projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
userns: Convert in_group_p and in_egroup_p to use kgid_t
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / fs / namei.c
CommitLineData
1da177e4
LT		 1/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
630d9c47 18#include <linux/export.h>
1da177e4
LT		 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{
f7493e5d 164 return getname_flags(filename, 0, NULL);
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
14067ff5 231 if (likely(current_fsuid() == inode->i_uid))
1da177e4
LT		 232 mode >>= 6;
233 else {
e77819e5 234 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
7e40145e 235 int error = check_acl(inode, mask);
b74c79e9
NP		 236 if (error != -EAGAIN)
237 return error;
1da177e4
LT		 238 }
239
240 if (in_group_p(inode->i_gid))
241 mode >>= 3;
242 }
243
244 /*
245 * If the DACs are ok we don't need any capability check.
246 */
9c2c7039 247 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
1da177e4 248 return 0;
5909ccaa
LT		 249 return -EACCES;
250}
251
252/**
b74c79e9 253 * generic_permission - check for access rights on a Posix-like filesystem
5909ccaa 254 * @inode: inode to check access rights for
8fd90c8d 255 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
5909ccaa
LT		 256 *
257 * Used to check for read/write/execute permissions on a file.
258 * We use "fsuid" for this, letting us set arbitrary permissions
259 * for filesystem access without changing the "normal" uids which
b74c79e9
NP		 260 * are used for other things.
261 *
262 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
263 * request cannot be satisfied (eg. requires blocking or too much complexity).
264 * It would then be called again in ref-walk mode.
5909ccaa 265 */
2830ba7f 266int generic_permission(struct inode *inode, int mask)
5909ccaa
LT		 267{
268 int ret;
269
270 /*
948409c7 271 * Do the basic permission checks.
5909ccaa 272 */
7e40145e 273 ret = acl_permission_check(inode, mask);
5909ccaa
LT		 274 if (ret != -EACCES)
275 return ret;
1da177e4 276
d594e7ec
AV		 277 if (S_ISDIR(inode->i_mode)) {
278 /* DACs are overridable for directories */
1a48e2ac 279 if (inode_capable(inode, CAP_DAC_OVERRIDE))
d594e7ec
AV		 280 return 0;
281 if (!(mask & MAY_WRITE))
1a48e2ac 282 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
d594e7ec
AV		 283 return 0;
284 return -EACCES;
285 }
1da177e4
LT		 286 /*
287 * Read/write DACs are always overridable.
d594e7ec
AV		 288 * Executable DACs are overridable when there is
289 * at least one exec bit set.
1da177e4 290 */
d594e7ec 291 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
1a48e2ac 292 if (inode_capable(inode, CAP_DAC_OVERRIDE))
1da177e4
LT		 293 return 0;
294
295 /*
296 * Searching includes executable on directories, else just read.
297 */
7ea66001 298 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
d594e7ec 299 if (mask == MAY_READ)
1a48e2ac 300 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
1da177e4
LT		 301 return 0;
302
303 return -EACCES;
304}
305
3ddcd056
LT		 306/*
307 * We _really_ want to just do "generic_permission()" without
308 * even looking at the inode->i_op values. So we keep a cache
309 * flag in inode->i_opflags, that says "this has not special
310 * permission function, use the fast case".
311 */
312static inline int do_inode_permission(struct inode *inode, int mask)
313{
314 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
315 if (likely(inode->i_op->permission))
316 return inode->i_op->permission(inode, mask);
317
318 /* This gets set once for the inode lifetime */
319 spin_lock(&inode->i_lock);
320 inode->i_opflags |= IOP_FASTPERM;
321 spin_unlock(&inode->i_lock);
322 }
323 return generic_permission(inode, mask);
324}
325
cb23beb5
CH		 326/**
327 * inode_permission - check for access rights to a given inode
328 * @inode: inode to check permission on
8fd90c8d 329 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
cb23beb5
CH		 330 *
331 * Used to check for read/write/execute permissions on an inode.
332 * We use "fsuid" for this, letting us set arbitrary permissions
333 * for filesystem access without changing the "normal" uids which
334 * are used for other things.
948409c7
AG		 335 *
336 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
cb23beb5 337 */
f419a2e3 338int inode_permission(struct inode *inode, int mask)
1da177e4 339{
e6305c43 340 int retval;
1da177e4 341
3ddcd056 342 if (unlikely(mask & MAY_WRITE)) {
22590e41 343 umode_t mode = inode->i_mode;
1da177e4
LT		 344
345 /*
346 * Nobody gets write access to a read-only fs.
347 */
348 if (IS_RDONLY(inode) &&
349 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
350 return -EROFS;
351
352 /*
353 * Nobody gets write access to an immutable file.
354 */
355 if (IS_IMMUTABLE(inode))
356 return -EACCES;
357 }
358
3ddcd056 359 retval = do_inode_permission(inode, mask);
1da177e4
LT		 360 if (retval)
361 return retval;
362
08ce5f16
SH		 363 retval = devcgroup_inode_permission(inode, mask);
364 if (retval)
365 return retval;
366
d09ca739 367 return security_inode_permission(inode, mask);
1da177e4
LT		 368}
369
5dd784d0
JB		 370/**
371 * path_get - get a reference to a path
372 * @path: path to get the reference to
373 *
374 * Given a path increment the reference count to the dentry and the vfsmount.
375 */
376void path_get(struct path *path)
377{
378 mntget(path->mnt);
379 dget(path->dentry);
380}
381EXPORT_SYMBOL(path_get);
382
1d957f9b
JB		 383/**
384 * path_put - put a reference to a path
385 * @path: path to put the reference to
386 *
387 * Given a path decrement the reference count to the dentry and the vfsmount.
388 */
389void path_put(struct path *path)
1da177e4 390{
1d957f9b
JB		 391 dput(path->dentry);
392 mntput(path->mnt);
1da177e4 393}
1d957f9b 394EXPORT_SYMBOL(path_put);
1da177e4 395
19660af7 396/*
31e6b01f 397 * Path walking has 2 modes, rcu-walk and ref-walk (see
19660af7
AV		 398 * Documentation/filesystems/path-lookup.txt). In situations when we can't
399 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
400 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
401 * mode. Refcounts are grabbed at the last known good point before rcu-walk
402 * got stuck, so ref-walk may continue from there. If this is not successful
403 * (eg. a seqcount has changed), then failure is returned and it's up to caller
404 * to restart the path walk from the beginning in ref-walk mode.
31e6b01f 405 */
31e6b01f
NP		 406
407/**
19660af7
AV		 408 * unlazy_walk - try to switch to ref-walk mode.
409 * @nd: nameidata pathwalk data
410 * @dentry: child of nd->path.dentry or NULL
39191628 411 * Returns: 0 on success, -ECHILD on failure
31e6b01f 412 *
19660af7
AV		 413 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
414 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
415 * @nd or NULL. Must be called from rcu-walk context.
31e6b01f 416 */
19660af7 417static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
31e6b01f
NP		 418{
419 struct fs_struct *fs = current->fs;
420 struct dentry *parent = nd->path.dentry;
5b6ca027 421 int want_root = 0;
31e6b01f
NP		 422
423 BUG_ON(!(nd->flags & LOOKUP_RCU));
5b6ca027
AV		 424 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
425 want_root = 1;
31e6b01f
NP		 426 spin_lock(&fs->lock);
427 if (nd->root.mnt != fs->root.mnt ||
428 nd->root.dentry != fs->root.dentry)
429 goto err_root;
430 }
431 spin_lock(&parent->d_lock);
19660af7
AV		 432 if (!dentry) {
433 if (!__d_rcu_to_refcount(parent, nd->seq))
434 goto err_parent;
435 BUG_ON(nd->inode != parent->d_inode);
436 } else {
94c0d4ec
AV		 437 if (dentry->d_parent != parent)
438 goto err_parent;
19660af7
AV		 439 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
440 if (!__d_rcu_to_refcount(dentry, nd->seq))
441 goto err_child;
442 /*
443 * If the sequence check on the child dentry passed, then
444 * the child has not been removed from its parent. This
445 * means the parent dentry must be valid and able to take
446 * a reference at this point.
447 */
448 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
449 BUG_ON(!parent->d_count);
450 parent->d_count++;
451 spin_unlock(&dentry->d_lock);
452 }
31e6b01f 453 spin_unlock(&parent->d_lock);
5b6ca027 454 if (want_root) {
31e6b01f
NP		 455 path_get(&nd->root);
456 spin_unlock(&fs->lock);
457 }
458 mntget(nd->path.mnt);
459
460 rcu_read_unlock();
461 br_read_unlock(vfsmount_lock);
462 nd->flags &= ~LOOKUP_RCU;
463 return 0;
19660af7
AV		 464
465err_child:
31e6b01f 466 spin_unlock(&dentry->d_lock);
19660af7 467err_parent:
31e6b01f
NP		 468 spin_unlock(&parent->d_lock);
469err_root:
5b6ca027 470 if (want_root)
31e6b01f
NP		 471 spin_unlock(&fs->lock);
472 return -ECHILD;
473}
474
834f2a4a
TM		 475/**
476 * release_open_intent - free up open intent resources
477 * @nd: pointer to nameidata
478 */
479void release_open_intent(struct nameidata *nd)
480{
2dab5974
LT		 481 struct file *file = nd->intent.open.file;
482
483 if (file && !IS_ERR(file)) {
484 if (file->f_path.dentry == NULL)
485 put_filp(file);
486 else
487 fput(file);
488 }
834f2a4a
TM		 489}
490
f60aef7e 491static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
34286d66 492{
f60aef7e 493 return dentry->d_op->d_revalidate(dentry, nd);
34286d66
NP		 494}
495
9f1fafee
AV		 496/**
497 * complete_walk - successful completion of path walk
498 * @nd: pointer nameidata
39159de2 499 *
9f1fafee
AV		 500 * If we had been in RCU mode, drop out of it and legitimize nd->path.
501 * Revalidate the final result, unless we'd already done that during
502 * the path walk or the filesystem doesn't ask for it. Return 0 on
503 * success, -error on failure. In case of failure caller does not
504 * need to drop nd->path.
39159de2 505 */
9f1fafee 506static int complete_walk(struct nameidata *nd)
39159de2 507{
16c2cd71 508 struct dentry *dentry = nd->path.dentry;
39159de2 509 int status;
39159de2 510
9f1fafee
AV		 511 if (nd->flags & LOOKUP_RCU) {
512 nd->flags &= ~LOOKUP_RCU;
513 if (!(nd->flags & LOOKUP_ROOT))
514 nd->root.mnt = NULL;
515 spin_lock(&dentry->d_lock);
516 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
517 spin_unlock(&dentry->d_lock);
518 rcu_read_unlock();
519 br_read_unlock(vfsmount_lock);
520 return -ECHILD;
521 }
522 BUG_ON(nd->inode != dentry->d_inode);
523 spin_unlock(&dentry->d_lock);
524 mntget(nd->path.mnt);
525 rcu_read_unlock();
526 br_read_unlock(vfsmount_lock);
527 }
528
16c2cd71
AV		 529 if (likely(!(nd->flags & LOOKUP_JUMPED)))
530 return 0;
531
532 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
39159de2
JL		 533 return 0;
534
16c2cd71
AV		 535 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
536 return 0;
537
538 /* Note: we do not d_invalidate() */
34286d66 539 status = d_revalidate(dentry, nd);
39159de2
JL		 540 if (status > 0)
541 return 0;
542
16c2cd71 543 if (!status)
39159de2 544 status = -ESTALE;
16c2cd71 545
9f1fafee 546 path_put(&nd->path);
39159de2
JL		 547 return status;
548}
549
2a737871
AV		 550static __always_inline void set_root(struct nameidata *nd)
551{
f7ad3c6b
MS		 552 if (!nd->root.mnt)
553 get_fs_root(current->fs, &nd->root);
2a737871
AV		 554}
555
6de88d72
AV		 556static int link_path_walk(const char *, struct nameidata *);
557
31e6b01f
NP		 558static __always_inline void set_root_rcu(struct nameidata *nd)
559{
560 if (!nd->root.mnt) {
561 struct fs_struct *fs = current->fs;
c28cc364
NP		 562 unsigned seq;
563
564 do {
565 seq = read_seqcount_begin(&fs->seq);
566 nd->root = fs->root;
c1530019 567 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
c28cc364 568 } while (read_seqcount_retry(&fs->seq, seq));
31e6b01f
NP		 569 }
570}
571
f1662356 572static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
1da177e4 573{
31e6b01f
NP		 574 int ret;
575
1da177e4
LT		 576 if (IS_ERR(link))
577 goto fail;
578
579 if (*link == '/') {
2a737871 580 set_root(nd);
1d957f9b 581 path_put(&nd->path);
2a737871
AV		 582 nd->path = nd->root;
583 path_get(&nd->root);
16c2cd71 584 nd->flags |= LOOKUP_JUMPED;
1da177e4 585 }
31e6b01f 586 nd->inode = nd->path.dentry->d_inode;
b4091d5f 587
31e6b01f
NP		 588 ret = link_path_walk(link, nd);
589 return ret;
1da177e4 590fail:
1d957f9b 591 path_put(&nd->path);
1da177e4
LT		 592 return PTR_ERR(link);
593}
594
1d957f9b 595static void path_put_conditional(struct path *path, struct nameidata *nd)
051d3812
IK		 596{
597 dput(path->dentry);
4ac91378 598 if (path->mnt != nd->path.mnt)
051d3812
IK		 599 mntput(path->mnt);
600}
601
7b9337aa
NP		 602static inline void path_to_nameidata(const struct path *path,
603 struct nameidata *nd)
051d3812 604{
31e6b01f
NP		 605 if (!(nd->flags & LOOKUP_RCU)) {
606 dput(nd->path.dentry);
607 if (nd->path.mnt != path->mnt)
608 mntput(nd->path.mnt);
9a229683 609 }
31e6b01f 610 nd->path.mnt = path->mnt;
4ac91378 611 nd->path.dentry = path->dentry;
051d3812
IK		 612}
613
574197e0
AV		 614static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
615{
616 struct inode *inode = link->dentry->d_inode;
617 if (!IS_ERR(cookie) && inode->i_op->put_link)
618 inode->i_op->put_link(link->dentry, nd, cookie);
619 path_put(link);
620}
621
def4af30 622static __always_inline int
574197e0 623follow_link(struct path *link, struct nameidata *nd, void **p)
1da177e4
LT		 624{
625 int error;
7b9337aa 626 struct dentry *dentry = link->dentry;
1da177e4 627
844a3917
AV		 628 BUG_ON(nd->flags & LOOKUP_RCU);
629
0e794589
AV		 630 if (link->mnt == nd->path.mnt)
631 mntget(link->mnt);
632
574197e0
AV		 633 if (unlikely(current->total_link_count >= 40)) {
634 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
574197e0
AV		 635 path_put(&nd->path);
636 return -ELOOP;
637 }
638 cond_resched();
639 current->total_link_count++;
640
68ac1234 641 touch_atime(link);
1da177e4 642 nd_set_link(nd, NULL);
cd4e91d3 643
36f3b4f6
AV		 644 error = security_inode_follow_link(link->dentry, nd);
645 if (error) {
646 *p = ERR_PTR(error); /* no ->put_link(), please */
647 path_put(&nd->path);
648 return error;
649 }
650
86acdca1 651 nd->last_type = LAST_BIND;
def4af30
AV		 652 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
653 error = PTR_ERR(*p);
654 if (!IS_ERR(*p)) {
1da177e4 655 char *s = nd_get_link(nd);
cc314eef 656 error = 0;
1da177e4
LT		 657 if (s)
658 error = __vfs_follow_link(nd, s);
bcda7652 659 else if (nd->last_type == LAST_BIND) {
16c2cd71 660 nd->flags |= LOOKUP_JUMPED;
b21041d0
AV		 661 nd->inode = nd->path.dentry->d_inode;
662 if (nd->inode->i_op->follow_link) {
bcda7652
AV		 663 /* stepped on a _really_ weird one */
664 path_put(&nd->path);
665 error = -ELOOP;
666 }
667 }
1da177e4 668 }
1da177e4
LT		 669 return error;
670}
671
31e6b01f
NP		 672static int follow_up_rcu(struct path *path)
673{
0714a533
AV		 674 struct mount *mnt = real_mount(path->mnt);
675 struct mount *parent;
31e6b01f
NP		 676 struct dentry *mountpoint;
677
0714a533
AV		 678 parent = mnt->mnt_parent;
679 if (&parent->mnt == path->mnt)
31e6b01f 680 return 0;
a73324da 681 mountpoint = mnt->mnt_mountpoint;
31e6b01f 682 path->dentry = mountpoint;
0714a533 683 path->mnt = &parent->mnt;
31e6b01f
NP		 684 return 1;
685}
686
bab77ebf 687int follow_up(struct path *path)
1da177e4 688{
0714a533
AV		 689 struct mount *mnt = real_mount(path->mnt);
690 struct mount *parent;
1da177e4 691 struct dentry *mountpoint;
99b7db7b
NP		 692
693 br_read_lock(vfsmount_lock);
0714a533
AV		 694 parent = mnt->mnt_parent;
695 if (&parent->mnt == path->mnt) {
99b7db7b 696 br_read_unlock(vfsmount_lock);
1da177e4
LT		 697 return 0;
698 }
0714a533 699 mntget(&parent->mnt);
a73324da 700 mountpoint = dget(mnt->mnt_mountpoint);
99b7db7b 701 br_read_unlock(vfsmount_lock);
bab77ebf
AV		 702 dput(path->dentry);
703 path->dentry = mountpoint;
704 mntput(path->mnt);
0714a533 705 path->mnt = &parent->mnt;
1da177e4
LT		 706 return 1;
707}
708
b5c84bf6 709/*
9875cf80
DH		 710 * Perform an automount
711 * - return -EISDIR to tell follow_managed() to stop and return the path we
712 * were called with.
1da177e4 713 */
9875cf80
DH		 714static int follow_automount(struct path *path, unsigned flags,
715 bool *need_mntput)
31e6b01f 716{
9875cf80 717 struct vfsmount *mnt;
ea5b778a 718 int err;
9875cf80
DH		 719
720 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
721 return -EREMOTE;
722
0ec26fd0
MS		 723 /* We don't want to mount if someone's just doing a stat -
724 * unless they're stat'ing a directory and appended a '/' to
725 * the name.
726 *
727 * We do, however, want to mount if someone wants to open or
728 * create a file of any type under the mountpoint, wants to
729 * traverse through the mountpoint or wants to open the
730 * mounted directory. Also, autofs may mark negative dentries
731 * as being automount points. These will need the attentions
732 * of the daemon to instantiate them before they can be used.
9875cf80 733 */
0ec26fd0 734 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
d94c177b 735 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
0ec26fd0
MS		 736 path->dentry->d_inode)
737 return -EISDIR;
738
9875cf80
DH		 739 current->total_link_count++;
740 if (current->total_link_count >= 40)
741 return -ELOOP;
742
743 mnt = path->dentry->d_op->d_automount(path);
744 if (IS_ERR(mnt)) {
745 /*
746 * The filesystem is allowed to return -EISDIR here to indicate
747 * it doesn't want to automount. For instance, autofs would do
748 * this so that its userspace daemon can mount on this dentry.
749 *
750 * However, we can only permit this if it's a terminal point in
751 * the path being looked up; if it wasn't then the remainder of
752 * the path is inaccessible and we should say so.
753 */
49084c3b 754 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
9875cf80
DH		 755 return -EREMOTE;
756 return PTR_ERR(mnt);
31e6b01f 757 }
ea5b778a 758
9875cf80
DH		 759 if (!mnt) /* mount collision */
760 return 0;
31e6b01f 761
8aef1884
AV		 762 if (!*need_mntput) {
763 /* lock_mount() may release path->mnt on error */
764 mntget(path->mnt);
765 *need_mntput = true;
766 }
19a167af 767 err = finish_automount(mnt, path);
9875cf80 768
ea5b778a
DH		 769 switch (err) {
770 case -EBUSY:
771 /* Someone else made a mount here whilst we were busy */
19a167af 772 return 0;
ea5b778a 773 case 0:
8aef1884 774 path_put(path);
ea5b778a
DH		 775 path->mnt = mnt;
776 path->dentry = dget(mnt->mnt_root);
ea5b778a 777 return 0;
19a167af
AV		 778 default:
779 return err;
ea5b778a 780 }
19a167af 781
463ffb2e
AV		 782}
783
9875cf80
DH		 784/*
785 * Handle a dentry that is managed in some way.
cc53ce53 786 * - Flagged for transit management (autofs)
9875cf80
DH		 787 * - Flagged as mountpoint
788 * - Flagged as automount point
789 *
790 * This may only be called in refwalk mode.
791 *
792 * Serialization is taken care of in namespace.c
793 */
794static int follow_managed(struct path *path, unsigned flags)
1da177e4 795{
8aef1884 796 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
9875cf80
DH		 797 unsigned managed;
798 bool need_mntput = false;
8aef1884 799 int ret = 0;
9875cf80
DH		 800
801 /* Given that we're not holding a lock here, we retain the value in a
802 * local variable for each dentry as we look at it so that we don't see
803 * the components of that value change under us */
804 while (managed = ACCESS_ONCE(path->dentry->d_flags),
805 managed &= DCACHE_MANAGED_DENTRY,
806 unlikely(managed != 0)) {
cc53ce53
DH		 807 /* Allow the filesystem to manage the transit without i_mutex
808 * being held. */
809 if (managed & DCACHE_MANAGE_TRANSIT) {
810 BUG_ON(!path->dentry->d_op);
811 BUG_ON(!path->dentry->d_op->d_manage);
1aed3e42 812 ret = path->dentry->d_op->d_manage(path->dentry, false);
cc53ce53 813 if (ret < 0)
8aef1884 814 break;
cc53ce53
DH		 815 }
816
9875cf80
DH		 817 /* Transit to a mounted filesystem. */
818 if (managed & DCACHE_MOUNTED) {
819 struct vfsmount *mounted = lookup_mnt(path);
820 if (mounted) {
821 dput(path->dentry);
822 if (need_mntput)
823 mntput(path->mnt);
824 path->mnt = mounted;
825 path->dentry = dget(mounted->mnt_root);
826 need_mntput = true;
827 continue;
828 }
829
830 /* Something is mounted on this dentry in another
831 * namespace and/or whatever was mounted there in this
832 * namespace got unmounted before we managed to get the
833 * vfsmount_lock */
834 }
835
836 /* Handle an automount point */
837 if (managed & DCACHE_NEED_AUTOMOUNT) {
838 ret = follow_automount(path, flags, &need_mntput);
839 if (ret < 0)
8aef1884 840 break;
9875cf80
DH		 841 continue;
842 }
843
844 /* We didn't change the current path point */
845 break;
1da177e4 846 }
8aef1884
AV		 847
848 if (need_mntput && path->mnt == mnt)
849 mntput(path->mnt);
850 if (ret == -EISDIR)
851 ret = 0;
a3fbbde7 852 return ret < 0 ? ret : need_mntput;
1da177e4
LT		 853}
854
cc53ce53 855int follow_down_one(struct path *path)
1da177e4
LT		 856{
857 struct vfsmount *mounted;
858
1c755af4 859 mounted = lookup_mnt(path);
1da177e4 860 if (mounted) {
9393bd07
AV		 861 dput(path->dentry);
862 mntput(path->mnt);
863 path->mnt = mounted;
864 path->dentry = dget(mounted->mnt_root);
1da177e4
LT		 865 return 1;
866 }
867 return 0;
868}
869
62a7375e
IK		 870static inline bool managed_dentry_might_block(struct dentry *dentry)
871{
872 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
873 dentry->d_op->d_manage(dentry, true) < 0);
874}
875
9875cf80 876/*
287548e4
AV		 877 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
878 * we meet a managed dentry that would need blocking.
9875cf80
DH		 879 */
880static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
287548e4 881 struct inode **inode)
9875cf80 882{
62a7375e 883 for (;;) {
c7105365 884 struct mount *mounted;
62a7375e
IK		 885 /*
886 * Don't forget we might have a non-mountpoint managed dentry
887 * that wants to block transit.
888 */
287548e4 889 if (unlikely(managed_dentry_might_block(path->dentry)))
ab90911f 890 return false;
62a7375e
IK		 891
892 if (!d_mountpoint(path->dentry))
893 break;
894
9875cf80
DH		 895 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
896 if (!mounted)
897 break;
c7105365
AV		 898 path->mnt = &mounted->mnt;
899 path->dentry = mounted->mnt.mnt_root;
a3fbbde7 900 nd->flags |= LOOKUP_JUMPED;
9875cf80 901 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
59430262
LT		 902 /*
903 * Update the inode too. We don't need to re-check the
904 * dentry sequence number here after this d_inode read,
905 * because a mount-point is always pinned.
906 */
907 *inode = path->dentry->d_inode;
9875cf80 908 }
9875cf80
DH		 909 return true;
910}
911
dea39376 912static void follow_mount_rcu(struct nameidata *nd)
287548e4 913{
dea39376 914 while (d_mountpoint(nd->path.dentry)) {
c7105365 915 struct mount *mounted;
dea39376 916 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
287548e4
AV		 917 if (!mounted)
918 break;
c7105365
AV		 919 nd->path.mnt = &mounted->mnt;
920 nd->path.dentry = mounted->mnt.mnt_root;
dea39376 921 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
287548e4
AV		 922 }
923}
924
31e6b01f
NP		 925static int follow_dotdot_rcu(struct nameidata *nd)
926{
31e6b01f
NP		 927 set_root_rcu(nd);
928
9875cf80 929 while (1) {
31e6b01f
NP		 930 if (nd->path.dentry == nd->root.dentry &&
931 nd->path.mnt == nd->root.mnt) {
932 break;
933 }
934 if (nd->path.dentry != nd->path.mnt->mnt_root) {
935 struct dentry *old = nd->path.dentry;
936 struct dentry *parent = old->d_parent;
937 unsigned seq;
938
939 seq = read_seqcount_begin(&parent->d_seq);
940 if (read_seqcount_retry(&old->d_seq, nd->seq))
ef7562d5 941 goto failed;
31e6b01f
NP		 942 nd->path.dentry = parent;
943 nd->seq = seq;
944 break;
945 }
946 if (!follow_up_rcu(&nd->path))
947 break;
948 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
31e6b01f 949 }
dea39376
AV		 950 follow_mount_rcu(nd);
951 nd->inode = nd->path.dentry->d_inode;
31e6b01f 952 return 0;
ef7562d5
AV		 953
954failed:
955 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 956 if (!(nd->flags & LOOKUP_ROOT))
957 nd->root.mnt = NULL;
ef7562d5
AV		 958 rcu_read_unlock();
959 br_read_unlock(vfsmount_lock);
960 return -ECHILD;
31e6b01f
NP		 961}
962
cc53ce53
DH		 963/*
964 * Follow down to the covering mount currently visible to userspace. At each
965 * point, the filesystem owning that dentry may be queried as to whether the
966 * caller is permitted to proceed or not.
cc53ce53 967 */
7cc90cc3 968int follow_down(struct path *path)
cc53ce53
DH		 969{
970 unsigned managed;
971 int ret;
972
973 while (managed = ACCESS_ONCE(path->dentry->d_flags),
974 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
975 /* Allow the filesystem to manage the transit without i_mutex
976 * being held.
977 *
978 * We indicate to the filesystem if someone is trying to mount
979 * something here. This gives autofs the chance to deny anyone
980 * other than its daemon the right to mount on its
981 * superstructure.
982 *
983 * The filesystem may sleep at this point.
984 */
985 if (managed & DCACHE_MANAGE_TRANSIT) {
986 BUG_ON(!path->dentry->d_op);
987 BUG_ON(!path->dentry->d_op->d_manage);
ab90911f 988 ret = path->dentry->d_op->d_manage(
1aed3e42 989 path->dentry, false);
cc53ce53
DH		 990 if (ret < 0)
991 return ret == -EISDIR ? 0 : ret;
992 }
993
994 /* Transit to a mounted filesystem. */
995 if (managed & DCACHE_MOUNTED) {
996 struct vfsmount *mounted = lookup_mnt(path);
997 if (!mounted)
998 break;
999 dput(path->dentry);
1000 mntput(path->mnt);
1001 path->mnt = mounted;
1002 path->dentry = dget(mounted->mnt_root);
1003 continue;
1004 }
1005
1006 /* Don't handle automount points here */
1007 break;
1008 }
1009 return 0;
1010}
1011
9875cf80
DH		 1012/*
1013 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1014 */
1015static void follow_mount(struct path *path)
1016{
1017 while (d_mountpoint(path->dentry)) {
1018 struct vfsmount *mounted = lookup_mnt(path);
1019 if (!mounted)
1020 break;
1021 dput(path->dentry);
1022 mntput(path->mnt);
1023 path->mnt = mounted;
1024 path->dentry = dget(mounted->mnt_root);
1025 }
1026}
1027
31e6b01f 1028static void follow_dotdot(struct nameidata *nd)
1da177e4 1029{
2a737871 1030 set_root(nd);
e518ddb7 1031
1da177e4 1032 while(1) {
4ac91378 1033 struct dentry *old = nd->path.dentry;
1da177e4 1034
2a737871
AV		 1035 if (nd->path.dentry == nd->root.dentry &&
1036 nd->path.mnt == nd->root.mnt) {
1da177e4
LT		 1037 break;
1038 }
4ac91378 1039 if (nd->path.dentry != nd->path.mnt->mnt_root) {
3088dd70
AV		 1040 /* rare case of legitimate dget_parent()... */
1041 nd->path.dentry = dget_parent(nd->path.dentry);
1da177e4
LT		 1042 dput(old);
1043 break;
1044 }
3088dd70 1045 if (!follow_up(&nd->path))
1da177e4 1046 break;
1da177e4 1047 }
79ed0226 1048 follow_mount(&nd->path);
31e6b01f 1049 nd->inode = nd->path.dentry->d_inode;
1da177e4
LT		 1050}
1051
baa03890 1052/*
bad61189
MS		 1053 * This looks up the name in dcache, possibly revalidates the old dentry and
1054 * allocates a new one if not found or not valid. In the need_lookup argument
1055 * returns whether i_op->lookup is necessary.
1056 *
1057 * dir->d_inode->i_mutex must be held
baa03890 1058 */
bad61189
MS		 1059static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1060 struct nameidata *nd, bool *need_lookup)
baa03890 1061{
baa03890 1062 struct dentry *dentry;
bad61189 1063 int error;
baa03890 1064
bad61189
MS		 1065 *need_lookup = false;
1066 dentry = d_lookup(dir, name);
1067 if (dentry) {
1068 if (d_need_lookup(dentry)) {
1069 *need_lookup = true;
1070 } else if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1071 error = d_revalidate(dentry, nd);
1072 if (unlikely(error <= 0)) {
1073 if (error < 0) {
1074 dput(dentry);
1075 return ERR_PTR(error);
1076 } else if (!d_invalidate(dentry)) {
1077 dput(dentry);
1078 dentry = NULL;
1079 }
1080 }
1081 }
1082 }
baa03890 1083
bad61189
MS		 1084 if (!dentry) {
1085 dentry = d_alloc(dir, name);
1086 if (unlikely(!dentry))
1087 return ERR_PTR(-ENOMEM);
baa03890 1088
bad61189 1089 *need_lookup = true;
baa03890
NP		 1090 }
1091 return dentry;
1092}
1093
44396f4b 1094/*
bad61189
MS		 1095 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1096 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1097 *
1098 * dir->d_inode->i_mutex must be held
44396f4b 1099 */
bad61189
MS		 1100static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1101 struct nameidata *nd)
44396f4b 1102{
44396f4b
JB		 1103 struct dentry *old;
1104
1105 /* Don't create child dentry for a dead directory. */
bad61189 1106 if (unlikely(IS_DEADDIR(dir))) {
e188dc02 1107 dput(dentry);
44396f4b 1108 return ERR_PTR(-ENOENT);
e188dc02 1109 }
44396f4b 1110
bad61189 1111 old = dir->i_op->lookup(dir, dentry, nd);
44396f4b
JB		 1112 if (unlikely(old)) {
1113 dput(dentry);
1114 dentry = old;
1115 }
1116 return dentry;
1117}
1118
a3255546
AV		 1119static struct dentry *__lookup_hash(struct qstr *name,
1120 struct dentry *base, struct nameidata *nd)
1121{
bad61189 1122 bool need_lookup;
a3255546
AV		 1123 struct dentry *dentry;
1124
bad61189
MS		 1125 dentry = lookup_dcache(name, base, nd, &need_lookup);
1126 if (!need_lookup)
1127 return dentry;
a3255546 1128
bad61189 1129 return lookup_real(base->d_inode, dentry, nd);
a3255546
AV		 1130}
1131
1da177e4
LT		 1132/*
1133 * It's more convoluted than I'd like it to be, but... it's still fairly
1134 * small and for now I'd prefer to have fast path as straight as possible.
1135 * It _is_ time-critical.
1136 */
1137static int do_lookup(struct nameidata *nd, struct qstr *name,
31e6b01f 1138 struct path *path, struct inode **inode)
1da177e4 1139{
4ac91378 1140 struct vfsmount *mnt = nd->path.mnt;
31e6b01f 1141 struct dentry *dentry, *parent = nd->path.dentry;
5a18fff2
AV		 1142 int need_reval = 1;
1143 int status = 1;
9875cf80
DH		 1144 int err;
1145
b04f784e
NP		 1146 /*
1147 * Rename seqlock is not required here because in the off chance
1148 * of a false negative due to a concurrent rename, we're going to
1149 * do the non-racy lookup, below.
1150 */
31e6b01f
NP		 1151 if (nd->flags & LOOKUP_RCU) {
1152 unsigned seq;
31e6b01f
NP		 1153 *inode = nd->inode;
1154 dentry = __d_lookup_rcu(parent, name, &seq, inode);
5a18fff2
AV		 1155 if (!dentry)
1156 goto unlazy;
1157
31e6b01f
NP		 1158 /* Memory barrier in read_seqcount_begin of child is enough */
1159 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1160 return -ECHILD;
31e6b01f 1161 nd->seq = seq;
5a18fff2 1162
fa4ee159
MS		 1163 if (unlikely(d_need_lookup(dentry)))
1164 goto unlazy;
24643087 1165 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
5a18fff2
AV		 1166 status = d_revalidate(dentry, nd);
1167 if (unlikely(status <= 0)) {
1168 if (status != -ECHILD)
1169 need_reval = 0;
1170 goto unlazy;
1171 }
24643087 1172 }
31e6b01f
NP		 1173 path->mnt = mnt;
1174 path->dentry = dentry;
d6e9bd25
AV		 1175 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1176 goto unlazy;
1177 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1178 goto unlazy;
1179 return 0;
5a18fff2 1180unlazy:
19660af7
AV		 1181 if (unlazy_walk(nd, dentry))
1182 return -ECHILD;
5a18fff2
AV		 1183 } else {
1184 dentry = __d_lookup(parent, name);
9875cf80 1185 }
5a18fff2 1186
81e6f520
AV		 1187 if (unlikely(!dentry))
1188 goto need_lookup;
1189
1190 if (unlikely(d_need_lookup(dentry))) {
44396f4b 1191 dput(dentry);
81e6f520 1192 goto need_lookup;
5a18fff2 1193 }
81e6f520 1194
5a18fff2
AV		 1195 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1196 status = d_revalidate(dentry, nd);
1197 if (unlikely(status <= 0)) {
1198 if (status < 0) {
1199 dput(dentry);
1200 return status;
1201 }
1202 if (!d_invalidate(dentry)) {
1203 dput(dentry);
81e6f520 1204 goto need_lookup;
5a18fff2 1205 }
24643087 1206 }
3f6c7c71 1207done:
9875cf80
DH		 1208 path->mnt = mnt;
1209 path->dentry = dentry;
1210 err = follow_managed(path, nd->flags);
89312214
IK		 1211 if (unlikely(err < 0)) {
1212 path_put_conditional(path, nd);
9875cf80 1213 return err;
89312214 1214 }
a3fbbde7
AV		 1215 if (err)
1216 nd->flags |= LOOKUP_JUMPED;
9875cf80 1217 *inode = path->dentry->d_inode;
1da177e4 1218 return 0;
81e6f520
AV		 1219
1220need_lookup:
1221 BUG_ON(nd->inode != parent->d_inode);
1222
1223 mutex_lock(&parent->d_inode->i_mutex);
1224 dentry = __lookup_hash(name, parent, nd);
1225 mutex_unlock(&parent->d_inode->i_mutex);
1226 if (IS_ERR(dentry))
1227 return PTR_ERR(dentry);
1228 goto done;
1da177e4
LT		 1229}
1230
52094c8a
AV		 1231static inline int may_lookup(struct nameidata *nd)
1232{
1233 if (nd->flags & LOOKUP_RCU) {
4ad5abb3 1234 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
52094c8a
AV		 1235 if (err != -ECHILD)
1236 return err;
19660af7 1237 if (unlazy_walk(nd, NULL))
52094c8a
AV		 1238 return -ECHILD;
1239 }
4ad5abb3 1240 return inode_permission(nd->inode, MAY_EXEC);
52094c8a
AV		 1241}
1242
9856fa1b
AV		 1243static inline int handle_dots(struct nameidata *nd, int type)
1244{
1245 if (type == LAST_DOTDOT) {
1246 if (nd->flags & LOOKUP_RCU) {
1247 if (follow_dotdot_rcu(nd))
1248 return -ECHILD;
1249 } else
1250 follow_dotdot(nd);
1251 }
1252 return 0;
1253}
1254
951361f9
AV		 1255static void terminate_walk(struct nameidata *nd)
1256{
1257 if (!(nd->flags & LOOKUP_RCU)) {
1258 path_put(&nd->path);
1259 } else {
1260 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 1261 if (!(nd->flags & LOOKUP_ROOT))
1262 nd->root.mnt = NULL;
951361f9
AV		 1263 rcu_read_unlock();
1264 br_read_unlock(vfsmount_lock);
1265 }
1266}
1267
3ddcd056
LT		 1268/*
1269 * Do we need to follow links? We _really_ want to be able
1270 * to do this check without having to look at inode->i_op,
1271 * so we keep a cache of "no, this doesn't need follow_link"
1272 * for the common case.
1273 */
7813b94a 1274static inline int should_follow_link(struct inode *inode, int follow)
3ddcd056
LT		 1275{
1276 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1277 if (likely(inode->i_op->follow_link))
1278 return follow;
1279
1280 /* This gets set once for the inode lifetime */
1281 spin_lock(&inode->i_lock);
1282 inode->i_opflags |= IOP_NOFOLLOW;
1283 spin_unlock(&inode->i_lock);
1284 }
1285 return 0;
1286}
1287
ce57dfc1
AV		 1288static inline int walk_component(struct nameidata *nd, struct path *path,
1289 struct qstr *name, int type, int follow)
1290{
1291 struct inode *inode;
1292 int err;
1293 /*
1294 * "." and ".." are special - ".." especially so because it has
1295 * to be able to know about the current root directory and
1296 * parent relationships.
1297 */
1298 if (unlikely(type != LAST_NORM))
1299 return handle_dots(nd, type);
1300 err = do_lookup(nd, name, path, &inode);
1301 if (unlikely(err)) {
1302 terminate_walk(nd);
1303 return err;
1304 }
1305 if (!inode) {
1306 path_to_nameidata(path, nd);
1307 terminate_walk(nd);
1308 return -ENOENT;
1309 }
7813b94a 1310 if (should_follow_link(inode, follow)) {
19660af7
AV		 1311 if (nd->flags & LOOKUP_RCU) {
1312 if (unlikely(unlazy_walk(nd, path->dentry))) {
1313 terminate_walk(nd);
1314 return -ECHILD;
1315 }
1316 }
ce57dfc1
AV		 1317 BUG_ON(inode != path->dentry->d_inode);
1318 return 1;
1319 }
1320 path_to_nameidata(path, nd);
1321 nd->inode = inode;
1322 return 0;
1323}
1324
b356379a
AV		 1325/*
1326 * This limits recursive symlink follows to 8, while
1327 * limiting consecutive symlinks to 40.
1328 *
1329 * Without that kind of total limit, nasty chains of consecutive
1330 * symlinks can cause almost arbitrarily long lookups.
1331 */
1332static inline int nested_symlink(struct path *path, struct nameidata *nd)
1333{
1334 int res;
1335
b356379a
AV		 1336 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1337 path_put_conditional(path, nd);
1338 path_put(&nd->path);
1339 return -ELOOP;
1340 }
1a4022f8 1341 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
b356379a
AV		 1342
1343 nd->depth++;
1344 current->link_count++;
1345
1346 do {
1347 struct path link = *path;
1348 void *cookie;
574197e0
AV		 1349
1350 res = follow_link(&link, nd, &cookie);
b356379a
AV		 1351 if (!res)
1352 res = walk_component(nd, path, &nd->last,
1353 nd->last_type, LOOKUP_FOLLOW);
574197e0 1354 put_link(nd, &link, cookie);
b356379a
AV		 1355 } while (res > 0);
1356
1357 current->link_count--;
1358 nd->depth--;
1359 return res;
1360}
1361
3ddcd056
LT		 1362/*
1363 * We really don't want to look at inode->i_op->lookup
1364 * when we don't have to. So we keep a cache bit in
1365 * the inode ->i_opflags field that says "yes, we can
1366 * do lookup on this inode".
1367 */
1368static inline int can_lookup(struct inode *inode)
1369{
1370 if (likely(inode->i_opflags & IOP_LOOKUP))
1371 return 1;
1372 if (likely(!inode->i_op->lookup))
1373 return 0;
1374
1375 /* We do this once for the lifetime of the inode */
1376 spin_lock(&inode->i_lock);
1377 inode->i_opflags |= IOP_LOOKUP;
1378 spin_unlock(&inode->i_lock);
1379 return 1;
1380}
1381
bfcfaa77
LT		 1382/*
1383 * We can do the critical dentry name comparison and hashing
1384 * operations one word at a time, but we are limited to:
1385 *
1386 * - Architectures with fast unaligned word accesses. We could
1387 * do a "get_unaligned()" if this helps and is sufficiently
1388 * fast.
1389 *
1390 * - Little-endian machines (so that we can generate the mask
1391 * of low bytes efficiently). Again, we *could* do a byte
1392 * swapping load on big-endian architectures if that is not
1393 * expensive enough to make the optimization worthless.
1394 *
1395 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1396 * do not trap on the (extremely unlikely) case of a page
1397 * crossing operation.
1398 *
1399 * - Furthermore, we need an efficient 64-bit compile for the
1400 * 64-bit case in order to generate the "number of bytes in
1401 * the final mask". Again, that could be replaced with a
1402 * efficient population count instruction or similar.
1403 */
1404#ifdef CONFIG_DCACHE_WORD_ACCESS
1405
1406#ifdef CONFIG_64BIT
1407
1408/*
1409 * Jan Achrenius on G+: microoptimized version of
1410 * the simpler "(mask & ONEBYTES) * ONEBYTES >> 56"
1411 * that works for the bytemasks without having to
1412 * mask them first.
1413 */
1414static inline long count_masked_bytes(unsigned long mask)
1415{
f7493e5d 1416 return mask*0x0001020304050608ul >> 56;
bfcfaa77
LT		 1417}
1418
1419static inline unsigned int fold_hash(unsigned long hash)
1420{
1421 hash += hash >> (8*sizeof(int));
1422 return hash;
1423}
1424
1425#else /* 32-bit case */
1426
1427/* Carl Chatfield / Jan Achrenius G+ version for 32-bit */
1428static inline long count_masked_bytes(long mask)
1429{
1430 /* (000000 0000ff 00ffff ffffff) -> ( 1 1 2 3 ) */
1431 long a = (0x0ff0001+mask) >> 23;
1432 /* Fix the 1 for 00 case */
1433 return a & mask;
1434}
1435
1436#define fold_hash(x) (x)
1437
1438#endif
1439
1440unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1441{
1442 unsigned long a, mask;
1443 unsigned long hash = 0;
1444
1445 for (;;) {
1446 a = *(unsigned long *)name;
bfcfaa77
LT		 1447 if (len < sizeof(unsigned long))
1448 break;
1449 hash += a;
f132c5be 1450 hash *= 9;
bfcfaa77
LT		 1451 name += sizeof(unsigned long);
1452 len -= sizeof(unsigned long);
1453 if (!len)
1454 goto done;
1455 }
1456 mask = ~(~0ul << len*8);
1457 hash += mask & a;
1458done:
1459 return fold_hash(hash);
1460}
1461EXPORT_SYMBOL(full_name_hash);
1462
989412bb
LT		 1463#define REPEAT_BYTE(x) ((~0ul / 0xff) * (x))
1464#define ONEBYTES REPEAT_BYTE(0x01)
1465#define SLASHBYTES REPEAT_BYTE('/')
1466#define HIGHBITS REPEAT_BYTE(0x80)
bfcfaa77
LT		 1467
1468/* Return the high bit set in the first byte that is a zero */
1469static inline unsigned long has_zero(unsigned long a)
1470{
1471 return ((a - ONEBYTES) & ~a) & HIGHBITS;
1472}
1473
1474/*
1475 * Calculate the length and hash of the path component, and
1476 * return the length of the component;
1477 */
1478static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1479{
1480 unsigned long a, mask, hash, len;
1481
1482 hash = a = 0;
1483 len = -sizeof(unsigned long);
1484 do {
1485 hash = (hash + a) * 9;
1486 len += sizeof(unsigned long);
1487 a = *(unsigned long *)(name+len);
1488 /* Do we have any NUL or '/' bytes in this word? */
1489 mask = has_zero(a) | has_zero(a ^ SLASHBYTES);
1490 } while (!mask);
1491
1492 /* The mask *below* the first high bit set */
1493 mask = (mask - 1) & ~mask;
1494 mask >>= 7;
1495 hash += a & mask;
1496 *hashp = fold_hash(hash);
1497
1498 return len + count_masked_bytes(mask);
1499}
1500
1501#else
1502
0145acc2
LT		 1503unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1504{
1505 unsigned long hash = init_name_hash();
1506 while (len--)
1507 hash = partial_name_hash(*name++, hash);
1508 return end_name_hash(hash);
1509}
ae942ae7 1510EXPORT_SYMBOL(full_name_hash);
0145acc2 1511
200e9ef7
LT		 1512/*
1513 * We know there's a real path component here of at least
1514 * one character.
1515 */
1516static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1517{
1518 unsigned long hash = init_name_hash();
1519 unsigned long len = 0, c;
1520
1521 c = (unsigned char)*name;
1522 do {
1523 len++;
1524 hash = partial_name_hash(c, hash);
1525 c = (unsigned char)name[len];
1526 } while (c && c != '/');
1527 *hashp = end_name_hash(hash);
1528 return len;
1529}
1530
bfcfaa77
LT		 1531#endif
1532
1da177e4
LT		 1533/*
1534 * Name resolution.
ea3834d9
PM		 1535 * This is the basic name resolution function, turning a pathname into
1536 * the final dentry. We expect 'base' to be positive and a directory.
1da177e4 1537 *
ea3834d9
PM		 1538 * Returns 0 and nd will have valid dentry and mnt on success.
1539 * Returns error and drops reference to input namei data on failure.
1da177e4 1540 */
6de88d72 1541static int link_path_walk(const char *name, struct nameidata *nd)
1da177e4
LT		 1542{
1543 struct path next;
1da177e4 1544 int err;
1da177e4
LT		 1545
1546 while (*name=='/')
1547 name++;
1548 if (!*name)
086e183a 1549 return 0;
1da177e4 1550
1da177e4
LT		 1551 /* At this point we know we have a real path component. */
1552 for(;;) {
1da177e4 1553 struct qstr this;
200e9ef7 1554 long len;
fe479a58 1555 int type;
1da177e4 1556
52094c8a 1557 err = may_lookup(nd);
1da177e4
LT		 1558 if (err)
1559 break;
1560
200e9ef7 1561 len = hash_name(name, &this.hash);
1da177e4 1562 this.name = name;
200e9ef7 1563 this.len = len;
1da177e4 1564
fe479a58 1565 type = LAST_NORM;
200e9ef7 1566 if (name[0] == '.') switch (len) {
fe479a58 1567 case 2:
200e9ef7 1568 if (name[1] == '.') {
fe479a58 1569 type = LAST_DOTDOT;
16c2cd71
AV		 1570 nd->flags |= LOOKUP_JUMPED;
1571 }
fe479a58
AV		 1572 break;
1573 case 1:
1574 type = LAST_DOT;
1575 }
5a202bcd
AV		 1576 if (likely(type == LAST_NORM)) {
1577 struct dentry *parent = nd->path.dentry;
16c2cd71 1578 nd->flags &= ~LOOKUP_JUMPED;
5a202bcd
AV		 1579 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1580 err = parent->d_op->d_hash(parent, nd->inode,
1581 &this);
1582 if (err < 0)
1583 break;
1584 }
1585 }
fe479a58 1586
200e9ef7 1587 if (!name[len])
1da177e4 1588 goto last_component;
200e9ef7
LT		 1589 /*
1590 * If it wasn't NUL, we know it was '/'. Skip that
1591 * slash, and continue until no more slashes.
1592 */
1593 do {
1594 len++;
1595 } while (unlikely(name[len] == '/'));
1596 if (!name[len])
b356379a 1597 goto last_component;
200e9ef7 1598 name += len;
1da177e4 1599
ce57dfc1
AV		 1600 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1601 if (err < 0)
1602 return err;
1da177e4 1603
ce57dfc1 1604 if (err) {
b356379a 1605 err = nested_symlink(&next, nd);
1da177e4 1606 if (err)
a7472bab 1607 return err;
31e6b01f 1608 }
3ddcd056
LT		 1609 if (can_lookup(nd->inode))
1610 continue;
1da177e4 1611 err = -ENOTDIR;
3ddcd056 1612 break;
1da177e4
LT		 1613 /* here ends the main loop */
1614
1da177e4 1615last_component:
b356379a
AV		 1616 nd->last = this;
1617 nd->last_type = type;
086e183a 1618 return 0;
1da177e4 1619 }
951361f9 1620 terminate_walk(nd);
1da177e4
LT		 1621 return err;
1622}
1623
70e9b357
AV		 1624static int path_init(int dfd, const char *name, unsigned int flags,
1625 struct nameidata *nd, struct file **fp)
31e6b01f
NP		 1626{
1627 int retval = 0;
1628 int fput_needed;
1629 struct file *file;
1630
1631 nd->last_type = LAST_ROOT; /* if there are only slashes... */
16c2cd71 1632 nd->flags = flags | LOOKUP_JUMPED;
31e6b01f 1633 nd->depth = 0;
5b6ca027
AV		 1634 if (flags & LOOKUP_ROOT) {
1635 struct inode *inode = nd->root.dentry->d_inode;
73d049a4
AV		 1636 if (*name) {
1637 if (!inode->i_op->lookup)
1638 return -ENOTDIR;
1639 retval = inode_permission(inode, MAY_EXEC);
1640 if (retval)
1641 return retval;
1642 }
5b6ca027
AV		 1643 nd->path = nd->root;
1644 nd->inode = inode;
1645 if (flags & LOOKUP_RCU) {
1646 br_read_lock(vfsmount_lock);
1647 rcu_read_lock();
1648 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1649 } else {
1650 path_get(&nd->path);
1651 }
1652 return 0;
1653 }
1654
31e6b01f 1655 nd->root.mnt = NULL;
31e6b01f
NP		 1656
1657 if (*name=='/') {
e41f7d4e
AV		 1658 if (flags & LOOKUP_RCU) {
1659 br_read_lock(vfsmount_lock);
1660 rcu_read_lock();
1661 set_root_rcu(nd);
1662 } else {
1663 set_root(nd);
1664 path_get(&nd->root);
1665 }
1666 nd->path = nd->root;
31e6b01f 1667 } else if (dfd == AT_FDCWD) {
e41f7d4e
AV		 1668 if (flags & LOOKUP_RCU) {
1669 struct fs_struct *fs = current->fs;
1670 unsigned seq;
31e6b01f 1671
e41f7d4e
AV		 1672 br_read_lock(vfsmount_lock);
1673 rcu_read_lock();
c28cc364 1674
e41f7d4e
AV		 1675 do {
1676 seq = read_seqcount_begin(&fs->seq);
1677 nd->path = fs->pwd;
1678 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1679 } while (read_seqcount_retry(&fs->seq, seq));
1680 } else {
1681 get_fs_pwd(current->fs, &nd->path);
1682 }
31e6b01f
NP		 1683 } else {
1684 struct dentry *dentry;
1685
1abf0c71 1686 file = fget_raw_light(dfd, &fput_needed);
31e6b01f
NP		 1687 retval = -EBADF;
1688 if (!file)
1689 goto out_fail;
1690
1691 dentry = file->f_path.dentry;
1692
f52e0c11
AV		 1693 if (*name) {
1694 retval = -ENOTDIR;
1695 if (!S_ISDIR(dentry->d_inode->i_mode))
1696 goto fput_fail;
31e6b01f 1697
4ad5abb3 1698 retval = inode_permission(dentry->d_inode, MAY_EXEC);
f52e0c11
AV		 1699 if (retval)
1700 goto fput_fail;
1701 }
31e6b01f
NP		 1702
1703 nd->path = file->f_path;
e41f7d4e
AV		 1704 if (flags & LOOKUP_RCU) {
1705 if (fput_needed)
70e9b357 1706 *fp = file;
e41f7d4e
AV		 1707 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1708 br_read_lock(vfsmount_lock);
1709 rcu_read_lock();
1710 } else {
1711 path_get(&file->f_path);
1712 fput_light(file, fput_needed);
1713 }
31e6b01f 1714 }
31e6b01f 1715
31e6b01f 1716 nd->inode = nd->path.dentry->d_inode;
9b4a9b14 1717 return 0;
2dfdd266 1718
9b4a9b14
AV		 1719fput_fail:
1720 fput_light(file, fput_needed);
1721out_fail:
1722 return retval;
1723}
1724
bd92d7fe
AV		 1725static inline int lookup_last(struct nameidata *nd, struct path *path)
1726{
1727 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1728 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1729
1730 nd->flags &= ~LOOKUP_PARENT;
1731 return walk_component(nd, path, &nd->last, nd->last_type,
1732 nd->flags & LOOKUP_FOLLOW);
1733}
1734
9b4a9b14 1735/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
ee0827cd 1736static int path_lookupat(int dfd, const char *name,
9b4a9b14
AV		 1737 unsigned int flags, struct nameidata *nd)
1738{
70e9b357 1739 struct file *base = NULL;
bd92d7fe
AV		 1740 struct path path;
1741 int err;
31e6b01f
NP		 1742
1743 /*
1744 * Path walking is largely split up into 2 different synchronisation
1745 * schemes, rcu-walk and ref-walk (explained in
1746 * Documentation/filesystems/path-lookup.txt). These share much of the
1747 * path walk code, but some things particularly setup, cleanup, and
1748 * following mounts are sufficiently divergent that functions are
1749 * duplicated. Typically there is a function foo(), and its RCU
1750 * analogue, foo_rcu().
1751 *
1752 * -ECHILD is the error number of choice (just to avoid clashes) that
1753 * is returned if some aspect of an rcu-walk fails. Such an error must
1754 * be handled by restarting a traditional ref-walk (which will always
1755 * be able to complete).
1756 */
bd92d7fe 1757 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
ee0827cd 1758
bd92d7fe
AV		 1759 if (unlikely(err))
1760 return err;
ee0827cd
AV		 1761
1762 current->total_link_count = 0;
bd92d7fe
AV		 1763 err = link_path_walk(name, nd);
1764
1765 if (!err && !(flags & LOOKUP_PARENT)) {
bd92d7fe
AV		 1766 err = lookup_last(nd, &path);
1767 while (err > 0) {
1768 void *cookie;
1769 struct path link = path;
bd92d7fe 1770 nd->flags |= LOOKUP_PARENT;
574197e0 1771 err = follow_link(&link, nd, &cookie);
bd92d7fe
AV		 1772 if (!err)
1773 err = lookup_last(nd, &path);
574197e0 1774 put_link(nd, &link, cookie);
bd92d7fe
AV		 1775 }
1776 }
ee0827cd 1777
9f1fafee
AV		 1778 if (!err)
1779 err = complete_walk(nd);
bd92d7fe
AV		 1780
1781 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1782 if (!nd->inode->i_op->lookup) {
1783 path_put(&nd->path);
bd23a539 1784 err = -ENOTDIR;
bd92d7fe
AV		 1785 }
1786 }
16c2cd71 1787
70e9b357
AV		 1788 if (base)
1789 fput(base);
ee0827cd 1790
5b6ca027 1791 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
2a737871
AV		 1792 path_put(&nd->root);
1793 nd->root.mnt = NULL;
1794 }
bd92d7fe 1795 return err;
ee0827cd 1796}
31e6b01f 1797
ee0827cd
AV		 1798static int do_path_lookup(int dfd, const char *name,
1799 unsigned int flags, struct nameidata *nd)
1800{
1801 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1802 if (unlikely(retval == -ECHILD))
1803 retval = path_lookupat(dfd, name, flags, nd);
1804 if (unlikely(retval == -ESTALE))
1805 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
31e6b01f
NP		 1806
1807 if (likely(!retval)) {
1808 if (unlikely(!audit_dummy_context())) {
1809 if (nd->path.dentry && nd->inode)
1810 audit_inode(name, nd->path.dentry);
1811 }
1812 }
170aa3d0 1813 return retval;
1da177e4
LT		 1814}
1815
c9c6cac0 1816int kern_path_parent(const char *name, struct nameidata *nd)
5590ff0d 1817{
c9c6cac0 1818 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
5590ff0d
UD		 1819}
1820
d1811465
AV		 1821int kern_path(const char *name, unsigned int flags, struct path *path)
1822{
1823 struct nameidata nd;
1824 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1825 if (!res)
1826 *path = nd.path;
1827 return res;
1828}
1829
16f18200
JJS		 1830/**
1831 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1832 * @dentry: pointer to dentry of the base directory
1833 * @mnt: pointer to vfs mount of the base directory
1834 * @name: pointer to file name
1835 * @flags: lookup flags
e0a01249 1836 * @path: pointer to struct path to fill
16f18200
JJS		 1837 */
1838int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1839 const char *name, unsigned int flags,
e0a01249 1840 struct path *path)
16f18200 1841{
e0a01249
AV		 1842 struct nameidata nd;
1843 int err;
1844 nd.root.dentry = dentry;
1845 nd.root.mnt = mnt;
1846 BUG_ON(flags & LOOKUP_PARENT);
5b6ca027 1847 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
e0a01249
AV		 1848 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1849 if (!err)
1850 *path = nd.path;
1851 return err;
16f18200
JJS		 1852}
1853
057f6c01
JM		 1854/*
1855 * Restricted form of lookup. Doesn't follow links, single-component only,
1856 * needs parent already locked. Doesn't follow mounts.
1857 * SMP-safe.
1858 */
eead1911 1859static struct dentry *lookup_hash(struct nameidata *nd)
057f6c01 1860{
4ac91378 1861 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1da177e4
LT		 1862}
1863
eead1911 1864/**
a6b91919 1865 * lookup_one_len - filesystem helper to lookup single pathname component
eead1911
CH		 1866 * @name: pathname component to lookup
1867 * @base: base directory to lookup from
1868 * @len: maximum length @len should be interpreted to
1869 *
a6b91919
RD		 1870 * Note that this routine is purely a helper for filesystem usage and should
1871 * not be called by generic code. Also note that by using this function the
eead1911
CH		 1872 * nameidata argument is passed to the filesystem methods and a filesystem
1873 * using this helper needs to be prepared for that.
1874 */
057f6c01
JM		 1875struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1876{
057f6c01 1877 struct qstr this;
6a96ba54 1878 unsigned int c;
cda309de 1879 int err;
057f6c01 1880
2f9092e1
DW		 1881 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1882
6a96ba54
AV		 1883 this.name = name;
1884 this.len = len;
0145acc2 1885 this.hash = full_name_hash(name, len);
6a96ba54
AV		 1886 if (!len)
1887 return ERR_PTR(-EACCES);
1888
6a96ba54
AV		 1889 while (len--) {
1890 c = *(const unsigned char *)name++;
1891 if (c == '/' || c == '\0')
1892 return ERR_PTR(-EACCES);
6a96ba54 1893 }
5a202bcd
AV		 1894 /*
1895 * See if the low-level filesystem might want
1896 * to use its own hash..
1897 */
1898 if (base->d_flags & DCACHE_OP_HASH) {
1899 int err = base->d_op->d_hash(base, base->d_inode, &this);
1900 if (err < 0)
1901 return ERR_PTR(err);
1902 }
eead1911 1903
cda309de
MS		 1904 err = inode_permission(base->d_inode, MAY_EXEC);
1905 if (err)
1906 return ERR_PTR(err);
1907
49705b77 1908 return __lookup_hash(&this, base, NULL);
057f6c01
JM		 1909}
1910
1fa1e7f6
AW		 1911int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
1912 struct path *path, int *empty)
1da177e4 1913{
2d8f3038 1914 struct nameidata nd;
1fa1e7f6 1915 char *tmp = getname_flags(name, flags, empty);
1da177e4 1916 int err = PTR_ERR(tmp);
1da177e4 1917 if (!IS_ERR(tmp)) {
2d8f3038
AV		 1918
1919 BUG_ON(flags & LOOKUP_PARENT);
1920
1921 err = do_path_lookup(dfd, tmp, flags, &nd);
1da177e4 1922 putname(tmp);
2d8f3038
AV		 1923 if (!err)
1924 *path = nd.path;
1da177e4
LT		 1925 }
1926 return err;
1927}
1928
1fa1e7f6
AW		 1929int user_path_at(int dfd, const char __user *name, unsigned flags,
1930 struct path *path)
1931{
f7493e5d 1932 return user_path_at_empty(dfd, name, flags, path, NULL);
1fa1e7f6
AW		 1933}
1934
2ad94ae6
AV		 1935static int user_path_parent(int dfd, const char __user *path,
1936 struct nameidata *nd, char **name)
1937{
1938 char *s = getname(path);
1939 int error;
1940
1941 if (IS_ERR(s))
1942 return PTR_ERR(s);
1943
1944 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1945 if (error)
1946 putname(s);
1947 else
1948 *name = s;
1949
1950 return error;
1951}
1952
1da177e4
LT		 1953/*
1954 * It's inline, so penalty for filesystems that don't use sticky bit is
1955 * minimal.
1956 */
1957static inline int check_sticky(struct inode *dir, struct inode *inode)
1958{
da9592ed
DH		 1959 uid_t fsuid = current_fsuid();
1960
1da177e4
LT		 1961 if (!(dir->i_mode & S_ISVTX))
1962 return 0;
da9592ed 1963 if (inode->i_uid == fsuid)
1da177e4 1964 return 0;
da9592ed 1965 if (dir->i_uid == fsuid)
1da177e4 1966 return 0;
1a48e2ac 1967 return !inode_capable(inode, CAP_FOWNER);
1da177e4
LT		 1968}
1969
1970/*
1971 * Check whether we can remove a link victim from directory dir, check
1972 * whether the type of victim is right.
1973 * 1. We can't do it if dir is read-only (done in permission())
1974 * 2. We should have write and exec permissions on dir
1975 * 3. We can't remove anything from append-only dir
1976 * 4. We can't do anything with immutable dir (done in permission())
1977 * 5. If the sticky bit on dir is set we should either
1978 * a. be owner of dir, or
1979 * b. be owner of victim, or
1980 * c. have CAP_FOWNER capability
1981 * 6. If the victim is append-only or immutable we can't do antyhing with
1982 * links pointing to it.
1983 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1984 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1985 * 9. We can't remove a root or mountpoint.
1986 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1987 * nfs_async_unlink().
1988 */
858119e1 1989static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1da177e4
LT		 1990{
1991 int error;
1992
1993 if (!victim->d_inode)
1994 return -ENOENT;
1995
1996 BUG_ON(victim->d_parent->d_inode != dir);
cccc6bba 1997 audit_inode_child(victim, dir);
1da177e4 1998
f419a2e3 1999 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 2000 if (error)
2001 return error;
2002 if (IS_APPEND(dir))
2003 return -EPERM;
2004 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
f9454548 2005 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1da177e4
LT		 2006 return -EPERM;
2007 if (isdir) {
2008 if (!S_ISDIR(victim->d_inode->i_mode))
2009 return -ENOTDIR;
2010 if (IS_ROOT(victim))
2011 return -EBUSY;
2012 } else if (S_ISDIR(victim->d_inode->i_mode))
2013 return -EISDIR;
2014 if (IS_DEADDIR(dir))
2015 return -ENOENT;
2016 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2017 return -EBUSY;
2018 return 0;
2019}
2020
2021/* Check whether we can create an object with dentry child in directory
2022 * dir.
2023 * 1. We can't do it if child already exists (open has special treatment for
2024 * this case, but since we are inlined it's OK)
2025 * 2. We can't do it if dir is read-only (done in permission())
2026 * 3. We should have write and exec permissions on dir
2027 * 4. We can't do it if dir is immutable (done in permission())
2028 */
a95164d9 2029static inline int may_create(struct inode *dir, struct dentry *child)
1da177e4
LT		 2030{
2031 if (child->d_inode)
2032 return -EEXIST;
2033 if (IS_DEADDIR(dir))
2034 return -ENOENT;
f419a2e3 2035 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 2036}
2037
1da177e4
LT		 2038/*
2039 * p1 and p2 should be directories on the same fs.
2040 */
2041struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2042{
2043 struct dentry *p;
2044
2045 if (p1 == p2) {
f2eace23 2046 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1da177e4
LT		 2047 return NULL;
2048 }
2049
a11f3a05 2050 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4 2051
e2761a11
OH		 2052 p = d_ancestor(p2, p1);
2053 if (p) {
2054 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2055 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2056 return p;
1da177e4
LT		 2057 }
2058
e2761a11
OH		 2059 p = d_ancestor(p1, p2);
2060 if (p) {
2061 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2062 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2063 return p;
1da177e4
LT		 2064 }
2065
f2eace23
IM		 2066 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2067 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1da177e4
LT		 2068 return NULL;
2069}
2070
2071void unlock_rename(struct dentry *p1, struct dentry *p2)
2072{
1b1dcc1b 2073 mutex_unlock(&p1->d_inode->i_mutex);
1da177e4 2074 if (p1 != p2) {
1b1dcc1b 2075 mutex_unlock(&p2->d_inode->i_mutex);
a11f3a05 2076 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4
LT		 2077 }
2078}
2079
4acdaf27 2080int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
1da177e4
LT		 2081 struct nameidata *nd)
2082{
a95164d9 2083 int error = may_create(dir, dentry);
1da177e4
LT		 2084
2085 if (error)
2086 return error;
2087
acfa4380 2088 if (!dir->i_op->create)
1da177e4
LT		 2089 return -EACCES; /* shouldn't it be ENOSYS? */
2090 mode &= S_IALLUGO;
2091 mode |= S_IFREG;
2092 error = security_inode_create(dir, dentry, mode);
2093 if (error)
2094 return error;
1da177e4 2095 error = dir->i_op->create(dir, dentry, mode, nd);
a74574aa 2096 if (!error)
f38aa942 2097 fsnotify_create(dir, dentry);
1da177e4
LT		 2098 return error;
2099}
2100
73d049a4 2101static int may_open(struct path *path, int acc_mode, int flag)
1da177e4 2102{
3fb64190 2103 struct dentry *dentry = path->dentry;
1da177e4
LT		 2104 struct inode *inode = dentry->d_inode;
2105 int error;
2106
bcda7652
AV		 2107 /* O_PATH? */
2108 if (!acc_mode)
2109 return 0;
2110
1da177e4
LT		 2111 if (!inode)
2112 return -ENOENT;
2113
c8fe8f30
CH		 2114 switch (inode->i_mode & S_IFMT) {
2115 case S_IFLNK:
1da177e4 2116 return -ELOOP;
c8fe8f30
CH		 2117 case S_IFDIR:
2118 if (acc_mode & MAY_WRITE)
2119 return -EISDIR;
2120 break;
2121 case S_IFBLK:
2122 case S_IFCHR:
3fb64190 2123 if (path->mnt->mnt_flags & MNT_NODEV)
1da177e4 2124 return -EACCES;
c8fe8f30
CH		 2125 /*FALLTHRU*/
2126 case S_IFIFO:
2127 case S_IFSOCK:
1da177e4 2128 flag &= ~O_TRUNC;
c8fe8f30 2129 break;
4a3fd211 2130 }
b41572e9 2131
3fb64190 2132 error = inode_permission(inode, acc_mode);
b41572e9
DH		 2133 if (error)
2134 return error;
6146f0d5 2135
1da177e4
LT		 2136 /*
2137 * An append-only file must be opened in append mode for writing.
2138 */
2139 if (IS_APPEND(inode)) {
8737c930 2140 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
7715b521 2141 return -EPERM;
1da177e4 2142 if (flag & O_TRUNC)
7715b521 2143 return -EPERM;
1da177e4
LT		 2144 }
2145
2146 /* O_NOATIME can only be set by the owner or superuser */
2e149670 2147 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
7715b521 2148 return -EPERM;
1da177e4 2149
f3c7691e 2150 return 0;
7715b521 2151}
1da177e4 2152
e1181ee6 2153static int handle_truncate(struct file *filp)
7715b521 2154{
e1181ee6 2155 struct path *path = &filp->f_path;
7715b521
AV		 2156 struct inode *inode = path->dentry->d_inode;
2157 int error = get_write_access(inode);
2158 if (error)
2159 return error;
2160 /*
2161 * Refuse to truncate files with mandatory locks held on them.
2162 */
2163 error = locks_verify_locked(inode);
2164 if (!error)
ea0d3ab2 2165 error = security_path_truncate(path);
7715b521
AV		 2166 if (!error) {
2167 error = do_truncate(path->dentry, 0,
2168 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
e1181ee6 2169 filp);
7715b521
AV		 2170 }
2171 put_write_access(inode);
acd0c935 2172 return error;
1da177e4
LT		 2173}
2174
d57999e1
DH		 2175static inline int open_to_namei_flags(int flag)
2176{
8a5e929d
AV		 2177 if ((flag & O_ACCMODE) == 3)
2178 flag--;
d57999e1
DH		 2179 return flag;
2180}
2181
31e6b01f 2182/*
fe2d35ff 2183 * Handle the last step of open()
31e6b01f 2184 */
fb1cc555 2185static struct file *do_last(struct nameidata *nd, struct path *path,
c3e380b0 2186 const struct open_flags *op, const char *pathname)
fb1cc555 2187{
a1e28038 2188 struct dentry *dir = nd->path.dentry;
6c0d46c4 2189 struct dentry *dentry;
ca344a89 2190 int open_flag = op->open_flag;
6c0d46c4 2191 int will_truncate = open_flag & O_TRUNC;
ca344a89 2192 int want_write = 0;
bcda7652 2193 int acc_mode = op->acc_mode;
fb1cc555 2194 struct file *filp;
16c2cd71 2195 int error;
1f36f774 2196
c3e380b0
AV		 2197 nd->flags &= ~LOOKUP_PARENT;
2198 nd->flags |= op->intent;
2199
1f36f774
AV		 2200 switch (nd->last_type) {
2201 case LAST_DOTDOT:
176306f5 2202 case LAST_DOT:
fe2d35ff
AV		 2203 error = handle_dots(nd, nd->last_type);
2204 if (error)
2205 return ERR_PTR(error);
1f36f774 2206 /* fallthrough */
1f36f774 2207 case LAST_ROOT:
9f1fafee 2208 error = complete_walk(nd);
16c2cd71 2209 if (error)
9f1fafee 2210 return ERR_PTR(error);
fe2d35ff 2211 audit_inode(pathname, nd->path.dentry);
ca344a89 2212 if (open_flag & O_CREAT) {
fe2d35ff
AV		 2213 error = -EISDIR;
2214 goto exit;
2215 }
2216 goto ok;
1f36f774 2217 case LAST_BIND:
9f1fafee 2218 error = complete_walk(nd);
16c2cd71 2219 if (error)
9f1fafee 2220 return ERR_PTR(error);
1f36f774 2221 audit_inode(pathname, dir);
67ee3ad2 2222 goto ok;
1f36f774 2223 }
67ee3ad2 2224
ca344a89 2225 if (!(open_flag & O_CREAT)) {
bcda7652 2226 int symlink_ok = 0;
fe2d35ff
AV		 2227 if (nd->last.name[nd->last.len])
2228 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
bcda7652
AV		 2229 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2230 symlink_ok = 1;
fe2d35ff 2231 /* we _can_ be in RCU mode here */
ce57dfc1
AV		 2232 error = walk_component(nd, path, &nd->last, LAST_NORM,
2233 !symlink_ok);
2234 if (error < 0)
fe2d35ff 2235 return ERR_PTR(error);
ce57dfc1 2236 if (error) /* symlink */
fe2d35ff 2237 return NULL;
fe2d35ff 2238 /* sayonara */
9f1fafee
AV		 2239 error = complete_walk(nd);
2240 if (error)
7f6c7e62 2241 return ERR_PTR(error);
fe2d35ff
AV		 2242
2243 error = -ENOTDIR;
2244 if (nd->flags & LOOKUP_DIRECTORY) {
ce57dfc1 2245 if (!nd->inode->i_op->lookup)
fe2d35ff
AV		 2246 goto exit;
2247 }
2248 audit_inode(pathname, nd->path.dentry);
2249 goto ok;
2250 }
2251
2252 /* create side of things */
a3fbbde7
AV		 2253 /*
2254 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED has been
2255 * cleared when we got to the last component we are about to look up
2256 */
9f1fafee
AV		 2257 error = complete_walk(nd);
2258 if (error)
2259 return ERR_PTR(error);
fe2d35ff
AV		 2260
2261 audit_inode(pathname, dir);
16c2cd71 2262 error = -EISDIR;
1f36f774 2263 /* trailing slashes? */
31e6b01f
NP		 2264 if (nd->last.name[nd->last.len])
2265 goto exit;
a2c36b45 2266
a1e28038
AV		 2267 mutex_lock(&dir->d_inode->i_mutex);
2268
6c0d46c4
AV		 2269 dentry = lookup_hash(nd);
2270 error = PTR_ERR(dentry);
2271 if (IS_ERR(dentry)) {
fb1cc555
AV		 2272 mutex_unlock(&dir->d_inode->i_mutex);
2273 goto exit;
2274 }
2275
6c0d46c4
AV		 2276 path->dentry = dentry;
2277 path->mnt = nd->path.mnt;
2278
fb1cc555 2279 /* Negative dentry, just create the file */
6c0d46c4 2280 if (!dentry->d_inode) {
a218d0fd 2281 umode_t mode = op->mode;
6c0d46c4
AV		 2282 if (!IS_POSIXACL(dir->d_inode))
2283 mode &= ~current_umask();
fb1cc555
AV		 2284 /*
2285 * This write is needed to ensure that a
6c0d46c4 2286 * rw->ro transition does not occur between
fb1cc555
AV		 2287 * the time when the file is created and when
2288 * a permanent write count is taken through
2289 * the 'struct file' in nameidata_to_filp().
2290 */
2291 error = mnt_want_write(nd->path.mnt);
2292 if (error)
2293 goto exit_mutex_unlock;
ca344a89 2294 want_write = 1;
9b44f1b3 2295 /* Don't check for write permission, don't truncate */
ca344a89 2296 open_flag &= ~O_TRUNC;
6c0d46c4 2297 will_truncate = 0;
bcda7652 2298 acc_mode = MAY_OPEN;
6c0d46c4
AV		 2299 error = security_path_mknod(&nd->path, dentry, mode, 0);
2300 if (error)
2301 goto exit_mutex_unlock;
2302 error = vfs_create(dir->d_inode, dentry, mode, nd);
2303 if (error)
2304 goto exit_mutex_unlock;
2305 mutex_unlock(&dir->d_inode->i_mutex);
2306 dput(nd->path.dentry);
2307 nd->path.dentry = dentry;
ca344a89 2308 goto common;
fb1cc555
AV		 2309 }
2310
2311 /*
2312 * It already exists.
2313 */
2314 mutex_unlock(&dir->d_inode->i_mutex);
2315 audit_inode(pathname, path->dentry);
2316
2317 error = -EEXIST;
ca344a89 2318 if (open_flag & O_EXCL)
fb1cc555
AV		 2319 goto exit_dput;
2320
9875cf80
DH		 2321 error = follow_managed(path, nd->flags);
2322 if (error < 0)
2323 goto exit_dput;
fb1cc555 2324
a3fbbde7
AV		 2325 if (error)
2326 nd->flags |= LOOKUP_JUMPED;
2327
fb1cc555
AV		 2328 error = -ENOENT;
2329 if (!path->dentry->d_inode)
2330 goto exit_dput;
9e67f361
AV		 2331
2332 if (path->dentry->d_inode->i_op->follow_link)
fb1cc555 2333 return NULL;
fb1cc555
AV		 2334
2335 path_to_nameidata(path, nd);
31e6b01f 2336 nd->inode = path->dentry->d_inode;
a3fbbde7
AV		 2337 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2338 error = complete_walk(nd);
2339 if (error)
097b180c 2340 return ERR_PTR(error);
fb1cc555 2341 error = -EISDIR;
31e6b01f 2342 if (S_ISDIR(nd->inode->i_mode))
fb1cc555 2343 goto exit;
67ee3ad2 2344ok:
6c0d46c4
AV		 2345 if (!S_ISREG(nd->inode->i_mode))
2346 will_truncate = 0;
2347
0f9d1a10
AV		 2348 if (will_truncate) {
2349 error = mnt_want_write(nd->path.mnt);
2350 if (error)
2351 goto exit;
ca344a89 2352 want_write = 1;
0f9d1a10 2353 }
ca344a89 2354common:
bcda7652 2355 error = may_open(&nd->path, acc_mode, open_flag);
ca344a89 2356 if (error)
0f9d1a10 2357 goto exit;
0f9d1a10
AV		 2358 filp = nameidata_to_filp(nd);
2359 if (!IS_ERR(filp)) {
2360 error = ima_file_check(filp, op->acc_mode);
2361 if (error) {
2362 fput(filp);
2363 filp = ERR_PTR(error);
2364 }
2365 }
2366 if (!IS_ERR(filp)) {
2367 if (will_truncate) {
2368 error = handle_truncate(filp);
2369 if (error) {
2370 fput(filp);
2371 filp = ERR_PTR(error);
2372 }
2373 }
2374 }
ca344a89
AV		 2375out:
2376 if (want_write)
0f9d1a10
AV		 2377 mnt_drop_write(nd->path.mnt);
2378 path_put(&nd->path);
fb1cc555
AV		 2379 return filp;
2380
2381exit_mutex_unlock:
2382 mutex_unlock(&dir->d_inode->i_mutex);
2383exit_dput:
2384 path_put_conditional(path, nd);
2385exit:
ca344a89
AV		 2386 filp = ERR_PTR(error);
2387 goto out;
fb1cc555
AV		 2388}
2389
13aab428 2390static struct file *path_openat(int dfd, const char *pathname,
73d049a4 2391 struct nameidata *nd, const struct open_flags *op, int flags)
1da177e4 2392{
fe2d35ff 2393 struct file *base = NULL;
4a3fd211 2394 struct file *filp;
9850c056 2395 struct path path;
13aab428 2396 int error;
31e6b01f
NP		 2397
2398 filp = get_empty_filp();
2399 if (!filp)
2400 return ERR_PTR(-ENFILE);
2401
47c805dc 2402 filp->f_flags = op->open_flag;
73d049a4
AV		 2403 nd->intent.open.file = filp;
2404 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2405 nd->intent.open.create_mode = op->mode;
31e6b01f 2406
73d049a4 2407 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
31e6b01f 2408 if (unlikely(error))
13aab428 2409 goto out_filp;
31e6b01f 2410
fe2d35ff 2411 current->total_link_count = 0;
73d049a4 2412 error = link_path_walk(pathname, nd);
31e6b01f
NP		 2413 if (unlikely(error))
2414 goto out_filp;
1da177e4 2415
73d049a4 2416 filp = do_last(nd, &path, op, pathname);
806b681c 2417 while (unlikely(!filp)) { /* trailing symlink */
7b9337aa 2418 struct path link = path;
def4af30 2419 void *cookie;
574197e0 2420 if (!(nd->flags & LOOKUP_FOLLOW)) {
73d049a4
AV		 2421 path_put_conditional(&path, nd);
2422 path_put(&nd->path);
40b39136
AV		 2423 filp = ERR_PTR(-ELOOP);
2424 break;
2425 }
73d049a4
AV		 2426 nd->flags |= LOOKUP_PARENT;
2427 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
574197e0 2428 error = follow_link(&link, nd, &cookie);
c3e380b0 2429 if (unlikely(error))
f1afe9ef 2430 filp = ERR_PTR(error);
c3e380b0 2431 else
73d049a4 2432 filp = do_last(nd, &path, op, pathname);
574197e0 2433 put_link(nd, &link, cookie);
806b681c 2434 }
10fa8e62 2435out:
73d049a4
AV		 2436 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2437 path_put(&nd->root);
fe2d35ff
AV		 2438 if (base)
2439 fput(base);
73d049a4 2440 release_open_intent(nd);
10fa8e62 2441 return filp;
1da177e4 2442
31e6b01f 2443out_filp:
806b681c 2444 filp = ERR_PTR(error);
10fa8e62 2445 goto out;
1da177e4
LT		 2446}
2447
13aab428
AV		 2448struct file *do_filp_open(int dfd, const char *pathname,
2449 const struct open_flags *op, int flags)
2450{
73d049a4 2451 struct nameidata nd;
13aab428
AV		 2452 struct file *filp;
2453
73d049a4 2454 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
13aab428 2455 if (unlikely(filp == ERR_PTR(-ECHILD)))
73d049a4 2456 filp = path_openat(dfd, pathname, &nd, op, flags);
13aab428 2457 if (unlikely(filp == ERR_PTR(-ESTALE)))
73d049a4 2458 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
13aab428
AV		 2459 return filp;
2460}
2461
73d049a4
AV		 2462struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2463 const char *name, const struct open_flags *op, int flags)
2464{
2465 struct nameidata nd;
2466 struct file *file;
2467
2468 nd.root.mnt = mnt;
2469 nd.root.dentry = dentry;
2470
2471 flags |= LOOKUP_ROOT;
2472
bcda7652 2473 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
73d049a4
AV		 2474 return ERR_PTR(-ELOOP);
2475
2476 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2477 if (unlikely(file == ERR_PTR(-ECHILD)))
2478 file = path_openat(-1, name, &nd, op, flags);
2479 if (unlikely(file == ERR_PTR(-ESTALE)))
2480 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2481 return file;
2482}
2483
ed75e95d 2484struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
1da177e4 2485{
c663e5d8 2486 struct dentry *dentry = ERR_PTR(-EEXIST);
ed75e95d
AV		 2487 struct nameidata nd;
2488 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2489 if (error)
2490 return ERR_PTR(error);
1da177e4 2491
c663e5d8
CH		 2492 /*
2493 * Yucky last component or no last component at all?
2494 * (foo/., foo/.., /////)
2495 */
ed75e95d
AV		 2496 if (nd.last_type != LAST_NORM)
2497 goto out;
2498 nd.flags &= ~LOOKUP_PARENT;
2499 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2500 nd.intent.open.flags = O_EXCL;
c663e5d8
CH		 2501
2502 /*
2503 * Do the final lookup.
2504 */
ed75e95d
AV		 2505 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2506 dentry = lookup_hash(&nd);
1da177e4
LT		 2507 if (IS_ERR(dentry))
2508 goto fail;
c663e5d8 2509
e9baf6e5
AV		 2510 if (dentry->d_inode)
2511 goto eexist;
c663e5d8
CH		 2512 /*
2513 * Special case - lookup gave negative, but... we had foo/bar/
2514 * From the vfs_mknod() POV we just have a negative dentry -
2515 * all is fine. Let's be bastards - you had / on the end, you've
2516 * been asking for (non-existent) directory. -ENOENT for you.
2517 */
ed75e95d 2518 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
e9baf6e5
AV		 2519 dput(dentry);
2520 dentry = ERR_PTR(-ENOENT);
ed75e95d 2521 goto fail;
e9baf6e5 2522 }
ed75e95d 2523 *path = nd.path;
1da177e4 2524 return dentry;
e9baf6e5 2525eexist:
1da177e4 2526 dput(dentry);
e9baf6e5 2527 dentry = ERR_PTR(-EEXIST);
1da177e4 2528fail:
ed75e95d
AV		 2529 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2530out:
2531 path_put(&nd.path);
1da177e4
LT		 2532 return dentry;
2533}
dae6ad8f
AV		 2534EXPORT_SYMBOL(kern_path_create);
2535
2536struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2537{
2538 char *tmp = getname(pathname);
2539 struct dentry *res;
2540 if (IS_ERR(tmp))
2541 return ERR_CAST(tmp);
2542 res = kern_path_create(dfd, tmp, path, is_dir);
2543 putname(tmp);
2544 return res;
2545}
2546EXPORT_SYMBOL(user_path_create);
2547
1a67aafb 2548int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
1da177e4 2549{
a95164d9 2550 int error = may_create(dir, dentry);
1da177e4
LT		 2551
2552 if (error)
2553 return error;
2554
975d6b39 2555 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1da177e4
LT		 2556 return -EPERM;
2557
acfa4380 2558 if (!dir->i_op->mknod)
1da177e4
LT		 2559 return -EPERM;
2560
08ce5f16
SH		 2561 error = devcgroup_inode_mknod(mode, dev);
2562 if (error)
2563 return error;
2564
1da177e4
LT		 2565 error = security_inode_mknod(dir, dentry, mode, dev);
2566 if (error)
2567 return error;
2568
1da177e4 2569 error = dir->i_op->mknod(dir, dentry, mode, dev);
a74574aa 2570 if (!error)
f38aa942 2571 fsnotify_create(dir, dentry);
1da177e4
LT		 2572 return error;
2573}
2574
f69aac00 2575static int may_mknod(umode_t mode)
463c3197
DH		 2576{
2577 switch (mode & S_IFMT) {
2578 case S_IFREG:
2579 case S_IFCHR:
2580 case S_IFBLK:
2581 case S_IFIFO:
2582 case S_IFSOCK:
2583 case 0: /* zero mode translates to S_IFREG */
2584 return 0;
2585 case S_IFDIR:
2586 return -EPERM;
2587 default:
2588 return -EINVAL;
2589 }
2590}
2591
8208a22b 2592SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
2e4d0924 2593 unsigned, dev)
1da177e4 2594{
2ad94ae6 2595 struct dentry *dentry;
dae6ad8f
AV		 2596 struct path path;
2597 int error;
1da177e4
LT		 2598
2599 if (S_ISDIR(mode))
2600 return -EPERM;
1da177e4 2601
dae6ad8f
AV		 2602 dentry = user_path_create(dfd, filename, &path, 0);
2603 if (IS_ERR(dentry))
2604 return PTR_ERR(dentry);
2ad94ae6 2605
dae6ad8f 2606 if (!IS_POSIXACL(path.dentry->d_inode))
ce3b0f8d 2607 mode &= ~current_umask();
463c3197
DH		 2608 error = may_mknod(mode);
2609 if (error)
2610 goto out_dput;
dae6ad8f 2611 error = mnt_want_write(path.mnt);
463c3197
DH		 2612 if (error)
2613 goto out_dput;
dae6ad8f 2614 error = security_path_mknod(&path, dentry, mode, dev);
be6d3e56
KT		 2615 if (error)
2616 goto out_drop_write;
463c3197 2617 switch (mode & S_IFMT) {
1da177e4 2618 case 0: case S_IFREG:
dae6ad8f 2619 error = vfs_create(path.dentry->d_inode,dentry,mode,NULL);
1da177e4
LT		 2620 break;
2621 case S_IFCHR: case S_IFBLK:
dae6ad8f 2622 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
1da177e4
LT		 2623 new_decode_dev(dev));
2624 break;
2625 case S_IFIFO: case S_IFSOCK:
dae6ad8f 2626 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
1da177e4 2627 break;
1da177e4 2628 }
be6d3e56 2629out_drop_write:
dae6ad8f 2630 mnt_drop_write(path.mnt);
463c3197
DH		 2631out_dput:
2632 dput(dentry);
dae6ad8f
AV		 2633 mutex_unlock(&path.dentry->d_inode->i_mutex);
2634 path_put(&path);
1da177e4
LT		 2635
2636 return error;
2637}
2638
8208a22b 2639SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
5590ff0d
UD		 2640{
2641 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2642}
2643
18bb1db3 2644int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1da177e4 2645{
a95164d9 2646 int error = may_create(dir, dentry);
8de52778 2647 unsigned max_links = dir->i_sb->s_max_links;
1da177e4
LT		 2648
2649 if (error)
2650 return error;
2651
acfa4380 2652 if (!dir->i_op->mkdir)
1da177e4
LT		 2653 return -EPERM;
2654
2655 mode &= (S_IRWXUGO|S_ISVTX);
2656 error = security_inode_mkdir(dir, dentry, mode);
2657 if (error)
2658 return error;
2659
8de52778
AV		 2660 if (max_links && dir->i_nlink >= max_links)
2661 return -EMLINK;
2662
1da177e4 2663 error = dir->i_op->mkdir(dir, dentry, mode);
a74574aa 2664 if (!error)
f38aa942 2665 fsnotify_mkdir(dir, dentry);
1da177e4
LT		 2666 return error;
2667}
2668
a218d0fd 2669SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
1da177e4 2670{
6902d925 2671 struct dentry *dentry;
dae6ad8f
AV		 2672 struct path path;
2673 int error;
1da177e4 2674
dae6ad8f 2675 dentry = user_path_create(dfd, pathname, &path, 1);
6902d925 2676 if (IS_ERR(dentry))
dae6ad8f 2677 return PTR_ERR(dentry);
1da177e4 2678
dae6ad8f 2679 if (!IS_POSIXACL(path.dentry->d_inode))
ce3b0f8d 2680 mode &= ~current_umask();
dae6ad8f 2681 error = mnt_want_write(path.mnt);
463c3197
DH		 2682 if (error)
2683 goto out_dput;
dae6ad8f 2684 error = security_path_mkdir(&path, dentry, mode);
be6d3e56
KT		 2685 if (error)
2686 goto out_drop_write;
dae6ad8f 2687 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
be6d3e56 2688out_drop_write:
dae6ad8f 2689 mnt_drop_write(path.mnt);
463c3197 2690out_dput:
6902d925 2691 dput(dentry);
dae6ad8f
AV		 2692 mutex_unlock(&path.dentry->d_inode->i_mutex);
2693 path_put(&path);
1da177e4
LT		 2694 return error;
2695}
2696
a218d0fd 2697SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
5590ff0d
UD		 2698{
2699 return sys_mkdirat(AT_FDCWD, pathname, mode);
2700}
2701
1da177e4 2702/*
a71905f0 2703 * The dentry_unhash() helper will try to drop the dentry early: we
c0d02594 2704 * should have a usage count of 1 if we're the only user of this
a71905f0
SW		 2705 * dentry, and if that is true (possibly after pruning the dcache),
2706 * then we drop the dentry now.
1da177e4
LT		 2707 *
2708 * A low-level filesystem can, if it choses, legally
2709 * do a
2710 *
2711 * if (!d_unhashed(dentry))
2712 * return -EBUSY;
2713 *
2714 * if it cannot handle the case of removing a directory
2715 * that is still in use by something else..
2716 */
2717void dentry_unhash(struct dentry *dentry)
2718{
dc168427 2719 shrink_dcache_parent(dentry);
1da177e4 2720 spin_lock(&dentry->d_lock);
64252c75 2721 if (dentry->d_count == 1)
1da177e4
LT		 2722 __d_drop(dentry);
2723 spin_unlock(&dentry->d_lock);
1da177e4
LT		 2724}
2725
2726int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2727{
2728 int error = may_delete(dir, dentry, 1);
2729
2730 if (error)
2731 return error;
2732
acfa4380 2733 if (!dir->i_op->rmdir)
1da177e4
LT		 2734 return -EPERM;
2735
1d2ef590 2736 dget(dentry);
1b1dcc1b 2737 mutex_lock(&dentry->d_inode->i_mutex);
912dbc15
SW		 2738
2739 error = -EBUSY;
1da177e4 2740 if (d_mountpoint(dentry))
912dbc15
SW		 2741 goto out;
2742
2743 error = security_inode_rmdir(dir, dentry);
2744 if (error)
2745 goto out;
2746
3cebde24 2747 shrink_dcache_parent(dentry);
912dbc15
SW		 2748 error = dir->i_op->rmdir(dir, dentry);
2749 if (error)
2750 goto out;
2751
2752 dentry->d_inode->i_flags |= S_DEAD;
2753 dont_mount(dentry);
2754
2755out:
1b1dcc1b 2756 mutex_unlock(&dentry->d_inode->i_mutex);
1d2ef590 2757 dput(dentry);
912dbc15 2758 if (!error)
1da177e4 2759 d_delete(dentry);
1da177e4
LT		 2760 return error;
2761}
2762
5590ff0d 2763static long do_rmdir(int dfd, const char __user *pathname)
1da177e4
LT		 2764{
2765 int error = 0;
2766 char * name;
2767 struct dentry *dentry;
2768 struct nameidata nd;
2769
2ad94ae6 2770 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2771 if (error)
2ad94ae6 2772 return error;
1da177e4
LT		 2773
2774 switch(nd.last_type) {
0612d9fb
OH		 2775 case LAST_DOTDOT:
2776 error = -ENOTEMPTY;
2777 goto exit1;
2778 case LAST_DOT:
2779 error = -EINVAL;
2780 goto exit1;
2781 case LAST_ROOT:
2782 error = -EBUSY;
2783 goto exit1;
1da177e4 2784 }
0612d9fb
OH		 2785
2786 nd.flags &= ~LOOKUP_PARENT;
2787
4ac91378 2788 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2789 dentry = lookup_hash(&nd);
1da177e4 2790 error = PTR_ERR(dentry);
6902d925
DH		 2791 if (IS_ERR(dentry))
2792 goto exit2;
e6bc45d6
TT		 2793 if (!dentry->d_inode) {
2794 error = -ENOENT;
2795 goto exit3;
2796 }
0622753b
DH		 2797 error = mnt_want_write(nd.path.mnt);
2798 if (error)
2799 goto exit3;
be6d3e56
KT		 2800 error = security_path_rmdir(&nd.path, dentry);
2801 if (error)
2802 goto exit4;
4ac91378 2803 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
be6d3e56 2804exit4:
0622753b
DH		 2805 mnt_drop_write(nd.path.mnt);
2806exit3:
6902d925
DH		 2807 dput(dentry);
2808exit2:
4ac91378 2809 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2810exit1:
1d957f9b 2811 path_put(&nd.path);
1da177e4
LT		 2812 putname(name);
2813 return error;
2814}
2815
3cdad428 2816SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
5590ff0d
UD		 2817{
2818 return do_rmdir(AT_FDCWD, pathname);
2819}
2820
1da177e4
LT		 2821int vfs_unlink(struct inode *dir, struct dentry *dentry)
2822{
2823 int error = may_delete(dir, dentry, 0);
2824
2825 if (error)
2826 return error;
2827
acfa4380 2828 if (!dir->i_op->unlink)
1da177e4
LT		 2829 return -EPERM;
2830
1b1dcc1b 2831 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2832 if (d_mountpoint(dentry))
2833 error = -EBUSY;
2834 else {
2835 error = security_inode_unlink(dir, dentry);
bec1052e 2836 if (!error) {
1da177e4 2837 error = dir->i_op->unlink(dir, dentry);
bec1052e 2838 if (!error)
d83c49f3 2839 dont_mount(dentry);
bec1052e 2840 }
1da177e4 2841 }
1b1dcc1b 2842 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2843
2844 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2845 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
ece95912 2846 fsnotify_link_count(dentry->d_inode);
e234f35c 2847 d_delete(dentry);
1da177e4 2848 }
0eeca283 2849
1da177e4
LT		 2850 return error;
2851}
2852
2853/*
2854 * Make sure that the actual truncation of the file will occur outside its
1b1dcc1b 2855 * directory's i_mutex. Truncate can take a long time if there is a lot of
1da177e4
LT		 2856 * writeout happening, and we don't want to prevent access to the directory
2857 * while waiting on the I/O.
2858 */
5590ff0d 2859static long do_unlinkat(int dfd, const char __user *pathname)
1da177e4 2860{
2ad94ae6
AV		 2861 int error;
2862 char *name;
1da177e4
LT		 2863 struct dentry *dentry;
2864 struct nameidata nd;
2865 struct inode *inode = NULL;
2866
2ad94ae6 2867 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2868 if (error)
2ad94ae6
AV		 2869 return error;
2870
1da177e4
LT		 2871 error = -EISDIR;
2872 if (nd.last_type != LAST_NORM)
2873 goto exit1;
0612d9fb
OH		 2874
2875 nd.flags &= ~LOOKUP_PARENT;
2876
4ac91378 2877 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2878 dentry = lookup_hash(&nd);
1da177e4
LT		 2879 error = PTR_ERR(dentry);
2880 if (!IS_ERR(dentry)) {
2881 /* Why not before? Because we want correct error value */
50338b88
TE		 2882 if (nd.last.name[nd.last.len])
2883 goto slashes;
1da177e4 2884 inode = dentry->d_inode;
50338b88 2885 if (!inode)
e6bc45d6
TT		 2886 goto slashes;
2887 ihold(inode);
0622753b
DH		 2888 error = mnt_want_write(nd.path.mnt);
2889 if (error)
2890 goto exit2;
be6d3e56
KT		 2891 error = security_path_unlink(&nd.path, dentry);
2892 if (error)
2893 goto exit3;
4ac91378 2894 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
be6d3e56 2895exit3:
0622753b 2896 mnt_drop_write(nd.path.mnt);
1da177e4
LT		 2897 exit2:
2898 dput(dentry);
2899 }
4ac91378 2900 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4
LT		 2901 if (inode)
2902 iput(inode); /* truncate the inode here */
2903exit1:
1d957f9b 2904 path_put(&nd.path);
1da177e4
LT		 2905 putname(name);
2906 return error;
2907
2908slashes:
2909 error = !dentry->d_inode ? -ENOENT :
2910 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2911 goto exit2;
2912}
2913
2e4d0924 2914SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
5590ff0d
UD		 2915{
2916 if ((flag & ~AT_REMOVEDIR) != 0)
2917 return -EINVAL;
2918
2919 if (flag & AT_REMOVEDIR)
2920 return do_rmdir(dfd, pathname);
2921
2922 return do_unlinkat(dfd, pathname);
2923}
2924
3480b257 2925SYSCALL_DEFINE1(unlink, const char __user *, pathname)
5590ff0d
UD		 2926{
2927 return do_unlinkat(AT_FDCWD, pathname);
2928}
2929
db2e747b 2930int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
1da177e4 2931{
a95164d9 2932 int error = may_create(dir, dentry);
1da177e4
LT		 2933
2934 if (error)
2935 return error;
2936
acfa4380 2937 if (!dir->i_op->symlink)
1da177e4
LT		 2938 return -EPERM;
2939
2940 error = security_inode_symlink(dir, dentry, oldname);
2941 if (error)
2942 return error;
2943
1da177e4 2944 error = dir->i_op->symlink(dir, dentry, oldname);
a74574aa 2945 if (!error)
f38aa942 2946 fsnotify_create(dir, dentry);
1da177e4
LT		 2947 return error;
2948}
2949
2e4d0924
HC		 2950SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2951 int, newdfd, const char __user *, newname)
1da177e4 2952{
2ad94ae6
AV		 2953 int error;
2954 char *from;
6902d925 2955 struct dentry *dentry;
dae6ad8f 2956 struct path path;
1da177e4
LT		 2957
2958 from = getname(oldname);
2ad94ae6 2959 if (IS_ERR(from))
1da177e4 2960 return PTR_ERR(from);
1da177e4 2961
dae6ad8f 2962 dentry = user_path_create(newdfd, newname, &path, 0);
6902d925
DH		 2963 error = PTR_ERR(dentry);
2964 if (IS_ERR(dentry))
dae6ad8f 2965 goto out_putname;
6902d925 2966
dae6ad8f 2967 error = mnt_want_write(path.mnt);
75c3f29d
DH		 2968 if (error)
2969 goto out_dput;
dae6ad8f 2970 error = security_path_symlink(&path, dentry, from);
be6d3e56
KT		 2971 if (error)
2972 goto out_drop_write;
dae6ad8f 2973 error = vfs_symlink(path.dentry->d_inode, dentry, from);
be6d3e56 2974out_drop_write:
dae6ad8f 2975 mnt_drop_write(path.mnt);
75c3f29d 2976out_dput:
6902d925 2977 dput(dentry);
dae6ad8f
AV		 2978 mutex_unlock(&path.dentry->d_inode->i_mutex);
2979 path_put(&path);
6902d925 2980out_putname:
1da177e4
LT		 2981 putname(from);
2982 return error;
2983}
2984
3480b257 2985SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 2986{
2987 return sys_symlinkat(oldname, AT_FDCWD, newname);
2988}
2989
1da177e4
LT		 2990int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2991{
2992 struct inode *inode = old_dentry->d_inode;
8de52778 2993 unsigned max_links = dir->i_sb->s_max_links;
1da177e4
LT		 2994 int error;
2995
2996 if (!inode)
2997 return -ENOENT;
2998
a95164d9 2999 error = may_create(dir, new_dentry);
1da177e4
LT		 3000 if (error)
3001 return error;
3002
3003 if (dir->i_sb != inode->i_sb)
3004 return -EXDEV;
3005
3006 /*
3007 * A link to an append-only or immutable file cannot be created.
3008 */
3009 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3010 return -EPERM;
acfa4380 3011 if (!dir->i_op->link)
1da177e4 3012 return -EPERM;
7e79eedb 3013 if (S_ISDIR(inode->i_mode))
1da177e4
LT		 3014 return -EPERM;
3015
3016 error = security_inode_link(old_dentry, dir, new_dentry);
3017 if (error)
3018 return error;
3019
7e79eedb 3020 mutex_lock(&inode->i_mutex);
aae8a97d
AK		 3021 /* Make sure we don't allow creating hardlink to an unlinked file */
3022 if (inode->i_nlink == 0)
3023 error = -ENOENT;
8de52778
AV		 3024 else if (max_links && inode->i_nlink >= max_links)
3025 error = -EMLINK;
aae8a97d
AK		 3026 else
3027 error = dir->i_op->link(old_dentry, dir, new_dentry);
7e79eedb 3028 mutex_unlock(&inode->i_mutex);
e31e14ec 3029 if (!error)
7e79eedb 3030 fsnotify_link(dir, inode, new_dentry);
1da177e4
LT		 3031 return error;
3032}
3033
3034/*
3035 * Hardlinks are often used in delicate situations. We avoid
3036 * security-related surprises by not following symlinks on the
3037 * newname. --KAB
3038 *
3039 * We don't follow them on the oldname either to be compatible
3040 * with linux 2.0, and to avoid hard-linking to directories
3041 * and other special files. --ADM
3042 */
2e4d0924
HC		 3043SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3044 int, newdfd, const char __user *, newname, int, flags)
1da177e4
LT		 3045{
3046 struct dentry *new_dentry;
dae6ad8f 3047 struct path old_path, new_path;
11a7b371 3048 int how = 0;
1da177e4 3049 int error;
1da177e4 3050
11a7b371 3051 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
c04030e1 3052 return -EINVAL;
11a7b371
AK		 3053 /*
3054 * To use null names we require CAP_DAC_READ_SEARCH
3055 * This ensures that not everyone will be able to create
3056 * handlink using the passed filedescriptor.
3057 */
3058 if (flags & AT_EMPTY_PATH) {
3059 if (!capable(CAP_DAC_READ_SEARCH))
3060 return -ENOENT;
3061 how = LOOKUP_EMPTY;
3062 }
3063
3064 if (flags & AT_SYMLINK_FOLLOW)
3065 how |= LOOKUP_FOLLOW;
c04030e1 3066
11a7b371 3067 error = user_path_at(olddfd, oldname, how, &old_path);
1da177e4 3068 if (error)
2ad94ae6
AV		 3069 return error;
3070
dae6ad8f 3071 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
1da177e4 3072 error = PTR_ERR(new_dentry);
6902d925 3073 if (IS_ERR(new_dentry))
dae6ad8f
AV		 3074 goto out;
3075
3076 error = -EXDEV;
3077 if (old_path.mnt != new_path.mnt)
3078 goto out_dput;
3079 error = mnt_want_write(new_path.mnt);
75c3f29d
DH		 3080 if (error)
3081 goto out_dput;
dae6ad8f 3082 error = security_path_link(old_path.dentry, &new_path, new_dentry);
be6d3e56
KT		 3083 if (error)
3084 goto out_drop_write;
dae6ad8f 3085 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
be6d3e56 3086out_drop_write:
dae6ad8f 3087 mnt_drop_write(new_path.mnt);
75c3f29d 3088out_dput:
6902d925 3089 dput(new_dentry);
dae6ad8f
AV		 3090 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
3091 path_put(&new_path);
1da177e4 3092out:
2d8f3038 3093 path_put(&old_path);
1da177e4
LT		 3094
3095 return error;
3096}
3097
3480b257 3098SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
5590ff0d 3099{
c04030e1 3100 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
5590ff0d
UD		 3101}
3102
1da177e4
LT		 3103/*
3104 * The worst of all namespace operations - renaming directory. "Perverted"
3105 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3106 * Problems:
3107 * a) we can get into loop creation. Check is done in is_subdir().
3108 * b) race potential - two innocent renames can create a loop together.
3109 * That's where 4.4 screws up. Current fix: serialization on
a11f3a05 3110 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
1da177e4
LT		 3111 * story.
3112 * c) we have to lock _three_ objects - parents and victim (if it exists).
1b1dcc1b 3113 * And that - after we got ->i_mutex on parents (until then we don't know
1da177e4
LT		 3114 * whether the target exists). Solution: try to be smart with locking
3115 * order for inodes. We rely on the fact that tree topology may change
a11f3a05 3116 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
1da177e4
LT		 3117 * move will be locked. Thus we can rank directories by the tree
3118 * (ancestors first) and rank all non-directories after them.
3119 * That works since everybody except rename does "lock parent, lookup,
a11f3a05 3120 * lock child" and rename is under ->s_vfs_rename_mutex.
1da177e4
LT		 3121 * HOWEVER, it relies on the assumption that any object with ->lookup()
3122 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3123 * we'd better make sure that there's no link(2) for them.
e4eaac06 3124 * d) conversion from fhandle to dentry may come in the wrong moment - when
1b1dcc1b 3125 * we are removing the target. Solution: we will have to grab ->i_mutex
1da177e4 3126 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
c41b20e7 3127 * ->i_mutex on parents, which works but leads to some truly excessive
1da177e4
LT		 3128 * locking].
3129 */
75c96f85
AB		 3130static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3131 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3132{
3133 int error = 0;
9055cba7 3134 struct inode *target = new_dentry->d_inode;
8de52778 3135 unsigned max_links = new_dir->i_sb->s_max_links;
1da177e4
LT		 3136
3137 /*
3138 * If we are going to change the parent - check write permissions,
3139 * we'll need to flip '..'.
3140 */
3141 if (new_dir != old_dir) {
f419a2e3 3142 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
1da177e4
LT		 3143 if (error)
3144 return error;
3145 }
3146
3147 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3148 if (error)
3149 return error;
3150
1d2ef590 3151 dget(new_dentry);
d83c49f3 3152 if (target)
1b1dcc1b 3153 mutex_lock(&target->i_mutex);
9055cba7
SW		 3154
3155 error = -EBUSY;
3156 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3157 goto out;
3158
8de52778
AV		 3159 error = -EMLINK;
3160 if (max_links && !target && new_dir != old_dir &&
3161 new_dir->i_nlink >= max_links)
3162 goto out;
3163
3cebde24
SW		 3164 if (target)
3165 shrink_dcache_parent(new_dentry);
9055cba7
SW		 3166 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3167 if (error)
3168 goto out;
3169
1da177e4 3170 if (target) {
9055cba7
SW		 3171 target->i_flags |= S_DEAD;
3172 dont_mount(new_dentry);
1da177e4 3173 }
9055cba7
SW		 3174out:
3175 if (target)
3176 mutex_unlock(&target->i_mutex);
1d2ef590 3177 dput(new_dentry);
e31e14ec 3178 if (!error)
349457cc
MF		 3179 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3180 d_move(old_dentry,new_dentry);
1da177e4
LT		 3181 return error;
3182}
3183
75c96f85
AB		 3184static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3185 struct inode *new_dir, struct dentry *new_dentry)
1da177e4 3186{
51892bbb 3187 struct inode *target = new_dentry->d_inode;
1da177e4
LT		 3188 int error;
3189
3190 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3191 if (error)
3192 return error;
3193
3194 dget(new_dentry);
1da177e4 3195 if (target)
1b1dcc1b 3196 mutex_lock(&target->i_mutex);
51892bbb
SW		 3197
3198 error = -EBUSY;
1da177e4 3199 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
51892bbb
SW		 3200 goto out;
3201
3202 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3203 if (error)
3204 goto out;
3205
3206 if (target)
3207 dont_mount(new_dentry);
3208 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3209 d_move(old_dentry, new_dentry);
3210out:
1da177e4 3211 if (target)
1b1dcc1b 3212 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3213 dput(new_dentry);
3214 return error;
3215}
3216
3217int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3218 struct inode *new_dir, struct dentry *new_dentry)
3219{
3220 int error;
3221 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
59b0df21 3222 const unsigned char *old_name;
1da177e4
LT		 3223
3224 if (old_dentry->d_inode == new_dentry->d_inode)
3225 return 0;
3226
3227 error = may_delete(old_dir, old_dentry, is_dir);
3228 if (error)
3229 return error;
3230
3231 if (!new_dentry->d_inode)
a95164d9 3232 error = may_create(new_dir, new_dentry);
1da177e4
LT		 3233 else
3234 error = may_delete(new_dir, new_dentry, is_dir);
3235 if (error)
3236 return error;
3237
acfa4380 3238 if (!old_dir->i_op->rename)
1da177e4
LT		 3239 return -EPERM;
3240
0eeca283
RL		 3241 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3242
1da177e4
LT		 3243 if (is_dir)
3244 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3245 else
3246 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
123df294
AV		 3247 if (!error)
3248 fsnotify_move(old_dir, new_dir, old_name, is_dir,
5a190ae6 3249 new_dentry->d_inode, old_dentry);
0eeca283
RL		 3250 fsnotify_oldname_free(old_name);
3251
1da177e4
LT		 3252 return error;
3253}
3254
2e4d0924
HC		 3255SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3256 int, newdfd, const char __user *, newname)
1da177e4 3257{
2ad94ae6
AV		 3258 struct dentry *old_dir, *new_dir;
3259 struct dentry *old_dentry, *new_dentry;
3260 struct dentry *trap;
1da177e4 3261 struct nameidata oldnd, newnd;
2ad94ae6
AV		 3262 char *from;
3263 char *to;
3264 int error;
1da177e4 3265
2ad94ae6 3266 error = user_path_parent(olddfd, oldname, &oldnd, &from);
1da177e4
LT		 3267 if (error)
3268 goto exit;
3269
2ad94ae6 3270 error = user_path_parent(newdfd, newname, &newnd, &to);
1da177e4
LT		 3271 if (error)
3272 goto exit1;
3273
3274 error = -EXDEV;
4ac91378 3275 if (oldnd.path.mnt != newnd.path.mnt)
1da177e4
LT		 3276 goto exit2;
3277
4ac91378 3278 old_dir = oldnd.path.dentry;
1da177e4
LT		 3279 error = -EBUSY;
3280 if (oldnd.last_type != LAST_NORM)
3281 goto exit2;
3282
4ac91378 3283 new_dir = newnd.path.dentry;
1da177e4
LT		 3284 if (newnd.last_type != LAST_NORM)
3285 goto exit2;
3286
0612d9fb
OH		 3287 oldnd.flags &= ~LOOKUP_PARENT;
3288 newnd.flags &= ~LOOKUP_PARENT;
4e9ed2f8 3289 newnd.flags |= LOOKUP_RENAME_TARGET;
0612d9fb 3290
1da177e4
LT		 3291 trap = lock_rename(new_dir, old_dir);
3292
49705b77 3293 old_dentry = lookup_hash(&oldnd);
1da177e4
LT		 3294 error = PTR_ERR(old_dentry);
3295 if (IS_ERR(old_dentry))
3296 goto exit3;
3297 /* source must exist */
3298 error = -ENOENT;
3299 if (!old_dentry->d_inode)
3300 goto exit4;
3301 /* unless the source is a directory trailing slashes give -ENOTDIR */
3302 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3303 error = -ENOTDIR;
3304 if (oldnd.last.name[oldnd.last.len])
3305 goto exit4;
3306 if (newnd.last.name[newnd.last.len])
3307 goto exit4;
3308 }
3309 /* source should not be ancestor of target */
3310 error = -EINVAL;
3311 if (old_dentry == trap)
3312 goto exit4;
49705b77 3313 new_dentry = lookup_hash(&newnd);
1da177e4
LT		 3314 error = PTR_ERR(new_dentry);
3315 if (IS_ERR(new_dentry))
3316 goto exit4;
3317 /* target should not be an ancestor of source */
3318 error = -ENOTEMPTY;
3319 if (new_dentry == trap)
3320 goto exit5;
3321
9079b1eb
DH		 3322 error = mnt_want_write(oldnd.path.mnt);
3323 if (error)
3324 goto exit5;
be6d3e56
KT		 3325 error = security_path_rename(&oldnd.path, old_dentry,
3326 &newnd.path, new_dentry);
3327 if (error)
3328 goto exit6;
1da177e4
LT		 3329 error = vfs_rename(old_dir->d_inode, old_dentry,
3330 new_dir->d_inode, new_dentry);
be6d3e56 3331exit6:
9079b1eb 3332 mnt_drop_write(oldnd.path.mnt);
1da177e4
LT		 3333exit5:
3334 dput(new_dentry);
3335exit4:
3336 dput(old_dentry);
3337exit3:
3338 unlock_rename(new_dir, old_dir);
3339exit2:
1d957f9b 3340 path_put(&newnd.path);
2ad94ae6 3341 putname(to);
1da177e4 3342exit1:
1d957f9b 3343 path_put(&oldnd.path);
1da177e4 3344 putname(from);
2ad94ae6 3345exit:
1da177e4
LT		 3346 return error;
3347}
3348
a26eab24 3349SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 3350{
3351 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3352}
3353
1da177e4
LT		 3354int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3355{
3356 int len;
3357
3358 len = PTR_ERR(link);
3359 if (IS_ERR(link))
3360 goto out;
3361
3362 len = strlen(link);
3363 if (len > (unsigned) buflen)
3364 len = buflen;
3365 if (copy_to_user(buffer, link, len))
3366 len = -EFAULT;
3367out:
3368 return len;
3369}
3370
3371/*
3372 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3373 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3374 * using) it for any given inode is up to filesystem.
3375 */
3376int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3377{
3378 struct nameidata nd;
cc314eef 3379 void *cookie;
694a1764 3380 int res;
cc314eef 3381
1da177e4 3382 nd.depth = 0;
cc314eef 3383 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
694a1764
MS		 3384 if (IS_ERR(cookie))
3385 return PTR_ERR(cookie);
3386
3387 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3388 if (dentry->d_inode->i_op->put_link)
3389 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3390 return res;
1da177e4
LT		 3391}
3392
3393int vfs_follow_link(struct nameidata *nd, const char *link)
3394{
3395 return __vfs_follow_link(nd, link);
3396}
3397
3398/* get the link contents into pagecache */
3399static char *page_getlink(struct dentry * dentry, struct page **ppage)
3400{
ebd09abb
DG		 3401 char *kaddr;
3402 struct page *page;
1da177e4 3403 struct address_space *mapping = dentry->d_inode->i_mapping;
090d2b18 3404 page = read_mapping_page(mapping, 0, NULL);
1da177e4 3405 if (IS_ERR(page))
6fe6900e 3406 return (char*)page;
1da177e4 3407 *ppage = page;
ebd09abb
DG		 3408 kaddr = kmap(page);
3409 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3410 return kaddr;
1da177e4
LT		 3411}
3412
3413int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3414{
3415 struct page *page = NULL;
3416 char *s = page_getlink(dentry, &page);
3417 int res = vfs_readlink(dentry,buffer,buflen,s);
3418 if (page) {
3419 kunmap(page);
3420 page_cache_release(page);
3421 }
3422 return res;
3423}
3424
cc314eef 3425void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
1da177e4 3426{
cc314eef 3427 struct page *page = NULL;
1da177e4 3428 nd_set_link(nd, page_getlink(dentry, &page));
cc314eef 3429 return page;
1da177e4
LT		 3430}
3431
cc314eef 3432void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1da177e4 3433{
cc314eef
LT		 3434 struct page *page = cookie;
3435
3436 if (page) {
1da177e4
LT		 3437 kunmap(page);
3438 page_cache_release(page);
1da177e4
LT		 3439 }
3440}
3441
54566b2c
NP		 3442/*
3443 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3444 */
3445int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
1da177e4
LT		 3446{
3447 struct address_space *mapping = inode->i_mapping;
0adb25d2 3448 struct page *page;
afddba49 3449 void *fsdata;
beb497ab 3450 int err;
1da177e4 3451 char *kaddr;
54566b2c
NP		 3452 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3453 if (nofs)
3454 flags |= AOP_FLAG_NOFS;
1da177e4 3455
7e53cac4 3456retry:
afddba49 3457 err = pagecache_write_begin(NULL, mapping, 0, len-1,
54566b2c 3458 flags, &page, &fsdata);
1da177e4 3459 if (err)
afddba49
NP		 3460 goto fail;
3461
e8e3c3d6 3462 kaddr = kmap_atomic(page);
1da177e4 3463 memcpy(kaddr, symname, len-1);
e8e3c3d6 3464 kunmap_atomic(kaddr);
afddba49
NP		 3465
3466 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3467 page, fsdata);
1da177e4
LT		 3468 if (err < 0)
3469 goto fail;
afddba49
NP		 3470 if (err < len-1)
3471 goto retry;
3472
1da177e4
LT		 3473 mark_inode_dirty(inode);
3474 return 0;
1da177e4
LT		 3475fail:
3476 return err;
3477}
3478
0adb25d2
KK		 3479int page_symlink(struct inode *inode, const char *symname, int len)
3480{
3481 return __page_symlink(inode, symname, len,
54566b2c 3482 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
0adb25d2
KK		 3483}
3484
92e1d5be 3485const struct inode_operations page_symlink_inode_operations = {
1da177e4
LT		 3486 .readlink = generic_readlink,
3487 .follow_link = page_follow_link_light,
3488 .put_link = page_put_link,
3489};
3490
2d8f3038 3491EXPORT_SYMBOL(user_path_at);
cc53ce53 3492EXPORT_SYMBOL(follow_down_one);
1da177e4
LT		 3493EXPORT_SYMBOL(follow_down);
3494EXPORT_SYMBOL(follow_up);
3495EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3496EXPORT_SYMBOL(getname);
3497EXPORT_SYMBOL(lock_rename);
1da177e4
LT		 3498EXPORT_SYMBOL(lookup_one_len);
3499EXPORT_SYMBOL(page_follow_link_light);
3500EXPORT_SYMBOL(page_put_link);
3501EXPORT_SYMBOL(page_readlink);
0adb25d2 3502EXPORT_SYMBOL(__page_symlink);
1da177e4
LT		 3503EXPORT_SYMBOL(page_symlink);
3504EXPORT_SYMBOL(page_symlink_inode_operations);
d1811465 3505EXPORT_SYMBOL(kern_path);
16f18200 3506EXPORT_SYMBOL(vfs_path_lookup);
f419a2e3 3507EXPORT_SYMBOL(inode_permission);
1da177e4
LT		 3508EXPORT_SYMBOL(unlock_rename);
3509EXPORT_SYMBOL(vfs_create);
3510EXPORT_SYMBOL(vfs_follow_link);
3511EXPORT_SYMBOL(vfs_link);
3512EXPORT_SYMBOL(vfs_mkdir);
3513EXPORT_SYMBOL(vfs_mknod);
3514EXPORT_SYMBOL(generic_permission);
3515EXPORT_SYMBOL(vfs_readlink);
3516EXPORT_SYMBOL(vfs_rename);
3517EXPORT_SYMBOL(vfs_rmdir);
3518EXPORT_SYMBOL(vfs_symlink);
3519EXPORT_SYMBOL(vfs_unlink);
3520EXPORT_SYMBOL(dentry_unhash);
3521EXPORT_SYMBOL(generic_readlink);
kernel source for telekom puls (alcatel/mediatek)