projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
userns: Start out with a full set of capabilities.
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / fs / namei.c
CommitLineData
1da177e4
LT		 1/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
630d9c47 18#include <linux/export.h>
1da177e4
LT		 19#include <linux/slab.h>
20#include <linux/fs.h>
21#include <linux/namei.h>
1da177e4 22#include <linux/pagemap.h>
0eeca283 23#include <linux/fsnotify.h>
1da177e4
LT		 24#include <linux/personality.h>
25#include <linux/security.h>
6146f0d5 26#include <linux/ima.h>
1da177e4
LT		 27#include <linux/syscalls.h>
28#include <linux/mount.h>
29#include <linux/audit.h>
16f7e0fe 30#include <linux/capability.h>
834f2a4a 31#include <linux/file.h>
5590ff0d 32#include <linux/fcntl.h>
08ce5f16 33#include <linux/device_cgroup.h>
5ad4e53b 34#include <linux/fs_struct.h>
e77819e5 35#include <linux/posix_acl.h>
1da177e4
LT		 36#include <asm/uaccess.h>
37
e81e3f4d 38#include "internal.h"
c7105365 39#include "mount.h"
e81e3f4d 40
1da177e4
LT		 41/* [Feb-1997 T. Schoebel-Theuer]
42 * Fundamental changes in the pathname lookup mechanisms (namei)
43 * were necessary because of omirr. The reason is that omirr needs
44 * to know the _real_ pathname, not the user-supplied one, in case
45 * of symlinks (and also when transname replacements occur).
46 *
47 * The new code replaces the old recursive symlink resolution with
48 * an iterative one (in case of non-nested symlink chains). It does
49 * this with calls to <fs>_follow_link().
50 * As a side effect, dir_namei(), _namei() and follow_link() are now
51 * replaced with a single function lookup_dentry() that can handle all
52 * the special cases of the former code.
53 *
54 * With the new dcache, the pathname is stored at each inode, at least as
55 * long as the refcount of the inode is positive. As a side effect, the
56 * size of the dcache depends on the inode cache and thus is dynamic.
57 *
58 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
59 * resolution to correspond with current state of the code.
60 *
61 * Note that the symlink resolution is not *completely* iterative.
62 * There is still a significant amount of tail- and mid- recursion in
63 * the algorithm. Also, note that <fs>_readlink() is not used in
64 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
65 * may return different results than <fs>_follow_link(). Many virtual
66 * filesystems (including /proc) exhibit this behavior.
67 */
68
69/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
70 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
71 * and the name already exists in form of a symlink, try to create the new
72 * name indicated by the symlink. The old code always complained that the
73 * name already exists, due to not following the symlink even if its target
74 * is nonexistent. The new semantics affects also mknod() and link() when
25985edc 75 * the name is a symlink pointing to a non-existent name.
1da177e4
LT		 76 *
77 * I don't know which semantics is the right one, since I have no access
78 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
79 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
80 * "old" one. Personally, I think the new semantics is much more logical.
81 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
82 * file does succeed in both HP-UX and SunOs, but not in Solaris
83 * and in the old Linux semantics.
84 */
85
86/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
87 * semantics. See the comments in "open_namei" and "do_link" below.
88 *
89 * [10-Sep-98 Alan Modra] Another symlink change.
90 */
91
92/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
93 * inside the path - always follow.
94 * in the last component in creation/removal/renaming - never follow.
95 * if LOOKUP_FOLLOW passed - follow.
96 * if the pathname has trailing slashes - follow.
97 * otherwise - don't follow.
98 * (applied in that order).
99 *
100 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
101 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
102 * During the 2.4 we need to fix the userland stuff depending on it -
103 * hopefully we will be able to get rid of that wart in 2.5. So far only
104 * XEmacs seems to be relying on it...
105 */
106/*
107 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
a11f3a05 108 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
1da177e4
LT		 109 * any extra contention...
110 */
111
112/* In order to reduce some races, while at the same time doing additional
113 * checking and hopefully speeding things up, we copy filenames to the
114 * kernel data space before using them..
115 *
116 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
117 * PATH_MAX includes the nul terminator --RR.
118 */
858119e1 119static int do_getname(const char __user *filename, char *page)
1da177e4
LT		 120{
121 int retval;
122 unsigned long len = PATH_MAX;
123
124 if (!segment_eq(get_fs(), KERNEL_DS)) {
125 if ((unsigned long) filename >= TASK_SIZE)
126 return -EFAULT;
127 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
128 len = TASK_SIZE - (unsigned long) filename;
129 }
130
131 retval = strncpy_from_user(page, filename, len);
132 if (retval > 0) {
133 if (retval < len)
134 return 0;
135 return -ENAMETOOLONG;
136 } else if (!retval)
137 retval = -ENOENT;
138 return retval;
139}
140
1fa1e7f6 141static char *getname_flags(const char __user *filename, int flags, int *empty)
1da177e4 142{
4043cde8
EP		 143 char *result = __getname();
144 int retval;
145
146 if (!result)
147 return ERR_PTR(-ENOMEM);
148
149 retval = do_getname(filename, result);
150 if (retval < 0) {
151 if (retval == -ENOENT && empty)
152 *empty = 1;
153 if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) {
154 __putname(result);
155 return ERR_PTR(retval);
1da177e4
LT		 156 }
157 }
158 audit_getname(result);
159 return result;
160}
161
f52e0c11
AV		 162char *getname(const char __user * filename)
163{
f7493e5d 164 return getname_flags(filename, 0, NULL);
f52e0c11
AV		 165}
166
1da177e4
LT		 167#ifdef CONFIG_AUDITSYSCALL
168void putname(const char *name)
169{
5ac3a9c2 170 if (unlikely(!audit_dummy_context()))
1da177e4
LT		 171 audit_putname(name);
172 else
173 __putname(name);
174}
175EXPORT_SYMBOL(putname);
176#endif
177
e77819e5
LT		 178static int check_acl(struct inode *inode, int mask)
179{
84635d68 180#ifdef CONFIG_FS_POSIX_ACL
e77819e5
LT		 181 struct posix_acl *acl;
182
e77819e5 183 if (mask & MAY_NOT_BLOCK) {
3567866b
AV		 184 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
185 if (!acl)
e77819e5 186 return -EAGAIN;
3567866b
AV		 187 /* no ->get_acl() calls in RCU mode... */
188 if (acl == ACL_NOT_CACHED)
189 return -ECHILD;
206b1d09 190 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
e77819e5
LT		 191 }
192
193 acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
194
195 /*
4e34e719
CH		 196 * A filesystem can force a ACL callback by just never filling the
197 * ACL cache. But normally you'd fill the cache either at inode
198 * instantiation time, or on the first ->get_acl call.
e77819e5 199 *
4e34e719
CH		 200 * If the filesystem doesn't have a get_acl() function at all, we'll
201 * just create the negative cache entry.
e77819e5
LT		 202 */
203 if (acl == ACL_NOT_CACHED) {
4e34e719
CH		 204 if (inode->i_op->get_acl) {
205 acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS);
206 if (IS_ERR(acl))
207 return PTR_ERR(acl);
208 } else {
209 set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
210 return -EAGAIN;
211 }
e77819e5
LT		 212 }
213
214 if (acl) {
215 int error = posix_acl_permission(inode, acl, mask);
216 posix_acl_release(acl);
217 return error;
218 }
84635d68 219#endif
e77819e5
LT		 220
221 return -EAGAIN;
222}
223
5909ccaa 224/*
948409c7 225 * This does the basic permission checking
1da177e4 226 */
7e40145e 227static int acl_permission_check(struct inode *inode, int mask)
1da177e4 228{
26cf46be 229 unsigned int mode = inode->i_mode;
1da177e4 230
e795b717
SH		 231 if (current_user_ns() != inode_userns(inode))
232 goto other_perms;
233
14067ff5 234 if (likely(current_fsuid() == inode->i_uid))
1da177e4
LT		 235 mode >>= 6;
236 else {
e77819e5 237 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
7e40145e 238 int error = check_acl(inode, mask);
b74c79e9
NP		 239 if (error != -EAGAIN)
240 return error;
1da177e4
LT		 241 }
242
243 if (in_group_p(inode->i_gid))
244 mode >>= 3;
245 }
246
e795b717 247other_perms:
1da177e4
LT		 248 /*
249 * If the DACs are ok we don't need any capability check.
250 */
9c2c7039 251 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
1da177e4 252 return 0;
5909ccaa
LT		 253 return -EACCES;
254}
255
256/**
b74c79e9 257 * generic_permission - check for access rights on a Posix-like filesystem
5909ccaa 258 * @inode: inode to check access rights for
8fd90c8d 259 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
5909ccaa
LT		 260 *
261 * Used to check for read/write/execute permissions on a file.
262 * We use "fsuid" for this, letting us set arbitrary permissions
263 * for filesystem access without changing the "normal" uids which
b74c79e9
NP		 264 * are used for other things.
265 *
266 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
267 * request cannot be satisfied (eg. requires blocking or too much complexity).
268 * It would then be called again in ref-walk mode.
5909ccaa 269 */
2830ba7f 270int generic_permission(struct inode *inode, int mask)
5909ccaa
LT		 271{
272 int ret;
273
274 /*
948409c7 275 * Do the basic permission checks.
5909ccaa 276 */
7e40145e 277 ret = acl_permission_check(inode, mask);
5909ccaa
LT		 278 if (ret != -EACCES)
279 return ret;
1da177e4 280
d594e7ec
AV		 281 if (S_ISDIR(inode->i_mode)) {
282 /* DACs are overridable for directories */
283 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
284 return 0;
285 if (!(mask & MAY_WRITE))
286 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
287 return 0;
288 return -EACCES;
289 }
1da177e4
LT		 290 /*
291 * Read/write DACs are always overridable.
d594e7ec
AV		 292 * Executable DACs are overridable when there is
293 * at least one exec bit set.
1da177e4 294 */
d594e7ec 295 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
e795b717 296 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
1da177e4
LT		 297 return 0;
298
299 /*
300 * Searching includes executable on directories, else just read.
301 */
7ea66001 302 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
d594e7ec 303 if (mask == MAY_READ)
e795b717 304 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
1da177e4
LT		 305 return 0;
306
307 return -EACCES;
308}
309
3ddcd056
LT		 310/*
311 * We _really_ want to just do "generic_permission()" without
312 * even looking at the inode->i_op values. So we keep a cache
313 * flag in inode->i_opflags, that says "this has not special
314 * permission function, use the fast case".
315 */
316static inline int do_inode_permission(struct inode *inode, int mask)
317{
318 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
319 if (likely(inode->i_op->permission))
320 return inode->i_op->permission(inode, mask);
321
322 /* This gets set once for the inode lifetime */
323 spin_lock(&inode->i_lock);
324 inode->i_opflags |= IOP_FASTPERM;
325 spin_unlock(&inode->i_lock);
326 }
327 return generic_permission(inode, mask);
328}
329
cb23beb5
CH		 330/**
331 * inode_permission - check for access rights to a given inode
332 * @inode: inode to check permission on
8fd90c8d 333 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
cb23beb5
CH		 334 *
335 * Used to check for read/write/execute permissions on an inode.
336 * We use "fsuid" for this, letting us set arbitrary permissions
337 * for filesystem access without changing the "normal" uids which
338 * are used for other things.
948409c7
AG		 339 *
340 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
cb23beb5 341 */
f419a2e3 342int inode_permission(struct inode *inode, int mask)
1da177e4 343{
e6305c43 344 int retval;
1da177e4 345
3ddcd056 346 if (unlikely(mask & MAY_WRITE)) {
22590e41 347 umode_t mode = inode->i_mode;
1da177e4
LT		 348
349 /*
350 * Nobody gets write access to a read-only fs.
351 */
352 if (IS_RDONLY(inode) &&
353 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
354 return -EROFS;
355
356 /*
357 * Nobody gets write access to an immutable file.
358 */
359 if (IS_IMMUTABLE(inode))
360 return -EACCES;
361 }
362
3ddcd056 363 retval = do_inode_permission(inode, mask);
1da177e4
LT		 364 if (retval)
365 return retval;
366
08ce5f16
SH		 367 retval = devcgroup_inode_permission(inode, mask);
368 if (retval)
369 return retval;
370
d09ca739 371 return security_inode_permission(inode, mask);
1da177e4
LT		 372}
373
5dd784d0
JB		 374/**
375 * path_get - get a reference to a path
376 * @path: path to get the reference to
377 *
378 * Given a path increment the reference count to the dentry and the vfsmount.
379 */
380void path_get(struct path *path)
381{
382 mntget(path->mnt);
383 dget(path->dentry);
384}
385EXPORT_SYMBOL(path_get);
386
1d957f9b
JB		 387/**
388 * path_put - put a reference to a path
389 * @path: path to put the reference to
390 *
391 * Given a path decrement the reference count to the dentry and the vfsmount.
392 */
393void path_put(struct path *path)
1da177e4 394{
1d957f9b
JB		 395 dput(path->dentry);
396 mntput(path->mnt);
1da177e4 397}
1d957f9b 398EXPORT_SYMBOL(path_put);
1da177e4 399
19660af7 400/*
31e6b01f 401 * Path walking has 2 modes, rcu-walk and ref-walk (see
19660af7
AV		 402 * Documentation/filesystems/path-lookup.txt). In situations when we can't
403 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
404 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
405 * mode. Refcounts are grabbed at the last known good point before rcu-walk
406 * got stuck, so ref-walk may continue from there. If this is not successful
407 * (eg. a seqcount has changed), then failure is returned and it's up to caller
408 * to restart the path walk from the beginning in ref-walk mode.
31e6b01f 409 */
31e6b01f
NP		 410
411/**
19660af7
AV		 412 * unlazy_walk - try to switch to ref-walk mode.
413 * @nd: nameidata pathwalk data
414 * @dentry: child of nd->path.dentry or NULL
39191628 415 * Returns: 0 on success, -ECHILD on failure
31e6b01f 416 *
19660af7
AV		 417 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
418 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
419 * @nd or NULL. Must be called from rcu-walk context.
31e6b01f 420 */
19660af7 421static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
31e6b01f
NP		 422{
423 struct fs_struct *fs = current->fs;
424 struct dentry *parent = nd->path.dentry;
5b6ca027 425 int want_root = 0;
31e6b01f
NP		 426
427 BUG_ON(!(nd->flags & LOOKUP_RCU));
5b6ca027
AV		 428 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
429 want_root = 1;
31e6b01f
NP		 430 spin_lock(&fs->lock);
431 if (nd->root.mnt != fs->root.mnt ||
432 nd->root.dentry != fs->root.dentry)
433 goto err_root;
434 }
435 spin_lock(&parent->d_lock);
19660af7
AV		 436 if (!dentry) {
437 if (!__d_rcu_to_refcount(parent, nd->seq))
438 goto err_parent;
439 BUG_ON(nd->inode != parent->d_inode);
440 } else {
94c0d4ec
AV		 441 if (dentry->d_parent != parent)
442 goto err_parent;
19660af7
AV		 443 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
444 if (!__d_rcu_to_refcount(dentry, nd->seq))
445 goto err_child;
446 /*
447 * If the sequence check on the child dentry passed, then
448 * the child has not been removed from its parent. This
449 * means the parent dentry must be valid and able to take
450 * a reference at this point.
451 */
452 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
453 BUG_ON(!parent->d_count);
454 parent->d_count++;
455 spin_unlock(&dentry->d_lock);
456 }
31e6b01f 457 spin_unlock(&parent->d_lock);
5b6ca027 458 if (want_root) {
31e6b01f
NP		 459 path_get(&nd->root);
460 spin_unlock(&fs->lock);
461 }
462 mntget(nd->path.mnt);
463
464 rcu_read_unlock();
465 br_read_unlock(vfsmount_lock);
466 nd->flags &= ~LOOKUP_RCU;
467 return 0;
19660af7
AV		 468
469err_child:
31e6b01f 470 spin_unlock(&dentry->d_lock);
19660af7 471err_parent:
31e6b01f
NP		 472 spin_unlock(&parent->d_lock);
473err_root:
5b6ca027 474 if (want_root)
31e6b01f
NP		 475 spin_unlock(&fs->lock);
476 return -ECHILD;
477}
478
834f2a4a
TM		 479/**
480 * release_open_intent - free up open intent resources
481 * @nd: pointer to nameidata
482 */
483void release_open_intent(struct nameidata *nd)
484{
2dab5974
LT		 485 struct file *file = nd->intent.open.file;
486
487 if (file && !IS_ERR(file)) {
488 if (file->f_path.dentry == NULL)
489 put_filp(file);
490 else
491 fput(file);
492 }
834f2a4a
TM		 493}
494
f60aef7e 495static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
34286d66 496{
f60aef7e 497 return dentry->d_op->d_revalidate(dentry, nd);
34286d66
NP		 498}
499
9f1fafee
AV		 500/**
501 * complete_walk - successful completion of path walk
502 * @nd: pointer nameidata
39159de2 503 *
9f1fafee
AV		 504 * If we had been in RCU mode, drop out of it and legitimize nd->path.
505 * Revalidate the final result, unless we'd already done that during
506 * the path walk or the filesystem doesn't ask for it. Return 0 on
507 * success, -error on failure. In case of failure caller does not
508 * need to drop nd->path.
39159de2 509 */
9f1fafee 510static int complete_walk(struct nameidata *nd)
39159de2 511{
16c2cd71 512 struct dentry *dentry = nd->path.dentry;
39159de2 513 int status;
39159de2 514
9f1fafee
AV		 515 if (nd->flags & LOOKUP_RCU) {
516 nd->flags &= ~LOOKUP_RCU;
517 if (!(nd->flags & LOOKUP_ROOT))
518 nd->root.mnt = NULL;
519 spin_lock(&dentry->d_lock);
520 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
521 spin_unlock(&dentry->d_lock);
522 rcu_read_unlock();
523 br_read_unlock(vfsmount_lock);
524 return -ECHILD;
525 }
526 BUG_ON(nd->inode != dentry->d_inode);
527 spin_unlock(&dentry->d_lock);
528 mntget(nd->path.mnt);
529 rcu_read_unlock();
530 br_read_unlock(vfsmount_lock);
531 }
532
16c2cd71
AV		 533 if (likely(!(nd->flags & LOOKUP_JUMPED)))
534 return 0;
535
536 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
39159de2
JL		 537 return 0;
538
16c2cd71
AV		 539 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
540 return 0;
541
542 /* Note: we do not d_invalidate() */
34286d66 543 status = d_revalidate(dentry, nd);
39159de2
JL		 544 if (status > 0)
545 return 0;
546
16c2cd71 547 if (!status)
39159de2 548 status = -ESTALE;
16c2cd71 549
9f1fafee 550 path_put(&nd->path);
39159de2
JL		 551 return status;
552}
553
2a737871
AV		 554static __always_inline void set_root(struct nameidata *nd)
555{
f7ad3c6b
MS		 556 if (!nd->root.mnt)
557 get_fs_root(current->fs, &nd->root);
2a737871
AV		 558}
559
6de88d72
AV		 560static int link_path_walk(const char *, struct nameidata *);
561
31e6b01f
NP		 562static __always_inline void set_root_rcu(struct nameidata *nd)
563{
564 if (!nd->root.mnt) {
565 struct fs_struct *fs = current->fs;
c28cc364
NP		 566 unsigned seq;
567
568 do {
569 seq = read_seqcount_begin(&fs->seq);
570 nd->root = fs->root;
c1530019 571 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
c28cc364 572 } while (read_seqcount_retry(&fs->seq, seq));
31e6b01f
NP		 573 }
574}
575
f1662356 576static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
1da177e4 577{
31e6b01f
NP		 578 int ret;
579
1da177e4
LT		 580 if (IS_ERR(link))
581 goto fail;
582
583 if (*link == '/') {
2a737871 584 set_root(nd);
1d957f9b 585 path_put(&nd->path);
2a737871
AV		 586 nd->path = nd->root;
587 path_get(&nd->root);
16c2cd71 588 nd->flags |= LOOKUP_JUMPED;
1da177e4 589 }
31e6b01f 590 nd->inode = nd->path.dentry->d_inode;
b4091d5f 591
31e6b01f
NP		 592 ret = link_path_walk(link, nd);
593 return ret;
1da177e4 594fail:
1d957f9b 595 path_put(&nd->path);
1da177e4
LT		 596 return PTR_ERR(link);
597}
598
1d957f9b 599static void path_put_conditional(struct path *path, struct nameidata *nd)
051d3812
IK		 600{
601 dput(path->dentry);
4ac91378 602 if (path->mnt != nd->path.mnt)
051d3812
IK		 603 mntput(path->mnt);
604}
605
7b9337aa
NP		 606static inline void path_to_nameidata(const struct path *path,
607 struct nameidata *nd)
051d3812 608{
31e6b01f
NP		 609 if (!(nd->flags & LOOKUP_RCU)) {
610 dput(nd->path.dentry);
611 if (nd->path.mnt != path->mnt)
612 mntput(nd->path.mnt);
9a229683 613 }
31e6b01f 614 nd->path.mnt = path->mnt;
4ac91378 615 nd->path.dentry = path->dentry;
051d3812
IK		 616}
617
574197e0
AV		 618static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
619{
620 struct inode *inode = link->dentry->d_inode;
621 if (!IS_ERR(cookie) && inode->i_op->put_link)
622 inode->i_op->put_link(link->dentry, nd, cookie);
623 path_put(link);
624}
625
def4af30 626static __always_inline int
574197e0 627follow_link(struct path *link, struct nameidata *nd, void **p)
1da177e4
LT		 628{
629 int error;
7b9337aa 630 struct dentry *dentry = link->dentry;
1da177e4 631
844a3917
AV		 632 BUG_ON(nd->flags & LOOKUP_RCU);
633
0e794589
AV		 634 if (link->mnt == nd->path.mnt)
635 mntget(link->mnt);
636
574197e0
AV		 637 if (unlikely(current->total_link_count >= 40)) {
638 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
574197e0
AV		 639 path_put(&nd->path);
640 return -ELOOP;
641 }
642 cond_resched();
643 current->total_link_count++;
644
68ac1234 645 touch_atime(link);
1da177e4 646 nd_set_link(nd, NULL);
cd4e91d3 647
36f3b4f6
AV		 648 error = security_inode_follow_link(link->dentry, nd);
649 if (error) {
650 *p = ERR_PTR(error); /* no ->put_link(), please */
651 path_put(&nd->path);
652 return error;
653 }
654
86acdca1 655 nd->last_type = LAST_BIND;
def4af30
AV		 656 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
657 error = PTR_ERR(*p);
658 if (!IS_ERR(*p)) {
1da177e4 659 char *s = nd_get_link(nd);
cc314eef 660 error = 0;
1da177e4
LT		 661 if (s)
662 error = __vfs_follow_link(nd, s);
bcda7652 663 else if (nd->last_type == LAST_BIND) {
16c2cd71 664 nd->flags |= LOOKUP_JUMPED;
b21041d0
AV		 665 nd->inode = nd->path.dentry->d_inode;
666 if (nd->inode->i_op->follow_link) {
bcda7652
AV		 667 /* stepped on a _really_ weird one */
668 path_put(&nd->path);
669 error = -ELOOP;
670 }
671 }
1da177e4 672 }
1da177e4
LT		 673 return error;
674}
675
31e6b01f
NP		 676static int follow_up_rcu(struct path *path)
677{
0714a533
AV		 678 struct mount *mnt = real_mount(path->mnt);
679 struct mount *parent;
31e6b01f
NP		 680 struct dentry *mountpoint;
681
0714a533
AV		 682 parent = mnt->mnt_parent;
683 if (&parent->mnt == path->mnt)
31e6b01f 684 return 0;
a73324da 685 mountpoint = mnt->mnt_mountpoint;
31e6b01f 686 path->dentry = mountpoint;
0714a533 687 path->mnt = &parent->mnt;
31e6b01f
NP		 688 return 1;
689}
690
bab77ebf 691int follow_up(struct path *path)
1da177e4 692{
0714a533
AV		 693 struct mount *mnt = real_mount(path->mnt);
694 struct mount *parent;
1da177e4 695 struct dentry *mountpoint;
99b7db7b
NP		 696
697 br_read_lock(vfsmount_lock);
0714a533
AV		 698 parent = mnt->mnt_parent;
699 if (&parent->mnt == path->mnt) {
99b7db7b 700 br_read_unlock(vfsmount_lock);
1da177e4
LT		 701 return 0;
702 }
0714a533 703 mntget(&parent->mnt);
a73324da 704 mountpoint = dget(mnt->mnt_mountpoint);
99b7db7b 705 br_read_unlock(vfsmount_lock);
bab77ebf
AV		 706 dput(path->dentry);
707 path->dentry = mountpoint;
708 mntput(path->mnt);
0714a533 709 path->mnt = &parent->mnt;
1da177e4
LT		 710 return 1;
711}
712
b5c84bf6 713/*
9875cf80
DH		 714 * Perform an automount
715 * - return -EISDIR to tell follow_managed() to stop and return the path we
716 * were called with.
1da177e4 717 */
9875cf80
DH		 718static int follow_automount(struct path *path, unsigned flags,
719 bool *need_mntput)
31e6b01f 720{
9875cf80 721 struct vfsmount *mnt;
ea5b778a 722 int err;
9875cf80
DH		 723
724 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
725 return -EREMOTE;
726
0ec26fd0
MS		 727 /* We don't want to mount if someone's just doing a stat -
728 * unless they're stat'ing a directory and appended a '/' to
729 * the name.
730 *
731 * We do, however, want to mount if someone wants to open or
732 * create a file of any type under the mountpoint, wants to
733 * traverse through the mountpoint or wants to open the
734 * mounted directory. Also, autofs may mark negative dentries
735 * as being automount points. These will need the attentions
736 * of the daemon to instantiate them before they can be used.
9875cf80 737 */
0ec26fd0 738 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
d94c177b 739 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
0ec26fd0
MS		 740 path->dentry->d_inode)
741 return -EISDIR;
742
9875cf80
DH		 743 current->total_link_count++;
744 if (current->total_link_count >= 40)
745 return -ELOOP;
746
747 mnt = path->dentry->d_op->d_automount(path);
748 if (IS_ERR(mnt)) {
749 /*
750 * The filesystem is allowed to return -EISDIR here to indicate
751 * it doesn't want to automount. For instance, autofs would do
752 * this so that its userspace daemon can mount on this dentry.
753 *
754 * However, we can only permit this if it's a terminal point in
755 * the path being looked up; if it wasn't then the remainder of
756 * the path is inaccessible and we should say so.
757 */
49084c3b 758 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
9875cf80
DH		 759 return -EREMOTE;
760 return PTR_ERR(mnt);
31e6b01f 761 }
ea5b778a 762
9875cf80
DH		 763 if (!mnt) /* mount collision */
764 return 0;
31e6b01f 765
8aef1884
AV		 766 if (!*need_mntput) {
767 /* lock_mount() may release path->mnt on error */
768 mntget(path->mnt);
769 *need_mntput = true;
770 }
19a167af 771 err = finish_automount(mnt, path);
9875cf80 772
ea5b778a
DH		 773 switch (err) {
774 case -EBUSY:
775 /* Someone else made a mount here whilst we were busy */
19a167af 776 return 0;
ea5b778a 777 case 0:
8aef1884 778 path_put(path);
ea5b778a
DH		 779 path->mnt = mnt;
780 path->dentry = dget(mnt->mnt_root);
ea5b778a 781 return 0;
19a167af
AV		 782 default:
783 return err;
ea5b778a 784 }
19a167af 785
463ffb2e
AV		 786}
787
9875cf80
DH		 788/*
789 * Handle a dentry that is managed in some way.
cc53ce53 790 * - Flagged for transit management (autofs)
9875cf80
DH		 791 * - Flagged as mountpoint
792 * - Flagged as automount point
793 *
794 * This may only be called in refwalk mode.
795 *
796 * Serialization is taken care of in namespace.c
797 */
798static int follow_managed(struct path *path, unsigned flags)
1da177e4 799{
8aef1884 800 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
9875cf80
DH		 801 unsigned managed;
802 bool need_mntput = false;
8aef1884 803 int ret = 0;
9875cf80
DH		 804
805 /* Given that we're not holding a lock here, we retain the value in a
806 * local variable for each dentry as we look at it so that we don't see
807 * the components of that value change under us */
808 while (managed = ACCESS_ONCE(path->dentry->d_flags),
809 managed &= DCACHE_MANAGED_DENTRY,
810 unlikely(managed != 0)) {
cc53ce53
DH		 811 /* Allow the filesystem to manage the transit without i_mutex
812 * being held. */
813 if (managed & DCACHE_MANAGE_TRANSIT) {
814 BUG_ON(!path->dentry->d_op);
815 BUG_ON(!path->dentry->d_op->d_manage);
1aed3e42 816 ret = path->dentry->d_op->d_manage(path->dentry, false);
cc53ce53 817 if (ret < 0)
8aef1884 818 break;
cc53ce53
DH		 819 }
820
9875cf80
DH		 821 /* Transit to a mounted filesystem. */
822 if (managed & DCACHE_MOUNTED) {
823 struct vfsmount *mounted = lookup_mnt(path);
824 if (mounted) {
825 dput(path->dentry);
826 if (need_mntput)
827 mntput(path->mnt);
828 path->mnt = mounted;
829 path->dentry = dget(mounted->mnt_root);
830 need_mntput = true;
831 continue;
832 }
833
834 /* Something is mounted on this dentry in another
835 * namespace and/or whatever was mounted there in this
836 * namespace got unmounted before we managed to get the
837 * vfsmount_lock */
838 }
839
840 /* Handle an automount point */
841 if (managed & DCACHE_NEED_AUTOMOUNT) {
842 ret = follow_automount(path, flags, &need_mntput);
843 if (ret < 0)
8aef1884 844 break;
9875cf80
DH		 845 continue;
846 }
847
848 /* We didn't change the current path point */
849 break;
1da177e4 850 }
8aef1884
AV		 851
852 if (need_mntput && path->mnt == mnt)
853 mntput(path->mnt);
854 if (ret == -EISDIR)
855 ret = 0;
a3fbbde7 856 return ret < 0 ? ret : need_mntput;
1da177e4
LT		 857}
858
cc53ce53 859int follow_down_one(struct path *path)
1da177e4
LT		 860{
861 struct vfsmount *mounted;
862
1c755af4 863 mounted = lookup_mnt(path);
1da177e4 864 if (mounted) {
9393bd07
AV		 865 dput(path->dentry);
866 mntput(path->mnt);
867 path->mnt = mounted;
868 path->dentry = dget(mounted->mnt_root);
1da177e4
LT		 869 return 1;
870 }
871 return 0;
872}
873
62a7375e
IK		 874static inline bool managed_dentry_might_block(struct dentry *dentry)
875{
876 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
877 dentry->d_op->d_manage(dentry, true) < 0);
878}
879
9875cf80 880/*
287548e4
AV		 881 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
882 * we meet a managed dentry that would need blocking.
9875cf80
DH		 883 */
884static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
287548e4 885 struct inode **inode)
9875cf80 886{
62a7375e 887 for (;;) {
c7105365 888 struct mount *mounted;
62a7375e
IK		 889 /*
890 * Don't forget we might have a non-mountpoint managed dentry
891 * that wants to block transit.
892 */
287548e4 893 if (unlikely(managed_dentry_might_block(path->dentry)))
ab90911f 894 return false;
62a7375e
IK		 895
896 if (!d_mountpoint(path->dentry))
897 break;
898
9875cf80
DH		 899 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
900 if (!mounted)
901 break;
c7105365
AV		 902 path->mnt = &mounted->mnt;
903 path->dentry = mounted->mnt.mnt_root;
a3fbbde7 904 nd->flags |= LOOKUP_JUMPED;
9875cf80 905 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
59430262
LT		 906 /*
907 * Update the inode too. We don't need to re-check the
908 * dentry sequence number here after this d_inode read,
909 * because a mount-point is always pinned.
910 */
911 *inode = path->dentry->d_inode;
9875cf80 912 }
9875cf80
DH		 913 return true;
914}
915
dea39376 916static void follow_mount_rcu(struct nameidata *nd)
287548e4 917{
dea39376 918 while (d_mountpoint(nd->path.dentry)) {
c7105365 919 struct mount *mounted;
dea39376 920 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
287548e4
AV		 921 if (!mounted)
922 break;
c7105365
AV		 923 nd->path.mnt = &mounted->mnt;
924 nd->path.dentry = mounted->mnt.mnt_root;
dea39376 925 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
287548e4
AV		 926 }
927}
928
31e6b01f
NP		 929static int follow_dotdot_rcu(struct nameidata *nd)
930{
31e6b01f
NP		 931 set_root_rcu(nd);
932
9875cf80 933 while (1) {
31e6b01f
NP		 934 if (nd->path.dentry == nd->root.dentry &&
935 nd->path.mnt == nd->root.mnt) {
936 break;
937 }
938 if (nd->path.dentry != nd->path.mnt->mnt_root) {
939 struct dentry *old = nd->path.dentry;
940 struct dentry *parent = old->d_parent;
941 unsigned seq;
942
943 seq = read_seqcount_begin(&parent->d_seq);
944 if (read_seqcount_retry(&old->d_seq, nd->seq))
ef7562d5 945 goto failed;
31e6b01f
NP		 946 nd->path.dentry = parent;
947 nd->seq = seq;
948 break;
949 }
950 if (!follow_up_rcu(&nd->path))
951 break;
952 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
31e6b01f 953 }
dea39376
AV		 954 follow_mount_rcu(nd);
955 nd->inode = nd->path.dentry->d_inode;
31e6b01f 956 return 0;
ef7562d5
AV		 957
958failed:
959 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 960 if (!(nd->flags & LOOKUP_ROOT))
961 nd->root.mnt = NULL;
ef7562d5
AV		 962 rcu_read_unlock();
963 br_read_unlock(vfsmount_lock);
964 return -ECHILD;
31e6b01f
NP		 965}
966
cc53ce53
DH		 967/*
968 * Follow down to the covering mount currently visible to userspace. At each
969 * point, the filesystem owning that dentry may be queried as to whether the
970 * caller is permitted to proceed or not.
cc53ce53 971 */
7cc90cc3 972int follow_down(struct path *path)
cc53ce53
DH		 973{
974 unsigned managed;
975 int ret;
976
977 while (managed = ACCESS_ONCE(path->dentry->d_flags),
978 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
979 /* Allow the filesystem to manage the transit without i_mutex
980 * being held.
981 *
982 * We indicate to the filesystem if someone is trying to mount
983 * something here. This gives autofs the chance to deny anyone
984 * other than its daemon the right to mount on its
985 * superstructure.
986 *
987 * The filesystem may sleep at this point.
988 */
989 if (managed & DCACHE_MANAGE_TRANSIT) {
990 BUG_ON(!path->dentry->d_op);
991 BUG_ON(!path->dentry->d_op->d_manage);
ab90911f 992 ret = path->dentry->d_op->d_manage(
1aed3e42 993 path->dentry, false);
cc53ce53
DH		 994 if (ret < 0)
995 return ret == -EISDIR ? 0 : ret;
996 }
997
998 /* Transit to a mounted filesystem. */
999 if (managed & DCACHE_MOUNTED) {
1000 struct vfsmount *mounted = lookup_mnt(path);
1001 if (!mounted)
1002 break;
1003 dput(path->dentry);
1004 mntput(path->mnt);
1005 path->mnt = mounted;
1006 path->dentry = dget(mounted->mnt_root);
1007 continue;
1008 }
1009
1010 /* Don't handle automount points here */
1011 break;
1012 }
1013 return 0;
1014}
1015
9875cf80
DH		 1016/*
1017 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1018 */
1019static void follow_mount(struct path *path)
1020{
1021 while (d_mountpoint(path->dentry)) {
1022 struct vfsmount *mounted = lookup_mnt(path);
1023 if (!mounted)
1024 break;
1025 dput(path->dentry);
1026 mntput(path->mnt);
1027 path->mnt = mounted;
1028 path->dentry = dget(mounted->mnt_root);
1029 }
1030}
1031
31e6b01f 1032static void follow_dotdot(struct nameidata *nd)
1da177e4 1033{
2a737871 1034 set_root(nd);
e518ddb7 1035
1da177e4 1036 while(1) {
4ac91378 1037 struct dentry *old = nd->path.dentry;
1da177e4 1038
2a737871
AV		 1039 if (nd->path.dentry == nd->root.dentry &&
1040 nd->path.mnt == nd->root.mnt) {
1da177e4
LT		 1041 break;
1042 }
4ac91378 1043 if (nd->path.dentry != nd->path.mnt->mnt_root) {
3088dd70
AV		 1044 /* rare case of legitimate dget_parent()... */
1045 nd->path.dentry = dget_parent(nd->path.dentry);
1da177e4
LT		 1046 dput(old);
1047 break;
1048 }
3088dd70 1049 if (!follow_up(&nd->path))
1da177e4 1050 break;
1da177e4 1051 }
79ed0226 1052 follow_mount(&nd->path);
31e6b01f 1053 nd->inode = nd->path.dentry->d_inode;
1da177e4
LT		 1054}
1055
baa03890 1056/*
bad61189
MS		 1057 * This looks up the name in dcache, possibly revalidates the old dentry and
1058 * allocates a new one if not found or not valid. In the need_lookup argument
1059 * returns whether i_op->lookup is necessary.
1060 *
1061 * dir->d_inode->i_mutex must be held
baa03890 1062 */
bad61189
MS		 1063static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1064 struct nameidata *nd, bool *need_lookup)
baa03890 1065{
baa03890 1066 struct dentry *dentry;
bad61189 1067 int error;
baa03890 1068
bad61189
MS		 1069 *need_lookup = false;
1070 dentry = d_lookup(dir, name);
1071 if (dentry) {
1072 if (d_need_lookup(dentry)) {
1073 *need_lookup = true;
1074 } else if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1075 error = d_revalidate(dentry, nd);
1076 if (unlikely(error <= 0)) {
1077 if (error < 0) {
1078 dput(dentry);
1079 return ERR_PTR(error);
1080 } else if (!d_invalidate(dentry)) {
1081 dput(dentry);
1082 dentry = NULL;
1083 }
1084 }
1085 }
1086 }
baa03890 1087
bad61189
MS		 1088 if (!dentry) {
1089 dentry = d_alloc(dir, name);
1090 if (unlikely(!dentry))
1091 return ERR_PTR(-ENOMEM);
baa03890 1092
bad61189 1093 *need_lookup = true;
baa03890
NP		 1094 }
1095 return dentry;
1096}
1097
44396f4b 1098/*
bad61189
MS		 1099 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1100 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1101 *
1102 * dir->d_inode->i_mutex must be held
44396f4b 1103 */
bad61189
MS		 1104static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1105 struct nameidata *nd)
44396f4b 1106{
44396f4b
JB		 1107 struct dentry *old;
1108
1109 /* Don't create child dentry for a dead directory. */
bad61189 1110 if (unlikely(IS_DEADDIR(dir))) {
e188dc02 1111 dput(dentry);
44396f4b 1112 return ERR_PTR(-ENOENT);
e188dc02 1113 }
44396f4b 1114
bad61189 1115 old = dir->i_op->lookup(dir, dentry, nd);
44396f4b
JB		 1116 if (unlikely(old)) {
1117 dput(dentry);
1118 dentry = old;
1119 }
1120 return dentry;
1121}
1122
a3255546
AV		 1123static struct dentry *__lookup_hash(struct qstr *name,
1124 struct dentry *base, struct nameidata *nd)
1125{
bad61189 1126 bool need_lookup;
a3255546
AV		 1127 struct dentry *dentry;
1128
bad61189
MS		 1129 dentry = lookup_dcache(name, base, nd, &need_lookup);
1130 if (!need_lookup)
1131 return dentry;
a3255546 1132
bad61189 1133 return lookup_real(base->d_inode, dentry, nd);
a3255546
AV		 1134}
1135
1da177e4
LT		 1136/*
1137 * It's more convoluted than I'd like it to be, but... it's still fairly
1138 * small and for now I'd prefer to have fast path as straight as possible.
1139 * It _is_ time-critical.
1140 */
1141static int do_lookup(struct nameidata *nd, struct qstr *name,
31e6b01f 1142 struct path *path, struct inode **inode)
1da177e4 1143{
4ac91378 1144 struct vfsmount *mnt = nd->path.mnt;
31e6b01f 1145 struct dentry *dentry, *parent = nd->path.dentry;
5a18fff2
AV		 1146 int need_reval = 1;
1147 int status = 1;
9875cf80
DH		 1148 int err;
1149
b04f784e
NP		 1150 /*
1151 * Rename seqlock is not required here because in the off chance
1152 * of a false negative due to a concurrent rename, we're going to
1153 * do the non-racy lookup, below.
1154 */
31e6b01f
NP		 1155 if (nd->flags & LOOKUP_RCU) {
1156 unsigned seq;
31e6b01f
NP		 1157 *inode = nd->inode;
1158 dentry = __d_lookup_rcu(parent, name, &seq, inode);
5a18fff2
AV		 1159 if (!dentry)
1160 goto unlazy;
1161
31e6b01f
NP		 1162 /* Memory barrier in read_seqcount_begin of child is enough */
1163 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1164 return -ECHILD;
31e6b01f 1165 nd->seq = seq;
5a18fff2 1166
fa4ee159
MS		 1167 if (unlikely(d_need_lookup(dentry)))
1168 goto unlazy;
24643087 1169 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
5a18fff2
AV		 1170 status = d_revalidate(dentry, nd);
1171 if (unlikely(status <= 0)) {
1172 if (status != -ECHILD)
1173 need_reval = 0;
1174 goto unlazy;
1175 }
24643087 1176 }
31e6b01f
NP		 1177 path->mnt = mnt;
1178 path->dentry = dentry;
d6e9bd25
AV		 1179 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1180 goto unlazy;
1181 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1182 goto unlazy;
1183 return 0;
5a18fff2 1184unlazy:
19660af7
AV		 1185 if (unlazy_walk(nd, dentry))
1186 return -ECHILD;
5a18fff2
AV		 1187 } else {
1188 dentry = __d_lookup(parent, name);
9875cf80 1189 }
5a18fff2 1190
81e6f520
AV		 1191 if (unlikely(!dentry))
1192 goto need_lookup;
1193
1194 if (unlikely(d_need_lookup(dentry))) {
44396f4b 1195 dput(dentry);
81e6f520 1196 goto need_lookup;
5a18fff2 1197 }
81e6f520 1198
5a18fff2
AV		 1199 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1200 status = d_revalidate(dentry, nd);
1201 if (unlikely(status <= 0)) {
1202 if (status < 0) {
1203 dput(dentry);
1204 return status;
1205 }
1206 if (!d_invalidate(dentry)) {
1207 dput(dentry);
81e6f520 1208 goto need_lookup;
5a18fff2 1209 }
24643087 1210 }
3f6c7c71 1211done:
9875cf80
DH		 1212 path->mnt = mnt;
1213 path->dentry = dentry;
1214 err = follow_managed(path, nd->flags);
89312214
IK		 1215 if (unlikely(err < 0)) {
1216 path_put_conditional(path, nd);
9875cf80 1217 return err;
89312214 1218 }
a3fbbde7
AV		 1219 if (err)
1220 nd->flags |= LOOKUP_JUMPED;
9875cf80 1221 *inode = path->dentry->d_inode;
1da177e4 1222 return 0;
81e6f520
AV		 1223
1224need_lookup:
1225 BUG_ON(nd->inode != parent->d_inode);
1226
1227 mutex_lock(&parent->d_inode->i_mutex);
1228 dentry = __lookup_hash(name, parent, nd);
1229 mutex_unlock(&parent->d_inode->i_mutex);
1230 if (IS_ERR(dentry))
1231 return PTR_ERR(dentry);
1232 goto done;
1da177e4
LT		 1233}
1234
52094c8a
AV		 1235static inline int may_lookup(struct nameidata *nd)
1236{
1237 if (nd->flags & LOOKUP_RCU) {
4ad5abb3 1238 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
52094c8a
AV		 1239 if (err != -ECHILD)
1240 return err;
19660af7 1241 if (unlazy_walk(nd, NULL))
52094c8a
AV		 1242 return -ECHILD;
1243 }
4ad5abb3 1244 return inode_permission(nd->inode, MAY_EXEC);
52094c8a
AV		 1245}
1246
9856fa1b
AV		 1247static inline int handle_dots(struct nameidata *nd, int type)
1248{
1249 if (type == LAST_DOTDOT) {
1250 if (nd->flags & LOOKUP_RCU) {
1251 if (follow_dotdot_rcu(nd))
1252 return -ECHILD;
1253 } else
1254 follow_dotdot(nd);
1255 }
1256 return 0;
1257}
1258
951361f9
AV		 1259static void terminate_walk(struct nameidata *nd)
1260{
1261 if (!(nd->flags & LOOKUP_RCU)) {
1262 path_put(&nd->path);
1263 } else {
1264 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 1265 if (!(nd->flags & LOOKUP_ROOT))
1266 nd->root.mnt = NULL;
951361f9
AV		 1267 rcu_read_unlock();
1268 br_read_unlock(vfsmount_lock);
1269 }
1270}
1271
3ddcd056
LT		 1272/*
1273 * Do we need to follow links? We _really_ want to be able
1274 * to do this check without having to look at inode->i_op,
1275 * so we keep a cache of "no, this doesn't need follow_link"
1276 * for the common case.
1277 */
7813b94a 1278static inline int should_follow_link(struct inode *inode, int follow)
3ddcd056
LT		 1279{
1280 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1281 if (likely(inode->i_op->follow_link))
1282 return follow;
1283
1284 /* This gets set once for the inode lifetime */
1285 spin_lock(&inode->i_lock);
1286 inode->i_opflags |= IOP_NOFOLLOW;
1287 spin_unlock(&inode->i_lock);
1288 }
1289 return 0;
1290}
1291
ce57dfc1
AV		 1292static inline int walk_component(struct nameidata *nd, struct path *path,
1293 struct qstr *name, int type, int follow)
1294{
1295 struct inode *inode;
1296 int err;
1297 /*
1298 * "." and ".." are special - ".." especially so because it has
1299 * to be able to know about the current root directory and
1300 * parent relationships.
1301 */
1302 if (unlikely(type != LAST_NORM))
1303 return handle_dots(nd, type);
1304 err = do_lookup(nd, name, path, &inode);
1305 if (unlikely(err)) {
1306 terminate_walk(nd);
1307 return err;
1308 }
1309 if (!inode) {
1310 path_to_nameidata(path, nd);
1311 terminate_walk(nd);
1312 return -ENOENT;
1313 }
7813b94a 1314 if (should_follow_link(inode, follow)) {
19660af7
AV		 1315 if (nd->flags & LOOKUP_RCU) {
1316 if (unlikely(unlazy_walk(nd, path->dentry))) {
1317 terminate_walk(nd);
1318 return -ECHILD;
1319 }
1320 }
ce57dfc1
AV		 1321 BUG_ON(inode != path->dentry->d_inode);
1322 return 1;
1323 }
1324 path_to_nameidata(path, nd);
1325 nd->inode = inode;
1326 return 0;
1327}
1328
b356379a
AV		 1329/*
1330 * This limits recursive symlink follows to 8, while
1331 * limiting consecutive symlinks to 40.
1332 *
1333 * Without that kind of total limit, nasty chains of consecutive
1334 * symlinks can cause almost arbitrarily long lookups.
1335 */
1336static inline int nested_symlink(struct path *path, struct nameidata *nd)
1337{
1338 int res;
1339
b356379a
AV		 1340 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1341 path_put_conditional(path, nd);
1342 path_put(&nd->path);
1343 return -ELOOP;
1344 }
1a4022f8 1345 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
b356379a
AV		 1346
1347 nd->depth++;
1348 current->link_count++;
1349
1350 do {
1351 struct path link = *path;
1352 void *cookie;
574197e0
AV		 1353
1354 res = follow_link(&link, nd, &cookie);
b356379a
AV		 1355 if (!res)
1356 res = walk_component(nd, path, &nd->last,
1357 nd->last_type, LOOKUP_FOLLOW);
574197e0 1358 put_link(nd, &link, cookie);
b356379a
AV		 1359 } while (res > 0);
1360
1361 current->link_count--;
1362 nd->depth--;
1363 return res;
1364}
1365
3ddcd056
LT		 1366/*
1367 * We really don't want to look at inode->i_op->lookup
1368 * when we don't have to. So we keep a cache bit in
1369 * the inode ->i_opflags field that says "yes, we can
1370 * do lookup on this inode".
1371 */
1372static inline int can_lookup(struct inode *inode)
1373{
1374 if (likely(inode->i_opflags & IOP_LOOKUP))
1375 return 1;
1376 if (likely(!inode->i_op->lookup))
1377 return 0;
1378
1379 /* We do this once for the lifetime of the inode */
1380 spin_lock(&inode->i_lock);
1381 inode->i_opflags |= IOP_LOOKUP;
1382 spin_unlock(&inode->i_lock);
1383 return 1;
1384}
1385
bfcfaa77
LT		 1386/*
1387 * We can do the critical dentry name comparison and hashing
1388 * operations one word at a time, but we are limited to:
1389 *
1390 * - Architectures with fast unaligned word accesses. We could
1391 * do a "get_unaligned()" if this helps and is sufficiently
1392 * fast.
1393 *
1394 * - Little-endian machines (so that we can generate the mask
1395 * of low bytes efficiently). Again, we *could* do a byte
1396 * swapping load on big-endian architectures if that is not
1397 * expensive enough to make the optimization worthless.
1398 *
1399 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1400 * do not trap on the (extremely unlikely) case of a page
1401 * crossing operation.
1402 *
1403 * - Furthermore, we need an efficient 64-bit compile for the
1404 * 64-bit case in order to generate the "number of bytes in
1405 * the final mask". Again, that could be replaced with a
1406 * efficient population count instruction or similar.
1407 */
1408#ifdef CONFIG_DCACHE_WORD_ACCESS
1409
1410#ifdef CONFIG_64BIT
1411
1412/*
1413 * Jan Achrenius on G+: microoptimized version of
1414 * the simpler "(mask & ONEBYTES) * ONEBYTES >> 56"
1415 * that works for the bytemasks without having to
1416 * mask them first.
1417 */
1418static inline long count_masked_bytes(unsigned long mask)
1419{
f7493e5d 1420 return mask*0x0001020304050608ul >> 56;
bfcfaa77
LT		 1421}
1422
1423static inline unsigned int fold_hash(unsigned long hash)
1424{
1425 hash += hash >> (8*sizeof(int));
1426 return hash;
1427}
1428
1429#else /* 32-bit case */
1430
1431/* Carl Chatfield / Jan Achrenius G+ version for 32-bit */
1432static inline long count_masked_bytes(long mask)
1433{
1434 /* (000000 0000ff 00ffff ffffff) -> ( 1 1 2 3 ) */
1435 long a = (0x0ff0001+mask) >> 23;
1436 /* Fix the 1 for 00 case */
1437 return a & mask;
1438}
1439
1440#define fold_hash(x) (x)
1441
1442#endif
1443
1444unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1445{
1446 unsigned long a, mask;
1447 unsigned long hash = 0;
1448
1449 for (;;) {
1450 a = *(unsigned long *)name;
bfcfaa77
LT		 1451 if (len < sizeof(unsigned long))
1452 break;
1453 hash += a;
f132c5be 1454 hash *= 9;
bfcfaa77
LT		 1455 name += sizeof(unsigned long);
1456 len -= sizeof(unsigned long);
1457 if (!len)
1458 goto done;
1459 }
1460 mask = ~(~0ul << len*8);
1461 hash += mask & a;
1462done:
1463 return fold_hash(hash);
1464}
1465EXPORT_SYMBOL(full_name_hash);
1466
989412bb
LT		 1467#define REPEAT_BYTE(x) ((~0ul / 0xff) * (x))
1468#define ONEBYTES REPEAT_BYTE(0x01)
1469#define SLASHBYTES REPEAT_BYTE('/')
1470#define HIGHBITS REPEAT_BYTE(0x80)
bfcfaa77
LT		 1471
1472/* Return the high bit set in the first byte that is a zero */
1473static inline unsigned long has_zero(unsigned long a)
1474{
1475 return ((a - ONEBYTES) & ~a) & HIGHBITS;
1476}
1477
1478/*
1479 * Calculate the length and hash of the path component, and
1480 * return the length of the component;
1481 */
1482static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1483{
1484 unsigned long a, mask, hash, len;
1485
1486 hash = a = 0;
1487 len = -sizeof(unsigned long);
1488 do {
1489 hash = (hash + a) * 9;
1490 len += sizeof(unsigned long);
1491 a = *(unsigned long *)(name+len);
1492 /* Do we have any NUL or '/' bytes in this word? */
1493 mask = has_zero(a) | has_zero(a ^ SLASHBYTES);
1494 } while (!mask);
1495
1496 /* The mask *below* the first high bit set */
1497 mask = (mask - 1) & ~mask;
1498 mask >>= 7;
1499 hash += a & mask;
1500 *hashp = fold_hash(hash);
1501
1502 return len + count_masked_bytes(mask);
1503}
1504
1505#else
1506
0145acc2
LT		 1507unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1508{
1509 unsigned long hash = init_name_hash();
1510 while (len--)
1511 hash = partial_name_hash(*name++, hash);
1512 return end_name_hash(hash);
1513}
ae942ae7 1514EXPORT_SYMBOL(full_name_hash);
0145acc2 1515
200e9ef7
LT		 1516/*
1517 * We know there's a real path component here of at least
1518 * one character.
1519 */
1520static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1521{
1522 unsigned long hash = init_name_hash();
1523 unsigned long len = 0, c;
1524
1525 c = (unsigned char)*name;
1526 do {
1527 len++;
1528 hash = partial_name_hash(c, hash);
1529 c = (unsigned char)name[len];
1530 } while (c && c != '/');
1531 *hashp = end_name_hash(hash);
1532 return len;
1533}
1534
bfcfaa77
LT		 1535#endif
1536
1da177e4
LT		 1537/*
1538 * Name resolution.
ea3834d9
PM		 1539 * This is the basic name resolution function, turning a pathname into
1540 * the final dentry. We expect 'base' to be positive and a directory.
1da177e4 1541 *
ea3834d9
PM		 1542 * Returns 0 and nd will have valid dentry and mnt on success.
1543 * Returns error and drops reference to input namei data on failure.
1da177e4 1544 */
6de88d72 1545static int link_path_walk(const char *name, struct nameidata *nd)
1da177e4
LT		 1546{
1547 struct path next;
1da177e4 1548 int err;
1da177e4
LT		 1549
1550 while (*name=='/')
1551 name++;
1552 if (!*name)
086e183a 1553 return 0;
1da177e4 1554
1da177e4
LT		 1555 /* At this point we know we have a real path component. */
1556 for(;;) {
1da177e4 1557 struct qstr this;
200e9ef7 1558 long len;
fe479a58 1559 int type;
1da177e4 1560
52094c8a 1561 err = may_lookup(nd);
1da177e4
LT		 1562 if (err)
1563 break;
1564
200e9ef7 1565 len = hash_name(name, &this.hash);
1da177e4 1566 this.name = name;
200e9ef7 1567 this.len = len;
1da177e4 1568
fe479a58 1569 type = LAST_NORM;
200e9ef7 1570 if (name[0] == '.') switch (len) {
fe479a58 1571 case 2:
200e9ef7 1572 if (name[1] == '.') {
fe479a58 1573 type = LAST_DOTDOT;
16c2cd71
AV		 1574 nd->flags |= LOOKUP_JUMPED;
1575 }
fe479a58
AV		 1576 break;
1577 case 1:
1578 type = LAST_DOT;
1579 }
5a202bcd
AV		 1580 if (likely(type == LAST_NORM)) {
1581 struct dentry *parent = nd->path.dentry;
16c2cd71 1582 nd->flags &= ~LOOKUP_JUMPED;
5a202bcd
AV		 1583 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1584 err = parent->d_op->d_hash(parent, nd->inode,
1585 &this);
1586 if (err < 0)
1587 break;
1588 }
1589 }
fe479a58 1590
200e9ef7 1591 if (!name[len])
1da177e4 1592 goto last_component;
200e9ef7
LT		 1593 /*
1594 * If it wasn't NUL, we know it was '/'. Skip that
1595 * slash, and continue until no more slashes.
1596 */
1597 do {
1598 len++;
1599 } while (unlikely(name[len] == '/'));
1600 if (!name[len])
b356379a 1601 goto last_component;
200e9ef7 1602 name += len;
1da177e4 1603
ce57dfc1
AV		 1604 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1605 if (err < 0)
1606 return err;
1da177e4 1607
ce57dfc1 1608 if (err) {
b356379a 1609 err = nested_symlink(&next, nd);
1da177e4 1610 if (err)
a7472bab 1611 return err;
31e6b01f 1612 }
3ddcd056
LT		 1613 if (can_lookup(nd->inode))
1614 continue;
1da177e4 1615 err = -ENOTDIR;
3ddcd056 1616 break;
1da177e4
LT		 1617 /* here ends the main loop */
1618
1da177e4 1619last_component:
b356379a
AV		 1620 nd->last = this;
1621 nd->last_type = type;
086e183a 1622 return 0;
1da177e4 1623 }
951361f9 1624 terminate_walk(nd);
1da177e4
LT		 1625 return err;
1626}
1627
70e9b357
AV		 1628static int path_init(int dfd, const char *name, unsigned int flags,
1629 struct nameidata *nd, struct file **fp)
31e6b01f
NP		 1630{
1631 int retval = 0;
1632 int fput_needed;
1633 struct file *file;
1634
1635 nd->last_type = LAST_ROOT; /* if there are only slashes... */
16c2cd71 1636 nd->flags = flags | LOOKUP_JUMPED;
31e6b01f 1637 nd->depth = 0;
5b6ca027
AV		 1638 if (flags & LOOKUP_ROOT) {
1639 struct inode *inode = nd->root.dentry->d_inode;
73d049a4
AV		 1640 if (*name) {
1641 if (!inode->i_op->lookup)
1642 return -ENOTDIR;
1643 retval = inode_permission(inode, MAY_EXEC);
1644 if (retval)
1645 return retval;
1646 }
5b6ca027
AV		 1647 nd->path = nd->root;
1648 nd->inode = inode;
1649 if (flags & LOOKUP_RCU) {
1650 br_read_lock(vfsmount_lock);
1651 rcu_read_lock();
1652 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1653 } else {
1654 path_get(&nd->path);
1655 }
1656 return 0;
1657 }
1658
31e6b01f 1659 nd->root.mnt = NULL;
31e6b01f
NP		 1660
1661 if (*name=='/') {
e41f7d4e
AV		 1662 if (flags & LOOKUP_RCU) {
1663 br_read_lock(vfsmount_lock);
1664 rcu_read_lock();
1665 set_root_rcu(nd);
1666 } else {
1667 set_root(nd);
1668 path_get(&nd->root);
1669 }
1670 nd->path = nd->root;
31e6b01f 1671 } else if (dfd == AT_FDCWD) {
e41f7d4e
AV		 1672 if (flags & LOOKUP_RCU) {
1673 struct fs_struct *fs = current->fs;
1674 unsigned seq;
31e6b01f 1675
e41f7d4e
AV		 1676 br_read_lock(vfsmount_lock);
1677 rcu_read_lock();
c28cc364 1678
e41f7d4e
AV		 1679 do {
1680 seq = read_seqcount_begin(&fs->seq);
1681 nd->path = fs->pwd;
1682 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1683 } while (read_seqcount_retry(&fs->seq, seq));
1684 } else {
1685 get_fs_pwd(current->fs, &nd->path);
1686 }
31e6b01f
NP		 1687 } else {
1688 struct dentry *dentry;
1689
1abf0c71 1690 file = fget_raw_light(dfd, &fput_needed);
31e6b01f
NP		 1691 retval = -EBADF;
1692 if (!file)
1693 goto out_fail;
1694
1695 dentry = file->f_path.dentry;
1696
f52e0c11
AV		 1697 if (*name) {
1698 retval = -ENOTDIR;
1699 if (!S_ISDIR(dentry->d_inode->i_mode))
1700 goto fput_fail;
31e6b01f 1701
4ad5abb3 1702 retval = inode_permission(dentry->d_inode, MAY_EXEC);
f52e0c11
AV		 1703 if (retval)
1704 goto fput_fail;
1705 }
31e6b01f
NP		 1706
1707 nd->path = file->f_path;
e41f7d4e
AV		 1708 if (flags & LOOKUP_RCU) {
1709 if (fput_needed)
70e9b357 1710 *fp = file;
e41f7d4e
AV		 1711 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1712 br_read_lock(vfsmount_lock);
1713 rcu_read_lock();
1714 } else {
1715 path_get(&file->f_path);
1716 fput_light(file, fput_needed);
1717 }
31e6b01f 1718 }
31e6b01f 1719
31e6b01f 1720 nd->inode = nd->path.dentry->d_inode;
9b4a9b14 1721 return 0;
2dfdd266 1722
9b4a9b14
AV		 1723fput_fail:
1724 fput_light(file, fput_needed);
1725out_fail:
1726 return retval;
1727}
1728
bd92d7fe
AV		 1729static inline int lookup_last(struct nameidata *nd, struct path *path)
1730{
1731 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1732 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1733
1734 nd->flags &= ~LOOKUP_PARENT;
1735 return walk_component(nd, path, &nd->last, nd->last_type,
1736 nd->flags & LOOKUP_FOLLOW);
1737}
1738
9b4a9b14 1739/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
ee0827cd 1740static int path_lookupat(int dfd, const char *name,
9b4a9b14
AV		 1741 unsigned int flags, struct nameidata *nd)
1742{
70e9b357 1743 struct file *base = NULL;
bd92d7fe
AV		 1744 struct path path;
1745 int err;
31e6b01f
NP		 1746
1747 /*
1748 * Path walking is largely split up into 2 different synchronisation
1749 * schemes, rcu-walk and ref-walk (explained in
1750 * Documentation/filesystems/path-lookup.txt). These share much of the
1751 * path walk code, but some things particularly setup, cleanup, and
1752 * following mounts are sufficiently divergent that functions are
1753 * duplicated. Typically there is a function foo(), and its RCU
1754 * analogue, foo_rcu().
1755 *
1756 * -ECHILD is the error number of choice (just to avoid clashes) that
1757 * is returned if some aspect of an rcu-walk fails. Such an error must
1758 * be handled by restarting a traditional ref-walk (which will always
1759 * be able to complete).
1760 */
bd92d7fe 1761 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
ee0827cd 1762
bd92d7fe
AV		 1763 if (unlikely(err))
1764 return err;
ee0827cd
AV		 1765
1766 current->total_link_count = 0;
bd92d7fe
AV		 1767 err = link_path_walk(name, nd);
1768
1769 if (!err && !(flags & LOOKUP_PARENT)) {
bd92d7fe
AV		 1770 err = lookup_last(nd, &path);
1771 while (err > 0) {
1772 void *cookie;
1773 struct path link = path;
bd92d7fe 1774 nd->flags |= LOOKUP_PARENT;
574197e0 1775 err = follow_link(&link, nd, &cookie);
bd92d7fe
AV		 1776 if (!err)
1777 err = lookup_last(nd, &path);
574197e0 1778 put_link(nd, &link, cookie);
bd92d7fe
AV		 1779 }
1780 }
ee0827cd 1781
9f1fafee
AV		 1782 if (!err)
1783 err = complete_walk(nd);
bd92d7fe
AV		 1784
1785 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1786 if (!nd->inode->i_op->lookup) {
1787 path_put(&nd->path);
bd23a539 1788 err = -ENOTDIR;
bd92d7fe
AV		 1789 }
1790 }
16c2cd71 1791
70e9b357
AV		 1792 if (base)
1793 fput(base);
ee0827cd 1794
5b6ca027 1795 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
2a737871
AV		 1796 path_put(&nd->root);
1797 nd->root.mnt = NULL;
1798 }
bd92d7fe 1799 return err;
ee0827cd 1800}
31e6b01f 1801
ee0827cd
AV		 1802static int do_path_lookup(int dfd, const char *name,
1803 unsigned int flags, struct nameidata *nd)
1804{
1805 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1806 if (unlikely(retval == -ECHILD))
1807 retval = path_lookupat(dfd, name, flags, nd);
1808 if (unlikely(retval == -ESTALE))
1809 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
31e6b01f
NP		 1810
1811 if (likely(!retval)) {
1812 if (unlikely(!audit_dummy_context())) {
1813 if (nd->path.dentry && nd->inode)
1814 audit_inode(name, nd->path.dentry);
1815 }
1816 }
170aa3d0 1817 return retval;
1da177e4
LT		 1818}
1819
c9c6cac0 1820int kern_path_parent(const char *name, struct nameidata *nd)
5590ff0d 1821{
c9c6cac0 1822 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
5590ff0d
UD		 1823}
1824
d1811465
AV		 1825int kern_path(const char *name, unsigned int flags, struct path *path)
1826{
1827 struct nameidata nd;
1828 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1829 if (!res)
1830 *path = nd.path;
1831 return res;
1832}
1833
16f18200
JJS		 1834/**
1835 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1836 * @dentry: pointer to dentry of the base directory
1837 * @mnt: pointer to vfs mount of the base directory
1838 * @name: pointer to file name
1839 * @flags: lookup flags
e0a01249 1840 * @path: pointer to struct path to fill
16f18200
JJS		 1841 */
1842int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1843 const char *name, unsigned int flags,
e0a01249 1844 struct path *path)
16f18200 1845{
e0a01249
AV		 1846 struct nameidata nd;
1847 int err;
1848 nd.root.dentry = dentry;
1849 nd.root.mnt = mnt;
1850 BUG_ON(flags & LOOKUP_PARENT);
5b6ca027 1851 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
e0a01249
AV		 1852 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1853 if (!err)
1854 *path = nd.path;
1855 return err;
16f18200
JJS		 1856}
1857
057f6c01
JM		 1858/*
1859 * Restricted form of lookup. Doesn't follow links, single-component only,
1860 * needs parent already locked. Doesn't follow mounts.
1861 * SMP-safe.
1862 */
eead1911 1863static struct dentry *lookup_hash(struct nameidata *nd)
057f6c01 1864{
4ac91378 1865 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1da177e4
LT		 1866}
1867
eead1911 1868/**
a6b91919 1869 * lookup_one_len - filesystem helper to lookup single pathname component
eead1911
CH		 1870 * @name: pathname component to lookup
1871 * @base: base directory to lookup from
1872 * @len: maximum length @len should be interpreted to
1873 *
a6b91919
RD		 1874 * Note that this routine is purely a helper for filesystem usage and should
1875 * not be called by generic code. Also note that by using this function the
eead1911
CH		 1876 * nameidata argument is passed to the filesystem methods and a filesystem
1877 * using this helper needs to be prepared for that.
1878 */
057f6c01
JM		 1879struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1880{
057f6c01 1881 struct qstr this;
6a96ba54 1882 unsigned int c;
cda309de 1883 int err;
057f6c01 1884
2f9092e1
DW		 1885 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1886
6a96ba54
AV		 1887 this.name = name;
1888 this.len = len;
0145acc2 1889 this.hash = full_name_hash(name, len);
6a96ba54
AV		 1890 if (!len)
1891 return ERR_PTR(-EACCES);
1892
6a96ba54
AV		 1893 while (len--) {
1894 c = *(const unsigned char *)name++;
1895 if (c == '/' || c == '\0')
1896 return ERR_PTR(-EACCES);
6a96ba54 1897 }
5a202bcd
AV		 1898 /*
1899 * See if the low-level filesystem might want
1900 * to use its own hash..
1901 */
1902 if (base->d_flags & DCACHE_OP_HASH) {
1903 int err = base->d_op->d_hash(base, base->d_inode, &this);
1904 if (err < 0)
1905 return ERR_PTR(err);
1906 }
eead1911 1907
cda309de
MS		 1908 err = inode_permission(base->d_inode, MAY_EXEC);
1909 if (err)
1910 return ERR_PTR(err);
1911
49705b77 1912 return __lookup_hash(&this, base, NULL);
057f6c01
JM		 1913}
1914
1fa1e7f6
AW		 1915int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
1916 struct path *path, int *empty)
1da177e4 1917{
2d8f3038 1918 struct nameidata nd;
1fa1e7f6 1919 char *tmp = getname_flags(name, flags, empty);
1da177e4 1920 int err = PTR_ERR(tmp);
1da177e4 1921 if (!IS_ERR(tmp)) {
2d8f3038
AV		 1922
1923 BUG_ON(flags & LOOKUP_PARENT);
1924
1925 err = do_path_lookup(dfd, tmp, flags, &nd);
1da177e4 1926 putname(tmp);
2d8f3038
AV		 1927 if (!err)
1928 *path = nd.path;
1da177e4
LT		 1929 }
1930 return err;
1931}
1932
1fa1e7f6
AW		 1933int user_path_at(int dfd, const char __user *name, unsigned flags,
1934 struct path *path)
1935{
f7493e5d 1936 return user_path_at_empty(dfd, name, flags, path, NULL);
1fa1e7f6
AW		 1937}
1938
2ad94ae6
AV		 1939static int user_path_parent(int dfd, const char __user *path,
1940 struct nameidata *nd, char **name)
1941{
1942 char *s = getname(path);
1943 int error;
1944
1945 if (IS_ERR(s))
1946 return PTR_ERR(s);
1947
1948 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1949 if (error)
1950 putname(s);
1951 else
1952 *name = s;
1953
1954 return error;
1955}
1956
1da177e4
LT		 1957/*
1958 * It's inline, so penalty for filesystems that don't use sticky bit is
1959 * minimal.
1960 */
1961static inline int check_sticky(struct inode *dir, struct inode *inode)
1962{
da9592ed
DH		 1963 uid_t fsuid = current_fsuid();
1964
1da177e4
LT		 1965 if (!(dir->i_mode & S_ISVTX))
1966 return 0;
e795b717
SH		 1967 if (current_user_ns() != inode_userns(inode))
1968 goto other_userns;
da9592ed 1969 if (inode->i_uid == fsuid)
1da177e4 1970 return 0;
da9592ed 1971 if (dir->i_uid == fsuid)
1da177e4 1972 return 0;
e795b717
SH		 1973
1974other_userns:
1975 return !ns_capable(inode_userns(inode), CAP_FOWNER);
1da177e4
LT		 1976}
1977
1978/*
1979 * Check whether we can remove a link victim from directory dir, check
1980 * whether the type of victim is right.
1981 * 1. We can't do it if dir is read-only (done in permission())
1982 * 2. We should have write and exec permissions on dir
1983 * 3. We can't remove anything from append-only dir
1984 * 4. We can't do anything with immutable dir (done in permission())
1985 * 5. If the sticky bit on dir is set we should either
1986 * a. be owner of dir, or
1987 * b. be owner of victim, or
1988 * c. have CAP_FOWNER capability
1989 * 6. If the victim is append-only or immutable we can't do antyhing with
1990 * links pointing to it.
1991 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1992 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1993 * 9. We can't remove a root or mountpoint.
1994 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1995 * nfs_async_unlink().
1996 */
858119e1 1997static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1da177e4
LT		 1998{
1999 int error;
2000
2001 if (!victim->d_inode)
2002 return -ENOENT;
2003
2004 BUG_ON(victim->d_parent->d_inode != dir);
cccc6bba 2005 audit_inode_child(victim, dir);
1da177e4 2006
f419a2e3 2007 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 2008 if (error)
2009 return error;
2010 if (IS_APPEND(dir))
2011 return -EPERM;
2012 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
f9454548 2013 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1da177e4
LT		 2014 return -EPERM;
2015 if (isdir) {
2016 if (!S_ISDIR(victim->d_inode->i_mode))
2017 return -ENOTDIR;
2018 if (IS_ROOT(victim))
2019 return -EBUSY;
2020 } else if (S_ISDIR(victim->d_inode->i_mode))
2021 return -EISDIR;
2022 if (IS_DEADDIR(dir))
2023 return -ENOENT;
2024 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2025 return -EBUSY;
2026 return 0;
2027}
2028
2029/* Check whether we can create an object with dentry child in directory
2030 * dir.
2031 * 1. We can't do it if child already exists (open has special treatment for
2032 * this case, but since we are inlined it's OK)
2033 * 2. We can't do it if dir is read-only (done in permission())
2034 * 3. We should have write and exec permissions on dir
2035 * 4. We can't do it if dir is immutable (done in permission())
2036 */
a95164d9 2037static inline int may_create(struct inode *dir, struct dentry *child)
1da177e4
LT		 2038{
2039 if (child->d_inode)
2040 return -EEXIST;
2041 if (IS_DEADDIR(dir))
2042 return -ENOENT;
f419a2e3 2043 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 2044}
2045
1da177e4
LT		 2046/*
2047 * p1 and p2 should be directories on the same fs.
2048 */
2049struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2050{
2051 struct dentry *p;
2052
2053 if (p1 == p2) {
f2eace23 2054 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1da177e4
LT		 2055 return NULL;
2056 }
2057
a11f3a05 2058 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4 2059
e2761a11
OH		 2060 p = d_ancestor(p2, p1);
2061 if (p) {
2062 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2063 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2064 return p;
1da177e4
LT		 2065 }
2066
e2761a11
OH		 2067 p = d_ancestor(p1, p2);
2068 if (p) {
2069 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2070 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2071 return p;
1da177e4
LT		 2072 }
2073
f2eace23
IM		 2074 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2075 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1da177e4
LT		 2076 return NULL;
2077}
2078
2079void unlock_rename(struct dentry *p1, struct dentry *p2)
2080{
1b1dcc1b 2081 mutex_unlock(&p1->d_inode->i_mutex);
1da177e4 2082 if (p1 != p2) {
1b1dcc1b 2083 mutex_unlock(&p2->d_inode->i_mutex);
a11f3a05 2084 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4
LT		 2085 }
2086}
2087
4acdaf27 2088int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
1da177e4
LT		 2089 struct nameidata *nd)
2090{
a95164d9 2091 int error = may_create(dir, dentry);
1da177e4
LT		 2092
2093 if (error)
2094 return error;
2095
acfa4380 2096 if (!dir->i_op->create)
1da177e4
LT		 2097 return -EACCES; /* shouldn't it be ENOSYS? */
2098 mode &= S_IALLUGO;
2099 mode |= S_IFREG;
2100 error = security_inode_create(dir, dentry, mode);
2101 if (error)
2102 return error;
1da177e4 2103 error = dir->i_op->create(dir, dentry, mode, nd);
a74574aa 2104 if (!error)
f38aa942 2105 fsnotify_create(dir, dentry);
1da177e4
LT		 2106 return error;
2107}
2108
73d049a4 2109static int may_open(struct path *path, int acc_mode, int flag)
1da177e4 2110{
3fb64190 2111 struct dentry *dentry = path->dentry;
1da177e4
LT		 2112 struct inode *inode = dentry->d_inode;
2113 int error;
2114
bcda7652
AV		 2115 /* O_PATH? */
2116 if (!acc_mode)
2117 return 0;
2118
1da177e4
LT		 2119 if (!inode)
2120 return -ENOENT;
2121
c8fe8f30
CH		 2122 switch (inode->i_mode & S_IFMT) {
2123 case S_IFLNK:
1da177e4 2124 return -ELOOP;
c8fe8f30
CH		 2125 case S_IFDIR:
2126 if (acc_mode & MAY_WRITE)
2127 return -EISDIR;
2128 break;
2129 case S_IFBLK:
2130 case S_IFCHR:
3fb64190 2131 if (path->mnt->mnt_flags & MNT_NODEV)
1da177e4 2132 return -EACCES;
c8fe8f30
CH		 2133 /*FALLTHRU*/
2134 case S_IFIFO:
2135 case S_IFSOCK:
1da177e4 2136 flag &= ~O_TRUNC;
c8fe8f30 2137 break;
4a3fd211 2138 }
b41572e9 2139
3fb64190 2140 error = inode_permission(inode, acc_mode);
b41572e9
DH		 2141 if (error)
2142 return error;
6146f0d5 2143
1da177e4
LT		 2144 /*
2145 * An append-only file must be opened in append mode for writing.
2146 */
2147 if (IS_APPEND(inode)) {
8737c930 2148 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
7715b521 2149 return -EPERM;
1da177e4 2150 if (flag & O_TRUNC)
7715b521 2151 return -EPERM;
1da177e4
LT		 2152 }
2153
2154 /* O_NOATIME can only be set by the owner or superuser */
2e149670 2155 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
7715b521 2156 return -EPERM;
1da177e4 2157
f3c7691e 2158 return 0;
7715b521 2159}
1da177e4 2160
e1181ee6 2161static int handle_truncate(struct file *filp)
7715b521 2162{
e1181ee6 2163 struct path *path = &filp->f_path;
7715b521
AV		 2164 struct inode *inode = path->dentry->d_inode;
2165 int error = get_write_access(inode);
2166 if (error)
2167 return error;
2168 /*
2169 * Refuse to truncate files with mandatory locks held on them.
2170 */
2171 error = locks_verify_locked(inode);
2172 if (!error)
ea0d3ab2 2173 error = security_path_truncate(path);
7715b521
AV		 2174 if (!error) {
2175 error = do_truncate(path->dentry, 0,
2176 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
e1181ee6 2177 filp);
7715b521
AV		 2178 }
2179 put_write_access(inode);
acd0c935 2180 return error;
1da177e4
LT		 2181}
2182
d57999e1
DH		 2183static inline int open_to_namei_flags(int flag)
2184{
8a5e929d
AV		 2185 if ((flag & O_ACCMODE) == 3)
2186 flag--;
d57999e1
DH		 2187 return flag;
2188}
2189
31e6b01f 2190/*
fe2d35ff 2191 * Handle the last step of open()
31e6b01f 2192 */
fb1cc555 2193static struct file *do_last(struct nameidata *nd, struct path *path,
c3e380b0 2194 const struct open_flags *op, const char *pathname)
fb1cc555 2195{
a1e28038 2196 struct dentry *dir = nd->path.dentry;
6c0d46c4 2197 struct dentry *dentry;
ca344a89 2198 int open_flag = op->open_flag;
6c0d46c4 2199 int will_truncate = open_flag & O_TRUNC;
ca344a89 2200 int want_write = 0;
bcda7652 2201 int acc_mode = op->acc_mode;
fb1cc555 2202 struct file *filp;
16c2cd71 2203 int error;
1f36f774 2204
c3e380b0
AV		 2205 nd->flags &= ~LOOKUP_PARENT;
2206 nd->flags |= op->intent;
2207
1f36f774
AV		 2208 switch (nd->last_type) {
2209 case LAST_DOTDOT:
176306f5 2210 case LAST_DOT:
fe2d35ff
AV		 2211 error = handle_dots(nd, nd->last_type);
2212 if (error)
2213 return ERR_PTR(error);
1f36f774 2214 /* fallthrough */
1f36f774 2215 case LAST_ROOT:
9f1fafee 2216 error = complete_walk(nd);
16c2cd71 2217 if (error)
9f1fafee 2218 return ERR_PTR(error);
fe2d35ff 2219 audit_inode(pathname, nd->path.dentry);
ca344a89 2220 if (open_flag & O_CREAT) {
fe2d35ff
AV		 2221 error = -EISDIR;
2222 goto exit;
2223 }
2224 goto ok;
1f36f774 2225 case LAST_BIND:
9f1fafee 2226 error = complete_walk(nd);
16c2cd71 2227 if (error)
9f1fafee 2228 return ERR_PTR(error);
1f36f774 2229 audit_inode(pathname, dir);
67ee3ad2 2230 goto ok;
1f36f774 2231 }
67ee3ad2 2232
ca344a89 2233 if (!(open_flag & O_CREAT)) {
bcda7652 2234 int symlink_ok = 0;
fe2d35ff
AV		 2235 if (nd->last.name[nd->last.len])
2236 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
bcda7652
AV		 2237 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2238 symlink_ok = 1;
fe2d35ff 2239 /* we _can_ be in RCU mode here */
ce57dfc1
AV		 2240 error = walk_component(nd, path, &nd->last, LAST_NORM,
2241 !symlink_ok);
2242 if (error < 0)
fe2d35ff 2243 return ERR_PTR(error);
ce57dfc1 2244 if (error) /* symlink */
fe2d35ff 2245 return NULL;
fe2d35ff 2246 /* sayonara */
9f1fafee
AV		 2247 error = complete_walk(nd);
2248 if (error)
7f6c7e62 2249 return ERR_PTR(error);
fe2d35ff
AV		 2250
2251 error = -ENOTDIR;
2252 if (nd->flags & LOOKUP_DIRECTORY) {
ce57dfc1 2253 if (!nd->inode->i_op->lookup)
fe2d35ff
AV		 2254 goto exit;
2255 }
2256 audit_inode(pathname, nd->path.dentry);
2257 goto ok;
2258 }
2259
2260 /* create side of things */
a3fbbde7
AV		 2261 /*
2262 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED has been
2263 * cleared when we got to the last component we are about to look up
2264 */
9f1fafee
AV		 2265 error = complete_walk(nd);
2266 if (error)
2267 return ERR_PTR(error);
fe2d35ff
AV		 2268
2269 audit_inode(pathname, dir);
16c2cd71 2270 error = -EISDIR;
1f36f774 2271 /* trailing slashes? */
31e6b01f
NP		 2272 if (nd->last.name[nd->last.len])
2273 goto exit;
a2c36b45 2274
a1e28038
AV		 2275 mutex_lock(&dir->d_inode->i_mutex);
2276
6c0d46c4
AV		 2277 dentry = lookup_hash(nd);
2278 error = PTR_ERR(dentry);
2279 if (IS_ERR(dentry)) {
fb1cc555
AV		 2280 mutex_unlock(&dir->d_inode->i_mutex);
2281 goto exit;
2282 }
2283
6c0d46c4
AV		 2284 path->dentry = dentry;
2285 path->mnt = nd->path.mnt;
2286
fb1cc555 2287 /* Negative dentry, just create the file */
6c0d46c4 2288 if (!dentry->d_inode) {
a218d0fd 2289 umode_t mode = op->mode;
6c0d46c4
AV		 2290 if (!IS_POSIXACL(dir->d_inode))
2291 mode &= ~current_umask();
fb1cc555
AV		 2292 /*
2293 * This write is needed to ensure that a
6c0d46c4 2294 * rw->ro transition does not occur between
fb1cc555
AV		 2295 * the time when the file is created and when
2296 * a permanent write count is taken through
2297 * the 'struct file' in nameidata_to_filp().
2298 */
2299 error = mnt_want_write(nd->path.mnt);
2300 if (error)
2301 goto exit_mutex_unlock;
ca344a89 2302 want_write = 1;
9b44f1b3 2303 /* Don't check for write permission, don't truncate */
ca344a89 2304 open_flag &= ~O_TRUNC;
6c0d46c4 2305 will_truncate = 0;
bcda7652 2306 acc_mode = MAY_OPEN;
6c0d46c4
AV		 2307 error = security_path_mknod(&nd->path, dentry, mode, 0);
2308 if (error)
2309 goto exit_mutex_unlock;
2310 error = vfs_create(dir->d_inode, dentry, mode, nd);
2311 if (error)
2312 goto exit_mutex_unlock;
2313 mutex_unlock(&dir->d_inode->i_mutex);
2314 dput(nd->path.dentry);
2315 nd->path.dentry = dentry;
ca344a89 2316 goto common;
fb1cc555
AV		 2317 }
2318
2319 /*
2320 * It already exists.
2321 */
2322 mutex_unlock(&dir->d_inode->i_mutex);
2323 audit_inode(pathname, path->dentry);
2324
2325 error = -EEXIST;
ca344a89 2326 if (open_flag & O_EXCL)
fb1cc555
AV		 2327 goto exit_dput;
2328
9875cf80
DH		 2329 error = follow_managed(path, nd->flags);
2330 if (error < 0)
2331 goto exit_dput;
fb1cc555 2332
a3fbbde7
AV		 2333 if (error)
2334 nd->flags |= LOOKUP_JUMPED;
2335
fb1cc555
AV		 2336 error = -ENOENT;
2337 if (!path->dentry->d_inode)
2338 goto exit_dput;
9e67f361
AV		 2339
2340 if (path->dentry->d_inode->i_op->follow_link)
fb1cc555 2341 return NULL;
fb1cc555
AV		 2342
2343 path_to_nameidata(path, nd);
31e6b01f 2344 nd->inode = path->dentry->d_inode;
a3fbbde7
AV		 2345 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2346 error = complete_walk(nd);
2347 if (error)
097b180c 2348 return ERR_PTR(error);
fb1cc555 2349 error = -EISDIR;
31e6b01f 2350 if (S_ISDIR(nd->inode->i_mode))
fb1cc555 2351 goto exit;
67ee3ad2 2352ok:
6c0d46c4
AV		 2353 if (!S_ISREG(nd->inode->i_mode))
2354 will_truncate = 0;
2355
0f9d1a10
AV		 2356 if (will_truncate) {
2357 error = mnt_want_write(nd->path.mnt);
2358 if (error)
2359 goto exit;
ca344a89 2360 want_write = 1;
0f9d1a10 2361 }
ca344a89 2362common:
bcda7652 2363 error = may_open(&nd->path, acc_mode, open_flag);
ca344a89 2364 if (error)
0f9d1a10 2365 goto exit;
0f9d1a10
AV		 2366 filp = nameidata_to_filp(nd);
2367 if (!IS_ERR(filp)) {
2368 error = ima_file_check(filp, op->acc_mode);
2369 if (error) {
2370 fput(filp);
2371 filp = ERR_PTR(error);
2372 }
2373 }
2374 if (!IS_ERR(filp)) {
2375 if (will_truncate) {
2376 error = handle_truncate(filp);
2377 if (error) {
2378 fput(filp);
2379 filp = ERR_PTR(error);
2380 }
2381 }
2382 }
ca344a89
AV		 2383out:
2384 if (want_write)
0f9d1a10
AV		 2385 mnt_drop_write(nd->path.mnt);
2386 path_put(&nd->path);
fb1cc555
AV		 2387 return filp;
2388
2389exit_mutex_unlock:
2390 mutex_unlock(&dir->d_inode->i_mutex);
2391exit_dput:
2392 path_put_conditional(path, nd);
2393exit:
ca344a89
AV		 2394 filp = ERR_PTR(error);
2395 goto out;
fb1cc555
AV		 2396}
2397
13aab428 2398static struct file *path_openat(int dfd, const char *pathname,
73d049a4 2399 struct nameidata *nd, const struct open_flags *op, int flags)
1da177e4 2400{
fe2d35ff 2401 struct file *base = NULL;
4a3fd211 2402 struct file *filp;
9850c056 2403 struct path path;
13aab428 2404 int error;
31e6b01f
NP		 2405
2406 filp = get_empty_filp();
2407 if (!filp)
2408 return ERR_PTR(-ENFILE);
2409
47c805dc 2410 filp->f_flags = op->open_flag;
73d049a4
AV		 2411 nd->intent.open.file = filp;
2412 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2413 nd->intent.open.create_mode = op->mode;
31e6b01f 2414
73d049a4 2415 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
31e6b01f 2416 if (unlikely(error))
13aab428 2417 goto out_filp;
31e6b01f 2418
fe2d35ff 2419 current->total_link_count = 0;
73d049a4 2420 error = link_path_walk(pathname, nd);
31e6b01f
NP		 2421 if (unlikely(error))
2422 goto out_filp;
1da177e4 2423
73d049a4 2424 filp = do_last(nd, &path, op, pathname);
806b681c 2425 while (unlikely(!filp)) { /* trailing symlink */
7b9337aa 2426 struct path link = path;
def4af30 2427 void *cookie;
574197e0 2428 if (!(nd->flags & LOOKUP_FOLLOW)) {
73d049a4
AV		 2429 path_put_conditional(&path, nd);
2430 path_put(&nd->path);
40b39136
AV		 2431 filp = ERR_PTR(-ELOOP);
2432 break;
2433 }
73d049a4
AV		 2434 nd->flags |= LOOKUP_PARENT;
2435 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
574197e0 2436 error = follow_link(&link, nd, &cookie);
c3e380b0 2437 if (unlikely(error))
f1afe9ef 2438 filp = ERR_PTR(error);
c3e380b0 2439 else
73d049a4 2440 filp = do_last(nd, &path, op, pathname);
574197e0 2441 put_link(nd, &link, cookie);
806b681c 2442 }
10fa8e62 2443out:
73d049a4
AV		 2444 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2445 path_put(&nd->root);
fe2d35ff
AV		 2446 if (base)
2447 fput(base);
73d049a4 2448 release_open_intent(nd);
10fa8e62 2449 return filp;
1da177e4 2450
31e6b01f 2451out_filp:
806b681c 2452 filp = ERR_PTR(error);
10fa8e62 2453 goto out;
1da177e4
LT		 2454}
2455
13aab428
AV		 2456struct file *do_filp_open(int dfd, const char *pathname,
2457 const struct open_flags *op, int flags)
2458{
73d049a4 2459 struct nameidata nd;
13aab428
AV		 2460 struct file *filp;
2461
73d049a4 2462 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
13aab428 2463 if (unlikely(filp == ERR_PTR(-ECHILD)))
73d049a4 2464 filp = path_openat(dfd, pathname, &nd, op, flags);
13aab428 2465 if (unlikely(filp == ERR_PTR(-ESTALE)))
73d049a4 2466 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
13aab428
AV		 2467 return filp;
2468}
2469
73d049a4
AV		 2470struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2471 const char *name, const struct open_flags *op, int flags)
2472{
2473 struct nameidata nd;
2474 struct file *file;
2475
2476 nd.root.mnt = mnt;
2477 nd.root.dentry = dentry;
2478
2479 flags |= LOOKUP_ROOT;
2480
bcda7652 2481 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
73d049a4
AV		 2482 return ERR_PTR(-ELOOP);
2483
2484 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2485 if (unlikely(file == ERR_PTR(-ECHILD)))
2486 file = path_openat(-1, name, &nd, op, flags);
2487 if (unlikely(file == ERR_PTR(-ESTALE)))
2488 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2489 return file;
2490}
2491
ed75e95d 2492struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
1da177e4 2493{
c663e5d8 2494 struct dentry *dentry = ERR_PTR(-EEXIST);
ed75e95d
AV		 2495 struct nameidata nd;
2496 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2497 if (error)
2498 return ERR_PTR(error);
1da177e4 2499
c663e5d8
CH		 2500 /*
2501 * Yucky last component or no last component at all?
2502 * (foo/., foo/.., /////)
2503 */
ed75e95d
AV		 2504 if (nd.last_type != LAST_NORM)
2505 goto out;
2506 nd.flags &= ~LOOKUP_PARENT;
2507 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2508 nd.intent.open.flags = O_EXCL;
c663e5d8
CH		 2509
2510 /*
2511 * Do the final lookup.
2512 */
ed75e95d
AV		 2513 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2514 dentry = lookup_hash(&nd);
1da177e4
LT		 2515 if (IS_ERR(dentry))
2516 goto fail;
c663e5d8 2517
e9baf6e5
AV		 2518 if (dentry->d_inode)
2519 goto eexist;
c663e5d8
CH		 2520 /*
2521 * Special case - lookup gave negative, but... we had foo/bar/
2522 * From the vfs_mknod() POV we just have a negative dentry -
2523 * all is fine. Let's be bastards - you had / on the end, you've
2524 * been asking for (non-existent) directory. -ENOENT for you.
2525 */
ed75e95d 2526 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
e9baf6e5
AV		 2527 dput(dentry);
2528 dentry = ERR_PTR(-ENOENT);
ed75e95d 2529 goto fail;
e9baf6e5 2530 }
ed75e95d 2531 *path = nd.path;
1da177e4 2532 return dentry;
e9baf6e5 2533eexist:
1da177e4 2534 dput(dentry);
e9baf6e5 2535 dentry = ERR_PTR(-EEXIST);
1da177e4 2536fail:
ed75e95d
AV		 2537 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2538out:
2539 path_put(&nd.path);
1da177e4
LT		 2540 return dentry;
2541}
dae6ad8f
AV		 2542EXPORT_SYMBOL(kern_path_create);
2543
2544struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2545{
2546 char *tmp = getname(pathname);
2547 struct dentry *res;
2548 if (IS_ERR(tmp))
2549 return ERR_CAST(tmp);
2550 res = kern_path_create(dfd, tmp, path, is_dir);
2551 putname(tmp);
2552 return res;
2553}
2554EXPORT_SYMBOL(user_path_create);
2555
1a67aafb 2556int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
1da177e4 2557{
a95164d9 2558 int error = may_create(dir, dentry);
1da177e4
LT		 2559
2560 if (error)
2561 return error;
2562
975d6b39 2563 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1da177e4
LT		 2564 return -EPERM;
2565
acfa4380 2566 if (!dir->i_op->mknod)
1da177e4
LT		 2567 return -EPERM;
2568
08ce5f16
SH		 2569 error = devcgroup_inode_mknod(mode, dev);
2570 if (error)
2571 return error;
2572
1da177e4
LT		 2573 error = security_inode_mknod(dir, dentry, mode, dev);
2574 if (error)
2575 return error;
2576
1da177e4 2577 error = dir->i_op->mknod(dir, dentry, mode, dev);
a74574aa 2578 if (!error)
f38aa942 2579 fsnotify_create(dir, dentry);
1da177e4
LT		 2580 return error;
2581}
2582
f69aac00 2583static int may_mknod(umode_t mode)
463c3197
DH		 2584{
2585 switch (mode & S_IFMT) {
2586 case S_IFREG:
2587 case S_IFCHR:
2588 case S_IFBLK:
2589 case S_IFIFO:
2590 case S_IFSOCK:
2591 case 0: /* zero mode translates to S_IFREG */
2592 return 0;
2593 case S_IFDIR:
2594 return -EPERM;
2595 default:
2596 return -EINVAL;
2597 }
2598}
2599
8208a22b 2600SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
2e4d0924 2601 unsigned, dev)
1da177e4 2602{
2ad94ae6 2603 struct dentry *dentry;
dae6ad8f
AV		 2604 struct path path;
2605 int error;
1da177e4
LT		 2606
2607 if (S_ISDIR(mode))
2608 return -EPERM;
1da177e4 2609
dae6ad8f
AV		 2610 dentry = user_path_create(dfd, filename, &path, 0);
2611 if (IS_ERR(dentry))
2612 return PTR_ERR(dentry);
2ad94ae6 2613
dae6ad8f 2614 if (!IS_POSIXACL(path.dentry->d_inode))
ce3b0f8d 2615 mode &= ~current_umask();
463c3197
DH		 2616 error = may_mknod(mode);
2617 if (error)
2618 goto out_dput;
dae6ad8f 2619 error = mnt_want_write(path.mnt);
463c3197
DH		 2620 if (error)
2621 goto out_dput;
dae6ad8f 2622 error = security_path_mknod(&path, dentry, mode, dev);
be6d3e56
KT		 2623 if (error)
2624 goto out_drop_write;
463c3197 2625 switch (mode & S_IFMT) {
1da177e4 2626 case 0: case S_IFREG:
dae6ad8f 2627 error = vfs_create(path.dentry->d_inode,dentry,mode,NULL);
1da177e4
LT		 2628 break;
2629 case S_IFCHR: case S_IFBLK:
dae6ad8f 2630 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
1da177e4
LT		 2631 new_decode_dev(dev));
2632 break;
2633 case S_IFIFO: case S_IFSOCK:
dae6ad8f 2634 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
1da177e4 2635 break;
1da177e4 2636 }
be6d3e56 2637out_drop_write:
dae6ad8f 2638 mnt_drop_write(path.mnt);
463c3197
DH		 2639out_dput:
2640 dput(dentry);
dae6ad8f
AV		 2641 mutex_unlock(&path.dentry->d_inode->i_mutex);
2642 path_put(&path);
1da177e4
LT		 2643
2644 return error;
2645}
2646
8208a22b 2647SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
5590ff0d
UD		 2648{
2649 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2650}
2651
18bb1db3 2652int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1da177e4 2653{
a95164d9 2654 int error = may_create(dir, dentry);
8de52778 2655 unsigned max_links = dir->i_sb->s_max_links;
1da177e4
LT		 2656
2657 if (error)
2658 return error;
2659
acfa4380 2660 if (!dir->i_op->mkdir)
1da177e4
LT		 2661 return -EPERM;
2662
2663 mode &= (S_IRWXUGO|S_ISVTX);
2664 error = security_inode_mkdir(dir, dentry, mode);
2665 if (error)
2666 return error;
2667
8de52778
AV		 2668 if (max_links && dir->i_nlink >= max_links)
2669 return -EMLINK;
2670
1da177e4 2671 error = dir->i_op->mkdir(dir, dentry, mode);
a74574aa 2672 if (!error)
f38aa942 2673 fsnotify_mkdir(dir, dentry);
1da177e4
LT		 2674 return error;
2675}
2676
a218d0fd 2677SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
1da177e4 2678{
6902d925 2679 struct dentry *dentry;
dae6ad8f
AV		 2680 struct path path;
2681 int error;
1da177e4 2682
dae6ad8f 2683 dentry = user_path_create(dfd, pathname, &path, 1);
6902d925 2684 if (IS_ERR(dentry))
dae6ad8f 2685 return PTR_ERR(dentry);
1da177e4 2686
dae6ad8f 2687 if (!IS_POSIXACL(path.dentry->d_inode))
ce3b0f8d 2688 mode &= ~current_umask();
dae6ad8f 2689 error = mnt_want_write(path.mnt);
463c3197
DH		 2690 if (error)
2691 goto out_dput;
dae6ad8f 2692 error = security_path_mkdir(&path, dentry, mode);
be6d3e56
KT		 2693 if (error)
2694 goto out_drop_write;
dae6ad8f 2695 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
be6d3e56 2696out_drop_write:
dae6ad8f 2697 mnt_drop_write(path.mnt);
463c3197 2698out_dput:
6902d925 2699 dput(dentry);
dae6ad8f
AV		 2700 mutex_unlock(&path.dentry->d_inode->i_mutex);
2701 path_put(&path);
1da177e4
LT		 2702 return error;
2703}
2704
a218d0fd 2705SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
5590ff0d
UD		 2706{
2707 return sys_mkdirat(AT_FDCWD, pathname, mode);
2708}
2709
1da177e4 2710/*
a71905f0 2711 * The dentry_unhash() helper will try to drop the dentry early: we
c0d02594 2712 * should have a usage count of 1 if we're the only user of this
a71905f0
SW		 2713 * dentry, and if that is true (possibly after pruning the dcache),
2714 * then we drop the dentry now.
1da177e4
LT		 2715 *
2716 * A low-level filesystem can, if it choses, legally
2717 * do a
2718 *
2719 * if (!d_unhashed(dentry))
2720 * return -EBUSY;
2721 *
2722 * if it cannot handle the case of removing a directory
2723 * that is still in use by something else..
2724 */
2725void dentry_unhash(struct dentry *dentry)
2726{
dc168427 2727 shrink_dcache_parent(dentry);
1da177e4 2728 spin_lock(&dentry->d_lock);
64252c75 2729 if (dentry->d_count == 1)
1da177e4
LT		 2730 __d_drop(dentry);
2731 spin_unlock(&dentry->d_lock);
1da177e4
LT		 2732}
2733
2734int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2735{
2736 int error = may_delete(dir, dentry, 1);
2737
2738 if (error)
2739 return error;
2740
acfa4380 2741 if (!dir->i_op->rmdir)
1da177e4
LT		 2742 return -EPERM;
2743
1d2ef590 2744 dget(dentry);
1b1dcc1b 2745 mutex_lock(&dentry->d_inode->i_mutex);
912dbc15
SW		 2746
2747 error = -EBUSY;
1da177e4 2748 if (d_mountpoint(dentry))
912dbc15
SW		 2749 goto out;
2750
2751 error = security_inode_rmdir(dir, dentry);
2752 if (error)
2753 goto out;
2754
3cebde24 2755 shrink_dcache_parent(dentry);
912dbc15
SW		 2756 error = dir->i_op->rmdir(dir, dentry);
2757 if (error)
2758 goto out;
2759
2760 dentry->d_inode->i_flags |= S_DEAD;
2761 dont_mount(dentry);
2762
2763out:
1b1dcc1b 2764 mutex_unlock(&dentry->d_inode->i_mutex);
1d2ef590 2765 dput(dentry);
912dbc15 2766 if (!error)
1da177e4 2767 d_delete(dentry);
1da177e4
LT		 2768 return error;
2769}
2770
5590ff0d 2771static long do_rmdir(int dfd, const char __user *pathname)
1da177e4
LT		 2772{
2773 int error = 0;
2774 char * name;
2775 struct dentry *dentry;
2776 struct nameidata nd;
2777
2ad94ae6 2778 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2779 if (error)
2ad94ae6 2780 return error;
1da177e4
LT		 2781
2782 switch(nd.last_type) {
0612d9fb
OH		 2783 case LAST_DOTDOT:
2784 error = -ENOTEMPTY;
2785 goto exit1;
2786 case LAST_DOT:
2787 error = -EINVAL;
2788 goto exit1;
2789 case LAST_ROOT:
2790 error = -EBUSY;
2791 goto exit1;
1da177e4 2792 }
0612d9fb
OH		 2793
2794 nd.flags &= ~LOOKUP_PARENT;
2795
4ac91378 2796 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2797 dentry = lookup_hash(&nd);
1da177e4 2798 error = PTR_ERR(dentry);
6902d925
DH		 2799 if (IS_ERR(dentry))
2800 goto exit2;
e6bc45d6
TT		 2801 if (!dentry->d_inode) {
2802 error = -ENOENT;
2803 goto exit3;
2804 }
0622753b
DH		 2805 error = mnt_want_write(nd.path.mnt);
2806 if (error)
2807 goto exit3;
be6d3e56
KT		 2808 error = security_path_rmdir(&nd.path, dentry);
2809 if (error)
2810 goto exit4;
4ac91378 2811 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
be6d3e56 2812exit4:
0622753b
DH		 2813 mnt_drop_write(nd.path.mnt);
2814exit3:
6902d925
DH		 2815 dput(dentry);
2816exit2:
4ac91378 2817 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2818exit1:
1d957f9b 2819 path_put(&nd.path);
1da177e4
LT		 2820 putname(name);
2821 return error;
2822}
2823
3cdad428 2824SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
5590ff0d
UD		 2825{
2826 return do_rmdir(AT_FDCWD, pathname);
2827}
2828
1da177e4
LT		 2829int vfs_unlink(struct inode *dir, struct dentry *dentry)
2830{
2831 int error = may_delete(dir, dentry, 0);
2832
2833 if (error)
2834 return error;
2835
acfa4380 2836 if (!dir->i_op->unlink)
1da177e4
LT		 2837 return -EPERM;
2838
1b1dcc1b 2839 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2840 if (d_mountpoint(dentry))
2841 error = -EBUSY;
2842 else {
2843 error = security_inode_unlink(dir, dentry);
bec1052e 2844 if (!error) {
1da177e4 2845 error = dir->i_op->unlink(dir, dentry);
bec1052e 2846 if (!error)
d83c49f3 2847 dont_mount(dentry);
bec1052e 2848 }
1da177e4 2849 }
1b1dcc1b 2850 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2851
2852 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2853 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
ece95912 2854 fsnotify_link_count(dentry->d_inode);
e234f35c 2855 d_delete(dentry);
1da177e4 2856 }
0eeca283 2857
1da177e4
LT		 2858 return error;
2859}
2860
2861/*
2862 * Make sure that the actual truncation of the file will occur outside its
1b1dcc1b 2863 * directory's i_mutex. Truncate can take a long time if there is a lot of
1da177e4
LT		 2864 * writeout happening, and we don't want to prevent access to the directory
2865 * while waiting on the I/O.
2866 */
5590ff0d 2867static long do_unlinkat(int dfd, const char __user *pathname)
1da177e4 2868{
2ad94ae6
AV		 2869 int error;
2870 char *name;
1da177e4
LT		 2871 struct dentry *dentry;
2872 struct nameidata nd;
2873 struct inode *inode = NULL;
2874
2ad94ae6 2875 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2876 if (error)
2ad94ae6
AV		 2877 return error;
2878
1da177e4
LT		 2879 error = -EISDIR;
2880 if (nd.last_type != LAST_NORM)
2881 goto exit1;
0612d9fb
OH		 2882
2883 nd.flags &= ~LOOKUP_PARENT;
2884
4ac91378 2885 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2886 dentry = lookup_hash(&nd);
1da177e4
LT		 2887 error = PTR_ERR(dentry);
2888 if (!IS_ERR(dentry)) {
2889 /* Why not before? Because we want correct error value */
50338b88
TE		 2890 if (nd.last.name[nd.last.len])
2891 goto slashes;
1da177e4 2892 inode = dentry->d_inode;
50338b88 2893 if (!inode)
e6bc45d6
TT		 2894 goto slashes;
2895 ihold(inode);
0622753b
DH		 2896 error = mnt_want_write(nd.path.mnt);
2897 if (error)
2898 goto exit2;
be6d3e56
KT		 2899 error = security_path_unlink(&nd.path, dentry);
2900 if (error)
2901 goto exit3;
4ac91378 2902 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
be6d3e56 2903exit3:
0622753b 2904 mnt_drop_write(nd.path.mnt);
1da177e4
LT		 2905 exit2:
2906 dput(dentry);
2907 }
4ac91378 2908 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4
LT		 2909 if (inode)
2910 iput(inode); /* truncate the inode here */
2911exit1:
1d957f9b 2912 path_put(&nd.path);
1da177e4
LT		 2913 putname(name);
2914 return error;
2915
2916slashes:
2917 error = !dentry->d_inode ? -ENOENT :
2918 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2919 goto exit2;
2920}
2921
2e4d0924 2922SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
5590ff0d
UD		 2923{
2924 if ((flag & ~AT_REMOVEDIR) != 0)
2925 return -EINVAL;
2926
2927 if (flag & AT_REMOVEDIR)
2928 return do_rmdir(dfd, pathname);
2929
2930 return do_unlinkat(dfd, pathname);
2931}
2932
3480b257 2933SYSCALL_DEFINE1(unlink, const char __user *, pathname)
5590ff0d
UD		 2934{
2935 return do_unlinkat(AT_FDCWD, pathname);
2936}
2937
db2e747b 2938int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
1da177e4 2939{
a95164d9 2940 int error = may_create(dir, dentry);
1da177e4
LT		 2941
2942 if (error)
2943 return error;
2944
acfa4380 2945 if (!dir->i_op->symlink)
1da177e4
LT		 2946 return -EPERM;
2947
2948 error = security_inode_symlink(dir, dentry, oldname);
2949 if (error)
2950 return error;
2951
1da177e4 2952 error = dir->i_op->symlink(dir, dentry, oldname);
a74574aa 2953 if (!error)
f38aa942 2954 fsnotify_create(dir, dentry);
1da177e4
LT		 2955 return error;
2956}
2957
2e4d0924
HC		 2958SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2959 int, newdfd, const char __user *, newname)
1da177e4 2960{
2ad94ae6
AV		 2961 int error;
2962 char *from;
6902d925 2963 struct dentry *dentry;
dae6ad8f 2964 struct path path;
1da177e4
LT		 2965
2966 from = getname(oldname);
2ad94ae6 2967 if (IS_ERR(from))
1da177e4 2968 return PTR_ERR(from);
1da177e4 2969
dae6ad8f 2970 dentry = user_path_create(newdfd, newname, &path, 0);
6902d925
DH		 2971 error = PTR_ERR(dentry);
2972 if (IS_ERR(dentry))
dae6ad8f 2973 goto out_putname;
6902d925 2974
dae6ad8f 2975 error = mnt_want_write(path.mnt);
75c3f29d
DH		 2976 if (error)
2977 goto out_dput;
dae6ad8f 2978 error = security_path_symlink(&path, dentry, from);
be6d3e56
KT		 2979 if (error)
2980 goto out_drop_write;
dae6ad8f 2981 error = vfs_symlink(path.dentry->d_inode, dentry, from);
be6d3e56 2982out_drop_write:
dae6ad8f 2983 mnt_drop_write(path.mnt);
75c3f29d 2984out_dput:
6902d925 2985 dput(dentry);
dae6ad8f
AV		 2986 mutex_unlock(&path.dentry->d_inode->i_mutex);
2987 path_put(&path);
6902d925 2988out_putname:
1da177e4
LT		 2989 putname(from);
2990 return error;
2991}
2992
3480b257 2993SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 2994{
2995 return sys_symlinkat(oldname, AT_FDCWD, newname);
2996}
2997
1da177e4
LT		 2998int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2999{
3000 struct inode *inode = old_dentry->d_inode;
8de52778 3001 unsigned max_links = dir->i_sb->s_max_links;
1da177e4
LT		 3002 int error;
3003
3004 if (!inode)
3005 return -ENOENT;
3006
a95164d9 3007 error = may_create(dir, new_dentry);
1da177e4
LT		 3008 if (error)
3009 return error;
3010
3011 if (dir->i_sb != inode->i_sb)
3012 return -EXDEV;
3013
3014 /*
3015 * A link to an append-only or immutable file cannot be created.
3016 */
3017 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3018 return -EPERM;
acfa4380 3019 if (!dir->i_op->link)
1da177e4 3020 return -EPERM;
7e79eedb 3021 if (S_ISDIR(inode->i_mode))
1da177e4
LT		 3022 return -EPERM;
3023
3024 error = security_inode_link(old_dentry, dir, new_dentry);
3025 if (error)
3026 return error;
3027
7e79eedb 3028 mutex_lock(&inode->i_mutex);
aae8a97d
AK		 3029 /* Make sure we don't allow creating hardlink to an unlinked file */
3030 if (inode->i_nlink == 0)
3031 error = -ENOENT;
8de52778
AV		 3032 else if (max_links && inode->i_nlink >= max_links)
3033 error = -EMLINK;
aae8a97d
AK		 3034 else
3035 error = dir->i_op->link(old_dentry, dir, new_dentry);
7e79eedb 3036 mutex_unlock(&inode->i_mutex);
e31e14ec 3037 if (!error)
7e79eedb 3038 fsnotify_link(dir, inode, new_dentry);
1da177e4
LT		 3039 return error;
3040}
3041
3042/*
3043 * Hardlinks are often used in delicate situations. We avoid
3044 * security-related surprises by not following symlinks on the
3045 * newname. --KAB
3046 *
3047 * We don't follow them on the oldname either to be compatible
3048 * with linux 2.0, and to avoid hard-linking to directories
3049 * and other special files. --ADM
3050 */
2e4d0924
HC		 3051SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3052 int, newdfd, const char __user *, newname, int, flags)
1da177e4
LT		 3053{
3054 struct dentry *new_dentry;
dae6ad8f 3055 struct path old_path, new_path;
11a7b371 3056 int how = 0;
1da177e4 3057 int error;
1da177e4 3058
11a7b371 3059 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
c04030e1 3060 return -EINVAL;
11a7b371
AK		 3061 /*
3062 * To use null names we require CAP_DAC_READ_SEARCH
3063 * This ensures that not everyone will be able to create
3064 * handlink using the passed filedescriptor.
3065 */
3066 if (flags & AT_EMPTY_PATH) {
3067 if (!capable(CAP_DAC_READ_SEARCH))
3068 return -ENOENT;
3069 how = LOOKUP_EMPTY;
3070 }
3071
3072 if (flags & AT_SYMLINK_FOLLOW)
3073 how |= LOOKUP_FOLLOW;
c04030e1 3074
11a7b371 3075 error = user_path_at(olddfd, oldname, how, &old_path);
1da177e4 3076 if (error)
2ad94ae6
AV		 3077 return error;
3078
dae6ad8f 3079 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
1da177e4 3080 error = PTR_ERR(new_dentry);
6902d925 3081 if (IS_ERR(new_dentry))
dae6ad8f
AV		 3082 goto out;
3083
3084 error = -EXDEV;
3085 if (old_path.mnt != new_path.mnt)
3086 goto out_dput;
3087 error = mnt_want_write(new_path.mnt);
75c3f29d
DH		 3088 if (error)
3089 goto out_dput;
dae6ad8f 3090 error = security_path_link(old_path.dentry, &new_path, new_dentry);
be6d3e56
KT		 3091 if (error)
3092 goto out_drop_write;
dae6ad8f 3093 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
be6d3e56 3094out_drop_write:
dae6ad8f 3095 mnt_drop_write(new_path.mnt);
75c3f29d 3096out_dput:
6902d925 3097 dput(new_dentry);
dae6ad8f
AV		 3098 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
3099 path_put(&new_path);
1da177e4 3100out:
2d8f3038 3101 path_put(&old_path);
1da177e4
LT		 3102
3103 return error;
3104}
3105
3480b257 3106SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
5590ff0d 3107{
c04030e1 3108 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
5590ff0d
UD		 3109}
3110
1da177e4
LT		 3111/*
3112 * The worst of all namespace operations - renaming directory. "Perverted"
3113 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3114 * Problems:
3115 * a) we can get into loop creation. Check is done in is_subdir().
3116 * b) race potential - two innocent renames can create a loop together.
3117 * That's where 4.4 screws up. Current fix: serialization on
a11f3a05 3118 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
1da177e4
LT		 3119 * story.
3120 * c) we have to lock _three_ objects - parents and victim (if it exists).
1b1dcc1b 3121 * And that - after we got ->i_mutex on parents (until then we don't know
1da177e4
LT		 3122 * whether the target exists). Solution: try to be smart with locking
3123 * order for inodes. We rely on the fact that tree topology may change
a11f3a05 3124 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
1da177e4
LT		 3125 * move will be locked. Thus we can rank directories by the tree
3126 * (ancestors first) and rank all non-directories after them.
3127 * That works since everybody except rename does "lock parent, lookup,
a11f3a05 3128 * lock child" and rename is under ->s_vfs_rename_mutex.
1da177e4
LT		 3129 * HOWEVER, it relies on the assumption that any object with ->lookup()
3130 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3131 * we'd better make sure that there's no link(2) for them.
e4eaac06 3132 * d) conversion from fhandle to dentry may come in the wrong moment - when
1b1dcc1b 3133 * we are removing the target. Solution: we will have to grab ->i_mutex
1da177e4 3134 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
c41b20e7 3135 * ->i_mutex on parents, which works but leads to some truly excessive
1da177e4
LT		 3136 * locking].
3137 */
75c96f85
AB		 3138static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3139 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3140{
3141 int error = 0;
9055cba7 3142 struct inode *target = new_dentry->d_inode;
8de52778 3143 unsigned max_links = new_dir->i_sb->s_max_links;
1da177e4
LT		 3144
3145 /*
3146 * If we are going to change the parent - check write permissions,
3147 * we'll need to flip '..'.
3148 */
3149 if (new_dir != old_dir) {
f419a2e3 3150 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
1da177e4
LT		 3151 if (error)
3152 return error;
3153 }
3154
3155 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3156 if (error)
3157 return error;
3158
1d2ef590 3159 dget(new_dentry);
d83c49f3 3160 if (target)
1b1dcc1b 3161 mutex_lock(&target->i_mutex);
9055cba7
SW		 3162
3163 error = -EBUSY;
3164 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3165 goto out;
3166
8de52778
AV		 3167 error = -EMLINK;
3168 if (max_links && !target && new_dir != old_dir &&
3169 new_dir->i_nlink >= max_links)
3170 goto out;
3171
3cebde24
SW		 3172 if (target)
3173 shrink_dcache_parent(new_dentry);
9055cba7
SW		 3174 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3175 if (error)
3176 goto out;
3177
1da177e4 3178 if (target) {
9055cba7
SW		 3179 target->i_flags |= S_DEAD;
3180 dont_mount(new_dentry);
1da177e4 3181 }
9055cba7
SW		 3182out:
3183 if (target)
3184 mutex_unlock(&target->i_mutex);
1d2ef590 3185 dput(new_dentry);
e31e14ec 3186 if (!error)
349457cc
MF		 3187 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3188 d_move(old_dentry,new_dentry);
1da177e4
LT		 3189 return error;
3190}
3191
75c96f85
AB		 3192static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3193 struct inode *new_dir, struct dentry *new_dentry)
1da177e4 3194{
51892bbb 3195 struct inode *target = new_dentry->d_inode;
1da177e4
LT		 3196 int error;
3197
3198 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3199 if (error)
3200 return error;
3201
3202 dget(new_dentry);
1da177e4 3203 if (target)
1b1dcc1b 3204 mutex_lock(&target->i_mutex);
51892bbb
SW		 3205
3206 error = -EBUSY;
1da177e4 3207 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
51892bbb
SW		 3208 goto out;
3209
3210 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3211 if (error)
3212 goto out;
3213
3214 if (target)
3215 dont_mount(new_dentry);
3216 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3217 d_move(old_dentry, new_dentry);
3218out:
1da177e4 3219 if (target)
1b1dcc1b 3220 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3221 dput(new_dentry);
3222 return error;
3223}
3224
3225int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3226 struct inode *new_dir, struct dentry *new_dentry)
3227{
3228 int error;
3229 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
59b0df21 3230 const unsigned char *old_name;
1da177e4
LT		 3231
3232 if (old_dentry->d_inode == new_dentry->d_inode)
3233 return 0;
3234
3235 error = may_delete(old_dir, old_dentry, is_dir);
3236 if (error)
3237 return error;
3238
3239 if (!new_dentry->d_inode)
a95164d9 3240 error = may_create(new_dir, new_dentry);
1da177e4
LT		 3241 else
3242 error = may_delete(new_dir, new_dentry, is_dir);
3243 if (error)
3244 return error;
3245
acfa4380 3246 if (!old_dir->i_op->rename)
1da177e4
LT		 3247 return -EPERM;
3248
0eeca283
RL		 3249 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3250
1da177e4
LT		 3251 if (is_dir)
3252 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3253 else
3254 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
123df294
AV		 3255 if (!error)
3256 fsnotify_move(old_dir, new_dir, old_name, is_dir,
5a190ae6 3257 new_dentry->d_inode, old_dentry);
0eeca283
RL		 3258 fsnotify_oldname_free(old_name);
3259
1da177e4
LT		 3260 return error;
3261}
3262
2e4d0924
HC		 3263SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3264 int, newdfd, const char __user *, newname)
1da177e4 3265{
2ad94ae6
AV		 3266 struct dentry *old_dir, *new_dir;
3267 struct dentry *old_dentry, *new_dentry;
3268 struct dentry *trap;
1da177e4 3269 struct nameidata oldnd, newnd;
2ad94ae6
AV		 3270 char *from;
3271 char *to;
3272 int error;
1da177e4 3273
2ad94ae6 3274 error = user_path_parent(olddfd, oldname, &oldnd, &from);
1da177e4
LT		 3275 if (error)
3276 goto exit;
3277
2ad94ae6 3278 error = user_path_parent(newdfd, newname, &newnd, &to);
1da177e4
LT		 3279 if (error)
3280 goto exit1;
3281
3282 error = -EXDEV;
4ac91378 3283 if (oldnd.path.mnt != newnd.path.mnt)
1da177e4
LT		 3284 goto exit2;
3285
4ac91378 3286 old_dir = oldnd.path.dentry;
1da177e4
LT		 3287 error = -EBUSY;
3288 if (oldnd.last_type != LAST_NORM)
3289 goto exit2;
3290
4ac91378 3291 new_dir = newnd.path.dentry;
1da177e4
LT		 3292 if (newnd.last_type != LAST_NORM)
3293 goto exit2;
3294
0612d9fb
OH		 3295 oldnd.flags &= ~LOOKUP_PARENT;
3296 newnd.flags &= ~LOOKUP_PARENT;
4e9ed2f8 3297 newnd.flags |= LOOKUP_RENAME_TARGET;
0612d9fb 3298
1da177e4
LT		 3299 trap = lock_rename(new_dir, old_dir);
3300
49705b77 3301 old_dentry = lookup_hash(&oldnd);
1da177e4
LT		 3302 error = PTR_ERR(old_dentry);
3303 if (IS_ERR(old_dentry))
3304 goto exit3;
3305 /* source must exist */
3306 error = -ENOENT;
3307 if (!old_dentry->d_inode)
3308 goto exit4;
3309 /* unless the source is a directory trailing slashes give -ENOTDIR */
3310 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3311 error = -ENOTDIR;
3312 if (oldnd.last.name[oldnd.last.len])
3313 goto exit4;
3314 if (newnd.last.name[newnd.last.len])
3315 goto exit4;
3316 }
3317 /* source should not be ancestor of target */
3318 error = -EINVAL;
3319 if (old_dentry == trap)
3320 goto exit4;
49705b77 3321 new_dentry = lookup_hash(&newnd);
1da177e4
LT		 3322 error = PTR_ERR(new_dentry);
3323 if (IS_ERR(new_dentry))
3324 goto exit4;
3325 /* target should not be an ancestor of source */
3326 error = -ENOTEMPTY;
3327 if (new_dentry == trap)
3328 goto exit5;
3329
9079b1eb
DH		 3330 error = mnt_want_write(oldnd.path.mnt);
3331 if (error)
3332 goto exit5;
be6d3e56
KT		 3333 error = security_path_rename(&oldnd.path, old_dentry,
3334 &newnd.path, new_dentry);
3335 if (error)
3336 goto exit6;
1da177e4
LT		 3337 error = vfs_rename(old_dir->d_inode, old_dentry,
3338 new_dir->d_inode, new_dentry);
be6d3e56 3339exit6:
9079b1eb 3340 mnt_drop_write(oldnd.path.mnt);
1da177e4
LT		 3341exit5:
3342 dput(new_dentry);
3343exit4:
3344 dput(old_dentry);
3345exit3:
3346 unlock_rename(new_dir, old_dir);
3347exit2:
1d957f9b 3348 path_put(&newnd.path);
2ad94ae6 3349 putname(to);
1da177e4 3350exit1:
1d957f9b 3351 path_put(&oldnd.path);
1da177e4 3352 putname(from);
2ad94ae6 3353exit:
1da177e4
LT		 3354 return error;
3355}
3356
a26eab24 3357SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 3358{
3359 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3360}
3361
1da177e4
LT		 3362int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3363{
3364 int len;
3365
3366 len = PTR_ERR(link);
3367 if (IS_ERR(link))
3368 goto out;
3369
3370 len = strlen(link);
3371 if (len > (unsigned) buflen)
3372 len = buflen;
3373 if (copy_to_user(buffer, link, len))
3374 len = -EFAULT;
3375out:
3376 return len;
3377}
3378
3379/*
3380 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3381 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3382 * using) it for any given inode is up to filesystem.
3383 */
3384int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3385{
3386 struct nameidata nd;
cc314eef 3387 void *cookie;
694a1764 3388 int res;
cc314eef 3389
1da177e4 3390 nd.depth = 0;
cc314eef 3391 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
694a1764
MS		 3392 if (IS_ERR(cookie))
3393 return PTR_ERR(cookie);
3394
3395 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3396 if (dentry->d_inode->i_op->put_link)
3397 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3398 return res;
1da177e4
LT		 3399}
3400
3401int vfs_follow_link(struct nameidata *nd, const char *link)
3402{
3403 return __vfs_follow_link(nd, link);
3404}
3405
3406/* get the link contents into pagecache */
3407static char *page_getlink(struct dentry * dentry, struct page **ppage)
3408{
ebd09abb
DG		 3409 char *kaddr;
3410 struct page *page;
1da177e4 3411 struct address_space *mapping = dentry->d_inode->i_mapping;
090d2b18 3412 page = read_mapping_page(mapping, 0, NULL);
1da177e4 3413 if (IS_ERR(page))
6fe6900e 3414 return (char*)page;
1da177e4 3415 *ppage = page;
ebd09abb
DG		 3416 kaddr = kmap(page);
3417 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3418 return kaddr;
1da177e4
LT		 3419}
3420
3421int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3422{
3423 struct page *page = NULL;
3424 char *s = page_getlink(dentry, &page);
3425 int res = vfs_readlink(dentry,buffer,buflen,s);
3426 if (page) {
3427 kunmap(page);
3428 page_cache_release(page);
3429 }
3430 return res;
3431}
3432
cc314eef 3433void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
1da177e4 3434{
cc314eef 3435 struct page *page = NULL;
1da177e4 3436 nd_set_link(nd, page_getlink(dentry, &page));
cc314eef 3437 return page;
1da177e4
LT		 3438}
3439
cc314eef 3440void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1da177e4 3441{
cc314eef
LT		 3442 struct page *page = cookie;
3443
3444 if (page) {
1da177e4
LT		 3445 kunmap(page);
3446 page_cache_release(page);
1da177e4
LT		 3447 }
3448}
3449
54566b2c
NP		 3450/*
3451 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3452 */
3453int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
1da177e4
LT		 3454{
3455 struct address_space *mapping = inode->i_mapping;
0adb25d2 3456 struct page *page;
afddba49 3457 void *fsdata;
beb497ab 3458 int err;
1da177e4 3459 char *kaddr;
54566b2c
NP		 3460 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3461 if (nofs)
3462 flags |= AOP_FLAG_NOFS;
1da177e4 3463
7e53cac4 3464retry:
afddba49 3465 err = pagecache_write_begin(NULL, mapping, 0, len-1,
54566b2c 3466 flags, &page, &fsdata);
1da177e4 3467 if (err)
afddba49
NP		 3468 goto fail;
3469
e8e3c3d6 3470 kaddr = kmap_atomic(page);
1da177e4 3471 memcpy(kaddr, symname, len-1);
e8e3c3d6 3472 kunmap_atomic(kaddr);
afddba49
NP		 3473
3474 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3475 page, fsdata);
1da177e4
LT		 3476 if (err < 0)
3477 goto fail;
afddba49
NP		 3478 if (err < len-1)
3479 goto retry;
3480
1da177e4
LT		 3481 mark_inode_dirty(inode);
3482 return 0;
1da177e4
LT		 3483fail:
3484 return err;
3485}
3486
0adb25d2
KK		 3487int page_symlink(struct inode *inode, const char *symname, int len)
3488{
3489 return __page_symlink(inode, symname, len,
54566b2c 3490 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
0adb25d2
KK		 3491}
3492
92e1d5be 3493const struct inode_operations page_symlink_inode_operations = {
1da177e4
LT		 3494 .readlink = generic_readlink,
3495 .follow_link = page_follow_link_light,
3496 .put_link = page_put_link,
3497};
3498
2d8f3038 3499EXPORT_SYMBOL(user_path_at);
cc53ce53 3500EXPORT_SYMBOL(follow_down_one);
1da177e4
LT		 3501EXPORT_SYMBOL(follow_down);
3502EXPORT_SYMBOL(follow_up);
3503EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3504EXPORT_SYMBOL(getname);
3505EXPORT_SYMBOL(lock_rename);
1da177e4
LT		 3506EXPORT_SYMBOL(lookup_one_len);
3507EXPORT_SYMBOL(page_follow_link_light);
3508EXPORT_SYMBOL(page_put_link);
3509EXPORT_SYMBOL(page_readlink);
0adb25d2 3510EXPORT_SYMBOL(__page_symlink);
1da177e4
LT		 3511EXPORT_SYMBOL(page_symlink);
3512EXPORT_SYMBOL(page_symlink_inode_operations);
d1811465 3513EXPORT_SYMBOL(kern_path);
16f18200 3514EXPORT_SYMBOL(vfs_path_lookup);
f419a2e3 3515EXPORT_SYMBOL(inode_permission);
1da177e4
LT		 3516EXPORT_SYMBOL(unlock_rename);
3517EXPORT_SYMBOL(vfs_create);
3518EXPORT_SYMBOL(vfs_follow_link);
3519EXPORT_SYMBOL(vfs_link);
3520EXPORT_SYMBOL(vfs_mkdir);
3521EXPORT_SYMBOL(vfs_mknod);
3522EXPORT_SYMBOL(generic_permission);
3523EXPORT_SYMBOL(vfs_readlink);
3524EXPORT_SYMBOL(vfs_rename);
3525EXPORT_SYMBOL(vfs_rmdir);
3526EXPORT_SYMBOL(vfs_symlink);
3527EXPORT_SYMBOL(vfs_unlink);
3528EXPORT_SYMBOL(dentry_unhash);
3529EXPORT_SYMBOL(generic_readlink);
kernel source for telekom puls (alcatel/mediatek)