projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
kill LOOKUP_CONTINUE
[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>
1da177e4 22#include <linux/pagemap.h>
0eeca283 23#include <linux/fsnotify.h>
1da177e4
LT		 24#include <linux/personality.h>
25#include <linux/security.h>
6146f0d5 26#include <linux/ima.h>
1da177e4
LT		 27#include <linux/syscalls.h>
28#include <linux/mount.h>
29#include <linux/audit.h>
16f7e0fe 30#include <linux/capability.h>
834f2a4a 31#include <linux/file.h>
5590ff0d 32#include <linux/fcntl.h>
08ce5f16 33#include <linux/device_cgroup.h>
5ad4e53b 34#include <linux/fs_struct.h>
1da177e4
LT		 35#include <asm/uaccess.h>
36
e81e3f4d
EP		 37#include "internal.h"
38
1da177e4
LT		 39/* [Feb-1997 T. Schoebel-Theuer]
40 * Fundamental changes in the pathname lookup mechanisms (namei)
41 * were necessary because of omirr. The reason is that omirr needs
42 * to know the _real_ pathname, not the user-supplied one, in case
43 * of symlinks (and also when transname replacements occur).
44 *
45 * The new code replaces the old recursive symlink resolution with
46 * an iterative one (in case of non-nested symlink chains). It does
47 * this with calls to <fs>_follow_link().
48 * As a side effect, dir_namei(), _namei() and follow_link() are now
49 * replaced with a single function lookup_dentry() that can handle all
50 * the special cases of the former code.
51 *
52 * With the new dcache, the pathname is stored at each inode, at least as
53 * long as the refcount of the inode is positive. As a side effect, the
54 * size of the dcache depends on the inode cache and thus is dynamic.
55 *
56 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57 * resolution to correspond with current state of the code.
58 *
59 * Note that the symlink resolution is not *completely* iterative.
60 * There is still a significant amount of tail- and mid- recursion in
61 * the algorithm. Also, note that <fs>_readlink() is not used in
62 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63 * may return different results than <fs>_follow_link(). Many virtual
64 * filesystems (including /proc) exhibit this behavior.
65 */
66
67/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69 * and the name already exists in form of a symlink, try to create the new
70 * name indicated by the symlink. The old code always complained that the
71 * name already exists, due to not following the symlink even if its target
72 * is nonexistent. The new semantics affects also mknod() and link() when
25985edc 73 * the name is a symlink pointing to a non-existent name.
1da177e4
LT		 74 *
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
82 */
83
84/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
86 *
87 * [10-Sep-98 Alan Modra] Another symlink change.
88 */
89
90/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
97 *
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
103 */
104/*
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
a11f3a05 106 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
1da177e4
LT		 107 * any extra contention...
108 */
109
110/* In order to reduce some races, while at the same time doing additional
111 * checking and hopefully speeding things up, we copy filenames to the
112 * kernel data space before using them..
113 *
114 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
115 * PATH_MAX includes the nul terminator --RR.
116 */
858119e1 117static int do_getname(const char __user *filename, char *page)
1da177e4
LT		 118{
119 int retval;
120 unsigned long len = PATH_MAX;
121
122 if (!segment_eq(get_fs(), KERNEL_DS)) {
123 if ((unsigned long) filename >= TASK_SIZE)
124 return -EFAULT;
125 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
126 len = TASK_SIZE - (unsigned long) filename;
127 }
128
129 retval = strncpy_from_user(page, filename, len);
130 if (retval > 0) {
131 if (retval < len)
132 return 0;
133 return -ENAMETOOLONG;
134 } else if (!retval)
135 retval = -ENOENT;
136 return retval;
137}
138
f52e0c11 139static char *getname_flags(const char __user * filename, int flags)
1da177e4
LT		 140{
141 char *tmp, *result;
142
143 result = ERR_PTR(-ENOMEM);
144 tmp = __getname();
145 if (tmp) {
146 int retval = do_getname(filename, tmp);
147
148 result = tmp;
149 if (retval < 0) {
f52e0c11
AV		 150 if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) {
151 __putname(tmp);
152 result = ERR_PTR(retval);
153 }
1da177e4
LT		 154 }
155 }
156 audit_getname(result);
157 return result;
158}
159
f52e0c11
AV		 160char *getname(const char __user * filename)
161{
162 return getname_flags(filename, 0);
163}
164
1da177e4
LT		 165#ifdef CONFIG_AUDITSYSCALL
166void putname(const char *name)
167{
5ac3a9c2 168 if (unlikely(!audit_dummy_context()))
1da177e4
LT		 169 audit_putname(name);
170 else
171 __putname(name);
172}
173EXPORT_SYMBOL(putname);
174#endif
175
5909ccaa
LT		 176/*
177 * This does basic POSIX ACL permission checking
1da177e4 178 */
7e40145e 179static int acl_permission_check(struct inode *inode, int mask)
1da177e4 180{
7e40145e 181 int (*check_acl)(struct inode *inode, int mask);
26cf46be 182 unsigned int mode = inode->i_mode;
1da177e4 183
9c2c7039 184 mask &= MAY_READ | MAY_WRITE | MAY_EXEC | MAY_NOT_BLOCK;
e6305c43 185
e795b717
SH		 186 if (current_user_ns() != inode_userns(inode))
187 goto other_perms;
188
da9592ed 189 if (current_fsuid() == inode->i_uid)
1da177e4
LT		 190 mode >>= 6;
191 else {
178ea735 192 check_acl = inode->i_op->check_acl;
1da177e4 193 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
7e40145e 194 int error = check_acl(inode, mask);
b74c79e9
NP		 195 if (error != -EAGAIN)
196 return error;
1da177e4
LT		 197 }
198
199 if (in_group_p(inode->i_gid))
200 mode >>= 3;
201 }
202
e795b717 203other_perms:
1da177e4
LT		 204 /*
205 * If the DACs are ok we don't need any capability check.
206 */
9c2c7039 207 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
1da177e4 208 return 0;
5909ccaa
LT		 209 return -EACCES;
210}
211
212/**
b74c79e9 213 * generic_permission - check for access rights on a Posix-like filesystem
5909ccaa
LT		 214 * @inode: inode to check access rights for
215 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
39191628 216 * @flags: IPERM_FLAG_ flags.
5909ccaa
LT		 217 *
218 * Used to check for read/write/execute permissions on a file.
219 * We use "fsuid" for this, letting us set arbitrary permissions
220 * for filesystem access without changing the "normal" uids which
b74c79e9
NP		 221 * are used for other things.
222 *
223 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
224 * request cannot be satisfied (eg. requires blocking or too much complexity).
225 * It would then be called again in ref-walk mode.
5909ccaa 226 */
2830ba7f 227int generic_permission(struct inode *inode, int mask)
5909ccaa
LT		 228{
229 int ret;
230
231 /*
232 * Do the basic POSIX ACL permission checks.
233 */
7e40145e 234 ret = acl_permission_check(inode, mask);
5909ccaa
LT		 235 if (ret != -EACCES)
236 return ret;
1da177e4 237
d594e7ec
AV		 238 if (S_ISDIR(inode->i_mode)) {
239 /* DACs are overridable for directories */
240 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
241 return 0;
242 if (!(mask & MAY_WRITE))
243 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
244 return 0;
245 return -EACCES;
246 }
1da177e4
LT		 247 /*
248 * Read/write DACs are always overridable.
d594e7ec
AV		 249 * Executable DACs are overridable when there is
250 * at least one exec bit set.
1da177e4 251 */
d594e7ec 252 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
e795b717 253 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
1da177e4
LT		 254 return 0;
255
256 /*
257 * Searching includes executable on directories, else just read.
258 */
7ea66001 259 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
d594e7ec 260 if (mask == MAY_READ)
e795b717 261 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
1da177e4
LT		 262 return 0;
263
264 return -EACCES;
265}
266
cb23beb5
CH		 267/**
268 * inode_permission - check for access rights to a given inode
269 * @inode: inode to check permission on
270 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
271 *
272 * Used to check for read/write/execute permissions on an inode.
273 * We use "fsuid" for this, letting us set arbitrary permissions
274 * for filesystem access without changing the "normal" uids which
275 * are used for other things.
276 */
f419a2e3 277int inode_permission(struct inode *inode, int mask)
1da177e4 278{
e6305c43 279 int retval;
1da177e4
LT		 280
281 if (mask & MAY_WRITE) {
22590e41 282 umode_t mode = inode->i_mode;
1da177e4
LT		 283
284 /*
285 * Nobody gets write access to a read-only fs.
286 */
287 if (IS_RDONLY(inode) &&
288 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
289 return -EROFS;
290
291 /*
292 * Nobody gets write access to an immutable file.
293 */
294 if (IS_IMMUTABLE(inode))
295 return -EACCES;
296 }
297
acfa4380 298 if (inode->i_op->permission)
10556cb2 299 retval = inode->i_op->permission(inode, mask);
f696a365 300 else
2830ba7f 301 retval = generic_permission(inode, mask);
f696a365 302
1da177e4
LT		 303 if (retval)
304 return retval;
305
08ce5f16
SH		 306 retval = devcgroup_inode_permission(inode, mask);
307 if (retval)
308 return retval;
309
d09ca739 310 return security_inode_permission(inode, mask);
1da177e4
LT		 311}
312
5dd784d0
JB		 313/**
314 * path_get - get a reference to a path
315 * @path: path to get the reference to
316 *
317 * Given a path increment the reference count to the dentry and the vfsmount.
318 */
319void path_get(struct path *path)
320{
321 mntget(path->mnt);
322 dget(path->dentry);
323}
324EXPORT_SYMBOL(path_get);
325
1d957f9b
JB		 326/**
327 * path_put - put a reference to a path
328 * @path: path to put the reference to
329 *
330 * Given a path decrement the reference count to the dentry and the vfsmount.
331 */
332void path_put(struct path *path)
1da177e4 333{
1d957f9b
JB		 334 dput(path->dentry);
335 mntput(path->mnt);
1da177e4 336}
1d957f9b 337EXPORT_SYMBOL(path_put);
1da177e4 338
19660af7 339/*
31e6b01f 340 * Path walking has 2 modes, rcu-walk and ref-walk (see
19660af7
AV		 341 * Documentation/filesystems/path-lookup.txt). In situations when we can't
342 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
343 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
344 * mode. Refcounts are grabbed at the last known good point before rcu-walk
345 * got stuck, so ref-walk may continue from there. If this is not successful
346 * (eg. a seqcount has changed), then failure is returned and it's up to caller
347 * to restart the path walk from the beginning in ref-walk mode.
31e6b01f 348 */
31e6b01f
NP		 349
350/**
19660af7
AV		 351 * unlazy_walk - try to switch to ref-walk mode.
352 * @nd: nameidata pathwalk data
353 * @dentry: child of nd->path.dentry or NULL
39191628 354 * Returns: 0 on success, -ECHILD on failure
31e6b01f 355 *
19660af7
AV		 356 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
357 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
358 * @nd or NULL. Must be called from rcu-walk context.
31e6b01f 359 */
19660af7 360static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
31e6b01f
NP		 361{
362 struct fs_struct *fs = current->fs;
363 struct dentry *parent = nd->path.dentry;
5b6ca027 364 int want_root = 0;
31e6b01f
NP		 365
366 BUG_ON(!(nd->flags & LOOKUP_RCU));
5b6ca027
AV		 367 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
368 want_root = 1;
31e6b01f
NP		 369 spin_lock(&fs->lock);
370 if (nd->root.mnt != fs->root.mnt ||
371 nd->root.dentry != fs->root.dentry)
372 goto err_root;
373 }
374 spin_lock(&parent->d_lock);
19660af7
AV		 375 if (!dentry) {
376 if (!__d_rcu_to_refcount(parent, nd->seq))
377 goto err_parent;
378 BUG_ON(nd->inode != parent->d_inode);
379 } else {
94c0d4ec
AV		 380 if (dentry->d_parent != parent)
381 goto err_parent;
19660af7
AV		 382 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
383 if (!__d_rcu_to_refcount(dentry, nd->seq))
384 goto err_child;
385 /*
386 * If the sequence check on the child dentry passed, then
387 * the child has not been removed from its parent. This
388 * means the parent dentry must be valid and able to take
389 * a reference at this point.
390 */
391 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
392 BUG_ON(!parent->d_count);
393 parent->d_count++;
394 spin_unlock(&dentry->d_lock);
395 }
31e6b01f 396 spin_unlock(&parent->d_lock);
5b6ca027 397 if (want_root) {
31e6b01f
NP		 398 path_get(&nd->root);
399 spin_unlock(&fs->lock);
400 }
401 mntget(nd->path.mnt);
402
403 rcu_read_unlock();
404 br_read_unlock(vfsmount_lock);
405 nd->flags &= ~LOOKUP_RCU;
406 return 0;
19660af7
AV		 407
408err_child:
31e6b01f 409 spin_unlock(&dentry->d_lock);
19660af7 410err_parent:
31e6b01f
NP		 411 spin_unlock(&parent->d_lock);
412err_root:
5b6ca027 413 if (want_root)
31e6b01f
NP		 414 spin_unlock(&fs->lock);
415 return -ECHILD;
416}
417
834f2a4a
TM		 418/**
419 * release_open_intent - free up open intent resources
420 * @nd: pointer to nameidata
421 */
422void release_open_intent(struct nameidata *nd)
423{
2dab5974
LT		 424 struct file *file = nd->intent.open.file;
425
426 if (file && !IS_ERR(file)) {
427 if (file->f_path.dentry == NULL)
428 put_filp(file);
429 else
430 fput(file);
431 }
834f2a4a
TM		 432}
433
f60aef7e 434static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
34286d66 435{
f60aef7e 436 return dentry->d_op->d_revalidate(dentry, nd);
34286d66
NP		 437}
438
9f1fafee
AV		 439/**
440 * complete_walk - successful completion of path walk
441 * @nd: pointer nameidata
39159de2 442 *
9f1fafee
AV		 443 * If we had been in RCU mode, drop out of it and legitimize nd->path.
444 * Revalidate the final result, unless we'd already done that during
445 * the path walk or the filesystem doesn't ask for it. Return 0 on
446 * success, -error on failure. In case of failure caller does not
447 * need to drop nd->path.
39159de2 448 */
9f1fafee 449static int complete_walk(struct nameidata *nd)
39159de2 450{
16c2cd71 451 struct dentry *dentry = nd->path.dentry;
39159de2 452 int status;
39159de2 453
9f1fafee
AV		 454 if (nd->flags & LOOKUP_RCU) {
455 nd->flags &= ~LOOKUP_RCU;
456 if (!(nd->flags & LOOKUP_ROOT))
457 nd->root.mnt = NULL;
458 spin_lock(&dentry->d_lock);
459 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
460 spin_unlock(&dentry->d_lock);
461 rcu_read_unlock();
462 br_read_unlock(vfsmount_lock);
463 return -ECHILD;
464 }
465 BUG_ON(nd->inode != dentry->d_inode);
466 spin_unlock(&dentry->d_lock);
467 mntget(nd->path.mnt);
468 rcu_read_unlock();
469 br_read_unlock(vfsmount_lock);
470 }
471
16c2cd71
AV		 472 if (likely(!(nd->flags & LOOKUP_JUMPED)))
473 return 0;
474
475 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
39159de2
JL		 476 return 0;
477
16c2cd71
AV		 478 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
479 return 0;
480
481 /* Note: we do not d_invalidate() */
34286d66 482 status = d_revalidate(dentry, nd);
39159de2
JL		 483 if (status > 0)
484 return 0;
485
16c2cd71 486 if (!status)
39159de2 487 status = -ESTALE;
16c2cd71 488
9f1fafee 489 path_put(&nd->path);
39159de2
JL		 490 return status;
491}
492
2a737871
AV		 493static __always_inline void set_root(struct nameidata *nd)
494{
f7ad3c6b
MS		 495 if (!nd->root.mnt)
496 get_fs_root(current->fs, &nd->root);
2a737871
AV		 497}
498
6de88d72
AV		 499static int link_path_walk(const char *, struct nameidata *);
500
31e6b01f
NP		 501static __always_inline void set_root_rcu(struct nameidata *nd)
502{
503 if (!nd->root.mnt) {
504 struct fs_struct *fs = current->fs;
c28cc364
NP		 505 unsigned seq;
506
507 do {
508 seq = read_seqcount_begin(&fs->seq);
509 nd->root = fs->root;
c1530019 510 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
c28cc364 511 } while (read_seqcount_retry(&fs->seq, seq));
31e6b01f
NP		 512 }
513}
514
f1662356 515static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
1da177e4 516{
31e6b01f
NP		 517 int ret;
518
1da177e4
LT		 519 if (IS_ERR(link))
520 goto fail;
521
522 if (*link == '/') {
2a737871 523 set_root(nd);
1d957f9b 524 path_put(&nd->path);
2a737871
AV		 525 nd->path = nd->root;
526 path_get(&nd->root);
16c2cd71 527 nd->flags |= LOOKUP_JUMPED;
1da177e4 528 }
31e6b01f 529 nd->inode = nd->path.dentry->d_inode;
b4091d5f 530
31e6b01f
NP		 531 ret = link_path_walk(link, nd);
532 return ret;
1da177e4 533fail:
1d957f9b 534 path_put(&nd->path);
1da177e4
LT		 535 return PTR_ERR(link);
536}
537
1d957f9b 538static void path_put_conditional(struct path *path, struct nameidata *nd)
051d3812
IK		 539{
540 dput(path->dentry);
4ac91378 541 if (path->mnt != nd->path.mnt)
051d3812
IK		 542 mntput(path->mnt);
543}
544
7b9337aa
NP		 545static inline void path_to_nameidata(const struct path *path,
546 struct nameidata *nd)
051d3812 547{
31e6b01f
NP		 548 if (!(nd->flags & LOOKUP_RCU)) {
549 dput(nd->path.dentry);
550 if (nd->path.mnt != path->mnt)
551 mntput(nd->path.mnt);
9a229683 552 }
31e6b01f 553 nd->path.mnt = path->mnt;
4ac91378 554 nd->path.dentry = path->dentry;
051d3812
IK		 555}
556
574197e0
AV		 557static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
558{
559 struct inode *inode = link->dentry->d_inode;
560 if (!IS_ERR(cookie) && inode->i_op->put_link)
561 inode->i_op->put_link(link->dentry, nd, cookie);
562 path_put(link);
563}
564
def4af30 565static __always_inline int
574197e0 566follow_link(struct path *link, struct nameidata *nd, void **p)
1da177e4
LT		 567{
568 int error;
7b9337aa 569 struct dentry *dentry = link->dentry;
1da177e4 570
844a3917
AV		 571 BUG_ON(nd->flags & LOOKUP_RCU);
572
0e794589
AV		 573 if (link->mnt == nd->path.mnt)
574 mntget(link->mnt);
575
574197e0
AV		 576 if (unlikely(current->total_link_count >= 40)) {
577 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
574197e0
AV		 578 path_put(&nd->path);
579 return -ELOOP;
580 }
581 cond_resched();
582 current->total_link_count++;
583
7b9337aa 584 touch_atime(link->mnt, dentry);
1da177e4 585 nd_set_link(nd, NULL);
cd4e91d3 586
36f3b4f6
AV		 587 error = security_inode_follow_link(link->dentry, nd);
588 if (error) {
589 *p = ERR_PTR(error); /* no ->put_link(), please */
590 path_put(&nd->path);
591 return error;
592 }
593
86acdca1 594 nd->last_type = LAST_BIND;
def4af30
AV		 595 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
596 error = PTR_ERR(*p);
597 if (!IS_ERR(*p)) {
1da177e4 598 char *s = nd_get_link(nd);
cc314eef 599 error = 0;
1da177e4
LT		 600 if (s)
601 error = __vfs_follow_link(nd, s);
bcda7652 602 else if (nd->last_type == LAST_BIND) {
16c2cd71 603 nd->flags |= LOOKUP_JUMPED;
b21041d0
AV		 604 nd->inode = nd->path.dentry->d_inode;
605 if (nd->inode->i_op->follow_link) {
bcda7652
AV		 606 /* stepped on a _really_ weird one */
607 path_put(&nd->path);
608 error = -ELOOP;
609 }
610 }
1da177e4 611 }
1da177e4
LT		 612 return error;
613}
614
31e6b01f
NP		 615static int follow_up_rcu(struct path *path)
616{
617 struct vfsmount *parent;
618 struct dentry *mountpoint;
619
620 parent = path->mnt->mnt_parent;
621 if (parent == path->mnt)
622 return 0;
623 mountpoint = path->mnt->mnt_mountpoint;
624 path->dentry = mountpoint;
625 path->mnt = parent;
626 return 1;
627}
628
bab77ebf 629int follow_up(struct path *path)
1da177e4
LT		 630{
631 struct vfsmount *parent;
632 struct dentry *mountpoint;
99b7db7b
NP		 633
634 br_read_lock(vfsmount_lock);
bab77ebf
AV		 635 parent = path->mnt->mnt_parent;
636 if (parent == path->mnt) {
99b7db7b 637 br_read_unlock(vfsmount_lock);
1da177e4
LT		 638 return 0;
639 }
640 mntget(parent);
bab77ebf 641 mountpoint = dget(path->mnt->mnt_mountpoint);
99b7db7b 642 br_read_unlock(vfsmount_lock);
bab77ebf
AV		 643 dput(path->dentry);
644 path->dentry = mountpoint;
645 mntput(path->mnt);
646 path->mnt = parent;
1da177e4
LT		 647 return 1;
648}
649
b5c84bf6 650/*
9875cf80
DH		 651 * Perform an automount
652 * - return -EISDIR to tell follow_managed() to stop and return the path we
653 * were called with.
1da177e4 654 */
9875cf80
DH		 655static int follow_automount(struct path *path, unsigned flags,
656 bool *need_mntput)
31e6b01f 657{
9875cf80 658 struct vfsmount *mnt;
ea5b778a 659 int err;
9875cf80
DH		 660
661 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
662 return -EREMOTE;
663
6f45b656
DH		 664 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
665 * and this is the terminal part of the path.
666 */
49084c3b 667 if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_PARENT))
6f45b656
DH		 668 return -EISDIR; /* we actually want to stop here */
669
9875cf80
DH		 670 /* We want to mount if someone is trying to open/create a file of any
671 * type under the mountpoint, wants to traverse through the mountpoint
672 * or wants to open the mounted directory.
673 *
674 * We don't want to mount if someone's just doing a stat and they've
675 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
676 * appended a '/' to the name.
677 */
678 if (!(flags & LOOKUP_FOLLOW) &&
49084c3b 679 !(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
9875cf80
DH		 680 LOOKUP_OPEN | LOOKUP_CREATE)))
681 return -EISDIR;
682
683 current->total_link_count++;
684 if (current->total_link_count >= 40)
685 return -ELOOP;
686
687 mnt = path->dentry->d_op->d_automount(path);
688 if (IS_ERR(mnt)) {
689 /*
690 * The filesystem is allowed to return -EISDIR here to indicate
691 * it doesn't want to automount. For instance, autofs would do
692 * this so that its userspace daemon can mount on this dentry.
693 *
694 * However, we can only permit this if it's a terminal point in
695 * the path being looked up; if it wasn't then the remainder of
696 * the path is inaccessible and we should say so.
697 */
49084c3b 698 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
9875cf80
DH		 699 return -EREMOTE;
700 return PTR_ERR(mnt);
31e6b01f 701 }
ea5b778a 702
9875cf80
DH		 703 if (!mnt) /* mount collision */
704 return 0;
31e6b01f 705
8aef1884
AV		 706 if (!*need_mntput) {
707 /* lock_mount() may release path->mnt on error */
708 mntget(path->mnt);
709 *need_mntput = true;
710 }
19a167af 711 err = finish_automount(mnt, path);
9875cf80 712
ea5b778a
DH		 713 switch (err) {
714 case -EBUSY:
715 /* Someone else made a mount here whilst we were busy */
19a167af 716 return 0;
ea5b778a 717 case 0:
8aef1884 718 path_put(path);
ea5b778a
DH		 719 path->mnt = mnt;
720 path->dentry = dget(mnt->mnt_root);
ea5b778a 721 return 0;
19a167af
AV		 722 default:
723 return err;
ea5b778a 724 }
19a167af 725
463ffb2e
AV		 726}
727
9875cf80
DH		 728/*
729 * Handle a dentry that is managed in some way.
cc53ce53 730 * - Flagged for transit management (autofs)
9875cf80
DH		 731 * - Flagged as mountpoint
732 * - Flagged as automount point
733 *
734 * This may only be called in refwalk mode.
735 *
736 * Serialization is taken care of in namespace.c
737 */
738static int follow_managed(struct path *path, unsigned flags)
1da177e4 739{
8aef1884 740 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
9875cf80
DH		 741 unsigned managed;
742 bool need_mntput = false;
8aef1884 743 int ret = 0;
9875cf80
DH		 744
745 /* Given that we're not holding a lock here, we retain the value in a
746 * local variable for each dentry as we look at it so that we don't see
747 * the components of that value change under us */
748 while (managed = ACCESS_ONCE(path->dentry->d_flags),
749 managed &= DCACHE_MANAGED_DENTRY,
750 unlikely(managed != 0)) {
cc53ce53
DH		 751 /* Allow the filesystem to manage the transit without i_mutex
752 * being held. */
753 if (managed & DCACHE_MANAGE_TRANSIT) {
754 BUG_ON(!path->dentry->d_op);
755 BUG_ON(!path->dentry->d_op->d_manage);
1aed3e42 756 ret = path->dentry->d_op->d_manage(path->dentry, false);
cc53ce53 757 if (ret < 0)
8aef1884 758 break;
cc53ce53
DH		 759 }
760
9875cf80
DH		 761 /* Transit to a mounted filesystem. */
762 if (managed & DCACHE_MOUNTED) {
763 struct vfsmount *mounted = lookup_mnt(path);
764 if (mounted) {
765 dput(path->dentry);
766 if (need_mntput)
767 mntput(path->mnt);
768 path->mnt = mounted;
769 path->dentry = dget(mounted->mnt_root);
770 need_mntput = true;
771 continue;
772 }
773
774 /* Something is mounted on this dentry in another
775 * namespace and/or whatever was mounted there in this
776 * namespace got unmounted before we managed to get the
777 * vfsmount_lock */
778 }
779
780 /* Handle an automount point */
781 if (managed & DCACHE_NEED_AUTOMOUNT) {
782 ret = follow_automount(path, flags, &need_mntput);
783 if (ret < 0)
8aef1884 784 break;
9875cf80
DH		 785 continue;
786 }
787
788 /* We didn't change the current path point */
789 break;
1da177e4 790 }
8aef1884
AV		 791
792 if (need_mntput && path->mnt == mnt)
793 mntput(path->mnt);
794 if (ret == -EISDIR)
795 ret = 0;
796 return ret;
1da177e4
LT		 797}
798
cc53ce53 799int follow_down_one(struct path *path)
1da177e4
LT		 800{
801 struct vfsmount *mounted;
802
1c755af4 803 mounted = lookup_mnt(path);
1da177e4 804 if (mounted) {
9393bd07
AV		 805 dput(path->dentry);
806 mntput(path->mnt);
807 path->mnt = mounted;
808 path->dentry = dget(mounted->mnt_root);
1da177e4
LT		 809 return 1;
810 }
811 return 0;
812}
813
62a7375e
IK		 814static inline bool managed_dentry_might_block(struct dentry *dentry)
815{
816 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
817 dentry->d_op->d_manage(dentry, true) < 0);
818}
819
9875cf80 820/*
287548e4
AV		 821 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
822 * we meet a managed dentry that would need blocking.
9875cf80
DH		 823 */
824static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
287548e4 825 struct inode **inode)
9875cf80 826{
62a7375e 827 for (;;) {
9875cf80 828 struct vfsmount *mounted;
62a7375e
IK		 829 /*
830 * Don't forget we might have a non-mountpoint managed dentry
831 * that wants to block transit.
832 */
287548e4 833 if (unlikely(managed_dentry_might_block(path->dentry)))
ab90911f 834 return false;
62a7375e
IK		 835
836 if (!d_mountpoint(path->dentry))
837 break;
838
9875cf80
DH		 839 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
840 if (!mounted)
841 break;
842 path->mnt = mounted;
843 path->dentry = mounted->mnt_root;
844 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
59430262
LT		 845 /*
846 * Update the inode too. We don't need to re-check the
847 * dentry sequence number here after this d_inode read,
848 * because a mount-point is always pinned.
849 */
850 *inode = path->dentry->d_inode;
9875cf80 851 }
9875cf80
DH		 852 return true;
853}
854
dea39376 855static void follow_mount_rcu(struct nameidata *nd)
287548e4 856{
dea39376 857 while (d_mountpoint(nd->path.dentry)) {
287548e4 858 struct vfsmount *mounted;
dea39376 859 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
287548e4
AV		 860 if (!mounted)
861 break;
dea39376
AV		 862 nd->path.mnt = mounted;
863 nd->path.dentry = mounted->mnt_root;
864 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
287548e4
AV		 865 }
866}
867
31e6b01f
NP		 868static int follow_dotdot_rcu(struct nameidata *nd)
869{
31e6b01f
NP		 870 set_root_rcu(nd);
871
9875cf80 872 while (1) {
31e6b01f
NP		 873 if (nd->path.dentry == nd->root.dentry &&
874 nd->path.mnt == nd->root.mnt) {
875 break;
876 }
877 if (nd->path.dentry != nd->path.mnt->mnt_root) {
878 struct dentry *old = nd->path.dentry;
879 struct dentry *parent = old->d_parent;
880 unsigned seq;
881
882 seq = read_seqcount_begin(&parent->d_seq);
883 if (read_seqcount_retry(&old->d_seq, nd->seq))
ef7562d5 884 goto failed;
31e6b01f
NP		 885 nd->path.dentry = parent;
886 nd->seq = seq;
887 break;
888 }
889 if (!follow_up_rcu(&nd->path))
890 break;
891 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
31e6b01f 892 }
dea39376
AV		 893 follow_mount_rcu(nd);
894 nd->inode = nd->path.dentry->d_inode;
31e6b01f 895 return 0;
ef7562d5
AV		 896
897failed:
898 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 899 if (!(nd->flags & LOOKUP_ROOT))
900 nd->root.mnt = NULL;
ef7562d5
AV		 901 rcu_read_unlock();
902 br_read_unlock(vfsmount_lock);
903 return -ECHILD;
31e6b01f
NP		 904}
905
cc53ce53
DH		 906/*
907 * Follow down to the covering mount currently visible to userspace. At each
908 * point, the filesystem owning that dentry may be queried as to whether the
909 * caller is permitted to proceed or not.
cc53ce53 910 */
7cc90cc3 911int follow_down(struct path *path)
cc53ce53
DH		 912{
913 unsigned managed;
914 int ret;
915
916 while (managed = ACCESS_ONCE(path->dentry->d_flags),
917 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
918 /* Allow the filesystem to manage the transit without i_mutex
919 * being held.
920 *
921 * We indicate to the filesystem if someone is trying to mount
922 * something here. This gives autofs the chance to deny anyone
923 * other than its daemon the right to mount on its
924 * superstructure.
925 *
926 * The filesystem may sleep at this point.
927 */
928 if (managed & DCACHE_MANAGE_TRANSIT) {
929 BUG_ON(!path->dentry->d_op);
930 BUG_ON(!path->dentry->d_op->d_manage);
ab90911f 931 ret = path->dentry->d_op->d_manage(
1aed3e42 932 path->dentry, false);
cc53ce53
DH		 933 if (ret < 0)
934 return ret == -EISDIR ? 0 : ret;
935 }
936
937 /* Transit to a mounted filesystem. */
938 if (managed & DCACHE_MOUNTED) {
939 struct vfsmount *mounted = lookup_mnt(path);
940 if (!mounted)
941 break;
942 dput(path->dentry);
943 mntput(path->mnt);
944 path->mnt = mounted;
945 path->dentry = dget(mounted->mnt_root);
946 continue;
947 }
948
949 /* Don't handle automount points here */
950 break;
951 }
952 return 0;
953}
954
9875cf80
DH		 955/*
956 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
957 */
958static void follow_mount(struct path *path)
959{
960 while (d_mountpoint(path->dentry)) {
961 struct vfsmount *mounted = lookup_mnt(path);
962 if (!mounted)
963 break;
964 dput(path->dentry);
965 mntput(path->mnt);
966 path->mnt = mounted;
967 path->dentry = dget(mounted->mnt_root);
968 }
969}
970
31e6b01f 971static void follow_dotdot(struct nameidata *nd)
1da177e4 972{
2a737871 973 set_root(nd);
e518ddb7 974
1da177e4 975 while(1) {
4ac91378 976 struct dentry *old = nd->path.dentry;
1da177e4 977
2a737871
AV		 978 if (nd->path.dentry == nd->root.dentry &&
979 nd->path.mnt == nd->root.mnt) {
1da177e4
LT		 980 break;
981 }
4ac91378 982 if (nd->path.dentry != nd->path.mnt->mnt_root) {
3088dd70
AV		 983 /* rare case of legitimate dget_parent()... */
984 nd->path.dentry = dget_parent(nd->path.dentry);
1da177e4
LT		 985 dput(old);
986 break;
987 }
3088dd70 988 if (!follow_up(&nd->path))
1da177e4 989 break;
1da177e4 990 }
79ed0226 991 follow_mount(&nd->path);
31e6b01f 992 nd->inode = nd->path.dentry->d_inode;
1da177e4
LT		 993}
994
baa03890
NP		 995/*
996 * Allocate a dentry with name and parent, and perform a parent
997 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
998 * on error. parent->d_inode->i_mutex must be held. d_lookup must
999 * have verified that no child exists while under i_mutex.
1000 */
1001static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1002 struct qstr *name, struct nameidata *nd)
1003{
1004 struct inode *inode = parent->d_inode;
1005 struct dentry *dentry;
1006 struct dentry *old;
1007
1008 /* Don't create child dentry for a dead directory. */
1009 if (unlikely(IS_DEADDIR(inode)))
1010 return ERR_PTR(-ENOENT);
1011
1012 dentry = d_alloc(parent, name);
1013 if (unlikely(!dentry))
1014 return ERR_PTR(-ENOMEM);
1015
1016 old = inode->i_op->lookup(inode, dentry, nd);
1017 if (unlikely(old)) {
1018 dput(dentry);
1019 dentry = old;
1020 }
1021 return dentry;
1022}
1023
44396f4b
JB		 1024/*
1025 * We already have a dentry, but require a lookup to be performed on the parent
1026 * directory to fill in d_inode. Returns the new dentry, or ERR_PTR on error.
1027 * parent->d_inode->i_mutex must be held. d_lookup must have verified that no
1028 * child exists while under i_mutex.
1029 */
1030static struct dentry *d_inode_lookup(struct dentry *parent, struct dentry *dentry,
1031 struct nameidata *nd)
1032{
1033 struct inode *inode = parent->d_inode;
1034 struct dentry *old;
1035
1036 /* Don't create child dentry for a dead directory. */
1037 if (unlikely(IS_DEADDIR(inode)))
1038 return ERR_PTR(-ENOENT);
1039
1040 old = inode->i_op->lookup(inode, dentry, nd);
1041 if (unlikely(old)) {
1042 dput(dentry);
1043 dentry = old;
1044 }
1045 return dentry;
1046}
1047
1da177e4
LT		 1048/*
1049 * It's more convoluted than I'd like it to be, but... it's still fairly
1050 * small and for now I'd prefer to have fast path as straight as possible.
1051 * It _is_ time-critical.
1052 */
1053static int do_lookup(struct nameidata *nd, struct qstr *name,
31e6b01f 1054 struct path *path, struct inode **inode)
1da177e4 1055{
4ac91378 1056 struct vfsmount *mnt = nd->path.mnt;
31e6b01f 1057 struct dentry *dentry, *parent = nd->path.dentry;
5a18fff2
AV		 1058 int need_reval = 1;
1059 int status = 1;
9875cf80
DH		 1060 int err;
1061
b04f784e
NP		 1062 /*
1063 * Rename seqlock is not required here because in the off chance
1064 * of a false negative due to a concurrent rename, we're going to
1065 * do the non-racy lookup, below.
1066 */
31e6b01f
NP		 1067 if (nd->flags & LOOKUP_RCU) {
1068 unsigned seq;
31e6b01f
NP		 1069 *inode = nd->inode;
1070 dentry = __d_lookup_rcu(parent, name, &seq, inode);
5a18fff2
AV		 1071 if (!dentry)
1072 goto unlazy;
1073
31e6b01f
NP		 1074 /* Memory barrier in read_seqcount_begin of child is enough */
1075 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1076 return -ECHILD;
31e6b01f 1077 nd->seq = seq;
5a18fff2 1078
24643087 1079 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
5a18fff2
AV		 1080 status = d_revalidate(dentry, nd);
1081 if (unlikely(status <= 0)) {
1082 if (status != -ECHILD)
1083 need_reval = 0;
1084 goto unlazy;
1085 }
24643087 1086 }
44396f4b
JB		 1087 if (unlikely(d_need_lookup(dentry)))
1088 goto unlazy;
31e6b01f
NP		 1089 path->mnt = mnt;
1090 path->dentry = dentry;
d6e9bd25
AV		 1091 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1092 goto unlazy;
1093 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1094 goto unlazy;
1095 return 0;
5a18fff2 1096unlazy:
19660af7
AV		 1097 if (unlazy_walk(nd, dentry))
1098 return -ECHILD;
5a18fff2
AV		 1099 } else {
1100 dentry = __d_lookup(parent, name);
9875cf80 1101 }
5a18fff2 1102
44396f4b
JB		 1103 if (dentry && unlikely(d_need_lookup(dentry))) {
1104 dput(dentry);
1105 dentry = NULL;
1106 }
5a18fff2
AV		 1107retry:
1108 if (unlikely(!dentry)) {
1109 struct inode *dir = parent->d_inode;
1110 BUG_ON(nd->inode != dir);
1111
1112 mutex_lock(&dir->i_mutex);
1113 dentry = d_lookup(parent, name);
1114 if (likely(!dentry)) {
1115 dentry = d_alloc_and_lookup(parent, name, nd);
1116 if (IS_ERR(dentry)) {
1117 mutex_unlock(&dir->i_mutex);
1118 return PTR_ERR(dentry);
1119 }
1120 /* known good */
1121 need_reval = 0;
1122 status = 1;
44396f4b
JB		 1123 } else if (unlikely(d_need_lookup(dentry))) {
1124 dentry = d_inode_lookup(parent, dentry, nd);
1125 if (IS_ERR(dentry)) {
1126 mutex_unlock(&dir->i_mutex);
1127 return PTR_ERR(dentry);
1128 }
1129 /* known good */
1130 need_reval = 0;
1131 status = 1;
5a18fff2
AV		 1132 }
1133 mutex_unlock(&dir->i_mutex);
1134 }
1135 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1136 status = d_revalidate(dentry, nd);
1137 if (unlikely(status <= 0)) {
1138 if (status < 0) {
1139 dput(dentry);
1140 return status;
1141 }
1142 if (!d_invalidate(dentry)) {
1143 dput(dentry);
1144 dentry = NULL;
1145 need_reval = 1;
1146 goto retry;
1147 }
24643087 1148 }
5a18fff2 1149
9875cf80
DH		 1150 path->mnt = mnt;
1151 path->dentry = dentry;
1152 err = follow_managed(path, nd->flags);
89312214
IK		 1153 if (unlikely(err < 0)) {
1154 path_put_conditional(path, nd);
9875cf80 1155 return err;
89312214 1156 }
9875cf80 1157 *inode = path->dentry->d_inode;
1da177e4 1158 return 0;
1da177e4
LT		 1159}
1160
52094c8a
AV		 1161static inline int may_lookup(struct nameidata *nd)
1162{
1163 if (nd->flags & LOOKUP_RCU) {
4ad5abb3 1164 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
52094c8a
AV		 1165 if (err != -ECHILD)
1166 return err;
19660af7 1167 if (unlazy_walk(nd, NULL))
52094c8a
AV		 1168 return -ECHILD;
1169 }
4ad5abb3 1170 return inode_permission(nd->inode, MAY_EXEC);
52094c8a
AV		 1171}
1172
9856fa1b
AV		 1173static inline int handle_dots(struct nameidata *nd, int type)
1174{
1175 if (type == LAST_DOTDOT) {
1176 if (nd->flags & LOOKUP_RCU) {
1177 if (follow_dotdot_rcu(nd))
1178 return -ECHILD;
1179 } else
1180 follow_dotdot(nd);
1181 }
1182 return 0;
1183}
1184
951361f9
AV		 1185static void terminate_walk(struct nameidata *nd)
1186{
1187 if (!(nd->flags & LOOKUP_RCU)) {
1188 path_put(&nd->path);
1189 } else {
1190 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 1191 if (!(nd->flags & LOOKUP_ROOT))
1192 nd->root.mnt = NULL;
951361f9
AV		 1193 rcu_read_unlock();
1194 br_read_unlock(vfsmount_lock);
1195 }
1196}
1197
ce57dfc1
AV		 1198static inline int walk_component(struct nameidata *nd, struct path *path,
1199 struct qstr *name, int type, int follow)
1200{
1201 struct inode *inode;
1202 int err;
1203 /*
1204 * "." and ".." are special - ".." especially so because it has
1205 * to be able to know about the current root directory and
1206 * parent relationships.
1207 */
1208 if (unlikely(type != LAST_NORM))
1209 return handle_dots(nd, type);
1210 err = do_lookup(nd, name, path, &inode);
1211 if (unlikely(err)) {
1212 terminate_walk(nd);
1213 return err;
1214 }
1215 if (!inode) {
1216 path_to_nameidata(path, nd);
1217 terminate_walk(nd);
1218 return -ENOENT;
1219 }
1220 if (unlikely(inode->i_op->follow_link) && follow) {
19660af7
AV		 1221 if (nd->flags & LOOKUP_RCU) {
1222 if (unlikely(unlazy_walk(nd, path->dentry))) {
1223 terminate_walk(nd);
1224 return -ECHILD;
1225 }
1226 }
ce57dfc1
AV		 1227 BUG_ON(inode != path->dentry->d_inode);
1228 return 1;
1229 }
1230 path_to_nameidata(path, nd);
1231 nd->inode = inode;
1232 return 0;
1233}
1234
b356379a
AV		 1235/*
1236 * This limits recursive symlink follows to 8, while
1237 * limiting consecutive symlinks to 40.
1238 *
1239 * Without that kind of total limit, nasty chains of consecutive
1240 * symlinks can cause almost arbitrarily long lookups.
1241 */
1242static inline int nested_symlink(struct path *path, struct nameidata *nd)
1243{
1244 int res;
1245
b356379a
AV		 1246 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1247 path_put_conditional(path, nd);
1248 path_put(&nd->path);
1249 return -ELOOP;
1250 }
1a4022f8 1251 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
b356379a
AV		 1252
1253 nd->depth++;
1254 current->link_count++;
1255
1256 do {
1257 struct path link = *path;
1258 void *cookie;
574197e0
AV		 1259
1260 res = follow_link(&link, nd, &cookie);
b356379a
AV		 1261 if (!res)
1262 res = walk_component(nd, path, &nd->last,
1263 nd->last_type, LOOKUP_FOLLOW);
574197e0 1264 put_link(nd, &link, cookie);
b356379a
AV		 1265 } while (res > 0);
1266
1267 current->link_count--;
1268 nd->depth--;
1269 return res;
1270}
1271
1da177e4
LT		 1272/*
1273 * Name resolution.
ea3834d9
PM		 1274 * This is the basic name resolution function, turning a pathname into
1275 * the final dentry. We expect 'base' to be positive and a directory.
1da177e4 1276 *
ea3834d9
PM		 1277 * Returns 0 and nd will have valid dentry and mnt on success.
1278 * Returns error and drops reference to input namei data on failure.
1da177e4 1279 */
6de88d72 1280static int link_path_walk(const char *name, struct nameidata *nd)
1da177e4
LT		 1281{
1282 struct path next;
1da177e4 1283 int err;
1da177e4
LT		 1284
1285 while (*name=='/')
1286 name++;
1287 if (!*name)
086e183a 1288 return 0;
1da177e4 1289
1da177e4
LT		 1290 /* At this point we know we have a real path component. */
1291 for(;;) {
1292 unsigned long hash;
1293 struct qstr this;
1294 unsigned int c;
fe479a58 1295 int type;
1da177e4 1296
52094c8a 1297 err = may_lookup(nd);
1da177e4
LT		 1298 if (err)
1299 break;
1300
1301 this.name = name;
1302 c = *(const unsigned char *)name;
1303
1304 hash = init_name_hash();
1305 do {
1306 name++;
1307 hash = partial_name_hash(c, hash);
1308 c = *(const unsigned char *)name;
1309 } while (c && (c != '/'));
1310 this.len = name - (const char *) this.name;
1311 this.hash = end_name_hash(hash);
1312
fe479a58
AV		 1313 type = LAST_NORM;
1314 if (this.name[0] == '.') switch (this.len) {
1315 case 2:
16c2cd71 1316 if (this.name[1] == '.') {
fe479a58 1317 type = LAST_DOTDOT;
16c2cd71
AV		 1318 nd->flags |= LOOKUP_JUMPED;
1319 }
fe479a58
AV		 1320 break;
1321 case 1:
1322 type = LAST_DOT;
1323 }
5a202bcd
AV		 1324 if (likely(type == LAST_NORM)) {
1325 struct dentry *parent = nd->path.dentry;
16c2cd71 1326 nd->flags &= ~LOOKUP_JUMPED;
5a202bcd
AV		 1327 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1328 err = parent->d_op->d_hash(parent, nd->inode,
1329 &this);
1330 if (err < 0)
1331 break;
1332 }
1333 }
fe479a58 1334
1da177e4
LT		 1335 /* remove trailing slashes? */
1336 if (!c)
1337 goto last_component;
1338 while (*++name == '/');
1339 if (!*name)
b356379a 1340 goto last_component;
1da177e4 1341
ce57dfc1
AV		 1342 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1343 if (err < 0)
1344 return err;
1da177e4 1345
ce57dfc1 1346 if (err) {
b356379a 1347 err = nested_symlink(&next, nd);
1da177e4 1348 if (err)
a7472bab 1349 return err;
31e6b01f 1350 }
1da177e4 1351 err = -ENOTDIR;
31e6b01f 1352 if (!nd->inode->i_op->lookup)
1da177e4
LT		 1353 break;
1354 continue;
1355 /* here ends the main loop */
1356
1da177e4 1357last_component:
b356379a
AV		 1358 nd->last = this;
1359 nd->last_type = type;
086e183a 1360 return 0;
1da177e4 1361 }
951361f9 1362 terminate_walk(nd);
1da177e4
LT		 1363 return err;
1364}
1365
70e9b357
AV		 1366static int path_init(int dfd, const char *name, unsigned int flags,
1367 struct nameidata *nd, struct file **fp)
31e6b01f
NP		 1368{
1369 int retval = 0;
1370 int fput_needed;
1371 struct file *file;
1372
1373 nd->last_type = LAST_ROOT; /* if there are only slashes... */
16c2cd71 1374 nd->flags = flags | LOOKUP_JUMPED;
31e6b01f 1375 nd->depth = 0;
5b6ca027
AV		 1376 if (flags & LOOKUP_ROOT) {
1377 struct inode *inode = nd->root.dentry->d_inode;
73d049a4
AV		 1378 if (*name) {
1379 if (!inode->i_op->lookup)
1380 return -ENOTDIR;
1381 retval = inode_permission(inode, MAY_EXEC);
1382 if (retval)
1383 return retval;
1384 }
5b6ca027
AV		 1385 nd->path = nd->root;
1386 nd->inode = inode;
1387 if (flags & LOOKUP_RCU) {
1388 br_read_lock(vfsmount_lock);
1389 rcu_read_lock();
1390 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1391 } else {
1392 path_get(&nd->path);
1393 }
1394 return 0;
1395 }
1396
31e6b01f 1397 nd->root.mnt = NULL;
31e6b01f
NP		 1398
1399 if (*name=='/') {
e41f7d4e
AV		 1400 if (flags & LOOKUP_RCU) {
1401 br_read_lock(vfsmount_lock);
1402 rcu_read_lock();
1403 set_root_rcu(nd);
1404 } else {
1405 set_root(nd);
1406 path_get(&nd->root);
1407 }
1408 nd->path = nd->root;
31e6b01f 1409 } else if (dfd == AT_FDCWD) {
e41f7d4e
AV		 1410 if (flags & LOOKUP_RCU) {
1411 struct fs_struct *fs = current->fs;
1412 unsigned seq;
31e6b01f 1413
e41f7d4e
AV		 1414 br_read_lock(vfsmount_lock);
1415 rcu_read_lock();
c28cc364 1416
e41f7d4e
AV		 1417 do {
1418 seq = read_seqcount_begin(&fs->seq);
1419 nd->path = fs->pwd;
1420 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1421 } while (read_seqcount_retry(&fs->seq, seq));
1422 } else {
1423 get_fs_pwd(current->fs, &nd->path);
1424 }
31e6b01f
NP		 1425 } else {
1426 struct dentry *dentry;
1427
1abf0c71 1428 file = fget_raw_light(dfd, &fput_needed);
31e6b01f
NP		 1429 retval = -EBADF;
1430 if (!file)
1431 goto out_fail;
1432
1433 dentry = file->f_path.dentry;
1434
f52e0c11
AV		 1435 if (*name) {
1436 retval = -ENOTDIR;
1437 if (!S_ISDIR(dentry->d_inode->i_mode))
1438 goto fput_fail;
31e6b01f 1439
4ad5abb3 1440 retval = inode_permission(dentry->d_inode, MAY_EXEC);
f52e0c11
AV		 1441 if (retval)
1442 goto fput_fail;
1443 }
31e6b01f
NP		 1444
1445 nd->path = file->f_path;
e41f7d4e
AV		 1446 if (flags & LOOKUP_RCU) {
1447 if (fput_needed)
70e9b357 1448 *fp = file;
e41f7d4e
AV		 1449 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1450 br_read_lock(vfsmount_lock);
1451 rcu_read_lock();
1452 } else {
1453 path_get(&file->f_path);
1454 fput_light(file, fput_needed);
1455 }
31e6b01f 1456 }
31e6b01f 1457
31e6b01f 1458 nd->inode = nd->path.dentry->d_inode;
9b4a9b14 1459 return 0;
2dfdd266 1460
9b4a9b14
AV		 1461fput_fail:
1462 fput_light(file, fput_needed);
1463out_fail:
1464 return retval;
1465}
1466
bd92d7fe
AV		 1467static inline int lookup_last(struct nameidata *nd, struct path *path)
1468{
1469 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1470 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1471
1472 nd->flags &= ~LOOKUP_PARENT;
1473 return walk_component(nd, path, &nd->last, nd->last_type,
1474 nd->flags & LOOKUP_FOLLOW);
1475}
1476
9b4a9b14 1477/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
ee0827cd 1478static int path_lookupat(int dfd, const char *name,
9b4a9b14
AV		 1479 unsigned int flags, struct nameidata *nd)
1480{
70e9b357 1481 struct file *base = NULL;
bd92d7fe
AV		 1482 struct path path;
1483 int err;
31e6b01f
NP		 1484
1485 /*
1486 * Path walking is largely split up into 2 different synchronisation
1487 * schemes, rcu-walk and ref-walk (explained in
1488 * Documentation/filesystems/path-lookup.txt). These share much of the
1489 * path walk code, but some things particularly setup, cleanup, and
1490 * following mounts are sufficiently divergent that functions are
1491 * duplicated. Typically there is a function foo(), and its RCU
1492 * analogue, foo_rcu().
1493 *
1494 * -ECHILD is the error number of choice (just to avoid clashes) that
1495 * is returned if some aspect of an rcu-walk fails. Such an error must
1496 * be handled by restarting a traditional ref-walk (which will always
1497 * be able to complete).
1498 */
bd92d7fe 1499 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
ee0827cd 1500
bd92d7fe
AV		 1501 if (unlikely(err))
1502 return err;
ee0827cd
AV		 1503
1504 current->total_link_count = 0;
bd92d7fe
AV		 1505 err = link_path_walk(name, nd);
1506
1507 if (!err && !(flags & LOOKUP_PARENT)) {
bd92d7fe
AV		 1508 err = lookup_last(nd, &path);
1509 while (err > 0) {
1510 void *cookie;
1511 struct path link = path;
bd92d7fe 1512 nd->flags |= LOOKUP_PARENT;
574197e0 1513 err = follow_link(&link, nd, &cookie);
bd92d7fe
AV		 1514 if (!err)
1515 err = lookup_last(nd, &path);
574197e0 1516 put_link(nd, &link, cookie);
bd92d7fe
AV		 1517 }
1518 }
ee0827cd 1519
9f1fafee
AV		 1520 if (!err)
1521 err = complete_walk(nd);
bd92d7fe
AV		 1522
1523 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1524 if (!nd->inode->i_op->lookup) {
1525 path_put(&nd->path);
bd23a539 1526 err = -ENOTDIR;
bd92d7fe
AV		 1527 }
1528 }
16c2cd71 1529
70e9b357
AV		 1530 if (base)
1531 fput(base);
ee0827cd 1532
5b6ca027 1533 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
2a737871
AV		 1534 path_put(&nd->root);
1535 nd->root.mnt = NULL;
1536 }
bd92d7fe 1537 return err;
ee0827cd 1538}
31e6b01f 1539
ee0827cd
AV		 1540static int do_path_lookup(int dfd, const char *name,
1541 unsigned int flags, struct nameidata *nd)
1542{
1543 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1544 if (unlikely(retval == -ECHILD))
1545 retval = path_lookupat(dfd, name, flags, nd);
1546 if (unlikely(retval == -ESTALE))
1547 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
31e6b01f
NP		 1548
1549 if (likely(!retval)) {
1550 if (unlikely(!audit_dummy_context())) {
1551 if (nd->path.dentry && nd->inode)
1552 audit_inode(name, nd->path.dentry);
1553 }
1554 }
170aa3d0 1555 return retval;
1da177e4
LT		 1556}
1557
c9c6cac0 1558int kern_path_parent(const char *name, struct nameidata *nd)
5590ff0d 1559{
c9c6cac0 1560 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
5590ff0d
UD		 1561}
1562
d1811465
AV		 1563int kern_path(const char *name, unsigned int flags, struct path *path)
1564{
1565 struct nameidata nd;
1566 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1567 if (!res)
1568 *path = nd.path;
1569 return res;
1570}
1571
16f18200
JJS		 1572/**
1573 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1574 * @dentry: pointer to dentry of the base directory
1575 * @mnt: pointer to vfs mount of the base directory
1576 * @name: pointer to file name
1577 * @flags: lookup flags
1578 * @nd: pointer to nameidata
1579 */
1580int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1581 const char *name, unsigned int flags,
1582 struct nameidata *nd)
1583{
5b6ca027
AV		 1584 nd->root.dentry = dentry;
1585 nd->root.mnt = mnt;
1586 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1587 return do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, nd);
16f18200
JJS		 1588}
1589
eead1911
CH		 1590static struct dentry *__lookup_hash(struct qstr *name,
1591 struct dentry *base, struct nameidata *nd)
1da177e4 1592{
81fca444 1593 struct inode *inode = base->d_inode;
057f6c01 1594 struct dentry *dentry;
1da177e4
LT		 1595 int err;
1596
4ad5abb3 1597 err = inode_permission(inode, MAY_EXEC);
81fca444
CH		 1598 if (err)
1599 return ERR_PTR(err);
1da177e4 1600
b04f784e
NP		 1601 /*
1602 * Don't bother with __d_lookup: callers are for creat as
1603 * well as unlink, so a lot of the time it would cost
1604 * a double lookup.
6e6b1bd1 1605 */
b04f784e 1606 dentry = d_lookup(base, name);
6e6b1bd1 1607
44396f4b
JB		 1608 if (dentry && d_need_lookup(dentry)) {
1609 /*
1610 * __lookup_hash is called with the parent dir's i_mutex already
1611 * held, so we are good to go here.
1612 */
1613 dentry = d_inode_lookup(base, dentry, nd);
1614 if (IS_ERR(dentry))
1615 return dentry;
1616 }
1617
d2d9e9fb
AV		 1618 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1619 int status = d_revalidate(dentry, nd);
1620 if (unlikely(status <= 0)) {
1621 /*
1622 * The dentry failed validation.
1623 * If d_revalidate returned 0 attempt to invalidate
1624 * the dentry otherwise d_revalidate is asking us
1625 * to return a fail status.
1626 */
1627 if (status < 0) {
1628 dput(dentry);
1629 return ERR_PTR(status);
1630 } else if (!d_invalidate(dentry)) {
1631 dput(dentry);
1632 dentry = NULL;
1633 }
1634 }
1635 }
6e6b1bd1 1636
baa03890
NP		 1637 if (!dentry)
1638 dentry = d_alloc_and_lookup(base, name, nd);
5a202bcd 1639
1da177e4
LT		 1640 return dentry;
1641}
1642
057f6c01
JM		 1643/*
1644 * Restricted form of lookup. Doesn't follow links, single-component only,
1645 * needs parent already locked. Doesn't follow mounts.
1646 * SMP-safe.
1647 */
eead1911 1648static struct dentry *lookup_hash(struct nameidata *nd)
057f6c01 1649{
4ac91378 1650 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1da177e4
LT		 1651}
1652
eead1911 1653/**
a6b91919 1654 * lookup_one_len - filesystem helper to lookup single pathname component
eead1911
CH		 1655 * @name: pathname component to lookup
1656 * @base: base directory to lookup from
1657 * @len: maximum length @len should be interpreted to
1658 *
a6b91919
RD		 1659 * Note that this routine is purely a helper for filesystem usage and should
1660 * not be called by generic code. Also note that by using this function the
eead1911
CH		 1661 * nameidata argument is passed to the filesystem methods and a filesystem
1662 * using this helper needs to be prepared for that.
1663 */
057f6c01
JM		 1664struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1665{
057f6c01 1666 struct qstr this;
6a96ba54
AV		 1667 unsigned long hash;
1668 unsigned int c;
057f6c01 1669
2f9092e1
DW		 1670 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1671
6a96ba54
AV		 1672 this.name = name;
1673 this.len = len;
1674 if (!len)
1675 return ERR_PTR(-EACCES);
1676
1677 hash = init_name_hash();
1678 while (len--) {
1679 c = *(const unsigned char *)name++;
1680 if (c == '/' || c == '\0')
1681 return ERR_PTR(-EACCES);
1682 hash = partial_name_hash(c, hash);
1683 }
1684 this.hash = end_name_hash(hash);
5a202bcd
AV		 1685 /*
1686 * See if the low-level filesystem might want
1687 * to use its own hash..
1688 */
1689 if (base->d_flags & DCACHE_OP_HASH) {
1690 int err = base->d_op->d_hash(base, base->d_inode, &this);
1691 if (err < 0)
1692 return ERR_PTR(err);
1693 }
eead1911 1694
49705b77 1695 return __lookup_hash(&this, base, NULL);
057f6c01
JM		 1696}
1697
2d8f3038
AV		 1698int user_path_at(int dfd, const char __user *name, unsigned flags,
1699 struct path *path)
1da177e4 1700{
2d8f3038 1701 struct nameidata nd;
f52e0c11 1702 char *tmp = getname_flags(name, flags);
1da177e4 1703 int err = PTR_ERR(tmp);
1da177e4 1704 if (!IS_ERR(tmp)) {
2d8f3038
AV		 1705
1706 BUG_ON(flags & LOOKUP_PARENT);
1707
1708 err = do_path_lookup(dfd, tmp, flags, &nd);
1da177e4 1709 putname(tmp);
2d8f3038
AV		 1710 if (!err)
1711 *path = nd.path;
1da177e4
LT		 1712 }
1713 return err;
1714}
1715
2ad94ae6
AV		 1716static int user_path_parent(int dfd, const char __user *path,
1717 struct nameidata *nd, char **name)
1718{
1719 char *s = getname(path);
1720 int error;
1721
1722 if (IS_ERR(s))
1723 return PTR_ERR(s);
1724
1725 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1726 if (error)
1727 putname(s);
1728 else
1729 *name = s;
1730
1731 return error;
1732}
1733
1da177e4
LT		 1734/*
1735 * It's inline, so penalty for filesystems that don't use sticky bit is
1736 * minimal.
1737 */
1738static inline int check_sticky(struct inode *dir, struct inode *inode)
1739{
da9592ed
DH		 1740 uid_t fsuid = current_fsuid();
1741
1da177e4
LT		 1742 if (!(dir->i_mode & S_ISVTX))
1743 return 0;
e795b717
SH		 1744 if (current_user_ns() != inode_userns(inode))
1745 goto other_userns;
da9592ed 1746 if (inode->i_uid == fsuid)
1da177e4 1747 return 0;
da9592ed 1748 if (dir->i_uid == fsuid)
1da177e4 1749 return 0;
e795b717
SH		 1750
1751other_userns:
1752 return !ns_capable(inode_userns(inode), CAP_FOWNER);
1da177e4
LT		 1753}
1754
1755/*
1756 * Check whether we can remove a link victim from directory dir, check
1757 * whether the type of victim is right.
1758 * 1. We can't do it if dir is read-only (done in permission())
1759 * 2. We should have write and exec permissions on dir
1760 * 3. We can't remove anything from append-only dir
1761 * 4. We can't do anything with immutable dir (done in permission())
1762 * 5. If the sticky bit on dir is set we should either
1763 * a. be owner of dir, or
1764 * b. be owner of victim, or
1765 * c. have CAP_FOWNER capability
1766 * 6. If the victim is append-only or immutable we can't do antyhing with
1767 * links pointing to it.
1768 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1769 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1770 * 9. We can't remove a root or mountpoint.
1771 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1772 * nfs_async_unlink().
1773 */
858119e1 1774static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1da177e4
LT		 1775{
1776 int error;
1777
1778 if (!victim->d_inode)
1779 return -ENOENT;
1780
1781 BUG_ON(victim->d_parent->d_inode != dir);
cccc6bba 1782 audit_inode_child(victim, dir);
1da177e4 1783
f419a2e3 1784 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1785 if (error)
1786 return error;
1787 if (IS_APPEND(dir))
1788 return -EPERM;
1789 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
f9454548 1790 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1da177e4
LT		 1791 return -EPERM;
1792 if (isdir) {
1793 if (!S_ISDIR(victim->d_inode->i_mode))
1794 return -ENOTDIR;
1795 if (IS_ROOT(victim))
1796 return -EBUSY;
1797 } else if (S_ISDIR(victim->d_inode->i_mode))
1798 return -EISDIR;
1799 if (IS_DEADDIR(dir))
1800 return -ENOENT;
1801 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1802 return -EBUSY;
1803 return 0;
1804}
1805
1806/* Check whether we can create an object with dentry child in directory
1807 * dir.
1808 * 1. We can't do it if child already exists (open has special treatment for
1809 * this case, but since we are inlined it's OK)
1810 * 2. We can't do it if dir is read-only (done in permission())
1811 * 3. We should have write and exec permissions on dir
1812 * 4. We can't do it if dir is immutable (done in permission())
1813 */
a95164d9 1814static inline int may_create(struct inode *dir, struct dentry *child)
1da177e4
LT		 1815{
1816 if (child->d_inode)
1817 return -EEXIST;
1818 if (IS_DEADDIR(dir))
1819 return -ENOENT;
f419a2e3 1820 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1821}
1822
1da177e4
LT		 1823/*
1824 * p1 and p2 should be directories on the same fs.
1825 */
1826struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1827{
1828 struct dentry *p;
1829
1830 if (p1 == p2) {
f2eace23 1831 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1da177e4
LT		 1832 return NULL;
1833 }
1834
a11f3a05 1835 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4 1836
e2761a11
OH		 1837 p = d_ancestor(p2, p1);
1838 if (p) {
1839 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1840 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1841 return p;
1da177e4
LT		 1842 }
1843
e2761a11
OH		 1844 p = d_ancestor(p1, p2);
1845 if (p) {
1846 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1847 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1848 return p;
1da177e4
LT		 1849 }
1850
f2eace23
IM		 1851 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1852 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1da177e4
LT		 1853 return NULL;
1854}
1855
1856void unlock_rename(struct dentry *p1, struct dentry *p2)
1857{
1b1dcc1b 1858 mutex_unlock(&p1->d_inode->i_mutex);
1da177e4 1859 if (p1 != p2) {
1b1dcc1b 1860 mutex_unlock(&p2->d_inode->i_mutex);
a11f3a05 1861 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4
LT		 1862 }
1863}
1864
1865int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1866 struct nameidata *nd)
1867{
a95164d9 1868 int error = may_create(dir, dentry);
1da177e4
LT		 1869
1870 if (error)
1871 return error;
1872
acfa4380 1873 if (!dir->i_op->create)
1da177e4
LT		 1874 return -EACCES; /* shouldn't it be ENOSYS? */
1875 mode &= S_IALLUGO;
1876 mode |= S_IFREG;
1877 error = security_inode_create(dir, dentry, mode);
1878 if (error)
1879 return error;
1da177e4 1880 error = dir->i_op->create(dir, dentry, mode, nd);
a74574aa 1881 if (!error)
f38aa942 1882 fsnotify_create(dir, dentry);
1da177e4
LT		 1883 return error;
1884}
1885
73d049a4 1886static int may_open(struct path *path, int acc_mode, int flag)
1da177e4 1887{
3fb64190 1888 struct dentry *dentry = path->dentry;
1da177e4
LT		 1889 struct inode *inode = dentry->d_inode;
1890 int error;
1891
bcda7652
AV		 1892 /* O_PATH? */
1893 if (!acc_mode)
1894 return 0;
1895
1da177e4
LT		 1896 if (!inode)
1897 return -ENOENT;
1898
c8fe8f30
CH		 1899 switch (inode->i_mode & S_IFMT) {
1900 case S_IFLNK:
1da177e4 1901 return -ELOOP;
c8fe8f30
CH		 1902 case S_IFDIR:
1903 if (acc_mode & MAY_WRITE)
1904 return -EISDIR;
1905 break;
1906 case S_IFBLK:
1907 case S_IFCHR:
3fb64190 1908 if (path->mnt->mnt_flags & MNT_NODEV)
1da177e4 1909 return -EACCES;
c8fe8f30
CH		 1910 /*FALLTHRU*/
1911 case S_IFIFO:
1912 case S_IFSOCK:
1da177e4 1913 flag &= ~O_TRUNC;
c8fe8f30 1914 break;
4a3fd211 1915 }
b41572e9 1916
3fb64190 1917 error = inode_permission(inode, acc_mode);
b41572e9
DH		 1918 if (error)
1919 return error;
6146f0d5 1920
1da177e4
LT		 1921 /*
1922 * An append-only file must be opened in append mode for writing.
1923 */
1924 if (IS_APPEND(inode)) {
8737c930 1925 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
7715b521 1926 return -EPERM;
1da177e4 1927 if (flag & O_TRUNC)
7715b521 1928 return -EPERM;
1da177e4
LT		 1929 }
1930
1931 /* O_NOATIME can only be set by the owner or superuser */
2e149670 1932 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
7715b521 1933 return -EPERM;
1da177e4
LT		 1934
1935 /*
1936 * Ensure there are no outstanding leases on the file.
1937 */
b65a9cfc 1938 return break_lease(inode, flag);
7715b521 1939}
1da177e4 1940
e1181ee6 1941static int handle_truncate(struct file *filp)
7715b521 1942{
e1181ee6 1943 struct path *path = &filp->f_path;
7715b521
AV		 1944 struct inode *inode = path->dentry->d_inode;
1945 int error = get_write_access(inode);
1946 if (error)
1947 return error;
1948 /*
1949 * Refuse to truncate files with mandatory locks held on them.
1950 */
1951 error = locks_verify_locked(inode);
1952 if (!error)
ea0d3ab2 1953 error = security_path_truncate(path);
7715b521
AV		 1954 if (!error) {
1955 error = do_truncate(path->dentry, 0,
1956 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
e1181ee6 1957 filp);
7715b521
AV		 1958 }
1959 put_write_access(inode);
acd0c935 1960 return error;
1da177e4
LT		 1961}
1962
d57999e1
DH		 1963static inline int open_to_namei_flags(int flag)
1964{
8a5e929d
AV		 1965 if ((flag & O_ACCMODE) == 3)
1966 flag--;
d57999e1
DH		 1967 return flag;
1968}
1969
31e6b01f 1970/*
fe2d35ff 1971 * Handle the last step of open()
31e6b01f 1972 */
fb1cc555 1973static struct file *do_last(struct nameidata *nd, struct path *path,
c3e380b0 1974 const struct open_flags *op, const char *pathname)
fb1cc555 1975{
a1e28038 1976 struct dentry *dir = nd->path.dentry;
6c0d46c4 1977 struct dentry *dentry;
ca344a89 1978 int open_flag = op->open_flag;
6c0d46c4 1979 int will_truncate = open_flag & O_TRUNC;
ca344a89 1980 int want_write = 0;
bcda7652 1981 int acc_mode = op->acc_mode;
fb1cc555 1982 struct file *filp;
16c2cd71 1983 int error;
1f36f774 1984
c3e380b0
AV		 1985 nd->flags &= ~LOOKUP_PARENT;
1986 nd->flags |= op->intent;
1987
1f36f774
AV		 1988 switch (nd->last_type) {
1989 case LAST_DOTDOT:
176306f5 1990 case LAST_DOT:
fe2d35ff
AV		 1991 error = handle_dots(nd, nd->last_type);
1992 if (error)
1993 return ERR_PTR(error);
1f36f774 1994 /* fallthrough */
1f36f774 1995 case LAST_ROOT:
9f1fafee 1996 error = complete_walk(nd);
16c2cd71 1997 if (error)
9f1fafee 1998 return ERR_PTR(error);
fe2d35ff 1999 audit_inode(pathname, nd->path.dentry);
ca344a89 2000 if (open_flag & O_CREAT) {
fe2d35ff
AV		 2001 error = -EISDIR;
2002 goto exit;
2003 }
2004 goto ok;
1f36f774 2005 case LAST_BIND:
9f1fafee 2006 error = complete_walk(nd);
16c2cd71 2007 if (error)
9f1fafee 2008 return ERR_PTR(error);
1f36f774 2009 audit_inode(pathname, dir);
67ee3ad2 2010 goto ok;
1f36f774 2011 }
67ee3ad2 2012
ca344a89 2013 if (!(open_flag & O_CREAT)) {
bcda7652 2014 int symlink_ok = 0;
fe2d35ff
AV		 2015 if (nd->last.name[nd->last.len])
2016 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
bcda7652
AV		 2017 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2018 symlink_ok = 1;
fe2d35ff 2019 /* we _can_ be in RCU mode here */
ce57dfc1
AV		 2020 error = walk_component(nd, path, &nd->last, LAST_NORM,
2021 !symlink_ok);
2022 if (error < 0)
fe2d35ff 2023 return ERR_PTR(error);
ce57dfc1 2024 if (error) /* symlink */
fe2d35ff 2025 return NULL;
fe2d35ff 2026 /* sayonara */
9f1fafee
AV		 2027 error = complete_walk(nd);
2028 if (error)
2029 return ERR_PTR(-ECHILD);
fe2d35ff
AV		 2030
2031 error = -ENOTDIR;
2032 if (nd->flags & LOOKUP_DIRECTORY) {
ce57dfc1 2033 if (!nd->inode->i_op->lookup)
fe2d35ff
AV		 2034 goto exit;
2035 }
2036 audit_inode(pathname, nd->path.dentry);
2037 goto ok;
2038 }
2039
2040 /* create side of things */
9f1fafee
AV		 2041 error = complete_walk(nd);
2042 if (error)
2043 return ERR_PTR(error);
fe2d35ff
AV		 2044
2045 audit_inode(pathname, dir);
16c2cd71 2046 error = -EISDIR;
1f36f774 2047 /* trailing slashes? */
31e6b01f
NP		 2048 if (nd->last.name[nd->last.len])
2049 goto exit;
a2c36b45 2050
a1e28038
AV		 2051 mutex_lock(&dir->d_inode->i_mutex);
2052
6c0d46c4
AV		 2053 dentry = lookup_hash(nd);
2054 error = PTR_ERR(dentry);
2055 if (IS_ERR(dentry)) {
fb1cc555
AV		 2056 mutex_unlock(&dir->d_inode->i_mutex);
2057 goto exit;
2058 }
2059
6c0d46c4
AV		 2060 path->dentry = dentry;
2061 path->mnt = nd->path.mnt;
2062
fb1cc555 2063 /* Negative dentry, just create the file */
6c0d46c4
AV		 2064 if (!dentry->d_inode) {
2065 int mode = op->mode;
2066 if (!IS_POSIXACL(dir->d_inode))
2067 mode &= ~current_umask();
fb1cc555
AV		 2068 /*
2069 * This write is needed to ensure that a
6c0d46c4 2070 * rw->ro transition does not occur between
fb1cc555
AV		 2071 * the time when the file is created and when
2072 * a permanent write count is taken through
2073 * the 'struct file' in nameidata_to_filp().
2074 */
2075 error = mnt_want_write(nd->path.mnt);
2076 if (error)
2077 goto exit_mutex_unlock;
ca344a89 2078 want_write = 1;
9b44f1b3 2079 /* Don't check for write permission, don't truncate */
ca344a89 2080 open_flag &= ~O_TRUNC;
6c0d46c4 2081 will_truncate = 0;
bcda7652 2082 acc_mode = MAY_OPEN;
6c0d46c4
AV		 2083 error = security_path_mknod(&nd->path, dentry, mode, 0);
2084 if (error)
2085 goto exit_mutex_unlock;
2086 error = vfs_create(dir->d_inode, dentry, mode, nd);
2087 if (error)
2088 goto exit_mutex_unlock;
2089 mutex_unlock(&dir->d_inode->i_mutex);
2090 dput(nd->path.dentry);
2091 nd->path.dentry = dentry;
ca344a89 2092 goto common;
fb1cc555
AV		 2093 }
2094
2095 /*
2096 * It already exists.
2097 */
2098 mutex_unlock(&dir->d_inode->i_mutex);
2099 audit_inode(pathname, path->dentry);
2100
2101 error = -EEXIST;
ca344a89 2102 if (open_flag & O_EXCL)
fb1cc555
AV		 2103 goto exit_dput;
2104
9875cf80
DH		 2105 error = follow_managed(path, nd->flags);
2106 if (error < 0)
2107 goto exit_dput;
fb1cc555
AV		 2108
2109 error = -ENOENT;
2110 if (!path->dentry->d_inode)
2111 goto exit_dput;
9e67f361
AV		 2112
2113 if (path->dentry->d_inode->i_op->follow_link)
fb1cc555 2114 return NULL;
fb1cc555
AV		 2115
2116 path_to_nameidata(path, nd);
31e6b01f 2117 nd->inode = path->dentry->d_inode;
fb1cc555 2118 error = -EISDIR;
31e6b01f 2119 if (S_ISDIR(nd->inode->i_mode))
fb1cc555 2120 goto exit;
67ee3ad2 2121ok:
6c0d46c4
AV		 2122 if (!S_ISREG(nd->inode->i_mode))
2123 will_truncate = 0;
2124
0f9d1a10
AV		 2125 if (will_truncate) {
2126 error = mnt_want_write(nd->path.mnt);
2127 if (error)
2128 goto exit;
ca344a89 2129 want_write = 1;
0f9d1a10 2130 }
ca344a89 2131common:
bcda7652 2132 error = may_open(&nd->path, acc_mode, open_flag);
ca344a89 2133 if (error)
0f9d1a10 2134 goto exit;
0f9d1a10
AV		 2135 filp = nameidata_to_filp(nd);
2136 if (!IS_ERR(filp)) {
2137 error = ima_file_check(filp, op->acc_mode);
2138 if (error) {
2139 fput(filp);
2140 filp = ERR_PTR(error);
2141 }
2142 }
2143 if (!IS_ERR(filp)) {
2144 if (will_truncate) {
2145 error = handle_truncate(filp);
2146 if (error) {
2147 fput(filp);
2148 filp = ERR_PTR(error);
2149 }
2150 }
2151 }
ca344a89
AV		 2152out:
2153 if (want_write)
0f9d1a10
AV		 2154 mnt_drop_write(nd->path.mnt);
2155 path_put(&nd->path);
fb1cc555
AV		 2156 return filp;
2157
2158exit_mutex_unlock:
2159 mutex_unlock(&dir->d_inode->i_mutex);
2160exit_dput:
2161 path_put_conditional(path, nd);
2162exit:
ca344a89
AV		 2163 filp = ERR_PTR(error);
2164 goto out;
fb1cc555
AV		 2165}
2166
13aab428 2167static struct file *path_openat(int dfd, const char *pathname,
73d049a4 2168 struct nameidata *nd, const struct open_flags *op, int flags)
1da177e4 2169{
fe2d35ff 2170 struct file *base = NULL;
4a3fd211 2171 struct file *filp;
9850c056 2172 struct path path;
13aab428 2173 int error;
31e6b01f
NP		 2174
2175 filp = get_empty_filp();
2176 if (!filp)
2177 return ERR_PTR(-ENFILE);
2178
47c805dc 2179 filp->f_flags = op->open_flag;
73d049a4
AV		 2180 nd->intent.open.file = filp;
2181 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2182 nd->intent.open.create_mode = op->mode;
31e6b01f 2183
73d049a4 2184 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
31e6b01f 2185 if (unlikely(error))
13aab428 2186 goto out_filp;
31e6b01f 2187
fe2d35ff 2188 current->total_link_count = 0;
73d049a4 2189 error = link_path_walk(pathname, nd);
31e6b01f
NP		 2190 if (unlikely(error))
2191 goto out_filp;
1da177e4 2192
73d049a4 2193 filp = do_last(nd, &path, op, pathname);
806b681c 2194 while (unlikely(!filp)) { /* trailing symlink */
7b9337aa 2195 struct path link = path;
def4af30 2196 void *cookie;
574197e0 2197 if (!(nd->flags & LOOKUP_FOLLOW)) {
73d049a4
AV		 2198 path_put_conditional(&path, nd);
2199 path_put(&nd->path);
40b39136
AV		 2200 filp = ERR_PTR(-ELOOP);
2201 break;
2202 }
73d049a4
AV		 2203 nd->flags |= LOOKUP_PARENT;
2204 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
574197e0 2205 error = follow_link(&link, nd, &cookie);
c3e380b0 2206 if (unlikely(error))
f1afe9ef 2207 filp = ERR_PTR(error);
c3e380b0 2208 else
73d049a4 2209 filp = do_last(nd, &path, op, pathname);
574197e0 2210 put_link(nd, &link, cookie);
806b681c 2211 }
10fa8e62 2212out:
73d049a4
AV		 2213 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2214 path_put(&nd->root);
fe2d35ff
AV		 2215 if (base)
2216 fput(base);
73d049a4 2217 release_open_intent(nd);
10fa8e62 2218 return filp;
1da177e4 2219
31e6b01f 2220out_filp:
806b681c 2221 filp = ERR_PTR(error);
10fa8e62 2222 goto out;
1da177e4
LT		 2223}
2224
13aab428
AV		 2225struct file *do_filp_open(int dfd, const char *pathname,
2226 const struct open_flags *op, int flags)
2227{
73d049a4 2228 struct nameidata nd;
13aab428
AV		 2229 struct file *filp;
2230
73d049a4 2231 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
13aab428 2232 if (unlikely(filp == ERR_PTR(-ECHILD)))
73d049a4 2233 filp = path_openat(dfd, pathname, &nd, op, flags);
13aab428 2234 if (unlikely(filp == ERR_PTR(-ESTALE)))
73d049a4 2235 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
13aab428
AV		 2236 return filp;
2237}
2238
73d049a4
AV		 2239struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2240 const char *name, const struct open_flags *op, int flags)
2241{
2242 struct nameidata nd;
2243 struct file *file;
2244
2245 nd.root.mnt = mnt;
2246 nd.root.dentry = dentry;
2247
2248 flags |= LOOKUP_ROOT;
2249
bcda7652 2250 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
73d049a4
AV		 2251 return ERR_PTR(-ELOOP);
2252
2253 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2254 if (unlikely(file == ERR_PTR(-ECHILD)))
2255 file = path_openat(-1, name, &nd, op, flags);
2256 if (unlikely(file == ERR_PTR(-ESTALE)))
2257 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2258 return file;
2259}
2260
1da177e4
LT		 2261/**
2262 * lookup_create - lookup a dentry, creating it if it doesn't exist
2263 * @nd: nameidata info
2264 * @is_dir: directory flag
2265 *
2266 * Simple function to lookup and return a dentry and create it
2267 * if it doesn't exist. Is SMP-safe.
c663e5d8 2268 *
4ac91378 2269 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1da177e4
LT		 2270 */
2271struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2272{
c663e5d8 2273 struct dentry *dentry = ERR_PTR(-EEXIST);
1da177e4 2274
4ac91378 2275 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
c663e5d8
CH		 2276 /*
2277 * Yucky last component or no last component at all?
2278 * (foo/., foo/.., /////)
2279 */
1da177e4
LT		 2280 if (nd->last_type != LAST_NORM)
2281 goto fail;
2282 nd->flags &= ~LOOKUP_PARENT;
3516586a 2283 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
a634904a 2284 nd->intent.open.flags = O_EXCL;
c663e5d8
CH		 2285
2286 /*
2287 * Do the final lookup.
2288 */
49705b77 2289 dentry = lookup_hash(nd);
1da177e4
LT		 2290 if (IS_ERR(dentry))
2291 goto fail;
c663e5d8 2292
e9baf6e5
AV		 2293 if (dentry->d_inode)
2294 goto eexist;
c663e5d8
CH		 2295 /*
2296 * Special case - lookup gave negative, but... we had foo/bar/
2297 * From the vfs_mknod() POV we just have a negative dentry -
2298 * all is fine. Let's be bastards - you had / on the end, you've
2299 * been asking for (non-existent) directory. -ENOENT for you.
2300 */
e9baf6e5
AV		 2301 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2302 dput(dentry);
2303 dentry = ERR_PTR(-ENOENT);
2304 }
1da177e4 2305 return dentry;
e9baf6e5 2306eexist:
1da177e4 2307 dput(dentry);
e9baf6e5 2308 dentry = ERR_PTR(-EEXIST);
1da177e4
LT		 2309fail:
2310 return dentry;
2311}
f81a0bff 2312EXPORT_SYMBOL_GPL(lookup_create);
1da177e4
LT		 2313
2314int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2315{
a95164d9 2316 int error = may_create(dir, dentry);
1da177e4
LT		 2317
2318 if (error)
2319 return error;
2320
e795b717
SH		 2321 if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
2322 !ns_capable(inode_userns(dir), CAP_MKNOD))
1da177e4
LT		 2323 return -EPERM;
2324
acfa4380 2325 if (!dir->i_op->mknod)
1da177e4
LT		 2326 return -EPERM;
2327
08ce5f16
SH		 2328 error = devcgroup_inode_mknod(mode, dev);
2329 if (error)
2330 return error;
2331
1da177e4
LT		 2332 error = security_inode_mknod(dir, dentry, mode, dev);
2333 if (error)
2334 return error;
2335
1da177e4 2336 error = dir->i_op->mknod(dir, dentry, mode, dev);
a74574aa 2337 if (!error)
f38aa942 2338 fsnotify_create(dir, dentry);
1da177e4
LT		 2339 return error;
2340}
2341
463c3197
DH		 2342static int may_mknod(mode_t mode)
2343{
2344 switch (mode & S_IFMT) {
2345 case S_IFREG:
2346 case S_IFCHR:
2347 case S_IFBLK:
2348 case S_IFIFO:
2349 case S_IFSOCK:
2350 case 0: /* zero mode translates to S_IFREG */
2351 return 0;
2352 case S_IFDIR:
2353 return -EPERM;
2354 default:
2355 return -EINVAL;
2356 }
2357}
2358
2e4d0924
HC		 2359SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2360 unsigned, dev)
1da177e4 2361{
2ad94ae6
AV		 2362 int error;
2363 char *tmp;
2364 struct dentry *dentry;
1da177e4
LT		 2365 struct nameidata nd;
2366
2367 if (S_ISDIR(mode))
2368 return -EPERM;
1da177e4 2369
2ad94ae6 2370 error = user_path_parent(dfd, filename, &nd, &tmp);
1da177e4 2371 if (error)
2ad94ae6
AV		 2372 return error;
2373
1da177e4 2374 dentry = lookup_create(&nd, 0);
463c3197
DH		 2375 if (IS_ERR(dentry)) {
2376 error = PTR_ERR(dentry);
2377 goto out_unlock;
2378 }
4ac91378 2379 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2380 mode &= ~current_umask();
463c3197
DH		 2381 error = may_mknod(mode);
2382 if (error)
2383 goto out_dput;
2384 error = mnt_want_write(nd.path.mnt);
2385 if (error)
2386 goto out_dput;
be6d3e56
KT		 2387 error = security_path_mknod(&nd.path, dentry, mode, dev);
2388 if (error)
2389 goto out_drop_write;
463c3197 2390 switch (mode & S_IFMT) {
1da177e4 2391 case 0: case S_IFREG:
554a8b9f 2392 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,NULL);
1da177e4
LT		 2393 break;
2394 case S_IFCHR: case S_IFBLK:
4ac91378 2395 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
1da177e4
LT		 2396 new_decode_dev(dev));
2397 break;
2398 case S_IFIFO: case S_IFSOCK:
4ac91378 2399 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
1da177e4 2400 break;
1da177e4 2401 }
be6d3e56 2402out_drop_write:
463c3197
DH		 2403 mnt_drop_write(nd.path.mnt);
2404out_dput:
2405 dput(dentry);
2406out_unlock:
4ac91378 2407 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2408 path_put(&nd.path);
1da177e4
LT		 2409 putname(tmp);
2410
2411 return error;
2412}
2413
3480b257 2414SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
5590ff0d
UD		 2415{
2416 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2417}
2418
1da177e4
LT		 2419int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2420{
a95164d9 2421 int error = may_create(dir, dentry);
1da177e4
LT		 2422
2423 if (error)
2424 return error;
2425
acfa4380 2426 if (!dir->i_op->mkdir)
1da177e4
LT		 2427 return -EPERM;
2428
2429 mode &= (S_IRWXUGO|S_ISVTX);
2430 error = security_inode_mkdir(dir, dentry, mode);
2431 if (error)
2432 return error;
2433
1da177e4 2434 error = dir->i_op->mkdir(dir, dentry, mode);
a74574aa 2435 if (!error)
f38aa942 2436 fsnotify_mkdir(dir, dentry);
1da177e4
LT		 2437 return error;
2438}
2439
2e4d0924 2440SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
1da177e4
LT		 2441{
2442 int error = 0;
2443 char * tmp;
6902d925
DH		 2444 struct dentry *dentry;
2445 struct nameidata nd;
1da177e4 2446
2ad94ae6
AV		 2447 error = user_path_parent(dfd, pathname, &nd, &tmp);
2448 if (error)
6902d925 2449 goto out_err;
1da177e4 2450
6902d925
DH		 2451 dentry = lookup_create(&nd, 1);
2452 error = PTR_ERR(dentry);
2453 if (IS_ERR(dentry))
2454 goto out_unlock;
1da177e4 2455
4ac91378 2456 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2457 mode &= ~current_umask();
463c3197
DH		 2458 error = mnt_want_write(nd.path.mnt);
2459 if (error)
2460 goto out_dput;
be6d3e56
KT		 2461 error = security_path_mkdir(&nd.path, dentry, mode);
2462 if (error)
2463 goto out_drop_write;
4ac91378 2464 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
be6d3e56 2465out_drop_write:
463c3197
DH		 2466 mnt_drop_write(nd.path.mnt);
2467out_dput:
6902d925
DH		 2468 dput(dentry);
2469out_unlock:
4ac91378 2470 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2471 path_put(&nd.path);
6902d925
DH		 2472 putname(tmp);
2473out_err:
1da177e4
LT		 2474 return error;
2475}
2476
3cdad428 2477SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
5590ff0d
UD		 2478{
2479 return sys_mkdirat(AT_FDCWD, pathname, mode);
2480}
2481
1da177e4 2482/*
a71905f0
SW		 2483 * The dentry_unhash() helper will try to drop the dentry early: we
2484 * should have a usage count of 2 if we're the only user of this
2485 * dentry, and if that is true (possibly after pruning the dcache),
2486 * then we drop the dentry now.
1da177e4
LT		 2487 *
2488 * A low-level filesystem can, if it choses, legally
2489 * do a
2490 *
2491 * if (!d_unhashed(dentry))
2492 * return -EBUSY;
2493 *
2494 * if it cannot handle the case of removing a directory
2495 * that is still in use by something else..
2496 */
2497void dentry_unhash(struct dentry *dentry)
2498{
dc168427 2499 shrink_dcache_parent(dentry);
1da177e4 2500 spin_lock(&dentry->d_lock);
64252c75 2501 if (dentry->d_count == 1)
1da177e4
LT		 2502 __d_drop(dentry);
2503 spin_unlock(&dentry->d_lock);
1da177e4
LT		 2504}
2505
2506int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2507{
2508 int error = may_delete(dir, dentry, 1);
2509
2510 if (error)
2511 return error;
2512
acfa4380 2513 if (!dir->i_op->rmdir)
1da177e4
LT		 2514 return -EPERM;
2515
1b1dcc1b 2516 mutex_lock(&dentry->d_inode->i_mutex);
912dbc15
SW		 2517
2518 error = -EBUSY;
1da177e4 2519 if (d_mountpoint(dentry))
912dbc15
SW		 2520 goto out;
2521
2522 error = security_inode_rmdir(dir, dentry);
2523 if (error)
2524 goto out;
2525
3cebde24 2526 shrink_dcache_parent(dentry);
912dbc15
SW		 2527 error = dir->i_op->rmdir(dir, dentry);
2528 if (error)
2529 goto out;
2530
2531 dentry->d_inode->i_flags |= S_DEAD;
2532 dont_mount(dentry);
2533
2534out:
1b1dcc1b 2535 mutex_unlock(&dentry->d_inode->i_mutex);
912dbc15 2536 if (!error)
1da177e4 2537 d_delete(dentry);
1da177e4
LT		 2538 return error;
2539}
2540
5590ff0d 2541static long do_rmdir(int dfd, const char __user *pathname)
1da177e4
LT		 2542{
2543 int error = 0;
2544 char * name;
2545 struct dentry *dentry;
2546 struct nameidata nd;
2547
2ad94ae6 2548 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2549 if (error)
2ad94ae6 2550 return error;
1da177e4
LT		 2551
2552 switch(nd.last_type) {
0612d9fb
OH		 2553 case LAST_DOTDOT:
2554 error = -ENOTEMPTY;
2555 goto exit1;
2556 case LAST_DOT:
2557 error = -EINVAL;
2558 goto exit1;
2559 case LAST_ROOT:
2560 error = -EBUSY;
2561 goto exit1;
1da177e4 2562 }
0612d9fb
OH		 2563
2564 nd.flags &= ~LOOKUP_PARENT;
2565
4ac91378 2566 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2567 dentry = lookup_hash(&nd);
1da177e4 2568 error = PTR_ERR(dentry);
6902d925
DH		 2569 if (IS_ERR(dentry))
2570 goto exit2;
e6bc45d6
TT		 2571 if (!dentry->d_inode) {
2572 error = -ENOENT;
2573 goto exit3;
2574 }
0622753b
DH		 2575 error = mnt_want_write(nd.path.mnt);
2576 if (error)
2577 goto exit3;
be6d3e56
KT		 2578 error = security_path_rmdir(&nd.path, dentry);
2579 if (error)
2580 goto exit4;
4ac91378 2581 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
be6d3e56 2582exit4:
0622753b
DH		 2583 mnt_drop_write(nd.path.mnt);
2584exit3:
6902d925
DH		 2585 dput(dentry);
2586exit2:
4ac91378 2587 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2588exit1:
1d957f9b 2589 path_put(&nd.path);
1da177e4
LT		 2590 putname(name);
2591 return error;
2592}
2593
3cdad428 2594SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
5590ff0d
UD		 2595{
2596 return do_rmdir(AT_FDCWD, pathname);
2597}
2598
1da177e4
LT		 2599int vfs_unlink(struct inode *dir, struct dentry *dentry)
2600{
2601 int error = may_delete(dir, dentry, 0);
2602
2603 if (error)
2604 return error;
2605
acfa4380 2606 if (!dir->i_op->unlink)
1da177e4
LT		 2607 return -EPERM;
2608
1b1dcc1b 2609 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2610 if (d_mountpoint(dentry))
2611 error = -EBUSY;
2612 else {
2613 error = security_inode_unlink(dir, dentry);
bec1052e 2614 if (!error) {
1da177e4 2615 error = dir->i_op->unlink(dir, dentry);
bec1052e 2616 if (!error)
d83c49f3 2617 dont_mount(dentry);
bec1052e 2618 }
1da177e4 2619 }
1b1dcc1b 2620 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2621
2622 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2623 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
ece95912 2624 fsnotify_link_count(dentry->d_inode);
e234f35c 2625 d_delete(dentry);
1da177e4 2626 }
0eeca283 2627
1da177e4
LT		 2628 return error;
2629}
2630
2631/*
2632 * Make sure that the actual truncation of the file will occur outside its
1b1dcc1b 2633 * directory's i_mutex. Truncate can take a long time if there is a lot of
1da177e4
LT		 2634 * writeout happening, and we don't want to prevent access to the directory
2635 * while waiting on the I/O.
2636 */
5590ff0d 2637static long do_unlinkat(int dfd, const char __user *pathname)
1da177e4 2638{
2ad94ae6
AV		 2639 int error;
2640 char *name;
1da177e4
LT		 2641 struct dentry *dentry;
2642 struct nameidata nd;
2643 struct inode *inode = NULL;
2644
2ad94ae6 2645 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2646 if (error)
2ad94ae6
AV		 2647 return error;
2648
1da177e4
LT		 2649 error = -EISDIR;
2650 if (nd.last_type != LAST_NORM)
2651 goto exit1;
0612d9fb
OH		 2652
2653 nd.flags &= ~LOOKUP_PARENT;
2654
4ac91378 2655 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2656 dentry = lookup_hash(&nd);
1da177e4
LT		 2657 error = PTR_ERR(dentry);
2658 if (!IS_ERR(dentry)) {
2659 /* Why not before? Because we want correct error value */
50338b88
TE		 2660 if (nd.last.name[nd.last.len])
2661 goto slashes;
1da177e4 2662 inode = dentry->d_inode;
50338b88 2663 if (!inode)
e6bc45d6
TT		 2664 goto slashes;
2665 ihold(inode);
0622753b
DH		 2666 error = mnt_want_write(nd.path.mnt);
2667 if (error)
2668 goto exit2;
be6d3e56
KT		 2669 error = security_path_unlink(&nd.path, dentry);
2670 if (error)
2671 goto exit3;
4ac91378 2672 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
be6d3e56 2673exit3:
0622753b 2674 mnt_drop_write(nd.path.mnt);
1da177e4
LT		 2675 exit2:
2676 dput(dentry);
2677 }
4ac91378 2678 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4
LT		 2679 if (inode)
2680 iput(inode); /* truncate the inode here */
2681exit1:
1d957f9b 2682 path_put(&nd.path);
1da177e4
LT		 2683 putname(name);
2684 return error;
2685
2686slashes:
2687 error = !dentry->d_inode ? -ENOENT :
2688 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2689 goto exit2;
2690}
2691
2e4d0924 2692SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
5590ff0d
UD		 2693{
2694 if ((flag & ~AT_REMOVEDIR) != 0)
2695 return -EINVAL;
2696
2697 if (flag & AT_REMOVEDIR)
2698 return do_rmdir(dfd, pathname);
2699
2700 return do_unlinkat(dfd, pathname);
2701}
2702
3480b257 2703SYSCALL_DEFINE1(unlink, const char __user *, pathname)
5590ff0d
UD		 2704{
2705 return do_unlinkat(AT_FDCWD, pathname);
2706}
2707
db2e747b 2708int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
1da177e4 2709{
a95164d9 2710 int error = may_create(dir, dentry);
1da177e4
LT		 2711
2712 if (error)
2713 return error;
2714
acfa4380 2715 if (!dir->i_op->symlink)
1da177e4
LT		 2716 return -EPERM;
2717
2718 error = security_inode_symlink(dir, dentry, oldname);
2719 if (error)
2720 return error;
2721
1da177e4 2722 error = dir->i_op->symlink(dir, dentry, oldname);
a74574aa 2723 if (!error)
f38aa942 2724 fsnotify_create(dir, dentry);
1da177e4
LT		 2725 return error;
2726}
2727
2e4d0924
HC		 2728SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2729 int, newdfd, const char __user *, newname)
1da177e4 2730{
2ad94ae6
AV		 2731 int error;
2732 char *from;
2733 char *to;
6902d925
DH		 2734 struct dentry *dentry;
2735 struct nameidata nd;
1da177e4
LT		 2736
2737 from = getname(oldname);
2ad94ae6 2738 if (IS_ERR(from))
1da177e4 2739 return PTR_ERR(from);
1da177e4 2740
2ad94ae6 2741 error = user_path_parent(newdfd, newname, &nd, &to);
6902d925 2742 if (error)
2ad94ae6
AV		 2743 goto out_putname;
2744
6902d925
DH		 2745 dentry = lookup_create(&nd, 0);
2746 error = PTR_ERR(dentry);
2747 if (IS_ERR(dentry))
2748 goto out_unlock;
2749
75c3f29d
DH		 2750 error = mnt_want_write(nd.path.mnt);
2751 if (error)
2752 goto out_dput;
be6d3e56
KT		 2753 error = security_path_symlink(&nd.path, dentry, from);
2754 if (error)
2755 goto out_drop_write;
db2e747b 2756 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
be6d3e56 2757out_drop_write:
75c3f29d
DH		 2758 mnt_drop_write(nd.path.mnt);
2759out_dput:
6902d925
DH		 2760 dput(dentry);
2761out_unlock:
4ac91378 2762 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2763 path_put(&nd.path);
6902d925
DH		 2764 putname(to);
2765out_putname:
1da177e4
LT		 2766 putname(from);
2767 return error;
2768}
2769
3480b257 2770SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 2771{
2772 return sys_symlinkat(oldname, AT_FDCWD, newname);
2773}
2774
1da177e4
LT		 2775int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2776{
2777 struct inode *inode = old_dentry->d_inode;
2778 int error;
2779
2780 if (!inode)
2781 return -ENOENT;
2782
a95164d9 2783 error = may_create(dir, new_dentry);
1da177e4
LT		 2784 if (error)
2785 return error;
2786
2787 if (dir->i_sb != inode->i_sb)
2788 return -EXDEV;
2789
2790 /*
2791 * A link to an append-only or immutable file cannot be created.
2792 */
2793 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2794 return -EPERM;
acfa4380 2795 if (!dir->i_op->link)
1da177e4 2796 return -EPERM;
7e79eedb 2797 if (S_ISDIR(inode->i_mode))
1da177e4
LT		 2798 return -EPERM;
2799
2800 error = security_inode_link(old_dentry, dir, new_dentry);
2801 if (error)
2802 return error;
2803
7e79eedb 2804 mutex_lock(&inode->i_mutex);
aae8a97d
AK		 2805 /* Make sure we don't allow creating hardlink to an unlinked file */
2806 if (inode->i_nlink == 0)
2807 error = -ENOENT;
2808 else
2809 error = dir->i_op->link(old_dentry, dir, new_dentry);
7e79eedb 2810 mutex_unlock(&inode->i_mutex);
e31e14ec 2811 if (!error)
7e79eedb 2812 fsnotify_link(dir, inode, new_dentry);
1da177e4
LT		 2813 return error;
2814}
2815
2816/*
2817 * Hardlinks are often used in delicate situations. We avoid
2818 * security-related surprises by not following symlinks on the
2819 * newname. --KAB
2820 *
2821 * We don't follow them on the oldname either to be compatible
2822 * with linux 2.0, and to avoid hard-linking to directories
2823 * and other special files. --ADM
2824 */
2e4d0924
HC		 2825SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2826 int, newdfd, const char __user *, newname, int, flags)
1da177e4
LT		 2827{
2828 struct dentry *new_dentry;
2d8f3038
AV		 2829 struct nameidata nd;
2830 struct path old_path;
11a7b371 2831 int how = 0;
1da177e4 2832 int error;
2ad94ae6 2833 char *to;
1da177e4 2834
11a7b371 2835 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
c04030e1 2836 return -EINVAL;
11a7b371
AK		 2837 /*
2838 * To use null names we require CAP_DAC_READ_SEARCH
2839 * This ensures that not everyone will be able to create
2840 * handlink using the passed filedescriptor.
2841 */
2842 if (flags & AT_EMPTY_PATH) {
2843 if (!capable(CAP_DAC_READ_SEARCH))
2844 return -ENOENT;
2845 how = LOOKUP_EMPTY;
2846 }
2847
2848 if (flags & AT_SYMLINK_FOLLOW)
2849 how |= LOOKUP_FOLLOW;
c04030e1 2850
11a7b371 2851 error = user_path_at(olddfd, oldname, how, &old_path);
1da177e4 2852 if (error)
2ad94ae6
AV		 2853 return error;
2854
2855 error = user_path_parent(newdfd, newname, &nd, &to);
1da177e4
LT		 2856 if (error)
2857 goto out;
2858 error = -EXDEV;
2d8f3038 2859 if (old_path.mnt != nd.path.mnt)
1da177e4
LT		 2860 goto out_release;
2861 new_dentry = lookup_create(&nd, 0);
2862 error = PTR_ERR(new_dentry);
6902d925
DH		 2863 if (IS_ERR(new_dentry))
2864 goto out_unlock;
75c3f29d
DH		 2865 error = mnt_want_write(nd.path.mnt);
2866 if (error)
2867 goto out_dput;
be6d3e56
KT		 2868 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2869 if (error)
2870 goto out_drop_write;
2d8f3038 2871 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
be6d3e56 2872out_drop_write:
75c3f29d
DH		 2873 mnt_drop_write(nd.path.mnt);
2874out_dput:
6902d925
DH		 2875 dput(new_dentry);
2876out_unlock:
4ac91378 2877 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2878out_release:
1d957f9b 2879 path_put(&nd.path);
2ad94ae6 2880 putname(to);
1da177e4 2881out:
2d8f3038 2882 path_put(&old_path);
1da177e4
LT		 2883
2884 return error;
2885}
2886
3480b257 2887SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
5590ff0d 2888{
c04030e1 2889 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
5590ff0d
UD		 2890}
2891
1da177e4
LT		 2892/*
2893 * The worst of all namespace operations - renaming directory. "Perverted"
2894 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2895 * Problems:
2896 * a) we can get into loop creation. Check is done in is_subdir().
2897 * b) race potential - two innocent renames can create a loop together.
2898 * That's where 4.4 screws up. Current fix: serialization on
a11f3a05 2899 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
1da177e4
LT		 2900 * story.
2901 * c) we have to lock _three_ objects - parents and victim (if it exists).
1b1dcc1b 2902 * And that - after we got ->i_mutex on parents (until then we don't know
1da177e4
LT		 2903 * whether the target exists). Solution: try to be smart with locking
2904 * order for inodes. We rely on the fact that tree topology may change
a11f3a05 2905 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
1da177e4
LT		 2906 * move will be locked. Thus we can rank directories by the tree
2907 * (ancestors first) and rank all non-directories after them.
2908 * That works since everybody except rename does "lock parent, lookup,
a11f3a05 2909 * lock child" and rename is under ->s_vfs_rename_mutex.
1da177e4
LT		 2910 * HOWEVER, it relies on the assumption that any object with ->lookup()
2911 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2912 * we'd better make sure that there's no link(2) for them.
e4eaac06 2913 * d) conversion from fhandle to dentry may come in the wrong moment - when
1b1dcc1b 2914 * we are removing the target. Solution: we will have to grab ->i_mutex
1da177e4 2915 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
c41b20e7 2916 * ->i_mutex on parents, which works but leads to some truly excessive
1da177e4
LT		 2917 * locking].
2918 */
75c96f85
AB		 2919static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2920 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 2921{
2922 int error = 0;
9055cba7 2923 struct inode *target = new_dentry->d_inode;
1da177e4
LT		 2924
2925 /*
2926 * If we are going to change the parent - check write permissions,
2927 * we'll need to flip '..'.
2928 */
2929 if (new_dir != old_dir) {
f419a2e3 2930 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
1da177e4
LT		 2931 if (error)
2932 return error;
2933 }
2934
2935 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2936 if (error)
2937 return error;
2938
d83c49f3 2939 if (target)
1b1dcc1b 2940 mutex_lock(&target->i_mutex);
9055cba7
SW		 2941
2942 error = -EBUSY;
2943 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
2944 goto out;
2945
3cebde24
SW		 2946 if (target)
2947 shrink_dcache_parent(new_dentry);
9055cba7
SW		 2948 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2949 if (error)
2950 goto out;
2951
1da177e4 2952 if (target) {
9055cba7
SW		 2953 target->i_flags |= S_DEAD;
2954 dont_mount(new_dentry);
1da177e4 2955 }
9055cba7
SW		 2956out:
2957 if (target)
2958 mutex_unlock(&target->i_mutex);
e31e14ec 2959 if (!error)
349457cc
MF		 2960 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2961 d_move(old_dentry,new_dentry);
1da177e4
LT		 2962 return error;
2963}
2964
75c96f85
AB		 2965static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2966 struct inode *new_dir, struct dentry *new_dentry)
1da177e4 2967{
51892bbb 2968 struct inode *target = new_dentry->d_inode;
1da177e4
LT		 2969 int error;
2970
2971 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2972 if (error)
2973 return error;
2974
2975 dget(new_dentry);
1da177e4 2976 if (target)
1b1dcc1b 2977 mutex_lock(&target->i_mutex);
51892bbb
SW		 2978
2979 error = -EBUSY;
1da177e4 2980 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
51892bbb
SW		 2981 goto out;
2982
2983 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2984 if (error)
2985 goto out;
2986
2987 if (target)
2988 dont_mount(new_dentry);
2989 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2990 d_move(old_dentry, new_dentry);
2991out:
1da177e4 2992 if (target)
1b1dcc1b 2993 mutex_unlock(&target->i_mutex);
1da177e4
LT		 2994 dput(new_dentry);
2995 return error;
2996}
2997
2998int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2999 struct inode *new_dir, struct dentry *new_dentry)
3000{
3001 int error;
3002 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
59b0df21 3003 const unsigned char *old_name;
1da177e4
LT		 3004
3005 if (old_dentry->d_inode == new_dentry->d_inode)
3006 return 0;
3007
3008 error = may_delete(old_dir, old_dentry, is_dir);
3009 if (error)
3010 return error;
3011
3012 if (!new_dentry->d_inode)
a95164d9 3013 error = may_create(new_dir, new_dentry);
1da177e4
LT		 3014 else
3015 error = may_delete(new_dir, new_dentry, is_dir);
3016 if (error)
3017 return error;
3018
acfa4380 3019 if (!old_dir->i_op->rename)
1da177e4
LT		 3020 return -EPERM;
3021
0eeca283
RL		 3022 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3023
1da177e4
LT		 3024 if (is_dir)
3025 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3026 else
3027 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
123df294
AV		 3028 if (!error)
3029 fsnotify_move(old_dir, new_dir, old_name, is_dir,
5a190ae6 3030 new_dentry->d_inode, old_dentry);
0eeca283
RL		 3031 fsnotify_oldname_free(old_name);
3032
1da177e4
LT		 3033 return error;
3034}
3035
2e4d0924
HC		 3036SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3037 int, newdfd, const char __user *, newname)
1da177e4 3038{
2ad94ae6
AV		 3039 struct dentry *old_dir, *new_dir;
3040 struct dentry *old_dentry, *new_dentry;
3041 struct dentry *trap;
1da177e4 3042 struct nameidata oldnd, newnd;
2ad94ae6
AV		 3043 char *from;
3044 char *to;
3045 int error;
1da177e4 3046
2ad94ae6 3047 error = user_path_parent(olddfd, oldname, &oldnd, &from);
1da177e4
LT		 3048 if (error)
3049 goto exit;
3050
2ad94ae6 3051 error = user_path_parent(newdfd, newname, &newnd, &to);
1da177e4
LT		 3052 if (error)
3053 goto exit1;
3054
3055 error = -EXDEV;
4ac91378 3056 if (oldnd.path.mnt != newnd.path.mnt)
1da177e4
LT		 3057 goto exit2;
3058
4ac91378 3059 old_dir = oldnd.path.dentry;
1da177e4
LT		 3060 error = -EBUSY;
3061 if (oldnd.last_type != LAST_NORM)
3062 goto exit2;
3063
4ac91378 3064 new_dir = newnd.path.dentry;
1da177e4
LT		 3065 if (newnd.last_type != LAST_NORM)
3066 goto exit2;
3067
0612d9fb
OH		 3068 oldnd.flags &= ~LOOKUP_PARENT;
3069 newnd.flags &= ~LOOKUP_PARENT;
4e9ed2f8 3070 newnd.flags |= LOOKUP_RENAME_TARGET;
0612d9fb 3071
1da177e4
LT		 3072 trap = lock_rename(new_dir, old_dir);
3073
49705b77 3074 old_dentry = lookup_hash(&oldnd);
1da177e4
LT		 3075 error = PTR_ERR(old_dentry);
3076 if (IS_ERR(old_dentry))
3077 goto exit3;
3078 /* source must exist */
3079 error = -ENOENT;
3080 if (!old_dentry->d_inode)
3081 goto exit4;
3082 /* unless the source is a directory trailing slashes give -ENOTDIR */
3083 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3084 error = -ENOTDIR;
3085 if (oldnd.last.name[oldnd.last.len])
3086 goto exit4;
3087 if (newnd.last.name[newnd.last.len])
3088 goto exit4;
3089 }
3090 /* source should not be ancestor of target */
3091 error = -EINVAL;
3092 if (old_dentry == trap)
3093 goto exit4;
49705b77 3094 new_dentry = lookup_hash(&newnd);
1da177e4
LT		 3095 error = PTR_ERR(new_dentry);
3096 if (IS_ERR(new_dentry))
3097 goto exit4;
3098 /* target should not be an ancestor of source */
3099 error = -ENOTEMPTY;
3100 if (new_dentry == trap)
3101 goto exit5;
3102
9079b1eb
DH		 3103 error = mnt_want_write(oldnd.path.mnt);
3104 if (error)
3105 goto exit5;
be6d3e56
KT		 3106 error = security_path_rename(&oldnd.path, old_dentry,
3107 &newnd.path, new_dentry);
3108 if (error)
3109 goto exit6;
1da177e4
LT		 3110 error = vfs_rename(old_dir->d_inode, old_dentry,
3111 new_dir->d_inode, new_dentry);
be6d3e56 3112exit6:
9079b1eb 3113 mnt_drop_write(oldnd.path.mnt);
1da177e4
LT		 3114exit5:
3115 dput(new_dentry);
3116exit4:
3117 dput(old_dentry);
3118exit3:
3119 unlock_rename(new_dir, old_dir);
3120exit2:
1d957f9b 3121 path_put(&newnd.path);
2ad94ae6 3122 putname(to);
1da177e4 3123exit1:
1d957f9b 3124 path_put(&oldnd.path);
1da177e4 3125 putname(from);
2ad94ae6 3126exit:
1da177e4
LT		 3127 return error;
3128}
3129
a26eab24 3130SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 3131{
3132 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3133}
3134
1da177e4
LT		 3135int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3136{
3137 int len;
3138
3139 len = PTR_ERR(link);
3140 if (IS_ERR(link))
3141 goto out;
3142
3143 len = strlen(link);
3144 if (len > (unsigned) buflen)
3145 len = buflen;
3146 if (copy_to_user(buffer, link, len))
3147 len = -EFAULT;
3148out:
3149 return len;
3150}
3151
3152/*
3153 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3154 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3155 * using) it for any given inode is up to filesystem.
3156 */
3157int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3158{
3159 struct nameidata nd;
cc314eef 3160 void *cookie;
694a1764 3161 int res;
cc314eef 3162
1da177e4 3163 nd.depth = 0;
cc314eef 3164 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
694a1764
MS		 3165 if (IS_ERR(cookie))
3166 return PTR_ERR(cookie);
3167
3168 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3169 if (dentry->d_inode->i_op->put_link)
3170 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3171 return res;
1da177e4
LT		 3172}
3173
3174int vfs_follow_link(struct nameidata *nd, const char *link)
3175{
3176 return __vfs_follow_link(nd, link);
3177}
3178
3179/* get the link contents into pagecache */
3180static char *page_getlink(struct dentry * dentry, struct page **ppage)
3181{
ebd09abb
DG		 3182 char *kaddr;
3183 struct page *page;
1da177e4 3184 struct address_space *mapping = dentry->d_inode->i_mapping;
090d2b18 3185 page = read_mapping_page(mapping, 0, NULL);
1da177e4 3186 if (IS_ERR(page))
6fe6900e 3187 return (char*)page;
1da177e4 3188 *ppage = page;
ebd09abb
DG		 3189 kaddr = kmap(page);
3190 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3191 return kaddr;
1da177e4
LT		 3192}
3193
3194int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3195{
3196 struct page *page = NULL;
3197 char *s = page_getlink(dentry, &page);
3198 int res = vfs_readlink(dentry,buffer,buflen,s);
3199 if (page) {
3200 kunmap(page);
3201 page_cache_release(page);
3202 }
3203 return res;
3204}
3205
cc314eef 3206void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
1da177e4 3207{
cc314eef 3208 struct page *page = NULL;
1da177e4 3209 nd_set_link(nd, page_getlink(dentry, &page));
cc314eef 3210 return page;
1da177e4
LT		 3211}
3212
cc314eef 3213void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1da177e4 3214{
cc314eef
LT		 3215 struct page *page = cookie;
3216
3217 if (page) {
1da177e4
LT		 3218 kunmap(page);
3219 page_cache_release(page);
1da177e4
LT		 3220 }
3221}
3222
54566b2c
NP		 3223/*
3224 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3225 */
3226int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
1da177e4
LT		 3227{
3228 struct address_space *mapping = inode->i_mapping;
0adb25d2 3229 struct page *page;
afddba49 3230 void *fsdata;
beb497ab 3231 int err;
1da177e4 3232 char *kaddr;
54566b2c
NP		 3233 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3234 if (nofs)
3235 flags |= AOP_FLAG_NOFS;
1da177e4 3236
7e53cac4 3237retry:
afddba49 3238 err = pagecache_write_begin(NULL, mapping, 0, len-1,
54566b2c 3239 flags, &page, &fsdata);
1da177e4 3240 if (err)
afddba49
NP		 3241 goto fail;
3242
1da177e4
LT		 3243 kaddr = kmap_atomic(page, KM_USER0);
3244 memcpy(kaddr, symname, len-1);
3245 kunmap_atomic(kaddr, KM_USER0);
afddba49
NP		 3246
3247 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3248 page, fsdata);
1da177e4
LT		 3249 if (err < 0)
3250 goto fail;
afddba49
NP		 3251 if (err < len-1)
3252 goto retry;
3253
1da177e4
LT		 3254 mark_inode_dirty(inode);
3255 return 0;
1da177e4
LT		 3256fail:
3257 return err;
3258}
3259
0adb25d2
KK		 3260int page_symlink(struct inode *inode, const char *symname, int len)
3261{
3262 return __page_symlink(inode, symname, len,
54566b2c 3263 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
0adb25d2
KK		 3264}
3265
92e1d5be 3266const struct inode_operations page_symlink_inode_operations = {
1da177e4
LT		 3267 .readlink = generic_readlink,
3268 .follow_link = page_follow_link_light,
3269 .put_link = page_put_link,
3270};
3271
2d8f3038 3272EXPORT_SYMBOL(user_path_at);
cc53ce53 3273EXPORT_SYMBOL(follow_down_one);
1da177e4
LT		 3274EXPORT_SYMBOL(follow_down);
3275EXPORT_SYMBOL(follow_up);
3276EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3277EXPORT_SYMBOL(getname);
3278EXPORT_SYMBOL(lock_rename);
1da177e4
LT		 3279EXPORT_SYMBOL(lookup_one_len);
3280EXPORT_SYMBOL(page_follow_link_light);
3281EXPORT_SYMBOL(page_put_link);
3282EXPORT_SYMBOL(page_readlink);
0adb25d2 3283EXPORT_SYMBOL(__page_symlink);
1da177e4
LT		 3284EXPORT_SYMBOL(page_symlink);
3285EXPORT_SYMBOL(page_symlink_inode_operations);
c9c6cac0 3286EXPORT_SYMBOL(kern_path_parent);
d1811465 3287EXPORT_SYMBOL(kern_path);
16f18200 3288EXPORT_SYMBOL(vfs_path_lookup);
f419a2e3 3289EXPORT_SYMBOL(inode_permission);
1da177e4
LT		 3290EXPORT_SYMBOL(unlock_rename);
3291EXPORT_SYMBOL(vfs_create);
3292EXPORT_SYMBOL(vfs_follow_link);
3293EXPORT_SYMBOL(vfs_link);
3294EXPORT_SYMBOL(vfs_mkdir);
3295EXPORT_SYMBOL(vfs_mknod);
3296EXPORT_SYMBOL(generic_permission);
3297EXPORT_SYMBOL(vfs_readlink);
3298EXPORT_SYMBOL(vfs_rename);
3299EXPORT_SYMBOL(vfs_rmdir);
3300EXPORT_SYMBOL(vfs_symlink);
3301EXPORT_SYMBOL(vfs_unlink);
3302EXPORT_SYMBOL(dentry_unhash);
3303EXPORT_SYMBOL(generic_readlink);
kernel source for telekom puls (alcatel/mediatek)