projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
vfs: move O_DIRECT check to common code
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / fs / namei.c
CommitLineData
1da177e4
LT		 1/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
630d9c47 18#include <linux/export.h>
44696908 19#include <linux/kernel.h>
1da177e4
LT		 20#include <linux/slab.h>
21#include <linux/fs.h>
22#include <linux/namei.h>
1da177e4 23#include <linux/pagemap.h>
0eeca283 24#include <linux/fsnotify.h>
1da177e4
LT		 25#include <linux/personality.h>
26#include <linux/security.h>
6146f0d5 27#include <linux/ima.h>
1da177e4
LT		 28#include <linux/syscalls.h>
29#include <linux/mount.h>
30#include <linux/audit.h>
16f7e0fe 31#include <linux/capability.h>
834f2a4a 32#include <linux/file.h>
5590ff0d 33#include <linux/fcntl.h>
08ce5f16 34#include <linux/device_cgroup.h>
5ad4e53b 35#include <linux/fs_struct.h>
e77819e5 36#include <linux/posix_acl.h>
1da177e4
LT		 37#include <asm/uaccess.h>
38
e81e3f4d 39#include "internal.h"
c7105365 40#include "mount.h"
e81e3f4d 41
1da177e4
LT		 42/* [Feb-1997 T. Schoebel-Theuer]
43 * Fundamental changes in the pathname lookup mechanisms (namei)
44 * were necessary because of omirr. The reason is that omirr needs
45 * to know the _real_ pathname, not the user-supplied one, in case
46 * of symlinks (and also when transname replacements occur).
47 *
48 * The new code replaces the old recursive symlink resolution with
49 * an iterative one (in case of non-nested symlink chains). It does
50 * this with calls to <fs>_follow_link().
51 * As a side effect, dir_namei(), _namei() and follow_link() are now
52 * replaced with a single function lookup_dentry() that can handle all
53 * the special cases of the former code.
54 *
55 * With the new dcache, the pathname is stored at each inode, at least as
56 * long as the refcount of the inode is positive. As a side effect, the
57 * size of the dcache depends on the inode cache and thus is dynamic.
58 *
59 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
60 * resolution to correspond with current state of the code.
61 *
62 * Note that the symlink resolution is not *completely* iterative.
63 * There is still a significant amount of tail- and mid- recursion in
64 * the algorithm. Also, note that <fs>_readlink() is not used in
65 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
66 * may return different results than <fs>_follow_link(). Many virtual
67 * filesystems (including /proc) exhibit this behavior.
68 */
69
70/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
71 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
72 * and the name already exists in form of a symlink, try to create the new
73 * name indicated by the symlink. The old code always complained that the
74 * name already exists, due to not following the symlink even if its target
75 * is nonexistent. The new semantics affects also mknod() and link() when
25985edc 76 * the name is a symlink pointing to a non-existent name.
1da177e4
LT		 77 *
78 * I don't know which semantics is the right one, since I have no access
79 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
80 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
81 * "old" one. Personally, I think the new semantics is much more logical.
82 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
83 * file does succeed in both HP-UX and SunOs, but not in Solaris
84 * and in the old Linux semantics.
85 */
86
87/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
88 * semantics. See the comments in "open_namei" and "do_link" below.
89 *
90 * [10-Sep-98 Alan Modra] Another symlink change.
91 */
92
93/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
94 * inside the path - always follow.
95 * in the last component in creation/removal/renaming - never follow.
96 * if LOOKUP_FOLLOW passed - follow.
97 * if the pathname has trailing slashes - follow.
98 * otherwise - don't follow.
99 * (applied in that order).
100 *
101 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
102 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
103 * During the 2.4 we need to fix the userland stuff depending on it -
104 * hopefully we will be able to get rid of that wart in 2.5. So far only
105 * XEmacs seems to be relying on it...
106 */
107/*
108 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
a11f3a05 109 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
1da177e4
LT		 110 * any extra contention...
111 */
112
113/* In order to reduce some races, while at the same time doing additional
114 * checking and hopefully speeding things up, we copy filenames to the
115 * kernel data space before using them..
116 *
117 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
118 * PATH_MAX includes the nul terminator --RR.
119 */
1fa1e7f6 120static char *getname_flags(const char __user *filename, int flags, int *empty)
1da177e4 121{
3f9f0aa6
LT		 122 char *result = __getname(), *err;
123 int len;
4043cde8 124
3f9f0aa6 125 if (unlikely(!result))
4043cde8
EP		 126 return ERR_PTR(-ENOMEM);
127
3f9f0aa6
LT		 128 len = strncpy_from_user(result, filename, PATH_MAX);
129 err = ERR_PTR(len);
130 if (unlikely(len < 0))
131 goto error;
132
133 /* The empty path is special. */
134 if (unlikely(!len)) {
135 if (empty)
4043cde8 136 *empty = 1;
3f9f0aa6
LT		 137 err = ERR_PTR(-ENOENT);
138 if (!(flags & LOOKUP_EMPTY))
139 goto error;
1da177e4 140 }
3f9f0aa6
LT		 141
142 err = ERR_PTR(-ENAMETOOLONG);
143 if (likely(len < PATH_MAX)) {
144 audit_getname(result);
145 return result;
146 }
147
148error:
149 __putname(result);
150 return err;
1da177e4
LT		 151}
152
f52e0c11
AV		 153char *getname(const char __user * filename)
154{
f7493e5d 155 return getname_flags(filename, 0, NULL);
f52e0c11
AV		 156}
157
1da177e4
LT		 158#ifdef CONFIG_AUDITSYSCALL
159void putname(const char *name)
160{
5ac3a9c2 161 if (unlikely(!audit_dummy_context()))
1da177e4
LT		 162 audit_putname(name);
163 else
164 __putname(name);
165}
166EXPORT_SYMBOL(putname);
167#endif
168
e77819e5
LT		 169static int check_acl(struct inode *inode, int mask)
170{
84635d68 171#ifdef CONFIG_FS_POSIX_ACL
e77819e5
LT		 172 struct posix_acl *acl;
173
e77819e5 174 if (mask & MAY_NOT_BLOCK) {
3567866b
AV		 175 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
176 if (!acl)
e77819e5 177 return -EAGAIN;
3567866b
AV		 178 /* no ->get_acl() calls in RCU mode... */
179 if (acl == ACL_NOT_CACHED)
180 return -ECHILD;
206b1d09 181 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
e77819e5
LT		 182 }
183
184 acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
185
186 /*
4e34e719
CH		 187 * A filesystem can force a ACL callback by just never filling the
188 * ACL cache. But normally you'd fill the cache either at inode
189 * instantiation time, or on the first ->get_acl call.
e77819e5 190 *
4e34e719
CH		 191 * If the filesystem doesn't have a get_acl() function at all, we'll
192 * just create the negative cache entry.
e77819e5
LT		 193 */
194 if (acl == ACL_NOT_CACHED) {
4e34e719
CH		 195 if (inode->i_op->get_acl) {
196 acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS);
197 if (IS_ERR(acl))
198 return PTR_ERR(acl);
199 } else {
200 set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
201 return -EAGAIN;
202 }
e77819e5
LT		 203 }
204
205 if (acl) {
206 int error = posix_acl_permission(inode, acl, mask);
207 posix_acl_release(acl);
208 return error;
209 }
84635d68 210#endif
e77819e5
LT		 211
212 return -EAGAIN;
213}
214
5909ccaa 215/*
948409c7 216 * This does the basic permission checking
1da177e4 217 */
7e40145e 218static int acl_permission_check(struct inode *inode, int mask)
1da177e4 219{
26cf46be 220 unsigned int mode = inode->i_mode;
1da177e4 221
8e96e3b7 222 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
1da177e4
LT		 223 mode >>= 6;
224 else {
e77819e5 225 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
7e40145e 226 int error = check_acl(inode, mask);
b74c79e9
NP		 227 if (error != -EAGAIN)
228 return error;
1da177e4
LT		 229 }
230
231 if (in_group_p(inode->i_gid))
232 mode >>= 3;
233 }
234
235 /*
236 * If the DACs are ok we don't need any capability check.
237 */
9c2c7039 238 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
1da177e4 239 return 0;
5909ccaa
LT		 240 return -EACCES;
241}
242
243/**
b74c79e9 244 * generic_permission - check for access rights on a Posix-like filesystem
5909ccaa 245 * @inode: inode to check access rights for
8fd90c8d 246 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
5909ccaa
LT		 247 *
248 * Used to check for read/write/execute permissions on a file.
249 * We use "fsuid" for this, letting us set arbitrary permissions
250 * for filesystem access without changing the "normal" uids which
b74c79e9
NP		 251 * are used for other things.
252 *
253 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
254 * request cannot be satisfied (eg. requires blocking or too much complexity).
255 * It would then be called again in ref-walk mode.
5909ccaa 256 */
2830ba7f 257int generic_permission(struct inode *inode, int mask)
5909ccaa
LT		 258{
259 int ret;
260
261 /*
948409c7 262 * Do the basic permission checks.
5909ccaa 263 */
7e40145e 264 ret = acl_permission_check(inode, mask);
5909ccaa
LT		 265 if (ret != -EACCES)
266 return ret;
1da177e4 267
d594e7ec
AV		 268 if (S_ISDIR(inode->i_mode)) {
269 /* DACs are overridable for directories */
1a48e2ac 270 if (inode_capable(inode, CAP_DAC_OVERRIDE))
d594e7ec
AV		 271 return 0;
272 if (!(mask & MAY_WRITE))
1a48e2ac 273 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
d594e7ec
AV		 274 return 0;
275 return -EACCES;
276 }
1da177e4
LT		 277 /*
278 * Read/write DACs are always overridable.
d594e7ec
AV		 279 * Executable DACs are overridable when there is
280 * at least one exec bit set.
1da177e4 281 */
d594e7ec 282 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
1a48e2ac 283 if (inode_capable(inode, CAP_DAC_OVERRIDE))
1da177e4
LT		 284 return 0;
285
286 /*
287 * Searching includes executable on directories, else just read.
288 */
7ea66001 289 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
d594e7ec 290 if (mask == MAY_READ)
1a48e2ac 291 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
1da177e4
LT		 292 return 0;
293
294 return -EACCES;
295}
296
3ddcd056
LT		 297/*
298 * We _really_ want to just do "generic_permission()" without
299 * even looking at the inode->i_op values. So we keep a cache
300 * flag in inode->i_opflags, that says "this has not special
301 * permission function, use the fast case".
302 */
303static inline int do_inode_permission(struct inode *inode, int mask)
304{
305 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
306 if (likely(inode->i_op->permission))
307 return inode->i_op->permission(inode, mask);
308
309 /* This gets set once for the inode lifetime */
310 spin_lock(&inode->i_lock);
311 inode->i_opflags |= IOP_FASTPERM;
312 spin_unlock(&inode->i_lock);
313 }
314 return generic_permission(inode, mask);
315}
316
cb23beb5
CH		 317/**
318 * inode_permission - check for access rights to a given inode
319 * @inode: inode to check permission on
8fd90c8d 320 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
cb23beb5
CH		 321 *
322 * Used to check for read/write/execute permissions on an inode.
323 * We use "fsuid" for this, letting us set arbitrary permissions
324 * for filesystem access without changing the "normal" uids which
325 * are used for other things.
948409c7
AG		 326 *
327 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
cb23beb5 328 */
f419a2e3 329int inode_permission(struct inode *inode, int mask)
1da177e4 330{
e6305c43 331 int retval;
1da177e4 332
3ddcd056 333 if (unlikely(mask & MAY_WRITE)) {
22590e41 334 umode_t mode = inode->i_mode;
1da177e4
LT		 335
336 /*
337 * Nobody gets write access to a read-only fs.
338 */
339 if (IS_RDONLY(inode) &&
340 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
341 return -EROFS;
342
343 /*
344 * Nobody gets write access to an immutable file.
345 */
346 if (IS_IMMUTABLE(inode))
347 return -EACCES;
348 }
349
3ddcd056 350 retval = do_inode_permission(inode, mask);
1da177e4
LT		 351 if (retval)
352 return retval;
353
08ce5f16
SH		 354 retval = devcgroup_inode_permission(inode, mask);
355 if (retval)
356 return retval;
357
d09ca739 358 return security_inode_permission(inode, mask);
1da177e4
LT		 359}
360
5dd784d0
JB		 361/**
362 * path_get - get a reference to a path
363 * @path: path to get the reference to
364 *
365 * Given a path increment the reference count to the dentry and the vfsmount.
366 */
367void path_get(struct path *path)
368{
369 mntget(path->mnt);
370 dget(path->dentry);
371}
372EXPORT_SYMBOL(path_get);
373
1d957f9b
JB		 374/**
375 * path_put - put a reference to a path
376 * @path: path to put the reference to
377 *
378 * Given a path decrement the reference count to the dentry and the vfsmount.
379 */
380void path_put(struct path *path)
1da177e4 381{
1d957f9b
JB		 382 dput(path->dentry);
383 mntput(path->mnt);
1da177e4 384}
1d957f9b 385EXPORT_SYMBOL(path_put);
1da177e4 386
19660af7 387/*
31e6b01f 388 * Path walking has 2 modes, rcu-walk and ref-walk (see
19660af7
AV		 389 * Documentation/filesystems/path-lookup.txt). In situations when we can't
390 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
391 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
392 * mode. Refcounts are grabbed at the last known good point before rcu-walk
393 * got stuck, so ref-walk may continue from there. If this is not successful
394 * (eg. a seqcount has changed), then failure is returned and it's up to caller
395 * to restart the path walk from the beginning in ref-walk mode.
31e6b01f 396 */
31e6b01f
NP		 397
398/**
19660af7
AV		 399 * unlazy_walk - try to switch to ref-walk mode.
400 * @nd: nameidata pathwalk data
401 * @dentry: child of nd->path.dentry or NULL
39191628 402 * Returns: 0 on success, -ECHILD on failure
31e6b01f 403 *
19660af7
AV		 404 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
405 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
406 * @nd or NULL. Must be called from rcu-walk context.
31e6b01f 407 */
19660af7 408static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
31e6b01f
NP		 409{
410 struct fs_struct *fs = current->fs;
411 struct dentry *parent = nd->path.dentry;
5b6ca027 412 int want_root = 0;
31e6b01f
NP		 413
414 BUG_ON(!(nd->flags & LOOKUP_RCU));
5b6ca027
AV		 415 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
416 want_root = 1;
31e6b01f
NP		 417 spin_lock(&fs->lock);
418 if (nd->root.mnt != fs->root.mnt ||
419 nd->root.dentry != fs->root.dentry)
420 goto err_root;
421 }
422 spin_lock(&parent->d_lock);
19660af7
AV		 423 if (!dentry) {
424 if (!__d_rcu_to_refcount(parent, nd->seq))
425 goto err_parent;
426 BUG_ON(nd->inode != parent->d_inode);
427 } else {
94c0d4ec
AV		 428 if (dentry->d_parent != parent)
429 goto err_parent;
19660af7
AV		 430 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
431 if (!__d_rcu_to_refcount(dentry, nd->seq))
432 goto err_child;
433 /*
434 * If the sequence check on the child dentry passed, then
435 * the child has not been removed from its parent. This
436 * means the parent dentry must be valid and able to take
437 * a reference at this point.
438 */
439 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
440 BUG_ON(!parent->d_count);
441 parent->d_count++;
442 spin_unlock(&dentry->d_lock);
443 }
31e6b01f 444 spin_unlock(&parent->d_lock);
5b6ca027 445 if (want_root) {
31e6b01f
NP		 446 path_get(&nd->root);
447 spin_unlock(&fs->lock);
448 }
449 mntget(nd->path.mnt);
450
451 rcu_read_unlock();
962830df 452 br_read_unlock(&vfsmount_lock);
31e6b01f
NP		 453 nd->flags &= ~LOOKUP_RCU;
454 return 0;
19660af7
AV		 455
456err_child:
31e6b01f 457 spin_unlock(&dentry->d_lock);
19660af7 458err_parent:
31e6b01f
NP		 459 spin_unlock(&parent->d_lock);
460err_root:
5b6ca027 461 if (want_root)
31e6b01f
NP		 462 spin_unlock(&fs->lock);
463 return -ECHILD;
464}
465
f60aef7e 466static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
34286d66 467{
f60aef7e 468 return dentry->d_op->d_revalidate(dentry, nd);
34286d66
NP		 469}
470
9f1fafee
AV		 471/**
472 * complete_walk - successful completion of path walk
473 * @nd: pointer nameidata
39159de2 474 *
9f1fafee
AV		 475 * If we had been in RCU mode, drop out of it and legitimize nd->path.
476 * Revalidate the final result, unless we'd already done that during
477 * the path walk or the filesystem doesn't ask for it. Return 0 on
478 * success, -error on failure. In case of failure caller does not
479 * need to drop nd->path.
39159de2 480 */
9f1fafee 481static int complete_walk(struct nameidata *nd)
39159de2 482{
16c2cd71 483 struct dentry *dentry = nd->path.dentry;
39159de2 484 int status;
39159de2 485
9f1fafee
AV		 486 if (nd->flags & LOOKUP_RCU) {
487 nd->flags &= ~LOOKUP_RCU;
488 if (!(nd->flags & LOOKUP_ROOT))
489 nd->root.mnt = NULL;
490 spin_lock(&dentry->d_lock);
491 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
492 spin_unlock(&dentry->d_lock);
493 rcu_read_unlock();
962830df 494 br_read_unlock(&vfsmount_lock);
9f1fafee
AV		 495 return -ECHILD;
496 }
497 BUG_ON(nd->inode != dentry->d_inode);
498 spin_unlock(&dentry->d_lock);
499 mntget(nd->path.mnt);
500 rcu_read_unlock();
962830df 501 br_read_unlock(&vfsmount_lock);
9f1fafee
AV		 502 }
503
16c2cd71
AV		 504 if (likely(!(nd->flags & LOOKUP_JUMPED)))
505 return 0;
506
507 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
39159de2
JL		 508 return 0;
509
16c2cd71
AV		 510 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
511 return 0;
512
513 /* Note: we do not d_invalidate() */
34286d66 514 status = d_revalidate(dentry, nd);
39159de2
JL		 515 if (status > 0)
516 return 0;
517
16c2cd71 518 if (!status)
39159de2 519 status = -ESTALE;
16c2cd71 520
9f1fafee 521 path_put(&nd->path);
39159de2
JL		 522 return status;
523}
524
2a737871
AV		 525static __always_inline void set_root(struct nameidata *nd)
526{
f7ad3c6b
MS		 527 if (!nd->root.mnt)
528 get_fs_root(current->fs, &nd->root);
2a737871
AV		 529}
530
6de88d72
AV		 531static int link_path_walk(const char *, struct nameidata *);
532
31e6b01f
NP		 533static __always_inline void set_root_rcu(struct nameidata *nd)
534{
535 if (!nd->root.mnt) {
536 struct fs_struct *fs = current->fs;
c28cc364
NP		 537 unsigned seq;
538
539 do {
540 seq = read_seqcount_begin(&fs->seq);
541 nd->root = fs->root;
c1530019 542 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
c28cc364 543 } while (read_seqcount_retry(&fs->seq, seq));
31e6b01f
NP		 544 }
545}
546
f1662356 547static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
1da177e4 548{
31e6b01f
NP		 549 int ret;
550
1da177e4
LT		 551 if (IS_ERR(link))
552 goto fail;
553
554 if (*link == '/') {
2a737871 555 set_root(nd);
1d957f9b 556 path_put(&nd->path);
2a737871
AV		 557 nd->path = nd->root;
558 path_get(&nd->root);
16c2cd71 559 nd->flags |= LOOKUP_JUMPED;
1da177e4 560 }
31e6b01f 561 nd->inode = nd->path.dentry->d_inode;
b4091d5f 562
31e6b01f
NP		 563 ret = link_path_walk(link, nd);
564 return ret;
1da177e4 565fail:
1d957f9b 566 path_put(&nd->path);
1da177e4
LT		 567 return PTR_ERR(link);
568}
569
1d957f9b 570static void path_put_conditional(struct path *path, struct nameidata *nd)
051d3812
IK		 571{
572 dput(path->dentry);
4ac91378 573 if (path->mnt != nd->path.mnt)
051d3812
IK		 574 mntput(path->mnt);
575}
576
7b9337aa
NP		 577static inline void path_to_nameidata(const struct path *path,
578 struct nameidata *nd)
051d3812 579{
31e6b01f
NP		 580 if (!(nd->flags & LOOKUP_RCU)) {
581 dput(nd->path.dentry);
582 if (nd->path.mnt != path->mnt)
583 mntput(nd->path.mnt);
9a229683 584 }
31e6b01f 585 nd->path.mnt = path->mnt;
4ac91378 586 nd->path.dentry = path->dentry;
051d3812
IK		 587}
588
574197e0
AV		 589static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
590{
591 struct inode *inode = link->dentry->d_inode;
6d7b5aae 592 if (inode->i_op->put_link)
574197e0
AV		 593 inode->i_op->put_link(link->dentry, nd, cookie);
594 path_put(link);
595}
596
def4af30 597static __always_inline int
574197e0 598follow_link(struct path *link, struct nameidata *nd, void **p)
1da177e4 599{
7b9337aa 600 struct dentry *dentry = link->dentry;
6d7b5aae
AV		 601 int error;
602 char *s;
1da177e4 603
844a3917
AV		 604 BUG_ON(nd->flags & LOOKUP_RCU);
605
0e794589
AV		 606 if (link->mnt == nd->path.mnt)
607 mntget(link->mnt);
608
6d7b5aae
AV		 609 error = -ELOOP;
610 if (unlikely(current->total_link_count >= 40))
611 goto out_put_nd_path;
612
574197e0
AV		 613 cond_resched();
614 current->total_link_count++;
615
68ac1234 616 touch_atime(link);
1da177e4 617 nd_set_link(nd, NULL);
cd4e91d3 618
36f3b4f6 619 error = security_inode_follow_link(link->dentry, nd);
6d7b5aae
AV		 620 if (error)
621 goto out_put_nd_path;
36f3b4f6 622
86acdca1 623 nd->last_type = LAST_BIND;
def4af30
AV		 624 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
625 error = PTR_ERR(*p);
6d7b5aae
AV		 626 if (IS_ERR(*p))
627 goto out_put_link;
628
629 error = 0;
630 s = nd_get_link(nd);
631 if (s) {
632 error = __vfs_follow_link(nd, s);
633 } else if (nd->last_type == LAST_BIND) {
634 nd->flags |= LOOKUP_JUMPED;
635 nd->inode = nd->path.dentry->d_inode;
636 if (nd->inode->i_op->follow_link) {
637 /* stepped on a _really_ weird one */
638 path_put(&nd->path);
639 error = -ELOOP;
bcda7652 640 }
1da177e4 641 }
6d7b5aae
AV		 642 if (unlikely(error))
643 put_link(nd, link, *p);
644
645 return error;
646
647out_put_nd_path:
648 path_put(&nd->path);
649out_put_link:
650 path_put(link);
1da177e4
LT		 651 return error;
652}
653
31e6b01f
NP		 654static int follow_up_rcu(struct path *path)
655{
0714a533
AV		 656 struct mount *mnt = real_mount(path->mnt);
657 struct mount *parent;
31e6b01f
NP		 658 struct dentry *mountpoint;
659
0714a533
AV		 660 parent = mnt->mnt_parent;
661 if (&parent->mnt == path->mnt)
31e6b01f 662 return 0;
a73324da 663 mountpoint = mnt->mnt_mountpoint;
31e6b01f 664 path->dentry = mountpoint;
0714a533 665 path->mnt = &parent->mnt;
31e6b01f
NP		 666 return 1;
667}
668
bab77ebf 669int follow_up(struct path *path)
1da177e4 670{
0714a533
AV		 671 struct mount *mnt = real_mount(path->mnt);
672 struct mount *parent;
1da177e4 673 struct dentry *mountpoint;
99b7db7b 674
962830df 675 br_read_lock(&vfsmount_lock);
0714a533
AV		 676 parent = mnt->mnt_parent;
677 if (&parent->mnt == path->mnt) {
962830df 678 br_read_unlock(&vfsmount_lock);
1da177e4
LT		 679 return 0;
680 }
0714a533 681 mntget(&parent->mnt);
a73324da 682 mountpoint = dget(mnt->mnt_mountpoint);
962830df 683 br_read_unlock(&vfsmount_lock);
bab77ebf
AV		 684 dput(path->dentry);
685 path->dentry = mountpoint;
686 mntput(path->mnt);
0714a533 687 path->mnt = &parent->mnt;
1da177e4
LT		 688 return 1;
689}
690
b5c84bf6 691/*
9875cf80
DH		 692 * Perform an automount
693 * - return -EISDIR to tell follow_managed() to stop and return the path we
694 * were called with.
1da177e4 695 */
9875cf80
DH		 696static int follow_automount(struct path *path, unsigned flags,
697 bool *need_mntput)
31e6b01f 698{
9875cf80 699 struct vfsmount *mnt;
ea5b778a 700 int err;
9875cf80
DH		 701
702 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
703 return -EREMOTE;
704
0ec26fd0
MS		 705 /* We don't want to mount if someone's just doing a stat -
706 * unless they're stat'ing a directory and appended a '/' to
707 * the name.
708 *
709 * We do, however, want to mount if someone wants to open or
710 * create a file of any type under the mountpoint, wants to
711 * traverse through the mountpoint or wants to open the
712 * mounted directory. Also, autofs may mark negative dentries
713 * as being automount points. These will need the attentions
714 * of the daemon to instantiate them before they can be used.
9875cf80 715 */
0ec26fd0 716 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
d94c177b 717 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
0ec26fd0
MS		 718 path->dentry->d_inode)
719 return -EISDIR;
720
9875cf80
DH		 721 current->total_link_count++;
722 if (current->total_link_count >= 40)
723 return -ELOOP;
724
725 mnt = path->dentry->d_op->d_automount(path);
726 if (IS_ERR(mnt)) {
727 /*
728 * The filesystem is allowed to return -EISDIR here to indicate
729 * it doesn't want to automount. For instance, autofs would do
730 * this so that its userspace daemon can mount on this dentry.
731 *
732 * However, we can only permit this if it's a terminal point in
733 * the path being looked up; if it wasn't then the remainder of
734 * the path is inaccessible and we should say so.
735 */
49084c3b 736 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
9875cf80
DH		 737 return -EREMOTE;
738 return PTR_ERR(mnt);
31e6b01f 739 }
ea5b778a 740
9875cf80
DH		 741 if (!mnt) /* mount collision */
742 return 0;
31e6b01f 743
8aef1884
AV		 744 if (!*need_mntput) {
745 /* lock_mount() may release path->mnt on error */
746 mntget(path->mnt);
747 *need_mntput = true;
748 }
19a167af 749 err = finish_automount(mnt, path);
9875cf80 750
ea5b778a
DH		 751 switch (err) {
752 case -EBUSY:
753 /* Someone else made a mount here whilst we were busy */
19a167af 754 return 0;
ea5b778a 755 case 0:
8aef1884 756 path_put(path);
ea5b778a
DH		 757 path->mnt = mnt;
758 path->dentry = dget(mnt->mnt_root);
ea5b778a 759 return 0;
19a167af
AV		 760 default:
761 return err;
ea5b778a 762 }
19a167af 763
463ffb2e
AV		 764}
765
9875cf80
DH		 766/*
767 * Handle a dentry that is managed in some way.
cc53ce53 768 * - Flagged for transit management (autofs)
9875cf80
DH		 769 * - Flagged as mountpoint
770 * - Flagged as automount point
771 *
772 * This may only be called in refwalk mode.
773 *
774 * Serialization is taken care of in namespace.c
775 */
776static int follow_managed(struct path *path, unsigned flags)
1da177e4 777{
8aef1884 778 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
9875cf80
DH		 779 unsigned managed;
780 bool need_mntput = false;
8aef1884 781 int ret = 0;
9875cf80
DH		 782
783 /* Given that we're not holding a lock here, we retain the value in a
784 * local variable for each dentry as we look at it so that we don't see
785 * the components of that value change under us */
786 while (managed = ACCESS_ONCE(path->dentry->d_flags),
787 managed &= DCACHE_MANAGED_DENTRY,
788 unlikely(managed != 0)) {
cc53ce53
DH		 789 /* Allow the filesystem to manage the transit without i_mutex
790 * being held. */
791 if (managed & DCACHE_MANAGE_TRANSIT) {
792 BUG_ON(!path->dentry->d_op);
793 BUG_ON(!path->dentry->d_op->d_manage);
1aed3e42 794 ret = path->dentry->d_op->d_manage(path->dentry, false);
cc53ce53 795 if (ret < 0)
8aef1884 796 break;
cc53ce53
DH		 797 }
798
9875cf80
DH		 799 /* Transit to a mounted filesystem. */
800 if (managed & DCACHE_MOUNTED) {
801 struct vfsmount *mounted = lookup_mnt(path);
802 if (mounted) {
803 dput(path->dentry);
804 if (need_mntput)
805 mntput(path->mnt);
806 path->mnt = mounted;
807 path->dentry = dget(mounted->mnt_root);
808 need_mntput = true;
809 continue;
810 }
811
812 /* Something is mounted on this dentry in another
813 * namespace and/or whatever was mounted there in this
814 * namespace got unmounted before we managed to get the
815 * vfsmount_lock */
816 }
817
818 /* Handle an automount point */
819 if (managed & DCACHE_NEED_AUTOMOUNT) {
820 ret = follow_automount(path, flags, &need_mntput);
821 if (ret < 0)
8aef1884 822 break;
9875cf80
DH		 823 continue;
824 }
825
826 /* We didn't change the current path point */
827 break;
1da177e4 828 }
8aef1884
AV		 829
830 if (need_mntput && path->mnt == mnt)
831 mntput(path->mnt);
832 if (ret == -EISDIR)
833 ret = 0;
a3fbbde7 834 return ret < 0 ? ret : need_mntput;
1da177e4
LT		 835}
836
cc53ce53 837int follow_down_one(struct path *path)
1da177e4
LT		 838{
839 struct vfsmount *mounted;
840
1c755af4 841 mounted = lookup_mnt(path);
1da177e4 842 if (mounted) {
9393bd07
AV		 843 dput(path->dentry);
844 mntput(path->mnt);
845 path->mnt = mounted;
846 path->dentry = dget(mounted->mnt_root);
1da177e4
LT		 847 return 1;
848 }
849 return 0;
850}
851
62a7375e
IK		 852static inline bool managed_dentry_might_block(struct dentry *dentry)
853{
854 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
855 dentry->d_op->d_manage(dentry, true) < 0);
856}
857
9875cf80 858/*
287548e4
AV		 859 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
860 * we meet a managed dentry that would need blocking.
9875cf80
DH		 861 */
862static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
287548e4 863 struct inode **inode)
9875cf80 864{
62a7375e 865 for (;;) {
c7105365 866 struct mount *mounted;
62a7375e
IK		 867 /*
868 * Don't forget we might have a non-mountpoint managed dentry
869 * that wants to block transit.
870 */
287548e4 871 if (unlikely(managed_dentry_might_block(path->dentry)))
ab90911f 872 return false;
62a7375e
IK		 873
874 if (!d_mountpoint(path->dentry))
875 break;
876
9875cf80
DH		 877 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
878 if (!mounted)
879 break;
c7105365
AV		 880 path->mnt = &mounted->mnt;
881 path->dentry = mounted->mnt.mnt_root;
a3fbbde7 882 nd->flags |= LOOKUP_JUMPED;
9875cf80 883 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
59430262
LT		 884 /*
885 * Update the inode too. We don't need to re-check the
886 * dentry sequence number here after this d_inode read,
887 * because a mount-point is always pinned.
888 */
889 *inode = path->dentry->d_inode;
9875cf80 890 }
9875cf80
DH		 891 return true;
892}
893
dea39376 894static void follow_mount_rcu(struct nameidata *nd)
287548e4 895{
dea39376 896 while (d_mountpoint(nd->path.dentry)) {
c7105365 897 struct mount *mounted;
dea39376 898 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
287548e4
AV		 899 if (!mounted)
900 break;
c7105365
AV		 901 nd->path.mnt = &mounted->mnt;
902 nd->path.dentry = mounted->mnt.mnt_root;
dea39376 903 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
287548e4
AV		 904 }
905}
906
31e6b01f
NP		 907static int follow_dotdot_rcu(struct nameidata *nd)
908{
31e6b01f
NP		 909 set_root_rcu(nd);
910
9875cf80 911 while (1) {
31e6b01f
NP		 912 if (nd->path.dentry == nd->root.dentry &&
913 nd->path.mnt == nd->root.mnt) {
914 break;
915 }
916 if (nd->path.dentry != nd->path.mnt->mnt_root) {
917 struct dentry *old = nd->path.dentry;
918 struct dentry *parent = old->d_parent;
919 unsigned seq;
920
921 seq = read_seqcount_begin(&parent->d_seq);
922 if (read_seqcount_retry(&old->d_seq, nd->seq))
ef7562d5 923 goto failed;
31e6b01f
NP		 924 nd->path.dentry = parent;
925 nd->seq = seq;
926 break;
927 }
928 if (!follow_up_rcu(&nd->path))
929 break;
930 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
31e6b01f 931 }
dea39376
AV		 932 follow_mount_rcu(nd);
933 nd->inode = nd->path.dentry->d_inode;
31e6b01f 934 return 0;
ef7562d5
AV		 935
936failed:
937 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 938 if (!(nd->flags & LOOKUP_ROOT))
939 nd->root.mnt = NULL;
ef7562d5 940 rcu_read_unlock();
962830df 941 br_read_unlock(&vfsmount_lock);
ef7562d5 942 return -ECHILD;
31e6b01f
NP		 943}
944
cc53ce53
DH		 945/*
946 * Follow down to the covering mount currently visible to userspace. At each
947 * point, the filesystem owning that dentry may be queried as to whether the
948 * caller is permitted to proceed or not.
cc53ce53 949 */
7cc90cc3 950int follow_down(struct path *path)
cc53ce53
DH		 951{
952 unsigned managed;
953 int ret;
954
955 while (managed = ACCESS_ONCE(path->dentry->d_flags),
956 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
957 /* Allow the filesystem to manage the transit without i_mutex
958 * being held.
959 *
960 * We indicate to the filesystem if someone is trying to mount
961 * something here. This gives autofs the chance to deny anyone
962 * other than its daemon the right to mount on its
963 * superstructure.
964 *
965 * The filesystem may sleep at this point.
966 */
967 if (managed & DCACHE_MANAGE_TRANSIT) {
968 BUG_ON(!path->dentry->d_op);
969 BUG_ON(!path->dentry->d_op->d_manage);
ab90911f 970 ret = path->dentry->d_op->d_manage(
1aed3e42 971 path->dentry, false);
cc53ce53
DH		 972 if (ret < 0)
973 return ret == -EISDIR ? 0 : ret;
974 }
975
976 /* Transit to a mounted filesystem. */
977 if (managed & DCACHE_MOUNTED) {
978 struct vfsmount *mounted = lookup_mnt(path);
979 if (!mounted)
980 break;
981 dput(path->dentry);
982 mntput(path->mnt);
983 path->mnt = mounted;
984 path->dentry = dget(mounted->mnt_root);
985 continue;
986 }
987
988 /* Don't handle automount points here */
989 break;
990 }
991 return 0;
992}
993
9875cf80
DH		 994/*
995 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
996 */
997static void follow_mount(struct path *path)
998{
999 while (d_mountpoint(path->dentry)) {
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 }
1008}
1009
31e6b01f 1010static void follow_dotdot(struct nameidata *nd)
1da177e4 1011{
2a737871 1012 set_root(nd);
e518ddb7 1013
1da177e4 1014 while(1) {
4ac91378 1015 struct dentry *old = nd->path.dentry;
1da177e4 1016
2a737871
AV		 1017 if (nd->path.dentry == nd->root.dentry &&
1018 nd->path.mnt == nd->root.mnt) {
1da177e4
LT		 1019 break;
1020 }
4ac91378 1021 if (nd->path.dentry != nd->path.mnt->mnt_root) {
3088dd70
AV		 1022 /* rare case of legitimate dget_parent()... */
1023 nd->path.dentry = dget_parent(nd->path.dentry);
1da177e4
LT		 1024 dput(old);
1025 break;
1026 }
3088dd70 1027 if (!follow_up(&nd->path))
1da177e4 1028 break;
1da177e4 1029 }
79ed0226 1030 follow_mount(&nd->path);
31e6b01f 1031 nd->inode = nd->path.dentry->d_inode;
1da177e4
LT		 1032}
1033
baa03890 1034/*
bad61189
MS		 1035 * This looks up the name in dcache, possibly revalidates the old dentry and
1036 * allocates a new one if not found or not valid. In the need_lookup argument
1037 * returns whether i_op->lookup is necessary.
1038 *
1039 * dir->d_inode->i_mutex must be held
baa03890 1040 */
bad61189
MS		 1041static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1042 struct nameidata *nd, bool *need_lookup)
baa03890 1043{
baa03890 1044 struct dentry *dentry;
bad61189 1045 int error;
baa03890 1046
bad61189
MS		 1047 *need_lookup = false;
1048 dentry = d_lookup(dir, name);
1049 if (dentry) {
1050 if (d_need_lookup(dentry)) {
1051 *need_lookup = true;
1052 } else if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1053 error = d_revalidate(dentry, nd);
1054 if (unlikely(error <= 0)) {
1055 if (error < 0) {
1056 dput(dentry);
1057 return ERR_PTR(error);
1058 } else if (!d_invalidate(dentry)) {
1059 dput(dentry);
1060 dentry = NULL;
1061 }
1062 }
1063 }
1064 }
baa03890 1065
bad61189
MS		 1066 if (!dentry) {
1067 dentry = d_alloc(dir, name);
1068 if (unlikely(!dentry))
1069 return ERR_PTR(-ENOMEM);
baa03890 1070
bad61189 1071 *need_lookup = true;
baa03890
NP		 1072 }
1073 return dentry;
1074}
1075
44396f4b 1076/*
bad61189
MS		 1077 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1078 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1079 *
1080 * dir->d_inode->i_mutex must be held
44396f4b 1081 */
bad61189
MS		 1082static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1083 struct nameidata *nd)
44396f4b 1084{
44396f4b
JB		 1085 struct dentry *old;
1086
1087 /* Don't create child dentry for a dead directory. */
bad61189 1088 if (unlikely(IS_DEADDIR(dir))) {
e188dc02 1089 dput(dentry);
44396f4b 1090 return ERR_PTR(-ENOENT);
e188dc02 1091 }
44396f4b 1092
bad61189 1093 old = dir->i_op->lookup(dir, dentry, nd);
44396f4b
JB		 1094 if (unlikely(old)) {
1095 dput(dentry);
1096 dentry = old;
1097 }
1098 return dentry;
1099}
1100
a3255546
AV		 1101static struct dentry *__lookup_hash(struct qstr *name,
1102 struct dentry *base, struct nameidata *nd)
1103{
bad61189 1104 bool need_lookup;
a3255546
AV		 1105 struct dentry *dentry;
1106
bad61189
MS		 1107 dentry = lookup_dcache(name, base, nd, &need_lookup);
1108 if (!need_lookup)
1109 return dentry;
a3255546 1110
bad61189 1111 return lookup_real(base->d_inode, dentry, nd);
a3255546
AV		 1112}
1113
1da177e4
LT		 1114/*
1115 * It's more convoluted than I'd like it to be, but... it's still fairly
1116 * small and for now I'd prefer to have fast path as straight as possible.
1117 * It _is_ time-critical.
1118 */
697f514d
MS		 1119static int lookup_fast(struct nameidata *nd, struct qstr *name,
1120 struct path *path, struct inode **inode)
1da177e4 1121{
4ac91378 1122 struct vfsmount *mnt = nd->path.mnt;
31e6b01f 1123 struct dentry *dentry, *parent = nd->path.dentry;
5a18fff2
AV		 1124 int need_reval = 1;
1125 int status = 1;
9875cf80
DH		 1126 int err;
1127
b04f784e
NP		 1128 /*
1129 * Rename seqlock is not required here because in the off chance
1130 * of a false negative due to a concurrent rename, we're going to
1131 * do the non-racy lookup, below.
1132 */
31e6b01f
NP		 1133 if (nd->flags & LOOKUP_RCU) {
1134 unsigned seq;
12f8ad4b 1135 dentry = __d_lookup_rcu(parent, name, &seq, nd->inode);
5a18fff2
AV		 1136 if (!dentry)
1137 goto unlazy;
1138
12f8ad4b
LT		 1139 /*
1140 * This sequence count validates that the inode matches
1141 * the dentry name information from lookup.
1142 */
1143 *inode = dentry->d_inode;
1144 if (read_seqcount_retry(&dentry->d_seq, seq))
1145 return -ECHILD;
1146
1147 /*
1148 * This sequence count validates that the parent had no
1149 * changes while we did the lookup of the dentry above.
1150 *
1151 * The memory barrier in read_seqcount_begin of child is
1152 * enough, we can use __read_seqcount_retry here.
1153 */
31e6b01f
NP		 1154 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1155 return -ECHILD;
31e6b01f 1156 nd->seq = seq;
5a18fff2 1157
fa4ee159
MS		 1158 if (unlikely(d_need_lookup(dentry)))
1159 goto unlazy;
24643087 1160 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
5a18fff2
AV		 1161 status = d_revalidate(dentry, nd);
1162 if (unlikely(status <= 0)) {
1163 if (status != -ECHILD)
1164 need_reval = 0;
1165 goto unlazy;
1166 }
24643087 1167 }
31e6b01f
NP		 1168 path->mnt = mnt;
1169 path->dentry = dentry;
d6e9bd25
AV		 1170 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1171 goto unlazy;
1172 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1173 goto unlazy;
1174 return 0;
5a18fff2 1175unlazy:
19660af7
AV		 1176 if (unlazy_walk(nd, dentry))
1177 return -ECHILD;
5a18fff2
AV		 1178 } else {
1179 dentry = __d_lookup(parent, name);
9875cf80 1180 }
5a18fff2 1181
81e6f520
AV		 1182 if (unlikely(!dentry))
1183 goto need_lookup;
1184
1185 if (unlikely(d_need_lookup(dentry))) {
44396f4b 1186 dput(dentry);
81e6f520 1187 goto need_lookup;
5a18fff2 1188 }
81e6f520 1189
5a18fff2
AV		 1190 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1191 status = d_revalidate(dentry, nd);
1192 if (unlikely(status <= 0)) {
1193 if (status < 0) {
1194 dput(dentry);
1195 return status;
1196 }
1197 if (!d_invalidate(dentry)) {
1198 dput(dentry);
81e6f520 1199 goto need_lookup;
5a18fff2 1200 }
24643087 1201 }
697f514d 1202
9875cf80
DH		 1203 path->mnt = mnt;
1204 path->dentry = dentry;
1205 err = follow_managed(path, nd->flags);
89312214
IK		 1206 if (unlikely(err < 0)) {
1207 path_put_conditional(path, nd);
9875cf80 1208 return err;
89312214 1209 }
a3fbbde7
AV		 1210 if (err)
1211 nd->flags |= LOOKUP_JUMPED;
9875cf80 1212 *inode = path->dentry->d_inode;
1da177e4 1213 return 0;
81e6f520
AV		 1214
1215need_lookup:
697f514d
MS		 1216 return 1;
1217}
1218
1219/* Fast lookup failed, do it the slow way */
1220static int lookup_slow(struct nameidata *nd, struct qstr *name,
1221 struct path *path)
1222{
1223 struct dentry *dentry, *parent;
1224 int err;
1225
1226 parent = nd->path.dentry;
81e6f520
AV		 1227 BUG_ON(nd->inode != parent->d_inode);
1228
1229 mutex_lock(&parent->d_inode->i_mutex);
1230 dentry = __lookup_hash(name, parent, nd);
1231 mutex_unlock(&parent->d_inode->i_mutex);
1232 if (IS_ERR(dentry))
1233 return PTR_ERR(dentry);
697f514d
MS		 1234 path->mnt = nd->path.mnt;
1235 path->dentry = dentry;
1236 err = follow_managed(path, nd->flags);
1237 if (unlikely(err < 0)) {
1238 path_put_conditional(path, nd);
1239 return err;
1240 }
1241 if (err)
1242 nd->flags |= LOOKUP_JUMPED;
1243 return 0;
1da177e4
LT		 1244}
1245
52094c8a
AV		 1246static inline int may_lookup(struct nameidata *nd)
1247{
1248 if (nd->flags & LOOKUP_RCU) {
4ad5abb3 1249 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
52094c8a
AV		 1250 if (err != -ECHILD)
1251 return err;
19660af7 1252 if (unlazy_walk(nd, NULL))
52094c8a
AV		 1253 return -ECHILD;
1254 }
4ad5abb3 1255 return inode_permission(nd->inode, MAY_EXEC);
52094c8a
AV		 1256}
1257
9856fa1b
AV		 1258static inline int handle_dots(struct nameidata *nd, int type)
1259{
1260 if (type == LAST_DOTDOT) {
1261 if (nd->flags & LOOKUP_RCU) {
1262 if (follow_dotdot_rcu(nd))
1263 return -ECHILD;
1264 } else
1265 follow_dotdot(nd);
1266 }
1267 return 0;
1268}
1269
951361f9
AV		 1270static void terminate_walk(struct nameidata *nd)
1271{
1272 if (!(nd->flags & LOOKUP_RCU)) {
1273 path_put(&nd->path);
1274 } else {
1275 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 1276 if (!(nd->flags & LOOKUP_ROOT))
1277 nd->root.mnt = NULL;
951361f9 1278 rcu_read_unlock();
962830df 1279 br_read_unlock(&vfsmount_lock);
951361f9
AV		 1280 }
1281}
1282
3ddcd056
LT		 1283/*
1284 * Do we need to follow links? We _really_ want to be able
1285 * to do this check without having to look at inode->i_op,
1286 * so we keep a cache of "no, this doesn't need follow_link"
1287 * for the common case.
1288 */
7813b94a 1289static inline int should_follow_link(struct inode *inode, int follow)
3ddcd056
LT		 1290{
1291 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1292 if (likely(inode->i_op->follow_link))
1293 return follow;
1294
1295 /* This gets set once for the inode lifetime */
1296 spin_lock(&inode->i_lock);
1297 inode->i_opflags |= IOP_NOFOLLOW;
1298 spin_unlock(&inode->i_lock);
1299 }
1300 return 0;
1301}
1302
ce57dfc1
AV		 1303static inline int walk_component(struct nameidata *nd, struct path *path,
1304 struct qstr *name, int type, int follow)
1305{
1306 struct inode *inode;
1307 int err;
1308 /*
1309 * "." and ".." are special - ".." especially so because it has
1310 * to be able to know about the current root directory and
1311 * parent relationships.
1312 */
1313 if (unlikely(type != LAST_NORM))
1314 return handle_dots(nd, type);
697f514d 1315 err = lookup_fast(nd, name, path, &inode);
ce57dfc1 1316 if (unlikely(err)) {
697f514d
MS		 1317 if (err < 0)
1318 goto out_err;
1319
1320 err = lookup_slow(nd, name, path);
1321 if (err < 0)
1322 goto out_err;
1323
1324 inode = path->dentry->d_inode;
ce57dfc1 1325 }
697f514d
MS		 1326 err = -ENOENT;
1327 if (!inode)
1328 goto out_path_put;
1329
7813b94a 1330 if (should_follow_link(inode, follow)) {
19660af7
AV		 1331 if (nd->flags & LOOKUP_RCU) {
1332 if (unlikely(unlazy_walk(nd, path->dentry))) {
697f514d
MS		 1333 err = -ECHILD;
1334 goto out_err;
19660af7
AV		 1335 }
1336 }
ce57dfc1
AV		 1337 BUG_ON(inode != path->dentry->d_inode);
1338 return 1;
1339 }
1340 path_to_nameidata(path, nd);
1341 nd->inode = inode;
1342 return 0;
697f514d
MS		 1343
1344out_path_put:
1345 path_to_nameidata(path, nd);
1346out_err:
1347 terminate_walk(nd);
1348 return err;
ce57dfc1
AV		 1349}
1350
b356379a
AV		 1351/*
1352 * This limits recursive symlink follows to 8, while
1353 * limiting consecutive symlinks to 40.
1354 *
1355 * Without that kind of total limit, nasty chains of consecutive
1356 * symlinks can cause almost arbitrarily long lookups.
1357 */
1358static inline int nested_symlink(struct path *path, struct nameidata *nd)
1359{
1360 int res;
1361
b356379a
AV		 1362 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1363 path_put_conditional(path, nd);
1364 path_put(&nd->path);
1365 return -ELOOP;
1366 }
1a4022f8 1367 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
b356379a
AV		 1368
1369 nd->depth++;
1370 current->link_count++;
1371
1372 do {
1373 struct path link = *path;
1374 void *cookie;
574197e0
AV		 1375
1376 res = follow_link(&link, nd, &cookie);
6d7b5aae
AV		 1377 if (res)
1378 break;
1379 res = walk_component(nd, path, &nd->last,
1380 nd->last_type, LOOKUP_FOLLOW);
574197e0 1381 put_link(nd, &link, cookie);
b356379a
AV		 1382 } while (res > 0);
1383
1384 current->link_count--;
1385 nd->depth--;
1386 return res;
1387}
1388
3ddcd056
LT		 1389/*
1390 * We really don't want to look at inode->i_op->lookup
1391 * when we don't have to. So we keep a cache bit in
1392 * the inode ->i_opflags field that says "yes, we can
1393 * do lookup on this inode".
1394 */
1395static inline int can_lookup(struct inode *inode)
1396{
1397 if (likely(inode->i_opflags & IOP_LOOKUP))
1398 return 1;
1399 if (likely(!inode->i_op->lookup))
1400 return 0;
1401
1402 /* We do this once for the lifetime of the inode */
1403 spin_lock(&inode->i_lock);
1404 inode->i_opflags |= IOP_LOOKUP;
1405 spin_unlock(&inode->i_lock);
1406 return 1;
1407}
1408
bfcfaa77
LT		 1409/*
1410 * We can do the critical dentry name comparison and hashing
1411 * operations one word at a time, but we are limited to:
1412 *
1413 * - Architectures with fast unaligned word accesses. We could
1414 * do a "get_unaligned()" if this helps and is sufficiently
1415 * fast.
1416 *
1417 * - Little-endian machines (so that we can generate the mask
1418 * of low bytes efficiently). Again, we *could* do a byte
1419 * swapping load on big-endian architectures if that is not
1420 * expensive enough to make the optimization worthless.
1421 *
1422 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1423 * do not trap on the (extremely unlikely) case of a page
1424 * crossing operation.
1425 *
1426 * - Furthermore, we need an efficient 64-bit compile for the
1427 * 64-bit case in order to generate the "number of bytes in
1428 * the final mask". Again, that could be replaced with a
1429 * efficient population count instruction or similar.
1430 */
1431#ifdef CONFIG_DCACHE_WORD_ACCESS
1432
f68e556e 1433#include <asm/word-at-a-time.h>
bfcfaa77 1434
f68e556e 1435#ifdef CONFIG_64BIT
bfcfaa77
LT		 1436
1437static inline unsigned int fold_hash(unsigned long hash)
1438{
1439 hash += hash >> (8*sizeof(int));
1440 return hash;
1441}
1442
1443#else /* 32-bit case */
1444
bfcfaa77
LT		 1445#define fold_hash(x) (x)
1446
1447#endif
1448
1449unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1450{
1451 unsigned long a, mask;
1452 unsigned long hash = 0;
1453
1454 for (;;) {
e419b4cc 1455 a = load_unaligned_zeropad(name);
bfcfaa77
LT		 1456 if (len < sizeof(unsigned long))
1457 break;
1458 hash += a;
f132c5be 1459 hash *= 9;
bfcfaa77
LT		 1460 name += sizeof(unsigned long);
1461 len -= sizeof(unsigned long);
1462 if (!len)
1463 goto done;
1464 }
1465 mask = ~(~0ul << len*8);
1466 hash += mask & a;
1467done:
1468 return fold_hash(hash);
1469}
1470EXPORT_SYMBOL(full_name_hash);
1471
bfcfaa77
LT		 1472/*
1473 * Calculate the length and hash of the path component, and
1474 * return the length of the component;
1475 */
1476static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1477{
36126f8f
LT		 1478 unsigned long a, b, adata, bdata, mask, hash, len;
1479 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
bfcfaa77
LT		 1480
1481 hash = a = 0;
1482 len = -sizeof(unsigned long);
1483 do {
1484 hash = (hash + a) * 9;
1485 len += sizeof(unsigned long);
e419b4cc 1486 a = load_unaligned_zeropad(name+len);
36126f8f
LT		 1487 b = a ^ REPEAT_BYTE('/');
1488 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1489
1490 adata = prep_zero_mask(a, adata, &constants);
1491 bdata = prep_zero_mask(b, bdata, &constants);
1492
1493 mask = create_zero_mask(adata | bdata);
1494
1495 hash += a & zero_bytemask(mask);
bfcfaa77
LT		 1496 *hashp = fold_hash(hash);
1497
36126f8f 1498 return len + find_zero(mask);
bfcfaa77
LT		 1499}
1500
1501#else
1502
0145acc2
LT		 1503unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1504{
1505 unsigned long hash = init_name_hash();
1506 while (len--)
1507 hash = partial_name_hash(*name++, hash);
1508 return end_name_hash(hash);
1509}
ae942ae7 1510EXPORT_SYMBOL(full_name_hash);
0145acc2 1511
200e9ef7
LT		 1512/*
1513 * We know there's a real path component here of at least
1514 * one character.
1515 */
1516static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1517{
1518 unsigned long hash = init_name_hash();
1519 unsigned long len = 0, c;
1520
1521 c = (unsigned char)*name;
1522 do {
1523 len++;
1524 hash = partial_name_hash(c, hash);
1525 c = (unsigned char)name[len];
1526 } while (c && c != '/');
1527 *hashp = end_name_hash(hash);
1528 return len;
1529}
1530
bfcfaa77
LT		 1531#endif
1532
1da177e4
LT		 1533/*
1534 * Name resolution.
ea3834d9
PM		 1535 * This is the basic name resolution function, turning a pathname into
1536 * the final dentry. We expect 'base' to be positive and a directory.
1da177e4 1537 *
ea3834d9
PM		 1538 * Returns 0 and nd will have valid dentry and mnt on success.
1539 * Returns error and drops reference to input namei data on failure.
1da177e4 1540 */
6de88d72 1541static int link_path_walk(const char *name, struct nameidata *nd)
1da177e4
LT		 1542{
1543 struct path next;
1da177e4 1544 int err;
1da177e4
LT		 1545
1546 while (*name=='/')
1547 name++;
1548 if (!*name)
086e183a 1549 return 0;
1da177e4 1550
1da177e4
LT		 1551 /* At this point we know we have a real path component. */
1552 for(;;) {
1da177e4 1553 struct qstr this;
200e9ef7 1554 long len;
fe479a58 1555 int type;
1da177e4 1556
52094c8a 1557 err = may_lookup(nd);
1da177e4
LT		 1558 if (err)
1559 break;
1560
200e9ef7 1561 len = hash_name(name, &this.hash);
1da177e4 1562 this.name = name;
200e9ef7 1563 this.len = len;
1da177e4 1564
fe479a58 1565 type = LAST_NORM;
200e9ef7 1566 if (name[0] == '.') switch (len) {
fe479a58 1567 case 2:
200e9ef7 1568 if (name[1] == '.') {
fe479a58 1569 type = LAST_DOTDOT;
16c2cd71
AV		 1570 nd->flags |= LOOKUP_JUMPED;
1571 }
fe479a58
AV		 1572 break;
1573 case 1:
1574 type = LAST_DOT;
1575 }
5a202bcd
AV		 1576 if (likely(type == LAST_NORM)) {
1577 struct dentry *parent = nd->path.dentry;
16c2cd71 1578 nd->flags &= ~LOOKUP_JUMPED;
5a202bcd
AV		 1579 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1580 err = parent->d_op->d_hash(parent, nd->inode,
1581 &this);
1582 if (err < 0)
1583 break;
1584 }
1585 }
fe479a58 1586
200e9ef7 1587 if (!name[len])
1da177e4 1588 goto last_component;
200e9ef7
LT		 1589 /*
1590 * If it wasn't NUL, we know it was '/'. Skip that
1591 * slash, and continue until no more slashes.
1592 */
1593 do {
1594 len++;
1595 } while (unlikely(name[len] == '/'));
1596 if (!name[len])
b356379a 1597 goto last_component;
200e9ef7 1598 name += len;
1da177e4 1599
ce57dfc1
AV		 1600 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1601 if (err < 0)
1602 return err;
1da177e4 1603
ce57dfc1 1604 if (err) {
b356379a 1605 err = nested_symlink(&next, nd);
1da177e4 1606 if (err)
a7472bab 1607 return err;
31e6b01f 1608 }
3ddcd056
LT		 1609 if (can_lookup(nd->inode))
1610 continue;
1da177e4 1611 err = -ENOTDIR;
3ddcd056 1612 break;
1da177e4
LT		 1613 /* here ends the main loop */
1614
1da177e4 1615last_component:
b356379a
AV		 1616 nd->last = this;
1617 nd->last_type = type;
086e183a 1618 return 0;
1da177e4 1619 }
951361f9 1620 terminate_walk(nd);
1da177e4
LT		 1621 return err;
1622}
1623
70e9b357
AV		 1624static int path_init(int dfd, const char *name, unsigned int flags,
1625 struct nameidata *nd, struct file **fp)
31e6b01f
NP		 1626{
1627 int retval = 0;
1628 int fput_needed;
1629 struct file *file;
1630
1631 nd->last_type = LAST_ROOT; /* if there are only slashes... */
16c2cd71 1632 nd->flags = flags | LOOKUP_JUMPED;
31e6b01f 1633 nd->depth = 0;
5b6ca027
AV		 1634 if (flags & LOOKUP_ROOT) {
1635 struct inode *inode = nd->root.dentry->d_inode;
73d049a4
AV		 1636 if (*name) {
1637 if (!inode->i_op->lookup)
1638 return -ENOTDIR;
1639 retval = inode_permission(inode, MAY_EXEC);
1640 if (retval)
1641 return retval;
1642 }
5b6ca027
AV		 1643 nd->path = nd->root;
1644 nd->inode = inode;
1645 if (flags & LOOKUP_RCU) {
962830df 1646 br_read_lock(&vfsmount_lock);
5b6ca027
AV		 1647 rcu_read_lock();
1648 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1649 } else {
1650 path_get(&nd->path);
1651 }
1652 return 0;
1653 }
1654
31e6b01f 1655 nd->root.mnt = NULL;
31e6b01f
NP		 1656
1657 if (*name=='/') {
e41f7d4e 1658 if (flags & LOOKUP_RCU) {
962830df 1659 br_read_lock(&vfsmount_lock);
e41f7d4e
AV		 1660 rcu_read_lock();
1661 set_root_rcu(nd);
1662 } else {
1663 set_root(nd);
1664 path_get(&nd->root);
1665 }
1666 nd->path = nd->root;
31e6b01f 1667 } else if (dfd == AT_FDCWD) {
e41f7d4e
AV		 1668 if (flags & LOOKUP_RCU) {
1669 struct fs_struct *fs = current->fs;
1670 unsigned seq;
31e6b01f 1671
962830df 1672 br_read_lock(&vfsmount_lock);
e41f7d4e 1673 rcu_read_lock();
c28cc364 1674
e41f7d4e
AV		 1675 do {
1676 seq = read_seqcount_begin(&fs->seq);
1677 nd->path = fs->pwd;
1678 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1679 } while (read_seqcount_retry(&fs->seq, seq));
1680 } else {
1681 get_fs_pwd(current->fs, &nd->path);
1682 }
31e6b01f
NP		 1683 } else {
1684 struct dentry *dentry;
1685
1abf0c71 1686 file = fget_raw_light(dfd, &fput_needed);
31e6b01f
NP		 1687 retval = -EBADF;
1688 if (!file)
1689 goto out_fail;
1690
1691 dentry = file->f_path.dentry;
1692
f52e0c11
AV		 1693 if (*name) {
1694 retval = -ENOTDIR;
1695 if (!S_ISDIR(dentry->d_inode->i_mode))
1696 goto fput_fail;
31e6b01f 1697
4ad5abb3 1698 retval = inode_permission(dentry->d_inode, MAY_EXEC);
f52e0c11
AV		 1699 if (retval)
1700 goto fput_fail;
1701 }
31e6b01f
NP		 1702
1703 nd->path = file->f_path;
e41f7d4e
AV		 1704 if (flags & LOOKUP_RCU) {
1705 if (fput_needed)
70e9b357 1706 *fp = file;
e41f7d4e 1707 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
962830df 1708 br_read_lock(&vfsmount_lock);
e41f7d4e
AV		 1709 rcu_read_lock();
1710 } else {
1711 path_get(&file->f_path);
1712 fput_light(file, fput_needed);
1713 }
31e6b01f 1714 }
31e6b01f 1715
31e6b01f 1716 nd->inode = nd->path.dentry->d_inode;
9b4a9b14 1717 return 0;
2dfdd266 1718
9b4a9b14
AV		 1719fput_fail:
1720 fput_light(file, fput_needed);
1721out_fail:
1722 return retval;
1723}
1724
bd92d7fe
AV		 1725static inline int lookup_last(struct nameidata *nd, struct path *path)
1726{
1727 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1728 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1729
1730 nd->flags &= ~LOOKUP_PARENT;
1731 return walk_component(nd, path, &nd->last, nd->last_type,
1732 nd->flags & LOOKUP_FOLLOW);
1733}
1734
9b4a9b14 1735/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
ee0827cd 1736static int path_lookupat(int dfd, const char *name,
9b4a9b14
AV		 1737 unsigned int flags, struct nameidata *nd)
1738{
70e9b357 1739 struct file *base = NULL;
bd92d7fe
AV		 1740 struct path path;
1741 int err;
31e6b01f
NP		 1742
1743 /*
1744 * Path walking is largely split up into 2 different synchronisation
1745 * schemes, rcu-walk and ref-walk (explained in
1746 * Documentation/filesystems/path-lookup.txt). These share much of the
1747 * path walk code, but some things particularly setup, cleanup, and
1748 * following mounts are sufficiently divergent that functions are
1749 * duplicated. Typically there is a function foo(), and its RCU
1750 * analogue, foo_rcu().
1751 *
1752 * -ECHILD is the error number of choice (just to avoid clashes) that
1753 * is returned if some aspect of an rcu-walk fails. Such an error must
1754 * be handled by restarting a traditional ref-walk (which will always
1755 * be able to complete).
1756 */
bd92d7fe 1757 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
ee0827cd 1758
bd92d7fe
AV		 1759 if (unlikely(err))
1760 return err;
ee0827cd
AV		 1761
1762 current->total_link_count = 0;
bd92d7fe
AV		 1763 err = link_path_walk(name, nd);
1764
1765 if (!err && !(flags & LOOKUP_PARENT)) {
bd92d7fe
AV		 1766 err = lookup_last(nd, &path);
1767 while (err > 0) {
1768 void *cookie;
1769 struct path link = path;
bd92d7fe 1770 nd->flags |= LOOKUP_PARENT;
574197e0 1771 err = follow_link(&link, nd, &cookie);
6d7b5aae
AV		 1772 if (err)
1773 break;
1774 err = lookup_last(nd, &path);
574197e0 1775 put_link(nd, &link, cookie);
bd92d7fe
AV		 1776 }
1777 }
ee0827cd 1778
9f1fafee
AV		 1779 if (!err)
1780 err = complete_walk(nd);
bd92d7fe
AV		 1781
1782 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1783 if (!nd->inode->i_op->lookup) {
1784 path_put(&nd->path);
bd23a539 1785 err = -ENOTDIR;
bd92d7fe
AV		 1786 }
1787 }
16c2cd71 1788
70e9b357
AV		 1789 if (base)
1790 fput(base);
ee0827cd 1791
5b6ca027 1792 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
2a737871
AV		 1793 path_put(&nd->root);
1794 nd->root.mnt = NULL;
1795 }
bd92d7fe 1796 return err;
ee0827cd 1797}
31e6b01f 1798
ee0827cd
AV		 1799static int do_path_lookup(int dfd, const char *name,
1800 unsigned int flags, struct nameidata *nd)
1801{
1802 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1803 if (unlikely(retval == -ECHILD))
1804 retval = path_lookupat(dfd, name, flags, nd);
1805 if (unlikely(retval == -ESTALE))
1806 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
31e6b01f
NP		 1807
1808 if (likely(!retval)) {
1809 if (unlikely(!audit_dummy_context())) {
1810 if (nd->path.dentry && nd->inode)
1811 audit_inode(name, nd->path.dentry);
1812 }
1813 }
170aa3d0 1814 return retval;
1da177e4
LT		 1815}
1816
c9c6cac0 1817int kern_path_parent(const char *name, struct nameidata *nd)
5590ff0d 1818{
c9c6cac0 1819 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
5590ff0d
UD		 1820}
1821
d1811465
AV		 1822int kern_path(const char *name, unsigned int flags, struct path *path)
1823{
1824 struct nameidata nd;
1825 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1826 if (!res)
1827 *path = nd.path;
1828 return res;
1829}
1830
16f18200
JJS		 1831/**
1832 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1833 * @dentry: pointer to dentry of the base directory
1834 * @mnt: pointer to vfs mount of the base directory
1835 * @name: pointer to file name
1836 * @flags: lookup flags
e0a01249 1837 * @path: pointer to struct path to fill
16f18200
JJS		 1838 */
1839int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1840 const char *name, unsigned int flags,
e0a01249 1841 struct path *path)
16f18200 1842{
e0a01249
AV		 1843 struct nameidata nd;
1844 int err;
1845 nd.root.dentry = dentry;
1846 nd.root.mnt = mnt;
1847 BUG_ON(flags & LOOKUP_PARENT);
5b6ca027 1848 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
e0a01249
AV		 1849 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1850 if (!err)
1851 *path = nd.path;
1852 return err;
16f18200
JJS		 1853}
1854
057f6c01
JM		 1855/*
1856 * Restricted form of lookup. Doesn't follow links, single-component only,
1857 * needs parent already locked. Doesn't follow mounts.
1858 * SMP-safe.
1859 */
eead1911 1860static struct dentry *lookup_hash(struct nameidata *nd)
057f6c01 1861{
4ac91378 1862 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1da177e4
LT		 1863}
1864
eead1911 1865/**
a6b91919 1866 * lookup_one_len - filesystem helper to lookup single pathname component
eead1911
CH		 1867 * @name: pathname component to lookup
1868 * @base: base directory to lookup from
1869 * @len: maximum length @len should be interpreted to
1870 *
a6b91919
RD		 1871 * Note that this routine is purely a helper for filesystem usage and should
1872 * not be called by generic code. Also note that by using this function the
eead1911
CH		 1873 * nameidata argument is passed to the filesystem methods and a filesystem
1874 * using this helper needs to be prepared for that.
1875 */
057f6c01
JM		 1876struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1877{
057f6c01 1878 struct qstr this;
6a96ba54 1879 unsigned int c;
cda309de 1880 int err;
057f6c01 1881
2f9092e1
DW		 1882 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1883
6a96ba54
AV		 1884 this.name = name;
1885 this.len = len;
0145acc2 1886 this.hash = full_name_hash(name, len);
6a96ba54
AV		 1887 if (!len)
1888 return ERR_PTR(-EACCES);
1889
6a96ba54
AV		 1890 while (len--) {
1891 c = *(const unsigned char *)name++;
1892 if (c == '/' || c == '\0')
1893 return ERR_PTR(-EACCES);
6a96ba54 1894 }
5a202bcd
AV		 1895 /*
1896 * See if the low-level filesystem might want
1897 * to use its own hash..
1898 */
1899 if (base->d_flags & DCACHE_OP_HASH) {
1900 int err = base->d_op->d_hash(base, base->d_inode, &this);
1901 if (err < 0)
1902 return ERR_PTR(err);
1903 }
eead1911 1904
cda309de
MS		 1905 err = inode_permission(base->d_inode, MAY_EXEC);
1906 if (err)
1907 return ERR_PTR(err);
1908
49705b77 1909 return __lookup_hash(&this, base, NULL);
057f6c01
JM		 1910}
1911
1fa1e7f6
AW		 1912int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
1913 struct path *path, int *empty)
1da177e4 1914{
2d8f3038 1915 struct nameidata nd;
1fa1e7f6 1916 char *tmp = getname_flags(name, flags, empty);
1da177e4 1917 int err = PTR_ERR(tmp);
1da177e4 1918 if (!IS_ERR(tmp)) {
2d8f3038
AV		 1919
1920 BUG_ON(flags & LOOKUP_PARENT);
1921
1922 err = do_path_lookup(dfd, tmp, flags, &nd);
1da177e4 1923 putname(tmp);
2d8f3038
AV		 1924 if (!err)
1925 *path = nd.path;
1da177e4
LT		 1926 }
1927 return err;
1928}
1929
1fa1e7f6
AW		 1930int user_path_at(int dfd, const char __user *name, unsigned flags,
1931 struct path *path)
1932{
f7493e5d 1933 return user_path_at_empty(dfd, name, flags, path, NULL);
1fa1e7f6
AW		 1934}
1935
2ad94ae6
AV		 1936static int user_path_parent(int dfd, const char __user *path,
1937 struct nameidata *nd, char **name)
1938{
1939 char *s = getname(path);
1940 int error;
1941
1942 if (IS_ERR(s))
1943 return PTR_ERR(s);
1944
1945 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1946 if (error)
1947 putname(s);
1948 else
1949 *name = s;
1950
1951 return error;
1952}
1953
1da177e4
LT		 1954/*
1955 * It's inline, so penalty for filesystems that don't use sticky bit is
1956 * minimal.
1957 */
1958static inline int check_sticky(struct inode *dir, struct inode *inode)
1959{
8e96e3b7 1960 kuid_t fsuid = current_fsuid();
da9592ed 1961
1da177e4
LT		 1962 if (!(dir->i_mode & S_ISVTX))
1963 return 0;
8e96e3b7 1964 if (uid_eq(inode->i_uid, fsuid))
1da177e4 1965 return 0;
8e96e3b7 1966 if (uid_eq(dir->i_uid, fsuid))
1da177e4 1967 return 0;
1a48e2ac 1968 return !inode_capable(inode, CAP_FOWNER);
1da177e4
LT		 1969}
1970
1971/*
1972 * Check whether we can remove a link victim from directory dir, check
1973 * whether the type of victim is right.
1974 * 1. We can't do it if dir is read-only (done in permission())
1975 * 2. We should have write and exec permissions on dir
1976 * 3. We can't remove anything from append-only dir
1977 * 4. We can't do anything with immutable dir (done in permission())
1978 * 5. If the sticky bit on dir is set we should either
1979 * a. be owner of dir, or
1980 * b. be owner of victim, or
1981 * c. have CAP_FOWNER capability
1982 * 6. If the victim is append-only or immutable we can't do antyhing with
1983 * links pointing to it.
1984 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1985 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1986 * 9. We can't remove a root or mountpoint.
1987 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1988 * nfs_async_unlink().
1989 */
858119e1 1990static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1da177e4
LT		 1991{
1992 int error;
1993
1994 if (!victim->d_inode)
1995 return -ENOENT;
1996
1997 BUG_ON(victim->d_parent->d_inode != dir);
cccc6bba 1998 audit_inode_child(victim, dir);
1da177e4 1999
f419a2e3 2000 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 2001 if (error)
2002 return error;
2003 if (IS_APPEND(dir))
2004 return -EPERM;
2005 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
f9454548 2006 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1da177e4
LT		 2007 return -EPERM;
2008 if (isdir) {
2009 if (!S_ISDIR(victim->d_inode->i_mode))
2010 return -ENOTDIR;
2011 if (IS_ROOT(victim))
2012 return -EBUSY;
2013 } else if (S_ISDIR(victim->d_inode->i_mode))
2014 return -EISDIR;
2015 if (IS_DEADDIR(dir))
2016 return -ENOENT;
2017 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2018 return -EBUSY;
2019 return 0;
2020}
2021
2022/* Check whether we can create an object with dentry child in directory
2023 * dir.
2024 * 1. We can't do it if child already exists (open has special treatment for
2025 * this case, but since we are inlined it's OK)
2026 * 2. We can't do it if dir is read-only (done in permission())
2027 * 3. We should have write and exec permissions on dir
2028 * 4. We can't do it if dir is immutable (done in permission())
2029 */
a95164d9 2030static inline int may_create(struct inode *dir, struct dentry *child)
1da177e4
LT		 2031{
2032 if (child->d_inode)
2033 return -EEXIST;
2034 if (IS_DEADDIR(dir))
2035 return -ENOENT;
f419a2e3 2036 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 2037}
2038
1da177e4
LT		 2039/*
2040 * p1 and p2 should be directories on the same fs.
2041 */
2042struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2043{
2044 struct dentry *p;
2045
2046 if (p1 == p2) {
f2eace23 2047 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1da177e4
LT		 2048 return NULL;
2049 }
2050
a11f3a05 2051 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4 2052
e2761a11
OH		 2053 p = d_ancestor(p2, p1);
2054 if (p) {
2055 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2056 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2057 return p;
1da177e4
LT		 2058 }
2059
e2761a11
OH		 2060 p = d_ancestor(p1, p2);
2061 if (p) {
2062 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2063 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2064 return p;
1da177e4
LT		 2065 }
2066
f2eace23
IM		 2067 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2068 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1da177e4
LT		 2069 return NULL;
2070}
2071
2072void unlock_rename(struct dentry *p1, struct dentry *p2)
2073{
1b1dcc1b 2074 mutex_unlock(&p1->d_inode->i_mutex);
1da177e4 2075 if (p1 != p2) {
1b1dcc1b 2076 mutex_unlock(&p2->d_inode->i_mutex);
a11f3a05 2077 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4
LT		 2078 }
2079}
2080
4acdaf27 2081int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
1da177e4
LT		 2082 struct nameidata *nd)
2083{
a95164d9 2084 int error = may_create(dir, dentry);
1da177e4
LT		 2085
2086 if (error)
2087 return error;
2088
acfa4380 2089 if (!dir->i_op->create)
1da177e4
LT		 2090 return -EACCES; /* shouldn't it be ENOSYS? */
2091 mode &= S_IALLUGO;
2092 mode |= S_IFREG;
2093 error = security_inode_create(dir, dentry, mode);
2094 if (error)
2095 return error;
1da177e4 2096 error = dir->i_op->create(dir, dentry, mode, nd);
a74574aa 2097 if (!error)
f38aa942 2098 fsnotify_create(dir, dentry);
1da177e4
LT		 2099 return error;
2100}
2101
73d049a4 2102static int may_open(struct path *path, int acc_mode, int flag)
1da177e4 2103{
3fb64190 2104 struct dentry *dentry = path->dentry;
1da177e4
LT		 2105 struct inode *inode = dentry->d_inode;
2106 int error;
2107
bcda7652
AV		 2108 /* O_PATH? */
2109 if (!acc_mode)
2110 return 0;
2111
1da177e4
LT		 2112 if (!inode)
2113 return -ENOENT;
2114
c8fe8f30
CH		 2115 switch (inode->i_mode & S_IFMT) {
2116 case S_IFLNK:
1da177e4 2117 return -ELOOP;
c8fe8f30
CH		 2118 case S_IFDIR:
2119 if (acc_mode & MAY_WRITE)
2120 return -EISDIR;
2121 break;
2122 case S_IFBLK:
2123 case S_IFCHR:
3fb64190 2124 if (path->mnt->mnt_flags & MNT_NODEV)
1da177e4 2125 return -EACCES;
c8fe8f30
CH		 2126 /*FALLTHRU*/
2127 case S_IFIFO:
2128 case S_IFSOCK:
1da177e4 2129 flag &= ~O_TRUNC;
c8fe8f30 2130 break;
4a3fd211 2131 }
b41572e9 2132
3fb64190 2133 error = inode_permission(inode, acc_mode);
b41572e9
DH		 2134 if (error)
2135 return error;
6146f0d5 2136
1da177e4
LT		 2137 /*
2138 * An append-only file must be opened in append mode for writing.
2139 */
2140 if (IS_APPEND(inode)) {
8737c930 2141 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
7715b521 2142 return -EPERM;
1da177e4 2143 if (flag & O_TRUNC)
7715b521 2144 return -EPERM;
1da177e4
LT		 2145 }
2146
2147 /* O_NOATIME can only be set by the owner or superuser */
2e149670 2148 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
7715b521 2149 return -EPERM;
1da177e4 2150
f3c7691e 2151 return 0;
7715b521 2152}
1da177e4 2153
e1181ee6 2154static int handle_truncate(struct file *filp)
7715b521 2155{
e1181ee6 2156 struct path *path = &filp->f_path;
7715b521
AV		 2157 struct inode *inode = path->dentry->d_inode;
2158 int error = get_write_access(inode);
2159 if (error)
2160 return error;
2161 /*
2162 * Refuse to truncate files with mandatory locks held on them.
2163 */
2164 error = locks_verify_locked(inode);
2165 if (!error)
ea0d3ab2 2166 error = security_path_truncate(path);
7715b521
AV		 2167 if (!error) {
2168 error = do_truncate(path->dentry, 0,
2169 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
e1181ee6 2170 filp);
7715b521
AV		 2171 }
2172 put_write_access(inode);
acd0c935 2173 return error;
1da177e4
LT		 2174}
2175
d57999e1
DH		 2176static inline int open_to_namei_flags(int flag)
2177{
8a5e929d
AV		 2178 if ((flag & O_ACCMODE) == 3)
2179 flag--;
d57999e1
DH		 2180 return flag;
2181}
2182
d18e9008
MS		 2183static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2184{
2185 int error = security_path_mknod(dir, dentry, mode, 0);
2186 if (error)
2187 return error;
2188
2189 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2190 if (error)
2191 return error;
2192
2193 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2194}
2195
2196static struct file *atomic_open(struct nameidata *nd, struct dentry *dentry,
015c3bbc
MS		 2197 struct path *path, struct opendata *od,
2198 const struct open_flags *op,
77d660a8 2199 bool *want_write, bool need_lookup,
d18e9008
MS		 2200 bool *created)
2201{
2202 struct inode *dir = nd->path.dentry->d_inode;
2203 unsigned open_flag = open_to_namei_flags(op->open_flag);
2204 umode_t mode;
2205 int error;
2206 int acc_mode;
d18e9008
MS		 2207 struct file *filp;
2208 int create_error = 0;
2209 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2210
2211 BUG_ON(dentry->d_inode);
2212
2213 /* Don't create child dentry for a dead directory. */
2214 if (unlikely(IS_DEADDIR(dir))) {
2215 filp = ERR_PTR(-ENOENT);
2216 goto out;
2217 }
2218
2219 mode = op->mode & S_IALLUGO;
2220 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2221 mode &= ~current_umask();
2222
2223 if (open_flag & O_EXCL) {
2224 open_flag &= ~O_TRUNC;
2225 *created = true;
2226 }
2227
2228 /*
2229 * Checking write permission is tricky, bacuse we don't know if we are
2230 * going to actually need it: O_CREAT opens should work as long as the
2231 * file exists. But checking existence breaks atomicity. The trick is
2232 * to check access and if not granted clear O_CREAT from the flags.
2233 *
2234 * Another problem is returing the "right" error value (e.g. for an
2235 * O_EXCL open we want to return EEXIST not EROFS).
2236 */
2237 if ((open_flag & (O_CREAT | O_TRUNC)) ||
2238 (open_flag & O_ACCMODE) != O_RDONLY) {
2239 error = mnt_want_write(nd->path.mnt);
2240 if (!error) {
77d660a8 2241 *want_write = true;
d18e9008
MS		 2242 } else if (!(open_flag & O_CREAT)) {
2243 /*
2244 * No O_CREATE -> atomicity not a requirement -> fall
2245 * back to lookup + open
2246 */
2247 goto no_open;
2248 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2249 /* Fall back and fail with the right error */
2250 create_error = error;
2251 goto no_open;
2252 } else {
2253 /* No side effects, safe to clear O_CREAT */
2254 create_error = error;
2255 open_flag &= ~O_CREAT;
2256 }
2257 }
2258
2259 if (open_flag & O_CREAT) {
2260 error = may_o_create(&nd->path, dentry, op->mode);
2261 if (error) {
2262 create_error = error;
2263 if (open_flag & O_EXCL)
2264 goto no_open;
2265 open_flag &= ~O_CREAT;
2266 }
2267 }
2268
2269 if (nd->flags & LOOKUP_DIRECTORY)
2270 open_flag |= O_DIRECTORY;
2271
015c3bbc
MS		 2272 od->dentry = DENTRY_NOT_SET;
2273 od->mnt = nd->path.mnt;
2274 filp = dir->i_op->atomic_open(dir, dentry, od, open_flag, mode,
d18e9008
MS		 2275 created);
2276 if (IS_ERR(filp)) {
015c3bbc
MS		 2277 if (WARN_ON(od->dentry != DENTRY_NOT_SET))
2278 dput(od->dentry);
d18e9008
MS		 2279
2280 if (create_error && PTR_ERR(filp) == -ENOENT)
2281 filp = ERR_PTR(create_error);
2282 goto out;
2283 }
2284
2285 acc_mode = op->acc_mode;
2286 if (*created) {
2287 fsnotify_create(dir, dentry);
2288 acc_mode = MAY_OPEN;
2289 }
2290
2291 if (!filp) {
015c3bbc 2292 if (WARN_ON(od->dentry == DENTRY_NOT_SET)) {
d18e9008
MS		 2293 filp = ERR_PTR(-EIO);
2294 goto out;
2295 }
015c3bbc 2296 if (od->dentry) {
d18e9008 2297 dput(dentry);
015c3bbc 2298 dentry = od->dentry;
d18e9008
MS		 2299 }
2300 goto looked_up;
2301 }
2302
2303 /*
2304 * We didn't have the inode before the open, so check open permission
2305 * here.
2306 */
2307 error = may_open(&filp->f_path, acc_mode, open_flag);
a8277b9b
MS		 2308 if (error) {
2309 fput(filp);
2310 filp = ERR_PTR(error);
2311 }
d18e9008
MS		 2312
2313out:
2314 dput(dentry);
2315 return filp;
2316
d18e9008
MS		 2317no_open:
2318 if (need_lookup) {
2319 dentry = lookup_real(dir, dentry, nd);
2320 if (IS_ERR(dentry))
2321 return ERR_CAST(dentry);
2322
2323 if (create_error) {
2324 int open_flag = op->open_flag;
2325
2326 filp = ERR_PTR(create_error);
2327 if ((open_flag & O_EXCL)) {
2328 if (!dentry->d_inode)
2329 goto out;
2330 } else if (!dentry->d_inode) {
2331 goto out;
2332 } else if ((open_flag & O_TRUNC) &&
2333 S_ISREG(dentry->d_inode->i_mode)) {
2334 goto out;
2335 }
2336 /* will fail later, go on to get the right error */
2337 }
2338 }
2339looked_up:
2340 path->dentry = dentry;
2341 path->mnt = nd->path.mnt;
2342 return NULL;
2343}
2344
d58ffd35
MS		 2345/*
2346 * Lookup, maybe create and open the last component
2347 *
2348 * Must be called with i_mutex held on parent.
2349 *
2350 * Returns open file or NULL on success, error otherwise. NULL means no open
2351 * was performed, only lookup.
2352 */
2353static struct file *lookup_open(struct nameidata *nd, struct path *path,
015c3bbc 2354 struct opendata *od,
d58ffd35 2355 const struct open_flags *op,
77d660a8 2356 bool *want_write, bool *created)
d58ffd35
MS		 2357{
2358 struct dentry *dir = nd->path.dentry;
54ef4872 2359 struct inode *dir_inode = dir->d_inode;
d58ffd35
MS		 2360 struct dentry *dentry;
2361 int error;
54ef4872 2362 bool need_lookup;
d58ffd35
MS		 2363
2364 *created = false;
54ef4872 2365 dentry = lookup_dcache(&nd->last, dir, nd, &need_lookup);
d58ffd35
MS		 2366 if (IS_ERR(dentry))
2367 return ERR_CAST(dentry);
2368
d18e9008
MS		 2369 /* Cached positive dentry: will open in f_op->open */
2370 if (!need_lookup && dentry->d_inode)
2371 goto out_no_open;
2372
2373 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
015c3bbc 2374 return atomic_open(nd, dentry, path, od, op, want_write,
d18e9008
MS		 2375 need_lookup, created);
2376 }
2377
54ef4872
MS		 2378 if (need_lookup) {
2379 BUG_ON(dentry->d_inode);
2380
2381 dentry = lookup_real(dir_inode, dentry, nd);
2382 if (IS_ERR(dentry))
2383 return ERR_CAST(dentry);
2384 }
2385
d58ffd35
MS		 2386 /* Negative dentry, just create the file */
2387 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2388 umode_t mode = op->mode;
2389 if (!IS_POSIXACL(dir->d_inode))
2390 mode &= ~current_umask();
2391 /*
2392 * This write is needed to ensure that a
2393 * rw->ro transition does not occur between
2394 * the time when the file is created and when
2395 * a permanent write count is taken through
015c3bbc 2396 * the 'struct file' in finish_open().
d58ffd35
MS		 2397 */
2398 error = mnt_want_write(nd->path.mnt);
2399 if (error)
2400 goto out_dput;
77d660a8 2401 *want_write = true;
d58ffd35
MS		 2402 *created = true;
2403 error = security_path_mknod(&nd->path, dentry, mode, 0);
2404 if (error)
2405 goto out_dput;
2406 error = vfs_create(dir->d_inode, dentry, mode, nd);
2407 if (error)
2408 goto out_dput;
2409 }
d18e9008 2410out_no_open:
d58ffd35
MS		 2411 path->dentry = dentry;
2412 path->mnt = nd->path.mnt;
2413 return NULL;
2414
2415out_dput:
2416 dput(dentry);
2417 return ERR_PTR(error);
2418}
2419
31e6b01f 2420/*
fe2d35ff 2421 * Handle the last step of open()
31e6b01f 2422 */
fb1cc555 2423static struct file *do_last(struct nameidata *nd, struct path *path,
015c3bbc
MS		 2424 struct opendata *od, const struct open_flags *op,
2425 const char *pathname)
fb1cc555 2426{
a1e28038 2427 struct dentry *dir = nd->path.dentry;
ca344a89 2428 int open_flag = op->open_flag;
77d660a8
MS		 2429 bool will_truncate = (open_flag & O_TRUNC) != 0;
2430 bool want_write = false;
bcda7652 2431 int acc_mode = op->acc_mode;
fb1cc555 2432 struct file *filp;
a1eb3315 2433 struct inode *inode;
d58ffd35 2434 bool created;
77d660a8 2435 bool symlink_ok = false;
16b1c1cd
MS		 2436 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2437 bool retried = false;
16c2cd71 2438 int error;
1f36f774 2439
c3e380b0
AV		 2440 nd->flags &= ~LOOKUP_PARENT;
2441 nd->flags |= op->intent;
2442
1f36f774
AV		 2443 switch (nd->last_type) {
2444 case LAST_DOTDOT:
176306f5 2445 case LAST_DOT:
fe2d35ff
AV		 2446 error = handle_dots(nd, nd->last_type);
2447 if (error)
2448 return ERR_PTR(error);
1f36f774 2449 /* fallthrough */
1f36f774 2450 case LAST_ROOT:
9f1fafee 2451 error = complete_walk(nd);
16c2cd71 2452 if (error)
9f1fafee 2453 return ERR_PTR(error);
fe2d35ff 2454 audit_inode(pathname, nd->path.dentry);
ca344a89 2455 if (open_flag & O_CREAT) {
fe2d35ff
AV		 2456 error = -EISDIR;
2457 goto exit;
2458 }
e83db167 2459 goto finish_open;
1f36f774 2460 case LAST_BIND:
9f1fafee 2461 error = complete_walk(nd);
16c2cd71 2462 if (error)
9f1fafee 2463 return ERR_PTR(error);
1f36f774 2464 audit_inode(pathname, dir);
e83db167 2465 goto finish_open;
1f36f774 2466 }
67ee3ad2 2467
ca344a89 2468 if (!(open_flag & O_CREAT)) {
fe2d35ff
AV		 2469 if (nd->last.name[nd->last.len])
2470 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
bcda7652 2471 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
77d660a8 2472 symlink_ok = true;
fe2d35ff 2473 /* we _can_ be in RCU mode here */
a1eb3315 2474 error = lookup_fast(nd, &nd->last, path, &inode);
71574865
MS		 2475 if (likely(!error))
2476 goto finish_lookup;
2477
2478 if (error < 0)
2479 goto exit;
2480
2481 BUG_ON(nd->inode != dir->d_inode);
b6183df7
MS		 2482 } else {
2483 /* create side of things */
2484 /*
2485 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2486 * has been cleared when we got to the last component we are
2487 * about to look up
2488 */
2489 error = complete_walk(nd);
2490 if (error)
2491 return ERR_PTR(error);
fe2d35ff 2492
b6183df7
MS		 2493 audit_inode(pathname, dir);
2494 error = -EISDIR;
2495 /* trailing slashes? */
2496 if (nd->last.name[nd->last.len])
2497 goto exit;
2498 }
a2c36b45 2499
16b1c1cd 2500retry_lookup:
a1e28038 2501 mutex_lock(&dir->d_inode->i_mutex);
015c3bbc 2502 filp = lookup_open(nd, path, od, op, &want_write, &created);
d58ffd35 2503 mutex_unlock(&dir->d_inode->i_mutex);
a1e28038 2504
d18e9008
MS		 2505 if (filp) {
2506 if (IS_ERR(filp))
2507 goto out;
2508
2509 if (created || !S_ISREG(filp->f_path.dentry->d_inode->i_mode))
77d660a8 2510 will_truncate = false;
d18e9008
MS		 2511
2512 audit_inode(pathname, filp->f_path.dentry);
2513 goto opened;
2514 }
fb1cc555 2515
d58ffd35 2516 if (created) {
9b44f1b3 2517 /* Don't check for write permission, don't truncate */
ca344a89 2518 open_flag &= ~O_TRUNC;
77d660a8 2519 will_truncate = false;
bcda7652 2520 acc_mode = MAY_OPEN;
d58ffd35 2521 path_to_nameidata(path, nd);
e83db167 2522 goto finish_open_created;
fb1cc555
AV		 2523 }
2524
2525 /*
2526 * It already exists.
2527 */
fb1cc555
AV		 2528 audit_inode(pathname, path->dentry);
2529
d18e9008
MS		 2530 /*
2531 * If atomic_open() acquired write access it is dropped now due to
2532 * possible mount and symlink following (this might be optimized away if
2533 * necessary...)
2534 */
2535 if (want_write) {
2536 mnt_drop_write(nd->path.mnt);
77d660a8 2537 want_write = false;
d18e9008
MS		 2538 }
2539
fb1cc555 2540 error = -EEXIST;
ca344a89 2541 if (open_flag & O_EXCL)
fb1cc555
AV		 2542 goto exit_dput;
2543
9875cf80
DH		 2544 error = follow_managed(path, nd->flags);
2545 if (error < 0)
2546 goto exit_dput;
fb1cc555 2547
a3fbbde7
AV		 2548 if (error)
2549 nd->flags |= LOOKUP_JUMPED;
2550
decf3400
MS		 2551 BUG_ON(nd->flags & LOOKUP_RCU);
2552 inode = path->dentry->d_inode;
5f5daac1
MS		 2553finish_lookup:
2554 /* we _can_ be in RCU mode here */
fb1cc555 2555 error = -ENOENT;
54c33e7f
MS		 2556 if (!inode) {
2557 path_to_nameidata(path, nd);
2558 goto exit;
2559 }
9e67f361 2560
d45ea867
MS		 2561 if (should_follow_link(inode, !symlink_ok)) {
2562 if (nd->flags & LOOKUP_RCU) {
2563 if (unlikely(unlazy_walk(nd, path->dentry))) {
2564 error = -ECHILD;
2565 goto exit;
2566 }
2567 }
2568 BUG_ON(inode != path->dentry->d_inode);
fb1cc555 2569 return NULL;
d45ea867 2570 }
fb1cc555 2571
16b1c1cd
MS		 2572 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
2573 path_to_nameidata(path, nd);
2574 } else {
2575 save_parent.dentry = nd->path.dentry;
2576 save_parent.mnt = mntget(path->mnt);
2577 nd->path.dentry = path->dentry;
2578
2579 }
decf3400 2580 nd->inode = inode;
a3fbbde7
AV		 2581 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2582 error = complete_walk(nd);
16b1c1cd
MS		 2583 if (error) {
2584 path_put(&save_parent);
097b180c 2585 return ERR_PTR(error);
16b1c1cd 2586 }
fb1cc555 2587 error = -EISDIR;
050ac841 2588 if ((open_flag & O_CREAT) && S_ISDIR(nd->inode->i_mode))
fb1cc555 2589 goto exit;
af2f5542
MS		 2590 error = -ENOTDIR;
2591 if ((nd->flags & LOOKUP_DIRECTORY) && !nd->inode->i_op->lookup)
2592 goto exit;
d7fdd7f6 2593 audit_inode(pathname, nd->path.dentry);
e83db167 2594finish_open:
6c0d46c4 2595 if (!S_ISREG(nd->inode->i_mode))
77d660a8 2596 will_truncate = false;
6c0d46c4 2597
0f9d1a10
AV		 2598 if (will_truncate) {
2599 error = mnt_want_write(nd->path.mnt);
2600 if (error)
2601 goto exit;
77d660a8 2602 want_write = true;
0f9d1a10 2603 }
e83db167 2604finish_open_created:
bcda7652 2605 error = may_open(&nd->path, acc_mode, open_flag);
ca344a89 2606 if (error)
0f9d1a10 2607 goto exit;
015c3bbc
MS		 2608 od->mnt = nd->path.mnt;
2609 filp = finish_open(od, nd->path.dentry, NULL);
f60dc3db
MS		 2610 if (IS_ERR(filp)) {
2611 if (filp == ERR_PTR(-EOPENSTALE))
2612 goto stale_open;
015c3bbc 2613 goto out;
f60dc3db 2614 }
a8277b9b 2615opened:
015c3bbc
MS		 2616 error = open_check_o_direct(filp);
2617 if (error)
2618 goto exit_fput;
aa4caadb
MS		 2619 error = ima_file_check(filp, op->acc_mode);
2620 if (error)
2621 goto exit_fput;
2622
2623 if (will_truncate) {
2624 error = handle_truncate(filp);
2625 if (error)
2626 goto exit_fput;
0f9d1a10 2627 }
ca344a89
AV		 2628out:
2629 if (want_write)
0f9d1a10 2630 mnt_drop_write(nd->path.mnt);
16b1c1cd 2631 path_put(&save_parent);
e276ae67 2632 terminate_walk(nd);
fb1cc555
AV		 2633 return filp;
2634
fb1cc555
AV		 2635exit_dput:
2636 path_put_conditional(path, nd);
2637exit:
ca344a89
AV		 2638 filp = ERR_PTR(error);
2639 goto out;
015c3bbc
MS		 2640exit_fput:
2641 fput(filp);
2642 goto exit;
2643
f60dc3db
MS		 2644stale_open:
2645 /* If no saved parent or already retried then can't retry */
2646 if (!save_parent.dentry || retried)
2647 goto out;
2648
2649 BUG_ON(save_parent.dentry != dir);
2650 path_put(&nd->path);
2651 nd->path = save_parent;
2652 nd->inode = dir->d_inode;
2653 save_parent.mnt = NULL;
2654 save_parent.dentry = NULL;
2655 if (want_write) {
2656 mnt_drop_write(nd->path.mnt);
2657 want_write = false;
2658 }
2659 retried = true;
2660 goto retry_lookup;
fb1cc555
AV		 2661}
2662
13aab428 2663static struct file *path_openat(int dfd, const char *pathname,
73d049a4 2664 struct nameidata *nd, const struct open_flags *op, int flags)
1da177e4 2665{
fe2d35ff 2666 struct file *base = NULL;
015c3bbc
MS		 2667 struct opendata od;
2668 struct file *res;
9850c056 2669 struct path path;
13aab428 2670 int error;
31e6b01f 2671
015c3bbc
MS		 2672 od.filp = get_empty_filp();
2673 if (!od.filp)
31e6b01f
NP		 2674 return ERR_PTR(-ENFILE);
2675
015c3bbc 2676 od.filp->f_flags = op->open_flag;
31e6b01f 2677
73d049a4 2678 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
31e6b01f 2679 if (unlikely(error))
13aab428 2680 goto out_filp;
31e6b01f 2681
fe2d35ff 2682 current->total_link_count = 0;
73d049a4 2683 error = link_path_walk(pathname, nd);
31e6b01f
NP		 2684 if (unlikely(error))
2685 goto out_filp;
1da177e4 2686
015c3bbc
MS		 2687 res = do_last(nd, &path, &od, op, pathname);
2688 while (unlikely(!res)) { /* trailing symlink */
7b9337aa 2689 struct path link = path;
def4af30 2690 void *cookie;
574197e0 2691 if (!(nd->flags & LOOKUP_FOLLOW)) {
73d049a4
AV		 2692 path_put_conditional(&path, nd);
2693 path_put(&nd->path);
015c3bbc 2694 res = ERR_PTR(-ELOOP);
40b39136
AV		 2695 break;
2696 }
73d049a4
AV		 2697 nd->flags |= LOOKUP_PARENT;
2698 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
574197e0 2699 error = follow_link(&link, nd, &cookie);
c3e380b0 2700 if (unlikely(error))
6d7b5aae 2701 goto out_filp;
015c3bbc 2702 res = do_last(nd, &path, &od, op, pathname);
574197e0 2703 put_link(nd, &link, cookie);
806b681c 2704 }
10fa8e62 2705out:
73d049a4
AV		 2706 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2707 path_put(&nd->root);
fe2d35ff
AV		 2708 if (base)
2709 fput(base);
015c3bbc
MS		 2710 if (od.filp) {
2711 BUG_ON(od.filp->f_path.dentry);
2712 put_filp(od.filp);
2713 }
2714 if (res == ERR_PTR(-EOPENSTALE)) {
16b1c1cd 2715 if (flags & LOOKUP_RCU)
015c3bbc 2716 res = ERR_PTR(-ECHILD);
16b1c1cd 2717 else
015c3bbc 2718 res = ERR_PTR(-ESTALE);
16b1c1cd 2719 }
015c3bbc 2720 return res;
1da177e4 2721
31e6b01f 2722out_filp:
015c3bbc 2723 res = ERR_PTR(error);
10fa8e62 2724 goto out;
1da177e4
LT		 2725}
2726
13aab428
AV		 2727struct file *do_filp_open(int dfd, const char *pathname,
2728 const struct open_flags *op, int flags)
2729{
73d049a4 2730 struct nameidata nd;
13aab428
AV		 2731 struct file *filp;
2732
73d049a4 2733 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
13aab428 2734 if (unlikely(filp == ERR_PTR(-ECHILD)))
73d049a4 2735 filp = path_openat(dfd, pathname, &nd, op, flags);
13aab428 2736 if (unlikely(filp == ERR_PTR(-ESTALE)))
73d049a4 2737 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
13aab428
AV		 2738 return filp;
2739}
2740
73d049a4
AV		 2741struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2742 const char *name, const struct open_flags *op, int flags)
2743{
2744 struct nameidata nd;
2745 struct file *file;
2746
2747 nd.root.mnt = mnt;
2748 nd.root.dentry = dentry;
2749
2750 flags |= LOOKUP_ROOT;
2751
bcda7652 2752 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
73d049a4
AV		 2753 return ERR_PTR(-ELOOP);
2754
2755 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2756 if (unlikely(file == ERR_PTR(-ECHILD)))
2757 file = path_openat(-1, name, &nd, op, flags);
2758 if (unlikely(file == ERR_PTR(-ESTALE)))
2759 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2760 return file;
2761}
2762
ed75e95d 2763struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
1da177e4 2764{
c663e5d8 2765 struct dentry *dentry = ERR_PTR(-EEXIST);
ed75e95d
AV		 2766 struct nameidata nd;
2767 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2768 if (error)
2769 return ERR_PTR(error);
1da177e4 2770
c663e5d8
CH		 2771 /*
2772 * Yucky last component or no last component at all?
2773 * (foo/., foo/.., /////)
2774 */
ed75e95d
AV		 2775 if (nd.last_type != LAST_NORM)
2776 goto out;
2777 nd.flags &= ~LOOKUP_PARENT;
2778 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
c663e5d8
CH		 2779
2780 /*
2781 * Do the final lookup.
2782 */
ed75e95d
AV		 2783 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2784 dentry = lookup_hash(&nd);
1da177e4
LT		 2785 if (IS_ERR(dentry))
2786 goto fail;
c663e5d8 2787
e9baf6e5
AV		 2788 if (dentry->d_inode)
2789 goto eexist;
c663e5d8
CH		 2790 /*
2791 * Special case - lookup gave negative, but... we had foo/bar/
2792 * From the vfs_mknod() POV we just have a negative dentry -
2793 * all is fine. Let's be bastards - you had / on the end, you've
2794 * been asking for (non-existent) directory. -ENOENT for you.
2795 */
ed75e95d 2796 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
e9baf6e5
AV		 2797 dput(dentry);
2798 dentry = ERR_PTR(-ENOENT);
ed75e95d 2799 goto fail;
e9baf6e5 2800 }
ed75e95d 2801 *path = nd.path;
1da177e4 2802 return dentry;
e9baf6e5 2803eexist:
1da177e4 2804 dput(dentry);
e9baf6e5 2805 dentry = ERR_PTR(-EEXIST);
1da177e4 2806fail:
ed75e95d
AV		 2807 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2808out:
2809 path_put(&nd.path);
1da177e4
LT		 2810 return dentry;
2811}
dae6ad8f
AV		 2812EXPORT_SYMBOL(kern_path_create);
2813
2814struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2815{
2816 char *tmp = getname(pathname);
2817 struct dentry *res;
2818 if (IS_ERR(tmp))
2819 return ERR_CAST(tmp);
2820 res = kern_path_create(dfd, tmp, path, is_dir);
2821 putname(tmp);
2822 return res;
2823}
2824EXPORT_SYMBOL(user_path_create);
2825
1a67aafb 2826int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
1da177e4 2827{
a95164d9 2828 int error = may_create(dir, dentry);
1da177e4
LT		 2829
2830 if (error)
2831 return error;
2832
975d6b39 2833 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1da177e4
LT		 2834 return -EPERM;
2835
acfa4380 2836 if (!dir->i_op->mknod)
1da177e4
LT		 2837 return -EPERM;
2838
08ce5f16
SH		 2839 error = devcgroup_inode_mknod(mode, dev);
2840 if (error)
2841 return error;
2842
1da177e4
LT		 2843 error = security_inode_mknod(dir, dentry, mode, dev);
2844 if (error)
2845 return error;
2846
1da177e4 2847 error = dir->i_op->mknod(dir, dentry, mode, dev);
a74574aa 2848 if (!error)
f38aa942 2849 fsnotify_create(dir, dentry);
1da177e4
LT		 2850 return error;
2851}
2852
f69aac00 2853static int may_mknod(umode_t mode)
463c3197
DH		 2854{
2855 switch (mode & S_IFMT) {
2856 case S_IFREG:
2857 case S_IFCHR:
2858 case S_IFBLK:
2859 case S_IFIFO:
2860 case S_IFSOCK:
2861 case 0: /* zero mode translates to S_IFREG */
2862 return 0;
2863 case S_IFDIR:
2864 return -EPERM;
2865 default:
2866 return -EINVAL;
2867 }
2868}
2869
8208a22b 2870SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
2e4d0924 2871 unsigned, dev)
1da177e4 2872{
2ad94ae6 2873 struct dentry *dentry;
dae6ad8f
AV		 2874 struct path path;
2875 int error;
1da177e4
LT		 2876
2877 if (S_ISDIR(mode))
2878 return -EPERM;
1da177e4 2879
dae6ad8f
AV		 2880 dentry = user_path_create(dfd, filename, &path, 0);
2881 if (IS_ERR(dentry))
2882 return PTR_ERR(dentry);
2ad94ae6 2883
dae6ad8f 2884 if (!IS_POSIXACL(path.dentry->d_inode))
ce3b0f8d 2885 mode &= ~current_umask();
463c3197
DH		 2886 error = may_mknod(mode);
2887 if (error)
2888 goto out_dput;
dae6ad8f 2889 error = mnt_want_write(path.mnt);
463c3197
DH		 2890 if (error)
2891 goto out_dput;
dae6ad8f 2892 error = security_path_mknod(&path, dentry, mode, dev);
be6d3e56
KT		 2893 if (error)
2894 goto out_drop_write;
463c3197 2895 switch (mode & S_IFMT) {
1da177e4 2896 case 0: case S_IFREG:
dae6ad8f 2897 error = vfs_create(path.dentry->d_inode,dentry,mode,NULL);
1da177e4
LT		 2898 break;
2899 case S_IFCHR: case S_IFBLK:
dae6ad8f 2900 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
1da177e4
LT		 2901 new_decode_dev(dev));
2902 break;
2903 case S_IFIFO: case S_IFSOCK:
dae6ad8f 2904 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
1da177e4 2905 break;
1da177e4 2906 }
be6d3e56 2907out_drop_write:
dae6ad8f 2908 mnt_drop_write(path.mnt);
463c3197
DH		 2909out_dput:
2910 dput(dentry);
dae6ad8f
AV		 2911 mutex_unlock(&path.dentry->d_inode->i_mutex);
2912 path_put(&path);
1da177e4
LT		 2913
2914 return error;
2915}
2916
8208a22b 2917SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
5590ff0d
UD		 2918{
2919 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2920}
2921
18bb1db3 2922int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1da177e4 2923{
a95164d9 2924 int error = may_create(dir, dentry);
8de52778 2925 unsigned max_links = dir->i_sb->s_max_links;
1da177e4
LT		 2926
2927 if (error)
2928 return error;
2929
acfa4380 2930 if (!dir->i_op->mkdir)
1da177e4
LT		 2931 return -EPERM;
2932
2933 mode &= (S_IRWXUGO|S_ISVTX);
2934 error = security_inode_mkdir(dir, dentry, mode);
2935 if (error)
2936 return error;
2937
8de52778
AV		 2938 if (max_links && dir->i_nlink >= max_links)
2939 return -EMLINK;
2940
1da177e4 2941 error = dir->i_op->mkdir(dir, dentry, mode);
a74574aa 2942 if (!error)
f38aa942 2943 fsnotify_mkdir(dir, dentry);
1da177e4
LT		 2944 return error;
2945}
2946
a218d0fd 2947SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
1da177e4 2948{
6902d925 2949 struct dentry *dentry;
dae6ad8f
AV		 2950 struct path path;
2951 int error;
1da177e4 2952
dae6ad8f 2953 dentry = user_path_create(dfd, pathname, &path, 1);
6902d925 2954 if (IS_ERR(dentry))
dae6ad8f 2955 return PTR_ERR(dentry);
1da177e4 2956
dae6ad8f 2957 if (!IS_POSIXACL(path.dentry->d_inode))
ce3b0f8d 2958 mode &= ~current_umask();
dae6ad8f 2959 error = mnt_want_write(path.mnt);
463c3197
DH		 2960 if (error)
2961 goto out_dput;
dae6ad8f 2962 error = security_path_mkdir(&path, dentry, mode);
be6d3e56
KT		 2963 if (error)
2964 goto out_drop_write;
dae6ad8f 2965 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
be6d3e56 2966out_drop_write:
dae6ad8f 2967 mnt_drop_write(path.mnt);
463c3197 2968out_dput:
6902d925 2969 dput(dentry);
dae6ad8f
AV		 2970 mutex_unlock(&path.dentry->d_inode->i_mutex);
2971 path_put(&path);
1da177e4
LT		 2972 return error;
2973}
2974
a218d0fd 2975SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
5590ff0d
UD		 2976{
2977 return sys_mkdirat(AT_FDCWD, pathname, mode);
2978}
2979
1da177e4 2980/*
a71905f0 2981 * The dentry_unhash() helper will try to drop the dentry early: we
c0d02594 2982 * should have a usage count of 1 if we're the only user of this
a71905f0
SW		 2983 * dentry, and if that is true (possibly after pruning the dcache),
2984 * then we drop the dentry now.
1da177e4
LT		 2985 *
2986 * A low-level filesystem can, if it choses, legally
2987 * do a
2988 *
2989 * if (!d_unhashed(dentry))
2990 * return -EBUSY;
2991 *
2992 * if it cannot handle the case of removing a directory
2993 * that is still in use by something else..
2994 */
2995void dentry_unhash(struct dentry *dentry)
2996{
dc168427 2997 shrink_dcache_parent(dentry);
1da177e4 2998 spin_lock(&dentry->d_lock);
64252c75 2999 if (dentry->d_count == 1)
1da177e4
LT		 3000 __d_drop(dentry);
3001 spin_unlock(&dentry->d_lock);
1da177e4
LT		 3002}
3003
3004int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3005{
3006 int error = may_delete(dir, dentry, 1);
3007
3008 if (error)
3009 return error;
3010
acfa4380 3011 if (!dir->i_op->rmdir)
1da177e4
LT		 3012 return -EPERM;
3013
1d2ef590 3014 dget(dentry);
1b1dcc1b 3015 mutex_lock(&dentry->d_inode->i_mutex);
912dbc15
SW		 3016
3017 error = -EBUSY;
1da177e4 3018 if (d_mountpoint(dentry))
912dbc15
SW		 3019 goto out;
3020
3021 error = security_inode_rmdir(dir, dentry);
3022 if (error)
3023 goto out;
3024
3cebde24 3025 shrink_dcache_parent(dentry);
912dbc15
SW		 3026 error = dir->i_op->rmdir(dir, dentry);
3027 if (error)
3028 goto out;
3029
3030 dentry->d_inode->i_flags |= S_DEAD;
3031 dont_mount(dentry);
3032
3033out:
1b1dcc1b 3034 mutex_unlock(&dentry->d_inode->i_mutex);
1d2ef590 3035 dput(dentry);
912dbc15 3036 if (!error)
1da177e4 3037 d_delete(dentry);
1da177e4
LT		 3038 return error;
3039}
3040
5590ff0d 3041static long do_rmdir(int dfd, const char __user *pathname)
1da177e4
LT		 3042{
3043 int error = 0;
3044 char * name;
3045 struct dentry *dentry;
3046 struct nameidata nd;
3047
2ad94ae6 3048 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 3049 if (error)
2ad94ae6 3050 return error;
1da177e4
LT		 3051
3052 switch(nd.last_type) {
0612d9fb
OH		 3053 case LAST_DOTDOT:
3054 error = -ENOTEMPTY;
3055 goto exit1;
3056 case LAST_DOT:
3057 error = -EINVAL;
3058 goto exit1;
3059 case LAST_ROOT:
3060 error = -EBUSY;
3061 goto exit1;
1da177e4 3062 }
0612d9fb
OH		 3063
3064 nd.flags &= ~LOOKUP_PARENT;
3065
4ac91378 3066 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 3067 dentry = lookup_hash(&nd);
1da177e4 3068 error = PTR_ERR(dentry);
6902d925
DH		 3069 if (IS_ERR(dentry))
3070 goto exit2;
e6bc45d6
TT		 3071 if (!dentry->d_inode) {
3072 error = -ENOENT;
3073 goto exit3;
3074 }
0622753b
DH		 3075 error = mnt_want_write(nd.path.mnt);
3076 if (error)
3077 goto exit3;
be6d3e56
KT		 3078 error = security_path_rmdir(&nd.path, dentry);
3079 if (error)
3080 goto exit4;
4ac91378 3081 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
be6d3e56 3082exit4:
0622753b
DH		 3083 mnt_drop_write(nd.path.mnt);
3084exit3:
6902d925
DH		 3085 dput(dentry);
3086exit2:
4ac91378 3087 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 3088exit1:
1d957f9b 3089 path_put(&nd.path);
1da177e4
LT		 3090 putname(name);
3091 return error;
3092}
3093
3cdad428 3094SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
5590ff0d
UD		 3095{
3096 return do_rmdir(AT_FDCWD, pathname);
3097}
3098
1da177e4
LT		 3099int vfs_unlink(struct inode *dir, struct dentry *dentry)
3100{
3101 int error = may_delete(dir, dentry, 0);
3102
3103 if (error)
3104 return error;
3105
acfa4380 3106 if (!dir->i_op->unlink)
1da177e4
LT		 3107 return -EPERM;
3108
1b1dcc1b 3109 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 3110 if (d_mountpoint(dentry))
3111 error = -EBUSY;
3112 else {
3113 error = security_inode_unlink(dir, dentry);
bec1052e 3114 if (!error) {
1da177e4 3115 error = dir->i_op->unlink(dir, dentry);
bec1052e 3116 if (!error)
d83c49f3 3117 dont_mount(dentry);
bec1052e 3118 }
1da177e4 3119 }
1b1dcc1b 3120 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4
LT		 3121
3122 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3123 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
ece95912 3124 fsnotify_link_count(dentry->d_inode);
e234f35c 3125 d_delete(dentry);
1da177e4 3126 }
0eeca283 3127
1da177e4
LT		 3128 return error;
3129}
3130
3131/*
3132 * Make sure that the actual truncation of the file will occur outside its
1b1dcc1b 3133 * directory's i_mutex. Truncate can take a long time if there is a lot of
1da177e4
LT		 3134 * writeout happening, and we don't want to prevent access to the directory
3135 * while waiting on the I/O.
3136 */
5590ff0d 3137static long do_unlinkat(int dfd, const char __user *pathname)
1da177e4 3138{
2ad94ae6
AV		 3139 int error;
3140 char *name;
1da177e4
LT		 3141 struct dentry *dentry;
3142 struct nameidata nd;
3143 struct inode *inode = NULL;
3144
2ad94ae6 3145 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 3146 if (error)
2ad94ae6
AV		 3147 return error;
3148
1da177e4
LT		 3149 error = -EISDIR;
3150 if (nd.last_type != LAST_NORM)
3151 goto exit1;
0612d9fb
OH		 3152
3153 nd.flags &= ~LOOKUP_PARENT;
3154
4ac91378 3155 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 3156 dentry = lookup_hash(&nd);
1da177e4
LT		 3157 error = PTR_ERR(dentry);
3158 if (!IS_ERR(dentry)) {
3159 /* Why not before? Because we want correct error value */
50338b88
TE		 3160 if (nd.last.name[nd.last.len])
3161 goto slashes;
1da177e4 3162 inode = dentry->d_inode;
50338b88 3163 if (!inode)
e6bc45d6
TT		 3164 goto slashes;
3165 ihold(inode);
0622753b
DH		 3166 error = mnt_want_write(nd.path.mnt);
3167 if (error)
3168 goto exit2;
be6d3e56
KT		 3169 error = security_path_unlink(&nd.path, dentry);
3170 if (error)
3171 goto exit3;
4ac91378 3172 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
be6d3e56 3173exit3:
0622753b 3174 mnt_drop_write(nd.path.mnt);
1da177e4
LT		 3175 exit2:
3176 dput(dentry);
3177 }
4ac91378 3178 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4
LT		 3179 if (inode)
3180 iput(inode); /* truncate the inode here */
3181exit1:
1d957f9b 3182 path_put(&nd.path);
1da177e4
LT		 3183 putname(name);
3184 return error;
3185
3186slashes:
3187 error = !dentry->d_inode ? -ENOENT :
3188 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
3189 goto exit2;
3190}
3191
2e4d0924 3192SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
5590ff0d
UD		 3193{
3194 if ((flag & ~AT_REMOVEDIR) != 0)
3195 return -EINVAL;
3196
3197 if (flag & AT_REMOVEDIR)
3198 return do_rmdir(dfd, pathname);
3199
3200 return do_unlinkat(dfd, pathname);
3201}
3202
3480b257 3203SYSCALL_DEFINE1(unlink, const char __user *, pathname)
5590ff0d
UD		 3204{
3205 return do_unlinkat(AT_FDCWD, pathname);
3206}
3207
db2e747b 3208int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
1da177e4 3209{
a95164d9 3210 int error = may_create(dir, dentry);
1da177e4
LT		 3211
3212 if (error)
3213 return error;
3214
acfa4380 3215 if (!dir->i_op->symlink)
1da177e4
LT		 3216 return -EPERM;
3217
3218 error = security_inode_symlink(dir, dentry, oldname);
3219 if (error)
3220 return error;
3221
1da177e4 3222 error = dir->i_op->symlink(dir, dentry, oldname);
a74574aa 3223 if (!error)
f38aa942 3224 fsnotify_create(dir, dentry);
1da177e4
LT		 3225 return error;
3226}
3227
2e4d0924
HC		 3228SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3229 int, newdfd, const char __user *, newname)
1da177e4 3230{
2ad94ae6
AV		 3231 int error;
3232 char *from;
6902d925 3233 struct dentry *dentry;
dae6ad8f 3234 struct path path;
1da177e4
LT		 3235
3236 from = getname(oldname);
2ad94ae6 3237 if (IS_ERR(from))
1da177e4 3238 return PTR_ERR(from);
1da177e4 3239
dae6ad8f 3240 dentry = user_path_create(newdfd, newname, &path, 0);
6902d925
DH		 3241 error = PTR_ERR(dentry);
3242 if (IS_ERR(dentry))
dae6ad8f 3243 goto out_putname;
6902d925 3244
dae6ad8f 3245 error = mnt_want_write(path.mnt);
75c3f29d
DH		 3246 if (error)
3247 goto out_dput;
dae6ad8f 3248 error = security_path_symlink(&path, dentry, from);
be6d3e56
KT		 3249 if (error)
3250 goto out_drop_write;
dae6ad8f 3251 error = vfs_symlink(path.dentry->d_inode, dentry, from);
be6d3e56 3252out_drop_write:
dae6ad8f 3253 mnt_drop_write(path.mnt);
75c3f29d 3254out_dput:
6902d925 3255 dput(dentry);
dae6ad8f
AV		 3256 mutex_unlock(&path.dentry->d_inode->i_mutex);
3257 path_put(&path);
6902d925 3258out_putname:
1da177e4
LT		 3259 putname(from);
3260 return error;
3261}
3262
3480b257 3263SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 3264{
3265 return sys_symlinkat(oldname, AT_FDCWD, newname);
3266}
3267
1da177e4
LT		 3268int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
3269{
3270 struct inode *inode = old_dentry->d_inode;
8de52778 3271 unsigned max_links = dir->i_sb->s_max_links;
1da177e4
LT		 3272 int error;
3273
3274 if (!inode)
3275 return -ENOENT;
3276
a95164d9 3277 error = may_create(dir, new_dentry);
1da177e4
LT		 3278 if (error)
3279 return error;
3280
3281 if (dir->i_sb != inode->i_sb)
3282 return -EXDEV;
3283
3284 /*
3285 * A link to an append-only or immutable file cannot be created.
3286 */
3287 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3288 return -EPERM;
acfa4380 3289 if (!dir->i_op->link)
1da177e4 3290 return -EPERM;
7e79eedb 3291 if (S_ISDIR(inode->i_mode))
1da177e4
LT		 3292 return -EPERM;
3293
3294 error = security_inode_link(old_dentry, dir, new_dentry);
3295 if (error)
3296 return error;
3297
7e79eedb 3298 mutex_lock(&inode->i_mutex);
aae8a97d
AK		 3299 /* Make sure we don't allow creating hardlink to an unlinked file */
3300 if (inode->i_nlink == 0)
3301 error = -ENOENT;
8de52778
AV		 3302 else if (max_links && inode->i_nlink >= max_links)
3303 error = -EMLINK;
aae8a97d
AK		 3304 else
3305 error = dir->i_op->link(old_dentry, dir, new_dentry);
7e79eedb 3306 mutex_unlock(&inode->i_mutex);
e31e14ec 3307 if (!error)
7e79eedb 3308 fsnotify_link(dir, inode, new_dentry);
1da177e4
LT		 3309 return error;
3310}
3311
3312/*
3313 * Hardlinks are often used in delicate situations. We avoid
3314 * security-related surprises by not following symlinks on the
3315 * newname. --KAB
3316 *
3317 * We don't follow them on the oldname either to be compatible
3318 * with linux 2.0, and to avoid hard-linking to directories
3319 * and other special files. --ADM
3320 */
2e4d0924
HC		 3321SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3322 int, newdfd, const char __user *, newname, int, flags)
1da177e4
LT		 3323{
3324 struct dentry *new_dentry;
dae6ad8f 3325 struct path old_path, new_path;
11a7b371 3326 int how = 0;
1da177e4 3327 int error;
1da177e4 3328
11a7b371 3329 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
c04030e1 3330 return -EINVAL;
11a7b371
AK		 3331 /*
3332 * To use null names we require CAP_DAC_READ_SEARCH
3333 * This ensures that not everyone will be able to create
3334 * handlink using the passed filedescriptor.
3335 */
3336 if (flags & AT_EMPTY_PATH) {
3337 if (!capable(CAP_DAC_READ_SEARCH))
3338 return -ENOENT;
3339 how = LOOKUP_EMPTY;
3340 }
3341
3342 if (flags & AT_SYMLINK_FOLLOW)
3343 how |= LOOKUP_FOLLOW;
c04030e1 3344
11a7b371 3345 error = user_path_at(olddfd, oldname, how, &old_path);
1da177e4 3346 if (error)
2ad94ae6
AV		 3347 return error;
3348
dae6ad8f 3349 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
1da177e4 3350 error = PTR_ERR(new_dentry);
6902d925 3351 if (IS_ERR(new_dentry))
dae6ad8f
AV		 3352 goto out;
3353
3354 error = -EXDEV;
3355 if (old_path.mnt != new_path.mnt)
3356 goto out_dput;
3357 error = mnt_want_write(new_path.mnt);
75c3f29d
DH		 3358 if (error)
3359 goto out_dput;
dae6ad8f 3360 error = security_path_link(old_path.dentry, &new_path, new_dentry);
be6d3e56
KT		 3361 if (error)
3362 goto out_drop_write;
dae6ad8f 3363 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
be6d3e56 3364out_drop_write:
dae6ad8f 3365 mnt_drop_write(new_path.mnt);
75c3f29d 3366out_dput:
6902d925 3367 dput(new_dentry);
dae6ad8f
AV		 3368 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
3369 path_put(&new_path);
1da177e4 3370out:
2d8f3038 3371 path_put(&old_path);
1da177e4
LT		 3372
3373 return error;
3374}
3375
3480b257 3376SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
5590ff0d 3377{
c04030e1 3378 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
5590ff0d
UD		 3379}
3380
1da177e4
LT		 3381/*
3382 * The worst of all namespace operations - renaming directory. "Perverted"
3383 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3384 * Problems:
3385 * a) we can get into loop creation. Check is done in is_subdir().
3386 * b) race potential - two innocent renames can create a loop together.
3387 * That's where 4.4 screws up. Current fix: serialization on
a11f3a05 3388 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
1da177e4
LT		 3389 * story.
3390 * c) we have to lock _three_ objects - parents and victim (if it exists).
1b1dcc1b 3391 * And that - after we got ->i_mutex on parents (until then we don't know
1da177e4
LT		 3392 * whether the target exists). Solution: try to be smart with locking
3393 * order for inodes. We rely on the fact that tree topology may change
a11f3a05 3394 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
1da177e4
LT		 3395 * move will be locked. Thus we can rank directories by the tree
3396 * (ancestors first) and rank all non-directories after them.
3397 * That works since everybody except rename does "lock parent, lookup,
a11f3a05 3398 * lock child" and rename is under ->s_vfs_rename_mutex.
1da177e4
LT		 3399 * HOWEVER, it relies on the assumption that any object with ->lookup()
3400 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3401 * we'd better make sure that there's no link(2) for them.
e4eaac06 3402 * d) conversion from fhandle to dentry may come in the wrong moment - when
1b1dcc1b 3403 * we are removing the target. Solution: we will have to grab ->i_mutex
1da177e4 3404 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
c41b20e7 3405 * ->i_mutex on parents, which works but leads to some truly excessive
1da177e4
LT		 3406 * locking].
3407 */
75c96f85
AB		 3408static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3409 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3410{
3411 int error = 0;
9055cba7 3412 struct inode *target = new_dentry->d_inode;
8de52778 3413 unsigned max_links = new_dir->i_sb->s_max_links;
1da177e4
LT		 3414
3415 /*
3416 * If we are going to change the parent - check write permissions,
3417 * we'll need to flip '..'.
3418 */
3419 if (new_dir != old_dir) {
f419a2e3 3420 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
1da177e4
LT		 3421 if (error)
3422 return error;
3423 }
3424
3425 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3426 if (error)
3427 return error;
3428
1d2ef590 3429 dget(new_dentry);
d83c49f3 3430 if (target)
1b1dcc1b 3431 mutex_lock(&target->i_mutex);
9055cba7
SW		 3432
3433 error = -EBUSY;
3434 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3435 goto out;
3436
8de52778
AV		 3437 error = -EMLINK;
3438 if (max_links && !target && new_dir != old_dir &&
3439 new_dir->i_nlink >= max_links)
3440 goto out;
3441
3cebde24
SW		 3442 if (target)
3443 shrink_dcache_parent(new_dentry);
9055cba7
SW		 3444 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3445 if (error)
3446 goto out;
3447
1da177e4 3448 if (target) {
9055cba7
SW		 3449 target->i_flags |= S_DEAD;
3450 dont_mount(new_dentry);
1da177e4 3451 }
9055cba7
SW		 3452out:
3453 if (target)
3454 mutex_unlock(&target->i_mutex);
1d2ef590 3455 dput(new_dentry);
e31e14ec 3456 if (!error)
349457cc
MF		 3457 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3458 d_move(old_dentry,new_dentry);
1da177e4
LT		 3459 return error;
3460}
3461
75c96f85
AB		 3462static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3463 struct inode *new_dir, struct dentry *new_dentry)
1da177e4 3464{
51892bbb 3465 struct inode *target = new_dentry->d_inode;
1da177e4
LT		 3466 int error;
3467
3468 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3469 if (error)
3470 return error;
3471
3472 dget(new_dentry);
1da177e4 3473 if (target)
1b1dcc1b 3474 mutex_lock(&target->i_mutex);
51892bbb
SW		 3475
3476 error = -EBUSY;
1da177e4 3477 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
51892bbb
SW		 3478 goto out;
3479
3480 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3481 if (error)
3482 goto out;
3483
3484 if (target)
3485 dont_mount(new_dentry);
3486 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3487 d_move(old_dentry, new_dentry);
3488out:
1da177e4 3489 if (target)
1b1dcc1b 3490 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3491 dput(new_dentry);
3492 return error;
3493}
3494
3495int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3496 struct inode *new_dir, struct dentry *new_dentry)
3497{
3498 int error;
3499 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
59b0df21 3500 const unsigned char *old_name;
1da177e4
LT		 3501
3502 if (old_dentry->d_inode == new_dentry->d_inode)
3503 return 0;
3504
3505 error = may_delete(old_dir, old_dentry, is_dir);
3506 if (error)
3507 return error;
3508
3509 if (!new_dentry->d_inode)
a95164d9 3510 error = may_create(new_dir, new_dentry);
1da177e4
LT		 3511 else
3512 error = may_delete(new_dir, new_dentry, is_dir);
3513 if (error)
3514 return error;
3515
acfa4380 3516 if (!old_dir->i_op->rename)
1da177e4
LT		 3517 return -EPERM;
3518
0eeca283
RL		 3519 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3520
1da177e4
LT		 3521 if (is_dir)
3522 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3523 else
3524 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
123df294
AV		 3525 if (!error)
3526 fsnotify_move(old_dir, new_dir, old_name, is_dir,
5a190ae6 3527 new_dentry->d_inode, old_dentry);
0eeca283
RL		 3528 fsnotify_oldname_free(old_name);
3529
1da177e4
LT		 3530 return error;
3531}
3532
2e4d0924
HC		 3533SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3534 int, newdfd, const char __user *, newname)
1da177e4 3535{
2ad94ae6
AV		 3536 struct dentry *old_dir, *new_dir;
3537 struct dentry *old_dentry, *new_dentry;
3538 struct dentry *trap;
1da177e4 3539 struct nameidata oldnd, newnd;
2ad94ae6
AV		 3540 char *from;
3541 char *to;
3542 int error;
1da177e4 3543
2ad94ae6 3544 error = user_path_parent(olddfd, oldname, &oldnd, &from);
1da177e4
LT		 3545 if (error)
3546 goto exit;
3547
2ad94ae6 3548 error = user_path_parent(newdfd, newname, &newnd, &to);
1da177e4
LT		 3549 if (error)
3550 goto exit1;
3551
3552 error = -EXDEV;
4ac91378 3553 if (oldnd.path.mnt != newnd.path.mnt)
1da177e4
LT		 3554 goto exit2;
3555
4ac91378 3556 old_dir = oldnd.path.dentry;
1da177e4
LT		 3557 error = -EBUSY;
3558 if (oldnd.last_type != LAST_NORM)
3559 goto exit2;
3560
4ac91378 3561 new_dir = newnd.path.dentry;
1da177e4
LT		 3562 if (newnd.last_type != LAST_NORM)
3563 goto exit2;
3564
0612d9fb
OH		 3565 oldnd.flags &= ~LOOKUP_PARENT;
3566 newnd.flags &= ~LOOKUP_PARENT;
4e9ed2f8 3567 newnd.flags |= LOOKUP_RENAME_TARGET;
0612d9fb 3568
1da177e4
LT		 3569 trap = lock_rename(new_dir, old_dir);
3570
49705b77 3571 old_dentry = lookup_hash(&oldnd);
1da177e4
LT		 3572 error = PTR_ERR(old_dentry);
3573 if (IS_ERR(old_dentry))
3574 goto exit3;
3575 /* source must exist */
3576 error = -ENOENT;
3577 if (!old_dentry->d_inode)
3578 goto exit4;
3579 /* unless the source is a directory trailing slashes give -ENOTDIR */
3580 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3581 error = -ENOTDIR;
3582 if (oldnd.last.name[oldnd.last.len])
3583 goto exit4;
3584 if (newnd.last.name[newnd.last.len])
3585 goto exit4;
3586 }
3587 /* source should not be ancestor of target */
3588 error = -EINVAL;
3589 if (old_dentry == trap)
3590 goto exit4;
49705b77 3591 new_dentry = lookup_hash(&newnd);
1da177e4
LT		 3592 error = PTR_ERR(new_dentry);
3593 if (IS_ERR(new_dentry))
3594 goto exit4;
3595 /* target should not be an ancestor of source */
3596 error = -ENOTEMPTY;
3597 if (new_dentry == trap)
3598 goto exit5;
3599
9079b1eb
DH		 3600 error = mnt_want_write(oldnd.path.mnt);
3601 if (error)
3602 goto exit5;
be6d3e56
KT		 3603 error = security_path_rename(&oldnd.path, old_dentry,
3604 &newnd.path, new_dentry);
3605 if (error)
3606 goto exit6;
1da177e4
LT		 3607 error = vfs_rename(old_dir->d_inode, old_dentry,
3608 new_dir->d_inode, new_dentry);
be6d3e56 3609exit6:
9079b1eb 3610 mnt_drop_write(oldnd.path.mnt);
1da177e4
LT		 3611exit5:
3612 dput(new_dentry);
3613exit4:
3614 dput(old_dentry);
3615exit3:
3616 unlock_rename(new_dir, old_dir);
3617exit2:
1d957f9b 3618 path_put(&newnd.path);
2ad94ae6 3619 putname(to);
1da177e4 3620exit1:
1d957f9b 3621 path_put(&oldnd.path);
1da177e4 3622 putname(from);
2ad94ae6 3623exit:
1da177e4
LT		 3624 return error;
3625}
3626
a26eab24 3627SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 3628{
3629 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3630}
3631
1da177e4
LT		 3632int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3633{
3634 int len;
3635
3636 len = PTR_ERR(link);
3637 if (IS_ERR(link))
3638 goto out;
3639
3640 len = strlen(link);
3641 if (len > (unsigned) buflen)
3642 len = buflen;
3643 if (copy_to_user(buffer, link, len))
3644 len = -EFAULT;
3645out:
3646 return len;
3647}
3648
3649/*
3650 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3651 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3652 * using) it for any given inode is up to filesystem.
3653 */
3654int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3655{
3656 struct nameidata nd;
cc314eef 3657 void *cookie;
694a1764 3658 int res;
cc314eef 3659
1da177e4 3660 nd.depth = 0;
cc314eef 3661 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
694a1764
MS		 3662 if (IS_ERR(cookie))
3663 return PTR_ERR(cookie);
3664
3665 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3666 if (dentry->d_inode->i_op->put_link)
3667 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3668 return res;
1da177e4
LT		 3669}
3670
3671int vfs_follow_link(struct nameidata *nd, const char *link)
3672{
3673 return __vfs_follow_link(nd, link);
3674}
3675
3676/* get the link contents into pagecache */
3677static char *page_getlink(struct dentry * dentry, struct page **ppage)
3678{
ebd09abb
DG		 3679 char *kaddr;
3680 struct page *page;
1da177e4 3681 struct address_space *mapping = dentry->d_inode->i_mapping;
090d2b18 3682 page = read_mapping_page(mapping, 0, NULL);
1da177e4 3683 if (IS_ERR(page))
6fe6900e 3684 return (char*)page;
1da177e4 3685 *ppage = page;
ebd09abb
DG		 3686 kaddr = kmap(page);
3687 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3688 return kaddr;
1da177e4
LT		 3689}
3690
3691int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3692{
3693 struct page *page = NULL;
3694 char *s = page_getlink(dentry, &page);
3695 int res = vfs_readlink(dentry,buffer,buflen,s);
3696 if (page) {
3697 kunmap(page);
3698 page_cache_release(page);
3699 }
3700 return res;
3701}
3702
cc314eef 3703void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
1da177e4 3704{
cc314eef 3705 struct page *page = NULL;
1da177e4 3706 nd_set_link(nd, page_getlink(dentry, &page));
cc314eef 3707 return page;
1da177e4
LT		 3708}
3709
cc314eef 3710void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1da177e4 3711{
cc314eef
LT		 3712 struct page *page = cookie;
3713
3714 if (page) {
1da177e4
LT		 3715 kunmap(page);
3716 page_cache_release(page);
1da177e4
LT		 3717 }
3718}
3719
54566b2c
NP		 3720/*
3721 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3722 */
3723int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
1da177e4
LT		 3724{
3725 struct address_space *mapping = inode->i_mapping;
0adb25d2 3726 struct page *page;
afddba49 3727 void *fsdata;
beb497ab 3728 int err;
1da177e4 3729 char *kaddr;
54566b2c
NP		 3730 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3731 if (nofs)
3732 flags |= AOP_FLAG_NOFS;
1da177e4 3733
7e53cac4 3734retry:
afddba49 3735 err = pagecache_write_begin(NULL, mapping, 0, len-1,
54566b2c 3736 flags, &page, &fsdata);
1da177e4 3737 if (err)
afddba49
NP		 3738 goto fail;
3739
e8e3c3d6 3740 kaddr = kmap_atomic(page);
1da177e4 3741 memcpy(kaddr, symname, len-1);
e8e3c3d6 3742 kunmap_atomic(kaddr);
afddba49
NP		 3743
3744 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3745 page, fsdata);
1da177e4
LT		 3746 if (err < 0)
3747 goto fail;
afddba49
NP		 3748 if (err < len-1)
3749 goto retry;
3750
1da177e4
LT		 3751 mark_inode_dirty(inode);
3752 return 0;
1da177e4
LT		 3753fail:
3754 return err;
3755}
3756
0adb25d2
KK		 3757int page_symlink(struct inode *inode, const char *symname, int len)
3758{
3759 return __page_symlink(inode, symname, len,
54566b2c 3760 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
0adb25d2
KK		 3761}
3762
92e1d5be 3763const struct inode_operations page_symlink_inode_operations = {
1da177e4
LT		 3764 .readlink = generic_readlink,
3765 .follow_link = page_follow_link_light,
3766 .put_link = page_put_link,
3767};
3768
2d8f3038 3769EXPORT_SYMBOL(user_path_at);
cc53ce53 3770EXPORT_SYMBOL(follow_down_one);
1da177e4
LT		 3771EXPORT_SYMBOL(follow_down);
3772EXPORT_SYMBOL(follow_up);
3773EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3774EXPORT_SYMBOL(getname);
3775EXPORT_SYMBOL(lock_rename);
1da177e4
LT		 3776EXPORT_SYMBOL(lookup_one_len);
3777EXPORT_SYMBOL(page_follow_link_light);
3778EXPORT_SYMBOL(page_put_link);
3779EXPORT_SYMBOL(page_readlink);
0adb25d2 3780EXPORT_SYMBOL(__page_symlink);
1da177e4
LT		 3781EXPORT_SYMBOL(page_symlink);
3782EXPORT_SYMBOL(page_symlink_inode_operations);
d1811465 3783EXPORT_SYMBOL(kern_path);
16f18200 3784EXPORT_SYMBOL(vfs_path_lookup);
f419a2e3 3785EXPORT_SYMBOL(inode_permission);
1da177e4
LT		 3786EXPORT_SYMBOL(unlock_rename);
3787EXPORT_SYMBOL(vfs_create);
3788EXPORT_SYMBOL(vfs_follow_link);
3789EXPORT_SYMBOL(vfs_link);
3790EXPORT_SYMBOL(vfs_mkdir);
3791EXPORT_SYMBOL(vfs_mknod);
3792EXPORT_SYMBOL(generic_permission);
3793EXPORT_SYMBOL(vfs_readlink);
3794EXPORT_SYMBOL(vfs_rename);
3795EXPORT_SYMBOL(vfs_rmdir);
3796EXPORT_SYMBOL(vfs_symlink);
3797EXPORT_SYMBOL(vfs_unlink);
3798EXPORT_SYMBOL(dentry_unhash);
3799EXPORT_SYMBOL(generic_readlink);
kernel source for telekom puls (alcatel/mediatek)