projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
switch vfs_mkdir() and ->mkdir() to umode_t
[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>
e77819e5 35#include <linux/posix_acl.h>
1da177e4
LT		 36#include <asm/uaccess.h>
37
e81e3f4d
EP		 38#include "internal.h"
39
1da177e4
LT		 40/* [Feb-1997 T. Schoebel-Theuer]
41 * Fundamental changes in the pathname lookup mechanisms (namei)
42 * were necessary because of omirr. The reason is that omirr needs
43 * to know the _real_ pathname, not the user-supplied one, in case
44 * of symlinks (and also when transname replacements occur).
45 *
46 * The new code replaces the old recursive symlink resolution with
47 * an iterative one (in case of non-nested symlink chains). It does
48 * this with calls to <fs>_follow_link().
49 * As a side effect, dir_namei(), _namei() and follow_link() are now
50 * replaced with a single function lookup_dentry() that can handle all
51 * the special cases of the former code.
52 *
53 * With the new dcache, the pathname is stored at each inode, at least as
54 * long as the refcount of the inode is positive. As a side effect, the
55 * size of the dcache depends on the inode cache and thus is dynamic.
56 *
57 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
58 * resolution to correspond with current state of the code.
59 *
60 * Note that the symlink resolution is not *completely* iterative.
61 * There is still a significant amount of tail- and mid- recursion in
62 * the algorithm. Also, note that <fs>_readlink() is not used in
63 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
64 * may return different results than <fs>_follow_link(). Many virtual
65 * filesystems (including /proc) exhibit this behavior.
66 */
67
68/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
69 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
70 * and the name already exists in form of a symlink, try to create the new
71 * name indicated by the symlink. The old code always complained that the
72 * name already exists, due to not following the symlink even if its target
73 * is nonexistent. The new semantics affects also mknod() and link() when
25985edc 74 * the name is a symlink pointing to a non-existent name.
1da177e4
LT		 75 *
76 * I don't know which semantics is the right one, since I have no access
77 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
78 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
79 * "old" one. Personally, I think the new semantics is much more logical.
80 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
81 * file does succeed in both HP-UX and SunOs, but not in Solaris
82 * and in the old Linux semantics.
83 */
84
85/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
86 * semantics. See the comments in "open_namei" and "do_link" below.
87 *
88 * [10-Sep-98 Alan Modra] Another symlink change.
89 */
90
91/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
92 * inside the path - always follow.
93 * in the last component in creation/removal/renaming - never follow.
94 * if LOOKUP_FOLLOW passed - follow.
95 * if the pathname has trailing slashes - follow.
96 * otherwise - don't follow.
97 * (applied in that order).
98 *
99 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
100 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
101 * During the 2.4 we need to fix the userland stuff depending on it -
102 * hopefully we will be able to get rid of that wart in 2.5. So far only
103 * XEmacs seems to be relying on it...
104 */
105/*
106 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
a11f3a05 107 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
1da177e4
LT		 108 * any extra contention...
109 */
110
111/* In order to reduce some races, while at the same time doing additional
112 * checking and hopefully speeding things up, we copy filenames to the
113 * kernel data space before using them..
114 *
115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
116 * PATH_MAX includes the nul terminator --RR.
117 */
858119e1 118static int do_getname(const char __user *filename, char *page)
1da177e4
LT		 119{
120 int retval;
121 unsigned long len = PATH_MAX;
122
123 if (!segment_eq(get_fs(), KERNEL_DS)) {
124 if ((unsigned long) filename >= TASK_SIZE)
125 return -EFAULT;
126 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
127 len = TASK_SIZE - (unsigned long) filename;
128 }
129
130 retval = strncpy_from_user(page, filename, len);
131 if (retval > 0) {
132 if (retval < len)
133 return 0;
134 return -ENAMETOOLONG;
135 } else if (!retval)
136 retval = -ENOENT;
137 return retval;
138}
139
1fa1e7f6 140static char *getname_flags(const char __user *filename, int flags, int *empty)
1da177e4
LT		 141{
142 char *tmp, *result;
143
144 result = ERR_PTR(-ENOMEM);
145 tmp = __getname();
146 if (tmp) {
147 int retval = do_getname(filename, tmp);
148
149 result = tmp;
150 if (retval < 0) {
1fa1e7f6
AW		 151 if (retval == -ENOENT && empty)
152 *empty = 1;
f52e0c11
AV		 153 if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) {
154 __putname(tmp);
155 result = ERR_PTR(retval);
156 }
1da177e4
LT		 157 }
158 }
159 audit_getname(result);
160 return result;
161}
162
f52e0c11
AV		 163char *getname(const char __user * filename)
164{
1fa1e7f6 165 return getname_flags(filename, 0, 0);
f52e0c11
AV		 166}
167
1da177e4
LT		 168#ifdef CONFIG_AUDITSYSCALL
169void putname(const char *name)
170{
5ac3a9c2 171 if (unlikely(!audit_dummy_context()))
1da177e4
LT		 172 audit_putname(name);
173 else
174 __putname(name);
175}
176EXPORT_SYMBOL(putname);
177#endif
178
e77819e5
LT		 179static int check_acl(struct inode *inode, int mask)
180{
84635d68 181#ifdef CONFIG_FS_POSIX_ACL
e77819e5
LT		 182 struct posix_acl *acl;
183
e77819e5 184 if (mask & MAY_NOT_BLOCK) {
3567866b
AV		 185 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
186 if (!acl)
e77819e5 187 return -EAGAIN;
3567866b
AV		 188 /* no ->get_acl() calls in RCU mode... */
189 if (acl == ACL_NOT_CACHED)
190 return -ECHILD;
206b1d09 191 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
e77819e5
LT		 192 }
193
194 acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
195
196 /*
4e34e719
CH		 197 * A filesystem can force a ACL callback by just never filling the
198 * ACL cache. But normally you'd fill the cache either at inode
199 * instantiation time, or on the first ->get_acl call.
e77819e5 200 *
4e34e719
CH		 201 * If the filesystem doesn't have a get_acl() function at all, we'll
202 * just create the negative cache entry.
e77819e5
LT		 203 */
204 if (acl == ACL_NOT_CACHED) {
4e34e719
CH		 205 if (inode->i_op->get_acl) {
206 acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS);
207 if (IS_ERR(acl))
208 return PTR_ERR(acl);
209 } else {
210 set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
211 return -EAGAIN;
212 }
e77819e5
LT		 213 }
214
215 if (acl) {
216 int error = posix_acl_permission(inode, acl, mask);
217 posix_acl_release(acl);
218 return error;
219 }
84635d68 220#endif
e77819e5
LT		 221
222 return -EAGAIN;
223}
224
5909ccaa 225/*
948409c7 226 * This does the basic permission checking
1da177e4 227 */
7e40145e 228static int acl_permission_check(struct inode *inode, int mask)
1da177e4 229{
26cf46be 230 unsigned int mode = inode->i_mode;
1da177e4 231
e795b717
SH		 232 if (current_user_ns() != inode_userns(inode))
233 goto other_perms;
234
14067ff5 235 if (likely(current_fsuid() == inode->i_uid))
1da177e4
LT		 236 mode >>= 6;
237 else {
e77819e5 238 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
7e40145e 239 int error = check_acl(inode, mask);
b74c79e9
NP		 240 if (error != -EAGAIN)
241 return error;
1da177e4
LT		 242 }
243
244 if (in_group_p(inode->i_gid))
245 mode >>= 3;
246 }
247
e795b717 248other_perms:
1da177e4
LT		 249 /*
250 * If the DACs are ok we don't need any capability check.
251 */
9c2c7039 252 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
1da177e4 253 return 0;
5909ccaa
LT		 254 return -EACCES;
255}
256
257/**
b74c79e9 258 * generic_permission - check for access rights on a Posix-like filesystem
5909ccaa 259 * @inode: inode to check access rights for
8fd90c8d 260 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
5909ccaa
LT		 261 *
262 * Used to check for read/write/execute permissions on a file.
263 * We use "fsuid" for this, letting us set arbitrary permissions
264 * for filesystem access without changing the "normal" uids which
b74c79e9
NP		 265 * are used for other things.
266 *
267 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
268 * request cannot be satisfied (eg. requires blocking or too much complexity).
269 * It would then be called again in ref-walk mode.
5909ccaa 270 */
2830ba7f 271int generic_permission(struct inode *inode, int mask)
5909ccaa
LT		 272{
273 int ret;
274
275 /*
948409c7 276 * Do the basic permission checks.
5909ccaa 277 */
7e40145e 278 ret = acl_permission_check(inode, mask);
5909ccaa
LT		 279 if (ret != -EACCES)
280 return ret;
1da177e4 281
d594e7ec
AV		 282 if (S_ISDIR(inode->i_mode)) {
283 /* DACs are overridable for directories */
284 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
285 return 0;
286 if (!(mask & MAY_WRITE))
287 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
288 return 0;
289 return -EACCES;
290 }
1da177e4
LT		 291 /*
292 * Read/write DACs are always overridable.
d594e7ec
AV		 293 * Executable DACs are overridable when there is
294 * at least one exec bit set.
1da177e4 295 */
d594e7ec 296 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
e795b717 297 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
1da177e4
LT		 298 return 0;
299
300 /*
301 * Searching includes executable on directories, else just read.
302 */
7ea66001 303 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
d594e7ec 304 if (mask == MAY_READ)
e795b717 305 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
1da177e4
LT		 306 return 0;
307
308 return -EACCES;
309}
310
3ddcd056
LT		 311/*
312 * We _really_ want to just do "generic_permission()" without
313 * even looking at the inode->i_op values. So we keep a cache
314 * flag in inode->i_opflags, that says "this has not special
315 * permission function, use the fast case".
316 */
317static inline int do_inode_permission(struct inode *inode, int mask)
318{
319 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
320 if (likely(inode->i_op->permission))
321 return inode->i_op->permission(inode, mask);
322
323 /* This gets set once for the inode lifetime */
324 spin_lock(&inode->i_lock);
325 inode->i_opflags |= IOP_FASTPERM;
326 spin_unlock(&inode->i_lock);
327 }
328 return generic_permission(inode, mask);
329}
330
cb23beb5
CH		 331/**
332 * inode_permission - check for access rights to a given inode
333 * @inode: inode to check permission on
8fd90c8d 334 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
cb23beb5
CH		 335 *
336 * Used to check for read/write/execute permissions on an inode.
337 * We use "fsuid" for this, letting us set arbitrary permissions
338 * for filesystem access without changing the "normal" uids which
339 * are used for other things.
948409c7
AG		 340 *
341 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
cb23beb5 342 */
f419a2e3 343int inode_permission(struct inode *inode, int mask)
1da177e4 344{
e6305c43 345 int retval;
1da177e4 346
3ddcd056 347 if (unlikely(mask & MAY_WRITE)) {
22590e41 348 umode_t mode = inode->i_mode;
1da177e4
LT		 349
350 /*
351 * Nobody gets write access to a read-only fs.
352 */
353 if (IS_RDONLY(inode) &&
354 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
355 return -EROFS;
356
357 /*
358 * Nobody gets write access to an immutable file.
359 */
360 if (IS_IMMUTABLE(inode))
361 return -EACCES;
362 }
363
3ddcd056 364 retval = do_inode_permission(inode, mask);
1da177e4
LT		 365 if (retval)
366 return retval;
367
08ce5f16
SH		 368 retval = devcgroup_inode_permission(inode, mask);
369 if (retval)
370 return retval;
371
d09ca739 372 return security_inode_permission(inode, mask);
1da177e4
LT		 373}
374
5dd784d0
JB		 375/**
376 * path_get - get a reference to a path
377 * @path: path to get the reference to
378 *
379 * Given a path increment the reference count to the dentry and the vfsmount.
380 */
381void path_get(struct path *path)
382{
383 mntget(path->mnt);
384 dget(path->dentry);
385}
386EXPORT_SYMBOL(path_get);
387
1d957f9b
JB		 388/**
389 * path_put - put a reference to a path
390 * @path: path to put the reference to
391 *
392 * Given a path decrement the reference count to the dentry and the vfsmount.
393 */
394void path_put(struct path *path)
1da177e4 395{
1d957f9b
JB		 396 dput(path->dentry);
397 mntput(path->mnt);
1da177e4 398}
1d957f9b 399EXPORT_SYMBOL(path_put);
1da177e4 400
19660af7 401/*
31e6b01f 402 * Path walking has 2 modes, rcu-walk and ref-walk (see
19660af7
AV		 403 * Documentation/filesystems/path-lookup.txt). In situations when we can't
404 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
405 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
406 * mode. Refcounts are grabbed at the last known good point before rcu-walk
407 * got stuck, so ref-walk may continue from there. If this is not successful
408 * (eg. a seqcount has changed), then failure is returned and it's up to caller
409 * to restart the path walk from the beginning in ref-walk mode.
31e6b01f 410 */
31e6b01f
NP		 411
412/**
19660af7
AV		 413 * unlazy_walk - try to switch to ref-walk mode.
414 * @nd: nameidata pathwalk data
415 * @dentry: child of nd->path.dentry or NULL
39191628 416 * Returns: 0 on success, -ECHILD on failure
31e6b01f 417 *
19660af7
AV		 418 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
419 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
420 * @nd or NULL. Must be called from rcu-walk context.
31e6b01f 421 */
19660af7 422static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
31e6b01f
NP		 423{
424 struct fs_struct *fs = current->fs;
425 struct dentry *parent = nd->path.dentry;
5b6ca027 426 int want_root = 0;
31e6b01f
NP		 427
428 BUG_ON(!(nd->flags & LOOKUP_RCU));
5b6ca027
AV		 429 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
430 want_root = 1;
31e6b01f
NP		 431 spin_lock(&fs->lock);
432 if (nd->root.mnt != fs->root.mnt ||
433 nd->root.dentry != fs->root.dentry)
434 goto err_root;
435 }
436 spin_lock(&parent->d_lock);
19660af7
AV		 437 if (!dentry) {
438 if (!__d_rcu_to_refcount(parent, nd->seq))
439 goto err_parent;
440 BUG_ON(nd->inode != parent->d_inode);
441 } else {
94c0d4ec
AV		 442 if (dentry->d_parent != parent)
443 goto err_parent;
19660af7
AV		 444 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
445 if (!__d_rcu_to_refcount(dentry, nd->seq))
446 goto err_child;
447 /*
448 * If the sequence check on the child dentry passed, then
449 * the child has not been removed from its parent. This
450 * means the parent dentry must be valid and able to take
451 * a reference at this point.
452 */
453 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
454 BUG_ON(!parent->d_count);
455 parent->d_count++;
456 spin_unlock(&dentry->d_lock);
457 }
31e6b01f 458 spin_unlock(&parent->d_lock);
5b6ca027 459 if (want_root) {
31e6b01f
NP		 460 path_get(&nd->root);
461 spin_unlock(&fs->lock);
462 }
463 mntget(nd->path.mnt);
464
465 rcu_read_unlock();
466 br_read_unlock(vfsmount_lock);
467 nd->flags &= ~LOOKUP_RCU;
468 return 0;
19660af7
AV		 469
470err_child:
31e6b01f 471 spin_unlock(&dentry->d_lock);
19660af7 472err_parent:
31e6b01f
NP		 473 spin_unlock(&parent->d_lock);
474err_root:
5b6ca027 475 if (want_root)
31e6b01f
NP		 476 spin_unlock(&fs->lock);
477 return -ECHILD;
478}
479
834f2a4a
TM		 480/**
481 * release_open_intent - free up open intent resources
482 * @nd: pointer to nameidata
483 */
484void release_open_intent(struct nameidata *nd)
485{
2dab5974
LT		 486 struct file *file = nd->intent.open.file;
487
488 if (file && !IS_ERR(file)) {
489 if (file->f_path.dentry == NULL)
490 put_filp(file);
491 else
492 fput(file);
493 }
834f2a4a
TM		 494}
495
f60aef7e 496static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
34286d66 497{
f60aef7e 498 return dentry->d_op->d_revalidate(dentry, nd);
34286d66
NP		 499}
500
9f1fafee
AV		 501/**
502 * complete_walk - successful completion of path walk
503 * @nd: pointer nameidata
39159de2 504 *
9f1fafee
AV		 505 * If we had been in RCU mode, drop out of it and legitimize nd->path.
506 * Revalidate the final result, unless we'd already done that during
507 * the path walk or the filesystem doesn't ask for it. Return 0 on
508 * success, -error on failure. In case of failure caller does not
509 * need to drop nd->path.
39159de2 510 */
9f1fafee 511static int complete_walk(struct nameidata *nd)
39159de2 512{
16c2cd71 513 struct dentry *dentry = nd->path.dentry;
39159de2 514 int status;
39159de2 515
9f1fafee
AV		 516 if (nd->flags & LOOKUP_RCU) {
517 nd->flags &= ~LOOKUP_RCU;
518 if (!(nd->flags & LOOKUP_ROOT))
519 nd->root.mnt = NULL;
520 spin_lock(&dentry->d_lock);
521 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
522 spin_unlock(&dentry->d_lock);
523 rcu_read_unlock();
524 br_read_unlock(vfsmount_lock);
525 return -ECHILD;
526 }
527 BUG_ON(nd->inode != dentry->d_inode);
528 spin_unlock(&dentry->d_lock);
529 mntget(nd->path.mnt);
530 rcu_read_unlock();
531 br_read_unlock(vfsmount_lock);
532 }
533
16c2cd71
AV		 534 if (likely(!(nd->flags & LOOKUP_JUMPED)))
535 return 0;
536
537 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
39159de2
JL		 538 return 0;
539
16c2cd71
AV		 540 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
541 return 0;
542
543 /* Note: we do not d_invalidate() */
34286d66 544 status = d_revalidate(dentry, nd);
39159de2
JL		 545 if (status > 0)
546 return 0;
547
16c2cd71 548 if (!status)
39159de2 549 status = -ESTALE;
16c2cd71 550
9f1fafee 551 path_put(&nd->path);
39159de2
JL		 552 return status;
553}
554
2a737871
AV		 555static __always_inline void set_root(struct nameidata *nd)
556{
f7ad3c6b
MS		 557 if (!nd->root.mnt)
558 get_fs_root(current->fs, &nd->root);
2a737871
AV		 559}
560
6de88d72
AV		 561static int link_path_walk(const char *, struct nameidata *);
562
31e6b01f
NP		 563static __always_inline void set_root_rcu(struct nameidata *nd)
564{
565 if (!nd->root.mnt) {
566 struct fs_struct *fs = current->fs;
c28cc364
NP		 567 unsigned seq;
568
569 do {
570 seq = read_seqcount_begin(&fs->seq);
571 nd->root = fs->root;
c1530019 572 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
c28cc364 573 } while (read_seqcount_retry(&fs->seq, seq));
31e6b01f
NP		 574 }
575}
576
f1662356 577static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
1da177e4 578{
31e6b01f
NP		 579 int ret;
580
1da177e4
LT		 581 if (IS_ERR(link))
582 goto fail;
583
584 if (*link == '/') {
2a737871 585 set_root(nd);
1d957f9b 586 path_put(&nd->path);
2a737871
AV		 587 nd->path = nd->root;
588 path_get(&nd->root);
16c2cd71 589 nd->flags |= LOOKUP_JUMPED;
1da177e4 590 }
31e6b01f 591 nd->inode = nd->path.dentry->d_inode;
b4091d5f 592
31e6b01f
NP		 593 ret = link_path_walk(link, nd);
594 return ret;
1da177e4 595fail:
1d957f9b 596 path_put(&nd->path);
1da177e4
LT		 597 return PTR_ERR(link);
598}
599
1d957f9b 600static void path_put_conditional(struct path *path, struct nameidata *nd)
051d3812
IK		 601{
602 dput(path->dentry);
4ac91378 603 if (path->mnt != nd->path.mnt)
051d3812
IK		 604 mntput(path->mnt);
605}
606
7b9337aa
NP		 607static inline void path_to_nameidata(const struct path *path,
608 struct nameidata *nd)
051d3812 609{
31e6b01f
NP		 610 if (!(nd->flags & LOOKUP_RCU)) {
611 dput(nd->path.dentry);
612 if (nd->path.mnt != path->mnt)
613 mntput(nd->path.mnt);
9a229683 614 }
31e6b01f 615 nd->path.mnt = path->mnt;
4ac91378 616 nd->path.dentry = path->dentry;
051d3812
IK		 617}
618
574197e0
AV		 619static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
620{
621 struct inode *inode = link->dentry->d_inode;
622 if (!IS_ERR(cookie) && inode->i_op->put_link)
623 inode->i_op->put_link(link->dentry, nd, cookie);
624 path_put(link);
625}
626
def4af30 627static __always_inline int
574197e0 628follow_link(struct path *link, struct nameidata *nd, void **p)
1da177e4
LT		 629{
630 int error;
7b9337aa 631 struct dentry *dentry = link->dentry;
1da177e4 632
844a3917
AV		 633 BUG_ON(nd->flags & LOOKUP_RCU);
634
0e794589
AV		 635 if (link->mnt == nd->path.mnt)
636 mntget(link->mnt);
637
574197e0
AV		 638 if (unlikely(current->total_link_count >= 40)) {
639 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
574197e0
AV		 640 path_put(&nd->path);
641 return -ELOOP;
642 }
643 cond_resched();
644 current->total_link_count++;
645
7b9337aa 646 touch_atime(link->mnt, dentry);
1da177e4 647 nd_set_link(nd, NULL);
cd4e91d3 648
36f3b4f6
AV		 649 error = security_inode_follow_link(link->dentry, nd);
650 if (error) {
651 *p = ERR_PTR(error); /* no ->put_link(), please */
652 path_put(&nd->path);
653 return error;
654 }
655
86acdca1 656 nd->last_type = LAST_BIND;
def4af30
AV		 657 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
658 error = PTR_ERR(*p);
659 if (!IS_ERR(*p)) {
1da177e4 660 char *s = nd_get_link(nd);
cc314eef 661 error = 0;
1da177e4
LT		 662 if (s)
663 error = __vfs_follow_link(nd, s);
bcda7652 664 else if (nd->last_type == LAST_BIND) {
16c2cd71 665 nd->flags |= LOOKUP_JUMPED;
b21041d0
AV		 666 nd->inode = nd->path.dentry->d_inode;
667 if (nd->inode->i_op->follow_link) {
bcda7652
AV		 668 /* stepped on a _really_ weird one */
669 path_put(&nd->path);
670 error = -ELOOP;
671 }
672 }
1da177e4 673 }
1da177e4
LT		 674 return error;
675}
676
31e6b01f
NP		 677static int follow_up_rcu(struct path *path)
678{
679 struct vfsmount *parent;
680 struct dentry *mountpoint;
681
682 parent = path->mnt->mnt_parent;
683 if (parent == path->mnt)
684 return 0;
685 mountpoint = path->mnt->mnt_mountpoint;
686 path->dentry = mountpoint;
687 path->mnt = parent;
688 return 1;
689}
690
bab77ebf 691int follow_up(struct path *path)
1da177e4
LT		 692{
693 struct vfsmount *parent;
694 struct dentry *mountpoint;
99b7db7b
NP		 695
696 br_read_lock(vfsmount_lock);
bab77ebf
AV		 697 parent = path->mnt->mnt_parent;
698 if (parent == path->mnt) {
99b7db7b 699 br_read_unlock(vfsmount_lock);
1da177e4
LT		 700 return 0;
701 }
702 mntget(parent);
bab77ebf 703 mountpoint = dget(path->mnt->mnt_mountpoint);
99b7db7b 704 br_read_unlock(vfsmount_lock);
bab77ebf
AV		 705 dput(path->dentry);
706 path->dentry = mountpoint;
707 mntput(path->mnt);
708 path->mnt = parent;
1da177e4
LT		 709 return 1;
710}
711
b5c84bf6 712/*
9875cf80
DH		 713 * Perform an automount
714 * - return -EISDIR to tell follow_managed() to stop and return the path we
715 * were called with.
1da177e4 716 */
9875cf80
DH		 717static int follow_automount(struct path *path, unsigned flags,
718 bool *need_mntput)
31e6b01f 719{
9875cf80 720 struct vfsmount *mnt;
ea5b778a 721 int err;
9875cf80
DH		 722
723 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
724 return -EREMOTE;
725
0ec26fd0
MS		 726 /* We don't want to mount if someone's just doing a stat -
727 * unless they're stat'ing a directory and appended a '/' to
728 * the name.
729 *
730 * We do, however, want to mount if someone wants to open or
731 * create a file of any type under the mountpoint, wants to
732 * traverse through the mountpoint or wants to open the
733 * mounted directory. Also, autofs may mark negative dentries
734 * as being automount points. These will need the attentions
735 * of the daemon to instantiate them before they can be used.
9875cf80 736 */
0ec26fd0 737 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
d94c177b 738 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
0ec26fd0
MS		 739 path->dentry->d_inode)
740 return -EISDIR;
741
9875cf80
DH		 742 current->total_link_count++;
743 if (current->total_link_count >= 40)
744 return -ELOOP;
745
746 mnt = path->dentry->d_op->d_automount(path);
747 if (IS_ERR(mnt)) {
748 /*
749 * The filesystem is allowed to return -EISDIR here to indicate
750 * it doesn't want to automount. For instance, autofs would do
751 * this so that its userspace daemon can mount on this dentry.
752 *
753 * However, we can only permit this if it's a terminal point in
754 * the path being looked up; if it wasn't then the remainder of
755 * the path is inaccessible and we should say so.
756 */
49084c3b 757 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
9875cf80
DH		 758 return -EREMOTE;
759 return PTR_ERR(mnt);
31e6b01f 760 }
ea5b778a 761
9875cf80
DH		 762 if (!mnt) /* mount collision */
763 return 0;
31e6b01f 764
8aef1884
AV		 765 if (!*need_mntput) {
766 /* lock_mount() may release path->mnt on error */
767 mntget(path->mnt);
768 *need_mntput = true;
769 }
19a167af 770 err = finish_automount(mnt, path);
9875cf80 771
ea5b778a
DH		 772 switch (err) {
773 case -EBUSY:
774 /* Someone else made a mount here whilst we were busy */
19a167af 775 return 0;
ea5b778a 776 case 0:
8aef1884 777 path_put(path);
ea5b778a
DH		 778 path->mnt = mnt;
779 path->dentry = dget(mnt->mnt_root);
ea5b778a 780 return 0;
19a167af
AV		 781 default:
782 return err;
ea5b778a 783 }
19a167af 784
463ffb2e
AV		 785}
786
9875cf80
DH		 787/*
788 * Handle a dentry that is managed in some way.
cc53ce53 789 * - Flagged for transit management (autofs)
9875cf80
DH		 790 * - Flagged as mountpoint
791 * - Flagged as automount point
792 *
793 * This may only be called in refwalk mode.
794 *
795 * Serialization is taken care of in namespace.c
796 */
797static int follow_managed(struct path *path, unsigned flags)
1da177e4 798{
8aef1884 799 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
9875cf80
DH		 800 unsigned managed;
801 bool need_mntput = false;
8aef1884 802 int ret = 0;
9875cf80
DH		 803
804 /* Given that we're not holding a lock here, we retain the value in a
805 * local variable for each dentry as we look at it so that we don't see
806 * the components of that value change under us */
807 while (managed = ACCESS_ONCE(path->dentry->d_flags),
808 managed &= DCACHE_MANAGED_DENTRY,
809 unlikely(managed != 0)) {
cc53ce53
DH		 810 /* Allow the filesystem to manage the transit without i_mutex
811 * being held. */
812 if (managed & DCACHE_MANAGE_TRANSIT) {
813 BUG_ON(!path->dentry->d_op);
814 BUG_ON(!path->dentry->d_op->d_manage);
1aed3e42 815 ret = path->dentry->d_op->d_manage(path->dentry, false);
cc53ce53 816 if (ret < 0)
8aef1884 817 break;
cc53ce53
DH		 818 }
819
9875cf80
DH		 820 /* Transit to a mounted filesystem. */
821 if (managed & DCACHE_MOUNTED) {
822 struct vfsmount *mounted = lookup_mnt(path);
823 if (mounted) {
824 dput(path->dentry);
825 if (need_mntput)
826 mntput(path->mnt);
827 path->mnt = mounted;
828 path->dentry = dget(mounted->mnt_root);
829 need_mntput = true;
830 continue;
831 }
832
833 /* Something is mounted on this dentry in another
834 * namespace and/or whatever was mounted there in this
835 * namespace got unmounted before we managed to get the
836 * vfsmount_lock */
837 }
838
839 /* Handle an automount point */
840 if (managed & DCACHE_NEED_AUTOMOUNT) {
841 ret = follow_automount(path, flags, &need_mntput);
842 if (ret < 0)
8aef1884 843 break;
9875cf80
DH		 844 continue;
845 }
846
847 /* We didn't change the current path point */
848 break;
1da177e4 849 }
8aef1884
AV		 850
851 if (need_mntput && path->mnt == mnt)
852 mntput(path->mnt);
853 if (ret == -EISDIR)
854 ret = 0;
a3fbbde7 855 return ret < 0 ? ret : need_mntput;
1da177e4
LT		 856}
857
cc53ce53 858int follow_down_one(struct path *path)
1da177e4
LT		 859{
860 struct vfsmount *mounted;
861
1c755af4 862 mounted = lookup_mnt(path);
1da177e4 863 if (mounted) {
9393bd07
AV		 864 dput(path->dentry);
865 mntput(path->mnt);
866 path->mnt = mounted;
867 path->dentry = dget(mounted->mnt_root);
1da177e4
LT		 868 return 1;
869 }
870 return 0;
871}
872
62a7375e
IK		 873static inline bool managed_dentry_might_block(struct dentry *dentry)
874{
875 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
876 dentry->d_op->d_manage(dentry, true) < 0);
877}
878
9875cf80 879/*
287548e4
AV		 880 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
881 * we meet a managed dentry that would need blocking.
9875cf80
DH		 882 */
883static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
287548e4 884 struct inode **inode)
9875cf80 885{
62a7375e 886 for (;;) {
9875cf80 887 struct vfsmount *mounted;
62a7375e
IK		 888 /*
889 * Don't forget we might have a non-mountpoint managed dentry
890 * that wants to block transit.
891 */
287548e4 892 if (unlikely(managed_dentry_might_block(path->dentry)))
ab90911f 893 return false;
62a7375e
IK		 894
895 if (!d_mountpoint(path->dentry))
896 break;
897
9875cf80
DH		 898 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
899 if (!mounted)
900 break;
901 path->mnt = mounted;
902 path->dentry = mounted->mnt_root;
a3fbbde7 903 nd->flags |= LOOKUP_JUMPED;
9875cf80 904 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
59430262
LT		 905 /*
906 * Update the inode too. We don't need to re-check the
907 * dentry sequence number here after this d_inode read,
908 * because a mount-point is always pinned.
909 */
910 *inode = path->dentry->d_inode;
9875cf80 911 }
9875cf80
DH		 912 return true;
913}
914
dea39376 915static void follow_mount_rcu(struct nameidata *nd)
287548e4 916{
dea39376 917 while (d_mountpoint(nd->path.dentry)) {
287548e4 918 struct vfsmount *mounted;
dea39376 919 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
287548e4
AV		 920 if (!mounted)
921 break;
dea39376
AV		 922 nd->path.mnt = mounted;
923 nd->path.dentry = mounted->mnt_root;
924 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
287548e4
AV		 925 }
926}
927
31e6b01f
NP		 928static int follow_dotdot_rcu(struct nameidata *nd)
929{
31e6b01f
NP		 930 set_root_rcu(nd);
931
9875cf80 932 while (1) {
31e6b01f
NP		 933 if (nd->path.dentry == nd->root.dentry &&
934 nd->path.mnt == nd->root.mnt) {
935 break;
936 }
937 if (nd->path.dentry != nd->path.mnt->mnt_root) {
938 struct dentry *old = nd->path.dentry;
939 struct dentry *parent = old->d_parent;
940 unsigned seq;
941
942 seq = read_seqcount_begin(&parent->d_seq);
943 if (read_seqcount_retry(&old->d_seq, nd->seq))
ef7562d5 944 goto failed;
31e6b01f
NP		 945 nd->path.dentry = parent;
946 nd->seq = seq;
947 break;
948 }
949 if (!follow_up_rcu(&nd->path))
950 break;
951 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
31e6b01f 952 }
dea39376
AV		 953 follow_mount_rcu(nd);
954 nd->inode = nd->path.dentry->d_inode;
31e6b01f 955 return 0;
ef7562d5
AV		 956
957failed:
958 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 959 if (!(nd->flags & LOOKUP_ROOT))
960 nd->root.mnt = NULL;
ef7562d5
AV		 961 rcu_read_unlock();
962 br_read_unlock(vfsmount_lock);
963 return -ECHILD;
31e6b01f
NP		 964}
965
cc53ce53
DH		 966/*
967 * Follow down to the covering mount currently visible to userspace. At each
968 * point, the filesystem owning that dentry may be queried as to whether the
969 * caller is permitted to proceed or not.
cc53ce53 970 */
7cc90cc3 971int follow_down(struct path *path)
cc53ce53
DH		 972{
973 unsigned managed;
974 int ret;
975
976 while (managed = ACCESS_ONCE(path->dentry->d_flags),
977 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
978 /* Allow the filesystem to manage the transit without i_mutex
979 * being held.
980 *
981 * We indicate to the filesystem if someone is trying to mount
982 * something here. This gives autofs the chance to deny anyone
983 * other than its daemon the right to mount on its
984 * superstructure.
985 *
986 * The filesystem may sleep at this point.
987 */
988 if (managed & DCACHE_MANAGE_TRANSIT) {
989 BUG_ON(!path->dentry->d_op);
990 BUG_ON(!path->dentry->d_op->d_manage);
ab90911f 991 ret = path->dentry->d_op->d_manage(
1aed3e42 992 path->dentry, false);
cc53ce53
DH		 993 if (ret < 0)
994 return ret == -EISDIR ? 0 : ret;
995 }
996
997 /* Transit to a mounted filesystem. */
998 if (managed & DCACHE_MOUNTED) {
999 struct vfsmount *mounted = lookup_mnt(path);
1000 if (!mounted)
1001 break;
1002 dput(path->dentry);
1003 mntput(path->mnt);
1004 path->mnt = mounted;
1005 path->dentry = dget(mounted->mnt_root);
1006 continue;
1007 }
1008
1009 /* Don't handle automount points here */
1010 break;
1011 }
1012 return 0;
1013}
1014
9875cf80
DH		 1015/*
1016 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1017 */
1018static void follow_mount(struct path *path)
1019{
1020 while (d_mountpoint(path->dentry)) {
1021 struct vfsmount *mounted = lookup_mnt(path);
1022 if (!mounted)
1023 break;
1024 dput(path->dentry);
1025 mntput(path->mnt);
1026 path->mnt = mounted;
1027 path->dentry = dget(mounted->mnt_root);
1028 }
1029}
1030
31e6b01f 1031static void follow_dotdot(struct nameidata *nd)
1da177e4 1032{
2a737871 1033 set_root(nd);
e518ddb7 1034
1da177e4 1035 while(1) {
4ac91378 1036 struct dentry *old = nd->path.dentry;
1da177e4 1037
2a737871
AV		 1038 if (nd->path.dentry == nd->root.dentry &&
1039 nd->path.mnt == nd->root.mnt) {
1da177e4
LT		 1040 break;
1041 }
4ac91378 1042 if (nd->path.dentry != nd->path.mnt->mnt_root) {
3088dd70
AV		 1043 /* rare case of legitimate dget_parent()... */
1044 nd->path.dentry = dget_parent(nd->path.dentry);
1da177e4
LT		 1045 dput(old);
1046 break;
1047 }
3088dd70 1048 if (!follow_up(&nd->path))
1da177e4 1049 break;
1da177e4 1050 }
79ed0226 1051 follow_mount(&nd->path);
31e6b01f 1052 nd->inode = nd->path.dentry->d_inode;
1da177e4
LT		 1053}
1054
baa03890
NP		 1055/*
1056 * Allocate a dentry with name and parent, and perform a parent
1057 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1058 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1059 * have verified that no child exists while under i_mutex.
1060 */
1061static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1062 struct qstr *name, struct nameidata *nd)
1063{
1064 struct inode *inode = parent->d_inode;
1065 struct dentry *dentry;
1066 struct dentry *old;
1067
1068 /* Don't create child dentry for a dead directory. */
1069 if (unlikely(IS_DEADDIR(inode)))
1070 return ERR_PTR(-ENOENT);
1071
1072 dentry = d_alloc(parent, name);
1073 if (unlikely(!dentry))
1074 return ERR_PTR(-ENOMEM);
1075
1076 old = inode->i_op->lookup(inode, dentry, nd);
1077 if (unlikely(old)) {
1078 dput(dentry);
1079 dentry = old;
1080 }
1081 return dentry;
1082}
1083
44396f4b
JB		 1084/*
1085 * We already have a dentry, but require a lookup to be performed on the parent
1086 * directory to fill in d_inode. Returns the new dentry, or ERR_PTR on error.
1087 * parent->d_inode->i_mutex must be held. d_lookup must have verified that no
1088 * child exists while under i_mutex.
1089 */
1090static struct dentry *d_inode_lookup(struct dentry *parent, struct dentry *dentry,
1091 struct nameidata *nd)
1092{
1093 struct inode *inode = parent->d_inode;
1094 struct dentry *old;
1095
1096 /* Don't create child dentry for a dead directory. */
1097 if (unlikely(IS_DEADDIR(inode)))
1098 return ERR_PTR(-ENOENT);
1099
1100 old = inode->i_op->lookup(inode, dentry, nd);
1101 if (unlikely(old)) {
1102 dput(dentry);
1103 dentry = old;
1104 }
1105 return dentry;
1106}
1107
1da177e4
LT		 1108/*
1109 * It's more convoluted than I'd like it to be, but... it's still fairly
1110 * small and for now I'd prefer to have fast path as straight as possible.
1111 * It _is_ time-critical.
1112 */
1113static int do_lookup(struct nameidata *nd, struct qstr *name,
31e6b01f 1114 struct path *path, struct inode **inode)
1da177e4 1115{
4ac91378 1116 struct vfsmount *mnt = nd->path.mnt;
31e6b01f 1117 struct dentry *dentry, *parent = nd->path.dentry;
5a18fff2
AV		 1118 int need_reval = 1;
1119 int status = 1;
9875cf80
DH		 1120 int err;
1121
b04f784e
NP		 1122 /*
1123 * Rename seqlock is not required here because in the off chance
1124 * of a false negative due to a concurrent rename, we're going to
1125 * do the non-racy lookup, below.
1126 */
31e6b01f
NP		 1127 if (nd->flags & LOOKUP_RCU) {
1128 unsigned seq;
31e6b01f
NP		 1129 *inode = nd->inode;
1130 dentry = __d_lookup_rcu(parent, name, &seq, inode);
5a18fff2
AV		 1131 if (!dentry)
1132 goto unlazy;
1133
31e6b01f
NP		 1134 /* Memory barrier in read_seqcount_begin of child is enough */
1135 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1136 return -ECHILD;
31e6b01f 1137 nd->seq = seq;
5a18fff2 1138
24643087 1139 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
5a18fff2
AV		 1140 status = d_revalidate(dentry, nd);
1141 if (unlikely(status <= 0)) {
1142 if (status != -ECHILD)
1143 need_reval = 0;
1144 goto unlazy;
1145 }
24643087 1146 }
44396f4b
JB		 1147 if (unlikely(d_need_lookup(dentry)))
1148 goto unlazy;
31e6b01f
NP		 1149 path->mnt = mnt;
1150 path->dentry = dentry;
d6e9bd25
AV		 1151 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1152 goto unlazy;
1153 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1154 goto unlazy;
1155 return 0;
5a18fff2 1156unlazy:
19660af7
AV		 1157 if (unlazy_walk(nd, dentry))
1158 return -ECHILD;
5a18fff2
AV		 1159 } else {
1160 dentry = __d_lookup(parent, name);
9875cf80 1161 }
5a18fff2 1162
44396f4b
JB		 1163 if (dentry && unlikely(d_need_lookup(dentry))) {
1164 dput(dentry);
1165 dentry = NULL;
1166 }
5a18fff2
AV		 1167retry:
1168 if (unlikely(!dentry)) {
1169 struct inode *dir = parent->d_inode;
1170 BUG_ON(nd->inode != dir);
1171
1172 mutex_lock(&dir->i_mutex);
1173 dentry = d_lookup(parent, name);
1174 if (likely(!dentry)) {
1175 dentry = d_alloc_and_lookup(parent, name, nd);
1176 if (IS_ERR(dentry)) {
1177 mutex_unlock(&dir->i_mutex);
1178 return PTR_ERR(dentry);
1179 }
1180 /* known good */
1181 need_reval = 0;
1182 status = 1;
44396f4b
JB		 1183 } else if (unlikely(d_need_lookup(dentry))) {
1184 dentry = d_inode_lookup(parent, dentry, nd);
1185 if (IS_ERR(dentry)) {
1186 mutex_unlock(&dir->i_mutex);
1187 return PTR_ERR(dentry);
1188 }
1189 /* known good */
1190 need_reval = 0;
1191 status = 1;
5a18fff2
AV		 1192 }
1193 mutex_unlock(&dir->i_mutex);
1194 }
1195 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1196 status = d_revalidate(dentry, nd);
1197 if (unlikely(status <= 0)) {
1198 if (status < 0) {
1199 dput(dentry);
1200 return status;
1201 }
1202 if (!d_invalidate(dentry)) {
1203 dput(dentry);
1204 dentry = NULL;
1205 need_reval = 1;
1206 goto retry;
1207 }
24643087 1208 }
5a18fff2 1209
9875cf80
DH		 1210 path->mnt = mnt;
1211 path->dentry = dentry;
1212 err = follow_managed(path, nd->flags);
89312214
IK		 1213 if (unlikely(err < 0)) {
1214 path_put_conditional(path, nd);
9875cf80 1215 return err;
89312214 1216 }
a3fbbde7
AV		 1217 if (err)
1218 nd->flags |= LOOKUP_JUMPED;
9875cf80 1219 *inode = path->dentry->d_inode;
1da177e4 1220 return 0;
1da177e4
LT		 1221}
1222
52094c8a
AV		 1223static inline int may_lookup(struct nameidata *nd)
1224{
1225 if (nd->flags & LOOKUP_RCU) {
4ad5abb3 1226 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
52094c8a
AV		 1227 if (err != -ECHILD)
1228 return err;
19660af7 1229 if (unlazy_walk(nd, NULL))
52094c8a
AV		 1230 return -ECHILD;
1231 }
4ad5abb3 1232 return inode_permission(nd->inode, MAY_EXEC);
52094c8a
AV		 1233}
1234
9856fa1b
AV		 1235static inline int handle_dots(struct nameidata *nd, int type)
1236{
1237 if (type == LAST_DOTDOT) {
1238 if (nd->flags & LOOKUP_RCU) {
1239 if (follow_dotdot_rcu(nd))
1240 return -ECHILD;
1241 } else
1242 follow_dotdot(nd);
1243 }
1244 return 0;
1245}
1246
951361f9
AV		 1247static void terminate_walk(struct nameidata *nd)
1248{
1249 if (!(nd->flags & LOOKUP_RCU)) {
1250 path_put(&nd->path);
1251 } else {
1252 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 1253 if (!(nd->flags & LOOKUP_ROOT))
1254 nd->root.mnt = NULL;
951361f9
AV		 1255 rcu_read_unlock();
1256 br_read_unlock(vfsmount_lock);
1257 }
1258}
1259
3ddcd056
LT		 1260/*
1261 * Do we need to follow links? We _really_ want to be able
1262 * to do this check without having to look at inode->i_op,
1263 * so we keep a cache of "no, this doesn't need follow_link"
1264 * for the common case.
1265 */
7813b94a 1266static inline int should_follow_link(struct inode *inode, int follow)
3ddcd056
LT		 1267{
1268 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1269 if (likely(inode->i_op->follow_link))
1270 return follow;
1271
1272 /* This gets set once for the inode lifetime */
1273 spin_lock(&inode->i_lock);
1274 inode->i_opflags |= IOP_NOFOLLOW;
1275 spin_unlock(&inode->i_lock);
1276 }
1277 return 0;
1278}
1279
ce57dfc1
AV		 1280static inline int walk_component(struct nameidata *nd, struct path *path,
1281 struct qstr *name, int type, int follow)
1282{
1283 struct inode *inode;
1284 int err;
1285 /*
1286 * "." and ".." are special - ".." especially so because it has
1287 * to be able to know about the current root directory and
1288 * parent relationships.
1289 */
1290 if (unlikely(type != LAST_NORM))
1291 return handle_dots(nd, type);
1292 err = do_lookup(nd, name, path, &inode);
1293 if (unlikely(err)) {
1294 terminate_walk(nd);
1295 return err;
1296 }
1297 if (!inode) {
1298 path_to_nameidata(path, nd);
1299 terminate_walk(nd);
1300 return -ENOENT;
1301 }
7813b94a 1302 if (should_follow_link(inode, follow)) {
19660af7
AV		 1303 if (nd->flags & LOOKUP_RCU) {
1304 if (unlikely(unlazy_walk(nd, path->dentry))) {
1305 terminate_walk(nd);
1306 return -ECHILD;
1307 }
1308 }
ce57dfc1
AV		 1309 BUG_ON(inode != path->dentry->d_inode);
1310 return 1;
1311 }
1312 path_to_nameidata(path, nd);
1313 nd->inode = inode;
1314 return 0;
1315}
1316
b356379a
AV		 1317/*
1318 * This limits recursive symlink follows to 8, while
1319 * limiting consecutive symlinks to 40.
1320 *
1321 * Without that kind of total limit, nasty chains of consecutive
1322 * symlinks can cause almost arbitrarily long lookups.
1323 */
1324static inline int nested_symlink(struct path *path, struct nameidata *nd)
1325{
1326 int res;
1327
b356379a
AV		 1328 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1329 path_put_conditional(path, nd);
1330 path_put(&nd->path);
1331 return -ELOOP;
1332 }
1a4022f8 1333 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
b356379a
AV		 1334
1335 nd->depth++;
1336 current->link_count++;
1337
1338 do {
1339 struct path link = *path;
1340 void *cookie;
574197e0
AV		 1341
1342 res = follow_link(&link, nd, &cookie);
b356379a
AV		 1343 if (!res)
1344 res = walk_component(nd, path, &nd->last,
1345 nd->last_type, LOOKUP_FOLLOW);
574197e0 1346 put_link(nd, &link, cookie);
b356379a
AV		 1347 } while (res > 0);
1348
1349 current->link_count--;
1350 nd->depth--;
1351 return res;
1352}
1353
3ddcd056
LT		 1354/*
1355 * We really don't want to look at inode->i_op->lookup
1356 * when we don't have to. So we keep a cache bit in
1357 * the inode ->i_opflags field that says "yes, we can
1358 * do lookup on this inode".
1359 */
1360static inline int can_lookup(struct inode *inode)
1361{
1362 if (likely(inode->i_opflags & IOP_LOOKUP))
1363 return 1;
1364 if (likely(!inode->i_op->lookup))
1365 return 0;
1366
1367 /* We do this once for the lifetime of the inode */
1368 spin_lock(&inode->i_lock);
1369 inode->i_opflags |= IOP_LOOKUP;
1370 spin_unlock(&inode->i_lock);
1371 return 1;
1372}
1373
1da177e4
LT		 1374/*
1375 * Name resolution.
ea3834d9
PM		 1376 * This is the basic name resolution function, turning a pathname into
1377 * the final dentry. We expect 'base' to be positive and a directory.
1da177e4 1378 *
ea3834d9
PM		 1379 * Returns 0 and nd will have valid dentry and mnt on success.
1380 * Returns error and drops reference to input namei data on failure.
1da177e4 1381 */
6de88d72 1382static int link_path_walk(const char *name, struct nameidata *nd)
1da177e4
LT		 1383{
1384 struct path next;
1da177e4 1385 int err;
1da177e4
LT		 1386
1387 while (*name=='/')
1388 name++;
1389 if (!*name)
086e183a 1390 return 0;
1da177e4 1391
1da177e4
LT		 1392 /* At this point we know we have a real path component. */
1393 for(;;) {
1394 unsigned long hash;
1395 struct qstr this;
1396 unsigned int c;
fe479a58 1397 int type;
1da177e4 1398
52094c8a 1399 err = may_lookup(nd);
1da177e4
LT		 1400 if (err)
1401 break;
1402
1403 this.name = name;
1404 c = *(const unsigned char *)name;
1405
1406 hash = init_name_hash();
1407 do {
1408 name++;
1409 hash = partial_name_hash(c, hash);
1410 c = *(const unsigned char *)name;
1411 } while (c && (c != '/'));
1412 this.len = name - (const char *) this.name;
1413 this.hash = end_name_hash(hash);
1414
fe479a58
AV		 1415 type = LAST_NORM;
1416 if (this.name[0] == '.') switch (this.len) {
1417 case 2:
16c2cd71 1418 if (this.name[1] == '.') {
fe479a58 1419 type = LAST_DOTDOT;
16c2cd71
AV		 1420 nd->flags |= LOOKUP_JUMPED;
1421 }
fe479a58
AV		 1422 break;
1423 case 1:
1424 type = LAST_DOT;
1425 }
5a202bcd
AV		 1426 if (likely(type == LAST_NORM)) {
1427 struct dentry *parent = nd->path.dentry;
16c2cd71 1428 nd->flags &= ~LOOKUP_JUMPED;
5a202bcd
AV		 1429 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1430 err = parent->d_op->d_hash(parent, nd->inode,
1431 &this);
1432 if (err < 0)
1433 break;
1434 }
1435 }
fe479a58 1436
1da177e4
LT		 1437 /* remove trailing slashes? */
1438 if (!c)
1439 goto last_component;
1440 while (*++name == '/');
1441 if (!*name)
b356379a 1442 goto last_component;
1da177e4 1443
ce57dfc1
AV		 1444 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1445 if (err < 0)
1446 return err;
1da177e4 1447
ce57dfc1 1448 if (err) {
b356379a 1449 err = nested_symlink(&next, nd);
1da177e4 1450 if (err)
a7472bab 1451 return err;
31e6b01f 1452 }
3ddcd056
LT		 1453 if (can_lookup(nd->inode))
1454 continue;
1da177e4 1455 err = -ENOTDIR;
3ddcd056 1456 break;
1da177e4
LT		 1457 /* here ends the main loop */
1458
1da177e4 1459last_component:
b356379a
AV		 1460 nd->last = this;
1461 nd->last_type = type;
086e183a 1462 return 0;
1da177e4 1463 }
951361f9 1464 terminate_walk(nd);
1da177e4
LT		 1465 return err;
1466}
1467
70e9b357
AV		 1468static int path_init(int dfd, const char *name, unsigned int flags,
1469 struct nameidata *nd, struct file **fp)
31e6b01f
NP		 1470{
1471 int retval = 0;
1472 int fput_needed;
1473 struct file *file;
1474
1475 nd->last_type = LAST_ROOT; /* if there are only slashes... */
16c2cd71 1476 nd->flags = flags | LOOKUP_JUMPED;
31e6b01f 1477 nd->depth = 0;
5b6ca027
AV		 1478 if (flags & LOOKUP_ROOT) {
1479 struct inode *inode = nd->root.dentry->d_inode;
73d049a4
AV		 1480 if (*name) {
1481 if (!inode->i_op->lookup)
1482 return -ENOTDIR;
1483 retval = inode_permission(inode, MAY_EXEC);
1484 if (retval)
1485 return retval;
1486 }
5b6ca027
AV		 1487 nd->path = nd->root;
1488 nd->inode = inode;
1489 if (flags & LOOKUP_RCU) {
1490 br_read_lock(vfsmount_lock);
1491 rcu_read_lock();
1492 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1493 } else {
1494 path_get(&nd->path);
1495 }
1496 return 0;
1497 }
1498
31e6b01f 1499 nd->root.mnt = NULL;
31e6b01f
NP		 1500
1501 if (*name=='/') {
e41f7d4e
AV		 1502 if (flags & LOOKUP_RCU) {
1503 br_read_lock(vfsmount_lock);
1504 rcu_read_lock();
1505 set_root_rcu(nd);
1506 } else {
1507 set_root(nd);
1508 path_get(&nd->root);
1509 }
1510 nd->path = nd->root;
31e6b01f 1511 } else if (dfd == AT_FDCWD) {
e41f7d4e
AV		 1512 if (flags & LOOKUP_RCU) {
1513 struct fs_struct *fs = current->fs;
1514 unsigned seq;
31e6b01f 1515
e41f7d4e
AV		 1516 br_read_lock(vfsmount_lock);
1517 rcu_read_lock();
c28cc364 1518
e41f7d4e
AV		 1519 do {
1520 seq = read_seqcount_begin(&fs->seq);
1521 nd->path = fs->pwd;
1522 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1523 } while (read_seqcount_retry(&fs->seq, seq));
1524 } else {
1525 get_fs_pwd(current->fs, &nd->path);
1526 }
31e6b01f
NP		 1527 } else {
1528 struct dentry *dentry;
1529
1abf0c71 1530 file = fget_raw_light(dfd, &fput_needed);
31e6b01f
NP		 1531 retval = -EBADF;
1532 if (!file)
1533 goto out_fail;
1534
1535 dentry = file->f_path.dentry;
1536
f52e0c11
AV		 1537 if (*name) {
1538 retval = -ENOTDIR;
1539 if (!S_ISDIR(dentry->d_inode->i_mode))
1540 goto fput_fail;
31e6b01f 1541
4ad5abb3 1542 retval = inode_permission(dentry->d_inode, MAY_EXEC);
f52e0c11
AV		 1543 if (retval)
1544 goto fput_fail;
1545 }
31e6b01f
NP		 1546
1547 nd->path = file->f_path;
e41f7d4e
AV		 1548 if (flags & LOOKUP_RCU) {
1549 if (fput_needed)
70e9b357 1550 *fp = file;
e41f7d4e
AV		 1551 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1552 br_read_lock(vfsmount_lock);
1553 rcu_read_lock();
1554 } else {
1555 path_get(&file->f_path);
1556 fput_light(file, fput_needed);
1557 }
31e6b01f 1558 }
31e6b01f 1559
31e6b01f 1560 nd->inode = nd->path.dentry->d_inode;
9b4a9b14 1561 return 0;
2dfdd266 1562
9b4a9b14
AV		 1563fput_fail:
1564 fput_light(file, fput_needed);
1565out_fail:
1566 return retval;
1567}
1568
bd92d7fe
AV		 1569static inline int lookup_last(struct nameidata *nd, struct path *path)
1570{
1571 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1572 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1573
1574 nd->flags &= ~LOOKUP_PARENT;
1575 return walk_component(nd, path, &nd->last, nd->last_type,
1576 nd->flags & LOOKUP_FOLLOW);
1577}
1578
9b4a9b14 1579/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
ee0827cd 1580static int path_lookupat(int dfd, const char *name,
9b4a9b14
AV		 1581 unsigned int flags, struct nameidata *nd)
1582{
70e9b357 1583 struct file *base = NULL;
bd92d7fe
AV		 1584 struct path path;
1585 int err;
31e6b01f
NP		 1586
1587 /*
1588 * Path walking is largely split up into 2 different synchronisation
1589 * schemes, rcu-walk and ref-walk (explained in
1590 * Documentation/filesystems/path-lookup.txt). These share much of the
1591 * path walk code, but some things particularly setup, cleanup, and
1592 * following mounts are sufficiently divergent that functions are
1593 * duplicated. Typically there is a function foo(), and its RCU
1594 * analogue, foo_rcu().
1595 *
1596 * -ECHILD is the error number of choice (just to avoid clashes) that
1597 * is returned if some aspect of an rcu-walk fails. Such an error must
1598 * be handled by restarting a traditional ref-walk (which will always
1599 * be able to complete).
1600 */
bd92d7fe 1601 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
ee0827cd 1602
bd92d7fe
AV		 1603 if (unlikely(err))
1604 return err;
ee0827cd
AV		 1605
1606 current->total_link_count = 0;
bd92d7fe
AV		 1607 err = link_path_walk(name, nd);
1608
1609 if (!err && !(flags & LOOKUP_PARENT)) {
bd92d7fe
AV		 1610 err = lookup_last(nd, &path);
1611 while (err > 0) {
1612 void *cookie;
1613 struct path link = path;
bd92d7fe 1614 nd->flags |= LOOKUP_PARENT;
574197e0 1615 err = follow_link(&link, nd, &cookie);
bd92d7fe
AV		 1616 if (!err)
1617 err = lookup_last(nd, &path);
574197e0 1618 put_link(nd, &link, cookie);
bd92d7fe
AV		 1619 }
1620 }
ee0827cd 1621
9f1fafee
AV		 1622 if (!err)
1623 err = complete_walk(nd);
bd92d7fe
AV		 1624
1625 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1626 if (!nd->inode->i_op->lookup) {
1627 path_put(&nd->path);
bd23a539 1628 err = -ENOTDIR;
bd92d7fe
AV		 1629 }
1630 }
16c2cd71 1631
70e9b357
AV		 1632 if (base)
1633 fput(base);
ee0827cd 1634
5b6ca027 1635 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
2a737871
AV		 1636 path_put(&nd->root);
1637 nd->root.mnt = NULL;
1638 }
bd92d7fe 1639 return err;
ee0827cd 1640}
31e6b01f 1641
ee0827cd
AV		 1642static int do_path_lookup(int dfd, const char *name,
1643 unsigned int flags, struct nameidata *nd)
1644{
1645 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1646 if (unlikely(retval == -ECHILD))
1647 retval = path_lookupat(dfd, name, flags, nd);
1648 if (unlikely(retval == -ESTALE))
1649 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
31e6b01f
NP		 1650
1651 if (likely(!retval)) {
1652 if (unlikely(!audit_dummy_context())) {
1653 if (nd->path.dentry && nd->inode)
1654 audit_inode(name, nd->path.dentry);
1655 }
1656 }
170aa3d0 1657 return retval;
1da177e4
LT		 1658}
1659
c9c6cac0 1660int kern_path_parent(const char *name, struct nameidata *nd)
5590ff0d 1661{
c9c6cac0 1662 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
5590ff0d
UD		 1663}
1664
d1811465
AV		 1665int kern_path(const char *name, unsigned int flags, struct path *path)
1666{
1667 struct nameidata nd;
1668 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1669 if (!res)
1670 *path = nd.path;
1671 return res;
1672}
1673
16f18200
JJS		 1674/**
1675 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1676 * @dentry: pointer to dentry of the base directory
1677 * @mnt: pointer to vfs mount of the base directory
1678 * @name: pointer to file name
1679 * @flags: lookup flags
e0a01249 1680 * @path: pointer to struct path to fill
16f18200
JJS		 1681 */
1682int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1683 const char *name, unsigned int flags,
e0a01249 1684 struct path *path)
16f18200 1685{
e0a01249
AV		 1686 struct nameidata nd;
1687 int err;
1688 nd.root.dentry = dentry;
1689 nd.root.mnt = mnt;
1690 BUG_ON(flags & LOOKUP_PARENT);
5b6ca027 1691 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
e0a01249
AV		 1692 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1693 if (!err)
1694 *path = nd.path;
1695 return err;
16f18200
JJS		 1696}
1697
eead1911
CH		 1698static struct dentry *__lookup_hash(struct qstr *name,
1699 struct dentry *base, struct nameidata *nd)
1da177e4 1700{
81fca444 1701 struct inode *inode = base->d_inode;
057f6c01 1702 struct dentry *dentry;
1da177e4
LT		 1703 int err;
1704
4ad5abb3 1705 err = inode_permission(inode, MAY_EXEC);
81fca444
CH		 1706 if (err)
1707 return ERR_PTR(err);
1da177e4 1708
b04f784e
NP		 1709 /*
1710 * Don't bother with __d_lookup: callers are for creat as
1711 * well as unlink, so a lot of the time it would cost
1712 * a double lookup.
6e6b1bd1 1713 */
b04f784e 1714 dentry = d_lookup(base, name);
6e6b1bd1 1715
44396f4b
JB		 1716 if (dentry && d_need_lookup(dentry)) {
1717 /*
1718 * __lookup_hash is called with the parent dir's i_mutex already
1719 * held, so we are good to go here.
1720 */
1721 dentry = d_inode_lookup(base, dentry, nd);
1722 if (IS_ERR(dentry))
1723 return dentry;
1724 }
1725
d2d9e9fb
AV		 1726 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1727 int status = d_revalidate(dentry, nd);
1728 if (unlikely(status <= 0)) {
1729 /*
1730 * The dentry failed validation.
1731 * If d_revalidate returned 0 attempt to invalidate
1732 * the dentry otherwise d_revalidate is asking us
1733 * to return a fail status.
1734 */
1735 if (status < 0) {
1736 dput(dentry);
1737 return ERR_PTR(status);
1738 } else if (!d_invalidate(dentry)) {
1739 dput(dentry);
1740 dentry = NULL;
1741 }
1742 }
1743 }
6e6b1bd1 1744
baa03890
NP		 1745 if (!dentry)
1746 dentry = d_alloc_and_lookup(base, name, nd);
5a202bcd 1747
1da177e4
LT		 1748 return dentry;
1749}
1750
057f6c01
JM		 1751/*
1752 * Restricted form of lookup. Doesn't follow links, single-component only,
1753 * needs parent already locked. Doesn't follow mounts.
1754 * SMP-safe.
1755 */
eead1911 1756static struct dentry *lookup_hash(struct nameidata *nd)
057f6c01 1757{
4ac91378 1758 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1da177e4
LT		 1759}
1760
eead1911 1761/**
a6b91919 1762 * lookup_one_len - filesystem helper to lookup single pathname component
eead1911
CH		 1763 * @name: pathname component to lookup
1764 * @base: base directory to lookup from
1765 * @len: maximum length @len should be interpreted to
1766 *
a6b91919
RD		 1767 * Note that this routine is purely a helper for filesystem usage and should
1768 * not be called by generic code. Also note that by using this function the
eead1911
CH		 1769 * nameidata argument is passed to the filesystem methods and a filesystem
1770 * using this helper needs to be prepared for that.
1771 */
057f6c01
JM		 1772struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1773{
057f6c01 1774 struct qstr this;
6a96ba54
AV		 1775 unsigned long hash;
1776 unsigned int c;
057f6c01 1777
2f9092e1
DW		 1778 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1779
6a96ba54
AV		 1780 this.name = name;
1781 this.len = len;
1782 if (!len)
1783 return ERR_PTR(-EACCES);
1784
1785 hash = init_name_hash();
1786 while (len--) {
1787 c = *(const unsigned char *)name++;
1788 if (c == '/' || c == '\0')
1789 return ERR_PTR(-EACCES);
1790 hash = partial_name_hash(c, hash);
1791 }
1792 this.hash = end_name_hash(hash);
5a202bcd
AV		 1793 /*
1794 * See if the low-level filesystem might want
1795 * to use its own hash..
1796 */
1797 if (base->d_flags & DCACHE_OP_HASH) {
1798 int err = base->d_op->d_hash(base, base->d_inode, &this);
1799 if (err < 0)
1800 return ERR_PTR(err);
1801 }
eead1911 1802
49705b77 1803 return __lookup_hash(&this, base, NULL);
057f6c01
JM		 1804}
1805
1fa1e7f6
AW		 1806int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
1807 struct path *path, int *empty)
1da177e4 1808{
2d8f3038 1809 struct nameidata nd;
1fa1e7f6 1810 char *tmp = getname_flags(name, flags, empty);
1da177e4 1811 int err = PTR_ERR(tmp);
1da177e4 1812 if (!IS_ERR(tmp)) {
2d8f3038
AV		 1813
1814 BUG_ON(flags & LOOKUP_PARENT);
1815
1816 err = do_path_lookup(dfd, tmp, flags, &nd);
1da177e4 1817 putname(tmp);
2d8f3038
AV		 1818 if (!err)
1819 *path = nd.path;
1da177e4
LT		 1820 }
1821 return err;
1822}
1823
1fa1e7f6
AW		 1824int user_path_at(int dfd, const char __user *name, unsigned flags,
1825 struct path *path)
1826{
1827 return user_path_at_empty(dfd, name, flags, path, 0);
1828}
1829
2ad94ae6
AV		 1830static int user_path_parent(int dfd, const char __user *path,
1831 struct nameidata *nd, char **name)
1832{
1833 char *s = getname(path);
1834 int error;
1835
1836 if (IS_ERR(s))
1837 return PTR_ERR(s);
1838
1839 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1840 if (error)
1841 putname(s);
1842 else
1843 *name = s;
1844
1845 return error;
1846}
1847
1da177e4
LT		 1848/*
1849 * It's inline, so penalty for filesystems that don't use sticky bit is
1850 * minimal.
1851 */
1852static inline int check_sticky(struct inode *dir, struct inode *inode)
1853{
da9592ed
DH		 1854 uid_t fsuid = current_fsuid();
1855
1da177e4
LT		 1856 if (!(dir->i_mode & S_ISVTX))
1857 return 0;
e795b717
SH		 1858 if (current_user_ns() != inode_userns(inode))
1859 goto other_userns;
da9592ed 1860 if (inode->i_uid == fsuid)
1da177e4 1861 return 0;
da9592ed 1862 if (dir->i_uid == fsuid)
1da177e4 1863 return 0;
e795b717
SH		 1864
1865other_userns:
1866 return !ns_capable(inode_userns(inode), CAP_FOWNER);
1da177e4
LT		 1867}
1868
1869/*
1870 * Check whether we can remove a link victim from directory dir, check
1871 * whether the type of victim is right.
1872 * 1. We can't do it if dir is read-only (done in permission())
1873 * 2. We should have write and exec permissions on dir
1874 * 3. We can't remove anything from append-only dir
1875 * 4. We can't do anything with immutable dir (done in permission())
1876 * 5. If the sticky bit on dir is set we should either
1877 * a. be owner of dir, or
1878 * b. be owner of victim, or
1879 * c. have CAP_FOWNER capability
1880 * 6. If the victim is append-only or immutable we can't do antyhing with
1881 * links pointing to it.
1882 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1883 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1884 * 9. We can't remove a root or mountpoint.
1885 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1886 * nfs_async_unlink().
1887 */
858119e1 1888static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1da177e4
LT		 1889{
1890 int error;
1891
1892 if (!victim->d_inode)
1893 return -ENOENT;
1894
1895 BUG_ON(victim->d_parent->d_inode != dir);
cccc6bba 1896 audit_inode_child(victim, dir);
1da177e4 1897
f419a2e3 1898 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1899 if (error)
1900 return error;
1901 if (IS_APPEND(dir))
1902 return -EPERM;
1903 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
f9454548 1904 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1da177e4
LT		 1905 return -EPERM;
1906 if (isdir) {
1907 if (!S_ISDIR(victim->d_inode->i_mode))
1908 return -ENOTDIR;
1909 if (IS_ROOT(victim))
1910 return -EBUSY;
1911 } else if (S_ISDIR(victim->d_inode->i_mode))
1912 return -EISDIR;
1913 if (IS_DEADDIR(dir))
1914 return -ENOENT;
1915 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1916 return -EBUSY;
1917 return 0;
1918}
1919
1920/* Check whether we can create an object with dentry child in directory
1921 * dir.
1922 * 1. We can't do it if child already exists (open has special treatment for
1923 * this case, but since we are inlined it's OK)
1924 * 2. We can't do it if dir is read-only (done in permission())
1925 * 3. We should have write and exec permissions on dir
1926 * 4. We can't do it if dir is immutable (done in permission())
1927 */
a95164d9 1928static inline int may_create(struct inode *dir, struct dentry *child)
1da177e4
LT		 1929{
1930 if (child->d_inode)
1931 return -EEXIST;
1932 if (IS_DEADDIR(dir))
1933 return -ENOENT;
f419a2e3 1934 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1935}
1936
1da177e4
LT		 1937/*
1938 * p1 and p2 should be directories on the same fs.
1939 */
1940struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1941{
1942 struct dentry *p;
1943
1944 if (p1 == p2) {
f2eace23 1945 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1da177e4
LT		 1946 return NULL;
1947 }
1948
a11f3a05 1949 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4 1950
e2761a11
OH		 1951 p = d_ancestor(p2, p1);
1952 if (p) {
1953 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1954 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1955 return p;
1da177e4
LT		 1956 }
1957
e2761a11
OH		 1958 p = d_ancestor(p1, p2);
1959 if (p) {
1960 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1961 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1962 return p;
1da177e4
LT		 1963 }
1964
f2eace23
IM		 1965 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1966 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1da177e4
LT		 1967 return NULL;
1968}
1969
1970void unlock_rename(struct dentry *p1, struct dentry *p2)
1971{
1b1dcc1b 1972 mutex_unlock(&p1->d_inode->i_mutex);
1da177e4 1973 if (p1 != p2) {
1b1dcc1b 1974 mutex_unlock(&p2->d_inode->i_mutex);
a11f3a05 1975 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4
LT		 1976 }
1977}
1978
1979int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1980 struct nameidata *nd)
1981{
a95164d9 1982 int error = may_create(dir, dentry);
1da177e4
LT		 1983
1984 if (error)
1985 return error;
1986
acfa4380 1987 if (!dir->i_op->create)
1da177e4
LT		 1988 return -EACCES; /* shouldn't it be ENOSYS? */
1989 mode &= S_IALLUGO;
1990 mode |= S_IFREG;
1991 error = security_inode_create(dir, dentry, mode);
1992 if (error)
1993 return error;
1da177e4 1994 error = dir->i_op->create(dir, dentry, mode, nd);
a74574aa 1995 if (!error)
f38aa942 1996 fsnotify_create(dir, dentry);
1da177e4
LT		 1997 return error;
1998}
1999
73d049a4 2000static int may_open(struct path *path, int acc_mode, int flag)
1da177e4 2001{
3fb64190 2002 struct dentry *dentry = path->dentry;
1da177e4
LT		 2003 struct inode *inode = dentry->d_inode;
2004 int error;
2005
bcda7652
AV		 2006 /* O_PATH? */
2007 if (!acc_mode)
2008 return 0;
2009
1da177e4
LT		 2010 if (!inode)
2011 return -ENOENT;
2012
c8fe8f30
CH		 2013 switch (inode->i_mode & S_IFMT) {
2014 case S_IFLNK:
1da177e4 2015 return -ELOOP;
c8fe8f30
CH		 2016 case S_IFDIR:
2017 if (acc_mode & MAY_WRITE)
2018 return -EISDIR;
2019 break;
2020 case S_IFBLK:
2021 case S_IFCHR:
3fb64190 2022 if (path->mnt->mnt_flags & MNT_NODEV)
1da177e4 2023 return -EACCES;
c8fe8f30
CH		 2024 /*FALLTHRU*/
2025 case S_IFIFO:
2026 case S_IFSOCK:
1da177e4 2027 flag &= ~O_TRUNC;
c8fe8f30 2028 break;
4a3fd211 2029 }
b41572e9 2030
3fb64190 2031 error = inode_permission(inode, acc_mode);
b41572e9
DH		 2032 if (error)
2033 return error;
6146f0d5 2034
1da177e4
LT		 2035 /*
2036 * An append-only file must be opened in append mode for writing.
2037 */
2038 if (IS_APPEND(inode)) {
8737c930 2039 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
7715b521 2040 return -EPERM;
1da177e4 2041 if (flag & O_TRUNC)
7715b521 2042 return -EPERM;
1da177e4
LT		 2043 }
2044
2045 /* O_NOATIME can only be set by the owner or superuser */
2e149670 2046 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
7715b521 2047 return -EPERM;
1da177e4 2048
f3c7691e 2049 return 0;
7715b521 2050}
1da177e4 2051
e1181ee6 2052static int handle_truncate(struct file *filp)
7715b521 2053{
e1181ee6 2054 struct path *path = &filp->f_path;
7715b521
AV		 2055 struct inode *inode = path->dentry->d_inode;
2056 int error = get_write_access(inode);
2057 if (error)
2058 return error;
2059 /*
2060 * Refuse to truncate files with mandatory locks held on them.
2061 */
2062 error = locks_verify_locked(inode);
2063 if (!error)
ea0d3ab2 2064 error = security_path_truncate(path);
7715b521
AV		 2065 if (!error) {
2066 error = do_truncate(path->dentry, 0,
2067 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
e1181ee6 2068 filp);
7715b521
AV		 2069 }
2070 put_write_access(inode);
acd0c935 2071 return error;
1da177e4
LT		 2072}
2073
d57999e1
DH		 2074static inline int open_to_namei_flags(int flag)
2075{
8a5e929d
AV		 2076 if ((flag & O_ACCMODE) == 3)
2077 flag--;
d57999e1
DH		 2078 return flag;
2079}
2080
31e6b01f 2081/*
fe2d35ff 2082 * Handle the last step of open()
31e6b01f 2083 */
fb1cc555 2084static struct file *do_last(struct nameidata *nd, struct path *path,
c3e380b0 2085 const struct open_flags *op, const char *pathname)
fb1cc555 2086{
a1e28038 2087 struct dentry *dir = nd->path.dentry;
6c0d46c4 2088 struct dentry *dentry;
ca344a89 2089 int open_flag = op->open_flag;
6c0d46c4 2090 int will_truncate = open_flag & O_TRUNC;
ca344a89 2091 int want_write = 0;
bcda7652 2092 int acc_mode = op->acc_mode;
fb1cc555 2093 struct file *filp;
16c2cd71 2094 int error;
1f36f774 2095
c3e380b0
AV		 2096 nd->flags &= ~LOOKUP_PARENT;
2097 nd->flags |= op->intent;
2098
1f36f774
AV		 2099 switch (nd->last_type) {
2100 case LAST_DOTDOT:
176306f5 2101 case LAST_DOT:
fe2d35ff
AV		 2102 error = handle_dots(nd, nd->last_type);
2103 if (error)
2104 return ERR_PTR(error);
1f36f774 2105 /* fallthrough */
1f36f774 2106 case LAST_ROOT:
9f1fafee 2107 error = complete_walk(nd);
16c2cd71 2108 if (error)
9f1fafee 2109 return ERR_PTR(error);
fe2d35ff 2110 audit_inode(pathname, nd->path.dentry);
ca344a89 2111 if (open_flag & O_CREAT) {
fe2d35ff
AV		 2112 error = -EISDIR;
2113 goto exit;
2114 }
2115 goto ok;
1f36f774 2116 case LAST_BIND:
9f1fafee 2117 error = complete_walk(nd);
16c2cd71 2118 if (error)
9f1fafee 2119 return ERR_PTR(error);
1f36f774 2120 audit_inode(pathname, dir);
67ee3ad2 2121 goto ok;
1f36f774 2122 }
67ee3ad2 2123
ca344a89 2124 if (!(open_flag & O_CREAT)) {
bcda7652 2125 int symlink_ok = 0;
fe2d35ff
AV		 2126 if (nd->last.name[nd->last.len])
2127 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
bcda7652
AV		 2128 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2129 symlink_ok = 1;
fe2d35ff 2130 /* we _can_ be in RCU mode here */
ce57dfc1
AV		 2131 error = walk_component(nd, path, &nd->last, LAST_NORM,
2132 !symlink_ok);
2133 if (error < 0)
fe2d35ff 2134 return ERR_PTR(error);
ce57dfc1 2135 if (error) /* symlink */
fe2d35ff 2136 return NULL;
fe2d35ff 2137 /* sayonara */
9f1fafee
AV		 2138 error = complete_walk(nd);
2139 if (error)
2140 return ERR_PTR(-ECHILD);
fe2d35ff
AV		 2141
2142 error = -ENOTDIR;
2143 if (nd->flags & LOOKUP_DIRECTORY) {
ce57dfc1 2144 if (!nd->inode->i_op->lookup)
fe2d35ff
AV		 2145 goto exit;
2146 }
2147 audit_inode(pathname, nd->path.dentry);
2148 goto ok;
2149 }
2150
2151 /* create side of things */
a3fbbde7
AV		 2152 /*
2153 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED has been
2154 * cleared when we got to the last component we are about to look up
2155 */
9f1fafee
AV		 2156 error = complete_walk(nd);
2157 if (error)
2158 return ERR_PTR(error);
fe2d35ff
AV		 2159
2160 audit_inode(pathname, dir);
16c2cd71 2161 error = -EISDIR;
1f36f774 2162 /* trailing slashes? */
31e6b01f
NP		 2163 if (nd->last.name[nd->last.len])
2164 goto exit;
a2c36b45 2165
a1e28038
AV		 2166 mutex_lock(&dir->d_inode->i_mutex);
2167
6c0d46c4
AV		 2168 dentry = lookup_hash(nd);
2169 error = PTR_ERR(dentry);
2170 if (IS_ERR(dentry)) {
fb1cc555
AV		 2171 mutex_unlock(&dir->d_inode->i_mutex);
2172 goto exit;
2173 }
2174
6c0d46c4
AV		 2175 path->dentry = dentry;
2176 path->mnt = nd->path.mnt;
2177
fb1cc555 2178 /* Negative dentry, just create the file */
6c0d46c4
AV		 2179 if (!dentry->d_inode) {
2180 int mode = op->mode;
2181 if (!IS_POSIXACL(dir->d_inode))
2182 mode &= ~current_umask();
fb1cc555
AV		 2183 /*
2184 * This write is needed to ensure that a
6c0d46c4 2185 * rw->ro transition does not occur between
fb1cc555
AV		 2186 * the time when the file is created and when
2187 * a permanent write count is taken through
2188 * the 'struct file' in nameidata_to_filp().
2189 */
2190 error = mnt_want_write(nd->path.mnt);
2191 if (error)
2192 goto exit_mutex_unlock;
ca344a89 2193 want_write = 1;
9b44f1b3 2194 /* Don't check for write permission, don't truncate */
ca344a89 2195 open_flag &= ~O_TRUNC;
6c0d46c4 2196 will_truncate = 0;
bcda7652 2197 acc_mode = MAY_OPEN;
6c0d46c4
AV		 2198 error = security_path_mknod(&nd->path, dentry, mode, 0);
2199 if (error)
2200 goto exit_mutex_unlock;
2201 error = vfs_create(dir->d_inode, dentry, mode, nd);
2202 if (error)
2203 goto exit_mutex_unlock;
2204 mutex_unlock(&dir->d_inode->i_mutex);
2205 dput(nd->path.dentry);
2206 nd->path.dentry = dentry;
ca344a89 2207 goto common;
fb1cc555
AV		 2208 }
2209
2210 /*
2211 * It already exists.
2212 */
2213 mutex_unlock(&dir->d_inode->i_mutex);
2214 audit_inode(pathname, path->dentry);
2215
2216 error = -EEXIST;
ca344a89 2217 if (open_flag & O_EXCL)
fb1cc555
AV		 2218 goto exit_dput;
2219
9875cf80
DH		 2220 error = follow_managed(path, nd->flags);
2221 if (error < 0)
2222 goto exit_dput;
fb1cc555 2223
a3fbbde7
AV		 2224 if (error)
2225 nd->flags |= LOOKUP_JUMPED;
2226
fb1cc555
AV		 2227 error = -ENOENT;
2228 if (!path->dentry->d_inode)
2229 goto exit_dput;
9e67f361
AV		 2230
2231 if (path->dentry->d_inode->i_op->follow_link)
fb1cc555 2232 return NULL;
fb1cc555
AV		 2233
2234 path_to_nameidata(path, nd);
31e6b01f 2235 nd->inode = path->dentry->d_inode;
a3fbbde7
AV		 2236 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2237 error = complete_walk(nd);
2238 if (error)
2239 goto exit;
fb1cc555 2240 error = -EISDIR;
31e6b01f 2241 if (S_ISDIR(nd->inode->i_mode))
fb1cc555 2242 goto exit;
67ee3ad2 2243ok:
6c0d46c4
AV		 2244 if (!S_ISREG(nd->inode->i_mode))
2245 will_truncate = 0;
2246
0f9d1a10
AV		 2247 if (will_truncate) {
2248 error = mnt_want_write(nd->path.mnt);
2249 if (error)
2250 goto exit;
ca344a89 2251 want_write = 1;
0f9d1a10 2252 }
ca344a89 2253common:
bcda7652 2254 error = may_open(&nd->path, acc_mode, open_flag);
ca344a89 2255 if (error)
0f9d1a10 2256 goto exit;
0f9d1a10
AV		 2257 filp = nameidata_to_filp(nd);
2258 if (!IS_ERR(filp)) {
2259 error = ima_file_check(filp, op->acc_mode);
2260 if (error) {
2261 fput(filp);
2262 filp = ERR_PTR(error);
2263 }
2264 }
2265 if (!IS_ERR(filp)) {
2266 if (will_truncate) {
2267 error = handle_truncate(filp);
2268 if (error) {
2269 fput(filp);
2270 filp = ERR_PTR(error);
2271 }
2272 }
2273 }
ca344a89
AV		 2274out:
2275 if (want_write)
0f9d1a10
AV		 2276 mnt_drop_write(nd->path.mnt);
2277 path_put(&nd->path);
fb1cc555
AV		 2278 return filp;
2279
2280exit_mutex_unlock:
2281 mutex_unlock(&dir->d_inode->i_mutex);
2282exit_dput:
2283 path_put_conditional(path, nd);
2284exit:
ca344a89
AV		 2285 filp = ERR_PTR(error);
2286 goto out;
fb1cc555
AV		 2287}
2288
13aab428 2289static struct file *path_openat(int dfd, const char *pathname,
73d049a4 2290 struct nameidata *nd, const struct open_flags *op, int flags)
1da177e4 2291{
fe2d35ff 2292 struct file *base = NULL;
4a3fd211 2293 struct file *filp;
9850c056 2294 struct path path;
13aab428 2295 int error;
31e6b01f
NP		 2296
2297 filp = get_empty_filp();
2298 if (!filp)
2299 return ERR_PTR(-ENFILE);
2300
47c805dc 2301 filp->f_flags = op->open_flag;
73d049a4
AV		 2302 nd->intent.open.file = filp;
2303 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2304 nd->intent.open.create_mode = op->mode;
31e6b01f 2305
73d049a4 2306 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
31e6b01f 2307 if (unlikely(error))
13aab428 2308 goto out_filp;
31e6b01f 2309
fe2d35ff 2310 current->total_link_count = 0;
73d049a4 2311 error = link_path_walk(pathname, nd);
31e6b01f
NP		 2312 if (unlikely(error))
2313 goto out_filp;
1da177e4 2314
73d049a4 2315 filp = do_last(nd, &path, op, pathname);
806b681c 2316 while (unlikely(!filp)) { /* trailing symlink */
7b9337aa 2317 struct path link = path;
def4af30 2318 void *cookie;
574197e0 2319 if (!(nd->flags & LOOKUP_FOLLOW)) {
73d049a4
AV		 2320 path_put_conditional(&path, nd);
2321 path_put(&nd->path);
40b39136
AV		 2322 filp = ERR_PTR(-ELOOP);
2323 break;
2324 }
73d049a4
AV		 2325 nd->flags |= LOOKUP_PARENT;
2326 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
574197e0 2327 error = follow_link(&link, nd, &cookie);
c3e380b0 2328 if (unlikely(error))
f1afe9ef 2329 filp = ERR_PTR(error);
c3e380b0 2330 else
73d049a4 2331 filp = do_last(nd, &path, op, pathname);
574197e0 2332 put_link(nd, &link, cookie);
806b681c 2333 }
10fa8e62 2334out:
73d049a4
AV		 2335 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2336 path_put(&nd->root);
fe2d35ff
AV		 2337 if (base)
2338 fput(base);
73d049a4 2339 release_open_intent(nd);
10fa8e62 2340 return filp;
1da177e4 2341
31e6b01f 2342out_filp:
806b681c 2343 filp = ERR_PTR(error);
10fa8e62 2344 goto out;
1da177e4
LT		 2345}
2346
13aab428
AV		 2347struct file *do_filp_open(int dfd, const char *pathname,
2348 const struct open_flags *op, int flags)
2349{
73d049a4 2350 struct nameidata nd;
13aab428
AV		 2351 struct file *filp;
2352
73d049a4 2353 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
13aab428 2354 if (unlikely(filp == ERR_PTR(-ECHILD)))
73d049a4 2355 filp = path_openat(dfd, pathname, &nd, op, flags);
13aab428 2356 if (unlikely(filp == ERR_PTR(-ESTALE)))
73d049a4 2357 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
13aab428
AV		 2358 return filp;
2359}
2360
73d049a4
AV		 2361struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2362 const char *name, const struct open_flags *op, int flags)
2363{
2364 struct nameidata nd;
2365 struct file *file;
2366
2367 nd.root.mnt = mnt;
2368 nd.root.dentry = dentry;
2369
2370 flags |= LOOKUP_ROOT;
2371
bcda7652 2372 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
73d049a4
AV		 2373 return ERR_PTR(-ELOOP);
2374
2375 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2376 if (unlikely(file == ERR_PTR(-ECHILD)))
2377 file = path_openat(-1, name, &nd, op, flags);
2378 if (unlikely(file == ERR_PTR(-ESTALE)))
2379 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2380 return file;
2381}
2382
ed75e95d 2383struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
1da177e4 2384{
c663e5d8 2385 struct dentry *dentry = ERR_PTR(-EEXIST);
ed75e95d
AV		 2386 struct nameidata nd;
2387 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2388 if (error)
2389 return ERR_PTR(error);
1da177e4 2390
c663e5d8
CH		 2391 /*
2392 * Yucky last component or no last component at all?
2393 * (foo/., foo/.., /////)
2394 */
ed75e95d
AV		 2395 if (nd.last_type != LAST_NORM)
2396 goto out;
2397 nd.flags &= ~LOOKUP_PARENT;
2398 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2399 nd.intent.open.flags = O_EXCL;
c663e5d8
CH		 2400
2401 /*
2402 * Do the final lookup.
2403 */
ed75e95d
AV		 2404 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2405 dentry = lookup_hash(&nd);
1da177e4
LT		 2406 if (IS_ERR(dentry))
2407 goto fail;
c663e5d8 2408
e9baf6e5
AV		 2409 if (dentry->d_inode)
2410 goto eexist;
c663e5d8
CH		 2411 /*
2412 * Special case - lookup gave negative, but... we had foo/bar/
2413 * From the vfs_mknod() POV we just have a negative dentry -
2414 * all is fine. Let's be bastards - you had / on the end, you've
2415 * been asking for (non-existent) directory. -ENOENT for you.
2416 */
ed75e95d 2417 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
e9baf6e5
AV		 2418 dput(dentry);
2419 dentry = ERR_PTR(-ENOENT);
ed75e95d 2420 goto fail;
e9baf6e5 2421 }
ed75e95d 2422 *path = nd.path;
1da177e4 2423 return dentry;
e9baf6e5 2424eexist:
1da177e4 2425 dput(dentry);
e9baf6e5 2426 dentry = ERR_PTR(-EEXIST);
1da177e4 2427fail:
ed75e95d
AV		 2428 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2429out:
2430 path_put(&nd.path);
1da177e4
LT		 2431 return dentry;
2432}
dae6ad8f
AV		 2433EXPORT_SYMBOL(kern_path_create);
2434
2435struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2436{
2437 char *tmp = getname(pathname);
2438 struct dentry *res;
2439 if (IS_ERR(tmp))
2440 return ERR_CAST(tmp);
2441 res = kern_path_create(dfd, tmp, path, is_dir);
2442 putname(tmp);
2443 return res;
2444}
2445EXPORT_SYMBOL(user_path_create);
2446
1da177e4
LT		 2447int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2448{
a95164d9 2449 int error = may_create(dir, dentry);
1da177e4
LT		 2450
2451 if (error)
2452 return error;
2453
e795b717
SH		 2454 if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
2455 !ns_capable(inode_userns(dir), CAP_MKNOD))
1da177e4
LT		 2456 return -EPERM;
2457
acfa4380 2458 if (!dir->i_op->mknod)
1da177e4
LT		 2459 return -EPERM;
2460
08ce5f16
SH		 2461 error = devcgroup_inode_mknod(mode, dev);
2462 if (error)
2463 return error;
2464
1da177e4
LT		 2465 error = security_inode_mknod(dir, dentry, mode, dev);
2466 if (error)
2467 return error;
2468
1da177e4 2469 error = dir->i_op->mknod(dir, dentry, mode, dev);
a74574aa 2470 if (!error)
f38aa942 2471 fsnotify_create(dir, dentry);
1da177e4
LT		 2472 return error;
2473}
2474
463c3197
DH		 2475static int may_mknod(mode_t mode)
2476{
2477 switch (mode & S_IFMT) {
2478 case S_IFREG:
2479 case S_IFCHR:
2480 case S_IFBLK:
2481 case S_IFIFO:
2482 case S_IFSOCK:
2483 case 0: /* zero mode translates to S_IFREG */
2484 return 0;
2485 case S_IFDIR:
2486 return -EPERM;
2487 default:
2488 return -EINVAL;
2489 }
2490}
2491
8208a22b 2492SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
2e4d0924 2493 unsigned, dev)
1da177e4 2494{
2ad94ae6 2495 struct dentry *dentry;
dae6ad8f
AV		 2496 struct path path;
2497 int error;
1da177e4
LT		 2498
2499 if (S_ISDIR(mode))
2500 return -EPERM;
1da177e4 2501
dae6ad8f
AV		 2502 dentry = user_path_create(dfd, filename, &path, 0);
2503 if (IS_ERR(dentry))
2504 return PTR_ERR(dentry);
2ad94ae6 2505
dae6ad8f 2506 if (!IS_POSIXACL(path.dentry->d_inode))
ce3b0f8d 2507 mode &= ~current_umask();
463c3197
DH		 2508 error = may_mknod(mode);
2509 if (error)
2510 goto out_dput;
dae6ad8f 2511 error = mnt_want_write(path.mnt);
463c3197
DH		 2512 if (error)
2513 goto out_dput;
dae6ad8f 2514 error = security_path_mknod(&path, dentry, mode, dev);
be6d3e56
KT		 2515 if (error)
2516 goto out_drop_write;
463c3197 2517 switch (mode & S_IFMT) {
1da177e4 2518 case 0: case S_IFREG:
dae6ad8f 2519 error = vfs_create(path.dentry->d_inode,dentry,mode,NULL);
1da177e4
LT		 2520 break;
2521 case S_IFCHR: case S_IFBLK:
dae6ad8f 2522 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
1da177e4
LT		 2523 new_decode_dev(dev));
2524 break;
2525 case S_IFIFO: case S_IFSOCK:
dae6ad8f 2526 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
1da177e4 2527 break;
1da177e4 2528 }
be6d3e56 2529out_drop_write:
dae6ad8f 2530 mnt_drop_write(path.mnt);
463c3197
DH		 2531out_dput:
2532 dput(dentry);
dae6ad8f
AV		 2533 mutex_unlock(&path.dentry->d_inode->i_mutex);
2534 path_put(&path);
1da177e4
LT		 2535
2536 return error;
2537}
2538
8208a22b 2539SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
5590ff0d
UD		 2540{
2541 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2542}
2543
18bb1db3 2544int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1da177e4 2545{
a95164d9 2546 int error = may_create(dir, dentry);
1da177e4
LT		 2547
2548 if (error)
2549 return error;
2550
acfa4380 2551 if (!dir->i_op->mkdir)
1da177e4
LT		 2552 return -EPERM;
2553
2554 mode &= (S_IRWXUGO|S_ISVTX);
2555 error = security_inode_mkdir(dir, dentry, mode);
2556 if (error)
2557 return error;
2558
1da177e4 2559 error = dir->i_op->mkdir(dir, dentry, mode);
a74574aa 2560 if (!error)
f38aa942 2561 fsnotify_mkdir(dir, dentry);
1da177e4
LT		 2562 return error;
2563}
2564
2e4d0924 2565SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
1da177e4 2566{
6902d925 2567 struct dentry *dentry;
dae6ad8f
AV		 2568 struct path path;
2569 int error;
1da177e4 2570
dae6ad8f 2571 dentry = user_path_create(dfd, pathname, &path, 1);
6902d925 2572 if (IS_ERR(dentry))
dae6ad8f 2573 return PTR_ERR(dentry);
1da177e4 2574
dae6ad8f 2575 if (!IS_POSIXACL(path.dentry->d_inode))
ce3b0f8d 2576 mode &= ~current_umask();
dae6ad8f 2577 error = mnt_want_write(path.mnt);
463c3197
DH		 2578 if (error)
2579 goto out_dput;
dae6ad8f 2580 error = security_path_mkdir(&path, dentry, mode);
be6d3e56
KT		 2581 if (error)
2582 goto out_drop_write;
dae6ad8f 2583 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
be6d3e56 2584out_drop_write:
dae6ad8f 2585 mnt_drop_write(path.mnt);
463c3197 2586out_dput:
6902d925 2587 dput(dentry);
dae6ad8f
AV		 2588 mutex_unlock(&path.dentry->d_inode->i_mutex);
2589 path_put(&path);
1da177e4
LT		 2590 return error;
2591}
2592
3cdad428 2593SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
5590ff0d
UD		 2594{
2595 return sys_mkdirat(AT_FDCWD, pathname, mode);
2596}
2597
1da177e4 2598/*
a71905f0
SW		 2599 * The dentry_unhash() helper will try to drop the dentry early: we
2600 * should have a usage count of 2 if we're the only user of this
2601 * dentry, and if that is true (possibly after pruning the dcache),
2602 * then we drop the dentry now.
1da177e4
LT		 2603 *
2604 * A low-level filesystem can, if it choses, legally
2605 * do a
2606 *
2607 * if (!d_unhashed(dentry))
2608 * return -EBUSY;
2609 *
2610 * if it cannot handle the case of removing a directory
2611 * that is still in use by something else..
2612 */
2613void dentry_unhash(struct dentry *dentry)
2614{
dc168427 2615 shrink_dcache_parent(dentry);
1da177e4 2616 spin_lock(&dentry->d_lock);
64252c75 2617 if (dentry->d_count == 1)
1da177e4
LT		 2618 __d_drop(dentry);
2619 spin_unlock(&dentry->d_lock);
1da177e4
LT		 2620}
2621
2622int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2623{
2624 int error = may_delete(dir, dentry, 1);
2625
2626 if (error)
2627 return error;
2628
acfa4380 2629 if (!dir->i_op->rmdir)
1da177e4
LT		 2630 return -EPERM;
2631
1d2ef590 2632 dget(dentry);
1b1dcc1b 2633 mutex_lock(&dentry->d_inode->i_mutex);
912dbc15
SW		 2634
2635 error = -EBUSY;
1da177e4 2636 if (d_mountpoint(dentry))
912dbc15
SW		 2637 goto out;
2638
2639 error = security_inode_rmdir(dir, dentry);
2640 if (error)
2641 goto out;
2642
3cebde24 2643 shrink_dcache_parent(dentry);
912dbc15
SW		 2644 error = dir->i_op->rmdir(dir, dentry);
2645 if (error)
2646 goto out;
2647
2648 dentry->d_inode->i_flags |= S_DEAD;
2649 dont_mount(dentry);
2650
2651out:
1b1dcc1b 2652 mutex_unlock(&dentry->d_inode->i_mutex);
1d2ef590 2653 dput(dentry);
912dbc15 2654 if (!error)
1da177e4 2655 d_delete(dentry);
1da177e4
LT		 2656 return error;
2657}
2658
5590ff0d 2659static long do_rmdir(int dfd, const char __user *pathname)
1da177e4
LT		 2660{
2661 int error = 0;
2662 char * name;
2663 struct dentry *dentry;
2664 struct nameidata nd;
2665
2ad94ae6 2666 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2667 if (error)
2ad94ae6 2668 return error;
1da177e4
LT		 2669
2670 switch(nd.last_type) {
0612d9fb
OH		 2671 case LAST_DOTDOT:
2672 error = -ENOTEMPTY;
2673 goto exit1;
2674 case LAST_DOT:
2675 error = -EINVAL;
2676 goto exit1;
2677 case LAST_ROOT:
2678 error = -EBUSY;
2679 goto exit1;
1da177e4 2680 }
0612d9fb
OH		 2681
2682 nd.flags &= ~LOOKUP_PARENT;
2683
4ac91378 2684 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2685 dentry = lookup_hash(&nd);
1da177e4 2686 error = PTR_ERR(dentry);
6902d925
DH		 2687 if (IS_ERR(dentry))
2688 goto exit2;
e6bc45d6
TT		 2689 if (!dentry->d_inode) {
2690 error = -ENOENT;
2691 goto exit3;
2692 }
0622753b
DH		 2693 error = mnt_want_write(nd.path.mnt);
2694 if (error)
2695 goto exit3;
be6d3e56
KT		 2696 error = security_path_rmdir(&nd.path, dentry);
2697 if (error)
2698 goto exit4;
4ac91378 2699 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
be6d3e56 2700exit4:
0622753b
DH		 2701 mnt_drop_write(nd.path.mnt);
2702exit3:
6902d925
DH		 2703 dput(dentry);
2704exit2:
4ac91378 2705 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2706exit1:
1d957f9b 2707 path_put(&nd.path);
1da177e4
LT		 2708 putname(name);
2709 return error;
2710}
2711
3cdad428 2712SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
5590ff0d
UD		 2713{
2714 return do_rmdir(AT_FDCWD, pathname);
2715}
2716
1da177e4
LT		 2717int vfs_unlink(struct inode *dir, struct dentry *dentry)
2718{
2719 int error = may_delete(dir, dentry, 0);
2720
2721 if (error)
2722 return error;
2723
acfa4380 2724 if (!dir->i_op->unlink)
1da177e4
LT		 2725 return -EPERM;
2726
1b1dcc1b 2727 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2728 if (d_mountpoint(dentry))
2729 error = -EBUSY;
2730 else {
2731 error = security_inode_unlink(dir, dentry);
bec1052e 2732 if (!error) {
1da177e4 2733 error = dir->i_op->unlink(dir, dentry);
bec1052e 2734 if (!error)
d83c49f3 2735 dont_mount(dentry);
bec1052e 2736 }
1da177e4 2737 }
1b1dcc1b 2738 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2739
2740 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2741 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
ece95912 2742 fsnotify_link_count(dentry->d_inode);
e234f35c 2743 d_delete(dentry);
1da177e4 2744 }
0eeca283 2745
1da177e4
LT		 2746 return error;
2747}
2748
2749/*
2750 * Make sure that the actual truncation of the file will occur outside its
1b1dcc1b 2751 * directory's i_mutex. Truncate can take a long time if there is a lot of
1da177e4
LT		 2752 * writeout happening, and we don't want to prevent access to the directory
2753 * while waiting on the I/O.
2754 */
5590ff0d 2755static long do_unlinkat(int dfd, const char __user *pathname)
1da177e4 2756{
2ad94ae6
AV		 2757 int error;
2758 char *name;
1da177e4
LT		 2759 struct dentry *dentry;
2760 struct nameidata nd;
2761 struct inode *inode = NULL;
2762
2ad94ae6 2763 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2764 if (error)
2ad94ae6
AV		 2765 return error;
2766
1da177e4
LT		 2767 error = -EISDIR;
2768 if (nd.last_type != LAST_NORM)
2769 goto exit1;
0612d9fb
OH		 2770
2771 nd.flags &= ~LOOKUP_PARENT;
2772
4ac91378 2773 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2774 dentry = lookup_hash(&nd);
1da177e4
LT		 2775 error = PTR_ERR(dentry);
2776 if (!IS_ERR(dentry)) {
2777 /* Why not before? Because we want correct error value */
50338b88
TE		 2778 if (nd.last.name[nd.last.len])
2779 goto slashes;
1da177e4 2780 inode = dentry->d_inode;
50338b88 2781 if (!inode)
e6bc45d6
TT		 2782 goto slashes;
2783 ihold(inode);
0622753b
DH		 2784 error = mnt_want_write(nd.path.mnt);
2785 if (error)
2786 goto exit2;
be6d3e56
KT		 2787 error = security_path_unlink(&nd.path, dentry);
2788 if (error)
2789 goto exit3;
4ac91378 2790 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
be6d3e56 2791exit3:
0622753b 2792 mnt_drop_write(nd.path.mnt);
1da177e4
LT		 2793 exit2:
2794 dput(dentry);
2795 }
4ac91378 2796 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4
LT		 2797 if (inode)
2798 iput(inode); /* truncate the inode here */
2799exit1:
1d957f9b 2800 path_put(&nd.path);
1da177e4
LT		 2801 putname(name);
2802 return error;
2803
2804slashes:
2805 error = !dentry->d_inode ? -ENOENT :
2806 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2807 goto exit2;
2808}
2809
2e4d0924 2810SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
5590ff0d
UD		 2811{
2812 if ((flag & ~AT_REMOVEDIR) != 0)
2813 return -EINVAL;
2814
2815 if (flag & AT_REMOVEDIR)
2816 return do_rmdir(dfd, pathname);
2817
2818 return do_unlinkat(dfd, pathname);
2819}
2820
3480b257 2821SYSCALL_DEFINE1(unlink, const char __user *, pathname)
5590ff0d
UD		 2822{
2823 return do_unlinkat(AT_FDCWD, pathname);
2824}
2825
db2e747b 2826int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
1da177e4 2827{
a95164d9 2828 int error = may_create(dir, dentry);
1da177e4
LT		 2829
2830 if (error)
2831 return error;
2832
acfa4380 2833 if (!dir->i_op->symlink)
1da177e4
LT		 2834 return -EPERM;
2835
2836 error = security_inode_symlink(dir, dentry, oldname);
2837 if (error)
2838 return error;
2839
1da177e4 2840 error = dir->i_op->symlink(dir, dentry, oldname);
a74574aa 2841 if (!error)
f38aa942 2842 fsnotify_create(dir, dentry);
1da177e4
LT		 2843 return error;
2844}
2845
2e4d0924
HC		 2846SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2847 int, newdfd, const char __user *, newname)
1da177e4 2848{
2ad94ae6
AV		 2849 int error;
2850 char *from;
6902d925 2851 struct dentry *dentry;
dae6ad8f 2852 struct path path;
1da177e4
LT		 2853
2854 from = getname(oldname);
2ad94ae6 2855 if (IS_ERR(from))
1da177e4 2856 return PTR_ERR(from);
1da177e4 2857
dae6ad8f 2858 dentry = user_path_create(newdfd, newname, &path, 0);
6902d925
DH		 2859 error = PTR_ERR(dentry);
2860 if (IS_ERR(dentry))
dae6ad8f 2861 goto out_putname;
6902d925 2862
dae6ad8f 2863 error = mnt_want_write(path.mnt);
75c3f29d
DH		 2864 if (error)
2865 goto out_dput;
dae6ad8f 2866 error = security_path_symlink(&path, dentry, from);
be6d3e56
KT		 2867 if (error)
2868 goto out_drop_write;
dae6ad8f 2869 error = vfs_symlink(path.dentry->d_inode, dentry, from);
be6d3e56 2870out_drop_write:
dae6ad8f 2871 mnt_drop_write(path.mnt);
75c3f29d 2872out_dput:
6902d925 2873 dput(dentry);
dae6ad8f
AV		 2874 mutex_unlock(&path.dentry->d_inode->i_mutex);
2875 path_put(&path);
6902d925 2876out_putname:
1da177e4
LT		 2877 putname(from);
2878 return error;
2879}
2880
3480b257 2881SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 2882{
2883 return sys_symlinkat(oldname, AT_FDCWD, newname);
2884}
2885
1da177e4
LT		 2886int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2887{
2888 struct inode *inode = old_dentry->d_inode;
2889 int error;
2890
2891 if (!inode)
2892 return -ENOENT;
2893
a95164d9 2894 error = may_create(dir, new_dentry);
1da177e4
LT		 2895 if (error)
2896 return error;
2897
2898 if (dir->i_sb != inode->i_sb)
2899 return -EXDEV;
2900
2901 /*
2902 * A link to an append-only or immutable file cannot be created.
2903 */
2904 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2905 return -EPERM;
acfa4380 2906 if (!dir->i_op->link)
1da177e4 2907 return -EPERM;
7e79eedb 2908 if (S_ISDIR(inode->i_mode))
1da177e4
LT		 2909 return -EPERM;
2910
2911 error = security_inode_link(old_dentry, dir, new_dentry);
2912 if (error)
2913 return error;
2914
7e79eedb 2915 mutex_lock(&inode->i_mutex);
aae8a97d
AK		 2916 /* Make sure we don't allow creating hardlink to an unlinked file */
2917 if (inode->i_nlink == 0)
2918 error = -ENOENT;
2919 else
2920 error = dir->i_op->link(old_dentry, dir, new_dentry);
7e79eedb 2921 mutex_unlock(&inode->i_mutex);
e31e14ec 2922 if (!error)
7e79eedb 2923 fsnotify_link(dir, inode, new_dentry);
1da177e4
LT		 2924 return error;
2925}
2926
2927/*
2928 * Hardlinks are often used in delicate situations. We avoid
2929 * security-related surprises by not following symlinks on the
2930 * newname. --KAB
2931 *
2932 * We don't follow them on the oldname either to be compatible
2933 * with linux 2.0, and to avoid hard-linking to directories
2934 * and other special files. --ADM
2935 */
2e4d0924
HC		 2936SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2937 int, newdfd, const char __user *, newname, int, flags)
1da177e4
LT		 2938{
2939 struct dentry *new_dentry;
dae6ad8f 2940 struct path old_path, new_path;
11a7b371 2941 int how = 0;
1da177e4 2942 int error;
1da177e4 2943
11a7b371 2944 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
c04030e1 2945 return -EINVAL;
11a7b371
AK		 2946 /*
2947 * To use null names we require CAP_DAC_READ_SEARCH
2948 * This ensures that not everyone will be able to create
2949 * handlink using the passed filedescriptor.
2950 */
2951 if (flags & AT_EMPTY_PATH) {
2952 if (!capable(CAP_DAC_READ_SEARCH))
2953 return -ENOENT;
2954 how = LOOKUP_EMPTY;
2955 }
2956
2957 if (flags & AT_SYMLINK_FOLLOW)
2958 how |= LOOKUP_FOLLOW;
c04030e1 2959
11a7b371 2960 error = user_path_at(olddfd, oldname, how, &old_path);
1da177e4 2961 if (error)
2ad94ae6
AV		 2962 return error;
2963
dae6ad8f 2964 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
1da177e4 2965 error = PTR_ERR(new_dentry);
6902d925 2966 if (IS_ERR(new_dentry))
dae6ad8f
AV		 2967 goto out;
2968
2969 error = -EXDEV;
2970 if (old_path.mnt != new_path.mnt)
2971 goto out_dput;
2972 error = mnt_want_write(new_path.mnt);
75c3f29d
DH		 2973 if (error)
2974 goto out_dput;
dae6ad8f 2975 error = security_path_link(old_path.dentry, &new_path, new_dentry);
be6d3e56
KT		 2976 if (error)
2977 goto out_drop_write;
dae6ad8f 2978 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
be6d3e56 2979out_drop_write:
dae6ad8f 2980 mnt_drop_write(new_path.mnt);
75c3f29d 2981out_dput:
6902d925 2982 dput(new_dentry);
dae6ad8f
AV		 2983 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
2984 path_put(&new_path);
1da177e4 2985out:
2d8f3038 2986 path_put(&old_path);
1da177e4
LT		 2987
2988 return error;
2989}
2990
3480b257 2991SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
5590ff0d 2992{
c04030e1 2993 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
5590ff0d
UD		 2994}
2995
1da177e4
LT		 2996/*
2997 * The worst of all namespace operations - renaming directory. "Perverted"
2998 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2999 * Problems:
3000 * a) we can get into loop creation. Check is done in is_subdir().
3001 * b) race potential - two innocent renames can create a loop together.
3002 * That's where 4.4 screws up. Current fix: serialization on
a11f3a05 3003 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
1da177e4
LT		 3004 * story.
3005 * c) we have to lock _three_ objects - parents and victim (if it exists).
1b1dcc1b 3006 * And that - after we got ->i_mutex on parents (until then we don't know
1da177e4
LT		 3007 * whether the target exists). Solution: try to be smart with locking
3008 * order for inodes. We rely on the fact that tree topology may change
a11f3a05 3009 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
1da177e4
LT		 3010 * move will be locked. Thus we can rank directories by the tree
3011 * (ancestors first) and rank all non-directories after them.
3012 * That works since everybody except rename does "lock parent, lookup,
a11f3a05 3013 * lock child" and rename is under ->s_vfs_rename_mutex.
1da177e4
LT		 3014 * HOWEVER, it relies on the assumption that any object with ->lookup()
3015 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3016 * we'd better make sure that there's no link(2) for them.
e4eaac06 3017 * d) conversion from fhandle to dentry may come in the wrong moment - when
1b1dcc1b 3018 * we are removing the target. Solution: we will have to grab ->i_mutex
1da177e4 3019 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
c41b20e7 3020 * ->i_mutex on parents, which works but leads to some truly excessive
1da177e4
LT		 3021 * locking].
3022 */
75c96f85
AB		 3023static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3024 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3025{
3026 int error = 0;
9055cba7 3027 struct inode *target = new_dentry->d_inode;
1da177e4
LT		 3028
3029 /*
3030 * If we are going to change the parent - check write permissions,
3031 * we'll need to flip '..'.
3032 */
3033 if (new_dir != old_dir) {
f419a2e3 3034 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
1da177e4
LT		 3035 if (error)
3036 return error;
3037 }
3038
3039 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3040 if (error)
3041 return error;
3042
1d2ef590 3043 dget(new_dentry);
d83c49f3 3044 if (target)
1b1dcc1b 3045 mutex_lock(&target->i_mutex);
9055cba7
SW		 3046
3047 error = -EBUSY;
3048 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3049 goto out;
3050
3cebde24
SW		 3051 if (target)
3052 shrink_dcache_parent(new_dentry);
9055cba7
SW		 3053 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3054 if (error)
3055 goto out;
3056
1da177e4 3057 if (target) {
9055cba7
SW		 3058 target->i_flags |= S_DEAD;
3059 dont_mount(new_dentry);
1da177e4 3060 }
9055cba7
SW		 3061out:
3062 if (target)
3063 mutex_unlock(&target->i_mutex);
1d2ef590 3064 dput(new_dentry);
e31e14ec 3065 if (!error)
349457cc
MF		 3066 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3067 d_move(old_dentry,new_dentry);
1da177e4
LT		 3068 return error;
3069}
3070
75c96f85
AB		 3071static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3072 struct inode *new_dir, struct dentry *new_dentry)
1da177e4 3073{
51892bbb 3074 struct inode *target = new_dentry->d_inode;
1da177e4
LT		 3075 int error;
3076
3077 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3078 if (error)
3079 return error;
3080
3081 dget(new_dentry);
1da177e4 3082 if (target)
1b1dcc1b 3083 mutex_lock(&target->i_mutex);
51892bbb
SW		 3084
3085 error = -EBUSY;
1da177e4 3086 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
51892bbb
SW		 3087 goto out;
3088
3089 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3090 if (error)
3091 goto out;
3092
3093 if (target)
3094 dont_mount(new_dentry);
3095 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3096 d_move(old_dentry, new_dentry);
3097out:
1da177e4 3098 if (target)
1b1dcc1b 3099 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3100 dput(new_dentry);
3101 return error;
3102}
3103
3104int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3105 struct inode *new_dir, struct dentry *new_dentry)
3106{
3107 int error;
3108 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
59b0df21 3109 const unsigned char *old_name;
1da177e4
LT		 3110
3111 if (old_dentry->d_inode == new_dentry->d_inode)
3112 return 0;
3113
3114 error = may_delete(old_dir, old_dentry, is_dir);
3115 if (error)
3116 return error;
3117
3118 if (!new_dentry->d_inode)
a95164d9 3119 error = may_create(new_dir, new_dentry);
1da177e4
LT		 3120 else
3121 error = may_delete(new_dir, new_dentry, is_dir);
3122 if (error)
3123 return error;
3124
acfa4380 3125 if (!old_dir->i_op->rename)
1da177e4
LT		 3126 return -EPERM;
3127
0eeca283
RL		 3128 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3129
1da177e4
LT		 3130 if (is_dir)
3131 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3132 else
3133 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
123df294
AV		 3134 if (!error)
3135 fsnotify_move(old_dir, new_dir, old_name, is_dir,
5a190ae6 3136 new_dentry->d_inode, old_dentry);
0eeca283
RL		 3137 fsnotify_oldname_free(old_name);
3138
1da177e4
LT		 3139 return error;
3140}
3141
2e4d0924
HC		 3142SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3143 int, newdfd, const char __user *, newname)
1da177e4 3144{
2ad94ae6
AV		 3145 struct dentry *old_dir, *new_dir;
3146 struct dentry *old_dentry, *new_dentry;
3147 struct dentry *trap;
1da177e4 3148 struct nameidata oldnd, newnd;
2ad94ae6
AV		 3149 char *from;
3150 char *to;
3151 int error;
1da177e4 3152
2ad94ae6 3153 error = user_path_parent(olddfd, oldname, &oldnd, &from);
1da177e4
LT		 3154 if (error)
3155 goto exit;
3156
2ad94ae6 3157 error = user_path_parent(newdfd, newname, &newnd, &to);
1da177e4
LT		 3158 if (error)
3159 goto exit1;
3160
3161 error = -EXDEV;
4ac91378 3162 if (oldnd.path.mnt != newnd.path.mnt)
1da177e4
LT		 3163 goto exit2;
3164
4ac91378 3165 old_dir = oldnd.path.dentry;
1da177e4
LT		 3166 error = -EBUSY;
3167 if (oldnd.last_type != LAST_NORM)
3168 goto exit2;
3169
4ac91378 3170 new_dir = newnd.path.dentry;
1da177e4
LT		 3171 if (newnd.last_type != LAST_NORM)
3172 goto exit2;
3173
0612d9fb
OH		 3174 oldnd.flags &= ~LOOKUP_PARENT;
3175 newnd.flags &= ~LOOKUP_PARENT;
4e9ed2f8 3176 newnd.flags |= LOOKUP_RENAME_TARGET;
0612d9fb 3177
1da177e4
LT		 3178 trap = lock_rename(new_dir, old_dir);
3179
49705b77 3180 old_dentry = lookup_hash(&oldnd);
1da177e4
LT		 3181 error = PTR_ERR(old_dentry);
3182 if (IS_ERR(old_dentry))
3183 goto exit3;
3184 /* source must exist */
3185 error = -ENOENT;
3186 if (!old_dentry->d_inode)
3187 goto exit4;
3188 /* unless the source is a directory trailing slashes give -ENOTDIR */
3189 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3190 error = -ENOTDIR;
3191 if (oldnd.last.name[oldnd.last.len])
3192 goto exit4;
3193 if (newnd.last.name[newnd.last.len])
3194 goto exit4;
3195 }
3196 /* source should not be ancestor of target */
3197 error = -EINVAL;
3198 if (old_dentry == trap)
3199 goto exit4;
49705b77 3200 new_dentry = lookup_hash(&newnd);
1da177e4
LT		 3201 error = PTR_ERR(new_dentry);
3202 if (IS_ERR(new_dentry))
3203 goto exit4;
3204 /* target should not be an ancestor of source */
3205 error = -ENOTEMPTY;
3206 if (new_dentry == trap)
3207 goto exit5;
3208
9079b1eb
DH		 3209 error = mnt_want_write(oldnd.path.mnt);
3210 if (error)
3211 goto exit5;
be6d3e56
KT		 3212 error = security_path_rename(&oldnd.path, old_dentry,
3213 &newnd.path, new_dentry);
3214 if (error)
3215 goto exit6;
1da177e4
LT		 3216 error = vfs_rename(old_dir->d_inode, old_dentry,
3217 new_dir->d_inode, new_dentry);
be6d3e56 3218exit6:
9079b1eb 3219 mnt_drop_write(oldnd.path.mnt);
1da177e4
LT		 3220exit5:
3221 dput(new_dentry);
3222exit4:
3223 dput(old_dentry);
3224exit3:
3225 unlock_rename(new_dir, old_dir);
3226exit2:
1d957f9b 3227 path_put(&newnd.path);
2ad94ae6 3228 putname(to);
1da177e4 3229exit1:
1d957f9b 3230 path_put(&oldnd.path);
1da177e4 3231 putname(from);
2ad94ae6 3232exit:
1da177e4
LT		 3233 return error;
3234}
3235
a26eab24 3236SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 3237{
3238 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3239}
3240
1da177e4
LT		 3241int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3242{
3243 int len;
3244
3245 len = PTR_ERR(link);
3246 if (IS_ERR(link))
3247 goto out;
3248
3249 len = strlen(link);
3250 if (len > (unsigned) buflen)
3251 len = buflen;
3252 if (copy_to_user(buffer, link, len))
3253 len = -EFAULT;
3254out:
3255 return len;
3256}
3257
3258/*
3259 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3260 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3261 * using) it for any given inode is up to filesystem.
3262 */
3263int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3264{
3265 struct nameidata nd;
cc314eef 3266 void *cookie;
694a1764 3267 int res;
cc314eef 3268
1da177e4 3269 nd.depth = 0;
cc314eef 3270 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
694a1764
MS		 3271 if (IS_ERR(cookie))
3272 return PTR_ERR(cookie);
3273
3274 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3275 if (dentry->d_inode->i_op->put_link)
3276 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3277 return res;
1da177e4
LT		 3278}
3279
3280int vfs_follow_link(struct nameidata *nd, const char *link)
3281{
3282 return __vfs_follow_link(nd, link);
3283}
3284
3285/* get the link contents into pagecache */
3286static char *page_getlink(struct dentry * dentry, struct page **ppage)
3287{
ebd09abb
DG		 3288 char *kaddr;
3289 struct page *page;
1da177e4 3290 struct address_space *mapping = dentry->d_inode->i_mapping;
090d2b18 3291 page = read_mapping_page(mapping, 0, NULL);
1da177e4 3292 if (IS_ERR(page))
6fe6900e 3293 return (char*)page;
1da177e4 3294 *ppage = page;
ebd09abb
DG		 3295 kaddr = kmap(page);
3296 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3297 return kaddr;
1da177e4
LT		 3298}
3299
3300int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3301{
3302 struct page *page = NULL;
3303 char *s = page_getlink(dentry, &page);
3304 int res = vfs_readlink(dentry,buffer,buflen,s);
3305 if (page) {
3306 kunmap(page);
3307 page_cache_release(page);
3308 }
3309 return res;
3310}
3311
cc314eef 3312void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
1da177e4 3313{
cc314eef 3314 struct page *page = NULL;
1da177e4 3315 nd_set_link(nd, page_getlink(dentry, &page));
cc314eef 3316 return page;
1da177e4
LT		 3317}
3318
cc314eef 3319void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1da177e4 3320{
cc314eef
LT		 3321 struct page *page = cookie;
3322
3323 if (page) {
1da177e4
LT		 3324 kunmap(page);
3325 page_cache_release(page);
1da177e4
LT		 3326 }
3327}
3328
54566b2c
NP		 3329/*
3330 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3331 */
3332int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
1da177e4
LT		 3333{
3334 struct address_space *mapping = inode->i_mapping;
0adb25d2 3335 struct page *page;
afddba49 3336 void *fsdata;
beb497ab 3337 int err;
1da177e4 3338 char *kaddr;
54566b2c
NP		 3339 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3340 if (nofs)
3341 flags |= AOP_FLAG_NOFS;
1da177e4 3342
7e53cac4 3343retry:
afddba49 3344 err = pagecache_write_begin(NULL, mapping, 0, len-1,
54566b2c 3345 flags, &page, &fsdata);
1da177e4 3346 if (err)
afddba49
NP		 3347 goto fail;
3348
1da177e4
LT		 3349 kaddr = kmap_atomic(page, KM_USER0);
3350 memcpy(kaddr, symname, len-1);
3351 kunmap_atomic(kaddr, KM_USER0);
afddba49
NP		 3352
3353 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3354 page, fsdata);
1da177e4
LT		 3355 if (err < 0)
3356 goto fail;
afddba49
NP		 3357 if (err < len-1)
3358 goto retry;
3359
1da177e4
LT		 3360 mark_inode_dirty(inode);
3361 return 0;
1da177e4
LT		 3362fail:
3363 return err;
3364}
3365
0adb25d2
KK		 3366int page_symlink(struct inode *inode, const char *symname, int len)
3367{
3368 return __page_symlink(inode, symname, len,
54566b2c 3369 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
0adb25d2
KK		 3370}
3371
92e1d5be 3372const struct inode_operations page_symlink_inode_operations = {
1da177e4
LT		 3373 .readlink = generic_readlink,
3374 .follow_link = page_follow_link_light,
3375 .put_link = page_put_link,
3376};
3377
2d8f3038 3378EXPORT_SYMBOL(user_path_at);
cc53ce53 3379EXPORT_SYMBOL(follow_down_one);
1da177e4
LT		 3380EXPORT_SYMBOL(follow_down);
3381EXPORT_SYMBOL(follow_up);
3382EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3383EXPORT_SYMBOL(getname);
3384EXPORT_SYMBOL(lock_rename);
1da177e4
LT		 3385EXPORT_SYMBOL(lookup_one_len);
3386EXPORT_SYMBOL(page_follow_link_light);
3387EXPORT_SYMBOL(page_put_link);
3388EXPORT_SYMBOL(page_readlink);
0adb25d2 3389EXPORT_SYMBOL(__page_symlink);
1da177e4
LT		 3390EXPORT_SYMBOL(page_symlink);
3391EXPORT_SYMBOL(page_symlink_inode_operations);
d1811465 3392EXPORT_SYMBOL(kern_path);
16f18200 3393EXPORT_SYMBOL(vfs_path_lookup);
f419a2e3 3394EXPORT_SYMBOL(inode_permission);
1da177e4
LT		 3395EXPORT_SYMBOL(unlock_rename);
3396EXPORT_SYMBOL(vfs_create);
3397EXPORT_SYMBOL(vfs_follow_link);
3398EXPORT_SYMBOL(vfs_link);
3399EXPORT_SYMBOL(vfs_mkdir);
3400EXPORT_SYMBOL(vfs_mknod);
3401EXPORT_SYMBOL(generic_permission);
3402EXPORT_SYMBOL(vfs_readlink);
3403EXPORT_SYMBOL(vfs_rename);
3404EXPORT_SYMBOL(vfs_rmdir);
3405EXPORT_SYMBOL(vfs_symlink);
3406EXPORT_SYMBOL(vfs_unlink);
3407EXPORT_SYMBOL(dentry_unhash);
3408EXPORT_SYMBOL(generic_readlink);
kernel source for telekom puls (alcatel/mediatek)