]> Git Repo - linux.git/blame - fs/namei.c
Merge tag 'gpio-fixes-for-v5.19-rc1' of git://git.kernel.org/pub/scm/linux/kernel...
[linux.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>
2d878178 25#include <linux/sched/mm.h>
0eeca283 26#include <linux/fsnotify.h>
1da177e4
LT
27#include <linux/personality.h>
28#include <linux/security.h>
6146f0d5 29#include <linux/ima.h>
1da177e4
LT
30#include <linux/syscalls.h>
31#include <linux/mount.h>
32#include <linux/audit.h>
16f7e0fe 33#include <linux/capability.h>
834f2a4a 34#include <linux/file.h>
5590ff0d 35#include <linux/fcntl.h>
08ce5f16 36#include <linux/device_cgroup.h>
5ad4e53b 37#include <linux/fs_struct.h>
e77819e5 38#include <linux/posix_acl.h>
99d263d4 39#include <linux/hash.h>
2a18da7a 40#include <linux/bitops.h>
aeaa4a79 41#include <linux/init_task.h>
7c0f6ba6 42#include <linux/uaccess.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;
4043cde8 134
7ac86265
JL
135 result = audit_reusename(filename);
136 if (result)
137 return result;
138
7950e385 139 result = __getname();
3f9f0aa6 140 if (unlikely(!result))
4043cde8
EP
141 return ERR_PTR(-ENOMEM);
142
7950e385
JL
143 /*
144 * First, try to embed the struct filename inside the names_cache
145 * allocation
146 */
fd2f7cb5 147 kname = (char *)result->iname;
91a27b2a 148 result->name = kname;
7950e385 149
94b5d262 150 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
91a27b2a 151 if (unlikely(len < 0)) {
94b5d262
AV
152 __putname(result);
153 return ERR_PTR(len);
91a27b2a 154 }
3f9f0aa6 155
7950e385
JL
156 /*
157 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
158 * separate struct filename so we can dedicate the entire
159 * names_cache allocation for the pathname, and re-do the copy from
160 * userland.
161 */
94b5d262 162 if (unlikely(len == EMBEDDED_NAME_MAX)) {
fd2f7cb5 163 const size_t size = offsetof(struct filename, iname[1]);
7950e385
JL
164 kname = (char *)result;
165
fd2f7cb5
AV
166 /*
167 * size is chosen that way we to guarantee that
168 * result->iname[0] is within the same object and that
169 * kname can't be equal to result->iname, no matter what.
170 */
171 result = kzalloc(size, GFP_KERNEL);
94b5d262
AV
172 if (unlikely(!result)) {
173 __putname(kname);
174 return ERR_PTR(-ENOMEM);
7950e385
JL
175 }
176 result->name = kname;
94b5d262
AV
177 len = strncpy_from_user(kname, filename, PATH_MAX);
178 if (unlikely(len < 0)) {
179 __putname(kname);
180 kfree(result);
181 return ERR_PTR(len);
182 }
183 if (unlikely(len == PATH_MAX)) {
184 __putname(kname);
185 kfree(result);
186 return ERR_PTR(-ENAMETOOLONG);
187 }
7950e385
JL
188 }
189
94b5d262 190 result->refcnt = 1;
3f9f0aa6
LT
191 /* The empty path is special. */
192 if (unlikely(!len)) {
193 if (empty)
4043cde8 194 *empty = 1;
94b5d262
AV
195 if (!(flags & LOOKUP_EMPTY)) {
196 putname(result);
197 return ERR_PTR(-ENOENT);
198 }
1da177e4 199 }
3f9f0aa6 200
7950e385 201 result->uptr = filename;
c4ad8f98 202 result->aname = NULL;
7950e385
JL
203 audit_getname(result);
204 return result;
1da177e4
LT
205}
206
8228e2c3
DK
207struct filename *
208getname_uflags(const char __user *filename, int uflags)
209{
210 int flags = (uflags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
211
212 return getname_flags(filename, flags, NULL);
213}
214
91a27b2a
JL
215struct filename *
216getname(const char __user * filename)
f52e0c11 217{
f7493e5d 218 return getname_flags(filename, 0, NULL);
f52e0c11
AV
219}
220
c4ad8f98
LT
221struct filename *
222getname_kernel(const char * filename)
223{
224 struct filename *result;
08518549 225 int len = strlen(filename) + 1;
c4ad8f98
LT
226
227 result = __getname();
228 if (unlikely(!result))
229 return ERR_PTR(-ENOMEM);
230
08518549 231 if (len <= EMBEDDED_NAME_MAX) {
fd2f7cb5 232 result->name = (char *)result->iname;
08518549 233 } else if (len <= PATH_MAX) {
30ce4d19 234 const size_t size = offsetof(struct filename, iname[1]);
08518549
PM
235 struct filename *tmp;
236
30ce4d19 237 tmp = kmalloc(size, GFP_KERNEL);
08518549
PM
238 if (unlikely(!tmp)) {
239 __putname(result);
240 return ERR_PTR(-ENOMEM);
241 }
242 tmp->name = (char *)result;
08518549
PM
243 result = tmp;
244 } else {
245 __putname(result);
246 return ERR_PTR(-ENAMETOOLONG);
247 }
248 memcpy((char *)result->name, filename, len);
c4ad8f98
LT
249 result->uptr = NULL;
250 result->aname = NULL;
55422d0b 251 result->refcnt = 1;
fd3522fd 252 audit_getname(result);
c4ad8f98 253
c4ad8f98
LT
254 return result;
255}
256
91a27b2a 257void putname(struct filename *name)
1da177e4 258{
ea47ab11 259 if (IS_ERR(name))
91ef658f
DK
260 return;
261
55422d0b
PM
262 BUG_ON(name->refcnt <= 0);
263
264 if (--name->refcnt > 0)
265 return;
266
fd2f7cb5 267 if (name->name != name->iname) {
55422d0b
PM
268 __putname(name->name);
269 kfree(name);
270 } else
271 __putname(name);
1da177e4 272}
1da177e4 273
47291baa
CB
274/**
275 * check_acl - perform ACL permission checking
276 * @mnt_userns: user namespace of the mount the inode was found from
277 * @inode: inode to check permissions on
278 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
279 *
280 * This function performs the ACL permission checking. Since this function
281 * retrieve POSIX acls it needs to know whether it is called from a blocking or
282 * non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
283 *
284 * If the inode has been found through an idmapped mount the user namespace of
285 * the vfsmount must be passed through @mnt_userns. This function will then take
286 * care to map the inode according to @mnt_userns before checking permissions.
287 * On non-idmapped mounts or if permission checking is to be performed on the
288 * raw inode simply passs init_user_ns.
289 */
290static int check_acl(struct user_namespace *mnt_userns,
291 struct inode *inode, int mask)
e77819e5 292{
84635d68 293#ifdef CONFIG_FS_POSIX_ACL
e77819e5
LT
294 struct posix_acl *acl;
295
e77819e5 296 if (mask & MAY_NOT_BLOCK) {
3567866b
AV
297 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
298 if (!acl)
e77819e5 299 return -EAGAIN;
3567866b 300 /* no ->get_acl() calls in RCU mode... */
b8a7a3a6 301 if (is_uncached_acl(acl))
3567866b 302 return -ECHILD;
47291baa 303 return posix_acl_permission(mnt_userns, inode, acl, mask);
e77819e5
LT
304 }
305
2982baa2
CH
306 acl = get_acl(inode, ACL_TYPE_ACCESS);
307 if (IS_ERR(acl))
308 return PTR_ERR(acl);
e77819e5 309 if (acl) {
47291baa 310 int error = posix_acl_permission(mnt_userns, inode, acl, mask);
e77819e5
LT
311 posix_acl_release(acl);
312 return error;
313 }
84635d68 314#endif
e77819e5
LT
315
316 return -EAGAIN;
317}
318
47291baa
CB
319/**
320 * acl_permission_check - perform basic UNIX permission checking
321 * @mnt_userns: user namespace of the mount the inode was found from
322 * @inode: inode to check permissions on
323 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
324 *
325 * This function performs the basic UNIX permission checking. Since this
326 * function may retrieve POSIX acls it needs to know whether it is called from a
327 * blocking or non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
5fc475b7 328 *
47291baa
CB
329 * If the inode has been found through an idmapped mount the user namespace of
330 * the vfsmount must be passed through @mnt_userns. This function will then take
331 * care to map the inode according to @mnt_userns before checking permissions.
332 * On non-idmapped mounts or if permission checking is to be performed on the
333 * raw inode simply passs init_user_ns.
1da177e4 334 */
47291baa
CB
335static int acl_permission_check(struct user_namespace *mnt_userns,
336 struct inode *inode, int mask)
1da177e4 337{
26cf46be 338 unsigned int mode = inode->i_mode;
47291baa 339 kuid_t i_uid;
1da177e4 340
5fc475b7 341 /* Are we the owner? If so, ACL's don't matter */
47291baa
CB
342 i_uid = i_uid_into_mnt(mnt_userns, inode);
343 if (likely(uid_eq(current_fsuid(), i_uid))) {
5fc475b7 344 mask &= 7;
1da177e4 345 mode >>= 6;
5fc475b7
LT
346 return (mask & ~mode) ? -EACCES : 0;
347 }
1da177e4 348
5fc475b7
LT
349 /* Do we have ACL's? */
350 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
47291baa 351 int error = check_acl(mnt_userns, inode, mask);
5fc475b7
LT
352 if (error != -EAGAIN)
353 return error;
1da177e4
LT
354 }
355
5fc475b7
LT
356 /* Only RWX matters for group/other mode bits */
357 mask &= 7;
358
1da177e4 359 /*
5fc475b7
LT
360 * Are the group permissions different from
361 * the other permissions in the bits we care
362 * about? Need to check group ownership if so.
1da177e4 363 */
5fc475b7 364 if (mask & (mode ^ (mode >> 3))) {
47291baa
CB
365 kgid_t kgid = i_gid_into_mnt(mnt_userns, inode);
366 if (in_group_p(kgid))
5fc475b7
LT
367 mode >>= 3;
368 }
369
370 /* Bits in 'mode' clear that we require? */
371 return (mask & ~mode) ? -EACCES : 0;
5909ccaa
LT
372}
373
374/**
b74c79e9 375 * generic_permission - check for access rights on a Posix-like filesystem
47291baa 376 * @mnt_userns: user namespace of the mount the inode was found from
5909ccaa 377 * @inode: inode to check access rights for
5fc475b7
LT
378 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC,
379 * %MAY_NOT_BLOCK ...)
5909ccaa
LT
380 *
381 * Used to check for read/write/execute permissions on a file.
382 * We use "fsuid" for this, letting us set arbitrary permissions
383 * for filesystem access without changing the "normal" uids which
b74c79e9
NP
384 * are used for other things.
385 *
386 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
387 * request cannot be satisfied (eg. requires blocking or too much complexity).
388 * It would then be called again in ref-walk mode.
47291baa
CB
389 *
390 * If the inode has been found through an idmapped mount the user namespace of
391 * the vfsmount must be passed through @mnt_userns. This function will then take
392 * care to map the inode according to @mnt_userns before checking permissions.
393 * On non-idmapped mounts or if permission checking is to be performed on the
394 * raw inode simply passs init_user_ns.
5909ccaa 395 */
47291baa
CB
396int generic_permission(struct user_namespace *mnt_userns, struct inode *inode,
397 int mask)
5909ccaa
LT
398{
399 int ret;
400
401 /*
948409c7 402 * Do the basic permission checks.
5909ccaa 403 */
47291baa 404 ret = acl_permission_check(mnt_userns, inode, mask);
5909ccaa
LT
405 if (ret != -EACCES)
406 return ret;
1da177e4 407
d594e7ec
AV
408 if (S_ISDIR(inode->i_mode)) {
409 /* DACs are overridable for directories */
d594e7ec 410 if (!(mask & MAY_WRITE))
47291baa 411 if (capable_wrt_inode_uidgid(mnt_userns, inode,
23adbe12 412 CAP_DAC_READ_SEARCH))
d594e7ec 413 return 0;
47291baa 414 if (capable_wrt_inode_uidgid(mnt_userns, inode,
0558c1bf 415 CAP_DAC_OVERRIDE))
1da177e4 416 return 0;
2a4c2242
SS
417 return -EACCES;
418 }
1da177e4
LT
419
420 /*
421 * Searching includes executable on directories, else just read.
422 */
7ea66001 423 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
d594e7ec 424 if (mask == MAY_READ)
47291baa 425 if (capable_wrt_inode_uidgid(mnt_userns, inode,
0558c1bf 426 CAP_DAC_READ_SEARCH))
1da177e4 427 return 0;
2a4c2242
SS
428 /*
429 * Read/write DACs are always overridable.
430 * Executable DACs are overridable when there is
431 * at least one exec bit set.
432 */
433 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
47291baa 434 if (capable_wrt_inode_uidgid(mnt_userns, inode,
0558c1bf 435 CAP_DAC_OVERRIDE))
2a4c2242 436 return 0;
1da177e4
LT
437
438 return -EACCES;
439}
4d359507 440EXPORT_SYMBOL(generic_permission);
1da177e4 441
47291baa
CB
442/**
443 * do_inode_permission - UNIX permission checking
444 * @mnt_userns: user namespace of the mount the inode was found from
445 * @inode: inode to check permissions on
446 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
447 *
3ddcd056
LT
448 * We _really_ want to just do "generic_permission()" without
449 * even looking at the inode->i_op values. So we keep a cache
450 * flag in inode->i_opflags, that says "this has not special
451 * permission function, use the fast case".
452 */
47291baa
CB
453static inline int do_inode_permission(struct user_namespace *mnt_userns,
454 struct inode *inode, int mask)
3ddcd056
LT
455{
456 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
457 if (likely(inode->i_op->permission))
549c7297 458 return inode->i_op->permission(mnt_userns, inode, mask);
3ddcd056
LT
459
460 /* This gets set once for the inode lifetime */
461 spin_lock(&inode->i_lock);
462 inode->i_opflags |= IOP_FASTPERM;
463 spin_unlock(&inode->i_lock);
464 }
47291baa 465 return generic_permission(mnt_userns, inode, mask);
3ddcd056
LT
466}
467
0bdaea90
DH
468/**
469 * sb_permission - Check superblock-level permissions
470 * @sb: Superblock of inode to check permission on
55852635 471 * @inode: Inode to check permission on
0bdaea90
DH
472 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
473 *
474 * Separate out file-system wide checks from inode-specific permission checks.
475 */
476static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
477{
478 if (unlikely(mask & MAY_WRITE)) {
479 umode_t mode = inode->i_mode;
480
481 /* Nobody gets write access to a read-only fs. */
bc98a42c 482 if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
0bdaea90
DH
483 return -EROFS;
484 }
485 return 0;
486}
487
488/**
489 * inode_permission - Check for access rights to a given inode
47291baa
CB
490 * @mnt_userns: User namespace of the mount the inode was found from
491 * @inode: Inode to check permission on
492 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
0bdaea90
DH
493 *
494 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
495 * this, letting us set arbitrary permissions for filesystem access without
496 * changing the "normal" UIDs which are used for other things.
497 *
498 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
499 */
47291baa
CB
500int inode_permission(struct user_namespace *mnt_userns,
501 struct inode *inode, int mask)
0bdaea90
DH
502{
503 int retval;
504
505 retval = sb_permission(inode->i_sb, inode, mask);
506 if (retval)
507 return retval;
4bfd054a
EB
508
509 if (unlikely(mask & MAY_WRITE)) {
510 /*
511 * Nobody gets write access to an immutable file.
512 */
513 if (IS_IMMUTABLE(inode))
514 return -EPERM;
515
516 /*
517 * Updating mtime will likely cause i_uid and i_gid to be
518 * written back improperly if their true value is unknown
519 * to the vfs.
520 */
ba73d987 521 if (HAS_UNMAPPED_ID(mnt_userns, inode))
4bfd054a
EB
522 return -EACCES;
523 }
524
47291baa 525 retval = do_inode_permission(mnt_userns, inode, mask);
4bfd054a
EB
526 if (retval)
527 return retval;
528
529 retval = devcgroup_inode_permission(inode, mask);
530 if (retval)
531 return retval;
532
533 return security_inode_permission(inode, mask);
0bdaea90 534}
4d359507 535EXPORT_SYMBOL(inode_permission);
0bdaea90 536
5dd784d0
JB
537/**
538 * path_get - get a reference to a path
539 * @path: path to get the reference to
540 *
541 * Given a path increment the reference count to the dentry and the vfsmount.
542 */
dcf787f3 543void path_get(const struct path *path)
5dd784d0
JB
544{
545 mntget(path->mnt);
546 dget(path->dentry);
547}
548EXPORT_SYMBOL(path_get);
549
1d957f9b
JB
550/**
551 * path_put - put a reference to a path
552 * @path: path to put the reference to
553 *
554 * Given a path decrement the reference count to the dentry and the vfsmount.
555 */
dcf787f3 556void path_put(const struct path *path)
1da177e4 557{
1d957f9b
JB
558 dput(path->dentry);
559 mntput(path->mnt);
1da177e4 560}
1d957f9b 561EXPORT_SYMBOL(path_put);
1da177e4 562
894bc8c4 563#define EMBEDDED_LEVELS 2
1f55a6ec
AV
564struct nameidata {
565 struct path path;
1cf2665b 566 struct qstr last;
1f55a6ec
AV
567 struct path root;
568 struct inode *inode; /* path.dentry.d_inode */
bcba1e7d 569 unsigned int flags, state;
ab87f9a5 570 unsigned seq, m_seq, r_seq;
1f55a6ec
AV
571 int last_type;
572 unsigned depth;
756daf26 573 int total_link_count;
697fc6ca
AV
574 struct saved {
575 struct path link;
fceef393 576 struct delayed_call done;
697fc6ca 577 const char *name;
0450b2d1 578 unsigned seq;
894bc8c4 579 } *stack, internal[EMBEDDED_LEVELS];
9883d185
AV
580 struct filename *name;
581 struct nameidata *saved;
582 unsigned root_seq;
583 int dfd;
0f705953
AV
584 kuid_t dir_uid;
585 umode_t dir_mode;
3859a271 586} __randomize_layout;
1f55a6ec 587
bcba1e7d
AV
588#define ND_ROOT_PRESET 1
589#define ND_ROOT_GRABBED 2
590#define ND_JUMPED 4
591
06422964 592static void __set_nameidata(struct nameidata *p, int dfd, struct filename *name)
894bc8c4 593{
756daf26
N
594 struct nameidata *old = current->nameidata;
595 p->stack = p->internal;
7962c7d1 596 p->depth = 0;
c8a53ee5
AV
597 p->dfd = dfd;
598 p->name = name;
7d01ef75
AV
599 p->path.mnt = NULL;
600 p->path.dentry = NULL;
756daf26 601 p->total_link_count = old ? old->total_link_count : 0;
9883d185 602 p->saved = old;
756daf26 603 current->nameidata = p;
894bc8c4
AV
604}
605
06422964
AV
606static inline void set_nameidata(struct nameidata *p, int dfd, struct filename *name,
607 const struct path *root)
608{
609 __set_nameidata(p, dfd, name);
610 p->state = 0;
611 if (unlikely(root)) {
612 p->state = ND_ROOT_PRESET;
613 p->root = *root;
614 }
615}
616
9883d185 617static void restore_nameidata(void)
894bc8c4 618{
9883d185 619 struct nameidata *now = current->nameidata, *old = now->saved;
756daf26
N
620
621 current->nameidata = old;
622 if (old)
623 old->total_link_count = now->total_link_count;
e1a63bbc 624 if (now->stack != now->internal)
756daf26 625 kfree(now->stack);
894bc8c4
AV
626}
627
60ef60c7 628static bool nd_alloc_stack(struct nameidata *nd)
894bc8c4 629{
bc40aee0
AV
630 struct saved *p;
631
60ef60c7
AV
632 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
633 nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL);
634 if (unlikely(!p))
635 return false;
894bc8c4
AV
636 memcpy(p, nd->internal, sizeof(nd->internal));
637 nd->stack = p;
60ef60c7 638 return true;
894bc8c4
AV
639}
640
397d425d 641/**
6b03f7ed 642 * path_connected - Verify that a dentry is below mnt.mnt_root
397d425d
EB
643 *
644 * Rename can sometimes move a file or directory outside of a bind
645 * mount, path_connected allows those cases to be detected.
646 */
6b03f7ed 647static bool path_connected(struct vfsmount *mnt, struct dentry *dentry)
397d425d 648{
95dd7758 649 struct super_block *sb = mnt->mnt_sb;
397d425d 650
402dd2cf
CH
651 /* Bind mounts can have disconnected paths */
652 if (mnt->mnt_root == sb->s_root)
397d425d
EB
653 return true;
654
6b03f7ed 655 return is_subdir(dentry, mnt->mnt_root);
397d425d
EB
656}
657
7973387a
AV
658static void drop_links(struct nameidata *nd)
659{
660 int i = nd->depth;
661 while (i--) {
662 struct saved *last = nd->stack + i;
fceef393
AV
663 do_delayed_call(&last->done);
664 clear_delayed_call(&last->done);
7973387a
AV
665 }
666}
667
668static void terminate_walk(struct nameidata *nd)
669{
670 drop_links(nd);
671 if (!(nd->flags & LOOKUP_RCU)) {
672 int i;
673 path_put(&nd->path);
674 for (i = 0; i < nd->depth; i++)
675 path_put(&nd->stack[i].link);
bcba1e7d 676 if (nd->state & ND_ROOT_GRABBED) {
102b8af2 677 path_put(&nd->root);
bcba1e7d 678 nd->state &= ~ND_ROOT_GRABBED;
102b8af2 679 }
7973387a
AV
680 } else {
681 nd->flags &= ~LOOKUP_RCU;
7973387a
AV
682 rcu_read_unlock();
683 }
684 nd->depth = 0;
7d01ef75
AV
685 nd->path.mnt = NULL;
686 nd->path.dentry = NULL;
7973387a
AV
687}
688
689/* path_put is needed afterwards regardless of success or failure */
2aa38470 690static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq)
7973387a 691{
2aa38470 692 int res = __legitimize_mnt(path->mnt, mseq);
7973387a
AV
693 if (unlikely(res)) {
694 if (res > 0)
695 path->mnt = NULL;
696 path->dentry = NULL;
697 return false;
698 }
699 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
700 path->dentry = NULL;
701 return false;
702 }
703 return !read_seqcount_retry(&path->dentry->d_seq, seq);
704}
705
2aa38470
AV
706static inline bool legitimize_path(struct nameidata *nd,
707 struct path *path, unsigned seq)
708{
5bd73286 709 return __legitimize_path(path, seq, nd->m_seq);
2aa38470
AV
710}
711
7973387a
AV
712static bool legitimize_links(struct nameidata *nd)
713{
714 int i;
eacd9aa8
AV
715 if (unlikely(nd->flags & LOOKUP_CACHED)) {
716 drop_links(nd);
717 nd->depth = 0;
718 return false;
719 }
7973387a
AV
720 for (i = 0; i < nd->depth; i++) {
721 struct saved *last = nd->stack + i;
722 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
723 drop_links(nd);
724 nd->depth = i + 1;
725 return false;
726 }
727 }
728 return true;
729}
730
ee594bff
AV
731static bool legitimize_root(struct nameidata *nd)
732{
adb21d2b
AS
733 /*
734 * For scoped-lookups (where nd->root has been zeroed), we need to
735 * restart the whole lookup from scratch -- because set_root() is wrong
736 * for these lookups (nd->dfd is the root, not the filesystem root).
737 */
738 if (!nd->root.mnt && (nd->flags & LOOKUP_IS_SCOPED))
739 return false;
740 /* Nothing to do if nd->root is zero or is managed by the VFS user. */
bcba1e7d 741 if (!nd->root.mnt || (nd->state & ND_ROOT_PRESET))
ee594bff 742 return true;
bcba1e7d 743 nd->state |= ND_ROOT_GRABBED;
ee594bff
AV
744 return legitimize_path(nd, &nd->root, nd->root_seq);
745}
746
19660af7 747/*
31e6b01f 748 * Path walking has 2 modes, rcu-walk and ref-walk (see
19660af7
AV
749 * Documentation/filesystems/path-lookup.txt). In situations when we can't
750 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
57e3715c 751 * normal reference counts on dentries and vfsmounts to transition to ref-walk
19660af7
AV
752 * mode. Refcounts are grabbed at the last known good point before rcu-walk
753 * got stuck, so ref-walk may continue from there. If this is not successful
754 * (eg. a seqcount has changed), then failure is returned and it's up to caller
755 * to restart the path walk from the beginning in ref-walk mode.
31e6b01f 756 */
31e6b01f
NP
757
758/**
e36cffed 759 * try_to_unlazy - try to switch to ref-walk mode.
19660af7 760 * @nd: nameidata pathwalk data
e36cffed 761 * Returns: true on success, false on failure
31e6b01f 762 *
e36cffed 763 * try_to_unlazy attempts to legitimize the current nd->path and nd->root
4675ac39
AV
764 * for ref-walk mode.
765 * Must be called from rcu-walk context.
e36cffed 766 * Nothing should touch nameidata between try_to_unlazy() failure and
7973387a 767 * terminate_walk().
31e6b01f 768 */
e36cffed 769static bool try_to_unlazy(struct nameidata *nd)
31e6b01f 770{
31e6b01f
NP
771 struct dentry *parent = nd->path.dentry;
772
773 BUG_ON(!(nd->flags & LOOKUP_RCU));
e5c832d5 774
4675ac39
AV
775 nd->flags &= ~LOOKUP_RCU;
776 if (unlikely(!legitimize_links(nd)))
4675ac39 777 goto out1;
84a2bd39
AV
778 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
779 goto out;
ee594bff
AV
780 if (unlikely(!legitimize_root(nd)))
781 goto out;
4675ac39
AV
782 rcu_read_unlock();
783 BUG_ON(nd->inode != parent->d_inode);
e36cffed 784 return true;
4675ac39 785
84a2bd39 786out1:
4675ac39
AV
787 nd->path.mnt = NULL;
788 nd->path.dentry = NULL;
4675ac39
AV
789out:
790 rcu_read_unlock();
e36cffed 791 return false;
4675ac39
AV
792}
793
794/**
ae66db45 795 * try_to_unlazy_next - try to switch to ref-walk mode.
4675ac39 796 * @nd: nameidata pathwalk data
ae66db45
AV
797 * @dentry: next dentry to step into
798 * @seq: seq number to check @dentry against
799 * Returns: true on success, false on failure
4675ac39 800 *
ae66db45
AV
801 * Similar to to try_to_unlazy(), but here we have the next dentry already
802 * picked by rcu-walk and want to legitimize that in addition to the current
803 * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context.
804 * Nothing should touch nameidata between try_to_unlazy_next() failure and
4675ac39
AV
805 * terminate_walk().
806 */
ae66db45 807static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry, unsigned seq)
4675ac39
AV
808{
809 BUG_ON(!(nd->flags & LOOKUP_RCU));
810
e5c832d5 811 nd->flags &= ~LOOKUP_RCU;
7973387a
AV
812 if (unlikely(!legitimize_links(nd)))
813 goto out2;
814 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
815 goto out2;
4675ac39 816 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
7973387a 817 goto out1;
48a066e7 818
15570086 819 /*
4675ac39
AV
820 * We need to move both the parent and the dentry from the RCU domain
821 * to be properly refcounted. And the sequence number in the dentry
822 * validates *both* dentry counters, since we checked the sequence
823 * number of the parent after we got the child sequence number. So we
824 * know the parent must still be valid if the child sequence number is
15570086 825 */
4675ac39
AV
826 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
827 goto out;
84a2bd39
AV
828 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
829 goto out_dput;
e5c832d5
LT
830 /*
831 * Sequence counts matched. Now make sure that the root is
832 * still valid and get it if required.
833 */
84a2bd39
AV
834 if (unlikely(!legitimize_root(nd)))
835 goto out_dput;
8b61e74f 836 rcu_read_unlock();
ae66db45 837 return true;
19660af7 838
7973387a
AV
839out2:
840 nd->path.mnt = NULL;
841out1:
842 nd->path.dentry = NULL;
e5c832d5 843out:
8b61e74f 844 rcu_read_unlock();
ae66db45 845 return false;
84a2bd39
AV
846out_dput:
847 rcu_read_unlock();
848 dput(dentry);
ae66db45 849 return false;
31e6b01f
NP
850}
851
4ce16ef3 852static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
34286d66 853{
a89f8337
AV
854 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
855 return dentry->d_op->d_revalidate(dentry, flags);
856 else
857 return 1;
34286d66
NP
858}
859
9f1fafee
AV
860/**
861 * complete_walk - successful completion of path walk
862 * @nd: pointer nameidata
39159de2 863 *
9f1fafee
AV
864 * If we had been in RCU mode, drop out of it and legitimize nd->path.
865 * Revalidate the final result, unless we'd already done that during
866 * the path walk or the filesystem doesn't ask for it. Return 0 on
867 * success, -error on failure. In case of failure caller does not
868 * need to drop nd->path.
39159de2 869 */
9f1fafee 870static int complete_walk(struct nameidata *nd)
39159de2 871{
16c2cd71 872 struct dentry *dentry = nd->path.dentry;
39159de2 873 int status;
39159de2 874
9f1fafee 875 if (nd->flags & LOOKUP_RCU) {
adb21d2b
AS
876 /*
877 * We don't want to zero nd->root for scoped-lookups or
878 * externally-managed nd->root.
879 */
bcba1e7d
AV
880 if (!(nd->state & ND_ROOT_PRESET))
881 if (!(nd->flags & LOOKUP_IS_SCOPED))
882 nd->root.mnt = NULL;
6c6ec2b0 883 nd->flags &= ~LOOKUP_CACHED;
e36cffed 884 if (!try_to_unlazy(nd))
9f1fafee 885 return -ECHILD;
9f1fafee
AV
886 }
887
adb21d2b
AS
888 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
889 /*
890 * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't
891 * ever step outside the root during lookup" and should already
892 * be guaranteed by the rest of namei, we want to avoid a namei
893 * BUG resulting in userspace being given a path that was not
894 * scoped within the root at some point during the lookup.
895 *
896 * So, do a final sanity-check to make sure that in the
897 * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED)
898 * we won't silently return an fd completely outside of the
899 * requested root to userspace.
900 *
901 * Userspace could move the path outside the root after this
902 * check, but as discussed elsewhere this is not a concern (the
903 * resolved file was inside the root at some point).
904 */
905 if (!path_is_under(&nd->path, &nd->root))
906 return -EXDEV;
907 }
908
bcba1e7d 909 if (likely(!(nd->state & ND_JUMPED)))
16c2cd71
AV
910 return 0;
911
ecf3d1f1 912 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
39159de2
JL
913 return 0;
914
ecf3d1f1 915 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
39159de2
JL
916 if (status > 0)
917 return 0;
918
16c2cd71 919 if (!status)
39159de2 920 status = -ESTALE;
16c2cd71 921
39159de2
JL
922 return status;
923}
924
740a1678 925static int set_root(struct nameidata *nd)
31e6b01f 926{
7bd88377 927 struct fs_struct *fs = current->fs;
c28cc364 928
adb21d2b
AS
929 /*
930 * Jumping to the real root in a scoped-lookup is a BUG in namei, but we
931 * still have to ensure it doesn't happen because it will cause a breakout
932 * from the dirfd.
933 */
934 if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED))
935 return -ENOTRECOVERABLE;
936
9e6697e2
AV
937 if (nd->flags & LOOKUP_RCU) {
938 unsigned seq;
939
940 do {
941 seq = read_seqcount_begin(&fs->seq);
942 nd->root = fs->root;
943 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
944 } while (read_seqcount_retry(&fs->seq, seq));
945 } else {
946 get_fs_root(fs, &nd->root);
bcba1e7d 947 nd->state |= ND_ROOT_GRABBED;
9e6697e2 948 }
740a1678 949 return 0;
31e6b01f
NP
950}
951
248fb5b9
AV
952static int nd_jump_root(struct nameidata *nd)
953{
adb21d2b
AS
954 if (unlikely(nd->flags & LOOKUP_BENEATH))
955 return -EXDEV;
72ba2929
AS
956 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
957 /* Absolute path arguments to path_init() are allowed. */
958 if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt)
959 return -EXDEV;
960 }
740a1678
AS
961 if (!nd->root.mnt) {
962 int error = set_root(nd);
963 if (error)
964 return error;
965 }
248fb5b9
AV
966 if (nd->flags & LOOKUP_RCU) {
967 struct dentry *d;
968 nd->path = nd->root;
969 d = nd->path.dentry;
970 nd->inode = d->d_inode;
971 nd->seq = nd->root_seq;
972 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
973 return -ECHILD;
974 } else {
975 path_put(&nd->path);
976 nd->path = nd->root;
977 path_get(&nd->path);
978 nd->inode = nd->path.dentry->d_inode;
979 }
bcba1e7d 980 nd->state |= ND_JUMPED;
248fb5b9
AV
981 return 0;
982}
983
b5fb63c1 984/*
6b255391 985 * Helper to directly jump to a known parsed path from ->get_link,
b5fb63c1
CH
986 * caller must have taken a reference to path beforehand.
987 */
1bc82070 988int nd_jump_link(struct path *path)
b5fb63c1 989{
4b99d499 990 int error = -ELOOP;
6e77137b 991 struct nameidata *nd = current->nameidata;
b5fb63c1 992
4b99d499
AS
993 if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS))
994 goto err;
995
72ba2929
AS
996 error = -EXDEV;
997 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
998 if (nd->path.mnt != path->mnt)
999 goto err;
1000 }
adb21d2b
AS
1001 /* Not currently safe for scoped-lookups. */
1002 if (unlikely(nd->flags & LOOKUP_IS_SCOPED))
1003 goto err;
72ba2929 1004
4b99d499 1005 path_put(&nd->path);
b5fb63c1
CH
1006 nd->path = *path;
1007 nd->inode = nd->path.dentry->d_inode;
bcba1e7d 1008 nd->state |= ND_JUMPED;
1bc82070 1009 return 0;
4b99d499
AS
1010
1011err:
1012 path_put(path);
1013 return error;
b5fb63c1
CH
1014}
1015
b9ff4429 1016static inline void put_link(struct nameidata *nd)
574197e0 1017{
21c3003d 1018 struct saved *last = nd->stack + --nd->depth;
fceef393 1019 do_delayed_call(&last->done);
6548fae2
AV
1020 if (!(nd->flags & LOOKUP_RCU))
1021 path_put(&last->link);
574197e0
AV
1022}
1023
9c011be1
LC
1024static int sysctl_protected_symlinks __read_mostly;
1025static int sysctl_protected_hardlinks __read_mostly;
1026static int sysctl_protected_fifos __read_mostly;
1027static int sysctl_protected_regular __read_mostly;
1028
1029#ifdef CONFIG_SYSCTL
1030static struct ctl_table namei_sysctls[] = {
1031 {
1032 .procname = "protected_symlinks",
1033 .data = &sysctl_protected_symlinks,
1034 .maxlen = sizeof(int),
c7031c14 1035 .mode = 0644,
9c011be1
LC
1036 .proc_handler = proc_dointvec_minmax,
1037 .extra1 = SYSCTL_ZERO,
1038 .extra2 = SYSCTL_ONE,
1039 },
1040 {
1041 .procname = "protected_hardlinks",
1042 .data = &sysctl_protected_hardlinks,
1043 .maxlen = sizeof(int),
c7031c14 1044 .mode = 0644,
9c011be1
LC
1045 .proc_handler = proc_dointvec_minmax,
1046 .extra1 = SYSCTL_ZERO,
1047 .extra2 = SYSCTL_ONE,
1048 },
1049 {
1050 .procname = "protected_fifos",
1051 .data = &sysctl_protected_fifos,
1052 .maxlen = sizeof(int),
c7031c14 1053 .mode = 0644,
9c011be1
LC
1054 .proc_handler = proc_dointvec_minmax,
1055 .extra1 = SYSCTL_ZERO,
1056 .extra2 = SYSCTL_TWO,
1057 },
1058 {
1059 .procname = "protected_regular",
1060 .data = &sysctl_protected_regular,
1061 .maxlen = sizeof(int),
c7031c14 1062 .mode = 0644,
9c011be1
LC
1063 .proc_handler = proc_dointvec_minmax,
1064 .extra1 = SYSCTL_ZERO,
1065 .extra2 = SYSCTL_TWO,
1066 },
1067 { }
1068};
1069
1070static int __init init_fs_namei_sysctls(void)
1071{
1072 register_sysctl_init("fs", namei_sysctls);
1073 return 0;
1074}
1075fs_initcall(init_fs_namei_sysctls);
1076
1077#endif /* CONFIG_SYSCTL */
800179c9
KC
1078
1079/**
1080 * may_follow_link - Check symlink following for unsafe situations
55852635 1081 * @nd: nameidata pathwalk data
800179c9
KC
1082 *
1083 * In the case of the sysctl_protected_symlinks sysctl being enabled,
1084 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
1085 * in a sticky world-writable directory. This is to protect privileged
1086 * processes from failing races against path names that may change out
1087 * from under them by way of other users creating malicious symlinks.
1088 * It will permit symlinks to be followed only when outside a sticky
1089 * world-writable directory, or when the uid of the symlink and follower
1090 * match, or when the directory owner matches the symlink's owner.
1091 *
1092 * Returns 0 if following the symlink is allowed, -ve on error.
1093 */
ad6cc4c3 1094static inline int may_follow_link(struct nameidata *nd, const struct inode *inode)
800179c9 1095{
ba73d987
CB
1096 struct user_namespace *mnt_userns;
1097 kuid_t i_uid;
1098
800179c9
KC
1099 if (!sysctl_protected_symlinks)
1100 return 0;
1101
ba73d987
CB
1102 mnt_userns = mnt_user_ns(nd->path.mnt);
1103 i_uid = i_uid_into_mnt(mnt_userns, inode);
800179c9 1104 /* Allowed if owner and follower match. */
ba73d987 1105 if (uid_eq(current_cred()->fsuid, i_uid))
800179c9
KC
1106 return 0;
1107
1108 /* Allowed if parent directory not sticky and world-writable. */
0f705953 1109 if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
800179c9
KC
1110 return 0;
1111
1112 /* Allowed if parent directory and link owner match. */
ba73d987 1113 if (uid_valid(nd->dir_uid) && uid_eq(nd->dir_uid, i_uid))
800179c9
KC
1114 return 0;
1115
31956502
AV
1116 if (nd->flags & LOOKUP_RCU)
1117 return -ECHILD;
1118
ea841baf 1119 audit_inode(nd->name, nd->stack[0].link.dentry, 0);
245d7369 1120 audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link");
800179c9
KC
1121 return -EACCES;
1122}
1123
1124/**
1125 * safe_hardlink_source - Check for safe hardlink conditions
ba73d987 1126 * @mnt_userns: user namespace of the mount the inode was found from
800179c9
KC
1127 * @inode: the source inode to hardlink from
1128 *
1129 * Return false if at least one of the following conditions:
1130 * - inode is not a regular file
1131 * - inode is setuid
1132 * - inode is setgid and group-exec
1133 * - access failure for read and write
1134 *
1135 * Otherwise returns true.
1136 */
ba73d987
CB
1137static bool safe_hardlink_source(struct user_namespace *mnt_userns,
1138 struct inode *inode)
800179c9
KC
1139{
1140 umode_t mode = inode->i_mode;
1141
1142 /* Special files should not get pinned to the filesystem. */
1143 if (!S_ISREG(mode))
1144 return false;
1145
1146 /* Setuid files should not get pinned to the filesystem. */
1147 if (mode & S_ISUID)
1148 return false;
1149
1150 /* Executable setgid files should not get pinned to the filesystem. */
1151 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
1152 return false;
1153
1154 /* Hardlinking to unreadable or unwritable sources is dangerous. */
ba73d987 1155 if (inode_permission(mnt_userns, inode, MAY_READ | MAY_WRITE))
800179c9
KC
1156 return false;
1157
1158 return true;
1159}
1160
1161/**
1162 * may_linkat - Check permissions for creating a hardlink
ba73d987 1163 * @mnt_userns: user namespace of the mount the inode was found from
800179c9
KC
1164 * @link: the source to hardlink from
1165 *
1166 * Block hardlink when all of:
1167 * - sysctl_protected_hardlinks enabled
1168 * - fsuid does not match inode
1169 * - hardlink source is unsafe (see safe_hardlink_source() above)
f2ca3796 1170 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
800179c9 1171 *
ba73d987
CB
1172 * If the inode has been found through an idmapped mount the user namespace of
1173 * the vfsmount must be passed through @mnt_userns. This function will then take
1174 * care to map the inode according to @mnt_userns before checking permissions.
1175 * On non-idmapped mounts or if permission checking is to be performed on the
1176 * raw inode simply passs init_user_ns.
1177 *
800179c9
KC
1178 * Returns 0 if successful, -ve on error.
1179 */
ba73d987 1180int may_linkat(struct user_namespace *mnt_userns, struct path *link)
800179c9 1181{
593d1ce8
EB
1182 struct inode *inode = link->dentry->d_inode;
1183
1184 /* Inode writeback is not safe when the uid or gid are invalid. */
ba73d987
CB
1185 if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) ||
1186 !gid_valid(i_gid_into_mnt(mnt_userns, inode)))
593d1ce8 1187 return -EOVERFLOW;
800179c9
KC
1188
1189 if (!sysctl_protected_hardlinks)
1190 return 0;
1191
800179c9
KC
1192 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1193 * otherwise, it must be a safe source.
1194 */
ba73d987
CB
1195 if (safe_hardlink_source(mnt_userns, inode) ||
1196 inode_owner_or_capable(mnt_userns, inode))
800179c9
KC
1197 return 0;
1198
245d7369 1199 audit_log_path_denied(AUDIT_ANOM_LINK, "linkat");
800179c9
KC
1200 return -EPERM;
1201}
1202
30aba665
SM
1203/**
1204 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1205 * should be allowed, or not, on files that already
1206 * exist.
ba73d987 1207 * @mnt_userns: user namespace of the mount the inode was found from
2111c3c0 1208 * @nd: nameidata pathwalk data
30aba665
SM
1209 * @inode: the inode of the file to open
1210 *
1211 * Block an O_CREAT open of a FIFO (or a regular file) when:
1212 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1213 * - the file already exists
1214 * - we are in a sticky directory
1215 * - we don't own the file
1216 * - the owner of the directory doesn't own the file
1217 * - the directory is world writable
1218 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1219 * the directory doesn't have to be world writable: being group writable will
1220 * be enough.
1221 *
ba73d987
CB
1222 * If the inode has been found through an idmapped mount the user namespace of
1223 * the vfsmount must be passed through @mnt_userns. This function will then take
1224 * care to map the inode according to @mnt_userns before checking permissions.
1225 * On non-idmapped mounts or if permission checking is to be performed on the
1226 * raw inode simply passs init_user_ns.
1227 *
30aba665
SM
1228 * Returns 0 if the open is allowed, -ve on error.
1229 */
ba73d987
CB
1230static int may_create_in_sticky(struct user_namespace *mnt_userns,
1231 struct nameidata *nd, struct inode *const inode)
30aba665 1232{
ba73d987
CB
1233 umode_t dir_mode = nd->dir_mode;
1234 kuid_t dir_uid = nd->dir_uid;
1235
30aba665
SM
1236 if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1237 (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
d0cb5018 1238 likely(!(dir_mode & S_ISVTX)) ||
ba73d987
CB
1239 uid_eq(i_uid_into_mnt(mnt_userns, inode), dir_uid) ||
1240 uid_eq(current_fsuid(), i_uid_into_mnt(mnt_userns, inode)))
30aba665
SM
1241 return 0;
1242
d0cb5018
AV
1243 if (likely(dir_mode & 0002) ||
1244 (dir_mode & 0020 &&
30aba665
SM
1245 ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1246 (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
245d7369
KC
1247 const char *operation = S_ISFIFO(inode->i_mode) ?
1248 "sticky_create_fifo" :
1249 "sticky_create_regular";
1250 audit_log_path_denied(AUDIT_ANOM_CREAT, operation);
30aba665
SM
1251 return -EACCES;
1252 }
1253 return 0;
1254}
1255
f015f126
DH
1256/*
1257 * follow_up - Find the mountpoint of path's vfsmount
1258 *
1259 * Given a path, find the mountpoint of its source file system.
1260 * Replace @path with the path of the mountpoint in the parent mount.
1261 * Up is towards /.
1262 *
1263 * Return 1 if we went up a level and 0 if we were already at the
1264 * root.
1265 */
bab77ebf 1266int follow_up(struct path *path)
1da177e4 1267{
0714a533
AV
1268 struct mount *mnt = real_mount(path->mnt);
1269 struct mount *parent;
1da177e4 1270 struct dentry *mountpoint;
99b7db7b 1271
48a066e7 1272 read_seqlock_excl(&mount_lock);
0714a533 1273 parent = mnt->mnt_parent;
3c0a6163 1274 if (parent == mnt) {
48a066e7 1275 read_sequnlock_excl(&mount_lock);
1da177e4
LT
1276 return 0;
1277 }
0714a533 1278 mntget(&parent->mnt);
a73324da 1279 mountpoint = dget(mnt->mnt_mountpoint);
48a066e7 1280 read_sequnlock_excl(&mount_lock);
bab77ebf
AV
1281 dput(path->dentry);
1282 path->dentry = mountpoint;
1283 mntput(path->mnt);
0714a533 1284 path->mnt = &parent->mnt;
1da177e4
LT
1285 return 1;
1286}
4d359507 1287EXPORT_SYMBOL(follow_up);
1da177e4 1288
7ef482fa
AV
1289static bool choose_mountpoint_rcu(struct mount *m, const struct path *root,
1290 struct path *path, unsigned *seqp)
1291{
1292 while (mnt_has_parent(m)) {
1293 struct dentry *mountpoint = m->mnt_mountpoint;
1294
1295 m = m->mnt_parent;
1296 if (unlikely(root->dentry == mountpoint &&
1297 root->mnt == &m->mnt))
1298 break;
1299 if (mountpoint != m->mnt.mnt_root) {
1300 path->mnt = &m->mnt;
1301 path->dentry = mountpoint;
1302 *seqp = read_seqcount_begin(&mountpoint->d_seq);
1303 return true;
1304 }
1305 }
1306 return false;
1307}
1308
2aa38470
AV
1309static bool choose_mountpoint(struct mount *m, const struct path *root,
1310 struct path *path)
1311{
1312 bool found;
1313
1314 rcu_read_lock();
1315 while (1) {
1316 unsigned seq, mseq = read_seqbegin(&mount_lock);
1317
1318 found = choose_mountpoint_rcu(m, root, path, &seq);
1319 if (unlikely(!found)) {
1320 if (!read_seqretry(&mount_lock, mseq))
1321 break;
1322 } else {
1323 if (likely(__legitimize_path(path, seq, mseq)))
1324 break;
1325 rcu_read_unlock();
1326 path_put(path);
1327 rcu_read_lock();
1328 }
1329 }
1330 rcu_read_unlock();
1331 return found;
1332}
1333
b5c84bf6 1334/*
9875cf80
DH
1335 * Perform an automount
1336 * - return -EISDIR to tell follow_managed() to stop and return the path we
1337 * were called with.
1da177e4 1338 */
1c9f5e06 1339static int follow_automount(struct path *path, int *count, unsigned lookup_flags)
31e6b01f 1340{
25e195aa 1341 struct dentry *dentry = path->dentry;
9875cf80 1342
0ec26fd0
MS
1343 /* We don't want to mount if someone's just doing a stat -
1344 * unless they're stat'ing a directory and appended a '/' to
1345 * the name.
1346 *
1347 * We do, however, want to mount if someone wants to open or
1348 * create a file of any type under the mountpoint, wants to
1349 * traverse through the mountpoint or wants to open the
1350 * mounted directory. Also, autofs may mark negative dentries
1351 * as being automount points. These will need the attentions
1352 * of the daemon to instantiate them before they can be used.
9875cf80 1353 */
1c9f5e06 1354 if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
5d38f049 1355 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
25e195aa 1356 dentry->d_inode)
5d38f049 1357 return -EISDIR;
0ec26fd0 1358
1c9f5e06 1359 if (count && (*count)++ >= MAXSYMLINKS)
9875cf80
DH
1360 return -ELOOP;
1361
25e195aa 1362 return finish_automount(dentry->d_op->d_automount(path), path);
463ffb2e
AV
1363}
1364
9875cf80 1365/*
9deed3eb
AV
1366 * mount traversal - out-of-line part. One note on ->d_flags accesses -
1367 * dentries are pinned but not locked here, so negative dentry can go
1368 * positive right under us. Use of smp_load_acquire() provides a barrier
1369 * sufficient for ->d_inode and ->d_flags consistency.
9875cf80 1370 */
9deed3eb
AV
1371static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped,
1372 int *count, unsigned lookup_flags)
1da177e4 1373{
9deed3eb 1374 struct vfsmount *mnt = path->mnt;
9875cf80 1375 bool need_mntput = false;
8aef1884 1376 int ret = 0;
9875cf80 1377
9deed3eb 1378 while (flags & DCACHE_MANAGED_DENTRY) {
cc53ce53
DH
1379 /* Allow the filesystem to manage the transit without i_mutex
1380 * being held. */
d41efb52 1381 if (flags & DCACHE_MANAGE_TRANSIT) {
fb5f51c7 1382 ret = path->dentry->d_op->d_manage(path, false);
508c8772 1383 flags = smp_load_acquire(&path->dentry->d_flags);
cc53ce53 1384 if (ret < 0)
8aef1884 1385 break;
cc53ce53
DH
1386 }
1387
9deed3eb 1388 if (flags & DCACHE_MOUNTED) { // something's mounted on it..
9875cf80 1389 struct vfsmount *mounted = lookup_mnt(path);
9deed3eb 1390 if (mounted) { // ... in our namespace
9875cf80
DH
1391 dput(path->dentry);
1392 if (need_mntput)
1393 mntput(path->mnt);
1394 path->mnt = mounted;
1395 path->dentry = dget(mounted->mnt_root);
9deed3eb
AV
1396 // here we know it's positive
1397 flags = path->dentry->d_flags;
9875cf80
DH
1398 need_mntput = true;
1399 continue;
1400 }
9875cf80
DH
1401 }
1402
9deed3eb
AV
1403 if (!(flags & DCACHE_NEED_AUTOMOUNT))
1404 break;
9875cf80 1405
9deed3eb
AV
1406 // uncovered automount point
1407 ret = follow_automount(path, count, lookup_flags);
1408 flags = smp_load_acquire(&path->dentry->d_flags);
1409 if (ret < 0)
1410 break;
1da177e4 1411 }
8aef1884 1412
9deed3eb
AV
1413 if (ret == -EISDIR)
1414 ret = 0;
1415 // possible if you race with several mount --move
1416 if (need_mntput && path->mnt == mnt)
1417 mntput(path->mnt);
1418 if (!ret && unlikely(d_flags_negative(flags)))
d41efb52 1419 ret = -ENOENT;
9deed3eb 1420 *jumped = need_mntput;
8402752e 1421 return ret;
1da177e4
LT
1422}
1423
9deed3eb
AV
1424static inline int traverse_mounts(struct path *path, bool *jumped,
1425 int *count, unsigned lookup_flags)
1426{
1427 unsigned flags = smp_load_acquire(&path->dentry->d_flags);
1428
1429 /* fastpath */
1430 if (likely(!(flags & DCACHE_MANAGED_DENTRY))) {
1431 *jumped = false;
1432 if (unlikely(d_flags_negative(flags)))
1433 return -ENOENT;
1434 return 0;
1435 }
1436 return __traverse_mounts(path, flags, jumped, count, lookup_flags);
1437}
1438
cc53ce53 1439int follow_down_one(struct path *path)
1da177e4
LT
1440{
1441 struct vfsmount *mounted;
1442
1c755af4 1443 mounted = lookup_mnt(path);
1da177e4 1444 if (mounted) {
9393bd07
AV
1445 dput(path->dentry);
1446 mntput(path->mnt);
1447 path->mnt = mounted;
1448 path->dentry = dget(mounted->mnt_root);
1da177e4
LT
1449 return 1;
1450 }
1451 return 0;
1452}
4d359507 1453EXPORT_SYMBOL(follow_down_one);
1da177e4 1454
9deed3eb
AV
1455/*
1456 * Follow down to the covering mount currently visible to userspace. At each
1457 * point, the filesystem owning that dentry may be queried as to whether the
1458 * caller is permitted to proceed or not.
1459 */
1460int follow_down(struct path *path)
1461{
1462 struct vfsmount *mnt = path->mnt;
1463 bool jumped;
1464 int ret = traverse_mounts(path, &jumped, NULL, 0);
1465
1466 if (path->mnt != mnt)
1467 mntput(mnt);
1468 return ret;
1469}
1470EXPORT_SYMBOL(follow_down);
1471
9875cf80 1472/*
287548e4
AV
1473 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1474 * we meet a managed dentry that would need blocking.
9875cf80
DH
1475 */
1476static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
254cf582 1477 struct inode **inode, unsigned *seqp)
9875cf80 1478{
ea936aeb
AV
1479 struct dentry *dentry = path->dentry;
1480 unsigned int flags = dentry->d_flags;
1481
1482 if (likely(!(flags & DCACHE_MANAGED_DENTRY)))
1483 return true;
1484
1485 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1486 return false;
1487
62a7375e 1488 for (;;) {
62a7375e
IK
1489 /*
1490 * Don't forget we might have a non-mountpoint managed dentry
1491 * that wants to block transit.
1492 */
ea936aeb
AV
1493 if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) {
1494 int res = dentry->d_op->d_manage(path, true);
1495 if (res)
1496 return res == -EISDIR;
1497 flags = dentry->d_flags;
b8faf035 1498 }
62a7375e 1499
ea936aeb
AV
1500 if (flags & DCACHE_MOUNTED) {
1501 struct mount *mounted = __lookup_mnt(path->mnt, dentry);
1502 if (mounted) {
1503 path->mnt = &mounted->mnt;
1504 dentry = path->dentry = mounted->mnt.mnt_root;
bcba1e7d 1505 nd->state |= ND_JUMPED;
ea936aeb
AV
1506 *seqp = read_seqcount_begin(&dentry->d_seq);
1507 *inode = dentry->d_inode;
1508 /*
1509 * We don't need to re-check ->d_seq after this
1510 * ->d_inode read - there will be an RCU delay
1511 * between mount hash removal and ->mnt_root
1512 * becoming unpinned.
1513 */
1514 flags = dentry->d_flags;
1515 continue;
1516 }
1517 if (read_seqretry(&mount_lock, nd->m_seq))
1518 return false;
1519 }
1520 return !(flags & DCACHE_NEED_AUTOMOUNT);
9875cf80 1521 }
287548e4
AV
1522}
1523
db3c9ade
AV
1524static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry,
1525 struct path *path, struct inode **inode,
1526 unsigned int *seqp)
bd7c4b50 1527{
9deed3eb 1528 bool jumped;
db3c9ade 1529 int ret;
bd7c4b50 1530
db3c9ade
AV
1531 path->mnt = nd->path.mnt;
1532 path->dentry = dentry;
c153007b
AV
1533 if (nd->flags & LOOKUP_RCU) {
1534 unsigned int seq = *seqp;
1535 if (unlikely(!*inode))
1536 return -ENOENT;
1537 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
9deed3eb 1538 return 0;
ae66db45 1539 if (!try_to_unlazy_next(nd, dentry, seq))
c153007b
AV
1540 return -ECHILD;
1541 // *path might've been clobbered by __follow_mount_rcu()
1542 path->mnt = nd->path.mnt;
1543 path->dentry = dentry;
1544 }
9deed3eb
AV
1545 ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags);
1546 if (jumped) {
1547 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1548 ret = -EXDEV;
1549 else
bcba1e7d 1550 nd->state |= ND_JUMPED;
9deed3eb
AV
1551 }
1552 if (unlikely(ret)) {
1553 dput(path->dentry);
1554 if (path->mnt != nd->path.mnt)
1555 mntput(path->mnt);
1556 } else {
bd7c4b50
AV
1557 *inode = d_backing_inode(path->dentry);
1558 *seqp = 0; /* out of RCU mode, so the value doesn't matter */
1559 }
1560 return ret;
1561}
1562
baa03890 1563/*
f4fdace9
OD
1564 * This looks up the name in dcache and possibly revalidates the found dentry.
1565 * NULL is returned if the dentry does not exist in the cache.
baa03890 1566 */
e3c13928
AV
1567static struct dentry *lookup_dcache(const struct qstr *name,
1568 struct dentry *dir,
6c51e513 1569 unsigned int flags)
baa03890 1570{
a89f8337 1571 struct dentry *dentry = d_lookup(dir, name);
bad61189 1572 if (dentry) {
a89f8337
AV
1573 int error = d_revalidate(dentry, flags);
1574 if (unlikely(error <= 0)) {
1575 if (!error)
1576 d_invalidate(dentry);
1577 dput(dentry);
1578 return ERR_PTR(error);
bad61189
MS
1579 }
1580 }
baa03890
NP
1581 return dentry;
1582}
1583
44396f4b 1584/*
a03ece5f
AV
1585 * Parent directory has inode locked exclusive. This is one
1586 * and only case when ->lookup() gets called on non in-lookup
1587 * dentries - as the matter of fact, this only gets called
1588 * when directory is guaranteed to have no in-lookup children
1589 * at all.
44396f4b 1590 */
e3c13928 1591static struct dentry *__lookup_hash(const struct qstr *name,
72bd866a 1592 struct dentry *base, unsigned int flags)
a3255546 1593{
6c51e513 1594 struct dentry *dentry = lookup_dcache(name, base, flags);
a03ece5f
AV
1595 struct dentry *old;
1596 struct inode *dir = base->d_inode;
a3255546 1597
6c51e513 1598 if (dentry)
bad61189 1599 return dentry;
a3255546 1600
a03ece5f
AV
1601 /* Don't create child dentry for a dead directory. */
1602 if (unlikely(IS_DEADDIR(dir)))
1603 return ERR_PTR(-ENOENT);
1604
6c51e513
AV
1605 dentry = d_alloc(base, name);
1606 if (unlikely(!dentry))
1607 return ERR_PTR(-ENOMEM);
1608
a03ece5f
AV
1609 old = dir->i_op->lookup(dir, dentry, flags);
1610 if (unlikely(old)) {
1611 dput(dentry);
1612 dentry = old;
1613 }
1614 return dentry;
a3255546
AV
1615}
1616
20e34357
AV
1617static struct dentry *lookup_fast(struct nameidata *nd,
1618 struct inode **inode,
1619 unsigned *seqp)
1da177e4 1620{
31e6b01f 1621 struct dentry *dentry, *parent = nd->path.dentry;
5a18fff2 1622 int status = 1;
9875cf80 1623
b04f784e
NP
1624 /*
1625 * Rename seqlock is not required here because in the off chance
5d0f49c1
AV
1626 * of a false negative due to a concurrent rename, the caller is
1627 * going to fall back to non-racy lookup.
b04f784e 1628 */
31e6b01f
NP
1629 if (nd->flags & LOOKUP_RCU) {
1630 unsigned seq;
da53be12 1631 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
5d0f49c1 1632 if (unlikely(!dentry)) {
e36cffed 1633 if (!try_to_unlazy(nd))
20e34357
AV
1634 return ERR_PTR(-ECHILD);
1635 return NULL;
5d0f49c1 1636 }
5a18fff2 1637
12f8ad4b
LT
1638 /*
1639 * This sequence count validates that the inode matches
1640 * the dentry name information from lookup.
1641 */
63afdfc7 1642 *inode = d_backing_inode(dentry);
5d0f49c1 1643 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
20e34357 1644 return ERR_PTR(-ECHILD);
12f8ad4b
LT
1645
1646 /*
1647 * This sequence count validates that the parent had no
1648 * changes while we did the lookup of the dentry above.
1649 *
1650 * The memory barrier in read_seqcount_begin of child is
1651 * enough, we can use __read_seqcount_retry here.
1652 */
5d0f49c1 1653 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
20e34357 1654 return ERR_PTR(-ECHILD);
5a18fff2 1655
254cf582 1656 *seqp = seq;
a89f8337 1657 status = d_revalidate(dentry, nd->flags);
c153007b 1658 if (likely(status > 0))
20e34357 1659 return dentry;
ae66db45 1660 if (!try_to_unlazy_next(nd, dentry, seq))
20e34357 1661 return ERR_PTR(-ECHILD);
26ddb45e 1662 if (status == -ECHILD)
209a7fb2
AV
1663 /* we'd been told to redo it in non-rcu mode */
1664 status = d_revalidate(dentry, nd->flags);
5a18fff2 1665 } else {
e97cdc87 1666 dentry = __d_lookup(parent, &nd->last);
5d0f49c1 1667 if (unlikely(!dentry))
20e34357 1668 return NULL;
a89f8337 1669 status = d_revalidate(dentry, nd->flags);
9875cf80 1670 }
5a18fff2 1671 if (unlikely(status <= 0)) {
e9742b53 1672 if (!status)
5d0f49c1 1673 d_invalidate(dentry);
5542aa2f 1674 dput(dentry);
20e34357 1675 return ERR_PTR(status);
24643087 1676 }
20e34357 1677 return dentry;
697f514d
MS
1678}
1679
1680/* Fast lookup failed, do it the slow way */
88d8331a
AV
1681static struct dentry *__lookup_slow(const struct qstr *name,
1682 struct dentry *dir,
1683 unsigned int flags)
697f514d 1684{
88d8331a 1685 struct dentry *dentry, *old;
1936386e 1686 struct inode *inode = dir->d_inode;
d9171b93 1687 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1936386e 1688
1936386e 1689 /* Don't go there if it's already dead */
94bdd655 1690 if (unlikely(IS_DEADDIR(inode)))
88d8331a 1691 return ERR_PTR(-ENOENT);
94bdd655 1692again:
d9171b93 1693 dentry = d_alloc_parallel(dir, name, &wq);
94bdd655 1694 if (IS_ERR(dentry))
88d8331a 1695 return dentry;
94bdd655 1696 if (unlikely(!d_in_lookup(dentry))) {
c64cd6e3
AV
1697 int error = d_revalidate(dentry, flags);
1698 if (unlikely(error <= 0)) {
1699 if (!error) {
1700 d_invalidate(dentry);
949a852e 1701 dput(dentry);
c64cd6e3 1702 goto again;
949a852e 1703 }
c64cd6e3
AV
1704 dput(dentry);
1705 dentry = ERR_PTR(error);
949a852e 1706 }
94bdd655
AV
1707 } else {
1708 old = inode->i_op->lookup(inode, dentry, flags);
1709 d_lookup_done(dentry);
1710 if (unlikely(old)) {
1711 dput(dentry);
1712 dentry = old;
949a852e
AV
1713 }
1714 }
e3c13928 1715 return dentry;
1da177e4
LT
1716}
1717
88d8331a
AV
1718static struct dentry *lookup_slow(const struct qstr *name,
1719 struct dentry *dir,
1720 unsigned int flags)
1721{
1722 struct inode *inode = dir->d_inode;
1723 struct dentry *res;
1724 inode_lock_shared(inode);
1725 res = __lookup_slow(name, dir, flags);
1726 inode_unlock_shared(inode);
1727 return res;
1728}
1729
ba73d987
CB
1730static inline int may_lookup(struct user_namespace *mnt_userns,
1731 struct nameidata *nd)
52094c8a
AV
1732{
1733 if (nd->flags & LOOKUP_RCU) {
7d6beb71 1734 int err = inode_permission(mnt_userns, nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
e36cffed 1735 if (err != -ECHILD || !try_to_unlazy(nd))
52094c8a 1736 return err;
52094c8a 1737 }
ba73d987 1738 return inode_permission(mnt_userns, nd->inode, MAY_EXEC);
52094c8a
AV
1739}
1740
49055906
AV
1741static int reserve_stack(struct nameidata *nd, struct path *link, unsigned seq)
1742{
49055906
AV
1743 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS))
1744 return -ELOOP;
4542576b
AV
1745
1746 if (likely(nd->depth != EMBEDDED_LEVELS))
1747 return 0;
1748 if (likely(nd->stack != nd->internal))
1749 return 0;
60ef60c7 1750 if (likely(nd_alloc_stack(nd)))
49055906 1751 return 0;
60ef60c7
AV
1752
1753 if (nd->flags & LOOKUP_RCU) {
1754 // we need to grab link before we do unlazy. And we can't skip
1755 // unlazy even if we fail to grab the link - cleanup needs it
49055906 1756 bool grabbed_link = legitimize_path(nd, link, seq);
60ef60c7 1757
e36cffed 1758 if (!try_to_unlazy(nd) != 0 || !grabbed_link)
60ef60c7
AV
1759 return -ECHILD;
1760
1761 if (nd_alloc_stack(nd))
1762 return 0;
49055906 1763 }
60ef60c7 1764 return -ENOMEM;
49055906
AV
1765}
1766
b1a81972
AV
1767enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4};
1768
06708adb 1769static const char *pick_link(struct nameidata *nd, struct path *link,
b1a81972 1770 struct inode *inode, unsigned seq, int flags)
d63ff28f 1771{
1cf2665b 1772 struct saved *last;
ad6cc4c3 1773 const char *res;
49055906 1774 int error = reserve_stack(nd, link, seq);
ad6cc4c3 1775
626de996 1776 if (unlikely(error)) {
49055906 1777 if (!(nd->flags & LOOKUP_RCU))
bc40aee0 1778 path_put(link);
49055906 1779 return ERR_PTR(error);
626de996 1780 }
ab104923 1781 last = nd->stack + nd->depth++;
1cf2665b 1782 last->link = *link;
fceef393 1783 clear_delayed_call(&last->done);
0450b2d1 1784 last->seq = seq;
ad6cc4c3 1785
b1a81972 1786 if (flags & WALK_TRAILING) {
ad6cc4c3
AV
1787 error = may_follow_link(nd, inode);
1788 if (unlikely(error))
1789 return ERR_PTR(error);
1790 }
1791
dab741e0
MN
1792 if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) ||
1793 unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW))
ad6cc4c3
AV
1794 return ERR_PTR(-ELOOP);
1795
1796 if (!(nd->flags & LOOKUP_RCU)) {
1797 touch_atime(&last->link);
1798 cond_resched();
1799 } else if (atime_needs_update(&last->link, inode)) {
e36cffed 1800 if (!try_to_unlazy(nd))
ad6cc4c3
AV
1801 return ERR_PTR(-ECHILD);
1802 touch_atime(&last->link);
1803 }
1804
1805 error = security_inode_follow_link(link->dentry, inode,
1806 nd->flags & LOOKUP_RCU);
1807 if (unlikely(error))
1808 return ERR_PTR(error);
1809
ad6cc4c3
AV
1810 res = READ_ONCE(inode->i_link);
1811 if (!res) {
1812 const char * (*get)(struct dentry *, struct inode *,
1813 struct delayed_call *);
1814 get = inode->i_op->get_link;
1815 if (nd->flags & LOOKUP_RCU) {
1816 res = get(NULL, inode, &last->done);
e36cffed 1817 if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd))
ad6cc4c3 1818 res = get(link->dentry, inode, &last->done);
ad6cc4c3
AV
1819 } else {
1820 res = get(link->dentry, inode, &last->done);
1821 }
1822 if (!res)
1823 goto all_done;
1824 if (IS_ERR(res))
1825 return res;
1826 }
1827 if (*res == '/') {
1828 error = nd_jump_root(nd);
1829 if (unlikely(error))
1830 return ERR_PTR(error);
1831 while (unlikely(*++res == '/'))
1832 ;
1833 }
1834 if (*res)
1835 return res;
1836all_done: // pure jump
1837 put_link(nd);
1838 return NULL;
d63ff28f
AV
1839}
1840
3ddcd056
LT
1841/*
1842 * Do we need to follow links? We _really_ want to be able
1843 * to do this check without having to look at inode->i_op,
1844 * so we keep a cache of "no, this doesn't need follow_link"
1845 * for the common case.
1846 */
b0417d2c 1847static const char *step_into(struct nameidata *nd, int flags,
cbae4d12 1848 struct dentry *dentry, struct inode *inode, unsigned seq)
3ddcd056 1849{
cbae4d12
AV
1850 struct path path;
1851 int err = handle_mounts(nd, dentry, &path, &inode, &seq);
1852
1853 if (err < 0)
b0417d2c 1854 return ERR_PTR(err);
cbae4d12 1855 if (likely(!d_is_symlink(path.dentry)) ||
8c4efe22 1856 ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) ||
aca2903e 1857 (flags & WALK_NOFOLLOW)) {
8f64fb1c 1858 /* not a symlink or should not follow */
c99687a0
AV
1859 if (!(nd->flags & LOOKUP_RCU)) {
1860 dput(nd->path.dentry);
1861 if (nd->path.mnt != path.mnt)
1862 mntput(nd->path.mnt);
1863 }
1864 nd->path = path;
8f64fb1c
AV
1865 nd->inode = inode;
1866 nd->seq = seq;
b0417d2c 1867 return NULL;
8f64fb1c 1868 }
a7f77542 1869 if (nd->flags & LOOKUP_RCU) {
84f0cd9e 1870 /* make sure that d_is_symlink above matches inode */
cbae4d12 1871 if (read_seqcount_retry(&path.dentry->d_seq, seq))
b0417d2c 1872 return ERR_PTR(-ECHILD);
84f0cd9e
AV
1873 } else {
1874 if (path.mnt == nd->path.mnt)
1875 mntget(path.mnt);
a7f77542 1876 }
b1a81972 1877 return pick_link(nd, &path, inode, seq, flags);
3ddcd056
LT
1878}
1879
c2df1968
AV
1880static struct dentry *follow_dotdot_rcu(struct nameidata *nd,
1881 struct inode **inodep,
1882 unsigned *seqp)
957dd41d 1883{
12487f30 1884 struct dentry *parent, *old;
957dd41d 1885
12487f30
AV
1886 if (path_equal(&nd->path, &nd->root))
1887 goto in_root;
1888 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
7ef482fa 1889 struct path path;
efe772d6 1890 unsigned seq;
7ef482fa
AV
1891 if (!choose_mountpoint_rcu(real_mount(nd->path.mnt),
1892 &nd->root, &path, &seq))
1893 goto in_root;
efe772d6
AV
1894 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1895 return ERR_PTR(-ECHILD);
1896 nd->path = path;
1897 nd->inode = path.dentry->d_inode;
1898 nd->seq = seq;
1899 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1900 return ERR_PTR(-ECHILD);
1901 /* we know that mountpoint was pinned */
957dd41d 1902 }
12487f30
AV
1903 old = nd->path.dentry;
1904 parent = old->d_parent;
1905 *inodep = parent->d_inode;
1906 *seqp = read_seqcount_begin(&parent->d_seq);
1907 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1908 return ERR_PTR(-ECHILD);
1909 if (unlikely(!path_connected(nd->path.mnt, parent)))
1910 return ERR_PTR(-ECHILD);
1911 return parent;
1912in_root:
efe772d6
AV
1913 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1914 return ERR_PTR(-ECHILD);
c2df1968
AV
1915 if (unlikely(nd->flags & LOOKUP_BENEATH))
1916 return ERR_PTR(-ECHILD);
1917 return NULL;
957dd41d
AV
1918}
1919
c2df1968
AV
1920static struct dentry *follow_dotdot(struct nameidata *nd,
1921 struct inode **inodep,
1922 unsigned *seqp)
957dd41d 1923{
12487f30
AV
1924 struct dentry *parent;
1925
1926 if (path_equal(&nd->path, &nd->root))
1927 goto in_root;
1928 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
2aa38470
AV
1929 struct path path;
1930
1931 if (!choose_mountpoint(real_mount(nd->path.mnt),
1932 &nd->root, &path))
1933 goto in_root;
165200d6
AV
1934 path_put(&nd->path);
1935 nd->path = path;
2aa38470 1936 nd->inode = path.dentry->d_inode;
165200d6
AV
1937 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1938 return ERR_PTR(-EXDEV);
957dd41d 1939 }
12487f30
AV
1940 /* rare case of legitimate dget_parent()... */
1941 parent = dget_parent(nd->path.dentry);
1942 if (unlikely(!path_connected(nd->path.mnt, parent))) {
1943 dput(parent);
1944 return ERR_PTR(-ENOENT);
1945 }
1946 *seqp = 0;
1947 *inodep = parent->d_inode;
1948 return parent;
1949
1950in_root:
c2df1968
AV
1951 if (unlikely(nd->flags & LOOKUP_BENEATH))
1952 return ERR_PTR(-EXDEV);
1953 dget(nd->path.dentry);
1954 return NULL;
957dd41d
AV
1955}
1956
7521f22b 1957static const char *handle_dots(struct nameidata *nd, int type)
957dd41d
AV
1958{
1959 if (type == LAST_DOTDOT) {
7521f22b 1960 const char *error = NULL;
c2df1968
AV
1961 struct dentry *parent;
1962 struct inode *inode;
1963 unsigned seq;
957dd41d
AV
1964
1965 if (!nd->root.mnt) {
7521f22b 1966 error = ERR_PTR(set_root(nd));
957dd41d
AV
1967 if (error)
1968 return error;
1969 }
1970 if (nd->flags & LOOKUP_RCU)
c2df1968 1971 parent = follow_dotdot_rcu(nd, &inode, &seq);
957dd41d 1972 else
c2df1968
AV
1973 parent = follow_dotdot(nd, &inode, &seq);
1974 if (IS_ERR(parent))
1975 return ERR_CAST(parent);
1976 if (unlikely(!parent))
1977 error = step_into(nd, WALK_NOFOLLOW,
1978 nd->path.dentry, nd->inode, nd->seq);
1979 else
1980 error = step_into(nd, WALK_NOFOLLOW,
1981 parent, inode, seq);
1982 if (unlikely(error))
957dd41d
AV
1983 return error;
1984
1985 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
1986 /*
1987 * If there was a racing rename or mount along our
1988 * path, then we can't be sure that ".." hasn't jumped
1989 * above nd->root (and so userspace should retry or use
1990 * some fallback).
1991 */
1992 smp_rmb();
1993 if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq)))
7521f22b 1994 return ERR_PTR(-EAGAIN);
957dd41d 1995 if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq)))
7521f22b 1996 return ERR_PTR(-EAGAIN);
957dd41d
AV
1997 }
1998 }
7521f22b 1999 return NULL;
957dd41d
AV
2000}
2001
92d27016 2002static const char *walk_component(struct nameidata *nd, int flags)
ce57dfc1 2003{
db3c9ade 2004 struct dentry *dentry;
ce57dfc1 2005 struct inode *inode;
254cf582 2006 unsigned seq;
ce57dfc1
AV
2007 /*
2008 * "." and ".." are special - ".." especially so because it has
2009 * to be able to know about the current root directory and
2010 * parent relationships.
2011 */
4693a547 2012 if (unlikely(nd->last_type != LAST_NORM)) {
1c4ff1a8 2013 if (!(flags & WALK_MORE) && nd->depth)
4693a547 2014 put_link(nd);
7521f22b 2015 return handle_dots(nd, nd->last_type);
4693a547 2016 }
20e34357
AV
2017 dentry = lookup_fast(nd, &inode, &seq);
2018 if (IS_ERR(dentry))
92d27016 2019 return ERR_CAST(dentry);
20e34357 2020 if (unlikely(!dentry)) {
db3c9ade
AV
2021 dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags);
2022 if (IS_ERR(dentry))
92d27016 2023 return ERR_CAST(dentry);
ce57dfc1 2024 }
56676ec3
AV
2025 if (!(flags & WALK_MORE) && nd->depth)
2026 put_link(nd);
b0417d2c 2027 return step_into(nd, flags, dentry, inode, seq);
ce57dfc1
AV
2028}
2029
bfcfaa77
LT
2030/*
2031 * We can do the critical dentry name comparison and hashing
2032 * operations one word at a time, but we are limited to:
2033 *
2034 * - Architectures with fast unaligned word accesses. We could
2035 * do a "get_unaligned()" if this helps and is sufficiently
2036 * fast.
2037 *
bfcfaa77
LT
2038 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
2039 * do not trap on the (extremely unlikely) case of a page
2040 * crossing operation.
2041 *
2042 * - Furthermore, we need an efficient 64-bit compile for the
2043 * 64-bit case in order to generate the "number of bytes in
2044 * the final mask". Again, that could be replaced with a
2045 * efficient population count instruction or similar.
2046 */
2047#ifdef CONFIG_DCACHE_WORD_ACCESS
2048
f68e556e 2049#include <asm/word-at-a-time.h>
bfcfaa77 2050
468a9428 2051#ifdef HASH_MIX
bfcfaa77 2052
468a9428 2053/* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
bfcfaa77 2054
468a9428 2055#elif defined(CONFIG_64BIT)
0fed3ac8 2056/*
2a18da7a
GS
2057 * Register pressure in the mixing function is an issue, particularly
2058 * on 32-bit x86, but almost any function requires one state value and
2059 * one temporary. Instead, use a function designed for two state values
2060 * and no temporaries.
2061 *
2062 * This function cannot create a collision in only two iterations, so
2063 * we have two iterations to achieve avalanche. In those two iterations,
2064 * we have six layers of mixing, which is enough to spread one bit's
2065 * influence out to 2^6 = 64 state bits.
2066 *
2067 * Rotate constants are scored by considering either 64 one-bit input
2068 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
2069 * probability of that delta causing a change to each of the 128 output
2070 * bits, using a sample of random initial states.
2071 *
2072 * The Shannon entropy of the computed probabilities is then summed
2073 * to produce a score. Ideally, any input change has a 50% chance of
2074 * toggling any given output bit.
2075 *
2076 * Mixing scores (in bits) for (12,45):
2077 * Input delta: 1-bit 2-bit
2078 * 1 round: 713.3 42542.6
2079 * 2 rounds: 2753.7 140389.8
2080 * 3 rounds: 5954.1 233458.2
2081 * 4 rounds: 7862.6 256672.2
2082 * Perfect: 8192 258048
2083 * (64*128) (64*63/2 * 128)
0fed3ac8 2084 */
2a18da7a
GS
2085#define HASH_MIX(x, y, a) \
2086 ( x ^= (a), \
2087 y ^= x, x = rol64(x,12),\
2088 x += y, y = rol64(y,45),\
2089 y *= 9 )
bfcfaa77 2090
0fed3ac8 2091/*
2a18da7a
GS
2092 * Fold two longs into one 32-bit hash value. This must be fast, but
2093 * latency isn't quite as critical, as there is a fair bit of additional
2094 * work done before the hash value is used.
0fed3ac8 2095 */
2a18da7a 2096static inline unsigned int fold_hash(unsigned long x, unsigned long y)
0fed3ac8 2097{
2a18da7a
GS
2098 y ^= x * GOLDEN_RATIO_64;
2099 y *= GOLDEN_RATIO_64;
2100 return y >> 32;
0fed3ac8
GS
2101}
2102
bfcfaa77
LT
2103#else /* 32-bit case */
2104
2a18da7a
GS
2105/*
2106 * Mixing scores (in bits) for (7,20):
2107 * Input delta: 1-bit 2-bit
2108 * 1 round: 330.3 9201.6
2109 * 2 rounds: 1246.4 25475.4
2110 * 3 rounds: 1907.1 31295.1
2111 * 4 rounds: 2042.3 31718.6
2112 * Perfect: 2048 31744
2113 * (32*64) (32*31/2 * 64)
2114 */
2115#define HASH_MIX(x, y, a) \
2116 ( x ^= (a), \
2117 y ^= x, x = rol32(x, 7),\
2118 x += y, y = rol32(y,20),\
2119 y *= 9 )
bfcfaa77 2120
2a18da7a 2121static inline unsigned int fold_hash(unsigned long x, unsigned long y)
0fed3ac8 2122{
2a18da7a
GS
2123 /* Use arch-optimized multiply if one exists */
2124 return __hash_32(y ^ __hash_32(x));
0fed3ac8
GS
2125}
2126
bfcfaa77
LT
2127#endif
2128
2a18da7a
GS
2129/*
2130 * Return the hash of a string of known length. This is carfully
2131 * designed to match hash_name(), which is the more critical function.
2132 * In particular, we must end by hashing a final word containing 0..7
2133 * payload bytes, to match the way that hash_name() iterates until it
2134 * finds the delimiter after the name.
2135 */
8387ff25 2136unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
bfcfaa77 2137{
8387ff25 2138 unsigned long a, x = 0, y = (unsigned long)salt;
bfcfaa77
LT
2139
2140 for (;;) {
fcfd2fbf
GS
2141 if (!len)
2142 goto done;
e419b4cc 2143 a = load_unaligned_zeropad(name);
bfcfaa77
LT
2144 if (len < sizeof(unsigned long))
2145 break;
2a18da7a 2146 HASH_MIX(x, y, a);
bfcfaa77
LT
2147 name += sizeof(unsigned long);
2148 len -= sizeof(unsigned long);
bfcfaa77 2149 }
2a18da7a 2150 x ^= a & bytemask_from_count(len);
bfcfaa77 2151done:
2a18da7a 2152 return fold_hash(x, y);
bfcfaa77
LT
2153}
2154EXPORT_SYMBOL(full_name_hash);
2155
fcfd2fbf 2156/* Return the "hash_len" (hash and length) of a null-terminated string */
8387ff25 2157u64 hashlen_string(const void *salt, const char *name)
fcfd2fbf 2158{
8387ff25
LT
2159 unsigned long a = 0, x = 0, y = (unsigned long)salt;
2160 unsigned long adata, mask, len;
fcfd2fbf
GS
2161 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2162
8387ff25
LT
2163 len = 0;
2164 goto inside;
2165
fcfd2fbf 2166 do {
2a18da7a 2167 HASH_MIX(x, y, a);
fcfd2fbf 2168 len += sizeof(unsigned long);
8387ff25 2169inside:
fcfd2fbf
GS
2170 a = load_unaligned_zeropad(name+len);
2171 } while (!has_zero(a, &adata, &constants));
2172
2173 adata = prep_zero_mask(a, adata, &constants);
2174 mask = create_zero_mask(adata);
2a18da7a 2175 x ^= a & zero_bytemask(mask);
fcfd2fbf 2176
2a18da7a 2177 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
fcfd2fbf
GS
2178}
2179EXPORT_SYMBOL(hashlen_string);
2180
bfcfaa77
LT
2181/*
2182 * Calculate the length and hash of the path component, and
d6bb3e90 2183 * return the "hash_len" as the result.
bfcfaa77 2184 */
8387ff25 2185static inline u64 hash_name(const void *salt, const char *name)
bfcfaa77 2186{
8387ff25
LT
2187 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
2188 unsigned long adata, bdata, mask, len;
36126f8f 2189 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
bfcfaa77 2190
8387ff25
LT
2191 len = 0;
2192 goto inside;
2193
bfcfaa77 2194 do {
2a18da7a 2195 HASH_MIX(x, y, a);
bfcfaa77 2196 len += sizeof(unsigned long);
8387ff25 2197inside:
e419b4cc 2198 a = load_unaligned_zeropad(name+len);
36126f8f
LT
2199 b = a ^ REPEAT_BYTE('/');
2200 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
2201
2202 adata = prep_zero_mask(a, adata, &constants);
2203 bdata = prep_zero_mask(b, bdata, &constants);
36126f8f 2204 mask = create_zero_mask(adata | bdata);
2a18da7a 2205 x ^= a & zero_bytemask(mask);
36126f8f 2206
2a18da7a 2207 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
bfcfaa77
LT
2208}
2209
2a18da7a 2210#else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
bfcfaa77 2211
fcfd2fbf 2212/* Return the hash of a string of known length */
8387ff25 2213unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
0145acc2 2214{
8387ff25 2215 unsigned long hash = init_name_hash(salt);
0145acc2 2216 while (len--)
fcfd2fbf 2217 hash = partial_name_hash((unsigned char)*name++, hash);
0145acc2
LT
2218 return end_name_hash(hash);
2219}
ae942ae7 2220EXPORT_SYMBOL(full_name_hash);
0145acc2 2221
fcfd2fbf 2222/* Return the "hash_len" (hash and length) of a null-terminated string */
8387ff25 2223u64 hashlen_string(const void *salt, const char *name)
fcfd2fbf 2224{
8387ff25 2225 unsigned long hash = init_name_hash(salt);
fcfd2fbf
GS
2226 unsigned long len = 0, c;
2227
2228 c = (unsigned char)*name;
e0ab7af9 2229 while (c) {
fcfd2fbf
GS
2230 len++;
2231 hash = partial_name_hash(c, hash);
2232 c = (unsigned char)name[len];
e0ab7af9 2233 }
fcfd2fbf
GS
2234 return hashlen_create(end_name_hash(hash), len);
2235}
f2a031b6 2236EXPORT_SYMBOL(hashlen_string);
fcfd2fbf 2237
200e9ef7
LT
2238/*
2239 * We know there's a real path component here of at least
2240 * one character.
2241 */
8387ff25 2242static inline u64 hash_name(const void *salt, const char *name)
200e9ef7 2243{
8387ff25 2244 unsigned long hash = init_name_hash(salt);
200e9ef7
LT
2245 unsigned long len = 0, c;
2246
2247 c = (unsigned char)*name;
2248 do {
2249 len++;
2250 hash = partial_name_hash(c, hash);
2251 c = (unsigned char)name[len];
2252 } while (c && c != '/');
d6bb3e90 2253 return hashlen_create(end_name_hash(hash), len);
200e9ef7
LT
2254}
2255
bfcfaa77
LT
2256#endif
2257
1da177e4
LT
2258/*
2259 * Name resolution.
ea3834d9
PM
2260 * This is the basic name resolution function, turning a pathname into
2261 * the final dentry. We expect 'base' to be positive and a directory.
1da177e4 2262 *
ea3834d9
PM
2263 * Returns 0 and nd will have valid dentry and mnt on success.
2264 * Returns error and drops reference to input namei data on failure.
1da177e4 2265 */
6de88d72 2266static int link_path_walk(const char *name, struct nameidata *nd)
1da177e4 2267{
d8d4611a 2268 int depth = 0; // depth <= nd->depth
1da177e4 2269 int err;
32cd7468 2270
b4c03536 2271 nd->last_type = LAST_ROOT;
c108837e 2272 nd->flags |= LOOKUP_PARENT;
9b5858e9
AV
2273 if (IS_ERR(name))
2274 return PTR_ERR(name);
1da177e4
LT
2275 while (*name=='/')
2276 name++;
1a97d899
AV
2277 if (!*name) {
2278 nd->dir_mode = 0; // short-circuit the 'hardening' idiocy
9e18f10a 2279 return 0;
1a97d899 2280 }
1da177e4 2281
1da177e4
LT
2282 /* At this point we know we have a real path component. */
2283 for(;;) {
549c7297 2284 struct user_namespace *mnt_userns;
92d27016 2285 const char *link;
d6bb3e90 2286 u64 hash_len;
fe479a58 2287 int type;
1da177e4 2288
549c7297
CB
2289 mnt_userns = mnt_user_ns(nd->path.mnt);
2290 err = may_lookup(mnt_userns, nd);
2a18da7a 2291 if (err)
3595e234 2292 return err;
1da177e4 2293
8387ff25 2294 hash_len = hash_name(nd->path.dentry, name);
1da177e4 2295
fe479a58 2296 type = LAST_NORM;
d6bb3e90 2297 if (name[0] == '.') switch (hashlen_len(hash_len)) {
fe479a58 2298 case 2:
200e9ef7 2299 if (name[1] == '.') {
fe479a58 2300 type = LAST_DOTDOT;
bcba1e7d 2301 nd->state |= ND_JUMPED;
16c2cd71 2302 }
fe479a58
AV
2303 break;
2304 case 1:
2305 type = LAST_DOT;
2306 }
5a202bcd
AV
2307 if (likely(type == LAST_NORM)) {
2308 struct dentry *parent = nd->path.dentry;
bcba1e7d 2309 nd->state &= ~ND_JUMPED;
5a202bcd 2310 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
a060dc50 2311 struct qstr this = { { .hash_len = hash_len }, .name = name };
da53be12 2312 err = parent->d_op->d_hash(parent, &this);
5a202bcd 2313 if (err < 0)
3595e234 2314 return err;
d6bb3e90
LT
2315 hash_len = this.hash_len;
2316 name = this.name;
5a202bcd
AV
2317 }
2318 }
fe479a58 2319
d6bb3e90
LT
2320 nd->last.hash_len = hash_len;
2321 nd->last.name = name;
5f4a6a69
AV
2322 nd->last_type = type;
2323
d6bb3e90
LT
2324 name += hashlen_len(hash_len);
2325 if (!*name)
bdf6cbf1 2326 goto OK;
200e9ef7
LT
2327 /*
2328 * If it wasn't NUL, we know it was '/'. Skip that
2329 * slash, and continue until no more slashes.
2330 */
2331 do {
d6bb3e90
LT
2332 name++;
2333 } while (unlikely(*name == '/'));
8620c238
AV
2334 if (unlikely(!*name)) {
2335OK:
d8d4611a 2336 /* pathname or trailing symlink, done */
c108837e 2337 if (!depth) {
549c7297 2338 nd->dir_uid = i_uid_into_mnt(mnt_userns, nd->inode);
0f705953 2339 nd->dir_mode = nd->inode->i_mode;
c108837e 2340 nd->flags &= ~LOOKUP_PARENT;
8620c238 2341 return 0;
c108837e 2342 }
8620c238 2343 /* last component of nested symlink */
d8d4611a 2344 name = nd->stack[--depth].name;
8c4efe22 2345 link = walk_component(nd, 0);
1c4ff1a8
AV
2346 } else {
2347 /* not the last component */
8c4efe22 2348 link = walk_component(nd, WALK_MORE);
8620c238 2349 }
92d27016
AV
2350 if (unlikely(link)) {
2351 if (IS_ERR(link))
2352 return PTR_ERR(link);
2353 /* a symlink to follow */
d8d4611a 2354 nd->stack[depth++].name = name;
92d27016
AV
2355 name = link;
2356 continue;
31e6b01f 2357 }
97242f99
AV
2358 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2359 if (nd->flags & LOOKUP_RCU) {
e36cffed 2360 if (!try_to_unlazy(nd))
97242f99
AV
2361 return -ECHILD;
2362 }
3595e234 2363 return -ENOTDIR;
97242f99 2364 }
1da177e4 2365 }
1da177e4
LT
2366}
2367
edc2b1da 2368/* must be paired with terminate_walk() */
c8a53ee5 2369static const char *path_init(struct nameidata *nd, unsigned flags)
31e6b01f 2370{
740a1678 2371 int error;
c8a53ee5 2372 const char *s = nd->name->name;
31e6b01f 2373
6c6ec2b0
JA
2374 /* LOOKUP_CACHED requires RCU, ask caller to retry */
2375 if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED)
2376 return ERR_PTR(-EAGAIN);
2377
c0eb027e
LT
2378 if (!*s)
2379 flags &= ~LOOKUP_RCU;
edc2b1da
AV
2380 if (flags & LOOKUP_RCU)
2381 rcu_read_lock();
c0eb027e 2382
bcba1e7d
AV
2383 nd->flags = flags;
2384 nd->state |= ND_JUMPED;
ab87f9a5
AS
2385
2386 nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount);
2387 nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount);
2388 smp_rmb();
2389
bcba1e7d 2390 if (nd->state & ND_ROOT_PRESET) {
b18825a7
DH
2391 struct dentry *root = nd->root.dentry;
2392 struct inode *inode = root->d_inode;
93893862
AV
2393 if (*s && unlikely(!d_can_lookup(root)))
2394 return ERR_PTR(-ENOTDIR);
5b6ca027
AV
2395 nd->path = nd->root;
2396 nd->inode = inode;
2397 if (flags & LOOKUP_RCU) {
ab87f9a5 2398 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
8f47a016 2399 nd->root_seq = nd->seq;
5b6ca027
AV
2400 } else {
2401 path_get(&nd->path);
2402 }
368ee9ba 2403 return s;
5b6ca027
AV
2404 }
2405
31e6b01f 2406 nd->root.mnt = NULL;
31e6b01f 2407
8db52c7e
AS
2408 /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2409 if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) {
740a1678
AS
2410 error = nd_jump_root(nd);
2411 if (unlikely(error))
2412 return ERR_PTR(error);
2413 return s;
8db52c7e
AS
2414 }
2415
2416 /* Relative pathname -- get the starting-point it is relative to. */
2417 if (nd->dfd == AT_FDCWD) {
e41f7d4e
AV
2418 if (flags & LOOKUP_RCU) {
2419 struct fs_struct *fs = current->fs;
2420 unsigned seq;
31e6b01f 2421
e41f7d4e
AV
2422 do {
2423 seq = read_seqcount_begin(&fs->seq);
2424 nd->path = fs->pwd;
ef55d917 2425 nd->inode = nd->path.dentry->d_inode;
e41f7d4e
AV
2426 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2427 } while (read_seqcount_retry(&fs->seq, seq));
2428 } else {
2429 get_fs_pwd(current->fs, &nd->path);
ef55d917 2430 nd->inode = nd->path.dentry->d_inode;
e41f7d4e 2431 }
31e6b01f 2432 } else {
582aa64a 2433 /* Caller must check execute permissions on the starting path component */
c8a53ee5 2434 struct fd f = fdget_raw(nd->dfd);
31e6b01f
NP
2435 struct dentry *dentry;
2436
2903ff01 2437 if (!f.file)
368ee9ba 2438 return ERR_PTR(-EBADF);
31e6b01f 2439
2903ff01 2440 dentry = f.file->f_path.dentry;
31e6b01f 2441
edc2b1da
AV
2442 if (*s && unlikely(!d_can_lookup(dentry))) {
2443 fdput(f);
2444 return ERR_PTR(-ENOTDIR);
f52e0c11 2445 }
31e6b01f 2446
2903ff01 2447 nd->path = f.file->f_path;
e41f7d4e 2448 if (flags & LOOKUP_RCU) {
34a26b99
AV
2449 nd->inode = nd->path.dentry->d_inode;
2450 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
e41f7d4e 2451 } else {
2903ff01 2452 path_get(&nd->path);
34a26b99 2453 nd->inode = nd->path.dentry->d_inode;
e41f7d4e 2454 }
34a26b99 2455 fdput(f);
31e6b01f 2456 }
8db52c7e 2457
adb21d2b
AS
2458 /* For scoped-lookups we need to set the root to the dirfd as well. */
2459 if (flags & LOOKUP_IS_SCOPED) {
2460 nd->root = nd->path;
2461 if (flags & LOOKUP_RCU) {
2462 nd->root_seq = nd->seq;
2463 } else {
2464 path_get(&nd->root);
bcba1e7d 2465 nd->state |= ND_ROOT_GRABBED;
adb21d2b
AS
2466 }
2467 }
2468 return s;
9b4a9b14
AV
2469}
2470
1ccac622 2471static inline const char *lookup_last(struct nameidata *nd)
bd92d7fe
AV
2472{
2473 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2474 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2475
c108837e 2476 return walk_component(nd, WALK_TRAILING);
bd92d7fe
AV
2477}
2478
4f757f3c
AV
2479static int handle_lookup_down(struct nameidata *nd)
2480{
c153007b 2481 if (!(nd->flags & LOOKUP_RCU))
db3c9ade 2482 dget(nd->path.dentry);
b0417d2c
AV
2483 return PTR_ERR(step_into(nd, WALK_NOFOLLOW,
2484 nd->path.dentry, nd->inode, nd->seq));
4f757f3c
AV
2485}
2486
9b4a9b14 2487/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
c8a53ee5 2488static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
9b4a9b14 2489{
c8a53ee5 2490 const char *s = path_init(nd, flags);
bd92d7fe 2491 int err;
31e6b01f 2492
9b5858e9 2493 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
4f757f3c 2494 err = handle_lookup_down(nd);
5f336e72
AV
2495 if (unlikely(err < 0))
2496 s = ERR_PTR(err);
4f757f3c
AV
2497 }
2498
1ccac622
AV
2499 while (!(err = link_path_walk(s, nd)) &&
2500 (s = lookup_last(nd)) != NULL)
2501 ;
4f0ed93f
AV
2502 if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) {
2503 err = handle_lookup_down(nd);
bcba1e7d 2504 nd->state &= ~ND_JUMPED; // no d_weak_revalidate(), please...
4f0ed93f 2505 }
9f1fafee
AV
2506 if (!err)
2507 err = complete_walk(nd);
bd92d7fe 2508
deb106c6
AV
2509 if (!err && nd->flags & LOOKUP_DIRECTORY)
2510 if (!d_can_lookup(nd->path.dentry))
bd23a539 2511 err = -ENOTDIR;
625b6d10
AV
2512 if (!err) {
2513 *path = nd->path;
2514 nd->path.mnt = NULL;
2515 nd->path.dentry = NULL;
2516 }
2517 terminate_walk(nd);
bd92d7fe 2518 return err;
ee0827cd 2519}
31e6b01f 2520
794ebcea 2521int filename_lookup(int dfd, struct filename *name, unsigned flags,
31d921c7 2522 struct path *path, struct path *root)
ee0827cd 2523{
894bc8c4 2524 int retval;
9883d185 2525 struct nameidata nd;
abc9f5be
AV
2526 if (IS_ERR(name))
2527 return PTR_ERR(name);
06422964 2528 set_nameidata(&nd, dfd, name, root);
c8a53ee5 2529 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
ee0827cd 2530 if (unlikely(retval == -ECHILD))
c8a53ee5 2531 retval = path_lookupat(&nd, flags, path);
ee0827cd 2532 if (unlikely(retval == -ESTALE))
c8a53ee5 2533 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
31e6b01f 2534
f78570dd 2535 if (likely(!retval))
161aff1d
AV
2536 audit_inode(name, path->dentry,
2537 flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0);
9883d185 2538 restore_nameidata();
020250f3
DK
2539 return retval;
2540}
2541
8bcb77fa 2542/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
c8a53ee5 2543static int path_parentat(struct nameidata *nd, unsigned flags,
391172c4 2544 struct path *parent)
8bcb77fa 2545{
c8a53ee5 2546 const char *s = path_init(nd, flags);
9b5858e9 2547 int err = link_path_walk(s, nd);
8bcb77fa
AV
2548 if (!err)
2549 err = complete_walk(nd);
391172c4
AV
2550 if (!err) {
2551 *parent = nd->path;
2552 nd->path.mnt = NULL;
2553 nd->path.dentry = NULL;
2554 }
2555 terminate_walk(nd);
8bcb77fa
AV
2556 return err;
2557}
2558
0766ec82 2559/* Note: this does not consume "name" */
c5f563f9 2560static int filename_parentat(int dfd, struct filename *name,
0766ec82
SB
2561 unsigned int flags, struct path *parent,
2562 struct qstr *last, int *type)
8bcb77fa
AV
2563{
2564 int retval;
9883d185 2565 struct nameidata nd;
8bcb77fa 2566
5c31b6ce 2567 if (IS_ERR(name))
0ee50b47 2568 return PTR_ERR(name);
06422964 2569 set_nameidata(&nd, dfd, name, NULL);
c8a53ee5 2570 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
8bcb77fa 2571 if (unlikely(retval == -ECHILD))
c8a53ee5 2572 retval = path_parentat(&nd, flags, parent);
8bcb77fa 2573 if (unlikely(retval == -ESTALE))
c8a53ee5 2574 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
391172c4
AV
2575 if (likely(!retval)) {
2576 *last = nd.last;
2577 *type = nd.last_type;
c9b07eab 2578 audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
391172c4 2579 }
9883d185 2580 restore_nameidata();
0ee50b47
DK
2581 return retval;
2582}
2583
79714f72 2584/* does lookup, returns the object with parent locked */
0766ec82 2585static struct dentry *__kern_path_locked(struct filename *name, struct path *path)
5590ff0d 2586{
5c31b6ce 2587 struct dentry *d;
391172c4 2588 struct qstr last;
0ee50b47 2589 int type, error;
51689104 2590
c5f563f9 2591 error = filename_parentat(AT_FDCWD, name, 0, path, &last, &type);
0ee50b47
DK
2592 if (error)
2593 return ERR_PTR(error);
5c31b6ce 2594 if (unlikely(type != LAST_NORM)) {
391172c4 2595 path_put(path);
5c31b6ce 2596 return ERR_PTR(-EINVAL);
79714f72 2597 }
5955102c 2598 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
391172c4 2599 d = __lookup_hash(&last, path->dentry, 0);
79714f72 2600 if (IS_ERR(d)) {
5955102c 2601 inode_unlock(path->dentry->d_inode);
391172c4 2602 path_put(path);
79714f72 2603 }
79714f72 2604 return d;
5590ff0d
UD
2605}
2606
0766ec82
SB
2607struct dentry *kern_path_locked(const char *name, struct path *path)
2608{
2609 struct filename *filename = getname_kernel(name);
2610 struct dentry *res = __kern_path_locked(filename, path);
2611
2612 putname(filename);
2613 return res;
2614}
2615
d1811465
AV
2616int kern_path(const char *name, unsigned int flags, struct path *path)
2617{
794ebcea
SB
2618 struct filename *filename = getname_kernel(name);
2619 int ret = filename_lookup(AT_FDCWD, filename, flags, path, NULL);
2620
2621 putname(filename);
2622 return ret;
2623
d1811465 2624}
4d359507 2625EXPORT_SYMBOL(kern_path);
d1811465 2626
16f18200
JJS
2627/**
2628 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2629 * @dentry: pointer to dentry of the base directory
2630 * @mnt: pointer to vfs mount of the base directory
2631 * @name: pointer to file name
2632 * @flags: lookup flags
e0a01249 2633 * @path: pointer to struct path to fill
16f18200
JJS
2634 */
2635int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2636 const char *name, unsigned int flags,
e0a01249 2637 struct path *path)
16f18200 2638{
794ebcea 2639 struct filename *filename;
9ad1aaa6 2640 struct path root = {.mnt = mnt, .dentry = dentry};
794ebcea
SB
2641 int ret;
2642
2643 filename = getname_kernel(name);
9ad1aaa6 2644 /* the first argument of filename_lookup() is ignored with root */
794ebcea
SB
2645 ret = filename_lookup(AT_FDCWD, filename, flags, path, &root);
2646 putname(filename);
2647 return ret;
16f18200 2648}
4d359507 2649EXPORT_SYMBOL(vfs_path_lookup);
16f18200 2650
c2fd68b6
CB
2651static int lookup_one_common(struct user_namespace *mnt_userns,
2652 const char *name, struct dentry *base, int len,
2653 struct qstr *this)
057f6c01 2654{
3c95f0dc
AV
2655 this->name = name;
2656 this->len = len;
2657 this->hash = full_name_hash(base, name, len);
6a96ba54 2658 if (!len)
3c95f0dc 2659 return -EACCES;
6a96ba54 2660
21d8a15a
AV
2661 if (unlikely(name[0] == '.')) {
2662 if (len < 2 || (len == 2 && name[1] == '.'))
3c95f0dc 2663 return -EACCES;
21d8a15a
AV
2664 }
2665
6a96ba54 2666 while (len--) {
3c95f0dc 2667 unsigned int c = *(const unsigned char *)name++;
6a96ba54 2668 if (c == '/' || c == '\0')
3c95f0dc 2669 return -EACCES;
6a96ba54 2670 }
5a202bcd
AV
2671 /*
2672 * See if the low-level filesystem might want
2673 * to use its own hash..
2674 */
2675 if (base->d_flags & DCACHE_OP_HASH) {
3c95f0dc 2676 int err = base->d_op->d_hash(base, this);
5a202bcd 2677 if (err < 0)
3c95f0dc 2678 return err;
5a202bcd 2679 }
eead1911 2680
c2fd68b6 2681 return inode_permission(mnt_userns, base->d_inode, MAY_EXEC);
3c95f0dc
AV
2682}
2683
0da0b7fd
DH
2684/**
2685 * try_lookup_one_len - filesystem helper to lookup single pathname component
2686 * @name: pathname component to lookup
2687 * @base: base directory to lookup from
2688 * @len: maximum length @len should be interpreted to
2689 *
2690 * Look up a dentry by name in the dcache, returning NULL if it does not
2691 * currently exist. The function does not try to create a dentry.
2692 *
2693 * Note that this routine is purely a helper for filesystem usage and should
2694 * not be called by generic code.
2695 *
2696 * The caller must hold base->i_mutex.
2697 */
2698struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2699{
2700 struct qstr this;
2701 int err;
2702
2703 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2704
c2fd68b6 2705 err = lookup_one_common(&init_user_ns, name, base, len, &this);
0da0b7fd
DH
2706 if (err)
2707 return ERR_PTR(err);
2708
2709 return lookup_dcache(&this, base, 0);
2710}
2711EXPORT_SYMBOL(try_lookup_one_len);
2712
3c95f0dc
AV
2713/**
2714 * lookup_one_len - filesystem helper to lookup single pathname component
2715 * @name: pathname component to lookup
2716 * @base: base directory to lookup from
2717 * @len: maximum length @len should be interpreted to
2718 *
2719 * Note that this routine is purely a helper for filesystem usage and should
2720 * not be called by generic code.
2721 *
2722 * The caller must hold base->i_mutex.
2723 */
2724struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2725{
8613a209 2726 struct dentry *dentry;
3c95f0dc
AV
2727 struct qstr this;
2728 int err;
2729
2730 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2731
c2fd68b6 2732 err = lookup_one_common(&init_user_ns, name, base, len, &this);
cda309de
MS
2733 if (err)
2734 return ERR_PTR(err);
2735
8613a209
AV
2736 dentry = lookup_dcache(&this, base, 0);
2737 return dentry ? dentry : __lookup_slow(&this, base, 0);
057f6c01 2738}
4d359507 2739EXPORT_SYMBOL(lookup_one_len);
057f6c01 2740
c2fd68b6
CB
2741/**
2742 * lookup_one - filesystem helper to lookup single pathname component
2743 * @mnt_userns: user namespace of the mount the lookup is performed from
2744 * @name: pathname component to lookup
2745 * @base: base directory to lookup from
2746 * @len: maximum length @len should be interpreted to
2747 *
2748 * Note that this routine is purely a helper for filesystem usage and should
2749 * not be called by generic code.
2750 *
2751 * The caller must hold base->i_mutex.
2752 */
2753struct dentry *lookup_one(struct user_namespace *mnt_userns, const char *name,
2754 struct dentry *base, int len)
2755{
2756 struct dentry *dentry;
2757 struct qstr this;
2758 int err;
2759
2760 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2761
2762 err = lookup_one_common(mnt_userns, name, base, len, &this);
2763 if (err)
2764 return ERR_PTR(err);
2765
2766 dentry = lookup_dcache(&this, base, 0);
2767 return dentry ? dentry : __lookup_slow(&this, base, 0);
2768}
2769EXPORT_SYMBOL(lookup_one);
2770
bbddca8e 2771/**
00675017
CB
2772 * lookup_one_unlocked - filesystem helper to lookup single pathname component
2773 * @mnt_userns: idmapping of the mount the lookup is performed from
bbddca8e
N
2774 * @name: pathname component to lookup
2775 * @base: base directory to lookup from
2776 * @len: maximum length @len should be interpreted to
2777 *
2778 * Note that this routine is purely a helper for filesystem usage and should
2779 * not be called by generic code.
2780 *
2781 * Unlike lookup_one_len, it should be called without the parent
2782 * i_mutex held, and will take the i_mutex itself if necessary.
2783 */
00675017
CB
2784struct dentry *lookup_one_unlocked(struct user_namespace *mnt_userns,
2785 const char *name, struct dentry *base,
2786 int len)
bbddca8e
N
2787{
2788 struct qstr this;
bbddca8e 2789 int err;
20d00ee8 2790 struct dentry *ret;
bbddca8e 2791
00675017 2792 err = lookup_one_common(mnt_userns, name, base, len, &this);
bbddca8e
N
2793 if (err)
2794 return ERR_PTR(err);
2795
20d00ee8
LT
2796 ret = lookup_dcache(&this, base, 0);
2797 if (!ret)
2798 ret = lookup_slow(&this, base, 0);
2799 return ret;
bbddca8e 2800}
00675017
CB
2801EXPORT_SYMBOL(lookup_one_unlocked);
2802
2803/**
2804 * lookup_one_positive_unlocked - filesystem helper to lookup single
2805 * pathname component
2806 * @mnt_userns: idmapping of the mount the lookup is performed from
2807 * @name: pathname component to lookup
2808 * @base: base directory to lookup from
2809 * @len: maximum length @len should be interpreted to
2810 *
2811 * This helper will yield ERR_PTR(-ENOENT) on negatives. The helper returns
2812 * known positive or ERR_PTR(). This is what most of the users want.
2813 *
2814 * Note that pinned negative with unlocked parent _can_ become positive at any
2815 * time, so callers of lookup_one_unlocked() need to be very careful; pinned
2816 * positives have >d_inode stable, so this one avoids such problems.
2817 *
2818 * Note that this routine is purely a helper for filesystem usage and should
2819 * not be called by generic code.
2820 *
2821 * The helper should be called without i_mutex held.
2822 */
2823struct dentry *lookup_one_positive_unlocked(struct user_namespace *mnt_userns,
2824 const char *name,
2825 struct dentry *base, int len)
2826{
2827 struct dentry *ret = lookup_one_unlocked(mnt_userns, name, base, len);
2828
2829 if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
2830 dput(ret);
2831 ret = ERR_PTR(-ENOENT);
2832 }
2833 return ret;
2834}
2835EXPORT_SYMBOL(lookup_one_positive_unlocked);
2836
2837/**
2838 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2839 * @name: pathname component to lookup
2840 * @base: base directory to lookup from
2841 * @len: maximum length @len should be interpreted to
2842 *
2843 * Note that this routine is purely a helper for filesystem usage and should
2844 * not be called by generic code.
2845 *
2846 * Unlike lookup_one_len, it should be called without the parent
2847 * i_mutex held, and will take the i_mutex itself if necessary.
2848 */
2849struct dentry *lookup_one_len_unlocked(const char *name,
2850 struct dentry *base, int len)
2851{
2852 return lookup_one_unlocked(&init_user_ns, name, base, len);
2853}
bbddca8e
N
2854EXPORT_SYMBOL(lookup_one_len_unlocked);
2855
6c2d4798
AV
2856/*
2857 * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2858 * on negatives. Returns known positive or ERR_PTR(); that's what
2859 * most of the users want. Note that pinned negative with unlocked parent
2860 * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2861 * need to be very careful; pinned positives have ->d_inode stable, so
2862 * this one avoids such problems.
2863 */
2864struct dentry *lookup_positive_unlocked(const char *name,
2865 struct dentry *base, int len)
2866{
00675017 2867 return lookup_one_positive_unlocked(&init_user_ns, name, base, len);
6c2d4798
AV
2868}
2869EXPORT_SYMBOL(lookup_positive_unlocked);
2870
eedf265a
EB
2871#ifdef CONFIG_UNIX98_PTYS
2872int path_pts(struct path *path)
2873{
2874 /* Find something mounted on "pts" in the same directory as
2875 * the input path.
2876 */
a6a7eb76
AV
2877 struct dentry *parent = dget_parent(path->dentry);
2878 struct dentry *child;
19f6028a 2879 struct qstr this = QSTR_INIT("pts", 3);
eedf265a 2880
a6a7eb76
AV
2881 if (unlikely(!path_connected(path->mnt, parent))) {
2882 dput(parent);
63b27720 2883 return -ENOENT;
a6a7eb76 2884 }
63b27720
AV
2885 dput(path->dentry);
2886 path->dentry = parent;
eedf265a
EB
2887 child = d_hash_and_lookup(parent, &this);
2888 if (!child)
2889 return -ENOENT;
2890
2891 path->dentry = child;
2892 dput(parent);
19f6028a 2893 follow_down(path);
eedf265a
EB
2894 return 0;
2895}
2896#endif
2897
1fa1e7f6
AW
2898int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2899 struct path *path, int *empty)
1da177e4 2900{
794ebcea
SB
2901 struct filename *filename = getname_flags(name, flags, empty);
2902 int ret = filename_lookup(dfd, filename, flags, path, NULL);
2903
2904 putname(filename);
2905 return ret;
1da177e4 2906}
b853a161 2907EXPORT_SYMBOL(user_path_at_empty);
1fa1e7f6 2908
ba73d987
CB
2909int __check_sticky(struct user_namespace *mnt_userns, struct inode *dir,
2910 struct inode *inode)
1da177e4 2911{
8e96e3b7 2912 kuid_t fsuid = current_fsuid();
da9592ed 2913
ba73d987 2914 if (uid_eq(i_uid_into_mnt(mnt_userns, inode), fsuid))
1da177e4 2915 return 0;
ba73d987 2916 if (uid_eq(i_uid_into_mnt(mnt_userns, dir), fsuid))
1da177e4 2917 return 0;
ba73d987 2918 return !capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FOWNER);
1da177e4 2919}
cbdf35bc 2920EXPORT_SYMBOL(__check_sticky);
1da177e4
LT
2921
2922/*
2923 * Check whether we can remove a link victim from directory dir, check
2924 * whether the type of victim is right.
2925 * 1. We can't do it if dir is read-only (done in permission())
2926 * 2. We should have write and exec permissions on dir
2927 * 3. We can't remove anything from append-only dir
2928 * 4. We can't do anything with immutable dir (done in permission())
2929 * 5. If the sticky bit on dir is set we should either
2930 * a. be owner of dir, or
2931 * b. be owner of victim, or
2932 * c. have CAP_FOWNER capability
2933 * 6. If the victim is append-only or immutable we can't do antyhing with
2934 * links pointing to it.
0bd23d09
EB
2935 * 7. If the victim has an unknown uid or gid we can't change the inode.
2936 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2937 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2938 * 10. We can't remove a root or mountpoint.
2939 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
1da177e4
LT
2940 * nfs_async_unlink().
2941 */
ba73d987
CB
2942static int may_delete(struct user_namespace *mnt_userns, struct inode *dir,
2943 struct dentry *victim, bool isdir)
1da177e4 2944{
63afdfc7 2945 struct inode *inode = d_backing_inode(victim);
1da177e4
LT
2946 int error;
2947
b18825a7 2948 if (d_is_negative(victim))
1da177e4 2949 return -ENOENT;
b18825a7 2950 BUG_ON(!inode);
1da177e4
LT
2951
2952 BUG_ON(victim->d_parent->d_inode != dir);
593d1ce8
EB
2953
2954 /* Inode writeback is not safe when the uid or gid are invalid. */
ba73d987
CB
2955 if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) ||
2956 !gid_valid(i_gid_into_mnt(mnt_userns, inode)))
593d1ce8
EB
2957 return -EOVERFLOW;
2958
4fa6b5ec 2959 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
1da177e4 2960
ba73d987 2961 error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT
2962 if (error)
2963 return error;
2964 if (IS_APPEND(dir))
2965 return -EPERM;
b18825a7 2966
ba73d987
CB
2967 if (check_sticky(mnt_userns, dir, inode) || IS_APPEND(inode) ||
2968 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) ||
2969 HAS_UNMAPPED_ID(mnt_userns, inode))
1da177e4
LT
2970 return -EPERM;
2971 if (isdir) {
44b1d530 2972 if (!d_is_dir(victim))
1da177e4
LT
2973 return -ENOTDIR;
2974 if (IS_ROOT(victim))
2975 return -EBUSY;
44b1d530 2976 } else if (d_is_dir(victim))
1da177e4
LT
2977 return -EISDIR;
2978 if (IS_DEADDIR(dir))
2979 return -ENOENT;
2980 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2981 return -EBUSY;
2982 return 0;
2983}
2984
2985/* Check whether we can create an object with dentry child in directory
2986 * dir.
2987 * 1. We can't do it if child already exists (open has special treatment for
2988 * this case, but since we are inlined it's OK)
2989 * 2. We can't do it if dir is read-only (done in permission())
036d5236
EB
2990 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2991 * 4. We should have write and exec permissions on dir
2992 * 5. We can't do it if dir is immutable (done in permission())
1da177e4 2993 */
ba73d987
CB
2994static inline int may_create(struct user_namespace *mnt_userns,
2995 struct inode *dir, struct dentry *child)
1da177e4 2996{
14e972b4 2997 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
1da177e4
LT
2998 if (child->d_inode)
2999 return -EEXIST;
3000 if (IS_DEADDIR(dir))
3001 return -ENOENT;
8e538913 3002 if (!fsuidgid_has_mapping(dir->i_sb, mnt_userns))
036d5236 3003 return -EOVERFLOW;
8e538913 3004
ba73d987 3005 return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT
3006}
3007
1da177e4
LT
3008/*
3009 * p1 and p2 should be directories on the same fs.
3010 */
3011struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
3012{
3013 struct dentry *p;
3014
3015 if (p1 == p2) {
5955102c 3016 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
1da177e4
LT
3017 return NULL;
3018 }
3019
fc64005c 3020 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
1da177e4 3021
e2761a11
OH
3022 p = d_ancestor(p2, p1);
3023 if (p) {
5955102c
AV
3024 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
3025 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
e2761a11 3026 return p;
1da177e4
LT
3027 }
3028
e2761a11
OH
3029 p = d_ancestor(p1, p2);
3030 if (p) {
5955102c
AV
3031 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3032 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
e2761a11 3033 return p;
1da177e4
LT
3034 }
3035
5955102c
AV
3036 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3037 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
1da177e4
LT
3038 return NULL;
3039}
4d359507 3040EXPORT_SYMBOL(lock_rename);
1da177e4
LT
3041
3042void unlock_rename(struct dentry *p1, struct dentry *p2)
3043{
5955102c 3044 inode_unlock(p1->d_inode);
1da177e4 3045 if (p1 != p2) {
5955102c 3046 inode_unlock(p2->d_inode);
fc64005c 3047 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
1da177e4
LT
3048 }
3049}
4d359507 3050EXPORT_SYMBOL(unlock_rename);
1da177e4 3051
6521f891
CB
3052/**
3053 * vfs_create - create new file
3054 * @mnt_userns: user namespace of the mount the inode was found from
3055 * @dir: inode of @dentry
3056 * @dentry: pointer to dentry of the base directory
3057 * @mode: mode of the new file
3058 * @want_excl: whether the file must not yet exist
3059 *
3060 * Create a new file.
3061 *
3062 * If the inode has been found through an idmapped mount the user namespace of
3063 * the vfsmount must be passed through @mnt_userns. This function will then take
3064 * care to map the inode according to @mnt_userns before checking permissions.
3065 * On non-idmapped mounts or if permission checking is to be performed on the
3066 * raw inode simply passs init_user_ns.
3067 */
3068int vfs_create(struct user_namespace *mnt_userns, struct inode *dir,
3069 struct dentry *dentry, umode_t mode, bool want_excl)
1da177e4 3070{
6521f891 3071 int error = may_create(mnt_userns, dir, dentry);
1da177e4
LT
3072 if (error)
3073 return error;
3074
acfa4380 3075 if (!dir->i_op->create)
1da177e4
LT
3076 return -EACCES; /* shouldn't it be ENOSYS? */
3077 mode &= S_IALLUGO;
3078 mode |= S_IFREG;
3079 error = security_inode_create(dir, dentry, mode);
3080 if (error)
3081 return error;
549c7297 3082 error = dir->i_op->create(mnt_userns, dir, dentry, mode, want_excl);
a74574aa 3083 if (!error)
f38aa942 3084 fsnotify_create(dir, dentry);
1da177e4
LT
3085 return error;
3086}
4d359507 3087EXPORT_SYMBOL(vfs_create);
1da177e4 3088
8e6c848e
AV
3089int vfs_mkobj(struct dentry *dentry, umode_t mode,
3090 int (*f)(struct dentry *, umode_t, void *),
3091 void *arg)
3092{
3093 struct inode *dir = dentry->d_parent->d_inode;
ba73d987 3094 int error = may_create(&init_user_ns, dir, dentry);
8e6c848e
AV
3095 if (error)
3096 return error;
3097
3098 mode &= S_IALLUGO;
3099 mode |= S_IFREG;
3100 error = security_inode_create(dir, dentry, mode);
3101 if (error)
3102 return error;
3103 error = f(dentry, mode, arg);
3104 if (!error)
3105 fsnotify_create(dir, dentry);
3106 return error;
3107}
3108EXPORT_SYMBOL(vfs_mkobj);
3109
a2982cc9
EB
3110bool may_open_dev(const struct path *path)
3111{
3112 return !(path->mnt->mnt_flags & MNT_NODEV) &&
3113 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
3114}
3115
ba73d987
CB
3116static int may_open(struct user_namespace *mnt_userns, const struct path *path,
3117 int acc_mode, int flag)
1da177e4 3118{
3fb64190 3119 struct dentry *dentry = path->dentry;
1da177e4
LT
3120 struct inode *inode = dentry->d_inode;
3121 int error;
3122
3123 if (!inode)
3124 return -ENOENT;
3125
c8fe8f30
CH
3126 switch (inode->i_mode & S_IFMT) {
3127 case S_IFLNK:
1da177e4 3128 return -ELOOP;
c8fe8f30 3129 case S_IFDIR:
fc4177be 3130 if (acc_mode & MAY_WRITE)
c8fe8f30 3131 return -EISDIR;
fc4177be
KC
3132 if (acc_mode & MAY_EXEC)
3133 return -EACCES;
c8fe8f30
CH
3134 break;
3135 case S_IFBLK:
3136 case S_IFCHR:
a2982cc9 3137 if (!may_open_dev(path))
1da177e4 3138 return -EACCES;
633fb6ac 3139 fallthrough;
c8fe8f30
CH
3140 case S_IFIFO:
3141 case S_IFSOCK:
633fb6ac
KC
3142 if (acc_mode & MAY_EXEC)
3143 return -EACCES;
1da177e4 3144 flag &= ~O_TRUNC;
c8fe8f30 3145 break;
0fd338b2
KC
3146 case S_IFREG:
3147 if ((acc_mode & MAY_EXEC) && path_noexec(path))
3148 return -EACCES;
3149 break;
4a3fd211 3150 }
b41572e9 3151
ba73d987 3152 error = inode_permission(mnt_userns, inode, MAY_OPEN | acc_mode);
b41572e9
DH
3153 if (error)
3154 return error;
6146f0d5 3155
1da177e4
LT
3156 /*
3157 * An append-only file must be opened in append mode for writing.
3158 */
3159 if (IS_APPEND(inode)) {
8737c930 3160 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
7715b521 3161 return -EPERM;
1da177e4 3162 if (flag & O_TRUNC)
7715b521 3163 return -EPERM;
1da177e4
LT
3164 }
3165
3166 /* O_NOATIME can only be set by the owner or superuser */
ba73d987 3167 if (flag & O_NOATIME && !inode_owner_or_capable(mnt_userns, inode))
7715b521 3168 return -EPERM;
1da177e4 3169
f3c7691e 3170 return 0;
7715b521 3171}
1da177e4 3172
549c7297 3173static int handle_truncate(struct user_namespace *mnt_userns, struct file *filp)
7715b521 3174{
f0bb5aaf 3175 const struct path *path = &filp->f_path;
7715b521
AV
3176 struct inode *inode = path->dentry->d_inode;
3177 int error = get_write_access(inode);
3178 if (error)
3179 return error;
482e0007 3180
f7e33bdb 3181 error = security_path_truncate(path);
7715b521 3182 if (!error) {
549c7297 3183 error = do_truncate(mnt_userns, path->dentry, 0,
7715b521 3184 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
e1181ee6 3185 filp);
7715b521
AV
3186 }
3187 put_write_access(inode);
acd0c935 3188 return error;
1da177e4
LT
3189}
3190
d57999e1
DH
3191static inline int open_to_namei_flags(int flag)
3192{
8a5e929d
AV
3193 if ((flag & O_ACCMODE) == 3)
3194 flag--;
d57999e1
DH
3195 return flag;
3196}
3197
ba73d987
CB
3198static int may_o_create(struct user_namespace *mnt_userns,
3199 const struct path *dir, struct dentry *dentry,
3200 umode_t mode)
d18e9008
MS
3201{
3202 int error = security_path_mknod(dir, dentry, mode, 0);
3203 if (error)
3204 return error;
3205
8e538913 3206 if (!fsuidgid_has_mapping(dir->dentry->d_sb, mnt_userns))
1328c727
SF
3207 return -EOVERFLOW;
3208
ba73d987 3209 error = inode_permission(mnt_userns, dir->dentry->d_inode,
47291baa 3210 MAY_WRITE | MAY_EXEC);
d18e9008
MS
3211 if (error)
3212 return error;
3213
3214 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3215}
3216
1acf0af9
DH
3217/*
3218 * Attempt to atomically look up, create and open a file from a negative
3219 * dentry.
3220 *
3221 * Returns 0 if successful. The file will have been created and attached to
3222 * @file by the filesystem calling finish_open().
3223 *
00a07c15
AV
3224 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3225 * be set. The caller will need to perform the open themselves. @path will
3226 * have been updated to point to the new dentry. This may be negative.
1acf0af9
DH
3227 *
3228 * Returns an error code otherwise.
3229 */
239eb983
AV
3230static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry,
3231 struct file *file,
239eb983 3232 int open_flag, umode_t mode)
d18e9008 3233{
384f26e2 3234 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
d18e9008 3235 struct inode *dir = nd->path.dentry->d_inode;
d18e9008 3236 int error;
d18e9008 3237
d18e9008
MS
3238 if (nd->flags & LOOKUP_DIRECTORY)
3239 open_flag |= O_DIRECTORY;
3240
30d90494
AV
3241 file->f_path.dentry = DENTRY_NOT_SET;
3242 file->f_path.mnt = nd->path.mnt;
0fb1ea09 3243 error = dir->i_op->atomic_open(dir, dentry, file,
44907d79 3244 open_to_namei_flags(open_flag), mode);
6fbd0714 3245 d_lookup_done(dentry);
384f26e2 3246 if (!error) {
64e1ac4d 3247 if (file->f_mode & FMODE_OPENED) {
6fb968cd
AV
3248 if (unlikely(dentry != file->f_path.dentry)) {
3249 dput(dentry);
3250 dentry = dget(file->f_path.dentry);
3251 }
64e1ac4d 3252 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2675a4eb 3253 error = -EIO;
03da633a 3254 } else {
384f26e2
AV
3255 if (file->f_path.dentry) {
3256 dput(dentry);
3257 dentry = file->f_path.dentry;
03da633a 3258 }
239eb983 3259 if (unlikely(d_is_negative(dentry)))
a01e718f 3260 error = -ENOENT;
62b2ce96 3261 }
d18e9008 3262 }
239eb983
AV
3263 if (error) {
3264 dput(dentry);
3265 dentry = ERR_PTR(error);
3266 }
3267 return dentry;
d18e9008
MS
3268}
3269
d58ffd35 3270/*
1acf0af9 3271 * Look up and maybe create and open the last component.
d58ffd35 3272 *
00a07c15 3273 * Must be called with parent locked (exclusive in O_CREAT case).
1acf0af9 3274 *
00a07c15
AV
3275 * Returns 0 on success, that is, if
3276 * the file was successfully atomically created (if necessary) and opened, or
3277 * the file was not completely opened at this time, though lookups and
3278 * creations were performed.
3279 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3280 * In the latter case dentry returned in @path might be negative if O_CREAT
3281 * hadn't been specified.
1acf0af9 3282 *
00a07c15 3283 * An error code is returned on failure.
d58ffd35 3284 */
da5ebf5a
AV
3285static struct dentry *lookup_open(struct nameidata *nd, struct file *file,
3286 const struct open_flags *op,
3287 bool got_write)
d58ffd35 3288{
549c7297 3289 struct user_namespace *mnt_userns;
d58ffd35 3290 struct dentry *dir = nd->path.dentry;
54ef4872 3291 struct inode *dir_inode = dir->d_inode;
1643b43f 3292 int open_flag = op->open_flag;
d58ffd35 3293 struct dentry *dentry;
1643b43f 3294 int error, create_error = 0;
1643b43f 3295 umode_t mode = op->mode;
6fbd0714 3296 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
d58ffd35 3297
ce8644fc 3298 if (unlikely(IS_DEADDIR(dir_inode)))
da5ebf5a 3299 return ERR_PTR(-ENOENT);
d58ffd35 3300
73a09dd9 3301 file->f_mode &= ~FMODE_CREATED;
6fbd0714
AV
3302 dentry = d_lookup(dir, &nd->last);
3303 for (;;) {
3304 if (!dentry) {
3305 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3306 if (IS_ERR(dentry))
da5ebf5a 3307 return dentry;
6fbd0714
AV
3308 }
3309 if (d_in_lookup(dentry))
3310 break;
d58ffd35 3311
6fbd0714
AV
3312 error = d_revalidate(dentry, nd->flags);
3313 if (likely(error > 0))
3314 break;
3315 if (error)
3316 goto out_dput;
3317 d_invalidate(dentry);
3318 dput(dentry);
3319 dentry = NULL;
3320 }
3321 if (dentry->d_inode) {
6c51e513 3322 /* Cached positive dentry: will open in f_op->open */
da5ebf5a 3323 return dentry;
6c51e513 3324 }
d18e9008 3325
1643b43f
AV
3326 /*
3327 * Checking write permission is tricky, bacuse we don't know if we are
3328 * going to actually need it: O_CREAT opens should work as long as the
3329 * file exists. But checking existence breaks atomicity. The trick is
3330 * to check access and if not granted clear O_CREAT from the flags.
3331 *
3332 * Another problem is returing the "right" error value (e.g. for an
3333 * O_EXCL open we want to return EEXIST not EROFS).
3334 */
99a4a90c
AV
3335 if (unlikely(!got_write))
3336 open_flag &= ~O_TRUNC;
549c7297 3337 mnt_userns = mnt_user_ns(nd->path.mnt);
1643b43f 3338 if (open_flag & O_CREAT) {
99a4a90c
AV
3339 if (open_flag & O_EXCL)
3340 open_flag &= ~O_TRUNC;
1643b43f
AV
3341 if (!IS_POSIXACL(dir->d_inode))
3342 mode &= ~current_umask();
99a4a90c 3343 if (likely(got_write))
549c7297 3344 create_error = may_o_create(mnt_userns, &nd->path,
ba73d987 3345 dentry, mode);
99a4a90c
AV
3346 else
3347 create_error = -EROFS;
d18e9008 3348 }
99a4a90c
AV
3349 if (create_error)
3350 open_flag &= ~O_CREAT;
6ac08709 3351 if (dir_inode->i_op->atomic_open) {
d489cf9a 3352 dentry = atomic_open(nd, dentry, file, open_flag, mode);
da5ebf5a
AV
3353 if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT))
3354 dentry = ERR_PTR(create_error);
3355 return dentry;
d18e9008 3356 }
54ef4872 3357
6fbd0714 3358 if (d_in_lookup(dentry)) {
12fa5e24
AV
3359 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3360 nd->flags);
6fbd0714 3361 d_lookup_done(dentry);
12fa5e24
AV
3362 if (unlikely(res)) {
3363 if (IS_ERR(res)) {
3364 error = PTR_ERR(res);
3365 goto out_dput;
3366 }
3367 dput(dentry);
3368 dentry = res;
3369 }
54ef4872
MS
3370 }
3371
d58ffd35 3372 /* Negative dentry, just create the file */
1643b43f 3373 if (!dentry->d_inode && (open_flag & O_CREAT)) {
73a09dd9 3374 file->f_mode |= FMODE_CREATED;
ce8644fc 3375 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
ce8644fc
AV
3376 if (!dir_inode->i_op->create) {
3377 error = -EACCES;
d58ffd35 3378 goto out_dput;
ce8644fc 3379 }
549c7297
CB
3380
3381 error = dir_inode->i_op->create(mnt_userns, dir_inode, dentry,
3382 mode, open_flag & O_EXCL);
d58ffd35
MS
3383 if (error)
3384 goto out_dput;
3385 }
1643b43f
AV
3386 if (unlikely(create_error) && !dentry->d_inode) {
3387 error = create_error;
3388 goto out_dput;
d58ffd35 3389 }
da5ebf5a 3390 return dentry;
d58ffd35
MS
3391
3392out_dput:
3393 dput(dentry);
da5ebf5a 3394 return ERR_PTR(error);
d58ffd35
MS
3395}
3396
c981a482 3397static const char *open_last_lookups(struct nameidata *nd,
3ec2eef1 3398 struct file *file, const struct open_flags *op)
fb1cc555 3399{
a1e28038 3400 struct dentry *dir = nd->path.dentry;
ca344a89 3401 int open_flag = op->open_flag;
64894cf8 3402 bool got_write = false;
254cf582 3403 unsigned seq;
a1eb3315 3404 struct inode *inode;
da5ebf5a 3405 struct dentry *dentry;
b0417d2c 3406 const char *res;
1f36f774 3407
c3e380b0
AV
3408 nd->flags |= op->intent;
3409
bc77daa7 3410 if (nd->last_type != LAST_NORM) {
56676ec3
AV
3411 if (nd->depth)
3412 put_link(nd);
ff326a32 3413 return handle_dots(nd, nd->last_type);
1f36f774 3414 }
67ee3ad2 3415
ca344a89 3416 if (!(open_flag & O_CREAT)) {
fe2d35ff
AV
3417 if (nd->last.name[nd->last.len])
3418 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3419 /* we _can_ be in RCU mode here */
20e34357
AV
3420 dentry = lookup_fast(nd, &inode, &seq);
3421 if (IS_ERR(dentry))
1ccac622 3422 return ERR_CAST(dentry);
20e34357 3423 if (likely(dentry))
71574865
MS
3424 goto finish_lookup;
3425
6583fe22 3426 BUG_ON(nd->flags & LOOKUP_RCU);
b6183df7
MS
3427 } else {
3428 /* create side of things */
72287417 3429 if (nd->flags & LOOKUP_RCU) {
e36cffed
JA
3430 if (!try_to_unlazy(nd))
3431 return ERR_PTR(-ECHILD);
72287417 3432 }
c9b07eab 3433 audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
b6183df7 3434 /* trailing slashes? */
deb106c6 3435 if (unlikely(nd->last.name[nd->last.len]))
1ccac622 3436 return ERR_PTR(-EISDIR);
b6183df7 3437 }
a2c36b45 3438
9cf843e3 3439 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
e36cffed 3440 got_write = !mnt_want_write(nd->path.mnt);
64894cf8
AV
3441 /*
3442 * do _not_ fail yet - we might not need that or fail with
3443 * a different error; let lookup_open() decide; we'll be
3444 * dropping this one anyway.
3445 */
3446 }
9cf843e3
AV
3447 if (open_flag & O_CREAT)
3448 inode_lock(dir->d_inode);
3449 else
3450 inode_lock_shared(dir->d_inode);
da5ebf5a 3451 dentry = lookup_open(nd, file, op, got_write);
f7bb959d
AV
3452 if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED))
3453 fsnotify_create(dir->d_inode, dentry);
9cf843e3
AV
3454 if (open_flag & O_CREAT)
3455 inode_unlock(dir->d_inode);
3456 else
3457 inode_unlock_shared(dir->d_inode);
a1e28038 3458
c981a482 3459 if (got_write)
59e96e65 3460 mnt_drop_write(nd->path.mnt);
d18e9008 3461
59e96e65
AV
3462 if (IS_ERR(dentry))
3463 return ERR_CAST(dentry);
3464
973d4b73 3465 if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) {
e73cabff
AV
3466 dput(nd->path.dentry);
3467 nd->path.dentry = dentry;
c981a482 3468 return NULL;
fb1cc555
AV
3469 }
3470
20e34357 3471finish_lookup:
56676ec3
AV
3472 if (nd->depth)
3473 put_link(nd);
8c4efe22 3474 res = step_into(nd, WALK_TRAILING, dentry, inode, seq);
ff326a32 3475 if (unlikely(res))
b0417d2c 3476 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
ff326a32 3477 return res;
c981a482
AV
3478}
3479
3480/*
3481 * Handle the last step of open()
3482 */
c5971b8c 3483static int do_open(struct nameidata *nd,
c981a482
AV
3484 struct file *file, const struct open_flags *op)
3485{
549c7297 3486 struct user_namespace *mnt_userns;
c981a482
AV
3487 int open_flag = op->open_flag;
3488 bool do_truncate;
3489 int acc_mode;
c981a482
AV
3490 int error;
3491
ff326a32
AV
3492 if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) {
3493 error = complete_walk(nd);
3494 if (error)
3495 return error;
3496 }
973d4b73
AV
3497 if (!(file->f_mode & FMODE_CREATED))
3498 audit_inode(nd->name, nd->path.dentry, 0);
549c7297 3499 mnt_userns = mnt_user_ns(nd->path.mnt);
30aba665 3500 if (open_flag & O_CREAT) {
b94e0b32
AV
3501 if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED))
3502 return -EEXIST;
30aba665 3503 if (d_is_dir(nd->path.dentry))
c5971b8c 3504 return -EISDIR;
549c7297 3505 error = may_create_in_sticky(mnt_userns, nd,
30aba665
SM
3506 d_backing_inode(nd->path.dentry));
3507 if (unlikely(error))
c5971b8c 3508 return error;
30aba665 3509 }
44b1d530 3510 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
c5971b8c 3511 return -ENOTDIR;
6c0d46c4 3512
8795e7d4
AV
3513 do_truncate = false;
3514 acc_mode = op->acc_mode;
5a2d3edd
AV
3515 if (file->f_mode & FMODE_CREATED) {
3516 /* Don't check for write permission, don't truncate */
3517 open_flag &= ~O_TRUNC;
5a2d3edd 3518 acc_mode = 0;
8795e7d4 3519 } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) {
0f9d1a10
AV
3520 error = mnt_want_write(nd->path.mnt);
3521 if (error)
c5971b8c 3522 return error;
8795e7d4 3523 do_truncate = true;
0f9d1a10 3524 }
549c7297 3525 error = may_open(mnt_userns, &nd->path, acc_mode, open_flag);
8795e7d4 3526 if (!error && !(file->f_mode & FMODE_OPENED))
3ad5615a 3527 error = vfs_open(&nd->path, file);
8795e7d4
AV
3528 if (!error)
3529 error = ima_file_check(file, op->acc_mode);
3530 if (!error && do_truncate)
549c7297 3531 error = handle_truncate(mnt_userns, file);
c80567c8
AV
3532 if (unlikely(error > 0)) {
3533 WARN_ON(1);
3534 error = -EINVAL;
3535 }
8795e7d4 3536 if (do_truncate)
0f9d1a10 3537 mnt_drop_write(nd->path.mnt);
c5971b8c 3538 return error;
fb1cc555
AV
3539}
3540
6521f891
CB
3541/**
3542 * vfs_tmpfile - create tmpfile
3543 * @mnt_userns: user namespace of the mount the inode was found from
3544 * @dentry: pointer to dentry of the base directory
3545 * @mode: mode of the new tmpfile
2111c3c0 3546 * @open_flag: flags
6521f891
CB
3547 *
3548 * Create a temporary file.
3549 *
3550 * If the inode has been found through an idmapped mount the user namespace of
3551 * the vfsmount must be passed through @mnt_userns. This function will then take
3552 * care to map the inode according to @mnt_userns before checking permissions.
3553 * On non-idmapped mounts or if permission checking is to be performed on the
3554 * raw inode simply passs init_user_ns.
3555 */
3556struct dentry *vfs_tmpfile(struct user_namespace *mnt_userns,
3557 struct dentry *dentry, umode_t mode, int open_flag)
af7bd4dc 3558{
af7bd4dc
AG
3559 struct dentry *child = NULL;
3560 struct inode *dir = dentry->d_inode;
3561 struct inode *inode;
3562 int error;
3563
3564 /* we want directory to be writable */
6521f891 3565 error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
af7bd4dc
AG
3566 if (error)
3567 goto out_err;
3568 error = -EOPNOTSUPP;
3569 if (!dir->i_op->tmpfile)
3570 goto out_err;
3571 error = -ENOMEM;
cdf01226 3572 child = d_alloc(dentry, &slash_name);
af7bd4dc
AG
3573 if (unlikely(!child))
3574 goto out_err;
549c7297 3575 error = dir->i_op->tmpfile(mnt_userns, dir, child, mode);
af7bd4dc
AG
3576 if (error)
3577 goto out_err;
3578 error = -ENOENT;
3579 inode = child->d_inode;
3580 if (unlikely(!inode))
3581 goto out_err;
3582 if (!(open_flag & O_EXCL)) {
3583 spin_lock(&inode->i_lock);
3584 inode->i_state |= I_LINKABLE;
3585 spin_unlock(&inode->i_lock);
3586 }
a2d2329e 3587 ima_post_create_tmpfile(mnt_userns, inode);
af7bd4dc
AG
3588 return child;
3589
3590out_err:
3591 dput(child);
3592 return ERR_PTR(error);
3593}
3594EXPORT_SYMBOL(vfs_tmpfile);
3595
c8a53ee5 3596static int do_tmpfile(struct nameidata *nd, unsigned flags,
60545d0d 3597 const struct open_flags *op,
3ec2eef1 3598 struct file *file)
60545d0d 3599{
6521f891 3600 struct user_namespace *mnt_userns;
625b6d10 3601 struct dentry *child;
625b6d10 3602 struct path path;
c8a53ee5 3603 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
60545d0d
AV
3604 if (unlikely(error))
3605 return error;
625b6d10 3606 error = mnt_want_write(path.mnt);
60545d0d
AV
3607 if (unlikely(error))
3608 goto out;
6521f891
CB
3609 mnt_userns = mnt_user_ns(path.mnt);
3610 child = vfs_tmpfile(mnt_userns, path.dentry, op->mode, op->open_flag);
af7bd4dc 3611 error = PTR_ERR(child);
684e73be 3612 if (IS_ERR(child))
60545d0d 3613 goto out2;
625b6d10
AV
3614 dput(path.dentry);
3615 path.dentry = child;
c8a53ee5 3616 audit_inode(nd->name, child, 0);
69a91c23 3617 /* Don't check for other permissions, the inode was just created */
549c7297 3618 error = may_open(mnt_userns, &path, 0, op->open_flag);
1e8f44f1
AV
3619 if (!error)
3620 error = vfs_open(&path, file);
60545d0d 3621out2:
625b6d10 3622 mnt_drop_write(path.mnt);
60545d0d 3623out:
625b6d10 3624 path_put(&path);
60545d0d
AV
3625 return error;
3626}
3627
6ac08709
AV
3628static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3629{
3630 struct path path;
3631 int error = path_lookupat(nd, flags, &path);
3632 if (!error) {
3633 audit_inode(nd->name, path.dentry, 0);
ae2bb293 3634 error = vfs_open(&path, file);
6ac08709
AV
3635 path_put(&path);
3636 }
3637 return error;
3638}
3639
c8a53ee5
AV
3640static struct file *path_openat(struct nameidata *nd,
3641 const struct open_flags *op, unsigned flags)
1da177e4 3642{
30d90494 3643 struct file *file;
13aab428 3644 int error;
31e6b01f 3645
ea73ea72 3646 file = alloc_empty_file(op->open_flag, current_cred());
1afc99be
AV
3647 if (IS_ERR(file))
3648 return file;
31e6b01f 3649
bb458c64 3650 if (unlikely(file->f_flags & __O_TMPFILE)) {
3ec2eef1 3651 error = do_tmpfile(nd, flags, op, file);
5f336e72 3652 } else if (unlikely(file->f_flags & O_PATH)) {
6ac08709 3653 error = do_o_path(nd, flags, file);
5f336e72
AV
3654 } else {
3655 const char *s = path_init(nd, flags);
3656 while (!(error = link_path_walk(s, nd)) &&
c5971b8c 3657 (s = open_last_lookups(nd, file, op)) != NULL)
1ccac622 3658 ;
c5971b8c
AV
3659 if (!error)
3660 error = do_open(nd, file, op);
5f336e72 3661 terminate_walk(nd);
806b681c 3662 }
7c1c01ec 3663 if (likely(!error)) {
aad888f8 3664 if (likely(file->f_mode & FMODE_OPENED))
7c1c01ec
AV
3665 return file;
3666 WARN_ON(1);
3667 error = -EINVAL;
16b1c1cd 3668 }
7c1c01ec
AV
3669 fput(file);
3670 if (error == -EOPENSTALE) {
3671 if (flags & LOOKUP_RCU)
3672 error = -ECHILD;
3673 else
3674 error = -ESTALE;
2675a4eb 3675 }
7c1c01ec 3676 return ERR_PTR(error);
1da177e4
LT
3677}
3678
669abf4e 3679struct file *do_filp_open(int dfd, struct filename *pathname,
f9652e10 3680 const struct open_flags *op)
13aab428 3681{
9883d185 3682 struct nameidata nd;
f9652e10 3683 int flags = op->lookup_flags;
13aab428
AV
3684 struct file *filp;
3685
06422964 3686 set_nameidata(&nd, dfd, pathname, NULL);
c8a53ee5 3687 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
13aab428 3688 if (unlikely(filp == ERR_PTR(-ECHILD)))
c8a53ee5 3689 filp = path_openat(&nd, op, flags);
13aab428 3690 if (unlikely(filp == ERR_PTR(-ESTALE)))
c8a53ee5 3691 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
9883d185 3692 restore_nameidata();
13aab428
AV
3693 return filp;
3694}
3695
ffb37ca3 3696struct file *do_file_open_root(const struct path *root,
f9652e10 3697 const char *name, const struct open_flags *op)
73d049a4 3698{
9883d185 3699 struct nameidata nd;
73d049a4 3700 struct file *file;
51689104 3701 struct filename *filename;
bcba1e7d 3702 int flags = op->lookup_flags;
73d049a4 3703
ffb37ca3 3704 if (d_is_symlink(root->dentry) && op->intent & LOOKUP_OPEN)
73d049a4
AV
3705 return ERR_PTR(-ELOOP);
3706
51689104 3707 filename = getname_kernel(name);
a1c83681 3708 if (IS_ERR(filename))
51689104
PM
3709 return ERR_CAST(filename);
3710
06422964 3711 set_nameidata(&nd, -1, filename, root);
c8a53ee5 3712 file = path_openat(&nd, op, flags | LOOKUP_RCU);
73d049a4 3713 if (unlikely(file == ERR_PTR(-ECHILD)))
c8a53ee5 3714 file = path_openat(&nd, op, flags);
73d049a4 3715 if (unlikely(file == ERR_PTR(-ESTALE)))
c8a53ee5 3716 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
9883d185 3717 restore_nameidata();
51689104 3718 putname(filename);
73d049a4
AV
3719 return file;
3720}
3721
b4a4f213
SB
3722static struct dentry *filename_create(int dfd, struct filename *name,
3723 struct path *path, unsigned int lookup_flags)
1da177e4 3724{
c663e5d8 3725 struct dentry *dentry = ERR_PTR(-EEXIST);
391172c4 3726 struct qstr last;
b3d4650d
N
3727 bool want_dir = lookup_flags & LOOKUP_DIRECTORY;
3728 unsigned int reval_flag = lookup_flags & LOOKUP_REVAL;
3729 unsigned int create_flags = LOOKUP_CREATE | LOOKUP_EXCL;
391172c4 3730 int type;
c30dabfe 3731 int err2;
1ac12b4b 3732 int error;
1ac12b4b 3733
b3d4650d 3734 error = filename_parentat(dfd, name, reval_flag, path, &last, &type);
0ee50b47
DK
3735 if (error)
3736 return ERR_PTR(error);
1da177e4 3737
c663e5d8
CH
3738 /*
3739 * Yucky last component or no last component at all?
3740 * (foo/., foo/.., /////)
3741 */
5c31b6ce 3742 if (unlikely(type != LAST_NORM))
ed75e95d 3743 goto out;
c663e5d8 3744
c30dabfe 3745 /* don't fail immediately if it's r/o, at least try to report other errors */
391172c4 3746 err2 = mnt_want_write(path->mnt);
c663e5d8 3747 /*
b3d4650d
N
3748 * Do the final lookup. Suppress 'create' if there is a trailing
3749 * '/', and a directory wasn't requested.
c663e5d8 3750 */
b3d4650d
N
3751 if (last.name[last.len] && !want_dir)
3752 create_flags = 0;
5955102c 3753 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
b3d4650d 3754 dentry = __lookup_hash(&last, path->dentry, reval_flag | create_flags);
1da177e4 3755 if (IS_ERR(dentry))
a8104a9f 3756 goto unlock;
c663e5d8 3757
a8104a9f 3758 error = -EEXIST;
b18825a7 3759 if (d_is_positive(dentry))
a8104a9f 3760 goto fail;
b18825a7 3761
c663e5d8
CH
3762 /*
3763 * Special case - lookup gave negative, but... we had foo/bar/
3764 * From the vfs_mknod() POV we just have a negative dentry -
3765 * all is fine. Let's be bastards - you had / on the end, you've
3766 * been asking for (non-existent) directory. -ENOENT for you.
3767 */
b3d4650d 3768 if (unlikely(!create_flags)) {
a8104a9f 3769 error = -ENOENT;
ed75e95d 3770 goto fail;
e9baf6e5 3771 }
c30dabfe
JK
3772 if (unlikely(err2)) {
3773 error = err2;
a8104a9f 3774 goto fail;
c30dabfe 3775 }
1da177e4 3776 return dentry;
1da177e4 3777fail:
a8104a9f
AV
3778 dput(dentry);
3779 dentry = ERR_PTR(error);
3780unlock:
5955102c 3781 inode_unlock(path->dentry->d_inode);
c30dabfe 3782 if (!err2)
391172c4 3783 mnt_drop_write(path->mnt);
ed75e95d 3784out:
391172c4 3785 path_put(path);
1da177e4
LT
3786 return dentry;
3787}
fa14a0b8 3788
b4a4f213 3789struct dentry *kern_path_create(int dfd, const char *pathname,
584d3226
DK
3790 struct path *path, unsigned int lookup_flags)
3791{
b4a4f213
SB
3792 struct filename *filename = getname_kernel(pathname);
3793 struct dentry *res = filename_create(dfd, filename, path, lookup_flags);
584d3226 3794
b4a4f213 3795 putname(filename);
584d3226
DK
3796 return res;
3797}
dae6ad8f
AV
3798EXPORT_SYMBOL(kern_path_create);
3799
921a1650
AV
3800void done_path_create(struct path *path, struct dentry *dentry)
3801{
3802 dput(dentry);
5955102c 3803 inode_unlock(path->dentry->d_inode);
a8104a9f 3804 mnt_drop_write(path->mnt);
921a1650
AV
3805 path_put(path);
3806}
3807EXPORT_SYMBOL(done_path_create);
3808
520ae687 3809inline struct dentry *user_path_create(int dfd, const char __user *pathname,
1ac12b4b 3810 struct path *path, unsigned int lookup_flags)
dae6ad8f 3811{
b4a4f213
SB
3812 struct filename *filename = getname(pathname);
3813 struct dentry *res = filename_create(dfd, filename, path, lookup_flags);
3814
3815 putname(filename);
3816 return res;
dae6ad8f
AV
3817}
3818EXPORT_SYMBOL(user_path_create);
3819
6521f891
CB
3820/**
3821 * vfs_mknod - create device node or file
3822 * @mnt_userns: user namespace of the mount the inode was found from
3823 * @dir: inode of @dentry
3824 * @dentry: pointer to dentry of the base directory
3825 * @mode: mode of the new device node or file
3826 * @dev: device number of device to create
3827 *
3828 * Create a device node or file.
3829 *
3830 * If the inode has been found through an idmapped mount the user namespace of
3831 * the vfsmount must be passed through @mnt_userns. This function will then take
3832 * care to map the inode according to @mnt_userns before checking permissions.
3833 * On non-idmapped mounts or if permission checking is to be performed on the
3834 * raw inode simply passs init_user_ns.
3835 */
3836int vfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
3837 struct dentry *dentry, umode_t mode, dev_t dev)
1da177e4 3838{
a3c751a5 3839 bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV;
6521f891 3840 int error = may_create(mnt_userns, dir, dentry);
1da177e4
LT
3841
3842 if (error)
3843 return error;
3844
a3c751a5
MS
3845 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout &&
3846 !capable(CAP_MKNOD))
1da177e4
LT
3847 return -EPERM;
3848
acfa4380 3849 if (!dir->i_op->mknod)
1da177e4
LT
3850 return -EPERM;
3851
08ce5f16
SH
3852 error = devcgroup_inode_mknod(mode, dev);
3853 if (error)
3854 return error;
3855
1da177e4
LT
3856 error = security_inode_mknod(dir, dentry, mode, dev);
3857 if (error)
3858 return error;
3859
549c7297 3860 error = dir->i_op->mknod(mnt_userns, dir, dentry, mode, dev);
a74574aa 3861 if (!error)
f38aa942 3862 fsnotify_create(dir, dentry);
1da177e4
LT
3863 return error;
3864}
4d359507 3865EXPORT_SYMBOL(vfs_mknod);
1da177e4 3866
f69aac00 3867static int may_mknod(umode_t mode)
463c3197
DH
3868{
3869 switch (mode & S_IFMT) {
3870 case S_IFREG:
3871 case S_IFCHR:
3872 case S_IFBLK:
3873 case S_IFIFO:
3874 case S_IFSOCK:
3875 case 0: /* zero mode translates to S_IFREG */
3876 return 0;
3877 case S_IFDIR:
3878 return -EPERM;
3879 default:
3880 return -EINVAL;
3881 }
3882}
3883
45f30dab 3884static int do_mknodat(int dfd, struct filename *name, umode_t mode,
87c4e192 3885 unsigned int dev)
1da177e4 3886{
6521f891 3887 struct user_namespace *mnt_userns;
2ad94ae6 3888 struct dentry *dentry;
dae6ad8f
AV
3889 struct path path;
3890 int error;
972567f1 3891 unsigned int lookup_flags = 0;
1da177e4 3892
8e4bfca1
AV
3893 error = may_mknod(mode);
3894 if (error)
7797251b 3895 goto out1;
972567f1 3896retry:
b4a4f213 3897 dentry = filename_create(dfd, name, &path, lookup_flags);
7797251b 3898 error = PTR_ERR(dentry);
dae6ad8f 3899 if (IS_ERR(dentry))
7797251b 3900 goto out1;
2ad94ae6 3901
dae6ad8f 3902 if (!IS_POSIXACL(path.dentry->d_inode))
ce3b0f8d 3903 mode &= ~current_umask();
dae6ad8f 3904 error = security_path_mknod(&path, dentry, mode, dev);
be6d3e56 3905 if (error)
7797251b 3906 goto out2;
6521f891
CB
3907
3908 mnt_userns = mnt_user_ns(path.mnt);
463c3197 3909 switch (mode & S_IFMT) {
1da177e4 3910 case 0: case S_IFREG:
6521f891
CB
3911 error = vfs_create(mnt_userns, path.dentry->d_inode,
3912 dentry, mode, true);
05d1a717 3913 if (!error)
a2d2329e 3914 ima_post_path_mknod(mnt_userns, dentry);
1da177e4
LT
3915 break;
3916 case S_IFCHR: case S_IFBLK:
6521f891
CB
3917 error = vfs_mknod(mnt_userns, path.dentry->d_inode,
3918 dentry, mode, new_decode_dev(dev));
1da177e4
LT
3919 break;
3920 case S_IFIFO: case S_IFSOCK:
6521f891
CB
3921 error = vfs_mknod(mnt_userns, path.dentry->d_inode,
3922 dentry, mode, 0);
1da177e4 3923 break;
1da177e4 3924 }
7797251b 3925out2:
921a1650 3926 done_path_create(&path, dentry);
972567f1
JL
3927 if (retry_estale(error, lookup_flags)) {
3928 lookup_flags |= LOOKUP_REVAL;
3929 goto retry;
3930 }
7797251b
DK
3931out1:
3932 putname(name);
1da177e4
LT
3933 return error;
3934}
3935
87c4e192
DB
3936SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3937 unsigned int, dev)
3938{
7797251b 3939 return do_mknodat(dfd, getname(filename), mode, dev);
87c4e192
DB
3940}
3941
8208a22b 3942SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
5590ff0d 3943{
7797251b 3944 return do_mknodat(AT_FDCWD, getname(filename), mode, dev);
5590ff0d
UD
3945}
3946
6521f891
CB
3947/**
3948 * vfs_mkdir - create directory
3949 * @mnt_userns: user namespace of the mount the inode was found from
3950 * @dir: inode of @dentry
3951 * @dentry: pointer to dentry of the base directory
3952 * @mode: mode of the new directory
3953 *
3954 * Create a directory.
3955 *
3956 * If the inode has been found through an idmapped mount the user namespace of
3957 * the vfsmount must be passed through @mnt_userns. This function will then take
3958 * care to map the inode according to @mnt_userns before checking permissions.
3959 * On non-idmapped mounts or if permission checking is to be performed on the
3960 * raw inode simply passs init_user_ns.
3961 */
3962int vfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
3963 struct dentry *dentry, umode_t mode)
1da177e4 3964{
6521f891 3965 int error = may_create(mnt_userns, dir, dentry);
8de52778 3966 unsigned max_links = dir->i_sb->s_max_links;
1da177e4
LT
3967
3968 if (error)
3969 return error;
3970
acfa4380 3971 if (!dir->i_op->mkdir)
1da177e4
LT
3972 return -EPERM;
3973
3974 mode &= (S_IRWXUGO|S_ISVTX);
3975 error = security_inode_mkdir(dir, dentry, mode);
3976 if (error)
3977 return error;
3978
8de52778
AV
3979 if (max_links && dir->i_nlink >= max_links)
3980 return -EMLINK;
3981
549c7297 3982 error = dir->i_op->mkdir(mnt_userns, dir, dentry, mode);
a74574aa 3983 if (!error)
f38aa942 3984 fsnotify_mkdir(dir, dentry);
1da177e4
LT
3985 return error;
3986}
4d359507 3987EXPORT_SYMBOL(vfs_mkdir);
1da177e4 3988
45f30dab 3989int do_mkdirat(int dfd, struct filename *name, umode_t mode)
1da177e4 3990{
6902d925 3991 struct dentry *dentry;
dae6ad8f
AV
3992 struct path path;
3993 int error;
b76d8b82 3994 unsigned int lookup_flags = LOOKUP_DIRECTORY;
1da177e4 3995
b76d8b82 3996retry:
b4a4f213 3997 dentry = filename_create(dfd, name, &path, lookup_flags);
584d3226 3998 error = PTR_ERR(dentry);
6902d925 3999 if (IS_ERR(dentry))
584d3226 4000 goto out_putname;
1da177e4 4001
dae6ad8f 4002 if (!IS_POSIXACL(path.dentry->d_inode))
ce3b0f8d 4003 mode &= ~current_umask();
dae6ad8f 4004 error = security_path_mkdir(&path, dentry, mode);
6521f891
CB
4005 if (!error) {
4006 struct user_namespace *mnt_userns;
4007 mnt_userns = mnt_user_ns(path.mnt);
549c7297
CB
4008 error = vfs_mkdir(mnt_userns, path.dentry->d_inode, dentry,
4009 mode);
6521f891 4010 }
921a1650 4011 done_path_create(&path, dentry);
b76d8b82
JL
4012 if (retry_estale(error, lookup_flags)) {
4013 lookup_flags |= LOOKUP_REVAL;
4014 goto retry;
4015 }
584d3226
DK
4016out_putname:
4017 putname(name);
1da177e4
LT
4018 return error;
4019}
4020
0101db7a
DB
4021SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
4022{
584d3226 4023 return do_mkdirat(dfd, getname(pathname), mode);
0101db7a
DB
4024}
4025
a218d0fd 4026SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
5590ff0d 4027{
584d3226 4028 return do_mkdirat(AT_FDCWD, getname(pathname), mode);
5590ff0d
UD
4029}
4030
6521f891
CB
4031/**
4032 * vfs_rmdir - remove directory
4033 * @mnt_userns: user namespace of the mount the inode was found from
4034 * @dir: inode of @dentry
4035 * @dentry: pointer to dentry of the base directory
4036 *
4037 * Remove a directory.
4038 *
4039 * If the inode has been found through an idmapped mount the user namespace of
4040 * the vfsmount must be passed through @mnt_userns. This function will then take
4041 * care to map the inode according to @mnt_userns before checking permissions.
4042 * On non-idmapped mounts or if permission checking is to be performed on the
4043 * raw inode simply passs init_user_ns.
4044 */
4045int vfs_rmdir(struct user_namespace *mnt_userns, struct inode *dir,
4046 struct dentry *dentry)
1da177e4 4047{
6521f891 4048 int error = may_delete(mnt_userns, dir, dentry, 1);
1da177e4
LT
4049
4050 if (error)
4051 return error;
4052
acfa4380 4053 if (!dir->i_op->rmdir)
1da177e4
LT
4054 return -EPERM;
4055
1d2ef590 4056 dget(dentry);
5955102c 4057 inode_lock(dentry->d_inode);
912dbc15
SW
4058
4059 error = -EBUSY;
1bd9c4e4
DH
4060 if (is_local_mountpoint(dentry) ||
4061 (dentry->d_inode->i_flags & S_KERNEL_FILE))
912dbc15
SW
4062 goto out;
4063
4064 error = security_inode_rmdir(dir, dentry);
4065 if (error)
4066 goto out;
4067
4068 error = dir->i_op->rmdir(dir, dentry);
4069 if (error)
4070 goto out;
4071
8767712f 4072 shrink_dcache_parent(dentry);
912dbc15
SW
4073 dentry->d_inode->i_flags |= S_DEAD;
4074 dont_mount(dentry);
8ed936b5 4075 detach_mounts(dentry);
912dbc15
SW
4076
4077out:
5955102c 4078 inode_unlock(dentry->d_inode);
1d2ef590 4079 dput(dentry);
912dbc15 4080 if (!error)
a37d9a17 4081 d_delete_notify(dir, dentry);
1da177e4
LT
4082 return error;
4083}
4d359507 4084EXPORT_SYMBOL(vfs_rmdir);
1da177e4 4085
45f30dab 4086int do_rmdir(int dfd, struct filename *name)
1da177e4 4087{
6521f891 4088 struct user_namespace *mnt_userns;
0ee50b47 4089 int error;
1da177e4 4090 struct dentry *dentry;
f5beed75
AV
4091 struct path path;
4092 struct qstr last;
4093 int type;
c6ee9206
JL
4094 unsigned int lookup_flags = 0;
4095retry:
c5f563f9 4096 error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
0ee50b47
DK
4097 if (error)
4098 goto exit1;
1da177e4 4099
f5beed75 4100 switch (type) {
0612d9fb
OH
4101 case LAST_DOTDOT:
4102 error = -ENOTEMPTY;
0ee50b47 4103 goto exit2;
0612d9fb
OH
4104 case LAST_DOT:
4105 error = -EINVAL;
0ee50b47 4106 goto exit2;
0612d9fb
OH
4107 case LAST_ROOT:
4108 error = -EBUSY;
0ee50b47 4109 goto exit2;
1da177e4 4110 }
0612d9fb 4111
f5beed75 4112 error = mnt_want_write(path.mnt);
c30dabfe 4113 if (error)
0ee50b47 4114 goto exit2;
0612d9fb 4115
5955102c 4116 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
f5beed75 4117 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
1da177e4 4118 error = PTR_ERR(dentry);
6902d925 4119 if (IS_ERR(dentry))
0ee50b47 4120 goto exit3;
e6bc45d6
TT
4121 if (!dentry->d_inode) {
4122 error = -ENOENT;
0ee50b47 4123 goto exit4;
e6bc45d6 4124 }
f5beed75 4125 error = security_path_rmdir(&path, dentry);
be6d3e56 4126 if (error)
0ee50b47 4127 goto exit4;
6521f891
CB
4128 mnt_userns = mnt_user_ns(path.mnt);
4129 error = vfs_rmdir(mnt_userns, path.dentry->d_inode, dentry);
0ee50b47 4130exit4:
6902d925 4131 dput(dentry);
0ee50b47 4132exit3:
5955102c 4133 inode_unlock(path.dentry->d_inode);
f5beed75 4134 mnt_drop_write(path.mnt);
0ee50b47 4135exit2:
f5beed75 4136 path_put(&path);
c6ee9206
JL
4137 if (retry_estale(error, lookup_flags)) {
4138 lookup_flags |= LOOKUP_REVAL;
4139 goto retry;
4140 }
0ee50b47 4141exit1:
24fb33d4 4142 putname(name);
1da177e4
LT
4143 return error;
4144}
4145
3cdad428 4146SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
5590ff0d 4147{
e24ab0ef 4148 return do_rmdir(AT_FDCWD, getname(pathname));
5590ff0d
UD
4149}
4150
b21996e3
BF
4151/**
4152 * vfs_unlink - unlink a filesystem object
6521f891 4153 * @mnt_userns: user namespace of the mount the inode was found from
b21996e3
BF
4154 * @dir: parent directory
4155 * @dentry: victim
4156 * @delegated_inode: returns victim inode, if the inode is delegated.
4157 *
4158 * The caller must hold dir->i_mutex.
4159 *
4160 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
4161 * return a reference to the inode in delegated_inode. The caller
4162 * should then break the delegation on that inode and retry. Because
4163 * breaking a delegation may take a long time, the caller should drop
4164 * dir->i_mutex before doing so.
4165 *
4166 * Alternatively, a caller may pass NULL for delegated_inode. This may
4167 * be appropriate for callers that expect the underlying filesystem not
4168 * to be NFS exported.
6521f891
CB
4169 *
4170 * If the inode has been found through an idmapped mount the user namespace of
4171 * the vfsmount must be passed through @mnt_userns. This function will then take
4172 * care to map the inode according to @mnt_userns before checking permissions.
4173 * On non-idmapped mounts or if permission checking is to be performed on the
4174 * raw inode simply passs init_user_ns.
b21996e3 4175 */
6521f891
CB
4176int vfs_unlink(struct user_namespace *mnt_userns, struct inode *dir,
4177 struct dentry *dentry, struct inode **delegated_inode)
1da177e4 4178{
9accbb97 4179 struct inode *target = dentry->d_inode;
6521f891 4180 int error = may_delete(mnt_userns, dir, dentry, 0);
1da177e4
LT
4181
4182 if (error)
4183 return error;
4184
acfa4380 4185 if (!dir->i_op->unlink)
1da177e4
LT
4186 return -EPERM;
4187
5955102c 4188 inode_lock(target);
51cc3a66
HD
4189 if (IS_SWAPFILE(target))
4190 error = -EPERM;
4191 else if (is_local_mountpoint(dentry))
1da177e4
LT
4192 error = -EBUSY;
4193 else {
4194 error = security_inode_unlink(dir, dentry);
bec1052e 4195 if (!error) {
5a14696c
BF
4196 error = try_break_deleg(target, delegated_inode);
4197 if (error)
b21996e3 4198 goto out;
1da177e4 4199 error = dir->i_op->unlink(dir, dentry);
8ed936b5 4200 if (!error) {
d83c49f3 4201 dont_mount(dentry);
8ed936b5
EB
4202 detach_mounts(dentry);
4203 }
bec1052e 4204 }
1da177e4 4205 }
b21996e3 4206out:
5955102c 4207 inode_unlock(target);
1da177e4
LT
4208
4209 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
a37d9a17
AG
4210 if (!error && dentry->d_flags & DCACHE_NFSFS_RENAMED) {
4211 fsnotify_unlink(dir, dentry);
4212 } else if (!error) {
9accbb97 4213 fsnotify_link_count(target);
a37d9a17 4214 d_delete_notify(dir, dentry);
1da177e4 4215 }
0eeca283 4216
1da177e4
LT
4217 return error;
4218}
4d359507 4219EXPORT_SYMBOL(vfs_unlink);
1da177e4
LT
4220
4221/*
4222 * Make sure that the actual truncation of the file will occur outside its
1b1dcc1b 4223 * directory's i_mutex. Truncate can take a long time if there is a lot of
1da177e4
LT
4224 * writeout happening, and we don't want to prevent access to the directory
4225 * while waiting on the I/O.
4226 */
45f30dab 4227int do_unlinkat(int dfd, struct filename *name)
1da177e4 4228{
2ad94ae6 4229 int error;
1da177e4 4230 struct dentry *dentry;
f5beed75
AV
4231 struct path path;
4232 struct qstr last;
4233 int type;
1da177e4 4234 struct inode *inode = NULL;
b21996e3 4235 struct inode *delegated_inode = NULL;
5d18f813
JL
4236 unsigned int lookup_flags = 0;
4237retry:
c5f563f9 4238 error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
0ee50b47
DK
4239 if (error)
4240 goto exit1;
2ad94ae6 4241
1da177e4 4242 error = -EISDIR;
f5beed75 4243 if (type != LAST_NORM)
0ee50b47 4244 goto exit2;
0612d9fb 4245
f5beed75 4246 error = mnt_want_write(path.mnt);
c30dabfe 4247 if (error)
0ee50b47 4248 goto exit2;
b21996e3 4249retry_deleg:
5955102c 4250 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
f5beed75 4251 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
1da177e4
LT
4252 error = PTR_ERR(dentry);
4253 if (!IS_ERR(dentry)) {
6521f891
CB
4254 struct user_namespace *mnt_userns;
4255
1da177e4 4256 /* Why not before? Because we want correct error value */
f5beed75 4257 if (last.name[last.len])
50338b88 4258 goto slashes;
1da177e4 4259 inode = dentry->d_inode;
b18825a7 4260 if (d_is_negative(dentry))
e6bc45d6
TT
4261 goto slashes;
4262 ihold(inode);
f5beed75 4263 error = security_path_unlink(&path, dentry);
be6d3e56 4264 if (error)
0ee50b47 4265 goto exit3;
6521f891 4266 mnt_userns = mnt_user_ns(path.mnt);
549c7297
CB
4267 error = vfs_unlink(mnt_userns, path.dentry->d_inode, dentry,
4268 &delegated_inode);
0ee50b47 4269exit3:
1da177e4
LT
4270 dput(dentry);
4271 }
5955102c 4272 inode_unlock(path.dentry->d_inode);
1da177e4
LT
4273 if (inode)
4274 iput(inode); /* truncate the inode here */
b21996e3
BF
4275 inode = NULL;
4276 if (delegated_inode) {
5a14696c 4277 error = break_deleg_wait(&delegated_inode);
b21996e3
BF
4278 if (!error)
4279 goto retry_deleg;
4280 }
f5beed75 4281 mnt_drop_write(path.mnt);
0ee50b47 4282exit2:
f5beed75 4283 path_put(&path);
5d18f813
JL
4284 if (retry_estale(error, lookup_flags)) {
4285 lookup_flags |= LOOKUP_REVAL;
4286 inode = NULL;
4287 goto retry;
4288 }
0ee50b47 4289exit1:
da2f1362 4290 putname(name);
1da177e4
LT
4291 return error;
4292
4293slashes:
b18825a7
DH
4294 if (d_is_negative(dentry))
4295 error = -ENOENT;
44b1d530 4296 else if (d_is_dir(dentry))
b18825a7
DH
4297 error = -EISDIR;
4298 else
4299 error = -ENOTDIR;
0ee50b47 4300 goto exit3;
1da177e4
LT
4301}
4302
2e4d0924 4303SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
5590ff0d
UD
4304{
4305 if ((flag & ~AT_REMOVEDIR) != 0)
4306 return -EINVAL;
4307
4308 if (flag & AT_REMOVEDIR)
e24ab0ef 4309 return do_rmdir(dfd, getname(pathname));
da2f1362 4310 return do_unlinkat(dfd, getname(pathname));
5590ff0d
UD
4311}
4312
3480b257 4313SYSCALL_DEFINE1(unlink, const char __user *, pathname)
5590ff0d 4314{
da2f1362 4315 return do_unlinkat(AT_FDCWD, getname(pathname));
5590ff0d
UD
4316}
4317
6521f891
CB
4318/**
4319 * vfs_symlink - create symlink
4320 * @mnt_userns: user namespace of the mount the inode was found from
4321 * @dir: inode of @dentry
4322 * @dentry: pointer to dentry of the base directory
4323 * @oldname: name of the file to link to
4324 *
4325 * Create a symlink.
4326 *
4327 * If the inode has been found through an idmapped mount the user namespace of
4328 * the vfsmount must be passed through @mnt_userns. This function will then take
4329 * care to map the inode according to @mnt_userns before checking permissions.
4330 * On non-idmapped mounts or if permission checking is to be performed on the
4331 * raw inode simply passs init_user_ns.
4332 */
4333int vfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
4334 struct dentry *dentry, const char *oldname)
1da177e4 4335{
6521f891 4336 int error = may_create(mnt_userns, dir, dentry);
1da177e4
LT
4337
4338 if (error)
4339 return error;
4340
acfa4380 4341 if (!dir->i_op->symlink)
1da177e4
LT
4342 return -EPERM;
4343
4344 error = security_inode_symlink(dir, dentry, oldname);
4345 if (error)
4346 return error;
4347
549c7297 4348 error = dir->i_op->symlink(mnt_userns, dir, dentry, oldname);
a74574aa 4349 if (!error)
f38aa942 4350 fsnotify_create(dir, dentry);
1da177e4
LT
4351 return error;
4352}
4d359507 4353EXPORT_SYMBOL(vfs_symlink);
1da177e4 4354
7a8721f8 4355int do_symlinkat(struct filename *from, int newdfd, struct filename *to)
1da177e4 4356{
2ad94ae6 4357 int error;
6902d925 4358 struct dentry *dentry;
dae6ad8f 4359 struct path path;
f46d3567 4360 unsigned int lookup_flags = 0;
1da177e4 4361
da2d0ced
DK
4362 if (IS_ERR(from)) {
4363 error = PTR_ERR(from);
4364 goto out_putnames;
4365 }
f46d3567 4366retry:
b4a4f213 4367 dentry = filename_create(newdfd, to, &path, lookup_flags);
6902d925
DH
4368 error = PTR_ERR(dentry);
4369 if (IS_ERR(dentry))
da2d0ced 4370 goto out_putnames;
6902d925 4371
91a27b2a 4372 error = security_path_symlink(&path, dentry, from->name);
6521f891
CB
4373 if (!error) {
4374 struct user_namespace *mnt_userns;
4375
4376 mnt_userns = mnt_user_ns(path.mnt);
4377 error = vfs_symlink(mnt_userns, path.dentry->d_inode, dentry,
4378 from->name);
4379 }
921a1650 4380 done_path_create(&path, dentry);
f46d3567
JL
4381 if (retry_estale(error, lookup_flags)) {
4382 lookup_flags |= LOOKUP_REVAL;
4383 goto retry;
4384 }
da2d0ced
DK
4385out_putnames:
4386 putname(to);
1da177e4
LT
4387 putname(from);
4388 return error;
4389}
4390
b724e846
DB
4391SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4392 int, newdfd, const char __user *, newname)
4393{
da2d0ced 4394 return do_symlinkat(getname(oldname), newdfd, getname(newname));
b724e846
DB
4395}
4396
3480b257 4397SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
5590ff0d 4398{
da2d0ced 4399 return do_symlinkat(getname(oldname), AT_FDCWD, getname(newname));
5590ff0d
UD
4400}
4401
146a8595
BF
4402/**
4403 * vfs_link - create a new link
4404 * @old_dentry: object to be linked
6521f891 4405 * @mnt_userns: the user namespace of the mount
146a8595
BF
4406 * @dir: new parent
4407 * @new_dentry: where to create the new link
4408 * @delegated_inode: returns inode needing a delegation break
4409 *
4410 * The caller must hold dir->i_mutex
4411 *
4412 * If vfs_link discovers a delegation on the to-be-linked file in need
4413 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4414 * inode in delegated_inode. The caller should then break the delegation
4415 * and retry. Because breaking a delegation may take a long time, the
4416 * caller should drop the i_mutex before doing so.
4417 *
4418 * Alternatively, a caller may pass NULL for delegated_inode. This may
4419 * be appropriate for callers that expect the underlying filesystem not
4420 * to be NFS exported.
6521f891
CB
4421 *
4422 * If the inode has been found through an idmapped mount the user namespace of
4423 * the vfsmount must be passed through @mnt_userns. This function will then take
4424 * care to map the inode according to @mnt_userns before checking permissions.
4425 * On non-idmapped mounts or if permission checking is to be performed on the
4426 * raw inode simply passs init_user_ns.
146a8595 4427 */
6521f891
CB
4428int vfs_link(struct dentry *old_dentry, struct user_namespace *mnt_userns,
4429 struct inode *dir, struct dentry *new_dentry,
4430 struct inode **delegated_inode)
1da177e4
LT
4431{
4432 struct inode *inode = old_dentry->d_inode;
8de52778 4433 unsigned max_links = dir->i_sb->s_max_links;
1da177e4
LT
4434 int error;
4435
4436 if (!inode)
4437 return -ENOENT;
4438
6521f891 4439 error = may_create(mnt_userns, dir, new_dentry);
1da177e4
LT
4440 if (error)
4441 return error;
4442
4443 if (dir->i_sb != inode->i_sb)
4444 return -EXDEV;
4445
4446 /*
4447 * A link to an append-only or immutable file cannot be created.
4448 */
4449 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4450 return -EPERM;
0bd23d09
EB
4451 /*
4452 * Updating the link count will likely cause i_uid and i_gid to
4453 * be writen back improperly if their true value is unknown to
4454 * the vfs.
4455 */
6521f891 4456 if (HAS_UNMAPPED_ID(mnt_userns, inode))
0bd23d09 4457 return -EPERM;
acfa4380 4458 if (!dir->i_op->link)
1da177e4 4459 return -EPERM;
7e79eedb 4460 if (S_ISDIR(inode->i_mode))
1da177e4
LT
4461 return -EPERM;
4462
4463 error = security_inode_link(old_dentry, dir, new_dentry);
4464 if (error)
4465 return error;
4466
5955102c 4467 inode_lock(inode);
aae8a97d 4468 /* Make sure we don't allow creating hardlink to an unlinked file */
f4e0c30c 4469 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
aae8a97d 4470 error = -ENOENT;
8de52778
AV
4471 else if (max_links && inode->i_nlink >= max_links)
4472 error = -EMLINK;
146a8595
BF
4473 else {
4474 error = try_break_deleg(inode, delegated_inode);
4475 if (!error)
4476 error = dir->i_op->link(old_dentry, dir, new_dentry);
4477 }
f4e0c30c
AV
4478
4479 if (!error && (inode->i_state & I_LINKABLE)) {
4480 spin_lock(&inode->i_lock);
4481 inode->i_state &= ~I_LINKABLE;
4482 spin_unlock(&inode->i_lock);
4483 }
5955102c 4484 inode_unlock(inode);
e31e14ec 4485 if (!error)
7e79eedb 4486 fsnotify_link(dir, inode, new_dentry);
1da177e4
LT
4487 return error;
4488}
4d359507 4489EXPORT_SYMBOL(vfs_link);
1da177e4
LT
4490
4491/*
4492 * Hardlinks are often used in delicate situations. We avoid
4493 * security-related surprises by not following symlinks on the
4494 * newname. --KAB
4495 *
4496 * We don't follow them on the oldname either to be compatible
4497 * with linux 2.0, and to avoid hard-linking to directories
4498 * and other special files. --ADM
4499 */
cf30da90 4500int do_linkat(int olddfd, struct filename *old, int newdfd,
020250f3 4501 struct filename *new, int flags)
1da177e4 4502{
6521f891 4503 struct user_namespace *mnt_userns;
1da177e4 4504 struct dentry *new_dentry;
dae6ad8f 4505 struct path old_path, new_path;
146a8595 4506 struct inode *delegated_inode = NULL;
11a7b371 4507 int how = 0;
1da177e4 4508 int error;
1da177e4 4509
020250f3
DK
4510 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) {
4511 error = -EINVAL;
4512 goto out_putnames;
4513 }
11a7b371 4514 /*
f0cc6ffb
LT
4515 * To use null names we require CAP_DAC_READ_SEARCH
4516 * This ensures that not everyone will be able to create
4517 * handlink using the passed filedescriptor.
11a7b371 4518 */
020250f3
DK
4519 if (flags & AT_EMPTY_PATH && !capable(CAP_DAC_READ_SEARCH)) {
4520 error = -ENOENT;
4521 goto out_putnames;
f0cc6ffb 4522 }
11a7b371
AK
4523
4524 if (flags & AT_SYMLINK_FOLLOW)
4525 how |= LOOKUP_FOLLOW;
442e31ca 4526retry:
794ebcea 4527 error = filename_lookup(olddfd, old, how, &old_path, NULL);
1da177e4 4528 if (error)
020250f3 4529 goto out_putnames;
2ad94ae6 4530
b4a4f213 4531 new_dentry = filename_create(newdfd, new, &new_path,
442e31ca 4532 (how & LOOKUP_REVAL));
1da177e4 4533 error = PTR_ERR(new_dentry);
6902d925 4534 if (IS_ERR(new_dentry))
020250f3 4535 goto out_putpath;
dae6ad8f
AV
4536
4537 error = -EXDEV;
4538 if (old_path.mnt != new_path.mnt)
4539 goto out_dput;
549c7297
CB
4540 mnt_userns = mnt_user_ns(new_path.mnt);
4541 error = may_linkat(mnt_userns, &old_path);
800179c9
KC
4542 if (unlikely(error))
4543 goto out_dput;
dae6ad8f 4544 error = security_path_link(old_path.dentry, &new_path, new_dentry);
be6d3e56 4545 if (error)
a8104a9f 4546 goto out_dput;
6521f891
CB
4547 error = vfs_link(old_path.dentry, mnt_userns, new_path.dentry->d_inode,
4548 new_dentry, &delegated_inode);
75c3f29d 4549out_dput:
921a1650 4550 done_path_create(&new_path, new_dentry);
146a8595
BF
4551 if (delegated_inode) {
4552 error = break_deleg_wait(&delegated_inode);
d22e6338
OD
4553 if (!error) {
4554 path_put(&old_path);
146a8595 4555 goto retry;
d22e6338 4556 }
146a8595 4557 }
442e31ca 4558 if (retry_estale(error, how)) {
d22e6338 4559 path_put(&old_path);
442e31ca
JL
4560 how |= LOOKUP_REVAL;
4561 goto retry;
4562 }
020250f3 4563out_putpath:
2d8f3038 4564 path_put(&old_path);
020250f3
DK
4565out_putnames:
4566 putname(old);
4567 putname(new);
1da177e4
LT
4568
4569 return error;
4570}
4571
46ea89eb
DB
4572SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4573 int, newdfd, const char __user *, newname, int, flags)
4574{
020250f3
DK
4575 return do_linkat(olddfd, getname_uflags(oldname, flags),
4576 newdfd, getname(newname), flags);
46ea89eb
DB
4577}
4578
3480b257 4579SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
5590ff0d 4580{
020250f3 4581 return do_linkat(AT_FDCWD, getname(oldname), AT_FDCWD, getname(newname), 0);
5590ff0d
UD
4582}
4583
bc27027a
MS
4584/**
4585 * vfs_rename - rename a filesystem object
2111c3c0 4586 * @rd: pointer to &struct renamedata info
bc27027a
MS
4587 *
4588 * The caller must hold multiple mutexes--see lock_rename()).
4589 *
4590 * If vfs_rename discovers a delegation in need of breaking at either
4591 * the source or destination, it will return -EWOULDBLOCK and return a
4592 * reference to the inode in delegated_inode. The caller should then
4593 * break the delegation and retry. Because breaking a delegation may
4594 * take a long time, the caller should drop all locks before doing
4595 * so.
4596 *
4597 * Alternatively, a caller may pass NULL for delegated_inode. This may
4598 * be appropriate for callers that expect the underlying filesystem not
4599 * to be NFS exported.
4600 *
1da177e4
LT
4601 * The worst of all namespace operations - renaming directory. "Perverted"
4602 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4603 * Problems:
0117d427 4604 *
d03b29a2 4605 * a) we can get into loop creation.
1da177e4
LT
4606 * b) race potential - two innocent renames can create a loop together.
4607 * That's where 4.4 screws up. Current fix: serialization on
a11f3a05 4608 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
1da177e4 4609 * story.
6cedba89
BF
4610 * c) we have to lock _four_ objects - parents and victim (if it exists),
4611 * and source (if it is not a directory).
1b1dcc1b 4612 * And that - after we got ->i_mutex on parents (until then we don't know
1da177e4
LT
4613 * whether the target exists). Solution: try to be smart with locking
4614 * order for inodes. We rely on the fact that tree topology may change
a11f3a05 4615 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
1da177e4
LT
4616 * move will be locked. Thus we can rank directories by the tree
4617 * (ancestors first) and rank all non-directories after them.
4618 * That works since everybody except rename does "lock parent, lookup,
a11f3a05 4619 * lock child" and rename is under ->s_vfs_rename_mutex.
1da177e4
LT
4620 * HOWEVER, it relies on the assumption that any object with ->lookup()
4621 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4622 * we'd better make sure that there's no link(2) for them.
e4eaac06 4623 * d) conversion from fhandle to dentry may come in the wrong moment - when
1b1dcc1b 4624 * we are removing the target. Solution: we will have to grab ->i_mutex
1da177e4 4625 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
c41b20e7 4626 * ->i_mutex on parents, which works but leads to some truly excessive
1da177e4
LT
4627 * locking].
4628 */
9fe61450 4629int vfs_rename(struct renamedata *rd)
1da177e4 4630{
bc27027a 4631 int error;
9fe61450
CB
4632 struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir;
4633 struct dentry *old_dentry = rd->old_dentry;
4634 struct dentry *new_dentry = rd->new_dentry;
4635 struct inode **delegated_inode = rd->delegated_inode;
4636 unsigned int flags = rd->flags;
bc27027a 4637 bool is_dir = d_is_dir(old_dentry);
bc27027a 4638 struct inode *source = old_dentry->d_inode;
9055cba7 4639 struct inode *target = new_dentry->d_inode;
da1ce067
MS
4640 bool new_is_dir = false;
4641 unsigned max_links = new_dir->i_sb->s_max_links;
49d31c2f 4642 struct name_snapshot old_name;
bc27027a 4643
8d3e2936 4644 if (source == target)
bc27027a
MS
4645 return 0;
4646
6521f891 4647 error = may_delete(rd->old_mnt_userns, old_dir, old_dentry, is_dir);
bc27027a
MS
4648 if (error)
4649 return error;
4650
da1ce067 4651 if (!target) {
6521f891 4652 error = may_create(rd->new_mnt_userns, new_dir, new_dentry);
da1ce067
MS
4653 } else {
4654 new_is_dir = d_is_dir(new_dentry);
4655
4656 if (!(flags & RENAME_EXCHANGE))
6521f891
CB
4657 error = may_delete(rd->new_mnt_userns, new_dir,
4658 new_dentry, is_dir);
da1ce067 4659 else
6521f891
CB
4660 error = may_delete(rd->new_mnt_userns, new_dir,
4661 new_dentry, new_is_dir);
da1ce067 4662 }
bc27027a
MS
4663 if (error)
4664 return error;
4665
2773bf00 4666 if (!old_dir->i_op->rename)
bc27027a 4667 return -EPERM;
1da177e4
LT
4668
4669 /*
4670 * If we are going to change the parent - check write permissions,
4671 * we'll need to flip '..'.
4672 */
da1ce067
MS
4673 if (new_dir != old_dir) {
4674 if (is_dir) {
6521f891 4675 error = inode_permission(rd->old_mnt_userns, source,
47291baa 4676 MAY_WRITE);
da1ce067
MS
4677 if (error)
4678 return error;
4679 }
4680 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
6521f891 4681 error = inode_permission(rd->new_mnt_userns, target,
47291baa 4682 MAY_WRITE);
da1ce067
MS
4683 if (error)
4684 return error;
4685 }
1da177e4
LT
4686 }
4687
0b3974eb
MS
4688 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4689 flags);
1da177e4
LT
4690 if (error)
4691 return error;
4692
49d31c2f 4693 take_dentry_name_snapshot(&old_name, old_dentry);
1d2ef590 4694 dget(new_dentry);
da1ce067 4695 if (!is_dir || (flags & RENAME_EXCHANGE))
bc27027a
MS
4696 lock_two_nondirectories(source, target);
4697 else if (target)
5955102c 4698 inode_lock(target);
9055cba7 4699
51cc3a66
HD
4700 error = -EPERM;
4701 if (IS_SWAPFILE(source) || (target && IS_SWAPFILE(target)))
4702 goto out;
4703
9055cba7 4704 error = -EBUSY;
7af1364f 4705 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
9055cba7
SW
4706 goto out;
4707
da1ce067 4708 if (max_links && new_dir != old_dir) {
bc27027a 4709 error = -EMLINK;
da1ce067 4710 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
bc27027a 4711 goto out;
da1ce067
MS
4712 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4713 old_dir->i_nlink >= max_links)
4714 goto out;
4715 }
da1ce067 4716 if (!is_dir) {
bc27027a 4717 error = try_break_deleg(source, delegated_inode);
8e6d782c
BF
4718 if (error)
4719 goto out;
da1ce067
MS
4720 }
4721 if (target && !new_is_dir) {
4722 error = try_break_deleg(target, delegated_inode);
4723 if (error)
4724 goto out;
8e6d782c 4725 }
549c7297
CB
4726 error = old_dir->i_op->rename(rd->new_mnt_userns, old_dir, old_dentry,
4727 new_dir, new_dentry, flags);
51892bbb
SW
4728 if (error)
4729 goto out;
4730
da1ce067 4731 if (!(flags & RENAME_EXCHANGE) && target) {
8767712f
AV
4732 if (is_dir) {
4733 shrink_dcache_parent(new_dentry);
bc27027a 4734 target->i_flags |= S_DEAD;
8767712f 4735 }
51892bbb 4736 dont_mount(new_dentry);
8ed936b5 4737 detach_mounts(new_dentry);
bc27027a 4738 }
da1ce067
MS
4739 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4740 if (!(flags & RENAME_EXCHANGE))
4741 d_move(old_dentry, new_dentry);
4742 else
4743 d_exchange(old_dentry, new_dentry);
4744 }
51892bbb 4745out:
da1ce067 4746 if (!is_dir || (flags & RENAME_EXCHANGE))
bc27027a
MS
4747 unlock_two_nondirectories(source, target);
4748 else if (target)
5955102c 4749 inode_unlock(target);
1da177e4 4750 dput(new_dentry);
da1ce067 4751 if (!error) {
f4ec3a3d 4752 fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
da1ce067
MS
4753 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4754 if (flags & RENAME_EXCHANGE) {
f4ec3a3d 4755 fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
da1ce067
MS
4756 new_is_dir, NULL, new_dentry);
4757 }
4758 }
49d31c2f 4759 release_dentry_name_snapshot(&old_name);
0eeca283 4760
1da177e4
LT
4761 return error;
4762}
4d359507 4763EXPORT_SYMBOL(vfs_rename);
1da177e4 4764
e886663c
JA
4765int do_renameat2(int olddfd, struct filename *from, int newdfd,
4766 struct filename *to, unsigned int flags)
1da177e4 4767{
9fe61450 4768 struct renamedata rd;
2ad94ae6
AV
4769 struct dentry *old_dentry, *new_dentry;
4770 struct dentry *trap;
f5beed75
AV
4771 struct path old_path, new_path;
4772 struct qstr old_last, new_last;
4773 int old_type, new_type;
8e6d782c 4774 struct inode *delegated_inode = NULL;
f5beed75 4775 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
c6a94284 4776 bool should_retry = false;
e886663c 4777 int error = -EINVAL;
520c8b16 4778
0d7a8555 4779 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
0ee50b47 4780 goto put_names;
da1ce067 4781
0d7a8555
MS
4782 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4783 (flags & RENAME_EXCHANGE))
0ee50b47 4784 goto put_names;
520c8b16 4785
f5beed75
AV
4786 if (flags & RENAME_EXCHANGE)
4787 target_flags = 0;
4788
c6a94284 4789retry:
c5f563f9
AV
4790 error = filename_parentat(olddfd, from, lookup_flags, &old_path,
4791 &old_last, &old_type);
0ee50b47
DK
4792 if (error)
4793 goto put_names;
1da177e4 4794
c5f563f9
AV
4795 error = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last,
4796 &new_type);
0ee50b47 4797 if (error)
1da177e4
LT
4798 goto exit1;
4799
4800 error = -EXDEV;
f5beed75 4801 if (old_path.mnt != new_path.mnt)
1da177e4
LT
4802 goto exit2;
4803
1da177e4 4804 error = -EBUSY;
f5beed75 4805 if (old_type != LAST_NORM)
1da177e4
LT
4806 goto exit2;
4807
0a7c3937
MS
4808 if (flags & RENAME_NOREPLACE)
4809 error = -EEXIST;
f5beed75 4810 if (new_type != LAST_NORM)
1da177e4
LT
4811 goto exit2;
4812
f5beed75 4813 error = mnt_want_write(old_path.mnt);
c30dabfe
JK
4814 if (error)
4815 goto exit2;
4816
8e6d782c 4817retry_deleg:
f5beed75 4818 trap = lock_rename(new_path.dentry, old_path.dentry);
1da177e4 4819
f5beed75 4820 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
1da177e4
LT
4821 error = PTR_ERR(old_dentry);
4822 if (IS_ERR(old_dentry))
4823 goto exit3;
4824 /* source must exist */
4825 error = -ENOENT;
b18825a7 4826 if (d_is_negative(old_dentry))
1da177e4 4827 goto exit4;
f5beed75 4828 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
0a7c3937
MS
4829 error = PTR_ERR(new_dentry);
4830 if (IS_ERR(new_dentry))
4831 goto exit4;
4832 error = -EEXIST;
4833 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4834 goto exit5;
da1ce067
MS
4835 if (flags & RENAME_EXCHANGE) {
4836 error = -ENOENT;
4837 if (d_is_negative(new_dentry))
4838 goto exit5;
4839
4840 if (!d_is_dir(new_dentry)) {
4841 error = -ENOTDIR;
f5beed75 4842 if (new_last.name[new_last.len])
da1ce067
MS
4843 goto exit5;
4844 }
4845 }
1da177e4 4846 /* unless the source is a directory trailing slashes give -ENOTDIR */
44b1d530 4847 if (!d_is_dir(old_dentry)) {
1da177e4 4848 error = -ENOTDIR;
f5beed75 4849 if (old_last.name[old_last.len])
0a7c3937 4850 goto exit5;
f5beed75 4851 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
0a7c3937 4852 goto exit5;
1da177e4
LT
4853 }
4854 /* source should not be ancestor of target */
4855 error = -EINVAL;
4856 if (old_dentry == trap)
0a7c3937 4857 goto exit5;
1da177e4 4858 /* target should not be an ancestor of source */
da1ce067
MS
4859 if (!(flags & RENAME_EXCHANGE))
4860 error = -ENOTEMPTY;
1da177e4
LT
4861 if (new_dentry == trap)
4862 goto exit5;
4863
f5beed75
AV
4864 error = security_path_rename(&old_path, old_dentry,
4865 &new_path, new_dentry, flags);
be6d3e56 4866 if (error)
c30dabfe 4867 goto exit5;
9fe61450
CB
4868
4869 rd.old_dir = old_path.dentry->d_inode;
4870 rd.old_dentry = old_dentry;
6521f891 4871 rd.old_mnt_userns = mnt_user_ns(old_path.mnt);
9fe61450
CB
4872 rd.new_dir = new_path.dentry->d_inode;
4873 rd.new_dentry = new_dentry;
6521f891 4874 rd.new_mnt_userns = mnt_user_ns(new_path.mnt);
9fe61450
CB
4875 rd.delegated_inode = &delegated_inode;
4876 rd.flags = flags;
4877 error = vfs_rename(&rd);
1da177e4
LT
4878exit5:
4879 dput(new_dentry);
4880exit4:
4881 dput(old_dentry);
4882exit3:
f5beed75 4883 unlock_rename(new_path.dentry, old_path.dentry);
8e6d782c
BF
4884 if (delegated_inode) {
4885 error = break_deleg_wait(&delegated_inode);
4886 if (!error)
4887 goto retry_deleg;
4888 }
f5beed75 4889 mnt_drop_write(old_path.mnt);
1da177e4 4890exit2:
c6a94284
JL
4891 if (retry_estale(error, lookup_flags))
4892 should_retry = true;
f5beed75 4893 path_put(&new_path);
1da177e4 4894exit1:
f5beed75 4895 path_put(&old_path);
c6a94284
JL
4896 if (should_retry) {
4897 should_retry = false;
4898 lookup_flags |= LOOKUP_REVAL;
4899 goto retry;
4900 }
0ee50b47 4901put_names:
91ef658f 4902 putname(from);
91ef658f 4903 putname(to);
1da177e4
LT
4904 return error;
4905}
4906
ee81feb6
DB
4907SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4908 int, newdfd, const char __user *, newname, unsigned int, flags)
4909{
e886663c
JA
4910 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
4911 flags);
ee81feb6
DB
4912}
4913
520c8b16
MS
4914SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4915 int, newdfd, const char __user *, newname)
4916{
e886663c
JA
4917 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
4918 0);
520c8b16
MS
4919}
4920
a26eab24 4921SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5590ff0d 4922{
e886663c
JA
4923 return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD,
4924 getname(newname), 0);
5590ff0d
UD
4925}
4926
5d826c84 4927int readlink_copy(char __user *buffer, int buflen, const char *link)
1da177e4 4928{
5d826c84 4929 int len = PTR_ERR(link);
1da177e4
LT
4930 if (IS_ERR(link))
4931 goto out;
4932
4933 len = strlen(link);
4934 if (len > (unsigned) buflen)
4935 len = buflen;
4936 if (copy_to_user(buffer, link, len))
4937 len = -EFAULT;
4938out:
4939 return len;
4940}
4941
fd4a0edf
MS
4942/**
4943 * vfs_readlink - copy symlink body into userspace buffer
4944 * @dentry: dentry on which to get symbolic link
4945 * @buffer: user memory pointer
4946 * @buflen: size of buffer
4947 *
4948 * Does not touch atime. That's up to the caller if necessary
4949 *
4950 * Does not call security hook.
4951 */
4952int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4953{
4954 struct inode *inode = d_inode(dentry);
f2df5da6
AV
4955 DEFINE_DELAYED_CALL(done);
4956 const char *link;
4957 int res;
fd4a0edf 4958
76fca90e
MS
4959 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
4960 if (unlikely(inode->i_op->readlink))
4961 return inode->i_op->readlink(dentry, buffer, buflen);
4962
4963 if (!d_is_symlink(dentry))
4964 return -EINVAL;
4965
4966 spin_lock(&inode->i_lock);
4967 inode->i_opflags |= IOP_DEFAULT_READLINK;
4968 spin_unlock(&inode->i_lock);
4969 }
fd4a0edf 4970
4c4f7c19 4971 link = READ_ONCE(inode->i_link);
f2df5da6
AV
4972 if (!link) {
4973 link = inode->i_op->get_link(dentry, inode, &done);
4974 if (IS_ERR(link))
4975 return PTR_ERR(link);
4976 }
4977 res = readlink_copy(buffer, buflen, link);
4978 do_delayed_call(&done);
4979 return res;
fd4a0edf
MS
4980}
4981EXPORT_SYMBOL(vfs_readlink);
1da177e4 4982
d60874cd
MS
4983/**
4984 * vfs_get_link - get symlink body
4985 * @dentry: dentry on which to get symbolic link
4986 * @done: caller needs to free returned data with this
4987 *
4988 * Calls security hook and i_op->get_link() on the supplied inode.
4989 *
4990 * It does not touch atime. That's up to the caller if necessary.
4991 *
4992 * Does not work on "special" symlinks like /proc/$$/fd/N
4993 */
4994const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4995{
4996 const char *res = ERR_PTR(-EINVAL);
4997 struct inode *inode = d_inode(dentry);
4998
4999 if (d_is_symlink(dentry)) {
5000 res = ERR_PTR(security_inode_readlink(dentry));
5001 if (!res)
5002 res = inode->i_op->get_link(dentry, inode, done);
5003 }
5004 return res;
5005}
5006EXPORT_SYMBOL(vfs_get_link);
5007
1da177e4 5008/* get the link contents into pagecache */
6b255391 5009const char *page_get_link(struct dentry *dentry, struct inode *inode,
fceef393 5010 struct delayed_call *callback)
1da177e4 5011{
ebd09abb
DG
5012 char *kaddr;
5013 struct page *page;
6b255391
AV
5014 struct address_space *mapping = inode->i_mapping;
5015
d3883d4f
AV
5016 if (!dentry) {
5017 page = find_get_page(mapping, 0);
5018 if (!page)
5019 return ERR_PTR(-ECHILD);
5020 if (!PageUptodate(page)) {
5021 put_page(page);
5022 return ERR_PTR(-ECHILD);
5023 }
5024 } else {
5025 page = read_mapping_page(mapping, 0, NULL);
5026 if (IS_ERR(page))
5027 return (char*)page;
5028 }
fceef393 5029 set_delayed_call(callback, page_put_link, page);
21fc61c7
AV
5030 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
5031 kaddr = page_address(page);
6b255391 5032 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
ebd09abb 5033 return kaddr;
1da177e4
LT
5034}
5035
6b255391 5036EXPORT_SYMBOL(page_get_link);
1da177e4 5037
fceef393 5038void page_put_link(void *arg)
1da177e4 5039{
fceef393 5040 put_page(arg);
1da177e4 5041}
4d359507 5042EXPORT_SYMBOL(page_put_link);
1da177e4 5043
aa80deab
AV
5044int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
5045{
fceef393 5046 DEFINE_DELAYED_CALL(done);
6b255391
AV
5047 int res = readlink_copy(buffer, buflen,
5048 page_get_link(dentry, d_inode(dentry),
fceef393
AV
5049 &done));
5050 do_delayed_call(&done);
aa80deab
AV
5051 return res;
5052}
5053EXPORT_SYMBOL(page_readlink);
5054
56f5746c 5055int page_symlink(struct inode *inode, const char *symname, int len)
1da177e4
LT
5056{
5057 struct address_space *mapping = inode->i_mapping;
27a77913 5058 const struct address_space_operations *aops = mapping->a_ops;
56f5746c 5059 bool nofs = !mapping_gfp_constraint(mapping, __GFP_FS);
0adb25d2 5060 struct page *page;
afddba49 5061 void *fsdata;
beb497ab 5062 int err;
2d878178 5063 unsigned int flags;
1da177e4 5064
7e53cac4 5065retry:
2d878178
MWO
5066 if (nofs)
5067 flags = memalloc_nofs_save();
27a77913 5068 err = aops->write_begin(NULL, mapping, 0, len-1, &page, &fsdata);
2d878178
MWO
5069 if (nofs)
5070 memalloc_nofs_restore(flags);
1da177e4 5071 if (err)
afddba49
NP
5072 goto fail;
5073
21fc61c7 5074 memcpy(page_address(page), symname, len-1);
afddba49 5075
27a77913 5076 err = aops->write_end(NULL, mapping, 0, len-1, len-1,
afddba49 5077 page, fsdata);
1da177e4
LT
5078 if (err < 0)
5079 goto fail;
afddba49
NP
5080 if (err < len-1)
5081 goto retry;
5082
1da177e4
LT
5083 mark_inode_dirty(inode);
5084 return 0;
1da177e4
LT
5085fail:
5086 return err;
5087}
4d359507 5088EXPORT_SYMBOL(page_symlink);
0adb25d2 5089
92e1d5be 5090const struct inode_operations page_symlink_inode_operations = {
6b255391 5091 .get_link = page_get_link,
1da177e4 5092};
1da177e4 5093EXPORT_SYMBOL(page_symlink_inode_operations);
This page took 2.650495 seconds and 4 git commands to generate.