projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'hwmon-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/groeck...
[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
0145acc2
LT		 1377unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1378{
1379 unsigned long hash = init_name_hash();
1380 while (len--)
1381 hash = partial_name_hash(*name++, hash);
1382 return end_name_hash(hash);
1383}
1384
200e9ef7
LT		 1385/*
1386 * We know there's a real path component here of at least
1387 * one character.
1388 */
1389static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1390{
1391 unsigned long hash = init_name_hash();
1392 unsigned long len = 0, c;
1393
1394 c = (unsigned char)*name;
1395 do {
1396 len++;
1397 hash = partial_name_hash(c, hash);
1398 c = (unsigned char)name[len];
1399 } while (c && c != '/');
1400 *hashp = end_name_hash(hash);
1401 return len;
1402}
1403
1da177e4
LT		 1404/*
1405 * Name resolution.
ea3834d9
PM		 1406 * This is the basic name resolution function, turning a pathname into
1407 * the final dentry. We expect 'base' to be positive and a directory.
1da177e4 1408 *
ea3834d9
PM		 1409 * Returns 0 and nd will have valid dentry and mnt on success.
1410 * Returns error and drops reference to input namei data on failure.
1da177e4 1411 */
6de88d72 1412static int link_path_walk(const char *name, struct nameidata *nd)
1da177e4
LT		 1413{
1414 struct path next;
1da177e4 1415 int err;
1da177e4
LT		 1416
1417 while (*name=='/')
1418 name++;
1419 if (!*name)
086e183a 1420 return 0;
1da177e4 1421
1da177e4
LT		 1422 /* At this point we know we have a real path component. */
1423 for(;;) {
1da177e4 1424 struct qstr this;
200e9ef7 1425 long len;
fe479a58 1426 int type;
1da177e4 1427
52094c8a 1428 err = may_lookup(nd);
1da177e4
LT		 1429 if (err)
1430 break;
1431
200e9ef7 1432 len = hash_name(name, &this.hash);
1da177e4 1433 this.name = name;
200e9ef7 1434 this.len = len;
1da177e4 1435
fe479a58 1436 type = LAST_NORM;
200e9ef7 1437 if (name[0] == '.') switch (len) {
fe479a58 1438 case 2:
200e9ef7 1439 if (name[1] == '.') {
fe479a58 1440 type = LAST_DOTDOT;
16c2cd71
AV		 1441 nd->flags |= LOOKUP_JUMPED;
1442 }
fe479a58
AV		 1443 break;
1444 case 1:
1445 type = LAST_DOT;
1446 }
5a202bcd
AV		 1447 if (likely(type == LAST_NORM)) {
1448 struct dentry *parent = nd->path.dentry;
16c2cd71 1449 nd->flags &= ~LOOKUP_JUMPED;
5a202bcd
AV		 1450 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1451 err = parent->d_op->d_hash(parent, nd->inode,
1452 &this);
1453 if (err < 0)
1454 break;
1455 }
1456 }
fe479a58 1457
200e9ef7 1458 if (!name[len])
1da177e4 1459 goto last_component;
200e9ef7
LT		 1460 /*
1461 * If it wasn't NUL, we know it was '/'. Skip that
1462 * slash, and continue until no more slashes.
1463 */
1464 do {
1465 len++;
1466 } while (unlikely(name[len] == '/'));
1467 if (!name[len])
b356379a 1468 goto last_component;
200e9ef7 1469 name += len;
1da177e4 1470
ce57dfc1
AV		 1471 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1472 if (err < 0)
1473 return err;
1da177e4 1474
ce57dfc1 1475 if (err) {
b356379a 1476 err = nested_symlink(&next, nd);
1da177e4 1477 if (err)
a7472bab 1478 return err;
31e6b01f 1479 }
3ddcd056
LT		 1480 if (can_lookup(nd->inode))
1481 continue;
1da177e4 1482 err = -ENOTDIR;
3ddcd056 1483 break;
1da177e4
LT		 1484 /* here ends the main loop */
1485
1da177e4 1486last_component:
b356379a
AV		 1487 nd->last = this;
1488 nd->last_type = type;
086e183a 1489 return 0;
1da177e4 1490 }
951361f9 1491 terminate_walk(nd);
1da177e4
LT		 1492 return err;
1493}
1494
70e9b357
AV		 1495static int path_init(int dfd, const char *name, unsigned int flags,
1496 struct nameidata *nd, struct file **fp)
31e6b01f
NP		 1497{
1498 int retval = 0;
1499 int fput_needed;
1500 struct file *file;
1501
1502 nd->last_type = LAST_ROOT; /* if there are only slashes... */
16c2cd71 1503 nd->flags = flags | LOOKUP_JUMPED;
31e6b01f 1504 nd->depth = 0;
5b6ca027
AV		 1505 if (flags & LOOKUP_ROOT) {
1506 struct inode *inode = nd->root.dentry->d_inode;
73d049a4
AV		 1507 if (*name) {
1508 if (!inode->i_op->lookup)
1509 return -ENOTDIR;
1510 retval = inode_permission(inode, MAY_EXEC);
1511 if (retval)
1512 return retval;
1513 }
5b6ca027
AV		 1514 nd->path = nd->root;
1515 nd->inode = inode;
1516 if (flags & LOOKUP_RCU) {
1517 br_read_lock(vfsmount_lock);
1518 rcu_read_lock();
1519 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1520 } else {
1521 path_get(&nd->path);
1522 }
1523 return 0;
1524 }
1525
31e6b01f 1526 nd->root.mnt = NULL;
31e6b01f
NP		 1527
1528 if (*name=='/') {
e41f7d4e
AV		 1529 if (flags & LOOKUP_RCU) {
1530 br_read_lock(vfsmount_lock);
1531 rcu_read_lock();
1532 set_root_rcu(nd);
1533 } else {
1534 set_root(nd);
1535 path_get(&nd->root);
1536 }
1537 nd->path = nd->root;
31e6b01f 1538 } else if (dfd == AT_FDCWD) {
e41f7d4e
AV		 1539 if (flags & LOOKUP_RCU) {
1540 struct fs_struct *fs = current->fs;
1541 unsigned seq;
31e6b01f 1542
e41f7d4e
AV		 1543 br_read_lock(vfsmount_lock);
1544 rcu_read_lock();
c28cc364 1545
e41f7d4e
AV		 1546 do {
1547 seq = read_seqcount_begin(&fs->seq);
1548 nd->path = fs->pwd;
1549 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1550 } while (read_seqcount_retry(&fs->seq, seq));
1551 } else {
1552 get_fs_pwd(current->fs, &nd->path);
1553 }
31e6b01f
NP		 1554 } else {
1555 struct dentry *dentry;
1556
1abf0c71 1557 file = fget_raw_light(dfd, &fput_needed);
31e6b01f
NP		 1558 retval = -EBADF;
1559 if (!file)
1560 goto out_fail;
1561
1562 dentry = file->f_path.dentry;
1563
f52e0c11
AV		 1564 if (*name) {
1565 retval = -ENOTDIR;
1566 if (!S_ISDIR(dentry->d_inode->i_mode))
1567 goto fput_fail;
31e6b01f 1568
4ad5abb3 1569 retval = inode_permission(dentry->d_inode, MAY_EXEC);
f52e0c11
AV		 1570 if (retval)
1571 goto fput_fail;
1572 }
31e6b01f
NP		 1573
1574 nd->path = file->f_path;
e41f7d4e
AV		 1575 if (flags & LOOKUP_RCU) {
1576 if (fput_needed)
70e9b357 1577 *fp = file;
e41f7d4e
AV		 1578 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1579 br_read_lock(vfsmount_lock);
1580 rcu_read_lock();
1581 } else {
1582 path_get(&file->f_path);
1583 fput_light(file, fput_needed);
1584 }
31e6b01f 1585 }
31e6b01f 1586
31e6b01f 1587 nd->inode = nd->path.dentry->d_inode;
9b4a9b14 1588 return 0;
2dfdd266 1589
9b4a9b14
AV		 1590fput_fail:
1591 fput_light(file, fput_needed);
1592out_fail:
1593 return retval;
1594}
1595
bd92d7fe
AV		 1596static inline int lookup_last(struct nameidata *nd, struct path *path)
1597{
1598 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1599 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1600
1601 nd->flags &= ~LOOKUP_PARENT;
1602 return walk_component(nd, path, &nd->last, nd->last_type,
1603 nd->flags & LOOKUP_FOLLOW);
1604}
1605
9b4a9b14 1606/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
ee0827cd 1607static int path_lookupat(int dfd, const char *name,
9b4a9b14
AV		 1608 unsigned int flags, struct nameidata *nd)
1609{
70e9b357 1610 struct file *base = NULL;
bd92d7fe
AV		 1611 struct path path;
1612 int err;
31e6b01f
NP		 1613
1614 /*
1615 * Path walking is largely split up into 2 different synchronisation
1616 * schemes, rcu-walk and ref-walk (explained in
1617 * Documentation/filesystems/path-lookup.txt). These share much of the
1618 * path walk code, but some things particularly setup, cleanup, and
1619 * following mounts are sufficiently divergent that functions are
1620 * duplicated. Typically there is a function foo(), and its RCU
1621 * analogue, foo_rcu().
1622 *
1623 * -ECHILD is the error number of choice (just to avoid clashes) that
1624 * is returned if some aspect of an rcu-walk fails. Such an error must
1625 * be handled by restarting a traditional ref-walk (which will always
1626 * be able to complete).
1627 */
bd92d7fe 1628 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
ee0827cd 1629
bd92d7fe
AV		 1630 if (unlikely(err))
1631 return err;
ee0827cd
AV		 1632
1633 current->total_link_count = 0;
bd92d7fe
AV		 1634 err = link_path_walk(name, nd);
1635
1636 if (!err && !(flags & LOOKUP_PARENT)) {
bd92d7fe
AV		 1637 err = lookup_last(nd, &path);
1638 while (err > 0) {
1639 void *cookie;
1640 struct path link = path;
bd92d7fe 1641 nd->flags |= LOOKUP_PARENT;
574197e0 1642 err = follow_link(&link, nd, &cookie);
bd92d7fe
AV		 1643 if (!err)
1644 err = lookup_last(nd, &path);
574197e0 1645 put_link(nd, &link, cookie);
bd92d7fe
AV		 1646 }
1647 }
ee0827cd 1648
9f1fafee
AV		 1649 if (!err)
1650 err = complete_walk(nd);
bd92d7fe
AV		 1651
1652 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1653 if (!nd->inode->i_op->lookup) {
1654 path_put(&nd->path);
bd23a539 1655 err = -ENOTDIR;
bd92d7fe
AV		 1656 }
1657 }
16c2cd71 1658
70e9b357
AV		 1659 if (base)
1660 fput(base);
ee0827cd 1661
5b6ca027 1662 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
2a737871
AV		 1663 path_put(&nd->root);
1664 nd->root.mnt = NULL;
1665 }
bd92d7fe 1666 return err;
ee0827cd 1667}
31e6b01f 1668
ee0827cd
AV		 1669static int do_path_lookup(int dfd, const char *name,
1670 unsigned int flags, struct nameidata *nd)
1671{
1672 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1673 if (unlikely(retval == -ECHILD))
1674 retval = path_lookupat(dfd, name, flags, nd);
1675 if (unlikely(retval == -ESTALE))
1676 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
31e6b01f
NP		 1677
1678 if (likely(!retval)) {
1679 if (unlikely(!audit_dummy_context())) {
1680 if (nd->path.dentry && nd->inode)
1681 audit_inode(name, nd->path.dentry);
1682 }
1683 }
170aa3d0 1684 return retval;
1da177e4
LT		 1685}
1686
c9c6cac0 1687int kern_path_parent(const char *name, struct nameidata *nd)
5590ff0d 1688{
c9c6cac0 1689 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
5590ff0d
UD		 1690}
1691
d1811465
AV		 1692int kern_path(const char *name, unsigned int flags, struct path *path)
1693{
1694 struct nameidata nd;
1695 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1696 if (!res)
1697 *path = nd.path;
1698 return res;
1699}
1700
16f18200
JJS		 1701/**
1702 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1703 * @dentry: pointer to dentry of the base directory
1704 * @mnt: pointer to vfs mount of the base directory
1705 * @name: pointer to file name
1706 * @flags: lookup flags
e0a01249 1707 * @path: pointer to struct path to fill
16f18200
JJS		 1708 */
1709int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1710 const char *name, unsigned int flags,
e0a01249 1711 struct path *path)
16f18200 1712{
e0a01249
AV		 1713 struct nameidata nd;
1714 int err;
1715 nd.root.dentry = dentry;
1716 nd.root.mnt = mnt;
1717 BUG_ON(flags & LOOKUP_PARENT);
5b6ca027 1718 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
e0a01249
AV		 1719 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1720 if (!err)
1721 *path = nd.path;
1722 return err;
16f18200
JJS		 1723}
1724
eead1911
CH		 1725static struct dentry *__lookup_hash(struct qstr *name,
1726 struct dentry *base, struct nameidata *nd)
1da177e4 1727{
81fca444 1728 struct inode *inode = base->d_inode;
057f6c01 1729 struct dentry *dentry;
1da177e4
LT		 1730 int err;
1731
4ad5abb3 1732 err = inode_permission(inode, MAY_EXEC);
81fca444
CH		 1733 if (err)
1734 return ERR_PTR(err);
1da177e4 1735
b04f784e
NP		 1736 /*
1737 * Don't bother with __d_lookup: callers are for creat as
1738 * well as unlink, so a lot of the time it would cost
1739 * a double lookup.
6e6b1bd1 1740 */
b04f784e 1741 dentry = d_lookup(base, name);
6e6b1bd1 1742
44396f4b
JB		 1743 if (dentry && d_need_lookup(dentry)) {
1744 /*
1745 * __lookup_hash is called with the parent dir's i_mutex already
1746 * held, so we are good to go here.
1747 */
1748 dentry = d_inode_lookup(base, dentry, nd);
1749 if (IS_ERR(dentry))
1750 return dentry;
1751 }
1752
d2d9e9fb
AV		 1753 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1754 int status = d_revalidate(dentry, nd);
1755 if (unlikely(status <= 0)) {
1756 /*
1757 * The dentry failed validation.
1758 * If d_revalidate returned 0 attempt to invalidate
1759 * the dentry otherwise d_revalidate is asking us
1760 * to return a fail status.
1761 */
1762 if (status < 0) {
1763 dput(dentry);
1764 return ERR_PTR(status);
1765 } else if (!d_invalidate(dentry)) {
1766 dput(dentry);
1767 dentry = NULL;
1768 }
1769 }
1770 }
6e6b1bd1 1771
baa03890
NP		 1772 if (!dentry)
1773 dentry = d_alloc_and_lookup(base, name, nd);
5a202bcd 1774
1da177e4
LT		 1775 return dentry;
1776}
1777
057f6c01
JM		 1778/*
1779 * Restricted form of lookup. Doesn't follow links, single-component only,
1780 * needs parent already locked. Doesn't follow mounts.
1781 * SMP-safe.
1782 */
eead1911 1783static struct dentry *lookup_hash(struct nameidata *nd)
057f6c01 1784{
4ac91378 1785 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1da177e4
LT		 1786}
1787
eead1911 1788/**
a6b91919 1789 * lookup_one_len - filesystem helper to lookup single pathname component
eead1911
CH		 1790 * @name: pathname component to lookup
1791 * @base: base directory to lookup from
1792 * @len: maximum length @len should be interpreted to
1793 *
a6b91919
RD		 1794 * Note that this routine is purely a helper for filesystem usage and should
1795 * not be called by generic code. Also note that by using this function the
eead1911
CH		 1796 * nameidata argument is passed to the filesystem methods and a filesystem
1797 * using this helper needs to be prepared for that.
1798 */
057f6c01
JM		 1799struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1800{
057f6c01 1801 struct qstr this;
6a96ba54 1802 unsigned int c;
057f6c01 1803
2f9092e1
DW		 1804 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1805
6a96ba54
AV		 1806 this.name = name;
1807 this.len = len;
0145acc2 1808 this.hash = full_name_hash(name, len);
6a96ba54
AV		 1809 if (!len)
1810 return ERR_PTR(-EACCES);
1811
6a96ba54
AV		 1812 while (len--) {
1813 c = *(const unsigned char *)name++;
1814 if (c == '/' || c == '\0')
1815 return ERR_PTR(-EACCES);
6a96ba54 1816 }
5a202bcd
AV		 1817 /*
1818 * See if the low-level filesystem might want
1819 * to use its own hash..
1820 */
1821 if (base->d_flags & DCACHE_OP_HASH) {
1822 int err = base->d_op->d_hash(base, base->d_inode, &this);
1823 if (err < 0)
1824 return ERR_PTR(err);
1825 }
eead1911 1826
49705b77 1827 return __lookup_hash(&this, base, NULL);
057f6c01
JM		 1828}
1829
1fa1e7f6
AW		 1830int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
1831 struct path *path, int *empty)
1da177e4 1832{
2d8f3038 1833 struct nameidata nd;
1fa1e7f6 1834 char *tmp = getname_flags(name, flags, empty);
1da177e4 1835 int err = PTR_ERR(tmp);
1da177e4 1836 if (!IS_ERR(tmp)) {
2d8f3038
AV		 1837
1838 BUG_ON(flags & LOOKUP_PARENT);
1839
1840 err = do_path_lookup(dfd, tmp, flags, &nd);
1da177e4 1841 putname(tmp);
2d8f3038
AV		 1842 if (!err)
1843 *path = nd.path;
1da177e4
LT		 1844 }
1845 return err;
1846}
1847
1fa1e7f6
AW		 1848int user_path_at(int dfd, const char __user *name, unsigned flags,
1849 struct path *path)
1850{
1851 return user_path_at_empty(dfd, name, flags, path, 0);
1852}
1853
2ad94ae6
AV		 1854static int user_path_parent(int dfd, const char __user *path,
1855 struct nameidata *nd, char **name)
1856{
1857 char *s = getname(path);
1858 int error;
1859
1860 if (IS_ERR(s))
1861 return PTR_ERR(s);
1862
1863 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1864 if (error)
1865 putname(s);
1866 else
1867 *name = s;
1868
1869 return error;
1870}
1871
1da177e4
LT		 1872/*
1873 * It's inline, so penalty for filesystems that don't use sticky bit is
1874 * minimal.
1875 */
1876static inline int check_sticky(struct inode *dir, struct inode *inode)
1877{
da9592ed
DH		 1878 uid_t fsuid = current_fsuid();
1879
1da177e4
LT		 1880 if (!(dir->i_mode & S_ISVTX))
1881 return 0;
e795b717
SH		 1882 if (current_user_ns() != inode_userns(inode))
1883 goto other_userns;
da9592ed 1884 if (inode->i_uid == fsuid)
1da177e4 1885 return 0;
da9592ed 1886 if (dir->i_uid == fsuid)
1da177e4 1887 return 0;
e795b717
SH		 1888
1889other_userns:
1890 return !ns_capable(inode_userns(inode), CAP_FOWNER);
1da177e4
LT		 1891}
1892
1893/*
1894 * Check whether we can remove a link victim from directory dir, check
1895 * whether the type of victim is right.
1896 * 1. We can't do it if dir is read-only (done in permission())
1897 * 2. We should have write and exec permissions on dir
1898 * 3. We can't remove anything from append-only dir
1899 * 4. We can't do anything with immutable dir (done in permission())
1900 * 5. If the sticky bit on dir is set we should either
1901 * a. be owner of dir, or
1902 * b. be owner of victim, or
1903 * c. have CAP_FOWNER capability
1904 * 6. If the victim is append-only or immutable we can't do antyhing with
1905 * links pointing to it.
1906 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1907 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1908 * 9. We can't remove a root or mountpoint.
1909 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1910 * nfs_async_unlink().
1911 */
858119e1 1912static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1da177e4
LT		 1913{
1914 int error;
1915
1916 if (!victim->d_inode)
1917 return -ENOENT;
1918
1919 BUG_ON(victim->d_parent->d_inode != dir);
cccc6bba 1920 audit_inode_child(victim, dir);
1da177e4 1921
f419a2e3 1922 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1923 if (error)
1924 return error;
1925 if (IS_APPEND(dir))
1926 return -EPERM;
1927 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
f9454548 1928 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1da177e4
LT		 1929 return -EPERM;
1930 if (isdir) {
1931 if (!S_ISDIR(victim->d_inode->i_mode))
1932 return -ENOTDIR;
1933 if (IS_ROOT(victim))
1934 return -EBUSY;
1935 } else if (S_ISDIR(victim->d_inode->i_mode))
1936 return -EISDIR;
1937 if (IS_DEADDIR(dir))
1938 return -ENOENT;
1939 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1940 return -EBUSY;
1941 return 0;
1942}
1943
1944/* Check whether we can create an object with dentry child in directory
1945 * dir.
1946 * 1. We can't do it if child already exists (open has special treatment for
1947 * this case, but since we are inlined it's OK)
1948 * 2. We can't do it if dir is read-only (done in permission())
1949 * 3. We should have write and exec permissions on dir
1950 * 4. We can't do it if dir is immutable (done in permission())
1951 */
a95164d9 1952static inline int may_create(struct inode *dir, struct dentry *child)
1da177e4
LT		 1953{
1954 if (child->d_inode)
1955 return -EEXIST;
1956 if (IS_DEADDIR(dir))
1957 return -ENOENT;
f419a2e3 1958 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1959}
1960
1da177e4
LT		 1961/*
1962 * p1 and p2 should be directories on the same fs.
1963 */
1964struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1965{
1966 struct dentry *p;
1967
1968 if (p1 == p2) {
f2eace23 1969 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1da177e4
LT		 1970 return NULL;
1971 }
1972
a11f3a05 1973 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4 1974
e2761a11
OH		 1975 p = d_ancestor(p2, p1);
1976 if (p) {
1977 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1978 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1979 return p;
1da177e4
LT		 1980 }
1981
e2761a11
OH		 1982 p = d_ancestor(p1, p2);
1983 if (p) {
1984 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1985 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1986 return p;
1da177e4
LT		 1987 }
1988
f2eace23
IM		 1989 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1990 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1da177e4
LT		 1991 return NULL;
1992}
1993
1994void unlock_rename(struct dentry *p1, struct dentry *p2)
1995{
1b1dcc1b 1996 mutex_unlock(&p1->d_inode->i_mutex);
1da177e4 1997 if (p1 != p2) {
1b1dcc1b 1998 mutex_unlock(&p2->d_inode->i_mutex);
a11f3a05 1999 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4
LT		 2000 }
2001}
2002
4acdaf27 2003int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
1da177e4
LT		 2004 struct nameidata *nd)
2005{
a95164d9 2006 int error = may_create(dir, dentry);
1da177e4
LT		 2007
2008 if (error)
2009 return error;
2010
acfa4380 2011 if (!dir->i_op->create)
1da177e4
LT		 2012 return -EACCES; /* shouldn't it be ENOSYS? */
2013 mode &= S_IALLUGO;
2014 mode |= S_IFREG;
2015 error = security_inode_create(dir, dentry, mode);
2016 if (error)
2017 return error;
1da177e4 2018 error = dir->i_op->create(dir, dentry, mode, nd);
a74574aa 2019 if (!error)
f38aa942 2020 fsnotify_create(dir, dentry);
1da177e4
LT		 2021 return error;
2022}
2023
73d049a4 2024static int may_open(struct path *path, int acc_mode, int flag)
1da177e4 2025{
3fb64190 2026 struct dentry *dentry = path->dentry;
1da177e4
LT		 2027 struct inode *inode = dentry->d_inode;
2028 int error;
2029
bcda7652
AV		 2030 /* O_PATH? */
2031 if (!acc_mode)
2032 return 0;
2033
1da177e4
LT		 2034 if (!inode)
2035 return -ENOENT;
2036
c8fe8f30
CH		 2037 switch (inode->i_mode & S_IFMT) {
2038 case S_IFLNK:
1da177e4 2039 return -ELOOP;
c8fe8f30
CH		 2040 case S_IFDIR:
2041 if (acc_mode & MAY_WRITE)
2042 return -EISDIR;
2043 break;
2044 case S_IFBLK:
2045 case S_IFCHR:
3fb64190 2046 if (path->mnt->mnt_flags & MNT_NODEV)
1da177e4 2047 return -EACCES;
c8fe8f30
CH		 2048 /*FALLTHRU*/
2049 case S_IFIFO:
2050 case S_IFSOCK:
1da177e4 2051 flag &= ~O_TRUNC;
c8fe8f30 2052 break;
4a3fd211 2053 }
b41572e9 2054
3fb64190 2055 error = inode_permission(inode, acc_mode);
b41572e9
DH		 2056 if (error)
2057 return error;
6146f0d5 2058
1da177e4
LT		 2059 /*
2060 * An append-only file must be opened in append mode for writing.
2061 */
2062 if (IS_APPEND(inode)) {
8737c930 2063 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
7715b521 2064 return -EPERM;
1da177e4 2065 if (flag & O_TRUNC)
7715b521 2066 return -EPERM;
1da177e4
LT		 2067 }
2068
2069 /* O_NOATIME can only be set by the owner or superuser */
2e149670 2070 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
7715b521 2071 return -EPERM;
1da177e4 2072
f3c7691e 2073 return 0;
7715b521 2074}
1da177e4 2075
e1181ee6 2076static int handle_truncate(struct file *filp)
7715b521 2077{
e1181ee6 2078 struct path *path = &filp->f_path;
7715b521
AV		 2079 struct inode *inode = path->dentry->d_inode;
2080 int error = get_write_access(inode);
2081 if (error)
2082 return error;
2083 /*
2084 * Refuse to truncate files with mandatory locks held on them.
2085 */
2086 error = locks_verify_locked(inode);
2087 if (!error)
ea0d3ab2 2088 error = security_path_truncate(path);
7715b521
AV		 2089 if (!error) {
2090 error = do_truncate(path->dentry, 0,
2091 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
e1181ee6 2092 filp);
7715b521
AV		 2093 }
2094 put_write_access(inode);
acd0c935 2095 return error;
1da177e4
LT		 2096}
2097
d57999e1
DH		 2098static inline int open_to_namei_flags(int flag)
2099{
8a5e929d
AV		 2100 if ((flag & O_ACCMODE) == 3)
2101 flag--;
d57999e1
DH		 2102 return flag;
2103}
2104
31e6b01f 2105/*
fe2d35ff 2106 * Handle the last step of open()
31e6b01f 2107 */
fb1cc555 2108static struct file *do_last(struct nameidata *nd, struct path *path,
c3e380b0 2109 const struct open_flags *op, const char *pathname)
fb1cc555 2110{
a1e28038 2111 struct dentry *dir = nd->path.dentry;
6c0d46c4 2112 struct dentry *dentry;
ca344a89 2113 int open_flag = op->open_flag;
6c0d46c4 2114 int will_truncate = open_flag & O_TRUNC;
ca344a89 2115 int want_write = 0;
bcda7652 2116 int acc_mode = op->acc_mode;
fb1cc555 2117 struct file *filp;
16c2cd71 2118 int error;
1f36f774 2119
c3e380b0
AV		 2120 nd->flags &= ~LOOKUP_PARENT;
2121 nd->flags |= op->intent;
2122
1f36f774
AV		 2123 switch (nd->last_type) {
2124 case LAST_DOTDOT:
176306f5 2125 case LAST_DOT:
fe2d35ff
AV		 2126 error = handle_dots(nd, nd->last_type);
2127 if (error)
2128 return ERR_PTR(error);
1f36f774 2129 /* fallthrough */
1f36f774 2130 case LAST_ROOT:
9f1fafee 2131 error = complete_walk(nd);
16c2cd71 2132 if (error)
9f1fafee 2133 return ERR_PTR(error);
fe2d35ff 2134 audit_inode(pathname, nd->path.dentry);
ca344a89 2135 if (open_flag & O_CREAT) {
fe2d35ff
AV		 2136 error = -EISDIR;
2137 goto exit;
2138 }
2139 goto ok;
1f36f774 2140 case LAST_BIND:
9f1fafee 2141 error = complete_walk(nd);
16c2cd71 2142 if (error)
9f1fafee 2143 return ERR_PTR(error);
1f36f774 2144 audit_inode(pathname, dir);
67ee3ad2 2145 goto ok;
1f36f774 2146 }
67ee3ad2 2147
ca344a89 2148 if (!(open_flag & O_CREAT)) {
bcda7652 2149 int symlink_ok = 0;
fe2d35ff
AV		 2150 if (nd->last.name[nd->last.len])
2151 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
bcda7652
AV		 2152 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2153 symlink_ok = 1;
fe2d35ff 2154 /* we _can_ be in RCU mode here */
ce57dfc1
AV		 2155 error = walk_component(nd, path, &nd->last, LAST_NORM,
2156 !symlink_ok);
2157 if (error < 0)
fe2d35ff 2158 return ERR_PTR(error);
ce57dfc1 2159 if (error) /* symlink */
fe2d35ff 2160 return NULL;
fe2d35ff 2161 /* sayonara */
9f1fafee
AV		 2162 error = complete_walk(nd);
2163 if (error)
2164 return ERR_PTR(-ECHILD);
fe2d35ff
AV		 2165
2166 error = -ENOTDIR;
2167 if (nd->flags & LOOKUP_DIRECTORY) {
ce57dfc1 2168 if (!nd->inode->i_op->lookup)
fe2d35ff
AV		 2169 goto exit;
2170 }
2171 audit_inode(pathname, nd->path.dentry);
2172 goto ok;
2173 }
2174
2175 /* create side of things */
a3fbbde7
AV		 2176 /*
2177 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED has been
2178 * cleared when we got to the last component we are about to look up
2179 */
9f1fafee
AV		 2180 error = complete_walk(nd);
2181 if (error)
2182 return ERR_PTR(error);
fe2d35ff
AV		 2183
2184 audit_inode(pathname, dir);
16c2cd71 2185 error = -EISDIR;
1f36f774 2186 /* trailing slashes? */
31e6b01f
NP		 2187 if (nd->last.name[nd->last.len])
2188 goto exit;
a2c36b45 2189
a1e28038
AV		 2190 mutex_lock(&dir->d_inode->i_mutex);
2191
6c0d46c4
AV		 2192 dentry = lookup_hash(nd);
2193 error = PTR_ERR(dentry);
2194 if (IS_ERR(dentry)) {
fb1cc555
AV		 2195 mutex_unlock(&dir->d_inode->i_mutex);
2196 goto exit;
2197 }
2198
6c0d46c4
AV		 2199 path->dentry = dentry;
2200 path->mnt = nd->path.mnt;
2201
fb1cc555 2202 /* Negative dentry, just create the file */
6c0d46c4 2203 if (!dentry->d_inode) {
a218d0fd 2204 umode_t mode = op->mode;
6c0d46c4
AV		 2205 if (!IS_POSIXACL(dir->d_inode))
2206 mode &= ~current_umask();
fb1cc555
AV		 2207 /*
2208 * This write is needed to ensure that a
6c0d46c4 2209 * rw->ro transition does not occur between
fb1cc555
AV		 2210 * the time when the file is created and when
2211 * a permanent write count is taken through
2212 * the 'struct file' in nameidata_to_filp().
2213 */
2214 error = mnt_want_write(nd->path.mnt);
2215 if (error)
2216 goto exit_mutex_unlock;
ca344a89 2217 want_write = 1;
9b44f1b3 2218 /* Don't check for write permission, don't truncate */
ca344a89 2219 open_flag &= ~O_TRUNC;
6c0d46c4 2220 will_truncate = 0;
bcda7652 2221 acc_mode = MAY_OPEN;
6c0d46c4
AV		 2222 error = security_path_mknod(&nd->path, dentry, mode, 0);
2223 if (error)
2224 goto exit_mutex_unlock;
2225 error = vfs_create(dir->d_inode, dentry, mode, nd);
2226 if (error)
2227 goto exit_mutex_unlock;
2228 mutex_unlock(&dir->d_inode->i_mutex);
2229 dput(nd->path.dentry);
2230 nd->path.dentry = dentry;
ca344a89 2231 goto common;
fb1cc555
AV		 2232 }
2233
2234 /*
2235 * It already exists.
2236 */
2237 mutex_unlock(&dir->d_inode->i_mutex);
2238 audit_inode(pathname, path->dentry);
2239
2240 error = -EEXIST;
ca344a89 2241 if (open_flag & O_EXCL)
fb1cc555
AV		 2242 goto exit_dput;
2243
9875cf80
DH		 2244 error = follow_managed(path, nd->flags);
2245 if (error < 0)
2246 goto exit_dput;
fb1cc555 2247
a3fbbde7
AV		 2248 if (error)
2249 nd->flags |= LOOKUP_JUMPED;
2250
fb1cc555
AV		 2251 error = -ENOENT;
2252 if (!path->dentry->d_inode)
2253 goto exit_dput;
9e67f361
AV		 2254
2255 if (path->dentry->d_inode->i_op->follow_link)
fb1cc555 2256 return NULL;
fb1cc555
AV		 2257
2258 path_to_nameidata(path, nd);
31e6b01f 2259 nd->inode = path->dentry->d_inode;
a3fbbde7
AV		 2260 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2261 error = complete_walk(nd);
2262 if (error)
2263 goto exit;
fb1cc555 2264 error = -EISDIR;
31e6b01f 2265 if (S_ISDIR(nd->inode->i_mode))
fb1cc555 2266 goto exit;
67ee3ad2 2267ok:
6c0d46c4
AV		 2268 if (!S_ISREG(nd->inode->i_mode))
2269 will_truncate = 0;
2270
0f9d1a10
AV		 2271 if (will_truncate) {
2272 error = mnt_want_write(nd->path.mnt);
2273 if (error)
2274 goto exit;
ca344a89 2275 want_write = 1;
0f9d1a10 2276 }
ca344a89 2277common:
bcda7652 2278 error = may_open(&nd->path, acc_mode, open_flag);
ca344a89 2279 if (error)
0f9d1a10 2280 goto exit;
0f9d1a10
AV		 2281 filp = nameidata_to_filp(nd);
2282 if (!IS_ERR(filp)) {
2283 error = ima_file_check(filp, op->acc_mode);
2284 if (error) {
2285 fput(filp);
2286 filp = ERR_PTR(error);
2287 }
2288 }
2289 if (!IS_ERR(filp)) {
2290 if (will_truncate) {
2291 error = handle_truncate(filp);
2292 if (error) {
2293 fput(filp);
2294 filp = ERR_PTR(error);
2295 }
2296 }
2297 }
ca344a89
AV		 2298out:
2299 if (want_write)
0f9d1a10
AV		 2300 mnt_drop_write(nd->path.mnt);
2301 path_put(&nd->path);
fb1cc555
AV		 2302 return filp;
2303
2304exit_mutex_unlock:
2305 mutex_unlock(&dir->d_inode->i_mutex);
2306exit_dput:
2307 path_put_conditional(path, nd);
2308exit:
ca344a89
AV		 2309 filp = ERR_PTR(error);
2310 goto out;
fb1cc555
AV		 2311}
2312
13aab428 2313static struct file *path_openat(int dfd, const char *pathname,
73d049a4 2314 struct nameidata *nd, const struct open_flags *op, int flags)
1da177e4 2315{
fe2d35ff 2316 struct file *base = NULL;
4a3fd211 2317 struct file *filp;
9850c056 2318 struct path path;
13aab428 2319 int error;
31e6b01f
NP		 2320
2321 filp = get_empty_filp();
2322 if (!filp)
2323 return ERR_PTR(-ENFILE);
2324
47c805dc 2325 filp->f_flags = op->open_flag;
73d049a4
AV		 2326 nd->intent.open.file = filp;
2327 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2328 nd->intent.open.create_mode = op->mode;
31e6b01f 2329
73d049a4 2330 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
31e6b01f 2331 if (unlikely(error))
13aab428 2332 goto out_filp;
31e6b01f 2333
fe2d35ff 2334 current->total_link_count = 0;
73d049a4 2335 error = link_path_walk(pathname, nd);
31e6b01f
NP		 2336 if (unlikely(error))
2337 goto out_filp;
1da177e4 2338
73d049a4 2339 filp = do_last(nd, &path, op, pathname);
806b681c 2340 while (unlikely(!filp)) { /* trailing symlink */
7b9337aa 2341 struct path link = path;
def4af30 2342 void *cookie;
574197e0 2343 if (!(nd->flags & LOOKUP_FOLLOW)) {
73d049a4
AV		 2344 path_put_conditional(&path, nd);
2345 path_put(&nd->path);
40b39136
AV		 2346 filp = ERR_PTR(-ELOOP);
2347 break;
2348 }
73d049a4
AV		 2349 nd->flags |= LOOKUP_PARENT;
2350 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
574197e0 2351 error = follow_link(&link, nd, &cookie);
c3e380b0 2352 if (unlikely(error))
f1afe9ef 2353 filp = ERR_PTR(error);
c3e380b0 2354 else
73d049a4 2355 filp = do_last(nd, &path, op, pathname);
574197e0 2356 put_link(nd, &link, cookie);
806b681c 2357 }
10fa8e62 2358out:
73d049a4
AV		 2359 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2360 path_put(&nd->root);
fe2d35ff
AV		 2361 if (base)
2362 fput(base);
73d049a4 2363 release_open_intent(nd);
10fa8e62 2364 return filp;
1da177e4 2365
31e6b01f 2366out_filp:
806b681c 2367 filp = ERR_PTR(error);
10fa8e62 2368 goto out;
1da177e4
LT		 2369}
2370
13aab428
AV		 2371struct file *do_filp_open(int dfd, const char *pathname,
2372 const struct open_flags *op, int flags)
2373{
73d049a4 2374 struct nameidata nd;
13aab428
AV		 2375 struct file *filp;
2376
73d049a4 2377 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
13aab428 2378 if (unlikely(filp == ERR_PTR(-ECHILD)))
73d049a4 2379 filp = path_openat(dfd, pathname, &nd, op, flags);
13aab428 2380 if (unlikely(filp == ERR_PTR(-ESTALE)))
73d049a4 2381 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
13aab428
AV		 2382 return filp;
2383}
2384
73d049a4
AV		 2385struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2386 const char *name, const struct open_flags *op, int flags)
2387{
2388 struct nameidata nd;
2389 struct file *file;
2390
2391 nd.root.mnt = mnt;
2392 nd.root.dentry = dentry;
2393
2394 flags |= LOOKUP_ROOT;
2395
bcda7652 2396 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
73d049a4
AV		 2397 return ERR_PTR(-ELOOP);
2398
2399 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2400 if (unlikely(file == ERR_PTR(-ECHILD)))
2401 file = path_openat(-1, name, &nd, op, flags);
2402 if (unlikely(file == ERR_PTR(-ESTALE)))
2403 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2404 return file;
2405}
2406
ed75e95d 2407struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
1da177e4 2408{
c663e5d8 2409 struct dentry *dentry = ERR_PTR(-EEXIST);
ed75e95d
AV		 2410 struct nameidata nd;
2411 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2412 if (error)
2413 return ERR_PTR(error);
1da177e4 2414
c663e5d8
CH		 2415 /*
2416 * Yucky last component or no last component at all?
2417 * (foo/., foo/.., /////)
2418 */
ed75e95d
AV		 2419 if (nd.last_type != LAST_NORM)
2420 goto out;
2421 nd.flags &= ~LOOKUP_PARENT;
2422 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2423 nd.intent.open.flags = O_EXCL;
c663e5d8
CH		 2424
2425 /*
2426 * Do the final lookup.
2427 */
ed75e95d
AV		 2428 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2429 dentry = lookup_hash(&nd);
1da177e4
LT		 2430 if (IS_ERR(dentry))
2431 goto fail;
c663e5d8 2432
e9baf6e5
AV		 2433 if (dentry->d_inode)
2434 goto eexist;
c663e5d8
CH		 2435 /*
2436 * Special case - lookup gave negative, but... we had foo/bar/
2437 * From the vfs_mknod() POV we just have a negative dentry -
2438 * all is fine. Let's be bastards - you had / on the end, you've
2439 * been asking for (non-existent) directory. -ENOENT for you.
2440 */
ed75e95d 2441 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
e9baf6e5
AV		 2442 dput(dentry);
2443 dentry = ERR_PTR(-ENOENT);
ed75e95d 2444 goto fail;
e9baf6e5 2445 }
ed75e95d 2446 *path = nd.path;
1da177e4 2447 return dentry;
e9baf6e5 2448eexist:
1da177e4 2449 dput(dentry);
e9baf6e5 2450 dentry = ERR_PTR(-EEXIST);
1da177e4 2451fail:
ed75e95d
AV		 2452 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2453out:
2454 path_put(&nd.path);
1da177e4
LT		 2455 return dentry;
2456}
dae6ad8f
AV		 2457EXPORT_SYMBOL(kern_path_create);
2458
2459struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2460{
2461 char *tmp = getname(pathname);
2462 struct dentry *res;
2463 if (IS_ERR(tmp))
2464 return ERR_CAST(tmp);
2465 res = kern_path_create(dfd, tmp, path, is_dir);
2466 putname(tmp);
2467 return res;
2468}
2469EXPORT_SYMBOL(user_path_create);
2470
1a67aafb 2471int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
1da177e4 2472{
a95164d9 2473 int error = may_create(dir, dentry);
1da177e4
LT		 2474
2475 if (error)
2476 return error;
2477
e795b717
SH		 2478 if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
2479 !ns_capable(inode_userns(dir), CAP_MKNOD))
1da177e4
LT		 2480 return -EPERM;
2481
acfa4380 2482 if (!dir->i_op->mknod)
1da177e4
LT		 2483 return -EPERM;
2484
08ce5f16
SH		 2485 error = devcgroup_inode_mknod(mode, dev);
2486 if (error)
2487 return error;
2488
1da177e4
LT		 2489 error = security_inode_mknod(dir, dentry, mode, dev);
2490 if (error)
2491 return error;
2492
1da177e4 2493 error = dir->i_op->mknod(dir, dentry, mode, dev);
a74574aa 2494 if (!error)
f38aa942 2495 fsnotify_create(dir, dentry);
1da177e4
LT		 2496 return error;
2497}
2498
f69aac00 2499static int may_mknod(umode_t mode)
463c3197
DH		 2500{
2501 switch (mode & S_IFMT) {
2502 case S_IFREG:
2503 case S_IFCHR:
2504 case S_IFBLK:
2505 case S_IFIFO:
2506 case S_IFSOCK:
2507 case 0: /* zero mode translates to S_IFREG */
2508 return 0;
2509 case S_IFDIR:
2510 return -EPERM;
2511 default:
2512 return -EINVAL;
2513 }
2514}
2515
8208a22b 2516SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
2e4d0924 2517 unsigned, dev)
1da177e4 2518{
2ad94ae6 2519 struct dentry *dentry;
dae6ad8f
AV		 2520 struct path path;
2521 int error;
1da177e4
LT		 2522
2523 if (S_ISDIR(mode))
2524 return -EPERM;
1da177e4 2525
dae6ad8f
AV		 2526 dentry = user_path_create(dfd, filename, &path, 0);
2527 if (IS_ERR(dentry))
2528 return PTR_ERR(dentry);
2ad94ae6 2529
dae6ad8f 2530 if (!IS_POSIXACL(path.dentry->d_inode))
ce3b0f8d 2531 mode &= ~current_umask();
463c3197
DH		 2532 error = may_mknod(mode);
2533 if (error)
2534 goto out_dput;
dae6ad8f 2535 error = mnt_want_write(path.mnt);
463c3197
DH		 2536 if (error)
2537 goto out_dput;
dae6ad8f 2538 error = security_path_mknod(&path, dentry, mode, dev);
be6d3e56
KT		 2539 if (error)
2540 goto out_drop_write;
463c3197 2541 switch (mode & S_IFMT) {
1da177e4 2542 case 0: case S_IFREG:
dae6ad8f 2543 error = vfs_create(path.dentry->d_inode,dentry,mode,NULL);
1da177e4
LT		 2544 break;
2545 case S_IFCHR: case S_IFBLK:
dae6ad8f 2546 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
1da177e4
LT		 2547 new_decode_dev(dev));
2548 break;
2549 case S_IFIFO: case S_IFSOCK:
dae6ad8f 2550 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
1da177e4 2551 break;
1da177e4 2552 }
be6d3e56 2553out_drop_write:
dae6ad8f 2554 mnt_drop_write(path.mnt);
463c3197
DH		 2555out_dput:
2556 dput(dentry);
dae6ad8f
AV		 2557 mutex_unlock(&path.dentry->d_inode->i_mutex);
2558 path_put(&path);
1da177e4
LT		 2559
2560 return error;
2561}
2562
8208a22b 2563SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
5590ff0d
UD		 2564{
2565 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2566}
2567
18bb1db3 2568int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1da177e4 2569{
a95164d9 2570 int error = may_create(dir, dentry);
1da177e4
LT		 2571
2572 if (error)
2573 return error;
2574
acfa4380 2575 if (!dir->i_op->mkdir)
1da177e4
LT		 2576 return -EPERM;
2577
2578 mode &= (S_IRWXUGO|S_ISVTX);
2579 error = security_inode_mkdir(dir, dentry, mode);
2580 if (error)
2581 return error;
2582
1da177e4 2583 error = dir->i_op->mkdir(dir, dentry, mode);
a74574aa 2584 if (!error)
f38aa942 2585 fsnotify_mkdir(dir, dentry);
1da177e4
LT		 2586 return error;
2587}
2588
a218d0fd 2589SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
1da177e4 2590{
6902d925 2591 struct dentry *dentry;
dae6ad8f
AV		 2592 struct path path;
2593 int error;
1da177e4 2594
dae6ad8f 2595 dentry = user_path_create(dfd, pathname, &path, 1);
6902d925 2596 if (IS_ERR(dentry))
dae6ad8f 2597 return PTR_ERR(dentry);
1da177e4 2598
dae6ad8f 2599 if (!IS_POSIXACL(path.dentry->d_inode))
ce3b0f8d 2600 mode &= ~current_umask();
dae6ad8f 2601 error = mnt_want_write(path.mnt);
463c3197
DH		 2602 if (error)
2603 goto out_dput;
dae6ad8f 2604 error = security_path_mkdir(&path, dentry, mode);
be6d3e56
KT		 2605 if (error)
2606 goto out_drop_write;
dae6ad8f 2607 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
be6d3e56 2608out_drop_write:
dae6ad8f 2609 mnt_drop_write(path.mnt);
463c3197 2610out_dput:
6902d925 2611 dput(dentry);
dae6ad8f
AV		 2612 mutex_unlock(&path.dentry->d_inode->i_mutex);
2613 path_put(&path);
1da177e4
LT		 2614 return error;
2615}
2616
a218d0fd 2617SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
5590ff0d
UD		 2618{
2619 return sys_mkdirat(AT_FDCWD, pathname, mode);
2620}
2621
1da177e4 2622/*
a71905f0
SW		 2623 * The dentry_unhash() helper will try to drop the dentry early: we
2624 * should have a usage count of 2 if we're the only user of this
2625 * dentry, and if that is true (possibly after pruning the dcache),
2626 * then we drop the dentry now.
1da177e4
LT		 2627 *
2628 * A low-level filesystem can, if it choses, legally
2629 * do a
2630 *
2631 * if (!d_unhashed(dentry))
2632 * return -EBUSY;
2633 *
2634 * if it cannot handle the case of removing a directory
2635 * that is still in use by something else..
2636 */
2637void dentry_unhash(struct dentry *dentry)
2638{
dc168427 2639 shrink_dcache_parent(dentry);
1da177e4 2640 spin_lock(&dentry->d_lock);
64252c75 2641 if (dentry->d_count == 1)
1da177e4
LT		 2642 __d_drop(dentry);
2643 spin_unlock(&dentry->d_lock);
1da177e4
LT		 2644}
2645
2646int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2647{
2648 int error = may_delete(dir, dentry, 1);
2649
2650 if (error)
2651 return error;
2652
acfa4380 2653 if (!dir->i_op->rmdir)
1da177e4
LT		 2654 return -EPERM;
2655
1d2ef590 2656 dget(dentry);
1b1dcc1b 2657 mutex_lock(&dentry->d_inode->i_mutex);
912dbc15
SW		 2658
2659 error = -EBUSY;
1da177e4 2660 if (d_mountpoint(dentry))
912dbc15
SW		 2661 goto out;
2662
2663 error = security_inode_rmdir(dir, dentry);
2664 if (error)
2665 goto out;
2666
3cebde24 2667 shrink_dcache_parent(dentry);
912dbc15
SW		 2668 error = dir->i_op->rmdir(dir, dentry);
2669 if (error)
2670 goto out;
2671
2672 dentry->d_inode->i_flags |= S_DEAD;
2673 dont_mount(dentry);
2674
2675out:
1b1dcc1b 2676 mutex_unlock(&dentry->d_inode->i_mutex);
1d2ef590 2677 dput(dentry);
912dbc15 2678 if (!error)
1da177e4 2679 d_delete(dentry);
1da177e4
LT		 2680 return error;
2681}
2682
5590ff0d 2683static long do_rmdir(int dfd, const char __user *pathname)
1da177e4
LT		 2684{
2685 int error = 0;
2686 char * name;
2687 struct dentry *dentry;
2688 struct nameidata nd;
2689
2ad94ae6 2690 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2691 if (error)
2ad94ae6 2692 return error;
1da177e4
LT		 2693
2694 switch(nd.last_type) {
0612d9fb
OH		 2695 case LAST_DOTDOT:
2696 error = -ENOTEMPTY;
2697 goto exit1;
2698 case LAST_DOT:
2699 error = -EINVAL;
2700 goto exit1;
2701 case LAST_ROOT:
2702 error = -EBUSY;
2703 goto exit1;
1da177e4 2704 }
0612d9fb
OH		 2705
2706 nd.flags &= ~LOOKUP_PARENT;
2707
4ac91378 2708 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2709 dentry = lookup_hash(&nd);
1da177e4 2710 error = PTR_ERR(dentry);
6902d925
DH		 2711 if (IS_ERR(dentry))
2712 goto exit2;
e6bc45d6
TT		 2713 if (!dentry->d_inode) {
2714 error = -ENOENT;
2715 goto exit3;
2716 }
0622753b
DH		 2717 error = mnt_want_write(nd.path.mnt);
2718 if (error)
2719 goto exit3;
be6d3e56
KT		 2720 error = security_path_rmdir(&nd.path, dentry);
2721 if (error)
2722 goto exit4;
4ac91378 2723 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
be6d3e56 2724exit4:
0622753b
DH		 2725 mnt_drop_write(nd.path.mnt);
2726exit3:
6902d925
DH		 2727 dput(dentry);
2728exit2:
4ac91378 2729 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2730exit1:
1d957f9b 2731 path_put(&nd.path);
1da177e4
LT		 2732 putname(name);
2733 return error;
2734}
2735
3cdad428 2736SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
5590ff0d
UD		 2737{
2738 return do_rmdir(AT_FDCWD, pathname);
2739}
2740
1da177e4
LT		 2741int vfs_unlink(struct inode *dir, struct dentry *dentry)
2742{
2743 int error = may_delete(dir, dentry, 0);
2744
2745 if (error)
2746 return error;
2747
acfa4380 2748 if (!dir->i_op->unlink)
1da177e4
LT		 2749 return -EPERM;
2750
1b1dcc1b 2751 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2752 if (d_mountpoint(dentry))
2753 error = -EBUSY;
2754 else {
2755 error = security_inode_unlink(dir, dentry);
bec1052e 2756 if (!error) {
1da177e4 2757 error = dir->i_op->unlink(dir, dentry);
bec1052e 2758 if (!error)
d83c49f3 2759 dont_mount(dentry);
bec1052e 2760 }
1da177e4 2761 }
1b1dcc1b 2762 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2763
2764 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2765 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
ece95912 2766 fsnotify_link_count(dentry->d_inode);
e234f35c 2767 d_delete(dentry);
1da177e4 2768 }
0eeca283 2769
1da177e4
LT		 2770 return error;
2771}
2772
2773/*
2774 * Make sure that the actual truncation of the file will occur outside its
1b1dcc1b 2775 * directory's i_mutex. Truncate can take a long time if there is a lot of
1da177e4
LT		 2776 * writeout happening, and we don't want to prevent access to the directory
2777 * while waiting on the I/O.
2778 */
5590ff0d 2779static long do_unlinkat(int dfd, const char __user *pathname)
1da177e4 2780{
2ad94ae6
AV		 2781 int error;
2782 char *name;
1da177e4
LT		 2783 struct dentry *dentry;
2784 struct nameidata nd;
2785 struct inode *inode = NULL;
2786
2ad94ae6 2787 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2788 if (error)
2ad94ae6
AV		 2789 return error;
2790
1da177e4
LT		 2791 error = -EISDIR;
2792 if (nd.last_type != LAST_NORM)
2793 goto exit1;
0612d9fb
OH		 2794
2795 nd.flags &= ~LOOKUP_PARENT;
2796
4ac91378 2797 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2798 dentry = lookup_hash(&nd);
1da177e4
LT		 2799 error = PTR_ERR(dentry);
2800 if (!IS_ERR(dentry)) {
2801 /* Why not before? Because we want correct error value */
50338b88
TE		 2802 if (nd.last.name[nd.last.len])
2803 goto slashes;
1da177e4 2804 inode = dentry->d_inode;
50338b88 2805 if (!inode)
e6bc45d6
TT		 2806 goto slashes;
2807 ihold(inode);
0622753b
DH		 2808 error = mnt_want_write(nd.path.mnt);
2809 if (error)
2810 goto exit2;
be6d3e56
KT		 2811 error = security_path_unlink(&nd.path, dentry);
2812 if (error)
2813 goto exit3;
4ac91378 2814 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
be6d3e56 2815exit3:
0622753b 2816 mnt_drop_write(nd.path.mnt);
1da177e4
LT		 2817 exit2:
2818 dput(dentry);
2819 }
4ac91378 2820 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4
LT		 2821 if (inode)
2822 iput(inode); /* truncate the inode here */
2823exit1:
1d957f9b 2824 path_put(&nd.path);
1da177e4
LT		 2825 putname(name);
2826 return error;
2827
2828slashes:
2829 error = !dentry->d_inode ? -ENOENT :
2830 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2831 goto exit2;
2832}
2833
2e4d0924 2834SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
5590ff0d
UD		 2835{
2836 if ((flag & ~AT_REMOVEDIR) != 0)
2837 return -EINVAL;
2838
2839 if (flag & AT_REMOVEDIR)
2840 return do_rmdir(dfd, pathname);
2841
2842 return do_unlinkat(dfd, pathname);
2843}
2844
3480b257 2845SYSCALL_DEFINE1(unlink, const char __user *, pathname)
5590ff0d
UD		 2846{
2847 return do_unlinkat(AT_FDCWD, pathname);
2848}
2849
db2e747b 2850int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
1da177e4 2851{
a95164d9 2852 int error = may_create(dir, dentry);
1da177e4
LT		 2853
2854 if (error)
2855 return error;
2856
acfa4380 2857 if (!dir->i_op->symlink)
1da177e4
LT		 2858 return -EPERM;
2859
2860 error = security_inode_symlink(dir, dentry, oldname);
2861 if (error)
2862 return error;
2863
1da177e4 2864 error = dir->i_op->symlink(dir, dentry, oldname);
a74574aa 2865 if (!error)
f38aa942 2866 fsnotify_create(dir, dentry);
1da177e4
LT		 2867 return error;
2868}
2869
2e4d0924
HC		 2870SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2871 int, newdfd, const char __user *, newname)
1da177e4 2872{
2ad94ae6
AV		 2873 int error;
2874 char *from;
6902d925 2875 struct dentry *dentry;
dae6ad8f 2876 struct path path;
1da177e4
LT		 2877
2878 from = getname(oldname);
2ad94ae6 2879 if (IS_ERR(from))
1da177e4 2880 return PTR_ERR(from);
1da177e4 2881
dae6ad8f 2882 dentry = user_path_create(newdfd, newname, &path, 0);
6902d925
DH		 2883 error = PTR_ERR(dentry);
2884 if (IS_ERR(dentry))
dae6ad8f 2885 goto out_putname;
6902d925 2886
dae6ad8f 2887 error = mnt_want_write(path.mnt);
75c3f29d
DH		 2888 if (error)
2889 goto out_dput;
dae6ad8f 2890 error = security_path_symlink(&path, dentry, from);
be6d3e56
KT		 2891 if (error)
2892 goto out_drop_write;
dae6ad8f 2893 error = vfs_symlink(path.dentry->d_inode, dentry, from);
be6d3e56 2894out_drop_write:
dae6ad8f 2895 mnt_drop_write(path.mnt);
75c3f29d 2896out_dput:
6902d925 2897 dput(dentry);
dae6ad8f
AV		 2898 mutex_unlock(&path.dentry->d_inode->i_mutex);
2899 path_put(&path);
6902d925 2900out_putname:
1da177e4
LT		 2901 putname(from);
2902 return error;
2903}
2904
3480b257 2905SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 2906{
2907 return sys_symlinkat(oldname, AT_FDCWD, newname);
2908}
2909
1da177e4
LT		 2910int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2911{
2912 struct inode *inode = old_dentry->d_inode;
2913 int error;
2914
2915 if (!inode)
2916 return -ENOENT;
2917
a95164d9 2918 error = may_create(dir, new_dentry);
1da177e4
LT		 2919 if (error)
2920 return error;
2921
2922 if (dir->i_sb != inode->i_sb)
2923 return -EXDEV;
2924
2925 /*
2926 * A link to an append-only or immutable file cannot be created.
2927 */
2928 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2929 return -EPERM;
acfa4380 2930 if (!dir->i_op->link)
1da177e4 2931 return -EPERM;
7e79eedb 2932 if (S_ISDIR(inode->i_mode))
1da177e4
LT		 2933 return -EPERM;
2934
2935 error = security_inode_link(old_dentry, dir, new_dentry);
2936 if (error)
2937 return error;
2938
7e79eedb 2939 mutex_lock(&inode->i_mutex);
aae8a97d
AK		 2940 /* Make sure we don't allow creating hardlink to an unlinked file */
2941 if (inode->i_nlink == 0)
2942 error = -ENOENT;
2943 else
2944 error = dir->i_op->link(old_dentry, dir, new_dentry);
7e79eedb 2945 mutex_unlock(&inode->i_mutex);
e31e14ec 2946 if (!error)
7e79eedb 2947 fsnotify_link(dir, inode, new_dentry);
1da177e4
LT		 2948 return error;
2949}
2950
2951/*
2952 * Hardlinks are often used in delicate situations. We avoid
2953 * security-related surprises by not following symlinks on the
2954 * newname. --KAB
2955 *
2956 * We don't follow them on the oldname either to be compatible
2957 * with linux 2.0, and to avoid hard-linking to directories
2958 * and other special files. --ADM
2959 */
2e4d0924
HC		 2960SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2961 int, newdfd, const char __user *, newname, int, flags)
1da177e4
LT		 2962{
2963 struct dentry *new_dentry;
dae6ad8f 2964 struct path old_path, new_path;
11a7b371 2965 int how = 0;
1da177e4 2966 int error;
1da177e4 2967
11a7b371 2968 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
c04030e1 2969 return -EINVAL;
11a7b371
AK		 2970 /*
2971 * To use null names we require CAP_DAC_READ_SEARCH
2972 * This ensures that not everyone will be able to create
2973 * handlink using the passed filedescriptor.
2974 */
2975 if (flags & AT_EMPTY_PATH) {
2976 if (!capable(CAP_DAC_READ_SEARCH))
2977 return -ENOENT;
2978 how = LOOKUP_EMPTY;
2979 }
2980
2981 if (flags & AT_SYMLINK_FOLLOW)
2982 how |= LOOKUP_FOLLOW;
c04030e1 2983
11a7b371 2984 error = user_path_at(olddfd, oldname, how, &old_path);
1da177e4 2985 if (error)
2ad94ae6
AV		 2986 return error;
2987
dae6ad8f 2988 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
1da177e4 2989 error = PTR_ERR(new_dentry);
6902d925 2990 if (IS_ERR(new_dentry))
dae6ad8f
AV		 2991 goto out;
2992
2993 error = -EXDEV;
2994 if (old_path.mnt != new_path.mnt)
2995 goto out_dput;
2996 error = mnt_want_write(new_path.mnt);
75c3f29d
DH		 2997 if (error)
2998 goto out_dput;
dae6ad8f 2999 error = security_path_link(old_path.dentry, &new_path, new_dentry);
be6d3e56
KT		 3000 if (error)
3001 goto out_drop_write;
dae6ad8f 3002 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
be6d3e56 3003out_drop_write:
dae6ad8f 3004 mnt_drop_write(new_path.mnt);
75c3f29d 3005out_dput:
6902d925 3006 dput(new_dentry);
dae6ad8f
AV		 3007 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
3008 path_put(&new_path);
1da177e4 3009out:
2d8f3038 3010 path_put(&old_path);
1da177e4
LT		 3011
3012 return error;
3013}
3014
3480b257 3015SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
5590ff0d 3016{
c04030e1 3017 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
5590ff0d
UD		 3018}
3019
1da177e4
LT		 3020/*
3021 * The worst of all namespace operations - renaming directory. "Perverted"
3022 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3023 * Problems:
3024 * a) we can get into loop creation. Check is done in is_subdir().
3025 * b) race potential - two innocent renames can create a loop together.
3026 * That's where 4.4 screws up. Current fix: serialization on
a11f3a05 3027 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
1da177e4
LT		 3028 * story.
3029 * c) we have to lock _three_ objects - parents and victim (if it exists).
1b1dcc1b 3030 * And that - after we got ->i_mutex on parents (until then we don't know
1da177e4
LT		 3031 * whether the target exists). Solution: try to be smart with locking
3032 * order for inodes. We rely on the fact that tree topology may change
a11f3a05 3033 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
1da177e4
LT		 3034 * move will be locked. Thus we can rank directories by the tree
3035 * (ancestors first) and rank all non-directories after them.
3036 * That works since everybody except rename does "lock parent, lookup,
a11f3a05 3037 * lock child" and rename is under ->s_vfs_rename_mutex.
1da177e4
LT		 3038 * HOWEVER, it relies on the assumption that any object with ->lookup()
3039 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3040 * we'd better make sure that there's no link(2) for them.
e4eaac06 3041 * d) conversion from fhandle to dentry may come in the wrong moment - when
1b1dcc1b 3042 * we are removing the target. Solution: we will have to grab ->i_mutex
1da177e4 3043 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
c41b20e7 3044 * ->i_mutex on parents, which works but leads to some truly excessive
1da177e4
LT		 3045 * locking].
3046 */
75c96f85
AB		 3047static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3048 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3049{
3050 int error = 0;
9055cba7 3051 struct inode *target = new_dentry->d_inode;
1da177e4
LT		 3052
3053 /*
3054 * If we are going to change the parent - check write permissions,
3055 * we'll need to flip '..'.
3056 */
3057 if (new_dir != old_dir) {
f419a2e3 3058 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
1da177e4
LT		 3059 if (error)
3060 return error;
3061 }
3062
3063 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3064 if (error)
3065 return error;
3066
1d2ef590 3067 dget(new_dentry);
d83c49f3 3068 if (target)
1b1dcc1b 3069 mutex_lock(&target->i_mutex);
9055cba7
SW		 3070
3071 error = -EBUSY;
3072 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3073 goto out;
3074
3cebde24
SW		 3075 if (target)
3076 shrink_dcache_parent(new_dentry);
9055cba7
SW		 3077 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3078 if (error)
3079 goto out;
3080
1da177e4 3081 if (target) {
9055cba7
SW		 3082 target->i_flags |= S_DEAD;
3083 dont_mount(new_dentry);
1da177e4 3084 }
9055cba7
SW		 3085out:
3086 if (target)
3087 mutex_unlock(&target->i_mutex);
1d2ef590 3088 dput(new_dentry);
e31e14ec 3089 if (!error)
349457cc
MF		 3090 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3091 d_move(old_dentry,new_dentry);
1da177e4
LT		 3092 return error;
3093}
3094
75c96f85
AB		 3095static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3096 struct inode *new_dir, struct dentry *new_dentry)
1da177e4 3097{
51892bbb 3098 struct inode *target = new_dentry->d_inode;
1da177e4
LT		 3099 int error;
3100
3101 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3102 if (error)
3103 return error;
3104
3105 dget(new_dentry);
1da177e4 3106 if (target)
1b1dcc1b 3107 mutex_lock(&target->i_mutex);
51892bbb
SW		 3108
3109 error = -EBUSY;
1da177e4 3110 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
51892bbb
SW		 3111 goto out;
3112
3113 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3114 if (error)
3115 goto out;
3116
3117 if (target)
3118 dont_mount(new_dentry);
3119 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3120 d_move(old_dentry, new_dentry);
3121out:
1da177e4 3122 if (target)
1b1dcc1b 3123 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3124 dput(new_dentry);
3125 return error;
3126}
3127
3128int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3129 struct inode *new_dir, struct dentry *new_dentry)
3130{
3131 int error;
3132 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
59b0df21 3133 const unsigned char *old_name;
1da177e4
LT		 3134
3135 if (old_dentry->d_inode == new_dentry->d_inode)
3136 return 0;
3137
3138 error = may_delete(old_dir, old_dentry, is_dir);
3139 if (error)
3140 return error;
3141
3142 if (!new_dentry->d_inode)
a95164d9 3143 error = may_create(new_dir, new_dentry);
1da177e4
LT		 3144 else
3145 error = may_delete(new_dir, new_dentry, is_dir);
3146 if (error)
3147 return error;
3148
acfa4380 3149 if (!old_dir->i_op->rename)
1da177e4
LT		 3150 return -EPERM;
3151
0eeca283
RL		 3152 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3153
1da177e4
LT		 3154 if (is_dir)
3155 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3156 else
3157 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
123df294
AV		 3158 if (!error)
3159 fsnotify_move(old_dir, new_dir, old_name, is_dir,
5a190ae6 3160 new_dentry->d_inode, old_dentry);
0eeca283
RL		 3161 fsnotify_oldname_free(old_name);
3162
1da177e4
LT		 3163 return error;
3164}
3165
2e4d0924
HC		 3166SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3167 int, newdfd, const char __user *, newname)
1da177e4 3168{
2ad94ae6
AV		 3169 struct dentry *old_dir, *new_dir;
3170 struct dentry *old_dentry, *new_dentry;
3171 struct dentry *trap;
1da177e4 3172 struct nameidata oldnd, newnd;
2ad94ae6
AV		 3173 char *from;
3174 char *to;
3175 int error;
1da177e4 3176
2ad94ae6 3177 error = user_path_parent(olddfd, oldname, &oldnd, &from);
1da177e4
LT		 3178 if (error)
3179 goto exit;
3180
2ad94ae6 3181 error = user_path_parent(newdfd, newname, &newnd, &to);
1da177e4
LT		 3182 if (error)
3183 goto exit1;
3184
3185 error = -EXDEV;
4ac91378 3186 if (oldnd.path.mnt != newnd.path.mnt)
1da177e4
LT		 3187 goto exit2;
3188
4ac91378 3189 old_dir = oldnd.path.dentry;
1da177e4
LT		 3190 error = -EBUSY;
3191 if (oldnd.last_type != LAST_NORM)
3192 goto exit2;
3193
4ac91378 3194 new_dir = newnd.path.dentry;
1da177e4
LT		 3195 if (newnd.last_type != LAST_NORM)
3196 goto exit2;
3197
0612d9fb
OH		 3198 oldnd.flags &= ~LOOKUP_PARENT;
3199 newnd.flags &= ~LOOKUP_PARENT;
4e9ed2f8 3200 newnd.flags |= LOOKUP_RENAME_TARGET;
0612d9fb 3201
1da177e4
LT		 3202 trap = lock_rename(new_dir, old_dir);
3203
49705b77 3204 old_dentry = lookup_hash(&oldnd);
1da177e4
LT		 3205 error = PTR_ERR(old_dentry);
3206 if (IS_ERR(old_dentry))
3207 goto exit3;
3208 /* source must exist */
3209 error = -ENOENT;
3210 if (!old_dentry->d_inode)
3211 goto exit4;
3212 /* unless the source is a directory trailing slashes give -ENOTDIR */
3213 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3214 error = -ENOTDIR;
3215 if (oldnd.last.name[oldnd.last.len])
3216 goto exit4;
3217 if (newnd.last.name[newnd.last.len])
3218 goto exit4;
3219 }
3220 /* source should not be ancestor of target */
3221 error = -EINVAL;
3222 if (old_dentry == trap)
3223 goto exit4;
49705b77 3224 new_dentry = lookup_hash(&newnd);
1da177e4
LT		 3225 error = PTR_ERR(new_dentry);
3226 if (IS_ERR(new_dentry))
3227 goto exit4;
3228 /* target should not be an ancestor of source */
3229 error = -ENOTEMPTY;
3230 if (new_dentry == trap)
3231 goto exit5;
3232
9079b1eb
DH		 3233 error = mnt_want_write(oldnd.path.mnt);
3234 if (error)
3235 goto exit5;
be6d3e56
KT		 3236 error = security_path_rename(&oldnd.path, old_dentry,
3237 &newnd.path, new_dentry);
3238 if (error)
3239 goto exit6;
1da177e4
LT		 3240 error = vfs_rename(old_dir->d_inode, old_dentry,
3241 new_dir->d_inode, new_dentry);
be6d3e56 3242exit6:
9079b1eb 3243 mnt_drop_write(oldnd.path.mnt);
1da177e4
LT		 3244exit5:
3245 dput(new_dentry);
3246exit4:
3247 dput(old_dentry);
3248exit3:
3249 unlock_rename(new_dir, old_dir);
3250exit2:
1d957f9b 3251 path_put(&newnd.path);
2ad94ae6 3252 putname(to);
1da177e4 3253exit1:
1d957f9b 3254 path_put(&oldnd.path);
1da177e4 3255 putname(from);
2ad94ae6 3256exit:
1da177e4
LT		 3257 return error;
3258}
3259
a26eab24 3260SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 3261{
3262 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3263}
3264
1da177e4
LT		 3265int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3266{
3267 int len;
3268
3269 len = PTR_ERR(link);
3270 if (IS_ERR(link))
3271 goto out;
3272
3273 len = strlen(link);
3274 if (len > (unsigned) buflen)
3275 len = buflen;
3276 if (copy_to_user(buffer, link, len))
3277 len = -EFAULT;
3278out:
3279 return len;
3280}
3281
3282/*
3283 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3284 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3285 * using) it for any given inode is up to filesystem.
3286 */
3287int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3288{
3289 struct nameidata nd;
cc314eef 3290 void *cookie;
694a1764 3291 int res;
cc314eef 3292
1da177e4 3293 nd.depth = 0;
cc314eef 3294 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
694a1764
MS		 3295 if (IS_ERR(cookie))
3296 return PTR_ERR(cookie);
3297
3298 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3299 if (dentry->d_inode->i_op->put_link)
3300 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3301 return res;
1da177e4
LT		 3302}
3303
3304int vfs_follow_link(struct nameidata *nd, const char *link)
3305{
3306 return __vfs_follow_link(nd, link);
3307}
3308
3309/* get the link contents into pagecache */
3310static char *page_getlink(struct dentry * dentry, struct page **ppage)
3311{
ebd09abb
DG		 3312 char *kaddr;
3313 struct page *page;
1da177e4 3314 struct address_space *mapping = dentry->d_inode->i_mapping;
090d2b18 3315 page = read_mapping_page(mapping, 0, NULL);
1da177e4 3316 if (IS_ERR(page))
6fe6900e 3317 return (char*)page;
1da177e4 3318 *ppage = page;
ebd09abb
DG		 3319 kaddr = kmap(page);
3320 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3321 return kaddr;
1da177e4
LT		 3322}
3323
3324int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3325{
3326 struct page *page = NULL;
3327 char *s = page_getlink(dentry, &page);
3328 int res = vfs_readlink(dentry,buffer,buflen,s);
3329 if (page) {
3330 kunmap(page);
3331 page_cache_release(page);
3332 }
3333 return res;
3334}
3335
cc314eef 3336void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
1da177e4 3337{
cc314eef 3338 struct page *page = NULL;
1da177e4 3339 nd_set_link(nd, page_getlink(dentry, &page));
cc314eef 3340 return page;
1da177e4
LT		 3341}
3342
cc314eef 3343void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1da177e4 3344{
cc314eef
LT		 3345 struct page *page = cookie;
3346
3347 if (page) {
1da177e4
LT		 3348 kunmap(page);
3349 page_cache_release(page);
1da177e4
LT		 3350 }
3351}
3352
54566b2c
NP		 3353/*
3354 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3355 */
3356int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
1da177e4
LT		 3357{
3358 struct address_space *mapping = inode->i_mapping;
0adb25d2 3359 struct page *page;
afddba49 3360 void *fsdata;
beb497ab 3361 int err;
1da177e4 3362 char *kaddr;
54566b2c
NP		 3363 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3364 if (nofs)
3365 flags |= AOP_FLAG_NOFS;
1da177e4 3366
7e53cac4 3367retry:
afddba49 3368 err = pagecache_write_begin(NULL, mapping, 0, len-1,
54566b2c 3369 flags, &page, &fsdata);
1da177e4 3370 if (err)
afddba49
NP		 3371 goto fail;
3372
1da177e4
LT		 3373 kaddr = kmap_atomic(page, KM_USER0);
3374 memcpy(kaddr, symname, len-1);
3375 kunmap_atomic(kaddr, KM_USER0);
afddba49
NP		 3376
3377 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3378 page, fsdata);
1da177e4
LT		 3379 if (err < 0)
3380 goto fail;
afddba49
NP		 3381 if (err < len-1)
3382 goto retry;
3383
1da177e4
LT		 3384 mark_inode_dirty(inode);
3385 return 0;
1da177e4
LT		 3386fail:
3387 return err;
3388}
3389
0adb25d2
KK		 3390int page_symlink(struct inode *inode, const char *symname, int len)
3391{
3392 return __page_symlink(inode, symname, len,
54566b2c 3393 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
0adb25d2
KK		 3394}
3395
92e1d5be 3396const struct inode_operations page_symlink_inode_operations = {
1da177e4
LT		 3397 .readlink = generic_readlink,
3398 .follow_link = page_follow_link_light,
3399 .put_link = page_put_link,
3400};
3401
2d8f3038 3402EXPORT_SYMBOL(user_path_at);
cc53ce53 3403EXPORT_SYMBOL(follow_down_one);
1da177e4
LT		 3404EXPORT_SYMBOL(follow_down);
3405EXPORT_SYMBOL(follow_up);
3406EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3407EXPORT_SYMBOL(getname);
3408EXPORT_SYMBOL(lock_rename);
1da177e4
LT		 3409EXPORT_SYMBOL(lookup_one_len);
3410EXPORT_SYMBOL(page_follow_link_light);
3411EXPORT_SYMBOL(page_put_link);
3412EXPORT_SYMBOL(page_readlink);
0adb25d2 3413EXPORT_SYMBOL(__page_symlink);
1da177e4
LT		 3414EXPORT_SYMBOL(page_symlink);
3415EXPORT_SYMBOL(page_symlink_inode_operations);
d1811465 3416EXPORT_SYMBOL(kern_path);
16f18200 3417EXPORT_SYMBOL(vfs_path_lookup);
f419a2e3 3418EXPORT_SYMBOL(inode_permission);
1da177e4
LT		 3419EXPORT_SYMBOL(unlock_rename);
3420EXPORT_SYMBOL(vfs_create);
3421EXPORT_SYMBOL(vfs_follow_link);
3422EXPORT_SYMBOL(vfs_link);
3423EXPORT_SYMBOL(vfs_mkdir);
3424EXPORT_SYMBOL(vfs_mknod);
3425EXPORT_SYMBOL(generic_permission);
3426EXPORT_SYMBOL(vfs_readlink);
3427EXPORT_SYMBOL(vfs_rename);
3428EXPORT_SYMBOL(vfs_rmdir);
3429EXPORT_SYMBOL(vfs_symlink);
3430EXPORT_SYMBOL(vfs_unlink);
3431EXPORT_SYMBOL(dentry_unhash);
3432EXPORT_SYMBOL(generic_readlink);
kernel source for telekom puls (alcatel/mediatek)