXPORT_SYMBOL(lock_rename);
1da177e4
LT		 2940EXPORT_SYMBOL(lookup_one_len);
2941EXPORT_SYMBOL(page_follow_link_light);
2942EXPORT_SYMBOL(page_put_link);
2943EXPORT_SYMBOL(page_readlink);
0adb25d2 2944EXPORT_SYMBOL(__page_symlink);
1da177e4
LT		 2945EXPORT_SYMBOL(page_symlink);
2946EXPORT_SYMBOL(page_symlink_inode_operations);
2947EXPORT_SYMBOL(path_lookup);
d1811465 2948EXPORT_SYMBOL(kern_path);
16f18200 2949EXPORT_SYMBOL(vfs_path_lookup);
f419a2e3 2950EXPORT_SYMBOL(inode_permission);
8c744fb8 2951EXPORT_SYMBOL(file_permission);
1da177e4
LT		 2952EXPORT_SYMBOL(unlock_rename);
2953EXPORT_SYMBOL(vfs_create);
2954EXPORT_SYMBOL(vfs_follow_link);
2955EXPORT_SYMBOL(vfs_link);
2956EXPORT_SYMBOL(vfs_mkdir);
2957EXPORT_SYMBOL(vfs_mknod);
2958EXPORT_SYMBOL(generic_permission);
2959EXPORT_SYMBOL(vfs_readlink);
2960EXPORT_SYMBOL(vfs_rename);
2961EXPORT_SYMBOL(vfs_rmdir);
2962EXPORT_SYMBOL(vfs_symlink);
2963EXPORT_SYMBOL(vfs_unlink);
2964EXPORT_SYMBOL(dentry_unhash);
2965EXPORT_SYMBOL(generic_readlink);
kernel source for telekom puls (alcatel/mediatek)