Loading...
1// SPDX-License-Identifier: GPL-2.0
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>
19#include <linux/export.h>
20#include <linux/slab.h>
21#include <linux/wordpart.h>
22#include <linux/fs.h>
23#include <linux/filelock.h>
24#include <linux/namei.h>
25#include <linux/pagemap.h>
26#include <linux/sched/mm.h>
27#include <linux/fsnotify.h>
28#include <linux/personality.h>
29#include <linux/security.h>
30#include <linux/syscalls.h>
31#include <linux/mount.h>
32#include <linux/audit.h>
33#include <linux/capability.h>
34#include <linux/file.h>
35#include <linux/fcntl.h>
36#include <linux/device_cgroup.h>
37#include <linux/fs_struct.h>
38#include <linux/posix_acl.h>
39#include <linux/hash.h>
40#include <linux/bitops.h>
41#include <linux/init_task.h>
42#include <linux/uaccess.h>
43
44#include "internal.h"
45#include "mount.h"
46
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
81 * the name is a symlink pointing to a non-existent name.
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)
114 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
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 */
125
126#define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
127
128struct filename *
129getname_flags(const char __user *filename, int flags)
130{
131 struct filename *result;
132 char *kname;
133 int len;
134
135 result = audit_reusename(filename);
136 if (result)
137 return result;
138
139 result = __getname();
140 if (unlikely(!result))
141 return ERR_PTR(-ENOMEM);
142
143 /*
144 * First, try to embed the struct filename inside the names_cache
145 * allocation
146 */
147 kname = (char *)result->iname;
148 result->name = kname;
149
150 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
151 /*
152 * Handle both empty path and copy failure in one go.
153 */
154 if (unlikely(len <= 0)) {
155 if (unlikely(len < 0)) {
156 __putname(result);
157 return ERR_PTR(len);
158 }
159
160 /* The empty path is special. */
161 if (!(flags & LOOKUP_EMPTY)) {
162 __putname(result);
163 return ERR_PTR(-ENOENT);
164 }
165 }
166
167 /*
168 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
169 * separate struct filename so we can dedicate the entire
170 * names_cache allocation for the pathname, and re-do the copy from
171 * userland.
172 */
173 if (unlikely(len == EMBEDDED_NAME_MAX)) {
174 const size_t size = offsetof(struct filename, iname[1]);
175 kname = (char *)result;
176
177 /*
178 * size is chosen that way we to guarantee that
179 * result->iname[0] is within the same object and that
180 * kname can't be equal to result->iname, no matter what.
181 */
182 result = kzalloc(size, GFP_KERNEL);
183 if (unlikely(!result)) {
184 __putname(kname);
185 return ERR_PTR(-ENOMEM);
186 }
187 result->name = kname;
188 len = strncpy_from_user(kname, filename, PATH_MAX);
189 if (unlikely(len < 0)) {
190 __putname(kname);
191 kfree(result);
192 return ERR_PTR(len);
193 }
194 /* The empty path is special. */
195 if (unlikely(!len) && !(flags & LOOKUP_EMPTY)) {
196 __putname(kname);
197 kfree(result);
198 return ERR_PTR(-ENOENT);
199 }
200 if (unlikely(len == PATH_MAX)) {
201 __putname(kname);
202 kfree(result);
203 return ERR_PTR(-ENAMETOOLONG);
204 }
205 }
206
207 atomic_set(&result->refcnt, 1);
208 result->uptr = filename;
209 result->aname = NULL;
210 audit_getname(result);
211 return result;
212}
213
214struct filename *getname_uflags(const char __user *filename, int uflags)
215{
216 int flags = (uflags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
217
218 return getname_flags(filename, flags);
219}
220
221struct filename *getname(const char __user * filename)
222{
223 return getname_flags(filename, 0);
224}
225
226struct filename *__getname_maybe_null(const char __user *pathname)
227{
228 struct filename *name;
229 char c;
230
231 /* try to save on allocations; loss on um, though */
232 if (get_user(c, pathname))
233 return ERR_PTR(-EFAULT);
234 if (!c)
235 return NULL;
236
237 name = getname_flags(pathname, LOOKUP_EMPTY);
238 if (!IS_ERR(name) && !(name->name[0])) {
239 putname(name);
240 name = NULL;
241 }
242 return name;
243}
244
245struct filename *getname_kernel(const char * filename)
246{
247 struct filename *result;
248 int len = strlen(filename) + 1;
249
250 result = __getname();
251 if (unlikely(!result))
252 return ERR_PTR(-ENOMEM);
253
254 if (len <= EMBEDDED_NAME_MAX) {
255 result->name = (char *)result->iname;
256 } else if (len <= PATH_MAX) {
257 const size_t size = offsetof(struct filename, iname[1]);
258 struct filename *tmp;
259
260 tmp = kmalloc(size, GFP_KERNEL);
261 if (unlikely(!tmp)) {
262 __putname(result);
263 return ERR_PTR(-ENOMEM);
264 }
265 tmp->name = (char *)result;
266 result = tmp;
267 } else {
268 __putname(result);
269 return ERR_PTR(-ENAMETOOLONG);
270 }
271 memcpy((char *)result->name, filename, len);
272 result->uptr = NULL;
273 result->aname = NULL;
274 atomic_set(&result->refcnt, 1);
275 audit_getname(result);
276
277 return result;
278}
279EXPORT_SYMBOL(getname_kernel);
280
281void putname(struct filename *name)
282{
283 if (IS_ERR_OR_NULL(name))
284 return;
285
286 if (WARN_ON_ONCE(!atomic_read(&name->refcnt)))
287 return;
288
289 if (!atomic_dec_and_test(&name->refcnt))
290 return;
291
292 if (name->name != name->iname) {
293 __putname(name->name);
294 kfree(name);
295 } else
296 __putname(name);
297}
298EXPORT_SYMBOL(putname);
299
300/**
301 * check_acl - perform ACL permission checking
302 * @idmap: idmap of the mount the inode was found from
303 * @inode: inode to check permissions on
304 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
305 *
306 * This function performs the ACL permission checking. Since this function
307 * retrieve POSIX acls it needs to know whether it is called from a blocking or
308 * non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
309 *
310 * If the inode has been found through an idmapped mount the idmap of
311 * the vfsmount must be passed through @idmap. This function will then take
312 * care to map the inode according to @idmap before checking permissions.
313 * On non-idmapped mounts or if permission checking is to be performed on the
314 * raw inode simply pass @nop_mnt_idmap.
315 */
316static int check_acl(struct mnt_idmap *idmap,
317 struct inode *inode, int mask)
318{
319#ifdef CONFIG_FS_POSIX_ACL
320 struct posix_acl *acl;
321
322 if (mask & MAY_NOT_BLOCK) {
323 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
324 if (!acl)
325 return -EAGAIN;
326 /* no ->get_inode_acl() calls in RCU mode... */
327 if (is_uncached_acl(acl))
328 return -ECHILD;
329 return posix_acl_permission(idmap, inode, acl, mask);
330 }
331
332 acl = get_inode_acl(inode, ACL_TYPE_ACCESS);
333 if (IS_ERR(acl))
334 return PTR_ERR(acl);
335 if (acl) {
336 int error = posix_acl_permission(idmap, inode, acl, mask);
337 posix_acl_release(acl);
338 return error;
339 }
340#endif
341
342 return -EAGAIN;
343}
344
345/*
346 * Very quick optimistic "we know we have no ACL's" check.
347 *
348 * Note that this is purely for ACL_TYPE_ACCESS, and purely
349 * for the "we have cached that there are no ACLs" case.
350 *
351 * If this returns true, we know there are no ACLs. But if
352 * it returns false, we might still not have ACLs (it could
353 * be the is_uncached_acl() case).
354 */
355static inline bool no_acl_inode(struct inode *inode)
356{
357#ifdef CONFIG_FS_POSIX_ACL
358 return likely(!READ_ONCE(inode->i_acl));
359#else
360 return true;
361#endif
362}
363
364/**
365 * acl_permission_check - perform basic UNIX permission checking
366 * @idmap: idmap of the mount the inode was found from
367 * @inode: inode to check permissions on
368 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
369 *
370 * This function performs the basic UNIX permission checking. Since this
371 * function may retrieve POSIX acls it needs to know whether it is called from a
372 * blocking or non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
373 *
374 * If the inode has been found through an idmapped mount the idmap of
375 * the vfsmount must be passed through @idmap. This function will then take
376 * care to map the inode according to @idmap before checking permissions.
377 * On non-idmapped mounts or if permission checking is to be performed on the
378 * raw inode simply pass @nop_mnt_idmap.
379 */
380static int acl_permission_check(struct mnt_idmap *idmap,
381 struct inode *inode, int mask)
382{
383 unsigned int mode = inode->i_mode;
384 vfsuid_t vfsuid;
385
386 /*
387 * Common cheap case: everybody has the requested
388 * rights, and there are no ACLs to check. No need
389 * to do any owner/group checks in that case.
390 *
391 * - 'mask&7' is the requested permission bit set
392 * - multiplying by 0111 spreads them out to all of ugo
393 * - '& ~mode' looks for missing inode permission bits
394 * - the '!' is for "no missing permissions"
395 *
396 * After that, we just need to check that there are no
397 * ACL's on the inode - do the 'IS_POSIXACL()' check last
398 * because it will dereference the ->i_sb pointer and we
399 * want to avoid that if at all possible.
400 */
401 if (!((mask & 7) * 0111 & ~mode)) {
402 if (no_acl_inode(inode))
403 return 0;
404 if (!IS_POSIXACL(inode))
405 return 0;
406 }
407
408 /* Are we the owner? If so, ACL's don't matter */
409 vfsuid = i_uid_into_vfsuid(idmap, inode);
410 if (likely(vfsuid_eq_kuid(vfsuid, current_fsuid()))) {
411 mask &= 7;
412 mode >>= 6;
413 return (mask & ~mode) ? -EACCES : 0;
414 }
415
416 /* Do we have ACL's? */
417 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
418 int error = check_acl(idmap, inode, mask);
419 if (error != -EAGAIN)
420 return error;
421 }
422
423 /* Only RWX matters for group/other mode bits */
424 mask &= 7;
425
426 /*
427 * Are the group permissions different from
428 * the other permissions in the bits we care
429 * about? Need to check group ownership if so.
430 */
431 if (mask & (mode ^ (mode >> 3))) {
432 vfsgid_t vfsgid = i_gid_into_vfsgid(idmap, inode);
433 if (vfsgid_in_group_p(vfsgid))
434 mode >>= 3;
435 }
436
437 /* Bits in 'mode' clear that we require? */
438 return (mask & ~mode) ? -EACCES : 0;
439}
440
441/**
442 * generic_permission - check for access rights on a Posix-like filesystem
443 * @idmap: idmap of the mount the inode was found from
444 * @inode: inode to check access rights for
445 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC,
446 * %MAY_NOT_BLOCK ...)
447 *
448 * Used to check for read/write/execute permissions on a file.
449 * We use "fsuid" for this, letting us set arbitrary permissions
450 * for filesystem access without changing the "normal" uids which
451 * are used for other things.
452 *
453 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
454 * request cannot be satisfied (eg. requires blocking or too much complexity).
455 * It would then be called again in ref-walk mode.
456 *
457 * If the inode has been found through an idmapped mount the idmap of
458 * the vfsmount must be passed through @idmap. This function will then take
459 * care to map the inode according to @idmap before checking permissions.
460 * On non-idmapped mounts or if permission checking is to be performed on the
461 * raw inode simply pass @nop_mnt_idmap.
462 */
463int generic_permission(struct mnt_idmap *idmap, struct inode *inode,
464 int mask)
465{
466 int ret;
467
468 /*
469 * Do the basic permission checks.
470 */
471 ret = acl_permission_check(idmap, inode, mask);
472 if (ret != -EACCES)
473 return ret;
474
475 if (S_ISDIR(inode->i_mode)) {
476 /* DACs are overridable for directories */
477 if (!(mask & MAY_WRITE))
478 if (capable_wrt_inode_uidgid(idmap, inode,
479 CAP_DAC_READ_SEARCH))
480 return 0;
481 if (capable_wrt_inode_uidgid(idmap, inode,
482 CAP_DAC_OVERRIDE))
483 return 0;
484 return -EACCES;
485 }
486
487 /*
488 * Searching includes executable on directories, else just read.
489 */
490 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
491 if (mask == MAY_READ)
492 if (capable_wrt_inode_uidgid(idmap, inode,
493 CAP_DAC_READ_SEARCH))
494 return 0;
495 /*
496 * Read/write DACs are always overridable.
497 * Executable DACs are overridable when there is
498 * at least one exec bit set.
499 */
500 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
501 if (capable_wrt_inode_uidgid(idmap, inode,
502 CAP_DAC_OVERRIDE))
503 return 0;
504
505 return -EACCES;
506}
507EXPORT_SYMBOL(generic_permission);
508
509/**
510 * do_inode_permission - UNIX permission checking
511 * @idmap: idmap of the mount the inode was found from
512 * @inode: inode to check permissions on
513 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
514 *
515 * We _really_ want to just do "generic_permission()" without
516 * even looking at the inode->i_op values. So we keep a cache
517 * flag in inode->i_opflags, that says "this has not special
518 * permission function, use the fast case".
519 */
520static inline int do_inode_permission(struct mnt_idmap *idmap,
521 struct inode *inode, int mask)
522{
523 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
524 if (likely(inode->i_op->permission))
525 return inode->i_op->permission(idmap, inode, mask);
526
527 /* This gets set once for the inode lifetime */
528 spin_lock(&inode->i_lock);
529 inode->i_opflags |= IOP_FASTPERM;
530 spin_unlock(&inode->i_lock);
531 }
532 return generic_permission(idmap, inode, mask);
533}
534
535/**
536 * sb_permission - Check superblock-level permissions
537 * @sb: Superblock of inode to check permission on
538 * @inode: Inode to check permission on
539 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
540 *
541 * Separate out file-system wide checks from inode-specific permission checks.
542 */
543static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
544{
545 if (unlikely(mask & MAY_WRITE)) {
546 umode_t mode = inode->i_mode;
547
548 /* Nobody gets write access to a read-only fs. */
549 if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
550 return -EROFS;
551 }
552 return 0;
553}
554
555/**
556 * inode_permission - Check for access rights to a given inode
557 * @idmap: idmap of the mount the inode was found from
558 * @inode: Inode to check permission on
559 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
560 *
561 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
562 * this, letting us set arbitrary permissions for filesystem access without
563 * changing the "normal" UIDs which are used for other things.
564 *
565 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
566 */
567int inode_permission(struct mnt_idmap *idmap,
568 struct inode *inode, int mask)
569{
570 int retval;
571
572 retval = sb_permission(inode->i_sb, inode, mask);
573 if (retval)
574 return retval;
575
576 if (unlikely(mask & MAY_WRITE)) {
577 /*
578 * Nobody gets write access to an immutable file.
579 */
580 if (IS_IMMUTABLE(inode))
581 return -EPERM;
582
583 /*
584 * Updating mtime will likely cause i_uid and i_gid to be
585 * written back improperly if their true value is unknown
586 * to the vfs.
587 */
588 if (HAS_UNMAPPED_ID(idmap, inode))
589 return -EACCES;
590 }
591
592 retval = do_inode_permission(idmap, inode, mask);
593 if (retval)
594 return retval;
595
596 retval = devcgroup_inode_permission(inode, mask);
597 if (retval)
598 return retval;
599
600 return security_inode_permission(inode, mask);
601}
602EXPORT_SYMBOL(inode_permission);
603
604/**
605 * path_get - get a reference to a path
606 * @path: path to get the reference to
607 *
608 * Given a path increment the reference count to the dentry and the vfsmount.
609 */
610void path_get(const struct path *path)
611{
612 mntget(path->mnt);
613 dget(path->dentry);
614}
615EXPORT_SYMBOL(path_get);
616
617/**
618 * path_put - put a reference to a path
619 * @path: path to put the reference to
620 *
621 * Given a path decrement the reference count to the dentry and the vfsmount.
622 */
623void path_put(const struct path *path)
624{
625 dput(path->dentry);
626 mntput(path->mnt);
627}
628EXPORT_SYMBOL(path_put);
629
630#define EMBEDDED_LEVELS 2
631struct nameidata {
632 struct path path;
633 struct qstr last;
634 struct path root;
635 struct inode *inode; /* path.dentry.d_inode */
636 unsigned int flags, state;
637 unsigned seq, next_seq, m_seq, r_seq;
638 int last_type;
639 unsigned depth;
640 int total_link_count;
641 struct saved {
642 struct path link;
643 struct delayed_call done;
644 const char *name;
645 unsigned seq;
646 } *stack, internal[EMBEDDED_LEVELS];
647 struct filename *name;
648 const char *pathname;
649 struct nameidata *saved;
650 unsigned root_seq;
651 int dfd;
652 vfsuid_t dir_vfsuid;
653 umode_t dir_mode;
654} __randomize_layout;
655
656#define ND_ROOT_PRESET 1
657#define ND_ROOT_GRABBED 2
658#define ND_JUMPED 4
659
660static void __set_nameidata(struct nameidata *p, int dfd, struct filename *name)
661{
662 struct nameidata *old = current->nameidata;
663 p->stack = p->internal;
664 p->depth = 0;
665 p->dfd = dfd;
666 p->name = name;
667 p->pathname = likely(name) ? name->name : "";
668 p->path.mnt = NULL;
669 p->path.dentry = NULL;
670 p->total_link_count = old ? old->total_link_count : 0;
671 p->saved = old;
672 current->nameidata = p;
673}
674
675static inline void set_nameidata(struct nameidata *p, int dfd, struct filename *name,
676 const struct path *root)
677{
678 __set_nameidata(p, dfd, name);
679 p->state = 0;
680 if (unlikely(root)) {
681 p->state = ND_ROOT_PRESET;
682 p->root = *root;
683 }
684}
685
686static void restore_nameidata(void)
687{
688 struct nameidata *now = current->nameidata, *old = now->saved;
689
690 current->nameidata = old;
691 if (old)
692 old->total_link_count = now->total_link_count;
693 if (now->stack != now->internal)
694 kfree(now->stack);
695}
696
697static bool nd_alloc_stack(struct nameidata *nd)
698{
699 struct saved *p;
700
701 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
702 nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL);
703 if (unlikely(!p))
704 return false;
705 memcpy(p, nd->internal, sizeof(nd->internal));
706 nd->stack = p;
707 return true;
708}
709
710/**
711 * path_connected - Verify that a dentry is below mnt.mnt_root
712 * @mnt: The mountpoint to check.
713 * @dentry: The dentry to check.
714 *
715 * Rename can sometimes move a file or directory outside of a bind
716 * mount, path_connected allows those cases to be detected.
717 */
718static bool path_connected(struct vfsmount *mnt, struct dentry *dentry)
719{
720 struct super_block *sb = mnt->mnt_sb;
721
722 /* Bind mounts can have disconnected paths */
723 if (mnt->mnt_root == sb->s_root)
724 return true;
725
726 return is_subdir(dentry, mnt->mnt_root);
727}
728
729static void drop_links(struct nameidata *nd)
730{
731 int i = nd->depth;
732 while (i--) {
733 struct saved *last = nd->stack + i;
734 do_delayed_call(&last->done);
735 clear_delayed_call(&last->done);
736 }
737}
738
739static void leave_rcu(struct nameidata *nd)
740{
741 nd->flags &= ~LOOKUP_RCU;
742 nd->seq = nd->next_seq = 0;
743 rcu_read_unlock();
744}
745
746static void terminate_walk(struct nameidata *nd)
747{
748 drop_links(nd);
749 if (!(nd->flags & LOOKUP_RCU)) {
750 int i;
751 path_put(&nd->path);
752 for (i = 0; i < nd->depth; i++)
753 path_put(&nd->stack[i].link);
754 if (nd->state & ND_ROOT_GRABBED) {
755 path_put(&nd->root);
756 nd->state &= ~ND_ROOT_GRABBED;
757 }
758 } else {
759 leave_rcu(nd);
760 }
761 nd->depth = 0;
762 nd->path.mnt = NULL;
763 nd->path.dentry = NULL;
764}
765
766/* path_put is needed afterwards regardless of success or failure */
767static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq)
768{
769 int res = __legitimize_mnt(path->mnt, mseq);
770 if (unlikely(res)) {
771 if (res > 0)
772 path->mnt = NULL;
773 path->dentry = NULL;
774 return false;
775 }
776 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
777 path->dentry = NULL;
778 return false;
779 }
780 return !read_seqcount_retry(&path->dentry->d_seq, seq);
781}
782
783static inline bool legitimize_path(struct nameidata *nd,
784 struct path *path, unsigned seq)
785{
786 return __legitimize_path(path, seq, nd->m_seq);
787}
788
789static bool legitimize_links(struct nameidata *nd)
790{
791 int i;
792 if (unlikely(nd->flags & LOOKUP_CACHED)) {
793 drop_links(nd);
794 nd->depth = 0;
795 return false;
796 }
797 for (i = 0; i < nd->depth; i++) {
798 struct saved *last = nd->stack + i;
799 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
800 drop_links(nd);
801 nd->depth = i + 1;
802 return false;
803 }
804 }
805 return true;
806}
807
808static bool legitimize_root(struct nameidata *nd)
809{
810 /* Nothing to do if nd->root is zero or is managed by the VFS user. */
811 if (!nd->root.mnt || (nd->state & ND_ROOT_PRESET))
812 return true;
813 nd->state |= ND_ROOT_GRABBED;
814 return legitimize_path(nd, &nd->root, nd->root_seq);
815}
816
817/*
818 * Path walking has 2 modes, rcu-walk and ref-walk (see
819 * Documentation/filesystems/path-lookup.txt). In situations when we can't
820 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
821 * normal reference counts on dentries and vfsmounts to transition to ref-walk
822 * mode. Refcounts are grabbed at the last known good point before rcu-walk
823 * got stuck, so ref-walk may continue from there. If this is not successful
824 * (eg. a seqcount has changed), then failure is returned and it's up to caller
825 * to restart the path walk from the beginning in ref-walk mode.
826 */
827
828/**
829 * try_to_unlazy - try to switch to ref-walk mode.
830 * @nd: nameidata pathwalk data
831 * Returns: true on success, false on failure
832 *
833 * try_to_unlazy attempts to legitimize the current nd->path and nd->root
834 * for ref-walk mode.
835 * Must be called from rcu-walk context.
836 * Nothing should touch nameidata between try_to_unlazy() failure and
837 * terminate_walk().
838 */
839static bool try_to_unlazy(struct nameidata *nd)
840{
841 struct dentry *parent = nd->path.dentry;
842
843 BUG_ON(!(nd->flags & LOOKUP_RCU));
844
845 if (unlikely(!legitimize_links(nd)))
846 goto out1;
847 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
848 goto out;
849 if (unlikely(!legitimize_root(nd)))
850 goto out;
851 leave_rcu(nd);
852 BUG_ON(nd->inode != parent->d_inode);
853 return true;
854
855out1:
856 nd->path.mnt = NULL;
857 nd->path.dentry = NULL;
858out:
859 leave_rcu(nd);
860 return false;
861}
862
863/**
864 * try_to_unlazy_next - try to switch to ref-walk mode.
865 * @nd: nameidata pathwalk data
866 * @dentry: next dentry to step into
867 * Returns: true on success, false on failure
868 *
869 * Similar to try_to_unlazy(), but here we have the next dentry already
870 * picked by rcu-walk and want to legitimize that in addition to the current
871 * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context.
872 * Nothing should touch nameidata between try_to_unlazy_next() failure and
873 * terminate_walk().
874 */
875static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry)
876{
877 int res;
878 BUG_ON(!(nd->flags & LOOKUP_RCU));
879
880 if (unlikely(!legitimize_links(nd)))
881 goto out2;
882 res = __legitimize_mnt(nd->path.mnt, nd->m_seq);
883 if (unlikely(res)) {
884 if (res > 0)
885 goto out2;
886 goto out1;
887 }
888 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
889 goto out1;
890
891 /*
892 * We need to move both the parent and the dentry from the RCU domain
893 * to be properly refcounted. And the sequence number in the dentry
894 * validates *both* dentry counters, since we checked the sequence
895 * number of the parent after we got the child sequence number. So we
896 * know the parent must still be valid if the child sequence number is
897 */
898 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
899 goto out;
900 if (read_seqcount_retry(&dentry->d_seq, nd->next_seq))
901 goto out_dput;
902 /*
903 * Sequence counts matched. Now make sure that the root is
904 * still valid and get it if required.
905 */
906 if (unlikely(!legitimize_root(nd)))
907 goto out_dput;
908 leave_rcu(nd);
909 return true;
910
911out2:
912 nd->path.mnt = NULL;
913out1:
914 nd->path.dentry = NULL;
915out:
916 leave_rcu(nd);
917 return false;
918out_dput:
919 leave_rcu(nd);
920 dput(dentry);
921 return false;
922}
923
924static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
925{
926 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
927 return dentry->d_op->d_revalidate(dentry, flags);
928 else
929 return 1;
930}
931
932/**
933 * complete_walk - successful completion of path walk
934 * @nd: pointer nameidata
935 *
936 * If we had been in RCU mode, drop out of it and legitimize nd->path.
937 * Revalidate the final result, unless we'd already done that during
938 * the path walk or the filesystem doesn't ask for it. Return 0 on
939 * success, -error on failure. In case of failure caller does not
940 * need to drop nd->path.
941 */
942static int complete_walk(struct nameidata *nd)
943{
944 struct dentry *dentry = nd->path.dentry;
945 int status;
946
947 if (nd->flags & LOOKUP_RCU) {
948 /*
949 * We don't want to zero nd->root for scoped-lookups or
950 * externally-managed nd->root.
951 */
952 if (!(nd->state & ND_ROOT_PRESET))
953 if (!(nd->flags & LOOKUP_IS_SCOPED))
954 nd->root.mnt = NULL;
955 nd->flags &= ~LOOKUP_CACHED;
956 if (!try_to_unlazy(nd))
957 return -ECHILD;
958 }
959
960 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
961 /*
962 * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't
963 * ever step outside the root during lookup" and should already
964 * be guaranteed by the rest of namei, we want to avoid a namei
965 * BUG resulting in userspace being given a path that was not
966 * scoped within the root at some point during the lookup.
967 *
968 * So, do a final sanity-check to make sure that in the
969 * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED)
970 * we won't silently return an fd completely outside of the
971 * requested root to userspace.
972 *
973 * Userspace could move the path outside the root after this
974 * check, but as discussed elsewhere this is not a concern (the
975 * resolved file was inside the root at some point).
976 */
977 if (!path_is_under(&nd->path, &nd->root))
978 return -EXDEV;
979 }
980
981 if (likely(!(nd->state & ND_JUMPED)))
982 return 0;
983
984 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
985 return 0;
986
987 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
988 if (status > 0)
989 return 0;
990
991 if (!status)
992 status = -ESTALE;
993
994 return status;
995}
996
997static int set_root(struct nameidata *nd)
998{
999 struct fs_struct *fs = current->fs;
1000
1001 /*
1002 * Jumping to the real root in a scoped-lookup is a BUG in namei, but we
1003 * still have to ensure it doesn't happen because it will cause a breakout
1004 * from the dirfd.
1005 */
1006 if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED))
1007 return -ENOTRECOVERABLE;
1008
1009 if (nd->flags & LOOKUP_RCU) {
1010 unsigned seq;
1011
1012 do {
1013 seq = read_seqcount_begin(&fs->seq);
1014 nd->root = fs->root;
1015 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
1016 } while (read_seqcount_retry(&fs->seq, seq));
1017 } else {
1018 get_fs_root(fs, &nd->root);
1019 nd->state |= ND_ROOT_GRABBED;
1020 }
1021 return 0;
1022}
1023
1024static int nd_jump_root(struct nameidata *nd)
1025{
1026 if (unlikely(nd->flags & LOOKUP_BENEATH))
1027 return -EXDEV;
1028 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
1029 /* Absolute path arguments to path_init() are allowed. */
1030 if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt)
1031 return -EXDEV;
1032 }
1033 if (!nd->root.mnt) {
1034 int error = set_root(nd);
1035 if (error)
1036 return error;
1037 }
1038 if (nd->flags & LOOKUP_RCU) {
1039 struct dentry *d;
1040 nd->path = nd->root;
1041 d = nd->path.dentry;
1042 nd->inode = d->d_inode;
1043 nd->seq = nd->root_seq;
1044 if (read_seqcount_retry(&d->d_seq, nd->seq))
1045 return -ECHILD;
1046 } else {
1047 path_put(&nd->path);
1048 nd->path = nd->root;
1049 path_get(&nd->path);
1050 nd->inode = nd->path.dentry->d_inode;
1051 }
1052 nd->state |= ND_JUMPED;
1053 return 0;
1054}
1055
1056/*
1057 * Helper to directly jump to a known parsed path from ->get_link,
1058 * caller must have taken a reference to path beforehand.
1059 */
1060int nd_jump_link(const struct path *path)
1061{
1062 int error = -ELOOP;
1063 struct nameidata *nd = current->nameidata;
1064
1065 if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS))
1066 goto err;
1067
1068 error = -EXDEV;
1069 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
1070 if (nd->path.mnt != path->mnt)
1071 goto err;
1072 }
1073 /* Not currently safe for scoped-lookups. */
1074 if (unlikely(nd->flags & LOOKUP_IS_SCOPED))
1075 goto err;
1076
1077 path_put(&nd->path);
1078 nd->path = *path;
1079 nd->inode = nd->path.dentry->d_inode;
1080 nd->state |= ND_JUMPED;
1081 return 0;
1082
1083err:
1084 path_put(path);
1085 return error;
1086}
1087
1088static inline void put_link(struct nameidata *nd)
1089{
1090 struct saved *last = nd->stack + --nd->depth;
1091 do_delayed_call(&last->done);
1092 if (!(nd->flags & LOOKUP_RCU))
1093 path_put(&last->link);
1094}
1095
1096static int sysctl_protected_symlinks __read_mostly;
1097static int sysctl_protected_hardlinks __read_mostly;
1098static int sysctl_protected_fifos __read_mostly;
1099static int sysctl_protected_regular __read_mostly;
1100
1101#ifdef CONFIG_SYSCTL
1102static struct ctl_table namei_sysctls[] = {
1103 {
1104 .procname = "protected_symlinks",
1105 .data = &sysctl_protected_symlinks,
1106 .maxlen = sizeof(int),
1107 .mode = 0644,
1108 .proc_handler = proc_dointvec_minmax,
1109 .extra1 = SYSCTL_ZERO,
1110 .extra2 = SYSCTL_ONE,
1111 },
1112 {
1113 .procname = "protected_hardlinks",
1114 .data = &sysctl_protected_hardlinks,
1115 .maxlen = sizeof(int),
1116 .mode = 0644,
1117 .proc_handler = proc_dointvec_minmax,
1118 .extra1 = SYSCTL_ZERO,
1119 .extra2 = SYSCTL_ONE,
1120 },
1121 {
1122 .procname = "protected_fifos",
1123 .data = &sysctl_protected_fifos,
1124 .maxlen = sizeof(int),
1125 .mode = 0644,
1126 .proc_handler = proc_dointvec_minmax,
1127 .extra1 = SYSCTL_ZERO,
1128 .extra2 = SYSCTL_TWO,
1129 },
1130 {
1131 .procname = "protected_regular",
1132 .data = &sysctl_protected_regular,
1133 .maxlen = sizeof(int),
1134 .mode = 0644,
1135 .proc_handler = proc_dointvec_minmax,
1136 .extra1 = SYSCTL_ZERO,
1137 .extra2 = SYSCTL_TWO,
1138 },
1139};
1140
1141static int __init init_fs_namei_sysctls(void)
1142{
1143 register_sysctl_init("fs", namei_sysctls);
1144 return 0;
1145}
1146fs_initcall(init_fs_namei_sysctls);
1147
1148#endif /* CONFIG_SYSCTL */
1149
1150/**
1151 * may_follow_link - Check symlink following for unsafe situations
1152 * @nd: nameidata pathwalk data
1153 * @inode: Used for idmapping.
1154 *
1155 * In the case of the sysctl_protected_symlinks sysctl being enabled,
1156 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
1157 * in a sticky world-writable directory. This is to protect privileged
1158 * processes from failing races against path names that may change out
1159 * from under them by way of other users creating malicious symlinks.
1160 * It will permit symlinks to be followed only when outside a sticky
1161 * world-writable directory, or when the uid of the symlink and follower
1162 * match, or when the directory owner matches the symlink's owner.
1163 *
1164 * Returns 0 if following the symlink is allowed, -ve on error.
1165 */
1166static inline int may_follow_link(struct nameidata *nd, const struct inode *inode)
1167{
1168 struct mnt_idmap *idmap;
1169 vfsuid_t vfsuid;
1170
1171 if (!sysctl_protected_symlinks)
1172 return 0;
1173
1174 idmap = mnt_idmap(nd->path.mnt);
1175 vfsuid = i_uid_into_vfsuid(idmap, inode);
1176 /* Allowed if owner and follower match. */
1177 if (vfsuid_eq_kuid(vfsuid, current_fsuid()))
1178 return 0;
1179
1180 /* Allowed if parent directory not sticky and world-writable. */
1181 if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
1182 return 0;
1183
1184 /* Allowed if parent directory and link owner match. */
1185 if (vfsuid_valid(nd->dir_vfsuid) && vfsuid_eq(nd->dir_vfsuid, vfsuid))
1186 return 0;
1187
1188 if (nd->flags & LOOKUP_RCU)
1189 return -ECHILD;
1190
1191 audit_inode(nd->name, nd->stack[0].link.dentry, 0);
1192 audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link");
1193 return -EACCES;
1194}
1195
1196/**
1197 * safe_hardlink_source - Check for safe hardlink conditions
1198 * @idmap: idmap of the mount the inode was found from
1199 * @inode: the source inode to hardlink from
1200 *
1201 * Return false if at least one of the following conditions:
1202 * - inode is not a regular file
1203 * - inode is setuid
1204 * - inode is setgid and group-exec
1205 * - access failure for read and write
1206 *
1207 * Otherwise returns true.
1208 */
1209static bool safe_hardlink_source(struct mnt_idmap *idmap,
1210 struct inode *inode)
1211{
1212 umode_t mode = inode->i_mode;
1213
1214 /* Special files should not get pinned to the filesystem. */
1215 if (!S_ISREG(mode))
1216 return false;
1217
1218 /* Setuid files should not get pinned to the filesystem. */
1219 if (mode & S_ISUID)
1220 return false;
1221
1222 /* Executable setgid files should not get pinned to the filesystem. */
1223 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
1224 return false;
1225
1226 /* Hardlinking to unreadable or unwritable sources is dangerous. */
1227 if (inode_permission(idmap, inode, MAY_READ | MAY_WRITE))
1228 return false;
1229
1230 return true;
1231}
1232
1233/**
1234 * may_linkat - Check permissions for creating a hardlink
1235 * @idmap: idmap of the mount the inode was found from
1236 * @link: the source to hardlink from
1237 *
1238 * Block hardlink when all of:
1239 * - sysctl_protected_hardlinks enabled
1240 * - fsuid does not match inode
1241 * - hardlink source is unsafe (see safe_hardlink_source() above)
1242 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
1243 *
1244 * If the inode has been found through an idmapped mount the idmap of
1245 * the vfsmount must be passed through @idmap. This function will then take
1246 * care to map the inode according to @idmap before checking permissions.
1247 * On non-idmapped mounts or if permission checking is to be performed on the
1248 * raw inode simply pass @nop_mnt_idmap.
1249 *
1250 * Returns 0 if successful, -ve on error.
1251 */
1252int may_linkat(struct mnt_idmap *idmap, const struct path *link)
1253{
1254 struct inode *inode = link->dentry->d_inode;
1255
1256 /* Inode writeback is not safe when the uid or gid are invalid. */
1257 if (!vfsuid_valid(i_uid_into_vfsuid(idmap, inode)) ||
1258 !vfsgid_valid(i_gid_into_vfsgid(idmap, inode)))
1259 return -EOVERFLOW;
1260
1261 if (!sysctl_protected_hardlinks)
1262 return 0;
1263
1264 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1265 * otherwise, it must be a safe source.
1266 */
1267 if (safe_hardlink_source(idmap, inode) ||
1268 inode_owner_or_capable(idmap, inode))
1269 return 0;
1270
1271 audit_log_path_denied(AUDIT_ANOM_LINK, "linkat");
1272 return -EPERM;
1273}
1274
1275/**
1276 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1277 * should be allowed, or not, on files that already
1278 * exist.
1279 * @idmap: idmap of the mount the inode was found from
1280 * @nd: nameidata pathwalk data
1281 * @inode: the inode of the file to open
1282 *
1283 * Block an O_CREAT open of a FIFO (or a regular file) when:
1284 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1285 * - the file already exists
1286 * - we are in a sticky directory
1287 * - we don't own the file
1288 * - the owner of the directory doesn't own the file
1289 * - the directory is world writable
1290 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1291 * the directory doesn't have to be world writable: being group writable will
1292 * be enough.
1293 *
1294 * If the inode has been found through an idmapped mount the idmap of
1295 * the vfsmount must be passed through @idmap. This function will then take
1296 * care to map the inode according to @idmap before checking permissions.
1297 * On non-idmapped mounts or if permission checking is to be performed on the
1298 * raw inode simply pass @nop_mnt_idmap.
1299 *
1300 * Returns 0 if the open is allowed, -ve on error.
1301 */
1302static int may_create_in_sticky(struct mnt_idmap *idmap, struct nameidata *nd,
1303 struct inode *const inode)
1304{
1305 umode_t dir_mode = nd->dir_mode;
1306 vfsuid_t dir_vfsuid = nd->dir_vfsuid, i_vfsuid;
1307
1308 if (likely(!(dir_mode & S_ISVTX)))
1309 return 0;
1310
1311 if (S_ISREG(inode->i_mode) && !sysctl_protected_regular)
1312 return 0;
1313
1314 if (S_ISFIFO(inode->i_mode) && !sysctl_protected_fifos)
1315 return 0;
1316
1317 i_vfsuid = i_uid_into_vfsuid(idmap, inode);
1318
1319 if (vfsuid_eq(i_vfsuid, dir_vfsuid))
1320 return 0;
1321
1322 if (vfsuid_eq_kuid(i_vfsuid, current_fsuid()))
1323 return 0;
1324
1325 if (likely(dir_mode & 0002)) {
1326 audit_log_path_denied(AUDIT_ANOM_CREAT, "sticky_create");
1327 return -EACCES;
1328 }
1329
1330 if (dir_mode & 0020) {
1331 if (sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) {
1332 audit_log_path_denied(AUDIT_ANOM_CREAT,
1333 "sticky_create_fifo");
1334 return -EACCES;
1335 }
1336
1337 if (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode)) {
1338 audit_log_path_denied(AUDIT_ANOM_CREAT,
1339 "sticky_create_regular");
1340 return -EACCES;
1341 }
1342 }
1343
1344 return 0;
1345}
1346
1347/*
1348 * follow_up - Find the mountpoint of path's vfsmount
1349 *
1350 * Given a path, find the mountpoint of its source file system.
1351 * Replace @path with the path of the mountpoint in the parent mount.
1352 * Up is towards /.
1353 *
1354 * Return 1 if we went up a level and 0 if we were already at the
1355 * root.
1356 */
1357int follow_up(struct path *path)
1358{
1359 struct mount *mnt = real_mount(path->mnt);
1360 struct mount *parent;
1361 struct dentry *mountpoint;
1362
1363 read_seqlock_excl(&mount_lock);
1364 parent = mnt->mnt_parent;
1365 if (parent == mnt) {
1366 read_sequnlock_excl(&mount_lock);
1367 return 0;
1368 }
1369 mntget(&parent->mnt);
1370 mountpoint = dget(mnt->mnt_mountpoint);
1371 read_sequnlock_excl(&mount_lock);
1372 dput(path->dentry);
1373 path->dentry = mountpoint;
1374 mntput(path->mnt);
1375 path->mnt = &parent->mnt;
1376 return 1;
1377}
1378EXPORT_SYMBOL(follow_up);
1379
1380static bool choose_mountpoint_rcu(struct mount *m, const struct path *root,
1381 struct path *path, unsigned *seqp)
1382{
1383 while (mnt_has_parent(m)) {
1384 struct dentry *mountpoint = m->mnt_mountpoint;
1385
1386 m = m->mnt_parent;
1387 if (unlikely(root->dentry == mountpoint &&
1388 root->mnt == &m->mnt))
1389 break;
1390 if (mountpoint != m->mnt.mnt_root) {
1391 path->mnt = &m->mnt;
1392 path->dentry = mountpoint;
1393 *seqp = read_seqcount_begin(&mountpoint->d_seq);
1394 return true;
1395 }
1396 }
1397 return false;
1398}
1399
1400static bool choose_mountpoint(struct mount *m, const struct path *root,
1401 struct path *path)
1402{
1403 bool found;
1404
1405 rcu_read_lock();
1406 while (1) {
1407 unsigned seq, mseq = read_seqbegin(&mount_lock);
1408
1409 found = choose_mountpoint_rcu(m, root, path, &seq);
1410 if (unlikely(!found)) {
1411 if (!read_seqretry(&mount_lock, mseq))
1412 break;
1413 } else {
1414 if (likely(__legitimize_path(path, seq, mseq)))
1415 break;
1416 rcu_read_unlock();
1417 path_put(path);
1418 rcu_read_lock();
1419 }
1420 }
1421 rcu_read_unlock();
1422 return found;
1423}
1424
1425/*
1426 * Perform an automount
1427 * - return -EISDIR to tell follow_managed() to stop and return the path we
1428 * were called with.
1429 */
1430static int follow_automount(struct path *path, int *count, unsigned lookup_flags)
1431{
1432 struct dentry *dentry = path->dentry;
1433
1434 /* We don't want to mount if someone's just doing a stat -
1435 * unless they're stat'ing a directory and appended a '/' to
1436 * the name.
1437 *
1438 * We do, however, want to mount if someone wants to open or
1439 * create a file of any type under the mountpoint, wants to
1440 * traverse through the mountpoint or wants to open the
1441 * mounted directory. Also, autofs may mark negative dentries
1442 * as being automount points. These will need the attentions
1443 * of the daemon to instantiate them before they can be used.
1444 */
1445 if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1446 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1447 dentry->d_inode)
1448 return -EISDIR;
1449
1450 if (count && (*count)++ >= MAXSYMLINKS)
1451 return -ELOOP;
1452
1453 return finish_automount(dentry->d_op->d_automount(path), path);
1454}
1455
1456/*
1457 * mount traversal - out-of-line part. One note on ->d_flags accesses -
1458 * dentries are pinned but not locked here, so negative dentry can go
1459 * positive right under us. Use of smp_load_acquire() provides a barrier
1460 * sufficient for ->d_inode and ->d_flags consistency.
1461 */
1462static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped,
1463 int *count, unsigned lookup_flags)
1464{
1465 struct vfsmount *mnt = path->mnt;
1466 bool need_mntput = false;
1467 int ret = 0;
1468
1469 while (flags & DCACHE_MANAGED_DENTRY) {
1470 /* Allow the filesystem to manage the transit without i_mutex
1471 * being held. */
1472 if (flags & DCACHE_MANAGE_TRANSIT) {
1473 ret = path->dentry->d_op->d_manage(path, false);
1474 flags = smp_load_acquire(&path->dentry->d_flags);
1475 if (ret < 0)
1476 break;
1477 }
1478
1479 if (flags & DCACHE_MOUNTED) { // something's mounted on it..
1480 struct vfsmount *mounted = lookup_mnt(path);
1481 if (mounted) { // ... in our namespace
1482 dput(path->dentry);
1483 if (need_mntput)
1484 mntput(path->mnt);
1485 path->mnt = mounted;
1486 path->dentry = dget(mounted->mnt_root);
1487 // here we know it's positive
1488 flags = path->dentry->d_flags;
1489 need_mntput = true;
1490 continue;
1491 }
1492 }
1493
1494 if (!(flags & DCACHE_NEED_AUTOMOUNT))
1495 break;
1496
1497 // uncovered automount point
1498 ret = follow_automount(path, count, lookup_flags);
1499 flags = smp_load_acquire(&path->dentry->d_flags);
1500 if (ret < 0)
1501 break;
1502 }
1503
1504 if (ret == -EISDIR)
1505 ret = 0;
1506 // possible if you race with several mount --move
1507 if (need_mntput && path->mnt == mnt)
1508 mntput(path->mnt);
1509 if (!ret && unlikely(d_flags_negative(flags)))
1510 ret = -ENOENT;
1511 *jumped = need_mntput;
1512 return ret;
1513}
1514
1515static inline int traverse_mounts(struct path *path, bool *jumped,
1516 int *count, unsigned lookup_flags)
1517{
1518 unsigned flags = smp_load_acquire(&path->dentry->d_flags);
1519
1520 /* fastpath */
1521 if (likely(!(flags & DCACHE_MANAGED_DENTRY))) {
1522 *jumped = false;
1523 if (unlikely(d_flags_negative(flags)))
1524 return -ENOENT;
1525 return 0;
1526 }
1527 return __traverse_mounts(path, flags, jumped, count, lookup_flags);
1528}
1529
1530int follow_down_one(struct path *path)
1531{
1532 struct vfsmount *mounted;
1533
1534 mounted = lookup_mnt(path);
1535 if (mounted) {
1536 dput(path->dentry);
1537 mntput(path->mnt);
1538 path->mnt = mounted;
1539 path->dentry = dget(mounted->mnt_root);
1540 return 1;
1541 }
1542 return 0;
1543}
1544EXPORT_SYMBOL(follow_down_one);
1545
1546/*
1547 * Follow down to the covering mount currently visible to userspace. At each
1548 * point, the filesystem owning that dentry may be queried as to whether the
1549 * caller is permitted to proceed or not.
1550 */
1551int follow_down(struct path *path, unsigned int flags)
1552{
1553 struct vfsmount *mnt = path->mnt;
1554 bool jumped;
1555 int ret = traverse_mounts(path, &jumped, NULL, flags);
1556
1557 if (path->mnt != mnt)
1558 mntput(mnt);
1559 return ret;
1560}
1561EXPORT_SYMBOL(follow_down);
1562
1563/*
1564 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1565 * we meet a managed dentry that would need blocking.
1566 */
1567static bool __follow_mount_rcu(struct nameidata *nd, struct path *path)
1568{
1569 struct dentry *dentry = path->dentry;
1570 unsigned int flags = dentry->d_flags;
1571
1572 if (likely(!(flags & DCACHE_MANAGED_DENTRY)))
1573 return true;
1574
1575 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1576 return false;
1577
1578 for (;;) {
1579 /*
1580 * Don't forget we might have a non-mountpoint managed dentry
1581 * that wants to block transit.
1582 */
1583 if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) {
1584 int res = dentry->d_op->d_manage(path, true);
1585 if (res)
1586 return res == -EISDIR;
1587 flags = dentry->d_flags;
1588 }
1589
1590 if (flags & DCACHE_MOUNTED) {
1591 struct mount *mounted = __lookup_mnt(path->mnt, dentry);
1592 if (mounted) {
1593 path->mnt = &mounted->mnt;
1594 dentry = path->dentry = mounted->mnt.mnt_root;
1595 nd->state |= ND_JUMPED;
1596 nd->next_seq = read_seqcount_begin(&dentry->d_seq);
1597 flags = dentry->d_flags;
1598 // makes sure that non-RCU pathwalk could reach
1599 // this state.
1600 if (read_seqretry(&mount_lock, nd->m_seq))
1601 return false;
1602 continue;
1603 }
1604 if (read_seqretry(&mount_lock, nd->m_seq))
1605 return false;
1606 }
1607 return !(flags & DCACHE_NEED_AUTOMOUNT);
1608 }
1609}
1610
1611static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry,
1612 struct path *path)
1613{
1614 bool jumped;
1615 int ret;
1616
1617 path->mnt = nd->path.mnt;
1618 path->dentry = dentry;
1619 if (nd->flags & LOOKUP_RCU) {
1620 unsigned int seq = nd->next_seq;
1621 if (likely(__follow_mount_rcu(nd, path)))
1622 return 0;
1623 // *path and nd->next_seq might've been clobbered
1624 path->mnt = nd->path.mnt;
1625 path->dentry = dentry;
1626 nd->next_seq = seq;
1627 if (!try_to_unlazy_next(nd, dentry))
1628 return -ECHILD;
1629 }
1630 ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags);
1631 if (jumped) {
1632 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1633 ret = -EXDEV;
1634 else
1635 nd->state |= ND_JUMPED;
1636 }
1637 if (unlikely(ret)) {
1638 dput(path->dentry);
1639 if (path->mnt != nd->path.mnt)
1640 mntput(path->mnt);
1641 }
1642 return ret;
1643}
1644
1645/*
1646 * This looks up the name in dcache and possibly revalidates the found dentry.
1647 * NULL is returned if the dentry does not exist in the cache.
1648 */
1649static struct dentry *lookup_dcache(const struct qstr *name,
1650 struct dentry *dir,
1651 unsigned int flags)
1652{
1653 struct dentry *dentry = d_lookup(dir, name);
1654 if (dentry) {
1655 int error = d_revalidate(dentry, flags);
1656 if (unlikely(error <= 0)) {
1657 if (!error)
1658 d_invalidate(dentry);
1659 dput(dentry);
1660 return ERR_PTR(error);
1661 }
1662 }
1663 return dentry;
1664}
1665
1666/*
1667 * Parent directory has inode locked exclusive. This is one
1668 * and only case when ->lookup() gets called on non in-lookup
1669 * dentries - as the matter of fact, this only gets called
1670 * when directory is guaranteed to have no in-lookup children
1671 * at all.
1672 */
1673struct dentry *lookup_one_qstr_excl(const struct qstr *name,
1674 struct dentry *base,
1675 unsigned int flags)
1676{
1677 struct dentry *dentry = lookup_dcache(name, base, flags);
1678 struct dentry *old;
1679 struct inode *dir = base->d_inode;
1680
1681 if (dentry)
1682 return dentry;
1683
1684 /* Don't create child dentry for a dead directory. */
1685 if (unlikely(IS_DEADDIR(dir)))
1686 return ERR_PTR(-ENOENT);
1687
1688 dentry = d_alloc(base, name);
1689 if (unlikely(!dentry))
1690 return ERR_PTR(-ENOMEM);
1691
1692 old = dir->i_op->lookup(dir, dentry, flags);
1693 if (unlikely(old)) {
1694 dput(dentry);
1695 dentry = old;
1696 }
1697 return dentry;
1698}
1699EXPORT_SYMBOL(lookup_one_qstr_excl);
1700
1701/**
1702 * lookup_fast - do fast lockless (but racy) lookup of a dentry
1703 * @nd: current nameidata
1704 *
1705 * Do a fast, but racy lookup in the dcache for the given dentry, and
1706 * revalidate it. Returns a valid dentry pointer or NULL if one wasn't
1707 * found. On error, an ERR_PTR will be returned.
1708 *
1709 * If this function returns a valid dentry and the walk is no longer
1710 * lazy, the dentry will carry a reference that must later be put. If
1711 * RCU mode is still in force, then this is not the case and the dentry
1712 * must be legitimized before use. If this returns NULL, then the walk
1713 * will no longer be in RCU mode.
1714 */
1715static struct dentry *lookup_fast(struct nameidata *nd)
1716{
1717 struct dentry *dentry, *parent = nd->path.dentry;
1718 int status = 1;
1719
1720 /*
1721 * Rename seqlock is not required here because in the off chance
1722 * of a false negative due to a concurrent rename, the caller is
1723 * going to fall back to non-racy lookup.
1724 */
1725 if (nd->flags & LOOKUP_RCU) {
1726 dentry = __d_lookup_rcu(parent, &nd->last, &nd->next_seq);
1727 if (unlikely(!dentry)) {
1728 if (!try_to_unlazy(nd))
1729 return ERR_PTR(-ECHILD);
1730 return NULL;
1731 }
1732
1733 /*
1734 * This sequence count validates that the parent had no
1735 * changes while we did the lookup of the dentry above.
1736 */
1737 if (read_seqcount_retry(&parent->d_seq, nd->seq))
1738 return ERR_PTR(-ECHILD);
1739
1740 status = d_revalidate(dentry, nd->flags);
1741 if (likely(status > 0))
1742 return dentry;
1743 if (!try_to_unlazy_next(nd, dentry))
1744 return ERR_PTR(-ECHILD);
1745 if (status == -ECHILD)
1746 /* we'd been told to redo it in non-rcu mode */
1747 status = d_revalidate(dentry, nd->flags);
1748 } else {
1749 dentry = __d_lookup(parent, &nd->last);
1750 if (unlikely(!dentry))
1751 return NULL;
1752 status = d_revalidate(dentry, nd->flags);
1753 }
1754 if (unlikely(status <= 0)) {
1755 if (!status)
1756 d_invalidate(dentry);
1757 dput(dentry);
1758 return ERR_PTR(status);
1759 }
1760 return dentry;
1761}
1762
1763/* Fast lookup failed, do it the slow way */
1764static struct dentry *__lookup_slow(const struct qstr *name,
1765 struct dentry *dir,
1766 unsigned int flags)
1767{
1768 struct dentry *dentry, *old;
1769 struct inode *inode = dir->d_inode;
1770 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1771
1772 /* Don't go there if it's already dead */
1773 if (unlikely(IS_DEADDIR(inode)))
1774 return ERR_PTR(-ENOENT);
1775again:
1776 dentry = d_alloc_parallel(dir, name, &wq);
1777 if (IS_ERR(dentry))
1778 return dentry;
1779 if (unlikely(!d_in_lookup(dentry))) {
1780 int error = d_revalidate(dentry, flags);
1781 if (unlikely(error <= 0)) {
1782 if (!error) {
1783 d_invalidate(dentry);
1784 dput(dentry);
1785 goto again;
1786 }
1787 dput(dentry);
1788 dentry = ERR_PTR(error);
1789 }
1790 } else {
1791 old = inode->i_op->lookup(inode, dentry, flags);
1792 d_lookup_done(dentry);
1793 if (unlikely(old)) {
1794 dput(dentry);
1795 dentry = old;
1796 }
1797 }
1798 return dentry;
1799}
1800
1801static struct dentry *lookup_slow(const struct qstr *name,
1802 struct dentry *dir,
1803 unsigned int flags)
1804{
1805 struct inode *inode = dir->d_inode;
1806 struct dentry *res;
1807 inode_lock_shared(inode);
1808 res = __lookup_slow(name, dir, flags);
1809 inode_unlock_shared(inode);
1810 return res;
1811}
1812
1813static inline int may_lookup(struct mnt_idmap *idmap,
1814 struct nameidata *restrict nd)
1815{
1816 int err, mask;
1817
1818 mask = nd->flags & LOOKUP_RCU ? MAY_NOT_BLOCK : 0;
1819 err = inode_permission(idmap, nd->inode, mask | MAY_EXEC);
1820 if (likely(!err))
1821 return 0;
1822
1823 // If we failed, and we weren't in LOOKUP_RCU, it's final
1824 if (!(nd->flags & LOOKUP_RCU))
1825 return err;
1826
1827 // Drop out of RCU mode to make sure it wasn't transient
1828 if (!try_to_unlazy(nd))
1829 return -ECHILD; // redo it all non-lazy
1830
1831 if (err != -ECHILD) // hard error
1832 return err;
1833
1834 return inode_permission(idmap, nd->inode, MAY_EXEC);
1835}
1836
1837static int reserve_stack(struct nameidata *nd, struct path *link)
1838{
1839 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS))
1840 return -ELOOP;
1841
1842 if (likely(nd->depth != EMBEDDED_LEVELS))
1843 return 0;
1844 if (likely(nd->stack != nd->internal))
1845 return 0;
1846 if (likely(nd_alloc_stack(nd)))
1847 return 0;
1848
1849 if (nd->flags & LOOKUP_RCU) {
1850 // we need to grab link before we do unlazy. And we can't skip
1851 // unlazy even if we fail to grab the link - cleanup needs it
1852 bool grabbed_link = legitimize_path(nd, link, nd->next_seq);
1853
1854 if (!try_to_unlazy(nd) || !grabbed_link)
1855 return -ECHILD;
1856
1857 if (nd_alloc_stack(nd))
1858 return 0;
1859 }
1860 return -ENOMEM;
1861}
1862
1863enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4};
1864
1865static const char *pick_link(struct nameidata *nd, struct path *link,
1866 struct inode *inode, int flags)
1867{
1868 struct saved *last;
1869 const char *res;
1870 int error = reserve_stack(nd, link);
1871
1872 if (unlikely(error)) {
1873 if (!(nd->flags & LOOKUP_RCU))
1874 path_put(link);
1875 return ERR_PTR(error);
1876 }
1877 last = nd->stack + nd->depth++;
1878 last->link = *link;
1879 clear_delayed_call(&last->done);
1880 last->seq = nd->next_seq;
1881
1882 if (flags & WALK_TRAILING) {
1883 error = may_follow_link(nd, inode);
1884 if (unlikely(error))
1885 return ERR_PTR(error);
1886 }
1887
1888 if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) ||
1889 unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW))
1890 return ERR_PTR(-ELOOP);
1891
1892 if (!(nd->flags & LOOKUP_RCU)) {
1893 touch_atime(&last->link);
1894 cond_resched();
1895 } else if (atime_needs_update(&last->link, inode)) {
1896 if (!try_to_unlazy(nd))
1897 return ERR_PTR(-ECHILD);
1898 touch_atime(&last->link);
1899 }
1900
1901 error = security_inode_follow_link(link->dentry, inode,
1902 nd->flags & LOOKUP_RCU);
1903 if (unlikely(error))
1904 return ERR_PTR(error);
1905
1906 res = READ_ONCE(inode->i_link);
1907 if (!res) {
1908 const char * (*get)(struct dentry *, struct inode *,
1909 struct delayed_call *);
1910 get = inode->i_op->get_link;
1911 if (nd->flags & LOOKUP_RCU) {
1912 res = get(NULL, inode, &last->done);
1913 if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd))
1914 res = get(link->dentry, inode, &last->done);
1915 } else {
1916 res = get(link->dentry, inode, &last->done);
1917 }
1918 if (!res)
1919 goto all_done;
1920 if (IS_ERR(res))
1921 return res;
1922 }
1923 if (*res == '/') {
1924 error = nd_jump_root(nd);
1925 if (unlikely(error))
1926 return ERR_PTR(error);
1927 while (unlikely(*++res == '/'))
1928 ;
1929 }
1930 if (*res)
1931 return res;
1932all_done: // pure jump
1933 put_link(nd);
1934 return NULL;
1935}
1936
1937/*
1938 * Do we need to follow links? We _really_ want to be able
1939 * to do this check without having to look at inode->i_op,
1940 * so we keep a cache of "no, this doesn't need follow_link"
1941 * for the common case.
1942 *
1943 * NOTE: dentry must be what nd->next_seq had been sampled from.
1944 */
1945static const char *step_into(struct nameidata *nd, int flags,
1946 struct dentry *dentry)
1947{
1948 struct path path;
1949 struct inode *inode;
1950 int err = handle_mounts(nd, dentry, &path);
1951
1952 if (err < 0)
1953 return ERR_PTR(err);
1954 inode = path.dentry->d_inode;
1955 if (likely(!d_is_symlink(path.dentry)) ||
1956 ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) ||
1957 (flags & WALK_NOFOLLOW)) {
1958 /* not a symlink or should not follow */
1959 if (nd->flags & LOOKUP_RCU) {
1960 if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq))
1961 return ERR_PTR(-ECHILD);
1962 if (unlikely(!inode))
1963 return ERR_PTR(-ENOENT);
1964 } else {
1965 dput(nd->path.dentry);
1966 if (nd->path.mnt != path.mnt)
1967 mntput(nd->path.mnt);
1968 }
1969 nd->path = path;
1970 nd->inode = inode;
1971 nd->seq = nd->next_seq;
1972 return NULL;
1973 }
1974 if (nd->flags & LOOKUP_RCU) {
1975 /* make sure that d_is_symlink above matches inode */
1976 if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq))
1977 return ERR_PTR(-ECHILD);
1978 } else {
1979 if (path.mnt == nd->path.mnt)
1980 mntget(path.mnt);
1981 }
1982 return pick_link(nd, &path, inode, flags);
1983}
1984
1985static struct dentry *follow_dotdot_rcu(struct nameidata *nd)
1986{
1987 struct dentry *parent, *old;
1988
1989 if (path_equal(&nd->path, &nd->root))
1990 goto in_root;
1991 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1992 struct path path;
1993 unsigned seq;
1994 if (!choose_mountpoint_rcu(real_mount(nd->path.mnt),
1995 &nd->root, &path, &seq))
1996 goto in_root;
1997 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1998 return ERR_PTR(-ECHILD);
1999 nd->path = path;
2000 nd->inode = path.dentry->d_inode;
2001 nd->seq = seq;
2002 // makes sure that non-RCU pathwalk could reach this state
2003 if (read_seqretry(&mount_lock, nd->m_seq))
2004 return ERR_PTR(-ECHILD);
2005 /* we know that mountpoint was pinned */
2006 }
2007 old = nd->path.dentry;
2008 parent = old->d_parent;
2009 nd->next_seq = read_seqcount_begin(&parent->d_seq);
2010 // makes sure that non-RCU pathwalk could reach this state
2011 if (read_seqcount_retry(&old->d_seq, nd->seq))
2012 return ERR_PTR(-ECHILD);
2013 if (unlikely(!path_connected(nd->path.mnt, parent)))
2014 return ERR_PTR(-ECHILD);
2015 return parent;
2016in_root:
2017 if (read_seqretry(&mount_lock, nd->m_seq))
2018 return ERR_PTR(-ECHILD);
2019 if (unlikely(nd->flags & LOOKUP_BENEATH))
2020 return ERR_PTR(-ECHILD);
2021 nd->next_seq = nd->seq;
2022 return nd->path.dentry;
2023}
2024
2025static struct dentry *follow_dotdot(struct nameidata *nd)
2026{
2027 struct dentry *parent;
2028
2029 if (path_equal(&nd->path, &nd->root))
2030 goto in_root;
2031 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
2032 struct path path;
2033
2034 if (!choose_mountpoint(real_mount(nd->path.mnt),
2035 &nd->root, &path))
2036 goto in_root;
2037 path_put(&nd->path);
2038 nd->path = path;
2039 nd->inode = path.dentry->d_inode;
2040 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
2041 return ERR_PTR(-EXDEV);
2042 }
2043 /* rare case of legitimate dget_parent()... */
2044 parent = dget_parent(nd->path.dentry);
2045 if (unlikely(!path_connected(nd->path.mnt, parent))) {
2046 dput(parent);
2047 return ERR_PTR(-ENOENT);
2048 }
2049 return parent;
2050
2051in_root:
2052 if (unlikely(nd->flags & LOOKUP_BENEATH))
2053 return ERR_PTR(-EXDEV);
2054 return dget(nd->path.dentry);
2055}
2056
2057static const char *handle_dots(struct nameidata *nd, int type)
2058{
2059 if (type == LAST_DOTDOT) {
2060 const char *error = NULL;
2061 struct dentry *parent;
2062
2063 if (!nd->root.mnt) {
2064 error = ERR_PTR(set_root(nd));
2065 if (error)
2066 return error;
2067 }
2068 if (nd->flags & LOOKUP_RCU)
2069 parent = follow_dotdot_rcu(nd);
2070 else
2071 parent = follow_dotdot(nd);
2072 if (IS_ERR(parent))
2073 return ERR_CAST(parent);
2074 error = step_into(nd, WALK_NOFOLLOW, parent);
2075 if (unlikely(error))
2076 return error;
2077
2078 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
2079 /*
2080 * If there was a racing rename or mount along our
2081 * path, then we can't be sure that ".." hasn't jumped
2082 * above nd->root (and so userspace should retry or use
2083 * some fallback).
2084 */
2085 smp_rmb();
2086 if (__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq))
2087 return ERR_PTR(-EAGAIN);
2088 if (__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq))
2089 return ERR_PTR(-EAGAIN);
2090 }
2091 }
2092 return NULL;
2093}
2094
2095static const char *walk_component(struct nameidata *nd, int flags)
2096{
2097 struct dentry *dentry;
2098 /*
2099 * "." and ".." are special - ".." especially so because it has
2100 * to be able to know about the current root directory and
2101 * parent relationships.
2102 */
2103 if (unlikely(nd->last_type != LAST_NORM)) {
2104 if (!(flags & WALK_MORE) && nd->depth)
2105 put_link(nd);
2106 return handle_dots(nd, nd->last_type);
2107 }
2108 dentry = lookup_fast(nd);
2109 if (IS_ERR(dentry))
2110 return ERR_CAST(dentry);
2111 if (unlikely(!dentry)) {
2112 dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags);
2113 if (IS_ERR(dentry))
2114 return ERR_CAST(dentry);
2115 }
2116 if (!(flags & WALK_MORE) && nd->depth)
2117 put_link(nd);
2118 return step_into(nd, flags, dentry);
2119}
2120
2121/*
2122 * We can do the critical dentry name comparison and hashing
2123 * operations one word at a time, but we are limited to:
2124 *
2125 * - Architectures with fast unaligned word accesses. We could
2126 * do a "get_unaligned()" if this helps and is sufficiently
2127 * fast.
2128 *
2129 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
2130 * do not trap on the (extremely unlikely) case of a page
2131 * crossing operation.
2132 *
2133 * - Furthermore, we need an efficient 64-bit compile for the
2134 * 64-bit case in order to generate the "number of bytes in
2135 * the final mask". Again, that could be replaced with a
2136 * efficient population count instruction or similar.
2137 */
2138#ifdef CONFIG_DCACHE_WORD_ACCESS
2139
2140#include <asm/word-at-a-time.h>
2141
2142#ifdef HASH_MIX
2143
2144/* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
2145
2146#elif defined(CONFIG_64BIT)
2147/*
2148 * Register pressure in the mixing function is an issue, particularly
2149 * on 32-bit x86, but almost any function requires one state value and
2150 * one temporary. Instead, use a function designed for two state values
2151 * and no temporaries.
2152 *
2153 * This function cannot create a collision in only two iterations, so
2154 * we have two iterations to achieve avalanche. In those two iterations,
2155 * we have six layers of mixing, which is enough to spread one bit's
2156 * influence out to 2^6 = 64 state bits.
2157 *
2158 * Rotate constants are scored by considering either 64 one-bit input
2159 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
2160 * probability of that delta causing a change to each of the 128 output
2161 * bits, using a sample of random initial states.
2162 *
2163 * The Shannon entropy of the computed probabilities is then summed
2164 * to produce a score. Ideally, any input change has a 50% chance of
2165 * toggling any given output bit.
2166 *
2167 * Mixing scores (in bits) for (12,45):
2168 * Input delta: 1-bit 2-bit
2169 * 1 round: 713.3 42542.6
2170 * 2 rounds: 2753.7 140389.8
2171 * 3 rounds: 5954.1 233458.2
2172 * 4 rounds: 7862.6 256672.2
2173 * Perfect: 8192 258048
2174 * (64*128) (64*63/2 * 128)
2175 */
2176#define HASH_MIX(x, y, a) \
2177 ( x ^= (a), \
2178 y ^= x, x = rol64(x,12),\
2179 x += y, y = rol64(y,45),\
2180 y *= 9 )
2181
2182/*
2183 * Fold two longs into one 32-bit hash value. This must be fast, but
2184 * latency isn't quite as critical, as there is a fair bit of additional
2185 * work done before the hash value is used.
2186 */
2187static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2188{
2189 y ^= x * GOLDEN_RATIO_64;
2190 y *= GOLDEN_RATIO_64;
2191 return y >> 32;
2192}
2193
2194#else /* 32-bit case */
2195
2196/*
2197 * Mixing scores (in bits) for (7,20):
2198 * Input delta: 1-bit 2-bit
2199 * 1 round: 330.3 9201.6
2200 * 2 rounds: 1246.4 25475.4
2201 * 3 rounds: 1907.1 31295.1
2202 * 4 rounds: 2042.3 31718.6
2203 * Perfect: 2048 31744
2204 * (32*64) (32*31/2 * 64)
2205 */
2206#define HASH_MIX(x, y, a) \
2207 ( x ^= (a), \
2208 y ^= x, x = rol32(x, 7),\
2209 x += y, y = rol32(y,20),\
2210 y *= 9 )
2211
2212static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2213{
2214 /* Use arch-optimized multiply if one exists */
2215 return __hash_32(y ^ __hash_32(x));
2216}
2217
2218#endif
2219
2220/*
2221 * Return the hash of a string of known length. This is carfully
2222 * designed to match hash_name(), which is the more critical function.
2223 * In particular, we must end by hashing a final word containing 0..7
2224 * payload bytes, to match the way that hash_name() iterates until it
2225 * finds the delimiter after the name.
2226 */
2227unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2228{
2229 unsigned long a, x = 0, y = (unsigned long)salt;
2230
2231 for (;;) {
2232 if (!len)
2233 goto done;
2234 a = load_unaligned_zeropad(name);
2235 if (len < sizeof(unsigned long))
2236 break;
2237 HASH_MIX(x, y, a);
2238 name += sizeof(unsigned long);
2239 len -= sizeof(unsigned long);
2240 }
2241 x ^= a & bytemask_from_count(len);
2242done:
2243 return fold_hash(x, y);
2244}
2245EXPORT_SYMBOL(full_name_hash);
2246
2247/* Return the "hash_len" (hash and length) of a null-terminated string */
2248u64 hashlen_string(const void *salt, const char *name)
2249{
2250 unsigned long a = 0, x = 0, y = (unsigned long)salt;
2251 unsigned long adata, mask, len;
2252 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2253
2254 len = 0;
2255 goto inside;
2256
2257 do {
2258 HASH_MIX(x, y, a);
2259 len += sizeof(unsigned long);
2260inside:
2261 a = load_unaligned_zeropad(name+len);
2262 } while (!has_zero(a, &adata, &constants));
2263
2264 adata = prep_zero_mask(a, adata, &constants);
2265 mask = create_zero_mask(adata);
2266 x ^= a & zero_bytemask(mask);
2267
2268 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2269}
2270EXPORT_SYMBOL(hashlen_string);
2271
2272/*
2273 * Calculate the length and hash of the path component, and
2274 * return the length as the result.
2275 */
2276static inline const char *hash_name(struct nameidata *nd,
2277 const char *name,
2278 unsigned long *lastword)
2279{
2280 unsigned long a, b, x, y = (unsigned long)nd->path.dentry;
2281 unsigned long adata, bdata, mask, len;
2282 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2283
2284 /*
2285 * The first iteration is special, because it can result in
2286 * '.' and '..' and has no mixing other than the final fold.
2287 */
2288 a = load_unaligned_zeropad(name);
2289 b = a ^ REPEAT_BYTE('/');
2290 if (has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)) {
2291 adata = prep_zero_mask(a, adata, &constants);
2292 bdata = prep_zero_mask(b, bdata, &constants);
2293 mask = create_zero_mask(adata | bdata);
2294 a &= zero_bytemask(mask);
2295 *lastword = a;
2296 len = find_zero(mask);
2297 nd->last.hash = fold_hash(a, y);
2298 nd->last.len = len;
2299 return name + len;
2300 }
2301
2302 len = 0;
2303 x = 0;
2304 do {
2305 HASH_MIX(x, y, a);
2306 len += sizeof(unsigned long);
2307 a = load_unaligned_zeropad(name+len);
2308 b = a ^ REPEAT_BYTE('/');
2309 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
2310
2311 adata = prep_zero_mask(a, adata, &constants);
2312 bdata = prep_zero_mask(b, bdata, &constants);
2313 mask = create_zero_mask(adata | bdata);
2314 a &= zero_bytemask(mask);
2315 x ^= a;
2316 len += find_zero(mask);
2317 *lastword = 0; // Multi-word components cannot be DOT or DOTDOT
2318
2319 nd->last.hash = fold_hash(x, y);
2320 nd->last.len = len;
2321 return name + len;
2322}
2323
2324/*
2325 * Note that the 'last' word is always zero-masked, but
2326 * was loaded as a possibly big-endian word.
2327 */
2328#ifdef __BIG_ENDIAN
2329 #define LAST_WORD_IS_DOT (0x2eul << (BITS_PER_LONG-8))
2330 #define LAST_WORD_IS_DOTDOT (0x2e2eul << (BITS_PER_LONG-16))
2331#endif
2332
2333#else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2334
2335/* Return the hash of a string of known length */
2336unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2337{
2338 unsigned long hash = init_name_hash(salt);
2339 while (len--)
2340 hash = partial_name_hash((unsigned char)*name++, hash);
2341 return end_name_hash(hash);
2342}
2343EXPORT_SYMBOL(full_name_hash);
2344
2345/* Return the "hash_len" (hash and length) of a null-terminated string */
2346u64 hashlen_string(const void *salt, const char *name)
2347{
2348 unsigned long hash = init_name_hash(salt);
2349 unsigned long len = 0, c;
2350
2351 c = (unsigned char)*name;
2352 while (c) {
2353 len++;
2354 hash = partial_name_hash(c, hash);
2355 c = (unsigned char)name[len];
2356 }
2357 return hashlen_create(end_name_hash(hash), len);
2358}
2359EXPORT_SYMBOL(hashlen_string);
2360
2361/*
2362 * We know there's a real path component here of at least
2363 * one character.
2364 */
2365static inline const char *hash_name(struct nameidata *nd, const char *name, unsigned long *lastword)
2366{
2367 unsigned long hash = init_name_hash(nd->path.dentry);
2368 unsigned long len = 0, c, last = 0;
2369
2370 c = (unsigned char)*name;
2371 do {
2372 last = (last << 8) + c;
2373 len++;
2374 hash = partial_name_hash(c, hash);
2375 c = (unsigned char)name[len];
2376 } while (c && c != '/');
2377
2378 // This is reliable for DOT or DOTDOT, since the component
2379 // cannot contain NUL characters - top bits being zero means
2380 // we cannot have had any other pathnames.
2381 *lastword = last;
2382 nd->last.hash = end_name_hash(hash);
2383 nd->last.len = len;
2384 return name + len;
2385}
2386
2387#endif
2388
2389#ifndef LAST_WORD_IS_DOT
2390 #define LAST_WORD_IS_DOT 0x2e
2391 #define LAST_WORD_IS_DOTDOT 0x2e2e
2392#endif
2393
2394/*
2395 * Name resolution.
2396 * This is the basic name resolution function, turning a pathname into
2397 * the final dentry. We expect 'base' to be positive and a directory.
2398 *
2399 * Returns 0 and nd will have valid dentry and mnt on success.
2400 * Returns error and drops reference to input namei data on failure.
2401 */
2402static int link_path_walk(const char *name, struct nameidata *nd)
2403{
2404 int depth = 0; // depth <= nd->depth
2405 int err;
2406
2407 nd->last_type = LAST_ROOT;
2408 nd->flags |= LOOKUP_PARENT;
2409 if (IS_ERR(name))
2410 return PTR_ERR(name);
2411 while (*name=='/')
2412 name++;
2413 if (!*name) {
2414 nd->dir_mode = 0; // short-circuit the 'hardening' idiocy
2415 return 0;
2416 }
2417
2418 /* At this point we know we have a real path component. */
2419 for(;;) {
2420 struct mnt_idmap *idmap;
2421 const char *link;
2422 unsigned long lastword;
2423
2424 idmap = mnt_idmap(nd->path.mnt);
2425 err = may_lookup(idmap, nd);
2426 if (err)
2427 return err;
2428
2429 nd->last.name = name;
2430 name = hash_name(nd, name, &lastword);
2431
2432 switch(lastword) {
2433 case LAST_WORD_IS_DOTDOT:
2434 nd->last_type = LAST_DOTDOT;
2435 nd->state |= ND_JUMPED;
2436 break;
2437
2438 case LAST_WORD_IS_DOT:
2439 nd->last_type = LAST_DOT;
2440 break;
2441
2442 default:
2443 nd->last_type = LAST_NORM;
2444 nd->state &= ~ND_JUMPED;
2445
2446 struct dentry *parent = nd->path.dentry;
2447 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2448 err = parent->d_op->d_hash(parent, &nd->last);
2449 if (err < 0)
2450 return err;
2451 }
2452 }
2453
2454 if (!*name)
2455 goto OK;
2456 /*
2457 * If it wasn't NUL, we know it was '/'. Skip that
2458 * slash, and continue until no more slashes.
2459 */
2460 do {
2461 name++;
2462 } while (unlikely(*name == '/'));
2463 if (unlikely(!*name)) {
2464OK:
2465 /* pathname or trailing symlink, done */
2466 if (!depth) {
2467 nd->dir_vfsuid = i_uid_into_vfsuid(idmap, nd->inode);
2468 nd->dir_mode = nd->inode->i_mode;
2469 nd->flags &= ~LOOKUP_PARENT;
2470 return 0;
2471 }
2472 /* last component of nested symlink */
2473 name = nd->stack[--depth].name;
2474 link = walk_component(nd, 0);
2475 } else {
2476 /* not the last component */
2477 link = walk_component(nd, WALK_MORE);
2478 }
2479 if (unlikely(link)) {
2480 if (IS_ERR(link))
2481 return PTR_ERR(link);
2482 /* a symlink to follow */
2483 nd->stack[depth++].name = name;
2484 name = link;
2485 continue;
2486 }
2487 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2488 if (nd->flags & LOOKUP_RCU) {
2489 if (!try_to_unlazy(nd))
2490 return -ECHILD;
2491 }
2492 return -ENOTDIR;
2493 }
2494 }
2495}
2496
2497/* must be paired with terminate_walk() */
2498static const char *path_init(struct nameidata *nd, unsigned flags)
2499{
2500 int error;
2501 const char *s = nd->pathname;
2502
2503 /* LOOKUP_CACHED requires RCU, ask caller to retry */
2504 if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED)
2505 return ERR_PTR(-EAGAIN);
2506
2507 if (!*s)
2508 flags &= ~LOOKUP_RCU;
2509 if (flags & LOOKUP_RCU)
2510 rcu_read_lock();
2511 else
2512 nd->seq = nd->next_seq = 0;
2513
2514 nd->flags = flags;
2515 nd->state |= ND_JUMPED;
2516
2517 nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount);
2518 nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount);
2519 smp_rmb();
2520
2521 if (nd->state & ND_ROOT_PRESET) {
2522 struct dentry *root = nd->root.dentry;
2523 struct inode *inode = root->d_inode;
2524 if (*s && unlikely(!d_can_lookup(root)))
2525 return ERR_PTR(-ENOTDIR);
2526 nd->path = nd->root;
2527 nd->inode = inode;
2528 if (flags & LOOKUP_RCU) {
2529 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2530 nd->root_seq = nd->seq;
2531 } else {
2532 path_get(&nd->path);
2533 }
2534 return s;
2535 }
2536
2537 nd->root.mnt = NULL;
2538
2539 /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2540 if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) {
2541 error = nd_jump_root(nd);
2542 if (unlikely(error))
2543 return ERR_PTR(error);
2544 return s;
2545 }
2546
2547 /* Relative pathname -- get the starting-point it is relative to. */
2548 if (nd->dfd == AT_FDCWD) {
2549 if (flags & LOOKUP_RCU) {
2550 struct fs_struct *fs = current->fs;
2551 unsigned seq;
2552
2553 do {
2554 seq = read_seqcount_begin(&fs->seq);
2555 nd->path = fs->pwd;
2556 nd->inode = nd->path.dentry->d_inode;
2557 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2558 } while (read_seqcount_retry(&fs->seq, seq));
2559 } else {
2560 get_fs_pwd(current->fs, &nd->path);
2561 nd->inode = nd->path.dentry->d_inode;
2562 }
2563 } else {
2564 /* Caller must check execute permissions on the starting path component */
2565 CLASS(fd_raw, f)(nd->dfd);
2566 struct dentry *dentry;
2567
2568 if (fd_empty(f))
2569 return ERR_PTR(-EBADF);
2570
2571 if (flags & LOOKUP_LINKAT_EMPTY) {
2572 if (fd_file(f)->f_cred != current_cred() &&
2573 !ns_capable(fd_file(f)->f_cred->user_ns, CAP_DAC_READ_SEARCH))
2574 return ERR_PTR(-ENOENT);
2575 }
2576
2577 dentry = fd_file(f)->f_path.dentry;
2578
2579 if (*s && unlikely(!d_can_lookup(dentry)))
2580 return ERR_PTR(-ENOTDIR);
2581
2582 nd->path = fd_file(f)->f_path;
2583 if (flags & LOOKUP_RCU) {
2584 nd->inode = nd->path.dentry->d_inode;
2585 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2586 } else {
2587 path_get(&nd->path);
2588 nd->inode = nd->path.dentry->d_inode;
2589 }
2590 }
2591
2592 /* For scoped-lookups we need to set the root to the dirfd as well. */
2593 if (flags & LOOKUP_IS_SCOPED) {
2594 nd->root = nd->path;
2595 if (flags & LOOKUP_RCU) {
2596 nd->root_seq = nd->seq;
2597 } else {
2598 path_get(&nd->root);
2599 nd->state |= ND_ROOT_GRABBED;
2600 }
2601 }
2602 return s;
2603}
2604
2605static inline const char *lookup_last(struct nameidata *nd)
2606{
2607 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2608 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2609
2610 return walk_component(nd, WALK_TRAILING);
2611}
2612
2613static int handle_lookup_down(struct nameidata *nd)
2614{
2615 if (!(nd->flags & LOOKUP_RCU))
2616 dget(nd->path.dentry);
2617 nd->next_seq = nd->seq;
2618 return PTR_ERR(step_into(nd, WALK_NOFOLLOW, nd->path.dentry));
2619}
2620
2621/* Returns 0 and nd will be valid on success; Returns error, otherwise. */
2622static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2623{
2624 const char *s = path_init(nd, flags);
2625 int err;
2626
2627 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2628 err = handle_lookup_down(nd);
2629 if (unlikely(err < 0))
2630 s = ERR_PTR(err);
2631 }
2632
2633 while (!(err = link_path_walk(s, nd)) &&
2634 (s = lookup_last(nd)) != NULL)
2635 ;
2636 if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) {
2637 err = handle_lookup_down(nd);
2638 nd->state &= ~ND_JUMPED; // no d_weak_revalidate(), please...
2639 }
2640 if (!err)
2641 err = complete_walk(nd);
2642
2643 if (!err && nd->flags & LOOKUP_DIRECTORY)
2644 if (!d_can_lookup(nd->path.dentry))
2645 err = -ENOTDIR;
2646 if (!err) {
2647 *path = nd->path;
2648 nd->path.mnt = NULL;
2649 nd->path.dentry = NULL;
2650 }
2651 terminate_walk(nd);
2652 return err;
2653}
2654
2655int filename_lookup(int dfd, struct filename *name, unsigned flags,
2656 struct path *path, struct path *root)
2657{
2658 int retval;
2659 struct nameidata nd;
2660 if (IS_ERR(name))
2661 return PTR_ERR(name);
2662 set_nameidata(&nd, dfd, name, root);
2663 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2664 if (unlikely(retval == -ECHILD))
2665 retval = path_lookupat(&nd, flags, path);
2666 if (unlikely(retval == -ESTALE))
2667 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2668
2669 if (likely(!retval))
2670 audit_inode(name, path->dentry,
2671 flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0);
2672 restore_nameidata();
2673 return retval;
2674}
2675
2676/* Returns 0 and nd will be valid on success; Returns error, otherwise. */
2677static int path_parentat(struct nameidata *nd, unsigned flags,
2678 struct path *parent)
2679{
2680 const char *s = path_init(nd, flags);
2681 int err = link_path_walk(s, nd);
2682 if (!err)
2683 err = complete_walk(nd);
2684 if (!err) {
2685 *parent = nd->path;
2686 nd->path.mnt = NULL;
2687 nd->path.dentry = NULL;
2688 }
2689 terminate_walk(nd);
2690 return err;
2691}
2692
2693/* Note: this does not consume "name" */
2694static int __filename_parentat(int dfd, struct filename *name,
2695 unsigned int flags, struct path *parent,
2696 struct qstr *last, int *type,
2697 const struct path *root)
2698{
2699 int retval;
2700 struct nameidata nd;
2701
2702 if (IS_ERR(name))
2703 return PTR_ERR(name);
2704 set_nameidata(&nd, dfd, name, root);
2705 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2706 if (unlikely(retval == -ECHILD))
2707 retval = path_parentat(&nd, flags, parent);
2708 if (unlikely(retval == -ESTALE))
2709 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2710 if (likely(!retval)) {
2711 *last = nd.last;
2712 *type = nd.last_type;
2713 audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
2714 }
2715 restore_nameidata();
2716 return retval;
2717}
2718
2719static int filename_parentat(int dfd, struct filename *name,
2720 unsigned int flags, struct path *parent,
2721 struct qstr *last, int *type)
2722{
2723 return __filename_parentat(dfd, name, flags, parent, last, type, NULL);
2724}
2725
2726/* does lookup, returns the object with parent locked */
2727static struct dentry *__kern_path_locked(int dfd, struct filename *name, struct path *path)
2728{
2729 struct dentry *d;
2730 struct qstr last;
2731 int type, error;
2732
2733 error = filename_parentat(dfd, name, 0, path, &last, &type);
2734 if (error)
2735 return ERR_PTR(error);
2736 if (unlikely(type != LAST_NORM)) {
2737 path_put(path);
2738 return ERR_PTR(-EINVAL);
2739 }
2740 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2741 d = lookup_one_qstr_excl(&last, path->dentry, 0);
2742 if (IS_ERR(d)) {
2743 inode_unlock(path->dentry->d_inode);
2744 path_put(path);
2745 }
2746 return d;
2747}
2748
2749struct dentry *kern_path_locked(const char *name, struct path *path)
2750{
2751 struct filename *filename = getname_kernel(name);
2752 struct dentry *res = __kern_path_locked(AT_FDCWD, filename, path);
2753
2754 putname(filename);
2755 return res;
2756}
2757
2758struct dentry *user_path_locked_at(int dfd, const char __user *name, struct path *path)
2759{
2760 struct filename *filename = getname(name);
2761 struct dentry *res = __kern_path_locked(dfd, filename, path);
2762
2763 putname(filename);
2764 return res;
2765}
2766EXPORT_SYMBOL(user_path_locked_at);
2767
2768int kern_path(const char *name, unsigned int flags, struct path *path)
2769{
2770 struct filename *filename = getname_kernel(name);
2771 int ret = filename_lookup(AT_FDCWD, filename, flags, path, NULL);
2772
2773 putname(filename);
2774 return ret;
2775
2776}
2777EXPORT_SYMBOL(kern_path);
2778
2779/**
2780 * vfs_path_parent_lookup - lookup a parent path relative to a dentry-vfsmount pair
2781 * @filename: filename structure
2782 * @flags: lookup flags
2783 * @parent: pointer to struct path to fill
2784 * @last: last component
2785 * @type: type of the last component
2786 * @root: pointer to struct path of the base directory
2787 */
2788int vfs_path_parent_lookup(struct filename *filename, unsigned int flags,
2789 struct path *parent, struct qstr *last, int *type,
2790 const struct path *root)
2791{
2792 return __filename_parentat(AT_FDCWD, filename, flags, parent, last,
2793 type, root);
2794}
2795EXPORT_SYMBOL(vfs_path_parent_lookup);
2796
2797/**
2798 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2799 * @dentry: pointer to dentry of the base directory
2800 * @mnt: pointer to vfs mount of the base directory
2801 * @name: pointer to file name
2802 * @flags: lookup flags
2803 * @path: pointer to struct path to fill
2804 */
2805int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2806 const char *name, unsigned int flags,
2807 struct path *path)
2808{
2809 struct filename *filename;
2810 struct path root = {.mnt = mnt, .dentry = dentry};
2811 int ret;
2812
2813 filename = getname_kernel(name);
2814 /* the first argument of filename_lookup() is ignored with root */
2815 ret = filename_lookup(AT_FDCWD, filename, flags, path, &root);
2816 putname(filename);
2817 return ret;
2818}
2819EXPORT_SYMBOL(vfs_path_lookup);
2820
2821static int lookup_one_common(struct mnt_idmap *idmap,
2822 const char *name, struct dentry *base, int len,
2823 struct qstr *this)
2824{
2825 this->name = name;
2826 this->len = len;
2827 this->hash = full_name_hash(base, name, len);
2828 if (!len)
2829 return -EACCES;
2830
2831 if (is_dot_dotdot(name, len))
2832 return -EACCES;
2833
2834 while (len--) {
2835 unsigned int c = *(const unsigned char *)name++;
2836 if (c == '/' || c == '\0')
2837 return -EACCES;
2838 }
2839 /*
2840 * See if the low-level filesystem might want
2841 * to use its own hash..
2842 */
2843 if (base->d_flags & DCACHE_OP_HASH) {
2844 int err = base->d_op->d_hash(base, this);
2845 if (err < 0)
2846 return err;
2847 }
2848
2849 return inode_permission(idmap, base->d_inode, MAY_EXEC);
2850}
2851
2852/**
2853 * try_lookup_one_len - filesystem helper to lookup single pathname component
2854 * @name: pathname component to lookup
2855 * @base: base directory to lookup from
2856 * @len: maximum length @len should be interpreted to
2857 *
2858 * Look up a dentry by name in the dcache, returning NULL if it does not
2859 * currently exist. The function does not try to create a dentry.
2860 *
2861 * Note that this routine is purely a helper for filesystem usage and should
2862 * not be called by generic code.
2863 *
2864 * The caller must hold base->i_mutex.
2865 */
2866struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2867{
2868 struct qstr this;
2869 int err;
2870
2871 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2872
2873 err = lookup_one_common(&nop_mnt_idmap, name, base, len, &this);
2874 if (err)
2875 return ERR_PTR(err);
2876
2877 return lookup_dcache(&this, base, 0);
2878}
2879EXPORT_SYMBOL(try_lookup_one_len);
2880
2881/**
2882 * lookup_one_len - filesystem helper to lookup single pathname component
2883 * @name: pathname component to lookup
2884 * @base: base directory to lookup from
2885 * @len: maximum length @len should be interpreted to
2886 *
2887 * Note that this routine is purely a helper for filesystem usage and should
2888 * not be called by generic code.
2889 *
2890 * The caller must hold base->i_mutex.
2891 */
2892struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2893{
2894 struct dentry *dentry;
2895 struct qstr this;
2896 int err;
2897
2898 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2899
2900 err = lookup_one_common(&nop_mnt_idmap, name, base, len, &this);
2901 if (err)
2902 return ERR_PTR(err);
2903
2904 dentry = lookup_dcache(&this, base, 0);
2905 return dentry ? dentry : __lookup_slow(&this, base, 0);
2906}
2907EXPORT_SYMBOL(lookup_one_len);
2908
2909/**
2910 * lookup_one - filesystem helper to lookup single pathname component
2911 * @idmap: idmap of the mount the lookup is performed from
2912 * @name: pathname component to lookup
2913 * @base: base directory to lookup from
2914 * @len: maximum length @len should be interpreted to
2915 *
2916 * Note that this routine is purely a helper for filesystem usage and should
2917 * not be called by generic code.
2918 *
2919 * The caller must hold base->i_mutex.
2920 */
2921struct dentry *lookup_one(struct mnt_idmap *idmap, const char *name,
2922 struct dentry *base, int len)
2923{
2924 struct dentry *dentry;
2925 struct qstr this;
2926 int err;
2927
2928 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2929
2930 err = lookup_one_common(idmap, name, base, len, &this);
2931 if (err)
2932 return ERR_PTR(err);
2933
2934 dentry = lookup_dcache(&this, base, 0);
2935 return dentry ? dentry : __lookup_slow(&this, base, 0);
2936}
2937EXPORT_SYMBOL(lookup_one);
2938
2939/**
2940 * lookup_one_unlocked - filesystem helper to lookup single pathname component
2941 * @idmap: idmap of the mount the lookup is performed from
2942 * @name: pathname component to lookup
2943 * @base: base directory to lookup from
2944 * @len: maximum length @len should be interpreted to
2945 *
2946 * Note that this routine is purely a helper for filesystem usage and should
2947 * not be called by generic code.
2948 *
2949 * Unlike lookup_one_len, it should be called without the parent
2950 * i_mutex held, and will take the i_mutex itself if necessary.
2951 */
2952struct dentry *lookup_one_unlocked(struct mnt_idmap *idmap,
2953 const char *name, struct dentry *base,
2954 int len)
2955{
2956 struct qstr this;
2957 int err;
2958 struct dentry *ret;
2959
2960 err = lookup_one_common(idmap, name, base, len, &this);
2961 if (err)
2962 return ERR_PTR(err);
2963
2964 ret = lookup_dcache(&this, base, 0);
2965 if (!ret)
2966 ret = lookup_slow(&this, base, 0);
2967 return ret;
2968}
2969EXPORT_SYMBOL(lookup_one_unlocked);
2970
2971/**
2972 * lookup_one_positive_unlocked - filesystem helper to lookup single
2973 * pathname component
2974 * @idmap: idmap of the mount the lookup is performed from
2975 * @name: pathname component to lookup
2976 * @base: base directory to lookup from
2977 * @len: maximum length @len should be interpreted to
2978 *
2979 * This helper will yield ERR_PTR(-ENOENT) on negatives. The helper returns
2980 * known positive or ERR_PTR(). This is what most of the users want.
2981 *
2982 * Note that pinned negative with unlocked parent _can_ become positive at any
2983 * time, so callers of lookup_one_unlocked() need to be very careful; pinned
2984 * positives have >d_inode stable, so this one avoids such problems.
2985 *
2986 * Note that this routine is purely a helper for filesystem usage and should
2987 * not be called by generic code.
2988 *
2989 * The helper should be called without i_mutex held.
2990 */
2991struct dentry *lookup_one_positive_unlocked(struct mnt_idmap *idmap,
2992 const char *name,
2993 struct dentry *base, int len)
2994{
2995 struct dentry *ret = lookup_one_unlocked(idmap, name, base, len);
2996
2997 if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
2998 dput(ret);
2999 ret = ERR_PTR(-ENOENT);
3000 }
3001 return ret;
3002}
3003EXPORT_SYMBOL(lookup_one_positive_unlocked);
3004
3005/**
3006 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
3007 * @name: pathname component to lookup
3008 * @base: base directory to lookup from
3009 * @len: maximum length @len should be interpreted to
3010 *
3011 * Note that this routine is purely a helper for filesystem usage and should
3012 * not be called by generic code.
3013 *
3014 * Unlike lookup_one_len, it should be called without the parent
3015 * i_mutex held, and will take the i_mutex itself if necessary.
3016 */
3017struct dentry *lookup_one_len_unlocked(const char *name,
3018 struct dentry *base, int len)
3019{
3020 return lookup_one_unlocked(&nop_mnt_idmap, name, base, len);
3021}
3022EXPORT_SYMBOL(lookup_one_len_unlocked);
3023
3024/*
3025 * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
3026 * on negatives. Returns known positive or ERR_PTR(); that's what
3027 * most of the users want. Note that pinned negative with unlocked parent
3028 * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
3029 * need to be very careful; pinned positives have ->d_inode stable, so
3030 * this one avoids such problems.
3031 */
3032struct dentry *lookup_positive_unlocked(const char *name,
3033 struct dentry *base, int len)
3034{
3035 return lookup_one_positive_unlocked(&nop_mnt_idmap, name, base, len);
3036}
3037EXPORT_SYMBOL(lookup_positive_unlocked);
3038
3039#ifdef CONFIG_UNIX98_PTYS
3040int path_pts(struct path *path)
3041{
3042 /* Find something mounted on "pts" in the same directory as
3043 * the input path.
3044 */
3045 struct dentry *parent = dget_parent(path->dentry);
3046 struct dentry *child;
3047 struct qstr this = QSTR_INIT("pts", 3);
3048
3049 if (unlikely(!path_connected(path->mnt, parent))) {
3050 dput(parent);
3051 return -ENOENT;
3052 }
3053 dput(path->dentry);
3054 path->dentry = parent;
3055 child = d_hash_and_lookup(parent, &this);
3056 if (IS_ERR_OR_NULL(child))
3057 return -ENOENT;
3058
3059 path->dentry = child;
3060 dput(parent);
3061 follow_down(path, 0);
3062 return 0;
3063}
3064#endif
3065
3066int user_path_at(int dfd, const char __user *name, unsigned flags,
3067 struct path *path)
3068{
3069 struct filename *filename = getname_flags(name, flags);
3070 int ret = filename_lookup(dfd, filename, flags, path, NULL);
3071
3072 putname(filename);
3073 return ret;
3074}
3075EXPORT_SYMBOL(user_path_at);
3076
3077int __check_sticky(struct mnt_idmap *idmap, struct inode *dir,
3078 struct inode *inode)
3079{
3080 kuid_t fsuid = current_fsuid();
3081
3082 if (vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, inode), fsuid))
3083 return 0;
3084 if (vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, dir), fsuid))
3085 return 0;
3086 return !capable_wrt_inode_uidgid(idmap, inode, CAP_FOWNER);
3087}
3088EXPORT_SYMBOL(__check_sticky);
3089
3090/*
3091 * Check whether we can remove a link victim from directory dir, check
3092 * whether the type of victim is right.
3093 * 1. We can't do it if dir is read-only (done in permission())
3094 * 2. We should have write and exec permissions on dir
3095 * 3. We can't remove anything from append-only dir
3096 * 4. We can't do anything with immutable dir (done in permission())
3097 * 5. If the sticky bit on dir is set we should either
3098 * a. be owner of dir, or
3099 * b. be owner of victim, or
3100 * c. have CAP_FOWNER capability
3101 * 6. If the victim is append-only or immutable we can't do antyhing with
3102 * links pointing to it.
3103 * 7. If the victim has an unknown uid or gid we can't change the inode.
3104 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
3105 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
3106 * 10. We can't remove a root or mountpoint.
3107 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
3108 * nfs_async_unlink().
3109 */
3110static int may_delete(struct mnt_idmap *idmap, struct inode *dir,
3111 struct dentry *victim, bool isdir)
3112{
3113 struct inode *inode = d_backing_inode(victim);
3114 int error;
3115
3116 if (d_is_negative(victim))
3117 return -ENOENT;
3118 BUG_ON(!inode);
3119
3120 BUG_ON(victim->d_parent->d_inode != dir);
3121
3122 /* Inode writeback is not safe when the uid or gid are invalid. */
3123 if (!vfsuid_valid(i_uid_into_vfsuid(idmap, inode)) ||
3124 !vfsgid_valid(i_gid_into_vfsgid(idmap, inode)))
3125 return -EOVERFLOW;
3126
3127 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
3128
3129 error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
3130 if (error)
3131 return error;
3132 if (IS_APPEND(dir))
3133 return -EPERM;
3134
3135 if (check_sticky(idmap, dir, inode) || IS_APPEND(inode) ||
3136 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) ||
3137 HAS_UNMAPPED_ID(idmap, inode))
3138 return -EPERM;
3139 if (isdir) {
3140 if (!d_is_dir(victim))
3141 return -ENOTDIR;
3142 if (IS_ROOT(victim))
3143 return -EBUSY;
3144 } else if (d_is_dir(victim))
3145 return -EISDIR;
3146 if (IS_DEADDIR(dir))
3147 return -ENOENT;
3148 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
3149 return -EBUSY;
3150 return 0;
3151}
3152
3153/* Check whether we can create an object with dentry child in directory
3154 * dir.
3155 * 1. We can't do it if child already exists (open has special treatment for
3156 * this case, but since we are inlined it's OK)
3157 * 2. We can't do it if dir is read-only (done in permission())
3158 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
3159 * 4. We should have write and exec permissions on dir
3160 * 5. We can't do it if dir is immutable (done in permission())
3161 */
3162static inline int may_create(struct mnt_idmap *idmap,
3163 struct inode *dir, struct dentry *child)
3164{
3165 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
3166 if (child->d_inode)
3167 return -EEXIST;
3168 if (IS_DEADDIR(dir))
3169 return -ENOENT;
3170 if (!fsuidgid_has_mapping(dir->i_sb, idmap))
3171 return -EOVERFLOW;
3172
3173 return inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
3174}
3175
3176// p1 != p2, both are on the same filesystem, ->s_vfs_rename_mutex is held
3177static struct dentry *lock_two_directories(struct dentry *p1, struct dentry *p2)
3178{
3179 struct dentry *p = p1, *q = p2, *r;
3180
3181 while ((r = p->d_parent) != p2 && r != p)
3182 p = r;
3183 if (r == p2) {
3184 // p is a child of p2 and an ancestor of p1 or p1 itself
3185 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
3186 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT2);
3187 return p;
3188 }
3189 // p is the root of connected component that contains p1
3190 // p2 does not occur on the path from p to p1
3191 while ((r = q->d_parent) != p1 && r != p && r != q)
3192 q = r;
3193 if (r == p1) {
3194 // q is a child of p1 and an ancestor of p2 or p2 itself
3195 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3196 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
3197 return q;
3198 } else if (likely(r == p)) {
3199 // both p2 and p1 are descendents of p
3200 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3201 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
3202 return NULL;
3203 } else { // no common ancestor at the time we'd been called
3204 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
3205 return ERR_PTR(-EXDEV);
3206 }
3207}
3208
3209/*
3210 * p1 and p2 should be directories on the same fs.
3211 */
3212struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
3213{
3214 if (p1 == p2) {
3215 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3216 return NULL;
3217 }
3218
3219 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
3220 return lock_two_directories(p1, p2);
3221}
3222EXPORT_SYMBOL(lock_rename);
3223
3224/*
3225 * c1 and p2 should be on the same fs.
3226 */
3227struct dentry *lock_rename_child(struct dentry *c1, struct dentry *p2)
3228{
3229 if (READ_ONCE(c1->d_parent) == p2) {
3230 /*
3231 * hopefully won't need to touch ->s_vfs_rename_mutex at all.
3232 */
3233 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
3234 /*
3235 * now that p2 is locked, nobody can move in or out of it,
3236 * so the test below is safe.
3237 */
3238 if (likely(c1->d_parent == p2))
3239 return NULL;
3240
3241 /*
3242 * c1 got moved out of p2 while we'd been taking locks;
3243 * unlock and fall back to slow case.
3244 */
3245 inode_unlock(p2->d_inode);
3246 }
3247
3248 mutex_lock(&c1->d_sb->s_vfs_rename_mutex);
3249 /*
3250 * nobody can move out of any directories on this fs.
3251 */
3252 if (likely(c1->d_parent != p2))
3253 return lock_two_directories(c1->d_parent, p2);
3254
3255 /*
3256 * c1 got moved into p2 while we were taking locks;
3257 * we need p2 locked and ->s_vfs_rename_mutex unlocked,
3258 * for consistency with lock_rename().
3259 */
3260 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
3261 mutex_unlock(&c1->d_sb->s_vfs_rename_mutex);
3262 return NULL;
3263}
3264EXPORT_SYMBOL(lock_rename_child);
3265
3266void unlock_rename(struct dentry *p1, struct dentry *p2)
3267{
3268 inode_unlock(p1->d_inode);
3269 if (p1 != p2) {
3270 inode_unlock(p2->d_inode);
3271 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
3272 }
3273}
3274EXPORT_SYMBOL(unlock_rename);
3275
3276/**
3277 * vfs_prepare_mode - prepare the mode to be used for a new inode
3278 * @idmap: idmap of the mount the inode was found from
3279 * @dir: parent directory of the new inode
3280 * @mode: mode of the new inode
3281 * @mask_perms: allowed permission by the vfs
3282 * @type: type of file to be created
3283 *
3284 * This helper consolidates and enforces vfs restrictions on the @mode of a new
3285 * object to be created.
3286 *
3287 * Umask stripping depends on whether the filesystem supports POSIX ACLs (see
3288 * the kernel documentation for mode_strip_umask()). Moving umask stripping
3289 * after setgid stripping allows the same ordering for both non-POSIX ACL and
3290 * POSIX ACL supporting filesystems.
3291 *
3292 * Note that it's currently valid for @type to be 0 if a directory is created.
3293 * Filesystems raise that flag individually and we need to check whether each
3294 * filesystem can deal with receiving S_IFDIR from the vfs before we enforce a
3295 * non-zero type.
3296 *
3297 * Returns: mode to be passed to the filesystem
3298 */
3299static inline umode_t vfs_prepare_mode(struct mnt_idmap *idmap,
3300 const struct inode *dir, umode_t mode,
3301 umode_t mask_perms, umode_t type)
3302{
3303 mode = mode_strip_sgid(idmap, dir, mode);
3304 mode = mode_strip_umask(dir, mode);
3305
3306 /*
3307 * Apply the vfs mandated allowed permission mask and set the type of
3308 * file to be created before we call into the filesystem.
3309 */
3310 mode &= (mask_perms & ~S_IFMT);
3311 mode |= (type & S_IFMT);
3312
3313 return mode;
3314}
3315
3316/**
3317 * vfs_create - create new file
3318 * @idmap: idmap of the mount the inode was found from
3319 * @dir: inode of the parent directory
3320 * @dentry: dentry of the child file
3321 * @mode: mode of the child file
3322 * @want_excl: whether the file must not yet exist
3323 *
3324 * Create a new file.
3325 *
3326 * If the inode has been found through an idmapped mount the idmap of
3327 * the vfsmount must be passed through @idmap. This function will then take
3328 * care to map the inode according to @idmap before checking permissions.
3329 * On non-idmapped mounts or if permission checking is to be performed on the
3330 * raw inode simply pass @nop_mnt_idmap.
3331 */
3332int vfs_create(struct mnt_idmap *idmap, struct inode *dir,
3333 struct dentry *dentry, umode_t mode, bool want_excl)
3334{
3335 int error;
3336
3337 error = may_create(idmap, dir, dentry);
3338 if (error)
3339 return error;
3340
3341 if (!dir->i_op->create)
3342 return -EACCES; /* shouldn't it be ENOSYS? */
3343
3344 mode = vfs_prepare_mode(idmap, dir, mode, S_IALLUGO, S_IFREG);
3345 error = security_inode_create(dir, dentry, mode);
3346 if (error)
3347 return error;
3348 error = dir->i_op->create(idmap, dir, dentry, mode, want_excl);
3349 if (!error)
3350 fsnotify_create(dir, dentry);
3351 return error;
3352}
3353EXPORT_SYMBOL(vfs_create);
3354
3355int vfs_mkobj(struct dentry *dentry, umode_t mode,
3356 int (*f)(struct dentry *, umode_t, void *),
3357 void *arg)
3358{
3359 struct inode *dir = dentry->d_parent->d_inode;
3360 int error = may_create(&nop_mnt_idmap, dir, dentry);
3361 if (error)
3362 return error;
3363
3364 mode &= S_IALLUGO;
3365 mode |= S_IFREG;
3366 error = security_inode_create(dir, dentry, mode);
3367 if (error)
3368 return error;
3369 error = f(dentry, mode, arg);
3370 if (!error)
3371 fsnotify_create(dir, dentry);
3372 return error;
3373}
3374EXPORT_SYMBOL(vfs_mkobj);
3375
3376bool may_open_dev(const struct path *path)
3377{
3378 return !(path->mnt->mnt_flags & MNT_NODEV) &&
3379 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
3380}
3381
3382static int may_open(struct mnt_idmap *idmap, const struct path *path,
3383 int acc_mode, int flag)
3384{
3385 struct dentry *dentry = path->dentry;
3386 struct inode *inode = dentry->d_inode;
3387 int error;
3388
3389 if (!inode)
3390 return -ENOENT;
3391
3392 switch (inode->i_mode & S_IFMT) {
3393 case S_IFLNK:
3394 return -ELOOP;
3395 case S_IFDIR:
3396 if (acc_mode & MAY_WRITE)
3397 return -EISDIR;
3398 if (acc_mode & MAY_EXEC)
3399 return -EACCES;
3400 break;
3401 case S_IFBLK:
3402 case S_IFCHR:
3403 if (!may_open_dev(path))
3404 return -EACCES;
3405 fallthrough;
3406 case S_IFIFO:
3407 case S_IFSOCK:
3408 if (acc_mode & MAY_EXEC)
3409 return -EACCES;
3410 flag &= ~O_TRUNC;
3411 break;
3412 case S_IFREG:
3413 if ((acc_mode & MAY_EXEC) && path_noexec(path))
3414 return -EACCES;
3415 break;
3416 }
3417
3418 error = inode_permission(idmap, inode, MAY_OPEN | acc_mode);
3419 if (error)
3420 return error;
3421
3422 /*
3423 * An append-only file must be opened in append mode for writing.
3424 */
3425 if (IS_APPEND(inode)) {
3426 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
3427 return -EPERM;
3428 if (flag & O_TRUNC)
3429 return -EPERM;
3430 }
3431
3432 /* O_NOATIME can only be set by the owner or superuser */
3433 if (flag & O_NOATIME && !inode_owner_or_capable(idmap, inode))
3434 return -EPERM;
3435
3436 return 0;
3437}
3438
3439static int handle_truncate(struct mnt_idmap *idmap, struct file *filp)
3440{
3441 const struct path *path = &filp->f_path;
3442 struct inode *inode = path->dentry->d_inode;
3443 int error = get_write_access(inode);
3444 if (error)
3445 return error;
3446
3447 error = security_file_truncate(filp);
3448 if (!error) {
3449 error = do_truncate(idmap, path->dentry, 0,
3450 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
3451 filp);
3452 }
3453 put_write_access(inode);
3454 return error;
3455}
3456
3457static inline int open_to_namei_flags(int flag)
3458{
3459 if ((flag & O_ACCMODE) == 3)
3460 flag--;
3461 return flag;
3462}
3463
3464static int may_o_create(struct mnt_idmap *idmap,
3465 const struct path *dir, struct dentry *dentry,
3466 umode_t mode)
3467{
3468 int error = security_path_mknod(dir, dentry, mode, 0);
3469 if (error)
3470 return error;
3471
3472 if (!fsuidgid_has_mapping(dir->dentry->d_sb, idmap))
3473 return -EOVERFLOW;
3474
3475 error = inode_permission(idmap, dir->dentry->d_inode,
3476 MAY_WRITE | MAY_EXEC);
3477 if (error)
3478 return error;
3479
3480 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3481}
3482
3483/*
3484 * Attempt to atomically look up, create and open a file from a negative
3485 * dentry.
3486 *
3487 * Returns 0 if successful. The file will have been created and attached to
3488 * @file by the filesystem calling finish_open().
3489 *
3490 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3491 * be set. The caller will need to perform the open themselves. @path will
3492 * have been updated to point to the new dentry. This may be negative.
3493 *
3494 * Returns an error code otherwise.
3495 */
3496static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry,
3497 struct file *file,
3498 int open_flag, umode_t mode)
3499{
3500 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3501 struct inode *dir = nd->path.dentry->d_inode;
3502 int error;
3503
3504 if (nd->flags & LOOKUP_DIRECTORY)
3505 open_flag |= O_DIRECTORY;
3506
3507 file->f_path.dentry = DENTRY_NOT_SET;
3508 file->f_path.mnt = nd->path.mnt;
3509 error = dir->i_op->atomic_open(dir, dentry, file,
3510 open_to_namei_flags(open_flag), mode);
3511 d_lookup_done(dentry);
3512 if (!error) {
3513 if (file->f_mode & FMODE_OPENED) {
3514 if (unlikely(dentry != file->f_path.dentry)) {
3515 dput(dentry);
3516 dentry = dget(file->f_path.dentry);
3517 }
3518 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3519 error = -EIO;
3520 } else {
3521 if (file->f_path.dentry) {
3522 dput(dentry);
3523 dentry = file->f_path.dentry;
3524 }
3525 if (unlikely(d_is_negative(dentry)))
3526 error = -ENOENT;
3527 }
3528 }
3529 if (error) {
3530 dput(dentry);
3531 dentry = ERR_PTR(error);
3532 }
3533 return dentry;
3534}
3535
3536/*
3537 * Look up and maybe create and open the last component.
3538 *
3539 * Must be called with parent locked (exclusive in O_CREAT case).
3540 *
3541 * Returns 0 on success, that is, if
3542 * the file was successfully atomically created (if necessary) and opened, or
3543 * the file was not completely opened at this time, though lookups and
3544 * creations were performed.
3545 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3546 * In the latter case dentry returned in @path might be negative if O_CREAT
3547 * hadn't been specified.
3548 *
3549 * An error code is returned on failure.
3550 */
3551static struct dentry *lookup_open(struct nameidata *nd, struct file *file,
3552 const struct open_flags *op,
3553 bool got_write)
3554{
3555 struct mnt_idmap *idmap;
3556 struct dentry *dir = nd->path.dentry;
3557 struct inode *dir_inode = dir->d_inode;
3558 int open_flag = op->open_flag;
3559 struct dentry *dentry;
3560 int error, create_error = 0;
3561 umode_t mode = op->mode;
3562 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3563
3564 if (unlikely(IS_DEADDIR(dir_inode)))
3565 return ERR_PTR(-ENOENT);
3566
3567 file->f_mode &= ~FMODE_CREATED;
3568 dentry = d_lookup(dir, &nd->last);
3569 for (;;) {
3570 if (!dentry) {
3571 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3572 if (IS_ERR(dentry))
3573 return dentry;
3574 }
3575 if (d_in_lookup(dentry))
3576 break;
3577
3578 error = d_revalidate(dentry, nd->flags);
3579 if (likely(error > 0))
3580 break;
3581 if (error)
3582 goto out_dput;
3583 d_invalidate(dentry);
3584 dput(dentry);
3585 dentry = NULL;
3586 }
3587 if (dentry->d_inode) {
3588 /* Cached positive dentry: will open in f_op->open */
3589 return dentry;
3590 }
3591
3592 if (open_flag & O_CREAT)
3593 audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
3594
3595 /*
3596 * Checking write permission is tricky, bacuse we don't know if we are
3597 * going to actually need it: O_CREAT opens should work as long as the
3598 * file exists. But checking existence breaks atomicity. The trick is
3599 * to check access and if not granted clear O_CREAT from the flags.
3600 *
3601 * Another problem is returing the "right" error value (e.g. for an
3602 * O_EXCL open we want to return EEXIST not EROFS).
3603 */
3604 if (unlikely(!got_write))
3605 open_flag &= ~O_TRUNC;
3606 idmap = mnt_idmap(nd->path.mnt);
3607 if (open_flag & O_CREAT) {
3608 if (open_flag & O_EXCL)
3609 open_flag &= ~O_TRUNC;
3610 mode = vfs_prepare_mode(idmap, dir->d_inode, mode, mode, mode);
3611 if (likely(got_write))
3612 create_error = may_o_create(idmap, &nd->path,
3613 dentry, mode);
3614 else
3615 create_error = -EROFS;
3616 }
3617 if (create_error)
3618 open_flag &= ~O_CREAT;
3619 if (dir_inode->i_op->atomic_open) {
3620 dentry = atomic_open(nd, dentry, file, open_flag, mode);
3621 if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT))
3622 dentry = ERR_PTR(create_error);
3623 return dentry;
3624 }
3625
3626 if (d_in_lookup(dentry)) {
3627 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3628 nd->flags);
3629 d_lookup_done(dentry);
3630 if (unlikely(res)) {
3631 if (IS_ERR(res)) {
3632 error = PTR_ERR(res);
3633 goto out_dput;
3634 }
3635 dput(dentry);
3636 dentry = res;
3637 }
3638 }
3639
3640 /* Negative dentry, just create the file */
3641 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3642 file->f_mode |= FMODE_CREATED;
3643 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3644 if (!dir_inode->i_op->create) {
3645 error = -EACCES;
3646 goto out_dput;
3647 }
3648
3649 error = dir_inode->i_op->create(idmap, dir_inode, dentry,
3650 mode, open_flag & O_EXCL);
3651 if (error)
3652 goto out_dput;
3653 }
3654 if (unlikely(create_error) && !dentry->d_inode) {
3655 error = create_error;
3656 goto out_dput;
3657 }
3658 return dentry;
3659
3660out_dput:
3661 dput(dentry);
3662 return ERR_PTR(error);
3663}
3664
3665static inline bool trailing_slashes(struct nameidata *nd)
3666{
3667 return (bool)nd->last.name[nd->last.len];
3668}
3669
3670static struct dentry *lookup_fast_for_open(struct nameidata *nd, int open_flag)
3671{
3672 struct dentry *dentry;
3673
3674 if (open_flag & O_CREAT) {
3675 if (trailing_slashes(nd))
3676 return ERR_PTR(-EISDIR);
3677
3678 /* Don't bother on an O_EXCL create */
3679 if (open_flag & O_EXCL)
3680 return NULL;
3681 }
3682
3683 if (trailing_slashes(nd))
3684 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3685
3686 dentry = lookup_fast(nd);
3687 if (IS_ERR_OR_NULL(dentry))
3688 return dentry;
3689
3690 if (open_flag & O_CREAT) {
3691 /* Discard negative dentries. Need inode_lock to do the create */
3692 if (!dentry->d_inode) {
3693 if (!(nd->flags & LOOKUP_RCU))
3694 dput(dentry);
3695 dentry = NULL;
3696 }
3697 }
3698 return dentry;
3699}
3700
3701static const char *open_last_lookups(struct nameidata *nd,
3702 struct file *file, const struct open_flags *op)
3703{
3704 struct dentry *dir = nd->path.dentry;
3705 int open_flag = op->open_flag;
3706 bool got_write = false;
3707 struct dentry *dentry;
3708 const char *res;
3709
3710 nd->flags |= op->intent;
3711
3712 if (nd->last_type != LAST_NORM) {
3713 if (nd->depth)
3714 put_link(nd);
3715 return handle_dots(nd, nd->last_type);
3716 }
3717
3718 /* We _can_ be in RCU mode here */
3719 dentry = lookup_fast_for_open(nd, open_flag);
3720 if (IS_ERR(dentry))
3721 return ERR_CAST(dentry);
3722
3723 if (likely(dentry))
3724 goto finish_lookup;
3725
3726 if (!(open_flag & O_CREAT)) {
3727 if (WARN_ON_ONCE(nd->flags & LOOKUP_RCU))
3728 return ERR_PTR(-ECHILD);
3729 } else {
3730 if (nd->flags & LOOKUP_RCU) {
3731 if (!try_to_unlazy(nd))
3732 return ERR_PTR(-ECHILD);
3733 }
3734 }
3735
3736 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3737 got_write = !mnt_want_write(nd->path.mnt);
3738 /*
3739 * do _not_ fail yet - we might not need that or fail with
3740 * a different error; let lookup_open() decide; we'll be
3741 * dropping this one anyway.
3742 */
3743 }
3744 if (open_flag & O_CREAT)
3745 inode_lock(dir->d_inode);
3746 else
3747 inode_lock_shared(dir->d_inode);
3748 dentry = lookup_open(nd, file, op, got_write);
3749 if (!IS_ERR(dentry)) {
3750 if (file->f_mode & FMODE_CREATED)
3751 fsnotify_create(dir->d_inode, dentry);
3752 if (file->f_mode & FMODE_OPENED)
3753 fsnotify_open(file);
3754 }
3755 if (open_flag & O_CREAT)
3756 inode_unlock(dir->d_inode);
3757 else
3758 inode_unlock_shared(dir->d_inode);
3759
3760 if (got_write)
3761 mnt_drop_write(nd->path.mnt);
3762
3763 if (IS_ERR(dentry))
3764 return ERR_CAST(dentry);
3765
3766 if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) {
3767 dput(nd->path.dentry);
3768 nd->path.dentry = dentry;
3769 return NULL;
3770 }
3771
3772finish_lookup:
3773 if (nd->depth)
3774 put_link(nd);
3775 res = step_into(nd, WALK_TRAILING, dentry);
3776 if (unlikely(res))
3777 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3778 return res;
3779}
3780
3781/*
3782 * Handle the last step of open()
3783 */
3784static int do_open(struct nameidata *nd,
3785 struct file *file, const struct open_flags *op)
3786{
3787 struct mnt_idmap *idmap;
3788 int open_flag = op->open_flag;
3789 bool do_truncate;
3790 int acc_mode;
3791 int error;
3792
3793 if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) {
3794 error = complete_walk(nd);
3795 if (error)
3796 return error;
3797 }
3798 if (!(file->f_mode & FMODE_CREATED))
3799 audit_inode(nd->name, nd->path.dentry, 0);
3800 idmap = mnt_idmap(nd->path.mnt);
3801 if (open_flag & O_CREAT) {
3802 if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED))
3803 return -EEXIST;
3804 if (d_is_dir(nd->path.dentry))
3805 return -EISDIR;
3806 error = may_create_in_sticky(idmap, nd,
3807 d_backing_inode(nd->path.dentry));
3808 if (unlikely(error))
3809 return error;
3810 }
3811 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3812 return -ENOTDIR;
3813
3814 do_truncate = false;
3815 acc_mode = op->acc_mode;
3816 if (file->f_mode & FMODE_CREATED) {
3817 /* Don't check for write permission, don't truncate */
3818 open_flag &= ~O_TRUNC;
3819 acc_mode = 0;
3820 } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) {
3821 error = mnt_want_write(nd->path.mnt);
3822 if (error)
3823 return error;
3824 do_truncate = true;
3825 }
3826 error = may_open(idmap, &nd->path, acc_mode, open_flag);
3827 if (!error && !(file->f_mode & FMODE_OPENED))
3828 error = vfs_open(&nd->path, file);
3829 if (!error)
3830 error = security_file_post_open(file, op->acc_mode);
3831 if (!error && do_truncate)
3832 error = handle_truncate(idmap, file);
3833 if (unlikely(error > 0)) {
3834 WARN_ON(1);
3835 error = -EINVAL;
3836 }
3837 if (do_truncate)
3838 mnt_drop_write(nd->path.mnt);
3839 return error;
3840}
3841
3842/**
3843 * vfs_tmpfile - create tmpfile
3844 * @idmap: idmap of the mount the inode was found from
3845 * @parentpath: pointer to the path of the base directory
3846 * @file: file descriptor of the new tmpfile
3847 * @mode: mode of the new tmpfile
3848 *
3849 * Create a temporary file.
3850 *
3851 * If the inode has been found through an idmapped mount the idmap of
3852 * the vfsmount must be passed through @idmap. This function will then take
3853 * care to map the inode according to @idmap before checking permissions.
3854 * On non-idmapped mounts or if permission checking is to be performed on the
3855 * raw inode simply pass @nop_mnt_idmap.
3856 */
3857int vfs_tmpfile(struct mnt_idmap *idmap,
3858 const struct path *parentpath,
3859 struct file *file, umode_t mode)
3860{
3861 struct dentry *child;
3862 struct inode *dir = d_inode(parentpath->dentry);
3863 struct inode *inode;
3864 int error;
3865 int open_flag = file->f_flags;
3866
3867 /* we want directory to be writable */
3868 error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
3869 if (error)
3870 return error;
3871 if (!dir->i_op->tmpfile)
3872 return -EOPNOTSUPP;
3873 child = d_alloc(parentpath->dentry, &slash_name);
3874 if (unlikely(!child))
3875 return -ENOMEM;
3876 file->f_path.mnt = parentpath->mnt;
3877 file->f_path.dentry = child;
3878 mode = vfs_prepare_mode(idmap, dir, mode, mode, mode);
3879 error = dir->i_op->tmpfile(idmap, dir, file, mode);
3880 dput(child);
3881 if (file->f_mode & FMODE_OPENED)
3882 fsnotify_open(file);
3883 if (error)
3884 return error;
3885 /* Don't check for other permissions, the inode was just created */
3886 error = may_open(idmap, &file->f_path, 0, file->f_flags);
3887 if (error)
3888 return error;
3889 inode = file_inode(file);
3890 if (!(open_flag & O_EXCL)) {
3891 spin_lock(&inode->i_lock);
3892 inode->i_state |= I_LINKABLE;
3893 spin_unlock(&inode->i_lock);
3894 }
3895 security_inode_post_create_tmpfile(idmap, inode);
3896 return 0;
3897}
3898
3899/**
3900 * kernel_tmpfile_open - open a tmpfile for kernel internal use
3901 * @idmap: idmap of the mount the inode was found from
3902 * @parentpath: path of the base directory
3903 * @mode: mode of the new tmpfile
3904 * @open_flag: flags
3905 * @cred: credentials for open
3906 *
3907 * Create and open a temporary file. The file is not accounted in nr_files,
3908 * hence this is only for kernel internal use, and must not be installed into
3909 * file tables or such.
3910 */
3911struct file *kernel_tmpfile_open(struct mnt_idmap *idmap,
3912 const struct path *parentpath,
3913 umode_t mode, int open_flag,
3914 const struct cred *cred)
3915{
3916 struct file *file;
3917 int error;
3918
3919 file = alloc_empty_file_noaccount(open_flag, cred);
3920 if (IS_ERR(file))
3921 return file;
3922
3923 error = vfs_tmpfile(idmap, parentpath, file, mode);
3924 if (error) {
3925 fput(file);
3926 file = ERR_PTR(error);
3927 }
3928 return file;
3929}
3930EXPORT_SYMBOL(kernel_tmpfile_open);
3931
3932static int do_tmpfile(struct nameidata *nd, unsigned flags,
3933 const struct open_flags *op,
3934 struct file *file)
3935{
3936 struct path path;
3937 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3938
3939 if (unlikely(error))
3940 return error;
3941 error = mnt_want_write(path.mnt);
3942 if (unlikely(error))
3943 goto out;
3944 error = vfs_tmpfile(mnt_idmap(path.mnt), &path, file, op->mode);
3945 if (error)
3946 goto out2;
3947 audit_inode(nd->name, file->f_path.dentry, 0);
3948out2:
3949 mnt_drop_write(path.mnt);
3950out:
3951 path_put(&path);
3952 return error;
3953}
3954
3955static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3956{
3957 struct path path;
3958 int error = path_lookupat(nd, flags, &path);
3959 if (!error) {
3960 audit_inode(nd->name, path.dentry, 0);
3961 error = vfs_open(&path, file);
3962 path_put(&path);
3963 }
3964 return error;
3965}
3966
3967static struct file *path_openat(struct nameidata *nd,
3968 const struct open_flags *op, unsigned flags)
3969{
3970 struct file *file;
3971 int error;
3972
3973 file = alloc_empty_file(op->open_flag, current_cred());
3974 if (IS_ERR(file))
3975 return file;
3976
3977 if (unlikely(file->f_flags & __O_TMPFILE)) {
3978 error = do_tmpfile(nd, flags, op, file);
3979 } else if (unlikely(file->f_flags & O_PATH)) {
3980 error = do_o_path(nd, flags, file);
3981 } else {
3982 const char *s = path_init(nd, flags);
3983 while (!(error = link_path_walk(s, nd)) &&
3984 (s = open_last_lookups(nd, file, op)) != NULL)
3985 ;
3986 if (!error)
3987 error = do_open(nd, file, op);
3988 terminate_walk(nd);
3989 }
3990 if (likely(!error)) {
3991 if (likely(file->f_mode & FMODE_OPENED))
3992 return file;
3993 WARN_ON(1);
3994 error = -EINVAL;
3995 }
3996 fput(file);
3997 if (error == -EOPENSTALE) {
3998 if (flags & LOOKUP_RCU)
3999 error = -ECHILD;
4000 else
4001 error = -ESTALE;
4002 }
4003 return ERR_PTR(error);
4004}
4005
4006struct file *do_filp_open(int dfd, struct filename *pathname,
4007 const struct open_flags *op)
4008{
4009 struct nameidata nd;
4010 int flags = op->lookup_flags;
4011 struct file *filp;
4012
4013 set_nameidata(&nd, dfd, pathname, NULL);
4014 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
4015 if (unlikely(filp == ERR_PTR(-ECHILD)))
4016 filp = path_openat(&nd, op, flags);
4017 if (unlikely(filp == ERR_PTR(-ESTALE)))
4018 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
4019 restore_nameidata();
4020 return filp;
4021}
4022
4023struct file *do_file_open_root(const struct path *root,
4024 const char *name, const struct open_flags *op)
4025{
4026 struct nameidata nd;
4027 struct file *file;
4028 struct filename *filename;
4029 int flags = op->lookup_flags;
4030
4031 if (d_is_symlink(root->dentry) && op->intent & LOOKUP_OPEN)
4032 return ERR_PTR(-ELOOP);
4033
4034 filename = getname_kernel(name);
4035 if (IS_ERR(filename))
4036 return ERR_CAST(filename);
4037
4038 set_nameidata(&nd, -1, filename, root);
4039 file = path_openat(&nd, op, flags | LOOKUP_RCU);
4040 if (unlikely(file == ERR_PTR(-ECHILD)))
4041 file = path_openat(&nd, op, flags);
4042 if (unlikely(file == ERR_PTR(-ESTALE)))
4043 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
4044 restore_nameidata();
4045 putname(filename);
4046 return file;
4047}
4048
4049static struct dentry *filename_create(int dfd, struct filename *name,
4050 struct path *path, unsigned int lookup_flags)
4051{
4052 struct dentry *dentry = ERR_PTR(-EEXIST);
4053 struct qstr last;
4054 bool want_dir = lookup_flags & LOOKUP_DIRECTORY;
4055 unsigned int reval_flag = lookup_flags & LOOKUP_REVAL;
4056 unsigned int create_flags = LOOKUP_CREATE | LOOKUP_EXCL;
4057 int type;
4058 int err2;
4059 int error;
4060
4061 error = filename_parentat(dfd, name, reval_flag, path, &last, &type);
4062 if (error)
4063 return ERR_PTR(error);
4064
4065 /*
4066 * Yucky last component or no last component at all?
4067 * (foo/., foo/.., /////)
4068 */
4069 if (unlikely(type != LAST_NORM))
4070 goto out;
4071
4072 /* don't fail immediately if it's r/o, at least try to report other errors */
4073 err2 = mnt_want_write(path->mnt);
4074 /*
4075 * Do the final lookup. Suppress 'create' if there is a trailing
4076 * '/', and a directory wasn't requested.
4077 */
4078 if (last.name[last.len] && !want_dir)
4079 create_flags = 0;
4080 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
4081 dentry = lookup_one_qstr_excl(&last, path->dentry,
4082 reval_flag | create_flags);
4083 if (IS_ERR(dentry))
4084 goto unlock;
4085
4086 error = -EEXIST;
4087 if (d_is_positive(dentry))
4088 goto fail;
4089
4090 /*
4091 * Special case - lookup gave negative, but... we had foo/bar/
4092 * From the vfs_mknod() POV we just have a negative dentry -
4093 * all is fine. Let's be bastards - you had / on the end, you've
4094 * been asking for (non-existent) directory. -ENOENT for you.
4095 */
4096 if (unlikely(!create_flags)) {
4097 error = -ENOENT;
4098 goto fail;
4099 }
4100 if (unlikely(err2)) {
4101 error = err2;
4102 goto fail;
4103 }
4104 return dentry;
4105fail:
4106 dput(dentry);
4107 dentry = ERR_PTR(error);
4108unlock:
4109 inode_unlock(path->dentry->d_inode);
4110 if (!err2)
4111 mnt_drop_write(path->mnt);
4112out:
4113 path_put(path);
4114 return dentry;
4115}
4116
4117struct dentry *kern_path_create(int dfd, const char *pathname,
4118 struct path *path, unsigned int lookup_flags)
4119{
4120 struct filename *filename = getname_kernel(pathname);
4121 struct dentry *res = filename_create(dfd, filename, path, lookup_flags);
4122
4123 putname(filename);
4124 return res;
4125}
4126EXPORT_SYMBOL(kern_path_create);
4127
4128void done_path_create(struct path *path, struct dentry *dentry)
4129{
4130 dput(dentry);
4131 inode_unlock(path->dentry->d_inode);
4132 mnt_drop_write(path->mnt);
4133 path_put(path);
4134}
4135EXPORT_SYMBOL(done_path_create);
4136
4137inline struct dentry *user_path_create(int dfd, const char __user *pathname,
4138 struct path *path, unsigned int lookup_flags)
4139{
4140 struct filename *filename = getname(pathname);
4141 struct dentry *res = filename_create(dfd, filename, path, lookup_flags);
4142
4143 putname(filename);
4144 return res;
4145}
4146EXPORT_SYMBOL(user_path_create);
4147
4148/**
4149 * vfs_mknod - create device node or file
4150 * @idmap: idmap of the mount the inode was found from
4151 * @dir: inode of the parent directory
4152 * @dentry: dentry of the child device node
4153 * @mode: mode of the child device node
4154 * @dev: device number of device to create
4155 *
4156 * Create a device node or file.
4157 *
4158 * If the inode has been found through an idmapped mount the idmap of
4159 * the vfsmount must be passed through @idmap. This function will then take
4160 * care to map the inode according to @idmap before checking permissions.
4161 * On non-idmapped mounts or if permission checking is to be performed on the
4162 * raw inode simply pass @nop_mnt_idmap.
4163 */
4164int vfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
4165 struct dentry *dentry, umode_t mode, dev_t dev)
4166{
4167 bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV;
4168 int error = may_create(idmap, dir, dentry);
4169
4170 if (error)
4171 return error;
4172
4173 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout &&
4174 !capable(CAP_MKNOD))
4175 return -EPERM;
4176
4177 if (!dir->i_op->mknod)
4178 return -EPERM;
4179
4180 mode = vfs_prepare_mode(idmap, dir, mode, mode, mode);
4181 error = devcgroup_inode_mknod(mode, dev);
4182 if (error)
4183 return error;
4184
4185 error = security_inode_mknod(dir, dentry, mode, dev);
4186 if (error)
4187 return error;
4188
4189 error = dir->i_op->mknod(idmap, dir, dentry, mode, dev);
4190 if (!error)
4191 fsnotify_create(dir, dentry);
4192 return error;
4193}
4194EXPORT_SYMBOL(vfs_mknod);
4195
4196static int may_mknod(umode_t mode)
4197{
4198 switch (mode & S_IFMT) {
4199 case S_IFREG:
4200 case S_IFCHR:
4201 case S_IFBLK:
4202 case S_IFIFO:
4203 case S_IFSOCK:
4204 case 0: /* zero mode translates to S_IFREG */
4205 return 0;
4206 case S_IFDIR:
4207 return -EPERM;
4208 default:
4209 return -EINVAL;
4210 }
4211}
4212
4213static int do_mknodat(int dfd, struct filename *name, umode_t mode,
4214 unsigned int dev)
4215{
4216 struct mnt_idmap *idmap;
4217 struct dentry *dentry;
4218 struct path path;
4219 int error;
4220 unsigned int lookup_flags = 0;
4221
4222 error = may_mknod(mode);
4223 if (error)
4224 goto out1;
4225retry:
4226 dentry = filename_create(dfd, name, &path, lookup_flags);
4227 error = PTR_ERR(dentry);
4228 if (IS_ERR(dentry))
4229 goto out1;
4230
4231 error = security_path_mknod(&path, dentry,
4232 mode_strip_umask(path.dentry->d_inode, mode), dev);
4233 if (error)
4234 goto out2;
4235
4236 idmap = mnt_idmap(path.mnt);
4237 switch (mode & S_IFMT) {
4238 case 0: case S_IFREG:
4239 error = vfs_create(idmap, path.dentry->d_inode,
4240 dentry, mode, true);
4241 if (!error)
4242 security_path_post_mknod(idmap, dentry);
4243 break;
4244 case S_IFCHR: case S_IFBLK:
4245 error = vfs_mknod(idmap, path.dentry->d_inode,
4246 dentry, mode, new_decode_dev(dev));
4247 break;
4248 case S_IFIFO: case S_IFSOCK:
4249 error = vfs_mknod(idmap, path.dentry->d_inode,
4250 dentry, mode, 0);
4251 break;
4252 }
4253out2:
4254 done_path_create(&path, dentry);
4255 if (retry_estale(error, lookup_flags)) {
4256 lookup_flags |= LOOKUP_REVAL;
4257 goto retry;
4258 }
4259out1:
4260 putname(name);
4261 return error;
4262}
4263
4264SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
4265 unsigned int, dev)
4266{
4267 return do_mknodat(dfd, getname(filename), mode, dev);
4268}
4269
4270SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
4271{
4272 return do_mknodat(AT_FDCWD, getname(filename), mode, dev);
4273}
4274
4275/**
4276 * vfs_mkdir - create directory
4277 * @idmap: idmap of the mount the inode was found from
4278 * @dir: inode of the parent directory
4279 * @dentry: dentry of the child directory
4280 * @mode: mode of the child directory
4281 *
4282 * Create a directory.
4283 *
4284 * If the inode has been found through an idmapped mount the idmap of
4285 * the vfsmount must be passed through @idmap. This function will then take
4286 * care to map the inode according to @idmap before checking permissions.
4287 * On non-idmapped mounts or if permission checking is to be performed on the
4288 * raw inode simply pass @nop_mnt_idmap.
4289 */
4290int vfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
4291 struct dentry *dentry, umode_t mode)
4292{
4293 int error;
4294 unsigned max_links = dir->i_sb->s_max_links;
4295
4296 error = may_create(idmap, dir, dentry);
4297 if (error)
4298 return error;
4299
4300 if (!dir->i_op->mkdir)
4301 return -EPERM;
4302
4303 mode = vfs_prepare_mode(idmap, dir, mode, S_IRWXUGO | S_ISVTX, 0);
4304 error = security_inode_mkdir(dir, dentry, mode);
4305 if (error)
4306 return error;
4307
4308 if (max_links && dir->i_nlink >= max_links)
4309 return -EMLINK;
4310
4311 error = dir->i_op->mkdir(idmap, dir, dentry, mode);
4312 if (!error)
4313 fsnotify_mkdir(dir, dentry);
4314 return error;
4315}
4316EXPORT_SYMBOL(vfs_mkdir);
4317
4318int do_mkdirat(int dfd, struct filename *name, umode_t mode)
4319{
4320 struct dentry *dentry;
4321 struct path path;
4322 int error;
4323 unsigned int lookup_flags = LOOKUP_DIRECTORY;
4324
4325retry:
4326 dentry = filename_create(dfd, name, &path, lookup_flags);
4327 error = PTR_ERR(dentry);
4328 if (IS_ERR(dentry))
4329 goto out_putname;
4330
4331 error = security_path_mkdir(&path, dentry,
4332 mode_strip_umask(path.dentry->d_inode, mode));
4333 if (!error) {
4334 error = vfs_mkdir(mnt_idmap(path.mnt), path.dentry->d_inode,
4335 dentry, mode);
4336 }
4337 done_path_create(&path, dentry);
4338 if (retry_estale(error, lookup_flags)) {
4339 lookup_flags |= LOOKUP_REVAL;
4340 goto retry;
4341 }
4342out_putname:
4343 putname(name);
4344 return error;
4345}
4346
4347SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
4348{
4349 return do_mkdirat(dfd, getname(pathname), mode);
4350}
4351
4352SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
4353{
4354 return do_mkdirat(AT_FDCWD, getname(pathname), mode);
4355}
4356
4357/**
4358 * vfs_rmdir - remove directory
4359 * @idmap: idmap of the mount the inode was found from
4360 * @dir: inode of the parent directory
4361 * @dentry: dentry of the child directory
4362 *
4363 * Remove a directory.
4364 *
4365 * If the inode has been found through an idmapped mount the idmap of
4366 * the vfsmount must be passed through @idmap. This function will then take
4367 * care to map the inode according to @idmap before checking permissions.
4368 * On non-idmapped mounts or if permission checking is to be performed on the
4369 * raw inode simply pass @nop_mnt_idmap.
4370 */
4371int vfs_rmdir(struct mnt_idmap *idmap, struct inode *dir,
4372 struct dentry *dentry)
4373{
4374 int error = may_delete(idmap, dir, dentry, 1);
4375
4376 if (error)
4377 return error;
4378
4379 if (!dir->i_op->rmdir)
4380 return -EPERM;
4381
4382 dget(dentry);
4383 inode_lock(dentry->d_inode);
4384
4385 error = -EBUSY;
4386 if (is_local_mountpoint(dentry) ||
4387 (dentry->d_inode->i_flags & S_KERNEL_FILE))
4388 goto out;
4389
4390 error = security_inode_rmdir(dir, dentry);
4391 if (error)
4392 goto out;
4393
4394 error = dir->i_op->rmdir(dir, dentry);
4395 if (error)
4396 goto out;
4397
4398 shrink_dcache_parent(dentry);
4399 dentry->d_inode->i_flags |= S_DEAD;
4400 dont_mount(dentry);
4401 detach_mounts(dentry);
4402
4403out:
4404 inode_unlock(dentry->d_inode);
4405 dput(dentry);
4406 if (!error)
4407 d_delete_notify(dir, dentry);
4408 return error;
4409}
4410EXPORT_SYMBOL(vfs_rmdir);
4411
4412int do_rmdir(int dfd, struct filename *name)
4413{
4414 int error;
4415 struct dentry *dentry;
4416 struct path path;
4417 struct qstr last;
4418 int type;
4419 unsigned int lookup_flags = 0;
4420retry:
4421 error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4422 if (error)
4423 goto exit1;
4424
4425 switch (type) {
4426 case LAST_DOTDOT:
4427 error = -ENOTEMPTY;
4428 goto exit2;
4429 case LAST_DOT:
4430 error = -EINVAL;
4431 goto exit2;
4432 case LAST_ROOT:
4433 error = -EBUSY;
4434 goto exit2;
4435 }
4436
4437 error = mnt_want_write(path.mnt);
4438 if (error)
4439 goto exit2;
4440
4441 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4442 dentry = lookup_one_qstr_excl(&last, path.dentry, lookup_flags);
4443 error = PTR_ERR(dentry);
4444 if (IS_ERR(dentry))
4445 goto exit3;
4446 if (!dentry->d_inode) {
4447 error = -ENOENT;
4448 goto exit4;
4449 }
4450 error = security_path_rmdir(&path, dentry);
4451 if (error)
4452 goto exit4;
4453 error = vfs_rmdir(mnt_idmap(path.mnt), path.dentry->d_inode, dentry);
4454exit4:
4455 dput(dentry);
4456exit3:
4457 inode_unlock(path.dentry->d_inode);
4458 mnt_drop_write(path.mnt);
4459exit2:
4460 path_put(&path);
4461 if (retry_estale(error, lookup_flags)) {
4462 lookup_flags |= LOOKUP_REVAL;
4463 goto retry;
4464 }
4465exit1:
4466 putname(name);
4467 return error;
4468}
4469
4470SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
4471{
4472 return do_rmdir(AT_FDCWD, getname(pathname));
4473}
4474
4475/**
4476 * vfs_unlink - unlink a filesystem object
4477 * @idmap: idmap of the mount the inode was found from
4478 * @dir: parent directory
4479 * @dentry: victim
4480 * @delegated_inode: returns victim inode, if the inode is delegated.
4481 *
4482 * The caller must hold dir->i_mutex.
4483 *
4484 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
4485 * return a reference to the inode in delegated_inode. The caller
4486 * should then break the delegation on that inode and retry. Because
4487 * breaking a delegation may take a long time, the caller should drop
4488 * dir->i_mutex before doing so.
4489 *
4490 * Alternatively, a caller may pass NULL for delegated_inode. This may
4491 * be appropriate for callers that expect the underlying filesystem not
4492 * to be NFS exported.
4493 *
4494 * If the inode has been found through an idmapped mount the idmap of
4495 * the vfsmount must be passed through @idmap. This function will then take
4496 * care to map the inode according to @idmap before checking permissions.
4497 * On non-idmapped mounts or if permission checking is to be performed on the
4498 * raw inode simply pass @nop_mnt_idmap.
4499 */
4500int vfs_unlink(struct mnt_idmap *idmap, struct inode *dir,
4501 struct dentry *dentry, struct inode **delegated_inode)
4502{
4503 struct inode *target = dentry->d_inode;
4504 int error = may_delete(idmap, dir, dentry, 0);
4505
4506 if (error)
4507 return error;
4508
4509 if (!dir->i_op->unlink)
4510 return -EPERM;
4511
4512 inode_lock(target);
4513 if (IS_SWAPFILE(target))
4514 error = -EPERM;
4515 else if (is_local_mountpoint(dentry))
4516 error = -EBUSY;
4517 else {
4518 error = security_inode_unlink(dir, dentry);
4519 if (!error) {
4520 error = try_break_deleg(target, delegated_inode);
4521 if (error)
4522 goto out;
4523 error = dir->i_op->unlink(dir, dentry);
4524 if (!error) {
4525 dont_mount(dentry);
4526 detach_mounts(dentry);
4527 }
4528 }
4529 }
4530out:
4531 inode_unlock(target);
4532
4533 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
4534 if (!error && dentry->d_flags & DCACHE_NFSFS_RENAMED) {
4535 fsnotify_unlink(dir, dentry);
4536 } else if (!error) {
4537 fsnotify_link_count(target);
4538 d_delete_notify(dir, dentry);
4539 }
4540
4541 return error;
4542}
4543EXPORT_SYMBOL(vfs_unlink);
4544
4545/*
4546 * Make sure that the actual truncation of the file will occur outside its
4547 * directory's i_mutex. Truncate can take a long time if there is a lot of
4548 * writeout happening, and we don't want to prevent access to the directory
4549 * while waiting on the I/O.
4550 */
4551int do_unlinkat(int dfd, struct filename *name)
4552{
4553 int error;
4554 struct dentry *dentry;
4555 struct path path;
4556 struct qstr last;
4557 int type;
4558 struct inode *inode = NULL;
4559 struct inode *delegated_inode = NULL;
4560 unsigned int lookup_flags = 0;
4561retry:
4562 error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4563 if (error)
4564 goto exit1;
4565
4566 error = -EISDIR;
4567 if (type != LAST_NORM)
4568 goto exit2;
4569
4570 error = mnt_want_write(path.mnt);
4571 if (error)
4572 goto exit2;
4573retry_deleg:
4574 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4575 dentry = lookup_one_qstr_excl(&last, path.dentry, lookup_flags);
4576 error = PTR_ERR(dentry);
4577 if (!IS_ERR(dentry)) {
4578
4579 /* Why not before? Because we want correct error value */
4580 if (last.name[last.len] || d_is_negative(dentry))
4581 goto slashes;
4582 inode = dentry->d_inode;
4583 ihold(inode);
4584 error = security_path_unlink(&path, dentry);
4585 if (error)
4586 goto exit3;
4587 error = vfs_unlink(mnt_idmap(path.mnt), path.dentry->d_inode,
4588 dentry, &delegated_inode);
4589exit3:
4590 dput(dentry);
4591 }
4592 inode_unlock(path.dentry->d_inode);
4593 if (inode)
4594 iput(inode); /* truncate the inode here */
4595 inode = NULL;
4596 if (delegated_inode) {
4597 error = break_deleg_wait(&delegated_inode);
4598 if (!error)
4599 goto retry_deleg;
4600 }
4601 mnt_drop_write(path.mnt);
4602exit2:
4603 path_put(&path);
4604 if (retry_estale(error, lookup_flags)) {
4605 lookup_flags |= LOOKUP_REVAL;
4606 inode = NULL;
4607 goto retry;
4608 }
4609exit1:
4610 putname(name);
4611 return error;
4612
4613slashes:
4614 if (d_is_negative(dentry))
4615 error = -ENOENT;
4616 else if (d_is_dir(dentry))
4617 error = -EISDIR;
4618 else
4619 error = -ENOTDIR;
4620 goto exit3;
4621}
4622
4623SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4624{
4625 if ((flag & ~AT_REMOVEDIR) != 0)
4626 return -EINVAL;
4627
4628 if (flag & AT_REMOVEDIR)
4629 return do_rmdir(dfd, getname(pathname));
4630 return do_unlinkat(dfd, getname(pathname));
4631}
4632
4633SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4634{
4635 return do_unlinkat(AT_FDCWD, getname(pathname));
4636}
4637
4638/**
4639 * vfs_symlink - create symlink
4640 * @idmap: idmap of the mount the inode was found from
4641 * @dir: inode of the parent directory
4642 * @dentry: dentry of the child symlink file
4643 * @oldname: name of the file to link to
4644 *
4645 * Create a symlink.
4646 *
4647 * If the inode has been found through an idmapped mount the idmap of
4648 * the vfsmount must be passed through @idmap. This function will then take
4649 * care to map the inode according to @idmap before checking permissions.
4650 * On non-idmapped mounts or if permission checking is to be performed on the
4651 * raw inode simply pass @nop_mnt_idmap.
4652 */
4653int vfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
4654 struct dentry *dentry, const char *oldname)
4655{
4656 int error;
4657
4658 error = may_create(idmap, dir, dentry);
4659 if (error)
4660 return error;
4661
4662 if (!dir->i_op->symlink)
4663 return -EPERM;
4664
4665 error = security_inode_symlink(dir, dentry, oldname);
4666 if (error)
4667 return error;
4668
4669 error = dir->i_op->symlink(idmap, dir, dentry, oldname);
4670 if (!error)
4671 fsnotify_create(dir, dentry);
4672 return error;
4673}
4674EXPORT_SYMBOL(vfs_symlink);
4675
4676int do_symlinkat(struct filename *from, int newdfd, struct filename *to)
4677{
4678 int error;
4679 struct dentry *dentry;
4680 struct path path;
4681 unsigned int lookup_flags = 0;
4682
4683 if (IS_ERR(from)) {
4684 error = PTR_ERR(from);
4685 goto out_putnames;
4686 }
4687retry:
4688 dentry = filename_create(newdfd, to, &path, lookup_flags);
4689 error = PTR_ERR(dentry);
4690 if (IS_ERR(dentry))
4691 goto out_putnames;
4692
4693 error = security_path_symlink(&path, dentry, from->name);
4694 if (!error)
4695 error = vfs_symlink(mnt_idmap(path.mnt), path.dentry->d_inode,
4696 dentry, from->name);
4697 done_path_create(&path, dentry);
4698 if (retry_estale(error, lookup_flags)) {
4699 lookup_flags |= LOOKUP_REVAL;
4700 goto retry;
4701 }
4702out_putnames:
4703 putname(to);
4704 putname(from);
4705 return error;
4706}
4707
4708SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4709 int, newdfd, const char __user *, newname)
4710{
4711 return do_symlinkat(getname(oldname), newdfd, getname(newname));
4712}
4713
4714SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4715{
4716 return do_symlinkat(getname(oldname), AT_FDCWD, getname(newname));
4717}
4718
4719/**
4720 * vfs_link - create a new link
4721 * @old_dentry: object to be linked
4722 * @idmap: idmap of the mount
4723 * @dir: new parent
4724 * @new_dentry: where to create the new link
4725 * @delegated_inode: returns inode needing a delegation break
4726 *
4727 * The caller must hold dir->i_mutex
4728 *
4729 * If vfs_link discovers a delegation on the to-be-linked file in need
4730 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4731 * inode in delegated_inode. The caller should then break the delegation
4732 * and retry. Because breaking a delegation may take a long time, the
4733 * caller should drop the i_mutex before doing so.
4734 *
4735 * Alternatively, a caller may pass NULL for delegated_inode. This may
4736 * be appropriate for callers that expect the underlying filesystem not
4737 * to be NFS exported.
4738 *
4739 * If the inode has been found through an idmapped mount the idmap of
4740 * the vfsmount must be passed through @idmap. This function will then take
4741 * care to map the inode according to @idmap before checking permissions.
4742 * On non-idmapped mounts or if permission checking is to be performed on the
4743 * raw inode simply pass @nop_mnt_idmap.
4744 */
4745int vfs_link(struct dentry *old_dentry, struct mnt_idmap *idmap,
4746 struct inode *dir, struct dentry *new_dentry,
4747 struct inode **delegated_inode)
4748{
4749 struct inode *inode = old_dentry->d_inode;
4750 unsigned max_links = dir->i_sb->s_max_links;
4751 int error;
4752
4753 if (!inode)
4754 return -ENOENT;
4755
4756 error = may_create(idmap, dir, new_dentry);
4757 if (error)
4758 return error;
4759
4760 if (dir->i_sb != inode->i_sb)
4761 return -EXDEV;
4762
4763 /*
4764 * A link to an append-only or immutable file cannot be created.
4765 */
4766 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4767 return -EPERM;
4768 /*
4769 * Updating the link count will likely cause i_uid and i_gid to
4770 * be writen back improperly if their true value is unknown to
4771 * the vfs.
4772 */
4773 if (HAS_UNMAPPED_ID(idmap, inode))
4774 return -EPERM;
4775 if (!dir->i_op->link)
4776 return -EPERM;
4777 if (S_ISDIR(inode->i_mode))
4778 return -EPERM;
4779
4780 error = security_inode_link(old_dentry, dir, new_dentry);
4781 if (error)
4782 return error;
4783
4784 inode_lock(inode);
4785 /* Make sure we don't allow creating hardlink to an unlinked file */
4786 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4787 error = -ENOENT;
4788 else if (max_links && inode->i_nlink >= max_links)
4789 error = -EMLINK;
4790 else {
4791 error = try_break_deleg(inode, delegated_inode);
4792 if (!error)
4793 error = dir->i_op->link(old_dentry, dir, new_dentry);
4794 }
4795
4796 if (!error && (inode->i_state & I_LINKABLE)) {
4797 spin_lock(&inode->i_lock);
4798 inode->i_state &= ~I_LINKABLE;
4799 spin_unlock(&inode->i_lock);
4800 }
4801 inode_unlock(inode);
4802 if (!error)
4803 fsnotify_link(dir, inode, new_dentry);
4804 return error;
4805}
4806EXPORT_SYMBOL(vfs_link);
4807
4808/*
4809 * Hardlinks are often used in delicate situations. We avoid
4810 * security-related surprises by not following symlinks on the
4811 * newname. --KAB
4812 *
4813 * We don't follow them on the oldname either to be compatible
4814 * with linux 2.0, and to avoid hard-linking to directories
4815 * and other special files. --ADM
4816 */
4817int do_linkat(int olddfd, struct filename *old, int newdfd,
4818 struct filename *new, int flags)
4819{
4820 struct mnt_idmap *idmap;
4821 struct dentry *new_dentry;
4822 struct path old_path, new_path;
4823 struct inode *delegated_inode = NULL;
4824 int how = 0;
4825 int error;
4826
4827 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) {
4828 error = -EINVAL;
4829 goto out_putnames;
4830 }
4831 /*
4832 * To use null names we require CAP_DAC_READ_SEARCH or
4833 * that the open-time creds of the dfd matches current.
4834 * This ensures that not everyone will be able to create
4835 * a hardlink using the passed file descriptor.
4836 */
4837 if (flags & AT_EMPTY_PATH)
4838 how |= LOOKUP_LINKAT_EMPTY;
4839
4840 if (flags & AT_SYMLINK_FOLLOW)
4841 how |= LOOKUP_FOLLOW;
4842retry:
4843 error = filename_lookup(olddfd, old, how, &old_path, NULL);
4844 if (error)
4845 goto out_putnames;
4846
4847 new_dentry = filename_create(newdfd, new, &new_path,
4848 (how & LOOKUP_REVAL));
4849 error = PTR_ERR(new_dentry);
4850 if (IS_ERR(new_dentry))
4851 goto out_putpath;
4852
4853 error = -EXDEV;
4854 if (old_path.mnt != new_path.mnt)
4855 goto out_dput;
4856 idmap = mnt_idmap(new_path.mnt);
4857 error = may_linkat(idmap, &old_path);
4858 if (unlikely(error))
4859 goto out_dput;
4860 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4861 if (error)
4862 goto out_dput;
4863 error = vfs_link(old_path.dentry, idmap, new_path.dentry->d_inode,
4864 new_dentry, &delegated_inode);
4865out_dput:
4866 done_path_create(&new_path, new_dentry);
4867 if (delegated_inode) {
4868 error = break_deleg_wait(&delegated_inode);
4869 if (!error) {
4870 path_put(&old_path);
4871 goto retry;
4872 }
4873 }
4874 if (retry_estale(error, how)) {
4875 path_put(&old_path);
4876 how |= LOOKUP_REVAL;
4877 goto retry;
4878 }
4879out_putpath:
4880 path_put(&old_path);
4881out_putnames:
4882 putname(old);
4883 putname(new);
4884
4885 return error;
4886}
4887
4888SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4889 int, newdfd, const char __user *, newname, int, flags)
4890{
4891 return do_linkat(olddfd, getname_uflags(oldname, flags),
4892 newdfd, getname(newname), flags);
4893}
4894
4895SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4896{
4897 return do_linkat(AT_FDCWD, getname(oldname), AT_FDCWD, getname(newname), 0);
4898}
4899
4900/**
4901 * vfs_rename - rename a filesystem object
4902 * @rd: pointer to &struct renamedata info
4903 *
4904 * The caller must hold multiple mutexes--see lock_rename()).
4905 *
4906 * If vfs_rename discovers a delegation in need of breaking at either
4907 * the source or destination, it will return -EWOULDBLOCK and return a
4908 * reference to the inode in delegated_inode. The caller should then
4909 * break the delegation and retry. Because breaking a delegation may
4910 * take a long time, the caller should drop all locks before doing
4911 * so.
4912 *
4913 * Alternatively, a caller may pass NULL for delegated_inode. This may
4914 * be appropriate for callers that expect the underlying filesystem not
4915 * to be NFS exported.
4916 *
4917 * The worst of all namespace operations - renaming directory. "Perverted"
4918 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4919 * Problems:
4920 *
4921 * a) we can get into loop creation.
4922 * b) race potential - two innocent renames can create a loop together.
4923 * That's where 4.4BSD screws up. Current fix: serialization on
4924 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4925 * story.
4926 * c) we may have to lock up to _four_ objects - parents and victim (if it exists),
4927 * and source (if it's a non-directory or a subdirectory that moves to
4928 * different parent).
4929 * And that - after we got ->i_mutex on parents (until then we don't know
4930 * whether the target exists). Solution: try to be smart with locking
4931 * order for inodes. We rely on the fact that tree topology may change
4932 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4933 * move will be locked. Thus we can rank directories by the tree
4934 * (ancestors first) and rank all non-directories after them.
4935 * That works since everybody except rename does "lock parent, lookup,
4936 * lock child" and rename is under ->s_vfs_rename_mutex.
4937 * HOWEVER, it relies on the assumption that any object with ->lookup()
4938 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4939 * we'd better make sure that there's no link(2) for them.
4940 * d) conversion from fhandle to dentry may come in the wrong moment - when
4941 * we are removing the target. Solution: we will have to grab ->i_mutex
4942 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4943 * ->i_mutex on parents, which works but leads to some truly excessive
4944 * locking].
4945 */
4946int vfs_rename(struct renamedata *rd)
4947{
4948 int error;
4949 struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir;
4950 struct dentry *old_dentry = rd->old_dentry;
4951 struct dentry *new_dentry = rd->new_dentry;
4952 struct inode **delegated_inode = rd->delegated_inode;
4953 unsigned int flags = rd->flags;
4954 bool is_dir = d_is_dir(old_dentry);
4955 struct inode *source = old_dentry->d_inode;
4956 struct inode *target = new_dentry->d_inode;
4957 bool new_is_dir = false;
4958 unsigned max_links = new_dir->i_sb->s_max_links;
4959 struct name_snapshot old_name;
4960 bool lock_old_subdir, lock_new_subdir;
4961
4962 if (source == target)
4963 return 0;
4964
4965 error = may_delete(rd->old_mnt_idmap, old_dir, old_dentry, is_dir);
4966 if (error)
4967 return error;
4968
4969 if (!target) {
4970 error = may_create(rd->new_mnt_idmap, new_dir, new_dentry);
4971 } else {
4972 new_is_dir = d_is_dir(new_dentry);
4973
4974 if (!(flags & RENAME_EXCHANGE))
4975 error = may_delete(rd->new_mnt_idmap, new_dir,
4976 new_dentry, is_dir);
4977 else
4978 error = may_delete(rd->new_mnt_idmap, new_dir,
4979 new_dentry, new_is_dir);
4980 }
4981 if (error)
4982 return error;
4983
4984 if (!old_dir->i_op->rename)
4985 return -EPERM;
4986
4987 /*
4988 * If we are going to change the parent - check write permissions,
4989 * we'll need to flip '..'.
4990 */
4991 if (new_dir != old_dir) {
4992 if (is_dir) {
4993 error = inode_permission(rd->old_mnt_idmap, source,
4994 MAY_WRITE);
4995 if (error)
4996 return error;
4997 }
4998 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4999 error = inode_permission(rd->new_mnt_idmap, target,
5000 MAY_WRITE);
5001 if (error)
5002 return error;
5003 }
5004 }
5005
5006 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
5007 flags);
5008 if (error)
5009 return error;
5010
5011 take_dentry_name_snapshot(&old_name, old_dentry);
5012 dget(new_dentry);
5013 /*
5014 * Lock children.
5015 * The source subdirectory needs to be locked on cross-directory
5016 * rename or cross-directory exchange since its parent changes.
5017 * The target subdirectory needs to be locked on cross-directory
5018 * exchange due to parent change and on any rename due to becoming
5019 * a victim.
5020 * Non-directories need locking in all cases (for NFS reasons);
5021 * they get locked after any subdirectories (in inode address order).
5022 *
5023 * NOTE: WE ONLY LOCK UNRELATED DIRECTORIES IN CROSS-DIRECTORY CASE.
5024 * NEVER, EVER DO THAT WITHOUT ->s_vfs_rename_mutex.
5025 */
5026 lock_old_subdir = new_dir != old_dir;
5027 lock_new_subdir = new_dir != old_dir || !(flags & RENAME_EXCHANGE);
5028 if (is_dir) {
5029 if (lock_old_subdir)
5030 inode_lock_nested(source, I_MUTEX_CHILD);
5031 if (target && (!new_is_dir || lock_new_subdir))
5032 inode_lock(target);
5033 } else if (new_is_dir) {
5034 if (lock_new_subdir)
5035 inode_lock_nested(target, I_MUTEX_CHILD);
5036 inode_lock(source);
5037 } else {
5038 lock_two_nondirectories(source, target);
5039 }
5040
5041 error = -EPERM;
5042 if (IS_SWAPFILE(source) || (target && IS_SWAPFILE(target)))
5043 goto out;
5044
5045 error = -EBUSY;
5046 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
5047 goto out;
5048
5049 if (max_links && new_dir != old_dir) {
5050 error = -EMLINK;
5051 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
5052 goto out;
5053 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
5054 old_dir->i_nlink >= max_links)
5055 goto out;
5056 }
5057 if (!is_dir) {
5058 error = try_break_deleg(source, delegated_inode);
5059 if (error)
5060 goto out;
5061 }
5062 if (target && !new_is_dir) {
5063 error = try_break_deleg(target, delegated_inode);
5064 if (error)
5065 goto out;
5066 }
5067 error = old_dir->i_op->rename(rd->new_mnt_idmap, old_dir, old_dentry,
5068 new_dir, new_dentry, flags);
5069 if (error)
5070 goto out;
5071
5072 if (!(flags & RENAME_EXCHANGE) && target) {
5073 if (is_dir) {
5074 shrink_dcache_parent(new_dentry);
5075 target->i_flags |= S_DEAD;
5076 }
5077 dont_mount(new_dentry);
5078 detach_mounts(new_dentry);
5079 }
5080 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
5081 if (!(flags & RENAME_EXCHANGE))
5082 d_move(old_dentry, new_dentry);
5083 else
5084 d_exchange(old_dentry, new_dentry);
5085 }
5086out:
5087 if (!is_dir || lock_old_subdir)
5088 inode_unlock(source);
5089 if (target && (!new_is_dir || lock_new_subdir))
5090 inode_unlock(target);
5091 dput(new_dentry);
5092 if (!error) {
5093 fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
5094 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
5095 if (flags & RENAME_EXCHANGE) {
5096 fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
5097 new_is_dir, NULL, new_dentry);
5098 }
5099 }
5100 release_dentry_name_snapshot(&old_name);
5101
5102 return error;
5103}
5104EXPORT_SYMBOL(vfs_rename);
5105
5106int do_renameat2(int olddfd, struct filename *from, int newdfd,
5107 struct filename *to, unsigned int flags)
5108{
5109 struct renamedata rd;
5110 struct dentry *old_dentry, *new_dentry;
5111 struct dentry *trap;
5112 struct path old_path, new_path;
5113 struct qstr old_last, new_last;
5114 int old_type, new_type;
5115 struct inode *delegated_inode = NULL;
5116 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
5117 bool should_retry = false;
5118 int error = -EINVAL;
5119
5120 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
5121 goto put_names;
5122
5123 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
5124 (flags & RENAME_EXCHANGE))
5125 goto put_names;
5126
5127 if (flags & RENAME_EXCHANGE)
5128 target_flags = 0;
5129
5130retry:
5131 error = filename_parentat(olddfd, from, lookup_flags, &old_path,
5132 &old_last, &old_type);
5133 if (error)
5134 goto put_names;
5135
5136 error = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last,
5137 &new_type);
5138 if (error)
5139 goto exit1;
5140
5141 error = -EXDEV;
5142 if (old_path.mnt != new_path.mnt)
5143 goto exit2;
5144
5145 error = -EBUSY;
5146 if (old_type != LAST_NORM)
5147 goto exit2;
5148
5149 if (flags & RENAME_NOREPLACE)
5150 error = -EEXIST;
5151 if (new_type != LAST_NORM)
5152 goto exit2;
5153
5154 error = mnt_want_write(old_path.mnt);
5155 if (error)
5156 goto exit2;
5157
5158retry_deleg:
5159 trap = lock_rename(new_path.dentry, old_path.dentry);
5160 if (IS_ERR(trap)) {
5161 error = PTR_ERR(trap);
5162 goto exit_lock_rename;
5163 }
5164
5165 old_dentry = lookup_one_qstr_excl(&old_last, old_path.dentry,
5166 lookup_flags);
5167 error = PTR_ERR(old_dentry);
5168 if (IS_ERR(old_dentry))
5169 goto exit3;
5170 /* source must exist */
5171 error = -ENOENT;
5172 if (d_is_negative(old_dentry))
5173 goto exit4;
5174 new_dentry = lookup_one_qstr_excl(&new_last, new_path.dentry,
5175 lookup_flags | target_flags);
5176 error = PTR_ERR(new_dentry);
5177 if (IS_ERR(new_dentry))
5178 goto exit4;
5179 error = -EEXIST;
5180 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
5181 goto exit5;
5182 if (flags & RENAME_EXCHANGE) {
5183 error = -ENOENT;
5184 if (d_is_negative(new_dentry))
5185 goto exit5;
5186
5187 if (!d_is_dir(new_dentry)) {
5188 error = -ENOTDIR;
5189 if (new_last.name[new_last.len])
5190 goto exit5;
5191 }
5192 }
5193 /* unless the source is a directory trailing slashes give -ENOTDIR */
5194 if (!d_is_dir(old_dentry)) {
5195 error = -ENOTDIR;
5196 if (old_last.name[old_last.len])
5197 goto exit5;
5198 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
5199 goto exit5;
5200 }
5201 /* source should not be ancestor of target */
5202 error = -EINVAL;
5203 if (old_dentry == trap)
5204 goto exit5;
5205 /* target should not be an ancestor of source */
5206 if (!(flags & RENAME_EXCHANGE))
5207 error = -ENOTEMPTY;
5208 if (new_dentry == trap)
5209 goto exit5;
5210
5211 error = security_path_rename(&old_path, old_dentry,
5212 &new_path, new_dentry, flags);
5213 if (error)
5214 goto exit5;
5215
5216 rd.old_dir = old_path.dentry->d_inode;
5217 rd.old_dentry = old_dentry;
5218 rd.old_mnt_idmap = mnt_idmap(old_path.mnt);
5219 rd.new_dir = new_path.dentry->d_inode;
5220 rd.new_dentry = new_dentry;
5221 rd.new_mnt_idmap = mnt_idmap(new_path.mnt);
5222 rd.delegated_inode = &delegated_inode;
5223 rd.flags = flags;
5224 error = vfs_rename(&rd);
5225exit5:
5226 dput(new_dentry);
5227exit4:
5228 dput(old_dentry);
5229exit3:
5230 unlock_rename(new_path.dentry, old_path.dentry);
5231exit_lock_rename:
5232 if (delegated_inode) {
5233 error = break_deleg_wait(&delegated_inode);
5234 if (!error)
5235 goto retry_deleg;
5236 }
5237 mnt_drop_write(old_path.mnt);
5238exit2:
5239 if (retry_estale(error, lookup_flags))
5240 should_retry = true;
5241 path_put(&new_path);
5242exit1:
5243 path_put(&old_path);
5244 if (should_retry) {
5245 should_retry = false;
5246 lookup_flags |= LOOKUP_REVAL;
5247 goto retry;
5248 }
5249put_names:
5250 putname(from);
5251 putname(to);
5252 return error;
5253}
5254
5255SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
5256 int, newdfd, const char __user *, newname, unsigned int, flags)
5257{
5258 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
5259 flags);
5260}
5261
5262SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
5263 int, newdfd, const char __user *, newname)
5264{
5265 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
5266 0);
5267}
5268
5269SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5270{
5271 return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD,
5272 getname(newname), 0);
5273}
5274
5275int readlink_copy(char __user *buffer, int buflen, const char *link)
5276{
5277 int len = PTR_ERR(link);
5278 if (IS_ERR(link))
5279 goto out;
5280
5281 len = strlen(link);
5282 if (len > (unsigned) buflen)
5283 len = buflen;
5284 if (copy_to_user(buffer, link, len))
5285 len = -EFAULT;
5286out:
5287 return len;
5288}
5289
5290/**
5291 * vfs_readlink - copy symlink body into userspace buffer
5292 * @dentry: dentry on which to get symbolic link
5293 * @buffer: user memory pointer
5294 * @buflen: size of buffer
5295 *
5296 * Does not touch atime. That's up to the caller if necessary
5297 *
5298 * Does not call security hook.
5299 */
5300int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
5301{
5302 struct inode *inode = d_inode(dentry);
5303 DEFINE_DELAYED_CALL(done);
5304 const char *link;
5305 int res;
5306
5307 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
5308 if (unlikely(inode->i_op->readlink))
5309 return inode->i_op->readlink(dentry, buffer, buflen);
5310
5311 if (!d_is_symlink(dentry))
5312 return -EINVAL;
5313
5314 spin_lock(&inode->i_lock);
5315 inode->i_opflags |= IOP_DEFAULT_READLINK;
5316 spin_unlock(&inode->i_lock);
5317 }
5318
5319 link = READ_ONCE(inode->i_link);
5320 if (!link) {
5321 link = inode->i_op->get_link(dentry, inode, &done);
5322 if (IS_ERR(link))
5323 return PTR_ERR(link);
5324 }
5325 res = readlink_copy(buffer, buflen, link);
5326 do_delayed_call(&done);
5327 return res;
5328}
5329EXPORT_SYMBOL(vfs_readlink);
5330
5331/**
5332 * vfs_get_link - get symlink body
5333 * @dentry: dentry on which to get symbolic link
5334 * @done: caller needs to free returned data with this
5335 *
5336 * Calls security hook and i_op->get_link() on the supplied inode.
5337 *
5338 * It does not touch atime. That's up to the caller if necessary.
5339 *
5340 * Does not work on "special" symlinks like /proc/$$/fd/N
5341 */
5342const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
5343{
5344 const char *res = ERR_PTR(-EINVAL);
5345 struct inode *inode = d_inode(dentry);
5346
5347 if (d_is_symlink(dentry)) {
5348 res = ERR_PTR(security_inode_readlink(dentry));
5349 if (!res)
5350 res = inode->i_op->get_link(dentry, inode, done);
5351 }
5352 return res;
5353}
5354EXPORT_SYMBOL(vfs_get_link);
5355
5356/* get the link contents into pagecache */
5357const char *page_get_link(struct dentry *dentry, struct inode *inode,
5358 struct delayed_call *callback)
5359{
5360 char *kaddr;
5361 struct page *page;
5362 struct address_space *mapping = inode->i_mapping;
5363
5364 if (!dentry) {
5365 page = find_get_page(mapping, 0);
5366 if (!page)
5367 return ERR_PTR(-ECHILD);
5368 if (!PageUptodate(page)) {
5369 put_page(page);
5370 return ERR_PTR(-ECHILD);
5371 }
5372 } else {
5373 page = read_mapping_page(mapping, 0, NULL);
5374 if (IS_ERR(page))
5375 return (char*)page;
5376 }
5377 set_delayed_call(callback, page_put_link, page);
5378 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
5379 kaddr = page_address(page);
5380 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
5381 return kaddr;
5382}
5383
5384EXPORT_SYMBOL(page_get_link);
5385
5386void page_put_link(void *arg)
5387{
5388 put_page(arg);
5389}
5390EXPORT_SYMBOL(page_put_link);
5391
5392int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
5393{
5394 DEFINE_DELAYED_CALL(done);
5395 int res = readlink_copy(buffer, buflen,
5396 page_get_link(dentry, d_inode(dentry),
5397 &done));
5398 do_delayed_call(&done);
5399 return res;
5400}
5401EXPORT_SYMBOL(page_readlink);
5402
5403int page_symlink(struct inode *inode, const char *symname, int len)
5404{
5405 struct address_space *mapping = inode->i_mapping;
5406 const struct address_space_operations *aops = mapping->a_ops;
5407 bool nofs = !mapping_gfp_constraint(mapping, __GFP_FS);
5408 struct folio *folio;
5409 void *fsdata = NULL;
5410 int err;
5411 unsigned int flags;
5412
5413retry:
5414 if (nofs)
5415 flags = memalloc_nofs_save();
5416 err = aops->write_begin(NULL, mapping, 0, len-1, &folio, &fsdata);
5417 if (nofs)
5418 memalloc_nofs_restore(flags);
5419 if (err)
5420 goto fail;
5421
5422 memcpy(folio_address(folio), symname, len - 1);
5423
5424 err = aops->write_end(NULL, mapping, 0, len - 1, len - 1,
5425 folio, fsdata);
5426 if (err < 0)
5427 goto fail;
5428 if (err < len-1)
5429 goto retry;
5430
5431 mark_inode_dirty(inode);
5432 return 0;
5433fail:
5434 return err;
5435}
5436EXPORT_SYMBOL(page_symlink);
5437
5438const struct inode_operations page_symlink_inode_operations = {
5439 .get_link = page_get_link,
5440};
5441EXPORT_SYMBOL(page_symlink_inode_operations);