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