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