projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Trim excessive arguments of follow_mount_rcu()
[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 */
b74c79e9
NP		 179static int acl_permission_check(struct inode *inode, int mask, unsigned int flags,
180 int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
1da177e4 181{
26cf46be 182 unsigned int mode = inode->i_mode;
1da177e4 183
e6305c43
AV		 184 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
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 {
192 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
b74c79e9
NP		 193 int error = check_acl(inode, mask, flags);
194 if (error != -EAGAIN)
195 return error;
1da177e4
LT		 196 }
197
198 if (in_group_p(inode->i_gid))
199 mode >>= 3;
200 }
201
e795b717 202other_perms:
1da177e4
LT		 203 /*
204 * If the DACs are ok we don't need any capability check.
205 */
e6305c43 206 if ((mask & ~mode) == 0)
1da177e4 207 return 0;
5909ccaa
LT		 208 return -EACCES;
209}
210
211/**
b74c79e9 212 * generic_permission - check for access rights on a Posix-like filesystem
5909ccaa
LT		 213 * @inode: inode to check access rights for
214 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
215 * @check_acl: optional callback to check for Posix ACLs
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 */
b74c79e9
NP		 227int generic_permission(struct inode *inode, int mask, unsigned int flags,
228 int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
5909ccaa
LT		 229{
230 int ret;
231
232 /*
233 * Do the basic POSIX ACL permission checks.
234 */
b74c79e9 235 ret = acl_permission_check(inode, mask, flags, check_acl);
5909ccaa
LT		 236 if (ret != -EACCES)
237 return ret;
1da177e4 238
1da177e4
LT		 239 /*
240 * Read/write DACs are always overridable.
241 * Executable DACs are overridable if at least one exec bit is set.
242 */
f696a365 243 if (!(mask & MAY_EXEC) || execute_ok(inode))
e795b717 244 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
1da177e4
LT		 245 return 0;
246
247 /*
248 * Searching includes executable on directories, else just read.
249 */
7ea66001 250 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
1da177e4 251 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
e795b717 252 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
1da177e4
LT		 253 return 0;
254
255 return -EACCES;
256}
257
cb23beb5
CH		 258/**
259 * inode_permission - check for access rights to a given inode
260 * @inode: inode to check permission on
261 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
262 *
263 * Used to check for read/write/execute permissions on an inode.
264 * We use "fsuid" for this, letting us set arbitrary permissions
265 * for filesystem access without changing the "normal" uids which
266 * are used for other things.
267 */
f419a2e3 268int inode_permission(struct inode *inode, int mask)
1da177e4 269{
e6305c43 270 int retval;
1da177e4
LT		 271
272 if (mask & MAY_WRITE) {
22590e41 273 umode_t mode = inode->i_mode;
1da177e4
LT		 274
275 /*
276 * Nobody gets write access to a read-only fs.
277 */
278 if (IS_RDONLY(inode) &&
279 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
280 return -EROFS;
281
282 /*
283 * Nobody gets write access to an immutable file.
284 */
285 if (IS_IMMUTABLE(inode))
286 return -EACCES;
287 }
288
acfa4380 289 if (inode->i_op->permission)
b74c79e9 290 retval = inode->i_op->permission(inode, mask, 0);
f696a365 291 else
b74c79e9
NP		 292 retval = generic_permission(inode, mask, 0,
293 inode->i_op->check_acl);
f696a365 294
1da177e4
LT		 295 if (retval)
296 return retval;
297
08ce5f16
SH		 298 retval = devcgroup_inode_permission(inode, mask);
299 if (retval)
300 return retval;
301
d09ca739 302 return security_inode_permission(inode, mask);
1da177e4
LT		 303}
304
8c744fb8
CH		 305/**
306 * file_permission - check for additional access rights to a given file
307 * @file: file to check access rights for
308 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
309 *
310 * Used to check for read/write/execute permissions on an already opened
311 * file.
312 *
313 * Note:
314 * Do not use this function in new code. All access checks should
cb23beb5 315 * be done using inode_permission().
8c744fb8
CH		 316 */
317int file_permission(struct file *file, int mask)
318{
f419a2e3 319 return inode_permission(file->f_path.dentry->d_inode, mask);
8c744fb8
CH		 320}
321
1da177e4
LT		 322/*
323 * get_write_access() gets write permission for a file.
324 * put_write_access() releases this write permission.
325 * This is used for regular files.
326 * We cannot support write (and maybe mmap read-write shared) accesses and
327 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
328 * can have the following values:
329 * 0: no writers, no VM_DENYWRITE mappings
330 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
331 * > 0: (i_writecount) users are writing to the file.
332 *
333 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
334 * except for the cases where we don't hold i_writecount yet. Then we need to
335 * use {get,deny}_write_access() - these functions check the sign and refuse
336 * to do the change if sign is wrong. Exclusion between them is provided by
337 * the inode->i_lock spinlock.
338 */
339
340int get_write_access(struct inode * inode)
341{
342 spin_lock(&inode->i_lock);
343 if (atomic_read(&inode->i_writecount) < 0) {
344 spin_unlock(&inode->i_lock);
345 return -ETXTBSY;
346 }
347 atomic_inc(&inode->i_writecount);
348 spin_unlock(&inode->i_lock);
349
350 return 0;
351}
352
353int deny_write_access(struct file * file)
354{
0f7fc9e4 355 struct inode *inode = file->f_path.dentry->d_inode;
1da177e4
LT		 356
357 spin_lock(&inode->i_lock);
358 if (atomic_read(&inode->i_writecount) > 0) {
359 spin_unlock(&inode->i_lock);
360 return -ETXTBSY;
361 }
362 atomic_dec(&inode->i_writecount);
363 spin_unlock(&inode->i_lock);
364
365 return 0;
366}
367
5dd784d0
JB		 368/**
369 * path_get - get a reference to a path
370 * @path: path to get the reference to
371 *
372 * Given a path increment the reference count to the dentry and the vfsmount.
373 */
374void path_get(struct path *path)
375{
376 mntget(path->mnt);
377 dget(path->dentry);
378}
379EXPORT_SYMBOL(path_get);
380
1d957f9b
JB		 381/**
382 * path_put - put a reference to a path
383 * @path: path to put the reference to
384 *
385 * Given a path decrement the reference count to the dentry and the vfsmount.
386 */
387void path_put(struct path *path)
1da177e4 388{
1d957f9b
JB		 389 dput(path->dentry);
390 mntput(path->mnt);
1da177e4 391}
1d957f9b 392EXPORT_SYMBOL(path_put);
1da177e4 393
19660af7 394/*
31e6b01f 395 * Path walking has 2 modes, rcu-walk and ref-walk (see
19660af7
AV		 396 * Documentation/filesystems/path-lookup.txt). In situations when we can't
397 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
398 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
399 * mode. Refcounts are grabbed at the last known good point before rcu-walk
400 * got stuck, so ref-walk may continue from there. If this is not successful
401 * (eg. a seqcount has changed), then failure is returned and it's up to caller
402 * to restart the path walk from the beginning in ref-walk mode.
31e6b01f 403 */
31e6b01f
NP		 404
405/**
19660af7
AV		 406 * unlazy_walk - try to switch to ref-walk mode.
407 * @nd: nameidata pathwalk data
408 * @dentry: child of nd->path.dentry or NULL
39191628 409 * Returns: 0 on success, -ECHILD on failure
31e6b01f 410 *
19660af7
AV		 411 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
412 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
413 * @nd or NULL. Must be called from rcu-walk context.
31e6b01f 414 */
19660af7 415static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
31e6b01f
NP		 416{
417 struct fs_struct *fs = current->fs;
418 struct dentry *parent = nd->path.dentry;
5b6ca027 419 int want_root = 0;
31e6b01f
NP		 420
421 BUG_ON(!(nd->flags & LOOKUP_RCU));
5b6ca027
AV		 422 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
423 want_root = 1;
31e6b01f
NP		 424 spin_lock(&fs->lock);
425 if (nd->root.mnt != fs->root.mnt ||
426 nd->root.dentry != fs->root.dentry)
427 goto err_root;
428 }
429 spin_lock(&parent->d_lock);
19660af7
AV		 430 if (!dentry) {
431 if (!__d_rcu_to_refcount(parent, nd->seq))
432 goto err_parent;
433 BUG_ON(nd->inode != parent->d_inode);
434 } else {
435 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
436 if (!__d_rcu_to_refcount(dentry, nd->seq))
437 goto err_child;
438 /*
439 * If the sequence check on the child dentry passed, then
440 * the child has not been removed from its parent. This
441 * means the parent dentry must be valid and able to take
442 * a reference at this point.
443 */
444 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
445 BUG_ON(!parent->d_count);
446 parent->d_count++;
447 spin_unlock(&dentry->d_lock);
448 }
31e6b01f 449 spin_unlock(&parent->d_lock);
5b6ca027 450 if (want_root) {
31e6b01f
NP		 451 path_get(&nd->root);
452 spin_unlock(&fs->lock);
453 }
454 mntget(nd->path.mnt);
455
456 rcu_read_unlock();
457 br_read_unlock(vfsmount_lock);
458 nd->flags &= ~LOOKUP_RCU;
459 return 0;
19660af7
AV		 460
461err_child:
31e6b01f 462 spin_unlock(&dentry->d_lock);
19660af7 463err_parent:
31e6b01f
NP		 464 spin_unlock(&parent->d_lock);
465err_root:
5b6ca027 466 if (want_root)
31e6b01f
NP		 467 spin_unlock(&fs->lock);
468 return -ECHILD;
469}
470
834f2a4a
TM		 471/**
472 * release_open_intent - free up open intent resources
473 * @nd: pointer to nameidata
474 */
475void release_open_intent(struct nameidata *nd)
476{
2dab5974
LT		 477 struct file *file = nd->intent.open.file;
478
479 if (file && !IS_ERR(file)) {
480 if (file->f_path.dentry == NULL)
481 put_filp(file);
482 else
483 fput(file);
484 }
834f2a4a
TM		 485}
486
f60aef7e 487static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
34286d66 488{
f60aef7e 489 return dentry->d_op->d_revalidate(dentry, nd);
34286d66
NP		 490}
491
f5e1c1c1 492static struct dentry *
bcdc5e01
IK		 493do_revalidate(struct dentry *dentry, struct nameidata *nd)
494{
f5e1c1c1 495 int status = d_revalidate(dentry, nd);
bcdc5e01
IK		 496 if (unlikely(status <= 0)) {
497 /*
498 * The dentry failed validation.
499 * If d_revalidate returned 0 attempt to invalidate
500 * the dentry otherwise d_revalidate is asking us
501 * to return a fail status.
502 */
34286d66 503 if (status < 0) {
f5e1c1c1 504 dput(dentry);
34286d66 505 dentry = ERR_PTR(status);
f5e1c1c1
AV		 506 } else if (!d_invalidate(dentry)) {
507 dput(dentry);
508 dentry = NULL;
bcdc5e01
IK		 509 }
510 }
511 return dentry;
512}
513
9f1fafee
AV		 514/**
515 * complete_walk - successful completion of path walk
516 * @nd: pointer nameidata
39159de2 517 *
9f1fafee
AV		 518 * If we had been in RCU mode, drop out of it and legitimize nd->path.
519 * Revalidate the final result, unless we'd already done that during
520 * the path walk or the filesystem doesn't ask for it. Return 0 on
521 * success, -error on failure. In case of failure caller does not
522 * need to drop nd->path.
39159de2 523 */
9f1fafee 524static int complete_walk(struct nameidata *nd)
39159de2 525{
16c2cd71 526 struct dentry *dentry = nd->path.dentry;
39159de2 527 int status;
39159de2 528
9f1fafee
AV		 529 if (nd->flags & LOOKUP_RCU) {
530 nd->flags &= ~LOOKUP_RCU;
531 if (!(nd->flags & LOOKUP_ROOT))
532 nd->root.mnt = NULL;
533 spin_lock(&dentry->d_lock);
534 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
535 spin_unlock(&dentry->d_lock);
536 rcu_read_unlock();
537 br_read_unlock(vfsmount_lock);
538 return -ECHILD;
539 }
540 BUG_ON(nd->inode != dentry->d_inode);
541 spin_unlock(&dentry->d_lock);
542 mntget(nd->path.mnt);
543 rcu_read_unlock();
544 br_read_unlock(vfsmount_lock);
545 }
546
16c2cd71
AV		 547 if (likely(!(nd->flags & LOOKUP_JUMPED)))
548 return 0;
549
550 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
39159de2
JL		 551 return 0;
552
16c2cd71
AV		 553 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
554 return 0;
555
556 /* Note: we do not d_invalidate() */
34286d66 557 status = d_revalidate(dentry, nd);
39159de2
JL		 558 if (status > 0)
559 return 0;
560
16c2cd71 561 if (!status)
39159de2 562 status = -ESTALE;
16c2cd71 563
9f1fafee 564 path_put(&nd->path);
39159de2
JL		 565 return status;
566}
567
1da177e4 568/*
b75b5086
AV		 569 * Short-cut version of permission(), for calling on directories
570 * during pathname resolution. Combines parts of permission()
571 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
1da177e4
LT		 572 *
573 * If appropriate, check DAC only. If not appropriate, or
b75b5086 574 * short-cut DAC fails, then call ->permission() to do more
1da177e4
LT		 575 * complete permission check.
576 */
b74c79e9 577static inline int exec_permission(struct inode *inode, unsigned int flags)
1da177e4 578{
5909ccaa 579 int ret;
e795b717 580 struct user_namespace *ns = inode_userns(inode);
1da177e4 581
cb9179ea 582 if (inode->i_op->permission) {
b74c79e9
NP		 583 ret = inode->i_op->permission(inode, MAY_EXEC, flags);
584 } else {
585 ret = acl_permission_check(inode, MAY_EXEC, flags,
586 inode->i_op->check_acl);
cb9179ea 587 }
b74c79e9 588 if (likely(!ret))
1da177e4 589 goto ok;
b74c79e9 590 if (ret == -ECHILD)
31e6b01f 591 return ret;
1da177e4 592
e795b717
SH		 593 if (ns_capable(ns, CAP_DAC_OVERRIDE) ||
594 ns_capable(ns, CAP_DAC_READ_SEARCH))
1da177e4
LT		 595 goto ok;
596
5909ccaa 597 return ret;
1da177e4 598ok:
b74c79e9 599 return security_inode_exec_permission(inode, flags);
1da177e4
LT		 600}
601
2a737871
AV		 602static __always_inline void set_root(struct nameidata *nd)
603{
f7ad3c6b
MS		 604 if (!nd->root.mnt)
605 get_fs_root(current->fs, &nd->root);
2a737871
AV		 606}
607
6de88d72
AV		 608static int link_path_walk(const char *, struct nameidata *);
609
31e6b01f
NP		 610static __always_inline void set_root_rcu(struct nameidata *nd)
611{
612 if (!nd->root.mnt) {
613 struct fs_struct *fs = current->fs;
c28cc364
NP		 614 unsigned seq;
615
616 do {
617 seq = read_seqcount_begin(&fs->seq);
618 nd->root = fs->root;
c1530019 619 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
c28cc364 620 } while (read_seqcount_retry(&fs->seq, seq));
31e6b01f
NP		 621 }
622}
623
f1662356 624static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
1da177e4 625{
31e6b01f
NP		 626 int ret;
627
1da177e4
LT		 628 if (IS_ERR(link))
629 goto fail;
630
631 if (*link == '/') {
2a737871 632 set_root(nd);
1d957f9b 633 path_put(&nd->path);
2a737871
AV		 634 nd->path = nd->root;
635 path_get(&nd->root);
16c2cd71 636 nd->flags |= LOOKUP_JUMPED;
1da177e4 637 }
31e6b01f 638 nd->inode = nd->path.dentry->d_inode;
b4091d5f 639
31e6b01f
NP		 640 ret = link_path_walk(link, nd);
641 return ret;
1da177e4 642fail:
1d957f9b 643 path_put(&nd->path);
1da177e4
LT		 644 return PTR_ERR(link);
645}
646
1d957f9b 647static void path_put_conditional(struct path *path, struct nameidata *nd)
051d3812
IK		 648{
649 dput(path->dentry);
4ac91378 650 if (path->mnt != nd->path.mnt)
051d3812
IK		 651 mntput(path->mnt);
652}
653
7b9337aa
NP		 654static inline void path_to_nameidata(const struct path *path,
655 struct nameidata *nd)
051d3812 656{
31e6b01f
NP		 657 if (!(nd->flags & LOOKUP_RCU)) {
658 dput(nd->path.dentry);
659 if (nd->path.mnt != path->mnt)
660 mntput(nd->path.mnt);
9a229683 661 }
31e6b01f 662 nd->path.mnt = path->mnt;
4ac91378 663 nd->path.dentry = path->dentry;
051d3812
IK		 664}
665
574197e0
AV		 666static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
667{
668 struct inode *inode = link->dentry->d_inode;
669 if (!IS_ERR(cookie) && inode->i_op->put_link)
670 inode->i_op->put_link(link->dentry, nd, cookie);
671 path_put(link);
672}
673
def4af30 674static __always_inline int
574197e0 675follow_link(struct path *link, struct nameidata *nd, void **p)
1da177e4
LT		 676{
677 int error;
7b9337aa 678 struct dentry *dentry = link->dentry;
1da177e4 679
844a3917
AV		 680 BUG_ON(nd->flags & LOOKUP_RCU);
681
0e794589
AV		 682 if (link->mnt == nd->path.mnt)
683 mntget(link->mnt);
684
574197e0
AV		 685 if (unlikely(current->total_link_count >= 40)) {
686 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
574197e0
AV		 687 path_put(&nd->path);
688 return -ELOOP;
689 }
690 cond_resched();
691 current->total_link_count++;
692
7b9337aa 693 touch_atime(link->mnt, dentry);
1da177e4 694 nd_set_link(nd, NULL);
cd4e91d3 695
36f3b4f6
AV		 696 error = security_inode_follow_link(link->dentry, nd);
697 if (error) {
698 *p = ERR_PTR(error); /* no ->put_link(), please */
699 path_put(&nd->path);
700 return error;
701 }
702
86acdca1 703 nd->last_type = LAST_BIND;
def4af30
AV		 704 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
705 error = PTR_ERR(*p);
706 if (!IS_ERR(*p)) {
1da177e4 707 char *s = nd_get_link(nd);
cc314eef 708 error = 0;
1da177e4
LT		 709 if (s)
710 error = __vfs_follow_link(nd, s);
bcda7652 711 else if (nd->last_type == LAST_BIND) {
16c2cd71 712 nd->flags |= LOOKUP_JUMPED;
b21041d0
AV		 713 nd->inode = nd->path.dentry->d_inode;
714 if (nd->inode->i_op->follow_link) {
bcda7652
AV		 715 /* stepped on a _really_ weird one */
716 path_put(&nd->path);
717 error = -ELOOP;
718 }
719 }
1da177e4 720 }
1da177e4
LT		 721 return error;
722}
723
31e6b01f
NP		 724static int follow_up_rcu(struct path *path)
725{
726 struct vfsmount *parent;
727 struct dentry *mountpoint;
728
729 parent = path->mnt->mnt_parent;
730 if (parent == path->mnt)
731 return 0;
732 mountpoint = path->mnt->mnt_mountpoint;
733 path->dentry = mountpoint;
734 path->mnt = parent;
735 return 1;
736}
737
bab77ebf 738int follow_up(struct path *path)
1da177e4
LT		 739{
740 struct vfsmount *parent;
741 struct dentry *mountpoint;
99b7db7b
NP		 742
743 br_read_lock(vfsmount_lock);
bab77ebf
AV		 744 parent = path->mnt->mnt_parent;
745 if (parent == path->mnt) {
99b7db7b 746 br_read_unlock(vfsmount_lock);
1da177e4
LT		 747 return 0;
748 }
749 mntget(parent);
bab77ebf 750 mountpoint = dget(path->mnt->mnt_mountpoint);
99b7db7b 751 br_read_unlock(vfsmount_lock);
bab77ebf
AV		 752 dput(path->dentry);
753 path->dentry = mountpoint;
754 mntput(path->mnt);
755 path->mnt = parent;
1da177e4
LT		 756 return 1;
757}
758
b5c84bf6 759/*
9875cf80
DH		 760 * Perform an automount
761 * - return -EISDIR to tell follow_managed() to stop and return the path we
762 * were called with.
1da177e4 763 */
9875cf80
DH		 764static int follow_automount(struct path *path, unsigned flags,
765 bool *need_mntput)
31e6b01f 766{
9875cf80 767 struct vfsmount *mnt;
ea5b778a 768 int err;
9875cf80
DH		 769
770 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
771 return -EREMOTE;
772
6f45b656
DH		 773 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
774 * and this is the terminal part of the path.
775 */
776 if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE))
777 return -EISDIR; /* we actually want to stop here */
778
9875cf80
DH		 779 /* We want to mount if someone is trying to open/create a file of any
780 * type under the mountpoint, wants to traverse through the mountpoint
781 * or wants to open the mounted directory.
782 *
783 * We don't want to mount if someone's just doing a stat and they've
784 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
785 * appended a '/' to the name.
786 */
787 if (!(flags & LOOKUP_FOLLOW) &&
788 !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY |
789 LOOKUP_OPEN | LOOKUP_CREATE)))
790 return -EISDIR;
791
792 current->total_link_count++;
793 if (current->total_link_count >= 40)
794 return -ELOOP;
795
796 mnt = path->dentry->d_op->d_automount(path);
797 if (IS_ERR(mnt)) {
798 /*
799 * The filesystem is allowed to return -EISDIR here to indicate
800 * it doesn't want to automount. For instance, autofs would do
801 * this so that its userspace daemon can mount on this dentry.
802 *
803 * However, we can only permit this if it's a terminal point in
804 * the path being looked up; if it wasn't then the remainder of
805 * the path is inaccessible and we should say so.
806 */
807 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE))
808 return -EREMOTE;
809 return PTR_ERR(mnt);
31e6b01f 810 }
ea5b778a 811
9875cf80
DH		 812 if (!mnt) /* mount collision */
813 return 0;
31e6b01f 814
19a167af 815 err = finish_automount(mnt, path);
9875cf80 816
ea5b778a
DH		 817 switch (err) {
818 case -EBUSY:
819 /* Someone else made a mount here whilst we were busy */
19a167af 820 return 0;
ea5b778a 821 case 0:
ea5b778a
DH		 822 dput(path->dentry);
823 if (*need_mntput)
824 mntput(path->mnt);
825 path->mnt = mnt;
826 path->dentry = dget(mnt->mnt_root);
827 *need_mntput = true;
828 return 0;
19a167af
AV		 829 default:
830 return err;
ea5b778a 831 }
19a167af 832
463ffb2e
AV		 833}
834
9875cf80
DH		 835/*
836 * Handle a dentry that is managed in some way.
cc53ce53 837 * - Flagged for transit management (autofs)
9875cf80
DH		 838 * - Flagged as mountpoint
839 * - Flagged as automount point
840 *
841 * This may only be called in refwalk mode.
842 *
843 * Serialization is taken care of in namespace.c
844 */
845static int follow_managed(struct path *path, unsigned flags)
1da177e4 846{
9875cf80
DH		 847 unsigned managed;
848 bool need_mntput = false;
849 int ret;
850
851 /* Given that we're not holding a lock here, we retain the value in a
852 * local variable for each dentry as we look at it so that we don't see
853 * the components of that value change under us */
854 while (managed = ACCESS_ONCE(path->dentry->d_flags),
855 managed &= DCACHE_MANAGED_DENTRY,
856 unlikely(managed != 0)) {
cc53ce53
DH		 857 /* Allow the filesystem to manage the transit without i_mutex
858 * being held. */
859 if (managed & DCACHE_MANAGE_TRANSIT) {
860 BUG_ON(!path->dentry->d_op);
861 BUG_ON(!path->dentry->d_op->d_manage);
1aed3e42 862 ret = path->dentry->d_op->d_manage(path->dentry, false);
cc53ce53
DH		 863 if (ret < 0)
864 return ret == -EISDIR ? 0 : ret;
865 }
866
9875cf80
DH		 867 /* Transit to a mounted filesystem. */
868 if (managed & DCACHE_MOUNTED) {
869 struct vfsmount *mounted = lookup_mnt(path);
870 if (mounted) {
871 dput(path->dentry);
872 if (need_mntput)
873 mntput(path->mnt);
874 path->mnt = mounted;
875 path->dentry = dget(mounted->mnt_root);
876 need_mntput = true;
877 continue;
878 }
879
880 /* Something is mounted on this dentry in another
881 * namespace and/or whatever was mounted there in this
882 * namespace got unmounted before we managed to get the
883 * vfsmount_lock */
884 }
885
886 /* Handle an automount point */
887 if (managed & DCACHE_NEED_AUTOMOUNT) {
888 ret = follow_automount(path, flags, &need_mntput);
889 if (ret < 0)
890 return ret == -EISDIR ? 0 : ret;
891 continue;
892 }
893
894 /* We didn't change the current path point */
895 break;
1da177e4 896 }
9875cf80 897 return 0;
1da177e4
LT		 898}
899
cc53ce53 900int follow_down_one(struct path *path)
1da177e4
LT		 901{
902 struct vfsmount *mounted;
903
1c755af4 904 mounted = lookup_mnt(path);
1da177e4 905 if (mounted) {
9393bd07
AV		 906 dput(path->dentry);
907 mntput(path->mnt);
908 path->mnt = mounted;
909 path->dentry = dget(mounted->mnt_root);
1da177e4
LT		 910 return 1;
911 }
912 return 0;
913}
914
62a7375e
IK		 915static inline bool managed_dentry_might_block(struct dentry *dentry)
916{
917 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
918 dentry->d_op->d_manage(dentry, true) < 0);
919}
920
9875cf80 921/*
287548e4
AV		 922 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
923 * we meet a managed dentry that would need blocking.
9875cf80
DH		 924 */
925static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
287548e4 926 struct inode **inode)
9875cf80 927{
62a7375e 928 for (;;) {
9875cf80 929 struct vfsmount *mounted;
62a7375e
IK		 930 /*
931 * Don't forget we might have a non-mountpoint managed dentry
932 * that wants to block transit.
933 */
934 *inode = path->dentry->d_inode;
287548e4 935 if (unlikely(managed_dentry_might_block(path->dentry)))
ab90911f 936 return false;
62a7375e
IK		 937
938 if (!d_mountpoint(path->dentry))
939 break;
940
9875cf80
DH		 941 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
942 if (!mounted)
943 break;
944 path->mnt = mounted;
945 path->dentry = mounted->mnt_root;
946 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
9875cf80
DH		 947 }
948
949 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
287548e4 950 return false;
9875cf80
DH		 951 return true;
952}
953
dea39376 954static void follow_mount_rcu(struct nameidata *nd)
287548e4 955{
dea39376 956 while (d_mountpoint(nd->path.dentry)) {
287548e4 957 struct vfsmount *mounted;
dea39376 958 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
287548e4
AV		 959 if (!mounted)
960 break;
dea39376
AV		 961 nd->path.mnt = mounted;
962 nd->path.dentry = mounted->mnt_root;
963 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
287548e4
AV		 964 }
965}
966
31e6b01f
NP		 967static int follow_dotdot_rcu(struct nameidata *nd)
968{
31e6b01f
NP		 969 set_root_rcu(nd);
970
9875cf80 971 while (1) {
31e6b01f
NP		 972 if (nd->path.dentry == nd->root.dentry &&
973 nd->path.mnt == nd->root.mnt) {
974 break;
975 }
976 if (nd->path.dentry != nd->path.mnt->mnt_root) {
977 struct dentry *old = nd->path.dentry;
978 struct dentry *parent = old->d_parent;
979 unsigned seq;
980
981 seq = read_seqcount_begin(&parent->d_seq);
982 if (read_seqcount_retry(&old->d_seq, nd->seq))
ef7562d5 983 goto failed;
31e6b01f
NP		 984 nd->path.dentry = parent;
985 nd->seq = seq;
986 break;
987 }
988 if (!follow_up_rcu(&nd->path))
989 break;
990 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
31e6b01f 991 }
dea39376
AV		 992 follow_mount_rcu(nd);
993 nd->inode = nd->path.dentry->d_inode;
31e6b01f 994 return 0;
ef7562d5
AV		 995
996failed:
997 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 998 if (!(nd->flags & LOOKUP_ROOT))
999 nd->root.mnt = NULL;
ef7562d5
AV		 1000 rcu_read_unlock();
1001 br_read_unlock(vfsmount_lock);
1002 return -ECHILD;
31e6b01f
NP		 1003}
1004
cc53ce53
DH		 1005/*
1006 * Follow down to the covering mount currently visible to userspace. At each
1007 * point, the filesystem owning that dentry may be queried as to whether the
1008 * caller is permitted to proceed or not.
1009 *
1010 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1011 * being true).
1012 */
7cc90cc3 1013int follow_down(struct path *path)
cc53ce53
DH		 1014{
1015 unsigned managed;
1016 int ret;
1017
1018 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1019 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1020 /* Allow the filesystem to manage the transit without i_mutex
1021 * being held.
1022 *
1023 * We indicate to the filesystem if someone is trying to mount
1024 * something here. This gives autofs the chance to deny anyone
1025 * other than its daemon the right to mount on its
1026 * superstructure.
1027 *
1028 * The filesystem may sleep at this point.
1029 */
1030 if (managed & DCACHE_MANAGE_TRANSIT) {
1031 BUG_ON(!path->dentry->d_op);
1032 BUG_ON(!path->dentry->d_op->d_manage);
ab90911f 1033 ret = path->dentry->d_op->d_manage(
1aed3e42 1034 path->dentry, false);
cc53ce53
DH		 1035 if (ret < 0)
1036 return ret == -EISDIR ? 0 : ret;
1037 }
1038
1039 /* Transit to a mounted filesystem. */
1040 if (managed & DCACHE_MOUNTED) {
1041 struct vfsmount *mounted = lookup_mnt(path);
1042 if (!mounted)
1043 break;
1044 dput(path->dentry);
1045 mntput(path->mnt);
1046 path->mnt = mounted;
1047 path->dentry = dget(mounted->mnt_root);
1048 continue;
1049 }
1050
1051 /* Don't handle automount points here */
1052 break;
1053 }
1054 return 0;
1055}
1056
9875cf80
DH		 1057/*
1058 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1059 */
1060static void follow_mount(struct path *path)
1061{
1062 while (d_mountpoint(path->dentry)) {
1063 struct vfsmount *mounted = lookup_mnt(path);
1064 if (!mounted)
1065 break;
1066 dput(path->dentry);
1067 mntput(path->mnt);
1068 path->mnt = mounted;
1069 path->dentry = dget(mounted->mnt_root);
1070 }
1071}
1072
31e6b01f 1073static void follow_dotdot(struct nameidata *nd)
1da177e4 1074{
2a737871 1075 set_root(nd);
e518ddb7 1076
1da177e4 1077 while(1) {
4ac91378 1078 struct dentry *old = nd->path.dentry;
1da177e4 1079
2a737871
AV		 1080 if (nd->path.dentry == nd->root.dentry &&
1081 nd->path.mnt == nd->root.mnt) {
1da177e4
LT		 1082 break;
1083 }
4ac91378 1084 if (nd->path.dentry != nd->path.mnt->mnt_root) {
3088dd70
AV		 1085 /* rare case of legitimate dget_parent()... */
1086 nd->path.dentry = dget_parent(nd->path.dentry);
1da177e4
LT		 1087 dput(old);
1088 break;
1089 }
3088dd70 1090 if (!follow_up(&nd->path))
1da177e4 1091 break;
1da177e4 1092 }
79ed0226 1093 follow_mount(&nd->path);
31e6b01f 1094 nd->inode = nd->path.dentry->d_inode;
1da177e4
LT		 1095}
1096
baa03890
NP		 1097/*
1098 * Allocate a dentry with name and parent, and perform a parent
1099 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1100 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1101 * have verified that no child exists while under i_mutex.
1102 */
1103static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1104 struct qstr *name, struct nameidata *nd)
1105{
1106 struct inode *inode = parent->d_inode;
1107 struct dentry *dentry;
1108 struct dentry *old;
1109
1110 /* Don't create child dentry for a dead directory. */
1111 if (unlikely(IS_DEADDIR(inode)))
1112 return ERR_PTR(-ENOENT);
1113
1114 dentry = d_alloc(parent, name);
1115 if (unlikely(!dentry))
1116 return ERR_PTR(-ENOMEM);
1117
1118 old = inode->i_op->lookup(inode, dentry, nd);
1119 if (unlikely(old)) {
1120 dput(dentry);
1121 dentry = old;
1122 }
1123 return dentry;
1124}
1125
1da177e4
LT		 1126/*
1127 * It's more convoluted than I'd like it to be, but... it's still fairly
1128 * small and for now I'd prefer to have fast path as straight as possible.
1129 * It _is_ time-critical.
1130 */
1131static int do_lookup(struct nameidata *nd, struct qstr *name,
31e6b01f 1132 struct path *path, struct inode **inode)
1da177e4 1133{
4ac91378 1134 struct vfsmount *mnt = nd->path.mnt;
31e6b01f 1135 struct dentry *dentry, *parent = nd->path.dentry;
5a18fff2
AV		 1136 int need_reval = 1;
1137 int status = 1;
9875cf80
DH		 1138 int err;
1139
b04f784e
NP		 1140 /*
1141 * Rename seqlock is not required here because in the off chance
1142 * of a false negative due to a concurrent rename, we're going to
1143 * do the non-racy lookup, below.
1144 */
31e6b01f
NP		 1145 if (nd->flags & LOOKUP_RCU) {
1146 unsigned seq;
31e6b01f
NP		 1147 *inode = nd->inode;
1148 dentry = __d_lookup_rcu(parent, name, &seq, inode);
5a18fff2
AV		 1149 if (!dentry)
1150 goto unlazy;
1151
31e6b01f
NP		 1152 /* Memory barrier in read_seqcount_begin of child is enough */
1153 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1154 return -ECHILD;
31e6b01f 1155 nd->seq = seq;
5a18fff2 1156
24643087 1157 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
5a18fff2
AV		 1158 status = d_revalidate(dentry, nd);
1159 if (unlikely(status <= 0)) {
1160 if (status != -ECHILD)
1161 need_reval = 0;
1162 goto unlazy;
1163 }
24643087 1164 }
31e6b01f
NP		 1165 path->mnt = mnt;
1166 path->dentry = dentry;
287548e4 1167 if (likely(__follow_mount_rcu(nd, path, inode)))
9875cf80 1168 return 0;
5a18fff2 1169unlazy:
19660af7
AV		 1170 if (unlazy_walk(nd, dentry))
1171 return -ECHILD;
5a18fff2
AV		 1172 } else {
1173 dentry = __d_lookup(parent, name);
9875cf80 1174 }
5a18fff2
AV		 1175
1176retry:
1177 if (unlikely(!dentry)) {
1178 struct inode *dir = parent->d_inode;
1179 BUG_ON(nd->inode != dir);
1180
1181 mutex_lock(&dir->i_mutex);
1182 dentry = d_lookup(parent, name);
1183 if (likely(!dentry)) {
1184 dentry = d_alloc_and_lookup(parent, name, nd);
1185 if (IS_ERR(dentry)) {
1186 mutex_unlock(&dir->i_mutex);
1187 return PTR_ERR(dentry);
1188 }
1189 /* known good */
1190 need_reval = 0;
1191 status = 1;
1192 }
1193 mutex_unlock(&dir->i_mutex);
1194 }
1195 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1196 status = d_revalidate(dentry, nd);
1197 if (unlikely(status <= 0)) {
1198 if (status < 0) {
1199 dput(dentry);
1200 return status;
1201 }
1202 if (!d_invalidate(dentry)) {
1203 dput(dentry);
1204 dentry = NULL;
1205 need_reval = 1;
1206 goto retry;
1207 }
24643087 1208 }
5a18fff2 1209
9875cf80
DH		 1210 path->mnt = mnt;
1211 path->dentry = dentry;
1212 err = follow_managed(path, nd->flags);
89312214
IK		 1213 if (unlikely(err < 0)) {
1214 path_put_conditional(path, nd);
9875cf80 1215 return err;
89312214 1216 }
9875cf80 1217 *inode = path->dentry->d_inode;
1da177e4 1218 return 0;
1da177e4
LT		 1219}
1220
52094c8a
AV		 1221static inline int may_lookup(struct nameidata *nd)
1222{
1223 if (nd->flags & LOOKUP_RCU) {
1224 int err = exec_permission(nd->inode, IPERM_FLAG_RCU);
1225 if (err != -ECHILD)
1226 return err;
19660af7 1227 if (unlazy_walk(nd, NULL))
52094c8a
AV		 1228 return -ECHILD;
1229 }
1230 return exec_permission(nd->inode, 0);
1231}
1232
9856fa1b
AV		 1233static inline int handle_dots(struct nameidata *nd, int type)
1234{
1235 if (type == LAST_DOTDOT) {
1236 if (nd->flags & LOOKUP_RCU) {
1237 if (follow_dotdot_rcu(nd))
1238 return -ECHILD;
1239 } else
1240 follow_dotdot(nd);
1241 }
1242 return 0;
1243}
1244
951361f9
AV		 1245static void terminate_walk(struct nameidata *nd)
1246{
1247 if (!(nd->flags & LOOKUP_RCU)) {
1248 path_put(&nd->path);
1249 } else {
1250 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 1251 if (!(nd->flags & LOOKUP_ROOT))
1252 nd->root.mnt = NULL;
951361f9
AV		 1253 rcu_read_unlock();
1254 br_read_unlock(vfsmount_lock);
1255 }
1256}
1257
ce57dfc1
AV		 1258static inline int walk_component(struct nameidata *nd, struct path *path,
1259 struct qstr *name, int type, int follow)
1260{
1261 struct inode *inode;
1262 int err;
1263 /*
1264 * "." and ".." are special - ".." especially so because it has
1265 * to be able to know about the current root directory and
1266 * parent relationships.
1267 */
1268 if (unlikely(type != LAST_NORM))
1269 return handle_dots(nd, type);
1270 err = do_lookup(nd, name, path, &inode);
1271 if (unlikely(err)) {
1272 terminate_walk(nd);
1273 return err;
1274 }
1275 if (!inode) {
1276 path_to_nameidata(path, nd);
1277 terminate_walk(nd);
1278 return -ENOENT;
1279 }
1280 if (unlikely(inode->i_op->follow_link) && follow) {
19660af7
AV		 1281 if (nd->flags & LOOKUP_RCU) {
1282 if (unlikely(unlazy_walk(nd, path->dentry))) {
1283 terminate_walk(nd);
1284 return -ECHILD;
1285 }
1286 }
ce57dfc1
AV		 1287 BUG_ON(inode != path->dentry->d_inode);
1288 return 1;
1289 }
1290 path_to_nameidata(path, nd);
1291 nd->inode = inode;
1292 return 0;
1293}
1294
b356379a
AV		 1295/*
1296 * This limits recursive symlink follows to 8, while
1297 * limiting consecutive symlinks to 40.
1298 *
1299 * Without that kind of total limit, nasty chains of consecutive
1300 * symlinks can cause almost arbitrarily long lookups.
1301 */
1302static inline int nested_symlink(struct path *path, struct nameidata *nd)
1303{
1304 int res;
1305
b356379a
AV		 1306 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1307 path_put_conditional(path, nd);
1308 path_put(&nd->path);
1309 return -ELOOP;
1310 }
1a4022f8 1311 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
b356379a
AV		 1312
1313 nd->depth++;
1314 current->link_count++;
1315
1316 do {
1317 struct path link = *path;
1318 void *cookie;
574197e0
AV		 1319
1320 res = follow_link(&link, nd, &cookie);
b356379a
AV		 1321 if (!res)
1322 res = walk_component(nd, path, &nd->last,
1323 nd->last_type, LOOKUP_FOLLOW);
574197e0 1324 put_link(nd, &link, cookie);
b356379a
AV		 1325 } while (res > 0);
1326
1327 current->link_count--;
1328 nd->depth--;
1329 return res;
1330}
1331
1da177e4
LT		 1332/*
1333 * Name resolution.
ea3834d9
PM		 1334 * This is the basic name resolution function, turning a pathname into
1335 * the final dentry. We expect 'base' to be positive and a directory.
1da177e4 1336 *
ea3834d9
PM		 1337 * Returns 0 and nd will have valid dentry and mnt on success.
1338 * Returns error and drops reference to input namei data on failure.
1da177e4 1339 */
6de88d72 1340static int link_path_walk(const char *name, struct nameidata *nd)
1da177e4
LT		 1341{
1342 struct path next;
1da177e4
LT		 1343 int err;
1344 unsigned int lookup_flags = nd->flags;
1345
1346 while (*name=='/')
1347 name++;
1348 if (!*name)
086e183a 1349 return 0;
1da177e4 1350
1da177e4
LT		 1351 /* At this point we know we have a real path component. */
1352 for(;;) {
1353 unsigned long hash;
1354 struct qstr this;
1355 unsigned int c;
fe479a58 1356 int type;
1da177e4 1357
cdce5d6b 1358 nd->flags |= LOOKUP_CONTINUE;
52094c8a
AV		 1359
1360 err = may_lookup(nd);
1da177e4
LT		 1361 if (err)
1362 break;
1363
1364 this.name = name;
1365 c = *(const unsigned char *)name;
1366
1367 hash = init_name_hash();
1368 do {
1369 name++;
1370 hash = partial_name_hash(c, hash);
1371 c = *(const unsigned char *)name;
1372 } while (c && (c != '/'));
1373 this.len = name - (const char *) this.name;
1374 this.hash = end_name_hash(hash);
1375
fe479a58
AV		 1376 type = LAST_NORM;
1377 if (this.name[0] == '.') switch (this.len) {
1378 case 2:
16c2cd71 1379 if (this.name[1] == '.') {
fe479a58 1380 type = LAST_DOTDOT;
16c2cd71
AV		 1381 nd->flags |= LOOKUP_JUMPED;
1382 }
fe479a58
AV		 1383 break;
1384 case 1:
1385 type = LAST_DOT;
1386 }
5a202bcd
AV		 1387 if (likely(type == LAST_NORM)) {
1388 struct dentry *parent = nd->path.dentry;
16c2cd71 1389 nd->flags &= ~LOOKUP_JUMPED;
5a202bcd
AV		 1390 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1391 err = parent->d_op->d_hash(parent, nd->inode,
1392 &this);
1393 if (err < 0)
1394 break;
1395 }
1396 }
fe479a58 1397
1da177e4
LT		 1398 /* remove trailing slashes? */
1399 if (!c)
1400 goto last_component;
1401 while (*++name == '/');
1402 if (!*name)
b356379a 1403 goto last_component;
1da177e4 1404
ce57dfc1
AV		 1405 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1406 if (err < 0)
1407 return err;
1da177e4 1408
ce57dfc1 1409 if (err) {
b356379a 1410 err = nested_symlink(&next, nd);
1da177e4 1411 if (err)
a7472bab 1412 return err;
31e6b01f 1413 }
1da177e4 1414 err = -ENOTDIR;
31e6b01f 1415 if (!nd->inode->i_op->lookup)
1da177e4
LT		 1416 break;
1417 continue;
1418 /* here ends the main loop */
1419
1da177e4 1420last_component:
f55eab82
TM		 1421 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1422 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
b356379a
AV		 1423 nd->last = this;
1424 nd->last_type = type;
086e183a 1425 return 0;
1da177e4 1426 }
951361f9 1427 terminate_walk(nd);
1da177e4
LT		 1428 return err;
1429}
1430
70e9b357
AV		 1431static int path_init(int dfd, const char *name, unsigned int flags,
1432 struct nameidata *nd, struct file **fp)
31e6b01f
NP		 1433{
1434 int retval = 0;
1435 int fput_needed;
1436 struct file *file;
1437
1438 nd->last_type = LAST_ROOT; /* if there are only slashes... */
16c2cd71 1439 nd->flags = flags | LOOKUP_JUMPED;
31e6b01f 1440 nd->depth = 0;
5b6ca027
AV		 1441 if (flags & LOOKUP_ROOT) {
1442 struct inode *inode = nd->root.dentry->d_inode;
73d049a4
AV		 1443 if (*name) {
1444 if (!inode->i_op->lookup)
1445 return -ENOTDIR;
1446 retval = inode_permission(inode, MAY_EXEC);
1447 if (retval)
1448 return retval;
1449 }
5b6ca027
AV		 1450 nd->path = nd->root;
1451 nd->inode = inode;
1452 if (flags & LOOKUP_RCU) {
1453 br_read_lock(vfsmount_lock);
1454 rcu_read_lock();
1455 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1456 } else {
1457 path_get(&nd->path);
1458 }
1459 return 0;
1460 }
1461
31e6b01f 1462 nd->root.mnt = NULL;
31e6b01f
NP		 1463
1464 if (*name=='/') {
e41f7d4e
AV		 1465 if (flags & LOOKUP_RCU) {
1466 br_read_lock(vfsmount_lock);
1467 rcu_read_lock();
1468 set_root_rcu(nd);
1469 } else {
1470 set_root(nd);
1471 path_get(&nd->root);
1472 }
1473 nd->path = nd->root;
31e6b01f 1474 } else if (dfd == AT_FDCWD) {
e41f7d4e
AV		 1475 if (flags & LOOKUP_RCU) {
1476 struct fs_struct *fs = current->fs;
1477 unsigned seq;
31e6b01f 1478
e41f7d4e
AV		 1479 br_read_lock(vfsmount_lock);
1480 rcu_read_lock();
c28cc364 1481
e41f7d4e
AV		 1482 do {
1483 seq = read_seqcount_begin(&fs->seq);
1484 nd->path = fs->pwd;
1485 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1486 } while (read_seqcount_retry(&fs->seq, seq));
1487 } else {
1488 get_fs_pwd(current->fs, &nd->path);
1489 }
31e6b01f
NP		 1490 } else {
1491 struct dentry *dentry;
1492
1abf0c71 1493 file = fget_raw_light(dfd, &fput_needed);
31e6b01f
NP		 1494 retval = -EBADF;
1495 if (!file)
1496 goto out_fail;
1497
1498 dentry = file->f_path.dentry;
1499
f52e0c11
AV		 1500 if (*name) {
1501 retval = -ENOTDIR;
1502 if (!S_ISDIR(dentry->d_inode->i_mode))
1503 goto fput_fail;
31e6b01f 1504
f52e0c11
AV		 1505 retval = file_permission(file, MAY_EXEC);
1506 if (retval)
1507 goto fput_fail;
1508 }
31e6b01f
NP		 1509
1510 nd->path = file->f_path;
e41f7d4e
AV		 1511 if (flags & LOOKUP_RCU) {
1512 if (fput_needed)
70e9b357 1513 *fp = file;
e41f7d4e
AV		 1514 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1515 br_read_lock(vfsmount_lock);
1516 rcu_read_lock();
1517 } else {
1518 path_get(&file->f_path);
1519 fput_light(file, fput_needed);
1520 }
31e6b01f 1521 }
31e6b01f 1522
31e6b01f 1523 nd->inode = nd->path.dentry->d_inode;
9b4a9b14 1524 return 0;
2dfdd266 1525
9b4a9b14
AV		 1526fput_fail:
1527 fput_light(file, fput_needed);
1528out_fail:
1529 return retval;
1530}
1531
bd92d7fe
AV		 1532static inline int lookup_last(struct nameidata *nd, struct path *path)
1533{
1534 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1535 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1536
1537 nd->flags &= ~LOOKUP_PARENT;
1538 return walk_component(nd, path, &nd->last, nd->last_type,
1539 nd->flags & LOOKUP_FOLLOW);
1540}
1541
9b4a9b14 1542/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
ee0827cd 1543static int path_lookupat(int dfd, const char *name,
9b4a9b14
AV		 1544 unsigned int flags, struct nameidata *nd)
1545{
70e9b357 1546 struct file *base = NULL;
bd92d7fe
AV		 1547 struct path path;
1548 int err;
31e6b01f
NP		 1549
1550 /*
1551 * Path walking is largely split up into 2 different synchronisation
1552 * schemes, rcu-walk and ref-walk (explained in
1553 * Documentation/filesystems/path-lookup.txt). These share much of the
1554 * path walk code, but some things particularly setup, cleanup, and
1555 * following mounts are sufficiently divergent that functions are
1556 * duplicated. Typically there is a function foo(), and its RCU
1557 * analogue, foo_rcu().
1558 *
1559 * -ECHILD is the error number of choice (just to avoid clashes) that
1560 * is returned if some aspect of an rcu-walk fails. Such an error must
1561 * be handled by restarting a traditional ref-walk (which will always
1562 * be able to complete).
1563 */
bd92d7fe 1564 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
ee0827cd 1565
bd92d7fe
AV		 1566 if (unlikely(err))
1567 return err;
ee0827cd
AV		 1568
1569 current->total_link_count = 0;
bd92d7fe
AV		 1570 err = link_path_walk(name, nd);
1571
1572 if (!err && !(flags & LOOKUP_PARENT)) {
bd92d7fe
AV		 1573 err = lookup_last(nd, &path);
1574 while (err > 0) {
1575 void *cookie;
1576 struct path link = path;
bd92d7fe 1577 nd->flags |= LOOKUP_PARENT;
574197e0 1578 err = follow_link(&link, nd, &cookie);
bd92d7fe
AV		 1579 if (!err)
1580 err = lookup_last(nd, &path);
574197e0 1581 put_link(nd, &link, cookie);
bd92d7fe
AV		 1582 }
1583 }
ee0827cd 1584
9f1fafee
AV		 1585 if (!err)
1586 err = complete_walk(nd);
bd92d7fe
AV		 1587
1588 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1589 if (!nd->inode->i_op->lookup) {
1590 path_put(&nd->path);
bd23a539 1591 err = -ENOTDIR;
bd92d7fe
AV		 1592 }
1593 }
16c2cd71 1594
70e9b357
AV		 1595 if (base)
1596 fput(base);
ee0827cd 1597
5b6ca027 1598 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
2a737871
AV		 1599 path_put(&nd->root);
1600 nd->root.mnt = NULL;
1601 }
bd92d7fe 1602 return err;
ee0827cd 1603}
31e6b01f 1604
ee0827cd
AV		 1605static int do_path_lookup(int dfd, const char *name,
1606 unsigned int flags, struct nameidata *nd)
1607{
1608 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1609 if (unlikely(retval == -ECHILD))
1610 retval = path_lookupat(dfd, name, flags, nd);
1611 if (unlikely(retval == -ESTALE))
1612 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
31e6b01f
NP		 1613
1614 if (likely(!retval)) {
1615 if (unlikely(!audit_dummy_context())) {
1616 if (nd->path.dentry && nd->inode)
1617 audit_inode(name, nd->path.dentry);
1618 }
1619 }
170aa3d0 1620 return retval;
1da177e4
LT		 1621}
1622
c9c6cac0 1623int kern_path_parent(const char *name, struct nameidata *nd)
5590ff0d 1624{
c9c6cac0 1625 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
5590ff0d
UD		 1626}
1627
d1811465
AV		 1628int kern_path(const char *name, unsigned int flags, struct path *path)
1629{
1630 struct nameidata nd;
1631 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1632 if (!res)
1633 *path = nd.path;
1634 return res;
1635}
1636
16f18200
JJS		 1637/**
1638 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1639 * @dentry: pointer to dentry of the base directory
1640 * @mnt: pointer to vfs mount of the base directory
1641 * @name: pointer to file name
1642 * @flags: lookup flags
1643 * @nd: pointer to nameidata
1644 */
1645int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1646 const char *name, unsigned int flags,
1647 struct nameidata *nd)
1648{
5b6ca027
AV		 1649 nd->root.dentry = dentry;
1650 nd->root.mnt = mnt;
1651 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1652 return do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, nd);
16f18200
JJS		 1653}
1654
eead1911
CH		 1655static struct dentry *__lookup_hash(struct qstr *name,
1656 struct dentry *base, struct nameidata *nd)
1da177e4 1657{
81fca444 1658 struct inode *inode = base->d_inode;
057f6c01 1659 struct dentry *dentry;
1da177e4
LT		 1660 int err;
1661
b74c79e9 1662 err = exec_permission(inode, 0);
81fca444
CH		 1663 if (err)
1664 return ERR_PTR(err);
1da177e4 1665
b04f784e
NP		 1666 /*
1667 * Don't bother with __d_lookup: callers are for creat as
1668 * well as unlink, so a lot of the time it would cost
1669 * a double lookup.
6e6b1bd1 1670 */
b04f784e 1671 dentry = d_lookup(base, name);
6e6b1bd1 1672
fb045adb 1673 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE))
6e6b1bd1
AV		 1674 dentry = do_revalidate(dentry, nd);
1675
baa03890
NP		 1676 if (!dentry)
1677 dentry = d_alloc_and_lookup(base, name, nd);
5a202bcd 1678
1da177e4
LT		 1679 return dentry;
1680}
1681
057f6c01
JM		 1682/*
1683 * Restricted form of lookup. Doesn't follow links, single-component only,
1684 * needs parent already locked. Doesn't follow mounts.
1685 * SMP-safe.
1686 */
eead1911 1687static struct dentry *lookup_hash(struct nameidata *nd)
057f6c01 1688{
4ac91378 1689 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1da177e4
LT		 1690}
1691
eead1911 1692/**
a6b91919 1693 * lookup_one_len - filesystem helper to lookup single pathname component
eead1911
CH		 1694 * @name: pathname component to lookup
1695 * @base: base directory to lookup from
1696 * @len: maximum length @len should be interpreted to
1697 *
a6b91919
RD		 1698 * Note that this routine is purely a helper for filesystem usage and should
1699 * not be called by generic code. Also note that by using this function the
eead1911
CH		 1700 * nameidata argument is passed to the filesystem methods and a filesystem
1701 * using this helper needs to be prepared for that.
1702 */
057f6c01
JM		 1703struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1704{
057f6c01 1705 struct qstr this;
6a96ba54
AV		 1706 unsigned long hash;
1707 unsigned int c;
057f6c01 1708
2f9092e1
DW		 1709 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1710
6a96ba54
AV		 1711 this.name = name;
1712 this.len = len;
1713 if (!len)
1714 return ERR_PTR(-EACCES);
1715
1716 hash = init_name_hash();
1717 while (len--) {
1718 c = *(const unsigned char *)name++;
1719 if (c == '/' || c == '\0')
1720 return ERR_PTR(-EACCES);
1721 hash = partial_name_hash(c, hash);
1722 }
1723 this.hash = end_name_hash(hash);
5a202bcd
AV		 1724 /*
1725 * See if the low-level filesystem might want
1726 * to use its own hash..
1727 */
1728 if (base->d_flags & DCACHE_OP_HASH) {
1729 int err = base->d_op->d_hash(base, base->d_inode, &this);
1730 if (err < 0)
1731 return ERR_PTR(err);
1732 }
eead1911 1733
49705b77 1734 return __lookup_hash(&this, base, NULL);
057f6c01
JM		 1735}
1736
2d8f3038
AV		 1737int user_path_at(int dfd, const char __user *name, unsigned flags,
1738 struct path *path)
1da177e4 1739{
2d8f3038 1740 struct nameidata nd;
f52e0c11 1741 char *tmp = getname_flags(name, flags);
1da177e4 1742 int err = PTR_ERR(tmp);
1da177e4 1743 if (!IS_ERR(tmp)) {
2d8f3038
AV		 1744
1745 BUG_ON(flags & LOOKUP_PARENT);
1746
1747 err = do_path_lookup(dfd, tmp, flags, &nd);
1da177e4 1748 putname(tmp);
2d8f3038
AV		 1749 if (!err)
1750 *path = nd.path;
1da177e4
LT		 1751 }
1752 return err;
1753}
1754
2ad94ae6
AV		 1755static int user_path_parent(int dfd, const char __user *path,
1756 struct nameidata *nd, char **name)
1757{
1758 char *s = getname(path);
1759 int error;
1760
1761 if (IS_ERR(s))
1762 return PTR_ERR(s);
1763
1764 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1765 if (error)
1766 putname(s);
1767 else
1768 *name = s;
1769
1770 return error;
1771}
1772
1da177e4
LT		 1773/*
1774 * It's inline, so penalty for filesystems that don't use sticky bit is
1775 * minimal.
1776 */
1777static inline int check_sticky(struct inode *dir, struct inode *inode)
1778{
da9592ed
DH		 1779 uid_t fsuid = current_fsuid();
1780
1da177e4
LT		 1781 if (!(dir->i_mode & S_ISVTX))
1782 return 0;
e795b717
SH		 1783 if (current_user_ns() != inode_userns(inode))
1784 goto other_userns;
da9592ed 1785 if (inode->i_uid == fsuid)
1da177e4 1786 return 0;
da9592ed 1787 if (dir->i_uid == fsuid)
1da177e4 1788 return 0;
e795b717
SH		 1789
1790other_userns:
1791 return !ns_capable(inode_userns(inode), CAP_FOWNER);
1da177e4
LT		 1792}
1793
1794/*
1795 * Check whether we can remove a link victim from directory dir, check
1796 * whether the type of victim is right.
1797 * 1. We can't do it if dir is read-only (done in permission())
1798 * 2. We should have write and exec permissions on dir
1799 * 3. We can't remove anything from append-only dir
1800 * 4. We can't do anything with immutable dir (done in permission())
1801 * 5. If the sticky bit on dir is set we should either
1802 * a. be owner of dir, or
1803 * b. be owner of victim, or
1804 * c. have CAP_FOWNER capability
1805 * 6. If the victim is append-only or immutable we can't do antyhing with
1806 * links pointing to it.
1807 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1808 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1809 * 9. We can't remove a root or mountpoint.
1810 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1811 * nfs_async_unlink().
1812 */
858119e1 1813static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1da177e4
LT		 1814{
1815 int error;
1816
1817 if (!victim->d_inode)
1818 return -ENOENT;
1819
1820 BUG_ON(victim->d_parent->d_inode != dir);
cccc6bba 1821 audit_inode_child(victim, dir);
1da177e4 1822
f419a2e3 1823 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1824 if (error)
1825 return error;
1826 if (IS_APPEND(dir))
1827 return -EPERM;
1828 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
f9454548 1829 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1da177e4
LT		 1830 return -EPERM;
1831 if (isdir) {
1832 if (!S_ISDIR(victim->d_inode->i_mode))
1833 return -ENOTDIR;
1834 if (IS_ROOT(victim))
1835 return -EBUSY;
1836 } else if (S_ISDIR(victim->d_inode->i_mode))
1837 return -EISDIR;
1838 if (IS_DEADDIR(dir))
1839 return -ENOENT;
1840 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1841 return -EBUSY;
1842 return 0;
1843}
1844
1845/* Check whether we can create an object with dentry child in directory
1846 * dir.
1847 * 1. We can't do it if child already exists (open has special treatment for
1848 * this case, but since we are inlined it's OK)
1849 * 2. We can't do it if dir is read-only (done in permission())
1850 * 3. We should have write and exec permissions on dir
1851 * 4. We can't do it if dir is immutable (done in permission())
1852 */
a95164d9 1853static inline int may_create(struct inode *dir, struct dentry *child)
1da177e4
LT		 1854{
1855 if (child->d_inode)
1856 return -EEXIST;
1857 if (IS_DEADDIR(dir))
1858 return -ENOENT;
f419a2e3 1859 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1860}
1861
1da177e4
LT		 1862/*
1863 * p1 and p2 should be directories on the same fs.
1864 */
1865struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1866{
1867 struct dentry *p;
1868
1869 if (p1 == p2) {
f2eace23 1870 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1da177e4
LT		 1871 return NULL;
1872 }
1873
a11f3a05 1874 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4 1875
e2761a11
OH		 1876 p = d_ancestor(p2, p1);
1877 if (p) {
1878 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1879 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1880 return p;
1da177e4
LT		 1881 }
1882
e2761a11
OH		 1883 p = d_ancestor(p1, p2);
1884 if (p) {
1885 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1886 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1887 return p;
1da177e4
LT		 1888 }
1889
f2eace23
IM		 1890 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1891 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1da177e4
LT		 1892 return NULL;
1893}
1894
1895void unlock_rename(struct dentry *p1, struct dentry *p2)
1896{
1b1dcc1b 1897 mutex_unlock(&p1->d_inode->i_mutex);
1da177e4 1898 if (p1 != p2) {
1b1dcc1b 1899 mutex_unlock(&p2->d_inode->i_mutex);
a11f3a05 1900 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4
LT		 1901 }
1902}
1903
1904int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1905 struct nameidata *nd)
1906{
a95164d9 1907 int error = may_create(dir, dentry);
1da177e4
LT		 1908
1909 if (error)
1910 return error;
1911
acfa4380 1912 if (!dir->i_op->create)
1da177e4
LT		 1913 return -EACCES; /* shouldn't it be ENOSYS? */
1914 mode &= S_IALLUGO;
1915 mode |= S_IFREG;
1916 error = security_inode_create(dir, dentry, mode);
1917 if (error)
1918 return error;
1da177e4 1919 error = dir->i_op->create(dir, dentry, mode, nd);
a74574aa 1920 if (!error)
f38aa942 1921 fsnotify_create(dir, dentry);
1da177e4
LT		 1922 return error;
1923}
1924
73d049a4 1925static int may_open(struct path *path, int acc_mode, int flag)
1da177e4 1926{
3fb64190 1927 struct dentry *dentry = path->dentry;
1da177e4
LT		 1928 struct inode *inode = dentry->d_inode;
1929 int error;
1930
bcda7652
AV		 1931 /* O_PATH? */
1932 if (!acc_mode)
1933 return 0;
1934
1da177e4
LT		 1935 if (!inode)
1936 return -ENOENT;
1937
c8fe8f30
CH		 1938 switch (inode->i_mode & S_IFMT) {
1939 case S_IFLNK:
1da177e4 1940 return -ELOOP;
c8fe8f30
CH		 1941 case S_IFDIR:
1942 if (acc_mode & MAY_WRITE)
1943 return -EISDIR;
1944 break;
1945 case S_IFBLK:
1946 case S_IFCHR:
3fb64190 1947 if (path->mnt->mnt_flags & MNT_NODEV)
1da177e4 1948 return -EACCES;
c8fe8f30
CH		 1949 /*FALLTHRU*/
1950 case S_IFIFO:
1951 case S_IFSOCK:
1da177e4 1952 flag &= ~O_TRUNC;
c8fe8f30 1953 break;
4a3fd211 1954 }
b41572e9 1955
3fb64190 1956 error = inode_permission(inode, acc_mode);
b41572e9
DH		 1957 if (error)
1958 return error;
6146f0d5 1959
1da177e4
LT		 1960 /*
1961 * An append-only file must be opened in append mode for writing.
1962 */
1963 if (IS_APPEND(inode)) {
8737c930 1964 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
7715b521 1965 return -EPERM;
1da177e4 1966 if (flag & O_TRUNC)
7715b521 1967 return -EPERM;
1da177e4
LT		 1968 }
1969
1970 /* O_NOATIME can only be set by the owner or superuser */
2e149670 1971 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
7715b521 1972 return -EPERM;
1da177e4
LT		 1973
1974 /*
1975 * Ensure there are no outstanding leases on the file.
1976 */
b65a9cfc 1977 return break_lease(inode, flag);
7715b521 1978}
1da177e4 1979
e1181ee6 1980static int handle_truncate(struct file *filp)
7715b521 1981{
e1181ee6 1982 struct path *path = &filp->f_path;
7715b521
AV		 1983 struct inode *inode = path->dentry->d_inode;
1984 int error = get_write_access(inode);
1985 if (error)
1986 return error;
1987 /*
1988 * Refuse to truncate files with mandatory locks held on them.
1989 */
1990 error = locks_verify_locked(inode);
1991 if (!error)
ea0d3ab2 1992 error = security_path_truncate(path);
7715b521
AV		 1993 if (!error) {
1994 error = do_truncate(path->dentry, 0,
1995 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
e1181ee6 1996 filp);
7715b521
AV		 1997 }
1998 put_write_access(inode);
acd0c935 1999 return error;
1da177e4
LT		 2000}
2001
d57999e1
DH		 2002/*
2003 * Note that while the flag value (low two bits) for sys_open means:
2004 * 00 - read-only
2005 * 01 - write-only
2006 * 10 - read-write
2007 * 11 - special
2008 * it is changed into
2009 * 00 - no permissions needed
2010 * 01 - read-permission
2011 * 10 - write-permission
2012 * 11 - read-write
2013 * for the internal routines (ie open_namei()/follow_link() etc)
2014 * This is more logical, and also allows the 00 "no perm needed"
2015 * to be used for symlinks (where the permissions are checked
2016 * later).
2017 *
2018*/
2019static inline int open_to_namei_flags(int flag)
2020{
2021 if ((flag+1) & O_ACCMODE)
2022 flag++;
2023 return flag;
2024}
2025
31e6b01f 2026/*
fe2d35ff 2027 * Handle the last step of open()
31e6b01f 2028 */
fb1cc555 2029static struct file *do_last(struct nameidata *nd, struct path *path,
c3e380b0 2030 const struct open_flags *op, const char *pathname)
fb1cc555 2031{
a1e28038 2032 struct dentry *dir = nd->path.dentry;
6c0d46c4 2033 struct dentry *dentry;
ca344a89 2034 int open_flag = op->open_flag;
6c0d46c4 2035 int will_truncate = open_flag & O_TRUNC;
ca344a89 2036 int want_write = 0;
bcda7652 2037 int acc_mode = op->acc_mode;
fb1cc555 2038 struct file *filp;
16c2cd71 2039 int error;
1f36f774 2040
c3e380b0
AV		 2041 nd->flags &= ~LOOKUP_PARENT;
2042 nd->flags |= op->intent;
2043
1f36f774
AV		 2044 switch (nd->last_type) {
2045 case LAST_DOTDOT:
176306f5 2046 case LAST_DOT:
fe2d35ff
AV		 2047 error = handle_dots(nd, nd->last_type);
2048 if (error)
2049 return ERR_PTR(error);
1f36f774 2050 /* fallthrough */
1f36f774 2051 case LAST_ROOT:
9f1fafee 2052 error = complete_walk(nd);
16c2cd71 2053 if (error)
9f1fafee 2054 return ERR_PTR(error);
fe2d35ff 2055 audit_inode(pathname, nd->path.dentry);
ca344a89 2056 if (open_flag & O_CREAT) {
fe2d35ff
AV		 2057 error = -EISDIR;
2058 goto exit;
2059 }
2060 goto ok;
1f36f774 2061 case LAST_BIND:
9f1fafee 2062 error = complete_walk(nd);
16c2cd71 2063 if (error)
9f1fafee 2064 return ERR_PTR(error);
1f36f774 2065 audit_inode(pathname, dir);
67ee3ad2 2066 goto ok;
1f36f774 2067 }
67ee3ad2 2068
ca344a89 2069 if (!(open_flag & O_CREAT)) {
bcda7652 2070 int symlink_ok = 0;
fe2d35ff
AV		 2071 if (nd->last.name[nd->last.len])
2072 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
bcda7652
AV		 2073 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2074 symlink_ok = 1;
fe2d35ff 2075 /* we _can_ be in RCU mode here */
ce57dfc1
AV		 2076 error = walk_component(nd, path, &nd->last, LAST_NORM,
2077 !symlink_ok);
2078 if (error < 0)
fe2d35ff 2079 return ERR_PTR(error);
ce57dfc1 2080 if (error) /* symlink */
fe2d35ff 2081 return NULL;
fe2d35ff 2082 /* sayonara */
9f1fafee
AV		 2083 error = complete_walk(nd);
2084 if (error)
2085 return ERR_PTR(-ECHILD);
fe2d35ff
AV		 2086
2087 error = -ENOTDIR;
2088 if (nd->flags & LOOKUP_DIRECTORY) {
ce57dfc1 2089 if (!nd->inode->i_op->lookup)
fe2d35ff
AV		 2090 goto exit;
2091 }
2092 audit_inode(pathname, nd->path.dentry);
2093 goto ok;
2094 }
2095
2096 /* create side of things */
9f1fafee
AV		 2097 error = complete_walk(nd);
2098 if (error)
2099 return ERR_PTR(error);
fe2d35ff
AV		 2100
2101 audit_inode(pathname, dir);
16c2cd71 2102 error = -EISDIR;
1f36f774 2103 /* trailing slashes? */
31e6b01f
NP		 2104 if (nd->last.name[nd->last.len])
2105 goto exit;
a2c36b45 2106
a1e28038
AV		 2107 mutex_lock(&dir->d_inode->i_mutex);
2108
6c0d46c4
AV		 2109 dentry = lookup_hash(nd);
2110 error = PTR_ERR(dentry);
2111 if (IS_ERR(dentry)) {
fb1cc555
AV		 2112 mutex_unlock(&dir->d_inode->i_mutex);
2113 goto exit;
2114 }
2115
6c0d46c4
AV		 2116 path->dentry = dentry;
2117 path->mnt = nd->path.mnt;
2118
fb1cc555 2119 /* Negative dentry, just create the file */
6c0d46c4
AV		 2120 if (!dentry->d_inode) {
2121 int mode = op->mode;
2122 if (!IS_POSIXACL(dir->d_inode))
2123 mode &= ~current_umask();
fb1cc555
AV		 2124 /*
2125 * This write is needed to ensure that a
6c0d46c4 2126 * rw->ro transition does not occur between
fb1cc555
AV		 2127 * the time when the file is created and when
2128 * a permanent write count is taken through
2129 * the 'struct file' in nameidata_to_filp().
2130 */
2131 error = mnt_want_write(nd->path.mnt);
2132 if (error)
2133 goto exit_mutex_unlock;
ca344a89 2134 want_write = 1;
9b44f1b3 2135 /* Don't check for write permission, don't truncate */
ca344a89 2136 open_flag &= ~O_TRUNC;
6c0d46c4 2137 will_truncate = 0;
bcda7652 2138 acc_mode = MAY_OPEN;
6c0d46c4
AV		 2139 error = security_path_mknod(&nd->path, dentry, mode, 0);
2140 if (error)
2141 goto exit_mutex_unlock;
2142 error = vfs_create(dir->d_inode, dentry, mode, nd);
2143 if (error)
2144 goto exit_mutex_unlock;
2145 mutex_unlock(&dir->d_inode->i_mutex);
2146 dput(nd->path.dentry);
2147 nd->path.dentry = dentry;
ca344a89 2148 goto common;
fb1cc555
AV		 2149 }
2150
2151 /*
2152 * It already exists.
2153 */
2154 mutex_unlock(&dir->d_inode->i_mutex);
2155 audit_inode(pathname, path->dentry);
2156
2157 error = -EEXIST;
ca344a89 2158 if (open_flag & O_EXCL)
fb1cc555
AV		 2159 goto exit_dput;
2160
9875cf80
DH		 2161 error = follow_managed(path, nd->flags);
2162 if (error < 0)
2163 goto exit_dput;
fb1cc555
AV		 2164
2165 error = -ENOENT;
2166 if (!path->dentry->d_inode)
2167 goto exit_dput;
9e67f361
AV		 2168
2169 if (path->dentry->d_inode->i_op->follow_link)
fb1cc555 2170 return NULL;
fb1cc555
AV		 2171
2172 path_to_nameidata(path, nd);
31e6b01f 2173 nd->inode = path->dentry->d_inode;
fb1cc555 2174 error = -EISDIR;
31e6b01f 2175 if (S_ISDIR(nd->inode->i_mode))
fb1cc555 2176 goto exit;
67ee3ad2 2177ok:
6c0d46c4
AV		 2178 if (!S_ISREG(nd->inode->i_mode))
2179 will_truncate = 0;
2180
0f9d1a10
AV		 2181 if (will_truncate) {
2182 error = mnt_want_write(nd->path.mnt);
2183 if (error)
2184 goto exit;
ca344a89 2185 want_write = 1;
0f9d1a10 2186 }
ca344a89 2187common:
bcda7652 2188 error = may_open(&nd->path, acc_mode, open_flag);
ca344a89 2189 if (error)
0f9d1a10 2190 goto exit;
0f9d1a10
AV		 2191 filp = nameidata_to_filp(nd);
2192 if (!IS_ERR(filp)) {
2193 error = ima_file_check(filp, op->acc_mode);
2194 if (error) {
2195 fput(filp);
2196 filp = ERR_PTR(error);
2197 }
2198 }
2199 if (!IS_ERR(filp)) {
2200 if (will_truncate) {
2201 error = handle_truncate(filp);
2202 if (error) {
2203 fput(filp);
2204 filp = ERR_PTR(error);
2205 }
2206 }
2207 }
ca344a89
AV		 2208out:
2209 if (want_write)
0f9d1a10
AV		 2210 mnt_drop_write(nd->path.mnt);
2211 path_put(&nd->path);
fb1cc555
AV		 2212 return filp;
2213
2214exit_mutex_unlock:
2215 mutex_unlock(&dir->d_inode->i_mutex);
2216exit_dput:
2217 path_put_conditional(path, nd);
2218exit:
ca344a89
AV		 2219 filp = ERR_PTR(error);
2220 goto out;
fb1cc555
AV		 2221}
2222
13aab428 2223static struct file *path_openat(int dfd, const char *pathname,
73d049a4 2224 struct nameidata *nd, const struct open_flags *op, int flags)
1da177e4 2225{
fe2d35ff 2226 struct file *base = NULL;
4a3fd211 2227 struct file *filp;
9850c056 2228 struct path path;
13aab428 2229 int error;
31e6b01f
NP		 2230
2231 filp = get_empty_filp();
2232 if (!filp)
2233 return ERR_PTR(-ENFILE);
2234
47c805dc 2235 filp->f_flags = op->open_flag;
73d049a4
AV		 2236 nd->intent.open.file = filp;
2237 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2238 nd->intent.open.create_mode = op->mode;
31e6b01f 2239
73d049a4 2240 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
31e6b01f 2241 if (unlikely(error))
13aab428 2242 goto out_filp;
31e6b01f 2243
fe2d35ff 2244 current->total_link_count = 0;
73d049a4 2245 error = link_path_walk(pathname, nd);
31e6b01f
NP		 2246 if (unlikely(error))
2247 goto out_filp;
1da177e4 2248
73d049a4 2249 filp = do_last(nd, &path, op, pathname);
806b681c 2250 while (unlikely(!filp)) { /* trailing symlink */
7b9337aa 2251 struct path link = path;
def4af30 2252 void *cookie;
574197e0 2253 if (!(nd->flags & LOOKUP_FOLLOW)) {
73d049a4
AV		 2254 path_put_conditional(&path, nd);
2255 path_put(&nd->path);
40b39136
AV		 2256 filp = ERR_PTR(-ELOOP);
2257 break;
2258 }
73d049a4
AV		 2259 nd->flags |= LOOKUP_PARENT;
2260 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
574197e0 2261 error = follow_link(&link, nd, &cookie);
c3e380b0 2262 if (unlikely(error))
f1afe9ef 2263 filp = ERR_PTR(error);
c3e380b0 2264 else
73d049a4 2265 filp = do_last(nd, &path, op, pathname);
574197e0 2266 put_link(nd, &link, cookie);
806b681c 2267 }
10fa8e62 2268out:
73d049a4
AV		 2269 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2270 path_put(&nd->root);
fe2d35ff
AV		 2271 if (base)
2272 fput(base);
73d049a4 2273 release_open_intent(nd);
10fa8e62 2274 return filp;
1da177e4 2275
31e6b01f 2276out_filp:
806b681c 2277 filp = ERR_PTR(error);
10fa8e62 2278 goto out;
1da177e4
LT		 2279}
2280
13aab428
AV		 2281struct file *do_filp_open(int dfd, const char *pathname,
2282 const struct open_flags *op, int flags)
2283{
73d049a4 2284 struct nameidata nd;
13aab428
AV		 2285 struct file *filp;
2286
73d049a4 2287 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
13aab428 2288 if (unlikely(filp == ERR_PTR(-ECHILD)))
73d049a4 2289 filp = path_openat(dfd, pathname, &nd, op, flags);
13aab428 2290 if (unlikely(filp == ERR_PTR(-ESTALE)))
73d049a4 2291 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
13aab428
AV		 2292 return filp;
2293}
2294
73d049a4
AV		 2295struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2296 const char *name, const struct open_flags *op, int flags)
2297{
2298 struct nameidata nd;
2299 struct file *file;
2300
2301 nd.root.mnt = mnt;
2302 nd.root.dentry = dentry;
2303
2304 flags |= LOOKUP_ROOT;
2305
bcda7652 2306 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
73d049a4
AV		 2307 return ERR_PTR(-ELOOP);
2308
2309 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2310 if (unlikely(file == ERR_PTR(-ECHILD)))
2311 file = path_openat(-1, name, &nd, op, flags);
2312 if (unlikely(file == ERR_PTR(-ESTALE)))
2313 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2314 return file;
2315}
2316
1da177e4
LT		 2317/**
2318 * lookup_create - lookup a dentry, creating it if it doesn't exist
2319 * @nd: nameidata info
2320 * @is_dir: directory flag
2321 *
2322 * Simple function to lookup and return a dentry and create it
2323 * if it doesn't exist. Is SMP-safe.
c663e5d8 2324 *
4ac91378 2325 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1da177e4
LT		 2326 */
2327struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2328{
c663e5d8 2329 struct dentry *dentry = ERR_PTR(-EEXIST);
1da177e4 2330
4ac91378 2331 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
c663e5d8
CH		 2332 /*
2333 * Yucky last component or no last component at all?
2334 * (foo/., foo/.., /////)
2335 */
1da177e4
LT		 2336 if (nd->last_type != LAST_NORM)
2337 goto fail;
2338 nd->flags &= ~LOOKUP_PARENT;
3516586a 2339 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
a634904a 2340 nd->intent.open.flags = O_EXCL;
c663e5d8
CH		 2341
2342 /*
2343 * Do the final lookup.
2344 */
49705b77 2345 dentry = lookup_hash(nd);
1da177e4
LT		 2346 if (IS_ERR(dentry))
2347 goto fail;
c663e5d8 2348
e9baf6e5
AV		 2349 if (dentry->d_inode)
2350 goto eexist;
c663e5d8
CH		 2351 /*
2352 * Special case - lookup gave negative, but... we had foo/bar/
2353 * From the vfs_mknod() POV we just have a negative dentry -
2354 * all is fine. Let's be bastards - you had / on the end, you've
2355 * been asking for (non-existent) directory. -ENOENT for you.
2356 */
e9baf6e5
AV		 2357 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2358 dput(dentry);
2359 dentry = ERR_PTR(-ENOENT);
2360 }
1da177e4 2361 return dentry;
e9baf6e5 2362eexist:
1da177e4 2363 dput(dentry);
e9baf6e5 2364 dentry = ERR_PTR(-EEXIST);
1da177e4
LT		 2365fail:
2366 return dentry;
2367}
f81a0bff 2368EXPORT_SYMBOL_GPL(lookup_create);
1da177e4
LT		 2369
2370int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2371{
a95164d9 2372 int error = may_create(dir, dentry);
1da177e4
LT		 2373
2374 if (error)
2375 return error;
2376
e795b717
SH		 2377 if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
2378 !ns_capable(inode_userns(dir), CAP_MKNOD))
1da177e4
LT		 2379 return -EPERM;
2380
acfa4380 2381 if (!dir->i_op->mknod)
1da177e4
LT		 2382 return -EPERM;
2383
08ce5f16
SH		 2384 error = devcgroup_inode_mknod(mode, dev);
2385 if (error)
2386 return error;
2387
1da177e4
LT		 2388 error = security_inode_mknod(dir, dentry, mode, dev);
2389 if (error)
2390 return error;
2391
1da177e4 2392 error = dir->i_op->mknod(dir, dentry, mode, dev);
a74574aa 2393 if (!error)
f38aa942 2394 fsnotify_create(dir, dentry);
1da177e4
LT		 2395 return error;
2396}
2397
463c3197
DH		 2398static int may_mknod(mode_t mode)
2399{
2400 switch (mode & S_IFMT) {
2401 case S_IFREG:
2402 case S_IFCHR:
2403 case S_IFBLK:
2404 case S_IFIFO:
2405 case S_IFSOCK:
2406 case 0: /* zero mode translates to S_IFREG */
2407 return 0;
2408 case S_IFDIR:
2409 return -EPERM;
2410 default:
2411 return -EINVAL;
2412 }
2413}
2414
2e4d0924
HC		 2415SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2416 unsigned, dev)
1da177e4 2417{
2ad94ae6
AV		 2418 int error;
2419 char *tmp;
2420 struct dentry *dentry;
1da177e4
LT		 2421 struct nameidata nd;
2422
2423 if (S_ISDIR(mode))
2424 return -EPERM;
1da177e4 2425
2ad94ae6 2426 error = user_path_parent(dfd, filename, &nd, &tmp);
1da177e4 2427 if (error)
2ad94ae6
AV		 2428 return error;
2429
1da177e4 2430 dentry = lookup_create(&nd, 0);
463c3197
DH		 2431 if (IS_ERR(dentry)) {
2432 error = PTR_ERR(dentry);
2433 goto out_unlock;
2434 }
4ac91378 2435 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2436 mode &= ~current_umask();
463c3197
DH		 2437 error = may_mknod(mode);
2438 if (error)
2439 goto out_dput;
2440 error = mnt_want_write(nd.path.mnt);
2441 if (error)
2442 goto out_dput;
be6d3e56
KT		 2443 error = security_path_mknod(&nd.path, dentry, mode, dev);
2444 if (error)
2445 goto out_drop_write;
463c3197 2446 switch (mode & S_IFMT) {
1da177e4 2447 case 0: case S_IFREG:
4ac91378 2448 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
1da177e4
LT		 2449 break;
2450 case S_IFCHR: case S_IFBLK:
4ac91378 2451 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
1da177e4
LT		 2452 new_decode_dev(dev));
2453 break;
2454 case S_IFIFO: case S_IFSOCK:
4ac91378 2455 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
1da177e4 2456 break;
1da177e4 2457 }
be6d3e56 2458out_drop_write:
463c3197
DH		 2459 mnt_drop_write(nd.path.mnt);
2460out_dput:
2461 dput(dentry);
2462out_unlock:
4ac91378 2463 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2464 path_put(&nd.path);
1da177e4
LT		 2465 putname(tmp);
2466
2467 return error;
2468}
2469
3480b257 2470SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
5590ff0d
UD		 2471{
2472 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2473}
2474
1da177e4
LT		 2475int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2476{
a95164d9 2477 int error = may_create(dir, dentry);
1da177e4
LT		 2478
2479 if (error)
2480 return error;
2481
acfa4380 2482 if (!dir->i_op->mkdir)
1da177e4
LT		 2483 return -EPERM;
2484
2485 mode &= (S_IRWXUGO|S_ISVTX);
2486 error = security_inode_mkdir(dir, dentry, mode);
2487 if (error)
2488 return error;
2489
1da177e4 2490 error = dir->i_op->mkdir(dir, dentry, mode);
a74574aa 2491 if (!error)
f38aa942 2492 fsnotify_mkdir(dir, dentry);
1da177e4
LT		 2493 return error;
2494}
2495
2e4d0924 2496SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
1da177e4
LT		 2497{
2498 int error = 0;
2499 char * tmp;
6902d925
DH		 2500 struct dentry *dentry;
2501 struct nameidata nd;
1da177e4 2502
2ad94ae6
AV		 2503 error = user_path_parent(dfd, pathname, &nd, &tmp);
2504 if (error)
6902d925 2505 goto out_err;
1da177e4 2506
6902d925
DH		 2507 dentry = lookup_create(&nd, 1);
2508 error = PTR_ERR(dentry);
2509 if (IS_ERR(dentry))
2510 goto out_unlock;
1da177e4 2511
4ac91378 2512 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2513 mode &= ~current_umask();
463c3197
DH		 2514 error = mnt_want_write(nd.path.mnt);
2515 if (error)
2516 goto out_dput;
be6d3e56
KT		 2517 error = security_path_mkdir(&nd.path, dentry, mode);
2518 if (error)
2519 goto out_drop_write;
4ac91378 2520 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
be6d3e56 2521out_drop_write:
463c3197
DH		 2522 mnt_drop_write(nd.path.mnt);
2523out_dput:
6902d925
DH		 2524 dput(dentry);
2525out_unlock:
4ac91378 2526 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2527 path_put(&nd.path);
6902d925
DH		 2528 putname(tmp);
2529out_err:
1da177e4
LT		 2530 return error;
2531}
2532
3cdad428 2533SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
5590ff0d
UD		 2534{
2535 return sys_mkdirat(AT_FDCWD, pathname, mode);
2536}
2537
1da177e4 2538/*
a71905f0
SW		 2539 * The dentry_unhash() helper will try to drop the dentry early: we
2540 * should have a usage count of 2 if we're the only user of this
2541 * dentry, and if that is true (possibly after pruning the dcache),
2542 * then we drop the dentry now.
1da177e4
LT		 2543 *
2544 * A low-level filesystem can, if it choses, legally
2545 * do a
2546 *
2547 * if (!d_unhashed(dentry))
2548 * return -EBUSY;
2549 *
2550 * if it cannot handle the case of removing a directory
2551 * that is still in use by something else..
2552 */
2553void dentry_unhash(struct dentry *dentry)
2554{
dc168427 2555 shrink_dcache_parent(dentry);
1da177e4 2556 spin_lock(&dentry->d_lock);
64252c75 2557 if (dentry->d_count == 1)
1da177e4
LT		 2558 __d_drop(dentry);
2559 spin_unlock(&dentry->d_lock);
1da177e4
LT		 2560}
2561
2562int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2563{
2564 int error = may_delete(dir, dentry, 1);
2565
2566 if (error)
2567 return error;
2568
acfa4380 2569 if (!dir->i_op->rmdir)
1da177e4
LT		 2570 return -EPERM;
2571
1b1dcc1b 2572 mutex_lock(&dentry->d_inode->i_mutex);
912dbc15
SW		 2573
2574 error = -EBUSY;
1da177e4 2575 if (d_mountpoint(dentry))
912dbc15
SW		 2576 goto out;
2577
2578 error = security_inode_rmdir(dir, dentry);
2579 if (error)
2580 goto out;
2581
2582 error = dir->i_op->rmdir(dir, dentry);
2583 if (error)
2584 goto out;
2585
2586 dentry->d_inode->i_flags |= S_DEAD;
2587 dont_mount(dentry);
2588
2589out:
1b1dcc1b 2590 mutex_unlock(&dentry->d_inode->i_mutex);
912dbc15 2591 if (!error)
1da177e4 2592 d_delete(dentry);
1da177e4
LT		 2593 return error;
2594}
2595
5590ff0d 2596static long do_rmdir(int dfd, const char __user *pathname)
1da177e4
LT		 2597{
2598 int error = 0;
2599 char * name;
2600 struct dentry *dentry;
2601 struct nameidata nd;
2602
2ad94ae6 2603 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2604 if (error)
2ad94ae6 2605 return error;
1da177e4
LT		 2606
2607 switch(nd.last_type) {
0612d9fb
OH		 2608 case LAST_DOTDOT:
2609 error = -ENOTEMPTY;
2610 goto exit1;
2611 case LAST_DOT:
2612 error = -EINVAL;
2613 goto exit1;
2614 case LAST_ROOT:
2615 error = -EBUSY;
2616 goto exit1;
1da177e4 2617 }
0612d9fb
OH		 2618
2619 nd.flags &= ~LOOKUP_PARENT;
2620
4ac91378 2621 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2622 dentry = lookup_hash(&nd);
1da177e4 2623 error = PTR_ERR(dentry);
6902d925
DH		 2624 if (IS_ERR(dentry))
2625 goto exit2;
0622753b
DH		 2626 error = mnt_want_write(nd.path.mnt);
2627 if (error)
2628 goto exit3;
be6d3e56
KT		 2629 error = security_path_rmdir(&nd.path, dentry);
2630 if (error)
2631 goto exit4;
4ac91378 2632 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
be6d3e56 2633exit4:
0622753b
DH		 2634 mnt_drop_write(nd.path.mnt);
2635exit3:
6902d925
DH		 2636 dput(dentry);
2637exit2:
4ac91378 2638 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2639exit1:
1d957f9b 2640 path_put(&nd.path);
1da177e4
LT		 2641 putname(name);
2642 return error;
2643}
2644
3cdad428 2645SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
5590ff0d
UD		 2646{
2647 return do_rmdir(AT_FDCWD, pathname);
2648}
2649
1da177e4
LT		 2650int vfs_unlink(struct inode *dir, struct dentry *dentry)
2651{
2652 int error = may_delete(dir, dentry, 0);
2653
2654 if (error)
2655 return error;
2656
acfa4380 2657 if (!dir->i_op->unlink)
1da177e4
LT		 2658 return -EPERM;
2659
1b1dcc1b 2660 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2661 if (d_mountpoint(dentry))
2662 error = -EBUSY;
2663 else {
2664 error = security_inode_unlink(dir, dentry);
bec1052e 2665 if (!error) {
1da177e4 2666 error = dir->i_op->unlink(dir, dentry);
bec1052e 2667 if (!error)
d83c49f3 2668 dont_mount(dentry);
bec1052e 2669 }
1da177e4 2670 }
1b1dcc1b 2671 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2672
2673 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2674 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
ece95912 2675 fsnotify_link_count(dentry->d_inode);
e234f35c 2676 d_delete(dentry);
1da177e4 2677 }
0eeca283 2678
1da177e4
LT		 2679 return error;
2680}
2681
2682/*
2683 * Make sure that the actual truncation of the file will occur outside its
1b1dcc1b 2684 * directory's i_mutex. Truncate can take a long time if there is a lot of
1da177e4
LT		 2685 * writeout happening, and we don't want to prevent access to the directory
2686 * while waiting on the I/O.
2687 */
5590ff0d 2688static long do_unlinkat(int dfd, const char __user *pathname)
1da177e4 2689{
2ad94ae6
AV		 2690 int error;
2691 char *name;
1da177e4
LT		 2692 struct dentry *dentry;
2693 struct nameidata nd;
2694 struct inode *inode = NULL;
2695
2ad94ae6 2696 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2697 if (error)
2ad94ae6
AV		 2698 return error;
2699
1da177e4
LT		 2700 error = -EISDIR;
2701 if (nd.last_type != LAST_NORM)
2702 goto exit1;
0612d9fb
OH		 2703
2704 nd.flags &= ~LOOKUP_PARENT;
2705
4ac91378 2706 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2707 dentry = lookup_hash(&nd);
1da177e4
LT		 2708 error = PTR_ERR(dentry);
2709 if (!IS_ERR(dentry)) {
2710 /* Why not before? Because we want correct error value */
2711 if (nd.last.name[nd.last.len])
2712 goto slashes;
2713 inode = dentry->d_inode;
2714 if (inode)
7de9c6ee 2715 ihold(inode);
0622753b
DH		 2716 error = mnt_want_write(nd.path.mnt);
2717 if (error)
2718 goto exit2;
be6d3e56
KT		 2719 error = security_path_unlink(&nd.path, dentry);
2720 if (error)
2721 goto exit3;
4ac91378 2722 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
be6d3e56 2723exit3:
0622753b 2724 mnt_drop_write(nd.path.mnt);
1da177e4
LT		 2725 exit2:
2726 dput(dentry);
2727 }
4ac91378 2728 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4
LT		 2729 if (inode)
2730 iput(inode); /* truncate the inode here */
2731exit1:
1d957f9b 2732 path_put(&nd.path);
1da177e4
LT		 2733 putname(name);
2734 return error;
2735
2736slashes:
2737 error = !dentry->d_inode ? -ENOENT :
2738 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2739 goto exit2;
2740}
2741
2e4d0924 2742SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
5590ff0d
UD		 2743{
2744 if ((flag & ~AT_REMOVEDIR) != 0)
2745 return -EINVAL;
2746
2747 if (flag & AT_REMOVEDIR)
2748 return do_rmdir(dfd, pathname);
2749
2750 return do_unlinkat(dfd, pathname);
2751}
2752
3480b257 2753SYSCALL_DEFINE1(unlink, const char __user *, pathname)
5590ff0d
UD		 2754{
2755 return do_unlinkat(AT_FDCWD, pathname);
2756}
2757
db2e747b 2758int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
1da177e4 2759{
a95164d9 2760 int error = may_create(dir, dentry);
1da177e4
LT		 2761
2762 if (error)
2763 return error;
2764
acfa4380 2765 if (!dir->i_op->symlink)
1da177e4
LT		 2766 return -EPERM;
2767
2768 error = security_inode_symlink(dir, dentry, oldname);
2769 if (error)
2770 return error;
2771
1da177e4 2772 error = dir->i_op->symlink(dir, dentry, oldname);
a74574aa 2773 if (!error)
f38aa942 2774 fsnotify_create(dir, dentry);
1da177e4
LT		 2775 return error;
2776}
2777
2e4d0924
HC		 2778SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2779 int, newdfd, const char __user *, newname)
1da177e4 2780{
2ad94ae6
AV		 2781 int error;
2782 char *from;
2783 char *to;
6902d925
DH		 2784 struct dentry *dentry;
2785 struct nameidata nd;
1da177e4
LT		 2786
2787 from = getname(oldname);
2ad94ae6 2788 if (IS_ERR(from))
1da177e4 2789 return PTR_ERR(from);
1da177e4 2790
2ad94ae6 2791 error = user_path_parent(newdfd, newname, &nd, &to);
6902d925 2792 if (error)
2ad94ae6
AV		 2793 goto out_putname;
2794
6902d925
DH		 2795 dentry = lookup_create(&nd, 0);
2796 error = PTR_ERR(dentry);
2797 if (IS_ERR(dentry))
2798 goto out_unlock;
2799
75c3f29d
DH		 2800 error = mnt_want_write(nd.path.mnt);
2801 if (error)
2802 goto out_dput;
be6d3e56
KT		 2803 error = security_path_symlink(&nd.path, dentry, from);
2804 if (error)
2805 goto out_drop_write;
db2e747b 2806 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
be6d3e56 2807out_drop_write:
75c3f29d
DH		 2808 mnt_drop_write(nd.path.mnt);
2809out_dput:
6902d925
DH		 2810 dput(dentry);
2811out_unlock:
4ac91378 2812 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2813 path_put(&nd.path);
6902d925
DH		 2814 putname(to);
2815out_putname:
1da177e4
LT		 2816 putname(from);
2817 return error;
2818}
2819
3480b257 2820SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 2821{
2822 return sys_symlinkat(oldname, AT_FDCWD, newname);
2823}
2824
1da177e4
LT		 2825int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2826{
2827 struct inode *inode = old_dentry->d_inode;
2828 int error;
2829
2830 if (!inode)
2831 return -ENOENT;
2832
a95164d9 2833 error = may_create(dir, new_dentry);
1da177e4
LT		 2834 if (error)
2835 return error;
2836
2837 if (dir->i_sb != inode->i_sb)
2838 return -EXDEV;
2839
2840 /*
2841 * A link to an append-only or immutable file cannot be created.
2842 */
2843 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2844 return -EPERM;
acfa4380 2845 if (!dir->i_op->link)
1da177e4 2846 return -EPERM;
7e79eedb 2847 if (S_ISDIR(inode->i_mode))
1da177e4
LT		 2848 return -EPERM;
2849
2850 error = security_inode_link(old_dentry, dir, new_dentry);
2851 if (error)
2852 return error;
2853
7e79eedb 2854 mutex_lock(&inode->i_mutex);
aae8a97d
AK		 2855 /* Make sure we don't allow creating hardlink to an unlinked file */
2856 if (inode->i_nlink == 0)
2857 error = -ENOENT;
2858 else
2859 error = dir->i_op->link(old_dentry, dir, new_dentry);
7e79eedb 2860 mutex_unlock(&inode->i_mutex);
e31e14ec 2861 if (!error)
7e79eedb 2862 fsnotify_link(dir, inode, new_dentry);
1da177e4
LT		 2863 return error;
2864}
2865
2866/*
2867 * Hardlinks are often used in delicate situations. We avoid
2868 * security-related surprises by not following symlinks on the
2869 * newname. --KAB
2870 *
2871 * We don't follow them on the oldname either to be compatible
2872 * with linux 2.0, and to avoid hard-linking to directories
2873 * and other special files. --ADM
2874 */
2e4d0924
HC		 2875SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2876 int, newdfd, const char __user *, newname, int, flags)
1da177e4
LT		 2877{
2878 struct dentry *new_dentry;
2d8f3038
AV		 2879 struct nameidata nd;
2880 struct path old_path;
11a7b371 2881 int how = 0;
1da177e4 2882 int error;
2ad94ae6 2883 char *to;
1da177e4 2884
11a7b371 2885 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
c04030e1 2886 return -EINVAL;
11a7b371
AK		 2887 /*
2888 * To use null names we require CAP_DAC_READ_SEARCH
2889 * This ensures that not everyone will be able to create
2890 * handlink using the passed filedescriptor.
2891 */
2892 if (flags & AT_EMPTY_PATH) {
2893 if (!capable(CAP_DAC_READ_SEARCH))
2894 return -ENOENT;
2895 how = LOOKUP_EMPTY;
2896 }
2897
2898 if (flags & AT_SYMLINK_FOLLOW)
2899 how |= LOOKUP_FOLLOW;
c04030e1 2900
11a7b371 2901 error = user_path_at(olddfd, oldname, how, &old_path);
1da177e4 2902 if (error)
2ad94ae6
AV		 2903 return error;
2904
2905 error = user_path_parent(newdfd, newname, &nd, &to);
1da177e4
LT		 2906 if (error)
2907 goto out;
2908 error = -EXDEV;
2d8f3038 2909 if (old_path.mnt != nd.path.mnt)
1da177e4
LT		 2910 goto out_release;
2911 new_dentry = lookup_create(&nd, 0);
2912 error = PTR_ERR(new_dentry);
6902d925
DH		 2913 if (IS_ERR(new_dentry))
2914 goto out_unlock;
75c3f29d
DH		 2915 error = mnt_want_write(nd.path.mnt);
2916 if (error)
2917 goto out_dput;
be6d3e56
KT		 2918 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2919 if (error)
2920 goto out_drop_write;
2d8f3038 2921 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
be6d3e56 2922out_drop_write:
75c3f29d
DH		 2923 mnt_drop_write(nd.path.mnt);
2924out_dput:
6902d925
DH		 2925 dput(new_dentry);
2926out_unlock:
4ac91378 2927 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2928out_release:
1d957f9b 2929 path_put(&nd.path);
2ad94ae6 2930 putname(to);
1da177e4 2931out:
2d8f3038 2932 path_put(&old_path);
1da177e4
LT		 2933
2934 return error;
2935}
2936
3480b257 2937SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
5590ff0d 2938{
c04030e1 2939 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
5590ff0d
UD		 2940}
2941
1da177e4
LT		 2942/*
2943 * The worst of all namespace operations - renaming directory. "Perverted"
2944 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2945 * Problems:
2946 * a) we can get into loop creation. Check is done in is_subdir().
2947 * b) race potential - two innocent renames can create a loop together.
2948 * That's where 4.4 screws up. Current fix: serialization on
a11f3a05 2949 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
1da177e4
LT		 2950 * story.
2951 * c) we have to lock _three_ objects - parents and victim (if it exists).
1b1dcc1b 2952 * And that - after we got ->i_mutex on parents (until then we don't know
1da177e4
LT		 2953 * whether the target exists). Solution: try to be smart with locking
2954 * order for inodes. We rely on the fact that tree topology may change
a11f3a05 2955 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
1da177e4
LT		 2956 * move will be locked. Thus we can rank directories by the tree
2957 * (ancestors first) and rank all non-directories after them.
2958 * That works since everybody except rename does "lock parent, lookup,
a11f3a05 2959 * lock child" and rename is under ->s_vfs_rename_mutex.
1da177e4
LT		 2960 * HOWEVER, it relies on the assumption that any object with ->lookup()
2961 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2962 * we'd better make sure that there's no link(2) for them.
e4eaac06 2963 * d) conversion from fhandle to dentry may come in the wrong moment - when
1b1dcc1b 2964 * we are removing the target. Solution: we will have to grab ->i_mutex
1da177e4 2965 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
c41b20e7 2966 * ->i_mutex on parents, which works but leads to some truly excessive
1da177e4
LT		 2967 * locking].
2968 */
75c96f85
AB		 2969static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2970 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 2971{
2972 int error = 0;
9055cba7 2973 struct inode *target = new_dentry->d_inode;
1da177e4
LT		 2974
2975 /*
2976 * If we are going to change the parent - check write permissions,
2977 * we'll need to flip '..'.
2978 */
2979 if (new_dir != old_dir) {
f419a2e3 2980 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
1da177e4
LT		 2981 if (error)
2982 return error;
2983 }
2984
2985 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2986 if (error)
2987 return error;
2988
d83c49f3 2989 if (target)
1b1dcc1b 2990 mutex_lock(&target->i_mutex);
9055cba7
SW		 2991
2992 error = -EBUSY;
2993 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
2994 goto out;
2995
2996 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2997 if (error)
2998 goto out;
2999
1da177e4 3000 if (target) {
9055cba7
SW		 3001 target->i_flags |= S_DEAD;
3002 dont_mount(new_dentry);
1da177e4 3003 }
9055cba7
SW		 3004out:
3005 if (target)
3006 mutex_unlock(&target->i_mutex);
e31e14ec 3007 if (!error)
349457cc
MF		 3008 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3009 d_move(old_dentry,new_dentry);
1da177e4
LT		 3010 return error;
3011}
3012
75c96f85
AB		 3013static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3014 struct inode *new_dir, struct dentry *new_dentry)
1da177e4 3015{
51892bbb 3016 struct inode *target = new_dentry->d_inode;
1da177e4
LT		 3017 int error;
3018
3019 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3020 if (error)
3021 return error;
3022
3023 dget(new_dentry);
1da177e4 3024 if (target)
1b1dcc1b 3025 mutex_lock(&target->i_mutex);
51892bbb
SW		 3026
3027 error = -EBUSY;
1da177e4 3028 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
51892bbb
SW		 3029 goto out;
3030
3031 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3032 if (error)
3033 goto out;
3034
3035 if (target)
3036 dont_mount(new_dentry);
3037 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3038 d_move(old_dentry, new_dentry);
3039out:
1da177e4 3040 if (target)
1b1dcc1b 3041 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3042 dput(new_dentry);
3043 return error;
3044}
3045
3046int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3047 struct inode *new_dir, struct dentry *new_dentry)
3048{
3049 int error;
3050 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
59b0df21 3051 const unsigned char *old_name;
1da177e4
LT		 3052
3053 if (old_dentry->d_inode == new_dentry->d_inode)
3054 return 0;
3055
3056 error = may_delete(old_dir, old_dentry, is_dir);
3057 if (error)
3058 return error;
3059
3060 if (!new_dentry->d_inode)
a95164d9 3061 error = may_create(new_dir, new_dentry);
1da177e4
LT		 3062 else
3063 error = may_delete(new_dir, new_dentry, is_dir);
3064 if (error)
3065 return error;
3066
acfa4380 3067 if (!old_dir->i_op->rename)
1da177e4
LT		 3068 return -EPERM;
3069
0eeca283
RL		 3070 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3071
1da177e4
LT		 3072 if (is_dir)
3073 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3074 else
3075 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
123df294
AV		 3076 if (!error)
3077 fsnotify_move(old_dir, new_dir, old_name, is_dir,
5a190ae6 3078 new_dentry->d_inode, old_dentry);
0eeca283
RL		 3079 fsnotify_oldname_free(old_name);
3080
1da177e4
LT		 3081 return error;
3082}
3083
2e4d0924
HC		 3084SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3085 int, newdfd, const char __user *, newname)
1da177e4 3086{
2ad94ae6
AV		 3087 struct dentry *old_dir, *new_dir;
3088 struct dentry *old_dentry, *new_dentry;
3089 struct dentry *trap;
1da177e4 3090 struct nameidata oldnd, newnd;
2ad94ae6
AV		 3091 char *from;
3092 char *to;
3093 int error;
1da177e4 3094
2ad94ae6 3095 error = user_path_parent(olddfd, oldname, &oldnd, &from);
1da177e4
LT		 3096 if (error)
3097 goto exit;
3098
2ad94ae6 3099 error = user_path_parent(newdfd, newname, &newnd, &to);
1da177e4
LT		 3100 if (error)
3101 goto exit1;
3102
3103 error = -EXDEV;
4ac91378 3104 if (oldnd.path.mnt != newnd.path.mnt)
1da177e4
LT		 3105 goto exit2;
3106
4ac91378 3107 old_dir = oldnd.path.dentry;
1da177e4
LT		 3108 error = -EBUSY;
3109 if (oldnd.last_type != LAST_NORM)
3110 goto exit2;
3111
4ac91378 3112 new_dir = newnd.path.dentry;
1da177e4
LT		 3113 if (newnd.last_type != LAST_NORM)
3114 goto exit2;
3115
0612d9fb
OH		 3116 oldnd.flags &= ~LOOKUP_PARENT;
3117 newnd.flags &= ~LOOKUP_PARENT;
4e9ed2f8 3118 newnd.flags |= LOOKUP_RENAME_TARGET;
0612d9fb 3119
1da177e4
LT		 3120 trap = lock_rename(new_dir, old_dir);
3121
49705b77 3122 old_dentry = lookup_hash(&oldnd);
1da177e4
LT		 3123 error = PTR_ERR(old_dentry);
3124 if (IS_ERR(old_dentry))
3125 goto exit3;
3126 /* source must exist */
3127 error = -ENOENT;
3128 if (!old_dentry->d_inode)
3129 goto exit4;
3130 /* unless the source is a directory trailing slashes give -ENOTDIR */
3131 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3132 error = -ENOTDIR;
3133 if (oldnd.last.name[oldnd.last.len])
3134 goto exit4;
3135 if (newnd.last.name[newnd.last.len])
3136 goto exit4;
3137 }
3138 /* source should not be ancestor of target */
3139 error = -EINVAL;
3140 if (old_dentry == trap)
3141 goto exit4;
49705b77 3142 new_dentry = lookup_hash(&newnd);
1da177e4
LT		 3143 error = PTR_ERR(new_dentry);
3144 if (IS_ERR(new_dentry))
3145 goto exit4;
3146 /* target should not be an ancestor of source */
3147 error = -ENOTEMPTY;
3148 if (new_dentry == trap)
3149 goto exit5;
3150
9079b1eb
DH		 3151 error = mnt_want_write(oldnd.path.mnt);
3152 if (error)
3153 goto exit5;
be6d3e56
KT		 3154 error = security_path_rename(&oldnd.path, old_dentry,
3155 &newnd.path, new_dentry);
3156 if (error)
3157 goto exit6;
1da177e4
LT		 3158 error = vfs_rename(old_dir->d_inode, old_dentry,
3159 new_dir->d_inode, new_dentry);
be6d3e56 3160exit6:
9079b1eb 3161 mnt_drop_write(oldnd.path.mnt);
1da177e4
LT		 3162exit5:
3163 dput(new_dentry);
3164exit4:
3165 dput(old_dentry);
3166exit3:
3167 unlock_rename(new_dir, old_dir);
3168exit2:
1d957f9b 3169 path_put(&newnd.path);
2ad94ae6 3170 putname(to);
1da177e4 3171exit1:
1d957f9b 3172 path_put(&oldnd.path);
1da177e4 3173 putname(from);
2ad94ae6 3174exit:
1da177e4
LT		 3175 return error;
3176}
3177
a26eab24 3178SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 3179{
3180 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3181}
3182
1da177e4
LT		 3183int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3184{
3185 int len;
3186
3187 len = PTR_ERR(link);
3188 if (IS_ERR(link))
3189 goto out;
3190
3191 len = strlen(link);
3192 if (len > (unsigned) buflen)
3193 len = buflen;
3194 if (copy_to_user(buffer, link, len))
3195 len = -EFAULT;
3196out:
3197 return len;
3198}
3199
3200/*
3201 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3202 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3203 * using) it for any given inode is up to filesystem.
3204 */
3205int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3206{
3207 struct nameidata nd;
cc314eef 3208 void *cookie;
694a1764 3209 int res;
cc314eef 3210
1da177e4 3211 nd.depth = 0;
cc314eef 3212 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
694a1764
MS		 3213 if (IS_ERR(cookie))
3214 return PTR_ERR(cookie);
3215
3216 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3217 if (dentry->d_inode->i_op->put_link)
3218 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3219 return res;
1da177e4
LT		 3220}
3221
3222int vfs_follow_link(struct nameidata *nd, const char *link)
3223{
3224 return __vfs_follow_link(nd, link);
3225}
3226
3227/* get the link contents into pagecache */
3228static char *page_getlink(struct dentry * dentry, struct page **ppage)
3229{
ebd09abb
DG		 3230 char *kaddr;
3231 struct page *page;
1da177e4 3232 struct address_space *mapping = dentry->d_inode->i_mapping;
090d2b18 3233 page = read_mapping_page(mapping, 0, NULL);
1da177e4 3234 if (IS_ERR(page))
6fe6900e 3235 return (char*)page;
1da177e4 3236 *ppage = page;
ebd09abb
DG		 3237 kaddr = kmap(page);
3238 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3239 return kaddr;
1da177e4
LT		 3240}
3241
3242int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3243{
3244 struct page *page = NULL;
3245 char *s = page_getlink(dentry, &page);
3246 int res = vfs_readlink(dentry,buffer,buflen,s);
3247 if (page) {
3248 kunmap(page);
3249 page_cache_release(page);
3250 }
3251 return res;
3252}
3253
cc314eef 3254void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
1da177e4 3255{
cc314eef 3256 struct page *page = NULL;
1da177e4 3257 nd_set_link(nd, page_getlink(dentry, &page));
cc314eef 3258 return page;
1da177e4
LT		 3259}
3260
cc314eef 3261void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1da177e4 3262{
cc314eef
LT		 3263 struct page *page = cookie;
3264
3265 if (page) {
1da177e4
LT		 3266 kunmap(page);
3267 page_cache_release(page);
1da177e4
LT		 3268 }
3269}
3270
54566b2c
NP		 3271/*
3272 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3273 */
3274int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
1da177e4
LT		 3275{
3276 struct address_space *mapping = inode->i_mapping;
0adb25d2 3277 struct page *page;
afddba49 3278 void *fsdata;
beb497ab 3279 int err;
1da177e4 3280 char *kaddr;
54566b2c
NP		 3281 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3282 if (nofs)
3283 flags |= AOP_FLAG_NOFS;
1da177e4 3284
7e53cac4 3285retry:
afddba49 3286 err = pagecache_write_begin(NULL, mapping, 0, len-1,
54566b2c 3287 flags, &page, &fsdata);
1da177e4 3288 if (err)
afddba49
NP		 3289 goto fail;
3290
1da177e4
LT		 3291 kaddr = kmap_atomic(page, KM_USER0);
3292 memcpy(kaddr, symname, len-1);
3293 kunmap_atomic(kaddr, KM_USER0);
afddba49
NP		 3294
3295 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3296 page, fsdata);
1da177e4
LT		 3297 if (err < 0)
3298 goto fail;
afddba49
NP		 3299 if (err < len-1)
3300 goto retry;
3301
1da177e4
LT		 3302 mark_inode_dirty(inode);
3303 return 0;
1da177e4
LT		 3304fail:
3305 return err;
3306}
3307
0adb25d2
KK		 3308int page_symlink(struct inode *inode, const char *symname, int len)
3309{
3310 return __page_symlink(inode, symname, len,
54566b2c 3311 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
0adb25d2
KK		 3312}
3313
92e1d5be 3314const struct inode_operations page_symlink_inode_operations = {
1da177e4
LT		 3315 .readlink = generic_readlink,
3316 .follow_link = page_follow_link_light,
3317 .put_link = page_put_link,
3318};
3319
2d8f3038 3320EXPORT_SYMBOL(user_path_at);
cc53ce53 3321EXPORT_SYMBOL(follow_down_one);
1da177e4
LT		 3322EXPORT_SYMBOL(follow_down);
3323EXPORT_SYMBOL(follow_up);
3324EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3325EXPORT_SYMBOL(getname);
3326EXPORT_SYMBOL(lock_rename);
1da177e4
LT		 3327EXPORT_SYMBOL(lookup_one_len);
3328EXPORT_SYMBOL(page_follow_link_light);
3329EXPORT_SYMBOL(page_put_link);
3330EXPORT_SYMBOL(page_readlink);
0adb25d2 3331EXPORT_SYMBOL(__page_symlink);
1da177e4
LT		 3332EXPORT_SYMBOL(page_symlink);
3333EXPORT_SYMBOL(page_symlink_inode_operations);
c9c6cac0 3334EXPORT_SYMBOL(kern_path_parent);
d1811465 3335EXPORT_SYMBOL(kern_path);
16f18200 3336EXPORT_SYMBOL(vfs_path_lookup);
f419a2e3 3337EXPORT_SYMBOL(inode_permission);
8c744fb8 3338EXPORT_SYMBOL(file_permission);
1da177e4
LT		 3339EXPORT_SYMBOL(unlock_rename);
3340EXPORT_SYMBOL(vfs_create);
3341EXPORT_SYMBOL(vfs_follow_link);
3342EXPORT_SYMBOL(vfs_link);
3343EXPORT_SYMBOL(vfs_mkdir);
3344EXPORT_SYMBOL(vfs_mknod);
3345EXPORT_SYMBOL(generic_permission);
3346EXPORT_SYMBOL(vfs_readlink);
3347EXPORT_SYMBOL(vfs_rename);
3348EXPORT_SYMBOL(vfs_rmdir);
3349EXPORT_SYMBOL(vfs_symlink);
3350EXPORT_SYMBOL(vfs_unlink);
3351EXPORT_SYMBOL(dentry_unhash);
3352EXPORT_SYMBOL(generic_readlink);
kernel source for telekom puls (alcatel/mediatek)