projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'fixes-urgent' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / fs / namei.c
CommitLineData
1da177e4
LT		 1/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
18#include <linux/module.h>
19#include <linux/slab.h>
20#include <linux/fs.h>
21#include <linux/namei.h>
1da177e4 22#include <linux/pagemap.h>
0eeca283 23#include <linux/fsnotify.h>
1da177e4
LT		 24#include <linux/personality.h>
25#include <linux/security.h>
6146f0d5 26#include <linux/ima.h>
1da177e4
LT		 27#include <linux/syscalls.h>
28#include <linux/mount.h>
29#include <linux/audit.h>
16f7e0fe 30#include <linux/capability.h>
834f2a4a 31#include <linux/file.h>
5590ff0d 32#include <linux/fcntl.h>
08ce5f16 33#include <linux/device_cgroup.h>
5ad4e53b 34#include <linux/fs_struct.h>
e77819e5 35#include <linux/posix_acl.h>
1da177e4
LT		 36#include <asm/uaccess.h>
37
e81e3f4d 38#include "internal.h"
c7105365 39#include "mount.h"
e81e3f4d 40
1da177e4
LT		 41/* [Feb-1997 T. Schoebel-Theuer]
42 * Fundamental changes in the pathname lookup mechanisms (namei)
43 * were necessary because of omirr. The reason is that omirr needs
44 * to know the _real_ pathname, not the user-supplied one, in case
45 * of symlinks (and also when transname replacements occur).
46 *
47 * The new code replaces the old recursive symlink resolution with
48 * an iterative one (in case of non-nested symlink chains). It does
49 * this with calls to <fs>_follow_link().
50 * As a side effect, dir_namei(), _namei() and follow_link() are now
51 * replaced with a single function lookup_dentry() that can handle all
52 * the special cases of the former code.
53 *
54 * With the new dcache, the pathname is stored at each inode, at least as
55 * long as the refcount of the inode is positive. As a side effect, the
56 * size of the dcache depends on the inode cache and thus is dynamic.
57 *
58 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
59 * resolution to correspond with current state of the code.
60 *
61 * Note that the symlink resolution is not *completely* iterative.
62 * There is still a significant amount of tail- and mid- recursion in
63 * the algorithm. Also, note that <fs>_readlink() is not used in
64 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
65 * may return different results than <fs>_follow_link(). Many virtual
66 * filesystems (including /proc) exhibit this behavior.
67 */
68
69/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
70 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
71 * and the name already exists in form of a symlink, try to create the new
72 * name indicated by the symlink. The old code always complained that the
73 * name already exists, due to not following the symlink even if its target
74 * is nonexistent. The new semantics affects also mknod() and link() when
25985edc 75 * the name is a symlink pointing to a non-existent name.
1da177e4
LT		 76 *
77 * I don't know which semantics is the right one, since I have no access
78 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
79 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
80 * "old" one. Personally, I think the new semantics is much more logical.
81 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
82 * file does succeed in both HP-UX and SunOs, but not in Solaris
83 * and in the old Linux semantics.
84 */
85
86/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
87 * semantics. See the comments in "open_namei" and "do_link" below.
88 *
89 * [10-Sep-98 Alan Modra] Another symlink change.
90 */
91
92/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
93 * inside the path - always follow.
94 * in the last component in creation/removal/renaming - never follow.
95 * if LOOKUP_FOLLOW passed - follow.
96 * if the pathname has trailing slashes - follow.
97 * otherwise - don't follow.
98 * (applied in that order).
99 *
100 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
101 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
102 * During the 2.4 we need to fix the userland stuff depending on it -
103 * hopefully we will be able to get rid of that wart in 2.5. So far only
104 * XEmacs seems to be relying on it...
105 */
106/*
107 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
a11f3a05 108 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
1da177e4
LT		 109 * any extra contention...
110 */
111
112/* In order to reduce some races, while at the same time doing additional
113 * checking and hopefully speeding things up, we copy filenames to the
114 * kernel data space before using them..
115 *
116 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
117 * PATH_MAX includes the nul terminator --RR.
118 */
858119e1 119static int do_getname(const char __user *filename, char *page)
1da177e4
LT		 120{
121 int retval;
122 unsigned long len = PATH_MAX;
123
124 if (!segment_eq(get_fs(), KERNEL_DS)) {
125 if ((unsigned long) filename >= TASK_SIZE)
126 return -EFAULT;
127 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
128 len = TASK_SIZE - (unsigned long) filename;
129 }
130
131 retval = strncpy_from_user(page, filename, len);
132 if (retval > 0) {
133 if (retval < len)
134 return 0;
135 return -ENAMETOOLONG;
136 } else if (!retval)
137 retval = -ENOENT;
138 return retval;
139}
140
1fa1e7f6 141static char *getname_flags(const char __user *filename, int flags, int *empty)
1da177e4 142{
4043cde8
EP		 143 char *result = __getname();
144 int retval;
145
146 if (!result)
147 return ERR_PTR(-ENOMEM);
148
149 retval = do_getname(filename, result);
150 if (retval < 0) {
151 if (retval == -ENOENT && empty)
152 *empty = 1;
153 if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) {
154 __putname(result);
155 return ERR_PTR(retval);
1da177e4
LT		 156 }
157 }
158 audit_getname(result);
159 return result;
160}
161
f52e0c11
AV		 162char *getname(const char __user * filename)
163{
1fa1e7f6 164 return getname_flags(filename, 0, 0);
f52e0c11
AV		 165}
166
1da177e4
LT		 167#ifdef CONFIG_AUDITSYSCALL
168void putname(const char *name)
169{
5ac3a9c2 170 if (unlikely(!audit_dummy_context()))
1da177e4
LT		 171 audit_putname(name);
172 else
173 __putname(name);
174}
175EXPORT_SYMBOL(putname);
176#endif
177
e77819e5
LT		 178static int check_acl(struct inode *inode, int mask)
179{
84635d68 180#ifdef CONFIG_FS_POSIX_ACL
e77819e5
LT		 181 struct posix_acl *acl;
182
e77819e5 183 if (mask & MAY_NOT_BLOCK) {
3567866b
AV		 184 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
185 if (!acl)
e77819e5 186 return -EAGAIN;
3567866b
AV		 187 /* no ->get_acl() calls in RCU mode... */
188 if (acl == ACL_NOT_CACHED)
189 return -ECHILD;
206b1d09 190 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
e77819e5
LT		 191 }
192
193 acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
194
195 /*
4e34e719
CH		 196 * A filesystem can force a ACL callback by just never filling the
197 * ACL cache. But normally you'd fill the cache either at inode
198 * instantiation time, or on the first ->get_acl call.
e77819e5 199 *
4e34e719
CH		 200 * If the filesystem doesn't have a get_acl() function at all, we'll
201 * just create the negative cache entry.
e77819e5
LT		 202 */
203 if (acl == ACL_NOT_CACHED) {
4e34e719
CH		 204 if (inode->i_op->get_acl) {
205 acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS);
206 if (IS_ERR(acl))
207 return PTR_ERR(acl);
208 } else {
209 set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
210 return -EAGAIN;
211 }
e77819e5
LT		 212 }
213
214 if (acl) {
215 int error = posix_acl_permission(inode, acl, mask);
216 posix_acl_release(acl);
217 return error;
218 }
84635d68 219#endif
e77819e5
LT		 220
221 return -EAGAIN;
222}
223
5909ccaa 224/*
948409c7 225 * This does the basic permission checking
1da177e4 226 */
7e40145e 227static int acl_permission_check(struct inode *inode, int mask)
1da177e4 228{
26cf46be 229 unsigned int mode = inode->i_mode;
1da177e4 230
e795b717
SH		 231 if (current_user_ns() != inode_userns(inode))
232 goto other_perms;
233
14067ff5 234 if (likely(current_fsuid() == inode->i_uid))
1da177e4
LT		 235 mode >>= 6;
236 else {
e77819e5 237 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
7e40145e 238 int error = check_acl(inode, mask);
b74c79e9
NP		 239 if (error != -EAGAIN)
240 return error;
1da177e4
LT		 241 }
242
243 if (in_group_p(inode->i_gid))
244 mode >>= 3;
245 }
246
e795b717 247other_perms:
1da177e4
LT		 248 /*
249 * If the DACs are ok we don't need any capability check.
250 */
9c2c7039 251 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
1da177e4 252 return 0;
5909ccaa
LT		 253 return -EACCES;
254}
255
256/**
b74c79e9 257 * generic_permission - check for access rights on a Posix-like filesystem
5909ccaa 258 * @inode: inode to check access rights for
8fd90c8d 259 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
5909ccaa
LT		 260 *
261 * Used to check for read/write/execute permissions on a file.
262 * We use "fsuid" for this, letting us set arbitrary permissions
263 * for filesystem access without changing the "normal" uids which
b74c79e9
NP		 264 * are used for other things.
265 *
266 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
267 * request cannot be satisfied (eg. requires blocking or too much complexity).
268 * It would then be called again in ref-walk mode.
5909ccaa 269 */
2830ba7f 270int generic_permission(struct inode *inode, int mask)
5909ccaa
LT		 271{
272 int ret;
273
274 /*
948409c7 275 * Do the basic permission checks.
5909ccaa 276 */
7e40145e 277 ret = acl_permission_check(inode, mask);
5909ccaa
LT		 278 if (ret != -EACCES)
279 return ret;
1da177e4 280
d594e7ec
AV		 281 if (S_ISDIR(inode->i_mode)) {
282 /* DACs are overridable for directories */
283 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
284 return 0;
285 if (!(mask & MAY_WRITE))
286 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
287 return 0;
288 return -EACCES;
289 }
1da177e4
LT		 290 /*
291 * Read/write DACs are always overridable.
d594e7ec
AV		 292 * Executable DACs are overridable when there is
293 * at least one exec bit set.
1da177e4 294 */
d594e7ec 295 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
e795b717 296 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
1da177e4
LT		 297 return 0;
298
299 /*
300 * Searching includes executable on directories, else just read.
301 */
7ea66001 302 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
d594e7ec 303 if (mask == MAY_READ)
e795b717 304 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
1da177e4
LT		 305 return 0;
306
307 return -EACCES;
308}
309
3ddcd056
LT		 310/*
311 * We _really_ want to just do "generic_permission()" without
312 * even looking at the inode->i_op values. So we keep a cache
313 * flag in inode->i_opflags, that says "this has not special
314 * permission function, use the fast case".
315 */
316static inline int do_inode_permission(struct inode *inode, int mask)
317{
318 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
319 if (likely(inode->i_op->permission))
320 return inode->i_op->permission(inode, mask);
321
322 /* This gets set once for the inode lifetime */
323 spin_lock(&inode->i_lock);
324 inode->i_opflags |= IOP_FASTPERM;
325 spin_unlock(&inode->i_lock);
326 }
327 return generic_permission(inode, mask);
328}
329
cb23beb5
CH		 330/**
331 * inode_permission - check for access rights to a given inode
332 * @inode: inode to check permission on
8fd90c8d 333 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
cb23beb5
CH		 334 *
335 * Used to check for read/write/execute permissions on an inode.
336 * We use "fsuid" for this, letting us set arbitrary permissions
337 * for filesystem access without changing the "normal" uids which
338 * are used for other things.
948409c7
AG		 339 *
340 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
cb23beb5 341 */
f419a2e3 342int inode_permission(struct inode *inode, int mask)
1da177e4 343{
e6305c43 344 int retval;
1da177e4 345
3ddcd056 346 if (unlikely(mask & MAY_WRITE)) {
22590e41 347 umode_t mode = inode->i_mode;
1da177e4
LT		 348
349 /*
350 * Nobody gets write access to a read-only fs.
351 */
352 if (IS_RDONLY(inode) &&
353 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
354 return -EROFS;
355
356 /*
357 * Nobody gets write access to an immutable file.
358 */
359 if (IS_IMMUTABLE(inode))
360 return -EACCES;
361 }
362
3ddcd056 363 retval = do_inode_permission(inode, mask);
1da177e4
LT		 364 if (retval)
365 return retval;
366
08ce5f16
SH		 367 retval = devcgroup_inode_permission(inode, mask);
368 if (retval)
369 return retval;
370
d09ca739 371 return security_inode_permission(inode, mask);
1da177e4
LT		 372}
373
5dd784d0
JB		 374/**
375 * path_get - get a reference to a path
376 * @path: path to get the reference to
377 *
378 * Given a path increment the reference count to the dentry and the vfsmount.
379 */
380void path_get(struct path *path)
381{
382 mntget(path->mnt);
383 dget(path->dentry);
384}
385EXPORT_SYMBOL(path_get);
386
1d957f9b
JB		 387/**
388 * path_put - put a reference to a path
389 * @path: path to put the reference to
390 *
391 * Given a path decrement the reference count to the dentry and the vfsmount.
392 */
393void path_put(struct path *path)
1da177e4 394{
1d957f9b
JB		 395 dput(path->dentry);
396 mntput(path->mnt);
1da177e4 397}
1d957f9b 398EXPORT_SYMBOL(path_put);
1da177e4 399
19660af7 400/*
31e6b01f 401 * Path walking has 2 modes, rcu-walk and ref-walk (see
19660af7
AV		 402 * Documentation/filesystems/path-lookup.txt). In situations when we can't
403 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
404 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
405 * mode. Refcounts are grabbed at the last known good point before rcu-walk
406 * got stuck, so ref-walk may continue from there. If this is not successful
407 * (eg. a seqcount has changed), then failure is returned and it's up to caller
408 * to restart the path walk from the beginning in ref-walk mode.
31e6b01f 409 */
31e6b01f
NP		 410
411/**
19660af7
AV		 412 * unlazy_walk - try to switch to ref-walk mode.
413 * @nd: nameidata pathwalk data
414 * @dentry: child of nd->path.dentry or NULL
39191628 415 * Returns: 0 on success, -ECHILD on failure
31e6b01f 416 *
19660af7
AV		 417 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
418 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
419 * @nd or NULL. Must be called from rcu-walk context.
31e6b01f 420 */
19660af7 421static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
31e6b01f
NP		 422{
423 struct fs_struct *fs = current->fs;
424 struct dentry *parent = nd->path.dentry;
5b6ca027 425 int want_root = 0;
31e6b01f
NP		 426
427 BUG_ON(!(nd->flags & LOOKUP_RCU));
5b6ca027
AV		 428 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
429 want_root = 1;
31e6b01f
NP		 430 spin_lock(&fs->lock);
431 if (nd->root.mnt != fs->root.mnt ||
432 nd->root.dentry != fs->root.dentry)
433 goto err_root;
434 }
435 spin_lock(&parent->d_lock);
19660af7
AV		 436 if (!dentry) {
437 if (!__d_rcu_to_refcount(parent, nd->seq))
438 goto err_parent;
439 BUG_ON(nd->inode != parent->d_inode);
440 } else {
94c0d4ec
AV		 441 if (dentry->d_parent != parent)
442 goto err_parent;
19660af7
AV		 443 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
444 if (!__d_rcu_to_refcount(dentry, nd->seq))
445 goto err_child;
446 /*
447 * If the sequence check on the child dentry passed, then
448 * the child has not been removed from its parent. This
449 * means the parent dentry must be valid and able to take
450 * a reference at this point.
451 */
452 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
453 BUG_ON(!parent->d_count);
454 parent->d_count++;
455 spin_unlock(&dentry->d_lock);
456 }
31e6b01f 457 spin_unlock(&parent->d_lock);
5b6ca027 458 if (want_root) {
31e6b01f
NP		 459 path_get(&nd->root);
460 spin_unlock(&fs->lock);
461 }
462 mntget(nd->path.mnt);
463
464 rcu_read_unlock();
465 br_read_unlock(vfsmount_lock);
466 nd->flags &= ~LOOKUP_RCU;
467 return 0;
19660af7
AV		 468
469err_child:
31e6b01f 470 spin_unlock(&dentry->d_lock);
19660af7 471err_parent:
31e6b01f
NP		 472 spin_unlock(&parent->d_lock);
473err_root:
5b6ca027 474 if (want_root)
31e6b01f
NP		 475 spin_unlock(&fs->lock);
476 return -ECHILD;
477}
478
834f2a4a
TM		 479/**
480 * release_open_intent - free up open intent resources
481 * @nd: pointer to nameidata
482 */
483void release_open_intent(struct nameidata *nd)
484{
2dab5974
LT		 485 struct file *file = nd->intent.open.file;
486
487 if (file && !IS_ERR(file)) {
488 if (file->f_path.dentry == NULL)
489 put_filp(file);
490 else
491 fput(file);
492 }
834f2a4a
TM		 493}
494
f60aef7e 495static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
34286d66 496{
f60aef7e 497 return dentry->d_op->d_revalidate(dentry, nd);
34286d66
NP		 498}
499
9f1fafee
AV		 500/**
501 * complete_walk - successful completion of path walk
502 * @nd: pointer nameidata
39159de2 503 *
9f1fafee
AV		 504 * If we had been in RCU mode, drop out of it and legitimize nd->path.
505 * Revalidate the final result, unless we'd already done that during
506 * the path walk or the filesystem doesn't ask for it. Return 0 on
507 * success, -error on failure. In case of failure caller does not
508 * need to drop nd->path.
39159de2 509 */
9f1fafee 510static int complete_walk(struct nameidata *nd)
39159de2 511{
16c2cd71 512 struct dentry *dentry = nd->path.dentry;
39159de2 513 int status;
39159de2 514
9f1fafee
AV		 515 if (nd->flags & LOOKUP_RCU) {
516 nd->flags &= ~LOOKUP_RCU;
517 if (!(nd->flags & LOOKUP_ROOT))
518 nd->root.mnt = NULL;
519 spin_lock(&dentry->d_lock);
520 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
521 spin_unlock(&dentry->d_lock);
522 rcu_read_unlock();
523 br_read_unlock(vfsmount_lock);
524 return -ECHILD;
525 }
526 BUG_ON(nd->inode != dentry->d_inode);
527 spin_unlock(&dentry->d_lock);
528 mntget(nd->path.mnt);
529 rcu_read_unlock();
530 br_read_unlock(vfsmount_lock);
531 }
532
16c2cd71
AV		 533 if (likely(!(nd->flags & LOOKUP_JUMPED)))
534 return 0;
535
536 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
39159de2
JL		 537 return 0;
538
16c2cd71
AV		 539 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
540 return 0;
541
542 /* Note: we do not d_invalidate() */
34286d66 543 status = d_revalidate(dentry, nd);
39159de2
JL		 544 if (status > 0)
545 return 0;
546
16c2cd71 547 if (!status)
39159de2 548 status = -ESTALE;
16c2cd71 549
9f1fafee 550 path_put(&nd->path);
39159de2
JL		 551 return status;
552}
553
2a737871
AV		 554static __always_inline void set_root(struct nameidata *nd)
555{
f7ad3c6b
MS		 556 if (!nd->root.mnt)
557 get_fs_root(current->fs, &nd->root);
2a737871
AV		 558}
559
6de88d72
AV		 560static int link_path_walk(const char *, struct nameidata *);
561
31e6b01f
NP		 562static __always_inline void set_root_rcu(struct nameidata *nd)
563{
564 if (!nd->root.mnt) {
565 struct fs_struct *fs = current->fs;
c28cc364
NP		 566 unsigned seq;
567
568 do {
569 seq = read_seqcount_begin(&fs->seq);
570 nd->root = fs->root;
c1530019 571 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
c28cc364 572 } while (read_seqcount_retry(&fs->seq, seq));
31e6b01f
NP		 573 }
574}
575
f1662356 576static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
1da177e4 577{
31e6b01f
NP		 578 int ret;
579
1da177e4
LT		 580 if (IS_ERR(link))
581 goto fail;
582
583 if (*link == '/') {
2a737871 584 set_root(nd);
1d957f9b 585 path_put(&nd->path);
2a737871
AV		 586 nd->path = nd->root;
587 path_get(&nd->root);
16c2cd71 588 nd->flags |= LOOKUP_JUMPED;
1da177e4 589 }
31e6b01f 590 nd->inode = nd->path.dentry->d_inode;
b4091d5f 591
31e6b01f
NP		 592 ret = link_path_walk(link, nd);
593 return ret;
1da177e4 594fail:
1d957f9b 595 path_put(&nd->path);
1da177e4
LT		 596 return PTR_ERR(link);
597}
598
1d957f9b 599static void path_put_conditional(struct path *path, struct nameidata *nd)
051d3812
IK		 600{
601 dput(path->dentry);
4ac91378 602 if (path->mnt != nd->path.mnt)
051d3812
IK		 603 mntput(path->mnt);
604}
605
7b9337aa
NP		 606static inline void path_to_nameidata(const struct path *path,
607 struct nameidata *nd)
051d3812 608{
31e6b01f
NP		 609 if (!(nd->flags & LOOKUP_RCU)) {
610 dput(nd->path.dentry);
611 if (nd->path.mnt != path->mnt)
612 mntput(nd->path.mnt);
9a229683 613 }
31e6b01f 614 nd->path.mnt = path->mnt;
4ac91378 615 nd->path.dentry = path->dentry;
051d3812
IK		 616}
617
574197e0
AV		 618static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
619{
620 struct inode *inode = link->dentry->d_inode;
621 if (!IS_ERR(cookie) && inode->i_op->put_link)
622 inode->i_op->put_link(link->dentry, nd, cookie);
623 path_put(link);
624}
625
def4af30 626static __always_inline int
574197e0 627follow_link(struct path *link, struct nameidata *nd, void **p)
1da177e4
LT		 628{
629 int error;
7b9337aa 630 struct dentry *dentry = link->dentry;
1da177e4 631
844a3917
AV		 632 BUG_ON(nd->flags & LOOKUP_RCU);
633
0e794589
AV		 634 if (link->mnt == nd->path.mnt)
635 mntget(link->mnt);
636
574197e0
AV		 637 if (unlikely(current->total_link_count >= 40)) {
638 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
574197e0
AV		 639 path_put(&nd->path);
640 return -ELOOP;
641 }
642 cond_resched();
643 current->total_link_count++;
644
7b9337aa 645 touch_atime(link->mnt, dentry);
1da177e4 646 nd_set_link(nd, NULL);
cd4e91d3 647
36f3b4f6
AV		 648 error = security_inode_follow_link(link->dentry, nd);
649 if (error) {
650 *p = ERR_PTR(error); /* no ->put_link(), please */
651 path_put(&nd->path);
652 return error;
653 }
654
86acdca1 655 nd->last_type = LAST_BIND;
def4af30
AV		 656 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
657 error = PTR_ERR(*p);
658 if (!IS_ERR(*p)) {
1da177e4 659 char *s = nd_get_link(nd);
cc314eef 660 error = 0;
1da177e4
LT		 661 if (s)
662 error = __vfs_follow_link(nd, s);
bcda7652 663 else if (nd->last_type == LAST_BIND) {
16c2cd71 664 nd->flags |= LOOKUP_JUMPED;
b21041d0
AV		 665 nd->inode = nd->path.dentry->d_inode;
666 if (nd->inode->i_op->follow_link) {
bcda7652
AV		 667 /* stepped on a _really_ weird one */
668 path_put(&nd->path);
669 error = -ELOOP;
670 }
671 }
1da177e4 672 }
1da177e4
LT		 673 return error;
674}
675
31e6b01f
NP		 676static int follow_up_rcu(struct path *path)
677{
0714a533
AV		 678 struct mount *mnt = real_mount(path->mnt);
679 struct mount *parent;
31e6b01f
NP		 680 struct dentry *mountpoint;
681
0714a533
AV		 682 parent = mnt->mnt_parent;
683 if (&parent->mnt == path->mnt)
31e6b01f 684 return 0;
a73324da 685 mountpoint = mnt->mnt_mountpoint;
31e6b01f 686 path->dentry = mountpoint;
0714a533 687 path->mnt = &parent->mnt;
31e6b01f
NP		 688 return 1;
689}
690
bab77ebf 691int follow_up(struct path *path)
1da177e4 692{
0714a533
AV		 693 struct mount *mnt = real_mount(path->mnt);
694 struct mount *parent;
1da177e4 695 struct dentry *mountpoint;
99b7db7b
NP		 696
697 br_read_lock(vfsmount_lock);
0714a533
AV		 698 parent = mnt->mnt_parent;
699 if (&parent->mnt == path->mnt) {
99b7db7b 700 br_read_unlock(vfsmount_lock);
1da177e4
LT		 701 return 0;
702 }
0714a533 703 mntget(&parent->mnt);
a73324da 704 mountpoint = dget(mnt->mnt_mountpoint);
99b7db7b 705 br_read_unlock(vfsmount_lock);
bab77ebf
AV		 706 dput(path->dentry);
707 path->dentry = mountpoint;
708 mntput(path->mnt);
0714a533 709 path->mnt = &parent->mnt;
1da177e4
LT		 710 return 1;
711}
712
b5c84bf6 713/*
9875cf80
DH		 714 * Perform an automount
715 * - return -EISDIR to tell follow_managed() to stop and return the path we
716 * were called with.
1da177e4 717 */
9875cf80
DH		 718static int follow_automount(struct path *path, unsigned flags,
719 bool *need_mntput)
31e6b01f 720{
9875cf80 721 struct vfsmount *mnt;
ea5b778a 722 int err;
9875cf80
DH		 723
724 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
725 return -EREMOTE;
726
0ec26fd0
MS		 727 /* We don't want to mount if someone's just doing a stat -
728 * unless they're stat'ing a directory and appended a '/' to
729 * the name.
730 *
731 * We do, however, want to mount if someone wants to open or
732 * create a file of any type under the mountpoint, wants to
733 * traverse through the mountpoint or wants to open the
734 * mounted directory. Also, autofs may mark negative dentries
735 * as being automount points. These will need the attentions
736 * of the daemon to instantiate them before they can be used.
9875cf80 737 */
0ec26fd0 738 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
d94c177b 739 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
0ec26fd0
MS		 740 path->dentry->d_inode)
741 return -EISDIR;
742
9875cf80
DH		 743 current->total_link_count++;
744 if (current->total_link_count >= 40)
745 return -ELOOP;
746
747 mnt = path->dentry->d_op->d_automount(path);
748 if (IS_ERR(mnt)) {
749 /*
750 * The filesystem is allowed to return -EISDIR here to indicate
751 * it doesn't want to automount. For instance, autofs would do
752 * this so that its userspace daemon can mount on this dentry.
753 *
754 * However, we can only permit this if it's a terminal point in
755 * the path being looked up; if it wasn't then the remainder of
756 * the path is inaccessible and we should say so.
757 */
49084c3b 758 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
9875cf80
DH		 759 return -EREMOTE;
760 return PTR_ERR(mnt);
31e6b01f 761 }
ea5b778a 762
9875cf80
DH		 763 if (!mnt) /* mount collision */
764 return 0;
31e6b01f 765
8aef1884
AV		 766 if (!*need_mntput) {
767 /* lock_mount() may release path->mnt on error */
768 mntget(path->mnt);
769 *need_mntput = true;
770 }
19a167af 771 err = finish_automount(mnt, path);
9875cf80 772
ea5b778a
DH		 773 switch (err) {
774 case -EBUSY:
775 /* Someone else made a mount here whilst we were busy */
19a167af 776 return 0;
ea5b778a 777 case 0:
8aef1884 778 path_put(path);
ea5b778a
DH		 779 path->mnt = mnt;
780 path->dentry = dget(mnt->mnt_root);
ea5b778a 781 return 0;
19a167af
AV		 782 default:
783 return err;
ea5b778a 784 }
19a167af 785
463ffb2e
AV		 786}
787
9875cf80
DH		 788/*
789 * Handle a dentry that is managed in some way.
cc53ce53 790 * - Flagged for transit management (autofs)
9875cf80
DH		 791 * - Flagged as mountpoint
792 * - Flagged as automount point
793 *
794 * This may only be called in refwalk mode.
795 *
796 * Serialization is taken care of in namespace.c
797 */
798static int follow_managed(struct path *path, unsigned flags)
1da177e4 799{
8aef1884 800 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
9875cf80
DH		 801 unsigned managed;
802 bool need_mntput = false;
8aef1884 803 int ret = 0;
9875cf80
DH		 804
805 /* Given that we're not holding a lock here, we retain the value in a
806 * local variable for each dentry as we look at it so that we don't see
807 * the components of that value change under us */
808 while (managed = ACCESS_ONCE(path->dentry->d_flags),
809 managed &= DCACHE_MANAGED_DENTRY,
810 unlikely(managed != 0)) {
cc53ce53
DH		 811 /* Allow the filesystem to manage the transit without i_mutex
812 * being held. */
813 if (managed & DCACHE_MANAGE_TRANSIT) {
814 BUG_ON(!path->dentry->d_op);
815 BUG_ON(!path->dentry->d_op->d_manage);
1aed3e42 816 ret = path->dentry->d_op->d_manage(path->dentry, false);
cc53ce53 817 if (ret < 0)
8aef1884 818 break;
cc53ce53
DH		 819 }
820
9875cf80
DH		 821 /* Transit to a mounted filesystem. */
822 if (managed & DCACHE_MOUNTED) {
823 struct vfsmount *mounted = lookup_mnt(path);
824 if (mounted) {
825 dput(path->dentry);
826 if (need_mntput)
827 mntput(path->mnt);
828 path->mnt = mounted;
829 path->dentry = dget(mounted->mnt_root);
830 need_mntput = true;
831 continue;
832 }
833
834 /* Something is mounted on this dentry in another
835 * namespace and/or whatever was mounted there in this
836 * namespace got unmounted before we managed to get the
837 * vfsmount_lock */
838 }
839
840 /* Handle an automount point */
841 if (managed & DCACHE_NEED_AUTOMOUNT) {
842 ret = follow_automount(path, flags, &need_mntput);
843 if (ret < 0)
8aef1884 844 break;
9875cf80
DH		 845 continue;
846 }
847
848 /* We didn't change the current path point */
849 break;
1da177e4 850 }
8aef1884
AV		 851
852 if (need_mntput && path->mnt == mnt)
853 mntput(path->mnt);
854 if (ret == -EISDIR)
855 ret = 0;
a3fbbde7 856 return ret < 0 ? ret : need_mntput;
1da177e4
LT		 857}
858
cc53ce53 859int follow_down_one(struct path *path)
1da177e4
LT		 860{
861 struct vfsmount *mounted;
862
1c755af4 863 mounted = lookup_mnt(path);
1da177e4 864 if (mounted) {
9393bd07
AV		 865 dput(path->dentry);
866 mntput(path->mnt);
867 path->mnt = mounted;
868 path->dentry = dget(mounted->mnt_root);
1da177e4
LT		 869 return 1;
870 }
871 return 0;
872}
873
62a7375e
IK		 874static inline bool managed_dentry_might_block(struct dentry *dentry)
875{
876 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
877 dentry->d_op->d_manage(dentry, true) < 0);
878}
879
9875cf80 880/*
287548e4
AV		 881 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
882 * we meet a managed dentry that would need blocking.
9875cf80
DH		 883 */
884static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
287548e4 885 struct inode **inode)
9875cf80 886{
62a7375e 887 for (;;) {
c7105365 888 struct mount *mounted;
62a7375e
IK		 889 /*
890 * Don't forget we might have a non-mountpoint managed dentry
891 * that wants to block transit.
892 */
287548e4 893 if (unlikely(managed_dentry_might_block(path->dentry)))
ab90911f 894 return false;
62a7375e
IK		 895
896 if (!d_mountpoint(path->dentry))
897 break;
898
9875cf80
DH		 899 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
900 if (!mounted)
901 break;
c7105365
AV		 902 path->mnt = &mounted->mnt;
903 path->dentry = mounted->mnt.mnt_root;
a3fbbde7 904 nd->flags |= LOOKUP_JUMPED;
9875cf80 905 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
59430262
LT		 906 /*
907 * Update the inode too. We don't need to re-check the
908 * dentry sequence number here after this d_inode read,
909 * because a mount-point is always pinned.
910 */
911 *inode = path->dentry->d_inode;
9875cf80 912 }
9875cf80
DH		 913 return true;
914}
915
dea39376 916static void follow_mount_rcu(struct nameidata *nd)
287548e4 917{
dea39376 918 while (d_mountpoint(nd->path.dentry)) {
c7105365 919 struct mount *mounted;
dea39376 920 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
287548e4
AV		 921 if (!mounted)
922 break;
c7105365
AV		 923 nd->path.mnt = &mounted->mnt;
924 nd->path.dentry = mounted->mnt.mnt_root;
dea39376 925 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
287548e4
AV		 926 }
927}
928
31e6b01f
NP		 929static int follow_dotdot_rcu(struct nameidata *nd)
930{
31e6b01f
NP		 931 set_root_rcu(nd);
932
9875cf80 933 while (1) {
31e6b01f
NP		 934 if (nd->path.dentry == nd->root.dentry &&
935 nd->path.mnt == nd->root.mnt) {
936 break;
937 }
938 if (nd->path.dentry != nd->path.mnt->mnt_root) {
939 struct dentry *old = nd->path.dentry;
940 struct dentry *parent = old->d_parent;
941 unsigned seq;
942
943 seq = read_seqcount_begin(&parent->d_seq);
944 if (read_seqcount_retry(&old->d_seq, nd->seq))
ef7562d5 945 goto failed;
31e6b01f
NP		 946 nd->path.dentry = parent;
947 nd->seq = seq;
948 break;
949 }
950 if (!follow_up_rcu(&nd->path))
951 break;
952 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
31e6b01f 953 }
dea39376
AV		 954 follow_mount_rcu(nd);
955 nd->inode = nd->path.dentry->d_inode;
31e6b01f 956 return 0;
ef7562d5
AV		 957
958failed:
959 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 960 if (!(nd->flags & LOOKUP_ROOT))
961 nd->root.mnt = NULL;
ef7562d5
AV		 962 rcu_read_unlock();
963 br_read_unlock(vfsmount_lock);
964 return -ECHILD;
31e6b01f
NP		 965}
966
cc53ce53
DH		 967/*
968 * Follow down to the covering mount currently visible to userspace. At each
969 * point, the filesystem owning that dentry may be queried as to whether the
970 * caller is permitted to proceed or not.
cc53ce53 971 */
7cc90cc3 972int follow_down(struct path *path)
cc53ce53
DH		 973{
974 unsigned managed;
975 int ret;
976
977 while (managed = ACCESS_ONCE(path->dentry->d_flags),
978 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
979 /* Allow the filesystem to manage the transit without i_mutex
980 * being held.
981 *
982 * We indicate to the filesystem if someone is trying to mount
983 * something here. This gives autofs the chance to deny anyone
984 * other than its daemon the right to mount on its
985 * superstructure.
986 *
987 * The filesystem may sleep at this point.
988 */
989 if (managed & DCACHE_MANAGE_TRANSIT) {
990 BUG_ON(!path->dentry->d_op);
991 BUG_ON(!path->dentry->d_op->d_manage);
ab90911f 992 ret = path->dentry->d_op->d_manage(
1aed3e42 993 path->dentry, false);
cc53ce53
DH		 994 if (ret < 0)
995 return ret == -EISDIR ? 0 : ret;
996 }
997
998 /* Transit to a mounted filesystem. */
999 if (managed & DCACHE_MOUNTED) {
1000 struct vfsmount *mounted = lookup_mnt(path);
1001 if (!mounted)
1002 break;
1003 dput(path->dentry);
1004 mntput(path->mnt);
1005 path->mnt = mounted;
1006 path->dentry = dget(mounted->mnt_root);
1007 continue;
1008 }
1009
1010 /* Don't handle automount points here */
1011 break;
1012 }
1013 return 0;
1014}
1015
9875cf80
DH		 1016/*
1017 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1018 */
1019static void follow_mount(struct path *path)
1020{
1021 while (d_mountpoint(path->dentry)) {
1022 struct vfsmount *mounted = lookup_mnt(path);
1023 if (!mounted)
1024 break;
1025 dput(path->dentry);
1026 mntput(path->mnt);
1027 path->mnt = mounted;
1028 path->dentry = dget(mounted->mnt_root);
1029 }
1030}
1031
31e6b01f 1032static void follow_dotdot(struct nameidata *nd)
1da177e4 1033{
2a737871 1034 set_root(nd);
e518ddb7 1035
1da177e4 1036 while(1) {
4ac91378 1037 struct dentry *old = nd->path.dentry;
1da177e4 1038
2a737871
AV		 1039 if (nd->path.dentry == nd->root.dentry &&
1040 nd->path.mnt == nd->root.mnt) {
1da177e4
LT		 1041 break;
1042 }
4ac91378 1043 if (nd->path.dentry != nd->path.mnt->mnt_root) {
3088dd70
AV		 1044 /* rare case of legitimate dget_parent()... */
1045 nd->path.dentry = dget_parent(nd->path.dentry);
1da177e4
LT		 1046 dput(old);
1047 break;
1048 }
3088dd70 1049 if (!follow_up(&nd->path))
1da177e4 1050 break;
1da177e4 1051 }
79ed0226 1052 follow_mount(&nd->path);
31e6b01f 1053 nd->inode = nd->path.dentry->d_inode;
1da177e4
LT		 1054}
1055
baa03890
NP		 1056/*
1057 * Allocate a dentry with name and parent, and perform a parent
1058 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1059 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1060 * have verified that no child exists while under i_mutex.
1061 */
1062static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1063 struct qstr *name, struct nameidata *nd)
1064{
1065 struct inode *inode = parent->d_inode;
1066 struct dentry *dentry;
1067 struct dentry *old;
1068
1069 /* Don't create child dentry for a dead directory. */
1070 if (unlikely(IS_DEADDIR(inode)))
1071 return ERR_PTR(-ENOENT);
1072
1073 dentry = d_alloc(parent, name);
1074 if (unlikely(!dentry))
1075 return ERR_PTR(-ENOMEM);
1076
1077 old = inode->i_op->lookup(inode, dentry, nd);
1078 if (unlikely(old)) {
1079 dput(dentry);
1080 dentry = old;
1081 }
1082 return dentry;
1083}
1084
44396f4b
JB		 1085/*
1086 * We already have a dentry, but require a lookup to be performed on the parent
1087 * directory to fill in d_inode. Returns the new dentry, or ERR_PTR on error.
1088 * parent->d_inode->i_mutex must be held. d_lookup must have verified that no
1089 * child exists while under i_mutex.
1090 */
1091static struct dentry *d_inode_lookup(struct dentry *parent, struct dentry *dentry,
1092 struct nameidata *nd)
1093{
1094 struct inode *inode = parent->d_inode;
1095 struct dentry *old;
1096
1097 /* Don't create child dentry for a dead directory. */
e188dc02
MS		 1098 if (unlikely(IS_DEADDIR(inode))) {
1099 dput(dentry);
44396f4b 1100 return ERR_PTR(-ENOENT);
e188dc02 1101 }
44396f4b
JB		 1102
1103 old = inode->i_op->lookup(inode, dentry, nd);
1104 if (unlikely(old)) {
1105 dput(dentry);
1106 dentry = old;
1107 }
1108 return dentry;
1109}
1110
1da177e4
LT		 1111/*
1112 * It's more convoluted than I'd like it to be, but... it's still fairly
1113 * small and for now I'd prefer to have fast path as straight as possible.
1114 * It _is_ time-critical.
1115 */
1116static int do_lookup(struct nameidata *nd, struct qstr *name,
31e6b01f 1117 struct path *path, struct inode **inode)
1da177e4 1118{
4ac91378 1119 struct vfsmount *mnt = nd->path.mnt;
31e6b01f 1120 struct dentry *dentry, *parent = nd->path.dentry;
5a18fff2
AV		 1121 int need_reval = 1;
1122 int status = 1;
9875cf80
DH		 1123 int err;
1124
b04f784e
NP		 1125 /*
1126 * Rename seqlock is not required here because in the off chance
1127 * of a false negative due to a concurrent rename, we're going to
1128 * do the non-racy lookup, below.
1129 */
31e6b01f
NP		 1130 if (nd->flags & LOOKUP_RCU) {
1131 unsigned seq;
31e6b01f
NP		 1132 *inode = nd->inode;
1133 dentry = __d_lookup_rcu(parent, name, &seq, inode);
5a18fff2
AV		 1134 if (!dentry)
1135 goto unlazy;
1136
31e6b01f
NP		 1137 /* Memory barrier in read_seqcount_begin of child is enough */
1138 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1139 return -ECHILD;
31e6b01f 1140 nd->seq = seq;
5a18fff2 1141
24643087 1142 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
5a18fff2
AV		 1143 status = d_revalidate(dentry, nd);
1144 if (unlikely(status <= 0)) {
1145 if (status != -ECHILD)
1146 need_reval = 0;
1147 goto unlazy;
1148 }
24643087 1149 }
44396f4b
JB		 1150 if (unlikely(d_need_lookup(dentry)))
1151 goto unlazy;
31e6b01f
NP		 1152 path->mnt = mnt;
1153 path->dentry = dentry;
d6e9bd25
AV		 1154 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1155 goto unlazy;
1156 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1157 goto unlazy;
1158 return 0;
5a18fff2 1159unlazy:
19660af7
AV		 1160 if (unlazy_walk(nd, dentry))
1161 return -ECHILD;
5a18fff2
AV		 1162 } else {
1163 dentry = __d_lookup(parent, name);
9875cf80 1164 }
5a18fff2 1165
44396f4b
JB		 1166 if (dentry && unlikely(d_need_lookup(dentry))) {
1167 dput(dentry);
1168 dentry = NULL;
1169 }
5a18fff2
AV		 1170retry:
1171 if (unlikely(!dentry)) {
1172 struct inode *dir = parent->d_inode;
1173 BUG_ON(nd->inode != dir);
1174
1175 mutex_lock(&dir->i_mutex);
1176 dentry = d_lookup(parent, name);
1177 if (likely(!dentry)) {
1178 dentry = d_alloc_and_lookup(parent, name, nd);
1179 if (IS_ERR(dentry)) {
1180 mutex_unlock(&dir->i_mutex);
1181 return PTR_ERR(dentry);
1182 }
1183 /* known good */
1184 need_reval = 0;
1185 status = 1;
44396f4b
JB		 1186 } else if (unlikely(d_need_lookup(dentry))) {
1187 dentry = d_inode_lookup(parent, dentry, nd);
1188 if (IS_ERR(dentry)) {
1189 mutex_unlock(&dir->i_mutex);
1190 return PTR_ERR(dentry);
1191 }
1192 /* known good */
1193 need_reval = 0;
1194 status = 1;
5a18fff2
AV		 1195 }
1196 mutex_unlock(&dir->i_mutex);
1197 }
1198 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1199 status = d_revalidate(dentry, nd);
1200 if (unlikely(status <= 0)) {
1201 if (status < 0) {
1202 dput(dentry);
1203 return status;
1204 }
1205 if (!d_invalidate(dentry)) {
1206 dput(dentry);
1207 dentry = NULL;
1208 need_reval = 1;
1209 goto retry;
1210 }
24643087 1211 }
5a18fff2 1212
9875cf80
DH		 1213 path->mnt = mnt;
1214 path->dentry = dentry;
1215 err = follow_managed(path, nd->flags);
89312214
IK		 1216 if (unlikely(err < 0)) {
1217 path_put_conditional(path, nd);
9875cf80 1218 return err;
89312214 1219 }
a3fbbde7
AV		 1220 if (err)
1221 nd->flags |= LOOKUP_JUMPED;
9875cf80 1222 *inode = path->dentry->d_inode;
1da177e4 1223 return 0;
1da177e4
LT		 1224}
1225
52094c8a
AV		 1226static inline int may_lookup(struct nameidata *nd)
1227{
1228 if (nd->flags & LOOKUP_RCU) {
4ad5abb3 1229 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
52094c8a
AV		 1230 if (err != -ECHILD)
1231 return err;
19660af7 1232 if (unlazy_walk(nd, NULL))
52094c8a
AV		 1233 return -ECHILD;
1234 }
4ad5abb3 1235 return inode_permission(nd->inode, MAY_EXEC);
52094c8a
AV		 1236}
1237
9856fa1b
AV		 1238static inline int handle_dots(struct nameidata *nd, int type)
1239{
1240 if (type == LAST_DOTDOT) {
1241 if (nd->flags & LOOKUP_RCU) {
1242 if (follow_dotdot_rcu(nd))
1243 return -ECHILD;
1244 } else
1245 follow_dotdot(nd);
1246 }
1247 return 0;
1248}
1249
951361f9
AV		 1250static void terminate_walk(struct nameidata *nd)
1251{
1252 if (!(nd->flags & LOOKUP_RCU)) {
1253 path_put(&nd->path);
1254 } else {
1255 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 1256 if (!(nd->flags & LOOKUP_ROOT))
1257 nd->root.mnt = NULL;
951361f9
AV		 1258 rcu_read_unlock();
1259 br_read_unlock(vfsmount_lock);
1260 }
1261}
1262
3ddcd056
LT		 1263/*
1264 * Do we need to follow links? We _really_ want to be able
1265 * to do this check without having to look at inode->i_op,
1266 * so we keep a cache of "no, this doesn't need follow_link"
1267 * for the common case.
1268 */
7813b94a 1269static inline int should_follow_link(struct inode *inode, int follow)
3ddcd056
LT		 1270{
1271 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1272 if (likely(inode->i_op->follow_link))
1273 return follow;
1274
1275 /* This gets set once for the inode lifetime */
1276 spin_lock(&inode->i_lock);
1277 inode->i_opflags |= IOP_NOFOLLOW;
1278 spin_unlock(&inode->i_lock);
1279 }
1280 return 0;
1281}
1282
ce57dfc1
AV		 1283static inline int walk_component(struct nameidata *nd, struct path *path,
1284 struct qstr *name, int type, int follow)
1285{
1286 struct inode *inode;
1287 int err;
1288 /*
1289 * "." and ".." are special - ".." especially so because it has
1290 * to be able to know about the current root directory and
1291 * parent relationships.
1292 */
1293 if (unlikely(type != LAST_NORM))
1294 return handle_dots(nd, type);
1295 err = do_lookup(nd, name, path, &inode);
1296 if (unlikely(err)) {
1297 terminate_walk(nd);
1298 return err;
1299 }
1300 if (!inode) {
1301 path_to_nameidata(path, nd);
1302 terminate_walk(nd);
1303 return -ENOENT;
1304 }
7813b94a 1305 if (should_follow_link(inode, follow)) {
19660af7
AV		 1306 if (nd->flags & LOOKUP_RCU) {
1307 if (unlikely(unlazy_walk(nd, path->dentry))) {
1308 terminate_walk(nd);
1309 return -ECHILD;
1310 }
1311 }
ce57dfc1
AV		 1312 BUG_ON(inode != path->dentry->d_inode);
1313 return 1;
1314 }
1315 path_to_nameidata(path, nd);
1316 nd->inode = inode;
1317 return 0;
1318}
1319
b356379a
AV		 1320/*
1321 * This limits recursive symlink follows to 8, while
1322 * limiting consecutive symlinks to 40.
1323 *
1324 * Without that kind of total limit, nasty chains of consecutive
1325 * symlinks can cause almost arbitrarily long lookups.
1326 */
1327static inline int nested_symlink(struct path *path, struct nameidata *nd)
1328{
1329 int res;
1330
b356379a
AV		 1331 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1332 path_put_conditional(path, nd);
1333 path_put(&nd->path);
1334 return -ELOOP;
1335 }
1a4022f8 1336 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
b356379a
AV		 1337
1338 nd->depth++;
1339 current->link_count++;
1340
1341 do {
1342 struct path link = *path;
1343 void *cookie;
574197e0
AV		 1344
1345 res = follow_link(&link, nd, &cookie);
b356379a
AV		 1346 if (!res)
1347 res = walk_component(nd, path, &nd->last,
1348 nd->last_type, LOOKUP_FOLLOW);
574197e0 1349 put_link(nd, &link, cookie);
b356379a
AV		 1350 } while (res > 0);
1351
1352 current->link_count--;
1353 nd->depth--;
1354 return res;
1355}
1356
3ddcd056
LT		 1357/*
1358 * We really don't want to look at inode->i_op->lookup
1359 * when we don't have to. So we keep a cache bit in
1360 * the inode ->i_opflags field that says "yes, we can
1361 * do lookup on this inode".
1362 */
1363static inline int can_lookup(struct inode *inode)
1364{
1365 if (likely(inode->i_opflags & IOP_LOOKUP))
1366 return 1;
1367 if (likely(!inode->i_op->lookup))
1368 return 0;
1369
1370 /* We do this once for the lifetime of the inode */
1371 spin_lock(&inode->i_lock);
1372 inode->i_opflags |= IOP_LOOKUP;
1373 spin_unlock(&inode->i_lock);
1374 return 1;
1375}
1376
0145acc2
LT		 1377unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1378{
1379 unsigned long hash = init_name_hash();
1380 while (len--)
1381 hash = partial_name_hash(*name++, hash);
1382 return end_name_hash(hash);
1383}
ae942ae7 1384EXPORT_SYMBOL(full_name_hash);
0145acc2 1385
200e9ef7
LT		 1386/*
1387 * We know there's a real path component here of at least
1388 * one character.
1389 */
1390static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1391{
1392 unsigned long hash = init_name_hash();
1393 unsigned long len = 0, c;
1394
1395 c = (unsigned char)*name;
1396 do {
1397 len++;
1398 hash = partial_name_hash(c, hash);
1399 c = (unsigned char)name[len];
1400 } while (c && c != '/');
1401 *hashp = end_name_hash(hash);
1402 return len;
1403}
1404
1da177e4
LT		 1405/*
1406 * Name resolution.
ea3834d9
PM		 1407 * This is the basic name resolution function, turning a pathname into
1408 * the final dentry. We expect 'base' to be positive and a directory.
1da177e4 1409 *
ea3834d9
PM		 1410 * Returns 0 and nd will have valid dentry and mnt on success.
1411 * Returns error and drops reference to input namei data on failure.
1da177e4 1412 */
6de88d72 1413static int link_path_walk(const char *name, struct nameidata *nd)
1da177e4
LT		 1414{
1415 struct path next;
1da177e4 1416 int err;
1da177e4
LT		 1417
1418 while (*name=='/')
1419 name++;
1420 if (!*name)
086e183a 1421 return 0;
1da177e4 1422
1da177e4
LT		 1423 /* At this point we know we have a real path component. */
1424 for(;;) {
1da177e4 1425 struct qstr this;
200e9ef7 1426 long len;
fe479a58 1427 int type;
1da177e4 1428
52094c8a 1429 err = may_lookup(nd);
1da177e4
LT		 1430 if (err)
1431 break;
1432
200e9ef7 1433 len = hash_name(name, &this.hash);
1da177e4 1434 this.name = name;
200e9ef7 1435 this.len = len;
1da177e4 1436
fe479a58 1437 type = LAST_NORM;
200e9ef7 1438 if (name[0] == '.') switch (len) {
fe479a58 1439 case 2:
200e9ef7 1440 if (name[1] == '.') {
fe479a58 1441 type = LAST_DOTDOT;
16c2cd71
AV		 1442 nd->flags |= LOOKUP_JUMPED;
1443 }
fe479a58
AV		 1444 break;
1445 case 1:
1446 type = LAST_DOT;
1447 }
5a202bcd
AV		 1448 if (likely(type == LAST_NORM)) {
1449 struct dentry *parent = nd->path.dentry;
16c2cd71 1450 nd->flags &= ~LOOKUP_JUMPED;
5a202bcd
AV		 1451 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1452 err = parent->d_op->d_hash(parent, nd->inode,
1453 &this);
1454 if (err < 0)
1455 break;
1456 }
1457 }
fe479a58 1458
200e9ef7 1459 if (!name[len])
1da177e4 1460 goto last_component;
200e9ef7
LT		 1461 /*
1462 * If it wasn't NUL, we know it was '/'. Skip that
1463 * slash, and continue until no more slashes.
1464 */
1465 do {
1466 len++;
1467 } while (unlikely(name[len] == '/'));
1468 if (!name[len])
b356379a 1469 goto last_component;
200e9ef7 1470 name += len;
1da177e4 1471
ce57dfc1
AV		 1472 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1473 if (err < 0)
1474 return err;
1da177e4 1475
ce57dfc1 1476 if (err) {
b356379a 1477 err = nested_symlink(&next, nd);
1da177e4 1478 if (err)
a7472bab 1479 return err;
31e6b01f 1480 }
3ddcd056
LT		 1481 if (can_lookup(nd->inode))
1482 continue;
1da177e4 1483 err = -ENOTDIR;
3ddcd056 1484 break;
1da177e4
LT		 1485 /* here ends the main loop */
1486
1da177e4 1487last_component:
b356379a
AV		 1488 nd->last = this;
1489 nd->last_type = type;
086e183a 1490 return 0;
1da177e4 1491 }
951361f9 1492 terminate_walk(nd);
1da177e4
LT		 1493 return err;
1494}
1495
70e9b357
AV		 1496static int path_init(int dfd, const char *name, unsigned int flags,
1497 struct nameidata *nd, struct file **fp)
31e6b01f
NP		 1498{
1499 int retval = 0;
1500 int fput_needed;
1501 struct file *file;
1502
1503 nd->last_type = LAST_ROOT; /* if there are only slashes... */
16c2cd71 1504 nd->flags = flags | LOOKUP_JUMPED;
31e6b01f 1505 nd->depth = 0;
5b6ca027
AV		 1506 if (flags & LOOKUP_ROOT) {
1507 struct inode *inode = nd->root.dentry->d_inode;
73d049a4
AV		 1508 if (*name) {
1509 if (!inode->i_op->lookup)
1510 return -ENOTDIR;
1511 retval = inode_permission(inode, MAY_EXEC);
1512 if (retval)
1513 return retval;
1514 }
5b6ca027
AV		 1515 nd->path = nd->root;
1516 nd->inode = inode;
1517 if (flags & LOOKUP_RCU) {
1518 br_read_lock(vfsmount_lock);
1519 rcu_read_lock();
1520 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1521 } else {
1522 path_get(&nd->path);
1523 }
1524 return 0;
1525 }
1526
31e6b01f 1527 nd->root.mnt = NULL;
31e6b01f
NP		 1528
1529 if (*name=='/') {
e41f7d4e
AV		 1530 if (flags & LOOKUP_RCU) {
1531 br_read_lock(vfsmount_lock);
1532 rcu_read_lock();
1533 set_root_rcu(nd);
1534 } else {
1535 set_root(nd);
1536 path_get(&nd->root);
1537 }
1538 nd->path = nd->root;
31e6b01f 1539 } else if (dfd == AT_FDCWD) {
e41f7d4e
AV		 1540 if (flags & LOOKUP_RCU) {
1541 struct fs_struct *fs = current->fs;
1542 unsigned seq;
31e6b01f 1543
e41f7d4e
AV		 1544 br_read_lock(vfsmount_lock);
1545 rcu_read_lock();
c28cc364 1546
e41f7d4e
AV		 1547 do {
1548 seq = read_seqcount_begin(&fs->seq);
1549 nd->path = fs->pwd;
1550 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1551 } while (read_seqcount_retry(&fs->seq, seq));
1552 } else {
1553 get_fs_pwd(current->fs, &nd->path);
1554 }
31e6b01f
NP		 1555 } else {
1556 struct dentry *dentry;
1557
1abf0c71 1558 file = fget_raw_light(dfd, &fput_needed);
31e6b01f
NP		 1559 retval = -EBADF;
1560 if (!file)
1561 goto out_fail;
1562
1563 dentry = file->f_path.dentry;
1564
f52e0c11
AV		 1565 if (*name) {
1566 retval = -ENOTDIR;
1567 if (!S_ISDIR(dentry->d_inode->i_mode))
1568 goto fput_fail;
31e6b01f 1569
4ad5abb3 1570 retval = inode_permission(dentry->d_inode, MAY_EXEC);
f52e0c11
AV		 1571 if (retval)
1572 goto fput_fail;
1573 }
31e6b01f
NP		 1574
1575 nd->path = file->f_path;
e41f7d4e
AV		 1576 if (flags & LOOKUP_RCU) {
1577 if (fput_needed)
70e9b357 1578 *fp = file;
e41f7d4e
AV		 1579 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1580 br_read_lock(vfsmount_lock);
1581 rcu_read_lock();
1582 } else {
1583 path_get(&file->f_path);
1584 fput_light(file, fput_needed);
1585 }
31e6b01f 1586 }
31e6b01f 1587
31e6b01f 1588 nd->inode = nd->path.dentry->d_inode;
9b4a9b14 1589 return 0;
2dfdd266 1590
9b4a9b14
AV		 1591fput_fail:
1592 fput_light(file, fput_needed);
1593out_fail:
1594 return retval;
1595}
1596
bd92d7fe
AV		 1597static inline int lookup_last(struct nameidata *nd, struct path *path)
1598{
1599 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1600 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1601
1602 nd->flags &= ~LOOKUP_PARENT;
1603 return walk_component(nd, path, &nd->last, nd->last_type,
1604 nd->flags & LOOKUP_FOLLOW);
1605}
1606
9b4a9b14 1607/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
ee0827cd 1608static int path_lookupat(int dfd, const char *name,
9b4a9b14
AV		 1609 unsigned int flags, struct nameidata *nd)
1610{
70e9b357 1611 struct file *base = NULL;
bd92d7fe
AV		 1612 struct path path;
1613 int err;
31e6b01f
NP		 1614
1615 /*
1616 * Path walking is largely split up into 2 different synchronisation
1617 * schemes, rcu-walk and ref-walk (explained in
1618 * Documentation/filesystems/path-lookup.txt). These share much of the
1619 * path walk code, but some things particularly setup, cleanup, and
1620 * following mounts are sufficiently divergent that functions are
1621 * duplicated. Typically there is a function foo(), and its RCU
1622 * analogue, foo_rcu().
1623 *
1624 * -ECHILD is the error number of choice (just to avoid clashes) that
1625 * is returned if some aspect of an rcu-walk fails. Such an error must
1626 * be handled by restarting a traditional ref-walk (which will always
1627 * be able to complete).
1628 */
bd92d7fe 1629 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
ee0827cd 1630
bd92d7fe
AV		 1631 if (unlikely(err))
1632 return err;
ee0827cd
AV		 1633
1634 current->total_link_count = 0;
bd92d7fe
AV		 1635 err = link_path_walk(name, nd);
1636
1637 if (!err && !(flags & LOOKUP_PARENT)) {
bd92d7fe
AV		 1638 err = lookup_last(nd, &path);
1639 while (err > 0) {
1640 void *cookie;
1641 struct path link = path;
bd92d7fe 1642 nd->flags |= LOOKUP_PARENT;
574197e0 1643 err = follow_link(&link, nd, &cookie);
bd92d7fe
AV		 1644 if (!err)
1645 err = lookup_last(nd, &path);
574197e0 1646 put_link(nd, &link, cookie);
bd92d7fe
AV		 1647 }
1648 }
ee0827cd 1649
9f1fafee
AV		 1650 if (!err)
1651 err = complete_walk(nd);
bd92d7fe
AV		 1652
1653 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1654 if (!nd->inode->i_op->lookup) {
1655 path_put(&nd->path);
bd23a539 1656 err = -ENOTDIR;
bd92d7fe
AV		 1657 }
1658 }
16c2cd71 1659
70e9b357
AV		 1660 if (base)
1661 fput(base);
ee0827cd 1662
5b6ca027 1663 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
2a737871
AV		 1664 path_put(&nd->root);
1665 nd->root.mnt = NULL;
1666 }
bd92d7fe 1667 return err;
ee0827cd 1668}
31e6b01f 1669
ee0827cd
AV		 1670static int do_path_lookup(int dfd, const char *name,
1671 unsigned int flags, struct nameidata *nd)
1672{
1673 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1674 if (unlikely(retval == -ECHILD))
1675 retval = path_lookupat(dfd, name, flags, nd);
1676 if (unlikely(retval == -ESTALE))
1677 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
31e6b01f
NP		 1678
1679 if (likely(!retval)) {
1680 if (unlikely(!audit_dummy_context())) {
1681 if (nd->path.dentry && nd->inode)
1682 audit_inode(name, nd->path.dentry);
1683 }
1684 }
170aa3d0 1685 return retval;
1da177e4
LT		 1686}
1687
c9c6cac0 1688int kern_path_parent(const char *name, struct nameidata *nd)
5590ff0d 1689{
c9c6cac0 1690 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
5590ff0d
UD		 1691}
1692
d1811465
AV		 1693int kern_path(const char *name, unsigned int flags, struct path *path)
1694{
1695 struct nameidata nd;
1696 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1697 if (!res)
1698 *path = nd.path;
1699 return res;
1700}
1701
16f18200
JJS		 1702/**
1703 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1704 * @dentry: pointer to dentry of the base directory
1705 * @mnt: pointer to vfs mount of the base directory
1706 * @name: pointer to file name
1707 * @flags: lookup flags
e0a01249 1708 * @path: pointer to struct path to fill
16f18200
JJS		 1709 */
1710int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1711 const char *name, unsigned int flags,
e0a01249 1712 struct path *path)
16f18200 1713{
e0a01249
AV		 1714 struct nameidata nd;
1715 int err;
1716 nd.root.dentry = dentry;
1717 nd.root.mnt = mnt;
1718 BUG_ON(flags & LOOKUP_PARENT);
5b6ca027 1719 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
e0a01249
AV		 1720 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1721 if (!err)
1722 *path = nd.path;
1723 return err;
16f18200
JJS		 1724}
1725
eead1911
CH		 1726static struct dentry *__lookup_hash(struct qstr *name,
1727 struct dentry *base, struct nameidata *nd)
1da177e4 1728{
81fca444 1729 struct inode *inode = base->d_inode;
057f6c01 1730 struct dentry *dentry;
1da177e4
LT		 1731 int err;
1732
4ad5abb3 1733 err = inode_permission(inode, MAY_EXEC);
81fca444
CH		 1734 if (err)
1735 return ERR_PTR(err);
1da177e4 1736
b04f784e
NP		 1737 /*
1738 * Don't bother with __d_lookup: callers are for creat as
1739 * well as unlink, so a lot of the time it would cost
1740 * a double lookup.
6e6b1bd1 1741 */
b04f784e 1742 dentry = d_lookup(base, name);
6e6b1bd1 1743
44396f4b
JB		 1744 if (dentry && d_need_lookup(dentry)) {
1745 /*
1746 * __lookup_hash is called with the parent dir's i_mutex already
1747 * held, so we are good to go here.
1748 */
1749 dentry = d_inode_lookup(base, dentry, nd);
1750 if (IS_ERR(dentry))
1751 return dentry;
1752 }
1753
d2d9e9fb
AV		 1754 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1755 int status = d_revalidate(dentry, nd);
1756 if (unlikely(status <= 0)) {
1757 /*
1758 * The dentry failed validation.
1759 * If d_revalidate returned 0 attempt to invalidate
1760 * the dentry otherwise d_revalidate is asking us
1761 * to return a fail status.
1762 */
1763 if (status < 0) {
1764 dput(dentry);
1765 return ERR_PTR(status);
1766 } else if (!d_invalidate(dentry)) {
1767 dput(dentry);
1768 dentry = NULL;
1769 }
1770 }
1771 }
6e6b1bd1 1772
baa03890
NP		 1773 if (!dentry)
1774 dentry = d_alloc_and_lookup(base, name, nd);
5a202bcd 1775
1da177e4
LT		 1776 return dentry;
1777}
1778
057f6c01
JM		 1779/*
1780 * Restricted form of lookup. Doesn't follow links, single-component only,
1781 * needs parent already locked. Doesn't follow mounts.
1782 * SMP-safe.
1783 */
eead1911 1784static struct dentry *lookup_hash(struct nameidata *nd)
057f6c01 1785{
4ac91378 1786 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1da177e4
LT		 1787}
1788
eead1911 1789/**
a6b91919 1790 * lookup_one_len - filesystem helper to lookup single pathname component
eead1911
CH		 1791 * @name: pathname component to lookup
1792 * @base: base directory to lookup from
1793 * @len: maximum length @len should be interpreted to
1794 *
a6b91919
RD		 1795 * Note that this routine is purely a helper for filesystem usage and should
1796 * not be called by generic code. Also note that by using this function the
eead1911
CH		 1797 * nameidata argument is passed to the filesystem methods and a filesystem
1798 * using this helper needs to be prepared for that.
1799 */
057f6c01
JM		 1800struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1801{
057f6c01 1802 struct qstr this;
6a96ba54 1803 unsigned int c;
057f6c01 1804
2f9092e1
DW		 1805 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1806
6a96ba54
AV		 1807 this.name = name;
1808 this.len = len;
0145acc2 1809 this.hash = full_name_hash(name, len);
6a96ba54
AV		 1810 if (!len)
1811 return ERR_PTR(-EACCES);
1812
6a96ba54
AV		 1813 while (len--) {
1814 c = *(const unsigned char *)name++;
1815 if (c == '/' || c == '\0')
1816 return ERR_PTR(-EACCES);
6a96ba54 1817 }
5a202bcd
AV		 1818 /*
1819 * See if the low-level filesystem might want
1820 * to use its own hash..
1821 */
1822 if (base->d_flags & DCACHE_OP_HASH) {
1823 int err = base->d_op->d_hash(base, base->d_inode, &this);
1824 if (err < 0)
1825 return ERR_PTR(err);
1826 }
eead1911 1827
49705b77 1828 return __lookup_hash(&this, base, NULL);
057f6c01
JM		 1829}
1830
1fa1e7f6
AW		 1831int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
1832 struct path *path, int *empty)
1da177e4 1833{
2d8f3038 1834 struct nameidata nd;
1fa1e7f6 1835 char *tmp = getname_flags(name, flags, empty);
1da177e4 1836 int err = PTR_ERR(tmp);
1da177e4 1837 if (!IS_ERR(tmp)) {
2d8f3038
AV		 1838
1839 BUG_ON(flags & LOOKUP_PARENT);
1840
1841 err = do_path_lookup(dfd, tmp, flags, &nd);
1da177e4 1842 putname(tmp);
2d8f3038
AV		 1843 if (!err)
1844 *path = nd.path;
1da177e4
LT		 1845 }
1846 return err;
1847}
1848
1fa1e7f6
AW		 1849int user_path_at(int dfd, const char __user *name, unsigned flags,
1850 struct path *path)
1851{
1852 return user_path_at_empty(dfd, name, flags, path, 0);
1853}
1854
2ad94ae6
AV		 1855static int user_path_parent(int dfd, const char __user *path,
1856 struct nameidata *nd, char **name)
1857{
1858 char *s = getname(path);
1859 int error;
1860
1861 if (IS_ERR(s))
1862 return PTR_ERR(s);
1863
1864 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1865 if (error)
1866 putname(s);
1867 else
1868 *name = s;
1869
1870 return error;
1871}
1872
1da177e4
LT		 1873/*
1874 * It's inline, so penalty for filesystems that don't use sticky bit is
1875 * minimal.
1876 */
1877static inline int check_sticky(struct inode *dir, struct inode *inode)
1878{
da9592ed
DH		 1879 uid_t fsuid = current_fsuid();
1880
1da177e4
LT		 1881 if (!(dir->i_mode & S_ISVTX))
1882 return 0;
e795b717
SH		 1883 if (current_user_ns() != inode_userns(inode))
1884 goto other_userns;
da9592ed 1885 if (inode->i_uid == fsuid)
1da177e4 1886 return 0;
da9592ed 1887 if (dir->i_uid == fsuid)
1da177e4 1888 return 0;
e795b717
SH		 1889
1890other_userns:
1891 return !ns_capable(inode_userns(inode), CAP_FOWNER);
1da177e4
LT		 1892}
1893
1894/*
1895 * Check whether we can remove a link victim from directory dir, check
1896 * whether the type of victim is right.
1897 * 1. We can't do it if dir is read-only (done in permission())
1898 * 2. We should have write and exec permissions on dir
1899 * 3. We can't remove anything from append-only dir
1900 * 4. We can't do anything with immutable dir (done in permission())
1901 * 5. If the sticky bit on dir is set we should either
1902 * a. be owner of dir, or
1903 * b. be owner of victim, or
1904 * c. have CAP_FOWNER capability
1905 * 6. If the victim is append-only or immutable we can't do antyhing with
1906 * links pointing to it.
1907 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1908 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1909 * 9. We can't remove a root or mountpoint.
1910 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1911 * nfs_async_unlink().
1912 */
858119e1 1913static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1da177e4
LT		 1914{
1915 int error;
1916
1917 if (!victim->d_inode)
1918 return -ENOENT;
1919
1920 BUG_ON(victim->d_parent->d_inode != dir);
cccc6bba 1921 audit_inode_child(victim, dir);
1da177e4 1922
f419a2e3 1923 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1924 if (error)
1925 return error;
1926 if (IS_APPEND(dir))
1927 return -EPERM;
1928 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
f9454548 1929 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1da177e4
LT		 1930 return -EPERM;
1931 if (isdir) {
1932 if (!S_ISDIR(victim->d_inode->i_mode))
1933 return -ENOTDIR;
1934 if (IS_ROOT(victim))
1935 return -EBUSY;
1936 } else if (S_ISDIR(victim->d_inode->i_mode))
1937 return -EISDIR;
1938 if (IS_DEADDIR(dir))
1939 return -ENOENT;
1940 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1941 return -EBUSY;
1942 return 0;
1943}
1944
1945/* Check whether we can create an object with dentry child in directory
1946 * dir.
1947 * 1. We can't do it if child already exists (open has special treatment for
1948 * this case, but since we are inlined it's OK)
1949 * 2. We can't do it if dir is read-only (done in permission())
1950 * 3. We should have write and exec permissions on dir
1951 * 4. We can't do it if dir is immutable (done in permission())
1952 */
a95164d9 1953static inline int may_create(struct inode *dir, struct dentry *child)
1da177e4
LT		 1954{
1955 if (child->d_inode)
1956 return -EEXIST;
1957 if (IS_DEADDIR(dir))
1958 return -ENOENT;
f419a2e3 1959 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1960}
1961
1da177e4
LT		 1962/*
1963 * p1 and p2 should be directories on the same fs.
1964 */
1965struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1966{
1967 struct dentry *p;
1968
1969 if (p1 == p2) {
f2eace23 1970 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1da177e4
LT		 1971 return NULL;
1972 }
1973
a11f3a05 1974 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4 1975
e2761a11
OH		 1976 p = d_ancestor(p2, p1);
1977 if (p) {
1978 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1979 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1980 return p;
1da177e4
LT		 1981 }
1982
e2761a11
OH		 1983 p = d_ancestor(p1, p2);
1984 if (p) {
1985 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1986 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1987 return p;
1da177e4
LT		 1988 }
1989
f2eace23
IM		 1990 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1991 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1da177e4
LT		 1992 return NULL;
1993}
1994
1995void unlock_rename(struct dentry *p1, struct dentry *p2)
1996{
1b1dcc1b 1997 mutex_unlock(&p1->d_inode->i_mutex);
1da177e4 1998 if (p1 != p2) {
1b1dcc1b 1999 mutex_unlock(&p2->d_inode->i_mutex);
a11f3a05 2000 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4
LT		 2001 }
2002}
2003
4acdaf27 2004int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
1da177e4
LT		 2005 struct nameidata *nd)
2006{
a95164d9 2007 int error = may_create(dir, dentry);
1da177e4
LT		 2008
2009 if (error)
2010 return error;
2011
acfa4380 2012 if (!dir->i_op->create)
1da177e4
LT		 2013 return -EACCES; /* shouldn't it be ENOSYS? */
2014 mode &= S_IALLUGO;
2015 mode |= S_IFREG;
2016 error = security_inode_create(dir, dentry, mode);
2017 if (error)
2018 return error;
1da177e4 2019 error = dir->i_op->create(dir, dentry, mode, nd);
a74574aa 2020 if (!error)
f38aa942 2021 fsnotify_create(dir, dentry);
1da177e4
LT		 2022 return error;
2023}
2024
73d049a4 2025static int may_open(struct path *path, int acc_mode, int flag)
1da177e4 2026{
3fb64190 2027 struct dentry *dentry = path->dentry;
1da177e4
LT		 2028 struct inode *inode = dentry->d_inode;
2029 int error;
2030
bcda7652
AV		 2031 /* O_PATH? */
2032 if (!acc_mode)
2033 return 0;
2034
1da177e4
LT		 2035 if (!inode)
2036 return -ENOENT;
2037
c8fe8f30
CH		 2038 switch (inode->i_mode & S_IFMT) {
2039 case S_IFLNK:
1da177e4 2040 return -ELOOP;
c8fe8f30
CH		 2041 case S_IFDIR:
2042 if (acc_mode & MAY_WRITE)
2043 return -EISDIR;
2044 break;
2045 case S_IFBLK:
2046 case S_IFCHR:
3fb64190 2047 if (path->mnt->mnt_flags & MNT_NODEV)
1da177e4 2048 return -EACCES;
c8fe8f30
CH		 2049 /*FALLTHRU*/
2050 case S_IFIFO:
2051 case S_IFSOCK:
1da177e4 2052 flag &= ~O_TRUNC;
c8fe8f30 2053 break;
4a3fd211 2054 }
b41572e9 2055
3fb64190 2056 error = inode_permission(inode, acc_mode);
b41572e9
DH		 2057 if (error)
2058 return error;
6146f0d5 2059
1da177e4
LT		 2060 /*
2061 * An append-only file must be opened in append mode for writing.
2062 */
2063 if (IS_APPEND(inode)) {
8737c930 2064 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
7715b521 2065 return -EPERM;
1da177e4 2066 if (flag & O_TRUNC)
7715b521 2067 return -EPERM;
1da177e4
LT		 2068 }
2069
2070 /* O_NOATIME can only be set by the owner or superuser */
2e149670 2071 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
7715b521 2072 return -EPERM;
1da177e4 2073
f3c7691e 2074 return 0;
7715b521 2075}
1da177e4 2076
e1181ee6 2077static int handle_truncate(struct file *filp)
7715b521 2078{
e1181ee6 2079 struct path *path = &filp->f_path;
7715b521
AV		 2080 struct inode *inode = path->dentry->d_inode;
2081 int error = get_write_access(inode);
2082 if (error)
2083 return error;
2084 /*
2085 * Refuse to truncate files with mandatory locks held on them.
2086 */
2087 error = locks_verify_locked(inode);
2088 if (!error)
ea0d3ab2 2089 error = security_path_truncate(path);
7715b521
AV		 2090 if (!error) {
2091 error = do_truncate(path->dentry, 0,
2092 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
e1181ee6 2093 filp);
7715b521
AV		 2094 }
2095 put_write_access(inode);
acd0c935 2096 return error;
1da177e4
LT		 2097}
2098
d57999e1
DH		 2099static inline int open_to_namei_flags(int flag)
2100{
8a5e929d
AV		 2101 if ((flag & O_ACCMODE) == 3)
2102 flag--;
d57999e1
DH		 2103 return flag;
2104}
2105
31e6b01f 2106/*
fe2d35ff 2107 * Handle the last step of open()
31e6b01f 2108 */
fb1cc555 2109static struct file *do_last(struct nameidata *nd, struct path *path,
c3e380b0 2110 const struct open_flags *op, const char *pathname)
fb1cc555 2111{
a1e28038 2112 struct dentry *dir = nd->path.dentry;
6c0d46c4 2113 struct dentry *dentry;
ca344a89 2114 int open_flag = op->open_flag;
6c0d46c4 2115 int will_truncate = open_flag & O_TRUNC;
ca344a89 2116 int want_write = 0;
bcda7652 2117 int acc_mode = op->acc_mode;
fb1cc555 2118 struct file *filp;
16c2cd71 2119 int error;
1f36f774 2120
c3e380b0
AV		 2121 nd->flags &= ~LOOKUP_PARENT;
2122 nd->flags |= op->intent;
2123
1f36f774
AV		 2124 switch (nd->last_type) {
2125 case LAST_DOTDOT:
176306f5 2126 case LAST_DOT:
fe2d35ff
AV		 2127 error = handle_dots(nd, nd->last_type);
2128 if (error)
2129 return ERR_PTR(error);
1f36f774 2130 /* fallthrough */
1f36f774 2131 case LAST_ROOT:
9f1fafee 2132 error = complete_walk(nd);
16c2cd71 2133 if (error)
9f1fafee 2134 return ERR_PTR(error);
fe2d35ff 2135 audit_inode(pathname, nd->path.dentry);
ca344a89 2136 if (open_flag & O_CREAT) {
fe2d35ff
AV		 2137 error = -EISDIR;
2138 goto exit;
2139 }
2140 goto ok;
1f36f774 2141 case LAST_BIND:
9f1fafee 2142 error = complete_walk(nd);
16c2cd71 2143 if (error)
9f1fafee 2144 return ERR_PTR(error);
1f36f774 2145 audit_inode(pathname, dir);
67ee3ad2 2146 goto ok;
1f36f774 2147 }
67ee3ad2 2148
ca344a89 2149 if (!(open_flag & O_CREAT)) {
bcda7652 2150 int symlink_ok = 0;
fe2d35ff
AV		 2151 if (nd->last.name[nd->last.len])
2152 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
bcda7652
AV		 2153 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2154 symlink_ok = 1;
fe2d35ff 2155 /* we _can_ be in RCU mode here */
ce57dfc1
AV		 2156 error = walk_component(nd, path, &nd->last, LAST_NORM,
2157 !symlink_ok);
2158 if (error < 0)
fe2d35ff 2159 return ERR_PTR(error);
ce57dfc1 2160 if (error) /* symlink */
fe2d35ff 2161 return NULL;
fe2d35ff 2162 /* sayonara */
9f1fafee
AV		 2163 error = complete_walk(nd);
2164 if (error)
2165 return ERR_PTR(-ECHILD);
fe2d35ff
AV		 2166
2167 error = -ENOTDIR;
2168 if (nd->flags & LOOKUP_DIRECTORY) {
ce57dfc1 2169 if (!nd->inode->i_op->lookup)
fe2d35ff
AV		 2170 goto exit;
2171 }
2172 audit_inode(pathname, nd->path.dentry);
2173 goto ok;
2174 }
2175
2176 /* create side of things */
a3fbbde7
AV		 2177 /*
2178 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED has been
2179 * cleared when we got to the last component we are about to look up
2180 */
9f1fafee
AV		 2181 error = complete_walk(nd);
2182 if (error)
2183 return ERR_PTR(error);
fe2d35ff
AV		 2184
2185 audit_inode(pathname, dir);
16c2cd71 2186 error = -EISDIR;
1f36f774 2187 /* trailing slashes? */
31e6b01f
NP		 2188 if (nd->last.name[nd->last.len])
2189 goto exit;
a2c36b45 2190
a1e28038
AV		 2191 mutex_lock(&dir->d_inode->i_mutex);
2192
6c0d46c4
AV		 2193 dentry = lookup_hash(nd);
2194 error = PTR_ERR(dentry);
2195 if (IS_ERR(dentry)) {
fb1cc555
AV		 2196 mutex_unlock(&dir->d_inode->i_mutex);
2197 goto exit;
2198 }
2199
6c0d46c4
AV		 2200 path->dentry = dentry;
2201 path->mnt = nd->path.mnt;
2202
fb1cc555 2203 /* Negative dentry, just create the file */
6c0d46c4 2204 if (!dentry->d_inode) {
a218d0fd 2205 umode_t mode = op->mode;
6c0d46c4
AV		 2206 if (!IS_POSIXACL(dir->d_inode))
2207 mode &= ~current_umask();
fb1cc555
AV		 2208 /*
2209 * This write is needed to ensure that a
6c0d46c4 2210 * rw->ro transition does not occur between
fb1cc555
AV		 2211 * the time when the file is created and when
2212 * a permanent write count is taken through
2213 * the 'struct file' in nameidata_to_filp().
2214 */
2215 error = mnt_want_write(nd->path.mnt);
2216 if (error)
2217 goto exit_mutex_unlock;
ca344a89 2218 want_write = 1;
9b44f1b3 2219 /* Don't check for write permission, don't truncate */
ca344a89 2220 open_flag &= ~O_TRUNC;
6c0d46c4 2221 will_truncate = 0;
bcda7652 2222 acc_mode = MAY_OPEN;
6c0d46c4
AV		 2223 error = security_path_mknod(&nd->path, dentry, mode, 0);
2224 if (error)
2225 goto exit_mutex_unlock;
2226 error = vfs_create(dir->d_inode, dentry, mode, nd);
2227 if (error)
2228 goto exit_mutex_unlock;
2229 mutex_unlock(&dir->d_inode->i_mutex);
2230 dput(nd->path.dentry);
2231 nd->path.dentry = dentry;
ca344a89 2232 goto common;
fb1cc555
AV		 2233 }
2234
2235 /*
2236 * It already exists.
2237 */
2238 mutex_unlock(&dir->d_inode->i_mutex);
2239 audit_inode(pathname, path->dentry);
2240
2241 error = -EEXIST;
ca344a89 2242 if (open_flag & O_EXCL)
fb1cc555
AV		 2243 goto exit_dput;
2244
9875cf80
DH		 2245 error = follow_managed(path, nd->flags);
2246 if (error < 0)
2247 goto exit_dput;
fb1cc555 2248
a3fbbde7
AV		 2249 if (error)
2250 nd->flags |= LOOKUP_JUMPED;
2251
fb1cc555
AV		 2252 error = -ENOENT;
2253 if (!path->dentry->d_inode)
2254 goto exit_dput;
9e67f361
AV		 2255
2256 if (path->dentry->d_inode->i_op->follow_link)
fb1cc555 2257 return NULL;
fb1cc555
AV		 2258
2259 path_to_nameidata(path, nd);
31e6b01f 2260 nd->inode = path->dentry->d_inode;
a3fbbde7
AV		 2261 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2262 error = complete_walk(nd);
2263 if (error)
2264 goto exit;
fb1cc555 2265 error = -EISDIR;
31e6b01f 2266 if (S_ISDIR(nd->inode->i_mode))
fb1cc555 2267 goto exit;
67ee3ad2 2268ok:
6c0d46c4
AV		 2269 if (!S_ISREG(nd->inode->i_mode))
2270 will_truncate = 0;
2271
0f9d1a10
AV		 2272 if (will_truncate) {
2273 error = mnt_want_write(nd->path.mnt);
2274 if (error)
2275 goto exit;
ca344a89 2276 want_write = 1;
0f9d1a10 2277 }
ca344a89 2278common:
bcda7652 2279 error = may_open(&nd->path, acc_mode, open_flag);
ca344a89 2280 if (error)
0f9d1a10 2281 goto exit;
0f9d1a10
AV		 2282 filp = nameidata_to_filp(nd);
2283 if (!IS_ERR(filp)) {
2284 error = ima_file_check(filp, op->acc_mode);
2285 if (error) {
2286 fput(filp);
2287 filp = ERR_PTR(error);
2288 }
2289 }
2290 if (!IS_ERR(filp)) {
2291 if (will_truncate) {
2292 error = handle_truncate(filp);
2293 if (error) {
2294 fput(filp);
2295 filp = ERR_PTR(error);
2296 }
2297 }
2298 }
ca344a89
AV		 2299out:
2300 if (want_write)
0f9d1a10
AV		 2301 mnt_drop_write(nd->path.mnt);
2302 path_put(&nd->path);
fb1cc555
AV		 2303 return filp;
2304
2305exit_mutex_unlock:
2306 mutex_unlock(&dir->d_inode->i_mutex);
2307exit_dput:
2308 path_put_conditional(path, nd);
2309exit:
ca344a89
AV		 2310 filp = ERR_PTR(error);
2311 goto out;
fb1cc555
AV		 2312}
2313
13aab428 2314static struct file *path_openat(int dfd, const char *pathname,
73d049a4 2315 struct nameidata *nd, const struct open_flags *op, int flags)
1da177e4 2316{
fe2d35ff 2317 struct file *base = NULL;
4a3fd211 2318 struct file *filp;
9850c056 2319 struct path path;
13aab428 2320 int error;
31e6b01f
NP		 2321
2322 filp = get_empty_filp();
2323 if (!filp)
2324 return ERR_PTR(-ENFILE);
2325
47c805dc 2326 filp->f_flags = op->open_flag;
73d049a4
AV		 2327 nd->intent.open.file = filp;
2328 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2329 nd->intent.open.create_mode = op->mode;
31e6b01f 2330
73d049a4 2331 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
31e6b01f 2332 if (unlikely(error))
13aab428 2333 goto out_filp;
31e6b01f 2334
fe2d35ff 2335 current->total_link_count = 0;
73d049a4 2336 error = link_path_walk(pathname, nd);
31e6b01f
NP		 2337 if (unlikely(error))
2338 goto out_filp;
1da177e4 2339
73d049a4 2340 filp = do_last(nd, &path, op, pathname);
806b681c 2341 while (unlikely(!filp)) { /* trailing symlink */
7b9337aa 2342 struct path link = path;
def4af30 2343 void *cookie;
574197e0 2344 if (!(nd->flags & LOOKUP_FOLLOW)) {
73d049a4
AV		 2345 path_put_conditional(&path, nd);
2346 path_put(&nd->path);
40b39136
AV		 2347 filp = ERR_PTR(-ELOOP);
2348 break;
2349 }
73d049a4
AV		 2350 nd->flags |= LOOKUP_PARENT;
2351 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
574197e0 2352 error = follow_link(&link, nd, &cookie);
c3e380b0 2353 if (unlikely(error))
f1afe9ef 2354 filp = ERR_PTR(error);
c3e380b0 2355 else
73d049a4 2356 filp = do_last(nd, &path, op, pathname);
574197e0 2357 put_link(nd, &link, cookie);
806b681c 2358 }
10fa8e62 2359out:
73d049a4
AV		 2360 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2361 path_put(&nd->root);
fe2d35ff
AV		 2362 if (base)
2363 fput(base);
73d049a4 2364 release_open_intent(nd);
10fa8e62 2365 return filp;
1da177e4 2366
31e6b01f 2367out_filp:
806b681c 2368 filp = ERR_PTR(error);
10fa8e62 2369 goto out;
1da177e4
LT		 2370}
2371
13aab428
AV		 2372struct file *do_filp_open(int dfd, const char *pathname,
2373 const struct open_flags *op, int flags)
2374{
73d049a4 2375 struct nameidata nd;
13aab428
AV		 2376 struct file *filp;
2377
73d049a4 2378 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
13aab428 2379 if (unlikely(filp == ERR_PTR(-ECHILD)))
73d049a4 2380 filp = path_openat(dfd, pathname, &nd, op, flags);
13aab428 2381 if (unlikely(filp == ERR_PTR(-ESTALE)))
73d049a4 2382 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
13aab428
AV		 2383 return filp;
2384}
2385
73d049a4
AV		 2386struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2387 const char *name, const struct open_flags *op, int flags)
2388{
2389 struct nameidata nd;
2390 struct file *file;
2391
2392 nd.root.mnt = mnt;
2393 nd.root.dentry = dentry;
2394
2395 flags |= LOOKUP_ROOT;
2396
bcda7652 2397 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
73d049a4
AV		 2398 return ERR_PTR(-ELOOP);
2399
2400 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2401 if (unlikely(file == ERR_PTR(-ECHILD)))
2402 file = path_openat(-1, name, &nd, op, flags);
2403 if (unlikely(file == ERR_PTR(-ESTALE)))
2404 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2405 return file;
2406}
2407
ed75e95d 2408struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
1da177e4 2409{
c663e5d8 2410 struct dentry *dentry = ERR_PTR(-EEXIST);
ed75e95d
AV		 2411 struct nameidata nd;
2412 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2413 if (error)
2414 return ERR_PTR(error);
1da177e4 2415
c663e5d8
CH		 2416 /*
2417 * Yucky last component or no last component at all?
2418 * (foo/., foo/.., /////)
2419 */
ed75e95d
AV		 2420 if (nd.last_type != LAST_NORM)
2421 goto out;
2422 nd.flags &= ~LOOKUP_PARENT;
2423 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2424 nd.intent.open.flags = O_EXCL;
c663e5d8
CH		 2425
2426 /*
2427 * Do the final lookup.
2428 */
ed75e95d
AV		 2429 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2430 dentry = lookup_hash(&nd);
1da177e4
LT		 2431 if (IS_ERR(dentry))
2432 goto fail;
c663e5d8 2433
e9baf6e5
AV		 2434 if (dentry->d_inode)
2435 goto eexist;
c663e5d8
CH		 2436 /*
2437 * Special case - lookup gave negative, but... we had foo/bar/
2438 * From the vfs_mknod() POV we just have a negative dentry -
2439 * all is fine. Let's be bastards - you had / on the end, you've
2440 * been asking for (non-existent) directory. -ENOENT for you.
2441 */
ed75e95d 2442 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
e9baf6e5
AV		 2443 dput(dentry);
2444 dentry = ERR_PTR(-ENOENT);
ed75e95d 2445 goto fail;
e9baf6e5 2446 }
ed75e95d 2447 *path = nd.path;
1da177e4 2448 return dentry;
e9baf6e5 2449eexist:
1da177e4 2450 dput(dentry);
e9baf6e5 2451 dentry = ERR_PTR(-EEXIST);
1da177e4 2452fail:
ed75e95d
AV		 2453 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2454out:
2455 path_put(&nd.path);
1da177e4
LT		 2456 return dentry;
2457}
dae6ad8f
AV		 2458EXPORT_SYMBOL(kern_path_create);
2459
2460struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2461{
2462 char *tmp = getname(pathname);
2463 struct dentry *res;
2464 if (IS_ERR(tmp))
2465 return ERR_CAST(tmp);
2466 res = kern_path_create(dfd, tmp, path, is_dir);
2467 putname(tmp);
2468 return res;
2469}
2470EXPORT_SYMBOL(user_path_create);
2471
1a67aafb 2472int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
1da177e4 2473{
a95164d9 2474 int error = may_create(dir, dentry);
1da177e4
LT		 2475
2476 if (error)
2477 return error;
2478
e795b717
SH		 2479 if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
2480 !ns_capable(inode_userns(dir), CAP_MKNOD))
1da177e4
LT		 2481 return -EPERM;
2482
acfa4380 2483 if (!dir->i_op->mknod)
1da177e4
LT		 2484 return -EPERM;
2485
08ce5f16
SH		 2486 error = devcgroup_inode_mknod(mode, dev);
2487 if (error)
2488 return error;
2489
1da177e4
LT		 2490 error = security_inode_mknod(dir, dentry, mode, dev);
2491 if (error)
2492 return error;
2493
1da177e4 2494 error = dir->i_op->mknod(dir, dentry, mode, dev);
a74574aa 2495 if (!error)
f38aa942 2496 fsnotify_create(dir, dentry);
1da177e4
LT		 2497 return error;
2498}
2499
f69aac00 2500static int may_mknod(umode_t mode)
463c3197
DH		 2501{
2502 switch (mode & S_IFMT) {
2503 case S_IFREG:
2504 case S_IFCHR:
2505 case S_IFBLK:
2506 case S_IFIFO:
2507 case S_IFSOCK:
2508 case 0: /* zero mode translates to S_IFREG */
2509 return 0;
2510 case S_IFDIR:
2511 return -EPERM;
2512 default:
2513 return -EINVAL;
2514 }
2515}
2516
8208a22b 2517SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
2e4d0924 2518 unsigned, dev)
1da177e4 2519{
2ad94ae6 2520 struct dentry *dentry;
dae6ad8f
AV		 2521 struct path path;
2522 int error;
1da177e4
LT		 2523
2524 if (S_ISDIR(mode))
2525 return -EPERM;
1da177e4 2526
dae6ad8f
AV		 2527 dentry = user_path_create(dfd, filename, &path, 0);
2528 if (IS_ERR(dentry))
2529 return PTR_ERR(dentry);
2ad94ae6 2530
dae6ad8f 2531 if (!IS_POSIXACL(path.dentry->d_inode))
ce3b0f8d 2532 mode &= ~current_umask();
463c3197
DH		 2533 error = may_mknod(mode);
2534 if (error)
2535 goto out_dput;
dae6ad8f 2536 error = mnt_want_write(path.mnt);
463c3197
DH		 2537 if (error)
2538 goto out_dput;
dae6ad8f 2539 error = security_path_mknod(&path, dentry, mode, dev);
be6d3e56
KT		 2540 if (error)
2541 goto out_drop_write;
463c3197 2542 switch (mode & S_IFMT) {
1da177e4 2543 case 0: case S_IFREG:
dae6ad8f 2544 error = vfs_create(path.dentry->d_inode,dentry,mode,NULL);
1da177e4
LT		 2545 break;
2546 case S_IFCHR: case S_IFBLK:
dae6ad8f 2547 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
1da177e4
LT		 2548 new_decode_dev(dev));
2549 break;
2550 case S_IFIFO: case S_IFSOCK:
dae6ad8f 2551 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
1da177e4 2552 break;
1da177e4 2553 }
be6d3e56 2554out_drop_write:
dae6ad8f 2555 mnt_drop_write(path.mnt);
463c3197
DH		 2556out_dput:
2557 dput(dentry);
dae6ad8f
AV		 2558 mutex_unlock(&path.dentry->d_inode->i_mutex);
2559 path_put(&path);
1da177e4
LT		 2560
2561 return error;
2562}
2563
8208a22b 2564SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
5590ff0d
UD		 2565{
2566 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2567}
2568
18bb1db3 2569int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1da177e4 2570{
a95164d9 2571 int error = may_create(dir, dentry);
1da177e4
LT		 2572
2573 if (error)
2574 return error;
2575
acfa4380 2576 if (!dir->i_op->mkdir)
1da177e4
LT		 2577 return -EPERM;
2578
2579 mode &= (S_IRWXUGO|S_ISVTX);
2580 error = security_inode_mkdir(dir, dentry, mode);
2581 if (error)
2582 return error;
2583
1da177e4 2584 error = dir->i_op->mkdir(dir, dentry, mode);
a74574aa 2585 if (!error)
f38aa942 2586 fsnotify_mkdir(dir, dentry);
1da177e4
LT		 2587 return error;
2588}
2589
a218d0fd 2590SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
1da177e4 2591{
6902d925 2592 struct dentry *dentry;
dae6ad8f
AV		 2593 struct path path;
2594 int error;
1da177e4 2595
dae6ad8f 2596 dentry = user_path_create(dfd, pathname, &path, 1);
6902d925 2597 if (IS_ERR(dentry))
dae6ad8f 2598 return PTR_ERR(dentry);
1da177e4 2599
dae6ad8f 2600 if (!IS_POSIXACL(path.dentry->d_inode))
ce3b0f8d 2601 mode &= ~current_umask();
dae6ad8f 2602 error = mnt_want_write(path.mnt);
463c3197
DH		 2603 if (error)
2604 goto out_dput;
dae6ad8f 2605 error = security_path_mkdir(&path, dentry, mode);
be6d3e56
KT		 2606 if (error)
2607 goto out_drop_write;
dae6ad8f 2608 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
be6d3e56 2609out_drop_write:
dae6ad8f 2610 mnt_drop_write(path.mnt);
463c3197 2611out_dput:
6902d925 2612 dput(dentry);
dae6ad8f
AV		 2613 mutex_unlock(&path.dentry->d_inode->i_mutex);
2614 path_put(&path);
1da177e4
LT		 2615 return error;
2616}
2617
a218d0fd 2618SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
5590ff0d
UD		 2619{
2620 return sys_mkdirat(AT_FDCWD, pathname, mode);
2621}
2622
1da177e4 2623/*
a71905f0
SW		 2624 * The dentry_unhash() helper will try to drop the dentry early: we
2625 * should have a usage count of 2 if we're the only user of this
2626 * dentry, and if that is true (possibly after pruning the dcache),
2627 * then we drop the dentry now.
1da177e4
LT		 2628 *
2629 * A low-level filesystem can, if it choses, legally
2630 * do a
2631 *
2632 * if (!d_unhashed(dentry))
2633 * return -EBUSY;
2634 *
2635 * if it cannot handle the case of removing a directory
2636 * that is still in use by something else..
2637 */
2638void dentry_unhash(struct dentry *dentry)
2639{
dc168427 2640 shrink_dcache_parent(dentry);
1da177e4 2641 spin_lock(&dentry->d_lock);
64252c75 2642 if (dentry->d_count == 1)
1da177e4
LT		 2643 __d_drop(dentry);
2644 spin_unlock(&dentry->d_lock);
1da177e4
LT		 2645}
2646
2647int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2648{
2649 int error = may_delete(dir, dentry, 1);
2650
2651 if (error)
2652 return error;
2653
acfa4380 2654 if (!dir->i_op->rmdir)
1da177e4
LT		 2655 return -EPERM;
2656
1d2ef590 2657 dget(dentry);
1b1dcc1b 2658 mutex_lock(&dentry->d_inode->i_mutex);
912dbc15
SW		 2659
2660 error = -EBUSY;
1da177e4 2661 if (d_mountpoint(dentry))
912dbc15
SW		 2662 goto out;
2663
2664 error = security_inode_rmdir(dir, dentry);
2665 if (error)
2666 goto out;
2667
3cebde24 2668 shrink_dcache_parent(dentry);
912dbc15
SW		 2669 error = dir->i_op->rmdir(dir, dentry);
2670 if (error)
2671 goto out;
2672
2673 dentry->d_inode->i_flags |= S_DEAD;
2674 dont_mount(dentry);
2675
2676out:
1b1dcc1b 2677 mutex_unlock(&dentry->d_inode->i_mutex);
1d2ef590 2678 dput(dentry);
912dbc15 2679 if (!error)
1da177e4 2680 d_delete(dentry);
1da177e4
LT		 2681 return error;
2682}
2683
5590ff0d 2684static long do_rmdir(int dfd, const char __user *pathname)
1da177e4
LT		 2685{
2686 int error = 0;
2687 char * name;
2688 struct dentry *dentry;
2689 struct nameidata nd;
2690
2ad94ae6 2691 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2692 if (error)
2ad94ae6 2693 return error;
1da177e4
LT		 2694
2695 switch(nd.last_type) {
0612d9fb
OH		 2696 case LAST_DOTDOT:
2697 error = -ENOTEMPTY;
2698 goto exit1;
2699 case LAST_DOT:
2700 error = -EINVAL;
2701 goto exit1;
2702 case LAST_ROOT:
2703 error = -EBUSY;
2704 goto exit1;
1da177e4 2705 }
0612d9fb
OH		 2706
2707 nd.flags &= ~LOOKUP_PARENT;
2708
4ac91378 2709 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2710 dentry = lookup_hash(&nd);
1da177e4 2711 error = PTR_ERR(dentry);
6902d925
DH		 2712 if (IS_ERR(dentry))
2713 goto exit2;
e6bc45d6
TT		 2714 if (!dentry->d_inode) {
2715 error = -ENOENT;
2716 goto exit3;
2717 }
0622753b
DH		 2718 error = mnt_want_write(nd.path.mnt);
2719 if (error)
2720 goto exit3;
be6d3e56
KT		 2721 error = security_path_rmdir(&nd.path, dentry);
2722 if (error)
2723 goto exit4;
4ac91378 2724 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
be6d3e56 2725exit4:
0622753b
DH		 2726 mnt_drop_write(nd.path.mnt);
2727exit3:
6902d925
DH		 2728 dput(dentry);
2729exit2:
4ac91378 2730 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2731exit1:
1d957f9b 2732 path_put(&nd.path);
1da177e4
LT		 2733 putname(name);
2734 return error;
2735}
2736
3cdad428 2737SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
5590ff0d
UD		 2738{
2739 return do_rmdir(AT_FDCWD, pathname);
2740}
2741
1da177e4
LT		 2742int vfs_unlink(struct inode *dir, struct dentry *dentry)
2743{
2744 int error = may_delete(dir, dentry, 0);
2745
2746 if (error)
2747 return error;
2748
acfa4380 2749 if (!dir->i_op->unlink)
1da177e4
LT		 2750 return -EPERM;
2751
1b1dcc1b 2752 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2753 if (d_mountpoint(dentry))
2754 error = -EBUSY;
2755 else {
2756 error = security_inode_unlink(dir, dentry);
bec1052e 2757 if (!error) {
1da177e4 2758 error = dir->i_op->unlink(dir, dentry);
bec1052e 2759 if (!error)
d83c49f3 2760 dont_mount(dentry);
bec1052e 2761 }
1da177e4 2762 }
1b1dcc1b 2763 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2764
2765 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2766 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
ece95912 2767 fsnotify_link_count(dentry->d_inode);
e234f35c 2768 d_delete(dentry);
1da177e4 2769 }
0eeca283 2770
1da177e4
LT		 2771 return error;
2772}
2773
2774/*
2775 * Make sure that the actual truncation of the file will occur outside its
1b1dcc1b 2776 * directory's i_mutex. Truncate can take a long time if there is a lot of
1da177e4
LT		 2777 * writeout happening, and we don't want to prevent access to the directory
2778 * while waiting on the I/O.
2779 */
5590ff0d 2780static long do_unlinkat(int dfd, const char __user *pathname)
1da177e4 2781{
2ad94ae6
AV		 2782 int error;
2783 char *name;
1da177e4
LT		 2784 struct dentry *dentry;
2785 struct nameidata nd;
2786 struct inode *inode = NULL;
2787
2ad94ae6 2788 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2789 if (error)
2ad94ae6
AV		 2790 return error;
2791
1da177e4
LT		 2792 error = -EISDIR;
2793 if (nd.last_type != LAST_NORM)
2794 goto exit1;
0612d9fb
OH		 2795
2796 nd.flags &= ~LOOKUP_PARENT;
2797
4ac91378 2798 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2799 dentry = lookup_hash(&nd);
1da177e4
LT		 2800 error = PTR_ERR(dentry);
2801 if (!IS_ERR(dentry)) {
2802 /* Why not before? Because we want correct error value */
50338b88
TE		 2803 if (nd.last.name[nd.last.len])
2804 goto slashes;
1da177e4 2805 inode = dentry->d_inode;
50338b88 2806 if (!inode)
e6bc45d6
TT		 2807 goto slashes;
2808 ihold(inode);
0622753b
DH		 2809 error = mnt_want_write(nd.path.mnt);
2810 if (error)
2811 goto exit2;
be6d3e56
KT		 2812 error = security_path_unlink(&nd.path, dentry);
2813 if (error)
2814 goto exit3;
4ac91378 2815 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
be6d3e56 2816exit3:
0622753b 2817 mnt_drop_write(nd.path.mnt);
1da177e4
LT		 2818 exit2:
2819 dput(dentry);
2820 }
4ac91378 2821 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4
LT		 2822 if (inode)
2823 iput(inode); /* truncate the inode here */
2824exit1:
1d957f9b 2825 path_put(&nd.path);
1da177e4
LT		 2826 putname(name);
2827 return error;
2828
2829slashes:
2830 error = !dentry->d_inode ? -ENOENT :
2831 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2832 goto exit2;
2833}
2834
2e4d0924 2835SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
5590ff0d
UD		 2836{
2837 if ((flag & ~AT_REMOVEDIR) != 0)
2838 return -EINVAL;
2839
2840 if (flag & AT_REMOVEDIR)
2841 return do_rmdir(dfd, pathname);
2842
2843 return do_unlinkat(dfd, pathname);
2844}
2845
3480b257 2846SYSCALL_DEFINE1(unlink, const char __user *, pathname)
5590ff0d
UD		 2847{
2848 return do_unlinkat(AT_FDCWD, pathname);
2849}
2850
db2e747b 2851int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
1da177e4 2852{
a95164d9 2853 int error = may_create(dir, dentry);
1da177e4
LT		 2854
2855 if (error)
2856 return error;
2857
acfa4380 2858 if (!dir->i_op->symlink)
1da177e4
LT		 2859 return -EPERM;
2860
2861 error = security_inode_symlink(dir, dentry, oldname);
2862 if (error)
2863 return error;
2864
1da177e4 2865 error = dir->i_op->symlink(dir, dentry, oldname);
a74574aa 2866 if (!error)
f38aa942 2867 fsnotify_create(dir, dentry);
1da177e4
LT		 2868 return error;
2869}
2870
2e4d0924
HC		 2871SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2872 int, newdfd, const char __user *, newname)
1da177e4 2873{
2ad94ae6
AV		 2874 int error;
2875 char *from;
6902d925 2876 struct dentry *dentry;
dae6ad8f 2877 struct path path;
1da177e4
LT		 2878
2879 from = getname(oldname);
2ad94ae6 2880 if (IS_ERR(from))
1da177e4 2881 return PTR_ERR(from);
1da177e4 2882
dae6ad8f 2883 dentry = user_path_create(newdfd, newname, &path, 0);
6902d925
DH		 2884 error = PTR_ERR(dentry);
2885 if (IS_ERR(dentry))
dae6ad8f 2886 goto out_putname;
6902d925 2887
dae6ad8f 2888 error = mnt_want_write(path.mnt);
75c3f29d
DH		 2889 if (error)
2890 goto out_dput;
dae6ad8f 2891 error = security_path_symlink(&path, dentry, from);
be6d3e56
KT		 2892 if (error)
2893 goto out_drop_write;
dae6ad8f 2894 error = vfs_symlink(path.dentry->d_inode, dentry, from);
be6d3e56 2895out_drop_write:
dae6ad8f 2896 mnt_drop_write(path.mnt);
75c3f29d 2897out_dput:
6902d925 2898 dput(dentry);
dae6ad8f
AV		 2899 mutex_unlock(&path.dentry->d_inode->i_mutex);
2900 path_put(&path);
6902d925 2901out_putname:
1da177e4
LT		 2902 putname(from);
2903 return error;
2904}
2905
3480b257 2906SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 2907{
2908 return sys_symlinkat(oldname, AT_FDCWD, newname);
2909}
2910
1da177e4
LT		 2911int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2912{
2913 struct inode *inode = old_dentry->d_inode;
2914 int error;
2915
2916 if (!inode)
2917 return -ENOENT;
2918
a95164d9 2919 error = may_create(dir, new_dentry);
1da177e4
LT		 2920 if (error)
2921 return error;
2922
2923 if (dir->i_sb != inode->i_sb)
2924 return -EXDEV;
2925
2926 /*
2927 * A link to an append-only or immutable file cannot be created.
2928 */
2929 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2930 return -EPERM;
acfa4380 2931 if (!dir->i_op->link)
1da177e4 2932 return -EPERM;
7e79eedb 2933 if (S_ISDIR(inode->i_mode))
1da177e4
LT		 2934 return -EPERM;
2935
2936 error = security_inode_link(old_dentry, dir, new_dentry);
2937 if (error)
2938 return error;
2939
7e79eedb 2940 mutex_lock(&inode->i_mutex);
aae8a97d
AK		 2941 /* Make sure we don't allow creating hardlink to an unlinked file */
2942 if (inode->i_nlink == 0)
2943 error = -ENOENT;
2944 else
2945 error = dir->i_op->link(old_dentry, dir, new_dentry);
7e79eedb 2946 mutex_unlock(&inode->i_mutex);
e31e14ec 2947 if (!error)
7e79eedb 2948 fsnotify_link(dir, inode, new_dentry);
1da177e4
LT		 2949 return error;
2950}
2951
2952/*
2953 * Hardlinks are often used in delicate situations. We avoid
2954 * security-related surprises by not following symlinks on the
2955 * newname. --KAB
2956 *
2957 * We don't follow them on the oldname either to be compatible
2958 * with linux 2.0, and to avoid hard-linking to directories
2959 * and other special files. --ADM
2960 */
2e4d0924
HC		 2961SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2962 int, newdfd, const char __user *, newname, int, flags)
1da177e4
LT		 2963{
2964 struct dentry *new_dentry;
dae6ad8f 2965 struct path old_path, new_path;
11a7b371 2966 int how = 0;
1da177e4 2967 int error;
1da177e4 2968
11a7b371 2969 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
c04030e1 2970 return -EINVAL;
11a7b371
AK		 2971 /*
2972 * To use null names we require CAP_DAC_READ_SEARCH
2973 * This ensures that not everyone will be able to create
2974 * handlink using the passed filedescriptor.
2975 */
2976 if (flags & AT_EMPTY_PATH) {
2977 if (!capable(CAP_DAC_READ_SEARCH))
2978 return -ENOENT;
2979 how = LOOKUP_EMPTY;
2980 }
2981
2982 if (flags & AT_SYMLINK_FOLLOW)
2983 how |= LOOKUP_FOLLOW;
c04030e1 2984
11a7b371 2985 error = user_path_at(olddfd, oldname, how, &old_path);
1da177e4 2986 if (error)
2ad94ae6
AV		 2987 return error;
2988
dae6ad8f 2989 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
1da177e4 2990 error = PTR_ERR(new_dentry);
6902d925 2991 if (IS_ERR(new_dentry))
dae6ad8f
AV		 2992 goto out;
2993
2994 error = -EXDEV;
2995 if (old_path.mnt != new_path.mnt)
2996 goto out_dput;
2997 error = mnt_want_write(new_path.mnt);
75c3f29d
DH		 2998 if (error)
2999 goto out_dput;
dae6ad8f 3000 error = security_path_link(old_path.dentry, &new_path, new_dentry);
be6d3e56
KT		 3001 if (error)
3002 goto out_drop_write;
dae6ad8f 3003 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
be6d3e56 3004out_drop_write:
dae6ad8f 3005 mnt_drop_write(new_path.mnt);
75c3f29d 3006out_dput:
6902d925 3007 dput(new_dentry);
dae6ad8f
AV		 3008 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
3009 path_put(&new_path);
1da177e4 3010out:
2d8f3038 3011 path_put(&old_path);
1da177e4
LT		 3012
3013 return error;
3014}
3015
3480b257 3016SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
5590ff0d 3017{
c04030e1 3018 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
5590ff0d
UD		 3019}
3020
1da177e4
LT		 3021/*
3022 * The worst of all namespace operations - renaming directory. "Perverted"
3023 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3024 * Problems:
3025 * a) we can get into loop creation. Check is done in is_subdir().
3026 * b) race potential - two innocent renames can create a loop together.
3027 * That's where 4.4 screws up. Current fix: serialization on
a11f3a05 3028 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
1da177e4
LT		 3029 * story.
3030 * c) we have to lock _three_ objects - parents and victim (if it exists).
1b1dcc1b 3031 * And that - after we got ->i_mutex on parents (until then we don't know
1da177e4
LT		 3032 * whether the target exists). Solution: try to be smart with locking
3033 * order for inodes. We rely on the fact that tree topology may change
a11f3a05 3034 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
1da177e4
LT		 3035 * move will be locked. Thus we can rank directories by the tree
3036 * (ancestors first) and rank all non-directories after them.
3037 * That works since everybody except rename does "lock parent, lookup,
a11f3a05 3038 * lock child" and rename is under ->s_vfs_rename_mutex.
1da177e4
LT		 3039 * HOWEVER, it relies on the assumption that any object with ->lookup()
3040 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3041 * we'd better make sure that there's no link(2) for them.
e4eaac06 3042 * d) conversion from fhandle to dentry may come in the wrong moment - when
1b1dcc1b 3043 * we are removing the target. Solution: we will have to grab ->i_mutex
1da177e4 3044 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
c41b20e7 3045 * ->i_mutex on parents, which works but leads to some truly excessive
1da177e4
LT		 3046 * locking].
3047 */
75c96f85
AB		 3048static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3049 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3050{
3051 int error = 0;
9055cba7 3052 struct inode *target = new_dentry->d_inode;
1da177e4
LT		 3053
3054 /*
3055 * If we are going to change the parent - check write permissions,
3056 * we'll need to flip '..'.
3057 */
3058 if (new_dir != old_dir) {
f419a2e3 3059 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
1da177e4
LT		 3060 if (error)
3061 return error;
3062 }
3063
3064 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3065 if (error)
3066 return error;
3067
1d2ef590 3068 dget(new_dentry);
d83c49f3 3069 if (target)
1b1dcc1b 3070 mutex_lock(&target->i_mutex);
9055cba7
SW		 3071
3072 error = -EBUSY;
3073 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3074 goto out;
3075
3cebde24
SW		 3076 if (target)
3077 shrink_dcache_parent(new_dentry);
9055cba7
SW		 3078 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3079 if (error)
3080 goto out;
3081
1da177e4 3082 if (target) {
9055cba7
SW		 3083 target->i_flags |= S_DEAD;
3084 dont_mount(new_dentry);
1da177e4 3085 }
9055cba7
SW		 3086out:
3087 if (target)
3088 mutex_unlock(&target->i_mutex);
1d2ef590 3089 dput(new_dentry);
e31e14ec 3090 if (!error)
349457cc
MF		 3091 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3092 d_move(old_dentry,new_dentry);
1da177e4
LT		 3093 return error;
3094}
3095
75c96f85
AB		 3096static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3097 struct inode *new_dir, struct dentry *new_dentry)
1da177e4 3098{
51892bbb 3099 struct inode *target = new_dentry->d_inode;
1da177e4
LT		 3100 int error;
3101
3102 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3103 if (error)
3104 return error;
3105
3106 dget(new_dentry);
1da177e4 3107 if (target)
1b1dcc1b 3108 mutex_lock(&target->i_mutex);
51892bbb
SW		 3109
3110 error = -EBUSY;
1da177e4 3111 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
51892bbb
SW		 3112 goto out;
3113
3114 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3115 if (error)
3116 goto out;
3117
3118 if (target)
3119 dont_mount(new_dentry);
3120 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3121 d_move(old_dentry, new_dentry);
3122out:
1da177e4 3123 if (target)
1b1dcc1b 3124 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3125 dput(new_dentry);
3126 return error;
3127}
3128
3129int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3130 struct inode *new_dir, struct dentry *new_dentry)
3131{
3132 int error;
3133 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
59b0df21 3134 const unsigned char *old_name;
1da177e4
LT		 3135
3136 if (old_dentry->d_inode == new_dentry->d_inode)
3137 return 0;
3138
3139 error = may_delete(old_dir, old_dentry, is_dir);
3140 if (error)
3141 return error;
3142
3143 if (!new_dentry->d_inode)
a95164d9 3144 error = may_create(new_dir, new_dentry);
1da177e4
LT		 3145 else
3146 error = may_delete(new_dir, new_dentry, is_dir);
3147 if (error)
3148 return error;
3149
acfa4380 3150 if (!old_dir->i_op->rename)
1da177e4
LT		 3151 return -EPERM;
3152
0eeca283
RL		 3153 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3154
1da177e4
LT		 3155 if (is_dir)
3156 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3157 else
3158 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
123df294
AV		 3159 if (!error)
3160 fsnotify_move(old_dir, new_dir, old_name, is_dir,
5a190ae6 3161 new_dentry->d_inode, old_dentry);
0eeca283
RL		 3162 fsnotify_oldname_free(old_name);
3163
1da177e4
LT		 3164 return error;
3165}
3166
2e4d0924
HC		 3167SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3168 int, newdfd, const char __user *, newname)
1da177e4 3169{
2ad94ae6
AV		 3170 struct dentry *old_dir, *new_dir;
3171 struct dentry *old_dentry, *new_dentry;
3172 struct dentry *trap;
1da177e4 3173 struct nameidata oldnd, newnd;
2ad94ae6
AV		 3174 char *from;
3175 char *to;
3176 int error;
1da177e4 3177
2ad94ae6 3178 error = user_path_parent(olddfd, oldname, &oldnd, &from);
1da177e4
LT		 3179 if (error)
3180 goto exit;
3181
2ad94ae6 3182 error = user_path_parent(newdfd, newname, &newnd, &to);
1da177e4
LT		 3183 if (error)
3184 goto exit1;
3185
3186 error = -EXDEV;
4ac91378 3187 if (oldnd.path.mnt != newnd.path.mnt)
1da177e4
LT		 3188 goto exit2;
3189
4ac91378 3190 old_dir = oldnd.path.dentry;
1da177e4
LT		 3191 error = -EBUSY;
3192 if (oldnd.last_type != LAST_NORM)
3193 goto exit2;
3194
4ac91378 3195 new_dir = newnd.path.dentry;
1da177e4
LT		 3196 if (newnd.last_type != LAST_NORM)
3197 goto exit2;
3198
0612d9fb
OH		 3199 oldnd.flags &= ~LOOKUP_PARENT;
3200 newnd.flags &= ~LOOKUP_PARENT;
4e9ed2f8 3201 newnd.flags |= LOOKUP_RENAME_TARGET;
0612d9fb 3202
1da177e4
LT		 3203 trap = lock_rename(new_dir, old_dir);
3204
49705b77 3205 old_dentry = lookup_hash(&oldnd);
1da177e4
LT		 3206 error = PTR_ERR(old_dentry);
3207 if (IS_ERR(old_dentry))
3208 goto exit3;
3209 /* source must exist */
3210 error = -ENOENT;
3211 if (!old_dentry->d_inode)
3212 goto exit4;
3213 /* unless the source is a directory trailing slashes give -ENOTDIR */
3214 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3215 error = -ENOTDIR;
3216 if (oldnd.last.name[oldnd.last.len])
3217 goto exit4;
3218 if (newnd.last.name[newnd.last.len])
3219 goto exit4;
3220 }
3221 /* source should not be ancestor of target */
3222 error = -EINVAL;
3223 if (old_dentry == trap)
3224 goto exit4;
49705b77 3225 new_dentry = lookup_hash(&newnd);
1da177e4
LT		 3226 error = PTR_ERR(new_dentry);
3227 if (IS_ERR(new_dentry))
3228 goto exit4;
3229 /* target should not be an ancestor of source */
3230 error = -ENOTEMPTY;
3231 if (new_dentry == trap)
3232 goto exit5;
3233
9079b1eb
DH		 3234 error = mnt_want_write(oldnd.path.mnt);
3235 if (error)
3236 goto exit5;
be6d3e56
KT		 3237 error = security_path_rename(&oldnd.path, old_dentry,
3238 &newnd.path, new_dentry);
3239 if (error)
3240 goto exit6;
1da177e4
LT		 3241 error = vfs_rename(old_dir->d_inode, old_dentry,
3242 new_dir->d_inode, new_dentry);
be6d3e56 3243exit6:
9079b1eb 3244 mnt_drop_write(oldnd.path.mnt);
1da177e4
LT		 3245exit5:
3246 dput(new_dentry);
3247exit4:
3248 dput(old_dentry);
3249exit3:
3250 unlock_rename(new_dir, old_dir);
3251exit2:
1d957f9b 3252 path_put(&newnd.path);
2ad94ae6 3253 putname(to);
1da177e4 3254exit1:
1d957f9b 3255 path_put(&oldnd.path);
1da177e4 3256 putname(from);
2ad94ae6 3257exit:
1da177e4
LT		 3258 return error;
3259}
3260
a26eab24 3261SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 3262{
3263 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3264}
3265
1da177e4
LT		 3266int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3267{
3268 int len;
3269
3270 len = PTR_ERR(link);
3271 if (IS_ERR(link))
3272 goto out;
3273
3274 len = strlen(link);
3275 if (len > (unsigned) buflen)
3276 len = buflen;
3277 if (copy_to_user(buffer, link, len))
3278 len = -EFAULT;
3279out:
3280 return len;
3281}
3282
3283/*
3284 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3285 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3286 * using) it for any given inode is up to filesystem.
3287 */
3288int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3289{
3290 struct nameidata nd;
cc314eef 3291 void *cookie;
694a1764 3292 int res;
cc314eef 3293
1da177e4 3294 nd.depth = 0;
cc314eef 3295 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
694a1764
MS		 3296 if (IS_ERR(cookie))
3297 return PTR_ERR(cookie);
3298
3299 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3300 if (dentry->d_inode->i_op->put_link)
3301 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3302 return res;
1da177e4
LT		 3303}
3304
3305int vfs_follow_link(struct nameidata *nd, const char *link)
3306{
3307 return __vfs_follow_link(nd, link);
3308}
3309
3310/* get the link contents into pagecache */
3311static char *page_getlink(struct dentry * dentry, struct page **ppage)
3312{
ebd09abb
DG		 3313 char *kaddr;
3314 struct page *page;
1da177e4 3315 struct address_space *mapping = dentry->d_inode->i_mapping;
090d2b18 3316 page = read_mapping_page(mapping, 0, NULL);
1da177e4 3317 if (IS_ERR(page))
6fe6900e 3318 return (char*)page;
1da177e4 3319 *ppage = page;
ebd09abb
DG		 3320 kaddr = kmap(page);
3321 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3322 return kaddr;
1da177e4
LT		 3323}
3324
3325int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3326{
3327 struct page *page = NULL;
3328 char *s = page_getlink(dentry, &page);
3329 int res = vfs_readlink(dentry,buffer,buflen,s);
3330 if (page) {
3331 kunmap(page);
3332 page_cache_release(page);
3333 }
3334 return res;
3335}
3336
cc314eef 3337void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
1da177e4 3338{
cc314eef 3339 struct page *page = NULL;
1da177e4 3340 nd_set_link(nd, page_getlink(dentry, &page));
cc314eef 3341 return page;
1da177e4
LT		 3342}
3343
cc314eef 3344void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1da177e4 3345{
cc314eef
LT		 3346 struct page *page = cookie;
3347
3348 if (page) {
1da177e4
LT		 3349 kunmap(page);
3350 page_cache_release(page);
1da177e4
LT		 3351 }
3352}
3353
54566b2c
NP		 3354/*
3355 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3356 */
3357int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
1da177e4
LT		 3358{
3359 struct address_space *mapping = inode->i_mapping;
0adb25d2 3360 struct page *page;
afddba49 3361 void *fsdata;
beb497ab 3362 int err;
1da177e4 3363 char *kaddr;
54566b2c
NP		 3364 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3365 if (nofs)
3366 flags |= AOP_FLAG_NOFS;
1da177e4 3367
7e53cac4 3368retry:
afddba49 3369 err = pagecache_write_begin(NULL, mapping, 0, len-1,
54566b2c 3370 flags, &page, &fsdata);
1da177e4 3371 if (err)
afddba49
NP		 3372 goto fail;
3373
1da177e4
LT		 3374 kaddr = kmap_atomic(page, KM_USER0);
3375 memcpy(kaddr, symname, len-1);
3376 kunmap_atomic(kaddr, KM_USER0);
afddba49
NP		 3377
3378 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3379 page, fsdata);
1da177e4
LT		 3380 if (err < 0)
3381 goto fail;
afddba49
NP		 3382 if (err < len-1)
3383 goto retry;
3384
1da177e4
LT		 3385 mark_inode_dirty(inode);
3386 return 0;
1da177e4
LT		 3387fail:
3388 return err;
3389}
3390
0adb25d2
KK		 3391int page_symlink(struct inode *inode, const char *symname, int len)
3392{
3393 return __page_symlink(inode, symname, len,
54566b2c 3394 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
0adb25d2
KK		 3395}
3396
92e1d5be 3397const struct inode_operations page_symlink_inode_operations = {
1da177e4
LT		 3398 .readlink = generic_readlink,
3399 .follow_link = page_follow_link_light,
3400 .put_link = page_put_link,
3401};
3402
2d8f3038 3403EXPORT_SYMBOL(user_path_at);
cc53ce53 3404EXPORT_SYMBOL(follow_down_one);
1da177e4
LT		 3405EXPORT_SYMBOL(follow_down);
3406EXPORT_SYMBOL(follow_up);
3407EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3408EXPORT_SYMBOL(getname);
3409EXPORT_SYMBOL(lock_rename);
1da177e4
LT		 3410EXPORT_SYMBOL(lookup_one_len);
3411EXPORT_SYMBOL(page_follow_link_light);
3412EXPORT_SYMBOL(page_put_link);
3413EXPORT_SYMBOL(page_readlink);
0adb25d2 3414EXPORT_SYMBOL(__page_symlink);
1da177e4
LT		 3415EXPORT_SYMBOL(page_symlink);
3416EXPORT_SYMBOL(page_symlink_inode_operations);
d1811465 3417EXPORT_SYMBOL(kern_path);
16f18200 3418EXPORT_SYMBOL(vfs_path_lookup);
f419a2e3 3419EXPORT_SYMBOL(inode_permission);
1da177e4
LT		 3420EXPORT_SYMBOL(unlock_rename);
3421EXPORT_SYMBOL(vfs_create);
3422EXPORT_SYMBOL(vfs_follow_link);
3423EXPORT_SYMBOL(vfs_link);
3424EXPORT_SYMBOL(vfs_mkdir);
3425EXPORT_SYMBOL(vfs_mknod);
3426EXPORT_SYMBOL(generic_permission);
3427EXPORT_SYMBOL(vfs_readlink);
3428EXPORT_SYMBOL(vfs_rename);
3429EXPORT_SYMBOL(vfs_rmdir);
3430EXPORT_SYMBOL(vfs_symlink);
3431EXPORT_SYMBOL(vfs_unlink);
3432EXPORT_SYMBOL(dentry_unhash);
3433EXPORT_SYMBOL(generic_readlink);
kernel source for telekom puls (alcatel/mediatek)