Merge tag 'modules-for-v4.18' of git://git.kernel.org/pub/scm/linux/kernel/git/jeyu...
[linux-2.6-block.git] / fs / namei.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
1da177e4
LT
2/*
3 * linux/fs/namei.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8/*
9 * Some corrections by tytso.
10 */
11
12/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
13 * lookup logic.
14 */
15/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
16 */
17
18#include <linux/init.h>
630d9c47 19#include <linux/export.h>
44696908 20#include <linux/kernel.h>
1da177e4
LT
21#include <linux/slab.h>
22#include <linux/fs.h>
23#include <linux/namei.h>
1da177e4 24#include <linux/pagemap.h>
0eeca283 25#include <linux/fsnotify.h>
1da177e4
LT
26#include <linux/personality.h>
27#include <linux/security.h>
6146f0d5 28#include <linux/ima.h>
1da177e4
LT
29#include <linux/syscalls.h>
30#include <linux/mount.h>
31#include <linux/audit.h>
16f7e0fe 32#include <linux/capability.h>
834f2a4a 33#include <linux/file.h>
5590ff0d 34#include <linux/fcntl.h>
08ce5f16 35#include <linux/device_cgroup.h>
5ad4e53b 36#include <linux/fs_struct.h>
e77819e5 37#include <linux/posix_acl.h>
99d263d4 38#include <linux/hash.h>
2a18da7a 39#include <linux/bitops.h>
aeaa4a79 40#include <linux/init_task.h>
7c0f6ba6 41#include <linux/uaccess.h>
1c949843 42#include <linux/build_bug.h>
1da177e4 43
e81e3f4d 44#include "internal.h"
c7105365 45#include "mount.h"
e81e3f4d 46
1da177e4
LT
47/* [Feb-1997 T. Schoebel-Theuer]
48 * Fundamental changes in the pathname lookup mechanisms (namei)
49 * were necessary because of omirr. The reason is that omirr needs
50 * to know the _real_ pathname, not the user-supplied one, in case
51 * of symlinks (and also when transname replacements occur).
52 *
53 * The new code replaces the old recursive symlink resolution with
54 * an iterative one (in case of non-nested symlink chains). It does
55 * this with calls to <fs>_follow_link().
56 * As a side effect, dir_namei(), _namei() and follow_link() are now
57 * replaced with a single function lookup_dentry() that can handle all
58 * the special cases of the former code.
59 *
60 * With the new dcache, the pathname is stored at each inode, at least as
61 * long as the refcount of the inode is positive. As a side effect, the
62 * size of the dcache depends on the inode cache and thus is dynamic.
63 *
64 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
65 * resolution to correspond with current state of the code.
66 *
67 * Note that the symlink resolution is not *completely* iterative.
68 * There is still a significant amount of tail- and mid- recursion in
69 * the algorithm. Also, note that <fs>_readlink() is not used in
70 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
71 * may return different results than <fs>_follow_link(). Many virtual
72 * filesystems (including /proc) exhibit this behavior.
73 */
74
75/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
76 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
77 * and the name already exists in form of a symlink, try to create the new
78 * name indicated by the symlink. The old code always complained that the
79 * name already exists, due to not following the symlink even if its target
80 * is nonexistent. The new semantics affects also mknod() and link() when
25985edc 81 * the name is a symlink pointing to a non-existent name.
1da177e4
LT
82 *
83 * I don't know which semantics is the right one, since I have no access
84 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
85 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
86 * "old" one. Personally, I think the new semantics is much more logical.
87 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
88 * file does succeed in both HP-UX and SunOs, but not in Solaris
89 * and in the old Linux semantics.
90 */
91
92/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
93 * semantics. See the comments in "open_namei" and "do_link" below.
94 *
95 * [10-Sep-98 Alan Modra] Another symlink change.
96 */
97
98/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
99 * inside the path - always follow.
100 * in the last component in creation/removal/renaming - never follow.
101 * if LOOKUP_FOLLOW passed - follow.
102 * if the pathname has trailing slashes - follow.
103 * otherwise - don't follow.
104 * (applied in that order).
105 *
106 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
107 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
108 * During the 2.4 we need to fix the userland stuff depending on it -
109 * hopefully we will be able to get rid of that wart in 2.5. So far only
110 * XEmacs seems to be relying on it...
111 */
112/*
113 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
a11f3a05 114 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
1da177e4
LT
115 * any extra contention...
116 */
117
118/* In order to reduce some races, while at the same time doing additional
119 * checking and hopefully speeding things up, we copy filenames to the
120 * kernel data space before using them..
121 *
122 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
123 * PATH_MAX includes the nul terminator --RR.
124 */
91a27b2a 125
fd2f7cb5 126#define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
7950e385 127
51f39a1f 128struct filename *
91a27b2a
JL
129getname_flags(const char __user *filename, int flags, int *empty)
130{
94b5d262 131 struct filename *result;
7950e385 132 char *kname;
94b5d262 133 int len;
1c949843 134 BUILD_BUG_ON(offsetof(struct filename, iname) % sizeof(long) != 0);
4043cde8 135
7ac86265
JL
136 result = audit_reusename(filename);
137 if (result)
138 return result;
139
7950e385 140 result = __getname();
3f9f0aa6 141 if (unlikely(!result))
4043cde8
EP
142 return ERR_PTR(-ENOMEM);
143
7950e385
JL
144 /*
145 * First, try to embed the struct filename inside the names_cache
146 * allocation
147 */
fd2f7cb5 148 kname = (char *)result->iname;
91a27b2a 149 result->name = kname;
7950e385 150
94b5d262 151 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
91a27b2a 152 if (unlikely(len < 0)) {
94b5d262
AV
153 __putname(result);
154 return ERR_PTR(len);
91a27b2a 155 }
3f9f0aa6 156
7950e385
JL
157 /*
158 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
159 * separate struct filename so we can dedicate the entire
160 * names_cache allocation for the pathname, and re-do the copy from
161 * userland.
162 */
94b5d262 163 if (unlikely(len == EMBEDDED_NAME_MAX)) {
fd2f7cb5 164 const size_t size = offsetof(struct filename, iname[1]);
7950e385
JL
165 kname = (char *)result;
166
fd2f7cb5
AV
167 /*
168 * size is chosen that way we to guarantee that
169 * result->iname[0] is within the same object and that
170 * kname can't be equal to result->iname, no matter what.
171 */
172 result = kzalloc(size, GFP_KERNEL);
94b5d262
AV
173 if (unlikely(!result)) {
174 __putname(kname);
175 return ERR_PTR(-ENOMEM);
7950e385
JL
176 }
177 result->name = kname;
94b5d262
AV
178 len = strncpy_from_user(kname, filename, PATH_MAX);
179 if (unlikely(len < 0)) {
180 __putname(kname);
181 kfree(result);
182 return ERR_PTR(len);
183 }
184 if (unlikely(len == PATH_MAX)) {
185 __putname(kname);
186 kfree(result);
187 return ERR_PTR(-ENAMETOOLONG);
188 }
7950e385
JL
189 }
190
94b5d262 191 result->refcnt = 1;
3f9f0aa6
LT
192 /* The empty path is special. */
193 if (unlikely(!len)) {
194 if (empty)
4043cde8 195 *empty = 1;
94b5d262
AV
196 if (!(flags & LOOKUP_EMPTY)) {
197 putname(result);
198 return ERR_PTR(-ENOENT);
199 }
1da177e4 200 }
3f9f0aa6 201
7950e385 202 result->uptr = filename;
c4ad8f98 203 result->aname = NULL;
7950e385
JL
204 audit_getname(result);
205 return result;
1da177e4
LT
206}
207
91a27b2a
JL
208struct filename *
209getname(const char __user * filename)
f52e0c11 210{
f7493e5d 211 return getname_flags(filename, 0, NULL);
f52e0c11
AV
212}
213
c4ad8f98
LT
214struct filename *
215getname_kernel(const char * filename)
216{
217 struct filename *result;
08518549 218 int len = strlen(filename) + 1;
c4ad8f98
LT
219
220 result = __getname();
221 if (unlikely(!result))
222 return ERR_PTR(-ENOMEM);
223
08518549 224 if (len <= EMBEDDED_NAME_MAX) {
fd2f7cb5 225 result->name = (char *)result->iname;
08518549 226 } else if (len <= PATH_MAX) {
30ce4d19 227 const size_t size = offsetof(struct filename, iname[1]);
08518549
PM
228 struct filename *tmp;
229
30ce4d19 230 tmp = kmalloc(size, GFP_KERNEL);
08518549
PM
231 if (unlikely(!tmp)) {
232 __putname(result);
233 return ERR_PTR(-ENOMEM);
234 }
235 tmp->name = (char *)result;
08518549
PM
236 result = tmp;
237 } else {
238 __putname(result);
239 return ERR_PTR(-ENAMETOOLONG);
240 }
241 memcpy((char *)result->name, filename, len);
c4ad8f98
LT
242 result->uptr = NULL;
243 result->aname = NULL;
55422d0b 244 result->refcnt = 1;
fd3522fd 245 audit_getname(result);
c4ad8f98 246
c4ad8f98
LT
247 return result;
248}
249
91a27b2a 250void putname(struct filename *name)
1da177e4 251{
55422d0b
PM
252 BUG_ON(name->refcnt <= 0);
253
254 if (--name->refcnt > 0)
255 return;
256
fd2f7cb5 257 if (name->name != name->iname) {
55422d0b
PM
258 __putname(name->name);
259 kfree(name);
260 } else
261 __putname(name);
1da177e4 262}
1da177e4 263
e77819e5
LT
264static int check_acl(struct inode *inode, int mask)
265{
84635d68 266#ifdef CONFIG_FS_POSIX_ACL
e77819e5
LT
267 struct posix_acl *acl;
268
e77819e5 269 if (mask & MAY_NOT_BLOCK) {
3567866b
AV
270 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
271 if (!acl)
e77819e5 272 return -EAGAIN;
3567866b 273 /* no ->get_acl() calls in RCU mode... */
b8a7a3a6 274 if (is_uncached_acl(acl))
3567866b 275 return -ECHILD;
206b1d09 276 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
e77819e5
LT
277 }
278
2982baa2
CH
279 acl = get_acl(inode, ACL_TYPE_ACCESS);
280 if (IS_ERR(acl))
281 return PTR_ERR(acl);
e77819e5
LT
282 if (acl) {
283 int error = posix_acl_permission(inode, acl, mask);
284 posix_acl_release(acl);
285 return error;
286 }
84635d68 287#endif
e77819e5
LT
288
289 return -EAGAIN;
290}
291
5909ccaa 292/*
948409c7 293 * This does the basic permission checking
1da177e4 294 */
7e40145e 295static int acl_permission_check(struct inode *inode, int mask)
1da177e4 296{
26cf46be 297 unsigned int mode = inode->i_mode;
1da177e4 298
8e96e3b7 299 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
1da177e4
LT
300 mode >>= 6;
301 else {
e77819e5 302 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
7e40145e 303 int error = check_acl(inode, mask);
b74c79e9
NP
304 if (error != -EAGAIN)
305 return error;
1da177e4
LT
306 }
307
308 if (in_group_p(inode->i_gid))
309 mode >>= 3;
310 }
311
312 /*
313 * If the DACs are ok we don't need any capability check.
314 */
9c2c7039 315 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
1da177e4 316 return 0;
5909ccaa
LT
317 return -EACCES;
318}
319
320/**
b74c79e9 321 * generic_permission - check for access rights on a Posix-like filesystem
5909ccaa 322 * @inode: inode to check access rights for
8fd90c8d 323 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
5909ccaa
LT
324 *
325 * Used to check for read/write/execute permissions on a file.
326 * We use "fsuid" for this, letting us set arbitrary permissions
327 * for filesystem access without changing the "normal" uids which
b74c79e9
NP
328 * are used for other things.
329 *
330 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
331 * request cannot be satisfied (eg. requires blocking or too much complexity).
332 * It would then be called again in ref-walk mode.
5909ccaa 333 */
2830ba7f 334int generic_permission(struct inode *inode, int mask)
5909ccaa
LT
335{
336 int ret;
337
338 /*
948409c7 339 * Do the basic permission checks.
5909ccaa 340 */
7e40145e 341 ret = acl_permission_check(inode, mask);
5909ccaa
LT
342 if (ret != -EACCES)
343 return ret;
1da177e4 344
d594e7ec
AV
345 if (S_ISDIR(inode->i_mode)) {
346 /* DACs are overridable for directories */
d594e7ec 347 if (!(mask & MAY_WRITE))
23adbe12
AL
348 if (capable_wrt_inode_uidgid(inode,
349 CAP_DAC_READ_SEARCH))
d594e7ec 350 return 0;
23adbe12 351 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
1da177e4 352 return 0;
2a4c2242
SS
353 return -EACCES;
354 }
1da177e4
LT
355
356 /*
357 * Searching includes executable on directories, else just read.
358 */
7ea66001 359 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
d594e7ec 360 if (mask == MAY_READ)
23adbe12 361 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
1da177e4 362 return 0;
2a4c2242
SS
363 /*
364 * Read/write DACs are always overridable.
365 * Executable DACs are overridable when there is
366 * at least one exec bit set.
367 */
368 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
369 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
370 return 0;
1da177e4
LT
371
372 return -EACCES;
373}
4d359507 374EXPORT_SYMBOL(generic_permission);
1da177e4 375
3ddcd056
LT
376/*
377 * We _really_ want to just do "generic_permission()" without
378 * even looking at the inode->i_op values. So we keep a cache
379 * flag in inode->i_opflags, that says "this has not special
380 * permission function, use the fast case".
381 */
382static inline int do_inode_permission(struct inode *inode, int mask)
383{
384 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
385 if (likely(inode->i_op->permission))
386 return inode->i_op->permission(inode, mask);
387
388 /* This gets set once for the inode lifetime */
389 spin_lock(&inode->i_lock);
390 inode->i_opflags |= IOP_FASTPERM;
391 spin_unlock(&inode->i_lock);
392 }
393 return generic_permission(inode, mask);
394}
395
0bdaea90
DH
396/**
397 * sb_permission - Check superblock-level permissions
398 * @sb: Superblock of inode to check permission on
55852635 399 * @inode: Inode to check permission on
0bdaea90
DH
400 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
401 *
402 * Separate out file-system wide checks from inode-specific permission checks.
403 */
404static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
405{
406 if (unlikely(mask & MAY_WRITE)) {
407 umode_t mode = inode->i_mode;
408
409 /* Nobody gets write access to a read-only fs. */
bc98a42c 410 if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
0bdaea90
DH
411 return -EROFS;
412 }
413 return 0;
414}
415
416/**
417 * inode_permission - Check for access rights to a given inode
418 * @inode: Inode to check permission on
419 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
420 *
421 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
422 * this, letting us set arbitrary permissions for filesystem access without
423 * changing the "normal" UIDs which are used for other things.
424 *
425 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
426 */
427int inode_permission(struct inode *inode, int mask)
428{
429 int retval;
430
431 retval = sb_permission(inode->i_sb, inode, mask);
432 if (retval)
433 return retval;
4bfd054a
EB
434
435 if (unlikely(mask & MAY_WRITE)) {
436 /*
437 * Nobody gets write access to an immutable file.
438 */
439 if (IS_IMMUTABLE(inode))
440 return -EPERM;
441
442 /*
443 * Updating mtime will likely cause i_uid and i_gid to be
444 * written back improperly if their true value is unknown
445 * to the vfs.
446 */
447 if (HAS_UNMAPPED_ID(inode))
448 return -EACCES;
449 }
450
451 retval = do_inode_permission(inode, mask);
452 if (retval)
453 return retval;
454
455 retval = devcgroup_inode_permission(inode, mask);
456 if (retval)
457 return retval;
458
459 return security_inode_permission(inode, mask);
0bdaea90 460}
4d359507 461EXPORT_SYMBOL(inode_permission);
0bdaea90 462
5dd784d0
JB
463/**
464 * path_get - get a reference to a path
465 * @path: path to get the reference to
466 *
467 * Given a path increment the reference count to the dentry and the vfsmount.
468 */
dcf787f3 469void path_get(const struct path *path)
5dd784d0
JB
470{
471 mntget(path->mnt);
472 dget(path->dentry);
473}
474EXPORT_SYMBOL(path_get);
475
1d957f9b
JB
476/**
477 * path_put - put a reference to a path
478 * @path: path to put the reference to
479 *
480 * Given a path decrement the reference count to the dentry and the vfsmount.
481 */
dcf787f3 482void path_put(const struct path *path)
1da177e4 483{
1d957f9b
JB
484 dput(path->dentry);
485 mntput(path->mnt);
1da177e4 486}
1d957f9b 487EXPORT_SYMBOL(path_put);
1da177e4 488
894bc8c4 489#define EMBEDDED_LEVELS 2
1f55a6ec
AV
490struct nameidata {
491 struct path path;
1cf2665b 492 struct qstr last;
1f55a6ec
AV
493 struct path root;
494 struct inode *inode; /* path.dentry.d_inode */
495 unsigned int flags;
9883d185 496 unsigned seq, m_seq;
1f55a6ec
AV
497 int last_type;
498 unsigned depth;
756daf26 499 int total_link_count;
697fc6ca
AV
500 struct saved {
501 struct path link;
fceef393 502 struct delayed_call done;
697fc6ca 503 const char *name;
0450b2d1 504 unsigned seq;
894bc8c4 505 } *stack, internal[EMBEDDED_LEVELS];
9883d185
AV
506 struct filename *name;
507 struct nameidata *saved;
fceef393 508 struct inode *link_inode;
9883d185
AV
509 unsigned root_seq;
510 int dfd;
3859a271 511} __randomize_layout;
1f55a6ec 512
9883d185 513static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
894bc8c4 514{
756daf26
N
515 struct nameidata *old = current->nameidata;
516 p->stack = p->internal;
c8a53ee5
AV
517 p->dfd = dfd;
518 p->name = name;
756daf26 519 p->total_link_count = old ? old->total_link_count : 0;
9883d185 520 p->saved = old;
756daf26 521 current->nameidata = p;
894bc8c4
AV
522}
523
9883d185 524static void restore_nameidata(void)
894bc8c4 525{
9883d185 526 struct nameidata *now = current->nameidata, *old = now->saved;
756daf26
N
527
528 current->nameidata = old;
529 if (old)
530 old->total_link_count = now->total_link_count;
e1a63bbc 531 if (now->stack != now->internal)
756daf26 532 kfree(now->stack);
894bc8c4
AV
533}
534
535static int __nd_alloc_stack(struct nameidata *nd)
536{
bc40aee0
AV
537 struct saved *p;
538
539 if (nd->flags & LOOKUP_RCU) {
6da2ec56 540 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
bc40aee0
AV
541 GFP_ATOMIC);
542 if (unlikely(!p))
543 return -ECHILD;
544 } else {
6da2ec56 545 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
894bc8c4 546 GFP_KERNEL);
bc40aee0
AV
547 if (unlikely(!p))
548 return -ENOMEM;
549 }
894bc8c4
AV
550 memcpy(p, nd->internal, sizeof(nd->internal));
551 nd->stack = p;
552 return 0;
553}
554
397d425d
EB
555/**
556 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
557 * @path: nameidate to verify
558 *
559 * Rename can sometimes move a file or directory outside of a bind
560 * mount, path_connected allows those cases to be detected.
561 */
562static bool path_connected(const struct path *path)
563{
564 struct vfsmount *mnt = path->mnt;
95dd7758 565 struct super_block *sb = mnt->mnt_sb;
397d425d 566
95dd7758
EB
567 /* Bind mounts and multi-root filesystems can have disconnected paths */
568 if (!(sb->s_iflags & SB_I_MULTIROOT) && (mnt->mnt_root == sb->s_root))
397d425d
EB
569 return true;
570
571 return is_subdir(path->dentry, mnt->mnt_root);
572}
573
894bc8c4
AV
574static inline int nd_alloc_stack(struct nameidata *nd)
575{
da4e0be0 576 if (likely(nd->depth != EMBEDDED_LEVELS))
894bc8c4
AV
577 return 0;
578 if (likely(nd->stack != nd->internal))
579 return 0;
580 return __nd_alloc_stack(nd);
581}
582
7973387a
AV
583static void drop_links(struct nameidata *nd)
584{
585 int i = nd->depth;
586 while (i--) {
587 struct saved *last = nd->stack + i;
fceef393
AV
588 do_delayed_call(&last->done);
589 clear_delayed_call(&last->done);
7973387a
AV
590 }
591}
592
593static void terminate_walk(struct nameidata *nd)
594{
595 drop_links(nd);
596 if (!(nd->flags & LOOKUP_RCU)) {
597 int i;
598 path_put(&nd->path);
599 for (i = 0; i < nd->depth; i++)
600 path_put(&nd->stack[i].link);
102b8af2
AV
601 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
602 path_put(&nd->root);
603 nd->root.mnt = NULL;
604 }
7973387a
AV
605 } else {
606 nd->flags &= ~LOOKUP_RCU;
607 if (!(nd->flags & LOOKUP_ROOT))
608 nd->root.mnt = NULL;
609 rcu_read_unlock();
610 }
611 nd->depth = 0;
612}
613
614/* path_put is needed afterwards regardless of success or failure */
615static bool legitimize_path(struct nameidata *nd,
616 struct path *path, unsigned seq)
617{
618 int res = __legitimize_mnt(path->mnt, nd->m_seq);
619 if (unlikely(res)) {
620 if (res > 0)
621 path->mnt = NULL;
622 path->dentry = NULL;
623 return false;
624 }
625 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
626 path->dentry = NULL;
627 return false;
628 }
629 return !read_seqcount_retry(&path->dentry->d_seq, seq);
630}
631
632static bool legitimize_links(struct nameidata *nd)
633{
634 int i;
635 for (i = 0; i < nd->depth; i++) {
636 struct saved *last = nd->stack + i;
637 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
638 drop_links(nd);
639 nd->depth = i + 1;
640 return false;
641 }
642 }
643 return true;
644}
645
19660af7 646/*
31e6b01f 647 * Path walking has 2 modes, rcu-walk and ref-walk (see
19660af7
AV
648 * Documentation/filesystems/path-lookup.txt). In situations when we can't
649 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
57e3715c 650 * normal reference counts on dentries and vfsmounts to transition to ref-walk
19660af7
AV
651 * mode. Refcounts are grabbed at the last known good point before rcu-walk
652 * got stuck, so ref-walk may continue from there. If this is not successful
653 * (eg. a seqcount has changed), then failure is returned and it's up to caller
654 * to restart the path walk from the beginning in ref-walk mode.
31e6b01f 655 */
31e6b01f
NP
656
657/**
19660af7
AV
658 * unlazy_walk - try to switch to ref-walk mode.
659 * @nd: nameidata pathwalk data
39191628 660 * Returns: 0 on success, -ECHILD on failure
31e6b01f 661 *
4675ac39
AV
662 * unlazy_walk attempts to legitimize the current nd->path and nd->root
663 * for ref-walk mode.
664 * Must be called from rcu-walk context.
7973387a
AV
665 * Nothing should touch nameidata between unlazy_walk() failure and
666 * terminate_walk().
31e6b01f 667 */
4675ac39 668static int unlazy_walk(struct nameidata *nd)
31e6b01f 669{
31e6b01f
NP
670 struct dentry *parent = nd->path.dentry;
671
672 BUG_ON(!(nd->flags & LOOKUP_RCU));
e5c832d5 673
4675ac39
AV
674 nd->flags &= ~LOOKUP_RCU;
675 if (unlikely(!legitimize_links(nd)))
676 goto out2;
677 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
678 goto out1;
679 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
680 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq)))
681 goto out;
682 }
683 rcu_read_unlock();
684 BUG_ON(nd->inode != parent->d_inode);
685 return 0;
686
687out2:
688 nd->path.mnt = NULL;
689 nd->path.dentry = NULL;
690out1:
691 if (!(nd->flags & LOOKUP_ROOT))
692 nd->root.mnt = NULL;
693out:
694 rcu_read_unlock();
695 return -ECHILD;
696}
697
698/**
699 * unlazy_child - try to switch to ref-walk mode.
700 * @nd: nameidata pathwalk data
701 * @dentry: child of nd->path.dentry
702 * @seq: seq number to check dentry against
703 * Returns: 0 on success, -ECHILD on failure
704 *
705 * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry
706 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
707 * @nd. Must be called from rcu-walk context.
708 * Nothing should touch nameidata between unlazy_child() failure and
709 * terminate_walk().
710 */
711static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq)
712{
713 BUG_ON(!(nd->flags & LOOKUP_RCU));
714
e5c832d5 715 nd->flags &= ~LOOKUP_RCU;
7973387a
AV
716 if (unlikely(!legitimize_links(nd)))
717 goto out2;
718 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
719 goto out2;
4675ac39 720 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
7973387a 721 goto out1;
48a066e7 722
15570086 723 /*
4675ac39
AV
724 * We need to move both the parent and the dentry from the RCU domain
725 * to be properly refcounted. And the sequence number in the dentry
726 * validates *both* dentry counters, since we checked the sequence
727 * number of the parent after we got the child sequence number. So we
728 * know the parent must still be valid if the child sequence number is
15570086 729 */
4675ac39
AV
730 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
731 goto out;
732 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) {
733 rcu_read_unlock();
734 dput(dentry);
735 goto drop_root_mnt;
19660af7 736 }
e5c832d5
LT
737 /*
738 * Sequence counts matched. Now make sure that the root is
739 * still valid and get it if required.
740 */
741 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
5a8d87e8
AV
742 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
743 rcu_read_unlock();
744 dput(dentry);
745 return -ECHILD;
7973387a 746 }
31e6b01f 747 }
31e6b01f 748
8b61e74f 749 rcu_read_unlock();
31e6b01f 750 return 0;
19660af7 751
7973387a
AV
752out2:
753 nd->path.mnt = NULL;
754out1:
755 nd->path.dentry = NULL;
e5c832d5 756out:
8b61e74f 757 rcu_read_unlock();
d0d27277
LT
758drop_root_mnt:
759 if (!(nd->flags & LOOKUP_ROOT))
760 nd->root.mnt = NULL;
31e6b01f
NP
761 return -ECHILD;
762}
763
4ce16ef3 764static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
34286d66 765{
a89f8337
AV
766 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
767 return dentry->d_op->d_revalidate(dentry, flags);
768 else
769 return 1;
34286d66
NP
770}
771
9f1fafee
AV
772/**
773 * complete_walk - successful completion of path walk
774 * @nd: pointer nameidata
39159de2 775 *
9f1fafee
AV
776 * If we had been in RCU mode, drop out of it and legitimize nd->path.
777 * Revalidate the final result, unless we'd already done that during
778 * the path walk or the filesystem doesn't ask for it. Return 0 on
779 * success, -error on failure. In case of failure caller does not
780 * need to drop nd->path.
39159de2 781 */
9f1fafee 782static int complete_walk(struct nameidata *nd)
39159de2 783{
16c2cd71 784 struct dentry *dentry = nd->path.dentry;
39159de2 785 int status;
39159de2 786
9f1fafee 787 if (nd->flags & LOOKUP_RCU) {
9f1fafee
AV
788 if (!(nd->flags & LOOKUP_ROOT))
789 nd->root.mnt = NULL;
4675ac39 790 if (unlikely(unlazy_walk(nd)))
9f1fafee 791 return -ECHILD;
9f1fafee
AV
792 }
793
16c2cd71
AV
794 if (likely(!(nd->flags & LOOKUP_JUMPED)))
795 return 0;
796
ecf3d1f1 797 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
39159de2
JL
798 return 0;
799
ecf3d1f1 800 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
39159de2
JL
801 if (status > 0)
802 return 0;
803
16c2cd71 804 if (!status)
39159de2 805 status = -ESTALE;
16c2cd71 806
39159de2
JL
807 return status;
808}
809
18d8c860 810static void set_root(struct nameidata *nd)
31e6b01f 811{
7bd88377 812 struct fs_struct *fs = current->fs;
c28cc364 813
9e6697e2
AV
814 if (nd->flags & LOOKUP_RCU) {
815 unsigned seq;
816
817 do {
818 seq = read_seqcount_begin(&fs->seq);
819 nd->root = fs->root;
820 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
821 } while (read_seqcount_retry(&fs->seq, seq));
822 } else {
823 get_fs_root(fs, &nd->root);
824 }
31e6b01f
NP
825}
826
1d957f9b 827static void path_put_conditional(struct path *path, struct nameidata *nd)
051d3812
IK
828{
829 dput(path->dentry);
4ac91378 830 if (path->mnt != nd->path.mnt)
051d3812
IK
831 mntput(path->mnt);
832}
833
7b9337aa
NP
834static inline void path_to_nameidata(const struct path *path,
835 struct nameidata *nd)
051d3812 836{
31e6b01f
NP
837 if (!(nd->flags & LOOKUP_RCU)) {
838 dput(nd->path.dentry);
839 if (nd->path.mnt != path->mnt)
840 mntput(nd->path.mnt);
9a229683 841 }
31e6b01f 842 nd->path.mnt = path->mnt;
4ac91378 843 nd->path.dentry = path->dentry;
051d3812
IK
844}
845
248fb5b9
AV
846static int nd_jump_root(struct nameidata *nd)
847{
848 if (nd->flags & LOOKUP_RCU) {
849 struct dentry *d;
850 nd->path = nd->root;
851 d = nd->path.dentry;
852 nd->inode = d->d_inode;
853 nd->seq = nd->root_seq;
854 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
855 return -ECHILD;
856 } else {
857 path_put(&nd->path);
858 nd->path = nd->root;
859 path_get(&nd->path);
860 nd->inode = nd->path.dentry->d_inode;
861 }
862 nd->flags |= LOOKUP_JUMPED;
863 return 0;
864}
865
b5fb63c1 866/*
6b255391 867 * Helper to directly jump to a known parsed path from ->get_link,
b5fb63c1
CH
868 * caller must have taken a reference to path beforehand.
869 */
6e77137b 870void nd_jump_link(struct path *path)
b5fb63c1 871{
6e77137b 872 struct nameidata *nd = current->nameidata;
b5fb63c1
CH
873 path_put(&nd->path);
874
875 nd->path = *path;
876 nd->inode = nd->path.dentry->d_inode;
877 nd->flags |= LOOKUP_JUMPED;
b5fb63c1
CH
878}
879
b9ff4429 880static inline void put_link(struct nameidata *nd)
574197e0 881{
21c3003d 882 struct saved *last = nd->stack + --nd->depth;
fceef393 883 do_delayed_call(&last->done);
6548fae2
AV
884 if (!(nd->flags & LOOKUP_RCU))
885 path_put(&last->link);
574197e0
AV
886}
887
561ec64a
LT
888int sysctl_protected_symlinks __read_mostly = 0;
889int sysctl_protected_hardlinks __read_mostly = 0;
800179c9
KC
890
891/**
892 * may_follow_link - Check symlink following for unsafe situations
55852635 893 * @nd: nameidata pathwalk data
800179c9
KC
894 *
895 * In the case of the sysctl_protected_symlinks sysctl being enabled,
896 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
897 * in a sticky world-writable directory. This is to protect privileged
898 * processes from failing races against path names that may change out
899 * from under them by way of other users creating malicious symlinks.
900 * It will permit symlinks to be followed only when outside a sticky
901 * world-writable directory, or when the uid of the symlink and follower
902 * match, or when the directory owner matches the symlink's owner.
903 *
904 * Returns 0 if following the symlink is allowed, -ve on error.
905 */
fec2fa24 906static inline int may_follow_link(struct nameidata *nd)
800179c9
KC
907{
908 const struct inode *inode;
909 const struct inode *parent;
2d7f9e2a 910 kuid_t puid;
800179c9
KC
911
912 if (!sysctl_protected_symlinks)
913 return 0;
914
915 /* Allowed if owner and follower match. */
fceef393 916 inode = nd->link_inode;
81abe27b 917 if (uid_eq(current_cred()->fsuid, inode->i_uid))
800179c9
KC
918 return 0;
919
920 /* Allowed if parent directory not sticky and world-writable. */
aa65fa35 921 parent = nd->inode;
800179c9
KC
922 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
923 return 0;
924
925 /* Allowed if parent directory and link owner match. */
2d7f9e2a
SF
926 puid = parent->i_uid;
927 if (uid_valid(puid) && uid_eq(puid, inode->i_uid))
800179c9
KC
928 return 0;
929
31956502
AV
930 if (nd->flags & LOOKUP_RCU)
931 return -ECHILD;
932
ea841baf 933 audit_inode(nd->name, nd->stack[0].link.dentry, 0);
94b9d9b7 934 audit_log_link_denied("follow_link");
800179c9
KC
935 return -EACCES;
936}
937
938/**
939 * safe_hardlink_source - Check for safe hardlink conditions
940 * @inode: the source inode to hardlink from
941 *
942 * Return false if at least one of the following conditions:
943 * - inode is not a regular file
944 * - inode is setuid
945 * - inode is setgid and group-exec
946 * - access failure for read and write
947 *
948 * Otherwise returns true.
949 */
950static bool safe_hardlink_source(struct inode *inode)
951{
952 umode_t mode = inode->i_mode;
953
954 /* Special files should not get pinned to the filesystem. */
955 if (!S_ISREG(mode))
956 return false;
957
958 /* Setuid files should not get pinned to the filesystem. */
959 if (mode & S_ISUID)
960 return false;
961
962 /* Executable setgid files should not get pinned to the filesystem. */
963 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
964 return false;
965
966 /* Hardlinking to unreadable or unwritable sources is dangerous. */
967 if (inode_permission(inode, MAY_READ | MAY_WRITE))
968 return false;
969
970 return true;
971}
972
973/**
974 * may_linkat - Check permissions for creating a hardlink
975 * @link: the source to hardlink from
976 *
977 * Block hardlink when all of:
978 * - sysctl_protected_hardlinks enabled
979 * - fsuid does not match inode
980 * - hardlink source is unsafe (see safe_hardlink_source() above)
f2ca3796 981 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
800179c9
KC
982 *
983 * Returns 0 if successful, -ve on error.
984 */
985static int may_linkat(struct path *link)
986{
593d1ce8
EB
987 struct inode *inode = link->dentry->d_inode;
988
989 /* Inode writeback is not safe when the uid or gid are invalid. */
990 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
991 return -EOVERFLOW;
800179c9
KC
992
993 if (!sysctl_protected_hardlinks)
994 return 0;
995
800179c9
KC
996 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
997 * otherwise, it must be a safe source.
998 */
cc658db4 999 if (safe_hardlink_source(inode) || inode_owner_or_capable(inode))
800179c9
KC
1000 return 0;
1001
94b9d9b7 1002 audit_log_link_denied("linkat");
800179c9
KC
1003 return -EPERM;
1004}
1005
3b2e7f75
AV
1006static __always_inline
1007const char *get_link(struct nameidata *nd)
1da177e4 1008{
ab104923 1009 struct saved *last = nd->stack + nd->depth - 1;
1cf2665b 1010 struct dentry *dentry = last->link.dentry;
fceef393 1011 struct inode *inode = nd->link_inode;
6d7b5aae 1012 int error;
0a959df5 1013 const char *res;
1da177e4 1014
8fa9dd24
N
1015 if (!(nd->flags & LOOKUP_RCU)) {
1016 touch_atime(&last->link);
1017 cond_resched();
598e3c8f 1018 } else if (atime_needs_update_rcu(&last->link, inode)) {
4675ac39 1019 if (unlikely(unlazy_walk(nd)))
bc40aee0 1020 return ERR_PTR(-ECHILD);
8fa9dd24 1021 touch_atime(&last->link);
bc40aee0 1022 }
574197e0 1023
bda0be7a
N
1024 error = security_inode_follow_link(dentry, inode,
1025 nd->flags & LOOKUP_RCU);
1026 if (unlikely(error))
6920a440 1027 return ERR_PTR(error);
36f3b4f6 1028
86acdca1 1029 nd->last_type = LAST_BIND;
d4dee48b
AV
1030 res = inode->i_link;
1031 if (!res) {
fceef393
AV
1032 const char * (*get)(struct dentry *, struct inode *,
1033 struct delayed_call *);
1034 get = inode->i_op->get_link;
8c1b4566 1035 if (nd->flags & LOOKUP_RCU) {
fceef393 1036 res = get(NULL, inode, &last->done);
6b255391 1037 if (res == ERR_PTR(-ECHILD)) {
4675ac39 1038 if (unlikely(unlazy_walk(nd)))
6b255391 1039 return ERR_PTR(-ECHILD);
fceef393 1040 res = get(dentry, inode, &last->done);
6b255391
AV
1041 }
1042 } else {
fceef393 1043 res = get(dentry, inode, &last->done);
8c1b4566 1044 }
fceef393 1045 if (IS_ERR_OR_NULL(res))
fab51e8a 1046 return res;
fab51e8a
AV
1047 }
1048 if (*res == '/') {
9e6697e2
AV
1049 if (!nd->root.mnt)
1050 set_root(nd);
248fb5b9
AV
1051 if (unlikely(nd_jump_root(nd)))
1052 return ERR_PTR(-ECHILD);
fab51e8a
AV
1053 while (unlikely(*++res == '/'))
1054 ;
1da177e4 1055 }
fab51e8a
AV
1056 if (!*res)
1057 res = NULL;
0a959df5
AV
1058 return res;
1059}
6d7b5aae 1060
f015f126
DH
1061/*
1062 * follow_up - Find the mountpoint of path's vfsmount
1063 *
1064 * Given a path, find the mountpoint of its source file system.
1065 * Replace @path with the path of the mountpoint in the parent mount.
1066 * Up is towards /.
1067 *
1068 * Return 1 if we went up a level and 0 if we were already at the
1069 * root.
1070 */
bab77ebf 1071int follow_up(struct path *path)
1da177e4 1072{
0714a533
AV
1073 struct mount *mnt = real_mount(path->mnt);
1074 struct mount *parent;
1da177e4 1075 struct dentry *mountpoint;
99b7db7b 1076
48a066e7 1077 read_seqlock_excl(&mount_lock);
0714a533 1078 parent = mnt->mnt_parent;
3c0a6163 1079 if (parent == mnt) {
48a066e7 1080 read_sequnlock_excl(&mount_lock);
1da177e4
LT
1081 return 0;
1082 }
0714a533 1083 mntget(&parent->mnt);
a73324da 1084 mountpoint = dget(mnt->mnt_mountpoint);
48a066e7 1085 read_sequnlock_excl(&mount_lock);
bab77ebf
AV
1086 dput(path->dentry);
1087 path->dentry = mountpoint;
1088 mntput(path->mnt);
0714a533 1089 path->mnt = &parent->mnt;
1da177e4
LT
1090 return 1;
1091}
4d359507 1092EXPORT_SYMBOL(follow_up);
1da177e4 1093
b5c84bf6 1094/*
9875cf80
DH
1095 * Perform an automount
1096 * - return -EISDIR to tell follow_managed() to stop and return the path we
1097 * were called with.
1da177e4 1098 */
756daf26 1099static int follow_automount(struct path *path, struct nameidata *nd,
9875cf80 1100 bool *need_mntput)
31e6b01f 1101{
9875cf80 1102 struct vfsmount *mnt;
ea5b778a 1103 int err;
9875cf80
DH
1104
1105 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1106 return -EREMOTE;
1107
0ec26fd0
MS
1108 /* We don't want to mount if someone's just doing a stat -
1109 * unless they're stat'ing a directory and appended a '/' to
1110 * the name.
1111 *
1112 * We do, however, want to mount if someone wants to open or
1113 * create a file of any type under the mountpoint, wants to
1114 * traverse through the mountpoint or wants to open the
1115 * mounted directory. Also, autofs may mark negative dentries
1116 * as being automount points. These will need the attentions
1117 * of the daemon to instantiate them before they can be used.
9875cf80 1118 */
756daf26 1119 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
5d38f049
IK
1120 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1121 path->dentry->d_inode)
1122 return -EISDIR;
0ec26fd0 1123
756daf26
N
1124 nd->total_link_count++;
1125 if (nd->total_link_count >= 40)
9875cf80
DH
1126 return -ELOOP;
1127
1128 mnt = path->dentry->d_op->d_automount(path);
1129 if (IS_ERR(mnt)) {
1130 /*
1131 * The filesystem is allowed to return -EISDIR here to indicate
1132 * it doesn't want to automount. For instance, autofs would do
1133 * this so that its userspace daemon can mount on this dentry.
1134 *
1135 * However, we can only permit this if it's a terminal point in
1136 * the path being looked up; if it wasn't then the remainder of
1137 * the path is inaccessible and we should say so.
1138 */
756daf26 1139 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
9875cf80
DH
1140 return -EREMOTE;
1141 return PTR_ERR(mnt);
31e6b01f 1142 }
ea5b778a 1143
9875cf80
DH
1144 if (!mnt) /* mount collision */
1145 return 0;
31e6b01f 1146
8aef1884
AV
1147 if (!*need_mntput) {
1148 /* lock_mount() may release path->mnt on error */
1149 mntget(path->mnt);
1150 *need_mntput = true;
1151 }
19a167af 1152 err = finish_automount(mnt, path);
9875cf80 1153
ea5b778a
DH
1154 switch (err) {
1155 case -EBUSY:
1156 /* Someone else made a mount here whilst we were busy */
19a167af 1157 return 0;
ea5b778a 1158 case 0:
8aef1884 1159 path_put(path);
ea5b778a
DH
1160 path->mnt = mnt;
1161 path->dentry = dget(mnt->mnt_root);
ea5b778a 1162 return 0;
19a167af
AV
1163 default:
1164 return err;
ea5b778a 1165 }
19a167af 1166
463ffb2e
AV
1167}
1168
9875cf80
DH
1169/*
1170 * Handle a dentry that is managed in some way.
cc53ce53 1171 * - Flagged for transit management (autofs)
9875cf80
DH
1172 * - Flagged as mountpoint
1173 * - Flagged as automount point
1174 *
1175 * This may only be called in refwalk mode.
1176 *
1177 * Serialization is taken care of in namespace.c
1178 */
756daf26 1179static int follow_managed(struct path *path, struct nameidata *nd)
1da177e4 1180{
8aef1884 1181 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
9875cf80
DH
1182 unsigned managed;
1183 bool need_mntput = false;
8aef1884 1184 int ret = 0;
9875cf80
DH
1185
1186 /* Given that we're not holding a lock here, we retain the value in a
1187 * local variable for each dentry as we look at it so that we don't see
1188 * the components of that value change under us */
6aa7de05 1189 while (managed = READ_ONCE(path->dentry->d_flags),
9875cf80
DH
1190 managed &= DCACHE_MANAGED_DENTRY,
1191 unlikely(managed != 0)) {
cc53ce53
DH
1192 /* Allow the filesystem to manage the transit without i_mutex
1193 * being held. */
1194 if (managed & DCACHE_MANAGE_TRANSIT) {
1195 BUG_ON(!path->dentry->d_op);
1196 BUG_ON(!path->dentry->d_op->d_manage);
fb5f51c7 1197 ret = path->dentry->d_op->d_manage(path, false);
cc53ce53 1198 if (ret < 0)
8aef1884 1199 break;
cc53ce53
DH
1200 }
1201
9875cf80
DH
1202 /* Transit to a mounted filesystem. */
1203 if (managed & DCACHE_MOUNTED) {
1204 struct vfsmount *mounted = lookup_mnt(path);
1205 if (mounted) {
1206 dput(path->dentry);
1207 if (need_mntput)
1208 mntput(path->mnt);
1209 path->mnt = mounted;
1210 path->dentry = dget(mounted->mnt_root);
1211 need_mntput = true;
1212 continue;
1213 }
1214
1215 /* Something is mounted on this dentry in another
1216 * namespace and/or whatever was mounted there in this
48a066e7
AV
1217 * namespace got unmounted before lookup_mnt() could
1218 * get it */
9875cf80
DH
1219 }
1220
1221 /* Handle an automount point */
1222 if (managed & DCACHE_NEED_AUTOMOUNT) {
756daf26 1223 ret = follow_automount(path, nd, &need_mntput);
9875cf80 1224 if (ret < 0)
8aef1884 1225 break;
9875cf80
DH
1226 continue;
1227 }
1228
1229 /* We didn't change the current path point */
1230 break;
1da177e4 1231 }
8aef1884
AV
1232
1233 if (need_mntput && path->mnt == mnt)
1234 mntput(path->mnt);
e9742b53
AV
1235 if (ret == -EISDIR || !ret)
1236 ret = 1;
8402752e
AV
1237 if (need_mntput)
1238 nd->flags |= LOOKUP_JUMPED;
1239 if (unlikely(ret < 0))
1240 path_put_conditional(path, nd);
1241 return ret;
1da177e4
LT
1242}
1243
cc53ce53 1244int follow_down_one(struct path *path)
1da177e4
LT
1245{
1246 struct vfsmount *mounted;
1247
1c755af4 1248 mounted = lookup_mnt(path);
1da177e4 1249 if (mounted) {
9393bd07
AV
1250 dput(path->dentry);
1251 mntput(path->mnt);
1252 path->mnt = mounted;
1253 path->dentry = dget(mounted->mnt_root);
1da177e4
LT
1254 return 1;
1255 }
1256 return 0;
1257}
4d359507 1258EXPORT_SYMBOL(follow_down_one);
1da177e4 1259
fb5f51c7 1260static inline int managed_dentry_rcu(const struct path *path)
62a7375e 1261{
fb5f51c7
IK
1262 return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1263 path->dentry->d_op->d_manage(path, true) : 0;
62a7375e
IK
1264}
1265
9875cf80 1266/*
287548e4
AV
1267 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1268 * we meet a managed dentry that would need blocking.
9875cf80
DH
1269 */
1270static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
254cf582 1271 struct inode **inode, unsigned *seqp)
9875cf80 1272{
62a7375e 1273 for (;;) {
c7105365 1274 struct mount *mounted;
62a7375e
IK
1275 /*
1276 * Don't forget we might have a non-mountpoint managed dentry
1277 * that wants to block transit.
1278 */
fb5f51c7 1279 switch (managed_dentry_rcu(path)) {
b8faf035
N
1280 case -ECHILD:
1281 default:
ab90911f 1282 return false;
b8faf035
N
1283 case -EISDIR:
1284 return true;
1285 case 0:
1286 break;
1287 }
62a7375e
IK
1288
1289 if (!d_mountpoint(path->dentry))
b8faf035 1290 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
62a7375e 1291
474279dc 1292 mounted = __lookup_mnt(path->mnt, path->dentry);
9875cf80
DH
1293 if (!mounted)
1294 break;
c7105365
AV
1295 path->mnt = &mounted->mnt;
1296 path->dentry = mounted->mnt.mnt_root;
a3fbbde7 1297 nd->flags |= LOOKUP_JUMPED;
254cf582 1298 *seqp = read_seqcount_begin(&path->dentry->d_seq);
59430262
LT
1299 /*
1300 * Update the inode too. We don't need to re-check the
1301 * dentry sequence number here after this d_inode read,
1302 * because a mount-point is always pinned.
1303 */
1304 *inode = path->dentry->d_inode;
9875cf80 1305 }
f5be3e29 1306 return !read_seqretry(&mount_lock, nd->m_seq) &&
b8faf035 1307 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
287548e4
AV
1308}
1309
31e6b01f
NP
1310static int follow_dotdot_rcu(struct nameidata *nd)
1311{
4023bfc9 1312 struct inode *inode = nd->inode;
31e6b01f 1313
9875cf80 1314 while (1) {
aed434ad 1315 if (path_equal(&nd->path, &nd->root))
31e6b01f 1316 break;
31e6b01f
NP
1317 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1318 struct dentry *old = nd->path.dentry;
1319 struct dentry *parent = old->d_parent;
1320 unsigned seq;
1321
4023bfc9 1322 inode = parent->d_inode;
31e6b01f 1323 seq = read_seqcount_begin(&parent->d_seq);
aed434ad
AV
1324 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1325 return -ECHILD;
31e6b01f
NP
1326 nd->path.dentry = parent;
1327 nd->seq = seq;
397d425d
EB
1328 if (unlikely(!path_connected(&nd->path)))
1329 return -ENOENT;
31e6b01f 1330 break;
aed434ad
AV
1331 } else {
1332 struct mount *mnt = real_mount(nd->path.mnt);
1333 struct mount *mparent = mnt->mnt_parent;
1334 struct dentry *mountpoint = mnt->mnt_mountpoint;
1335 struct inode *inode2 = mountpoint->d_inode;
1336 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1337 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1338 return -ECHILD;
1339 if (&mparent->mnt == nd->path.mnt)
1340 break;
1341 /* we know that mountpoint was pinned */
1342 nd->path.dentry = mountpoint;
1343 nd->path.mnt = &mparent->mnt;
1344 inode = inode2;
1345 nd->seq = seq;
31e6b01f 1346 }
31e6b01f 1347 }
aed434ad 1348 while (unlikely(d_mountpoint(nd->path.dentry))) {
b37199e6
AV
1349 struct mount *mounted;
1350 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
aed434ad
AV
1351 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1352 return -ECHILD;
b37199e6
AV
1353 if (!mounted)
1354 break;
1355 nd->path.mnt = &mounted->mnt;
1356 nd->path.dentry = mounted->mnt.mnt_root;
4023bfc9 1357 inode = nd->path.dentry->d_inode;
b37199e6 1358 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
b37199e6 1359 }
4023bfc9 1360 nd->inode = inode;
31e6b01f
NP
1361 return 0;
1362}
1363
cc53ce53
DH
1364/*
1365 * Follow down to the covering mount currently visible to userspace. At each
1366 * point, the filesystem owning that dentry may be queried as to whether the
1367 * caller is permitted to proceed or not.
cc53ce53 1368 */
7cc90cc3 1369int follow_down(struct path *path)
cc53ce53
DH
1370{
1371 unsigned managed;
1372 int ret;
1373
6aa7de05 1374 while (managed = READ_ONCE(path->dentry->d_flags),
cc53ce53
DH
1375 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1376 /* Allow the filesystem to manage the transit without i_mutex
1377 * being held.
1378 *
1379 * We indicate to the filesystem if someone is trying to mount
1380 * something here. This gives autofs the chance to deny anyone
1381 * other than its daemon the right to mount on its
1382 * superstructure.
1383 *
1384 * The filesystem may sleep at this point.
1385 */
1386 if (managed & DCACHE_MANAGE_TRANSIT) {
1387 BUG_ON(!path->dentry->d_op);
1388 BUG_ON(!path->dentry->d_op->d_manage);
fb5f51c7 1389 ret = path->dentry->d_op->d_manage(path, false);
cc53ce53
DH
1390 if (ret < 0)
1391 return ret == -EISDIR ? 0 : ret;
1392 }
1393
1394 /* Transit to a mounted filesystem. */
1395 if (managed & DCACHE_MOUNTED) {
1396 struct vfsmount *mounted = lookup_mnt(path);
1397 if (!mounted)
1398 break;
1399 dput(path->dentry);
1400 mntput(path->mnt);
1401 path->mnt = mounted;
1402 path->dentry = dget(mounted->mnt_root);
1403 continue;
1404 }
1405
1406 /* Don't handle automount points here */
1407 break;
1408 }
1409 return 0;
1410}
4d359507 1411EXPORT_SYMBOL(follow_down);
cc53ce53 1412
9875cf80
DH
1413/*
1414 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1415 */
1416static void follow_mount(struct path *path)
1417{
1418 while (d_mountpoint(path->dentry)) {
1419 struct vfsmount *mounted = lookup_mnt(path);
1420 if (!mounted)
1421 break;
1422 dput(path->dentry);
1423 mntput(path->mnt);
1424 path->mnt = mounted;
1425 path->dentry = dget(mounted->mnt_root);
1426 }
1427}
1428
eedf265a
EB
1429static int path_parent_directory(struct path *path)
1430{
1431 struct dentry *old = path->dentry;
1432 /* rare case of legitimate dget_parent()... */
1433 path->dentry = dget_parent(path->dentry);
1434 dput(old);
1435 if (unlikely(!path_connected(path)))
1436 return -ENOENT;
1437 return 0;
1438}
1439
397d425d 1440static int follow_dotdot(struct nameidata *nd)
1da177e4
LT
1441{
1442 while(1) {
030c7e0b 1443 if (path_equal(&nd->path, &nd->root))
1da177e4 1444 break;
4ac91378 1445 if (nd->path.dentry != nd->path.mnt->mnt_root) {
eedf265a
EB
1446 int ret = path_parent_directory(&nd->path);
1447 if (ret)
1448 return ret;
1da177e4
LT
1449 break;
1450 }
3088dd70 1451 if (!follow_up(&nd->path))
1da177e4 1452 break;
1da177e4 1453 }
79ed0226 1454 follow_mount(&nd->path);
31e6b01f 1455 nd->inode = nd->path.dentry->d_inode;
397d425d 1456 return 0;
1da177e4
LT
1457}
1458
baa03890 1459/*
f4fdace9
OD
1460 * This looks up the name in dcache and possibly revalidates the found dentry.
1461 * NULL is returned if the dentry does not exist in the cache.
baa03890 1462 */
e3c13928
AV
1463static struct dentry *lookup_dcache(const struct qstr *name,
1464 struct dentry *dir,
6c51e513 1465 unsigned int flags)
baa03890 1466{
a89f8337 1467 struct dentry *dentry = d_lookup(dir, name);
bad61189 1468 if (dentry) {
a89f8337
AV
1469 int error = d_revalidate(dentry, flags);
1470 if (unlikely(error <= 0)) {
1471 if (!error)
1472 d_invalidate(dentry);
1473 dput(dentry);
1474 return ERR_PTR(error);
bad61189
MS
1475 }
1476 }
baa03890
NP
1477 return dentry;
1478}
1479
44396f4b 1480/*
a03ece5f
AV
1481 * Parent directory has inode locked exclusive. This is one
1482 * and only case when ->lookup() gets called on non in-lookup
1483 * dentries - as the matter of fact, this only gets called
1484 * when directory is guaranteed to have no in-lookup children
1485 * at all.
44396f4b 1486 */
e3c13928 1487static struct dentry *__lookup_hash(const struct qstr *name,
72bd866a 1488 struct dentry *base, unsigned int flags)
a3255546 1489{
6c51e513 1490 struct dentry *dentry = lookup_dcache(name, base, flags);
a03ece5f
AV
1491 struct dentry *old;
1492 struct inode *dir = base->d_inode;
a3255546 1493
6c51e513 1494 if (dentry)
bad61189 1495 return dentry;
a3255546 1496
a03ece5f
AV
1497 /* Don't create child dentry for a dead directory. */
1498 if (unlikely(IS_DEADDIR(dir)))
1499 return ERR_PTR(-ENOENT);
1500
6c51e513
AV
1501 dentry = d_alloc(base, name);
1502 if (unlikely(!dentry))
1503 return ERR_PTR(-ENOMEM);
1504
a03ece5f
AV
1505 old = dir->i_op->lookup(dir, dentry, flags);
1506 if (unlikely(old)) {
1507 dput(dentry);
1508 dentry = old;
1509 }
1510 return dentry;
a3255546
AV
1511}
1512
e97cdc87 1513static int lookup_fast(struct nameidata *nd,
254cf582
AV
1514 struct path *path, struct inode **inode,
1515 unsigned *seqp)
1da177e4 1516{
4ac91378 1517 struct vfsmount *mnt = nd->path.mnt;
31e6b01f 1518 struct dentry *dentry, *parent = nd->path.dentry;
5a18fff2 1519 int status = 1;
9875cf80
DH
1520 int err;
1521
b04f784e
NP
1522 /*
1523 * Rename seqlock is not required here because in the off chance
5d0f49c1
AV
1524 * of a false negative due to a concurrent rename, the caller is
1525 * going to fall back to non-racy lookup.
b04f784e 1526 */
31e6b01f
NP
1527 if (nd->flags & LOOKUP_RCU) {
1528 unsigned seq;
766c4cbf 1529 bool negative;
da53be12 1530 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
5d0f49c1 1531 if (unlikely(!dentry)) {
4675ac39 1532 if (unlazy_walk(nd))
5d0f49c1 1533 return -ECHILD;
e9742b53 1534 return 0;
5d0f49c1 1535 }
5a18fff2 1536
12f8ad4b
LT
1537 /*
1538 * This sequence count validates that the inode matches
1539 * the dentry name information from lookup.
1540 */
63afdfc7 1541 *inode = d_backing_inode(dentry);
766c4cbf 1542 negative = d_is_negative(dentry);
5d0f49c1 1543 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
12f8ad4b
LT
1544 return -ECHILD;
1545
1546 /*
1547 * This sequence count validates that the parent had no
1548 * changes while we did the lookup of the dentry above.
1549 *
1550 * The memory barrier in read_seqcount_begin of child is
1551 * enough, we can use __read_seqcount_retry here.
1552 */
5d0f49c1 1553 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
31e6b01f 1554 return -ECHILD;
5a18fff2 1555
254cf582 1556 *seqp = seq;
a89f8337 1557 status = d_revalidate(dentry, nd->flags);
209a7fb2 1558 if (likely(status > 0)) {
5d0f49c1
AV
1559 /*
1560 * Note: do negative dentry check after revalidation in
1561 * case that drops it.
1562 */
1563 if (unlikely(negative))
1564 return -ENOENT;
1565 path->mnt = mnt;
1566 path->dentry = dentry;
1567 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
e9742b53 1568 return 1;
24643087 1569 }
4675ac39 1570 if (unlazy_child(nd, dentry, seq))
209a7fb2
AV
1571 return -ECHILD;
1572 if (unlikely(status == -ECHILD))
1573 /* we'd been told to redo it in non-rcu mode */
1574 status = d_revalidate(dentry, nd->flags);
5a18fff2 1575 } else {
e97cdc87 1576 dentry = __d_lookup(parent, &nd->last);
5d0f49c1 1577 if (unlikely(!dentry))
e9742b53 1578 return 0;
a89f8337 1579 status = d_revalidate(dentry, nd->flags);
9875cf80 1580 }
5a18fff2 1581 if (unlikely(status <= 0)) {
e9742b53 1582 if (!status)
5d0f49c1 1583 d_invalidate(dentry);
5542aa2f 1584 dput(dentry);
5d0f49c1 1585 return status;
24643087 1586 }
766c4cbf
AV
1587 if (unlikely(d_is_negative(dentry))) {
1588 dput(dentry);
1589 return -ENOENT;
1590 }
5d0f49c1 1591
9875cf80
DH
1592 path->mnt = mnt;
1593 path->dentry = dentry;
756daf26 1594 err = follow_managed(path, nd);
e9742b53 1595 if (likely(err > 0))
63afdfc7 1596 *inode = d_backing_inode(path->dentry);
8402752e 1597 return err;
697f514d
MS
1598}
1599
1600/* Fast lookup failed, do it the slow way */
88d8331a
AV
1601static struct dentry *__lookup_slow(const struct qstr *name,
1602 struct dentry *dir,
1603 unsigned int flags)
697f514d 1604{
88d8331a 1605 struct dentry *dentry, *old;
1936386e 1606 struct inode *inode = dir->d_inode;
d9171b93 1607 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1936386e 1608
1936386e 1609 /* Don't go there if it's already dead */
94bdd655 1610 if (unlikely(IS_DEADDIR(inode)))
88d8331a 1611 return ERR_PTR(-ENOENT);
94bdd655 1612again:
d9171b93 1613 dentry = d_alloc_parallel(dir, name, &wq);
94bdd655 1614 if (IS_ERR(dentry))
88d8331a 1615 return dentry;
94bdd655 1616 if (unlikely(!d_in_lookup(dentry))) {
a89f8337 1617 if (!(flags & LOOKUP_NO_REVAL)) {
949a852e
AV
1618 int error = d_revalidate(dentry, flags);
1619 if (unlikely(error <= 0)) {
94bdd655 1620 if (!error) {
949a852e 1621 d_invalidate(dentry);
94bdd655
AV
1622 dput(dentry);
1623 goto again;
1624 }
949a852e
AV
1625 dput(dentry);
1626 dentry = ERR_PTR(error);
1627 }
1628 }
94bdd655
AV
1629 } else {
1630 old = inode->i_op->lookup(inode, dentry, flags);
1631 d_lookup_done(dentry);
1632 if (unlikely(old)) {
1633 dput(dentry);
1634 dentry = old;
949a852e
AV
1635 }
1636 }
e3c13928 1637 return dentry;
1da177e4
LT
1638}
1639
88d8331a
AV
1640static struct dentry *lookup_slow(const struct qstr *name,
1641 struct dentry *dir,
1642 unsigned int flags)
1643{
1644 struct inode *inode = dir->d_inode;
1645 struct dentry *res;
1646 inode_lock_shared(inode);
1647 res = __lookup_slow(name, dir, flags);
1648 inode_unlock_shared(inode);
1649 return res;
1650}
1651
52094c8a
AV
1652static inline int may_lookup(struct nameidata *nd)
1653{
1654 if (nd->flags & LOOKUP_RCU) {
4ad5abb3 1655 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
52094c8a
AV
1656 if (err != -ECHILD)
1657 return err;
4675ac39 1658 if (unlazy_walk(nd))
52094c8a
AV
1659 return -ECHILD;
1660 }
4ad5abb3 1661 return inode_permission(nd->inode, MAY_EXEC);
52094c8a
AV
1662}
1663
9856fa1b
AV
1664static inline int handle_dots(struct nameidata *nd, int type)
1665{
1666 if (type == LAST_DOTDOT) {
9e6697e2
AV
1667 if (!nd->root.mnt)
1668 set_root(nd);
9856fa1b 1669 if (nd->flags & LOOKUP_RCU) {
70291aec 1670 return follow_dotdot_rcu(nd);
9856fa1b 1671 } else
397d425d 1672 return follow_dotdot(nd);
9856fa1b
AV
1673 }
1674 return 0;
1675}
1676
181548c0
AV
1677static int pick_link(struct nameidata *nd, struct path *link,
1678 struct inode *inode, unsigned seq)
d63ff28f 1679{
626de996 1680 int error;
1cf2665b 1681 struct saved *last;
756daf26 1682 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
626de996
AV
1683 path_to_nameidata(link, nd);
1684 return -ELOOP;
1685 }
bc40aee0 1686 if (!(nd->flags & LOOKUP_RCU)) {
7973387a
AV
1687 if (link->mnt == nd->path.mnt)
1688 mntget(link->mnt);
d63ff28f 1689 }
626de996
AV
1690 error = nd_alloc_stack(nd);
1691 if (unlikely(error)) {
bc40aee0 1692 if (error == -ECHILD) {
ad1633a1
AV
1693 if (unlikely(!legitimize_path(nd, link, seq))) {
1694 drop_links(nd);
1695 nd->depth = 0;
1696 nd->flags &= ~LOOKUP_RCU;
1697 nd->path.mnt = NULL;
1698 nd->path.dentry = NULL;
1699 if (!(nd->flags & LOOKUP_ROOT))
1700 nd->root.mnt = NULL;
1701 rcu_read_unlock();
4675ac39 1702 } else if (likely(unlazy_walk(nd)) == 0)
ad1633a1 1703 error = nd_alloc_stack(nd);
bc40aee0
AV
1704 }
1705 if (error) {
1706 path_put(link);
1707 return error;
1708 }
626de996
AV
1709 }
1710
ab104923 1711 last = nd->stack + nd->depth++;
1cf2665b 1712 last->link = *link;
fceef393
AV
1713 clear_delayed_call(&last->done);
1714 nd->link_inode = inode;
0450b2d1 1715 last->seq = seq;
d63ff28f
AV
1716 return 1;
1717}
1718
8f64fb1c 1719enum {WALK_FOLLOW = 1, WALK_MORE = 2};
31d66bcd 1720
3ddcd056
LT
1721/*
1722 * Do we need to follow links? We _really_ want to be able
1723 * to do this check without having to look at inode->i_op,
1724 * so we keep a cache of "no, this doesn't need follow_link"
1725 * for the common case.
1726 */
8f64fb1c
AV
1727static inline int step_into(struct nameidata *nd, struct path *path,
1728 int flags, struct inode *inode, unsigned seq)
3ddcd056 1729{
31d66bcd
AV
1730 if (!(flags & WALK_MORE) && nd->depth)
1731 put_link(nd);
8f64fb1c
AV
1732 if (likely(!d_is_symlink(path->dentry)) ||
1733 !(flags & WALK_FOLLOW || nd->flags & LOOKUP_FOLLOW)) {
1734 /* not a symlink or should not follow */
1735 path_to_nameidata(path, nd);
1736 nd->inode = inode;
1737 nd->seq = seq;
d63ff28f 1738 return 0;
8f64fb1c 1739 }
a7f77542
AV
1740 /* make sure that d_is_symlink above matches inode */
1741 if (nd->flags & LOOKUP_RCU) {
8f64fb1c 1742 if (read_seqcount_retry(&path->dentry->d_seq, seq))
a7f77542
AV
1743 return -ECHILD;
1744 }
8f64fb1c 1745 return pick_link(nd, path, inode, seq);
3ddcd056
LT
1746}
1747
4693a547 1748static int walk_component(struct nameidata *nd, int flags)
ce57dfc1 1749{
caa85634 1750 struct path path;
ce57dfc1 1751 struct inode *inode;
254cf582 1752 unsigned seq;
ce57dfc1
AV
1753 int err;
1754 /*
1755 * "." and ".." are special - ".." especially so because it has
1756 * to be able to know about the current root directory and
1757 * parent relationships.
1758 */
4693a547
AV
1759 if (unlikely(nd->last_type != LAST_NORM)) {
1760 err = handle_dots(nd, nd->last_type);
1c4ff1a8 1761 if (!(flags & WALK_MORE) && nd->depth)
4693a547
AV
1762 put_link(nd);
1763 return err;
1764 }
254cf582 1765 err = lookup_fast(nd, &path, &inode, &seq);
e9742b53 1766 if (unlikely(err <= 0)) {
697f514d 1767 if (err < 0)
f0a9ba70 1768 return err;
e3c13928
AV
1769 path.dentry = lookup_slow(&nd->last, nd->path.dentry,
1770 nd->flags);
1771 if (IS_ERR(path.dentry))
1772 return PTR_ERR(path.dentry);
7500c38a 1773
e3c13928
AV
1774 path.mnt = nd->path.mnt;
1775 err = follow_managed(&path, nd);
1776 if (unlikely(err < 0))
f0a9ba70 1777 return err;
697f514d 1778
7500c38a
AV
1779 if (unlikely(d_is_negative(path.dentry))) {
1780 path_to_nameidata(&path, nd);
1781 return -ENOENT;
1782 }
1783
254cf582 1784 seq = 0; /* we are already out of RCU mode */
d4565649 1785 inode = d_backing_inode(path.dentry);
ce57dfc1 1786 }
697f514d 1787
8f64fb1c 1788 return step_into(nd, &path, flags, inode, seq);
ce57dfc1
AV
1789}
1790
bfcfaa77
LT
1791/*
1792 * We can do the critical dentry name comparison and hashing
1793 * operations one word at a time, but we are limited to:
1794 *
1795 * - Architectures with fast unaligned word accesses. We could
1796 * do a "get_unaligned()" if this helps and is sufficiently
1797 * fast.
1798 *
bfcfaa77
LT
1799 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1800 * do not trap on the (extremely unlikely) case of a page
1801 * crossing operation.
1802 *
1803 * - Furthermore, we need an efficient 64-bit compile for the
1804 * 64-bit case in order to generate the "number of bytes in
1805 * the final mask". Again, that could be replaced with a
1806 * efficient population count instruction or similar.
1807 */
1808#ifdef CONFIG_DCACHE_WORD_ACCESS
1809
f68e556e 1810#include <asm/word-at-a-time.h>
bfcfaa77 1811
468a9428 1812#ifdef HASH_MIX
bfcfaa77 1813
468a9428 1814/* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
bfcfaa77 1815
468a9428 1816#elif defined(CONFIG_64BIT)
0fed3ac8 1817/*
2a18da7a
GS
1818 * Register pressure in the mixing function is an issue, particularly
1819 * on 32-bit x86, but almost any function requires one state value and
1820 * one temporary. Instead, use a function designed for two state values
1821 * and no temporaries.
1822 *
1823 * This function cannot create a collision in only two iterations, so
1824 * we have two iterations to achieve avalanche. In those two iterations,
1825 * we have six layers of mixing, which is enough to spread one bit's
1826 * influence out to 2^6 = 64 state bits.
1827 *
1828 * Rotate constants are scored by considering either 64 one-bit input
1829 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1830 * probability of that delta causing a change to each of the 128 output
1831 * bits, using a sample of random initial states.
1832 *
1833 * The Shannon entropy of the computed probabilities is then summed
1834 * to produce a score. Ideally, any input change has a 50% chance of
1835 * toggling any given output bit.
1836 *
1837 * Mixing scores (in bits) for (12,45):
1838 * Input delta: 1-bit 2-bit
1839 * 1 round: 713.3 42542.6
1840 * 2 rounds: 2753.7 140389.8
1841 * 3 rounds: 5954.1 233458.2
1842 * 4 rounds: 7862.6 256672.2
1843 * Perfect: 8192 258048
1844 * (64*128) (64*63/2 * 128)
0fed3ac8 1845 */
2a18da7a
GS
1846#define HASH_MIX(x, y, a) \
1847 ( x ^= (a), \
1848 y ^= x, x = rol64(x,12),\
1849 x += y, y = rol64(y,45),\
1850 y *= 9 )
bfcfaa77 1851
0fed3ac8 1852/*
2a18da7a
GS
1853 * Fold two longs into one 32-bit hash value. This must be fast, but
1854 * latency isn't quite as critical, as there is a fair bit of additional
1855 * work done before the hash value is used.
0fed3ac8 1856 */
2a18da7a 1857static inline unsigned int fold_hash(unsigned long x, unsigned long y)
0fed3ac8 1858{
2a18da7a
GS
1859 y ^= x * GOLDEN_RATIO_64;
1860 y *= GOLDEN_RATIO_64;
1861 return y >> 32;
0fed3ac8
GS
1862}
1863
bfcfaa77
LT
1864#else /* 32-bit case */
1865
2a18da7a
GS
1866/*
1867 * Mixing scores (in bits) for (7,20):
1868 * Input delta: 1-bit 2-bit
1869 * 1 round: 330.3 9201.6
1870 * 2 rounds: 1246.4 25475.4
1871 * 3 rounds: 1907.1 31295.1
1872 * 4 rounds: 2042.3 31718.6
1873 * Perfect: 2048 31744
1874 * (32*64) (32*31/2 * 64)
1875 */
1876#define HASH_MIX(x, y, a) \
1877 ( x ^= (a), \
1878 y ^= x, x = rol32(x, 7),\
1879 x += y, y = rol32(y,20),\
1880 y *= 9 )
bfcfaa77 1881
2a18da7a 1882static inline unsigned int fold_hash(unsigned long x, unsigned long y)
0fed3ac8 1883{
2a18da7a
GS
1884 /* Use arch-optimized multiply if one exists */
1885 return __hash_32(y ^ __hash_32(x));
0fed3ac8
GS
1886}
1887
bfcfaa77
LT
1888#endif
1889
2a18da7a
GS
1890/*
1891 * Return the hash of a string of known length. This is carfully
1892 * designed to match hash_name(), which is the more critical function.
1893 * In particular, we must end by hashing a final word containing 0..7
1894 * payload bytes, to match the way that hash_name() iterates until it
1895 * finds the delimiter after the name.
1896 */
8387ff25 1897unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
bfcfaa77 1898{
8387ff25 1899 unsigned long a, x = 0, y = (unsigned long)salt;
bfcfaa77
LT
1900
1901 for (;;) {
fcfd2fbf
GS
1902 if (!len)
1903 goto done;
e419b4cc 1904 a = load_unaligned_zeropad(name);
bfcfaa77
LT
1905 if (len < sizeof(unsigned long))
1906 break;
2a18da7a 1907 HASH_MIX(x, y, a);
bfcfaa77
LT
1908 name += sizeof(unsigned long);
1909 len -= sizeof(unsigned long);
bfcfaa77 1910 }
2a18da7a 1911 x ^= a & bytemask_from_count(len);
bfcfaa77 1912done:
2a18da7a 1913 return fold_hash(x, y);
bfcfaa77
LT
1914}
1915EXPORT_SYMBOL(full_name_hash);
1916
fcfd2fbf 1917/* Return the "hash_len" (hash and length) of a null-terminated string */
8387ff25 1918u64 hashlen_string(const void *salt, const char *name)
fcfd2fbf 1919{
8387ff25
LT
1920 unsigned long a = 0, x = 0, y = (unsigned long)salt;
1921 unsigned long adata, mask, len;
fcfd2fbf
GS
1922 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1923
8387ff25
LT
1924 len = 0;
1925 goto inside;
1926
fcfd2fbf 1927 do {
2a18da7a 1928 HASH_MIX(x, y, a);
fcfd2fbf 1929 len += sizeof(unsigned long);
8387ff25 1930inside:
fcfd2fbf
GS
1931 a = load_unaligned_zeropad(name+len);
1932 } while (!has_zero(a, &adata, &constants));
1933
1934 adata = prep_zero_mask(a, adata, &constants);
1935 mask = create_zero_mask(adata);
2a18da7a 1936 x ^= a & zero_bytemask(mask);
fcfd2fbf 1937
2a18da7a 1938 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
fcfd2fbf
GS
1939}
1940EXPORT_SYMBOL(hashlen_string);
1941
bfcfaa77
LT
1942/*
1943 * Calculate the length and hash of the path component, and
d6bb3e90 1944 * return the "hash_len" as the result.
bfcfaa77 1945 */
8387ff25 1946static inline u64 hash_name(const void *salt, const char *name)
bfcfaa77 1947{
8387ff25
LT
1948 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
1949 unsigned long adata, bdata, mask, len;
36126f8f 1950 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
bfcfaa77 1951
8387ff25
LT
1952 len = 0;
1953 goto inside;
1954
bfcfaa77 1955 do {
2a18da7a 1956 HASH_MIX(x, y, a);
bfcfaa77 1957 len += sizeof(unsigned long);
8387ff25 1958inside:
e419b4cc 1959 a = load_unaligned_zeropad(name+len);
36126f8f
LT
1960 b = a ^ REPEAT_BYTE('/');
1961 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1962
1963 adata = prep_zero_mask(a, adata, &constants);
1964 bdata = prep_zero_mask(b, bdata, &constants);
36126f8f 1965 mask = create_zero_mask(adata | bdata);
2a18da7a 1966 x ^= a & zero_bytemask(mask);
36126f8f 1967
2a18da7a 1968 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
bfcfaa77
LT
1969}
1970
2a18da7a 1971#else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
bfcfaa77 1972
fcfd2fbf 1973/* Return the hash of a string of known length */
8387ff25 1974unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
0145acc2 1975{
8387ff25 1976 unsigned long hash = init_name_hash(salt);
0145acc2 1977 while (len--)
fcfd2fbf 1978 hash = partial_name_hash((unsigned char)*name++, hash);
0145acc2
LT
1979 return end_name_hash(hash);
1980}
ae942ae7 1981EXPORT_SYMBOL(full_name_hash);
0145acc2 1982
fcfd2fbf 1983/* Return the "hash_len" (hash and length) of a null-terminated string */
8387ff25 1984u64 hashlen_string(const void *salt, const char *name)
fcfd2fbf 1985{
8387ff25 1986 unsigned long hash = init_name_hash(salt);
fcfd2fbf
GS
1987 unsigned long len = 0, c;
1988
1989 c = (unsigned char)*name;
e0ab7af9 1990 while (c) {
fcfd2fbf
GS
1991 len++;
1992 hash = partial_name_hash(c, hash);
1993 c = (unsigned char)name[len];
e0ab7af9 1994 }
fcfd2fbf
GS
1995 return hashlen_create(end_name_hash(hash), len);
1996}
f2a031b6 1997EXPORT_SYMBOL(hashlen_string);
fcfd2fbf 1998
200e9ef7
LT
1999/*
2000 * We know there's a real path component here of at least
2001 * one character.
2002 */
8387ff25 2003static inline u64 hash_name(const void *salt, const char *name)
200e9ef7 2004{
8387ff25 2005 unsigned long hash = init_name_hash(salt);
200e9ef7
LT
2006 unsigned long len = 0, c;
2007
2008 c = (unsigned char)*name;
2009 do {
2010 len++;
2011 hash = partial_name_hash(c, hash);
2012 c = (unsigned char)name[len];
2013 } while (c && c != '/');
d6bb3e90 2014 return hashlen_create(end_name_hash(hash), len);
200e9ef7
LT
2015}
2016
bfcfaa77
LT
2017#endif
2018
1da177e4
LT
2019/*
2020 * Name resolution.
ea3834d9
PM
2021 * This is the basic name resolution function, turning a pathname into
2022 * the final dentry. We expect 'base' to be positive and a directory.
1da177e4 2023 *
ea3834d9
PM
2024 * Returns 0 and nd will have valid dentry and mnt on success.
2025 * Returns error and drops reference to input namei data on failure.
1da177e4 2026 */
6de88d72 2027static int link_path_walk(const char *name, struct nameidata *nd)
1da177e4 2028{
1da177e4 2029 int err;
32cd7468 2030
1da177e4
LT
2031 while (*name=='/')
2032 name++;
2033 if (!*name)
9e18f10a 2034 return 0;
1da177e4 2035
1da177e4
LT
2036 /* At this point we know we have a real path component. */
2037 for(;;) {
d6bb3e90 2038 u64 hash_len;
fe479a58 2039 int type;
1da177e4 2040
52094c8a 2041 err = may_lookup(nd);
2a18da7a 2042 if (err)
3595e234 2043 return err;
1da177e4 2044
8387ff25 2045 hash_len = hash_name(nd->path.dentry, name);
1da177e4 2046
fe479a58 2047 type = LAST_NORM;
d6bb3e90 2048 if (name[0] == '.') switch (hashlen_len(hash_len)) {
fe479a58 2049 case 2:
200e9ef7 2050 if (name[1] == '.') {
fe479a58 2051 type = LAST_DOTDOT;
16c2cd71
AV
2052 nd->flags |= LOOKUP_JUMPED;
2053 }
fe479a58
AV
2054 break;
2055 case 1:
2056 type = LAST_DOT;
2057 }
5a202bcd
AV
2058 if (likely(type == LAST_NORM)) {
2059 struct dentry *parent = nd->path.dentry;
16c2cd71 2060 nd->flags &= ~LOOKUP_JUMPED;
5a202bcd 2061 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
a060dc50 2062 struct qstr this = { { .hash_len = hash_len }, .name = name };
da53be12 2063 err = parent->d_op->d_hash(parent, &this);
5a202bcd 2064 if (err < 0)
3595e234 2065 return err;
d6bb3e90
LT
2066 hash_len = this.hash_len;
2067 name = this.name;
5a202bcd
AV
2068 }
2069 }
fe479a58 2070
d6bb3e90
LT
2071 nd->last.hash_len = hash_len;
2072 nd->last.name = name;
5f4a6a69
AV
2073 nd->last_type = type;
2074
d6bb3e90
LT
2075 name += hashlen_len(hash_len);
2076 if (!*name)
bdf6cbf1 2077 goto OK;
200e9ef7
LT
2078 /*
2079 * If it wasn't NUL, we know it was '/'. Skip that
2080 * slash, and continue until no more slashes.
2081 */
2082 do {
d6bb3e90
LT
2083 name++;
2084 } while (unlikely(*name == '/'));
8620c238
AV
2085 if (unlikely(!*name)) {
2086OK:
368ee9ba 2087 /* pathname body, done */
8620c238
AV
2088 if (!nd->depth)
2089 return 0;
2090 name = nd->stack[nd->depth - 1].name;
368ee9ba 2091 /* trailing symlink, done */
8620c238
AV
2092 if (!name)
2093 return 0;
2094 /* last component of nested symlink */
8f64fb1c 2095 err = walk_component(nd, WALK_FOLLOW);
1c4ff1a8
AV
2096 } else {
2097 /* not the last component */
8f64fb1c 2098 err = walk_component(nd, WALK_FOLLOW | WALK_MORE);
8620c238 2099 }
ce57dfc1 2100 if (err < 0)
3595e234 2101 return err;
1da177e4 2102
ce57dfc1 2103 if (err) {
626de996 2104 const char *s = get_link(nd);
5a460275 2105
a1c83681 2106 if (IS_ERR(s))
3595e234 2107 return PTR_ERR(s);
d40bcc09
AV
2108 err = 0;
2109 if (unlikely(!s)) {
2110 /* jumped */
b9ff4429 2111 put_link(nd);
d40bcc09 2112 } else {
fab51e8a
AV
2113 nd->stack[nd->depth - 1].name = name;
2114 name = s;
2115 continue;
d40bcc09 2116 }
31e6b01f 2117 }
97242f99
AV
2118 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2119 if (nd->flags & LOOKUP_RCU) {
4675ac39 2120 if (unlazy_walk(nd))
97242f99
AV
2121 return -ECHILD;
2122 }
3595e234 2123 return -ENOTDIR;
97242f99 2124 }
1da177e4 2125 }
1da177e4
LT
2126}
2127
c8a53ee5 2128static const char *path_init(struct nameidata *nd, unsigned flags)
31e6b01f 2129{
c8a53ee5 2130 const char *s = nd->name->name;
31e6b01f 2131
c0eb027e
LT
2132 if (!*s)
2133 flags &= ~LOOKUP_RCU;
2134
31e6b01f 2135 nd->last_type = LAST_ROOT; /* if there are only slashes... */
980f3ea2 2136 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
31e6b01f 2137 nd->depth = 0;
5b6ca027 2138 if (flags & LOOKUP_ROOT) {
b18825a7
DH
2139 struct dentry *root = nd->root.dentry;
2140 struct inode *inode = root->d_inode;
93893862
AV
2141 if (*s && unlikely(!d_can_lookup(root)))
2142 return ERR_PTR(-ENOTDIR);
5b6ca027
AV
2143 nd->path = nd->root;
2144 nd->inode = inode;
2145 if (flags & LOOKUP_RCU) {
8b61e74f 2146 rcu_read_lock();
5b6ca027 2147 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
8f47a016 2148 nd->root_seq = nd->seq;
48a066e7 2149 nd->m_seq = read_seqbegin(&mount_lock);
5b6ca027
AV
2150 } else {
2151 path_get(&nd->path);
2152 }
368ee9ba 2153 return s;
5b6ca027
AV
2154 }
2155
31e6b01f 2156 nd->root.mnt = NULL;
248fb5b9
AV
2157 nd->path.mnt = NULL;
2158 nd->path.dentry = NULL;
31e6b01f 2159
48a066e7 2160 nd->m_seq = read_seqbegin(&mount_lock);
fd2f7cb5 2161 if (*s == '/') {
9e6697e2 2162 if (flags & LOOKUP_RCU)
8b61e74f 2163 rcu_read_lock();
9e6697e2 2164 set_root(nd);
248fb5b9 2165 if (likely(!nd_jump_root(nd)))
ef55d917 2166 return s;
248fb5b9 2167 nd->root.mnt = NULL;
ef55d917
AV
2168 rcu_read_unlock();
2169 return ERR_PTR(-ECHILD);
c8a53ee5 2170 } else if (nd->dfd == AT_FDCWD) {
e41f7d4e
AV
2171 if (flags & LOOKUP_RCU) {
2172 struct fs_struct *fs = current->fs;
2173 unsigned seq;
31e6b01f 2174
8b61e74f 2175 rcu_read_lock();
c28cc364 2176
e41f7d4e
AV
2177 do {
2178 seq = read_seqcount_begin(&fs->seq);
2179 nd->path = fs->pwd;
ef55d917 2180 nd->inode = nd->path.dentry->d_inode;
e41f7d4e
AV
2181 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2182 } while (read_seqcount_retry(&fs->seq, seq));
2183 } else {
2184 get_fs_pwd(current->fs, &nd->path);
ef55d917 2185 nd->inode = nd->path.dentry->d_inode;
e41f7d4e 2186 }
ef55d917 2187 return s;
31e6b01f 2188 } else {
582aa64a 2189 /* Caller must check execute permissions on the starting path component */
c8a53ee5 2190 struct fd f = fdget_raw(nd->dfd);
31e6b01f
NP
2191 struct dentry *dentry;
2192
2903ff01 2193 if (!f.file)
368ee9ba 2194 return ERR_PTR(-EBADF);
31e6b01f 2195
2903ff01 2196 dentry = f.file->f_path.dentry;
31e6b01f 2197
fd2f7cb5 2198 if (*s) {
44b1d530 2199 if (!d_can_lookup(dentry)) {
2903ff01 2200 fdput(f);
368ee9ba 2201 return ERR_PTR(-ENOTDIR);
2903ff01 2202 }
f52e0c11 2203 }
31e6b01f 2204
2903ff01 2205 nd->path = f.file->f_path;
e41f7d4e 2206 if (flags & LOOKUP_RCU) {
8b61e74f 2207 rcu_read_lock();
34a26b99
AV
2208 nd->inode = nd->path.dentry->d_inode;
2209 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
e41f7d4e 2210 } else {
2903ff01 2211 path_get(&nd->path);
34a26b99 2212 nd->inode = nd->path.dentry->d_inode;
e41f7d4e 2213 }
34a26b99 2214 fdput(f);
368ee9ba 2215 return s;
31e6b01f 2216 }
9b4a9b14
AV
2217}
2218
3bdba28b 2219static const char *trailing_symlink(struct nameidata *nd)
95fa25d9
AV
2220{
2221 const char *s;
fec2fa24 2222 int error = may_follow_link(nd);
deb106c6 2223 if (unlikely(error))
3bdba28b 2224 return ERR_PTR(error);
95fa25d9 2225 nd->flags |= LOOKUP_PARENT;
fab51e8a 2226 nd->stack[0].name = NULL;
3b2e7f75 2227 s = get_link(nd);
deb106c6 2228 return s ? s : "";
95fa25d9
AV
2229}
2230
caa85634 2231static inline int lookup_last(struct nameidata *nd)
bd92d7fe
AV
2232{
2233 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2234 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2235
2236 nd->flags &= ~LOOKUP_PARENT;
1c4ff1a8 2237 return walk_component(nd, 0);
bd92d7fe
AV
2238}
2239
4f757f3c
AV
2240static int handle_lookup_down(struct nameidata *nd)
2241{
2242 struct path path = nd->path;
2243 struct inode *inode = nd->inode;
2244 unsigned seq = nd->seq;
2245 int err;
2246
2247 if (nd->flags & LOOKUP_RCU) {
2248 /*
2249 * don't bother with unlazy_walk on failure - we are
2250 * at the very beginning of walk, so we lose nothing
2251 * if we simply redo everything in non-RCU mode
2252 */
2253 if (unlikely(!__follow_mount_rcu(nd, &path, &inode, &seq)))
2254 return -ECHILD;
2255 } else {
2256 dget(path.dentry);
2257 err = follow_managed(&path, nd);
2258 if (unlikely(err < 0))
2259 return err;
2260 inode = d_backing_inode(path.dentry);
2261 seq = 0;
2262 }
2263 path_to_nameidata(&path, nd);
2264 nd->inode = inode;
2265 nd->seq = seq;
2266 return 0;
2267}
2268
9b4a9b14 2269/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
c8a53ee5 2270static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
9b4a9b14 2271{
c8a53ee5 2272 const char *s = path_init(nd, flags);
bd92d7fe 2273 int err;
31e6b01f 2274
368ee9ba
AV
2275 if (IS_ERR(s))
2276 return PTR_ERR(s);
4f757f3c
AV
2277
2278 if (unlikely(flags & LOOKUP_DOWN)) {
2279 err = handle_lookup_down(nd);
2280 if (unlikely(err < 0)) {
2281 terminate_walk(nd);
2282 return err;
2283 }
2284 }
2285
3bdba28b
AV
2286 while (!(err = link_path_walk(s, nd))
2287 && ((err = lookup_last(nd)) > 0)) {
2288 s = trailing_symlink(nd);
2289 if (IS_ERR(s)) {
2290 err = PTR_ERR(s);
2291 break;
bd92d7fe
AV
2292 }
2293 }
9f1fafee
AV
2294 if (!err)
2295 err = complete_walk(nd);
bd92d7fe 2296
deb106c6
AV
2297 if (!err && nd->flags & LOOKUP_DIRECTORY)
2298 if (!d_can_lookup(nd->path.dentry))
bd23a539 2299 err = -ENOTDIR;
625b6d10
AV
2300 if (!err) {
2301 *path = nd->path;
2302 nd->path.mnt = NULL;
2303 nd->path.dentry = NULL;
2304 }
2305 terminate_walk(nd);
bd92d7fe 2306 return err;
ee0827cd 2307}
31e6b01f 2308
625b6d10 2309static int filename_lookup(int dfd, struct filename *name, unsigned flags,
9ad1aaa6 2310 struct path *path, struct path *root)
ee0827cd 2311{
894bc8c4 2312 int retval;
9883d185 2313 struct nameidata nd;
abc9f5be
AV
2314 if (IS_ERR(name))
2315 return PTR_ERR(name);
9ad1aaa6
AV
2316 if (unlikely(root)) {
2317 nd.root = *root;
2318 flags |= LOOKUP_ROOT;
2319 }
9883d185 2320 set_nameidata(&nd, dfd, name);
c8a53ee5 2321 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
ee0827cd 2322 if (unlikely(retval == -ECHILD))
c8a53ee5 2323 retval = path_lookupat(&nd, flags, path);
ee0827cd 2324 if (unlikely(retval == -ESTALE))
c8a53ee5 2325 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
31e6b01f 2326
f78570dd 2327 if (likely(!retval))
625b6d10 2328 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
9883d185 2329 restore_nameidata();
e4bd1c1a 2330 putname(name);
170aa3d0 2331 return retval;
1da177e4
LT
2332}
2333
8bcb77fa 2334/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
c8a53ee5 2335static int path_parentat(struct nameidata *nd, unsigned flags,
391172c4 2336 struct path *parent)
8bcb77fa 2337{
c8a53ee5 2338 const char *s = path_init(nd, flags);
368ee9ba
AV
2339 int err;
2340 if (IS_ERR(s))
2341 return PTR_ERR(s);
2342 err = link_path_walk(s, nd);
8bcb77fa
AV
2343 if (!err)
2344 err = complete_walk(nd);
391172c4
AV
2345 if (!err) {
2346 *parent = nd->path;
2347 nd->path.mnt = NULL;
2348 nd->path.dentry = NULL;
2349 }
2350 terminate_walk(nd);
8bcb77fa
AV
2351 return err;
2352}
2353
5c31b6ce 2354static struct filename *filename_parentat(int dfd, struct filename *name,
391172c4
AV
2355 unsigned int flags, struct path *parent,
2356 struct qstr *last, int *type)
8bcb77fa
AV
2357{
2358 int retval;
9883d185 2359 struct nameidata nd;
8bcb77fa 2360
5c31b6ce
AV
2361 if (IS_ERR(name))
2362 return name;
9883d185 2363 set_nameidata(&nd, dfd, name);
c8a53ee5 2364 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
8bcb77fa 2365 if (unlikely(retval == -ECHILD))
c8a53ee5 2366 retval = path_parentat(&nd, flags, parent);
8bcb77fa 2367 if (unlikely(retval == -ESTALE))
c8a53ee5 2368 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
391172c4
AV
2369 if (likely(!retval)) {
2370 *last = nd.last;
2371 *type = nd.last_type;
2372 audit_inode(name, parent->dentry, LOOKUP_PARENT);
5c31b6ce
AV
2373 } else {
2374 putname(name);
2375 name = ERR_PTR(retval);
391172c4 2376 }
9883d185 2377 restore_nameidata();
5c31b6ce 2378 return name;
8bcb77fa
AV
2379}
2380
79714f72
AV
2381/* does lookup, returns the object with parent locked */
2382struct dentry *kern_path_locked(const char *name, struct path *path)
5590ff0d 2383{
5c31b6ce
AV
2384 struct filename *filename;
2385 struct dentry *d;
391172c4
AV
2386 struct qstr last;
2387 int type;
51689104 2388
5c31b6ce
AV
2389 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2390 &last, &type);
51689104
PM
2391 if (IS_ERR(filename))
2392 return ERR_CAST(filename);
5c31b6ce 2393 if (unlikely(type != LAST_NORM)) {
391172c4 2394 path_put(path);
5c31b6ce
AV
2395 putname(filename);
2396 return ERR_PTR(-EINVAL);
79714f72 2397 }
5955102c 2398 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
391172c4 2399 d = __lookup_hash(&last, path->dentry, 0);
79714f72 2400 if (IS_ERR(d)) {
5955102c 2401 inode_unlock(path->dentry->d_inode);
391172c4 2402 path_put(path);
79714f72 2403 }
51689104 2404 putname(filename);
79714f72 2405 return d;
5590ff0d
UD
2406}
2407
d1811465
AV
2408int kern_path(const char *name, unsigned int flags, struct path *path)
2409{
abc9f5be
AV
2410 return filename_lookup(AT_FDCWD, getname_kernel(name),
2411 flags, path, NULL);
d1811465 2412}
4d359507 2413EXPORT_SYMBOL(kern_path);
d1811465 2414
16f18200
JJS
2415/**
2416 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2417 * @dentry: pointer to dentry of the base directory
2418 * @mnt: pointer to vfs mount of the base directory
2419 * @name: pointer to file name
2420 * @flags: lookup flags
e0a01249 2421 * @path: pointer to struct path to fill
16f18200
JJS
2422 */
2423int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2424 const char *name, unsigned int flags,
e0a01249 2425 struct path *path)
16f18200 2426{
9ad1aaa6 2427 struct path root = {.mnt = mnt, .dentry = dentry};
9ad1aaa6 2428 /* the first argument of filename_lookup() is ignored with root */
abc9f5be
AV
2429 return filename_lookup(AT_FDCWD, getname_kernel(name),
2430 flags , path, &root);
16f18200 2431}
4d359507 2432EXPORT_SYMBOL(vfs_path_lookup);
16f18200 2433
3c95f0dc
AV
2434static int lookup_one_len_common(const char *name, struct dentry *base,
2435 int len, struct qstr *this)
057f6c01 2436{
3c95f0dc
AV
2437 this->name = name;
2438 this->len = len;
2439 this->hash = full_name_hash(base, name, len);
6a96ba54 2440 if (!len)
3c95f0dc 2441 return -EACCES;
6a96ba54 2442
21d8a15a
AV
2443 if (unlikely(name[0] == '.')) {
2444 if (len < 2 || (len == 2 && name[1] == '.'))
3c95f0dc 2445 return -EACCES;
21d8a15a
AV
2446 }
2447
6a96ba54 2448 while (len--) {
3c95f0dc 2449 unsigned int c = *(const unsigned char *)name++;
6a96ba54 2450 if (c == '/' || c == '\0')
3c95f0dc 2451 return -EACCES;
6a96ba54 2452 }
5a202bcd
AV
2453 /*
2454 * See if the low-level filesystem might want
2455 * to use its own hash..
2456 */
2457 if (base->d_flags & DCACHE_OP_HASH) {
3c95f0dc 2458 int err = base->d_op->d_hash(base, this);
5a202bcd 2459 if (err < 0)
3c95f0dc 2460 return err;
5a202bcd 2461 }
eead1911 2462
3c95f0dc
AV
2463 return inode_permission(base->d_inode, MAY_EXEC);
2464}
2465
2466/**
2467 * lookup_one_len - filesystem helper to lookup single pathname component
2468 * @name: pathname component to lookup
2469 * @base: base directory to lookup from
2470 * @len: maximum length @len should be interpreted to
2471 *
2472 * Note that this routine is purely a helper for filesystem usage and should
2473 * not be called by generic code.
2474 *
2475 * The caller must hold base->i_mutex.
2476 */
2477struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2478{
8613a209 2479 struct dentry *dentry;
3c95f0dc
AV
2480 struct qstr this;
2481 int err;
2482
2483 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2484
2485 err = lookup_one_len_common(name, base, len, &this);
cda309de
MS
2486 if (err)
2487 return ERR_PTR(err);
2488
8613a209
AV
2489 dentry = lookup_dcache(&this, base, 0);
2490 return dentry ? dentry : __lookup_slow(&this, base, 0);
057f6c01 2491}
4d359507 2492EXPORT_SYMBOL(lookup_one_len);
057f6c01 2493
bbddca8e
N
2494/**
2495 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2496 * @name: pathname component to lookup
2497 * @base: base directory to lookup from
2498 * @len: maximum length @len should be interpreted to
2499 *
2500 * Note that this routine is purely a helper for filesystem usage and should
2501 * not be called by generic code.
2502 *
2503 * Unlike lookup_one_len, it should be called without the parent
2504 * i_mutex held, and will take the i_mutex itself if necessary.
2505 */
2506struct dentry *lookup_one_len_unlocked(const char *name,
2507 struct dentry *base, int len)
2508{
2509 struct qstr this;
bbddca8e 2510 int err;
20d00ee8 2511 struct dentry *ret;
bbddca8e 2512
3c95f0dc 2513 err = lookup_one_len_common(name, base, len, &this);
bbddca8e
N
2514 if (err)
2515 return ERR_PTR(err);
2516
20d00ee8
LT
2517 ret = lookup_dcache(&this, base, 0);
2518 if (!ret)
2519 ret = lookup_slow(&this, base, 0);
2520 return ret;
bbddca8e
N
2521}
2522EXPORT_SYMBOL(lookup_one_len_unlocked);
2523
eedf265a
EB
2524#ifdef CONFIG_UNIX98_PTYS
2525int path_pts(struct path *path)
2526{
2527 /* Find something mounted on "pts" in the same directory as
2528 * the input path.
2529 */
2530 struct dentry *child, *parent;
2531 struct qstr this;
2532 int ret;
2533
2534 ret = path_parent_directory(path);
2535 if (ret)
2536 return ret;
2537
2538 parent = path->dentry;
2539 this.name = "pts";
2540 this.len = 3;
2541 child = d_hash_and_lookup(parent, &this);
2542 if (!child)
2543 return -ENOENT;
2544
2545 path->dentry = child;
2546 dput(parent);
2547 follow_mount(path);
2548 return 0;
2549}
2550#endif
2551
1fa1e7f6
AW
2552int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2553 struct path *path, int *empty)
1da177e4 2554{
abc9f5be
AV
2555 return filename_lookup(dfd, getname_flags(name, flags, empty),
2556 flags, path, NULL);
1da177e4 2557}
b853a161 2558EXPORT_SYMBOL(user_path_at_empty);
1fa1e7f6 2559
8033426e 2560/**
197df04c 2561 * mountpoint_last - look up last component for umount
8033426e 2562 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
8033426e
JL
2563 *
2564 * This is a special lookup_last function just for umount. In this case, we
2565 * need to resolve the path without doing any revalidation.
2566 *
2567 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2568 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2569 * in almost all cases, this lookup will be served out of the dcache. The only
2570 * cases where it won't are if nd->last refers to a symlink or the path is
2571 * bogus and it doesn't exist.
2572 *
2573 * Returns:
2574 * -error: if there was an error during lookup. This includes -ENOENT if the
ba8f4613 2575 * lookup found a negative dentry.
8033426e 2576 *
ba8f4613
AV
2577 * 0: if we successfully resolved nd->last and found it to not to be a
2578 * symlink that needs to be followed.
8033426e
JL
2579 *
2580 * 1: if we successfully resolved nd->last and found it to be a symlink
ba8f4613 2581 * that needs to be followed.
8033426e
JL
2582 */
2583static int
ba8f4613 2584mountpoint_last(struct nameidata *nd)
8033426e
JL
2585{
2586 int error = 0;
8033426e 2587 struct dentry *dir = nd->path.dentry;
ba8f4613 2588 struct path path;
8033426e 2589
35759521
AV
2590 /* If we're in rcuwalk, drop out of it to handle last component */
2591 if (nd->flags & LOOKUP_RCU) {
4675ac39 2592 if (unlazy_walk(nd))
deb106c6 2593 return -ECHILD;
8033426e
JL
2594 }
2595
2596 nd->flags &= ~LOOKUP_PARENT;
2597
2598 if (unlikely(nd->last_type != LAST_NORM)) {
2599 error = handle_dots(nd, nd->last_type);
35759521 2600 if (error)
deb106c6 2601 return error;
ba8f4613 2602 path.dentry = dget(nd->path.dentry);
949a852e 2603 } else {
ba8f4613
AV
2604 path.dentry = d_lookup(dir, &nd->last);
2605 if (!path.dentry) {
949a852e
AV
2606 /*
2607 * No cached dentry. Mounted dentries are pinned in the
2608 * cache, so that means that this dentry is probably
2609 * a symlink or the path doesn't actually point
2610 * to a mounted dentry.
2611 */
ba8f4613 2612 path.dentry = lookup_slow(&nd->last, dir,
949a852e 2613 nd->flags | LOOKUP_NO_REVAL);
ba8f4613
AV
2614 if (IS_ERR(path.dentry))
2615 return PTR_ERR(path.dentry);
bcceeeba 2616 }
8033426e 2617 }
ba8f4613
AV
2618 if (d_is_negative(path.dentry)) {
2619 dput(path.dentry);
deb106c6 2620 return -ENOENT;
8033426e 2621 }
ba8f4613 2622 path.mnt = nd->path.mnt;
8f64fb1c 2623 return step_into(nd, &path, 0, d_backing_inode(path.dentry), 0);
8033426e
JL
2624}
2625
2626/**
197df04c 2627 * path_mountpoint - look up a path to be umounted
2a78b857 2628 * @nd: lookup context
8033426e 2629 * @flags: lookup flags
c8a53ee5 2630 * @path: pointer to container for result
8033426e
JL
2631 *
2632 * Look up the given name, but don't attempt to revalidate the last component.
606d6fe3 2633 * Returns 0 and "path" will be valid on success; Returns error otherwise.
8033426e
JL
2634 */
2635static int
c8a53ee5 2636path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
8033426e 2637{
c8a53ee5 2638 const char *s = path_init(nd, flags);
368ee9ba
AV
2639 int err;
2640 if (IS_ERR(s))
2641 return PTR_ERR(s);
3bdba28b 2642 while (!(err = link_path_walk(s, nd)) &&
ba8f4613 2643 (err = mountpoint_last(nd)) > 0) {
3bdba28b
AV
2644 s = trailing_symlink(nd);
2645 if (IS_ERR(s)) {
2646 err = PTR_ERR(s);
8033426e 2647 break;
3bdba28b 2648 }
8033426e 2649 }
ba8f4613
AV
2650 if (!err) {
2651 *path = nd->path;
2652 nd->path.mnt = NULL;
2653 nd->path.dentry = NULL;
2654 follow_mount(path);
2655 }
deb106c6 2656 terminate_walk(nd);
8033426e
JL
2657 return err;
2658}
2659
2d864651 2660static int
668696dc 2661filename_mountpoint(int dfd, struct filename *name, struct path *path,
2d864651
AV
2662 unsigned int flags)
2663{
9883d185 2664 struct nameidata nd;
cbaab2db 2665 int error;
668696dc
AV
2666 if (IS_ERR(name))
2667 return PTR_ERR(name);
9883d185 2668 set_nameidata(&nd, dfd, name);
c8a53ee5 2669 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2d864651 2670 if (unlikely(error == -ECHILD))
c8a53ee5 2671 error = path_mountpoint(&nd, flags, path);
2d864651 2672 if (unlikely(error == -ESTALE))
c8a53ee5 2673 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2d864651 2674 if (likely(!error))
668696dc 2675 audit_inode(name, path->dentry, 0);
9883d185 2676 restore_nameidata();
668696dc 2677 putname(name);
2d864651
AV
2678 return error;
2679}
2680
8033426e 2681/**
197df04c 2682 * user_path_mountpoint_at - lookup a path from userland in order to umount it
8033426e
JL
2683 * @dfd: directory file descriptor
2684 * @name: pathname from userland
2685 * @flags: lookup flags
2686 * @path: pointer to container to hold result
2687 *
2688 * A umount is a special case for path walking. We're not actually interested
2689 * in the inode in this situation, and ESTALE errors can be a problem. We
2690 * simply want track down the dentry and vfsmount attached at the mountpoint
2691 * and avoid revalidating the last component.
2692 *
2693 * Returns 0 and populates "path" on success.
2694 */
2695int
197df04c 2696user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
8033426e
JL
2697 struct path *path)
2698{
cbaab2db 2699 return filename_mountpoint(dfd, getname(name), path, flags);
8033426e
JL
2700}
2701
2d864651
AV
2702int
2703kern_path_mountpoint(int dfd, const char *name, struct path *path,
2704 unsigned int flags)
2705{
cbaab2db 2706 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2d864651
AV
2707}
2708EXPORT_SYMBOL(kern_path_mountpoint);
2709
cbdf35bc 2710int __check_sticky(struct inode *dir, struct inode *inode)
1da177e4 2711{
8e96e3b7 2712 kuid_t fsuid = current_fsuid();
da9592ed 2713
8e96e3b7 2714 if (uid_eq(inode->i_uid, fsuid))
1da177e4 2715 return 0;
8e96e3b7 2716 if (uid_eq(dir->i_uid, fsuid))
1da177e4 2717 return 0;
23adbe12 2718 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
1da177e4 2719}
cbdf35bc 2720EXPORT_SYMBOL(__check_sticky);
1da177e4
LT
2721
2722/*
2723 * Check whether we can remove a link victim from directory dir, check
2724 * whether the type of victim is right.
2725 * 1. We can't do it if dir is read-only (done in permission())
2726 * 2. We should have write and exec permissions on dir
2727 * 3. We can't remove anything from append-only dir
2728 * 4. We can't do anything with immutable dir (done in permission())
2729 * 5. If the sticky bit on dir is set we should either
2730 * a. be owner of dir, or
2731 * b. be owner of victim, or
2732 * c. have CAP_FOWNER capability
2733 * 6. If the victim is append-only or immutable we can't do antyhing with
2734 * links pointing to it.
0bd23d09
EB
2735 * 7. If the victim has an unknown uid or gid we can't change the inode.
2736 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2737 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2738 * 10. We can't remove a root or mountpoint.
2739 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
1da177e4
LT
2740 * nfs_async_unlink().
2741 */
b18825a7 2742static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
1da177e4 2743{
63afdfc7 2744 struct inode *inode = d_backing_inode(victim);
1da177e4
LT
2745 int error;
2746
b18825a7 2747 if (d_is_negative(victim))
1da177e4 2748 return -ENOENT;
b18825a7 2749 BUG_ON(!inode);
1da177e4
LT
2750
2751 BUG_ON(victim->d_parent->d_inode != dir);
593d1ce8
EB
2752
2753 /* Inode writeback is not safe when the uid or gid are invalid. */
2754 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
2755 return -EOVERFLOW;
2756
4fa6b5ec 2757 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
1da177e4 2758
f419a2e3 2759 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT
2760 if (error)
2761 return error;
2762 if (IS_APPEND(dir))
2763 return -EPERM;
b18825a7
DH
2764
2765 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
0bd23d09 2766 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
1da177e4
LT
2767 return -EPERM;
2768 if (isdir) {
44b1d530 2769 if (!d_is_dir(victim))
1da177e4
LT
2770 return -ENOTDIR;
2771 if (IS_ROOT(victim))
2772 return -EBUSY;
44b1d530 2773 } else if (d_is_dir(victim))
1da177e4
LT
2774 return -EISDIR;
2775 if (IS_DEADDIR(dir))
2776 return -ENOENT;
2777 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2778 return -EBUSY;
2779 return 0;
2780}
2781
2782/* Check whether we can create an object with dentry child in directory
2783 * dir.
2784 * 1. We can't do it if child already exists (open has special treatment for
2785 * this case, but since we are inlined it's OK)
2786 * 2. We can't do it if dir is read-only (done in permission())
036d5236
EB
2787 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2788 * 4. We should have write and exec permissions on dir
2789 * 5. We can't do it if dir is immutable (done in permission())
1da177e4 2790 */
a95164d9 2791static inline int may_create(struct inode *dir, struct dentry *child)
1da177e4 2792{
036d5236 2793 struct user_namespace *s_user_ns;
14e972b4 2794 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
1da177e4
LT
2795 if (child->d_inode)
2796 return -EEXIST;
2797 if (IS_DEADDIR(dir))
2798 return -ENOENT;
036d5236
EB
2799 s_user_ns = dir->i_sb->s_user_ns;
2800 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2801 !kgid_has_mapping(s_user_ns, current_fsgid()))
2802 return -EOVERFLOW;
f419a2e3 2803 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT
2804}
2805
1da177e4
LT
2806/*
2807 * p1 and p2 should be directories on the same fs.
2808 */
2809struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2810{
2811 struct dentry *p;
2812
2813 if (p1 == p2) {
5955102c 2814 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
1da177e4
LT
2815 return NULL;
2816 }
2817
fc64005c 2818 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
1da177e4 2819
e2761a11
OH
2820 p = d_ancestor(p2, p1);
2821 if (p) {
5955102c
AV
2822 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2823 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
e2761a11 2824 return p;
1da177e4
LT
2825 }
2826
e2761a11
OH
2827 p = d_ancestor(p1, p2);
2828 if (p) {
5955102c
AV
2829 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2830 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
e2761a11 2831 return p;
1da177e4
LT
2832 }
2833
5955102c
AV
2834 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2835 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
1da177e4
LT
2836 return NULL;
2837}
4d359507 2838EXPORT_SYMBOL(lock_rename);
1da177e4
LT
2839
2840void unlock_rename(struct dentry *p1, struct dentry *p2)
2841{
5955102c 2842 inode_unlock(p1->d_inode);
1da177e4 2843 if (p1 != p2) {
5955102c 2844 inode_unlock(p2->d_inode);
fc64005c 2845 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
1da177e4
LT
2846 }
2847}
4d359507 2848EXPORT_SYMBOL(unlock_rename);
1da177e4 2849
4acdaf27 2850int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
312b63fb 2851 bool want_excl)
1da177e4 2852{
a95164d9 2853 int error = may_create(dir, dentry);
1da177e4
LT
2854 if (error)
2855 return error;
2856
acfa4380 2857 if (!dir->i_op->create)
1da177e4
LT
2858 return -EACCES; /* shouldn't it be ENOSYS? */
2859 mode &= S_IALLUGO;
2860 mode |= S_IFREG;
2861 error = security_inode_create(dir, dentry, mode);
2862 if (error)
2863 return error;
312b63fb 2864 error = dir->i_op->create(dir, dentry, mode, want_excl);
a74574aa 2865 if (!error)
f38aa942 2866 fsnotify_create(dir, dentry);
1da177e4
LT
2867 return error;
2868}
4d359507 2869EXPORT_SYMBOL(vfs_create);
1da177e4 2870
8e6c848e
AV
2871int vfs_mkobj(struct dentry *dentry, umode_t mode,
2872 int (*f)(struct dentry *, umode_t, void *),
2873 void *arg)
2874{
2875 struct inode *dir = dentry->d_parent->d_inode;
2876 int error = may_create(dir, dentry);
2877 if (error)
2878 return error;
2879
2880 mode &= S_IALLUGO;
2881 mode |= S_IFREG;
2882 error = security_inode_create(dir, dentry, mode);
2883 if (error)
2884 return error;
2885 error = f(dentry, mode, arg);
2886 if (!error)
2887 fsnotify_create(dir, dentry);
2888 return error;
2889}
2890EXPORT_SYMBOL(vfs_mkobj);
2891
a2982cc9
EB
2892bool may_open_dev(const struct path *path)
2893{
2894 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2895 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2896}
2897
f0bb5aaf 2898static int may_open(const struct path *path, int acc_mode, int flag)
1da177e4 2899{
3fb64190 2900 struct dentry *dentry = path->dentry;
1da177e4
LT
2901 struct inode *inode = dentry->d_inode;
2902 int error;
2903
2904 if (!inode)
2905 return -ENOENT;
2906
c8fe8f30
CH
2907 switch (inode->i_mode & S_IFMT) {
2908 case S_IFLNK:
1da177e4 2909 return -ELOOP;
c8fe8f30
CH
2910 case S_IFDIR:
2911 if (acc_mode & MAY_WRITE)
2912 return -EISDIR;
2913 break;
2914 case S_IFBLK:
2915 case S_IFCHR:
a2982cc9 2916 if (!may_open_dev(path))
1da177e4 2917 return -EACCES;
c8fe8f30
CH
2918 /*FALLTHRU*/
2919 case S_IFIFO:
2920 case S_IFSOCK:
1da177e4 2921 flag &= ~O_TRUNC;
c8fe8f30 2922 break;
4a3fd211 2923 }
b41572e9 2924
62fb4a15 2925 error = inode_permission(inode, MAY_OPEN | acc_mode);
b41572e9
DH
2926 if (error)
2927 return error;
6146f0d5 2928
1da177e4
LT
2929 /*
2930 * An append-only file must be opened in append mode for writing.
2931 */
2932 if (IS_APPEND(inode)) {
8737c930 2933 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
7715b521 2934 return -EPERM;
1da177e4 2935 if (flag & O_TRUNC)
7715b521 2936 return -EPERM;
1da177e4
LT
2937 }
2938
2939 /* O_NOATIME can only be set by the owner or superuser */
2e149670 2940 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
7715b521 2941 return -EPERM;
1da177e4 2942
f3c7691e 2943 return 0;
7715b521 2944}
1da177e4 2945
e1181ee6 2946static int handle_truncate(struct file *filp)
7715b521 2947{
f0bb5aaf 2948 const struct path *path = &filp->f_path;
7715b521
AV
2949 struct inode *inode = path->dentry->d_inode;
2950 int error = get_write_access(inode);
2951 if (error)
2952 return error;
2953 /*
2954 * Refuse to truncate files with mandatory locks held on them.
2955 */
d7a06983 2956 error = locks_verify_locked(filp);
7715b521 2957 if (!error)
ea0d3ab2 2958 error = security_path_truncate(path);
7715b521
AV
2959 if (!error) {
2960 error = do_truncate(path->dentry, 0,
2961 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
e1181ee6 2962 filp);
7715b521
AV
2963 }
2964 put_write_access(inode);
acd0c935 2965 return error;
1da177e4
LT
2966}
2967
d57999e1
DH
2968static inline int open_to_namei_flags(int flag)
2969{
8a5e929d
AV
2970 if ((flag & O_ACCMODE) == 3)
2971 flag--;
d57999e1
DH
2972 return flag;
2973}
2974
d3607752 2975static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
d18e9008 2976{
1328c727 2977 struct user_namespace *s_user_ns;
d18e9008
MS
2978 int error = security_path_mknod(dir, dentry, mode, 0);
2979 if (error)
2980 return error;
2981
1328c727
SF
2982 s_user_ns = dir->dentry->d_sb->s_user_ns;
2983 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2984 !kgid_has_mapping(s_user_ns, current_fsgid()))
2985 return -EOVERFLOW;
2986
d18e9008
MS
2987 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2988 if (error)
2989 return error;
2990
2991 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2992}
2993
1acf0af9
DH
2994/*
2995 * Attempt to atomically look up, create and open a file from a negative
2996 * dentry.
2997 *
2998 * Returns 0 if successful. The file will have been created and attached to
2999 * @file by the filesystem calling finish_open().
3000 *
3001 * Returns 1 if the file was looked up only or didn't need creating. The
3002 * caller will need to perform the open themselves. @path will have been
3003 * updated to point to the new dentry. This may be negative.
3004 *
3005 * Returns an error code otherwise.
3006 */
2675a4eb
AV
3007static int atomic_open(struct nameidata *nd, struct dentry *dentry,
3008 struct path *path, struct file *file,
3009 const struct open_flags *op,
1643b43f 3010 int open_flag, umode_t mode,
2675a4eb 3011 int *opened)
d18e9008 3012{
384f26e2 3013 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
d18e9008 3014 struct inode *dir = nd->path.dentry->d_inode;
d18e9008 3015 int error;
d18e9008 3016
384f26e2 3017 if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */
d18e9008 3018 open_flag &= ~O_TRUNC;
d18e9008 3019
d18e9008
MS
3020 if (nd->flags & LOOKUP_DIRECTORY)
3021 open_flag |= O_DIRECTORY;
3022
30d90494
AV
3023 file->f_path.dentry = DENTRY_NOT_SET;
3024 file->f_path.mnt = nd->path.mnt;
0fb1ea09
AV
3025 error = dir->i_op->atomic_open(dir, dentry, file,
3026 open_to_namei_flags(open_flag),
3027 mode, opened);
6fbd0714 3028 d_lookup_done(dentry);
384f26e2
AV
3029 if (!error) {
3030 /*
3031 * We didn't have the inode before the open, so check open
3032 * permission here.
3033 */
3034 int acc_mode = op->acc_mode;
3035 if (*opened & FILE_CREATED) {
3036 WARN_ON(!(open_flag & O_CREAT));
3037 fsnotify_create(dir, dentry);
3038 acc_mode = 0;
3039 }
3040 error = may_open(&file->f_path, acc_mode, open_flag);
3041 if (WARN_ON(error > 0))
3042 error = -EINVAL;
3043 } else if (error > 0) {
30d90494 3044 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2675a4eb 3045 error = -EIO;
03da633a 3046 } else {
384f26e2
AV
3047 if (file->f_path.dentry) {
3048 dput(dentry);
3049 dentry = file->f_path.dentry;
03da633a 3050 }
384f26e2
AV
3051 if (*opened & FILE_CREATED)
3052 fsnotify_create(dir, dentry);
a01e718f
AV
3053 if (unlikely(d_is_negative(dentry))) {
3054 error = -ENOENT;
3055 } else {
3056 path->dentry = dentry;
3057 path->mnt = nd->path.mnt;
3058 return 1;
3059 }
62b2ce96 3060 }
d18e9008 3061 }
d18e9008 3062 dput(dentry);
2675a4eb 3063 return error;
d18e9008
MS
3064}
3065
d58ffd35 3066/*
1acf0af9 3067 * Look up and maybe create and open the last component.
d58ffd35
MS
3068 *
3069 * Must be called with i_mutex held on parent.
3070 *
1acf0af9
DH
3071 * Returns 0 if the file was successfully atomically created (if necessary) and
3072 * opened. In this case the file will be returned attached to @file.
3073 *
3074 * Returns 1 if the file was not completely opened at this time, though lookups
3075 * and creations will have been performed and the dentry returned in @path will
3076 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
3077 * specified then a negative dentry may be returned.
3078 *
3079 * An error code is returned otherwise.
3080 *
3081 * FILE_CREATE will be set in @*opened if the dentry was created and will be
3082 * cleared otherwise prior to returning.
d58ffd35 3083 */
2675a4eb
AV
3084static int lookup_open(struct nameidata *nd, struct path *path,
3085 struct file *file,
3086 const struct open_flags *op,
64894cf8 3087 bool got_write, int *opened)
d58ffd35
MS
3088{
3089 struct dentry *dir = nd->path.dentry;
54ef4872 3090 struct inode *dir_inode = dir->d_inode;
1643b43f 3091 int open_flag = op->open_flag;
d58ffd35 3092 struct dentry *dentry;
1643b43f 3093 int error, create_error = 0;
1643b43f 3094 umode_t mode = op->mode;
6fbd0714 3095 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
d58ffd35 3096
ce8644fc
AV
3097 if (unlikely(IS_DEADDIR(dir_inode)))
3098 return -ENOENT;
d58ffd35 3099
47237687 3100 *opened &= ~FILE_CREATED;
6fbd0714
AV
3101 dentry = d_lookup(dir, &nd->last);
3102 for (;;) {
3103 if (!dentry) {
3104 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3105 if (IS_ERR(dentry))
3106 return PTR_ERR(dentry);
3107 }
3108 if (d_in_lookup(dentry))
3109 break;
d58ffd35 3110
6fbd0714
AV
3111 error = d_revalidate(dentry, nd->flags);
3112 if (likely(error > 0))
3113 break;
3114 if (error)
3115 goto out_dput;
3116 d_invalidate(dentry);
3117 dput(dentry);
3118 dentry = NULL;
3119 }
3120 if (dentry->d_inode) {
6c51e513 3121 /* Cached positive dentry: will open in f_op->open */
d18e9008 3122 goto out_no_open;
6c51e513 3123 }
d18e9008 3124
1643b43f
AV
3125 /*
3126 * Checking write permission is tricky, bacuse we don't know if we are
3127 * going to actually need it: O_CREAT opens should work as long as the
3128 * file exists. But checking existence breaks atomicity. The trick is
3129 * to check access and if not granted clear O_CREAT from the flags.
3130 *
3131 * Another problem is returing the "right" error value (e.g. for an
3132 * O_EXCL open we want to return EEXIST not EROFS).
3133 */
3134 if (open_flag & O_CREAT) {
3135 if (!IS_POSIXACL(dir->d_inode))
3136 mode &= ~current_umask();
3137 if (unlikely(!got_write)) {
3138 create_error = -EROFS;
3139 open_flag &= ~O_CREAT;
3140 if (open_flag & (O_EXCL | O_TRUNC))
3141 goto no_open;
3142 /* No side effects, safe to clear O_CREAT */
3143 } else {
3144 create_error = may_o_create(&nd->path, dentry, mode);
3145 if (create_error) {
3146 open_flag &= ~O_CREAT;
3147 if (open_flag & O_EXCL)
3148 goto no_open;
3149 }
3150 }
3151 } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3152 unlikely(!got_write)) {
3153 /*
3154 * No O_CREATE -> atomicity not a requirement -> fall
3155 * back to lookup + open
3156 */
3157 goto no_open;
d18e9008
MS
3158 }
3159
6ac08709 3160 if (dir_inode->i_op->atomic_open) {
1643b43f
AV
3161 error = atomic_open(nd, dentry, path, file, op, open_flag,
3162 mode, opened);
3163 if (unlikely(error == -ENOENT) && create_error)
3164 error = create_error;
3165 return error;
d18e9008 3166 }
54ef4872 3167
1643b43f 3168no_open:
6fbd0714 3169 if (d_in_lookup(dentry)) {
12fa5e24
AV
3170 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3171 nd->flags);
6fbd0714 3172 d_lookup_done(dentry);
12fa5e24
AV
3173 if (unlikely(res)) {
3174 if (IS_ERR(res)) {
3175 error = PTR_ERR(res);
3176 goto out_dput;
3177 }
3178 dput(dentry);
3179 dentry = res;
3180 }
54ef4872
MS
3181 }
3182
d58ffd35 3183 /* Negative dentry, just create the file */
1643b43f 3184 if (!dentry->d_inode && (open_flag & O_CREAT)) {
47237687 3185 *opened |= FILE_CREATED;
ce8644fc 3186 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
ce8644fc
AV
3187 if (!dir_inode->i_op->create) {
3188 error = -EACCES;
d58ffd35 3189 goto out_dput;
ce8644fc
AV
3190 }
3191 error = dir_inode->i_op->create(dir_inode, dentry, mode,
1643b43f 3192 open_flag & O_EXCL);
d58ffd35
MS
3193 if (error)
3194 goto out_dput;
ce8644fc 3195 fsnotify_create(dir_inode, dentry);
d58ffd35 3196 }
1643b43f
AV
3197 if (unlikely(create_error) && !dentry->d_inode) {
3198 error = create_error;
3199 goto out_dput;
d58ffd35 3200 }
d18e9008 3201out_no_open:
d58ffd35
MS
3202 path->dentry = dentry;
3203 path->mnt = nd->path.mnt;
2675a4eb 3204 return 1;
d58ffd35
MS
3205
3206out_dput:
3207 dput(dentry);
2675a4eb 3208 return error;
d58ffd35
MS
3209}
3210
31e6b01f 3211/*
fe2d35ff 3212 * Handle the last step of open()
31e6b01f 3213 */
896475d5 3214static int do_last(struct nameidata *nd,
2675a4eb 3215 struct file *file, const struct open_flags *op,
76ae2a5a 3216 int *opened)
fb1cc555 3217{
a1e28038 3218 struct dentry *dir = nd->path.dentry;
ca344a89 3219 int open_flag = op->open_flag;
77d660a8 3220 bool will_truncate = (open_flag & O_TRUNC) != 0;
64894cf8 3221 bool got_write = false;
bcda7652 3222 int acc_mode = op->acc_mode;
254cf582 3223 unsigned seq;
a1eb3315 3224 struct inode *inode;
896475d5 3225 struct path path;
16c2cd71 3226 int error;
1f36f774 3227
c3e380b0
AV
3228 nd->flags &= ~LOOKUP_PARENT;
3229 nd->flags |= op->intent;
3230
bc77daa7 3231 if (nd->last_type != LAST_NORM) {
fe2d35ff 3232 error = handle_dots(nd, nd->last_type);
deb106c6 3233 if (unlikely(error))
2675a4eb 3234 return error;
e83db167 3235 goto finish_open;
1f36f774 3236 }
67ee3ad2 3237
ca344a89 3238 if (!(open_flag & O_CREAT)) {
fe2d35ff
AV
3239 if (nd->last.name[nd->last.len])
3240 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3241 /* we _can_ be in RCU mode here */
254cf582 3242 error = lookup_fast(nd, &path, &inode, &seq);
e9742b53 3243 if (likely(error > 0))
71574865
MS
3244 goto finish_lookup;
3245
3246 if (error < 0)
deb106c6 3247 return error;
71574865
MS
3248
3249 BUG_ON(nd->inode != dir->d_inode);
6583fe22 3250 BUG_ON(nd->flags & LOOKUP_RCU);
b6183df7
MS
3251 } else {
3252 /* create side of things */
3253 /*
3254 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3255 * has been cleared when we got to the last component we are
3256 * about to look up
3257 */
3258 error = complete_walk(nd);
e8bb73df 3259 if (error)
2675a4eb 3260 return error;
fe2d35ff 3261
76ae2a5a 3262 audit_inode(nd->name, dir, LOOKUP_PARENT);
b6183df7 3263 /* trailing slashes? */
deb106c6
AV
3264 if (unlikely(nd->last.name[nd->last.len]))
3265 return -EISDIR;
b6183df7 3266 }
a2c36b45 3267
9cf843e3 3268 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
64894cf8
AV
3269 error = mnt_want_write(nd->path.mnt);
3270 if (!error)
3271 got_write = true;
3272 /*
3273 * do _not_ fail yet - we might not need that or fail with
3274 * a different error; let lookup_open() decide; we'll be
3275 * dropping this one anyway.
3276 */
3277 }
9cf843e3
AV
3278 if (open_flag & O_CREAT)
3279 inode_lock(dir->d_inode);
3280 else
3281 inode_lock_shared(dir->d_inode);
896475d5 3282 error = lookup_open(nd, &path, file, op, got_write, opened);
9cf843e3
AV
3283 if (open_flag & O_CREAT)
3284 inode_unlock(dir->d_inode);
3285 else
3286 inode_unlock_shared(dir->d_inode);
a1e28038 3287
2675a4eb
AV
3288 if (error <= 0) {
3289 if (error)
d18e9008
MS
3290 goto out;
3291
47237687 3292 if ((*opened & FILE_CREATED) ||
496ad9aa 3293 !S_ISREG(file_inode(file)->i_mode))
77d660a8 3294 will_truncate = false;
d18e9008 3295
76ae2a5a 3296 audit_inode(nd->name, file->f_path.dentry, 0);
d18e9008
MS
3297 goto opened;
3298 }
fb1cc555 3299
47237687 3300 if (*opened & FILE_CREATED) {
9b44f1b3 3301 /* Don't check for write permission, don't truncate */
ca344a89 3302 open_flag &= ~O_TRUNC;
77d660a8 3303 will_truncate = false;
62fb4a15 3304 acc_mode = 0;
896475d5 3305 path_to_nameidata(&path, nd);
e83db167 3306 goto finish_open_created;
fb1cc555
AV
3307 }
3308
d18e9008
MS
3309 /*
3310 * If atomic_open() acquired write access it is dropped now due to
3311 * possible mount and symlink following (this might be optimized away if
3312 * necessary...)
3313 */
64894cf8 3314 if (got_write) {
d18e9008 3315 mnt_drop_write(nd->path.mnt);
64894cf8 3316 got_write = false;
d18e9008
MS
3317 }
3318
e6ec03a2
AV
3319 error = follow_managed(&path, nd);
3320 if (unlikely(error < 0))
3321 return error;
3322
6583fe22
AV
3323 if (unlikely(d_is_negative(path.dentry))) {
3324 path_to_nameidata(&path, nd);
3325 return -ENOENT;
3326 }
3327
3328 /*
3329 * create/update audit record if it already exists.
3330 */
3331 audit_inode(nd->name, path.dentry, 0);
3332
deb106c6
AV
3333 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3334 path_to_nameidata(&path, nd);
3335 return -EEXIST;
3336 }
fb1cc555 3337
254cf582 3338 seq = 0; /* out of RCU mode, so the value doesn't matter */
d4565649 3339 inode = d_backing_inode(path.dentry);
766c4cbf 3340finish_lookup:
8f64fb1c 3341 error = step_into(nd, &path, 0, inode, seq);
deb106c6 3342 if (unlikely(error))
d63ff28f 3343 return error;
bc77daa7 3344finish_open:
8f64fb1c 3345 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
a3fbbde7 3346 error = complete_walk(nd);
fac7d191 3347 if (error)
2675a4eb 3348 return error;
76ae2a5a 3349 audit_inode(nd->name, nd->path.dentry, 0);
fb1cc555 3350 error = -EISDIR;
44b1d530 3351 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
2675a4eb 3352 goto out;
af2f5542 3353 error = -ENOTDIR;
44b1d530 3354 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
2675a4eb 3355 goto out;
4bbcbd3b 3356 if (!d_is_reg(nd->path.dentry))
77d660a8 3357 will_truncate = false;
6c0d46c4 3358
0f9d1a10
AV
3359 if (will_truncate) {
3360 error = mnt_want_write(nd->path.mnt);
3361 if (error)
2675a4eb 3362 goto out;
64894cf8 3363 got_write = true;
0f9d1a10 3364 }
e83db167 3365finish_open_created:
6ac08709
AV
3366 error = may_open(&nd->path, acc_mode, open_flag);
3367 if (error)
3368 goto out;
4aa7c634
MS
3369 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3370 error = vfs_open(&nd->path, file, current_cred());
fac7d191 3371 if (error)
015c3bbc 3372 goto out;
fac7d191 3373 *opened |= FILE_OPENED;
a8277b9b 3374opened:
af04fadc
AV
3375 error = open_check_o_direct(file);
3376 if (!error)
3377 error = ima_file_check(file, op->acc_mode, *opened);
fe9ec829 3378 if (!error && will_truncate)
2675a4eb 3379 error = handle_truncate(file);
ca344a89 3380out:
fe9ec829
AV
3381 if (unlikely(error) && (*opened & FILE_OPENED))
3382 fput(file);
c80567c8
AV
3383 if (unlikely(error > 0)) {
3384 WARN_ON(1);
3385 error = -EINVAL;
3386 }
64894cf8 3387 if (got_write)
0f9d1a10 3388 mnt_drop_write(nd->path.mnt);
2675a4eb 3389 return error;
fb1cc555
AV
3390}
3391
af7bd4dc
AG
3392struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag)
3393{
af7bd4dc
AG
3394 struct dentry *child = NULL;
3395 struct inode *dir = dentry->d_inode;
3396 struct inode *inode;
3397 int error;
3398
3399 /* we want directory to be writable */
3400 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3401 if (error)
3402 goto out_err;
3403 error = -EOPNOTSUPP;
3404 if (!dir->i_op->tmpfile)
3405 goto out_err;
3406 error = -ENOMEM;
cdf01226 3407 child = d_alloc(dentry, &slash_name);
af7bd4dc
AG
3408 if (unlikely(!child))
3409 goto out_err;
3410 error = dir->i_op->tmpfile(dir, child, mode);
3411 if (error)
3412 goto out_err;
3413 error = -ENOENT;
3414 inode = child->d_inode;
3415 if (unlikely(!inode))
3416 goto out_err;
3417 if (!(open_flag & O_EXCL)) {
3418 spin_lock(&inode->i_lock);
3419 inode->i_state |= I_LINKABLE;
3420 spin_unlock(&inode->i_lock);
3421 }
3422 return child;
3423
3424out_err:
3425 dput(child);
3426 return ERR_PTR(error);
3427}
3428EXPORT_SYMBOL(vfs_tmpfile);
3429
c8a53ee5 3430static int do_tmpfile(struct nameidata *nd, unsigned flags,
60545d0d
AV
3431 const struct open_flags *op,
3432 struct file *file, int *opened)
3433{
625b6d10 3434 struct dentry *child;
625b6d10 3435 struct path path;
c8a53ee5 3436 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
60545d0d
AV
3437 if (unlikely(error))
3438 return error;
625b6d10 3439 error = mnt_want_write(path.mnt);
60545d0d
AV
3440 if (unlikely(error))
3441 goto out;
af7bd4dc
AG
3442 child = vfs_tmpfile(path.dentry, op->mode, op->open_flag);
3443 error = PTR_ERR(child);
684e73be 3444 if (IS_ERR(child))
60545d0d 3445 goto out2;
625b6d10
AV
3446 dput(path.dentry);
3447 path.dentry = child;
c8a53ee5 3448 audit_inode(nd->name, child, 0);
69a91c23 3449 /* Don't check for other permissions, the inode was just created */
62fb4a15 3450 error = may_open(&path, 0, op->open_flag);
60545d0d
AV
3451 if (error)
3452 goto out2;
625b6d10
AV
3453 file->f_path.mnt = path.mnt;
3454 error = finish_open(file, child, NULL, opened);
60545d0d
AV
3455 if (error)
3456 goto out2;
af04fadc
AV
3457 error = open_check_o_direct(file);
3458 if (error)
3459 fput(file);
60545d0d 3460out2:
625b6d10 3461 mnt_drop_write(path.mnt);
60545d0d 3462out:
625b6d10 3463 path_put(&path);
60545d0d
AV
3464 return error;
3465}
3466
6ac08709
AV
3467static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3468{
3469 struct path path;
3470 int error = path_lookupat(nd, flags, &path);
3471 if (!error) {
3472 audit_inode(nd->name, path.dentry, 0);
3473 error = vfs_open(&path, file, current_cred());
3474 path_put(&path);
3475 }
3476 return error;
3477}
3478
c8a53ee5
AV
3479static struct file *path_openat(struct nameidata *nd,
3480 const struct open_flags *op, unsigned flags)
1da177e4 3481{
368ee9ba 3482 const char *s;
30d90494 3483 struct file *file;
47237687 3484 int opened = 0;
13aab428 3485 int error;
31e6b01f 3486
30d90494 3487 file = get_empty_filp();
1afc99be
AV
3488 if (IS_ERR(file))
3489 return file;
31e6b01f 3490
30d90494 3491 file->f_flags = op->open_flag;
31e6b01f 3492
bb458c64 3493 if (unlikely(file->f_flags & __O_TMPFILE)) {
c8a53ee5 3494 error = do_tmpfile(nd, flags, op, file, &opened);
f15133df 3495 goto out2;
60545d0d
AV
3496 }
3497
6ac08709
AV
3498 if (unlikely(file->f_flags & O_PATH)) {
3499 error = do_o_path(nd, flags, file);
3500 if (!error)
3501 opened |= FILE_OPENED;
3502 goto out2;
3503 }
3504
c8a53ee5 3505 s = path_init(nd, flags);
368ee9ba
AV
3506 if (IS_ERR(s)) {
3507 put_filp(file);
3508 return ERR_CAST(s);
3509 }
3bdba28b 3510 while (!(error = link_path_walk(s, nd)) &&
76ae2a5a 3511 (error = do_last(nd, file, op, &opened)) > 0) {
73d049a4 3512 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3bdba28b
AV
3513 s = trailing_symlink(nd);
3514 if (IS_ERR(s)) {
3515 error = PTR_ERR(s);
2675a4eb 3516 break;
3bdba28b 3517 }
806b681c 3518 }
deb106c6 3519 terminate_walk(nd);
f15133df 3520out2:
2675a4eb
AV
3521 if (!(opened & FILE_OPENED)) {
3522 BUG_ON(!error);
30d90494 3523 put_filp(file);
16b1c1cd 3524 }
2675a4eb
AV
3525 if (unlikely(error)) {
3526 if (error == -EOPENSTALE) {
3527 if (flags & LOOKUP_RCU)
3528 error = -ECHILD;
3529 else
3530 error = -ESTALE;
3531 }
3532 file = ERR_PTR(error);
3533 }
3534 return file;
1da177e4
LT
3535}
3536
669abf4e 3537struct file *do_filp_open(int dfd, struct filename *pathname,
f9652e10 3538 const struct open_flags *op)
13aab428 3539{
9883d185 3540 struct nameidata nd;
f9652e10 3541 int flags = op->lookup_flags;
13aab428
AV
3542 struct file *filp;
3543
9883d185 3544 set_nameidata(&nd, dfd, pathname);
c8a53ee5 3545 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
13aab428 3546 if (unlikely(filp == ERR_PTR(-ECHILD)))
c8a53ee5 3547 filp = path_openat(&nd, op, flags);
13aab428 3548 if (unlikely(filp == ERR_PTR(-ESTALE)))
c8a53ee5 3549 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
9883d185 3550 restore_nameidata();
13aab428
AV
3551 return filp;
3552}
3553
73d049a4 3554struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
f9652e10 3555 const char *name, const struct open_flags *op)
73d049a4 3556{
9883d185 3557 struct nameidata nd;
73d049a4 3558 struct file *file;
51689104 3559 struct filename *filename;
f9652e10 3560 int flags = op->lookup_flags | LOOKUP_ROOT;
73d049a4
AV
3561
3562 nd.root.mnt = mnt;
3563 nd.root.dentry = dentry;
3564
b18825a7 3565 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
73d049a4
AV
3566 return ERR_PTR(-ELOOP);
3567
51689104 3568 filename = getname_kernel(name);
a1c83681 3569 if (IS_ERR(filename))
51689104
PM
3570 return ERR_CAST(filename);
3571
9883d185 3572 set_nameidata(&nd, -1, filename);
c8a53ee5 3573 file = path_openat(&nd, op, flags | LOOKUP_RCU);
73d049a4 3574 if (unlikely(file == ERR_PTR(-ECHILD)))
c8a53ee5 3575 file = path_openat(&nd, op, flags);
73d049a4 3576 if (unlikely(file == ERR_PTR(-ESTALE)))
c8a53ee5 3577 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
9883d185 3578 restore_nameidata();
51689104 3579 putname(filename);
73d049a4
AV
3580 return file;
3581}
3582
fa14a0b8 3583static struct dentry *filename_create(int dfd, struct filename *name,
1ac12b4b 3584 struct path *path, unsigned int lookup_flags)
1da177e4 3585{
c663e5d8 3586 struct dentry *dentry = ERR_PTR(-EEXIST);
391172c4
AV
3587 struct qstr last;
3588 int type;
c30dabfe 3589 int err2;
1ac12b4b
JL
3590 int error;
3591 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3592
3593 /*
3594 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3595 * other flags passed in are ignored!
3596 */
3597 lookup_flags &= LOOKUP_REVAL;
3598
5c31b6ce
AV
3599 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3600 if (IS_ERR(name))
3601 return ERR_CAST(name);
1da177e4 3602
c663e5d8
CH
3603 /*
3604 * Yucky last component or no last component at all?
3605 * (foo/., foo/.., /////)
3606 */
5c31b6ce 3607 if (unlikely(type != LAST_NORM))
ed75e95d 3608 goto out;
c663e5d8 3609
c30dabfe 3610 /* don't fail immediately if it's r/o, at least try to report other errors */
391172c4 3611 err2 = mnt_want_write(path->mnt);
c663e5d8
CH
3612 /*
3613 * Do the final lookup.
3614 */
391172c4 3615 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
5955102c 3616 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
391172c4 3617 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
1da177e4 3618 if (IS_ERR(dentry))
a8104a9f 3619 goto unlock;
c663e5d8 3620
a8104a9f 3621 error = -EEXIST;
b18825a7 3622 if (d_is_positive(dentry))
a8104a9f 3623 goto fail;
b18825a7 3624
c663e5d8
CH
3625 /*
3626 * Special case - lookup gave negative, but... we had foo/bar/
3627 * From the vfs_mknod() POV we just have a negative dentry -
3628 * all is fine. Let's be bastards - you had / on the end, you've
3629 * been asking for (non-existent) directory. -ENOENT for you.
3630 */
391172c4 3631 if (unlikely(!is_dir && last.name[last.len])) {
a8104a9f 3632 error = -ENOENT;
ed75e95d 3633 goto fail;
e9baf6e5 3634 }
c30dabfe
JK
3635 if (unlikely(err2)) {
3636 error = err2;
a8104a9f 3637 goto fail;
c30dabfe 3638 }
181c37b6 3639 putname(name);
1da177e4 3640 return dentry;
1da177e4 3641fail:
a8104a9f
AV
3642 dput(dentry);
3643 dentry = ERR_PTR(error);
3644unlock:
5955102c 3645 inode_unlock(path->dentry->d_inode);
c30dabfe 3646 if (!err2)
391172c4 3647 mnt_drop_write(path->mnt);
ed75e95d 3648out:
391172c4 3649 path_put(path);
181c37b6 3650 putname(name);
1da177e4
LT
3651 return dentry;
3652}
fa14a0b8
AV
3653
3654struct dentry *kern_path_create(int dfd, const char *pathname,
3655 struct path *path, unsigned int lookup_flags)
3656{
181c37b6
AV
3657 return filename_create(dfd, getname_kernel(pathname),
3658 path, lookup_flags);
fa14a0b8 3659}
dae6ad8f
AV
3660EXPORT_SYMBOL(kern_path_create);
3661
921a1650
AV
3662void done_path_create(struct path *path, struct dentry *dentry)
3663{
3664 dput(dentry);
5955102c 3665 inode_unlock(path->dentry->d_inode);
a8104a9f 3666 mnt_drop_write(path->mnt);
921a1650
AV
3667 path_put(path);
3668}
3669EXPORT_SYMBOL(done_path_create);
3670
520ae687 3671inline struct dentry *user_path_create(int dfd, const char __user *pathname,
1ac12b4b 3672 struct path *path, unsigned int lookup_flags)
dae6ad8f 3673{
181c37b6 3674 return filename_create(dfd, getname(pathname), path, lookup_flags);
dae6ad8f
AV
3675}
3676EXPORT_SYMBOL(user_path_create);
3677
1a67aafb 3678int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
1da177e4 3679{
a95164d9 3680 int error = may_create(dir, dentry);
1da177e4
LT
3681
3682 if (error)
3683 return error;
3684
55956b59
EB
3685 if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
3686 !ns_capable(dentry->d_sb->s_user_ns, CAP_MKNOD))
1da177e4
LT
3687 return -EPERM;
3688
acfa4380 3689 if (!dir->i_op->mknod)
1da177e4
LT
3690 return -EPERM;
3691
08ce5f16
SH
3692 error = devcgroup_inode_mknod(mode, dev);
3693 if (error)
3694 return error;
3695
1da177e4
LT
3696 error = security_inode_mknod(dir, dentry, mode, dev);
3697 if (error)
3698 return error;
3699
1da177e4 3700 error = dir->i_op->mknod(dir, dentry, mode, dev);
a74574aa 3701 if (!error)
f38aa942 3702 fsnotify_create(dir, dentry);
1da177e4
LT
3703 return error;
3704}
4d359507 3705EXPORT_SYMBOL(vfs_mknod);
1da177e4 3706
f69aac00 3707static int may_mknod(umode_t mode)
463c3197
DH
3708{
3709 switch (mode & S_IFMT) {
3710 case S_IFREG:
3711 case S_IFCHR:
3712 case S_IFBLK:
3713 case S_IFIFO:
3714 case S_IFSOCK:
3715 case 0: /* zero mode translates to S_IFREG */
3716 return 0;
3717 case S_IFDIR:
3718 return -EPERM;
3719 default:
3720 return -EINVAL;
3721 }
3722}
3723
87c4e192
DB
3724long do_mknodat(int dfd, const char __user *filename, umode_t mode,
3725 unsigned int dev)
1da177e4 3726{
2ad94ae6 3727 struct dentry *dentry;
dae6ad8f
AV
3728 struct path path;
3729 int error;
972567f1 3730 unsigned int lookup_flags = 0;
1da177e4 3731
8e4bfca1
AV
3732 error = may_mknod(mode);
3733 if (error)
3734 return error;
972567f1
JL
3735retry:
3736 dentry = user_path_create(dfd, filename, &path, lookup_flags);
dae6ad8f
AV
3737 if (IS_ERR(dentry))
3738 return PTR_ERR(dentry);
2ad94ae6 3739
dae6ad8f 3740 if (!IS_POSIXACL(path.dentry->d_inode))
ce3b0f8d 3741 mode &= ~current_umask();
dae6ad8f 3742 error = security_path_mknod(&path, dentry, mode, dev);
be6d3e56 3743 if (error)
a8104a9f 3744 goto out;
463c3197 3745 switch (mode & S_IFMT) {
1da177e4 3746 case 0: case S_IFREG:
312b63fb 3747 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
05d1a717
MZ
3748 if (!error)
3749 ima_post_path_mknod(dentry);
1da177e4
LT
3750 break;
3751 case S_IFCHR: case S_IFBLK:
dae6ad8f 3752 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
1da177e4
LT
3753 new_decode_dev(dev));
3754 break;
3755 case S_IFIFO: case S_IFSOCK:
dae6ad8f 3756 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
1da177e4 3757 break;
1da177e4 3758 }
a8104a9f 3759out:
921a1650 3760 done_path_create(&path, dentry);
972567f1
JL
3761 if (retry_estale(error, lookup_flags)) {
3762 lookup_flags |= LOOKUP_REVAL;
3763 goto retry;
3764 }
1da177e4
LT
3765 return error;
3766}
3767
87c4e192
DB
3768SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3769 unsigned int, dev)
3770{
3771 return do_mknodat(dfd, filename, mode, dev);
3772}
3773
8208a22b 3774SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
5590ff0d 3775{
87c4e192 3776 return do_mknodat(AT_FDCWD, filename, mode, dev);
5590ff0d
UD
3777}
3778
18bb1db3 3779int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1da177e4 3780{
a95164d9 3781 int error = may_create(dir, dentry);
8de52778 3782 unsigned max_links = dir->i_sb->s_max_links;
1da177e4
LT
3783
3784 if (error)
3785 return error;
3786
acfa4380 3787 if (!dir->i_op->mkdir)
1da177e4
LT
3788 return -EPERM;
3789
3790 mode &= (S_IRWXUGO|S_ISVTX);
3791 error = security_inode_mkdir(dir, dentry, mode);
3792 if (error)
3793 return error;
3794
8de52778
AV
3795 if (max_links && dir->i_nlink >= max_links)
3796 return -EMLINK;
3797
1da177e4 3798 error = dir->i_op->mkdir(dir, dentry, mode);
a74574aa 3799 if (!error)
f38aa942 3800 fsnotify_mkdir(dir, dentry);
1da177e4
LT
3801 return error;
3802}
4d359507 3803EXPORT_SYMBOL(vfs_mkdir);
1da177e4 3804
0101db7a 3805long do_mkdirat(int dfd, const char __user *pathname, umode_t mode)
1da177e4 3806{
6902d925 3807 struct dentry *dentry;
dae6ad8f
AV
3808 struct path path;
3809 int error;
b76d8b82 3810 unsigned int lookup_flags = LOOKUP_DIRECTORY;
1da177e4 3811
b76d8b82
JL
3812retry:
3813 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
6902d925 3814 if (IS_ERR(dentry))
dae6ad8f 3815 return PTR_ERR(dentry);
1da177e4 3816
dae6ad8f 3817 if (!IS_POSIXACL(path.dentry->d_inode))
ce3b0f8d 3818 mode &= ~current_umask();
dae6ad8f 3819 error = security_path_mkdir(&path, dentry, mode);
a8104a9f
AV
3820 if (!error)
3821 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
921a1650 3822 done_path_create(&path, dentry);
b76d8b82
JL
3823 if (retry_estale(error, lookup_flags)) {
3824 lookup_flags |= LOOKUP_REVAL;
3825 goto retry;
3826 }
1da177e4
LT
3827 return error;
3828}
3829
0101db7a
DB
3830SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3831{
3832 return do_mkdirat(dfd, pathname, mode);
3833}
3834
a218d0fd 3835SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
5590ff0d 3836{
0101db7a 3837 return do_mkdirat(AT_FDCWD, pathname, mode);
5590ff0d
UD
3838}
3839
1da177e4
LT
3840int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3841{
3842 int error = may_delete(dir, dentry, 1);
3843
3844 if (error)
3845 return error;
3846
acfa4380 3847 if (!dir->i_op->rmdir)
1da177e4
LT
3848 return -EPERM;
3849
1d2ef590 3850 dget(dentry);
5955102c 3851 inode_lock(dentry->d_inode);
912dbc15
SW
3852
3853 error = -EBUSY;
7af1364f 3854 if (is_local_mountpoint(dentry))
912dbc15
SW
3855 goto out;
3856
3857 error = security_inode_rmdir(dir, dentry);
3858 if (error)
3859 goto out;
3860
3861 error = dir->i_op->rmdir(dir, dentry);
3862 if (error)
3863 goto out;
3864
8767712f 3865 shrink_dcache_parent(dentry);
912dbc15
SW
3866 dentry->d_inode->i_flags |= S_DEAD;
3867 dont_mount(dentry);
8ed936b5 3868 detach_mounts(dentry);
912dbc15
SW
3869
3870out:
5955102c 3871 inode_unlock(dentry->d_inode);
1d2ef590 3872 dput(dentry);
912dbc15 3873 if (!error)
1da177e4 3874 d_delete(dentry);
1da177e4
LT
3875 return error;
3876}
4d359507 3877EXPORT_SYMBOL(vfs_rmdir);
1da177e4 3878
f459dffa 3879long do_rmdir(int dfd, const char __user *pathname)
1da177e4
LT
3880{
3881 int error = 0;
91a27b2a 3882 struct filename *name;
1da177e4 3883 struct dentry *dentry;
f5beed75
AV
3884 struct path path;
3885 struct qstr last;
3886 int type;
c6ee9206
JL
3887 unsigned int lookup_flags = 0;
3888retry:
c1d4dd27
AV
3889 name = filename_parentat(dfd, getname(pathname), lookup_flags,
3890 &path, &last, &type);
91a27b2a
JL
3891 if (IS_ERR(name))
3892 return PTR_ERR(name);
1da177e4 3893
f5beed75 3894 switch (type) {
0612d9fb
OH
3895 case LAST_DOTDOT:
3896 error = -ENOTEMPTY;
3897 goto exit1;
3898 case LAST_DOT:
3899 error = -EINVAL;
3900 goto exit1;
3901 case LAST_ROOT:
3902 error = -EBUSY;
3903 goto exit1;
1da177e4 3904 }
0612d9fb 3905
f5beed75 3906 error = mnt_want_write(path.mnt);
c30dabfe
JK
3907 if (error)
3908 goto exit1;
0612d9fb 3909
5955102c 3910 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
f5beed75 3911 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
1da177e4 3912 error = PTR_ERR(dentry);
6902d925
DH
3913 if (IS_ERR(dentry))
3914 goto exit2;
e6bc45d6
TT
3915 if (!dentry->d_inode) {
3916 error = -ENOENT;
3917 goto exit3;
3918 }
f5beed75 3919 error = security_path_rmdir(&path, dentry);
be6d3e56 3920 if (error)
c30dabfe 3921 goto exit3;
f5beed75 3922 error = vfs_rmdir(path.dentry->d_inode, dentry);
0622753b 3923exit3:
6902d925
DH
3924 dput(dentry);
3925exit2:
5955102c 3926 inode_unlock(path.dentry->d_inode);
f5beed75 3927 mnt_drop_write(path.mnt);
1da177e4 3928exit1:
f5beed75 3929 path_put(&path);
1da177e4 3930 putname(name);
c6ee9206
JL
3931 if (retry_estale(error, lookup_flags)) {
3932 lookup_flags |= LOOKUP_REVAL;
3933 goto retry;
3934 }
1da177e4
LT
3935 return error;
3936}
3937
3cdad428 3938SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
5590ff0d
UD
3939{
3940 return do_rmdir(AT_FDCWD, pathname);
3941}
3942
b21996e3
BF
3943/**
3944 * vfs_unlink - unlink a filesystem object
3945 * @dir: parent directory
3946 * @dentry: victim
3947 * @delegated_inode: returns victim inode, if the inode is delegated.
3948 *
3949 * The caller must hold dir->i_mutex.
3950 *
3951 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3952 * return a reference to the inode in delegated_inode. The caller
3953 * should then break the delegation on that inode and retry. Because
3954 * breaking a delegation may take a long time, the caller should drop
3955 * dir->i_mutex before doing so.
3956 *
3957 * Alternatively, a caller may pass NULL for delegated_inode. This may
3958 * be appropriate for callers that expect the underlying filesystem not
3959 * to be NFS exported.
3960 */
3961int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
1da177e4 3962{
9accbb97 3963 struct inode *target = dentry->d_inode;
1da177e4
LT
3964 int error = may_delete(dir, dentry, 0);
3965
3966 if (error)
3967 return error;
3968
acfa4380 3969 if (!dir->i_op->unlink)
1da177e4
LT
3970 return -EPERM;
3971
5955102c 3972 inode_lock(target);
8ed936b5 3973 if (is_local_mountpoint(dentry))
1da177e4
LT
3974 error = -EBUSY;
3975 else {
3976 error = security_inode_unlink(dir, dentry);
bec1052e 3977 if (!error) {
5a14696c
BF
3978 error = try_break_deleg(target, delegated_inode);
3979 if (error)
b21996e3 3980 goto out;
1da177e4 3981 error = dir->i_op->unlink(dir, dentry);
8ed936b5 3982 if (!error) {
d83c49f3 3983 dont_mount(dentry);
8ed936b5
EB
3984 detach_mounts(dentry);
3985 }
bec1052e 3986 }
1da177e4 3987 }
b21996e3 3988out:
5955102c 3989 inode_unlock(target);
1da177e4
LT
3990
3991 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3992 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
9accbb97 3993 fsnotify_link_count(target);
e234f35c 3994 d_delete(dentry);
1da177e4 3995 }
0eeca283 3996
1da177e4
LT
3997 return error;
3998}
4d359507 3999EXPORT_SYMBOL(vfs_unlink);
1da177e4
LT
4000
4001/*
4002 * Make sure that the actual truncation of the file will occur outside its
1b1dcc1b 4003 * directory's i_mutex. Truncate can take a long time if there is a lot of
1da177e4
LT
4004 * writeout happening, and we don't want to prevent access to the directory
4005 * while waiting on the I/O.
4006 */
da2f1362 4007long do_unlinkat(int dfd, struct filename *name)
1da177e4 4008{
2ad94ae6 4009 int error;
1da177e4 4010 struct dentry *dentry;
f5beed75
AV
4011 struct path path;
4012 struct qstr last;
4013 int type;
1da177e4 4014 struct inode *inode = NULL;
b21996e3 4015 struct inode *delegated_inode = NULL;
5d18f813
JL
4016 unsigned int lookup_flags = 0;
4017retry:
da2f1362 4018 name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
91a27b2a
JL
4019 if (IS_ERR(name))
4020 return PTR_ERR(name);
2ad94ae6 4021
1da177e4 4022 error = -EISDIR;
f5beed75 4023 if (type != LAST_NORM)
1da177e4 4024 goto exit1;
0612d9fb 4025
f5beed75 4026 error = mnt_want_write(path.mnt);
c30dabfe
JK
4027 if (error)
4028 goto exit1;
b21996e3 4029retry_deleg:
5955102c 4030 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
f5beed75 4031 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
1da177e4
LT
4032 error = PTR_ERR(dentry);
4033 if (!IS_ERR(dentry)) {
4034 /* Why not before? Because we want correct error value */
f5beed75 4035 if (last.name[last.len])
50338b88 4036 goto slashes;
1da177e4 4037 inode = dentry->d_inode;
b18825a7 4038 if (d_is_negative(dentry))
e6bc45d6
TT
4039 goto slashes;
4040 ihold(inode);
f5beed75 4041 error = security_path_unlink(&path, dentry);
be6d3e56 4042 if (error)
c30dabfe 4043 goto exit2;
f5beed75 4044 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
c30dabfe 4045exit2:
1da177e4
LT
4046 dput(dentry);
4047 }
5955102c 4048 inode_unlock(path.dentry->d_inode);
1da177e4
LT
4049 if (inode)
4050 iput(inode); /* truncate the inode here */
b21996e3
BF
4051 inode = NULL;
4052 if (delegated_inode) {
5a14696c 4053 error = break_deleg_wait(&delegated_inode);
b21996e3
BF
4054 if (!error)
4055 goto retry_deleg;
4056 }
f5beed75 4057 mnt_drop_write(path.mnt);
1da177e4 4058exit1:
f5beed75 4059 path_put(&path);
5d18f813
JL
4060 if (retry_estale(error, lookup_flags)) {
4061 lookup_flags |= LOOKUP_REVAL;
4062 inode = NULL;
4063 goto retry;
4064 }
da2f1362 4065 putname(name);
1da177e4
LT
4066 return error;
4067
4068slashes:
b18825a7
DH
4069 if (d_is_negative(dentry))
4070 error = -ENOENT;
44b1d530 4071 else if (d_is_dir(dentry))
b18825a7
DH
4072 error = -EISDIR;
4073 else
4074 error = -ENOTDIR;
1da177e4
LT
4075 goto exit2;
4076}
4077
2e4d0924 4078SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
5590ff0d
UD
4079{
4080 if ((flag & ~AT_REMOVEDIR) != 0)
4081 return -EINVAL;
4082
4083 if (flag & AT_REMOVEDIR)
4084 return do_rmdir(dfd, pathname);
4085
da2f1362 4086 return do_unlinkat(dfd, getname(pathname));
5590ff0d
UD
4087}
4088
3480b257 4089SYSCALL_DEFINE1(unlink, const char __user *, pathname)
5590ff0d 4090{
da2f1362 4091 return do_unlinkat(AT_FDCWD, getname(pathname));
5590ff0d
UD
4092}
4093
db2e747b 4094int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
1da177e4 4095{
a95164d9 4096 int error = may_create(dir, dentry);
1da177e4
LT
4097
4098 if (error)
4099 return error;
4100
acfa4380 4101 if (!dir->i_op->symlink)
1da177e4
LT
4102 return -EPERM;
4103
4104 error = security_inode_symlink(dir, dentry, oldname);
4105 if (error)
4106 return error;
4107
1da177e4 4108 error = dir->i_op->symlink(dir, dentry, oldname);
a74574aa 4109 if (!error)
f38aa942 4110 fsnotify_create(dir, dentry);
1da177e4
LT
4111 return error;
4112}
4d359507 4113EXPORT_SYMBOL(vfs_symlink);
1da177e4 4114
b724e846
DB
4115long do_symlinkat(const char __user *oldname, int newdfd,
4116 const char __user *newname)
1da177e4 4117{
2ad94ae6 4118 int error;
91a27b2a 4119 struct filename *from;
6902d925 4120 struct dentry *dentry;
dae6ad8f 4121 struct path path;
f46d3567 4122 unsigned int lookup_flags = 0;
1da177e4
LT
4123
4124 from = getname(oldname);
2ad94ae6 4125 if (IS_ERR(from))
1da177e4 4126 return PTR_ERR(from);
f46d3567
JL
4127retry:
4128 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
6902d925
DH
4129 error = PTR_ERR(dentry);
4130 if (IS_ERR(dentry))
dae6ad8f 4131 goto out_putname;
6902d925 4132
91a27b2a 4133 error = security_path_symlink(&path, dentry, from->name);
a8104a9f 4134 if (!error)
91a27b2a 4135 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
921a1650 4136 done_path_create(&path, dentry);
f46d3567
JL
4137 if (retry_estale(error, lookup_flags)) {
4138 lookup_flags |= LOOKUP_REVAL;
4139 goto retry;
4140 }
6902d925 4141out_putname:
1da177e4
LT
4142 putname(from);
4143 return error;
4144}
4145
b724e846
DB
4146SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4147 int, newdfd, const char __user *, newname)
4148{
4149 return do_symlinkat(oldname, newdfd, newname);
4150}
4151
3480b257 4152SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
5590ff0d 4153{
b724e846 4154 return do_symlinkat(oldname, AT_FDCWD, newname);
5590ff0d
UD
4155}
4156
146a8595
BF
4157/**
4158 * vfs_link - create a new link
4159 * @old_dentry: object to be linked
4160 * @dir: new parent
4161 * @new_dentry: where to create the new link
4162 * @delegated_inode: returns inode needing a delegation break
4163 *
4164 * The caller must hold dir->i_mutex
4165 *
4166 * If vfs_link discovers a delegation on the to-be-linked file in need
4167 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4168 * inode in delegated_inode. The caller should then break the delegation
4169 * and retry. Because breaking a delegation may take a long time, the
4170 * caller should drop the i_mutex before doing so.
4171 *
4172 * Alternatively, a caller may pass NULL for delegated_inode. This may
4173 * be appropriate for callers that expect the underlying filesystem not
4174 * to be NFS exported.
4175 */
4176int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
1da177e4
LT
4177{
4178 struct inode *inode = old_dentry->d_inode;
8de52778 4179 unsigned max_links = dir->i_sb->s_max_links;
1da177e4
LT
4180 int error;
4181
4182 if (!inode)
4183 return -ENOENT;
4184
a95164d9 4185 error = may_create(dir, new_dentry);
1da177e4
LT
4186 if (error)
4187 return error;
4188
4189 if (dir->i_sb != inode->i_sb)
4190 return -EXDEV;
4191
4192 /*
4193 * A link to an append-only or immutable file cannot be created.
4194 */
4195 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4196 return -EPERM;
0bd23d09
EB
4197 /*
4198 * Updating the link count will likely cause i_uid and i_gid to
4199 * be writen back improperly if their true value is unknown to
4200 * the vfs.
4201 */
4202 if (HAS_UNMAPPED_ID(inode))
4203 return -EPERM;
acfa4380 4204 if (!dir->i_op->link)
1da177e4 4205 return -EPERM;
7e79eedb 4206 if (S_ISDIR(inode->i_mode))
1da177e4
LT
4207 return -EPERM;
4208
4209 error = security_inode_link(old_dentry, dir, new_dentry);
4210 if (error)
4211 return error;
4212
5955102c 4213 inode_lock(inode);
aae8a97d 4214 /* Make sure we don't allow creating hardlink to an unlinked file */
f4e0c30c 4215 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
aae8a97d 4216 error = -ENOENT;
8de52778
AV
4217 else if (max_links && inode->i_nlink >= max_links)
4218 error = -EMLINK;
146a8595
BF
4219 else {
4220 error = try_break_deleg(inode, delegated_inode);
4221 if (!error)
4222 error = dir->i_op->link(old_dentry, dir, new_dentry);
4223 }
f4e0c30c
AV
4224
4225 if (!error && (inode->i_state & I_LINKABLE)) {
4226 spin_lock(&inode->i_lock);
4227 inode->i_state &= ~I_LINKABLE;
4228 spin_unlock(&inode->i_lock);
4229 }
5955102c 4230 inode_unlock(inode);
e31e14ec 4231 if (!error)
7e79eedb 4232 fsnotify_link(dir, inode, new_dentry);
1da177e4
LT
4233 return error;
4234}
4d359507 4235EXPORT_SYMBOL(vfs_link);
1da177e4
LT
4236
4237/*
4238 * Hardlinks are often used in delicate situations. We avoid
4239 * security-related surprises by not following symlinks on the
4240 * newname. --KAB
4241 *
4242 * We don't follow them on the oldname either to be compatible
4243 * with linux 2.0, and to avoid hard-linking to directories
4244 * and other special files. --ADM
4245 */
46ea89eb
DB
4246int do_linkat(int olddfd, const char __user *oldname, int newdfd,
4247 const char __user *newname, int flags)
1da177e4
LT
4248{
4249 struct dentry *new_dentry;
dae6ad8f 4250 struct path old_path, new_path;
146a8595 4251 struct inode *delegated_inode = NULL;
11a7b371 4252 int how = 0;
1da177e4 4253 int error;
1da177e4 4254
11a7b371 4255 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
c04030e1 4256 return -EINVAL;
11a7b371 4257 /*
f0cc6ffb
LT
4258 * To use null names we require CAP_DAC_READ_SEARCH
4259 * This ensures that not everyone will be able to create
4260 * handlink using the passed filedescriptor.
11a7b371 4261 */
f0cc6ffb
LT
4262 if (flags & AT_EMPTY_PATH) {
4263 if (!capable(CAP_DAC_READ_SEARCH))
4264 return -ENOENT;
11a7b371 4265 how = LOOKUP_EMPTY;
f0cc6ffb 4266 }
11a7b371
AK
4267
4268 if (flags & AT_SYMLINK_FOLLOW)
4269 how |= LOOKUP_FOLLOW;
442e31ca 4270retry:
11a7b371 4271 error = user_path_at(olddfd, oldname, how, &old_path);
1da177e4 4272 if (error)
2ad94ae6
AV
4273 return error;
4274
442e31ca
JL
4275 new_dentry = user_path_create(newdfd, newname, &new_path,
4276 (how & LOOKUP_REVAL));
1da177e4 4277 error = PTR_ERR(new_dentry);
6902d925 4278 if (IS_ERR(new_dentry))
dae6ad8f
AV
4279 goto out;
4280
4281 error = -EXDEV;
4282 if (old_path.mnt != new_path.mnt)
4283 goto out_dput;
800179c9
KC
4284 error = may_linkat(&old_path);
4285 if (unlikely(error))
4286 goto out_dput;
dae6ad8f 4287 error = security_path_link(old_path.dentry, &new_path, new_dentry);
be6d3e56 4288 if (error)
a8104a9f 4289 goto out_dput;
146a8595 4290 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
75c3f29d 4291out_dput:
921a1650 4292 done_path_create(&new_path, new_dentry);
146a8595
BF
4293 if (delegated_inode) {
4294 error = break_deleg_wait(&delegated_inode);
d22e6338
OD
4295 if (!error) {
4296 path_put(&old_path);
146a8595 4297 goto retry;
d22e6338 4298 }
146a8595 4299 }
442e31ca 4300 if (retry_estale(error, how)) {
d22e6338 4301 path_put(&old_path);
442e31ca
JL
4302 how |= LOOKUP_REVAL;
4303 goto retry;
4304 }
1da177e4 4305out:
2d8f3038 4306 path_put(&old_path);
1da177e4
LT
4307
4308 return error;
4309}
4310
46ea89eb
DB
4311SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4312 int, newdfd, const char __user *, newname, int, flags)
4313{
4314 return do_linkat(olddfd, oldname, newdfd, newname, flags);
4315}
4316
3480b257 4317SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
5590ff0d 4318{
46ea89eb 4319 return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
5590ff0d
UD
4320}
4321
bc27027a
MS
4322/**
4323 * vfs_rename - rename a filesystem object
4324 * @old_dir: parent of source
4325 * @old_dentry: source
4326 * @new_dir: parent of destination
4327 * @new_dentry: destination
4328 * @delegated_inode: returns an inode needing a delegation break
520c8b16 4329 * @flags: rename flags
bc27027a
MS
4330 *
4331 * The caller must hold multiple mutexes--see lock_rename()).
4332 *
4333 * If vfs_rename discovers a delegation in need of breaking at either
4334 * the source or destination, it will return -EWOULDBLOCK and return a
4335 * reference to the inode in delegated_inode. The caller should then
4336 * break the delegation and retry. Because breaking a delegation may
4337 * take a long time, the caller should drop all locks before doing
4338 * so.
4339 *
4340 * Alternatively, a caller may pass NULL for delegated_inode. This may
4341 * be appropriate for callers that expect the underlying filesystem not
4342 * to be NFS exported.
4343 *
1da177e4
LT
4344 * The worst of all namespace operations - renaming directory. "Perverted"
4345 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4346 * Problems:
0117d427 4347 *
d03b29a2 4348 * a) we can get into loop creation.
1da177e4
LT
4349 * b) race potential - two innocent renames can create a loop together.
4350 * That's where 4.4 screws up. Current fix: serialization on
a11f3a05 4351 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
1da177e4 4352 * story.
6cedba89
BF
4353 * c) we have to lock _four_ objects - parents and victim (if it exists),
4354 * and source (if it is not a directory).
1b1dcc1b 4355 * And that - after we got ->i_mutex on parents (until then we don't know
1da177e4
LT
4356 * whether the target exists). Solution: try to be smart with locking
4357 * order for inodes. We rely on the fact that tree topology may change
a11f3a05 4358 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
1da177e4
LT
4359 * move will be locked. Thus we can rank directories by the tree
4360 * (ancestors first) and rank all non-directories after them.
4361 * That works since everybody except rename does "lock parent, lookup,
a11f3a05 4362 * lock child" and rename is under ->s_vfs_rename_mutex.
1da177e4
LT
4363 * HOWEVER, it relies on the assumption that any object with ->lookup()
4364 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4365 * we'd better make sure that there's no link(2) for them.
e4eaac06 4366 * d) conversion from fhandle to dentry may come in the wrong moment - when
1b1dcc1b 4367 * we are removing the target. Solution: we will have to grab ->i_mutex
1da177e4 4368 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
c41b20e7 4369 * ->i_mutex on parents, which works but leads to some truly excessive
1da177e4
LT
4370 * locking].
4371 */
bc27027a
MS
4372int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4373 struct inode *new_dir, struct dentry *new_dentry,
520c8b16 4374 struct inode **delegated_inode, unsigned int flags)
1da177e4 4375{
bc27027a
MS
4376 int error;
4377 bool is_dir = d_is_dir(old_dentry);
bc27027a 4378 struct inode *source = old_dentry->d_inode;
9055cba7 4379 struct inode *target = new_dentry->d_inode;
da1ce067
MS
4380 bool new_is_dir = false;
4381 unsigned max_links = new_dir->i_sb->s_max_links;
49d31c2f 4382 struct name_snapshot old_name;
bc27027a 4383
8d3e2936 4384 if (source == target)
bc27027a
MS
4385 return 0;
4386
4387 error = may_delete(old_dir, old_dentry, is_dir);
4388 if (error)
4389 return error;
4390
da1ce067 4391 if (!target) {
bc27027a 4392 error = may_create(new_dir, new_dentry);
da1ce067
MS
4393 } else {
4394 new_is_dir = d_is_dir(new_dentry);
4395
4396 if (!(flags & RENAME_EXCHANGE))
4397 error = may_delete(new_dir, new_dentry, is_dir);
4398 else
4399 error = may_delete(new_dir, new_dentry, new_is_dir);
4400 }
bc27027a
MS
4401 if (error)
4402 return error;
4403
2773bf00 4404 if (!old_dir->i_op->rename)
bc27027a 4405 return -EPERM;
1da177e4
LT
4406
4407 /*
4408 * If we are going to change the parent - check write permissions,
4409 * we'll need to flip '..'.
4410 */
da1ce067
MS
4411 if (new_dir != old_dir) {
4412 if (is_dir) {
4413 error = inode_permission(source, MAY_WRITE);
4414 if (error)
4415 return error;
4416 }
4417 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4418 error = inode_permission(target, MAY_WRITE);
4419 if (error)
4420 return error;
4421 }
1da177e4
LT
4422 }
4423
0b3974eb
MS
4424 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4425 flags);
1da177e4
LT
4426 if (error)
4427 return error;
4428
49d31c2f 4429 take_dentry_name_snapshot(&old_name, old_dentry);
1d2ef590 4430 dget(new_dentry);
da1ce067 4431 if (!is_dir || (flags & RENAME_EXCHANGE))
bc27027a
MS
4432 lock_two_nondirectories(source, target);
4433 else if (target)
5955102c 4434 inode_lock(target);
9055cba7
SW
4435
4436 error = -EBUSY;
7af1364f 4437 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
9055cba7
SW
4438 goto out;
4439
da1ce067 4440 if (max_links && new_dir != old_dir) {
bc27027a 4441 error = -EMLINK;
da1ce067 4442 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
bc27027a 4443 goto out;
da1ce067
MS
4444 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4445 old_dir->i_nlink >= max_links)
4446 goto out;
4447 }
da1ce067 4448 if (!is_dir) {
bc27027a 4449 error = try_break_deleg(source, delegated_inode);
8e6d782c
BF
4450 if (error)
4451 goto out;
da1ce067
MS
4452 }
4453 if (target && !new_is_dir) {
4454 error = try_break_deleg(target, delegated_inode);
4455 if (error)
4456 goto out;
8e6d782c 4457 }
2773bf00 4458 error = old_dir->i_op->rename(old_dir, old_dentry,
18fc84da 4459 new_dir, new_dentry, flags);
51892bbb
SW
4460 if (error)
4461 goto out;
4462
da1ce067 4463 if (!(flags & RENAME_EXCHANGE) && target) {
8767712f
AV
4464 if (is_dir) {
4465 shrink_dcache_parent(new_dentry);
bc27027a 4466 target->i_flags |= S_DEAD;
8767712f 4467 }
51892bbb 4468 dont_mount(new_dentry);
8ed936b5 4469 detach_mounts(new_dentry);
bc27027a 4470 }
da1ce067
MS
4471 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4472 if (!(flags & RENAME_EXCHANGE))
4473 d_move(old_dentry, new_dentry);
4474 else
4475 d_exchange(old_dentry, new_dentry);
4476 }
51892bbb 4477out:
da1ce067 4478 if (!is_dir || (flags & RENAME_EXCHANGE))
bc27027a
MS
4479 unlock_two_nondirectories(source, target);
4480 else if (target)
5955102c 4481 inode_unlock(target);
1da177e4 4482 dput(new_dentry);
da1ce067 4483 if (!error) {
49d31c2f 4484 fsnotify_move(old_dir, new_dir, old_name.name, is_dir,
da1ce067
MS
4485 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4486 if (flags & RENAME_EXCHANGE) {
4487 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4488 new_is_dir, NULL, new_dentry);
4489 }
4490 }
49d31c2f 4491 release_dentry_name_snapshot(&old_name);
0eeca283 4492
1da177e4
LT
4493 return error;
4494}
4d359507 4495EXPORT_SYMBOL(vfs_rename);
1da177e4 4496
ee81feb6
DB
4497static int do_renameat2(int olddfd, const char __user *oldname, int newdfd,
4498 const char __user *newname, unsigned int flags)
1da177e4 4499{
2ad94ae6
AV
4500 struct dentry *old_dentry, *new_dentry;
4501 struct dentry *trap;
f5beed75
AV
4502 struct path old_path, new_path;
4503 struct qstr old_last, new_last;
4504 int old_type, new_type;
8e6d782c 4505 struct inode *delegated_inode = NULL;
91a27b2a
JL
4506 struct filename *from;
4507 struct filename *to;
f5beed75 4508 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
c6a94284 4509 bool should_retry = false;
2ad94ae6 4510 int error;
520c8b16 4511
0d7a8555 4512 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
da1ce067
MS
4513 return -EINVAL;
4514
0d7a8555
MS
4515 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4516 (flags & RENAME_EXCHANGE))
520c8b16
MS
4517 return -EINVAL;
4518
0d7a8555
MS
4519 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4520 return -EPERM;
4521
f5beed75
AV
4522 if (flags & RENAME_EXCHANGE)
4523 target_flags = 0;
4524
c6a94284 4525retry:
c1d4dd27
AV
4526 from = filename_parentat(olddfd, getname(oldname), lookup_flags,
4527 &old_path, &old_last, &old_type);
91a27b2a
JL
4528 if (IS_ERR(from)) {
4529 error = PTR_ERR(from);
1da177e4 4530 goto exit;
91a27b2a 4531 }
1da177e4 4532
c1d4dd27
AV
4533 to = filename_parentat(newdfd, getname(newname), lookup_flags,
4534 &new_path, &new_last, &new_type);
91a27b2a
JL
4535 if (IS_ERR(to)) {
4536 error = PTR_ERR(to);
1da177e4 4537 goto exit1;
91a27b2a 4538 }
1da177e4
LT
4539
4540 error = -EXDEV;
f5beed75 4541 if (old_path.mnt != new_path.mnt)
1da177e4
LT
4542 goto exit2;
4543
1da177e4 4544 error = -EBUSY;
f5beed75 4545 if (old_type != LAST_NORM)
1da177e4
LT
4546 goto exit2;
4547
0a7c3937
MS
4548 if (flags & RENAME_NOREPLACE)
4549 error = -EEXIST;
f5beed75 4550 if (new_type != LAST_NORM)
1da177e4
LT
4551 goto exit2;
4552
f5beed75 4553 error = mnt_want_write(old_path.mnt);
c30dabfe
JK
4554 if (error)
4555 goto exit2;
4556
8e6d782c 4557retry_deleg:
f5beed75 4558 trap = lock_rename(new_path.dentry, old_path.dentry);
1da177e4 4559
f5beed75 4560 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
1da177e4
LT
4561 error = PTR_ERR(old_dentry);
4562 if (IS_ERR(old_dentry))
4563 goto exit3;
4564 /* source must exist */
4565 error = -ENOENT;
b18825a7 4566 if (d_is_negative(old_dentry))
1da177e4 4567 goto exit4;
f5beed75 4568 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
0a7c3937
MS
4569 error = PTR_ERR(new_dentry);
4570 if (IS_ERR(new_dentry))
4571 goto exit4;
4572 error = -EEXIST;
4573 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4574 goto exit5;
da1ce067
MS
4575 if (flags & RENAME_EXCHANGE) {
4576 error = -ENOENT;
4577 if (d_is_negative(new_dentry))
4578 goto exit5;
4579
4580 if (!d_is_dir(new_dentry)) {
4581 error = -ENOTDIR;
f5beed75 4582 if (new_last.name[new_last.len])
da1ce067
MS
4583 goto exit5;
4584 }
4585 }
1da177e4 4586 /* unless the source is a directory trailing slashes give -ENOTDIR */
44b1d530 4587 if (!d_is_dir(old_dentry)) {
1da177e4 4588 error = -ENOTDIR;
f5beed75 4589 if (old_last.name[old_last.len])
0a7c3937 4590 goto exit5;
f5beed75 4591 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
0a7c3937 4592 goto exit5;
1da177e4
LT
4593 }
4594 /* source should not be ancestor of target */
4595 error = -EINVAL;
4596 if (old_dentry == trap)
0a7c3937 4597 goto exit5;
1da177e4 4598 /* target should not be an ancestor of source */
da1ce067
MS
4599 if (!(flags & RENAME_EXCHANGE))
4600 error = -ENOTEMPTY;
1da177e4
LT
4601 if (new_dentry == trap)
4602 goto exit5;
4603
f5beed75
AV
4604 error = security_path_rename(&old_path, old_dentry,
4605 &new_path, new_dentry, flags);
be6d3e56 4606 if (error)
c30dabfe 4607 goto exit5;
f5beed75
AV
4608 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4609 new_path.dentry->d_inode, new_dentry,
520c8b16 4610 &delegated_inode, flags);
1da177e4
LT
4611exit5:
4612 dput(new_dentry);
4613exit4:
4614 dput(old_dentry);
4615exit3:
f5beed75 4616 unlock_rename(new_path.dentry, old_path.dentry);
8e6d782c
BF
4617 if (delegated_inode) {
4618 error = break_deleg_wait(&delegated_inode);
4619 if (!error)
4620 goto retry_deleg;
4621 }
f5beed75 4622 mnt_drop_write(old_path.mnt);
1da177e4 4623exit2:
c6a94284
JL
4624 if (retry_estale(error, lookup_flags))
4625 should_retry = true;
f5beed75 4626 path_put(&new_path);
2ad94ae6 4627 putname(to);
1da177e4 4628exit1:
f5beed75 4629 path_put(&old_path);
1da177e4 4630 putname(from);
c6a94284
JL
4631 if (should_retry) {
4632 should_retry = false;
4633 lookup_flags |= LOOKUP_REVAL;
4634 goto retry;
4635 }
2ad94ae6 4636exit:
1da177e4
LT
4637 return error;
4638}
4639
ee81feb6
DB
4640SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4641 int, newdfd, const char __user *, newname, unsigned int, flags)
4642{
4643 return do_renameat2(olddfd, oldname, newdfd, newname, flags);
4644}
4645
520c8b16
MS
4646SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4647 int, newdfd, const char __user *, newname)
4648{
ee81feb6 4649 return do_renameat2(olddfd, oldname, newdfd, newname, 0);
520c8b16
MS
4650}
4651
a26eab24 4652SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5590ff0d 4653{
ee81feb6 4654 return do_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
5590ff0d
UD
4655}
4656
787fb6bc
MS
4657int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4658{
4659 int error = may_create(dir, dentry);
4660 if (error)
4661 return error;
4662
4663 if (!dir->i_op->mknod)
4664 return -EPERM;
4665
4666 return dir->i_op->mknod(dir, dentry,
4667 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4668}
4669EXPORT_SYMBOL(vfs_whiteout);
4670
5d826c84 4671int readlink_copy(char __user *buffer, int buflen, const char *link)
1da177e4 4672{
5d826c84 4673 int len = PTR_ERR(link);
1da177e4
LT
4674 if (IS_ERR(link))
4675 goto out;
4676
4677 len = strlen(link);
4678 if (len > (unsigned) buflen)
4679 len = buflen;
4680 if (copy_to_user(buffer, link, len))
4681 len = -EFAULT;
4682out:
4683 return len;
4684}
4685
4686/*
4687 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
6b255391
AV
4688 * have ->get_link() not calling nd_jump_link(). Using (or not using) it
4689 * for any given inode is up to filesystem.
1da177e4 4690 */
d16744ec
MS
4691static int generic_readlink(struct dentry *dentry, char __user *buffer,
4692 int buflen)
1da177e4 4693{
fceef393 4694 DEFINE_DELAYED_CALL(done);
5f2c4179
AV
4695 struct inode *inode = d_inode(dentry);
4696 const char *link = inode->i_link;
694a1764 4697 int res;
cc314eef 4698
d4dee48b 4699 if (!link) {
fceef393 4700 link = inode->i_op->get_link(dentry, inode, &done);
d4dee48b
AV
4701 if (IS_ERR(link))
4702 return PTR_ERR(link);
4703 }
680baacb 4704 res = readlink_copy(buffer, buflen, link);
fceef393 4705 do_delayed_call(&done);
694a1764 4706 return res;
1da177e4
LT
4707}
4708
fd4a0edf
MS
4709/**
4710 * vfs_readlink - copy symlink body into userspace buffer
4711 * @dentry: dentry on which to get symbolic link
4712 * @buffer: user memory pointer
4713 * @buflen: size of buffer
4714 *
4715 * Does not touch atime. That's up to the caller if necessary
4716 *
4717 * Does not call security hook.
4718 */
4719int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4720{
4721 struct inode *inode = d_inode(dentry);
4722
76fca90e
MS
4723 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
4724 if (unlikely(inode->i_op->readlink))
4725 return inode->i_op->readlink(dentry, buffer, buflen);
4726
4727 if (!d_is_symlink(dentry))
4728 return -EINVAL;
4729
4730 spin_lock(&inode->i_lock);
4731 inode->i_opflags |= IOP_DEFAULT_READLINK;
4732 spin_unlock(&inode->i_lock);
4733 }
fd4a0edf 4734
76fca90e 4735 return generic_readlink(dentry, buffer, buflen);
fd4a0edf
MS
4736}
4737EXPORT_SYMBOL(vfs_readlink);
1da177e4 4738
d60874cd
MS
4739/**
4740 * vfs_get_link - get symlink body
4741 * @dentry: dentry on which to get symbolic link
4742 * @done: caller needs to free returned data with this
4743 *
4744 * Calls security hook and i_op->get_link() on the supplied inode.
4745 *
4746 * It does not touch atime. That's up to the caller if necessary.
4747 *
4748 * Does not work on "special" symlinks like /proc/$$/fd/N
4749 */
4750const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4751{
4752 const char *res = ERR_PTR(-EINVAL);
4753 struct inode *inode = d_inode(dentry);
4754
4755 if (d_is_symlink(dentry)) {
4756 res = ERR_PTR(security_inode_readlink(dentry));
4757 if (!res)
4758 res = inode->i_op->get_link(dentry, inode, done);
4759 }
4760 return res;
4761}
4762EXPORT_SYMBOL(vfs_get_link);
4763
1da177e4 4764/* get the link contents into pagecache */
6b255391 4765const char *page_get_link(struct dentry *dentry, struct inode *inode,
fceef393 4766 struct delayed_call *callback)
1da177e4 4767{
ebd09abb
DG
4768 char *kaddr;
4769 struct page *page;
6b255391
AV
4770 struct address_space *mapping = inode->i_mapping;
4771
d3883d4f
AV
4772 if (!dentry) {
4773 page = find_get_page(mapping, 0);
4774 if (!page)
4775 return ERR_PTR(-ECHILD);
4776 if (!PageUptodate(page)) {
4777 put_page(page);
4778 return ERR_PTR(-ECHILD);
4779 }
4780 } else {
4781 page = read_mapping_page(mapping, 0, NULL);
4782 if (IS_ERR(page))
4783 return (char*)page;
4784 }
fceef393 4785 set_delayed_call(callback, page_put_link, page);
21fc61c7
AV
4786 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4787 kaddr = page_address(page);
6b255391 4788 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
ebd09abb 4789 return kaddr;
1da177e4
LT
4790}
4791
6b255391 4792EXPORT_SYMBOL(page_get_link);
1da177e4 4793
fceef393 4794void page_put_link(void *arg)
1da177e4 4795{
fceef393 4796 put_page(arg);
1da177e4 4797}
4d359507 4798EXPORT_SYMBOL(page_put_link);
1da177e4 4799
aa80deab
AV
4800int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4801{
fceef393 4802 DEFINE_DELAYED_CALL(done);
6b255391
AV
4803 int res = readlink_copy(buffer, buflen,
4804 page_get_link(dentry, d_inode(dentry),
fceef393
AV
4805 &done));
4806 do_delayed_call(&done);
aa80deab
AV
4807 return res;
4808}
4809EXPORT_SYMBOL(page_readlink);
4810
54566b2c
NP
4811/*
4812 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4813 */
4814int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
1da177e4
LT
4815{
4816 struct address_space *mapping = inode->i_mapping;
0adb25d2 4817 struct page *page;
afddba49 4818 void *fsdata;
beb497ab 4819 int err;
c718a975 4820 unsigned int flags = 0;
54566b2c
NP
4821 if (nofs)
4822 flags |= AOP_FLAG_NOFS;
1da177e4 4823
7e53cac4 4824retry:
afddba49 4825 err = pagecache_write_begin(NULL, mapping, 0, len-1,
54566b2c 4826 flags, &page, &fsdata);
1da177e4 4827 if (err)
afddba49
NP
4828 goto fail;
4829
21fc61c7 4830 memcpy(page_address(page), symname, len-1);
afddba49
NP
4831
4832 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4833 page, fsdata);
1da177e4
LT
4834 if (err < 0)
4835 goto fail;
afddba49
NP
4836 if (err < len-1)
4837 goto retry;
4838
1da177e4
LT
4839 mark_inode_dirty(inode);
4840 return 0;
1da177e4
LT
4841fail:
4842 return err;
4843}
4d359507 4844EXPORT_SYMBOL(__page_symlink);
1da177e4 4845
0adb25d2
KK
4846int page_symlink(struct inode *inode, const char *symname, int len)
4847{
4848 return __page_symlink(inode, symname, len,
c62d2555 4849 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
0adb25d2 4850}
4d359507 4851EXPORT_SYMBOL(page_symlink);
0adb25d2 4852
92e1d5be 4853const struct inode_operations page_symlink_inode_operations = {
6b255391 4854 .get_link = page_get_link,
1da177e4 4855};
1da177e4 4856EXPORT_SYMBOL(page_symlink_inode_operations);