projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
includecheck: delete any duplicate instances of module.h
[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 38#include "internal.h"
c7105365 39#include "mount.h"
e81e3f4d 40
1da177e4
LT		 41/* [Feb-1997 T. Schoebel-Theuer]
42 * Fundamental changes in the pathname lookup mechanisms (namei)
43 * were necessary because of omirr. The reason is that omirr needs
44 * to know the _real_ pathname, not the user-supplied one, in case
45 * of symlinks (and also when transname replacements occur).
46 *
47 * The new code replaces the old recursive symlink resolution with
48 * an iterative one (in case of non-nested symlink chains). It does
49 * this with calls to <fs>_follow_link().
50 * As a side effect, dir_namei(), _namei() and follow_link() are now
51 * replaced with a single function lookup_dentry() that can handle all
52 * the special cases of the former code.
53 *
54 * With the new dcache, the pathname is stored at each inode, at least as
55 * long as the refcount of the inode is positive. As a side effect, the
56 * size of the dcache depends on the inode cache and thus is dynamic.
57 *
58 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
59 * resolution to correspond with current state of the code.
60 *
61 * Note that the symlink resolution is not *completely* iterative.
62 * There is still a significant amount of tail- and mid- recursion in
63 * the algorithm. Also, note that <fs>_readlink() is not used in
64 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
65 * may return different results than <fs>_follow_link(). Many virtual
66 * filesystems (including /proc) exhibit this behavior.
67 */
68
69/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
70 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
71 * and the name already exists in form of a symlink, try to create the new
72 * name indicated by the symlink. The old code always complained that the
73 * name already exists, due to not following the symlink even if its target
74 * is nonexistent. The new semantics affects also mknod() and link() when
25985edc 75 * the name is a symlink pointing to a non-existent name.
1da177e4
LT		 76 *
77 * I don't know which semantics is the right one, since I have no access
78 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
79 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
80 * "old" one. Personally, I think the new semantics is much more logical.
81 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
82 * file does succeed in both HP-UX and SunOs, but not in Solaris
83 * and in the old Linux semantics.
84 */
85
86/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
87 * semantics. See the comments in "open_namei" and "do_link" below.
88 *
89 * [10-Sep-98 Alan Modra] Another symlink change.
90 */
91
92/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
93 * inside the path - always follow.
94 * in the last component in creation/removal/renaming - never follow.
95 * if LOOKUP_FOLLOW passed - follow.
96 * if the pathname has trailing slashes - follow.
97 * otherwise - don't follow.
98 * (applied in that order).
99 *
100 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
101 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
102 * During the 2.4 we need to fix the userland stuff depending on it -
103 * hopefully we will be able to get rid of that wart in 2.5. So far only
104 * XEmacs seems to be relying on it...
105 */
106/*
107 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
a11f3a05 108 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
1da177e4
LT		 109 * any extra contention...
110 */
111
112/* In order to reduce some races, while at the same time doing additional
113 * checking and hopefully speeding things up, we copy filenames to the
114 * kernel data space before using them..
115 *
116 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
117 * PATH_MAX includes the nul terminator --RR.
118 */
858119e1 119static int do_getname(const char __user *filename, char *page)
1da177e4
LT		 120{
121 int retval;
122 unsigned long len = PATH_MAX;
123
124 if (!segment_eq(get_fs(), KERNEL_DS)) {
125 if ((unsigned long) filename >= TASK_SIZE)
126 return -EFAULT;
127 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
128 len = TASK_SIZE - (unsigned long) filename;
129 }
130
131 retval = strncpy_from_user(page, filename, len);
132 if (retval > 0) {
133 if (retval < len)
134 return 0;
135 return -ENAMETOOLONG;
136 } else if (!retval)
137 retval = -ENOENT;
138 return retval;
139}
140
1fa1e7f6 141static char *getname_flags(const char __user *filename, int flags, int *empty)
1da177e4 142{
4043cde8
EP		 143 char *result = __getname();
144 int retval;
145
146 if (!result)
147 return ERR_PTR(-ENOMEM);
148
149 retval = do_getname(filename, result);
150 if (retval < 0) {
151 if (retval == -ENOENT && empty)
152 *empty = 1;
153 if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) {
154 __putname(result);
155 return ERR_PTR(retval);
1da177e4
LT		 156 }
157 }
158 audit_getname(result);
159 return result;
160}
161
f52e0c11
AV		 162char *getname(const char __user * filename)
163{
1fa1e7f6 164 return getname_flags(filename, 0, 0);
f52e0c11
AV		 165}
166
1da177e4
LT		 167#ifdef CONFIG_AUDITSYSCALL
168void putname(const char *name)
169{
5ac3a9c2 170 if (unlikely(!audit_dummy_context()))
1da177e4
LT		 171 audit_putname(name);
172 else
173 __putname(name);
174}
175EXPORT_SYMBOL(putname);
176#endif
177
e77819e5
LT		 178static int check_acl(struct inode *inode, int mask)
179{
84635d68 180#ifdef CONFIG_FS_POSIX_ACL
e77819e5
LT		 181 struct posix_acl *acl;
182
e77819e5 183 if (mask & MAY_NOT_BLOCK) {
3567866b
AV		 184 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
185 if (!acl)
e77819e5 186 return -EAGAIN;
3567866b
AV		 187 /* no ->get_acl() calls in RCU mode... */
188 if (acl == ACL_NOT_CACHED)
189 return -ECHILD;
206b1d09 190 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
e77819e5
LT		 191 }
192
193 acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
194
195 /*
4e34e719
CH		 196 * A filesystem can force a ACL callback by just never filling the
197 * ACL cache. But normally you'd fill the cache either at inode
198 * instantiation time, or on the first ->get_acl call.
e77819e5 199 *
4e34e719
CH		 200 * If the filesystem doesn't have a get_acl() function at all, we'll
201 * just create the negative cache entry.
e77819e5
LT		 202 */
203 if (acl == ACL_NOT_CACHED) {
4e34e719
CH		 204 if (inode->i_op->get_acl) {
205 acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS);
206 if (IS_ERR(acl))
207 return PTR_ERR(acl);
208 } else {
209 set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
210 return -EAGAIN;
211 }
e77819e5
LT		 212 }
213
214 if (acl) {
215 int error = posix_acl_permission(inode, acl, mask);
216 posix_acl_release(acl);
217 return error;
218 }
84635d68 219#endif
e77819e5
LT		 220
221 return -EAGAIN;
222}
223
5909ccaa 224/*
948409c7 225 * This does the basic permission checking
1da177e4 226 */
7e40145e 227static int acl_permission_check(struct inode *inode, int mask)
1da177e4 228{
26cf46be 229 unsigned int mode = inode->i_mode;
1da177e4 230
e795b717
SH		 231 if (current_user_ns() != inode_userns(inode))
232 goto other_perms;
233
14067ff5 234 if (likely(current_fsuid() == inode->i_uid))
1da177e4
LT		 235 mode >>= 6;
236 else {
e77819e5 237 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
7e40145e 238 int error = check_acl(inode, mask);
b74c79e9
NP		 239 if (error != -EAGAIN)
240 return error;
1da177e4
LT		 241 }
242
243 if (in_group_p(inode->i_gid))
244 mode >>= 3;
245 }
246
e795b717 247other_perms:
1da177e4
LT		 248 /*
249 * If the DACs are ok we don't need any capability check.
250 */
9c2c7039 251 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
1da177e4 252 return 0;
5909ccaa
LT		 253 return -EACCES;
254}
255
256/**
b74c79e9 257 * generic_permission - check for access rights on a Posix-like filesystem
5909ccaa 258 * @inode: inode to check access rights for
8fd90c8d 259 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
5909ccaa
LT		 260 *
261 * Used to check for read/write/execute permissions on a file.
262 * We use "fsuid" for this, letting us set arbitrary permissions
263 * for filesystem access without changing the "normal" uids which
b74c79e9
NP		 264 * are used for other things.
265 *
266 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
267 * request cannot be satisfied (eg. requires blocking or too much complexity).
268 * It would then be called again in ref-walk mode.
5909ccaa 269 */
2830ba7f 270int generic_permission(struct inode *inode, int mask)
5909ccaa
LT		 271{
272 int ret;
273
274 /*
948409c7 275 * Do the basic permission checks.
5909ccaa 276 */
7e40145e 277 ret = acl_permission_check(inode, mask);
5909ccaa
LT		 278 if (ret != -EACCES)
279 return ret;
1da177e4 280
d594e7ec
AV		 281 if (S_ISDIR(inode->i_mode)) {
282 /* DACs are overridable for directories */
283 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
284 return 0;
285 if (!(mask & MAY_WRITE))
286 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
287 return 0;
288 return -EACCES;
289 }
1da177e4
LT		 290 /*
291 * Read/write DACs are always overridable.
d594e7ec
AV		 292 * Executable DACs are overridable when there is
293 * at least one exec bit set.
1da177e4 294 */
d594e7ec 295 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
e795b717 296 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
1da177e4
LT		 297 return 0;
298
299 /*
300 * Searching includes executable on directories, else just read.
301 */
7ea66001 302 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
d594e7ec 303 if (mask == MAY_READ)
e795b717 304 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
1da177e4
LT		 305 return 0;
306
307 return -EACCES;
308}
309
3ddcd056
LT		 310/*
311 * We _really_ want to just do "generic_permission()" without
312 * even looking at the inode->i_op values. So we keep a cache
313 * flag in inode->i_opflags, that says "this has not special
314 * permission function, use the fast case".
315 */
316static inline int do_inode_permission(struct inode *inode, int mask)
317{
318 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
319 if (likely(inode->i_op->permission))
320 return inode->i_op->permission(inode, mask);
321
322 /* This gets set once for the inode lifetime */
323 spin_lock(&inode->i_lock);
324 inode->i_opflags |= IOP_FASTPERM;
325 spin_unlock(&inode->i_lock);
326 }
327 return generic_permission(inode, mask);
328}
329
cb23beb5
CH		 330/**
331 * inode_permission - check for access rights to a given inode
332 * @inode: inode to check permission on
8fd90c8d 333 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
cb23beb5
CH		 334 *
335 * Used to check for read/write/execute permissions on an inode.
336 * We use "fsuid" for this, letting us set arbitrary permissions
337 * for filesystem access without changing the "normal" uids which
338 * are used for other things.
948409c7
AG		 339 *
340 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
cb23beb5 341 */
f419a2e3 342int inode_permission(struct inode *inode, int mask)
1da177e4 343{
e6305c43 344 int retval;
1da177e4 345
3ddcd056 346 if (unlikely(mask & MAY_WRITE)) {
22590e41 347 umode_t mode = inode->i_mode;
1da177e4
LT		 348
349 /*
350 * Nobody gets write access to a read-only fs.
351 */
352 if (IS_RDONLY(inode) &&
353 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
354 return -EROFS;
355
356 /*
357 * Nobody gets write access to an immutable file.
358 */
359 if (IS_IMMUTABLE(inode))
360 return -EACCES;
361 }
362
3ddcd056 363 retval = do_inode_permission(inode, mask);
1da177e4
LT		 364 if (retval)
365 return retval;
366
08ce5f16
SH		 367 retval = devcgroup_inode_permission(inode, mask);
368 if (retval)
369 return retval;
370
d09ca739 371 return security_inode_permission(inode, mask);
1da177e4
LT		 372}
373
5dd784d0
JB		 374/**
375 * path_get - get a reference to a path
376 * @path: path to get the reference to
377 *
378 * Given a path increment the reference count to the dentry and the vfsmount.
379 */
380void path_get(struct path *path)
381{
382 mntget(path->mnt);
383 dget(path->dentry);
384}
385EXPORT_SYMBOL(path_get);
386
1d957f9b
JB		 387/**
388 * path_put - put a reference to a path
389 * @path: path to put the reference to
390 *
391 * Given a path decrement the reference count to the dentry and the vfsmount.
392 */
393void path_put(struct path *path)
1da177e4 394{
1d957f9b
JB		 395 dput(path->dentry);
396 mntput(path->mnt);
1da177e4 397}
1d957f9b 398EXPORT_SYMBOL(path_put);
1da177e4 399
19660af7 400/*
31e6b01f 401 * Path walking has 2 modes, rcu-walk and ref-walk (see
19660af7
AV		 402 * Documentation/filesystems/path-lookup.txt). In situations when we can't
403 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
404 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
405 * mode. Refcounts are grabbed at the last known good point before rcu-walk
406 * got stuck, so ref-walk may continue from there. If this is not successful
407 * (eg. a seqcount has changed), then failure is returned and it's up to caller
408 * to restart the path walk from the beginning in ref-walk mode.
31e6b01f 409 */
31e6b01f
NP		 410
411/**
19660af7
AV		 412 * unlazy_walk - try to switch to ref-walk mode.
413 * @nd: nameidata pathwalk data
414 * @dentry: child of nd->path.dentry or NULL
39191628 415 * Returns: 0 on success, -ECHILD on failure
31e6b01f 416 *
19660af7
AV		 417 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
418 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
419 * @nd or NULL. Must be called from rcu-walk context.
31e6b01f 420 */
19660af7 421static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
31e6b01f
NP		 422{
423 struct fs_struct *fs = current->fs;
424 struct dentry *parent = nd->path.dentry;
5b6ca027 425 int want_root = 0;
31e6b01f
NP		 426
427 BUG_ON(!(nd->flags & LOOKUP_RCU));
5b6ca027
AV		 428 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
429 want_root = 1;
31e6b01f
NP		 430 spin_lock(&fs->lock);
431 if (nd->root.mnt != fs->root.mnt ||
432 nd->root.dentry != fs->root.dentry)
433 goto err_root;
434 }
435 spin_lock(&parent->d_lock);
19660af7
AV		 436 if (!dentry) {
437 if (!__d_rcu_to_refcount(parent, nd->seq))
438 goto err_parent;
439 BUG_ON(nd->inode != parent->d_inode);
440 } else {
94c0d4ec
AV		 441 if (dentry->d_parent != parent)
442 goto err_parent;
19660af7
AV		 443 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
444 if (!__d_rcu_to_refcount(dentry, nd->seq))
445 goto err_child;
446 /*
447 * If the sequence check on the child dentry passed, then
448 * the child has not been removed from its parent. This
449 * means the parent dentry must be valid and able to take
450 * a reference at this point.
451 */
452 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
453 BUG_ON(!parent->d_count);
454 parent->d_count++;
455 spin_unlock(&dentry->d_lock);
456 }
31e6b01f 457 spin_unlock(&parent->d_lock);
5b6ca027 458 if (want_root) {
31e6b01f
NP		 459 path_get(&nd->root);
460 spin_unlock(&fs->lock);
461 }
462 mntget(nd->path.mnt);
463
464 rcu_read_unlock();
465 br_read_unlock(vfsmount_lock);
466 nd->flags &= ~LOOKUP_RCU;
467 return 0;
19660af7
AV		 468
469err_child:
31e6b01f 470 spin_unlock(&dentry->d_lock);
19660af7 471err_parent:
31e6b01f
NP		 472 spin_unlock(&parent->d_lock);
473err_root:
5b6ca027 474 if (want_root)
31e6b01f
NP		 475 spin_unlock(&fs->lock);
476 return -ECHILD;
477}
478
834f2a4a
TM		 479/**
480 * release_open_intent - free up open intent resources
481 * @nd: pointer to nameidata
482 */
483void release_open_intent(struct nameidata *nd)
484{
2dab5974
LT		 485 struct file *file = nd->intent.open.file;
486
487 if (file && !IS_ERR(file)) {
488 if (file->f_path.dentry == NULL)
489 put_filp(file);
490 else
491 fput(file);
492 }
834f2a4a
TM		 493}
494
f60aef7e 495static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
34286d66 496{
f60aef7e 497 return dentry->d_op->d_revalidate(dentry, nd);
34286d66
NP		 498}
499
9f1fafee
AV		 500/**
501 * complete_walk - successful completion of path walk
502 * @nd: pointer nameidata
39159de2 503 *
9f1fafee
AV		 504 * If we had been in RCU mode, drop out of it and legitimize nd->path.
505 * Revalidate the final result, unless we'd already done that during
506 * the path walk or the filesystem doesn't ask for it. Return 0 on
507 * success, -error on failure. In case of failure caller does not
508 * need to drop nd->path.
39159de2 509 */
9f1fafee 510static int complete_walk(struct nameidata *nd)
39159de2 511{
16c2cd71 512 struct dentry *dentry = nd->path.dentry;
39159de2 513 int status;
39159de2 514
9f1fafee
AV		 515 if (nd->flags & LOOKUP_RCU) {
516 nd->flags &= ~LOOKUP_RCU;
517 if (!(nd->flags & LOOKUP_ROOT))
518 nd->root.mnt = NULL;
519 spin_lock(&dentry->d_lock);
520 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
521 spin_unlock(&dentry->d_lock);
522 rcu_read_unlock();
523 br_read_unlock(vfsmount_lock);
524 return -ECHILD;
525 }
526 BUG_ON(nd->inode != dentry->d_inode);
527 spin_unlock(&dentry->d_lock);
528 mntget(nd->path.mnt);
529 rcu_read_unlock();
530 br_read_unlock(vfsmount_lock);
531 }
532
16c2cd71
AV		 533 if (likely(!(nd->flags & LOOKUP_JUMPED)))
534 return 0;
535
536 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
39159de2
JL		 537 return 0;
538
16c2cd71
AV		 539 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
540 return 0;
541
542 /* Note: we do not d_invalidate() */
34286d66 543 status = d_revalidate(dentry, nd);
39159de2
JL		 544 if (status > 0)
545 return 0;
546
16c2cd71 547 if (!status)
39159de2 548 status = -ESTALE;
16c2cd71 549
9f1fafee 550 path_put(&nd->path);
39159de2
JL		 551 return status;
552}
553
2a737871
AV		 554static __always_inline void set_root(struct nameidata *nd)
555{
f7ad3c6b
MS		 556 if (!nd->root.mnt)
557 get_fs_root(current->fs, &nd->root);
2a737871
AV		 558}
559
6de88d72
AV		 560static int link_path_walk(const char *, struct nameidata *);
561
31e6b01f
NP		 562static __always_inline void set_root_rcu(struct nameidata *nd)
563{
564 if (!nd->root.mnt) {
565 struct fs_struct *fs = current->fs;
c28cc364
NP		 566 unsigned seq;
567
568 do {
569 seq = read_seqcount_begin(&fs->seq);
570 nd->root = fs->root;
c1530019 571 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
c28cc364 572 } while (read_seqcount_retry(&fs->seq, seq));
31e6b01f
NP		 573 }
574}
575
f1662356 576static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
1da177e4 577{
31e6b01f
NP		 578 int ret;
579
1da177e4
LT		 580 if (IS_ERR(link))
581 goto fail;
582
583 if (*link == '/') {
2a737871 584 set_root(nd);
1d957f9b 585 path_put(&nd->path);
2a737871
AV		 586 nd->path = nd->root;
587 path_get(&nd->root);
16c2cd71 588 nd->flags |= LOOKUP_JUMPED;
1da177e4 589 }
31e6b01f 590 nd->inode = nd->path.dentry->d_inode;
b4091d5f 591
31e6b01f
NP		 592 ret = link_path_walk(link, nd);
593 return ret;
1da177e4 594fail:
1d957f9b 595 path_put(&nd->path);
1da177e4
LT		 596 return PTR_ERR(link);
597}
598
1d957f9b 599static void path_put_conditional(struct path *path, struct nameidata *nd)
051d3812
IK		 600{
601 dput(path->dentry);
4ac91378 602 if (path->mnt != nd->path.mnt)
051d3812
IK		 603 mntput(path->mnt);
604}
605
7b9337aa
NP		 606static inline void path_to_nameidata(const struct path *path,
607 struct nameidata *nd)
051d3812 608{
31e6b01f
NP		 609 if (!(nd->flags & LOOKUP_RCU)) {
610 dput(nd->path.dentry);
611 if (nd->path.mnt != path->mnt)
612 mntput(nd->path.mnt);
9a229683 613 }
31e6b01f 614 nd->path.mnt = path->mnt;
4ac91378 615 nd->path.dentry = path->dentry;
051d3812
IK		 616}
617
574197e0
AV		 618static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
619{
620 struct inode *inode = link->dentry->d_inode;
621 if (!IS_ERR(cookie) && inode->i_op->put_link)
622 inode->i_op->put_link(link->dentry, nd, cookie);
623 path_put(link);
624}
625
def4af30 626static __always_inline int
574197e0 627follow_link(struct path *link, struct nameidata *nd, void **p)
1da177e4
LT		 628{
629 int error;
7b9337aa 630 struct dentry *dentry = link->dentry;
1da177e4 631
844a3917
AV		 632 BUG_ON(nd->flags & LOOKUP_RCU);
633
0e794589
AV		 634 if (link->mnt == nd->path.mnt)
635 mntget(link->mnt);
636
574197e0
AV		 637 if (unlikely(current->total_link_count >= 40)) {
638 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
574197e0
AV		 639 path_put(&nd->path);
640 return -ELOOP;
641 }
642 cond_resched();
643 current->total_link_count++;
644
7b9337aa 645 touch_atime(link->mnt, dentry);
1da177e4 646 nd_set_link(nd, NULL);
cd4e91d3 647
36f3b4f6
AV		 648 error = security_inode_follow_link(link->dentry, nd);
649 if (error) {
650 *p = ERR_PTR(error); /* no ->put_link(), please */
651 path_put(&nd->path);
652 return error;
653 }
654
86acdca1 655 nd->last_type = LAST_BIND;
def4af30
AV		 656 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
657 error = PTR_ERR(*p);
658 if (!IS_ERR(*p)) {
1da177e4 659 char *s = nd_get_link(nd);
cc314eef 660 error = 0;
1da177e4
LT		 661 if (s)
662 error = __vfs_follow_link(nd, s);
bcda7652 663 else if (nd->last_type == LAST_BIND) {
16c2cd71 664 nd->flags |= LOOKUP_JUMPED;
b21041d0
AV		 665 nd->inode = nd->path.dentry->d_inode;
666 if (nd->inode->i_op->follow_link) {
bcda7652
AV		 667 /* stepped on a _really_ weird one */
668 path_put(&nd->path);
669 error = -ELOOP;
670 }
671 }
1da177e4 672 }
1da177e4
LT		 673 return error;
674}
675
31e6b01f
NP		 676static int follow_up_rcu(struct path *path)
677{
0714a533
AV		 678 struct mount *mnt = real_mount(path->mnt);
679 struct mount *parent;
31e6b01f
NP		 680 struct dentry *mountpoint;
681
0714a533
AV		 682 parent = mnt->mnt_parent;
683 if (&parent->mnt == path->mnt)
31e6b01f 684 return 0;
a73324da 685 mountpoint = mnt->mnt_mountpoint;
31e6b01f 686 path->dentry = mountpoint;
0714a533 687 path->mnt = &parent->mnt;
31e6b01f
NP		 688 return 1;
689}
690
bab77ebf 691int follow_up(struct path *path)
1da177e4 692{
0714a533
AV		 693 struct mount *mnt = real_mount(path->mnt);
694 struct mount *parent;
1da177e4 695 struct dentry *mountpoint;
99b7db7b
NP		 696
697 br_read_lock(vfsmount_lock);
0714a533
AV		 698 parent = mnt->mnt_parent;
699 if (&parent->mnt == path->mnt) {
99b7db7b 700 br_read_unlock(vfsmount_lock);
1da177e4
LT		 701 return 0;
702 }
0714a533 703 mntget(&parent->mnt);
a73324da 704 mountpoint = dget(mnt->mnt_mountpoint);
99b7db7b 705 br_read_unlock(vfsmount_lock);
bab77ebf
AV		 706 dput(path->dentry);
707 path->dentry = mountpoint;
708 mntput(path->mnt);
0714a533 709 path->mnt = &parent->mnt;
1da177e4
LT		 710 return 1;
711}
712
b5c84bf6 713/*
9875cf80
DH		 714 * Perform an automount
715 * - return -EISDIR to tell follow_managed() to stop and return the path we
716 * were called with.
1da177e4 717 */
9875cf80
DH		 718static int follow_automount(struct path *path, unsigned flags,
719 bool *need_mntput)
31e6b01f 720{
9875cf80 721 struct vfsmount *mnt;
ea5b778a 722 int err;
9875cf80
DH		 723
724 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
725 return -EREMOTE;
726
0ec26fd0
MS		 727 /* We don't want to mount if someone's just doing a stat -
728 * unless they're stat'ing a directory and appended a '/' to
729 * the name.
730 *
731 * We do, however, want to mount if someone wants to open or
732 * create a file of any type under the mountpoint, wants to
733 * traverse through the mountpoint or wants to open the
734 * mounted directory. Also, autofs may mark negative dentries
735 * as being automount points. These will need the attentions
736 * of the daemon to instantiate them before they can be used.
9875cf80 737 */
0ec26fd0 738 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
d94c177b 739 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
0ec26fd0
MS		 740 path->dentry->d_inode)
741 return -EISDIR;
742
9875cf80
DH		 743 current->total_link_count++;
744 if (current->total_link_count >= 40)
745 return -ELOOP;
746
747 mnt = path->dentry->d_op->d_automount(path);
748 if (IS_ERR(mnt)) {
749 /*
750 * The filesystem is allowed to return -EISDIR here to indicate
751 * it doesn't want to automount. For instance, autofs would do
752 * this so that its userspace daemon can mount on this dentry.
753 *
754 * However, we can only permit this if it's a terminal point in
755 * the path being looked up; if it wasn't then the remainder of
756 * the path is inaccessible and we should say so.
757 */
49084c3b 758 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
9875cf80
DH		 759 return -EREMOTE;
760 return PTR_ERR(mnt);
31e6b01f 761 }
ea5b778a 762
9875cf80
DH		 763 if (!mnt) /* mount collision */
764 return 0;
31e6b01f 765
8aef1884
AV		 766 if (!*need_mntput) {
767 /* lock_mount() may release path->mnt on error */
768 mntget(path->mnt);
769 *need_mntput = true;
770 }
19a167af 771 err = finish_automount(mnt, path);
9875cf80 772
ea5b778a
DH		 773 switch (err) {
774 case -EBUSY:
775 /* Someone else made a mount here whilst we were busy */
19a167af 776 return 0;
ea5b778a 777 case 0:
8aef1884 778 path_put(path);
ea5b778a
DH		 779 path->mnt = mnt;
780 path->dentry = dget(mnt->mnt_root);
ea5b778a 781 return 0;
19a167af
AV		 782 default:
783 return err;
ea5b778a 784 }
19a167af 785
463ffb2e
AV		 786}
787
9875cf80
DH		 788/*
789 * Handle a dentry that is managed in some way.
cc53ce53 790 * - Flagged for transit management (autofs)
9875cf80
DH		 791 * - Flagged as mountpoint
792 * - Flagged as automount point
793 *
794 * This may only be called in refwalk mode.
795 *
796 * Serialization is taken care of in namespace.c
797 */
798static int follow_managed(struct path *path, unsigned flags)
1da177e4 799{
8aef1884 800 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
9875cf80
DH		 801 unsigned managed;
802 bool need_mntput = false;
8aef1884 803 int ret = 0;
9875cf80
DH		 804
805 /* Given that we're not holding a lock here, we retain the value in a
806 * local variable for each dentry as we look at it so that we don't see
807 * the components of that value change under us */
808 while (managed = ACCESS_ONCE(path->dentry->d_flags),
809 managed &= DCACHE_MANAGED_DENTRY,
810 unlikely(managed != 0)) {
cc53ce53
DH		 811 /* Allow the filesystem to manage the transit without i_mutex
812 * being held. */
813 if (managed & DCACHE_MANAGE_TRANSIT) {
814 BUG_ON(!path->dentry->d_op);
815 BUG_ON(!path->dentry->d_op->d_manage);
1aed3e42 816 ret = path->dentry->d_op->d_manage(path->dentry, false);
cc53ce53 817 if (ret < 0)
8aef1884 818 break;
cc53ce53
DH		 819 }
820
9875cf80
DH		 821 /* Transit to a mounted filesystem. */
822 if (managed & DCACHE_MOUNTED) {
823 struct vfsmount *mounted = lookup_mnt(path);
824 if (mounted) {
825 dput(path->dentry);
826 if (need_mntput)
827 mntput(path->mnt);
828 path->mnt = mounted;
829 path->dentry = dget(mounted->mnt_root);
830 need_mntput = true;
831 continue;
832 }
833
834 /* Something is mounted on this dentry in another
835 * namespace and/or whatever was mounted there in this
836 * namespace got unmounted before we managed to get the
837 * vfsmount_lock */
838 }
839
840 /* Handle an automount point */
841 if (managed & DCACHE_NEED_AUTOMOUNT) {
842 ret = follow_automount(path, flags, &need_mntput);
843 if (ret < 0)
8aef1884 844 break;
9875cf80
DH		 845 continue;
846 }
847
848 /* We didn't change the current path point */
849 break;
1da177e4 850 }
8aef1884
AV		 851
852 if (need_mntput && path->mnt == mnt)
853 mntput(path->mnt);
854 if (ret == -EISDIR)
855 ret = 0;
a3fbbde7 856 return ret < 0 ? ret : need_mntput;
1da177e4
LT		 857}
858
cc53ce53 859int follow_down_one(struct path *path)
1da177e4
LT		 860{
861 struct vfsmount *mounted;
862
1c755af4 863 mounted = lookup_mnt(path);
1da177e4 864 if (mounted) {
9393bd07
AV		 865 dput(path->dentry);
866 mntput(path->mnt);
867 path->mnt = mounted;
868 path->dentry = dget(mounted->mnt_root);
1da177e4
LT		 869 return 1;
870 }
871 return 0;
872}
873
62a7375e
IK		 874static inline bool managed_dentry_might_block(struct dentry *dentry)
875{
876 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
877 dentry->d_op->d_manage(dentry, true) < 0);
878}
879
9875cf80 880/*
287548e4
AV		 881 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
882 * we meet a managed dentry that would need blocking.
9875cf80
DH		 883 */
884static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
287548e4 885 struct inode **inode)
9875cf80 886{
62a7375e 887 for (;;) {
c7105365 888 struct mount *mounted;
62a7375e
IK		 889 /*
890 * Don't forget we might have a non-mountpoint managed dentry
891 * that wants to block transit.
892 */
287548e4 893 if (unlikely(managed_dentry_might_block(path->dentry)))
ab90911f 894 return false;
62a7375e
IK		 895
896 if (!d_mountpoint(path->dentry))
897 break;
898
9875cf80
DH		 899 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
900 if (!mounted)
901 break;
c7105365
AV		 902 path->mnt = &mounted->mnt;
903 path->dentry = mounted->mnt.mnt_root;
a3fbbde7 904 nd->flags |= LOOKUP_JUMPED;
9875cf80 905 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
59430262
LT		 906 /*
907 * Update the inode too. We don't need to re-check the
908 * dentry sequence number here after this d_inode read,
909 * because a mount-point is always pinned.
910 */
911 *inode = path->dentry->d_inode;
9875cf80 912 }
9875cf80
DH		 913 return true;
914}
915
dea39376 916static void follow_mount_rcu(struct nameidata *nd)
287548e4 917{
dea39376 918 while (d_mountpoint(nd->path.dentry)) {
c7105365 919 struct mount *mounted;
dea39376 920 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
287548e4
AV		 921 if (!mounted)
922 break;
c7105365
AV		 923 nd->path.mnt = &mounted->mnt;
924 nd->path.dentry = mounted->mnt.mnt_root;
dea39376 925 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
287548e4
AV		 926 }
927}
928
31e6b01f
NP		 929static int follow_dotdot_rcu(struct nameidata *nd)
930{
31e6b01f
NP		 931 set_root_rcu(nd);
932
9875cf80 933 while (1) {
31e6b01f
NP		 934 if (nd->path.dentry == nd->root.dentry &&
935 nd->path.mnt == nd->root.mnt) {
936 break;
937 }
938 if (nd->path.dentry != nd->path.mnt->mnt_root) {
939 struct dentry *old = nd->path.dentry;
940 struct dentry *parent = old->d_parent;
941 unsigned seq;
942
943 seq = read_seqcount_begin(&parent->d_seq);
944 if (read_seqcount_retry(&old->d_seq, nd->seq))
ef7562d5 945 goto failed;
31e6b01f
NP		 946 nd->path.dentry = parent;
947 nd->seq = seq;
948 break;
949 }
950 if (!follow_up_rcu(&nd->path))
951 break;
952 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
31e6b01f 953 }
dea39376
AV		 954 follow_mount_rcu(nd);
955 nd->inode = nd->path.dentry->d_inode;
31e6b01f 956 return 0;
ef7562d5
AV		 957
958failed:
959 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 960 if (!(nd->flags & LOOKUP_ROOT))
961 nd->root.mnt = NULL;
ef7562d5
AV		 962 rcu_read_unlock();
963 br_read_unlock(vfsmount_lock);
964 return -ECHILD;
31e6b01f
NP		 965}
966
cc53ce53
DH		 967/*
968 * Follow down to the covering mount currently visible to userspace. At each
969 * point, the filesystem owning that dentry may be queried as to whether the
970 * caller is permitted to proceed or not.
cc53ce53 971 */
7cc90cc3 972int follow_down(struct path *path)
cc53ce53
DH		 973{
974 unsigned managed;
975 int ret;
976
977 while (managed = ACCESS_ONCE(path->dentry->d_flags),
978 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
979 /* Allow the filesystem to manage the transit without i_mutex
980 * being held.
981 *
982 * We indicate to the filesystem if someone is trying to mount
983 * something here. This gives autofs the chance to deny anyone
984 * other than its daemon the right to mount on its
985 * superstructure.
986 *
987 * The filesystem may sleep at this point.
988 */
989 if (managed & DCACHE_MANAGE_TRANSIT) {
990 BUG_ON(!path->dentry->d_op);
991 BUG_ON(!path->dentry->d_op->d_manage);
ab90911f 992 ret = path->dentry->d_op->d_manage(
1aed3e42 993 path->dentry, false);
cc53ce53
DH		 994 if (ret < 0)
995 return ret == -EISDIR ? 0 : ret;
996 }
997
998 /* Transit to a mounted filesystem. */
999 if (managed & DCACHE_MOUNTED) {
1000 struct vfsmount *mounted = lookup_mnt(path);
1001 if (!mounted)
1002 break;
1003 dput(path->dentry);
1004 mntput(path->mnt);
1005 path->mnt = mounted;
1006 path->dentry = dget(mounted->mnt_root);
1007 continue;
1008 }
1009
1010 /* Don't handle automount points here */
1011 break;
1012 }
1013 return 0;
1014}
1015
9875cf80
DH		 1016/*
1017 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1018 */
1019static void follow_mount(struct path *path)
1020{
1021 while (d_mountpoint(path->dentry)) {
1022 struct vfsmount *mounted = lookup_mnt(path);
1023 if (!mounted)
1024 break;
1025 dput(path->dentry);
1026 mntput(path->mnt);
1027 path->mnt = mounted;
1028 path->dentry = dget(mounted->mnt_root);
1029 }
1030}
1031
31e6b01f 1032static void follow_dotdot(struct nameidata *nd)
1da177e4 1033{
2a737871 1034 set_root(nd);
e518ddb7 1035
1da177e4 1036 while(1) {
4ac91378 1037 struct dentry *old = nd->path.dentry;
1da177e4 1038
2a737871
AV		 1039 if (nd->path.dentry == nd->root.dentry &&
1040 nd->path.mnt == nd->root.mnt) {
1da177e4
LT		 1041 break;
1042 }
4ac91378 1043 if (nd->path.dentry != nd->path.mnt->mnt_root) {
3088dd70
AV		 1044 /* rare case of legitimate dget_parent()... */
1045 nd->path.dentry = dget_parent(nd->path.dentry);
1da177e4
LT		 1046 dput(old);
1047 break;
1048 }
3088dd70 1049 if (!follow_up(&nd->path))
1da177e4 1050 break;
1da177e4 1051 }
79ed0226 1052 follow_mount(&nd->path);
31e6b01f 1053 nd->inode = nd->path.dentry->d_inode;
1da177e4
LT		 1054}
1055
baa03890
NP		 1056/*
1057 * Allocate a dentry with name and parent, and perform a parent
1058 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1059 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1060 * have verified that no child exists while under i_mutex.
1061 */
1062static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1063 struct qstr *name, struct nameidata *nd)
1064{
1065 struct inode *inode = parent->d_inode;
1066 struct dentry *dentry;
1067 struct dentry *old;
1068
1069 /* Don't create child dentry for a dead directory. */
1070 if (unlikely(IS_DEADDIR(inode)))
1071 return ERR_PTR(-ENOENT);
1072
1073 dentry = d_alloc(parent, name);
1074 if (unlikely(!dentry))
1075 return ERR_PTR(-ENOMEM);
1076
1077 old = inode->i_op->lookup(inode, dentry, nd);
1078 if (unlikely(old)) {
1079 dput(dentry);
1080 dentry = old;
1081 }
1082 return dentry;
1083}
1084
44396f4b
JB		 1085/*
1086 * We already have a dentry, but require a lookup to be performed on the parent
1087 * directory to fill in d_inode. Returns the new dentry, or ERR_PTR on error.
1088 * parent->d_inode->i_mutex must be held. d_lookup must have verified that no
1089 * child exists while under i_mutex.
1090 */
1091static struct dentry *d_inode_lookup(struct dentry *parent, struct dentry *dentry,
1092 struct nameidata *nd)
1093{
1094 struct inode *inode = parent->d_inode;
1095 struct dentry *old;
1096
1097 /* Don't create child dentry for a dead directory. */
e188dc02
MS		 1098 if (unlikely(IS_DEADDIR(inode))) {
1099 dput(dentry);
44396f4b 1100 return ERR_PTR(-ENOENT);
e188dc02 1101 }
44396f4b
JB		 1102
1103 old = inode->i_op->lookup(inode, dentry, nd);
1104 if (unlikely(old)) {
1105 dput(dentry);
1106 dentry = old;
1107 }
1108 return dentry;
1109}
1110
1da177e4
LT		 1111/*
1112 * It's more convoluted than I'd like it to be, but... it's still fairly
1113 * small and for now I'd prefer to have fast path as straight as possible.
1114 * It _is_ time-critical.
1115 */
1116static int do_lookup(struct nameidata *nd, struct qstr *name,
31e6b01f 1117 struct path *path, struct inode **inode)
1da177e4 1118{
4ac91378 1119 struct vfsmount *mnt = nd->path.mnt;
31e6b01f 1120 struct dentry *dentry, *parent = nd->path.dentry;
5a18fff2
AV		 1121 int need_reval = 1;
1122 int status = 1;
9875cf80
DH		 1123 int err;
1124
b04f784e
NP		 1125 /*
1126 * Rename seqlock is not required here because in the off chance
1127 * of a false negative due to a concurrent rename, we're going to
1128 * do the non-racy lookup, below.
1129 */
31e6b01f
NP		 1130 if (nd->flags & LOOKUP_RCU) {
1131 unsigned seq;
31e6b01f
NP		 1132 *inode = nd->inode;
1133 dentry = __d_lookup_rcu(parent, name, &seq, inode);
5a18fff2
AV		 1134 if (!dentry)
1135 goto unlazy;
1136
31e6b01f
NP		 1137 /* Memory barrier in read_seqcount_begin of child is enough */
1138 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1139 return -ECHILD;
31e6b01f 1140 nd->seq = seq;
5a18fff2 1141
24643087 1142 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
5a18fff2
AV		 1143 status = d_revalidate(dentry, nd);
1144 if (unlikely(status <= 0)) {
1145 if (status != -ECHILD)
1146 need_reval = 0;
1147 goto unlazy;
1148 }
24643087 1149 }
44396f4b
JB		 1150 if (unlikely(d_need_lookup(dentry)))
1151 goto unlazy;
31e6b01f
NP		 1152 path->mnt = mnt;
1153 path->dentry = dentry;
d6e9bd25
AV		 1154 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1155 goto unlazy;
1156 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1157 goto unlazy;
1158 return 0;
5a18fff2 1159unlazy:
19660af7
AV		 1160 if (unlazy_walk(nd, dentry))
1161 return -ECHILD;
5a18fff2
AV		 1162 } else {
1163 dentry = __d_lookup(parent, name);
9875cf80 1164 }
5a18fff2 1165
44396f4b
JB		 1166 if (dentry && unlikely(d_need_lookup(dentry))) {
1167 dput(dentry);
1168 dentry = NULL;
1169 }
5a18fff2
AV		 1170retry:
1171 if (unlikely(!dentry)) {
1172 struct inode *dir = parent->d_inode;
1173 BUG_ON(nd->inode != dir);
1174
1175 mutex_lock(&dir->i_mutex);
1176 dentry = d_lookup(parent, name);
1177 if (likely(!dentry)) {
1178 dentry = d_alloc_and_lookup(parent, name, nd);
1179 if (IS_ERR(dentry)) {
1180 mutex_unlock(&dir->i_mutex);
1181 return PTR_ERR(dentry);
1182 }
1183 /* known good */
1184 need_reval = 0;
1185 status = 1;
44396f4b
JB		 1186 } else if (unlikely(d_need_lookup(dentry))) {
1187 dentry = d_inode_lookup(parent, dentry, nd);
1188 if (IS_ERR(dentry)) {
1189 mutex_unlock(&dir->i_mutex);
1190 return PTR_ERR(dentry);
1191 }
1192 /* known good */
1193 need_reval = 0;
1194 status = 1;
5a18fff2
AV		 1195 }
1196 mutex_unlock(&dir->i_mutex);
1197 }
1198 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1199 status = d_revalidate(dentry, nd);
1200 if (unlikely(status <= 0)) {
1201 if (status < 0) {
1202 dput(dentry);
1203 return status;
1204 }
1205 if (!d_invalidate(dentry)) {
1206 dput(dentry);
1207 dentry = NULL;
1208 need_reval = 1;
1209 goto retry;
1210 }
24643087 1211 }
5a18fff2 1212
9875cf80
DH		 1213 path->mnt = mnt;
1214 path->dentry = dentry;
1215 err = follow_managed(path, nd->flags);
89312214
IK		 1216 if (unlikely(err < 0)) {
1217 path_put_conditional(path, nd);
9875cf80 1218 return err;
89312214 1219 }
a3fbbde7
AV		 1220 if (err)
1221 nd->flags |= LOOKUP_JUMPED;
9875cf80 1222 *inode = path->dentry->d_inode;
1da177e4 1223 return 0;
1da177e4
LT		 1224}
1225
52094c8a
AV		 1226static inline int may_lookup(struct nameidata *nd)
1227{
1228 if (nd->flags & LOOKUP_RCU) {
4ad5abb3 1229 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
52094c8a
AV		 1230 if (err != -ECHILD)
1231 return err;
19660af7 1232 if (unlazy_walk(nd, NULL))
52094c8a
AV		 1233 return -ECHILD;
1234 }
4ad5abb3 1235 return inode_permission(nd->inode, MAY_EXEC);
52094c8a
AV		 1236}
1237
9856fa1b
AV		 1238static inline int handle_dots(struct nameidata *nd, int type)
1239{
1240 if (type == LAST_DOTDOT) {
1241 if (nd->flags & LOOKUP_RCU) {
1242 if (follow_dotdot_rcu(nd))
1243 return -ECHILD;
1244 } else
1245 follow_dotdot(nd);
1246 }
1247 return 0;
1248}
1249
951361f9
AV		 1250static void terminate_walk(struct nameidata *nd)
1251{
1252 if (!(nd->flags & LOOKUP_RCU)) {
1253 path_put(&nd->path);
1254 } else {
1255 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 1256 if (!(nd->flags & LOOKUP_ROOT))
1257 nd->root.mnt = NULL;
951361f9
AV		 1258 rcu_read_unlock();
1259 br_read_unlock(vfsmount_lock);
1260 }
1261}
1262
3ddcd056
LT		 1263/*
1264 * Do we need to follow links? We _really_ want to be able
1265 * to do this check without having to look at inode->i_op,
1266 * so we keep a cache of "no, this doesn't need follow_link"
1267 * for the common case.
1268 */
7813b94a 1269static inline int should_follow_link(struct inode *inode, int follow)
3ddcd056
LT		 1270{
1271 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1272 if (likely(inode->i_op->follow_link))
1273 return follow;
1274
1275 /* This gets set once for the inode lifetime */
1276 spin_lock(&inode->i_lock);
1277 inode->i_opflags |= IOP_NOFOLLOW;
1278 spin_unlock(&inode->i_lock);
1279 }
1280 return 0;
1281}
1282
ce57dfc1
AV		 1283static inline int walk_component(struct nameidata *nd, struct path *path,
1284 struct qstr *name, int type, int follow)
1285{
1286 struct inode *inode;
1287 int err;
1288 /*
1289 * "." and ".." are special - ".." especially so because it has
1290 * to be able to know about the current root directory and
1291 * parent relationships.
1292 */
1293 if (unlikely(type != LAST_NORM))
1294 return handle_dots(nd, type);
1295 err = do_lookup(nd, name, path, &inode);
1296 if (unlikely(err)) {
1297 terminate_walk(nd);
1298 return err;
1299 }
1300 if (!inode) {
1301 path_to_nameidata(path, nd);
1302 terminate_walk(nd);
1303 return -ENOENT;
1304 }
7813b94a 1305 if (should_follow_link(inode, follow)) {
19660af7
AV		 1306 if (nd->flags & LOOKUP_RCU) {
1307 if (unlikely(unlazy_walk(nd, path->dentry))) {
1308 terminate_walk(nd);
1309 return -ECHILD;
1310 }
1311 }
ce57dfc1
AV		 1312 BUG_ON(inode != path->dentry->d_inode);
1313 return 1;
1314 }
1315 path_to_nameidata(path, nd);
1316 nd->inode = inode;
1317 return 0;
1318}
1319
b356379a
AV		 1320/*
1321 * This limits recursive symlink follows to 8, while
1322 * limiting consecutive symlinks to 40.
1323 *
1324 * Without that kind of total limit, nasty chains of consecutive
1325 * symlinks can cause almost arbitrarily long lookups.
1326 */
1327static inline int nested_symlink(struct path *path, struct nameidata *nd)
1328{
1329 int res;
1330
b356379a
AV		 1331 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1332 path_put_conditional(path, nd);
1333 path_put(&nd->path);
1334 return -ELOOP;
1335 }
1a4022f8 1336 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
b356379a
AV		 1337
1338 nd->depth++;
1339 current->link_count++;
1340
1341 do {
1342 struct path link = *path;
1343 void *cookie;
574197e0
AV		 1344
1345 res = follow_link(&link, nd, &cookie);
b356379a
AV		 1346 if (!res)
1347 res = walk_component(nd, path, &nd->last,
1348 nd->last_type, LOOKUP_FOLLOW);
574197e0 1349 put_link(nd, &link, cookie);
b356379a
AV		 1350 } while (res > 0);
1351
1352 current->link_count--;
1353 nd->depth--;
1354 return res;
1355}
1356
3ddcd056
LT		 1357/*
1358 * We really don't want to look at inode->i_op->lookup
1359 * when we don't have to. So we keep a cache bit in
1360 * the inode ->i_opflags field that says "yes, we can
1361 * do lookup on this inode".
1362 */
1363static inline int can_lookup(struct inode *inode)
1364{
1365 if (likely(inode->i_opflags & IOP_LOOKUP))
1366 return 1;
1367 if (likely(!inode->i_op->lookup))
1368 return 0;
1369
1370 /* We do this once for the lifetime of the inode */
1371 spin_lock(&inode->i_lock);
1372 inode->i_opflags |= IOP_LOOKUP;
1373 spin_unlock(&inode->i_lock);
1374 return 1;
1375}
1376
1da177e4
LT		 1377/*
1378 * Name resolution.
ea3834d9
PM		 1379 * This is the basic name resolution function, turning a pathname into
1380 * the final dentry. We expect 'base' to be positive and a directory.
1da177e4 1381 *
ea3834d9
PM		 1382 * Returns 0 and nd will have valid dentry and mnt on success.
1383 * Returns error and drops reference to input namei data on failure.
1da177e4 1384 */
6de88d72 1385static int link_path_walk(const char *name, struct nameidata *nd)
1da177e4
LT		 1386{
1387 struct path next;
1da177e4 1388 int err;
1da177e4
LT		 1389
1390 while (*name=='/')
1391 name++;
1392 if (!*name)
086e183a 1393 return 0;
1da177e4 1394
1da177e4
LT		 1395 /* At this point we know we have a real path component. */
1396 for(;;) {
1397 unsigned long hash;
1398 struct qstr this;
1399 unsigned int c;
fe479a58 1400 int type;
1da177e4 1401
52094c8a 1402 err = may_lookup(nd);
1da177e4
LT		 1403 if (err)
1404 break;
1405
1406 this.name = name;
1407 c = *(const unsigned char *)name;
1408
1409 hash = init_name_hash();
1410 do {
1411 name++;
1412 hash = partial_name_hash(c, hash);
1413 c = *(const unsigned char *)name;
1414 } while (c && (c != '/'));
1415 this.len = name - (const char *) this.name;
1416 this.hash = end_name_hash(hash);
1417
fe479a58
AV		 1418 type = LAST_NORM;
1419 if (this.name[0] == '.') switch (this.len) {
1420 case 2:
16c2cd71 1421 if (this.name[1] == '.') {
fe479a58 1422 type = LAST_DOTDOT;
16c2cd71
AV		 1423 nd->flags |= LOOKUP_JUMPED;
1424 }
fe479a58
AV		 1425 break;
1426 case 1:
1427 type = LAST_DOT;
1428 }
5a202bcd
AV		 1429 if (likely(type == LAST_NORM)) {
1430 struct dentry *parent = nd->path.dentry;
16c2cd71 1431 nd->flags &= ~LOOKUP_JUMPED;
5a202bcd
AV		 1432 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1433 err = parent->d_op->d_hash(parent, nd->inode,
1434 &this);
1435 if (err < 0)
1436 break;
1437 }
1438 }
fe479a58 1439
1da177e4
LT		 1440 /* remove trailing slashes? */
1441 if (!c)
1442 goto last_component;
1443 while (*++name == '/');
1444 if (!*name)
b356379a 1445 goto last_component;
1da177e4 1446
ce57dfc1
AV		 1447 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1448 if (err < 0)
1449 return err;
1da177e4 1450
ce57dfc1 1451 if (err) {
b356379a 1452 err = nested_symlink(&next, nd);
1da177e4 1453 if (err)
a7472bab 1454 return err;
31e6b01f 1455 }
3ddcd056
LT		 1456 if (can_lookup(nd->inode))
1457 continue;
1da177e4 1458 err = -ENOTDIR;
3ddcd056 1459 break;
1da177e4
LT		 1460 /* here ends the main loop */
1461
1da177e4 1462last_component:
b356379a
AV		 1463 nd->last = this;
1464 nd->last_type = type;
086e183a 1465 return 0;
1da177e4 1466 }
951361f9 1467 terminate_walk(nd);
1da177e4
LT		 1468 return err;
1469}
1470
70e9b357
AV		 1471static int path_init(int dfd, const char *name, unsigned int flags,
1472 struct nameidata *nd, struct file **fp)
31e6b01f
NP		 1473{
1474 int retval = 0;
1475 int fput_needed;
1476 struct file *file;
1477
1478 nd->last_type = LAST_ROOT; /* if there are only slashes... */
16c2cd71 1479 nd->flags = flags | LOOKUP_JUMPED;
31e6b01f 1480 nd->depth = 0;
5b6ca027
AV		 1481 if (flags & LOOKUP_ROOT) {
1482 struct inode *inode = nd->root.dentry->d_inode;
73d049a4
AV		 1483 if (*name) {
1484 if (!inode->i_op->lookup)
1485 return -ENOTDIR;
1486 retval = inode_permission(inode, MAY_EXEC);
1487 if (retval)
1488 return retval;
1489 }
5b6ca027
AV		 1490 nd->path = nd->root;
1491 nd->inode = inode;
1492 if (flags & LOOKUP_RCU) {
1493 br_read_lock(vfsmount_lock);
1494 rcu_read_lock();
1495 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1496 } else {
1497 path_get(&nd->path);
1498 }
1499 return 0;
1500 }
1501
31e6b01f 1502 nd->root.mnt = NULL;
31e6b01f
NP		 1503
1504 if (*name=='/') {
e41f7d4e
AV		 1505 if (flags & LOOKUP_RCU) {
1506 br_read_lock(vfsmount_lock);
1507 rcu_read_lock();
1508 set_root_rcu(nd);
1509 } else {
1510 set_root(nd);
1511 path_get(&nd->root);
1512 }
1513 nd->path = nd->root;
31e6b01f 1514 } else if (dfd == AT_FDCWD) {
e41f7d4e
AV		 1515 if (flags & LOOKUP_RCU) {
1516 struct fs_struct *fs = current->fs;
1517 unsigned seq;
31e6b01f 1518
e41f7d4e
AV		 1519 br_read_lock(vfsmount_lock);
1520 rcu_read_lock();
c28cc364 1521
e41f7d4e
AV		 1522 do {
1523 seq = read_seqcount_begin(&fs->seq);
1524 nd->path = fs->pwd;
1525 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1526 } while (read_seqcount_retry(&fs->seq, seq));
1527 } else {
1528 get_fs_pwd(current->fs, &nd->path);
1529 }
31e6b01f
NP		 1530 } else {
1531 struct dentry *dentry;
1532
1abf0c71 1533 file = fget_raw_light(dfd, &fput_needed);
31e6b01f
NP		 1534 retval = -EBADF;
1535 if (!file)
1536 goto out_fail;
1537
1538 dentry = file->f_path.dentry;
1539
f52e0c11
AV		 1540 if (*name) {
1541 retval = -ENOTDIR;
1542 if (!S_ISDIR(dentry->d_inode->i_mode))
1543 goto fput_fail;
31e6b01f 1544
4ad5abb3 1545 retval = inode_permission(dentry->d_inode, MAY_EXEC);
f52e0c11
AV		 1546 if (retval)
1547 goto fput_fail;
1548 }
31e6b01f
NP		 1549
1550 nd->path = file->f_path;
e41f7d4e
AV		 1551 if (flags & LOOKUP_RCU) {
1552 if (fput_needed)
70e9b357 1553 *fp = file;
e41f7d4e
AV		 1554 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1555 br_read_lock(vfsmount_lock);
1556 rcu_read_lock();
1557 } else {
1558 path_get(&file->f_path);
1559 fput_light(file, fput_needed);
1560 }
31e6b01f 1561 }
31e6b01f 1562
31e6b01f 1563 nd->inode = nd->path.dentry->d_inode;
9b4a9b14 1564 return 0;
2dfdd266 1565
9b4a9b14
AV		 1566fput_fail:
1567 fput_light(file, fput_needed);
1568out_fail:
1569 return retval;
1570}
1571
bd92d7fe
AV		 1572static inline int lookup_last(struct nameidata *nd, struct path *path)
1573{
1574 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1575 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1576
1577 nd->flags &= ~LOOKUP_PARENT;
1578 return walk_component(nd, path, &nd->last, nd->last_type,
1579 nd->flags & LOOKUP_FOLLOW);
1580}
1581
9b4a9b14 1582/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
ee0827cd 1583static int path_lookupat(int dfd, const char *name,
9b4a9b14
AV		 1584 unsigned int flags, struct nameidata *nd)
1585{
70e9b357 1586 struct file *base = NULL;
bd92d7fe
AV		 1587 struct path path;
1588 int err;
31e6b01f
NP		 1589
1590 /*
1591 * Path walking is largely split up into 2 different synchronisation
1592 * schemes, rcu-walk and ref-walk (explained in
1593 * Documentation/filesystems/path-lookup.txt). These share much of the
1594 * path walk code, but some things particularly setup, cleanup, and
1595 * following mounts are sufficiently divergent that functions are
1596 * duplicated. Typically there is a function foo(), and its RCU
1597 * analogue, foo_rcu().
1598 *
1599 * -ECHILD is the error number of choice (just to avoid clashes) that
1600 * is returned if some aspect of an rcu-walk fails. Such an error must
1601 * be handled by restarting a traditional ref-walk (which will always
1602 * be able to complete).
1603 */
bd92d7fe 1604 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
ee0827cd 1605
bd92d7fe
AV		 1606 if (unlikely(err))
1607 return err;
ee0827cd
AV		 1608
1609 current->total_link_count = 0;
bd92d7fe
AV		 1610 err = link_path_walk(name, nd);
1611
1612 if (!err && !(flags & LOOKUP_PARENT)) {
bd92d7fe
AV		 1613 err = lookup_last(nd, &path);
1614 while (err > 0) {
1615 void *cookie;
1616 struct path link = path;
bd92d7fe 1617 nd->flags |= LOOKUP_PARENT;
574197e0 1618 err = follow_link(&link, nd, &cookie);
bd92d7fe
AV		 1619 if (!err)
1620 err = lookup_last(nd, &path);
574197e0 1621 put_link(nd, &link, cookie);
bd92d7fe
AV		 1622 }
1623 }
ee0827cd 1624
9f1fafee
AV		 1625 if (!err)
1626 err = complete_walk(nd);
bd92d7fe
AV		 1627
1628 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1629 if (!nd->inode->i_op->lookup) {
1630 path_put(&nd->path);
bd23a539 1631 err = -ENOTDIR;
bd92d7fe
AV		 1632 }
1633 }
16c2cd71 1634
70e9b357
AV		 1635 if (base)
1636 fput(base);
ee0827cd 1637
5b6ca027 1638 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
2a737871
AV		 1639 path_put(&nd->root);
1640 nd->root.mnt = NULL;
1641 }
bd92d7fe 1642 return err;
ee0827cd 1643}
31e6b01f 1644
ee0827cd
AV		 1645static int do_path_lookup(int dfd, const char *name,
1646 unsigned int flags, struct nameidata *nd)
1647{
1648 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1649 if (unlikely(retval == -ECHILD))
1650 retval = path_lookupat(dfd, name, flags, nd);
1651 if (unlikely(retval == -ESTALE))
1652 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
31e6b01f
NP		 1653
1654 if (likely(!retval)) {
1655 if (unlikely(!audit_dummy_context())) {
1656 if (nd->path.dentry && nd->inode)
1657 audit_inode(name, nd->path.dentry);
1658 }
1659 }
170aa3d0 1660 return retval;
1da177e4
LT		 1661}
1662
c9c6cac0 1663int kern_path_parent(const char *name, struct nameidata *nd)
5590ff0d 1664{
c9c6cac0 1665 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
5590ff0d
UD		 1666}
1667
d1811465
AV		 1668int kern_path(const char *name, unsigned int flags, struct path *path)
1669{
1670 struct nameidata nd;
1671 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1672 if (!res)
1673 *path = nd.path;
1674 return res;
1675}
1676
16f18200
JJS		 1677/**
1678 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1679 * @dentry: pointer to dentry of the base directory
1680 * @mnt: pointer to vfs mount of the base directory
1681 * @name: pointer to file name
1682 * @flags: lookup flags
e0a01249 1683 * @path: pointer to struct path to fill
16f18200
JJS		 1684 */
1685int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1686 const char *name, unsigned int flags,
e0a01249 1687 struct path *path)
16f18200 1688{
e0a01249
AV		 1689 struct nameidata nd;
1690 int err;
1691 nd.root.dentry = dentry;
1692 nd.root.mnt = mnt;
1693 BUG_ON(flags & LOOKUP_PARENT);
5b6ca027 1694 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
e0a01249
AV		 1695 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1696 if (!err)
1697 *path = nd.path;
1698 return err;
16f18200
JJS		 1699}
1700
eead1911
CH		 1701static struct dentry *__lookup_hash(struct qstr *name,
1702 struct dentry *base, struct nameidata *nd)
1da177e4 1703{
81fca444 1704 struct inode *inode = base->d_inode;
057f6c01 1705 struct dentry *dentry;
1da177e4
LT		 1706 int err;
1707
4ad5abb3 1708 err = inode_permission(inode, MAY_EXEC);
81fca444
CH		 1709 if (err)
1710 return ERR_PTR(err);
1da177e4 1711
b04f784e
NP		 1712 /*
1713 * Don't bother with __d_lookup: callers are for creat as
1714 * well as unlink, so a lot of the time it would cost
1715 * a double lookup.
6e6b1bd1 1716 */
b04f784e 1717 dentry = d_lookup(base, name);
6e6b1bd1 1718
44396f4b
JB		 1719 if (dentry && d_need_lookup(dentry)) {
1720 /*
1721 * __lookup_hash is called with the parent dir's i_mutex already
1722 * held, so we are good to go here.
1723 */
1724 dentry = d_inode_lookup(base, dentry, nd);
1725 if (IS_ERR(dentry))
1726 return dentry;
1727 }
1728
d2d9e9fb
AV		 1729 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1730 int status = d_revalidate(dentry, nd);
1731 if (unlikely(status <= 0)) {
1732 /*
1733 * The dentry failed validation.
1734 * If d_revalidate returned 0 attempt to invalidate
1735 * the dentry otherwise d_revalidate is asking us
1736 * to return a fail status.
1737 */
1738 if (status < 0) {
1739 dput(dentry);
1740 return ERR_PTR(status);
1741 } else if (!d_invalidate(dentry)) {
1742 dput(dentry);
1743 dentry = NULL;
1744 }
1745 }
1746 }
6e6b1bd1 1747
baa03890
NP		 1748 if (!dentry)
1749 dentry = d_alloc_and_lookup(base, name, nd);
5a202bcd 1750
1da177e4
LT		 1751 return dentry;
1752}
1753
057f6c01
JM		 1754/*
1755 * Restricted form of lookup. Doesn't follow links, single-component only,
1756 * needs parent already locked. Doesn't follow mounts.
1757 * SMP-safe.
1758 */
eead1911 1759static struct dentry *lookup_hash(struct nameidata *nd)
057f6c01 1760{
4ac91378 1761 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1da177e4
LT		 1762}
1763
eead1911 1764/**
a6b91919 1765 * lookup_one_len - filesystem helper to lookup single pathname component
eead1911
CH		 1766 * @name: pathname component to lookup
1767 * @base: base directory to lookup from
1768 * @len: maximum length @len should be interpreted to
1769 *
a6b91919
RD		 1770 * Note that this routine is purely a helper for filesystem usage and should
1771 * not be called by generic code. Also note that by using this function the
eead1911
CH		 1772 * nameidata argument is passed to the filesystem methods and a filesystem
1773 * using this helper needs to be prepared for that.
1774 */
057f6c01
JM		 1775struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1776{
057f6c01 1777 struct qstr this;
6a96ba54
AV		 1778 unsigned long hash;
1779 unsigned int c;
057f6c01 1780
2f9092e1
DW		 1781 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1782
6a96ba54
AV		 1783 this.name = name;
1784 this.len = len;
1785 if (!len)
1786 return ERR_PTR(-EACCES);
1787
1788 hash = init_name_hash();
1789 while (len--) {
1790 c = *(const unsigned char *)name++;
1791 if (c == '/' || c == '\0')
1792 return ERR_PTR(-EACCES);
1793 hash = partial_name_hash(c, hash);
1794 }
1795 this.hash = end_name_hash(hash);
5a202bcd
AV		 1796 /*
1797 * See if the low-level filesystem might want
1798 * to use its own hash..
1799 */
1800 if (base->d_flags & DCACHE_OP_HASH) {
1801 int err = base->d_op->d_hash(base, base->d_inode, &this);
1802 if (err < 0)
1803 return ERR_PTR(err);
1804 }
eead1911 1805
49705b77 1806 return __lookup_hash(&this, base, NULL);
057f6c01
JM		 1807}
1808
1fa1e7f6
AW		 1809int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
1810 struct path *path, int *empty)
1da177e4 1811{
2d8f3038 1812 struct nameidata nd;
1fa1e7f6 1813 char *tmp = getname_flags(name, flags, empty);
1da177e4 1814 int err = PTR_ERR(tmp);
1da177e4 1815 if (!IS_ERR(tmp)) {
2d8f3038
AV		 1816
1817 BUG_ON(flags & LOOKUP_PARENT);
1818
1819 err = do_path_lookup(dfd, tmp, flags, &nd);
1da177e4 1820 putname(tmp);
2d8f3038
AV		 1821 if (!err)
1822 *path = nd.path;
1da177e4
LT		 1823 }
1824 return err;
1825}
1826
1fa1e7f6
AW		 1827int user_path_at(int dfd, const char __user *name, unsigned flags,
1828 struct path *path)
1829{
1830 return user_path_at_empty(dfd, name, flags, path, 0);
1831}
1832
2ad94ae6
AV		 1833static int user_path_parent(int dfd, const char __user *path,
1834 struct nameidata *nd, char **name)
1835{
1836 char *s = getname(path);
1837 int error;
1838
1839 if (IS_ERR(s))
1840 return PTR_ERR(s);
1841
1842 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1843 if (error)
1844 putname(s);
1845 else
1846 *name = s;
1847
1848 return error;
1849}
1850
1da177e4
LT		 1851/*
1852 * It's inline, so penalty for filesystems that don't use sticky bit is
1853 * minimal.
1854 */
1855static inline int check_sticky(struct inode *dir, struct inode *inode)
1856{
da9592ed
DH		 1857 uid_t fsuid = current_fsuid();
1858
1da177e4
LT		 1859 if (!(dir->i_mode & S_ISVTX))
1860 return 0;
e795b717
SH		 1861 if (current_user_ns() != inode_userns(inode))
1862 goto other_userns;
da9592ed 1863 if (inode->i_uid == fsuid)
1da177e4 1864 return 0;
da9592ed 1865 if (dir->i_uid == fsuid)
1da177e4 1866 return 0;
e795b717
SH		 1867
1868other_userns:
1869 return !ns_capable(inode_userns(inode), CAP_FOWNER);
1da177e4
LT		 1870}
1871
1872/*
1873 * Check whether we can remove a link victim from directory dir, check
1874 * whether the type of victim is right.
1875 * 1. We can't do it if dir is read-only (done in permission())
1876 * 2. We should have write and exec permissions on dir
1877 * 3. We can't remove anything from append-only dir
1878 * 4. We can't do anything with immutable dir (done in permission())
1879 * 5. If the sticky bit on dir is set we should either
1880 * a. be owner of dir, or
1881 * b. be owner of victim, or
1882 * c. have CAP_FOWNER capability
1883 * 6. If the victim is append-only or immutable we can't do antyhing with
1884 * links pointing to it.
1885 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1886 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1887 * 9. We can't remove a root or mountpoint.
1888 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1889 * nfs_async_unlink().
1890 */
858119e1 1891static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1da177e4
LT		 1892{
1893 int error;
1894
1895 if (!victim->d_inode)
1896 return -ENOENT;
1897
1898 BUG_ON(victim->d_parent->d_inode != dir);
cccc6bba 1899 audit_inode_child(victim, dir);
1da177e4 1900
f419a2e3 1901 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1902 if (error)
1903 return error;
1904 if (IS_APPEND(dir))
1905 return -EPERM;
1906 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
f9454548 1907 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1da177e4
LT		 1908 return -EPERM;
1909 if (isdir) {
1910 if (!S_ISDIR(victim->d_inode->i_mode))
1911 return -ENOTDIR;
1912 if (IS_ROOT(victim))
1913 return -EBUSY;
1914 } else if (S_ISDIR(victim->d_inode->i_mode))
1915 return -EISDIR;
1916 if (IS_DEADDIR(dir))
1917 return -ENOENT;
1918 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1919 return -EBUSY;
1920 return 0;
1921}
1922
1923/* Check whether we can create an object with dentry child in directory
1924 * dir.
1925 * 1. We can't do it if child already exists (open has special treatment for
1926 * this case, but since we are inlined it's OK)
1927 * 2. We can't do it if dir is read-only (done in permission())
1928 * 3. We should have write and exec permissions on dir
1929 * 4. We can't do it if dir is immutable (done in permission())
1930 */
a95164d9 1931static inline int may_create(struct inode *dir, struct dentry *child)
1da177e4
LT		 1932{
1933 if (child->d_inode)
1934 return -EEXIST;
1935 if (IS_DEADDIR(dir))
1936 return -ENOENT;
f419a2e3 1937 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1938}
1939
1da177e4
LT		 1940/*
1941 * p1 and p2 should be directories on the same fs.
1942 */
1943struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1944{
1945 struct dentry *p;
1946
1947 if (p1 == p2) {
f2eace23 1948 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1da177e4
LT		 1949 return NULL;
1950 }
1951
a11f3a05 1952 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4 1953
e2761a11
OH		 1954 p = d_ancestor(p2, p1);
1955 if (p) {
1956 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1957 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1958 return p;
1da177e4
LT		 1959 }
1960
e2761a11
OH		 1961 p = d_ancestor(p1, p2);
1962 if (p) {
1963 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1964 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1965 return p;
1da177e4
LT		 1966 }
1967
f2eace23
IM		 1968 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1969 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1da177e4
LT		 1970 return NULL;
1971}
1972
1973void unlock_rename(struct dentry *p1, struct dentry *p2)
1974{
1b1dcc1b 1975 mutex_unlock(&p1->d_inode->i_mutex);
1da177e4 1976 if (p1 != p2) {
1b1dcc1b 1977 mutex_unlock(&p2->d_inode->i_mutex);
a11f3a05 1978 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4
LT		 1979 }
1980}
1981
4acdaf27 1982int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
1da177e4
LT		 1983 struct nameidata *nd)
1984{
a95164d9 1985 int error = may_create(dir, dentry);
1da177e4
LT		 1986
1987 if (error)
1988 return error;
1989
acfa4380 1990 if (!dir->i_op->create)
1da177e4
LT		 1991 return -EACCES; /* shouldn't it be ENOSYS? */
1992 mode &= S_IALLUGO;
1993 mode |= S_IFREG;
1994 error = security_inode_create(dir, dentry, mode);
1995 if (error)
1996 return error;
1da177e4 1997 error = dir->i_op->create(dir, dentry, mode, nd);
a74574aa 1998 if (!error)
f38aa942 1999 fsnotify_create(dir, dentry);
1da177e4
LT		 2000 return error;
2001}
2002
73d049a4 2003static int may_open(struct path *path, int acc_mode, int flag)
1da177e4 2004{
3fb64190 2005 struct dentry *dentry = path->dentry;
1da177e4
LT		 2006 struct inode *inode = dentry->d_inode;
2007 int error;
2008
bcda7652
AV		 2009 /* O_PATH? */
2010 if (!acc_mode)
2011 return 0;
2012
1da177e4
LT		 2013 if (!inode)
2014 return -ENOENT;
2015
c8fe8f30
CH		 2016 switch (inode->i_mode & S_IFMT) {
2017 case S_IFLNK:
1da177e4 2018 return -ELOOP;
c8fe8f30
CH		 2019 case S_IFDIR:
2020 if (acc_mode & MAY_WRITE)
2021 return -EISDIR;
2022 break;
2023 case S_IFBLK:
2024 case S_IFCHR:
3fb64190 2025 if (path->mnt->mnt_flags & MNT_NODEV)
1da177e4 2026 return -EACCES;
c8fe8f30
CH		 2027 /*FALLTHRU*/
2028 case S_IFIFO:
2029 case S_IFSOCK:
1da177e4 2030 flag &= ~O_TRUNC;
c8fe8f30 2031 break;
4a3fd211 2032 }
b41572e9 2033
3fb64190 2034 error = inode_permission(inode, acc_mode);
b41572e9
DH		 2035 if (error)
2036 return error;
6146f0d5 2037
1da177e4
LT		 2038 /*
2039 * An append-only file must be opened in append mode for writing.
2040 */
2041 if (IS_APPEND(inode)) {
8737c930 2042 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
7715b521 2043 return -EPERM;
1da177e4 2044 if (flag & O_TRUNC)
7715b521 2045 return -EPERM;
1da177e4
LT		 2046 }
2047
2048 /* O_NOATIME can only be set by the owner or superuser */
2e149670 2049 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
7715b521 2050 return -EPERM;
1da177e4 2051
f3c7691e 2052 return 0;
7715b521 2053}
1da177e4 2054
e1181ee6 2055static int handle_truncate(struct file *filp)
7715b521 2056{
e1181ee6 2057 struct path *path = &filp->f_path;
7715b521
AV		 2058 struct inode *inode = path->dentry->d_inode;
2059 int error = get_write_access(inode);
2060 if (error)
2061 return error;
2062 /*
2063 * Refuse to truncate files with mandatory locks held on them.
2064 */
2065 error = locks_verify_locked(inode);
2066 if (!error)
ea0d3ab2 2067 error = security_path_truncate(path);
7715b521
AV		 2068 if (!error) {
2069 error = do_truncate(path->dentry, 0,
2070 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
e1181ee6 2071 filp);
7715b521
AV		 2072 }
2073 put_write_access(inode);
acd0c935 2074 return error;
1da177e4
LT		 2075}
2076
d57999e1
DH		 2077static inline int open_to_namei_flags(int flag)
2078{
8a5e929d
AV		 2079 if ((flag & O_ACCMODE) == 3)
2080 flag--;
d57999e1
DH		 2081 return flag;
2082}
2083
31e6b01f 2084/*
fe2d35ff 2085 * Handle the last step of open()
31e6b01f 2086 */
fb1cc555 2087static struct file *do_last(struct nameidata *nd, struct path *path,
c3e380b0 2088 const struct open_flags *op, const char *pathname)
fb1cc555 2089{
a1e28038 2090 struct dentry *dir = nd->path.dentry;
6c0d46c4 2091 struct dentry *dentry;
ca344a89 2092 int open_flag = op->open_flag;
6c0d46c4 2093 int will_truncate = open_flag & O_TRUNC;
ca344a89 2094 int want_write = 0;
bcda7652 2095 int acc_mode = op->acc_mode;
fb1cc555 2096 struct file *filp;
16c2cd71 2097 int error;
1f36f774 2098
c3e380b0
AV		 2099 nd->flags &= ~LOOKUP_PARENT;
2100 nd->flags |= op->intent;
2101
1f36f774
AV		 2102 switch (nd->last_type) {
2103 case LAST_DOTDOT:
176306f5 2104 case LAST_DOT:
fe2d35ff
AV		 2105 error = handle_dots(nd, nd->last_type);
2106 if (error)
2107 return ERR_PTR(error);
1f36f774 2108 /* fallthrough */
1f36f774 2109 case LAST_ROOT:
9f1fafee 2110 error = complete_walk(nd);
16c2cd71 2111 if (error)
9f1fafee 2112 return ERR_PTR(error);
fe2d35ff 2113 audit_inode(pathname, nd->path.dentry);
ca344a89 2114 if (open_flag & O_CREAT) {
fe2d35ff
AV		 2115 error = -EISDIR;
2116 goto exit;
2117 }
2118 goto ok;
1f36f774 2119 case LAST_BIND:
9f1fafee 2120 error = complete_walk(nd);
16c2cd71 2121 if (error)
9f1fafee 2122 return ERR_PTR(error);
1f36f774 2123 audit_inode(pathname, dir);
67ee3ad2 2124 goto ok;
1f36f774 2125 }
67ee3ad2 2126
ca344a89 2127 if (!(open_flag & O_CREAT)) {
bcda7652 2128 int symlink_ok = 0;
fe2d35ff
AV		 2129 if (nd->last.name[nd->last.len])
2130 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
bcda7652
AV		 2131 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2132 symlink_ok = 1;
fe2d35ff 2133 /* we _can_ be in RCU mode here */
ce57dfc1
AV		 2134 error = walk_component(nd, path, &nd->last, LAST_NORM,
2135 !symlink_ok);
2136 if (error < 0)
fe2d35ff 2137 return ERR_PTR(error);
ce57dfc1 2138 if (error) /* symlink */
fe2d35ff 2139 return NULL;
fe2d35ff 2140 /* sayonara */
9f1fafee
AV		 2141 error = complete_walk(nd);
2142 if (error)
2143 return ERR_PTR(-ECHILD);
fe2d35ff
AV		 2144
2145 error = -ENOTDIR;
2146 if (nd->flags & LOOKUP_DIRECTORY) {
ce57dfc1 2147 if (!nd->inode->i_op->lookup)
fe2d35ff
AV		 2148 goto exit;
2149 }
2150 audit_inode(pathname, nd->path.dentry);
2151 goto ok;
2152 }
2153
2154 /* create side of things */
a3fbbde7
AV		 2155 /*
2156 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED has been
2157 * cleared when we got to the last component we are about to look up
2158 */
9f1fafee
AV		 2159 error = complete_walk(nd);
2160 if (error)
2161 return ERR_PTR(error);
fe2d35ff
AV		 2162
2163 audit_inode(pathname, dir);
16c2cd71 2164 error = -EISDIR;
1f36f774 2165 /* trailing slashes? */
31e6b01f
NP		 2166 if (nd->last.name[nd->last.len])
2167 goto exit;
a2c36b45 2168
a1e28038
AV		 2169 mutex_lock(&dir->d_inode->i_mutex);
2170
6c0d46c4
AV		 2171 dentry = lookup_hash(nd);
2172 error = PTR_ERR(dentry);
2173 if (IS_ERR(dentry)) {
fb1cc555
AV		 2174 mutex_unlock(&dir->d_inode->i_mutex);
2175 goto exit;
2176 }
2177
6c0d46c4
AV		 2178 path->dentry = dentry;
2179 path->mnt = nd->path.mnt;
2180
fb1cc555 2181 /* Negative dentry, just create the file */
6c0d46c4 2182 if (!dentry->d_inode) {
a218d0fd 2183 umode_t mode = op->mode;
6c0d46c4
AV		 2184 if (!IS_POSIXACL(dir->d_inode))
2185 mode &= ~current_umask();
fb1cc555
AV		 2186 /*
2187 * This write is needed to ensure that a
6c0d46c4 2188 * rw->ro transition does not occur between
fb1cc555
AV		 2189 * the time when the file is created and when
2190 * a permanent write count is taken through
2191 * the 'struct file' in nameidata_to_filp().
2192 */
2193 error = mnt_want_write(nd->path.mnt);
2194 if (error)
2195 goto exit_mutex_unlock;
ca344a89 2196 want_write = 1;
9b44f1b3 2197 /* Don't check for write permission, don't truncate */
ca344a89 2198 open_flag &= ~O_TRUNC;
6c0d46c4 2199 will_truncate = 0;
bcda7652 2200 acc_mode = MAY_OPEN;
6c0d46c4
AV		 2201 error = security_path_mknod(&nd->path, dentry, mode, 0);
2202 if (error)
2203 goto exit_mutex_unlock;
2204 error = vfs_create(dir->d_inode, dentry, mode, nd);
2205 if (error)
2206 goto exit_mutex_unlock;
2207 mutex_unlock(&dir->d_inode->i_mutex);
2208 dput(nd->path.dentry);
2209 nd->path.dentry = dentry;
ca344a89 2210 goto common;
fb1cc555
AV		 2211 }
2212
2213 /*
2214 * It already exists.
2215 */
2216 mutex_unlock(&dir->d_inode->i_mutex);
2217 audit_inode(pathname, path->dentry);
2218
2219 error = -EEXIST;
ca344a89 2220 if (open_flag & O_EXCL)
fb1cc555
AV		 2221 goto exit_dput;
2222
9875cf80
DH		 2223 error = follow_managed(path, nd->flags);
2224 if (error < 0)
2225 goto exit_dput;
fb1cc555 2226
a3fbbde7
AV		 2227 if (error)
2228 nd->flags |= LOOKUP_JUMPED;
2229
fb1cc555
AV		 2230 error = -ENOENT;
2231 if (!path->dentry->d_inode)
2232 goto exit_dput;
9e67f361
AV		 2233
2234 if (path->dentry->d_inode->i_op->follow_link)
fb1cc555 2235 return NULL;
fb1cc555
AV		 2236
2237 path_to_nameidata(path, nd);
31e6b01f 2238 nd->inode = path->dentry->d_inode;
a3fbbde7
AV		 2239 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2240 error = complete_walk(nd);
2241 if (error)
2242 goto exit;
fb1cc555 2243 error = -EISDIR;
31e6b01f 2244 if (S_ISDIR(nd->inode->i_mode))
fb1cc555 2245 goto exit;
67ee3ad2 2246ok:
6c0d46c4
AV		 2247 if (!S_ISREG(nd->inode->i_mode))
2248 will_truncate = 0;
2249
0f9d1a10
AV		 2250 if (will_truncate) {
2251 error = mnt_want_write(nd->path.mnt);
2252 if (error)
2253 goto exit;
ca344a89 2254 want_write = 1;
0f9d1a10 2255 }
ca344a89 2256common:
bcda7652 2257 error = may_open(&nd->path, acc_mode, open_flag);
ca344a89 2258 if (error)
0f9d1a10 2259 goto exit;
0f9d1a10
AV		 2260 filp = nameidata_to_filp(nd);
2261 if (!IS_ERR(filp)) {
2262 error = ima_file_check(filp, op->acc_mode);
2263 if (error) {
2264 fput(filp);
2265 filp = ERR_PTR(error);
2266 }
2267 }
2268 if (!IS_ERR(filp)) {
2269 if (will_truncate) {
2270 error = handle_truncate(filp);
2271 if (error) {
2272 fput(filp);
2273 filp = ERR_PTR(error);
2274 }
2275 }
2276 }
ca344a89
AV		 2277out:
2278 if (want_write)
0f9d1a10
AV		 2279 mnt_drop_write(nd->path.mnt);
2280 path_put(&nd->path);
fb1cc555
AV		 2281 return filp;
2282
2283exit_mutex_unlock:
2284 mutex_unlock(&dir->d_inode->i_mutex);
2285exit_dput:
2286 path_put_conditional(path, nd);
2287exit:
ca344a89
AV		 2288 filp = ERR_PTR(error);
2289 goto out;
fb1cc555
AV		 2290}
2291
13aab428 2292static struct file *path_openat(int dfd, const char *pathname,
73d049a4 2293 struct nameidata *nd, const struct open_flags *op, int flags)
1da177e4 2294{
fe2d35ff 2295 struct file *base = NULL;
4a3fd211 2296 struct file *filp;
9850c056 2297 struct path path;
13aab428 2298 int error;
31e6b01f
NP		 2299
2300 filp = get_empty_filp();
2301 if (!filp)
2302 return ERR_PTR(-ENFILE);
2303
47c805dc 2304 filp->f_flags = op->open_flag;
73d049a4
AV		 2305 nd->intent.open.file = filp;
2306 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2307 nd->intent.open.create_mode = op->mode;
31e6b01f 2308
73d049a4 2309 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
31e6b01f 2310 if (unlikely(error))
13aab428 2311 goto out_filp;
31e6b01f 2312
fe2d35ff 2313 current->total_link_count = 0;
73d049a4 2314 error = link_path_walk(pathname, nd);
31e6b01f
NP		 2315 if (unlikely(error))
2316 goto out_filp;
1da177e4 2317
73d049a4 2318 filp = do_last(nd, &path, op, pathname);
806b681c 2319 while (unlikely(!filp)) { /* trailing symlink */
7b9337aa 2320 struct path link = path;
def4af30 2321 void *cookie;
574197e0 2322 if (!(nd->flags & LOOKUP_FOLLOW)) {
73d049a4
AV		 2323 path_put_conditional(&path, nd);
2324 path_put(&nd->path);
40b39136
AV		 2325 filp = ERR_PTR(-ELOOP);
2326 break;
2327 }
73d049a4
AV		 2328 nd->flags |= LOOKUP_PARENT;
2329 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
574197e0 2330 error = follow_link(&link, nd, &cookie);
c3e380b0 2331 if (unlikely(error))
f1afe9ef 2332 filp = ERR_PTR(error);
c3e380b0 2333 else
73d049a4 2334 filp = do_last(nd, &path, op, pathname);
574197e0 2335 put_link(nd, &link, cookie);
806b681c 2336 }
10fa8e62 2337out:
73d049a4
AV		 2338 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2339 path_put(&nd->root);
fe2d35ff
AV		 2340 if (base)
2341 fput(base);
73d049a4 2342 release_open_intent(nd);
10fa8e62 2343 return filp;
1da177e4 2344
31e6b01f 2345out_filp:
806b681c 2346 filp = ERR_PTR(error);
10fa8e62 2347 goto out;
1da177e4
LT		 2348}
2349
13aab428
AV		 2350struct file *do_filp_open(int dfd, const char *pathname,
2351 const struct open_flags *op, int flags)
2352{
73d049a4 2353 struct nameidata nd;
13aab428
AV		 2354 struct file *filp;
2355
73d049a4 2356 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
13aab428 2357 if (unlikely(filp == ERR_PTR(-ECHILD)))
73d049a4 2358 filp = path_openat(dfd, pathname, &nd, op, flags);
13aab428 2359 if (unlikely(filp == ERR_PTR(-ESTALE)))
73d049a4 2360 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
13aab428
AV		 2361 return filp;
2362}
2363
73d049a4
AV		 2364struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2365 const char *name, const struct open_flags *op, int flags)
2366{
2367 struct nameidata nd;
2368 struct file *file;
2369
2370 nd.root.mnt = mnt;
2371 nd.root.dentry = dentry;
2372
2373 flags |= LOOKUP_ROOT;
2374
bcda7652 2375 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
73d049a4
AV		 2376 return ERR_PTR(-ELOOP);
2377
2378 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2379 if (unlikely(file == ERR_PTR(-ECHILD)))
2380 file = path_openat(-1, name, &nd, op, flags);
2381 if (unlikely(file == ERR_PTR(-ESTALE)))
2382 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2383 return file;
2384}
2385
ed75e95d 2386struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
1da177e4 2387{
c663e5d8 2388 struct dentry *dentry = ERR_PTR(-EEXIST);
ed75e95d
AV		 2389 struct nameidata nd;
2390 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2391 if (error)
2392 return ERR_PTR(error);
1da177e4 2393
c663e5d8
CH		 2394 /*
2395 * Yucky last component or no last component at all?
2396 * (foo/., foo/.., /////)
2397 */
ed75e95d
AV		 2398 if (nd.last_type != LAST_NORM)
2399 goto out;
2400 nd.flags &= ~LOOKUP_PARENT;
2401 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2402 nd.intent.open.flags = O_EXCL;
c663e5d8
CH		 2403
2404 /*
2405 * Do the final lookup.
2406 */
ed75e95d
AV		 2407 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2408 dentry = lookup_hash(&nd);
1da177e4
LT		 2409 if (IS_ERR(dentry))
2410 goto fail;
c663e5d8 2411
e9baf6e5
AV		 2412 if (dentry->d_inode)
2413 goto eexist;
c663e5d8
CH		 2414 /*
2415 * Special case - lookup gave negative, but... we had foo/bar/
2416 * From the vfs_mknod() POV we just have a negative dentry -
2417 * all is fine. Let's be bastards - you had / on the end, you've
2418 * been asking for (non-existent) directory. -ENOENT for you.
2419 */
ed75e95d 2420 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
e9baf6e5
AV		 2421 dput(dentry);
2422 dentry = ERR_PTR(-ENOENT);
ed75e95d 2423 goto fail;
e9baf6e5 2424 }
ed75e95d 2425 *path = nd.path;
1da177e4 2426 return dentry;
e9baf6e5 2427eexist:
1da177e4 2428 dput(dentry);
e9baf6e5 2429 dentry = ERR_PTR(-EEXIST);
1da177e4 2430fail:
ed75e95d
AV		 2431 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2432out:
2433 path_put(&nd.path);
1da177e4
LT		 2434 return dentry;
2435}
dae6ad8f
AV		 2436EXPORT_SYMBOL(kern_path_create);
2437
2438struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2439{
2440 char *tmp = getname(pathname);
2441 struct dentry *res;
2442 if (IS_ERR(tmp))
2443 return ERR_CAST(tmp);
2444 res = kern_path_create(dfd, tmp, path, is_dir);
2445 putname(tmp);
2446 return res;
2447}
2448EXPORT_SYMBOL(user_path_create);
2449
1a67aafb 2450int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
1da177e4 2451{
a95164d9 2452 int error = may_create(dir, dentry);
1da177e4
LT		 2453
2454 if (error)
2455 return error;
2456
e795b717
SH		 2457 if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
2458 !ns_capable(inode_userns(dir), CAP_MKNOD))
1da177e4
LT		 2459 return -EPERM;
2460
acfa4380 2461 if (!dir->i_op->mknod)
1da177e4
LT		 2462 return -EPERM;
2463
08ce5f16
SH		 2464 error = devcgroup_inode_mknod(mode, dev);
2465 if (error)
2466 return error;
2467
1da177e4
LT		 2468 error = security_inode_mknod(dir, dentry, mode, dev);
2469 if (error)
2470 return error;
2471
1da177e4 2472 error = dir->i_op->mknod(dir, dentry, mode, dev);
a74574aa 2473 if (!error)
f38aa942 2474 fsnotify_create(dir, dentry);
1da177e4
LT		 2475 return error;
2476}
2477
f69aac00 2478static int may_mknod(umode_t mode)
463c3197
DH		 2479{
2480 switch (mode & S_IFMT) {
2481 case S_IFREG:
2482 case S_IFCHR:
2483 case S_IFBLK:
2484 case S_IFIFO:
2485 case S_IFSOCK:
2486 case 0: /* zero mode translates to S_IFREG */
2487 return 0;
2488 case S_IFDIR:
2489 return -EPERM;
2490 default:
2491 return -EINVAL;
2492 }
2493}
2494
8208a22b 2495SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
2e4d0924 2496 unsigned, dev)
1da177e4 2497{
2ad94ae6 2498 struct dentry *dentry;
dae6ad8f
AV		 2499 struct path path;
2500 int error;
1da177e4
LT		 2501
2502 if (S_ISDIR(mode))
2503 return -EPERM;
1da177e4 2504
dae6ad8f
AV		 2505 dentry = user_path_create(dfd, filename, &path, 0);
2506 if (IS_ERR(dentry))
2507 return PTR_ERR(dentry);
2ad94ae6 2508
dae6ad8f 2509 if (!IS_POSIXACL(path.dentry->d_inode))
ce3b0f8d 2510 mode &= ~current_umask();
463c3197
DH		 2511 error = may_mknod(mode);
2512 if (error)
2513 goto out_dput;
dae6ad8f 2514 error = mnt_want_write(path.mnt);
463c3197
DH		 2515 if (error)
2516 goto out_dput;
dae6ad8f 2517 error = security_path_mknod(&path, dentry, mode, dev);
be6d3e56
KT		 2518 if (error)
2519 goto out_drop_write;
463c3197 2520 switch (mode & S_IFMT) {
1da177e4 2521 case 0: case S_IFREG:
dae6ad8f 2522 error = vfs_create(path.dentry->d_inode,dentry,mode,NULL);
1da177e4
LT		 2523 break;
2524 case S_IFCHR: case S_IFBLK:
dae6ad8f 2525 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
1da177e4
LT		 2526 new_decode_dev(dev));
2527 break;
2528 case S_IFIFO: case S_IFSOCK:
dae6ad8f 2529 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
1da177e4 2530 break;
1da177e4 2531 }
be6d3e56 2532out_drop_write:
dae6ad8f 2533 mnt_drop_write(path.mnt);
463c3197
DH		 2534out_dput:
2535 dput(dentry);
dae6ad8f
AV		 2536 mutex_unlock(&path.dentry->d_inode->i_mutex);
2537 path_put(&path);
1da177e4
LT		 2538
2539 return error;
2540}
2541
8208a22b 2542SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
5590ff0d
UD		 2543{
2544 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2545}
2546
18bb1db3 2547int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1da177e4 2548{
a95164d9 2549 int error = may_create(dir, dentry);
1da177e4
LT		 2550
2551 if (error)
2552 return error;
2553
acfa4380 2554 if (!dir->i_op->mkdir)
1da177e4
LT		 2555 return -EPERM;
2556
2557 mode &= (S_IRWXUGO|S_ISVTX);
2558 error = security_inode_mkdir(dir, dentry, mode);
2559 if (error)
2560 return error;
2561
1da177e4 2562 error = dir->i_op->mkdir(dir, dentry, mode);
a74574aa 2563 if (!error)
f38aa942 2564 fsnotify_mkdir(dir, dentry);
1da177e4
LT		 2565 return error;
2566}
2567
a218d0fd 2568SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
1da177e4 2569{
6902d925 2570 struct dentry *dentry;
dae6ad8f
AV		 2571 struct path path;
2572 int error;
1da177e4 2573
dae6ad8f 2574 dentry = user_path_create(dfd, pathname, &path, 1);
6902d925 2575 if (IS_ERR(dentry))
dae6ad8f 2576 return PTR_ERR(dentry);
1da177e4 2577
dae6ad8f 2578 if (!IS_POSIXACL(path.dentry->d_inode))
ce3b0f8d 2579 mode &= ~current_umask();
dae6ad8f 2580 error = mnt_want_write(path.mnt);
463c3197
DH		 2581 if (error)
2582 goto out_dput;
dae6ad8f 2583 error = security_path_mkdir(&path, dentry, mode);
be6d3e56
KT		 2584 if (error)
2585 goto out_drop_write;
dae6ad8f 2586 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
be6d3e56 2587out_drop_write:
dae6ad8f 2588 mnt_drop_write(path.mnt);
463c3197 2589out_dput:
6902d925 2590 dput(dentry);
dae6ad8f
AV		 2591 mutex_unlock(&path.dentry->d_inode->i_mutex);
2592 path_put(&path);
1da177e4
LT		 2593 return error;
2594}
2595
a218d0fd 2596SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
5590ff0d
UD		 2597{
2598 return sys_mkdirat(AT_FDCWD, pathname, mode);
2599}
2600
1da177e4 2601/*
a71905f0
SW		 2602 * The dentry_unhash() helper will try to drop the dentry early: we
2603 * should have a usage count of 2 if we're the only user of this
2604 * dentry, and if that is true (possibly after pruning the dcache),
2605 * then we drop the dentry now.
1da177e4
LT		 2606 *
2607 * A low-level filesystem can, if it choses, legally
2608 * do a
2609 *
2610 * if (!d_unhashed(dentry))
2611 * return -EBUSY;
2612 *
2613 * if it cannot handle the case of removing a directory
2614 * that is still in use by something else..
2615 */
2616void dentry_unhash(struct dentry *dentry)
2617{
dc168427 2618 shrink_dcache_parent(dentry);
1da177e4 2619 spin_lock(&dentry->d_lock);
64252c75 2620 if (dentry->d_count == 1)
1da177e4
LT		 2621 __d_drop(dentry);
2622 spin_unlock(&dentry->d_lock);
1da177e4
LT		 2623}
2624
2625int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2626{
2627 int error = may_delete(dir, dentry, 1);
2628
2629 if (error)
2630 return error;
2631
acfa4380 2632 if (!dir->i_op->rmdir)
1da177e4
LT		 2633 return -EPERM;
2634
1d2ef590 2635 dget(dentry);
1b1dcc1b 2636 mutex_lock(&dentry->d_inode->i_mutex);
912dbc15
SW		 2637
2638 error = -EBUSY;
1da177e4 2639 if (d_mountpoint(dentry))
912dbc15
SW		 2640 goto out;
2641
2642 error = security_inode_rmdir(dir, dentry);
2643 if (error)
2644 goto out;
2645
3cebde24 2646 shrink_dcache_parent(dentry);
912dbc15
SW		 2647 error = dir->i_op->rmdir(dir, dentry);
2648 if (error)
2649 goto out;
2650
2651 dentry->d_inode->i_flags |= S_DEAD;
2652 dont_mount(dentry);
2653
2654out:
1b1dcc1b 2655 mutex_unlock(&dentry->d_inode->i_mutex);
1d2ef590 2656 dput(dentry);
912dbc15 2657 if (!error)
1da177e4 2658 d_delete(dentry);
1da177e4
LT		 2659 return error;
2660}
2661
5590ff0d 2662static long do_rmdir(int dfd, const char __user *pathname)
1da177e4
LT		 2663{
2664 int error = 0;
2665 char * name;
2666 struct dentry *dentry;
2667 struct nameidata nd;
2668
2ad94ae6 2669 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2670 if (error)
2ad94ae6 2671 return error;
1da177e4
LT		 2672
2673 switch(nd.last_type) {
0612d9fb
OH		 2674 case LAST_DOTDOT:
2675 error = -ENOTEMPTY;
2676 goto exit1;
2677 case LAST_DOT:
2678 error = -EINVAL;
2679 goto exit1;
2680 case LAST_ROOT:
2681 error = -EBUSY;
2682 goto exit1;
1da177e4 2683 }
0612d9fb
OH		 2684
2685 nd.flags &= ~LOOKUP_PARENT;
2686
4ac91378 2687 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2688 dentry = lookup_hash(&nd);
1da177e4 2689 error = PTR_ERR(dentry);
6902d925
DH		 2690 if (IS_ERR(dentry))
2691 goto exit2;
e6bc45d6
TT		 2692 if (!dentry->d_inode) {
2693 error = -ENOENT;
2694 goto exit3;
2695 }
0622753b
DH		 2696 error = mnt_want_write(nd.path.mnt);
2697 if (error)
2698 goto exit3;
be6d3e56
KT		 2699 error = security_path_rmdir(&nd.path, dentry);
2700 if (error)
2701 goto exit4;
4ac91378 2702 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
be6d3e56 2703exit4:
0622753b
DH		 2704 mnt_drop_write(nd.path.mnt);
2705exit3:
6902d925
DH		 2706 dput(dentry);
2707exit2:
4ac91378 2708 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2709exit1:
1d957f9b 2710 path_put(&nd.path);
1da177e4
LT		 2711 putname(name);
2712 return error;
2713}
2714
3cdad428 2715SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
5590ff0d
UD		 2716{
2717 return do_rmdir(AT_FDCWD, pathname);
2718}
2719
1da177e4
LT		 2720int vfs_unlink(struct inode *dir, struct dentry *dentry)
2721{
2722 int error = may_delete(dir, dentry, 0);
2723
2724 if (error)
2725 return error;
2726
acfa4380 2727 if (!dir->i_op->unlink)
1da177e4
LT		 2728 return -EPERM;
2729
1b1dcc1b 2730 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2731 if (d_mountpoint(dentry))
2732 error = -EBUSY;
2733 else {
2734 error = security_inode_unlink(dir, dentry);
bec1052e 2735 if (!error) {
1da177e4 2736 error = dir->i_op->unlink(dir, dentry);
bec1052e 2737 if (!error)
d83c49f3 2738 dont_mount(dentry);
bec1052e 2739 }
1da177e4 2740 }
1b1dcc1b 2741 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2742
2743 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2744 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
ece95912 2745 fsnotify_link_count(dentry->d_inode);
e234f35c 2746 d_delete(dentry);
1da177e4 2747 }
0eeca283 2748
1da177e4
LT		 2749 return error;
2750}
2751
2752/*
2753 * Make sure that the actual truncation of the file will occur outside its
1b1dcc1b 2754 * directory's i_mutex. Truncate can take a long time if there is a lot of
1da177e4
LT		 2755 * writeout happening, and we don't want to prevent access to the directory
2756 * while waiting on the I/O.
2757 */
5590ff0d 2758static long do_unlinkat(int dfd, const char __user *pathname)
1da177e4 2759{
2ad94ae6
AV		 2760 int error;
2761 char *name;
1da177e4
LT		 2762 struct dentry *dentry;
2763 struct nameidata nd;
2764 struct inode *inode = NULL;
2765
2ad94ae6 2766 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2767 if (error)
2ad94ae6
AV		 2768 return error;
2769
1da177e4
LT		 2770 error = -EISDIR;
2771 if (nd.last_type != LAST_NORM)
2772 goto exit1;
0612d9fb
OH		 2773
2774 nd.flags &= ~LOOKUP_PARENT;
2775
4ac91378 2776 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2777 dentry = lookup_hash(&nd);
1da177e4
LT		 2778 error = PTR_ERR(dentry);
2779 if (!IS_ERR(dentry)) {
2780 /* Why not before? Because we want correct error value */
50338b88
TE		 2781 if (nd.last.name[nd.last.len])
2782 goto slashes;
1da177e4 2783 inode = dentry->d_inode;
50338b88 2784 if (!inode)
e6bc45d6
TT		 2785 goto slashes;
2786 ihold(inode);
0622753b
DH		 2787 error = mnt_want_write(nd.path.mnt);
2788 if (error)
2789 goto exit2;
be6d3e56
KT		 2790 error = security_path_unlink(&nd.path, dentry);
2791 if (error)
2792 goto exit3;
4ac91378 2793 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
be6d3e56 2794exit3:
0622753b 2795 mnt_drop_write(nd.path.mnt);
1da177e4
LT		 2796 exit2:
2797 dput(dentry);
2798 }
4ac91378 2799 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4
LT		 2800 if (inode)
2801 iput(inode); /* truncate the inode here */
2802exit1:
1d957f9b 2803 path_put(&nd.path);
1da177e4
LT		 2804 putname(name);
2805 return error;
2806
2807slashes:
2808 error = !dentry->d_inode ? -ENOENT :
2809 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2810 goto exit2;
2811}
2812
2e4d0924 2813SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
5590ff0d
UD		 2814{
2815 if ((flag & ~AT_REMOVEDIR) != 0)
2816 return -EINVAL;
2817
2818 if (flag & AT_REMOVEDIR)
2819 return do_rmdir(dfd, pathname);
2820
2821 return do_unlinkat(dfd, pathname);
2822}
2823
3480b257 2824SYSCALL_DEFINE1(unlink, const char __user *, pathname)
5590ff0d
UD		 2825{
2826 return do_unlinkat(AT_FDCWD, pathname);
2827}
2828
db2e747b 2829int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
1da177e4 2830{
a95164d9 2831 int error = may_create(dir, dentry);
1da177e4
LT		 2832
2833 if (error)
2834 return error;
2835
acfa4380 2836 if (!dir->i_op->symlink)
1da177e4
LT		 2837 return -EPERM;
2838
2839 error = security_inode_symlink(dir, dentry, oldname);
2840 if (error)
2841 return error;
2842
1da177e4 2843 error = dir->i_op->symlink(dir, dentry, oldname);
a74574aa 2844 if (!error)
f38aa942 2845 fsnotify_create(dir, dentry);
1da177e4
LT		 2846 return error;
2847}
2848
2e4d0924
HC		 2849SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2850 int, newdfd, const char __user *, newname)
1da177e4 2851{
2ad94ae6
AV		 2852 int error;
2853 char *from;
6902d925 2854 struct dentry *dentry;
dae6ad8f 2855 struct path path;
1da177e4
LT		 2856
2857 from = getname(oldname);
2ad94ae6 2858 if (IS_ERR(from))
1da177e4 2859 return PTR_ERR(from);
1da177e4 2860
dae6ad8f 2861 dentry = user_path_create(newdfd, newname, &path, 0);
6902d925
DH		 2862 error = PTR_ERR(dentry);
2863 if (IS_ERR(dentry))
dae6ad8f 2864 goto out_putname;
6902d925 2865
dae6ad8f 2866 error = mnt_want_write(path.mnt);
75c3f29d
DH		 2867 if (error)
2868 goto out_dput;
dae6ad8f 2869 error = security_path_symlink(&path, dentry, from);
be6d3e56
KT		 2870 if (error)
2871 goto out_drop_write;
dae6ad8f 2872 error = vfs_symlink(path.dentry->d_inode, dentry, from);
be6d3e56 2873out_drop_write:
dae6ad8f 2874 mnt_drop_write(path.mnt);
75c3f29d 2875out_dput:
6902d925 2876 dput(dentry);
dae6ad8f
AV		 2877 mutex_unlock(&path.dentry->d_inode->i_mutex);
2878 path_put(&path);
6902d925 2879out_putname:
1da177e4
LT		 2880 putname(from);
2881 return error;
2882}
2883
3480b257 2884SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 2885{
2886 return sys_symlinkat(oldname, AT_FDCWD, newname);
2887}
2888
1da177e4
LT		 2889int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2890{
2891 struct inode *inode = old_dentry->d_inode;
2892 int error;
2893
2894 if (!inode)
2895 return -ENOENT;
2896
a95164d9 2897 error = may_create(dir, new_dentry);
1da177e4
LT		 2898 if (error)
2899 return error;
2900
2901 if (dir->i_sb != inode->i_sb)
2902 return -EXDEV;
2903
2904 /*
2905 * A link to an append-only or immutable file cannot be created.
2906 */
2907 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2908 return -EPERM;
acfa4380 2909 if (!dir->i_op->link)
1da177e4 2910 return -EPERM;
7e79eedb 2911 if (S_ISDIR(inode->i_mode))
1da177e4
LT		 2912 return -EPERM;
2913
2914 error = security_inode_link(old_dentry, dir, new_dentry);
2915 if (error)
2916 return error;
2917
7e79eedb 2918 mutex_lock(&inode->i_mutex);
aae8a97d
AK		 2919 /* Make sure we don't allow creating hardlink to an unlinked file */
2920 if (inode->i_nlink == 0)
2921 error = -ENOENT;
2922 else
2923 error = dir->i_op->link(old_dentry, dir, new_dentry);
7e79eedb 2924 mutex_unlock(&inode->i_mutex);
e31e14ec 2925 if (!error)
7e79eedb 2926 fsnotify_link(dir, inode, new_dentry);
1da177e4
LT		 2927 return error;
2928}
2929
2930/*
2931 * Hardlinks are often used in delicate situations. We avoid
2932 * security-related surprises by not following symlinks on the
2933 * newname. --KAB
2934 *
2935 * We don't follow them on the oldname either to be compatible
2936 * with linux 2.0, and to avoid hard-linking to directories
2937 * and other special files. --ADM
2938 */
2e4d0924
HC		 2939SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2940 int, newdfd, const char __user *, newname, int, flags)
1da177e4
LT		 2941{
2942 struct dentry *new_dentry;
dae6ad8f 2943 struct path old_path, new_path;
11a7b371 2944 int how = 0;
1da177e4 2945 int error;
1da177e4 2946
11a7b371 2947 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
c04030e1 2948 return -EINVAL;
11a7b371
AK		 2949 /*
2950 * To use null names we require CAP_DAC_READ_SEARCH
2951 * This ensures that not everyone will be able to create
2952 * handlink using the passed filedescriptor.
2953 */
2954 if (flags & AT_EMPTY_PATH) {
2955 if (!capable(CAP_DAC_READ_SEARCH))
2956 return -ENOENT;
2957 how = LOOKUP_EMPTY;
2958 }
2959
2960 if (flags & AT_SYMLINK_FOLLOW)
2961 how |= LOOKUP_FOLLOW;
c04030e1 2962
11a7b371 2963 error = user_path_at(olddfd, oldname, how, &old_path);
1da177e4 2964 if (error)
2ad94ae6
AV		 2965 return error;
2966
dae6ad8f 2967 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
1da177e4 2968 error = PTR_ERR(new_dentry);
6902d925 2969 if (IS_ERR(new_dentry))
dae6ad8f
AV		 2970 goto out;
2971
2972 error = -EXDEV;
2973 if (old_path.mnt != new_path.mnt)
2974 goto out_dput;
2975 error = mnt_want_write(new_path.mnt);
75c3f29d
DH		 2976 if (error)
2977 goto out_dput;
dae6ad8f 2978 error = security_path_link(old_path.dentry, &new_path, new_dentry);
be6d3e56
KT		 2979 if (error)
2980 goto out_drop_write;
dae6ad8f 2981 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
be6d3e56 2982out_drop_write:
dae6ad8f 2983 mnt_drop_write(new_path.mnt);
75c3f29d 2984out_dput:
6902d925 2985 dput(new_dentry);
dae6ad8f
AV		 2986 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
2987 path_put(&new_path);
1da177e4 2988out:
2d8f3038 2989 path_put(&old_path);
1da177e4
LT		 2990
2991 return error;
2992}
2993
3480b257 2994SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
5590ff0d 2995{
c04030e1 2996 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
5590ff0d
UD		 2997}
2998
1da177e4
LT		 2999/*
3000 * The worst of all namespace operations - renaming directory. "Perverted"
3001 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3002 * Problems:
3003 * a) we can get into loop creation. Check is done in is_subdir().
3004 * b) race potential - two innocent renames can create a loop together.
3005 * That's where 4.4 screws up. Current fix: serialization on
a11f3a05 3006 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
1da177e4
LT		 3007 * story.
3008 * c) we have to lock _three_ objects - parents and victim (if it exists).
1b1dcc1b 3009 * And that - after we got ->i_mutex on parents (until then we don't know
1da177e4
LT		 3010 * whether the target exists). Solution: try to be smart with locking
3011 * order for inodes. We rely on the fact that tree topology may change
a11f3a05 3012 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
1da177e4
LT		 3013 * move will be locked. Thus we can rank directories by the tree
3014 * (ancestors first) and rank all non-directories after them.
3015 * That works since everybody except rename does "lock parent, lookup,
a11f3a05 3016 * lock child" and rename is under ->s_vfs_rename_mutex.
1da177e4
LT		 3017 * HOWEVER, it relies on the assumption that any object with ->lookup()
3018 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3019 * we'd better make sure that there's no link(2) for them.
e4eaac06 3020 * d) conversion from fhandle to dentry may come in the wrong moment - when
1b1dcc1b 3021 * we are removing the target. Solution: we will have to grab ->i_mutex
1da177e4 3022 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
c41b20e7 3023 * ->i_mutex on parents, which works but leads to some truly excessive
1da177e4
LT		 3024 * locking].
3025 */
75c96f85
AB		 3026static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3027 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3028{
3029 int error = 0;
9055cba7 3030 struct inode *target = new_dentry->d_inode;
1da177e4
LT		 3031
3032 /*
3033 * If we are going to change the parent - check write permissions,
3034 * we'll need to flip '..'.
3035 */
3036 if (new_dir != old_dir) {
f419a2e3 3037 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
1da177e4
LT		 3038 if (error)
3039 return error;
3040 }
3041
3042 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3043 if (error)
3044 return error;
3045
1d2ef590 3046 dget(new_dentry);
d83c49f3 3047 if (target)
1b1dcc1b 3048 mutex_lock(&target->i_mutex);
9055cba7
SW		 3049
3050 error = -EBUSY;
3051 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3052 goto out;
3053
3cebde24
SW		 3054 if (target)
3055 shrink_dcache_parent(new_dentry);
9055cba7
SW		 3056 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3057 if (error)
3058 goto out;
3059
1da177e4 3060 if (target) {
9055cba7
SW		 3061 target->i_flags |= S_DEAD;
3062 dont_mount(new_dentry);
1da177e4 3063 }
9055cba7
SW		 3064out:
3065 if (target)
3066 mutex_unlock(&target->i_mutex);
1d2ef590 3067 dput(new_dentry);
e31e14ec 3068 if (!error)
349457cc
MF		 3069 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3070 d_move(old_dentry,new_dentry);
1da177e4
LT		 3071 return error;
3072}
3073
75c96f85
AB		 3074static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3075 struct inode *new_dir, struct dentry *new_dentry)
1da177e4 3076{
51892bbb 3077 struct inode *target = new_dentry->d_inode;
1da177e4
LT		 3078 int error;
3079
3080 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3081 if (error)
3082 return error;
3083
3084 dget(new_dentry);
1da177e4 3085 if (target)
1b1dcc1b 3086 mutex_lock(&target->i_mutex);
51892bbb
SW		 3087
3088 error = -EBUSY;
1da177e4 3089 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
51892bbb
SW		 3090 goto out;
3091
3092 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3093 if (error)
3094 goto out;
3095
3096 if (target)
3097 dont_mount(new_dentry);
3098 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3099 d_move(old_dentry, new_dentry);
3100out:
1da177e4 3101 if (target)
1b1dcc1b 3102 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3103 dput(new_dentry);
3104 return error;
3105}
3106
3107int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3108 struct inode *new_dir, struct dentry *new_dentry)
3109{
3110 int error;
3111 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
59b0df21 3112 const unsigned char *old_name;
1da177e4
LT		 3113
3114 if (old_dentry->d_inode == new_dentry->d_inode)
3115 return 0;
3116
3117 error = may_delete(old_dir, old_dentry, is_dir);
3118 if (error)
3119 return error;
3120
3121 if (!new_dentry->d_inode)
a95164d9 3122 error = may_create(new_dir, new_dentry);
1da177e4
LT		 3123 else
3124 error = may_delete(new_dir, new_dentry, is_dir);
3125 if (error)
3126 return error;
3127
acfa4380 3128 if (!old_dir->i_op->rename)
1da177e4
LT		 3129 return -EPERM;
3130
0eeca283
RL		 3131 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3132
1da177e4
LT		 3133 if (is_dir)
3134 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3135 else
3136 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
123df294
AV		 3137 if (!error)
3138 fsnotify_move(old_dir, new_dir, old_name, is_dir,
5a190ae6 3139 new_dentry->d_inode, old_dentry);
0eeca283
RL		 3140 fsnotify_oldname_free(old_name);
3141
1da177e4
LT		 3142 return error;
3143}
3144
2e4d0924
HC		 3145SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3146 int, newdfd, const char __user *, newname)
1da177e4 3147{
2ad94ae6
AV		 3148 struct dentry *old_dir, *new_dir;
3149 struct dentry *old_dentry, *new_dentry;
3150 struct dentry *trap;
1da177e4 3151 struct nameidata oldnd, newnd;
2ad94ae6
AV		 3152 char *from;
3153 char *to;
3154 int error;
1da177e4 3155
2ad94ae6 3156 error = user_path_parent(olddfd, oldname, &oldnd, &from);
1da177e4
LT		 3157 if (error)
3158 goto exit;
3159
2ad94ae6 3160 error = user_path_parent(newdfd, newname, &newnd, &to);
1da177e4
LT		 3161 if (error)
3162 goto exit1;
3163
3164 error = -EXDEV;
4ac91378 3165 if (oldnd.path.mnt != newnd.path.mnt)
1da177e4
LT		 3166 goto exit2;
3167
4ac91378 3168 old_dir = oldnd.path.dentry;
1da177e4
LT		 3169 error = -EBUSY;
3170 if (oldnd.last_type != LAST_NORM)
3171 goto exit2;
3172
4ac91378 3173 new_dir = newnd.path.dentry;
1da177e4
LT		 3174 if (newnd.last_type != LAST_NORM)
3175 goto exit2;
3176
0612d9fb
OH		 3177 oldnd.flags &= ~LOOKUP_PARENT;
3178 newnd.flags &= ~LOOKUP_PARENT;
4e9ed2f8 3179 newnd.flags |= LOOKUP_RENAME_TARGET;
0612d9fb 3180
1da177e4
LT		 3181 trap = lock_rename(new_dir, old_dir);
3182
49705b77 3183 old_dentry = lookup_hash(&oldnd);
1da177e4
LT		 3184 error = PTR_ERR(old_dentry);
3185 if (IS_ERR(old_dentry))
3186 goto exit3;
3187 /* source must exist */
3188 error = -ENOENT;
3189 if (!old_dentry->d_inode)
3190 goto exit4;
3191 /* unless the source is a directory trailing slashes give -ENOTDIR */
3192 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3193 error = -ENOTDIR;
3194 if (oldnd.last.name[oldnd.last.len])
3195 goto exit4;
3196 if (newnd.last.name[newnd.last.len])
3197 goto exit4;
3198 }
3199 /* source should not be ancestor of target */
3200 error = -EINVAL;
3201 if (old_dentry == trap)
3202 goto exit4;
49705b77 3203 new_dentry = lookup_hash(&newnd);
1da177e4
LT		 3204 error = PTR_ERR(new_dentry);
3205 if (IS_ERR(new_dentry))
3206 goto exit4;
3207 /* target should not be an ancestor of source */
3208 error = -ENOTEMPTY;
3209 if (new_dentry == trap)
3210 goto exit5;
3211
9079b1eb
DH		 3212 error = mnt_want_write(oldnd.path.mnt);
3213 if (error)
3214 goto exit5;
be6d3e56
KT		 3215 error = security_path_rename(&oldnd.path, old_dentry,
3216 &newnd.path, new_dentry);
3217 if (error)
3218 goto exit6;
1da177e4
LT		 3219 error = vfs_rename(old_dir->d_inode, old_dentry,
3220 new_dir->d_inode, new_dentry);
be6d3e56 3221exit6:
9079b1eb 3222 mnt_drop_write(oldnd.path.mnt);
1da177e4
LT		 3223exit5:
3224 dput(new_dentry);
3225exit4:
3226 dput(old_dentry);
3227exit3:
3228 unlock_rename(new_dir, old_dir);
3229exit2:
1d957f9b 3230 path_put(&newnd.path);
2ad94ae6 3231 putname(to);
1da177e4 3232exit1:
1d957f9b 3233 path_put(&oldnd.path);
1da177e4 3234 putname(from);
2ad94ae6 3235exit:
1da177e4
LT		 3236 return error;
3237}
3238
a26eab24 3239SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 3240{
3241 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3242}
3243
1da177e4
LT		 3244int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3245{
3246 int len;
3247
3248 len = PTR_ERR(link);
3249 if (IS_ERR(link))
3250 goto out;
3251
3252 len = strlen(link);
3253 if (len > (unsigned) buflen)
3254 len = buflen;
3255 if (copy_to_user(buffer, link, len))
3256 len = -EFAULT;
3257out:
3258 return len;
3259}
3260
3261/*
3262 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3263 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3264 * using) it for any given inode is up to filesystem.
3265 */
3266int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3267{
3268 struct nameidata nd;
cc314eef 3269 void *cookie;
694a1764 3270 int res;
cc314eef 3271
1da177e4 3272 nd.depth = 0;
cc314eef 3273 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
694a1764
MS		 3274 if (IS_ERR(cookie))
3275 return PTR_ERR(cookie);
3276
3277 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3278 if (dentry->d_inode->i_op->put_link)
3279 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3280 return res;
1da177e4
LT		 3281}
3282
3283int vfs_follow_link(struct nameidata *nd, const char *link)
3284{
3285 return __vfs_follow_link(nd, link);
3286}
3287
3288/* get the link contents into pagecache */
3289static char *page_getlink(struct dentry * dentry, struct page **ppage)
3290{
ebd09abb
DG		 3291 char *kaddr;
3292 struct page *page;
1da177e4 3293 struct address_space *mapping = dentry->d_inode->i_mapping;
090d2b18 3294 page = read_mapping_page(mapping, 0, NULL);
1da177e4 3295 if (IS_ERR(page))
6fe6900e 3296 return (char*)page;
1da177e4 3297 *ppage = page;
ebd09abb
DG		 3298 kaddr = kmap(page);
3299 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3300 return kaddr;
1da177e4
LT		 3301}
3302
3303int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3304{
3305 struct page *page = NULL;
3306 char *s = page_getlink(dentry, &page);
3307 int res = vfs_readlink(dentry,buffer,buflen,s);
3308 if (page) {
3309 kunmap(page);
3310 page_cache_release(page);
3311 }
3312 return res;
3313}
3314
cc314eef 3315void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
1da177e4 3316{
cc314eef 3317 struct page *page = NULL;
1da177e4 3318 nd_set_link(nd, page_getlink(dentry, &page));
cc314eef 3319 return page;
1da177e4
LT		 3320}
3321
cc314eef 3322void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1da177e4 3323{
cc314eef
LT		 3324 struct page *page = cookie;
3325
3326 if (page) {
1da177e4
LT		 3327 kunmap(page);
3328 page_cache_release(page);
1da177e4
LT		 3329 }
3330}
3331
54566b2c
NP		 3332/*
3333 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3334 */
3335int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
1da177e4
LT		 3336{
3337 struct address_space *mapping = inode->i_mapping;
0adb25d2 3338 struct page *page;
afddba49 3339 void *fsdata;
beb497ab 3340 int err;
1da177e4 3341 char *kaddr;
54566b2c
NP		 3342 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3343 if (nofs)
3344 flags |= AOP_FLAG_NOFS;
1da177e4 3345
7e53cac4 3346retry:
afddba49 3347 err = pagecache_write_begin(NULL, mapping, 0, len-1,
54566b2c 3348 flags, &page, &fsdata);
1da177e4 3349 if (err)
afddba49
NP		 3350 goto fail;
3351
1da177e4
LT		 3352 kaddr = kmap_atomic(page, KM_USER0);
3353 memcpy(kaddr, symname, len-1);
3354 kunmap_atomic(kaddr, KM_USER0);
afddba49
NP		 3355
3356 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3357 page, fsdata);
1da177e4
LT		 3358 if (err < 0)
3359 goto fail;
afddba49
NP		 3360 if (err < len-1)
3361 goto retry;
3362
1da177e4
LT		 3363 mark_inode_dirty(inode);
3364 return 0;
1da177e4
LT		 3365fail:
3366 return err;
3367}
3368
0adb25d2
KK		 3369int page_symlink(struct inode *inode, const char *symname, int len)
3370{
3371 return __page_symlink(inode, symname, len,
54566b2c 3372 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
0adb25d2
KK		 3373}
3374
92e1d5be 3375const struct inode_operations page_symlink_inode_operations = {
1da177e4
LT		 3376 .readlink = generic_readlink,
3377 .follow_link = page_follow_link_light,
3378 .put_link = page_put_link,
3379};
3380
2d8f3038 3381EXPORT_SYMBOL(user_path_at);
cc53ce53 3382EXPORT_SYMBOL(follow_down_one);
1da177e4
LT		 3383EXPORT_SYMBOL(follow_down);
3384EXPORT_SYMBOL(follow_up);
3385EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3386EXPORT_SYMBOL(getname);
3387EXPORT_SYMBOL(lock_rename);
1da177e4
LT		 3388EXPORT_SYMBOL(lookup_one_len);
3389EXPORT_SYMBOL(page_follow_link_light);
3390EXPORT_SYMBOL(page_put_link);
3391EXPORT_SYMBOL(page_readlink);
0adb25d2 3392EXPORT_SYMBOL(__page_symlink);
1da177e4
LT		 3393EXPORT_SYMBOL(page_symlink);
3394EXPORT_SYMBOL(page_symlink_inode_operations);
d1811465 3395EXPORT_SYMBOL(kern_path);
16f18200 3396EXPORT_SYMBOL(vfs_path_lookup);
f419a2e3 3397EXPORT_SYMBOL(inode_permission);
1da177e4
LT		 3398EXPORT_SYMBOL(unlock_rename);
3399EXPORT_SYMBOL(vfs_create);
3400EXPORT_SYMBOL(vfs_follow_link);
3401EXPORT_SYMBOL(vfs_link);
3402EXPORT_SYMBOL(vfs_mkdir);
3403EXPORT_SYMBOL(vfs_mknod);
3404EXPORT_SYMBOL(generic_permission);
3405EXPORT_SYMBOL(vfs_readlink);
3406EXPORT_SYMBOL(vfs_rename);
3407EXPORT_SYMBOL(vfs_rmdir);
3408EXPORT_SYMBOL(vfs_symlink);
3409EXPORT_SYMBOL(vfs_unlink);
3410EXPORT_SYMBOL(dentry_unhash);
3411EXPORT_SYMBOL(generic_readlink);
kernel source for telekom puls (alcatel/mediatek)