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/kernel.h>
21#include <linux/slab.h>
22#include <linux/fs.h>
23#include <linux/namei.h>
24#include <linux/pagemap.h>
25#include <linux/sched/mm.h>
26#include <linux/fsnotify.h>
27#include <linux/personality.h>
28#include <linux/security.h>
29#include <linux/ima.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, int *empty)
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 if (unlikely(len < 0)) {
152 __putname(result);
153 return ERR_PTR(len);
154 }
155
156 /*
157 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
158 * separate struct filename so we can dedicate the entire
159 * names_cache allocation for the pathname, and re-do the copy from
160 * userland.
161 */
162 if (unlikely(len == EMBEDDED_NAME_MAX)) {
163 const size_t size = offsetof(struct filename, iname[1]);
164 kname = (char *)result;
165
166 /*
167 * size is chosen that way we to guarantee that
168 * result->iname[0] is within the same object and that
169 * kname can't be equal to result->iname, no matter what.
170 */
171 result = kzalloc(size, GFP_KERNEL);
172 if (unlikely(!result)) {
173 __putname(kname);
174 return ERR_PTR(-ENOMEM);
175 }
176 result->name = kname;
177 len = strncpy_from_user(kname, filename, PATH_MAX);
178 if (unlikely(len < 0)) {
179 __putname(kname);
180 kfree(result);
181 return ERR_PTR(len);
182 }
183 if (unlikely(len == PATH_MAX)) {
184 __putname(kname);
185 kfree(result);
186 return ERR_PTR(-ENAMETOOLONG);
187 }
188 }
189
190 result->refcnt = 1;
191 /* The empty path is special. */
192 if (unlikely(!len)) {
193 if (empty)
194 *empty = 1;
195 if (!(flags & LOOKUP_EMPTY)) {
196 putname(result);
197 return ERR_PTR(-ENOENT);
198 }
199 }
200
201 result->uptr = filename;
202 result->aname = NULL;
203 audit_getname(result);
204 return result;
205}
206
207struct filename *
208getname_uflags(const char __user *filename, int uflags)
209{
210 int flags = (uflags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
211
212 return getname_flags(filename, flags, NULL);
213}
214
215struct filename *
216getname(const char __user * filename)
217{
218 return getname_flags(filename, 0, NULL);
219}
220
221struct filename *
222getname_kernel(const char * filename)
223{
224 struct filename *result;
225 int len = strlen(filename) + 1;
226
227 result = __getname();
228 if (unlikely(!result))
229 return ERR_PTR(-ENOMEM);
230
231 if (len <= EMBEDDED_NAME_MAX) {
232 result->name = (char *)result->iname;
233 } else if (len <= PATH_MAX) {
234 const size_t size = offsetof(struct filename, iname[1]);
235 struct filename *tmp;
236
237 tmp = kmalloc(size, GFP_KERNEL);
238 if (unlikely(!tmp)) {
239 __putname(result);
240 return ERR_PTR(-ENOMEM);
241 }
242 tmp->name = (char *)result;
243 result = tmp;
244 } else {
245 __putname(result);
246 return ERR_PTR(-ENAMETOOLONG);
247 }
248 memcpy((char *)result->name, filename, len);
249 result->uptr = NULL;
250 result->aname = NULL;
251 result->refcnt = 1;
252 audit_getname(result);
253
254 return result;
255}
256
257void putname(struct filename *name)
258{
259 if (IS_ERR(name))
260 return;
261
262 BUG_ON(name->refcnt <= 0);
263
264 if (--name->refcnt > 0)
265 return;
266
267 if (name->name != name->iname) {
268 __putname(name->name);
269 kfree(name);
270 } else
271 __putname(name);
272}
273
274/**
275 * check_acl - perform ACL permission checking
276 * @mnt_userns: user namespace of the mount the inode was found from
277 * @inode: inode to check permissions on
278 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
279 *
280 * This function performs the ACL permission checking. Since this function
281 * retrieve POSIX acls it needs to know whether it is called from a blocking or
282 * non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
283 *
284 * If the inode has been found through an idmapped mount the user namespace of
285 * the vfsmount must be passed through @mnt_userns. This function will then take
286 * care to map the inode according to @mnt_userns before checking permissions.
287 * On non-idmapped mounts or if permission checking is to be performed on the
288 * raw inode simply passs init_user_ns.
289 */
290static int check_acl(struct user_namespace *mnt_userns,
291 struct inode *inode, int mask)
292{
293#ifdef CONFIG_FS_POSIX_ACL
294 struct posix_acl *acl;
295
296 if (mask & MAY_NOT_BLOCK) {
297 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
298 if (!acl)
299 return -EAGAIN;
300 /* no ->get_inode_acl() calls in RCU mode... */
301 if (is_uncached_acl(acl))
302 return -ECHILD;
303 return posix_acl_permission(mnt_userns, inode, acl, mask);
304 }
305
306 acl = get_inode_acl(inode, ACL_TYPE_ACCESS);
307 if (IS_ERR(acl))
308 return PTR_ERR(acl);
309 if (acl) {
310 int error = posix_acl_permission(mnt_userns, inode, acl, mask);
311 posix_acl_release(acl);
312 return error;
313 }
314#endif
315
316 return -EAGAIN;
317}
318
319/**
320 * acl_permission_check - perform basic UNIX permission checking
321 * @mnt_userns: user namespace of the mount the inode was found from
322 * @inode: inode to check permissions on
323 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
324 *
325 * This function performs the basic UNIX permission checking. Since this
326 * function may retrieve POSIX acls it needs to know whether it is called from a
327 * blocking or non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
328 *
329 * If the inode has been found through an idmapped mount the user namespace of
330 * the vfsmount must be passed through @mnt_userns. This function will then take
331 * care to map the inode according to @mnt_userns before checking permissions.
332 * On non-idmapped mounts or if permission checking is to be performed on the
333 * raw inode simply passs init_user_ns.
334 */
335static int acl_permission_check(struct user_namespace *mnt_userns,
336 struct inode *inode, int mask)
337{
338 unsigned int mode = inode->i_mode;
339 vfsuid_t vfsuid;
340
341 /* Are we the owner? If so, ACL's don't matter */
342 vfsuid = i_uid_into_vfsuid(mnt_userns, inode);
343 if (likely(vfsuid_eq_kuid(vfsuid, current_fsuid()))) {
344 mask &= 7;
345 mode >>= 6;
346 return (mask & ~mode) ? -EACCES : 0;
347 }
348
349 /* Do we have ACL's? */
350 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
351 int error = check_acl(mnt_userns, inode, mask);
352 if (error != -EAGAIN)
353 return error;
354 }
355
356 /* Only RWX matters for group/other mode bits */
357 mask &= 7;
358
359 /*
360 * Are the group permissions different from
361 * the other permissions in the bits we care
362 * about? Need to check group ownership if so.
363 */
364 if (mask & (mode ^ (mode >> 3))) {
365 vfsgid_t vfsgid = i_gid_into_vfsgid(mnt_userns, inode);
366 if (vfsgid_in_group_p(vfsgid))
367 mode >>= 3;
368 }
369
370 /* Bits in 'mode' clear that we require? */
371 return (mask & ~mode) ? -EACCES : 0;
372}
373
374/**
375 * generic_permission - check for access rights on a Posix-like filesystem
376 * @mnt_userns: user namespace of the mount the inode was found from
377 * @inode: inode to check access rights for
378 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC,
379 * %MAY_NOT_BLOCK ...)
380 *
381 * Used to check for read/write/execute permissions on a file.
382 * We use "fsuid" for this, letting us set arbitrary permissions
383 * for filesystem access without changing the "normal" uids which
384 * are used for other things.
385 *
386 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
387 * request cannot be satisfied (eg. requires blocking or too much complexity).
388 * It would then be called again in ref-walk mode.
389 *
390 * If the inode has been found through an idmapped mount the user namespace of
391 * the vfsmount must be passed through @mnt_userns. This function will then take
392 * care to map the inode according to @mnt_userns before checking permissions.
393 * On non-idmapped mounts or if permission checking is to be performed on the
394 * raw inode simply passs init_user_ns.
395 */
396int generic_permission(struct user_namespace *mnt_userns, struct inode *inode,
397 int mask)
398{
399 int ret;
400
401 /*
402 * Do the basic permission checks.
403 */
404 ret = acl_permission_check(mnt_userns, inode, mask);
405 if (ret != -EACCES)
406 return ret;
407
408 if (S_ISDIR(inode->i_mode)) {
409 /* DACs are overridable for directories */
410 if (!(mask & MAY_WRITE))
411 if (capable_wrt_inode_uidgid(mnt_userns, inode,
412 CAP_DAC_READ_SEARCH))
413 return 0;
414 if (capable_wrt_inode_uidgid(mnt_userns, inode,
415 CAP_DAC_OVERRIDE))
416 return 0;
417 return -EACCES;
418 }
419
420 /*
421 * Searching includes executable on directories, else just read.
422 */
423 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
424 if (mask == MAY_READ)
425 if (capable_wrt_inode_uidgid(mnt_userns, inode,
426 CAP_DAC_READ_SEARCH))
427 return 0;
428 /*
429 * Read/write DACs are always overridable.
430 * Executable DACs are overridable when there is
431 * at least one exec bit set.
432 */
433 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
434 if (capable_wrt_inode_uidgid(mnt_userns, inode,
435 CAP_DAC_OVERRIDE))
436 return 0;
437
438 return -EACCES;
439}
440EXPORT_SYMBOL(generic_permission);
441
442/**
443 * do_inode_permission - UNIX permission checking
444 * @mnt_userns: user namespace of the mount the inode was found from
445 * @inode: inode to check permissions on
446 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
447 *
448 * We _really_ want to just do "generic_permission()" without
449 * even looking at the inode->i_op values. So we keep a cache
450 * flag in inode->i_opflags, that says "this has not special
451 * permission function, use the fast case".
452 */
453static inline int do_inode_permission(struct user_namespace *mnt_userns,
454 struct inode *inode, int mask)
455{
456 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
457 if (likely(inode->i_op->permission))
458 return inode->i_op->permission(mnt_userns, inode, mask);
459
460 /* This gets set once for the inode lifetime */
461 spin_lock(&inode->i_lock);
462 inode->i_opflags |= IOP_FASTPERM;
463 spin_unlock(&inode->i_lock);
464 }
465 return generic_permission(mnt_userns, inode, mask);
466}
467
468/**
469 * sb_permission - Check superblock-level permissions
470 * @sb: Superblock of inode to check permission on
471 * @inode: Inode to check permission on
472 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
473 *
474 * Separate out file-system wide checks from inode-specific permission checks.
475 */
476static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
477{
478 if (unlikely(mask & MAY_WRITE)) {
479 umode_t mode = inode->i_mode;
480
481 /* Nobody gets write access to a read-only fs. */
482 if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
483 return -EROFS;
484 }
485 return 0;
486}
487
488/**
489 * inode_permission - Check for access rights to a given inode
490 * @mnt_userns: User namespace of the mount the inode was found from
491 * @inode: Inode to check permission on
492 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
493 *
494 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
495 * this, letting us set arbitrary permissions for filesystem access without
496 * changing the "normal" UIDs which are used for other things.
497 *
498 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
499 */
500int inode_permission(struct user_namespace *mnt_userns,
501 struct inode *inode, int mask)
502{
503 int retval;
504
505 retval = sb_permission(inode->i_sb, inode, mask);
506 if (retval)
507 return retval;
508
509 if (unlikely(mask & MAY_WRITE)) {
510 /*
511 * Nobody gets write access to an immutable file.
512 */
513 if (IS_IMMUTABLE(inode))
514 return -EPERM;
515
516 /*
517 * Updating mtime will likely cause i_uid and i_gid to be
518 * written back improperly if their true value is unknown
519 * to the vfs.
520 */
521 if (HAS_UNMAPPED_ID(mnt_userns, inode))
522 return -EACCES;
523 }
524
525 retval = do_inode_permission(mnt_userns, inode, mask);
526 if (retval)
527 return retval;
528
529 retval = devcgroup_inode_permission(inode, mask);
530 if (retval)
531 return retval;
532
533 return security_inode_permission(inode, mask);
534}
535EXPORT_SYMBOL(inode_permission);
536
537/**
538 * path_get - get a reference to a path
539 * @path: path to get the reference to
540 *
541 * Given a path increment the reference count to the dentry and the vfsmount.
542 */
543void path_get(const struct path *path)
544{
545 mntget(path->mnt);
546 dget(path->dentry);
547}
548EXPORT_SYMBOL(path_get);
549
550/**
551 * path_put - put a reference to a path
552 * @path: path to put the reference to
553 *
554 * Given a path decrement the reference count to the dentry and the vfsmount.
555 */
556void path_put(const struct path *path)
557{
558 dput(path->dentry);
559 mntput(path->mnt);
560}
561EXPORT_SYMBOL(path_put);
562
563#define EMBEDDED_LEVELS 2
564struct nameidata {
565 struct path path;
566 struct qstr last;
567 struct path root;
568 struct inode *inode; /* path.dentry.d_inode */
569 unsigned int flags, state;
570 unsigned seq, next_seq, m_seq, r_seq;
571 int last_type;
572 unsigned depth;
573 int total_link_count;
574 struct saved {
575 struct path link;
576 struct delayed_call done;
577 const char *name;
578 unsigned seq;
579 } *stack, internal[EMBEDDED_LEVELS];
580 struct filename *name;
581 struct nameidata *saved;
582 unsigned root_seq;
583 int dfd;
584 vfsuid_t dir_vfsuid;
585 umode_t dir_mode;
586} __randomize_layout;
587
588#define ND_ROOT_PRESET 1
589#define ND_ROOT_GRABBED 2
590#define ND_JUMPED 4
591
592static void __set_nameidata(struct nameidata *p, int dfd, struct filename *name)
593{
594 struct nameidata *old = current->nameidata;
595 p->stack = p->internal;
596 p->depth = 0;
597 p->dfd = dfd;
598 p->name = name;
599 p->path.mnt = NULL;
600 p->path.dentry = NULL;
601 p->total_link_count = old ? old->total_link_count : 0;
602 p->saved = old;
603 current->nameidata = p;
604}
605
606static inline void set_nameidata(struct nameidata *p, int dfd, struct filename *name,
607 const struct path *root)
608{
609 __set_nameidata(p, dfd, name);
610 p->state = 0;
611 if (unlikely(root)) {
612 p->state = ND_ROOT_PRESET;
613 p->root = *root;
614 }
615}
616
617static void restore_nameidata(void)
618{
619 struct nameidata *now = current->nameidata, *old = now->saved;
620
621 current->nameidata = old;
622 if (old)
623 old->total_link_count = now->total_link_count;
624 if (now->stack != now->internal)
625 kfree(now->stack);
626}
627
628static bool nd_alloc_stack(struct nameidata *nd)
629{
630 struct saved *p;
631
632 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
633 nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL);
634 if (unlikely(!p))
635 return false;
636 memcpy(p, nd->internal, sizeof(nd->internal));
637 nd->stack = p;
638 return true;
639}
640
641/**
642 * path_connected - Verify that a dentry is below mnt.mnt_root
643 *
644 * Rename can sometimes move a file or directory outside of a bind
645 * mount, path_connected allows those cases to be detected.
646 */
647static bool path_connected(struct vfsmount *mnt, struct dentry *dentry)
648{
649 struct super_block *sb = mnt->mnt_sb;
650
651 /* Bind mounts can have disconnected paths */
652 if (mnt->mnt_root == sb->s_root)
653 return true;
654
655 return is_subdir(dentry, mnt->mnt_root);
656}
657
658static void drop_links(struct nameidata *nd)
659{
660 int i = nd->depth;
661 while (i--) {
662 struct saved *last = nd->stack + i;
663 do_delayed_call(&last->done);
664 clear_delayed_call(&last->done);
665 }
666}
667
668static void leave_rcu(struct nameidata *nd)
669{
670 nd->flags &= ~LOOKUP_RCU;
671 nd->seq = nd->next_seq = 0;
672 rcu_read_unlock();
673}
674
675static void terminate_walk(struct nameidata *nd)
676{
677 drop_links(nd);
678 if (!(nd->flags & LOOKUP_RCU)) {
679 int i;
680 path_put(&nd->path);
681 for (i = 0; i < nd->depth; i++)
682 path_put(&nd->stack[i].link);
683 if (nd->state & ND_ROOT_GRABBED) {
684 path_put(&nd->root);
685 nd->state &= ~ND_ROOT_GRABBED;
686 }
687 } else {
688 leave_rcu(nd);
689 }
690 nd->depth = 0;
691 nd->path.mnt = NULL;
692 nd->path.dentry = NULL;
693}
694
695/* path_put is needed afterwards regardless of success or failure */
696static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq)
697{
698 int res = __legitimize_mnt(path->mnt, mseq);
699 if (unlikely(res)) {
700 if (res > 0)
701 path->mnt = NULL;
702 path->dentry = NULL;
703 return false;
704 }
705 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
706 path->dentry = NULL;
707 return false;
708 }
709 return !read_seqcount_retry(&path->dentry->d_seq, seq);
710}
711
712static inline bool legitimize_path(struct nameidata *nd,
713 struct path *path, unsigned seq)
714{
715 return __legitimize_path(path, seq, nd->m_seq);
716}
717
718static bool legitimize_links(struct nameidata *nd)
719{
720 int i;
721 if (unlikely(nd->flags & LOOKUP_CACHED)) {
722 drop_links(nd);
723 nd->depth = 0;
724 return false;
725 }
726 for (i = 0; i < nd->depth; i++) {
727 struct saved *last = nd->stack + i;
728 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
729 drop_links(nd);
730 nd->depth = i + 1;
731 return false;
732 }
733 }
734 return true;
735}
736
737static bool legitimize_root(struct nameidata *nd)
738{
739 /* Nothing to do if nd->root is zero or is managed by the VFS user. */
740 if (!nd->root.mnt || (nd->state & ND_ROOT_PRESET))
741 return true;
742 nd->state |= ND_ROOT_GRABBED;
743 return legitimize_path(nd, &nd->root, nd->root_seq);
744}
745
746/*
747 * Path walking has 2 modes, rcu-walk and ref-walk (see
748 * Documentation/filesystems/path-lookup.txt). In situations when we can't
749 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
750 * normal reference counts on dentries and vfsmounts to transition to ref-walk
751 * mode. Refcounts are grabbed at the last known good point before rcu-walk
752 * got stuck, so ref-walk may continue from there. If this is not successful
753 * (eg. a seqcount has changed), then failure is returned and it's up to caller
754 * to restart the path walk from the beginning in ref-walk mode.
755 */
756
757/**
758 * try_to_unlazy - try to switch to ref-walk mode.
759 * @nd: nameidata pathwalk data
760 * Returns: true on success, false on failure
761 *
762 * try_to_unlazy attempts to legitimize the current nd->path and nd->root
763 * for ref-walk mode.
764 * Must be called from rcu-walk context.
765 * Nothing should touch nameidata between try_to_unlazy() failure and
766 * terminate_walk().
767 */
768static bool try_to_unlazy(struct nameidata *nd)
769{
770 struct dentry *parent = nd->path.dentry;
771
772 BUG_ON(!(nd->flags & LOOKUP_RCU));
773
774 if (unlikely(!legitimize_links(nd)))
775 goto out1;
776 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
777 goto out;
778 if (unlikely(!legitimize_root(nd)))
779 goto out;
780 leave_rcu(nd);
781 BUG_ON(nd->inode != parent->d_inode);
782 return true;
783
784out1:
785 nd->path.mnt = NULL;
786 nd->path.dentry = NULL;
787out:
788 leave_rcu(nd);
789 return false;
790}
791
792/**
793 * try_to_unlazy_next - try to switch to ref-walk mode.
794 * @nd: nameidata pathwalk data
795 * @dentry: next dentry to step into
796 * Returns: true on success, false on failure
797 *
798 * Similar to try_to_unlazy(), but here we have the next dentry already
799 * picked by rcu-walk and want to legitimize that in addition to the current
800 * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context.
801 * Nothing should touch nameidata between try_to_unlazy_next() failure and
802 * terminate_walk().
803 */
804static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry)
805{
806 int res;
807 BUG_ON(!(nd->flags & LOOKUP_RCU));
808
809 if (unlikely(!legitimize_links(nd)))
810 goto out2;
811 res = __legitimize_mnt(nd->path.mnt, nd->m_seq);
812 if (unlikely(res)) {
813 if (res > 0)
814 goto out2;
815 goto out1;
816 }
817 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
818 goto out1;
819
820 /*
821 * We need to move both the parent and the dentry from the RCU domain
822 * to be properly refcounted. And the sequence number in the dentry
823 * validates *both* dentry counters, since we checked the sequence
824 * number of the parent after we got the child sequence number. So we
825 * know the parent must still be valid if the child sequence number is
826 */
827 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
828 goto out;
829 if (read_seqcount_retry(&dentry->d_seq, nd->next_seq))
830 goto out_dput;
831 /*
832 * Sequence counts matched. Now make sure that the root is
833 * still valid and get it if required.
834 */
835 if (unlikely(!legitimize_root(nd)))
836 goto out_dput;
837 leave_rcu(nd);
838 return true;
839
840out2:
841 nd->path.mnt = NULL;
842out1:
843 nd->path.dentry = NULL;
844out:
845 leave_rcu(nd);
846 return false;
847out_dput:
848 leave_rcu(nd);
849 dput(dentry);
850 return false;
851}
852
853static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
854{
855 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
856 return dentry->d_op->d_revalidate(dentry, flags);
857 else
858 return 1;
859}
860
861/**
862 * complete_walk - successful completion of path walk
863 * @nd: pointer nameidata
864 *
865 * If we had been in RCU mode, drop out of it and legitimize nd->path.
866 * Revalidate the final result, unless we'd already done that during
867 * the path walk or the filesystem doesn't ask for it. Return 0 on
868 * success, -error on failure. In case of failure caller does not
869 * need to drop nd->path.
870 */
871static int complete_walk(struct nameidata *nd)
872{
873 struct dentry *dentry = nd->path.dentry;
874 int status;
875
876 if (nd->flags & LOOKUP_RCU) {
877 /*
878 * We don't want to zero nd->root for scoped-lookups or
879 * externally-managed nd->root.
880 */
881 if (!(nd->state & ND_ROOT_PRESET))
882 if (!(nd->flags & LOOKUP_IS_SCOPED))
883 nd->root.mnt = NULL;
884 nd->flags &= ~LOOKUP_CACHED;
885 if (!try_to_unlazy(nd))
886 return -ECHILD;
887 }
888
889 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
890 /*
891 * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't
892 * ever step outside the root during lookup" and should already
893 * be guaranteed by the rest of namei, we want to avoid a namei
894 * BUG resulting in userspace being given a path that was not
895 * scoped within the root at some point during the lookup.
896 *
897 * So, do a final sanity-check to make sure that in the
898 * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED)
899 * we won't silently return an fd completely outside of the
900 * requested root to userspace.
901 *
902 * Userspace could move the path outside the root after this
903 * check, but as discussed elsewhere this is not a concern (the
904 * resolved file was inside the root at some point).
905 */
906 if (!path_is_under(&nd->path, &nd->root))
907 return -EXDEV;
908 }
909
910 if (likely(!(nd->state & ND_JUMPED)))
911 return 0;
912
913 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
914 return 0;
915
916 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
917 if (status > 0)
918 return 0;
919
920 if (!status)
921 status = -ESTALE;
922
923 return status;
924}
925
926static int set_root(struct nameidata *nd)
927{
928 struct fs_struct *fs = current->fs;
929
930 /*
931 * Jumping to the real root in a scoped-lookup is a BUG in namei, but we
932 * still have to ensure it doesn't happen because it will cause a breakout
933 * from the dirfd.
934 */
935 if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED))
936 return -ENOTRECOVERABLE;
937
938 if (nd->flags & LOOKUP_RCU) {
939 unsigned seq;
940
941 do {
942 seq = read_seqcount_begin(&fs->seq);
943 nd->root = fs->root;
944 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
945 } while (read_seqcount_retry(&fs->seq, seq));
946 } else {
947 get_fs_root(fs, &nd->root);
948 nd->state |= ND_ROOT_GRABBED;
949 }
950 return 0;
951}
952
953static int nd_jump_root(struct nameidata *nd)
954{
955 if (unlikely(nd->flags & LOOKUP_BENEATH))
956 return -EXDEV;
957 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
958 /* Absolute path arguments to path_init() are allowed. */
959 if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt)
960 return -EXDEV;
961 }
962 if (!nd->root.mnt) {
963 int error = set_root(nd);
964 if (error)
965 return error;
966 }
967 if (nd->flags & LOOKUP_RCU) {
968 struct dentry *d;
969 nd->path = nd->root;
970 d = nd->path.dentry;
971 nd->inode = d->d_inode;
972 nd->seq = nd->root_seq;
973 if (read_seqcount_retry(&d->d_seq, nd->seq))
974 return -ECHILD;
975 } else {
976 path_put(&nd->path);
977 nd->path = nd->root;
978 path_get(&nd->path);
979 nd->inode = nd->path.dentry->d_inode;
980 }
981 nd->state |= ND_JUMPED;
982 return 0;
983}
984
985/*
986 * Helper to directly jump to a known parsed path from ->get_link,
987 * caller must have taken a reference to path beforehand.
988 */
989int nd_jump_link(const struct path *path)
990{
991 int error = -ELOOP;
992 struct nameidata *nd = current->nameidata;
993
994 if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS))
995 goto err;
996
997 error = -EXDEV;
998 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
999 if (nd->path.mnt != path->mnt)
1000 goto err;
1001 }
1002 /* Not currently safe for scoped-lookups. */
1003 if (unlikely(nd->flags & LOOKUP_IS_SCOPED))
1004 goto err;
1005
1006 path_put(&nd->path);
1007 nd->path = *path;
1008 nd->inode = nd->path.dentry->d_inode;
1009 nd->state |= ND_JUMPED;
1010 return 0;
1011
1012err:
1013 path_put(path);
1014 return error;
1015}
1016
1017static inline void put_link(struct nameidata *nd)
1018{
1019 struct saved *last = nd->stack + --nd->depth;
1020 do_delayed_call(&last->done);
1021 if (!(nd->flags & LOOKUP_RCU))
1022 path_put(&last->link);
1023}
1024
1025static int sysctl_protected_symlinks __read_mostly;
1026static int sysctl_protected_hardlinks __read_mostly;
1027static int sysctl_protected_fifos __read_mostly;
1028static int sysctl_protected_regular __read_mostly;
1029
1030#ifdef CONFIG_SYSCTL
1031static struct ctl_table namei_sysctls[] = {
1032 {
1033 .procname = "protected_symlinks",
1034 .data = &sysctl_protected_symlinks,
1035 .maxlen = sizeof(int),
1036 .mode = 0644,
1037 .proc_handler = proc_dointvec_minmax,
1038 .extra1 = SYSCTL_ZERO,
1039 .extra2 = SYSCTL_ONE,
1040 },
1041 {
1042 .procname = "protected_hardlinks",
1043 .data = &sysctl_protected_hardlinks,
1044 .maxlen = sizeof(int),
1045 .mode = 0644,
1046 .proc_handler = proc_dointvec_minmax,
1047 .extra1 = SYSCTL_ZERO,
1048 .extra2 = SYSCTL_ONE,
1049 },
1050 {
1051 .procname = "protected_fifos",
1052 .data = &sysctl_protected_fifos,
1053 .maxlen = sizeof(int),
1054 .mode = 0644,
1055 .proc_handler = proc_dointvec_minmax,
1056 .extra1 = SYSCTL_ZERO,
1057 .extra2 = SYSCTL_TWO,
1058 },
1059 {
1060 .procname = "protected_regular",
1061 .data = &sysctl_protected_regular,
1062 .maxlen = sizeof(int),
1063 .mode = 0644,
1064 .proc_handler = proc_dointvec_minmax,
1065 .extra1 = SYSCTL_ZERO,
1066 .extra2 = SYSCTL_TWO,
1067 },
1068 { }
1069};
1070
1071static int __init init_fs_namei_sysctls(void)
1072{
1073 register_sysctl_init("fs", namei_sysctls);
1074 return 0;
1075}
1076fs_initcall(init_fs_namei_sysctls);
1077
1078#endif /* CONFIG_SYSCTL */
1079
1080/**
1081 * may_follow_link - Check symlink following for unsafe situations
1082 * @nd: nameidata pathwalk data
1083 *
1084 * In the case of the sysctl_protected_symlinks sysctl being enabled,
1085 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
1086 * in a sticky world-writable directory. This is to protect privileged
1087 * processes from failing races against path names that may change out
1088 * from under them by way of other users creating malicious symlinks.
1089 * It will permit symlinks to be followed only when outside a sticky
1090 * world-writable directory, or when the uid of the symlink and follower
1091 * match, or when the directory owner matches the symlink's owner.
1092 *
1093 * Returns 0 if following the symlink is allowed, -ve on error.
1094 */
1095static inline int may_follow_link(struct nameidata *nd, const struct inode *inode)
1096{
1097 struct user_namespace *mnt_userns;
1098 vfsuid_t vfsuid;
1099
1100 if (!sysctl_protected_symlinks)
1101 return 0;
1102
1103 mnt_userns = mnt_user_ns(nd->path.mnt);
1104 vfsuid = i_uid_into_vfsuid(mnt_userns, inode);
1105 /* Allowed if owner and follower match. */
1106 if (vfsuid_eq_kuid(vfsuid, current_fsuid()))
1107 return 0;
1108
1109 /* Allowed if parent directory not sticky and world-writable. */
1110 if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
1111 return 0;
1112
1113 /* Allowed if parent directory and link owner match. */
1114 if (vfsuid_valid(nd->dir_vfsuid) && vfsuid_eq(nd->dir_vfsuid, vfsuid))
1115 return 0;
1116
1117 if (nd->flags & LOOKUP_RCU)
1118 return -ECHILD;
1119
1120 audit_inode(nd->name, nd->stack[0].link.dentry, 0);
1121 audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link");
1122 return -EACCES;
1123}
1124
1125/**
1126 * safe_hardlink_source - Check for safe hardlink conditions
1127 * @mnt_userns: user namespace of the mount the inode was found from
1128 * @inode: the source inode to hardlink from
1129 *
1130 * Return false if at least one of the following conditions:
1131 * - inode is not a regular file
1132 * - inode is setuid
1133 * - inode is setgid and group-exec
1134 * - access failure for read and write
1135 *
1136 * Otherwise returns true.
1137 */
1138static bool safe_hardlink_source(struct user_namespace *mnt_userns,
1139 struct inode *inode)
1140{
1141 umode_t mode = inode->i_mode;
1142
1143 /* Special files should not get pinned to the filesystem. */
1144 if (!S_ISREG(mode))
1145 return false;
1146
1147 /* Setuid files should not get pinned to the filesystem. */
1148 if (mode & S_ISUID)
1149 return false;
1150
1151 /* Executable setgid files should not get pinned to the filesystem. */
1152 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
1153 return false;
1154
1155 /* Hardlinking to unreadable or unwritable sources is dangerous. */
1156 if (inode_permission(mnt_userns, inode, MAY_READ | MAY_WRITE))
1157 return false;
1158
1159 return true;
1160}
1161
1162/**
1163 * may_linkat - Check permissions for creating a hardlink
1164 * @mnt_userns: user namespace of the mount the inode was found from
1165 * @link: the source to hardlink from
1166 *
1167 * Block hardlink when all of:
1168 * - sysctl_protected_hardlinks enabled
1169 * - fsuid does not match inode
1170 * - hardlink source is unsafe (see safe_hardlink_source() above)
1171 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
1172 *
1173 * If the inode has been found through an idmapped mount the user namespace of
1174 * the vfsmount must be passed through @mnt_userns. This function will then take
1175 * care to map the inode according to @mnt_userns before checking permissions.
1176 * On non-idmapped mounts or if permission checking is to be performed on the
1177 * raw inode simply passs init_user_ns.
1178 *
1179 * Returns 0 if successful, -ve on error.
1180 */
1181int may_linkat(struct user_namespace *mnt_userns, const struct path *link)
1182{
1183 struct inode *inode = link->dentry->d_inode;
1184
1185 /* Inode writeback is not safe when the uid or gid are invalid. */
1186 if (!vfsuid_valid(i_uid_into_vfsuid(mnt_userns, inode)) ||
1187 !vfsgid_valid(i_gid_into_vfsgid(mnt_userns, inode)))
1188 return -EOVERFLOW;
1189
1190 if (!sysctl_protected_hardlinks)
1191 return 0;
1192
1193 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1194 * otherwise, it must be a safe source.
1195 */
1196 if (safe_hardlink_source(mnt_userns, inode) ||
1197 inode_owner_or_capable(mnt_userns, inode))
1198 return 0;
1199
1200 audit_log_path_denied(AUDIT_ANOM_LINK, "linkat");
1201 return -EPERM;
1202}
1203
1204/**
1205 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1206 * should be allowed, or not, on files that already
1207 * exist.
1208 * @mnt_userns: user namespace of the mount the inode was found from
1209 * @nd: nameidata pathwalk data
1210 * @inode: the inode of the file to open
1211 *
1212 * Block an O_CREAT open of a FIFO (or a regular file) when:
1213 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1214 * - the file already exists
1215 * - we are in a sticky directory
1216 * - we don't own the file
1217 * - the owner of the directory doesn't own the file
1218 * - the directory is world writable
1219 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1220 * the directory doesn't have to be world writable: being group writable will
1221 * be enough.
1222 *
1223 * If the inode has been found through an idmapped mount the user namespace of
1224 * the vfsmount must be passed through @mnt_userns. This function will then take
1225 * care to map the inode according to @mnt_userns before checking permissions.
1226 * On non-idmapped mounts or if permission checking is to be performed on the
1227 * raw inode simply passs init_user_ns.
1228 *
1229 * Returns 0 if the open is allowed, -ve on error.
1230 */
1231static int may_create_in_sticky(struct user_namespace *mnt_userns,
1232 struct nameidata *nd, struct inode *const inode)
1233{
1234 umode_t dir_mode = nd->dir_mode;
1235 vfsuid_t dir_vfsuid = nd->dir_vfsuid;
1236
1237 if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1238 (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1239 likely(!(dir_mode & S_ISVTX)) ||
1240 vfsuid_eq(i_uid_into_vfsuid(mnt_userns, inode), dir_vfsuid) ||
1241 vfsuid_eq_kuid(i_uid_into_vfsuid(mnt_userns, inode), current_fsuid()))
1242 return 0;
1243
1244 if (likely(dir_mode & 0002) ||
1245 (dir_mode & 0020 &&
1246 ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1247 (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1248 const char *operation = S_ISFIFO(inode->i_mode) ?
1249 "sticky_create_fifo" :
1250 "sticky_create_regular";
1251 audit_log_path_denied(AUDIT_ANOM_CREAT, operation);
1252 return -EACCES;
1253 }
1254 return 0;
1255}
1256
1257/*
1258 * follow_up - Find the mountpoint of path's vfsmount
1259 *
1260 * Given a path, find the mountpoint of its source file system.
1261 * Replace @path with the path of the mountpoint in the parent mount.
1262 * Up is towards /.
1263 *
1264 * Return 1 if we went up a level and 0 if we were already at the
1265 * root.
1266 */
1267int follow_up(struct path *path)
1268{
1269 struct mount *mnt = real_mount(path->mnt);
1270 struct mount *parent;
1271 struct dentry *mountpoint;
1272
1273 read_seqlock_excl(&mount_lock);
1274 parent = mnt->mnt_parent;
1275 if (parent == mnt) {
1276 read_sequnlock_excl(&mount_lock);
1277 return 0;
1278 }
1279 mntget(&parent->mnt);
1280 mountpoint = dget(mnt->mnt_mountpoint);
1281 read_sequnlock_excl(&mount_lock);
1282 dput(path->dentry);
1283 path->dentry = mountpoint;
1284 mntput(path->mnt);
1285 path->mnt = &parent->mnt;
1286 return 1;
1287}
1288EXPORT_SYMBOL(follow_up);
1289
1290static bool choose_mountpoint_rcu(struct mount *m, const struct path *root,
1291 struct path *path, unsigned *seqp)
1292{
1293 while (mnt_has_parent(m)) {
1294 struct dentry *mountpoint = m->mnt_mountpoint;
1295
1296 m = m->mnt_parent;
1297 if (unlikely(root->dentry == mountpoint &&
1298 root->mnt == &m->mnt))
1299 break;
1300 if (mountpoint != m->mnt.mnt_root) {
1301 path->mnt = &m->mnt;
1302 path->dentry = mountpoint;
1303 *seqp = read_seqcount_begin(&mountpoint->d_seq);
1304 return true;
1305 }
1306 }
1307 return false;
1308}
1309
1310static bool choose_mountpoint(struct mount *m, const struct path *root,
1311 struct path *path)
1312{
1313 bool found;
1314
1315 rcu_read_lock();
1316 while (1) {
1317 unsigned seq, mseq = read_seqbegin(&mount_lock);
1318
1319 found = choose_mountpoint_rcu(m, root, path, &seq);
1320 if (unlikely(!found)) {
1321 if (!read_seqretry(&mount_lock, mseq))
1322 break;
1323 } else {
1324 if (likely(__legitimize_path(path, seq, mseq)))
1325 break;
1326 rcu_read_unlock();
1327 path_put(path);
1328 rcu_read_lock();
1329 }
1330 }
1331 rcu_read_unlock();
1332 return found;
1333}
1334
1335/*
1336 * Perform an automount
1337 * - return -EISDIR to tell follow_managed() to stop and return the path we
1338 * were called with.
1339 */
1340static int follow_automount(struct path *path, int *count, unsigned lookup_flags)
1341{
1342 struct dentry *dentry = path->dentry;
1343
1344 /* We don't want to mount if someone's just doing a stat -
1345 * unless they're stat'ing a directory and appended a '/' to
1346 * the name.
1347 *
1348 * We do, however, want to mount if someone wants to open or
1349 * create a file of any type under the mountpoint, wants to
1350 * traverse through the mountpoint or wants to open the
1351 * mounted directory. Also, autofs may mark negative dentries
1352 * as being automount points. These will need the attentions
1353 * of the daemon to instantiate them before they can be used.
1354 */
1355 if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1356 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1357 dentry->d_inode)
1358 return -EISDIR;
1359
1360 if (count && (*count)++ >= MAXSYMLINKS)
1361 return -ELOOP;
1362
1363 return finish_automount(dentry->d_op->d_automount(path), path);
1364}
1365
1366/*
1367 * mount traversal - out-of-line part. One note on ->d_flags accesses -
1368 * dentries are pinned but not locked here, so negative dentry can go
1369 * positive right under us. Use of smp_load_acquire() provides a barrier
1370 * sufficient for ->d_inode and ->d_flags consistency.
1371 */
1372static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped,
1373 int *count, unsigned lookup_flags)
1374{
1375 struct vfsmount *mnt = path->mnt;
1376 bool need_mntput = false;
1377 int ret = 0;
1378
1379 while (flags & DCACHE_MANAGED_DENTRY) {
1380 /* Allow the filesystem to manage the transit without i_mutex
1381 * being held. */
1382 if (flags & DCACHE_MANAGE_TRANSIT) {
1383 ret = path->dentry->d_op->d_manage(path, false);
1384 flags = smp_load_acquire(&path->dentry->d_flags);
1385 if (ret < 0)
1386 break;
1387 }
1388
1389 if (flags & DCACHE_MOUNTED) { // something's mounted on it..
1390 struct vfsmount *mounted = lookup_mnt(path);
1391 if (mounted) { // ... in our namespace
1392 dput(path->dentry);
1393 if (need_mntput)
1394 mntput(path->mnt);
1395 path->mnt = mounted;
1396 path->dentry = dget(mounted->mnt_root);
1397 // here we know it's positive
1398 flags = path->dentry->d_flags;
1399 need_mntput = true;
1400 continue;
1401 }
1402 }
1403
1404 if (!(flags & DCACHE_NEED_AUTOMOUNT))
1405 break;
1406
1407 // uncovered automount point
1408 ret = follow_automount(path, count, lookup_flags);
1409 flags = smp_load_acquire(&path->dentry->d_flags);
1410 if (ret < 0)
1411 break;
1412 }
1413
1414 if (ret == -EISDIR)
1415 ret = 0;
1416 // possible if you race with several mount --move
1417 if (need_mntput && path->mnt == mnt)
1418 mntput(path->mnt);
1419 if (!ret && unlikely(d_flags_negative(flags)))
1420 ret = -ENOENT;
1421 *jumped = need_mntput;
1422 return ret;
1423}
1424
1425static inline int traverse_mounts(struct path *path, bool *jumped,
1426 int *count, unsigned lookup_flags)
1427{
1428 unsigned flags = smp_load_acquire(&path->dentry->d_flags);
1429
1430 /* fastpath */
1431 if (likely(!(flags & DCACHE_MANAGED_DENTRY))) {
1432 *jumped = false;
1433 if (unlikely(d_flags_negative(flags)))
1434 return -ENOENT;
1435 return 0;
1436 }
1437 return __traverse_mounts(path, flags, jumped, count, lookup_flags);
1438}
1439
1440int follow_down_one(struct path *path)
1441{
1442 struct vfsmount *mounted;
1443
1444 mounted = lookup_mnt(path);
1445 if (mounted) {
1446 dput(path->dentry);
1447 mntput(path->mnt);
1448 path->mnt = mounted;
1449 path->dentry = dget(mounted->mnt_root);
1450 return 1;
1451 }
1452 return 0;
1453}
1454EXPORT_SYMBOL(follow_down_one);
1455
1456/*
1457 * Follow down to the covering mount currently visible to userspace. At each
1458 * point, the filesystem owning that dentry may be queried as to whether the
1459 * caller is permitted to proceed or not.
1460 */
1461int follow_down(struct path *path)
1462{
1463 struct vfsmount *mnt = path->mnt;
1464 bool jumped;
1465 int ret = traverse_mounts(path, &jumped, NULL, 0);
1466
1467 if (path->mnt != mnt)
1468 mntput(mnt);
1469 return ret;
1470}
1471EXPORT_SYMBOL(follow_down);
1472
1473/*
1474 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1475 * we meet a managed dentry that would need blocking.
1476 */
1477static bool __follow_mount_rcu(struct nameidata *nd, struct path *path)
1478{
1479 struct dentry *dentry = path->dentry;
1480 unsigned int flags = dentry->d_flags;
1481
1482 if (likely(!(flags & DCACHE_MANAGED_DENTRY)))
1483 return true;
1484
1485 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1486 return false;
1487
1488 for (;;) {
1489 /*
1490 * Don't forget we might have a non-mountpoint managed dentry
1491 * that wants to block transit.
1492 */
1493 if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) {
1494 int res = dentry->d_op->d_manage(path, true);
1495 if (res)
1496 return res == -EISDIR;
1497 flags = dentry->d_flags;
1498 }
1499
1500 if (flags & DCACHE_MOUNTED) {
1501 struct mount *mounted = __lookup_mnt(path->mnt, dentry);
1502 if (mounted) {
1503 path->mnt = &mounted->mnt;
1504 dentry = path->dentry = mounted->mnt.mnt_root;
1505 nd->state |= ND_JUMPED;
1506 nd->next_seq = read_seqcount_begin(&dentry->d_seq);
1507 flags = dentry->d_flags;
1508 // makes sure that non-RCU pathwalk could reach
1509 // this state.
1510 if (read_seqretry(&mount_lock, nd->m_seq))
1511 return false;
1512 continue;
1513 }
1514 if (read_seqretry(&mount_lock, nd->m_seq))
1515 return false;
1516 }
1517 return !(flags & DCACHE_NEED_AUTOMOUNT);
1518 }
1519}
1520
1521static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry,
1522 struct path *path)
1523{
1524 bool jumped;
1525 int ret;
1526
1527 path->mnt = nd->path.mnt;
1528 path->dentry = dentry;
1529 if (nd->flags & LOOKUP_RCU) {
1530 unsigned int seq = nd->next_seq;
1531 if (likely(__follow_mount_rcu(nd, path)))
1532 return 0;
1533 // *path and nd->next_seq might've been clobbered
1534 path->mnt = nd->path.mnt;
1535 path->dentry = dentry;
1536 nd->next_seq = seq;
1537 if (!try_to_unlazy_next(nd, dentry))
1538 return -ECHILD;
1539 }
1540 ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags);
1541 if (jumped) {
1542 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1543 ret = -EXDEV;
1544 else
1545 nd->state |= ND_JUMPED;
1546 }
1547 if (unlikely(ret)) {
1548 dput(path->dentry);
1549 if (path->mnt != nd->path.mnt)
1550 mntput(path->mnt);
1551 }
1552 return ret;
1553}
1554
1555/*
1556 * This looks up the name in dcache and possibly revalidates the found dentry.
1557 * NULL is returned if the dentry does not exist in the cache.
1558 */
1559static struct dentry *lookup_dcache(const struct qstr *name,
1560 struct dentry *dir,
1561 unsigned int flags)
1562{
1563 struct dentry *dentry = d_lookup(dir, name);
1564 if (dentry) {
1565 int error = d_revalidate(dentry, flags);
1566 if (unlikely(error <= 0)) {
1567 if (!error)
1568 d_invalidate(dentry);
1569 dput(dentry);
1570 return ERR_PTR(error);
1571 }
1572 }
1573 return dentry;
1574}
1575
1576/*
1577 * Parent directory has inode locked exclusive. This is one
1578 * and only case when ->lookup() gets called on non in-lookup
1579 * dentries - as the matter of fact, this only gets called
1580 * when directory is guaranteed to have no in-lookup children
1581 * at all.
1582 */
1583static struct dentry *__lookup_hash(const struct qstr *name,
1584 struct dentry *base, unsigned int flags)
1585{
1586 struct dentry *dentry = lookup_dcache(name, base, flags);
1587 struct dentry *old;
1588 struct inode *dir = base->d_inode;
1589
1590 if (dentry)
1591 return dentry;
1592
1593 /* Don't create child dentry for a dead directory. */
1594 if (unlikely(IS_DEADDIR(dir)))
1595 return ERR_PTR(-ENOENT);
1596
1597 dentry = d_alloc(base, name);
1598 if (unlikely(!dentry))
1599 return ERR_PTR(-ENOMEM);
1600
1601 old = dir->i_op->lookup(dir, dentry, flags);
1602 if (unlikely(old)) {
1603 dput(dentry);
1604 dentry = old;
1605 }
1606 return dentry;
1607}
1608
1609static struct dentry *lookup_fast(struct nameidata *nd)
1610{
1611 struct dentry *dentry, *parent = nd->path.dentry;
1612 int status = 1;
1613
1614 /*
1615 * Rename seqlock is not required here because in the off chance
1616 * of a false negative due to a concurrent rename, the caller is
1617 * going to fall back to non-racy lookup.
1618 */
1619 if (nd->flags & LOOKUP_RCU) {
1620 dentry = __d_lookup_rcu(parent, &nd->last, &nd->next_seq);
1621 if (unlikely(!dentry)) {
1622 if (!try_to_unlazy(nd))
1623 return ERR_PTR(-ECHILD);
1624 return NULL;
1625 }
1626
1627 /*
1628 * This sequence count validates that the parent had no
1629 * changes while we did the lookup of the dentry above.
1630 */
1631 if (read_seqcount_retry(&parent->d_seq, nd->seq))
1632 return ERR_PTR(-ECHILD);
1633
1634 status = d_revalidate(dentry, nd->flags);
1635 if (likely(status > 0))
1636 return dentry;
1637 if (!try_to_unlazy_next(nd, dentry))
1638 return ERR_PTR(-ECHILD);
1639 if (status == -ECHILD)
1640 /* we'd been told to redo it in non-rcu mode */
1641 status = d_revalidate(dentry, nd->flags);
1642 } else {
1643 dentry = __d_lookup(parent, &nd->last);
1644 if (unlikely(!dentry))
1645 return NULL;
1646 status = d_revalidate(dentry, nd->flags);
1647 }
1648 if (unlikely(status <= 0)) {
1649 if (!status)
1650 d_invalidate(dentry);
1651 dput(dentry);
1652 return ERR_PTR(status);
1653 }
1654 return dentry;
1655}
1656
1657/* Fast lookup failed, do it the slow way */
1658static struct dentry *__lookup_slow(const struct qstr *name,
1659 struct dentry *dir,
1660 unsigned int flags)
1661{
1662 struct dentry *dentry, *old;
1663 struct inode *inode = dir->d_inode;
1664 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1665
1666 /* Don't go there if it's already dead */
1667 if (unlikely(IS_DEADDIR(inode)))
1668 return ERR_PTR(-ENOENT);
1669again:
1670 dentry = d_alloc_parallel(dir, name, &wq);
1671 if (IS_ERR(dentry))
1672 return dentry;
1673 if (unlikely(!d_in_lookup(dentry))) {
1674 int error = d_revalidate(dentry, flags);
1675 if (unlikely(error <= 0)) {
1676 if (!error) {
1677 d_invalidate(dentry);
1678 dput(dentry);
1679 goto again;
1680 }
1681 dput(dentry);
1682 dentry = ERR_PTR(error);
1683 }
1684 } else {
1685 old = inode->i_op->lookup(inode, dentry, flags);
1686 d_lookup_done(dentry);
1687 if (unlikely(old)) {
1688 dput(dentry);
1689 dentry = old;
1690 }
1691 }
1692 return dentry;
1693}
1694
1695static struct dentry *lookup_slow(const struct qstr *name,
1696 struct dentry *dir,
1697 unsigned int flags)
1698{
1699 struct inode *inode = dir->d_inode;
1700 struct dentry *res;
1701 inode_lock_shared(inode);
1702 res = __lookup_slow(name, dir, flags);
1703 inode_unlock_shared(inode);
1704 return res;
1705}
1706
1707static inline int may_lookup(struct user_namespace *mnt_userns,
1708 struct nameidata *nd)
1709{
1710 if (nd->flags & LOOKUP_RCU) {
1711 int err = inode_permission(mnt_userns, nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1712 if (err != -ECHILD || !try_to_unlazy(nd))
1713 return err;
1714 }
1715 return inode_permission(mnt_userns, nd->inode, MAY_EXEC);
1716}
1717
1718static int reserve_stack(struct nameidata *nd, struct path *link)
1719{
1720 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS))
1721 return -ELOOP;
1722
1723 if (likely(nd->depth != EMBEDDED_LEVELS))
1724 return 0;
1725 if (likely(nd->stack != nd->internal))
1726 return 0;
1727 if (likely(nd_alloc_stack(nd)))
1728 return 0;
1729
1730 if (nd->flags & LOOKUP_RCU) {
1731 // we need to grab link before we do unlazy. And we can't skip
1732 // unlazy even if we fail to grab the link - cleanup needs it
1733 bool grabbed_link = legitimize_path(nd, link, nd->next_seq);
1734
1735 if (!try_to_unlazy(nd) || !grabbed_link)
1736 return -ECHILD;
1737
1738 if (nd_alloc_stack(nd))
1739 return 0;
1740 }
1741 return -ENOMEM;
1742}
1743
1744enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4};
1745
1746static const char *pick_link(struct nameidata *nd, struct path *link,
1747 struct inode *inode, int flags)
1748{
1749 struct saved *last;
1750 const char *res;
1751 int error = reserve_stack(nd, link);
1752
1753 if (unlikely(error)) {
1754 if (!(nd->flags & LOOKUP_RCU))
1755 path_put(link);
1756 return ERR_PTR(error);
1757 }
1758 last = nd->stack + nd->depth++;
1759 last->link = *link;
1760 clear_delayed_call(&last->done);
1761 last->seq = nd->next_seq;
1762
1763 if (flags & WALK_TRAILING) {
1764 error = may_follow_link(nd, inode);
1765 if (unlikely(error))
1766 return ERR_PTR(error);
1767 }
1768
1769 if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) ||
1770 unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW))
1771 return ERR_PTR(-ELOOP);
1772
1773 if (!(nd->flags & LOOKUP_RCU)) {
1774 touch_atime(&last->link);
1775 cond_resched();
1776 } else if (atime_needs_update(&last->link, inode)) {
1777 if (!try_to_unlazy(nd))
1778 return ERR_PTR(-ECHILD);
1779 touch_atime(&last->link);
1780 }
1781
1782 error = security_inode_follow_link(link->dentry, inode,
1783 nd->flags & LOOKUP_RCU);
1784 if (unlikely(error))
1785 return ERR_PTR(error);
1786
1787 res = READ_ONCE(inode->i_link);
1788 if (!res) {
1789 const char * (*get)(struct dentry *, struct inode *,
1790 struct delayed_call *);
1791 get = inode->i_op->get_link;
1792 if (nd->flags & LOOKUP_RCU) {
1793 res = get(NULL, inode, &last->done);
1794 if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd))
1795 res = get(link->dentry, inode, &last->done);
1796 } else {
1797 res = get(link->dentry, inode, &last->done);
1798 }
1799 if (!res)
1800 goto all_done;
1801 if (IS_ERR(res))
1802 return res;
1803 }
1804 if (*res == '/') {
1805 error = nd_jump_root(nd);
1806 if (unlikely(error))
1807 return ERR_PTR(error);
1808 while (unlikely(*++res == '/'))
1809 ;
1810 }
1811 if (*res)
1812 return res;
1813all_done: // pure jump
1814 put_link(nd);
1815 return NULL;
1816}
1817
1818/*
1819 * Do we need to follow links? We _really_ want to be able
1820 * to do this check without having to look at inode->i_op,
1821 * so we keep a cache of "no, this doesn't need follow_link"
1822 * for the common case.
1823 *
1824 * NOTE: dentry must be what nd->next_seq had been sampled from.
1825 */
1826static const char *step_into(struct nameidata *nd, int flags,
1827 struct dentry *dentry)
1828{
1829 struct path path;
1830 struct inode *inode;
1831 int err = handle_mounts(nd, dentry, &path);
1832
1833 if (err < 0)
1834 return ERR_PTR(err);
1835 inode = path.dentry->d_inode;
1836 if (likely(!d_is_symlink(path.dentry)) ||
1837 ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) ||
1838 (flags & WALK_NOFOLLOW)) {
1839 /* not a symlink or should not follow */
1840 if (nd->flags & LOOKUP_RCU) {
1841 if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq))
1842 return ERR_PTR(-ECHILD);
1843 if (unlikely(!inode))
1844 return ERR_PTR(-ENOENT);
1845 } else {
1846 dput(nd->path.dentry);
1847 if (nd->path.mnt != path.mnt)
1848 mntput(nd->path.mnt);
1849 }
1850 nd->path = path;
1851 nd->inode = inode;
1852 nd->seq = nd->next_seq;
1853 return NULL;
1854 }
1855 if (nd->flags & LOOKUP_RCU) {
1856 /* make sure that d_is_symlink above matches inode */
1857 if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq))
1858 return ERR_PTR(-ECHILD);
1859 } else {
1860 if (path.mnt == nd->path.mnt)
1861 mntget(path.mnt);
1862 }
1863 return pick_link(nd, &path, inode, flags);
1864}
1865
1866static struct dentry *follow_dotdot_rcu(struct nameidata *nd)
1867{
1868 struct dentry *parent, *old;
1869
1870 if (path_equal(&nd->path, &nd->root))
1871 goto in_root;
1872 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1873 struct path path;
1874 unsigned seq;
1875 if (!choose_mountpoint_rcu(real_mount(nd->path.mnt),
1876 &nd->root, &path, &seq))
1877 goto in_root;
1878 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1879 return ERR_PTR(-ECHILD);
1880 nd->path = path;
1881 nd->inode = path.dentry->d_inode;
1882 nd->seq = seq;
1883 // makes sure that non-RCU pathwalk could reach this state
1884 if (read_seqretry(&mount_lock, nd->m_seq))
1885 return ERR_PTR(-ECHILD);
1886 /* we know that mountpoint was pinned */
1887 }
1888 old = nd->path.dentry;
1889 parent = old->d_parent;
1890 nd->next_seq = read_seqcount_begin(&parent->d_seq);
1891 // makes sure that non-RCU pathwalk could reach this state
1892 if (read_seqcount_retry(&old->d_seq, nd->seq))
1893 return ERR_PTR(-ECHILD);
1894 if (unlikely(!path_connected(nd->path.mnt, parent)))
1895 return ERR_PTR(-ECHILD);
1896 return parent;
1897in_root:
1898 if (read_seqretry(&mount_lock, nd->m_seq))
1899 return ERR_PTR(-ECHILD);
1900 if (unlikely(nd->flags & LOOKUP_BENEATH))
1901 return ERR_PTR(-ECHILD);
1902 nd->next_seq = nd->seq;
1903 return nd->path.dentry;
1904}
1905
1906static struct dentry *follow_dotdot(struct nameidata *nd)
1907{
1908 struct dentry *parent;
1909
1910 if (path_equal(&nd->path, &nd->root))
1911 goto in_root;
1912 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1913 struct path path;
1914
1915 if (!choose_mountpoint(real_mount(nd->path.mnt),
1916 &nd->root, &path))
1917 goto in_root;
1918 path_put(&nd->path);
1919 nd->path = path;
1920 nd->inode = path.dentry->d_inode;
1921 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1922 return ERR_PTR(-EXDEV);
1923 }
1924 /* rare case of legitimate dget_parent()... */
1925 parent = dget_parent(nd->path.dentry);
1926 if (unlikely(!path_connected(nd->path.mnt, parent))) {
1927 dput(parent);
1928 return ERR_PTR(-ENOENT);
1929 }
1930 return parent;
1931
1932in_root:
1933 if (unlikely(nd->flags & LOOKUP_BENEATH))
1934 return ERR_PTR(-EXDEV);
1935 return dget(nd->path.dentry);
1936}
1937
1938static const char *handle_dots(struct nameidata *nd, int type)
1939{
1940 if (type == LAST_DOTDOT) {
1941 const char *error = NULL;
1942 struct dentry *parent;
1943
1944 if (!nd->root.mnt) {
1945 error = ERR_PTR(set_root(nd));
1946 if (error)
1947 return error;
1948 }
1949 if (nd->flags & LOOKUP_RCU)
1950 parent = follow_dotdot_rcu(nd);
1951 else
1952 parent = follow_dotdot(nd);
1953 if (IS_ERR(parent))
1954 return ERR_CAST(parent);
1955 error = step_into(nd, WALK_NOFOLLOW, parent);
1956 if (unlikely(error))
1957 return error;
1958
1959 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
1960 /*
1961 * If there was a racing rename or mount along our
1962 * path, then we can't be sure that ".." hasn't jumped
1963 * above nd->root (and so userspace should retry or use
1964 * some fallback).
1965 */
1966 smp_rmb();
1967 if (__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq))
1968 return ERR_PTR(-EAGAIN);
1969 if (__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq))
1970 return ERR_PTR(-EAGAIN);
1971 }
1972 }
1973 return NULL;
1974}
1975
1976static const char *walk_component(struct nameidata *nd, int flags)
1977{
1978 struct dentry *dentry;
1979 /*
1980 * "." and ".." are special - ".." especially so because it has
1981 * to be able to know about the current root directory and
1982 * parent relationships.
1983 */
1984 if (unlikely(nd->last_type != LAST_NORM)) {
1985 if (!(flags & WALK_MORE) && nd->depth)
1986 put_link(nd);
1987 return handle_dots(nd, nd->last_type);
1988 }
1989 dentry = lookup_fast(nd);
1990 if (IS_ERR(dentry))
1991 return ERR_CAST(dentry);
1992 if (unlikely(!dentry)) {
1993 dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags);
1994 if (IS_ERR(dentry))
1995 return ERR_CAST(dentry);
1996 }
1997 if (!(flags & WALK_MORE) && nd->depth)
1998 put_link(nd);
1999 return step_into(nd, flags, dentry);
2000}
2001
2002/*
2003 * We can do the critical dentry name comparison and hashing
2004 * operations one word at a time, but we are limited to:
2005 *
2006 * - Architectures with fast unaligned word accesses. We could
2007 * do a "get_unaligned()" if this helps and is sufficiently
2008 * fast.
2009 *
2010 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
2011 * do not trap on the (extremely unlikely) case of a page
2012 * crossing operation.
2013 *
2014 * - Furthermore, we need an efficient 64-bit compile for the
2015 * 64-bit case in order to generate the "number of bytes in
2016 * the final mask". Again, that could be replaced with a
2017 * efficient population count instruction or similar.
2018 */
2019#ifdef CONFIG_DCACHE_WORD_ACCESS
2020
2021#include <asm/word-at-a-time.h>
2022
2023#ifdef HASH_MIX
2024
2025/* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
2026
2027#elif defined(CONFIG_64BIT)
2028/*
2029 * Register pressure in the mixing function is an issue, particularly
2030 * on 32-bit x86, but almost any function requires one state value and
2031 * one temporary. Instead, use a function designed for two state values
2032 * and no temporaries.
2033 *
2034 * This function cannot create a collision in only two iterations, so
2035 * we have two iterations to achieve avalanche. In those two iterations,
2036 * we have six layers of mixing, which is enough to spread one bit's
2037 * influence out to 2^6 = 64 state bits.
2038 *
2039 * Rotate constants are scored by considering either 64 one-bit input
2040 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
2041 * probability of that delta causing a change to each of the 128 output
2042 * bits, using a sample of random initial states.
2043 *
2044 * The Shannon entropy of the computed probabilities is then summed
2045 * to produce a score. Ideally, any input change has a 50% chance of
2046 * toggling any given output bit.
2047 *
2048 * Mixing scores (in bits) for (12,45):
2049 * Input delta: 1-bit 2-bit
2050 * 1 round: 713.3 42542.6
2051 * 2 rounds: 2753.7 140389.8
2052 * 3 rounds: 5954.1 233458.2
2053 * 4 rounds: 7862.6 256672.2
2054 * Perfect: 8192 258048
2055 * (64*128) (64*63/2 * 128)
2056 */
2057#define HASH_MIX(x, y, a) \
2058 ( x ^= (a), \
2059 y ^= x, x = rol64(x,12),\
2060 x += y, y = rol64(y,45),\
2061 y *= 9 )
2062
2063/*
2064 * Fold two longs into one 32-bit hash value. This must be fast, but
2065 * latency isn't quite as critical, as there is a fair bit of additional
2066 * work done before the hash value is used.
2067 */
2068static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2069{
2070 y ^= x * GOLDEN_RATIO_64;
2071 y *= GOLDEN_RATIO_64;
2072 return y >> 32;
2073}
2074
2075#else /* 32-bit case */
2076
2077/*
2078 * Mixing scores (in bits) for (7,20):
2079 * Input delta: 1-bit 2-bit
2080 * 1 round: 330.3 9201.6
2081 * 2 rounds: 1246.4 25475.4
2082 * 3 rounds: 1907.1 31295.1
2083 * 4 rounds: 2042.3 31718.6
2084 * Perfect: 2048 31744
2085 * (32*64) (32*31/2 * 64)
2086 */
2087#define HASH_MIX(x, y, a) \
2088 ( x ^= (a), \
2089 y ^= x, x = rol32(x, 7),\
2090 x += y, y = rol32(y,20),\
2091 y *= 9 )
2092
2093static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2094{
2095 /* Use arch-optimized multiply if one exists */
2096 return __hash_32(y ^ __hash_32(x));
2097}
2098
2099#endif
2100
2101/*
2102 * Return the hash of a string of known length. This is carfully
2103 * designed to match hash_name(), which is the more critical function.
2104 * In particular, we must end by hashing a final word containing 0..7
2105 * payload bytes, to match the way that hash_name() iterates until it
2106 * finds the delimiter after the name.
2107 */
2108unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2109{
2110 unsigned long a, x = 0, y = (unsigned long)salt;
2111
2112 for (;;) {
2113 if (!len)
2114 goto done;
2115 a = load_unaligned_zeropad(name);
2116 if (len < sizeof(unsigned long))
2117 break;
2118 HASH_MIX(x, y, a);
2119 name += sizeof(unsigned long);
2120 len -= sizeof(unsigned long);
2121 }
2122 x ^= a & bytemask_from_count(len);
2123done:
2124 return fold_hash(x, y);
2125}
2126EXPORT_SYMBOL(full_name_hash);
2127
2128/* Return the "hash_len" (hash and length) of a null-terminated string */
2129u64 hashlen_string(const void *salt, const char *name)
2130{
2131 unsigned long a = 0, x = 0, y = (unsigned long)salt;
2132 unsigned long adata, mask, len;
2133 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2134
2135 len = 0;
2136 goto inside;
2137
2138 do {
2139 HASH_MIX(x, y, a);
2140 len += sizeof(unsigned long);
2141inside:
2142 a = load_unaligned_zeropad(name+len);
2143 } while (!has_zero(a, &adata, &constants));
2144
2145 adata = prep_zero_mask(a, adata, &constants);
2146 mask = create_zero_mask(adata);
2147 x ^= a & zero_bytemask(mask);
2148
2149 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2150}
2151EXPORT_SYMBOL(hashlen_string);
2152
2153/*
2154 * Calculate the length and hash of the path component, and
2155 * return the "hash_len" as the result.
2156 */
2157static inline u64 hash_name(const void *salt, const char *name)
2158{
2159 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
2160 unsigned long adata, bdata, mask, len;
2161 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2162
2163 len = 0;
2164 goto inside;
2165
2166 do {
2167 HASH_MIX(x, y, a);
2168 len += sizeof(unsigned long);
2169inside:
2170 a = load_unaligned_zeropad(name+len);
2171 b = a ^ REPEAT_BYTE('/');
2172 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
2173
2174 adata = prep_zero_mask(a, adata, &constants);
2175 bdata = prep_zero_mask(b, bdata, &constants);
2176 mask = create_zero_mask(adata | bdata);
2177 x ^= a & zero_bytemask(mask);
2178
2179 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2180}
2181
2182#else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2183
2184/* Return the hash of a string of known length */
2185unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2186{
2187 unsigned long hash = init_name_hash(salt);
2188 while (len--)
2189 hash = partial_name_hash((unsigned char)*name++, hash);
2190 return end_name_hash(hash);
2191}
2192EXPORT_SYMBOL(full_name_hash);
2193
2194/* Return the "hash_len" (hash and length) of a null-terminated string */
2195u64 hashlen_string(const void *salt, const char *name)
2196{
2197 unsigned long hash = init_name_hash(salt);
2198 unsigned long len = 0, c;
2199
2200 c = (unsigned char)*name;
2201 while (c) {
2202 len++;
2203 hash = partial_name_hash(c, hash);
2204 c = (unsigned char)name[len];
2205 }
2206 return hashlen_create(end_name_hash(hash), len);
2207}
2208EXPORT_SYMBOL(hashlen_string);
2209
2210/*
2211 * We know there's a real path component here of at least
2212 * one character.
2213 */
2214static inline u64 hash_name(const void *salt, const char *name)
2215{
2216 unsigned long hash = init_name_hash(salt);
2217 unsigned long len = 0, c;
2218
2219 c = (unsigned char)*name;
2220 do {
2221 len++;
2222 hash = partial_name_hash(c, hash);
2223 c = (unsigned char)name[len];
2224 } while (c && c != '/');
2225 return hashlen_create(end_name_hash(hash), len);
2226}
2227
2228#endif
2229
2230/*
2231 * Name resolution.
2232 * This is the basic name resolution function, turning a pathname into
2233 * the final dentry. We expect 'base' to be positive and a directory.
2234 *
2235 * Returns 0 and nd will have valid dentry and mnt on success.
2236 * Returns error and drops reference to input namei data on failure.
2237 */
2238static int link_path_walk(const char *name, struct nameidata *nd)
2239{
2240 int depth = 0; // depth <= nd->depth
2241 int err;
2242
2243 nd->last_type = LAST_ROOT;
2244 nd->flags |= LOOKUP_PARENT;
2245 if (IS_ERR(name))
2246 return PTR_ERR(name);
2247 while (*name=='/')
2248 name++;
2249 if (!*name) {
2250 nd->dir_mode = 0; // short-circuit the 'hardening' idiocy
2251 return 0;
2252 }
2253
2254 /* At this point we know we have a real path component. */
2255 for(;;) {
2256 struct user_namespace *mnt_userns;
2257 const char *link;
2258 u64 hash_len;
2259 int type;
2260
2261 mnt_userns = mnt_user_ns(nd->path.mnt);
2262 err = may_lookup(mnt_userns, nd);
2263 if (err)
2264 return err;
2265
2266 hash_len = hash_name(nd->path.dentry, name);
2267
2268 type = LAST_NORM;
2269 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2270 case 2:
2271 if (name[1] == '.') {
2272 type = LAST_DOTDOT;
2273 nd->state |= ND_JUMPED;
2274 }
2275 break;
2276 case 1:
2277 type = LAST_DOT;
2278 }
2279 if (likely(type == LAST_NORM)) {
2280 struct dentry *parent = nd->path.dentry;
2281 nd->state &= ~ND_JUMPED;
2282 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2283 struct qstr this = { { .hash_len = hash_len }, .name = name };
2284 err = parent->d_op->d_hash(parent, &this);
2285 if (err < 0)
2286 return err;
2287 hash_len = this.hash_len;
2288 name = this.name;
2289 }
2290 }
2291
2292 nd->last.hash_len = hash_len;
2293 nd->last.name = name;
2294 nd->last_type = type;
2295
2296 name += hashlen_len(hash_len);
2297 if (!*name)
2298 goto OK;
2299 /*
2300 * If it wasn't NUL, we know it was '/'. Skip that
2301 * slash, and continue until no more slashes.
2302 */
2303 do {
2304 name++;
2305 } while (unlikely(*name == '/'));
2306 if (unlikely(!*name)) {
2307OK:
2308 /* pathname or trailing symlink, done */
2309 if (!depth) {
2310 nd->dir_vfsuid = i_uid_into_vfsuid(mnt_userns, nd->inode);
2311 nd->dir_mode = nd->inode->i_mode;
2312 nd->flags &= ~LOOKUP_PARENT;
2313 return 0;
2314 }
2315 /* last component of nested symlink */
2316 name = nd->stack[--depth].name;
2317 link = walk_component(nd, 0);
2318 } else {
2319 /* not the last component */
2320 link = walk_component(nd, WALK_MORE);
2321 }
2322 if (unlikely(link)) {
2323 if (IS_ERR(link))
2324 return PTR_ERR(link);
2325 /* a symlink to follow */
2326 nd->stack[depth++].name = name;
2327 name = link;
2328 continue;
2329 }
2330 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2331 if (nd->flags & LOOKUP_RCU) {
2332 if (!try_to_unlazy(nd))
2333 return -ECHILD;
2334 }
2335 return -ENOTDIR;
2336 }
2337 }
2338}
2339
2340/* must be paired with terminate_walk() */
2341static const char *path_init(struct nameidata *nd, unsigned flags)
2342{
2343 int error;
2344 const char *s = nd->name->name;
2345
2346 /* LOOKUP_CACHED requires RCU, ask caller to retry */
2347 if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED)
2348 return ERR_PTR(-EAGAIN);
2349
2350 if (!*s)
2351 flags &= ~LOOKUP_RCU;
2352 if (flags & LOOKUP_RCU)
2353 rcu_read_lock();
2354 else
2355 nd->seq = nd->next_seq = 0;
2356
2357 nd->flags = flags;
2358 nd->state |= ND_JUMPED;
2359
2360 nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount);
2361 nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount);
2362 smp_rmb();
2363
2364 if (nd->state & ND_ROOT_PRESET) {
2365 struct dentry *root = nd->root.dentry;
2366 struct inode *inode = root->d_inode;
2367 if (*s && unlikely(!d_can_lookup(root)))
2368 return ERR_PTR(-ENOTDIR);
2369 nd->path = nd->root;
2370 nd->inode = inode;
2371 if (flags & LOOKUP_RCU) {
2372 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2373 nd->root_seq = nd->seq;
2374 } else {
2375 path_get(&nd->path);
2376 }
2377 return s;
2378 }
2379
2380 nd->root.mnt = NULL;
2381
2382 /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2383 if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) {
2384 error = nd_jump_root(nd);
2385 if (unlikely(error))
2386 return ERR_PTR(error);
2387 return s;
2388 }
2389
2390 /* Relative pathname -- get the starting-point it is relative to. */
2391 if (nd->dfd == AT_FDCWD) {
2392 if (flags & LOOKUP_RCU) {
2393 struct fs_struct *fs = current->fs;
2394 unsigned seq;
2395
2396 do {
2397 seq = read_seqcount_begin(&fs->seq);
2398 nd->path = fs->pwd;
2399 nd->inode = nd->path.dentry->d_inode;
2400 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2401 } while (read_seqcount_retry(&fs->seq, seq));
2402 } else {
2403 get_fs_pwd(current->fs, &nd->path);
2404 nd->inode = nd->path.dentry->d_inode;
2405 }
2406 } else {
2407 /* Caller must check execute permissions on the starting path component */
2408 struct fd f = fdget_raw(nd->dfd);
2409 struct dentry *dentry;
2410
2411 if (!f.file)
2412 return ERR_PTR(-EBADF);
2413
2414 dentry = f.file->f_path.dentry;
2415
2416 if (*s && unlikely(!d_can_lookup(dentry))) {
2417 fdput(f);
2418 return ERR_PTR(-ENOTDIR);
2419 }
2420
2421 nd->path = f.file->f_path;
2422 if (flags & LOOKUP_RCU) {
2423 nd->inode = nd->path.dentry->d_inode;
2424 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2425 } else {
2426 path_get(&nd->path);
2427 nd->inode = nd->path.dentry->d_inode;
2428 }
2429 fdput(f);
2430 }
2431
2432 /* For scoped-lookups we need to set the root to the dirfd as well. */
2433 if (flags & LOOKUP_IS_SCOPED) {
2434 nd->root = nd->path;
2435 if (flags & LOOKUP_RCU) {
2436 nd->root_seq = nd->seq;
2437 } else {
2438 path_get(&nd->root);
2439 nd->state |= ND_ROOT_GRABBED;
2440 }
2441 }
2442 return s;
2443}
2444
2445static inline const char *lookup_last(struct nameidata *nd)
2446{
2447 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2448 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2449
2450 return walk_component(nd, WALK_TRAILING);
2451}
2452
2453static int handle_lookup_down(struct nameidata *nd)
2454{
2455 if (!(nd->flags & LOOKUP_RCU))
2456 dget(nd->path.dentry);
2457 nd->next_seq = nd->seq;
2458 return PTR_ERR(step_into(nd, WALK_NOFOLLOW, nd->path.dentry));
2459}
2460
2461/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2462static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2463{
2464 const char *s = path_init(nd, flags);
2465 int err;
2466
2467 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2468 err = handle_lookup_down(nd);
2469 if (unlikely(err < 0))
2470 s = ERR_PTR(err);
2471 }
2472
2473 while (!(err = link_path_walk(s, nd)) &&
2474 (s = lookup_last(nd)) != NULL)
2475 ;
2476 if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) {
2477 err = handle_lookup_down(nd);
2478 nd->state &= ~ND_JUMPED; // no d_weak_revalidate(), please...
2479 }
2480 if (!err)
2481 err = complete_walk(nd);
2482
2483 if (!err && nd->flags & LOOKUP_DIRECTORY)
2484 if (!d_can_lookup(nd->path.dentry))
2485 err = -ENOTDIR;
2486 if (!err) {
2487 *path = nd->path;
2488 nd->path.mnt = NULL;
2489 nd->path.dentry = NULL;
2490 }
2491 terminate_walk(nd);
2492 return err;
2493}
2494
2495int filename_lookup(int dfd, struct filename *name, unsigned flags,
2496 struct path *path, struct path *root)
2497{
2498 int retval;
2499 struct nameidata nd;
2500 if (IS_ERR(name))
2501 return PTR_ERR(name);
2502 set_nameidata(&nd, dfd, name, root);
2503 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2504 if (unlikely(retval == -ECHILD))
2505 retval = path_lookupat(&nd, flags, path);
2506 if (unlikely(retval == -ESTALE))
2507 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2508
2509 if (likely(!retval))
2510 audit_inode(name, path->dentry,
2511 flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0);
2512 restore_nameidata();
2513 return retval;
2514}
2515
2516/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2517static int path_parentat(struct nameidata *nd, unsigned flags,
2518 struct path *parent)
2519{
2520 const char *s = path_init(nd, flags);
2521 int err = link_path_walk(s, nd);
2522 if (!err)
2523 err = complete_walk(nd);
2524 if (!err) {
2525 *parent = nd->path;
2526 nd->path.mnt = NULL;
2527 nd->path.dentry = NULL;
2528 }
2529 terminate_walk(nd);
2530 return err;
2531}
2532
2533/* Note: this does not consume "name" */
2534static int filename_parentat(int dfd, struct filename *name,
2535 unsigned int flags, struct path *parent,
2536 struct qstr *last, int *type)
2537{
2538 int retval;
2539 struct nameidata nd;
2540
2541 if (IS_ERR(name))
2542 return PTR_ERR(name);
2543 set_nameidata(&nd, dfd, name, NULL);
2544 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2545 if (unlikely(retval == -ECHILD))
2546 retval = path_parentat(&nd, flags, parent);
2547 if (unlikely(retval == -ESTALE))
2548 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2549 if (likely(!retval)) {
2550 *last = nd.last;
2551 *type = nd.last_type;
2552 audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
2553 }
2554 restore_nameidata();
2555 return retval;
2556}
2557
2558/* does lookup, returns the object with parent locked */
2559static struct dentry *__kern_path_locked(struct filename *name, struct path *path)
2560{
2561 struct dentry *d;
2562 struct qstr last;
2563 int type, error;
2564
2565 error = filename_parentat(AT_FDCWD, name, 0, path, &last, &type);
2566 if (error)
2567 return ERR_PTR(error);
2568 if (unlikely(type != LAST_NORM)) {
2569 path_put(path);
2570 return ERR_PTR(-EINVAL);
2571 }
2572 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2573 d = __lookup_hash(&last, path->dentry, 0);
2574 if (IS_ERR(d)) {
2575 inode_unlock(path->dentry->d_inode);
2576 path_put(path);
2577 }
2578 return d;
2579}
2580
2581struct dentry *kern_path_locked(const char *name, struct path *path)
2582{
2583 struct filename *filename = getname_kernel(name);
2584 struct dentry *res = __kern_path_locked(filename, path);
2585
2586 putname(filename);
2587 return res;
2588}
2589
2590int kern_path(const char *name, unsigned int flags, struct path *path)
2591{
2592 struct filename *filename = getname_kernel(name);
2593 int ret = filename_lookup(AT_FDCWD, filename, flags, path, NULL);
2594
2595 putname(filename);
2596 return ret;
2597
2598}
2599EXPORT_SYMBOL(kern_path);
2600
2601/**
2602 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2603 * @dentry: pointer to dentry of the base directory
2604 * @mnt: pointer to vfs mount of the base directory
2605 * @name: pointer to file name
2606 * @flags: lookup flags
2607 * @path: pointer to struct path to fill
2608 */
2609int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2610 const char *name, unsigned int flags,
2611 struct path *path)
2612{
2613 struct filename *filename;
2614 struct path root = {.mnt = mnt, .dentry = dentry};
2615 int ret;
2616
2617 filename = getname_kernel(name);
2618 /* the first argument of filename_lookup() is ignored with root */
2619 ret = filename_lookup(AT_FDCWD, filename, flags, path, &root);
2620 putname(filename);
2621 return ret;
2622}
2623EXPORT_SYMBOL(vfs_path_lookup);
2624
2625static int lookup_one_common(struct user_namespace *mnt_userns,
2626 const char *name, struct dentry *base, int len,
2627 struct qstr *this)
2628{
2629 this->name = name;
2630 this->len = len;
2631 this->hash = full_name_hash(base, name, len);
2632 if (!len)
2633 return -EACCES;
2634
2635 if (unlikely(name[0] == '.')) {
2636 if (len < 2 || (len == 2 && name[1] == '.'))
2637 return -EACCES;
2638 }
2639
2640 while (len--) {
2641 unsigned int c = *(const unsigned char *)name++;
2642 if (c == '/' || c == '\0')
2643 return -EACCES;
2644 }
2645 /*
2646 * See if the low-level filesystem might want
2647 * to use its own hash..
2648 */
2649 if (base->d_flags & DCACHE_OP_HASH) {
2650 int err = base->d_op->d_hash(base, this);
2651 if (err < 0)
2652 return err;
2653 }
2654
2655 return inode_permission(mnt_userns, base->d_inode, MAY_EXEC);
2656}
2657
2658/**
2659 * try_lookup_one_len - filesystem helper to lookup single pathname component
2660 * @name: pathname component to lookup
2661 * @base: base directory to lookup from
2662 * @len: maximum length @len should be interpreted to
2663 *
2664 * Look up a dentry by name in the dcache, returning NULL if it does not
2665 * currently exist. The function does not try to create a dentry.
2666 *
2667 * Note that this routine is purely a helper for filesystem usage and should
2668 * not be called by generic code.
2669 *
2670 * The caller must hold base->i_mutex.
2671 */
2672struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2673{
2674 struct qstr this;
2675 int err;
2676
2677 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2678
2679 err = lookup_one_common(&init_user_ns, name, base, len, &this);
2680 if (err)
2681 return ERR_PTR(err);
2682
2683 return lookup_dcache(&this, base, 0);
2684}
2685EXPORT_SYMBOL(try_lookup_one_len);
2686
2687/**
2688 * lookup_one_len - filesystem helper to lookup single pathname component
2689 * @name: pathname component to lookup
2690 * @base: base directory to lookup from
2691 * @len: maximum length @len should be interpreted to
2692 *
2693 * Note that this routine is purely a helper for filesystem usage and should
2694 * not be called by generic code.
2695 *
2696 * The caller must hold base->i_mutex.
2697 */
2698struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2699{
2700 struct dentry *dentry;
2701 struct qstr this;
2702 int err;
2703
2704 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2705
2706 err = lookup_one_common(&init_user_ns, name, base, len, &this);
2707 if (err)
2708 return ERR_PTR(err);
2709
2710 dentry = lookup_dcache(&this, base, 0);
2711 return dentry ? dentry : __lookup_slow(&this, base, 0);
2712}
2713EXPORT_SYMBOL(lookup_one_len);
2714
2715/**
2716 * lookup_one - filesystem helper to lookup single pathname component
2717 * @mnt_userns: user namespace of the mount the lookup is performed from
2718 * @name: pathname component to lookup
2719 * @base: base directory to lookup from
2720 * @len: maximum length @len should be interpreted to
2721 *
2722 * Note that this routine is purely a helper for filesystem usage and should
2723 * not be called by generic code.
2724 *
2725 * The caller must hold base->i_mutex.
2726 */
2727struct dentry *lookup_one(struct user_namespace *mnt_userns, const char *name,
2728 struct dentry *base, int len)
2729{
2730 struct dentry *dentry;
2731 struct qstr this;
2732 int err;
2733
2734 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2735
2736 err = lookup_one_common(mnt_userns, name, base, len, &this);
2737 if (err)
2738 return ERR_PTR(err);
2739
2740 dentry = lookup_dcache(&this, base, 0);
2741 return dentry ? dentry : __lookup_slow(&this, base, 0);
2742}
2743EXPORT_SYMBOL(lookup_one);
2744
2745/**
2746 * lookup_one_unlocked - filesystem helper to lookup single pathname component
2747 * @mnt_userns: idmapping of the mount the lookup is performed from
2748 * @name: pathname component to lookup
2749 * @base: base directory to lookup from
2750 * @len: maximum length @len should be interpreted to
2751 *
2752 * Note that this routine is purely a helper for filesystem usage and should
2753 * not be called by generic code.
2754 *
2755 * Unlike lookup_one_len, it should be called without the parent
2756 * i_mutex held, and will take the i_mutex itself if necessary.
2757 */
2758struct dentry *lookup_one_unlocked(struct user_namespace *mnt_userns,
2759 const char *name, struct dentry *base,
2760 int len)
2761{
2762 struct qstr this;
2763 int err;
2764 struct dentry *ret;
2765
2766 err = lookup_one_common(mnt_userns, name, base, len, &this);
2767 if (err)
2768 return ERR_PTR(err);
2769
2770 ret = lookup_dcache(&this, base, 0);
2771 if (!ret)
2772 ret = lookup_slow(&this, base, 0);
2773 return ret;
2774}
2775EXPORT_SYMBOL(lookup_one_unlocked);
2776
2777/**
2778 * lookup_one_positive_unlocked - filesystem helper to lookup single
2779 * pathname component
2780 * @mnt_userns: idmapping of the mount the lookup is performed from
2781 * @name: pathname component to lookup
2782 * @base: base directory to lookup from
2783 * @len: maximum length @len should be interpreted to
2784 *
2785 * This helper will yield ERR_PTR(-ENOENT) on negatives. The helper returns
2786 * known positive or ERR_PTR(). This is what most of the users want.
2787 *
2788 * Note that pinned negative with unlocked parent _can_ become positive at any
2789 * time, so callers of lookup_one_unlocked() need to be very careful; pinned
2790 * positives have >d_inode stable, so this one avoids such problems.
2791 *
2792 * Note that this routine is purely a helper for filesystem usage and should
2793 * not be called by generic code.
2794 *
2795 * The helper should be called without i_mutex held.
2796 */
2797struct dentry *lookup_one_positive_unlocked(struct user_namespace *mnt_userns,
2798 const char *name,
2799 struct dentry *base, int len)
2800{
2801 struct dentry *ret = lookup_one_unlocked(mnt_userns, name, base, len);
2802
2803 if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
2804 dput(ret);
2805 ret = ERR_PTR(-ENOENT);
2806 }
2807 return ret;
2808}
2809EXPORT_SYMBOL(lookup_one_positive_unlocked);
2810
2811/**
2812 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2813 * @name: pathname component to lookup
2814 * @base: base directory to lookup from
2815 * @len: maximum length @len should be interpreted to
2816 *
2817 * Note that this routine is purely a helper for filesystem usage and should
2818 * not be called by generic code.
2819 *
2820 * Unlike lookup_one_len, it should be called without the parent
2821 * i_mutex held, and will take the i_mutex itself if necessary.
2822 */
2823struct dentry *lookup_one_len_unlocked(const char *name,
2824 struct dentry *base, int len)
2825{
2826 return lookup_one_unlocked(&init_user_ns, name, base, len);
2827}
2828EXPORT_SYMBOL(lookup_one_len_unlocked);
2829
2830/*
2831 * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2832 * on negatives. Returns known positive or ERR_PTR(); that's what
2833 * most of the users want. Note that pinned negative with unlocked parent
2834 * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2835 * need to be very careful; pinned positives have ->d_inode stable, so
2836 * this one avoids such problems.
2837 */
2838struct dentry *lookup_positive_unlocked(const char *name,
2839 struct dentry *base, int len)
2840{
2841 return lookup_one_positive_unlocked(&init_user_ns, name, base, len);
2842}
2843EXPORT_SYMBOL(lookup_positive_unlocked);
2844
2845#ifdef CONFIG_UNIX98_PTYS
2846int path_pts(struct path *path)
2847{
2848 /* Find something mounted on "pts" in the same directory as
2849 * the input path.
2850 */
2851 struct dentry *parent = dget_parent(path->dentry);
2852 struct dentry *child;
2853 struct qstr this = QSTR_INIT("pts", 3);
2854
2855 if (unlikely(!path_connected(path->mnt, parent))) {
2856 dput(parent);
2857 return -ENOENT;
2858 }
2859 dput(path->dentry);
2860 path->dentry = parent;
2861 child = d_hash_and_lookup(parent, &this);
2862 if (!child)
2863 return -ENOENT;
2864
2865 path->dentry = child;
2866 dput(parent);
2867 follow_down(path);
2868 return 0;
2869}
2870#endif
2871
2872int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2873 struct path *path, int *empty)
2874{
2875 struct filename *filename = getname_flags(name, flags, empty);
2876 int ret = filename_lookup(dfd, filename, flags, path, NULL);
2877
2878 putname(filename);
2879 return ret;
2880}
2881EXPORT_SYMBOL(user_path_at_empty);
2882
2883int __check_sticky(struct user_namespace *mnt_userns, struct inode *dir,
2884 struct inode *inode)
2885{
2886 kuid_t fsuid = current_fsuid();
2887
2888 if (vfsuid_eq_kuid(i_uid_into_vfsuid(mnt_userns, inode), fsuid))
2889 return 0;
2890 if (vfsuid_eq_kuid(i_uid_into_vfsuid(mnt_userns, dir), fsuid))
2891 return 0;
2892 return !capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FOWNER);
2893}
2894EXPORT_SYMBOL(__check_sticky);
2895
2896/*
2897 * Check whether we can remove a link victim from directory dir, check
2898 * whether the type of victim is right.
2899 * 1. We can't do it if dir is read-only (done in permission())
2900 * 2. We should have write and exec permissions on dir
2901 * 3. We can't remove anything from append-only dir
2902 * 4. We can't do anything with immutable dir (done in permission())
2903 * 5. If the sticky bit on dir is set we should either
2904 * a. be owner of dir, or
2905 * b. be owner of victim, or
2906 * c. have CAP_FOWNER capability
2907 * 6. If the victim is append-only or immutable we can't do antyhing with
2908 * links pointing to it.
2909 * 7. If the victim has an unknown uid or gid we can't change the inode.
2910 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2911 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2912 * 10. We can't remove a root or mountpoint.
2913 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2914 * nfs_async_unlink().
2915 */
2916static int may_delete(struct user_namespace *mnt_userns, struct inode *dir,
2917 struct dentry *victim, bool isdir)
2918{
2919 struct inode *inode = d_backing_inode(victim);
2920 int error;
2921
2922 if (d_is_negative(victim))
2923 return -ENOENT;
2924 BUG_ON(!inode);
2925
2926 BUG_ON(victim->d_parent->d_inode != dir);
2927
2928 /* Inode writeback is not safe when the uid or gid are invalid. */
2929 if (!vfsuid_valid(i_uid_into_vfsuid(mnt_userns, inode)) ||
2930 !vfsgid_valid(i_gid_into_vfsgid(mnt_userns, inode)))
2931 return -EOVERFLOW;
2932
2933 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2934
2935 error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
2936 if (error)
2937 return error;
2938 if (IS_APPEND(dir))
2939 return -EPERM;
2940
2941 if (check_sticky(mnt_userns, dir, inode) || IS_APPEND(inode) ||
2942 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) ||
2943 HAS_UNMAPPED_ID(mnt_userns, inode))
2944 return -EPERM;
2945 if (isdir) {
2946 if (!d_is_dir(victim))
2947 return -ENOTDIR;
2948 if (IS_ROOT(victim))
2949 return -EBUSY;
2950 } else if (d_is_dir(victim))
2951 return -EISDIR;
2952 if (IS_DEADDIR(dir))
2953 return -ENOENT;
2954 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2955 return -EBUSY;
2956 return 0;
2957}
2958
2959/* Check whether we can create an object with dentry child in directory
2960 * dir.
2961 * 1. We can't do it if child already exists (open has special treatment for
2962 * this case, but since we are inlined it's OK)
2963 * 2. We can't do it if dir is read-only (done in permission())
2964 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2965 * 4. We should have write and exec permissions on dir
2966 * 5. We can't do it if dir is immutable (done in permission())
2967 */
2968static inline int may_create(struct user_namespace *mnt_userns,
2969 struct inode *dir, struct dentry *child)
2970{
2971 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2972 if (child->d_inode)
2973 return -EEXIST;
2974 if (IS_DEADDIR(dir))
2975 return -ENOENT;
2976 if (!fsuidgid_has_mapping(dir->i_sb, mnt_userns))
2977 return -EOVERFLOW;
2978
2979 return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
2980}
2981
2982/*
2983 * p1 and p2 should be directories on the same fs.
2984 */
2985struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2986{
2987 struct dentry *p;
2988
2989 if (p1 == p2) {
2990 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2991 return NULL;
2992 }
2993
2994 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2995
2996 p = d_ancestor(p2, p1);
2997 if (p) {
2998 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2999 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
3000 return p;
3001 }
3002
3003 p = d_ancestor(p1, p2);
3004 if (p) {
3005 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3006 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
3007 return p;
3008 }
3009
3010 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3011 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
3012 return NULL;
3013}
3014EXPORT_SYMBOL(lock_rename);
3015
3016void unlock_rename(struct dentry *p1, struct dentry *p2)
3017{
3018 inode_unlock(p1->d_inode);
3019 if (p1 != p2) {
3020 inode_unlock(p2->d_inode);
3021 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
3022 }
3023}
3024EXPORT_SYMBOL(unlock_rename);
3025
3026/**
3027 * mode_strip_umask - handle vfs umask stripping
3028 * @dir: parent directory of the new inode
3029 * @mode: mode of the new inode to be created in @dir
3030 *
3031 * Umask stripping depends on whether or not the filesystem supports POSIX
3032 * ACLs. If the filesystem doesn't support it umask stripping is done directly
3033 * in here. If the filesystem does support POSIX ACLs umask stripping is
3034 * deferred until the filesystem calls posix_acl_create().
3035 *
3036 * Returns: mode
3037 */
3038static inline umode_t mode_strip_umask(const struct inode *dir, umode_t mode)
3039{
3040 if (!IS_POSIXACL(dir))
3041 mode &= ~current_umask();
3042 return mode;
3043}
3044
3045/**
3046 * vfs_prepare_mode - prepare the mode to be used for a new inode
3047 * @mnt_userns: user namespace of the mount the inode was found from
3048 * @dir: parent directory of the new inode
3049 * @mode: mode of the new inode
3050 * @mask_perms: allowed permission by the vfs
3051 * @type: type of file to be created
3052 *
3053 * This helper consolidates and enforces vfs restrictions on the @mode of a new
3054 * object to be created.
3055 *
3056 * Umask stripping depends on whether the filesystem supports POSIX ACLs (see
3057 * the kernel documentation for mode_strip_umask()). Moving umask stripping
3058 * after setgid stripping allows the same ordering for both non-POSIX ACL and
3059 * POSIX ACL supporting filesystems.
3060 *
3061 * Note that it's currently valid for @type to be 0 if a directory is created.
3062 * Filesystems raise that flag individually and we need to check whether each
3063 * filesystem can deal with receiving S_IFDIR from the vfs before we enforce a
3064 * non-zero type.
3065 *
3066 * Returns: mode to be passed to the filesystem
3067 */
3068static inline umode_t vfs_prepare_mode(struct user_namespace *mnt_userns,
3069 const struct inode *dir, umode_t mode,
3070 umode_t mask_perms, umode_t type)
3071{
3072 mode = mode_strip_sgid(mnt_userns, dir, mode);
3073 mode = mode_strip_umask(dir, mode);
3074
3075 /*
3076 * Apply the vfs mandated allowed permission mask and set the type of
3077 * file to be created before we call into the filesystem.
3078 */
3079 mode &= (mask_perms & ~S_IFMT);
3080 mode |= (type & S_IFMT);
3081
3082 return mode;
3083}
3084
3085/**
3086 * vfs_create - create new file
3087 * @mnt_userns: user namespace of the mount the inode was found from
3088 * @dir: inode of @dentry
3089 * @dentry: pointer to dentry of the base directory
3090 * @mode: mode of the new file
3091 * @want_excl: whether the file must not yet exist
3092 *
3093 * Create a new file.
3094 *
3095 * If the inode has been found through an idmapped mount the user namespace of
3096 * the vfsmount must be passed through @mnt_userns. This function will then take
3097 * care to map the inode according to @mnt_userns before checking permissions.
3098 * On non-idmapped mounts or if permission checking is to be performed on the
3099 * raw inode simply passs init_user_ns.
3100 */
3101int vfs_create(struct user_namespace *mnt_userns, struct inode *dir,
3102 struct dentry *dentry, umode_t mode, bool want_excl)
3103{
3104 int error = may_create(mnt_userns, dir, dentry);
3105 if (error)
3106 return error;
3107
3108 if (!dir->i_op->create)
3109 return -EACCES; /* shouldn't it be ENOSYS? */
3110
3111 mode = vfs_prepare_mode(mnt_userns, dir, mode, S_IALLUGO, S_IFREG);
3112 error = security_inode_create(dir, dentry, mode);
3113 if (error)
3114 return error;
3115 error = dir->i_op->create(mnt_userns, dir, dentry, mode, want_excl);
3116 if (!error)
3117 fsnotify_create(dir, dentry);
3118 return error;
3119}
3120EXPORT_SYMBOL(vfs_create);
3121
3122int vfs_mkobj(struct dentry *dentry, umode_t mode,
3123 int (*f)(struct dentry *, umode_t, void *),
3124 void *arg)
3125{
3126 struct inode *dir = dentry->d_parent->d_inode;
3127 int error = may_create(&init_user_ns, dir, dentry);
3128 if (error)
3129 return error;
3130
3131 mode &= S_IALLUGO;
3132 mode |= S_IFREG;
3133 error = security_inode_create(dir, dentry, mode);
3134 if (error)
3135 return error;
3136 error = f(dentry, mode, arg);
3137 if (!error)
3138 fsnotify_create(dir, dentry);
3139 return error;
3140}
3141EXPORT_SYMBOL(vfs_mkobj);
3142
3143bool may_open_dev(const struct path *path)
3144{
3145 return !(path->mnt->mnt_flags & MNT_NODEV) &&
3146 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
3147}
3148
3149static int may_open(struct user_namespace *mnt_userns, const struct path *path,
3150 int acc_mode, int flag)
3151{
3152 struct dentry *dentry = path->dentry;
3153 struct inode *inode = dentry->d_inode;
3154 int error;
3155
3156 if (!inode)
3157 return -ENOENT;
3158
3159 switch (inode->i_mode & S_IFMT) {
3160 case S_IFLNK:
3161 return -ELOOP;
3162 case S_IFDIR:
3163 if (acc_mode & MAY_WRITE)
3164 return -EISDIR;
3165 if (acc_mode & MAY_EXEC)
3166 return -EACCES;
3167 break;
3168 case S_IFBLK:
3169 case S_IFCHR:
3170 if (!may_open_dev(path))
3171 return -EACCES;
3172 fallthrough;
3173 case S_IFIFO:
3174 case S_IFSOCK:
3175 if (acc_mode & MAY_EXEC)
3176 return -EACCES;
3177 flag &= ~O_TRUNC;
3178 break;
3179 case S_IFREG:
3180 if ((acc_mode & MAY_EXEC) && path_noexec(path))
3181 return -EACCES;
3182 break;
3183 }
3184
3185 error = inode_permission(mnt_userns, inode, MAY_OPEN | acc_mode);
3186 if (error)
3187 return error;
3188
3189 /*
3190 * An append-only file must be opened in append mode for writing.
3191 */
3192 if (IS_APPEND(inode)) {
3193 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
3194 return -EPERM;
3195 if (flag & O_TRUNC)
3196 return -EPERM;
3197 }
3198
3199 /* O_NOATIME can only be set by the owner or superuser */
3200 if (flag & O_NOATIME && !inode_owner_or_capable(mnt_userns, inode))
3201 return -EPERM;
3202
3203 return 0;
3204}
3205
3206static int handle_truncate(struct user_namespace *mnt_userns, struct file *filp)
3207{
3208 const struct path *path = &filp->f_path;
3209 struct inode *inode = path->dentry->d_inode;
3210 int error = get_write_access(inode);
3211 if (error)
3212 return error;
3213
3214 error = security_file_truncate(filp);
3215 if (!error) {
3216 error = do_truncate(mnt_userns, path->dentry, 0,
3217 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
3218 filp);
3219 }
3220 put_write_access(inode);
3221 return error;
3222}
3223
3224static inline int open_to_namei_flags(int flag)
3225{
3226 if ((flag & O_ACCMODE) == 3)
3227 flag--;
3228 return flag;
3229}
3230
3231static int may_o_create(struct user_namespace *mnt_userns,
3232 const struct path *dir, struct dentry *dentry,
3233 umode_t mode)
3234{
3235 int error = security_path_mknod(dir, dentry, mode, 0);
3236 if (error)
3237 return error;
3238
3239 if (!fsuidgid_has_mapping(dir->dentry->d_sb, mnt_userns))
3240 return -EOVERFLOW;
3241
3242 error = inode_permission(mnt_userns, dir->dentry->d_inode,
3243 MAY_WRITE | MAY_EXEC);
3244 if (error)
3245 return error;
3246
3247 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3248}
3249
3250/*
3251 * Attempt to atomically look up, create and open a file from a negative
3252 * dentry.
3253 *
3254 * Returns 0 if successful. The file will have been created and attached to
3255 * @file by the filesystem calling finish_open().
3256 *
3257 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3258 * be set. The caller will need to perform the open themselves. @path will
3259 * have been updated to point to the new dentry. This may be negative.
3260 *
3261 * Returns an error code otherwise.
3262 */
3263static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry,
3264 struct file *file,
3265 int open_flag, umode_t mode)
3266{
3267 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3268 struct inode *dir = nd->path.dentry->d_inode;
3269 int error;
3270
3271 if (nd->flags & LOOKUP_DIRECTORY)
3272 open_flag |= O_DIRECTORY;
3273
3274 file->f_path.dentry = DENTRY_NOT_SET;
3275 file->f_path.mnt = nd->path.mnt;
3276 error = dir->i_op->atomic_open(dir, dentry, file,
3277 open_to_namei_flags(open_flag), mode);
3278 d_lookup_done(dentry);
3279 if (!error) {
3280 if (file->f_mode & FMODE_OPENED) {
3281 if (unlikely(dentry != file->f_path.dentry)) {
3282 dput(dentry);
3283 dentry = dget(file->f_path.dentry);
3284 }
3285 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3286 error = -EIO;
3287 } else {
3288 if (file->f_path.dentry) {
3289 dput(dentry);
3290 dentry = file->f_path.dentry;
3291 }
3292 if (unlikely(d_is_negative(dentry)))
3293 error = -ENOENT;
3294 }
3295 }
3296 if (error) {
3297 dput(dentry);
3298 dentry = ERR_PTR(error);
3299 }
3300 return dentry;
3301}
3302
3303/*
3304 * Look up and maybe create and open the last component.
3305 *
3306 * Must be called with parent locked (exclusive in O_CREAT case).
3307 *
3308 * Returns 0 on success, that is, if
3309 * the file was successfully atomically created (if necessary) and opened, or
3310 * the file was not completely opened at this time, though lookups and
3311 * creations were performed.
3312 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3313 * In the latter case dentry returned in @path might be negative if O_CREAT
3314 * hadn't been specified.
3315 *
3316 * An error code is returned on failure.
3317 */
3318static struct dentry *lookup_open(struct nameidata *nd, struct file *file,
3319 const struct open_flags *op,
3320 bool got_write)
3321{
3322 struct user_namespace *mnt_userns;
3323 struct dentry *dir = nd->path.dentry;
3324 struct inode *dir_inode = dir->d_inode;
3325 int open_flag = op->open_flag;
3326 struct dentry *dentry;
3327 int error, create_error = 0;
3328 umode_t mode = op->mode;
3329 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3330
3331 if (unlikely(IS_DEADDIR(dir_inode)))
3332 return ERR_PTR(-ENOENT);
3333
3334 file->f_mode &= ~FMODE_CREATED;
3335 dentry = d_lookup(dir, &nd->last);
3336 for (;;) {
3337 if (!dentry) {
3338 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3339 if (IS_ERR(dentry))
3340 return dentry;
3341 }
3342 if (d_in_lookup(dentry))
3343 break;
3344
3345 error = d_revalidate(dentry, nd->flags);
3346 if (likely(error > 0))
3347 break;
3348 if (error)
3349 goto out_dput;
3350 d_invalidate(dentry);
3351 dput(dentry);
3352 dentry = NULL;
3353 }
3354 if (dentry->d_inode) {
3355 /* Cached positive dentry: will open in f_op->open */
3356 return dentry;
3357 }
3358
3359 /*
3360 * Checking write permission is tricky, bacuse we don't know if we are
3361 * going to actually need it: O_CREAT opens should work as long as the
3362 * file exists. But checking existence breaks atomicity. The trick is
3363 * to check access and if not granted clear O_CREAT from the flags.
3364 *
3365 * Another problem is returing the "right" error value (e.g. for an
3366 * O_EXCL open we want to return EEXIST not EROFS).
3367 */
3368 if (unlikely(!got_write))
3369 open_flag &= ~O_TRUNC;
3370 mnt_userns = mnt_user_ns(nd->path.mnt);
3371 if (open_flag & O_CREAT) {
3372 if (open_flag & O_EXCL)
3373 open_flag &= ~O_TRUNC;
3374 mode = vfs_prepare_mode(mnt_userns, dir->d_inode, mode, mode, mode);
3375 if (likely(got_write))
3376 create_error = may_o_create(mnt_userns, &nd->path,
3377 dentry, mode);
3378 else
3379 create_error = -EROFS;
3380 }
3381 if (create_error)
3382 open_flag &= ~O_CREAT;
3383 if (dir_inode->i_op->atomic_open) {
3384 dentry = atomic_open(nd, dentry, file, open_flag, mode);
3385 if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT))
3386 dentry = ERR_PTR(create_error);
3387 return dentry;
3388 }
3389
3390 if (d_in_lookup(dentry)) {
3391 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3392 nd->flags);
3393 d_lookup_done(dentry);
3394 if (unlikely(res)) {
3395 if (IS_ERR(res)) {
3396 error = PTR_ERR(res);
3397 goto out_dput;
3398 }
3399 dput(dentry);
3400 dentry = res;
3401 }
3402 }
3403
3404 /* Negative dentry, just create the file */
3405 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3406 file->f_mode |= FMODE_CREATED;
3407 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3408 if (!dir_inode->i_op->create) {
3409 error = -EACCES;
3410 goto out_dput;
3411 }
3412
3413 error = dir_inode->i_op->create(mnt_userns, dir_inode, dentry,
3414 mode, open_flag & O_EXCL);
3415 if (error)
3416 goto out_dput;
3417 }
3418 if (unlikely(create_error) && !dentry->d_inode) {
3419 error = create_error;
3420 goto out_dput;
3421 }
3422 return dentry;
3423
3424out_dput:
3425 dput(dentry);
3426 return ERR_PTR(error);
3427}
3428
3429static const char *open_last_lookups(struct nameidata *nd,
3430 struct file *file, const struct open_flags *op)
3431{
3432 struct dentry *dir = nd->path.dentry;
3433 int open_flag = op->open_flag;
3434 bool got_write = false;
3435 struct dentry *dentry;
3436 const char *res;
3437
3438 nd->flags |= op->intent;
3439
3440 if (nd->last_type != LAST_NORM) {
3441 if (nd->depth)
3442 put_link(nd);
3443 return handle_dots(nd, nd->last_type);
3444 }
3445
3446 if (!(open_flag & O_CREAT)) {
3447 if (nd->last.name[nd->last.len])
3448 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3449 /* we _can_ be in RCU mode here */
3450 dentry = lookup_fast(nd);
3451 if (IS_ERR(dentry))
3452 return ERR_CAST(dentry);
3453 if (likely(dentry))
3454 goto finish_lookup;
3455
3456 BUG_ON(nd->flags & LOOKUP_RCU);
3457 } else {
3458 /* create side of things */
3459 if (nd->flags & LOOKUP_RCU) {
3460 if (!try_to_unlazy(nd))
3461 return ERR_PTR(-ECHILD);
3462 }
3463 audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
3464 /* trailing slashes? */
3465 if (unlikely(nd->last.name[nd->last.len]))
3466 return ERR_PTR(-EISDIR);
3467 }
3468
3469 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3470 got_write = !mnt_want_write(nd->path.mnt);
3471 /*
3472 * do _not_ fail yet - we might not need that or fail with
3473 * a different error; let lookup_open() decide; we'll be
3474 * dropping this one anyway.
3475 */
3476 }
3477 if (open_flag & O_CREAT)
3478 inode_lock(dir->d_inode);
3479 else
3480 inode_lock_shared(dir->d_inode);
3481 dentry = lookup_open(nd, file, op, got_write);
3482 if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED))
3483 fsnotify_create(dir->d_inode, dentry);
3484 if (open_flag & O_CREAT)
3485 inode_unlock(dir->d_inode);
3486 else
3487 inode_unlock_shared(dir->d_inode);
3488
3489 if (got_write)
3490 mnt_drop_write(nd->path.mnt);
3491
3492 if (IS_ERR(dentry))
3493 return ERR_CAST(dentry);
3494
3495 if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) {
3496 dput(nd->path.dentry);
3497 nd->path.dentry = dentry;
3498 return NULL;
3499 }
3500
3501finish_lookup:
3502 if (nd->depth)
3503 put_link(nd);
3504 res = step_into(nd, WALK_TRAILING, dentry);
3505 if (unlikely(res))
3506 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3507 return res;
3508}
3509
3510/*
3511 * Handle the last step of open()
3512 */
3513static int do_open(struct nameidata *nd,
3514 struct file *file, const struct open_flags *op)
3515{
3516 struct user_namespace *mnt_userns;
3517 int open_flag = op->open_flag;
3518 bool do_truncate;
3519 int acc_mode;
3520 int error;
3521
3522 if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) {
3523 error = complete_walk(nd);
3524 if (error)
3525 return error;
3526 }
3527 if (!(file->f_mode & FMODE_CREATED))
3528 audit_inode(nd->name, nd->path.dentry, 0);
3529 mnt_userns = mnt_user_ns(nd->path.mnt);
3530 if (open_flag & O_CREAT) {
3531 if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED))
3532 return -EEXIST;
3533 if (d_is_dir(nd->path.dentry))
3534 return -EISDIR;
3535 error = may_create_in_sticky(mnt_userns, nd,
3536 d_backing_inode(nd->path.dentry));
3537 if (unlikely(error))
3538 return error;
3539 }
3540 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3541 return -ENOTDIR;
3542
3543 do_truncate = false;
3544 acc_mode = op->acc_mode;
3545 if (file->f_mode & FMODE_CREATED) {
3546 /* Don't check for write permission, don't truncate */
3547 open_flag &= ~O_TRUNC;
3548 acc_mode = 0;
3549 } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) {
3550 error = mnt_want_write(nd->path.mnt);
3551 if (error)
3552 return error;
3553 do_truncate = true;
3554 }
3555 error = may_open(mnt_userns, &nd->path, acc_mode, open_flag);
3556 if (!error && !(file->f_mode & FMODE_OPENED))
3557 error = vfs_open(&nd->path, file);
3558 if (!error)
3559 error = ima_file_check(file, op->acc_mode);
3560 if (!error && do_truncate)
3561 error = handle_truncate(mnt_userns, file);
3562 if (unlikely(error > 0)) {
3563 WARN_ON(1);
3564 error = -EINVAL;
3565 }
3566 if (do_truncate)
3567 mnt_drop_write(nd->path.mnt);
3568 return error;
3569}
3570
3571/**
3572 * vfs_tmpfile - create tmpfile
3573 * @mnt_userns: user namespace of the mount the inode was found from
3574 * @dentry: pointer to dentry of the base directory
3575 * @mode: mode of the new tmpfile
3576 * @open_flag: flags
3577 *
3578 * Create a temporary file.
3579 *
3580 * If the inode has been found through an idmapped mount the user namespace of
3581 * the vfsmount must be passed through @mnt_userns. This function will then take
3582 * care to map the inode according to @mnt_userns before checking permissions.
3583 * On non-idmapped mounts or if permission checking is to be performed on the
3584 * raw inode simply passs init_user_ns.
3585 */
3586static int vfs_tmpfile(struct user_namespace *mnt_userns,
3587 const struct path *parentpath,
3588 struct file *file, umode_t mode)
3589{
3590 struct dentry *child;
3591 struct inode *dir = d_inode(parentpath->dentry);
3592 struct inode *inode;
3593 int error;
3594 int open_flag = file->f_flags;
3595
3596 /* we want directory to be writable */
3597 error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
3598 if (error)
3599 return error;
3600 if (!dir->i_op->tmpfile)
3601 return -EOPNOTSUPP;
3602 child = d_alloc(parentpath->dentry, &slash_name);
3603 if (unlikely(!child))
3604 return -ENOMEM;
3605 file->f_path.mnt = parentpath->mnt;
3606 file->f_path.dentry = child;
3607 mode = vfs_prepare_mode(mnt_userns, dir, mode, mode, mode);
3608 error = dir->i_op->tmpfile(mnt_userns, dir, file, mode);
3609 dput(child);
3610 if (error)
3611 return error;
3612 /* Don't check for other permissions, the inode was just created */
3613 error = may_open(mnt_userns, &file->f_path, 0, file->f_flags);
3614 if (error)
3615 return error;
3616 inode = file_inode(file);
3617 if (!(open_flag & O_EXCL)) {
3618 spin_lock(&inode->i_lock);
3619 inode->i_state |= I_LINKABLE;
3620 spin_unlock(&inode->i_lock);
3621 }
3622 ima_post_create_tmpfile(mnt_userns, inode);
3623 return 0;
3624}
3625
3626/**
3627 * vfs_tmpfile_open - open a tmpfile for kernel internal use
3628 * @mnt_userns: user namespace of the mount the inode was found from
3629 * @parentpath: path of the base directory
3630 * @mode: mode of the new tmpfile
3631 * @open_flag: flags
3632 * @cred: credentials for open
3633 *
3634 * Create and open a temporary file. The file is not accounted in nr_files,
3635 * hence this is only for kernel internal use, and must not be installed into
3636 * file tables or such.
3637 */
3638struct file *vfs_tmpfile_open(struct user_namespace *mnt_userns,
3639 const struct path *parentpath,
3640 umode_t mode, int open_flag, const struct cred *cred)
3641{
3642 struct file *file;
3643 int error;
3644
3645 file = alloc_empty_file_noaccount(open_flag, cred);
3646 if (!IS_ERR(file)) {
3647 error = vfs_tmpfile(mnt_userns, parentpath, file, mode);
3648 if (error) {
3649 fput(file);
3650 file = ERR_PTR(error);
3651 }
3652 }
3653 return file;
3654}
3655EXPORT_SYMBOL(vfs_tmpfile_open);
3656
3657static int do_tmpfile(struct nameidata *nd, unsigned flags,
3658 const struct open_flags *op,
3659 struct file *file)
3660{
3661 struct user_namespace *mnt_userns;
3662 struct path path;
3663 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3664
3665 if (unlikely(error))
3666 return error;
3667 error = mnt_want_write(path.mnt);
3668 if (unlikely(error))
3669 goto out;
3670 mnt_userns = mnt_user_ns(path.mnt);
3671 error = vfs_tmpfile(mnt_userns, &path, file, op->mode);
3672 if (error)
3673 goto out2;
3674 audit_inode(nd->name, file->f_path.dentry, 0);
3675out2:
3676 mnt_drop_write(path.mnt);
3677out:
3678 path_put(&path);
3679 return error;
3680}
3681
3682static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3683{
3684 struct path path;
3685 int error = path_lookupat(nd, flags, &path);
3686 if (!error) {
3687 audit_inode(nd->name, path.dentry, 0);
3688 error = vfs_open(&path, file);
3689 path_put(&path);
3690 }
3691 return error;
3692}
3693
3694static struct file *path_openat(struct nameidata *nd,
3695 const struct open_flags *op, unsigned flags)
3696{
3697 struct file *file;
3698 int error;
3699
3700 file = alloc_empty_file(op->open_flag, current_cred());
3701 if (IS_ERR(file))
3702 return file;
3703
3704 if (unlikely(file->f_flags & __O_TMPFILE)) {
3705 error = do_tmpfile(nd, flags, op, file);
3706 } else if (unlikely(file->f_flags & O_PATH)) {
3707 error = do_o_path(nd, flags, file);
3708 } else {
3709 const char *s = path_init(nd, flags);
3710 while (!(error = link_path_walk(s, nd)) &&
3711 (s = open_last_lookups(nd, file, op)) != NULL)
3712 ;
3713 if (!error)
3714 error = do_open(nd, file, op);
3715 terminate_walk(nd);
3716 }
3717 if (likely(!error)) {
3718 if (likely(file->f_mode & FMODE_OPENED))
3719 return file;
3720 WARN_ON(1);
3721 error = -EINVAL;
3722 }
3723 fput(file);
3724 if (error == -EOPENSTALE) {
3725 if (flags & LOOKUP_RCU)
3726 error = -ECHILD;
3727 else
3728 error = -ESTALE;
3729 }
3730 return ERR_PTR(error);
3731}
3732
3733struct file *do_filp_open(int dfd, struct filename *pathname,
3734 const struct open_flags *op)
3735{
3736 struct nameidata nd;
3737 int flags = op->lookup_flags;
3738 struct file *filp;
3739
3740 set_nameidata(&nd, dfd, pathname, NULL);
3741 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3742 if (unlikely(filp == ERR_PTR(-ECHILD)))
3743 filp = path_openat(&nd, op, flags);
3744 if (unlikely(filp == ERR_PTR(-ESTALE)))
3745 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3746 restore_nameidata();
3747 return filp;
3748}
3749
3750struct file *do_file_open_root(const struct path *root,
3751 const char *name, const struct open_flags *op)
3752{
3753 struct nameidata nd;
3754 struct file *file;
3755 struct filename *filename;
3756 int flags = op->lookup_flags;
3757
3758 if (d_is_symlink(root->dentry) && op->intent & LOOKUP_OPEN)
3759 return ERR_PTR(-ELOOP);
3760
3761 filename = getname_kernel(name);
3762 if (IS_ERR(filename))
3763 return ERR_CAST(filename);
3764
3765 set_nameidata(&nd, -1, filename, root);
3766 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3767 if (unlikely(file == ERR_PTR(-ECHILD)))
3768 file = path_openat(&nd, op, flags);
3769 if (unlikely(file == ERR_PTR(-ESTALE)))
3770 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3771 restore_nameidata();
3772 putname(filename);
3773 return file;
3774}
3775
3776static struct dentry *filename_create(int dfd, struct filename *name,
3777 struct path *path, unsigned int lookup_flags)
3778{
3779 struct dentry *dentry = ERR_PTR(-EEXIST);
3780 struct qstr last;
3781 bool want_dir = lookup_flags & LOOKUP_DIRECTORY;
3782 unsigned int reval_flag = lookup_flags & LOOKUP_REVAL;
3783 unsigned int create_flags = LOOKUP_CREATE | LOOKUP_EXCL;
3784 int type;
3785 int err2;
3786 int error;
3787
3788 error = filename_parentat(dfd, name, reval_flag, path, &last, &type);
3789 if (error)
3790 return ERR_PTR(error);
3791
3792 /*
3793 * Yucky last component or no last component at all?
3794 * (foo/., foo/.., /////)
3795 */
3796 if (unlikely(type != LAST_NORM))
3797 goto out;
3798
3799 /* don't fail immediately if it's r/o, at least try to report other errors */
3800 err2 = mnt_want_write(path->mnt);
3801 /*
3802 * Do the final lookup. Suppress 'create' if there is a trailing
3803 * '/', and a directory wasn't requested.
3804 */
3805 if (last.name[last.len] && !want_dir)
3806 create_flags = 0;
3807 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3808 dentry = __lookup_hash(&last, path->dentry, reval_flag | create_flags);
3809 if (IS_ERR(dentry))
3810 goto unlock;
3811
3812 error = -EEXIST;
3813 if (d_is_positive(dentry))
3814 goto fail;
3815
3816 /*
3817 * Special case - lookup gave negative, but... we had foo/bar/
3818 * From the vfs_mknod() POV we just have a negative dentry -
3819 * all is fine. Let's be bastards - you had / on the end, you've
3820 * been asking for (non-existent) directory. -ENOENT for you.
3821 */
3822 if (unlikely(!create_flags)) {
3823 error = -ENOENT;
3824 goto fail;
3825 }
3826 if (unlikely(err2)) {
3827 error = err2;
3828 goto fail;
3829 }
3830 return dentry;
3831fail:
3832 dput(dentry);
3833 dentry = ERR_PTR(error);
3834unlock:
3835 inode_unlock(path->dentry->d_inode);
3836 if (!err2)
3837 mnt_drop_write(path->mnt);
3838out:
3839 path_put(path);
3840 return dentry;
3841}
3842
3843struct dentry *kern_path_create(int dfd, const char *pathname,
3844 struct path *path, unsigned int lookup_flags)
3845{
3846 struct filename *filename = getname_kernel(pathname);
3847 struct dentry *res = filename_create(dfd, filename, path, lookup_flags);
3848
3849 putname(filename);
3850 return res;
3851}
3852EXPORT_SYMBOL(kern_path_create);
3853
3854void done_path_create(struct path *path, struct dentry *dentry)
3855{
3856 dput(dentry);
3857 inode_unlock(path->dentry->d_inode);
3858 mnt_drop_write(path->mnt);
3859 path_put(path);
3860}
3861EXPORT_SYMBOL(done_path_create);
3862
3863inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3864 struct path *path, unsigned int lookup_flags)
3865{
3866 struct filename *filename = getname(pathname);
3867 struct dentry *res = filename_create(dfd, filename, path, lookup_flags);
3868
3869 putname(filename);
3870 return res;
3871}
3872EXPORT_SYMBOL(user_path_create);
3873
3874/**
3875 * vfs_mknod - create device node or file
3876 * @mnt_userns: user namespace of the mount the inode was found from
3877 * @dir: inode of @dentry
3878 * @dentry: pointer to dentry of the base directory
3879 * @mode: mode of the new device node or file
3880 * @dev: device number of device to create
3881 *
3882 * Create a device node or file.
3883 *
3884 * If the inode has been found through an idmapped mount the user namespace of
3885 * the vfsmount must be passed through @mnt_userns. This function will then take
3886 * care to map the inode according to @mnt_userns before checking permissions.
3887 * On non-idmapped mounts or if permission checking is to be performed on the
3888 * raw inode simply passs init_user_ns.
3889 */
3890int vfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
3891 struct dentry *dentry, umode_t mode, dev_t dev)
3892{
3893 bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV;
3894 int error = may_create(mnt_userns, dir, dentry);
3895
3896 if (error)
3897 return error;
3898
3899 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout &&
3900 !capable(CAP_MKNOD))
3901 return -EPERM;
3902
3903 if (!dir->i_op->mknod)
3904 return -EPERM;
3905
3906 mode = vfs_prepare_mode(mnt_userns, dir, mode, mode, mode);
3907 error = devcgroup_inode_mknod(mode, dev);
3908 if (error)
3909 return error;
3910
3911 error = security_inode_mknod(dir, dentry, mode, dev);
3912 if (error)
3913 return error;
3914
3915 error = dir->i_op->mknod(mnt_userns, dir, dentry, mode, dev);
3916 if (!error)
3917 fsnotify_create(dir, dentry);
3918 return error;
3919}
3920EXPORT_SYMBOL(vfs_mknod);
3921
3922static int may_mknod(umode_t mode)
3923{
3924 switch (mode & S_IFMT) {
3925 case S_IFREG:
3926 case S_IFCHR:
3927 case S_IFBLK:
3928 case S_IFIFO:
3929 case S_IFSOCK:
3930 case 0: /* zero mode translates to S_IFREG */
3931 return 0;
3932 case S_IFDIR:
3933 return -EPERM;
3934 default:
3935 return -EINVAL;
3936 }
3937}
3938
3939static int do_mknodat(int dfd, struct filename *name, umode_t mode,
3940 unsigned int dev)
3941{
3942 struct user_namespace *mnt_userns;
3943 struct dentry *dentry;
3944 struct path path;
3945 int error;
3946 unsigned int lookup_flags = 0;
3947
3948 error = may_mknod(mode);
3949 if (error)
3950 goto out1;
3951retry:
3952 dentry = filename_create(dfd, name, &path, lookup_flags);
3953 error = PTR_ERR(dentry);
3954 if (IS_ERR(dentry))
3955 goto out1;
3956
3957 error = security_path_mknod(&path, dentry,
3958 mode_strip_umask(path.dentry->d_inode, mode), dev);
3959 if (error)
3960 goto out2;
3961
3962 mnt_userns = mnt_user_ns(path.mnt);
3963 switch (mode & S_IFMT) {
3964 case 0: case S_IFREG:
3965 error = vfs_create(mnt_userns, path.dentry->d_inode,
3966 dentry, mode, true);
3967 if (!error)
3968 ima_post_path_mknod(mnt_userns, dentry);
3969 break;
3970 case S_IFCHR: case S_IFBLK:
3971 error = vfs_mknod(mnt_userns, path.dentry->d_inode,
3972 dentry, mode, new_decode_dev(dev));
3973 break;
3974 case S_IFIFO: case S_IFSOCK:
3975 error = vfs_mknod(mnt_userns, path.dentry->d_inode,
3976 dentry, mode, 0);
3977 break;
3978 }
3979out2:
3980 done_path_create(&path, dentry);
3981 if (retry_estale(error, lookup_flags)) {
3982 lookup_flags |= LOOKUP_REVAL;
3983 goto retry;
3984 }
3985out1:
3986 putname(name);
3987 return error;
3988}
3989
3990SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3991 unsigned int, dev)
3992{
3993 return do_mknodat(dfd, getname(filename), mode, dev);
3994}
3995
3996SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3997{
3998 return do_mknodat(AT_FDCWD, getname(filename), mode, dev);
3999}
4000
4001/**
4002 * vfs_mkdir - create directory
4003 * @mnt_userns: user namespace of the mount the inode was found from
4004 * @dir: inode of @dentry
4005 * @dentry: pointer to dentry of the base directory
4006 * @mode: mode of the new directory
4007 *
4008 * Create a directory.
4009 *
4010 * If the inode has been found through an idmapped mount the user namespace of
4011 * the vfsmount must be passed through @mnt_userns. This function will then take
4012 * care to map the inode according to @mnt_userns before checking permissions.
4013 * On non-idmapped mounts or if permission checking is to be performed on the
4014 * raw inode simply passs init_user_ns.
4015 */
4016int vfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
4017 struct dentry *dentry, umode_t mode)
4018{
4019 int error = may_create(mnt_userns, dir, dentry);
4020 unsigned max_links = dir->i_sb->s_max_links;
4021
4022 if (error)
4023 return error;
4024
4025 if (!dir->i_op->mkdir)
4026 return -EPERM;
4027
4028 mode = vfs_prepare_mode(mnt_userns, dir, mode, S_IRWXUGO | S_ISVTX, 0);
4029 error = security_inode_mkdir(dir, dentry, mode);
4030 if (error)
4031 return error;
4032
4033 if (max_links && dir->i_nlink >= max_links)
4034 return -EMLINK;
4035
4036 error = dir->i_op->mkdir(mnt_userns, dir, dentry, mode);
4037 if (!error)
4038 fsnotify_mkdir(dir, dentry);
4039 return error;
4040}
4041EXPORT_SYMBOL(vfs_mkdir);
4042
4043int do_mkdirat(int dfd, struct filename *name, umode_t mode)
4044{
4045 struct dentry *dentry;
4046 struct path path;
4047 int error;
4048 unsigned int lookup_flags = LOOKUP_DIRECTORY;
4049
4050retry:
4051 dentry = filename_create(dfd, name, &path, lookup_flags);
4052 error = PTR_ERR(dentry);
4053 if (IS_ERR(dentry))
4054 goto out_putname;
4055
4056 error = security_path_mkdir(&path, dentry,
4057 mode_strip_umask(path.dentry->d_inode, mode));
4058 if (!error) {
4059 struct user_namespace *mnt_userns;
4060 mnt_userns = mnt_user_ns(path.mnt);
4061 error = vfs_mkdir(mnt_userns, path.dentry->d_inode, dentry,
4062 mode);
4063 }
4064 done_path_create(&path, dentry);
4065 if (retry_estale(error, lookup_flags)) {
4066 lookup_flags |= LOOKUP_REVAL;
4067 goto retry;
4068 }
4069out_putname:
4070 putname(name);
4071 return error;
4072}
4073
4074SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
4075{
4076 return do_mkdirat(dfd, getname(pathname), mode);
4077}
4078
4079SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
4080{
4081 return do_mkdirat(AT_FDCWD, getname(pathname), mode);
4082}
4083
4084/**
4085 * vfs_rmdir - remove directory
4086 * @mnt_userns: user namespace of the mount the inode was found from
4087 * @dir: inode of @dentry
4088 * @dentry: pointer to dentry of the base directory
4089 *
4090 * Remove a directory.
4091 *
4092 * If the inode has been found through an idmapped mount the user namespace of
4093 * the vfsmount must be passed through @mnt_userns. This function will then take
4094 * care to map the inode according to @mnt_userns before checking permissions.
4095 * On non-idmapped mounts or if permission checking is to be performed on the
4096 * raw inode simply passs init_user_ns.
4097 */
4098int vfs_rmdir(struct user_namespace *mnt_userns, struct inode *dir,
4099 struct dentry *dentry)
4100{
4101 int error = may_delete(mnt_userns, dir, dentry, 1);
4102
4103 if (error)
4104 return error;
4105
4106 if (!dir->i_op->rmdir)
4107 return -EPERM;
4108
4109 dget(dentry);
4110 inode_lock(dentry->d_inode);
4111
4112 error = -EBUSY;
4113 if (is_local_mountpoint(dentry) ||
4114 (dentry->d_inode->i_flags & S_KERNEL_FILE))
4115 goto out;
4116
4117 error = security_inode_rmdir(dir, dentry);
4118 if (error)
4119 goto out;
4120
4121 error = dir->i_op->rmdir(dir, dentry);
4122 if (error)
4123 goto out;
4124
4125 shrink_dcache_parent(dentry);
4126 dentry->d_inode->i_flags |= S_DEAD;
4127 dont_mount(dentry);
4128 detach_mounts(dentry);
4129
4130out:
4131 inode_unlock(dentry->d_inode);
4132 dput(dentry);
4133 if (!error)
4134 d_delete_notify(dir, dentry);
4135 return error;
4136}
4137EXPORT_SYMBOL(vfs_rmdir);
4138
4139int do_rmdir(int dfd, struct filename *name)
4140{
4141 struct user_namespace *mnt_userns;
4142 int error;
4143 struct dentry *dentry;
4144 struct path path;
4145 struct qstr last;
4146 int type;
4147 unsigned int lookup_flags = 0;
4148retry:
4149 error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4150 if (error)
4151 goto exit1;
4152
4153 switch (type) {
4154 case LAST_DOTDOT:
4155 error = -ENOTEMPTY;
4156 goto exit2;
4157 case LAST_DOT:
4158 error = -EINVAL;
4159 goto exit2;
4160 case LAST_ROOT:
4161 error = -EBUSY;
4162 goto exit2;
4163 }
4164
4165 error = mnt_want_write(path.mnt);
4166 if (error)
4167 goto exit2;
4168
4169 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4170 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4171 error = PTR_ERR(dentry);
4172 if (IS_ERR(dentry))
4173 goto exit3;
4174 if (!dentry->d_inode) {
4175 error = -ENOENT;
4176 goto exit4;
4177 }
4178 error = security_path_rmdir(&path, dentry);
4179 if (error)
4180 goto exit4;
4181 mnt_userns = mnt_user_ns(path.mnt);
4182 error = vfs_rmdir(mnt_userns, path.dentry->d_inode, dentry);
4183exit4:
4184 dput(dentry);
4185exit3:
4186 inode_unlock(path.dentry->d_inode);
4187 mnt_drop_write(path.mnt);
4188exit2:
4189 path_put(&path);
4190 if (retry_estale(error, lookup_flags)) {
4191 lookup_flags |= LOOKUP_REVAL;
4192 goto retry;
4193 }
4194exit1:
4195 putname(name);
4196 return error;
4197}
4198
4199SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
4200{
4201 return do_rmdir(AT_FDCWD, getname(pathname));
4202}
4203
4204/**
4205 * vfs_unlink - unlink a filesystem object
4206 * @mnt_userns: user namespace of the mount the inode was found from
4207 * @dir: parent directory
4208 * @dentry: victim
4209 * @delegated_inode: returns victim inode, if the inode is delegated.
4210 *
4211 * The caller must hold dir->i_mutex.
4212 *
4213 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
4214 * return a reference to the inode in delegated_inode. The caller
4215 * should then break the delegation on that inode and retry. Because
4216 * breaking a delegation may take a long time, the caller should drop
4217 * dir->i_mutex before doing so.
4218 *
4219 * Alternatively, a caller may pass NULL for delegated_inode. This may
4220 * be appropriate for callers that expect the underlying filesystem not
4221 * to be NFS exported.
4222 *
4223 * If the inode has been found through an idmapped mount the user namespace of
4224 * the vfsmount must be passed through @mnt_userns. This function will then take
4225 * care to map the inode according to @mnt_userns before checking permissions.
4226 * On non-idmapped mounts or if permission checking is to be performed on the
4227 * raw inode simply passs init_user_ns.
4228 */
4229int vfs_unlink(struct user_namespace *mnt_userns, struct inode *dir,
4230 struct dentry *dentry, struct inode **delegated_inode)
4231{
4232 struct inode *target = dentry->d_inode;
4233 int error = may_delete(mnt_userns, dir, dentry, 0);
4234
4235 if (error)
4236 return error;
4237
4238 if (!dir->i_op->unlink)
4239 return -EPERM;
4240
4241 inode_lock(target);
4242 if (IS_SWAPFILE(target))
4243 error = -EPERM;
4244 else if (is_local_mountpoint(dentry))
4245 error = -EBUSY;
4246 else {
4247 error = security_inode_unlink(dir, dentry);
4248 if (!error) {
4249 error = try_break_deleg(target, delegated_inode);
4250 if (error)
4251 goto out;
4252 error = dir->i_op->unlink(dir, dentry);
4253 if (!error) {
4254 dont_mount(dentry);
4255 detach_mounts(dentry);
4256 }
4257 }
4258 }
4259out:
4260 inode_unlock(target);
4261
4262 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
4263 if (!error && dentry->d_flags & DCACHE_NFSFS_RENAMED) {
4264 fsnotify_unlink(dir, dentry);
4265 } else if (!error) {
4266 fsnotify_link_count(target);
4267 d_delete_notify(dir, dentry);
4268 }
4269
4270 return error;
4271}
4272EXPORT_SYMBOL(vfs_unlink);
4273
4274/*
4275 * Make sure that the actual truncation of the file will occur outside its
4276 * directory's i_mutex. Truncate can take a long time if there is a lot of
4277 * writeout happening, and we don't want to prevent access to the directory
4278 * while waiting on the I/O.
4279 */
4280int do_unlinkat(int dfd, struct filename *name)
4281{
4282 int error;
4283 struct dentry *dentry;
4284 struct path path;
4285 struct qstr last;
4286 int type;
4287 struct inode *inode = NULL;
4288 struct inode *delegated_inode = NULL;
4289 unsigned int lookup_flags = 0;
4290retry:
4291 error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4292 if (error)
4293 goto exit1;
4294
4295 error = -EISDIR;
4296 if (type != LAST_NORM)
4297 goto exit2;
4298
4299 error = mnt_want_write(path.mnt);
4300 if (error)
4301 goto exit2;
4302retry_deleg:
4303 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4304 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4305 error = PTR_ERR(dentry);
4306 if (!IS_ERR(dentry)) {
4307 struct user_namespace *mnt_userns;
4308
4309 /* Why not before? Because we want correct error value */
4310 if (last.name[last.len])
4311 goto slashes;
4312 inode = dentry->d_inode;
4313 if (d_is_negative(dentry))
4314 goto slashes;
4315 ihold(inode);
4316 error = security_path_unlink(&path, dentry);
4317 if (error)
4318 goto exit3;
4319 mnt_userns = mnt_user_ns(path.mnt);
4320 error = vfs_unlink(mnt_userns, path.dentry->d_inode, dentry,
4321 &delegated_inode);
4322exit3:
4323 dput(dentry);
4324 }
4325 inode_unlock(path.dentry->d_inode);
4326 if (inode)
4327 iput(inode); /* truncate the inode here */
4328 inode = NULL;
4329 if (delegated_inode) {
4330 error = break_deleg_wait(&delegated_inode);
4331 if (!error)
4332 goto retry_deleg;
4333 }
4334 mnt_drop_write(path.mnt);
4335exit2:
4336 path_put(&path);
4337 if (retry_estale(error, lookup_flags)) {
4338 lookup_flags |= LOOKUP_REVAL;
4339 inode = NULL;
4340 goto retry;
4341 }
4342exit1:
4343 putname(name);
4344 return error;
4345
4346slashes:
4347 if (d_is_negative(dentry))
4348 error = -ENOENT;
4349 else if (d_is_dir(dentry))
4350 error = -EISDIR;
4351 else
4352 error = -ENOTDIR;
4353 goto exit3;
4354}
4355
4356SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4357{
4358 if ((flag & ~AT_REMOVEDIR) != 0)
4359 return -EINVAL;
4360
4361 if (flag & AT_REMOVEDIR)
4362 return do_rmdir(dfd, getname(pathname));
4363 return do_unlinkat(dfd, getname(pathname));
4364}
4365
4366SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4367{
4368 return do_unlinkat(AT_FDCWD, getname(pathname));
4369}
4370
4371/**
4372 * vfs_symlink - create symlink
4373 * @mnt_userns: user namespace of the mount the inode was found from
4374 * @dir: inode of @dentry
4375 * @dentry: pointer to dentry of the base directory
4376 * @oldname: name of the file to link to
4377 *
4378 * Create a symlink.
4379 *
4380 * If the inode has been found through an idmapped mount the user namespace of
4381 * the vfsmount must be passed through @mnt_userns. This function will then take
4382 * care to map the inode according to @mnt_userns before checking permissions.
4383 * On non-idmapped mounts or if permission checking is to be performed on the
4384 * raw inode simply passs init_user_ns.
4385 */
4386int vfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
4387 struct dentry *dentry, const char *oldname)
4388{
4389 int error = may_create(mnt_userns, dir, dentry);
4390
4391 if (error)
4392 return error;
4393
4394 if (!dir->i_op->symlink)
4395 return -EPERM;
4396
4397 error = security_inode_symlink(dir, dentry, oldname);
4398 if (error)
4399 return error;
4400
4401 error = dir->i_op->symlink(mnt_userns, dir, dentry, oldname);
4402 if (!error)
4403 fsnotify_create(dir, dentry);
4404 return error;
4405}
4406EXPORT_SYMBOL(vfs_symlink);
4407
4408int do_symlinkat(struct filename *from, int newdfd, struct filename *to)
4409{
4410 int error;
4411 struct dentry *dentry;
4412 struct path path;
4413 unsigned int lookup_flags = 0;
4414
4415 if (IS_ERR(from)) {
4416 error = PTR_ERR(from);
4417 goto out_putnames;
4418 }
4419retry:
4420 dentry = filename_create(newdfd, to, &path, lookup_flags);
4421 error = PTR_ERR(dentry);
4422 if (IS_ERR(dentry))
4423 goto out_putnames;
4424
4425 error = security_path_symlink(&path, dentry, from->name);
4426 if (!error) {
4427 struct user_namespace *mnt_userns;
4428
4429 mnt_userns = mnt_user_ns(path.mnt);
4430 error = vfs_symlink(mnt_userns, path.dentry->d_inode, dentry,
4431 from->name);
4432 }
4433 done_path_create(&path, dentry);
4434 if (retry_estale(error, lookup_flags)) {
4435 lookup_flags |= LOOKUP_REVAL;
4436 goto retry;
4437 }
4438out_putnames:
4439 putname(to);
4440 putname(from);
4441 return error;
4442}
4443
4444SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4445 int, newdfd, const char __user *, newname)
4446{
4447 return do_symlinkat(getname(oldname), newdfd, getname(newname));
4448}
4449
4450SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4451{
4452 return do_symlinkat(getname(oldname), AT_FDCWD, getname(newname));
4453}
4454
4455/**
4456 * vfs_link - create a new link
4457 * @old_dentry: object to be linked
4458 * @mnt_userns: the user namespace of the mount
4459 * @dir: new parent
4460 * @new_dentry: where to create the new link
4461 * @delegated_inode: returns inode needing a delegation break
4462 *
4463 * The caller must hold dir->i_mutex
4464 *
4465 * If vfs_link discovers a delegation on the to-be-linked file in need
4466 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4467 * inode in delegated_inode. The caller should then break the delegation
4468 * and retry. Because breaking a delegation may take a long time, the
4469 * caller should drop the i_mutex before doing so.
4470 *
4471 * Alternatively, a caller may pass NULL for delegated_inode. This may
4472 * be appropriate for callers that expect the underlying filesystem not
4473 * to be NFS exported.
4474 *
4475 * If the inode has been found through an idmapped mount the user namespace of
4476 * the vfsmount must be passed through @mnt_userns. This function will then take
4477 * care to map the inode according to @mnt_userns before checking permissions.
4478 * On non-idmapped mounts or if permission checking is to be performed on the
4479 * raw inode simply passs init_user_ns.
4480 */
4481int vfs_link(struct dentry *old_dentry, struct user_namespace *mnt_userns,
4482 struct inode *dir, struct dentry *new_dentry,
4483 struct inode **delegated_inode)
4484{
4485 struct inode *inode = old_dentry->d_inode;
4486 unsigned max_links = dir->i_sb->s_max_links;
4487 int error;
4488
4489 if (!inode)
4490 return -ENOENT;
4491
4492 error = may_create(mnt_userns, dir, new_dentry);
4493 if (error)
4494 return error;
4495
4496 if (dir->i_sb != inode->i_sb)
4497 return -EXDEV;
4498
4499 /*
4500 * A link to an append-only or immutable file cannot be created.
4501 */
4502 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4503 return -EPERM;
4504 /*
4505 * Updating the link count will likely cause i_uid and i_gid to
4506 * be writen back improperly if their true value is unknown to
4507 * the vfs.
4508 */
4509 if (HAS_UNMAPPED_ID(mnt_userns, inode))
4510 return -EPERM;
4511 if (!dir->i_op->link)
4512 return -EPERM;
4513 if (S_ISDIR(inode->i_mode))
4514 return -EPERM;
4515
4516 error = security_inode_link(old_dentry, dir, new_dentry);
4517 if (error)
4518 return error;
4519
4520 inode_lock(inode);
4521 /* Make sure we don't allow creating hardlink to an unlinked file */
4522 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4523 error = -ENOENT;
4524 else if (max_links && inode->i_nlink >= max_links)
4525 error = -EMLINK;
4526 else {
4527 error = try_break_deleg(inode, delegated_inode);
4528 if (!error)
4529 error = dir->i_op->link(old_dentry, dir, new_dentry);
4530 }
4531
4532 if (!error && (inode->i_state & I_LINKABLE)) {
4533 spin_lock(&inode->i_lock);
4534 inode->i_state &= ~I_LINKABLE;
4535 spin_unlock(&inode->i_lock);
4536 }
4537 inode_unlock(inode);
4538 if (!error)
4539 fsnotify_link(dir, inode, new_dentry);
4540 return error;
4541}
4542EXPORT_SYMBOL(vfs_link);
4543
4544/*
4545 * Hardlinks are often used in delicate situations. We avoid
4546 * security-related surprises by not following symlinks on the
4547 * newname. --KAB
4548 *
4549 * We don't follow them on the oldname either to be compatible
4550 * with linux 2.0, and to avoid hard-linking to directories
4551 * and other special files. --ADM
4552 */
4553int do_linkat(int olddfd, struct filename *old, int newdfd,
4554 struct filename *new, int flags)
4555{
4556 struct user_namespace *mnt_userns;
4557 struct dentry *new_dentry;
4558 struct path old_path, new_path;
4559 struct inode *delegated_inode = NULL;
4560 int how = 0;
4561 int error;
4562
4563 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) {
4564 error = -EINVAL;
4565 goto out_putnames;
4566 }
4567 /*
4568 * To use null names we require CAP_DAC_READ_SEARCH
4569 * This ensures that not everyone will be able to create
4570 * handlink using the passed filedescriptor.
4571 */
4572 if (flags & AT_EMPTY_PATH && !capable(CAP_DAC_READ_SEARCH)) {
4573 error = -ENOENT;
4574 goto out_putnames;
4575 }
4576
4577 if (flags & AT_SYMLINK_FOLLOW)
4578 how |= LOOKUP_FOLLOW;
4579retry:
4580 error = filename_lookup(olddfd, old, how, &old_path, NULL);
4581 if (error)
4582 goto out_putnames;
4583
4584 new_dentry = filename_create(newdfd, new, &new_path,
4585 (how & LOOKUP_REVAL));
4586 error = PTR_ERR(new_dentry);
4587 if (IS_ERR(new_dentry))
4588 goto out_putpath;
4589
4590 error = -EXDEV;
4591 if (old_path.mnt != new_path.mnt)
4592 goto out_dput;
4593 mnt_userns = mnt_user_ns(new_path.mnt);
4594 error = may_linkat(mnt_userns, &old_path);
4595 if (unlikely(error))
4596 goto out_dput;
4597 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4598 if (error)
4599 goto out_dput;
4600 error = vfs_link(old_path.dentry, mnt_userns, new_path.dentry->d_inode,
4601 new_dentry, &delegated_inode);
4602out_dput:
4603 done_path_create(&new_path, new_dentry);
4604 if (delegated_inode) {
4605 error = break_deleg_wait(&delegated_inode);
4606 if (!error) {
4607 path_put(&old_path);
4608 goto retry;
4609 }
4610 }
4611 if (retry_estale(error, how)) {
4612 path_put(&old_path);
4613 how |= LOOKUP_REVAL;
4614 goto retry;
4615 }
4616out_putpath:
4617 path_put(&old_path);
4618out_putnames:
4619 putname(old);
4620 putname(new);
4621
4622 return error;
4623}
4624
4625SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4626 int, newdfd, const char __user *, newname, int, flags)
4627{
4628 return do_linkat(olddfd, getname_uflags(oldname, flags),
4629 newdfd, getname(newname), flags);
4630}
4631
4632SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4633{
4634 return do_linkat(AT_FDCWD, getname(oldname), AT_FDCWD, getname(newname), 0);
4635}
4636
4637/**
4638 * vfs_rename - rename a filesystem object
4639 * @rd: pointer to &struct renamedata info
4640 *
4641 * The caller must hold multiple mutexes--see lock_rename()).
4642 *
4643 * If vfs_rename discovers a delegation in need of breaking at either
4644 * the source or destination, it will return -EWOULDBLOCK and return a
4645 * reference to the inode in delegated_inode. The caller should then
4646 * break the delegation and retry. Because breaking a delegation may
4647 * take a long time, the caller should drop all locks before doing
4648 * so.
4649 *
4650 * Alternatively, a caller may pass NULL for delegated_inode. This may
4651 * be appropriate for callers that expect the underlying filesystem not
4652 * to be NFS exported.
4653 *
4654 * The worst of all namespace operations - renaming directory. "Perverted"
4655 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4656 * Problems:
4657 *
4658 * a) we can get into loop creation.
4659 * b) race potential - two innocent renames can create a loop together.
4660 * That's where 4.4 screws up. Current fix: serialization on
4661 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4662 * story.
4663 * c) we have to lock _four_ objects - parents and victim (if it exists),
4664 * and source (if it is not a directory).
4665 * And that - after we got ->i_mutex on parents (until then we don't know
4666 * whether the target exists). Solution: try to be smart with locking
4667 * order for inodes. We rely on the fact that tree topology may change
4668 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4669 * move will be locked. Thus we can rank directories by the tree
4670 * (ancestors first) and rank all non-directories after them.
4671 * That works since everybody except rename does "lock parent, lookup,
4672 * lock child" and rename is under ->s_vfs_rename_mutex.
4673 * HOWEVER, it relies on the assumption that any object with ->lookup()
4674 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4675 * we'd better make sure that there's no link(2) for them.
4676 * d) conversion from fhandle to dentry may come in the wrong moment - when
4677 * we are removing the target. Solution: we will have to grab ->i_mutex
4678 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4679 * ->i_mutex on parents, which works but leads to some truly excessive
4680 * locking].
4681 */
4682int vfs_rename(struct renamedata *rd)
4683{
4684 int error;
4685 struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir;
4686 struct dentry *old_dentry = rd->old_dentry;
4687 struct dentry *new_dentry = rd->new_dentry;
4688 struct inode **delegated_inode = rd->delegated_inode;
4689 unsigned int flags = rd->flags;
4690 bool is_dir = d_is_dir(old_dentry);
4691 struct inode *source = old_dentry->d_inode;
4692 struct inode *target = new_dentry->d_inode;
4693 bool new_is_dir = false;
4694 unsigned max_links = new_dir->i_sb->s_max_links;
4695 struct name_snapshot old_name;
4696
4697 if (source == target)
4698 return 0;
4699
4700 error = may_delete(rd->old_mnt_userns, old_dir, old_dentry, is_dir);
4701 if (error)
4702 return error;
4703
4704 if (!target) {
4705 error = may_create(rd->new_mnt_userns, new_dir, new_dentry);
4706 } else {
4707 new_is_dir = d_is_dir(new_dentry);
4708
4709 if (!(flags & RENAME_EXCHANGE))
4710 error = may_delete(rd->new_mnt_userns, new_dir,
4711 new_dentry, is_dir);
4712 else
4713 error = may_delete(rd->new_mnt_userns, new_dir,
4714 new_dentry, new_is_dir);
4715 }
4716 if (error)
4717 return error;
4718
4719 if (!old_dir->i_op->rename)
4720 return -EPERM;
4721
4722 /*
4723 * If we are going to change the parent - check write permissions,
4724 * we'll need to flip '..'.
4725 */
4726 if (new_dir != old_dir) {
4727 if (is_dir) {
4728 error = inode_permission(rd->old_mnt_userns, source,
4729 MAY_WRITE);
4730 if (error)
4731 return error;
4732 }
4733 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4734 error = inode_permission(rd->new_mnt_userns, target,
4735 MAY_WRITE);
4736 if (error)
4737 return error;
4738 }
4739 }
4740
4741 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4742 flags);
4743 if (error)
4744 return error;
4745
4746 take_dentry_name_snapshot(&old_name, old_dentry);
4747 dget(new_dentry);
4748 if (!is_dir || (flags & RENAME_EXCHANGE))
4749 lock_two_nondirectories(source, target);
4750 else if (target)
4751 inode_lock(target);
4752
4753 error = -EPERM;
4754 if (IS_SWAPFILE(source) || (target && IS_SWAPFILE(target)))
4755 goto out;
4756
4757 error = -EBUSY;
4758 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4759 goto out;
4760
4761 if (max_links && new_dir != old_dir) {
4762 error = -EMLINK;
4763 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4764 goto out;
4765 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4766 old_dir->i_nlink >= max_links)
4767 goto out;
4768 }
4769 if (!is_dir) {
4770 error = try_break_deleg(source, delegated_inode);
4771 if (error)
4772 goto out;
4773 }
4774 if (target && !new_is_dir) {
4775 error = try_break_deleg(target, delegated_inode);
4776 if (error)
4777 goto out;
4778 }
4779 error = old_dir->i_op->rename(rd->new_mnt_userns, old_dir, old_dentry,
4780 new_dir, new_dentry, flags);
4781 if (error)
4782 goto out;
4783
4784 if (!(flags & RENAME_EXCHANGE) && target) {
4785 if (is_dir) {
4786 shrink_dcache_parent(new_dentry);
4787 target->i_flags |= S_DEAD;
4788 }
4789 dont_mount(new_dentry);
4790 detach_mounts(new_dentry);
4791 }
4792 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4793 if (!(flags & RENAME_EXCHANGE))
4794 d_move(old_dentry, new_dentry);
4795 else
4796 d_exchange(old_dentry, new_dentry);
4797 }
4798out:
4799 if (!is_dir || (flags & RENAME_EXCHANGE))
4800 unlock_two_nondirectories(source, target);
4801 else if (target)
4802 inode_unlock(target);
4803 dput(new_dentry);
4804 if (!error) {
4805 fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
4806 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4807 if (flags & RENAME_EXCHANGE) {
4808 fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
4809 new_is_dir, NULL, new_dentry);
4810 }
4811 }
4812 release_dentry_name_snapshot(&old_name);
4813
4814 return error;
4815}
4816EXPORT_SYMBOL(vfs_rename);
4817
4818int do_renameat2(int olddfd, struct filename *from, int newdfd,
4819 struct filename *to, unsigned int flags)
4820{
4821 struct renamedata rd;
4822 struct dentry *old_dentry, *new_dentry;
4823 struct dentry *trap;
4824 struct path old_path, new_path;
4825 struct qstr old_last, new_last;
4826 int old_type, new_type;
4827 struct inode *delegated_inode = NULL;
4828 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4829 bool should_retry = false;
4830 int error = -EINVAL;
4831
4832 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4833 goto put_names;
4834
4835 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4836 (flags & RENAME_EXCHANGE))
4837 goto put_names;
4838
4839 if (flags & RENAME_EXCHANGE)
4840 target_flags = 0;
4841
4842retry:
4843 error = filename_parentat(olddfd, from, lookup_flags, &old_path,
4844 &old_last, &old_type);
4845 if (error)
4846 goto put_names;
4847
4848 error = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last,
4849 &new_type);
4850 if (error)
4851 goto exit1;
4852
4853 error = -EXDEV;
4854 if (old_path.mnt != new_path.mnt)
4855 goto exit2;
4856
4857 error = -EBUSY;
4858 if (old_type != LAST_NORM)
4859 goto exit2;
4860
4861 if (flags & RENAME_NOREPLACE)
4862 error = -EEXIST;
4863 if (new_type != LAST_NORM)
4864 goto exit2;
4865
4866 error = mnt_want_write(old_path.mnt);
4867 if (error)
4868 goto exit2;
4869
4870retry_deleg:
4871 trap = lock_rename(new_path.dentry, old_path.dentry);
4872
4873 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4874 error = PTR_ERR(old_dentry);
4875 if (IS_ERR(old_dentry))
4876 goto exit3;
4877 /* source must exist */
4878 error = -ENOENT;
4879 if (d_is_negative(old_dentry))
4880 goto exit4;
4881 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4882 error = PTR_ERR(new_dentry);
4883 if (IS_ERR(new_dentry))
4884 goto exit4;
4885 error = -EEXIST;
4886 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4887 goto exit5;
4888 if (flags & RENAME_EXCHANGE) {
4889 error = -ENOENT;
4890 if (d_is_negative(new_dentry))
4891 goto exit5;
4892
4893 if (!d_is_dir(new_dentry)) {
4894 error = -ENOTDIR;
4895 if (new_last.name[new_last.len])
4896 goto exit5;
4897 }
4898 }
4899 /* unless the source is a directory trailing slashes give -ENOTDIR */
4900 if (!d_is_dir(old_dentry)) {
4901 error = -ENOTDIR;
4902 if (old_last.name[old_last.len])
4903 goto exit5;
4904 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4905 goto exit5;
4906 }
4907 /* source should not be ancestor of target */
4908 error = -EINVAL;
4909 if (old_dentry == trap)
4910 goto exit5;
4911 /* target should not be an ancestor of source */
4912 if (!(flags & RENAME_EXCHANGE))
4913 error = -ENOTEMPTY;
4914 if (new_dentry == trap)
4915 goto exit5;
4916
4917 error = security_path_rename(&old_path, old_dentry,
4918 &new_path, new_dentry, flags);
4919 if (error)
4920 goto exit5;
4921
4922 rd.old_dir = old_path.dentry->d_inode;
4923 rd.old_dentry = old_dentry;
4924 rd.old_mnt_userns = mnt_user_ns(old_path.mnt);
4925 rd.new_dir = new_path.dentry->d_inode;
4926 rd.new_dentry = new_dentry;
4927 rd.new_mnt_userns = mnt_user_ns(new_path.mnt);
4928 rd.delegated_inode = &delegated_inode;
4929 rd.flags = flags;
4930 error = vfs_rename(&rd);
4931exit5:
4932 dput(new_dentry);
4933exit4:
4934 dput(old_dentry);
4935exit3:
4936 unlock_rename(new_path.dentry, old_path.dentry);
4937 if (delegated_inode) {
4938 error = break_deleg_wait(&delegated_inode);
4939 if (!error)
4940 goto retry_deleg;
4941 }
4942 mnt_drop_write(old_path.mnt);
4943exit2:
4944 if (retry_estale(error, lookup_flags))
4945 should_retry = true;
4946 path_put(&new_path);
4947exit1:
4948 path_put(&old_path);
4949 if (should_retry) {
4950 should_retry = false;
4951 lookup_flags |= LOOKUP_REVAL;
4952 goto retry;
4953 }
4954put_names:
4955 putname(from);
4956 putname(to);
4957 return error;
4958}
4959
4960SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4961 int, newdfd, const char __user *, newname, unsigned int, flags)
4962{
4963 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
4964 flags);
4965}
4966
4967SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4968 int, newdfd, const char __user *, newname)
4969{
4970 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
4971 0);
4972}
4973
4974SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4975{
4976 return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD,
4977 getname(newname), 0);
4978}
4979
4980int readlink_copy(char __user *buffer, int buflen, const char *link)
4981{
4982 int len = PTR_ERR(link);
4983 if (IS_ERR(link))
4984 goto out;
4985
4986 len = strlen(link);
4987 if (len > (unsigned) buflen)
4988 len = buflen;
4989 if (copy_to_user(buffer, link, len))
4990 len = -EFAULT;
4991out:
4992 return len;
4993}
4994
4995/**
4996 * vfs_readlink - copy symlink body into userspace buffer
4997 * @dentry: dentry on which to get symbolic link
4998 * @buffer: user memory pointer
4999 * @buflen: size of buffer
5000 *
5001 * Does not touch atime. That's up to the caller if necessary
5002 *
5003 * Does not call security hook.
5004 */
5005int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
5006{
5007 struct inode *inode = d_inode(dentry);
5008 DEFINE_DELAYED_CALL(done);
5009 const char *link;
5010 int res;
5011
5012 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
5013 if (unlikely(inode->i_op->readlink))
5014 return inode->i_op->readlink(dentry, buffer, buflen);
5015
5016 if (!d_is_symlink(dentry))
5017 return -EINVAL;
5018
5019 spin_lock(&inode->i_lock);
5020 inode->i_opflags |= IOP_DEFAULT_READLINK;
5021 spin_unlock(&inode->i_lock);
5022 }
5023
5024 link = READ_ONCE(inode->i_link);
5025 if (!link) {
5026 link = inode->i_op->get_link(dentry, inode, &done);
5027 if (IS_ERR(link))
5028 return PTR_ERR(link);
5029 }
5030 res = readlink_copy(buffer, buflen, link);
5031 do_delayed_call(&done);
5032 return res;
5033}
5034EXPORT_SYMBOL(vfs_readlink);
5035
5036/**
5037 * vfs_get_link - get symlink body
5038 * @dentry: dentry on which to get symbolic link
5039 * @done: caller needs to free returned data with this
5040 *
5041 * Calls security hook and i_op->get_link() on the supplied inode.
5042 *
5043 * It does not touch atime. That's up to the caller if necessary.
5044 *
5045 * Does not work on "special" symlinks like /proc/$$/fd/N
5046 */
5047const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
5048{
5049 const char *res = ERR_PTR(-EINVAL);
5050 struct inode *inode = d_inode(dentry);
5051
5052 if (d_is_symlink(dentry)) {
5053 res = ERR_PTR(security_inode_readlink(dentry));
5054 if (!res)
5055 res = inode->i_op->get_link(dentry, inode, done);
5056 }
5057 return res;
5058}
5059EXPORT_SYMBOL(vfs_get_link);
5060
5061/* get the link contents into pagecache */
5062const char *page_get_link(struct dentry *dentry, struct inode *inode,
5063 struct delayed_call *callback)
5064{
5065 char *kaddr;
5066 struct page *page;
5067 struct address_space *mapping = inode->i_mapping;
5068
5069 if (!dentry) {
5070 page = find_get_page(mapping, 0);
5071 if (!page)
5072 return ERR_PTR(-ECHILD);
5073 if (!PageUptodate(page)) {
5074 put_page(page);
5075 return ERR_PTR(-ECHILD);
5076 }
5077 } else {
5078 page = read_mapping_page(mapping, 0, NULL);
5079 if (IS_ERR(page))
5080 return (char*)page;
5081 }
5082 set_delayed_call(callback, page_put_link, page);
5083 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
5084 kaddr = page_address(page);
5085 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
5086 return kaddr;
5087}
5088
5089EXPORT_SYMBOL(page_get_link);
5090
5091void page_put_link(void *arg)
5092{
5093 put_page(arg);
5094}
5095EXPORT_SYMBOL(page_put_link);
5096
5097int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
5098{
5099 DEFINE_DELAYED_CALL(done);
5100 int res = readlink_copy(buffer, buflen,
5101 page_get_link(dentry, d_inode(dentry),
5102 &done));
5103 do_delayed_call(&done);
5104 return res;
5105}
5106EXPORT_SYMBOL(page_readlink);
5107
5108int page_symlink(struct inode *inode, const char *symname, int len)
5109{
5110 struct address_space *mapping = inode->i_mapping;
5111 const struct address_space_operations *aops = mapping->a_ops;
5112 bool nofs = !mapping_gfp_constraint(mapping, __GFP_FS);
5113 struct page *page;
5114 void *fsdata = NULL;
5115 int err;
5116 unsigned int flags;
5117
5118retry:
5119 if (nofs)
5120 flags = memalloc_nofs_save();
5121 err = aops->write_begin(NULL, mapping, 0, len-1, &page, &fsdata);
5122 if (nofs)
5123 memalloc_nofs_restore(flags);
5124 if (err)
5125 goto fail;
5126
5127 memcpy(page_address(page), symname, len-1);
5128
5129 err = aops->write_end(NULL, mapping, 0, len-1, len-1,
5130 page, fsdata);
5131 if (err < 0)
5132 goto fail;
5133 if (err < len-1)
5134 goto retry;
5135
5136 mark_inode_dirty(inode);
5137 return 0;
5138fail:
5139 return err;
5140}
5141EXPORT_SYMBOL(page_symlink);
5142
5143const struct inode_operations page_symlink_inode_operations = {
5144 .get_link = page_get_link,
5145};
5146EXPORT_SYMBOL(page_symlink_inode_operations);
1/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
18#include <linux/export.h>
19#include <linux/kernel.h>
20#include <linux/slab.h>
21#include <linux/fs.h>
22#include <linux/namei.h>
23#include <linux/pagemap.h>
24#include <linux/fsnotify.h>
25#include <linux/personality.h>
26#include <linux/security.h>
27#include <linux/ima.h>
28#include <linux/syscalls.h>
29#include <linux/mount.h>
30#include <linux/audit.h>
31#include <linux/capability.h>
32#include <linux/file.h>
33#include <linux/fcntl.h>
34#include <linux/device_cgroup.h>
35#include <linux/fs_struct.h>
36#include <linux/posix_acl.h>
37#include <asm/uaccess.h>
38
39#include "internal.h"
40#include "mount.h"
41
42/* [Feb-1997 T. Schoebel-Theuer]
43 * Fundamental changes in the pathname lookup mechanisms (namei)
44 * were necessary because of omirr. The reason is that omirr needs
45 * to know the _real_ pathname, not the user-supplied one, in case
46 * of symlinks (and also when transname replacements occur).
47 *
48 * The new code replaces the old recursive symlink resolution with
49 * an iterative one (in case of non-nested symlink chains). It does
50 * this with calls to <fs>_follow_link().
51 * As a side effect, dir_namei(), _namei() and follow_link() are now
52 * replaced with a single function lookup_dentry() that can handle all
53 * the special cases of the former code.
54 *
55 * With the new dcache, the pathname is stored at each inode, at least as
56 * long as the refcount of the inode is positive. As a side effect, the
57 * size of the dcache depends on the inode cache and thus is dynamic.
58 *
59 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
60 * resolution to correspond with current state of the code.
61 *
62 * Note that the symlink resolution is not *completely* iterative.
63 * There is still a significant amount of tail- and mid- recursion in
64 * the algorithm. Also, note that <fs>_readlink() is not used in
65 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
66 * may return different results than <fs>_follow_link(). Many virtual
67 * filesystems (including /proc) exhibit this behavior.
68 */
69
70/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
71 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
72 * and the name already exists in form of a symlink, try to create the new
73 * name indicated by the symlink. The old code always complained that the
74 * name already exists, due to not following the symlink even if its target
75 * is nonexistent. The new semantics affects also mknod() and link() when
76 * the name is a symlink pointing to a non-existent name.
77 *
78 * I don't know which semantics is the right one, since I have no access
79 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
80 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
81 * "old" one. Personally, I think the new semantics is much more logical.
82 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
83 * file does succeed in both HP-UX and SunOs, but not in Solaris
84 * and in the old Linux semantics.
85 */
86
87/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
88 * semantics. See the comments in "open_namei" and "do_link" below.
89 *
90 * [10-Sep-98 Alan Modra] Another symlink change.
91 */
92
93/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
94 * inside the path - always follow.
95 * in the last component in creation/removal/renaming - never follow.
96 * if LOOKUP_FOLLOW passed - follow.
97 * if the pathname has trailing slashes - follow.
98 * otherwise - don't follow.
99 * (applied in that order).
100 *
101 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
102 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
103 * During the 2.4 we need to fix the userland stuff depending on it -
104 * hopefully we will be able to get rid of that wart in 2.5. So far only
105 * XEmacs seems to be relying on it...
106 */
107/*
108 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
109 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
110 * any extra contention...
111 */
112
113/* In order to reduce some races, while at the same time doing additional
114 * checking and hopefully speeding things up, we copy filenames to the
115 * kernel data space before using them..
116 *
117 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
118 * PATH_MAX includes the nul terminator --RR.
119 */
120void final_putname(struct filename *name)
121{
122 if (name->separate) {
123 __putname(name->name);
124 kfree(name);
125 } else {
126 __putname(name);
127 }
128}
129
130#define EMBEDDED_NAME_MAX (PATH_MAX - sizeof(struct filename))
131
132static struct filename *
133getname_flags(const char __user *filename, int flags, int *empty)
134{
135 struct filename *result, *err;
136 int len;
137 long max;
138 char *kname;
139
140 result = audit_reusename(filename);
141 if (result)
142 return result;
143
144 result = __getname();
145 if (unlikely(!result))
146 return ERR_PTR(-ENOMEM);
147
148 /*
149 * First, try to embed the struct filename inside the names_cache
150 * allocation
151 */
152 kname = (char *)result + sizeof(*result);
153 result->name = kname;
154 result->separate = false;
155 max = EMBEDDED_NAME_MAX;
156
157recopy:
158 len = strncpy_from_user(kname, filename, max);
159 if (unlikely(len < 0)) {
160 err = ERR_PTR(len);
161 goto error;
162 }
163
164 /*
165 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
166 * separate struct filename so we can dedicate the entire
167 * names_cache allocation for the pathname, and re-do the copy from
168 * userland.
169 */
170 if (len == EMBEDDED_NAME_MAX && max == EMBEDDED_NAME_MAX) {
171 kname = (char *)result;
172
173 result = kzalloc(sizeof(*result), GFP_KERNEL);
174 if (!result) {
175 err = ERR_PTR(-ENOMEM);
176 result = (struct filename *)kname;
177 goto error;
178 }
179 result->name = kname;
180 result->separate = true;
181 max = PATH_MAX;
182 goto recopy;
183 }
184
185 /* The empty path is special. */
186 if (unlikely(!len)) {
187 if (empty)
188 *empty = 1;
189 err = ERR_PTR(-ENOENT);
190 if (!(flags & LOOKUP_EMPTY))
191 goto error;
192 }
193
194 err = ERR_PTR(-ENAMETOOLONG);
195 if (unlikely(len >= PATH_MAX))
196 goto error;
197
198 result->uptr = filename;
199 result->aname = NULL;
200 audit_getname(result);
201 return result;
202
203error:
204 final_putname(result);
205 return err;
206}
207
208struct filename *
209getname(const char __user * filename)
210{
211 return getname_flags(filename, 0, NULL);
212}
213
214/*
215 * The "getname_kernel()" interface doesn't do pathnames longer
216 * than EMBEDDED_NAME_MAX. Deal with it - you're a kernel user.
217 */
218struct filename *
219getname_kernel(const char * filename)
220{
221 struct filename *result;
222 char *kname;
223 int len;
224
225 len = strlen(filename);
226 if (len >= EMBEDDED_NAME_MAX)
227 return ERR_PTR(-ENAMETOOLONG);
228
229 result = __getname();
230 if (unlikely(!result))
231 return ERR_PTR(-ENOMEM);
232
233 kname = (char *)result + sizeof(*result);
234 result->name = kname;
235 result->uptr = NULL;
236 result->aname = NULL;
237 result->separate = false;
238
239 strlcpy(kname, filename, EMBEDDED_NAME_MAX);
240 return result;
241}
242
243#ifdef CONFIG_AUDITSYSCALL
244void putname(struct filename *name)
245{
246 if (unlikely(!audit_dummy_context()))
247 return audit_putname(name);
248 final_putname(name);
249}
250#endif
251
252static int check_acl(struct inode *inode, int mask)
253{
254#ifdef CONFIG_FS_POSIX_ACL
255 struct posix_acl *acl;
256
257 if (mask & MAY_NOT_BLOCK) {
258 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
259 if (!acl)
260 return -EAGAIN;
261 /* no ->get_acl() calls in RCU mode... */
262 if (acl == ACL_NOT_CACHED)
263 return -ECHILD;
264 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
265 }
266
267 acl = get_acl(inode, ACL_TYPE_ACCESS);
268 if (IS_ERR(acl))
269 return PTR_ERR(acl);
270 if (acl) {
271 int error = posix_acl_permission(inode, acl, mask);
272 posix_acl_release(acl);
273 return error;
274 }
275#endif
276
277 return -EAGAIN;
278}
279
280/*
281 * This does the basic permission checking
282 */
283static int acl_permission_check(struct inode *inode, int mask)
284{
285 unsigned int mode = inode->i_mode;
286
287 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
288 mode >>= 6;
289 else {
290 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
291 int error = check_acl(inode, mask);
292 if (error != -EAGAIN)
293 return error;
294 }
295
296 if (in_group_p(inode->i_gid))
297 mode >>= 3;
298 }
299
300 /*
301 * If the DACs are ok we don't need any capability check.
302 */
303 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
304 return 0;
305 return -EACCES;
306}
307
308/**
309 * generic_permission - check for access rights on a Posix-like filesystem
310 * @inode: inode to check access rights for
311 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
312 *
313 * Used to check for read/write/execute permissions on a file.
314 * We use "fsuid" for this, letting us set arbitrary permissions
315 * for filesystem access without changing the "normal" uids which
316 * are used for other things.
317 *
318 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
319 * request cannot be satisfied (eg. requires blocking or too much complexity).
320 * It would then be called again in ref-walk mode.
321 */
322int generic_permission(struct inode *inode, int mask)
323{
324 int ret;
325
326 /*
327 * Do the basic permission checks.
328 */
329 ret = acl_permission_check(inode, mask);
330 if (ret != -EACCES)
331 return ret;
332
333 if (S_ISDIR(inode->i_mode)) {
334 /* DACs are overridable for directories */
335 if (inode_capable(inode, CAP_DAC_OVERRIDE))
336 return 0;
337 if (!(mask & MAY_WRITE))
338 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
339 return 0;
340 return -EACCES;
341 }
342 /*
343 * Read/write DACs are always overridable.
344 * Executable DACs are overridable when there is
345 * at least one exec bit set.
346 */
347 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
348 if (inode_capable(inode, CAP_DAC_OVERRIDE))
349 return 0;
350
351 /*
352 * Searching includes executable on directories, else just read.
353 */
354 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
355 if (mask == MAY_READ)
356 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
357 return 0;
358
359 return -EACCES;
360}
361EXPORT_SYMBOL(generic_permission);
362
363/*
364 * We _really_ want to just do "generic_permission()" without
365 * even looking at the inode->i_op values. So we keep a cache
366 * flag in inode->i_opflags, that says "this has not special
367 * permission function, use the fast case".
368 */
369static inline int do_inode_permission(struct inode *inode, int mask)
370{
371 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
372 if (likely(inode->i_op->permission))
373 return inode->i_op->permission(inode, mask);
374
375 /* This gets set once for the inode lifetime */
376 spin_lock(&inode->i_lock);
377 inode->i_opflags |= IOP_FASTPERM;
378 spin_unlock(&inode->i_lock);
379 }
380 return generic_permission(inode, mask);
381}
382
383/**
384 * __inode_permission - Check for access rights to a given inode
385 * @inode: Inode to check permission on
386 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
387 *
388 * Check for read/write/execute permissions on an inode.
389 *
390 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
391 *
392 * This does not check for a read-only file system. You probably want
393 * inode_permission().
394 */
395int __inode_permission(struct inode *inode, int mask)
396{
397 int retval;
398
399 if (unlikely(mask & MAY_WRITE)) {
400 /*
401 * Nobody gets write access to an immutable file.
402 */
403 if (IS_IMMUTABLE(inode))
404 return -EACCES;
405 }
406
407 retval = do_inode_permission(inode, mask);
408 if (retval)
409 return retval;
410
411 retval = devcgroup_inode_permission(inode, mask);
412 if (retval)
413 return retval;
414
415 return security_inode_permission(inode, mask);
416}
417
418/**
419 * sb_permission - Check superblock-level permissions
420 * @sb: Superblock of inode to check permission on
421 * @inode: Inode to check permission on
422 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
423 *
424 * Separate out file-system wide checks from inode-specific permission checks.
425 */
426static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
427{
428 if (unlikely(mask & MAY_WRITE)) {
429 umode_t mode = inode->i_mode;
430
431 /* Nobody gets write access to a read-only fs. */
432 if ((sb->s_flags & MS_RDONLY) &&
433 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
434 return -EROFS;
435 }
436 return 0;
437}
438
439/**
440 * inode_permission - Check for access rights to a given inode
441 * @inode: Inode to check permission on
442 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
443 *
444 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
445 * this, letting us set arbitrary permissions for filesystem access without
446 * changing the "normal" UIDs which are used for other things.
447 *
448 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
449 */
450int inode_permission(struct inode *inode, int mask)
451{
452 int retval;
453
454 retval = sb_permission(inode->i_sb, inode, mask);
455 if (retval)
456 return retval;
457 return __inode_permission(inode, mask);
458}
459EXPORT_SYMBOL(inode_permission);
460
461/**
462 * path_get - get a reference to a path
463 * @path: path to get the reference to
464 *
465 * Given a path increment the reference count to the dentry and the vfsmount.
466 */
467void path_get(const struct path *path)
468{
469 mntget(path->mnt);
470 dget(path->dentry);
471}
472EXPORT_SYMBOL(path_get);
473
474/**
475 * path_put - put a reference to a path
476 * @path: path to put the reference to
477 *
478 * Given a path decrement the reference count to the dentry and the vfsmount.
479 */
480void path_put(const struct path *path)
481{
482 dput(path->dentry);
483 mntput(path->mnt);
484}
485EXPORT_SYMBOL(path_put);
486
487/*
488 * Path walking has 2 modes, rcu-walk and ref-walk (see
489 * Documentation/filesystems/path-lookup.txt). In situations when we can't
490 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
491 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
492 * mode. Refcounts are grabbed at the last known good point before rcu-walk
493 * got stuck, so ref-walk may continue from there. If this is not successful
494 * (eg. a seqcount has changed), then failure is returned and it's up to caller
495 * to restart the path walk from the beginning in ref-walk mode.
496 */
497
498/**
499 * unlazy_walk - try to switch to ref-walk mode.
500 * @nd: nameidata pathwalk data
501 * @dentry: child of nd->path.dentry or NULL
502 * Returns: 0 on success, -ECHILD on failure
503 *
504 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
505 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
506 * @nd or NULL. Must be called from rcu-walk context.
507 */
508static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
509{
510 struct fs_struct *fs = current->fs;
511 struct dentry *parent = nd->path.dentry;
512
513 BUG_ON(!(nd->flags & LOOKUP_RCU));
514
515 /*
516 * After legitimizing the bastards, terminate_walk()
517 * will do the right thing for non-RCU mode, and all our
518 * subsequent exit cases should rcu_read_unlock()
519 * before returning. Do vfsmount first; if dentry
520 * can't be legitimized, just set nd->path.dentry to NULL
521 * and rely on dput(NULL) being a no-op.
522 */
523 if (!legitimize_mnt(nd->path.mnt, nd->m_seq))
524 return -ECHILD;
525 nd->flags &= ~LOOKUP_RCU;
526
527 if (!lockref_get_not_dead(&parent->d_lockref)) {
528 nd->path.dentry = NULL;
529 goto out;
530 }
531
532 /*
533 * For a negative lookup, the lookup sequence point is the parents
534 * sequence point, and it only needs to revalidate the parent dentry.
535 *
536 * For a positive lookup, we need to move both the parent and the
537 * dentry from the RCU domain to be properly refcounted. And the
538 * sequence number in the dentry validates *both* dentry counters,
539 * since we checked the sequence number of the parent after we got
540 * the child sequence number. So we know the parent must still
541 * be valid if the child sequence number is still valid.
542 */
543 if (!dentry) {
544 if (read_seqcount_retry(&parent->d_seq, nd->seq))
545 goto out;
546 BUG_ON(nd->inode != parent->d_inode);
547 } else {
548 if (!lockref_get_not_dead(&dentry->d_lockref))
549 goto out;
550 if (read_seqcount_retry(&dentry->d_seq, nd->seq))
551 goto drop_dentry;
552 }
553
554 /*
555 * Sequence counts matched. Now make sure that the root is
556 * still valid and get it if required.
557 */
558 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
559 spin_lock(&fs->lock);
560 if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry)
561 goto unlock_and_drop_dentry;
562 path_get(&nd->root);
563 spin_unlock(&fs->lock);
564 }
565
566 rcu_read_unlock();
567 return 0;
568
569unlock_and_drop_dentry:
570 spin_unlock(&fs->lock);
571drop_dentry:
572 rcu_read_unlock();
573 dput(dentry);
574 goto drop_root_mnt;
575out:
576 rcu_read_unlock();
577drop_root_mnt:
578 if (!(nd->flags & LOOKUP_ROOT))
579 nd->root.mnt = NULL;
580 return -ECHILD;
581}
582
583static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
584{
585 return dentry->d_op->d_revalidate(dentry, flags);
586}
587
588/**
589 * complete_walk - successful completion of path walk
590 * @nd: pointer nameidata
591 *
592 * If we had been in RCU mode, drop out of it and legitimize nd->path.
593 * Revalidate the final result, unless we'd already done that during
594 * the path walk or the filesystem doesn't ask for it. Return 0 on
595 * success, -error on failure. In case of failure caller does not
596 * need to drop nd->path.
597 */
598static int complete_walk(struct nameidata *nd)
599{
600 struct dentry *dentry = nd->path.dentry;
601 int status;
602
603 if (nd->flags & LOOKUP_RCU) {
604 nd->flags &= ~LOOKUP_RCU;
605 if (!(nd->flags & LOOKUP_ROOT))
606 nd->root.mnt = NULL;
607
608 if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) {
609 rcu_read_unlock();
610 return -ECHILD;
611 }
612 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) {
613 rcu_read_unlock();
614 mntput(nd->path.mnt);
615 return -ECHILD;
616 }
617 if (read_seqcount_retry(&dentry->d_seq, nd->seq)) {
618 rcu_read_unlock();
619 dput(dentry);
620 mntput(nd->path.mnt);
621 return -ECHILD;
622 }
623 rcu_read_unlock();
624 }
625
626 if (likely(!(nd->flags & LOOKUP_JUMPED)))
627 return 0;
628
629 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
630 return 0;
631
632 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
633 if (status > 0)
634 return 0;
635
636 if (!status)
637 status = -ESTALE;
638
639 path_put(&nd->path);
640 return status;
641}
642
643static __always_inline void set_root(struct nameidata *nd)
644{
645 if (!nd->root.mnt)
646 get_fs_root(current->fs, &nd->root);
647}
648
649static int link_path_walk(const char *, struct nameidata *);
650
651static __always_inline void set_root_rcu(struct nameidata *nd)
652{
653 if (!nd->root.mnt) {
654 struct fs_struct *fs = current->fs;
655 unsigned seq;
656
657 do {
658 seq = read_seqcount_begin(&fs->seq);
659 nd->root = fs->root;
660 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
661 } while (read_seqcount_retry(&fs->seq, seq));
662 }
663}
664
665static void path_put_conditional(struct path *path, struct nameidata *nd)
666{
667 dput(path->dentry);
668 if (path->mnt != nd->path.mnt)
669 mntput(path->mnt);
670}
671
672static inline void path_to_nameidata(const struct path *path,
673 struct nameidata *nd)
674{
675 if (!(nd->flags & LOOKUP_RCU)) {
676 dput(nd->path.dentry);
677 if (nd->path.mnt != path->mnt)
678 mntput(nd->path.mnt);
679 }
680 nd->path.mnt = path->mnt;
681 nd->path.dentry = path->dentry;
682}
683
684/*
685 * Helper to directly jump to a known parsed path from ->follow_link,
686 * caller must have taken a reference to path beforehand.
687 */
688void nd_jump_link(struct nameidata *nd, struct path *path)
689{
690 path_put(&nd->path);
691
692 nd->path = *path;
693 nd->inode = nd->path.dentry->d_inode;
694 nd->flags |= LOOKUP_JUMPED;
695}
696
697static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
698{
699 struct inode *inode = link->dentry->d_inode;
700 if (inode->i_op->put_link)
701 inode->i_op->put_link(link->dentry, nd, cookie);
702 path_put(link);
703}
704
705int sysctl_protected_symlinks __read_mostly = 0;
706int sysctl_protected_hardlinks __read_mostly = 0;
707
708/**
709 * may_follow_link - Check symlink following for unsafe situations
710 * @link: The path of the symlink
711 * @nd: nameidata pathwalk data
712 *
713 * In the case of the sysctl_protected_symlinks sysctl being enabled,
714 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
715 * in a sticky world-writable directory. This is to protect privileged
716 * processes from failing races against path names that may change out
717 * from under them by way of other users creating malicious symlinks.
718 * It will permit symlinks to be followed only when outside a sticky
719 * world-writable directory, or when the uid of the symlink and follower
720 * match, or when the directory owner matches the symlink's owner.
721 *
722 * Returns 0 if following the symlink is allowed, -ve on error.
723 */
724static inline int may_follow_link(struct path *link, struct nameidata *nd)
725{
726 const struct inode *inode;
727 const struct inode *parent;
728
729 if (!sysctl_protected_symlinks)
730 return 0;
731
732 /* Allowed if owner and follower match. */
733 inode = link->dentry->d_inode;
734 if (uid_eq(current_cred()->fsuid, inode->i_uid))
735 return 0;
736
737 /* Allowed if parent directory not sticky and world-writable. */
738 parent = nd->path.dentry->d_inode;
739 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
740 return 0;
741
742 /* Allowed if parent directory and link owner match. */
743 if (uid_eq(parent->i_uid, inode->i_uid))
744 return 0;
745
746 audit_log_link_denied("follow_link", link);
747 path_put_conditional(link, nd);
748 path_put(&nd->path);
749 return -EACCES;
750}
751
752/**
753 * safe_hardlink_source - Check for safe hardlink conditions
754 * @inode: the source inode to hardlink from
755 *
756 * Return false if at least one of the following conditions:
757 * - inode is not a regular file
758 * - inode is setuid
759 * - inode is setgid and group-exec
760 * - access failure for read and write
761 *
762 * Otherwise returns true.
763 */
764static bool safe_hardlink_source(struct inode *inode)
765{
766 umode_t mode = inode->i_mode;
767
768 /* Special files should not get pinned to the filesystem. */
769 if (!S_ISREG(mode))
770 return false;
771
772 /* Setuid files should not get pinned to the filesystem. */
773 if (mode & S_ISUID)
774 return false;
775
776 /* Executable setgid files should not get pinned to the filesystem. */
777 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
778 return false;
779
780 /* Hardlinking to unreadable or unwritable sources is dangerous. */
781 if (inode_permission(inode, MAY_READ | MAY_WRITE))
782 return false;
783
784 return true;
785}
786
787/**
788 * may_linkat - Check permissions for creating a hardlink
789 * @link: the source to hardlink from
790 *
791 * Block hardlink when all of:
792 * - sysctl_protected_hardlinks enabled
793 * - fsuid does not match inode
794 * - hardlink source is unsafe (see safe_hardlink_source() above)
795 * - not CAP_FOWNER
796 *
797 * Returns 0 if successful, -ve on error.
798 */
799static int may_linkat(struct path *link)
800{
801 const struct cred *cred;
802 struct inode *inode;
803
804 if (!sysctl_protected_hardlinks)
805 return 0;
806
807 cred = current_cred();
808 inode = link->dentry->d_inode;
809
810 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
811 * otherwise, it must be a safe source.
812 */
813 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
814 capable(CAP_FOWNER))
815 return 0;
816
817 audit_log_link_denied("linkat", link);
818 return -EPERM;
819}
820
821static __always_inline int
822follow_link(struct path *link, struct nameidata *nd, void **p)
823{
824 struct dentry *dentry = link->dentry;
825 int error;
826 char *s;
827
828 BUG_ON(nd->flags & LOOKUP_RCU);
829
830 if (link->mnt == nd->path.mnt)
831 mntget(link->mnt);
832
833 error = -ELOOP;
834 if (unlikely(current->total_link_count >= 40))
835 goto out_put_nd_path;
836
837 cond_resched();
838 current->total_link_count++;
839
840 touch_atime(link);
841 nd_set_link(nd, NULL);
842
843 error = security_inode_follow_link(link->dentry, nd);
844 if (error)
845 goto out_put_nd_path;
846
847 nd->last_type = LAST_BIND;
848 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
849 error = PTR_ERR(*p);
850 if (IS_ERR(*p))
851 goto out_put_nd_path;
852
853 error = 0;
854 s = nd_get_link(nd);
855 if (s) {
856 if (unlikely(IS_ERR(s))) {
857 path_put(&nd->path);
858 put_link(nd, link, *p);
859 return PTR_ERR(s);
860 }
861 if (*s == '/') {
862 set_root(nd);
863 path_put(&nd->path);
864 nd->path = nd->root;
865 path_get(&nd->root);
866 nd->flags |= LOOKUP_JUMPED;
867 }
868 nd->inode = nd->path.dentry->d_inode;
869 error = link_path_walk(s, nd);
870 if (unlikely(error))
871 put_link(nd, link, *p);
872 }
873
874 return error;
875
876out_put_nd_path:
877 *p = NULL;
878 path_put(&nd->path);
879 path_put(link);
880 return error;
881}
882
883static int follow_up_rcu(struct path *path)
884{
885 struct mount *mnt = real_mount(path->mnt);
886 struct mount *parent;
887 struct dentry *mountpoint;
888
889 parent = mnt->mnt_parent;
890 if (&parent->mnt == path->mnt)
891 return 0;
892 mountpoint = mnt->mnt_mountpoint;
893 path->dentry = mountpoint;
894 path->mnt = &parent->mnt;
895 return 1;
896}
897
898/*
899 * follow_up - Find the mountpoint of path's vfsmount
900 *
901 * Given a path, find the mountpoint of its source file system.
902 * Replace @path with the path of the mountpoint in the parent mount.
903 * Up is towards /.
904 *
905 * Return 1 if we went up a level and 0 if we were already at the
906 * root.
907 */
908int follow_up(struct path *path)
909{
910 struct mount *mnt = real_mount(path->mnt);
911 struct mount *parent;
912 struct dentry *mountpoint;
913
914 read_seqlock_excl(&mount_lock);
915 parent = mnt->mnt_parent;
916 if (parent == mnt) {
917 read_sequnlock_excl(&mount_lock);
918 return 0;
919 }
920 mntget(&parent->mnt);
921 mountpoint = dget(mnt->mnt_mountpoint);
922 read_sequnlock_excl(&mount_lock);
923 dput(path->dentry);
924 path->dentry = mountpoint;
925 mntput(path->mnt);
926 path->mnt = &parent->mnt;
927 return 1;
928}
929EXPORT_SYMBOL(follow_up);
930
931/*
932 * Perform an automount
933 * - return -EISDIR to tell follow_managed() to stop and return the path we
934 * were called with.
935 */
936static int follow_automount(struct path *path, unsigned flags,
937 bool *need_mntput)
938{
939 struct vfsmount *mnt;
940 int err;
941
942 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
943 return -EREMOTE;
944
945 /* We don't want to mount if someone's just doing a stat -
946 * unless they're stat'ing a directory and appended a '/' to
947 * the name.
948 *
949 * We do, however, want to mount if someone wants to open or
950 * create a file of any type under the mountpoint, wants to
951 * traverse through the mountpoint or wants to open the
952 * mounted directory. Also, autofs may mark negative dentries
953 * as being automount points. These will need the attentions
954 * of the daemon to instantiate them before they can be used.
955 */
956 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
957 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
958 path->dentry->d_inode)
959 return -EISDIR;
960
961 current->total_link_count++;
962 if (current->total_link_count >= 40)
963 return -ELOOP;
964
965 mnt = path->dentry->d_op->d_automount(path);
966 if (IS_ERR(mnt)) {
967 /*
968 * The filesystem is allowed to return -EISDIR here to indicate
969 * it doesn't want to automount. For instance, autofs would do
970 * this so that its userspace daemon can mount on this dentry.
971 *
972 * However, we can only permit this if it's a terminal point in
973 * the path being looked up; if it wasn't then the remainder of
974 * the path is inaccessible and we should say so.
975 */
976 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
977 return -EREMOTE;
978 return PTR_ERR(mnt);
979 }
980
981 if (!mnt) /* mount collision */
982 return 0;
983
984 if (!*need_mntput) {
985 /* lock_mount() may release path->mnt on error */
986 mntget(path->mnt);
987 *need_mntput = true;
988 }
989 err = finish_automount(mnt, path);
990
991 switch (err) {
992 case -EBUSY:
993 /* Someone else made a mount here whilst we were busy */
994 return 0;
995 case 0:
996 path_put(path);
997 path->mnt = mnt;
998 path->dentry = dget(mnt->mnt_root);
999 return 0;
1000 default:
1001 return err;
1002 }
1003
1004}
1005
1006/*
1007 * Handle a dentry that is managed in some way.
1008 * - Flagged for transit management (autofs)
1009 * - Flagged as mountpoint
1010 * - Flagged as automount point
1011 *
1012 * This may only be called in refwalk mode.
1013 *
1014 * Serialization is taken care of in namespace.c
1015 */
1016static int follow_managed(struct path *path, unsigned flags)
1017{
1018 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1019 unsigned managed;
1020 bool need_mntput = false;
1021 int ret = 0;
1022
1023 /* Given that we're not holding a lock here, we retain the value in a
1024 * local variable for each dentry as we look at it so that we don't see
1025 * the components of that value change under us */
1026 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1027 managed &= DCACHE_MANAGED_DENTRY,
1028 unlikely(managed != 0)) {
1029 /* Allow the filesystem to manage the transit without i_mutex
1030 * being held. */
1031 if (managed & DCACHE_MANAGE_TRANSIT) {
1032 BUG_ON(!path->dentry->d_op);
1033 BUG_ON(!path->dentry->d_op->d_manage);
1034 ret = path->dentry->d_op->d_manage(path->dentry, false);
1035 if (ret < 0)
1036 break;
1037 }
1038
1039 /* Transit to a mounted filesystem. */
1040 if (managed & DCACHE_MOUNTED) {
1041 struct vfsmount *mounted = lookup_mnt(path);
1042 if (mounted) {
1043 dput(path->dentry);
1044 if (need_mntput)
1045 mntput(path->mnt);
1046 path->mnt = mounted;
1047 path->dentry = dget(mounted->mnt_root);
1048 need_mntput = true;
1049 continue;
1050 }
1051
1052 /* Something is mounted on this dentry in another
1053 * namespace and/or whatever was mounted there in this
1054 * namespace got unmounted before lookup_mnt() could
1055 * get it */
1056 }
1057
1058 /* Handle an automount point */
1059 if (managed & DCACHE_NEED_AUTOMOUNT) {
1060 ret = follow_automount(path, flags, &need_mntput);
1061 if (ret < 0)
1062 break;
1063 continue;
1064 }
1065
1066 /* We didn't change the current path point */
1067 break;
1068 }
1069
1070 if (need_mntput && path->mnt == mnt)
1071 mntput(path->mnt);
1072 if (ret == -EISDIR)
1073 ret = 0;
1074 return ret < 0 ? ret : need_mntput;
1075}
1076
1077int follow_down_one(struct path *path)
1078{
1079 struct vfsmount *mounted;
1080
1081 mounted = lookup_mnt(path);
1082 if (mounted) {
1083 dput(path->dentry);
1084 mntput(path->mnt);
1085 path->mnt = mounted;
1086 path->dentry = dget(mounted->mnt_root);
1087 return 1;
1088 }
1089 return 0;
1090}
1091EXPORT_SYMBOL(follow_down_one);
1092
1093static inline bool managed_dentry_might_block(struct dentry *dentry)
1094{
1095 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
1096 dentry->d_op->d_manage(dentry, true) < 0);
1097}
1098
1099/*
1100 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1101 * we meet a managed dentry that would need blocking.
1102 */
1103static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1104 struct inode **inode)
1105{
1106 for (;;) {
1107 struct mount *mounted;
1108 /*
1109 * Don't forget we might have a non-mountpoint managed dentry
1110 * that wants to block transit.
1111 */
1112 if (unlikely(managed_dentry_might_block(path->dentry)))
1113 return false;
1114
1115 if (!d_mountpoint(path->dentry))
1116 return true;
1117
1118 mounted = __lookup_mnt(path->mnt, path->dentry);
1119 if (!mounted)
1120 break;
1121 path->mnt = &mounted->mnt;
1122 path->dentry = mounted->mnt.mnt_root;
1123 nd->flags |= LOOKUP_JUMPED;
1124 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1125 /*
1126 * Update the inode too. We don't need to re-check the
1127 * dentry sequence number here after this d_inode read,
1128 * because a mount-point is always pinned.
1129 */
1130 *inode = path->dentry->d_inode;
1131 }
1132 return read_seqretry(&mount_lock, nd->m_seq);
1133}
1134
1135static int follow_dotdot_rcu(struct nameidata *nd)
1136{
1137 set_root_rcu(nd);
1138
1139 while (1) {
1140 if (nd->path.dentry == nd->root.dentry &&
1141 nd->path.mnt == nd->root.mnt) {
1142 break;
1143 }
1144 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1145 struct dentry *old = nd->path.dentry;
1146 struct dentry *parent = old->d_parent;
1147 unsigned seq;
1148
1149 seq = read_seqcount_begin(&parent->d_seq);
1150 if (read_seqcount_retry(&old->d_seq, nd->seq))
1151 goto failed;
1152 nd->path.dentry = parent;
1153 nd->seq = seq;
1154 break;
1155 }
1156 if (!follow_up_rcu(&nd->path))
1157 break;
1158 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1159 }
1160 while (d_mountpoint(nd->path.dentry)) {
1161 struct mount *mounted;
1162 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1163 if (!mounted)
1164 break;
1165 nd->path.mnt = &mounted->mnt;
1166 nd->path.dentry = mounted->mnt.mnt_root;
1167 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1168 if (!read_seqretry(&mount_lock, nd->m_seq))
1169 goto failed;
1170 }
1171 nd->inode = nd->path.dentry->d_inode;
1172 return 0;
1173
1174failed:
1175 nd->flags &= ~LOOKUP_RCU;
1176 if (!(nd->flags & LOOKUP_ROOT))
1177 nd->root.mnt = NULL;
1178 rcu_read_unlock();
1179 return -ECHILD;
1180}
1181
1182/*
1183 * Follow down to the covering mount currently visible to userspace. At each
1184 * point, the filesystem owning that dentry may be queried as to whether the
1185 * caller is permitted to proceed or not.
1186 */
1187int follow_down(struct path *path)
1188{
1189 unsigned managed;
1190 int ret;
1191
1192 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1193 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1194 /* Allow the filesystem to manage the transit without i_mutex
1195 * being held.
1196 *
1197 * We indicate to the filesystem if someone is trying to mount
1198 * something here. This gives autofs the chance to deny anyone
1199 * other than its daemon the right to mount on its
1200 * superstructure.
1201 *
1202 * The filesystem may sleep at this point.
1203 */
1204 if (managed & DCACHE_MANAGE_TRANSIT) {
1205 BUG_ON(!path->dentry->d_op);
1206 BUG_ON(!path->dentry->d_op->d_manage);
1207 ret = path->dentry->d_op->d_manage(
1208 path->dentry, false);
1209 if (ret < 0)
1210 return ret == -EISDIR ? 0 : ret;
1211 }
1212
1213 /* Transit to a mounted filesystem. */
1214 if (managed & DCACHE_MOUNTED) {
1215 struct vfsmount *mounted = lookup_mnt(path);
1216 if (!mounted)
1217 break;
1218 dput(path->dentry);
1219 mntput(path->mnt);
1220 path->mnt = mounted;
1221 path->dentry = dget(mounted->mnt_root);
1222 continue;
1223 }
1224
1225 /* Don't handle automount points here */
1226 break;
1227 }
1228 return 0;
1229}
1230EXPORT_SYMBOL(follow_down);
1231
1232/*
1233 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1234 */
1235static void follow_mount(struct path *path)
1236{
1237 while (d_mountpoint(path->dentry)) {
1238 struct vfsmount *mounted = lookup_mnt(path);
1239 if (!mounted)
1240 break;
1241 dput(path->dentry);
1242 mntput(path->mnt);
1243 path->mnt = mounted;
1244 path->dentry = dget(mounted->mnt_root);
1245 }
1246}
1247
1248static void follow_dotdot(struct nameidata *nd)
1249{
1250 set_root(nd);
1251
1252 while(1) {
1253 struct dentry *old = nd->path.dentry;
1254
1255 if (nd->path.dentry == nd->root.dentry &&
1256 nd->path.mnt == nd->root.mnt) {
1257 break;
1258 }
1259 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1260 /* rare case of legitimate dget_parent()... */
1261 nd->path.dentry = dget_parent(nd->path.dentry);
1262 dput(old);
1263 break;
1264 }
1265 if (!follow_up(&nd->path))
1266 break;
1267 }
1268 follow_mount(&nd->path);
1269 nd->inode = nd->path.dentry->d_inode;
1270}
1271
1272/*
1273 * This looks up the name in dcache, possibly revalidates the old dentry and
1274 * allocates a new one if not found or not valid. In the need_lookup argument
1275 * returns whether i_op->lookup is necessary.
1276 *
1277 * dir->d_inode->i_mutex must be held
1278 */
1279static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1280 unsigned int flags, bool *need_lookup)
1281{
1282 struct dentry *dentry;
1283 int error;
1284
1285 *need_lookup = false;
1286 dentry = d_lookup(dir, name);
1287 if (dentry) {
1288 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1289 error = d_revalidate(dentry, flags);
1290 if (unlikely(error <= 0)) {
1291 if (error < 0) {
1292 dput(dentry);
1293 return ERR_PTR(error);
1294 } else if (!d_invalidate(dentry)) {
1295 dput(dentry);
1296 dentry = NULL;
1297 }
1298 }
1299 }
1300 }
1301
1302 if (!dentry) {
1303 dentry = d_alloc(dir, name);
1304 if (unlikely(!dentry))
1305 return ERR_PTR(-ENOMEM);
1306
1307 *need_lookup = true;
1308 }
1309 return dentry;
1310}
1311
1312/*
1313 * Call i_op->lookup on the dentry. The dentry must be negative and
1314 * unhashed.
1315 *
1316 * dir->d_inode->i_mutex must be held
1317 */
1318static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1319 unsigned int flags)
1320{
1321 struct dentry *old;
1322
1323 /* Don't create child dentry for a dead directory. */
1324 if (unlikely(IS_DEADDIR(dir))) {
1325 dput(dentry);
1326 return ERR_PTR(-ENOENT);
1327 }
1328
1329 old = dir->i_op->lookup(dir, dentry, flags);
1330 if (unlikely(old)) {
1331 dput(dentry);
1332 dentry = old;
1333 }
1334 return dentry;
1335}
1336
1337static struct dentry *__lookup_hash(struct qstr *name,
1338 struct dentry *base, unsigned int flags)
1339{
1340 bool need_lookup;
1341 struct dentry *dentry;
1342
1343 dentry = lookup_dcache(name, base, flags, &need_lookup);
1344 if (!need_lookup)
1345 return dentry;
1346
1347 return lookup_real(base->d_inode, dentry, flags);
1348}
1349
1350/*
1351 * It's more convoluted than I'd like it to be, but... it's still fairly
1352 * small and for now I'd prefer to have fast path as straight as possible.
1353 * It _is_ time-critical.
1354 */
1355static int lookup_fast(struct nameidata *nd,
1356 struct path *path, struct inode **inode)
1357{
1358 struct vfsmount *mnt = nd->path.mnt;
1359 struct dentry *dentry, *parent = nd->path.dentry;
1360 int need_reval = 1;
1361 int status = 1;
1362 int err;
1363
1364 /*
1365 * Rename seqlock is not required here because in the off chance
1366 * of a false negative due to a concurrent rename, we're going to
1367 * do the non-racy lookup, below.
1368 */
1369 if (nd->flags & LOOKUP_RCU) {
1370 unsigned seq;
1371 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1372 if (!dentry)
1373 goto unlazy;
1374
1375 /*
1376 * This sequence count validates that the inode matches
1377 * the dentry name information from lookup.
1378 */
1379 *inode = dentry->d_inode;
1380 if (read_seqcount_retry(&dentry->d_seq, seq))
1381 return -ECHILD;
1382
1383 /*
1384 * This sequence count validates that the parent had no
1385 * changes while we did the lookup of the dentry above.
1386 *
1387 * The memory barrier in read_seqcount_begin of child is
1388 * enough, we can use __read_seqcount_retry here.
1389 */
1390 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1391 return -ECHILD;
1392 nd->seq = seq;
1393
1394 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1395 status = d_revalidate(dentry, nd->flags);
1396 if (unlikely(status <= 0)) {
1397 if (status != -ECHILD)
1398 need_reval = 0;
1399 goto unlazy;
1400 }
1401 }
1402 path->mnt = mnt;
1403 path->dentry = dentry;
1404 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1405 goto unlazy;
1406 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1407 goto unlazy;
1408 return 0;
1409unlazy:
1410 if (unlazy_walk(nd, dentry))
1411 return -ECHILD;
1412 } else {
1413 dentry = __d_lookup(parent, &nd->last);
1414 }
1415
1416 if (unlikely(!dentry))
1417 goto need_lookup;
1418
1419 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1420 status = d_revalidate(dentry, nd->flags);
1421 if (unlikely(status <= 0)) {
1422 if (status < 0) {
1423 dput(dentry);
1424 return status;
1425 }
1426 if (!d_invalidate(dentry)) {
1427 dput(dentry);
1428 goto need_lookup;
1429 }
1430 }
1431
1432 path->mnt = mnt;
1433 path->dentry = dentry;
1434 err = follow_managed(path, nd->flags);
1435 if (unlikely(err < 0)) {
1436 path_put_conditional(path, nd);
1437 return err;
1438 }
1439 if (err)
1440 nd->flags |= LOOKUP_JUMPED;
1441 *inode = path->dentry->d_inode;
1442 return 0;
1443
1444need_lookup:
1445 return 1;
1446}
1447
1448/* Fast lookup failed, do it the slow way */
1449static int lookup_slow(struct nameidata *nd, struct path *path)
1450{
1451 struct dentry *dentry, *parent;
1452 int err;
1453
1454 parent = nd->path.dentry;
1455 BUG_ON(nd->inode != parent->d_inode);
1456
1457 mutex_lock(&parent->d_inode->i_mutex);
1458 dentry = __lookup_hash(&nd->last, parent, nd->flags);
1459 mutex_unlock(&parent->d_inode->i_mutex);
1460 if (IS_ERR(dentry))
1461 return PTR_ERR(dentry);
1462 path->mnt = nd->path.mnt;
1463 path->dentry = dentry;
1464 err = follow_managed(path, nd->flags);
1465 if (unlikely(err < 0)) {
1466 path_put_conditional(path, nd);
1467 return err;
1468 }
1469 if (err)
1470 nd->flags |= LOOKUP_JUMPED;
1471 return 0;
1472}
1473
1474static inline int may_lookup(struct nameidata *nd)
1475{
1476 if (nd->flags & LOOKUP_RCU) {
1477 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1478 if (err != -ECHILD)
1479 return err;
1480 if (unlazy_walk(nd, NULL))
1481 return -ECHILD;
1482 }
1483 return inode_permission(nd->inode, MAY_EXEC);
1484}
1485
1486static inline int handle_dots(struct nameidata *nd, int type)
1487{
1488 if (type == LAST_DOTDOT) {
1489 if (nd->flags & LOOKUP_RCU) {
1490 if (follow_dotdot_rcu(nd))
1491 return -ECHILD;
1492 } else
1493 follow_dotdot(nd);
1494 }
1495 return 0;
1496}
1497
1498static void terminate_walk(struct nameidata *nd)
1499{
1500 if (!(nd->flags & LOOKUP_RCU)) {
1501 path_put(&nd->path);
1502 } else {
1503 nd->flags &= ~LOOKUP_RCU;
1504 if (!(nd->flags & LOOKUP_ROOT))
1505 nd->root.mnt = NULL;
1506 rcu_read_unlock();
1507 }
1508}
1509
1510/*
1511 * Do we need to follow links? We _really_ want to be able
1512 * to do this check without having to look at inode->i_op,
1513 * so we keep a cache of "no, this doesn't need follow_link"
1514 * for the common case.
1515 */
1516static inline int should_follow_link(struct dentry *dentry, int follow)
1517{
1518 return unlikely(d_is_symlink(dentry)) ? follow : 0;
1519}
1520
1521static inline int walk_component(struct nameidata *nd, struct path *path,
1522 int follow)
1523{
1524 struct inode *inode;
1525 int err;
1526 /*
1527 * "." and ".." are special - ".." especially so because it has
1528 * to be able to know about the current root directory and
1529 * parent relationships.
1530 */
1531 if (unlikely(nd->last_type != LAST_NORM))
1532 return handle_dots(nd, nd->last_type);
1533 err = lookup_fast(nd, path, &inode);
1534 if (unlikely(err)) {
1535 if (err < 0)
1536 goto out_err;
1537
1538 err = lookup_slow(nd, path);
1539 if (err < 0)
1540 goto out_err;
1541
1542 inode = path->dentry->d_inode;
1543 }
1544 err = -ENOENT;
1545 if (!inode || d_is_negative(path->dentry))
1546 goto out_path_put;
1547
1548 if (should_follow_link(path->dentry, follow)) {
1549 if (nd->flags & LOOKUP_RCU) {
1550 if (unlikely(unlazy_walk(nd, path->dentry))) {
1551 err = -ECHILD;
1552 goto out_err;
1553 }
1554 }
1555 BUG_ON(inode != path->dentry->d_inode);
1556 return 1;
1557 }
1558 path_to_nameidata(path, nd);
1559 nd->inode = inode;
1560 return 0;
1561
1562out_path_put:
1563 path_to_nameidata(path, nd);
1564out_err:
1565 terminate_walk(nd);
1566 return err;
1567}
1568
1569/*
1570 * This limits recursive symlink follows to 8, while
1571 * limiting consecutive symlinks to 40.
1572 *
1573 * Without that kind of total limit, nasty chains of consecutive
1574 * symlinks can cause almost arbitrarily long lookups.
1575 */
1576static inline int nested_symlink(struct path *path, struct nameidata *nd)
1577{
1578 int res;
1579
1580 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1581 path_put_conditional(path, nd);
1582 path_put(&nd->path);
1583 return -ELOOP;
1584 }
1585 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1586
1587 nd->depth++;
1588 current->link_count++;
1589
1590 do {
1591 struct path link = *path;
1592 void *cookie;
1593
1594 res = follow_link(&link, nd, &cookie);
1595 if (res)
1596 break;
1597 res = walk_component(nd, path, LOOKUP_FOLLOW);
1598 put_link(nd, &link, cookie);
1599 } while (res > 0);
1600
1601 current->link_count--;
1602 nd->depth--;
1603 return res;
1604}
1605
1606/*
1607 * We can do the critical dentry name comparison and hashing
1608 * operations one word at a time, but we are limited to:
1609 *
1610 * - Architectures with fast unaligned word accesses. We could
1611 * do a "get_unaligned()" if this helps and is sufficiently
1612 * fast.
1613 *
1614 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1615 * do not trap on the (extremely unlikely) case of a page
1616 * crossing operation.
1617 *
1618 * - Furthermore, we need an efficient 64-bit compile for the
1619 * 64-bit case in order to generate the "number of bytes in
1620 * the final mask". Again, that could be replaced with a
1621 * efficient population count instruction or similar.
1622 */
1623#ifdef CONFIG_DCACHE_WORD_ACCESS
1624
1625#include <asm/word-at-a-time.h>
1626
1627#ifdef CONFIG_64BIT
1628
1629static inline unsigned int fold_hash(unsigned long hash)
1630{
1631 hash += hash >> (8*sizeof(int));
1632 return hash;
1633}
1634
1635#else /* 32-bit case */
1636
1637#define fold_hash(x) (x)
1638
1639#endif
1640
1641unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1642{
1643 unsigned long a, mask;
1644 unsigned long hash = 0;
1645
1646 for (;;) {
1647 a = load_unaligned_zeropad(name);
1648 if (len < sizeof(unsigned long))
1649 break;
1650 hash += a;
1651 hash *= 9;
1652 name += sizeof(unsigned long);
1653 len -= sizeof(unsigned long);
1654 if (!len)
1655 goto done;
1656 }
1657 mask = bytemask_from_count(len);
1658 hash += mask & a;
1659done:
1660 return fold_hash(hash);
1661}
1662EXPORT_SYMBOL(full_name_hash);
1663
1664/*
1665 * Calculate the length and hash of the path component, and
1666 * return the length of the component;
1667 */
1668static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1669{
1670 unsigned long a, b, adata, bdata, mask, hash, len;
1671 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1672
1673 hash = a = 0;
1674 len = -sizeof(unsigned long);
1675 do {
1676 hash = (hash + a) * 9;
1677 len += sizeof(unsigned long);
1678 a = load_unaligned_zeropad(name+len);
1679 b = a ^ REPEAT_BYTE('/');
1680 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1681
1682 adata = prep_zero_mask(a, adata, &constants);
1683 bdata = prep_zero_mask(b, bdata, &constants);
1684
1685 mask = create_zero_mask(adata | bdata);
1686
1687 hash += a & zero_bytemask(mask);
1688 *hashp = fold_hash(hash);
1689
1690 return len + find_zero(mask);
1691}
1692
1693#else
1694
1695unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1696{
1697 unsigned long hash = init_name_hash();
1698 while (len--)
1699 hash = partial_name_hash(*name++, hash);
1700 return end_name_hash(hash);
1701}
1702EXPORT_SYMBOL(full_name_hash);
1703
1704/*
1705 * We know there's a real path component here of at least
1706 * one character.
1707 */
1708static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1709{
1710 unsigned long hash = init_name_hash();
1711 unsigned long len = 0, c;
1712
1713 c = (unsigned char)*name;
1714 do {
1715 len++;
1716 hash = partial_name_hash(c, hash);
1717 c = (unsigned char)name[len];
1718 } while (c && c != '/');
1719 *hashp = end_name_hash(hash);
1720 return len;
1721}
1722
1723#endif
1724
1725/*
1726 * Name resolution.
1727 * This is the basic name resolution function, turning a pathname into
1728 * the final dentry. We expect 'base' to be positive and a directory.
1729 *
1730 * Returns 0 and nd will have valid dentry and mnt on success.
1731 * Returns error and drops reference to input namei data on failure.
1732 */
1733static int link_path_walk(const char *name, struct nameidata *nd)
1734{
1735 struct path next;
1736 int err;
1737
1738 while (*name=='/')
1739 name++;
1740 if (!*name)
1741 return 0;
1742
1743 /* At this point we know we have a real path component. */
1744 for(;;) {
1745 struct qstr this;
1746 long len;
1747 int type;
1748
1749 err = may_lookup(nd);
1750 if (err)
1751 break;
1752
1753 len = hash_name(name, &this.hash);
1754 this.name = name;
1755 this.len = len;
1756
1757 type = LAST_NORM;
1758 if (name[0] == '.') switch (len) {
1759 case 2:
1760 if (name[1] == '.') {
1761 type = LAST_DOTDOT;
1762 nd->flags |= LOOKUP_JUMPED;
1763 }
1764 break;
1765 case 1:
1766 type = LAST_DOT;
1767 }
1768 if (likely(type == LAST_NORM)) {
1769 struct dentry *parent = nd->path.dentry;
1770 nd->flags &= ~LOOKUP_JUMPED;
1771 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1772 err = parent->d_op->d_hash(parent, &this);
1773 if (err < 0)
1774 break;
1775 }
1776 }
1777
1778 nd->last = this;
1779 nd->last_type = type;
1780
1781 if (!name[len])
1782 return 0;
1783 /*
1784 * If it wasn't NUL, we know it was '/'. Skip that
1785 * slash, and continue until no more slashes.
1786 */
1787 do {
1788 len++;
1789 } while (unlikely(name[len] == '/'));
1790 if (!name[len])
1791 return 0;
1792
1793 name += len;
1794
1795 err = walk_component(nd, &next, LOOKUP_FOLLOW);
1796 if (err < 0)
1797 return err;
1798
1799 if (err) {
1800 err = nested_symlink(&next, nd);
1801 if (err)
1802 return err;
1803 }
1804 if (!d_can_lookup(nd->path.dentry)) {
1805 err = -ENOTDIR;
1806 break;
1807 }
1808 }
1809 terminate_walk(nd);
1810 return err;
1811}
1812
1813static int path_init(int dfd, const char *name, unsigned int flags,
1814 struct nameidata *nd, struct file **fp)
1815{
1816 int retval = 0;
1817
1818 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1819 nd->flags = flags | LOOKUP_JUMPED;
1820 nd->depth = 0;
1821 if (flags & LOOKUP_ROOT) {
1822 struct dentry *root = nd->root.dentry;
1823 struct inode *inode = root->d_inode;
1824 if (*name) {
1825 if (!d_can_lookup(root))
1826 return -ENOTDIR;
1827 retval = inode_permission(inode, MAY_EXEC);
1828 if (retval)
1829 return retval;
1830 }
1831 nd->path = nd->root;
1832 nd->inode = inode;
1833 if (flags & LOOKUP_RCU) {
1834 rcu_read_lock();
1835 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1836 nd->m_seq = read_seqbegin(&mount_lock);
1837 } else {
1838 path_get(&nd->path);
1839 }
1840 return 0;
1841 }
1842
1843 nd->root.mnt = NULL;
1844
1845 nd->m_seq = read_seqbegin(&mount_lock);
1846 if (*name=='/') {
1847 if (flags & LOOKUP_RCU) {
1848 rcu_read_lock();
1849 set_root_rcu(nd);
1850 } else {
1851 set_root(nd);
1852 path_get(&nd->root);
1853 }
1854 nd->path = nd->root;
1855 } else if (dfd == AT_FDCWD) {
1856 if (flags & LOOKUP_RCU) {
1857 struct fs_struct *fs = current->fs;
1858 unsigned seq;
1859
1860 rcu_read_lock();
1861
1862 do {
1863 seq = read_seqcount_begin(&fs->seq);
1864 nd->path = fs->pwd;
1865 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1866 } while (read_seqcount_retry(&fs->seq, seq));
1867 } else {
1868 get_fs_pwd(current->fs, &nd->path);
1869 }
1870 } else {
1871 /* Caller must check execute permissions on the starting path component */
1872 struct fd f = fdget_raw(dfd);
1873 struct dentry *dentry;
1874
1875 if (!f.file)
1876 return -EBADF;
1877
1878 dentry = f.file->f_path.dentry;
1879
1880 if (*name) {
1881 if (!d_can_lookup(dentry)) {
1882 fdput(f);
1883 return -ENOTDIR;
1884 }
1885 }
1886
1887 nd->path = f.file->f_path;
1888 if (flags & LOOKUP_RCU) {
1889 if (f.flags & FDPUT_FPUT)
1890 *fp = f.file;
1891 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1892 rcu_read_lock();
1893 } else {
1894 path_get(&nd->path);
1895 fdput(f);
1896 }
1897 }
1898
1899 nd->inode = nd->path.dentry->d_inode;
1900 return 0;
1901}
1902
1903static inline int lookup_last(struct nameidata *nd, struct path *path)
1904{
1905 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1906 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1907
1908 nd->flags &= ~LOOKUP_PARENT;
1909 return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW);
1910}
1911
1912/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1913static int path_lookupat(int dfd, const char *name,
1914 unsigned int flags, struct nameidata *nd)
1915{
1916 struct file *base = NULL;
1917 struct path path;
1918 int err;
1919
1920 /*
1921 * Path walking is largely split up into 2 different synchronisation
1922 * schemes, rcu-walk and ref-walk (explained in
1923 * Documentation/filesystems/path-lookup.txt). These share much of the
1924 * path walk code, but some things particularly setup, cleanup, and
1925 * following mounts are sufficiently divergent that functions are
1926 * duplicated. Typically there is a function foo(), and its RCU
1927 * analogue, foo_rcu().
1928 *
1929 * -ECHILD is the error number of choice (just to avoid clashes) that
1930 * is returned if some aspect of an rcu-walk fails. Such an error must
1931 * be handled by restarting a traditional ref-walk (which will always
1932 * be able to complete).
1933 */
1934 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1935
1936 if (unlikely(err))
1937 return err;
1938
1939 current->total_link_count = 0;
1940 err = link_path_walk(name, nd);
1941
1942 if (!err && !(flags & LOOKUP_PARENT)) {
1943 err = lookup_last(nd, &path);
1944 while (err > 0) {
1945 void *cookie;
1946 struct path link = path;
1947 err = may_follow_link(&link, nd);
1948 if (unlikely(err))
1949 break;
1950 nd->flags |= LOOKUP_PARENT;
1951 err = follow_link(&link, nd, &cookie);
1952 if (err)
1953 break;
1954 err = lookup_last(nd, &path);
1955 put_link(nd, &link, cookie);
1956 }
1957 }
1958
1959 if (!err)
1960 err = complete_walk(nd);
1961
1962 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1963 if (!d_can_lookup(nd->path.dentry)) {
1964 path_put(&nd->path);
1965 err = -ENOTDIR;
1966 }
1967 }
1968
1969 if (base)
1970 fput(base);
1971
1972 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1973 path_put(&nd->root);
1974 nd->root.mnt = NULL;
1975 }
1976 return err;
1977}
1978
1979static int filename_lookup(int dfd, struct filename *name,
1980 unsigned int flags, struct nameidata *nd)
1981{
1982 int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd);
1983 if (unlikely(retval == -ECHILD))
1984 retval = path_lookupat(dfd, name->name, flags, nd);
1985 if (unlikely(retval == -ESTALE))
1986 retval = path_lookupat(dfd, name->name,
1987 flags | LOOKUP_REVAL, nd);
1988
1989 if (likely(!retval))
1990 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
1991 return retval;
1992}
1993
1994static int do_path_lookup(int dfd, const char *name,
1995 unsigned int flags, struct nameidata *nd)
1996{
1997 struct filename filename = { .name = name };
1998
1999 return filename_lookup(dfd, &filename, flags, nd);
2000}
2001
2002/* does lookup, returns the object with parent locked */
2003struct dentry *kern_path_locked(const char *name, struct path *path)
2004{
2005 struct nameidata nd;
2006 struct dentry *d;
2007 int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd);
2008 if (err)
2009 return ERR_PTR(err);
2010 if (nd.last_type != LAST_NORM) {
2011 path_put(&nd.path);
2012 return ERR_PTR(-EINVAL);
2013 }
2014 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2015 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2016 if (IS_ERR(d)) {
2017 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2018 path_put(&nd.path);
2019 return d;
2020 }
2021 *path = nd.path;
2022 return d;
2023}
2024
2025int kern_path(const char *name, unsigned int flags, struct path *path)
2026{
2027 struct nameidata nd;
2028 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
2029 if (!res)
2030 *path = nd.path;
2031 return res;
2032}
2033EXPORT_SYMBOL(kern_path);
2034
2035/**
2036 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2037 * @dentry: pointer to dentry of the base directory
2038 * @mnt: pointer to vfs mount of the base directory
2039 * @name: pointer to file name
2040 * @flags: lookup flags
2041 * @path: pointer to struct path to fill
2042 */
2043int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2044 const char *name, unsigned int flags,
2045 struct path *path)
2046{
2047 struct nameidata nd;
2048 int err;
2049 nd.root.dentry = dentry;
2050 nd.root.mnt = mnt;
2051 BUG_ON(flags & LOOKUP_PARENT);
2052 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2053 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
2054 if (!err)
2055 *path = nd.path;
2056 return err;
2057}
2058EXPORT_SYMBOL(vfs_path_lookup);
2059
2060/*
2061 * Restricted form of lookup. Doesn't follow links, single-component only,
2062 * needs parent already locked. Doesn't follow mounts.
2063 * SMP-safe.
2064 */
2065static struct dentry *lookup_hash(struct nameidata *nd)
2066{
2067 return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
2068}
2069
2070/**
2071 * lookup_one_len - filesystem helper to lookup single pathname component
2072 * @name: pathname component to lookup
2073 * @base: base directory to lookup from
2074 * @len: maximum length @len should be interpreted to
2075 *
2076 * Note that this routine is purely a helper for filesystem usage and should
2077 * not be called by generic code. Also note that by using this function the
2078 * nameidata argument is passed to the filesystem methods and a filesystem
2079 * using this helper needs to be prepared for that.
2080 */
2081struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2082{
2083 struct qstr this;
2084 unsigned int c;
2085 int err;
2086
2087 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2088
2089 this.name = name;
2090 this.len = len;
2091 this.hash = full_name_hash(name, len);
2092 if (!len)
2093 return ERR_PTR(-EACCES);
2094
2095 if (unlikely(name[0] == '.')) {
2096 if (len < 2 || (len == 2 && name[1] == '.'))
2097 return ERR_PTR(-EACCES);
2098 }
2099
2100 while (len--) {
2101 c = *(const unsigned char *)name++;
2102 if (c == '/' || c == '\0')
2103 return ERR_PTR(-EACCES);
2104 }
2105 /*
2106 * See if the low-level filesystem might want
2107 * to use its own hash..
2108 */
2109 if (base->d_flags & DCACHE_OP_HASH) {
2110 int err = base->d_op->d_hash(base, &this);
2111 if (err < 0)
2112 return ERR_PTR(err);
2113 }
2114
2115 err = inode_permission(base->d_inode, MAY_EXEC);
2116 if (err)
2117 return ERR_PTR(err);
2118
2119 return __lookup_hash(&this, base, 0);
2120}
2121EXPORT_SYMBOL(lookup_one_len);
2122
2123int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2124 struct path *path, int *empty)
2125{
2126 struct nameidata nd;
2127 struct filename *tmp = getname_flags(name, flags, empty);
2128 int err = PTR_ERR(tmp);
2129 if (!IS_ERR(tmp)) {
2130
2131 BUG_ON(flags & LOOKUP_PARENT);
2132
2133 err = filename_lookup(dfd, tmp, flags, &nd);
2134 putname(tmp);
2135 if (!err)
2136 *path = nd.path;
2137 }
2138 return err;
2139}
2140
2141int user_path_at(int dfd, const char __user *name, unsigned flags,
2142 struct path *path)
2143{
2144 return user_path_at_empty(dfd, name, flags, path, NULL);
2145}
2146EXPORT_SYMBOL(user_path_at);
2147
2148/*
2149 * NB: most callers don't do anything directly with the reference to the
2150 * to struct filename, but the nd->last pointer points into the name string
2151 * allocated by getname. So we must hold the reference to it until all
2152 * path-walking is complete.
2153 */
2154static struct filename *
2155user_path_parent(int dfd, const char __user *path, struct nameidata *nd,
2156 unsigned int flags)
2157{
2158 struct filename *s = getname(path);
2159 int error;
2160
2161 /* only LOOKUP_REVAL is allowed in extra flags */
2162 flags &= LOOKUP_REVAL;
2163
2164 if (IS_ERR(s))
2165 return s;
2166
2167 error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd);
2168 if (error) {
2169 putname(s);
2170 return ERR_PTR(error);
2171 }
2172
2173 return s;
2174}
2175
2176/**
2177 * mountpoint_last - look up last component for umount
2178 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2179 * @path: pointer to container for result
2180 *
2181 * This is a special lookup_last function just for umount. In this case, we
2182 * need to resolve the path without doing any revalidation.
2183 *
2184 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2185 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2186 * in almost all cases, this lookup will be served out of the dcache. The only
2187 * cases where it won't are if nd->last refers to a symlink or the path is
2188 * bogus and it doesn't exist.
2189 *
2190 * Returns:
2191 * -error: if there was an error during lookup. This includes -ENOENT if the
2192 * lookup found a negative dentry. The nd->path reference will also be
2193 * put in this case.
2194 *
2195 * 0: if we successfully resolved nd->path and found it to not to be a
2196 * symlink that needs to be followed. "path" will also be populated.
2197 * The nd->path reference will also be put.
2198 *
2199 * 1: if we successfully resolved nd->last and found it to be a symlink
2200 * that needs to be followed. "path" will be populated with the path
2201 * to the link, and nd->path will *not* be put.
2202 */
2203static int
2204mountpoint_last(struct nameidata *nd, struct path *path)
2205{
2206 int error = 0;
2207 struct dentry *dentry;
2208 struct dentry *dir = nd->path.dentry;
2209
2210 /* If we're in rcuwalk, drop out of it to handle last component */
2211 if (nd->flags & LOOKUP_RCU) {
2212 if (unlazy_walk(nd, NULL)) {
2213 error = -ECHILD;
2214 goto out;
2215 }
2216 }
2217
2218 nd->flags &= ~LOOKUP_PARENT;
2219
2220 if (unlikely(nd->last_type != LAST_NORM)) {
2221 error = handle_dots(nd, nd->last_type);
2222 if (error)
2223 goto out;
2224 dentry = dget(nd->path.dentry);
2225 goto done;
2226 }
2227
2228 mutex_lock(&dir->d_inode->i_mutex);
2229 dentry = d_lookup(dir, &nd->last);
2230 if (!dentry) {
2231 /*
2232 * No cached dentry. Mounted dentries are pinned in the cache,
2233 * so that means that this dentry is probably a symlink or the
2234 * path doesn't actually point to a mounted dentry.
2235 */
2236 dentry = d_alloc(dir, &nd->last);
2237 if (!dentry) {
2238 error = -ENOMEM;
2239 mutex_unlock(&dir->d_inode->i_mutex);
2240 goto out;
2241 }
2242 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2243 error = PTR_ERR(dentry);
2244 if (IS_ERR(dentry)) {
2245 mutex_unlock(&dir->d_inode->i_mutex);
2246 goto out;
2247 }
2248 }
2249 mutex_unlock(&dir->d_inode->i_mutex);
2250
2251done:
2252 if (!dentry->d_inode || d_is_negative(dentry)) {
2253 error = -ENOENT;
2254 dput(dentry);
2255 goto out;
2256 }
2257 path->dentry = dentry;
2258 path->mnt = mntget(nd->path.mnt);
2259 if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW))
2260 return 1;
2261 follow_mount(path);
2262 error = 0;
2263out:
2264 terminate_walk(nd);
2265 return error;
2266}
2267
2268/**
2269 * path_mountpoint - look up a path to be umounted
2270 * @dfd: directory file descriptor to start walk from
2271 * @name: full pathname to walk
2272 * @path: pointer to container for result
2273 * @flags: lookup flags
2274 *
2275 * Look up the given name, but don't attempt to revalidate the last component.
2276 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2277 */
2278static int
2279path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags)
2280{
2281 struct file *base = NULL;
2282 struct nameidata nd;
2283 int err;
2284
2285 err = path_init(dfd, name, flags | LOOKUP_PARENT, &nd, &base);
2286 if (unlikely(err))
2287 return err;
2288
2289 current->total_link_count = 0;
2290 err = link_path_walk(name, &nd);
2291 if (err)
2292 goto out;
2293
2294 err = mountpoint_last(&nd, path);
2295 while (err > 0) {
2296 void *cookie;
2297 struct path link = *path;
2298 err = may_follow_link(&link, &nd);
2299 if (unlikely(err))
2300 break;
2301 nd.flags |= LOOKUP_PARENT;
2302 err = follow_link(&link, &nd, &cookie);
2303 if (err)
2304 break;
2305 err = mountpoint_last(&nd, path);
2306 put_link(&nd, &link, cookie);
2307 }
2308out:
2309 if (base)
2310 fput(base);
2311
2312 if (nd.root.mnt && !(nd.flags & LOOKUP_ROOT))
2313 path_put(&nd.root);
2314
2315 return err;
2316}
2317
2318static int
2319filename_mountpoint(int dfd, struct filename *s, struct path *path,
2320 unsigned int flags)
2321{
2322 int error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU);
2323 if (unlikely(error == -ECHILD))
2324 error = path_mountpoint(dfd, s->name, path, flags);
2325 if (unlikely(error == -ESTALE))
2326 error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL);
2327 if (likely(!error))
2328 audit_inode(s, path->dentry, 0);
2329 return error;
2330}
2331
2332/**
2333 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2334 * @dfd: directory file descriptor
2335 * @name: pathname from userland
2336 * @flags: lookup flags
2337 * @path: pointer to container to hold result
2338 *
2339 * A umount is a special case for path walking. We're not actually interested
2340 * in the inode in this situation, and ESTALE errors can be a problem. We
2341 * simply want track down the dentry and vfsmount attached at the mountpoint
2342 * and avoid revalidating the last component.
2343 *
2344 * Returns 0 and populates "path" on success.
2345 */
2346int
2347user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2348 struct path *path)
2349{
2350 struct filename *s = getname(name);
2351 int error;
2352 if (IS_ERR(s))
2353 return PTR_ERR(s);
2354 error = filename_mountpoint(dfd, s, path, flags);
2355 putname(s);
2356 return error;
2357}
2358
2359int
2360kern_path_mountpoint(int dfd, const char *name, struct path *path,
2361 unsigned int flags)
2362{
2363 struct filename s = {.name = name};
2364 return filename_mountpoint(dfd, &s, path, flags);
2365}
2366EXPORT_SYMBOL(kern_path_mountpoint);
2367
2368/*
2369 * It's inline, so penalty for filesystems that don't use sticky bit is
2370 * minimal.
2371 */
2372static inline int check_sticky(struct inode *dir, struct inode *inode)
2373{
2374 kuid_t fsuid = current_fsuid();
2375
2376 if (!(dir->i_mode & S_ISVTX))
2377 return 0;
2378 if (uid_eq(inode->i_uid, fsuid))
2379 return 0;
2380 if (uid_eq(dir->i_uid, fsuid))
2381 return 0;
2382 return !inode_capable(inode, CAP_FOWNER);
2383}
2384
2385/*
2386 * Check whether we can remove a link victim from directory dir, check
2387 * whether the type of victim is right.
2388 * 1. We can't do it if dir is read-only (done in permission())
2389 * 2. We should have write and exec permissions on dir
2390 * 3. We can't remove anything from append-only dir
2391 * 4. We can't do anything with immutable dir (done in permission())
2392 * 5. If the sticky bit on dir is set we should either
2393 * a. be owner of dir, or
2394 * b. be owner of victim, or
2395 * c. have CAP_FOWNER capability
2396 * 6. If the victim is append-only or immutable we can't do antyhing with
2397 * links pointing to it.
2398 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2399 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2400 * 9. We can't remove a root or mountpoint.
2401 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2402 * nfs_async_unlink().
2403 */
2404static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2405{
2406 struct inode *inode = victim->d_inode;
2407 int error;
2408
2409 if (d_is_negative(victim))
2410 return -ENOENT;
2411 BUG_ON(!inode);
2412
2413 BUG_ON(victim->d_parent->d_inode != dir);
2414 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2415
2416 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2417 if (error)
2418 return error;
2419 if (IS_APPEND(dir))
2420 return -EPERM;
2421
2422 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2423 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2424 return -EPERM;
2425 if (isdir) {
2426 if (!d_is_dir(victim))
2427 return -ENOTDIR;
2428 if (IS_ROOT(victim))
2429 return -EBUSY;
2430 } else if (d_is_dir(victim))
2431 return -EISDIR;
2432 if (IS_DEADDIR(dir))
2433 return -ENOENT;
2434 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2435 return -EBUSY;
2436 return 0;
2437}
2438
2439/* Check whether we can create an object with dentry child in directory
2440 * dir.
2441 * 1. We can't do it if child already exists (open has special treatment for
2442 * this case, but since we are inlined it's OK)
2443 * 2. We can't do it if dir is read-only (done in permission())
2444 * 3. We should have write and exec permissions on dir
2445 * 4. We can't do it if dir is immutable (done in permission())
2446 */
2447static inline int may_create(struct inode *dir, struct dentry *child)
2448{
2449 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2450 if (child->d_inode)
2451 return -EEXIST;
2452 if (IS_DEADDIR(dir))
2453 return -ENOENT;
2454 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2455}
2456
2457/*
2458 * p1 and p2 should be directories on the same fs.
2459 */
2460struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2461{
2462 struct dentry *p;
2463
2464 if (p1 == p2) {
2465 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2466 return NULL;
2467 }
2468
2469 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2470
2471 p = d_ancestor(p2, p1);
2472 if (p) {
2473 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2474 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2475 return p;
2476 }
2477
2478 p = d_ancestor(p1, p2);
2479 if (p) {
2480 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2481 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2482 return p;
2483 }
2484
2485 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2486 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2487 return NULL;
2488}
2489EXPORT_SYMBOL(lock_rename);
2490
2491void unlock_rename(struct dentry *p1, struct dentry *p2)
2492{
2493 mutex_unlock(&p1->d_inode->i_mutex);
2494 if (p1 != p2) {
2495 mutex_unlock(&p2->d_inode->i_mutex);
2496 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2497 }
2498}
2499EXPORT_SYMBOL(unlock_rename);
2500
2501int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2502 bool want_excl)
2503{
2504 int error = may_create(dir, dentry);
2505 if (error)
2506 return error;
2507
2508 if (!dir->i_op->create)
2509 return -EACCES; /* shouldn't it be ENOSYS? */
2510 mode &= S_IALLUGO;
2511 mode |= S_IFREG;
2512 error = security_inode_create(dir, dentry, mode);
2513 if (error)
2514 return error;
2515 error = dir->i_op->create(dir, dentry, mode, want_excl);
2516 if (!error)
2517 fsnotify_create(dir, dentry);
2518 return error;
2519}
2520EXPORT_SYMBOL(vfs_create);
2521
2522static int may_open(struct path *path, int acc_mode, int flag)
2523{
2524 struct dentry *dentry = path->dentry;
2525 struct inode *inode = dentry->d_inode;
2526 int error;
2527
2528 /* O_PATH? */
2529 if (!acc_mode)
2530 return 0;
2531
2532 if (!inode)
2533 return -ENOENT;
2534
2535 switch (inode->i_mode & S_IFMT) {
2536 case S_IFLNK:
2537 return -ELOOP;
2538 case S_IFDIR:
2539 if (acc_mode & MAY_WRITE)
2540 return -EISDIR;
2541 break;
2542 case S_IFBLK:
2543 case S_IFCHR:
2544 if (path->mnt->mnt_flags & MNT_NODEV)
2545 return -EACCES;
2546 /*FALLTHRU*/
2547 case S_IFIFO:
2548 case S_IFSOCK:
2549 flag &= ~O_TRUNC;
2550 break;
2551 }
2552
2553 error = inode_permission(inode, acc_mode);
2554 if (error)
2555 return error;
2556
2557 /*
2558 * An append-only file must be opened in append mode for writing.
2559 */
2560 if (IS_APPEND(inode)) {
2561 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2562 return -EPERM;
2563 if (flag & O_TRUNC)
2564 return -EPERM;
2565 }
2566
2567 /* O_NOATIME can only be set by the owner or superuser */
2568 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2569 return -EPERM;
2570
2571 return 0;
2572}
2573
2574static int handle_truncate(struct file *filp)
2575{
2576 struct path *path = &filp->f_path;
2577 struct inode *inode = path->dentry->d_inode;
2578 int error = get_write_access(inode);
2579 if (error)
2580 return error;
2581 /*
2582 * Refuse to truncate files with mandatory locks held on them.
2583 */
2584 error = locks_verify_locked(filp);
2585 if (!error)
2586 error = security_path_truncate(path);
2587 if (!error) {
2588 error = do_truncate(path->dentry, 0,
2589 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2590 filp);
2591 }
2592 put_write_access(inode);
2593 return error;
2594}
2595
2596static inline int open_to_namei_flags(int flag)
2597{
2598 if ((flag & O_ACCMODE) == 3)
2599 flag--;
2600 return flag;
2601}
2602
2603static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2604{
2605 int error = security_path_mknod(dir, dentry, mode, 0);
2606 if (error)
2607 return error;
2608
2609 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2610 if (error)
2611 return error;
2612
2613 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2614}
2615
2616/*
2617 * Attempt to atomically look up, create and open a file from a negative
2618 * dentry.
2619 *
2620 * Returns 0 if successful. The file will have been created and attached to
2621 * @file by the filesystem calling finish_open().
2622 *
2623 * Returns 1 if the file was looked up only or didn't need creating. The
2624 * caller will need to perform the open themselves. @path will have been
2625 * updated to point to the new dentry. This may be negative.
2626 *
2627 * Returns an error code otherwise.
2628 */
2629static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2630 struct path *path, struct file *file,
2631 const struct open_flags *op,
2632 bool got_write, bool need_lookup,
2633 int *opened)
2634{
2635 struct inode *dir = nd->path.dentry->d_inode;
2636 unsigned open_flag = open_to_namei_flags(op->open_flag);
2637 umode_t mode;
2638 int error;
2639 int acc_mode;
2640 int create_error = 0;
2641 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2642 bool excl;
2643
2644 BUG_ON(dentry->d_inode);
2645
2646 /* Don't create child dentry for a dead directory. */
2647 if (unlikely(IS_DEADDIR(dir))) {
2648 error = -ENOENT;
2649 goto out;
2650 }
2651
2652 mode = op->mode;
2653 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2654 mode &= ~current_umask();
2655
2656 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2657 if (excl)
2658 open_flag &= ~O_TRUNC;
2659
2660 /*
2661 * Checking write permission is tricky, bacuse we don't know if we are
2662 * going to actually need it: O_CREAT opens should work as long as the
2663 * file exists. But checking existence breaks atomicity. The trick is
2664 * to check access and if not granted clear O_CREAT from the flags.
2665 *
2666 * Another problem is returing the "right" error value (e.g. for an
2667 * O_EXCL open we want to return EEXIST not EROFS).
2668 */
2669 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2670 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2671 if (!(open_flag & O_CREAT)) {
2672 /*
2673 * No O_CREATE -> atomicity not a requirement -> fall
2674 * back to lookup + open
2675 */
2676 goto no_open;
2677 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2678 /* Fall back and fail with the right error */
2679 create_error = -EROFS;
2680 goto no_open;
2681 } else {
2682 /* No side effects, safe to clear O_CREAT */
2683 create_error = -EROFS;
2684 open_flag &= ~O_CREAT;
2685 }
2686 }
2687
2688 if (open_flag & O_CREAT) {
2689 error = may_o_create(&nd->path, dentry, mode);
2690 if (error) {
2691 create_error = error;
2692 if (open_flag & O_EXCL)
2693 goto no_open;
2694 open_flag &= ~O_CREAT;
2695 }
2696 }
2697
2698 if (nd->flags & LOOKUP_DIRECTORY)
2699 open_flag |= O_DIRECTORY;
2700
2701 file->f_path.dentry = DENTRY_NOT_SET;
2702 file->f_path.mnt = nd->path.mnt;
2703 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2704 opened);
2705 if (error < 0) {
2706 if (create_error && error == -ENOENT)
2707 error = create_error;
2708 goto out;
2709 }
2710
2711 if (error) { /* returned 1, that is */
2712 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2713 error = -EIO;
2714 goto out;
2715 }
2716 if (file->f_path.dentry) {
2717 dput(dentry);
2718 dentry = file->f_path.dentry;
2719 }
2720 if (*opened & FILE_CREATED)
2721 fsnotify_create(dir, dentry);
2722 if (!dentry->d_inode) {
2723 WARN_ON(*opened & FILE_CREATED);
2724 if (create_error) {
2725 error = create_error;
2726 goto out;
2727 }
2728 } else {
2729 if (excl && !(*opened & FILE_CREATED)) {
2730 error = -EEXIST;
2731 goto out;
2732 }
2733 }
2734 goto looked_up;
2735 }
2736
2737 /*
2738 * We didn't have the inode before the open, so check open permission
2739 * here.
2740 */
2741 acc_mode = op->acc_mode;
2742 if (*opened & FILE_CREATED) {
2743 WARN_ON(!(open_flag & O_CREAT));
2744 fsnotify_create(dir, dentry);
2745 acc_mode = MAY_OPEN;
2746 }
2747 error = may_open(&file->f_path, acc_mode, open_flag);
2748 if (error)
2749 fput(file);
2750
2751out:
2752 dput(dentry);
2753 return error;
2754
2755no_open:
2756 if (need_lookup) {
2757 dentry = lookup_real(dir, dentry, nd->flags);
2758 if (IS_ERR(dentry))
2759 return PTR_ERR(dentry);
2760
2761 if (create_error) {
2762 int open_flag = op->open_flag;
2763
2764 error = create_error;
2765 if ((open_flag & O_EXCL)) {
2766 if (!dentry->d_inode)
2767 goto out;
2768 } else if (!dentry->d_inode) {
2769 goto out;
2770 } else if ((open_flag & O_TRUNC) &&
2771 S_ISREG(dentry->d_inode->i_mode)) {
2772 goto out;
2773 }
2774 /* will fail later, go on to get the right error */
2775 }
2776 }
2777looked_up:
2778 path->dentry = dentry;
2779 path->mnt = nd->path.mnt;
2780 return 1;
2781}
2782
2783/*
2784 * Look up and maybe create and open the last component.
2785 *
2786 * Must be called with i_mutex held on parent.
2787 *
2788 * Returns 0 if the file was successfully atomically created (if necessary) and
2789 * opened. In this case the file will be returned attached to @file.
2790 *
2791 * Returns 1 if the file was not completely opened at this time, though lookups
2792 * and creations will have been performed and the dentry returned in @path will
2793 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2794 * specified then a negative dentry may be returned.
2795 *
2796 * An error code is returned otherwise.
2797 *
2798 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2799 * cleared otherwise prior to returning.
2800 */
2801static int lookup_open(struct nameidata *nd, struct path *path,
2802 struct file *file,
2803 const struct open_flags *op,
2804 bool got_write, int *opened)
2805{
2806 struct dentry *dir = nd->path.dentry;
2807 struct inode *dir_inode = dir->d_inode;
2808 struct dentry *dentry;
2809 int error;
2810 bool need_lookup;
2811
2812 *opened &= ~FILE_CREATED;
2813 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2814 if (IS_ERR(dentry))
2815 return PTR_ERR(dentry);
2816
2817 /* Cached positive dentry: will open in f_op->open */
2818 if (!need_lookup && dentry->d_inode)
2819 goto out_no_open;
2820
2821 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2822 return atomic_open(nd, dentry, path, file, op, got_write,
2823 need_lookup, opened);
2824 }
2825
2826 if (need_lookup) {
2827 BUG_ON(dentry->d_inode);
2828
2829 dentry = lookup_real(dir_inode, dentry, nd->flags);
2830 if (IS_ERR(dentry))
2831 return PTR_ERR(dentry);
2832 }
2833
2834 /* Negative dentry, just create the file */
2835 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2836 umode_t mode = op->mode;
2837 if (!IS_POSIXACL(dir->d_inode))
2838 mode &= ~current_umask();
2839 /*
2840 * This write is needed to ensure that a
2841 * rw->ro transition does not occur between
2842 * the time when the file is created and when
2843 * a permanent write count is taken through
2844 * the 'struct file' in finish_open().
2845 */
2846 if (!got_write) {
2847 error = -EROFS;
2848 goto out_dput;
2849 }
2850 *opened |= FILE_CREATED;
2851 error = security_path_mknod(&nd->path, dentry, mode, 0);
2852 if (error)
2853 goto out_dput;
2854 error = vfs_create(dir->d_inode, dentry, mode,
2855 nd->flags & LOOKUP_EXCL);
2856 if (error)
2857 goto out_dput;
2858 }
2859out_no_open:
2860 path->dentry = dentry;
2861 path->mnt = nd->path.mnt;
2862 return 1;
2863
2864out_dput:
2865 dput(dentry);
2866 return error;
2867}
2868
2869/*
2870 * Handle the last step of open()
2871 */
2872static int do_last(struct nameidata *nd, struct path *path,
2873 struct file *file, const struct open_flags *op,
2874 int *opened, struct filename *name)
2875{
2876 struct dentry *dir = nd->path.dentry;
2877 int open_flag = op->open_flag;
2878 bool will_truncate = (open_flag & O_TRUNC) != 0;
2879 bool got_write = false;
2880 int acc_mode = op->acc_mode;
2881 struct inode *inode;
2882 bool symlink_ok = false;
2883 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2884 bool retried = false;
2885 int error;
2886
2887 nd->flags &= ~LOOKUP_PARENT;
2888 nd->flags |= op->intent;
2889
2890 if (nd->last_type != LAST_NORM) {
2891 error = handle_dots(nd, nd->last_type);
2892 if (error)
2893 return error;
2894 goto finish_open;
2895 }
2896
2897 if (!(open_flag & O_CREAT)) {
2898 if (nd->last.name[nd->last.len])
2899 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2900 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2901 symlink_ok = true;
2902 /* we _can_ be in RCU mode here */
2903 error = lookup_fast(nd, path, &inode);
2904 if (likely(!error))
2905 goto finish_lookup;
2906
2907 if (error < 0)
2908 goto out;
2909
2910 BUG_ON(nd->inode != dir->d_inode);
2911 } else {
2912 /* create side of things */
2913 /*
2914 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2915 * has been cleared when we got to the last component we are
2916 * about to look up
2917 */
2918 error = complete_walk(nd);
2919 if (error)
2920 return error;
2921
2922 audit_inode(name, dir, LOOKUP_PARENT);
2923 error = -EISDIR;
2924 /* trailing slashes? */
2925 if (nd->last.name[nd->last.len])
2926 goto out;
2927 }
2928
2929retry_lookup:
2930 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2931 error = mnt_want_write(nd->path.mnt);
2932 if (!error)
2933 got_write = true;
2934 /*
2935 * do _not_ fail yet - we might not need that or fail with
2936 * a different error; let lookup_open() decide; we'll be
2937 * dropping this one anyway.
2938 */
2939 }
2940 mutex_lock(&dir->d_inode->i_mutex);
2941 error = lookup_open(nd, path, file, op, got_write, opened);
2942 mutex_unlock(&dir->d_inode->i_mutex);
2943
2944 if (error <= 0) {
2945 if (error)
2946 goto out;
2947
2948 if ((*opened & FILE_CREATED) ||
2949 !S_ISREG(file_inode(file)->i_mode))
2950 will_truncate = false;
2951
2952 audit_inode(name, file->f_path.dentry, 0);
2953 goto opened;
2954 }
2955
2956 if (*opened & FILE_CREATED) {
2957 /* Don't check for write permission, don't truncate */
2958 open_flag &= ~O_TRUNC;
2959 will_truncate = false;
2960 acc_mode = MAY_OPEN;
2961 path_to_nameidata(path, nd);
2962 goto finish_open_created;
2963 }
2964
2965 /*
2966 * create/update audit record if it already exists.
2967 */
2968 if (d_is_positive(path->dentry))
2969 audit_inode(name, path->dentry, 0);
2970
2971 /*
2972 * If atomic_open() acquired write access it is dropped now due to
2973 * possible mount and symlink following (this might be optimized away if
2974 * necessary...)
2975 */
2976 if (got_write) {
2977 mnt_drop_write(nd->path.mnt);
2978 got_write = false;
2979 }
2980
2981 error = -EEXIST;
2982 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
2983 goto exit_dput;
2984
2985 error = follow_managed(path, nd->flags);
2986 if (error < 0)
2987 goto exit_dput;
2988
2989 if (error)
2990 nd->flags |= LOOKUP_JUMPED;
2991
2992 BUG_ON(nd->flags & LOOKUP_RCU);
2993 inode = path->dentry->d_inode;
2994finish_lookup:
2995 /* we _can_ be in RCU mode here */
2996 error = -ENOENT;
2997 if (!inode || d_is_negative(path->dentry)) {
2998 path_to_nameidata(path, nd);
2999 goto out;
3000 }
3001
3002 if (should_follow_link(path->dentry, !symlink_ok)) {
3003 if (nd->flags & LOOKUP_RCU) {
3004 if (unlikely(unlazy_walk(nd, path->dentry))) {
3005 error = -ECHILD;
3006 goto out;
3007 }
3008 }
3009 BUG_ON(inode != path->dentry->d_inode);
3010 return 1;
3011 }
3012
3013 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
3014 path_to_nameidata(path, nd);
3015 } else {
3016 save_parent.dentry = nd->path.dentry;
3017 save_parent.mnt = mntget(path->mnt);
3018 nd->path.dentry = path->dentry;
3019
3020 }
3021 nd->inode = inode;
3022 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3023finish_open:
3024 error = complete_walk(nd);
3025 if (error) {
3026 path_put(&save_parent);
3027 return error;
3028 }
3029 audit_inode(name, nd->path.dentry, 0);
3030 error = -EISDIR;
3031 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3032 goto out;
3033 error = -ENOTDIR;
3034 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3035 goto out;
3036 if (!S_ISREG(nd->inode->i_mode))
3037 will_truncate = false;
3038
3039 if (will_truncate) {
3040 error = mnt_want_write(nd->path.mnt);
3041 if (error)
3042 goto out;
3043 got_write = true;
3044 }
3045finish_open_created:
3046 error = may_open(&nd->path, acc_mode, open_flag);
3047 if (error)
3048 goto out;
3049 file->f_path.mnt = nd->path.mnt;
3050 error = finish_open(file, nd->path.dentry, NULL, opened);
3051 if (error) {
3052 if (error == -EOPENSTALE)
3053 goto stale_open;
3054 goto out;
3055 }
3056opened:
3057 error = open_check_o_direct(file);
3058 if (error)
3059 goto exit_fput;
3060 error = ima_file_check(file, op->acc_mode);
3061 if (error)
3062 goto exit_fput;
3063
3064 if (will_truncate) {
3065 error = handle_truncate(file);
3066 if (error)
3067 goto exit_fput;
3068 }
3069out:
3070 if (got_write)
3071 mnt_drop_write(nd->path.mnt);
3072 path_put(&save_parent);
3073 terminate_walk(nd);
3074 return error;
3075
3076exit_dput:
3077 path_put_conditional(path, nd);
3078 goto out;
3079exit_fput:
3080 fput(file);
3081 goto out;
3082
3083stale_open:
3084 /* If no saved parent or already retried then can't retry */
3085 if (!save_parent.dentry || retried)
3086 goto out;
3087
3088 BUG_ON(save_parent.dentry != dir);
3089 path_put(&nd->path);
3090 nd->path = save_parent;
3091 nd->inode = dir->d_inode;
3092 save_parent.mnt = NULL;
3093 save_parent.dentry = NULL;
3094 if (got_write) {
3095 mnt_drop_write(nd->path.mnt);
3096 got_write = false;
3097 }
3098 retried = true;
3099 goto retry_lookup;
3100}
3101
3102static int do_tmpfile(int dfd, struct filename *pathname,
3103 struct nameidata *nd, int flags,
3104 const struct open_flags *op,
3105 struct file *file, int *opened)
3106{
3107 static const struct qstr name = QSTR_INIT("/", 1);
3108 struct dentry *dentry, *child;
3109 struct inode *dir;
3110 int error = path_lookupat(dfd, pathname->name,
3111 flags | LOOKUP_DIRECTORY, nd);
3112 if (unlikely(error))
3113 return error;
3114 error = mnt_want_write(nd->path.mnt);
3115 if (unlikely(error))
3116 goto out;
3117 /* we want directory to be writable */
3118 error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3119 if (error)
3120 goto out2;
3121 dentry = nd->path.dentry;
3122 dir = dentry->d_inode;
3123 if (!dir->i_op->tmpfile) {
3124 error = -EOPNOTSUPP;
3125 goto out2;
3126 }
3127 child = d_alloc(dentry, &name);
3128 if (unlikely(!child)) {
3129 error = -ENOMEM;
3130 goto out2;
3131 }
3132 nd->flags &= ~LOOKUP_DIRECTORY;
3133 nd->flags |= op->intent;
3134 dput(nd->path.dentry);
3135 nd->path.dentry = child;
3136 error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3137 if (error)
3138 goto out2;
3139 audit_inode(pathname, nd->path.dentry, 0);
3140 error = may_open(&nd->path, op->acc_mode, op->open_flag);
3141 if (error)
3142 goto out2;
3143 file->f_path.mnt = nd->path.mnt;
3144 error = finish_open(file, nd->path.dentry, NULL, opened);
3145 if (error)
3146 goto out2;
3147 error = open_check_o_direct(file);
3148 if (error) {
3149 fput(file);
3150 } else if (!(op->open_flag & O_EXCL)) {
3151 struct inode *inode = file_inode(file);
3152 spin_lock(&inode->i_lock);
3153 inode->i_state |= I_LINKABLE;
3154 spin_unlock(&inode->i_lock);
3155 }
3156out2:
3157 mnt_drop_write(nd->path.mnt);
3158out:
3159 path_put(&nd->path);
3160 return error;
3161}
3162
3163static struct file *path_openat(int dfd, struct filename *pathname,
3164 struct nameidata *nd, const struct open_flags *op, int flags)
3165{
3166 struct file *base = NULL;
3167 struct file *file;
3168 struct path path;
3169 int opened = 0;
3170 int error;
3171
3172 file = get_empty_filp();
3173 if (IS_ERR(file))
3174 return file;
3175
3176 file->f_flags = op->open_flag;
3177
3178 if (unlikely(file->f_flags & __O_TMPFILE)) {
3179 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3180 goto out;
3181 }
3182
3183 error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base);
3184 if (unlikely(error))
3185 goto out;
3186
3187 current->total_link_count = 0;
3188 error = link_path_walk(pathname->name, nd);
3189 if (unlikely(error))
3190 goto out;
3191
3192 error = do_last(nd, &path, file, op, &opened, pathname);
3193 while (unlikely(error > 0)) { /* trailing symlink */
3194 struct path link = path;
3195 void *cookie;
3196 if (!(nd->flags & LOOKUP_FOLLOW)) {
3197 path_put_conditional(&path, nd);
3198 path_put(&nd->path);
3199 error = -ELOOP;
3200 break;
3201 }
3202 error = may_follow_link(&link, nd);
3203 if (unlikely(error))
3204 break;
3205 nd->flags |= LOOKUP_PARENT;
3206 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3207 error = follow_link(&link, nd, &cookie);
3208 if (unlikely(error))
3209 break;
3210 error = do_last(nd, &path, file, op, &opened, pathname);
3211 put_link(nd, &link, cookie);
3212 }
3213out:
3214 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
3215 path_put(&nd->root);
3216 if (base)
3217 fput(base);
3218 if (!(opened & FILE_OPENED)) {
3219 BUG_ON(!error);
3220 put_filp(file);
3221 }
3222 if (unlikely(error)) {
3223 if (error == -EOPENSTALE) {
3224 if (flags & LOOKUP_RCU)
3225 error = -ECHILD;
3226 else
3227 error = -ESTALE;
3228 }
3229 file = ERR_PTR(error);
3230 }
3231 return file;
3232}
3233
3234struct file *do_filp_open(int dfd, struct filename *pathname,
3235 const struct open_flags *op)
3236{
3237 struct nameidata nd;
3238 int flags = op->lookup_flags;
3239 struct file *filp;
3240
3241 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3242 if (unlikely(filp == ERR_PTR(-ECHILD)))
3243 filp = path_openat(dfd, pathname, &nd, op, flags);
3244 if (unlikely(filp == ERR_PTR(-ESTALE)))
3245 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3246 return filp;
3247}
3248
3249struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3250 const char *name, const struct open_flags *op)
3251{
3252 struct nameidata nd;
3253 struct file *file;
3254 struct filename filename = { .name = name };
3255 int flags = op->lookup_flags | LOOKUP_ROOT;
3256
3257 nd.root.mnt = mnt;
3258 nd.root.dentry = dentry;
3259
3260 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3261 return ERR_PTR(-ELOOP);
3262
3263 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU);
3264 if (unlikely(file == ERR_PTR(-ECHILD)))
3265 file = path_openat(-1, &filename, &nd, op, flags);
3266 if (unlikely(file == ERR_PTR(-ESTALE)))
3267 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL);
3268 return file;
3269}
3270
3271struct dentry *kern_path_create(int dfd, const char *pathname,
3272 struct path *path, unsigned int lookup_flags)
3273{
3274 struct dentry *dentry = ERR_PTR(-EEXIST);
3275 struct nameidata nd;
3276 int err2;
3277 int error;
3278 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3279
3280 /*
3281 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3282 * other flags passed in are ignored!
3283 */
3284 lookup_flags &= LOOKUP_REVAL;
3285
3286 error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd);
3287 if (error)
3288 return ERR_PTR(error);
3289
3290 /*
3291 * Yucky last component or no last component at all?
3292 * (foo/., foo/.., /////)
3293 */
3294 if (nd.last_type != LAST_NORM)
3295 goto out;
3296 nd.flags &= ~LOOKUP_PARENT;
3297 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3298
3299 /* don't fail immediately if it's r/o, at least try to report other errors */
3300 err2 = mnt_want_write(nd.path.mnt);
3301 /*
3302 * Do the final lookup.
3303 */
3304 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3305 dentry = lookup_hash(&nd);
3306 if (IS_ERR(dentry))
3307 goto unlock;
3308
3309 error = -EEXIST;
3310 if (d_is_positive(dentry))
3311 goto fail;
3312
3313 /*
3314 * Special case - lookup gave negative, but... we had foo/bar/
3315 * From the vfs_mknod() POV we just have a negative dentry -
3316 * all is fine. Let's be bastards - you had / on the end, you've
3317 * been asking for (non-existent) directory. -ENOENT for you.
3318 */
3319 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3320 error = -ENOENT;
3321 goto fail;
3322 }
3323 if (unlikely(err2)) {
3324 error = err2;
3325 goto fail;
3326 }
3327 *path = nd.path;
3328 return dentry;
3329fail:
3330 dput(dentry);
3331 dentry = ERR_PTR(error);
3332unlock:
3333 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3334 if (!err2)
3335 mnt_drop_write(nd.path.mnt);
3336out:
3337 path_put(&nd.path);
3338 return dentry;
3339}
3340EXPORT_SYMBOL(kern_path_create);
3341
3342void done_path_create(struct path *path, struct dentry *dentry)
3343{
3344 dput(dentry);
3345 mutex_unlock(&path->dentry->d_inode->i_mutex);
3346 mnt_drop_write(path->mnt);
3347 path_put(path);
3348}
3349EXPORT_SYMBOL(done_path_create);
3350
3351struct dentry *user_path_create(int dfd, const char __user *pathname,
3352 struct path *path, unsigned int lookup_flags)
3353{
3354 struct filename *tmp = getname(pathname);
3355 struct dentry *res;
3356 if (IS_ERR(tmp))
3357 return ERR_CAST(tmp);
3358 res = kern_path_create(dfd, tmp->name, path, lookup_flags);
3359 putname(tmp);
3360 return res;
3361}
3362EXPORT_SYMBOL(user_path_create);
3363
3364int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3365{
3366 int error = may_create(dir, dentry);
3367
3368 if (error)
3369 return error;
3370
3371 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3372 return -EPERM;
3373
3374 if (!dir->i_op->mknod)
3375 return -EPERM;
3376
3377 error = devcgroup_inode_mknod(mode, dev);
3378 if (error)
3379 return error;
3380
3381 error = security_inode_mknod(dir, dentry, mode, dev);
3382 if (error)
3383 return error;
3384
3385 error = dir->i_op->mknod(dir, dentry, mode, dev);
3386 if (!error)
3387 fsnotify_create(dir, dentry);
3388 return error;
3389}
3390EXPORT_SYMBOL(vfs_mknod);
3391
3392static int may_mknod(umode_t mode)
3393{
3394 switch (mode & S_IFMT) {
3395 case S_IFREG:
3396 case S_IFCHR:
3397 case S_IFBLK:
3398 case S_IFIFO:
3399 case S_IFSOCK:
3400 case 0: /* zero mode translates to S_IFREG */
3401 return 0;
3402 case S_IFDIR:
3403 return -EPERM;
3404 default:
3405 return -EINVAL;
3406 }
3407}
3408
3409SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3410 unsigned, dev)
3411{
3412 struct dentry *dentry;
3413 struct path path;
3414 int error;
3415 unsigned int lookup_flags = 0;
3416
3417 error = may_mknod(mode);
3418 if (error)
3419 return error;
3420retry:
3421 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3422 if (IS_ERR(dentry))
3423 return PTR_ERR(dentry);
3424
3425 if (!IS_POSIXACL(path.dentry->d_inode))
3426 mode &= ~current_umask();
3427 error = security_path_mknod(&path, dentry, mode, dev);
3428 if (error)
3429 goto out;
3430 switch (mode & S_IFMT) {
3431 case 0: case S_IFREG:
3432 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3433 break;
3434 case S_IFCHR: case S_IFBLK:
3435 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3436 new_decode_dev(dev));
3437 break;
3438 case S_IFIFO: case S_IFSOCK:
3439 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3440 break;
3441 }
3442out:
3443 done_path_create(&path, dentry);
3444 if (retry_estale(error, lookup_flags)) {
3445 lookup_flags |= LOOKUP_REVAL;
3446 goto retry;
3447 }
3448 return error;
3449}
3450
3451SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3452{
3453 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3454}
3455
3456int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3457{
3458 int error = may_create(dir, dentry);
3459 unsigned max_links = dir->i_sb->s_max_links;
3460
3461 if (error)
3462 return error;
3463
3464 if (!dir->i_op->mkdir)
3465 return -EPERM;
3466
3467 mode &= (S_IRWXUGO|S_ISVTX);
3468 error = security_inode_mkdir(dir, dentry, mode);
3469 if (error)
3470 return error;
3471
3472 if (max_links && dir->i_nlink >= max_links)
3473 return -EMLINK;
3474
3475 error = dir->i_op->mkdir(dir, dentry, mode);
3476 if (!error)
3477 fsnotify_mkdir(dir, dentry);
3478 return error;
3479}
3480EXPORT_SYMBOL(vfs_mkdir);
3481
3482SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3483{
3484 struct dentry *dentry;
3485 struct path path;
3486 int error;
3487 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3488
3489retry:
3490 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3491 if (IS_ERR(dentry))
3492 return PTR_ERR(dentry);
3493
3494 if (!IS_POSIXACL(path.dentry->d_inode))
3495 mode &= ~current_umask();
3496 error = security_path_mkdir(&path, dentry, mode);
3497 if (!error)
3498 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3499 done_path_create(&path, dentry);
3500 if (retry_estale(error, lookup_flags)) {
3501 lookup_flags |= LOOKUP_REVAL;
3502 goto retry;
3503 }
3504 return error;
3505}
3506
3507SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3508{
3509 return sys_mkdirat(AT_FDCWD, pathname, mode);
3510}
3511
3512/*
3513 * The dentry_unhash() helper will try to drop the dentry early: we
3514 * should have a usage count of 1 if we're the only user of this
3515 * dentry, and if that is true (possibly after pruning the dcache),
3516 * then we drop the dentry now.
3517 *
3518 * A low-level filesystem can, if it choses, legally
3519 * do a
3520 *
3521 * if (!d_unhashed(dentry))
3522 * return -EBUSY;
3523 *
3524 * if it cannot handle the case of removing a directory
3525 * that is still in use by something else..
3526 */
3527void dentry_unhash(struct dentry *dentry)
3528{
3529 shrink_dcache_parent(dentry);
3530 spin_lock(&dentry->d_lock);
3531 if (dentry->d_lockref.count == 1)
3532 __d_drop(dentry);
3533 spin_unlock(&dentry->d_lock);
3534}
3535EXPORT_SYMBOL(dentry_unhash);
3536
3537int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3538{
3539 int error = may_delete(dir, dentry, 1);
3540
3541 if (error)
3542 return error;
3543
3544 if (!dir->i_op->rmdir)
3545 return -EPERM;
3546
3547 dget(dentry);
3548 mutex_lock(&dentry->d_inode->i_mutex);
3549
3550 error = -EBUSY;
3551 if (d_mountpoint(dentry))
3552 goto out;
3553
3554 error = security_inode_rmdir(dir, dentry);
3555 if (error)
3556 goto out;
3557
3558 shrink_dcache_parent(dentry);
3559 error = dir->i_op->rmdir(dir, dentry);
3560 if (error)
3561 goto out;
3562
3563 dentry->d_inode->i_flags |= S_DEAD;
3564 dont_mount(dentry);
3565
3566out:
3567 mutex_unlock(&dentry->d_inode->i_mutex);
3568 dput(dentry);
3569 if (!error)
3570 d_delete(dentry);
3571 return error;
3572}
3573EXPORT_SYMBOL(vfs_rmdir);
3574
3575static long do_rmdir(int dfd, const char __user *pathname)
3576{
3577 int error = 0;
3578 struct filename *name;
3579 struct dentry *dentry;
3580 struct nameidata nd;
3581 unsigned int lookup_flags = 0;
3582retry:
3583 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3584 if (IS_ERR(name))
3585 return PTR_ERR(name);
3586
3587 switch(nd.last_type) {
3588 case LAST_DOTDOT:
3589 error = -ENOTEMPTY;
3590 goto exit1;
3591 case LAST_DOT:
3592 error = -EINVAL;
3593 goto exit1;
3594 case LAST_ROOT:
3595 error = -EBUSY;
3596 goto exit1;
3597 }
3598
3599 nd.flags &= ~LOOKUP_PARENT;
3600 error = mnt_want_write(nd.path.mnt);
3601 if (error)
3602 goto exit1;
3603
3604 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3605 dentry = lookup_hash(&nd);
3606 error = PTR_ERR(dentry);
3607 if (IS_ERR(dentry))
3608 goto exit2;
3609 if (!dentry->d_inode) {
3610 error = -ENOENT;
3611 goto exit3;
3612 }
3613 error = security_path_rmdir(&nd.path, dentry);
3614 if (error)
3615 goto exit3;
3616 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
3617exit3:
3618 dput(dentry);
3619exit2:
3620 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3621 mnt_drop_write(nd.path.mnt);
3622exit1:
3623 path_put(&nd.path);
3624 putname(name);
3625 if (retry_estale(error, lookup_flags)) {
3626 lookup_flags |= LOOKUP_REVAL;
3627 goto retry;
3628 }
3629 return error;
3630}
3631
3632SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3633{
3634 return do_rmdir(AT_FDCWD, pathname);
3635}
3636
3637/**
3638 * vfs_unlink - unlink a filesystem object
3639 * @dir: parent directory
3640 * @dentry: victim
3641 * @delegated_inode: returns victim inode, if the inode is delegated.
3642 *
3643 * The caller must hold dir->i_mutex.
3644 *
3645 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3646 * return a reference to the inode in delegated_inode. The caller
3647 * should then break the delegation on that inode and retry. Because
3648 * breaking a delegation may take a long time, the caller should drop
3649 * dir->i_mutex before doing so.
3650 *
3651 * Alternatively, a caller may pass NULL for delegated_inode. This may
3652 * be appropriate for callers that expect the underlying filesystem not
3653 * to be NFS exported.
3654 */
3655int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3656{
3657 struct inode *target = dentry->d_inode;
3658 int error = may_delete(dir, dentry, 0);
3659
3660 if (error)
3661 return error;
3662
3663 if (!dir->i_op->unlink)
3664 return -EPERM;
3665
3666 mutex_lock(&target->i_mutex);
3667 if (d_mountpoint(dentry))
3668 error = -EBUSY;
3669 else {
3670 error = security_inode_unlink(dir, dentry);
3671 if (!error) {
3672 error = try_break_deleg(target, delegated_inode);
3673 if (error)
3674 goto out;
3675 error = dir->i_op->unlink(dir, dentry);
3676 if (!error)
3677 dont_mount(dentry);
3678 }
3679 }
3680out:
3681 mutex_unlock(&target->i_mutex);
3682
3683 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3684 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3685 fsnotify_link_count(target);
3686 d_delete(dentry);
3687 }
3688
3689 return error;
3690}
3691EXPORT_SYMBOL(vfs_unlink);
3692
3693/*
3694 * Make sure that the actual truncation of the file will occur outside its
3695 * directory's i_mutex. Truncate can take a long time if there is a lot of
3696 * writeout happening, and we don't want to prevent access to the directory
3697 * while waiting on the I/O.
3698 */
3699static long do_unlinkat(int dfd, const char __user *pathname)
3700{
3701 int error;
3702 struct filename *name;
3703 struct dentry *dentry;
3704 struct nameidata nd;
3705 struct inode *inode = NULL;
3706 struct inode *delegated_inode = NULL;
3707 unsigned int lookup_flags = 0;
3708retry:
3709 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3710 if (IS_ERR(name))
3711 return PTR_ERR(name);
3712
3713 error = -EISDIR;
3714 if (nd.last_type != LAST_NORM)
3715 goto exit1;
3716
3717 nd.flags &= ~LOOKUP_PARENT;
3718 error = mnt_want_write(nd.path.mnt);
3719 if (error)
3720 goto exit1;
3721retry_deleg:
3722 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3723 dentry = lookup_hash(&nd);
3724 error = PTR_ERR(dentry);
3725 if (!IS_ERR(dentry)) {
3726 /* Why not before? Because we want correct error value */
3727 if (nd.last.name[nd.last.len])
3728 goto slashes;
3729 inode = dentry->d_inode;
3730 if (d_is_negative(dentry))
3731 goto slashes;
3732 ihold(inode);
3733 error = security_path_unlink(&nd.path, dentry);
3734 if (error)
3735 goto exit2;
3736 error = vfs_unlink(nd.path.dentry->d_inode, dentry, &delegated_inode);
3737exit2:
3738 dput(dentry);
3739 }
3740 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3741 if (inode)
3742 iput(inode); /* truncate the inode here */
3743 inode = NULL;
3744 if (delegated_inode) {
3745 error = break_deleg_wait(&delegated_inode);
3746 if (!error)
3747 goto retry_deleg;
3748 }
3749 mnt_drop_write(nd.path.mnt);
3750exit1:
3751 path_put(&nd.path);
3752 putname(name);
3753 if (retry_estale(error, lookup_flags)) {
3754 lookup_flags |= LOOKUP_REVAL;
3755 inode = NULL;
3756 goto retry;
3757 }
3758 return error;
3759
3760slashes:
3761 if (d_is_negative(dentry))
3762 error = -ENOENT;
3763 else if (d_is_dir(dentry))
3764 error = -EISDIR;
3765 else
3766 error = -ENOTDIR;
3767 goto exit2;
3768}
3769
3770SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3771{
3772 if ((flag & ~AT_REMOVEDIR) != 0)
3773 return -EINVAL;
3774
3775 if (flag & AT_REMOVEDIR)
3776 return do_rmdir(dfd, pathname);
3777
3778 return do_unlinkat(dfd, pathname);
3779}
3780
3781SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3782{
3783 return do_unlinkat(AT_FDCWD, pathname);
3784}
3785
3786int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3787{
3788 int error = may_create(dir, dentry);
3789
3790 if (error)
3791 return error;
3792
3793 if (!dir->i_op->symlink)
3794 return -EPERM;
3795
3796 error = security_inode_symlink(dir, dentry, oldname);
3797 if (error)
3798 return error;
3799
3800 error = dir->i_op->symlink(dir, dentry, oldname);
3801 if (!error)
3802 fsnotify_create(dir, dentry);
3803 return error;
3804}
3805EXPORT_SYMBOL(vfs_symlink);
3806
3807SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3808 int, newdfd, const char __user *, newname)
3809{
3810 int error;
3811 struct filename *from;
3812 struct dentry *dentry;
3813 struct path path;
3814 unsigned int lookup_flags = 0;
3815
3816 from = getname(oldname);
3817 if (IS_ERR(from))
3818 return PTR_ERR(from);
3819retry:
3820 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3821 error = PTR_ERR(dentry);
3822 if (IS_ERR(dentry))
3823 goto out_putname;
3824
3825 error = security_path_symlink(&path, dentry, from->name);
3826 if (!error)
3827 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3828 done_path_create(&path, dentry);
3829 if (retry_estale(error, lookup_flags)) {
3830 lookup_flags |= LOOKUP_REVAL;
3831 goto retry;
3832 }
3833out_putname:
3834 putname(from);
3835 return error;
3836}
3837
3838SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3839{
3840 return sys_symlinkat(oldname, AT_FDCWD, newname);
3841}
3842
3843/**
3844 * vfs_link - create a new link
3845 * @old_dentry: object to be linked
3846 * @dir: new parent
3847 * @new_dentry: where to create the new link
3848 * @delegated_inode: returns inode needing a delegation break
3849 *
3850 * The caller must hold dir->i_mutex
3851 *
3852 * If vfs_link discovers a delegation on the to-be-linked file in need
3853 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3854 * inode in delegated_inode. The caller should then break the delegation
3855 * and retry. Because breaking a delegation may take a long time, the
3856 * caller should drop the i_mutex before doing so.
3857 *
3858 * Alternatively, a caller may pass NULL for delegated_inode. This may
3859 * be appropriate for callers that expect the underlying filesystem not
3860 * to be NFS exported.
3861 */
3862int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
3863{
3864 struct inode *inode = old_dentry->d_inode;
3865 unsigned max_links = dir->i_sb->s_max_links;
3866 int error;
3867
3868 if (!inode)
3869 return -ENOENT;
3870
3871 error = may_create(dir, new_dentry);
3872 if (error)
3873 return error;
3874
3875 if (dir->i_sb != inode->i_sb)
3876 return -EXDEV;
3877
3878 /*
3879 * A link to an append-only or immutable file cannot be created.
3880 */
3881 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3882 return -EPERM;
3883 if (!dir->i_op->link)
3884 return -EPERM;
3885 if (S_ISDIR(inode->i_mode))
3886 return -EPERM;
3887
3888 error = security_inode_link(old_dentry, dir, new_dentry);
3889 if (error)
3890 return error;
3891
3892 mutex_lock(&inode->i_mutex);
3893 /* Make sure we don't allow creating hardlink to an unlinked file */
3894 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
3895 error = -ENOENT;
3896 else if (max_links && inode->i_nlink >= max_links)
3897 error = -EMLINK;
3898 else {
3899 error = try_break_deleg(inode, delegated_inode);
3900 if (!error)
3901 error = dir->i_op->link(old_dentry, dir, new_dentry);
3902 }
3903
3904 if (!error && (inode->i_state & I_LINKABLE)) {
3905 spin_lock(&inode->i_lock);
3906 inode->i_state &= ~I_LINKABLE;
3907 spin_unlock(&inode->i_lock);
3908 }
3909 mutex_unlock(&inode->i_mutex);
3910 if (!error)
3911 fsnotify_link(dir, inode, new_dentry);
3912 return error;
3913}
3914EXPORT_SYMBOL(vfs_link);
3915
3916/*
3917 * Hardlinks are often used in delicate situations. We avoid
3918 * security-related surprises by not following symlinks on the
3919 * newname. --KAB
3920 *
3921 * We don't follow them on the oldname either to be compatible
3922 * with linux 2.0, and to avoid hard-linking to directories
3923 * and other special files. --ADM
3924 */
3925SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3926 int, newdfd, const char __user *, newname, int, flags)
3927{
3928 struct dentry *new_dentry;
3929 struct path old_path, new_path;
3930 struct inode *delegated_inode = NULL;
3931 int how = 0;
3932 int error;
3933
3934 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3935 return -EINVAL;
3936 /*
3937 * To use null names we require CAP_DAC_READ_SEARCH
3938 * This ensures that not everyone will be able to create
3939 * handlink using the passed filedescriptor.
3940 */
3941 if (flags & AT_EMPTY_PATH) {
3942 if (!capable(CAP_DAC_READ_SEARCH))
3943 return -ENOENT;
3944 how = LOOKUP_EMPTY;
3945 }
3946
3947 if (flags & AT_SYMLINK_FOLLOW)
3948 how |= LOOKUP_FOLLOW;
3949retry:
3950 error = user_path_at(olddfd, oldname, how, &old_path);
3951 if (error)
3952 return error;
3953
3954 new_dentry = user_path_create(newdfd, newname, &new_path,
3955 (how & LOOKUP_REVAL));
3956 error = PTR_ERR(new_dentry);
3957 if (IS_ERR(new_dentry))
3958 goto out;
3959
3960 error = -EXDEV;
3961 if (old_path.mnt != new_path.mnt)
3962 goto out_dput;
3963 error = may_linkat(&old_path);
3964 if (unlikely(error))
3965 goto out_dput;
3966 error = security_path_link(old_path.dentry, &new_path, new_dentry);
3967 if (error)
3968 goto out_dput;
3969 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
3970out_dput:
3971 done_path_create(&new_path, new_dentry);
3972 if (delegated_inode) {
3973 error = break_deleg_wait(&delegated_inode);
3974 if (!error) {
3975 path_put(&old_path);
3976 goto retry;
3977 }
3978 }
3979 if (retry_estale(error, how)) {
3980 path_put(&old_path);
3981 how |= LOOKUP_REVAL;
3982 goto retry;
3983 }
3984out:
3985 path_put(&old_path);
3986
3987 return error;
3988}
3989
3990SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3991{
3992 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3993}
3994
3995/**
3996 * vfs_rename - rename a filesystem object
3997 * @old_dir: parent of source
3998 * @old_dentry: source
3999 * @new_dir: parent of destination
4000 * @new_dentry: destination
4001 * @delegated_inode: returns an inode needing a delegation break
4002 * @flags: rename flags
4003 *
4004 * The caller must hold multiple mutexes--see lock_rename()).
4005 *
4006 * If vfs_rename discovers a delegation in need of breaking at either
4007 * the source or destination, it will return -EWOULDBLOCK and return a
4008 * reference to the inode in delegated_inode. The caller should then
4009 * break the delegation and retry. Because breaking a delegation may
4010 * take a long time, the caller should drop all locks before doing
4011 * so.
4012 *
4013 * Alternatively, a caller may pass NULL for delegated_inode. This may
4014 * be appropriate for callers that expect the underlying filesystem not
4015 * to be NFS exported.
4016 *
4017 * The worst of all namespace operations - renaming directory. "Perverted"
4018 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4019 * Problems:
4020 * a) we can get into loop creation. Check is done in is_subdir().
4021 * b) race potential - two innocent renames can create a loop together.
4022 * That's where 4.4 screws up. Current fix: serialization on
4023 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4024 * story.
4025 * c) we have to lock _four_ objects - parents and victim (if it exists),
4026 * and source (if it is not a directory).
4027 * And that - after we got ->i_mutex on parents (until then we don't know
4028 * whether the target exists). Solution: try to be smart with locking
4029 * order for inodes. We rely on the fact that tree topology may change
4030 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4031 * move will be locked. Thus we can rank directories by the tree
4032 * (ancestors first) and rank all non-directories after them.
4033 * That works since everybody except rename does "lock parent, lookup,
4034 * lock child" and rename is under ->s_vfs_rename_mutex.
4035 * HOWEVER, it relies on the assumption that any object with ->lookup()
4036 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4037 * we'd better make sure that there's no link(2) for them.
4038 * d) conversion from fhandle to dentry may come in the wrong moment - when
4039 * we are removing the target. Solution: we will have to grab ->i_mutex
4040 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4041 * ->i_mutex on parents, which works but leads to some truly excessive
4042 * locking].
4043 */
4044int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4045 struct inode *new_dir, struct dentry *new_dentry,
4046 struct inode **delegated_inode, unsigned int flags)
4047{
4048 int error;
4049 bool is_dir = d_is_dir(old_dentry);
4050 const unsigned char *old_name;
4051 struct inode *source = old_dentry->d_inode;
4052 struct inode *target = new_dentry->d_inode;
4053 bool new_is_dir = false;
4054 unsigned max_links = new_dir->i_sb->s_max_links;
4055
4056 if (source == target)
4057 return 0;
4058
4059 error = may_delete(old_dir, old_dentry, is_dir);
4060 if (error)
4061 return error;
4062
4063 if (!target) {
4064 error = may_create(new_dir, new_dentry);
4065 } else {
4066 new_is_dir = d_is_dir(new_dentry);
4067
4068 if (!(flags & RENAME_EXCHANGE))
4069 error = may_delete(new_dir, new_dentry, is_dir);
4070 else
4071 error = may_delete(new_dir, new_dentry, new_is_dir);
4072 }
4073 if (error)
4074 return error;
4075
4076 if (!old_dir->i_op->rename)
4077 return -EPERM;
4078
4079 if (flags && !old_dir->i_op->rename2)
4080 return -EINVAL;
4081
4082 /*
4083 * If we are going to change the parent - check write permissions,
4084 * we'll need to flip '..'.
4085 */
4086 if (new_dir != old_dir) {
4087 if (is_dir) {
4088 error = inode_permission(source, MAY_WRITE);
4089 if (error)
4090 return error;
4091 }
4092 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4093 error = inode_permission(target, MAY_WRITE);
4094 if (error)
4095 return error;
4096 }
4097 }
4098
4099 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4100 flags);
4101 if (error)
4102 return error;
4103
4104 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4105 dget(new_dentry);
4106 if (!is_dir || (flags & RENAME_EXCHANGE))
4107 lock_two_nondirectories(source, target);
4108 else if (target)
4109 mutex_lock(&target->i_mutex);
4110
4111 error = -EBUSY;
4112 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
4113 goto out;
4114
4115 if (max_links && new_dir != old_dir) {
4116 error = -EMLINK;
4117 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4118 goto out;
4119 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4120 old_dir->i_nlink >= max_links)
4121 goto out;
4122 }
4123 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4124 shrink_dcache_parent(new_dentry);
4125 if (!is_dir) {
4126 error = try_break_deleg(source, delegated_inode);
4127 if (error)
4128 goto out;
4129 }
4130 if (target && !new_is_dir) {
4131 error = try_break_deleg(target, delegated_inode);
4132 if (error)
4133 goto out;
4134 }
4135 if (!flags) {
4136 error = old_dir->i_op->rename(old_dir, old_dentry,
4137 new_dir, new_dentry);
4138 } else {
4139 error = old_dir->i_op->rename2(old_dir, old_dentry,
4140 new_dir, new_dentry, flags);
4141 }
4142 if (error)
4143 goto out;
4144
4145 if (!(flags & RENAME_EXCHANGE) && target) {
4146 if (is_dir)
4147 target->i_flags |= S_DEAD;
4148 dont_mount(new_dentry);
4149 }
4150 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4151 if (!(flags & RENAME_EXCHANGE))
4152 d_move(old_dentry, new_dentry);
4153 else
4154 d_exchange(old_dentry, new_dentry);
4155 }
4156out:
4157 if (!is_dir || (flags & RENAME_EXCHANGE))
4158 unlock_two_nondirectories(source, target);
4159 else if (target)
4160 mutex_unlock(&target->i_mutex);
4161 dput(new_dentry);
4162 if (!error) {
4163 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4164 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4165 if (flags & RENAME_EXCHANGE) {
4166 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4167 new_is_dir, NULL, new_dentry);
4168 }
4169 }
4170 fsnotify_oldname_free(old_name);
4171
4172 return error;
4173}
4174EXPORT_SYMBOL(vfs_rename);
4175
4176SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4177 int, newdfd, const char __user *, newname, unsigned int, flags)
4178{
4179 struct dentry *old_dir, *new_dir;
4180 struct dentry *old_dentry, *new_dentry;
4181 struct dentry *trap;
4182 struct nameidata oldnd, newnd;
4183 struct inode *delegated_inode = NULL;
4184 struct filename *from;
4185 struct filename *to;
4186 unsigned int lookup_flags = 0;
4187 bool should_retry = false;
4188 int error;
4189
4190 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE))
4191 return -EINVAL;
4192
4193 if ((flags & RENAME_NOREPLACE) && (flags & RENAME_EXCHANGE))
4194 return -EINVAL;
4195
4196retry:
4197 from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags);
4198 if (IS_ERR(from)) {
4199 error = PTR_ERR(from);
4200 goto exit;
4201 }
4202
4203 to = user_path_parent(newdfd, newname, &newnd, lookup_flags);
4204 if (IS_ERR(to)) {
4205 error = PTR_ERR(to);
4206 goto exit1;
4207 }
4208
4209 error = -EXDEV;
4210 if (oldnd.path.mnt != newnd.path.mnt)
4211 goto exit2;
4212
4213 old_dir = oldnd.path.dentry;
4214 error = -EBUSY;
4215 if (oldnd.last_type != LAST_NORM)
4216 goto exit2;
4217
4218 new_dir = newnd.path.dentry;
4219 if (flags & RENAME_NOREPLACE)
4220 error = -EEXIST;
4221 if (newnd.last_type != LAST_NORM)
4222 goto exit2;
4223
4224 error = mnt_want_write(oldnd.path.mnt);
4225 if (error)
4226 goto exit2;
4227
4228 oldnd.flags &= ~LOOKUP_PARENT;
4229 newnd.flags &= ~LOOKUP_PARENT;
4230 if (!(flags & RENAME_EXCHANGE))
4231 newnd.flags |= LOOKUP_RENAME_TARGET;
4232
4233retry_deleg:
4234 trap = lock_rename(new_dir, old_dir);
4235
4236 old_dentry = lookup_hash(&oldnd);
4237 error = PTR_ERR(old_dentry);
4238 if (IS_ERR(old_dentry))
4239 goto exit3;
4240 /* source must exist */
4241 error = -ENOENT;
4242 if (d_is_negative(old_dentry))
4243 goto exit4;
4244 new_dentry = lookup_hash(&newnd);
4245 error = PTR_ERR(new_dentry);
4246 if (IS_ERR(new_dentry))
4247 goto exit4;
4248 error = -EEXIST;
4249 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4250 goto exit5;
4251 if (flags & RENAME_EXCHANGE) {
4252 error = -ENOENT;
4253 if (d_is_negative(new_dentry))
4254 goto exit5;
4255
4256 if (!d_is_dir(new_dentry)) {
4257 error = -ENOTDIR;
4258 if (newnd.last.name[newnd.last.len])
4259 goto exit5;
4260 }
4261 }
4262 /* unless the source is a directory trailing slashes give -ENOTDIR */
4263 if (!d_is_dir(old_dentry)) {
4264 error = -ENOTDIR;
4265 if (oldnd.last.name[oldnd.last.len])
4266 goto exit5;
4267 if (!(flags & RENAME_EXCHANGE) && newnd.last.name[newnd.last.len])
4268 goto exit5;
4269 }
4270 /* source should not be ancestor of target */
4271 error = -EINVAL;
4272 if (old_dentry == trap)
4273 goto exit5;
4274 /* target should not be an ancestor of source */
4275 if (!(flags & RENAME_EXCHANGE))
4276 error = -ENOTEMPTY;
4277 if (new_dentry == trap)
4278 goto exit5;
4279
4280 error = security_path_rename(&oldnd.path, old_dentry,
4281 &newnd.path, new_dentry, flags);
4282 if (error)
4283 goto exit5;
4284 error = vfs_rename(old_dir->d_inode, old_dentry,
4285 new_dir->d_inode, new_dentry,
4286 &delegated_inode, flags);
4287exit5:
4288 dput(new_dentry);
4289exit4:
4290 dput(old_dentry);
4291exit3:
4292 unlock_rename(new_dir, old_dir);
4293 if (delegated_inode) {
4294 error = break_deleg_wait(&delegated_inode);
4295 if (!error)
4296 goto retry_deleg;
4297 }
4298 mnt_drop_write(oldnd.path.mnt);
4299exit2:
4300 if (retry_estale(error, lookup_flags))
4301 should_retry = true;
4302 path_put(&newnd.path);
4303 putname(to);
4304exit1:
4305 path_put(&oldnd.path);
4306 putname(from);
4307 if (should_retry) {
4308 should_retry = false;
4309 lookup_flags |= LOOKUP_REVAL;
4310 goto retry;
4311 }
4312exit:
4313 return error;
4314}
4315
4316SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4317 int, newdfd, const char __user *, newname)
4318{
4319 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4320}
4321
4322SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4323{
4324 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4325}
4326
4327int readlink_copy(char __user *buffer, int buflen, const char *link)
4328{
4329 int len = PTR_ERR(link);
4330 if (IS_ERR(link))
4331 goto out;
4332
4333 len = strlen(link);
4334 if (len > (unsigned) buflen)
4335 len = buflen;
4336 if (copy_to_user(buffer, link, len))
4337 len = -EFAULT;
4338out:
4339 return len;
4340}
4341EXPORT_SYMBOL(readlink_copy);
4342
4343/*
4344 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4345 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4346 * using) it for any given inode is up to filesystem.
4347 */
4348int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4349{
4350 struct nameidata nd;
4351 void *cookie;
4352 int res;
4353
4354 nd.depth = 0;
4355 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
4356 if (IS_ERR(cookie))
4357 return PTR_ERR(cookie);
4358
4359 res = readlink_copy(buffer, buflen, nd_get_link(&nd));
4360 if (dentry->d_inode->i_op->put_link)
4361 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
4362 return res;
4363}
4364EXPORT_SYMBOL(generic_readlink);
4365
4366/* get the link contents into pagecache */
4367static char *page_getlink(struct dentry * dentry, struct page **ppage)
4368{
4369 char *kaddr;
4370 struct page *page;
4371 struct address_space *mapping = dentry->d_inode->i_mapping;
4372 page = read_mapping_page(mapping, 0, NULL);
4373 if (IS_ERR(page))
4374 return (char*)page;
4375 *ppage = page;
4376 kaddr = kmap(page);
4377 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4378 return kaddr;
4379}
4380
4381int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4382{
4383 struct page *page = NULL;
4384 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
4385 if (page) {
4386 kunmap(page);
4387 page_cache_release(page);
4388 }
4389 return res;
4390}
4391EXPORT_SYMBOL(page_readlink);
4392
4393void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
4394{
4395 struct page *page = NULL;
4396 nd_set_link(nd, page_getlink(dentry, &page));
4397 return page;
4398}
4399EXPORT_SYMBOL(page_follow_link_light);
4400
4401void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
4402{
4403 struct page *page = cookie;
4404
4405 if (page) {
4406 kunmap(page);
4407 page_cache_release(page);
4408 }
4409}
4410EXPORT_SYMBOL(page_put_link);
4411
4412/*
4413 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4414 */
4415int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4416{
4417 struct address_space *mapping = inode->i_mapping;
4418 struct page *page;
4419 void *fsdata;
4420 int err;
4421 char *kaddr;
4422 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4423 if (nofs)
4424 flags |= AOP_FLAG_NOFS;
4425
4426retry:
4427 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4428 flags, &page, &fsdata);
4429 if (err)
4430 goto fail;
4431
4432 kaddr = kmap_atomic(page);
4433 memcpy(kaddr, symname, len-1);
4434 kunmap_atomic(kaddr);
4435
4436 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4437 page, fsdata);
4438 if (err < 0)
4439 goto fail;
4440 if (err < len-1)
4441 goto retry;
4442
4443 mark_inode_dirty(inode);
4444 return 0;
4445fail:
4446 return err;
4447}
4448EXPORT_SYMBOL(__page_symlink);
4449
4450int page_symlink(struct inode *inode, const char *symname, int len)
4451{
4452 return __page_symlink(inode, symname, len,
4453 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4454}
4455EXPORT_SYMBOL(page_symlink);
4456
4457const struct inode_operations page_symlink_inode_operations = {
4458 .readlink = generic_readlink,
4459 .follow_link = page_follow_link_light,
4460 .put_link = page_put_link,
4461};
4462EXPORT_SYMBOL(page_symlink_inode_operations);