projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
split __follow_mount_rcu() into normal and .. cases
[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
287548e4
AV		 954static void follow_mount_rcu(struct nameidata *nd, struct path *path,
955 struct inode **inode)
956{
957 for (;;) {
958 struct vfsmount *mounted;
959 *inode = path->dentry->d_inode;
960
961 if (!d_mountpoint(path->dentry))
962 break;
963
964 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
965 if (!mounted)
966 break;
967 path->mnt = mounted;
968 path->dentry = mounted->mnt_root;
969 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
970 }
971}
972
31e6b01f
NP		 973static int follow_dotdot_rcu(struct nameidata *nd)
974{
975 struct inode *inode = nd->inode;
976
977 set_root_rcu(nd);
978
9875cf80 979 while (1) {
31e6b01f
NP		 980 if (nd->path.dentry == nd->root.dentry &&
981 nd->path.mnt == nd->root.mnt) {
982 break;
983 }
984 if (nd->path.dentry != nd->path.mnt->mnt_root) {
985 struct dentry *old = nd->path.dentry;
986 struct dentry *parent = old->d_parent;
987 unsigned seq;
988
989 seq = read_seqcount_begin(&parent->d_seq);
990 if (read_seqcount_retry(&old->d_seq, nd->seq))
ef7562d5 991 goto failed;
31e6b01f
NP		 992 inode = parent->d_inode;
993 nd->path.dentry = parent;
994 nd->seq = seq;
995 break;
996 }
997 if (!follow_up_rcu(&nd->path))
998 break;
999 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1000 inode = nd->path.dentry->d_inode;
1001 }
287548e4 1002 follow_mount_rcu(nd, &nd->path, &inode);
31e6b01f 1003 nd->inode = inode;
31e6b01f 1004 return 0;
ef7562d5
AV		 1005
1006failed:
1007 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 1008 if (!(nd->flags & LOOKUP_ROOT))
1009 nd->root.mnt = NULL;
ef7562d5
AV		 1010 rcu_read_unlock();
1011 br_read_unlock(vfsmount_lock);
1012 return -ECHILD;
31e6b01f
NP		 1013}
1014
cc53ce53
DH		 1015/*
1016 * Follow down to the covering mount currently visible to userspace. At each
1017 * point, the filesystem owning that dentry may be queried as to whether the
1018 * caller is permitted to proceed or not.
1019 *
1020 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1021 * being true).
1022 */
7cc90cc3 1023int follow_down(struct path *path)
cc53ce53
DH		 1024{
1025 unsigned managed;
1026 int ret;
1027
1028 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1029 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1030 /* Allow the filesystem to manage the transit without i_mutex
1031 * being held.
1032 *
1033 * We indicate to the filesystem if someone is trying to mount
1034 * something here. This gives autofs the chance to deny anyone
1035 * other than its daemon the right to mount on its
1036 * superstructure.
1037 *
1038 * The filesystem may sleep at this point.
1039 */
1040 if (managed & DCACHE_MANAGE_TRANSIT) {
1041 BUG_ON(!path->dentry->d_op);
1042 BUG_ON(!path->dentry->d_op->d_manage);
ab90911f 1043 ret = path->dentry->d_op->d_manage(
1aed3e42 1044 path->dentry, false);
cc53ce53
DH		 1045 if (ret < 0)
1046 return ret == -EISDIR ? 0 : ret;
1047 }
1048
1049 /* Transit to a mounted filesystem. */
1050 if (managed & DCACHE_MOUNTED) {
1051 struct vfsmount *mounted = lookup_mnt(path);
1052 if (!mounted)
1053 break;
1054 dput(path->dentry);
1055 mntput(path->mnt);
1056 path->mnt = mounted;
1057 path->dentry = dget(mounted->mnt_root);
1058 continue;
1059 }
1060
1061 /* Don't handle automount points here */
1062 break;
1063 }
1064 return 0;
1065}
1066
9875cf80
DH		 1067/*
1068 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1069 */
1070static void follow_mount(struct path *path)
1071{
1072 while (d_mountpoint(path->dentry)) {
1073 struct vfsmount *mounted = lookup_mnt(path);
1074 if (!mounted)
1075 break;
1076 dput(path->dentry);
1077 mntput(path->mnt);
1078 path->mnt = mounted;
1079 path->dentry = dget(mounted->mnt_root);
1080 }
1081}
1082
31e6b01f 1083static void follow_dotdot(struct nameidata *nd)
1da177e4 1084{
2a737871 1085 set_root(nd);
e518ddb7 1086
1da177e4 1087 while(1) {
4ac91378 1088 struct dentry *old = nd->path.dentry;
1da177e4 1089
2a737871
AV		 1090 if (nd->path.dentry == nd->root.dentry &&
1091 nd->path.mnt == nd->root.mnt) {
1da177e4
LT		 1092 break;
1093 }
4ac91378 1094 if (nd->path.dentry != nd->path.mnt->mnt_root) {
3088dd70
AV		 1095 /* rare case of legitimate dget_parent()... */
1096 nd->path.dentry = dget_parent(nd->path.dentry);
1da177e4
LT		 1097 dput(old);
1098 break;
1099 }
3088dd70 1100 if (!follow_up(&nd->path))
1da177e4 1101 break;
1da177e4 1102 }
79ed0226 1103 follow_mount(&nd->path);
31e6b01f 1104 nd->inode = nd->path.dentry->d_inode;
1da177e4
LT		 1105}
1106
baa03890
NP		 1107/*
1108 * Allocate a dentry with name and parent, and perform a parent
1109 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1110 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1111 * have verified that no child exists while under i_mutex.
1112 */
1113static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1114 struct qstr *name, struct nameidata *nd)
1115{
1116 struct inode *inode = parent->d_inode;
1117 struct dentry *dentry;
1118 struct dentry *old;
1119
1120 /* Don't create child dentry for a dead directory. */
1121 if (unlikely(IS_DEADDIR(inode)))
1122 return ERR_PTR(-ENOENT);
1123
1124 dentry = d_alloc(parent, name);
1125 if (unlikely(!dentry))
1126 return ERR_PTR(-ENOMEM);
1127
1128 old = inode->i_op->lookup(inode, dentry, nd);
1129 if (unlikely(old)) {
1130 dput(dentry);
1131 dentry = old;
1132 }
1133 return dentry;
1134}
1135
1da177e4
LT		 1136/*
1137 * It's more convoluted than I'd like it to be, but... it's still fairly
1138 * small and for now I'd prefer to have fast path as straight as possible.
1139 * It _is_ time-critical.
1140 */
1141static int do_lookup(struct nameidata *nd, struct qstr *name,
31e6b01f 1142 struct path *path, struct inode **inode)
1da177e4 1143{
4ac91378 1144 struct vfsmount *mnt = nd->path.mnt;
31e6b01f 1145 struct dentry *dentry, *parent = nd->path.dentry;
5a18fff2
AV		 1146 int need_reval = 1;
1147 int status = 1;
9875cf80
DH		 1148 int err;
1149
b04f784e
NP		 1150 /*
1151 * Rename seqlock is not required here because in the off chance
1152 * of a false negative due to a concurrent rename, we're going to
1153 * do the non-racy lookup, below.
1154 */
31e6b01f
NP		 1155 if (nd->flags & LOOKUP_RCU) {
1156 unsigned seq;
31e6b01f
NP		 1157 *inode = nd->inode;
1158 dentry = __d_lookup_rcu(parent, name, &seq, inode);
5a18fff2
AV		 1159 if (!dentry)
1160 goto unlazy;
1161
31e6b01f
NP		 1162 /* Memory barrier in read_seqcount_begin of child is enough */
1163 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1164 return -ECHILD;
31e6b01f 1165 nd->seq = seq;
5a18fff2 1166
24643087 1167 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
5a18fff2
AV		 1168 status = d_revalidate(dentry, nd);
1169 if (unlikely(status <= 0)) {
1170 if (status != -ECHILD)
1171 need_reval = 0;
1172 goto unlazy;
1173 }
24643087 1174 }
31e6b01f
NP		 1175 path->mnt = mnt;
1176 path->dentry = dentry;
287548e4 1177 if (likely(__follow_mount_rcu(nd, path, inode)))
9875cf80 1178 return 0;
5a18fff2 1179unlazy:
19660af7
AV		 1180 if (unlazy_walk(nd, dentry))
1181 return -ECHILD;
5a18fff2
AV		 1182 } else {
1183 dentry = __d_lookup(parent, name);
9875cf80 1184 }
5a18fff2
AV		 1185
1186retry:
1187 if (unlikely(!dentry)) {
1188 struct inode *dir = parent->d_inode;
1189 BUG_ON(nd->inode != dir);
1190
1191 mutex_lock(&dir->i_mutex);
1192 dentry = d_lookup(parent, name);
1193 if (likely(!dentry)) {
1194 dentry = d_alloc_and_lookup(parent, name, nd);
1195 if (IS_ERR(dentry)) {
1196 mutex_unlock(&dir->i_mutex);
1197 return PTR_ERR(dentry);
1198 }
1199 /* known good */
1200 need_reval = 0;
1201 status = 1;
1202 }
1203 mutex_unlock(&dir->i_mutex);
1204 }
1205 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1206 status = d_revalidate(dentry, nd);
1207 if (unlikely(status <= 0)) {
1208 if (status < 0) {
1209 dput(dentry);
1210 return status;
1211 }
1212 if (!d_invalidate(dentry)) {
1213 dput(dentry);
1214 dentry = NULL;
1215 need_reval = 1;
1216 goto retry;
1217 }
24643087 1218 }
5a18fff2 1219
9875cf80
DH		 1220 path->mnt = mnt;
1221 path->dentry = dentry;
1222 err = follow_managed(path, nd->flags);
89312214
IK		 1223 if (unlikely(err < 0)) {
1224 path_put_conditional(path, nd);
9875cf80 1225 return err;
89312214 1226 }
9875cf80 1227 *inode = path->dentry->d_inode;
1da177e4 1228 return 0;
1da177e4
LT		 1229}
1230
52094c8a
AV		 1231static inline int may_lookup(struct nameidata *nd)
1232{
1233 if (nd->flags & LOOKUP_RCU) {
1234 int err = exec_permission(nd->inode, IPERM_FLAG_RCU);
1235 if (err != -ECHILD)
1236 return err;
19660af7 1237 if (unlazy_walk(nd, NULL))
52094c8a
AV		 1238 return -ECHILD;
1239 }
1240 return exec_permission(nd->inode, 0);
1241}
1242
9856fa1b
AV		 1243static inline int handle_dots(struct nameidata *nd, int type)
1244{
1245 if (type == LAST_DOTDOT) {
1246 if (nd->flags & LOOKUP_RCU) {
1247 if (follow_dotdot_rcu(nd))
1248 return -ECHILD;
1249 } else
1250 follow_dotdot(nd);
1251 }
1252 return 0;
1253}
1254
951361f9
AV		 1255static void terminate_walk(struct nameidata *nd)
1256{
1257 if (!(nd->flags & LOOKUP_RCU)) {
1258 path_put(&nd->path);
1259 } else {
1260 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 1261 if (!(nd->flags & LOOKUP_ROOT))
1262 nd->root.mnt = NULL;
951361f9
AV		 1263 rcu_read_unlock();
1264 br_read_unlock(vfsmount_lock);
1265 }
1266}
1267
ce57dfc1
AV		 1268static inline int walk_component(struct nameidata *nd, struct path *path,
1269 struct qstr *name, int type, int follow)
1270{
1271 struct inode *inode;
1272 int err;
1273 /*
1274 * "." and ".." are special - ".." especially so because it has
1275 * to be able to know about the current root directory and
1276 * parent relationships.
1277 */
1278 if (unlikely(type != LAST_NORM))
1279 return handle_dots(nd, type);
1280 err = do_lookup(nd, name, path, &inode);
1281 if (unlikely(err)) {
1282 terminate_walk(nd);
1283 return err;
1284 }
1285 if (!inode) {
1286 path_to_nameidata(path, nd);
1287 terminate_walk(nd);
1288 return -ENOENT;
1289 }
1290 if (unlikely(inode->i_op->follow_link) && follow) {
19660af7
AV		 1291 if (nd->flags & LOOKUP_RCU) {
1292 if (unlikely(unlazy_walk(nd, path->dentry))) {
1293 terminate_walk(nd);
1294 return -ECHILD;
1295 }
1296 }
ce57dfc1
AV		 1297 BUG_ON(inode != path->dentry->d_inode);
1298 return 1;
1299 }
1300 path_to_nameidata(path, nd);
1301 nd->inode = inode;
1302 return 0;
1303}
1304
b356379a
AV		 1305/*
1306 * This limits recursive symlink follows to 8, while
1307 * limiting consecutive symlinks to 40.
1308 *
1309 * Without that kind of total limit, nasty chains of consecutive
1310 * symlinks can cause almost arbitrarily long lookups.
1311 */
1312static inline int nested_symlink(struct path *path, struct nameidata *nd)
1313{
1314 int res;
1315
b356379a
AV		 1316 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1317 path_put_conditional(path, nd);
1318 path_put(&nd->path);
1319 return -ELOOP;
1320 }
1a4022f8 1321 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
b356379a
AV		 1322
1323 nd->depth++;
1324 current->link_count++;
1325
1326 do {
1327 struct path link = *path;
1328 void *cookie;
574197e0
AV		 1329
1330 res = follow_link(&link, nd, &cookie);
b356379a
AV		 1331 if (!res)
1332 res = walk_component(nd, path, &nd->last,
1333 nd->last_type, LOOKUP_FOLLOW);
574197e0 1334 put_link(nd, &link, cookie);
b356379a
AV		 1335 } while (res > 0);
1336
1337 current->link_count--;
1338 nd->depth--;
1339 return res;
1340}
1341
1da177e4
LT		 1342/*
1343 * Name resolution.
ea3834d9
PM		 1344 * This is the basic name resolution function, turning a pathname into
1345 * the final dentry. We expect 'base' to be positive and a directory.
1da177e4 1346 *
ea3834d9
PM		 1347 * Returns 0 and nd will have valid dentry and mnt on success.
1348 * Returns error and drops reference to input namei data on failure.
1da177e4 1349 */
6de88d72 1350static int link_path_walk(const char *name, struct nameidata *nd)
1da177e4
LT		 1351{
1352 struct path next;
1da177e4
LT		 1353 int err;
1354 unsigned int lookup_flags = nd->flags;
1355
1356 while (*name=='/')
1357 name++;
1358 if (!*name)
086e183a 1359 return 0;
1da177e4 1360
1da177e4
LT		 1361 /* At this point we know we have a real path component. */
1362 for(;;) {
1363 unsigned long hash;
1364 struct qstr this;
1365 unsigned int c;
fe479a58 1366 int type;
1da177e4 1367
cdce5d6b 1368 nd->flags |= LOOKUP_CONTINUE;
52094c8a
AV		 1369
1370 err = may_lookup(nd);
1da177e4
LT		 1371 if (err)
1372 break;
1373
1374 this.name = name;
1375 c = *(const unsigned char *)name;
1376
1377 hash = init_name_hash();
1378 do {
1379 name++;
1380 hash = partial_name_hash(c, hash);
1381 c = *(const unsigned char *)name;
1382 } while (c && (c != '/'));
1383 this.len = name - (const char *) this.name;
1384 this.hash = end_name_hash(hash);
1385
fe479a58
AV		 1386 type = LAST_NORM;
1387 if (this.name[0] == '.') switch (this.len) {
1388 case 2:
16c2cd71 1389 if (this.name[1] == '.') {
fe479a58 1390 type = LAST_DOTDOT;
16c2cd71
AV		 1391 nd->flags |= LOOKUP_JUMPED;
1392 }
fe479a58
AV		 1393 break;
1394 case 1:
1395 type = LAST_DOT;
1396 }
5a202bcd
AV		 1397 if (likely(type == LAST_NORM)) {
1398 struct dentry *parent = nd->path.dentry;
16c2cd71 1399 nd->flags &= ~LOOKUP_JUMPED;
5a202bcd
AV		 1400 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1401 err = parent->d_op->d_hash(parent, nd->inode,
1402 &this);
1403 if (err < 0)
1404 break;
1405 }
1406 }
fe479a58 1407
1da177e4
LT		 1408 /* remove trailing slashes? */
1409 if (!c)
1410 goto last_component;
1411 while (*++name == '/');
1412 if (!*name)
b356379a 1413 goto last_component;
1da177e4 1414
ce57dfc1
AV		 1415 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1416 if (err < 0)
1417 return err;
1da177e4 1418
ce57dfc1 1419 if (err) {
b356379a 1420 err = nested_symlink(&next, nd);
1da177e4 1421 if (err)
a7472bab 1422 return err;
31e6b01f 1423 }
1da177e4 1424 err = -ENOTDIR;
31e6b01f 1425 if (!nd->inode->i_op->lookup)
1da177e4
LT		 1426 break;
1427 continue;
1428 /* here ends the main loop */
1429
1da177e4 1430last_component:
f55eab82
TM		 1431 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1432 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
b356379a
AV		 1433 nd->last = this;
1434 nd->last_type = type;
086e183a 1435 return 0;
1da177e4 1436 }
951361f9 1437 terminate_walk(nd);
1da177e4
LT		 1438 return err;
1439}
1440
70e9b357
AV		 1441static int path_init(int dfd, const char *name, unsigned int flags,
1442 struct nameidata *nd, struct file **fp)
31e6b01f
NP		 1443{
1444 int retval = 0;
1445 int fput_needed;
1446 struct file *file;
1447
1448 nd->last_type = LAST_ROOT; /* if there are only slashes... */
16c2cd71 1449 nd->flags = flags | LOOKUP_JUMPED;
31e6b01f 1450 nd->depth = 0;
5b6ca027
AV		 1451 if (flags & LOOKUP_ROOT) {
1452 struct inode *inode = nd->root.dentry->d_inode;
73d049a4
AV		 1453 if (*name) {
1454 if (!inode->i_op->lookup)
1455 return -ENOTDIR;
1456 retval = inode_permission(inode, MAY_EXEC);
1457 if (retval)
1458 return retval;
1459 }
5b6ca027
AV		 1460 nd->path = nd->root;
1461 nd->inode = inode;
1462 if (flags & LOOKUP_RCU) {
1463 br_read_lock(vfsmount_lock);
1464 rcu_read_lock();
1465 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1466 } else {
1467 path_get(&nd->path);
1468 }
1469 return 0;
1470 }
1471
31e6b01f 1472 nd->root.mnt = NULL;
31e6b01f
NP		 1473
1474 if (*name=='/') {
e41f7d4e
AV		 1475 if (flags & LOOKUP_RCU) {
1476 br_read_lock(vfsmount_lock);
1477 rcu_read_lock();
1478 set_root_rcu(nd);
1479 } else {
1480 set_root(nd);
1481 path_get(&nd->root);
1482 }
1483 nd->path = nd->root;
31e6b01f 1484 } else if (dfd == AT_FDCWD) {
e41f7d4e
AV		 1485 if (flags & LOOKUP_RCU) {
1486 struct fs_struct *fs = current->fs;
1487 unsigned seq;
31e6b01f 1488
e41f7d4e
AV		 1489 br_read_lock(vfsmount_lock);
1490 rcu_read_lock();
c28cc364 1491
e41f7d4e
AV		 1492 do {
1493 seq = read_seqcount_begin(&fs->seq);
1494 nd->path = fs->pwd;
1495 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1496 } while (read_seqcount_retry(&fs->seq, seq));
1497 } else {
1498 get_fs_pwd(current->fs, &nd->path);
1499 }
31e6b01f
NP		 1500 } else {
1501 struct dentry *dentry;
1502
1abf0c71 1503 file = fget_raw_light(dfd, &fput_needed);
31e6b01f
NP		 1504 retval = -EBADF;
1505 if (!file)
1506 goto out_fail;
1507
1508 dentry = file->f_path.dentry;
1509
f52e0c11
AV		 1510 if (*name) {
1511 retval = -ENOTDIR;
1512 if (!S_ISDIR(dentry->d_inode->i_mode))
1513 goto fput_fail;
31e6b01f 1514
f52e0c11
AV		 1515 retval = file_permission(file, MAY_EXEC);
1516 if (retval)
1517 goto fput_fail;
1518 }
31e6b01f
NP		 1519
1520 nd->path = file->f_path;
e41f7d4e
AV		 1521 if (flags & LOOKUP_RCU) {
1522 if (fput_needed)
70e9b357 1523 *fp = file;
e41f7d4e
AV		 1524 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1525 br_read_lock(vfsmount_lock);
1526 rcu_read_lock();
1527 } else {
1528 path_get(&file->f_path);
1529 fput_light(file, fput_needed);
1530 }
31e6b01f 1531 }
31e6b01f 1532
31e6b01f 1533 nd->inode = nd->path.dentry->d_inode;
9b4a9b14 1534 return 0;
2dfdd266 1535
9b4a9b14
AV		 1536fput_fail:
1537 fput_light(file, fput_needed);
1538out_fail:
1539 return retval;
1540}
1541
bd92d7fe
AV		 1542static inline int lookup_last(struct nameidata *nd, struct path *path)
1543{
1544 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1545 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1546
1547 nd->flags &= ~LOOKUP_PARENT;
1548 return walk_component(nd, path, &nd->last, nd->last_type,
1549 nd->flags & LOOKUP_FOLLOW);
1550}
1551
9b4a9b14 1552/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
ee0827cd 1553static int path_lookupat(int dfd, const char *name,
9b4a9b14
AV		 1554 unsigned int flags, struct nameidata *nd)
1555{
70e9b357 1556 struct file *base = NULL;
bd92d7fe
AV		 1557 struct path path;
1558 int err;
31e6b01f
NP		 1559
1560 /*
1561 * Path walking is largely split up into 2 different synchronisation
1562 * schemes, rcu-walk and ref-walk (explained in
1563 * Documentation/filesystems/path-lookup.txt). These share much of the
1564 * path walk code, but some things particularly setup, cleanup, and
1565 * following mounts are sufficiently divergent that functions are
1566 * duplicated. Typically there is a function foo(), and its RCU
1567 * analogue, foo_rcu().
1568 *
1569 * -ECHILD is the error number of choice (just to avoid clashes) that
1570 * is returned if some aspect of an rcu-walk fails. Such an error must
1571 * be handled by restarting a traditional ref-walk (which will always
1572 * be able to complete).
1573 */
bd92d7fe 1574 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
ee0827cd 1575
bd92d7fe
AV		 1576 if (unlikely(err))
1577 return err;
ee0827cd
AV		 1578
1579 current->total_link_count = 0;
bd92d7fe
AV		 1580 err = link_path_walk(name, nd);
1581
1582 if (!err && !(flags & LOOKUP_PARENT)) {
bd92d7fe
AV		 1583 err = lookup_last(nd, &path);
1584 while (err > 0) {
1585 void *cookie;
1586 struct path link = path;
bd92d7fe 1587 nd->flags |= LOOKUP_PARENT;
574197e0 1588 err = follow_link(&link, nd, &cookie);
bd92d7fe
AV		 1589 if (!err)
1590 err = lookup_last(nd, &path);
574197e0 1591 put_link(nd, &link, cookie);
bd92d7fe
AV		 1592 }
1593 }
ee0827cd 1594
9f1fafee
AV		 1595 if (!err)
1596 err = complete_walk(nd);
bd92d7fe
AV		 1597
1598 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1599 if (!nd->inode->i_op->lookup) {
1600 path_put(&nd->path);
bd23a539 1601 err = -ENOTDIR;
bd92d7fe
AV		 1602 }
1603 }
16c2cd71 1604
70e9b357
AV		 1605 if (base)
1606 fput(base);
ee0827cd 1607
5b6ca027 1608 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
2a737871
AV		 1609 path_put(&nd->root);
1610 nd->root.mnt = NULL;
1611 }
bd92d7fe 1612 return err;
ee0827cd 1613}
31e6b01f 1614
ee0827cd
AV		 1615static int do_path_lookup(int dfd, const char *name,
1616 unsigned int flags, struct nameidata *nd)
1617{
1618 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1619 if (unlikely(retval == -ECHILD))
1620 retval = path_lookupat(dfd, name, flags, nd);
1621 if (unlikely(retval == -ESTALE))
1622 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
31e6b01f
NP		 1623
1624 if (likely(!retval)) {
1625 if (unlikely(!audit_dummy_context())) {
1626 if (nd->path.dentry && nd->inode)
1627 audit_inode(name, nd->path.dentry);
1628 }
1629 }
170aa3d0 1630 return retval;
1da177e4
LT		 1631}
1632
c9c6cac0 1633int kern_path_parent(const char *name, struct nameidata *nd)
5590ff0d 1634{
c9c6cac0 1635 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
5590ff0d
UD		 1636}
1637
d1811465
AV		 1638int kern_path(const char *name, unsigned int flags, struct path *path)
1639{
1640 struct nameidata nd;
1641 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1642 if (!res)
1643 *path = nd.path;
1644 return res;
1645}
1646
16f18200
JJS		 1647/**
1648 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1649 * @dentry: pointer to dentry of the base directory
1650 * @mnt: pointer to vfs mount of the base directory
1651 * @name: pointer to file name
1652 * @flags: lookup flags
1653 * @nd: pointer to nameidata
1654 */
1655int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1656 const char *name, unsigned int flags,
1657 struct nameidata *nd)
1658{
5b6ca027
AV		 1659 nd->root.dentry = dentry;
1660 nd->root.mnt = mnt;
1661 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1662 return do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, nd);
16f18200
JJS		 1663}
1664
eead1911
CH		 1665static struct dentry *__lookup_hash(struct qstr *name,
1666 struct dentry *base, struct nameidata *nd)
1da177e4 1667{
81fca444 1668 struct inode *inode = base->d_inode;
057f6c01 1669 struct dentry *dentry;
1da177e4
LT		 1670 int err;
1671
b74c79e9 1672 err = exec_permission(inode, 0);
81fca444
CH		 1673 if (err)
1674 return ERR_PTR(err);
1da177e4 1675
b04f784e
NP		 1676 /*
1677 * Don't bother with __d_lookup: callers are for creat as
1678 * well as unlink, so a lot of the time it would cost
1679 * a double lookup.
6e6b1bd1 1680 */
b04f784e 1681 dentry = d_lookup(base, name);
6e6b1bd1 1682
fb045adb 1683 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE))
6e6b1bd1
AV		 1684 dentry = do_revalidate(dentry, nd);
1685
baa03890
NP		 1686 if (!dentry)
1687 dentry = d_alloc_and_lookup(base, name, nd);
5a202bcd 1688
1da177e4
LT		 1689 return dentry;
1690}
1691
057f6c01
JM		 1692/*
1693 * Restricted form of lookup. Doesn't follow links, single-component only,
1694 * needs parent already locked. Doesn't follow mounts.
1695 * SMP-safe.
1696 */
eead1911 1697static struct dentry *lookup_hash(struct nameidata *nd)
057f6c01 1698{
4ac91378 1699 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1da177e4
LT		 1700}
1701
eead1911 1702/**
a6b91919 1703 * lookup_one_len - filesystem helper to lookup single pathname component
eead1911
CH		 1704 * @name: pathname component to lookup
1705 * @base: base directory to lookup from
1706 * @len: maximum length @len should be interpreted to
1707 *
a6b91919
RD		 1708 * Note that this routine is purely a helper for filesystem usage and should
1709 * not be called by generic code. Also note that by using this function the
eead1911
CH		 1710 * nameidata argument is passed to the filesystem methods and a filesystem
1711 * using this helper needs to be prepared for that.
1712 */
057f6c01
JM		 1713struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1714{
057f6c01 1715 struct qstr this;
6a96ba54
AV		 1716 unsigned long hash;
1717 unsigned int c;
057f6c01 1718
2f9092e1
DW		 1719 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1720
6a96ba54
AV		 1721 this.name = name;
1722 this.len = len;
1723 if (!len)
1724 return ERR_PTR(-EACCES);
1725
1726 hash = init_name_hash();
1727 while (len--) {
1728 c = *(const unsigned char *)name++;
1729 if (c == '/' || c == '\0')
1730 return ERR_PTR(-EACCES);
1731 hash = partial_name_hash(c, hash);
1732 }
1733 this.hash = end_name_hash(hash);
5a202bcd
AV		 1734 /*
1735 * See if the low-level filesystem might want
1736 * to use its own hash..
1737 */
1738 if (base->d_flags & DCACHE_OP_HASH) {
1739 int err = base->d_op->d_hash(base, base->d_inode, &this);
1740 if (err < 0)
1741 return ERR_PTR(err);
1742 }
eead1911 1743
49705b77 1744 return __lookup_hash(&this, base, NULL);
057f6c01
JM		 1745}
1746
2d8f3038
AV		 1747int user_path_at(int dfd, const char __user *name, unsigned flags,
1748 struct path *path)
1da177e4 1749{
2d8f3038 1750 struct nameidata nd;
f52e0c11 1751 char *tmp = getname_flags(name, flags);
1da177e4 1752 int err = PTR_ERR(tmp);
1da177e4 1753 if (!IS_ERR(tmp)) {
2d8f3038
AV		 1754
1755 BUG_ON(flags & LOOKUP_PARENT);
1756
1757 err = do_path_lookup(dfd, tmp, flags, &nd);
1da177e4 1758 putname(tmp);
2d8f3038
AV		 1759 if (!err)
1760 *path = nd.path;
1da177e4
LT		 1761 }
1762 return err;
1763}
1764
2ad94ae6
AV		 1765static int user_path_parent(int dfd, const char __user *path,
1766 struct nameidata *nd, char **name)
1767{
1768 char *s = getname(path);
1769 int error;
1770
1771 if (IS_ERR(s))
1772 return PTR_ERR(s);
1773
1774 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1775 if (error)
1776 putname(s);
1777 else
1778 *name = s;
1779
1780 return error;
1781}
1782
1da177e4
LT		 1783/*
1784 * It's inline, so penalty for filesystems that don't use sticky bit is
1785 * minimal.
1786 */
1787static inline int check_sticky(struct inode *dir, struct inode *inode)
1788{
da9592ed
DH		 1789 uid_t fsuid = current_fsuid();
1790
1da177e4
LT		 1791 if (!(dir->i_mode & S_ISVTX))
1792 return 0;
e795b717
SH		 1793 if (current_user_ns() != inode_userns(inode))
1794 goto other_userns;
da9592ed 1795 if (inode->i_uid == fsuid)
1da177e4 1796 return 0;
da9592ed 1797 if (dir->i_uid == fsuid)
1da177e4 1798 return 0;
e795b717
SH		 1799
1800other_userns:
1801 return !ns_capable(inode_userns(inode), CAP_FOWNER);
1da177e4
LT		 1802}
1803
1804/*
1805 * Check whether we can remove a link victim from directory dir, check
1806 * whether the type of victim is right.
1807 * 1. We can't do it if dir is read-only (done in permission())
1808 * 2. We should have write and exec permissions on dir
1809 * 3. We can't remove anything from append-only dir
1810 * 4. We can't do anything with immutable dir (done in permission())
1811 * 5. If the sticky bit on dir is set we should either
1812 * a. be owner of dir, or
1813 * b. be owner of victim, or
1814 * c. have CAP_FOWNER capability
1815 * 6. If the victim is append-only or immutable we can't do antyhing with
1816 * links pointing to it.
1817 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1818 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1819 * 9. We can't remove a root or mountpoint.
1820 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1821 * nfs_async_unlink().
1822 */
858119e1 1823static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1da177e4
LT		 1824{
1825 int error;
1826
1827 if (!victim->d_inode)
1828 return -ENOENT;
1829
1830 BUG_ON(victim->d_parent->d_inode != dir);
cccc6bba 1831 audit_inode_child(victim, dir);
1da177e4 1832
f419a2e3 1833 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1834 if (error)
1835 return error;
1836 if (IS_APPEND(dir))
1837 return -EPERM;
1838 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
f9454548 1839 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1da177e4
LT		 1840 return -EPERM;
1841 if (isdir) {
1842 if (!S_ISDIR(victim->d_inode->i_mode))
1843 return -ENOTDIR;
1844 if (IS_ROOT(victim))
1845 return -EBUSY;
1846 } else if (S_ISDIR(victim->d_inode->i_mode))
1847 return -EISDIR;
1848 if (IS_DEADDIR(dir))
1849 return -ENOENT;
1850 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1851 return -EBUSY;
1852 return 0;
1853}
1854
1855/* Check whether we can create an object with dentry child in directory
1856 * dir.
1857 * 1. We can't do it if child already exists (open has special treatment for
1858 * this case, but since we are inlined it's OK)
1859 * 2. We can't do it if dir is read-only (done in permission())
1860 * 3. We should have write and exec permissions on dir
1861 * 4. We can't do it if dir is immutable (done in permission())
1862 */
a95164d9 1863static inline int may_create(struct inode *dir, struct dentry *child)
1da177e4
LT		 1864{
1865 if (child->d_inode)
1866 return -EEXIST;
1867 if (IS_DEADDIR(dir))
1868 return -ENOENT;
f419a2e3 1869 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1870}
1871
1da177e4
LT		 1872/*
1873 * p1 and p2 should be directories on the same fs.
1874 */
1875struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1876{
1877 struct dentry *p;
1878
1879 if (p1 == p2) {
f2eace23 1880 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1da177e4
LT		 1881 return NULL;
1882 }
1883
a11f3a05 1884 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4 1885
e2761a11
OH		 1886 p = d_ancestor(p2, p1);
1887 if (p) {
1888 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1889 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1890 return p;
1da177e4
LT		 1891 }
1892
e2761a11
OH		 1893 p = d_ancestor(p1, p2);
1894 if (p) {
1895 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1896 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1897 return p;
1da177e4
LT		 1898 }
1899
f2eace23
IM		 1900 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1901 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1da177e4
LT		 1902 return NULL;
1903}
1904
1905void unlock_rename(struct dentry *p1, struct dentry *p2)
1906{
1b1dcc1b 1907 mutex_unlock(&p1->d_inode->i_mutex);
1da177e4 1908 if (p1 != p2) {
1b1dcc1b 1909 mutex_unlock(&p2->d_inode->i_mutex);
a11f3a05 1910 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4
LT		 1911 }
1912}
1913
1914int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1915 struct nameidata *nd)
1916{
a95164d9 1917 int error = may_create(dir, dentry);
1da177e4
LT		 1918
1919 if (error)
1920 return error;
1921
acfa4380 1922 if (!dir->i_op->create)
1da177e4
LT		 1923 return -EACCES; /* shouldn't it be ENOSYS? */
1924 mode &= S_IALLUGO;
1925 mode |= S_IFREG;
1926 error = security_inode_create(dir, dentry, mode);
1927 if (error)
1928 return error;
1da177e4 1929 error = dir->i_op->create(dir, dentry, mode, nd);
a74574aa 1930 if (!error)
f38aa942 1931 fsnotify_create(dir, dentry);
1da177e4
LT		 1932 return error;
1933}
1934
73d049a4 1935static int may_open(struct path *path, int acc_mode, int flag)
1da177e4 1936{
3fb64190 1937 struct dentry *dentry = path->dentry;
1da177e4
LT		 1938 struct inode *inode = dentry->d_inode;
1939 int error;
1940
bcda7652
AV		 1941 /* O_PATH? */
1942 if (!acc_mode)
1943 return 0;
1944
1da177e4
LT		 1945 if (!inode)
1946 return -ENOENT;
1947
c8fe8f30
CH		 1948 switch (inode->i_mode & S_IFMT) {
1949 case S_IFLNK:
1da177e4 1950 return -ELOOP;
c8fe8f30
CH		 1951 case S_IFDIR:
1952 if (acc_mode & MAY_WRITE)
1953 return -EISDIR;
1954 break;
1955 case S_IFBLK:
1956 case S_IFCHR:
3fb64190 1957 if (path->mnt->mnt_flags & MNT_NODEV)
1da177e4 1958 return -EACCES;
c8fe8f30
CH		 1959 /*FALLTHRU*/
1960 case S_IFIFO:
1961 case S_IFSOCK:
1da177e4 1962 flag &= ~O_TRUNC;
c8fe8f30 1963 break;
4a3fd211 1964 }
b41572e9 1965
3fb64190 1966 error = inode_permission(inode, acc_mode);
b41572e9
DH		 1967 if (error)
1968 return error;
6146f0d5 1969
1da177e4
LT		 1970 /*
1971 * An append-only file must be opened in append mode for writing.
1972 */
1973 if (IS_APPEND(inode)) {
8737c930 1974 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
7715b521 1975 return -EPERM;
1da177e4 1976 if (flag & O_TRUNC)
7715b521 1977 return -EPERM;
1da177e4
LT		 1978 }
1979
1980 /* O_NOATIME can only be set by the owner or superuser */
2e149670 1981 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
7715b521 1982 return -EPERM;
1da177e4
LT		 1983
1984 /*
1985 * Ensure there are no outstanding leases on the file.
1986 */
b65a9cfc 1987 return break_lease(inode, flag);
7715b521 1988}
1da177e4 1989
e1181ee6 1990static int handle_truncate(struct file *filp)
7715b521 1991{
e1181ee6 1992 struct path *path = &filp->f_path;
7715b521
AV		 1993 struct inode *inode = path->dentry->d_inode;
1994 int error = get_write_access(inode);
1995 if (error)
1996 return error;
1997 /*
1998 * Refuse to truncate files with mandatory locks held on them.
1999 */
2000 error = locks_verify_locked(inode);
2001 if (!error)
ea0d3ab2 2002 error = security_path_truncate(path);
7715b521
AV		 2003 if (!error) {
2004 error = do_truncate(path->dentry, 0,
2005 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
e1181ee6 2006 filp);
7715b521
AV		 2007 }
2008 put_write_access(inode);
acd0c935 2009 return error;
1da177e4
LT		 2010}
2011
d57999e1
DH		 2012/*
2013 * Note that while the flag value (low two bits) for sys_open means:
2014 * 00 - read-only
2015 * 01 - write-only
2016 * 10 - read-write
2017 * 11 - special
2018 * it is changed into
2019 * 00 - no permissions needed
2020 * 01 - read-permission
2021 * 10 - write-permission
2022 * 11 - read-write
2023 * for the internal routines (ie open_namei()/follow_link() etc)
2024 * This is more logical, and also allows the 00 "no perm needed"
2025 * to be used for symlinks (where the permissions are checked
2026 * later).
2027 *
2028*/
2029static inline int open_to_namei_flags(int flag)
2030{
2031 if ((flag+1) & O_ACCMODE)
2032 flag++;
2033 return flag;
2034}
2035
31e6b01f 2036/*
fe2d35ff 2037 * Handle the last step of open()
31e6b01f 2038 */
fb1cc555 2039static struct file *do_last(struct nameidata *nd, struct path *path,
c3e380b0 2040 const struct open_flags *op, const char *pathname)
fb1cc555 2041{
a1e28038 2042 struct dentry *dir = nd->path.dentry;
6c0d46c4 2043 struct dentry *dentry;
ca344a89 2044 int open_flag = op->open_flag;
6c0d46c4 2045 int will_truncate = open_flag & O_TRUNC;
ca344a89 2046 int want_write = 0;
bcda7652 2047 int acc_mode = op->acc_mode;
fb1cc555 2048 struct file *filp;
16c2cd71 2049 int error;
1f36f774 2050
c3e380b0
AV		 2051 nd->flags &= ~LOOKUP_PARENT;
2052 nd->flags |= op->intent;
2053
1f36f774
AV		 2054 switch (nd->last_type) {
2055 case LAST_DOTDOT:
176306f5 2056 case LAST_DOT:
fe2d35ff
AV		 2057 error = handle_dots(nd, nd->last_type);
2058 if (error)
2059 return ERR_PTR(error);
1f36f774 2060 /* fallthrough */
1f36f774 2061 case LAST_ROOT:
9f1fafee 2062 error = complete_walk(nd);
16c2cd71 2063 if (error)
9f1fafee 2064 return ERR_PTR(error);
fe2d35ff 2065 audit_inode(pathname, nd->path.dentry);
ca344a89 2066 if (open_flag & O_CREAT) {
fe2d35ff
AV		 2067 error = -EISDIR;
2068 goto exit;
2069 }
2070 goto ok;
1f36f774 2071 case LAST_BIND:
9f1fafee 2072 error = complete_walk(nd);
16c2cd71 2073 if (error)
9f1fafee 2074 return ERR_PTR(error);
1f36f774 2075 audit_inode(pathname, dir);
67ee3ad2 2076 goto ok;
1f36f774 2077 }
67ee3ad2 2078
ca344a89 2079 if (!(open_flag & O_CREAT)) {
bcda7652 2080 int symlink_ok = 0;
fe2d35ff
AV		 2081 if (nd->last.name[nd->last.len])
2082 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
bcda7652
AV		 2083 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2084 symlink_ok = 1;
fe2d35ff 2085 /* we _can_ be in RCU mode here */
ce57dfc1
AV		 2086 error = walk_component(nd, path, &nd->last, LAST_NORM,
2087 !symlink_ok);
2088 if (error < 0)
fe2d35ff 2089 return ERR_PTR(error);
ce57dfc1 2090 if (error) /* symlink */
fe2d35ff 2091 return NULL;
fe2d35ff 2092 /* sayonara */
9f1fafee
AV		 2093 error = complete_walk(nd);
2094 if (error)
2095 return ERR_PTR(-ECHILD);
fe2d35ff
AV		 2096
2097 error = -ENOTDIR;
2098 if (nd->flags & LOOKUP_DIRECTORY) {
ce57dfc1 2099 if (!nd->inode->i_op->lookup)
fe2d35ff
AV		 2100 goto exit;
2101 }
2102 audit_inode(pathname, nd->path.dentry);
2103 goto ok;
2104 }
2105
2106 /* create side of things */
9f1fafee
AV		 2107 error = complete_walk(nd);
2108 if (error)
2109 return ERR_PTR(error);
fe2d35ff
AV		 2110
2111 audit_inode(pathname, dir);
16c2cd71 2112 error = -EISDIR;
1f36f774 2113 /* trailing slashes? */
31e6b01f
NP		 2114 if (nd->last.name[nd->last.len])
2115 goto exit;
a2c36b45 2116
a1e28038
AV		 2117 mutex_lock(&dir->d_inode->i_mutex);
2118
6c0d46c4
AV		 2119 dentry = lookup_hash(nd);
2120 error = PTR_ERR(dentry);
2121 if (IS_ERR(dentry)) {
fb1cc555
AV		 2122 mutex_unlock(&dir->d_inode->i_mutex);
2123 goto exit;
2124 }
2125
6c0d46c4
AV		 2126 path->dentry = dentry;
2127 path->mnt = nd->path.mnt;
2128
fb1cc555 2129 /* Negative dentry, just create the file */
6c0d46c4
AV		 2130 if (!dentry->d_inode) {
2131 int mode = op->mode;
2132 if (!IS_POSIXACL(dir->d_inode))
2133 mode &= ~current_umask();
fb1cc555
AV		 2134 /*
2135 * This write is needed to ensure that a
6c0d46c4 2136 * rw->ro transition does not occur between
fb1cc555
AV		 2137 * the time when the file is created and when
2138 * a permanent write count is taken through
2139 * the 'struct file' in nameidata_to_filp().
2140 */
2141 error = mnt_want_write(nd->path.mnt);
2142 if (error)
2143 goto exit_mutex_unlock;
ca344a89 2144 want_write = 1;
9b44f1b3 2145 /* Don't check for write permission, don't truncate */
ca344a89 2146 open_flag &= ~O_TRUNC;
6c0d46c4 2147 will_truncate = 0;
bcda7652 2148 acc_mode = MAY_OPEN;
6c0d46c4
AV		 2149 error = security_path_mknod(&nd->path, dentry, mode, 0);
2150 if (error)
2151 goto exit_mutex_unlock;
2152 error = vfs_create(dir->d_inode, dentry, mode, nd);
2153 if (error)
2154 goto exit_mutex_unlock;
2155 mutex_unlock(&dir->d_inode->i_mutex);
2156 dput(nd->path.dentry);
2157 nd->path.dentry = dentry;
ca344a89 2158 goto common;
fb1cc555
AV		 2159 }
2160
2161 /*
2162 * It already exists.
2163 */
2164 mutex_unlock(&dir->d_inode->i_mutex);
2165 audit_inode(pathname, path->dentry);
2166
2167 error = -EEXIST;
ca344a89 2168 if (open_flag & O_EXCL)
fb1cc555
AV		 2169 goto exit_dput;
2170
9875cf80
DH		 2171 error = follow_managed(path, nd->flags);
2172 if (error < 0)
2173 goto exit_dput;
fb1cc555
AV		 2174
2175 error = -ENOENT;
2176 if (!path->dentry->d_inode)
2177 goto exit_dput;
9e67f361
AV		 2178
2179 if (path->dentry->d_inode->i_op->follow_link)
fb1cc555 2180 return NULL;
fb1cc555
AV		 2181
2182 path_to_nameidata(path, nd);
31e6b01f 2183 nd->inode = path->dentry->d_inode;
fb1cc555 2184 error = -EISDIR;
31e6b01f 2185 if (S_ISDIR(nd->inode->i_mode))
fb1cc555 2186 goto exit;
67ee3ad2 2187ok:
6c0d46c4
AV		 2188 if (!S_ISREG(nd->inode->i_mode))
2189 will_truncate = 0;
2190
0f9d1a10
AV		 2191 if (will_truncate) {
2192 error = mnt_want_write(nd->path.mnt);
2193 if (error)
2194 goto exit;
ca344a89 2195 want_write = 1;
0f9d1a10 2196 }
ca344a89 2197common:
bcda7652 2198 error = may_open(&nd->path, acc_mode, open_flag);
ca344a89 2199 if (error)
0f9d1a10 2200 goto exit;
0f9d1a10
AV		 2201 filp = nameidata_to_filp(nd);
2202 if (!IS_ERR(filp)) {
2203 error = ima_file_check(filp, op->acc_mode);
2204 if (error) {
2205 fput(filp);
2206 filp = ERR_PTR(error);
2207 }
2208 }
2209 if (!IS_ERR(filp)) {
2210 if (will_truncate) {
2211 error = handle_truncate(filp);
2212 if (error) {
2213 fput(filp);
2214 filp = ERR_PTR(error);
2215 }
2216 }
2217 }
ca344a89
AV		 2218out:
2219 if (want_write)
0f9d1a10
AV		 2220 mnt_drop_write(nd->path.mnt);
2221 path_put(&nd->path);
fb1cc555
AV		 2222 return filp;
2223
2224exit_mutex_unlock:
2225 mutex_unlock(&dir->d_inode->i_mutex);
2226exit_dput:
2227 path_put_conditional(path, nd);
2228exit:
ca344a89
AV		 2229 filp = ERR_PTR(error);
2230 goto out;
fb1cc555
AV		 2231}
2232
13aab428 2233static struct file *path_openat(int dfd, const char *pathname,
73d049a4 2234 struct nameidata *nd, const struct open_flags *op, int flags)
1da177e4 2235{
fe2d35ff 2236 struct file *base = NULL;
4a3fd211 2237 struct file *filp;
9850c056 2238 struct path path;
13aab428 2239 int error;
31e6b01f
NP		 2240
2241 filp = get_empty_filp();
2242 if (!filp)
2243 return ERR_PTR(-ENFILE);
2244
47c805dc 2245 filp->f_flags = op->open_flag;
73d049a4
AV		 2246 nd->intent.open.file = filp;
2247 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2248 nd->intent.open.create_mode = op->mode;
31e6b01f 2249
73d049a4 2250 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
31e6b01f 2251 if (unlikely(error))
13aab428 2252 goto out_filp;
31e6b01f 2253
fe2d35ff 2254 current->total_link_count = 0;
73d049a4 2255 error = link_path_walk(pathname, nd);
31e6b01f
NP		 2256 if (unlikely(error))
2257 goto out_filp;
1da177e4 2258
73d049a4 2259 filp = do_last(nd, &path, op, pathname);
806b681c 2260 while (unlikely(!filp)) { /* trailing symlink */
7b9337aa 2261 struct path link = path;
def4af30 2262 void *cookie;
574197e0 2263 if (!(nd->flags & LOOKUP_FOLLOW)) {
73d049a4
AV		 2264 path_put_conditional(&path, nd);
2265 path_put(&nd->path);
40b39136
AV		 2266 filp = ERR_PTR(-ELOOP);
2267 break;
2268 }
73d049a4
AV		 2269 nd->flags |= LOOKUP_PARENT;
2270 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
574197e0 2271 error = follow_link(&link, nd, &cookie);
c3e380b0 2272 if (unlikely(error))
f1afe9ef 2273 filp = ERR_PTR(error);
c3e380b0 2274 else
73d049a4 2275 filp = do_last(nd, &path, op, pathname);
574197e0 2276 put_link(nd, &link, cookie);
806b681c 2277 }
10fa8e62 2278out:
73d049a4
AV		 2279 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2280 path_put(&nd->root);
fe2d35ff
AV		 2281 if (base)
2282 fput(base);
73d049a4 2283 release_open_intent(nd);
10fa8e62 2284 return filp;
1da177e4 2285
31e6b01f 2286out_filp:
806b681c 2287 filp = ERR_PTR(error);
10fa8e62 2288 goto out;
1da177e4
LT		 2289}
2290
13aab428
AV		 2291struct file *do_filp_open(int dfd, const char *pathname,
2292 const struct open_flags *op, int flags)
2293{
73d049a4 2294 struct nameidata nd;
13aab428
AV		 2295 struct file *filp;
2296
73d049a4 2297 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
13aab428 2298 if (unlikely(filp == ERR_PTR(-ECHILD)))
73d049a4 2299 filp = path_openat(dfd, pathname, &nd, op, flags);
13aab428 2300 if (unlikely(filp == ERR_PTR(-ESTALE)))
73d049a4 2301 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
13aab428
AV		 2302 return filp;
2303}
2304
73d049a4
AV		 2305struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2306 const char *name, const struct open_flags *op, int flags)
2307{
2308 struct nameidata nd;
2309 struct file *file;
2310
2311 nd.root.mnt = mnt;
2312 nd.root.dentry = dentry;
2313
2314 flags |= LOOKUP_ROOT;
2315
bcda7652 2316 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
73d049a4
AV		 2317 return ERR_PTR(-ELOOP);
2318
2319 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2320 if (unlikely(file == ERR_PTR(-ECHILD)))
2321 file = path_openat(-1, name, &nd, op, flags);
2322 if (unlikely(file == ERR_PTR(-ESTALE)))
2323 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2324 return file;
2325}
2326
1da177e4
LT		 2327/**
2328 * lookup_create - lookup a dentry, creating it if it doesn't exist
2329 * @nd: nameidata info
2330 * @is_dir: directory flag
2331 *
2332 * Simple function to lookup and return a dentry and create it
2333 * if it doesn't exist. Is SMP-safe.
c663e5d8 2334 *
4ac91378 2335 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1da177e4
LT		 2336 */
2337struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2338{
c663e5d8 2339 struct dentry *dentry = ERR_PTR(-EEXIST);
1da177e4 2340
4ac91378 2341 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
c663e5d8
CH		 2342 /*
2343 * Yucky last component or no last component at all?
2344 * (foo/., foo/.., /////)
2345 */
1da177e4
LT		 2346 if (nd->last_type != LAST_NORM)
2347 goto fail;
2348 nd->flags &= ~LOOKUP_PARENT;
3516586a 2349 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
a634904a 2350 nd->intent.open.flags = O_EXCL;
c663e5d8
CH		 2351
2352 /*
2353 * Do the final lookup.
2354 */
49705b77 2355 dentry = lookup_hash(nd);
1da177e4
LT		 2356 if (IS_ERR(dentry))
2357 goto fail;
c663e5d8 2358
e9baf6e5
AV		 2359 if (dentry->d_inode)
2360 goto eexist;
c663e5d8
CH		 2361 /*
2362 * Special case - lookup gave negative, but... we had foo/bar/
2363 * From the vfs_mknod() POV we just have a negative dentry -
2364 * all is fine. Let's be bastards - you had / on the end, you've
2365 * been asking for (non-existent) directory. -ENOENT for you.
2366 */
e9baf6e5
AV		 2367 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2368 dput(dentry);
2369 dentry = ERR_PTR(-ENOENT);
2370 }
1da177e4 2371 return dentry;
e9baf6e5 2372eexist:
1da177e4 2373 dput(dentry);
e9baf6e5 2374 dentry = ERR_PTR(-EEXIST);
1da177e4
LT		 2375fail:
2376 return dentry;
2377}
f81a0bff 2378EXPORT_SYMBOL_GPL(lookup_create);
1da177e4
LT		 2379
2380int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2381{
a95164d9 2382 int error = may_create(dir, dentry);
1da177e4
LT		 2383
2384 if (error)
2385 return error;
2386
e795b717
SH		 2387 if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
2388 !ns_capable(inode_userns(dir), CAP_MKNOD))
1da177e4
LT		 2389 return -EPERM;
2390
acfa4380 2391 if (!dir->i_op->mknod)
1da177e4
LT		 2392 return -EPERM;
2393
08ce5f16
SH		 2394 error = devcgroup_inode_mknod(mode, dev);
2395 if (error)
2396 return error;
2397
1da177e4
LT		 2398 error = security_inode_mknod(dir, dentry, mode, dev);
2399 if (error)
2400 return error;
2401
1da177e4 2402 error = dir->i_op->mknod(dir, dentry, mode, dev);
a74574aa 2403 if (!error)
f38aa942 2404 fsnotify_create(dir, dentry);
1da177e4
LT		 2405 return error;
2406}
2407
463c3197
DH		 2408static int may_mknod(mode_t mode)
2409{
2410 switch (mode & S_IFMT) {
2411 case S_IFREG:
2412 case S_IFCHR:
2413 case S_IFBLK:
2414 case S_IFIFO:
2415 case S_IFSOCK:
2416 case 0: /* zero mode translates to S_IFREG */
2417 return 0;
2418 case S_IFDIR:
2419 return -EPERM;
2420 default:
2421 return -EINVAL;
2422 }
2423}
2424
2e4d0924
HC		 2425SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2426 unsigned, dev)
1da177e4 2427{
2ad94ae6
AV		 2428 int error;
2429 char *tmp;
2430 struct dentry *dentry;
1da177e4
LT		 2431 struct nameidata nd;
2432
2433 if (S_ISDIR(mode))
2434 return -EPERM;
1da177e4 2435
2ad94ae6 2436 error = user_path_parent(dfd, filename, &nd, &tmp);
1da177e4 2437 if (error)
2ad94ae6
AV		 2438 return error;
2439
1da177e4 2440 dentry = lookup_create(&nd, 0);
463c3197
DH		 2441 if (IS_ERR(dentry)) {
2442 error = PTR_ERR(dentry);
2443 goto out_unlock;
2444 }
4ac91378 2445 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2446 mode &= ~current_umask();
463c3197
DH		 2447 error = may_mknod(mode);
2448 if (error)
2449 goto out_dput;
2450 error = mnt_want_write(nd.path.mnt);
2451 if (error)
2452 goto out_dput;
be6d3e56
KT		 2453 error = security_path_mknod(&nd.path, dentry, mode, dev);
2454 if (error)
2455 goto out_drop_write;
463c3197 2456 switch (mode & S_IFMT) {
1da177e4 2457 case 0: case S_IFREG:
4ac91378 2458 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
1da177e4
LT		 2459 break;
2460 case S_IFCHR: case S_IFBLK:
4ac91378 2461 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
1da177e4
LT		 2462 new_decode_dev(dev));
2463 break;
2464 case S_IFIFO: case S_IFSOCK:
4ac91378 2465 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
1da177e4 2466 break;
1da177e4 2467 }
be6d3e56 2468out_drop_write:
463c3197
DH		 2469 mnt_drop_write(nd.path.mnt);
2470out_dput:
2471 dput(dentry);
2472out_unlock:
4ac91378 2473 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2474 path_put(&nd.path);
1da177e4
LT		 2475 putname(tmp);
2476
2477 return error;
2478}
2479
3480b257 2480SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
5590ff0d
UD		 2481{
2482 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2483}
2484
1da177e4
LT		 2485int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2486{
a95164d9 2487 int error = may_create(dir, dentry);
1da177e4
LT		 2488
2489 if (error)
2490 return error;
2491
acfa4380 2492 if (!dir->i_op->mkdir)
1da177e4
LT		 2493 return -EPERM;
2494
2495 mode &= (S_IRWXUGO|S_ISVTX);
2496 error = security_inode_mkdir(dir, dentry, mode);
2497 if (error)
2498 return error;
2499
1da177e4 2500 error = dir->i_op->mkdir(dir, dentry, mode);
a74574aa 2501 if (!error)
f38aa942 2502 fsnotify_mkdir(dir, dentry);
1da177e4
LT		 2503 return error;
2504}
2505
2e4d0924 2506SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
1da177e4
LT		 2507{
2508 int error = 0;
2509 char * tmp;
6902d925
DH		 2510 struct dentry *dentry;
2511 struct nameidata nd;
1da177e4 2512
2ad94ae6
AV		 2513 error = user_path_parent(dfd, pathname, &nd, &tmp);
2514 if (error)
6902d925 2515 goto out_err;
1da177e4 2516
6902d925
DH		 2517 dentry = lookup_create(&nd, 1);
2518 error = PTR_ERR(dentry);
2519 if (IS_ERR(dentry))
2520 goto out_unlock;
1da177e4 2521
4ac91378 2522 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2523 mode &= ~current_umask();
463c3197
DH		 2524 error = mnt_want_write(nd.path.mnt);
2525 if (error)
2526 goto out_dput;
be6d3e56
KT		 2527 error = security_path_mkdir(&nd.path, dentry, mode);
2528 if (error)
2529 goto out_drop_write;
4ac91378 2530 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
be6d3e56 2531out_drop_write:
463c3197
DH		 2532 mnt_drop_write(nd.path.mnt);
2533out_dput:
6902d925
DH		 2534 dput(dentry);
2535out_unlock:
4ac91378 2536 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2537 path_put(&nd.path);
6902d925
DH		 2538 putname(tmp);
2539out_err:
1da177e4
LT		 2540 return error;
2541}
2542
3cdad428 2543SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
5590ff0d
UD		 2544{
2545 return sys_mkdirat(AT_FDCWD, pathname, mode);
2546}
2547
1da177e4 2548/*
a71905f0
SW		 2549 * The dentry_unhash() helper will try to drop the dentry early: we
2550 * should have a usage count of 2 if we're the only user of this
2551 * dentry, and if that is true (possibly after pruning the dcache),
2552 * then we drop the dentry now.
1da177e4
LT		 2553 *
2554 * A low-level filesystem can, if it choses, legally
2555 * do a
2556 *
2557 * if (!d_unhashed(dentry))
2558 * return -EBUSY;
2559 *
2560 * if it cannot handle the case of removing a directory
2561 * that is still in use by something else..
2562 */
2563void dentry_unhash(struct dentry *dentry)
2564{
dc168427 2565 shrink_dcache_parent(dentry);
1da177e4 2566 spin_lock(&dentry->d_lock);
64252c75 2567 if (dentry->d_count == 1)
1da177e4
LT		 2568 __d_drop(dentry);
2569 spin_unlock(&dentry->d_lock);
1da177e4
LT		 2570}
2571
2572int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2573{
2574 int error = may_delete(dir, dentry, 1);
2575
2576 if (error)
2577 return error;
2578
acfa4380 2579 if (!dir->i_op->rmdir)
1da177e4
LT		 2580 return -EPERM;
2581
1b1dcc1b 2582 mutex_lock(&dentry->d_inode->i_mutex);
912dbc15
SW		 2583
2584 error = -EBUSY;
1da177e4 2585 if (d_mountpoint(dentry))
912dbc15
SW		 2586 goto out;
2587
2588 error = security_inode_rmdir(dir, dentry);
2589 if (error)
2590 goto out;
2591
2592 error = dir->i_op->rmdir(dir, dentry);
2593 if (error)
2594 goto out;
2595
2596 dentry->d_inode->i_flags |= S_DEAD;
2597 dont_mount(dentry);
2598
2599out:
1b1dcc1b 2600 mutex_unlock(&dentry->d_inode->i_mutex);
912dbc15 2601 if (!error)
1da177e4 2602 d_delete(dentry);
1da177e4
LT		 2603 return error;
2604}
2605
5590ff0d 2606static long do_rmdir(int dfd, const char __user *pathname)
1da177e4
LT		 2607{
2608 int error = 0;
2609 char * name;
2610 struct dentry *dentry;
2611 struct nameidata nd;
2612
2ad94ae6 2613 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2614 if (error)
2ad94ae6 2615 return error;
1da177e4
LT		 2616
2617 switch(nd.last_type) {
0612d9fb
OH		 2618 case LAST_DOTDOT:
2619 error = -ENOTEMPTY;
2620 goto exit1;
2621 case LAST_DOT:
2622 error = -EINVAL;
2623 goto exit1;
2624 case LAST_ROOT:
2625 error = -EBUSY;
2626 goto exit1;
1da177e4 2627 }
0612d9fb
OH		 2628
2629 nd.flags &= ~LOOKUP_PARENT;
2630
4ac91378 2631 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2632 dentry = lookup_hash(&nd);
1da177e4 2633 error = PTR_ERR(dentry);
6902d925
DH		 2634 if (IS_ERR(dentry))
2635 goto exit2;
0622753b
DH		 2636 error = mnt_want_write(nd.path.mnt);
2637 if (error)
2638 goto exit3;
be6d3e56
KT		 2639 error = security_path_rmdir(&nd.path, dentry);
2640 if (error)
2641 goto exit4;
4ac91378 2642 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
be6d3e56 2643exit4:
0622753b
DH		 2644 mnt_drop_write(nd.path.mnt);
2645exit3:
6902d925
DH		 2646 dput(dentry);
2647exit2:
4ac91378 2648 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2649exit1:
1d957f9b 2650 path_put(&nd.path);
1da177e4
LT		 2651 putname(name);
2652 return error;
2653}
2654
3cdad428 2655SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
5590ff0d
UD		 2656{
2657 return do_rmdir(AT_FDCWD, pathname);
2658}
2659
1da177e4
LT		 2660int vfs_unlink(struct inode *dir, struct dentry *dentry)
2661{
2662 int error = may_delete(dir, dentry, 0);
2663
2664 if (error)
2665 return error;
2666
acfa4380 2667 if (!dir->i_op->unlink)
1da177e4
LT		 2668 return -EPERM;
2669
1b1dcc1b 2670 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2671 if (d_mountpoint(dentry))
2672 error = -EBUSY;
2673 else {
2674 error = security_inode_unlink(dir, dentry);
bec1052e 2675 if (!error) {
1da177e4 2676 error = dir->i_op->unlink(dir, dentry);
bec1052e 2677 if (!error)
d83c49f3 2678 dont_mount(dentry);
bec1052e 2679 }
1da177e4 2680 }
1b1dcc1b 2681 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2682
2683 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2684 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
ece95912 2685 fsnotify_link_count(dentry->d_inode);
e234f35c 2686 d_delete(dentry);
1da177e4 2687 }
0eeca283 2688
1da177e4
LT		 2689 return error;
2690}
2691
2692/*
2693 * Make sure that the actual truncation of the file will occur outside its
1b1dcc1b 2694 * directory's i_mutex. Truncate can take a long time if there is a lot of
1da177e4
LT		 2695 * writeout happening, and we don't want to prevent access to the directory
2696 * while waiting on the I/O.
2697 */
5590ff0d 2698static long do_unlinkat(int dfd, const char __user *pathname)
1da177e4 2699{
2ad94ae6
AV		 2700 int error;
2701 char *name;
1da177e4
LT		 2702 struct dentry *dentry;
2703 struct nameidata nd;
2704 struct inode *inode = NULL;
2705
2ad94ae6 2706 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2707 if (error)
2ad94ae6
AV		 2708 return error;
2709
1da177e4
LT		 2710 error = -EISDIR;
2711 if (nd.last_type != LAST_NORM)
2712 goto exit1;
0612d9fb
OH		 2713
2714 nd.flags &= ~LOOKUP_PARENT;
2715
4ac91378 2716 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2717 dentry = lookup_hash(&nd);
1da177e4
LT		 2718 error = PTR_ERR(dentry);
2719 if (!IS_ERR(dentry)) {
2720 /* Why not before? Because we want correct error value */
2721 if (nd.last.name[nd.last.len])
2722 goto slashes;
2723 inode = dentry->d_inode;
2724 if (inode)
7de9c6ee 2725 ihold(inode);
0622753b
DH		 2726 error = mnt_want_write(nd.path.mnt);
2727 if (error)
2728 goto exit2;
be6d3e56
KT		 2729 error = security_path_unlink(&nd.path, dentry);
2730 if (error)
2731 goto exit3;
4ac91378 2732 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
be6d3e56 2733exit3:
0622753b 2734 mnt_drop_write(nd.path.mnt);
1da177e4
LT		 2735 exit2:
2736 dput(dentry);
2737 }
4ac91378 2738 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4
LT		 2739 if (inode)
2740 iput(inode); /* truncate the inode here */
2741exit1:
1d957f9b 2742 path_put(&nd.path);
1da177e4
LT		 2743 putname(name);
2744 return error;
2745
2746slashes:
2747 error = !dentry->d_inode ? -ENOENT :
2748 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2749 goto exit2;
2750}
2751
2e4d0924 2752SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
5590ff0d
UD		 2753{
2754 if ((flag & ~AT_REMOVEDIR) != 0)
2755 return -EINVAL;
2756
2757 if (flag & AT_REMOVEDIR)
2758 return do_rmdir(dfd, pathname);
2759
2760 return do_unlinkat(dfd, pathname);
2761}
2762
3480b257 2763SYSCALL_DEFINE1(unlink, const char __user *, pathname)
5590ff0d
UD		 2764{
2765 return do_unlinkat(AT_FDCWD, pathname);
2766}
2767
db2e747b 2768int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
1da177e4 2769{
a95164d9 2770 int error = may_create(dir, dentry);
1da177e4
LT		 2771
2772 if (error)
2773 return error;
2774
acfa4380 2775 if (!dir->i_op->symlink)
1da177e4
LT		 2776 return -EPERM;
2777
2778 error = security_inode_symlink(dir, dentry, oldname);
2779 if (error)
2780 return error;
2781
1da177e4 2782 error = dir->i_op->symlink(dir, dentry, oldname);
a74574aa 2783 if (!error)
f38aa942 2784 fsnotify_create(dir, dentry);
1da177e4
LT		 2785 return error;
2786}
2787
2e4d0924
HC		 2788SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2789 int, newdfd, const char __user *, newname)
1da177e4 2790{
2ad94ae6
AV		 2791 int error;
2792 char *from;
2793 char *to;
6902d925
DH		 2794 struct dentry *dentry;
2795 struct nameidata nd;
1da177e4
LT		 2796
2797 from = getname(oldname);
2ad94ae6 2798 if (IS_ERR(from))
1da177e4 2799 return PTR_ERR(from);
1da177e4 2800
2ad94ae6 2801 error = user_path_parent(newdfd, newname, &nd, &to);
6902d925 2802 if (error)
2ad94ae6
AV		 2803 goto out_putname;
2804
6902d925
DH		 2805 dentry = lookup_create(&nd, 0);
2806 error = PTR_ERR(dentry);
2807 if (IS_ERR(dentry))
2808 goto out_unlock;
2809
75c3f29d
DH		 2810 error = mnt_want_write(nd.path.mnt);
2811 if (error)
2812 goto out_dput;
be6d3e56
KT		 2813 error = security_path_symlink(&nd.path, dentry, from);
2814 if (error)
2815 goto out_drop_write;
db2e747b 2816 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
be6d3e56 2817out_drop_write:
75c3f29d
DH		 2818 mnt_drop_write(nd.path.mnt);
2819out_dput:
6902d925
DH		 2820 dput(dentry);
2821out_unlock:
4ac91378 2822 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2823 path_put(&nd.path);
6902d925
DH		 2824 putname(to);
2825out_putname:
1da177e4
LT		 2826 putname(from);
2827 return error;
2828}
2829
3480b257 2830SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 2831{
2832 return sys_symlinkat(oldname, AT_FDCWD, newname);
2833}
2834
1da177e4
LT		 2835int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2836{
2837 struct inode *inode = old_dentry->d_inode;
2838 int error;
2839
2840 if (!inode)
2841 return -ENOENT;
2842
a95164d9 2843 error = may_create(dir, new_dentry);
1da177e4
LT		 2844 if (error)
2845 return error;
2846
2847 if (dir->i_sb != inode->i_sb)
2848 return -EXDEV;
2849
2850 /*
2851 * A link to an append-only or immutable file cannot be created.
2852 */
2853 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2854 return -EPERM;
acfa4380 2855 if (!dir->i_op->link)
1da177e4 2856 return -EPERM;
7e79eedb 2857 if (S_ISDIR(inode->i_mode))
1da177e4
LT		 2858 return -EPERM;
2859
2860 error = security_inode_link(old_dentry, dir, new_dentry);
2861 if (error)
2862 return error;
2863
7e79eedb 2864 mutex_lock(&inode->i_mutex);
aae8a97d
AK		 2865 /* Make sure we don't allow creating hardlink to an unlinked file */
2866 if (inode->i_nlink == 0)
2867 error = -ENOENT;
2868 else
2869 error = dir->i_op->link(old_dentry, dir, new_dentry);
7e79eedb 2870 mutex_unlock(&inode->i_mutex);
e31e14ec 2871 if (!error)
7e79eedb 2872 fsnotify_link(dir, inode, new_dentry);
1da177e4
LT		 2873 return error;
2874}
2875
2876/*
2877 * Hardlinks are often used in delicate situations. We avoid
2878 * security-related surprises by not following symlinks on the
2879 * newname. --KAB
2880 *
2881 * We don't follow them on the oldname either to be compatible
2882 * with linux 2.0, and to avoid hard-linking to directories
2883 * and other special files. --ADM
2884 */
2e4d0924
HC		 2885SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2886 int, newdfd, const char __user *, newname, int, flags)
1da177e4
LT		 2887{
2888 struct dentry *new_dentry;
2d8f3038
AV		 2889 struct nameidata nd;
2890 struct path old_path;
11a7b371 2891 int how = 0;
1da177e4 2892 int error;
2ad94ae6 2893 char *to;
1da177e4 2894
11a7b371 2895 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
c04030e1 2896 return -EINVAL;
11a7b371
AK		 2897 /*
2898 * To use null names we require CAP_DAC_READ_SEARCH
2899 * This ensures that not everyone will be able to create
2900 * handlink using the passed filedescriptor.
2901 */
2902 if (flags & AT_EMPTY_PATH) {
2903 if (!capable(CAP_DAC_READ_SEARCH))
2904 return -ENOENT;
2905 how = LOOKUP_EMPTY;
2906 }
2907
2908 if (flags & AT_SYMLINK_FOLLOW)
2909 how |= LOOKUP_FOLLOW;
c04030e1 2910
11a7b371 2911 error = user_path_at(olddfd, oldname, how, &old_path);
1da177e4 2912 if (error)
2ad94ae6
AV		 2913 return error;
2914
2915 error = user_path_parent(newdfd, newname, &nd, &to);
1da177e4
LT		 2916 if (error)
2917 goto out;
2918 error = -EXDEV;
2d8f3038 2919 if (old_path.mnt != nd.path.mnt)
1da177e4
LT		 2920 goto out_release;
2921 new_dentry = lookup_create(&nd, 0);
2922 error = PTR_ERR(new_dentry);
6902d925
DH		 2923 if (IS_ERR(new_dentry))
2924 goto out_unlock;
75c3f29d
DH		 2925 error = mnt_want_write(nd.path.mnt);
2926 if (error)
2927 goto out_dput;
be6d3e56
KT		 2928 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2929 if (error)
2930 goto out_drop_write;
2d8f3038 2931 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
be6d3e56 2932out_drop_write:
75c3f29d
DH		 2933 mnt_drop_write(nd.path.mnt);
2934out_dput:
6902d925
DH		 2935 dput(new_dentry);
2936out_unlock:
4ac91378 2937 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2938out_release:
1d957f9b 2939 path_put(&nd.path);
2ad94ae6 2940 putname(to);
1da177e4 2941out:
2d8f3038 2942 path_put(&old_path);
1da177e4
LT		 2943
2944 return error;
2945}
2946
3480b257 2947SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
5590ff0d 2948{
c04030e1 2949 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
5590ff0d
UD		 2950}
2951
1da177e4
LT		 2952/*
2953 * The worst of all namespace operations - renaming directory. "Perverted"
2954 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2955 * Problems:
2956 * a) we can get into loop creation. Check is done in is_subdir().
2957 * b) race potential - two innocent renames can create a loop together.
2958 * That's where 4.4 screws up. Current fix: serialization on
a11f3a05 2959 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
1da177e4
LT		 2960 * story.
2961 * c) we have to lock _three_ objects - parents and victim (if it exists).
1b1dcc1b 2962 * And that - after we got ->i_mutex on parents (until then we don't know
1da177e4
LT		 2963 * whether the target exists). Solution: try to be smart with locking
2964 * order for inodes. We rely on the fact that tree topology may change
a11f3a05 2965 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
1da177e4
LT		 2966 * move will be locked. Thus we can rank directories by the tree
2967 * (ancestors first) and rank all non-directories after them.
2968 * That works since everybody except rename does "lock parent, lookup,
a11f3a05 2969 * lock child" and rename is under ->s_vfs_rename_mutex.
1da177e4
LT		 2970 * HOWEVER, it relies on the assumption that any object with ->lookup()
2971 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2972 * we'd better make sure that there's no link(2) for them.
e4eaac06 2973 * d) conversion from fhandle to dentry may come in the wrong moment - when
1b1dcc1b 2974 * we are removing the target. Solution: we will have to grab ->i_mutex
1da177e4 2975 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
c41b20e7 2976 * ->i_mutex on parents, which works but leads to some truly excessive
1da177e4
LT		 2977 * locking].
2978 */
75c96f85
AB		 2979static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2980 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 2981{
2982 int error = 0;
9055cba7 2983 struct inode *target = new_dentry->d_inode;
1da177e4
LT		 2984
2985 /*
2986 * If we are going to change the parent - check write permissions,
2987 * we'll need to flip '..'.
2988 */
2989 if (new_dir != old_dir) {
f419a2e3 2990 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
1da177e4
LT		 2991 if (error)
2992 return error;
2993 }
2994
2995 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2996 if (error)
2997 return error;
2998
d83c49f3 2999 if (target)
1b1dcc1b 3000 mutex_lock(&target->i_mutex);
9055cba7
SW		 3001
3002 error = -EBUSY;
3003 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3004 goto out;
3005
3006 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3007 if (error)
3008 goto out;
3009
1da177e4 3010 if (target) {
9055cba7
SW		 3011 target->i_flags |= S_DEAD;
3012 dont_mount(new_dentry);
1da177e4 3013 }
9055cba7
SW		 3014out:
3015 if (target)
3016 mutex_unlock(&target->i_mutex);
e31e14ec 3017 if (!error)
349457cc
MF		 3018 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3019 d_move(old_dentry,new_dentry);
1da177e4
LT		 3020 return error;
3021}
3022
75c96f85
AB		 3023static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3024 struct inode *new_dir, struct dentry *new_dentry)
1da177e4 3025{
51892bbb 3026 struct inode *target = new_dentry->d_inode;
1da177e4
LT		 3027 int error;
3028
3029 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3030 if (error)
3031 return error;
3032
3033 dget(new_dentry);
1da177e4 3034 if (target)
1b1dcc1b 3035 mutex_lock(&target->i_mutex);
51892bbb
SW		 3036
3037 error = -EBUSY;
1da177e4 3038 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
51892bbb
SW		 3039 goto out;
3040
3041 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3042 if (error)
3043 goto out;
3044
3045 if (target)
3046 dont_mount(new_dentry);
3047 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3048 d_move(old_dentry, new_dentry);
3049out:
1da177e4 3050 if (target)
1b1dcc1b 3051 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3052 dput(new_dentry);
3053 return error;
3054}
3055
3056int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3057 struct inode *new_dir, struct dentry *new_dentry)
3058{
3059 int error;
3060 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
59b0df21 3061 const unsigned char *old_name;
1da177e4
LT		 3062
3063 if (old_dentry->d_inode == new_dentry->d_inode)
3064 return 0;
3065
3066 error = may_delete(old_dir, old_dentry, is_dir);
3067 if (error)
3068 return error;
3069
3070 if (!new_dentry->d_inode)
a95164d9 3071 error = may_create(new_dir, new_dentry);
1da177e4
LT		 3072 else
3073 error = may_delete(new_dir, new_dentry, is_dir);
3074 if (error)
3075 return error;
3076
acfa4380 3077 if (!old_dir->i_op->rename)
1da177e4
LT		 3078 return -EPERM;
3079
0eeca283
RL		 3080 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3081
1da177e4
LT		 3082 if (is_dir)
3083 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3084 else
3085 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
123df294
AV		 3086 if (!error)
3087 fsnotify_move(old_dir, new_dir, old_name, is_dir,
5a190ae6 3088 new_dentry->d_inode, old_dentry);
0eeca283
RL		 3089 fsnotify_oldname_free(old_name);
3090
1da177e4
LT		 3091 return error;
3092}
3093
2e4d0924
HC		 3094SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3095 int, newdfd, const char __user *, newname)
1da177e4 3096{
2ad94ae6
AV		 3097 struct dentry *old_dir, *new_dir;
3098 struct dentry *old_dentry, *new_dentry;
3099 struct dentry *trap;
1da177e4 3100 struct nameidata oldnd, newnd;
2ad94ae6
AV		 3101 char *from;
3102 char *to;
3103 int error;
1da177e4 3104
2ad94ae6 3105 error = user_path_parent(olddfd, oldname, &oldnd, &from);
1da177e4
LT		 3106 if (error)
3107 goto exit;
3108
2ad94ae6 3109 error = user_path_parent(newdfd, newname, &newnd, &to);
1da177e4
LT		 3110 if (error)
3111 goto exit1;
3112
3113 error = -EXDEV;
4ac91378 3114 if (oldnd.path.mnt != newnd.path.mnt)
1da177e4
LT		 3115 goto exit2;
3116
4ac91378 3117 old_dir = oldnd.path.dentry;
1da177e4
LT		 3118 error = -EBUSY;
3119 if (oldnd.last_type != LAST_NORM)
3120 goto exit2;
3121
4ac91378 3122 new_dir = newnd.path.dentry;
1da177e4
LT		 3123 if (newnd.last_type != LAST_NORM)
3124 goto exit2;
3125
0612d9fb
OH		 3126 oldnd.flags &= ~LOOKUP_PARENT;
3127 newnd.flags &= ~LOOKUP_PARENT;
4e9ed2f8 3128 newnd.flags |= LOOKUP_RENAME_TARGET;
0612d9fb 3129
1da177e4
LT		 3130 trap = lock_rename(new_dir, old_dir);
3131
49705b77 3132 old_dentry = lookup_hash(&oldnd);
1da177e4
LT		 3133 error = PTR_ERR(old_dentry);
3134 if (IS_ERR(old_dentry))
3135 goto exit3;
3136 /* source must exist */
3137 error = -ENOENT;
3138 if (!old_dentry->d_inode)
3139 goto exit4;
3140 /* unless the source is a directory trailing slashes give -ENOTDIR */
3141 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3142 error = -ENOTDIR;
3143 if (oldnd.last.name[oldnd.last.len])
3144 goto exit4;
3145 if (newnd.last.name[newnd.last.len])
3146 goto exit4;
3147 }
3148 /* source should not be ancestor of target */
3149 error = -EINVAL;
3150 if (old_dentry == trap)
3151 goto exit4;
49705b77 3152 new_dentry = lookup_hash(&newnd);
1da177e4
LT		 3153 error = PTR_ERR(new_dentry);
3154 if (IS_ERR(new_dentry))
3155 goto exit4;
3156 /* target should not be an ancestor of source */
3157 error = -ENOTEMPTY;
3158 if (new_dentry == trap)
3159 goto exit5;
3160
9079b1eb
DH		 3161 error = mnt_want_write(oldnd.path.mnt);
3162 if (error)
3163 goto exit5;
be6d3e56
KT		 3164 error = security_path_rename(&oldnd.path, old_dentry,
3165 &newnd.path, new_dentry);
3166 if (error)
3167 goto exit6;
1da177e4
LT		 3168 error = vfs_rename(old_dir->d_inode, old_dentry,
3169 new_dir->d_inode, new_dentry);
be6d3e56 3170exit6:
9079b1eb 3171 mnt_drop_write(oldnd.path.mnt);
1da177e4
LT		 3172exit5:
3173 dput(new_dentry);
3174exit4:
3175 dput(old_dentry);
3176exit3:
3177 unlock_rename(new_dir, old_dir);
3178exit2:
1d957f9b 3179 path_put(&newnd.path);
2ad94ae6 3180 putname(to);
1da177e4 3181exit1:
1d957f9b 3182 path_put(&oldnd.path);
1da177e4 3183 putname(from);
2ad94ae6 3184exit:
1da177e4
LT		 3185 return error;
3186}
3187
a26eab24 3188SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 3189{
3190 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3191}
3192
1da177e4
LT		 3193int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3194{
3195 int len;
3196
3197 len = PTR_ERR(link);
3198 if (IS_ERR(link))
3199 goto out;
3200
3201 len = strlen(link);
3202 if (len > (unsigned) buflen)
3203 len = buflen;
3204 if (copy_to_user(buffer, link, len))
3205 len = -EFAULT;
3206out:
3207 return len;
3208}
3209
3210/*
3211 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3212 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3213 * using) it for any given inode is up to filesystem.
3214 */
3215int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3216{
3217 struct nameidata nd;
cc314eef 3218 void *cookie;
694a1764 3219 int res;
cc314eef 3220
1da177e4 3221 nd.depth = 0;
cc314eef 3222 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
694a1764
MS		 3223 if (IS_ERR(cookie))
3224 return PTR_ERR(cookie);
3225
3226 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3227 if (dentry->d_inode->i_op->put_link)
3228 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3229 return res;
1da177e4
LT		 3230}
3231
3232int vfs_follow_link(struct nameidata *nd, const char *link)
3233{
3234 return __vfs_follow_link(nd, link);
3235}
3236
3237/* get the link contents into pagecache */
3238static char *page_getlink(struct dentry * dentry, struct page **ppage)
3239{
ebd09abb
DG		 3240 char *kaddr;
3241 struct page *page;
1da177e4 3242 struct address_space *mapping = dentry->d_inode->i_mapping;
090d2b18 3243 page = read_mapping_page(mapping, 0, NULL);
1da177e4 3244 if (IS_ERR(page))
6fe6900e 3245 return (char*)page;
1da177e4 3246 *ppage = page;
ebd09abb
DG		 3247 kaddr = kmap(page);
3248 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3249 return kaddr;
1da177e4
LT		 3250}
3251
3252int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3253{
3254 struct page *page = NULL;
3255 char *s = page_getlink(dentry, &page);
3256 int res = vfs_readlink(dentry,buffer,buflen,s);
3257 if (page) {
3258 kunmap(page);
3259 page_cache_release(page);
3260 }
3261 return res;
3262}
3263
cc314eef 3264void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
1da177e4 3265{
cc314eef 3266 struct page *page = NULL;
1da177e4 3267 nd_set_link(nd, page_getlink(dentry, &page));
cc314eef 3268 return page;
1da177e4
LT		 3269}
3270
cc314eef 3271void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1da177e4 3272{
cc314eef
LT		 3273 struct page *page = cookie;
3274
3275 if (page) {
1da177e4
LT		 3276 kunmap(page);
3277 page_cache_release(page);
1da177e4
LT		 3278 }
3279}
3280
54566b2c
NP		 3281/*
3282 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3283 */
3284int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
1da177e4
LT		 3285{
3286 struct address_space *mapping = inode->i_mapping;
0adb25d2 3287 struct page *page;
afddba49 3288 void *fsdata;
beb497ab 3289 int err;
1da177e4 3290 char *kaddr;
54566b2c
NP		 3291 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3292 if (nofs)
3293 flags |= AOP_FLAG_NOFS;
1da177e4 3294
7e53cac4 3295retry:
afddba49 3296 err = pagecache_write_begin(NULL, mapping, 0, len-1,
54566b2c 3297 flags, &page, &fsdata);
1da177e4 3298 if (err)
afddba49
NP		 3299 goto fail;
3300
1da177e4
LT		 3301 kaddr = kmap_atomic(page, KM_USER0);
3302 memcpy(kaddr, symname, len-1);
3303 kunmap_atomic(kaddr, KM_USER0);
afddba49
NP		 3304
3305 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3306 page, fsdata);
1da177e4
LT		 3307 if (err < 0)
3308 goto fail;
afddba49
NP		 3309 if (err < len-1)
3310 goto retry;
3311
1da177e4
LT		 3312 mark_inode_dirty(inode);
3313 return 0;
1da177e4
LT		 3314fail:
3315 return err;
3316}
3317
0adb25d2
KK		 3318int page_symlink(struct inode *inode, const char *symname, int len)
3319{
3320 return __page_symlink(inode, symname, len,
54566b2c 3321 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
0adb25d2
KK		 3322}
3323
92e1d5be 3324const struct inode_operations page_symlink_inode_operations = {
1da177e4
LT		 3325 .readlink = generic_readlink,
3326 .follow_link = page_follow_link_light,
3327 .put_link = page_put_link,
3328};
3329
2d8f3038 3330EXPORT_SYMBOL(user_path_at);
cc53ce53 3331EXPORT_SYMBOL(follow_down_one);
1da177e4
LT		 3332EXPORT_SYMBOL(follow_down);
3333EXPORT_SYMBOL(follow_up);
3334EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3335EXPORT_SYMBOL(getname);
3336EXPORT_SYMBOL(lock_rename);
1da177e4
LT		 3337EXPORT_SYMBOL(lookup_one_len);
3338EXPORT_SYMBOL(page_follow_link_light);
3339EXPORT_SYMBOL(page_put_link);
3340EXPORT_SYMBOL(page_readlink);
0adb25d2 3341EXPORT_SYMBOL(__page_symlink);
1da177e4
LT		 3342EXPORT_SYMBOL(page_symlink);
3343EXPORT_SYMBOL(page_symlink_inode_operations);
c9c6cac0 3344EXPORT_SYMBOL(kern_path_parent);
d1811465 3345EXPORT_SYMBOL(kern_path);
16f18200 3346EXPORT_SYMBOL(vfs_path_lookup);
f419a2e3 3347EXPORT_SYMBOL(inode_permission);
8c744fb8 3348EXPORT_SYMBOL(file_permission);
1da177e4
LT		 3349EXPORT_SYMBOL(unlock_rename);
3350EXPORT_SYMBOL(vfs_create);
3351EXPORT_SYMBOL(vfs_follow_link);
3352EXPORT_SYMBOL(vfs_link);
3353EXPORT_SYMBOL(vfs_mkdir);
3354EXPORT_SYMBOL(vfs_mknod);
3355EXPORT_SYMBOL(generic_permission);
3356EXPORT_SYMBOL(vfs_readlink);
3357EXPORT_SYMBOL(vfs_rename);
3358EXPORT_SYMBOL(vfs_rmdir);
3359EXPORT_SYMBOL(vfs_symlink);
3360EXPORT_SYMBOL(vfs_unlink);
3361EXPORT_SYMBOL(dentry_unhash);
3362EXPORT_SYMBOL(generic_readlink);
kernel source for telekom puls (alcatel/mediatek)